fs/xfs/xfs_bmap.c

   1 /*
   2  * Copyright (c) 2000-2004 Silicon Graphics, Inc.  All Rights Reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms of version 2 of the GNU General Public License as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it would be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
  11  *
  12  * Further, this software is distributed without any warranty that it is
  13  * free of the rightful claim of any third person regarding infringement
  14  * or the like.  Any license provided herein, whether implied or
  15  * otherwise, applies only to this software file.  Patent licenses, if
  16  * any, provided herein do not apply to combinations of this program with
  17  * other software, or any other product whatsoever.
  18  *
  19  * You should have received a copy of the GNU General Public License along
  20  * with this program; if not, write the Free Software Foundation, Inc., 59
  21  * Temple Place - Suite 330, Boston MA 02111-1307, USA.
  22  *
  23  * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
  24  * Mountain View, CA  94043, or:
  25  *
  26  * http://www.sgi.com
  27  *
  28  * For further information regarding this notice, see:
  29  *
  30  * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
  31  */
  32
  33 #include "xfs.h"
  34
  35 #include "xfs_macros.h"
  36 #include "xfs_types.h"
  37 #include "xfs_inum.h"
  38 #include "xfs_log.h"
  39 #include "xfs_trans.h"
  40 #include "xfs_sb.h"
  41 #include "xfs_ag.h"
  42 #include "xfs_dir.h"
  43 #include "xfs_dir2.h"
  44 #include "xfs_dmapi.h"
  45 #include "xfs_mount.h"
  46 #include "xfs_alloc_btree.h"
  47 #include "xfs_bmap_btree.h"
  48 #include "xfs_ialloc_btree.h"
  49 #include "xfs_btree.h"
  50 #include "xfs_ialloc.h"
  51 #include "xfs_attr_sf.h"
  52 #include "xfs_dir_sf.h"
  53 #include "xfs_dir2_sf.h"
  54 #include "xfs_dinode.h"
  55 #include "xfs_inode_item.h"
  56 #include "xfs_inode.h"
  57 #include "xfs_itable.h"
  58 #include "xfs_extfree_item.h"
  59 #include "xfs_alloc.h"
  60 #include "xfs_bmap.h"
  61 #include "xfs_rtalloc.h"
  62 #include "xfs_error.h"
  63 #include "xfs_da_btree.h"
  64 #include "xfs_dir_leaf.h"
  65 #include "xfs_bit.h"
  66 #include "xfs_rw.h"
  67 #include "xfs_quota.h"
  68 #include "xfs_trans_space.h"
  69 #include "xfs_buf_item.h"
  70
  71
  72 #ifdef DEBUG
  73 STATIC void
  74 xfs_bmap_check_leaf_extents(xfs_btree_cur_t *cur, xfs_inode_t *ip, int whichfork);
  75 #endif
  76
  77 kmem_zone_t             *xfs_bmap_free_item_zone;
  78
  79 /*
  80  * Prototypes for internal bmap routines.
  81  */
  82
  83
  84 /*
  85  * Called from xfs_bmap_add_attrfork to handle extents format files.
  86  */
  87 STATIC int                                      /* error */
  88 xfs_bmap_add_attrfork_extents(
  89         xfs_trans_t             *tp,            /* transaction pointer */
  90         xfs_inode_t             *ip,            /* incore inode pointer */
  91         xfs_fsblock_t           *firstblock,    /* first block allocated */
  92         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
  93         int                     *flags);        /* inode logging flags */
  94
  95 /*
  96  * Called from xfs_bmap_add_attrfork to handle local format files.
  97  */
  98 STATIC int                                      /* error */
  99 xfs_bmap_add_attrfork_local(
 100         xfs_trans_t             *tp,            /* transaction pointer */
 101         xfs_inode_t             *ip,            /* incore inode pointer */
 102         xfs_fsblock_t           *firstblock,    /* first block allocated */
 103         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 104         int                     *flags);        /* inode logging flags */
 105
 106 /*
 107  * Called by xfs_bmapi to update extent list structure and the btree
 108  * after allocating space (or doing a delayed allocation).
 109  */
 110 STATIC int                              /* error */
 111 xfs_bmap_add_extent(
 112         xfs_inode_t             *ip,    /* incore inode pointer */
 113         xfs_extnum_t            idx,    /* extent number to update/insert */
 114         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 115         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
 116         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 117         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 118         int                     *logflagsp, /* inode logging flags */
 119         int                     whichfork, /* data or attr fork */
 120         int                     rsvd);  /* OK to allocate reserved blocks */
 121
 122 /*
 123  * Called by xfs_bmap_add_extent to handle cases converting a delayed
 124  * allocation to a real allocation.
 125  */
 126 STATIC int                              /* error */
 127 xfs_bmap_add_extent_delay_real(
 128         xfs_inode_t             *ip,    /* incore inode pointer */
 129         xfs_extnum_t            idx,    /* extent number to update/insert */
 130         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 131         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
 132         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
 133         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 134         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 135         int                     *logflagsp, /* inode logging flags */
 136         int                     rsvd);  /* OK to allocate reserved blocks */
 137
 138 /*
 139  * Called by xfs_bmap_add_extent to handle cases converting a hole
 140  * to a delayed allocation.
 141  */
 142 STATIC int                              /* error */
 143 xfs_bmap_add_extent_hole_delay(
 144         xfs_inode_t             *ip,    /* incore inode pointer */
 145         xfs_extnum_t            idx,    /* extent number to update/insert */
 146         xfs_btree_cur_t         *cur,   /* if null, not a btree */
 147         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
 148         int                     *logflagsp,/* inode logging flags */
 149         int                     rsvd);  /* OK to allocate reserved blocks */
 150
 151 /*
 152  * Called by xfs_bmap_add_extent to handle cases converting a hole
 153  * to a real allocation.
 154  */
 155 STATIC int                              /* error */
 156 xfs_bmap_add_extent_hole_real(
 157         xfs_inode_t             *ip,    /* incore inode pointer */
 158         xfs_extnum_t            idx,    /* extent number to update/insert */
 159         xfs_btree_cur_t         *cur,   /* if null, not a btree */
 160         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
 161         int                     *logflagsp, /* inode logging flags */
 162         int                     whichfork); /* data or attr fork */
 163
 164 /*
 165  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
 166  * allocation to a real allocation or vice versa.
 167  */
 168 STATIC int                              /* error */
 169 xfs_bmap_add_extent_unwritten_real(
 170         xfs_inode_t             *ip,    /* incore inode pointer */
 171         xfs_extnum_t            idx,    /* extent number to update/insert */
 172         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 173         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
 174         int                     *logflagsp); /* inode logging flags */
 175
 176 /*
 177  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
 178  * It figures out where to ask the underlying allocator to put the new extent.
 179  */
 180 STATIC int                              /* error */
 181 xfs_bmap_alloc(
 182         xfs_bmalloca_t          *ap);   /* bmap alloc argument struct */
 183
 184 /*
 185  * Transform a btree format file with only one leaf node, where the
 186  * extents list will fit in the inode, into an extents format file.
 187  * Since the extent list is already in-core, all we have to do is
 188  * give up the space for the btree root and pitch the leaf block.
 189  */
 190 STATIC int                              /* error */
 191 xfs_bmap_btree_to_extents(
 192         xfs_trans_t             *tp,    /* transaction pointer */
 193         xfs_inode_t             *ip,    /* incore inode pointer */
 194         xfs_btree_cur_t         *cur,   /* btree cursor */
 195         int                     *logflagsp, /* inode logging flags */
 196         int                     whichfork); /* data or attr fork */
 197
 198 #ifdef DEBUG
 199 /*
 200  * Check that the extents list for the inode ip is in the right order.
 201  */
 202 STATIC void
 203 xfs_bmap_check_extents(
 204         xfs_inode_t             *ip,            /* incore inode pointer */
 205         int                     whichfork);     /* data or attr fork */
 206 #endif
 207
 208 /*
 209  * Called by xfs_bmapi to update extent list structure and the btree
 210  * after removing space (or undoing a delayed allocation).
 211  */
 212 STATIC int                              /* error */
 213 xfs_bmap_del_extent(
 214         xfs_inode_t             *ip,    /* incore inode pointer */
 215         xfs_trans_t             *tp,    /* current trans pointer */
 216         xfs_extnum_t            idx,    /* extent number to update/insert */
 217         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 218         xfs_btree_cur_t         *cur,   /* if null, not a btree */
 219         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
 220         int                     *logflagsp,/* inode logging flags */
 221         int                     whichfork, /* data or attr fork */
 222         int                     rsvd);   /* OK to allocate reserved blocks */
 223
 224 /*
 225  * Remove the entry "free" from the free item list.  Prev points to the
 226  * previous entry, unless "free" is the head of the list.
 227  */
 228 STATIC void
 229 xfs_bmap_del_free(
 230         xfs_bmap_free_t         *flist, /* free item list header */
 231         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
 232         xfs_bmap_free_item_t    *free); /* list item to be freed */
 233
 234 /*
 235  * Remove count entries from the extents array for inode "ip", starting
 236  * at index "idx".  Copies the remaining items down over the deleted ones,
 237  * and gives back the excess memory.
 238  */
 239 STATIC void
 240 xfs_bmap_delete_exlist(
 241         xfs_inode_t     *ip,            /* incode inode pointer */
 242         xfs_extnum_t    idx,            /* starting delete index */
 243         xfs_extnum_t    count,          /* count of items to delete */
 244         int             whichfork);     /* data or attr fork */
 245
 246 /*
 247  * Convert an extents-format file into a btree-format file.
 248  * The new file will have a root block (in the inode) and a single child block.
 249  */
 250 STATIC int                                      /* error */
 251 xfs_bmap_extents_to_btree(
 252         xfs_trans_t             *tp,            /* transaction pointer */
 253         xfs_inode_t             *ip,            /* incore inode pointer */
 254         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
 255         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
 256         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
 257         int                     wasdel,         /* converting a delayed alloc */
 258         int                     *logflagsp,     /* inode logging flags */
 259         int                     whichfork);     /* data or attr fork */
 260
 261 /*
 262  * Insert new item(s) in the extent list for inode "ip".
 263  * Count new items are inserted at offset idx.
 264  */
 265 STATIC void
 266 xfs_bmap_insert_exlist(
 267         xfs_inode_t     *ip,            /* incore inode pointer */
 268         xfs_extnum_t    idx,            /* starting index of new items */
 269         xfs_extnum_t    count,          /* number of inserted items */
 270         xfs_bmbt_irec_t *new,           /* items to insert */
 271         int             whichfork);     /* data or attr fork */
 272
 273 /*
 274  * Convert a local file to an extents file.
 275  * This code is sort of bogus, since the file data needs to get
 276  * logged so it won't be lost.  The bmap-level manipulations are ok, though.
 277  */
 278 STATIC int                              /* error */
 279 xfs_bmap_local_to_extents(
 280         xfs_trans_t     *tp,            /* transaction pointer */
 281         xfs_inode_t     *ip,            /* incore inode pointer */
 282         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
 283         xfs_extlen_t    total,          /* total blocks needed by transaction */
 284         int             *logflagsp,     /* inode logging flags */
 285         int             whichfork);     /* data or attr fork */
 286
 287 /*
 288  * Search the extents list for the inode, for the extent containing bno.
 289  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
 290  * *eofp will be set, and *prevp will contain the last entry (null if none).
 291  * Else, *lastxp will be set to the index of the found
 292  * entry; *gotp will contain the entry.
 293  */
 294 STATIC xfs_bmbt_rec_t *                 /* pointer to found extent entry */
 295 xfs_bmap_search_extents(
 296         xfs_inode_t     *ip,            /* incore inode pointer */
 297         xfs_fileoff_t   bno,            /* block number searched for */
 298         int             whichfork,      /* data or attr fork */
 299         int             *eofp,          /* out: end of file found */
 300         xfs_extnum_t    *lastxp,        /* out: last extent index */
 301         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
 302         xfs_bmbt_irec_t *prevp);        /* out: previous extent entry found */
 303
 304 #ifdef XFS_BMAP_TRACE
 305 /*
 306  * Add a bmap trace buffer entry.  Base routine for the others.
 307  */
 308 STATIC void
 309 xfs_bmap_trace_addentry(
 310         int             opcode,         /* operation */
 311         char            *fname,         /* function name */
 312         char            *desc,          /* operation description */
 313         xfs_inode_t     *ip,            /* incore inode pointer */
 314         xfs_extnum_t    idx,            /* index of entry(ies) */
 315         xfs_extnum_t    cnt,            /* count of entries, 1 or 2 */
 316         xfs_bmbt_rec_t  *r1,            /* first record */
 317         xfs_bmbt_rec_t  *r2,            /* second record or null */
 318         int             whichfork);     /* data or attr fork */
 319
 320 /*
 321  * Add bmap trace entry prior to a call to xfs_bmap_delete_exlist.
 322  */
 323 STATIC void
 324 xfs_bmap_trace_delete(
 325         char            *fname,         /* function name */
 326         char            *desc,          /* operation description */
 327         xfs_inode_t     *ip,            /* incore inode pointer */
 328         xfs_extnum_t    idx,            /* index of entry(entries) deleted */
 329         xfs_extnum_t    cnt,            /* count of entries deleted, 1 or 2 */
 330         int             whichfork);     /* data or attr fork */
 331
 332 /*
 333  * Add bmap trace entry prior to a call to xfs_bmap_insert_exlist, or
 334  * reading in the extents list from the disk (in the btree).
 335  */
 336 STATIC void
 337 xfs_bmap_trace_insert(
 338         char            *fname,         /* function name */
 339         char            *desc,          /* operation description */
 340         xfs_inode_t     *ip,            /* incore inode pointer */
 341         xfs_extnum_t    idx,            /* index of entry(entries) inserted */
 342         xfs_extnum_t    cnt,            /* count of entries inserted, 1 or 2 */
 343         xfs_bmbt_irec_t *r1,            /* inserted record 1 */
 344         xfs_bmbt_irec_t *r2,            /* inserted record 2 or null */
 345         int             whichfork);     /* data or attr fork */
 346
 347 /*
 348  * Add bmap trace entry after updating an extent list entry in place.
 349  */
 350 STATIC void
 351 xfs_bmap_trace_post_update(
 352         char            *fname,         /* function name */
 353         char            *desc,          /* operation description */
 354         xfs_inode_t     *ip,            /* incore inode pointer */
 355         xfs_extnum_t    idx,            /* index of entry updated */
 356         int             whichfork);     /* data or attr fork */
 357
 358 /*
 359  * Add bmap trace entry prior to updating an extent list entry in place.
 360  */
 361 STATIC void
 362 xfs_bmap_trace_pre_update(
 363         char            *fname,         /* function name */
 364         char            *desc,          /* operation description */
 365         xfs_inode_t     *ip,            /* incore inode pointer */
 366         xfs_extnum_t    idx,            /* index of entry to be updated */
 367         int             whichfork);     /* data or attr fork */
 368
 369 #else
 370 #define xfs_bmap_trace_delete(f,d,ip,i,c,w)
 371 #define xfs_bmap_trace_insert(f,d,ip,i,c,r1,r2,w)
 372 #define xfs_bmap_trace_post_update(f,d,ip,i,w)
 373 #define xfs_bmap_trace_pre_update(f,d,ip,i,w)
 374 #endif  /* XFS_BMAP_TRACE */
 375
 376 /*
 377  * Compute the worst-case number of indirect blocks that will be used
 378  * for ip's delayed extent of length "len".
 379  */
 380 STATIC xfs_filblks_t
 381 xfs_bmap_worst_indlen(
 382         xfs_inode_t             *ip,    /* incore inode pointer */
 383         xfs_filblks_t           len);   /* delayed extent length */
 384
 385 #ifdef DEBUG
 386 /*
 387  * Perform various validation checks on the values being returned
 388  * from xfs_bmapi().
 389  */
 390 STATIC void
 391 xfs_bmap_validate_ret(
 392         xfs_fileoff_t           bno,
 393         xfs_filblks_t           len,
 394         int                     flags,
 395         xfs_bmbt_irec_t         *mval,
 396         int                     nmap,
 397         int                     ret_nmap);
 398 #else
 399 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
 400 #endif /* DEBUG */
 401
 402 #if defined(XFS_RW_TRACE)
 403 STATIC void
 404 xfs_bunmap_trace(
 405         xfs_inode_t             *ip,
 406         xfs_fileoff_t           bno,
 407         xfs_filblks_t           len,
 408         int                     flags,
 409         inst_t                  *ra);
 410 #else
 411 #define xfs_bunmap_trace(ip, bno, len, flags, ra)
 412 #endif  /* XFS_RW_TRACE */
 413
 414 STATIC int
 415 xfs_bmap_count_tree(
 416         xfs_mount_t     *mp,
 417         xfs_trans_t     *tp,
 418         xfs_fsblock_t   blockno,
 419         int             levelin,
 420         int             *count);
 421
 422 STATIC int
 423 xfs_bmap_count_leaves(
 424         xfs_bmbt_rec_t          *frp,
 425         int                     numrecs,
 426         int                     *count);
 427
 428 /*
 429  * Bmap internal routines.
 430  */
 431
 432 /*
 433  * Called from xfs_bmap_add_attrfork to handle btree format files.
 434  */
 435 STATIC int                                      /* error */
 436 xfs_bmap_add_attrfork_btree(
 437         xfs_trans_t             *tp,            /* transaction pointer */
 438         xfs_inode_t             *ip,            /* incore inode pointer */
 439         xfs_fsblock_t           *firstblock,    /* first block allocated */
 440         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 441         int                     *flags)         /* inode logging flags */
 442 {
 443         xfs_btree_cur_t         *cur;           /* btree cursor */
 444         int                     error;          /* error return value */
 445         xfs_mount_t             *mp;            /* file system mount struct */
 446         int                     stat;           /* newroot status */
 447
 448         mp = ip->i_mount;
 449         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
 450                 *flags |= XFS_ILOG_DBROOT;
 451         else {
 452                 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
 453                         XFS_DATA_FORK);
 454                 cur->bc_private.b.flist = flist;
 455                 cur->bc_private.b.firstblock = *firstblock;
 456                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
 457                         goto error0;
 458                 ASSERT(stat == 1);      /* must be at least one entry */
 459                 if ((error = xfs_bmbt_newroot(cur, flags, &stat)))
 460                         goto error0;
 461                 if (stat == 0) {
 462                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 463                         return XFS_ERROR(ENOSPC);
 464                 }
 465                 *firstblock = cur->bc_private.b.firstblock;
 466                 cur->bc_private.b.allocated = 0;
 467                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 468         }
 469         return 0;
 470 error0:
 471         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 472         return error;
 473 }
 474
 475 /*
 476  * Called from xfs_bmap_add_attrfork to handle extents format files.
 477  */
 478 STATIC int                                      /* error */
 479 xfs_bmap_add_attrfork_extents(
 480         xfs_trans_t             *tp,            /* transaction pointer */
 481         xfs_inode_t             *ip,            /* incore inode pointer */
 482         xfs_fsblock_t           *firstblock,    /* first block allocated */
 483         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 484         int                     *flags)         /* inode logging flags */
 485 {
 486         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
 487         int                     error;          /* error return value */
 488
 489         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
 490                 return 0;
 491         cur = NULL;
 492         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
 493                 flags, XFS_DATA_FORK);
 494         if (cur) {
 495                 cur->bc_private.b.allocated = 0;
 496                 xfs_btree_del_cursor(cur,
 497                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
 498         }
 499         return error;
 500 }
 501
 502 /*
 503  * Called from xfs_bmap_add_attrfork to handle local format files.
 504  */
 505 STATIC int                                      /* error */
 506 xfs_bmap_add_attrfork_local(
 507         xfs_trans_t             *tp,            /* transaction pointer */
 508         xfs_inode_t             *ip,            /* incore inode pointer */
 509         xfs_fsblock_t           *firstblock,    /* first block allocated */
 510         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 511         int                     *flags)         /* inode logging flags */
 512 {
 513         xfs_da_args_t           dargs;          /* args for dir/attr code */
 514         int                     error;          /* error return value */
 515         xfs_mount_t             *mp;            /* mount structure pointer */
 516
 517         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
 518                 return 0;
 519         if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
 520                 mp = ip->i_mount;
 521                 memset(&dargs, 0, sizeof(dargs));
 522                 dargs.dp = ip;
 523                 dargs.firstblock = firstblock;
 524                 dargs.flist = flist;
 525                 dargs.total = mp->m_dirblkfsbs;
 526                 dargs.whichfork = XFS_DATA_FORK;
 527                 dargs.trans = tp;
 528                 error = XFS_DIR_SHORTFORM_TO_SINGLE(mp, &dargs);
 529         } else
 530                 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
 531                         XFS_DATA_FORK);
 532         return error;
 533 }
 534
 535 /*
 536  * Called by xfs_bmapi to update extent list structure and the btree
 537  * after allocating space (or doing a delayed allocation).
 538  */
 539 STATIC int                              /* error */
 540 xfs_bmap_add_extent(
 541         xfs_inode_t             *ip,    /* incore inode pointer */
 542         xfs_extnum_t            idx,    /* extent number to update/insert */
 543         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 544         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
 545         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 546         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 547         int                     *logflagsp, /* inode logging flags */
 548         int                     whichfork, /* data or attr fork */
 549         int                     rsvd)   /* OK to use reserved data blocks */
 550 {
 551         xfs_btree_cur_t         *cur;   /* btree cursor or null */
 552         xfs_filblks_t           da_new; /* new count del alloc blocks used */
 553         xfs_filblks_t           da_old; /* old count del alloc blocks used */
 554         int                     error;  /* error return value */
 555 #ifdef XFS_BMAP_TRACE
 556         static char             fname[] = "xfs_bmap_add_extent";
 557 #endif
 558         xfs_ifork_t             *ifp;   /* inode fork ptr */
 559         int                     logflags; /* returned value */
 560         xfs_extnum_t            nextents; /* number of extents in file now */
 561
 562         XFS_STATS_INC(xs_add_exlist);
 563         cur = *curp;
 564         ifp = XFS_IFORK_PTR(ip, whichfork);
 565         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
 566         ASSERT(idx <= nextents);
 567         da_old = da_new = 0;
 568         error = 0;
 569         /*
 570          * This is the first extent added to a new/empty file.
 571          * Special case this one, so other routines get to assume there are
 572          * already extents in the list.
 573          */
 574         if (nextents == 0) {
 575                 xfs_bmap_trace_insert(fname, "insert empty", ip, 0, 1, new,
 576                         NULL, whichfork);
 577                 xfs_bmap_insert_exlist(ip, 0, 1, new, whichfork);
 578                 ASSERT(cur == NULL);
 579                 ifp->if_lastex = 0;
 580                 if (!ISNULLSTARTBLOCK(new->br_startblock)) {
 581                         XFS_IFORK_NEXT_SET(ip, whichfork, 1);
 582                         logflags = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
 583                 } else
 584                         logflags = 0;
 585         }
 586         /*
 587          * Any kind of new delayed allocation goes here.
 588          */
 589         else if (ISNULLSTARTBLOCK(new->br_startblock)) {
 590                 if (cur)
 591                         ASSERT((cur->bc_private.b.flags &
 592                                 XFS_BTCUR_BPRV_WASDEL) == 0);
 593                 if ((error = xfs_bmap_add_extent_hole_delay(ip, idx, cur, new,
 594                                 &logflags, rsvd)))
 595                         goto done;
 596         }
 597         /*
 598          * Real allocation off the end of the file.
 599          */
 600         else if (idx == nextents) {
 601                 if (cur)
 602                         ASSERT((cur->bc_private.b.flags &
 603                                 XFS_BTCUR_BPRV_WASDEL) == 0);
 604                 if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur, new,
 605                                 &logflags, whichfork)))
 606                         goto done;
 607         } else {
 608                 xfs_bmbt_irec_t prev;   /* old extent at offset idx */
 609
 610                 /*
 611                  * Get the record referred to by idx.
 612                  */
 613                 xfs_bmbt_get_all(&ifp->if_u1.if_extents[idx], &prev);
 614                 /*
 615                  * If it's a real allocation record, and the new allocation ends
 616                  * after the start of the referred to record, then we're filling
 617                  * in a delayed or unwritten allocation with a real one, or
 618                  * converting real back to unwritten.
 619                  */
 620                 if (!ISNULLSTARTBLOCK(new->br_startblock) &&
 621                     new->br_startoff + new->br_blockcount > prev.br_startoff) {
 622                         if (prev.br_state != XFS_EXT_UNWRITTEN &&
 623                             ISNULLSTARTBLOCK(prev.br_startblock)) {
 624                                 da_old = STARTBLOCKVAL(prev.br_startblock);
 625                                 if (cur)
 626                                         ASSERT(cur->bc_private.b.flags &
 627                                                 XFS_BTCUR_BPRV_WASDEL);
 628                                 if ((error = xfs_bmap_add_extent_delay_real(ip,
 629                                         idx, &cur, new, &da_new, first, flist,
 630                                         &logflags, rsvd)))
 631                                         goto done;
 632                         } else if (new->br_state == XFS_EXT_NORM) {
 633                                 ASSERT(new->br_state == XFS_EXT_NORM);
 634                                 if ((error = xfs_bmap_add_extent_unwritten_real(
 635                                         ip, idx, &cur, new, &logflags)))
 636                                         goto done;
 637                         } else {
 638                                 ASSERT(new->br_state == XFS_EXT_UNWRITTEN);
 639                                 if ((error = xfs_bmap_add_extent_unwritten_real(
 640                                         ip, idx, &cur, new, &logflags)))
 641                                         goto done;
 642                         }
 643                         ASSERT(*curp == cur || *curp == NULL);
 644                 }
 645                 /*
 646                  * Otherwise we're filling in a hole with an allocation.
 647                  */
 648                 else {
 649                         if (cur)
 650                                 ASSERT((cur->bc_private.b.flags &
 651                                         XFS_BTCUR_BPRV_WASDEL) == 0);
 652                         if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur,
 653                                         new, &logflags, whichfork)))
 654                                 goto done;
 655                 }
 656         }
 657
 658         ASSERT(*curp == cur || *curp == NULL);
 659         /*
 660          * Convert to a btree if necessary.
 661          */
 662         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
 663             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
 664                 int     tmp_logflags;   /* partial log flag return val */
 665
 666                 ASSERT(cur == NULL);
 667                 error = xfs_bmap_extents_to_btree(ip->i_transp, ip, first,
 668                         flist, &cur, da_old > 0, &tmp_logflags, whichfork);
 669                 logflags |= tmp_logflags;
 670                 if (error)
 671                         goto done;
 672         }
 673         /*
 674          * Adjust for changes in reserved delayed indirect blocks.
 675          * Nothing to do for disk quotas here.
 676          */
 677         if (da_old || da_new) {
 678                 xfs_filblks_t   nblks;
 679
 680                 nblks = da_new;
 681                 if (cur)
 682                         nblks += cur->bc_private.b.allocated;
 683                 ASSERT(nblks <= da_old);
 684                 if (nblks < da_old)
 685                         xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
 686                                 (int)(da_old - nblks), rsvd);
 687         }
 688         /*
 689          * Clear out the allocated field, done with it now in any case.
 690          */
 691         if (cur) {
 692                 cur->bc_private.b.allocated = 0;
 693                 *curp = cur;
 694         }
 695 done:
 696 #ifdef DEBUG
 697         if (!error)
 698                 xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
 699 #endif
 700         *logflagsp = logflags;
 701         return error;
 702 }
 703
 704 /*
 705  * Called by xfs_bmap_add_extent to handle cases converting a delayed
 706  * allocation to a real allocation.
 707  */
 708 STATIC int                              /* error */
 709 xfs_bmap_add_extent_delay_real(
 710         xfs_inode_t             *ip,    /* incore inode pointer */
 711         xfs_extnum_t            idx,    /* extent number to update/insert */
 712         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 713         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
 714         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
 715         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 716         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 717         int                     *logflagsp, /* inode logging flags */
 718         int                     rsvd)   /* OK to use reserved data block allocation */
 719 {
 720         xfs_bmbt_rec_t          *base;  /* base of extent entry list */
 721         xfs_btree_cur_t         *cur;   /* btree cursor */
 722         int                     diff;   /* temp value */
 723         xfs_bmbt_rec_t          *ep;    /* extent entry for idx */
 724         int                     error;  /* error return value */
 725 #ifdef XFS_BMAP_TRACE
 726         static char             fname[] = "xfs_bmap_add_extent_delay_real";
 727 #endif
 728         int                     i;      /* temp state */
 729         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
 730         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
 731                                         /* left is 0, right is 1, prev is 2 */
 732         int                     rval=0; /* return value (logging flags) */
 733         int                     state = 0;/* state bits, accessed thru macros */
 734         xfs_filblks_t           temp;   /* value for dnew calculations */
 735         xfs_filblks_t           temp2;  /* value for dnew calculations */
 736         int                     tmp_rval;       /* partial logging flags */
 737         enum {                          /* bit number definitions for state */
 738                 LEFT_CONTIG,    RIGHT_CONTIG,
 739                 LEFT_FILLING,   RIGHT_FILLING,
 740                 LEFT_DELAY,     RIGHT_DELAY,
 741                 LEFT_VALID,     RIGHT_VALID
 742         };
 743
 744 #define LEFT            r[0]
 745 #define RIGHT           r[1]
 746 #define PREV            r[2]
 747 #define MASK(b)         (1 << (b))
 748 #define MASK2(a,b)      (MASK(a) | MASK(b))
 749 #define MASK3(a,b,c)    (MASK2(a,b) | MASK(c))
 750 #define MASK4(a,b,c,d)  (MASK3(a,b,c) | MASK(d))
 751 #define STATE_SET(b,v)  ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
 752 #define STATE_TEST(b)   (state & MASK(b))
 753 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
 754                                        ((state &= ~MASK(b)), 0))
 755 #define SWITCH_STATE            \
 756         (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
 757
 758         /*
 759          * Set up a bunch of variables to make the tests simpler.
 760          */
 761         cur = *curp;
 762         base = ip->i_df.if_u1.if_extents;
 763         ep = &base[idx];
 764         xfs_bmbt_get_all(ep, &PREV);
 765         new_endoff = new->br_startoff + new->br_blockcount;
 766         ASSERT(PREV.br_startoff <= new->br_startoff);
 767         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
 768         /*
 769          * Set flags determining what part of the previous delayed allocation
 770          * extent is being replaced by a real allocation.
 771          */
 772         STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
 773         STATE_SET(RIGHT_FILLING,
 774                 PREV.br_startoff + PREV.br_blockcount == new_endoff);
 775         /*
 776          * Check and set flags if this segment has a left neighbor.
 777          * Don't set contiguous if the combined extent would be too large.
 778          */
 779         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
 780                 xfs_bmbt_get_all(ep - 1, &LEFT);
 781                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
 782         }
 783         STATE_SET(LEFT_CONTIG,
 784                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
 785                 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
 786                 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
 787                 LEFT.br_state == new->br_state &&
 788                 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
 789         /*
 790          * Check and set flags if this segment has a right neighbor.
 791          * Don't set contiguous if the combined extent would be too large.
 792          * Also check for all-three-contiguous being too large.
 793          */
 794         if (STATE_SET_TEST(RIGHT_VALID,
 795                         idx <
 796                         ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
 797                 xfs_bmbt_get_all(ep + 1, &RIGHT);
 798                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
 799         }
 800         STATE_SET(RIGHT_CONTIG,
 801                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
 802                 new_endoff == RIGHT.br_startoff &&
 803                 new->br_startblock + new->br_blockcount ==
 804                     RIGHT.br_startblock &&
 805                 new->br_state == RIGHT.br_state &&
 806                 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
 807                 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
 808                   MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
 809                  LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
 810                      <= MAXEXTLEN));
 811         error = 0;
 812         /*
 813          * Switch out based on the FILLING and CONTIG state bits.
 814          */
 815         switch (SWITCH_STATE) {
 816
 817         case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
 818                 /*
 819                  * Filling in all of a previously delayed allocation extent.
 820                  * The left and right neighbors are both contiguous with new.
 821                  */
 822                 xfs_bmap_trace_pre_update(fname, "LF|RF|LC|RC", ip, idx - 1,
 823                         XFS_DATA_FORK);
 824                 xfs_bmbt_set_blockcount(ep - 1,
 825                         LEFT.br_blockcount + PREV.br_blockcount +
 826                         RIGHT.br_blockcount);
 827                 xfs_bmap_trace_post_update(fname, "LF|RF|LC|RC", ip, idx - 1,
 828                         XFS_DATA_FORK);
 829                 xfs_bmap_trace_delete(fname, "LF|RF|LC|RC", ip, idx, 2,
 830                         XFS_DATA_FORK);
 831                 xfs_bmap_delete_exlist(ip, idx, 2, XFS_DATA_FORK);
 832                 ip->i_df.if_lastex = idx - 1;
 833                 ip->i_d.di_nextents--;
 834                 if (cur == NULL)
 835                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 836                 else {
 837                         rval = XFS_ILOG_CORE;
 838                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 839                                         RIGHT.br_startblock,
 840                                         RIGHT.br_blockcount, &i)))
 841                                 goto done;
 842                         ASSERT(i == 1);
 843                         if ((error = xfs_bmbt_delete(cur, &i)))
 844                                 goto done;
 845                         ASSERT(i == 1);
 846                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
 847                                 goto done;
 848                         ASSERT(i == 1);
 849                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 850                                         LEFT.br_startblock,
 851                                         LEFT.br_blockcount +
 852                                         PREV.br_blockcount +
 853                                         RIGHT.br_blockcount, LEFT.br_state)))
 854                                 goto done;
 855                 }
 856                 *dnew = 0;
 857                 break;
 858
 859         case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
 860                 /*
 861                  * Filling in all of a previously delayed allocation extent.
 862                  * The left neighbor is contiguous, the right is not.
 863                  */
 864                 xfs_bmap_trace_pre_update(fname, "LF|RF|LC", ip, idx - 1,
 865                         XFS_DATA_FORK);
 866                 xfs_bmbt_set_blockcount(ep - 1,
 867                         LEFT.br_blockcount + PREV.br_blockcount);
 868                 xfs_bmap_trace_post_update(fname, "LF|RF|LC", ip, idx - 1,
 869                         XFS_DATA_FORK);
 870                 ip->i_df.if_lastex = idx - 1;
 871                 xfs_bmap_trace_delete(fname, "LF|RF|LC", ip, idx, 1,
 872                         XFS_DATA_FORK);
 873                 xfs_bmap_delete_exlist(ip, idx, 1, XFS_DATA_FORK);
 874                 if (cur == NULL)
 875                         rval = XFS_ILOG_DEXT;
 876                 else {
 877                         rval = 0;
 878                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
 879                                         LEFT.br_startblock, LEFT.br_blockcount,
 880                                         &i)))
 881                                 goto done;
 882                         ASSERT(i == 1);
 883                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 884                                         LEFT.br_startblock,
 885                                         LEFT.br_blockcount +
 886                                         PREV.br_blockcount, LEFT.br_state)))
 887                                 goto done;
 888                 }
 889                 *dnew = 0;
 890                 break;
 891
 892         case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
 893                 /*
 894                  * Filling in all of a previously delayed allocation extent.
 895                  * The right neighbor is contiguous, the left is not.
 896                  */
 897                 xfs_bmap_trace_pre_update(fname, "LF|RF|RC", ip, idx,
 898                         XFS_DATA_FORK);
 899                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 900                 xfs_bmbt_set_blockcount(ep,
 901                         PREV.br_blockcount + RIGHT.br_blockcount);
 902                 xfs_bmap_trace_post_update(fname, "LF|RF|RC", ip, idx,
 903                         XFS_DATA_FORK);
 904                 ip->i_df.if_lastex = idx;
 905                 xfs_bmap_trace_delete(fname, "LF|RF|RC", ip, idx + 1, 1,
 906                         XFS_DATA_FORK);
 907                 xfs_bmap_delete_exlist(ip, idx + 1, 1, XFS_DATA_FORK);
 908                 if (cur == NULL)
 909                         rval = XFS_ILOG_DEXT;
 910                 else {
 911                         rval = 0;
 912                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 913                                         RIGHT.br_startblock,
 914                                         RIGHT.br_blockcount, &i)))
 915                                 goto done;
 916                         ASSERT(i == 1);
 917                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
 918                                         new->br_startblock,
 919                                         PREV.br_blockcount +
 920                                         RIGHT.br_blockcount, PREV.br_state)))
 921                                 goto done;
 922                 }
 923                 *dnew = 0;
 924                 break;
 925
 926         case MASK2(LEFT_FILLING, RIGHT_FILLING):
 927                 /*
 928                  * Filling in all of a previously delayed allocation extent.
 929                  * Neither the left nor right neighbors are contiguous with
 930                  * the new one.
 931                  */
 932                 xfs_bmap_trace_pre_update(fname, "LF|RF", ip, idx,
 933                         XFS_DATA_FORK);
 934                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 935                 xfs_bmap_trace_post_update(fname, "LF|RF", ip, idx,
 936                         XFS_DATA_FORK);
 937                 ip->i_df.if_lastex = idx;
 938                 ip->i_d.di_nextents++;
 939                 if (cur == NULL)
 940                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 941                 else {
 942                         rval = XFS_ILOG_CORE;
 943                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
 944                                         new->br_startblock, new->br_blockcount,
 945                                         &i)))
 946                                 goto done;
 947                         ASSERT(i == 0);
 948                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
 949                         if ((error = xfs_bmbt_insert(cur, &i)))
 950                                 goto done;
 951                         ASSERT(i == 1);
 952                 }
 953                 *dnew = 0;
 954                 break;
 955
 956         case MASK2(LEFT_FILLING, LEFT_CONTIG):
 957                 /*
 958                  * Filling in the first part of a previous delayed allocation.
 959                  * The left neighbor is contiguous.
 960                  */
 961                 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx - 1,
 962                         XFS_DATA_FORK);
 963                 xfs_bmbt_set_blockcount(ep - 1,
 964                         LEFT.br_blockcount + new->br_blockcount);
 965                 xfs_bmbt_set_startoff(ep,
 966                         PREV.br_startoff + new->br_blockcount);
 967                 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx - 1,
 968                         XFS_DATA_FORK);
 969                 temp = PREV.br_blockcount - new->br_blockcount;
 970                 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx,
 971                         XFS_DATA_FORK);
 972                 xfs_bmbt_set_blockcount(ep, temp);
 973                 ip->i_df.if_lastex = idx - 1;
 974                 if (cur == NULL)
 975                         rval = XFS_ILOG_DEXT;
 976                 else {
 977                         rval = 0;
 978                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
 979                                         LEFT.br_startblock, LEFT.br_blockcount,
 980                                         &i)))
 981                                 goto done;
 982                         ASSERT(i == 1);
 983                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 984                                         LEFT.br_startblock,
 985                                         LEFT.br_blockcount +
 986                                         new->br_blockcount,
 987                                         LEFT.br_state)))
 988                                 goto done;
 989                 }
 990                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 991                         STARTBLOCKVAL(PREV.br_startblock));
 992                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
 993                 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx,
 994                         XFS_DATA_FORK);
 995                 *dnew = temp;
 996                 break;
 997
 998         case MASK(LEFT_FILLING):
 999                 /*
1000                  * Filling in the first part of a previous delayed allocation.
1001                  * The left neighbor is not contiguous.
1002                  */
1003                 xfs_bmap_trace_pre_update(fname, "LF", ip, idx, XFS_DATA_FORK);
1004                 xfs_bmbt_set_startoff(ep, new_endoff);
1005                 temp = PREV.br_blockcount - new->br_blockcount;
1006                 xfs_bmbt_set_blockcount(ep, temp);
1007                 xfs_bmap_trace_insert(fname, "LF", ip, idx, 1, new, NULL,
1008                         XFS_DATA_FORK);
1009                 xfs_bmap_insert_exlist(ip, idx, 1, new, XFS_DATA_FORK);
1010                 ip->i_df.if_lastex = idx;
1011                 ip->i_d.di_nextents++;
1012                 if (cur == NULL)
1013                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1014                 else {
1015                         rval = XFS_ILOG_CORE;
1016                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1017                                         new->br_startblock, new->br_blockcount,
1018                                         &i)))
1019                                 goto done;
1020                         ASSERT(i == 0);
1021                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1022                         if ((error = xfs_bmbt_insert(cur, &i)))
1023                                 goto done;
1024                         ASSERT(i == 1);
1025                 }
1026                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1027                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1028                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1029                                         first, flist, &cur, 1, &tmp_rval,
1030                                         XFS_DATA_FORK);
1031                         rval |= tmp_rval;
1032                         if (error)
1033                                 goto done;
1034                 }
1035                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1036                         STARTBLOCKVAL(PREV.br_startblock) -
1037                         (cur ? cur->bc_private.b.allocated : 0));
1038                 base = ip->i_df.if_u1.if_extents;
1039                 ep = &base[idx + 1];
1040                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1041                 xfs_bmap_trace_post_update(fname, "LF", ip, idx + 1,
1042                         XFS_DATA_FORK);
1043                 *dnew = temp;
1044                 break;
1045
1046         case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1047                 /*
1048                  * Filling in the last part of a previous delayed allocation.
1049                  * The right neighbor is contiguous with the new allocation.
1050                  */
1051                 temp = PREV.br_blockcount - new->br_blockcount;
1052                 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx,
1053                         XFS_DATA_FORK);
1054                 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx + 1,
1055                         XFS_DATA_FORK);
1056                 xfs_bmbt_set_blockcount(ep, temp);
1057                 xfs_bmbt_set_allf(ep + 1, new->br_startoff, new->br_startblock,
1058                         new->br_blockcount + RIGHT.br_blockcount,
1059                         RIGHT.br_state);
1060                 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx + 1,
1061                         XFS_DATA_FORK);
1062                 ip->i_df.if_lastex = idx + 1;
1063                 if (cur == NULL)
1064                         rval = XFS_ILOG_DEXT;
1065                 else {
1066                         rval = 0;
1067                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1068                                         RIGHT.br_startblock,
1069                                         RIGHT.br_blockcount, &i)))
1070                                 goto done;
1071                         ASSERT(i == 1);
1072                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1073                                         new->br_startblock,
1074                                         new->br_blockcount +
1075                                         RIGHT.br_blockcount,
1076                                         RIGHT.br_state)))
1077                                 goto done;
1078                 }
1079                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1080                         STARTBLOCKVAL(PREV.br_startblock));
1081                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1082                 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx,
1083                         XFS_DATA_FORK);
1084                 *dnew = temp;
1085                 break;
1086
1087         case MASK(RIGHT_FILLING):
1088                 /*
1089                  * Filling in the last part of a previous delayed allocation.
1090                  * The right neighbor is not contiguous.
1091                  */
1092                 temp = PREV.br_blockcount - new->br_blockcount;
1093                 xfs_bmap_trace_pre_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1094                 xfs_bmbt_set_blockcount(ep, temp);
1095                 xfs_bmap_trace_insert(fname, "RF", ip, idx + 1, 1,
1096                         new, NULL, XFS_DATA_FORK);
1097                 xfs_bmap_insert_exlist(ip, idx + 1, 1, new, XFS_DATA_FORK);
1098                 ip->i_df.if_lastex = idx + 1;
1099                 ip->i_d.di_nextents++;
1100                 if (cur == NULL)
1101                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1102                 else {
1103                         rval = XFS_ILOG_CORE;
1104                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1105                                         new->br_startblock, new->br_blockcount,
1106                                         &i)))
1107                                 goto done;
1108                         ASSERT(i == 0);
1109                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1110                         if ((error = xfs_bmbt_insert(cur, &i)))
1111                                 goto done;
1112                         ASSERT(i == 1);
1113                 }
1114                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1115                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1116                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1117                                 first, flist, &cur, 1, &tmp_rval,
1118                                 XFS_DATA_FORK);
1119                         rval |= tmp_rval;
1120                         if (error)
1121                                 goto done;
1122                 }
1123                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1124                         STARTBLOCKVAL(PREV.br_startblock) -
1125                         (cur ? cur->bc_private.b.allocated : 0));
1126                 base = ip->i_df.if_u1.if_extents;
1127                 ep = &base[idx];
1128                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1129                 xfs_bmap_trace_post_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1130                 *dnew = temp;
1131                 break;
1132
1133         case 0:
1134                 /*
1135                  * Filling in the middle part of a previous delayed allocation.
1136                  * Contiguity is impossible here.
1137                  * This case is avoided almost all the time.
1138                  */
1139                 temp = new->br_startoff - PREV.br_startoff;
1140                 xfs_bmap_trace_pre_update(fname, "0", ip, idx, XFS_DATA_FORK);
1141                 xfs_bmbt_set_blockcount(ep, temp);
1142                 r[0] = *new;
1143                 r[1].br_startoff = new_endoff;
1144                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
1145                 r[1].br_blockcount = temp2;
1146                 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 2, &r[0], &r[1],
1147                         XFS_DATA_FORK);
1148                 xfs_bmap_insert_exlist(ip, idx + 1, 2, &r[0], XFS_DATA_FORK);
1149                 ip->i_df.if_lastex = idx + 1;
1150                 ip->i_d.di_nextents++;
1151                 if (cur == NULL)
1152                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1153                 else {
1154                         rval = XFS_ILOG_CORE;
1155                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1156                                         new->br_startblock, new->br_blockcount,
1157                                         &i)))
1158                                 goto done;
1159                         ASSERT(i == 0);
1160                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1161                         if ((error = xfs_bmbt_insert(cur, &i)))
1162                                 goto done;
1163                         ASSERT(i == 1);
1164                 }
1165                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1166                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1167                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1168                                         first, flist, &cur, 1, &tmp_rval,
1169                                         XFS_DATA_FORK);
1170                         rval |= tmp_rval;
1171                         if (error)
1172                                 goto done;
1173                 }
1174                 temp = xfs_bmap_worst_indlen(ip, temp);
1175                 temp2 = xfs_bmap_worst_indlen(ip, temp2);
1176                 diff = (int)(temp + temp2 - STARTBLOCKVAL(PREV.br_startblock) -
1177                         (cur ? cur->bc_private.b.allocated : 0));
1178                 if (diff > 0 &&
1179                     xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS, -diff, rsvd)) {
1180                         /*
1181                          * Ick gross gag me with a spoon.
1182                          */
1183                         ASSERT(0);      /* want to see if this ever happens! */
1184                         while (diff > 0) {
1185                                 if (temp) {
1186                                         temp--;
1187                                         diff--;
1188                                         if (!diff ||
1189                                             !xfs_mod_incore_sb(ip->i_mount,
1190                                                     XFS_SBS_FDBLOCKS, -diff, rsvd))
1191                                                 break;
1192                                 }
1193                                 if (temp2) {
1194                                         temp2--;
1195                                         diff--;
1196                                         if (!diff ||
1197                                             !xfs_mod_incore_sb(ip->i_mount,
1198                                                     XFS_SBS_FDBLOCKS, -diff, rsvd))
1199                                                 break;
1200                                 }
1201                         }
1202                 }
1203                 base = ip->i_df.if_u1.if_extents;
1204                 ep = &base[idx];
1205                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1206                 xfs_bmap_trace_post_update(fname, "0", ip, idx, XFS_DATA_FORK);
1207                 xfs_bmap_trace_pre_update(fname, "0", ip, idx + 2,
1208                         XFS_DATA_FORK);
1209                 xfs_bmbt_set_startblock(ep + 2, NULLSTARTBLOCK((int)temp2));
1210                 xfs_bmap_trace_post_update(fname, "0", ip, idx + 2,
1211                         XFS_DATA_FORK);
1212                 *dnew = temp + temp2;
1213                 break;
1214
1215         case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1216         case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1217         case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1218         case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1219         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1220         case MASK(LEFT_CONTIG):
1221         case MASK(RIGHT_CONTIG):
1222                 /*
1223                  * These cases are all impossible.
1224                  */
1225                 ASSERT(0);
1226         }
1227         *curp = cur;
1228 done:
1229         *logflagsp = rval;
1230         return error;
1231 #undef  LEFT
1232 #undef  RIGHT
1233 #undef  PREV
1234 #undef  MASK
1235 #undef  MASK2
1236 #undef  MASK3
1237 #undef  MASK4
1238 #undef  STATE_SET
1239 #undef  STATE_TEST
1240 #undef  STATE_SET_TEST
1241 #undef  SWITCH_STATE
1242 }
1243
1244 /*
1245  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
1246  * allocation to a real allocation or vice versa.
1247  */
1248 STATIC int                              /* error */
1249 xfs_bmap_add_extent_unwritten_real(
1250         xfs_inode_t             *ip,    /* incore inode pointer */
1251         xfs_extnum_t            idx,    /* extent number to update/insert */
1252         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
1253         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
1254         int                     *logflagsp) /* inode logging flags */
1255 {
1256         xfs_bmbt_rec_t          *base;  /* base of extent entry list */
1257         xfs_btree_cur_t         *cur;   /* btree cursor */
1258         xfs_bmbt_rec_t          *ep;    /* extent entry for idx */
1259         int                     error;  /* error return value */
1260 #ifdef XFS_BMAP_TRACE
1261         static char             fname[] = "xfs_bmap_add_extent_unwritten_real";
1262 #endif
1263         int                     i;      /* temp state */
1264         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1265         xfs_exntst_t            newext; /* new extent state */
1266         xfs_exntst_t            oldext; /* old extent state */
1267         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1268                                         /* left is 0, right is 1, prev is 2 */
1269         int                     rval=0; /* return value (logging flags) */
1270         int                     state = 0;/* state bits, accessed thru macros */
1271         enum {                          /* bit number definitions for state */
1272                 LEFT_CONTIG,    RIGHT_CONTIG,
1273                 LEFT_FILLING,   RIGHT_FILLING,
1274                 LEFT_DELAY,     RIGHT_DELAY,
1275                 LEFT_VALID,     RIGHT_VALID
1276         };
1277
1278 #define LEFT            r[0]
1279 #define RIGHT           r[1]
1280 #define PREV            r[2]
1281 #define MASK(b)         (1 << (b))
1282 #define MASK2(a,b)      (MASK(a) | MASK(b))
1283 #define MASK3(a,b,c)    (MASK2(a,b) | MASK(c))
1284 #define MASK4(a,b,c,d)  (MASK3(a,b,c) | MASK(d))
1285 #define STATE_SET(b,v)  ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1286 #define STATE_TEST(b)   (state & MASK(b))
1287 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
1288                                        ((state &= ~MASK(b)), 0))
1289 #define SWITCH_STATE            \
1290         (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
1291
1292         /*
1293          * Set up a bunch of variables to make the tests simpler.
1294          */
1295         error = 0;
1296         cur = *curp;
1297         base = ip->i_df.if_u1.if_extents;
1298         ep = &base[idx];
1299         xfs_bmbt_get_all(ep, &PREV);
1300         newext = new->br_state;
1301         oldext = (newext == XFS_EXT_UNWRITTEN) ?
1302                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
1303         ASSERT(PREV.br_state == oldext);
1304         new_endoff = new->br_startoff + new->br_blockcount;
1305         ASSERT(PREV.br_startoff <= new->br_startoff);
1306         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1307         /*
1308          * Set flags determining what part of the previous oldext allocation
1309          * extent is being replaced by a newext allocation.
1310          */
1311         STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
1312         STATE_SET(RIGHT_FILLING,
1313                 PREV.br_startoff + PREV.br_blockcount == new_endoff);
1314         /*
1315          * Check and set flags if this segment has a left neighbor.
1316          * Don't set contiguous if the combined extent would be too large.
1317          */
1318         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1319                 xfs_bmbt_get_all(ep - 1, &LEFT);
1320                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
1321         }
1322         STATE_SET(LEFT_CONTIG,
1323                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
1324                 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1325                 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1326                 LEFT.br_state == newext &&
1327                 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1328         /*
1329          * Check and set flags if this segment has a right neighbor.
1330          * Don't set contiguous if the combined extent would be too large.
1331          * Also check for all-three-contiguous being too large.
1332          */
1333         if (STATE_SET_TEST(RIGHT_VALID,
1334                         idx <
1335                         ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
1336                 xfs_bmbt_get_all(ep + 1, &RIGHT);
1337                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
1338         }
1339         STATE_SET(RIGHT_CONTIG,
1340                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
1341                 new_endoff == RIGHT.br_startoff &&
1342                 new->br_startblock + new->br_blockcount ==
1343                     RIGHT.br_startblock &&
1344                 newext == RIGHT.br_state &&
1345                 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1346                 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
1347                   MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
1348                  LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1349                      <= MAXEXTLEN));
1350         /*
1351          * Switch out based on the FILLING and CONTIG state bits.
1352          */
1353         switch (SWITCH_STATE) {
1354
1355         case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1356                 /*
1357                  * Setting all of a previous oldext extent to newext.
1358                  * The left and right neighbors are both contiguous with new.
1359                  */
1360                 xfs_bmap_trace_pre_update(fname, "LF|RF|LC|RC", ip, idx - 1,
1361                         XFS_DATA_FORK);
1362                 xfs_bmbt_set_blockcount(ep - 1,
1363                         LEFT.br_blockcount + PREV.br_blockcount +
1364                         RIGHT.br_blockcount);
1365                 xfs_bmap_trace_post_update(fname, "LF|RF|LC|RC", ip, idx - 1,
1366                         XFS_DATA_FORK);
1367                 xfs_bmap_trace_delete(fname, "LF|RF|LC|RC", ip, idx, 2,
1368                         XFS_DATA_FORK);
1369                 xfs_bmap_delete_exlist(ip, idx, 2, XFS_DATA_FORK);
1370                 ip->i_df.if_lastex = idx - 1;
1371                 ip->i_d.di_nextents -= 2;
1372                 if (cur == NULL)
1373                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1374                 else {
1375                         rval = XFS_ILOG_CORE;
1376                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1377                                         RIGHT.br_startblock,
1378                                         RIGHT.br_blockcount, &i)))
1379                                 goto done;
1380                         ASSERT(i == 1);
1381                         if ((error = xfs_bmbt_delete(cur, &i)))
1382                                 goto done;
1383                         ASSERT(i == 1);
1384                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1385                                 goto done;
1386                         ASSERT(i == 1);
1387                         if ((error = xfs_bmbt_delete(cur, &i)))
1388                                 goto done;
1389                         ASSERT(i == 1);
1390                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1391                                 goto done;
1392                         ASSERT(i == 1);
1393                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1394                                 LEFT.br_startblock,
1395                                 LEFT.br_blockcount + PREV.br_blockcount +
1396                                 RIGHT.br_blockcount, LEFT.br_state)))
1397                                 goto done;
1398                 }
1399                 break;
1400
1401         case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
1402                 /*
1403                  * Setting all of a previous oldext extent to newext.
1404                  * The left neighbor is contiguous, the right is not.
1405                  */
1406                 xfs_bmap_trace_pre_update(fname, "LF|RF|LC", ip, idx - 1,
1407                         XFS_DATA_FORK);
1408                 xfs_bmbt_set_blockcount(ep - 1,
1409                         LEFT.br_blockcount + PREV.br_blockcount);
1410                 xfs_bmap_trace_post_update(fname, "LF|RF|LC", ip, idx - 1,
1411                         XFS_DATA_FORK);
1412                 ip->i_df.if_lastex = idx - 1;
1413                 xfs_bmap_trace_delete(fname, "LF|RF|LC", ip, idx, 1,
1414                         XFS_DATA_FORK);
1415                 xfs_bmap_delete_exlist(ip, idx, 1, XFS_DATA_FORK);
1416                 ip->i_d.di_nextents--;
1417                 if (cur == NULL)
1418                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1419                 else {
1420                         rval = XFS_ILOG_CORE;
1421                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1422                                         PREV.br_startblock, PREV.br_blockcount,
1423                                         &i)))
1424                                 goto done;
1425                         ASSERT(i == 1);
1426                         if ((error = xfs_bmbt_delete(cur, &i)))
1427                                 goto done;
1428                         ASSERT(i == 1);
1429                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1430                                 goto done;
1431                         ASSERT(i == 1);
1432                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1433                                 LEFT.br_startblock,
1434                                 LEFT.br_blockcount + PREV.br_blockcount,
1435                                 LEFT.br_state)))
1436                                 goto done;
1437                 }
1438                 break;
1439
1440         case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
1441                 /*
1442                  * Setting all of a previous oldext extent to newext.
1443                  * The right neighbor is contiguous, the left is not.
1444                  */
1445                 xfs_bmap_trace_pre_update(fname, "LF|RF|RC", ip, idx,
1446                         XFS_DATA_FORK);
1447                 xfs_bmbt_set_blockcount(ep,
1448                         PREV.br_blockcount + RIGHT.br_blockcount);
1449                 xfs_bmbt_set_state(ep, newext);
1450                 xfs_bmap_trace_post_update(fname, "LF|RF|RC", ip, idx,
1451                         XFS_DATA_FORK);
1452                 ip->i_df.if_lastex = idx;
1453                 xfs_bmap_trace_delete(fname, "LF|RF|RC", ip, idx + 1, 1,
1454                         XFS_DATA_FORK);
1455                 xfs_bmap_delete_exlist(ip, idx + 1, 1, XFS_DATA_FORK);
1456                 ip->i_d.di_nextents--;
1457                 if (cur == NULL)
1458                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1459                 else {
1460                         rval = XFS_ILOG_CORE;
1461                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1462                                         RIGHT.br_startblock,
1463                                         RIGHT.br_blockcount, &i)))
1464                                 goto done;
1465                         ASSERT(i == 1);
1466                         if ((error = xfs_bmbt_delete(cur, &i)))
1467                                 goto done;
1468                         ASSERT(i == 1);
1469                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1470                                 goto done;
1471                         ASSERT(i == 1);
1472                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1473                                 new->br_startblock,
1474                                 new->br_blockcount + RIGHT.br_blockcount,
1475                                 newext)))
1476                                 goto done;
1477                 }
1478                 break;
1479
1480         case MASK2(LEFT_FILLING, RIGHT_FILLING):
1481                 /*
1482                  * Setting all of a previous oldext extent to newext.
1483                  * Neither the left nor right neighbors are contiguous with
1484                  * the new one.
1485                  */
1486                 xfs_bmap_trace_pre_update(fname, "LF|RF", ip, idx,
1487                         XFS_DATA_FORK);
1488                 xfs_bmbt_set_state(ep, newext);
1489                 xfs_bmap_trace_post_update(fname, "LF|RF", ip, idx,
1490                         XFS_DATA_FORK);
1491                 ip->i_df.if_lastex = idx;
1492                 if (cur == NULL)
1493                         rval = XFS_ILOG_DEXT;
1494                 else {
1495                         rval = 0;
1496                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1497                                         new->br_startblock, new->br_blockcount,
1498                                         &i)))
1499                                 goto done;
1500                         ASSERT(i == 1);
1501                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1502                                 new->br_startblock, new->br_blockcount,
1503                                 newext)))
1504                                 goto done;
1505                 }
1506                 break;
1507
1508         case MASK2(LEFT_FILLING, LEFT_CONTIG):
1509                 /*
1510                  * Setting the first part of a previous oldext extent to newext.
1511                  * The left neighbor is contiguous.
1512                  */
1513                 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx - 1,
1514                         XFS_DATA_FORK);
1515                 xfs_bmbt_set_blockcount(ep - 1,
1516                         LEFT.br_blockcount + new->br_blockcount);
1517                 xfs_bmbt_set_startoff(ep,
1518                         PREV.br_startoff + new->br_blockcount);
1519                 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx - 1,
1520                         XFS_DATA_FORK);
1521                 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx,
1522                         XFS_DATA_FORK);
1523                 xfs_bmbt_set_startblock(ep,
1524                         new->br_startblock + new->br_blockcount);
1525                 xfs_bmbt_set_blockcount(ep,
1526                         PREV.br_blockcount - new->br_blockcount);
1527                 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx,
1528                         XFS_DATA_FORK);
1529                 ip->i_df.if_lastex = idx - 1;
1530                 if (cur == NULL)
1531                         rval = XFS_ILOG_DEXT;
1532                 else {
1533                         rval = 0;
1534                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1535                                         PREV.br_startblock, PREV.br_blockcount,
1536                                         &i)))
1537                                 goto done;
1538                         ASSERT(i == 1);
1539                         if ((error = xfs_bmbt_update(cur,
1540                                 PREV.br_startoff + new->br_blockcount,
1541                                 PREV.br_startblock + new->br_blockcount,
1542                                 PREV.br_blockcount - new->br_blockcount,
1543                                 oldext)))
1544                                 goto done;
1545                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1546                                 goto done;
1547                         if (xfs_bmbt_update(cur, LEFT.br_startoff,
1548                                 LEFT.br_startblock,
1549                                 LEFT.br_blockcount + new->br_blockcount,
1550                                 LEFT.br_state))
1551                                 goto done;
1552                 }
1553                 break;
1554
1555         case MASK(LEFT_FILLING):
1556                 /*
1557                  * Setting the first part of a previous oldext extent to newext.
1558                  * The left neighbor is not contiguous.
1559                  */
1560                 xfs_bmap_trace_pre_update(fname, "LF", ip, idx, XFS_DATA_FORK);
1561                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1562                 xfs_bmbt_set_startoff(ep, new_endoff);
1563                 xfs_bmbt_set_blockcount(ep,
1564                         PREV.br_blockcount - new->br_blockcount);
1565                 xfs_bmbt_set_startblock(ep,
1566                         new->br_startblock + new->br_blockcount);
1567                 xfs_bmap_trace_post_update(fname, "LF", ip, idx, XFS_DATA_FORK);
1568                 xfs_bmap_trace_insert(fname, "LF", ip, idx, 1, new, NULL,
1569                         XFS_DATA_FORK);
1570                 xfs_bmap_insert_exlist(ip, idx, 1, new, XFS_DATA_FORK);
1571                 ip->i_df.if_lastex = idx;
1572                 ip->i_d.di_nextents++;
1573                 if (cur == NULL)
1574                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1575                 else {
1576                         rval = XFS_ILOG_CORE;
1577                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1578                                         PREV.br_startblock, PREV.br_blockcount,
1579                                         &i)))
1580                                 goto done;
1581                         ASSERT(i == 1);
1582                         if ((error = xfs_bmbt_update(cur,
1583                                 PREV.br_startoff + new->br_blockcount,
1584                                 PREV.br_startblock + new->br_blockcount,
1585                                 PREV.br_blockcount - new->br_blockcount,
1586                                 oldext)))
1587                                 goto done;
1588                         cur->bc_rec.b = *new;
1589                         if ((error = xfs_bmbt_insert(cur, &i)))
1590                                 goto done;
1591                         ASSERT(i == 1);
1592                 }
1593                 break;
1594
1595         case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1596                 /*
1597                  * Setting the last part of a previous oldext extent to newext.
1598                  * The right neighbor is contiguous with the new allocation.
1599                  */
1600                 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx,
1601                         XFS_DATA_FORK);
1602                 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx + 1,
1603                         XFS_DATA_FORK);
1604                 xfs_bmbt_set_blockcount(ep,
1605                         PREV.br_blockcount - new->br_blockcount);
1606                 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx,
1607                         XFS_DATA_FORK);
1608                 xfs_bmbt_set_allf(ep + 1, new->br_startoff, new->br_startblock,
1609                         new->br_blockcount + RIGHT.br_blockcount, newext);
1610                 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx + 1,
1611                         XFS_DATA_FORK);
1612                 ip->i_df.if_lastex = idx + 1;
1613                 if (cur == NULL)
1614                         rval = XFS_ILOG_DEXT;
1615                 else {
1616                         rval = 0;
1617                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1618                                         PREV.br_startblock,
1619                                         PREV.br_blockcount, &i)))
1620                                 goto done;
1621                         ASSERT(i == 1);
1622                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1623                                 PREV.br_startblock,
1624                                 PREV.br_blockcount - new->br_blockcount,
1625                                 oldext)))
1626                                 goto done;
1627                         if ((error = xfs_bmbt_increment(cur, 0, &i)))
1628                                 goto done;
1629                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1630                                 new->br_startblock,
1631                                 new->br_blockcount + RIGHT.br_blockcount,
1632                                 newext)))
1633                                 goto done;
1634                 }
1635                 break;
1636
1637         case MASK(RIGHT_FILLING):
1638                 /*
1639                  * Setting the last part of a previous oldext extent to newext.
1640                  * The right neighbor is not contiguous.
1641                  */
1642                 xfs_bmap_trace_pre_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1643                 xfs_bmbt_set_blockcount(ep,
1644                         PREV.br_blockcount - new->br_blockcount);
1645                 xfs_bmap_trace_post_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1646                 xfs_bmap_trace_insert(fname, "RF", ip, idx + 1, 1,
1647                         new, NULL, XFS_DATA_FORK);
1648                 xfs_bmap_insert_exlist(ip, idx + 1, 1, new, XFS_DATA_FORK);
1649                 ip->i_df.if_lastex = idx + 1;
1650                 ip->i_d.di_nextents++;
1651                 if (cur == NULL)
1652                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1653                 else {
1654                         rval = XFS_ILOG_CORE;
1655                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1656                                         PREV.br_startblock, PREV.br_blockcount,
1657                                         &i)))
1658                                 goto done;
1659                         ASSERT(i == 1);
1660                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1661                                 PREV.br_startblock,
1662                                 PREV.br_blockcount - new->br_blockcount,
1663                                 oldext)))
1664                                 goto done;
1665                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1666                                         new->br_startblock, new->br_blockcount,
1667                                         &i)))
1668                                 goto done;
1669                         ASSERT(i == 0);
1670                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1671                         if ((error = xfs_bmbt_insert(cur, &i)))
1672                                 goto done;
1673                         ASSERT(i == 1);
1674                 }
1675                 break;
1676
1677         case 0:
1678                 /*
1679                  * Setting the middle part of a previous oldext extent to
1680                  * newext.  Contiguity is impossible here.
1681                  * One extent becomes three extents.
1682                  */
1683                 xfs_bmap_trace_pre_update(fname, "0", ip, idx, XFS_DATA_FORK);
1684                 xfs_bmbt_set_blockcount(ep,
1685                         new->br_startoff - PREV.br_startoff);
1686                 xfs_bmap_trace_post_update(fname, "0", ip, idx, XFS_DATA_FORK);
1687                 r[0] = *new;
1688                 r[1].br_startoff = new_endoff;
1689                 r[1].br_blockcount =
1690                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
1691                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1692                 r[1].br_state = oldext;
1693                 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 2, &r[0], &r[1],
1694                         XFS_DATA_FORK);
1695                 xfs_bmap_insert_exlist(ip, idx + 1, 2, &r[0], XFS_DATA_FORK);
1696                 ip->i_df.if_lastex = idx + 1;
1697                 ip->i_d.di_nextents += 2;
1698                 if (cur == NULL)
1699                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1700                 else {
1701                         rval = XFS_ILOG_CORE;
1702                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1703                                         PREV.br_startblock, PREV.br_blockcount,
1704                                         &i)))
1705                                 goto done;
1706                         ASSERT(i == 1);
1707                         /* new right extent - oldext */
1708                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1709                                 r[1].br_startblock, r[1].br_blockcount,
1710                                 r[1].br_state)))
1711                                 goto done;
1712                         /* new left extent - oldext */
1713                         PREV.br_blockcount =
1714                                 new->br_startoff - PREV.br_startoff;
1715                         cur->bc_rec.b = PREV;
1716                         if ((error = xfs_bmbt_insert(cur, &i)))
1717                                 goto done;
1718                         ASSERT(i == 1);
1719                         if ((error = xfs_bmbt_increment(cur, 0, &i)))
1720                                 goto done;
1721                         ASSERT(i == 1);
1722                         /* new middle extent - newext */
1723                         cur->bc_rec.b = *new;
1724                         if ((error = xfs_bmbt_insert(cur, &i)))
1725                                 goto done;
1726                         ASSERT(i == 1);
1727                 }
1728                 break;
1729
1730         case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1731         case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1732         case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1733         case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1734         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1735         case MASK(LEFT_CONTIG):
1736         case MASK(RIGHT_CONTIG):
1737                 /*
1738                  * These cases are all impossible.
1739                  */
1740                 ASSERT(0);
1741         }
1742         *curp = cur;
1743 done:
1744         *logflagsp = rval;
1745         return error;
1746 #undef  LEFT
1747 #undef  RIGHT
1748 #undef  PREV
1749 #undef  MASK
1750 #undef  MASK2
1751 #undef  MASK3
1752 #undef  MASK4
1753 #undef  STATE_SET
1754 #undef  STATE_TEST
1755 #undef  STATE_SET_TEST
1756 #undef  SWITCH_STATE
1757 }
1758
1759 /*
1760  * Called by xfs_bmap_add_extent to handle cases converting a hole
1761  * to a delayed allocation.
1762  */
1763 /*ARGSUSED*/
1764 STATIC int                              /* error */
1765 xfs_bmap_add_extent_hole_delay(
1766         xfs_inode_t             *ip,    /* incore inode pointer */
1767         xfs_extnum_t            idx,    /* extent number to update/insert */
1768         xfs_btree_cur_t         *cur,   /* if null, not a btree */
1769         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
1770         int                     *logflagsp, /* inode logging flags */
1771         int                     rsvd)           /* OK to allocate reserved blocks */
1772 {
1773         xfs_bmbt_rec_t          *base;  /* base of extent entry list */
1774         xfs_bmbt_rec_t          *ep;    /* extent list entry for idx */
1775 #ifdef XFS_BMAP_TRACE
1776         static char             fname[] = "xfs_bmap_add_extent_hole_delay";
1777 #endif
1778         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1779         xfs_filblks_t           newlen=0;       /* new indirect size */
1780         xfs_filblks_t           oldlen=0;       /* old indirect size */
1781         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1782         int                     state;  /* state bits, accessed thru macros */
1783         xfs_filblks_t           temp;   /* temp for indirect calculations */
1784         enum {                          /* bit number definitions for state */
1785                 LEFT_CONTIG,    RIGHT_CONTIG,
1786                 LEFT_DELAY,     RIGHT_DELAY,
1787                 LEFT_VALID,     RIGHT_VALID
1788         };
1789
1790 #define MASK(b)                 (1 << (b))
1791 #define MASK2(a,b)              (MASK(a) | MASK(b))
1792 #define STATE_SET(b,v)          ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1793 #define STATE_TEST(b)           (state & MASK(b))
1794 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
1795                                        ((state &= ~MASK(b)), 0))
1796 #define SWITCH_STATE            (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
1797
1798         base = ip->i_df.if_u1.if_extents;
1799         ep = &base[idx];
1800         state = 0;
1801         ASSERT(ISNULLSTARTBLOCK(new->br_startblock));
1802         /*
1803          * Check and set flags if this segment has a left neighbor
1804          */
1805         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1806                 xfs_bmbt_get_all(ep - 1, &left);
1807                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
1808         }
1809         /*
1810          * Check and set flags if the current (right) segment exists.
1811          * If it doesn't exist, we're converting the hole at end-of-file.
1812          */
1813         if (STATE_SET_TEST(RIGHT_VALID,
1814                            idx <
1815                            ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1816                 xfs_bmbt_get_all(ep, &right);
1817                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
1818         }
1819         /*
1820          * Set contiguity flags on the left and right neighbors.
1821          * Don't let extents get too large, even if the pieces are contiguous.
1822          */
1823         STATE_SET(LEFT_CONTIG,
1824                 STATE_TEST(LEFT_VALID) && STATE_TEST(LEFT_DELAY) &&
1825                 left.br_startoff + left.br_blockcount == new->br_startoff &&
1826                 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1827         STATE_SET(RIGHT_CONTIG,
1828                 STATE_TEST(RIGHT_VALID) && STATE_TEST(RIGHT_DELAY) &&
1829                 new->br_startoff + new->br_blockcount == right.br_startoff &&
1830                 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1831                 (!STATE_TEST(LEFT_CONTIG) ||
1832                  (left.br_blockcount + new->br_blockcount +
1833                      right.br_blockcount <= MAXEXTLEN)));
1834         /*
1835          * Switch out based on the contiguity flags.
1836          */
1837         switch (SWITCH_STATE) {
1838
1839         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1840                 /*
1841                  * New allocation is contiguous with delayed allocations
1842                  * on the left and on the right.
1843                  * Merge all three into a single extent list entry.
1844                  */
1845                 temp = left.br_blockcount + new->br_blockcount +
1846                         right.br_blockcount;
1847                 xfs_bmap_trace_pre_update(fname, "LC|RC", ip, idx - 1,
1848                         XFS_DATA_FORK);
1849                 xfs_bmbt_set_blockcount(ep - 1, temp);
1850                 oldlen = STARTBLOCKVAL(left.br_startblock) +
1851                         STARTBLOCKVAL(new->br_startblock) +
1852                         STARTBLOCKVAL(right.br_startblock);
1853                 newlen = xfs_bmap_worst_indlen(ip, temp);
1854                 xfs_bmbt_set_startblock(ep - 1, NULLSTARTBLOCK((int)newlen));
1855                 xfs_bmap_trace_post_update(fname, "LC|RC", ip, idx - 1,
1856                         XFS_DATA_FORK);
1857                 xfs_bmap_trace_delete(fname, "LC|RC", ip, idx, 1,
1858                         XFS_DATA_FORK);
1859                 xfs_bmap_delete_exlist(ip, idx, 1, XFS_DATA_FORK);
1860                 ip->i_df.if_lastex = idx - 1;
1861                 break;
1862
1863         case MASK(LEFT_CONTIG):
1864                 /*
1865                  * New allocation is contiguous with a delayed allocation
1866                  * on the left.
1867                  * Merge the new allocation with the left neighbor.
1868                  */
1869                 temp = left.br_blockcount + new->br_blockcount;
1870                 xfs_bmap_trace_pre_update(fname, "LC", ip, idx - 1,
1871                         XFS_DATA_FORK);
1872                 xfs_bmbt_set_blockcount(ep - 1, temp);
1873                 oldlen = STARTBLOCKVAL(left.br_startblock) +
1874                         STARTBLOCKVAL(new->br_startblock);
1875                 newlen = xfs_bmap_worst_indlen(ip, temp);
1876                 xfs_bmbt_set_startblock(ep - 1, NULLSTARTBLOCK((int)newlen));
1877                 xfs_bmap_trace_post_update(fname, "LC", ip, idx - 1,
1878                         XFS_DATA_FORK);
1879                 ip->i_df.if_lastex = idx - 1;
1880                 break;
1881
1882         case MASK(RIGHT_CONTIG):
1883                 /*
1884                  * New allocation is contiguous with a delayed allocation
1885                  * on the right.
1886                  * Merge the new allocation with the right neighbor.
1887                  */
1888                 xfs_bmap_trace_pre_update(fname, "RC", ip, idx, XFS_DATA_FORK);
1889                 temp = new->br_blockcount + right.br_blockcount;
1890                 oldlen = STARTBLOCKVAL(new->br_startblock) +
1891                         STARTBLOCKVAL(right.br_startblock);
1892                 newlen = xfs_bmap_worst_indlen(ip, temp);
1893                 xfs_bmbt_set_allf(ep, new->br_startoff,
1894                         NULLSTARTBLOCK((int)newlen), temp, right.br_state);
1895                 xfs_bmap_trace_post_update(fname, "RC", ip, idx, XFS_DATA_FORK);
1896                 ip->i_df.if_lastex = idx;
1897                 break;
1898
1899         case 0:
1900                 /*
1901                  * New allocation is not contiguous with another
1902                  * delayed allocation.
1903                  * Insert a new entry.
1904                  */
1905                 oldlen = newlen = 0;
1906                 xfs_bmap_trace_insert(fname, "0", ip, idx, 1, new, NULL,
1907                         XFS_DATA_FORK);
1908                 xfs_bmap_insert_exlist(ip, idx, 1, new, XFS_DATA_FORK);
1909                 ip->i_df.if_lastex = idx;
1910                 break;
1911         }
1912         if (oldlen != newlen) {
1913                 ASSERT(oldlen > newlen);
1914                 xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
1915                         (int)(oldlen - newlen), rsvd);
1916                 /*
1917                  * Nothing to do for disk quota accounting here.
1918                  */
1919         }
1920         *logflagsp = 0;
1921         return 0;
1922 #undef  MASK
1923 #undef  MASK2
1924 #undef  STATE_SET
1925 #undef  STATE_TEST
1926 #undef  STATE_SET_TEST
1927 #undef  SWITCH_STATE
1928 }
1929
1930 /*
1931  * Called by xfs_bmap_add_extent to handle cases converting a hole
1932  * to a real allocation.
1933  */
1934 STATIC int                              /* error */
1935 xfs_bmap_add_extent_hole_real(
1936         xfs_inode_t             *ip,    /* incore inode pointer */
1937         xfs_extnum_t            idx,    /* extent number to update/insert */
1938         xfs_btree_cur_t         *cur,   /* if null, not a btree */
1939         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
1940         int                     *logflagsp, /* inode logging flags */
1941         int                     whichfork) /* data or attr fork */
1942 {
1943         xfs_bmbt_rec_t          *ep;    /* pointer to extent entry ins. point */
1944         int                     error;  /* error return value */
1945 #ifdef XFS_BMAP_TRACE
1946         static char             fname[] = "xfs_bmap_add_extent_hole_real";
1947 #endif
1948         int                     i;      /* temp state */
1949         xfs_ifork_t             *ifp;   /* inode fork pointer */
1950         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1951         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1952         int                     state;  /* state bits, accessed thru macros */
1953         enum {                          /* bit number definitions for state */
1954                 LEFT_CONTIG,    RIGHT_CONTIG,
1955                 LEFT_DELAY,     RIGHT_DELAY,
1956                 LEFT_VALID,     RIGHT_VALID
1957         };
1958
1959 #define MASK(b)                 (1 << (b))
1960 #define MASK2(a,b)              (MASK(a) | MASK(b))
1961 #define STATE_SET(b,v)          ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1962 #define STATE_TEST(b)           (state & MASK(b))
1963 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
1964                                        ((state &= ~MASK(b)), 0))
1965 #define SWITCH_STATE            (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
1966
1967         ifp = XFS_IFORK_PTR(ip, whichfork);
1968         ASSERT(idx <= ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
1969         ep = &ifp->if_u1.if_extents[idx];
1970         state = 0;
1971         /*
1972          * Check and set flags if this segment has a left neighbor.
1973          */
1974         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1975                 xfs_bmbt_get_all(ep - 1, &left);
1976                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
1977         }
1978         /*
1979          * Check and set flags if this segment has a current value.
1980          * Not true if we're inserting into the "hole" at eof.
1981          */
1982         if (STATE_SET_TEST(RIGHT_VALID,
1983                            idx <
1984                            ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1985                 xfs_bmbt_get_all(ep, &right);
1986                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
1987         }
1988         /*
1989          * We're inserting a real allocation between "left" and "right".
1990          * Set the contiguity flags.  Don't let extents get too large.
1991          */
1992         STATE_SET(LEFT_CONTIG,
1993                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
1994                 left.br_startoff + left.br_blockcount == new->br_startoff &&
1995                 left.br_startblock + left.br_blockcount == new->br_startblock &&
1996                 left.br_state == new->br_state &&
1997                 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1998         STATE_SET(RIGHT_CONTIG,
1999                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
2000                 new->br_startoff + new->br_blockcount == right.br_startoff &&
2001                 new->br_startblock + new->br_blockcount ==
2002                     right.br_startblock &&
2003                 new->br_state == right.br_state &&
2004                 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2005                 (!STATE_TEST(LEFT_CONTIG) ||
2006                  left.br_blockcount + new->br_blockcount +
2007                      right.br_blockcount <= MAXEXTLEN));
2008
2009         /*
2010          * Select which case we're in here, and implement it.
2011          */
2012         switch (SWITCH_STATE) {
2013
2014         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
2015                 /*
2016                  * New allocation is contiguous with real allocations on the
2017                  * left and on the right.
2018                  * Merge all three into a single extent list entry.
2019                  */
2020                 xfs_bmap_trace_pre_update(fname, "LC|RC", ip, idx - 1,
2021                         whichfork);
2022                 xfs_bmbt_set_blockcount(ep - 1,
2023                         left.br_blockcount + new->br_blockcount +
2024                         right.br_blockcount);
2025                 xfs_bmap_trace_post_update(fname, "LC|RC", ip, idx - 1,
2026                         whichfork);
2027                 xfs_bmap_trace_delete(fname, "LC|RC", ip,
2028                         idx, 1, whichfork);
2029                 xfs_bmap_delete_exlist(ip, idx, 1, whichfork);
2030                 ifp->if_lastex = idx - 1;
2031                 XFS_IFORK_NEXT_SET(ip, whichfork,
2032                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2033                 if (cur == NULL) {
2034                         *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2035                         return 0;
2036                 }
2037                 *logflagsp = XFS_ILOG_CORE;
2038                 if ((error = xfs_bmbt_lookup_eq(cur, right.br_startoff,
2039                                 right.br_startblock, right.br_blockcount, &i)))
2040                         return error;
2041                 ASSERT(i == 1);
2042                 if ((error = xfs_bmbt_delete(cur, &i)))
2043                         return error;
2044                 ASSERT(i == 1);
2045                 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
2046                         return error;
2047                 ASSERT(i == 1);
2048                 error = xfs_bmbt_update(cur, left.br_startoff,
2049                                 left.br_startblock,
2050                                 left.br_blockcount + new->br_blockcount +
2051                                 right.br_blockcount, left.br_state);
2052                 return error;
2053
2054         case MASK(LEFT_CONTIG):
2055                 /*
2056                  * New allocation is contiguous with a real allocation
2057                  * on the left.
2058                  * Merge the new allocation with the left neighbor.
2059                  */
2060                 xfs_bmap_trace_pre_update(fname, "LC", ip, idx - 1, whichfork);
2061                 xfs_bmbt_set_blockcount(ep - 1,
2062                         left.br_blockcount + new->br_blockcount);
2063                 xfs_bmap_trace_post_update(fname, "LC", ip, idx - 1, whichfork);
2064                 ifp->if_lastex = idx - 1;
2065                 if (cur == NULL) {
2066                         *logflagsp = XFS_ILOG_FEXT(whichfork);
2067                         return 0;
2068                 }
2069                 *logflagsp = 0;
2070                 if ((error = xfs_bmbt_lookup_eq(cur, left.br_startoff,
2071                                 left.br_startblock, left.br_blockcount, &i)))
2072                         return error;
2073                 ASSERT(i == 1);
2074                 error = xfs_bmbt_update(cur, left.br_startoff,
2075                                 left.br_startblock,
2076                                 left.br_blockcount + new->br_blockcount,
2077                                 left.br_state);
2078                 return error;
2079
2080         case MASK(RIGHT_CONTIG):
2081                 /*
2082                  * New allocation is contiguous with a real allocation
2083                  * on the right.
2084                  * Merge the new allocation with the right neighbor.
2085                  */
2086                 xfs_bmap_trace_pre_update(fname, "RC", ip, idx, whichfork);
2087                 xfs_bmbt_set_allf(ep, new->br_startoff, new->br_startblock,
2088                         new->br_blockcount + right.br_blockcount,
2089                         right.br_state);
2090                 xfs_bmap_trace_post_update(fname, "RC", ip, idx, whichfork);
2091                 ifp->if_lastex = idx;
2092                 if (cur == NULL) {
2093                         *logflagsp = XFS_ILOG_FEXT(whichfork);
2094                         return 0;
2095                 }
2096                 *logflagsp = 0;
2097                 if ((error = xfs_bmbt_lookup_eq(cur, right.br_startoff,
2098                                 right.br_startblock, right.br_blockcount, &i)))
2099                         return error;
2100                 ASSERT(i == 1);
2101                 error = xfs_bmbt_update(cur, new->br_startoff,
2102                                 new->br_startblock,
2103                                 new->br_blockcount + right.br_blockcount,
2104                                 right.br_state);
2105                 return error;
2106
2107         case 0:
2108                 /*
2109                  * New allocation is not contiguous with another
2110                  * real allocation.
2111                  * Insert a new entry.
2112                  */
2113                 xfs_bmap_trace_insert(fname, "0", ip, idx, 1, new, NULL,
2114                         whichfork);
2115                 xfs_bmap_insert_exlist(ip, idx, 1, new, whichfork);
2116                 ifp->if_lastex = idx;
2117                 XFS_IFORK_NEXT_SET(ip, whichfork,
2118                         XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2119                 if (cur == NULL) {
2120                         *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2121                         return 0;
2122                 }
2123                 *logflagsp = XFS_ILOG_CORE;
2124                 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2125                                 new->br_startblock, new->br_blockcount, &i)))
2126                         return error;
2127                 ASSERT(i == 0);
2128                 cur->bc_rec.b.br_state = new->br_state;
2129                 if ((error = xfs_bmbt_insert(cur, &i)))
2130                         return error;
2131                 ASSERT(i == 1);
2132                 return 0;
2133         }
2134 #undef  MASK
2135 #undef  MASK2
2136 #undef  STATE_SET
2137 #undef  STATE_TEST
2138 #undef  STATE_SET_TEST
2139 #undef  SWITCH_STATE
2140         /* NOTREACHED */
2141         ASSERT(0);
2142         return 0; /* keep gcc quite */
2143 }
2144
2145 #define XFS_ALLOC_GAP_UNITS     4
2146
2147 /*
2148  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2149  * It figures out where to ask the underlying allocator to put the new extent.
2150  */
2151 STATIC int                              /* error */
2152 xfs_bmap_alloc(
2153         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2154 {
2155         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
2156         xfs_alloctype_t atype=0;        /* type for allocation routines */
2157         int             error;          /* error return value */
2158         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2159         xfs_mount_t     *mp;            /* mount point structure */
2160         int             nullfb;         /* true if ap->firstblock isn't set */
2161         int             rt;             /* true if inode is realtime */
2162 #ifdef __KERNEL__
2163         xfs_extlen_t    prod=0;         /* product factor for allocators */
2164         xfs_extlen_t    ralen=0;        /* realtime allocation length */
2165 #endif
2166
2167 #define ISVALID(x,y)    \
2168         (rt ? \
2169                 (x) < mp->m_sb.sb_rblocks : \
2170                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2171                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2172                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2173
2174         /*
2175          * Set up variables.
2176          */
2177         mp = ap->ip->i_mount;
2178         nullfb = ap->firstblock == NULLFSBLOCK;
2179         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2180         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2181 #ifdef __KERNEL__
2182         if (rt) {
2183                 xfs_extlen_t    extsz;          /* file extent size for rt */
2184                 xfs_fileoff_t   nexto;          /* next file offset */
2185                 xfs_extlen_t    orig_alen;      /* original ap->alen */
2186                 xfs_fileoff_t   orig_end;       /* original off+len */
2187                 xfs_fileoff_t   orig_off;       /* original ap->off */
2188                 xfs_extlen_t    mod_off;        /* modulus calculations */
2189                 xfs_fileoff_t   prevo;          /* previous file offset */
2190                 xfs_rtblock_t   rtx;            /* realtime extent number */
2191                 xfs_extlen_t    temp;           /* temp for rt calculations */
2192
2193                 /*
2194                  * Set prod to match the realtime extent size.
2195                  */
2196                 if (!(extsz = ap->ip->i_d.di_extsize))
2197                         extsz = mp->m_sb.sb_rextsize;
2198                 prod = extsz / mp->m_sb.sb_rextsize;
2199                 orig_off = ap->off;
2200                 orig_alen = ap->alen;
2201                 orig_end = orig_off + orig_alen;
2202                 /*
2203                  * If the file offset is unaligned vs. the extent size
2204                  * we need to align it.  This will be possible unless
2205                  * the file was previously written with a kernel that didn't
2206                  * perform this alignment.
2207                  */
2208                 mod_off = do_mod(orig_off, extsz);
2209                 if (mod_off) {
2210                         ap->alen += mod_off;
2211                         ap->off -= mod_off;
2212                 }
2213                 /*
2214                  * Same adjustment for the end of the requested area.
2215                  */
2216                 if ((temp = (ap->alen % extsz)))
2217                         ap->alen += extsz - temp;
2218                 /*
2219                  * If the previous block overlaps with this proposed allocation
2220                  * then move the start forward without adjusting the length.
2221                  */
2222                 prevo =
2223                         ap->prevp->br_startoff == NULLFILEOFF ?
2224                                 0 :
2225                                 (ap->prevp->br_startoff +
2226                                  ap->prevp->br_blockcount);
2227                 if (ap->off != orig_off && ap->off < prevo)
2228                         ap->off = prevo;
2229                 /*
2230                  * If the next block overlaps with this proposed allocation
2231                  * then move the start back without adjusting the length,
2232                  * but not before offset 0.
2233                  * This may of course make the start overlap previous block,
2234                  * and if we hit the offset 0 limit then the next block
2235                  * can still overlap too.
2236                  */
2237                 nexto = (ap->eof || ap->gotp->br_startoff == NULLFILEOFF) ?
2238                         NULLFILEOFF : ap->gotp->br_startoff;
2239                 if (!ap->eof &&
2240                     ap->off + ap->alen != orig_end &&
2241                     ap->off + ap->alen > nexto)
2242                         ap->off = nexto > ap->alen ? nexto - ap->alen : 0;
2243                 /*
2244                  * If we're now overlapping the next or previous extent that
2245                  * means we can't fit an extsz piece in this hole.  Just move
2246                  * the start forward to the first valid spot and set
2247                  * the length so we hit the end.
2248                  */
2249                 if ((ap->off != orig_off && ap->off < prevo) ||
2250                     (ap->off + ap->alen != orig_end &&
2251                      ap->off + ap->alen > nexto)) {
2252                         ap->off = prevo;
2253                         ap->alen = nexto - prevo;
2254                 }
2255                 /*
2256                  * If the result isn't a multiple of rtextents we need to
2257                  * remove blocks until it is.
2258                  */
2259                 if ((temp = (ap->alen % mp->m_sb.sb_rextsize))) {
2260                         /*
2261                          * We're not covering the original request, or
2262                          * we won't be able to once we fix the length.
2263                          */
2264                         if (orig_off < ap->off ||
2265                             orig_end > ap->off + ap->alen ||
2266                             ap->alen - temp < orig_alen)
2267                                 return XFS_ERROR(EINVAL);
2268                         /*
2269                          * Try to fix it by moving the start up.
2270                          */
2271                         if (ap->off + temp <= orig_off) {
2272                                 ap->alen -= temp;
2273                                 ap->off += temp;
2274                         }
2275                         /*
2276                          * Try to fix it by moving the end in.
2277                          */
2278                         else if (ap->off + ap->alen - temp >= orig_end)
2279                                 ap->alen -= temp;
2280                         /*
2281                          * Set the start to the minimum then trim the length.
2282                          */
2283                         else {
2284                                 ap->alen -= orig_off - ap->off;
2285                                 ap->off = orig_off;
2286                                 ap->alen -= ap->alen % mp->m_sb.sb_rextsize;
2287                         }
2288                         /*
2289                          * Result doesn't cover the request, fail it.
2290                          */
2291                         if (orig_off < ap->off || orig_end > ap->off + ap->alen)
2292                                 return XFS_ERROR(EINVAL);
2293                 }
2294                 ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
2295                 /*
2296                  * If the offset & length are not perfectly aligned
2297                  * then kill prod, it will just get us in trouble.
2298                  */
2299                 if (do_mod(ap->off, extsz) || ap->alen % extsz)
2300                         prod = 1;
2301                 /*
2302                  * Set ralen to be the actual requested length in rtextents.
2303                  */
2304                 ralen = ap->alen / mp->m_sb.sb_rextsize;
2305                 /*
2306                  * If the old value was close enough to MAXEXTLEN that
2307                  * we rounded up to it, cut it back so it's valid again.
2308                  * Note that if it's a really large request (bigger than
2309                  * MAXEXTLEN), we don't hear about that number, and can't
2310                  * adjust the starting point to match it.
2311                  */
2312                 if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2313                         ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2314                 /*
2315                  * If it's an allocation to an empty file at offset 0,
2316                  * pick an extent that will space things out in the rt area.
2317                  */
2318                 if (ap->eof && ap->off == 0) {
2319                         error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2320                         if (error)
2321                                 return error;
2322                         ap->rval = rtx * mp->m_sb.sb_rextsize;
2323                 } else
2324                         ap->rval = 0;
2325         }
2326 #else
2327         if (rt)
2328                 ap->rval = 0;
2329 #endif  /* __KERNEL__ */
2330         else if (nullfb)
2331                 ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2332         else
2333                 ap->rval = ap->firstblock;
2334         /*
2335          * If allocating at eof, and there's a previous real block,
2336          * try to use it's last block as our starting point.
2337          */
2338         if (ap->eof && ap->prevp->br_startoff != NULLFILEOFF &&
2339             !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2340             ISVALID(ap->prevp->br_startblock + ap->prevp->br_blockcount,
2341                     ap->prevp->br_startblock)) {
2342                 ap->rval = ap->prevp->br_startblock + ap->prevp->br_blockcount;
2343                 /*
2344                  * Adjust for the gap between prevp and us.
2345                  */
2346                 adjust = ap->off -
2347                         (ap->prevp->br_startoff + ap->prevp->br_blockcount);
2348                 if (adjust &&
2349                     ISVALID(ap->rval + adjust, ap->prevp->br_startblock))
2350                         ap->rval += adjust;
2351         }
2352         /*
2353          * If not at eof, then compare the two neighbor blocks.
2354          * Figure out whether either one gives us a good starting point,
2355          * and pick the better one.
2356          */
2357         else if (!ap->eof) {
2358                 xfs_fsblock_t   gotbno;         /* right side block number */
2359                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
2360                 xfs_fsblock_t   prevbno;        /* left side block number */
2361                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
2362
2363                 /*
2364                  * If there's a previous (left) block, select a requested
2365                  * start block based on it.
2366                  */
2367                 if (ap->prevp->br_startoff != NULLFILEOFF &&
2368                     !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2369                     (prevbno = ap->prevp->br_startblock +
2370                                ap->prevp->br_blockcount) &&
2371                     ISVALID(prevbno, ap->prevp->br_startblock)) {
2372                         /*
2373                          * Calculate gap to end of previous block.
2374                          */
2375                         adjust = prevdiff = ap->off -
2376                                 (ap->prevp->br_startoff +
2377                                  ap->prevp->br_blockcount);
2378                         /*
2379                          * Figure the startblock based on the previous block's
2380                          * end and the gap size.
2381                          * Heuristic!
2382                          * If the gap is large relative to the piece we're
2383                          * allocating, or using it gives us an invalid block
2384                          * number, then just use the end of the previous block.
2385                          */
2386                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2387                             ISVALID(prevbno + prevdiff,
2388                                     ap->prevp->br_startblock))
2389                                 prevbno += adjust;
2390                         else
2391                                 prevdiff += adjust;
2392                         /*
2393                          * If the firstblock forbids it, can't use it,
2394                          * must use default.
2395                          */
2396                         if (!rt && !nullfb &&
2397                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2398                                 prevbno = NULLFSBLOCK;
2399                 }
2400                 /*
2401                  * No previous block or can't follow it, just default.
2402                  */
2403                 else
2404                         prevbno = NULLFSBLOCK;
2405                 /*
2406                  * If there's a following (right) block, select a requested
2407                  * start block based on it.
2408                  */
2409                 if (!ISNULLSTARTBLOCK(ap->gotp->br_startblock)) {
2410                         /*
2411                          * Calculate gap to start of next block.
2412                          */
2413                         adjust = gotdiff = ap->gotp->br_startoff - ap->off;
2414                         /*
2415                          * Figure the startblock based on the next block's
2416                          * start and the gap size.
2417                          */
2418                         gotbno = ap->gotp->br_startblock;
2419                         /*
2420                          * Heuristic!
2421                          * If the gap is large relative to the piece we're
2422                          * allocating, or using it gives us an invalid block
2423                          * number, then just use the start of the next block
2424                          * offset by our length.
2425                          */
2426                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2427                             ISVALID(gotbno - gotdiff, gotbno))
2428                                 gotbno -= adjust;
2429                         else if (ISVALID(gotbno - ap->alen, gotbno)) {
2430                                 gotbno -= ap->alen;
2431                                 gotdiff += adjust - ap->alen;
2432                         } else
2433                                 gotdiff += adjust;
2434                         /*
2435                          * If the firstblock forbids it, can't use it,
2436                          * must use default.
2437                          */
2438                         if (!rt && !nullfb &&
2439                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2440                                 gotbno = NULLFSBLOCK;
2441                 }
2442                 /*
2443                  * No next block, just default.
2444                  */
2445                 else
2446                         gotbno = NULLFSBLOCK;
2447                 /*
2448                  * If both valid, pick the better one, else the only good
2449                  * one, else ap->rval is already set (to 0 or the inode block).
2450                  */
2451                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2452                         ap->rval = prevdiff <= gotdiff ? prevbno : gotbno;
2453                 else if (prevbno != NULLFSBLOCK)
2454                         ap->rval = prevbno;
2455                 else if (gotbno != NULLFSBLOCK)
2456                         ap->rval = gotbno;
2457         }
2458         /*
2459          * If allowed, use ap->rval; otherwise must use firstblock since
2460          * it's in the right allocation group.
2461          */
2462         if (nullfb || rt || XFS_FSB_TO_AGNO(mp, ap->rval) == fb_agno)
2463                 ;
2464         else
2465                 ap->rval = ap->firstblock;
2466         /*
2467          * Realtime allocation, done through xfs_rtallocate_extent.
2468          */
2469         if (rt) {
2470 #ifndef __KERNEL__
2471                 ASSERT(0);
2472 #else
2473                 xfs_rtblock_t   rtb;
2474
2475                 atype = ap->rval == 0 ?
2476                         XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2477                 do_div(ap->rval, mp->m_sb.sb_rextsize);
2478                 rtb = ap->rval;
2479                 ap->alen = ralen;
2480                 if ((error = xfs_rtallocate_extent(ap->tp, ap->rval, 1, ap->alen,
2481                                 &ralen, atype, ap->wasdel, prod, &rtb)))
2482                         return error;
2483                 if (rtb == NULLFSBLOCK && prod > 1 &&
2484                     (error = xfs_rtallocate_extent(ap->tp, ap->rval, 1,
2485                                                    ap->alen, &ralen, atype,
2486                                                    ap->wasdel, 1, &rtb)))
2487                         return error;
2488                 ap->rval = rtb;
2489                 if (ap->rval != NULLFSBLOCK) {
2490                         ap->rval *= mp->m_sb.sb_rextsize;
2491                         ralen *= mp->m_sb.sb_rextsize;
2492                         ap->alen = ralen;
2493                         ap->ip->i_d.di_nblocks += ralen;
2494                         xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2495                         if (ap->wasdel)
2496                                 ap->ip->i_delayed_blks -= ralen;
2497                         /*
2498                          * Adjust the disk quota also. This was reserved
2499                          * earlier.
2500                          */
2501                         XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2502                                 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2503                                                 XFS_TRANS_DQ_RTBCOUNT,
2504                                 (long) ralen);
2505                 } else
2506                         ap->alen = 0;
2507 #endif  /* __KERNEL__ */
2508         }
2509         /*
2510          * Normal allocation, done through xfs_alloc_vextent.
2511          */
2512         else {
2513                 xfs_agnumber_t  ag;
2514                 xfs_alloc_arg_t args;
2515                 xfs_extlen_t    blen;
2516                 xfs_extlen_t    delta;
2517                 int             isaligned;
2518                 xfs_extlen_t    longest;
2519                 xfs_extlen_t    need;
2520                 xfs_extlen_t    nextminlen=0;
2521                 int             notinit;
2522                 xfs_perag_t     *pag;
2523                 xfs_agnumber_t  startag;
2524                 int             tryagain;
2525
2526                 tryagain = isaligned = 0;
2527                 args.tp = ap->tp;
2528                 args.mp = mp;
2529                 args.fsbno = ap->rval;
2530                 args.maxlen = MIN(ap->alen, mp->m_sb.sb_agblocks);
2531                 blen = 0;
2532                 if (nullfb) {
2533                         args.type = XFS_ALLOCTYPE_START_BNO;
2534                         args.total = ap->total;
2535                         /*
2536                          * Find the longest available space.
2537                          * We're going to try for the whole allocation at once.
2538                          */
2539                         startag = ag = XFS_FSB_TO_AGNO(mp, args.fsbno);
2540                         notinit = 0;
2541                         down_read(&mp->m_peraglock);
2542                         while (blen < ap->alen) {
2543                                 pag = &mp->m_perag[ag];
2544                                 if (!pag->pagf_init &&
2545                                     (error = xfs_alloc_pagf_init(mp, args.tp,
2546                                             ag, XFS_ALLOC_FLAG_TRYLOCK))) {
2547                                         up_read(&mp->m_peraglock);
2548                                         return error;
2549                                 }
2550                                 /*
2551                                  * See xfs_alloc_fix_freelist...
2552                                  */
2553                                 if (pag->pagf_init) {
2554                                         need = XFS_MIN_FREELIST_PAG(pag, mp);
2555                                         delta = need > pag->pagf_flcount ?
2556                                                 need - pag->pagf_flcount : 0;
2557                                         longest = (pag->pagf_longest > delta) ?
2558                                                 (pag->pagf_longest - delta) :
2559                                                 (pag->pagf_flcount > 0 ||
2560                                                  pag->pagf_longest > 0);
2561                                         if (blen < longest)
2562                                                 blen = longest;
2563                                 } else
2564                                         notinit = 1;
2565                                 if (++ag == mp->m_sb.sb_agcount)
2566                                         ag = 0;
2567                                 if (ag == startag)
2568                                         break;
2569                         }
2570                         up_read(&mp->m_peraglock);
2571                         /*
2572                          * Since the above loop did a BUF_TRYLOCK, it is
2573                          * possible that there is space for this request.
2574                          */
2575                         if (notinit || blen < ap->minlen)
2576                                 args.minlen = ap->minlen;
2577                         /*
2578                          * If the best seen length is less than the request
2579                          * length, use the best as the minimum.
2580                          */
2581                         else if (blen < ap->alen)
2582                                 args.minlen = blen;
2583                         /*
2584                          * Otherwise we've seen an extent as big as alen,
2585                          * use that as the minimum.
2586                          */
2587                         else
2588                                 args.minlen = ap->alen;
2589                 } else if (ap->low) {
2590                         args.type = XFS_ALLOCTYPE_FIRST_AG;
2591                         args.total = args.minlen = ap->minlen;
2592                 } else {
2593                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
2594                         args.total = ap->total;
2595                         args.minlen = ap->minlen;
2596                 }
2597                 if (ap->ip->i_d.di_extsize) {
2598                         args.prod = ap->ip->i_d.di_extsize;
2599                         if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
2600                                 args.mod = (xfs_extlen_t)(args.prod - args.mod);
2601                 } else if (mp->m_sb.sb_blocksize >= NBPP) {
2602                         args.prod = 1;
2603                         args.mod = 0;
2604                 } else {
2605                         args.prod = NBPP >> mp->m_sb.sb_blocklog;
2606                         if ((args.mod = (xfs_extlen_t)(do_mod(ap->off, args.prod))))
2607                                 args.mod = (xfs_extlen_t)(args.prod - args.mod);
2608                 }
2609                 /*
2610                  * If we are not low on available data blocks, and the
2611                  * underlying logical volume manager is a stripe, and
2612                  * the file offset is zero then try to allocate data
2613                  * blocks on stripe unit boundary.
2614                  * NOTE: ap->aeof is only set if the allocation length
2615                  * is >= the stripe unit and the allocation offset is
2616                  * at the end of file.
2617                  */
2618                 if (!ap->low && ap->aeof) {
2619                         if (!ap->off) {
2620                                 args.alignment = mp->m_dalign;
2621                                 atype = args.type;
2622                                 isaligned = 1;
2623                                 /*
2624                                  * Adjust for alignment
2625                                  */
2626                                 if (blen > args.alignment && blen <= ap->alen)
2627                                         args.minlen = blen - args.alignment;
2628                                 args.minalignslop = 0;
2629                         } else {
2630                                 /*
2631                                  * First try an exact bno allocation.
2632                                  * If it fails then do a near or start bno
2633                                  * allocation with alignment turned on.
2634                                  */
2635                                 atype = args.type;
2636                                 tryagain = 1;
2637                                 args.type = XFS_ALLOCTYPE_THIS_BNO;
2638                                 args.alignment = 1;
2639                                 /*
2640                                  * Compute the minlen+alignment for the
2641                                  * next case.  Set slop so that the value
2642                                  * of minlen+alignment+slop doesn't go up
2643                                  * between the calls.
2644                                  */
2645                                 if (blen > mp->m_dalign && blen <= ap->alen)
2646                                         nextminlen = blen - mp->m_dalign;
2647                                 else
2648                                         nextminlen = args.minlen;
2649                                 if (nextminlen + mp->m_dalign > args.minlen + 1)
2650                                         args.minalignslop =
2651                                                 nextminlen + mp->m_dalign -
2652                                                 args.minlen - 1;
2653                                 else
2654                                         args.minalignslop = 0;
2655                         }
2656                 } else {
2657                         args.alignment = 1;
2658                         args.minalignslop = 0;
2659                 }
2660                 args.minleft = ap->minleft;
2661                 args.wasdel = ap->wasdel;
2662                 args.isfl = 0;
2663                 args.userdata = ap->userdata;
2664                 if ((error = xfs_alloc_vextent(&args)))
2665                         return error;
2666                 if (tryagain && args.fsbno == NULLFSBLOCK) {
2667                         /*
2668                          * Exact allocation failed. Now try with alignment
2669                          * turned on.
2670                          */
2671                         args.type = atype;
2672                         args.fsbno = ap->rval;
2673                         args.alignment = mp->m_dalign;
2674                         args.minlen = nextminlen;
2675                         args.minalignslop = 0;
2676                         isaligned = 1;
2677                         if ((error = xfs_alloc_vextent(&args)))
2678                                 return error;
2679                 }
2680                 if (isaligned && args.fsbno == NULLFSBLOCK) {
2681                         /*
2682                          * allocation failed, so turn off alignment and
2683                          * try again.
2684                          */
2685                         args.type = atype;
2686                         args.fsbno = ap->rval;
2687                         args.alignment = 0;
2688                         if ((error = xfs_alloc_vextent(&args)))
2689                                 return error;
2690                 }
2691                 if (args.fsbno == NULLFSBLOCK && nullfb &&
2692                     args.minlen > ap->minlen) {
2693                         args.minlen = ap->minlen;
2694                         args.type = XFS_ALLOCTYPE_START_BNO;
2695                         args.fsbno = ap->rval;
2696                         if ((error = xfs_alloc_vextent(&args)))
2697                                 return error;
2698                 }
2699                 if (args.fsbno == NULLFSBLOCK && nullfb) {
2700                         args.fsbno = 0;
2701                         args.type = XFS_ALLOCTYPE_FIRST_AG;
2702                         args.total = ap->minlen;
2703                         args.minleft = 0;
2704                         if ((error = xfs_alloc_vextent(&args)))
2705                                 return error;
2706                         ap->low = 1;
2707                 }
2708                 if (args.fsbno != NULLFSBLOCK) {
2709                         ap->firstblock = ap->rval = args.fsbno;
2710                         ASSERT(nullfb || fb_agno == args.agno ||
2711                                (ap->low && fb_agno < args.agno));
2712                         ap->alen = args.len;
2713                         ap->ip->i_d.di_nblocks += args.len;
2714                         xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2715                         if (ap->wasdel)
2716                                 ap->ip->i_delayed_blks -= args.len;
2717                         /*
2718                          * Adjust the disk quota also. This was reserved
2719                          * earlier.
2720                          */
2721                         XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2722                                 ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2723                                                 XFS_TRANS_DQ_BCOUNT,
2724                                 (long) args.len);
2725                 } else {
2726                         ap->rval = NULLFSBLOCK;
2727                         ap->alen = 0;
2728                 }
2729         }
2730         return 0;
2731 #undef  ISVALID
2732 }
2733
2734 /*
2735  * Transform a btree format file with only one leaf node, where the
2736  * extents list will fit in the inode, into an extents format file.
2737  * Since the extent list is already in-core, all we have to do is
2738  * give up the space for the btree root and pitch the leaf block.
2739  */
2740 STATIC int                              /* error */
2741 xfs_bmap_btree_to_extents(
2742         xfs_trans_t             *tp,    /* transaction pointer */
2743         xfs_inode_t             *ip,    /* incore inode pointer */
2744         xfs_btree_cur_t         *cur,   /* btree cursor */
2745         int                     *logflagsp, /* inode logging flags */
2746         int                     whichfork)  /* data or attr fork */
2747 {
2748         /* REFERENCED */
2749         xfs_bmbt_block_t        *cblock;/* child btree block */
2750         xfs_fsblock_t           cbno;   /* child block number */
2751         xfs_buf_t               *cbp;   /* child block's buffer */
2752         int                     error;  /* error return value */
2753         xfs_ifork_t             *ifp;   /* inode fork data */
2754         xfs_mount_t             *mp;    /* mount point structure */
2755         xfs_bmbt_ptr_t          *pp;    /* ptr to block address */
2756         xfs_bmbt_block_t        *rblock;/* root btree block */
2757
2758         ifp = XFS_IFORK_PTR(ip, whichfork);
2759         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2760         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2761         rblock = ifp->if_broot;
2762         ASSERT(INT_GET(rblock->bb_level, ARCH_CONVERT) == 1);
2763         ASSERT(INT_GET(rblock->bb_numrecs, ARCH_CONVERT) == 1);
2764         ASSERT(XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes) == 1);
2765         mp = ip->i_mount;
2766         pp = XFS_BMAP_BROOT_PTR_ADDR(rblock, 1, ifp->if_broot_bytes);
2767         *logflagsp = 0;
2768 #ifdef DEBUG
2769         if ((error = xfs_btree_check_lptr(cur, INT_GET(*pp, ARCH_CONVERT), 1)))
2770                 return error;
2771 #endif
2772         cbno = INT_GET(*pp, ARCH_CONVERT);
2773         if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
2774                         XFS_BMAP_BTREE_REF)))
2775                 return error;
2776         cblock = XFS_BUF_TO_BMBT_BLOCK(cbp);
2777         if ((error = xfs_btree_check_lblock(cur, cblock, 0, cbp)))
2778                 return error;
2779         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
2780         ip->i_d.di_nblocks--;
2781         XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
2782         xfs_trans_binval(tp, cbp);
2783         if (cur->bc_bufs[0] == cbp)
2784                 cur->bc_bufs[0] = NULL;
2785         xfs_iroot_realloc(ip, -1, whichfork);
2786         ASSERT(ifp->if_broot == NULL);
2787         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
2788         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
2789         *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2790         return 0;
2791 }
2792
2793 /*
2794  * Called by xfs_bmapi to update extent list structure and the btree
2795  * after removing space (or undoing a delayed allocation).
2796  */
2797 STATIC int                              /* error */
2798 xfs_bmap_del_extent(
2799         xfs_inode_t             *ip,    /* incore inode pointer */
2800         xfs_trans_t             *tp,    /* current transaction pointer */
2801         xfs_extnum_t            idx,    /* extent number to update/delete */
2802         xfs_bmap_free_t         *flist, /* list of extents to be freed */
2803         xfs_btree_cur_t         *cur,   /* if null, not a btree */
2804         xfs_bmbt_irec_t         *del,   /* data to remove from extent list */
2805         int                     *logflagsp, /* inode logging flags */
2806         int                     whichfork, /* data or attr fork */
2807         int                     rsvd)   /* OK to allocate reserved blocks */
2808 {
2809         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
2810         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
2811         xfs_fsblock_t           del_endblock=0; /* first block past del */
2812         xfs_fileoff_t           del_endoff;     /* first offset past del */
2813         int                     delay;  /* current block is delayed allocated */
2814         int                     do_fx;  /* free extent at end of routine */
2815         xfs_bmbt_rec_t          *ep;    /* current extent entry pointer */
2816         int                     error;  /* error return value */
2817         int                     flags;  /* inode logging flags */
2818 #ifdef XFS_BMAP_TRACE
2819         static char             fname[] = "xfs_bmap_del_extent";
2820 #endif
2821         xfs_bmbt_irec_t         got;    /* current extent entry */
2822         xfs_fileoff_t           got_endoff;     /* first offset past got */
2823         int                     i;      /* temp state */
2824         xfs_ifork_t             *ifp;   /* inode fork pointer */
2825         xfs_mount_t             *mp;    /* mount structure */
2826         xfs_filblks_t           nblks;  /* quota/sb block count */
2827         xfs_bmbt_irec_t         new;    /* new record to be inserted */
2828         /* REFERENCED */
2829         xfs_extnum_t            nextents;       /* number of extents in list */
2830         uint                    qfield; /* quota field to update */
2831         xfs_filblks_t           temp;   /* for indirect length calculations */
2832         xfs_filblks_t           temp2;  /* for indirect length calculations */
2833
2834         XFS_STATS_INC(xs_del_exlist);
2835         mp = ip->i_mount;
2836         ifp = XFS_IFORK_PTR(ip, whichfork);
2837         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2838         ASSERT(idx >= 0 && idx < nextents);
2839         ASSERT(del->br_blockcount > 0);
2840         ep = &ifp->if_u1.if_extents[idx];
2841         xfs_bmbt_get_all(ep, &got);
2842         ASSERT(got.br_startoff <= del->br_startoff);
2843         del_endoff = del->br_startoff + del->br_blockcount;
2844         got_endoff = got.br_startoff + got.br_blockcount;
2845         ASSERT(got_endoff >= del_endoff);
2846         delay = ISNULLSTARTBLOCK(got.br_startblock);
2847         ASSERT(ISNULLSTARTBLOCK(del->br_startblock) == delay);
2848         flags = 0;
2849         qfield = 0;
2850         error = 0;
2851         /*
2852          * If deleting a real allocation, must free up the disk space.
2853          */
2854         if (!delay) {
2855                 flags = XFS_ILOG_CORE;
2856                 /*
2857                  * Realtime allocation.  Free it and record di_nblocks update.
2858                  */
2859                 if (whichfork == XFS_DATA_FORK &&
2860                     (ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
2861                         xfs_fsblock_t   bno;
2862                         xfs_filblks_t   len;
2863
2864                         ASSERT(do_mod(del->br_blockcount,
2865                                       mp->m_sb.sb_rextsize) == 0);
2866                         ASSERT(do_mod(del->br_startblock,
2867                                       mp->m_sb.sb_rextsize) == 0);
2868                         bno = del->br_startblock;
2869                         len = del->br_blockcount;
2870                         do_div(bno, mp->m_sb.sb_rextsize);
2871                         do_div(len, mp->m_sb.sb_rextsize);
2872                         if ((error = xfs_rtfree_extent(ip->i_transp, bno,
2873                                         (xfs_extlen_t)len)))
2874                                 goto done;
2875                         do_fx = 0;
2876                         nblks = len * mp->m_sb.sb_rextsize;
2877                         qfield = XFS_TRANS_DQ_RTBCOUNT;
2878                 }
2879                 /*
2880                  * Ordinary allocation.
2881                  */
2882                 else {
2883                         do_fx = 1;
2884                         nblks = del->br_blockcount;
2885                         qfield = XFS_TRANS_DQ_BCOUNT;
2886                 }
2887                 /*
2888                  * Set up del_endblock and cur for later.
2889                  */
2890                 del_endblock = del->br_startblock + del->br_blockcount;
2891                 if (cur) {
2892                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
2893                                         got.br_startblock, got.br_blockcount,
2894                                         &i)))
2895                                 goto done;
2896                         ASSERT(i == 1);
2897                 }
2898                 da_old = da_new = 0;
2899         } else {
2900                 da_old = STARTBLOCKVAL(got.br_startblock);
2901                 da_new = 0;
2902                 nblks = 0;
2903                 do_fx = 0;
2904         }
2905         /*
2906          * Set flag value to use in switch statement.
2907          * Left-contig is 2, right-contig is 1.
2908          */
2909         switch (((got.br_startoff == del->br_startoff) << 1) |
2910                 (got_endoff == del_endoff)) {
2911         case 3:
2912                 /*
2913                  * Matches the whole extent.  Delete the entry.
2914                  */
2915                 xfs_bmap_trace_delete(fname, "3", ip, idx, 1, whichfork);
2916                 xfs_bmap_delete_exlist(ip, idx, 1, whichfork);
2917                 ifp->if_lastex = idx;
2918                 if (delay)
2919                         break;
2920                 XFS_IFORK_NEXT_SET(ip, whichfork,
2921                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2922                 flags |= XFS_ILOG_CORE;
2923                 if (!cur) {
2924                         flags |= XFS_ILOG_FEXT(whichfork);
2925                         break;
2926                 }
2927                 if ((error = xfs_bmbt_delete(cur, &i)))
2928                         goto done;
2929                 ASSERT(i == 1);
2930                 break;
2931
2932         case 2:
2933                 /*
2934                  * Deleting the first part of the extent.
2935                  */
2936                 xfs_bmap_trace_pre_update(fname, "2", ip, idx, whichfork);
2937                 xfs_bmbt_set_startoff(ep, del_endoff);
2938                 temp = got.br_blockcount - del->br_blockcount;
2939                 xfs_bmbt_set_blockcount(ep, temp);
2940                 ifp->if_lastex = idx;
2941                 if (delay) {
2942                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2943                                 da_old);
2944                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
2945                         xfs_bmap_trace_post_update(fname, "2", ip, idx,
2946                                 whichfork);
2947                         da_new = temp;
2948                         break;
2949                 }
2950                 xfs_bmbt_set_startblock(ep, del_endblock);
2951                 xfs_bmap_trace_post_update(fname, "2", ip, idx, whichfork);
2952                 if (!cur) {
2953                         flags |= XFS_ILOG_FEXT(whichfork);
2954                         break;
2955                 }
2956                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
2957                                 got.br_blockcount - del->br_blockcount,
2958                                 got.br_state)))
2959                         goto done;
2960                 break;
2961
2962         case 1:
2963                 /*
2964                  * Deleting the last part of the extent.
2965                  */
2966                 temp = got.br_blockcount - del->br_blockcount;
2967                 xfs_bmap_trace_pre_update(fname, "1", ip, idx, whichfork);
2968                 xfs_bmbt_set_blockcount(ep, temp);
2969                 ifp->if_lastex = idx;
2970                 if (delay) {
2971                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2972                                 da_old);
2973                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
2974                         xfs_bmap_trace_post_update(fname, "1", ip, idx,
2975                                 whichfork);
2976                         da_new = temp;
2977                         break;
2978                 }
2979                 xfs_bmap_trace_post_update(fname, "1", ip, idx, whichfork);
2980                 if (!cur) {
2981                         flags |= XFS_ILOG_FEXT(whichfork);
2982                         break;
2983                 }
2984                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
2985                                 got.br_startblock,
2986                                 got.br_blockcount - del->br_blockcount,
2987                                 got.br_state)))
2988                         goto done;
2989                 break;
2990
2991         case 0:
2992                 /*
2993                  * Deleting the middle of the extent.
2994                  */
2995                 temp = del->br_startoff - got.br_startoff;
2996                 xfs_bmap_trace_pre_update(fname, "0", ip, idx, whichfork);
2997                 xfs_bmbt_set_blockcount(ep, temp);
2998                 new.br_startoff = del_endoff;
2999                 temp2 = got_endoff - del_endoff;
3000                 new.br_blockcount = temp2;
3001                 new.br_state = got.br_state;
3002                 if (!delay) {
3003                         new.br_startblock = del_endblock;
3004                         flags |= XFS_ILOG_CORE;
3005                         if (cur) {
3006                                 if ((error = xfs_bmbt_update(cur,
3007                                                 got.br_startoff,
3008                                                 got.br_startblock, temp,
3009                                                 got.br_state)))
3010                                         goto done;
3011                                 if ((error = xfs_bmbt_increment(cur, 0, &i)))
3012                                         goto done;
3013                                 cur->bc_rec.b = new;
3014                                 error = xfs_bmbt_insert(cur, &i);
3015                                 if (error && error != ENOSPC)
3016                                         goto done;
3017                                 /*
3018                                  * If get no-space back from btree insert,
3019                                  * it tried a split, and we have a zero
3020                                  * block reservation.
3021                                  * Fix up our state and return the error.
3022                                  */
3023                                 if (error == ENOSPC) {
3024                                         /*
3025                                          * Reset the cursor, don't trust
3026                                          * it after any insert operation.
3027                                          */
3028                                         if ((error = xfs_bmbt_lookup_eq(cur,
3029                                                         got.br_startoff,
3030                                                         got.br_startblock,
3031                                                         temp, &i)))
3032                                                 goto done;
3033                                         ASSERT(i == 1);
3034                                         /*
3035                                          * Update the btree record back
3036                                          * to the original value.
3037                                          */
3038                                         if ((error = xfs_bmbt_update(cur,
3039                                                         got.br_startoff,
3040                                                         got.br_startblock,
3041                                                         got.br_blockcount,
3042                                                         got.br_state)))
3043                                                 goto done;
3044                                         /*
3045                                          * Reset the extent record back
3046                                          * to the original value.
3047                                          */
3048                                         xfs_bmbt_set_blockcount(ep,
3049                                                 got.br_blockcount);
3050                                         flags = 0;
3051                                         error = XFS_ERROR(ENOSPC);
3052                                         goto done;
3053                                 }
3054                                 ASSERT(i == 1);
3055                         } else
3056                                 flags |= XFS_ILOG_FEXT(whichfork);
3057                         XFS_IFORK_NEXT_SET(ip, whichfork,
3058                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3059                 } else {
3060                         ASSERT(whichfork == XFS_DATA_FORK);
3061                         temp = xfs_bmap_worst_indlen(ip, temp);
3062                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3063                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
3064                         new.br_startblock = NULLSTARTBLOCK((int)temp2);
3065                         da_new = temp + temp2;
3066                         while (da_new > da_old) {
3067                                 if (temp) {
3068                                         temp--;
3069                                         da_new--;
3070                                         xfs_bmbt_set_startblock(ep,
3071                                                 NULLSTARTBLOCK((int)temp));
3072                                 }
3073                                 if (da_new == da_old)
3074                                         break;
3075                                 if (temp2) {
3076                                         temp2--;
3077                                         da_new--;
3078                                         new.br_startblock =
3079                                                 NULLSTARTBLOCK((int)temp2);
3080                                 }
3081                         }
3082                 }
3083                 xfs_bmap_trace_post_update(fname, "0", ip, idx, whichfork);
3084                 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 1, &new, NULL,
3085                         whichfork);
3086                 xfs_bmap_insert_exlist(ip, idx + 1, 1, &new, whichfork);
3087                 ifp->if_lastex = idx + 1;
3088                 break;
3089         }
3090         /*
3091          * If we need to, add to list of extents to delete.
3092          */
3093         if (do_fx)
3094                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
3095                         mp);
3096         /*
3097          * Adjust inode # blocks in the file.
3098          */
3099         if (nblks)
3100                 ip->i_d.di_nblocks -= nblks;
3101         /*
3102          * Adjust quota data.
3103          */
3104         if (qfield)
3105                 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, qfield, (long)-nblks);
3106
3107         /*
3108          * Account for change in delayed indirect blocks.
3109          * Nothing to do for disk quota accounting here.
3110          */
3111         ASSERT(da_old >= da_new);
3112         if (da_old > da_new)
3113                 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS, (int)(da_old - da_new),
3114                         rsvd);
3115 done:
3116         *logflagsp = flags;
3117         return error;
3118 }
3119
3120 /*
3121  * Remove the entry "free" from the free item list.  Prev points to the
3122  * previous entry, unless "free" is the head of the list.
3123  */
3124 STATIC void
3125 xfs_bmap_del_free(
3126         xfs_bmap_free_t         *flist, /* free item list header */
3127         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
3128         xfs_bmap_free_item_t    *free)  /* list item to be freed */
3129 {
3130         if (prev)
3131                 prev->xbfi_next = free->xbfi_next;
3132         else
3133                 flist->xbf_first = free->xbfi_next;
3134         flist->xbf_count--;
3135         kmem_zone_free(xfs_bmap_free_item_zone, free);
3136 }
3137
3138 /*
3139  * Remove count entries from the extents array for inode "ip", starting
3140  * at index "idx".  Copies the remaining items down over the deleted ones,
3141  * and gives back the excess memory.
3142  */
3143 STATIC void
3144 xfs_bmap_delete_exlist(
3145         xfs_inode_t     *ip,            /* incore inode pointer */
3146         xfs_extnum_t    idx,            /* starting delete index */
3147         xfs_extnum_t    count,          /* count of items to delete */
3148         int             whichfork)      /* data or attr fork */
3149 {
3150         xfs_bmbt_rec_t  *base;          /* base of extent list */
3151         xfs_ifork_t     *ifp;           /* inode fork pointer */
3152         xfs_extnum_t    nextents;       /* number of extents in list after */
3153
3154         ifp = XFS_IFORK_PTR(ip, whichfork);
3155         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3156         base = ifp->if_u1.if_extents;
3157         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - count;
3158         memmove(&base[idx], &base[idx + count],
3159                 (nextents - idx) * sizeof(*base));
3160         xfs_iext_realloc(ip, -count, whichfork);
3161 }
3162
3163 /*
3164  * Convert an extents-format file into a btree-format file.
3165  * The new file will have a root block (in the inode) and a single child block.
3166  */
3167 STATIC int                                      /* error */
3168 xfs_bmap_extents_to_btree(
3169         xfs_trans_t             *tp,            /* transaction pointer */
3170         xfs_inode_t             *ip,            /* incore inode pointer */
3171         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
3172         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
3173         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
3174         int                     wasdel,         /* converting a delayed alloc */
3175         int                     *logflagsp,     /* inode logging flags */
3176         int                     whichfork)      /* data or attr fork */
3177 {
3178         xfs_bmbt_block_t        *ablock;        /* allocated (child) bt block */
3179         xfs_buf_t               *abp;           /* buffer for ablock */
3180         xfs_alloc_arg_t         args;           /* allocation arguments */
3181         xfs_bmbt_rec_t          *arp;           /* child record pointer */
3182         xfs_bmbt_block_t        *block;         /* btree root block */
3183         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
3184         xfs_bmbt_rec_t          *ep;            /* extent list pointer */
3185         int                     error;          /* error return value */
3186         xfs_extnum_t            i, cnt;         /* extent list index */
3187         xfs_ifork_t             *ifp;           /* inode fork pointer */
3188         xfs_bmbt_key_t          *kp;            /* root block key pointer */
3189         xfs_mount_t             *mp;            /* mount structure */
3190         xfs_extnum_t            nextents;       /* extent list size */
3191         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
3192
3193         ifp = XFS_IFORK_PTR(ip, whichfork);
3194         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3195         ASSERT(ifp->if_ext_max ==
3196                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
3197         /*
3198          * Make space in the inode incore.
3199          */
3200         xfs_iroot_realloc(ip, 1, whichfork);
3201         ifp->if_flags |= XFS_IFBROOT;
3202         /*
3203          * Fill in the root.
3204          */
3205         block = ifp->if_broot;
3206         INT_SET(block->bb_magic, ARCH_CONVERT, XFS_BMAP_MAGIC);
3207         INT_SET(block->bb_level, ARCH_CONVERT, 1);
3208         INT_SET(block->bb_numrecs, ARCH_CONVERT, 1);
3209         INT_SET(block->bb_leftsib, ARCH_CONVERT, NULLDFSBNO);
3210         INT_SET(block->bb_rightsib, ARCH_CONVERT, NULLDFSBNO);
3211         /*
3212          * Need a cursor.  Can't allocate until bb_level is filled in.
3213          */
3214         mp = ip->i_mount;
3215         cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
3216                 whichfork);
3217         cur->bc_private.b.firstblock = *firstblock;
3218         cur->bc_private.b.flist = flist;
3219         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3220         /*
3221          * Convert to a btree with two levels, one record in root.
3222          */
3223         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3224         args.tp = tp;
3225         args.mp = mp;
3226         if (*firstblock == NULLFSBLOCK) {
3227                 args.type = XFS_ALLOCTYPE_START_BNO;
3228                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3229         } else if (flist->xbf_low) {
3230                 args.type = XFS_ALLOCTYPE_START_BNO;
3231                 args.fsbno = *firstblock;
3232         } else {
3233                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3234                 args.fsbno = *firstblock;
3235         }
3236         args.minlen = args.maxlen = args.prod = 1;
3237         args.total = args.minleft = args.alignment = args.mod = args.isfl =
3238                 args.minalignslop = 0;
3239         args.wasdel = wasdel;
3240         *logflagsp = 0;
3241         if ((error = xfs_alloc_vextent(&args))) {
3242                 xfs_iroot_realloc(ip, -1, whichfork);
3243                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3244                 return error;
3245         }
3246         /*
3247          * Allocation can't fail, the space was reserved.
3248          */
3249         ASSERT(args.fsbno != NULLFSBLOCK);
3250         ASSERT(*firstblock == NULLFSBLOCK ||
3251                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3252                (flist->xbf_low &&
3253                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3254         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3255         cur->bc_private.b.allocated++;
3256         ip->i_d.di_nblocks++;
3257         XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3258         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3259         /*
3260          * Fill in the child block.
3261          */
3262         ablock = XFS_BUF_TO_BMBT_BLOCK(abp);
3263         INT_SET(ablock->bb_magic, ARCH_CONVERT, XFS_BMAP_MAGIC);
3264         ablock->bb_level = 0;
3265         INT_SET(ablock->bb_leftsib, ARCH_CONVERT, NULLDFSBNO);
3266         INT_SET(ablock->bb_rightsib, ARCH_CONVERT, NULLDFSBNO);
3267         arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3268         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3269         for (ep = ifp->if_u1.if_extents, cnt = i = 0; i < nextents; i++, ep++) {
3270                 if (!ISNULLSTARTBLOCK(xfs_bmbt_get_startblock(ep))) {
3271                         arp->l0 = INT_GET(ep->l0, ARCH_CONVERT);
3272                         arp->l1 = INT_GET(ep->l1, ARCH_CONVERT);
3273                         arp++; cnt++;
3274                 }
3275         }
3276         INT_SET(ablock->bb_numrecs, ARCH_CONVERT, cnt);
3277         ASSERT(INT_GET(ablock->bb_numrecs, ARCH_CONVERT) == XFS_IFORK_NEXTENTS(ip, whichfork));
3278         /*
3279          * Fill in the root key and pointer.
3280          */
3281         kp = XFS_BMAP_KEY_IADDR(block, 1, cur);
3282         arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3283         INT_SET(kp->br_startoff, ARCH_CONVERT, xfs_bmbt_disk_get_startoff(arp));
3284         pp = XFS_BMAP_PTR_IADDR(block, 1, cur);
3285         INT_SET(*pp, ARCH_CONVERT, args.fsbno);
3286         /*
3287          * Do all this logging at the end so that
3288          * the root is at the right level.
3289          */
3290         xfs_bmbt_log_block(cur, abp, XFS_BB_ALL_BITS);
3291         xfs_bmbt_log_recs(cur, abp, 1, INT_GET(ablock->bb_numrecs, ARCH_CONVERT));
3292         ASSERT(*curp == NULL);
3293         *curp = cur;
3294         *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FBROOT(whichfork);
3295         return 0;
3296 }
3297
3298 /*
3299  * Insert new item(s) in the extent list for inode "ip".
3300  * Count new items are inserted at offset idx.
3301  */
3302 STATIC void
3303 xfs_bmap_insert_exlist(
3304         xfs_inode_t     *ip,            /* incore inode pointer */
3305         xfs_extnum_t    idx,            /* starting index of new items */
3306         xfs_extnum_t    count,          /* number of inserted items */
3307         xfs_bmbt_irec_t *new,           /* items to insert */
3308         int             whichfork)      /* data or attr fork */
3309 {
3310         xfs_bmbt_rec_t  *base;          /* extent list base */
3311         xfs_ifork_t     *ifp;           /* inode fork pointer */
3312         xfs_extnum_t    nextents;       /* extent list size */
3313         xfs_extnum_t    to;             /* extent list index */
3314
3315         ifp = XFS_IFORK_PTR(ip, whichfork);
3316         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3317         xfs_iext_realloc(ip, count, whichfork);
3318         base = ifp->if_u1.if_extents;
3319         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3320         memmove(&base[idx + count], &base[idx],
3321                 (nextents - (idx + count)) * sizeof(*base));
3322         for (to = idx; to < idx + count; to++, new++)
3323                 xfs_bmbt_set_all(&base[to], new);
3324 }
3325
3326 /*
3327  * Convert a local file to an extents file.
3328  * This code is out of bounds for data forks of regular files,
3329  * since the file data needs to get logged so things will stay consistent.
3330  * (The bmap-level manipulations are ok, though).
3331  */
3332 STATIC int                              /* error */
3333 xfs_bmap_local_to_extents(
3334         xfs_trans_t     *tp,            /* transaction pointer */
3335         xfs_inode_t     *ip,            /* incore inode pointer */
3336         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
3337         xfs_extlen_t    total,          /* total blocks needed by transaction */
3338         int             *logflagsp,     /* inode logging flags */
3339         int             whichfork)      /* data or attr fork */
3340 {
3341         int             error;          /* error return value */
3342         int             flags;          /* logging flags returned */
3343 #ifdef XFS_BMAP_TRACE
3344         static char     fname[] = "xfs_bmap_local_to_extents";
3345 #endif
3346         xfs_ifork_t     *ifp;           /* inode fork pointer */
3347
3348         /*
3349          * We don't want to deal with the case of keeping inode data inline yet.
3350          * So sending the data fork of a regular inode is invalid.
3351          */
3352         ASSERT(!((ip->i_d.di_mode & S_IFMT) == S_IFREG &&
3353                  whichfork == XFS_DATA_FORK));
3354         ifp = XFS_IFORK_PTR(ip, whichfork);
3355         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3356         flags = 0;
3357         error = 0;
3358         if (ifp->if_bytes) {
3359                 xfs_alloc_arg_t args;   /* allocation arguments */
3360                 xfs_buf_t       *bp;    /* buffer for extent list block */
3361                 xfs_bmbt_rec_t  *ep;    /* extent list pointer */
3362
3363                 args.tp = tp;
3364                 args.mp = ip->i_mount;
3365                 ASSERT(ifp->if_flags & XFS_IFINLINE);
3366                 /*
3367                  * Allocate a block.  We know we need only one, since the
3368                  * file currently fits in an inode.
3369                  */
3370                 if (*firstblock == NULLFSBLOCK) {
3371                         args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3372                         args.type = XFS_ALLOCTYPE_START_BNO;
3373                 } else {
3374                         args.fsbno = *firstblock;
3375                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
3376                 }
3377                 args.total = total;
3378                 args.mod = args.minleft = args.alignment = args.wasdel =
3379                         args.isfl = args.minalignslop = 0;
3380                 args.minlen = args.maxlen = args.prod = 1;
3381                 if ((error = xfs_alloc_vextent(&args)))
3382                         goto done;
3383                 /*
3384                  * Can't fail, the space was reserved.
3385                  */
3386                 ASSERT(args.fsbno != NULLFSBLOCK);
3387                 ASSERT(args.len == 1);
3388                 *firstblock = args.fsbno;
3389                 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3390                 memcpy((char *)XFS_BUF_PTR(bp), ifp->if_u1.if_data,
3391                         ifp->if_bytes);
3392                 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3393                 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3394                 xfs_iext_realloc(ip, 1, whichfork);
3395                 ep = ifp->if_u1.if_extents;
3396                 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3397                 xfs_bmap_trace_post_update(fname, "new", ip, 0, whichfork);
3398                 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3399                 ip->i_d.di_nblocks = 1;
3400                 XFS_TRANS_MOD_DQUOT_BYINO(args.mp, tp, ip,
3401                         XFS_TRANS_DQ_BCOUNT, 1L);
3402                 flags |= XFS_ILOG_FEXT(whichfork);
3403         } else
3404                 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3405         ifp->if_flags &= ~XFS_IFINLINE;
3406         ifp->if_flags |= XFS_IFEXTENTS;
3407         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3408         flags |= XFS_ILOG_CORE;
3409 done:
3410         *logflagsp = flags;
3411         return error;
3412 }
3413
3414 xfs_bmbt_rec_t *                        /* pointer to found extent entry */
3415 xfs_bmap_do_search_extents(
3416         xfs_bmbt_rec_t  *base,          /* base of extent list */
3417         xfs_extnum_t    lastx,          /* last extent index used */
3418         xfs_extnum_t    nextents,       /* extent list size */
3419         xfs_fileoff_t   bno,            /* block number searched for */
3420         int             *eofp,          /* out: end of file found */
3421         xfs_extnum_t    *lastxp,        /* out: last extent index */
3422         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3423         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3424 {
3425         xfs_bmbt_rec_t  *ep;            /* extent list entry pointer */
3426         xfs_bmbt_irec_t got;            /* extent list entry, decoded */
3427         int             high;           /* high index of binary search */
3428         int             low;            /* low index of binary search */
3429
3430         /*
3431          * Initialize the extent entry structure to catch access to
3432          * uninitialized br_startblock field.
3433          */
3434         got.br_startoff = 0xffa5a5a5a5a5a5a5LL;
3435         got.br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3436         got.br_state = XFS_EXT_INVALID;
3437
3438 #if XFS_BIG_BLKNOS
3439         got.br_startblock = 0xffffa5a5a5a5a5a5LL;
3440 #else
3441         got.br_startblock = 0xffffa5a5;
3442 #endif
3443
3444         if (lastx != NULLEXTNUM && lastx < nextents)
3445                 ep = base + lastx;
3446         else
3447                 ep = NULL;
3448         prevp->br_startoff = NULLFILEOFF;
3449         if (ep && bno >= (got.br_startoff = xfs_bmbt_get_startoff(ep)) &&
3450             bno < got.br_startoff +
3451                   (got.br_blockcount = xfs_bmbt_get_blockcount(ep)))
3452                 *eofp = 0;
3453         else if (ep && lastx < nextents - 1 &&
3454                  bno >= (got.br_startoff = xfs_bmbt_get_startoff(ep + 1)) &&
3455                  bno < got.br_startoff +
3456                        (got.br_blockcount = xfs_bmbt_get_blockcount(ep + 1))) {
3457                 lastx++;
3458                 ep++;
3459                 *eofp = 0;
3460         } else if (nextents == 0)
3461                 *eofp = 1;
3462         else if (bno == 0 &&
3463                  (got.br_startoff = xfs_bmbt_get_startoff(base)) == 0) {
3464                 ep = base;
3465                 lastx = 0;
3466                 got.br_blockcount = xfs_bmbt_get_blockcount(ep);
3467                 *eofp = 0;
3468         } else {
3469                 /* binary search the extents array */
3470                 low = 0;
3471                 high = nextents - 1;
3472                 while (low <= high) {
3473                         XFS_STATS_INC(xs_cmp_exlist);
3474                         lastx = (low + high) >> 1;
3475                         ep = base + lastx;
3476                         got.br_startoff = xfs_bmbt_get_startoff(ep);
3477                         got.br_blockcount = xfs_bmbt_get_blockcount(ep);
3478                         if (bno < got.br_startoff)
3479                                 high = lastx - 1;
3480                         else if (bno >= got.br_startoff + got.br_blockcount)
3481                                 low = lastx + 1;
3482                         else {
3483                                 got.br_startblock = xfs_bmbt_get_startblock(ep);
3484                                 got.br_state = xfs_bmbt_get_state(ep);
3485                                 *eofp = 0;
3486                                 *lastxp = lastx;
3487                                 *gotp = got;
3488                                 return ep;
3489                         }
3490                 }
3491                 if (bno >= got.br_startoff + got.br_blockcount) {
3492                         lastx++;
3493                         if (lastx == nextents) {
3494                                 *eofp = 1;
3495                                 got.br_startblock = xfs_bmbt_get_startblock(ep);
3496                                 got.br_state = xfs_bmbt_get_state(ep);
3497                                 *prevp = got;
3498                                 ep = NULL;
3499                         } else {
3500                                 *eofp = 0;
3501                                 xfs_bmbt_get_all(ep, prevp);
3502                                 ep++;
3503                                 got.br_startoff = xfs_bmbt_get_startoff(ep);
3504                                 got.br_blockcount = xfs_bmbt_get_blockcount(ep);
3505                         }
3506                 } else {
3507                         *eofp = 0;
3508                         if (ep > base)
3509                                 xfs_bmbt_get_all(ep - 1, prevp);
3510                 }
3511         }
3512         if (ep) {
3513                 got.br_startblock = xfs_bmbt_get_startblock(ep);
3514                 got.br_state = xfs_bmbt_get_state(ep);
3515         }
3516         *lastxp = lastx;
3517         *gotp = got;
3518         return ep;
3519 }
3520
3521 /*
3522  * Search the extents list for the inode, for the extent containing bno.
3523  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
3524  * *eofp will be set, and *prevp will contain the last entry (null if none).
3525  * Else, *lastxp will be set to the index of the found
3526  * entry; *gotp will contain the entry.
3527  */
3528 STATIC xfs_bmbt_rec_t *                 /* pointer to found extent entry */
3529 xfs_bmap_search_extents(
3530         xfs_inode_t     *ip,            /* incore inode pointer */
3531         xfs_fileoff_t   bno,            /* block number searched for */
3532         int             whichfork,      /* data or attr fork */
3533         int             *eofp,          /* out: end of file found */
3534         xfs_extnum_t    *lastxp,        /* out: last extent index */
3535         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3536         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3537 {
3538         xfs_ifork_t     *ifp;           /* inode fork pointer */
3539         xfs_bmbt_rec_t  *base;          /* base of extent list */
3540         xfs_extnum_t    lastx;          /* last extent index used */
3541         xfs_extnum_t    nextents;       /* extent list size */
3542         xfs_bmbt_rec_t  *ep;            /* extent list entry pointer */
3543         int             rt;             /* realtime flag    */
3544
3545         XFS_STATS_INC(xs_look_exlist);
3546         ifp = XFS_IFORK_PTR(ip, whichfork);
3547         lastx = ifp->if_lastex;
3548         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3549         base = &ifp->if_u1.if_extents[0];
3550
3551         ep = xfs_bmap_do_search_extents(base, lastx, nextents, bno, eofp,
3552                                           lastxp, gotp, prevp);
3553         rt = ip->i_d.di_flags & XFS_DIFLAG_REALTIME;
3554         if(!rt && !gotp->br_startblock && (*lastxp != NULLEXTNUM)) {
3555                 cmn_err(CE_PANIC,"Access to block zero: fs: <%s> inode: %lld "
3556                         "start_block : %llx start_off : %llx blkcnt : %llx "
3557                         "extent-state : %x \n",
3558                         (ip->i_mount)->m_fsname,(long long)ip->i_ino,
3559                         gotp->br_startblock, gotp->br_startoff,
3560                         gotp->br_blockcount,gotp->br_state);
3561         }
3562         return ep;
3563 }
3564
3565
3566 #ifdef XFS_BMAP_TRACE
3567 ktrace_t        *xfs_bmap_trace_buf;
3568
3569 /*
3570  * Add a bmap trace buffer entry.  Base routine for the others.
3571  */
3572 STATIC void
3573 xfs_bmap_trace_addentry(
3574         int             opcode,         /* operation */
3575         char            *fname,         /* function name */
3576         char            *desc,          /* operation description */
3577         xfs_inode_t     *ip,            /* incore inode pointer */
3578         xfs_extnum_t    idx,            /* index of entry(ies) */
3579         xfs_extnum_t    cnt,            /* count of entries, 1 or 2 */
3580         xfs_bmbt_rec_t  *r1,            /* first record */
3581         xfs_bmbt_rec_t  *r2,            /* second record or null */
3582         int             whichfork)      /* data or attr fork */
3583 {
3584         xfs_bmbt_rec_t  tr2;
3585
3586         ASSERT(cnt == 1 || cnt == 2);
3587         ASSERT(r1 != NULL);
3588         if (cnt == 1) {
3589                 ASSERT(r2 == NULL);
3590                 r2 = &tr2;
3591                 memset(&tr2, 0, sizeof(tr2));
3592         } else
3593                 ASSERT(r2 != NULL);
3594         ktrace_enter(xfs_bmap_trace_buf,
3595                 (void *)(__psint_t)(opcode | (whichfork << 16)),
3596                 (void *)fname, (void *)desc, (void *)ip,
3597                 (void *)(__psint_t)idx,
3598                 (void *)(__psint_t)cnt,
3599                 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3600                 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3601                 (void *)(__psunsigned_t)(r1->l0 >> 32),
3602                 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3603                 (void *)(__psunsigned_t)(r1->l1 >> 32),
3604                 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3605                 (void *)(__psunsigned_t)(r2->l0 >> 32),
3606                 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3607                 (void *)(__psunsigned_t)(r2->l1 >> 32),
3608                 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3609                 );
3610         ASSERT(ip->i_xtrace);
3611         ktrace_enter(ip->i_xtrace,
3612                 (void *)(__psint_t)(opcode | (whichfork << 16)),
3613                 (void *)fname, (void *)desc, (void *)ip,
3614                 (void *)(__psint_t)idx,
3615                 (void *)(__psint_t)cnt,
3616                 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3617                 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3618                 (void *)(__psunsigned_t)(r1->l0 >> 32),
3619                 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3620                 (void *)(__psunsigned_t)(r1->l1 >> 32),
3621                 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3622                 (void *)(__psunsigned_t)(r2->l0 >> 32),
3623                 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3624                 (void *)(__psunsigned_t)(r2->l1 >> 32),
3625                 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3626                 );
3627 }
3628
3629 /*
3630  * Add bmap trace entry prior to a call to xfs_bmap_delete_exlist.
3631  */
3632 STATIC void
3633 xfs_bmap_trace_delete(
3634         char            *fname,         /* function name */
3635         char            *desc,          /* operation description */
3636         xfs_inode_t     *ip,            /* incore inode pointer */
3637         xfs_extnum_t    idx,            /* index of entry(entries) deleted */
3638         xfs_extnum_t    cnt,            /* count of entries deleted, 1 or 2 */
3639         int             whichfork)      /* data or attr fork */
3640 {
3641         xfs_ifork_t     *ifp;           /* inode fork pointer */
3642
3643         ifp = XFS_IFORK_PTR(ip, whichfork);
3644         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_DELETE, fname, desc, ip, idx,
3645                 cnt, &ifp->if_u1.if_extents[idx],
3646                 cnt == 2 ? &ifp->if_u1.if_extents[idx + 1] : NULL,
3647                 whichfork);
3648 }
3649
3650 /*
3651  * Add bmap trace entry prior to a call to xfs_bmap_insert_exlist, or
3652  * reading in the extents list from the disk (in the btree).
3653  */
3654 STATIC void
3655 xfs_bmap_trace_insert(
3656         char            *fname,         /* function name */
3657         char            *desc,          /* operation description */
3658         xfs_inode_t     *ip,            /* incore inode pointer */
3659         xfs_extnum_t    idx,            /* index of entry(entries) inserted */
3660         xfs_extnum_t    cnt,            /* count of entries inserted, 1 or 2 */
3661         xfs_bmbt_irec_t *r1,            /* inserted record 1 */
3662         xfs_bmbt_irec_t *r2,            /* inserted record 2 or null */
3663         int             whichfork)      /* data or attr fork */
3664 {
3665         xfs_bmbt_rec_t  tr1;            /* compressed record 1 */
3666         xfs_bmbt_rec_t  tr2;            /* compressed record 2 if needed */
3667
3668         xfs_bmbt_set_all(&tr1, r1);
3669         if (cnt == 2) {
3670                 ASSERT(r2 != NULL);
3671                 xfs_bmbt_set_all(&tr2, r2);
3672         } else {
3673                 ASSERT(cnt == 1);
3674                 ASSERT(r2 == NULL);
3675         }
3676         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_INSERT, fname, desc, ip, idx,
3677                 cnt, &tr1, cnt == 2 ? &tr2 : NULL, whichfork);
3678 }
3679
3680 /*
3681  * Add bmap trace entry after updating an extent list entry in place.
3682  */
3683 STATIC void
3684 xfs_bmap_trace_post_update(
3685         char            *fname,         /* function name */
3686         char            *desc,          /* operation description */
3687         xfs_inode_t     *ip,            /* incore inode pointer */
3688         xfs_extnum_t    idx,            /* index of entry updated */
3689         int             whichfork)      /* data or attr fork */
3690 {
3691         xfs_ifork_t     *ifp;           /* inode fork pointer */
3692
3693         ifp = XFS_IFORK_PTR(ip, whichfork);
3694         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_POST_UP, fname, desc, ip, idx,
3695                 1, &ifp->if_u1.if_extents[idx], NULL, whichfork);
3696 }
3697
3698 /*
3699  * Add bmap trace entry prior to updating an extent list entry in place.
3700  */
3701 STATIC void
3702 xfs_bmap_trace_pre_update(
3703         char            *fname,         /* function name */
3704         char            *desc,          /* operation description */
3705         xfs_inode_t     *ip,            /* incore inode pointer */
3706         xfs_extnum_t    idx,            /* index of entry to be updated */
3707         int             whichfork)      /* data or attr fork */
3708 {
3709         xfs_ifork_t     *ifp;           /* inode fork pointer */
3710
3711         ifp = XFS_IFORK_PTR(ip, whichfork);
3712         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_PRE_UP, fname, desc, ip, idx, 1,
3713                 &ifp->if_u1.if_extents[idx], NULL, whichfork);
3714 }
3715 #endif  /* XFS_BMAP_TRACE */
3716
3717 /*
3718  * Compute the worst-case number of indirect blocks that will be used
3719  * for ip's delayed extent of length "len".
3720  */
3721 STATIC xfs_filblks_t
3722 xfs_bmap_worst_indlen(
3723         xfs_inode_t     *ip,            /* incore inode pointer */
3724         xfs_filblks_t   len)            /* delayed extent length */
3725 {
3726         int             level;          /* btree level number */
3727         int             maxrecs;        /* maximum record count at this level */
3728         xfs_mount_t     *mp;            /* mount structure */
3729         xfs_filblks_t   rval;           /* return value */
3730
3731         mp = ip->i_mount;
3732         maxrecs = mp->m_bmap_dmxr[0];
3733         for (level = 0, rval = 0;
3734              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3735              level++) {
3736                 len += maxrecs - 1;
3737                 do_div(len, maxrecs);
3738                 rval += len;
3739                 if (len == 1)
3740                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3741                                 level - 1;
3742                 if (level == 0)
3743                         maxrecs = mp->m_bmap_dmxr[1];
3744         }
3745         return rval;
3746 }
3747
3748 #if defined(XFS_RW_TRACE)
3749 STATIC void
3750 xfs_bunmap_trace(
3751         xfs_inode_t             *ip,
3752         xfs_fileoff_t           bno,
3753         xfs_filblks_t           len,
3754         int                     flags,
3755         inst_t                  *ra)
3756 {
3757         if (ip->i_rwtrace == NULL)
3758                 return;
3759         ktrace_enter(ip->i_rwtrace,
3760                 (void *)(__psint_t)XFS_BUNMAPI,
3761                 (void *)ip,
3762                 (void *)(__psint_t)((ip->i_d.di_size >> 32) & 0xffffffff),
3763                 (void *)(__psint_t)(ip->i_d.di_size & 0xffffffff),
3764                 (void *)(__psint_t)(((xfs_dfiloff_t)bno >> 32) & 0xffffffff),
3765                 (void *)(__psint_t)((xfs_dfiloff_t)bno & 0xffffffff),
3766                 (void *)(__psint_t)len,
3767                 (void *)(__psint_t)flags,
3768                 (void *)(unsigned long)current_cpu(),
3769                 (void *)ra,
3770                 (void *)0,
3771                 (void *)0,
3772                 (void *)0,
3773                 (void *)0,
3774                 (void *)0,
3775                 (void *)0);
3776 }
3777 #endif
3778
3779 /*
3780  * Convert inode from non-attributed to attributed.
3781  * Must not be in a transaction, ip must not be locked.
3782  */
3783 int                                             /* error code */
3784 xfs_bmap_add_attrfork(
3785         xfs_inode_t             *ip,            /* incore inode pointer */
3786         int                     rsvd)           /* OK to allocated reserved blocks in trans */
3787 {
3788         int                     blks;           /* space reservation */
3789         int                     committed;      /* xaction was committed */
3790         int                     error;          /* error return value */
3791         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
3792         xfs_bmap_free_t         flist;          /* freed extent list */
3793         int                     logflags;       /* logging flags */
3794         xfs_mount_t             *mp;            /* mount structure */
3795         unsigned long           s;              /* spinlock spl value */
3796         xfs_trans_t             *tp;            /* transaction pointer */
3797
3798         ASSERT(ip->i_df.if_ext_max ==
3799                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3800         if (XFS_IFORK_Q(ip))
3801                 return 0;
3802         mp = ip->i_mount;
3803         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3804         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3805         blks = XFS_ADDAFORK_SPACE_RES(mp);
3806         if (rsvd)
3807                 tp->t_flags |= XFS_TRANS_RESERVE;
3808         if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3809                         XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3810                 goto error0;
3811         xfs_ilock(ip, XFS_ILOCK_EXCL);
3812         error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip, blks, 0, rsvd ?
3813                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3814                         XFS_QMOPT_RES_REGBLKS);
3815         if (error) {
3816                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3817                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3818                 return error;
3819         }
3820         if (XFS_IFORK_Q(ip))
3821                 goto error1;
3822         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3823                 /*
3824                  * For inodes coming from pre-6.2 filesystems.
3825                  */
3826                 ASSERT(ip->i_d.di_aformat == 0);
3827                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3828         }
3829         ASSERT(ip->i_d.di_anextents == 0);
3830         VN_HOLD(XFS_ITOV(ip));
3831         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3832         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3833         switch (ip->i_d.di_format) {
3834         case XFS_DINODE_FMT_DEV:
3835                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
3836                 break;
3837         case XFS_DINODE_FMT_UUID:
3838                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
3839                 break;
3840         case XFS_DINODE_FMT_LOCAL:
3841         case XFS_DINODE_FMT_EXTENTS:
3842         case XFS_DINODE_FMT_BTREE:
3843                 ip->i_d.di_forkoff = mp->m_attroffset >> 3;
3844                 break;
3845         default:
3846                 ASSERT(0);
3847                 error = XFS_ERROR(EINVAL);
3848                 goto error1;
3849         }
3850         ip->i_df.if_ext_max =
3851                 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3852         ASSERT(ip->i_afp == NULL);
3853         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
3854         ip->i_afp->if_ext_max =
3855                 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3856         ip->i_afp->if_flags = XFS_IFEXTENTS;
3857         logflags = 0;
3858         XFS_BMAP_INIT(&flist, &firstblock);
3859         switch (ip->i_d.di_format) {
3860         case XFS_DINODE_FMT_LOCAL:
3861                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
3862                         &logflags);
3863                 break;
3864         case XFS_DINODE_FMT_EXTENTS:
3865                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
3866                         &flist, &logflags);
3867                 break;
3868         case XFS_DINODE_FMT_BTREE:
3869                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
3870                         &logflags);
3871                 break;
3872         default:
3873                 error = 0;
3874                 break;
3875         }
3876         if (logflags)
3877                 xfs_trans_log_inode(tp, ip, logflags);
3878         if (error)
3879                 goto error2;
3880         if (!XFS_SB_VERSION_HASATTR(&mp->m_sb)) {
3881                 s = XFS_SB_LOCK(mp);
3882                 if (!XFS_SB_VERSION_HASATTR(&mp->m_sb)) {
3883                         XFS_SB_VERSION_ADDATTR(&mp->m_sb);
3884                         XFS_SB_UNLOCK(mp, s);
3885                         xfs_mod_sb(tp, XFS_SB_VERSIONNUM);
3886                 } else
3887                         XFS_SB_UNLOCK(mp, s);
3888         }
3889         if ((error = xfs_bmap_finish(&tp, &flist, firstblock, &committed)))
3890                 goto error2;
3891         error = xfs_trans_commit(tp, XFS_TRANS_PERM_LOG_RES, NULL);
3892         ASSERT(ip->i_df.if_ext_max ==
3893                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3894         return error;
3895 error2:
3896         xfs_bmap_cancel(&flist);
3897 error1:
3898         ASSERT(ismrlocked(&ip->i_lock,MR_UPDATE));
3899         xfs_iunlock(ip, XFS_ILOCK_EXCL);
3900 error0:
3901         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
3902         ASSERT(ip->i_df.if_ext_max ==
3903                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3904         return error;
3905 }
3906
3907 /*
3908  * Add the extent to the list of extents to be free at transaction end.
3909  * The list is maintained sorted (by block number).
3910  */
3911 /* ARGSUSED */
3912 void
3913 xfs_bmap_add_free(
3914         xfs_fsblock_t           bno,            /* fs block number of extent */
3915         xfs_filblks_t           len,            /* length of extent */
3916         xfs_bmap_free_t         *flist,         /* list of extents */
3917         xfs_mount_t             *mp)            /* mount point structure */
3918 {
3919         xfs_bmap_free_item_t    *cur;           /* current (next) element */
3920         xfs_bmap_free_item_t    *new;           /* new element */
3921         xfs_bmap_free_item_t    *prev;          /* previous element */
3922 #ifdef DEBUG
3923         xfs_agnumber_t          agno;
3924         xfs_agblock_t           agbno;
3925
3926         ASSERT(bno != NULLFSBLOCK);
3927         ASSERT(len > 0);
3928         ASSERT(len <= MAXEXTLEN);
3929         ASSERT(!ISNULLSTARTBLOCK(bno));
3930         agno = XFS_FSB_TO_AGNO(mp, bno);
3931         agbno = XFS_FSB_TO_AGBNO(mp, bno);
3932         ASSERT(agno < mp->m_sb.sb_agcount);
3933         ASSERT(agbno < mp->m_sb.sb_agblocks);
3934         ASSERT(len < mp->m_sb.sb_agblocks);
3935         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
3936 #endif
3937         ASSERT(xfs_bmap_free_item_zone != NULL);
3938         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
3939         new->xbfi_startblock = bno;
3940         new->xbfi_blockcount = (xfs_extlen_t)len;
3941         for (prev = NULL, cur = flist->xbf_first;
3942              cur != NULL;
3943              prev = cur, cur = cur->xbfi_next) {
3944                 if (cur->xbfi_startblock >= bno)
3945                         break;
3946         }
3947         if (prev)
3948                 prev->xbfi_next = new;
3949         else
3950                 flist->xbf_first = new;
3951         new->xbfi_next = cur;
3952         flist->xbf_count++;
3953 }
3954
3955 /*
3956  * Compute and fill in the value of the maximum depth of a bmap btree
3957  * in this filesystem.  Done once, during mount.
3958  */
3959 void
3960 xfs_bmap_compute_maxlevels(
3961         xfs_mount_t     *mp,            /* file system mount structure */
3962         int             whichfork)      /* data or attr fork */
3963 {
3964         int             level;          /* btree level */
3965         uint            maxblocks;      /* max blocks at this level */
3966         uint            maxleafents;    /* max leaf entries possible */
3967         int             maxrootrecs;    /* max records in root block */
3968         int             minleafrecs;    /* min records in leaf block */
3969         int             minnoderecs;    /* min records in node block */
3970         int             sz;             /* root block size */
3971
3972         /*
3973          * The maximum number of extents in a file, hence the maximum
3974          * number of leaf entries, is controlled by the type of di_nextents
3975          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
3976          * (a signed 16-bit number, xfs_aextnum_t).
3977          */
3978         maxleafents = (whichfork == XFS_DATA_FORK) ? MAXEXTNUM : MAXAEXTNUM;
3979         minleafrecs = mp->m_bmap_dmnr[0];
3980         minnoderecs = mp->m_bmap_dmnr[1];
3981         sz = (whichfork == XFS_DATA_FORK) ?
3982                 mp->m_attroffset :
3983                 mp->m_sb.sb_inodesize - mp->m_attroffset;
3984         maxrootrecs = (int)XFS_BTREE_BLOCK_MAXRECS(sz, xfs_bmdr, 0);
3985         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
3986         for (level = 1; maxblocks > 1; level++) {
3987                 if (maxblocks <= maxrootrecs)
3988                         maxblocks = 1;
3989                 else
3990                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
3991         }
3992         mp->m_bm_maxlevels[whichfork] = level;
3993 }
3994
3995 /*
3996  * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
3997  * caller.  Frees all the extents that need freeing, which must be done
3998  * last due to locking considerations.  We never free any extents in
3999  * the first transaction.  This is to allow the caller to make the first
4000  * transaction a synchronous one so that the pointers to the data being
4001  * broken in this transaction will be permanent before the data is actually
4002  * freed.  This is necessary to prevent blocks from being reallocated
4003  * and written to before the free and reallocation are actually permanent.
4004  * We do not just make the first transaction synchronous here, because
4005  * there are more efficient ways to gain the same protection in some cases
4006  * (see the file truncation code).
4007  *
4008  * Return 1 if the given transaction was committed and a new one
4009  * started, and 0 otherwise in the committed parameter.
4010  */
4011 /*ARGSUSED*/
4012 int                                             /* error */
4013 xfs_bmap_finish(
4014         xfs_trans_t             **tp,           /* transaction pointer addr */
4015         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
4016         xfs_fsblock_t           firstblock,     /* controlled ag for allocs */
4017         int                     *committed)     /* xact committed or not */
4018 {
4019         xfs_efd_log_item_t      *efd;           /* extent free data */
4020         xfs_efi_log_item_t      *efi;           /* extent free intention */
4021         int                     error;          /* error return value */
4022         xfs_bmap_free_item_t    *free;          /* free extent list item */
4023         unsigned int            logres;         /* new log reservation */
4024         unsigned int            logcount;       /* new log count */
4025         xfs_mount_t             *mp;            /* filesystem mount structure */
4026         xfs_bmap_free_item_t    *next;          /* next item on free list */
4027         xfs_trans_t             *ntp;           /* new transaction pointer */
4028
4029         ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
4030         if (flist->xbf_count == 0) {
4031                 *committed = 0;
4032                 return 0;
4033         }
4034         ntp = *tp;
4035         efi = xfs_trans_get_efi(ntp, flist->xbf_count);
4036         for (free = flist->xbf_first; free; free = free->xbfi_next)
4037                 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
4038                         free->xbfi_blockcount);
4039         logres = ntp->t_log_res;
4040         logcount = ntp->t_log_count;
4041         ntp = xfs_trans_dup(*tp);
4042         error = xfs_trans_commit(*tp, 0, NULL);
4043         *tp = ntp;
4044         *committed = 1;
4045         /*
4046          * We have a new transaction, so we should return committed=1,
4047          * even though we're returning an error.
4048          */
4049         if (error) {
4050                 return error;
4051         }
4052         if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
4053                         logcount)))
4054                 return error;
4055         efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
4056         for (free = flist->xbf_first; free != NULL; free = next) {
4057                 next = free->xbfi_next;
4058                 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
4059                                 free->xbfi_blockcount))) {
4060                         /*
4061                          * The bmap free list will be cleaned up at a
4062                          * higher level.  The EFI will be canceled when
4063                          * this transaction is aborted.
4064                          * Need to force shutdown here to make sure it
4065                          * happens, since this transaction may not be
4066                          * dirty yet.
4067                          */
4068                         mp = ntp->t_mountp;
4069                         if (!XFS_FORCED_SHUTDOWN(mp))
4070                                 xfs_force_shutdown(mp,
4071                                                    (error == EFSCORRUPTED) ?
4072                                                    XFS_CORRUPT_INCORE :
4073                                                    XFS_METADATA_IO_ERROR);
4074                         return error;
4075                 }
4076                 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
4077                         free->xbfi_blockcount);
4078                 xfs_bmap_del_free(flist, NULL, free);
4079         }
4080         return 0;
4081 }
4082
4083 /*
4084  * Free up any items left in the list.
4085  */
4086 void
4087 xfs_bmap_cancel(
4088         xfs_bmap_free_t         *flist) /* list of bmap_free_items */
4089 {
4090         xfs_bmap_free_item_t    *free;  /* free list item */
4091         xfs_bmap_free_item_t    *next;
4092
4093         if (flist->xbf_count == 0)
4094                 return;
4095         ASSERT(flist->xbf_first != NULL);
4096         for (free = flist->xbf_first; free; free = next) {
4097                 next = free->xbfi_next;
4098                 xfs_bmap_del_free(flist, NULL, free);
4099         }
4100         ASSERT(flist->xbf_count == 0);
4101 }
4102
4103 /*
4104  * Returns the file-relative block number of the first unused block(s)
4105  * in the file with at least "len" logically contiguous blocks free.
4106  * This is the lowest-address hole if the file has holes, else the first block
4107  * past the end of file.
4108  * Return 0 if the file is currently local (in-inode).
4109  */
4110 int                                             /* error */
4111 xfs_bmap_first_unused(
4112         xfs_trans_t     *tp,                    /* transaction pointer */
4113         xfs_inode_t     *ip,                    /* incore inode */
4114         xfs_extlen_t    len,                    /* size of hole to find */
4115         xfs_fileoff_t   *first_unused,          /* unused block */
4116         int             whichfork)              /* data or attr fork */
4117 {
4118         xfs_bmbt_rec_t  *base;                  /* base of extent array */
4119         xfs_bmbt_rec_t  *ep;                    /* pointer to an extent entry */
4120         int             error;                  /* error return value */
4121         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4122         xfs_fileoff_t   lastaddr;               /* last block number seen */
4123         xfs_fileoff_t   lowest;                 /* lowest useful block */
4124         xfs_fileoff_t   max;                    /* starting useful block */
4125         xfs_fileoff_t   off;                    /* offset for this block */
4126         xfs_extnum_t    nextents;               /* number of extent entries */
4127
4128         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
4129                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
4130                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
4131         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4132                 *first_unused = 0;
4133                 return 0;
4134         }
4135         ifp = XFS_IFORK_PTR(ip, whichfork);
4136         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4137             (error = xfs_iread_extents(tp, ip, whichfork)))
4138                 return error;
4139         lowest = *first_unused;
4140         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4141         base = &ifp->if_u1.if_extents[0];
4142         for (lastaddr = 0, max = lowest, ep = base;
4143              ep < &base[nextents];
4144              ep++) {
4145                 off = xfs_bmbt_get_startoff(ep);
4146                 /*
4147                  * See if the hole before this extent will work.
4148                  */
4149                 if (off >= lowest + len && off - max >= len) {
4150                         *first_unused = max;
4151                         return 0;
4152                 }
4153                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
4154                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
4155         }
4156         *first_unused = max;
4157         return 0;
4158 }
4159
4160 /*
4161  * Returns the file-relative block number of the last block + 1 before
4162  * last_block (input value) in the file.
4163  * This is not based on i_size, it is based on the extent list.
4164  * Returns 0 for local files, as they do not have an extent list.
4165  */
4166 int                                             /* error */
4167 xfs_bmap_last_before(
4168         xfs_trans_t     *tp,                    /* transaction pointer */
4169         xfs_inode_t     *ip,                    /* incore inode */
4170         xfs_fileoff_t   *last_block,            /* last block */
4171         int             whichfork)              /* data or attr fork */
4172 {
4173         xfs_fileoff_t   bno;                    /* input file offset */
4174         int             eof;                    /* hit end of file */
4175         xfs_bmbt_rec_t  *ep;                    /* pointer to last extent */
4176         int             error;                  /* error return value */
4177         xfs_bmbt_irec_t got;                    /* current extent value */
4178         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4179         xfs_extnum_t    lastx;                  /* last extent used */
4180         xfs_bmbt_irec_t prev;                   /* previous extent value */
4181
4182         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4183             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4184             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4185                return XFS_ERROR(EIO);
4186         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4187                 *last_block = 0;
4188                 return 0;
4189         }
4190         ifp = XFS_IFORK_PTR(ip, whichfork);
4191         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4192             (error = xfs_iread_extents(tp, ip, whichfork)))
4193                 return error;
4194         bno = *last_block - 1;
4195         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4196                 &prev);
4197         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
4198                 if (prev.br_startoff == NULLFILEOFF)
4199                         *last_block = 0;
4200                 else
4201                         *last_block = prev.br_startoff + prev.br_blockcount;
4202         }
4203         /*
4204          * Otherwise *last_block is already the right answer.
4205          */
4206         return 0;
4207 }
4208
4209 /*
4210  * Returns the file-relative block number of the first block past eof in
4211  * the file.  This is not based on i_size, it is based on the extent list.
4212  * Returns 0 for local files, as they do not have an extent list.
4213  */
4214 int                                             /* error */
4215 xfs_bmap_last_offset(
4216         xfs_trans_t     *tp,                    /* transaction pointer */
4217         xfs_inode_t     *ip,                    /* incore inode */
4218         xfs_fileoff_t   *last_block,            /* last block */
4219         int             whichfork)              /* data or attr fork */
4220 {
4221         xfs_bmbt_rec_t  *base;                  /* base of extent array */
4222         xfs_bmbt_rec_t  *ep;                    /* pointer to last extent */
4223         int             error;                  /* error return value */
4224         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4225         xfs_extnum_t    nextents;               /* number of extent entries */
4226
4227         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4228             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4229             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4230                return XFS_ERROR(EIO);
4231         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4232                 *last_block = 0;
4233                 return 0;
4234         }
4235         ifp = XFS_IFORK_PTR(ip, whichfork);
4236         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4237             (error = xfs_iread_extents(tp, ip, whichfork)))
4238                 return error;
4239         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4240         if (!nextents) {
4241                 *last_block = 0;
4242                 return 0;
4243         }
4244         base = &ifp->if_u1.if_extents[0];
4245         ASSERT(base != NULL);
4246         ep = &base[nextents - 1];
4247         *last_block = xfs_bmbt_get_startoff(ep) + xfs_bmbt_get_blockcount(ep);
4248         return 0;
4249 }
4250
4251 /*
4252  * Returns whether the selected fork of the inode has exactly one
4253  * block or not.  For the data fork we check this matches di_size,
4254  * implying the file's range is 0..bsize-1.
4255  */
4256 int                                     /* 1=>1 block, 0=>otherwise */
4257 xfs_bmap_one_block(
4258         xfs_inode_t     *ip,            /* incore inode */
4259         int             whichfork)      /* data or attr fork */
4260 {
4261         xfs_bmbt_rec_t  *ep;            /* ptr to fork's extent */
4262         xfs_ifork_t     *ifp;           /* inode fork pointer */
4263         int             rval;           /* return value */
4264         xfs_bmbt_irec_t s;              /* internal version of extent */
4265
4266 #ifndef DEBUG
4267         if (whichfork == XFS_DATA_FORK)
4268                 return ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize;
4269 #endif  /* !DEBUG */
4270         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4271                 return 0;
4272         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4273                 return 0;
4274         ifp = XFS_IFORK_PTR(ip, whichfork);
4275         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4276         ep = ifp->if_u1.if_extents;
4277         xfs_bmbt_get_all(ep, &s);
4278         rval = s.br_startoff == 0 && s.br_blockcount == 1;
4279         if (rval && whichfork == XFS_DATA_FORK)
4280                 ASSERT(ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
4281         return rval;
4282 }
4283
4284 /*
4285  * Read in the extents to if_extents.
4286  * All inode fields are set up by caller, we just traverse the btree
4287  * and copy the records in. If the file system cannot contain unwritten
4288  * extents, the records are checked for no "state" flags.
4289  */
4290 int                                     /* error */
4291 xfs_bmap_read_extents(
4292         xfs_trans_t             *tp,    /* transaction pointer */
4293         xfs_inode_t             *ip,    /* incore inode */
4294         int                     whichfork) /* data or attr fork */
4295 {
4296         xfs_bmbt_block_t        *block; /* current btree block */
4297         xfs_fsblock_t           bno;    /* block # of "block" */
4298         xfs_buf_t               *bp;    /* buffer for "block" */
4299         int                     error;  /* error return value */
4300         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
4301 #ifdef XFS_BMAP_TRACE
4302         static char             fname[] = "xfs_bmap_read_extents";
4303 #endif
4304         xfs_extnum_t            i, j;   /* index into the extents list */
4305         xfs_ifork_t             *ifp;   /* fork structure */
4306         int                     level;  /* btree level, for checking */
4307         xfs_mount_t             *mp;    /* file system mount structure */
4308         xfs_bmbt_ptr_t          *pp;    /* pointer to block address */
4309         /* REFERENCED */
4310         xfs_extnum_t            room;   /* number of entries there's room for */
4311         xfs_bmbt_rec_t          *trp;   /* target record pointer */
4312
4313         bno = NULLFSBLOCK;
4314         mp = ip->i_mount;
4315         ifp = XFS_IFORK_PTR(ip, whichfork);
4316         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4317                                         XFS_EXTFMT_INODE(ip);
4318         block = ifp->if_broot;
4319         /*
4320          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4321          */
4322         ASSERT(INT_GET(block->bb_level, ARCH_CONVERT) > 0);
4323         level = INT_GET(block->bb_level, ARCH_CONVERT);
4324         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
4325         ASSERT(INT_GET(*pp, ARCH_CONVERT) != NULLDFSBNO);
4326         ASSERT(XFS_FSB_TO_AGNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agcount);
4327         ASSERT(XFS_FSB_TO_AGBNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agblocks);
4328         bno = INT_GET(*pp, ARCH_CONVERT);
4329         /*
4330          * Go down the tree until leaf level is reached, following the first
4331          * pointer (leftmost) at each level.
4332          */
4333         while (level-- > 0) {
4334                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4335                                 XFS_BMAP_BTREE_REF)))
4336                         return error;
4337                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4338                 XFS_WANT_CORRUPTED_GOTO(
4339                         XFS_BMAP_SANITY_CHECK(mp, block, level),
4340                         error0);
4341                 if (level == 0)
4342                         break;
4343                 pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt, block,
4344                         1, mp->m_bmap_dmxr[1]);
4345                 XFS_WANT_CORRUPTED_GOTO(
4346                         XFS_FSB_SANITY_CHECK(mp, INT_GET(*pp, ARCH_CONVERT)),
4347                         error0);
4348                 bno = INT_GET(*pp, ARCH_CONVERT);
4349                 xfs_trans_brelse(tp, bp);
4350         }
4351         /*
4352          * Here with bp and block set to the leftmost leaf node in the tree.
4353          */
4354         room = ifp->if_bytes / (uint)sizeof(*trp);
4355         trp = ifp->if_u1.if_extents;
4356         i = 0;
4357         /*
4358          * Loop over all leaf nodes.  Copy information to the extent list.
4359          */
4360         for (;;) {
4361                 xfs_bmbt_rec_t  *frp, *temp;
4362                 xfs_fsblock_t   nextbno;
4363                 xfs_extnum_t    num_recs;
4364
4365
4366                 num_recs = INT_GET(block->bb_numrecs, ARCH_CONVERT);
4367                 if (unlikely(i + num_recs > room)) {
4368                         ASSERT(i + num_recs <= room);
4369                         xfs_fs_cmn_err(CE_WARN, ip->i_mount,
4370                                 "corrupt dinode %Lu, (btree extents).  Unmount and run xfs_repair.",
4371                                 (unsigned long long) ip->i_ino);
4372                         XFS_ERROR_REPORT("xfs_bmap_read_extents(1)",
4373                                          XFS_ERRLEVEL_LOW,
4374                                         ip->i_mount);
4375                         goto error0;
4376                 }
4377                 XFS_WANT_CORRUPTED_GOTO(
4378                         XFS_BMAP_SANITY_CHECK(mp, block, 0),
4379                         error0);
4380                 /*
4381                  * Read-ahead the next leaf block, if any.
4382                  */
4383                 nextbno = INT_GET(block->bb_rightsib, ARCH_CONVERT);
4384                 if (nextbno != NULLFSBLOCK)
4385                         xfs_btree_reada_bufl(mp, nextbno, 1);
4386                 /*
4387                  * Copy records into the extent list.
4388                  */
4389                 frp = XFS_BTREE_REC_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt,
4390                         block, 1, mp->m_bmap_dmxr[0]);
4391                 temp = trp;
4392                 for (j = 0; j < num_recs; j++, frp++, trp++) {
4393                         trp->l0 = INT_GET(frp->l0, ARCH_CONVERT);
4394                         trp->l1 = INT_GET(frp->l1, ARCH_CONVERT);
4395                 }
4396                 if (exntf == XFS_EXTFMT_NOSTATE) {
4397                         /*
4398                          * Check all attribute bmap btree records and
4399                          * any "older" data bmap btree records for a
4400                          * set bit in the "extent flag" position.
4401                          */
4402                         if (unlikely(xfs_check_nostate_extents(temp, num_recs))) {
4403                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4404                                                  XFS_ERRLEVEL_LOW,
4405                                                  ip->i_mount);
4406                                 goto error0;
4407                         }
4408                 }
4409                 i += num_recs;
4410                 xfs_trans_brelse(tp, bp);
4411                 bno = nextbno;
4412                 /*
4413                  * If we've reached the end, stop.
4414                  */
4415                 if (bno == NULLFSBLOCK)
4416                         break;
4417                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4418                                 XFS_BMAP_BTREE_REF)))
4419                         return error;
4420                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4421         }
4422         ASSERT(i == ifp->if_bytes / (uint)sizeof(*trp));
4423         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4424         xfs_bmap_trace_exlist(fname, ip, i, whichfork);
4425         return 0;
4426 error0:
4427         xfs_trans_brelse(tp, bp);
4428         return XFS_ERROR(EFSCORRUPTED);
4429 }
4430
4431 #ifdef XFS_BMAP_TRACE
4432 /*
4433  * Add bmap trace insert entries for all the contents of the extent list.
4434  */
4435 void
4436 xfs_bmap_trace_exlist(
4437         char            *fname,         /* function name */
4438         xfs_inode_t     *ip,            /* incore inode pointer */
4439         xfs_extnum_t    cnt,            /* count of entries in the list */
4440         int             whichfork)      /* data or attr fork */
4441 {
4442         xfs_bmbt_rec_t  *base;          /* base of extent list */
4443         xfs_bmbt_rec_t  *ep;            /* current entry in extent list */
4444         xfs_extnum_t    idx;            /* extent list entry number */
4445         xfs_ifork_t     *ifp;           /* inode fork pointer */
4446         xfs_bmbt_irec_t s;              /* extent list record */
4447
4448         ifp = XFS_IFORK_PTR(ip, whichfork);
4449         ASSERT(cnt == ifp->if_bytes / (uint)sizeof(*base));
4450         base = ifp->if_u1.if_extents;
4451         for (idx = 0, ep = base; idx < cnt; idx++, ep++) {
4452                 xfs_bmbt_get_all(ep, &s);
4453                 xfs_bmap_trace_insert(fname, "exlist", ip, idx, 1, &s, NULL,
4454                         whichfork);
4455         }
4456 }
4457 #endif
4458
4459 #ifdef DEBUG
4460 /*
4461  * Validate that the bmbt_irecs being returned from bmapi are valid
4462  * given the callers original parameters.  Specifically check the
4463  * ranges of the returned irecs to ensure that they only extent beyond
4464  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4465  */
4466 STATIC void
4467 xfs_bmap_validate_ret(
4468         xfs_fileoff_t           bno,
4469         xfs_filblks_t           len,
4470         int                     flags,
4471         xfs_bmbt_irec_t         *mval,
4472         int                     nmap,
4473         int                     ret_nmap)
4474 {
4475         int                     i;              /* index to map values */
4476
4477         ASSERT(ret_nmap <= nmap);
4478
4479         for (i = 0; i < ret_nmap; i++) {
4480                 ASSERT(mval[i].br_blockcount > 0);
4481                 if (!(flags & XFS_BMAPI_ENTIRE)) {
4482                         ASSERT(mval[i].br_startoff >= bno);
4483                         ASSERT(mval[i].br_blockcount <= len);
4484                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4485                                bno + len);
4486                 } else {
4487                         ASSERT(mval[i].br_startoff < bno + len);
4488                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4489                                bno);
4490                 }
4491                 ASSERT(i == 0 ||
4492                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4493                        mval[i].br_startoff);
4494                 if ((flags & XFS_BMAPI_WRITE) && !(flags & XFS_BMAPI_DELAY))
4495                         ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4496                                mval[i].br_startblock != HOLESTARTBLOCK);
4497                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4498                        mval[i].br_state == XFS_EXT_UNWRITTEN);
4499         }
4500 }
4501 #endif /* DEBUG */
4502
4503
4504 /*
4505  * Map file blocks to filesystem blocks.
4506  * File range is given by the bno/len pair.
4507  * Adds blocks to file if a write ("flags & XFS_BMAPI_WRITE" set)
4508  * into a hole or past eof.
4509  * Only allocates blocks from a single allocation group,
4510  * to avoid locking problems.
4511  * The returned value in "firstblock" from the first call in a transaction
4512  * must be remembered and presented to subsequent calls in "firstblock".
4513  * An upper bound for the number of blocks to be allocated is supplied to
4514  * the first call in "total"; if no allocation group has that many free
4515  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4516  */
4517 int                                     /* error */
4518 xfs_bmapi(
4519         xfs_trans_t     *tp,            /* transaction pointer */
4520         xfs_inode_t     *ip,            /* incore inode */
4521         xfs_fileoff_t   bno,            /* starting file offs. mapped */
4522         xfs_filblks_t   len,            /* length to map in file */
4523         int             flags,          /* XFS_BMAPI_... */
4524         xfs_fsblock_t   *firstblock,    /* first allocated block
4525                                            controls a.g. for allocs */
4526         xfs_extlen_t    total,          /* total blocks needed */
4527         xfs_bmbt_irec_t *mval,          /* output: map values */
4528         int             *nmap,          /* i/o: mval size/count */
4529         xfs_bmap_free_t *flist)         /* i/o: list extents to free */
4530 {
4531         xfs_fsblock_t   abno;           /* allocated block number */
4532         xfs_extlen_t    alen;           /* allocated extent length */
4533         xfs_fileoff_t   aoff;           /* allocated file offset */
4534         xfs_bmalloca_t  bma;            /* args for xfs_bmap_alloc */
4535         char            contig;         /* allocation must be one extent */
4536         xfs_btree_cur_t *cur;           /* bmap btree cursor */
4537         char            delay;          /* this request is for delayed alloc */
4538         xfs_fileoff_t   end;            /* end of mapped file region */
4539         int             eof;            /* we've hit the end of extent list */
4540         xfs_bmbt_rec_t  *ep;            /* extent list entry pointer */
4541         int             error;          /* error return */
4542         char            exact;          /* don't do all of wasdelayed extent */
4543         xfs_bmbt_irec_t got;            /* current extent list record */
4544         xfs_ifork_t     *ifp;           /* inode fork pointer */
4545         xfs_extlen_t    indlen;         /* indirect blocks length */
4546         char            inhole;         /* current location is hole in file */
4547         xfs_extnum_t    lastx;          /* last useful extent number */
4548         int             logflags;       /* flags for transaction logging */
4549         xfs_extlen_t    minleft;        /* min blocks left after allocation */
4550         xfs_extlen_t    minlen;         /* min allocation size */
4551         xfs_mount_t     *mp;            /* xfs mount structure */
4552         int             n;              /* current extent index */
4553         int             nallocs;        /* number of extents alloc\'d */
4554         xfs_extnum_t    nextents;       /* number of extents in file */
4555         xfs_fileoff_t   obno;           /* old block number (offset) */
4556         xfs_bmbt_irec_t prev;           /* previous extent list record */
4557         char            stateless;      /* ignore state flag set */
4558         int             tmp_logflags;   /* temp flags holder */
4559         char            trim;           /* output trimmed to match range */
4560         char            userdata;       /* allocating non-metadata */
4561         char            wasdelay;       /* old extent was delayed */
4562         int             whichfork;      /* data or attr fork */
4563         char            wr;             /* this is a write request */
4564         char            rsvd;           /* OK to allocate reserved blocks */
4565 #ifdef DEBUG
4566         xfs_fileoff_t   orig_bno;       /* original block number value */
4567         int             orig_flags;     /* original flags arg value */
4568         xfs_filblks_t   orig_len;       /* original value of len arg */
4569         xfs_bmbt_irec_t *orig_mval;     /* original value of mval */
4570         int             orig_nmap;      /* original value of *nmap */
4571
4572         orig_bno = bno;
4573         orig_len = len;
4574         orig_flags = flags;
4575         orig_mval = mval;
4576         orig_nmap = *nmap;
4577 #endif
4578         ASSERT(*nmap >= 1);
4579         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP || !(flags & XFS_BMAPI_WRITE));
4580         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4581                 XFS_ATTR_FORK : XFS_DATA_FORK;
4582         mp = ip->i_mount;
4583         if (unlikely(XFS_TEST_ERROR(
4584             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4585              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4586              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4587              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4588                 XFS_ERROR_REPORT("xfs_bmapi", XFS_ERRLEVEL_LOW, mp);
4589                 return XFS_ERROR(EFSCORRUPTED);
4590         }
4591         if (XFS_FORCED_SHUTDOWN(mp))
4592                 return XFS_ERROR(EIO);
4593         ifp = XFS_IFORK_PTR(ip, whichfork);
4594         ASSERT(ifp->if_ext_max ==
4595                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4596         if ((wr = (flags & XFS_BMAPI_WRITE)) != 0)
4597                 XFS_STATS_INC(xs_blk_mapw);
4598         else
4599                 XFS_STATS_INC(xs_blk_mapr);
4600         delay = (flags & XFS_BMAPI_DELAY) != 0;
4601         trim = (flags & XFS_BMAPI_ENTIRE) == 0;
4602         userdata = (flags & XFS_BMAPI_METADATA) == 0;
4603         exact = (flags & XFS_BMAPI_EXACT) != 0;
4604         rsvd = (flags & XFS_BMAPI_RSVBLOCKS) != 0;
4605         contig = (flags & XFS_BMAPI_CONTIG) != 0;
4606         /*
4607          * stateless is used to combine extents which
4608          * differ only due to the state of the extents.
4609          * This technique is used from xfs_getbmap()
4610          * when the caller does not wish to see the
4611          * separation (which is the default).
4612          *
4613          * This technique is also used when writing a
4614          * buffer which has been partially written,
4615          * (usually by being flushed during a chunkread),
4616          * to ensure one write takes place. This also
4617          * prevents a change in the xfs inode extents at
4618          * this time, intentionally. This change occurs
4619          * on completion of the write operation, in
4620          * xfs_strat_comp(), where the xfs_bmapi() call
4621          * is transactioned, and the extents combined.
4622          */
4623         stateless = (flags & XFS_BMAPI_IGSTATE) != 0;
4624         if (stateless && wr)    /* if writing unwritten space, no */
4625                 wr = 0;         /* allocations are allowed */
4626         ASSERT(wr || !delay);
4627         logflags = 0;
4628         nallocs = 0;
4629         cur = NULL;
4630         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4631                 ASSERT(wr && tp);
4632                 if ((error = xfs_bmap_local_to_extents(tp, ip,
4633                                 firstblock, total, &logflags, whichfork)))
4634                         goto error0;
4635         }
4636         if (wr && *firstblock == NULLFSBLOCK) {
4637                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4638                         minleft = INT_GET(ifp->if_broot->bb_level, ARCH_CONVERT) + 1;
4639                 else
4640                         minleft = 1;
4641         } else
4642                 minleft = 0;
4643         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4644             (error = xfs_iread_extents(tp, ip, whichfork)))
4645                 goto error0;
4646         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4647                 &prev);
4648         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4649         n = 0;
4650         end = bno + len;
4651         obno = bno;
4652         bma.ip = NULL;
4653         while (bno < end && n < *nmap) {
4654                 /*
4655                  * Reading past eof, act as though there's a hole
4656                  * up to end.
4657                  */
4658                 if (eof && !wr)
4659                         got.br_startoff = end;
4660                 inhole = eof || got.br_startoff > bno;
4661                 wasdelay = wr && !inhole && !delay &&
4662                         ISNULLSTARTBLOCK(got.br_startblock);
4663                 /*
4664                  * First, deal with the hole before the allocated space
4665                  * that we found, if any.
4666                  */
4667                 if (wr && (inhole || wasdelay)) {
4668                         /*
4669                          * For the wasdelay case, we could also just
4670                          * allocate the stuff asked for in this bmap call
4671                          * but that wouldn't be as good.
4672                          */
4673                         if (wasdelay && !exact) {
4674                                 alen = (xfs_extlen_t)got.br_blockcount;
4675                                 aoff = got.br_startoff;
4676                                 if (lastx != NULLEXTNUM && lastx) {
4677                                         ep = &ifp->if_u1.if_extents[lastx - 1];
4678                                         xfs_bmbt_get_all(ep, &prev);
4679                                 }
4680                         } else if (wasdelay) {
4681                                 alen = (xfs_extlen_t)
4682                                         XFS_FILBLKS_MIN(len,
4683                                                 (got.br_startoff +
4684                                                  got.br_blockcount) - bno);
4685                                 aoff = bno;
4686                         } else {
4687                                 alen = (xfs_extlen_t)
4688                                         XFS_FILBLKS_MIN(len, MAXEXTLEN);
4689                                 if (!eof)
4690                                         alen = (xfs_extlen_t)
4691                                                 XFS_FILBLKS_MIN(alen,
4692                                                         got.br_startoff - bno);
4693                                 aoff = bno;
4694                         }
4695                         minlen = contig ? alen : 1;
4696                         if (delay) {
4697                                 indlen = (xfs_extlen_t)
4698                                         xfs_bmap_worst_indlen(ip, alen);
4699                                 ASSERT(indlen > 0);
4700                                 /*
4701                                  * Make a transaction-less quota reservation for
4702                                  * delayed allocation blocks. This number gets
4703                                  * adjusted later.
4704                                  * We return EDQUOT if we haven't allocated
4705                                  * blks already inside this loop;
4706                                  */
4707                                 if (XFS_TRANS_RESERVE_BLKQUOTA(
4708                                                 mp, NULL, ip, (long)alen)) {
4709                                         if (n == 0) {
4710                                                 *nmap = 0;
4711                                                 ASSERT(cur == NULL);
4712                                                 return XFS_ERROR(EDQUOT);
4713                                         }
4714                                         break;
4715                                 }
4716
4717                                 /*
4718                                  * Split changing sb for alen and indlen since
4719                                  * they could be coming from different places.
4720                                  */
4721                                 if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) {
4722                                         xfs_extlen_t    extsz;
4723                                         xfs_extlen_t    ralen;
4724                                         if (!(extsz = ip->i_d.di_extsize))
4725                                                 extsz = mp->m_sb.sb_rextsize;
4726                                         ralen = roundup(alen, extsz);
4727                                         ralen = ralen / mp->m_sb.sb_rextsize;
4728                                         if (xfs_mod_incore_sb(mp,
4729                                                 XFS_SBS_FREXTENTS,
4730                                                 -(ralen), rsvd)) {
4731                                                 if (XFS_IS_QUOTA_ON(ip->i_mount))
4732                                                         XFS_TRANS_UNRESERVE_BLKQUOTA(
4733                                                                 mp, NULL, ip,
4734                                                                 (long)alen);
4735                                                 break;
4736                                         }
4737                                 } else {
4738                                         if (xfs_mod_incore_sb(mp,
4739                                                               XFS_SBS_FDBLOCKS,
4740                                                               -(alen), rsvd)) {
4741                                                 if (XFS_IS_QUOTA_ON(ip->i_mount))
4742                                                         XFS_TRANS_UNRESERVE_BLKQUOTA(
4743                                                                 mp, NULL, ip,
4744                                                                 (long)alen);
4745                                                 break;
4746                                         }
4747                                 }
4748
4749                                 if (xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS,
4750                                                 -(indlen), rsvd)) {
4751                                         XFS_TRANS_UNRESERVE_BLKQUOTA(
4752                                                 mp, NULL, ip, (long)alen);
4753                                         break;
4754                                 }
4755                                 ip->i_delayed_blks += alen;
4756                                 abno = NULLSTARTBLOCK(indlen);
4757                         } else {
4758                                 /*
4759                                  * If first time, allocate and fill in
4760                                  * once-only bma fields.
4761                                  */
4762                                 if (bma.ip == NULL) {
4763                                         bma.tp = tp;
4764                                         bma.ip = ip;
4765                                         bma.prevp = &prev;
4766                                         bma.gotp = &got;
4767                                         bma.total = total;
4768                                         bma.userdata = 0;
4769                                 }
4770                                 /* Indicate if this is the first user data
4771                                  * in the file, or just any user data.
4772                                  */
4773                                 if (userdata) {
4774                                         bma.userdata = (aoff == 0) ?
4775                                                 XFS_ALLOC_INITIAL_USER_DATA :
4776                                                 XFS_ALLOC_USERDATA;
4777                                 }
4778                                 /*
4779                                  * Fill in changeable bma fields.
4780                                  */
4781                                 bma.eof = eof;
4782                                 bma.firstblock = *firstblock;
4783                                 bma.alen = alen;
4784                                 bma.off = aoff;
4785                                 bma.wasdel = wasdelay;
4786                                 bma.minlen = minlen;
4787                                 bma.low = flist->xbf_low;
4788                                 bma.minleft = minleft;
4789                                 /*
4790                                  * Only want to do the alignment at the
4791                                  * eof if it is userdata and allocation length
4792                                  * is larger than a stripe unit.
4793                                  */
4794                                 if (mp->m_dalign && alen >= mp->m_dalign &&
4795                                     userdata && whichfork == XFS_DATA_FORK) {
4796                                         if ((error = xfs_bmap_isaeof(ip, aoff,
4797                                                         whichfork, &bma.aeof)))
4798                                                 goto error0;
4799                                 } else
4800                                         bma.aeof = 0;
4801                                 /*
4802                                  * Call allocator.
4803                                  */
4804                                 if ((error = xfs_bmap_alloc(&bma)))
4805                                         goto error0;
4806                                 /*
4807                                  * Copy out result fields.
4808                                  */
4809                                 abno = bma.rval;
4810                                 if ((flist->xbf_low = bma.low))
4811                                         minleft = 0;
4812                                 alen = bma.alen;
4813                                 aoff = bma.off;
4814                                 ASSERT(*firstblock == NULLFSBLOCK ||
4815                                        XFS_FSB_TO_AGNO(mp, *firstblock) ==
4816                                        XFS_FSB_TO_AGNO(mp, bma.firstblock) ||
4817                                        (flist->xbf_low &&
4818                                         XFS_FSB_TO_AGNO(mp, *firstblock) <
4819                                         XFS_FSB_TO_AGNO(mp, bma.firstblock)));
4820                                 *firstblock = bma.firstblock;
4821                                 if (cur)
4822                                         cur->bc_private.b.firstblock =
4823                                                 *firstblock;
4824                                 if (abno == NULLFSBLOCK)
4825                                         break;
4826                                 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4827                                         cur = xfs_btree_init_cursor(mp,
4828                                                 tp, NULL, 0, XFS_BTNUM_BMAP,
4829                                                 ip, whichfork);
4830                                         cur->bc_private.b.firstblock =
4831                                                 *firstblock;
4832                                         cur->bc_private.b.flist = flist;
4833                                 }
4834                                 /*
4835                                  * Bump the number of extents we've allocated
4836                                  * in this call.
4837                                  */
4838                                 nallocs++;
4839                         }
4840                         if (cur)
4841                                 cur->bc_private.b.flags =
4842                                         wasdelay ? XFS_BTCUR_BPRV_WASDEL : 0;
4843                         got.br_startoff = aoff;
4844                         got.br_startblock = abno;
4845                         got.br_blockcount = alen;
4846                         got.br_state = XFS_EXT_NORM;    /* assume normal */
4847                         /*
4848                          * Determine state of extent, and the filesystem.
4849                          * A wasdelay extent has been initialized, so
4850                          * shouldn't be flagged as unwritten.
4851                          */
4852                         if (wr && XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4853                                 if (!wasdelay && (flags & XFS_BMAPI_PREALLOC))
4854                                         got.br_state = XFS_EXT_UNWRITTEN;
4855                         }
4856                         error = xfs_bmap_add_extent(ip, lastx, &cur, &got,
4857                                 firstblock, flist, &tmp_logflags, whichfork,
4858                                 rsvd);
4859                         logflags |= tmp_logflags;
4860                         if (error)
4861                                 goto error0;
4862                         lastx = ifp->if_lastex;
4863                         ep = &ifp->if_u1.if_extents[lastx];
4864                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4865                         xfs_bmbt_get_all(ep, &got);
4866                         ASSERT(got.br_startoff <= aoff);
4867                         ASSERT(got.br_startoff + got.br_blockcount >=
4868                                 aoff + alen);
4869 #ifdef DEBUG
4870                         if (delay) {
4871                                 ASSERT(ISNULLSTARTBLOCK(got.br_startblock));
4872                                 ASSERT(STARTBLOCKVAL(got.br_startblock) > 0);
4873                         }
4874                         ASSERT(got.br_state == XFS_EXT_NORM ||
4875                                got.br_state == XFS_EXT_UNWRITTEN);
4876 #endif
4877                         /*
4878                          * Fall down into the found allocated space case.
4879                          */
4880                 } else if (inhole) {
4881                         /*
4882                          * Reading in a hole.
4883                          */
4884                         mval->br_startoff = bno;
4885                         mval->br_startblock = HOLESTARTBLOCK;
4886                         mval->br_blockcount =
4887                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4888                         mval->br_state = XFS_EXT_NORM;
4889                         bno += mval->br_blockcount;
4890                         len -= mval->br_blockcount;
4891                         mval++;
4892                         n++;
4893                         continue;
4894                 }
4895                 /*
4896                  * Then deal with the allocated space we found.
4897                  */
4898                 ASSERT(ep != NULL);
4899                 if (trim && (got.br_startoff + got.br_blockcount > obno)) {
4900                         if (obno > bno)
4901                                 bno = obno;
4902                         ASSERT((bno >= obno) || (n == 0));
4903                         ASSERT(bno < end);
4904                         mval->br_startoff = bno;
4905                         if (ISNULLSTARTBLOCK(got.br_startblock)) {
4906                                 ASSERT(!wr || delay);
4907                                 mval->br_startblock = DELAYSTARTBLOCK;
4908                         } else
4909                                 mval->br_startblock =
4910                                         got.br_startblock +
4911                                         (bno - got.br_startoff);
4912                         /*
4913                          * Return the minimum of what we got and what we
4914                          * asked for for the length.  We can use the len
4915                          * variable here because it is modified below
4916                          * and we could have been there before coming
4917                          * here if the first part of the allocation
4918                          * didn't overlap what was asked for.
4919                          */
4920                         mval->br_blockcount =
4921                                 XFS_FILBLKS_MIN(end - bno, got.br_blockcount -
4922                                         (bno - got.br_startoff));
4923                         mval->br_state = got.br_state;
4924                         ASSERT(mval->br_blockcount <= len);
4925                 } else {
4926                         *mval = got;
4927                         if (ISNULLSTARTBLOCK(mval->br_startblock)) {
4928                                 ASSERT(!wr || delay);
4929                                 mval->br_startblock = DELAYSTARTBLOCK;
4930                         }
4931                 }
4932
4933                 /*
4934                  * Check if writing previously allocated but
4935                  * unwritten extents.
4936                  */
4937                 if (wr && mval->br_state == XFS_EXT_UNWRITTEN &&
4938                     ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_DELAY)) == 0)) {
4939                         /*
4940                          * Modify (by adding) the state flag, if writing.
4941                          */
4942                         ASSERT(mval->br_blockcount <= len);
4943                         if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4944                                 cur = xfs_btree_init_cursor(mp,
4945                                         tp, NULL, 0, XFS_BTNUM_BMAP,
4946                                         ip, whichfork);
4947                                 cur->bc_private.b.firstblock =
4948                                         *firstblock;
4949                                 cur->bc_private.b.flist = flist;
4950                         }
4951                         mval->br_state = XFS_EXT_NORM;
4952                         error = xfs_bmap_add_extent(ip, lastx, &cur, mval,
4953                                 firstblock, flist, &tmp_logflags, whichfork,
4954                                 rsvd);
4955                         logflags |= tmp_logflags;
4956                         if (error)
4957                                 goto error0;
4958                         lastx = ifp->if_lastex;
4959                         ep = &ifp->if_u1.if_extents[lastx];
4960                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4961                         xfs_bmbt_get_all(ep, &got);
4962                         /*
4963                          * We may have combined previously unwritten
4964                          * space with written space, so generate
4965                          * another request.
4966                          */
4967                         if (mval->br_blockcount < len)
4968                                 continue;
4969                 }
4970
4971                 ASSERT(!trim ||
4972                        ((mval->br_startoff + mval->br_blockcount) <= end));
4973                 ASSERT(!trim || (mval->br_blockcount <= len) ||
4974                        (mval->br_startoff < obno));
4975                 bno = mval->br_startoff + mval->br_blockcount;
4976                 len = end - bno;
4977                 if (n > 0 && mval->br_startoff == mval[-1].br_startoff) {
4978                         ASSERT(mval->br_startblock == mval[-1].br_startblock);
4979                         ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
4980                         ASSERT(mval->br_state == mval[-1].br_state);
4981                         mval[-1].br_blockcount = mval->br_blockcount;
4982                         mval[-1].br_state = mval->br_state;
4983                 } else if (n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
4984                            mval[-1].br_startblock != DELAYSTARTBLOCK &&
4985                            mval[-1].br_startblock != HOLESTARTBLOCK &&
4986                            mval->br_startblock ==
4987                            mval[-1].br_startblock + mval[-1].br_blockcount &&
4988                            (stateless || mval[-1].br_state == mval->br_state)) {
4989                         ASSERT(mval->br_startoff ==
4990                                mval[-1].br_startoff + mval[-1].br_blockcount);
4991                         mval[-1].br_blockcount += mval->br_blockcount;
4992                 } else if (n > 0 &&
4993                            mval->br_startblock == DELAYSTARTBLOCK &&
4994                            mval[-1].br_startblock == DELAYSTARTBLOCK &&
4995                            mval->br_startoff ==
4996                            mval[-1].br_startoff + mval[-1].br_blockcount) {
4997                         mval[-1].br_blockcount += mval->br_blockcount;
4998                         mval[-1].br_state = mval->br_state;
4999                 } else if (!((n == 0) &&
5000                              ((mval->br_startoff + mval->br_blockcount) <=
5001                               obno))) {
5002                         mval++;
5003                         n++;
5004                 }
5005                 /*
5006                  * If we're done, stop now.  Stop when we've allocated
5007                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
5008                  * the transaction may get too big.
5009                  */
5010                 if (bno >= end || n >= *nmap || nallocs >= *nmap)
5011                         break;
5012                 /*
5013                  * Else go on to the next record.
5014                  */
5015                 ep++;
5016                 lastx++;
5017                 if (lastx >= nextents) {
5018                         eof = 1;
5019                         prev = got;
5020                 } else
5021                         xfs_bmbt_get_all(ep, &got);
5022         }
5023         ifp->if_lastex = lastx;
5024         *nmap = n;
5025         /*
5026          * Transform from btree to extents, give it cur.
5027          */
5028         if (tp && XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5029             XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5030                 ASSERT(wr && cur);
5031                 error = xfs_bmap_btree_to_extents(tp, ip, cur,
5032                         &tmp_logflags, whichfork);
5033                 logflags |= tmp_logflags;
5034                 if (error)
5035                         goto error0;
5036         }
5037         ASSERT(ifp->if_ext_max ==
5038                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5039         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
5040                XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max);
5041         error = 0;
5042
5043 error0:
5044         /*
5045          * Log everything.  Do this after conversion, there's no point in
5046          * logging the extent list if we've converted to btree format.
5047          */
5048         if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5049             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5050                 logflags &= ~XFS_ILOG_FEXT(whichfork);
5051         else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5052                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5053                 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5054         /*
5055          * Log whatever the flags say, even if error.  Otherwise we might miss
5056          * detecting a case where the data is changed, there's an error,
5057          * and it's not logged so we don't shutdown when we should.
5058          */
5059         if (logflags) {
5060                 ASSERT(tp && wr);
5061                 xfs_trans_log_inode(tp, ip, logflags);
5062         }
5063         if (cur) {
5064                 if (!error) {
5065                         ASSERT(*firstblock == NULLFSBLOCK ||
5066                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
5067                                XFS_FSB_TO_AGNO(mp,
5068                                        cur->bc_private.b.firstblock) ||
5069                                (flist->xbf_low &&
5070                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
5071                                 XFS_FSB_TO_AGNO(mp,
5072                                         cur->bc_private.b.firstblock)));
5073                         *firstblock = cur->bc_private.b.firstblock;
5074                 }
5075                 xfs_btree_del_cursor(cur,
5076                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5077         }
5078         if (!error)
5079                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
5080                         orig_nmap, *nmap);
5081         return error;
5082 }
5083
5084 /*
5085  * Map file blocks to filesystem blocks, simple version.
5086  * One block (extent) only, read-only.
5087  * For flags, only the XFS_BMAPI_ATTRFORK flag is examined.
5088  * For the other flag values, the effect is as if XFS_BMAPI_METADATA
5089  * was set and all the others were clear.
5090  */
5091 int                                             /* error */
5092 xfs_bmapi_single(
5093         xfs_trans_t     *tp,            /* transaction pointer */
5094         xfs_inode_t     *ip,            /* incore inode */
5095         int             whichfork,      /* data or attr fork */
5096         xfs_fsblock_t   *fsb,           /* output: mapped block */
5097         xfs_fileoff_t   bno)            /* starting file offs. mapped */
5098 {
5099         int             eof;            /* we've hit the end of extent list */
5100         int             error;          /* error return */
5101         xfs_bmbt_irec_t got;            /* current extent list record */
5102         xfs_ifork_t     *ifp;           /* inode fork pointer */
5103         xfs_extnum_t    lastx;          /* last useful extent number */
5104         xfs_bmbt_irec_t prev;           /* previous extent list record */
5105
5106         ifp = XFS_IFORK_PTR(ip, whichfork);
5107         if (unlikely(
5108             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
5109             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)) {
5110                XFS_ERROR_REPORT("xfs_bmapi_single", XFS_ERRLEVEL_LOW,
5111                                 ip->i_mount);
5112                return XFS_ERROR(EFSCORRUPTED);
5113         }
5114         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
5115                 return XFS_ERROR(EIO);
5116         XFS_STATS_INC(xs_blk_mapr);
5117         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5118             (error = xfs_iread_extents(tp, ip, whichfork)))
5119                 return error;
5120         (void)xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5121                 &prev);
5122         /*
5123          * Reading past eof, act as though there's a hole
5124          * up to end.
5125          */
5126         if (eof || got.br_startoff > bno) {
5127                 *fsb = NULLFSBLOCK;
5128                 return 0;
5129         }
5130         ASSERT(!ISNULLSTARTBLOCK(got.br_startblock));
5131         ASSERT(bno < got.br_startoff + got.br_blockcount);
5132         *fsb = got.br_startblock + (bno - got.br_startoff);
5133         ifp->if_lastex = lastx;
5134         return 0;
5135 }
5136
5137 /*
5138  * Unmap (remove) blocks from a file.
5139  * If nexts is nonzero then the number of extents to remove is limited to
5140  * that value.  If not all extents in the block range can be removed then
5141  * *done is set.
5142  */
5143 int                                             /* error */
5144 xfs_bunmapi(
5145         xfs_trans_t             *tp,            /* transaction pointer */
5146         struct xfs_inode        *ip,            /* incore inode */
5147         xfs_fileoff_t           bno,            /* starting offset to unmap */
5148         xfs_filblks_t           len,            /* length to unmap in file */
5149         int                     flags,          /* misc flags */
5150         xfs_extnum_t            nexts,          /* number of extents max */
5151         xfs_fsblock_t           *firstblock,    /* first allocated block
5152                                                    controls a.g. for allocs */
5153         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
5154         int                     *done)          /* set if not done yet */
5155 {
5156         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5157         xfs_bmbt_irec_t         del;            /* extent being deleted */
5158         int                     eof;            /* is deleting at eof */
5159         xfs_bmbt_rec_t          *ep;            /* extent list entry pointer */
5160         int                     error;          /* error return value */
5161         xfs_extnum_t            extno;          /* extent number in list */
5162         xfs_bmbt_irec_t         got;            /* current extent list entry */
5163         xfs_ifork_t             *ifp;           /* inode fork pointer */
5164         int                     isrt;           /* freeing in rt area */
5165         xfs_extnum_t            lastx;          /* last extent index used */
5166         int                     logflags;       /* transaction logging flags */
5167         xfs_extlen_t            mod;            /* rt extent offset */
5168         xfs_mount_t             *mp;            /* mount structure */
5169         xfs_extnum_t            nextents;       /* size of extent list */
5170         xfs_bmbt_irec_t         prev;           /* previous extent list entry */
5171         xfs_fileoff_t           start;          /* first file offset deleted */
5172         int                     tmp_logflags;   /* partial logging flags */
5173         int                     wasdel;         /* was a delayed alloc extent */
5174         int                     whichfork;      /* data or attribute fork */
5175         int                     rsvd;           /* OK to allocate reserved blocks */
5176         xfs_fsblock_t           sum;
5177
5178         xfs_bunmap_trace(ip, bno, len, flags, (inst_t *)__return_address);
5179         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5180                 XFS_ATTR_FORK : XFS_DATA_FORK;
5181         ifp = XFS_IFORK_PTR(ip, whichfork);
5182         if (unlikely(
5183             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5184             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5185                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5186                                  ip->i_mount);
5187                 return XFS_ERROR(EFSCORRUPTED);
5188         }
5189         mp = ip->i_mount;
5190         if (XFS_FORCED_SHUTDOWN(mp))
5191                 return XFS_ERROR(EIO);
5192         rsvd = (flags & XFS_BMAPI_RSVBLOCKS) != 0;
5193         ASSERT(len > 0);
5194         ASSERT(nexts >= 0);
5195         ASSERT(ifp->if_ext_max ==
5196                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5197         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5198             (error = xfs_iread_extents(tp, ip, whichfork)))
5199                 return error;
5200         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5201         if (nextents == 0) {
5202                 *done = 1;
5203                 return 0;
5204         }
5205         XFS_STATS_INC(xs_blk_unmap);
5206         isrt = (whichfork == XFS_DATA_FORK) &&
5207                (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
5208         start = bno;
5209         bno = start + len - 1;
5210         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5211                 &prev);
5212         /*
5213          * Check to see if the given block number is past the end of the
5214          * file, back up to the last block if so...
5215          */
5216         if (eof) {
5217                 ep = &ifp->if_u1.if_extents[--lastx];
5218                 xfs_bmbt_get_all(ep, &got);
5219                 bno = got.br_startoff + got.br_blockcount - 1;
5220         }
5221         logflags = 0;
5222         if (ifp->if_flags & XFS_IFBROOT) {
5223                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5224                 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
5225                         whichfork);
5226                 cur->bc_private.b.firstblock = *firstblock;
5227                 cur->bc_private.b.flist = flist;
5228                 cur->bc_private.b.flags = 0;
5229         } else
5230                 cur = NULL;
5231         extno = 0;
5232         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5233                (nexts == 0 || extno < nexts)) {
5234                 /*
5235                  * Is the found extent after a hole in which bno lives?
5236                  * Just back up to the previous extent, if so.
5237                  */
5238                 if (got.br_startoff > bno) {
5239                         if (--lastx < 0)
5240                                 break;
5241                         ep--;
5242                         xfs_bmbt_get_all(ep, &got);
5243                 }
5244                 /*
5245                  * Is the last block of this extent before the range
5246                  * we're supposed to delete?  If so, we're done.
5247                  */
5248                 bno = XFS_FILEOFF_MIN(bno,
5249                         got.br_startoff + got.br_blockcount - 1);
5250                 if (bno < start)
5251                         break;
5252                 /*
5253                  * Then deal with the (possibly delayed) allocated space
5254                  * we found.
5255                  */
5256                 ASSERT(ep != NULL);
5257                 del = got;
5258                 wasdel = ISNULLSTARTBLOCK(del.br_startblock);
5259                 if (got.br_startoff < start) {
5260                         del.br_startoff = start;
5261                         del.br_blockcount -= start - got.br_startoff;
5262                         if (!wasdel)
5263                                 del.br_startblock += start - got.br_startoff;
5264                 }
5265                 if (del.br_startoff + del.br_blockcount > bno + 1)
5266                         del.br_blockcount = bno + 1 - del.br_startoff;
5267                 sum = del.br_startblock + del.br_blockcount;
5268                 if (isrt &&
5269                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5270                         /*
5271                          * Realtime extent not lined up at the end.
5272                          * The extent could have been split into written
5273                          * and unwritten pieces, or we could just be
5274                          * unmapping part of it.  But we can't really
5275                          * get rid of part of a realtime extent.
5276                          */
5277                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5278                             !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
5279                                 /*
5280                                  * This piece is unwritten, or we're not
5281                                  * using unwritten extents.  Skip over it.
5282                                  */
5283                                 ASSERT(bno >= mod);
5284                                 bno -= mod > del.br_blockcount ?
5285                                         del.br_blockcount : mod;
5286                                 if (bno < got.br_startoff) {
5287                                         if (--lastx >= 0)
5288                                                 xfs_bmbt_get_all(--ep, &got);
5289                                 }
5290                                 continue;
5291                         }
5292                         /*
5293                          * It's written, turn it unwritten.
5294                          * This is better than zeroing it.
5295                          */
5296                         ASSERT(del.br_state == XFS_EXT_NORM);
5297                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5298                         /*
5299                          * If this spans a realtime extent boundary,
5300                          * chop it back to the start of the one we end at.
5301                          */
5302                         if (del.br_blockcount > mod) {
5303                                 del.br_startoff += del.br_blockcount - mod;
5304                                 del.br_startblock += del.br_blockcount - mod;
5305                                 del.br_blockcount = mod;
5306                         }
5307                         del.br_state = XFS_EXT_UNWRITTEN;
5308                         error = xfs_bmap_add_extent(ip, lastx, &cur, &del,
5309                                 firstblock, flist, &logflags, XFS_DATA_FORK, 0);
5310                         if (error)
5311                                 goto error0;
5312                         goto nodelete;
5313                 }
5314                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5315                         /*
5316                          * Realtime extent is lined up at the end but not
5317                          * at the front.  We'll get rid of full extents if
5318                          * we can.
5319                          */
5320                         mod = mp->m_sb.sb_rextsize - mod;
5321                         if (del.br_blockcount > mod) {
5322                                 del.br_blockcount -= mod;
5323                                 del.br_startoff += mod;
5324                                 del.br_startblock += mod;
5325                         } else if ((del.br_startoff == start &&
5326                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5327                                      xfs_trans_get_block_res(tp) == 0)) ||
5328                                    !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
5329                                 /*
5330                                  * Can't make it unwritten.  There isn't
5331                                  * a full extent here so just skip it.
5332                                  */
5333                                 ASSERT(bno >= del.br_blockcount);
5334                                 bno -= del.br_blockcount;
5335                                 if (bno < got.br_startoff) {
5336                                         if (--lastx >= 0)
5337                                                 xfs_bmbt_get_all(--ep, &got);
5338                                 }
5339                                 continue;
5340                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5341                                 /*
5342                                  * This one is already unwritten.
5343                                  * It must have a written left neighbor.
5344                                  * Unwrite the killed part of that one and
5345                                  * try again.
5346                                  */
5347                                 ASSERT(lastx > 0);
5348                                 xfs_bmbt_get_all(ep - 1, &prev);
5349                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5350                                 ASSERT(!ISNULLSTARTBLOCK(prev.br_startblock));
5351                                 ASSERT(del.br_startblock ==
5352                                        prev.br_startblock + prev.br_blockcount);
5353                                 if (prev.br_startoff < start) {
5354                                         mod = start - prev.br_startoff;
5355                                         prev.br_blockcount -= mod;
5356                                         prev.br_startblock += mod;
5357                                         prev.br_startoff = start;
5358                                 }
5359                                 prev.br_state = XFS_EXT_UNWRITTEN;
5360                                 error = xfs_bmap_add_extent(ip, lastx - 1, &cur,
5361                                         &prev, firstblock, flist, &logflags,
5362                                         XFS_DATA_FORK, 0);
5363                                 if (error)
5364                                         goto error0;
5365                                 goto nodelete;
5366                         } else {
5367                                 ASSERT(del.br_state == XFS_EXT_NORM);
5368                                 del.br_state = XFS_EXT_UNWRITTEN;
5369                                 error = xfs_bmap_add_extent(ip, lastx, &cur,
5370                                         &del, firstblock, flist, &logflags,
5371                                         XFS_DATA_FORK, 0);
5372                                 if (error)
5373                                         goto error0;
5374                                 goto nodelete;
5375                         }
5376                 }
5377                 if (wasdel) {
5378                         ASSERT(STARTBLOCKVAL(del.br_startblock) > 0);
5379                         xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS,
5380                                 (int)del.br_blockcount, rsvd);
5381                         /* Unreserve our quota space */
5382                         XFS_TRANS_RESERVE_QUOTA_NBLKS(
5383                                 mp, NULL, ip, -((long)del.br_blockcount), 0,
5384                                 isrt ?  XFS_QMOPT_RES_RTBLKS :
5385                                         XFS_QMOPT_RES_REGBLKS);
5386                         ip->i_delayed_blks -= del.br_blockcount;
5387                         if (cur)
5388                                 cur->bc_private.b.flags |=
5389                                         XFS_BTCUR_BPRV_WASDEL;
5390                 } else if (cur)
5391                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5392                 /*
5393                  * If it's the case where the directory code is running
5394                  * with no block reservation, and the deleted block is in
5395                  * the middle of its extent, and the resulting insert
5396                  * of an extent would cause transformation to btree format,
5397                  * then reject it.  The calling code will then swap
5398                  * blocks around instead.
5399                  * We have to do this now, rather than waiting for the
5400                  * conversion to btree format, since the transaction
5401                  * will be dirty.
5402                  */
5403                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5404                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5405                     XFS_IFORK_NEXTENTS(ip, whichfork) >= ifp->if_ext_max &&
5406                     del.br_startoff > got.br_startoff &&
5407                     del.br_startoff + del.br_blockcount <
5408                     got.br_startoff + got.br_blockcount) {
5409                         error = XFS_ERROR(ENOSPC);
5410                         goto error0;
5411                 }
5412                 error = xfs_bmap_del_extent(ip, tp, lastx, flist, cur, &del,
5413                         &tmp_logflags, whichfork, rsvd);
5414                 logflags |= tmp_logflags;
5415                 if (error)
5416                         goto error0;
5417                 bno = del.br_startoff - 1;
5418 nodelete:
5419                 lastx = ifp->if_lastex;
5420                 /*
5421                  * If not done go on to the next (previous) record.
5422                  * Reset ep in case the extents array was re-alloced.
5423                  */
5424                 ep = &ifp->if_u1.if_extents[lastx];
5425                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5426                         if (lastx >= XFS_IFORK_NEXTENTS(ip, whichfork) ||
5427                             xfs_bmbt_get_startoff(ep) > bno) {
5428                                 lastx--;
5429                                 ep--;
5430                         }
5431                         if (lastx >= 0)
5432                                 xfs_bmbt_get_all(ep, &got);
5433                         extno++;
5434                 }
5435         }
5436         ifp->if_lastex = lastx;
5437         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5438         ASSERT(ifp->if_ext_max ==
5439                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5440         /*
5441          * Convert to a btree if necessary.
5442          */
5443         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5444             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
5445                 ASSERT(cur == NULL);
5446                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5447                         &cur, 0, &tmp_logflags, whichfork);
5448                 logflags |= tmp_logflags;
5449                 if (error)
5450                         goto error0;
5451         }
5452         /*
5453          * transform from btree to extents, give it cur
5454          */
5455         else if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5456                  XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5457                 ASSERT(cur != NULL);
5458                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5459                         whichfork);
5460                 logflags |= tmp_logflags;
5461                 if (error)
5462                         goto error0;
5463         }
5464         /*
5465          * transform from extents to local?
5466          */
5467         ASSERT(ifp->if_ext_max ==
5468                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5469         error = 0;
5470 error0:
5471         /*
5472          * Log everything.  Do this after conversion, there's no point in
5473          * logging the extent list if we've converted to btree format.
5474          */
5475         if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5476             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5477                 logflags &= ~XFS_ILOG_FEXT(whichfork);
5478         else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5479                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5480                 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5481         /*
5482          * Log inode even in the error case, if the transaction
5483          * is dirty we'll need to shut down the filesystem.
5484          */
5485         if (logflags)
5486                 xfs_trans_log_inode(tp, ip, logflags);
5487         if (cur) {
5488                 if (!error) {
5489                         *firstblock = cur->bc_private.b.firstblock;
5490                         cur->bc_private.b.allocated = 0;
5491                 }
5492                 xfs_btree_del_cursor(cur,
5493                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5494         }
5495         return error;
5496 }
5497
5498 /*
5499  * Fcntl interface to xfs_bmapi.
5500  */
5501 int                                             /* error code */
5502 xfs_getbmap(
5503         bhv_desc_t              *bdp,           /* XFS behavior descriptor*/
5504         struct getbmap          *bmv,           /* user bmap structure */
5505         void                    __user *ap,     /* pointer to user's array */
5506         int                     interface)      /* interface flags */
5507 {
5508         __int64_t               bmvend;         /* last block requested */
5509         int                     error;          /* return value */
5510         __int64_t               fixlen;         /* length for -1 case */
5511         int                     i;              /* extent number */
5512         xfs_inode_t             *ip;            /* xfs incore inode pointer */
5513         vnode_t                 *vp;            /* corresponding vnode */
5514         int                     lock;           /* lock state */
5515         xfs_bmbt_irec_t         *map;           /* buffer for user's data */
5516         xfs_mount_t             *mp;            /* file system mount point */
5517         int                     nex;            /* # of user extents can do */
5518         int                     nexleft;        /* # of user extents left */
5519         int                     subnex;         /* # of bmapi's can do */
5520         int                     nmap;           /* number of map entries */
5521         struct getbmap          out;            /* output structure */
5522         int                     whichfork;      /* data or attr fork */
5523         int                     prealloced;     /* this is a file with
5524                                                  * preallocated data space */
5525         int                     sh_unwritten;   /* true, if unwritten */
5526                                                 /* extents listed separately */
5527         int                     bmapi_flags;    /* flags for xfs_bmapi */
5528         __int32_t               oflags;         /* getbmapx bmv_oflags field */
5529
5530         vp = BHV_TO_VNODE(bdp);
5531         ip = XFS_BHVTOI(bdp);
5532         mp = ip->i_mount;
5533
5534         whichfork = interface & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5535         sh_unwritten = (interface & BMV_IF_PREALLOC) != 0;
5536
5537         /*      If the BMV_IF_NO_DMAPI_READ interface bit specified, do not
5538          *      generate a DMAPI read event.  Otherwise, if the DM_EVENT_READ
5539          *      bit is set for the file, generate a read event in order
5540          *      that the DMAPI application may do its thing before we return
5541          *      the extents.  Usually this means restoring user file data to
5542          *      regions of the file that look like holes.
5543          *
5544          *      The "old behavior" (from XFS_IOC_GETBMAP) is to not specify
5545          *      BMV_IF_NO_DMAPI_READ so that read events are generated.
5546          *      If this were not true, callers of ioctl( XFS_IOC_GETBMAP )
5547          *      could misinterpret holes in a DMAPI file as true holes,
5548          *      when in fact they may represent offline user data.
5549          */
5550         if (   (interface & BMV_IF_NO_DMAPI_READ) == 0
5551             && DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_READ)
5552             && whichfork == XFS_DATA_FORK) {
5553
5554                 error = XFS_SEND_DATA(mp, DM_EVENT_READ, vp, 0, 0, 0, NULL);
5555                 if (error)
5556                         return XFS_ERROR(error);
5557         }
5558
5559         if (whichfork == XFS_ATTR_FORK) {
5560                 if (XFS_IFORK_Q(ip)) {
5561                         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5562                             ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5563                             ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5564                                 return XFS_ERROR(EINVAL);
5565                 } else if (unlikely(
5566                            ip->i_d.di_aformat != 0 &&
5567                            ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5568                         XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5569                                          ip->i_mount);
5570                         return XFS_ERROR(EFSCORRUPTED);
5571                 }
5572         } else if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5573                    ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5574                    ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5575                 return XFS_ERROR(EINVAL);
5576         if (whichfork == XFS_DATA_FORK) {
5577                 if (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC) {
5578                         prealloced = 1;
5579                         fixlen = XFS_MAXIOFFSET(mp);
5580                 } else {
5581                         prealloced = 0;
5582                         fixlen = ip->i_d.di_size;
5583                 }
5584         } else {
5585                 prealloced = 0;
5586                 fixlen = 1LL << 32;
5587         }
5588
5589         if (bmv->bmv_length == -1) {
5590                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5591                 bmv->bmv_length = MAX( (__int64_t)(fixlen - bmv->bmv_offset),
5592                                         (__int64_t)0);
5593         } else if (bmv->bmv_length < 0)
5594                 return XFS_ERROR(EINVAL);
5595         if (bmv->bmv_length == 0) {
5596                 bmv->bmv_entries = 0;
5597                 return 0;
5598         }
5599         nex = bmv->bmv_count - 1;
5600         if (nex <= 0)
5601                 return XFS_ERROR(EINVAL);
5602         bmvend = bmv->bmv_offset + bmv->bmv_length;
5603
5604         xfs_ilock(ip, XFS_IOLOCK_SHARED);
5605
5606         if (whichfork == XFS_DATA_FORK && ip->i_delayed_blks) {
5607                 /* xfs_fsize_t last_byte = xfs_file_last_byte(ip); */
5608                 VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1, 0, FI_REMAPF, error);
5609         }
5610
5611         ASSERT(whichfork == XFS_ATTR_FORK || ip->i_delayed_blks == 0);
5612
5613         lock = xfs_ilock_map_shared(ip);
5614
5615         /*
5616          * Don't let nex be bigger than the number of extents
5617          * we can have assuming alternating holes and real extents.
5618          */
5619         if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5620                 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5621
5622         bmapi_flags = XFS_BMAPI_AFLAG(whichfork) |
5623                         ((sh_unwritten) ? 0 : XFS_BMAPI_IGSTATE);
5624
5625         /*
5626          * Allocate enough space to handle "subnex" maps at a time.
5627          */
5628         subnex = 16;
5629         map = kmem_alloc(subnex * sizeof(*map), KM_SLEEP);
5630
5631         bmv->bmv_entries = 0;
5632
5633         if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0) {
5634                 error = 0;
5635                 goto unlock_and_return;
5636         }
5637
5638         nexleft = nex;
5639
5640         do {
5641                 nmap = (nexleft > subnex) ? subnex : nexleft;
5642                 error = xfs_bmapi(NULL, ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5643                                   XFS_BB_TO_FSB(mp, bmv->bmv_length),
5644                                   bmapi_flags, NULL, 0, map, &nmap, NULL);
5645                 if (error)
5646                         goto unlock_and_return;
5647                 ASSERT(nmap <= subnex);
5648
5649                 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5650                         nexleft--;
5651                         oflags = (map[i].br_state == XFS_EXT_UNWRITTEN) ?
5652                                         BMV_OF_PREALLOC : 0;
5653                         out.bmv_offset = XFS_FSB_TO_BB(mp, map[i].br_startoff);
5654                         out.bmv_length = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5655                         ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
5656                         if (prealloced &&
5657                             map[i].br_startblock == HOLESTARTBLOCK &&
5658                             out.bmv_offset + out.bmv_length == bmvend) {
5659                                 /*
5660                                  * came to hole at end of file
5661                                  */
5662                                 goto unlock_and_return;
5663                         } else {
5664                                 out.bmv_block =
5665                                     (map[i].br_startblock == HOLESTARTBLOCK) ?
5666                                         -1 :
5667                                         XFS_FSB_TO_DB(ip, map[i].br_startblock);
5668
5669                                 /* return either getbmap/getbmapx structure. */
5670                                 if (interface & BMV_IF_EXTENDED) {
5671                                         struct  getbmapx        outx;
5672
5673                                         GETBMAP_CONVERT(out,outx);
5674                                         outx.bmv_oflags = oflags;
5675                                         outx.bmv_unused1 = outx.bmv_unused2 = 0;
5676                                         if (copy_to_user(ap, &outx,
5677                                                         sizeof(outx))) {
5678                                                 error = XFS_ERROR(EFAULT);
5679                                                 goto unlock_and_return;
5680                                         }
5681                                 } else {
5682                                         if (copy_to_user(ap, &out,
5683                                                         sizeof(out))) {
5684                                                 error = XFS_ERROR(EFAULT);
5685                                                 goto unlock_and_return;
5686                                         }
5687                                 }
5688                                 bmv->bmv_offset =
5689                                         out.bmv_offset + out.bmv_length;
5690                                 bmv->bmv_length = MAX((__int64_t)0,
5691                                         (__int64_t)(bmvend - bmv->bmv_offset));
5692                                 bmv->bmv_entries++;
5693                                 ap = (interface & BMV_IF_EXTENDED) ?
5694                                                 (void __user *)
5695                                         ((struct getbmapx __user *)ap + 1) :
5696                                                 (void __user *)
5697                                         ((struct getbmap __user *)ap + 1);
5698                         }
5699                 }
5700         } while (nmap && nexleft && bmv->bmv_length);
5701
5702 unlock_and_return:
5703         xfs_iunlock_map_shared(ip, lock);
5704         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5705
5706         kmem_free(map, subnex * sizeof(*map));
5707
5708         return error;
5709 }
5710
5711 /*
5712  * Check the last inode extent to determine whether this allocation will result
5713  * in blocks being allocated at the end of the file. When we allocate new data
5714  * blocks at the end of the file which do not start at the previous data block,
5715  * we will try to align the new blocks at stripe unit boundaries.
5716  */
5717 int                                     /* error */
5718 xfs_bmap_isaeof(
5719         xfs_inode_t     *ip,            /* incore inode pointer */
5720         xfs_fileoff_t   off,            /* file offset in fsblocks */
5721         int             whichfork,      /* data or attribute fork */
5722         char            *aeof)          /* return value */
5723 {
5724         int             error;          /* error return value */
5725         xfs_ifork_t     *ifp;           /* inode fork pointer */
5726         xfs_bmbt_rec_t  *lastrec;       /* extent list entry pointer */
5727         xfs_extnum_t    nextents;       /* size of extent list */
5728         xfs_bmbt_irec_t s;              /* expanded extent list entry */
5729
5730         ASSERT(whichfork == XFS_DATA_FORK);
5731         ifp = XFS_IFORK_PTR(ip, whichfork);
5732         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5733             (error = xfs_iread_extents(NULL, ip, whichfork)))
5734                 return error;
5735         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5736         if (nextents == 0) {
5737                 *aeof = 1;
5738                 return 0;
5739         }
5740         /*
5741          * Go to the last extent
5742          */
5743         lastrec = &ifp->if_u1.if_extents[nextents - 1];
5744         xfs_bmbt_get_all(lastrec, &s);
5745         /*
5746          * Check we are allocating in the last extent (for delayed allocations)
5747          * or past the last extent for non-delayed allocations.
5748          */
5749         *aeof = (off >= s.br_startoff &&
5750                  off < s.br_startoff + s.br_blockcount &&
5751                  ISNULLSTARTBLOCK(s.br_startblock)) ||
5752                 off >= s.br_startoff + s.br_blockcount;
5753         return 0;
5754 }
5755
5756 /*
5757  * Check if the endoff is outside the last extent. If so the caller will grow
5758  * the allocation to a stripe unit boundary.
5759  */
5760 int                                     /* error */
5761 xfs_bmap_eof(
5762         xfs_inode_t     *ip,            /* incore inode pointer */
5763         xfs_fileoff_t   endoff,         /* file offset in fsblocks */
5764         int             whichfork,      /* data or attribute fork */
5765         int             *eof)           /* result value */
5766 {
5767         xfs_fsblock_t   blockcount;     /* extent block count */
5768         int             error;          /* error return value */
5769         xfs_ifork_t     *ifp;           /* inode fork pointer */
5770         xfs_bmbt_rec_t  *lastrec;       /* extent list entry pointer */
5771         xfs_extnum_t    nextents;       /* size of extent list */
5772         xfs_fileoff_t   startoff;       /* extent starting file offset */
5773
5774         ASSERT(whichfork == XFS_DATA_FORK);
5775         ifp = XFS_IFORK_PTR(ip, whichfork);
5776         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5777             (error = xfs_iread_extents(NULL, ip, whichfork)))
5778                 return error;
5779         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5780         if (nextents == 0) {
5781                 *eof = 1;
5782                 return 0;
5783         }
5784         /*
5785          * Go to the last extent
5786          */
5787         lastrec = &ifp->if_u1.if_extents[nextents - 1];
5788         startoff = xfs_bmbt_get_startoff(lastrec);
5789         blockcount = xfs_bmbt_get_blockcount(lastrec);
5790         *eof = endoff >= startoff + blockcount;
5791         return 0;
5792 }
5793
5794 #ifdef DEBUG
5795 /*
5796  * Check that the extents list for the inode ip is in the right order.
5797  */
5798 STATIC void
5799 xfs_bmap_check_extents(
5800         xfs_inode_t             *ip,            /* incore inode pointer */
5801         int                     whichfork)      /* data or attr fork */
5802 {
5803         xfs_bmbt_rec_t          *base;          /* base of extents list */
5804         xfs_bmbt_rec_t          *ep;            /* current extent entry */
5805         xfs_ifork_t             *ifp;           /* inode fork pointer */
5806         xfs_extnum_t            nextents;       /* number of extents in list */
5807
5808         ifp = XFS_IFORK_PTR(ip, whichfork);
5809         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
5810         base = ifp->if_u1.if_extents;
5811         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5812         for (ep = base; ep < &base[nextents - 1]; ep++) {
5813                 xfs_btree_check_rec(XFS_BTNUM_BMAP, (void *)ep,
5814                         (void *)(ep + 1));
5815         }
5816 }
5817
5818 STATIC
5819 xfs_buf_t *
5820 xfs_bmap_get_bp(
5821         xfs_btree_cur_t         *cur,
5822         xfs_fsblock_t           bno)
5823 {
5824         int i;
5825         xfs_buf_t *bp;
5826
5827         if (!cur)
5828                 return(NULL);
5829
5830         bp = NULL;
5831         for(i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
5832                 bp = cur->bc_bufs[i];
5833                 if (!bp) break;
5834                 if (XFS_BUF_ADDR(bp) == bno)
5835                         break;  /* Found it */
5836         }
5837         if (i == XFS_BTREE_MAXLEVELS)
5838                 bp = NULL;
5839
5840         if (!bp) { /* Chase down all the log items to see if the bp is there */
5841                 xfs_log_item_chunk_t    *licp;
5842                 xfs_trans_t             *tp;
5843
5844                 tp = cur->bc_tp;
5845                 licp = &tp->t_items;
5846                 while (!bp && licp != NULL) {
5847                         if (XFS_LIC_ARE_ALL_FREE(licp)) {
5848                                 licp = licp->lic_next;
5849                                 continue;
5850                         }
5851                         for (i = 0; i < licp->lic_unused; i++) {
5852                                 xfs_log_item_desc_t     *lidp;
5853                                 xfs_log_item_t          *lip;
5854                                 xfs_buf_log_item_t      *bip;
5855                                 xfs_buf_t               *lbp;
5856
5857                                 if (XFS_LIC_ISFREE(licp, i)) {
5858                                         continue;
5859                                 }
5860
5861                                 lidp = XFS_LIC_SLOT(licp, i);
5862                                 lip = lidp->lid_item;
5863                                 if (lip->li_type != XFS_LI_BUF)
5864                                         continue;
5865
5866                                 bip = (xfs_buf_log_item_t *)lip;
5867                                 lbp = bip->bli_buf;
5868
5869                                 if (XFS_BUF_ADDR(lbp) == bno) {
5870                                         bp = lbp;
5871                                         break; /* Found it */
5872                                 }
5873                         }
5874                         licp = licp->lic_next;
5875                 }
5876         }
5877         return(bp);
5878 }
5879
5880 void
5881 xfs_check_block(
5882         xfs_bmbt_block_t        *block,
5883         xfs_mount_t             *mp,
5884         int                     root,
5885         short                   sz)
5886 {
5887         int                     i, j, dmxr;
5888         xfs_bmbt_ptr_t          *pp, *thispa;   /* pointer to block address */
5889         xfs_bmbt_key_t          *prevp, *keyp;
5890
5891         ASSERT(INT_GET(block->bb_level, ARCH_CONVERT) > 0);
5892
5893         prevp = NULL;
5894         for( i = 1; i <= INT_GET(block->bb_numrecs, ARCH_CONVERT);i++) {
5895                 dmxr = mp->m_bmap_dmxr[0];
5896
5897                 if (root) {
5898                         keyp = XFS_BMAP_BROOT_KEY_ADDR(block, i, sz);
5899                 } else {
5900                         keyp = XFS_BTREE_KEY_ADDR(mp->m_sb.sb_blocksize,
5901                                 xfs_bmbt, block, i, dmxr);
5902                 }
5903
5904                 if (prevp) {
5905                         xfs_btree_check_key(XFS_BTNUM_BMAP, prevp, keyp);
5906                 }
5907                 prevp = keyp;
5908
5909                 /*
5910                  * Compare the block numbers to see if there are dups.
5911                  */
5912
5913                 if (root) {
5914                         pp = XFS_BMAP_BROOT_PTR_ADDR(block, i, sz);
5915                 } else {
5916                         pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize,
5917                                 xfs_bmbt, block, i, dmxr);
5918                 }
5919                 for (j = i+1; j <= INT_GET(block->bb_numrecs, ARCH_CONVERT); j++) {
5920                         if (root) {
5921                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(block, j, sz);
5922                         } else {
5923                                 thispa = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize,
5924                                         xfs_bmbt, block, j, dmxr);
5925                         }
5926                         if (INT_GET(*thispa, ARCH_CONVERT) ==
5927                             INT_GET(*pp, ARCH_CONVERT)) {
5928                                 cmn_err(CE_WARN, "%s: thispa(%d) == pp(%d) %Ld",
5929                                         __FUNCTION__, j, i,
5930                                         INT_GET(*thispa, ARCH_CONVERT));
5931                                 panic("%s: ptrs are equal in node\n",
5932                                         __FUNCTION__);
5933                         }
5934                 }
5935         }
5936 }
5937
5938 /*
5939  * Check that the extents for the inode ip are in the right order in all
5940  * btree leaves.
5941  */
5942
5943 STATIC void
5944 xfs_bmap_check_leaf_extents(
5945         xfs_btree_cur_t         *cur,   /* btree cursor or null */
5946         xfs_inode_t             *ip,            /* incore inode pointer */
5947         int                     whichfork)      /* data or attr fork */
5948 {
5949         xfs_bmbt_block_t        *block; /* current btree block */
5950         xfs_fsblock_t           bno;    /* block # of "block" */
5951         xfs_buf_t               *bp;    /* buffer for "block" */
5952         int                     error;  /* error return value */
5953         xfs_extnum_t            i=0;    /* index into the extents list */
5954         xfs_ifork_t             *ifp;   /* fork structure */
5955         int                     level;  /* btree level, for checking */
5956         xfs_mount_t             *mp;    /* file system mount structure */
5957         xfs_bmbt_ptr_t          *pp;    /* pointer to block address */
5958         xfs_bmbt_rec_t          *ep, *lastp;    /* extent pointers in block entry */
5959         int                     bp_release = 0;
5960
5961         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
5962                 return;
5963         }
5964
5965         bno = NULLFSBLOCK;
5966         mp = ip->i_mount;
5967         ifp = XFS_IFORK_PTR(ip, whichfork);
5968         block = ifp->if_broot;
5969         /*
5970          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5971          */
5972         ASSERT(INT_GET(block->bb_level, ARCH_CONVERT) > 0);
5973         level = INT_GET(block->bb_level, ARCH_CONVERT);
5974         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
5975         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
5976         ASSERT(INT_GET(*pp, ARCH_CONVERT) != NULLDFSBNO);
5977         ASSERT(XFS_FSB_TO_AGNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agcount);
5978         ASSERT(XFS_FSB_TO_AGBNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agblocks);
5979         bno = INT_GET(*pp, ARCH_CONVERT);
5980         /*
5981          * Go down the tree until leaf level is reached, following the first
5982          * pointer (leftmost) at each level.
5983          */
5984         while (level-- > 0) {
5985                 /* See if buf is in cur first */
5986                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5987                 if (bp) {
5988                         bp_release = 0;
5989                 } else {
5990                         bp_release = 1;
5991                 }
5992                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5993                                 XFS_BMAP_BTREE_REF)))
5994                         goto error_norelse;
5995                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
5996                 XFS_WANT_CORRUPTED_GOTO(
5997                         XFS_BMAP_SANITY_CHECK(mp, block, level),
5998                         error0);
5999                 if (level == 0)
6000                         break;
6001
6002                 /*
6003                  * Check this block for basic sanity (increasing keys and
6004                  * no duplicate blocks).
6005                  */
6006
6007                 xfs_check_block(block, mp, 0, 0);
6008                 pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt, block,
6009                         1, mp->m_bmap_dmxr[1]);
6010                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, INT_GET(*pp, ARCH_CONVERT)), error0);
6011                 bno = INT_GET(*pp, ARCH_CONVERT);
6012                 if (bp_release) {
6013                         bp_release = 0;
6014                         xfs_trans_brelse(NULL, bp);
6015                 }
6016         }
6017
6018         /*
6019          * Here with bp and block set to the leftmost leaf node in the tree.
6020          */
6021         i = 0;
6022
6023         /*
6024          * Loop over all leaf nodes checking that all extents are in the right order.
6025          */
6026         lastp = NULL;
6027         for (;;) {
6028                 xfs_bmbt_rec_t  *frp;
6029                 xfs_fsblock_t   nextbno;
6030                 xfs_extnum_t    num_recs;
6031
6032
6033                 num_recs = INT_GET(block->bb_numrecs, ARCH_CONVERT);
6034
6035                 /*
6036                  * Read-ahead the next leaf block, if any.
6037                  */
6038
6039                 nextbno = INT_GET(block->bb_rightsib, ARCH_CONVERT);
6040
6041                 /*
6042                  * Check all the extents to make sure they are OK.
6043                  * If we had a previous block, the last entry should
6044                  * conform with the first entry in this one.
6045                  */
6046
6047                 frp = XFS_BTREE_REC_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt,
6048                         block, 1, mp->m_bmap_dmxr[0]);
6049
6050                 for (ep = frp;ep < frp + (num_recs - 1); ep++) {
6051                         if (lastp) {
6052                                 xfs_btree_check_rec(XFS_BTNUM_BMAP,
6053                                         (void *)lastp, (void *)ep);
6054                         }
6055                         xfs_btree_check_rec(XFS_BTNUM_BMAP, (void *)ep,
6056                                 (void *)(ep + 1));
6057                 }
6058                 lastp = frp + num_recs - 1; /* For the next iteration */
6059
6060                 i += num_recs;
6061                 if (bp_release) {
6062                         bp_release = 0;
6063                         xfs_trans_brelse(NULL, bp);
6064                 }
6065                 bno = nextbno;
6066                 /*
6067                  * If we've reached the end, stop.
6068                  */
6069                 if (bno == NULLFSBLOCK)
6070                         break;
6071
6072                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6073                 if (bp) {
6074                         bp_release = 0;
6075                 } else {
6076                         bp_release = 1;
6077                 }
6078                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6079                                 XFS_BMAP_BTREE_REF)))
6080                         goto error_norelse;
6081                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6082         }
6083         if (bp_release) {
6084                 bp_release = 0;
6085                 xfs_trans_brelse(NULL, bp);
6086         }
6087         return;
6088
6089 error0:
6090         cmn_err(CE_WARN, "%s: at error0", __FUNCTION__);
6091         if (bp_release)
6092                 xfs_trans_brelse(NULL, bp);
6093 error_norelse:
6094         cmn_err(CE_WARN, "%s: BAD after btree leaves for %d extents",
6095                 i, __FUNCTION__);
6096         panic("%s: CORRUPTED BTREE OR SOMETHING", __FUNCTION__);
6097         return;
6098 }
6099 #endif
6100
6101 /*
6102  * Count fsblocks of the given fork.
6103  */
6104 int                                             /* error */
6105 xfs_bmap_count_blocks(
6106         xfs_trans_t             *tp,            /* transaction pointer */
6107         xfs_inode_t             *ip,            /* incore inode */
6108         int                     whichfork,      /* data or attr fork */
6109         int                     *count)         /* out: count of blocks */
6110 {
6111         xfs_bmbt_block_t        *block; /* current btree block */
6112         xfs_fsblock_t           bno;    /* block # of "block" */
6113         xfs_ifork_t             *ifp;   /* fork structure */
6114         int                     level;  /* btree level, for checking */
6115         xfs_mount_t             *mp;    /* file system mount structure */
6116         xfs_bmbt_ptr_t          *pp;    /* pointer to block address */
6117
6118         bno = NULLFSBLOCK;
6119         mp = ip->i_mount;
6120         ifp = XFS_IFORK_PTR(ip, whichfork);
6121         if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
6122                 if (unlikely(xfs_bmap_count_leaves(ifp->if_u1.if_extents,
6123                         ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
6124                         count) < 0)) {
6125                         XFS_ERROR_REPORT("xfs_bmap_count_blocks(1)",
6126                                          XFS_ERRLEVEL_LOW, mp);
6127                         return XFS_ERROR(EFSCORRUPTED);
6128                 }
6129                 return 0;
6130         }
6131
6132         /*
6133          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6134          */
6135         block = ifp->if_broot;
6136         ASSERT(INT_GET(block->bb_level, ARCH_CONVERT) > 0);
6137         level = INT_GET(block->bb_level, ARCH_CONVERT);
6138         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
6139         ASSERT(INT_GET(*pp, ARCH_CONVERT) != NULLDFSBNO);
6140         ASSERT(XFS_FSB_TO_AGNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agcount);
6141         ASSERT(XFS_FSB_TO_AGBNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agblocks);
6142         bno = INT_GET(*pp, ARCH_CONVERT);
6143
6144         if (unlikely(xfs_bmap_count_tree(mp, tp, bno, level, count) < 0)) {
6145                 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
6146                                  mp);
6147                 return XFS_ERROR(EFSCORRUPTED);
6148         }
6149
6150         return 0;
6151 }
6152
6153 /*
6154  * Recursively walks each level of a btree
6155  * to count total fsblocks is use.
6156  */
6157 int                                     /* error */
6158 xfs_bmap_count_tree(
6159         xfs_mount_t     *mp,            /* file system mount point */
6160         xfs_trans_t     *tp,            /* transaction pointer */
6161         xfs_fsblock_t   blockno,        /* file system block number */
6162         int             levelin,        /* level in btree */
6163         int             *count)         /* Count of blocks */
6164 {
6165         int                     error;
6166         xfs_buf_t               *bp, *nbp;
6167         int                     level = levelin;
6168         xfs_bmbt_ptr_t          *pp;
6169         xfs_fsblock_t           bno = blockno;
6170         xfs_fsblock_t           nextbno;
6171         xfs_bmbt_block_t        *block, *nextblock;
6172         int                     numrecs;
6173         xfs_bmbt_rec_t          *frp;
6174
6175         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
6176                 return error;
6177         *count += 1;
6178         block = XFS_BUF_TO_BMBT_BLOCK(bp);
6179
6180         if (--level) {
6181                 /* Not at node above leafs, count this level of nodes */
6182                 nextbno = INT_GET(block->bb_rightsib, ARCH_CONVERT);
6183                 while (nextbno != NULLFSBLOCK) {
6184                         if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
6185                                 0, &nbp, XFS_BMAP_BTREE_REF)))
6186                                 return error;
6187                         *count += 1;
6188                         nextblock = XFS_BUF_TO_BMBT_BLOCK(nbp);
6189                         nextbno = INT_GET(nextblock->bb_rightsib, ARCH_CONVERT);
6190                         xfs_trans_brelse(tp, nbp);
6191                 }
6192
6193                 /* Dive to the next level */
6194                 pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize,
6195                         xfs_bmbt, block, 1, mp->m_bmap_dmxr[1]);
6196                 bno = INT_GET(*pp, ARCH_CONVERT);
6197                 if (unlikely((error =
6198                      xfs_bmap_count_tree(mp, tp, bno, level, count)) < 0)) {
6199                         xfs_trans_brelse(tp, bp);
6200                         XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6201                                          XFS_ERRLEVEL_LOW, mp);
6202                         return XFS_ERROR(EFSCORRUPTED);
6203                 }
6204                 xfs_trans_brelse(tp, bp);
6205         } else {
6206                 /* count all level 1 nodes and their leaves */
6207                 for (;;) {
6208                         nextbno = INT_GET(block->bb_rightsib, ARCH_CONVERT);
6209                         numrecs = INT_GET(block->bb_numrecs, ARCH_CONVERT);
6210                         frp = XFS_BTREE_REC_ADDR(mp->m_sb.sb_blocksize,
6211                                 xfs_bmbt, block, 1, mp->m_bmap_dmxr[0]);
6212                         if (unlikely(xfs_bmap_count_leaves(frp, numrecs, count) < 0)) {
6213                                 xfs_trans_brelse(tp, bp);
6214                                 XFS_ERROR_REPORT("xfs_bmap_count_tree(2)",
6215                                                  XFS_ERRLEVEL_LOW, mp);
6216                                 return XFS_ERROR(EFSCORRUPTED);
6217                         }
6218                         xfs_trans_brelse(tp, bp);
6219                         if (nextbno == NULLFSBLOCK)
6220                                 break;
6221                         bno = nextbno;
6222                         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6223                                 XFS_BMAP_BTREE_REF)))
6224                                 return error;
6225                         *count += 1;
6226                         block = XFS_BUF_TO_BMBT_BLOCK(bp);
6227                 }
6228         }
6229         return 0;
6230 }
6231
6232 /*
6233  * Count leaf blocks given a pointer to an extent list.
6234  */
6235 int
6236 xfs_bmap_count_leaves(
6237         xfs_bmbt_rec_t          *frp,
6238         int                     numrecs,
6239         int                     *count)
6240 {
6241         int             b;
6242
6243         for ( b = 1; b <= numrecs; b++, frp++)
6244                 *count += xfs_bmbt_disk_get_blockcount(frp);
6245         return 0;
6246 }