fs/xfs/xfs_attr.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_alloc.h"
  50 #include "xfs_btree.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_bmap.h"
  58 #include "xfs_da_btree.h"
  59 #include "xfs_attr.h"
  60 #include "xfs_attr_leaf.h"
  61 #include "xfs_error.h"
  62 #include "xfs_bit.h"
  63 #include "xfs_quota.h"
  64 #include "xfs_rw.h"
  65 #include "xfs_trans_space.h"
  66 #include "xfs_acl.h"
  67
  68 /*
  69  * xfs_attr.c
  70  *
  71  * Provide the external interfaces to manage attribute lists.
  72  */
  73
  74 /*========================================================================
  75  * Function prototypes for the kernel.
  76  *========================================================================*/
  77
  78 /*
  79  * Internal routines when attribute list fits inside the inode.
  80  */
  81 STATIC int xfs_attr_shortform_addname(xfs_da_args_t *args);
  82
  83 /*
  84  * Internal routines when attribute list is one block.
  85  */
  86 STATIC int xfs_attr_leaf_addname(xfs_da_args_t *args);
  87 STATIC int xfs_attr_leaf_removename(xfs_da_args_t *args);
  88 STATIC int xfs_attr_leaf_list(xfs_attr_list_context_t *context);
  89
  90 /*
  91  * Internal routines when attribute list is more than one block.
  92  */
  93 STATIC int xfs_attr_node_addname(xfs_da_args_t *args);
  94 STATIC int xfs_attr_node_removename(xfs_da_args_t *args);
  95 STATIC int xfs_attr_node_list(xfs_attr_list_context_t *context);
  96 STATIC int xfs_attr_fillstate(xfs_da_state_t *state);
  97 STATIC int xfs_attr_refillstate(xfs_da_state_t *state);
  98
  99 /*
 100  * Routines to manipulate out-of-line attribute values.
 101  */
 102 STATIC int xfs_attr_rmtval_get(xfs_da_args_t *args);
 103 STATIC int xfs_attr_rmtval_set(xfs_da_args_t *args);
 104 STATIC int xfs_attr_rmtval_remove(xfs_da_args_t *args);
 105
 106 #define ATTR_RMTVALUE_MAPSIZE   1       /* # of map entries at once */
 107
 108 #if defined(XFS_ATTR_TRACE)
 109 ktrace_t *xfs_attr_trace_buf;
 110 #endif
 111
 112
 113 /*========================================================================
 114  * Overall external interface routines.
 115  *========================================================================*/
 116
 117 int
 118 xfs_attr_fetch(xfs_inode_t *ip, char *name, int namelen,
 119                char *value, int *valuelenp, int flags, struct cred *cred)
 120 {
 121         xfs_da_args_t   args;
 122         int             error;
 123
 124         if ((XFS_IFORK_Q(ip) == 0) ||
 125             (ip->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 126              ip->i_d.di_anextents == 0))
 127                 return(ENOATTR);
 128
 129         if (!(flags & (ATTR_KERNACCESS|ATTR_SECURE))) {
 130                 if ((error = xfs_iaccess(ip, S_IRUSR, cred)))
 131                         return(XFS_ERROR(error));
 132         }
 133
 134         /*
 135          * Fill in the arg structure for this request.
 136          */
 137         memset((char *)&args, 0, sizeof(args));
 138         args.name = name;
 139         args.namelen = namelen;
 140         args.value = value;
 141         args.valuelen = *valuelenp;
 142         args.flags = flags;
 143         args.hashval = xfs_da_hashname(args.name, args.namelen);
 144         args.dp = ip;
 145         args.whichfork = XFS_ATTR_FORK;
 146
 147         /*
 148          * Decide on what work routines to call based on the inode size.
 149          */
 150         if (XFS_IFORK_Q(ip) == 0 ||
 151             (ip->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 152              ip->i_d.di_anextents == 0)) {
 153                 error = XFS_ERROR(ENOATTR);
 154         } else if (ip->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
 155                 error = xfs_attr_shortform_getvalue(&args);
 156         } else if (xfs_bmap_one_block(ip, XFS_ATTR_FORK)) {
 157                 error = xfs_attr_leaf_get(&args);
 158         } else {
 159                 error = xfs_attr_node_get(&args);
 160         }
 161
 162         /*
 163          * Return the number of bytes in the value to the caller.
 164          */
 165         *valuelenp = args.valuelen;
 166
 167         if (error == EEXIST)
 168                 error = 0;
 169         return(error);
 170 }
 171
 172 int
 173 xfs_attr_get(bhv_desc_t *bdp, char *name, char *value, int *valuelenp,
 174              int flags, struct cred *cred)
 175 {
 176         xfs_inode_t     *ip = XFS_BHVTOI(bdp);
 177         int             error, namelen;
 178
 179         XFS_STATS_INC(xs_attr_get);
 180
 181         if (!name)
 182                 return(EINVAL);
 183         namelen = strlen(name);
 184         if (namelen >= MAXNAMELEN)
 185                 return(EFAULT);         /* match IRIX behaviour */
 186
 187         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
 188                 return(EIO);
 189
 190         xfs_ilock(ip, XFS_ILOCK_SHARED);
 191         error = xfs_attr_fetch(ip, name, namelen, value, valuelenp, flags, cred);
 192         xfs_iunlock(ip, XFS_ILOCK_SHARED);
 193         return(error);
 194 }
 195
 196 /*ARGSUSED*/
 197 int                                                             /* error */
 198 xfs_attr_set(bhv_desc_t *bdp, char *name, char *value, int valuelen, int flags,
 199                      struct cred *cred)
 200 {
 201         xfs_da_args_t   args;
 202         xfs_inode_t     *dp;
 203         xfs_fsblock_t   firstblock;
 204         xfs_bmap_free_t flist;
 205         int             error, err2, committed;
 206         int             local, size;
 207         uint            nblks;
 208         xfs_mount_t     *mp;
 209         int             rsvd = (flags & ATTR_ROOT) != 0;
 210         int             namelen;
 211
 212         namelen = strlen(name);
 213         if (namelen >= MAXNAMELEN)
 214                 return EFAULT;          /* match IRIX behaviour */
 215
 216         XFS_STATS_INC(xs_attr_set);
 217
 218         dp = XFS_BHVTOI(bdp);
 219         mp = dp->i_mount;
 220         if (XFS_FORCED_SHUTDOWN(mp))
 221                 return (EIO);
 222
 223         xfs_ilock(dp, XFS_ILOCK_SHARED);
 224         if (!(flags & ATTR_SECURE) &&
 225              (error = xfs_iaccess(dp, S_IWUSR, cred))) {
 226                 xfs_iunlock(dp, XFS_ILOCK_SHARED);
 227                 return(XFS_ERROR(error));
 228         }
 229         xfs_iunlock(dp, XFS_ILOCK_SHARED);
 230
 231         /*
 232          * Attach the dquots to the inode.
 233          */
 234         if ((error = XFS_QM_DQATTACH(mp, dp, 0)))
 235                 return (error);
 236
 237         /*
 238          * If the inode doesn't have an attribute fork, add one.
 239          * (inode must not be locked when we call this routine)
 240          */
 241         if (XFS_IFORK_Q(dp) == 0) {
 242                 error = xfs_bmap_add_attrfork(dp, rsvd);
 243                 if (error)
 244                         return(error);
 245         }
 246
 247         /*
 248          * Fill in the arg structure for this request.
 249          */
 250         memset((char *)&args, 0, sizeof(args));
 251         args.name = name;
 252         args.namelen = namelen;
 253         args.value = value;
 254         args.valuelen = valuelen;
 255         args.flags = flags;
 256         args.hashval = xfs_da_hashname(args.name, args.namelen);
 257         args.dp = dp;
 258         args.firstblock = &firstblock;
 259         args.flist = &flist;
 260         args.whichfork = XFS_ATTR_FORK;
 261         args.oknoent = 1;
 262
 263         /* Determine space new attribute will use, and if it will be inline
 264          * or out of line.
 265          */
 266         size = xfs_attr_leaf_newentsize(&args, mp->m_sb.sb_blocksize, &local);
 267
 268         nblks = XFS_DAENTER_SPACE_RES(mp, XFS_ATTR_FORK);
 269         if (local) {
 270                 if (size > (mp->m_sb.sb_blocksize >> 1)) {
 271                         /* Double split possible */
 272                         nblks <<= 1;
 273                 }
 274         } else {
 275                 uint    dblocks = XFS_B_TO_FSB(mp, valuelen);
 276                 /* Out of line attribute, cannot double split, but make
 277                  * room for the attribute value itself.
 278                  */
 279                 nblks += dblocks;
 280                 nblks += XFS_NEXTENTADD_SPACE_RES(mp, dblocks, XFS_ATTR_FORK);
 281         }
 282
 283         /* Size is now blocks for attribute data */
 284         args.total = nblks;
 285
 286         /*
 287          * Start our first transaction of the day.
 288          *
 289          * All future transactions during this code must be "chained" off
 290          * this one via the trans_dup() call.  All transactions will contain
 291          * the inode, and the inode will always be marked with trans_ihold().
 292          * Since the inode will be locked in all transactions, we must log
 293          * the inode in every transaction to let it float upward through
 294          * the log.
 295          */
 296         args.trans = xfs_trans_alloc(mp, XFS_TRANS_ATTR_SET);
 297
 298         /*
 299          * Root fork attributes can use reserved data blocks for this
 300          * operation if necessary
 301          */
 302
 303         if (rsvd)
 304                 args.trans->t_flags |= XFS_TRANS_RESERVE;
 305
 306         if ((error = xfs_trans_reserve(args.trans, (uint) nblks,
 307                                       XFS_ATTRSET_LOG_RES(mp, nblks),
 308                                       0, XFS_TRANS_PERM_LOG_RES,
 309                                       XFS_ATTRSET_LOG_COUNT))) {
 310                 xfs_trans_cancel(args.trans, 0);
 311                 return(error);
 312         }
 313         xfs_ilock(dp, XFS_ILOCK_EXCL);
 314
 315         error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, args.trans, dp, nblks, 0,
 316                          rsvd ? XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
 317                                 XFS_QMOPT_RES_REGBLKS);
 318         if (error) {
 319                 xfs_iunlock(dp, XFS_ILOCK_EXCL);
 320                 xfs_trans_cancel(args.trans, XFS_TRANS_RELEASE_LOG_RES);
 321                 return (error);
 322         }
 323
 324         xfs_trans_ijoin(args.trans, dp, XFS_ILOCK_EXCL);
 325         xfs_trans_ihold(args.trans, dp);
 326
 327         /*
 328          * If the attribute list is non-existant or a shortform list,
 329          * upgrade it to a single-leaf-block attribute list.
 330          */
 331         if ((dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) ||
 332             ((dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) &&
 333              (dp->i_d.di_anextents == 0))) {
 334
 335                 /*
 336                  * Build initial attribute list (if required).
 337                  */
 338                 if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS)
 339                         (void)xfs_attr_shortform_create(&args);
 340
 341                 /*
 342                  * Try to add the attr to the attribute list in
 343                  * the inode.
 344                  */
 345                 error = xfs_attr_shortform_addname(&args);
 346                 if (error != ENOSPC) {
 347                         /*
 348                          * Commit the shortform mods, and we're done.
 349                          * NOTE: this is also the error path (EEXIST, etc).
 350                          */
 351                         ASSERT(args.trans != NULL);
 352
 353                         /*
 354                          * If this is a synchronous mount, make sure that
 355                          * the transaction goes to disk before returning
 356                          * to the user.
 357                          */
 358                         if (mp->m_flags & XFS_MOUNT_WSYNC) {
 359                                 xfs_trans_set_sync(args.trans);
 360                         }
 361                         err2 = xfs_trans_commit(args.trans,
 362                                                  XFS_TRANS_RELEASE_LOG_RES,
 363                                                  NULL);
 364                         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 365
 366                         /*
 367                          * Hit the inode change time.
 368                          */
 369                         if (!error && (flags & ATTR_KERNOTIME) == 0) {
 370                                 xfs_ichgtime(dp, XFS_ICHGTIME_CHG);
 371                         }
 372                         return(error == 0 ? err2 : error);
 373                 }
 374
 375                 /*
 376                  * It won't fit in the shortform, transform to a leaf block.
 377                  * GROT: another possible req'mt for a double-split btree op.
 378                  */
 379                 XFS_BMAP_INIT(args.flist, args.firstblock);
 380                 error = xfs_attr_shortform_to_leaf(&args);
 381                 if (!error) {
 382                         error = xfs_bmap_finish(&args.trans, args.flist,
 383                                                 *args.firstblock, &committed);
 384                 }
 385                 if (error) {
 386                         ASSERT(committed);
 387                         args.trans = NULL;
 388                         xfs_bmap_cancel(&flist);
 389                         goto out;
 390                 }
 391
 392                 /*
 393                  * bmap_finish() may have committed the last trans and started
 394                  * a new one.  We need the inode to be in all transactions.
 395                  */
 396                 if (committed) {
 397                         xfs_trans_ijoin(args.trans, dp, XFS_ILOCK_EXCL);
 398                         xfs_trans_ihold(args.trans, dp);
 399                 }
 400
 401                 /*
 402                  * Commit the leaf transformation.  We'll need another (linked)
 403                  * transaction to add the new attribute to the leaf.
 404                  */
 405                 if ((error = xfs_attr_rolltrans(&args.trans, dp)))
 406                         goto out;
 407
 408         }
 409
 410         if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
 411                 error = xfs_attr_leaf_addname(&args);
 412         } else {
 413                 error = xfs_attr_node_addname(&args);
 414         }
 415         if (error) {
 416                 goto out;
 417         }
 418
 419         /*
 420          * If this is a synchronous mount, make sure that the
 421          * transaction goes to disk before returning to the user.
 422          */
 423         if (mp->m_flags & XFS_MOUNT_WSYNC) {
 424                 xfs_trans_set_sync(args.trans);
 425         }
 426
 427         /*
 428          * Commit the last in the sequence of transactions.
 429          */
 430         xfs_trans_log_inode(args.trans, dp, XFS_ILOG_CORE);
 431         error = xfs_trans_commit(args.trans, XFS_TRANS_RELEASE_LOG_RES,
 432                                  NULL);
 433         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 434
 435         /*
 436          * Hit the inode change time.
 437          */
 438         if (!error && (flags & ATTR_KERNOTIME) == 0) {
 439                 xfs_ichgtime(dp, XFS_ICHGTIME_CHG);
 440         }
 441
 442         return(error);
 443
 444 out:
 445         if (args.trans)
 446                 xfs_trans_cancel(args.trans,
 447                         XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
 448         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 449         return(error);
 450 }
 451
 452 /*
 453  * Generic handler routine to remove a name from an attribute list.
 454  * Transitions attribute list from Btree to shortform as necessary.
 455  */
 456 /*ARGSUSED*/
 457 int                                                             /* error */
 458 xfs_attr_remove(bhv_desc_t *bdp, char *name, int flags, struct cred *cred)
 459 {
 460         xfs_da_args_t       args;
 461         xfs_inode_t         *dp;
 462         xfs_fsblock_t       firstblock;
 463         xfs_bmap_free_t     flist;
 464         int                 error;
 465         xfs_mount_t         *mp;
 466         int                 namelen;
 467
 468         ASSERT(MAXNAMELEN-1<=0xff); /* length is stored in uint8 */
 469         namelen = strlen(name);
 470         if (namelen>=MAXNAMELEN)
 471                 return EFAULT; /* match irix behaviour */
 472
 473         XFS_STATS_INC(xs_attr_remove);
 474
 475         dp = XFS_BHVTOI(bdp);
 476         mp = dp->i_mount;
 477         if (XFS_FORCED_SHUTDOWN(mp))
 478                 return (EIO);
 479
 480         xfs_ilock(dp, XFS_ILOCK_SHARED);
 481         if (!(flags & ATTR_SECURE) &&
 482              (error = xfs_iaccess(dp, S_IWUSR, cred))) {
 483                 xfs_iunlock(dp, XFS_ILOCK_SHARED);
 484                 return(XFS_ERROR(error));
 485         } else if (XFS_IFORK_Q(dp) == 0 ||
 486                    (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 487                     dp->i_d.di_anextents == 0)) {
 488                 xfs_iunlock(dp, XFS_ILOCK_SHARED);
 489                 return(XFS_ERROR(ENOATTR));
 490         }
 491         xfs_iunlock(dp, XFS_ILOCK_SHARED);
 492
 493         /*
 494          * Fill in the arg structure for this request.
 495          */
 496         memset((char *)&args, 0, sizeof(args));
 497         args.name = name;
 498         args.namelen = namelen;
 499         args.flags = flags;
 500         args.hashval = xfs_da_hashname(args.name, args.namelen);
 501         args.dp = dp;
 502         args.firstblock = &firstblock;
 503         args.flist = &flist;
 504         args.total = 0;
 505         args.whichfork = XFS_ATTR_FORK;
 506
 507         /*
 508          * Attach the dquots to the inode.
 509          */
 510         if ((error = XFS_QM_DQATTACH(mp, dp, 0)))
 511                 return (error);
 512
 513         /*
 514          * Start our first transaction of the day.
 515          *
 516          * All future transactions during this code must be "chained" off
 517          * this one via the trans_dup() call.  All transactions will contain
 518          * the inode, and the inode will always be marked with trans_ihold().
 519          * Since the inode will be locked in all transactions, we must log
 520          * the inode in every transaction to let it float upward through
 521          * the log.
 522          */
 523         args.trans = xfs_trans_alloc(mp, XFS_TRANS_ATTR_RM);
 524
 525         /*
 526          * Root fork attributes can use reserved data blocks for this
 527          * operation if necessary
 528          */
 529
 530         if (flags & ATTR_ROOT)
 531                 args.trans->t_flags |= XFS_TRANS_RESERVE;
 532
 533         if ((error = xfs_trans_reserve(args.trans,
 534                                       XFS_ATTRRM_SPACE_RES(mp),
 535                                       XFS_ATTRRM_LOG_RES(mp),
 536                                       0, XFS_TRANS_PERM_LOG_RES,
 537                                       XFS_ATTRRM_LOG_COUNT))) {
 538                 xfs_trans_cancel(args.trans, 0);
 539                 return(error);
 540
 541         }
 542
 543         xfs_ilock(dp, XFS_ILOCK_EXCL);
 544         /*
 545          * No need to make quota reservations here. We expect to release some
 546          * blocks not allocate in the common case.
 547          */
 548         xfs_trans_ijoin(args.trans, dp, XFS_ILOCK_EXCL);
 549         xfs_trans_ihold(args.trans, dp);
 550
 551         /*
 552          * Decide on what work routines to call based on the inode size.
 553          */
 554         if (XFS_IFORK_Q(dp) == 0 ||
 555             (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 556              dp->i_d.di_anextents == 0)) {
 557                 error = XFS_ERROR(ENOATTR);
 558                 goto out;
 559         }
 560         if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
 561                 ASSERT(dp->i_afp->if_flags & XFS_IFINLINE);
 562                 error = xfs_attr_shortform_remove(&args);
 563                 if (error) {
 564                         goto out;
 565                 }
 566         } else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
 567                 error = xfs_attr_leaf_removename(&args);
 568         } else {
 569                 error = xfs_attr_node_removename(&args);
 570         }
 571         if (error) {
 572                 goto out;
 573         }
 574
 575         /*
 576          * If this is a synchronous mount, make sure that the
 577          * transaction goes to disk before returning to the user.
 578          */
 579         if (mp->m_flags & XFS_MOUNT_WSYNC) {
 580                 xfs_trans_set_sync(args.trans);
 581         }
 582
 583         /*
 584          * Commit the last in the sequence of transactions.
 585          */
 586         xfs_trans_log_inode(args.trans, dp, XFS_ILOG_CORE);
 587         error = xfs_trans_commit(args.trans, XFS_TRANS_RELEASE_LOG_RES,
 588                                  NULL);
 589         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 590
 591         /*
 592          * Hit the inode change time.
 593          */
 594         if (!error && (flags & ATTR_KERNOTIME) == 0) {
 595                 xfs_ichgtime(dp, XFS_ICHGTIME_CHG);
 596         }
 597
 598         return(error);
 599
 600 out:
 601         if (args.trans)
 602                 xfs_trans_cancel(args.trans,
 603                         XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
 604         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 605         return(error);
 606 }
 607
 608 /*
 609  * Generate a list of extended attribute names and optionally
 610  * also value lengths.  Positive return value follows the XFS
 611  * convention of being an error, zero or negative return code
 612  * is the length of the buffer returned (negated), indicating
 613  * success.
 614  */
 615 int
 616 xfs_attr_list(bhv_desc_t *bdp, char *buffer, int bufsize, int flags,
 617                       attrlist_cursor_kern_t *cursor, struct cred *cred)
 618 {
 619         xfs_attr_list_context_t context;
 620         xfs_inode_t *dp;
 621         int error;
 622
 623         XFS_STATS_INC(xs_attr_list);
 624
 625         /*
 626          * Validate the cursor.
 627          */
 628         if (cursor->pad1 || cursor->pad2)
 629                 return(XFS_ERROR(EINVAL));
 630         if ((cursor->initted == 0) &&
 631             (cursor->hashval || cursor->blkno || cursor->offset))
 632                 return(XFS_ERROR(EINVAL));
 633
 634         /*
 635          * Check for a properly aligned buffer.
 636          */
 637         if (((long)buffer) & (sizeof(int)-1))
 638                 return(XFS_ERROR(EFAULT));
 639         if (flags & ATTR_KERNOVAL)
 640                 bufsize = 0;
 641
 642         /*
 643          * Initialize the output buffer.
 644          */
 645         context.dp = dp = XFS_BHVTOI(bdp);
 646         context.cursor = cursor;
 647         context.count = 0;
 648         context.dupcnt = 0;
 649         context.resynch = 1;
 650         context.flags = flags;
 651         if (!(flags & ATTR_KERNAMELS)) {
 652                 context.bufsize = (bufsize & ~(sizeof(int)-1));  /* align */
 653                 context.firstu = context.bufsize;
 654                 context.alist = (attrlist_t *)buffer;
 655                 context.alist->al_count = 0;
 656                 context.alist->al_more = 0;
 657                 context.alist->al_offset[0] = context.bufsize;
 658         }
 659         else {
 660                 context.bufsize = bufsize;
 661                 context.firstu = context.bufsize;
 662                 context.alist = (attrlist_t *)buffer;
 663         }
 664
 665         if (XFS_FORCED_SHUTDOWN(dp->i_mount))
 666                 return (EIO);
 667
 668         xfs_ilock(dp, XFS_ILOCK_SHARED);
 669         if (!(flags & ATTR_SECURE) &&
 670              (error = xfs_iaccess(dp, S_IRUSR, cred))) {
 671                 xfs_iunlock(dp, XFS_ILOCK_SHARED);
 672                 return(XFS_ERROR(error));
 673         }
 674
 675         /*
 676          * Decide on what work routines to call based on the inode size.
 677          */
 678         xfs_attr_trace_l_c("syscall start", &context);
 679         if (XFS_IFORK_Q(dp) == 0 ||
 680             (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 681              dp->i_d.di_anextents == 0)) {
 682                 error = 0;
 683         } else if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
 684                 error = xfs_attr_shortform_list(&context);
 685         } else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
 686                 error = xfs_attr_leaf_list(&context);
 687         } else {
 688                 error = xfs_attr_node_list(&context);
 689         }
 690         xfs_iunlock(dp, XFS_ILOCK_SHARED);
 691         xfs_attr_trace_l_c("syscall end", &context);
 692
 693         if (!(context.flags & (ATTR_KERNOVAL|ATTR_KERNAMELS))) {
 694                 ASSERT(error >= 0);
 695         }
 696         else {  /* must return negated buffer size or the error */
 697                 if (context.count < 0)
 698                         error = XFS_ERROR(ERANGE);
 699                 else
 700                         error = -context.count;
 701         }
 702
 703         return(error);
 704 }
 705
 706 int                                                             /* error */
 707 xfs_attr_inactive(xfs_inode_t *dp)
 708 {
 709         xfs_trans_t *trans;
 710         xfs_mount_t *mp;
 711         int error;
 712
 713         mp = dp->i_mount;
 714         ASSERT(! XFS_NOT_DQATTACHED(mp, dp));
 715
 716         xfs_ilock(dp, XFS_ILOCK_SHARED);
 717         if ((XFS_IFORK_Q(dp) == 0) ||
 718             (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) ||
 719             (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 720              dp->i_d.di_anextents == 0)) {
 721                 xfs_iunlock(dp, XFS_ILOCK_SHARED);
 722                 return(0);
 723         }
 724         xfs_iunlock(dp, XFS_ILOCK_SHARED);
 725
 726         /*
 727          * Start our first transaction of the day.
 728          *
 729          * All future transactions during this code must be "chained" off
 730          * this one via the trans_dup() call.  All transactions will contain
 731          * the inode, and the inode will always be marked with trans_ihold().
 732          * Since the inode will be locked in all transactions, we must log
 733          * the inode in every transaction to let it float upward through
 734          * the log.
 735          */
 736         trans = xfs_trans_alloc(mp, XFS_TRANS_ATTRINVAL);
 737         if ((error = xfs_trans_reserve(trans, 0, XFS_ATTRINVAL_LOG_RES(mp), 0,
 738                                       XFS_TRANS_PERM_LOG_RES,
 739                                       XFS_ATTRINVAL_LOG_COUNT))) {
 740                 xfs_trans_cancel(trans, 0);
 741                 return(error);
 742         }
 743         xfs_ilock(dp, XFS_ILOCK_EXCL);
 744
 745         /*
 746          * No need to make quota reservations here. We expect to release some
 747          * blocks, not allocate, in the common case.
 748          */
 749         xfs_trans_ijoin(trans, dp, XFS_ILOCK_EXCL);
 750         xfs_trans_ihold(trans, dp);
 751
 752         /*
 753          * Decide on what work routines to call based on the inode size.
 754          */
 755         if ((XFS_IFORK_Q(dp) == 0) ||
 756             (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) ||
 757             (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 758              dp->i_d.di_anextents == 0)) {
 759                 error = 0;
 760                 goto out;
 761         }
 762         error = xfs_attr_root_inactive(&trans, dp);
 763         if (error)
 764                 goto out;
 765         /*
 766          * signal synchronous inactive transactions unless this
 767          * is a synchronous mount filesystem in which case we
 768          * know that we're here because we've been called out of
 769          * xfs_inactive which means that the last reference is gone
 770          * and the unlink transaction has already hit the disk so
 771          * async inactive transactions are safe.
 772          */
 773         if ((error = xfs_itruncate_finish(&trans, dp, 0LL, XFS_ATTR_FORK,
 774                                 (!(mp->m_flags & XFS_MOUNT_WSYNC)
 775                                  ? 1 : 0))))
 776                 goto out;
 777
 778         /*
 779          * Commit the last in the sequence of transactions.
 780          */
 781         xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE);
 782         error = xfs_trans_commit(trans, XFS_TRANS_RELEASE_LOG_RES,
 783                                  NULL);
 784         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 785
 786         return(error);
 787
 788 out:
 789         xfs_trans_cancel(trans, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
 790         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 791         return(error);
 792 }
 793
 794
 795
 796 /*========================================================================
 797  * External routines when attribute list is inside the inode
 798  *========================================================================*/
 799
 800 /*
 801  * Add a name to the shortform attribute list structure
 802  * This is the external routine.
 803  */
 804 STATIC int
 805 xfs_attr_shortform_addname(xfs_da_args_t *args)
 806 {
 807         int newsize, retval;
 808
 809         retval = xfs_attr_shortform_lookup(args);
 810         if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
 811                 return(retval);
 812         } else if (retval == EEXIST) {
 813                 if (args->flags & ATTR_CREATE)
 814                         return(retval);
 815                 retval = xfs_attr_shortform_remove(args);
 816                 ASSERT(retval == 0);
 817         }
 818
 819         newsize = XFS_ATTR_SF_TOTSIZE(args->dp);
 820         newsize += XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
 821         if ((newsize <= XFS_IFORK_ASIZE(args->dp)) &&
 822             (args->namelen < XFS_ATTR_SF_ENTSIZE_MAX) &&
 823             (args->valuelen < XFS_ATTR_SF_ENTSIZE_MAX)) {
 824                 retval = xfs_attr_shortform_add(args);
 825                 ASSERT(retval == 0);
 826         } else {
 827                 return(XFS_ERROR(ENOSPC));
 828         }
 829         return(0);
 830 }
 831
 832
 833 /*========================================================================
 834  * External routines when attribute list is one block
 835  *========================================================================*/
 836
 837 /*
 838  * Add a name to the leaf attribute list structure
 839  *
 840  * This leaf block cannot have a "remote" value, we only call this routine
 841  * if bmap_one_block() says there is only one block (ie: no remote blks).
 842  */
 843 int
 844 xfs_attr_leaf_addname(xfs_da_args_t *args)
 845 {
 846         xfs_inode_t *dp;
 847         xfs_dabuf_t *bp;
 848         int retval, error, committed;
 849
 850         /*
 851          * Read the (only) block in the attribute list in.
 852          */
 853         dp = args->dp;
 854         args->blkno = 0;
 855         error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
 856                                              XFS_ATTR_FORK);
 857         if (error)
 858                 return(error);
 859         ASSERT(bp != NULL);
 860
 861         /*
 862          * Look up the given attribute in the leaf block.  Figure out if
 863          * the given flags produce an error or call for an atomic rename.
 864          */
 865         retval = xfs_attr_leaf_lookup_int(bp, args);
 866         if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
 867                 xfs_da_brelse(args->trans, bp);
 868                 return(retval);
 869         } else if (retval == EEXIST) {
 870                 if (args->flags & ATTR_CREATE) {        /* pure create op */
 871                         xfs_da_brelse(args->trans, bp);
 872                         return(retval);
 873                 }
 874                 args->rename = 1;                       /* an atomic rename */
 875                 args->blkno2 = args->blkno;             /* set 2nd entry info*/
 876                 args->index2 = args->index;
 877                 args->rmtblkno2 = args->rmtblkno;
 878                 args->rmtblkcnt2 = args->rmtblkcnt;
 879         }
 880
 881         /*
 882          * Add the attribute to the leaf block, transitioning to a Btree
 883          * if required.
 884          */
 885         retval = xfs_attr_leaf_add(bp, args);
 886         xfs_da_buf_done(bp);
 887         if (retval == ENOSPC) {
 888                 /*
 889                  * Promote the attribute list to the Btree format, then
 890                  * Commit that transaction so that the node_addname() call
 891                  * can manage its own transactions.
 892                  */
 893                 XFS_BMAP_INIT(args->flist, args->firstblock);
 894                 error = xfs_attr_leaf_to_node(args);
 895                 if (!error) {
 896                         error = xfs_bmap_finish(&args->trans, args->flist,
 897                                                 *args->firstblock, &committed);
 898                 }
 899                 if (error) {
 900                         ASSERT(committed);
 901                         args->trans = NULL;
 902                         xfs_bmap_cancel(args->flist);
 903                         return(error);
 904                 }
 905
 906                 /*
 907                  * bmap_finish() may have committed the last trans and started
 908                  * a new one.  We need the inode to be in all transactions.
 909                  */
 910                 if (committed) {
 911                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
 912                         xfs_trans_ihold(args->trans, dp);
 913                 }
 914
 915                 /*
 916                  * Commit the current trans (including the inode) and start
 917                  * a new one.
 918                  */
 919                 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
 920                         return (error);
 921
 922                 /*
 923                  * Fob the whole rest of the problem off on the Btree code.
 924                  */
 925                 error = xfs_attr_node_addname(args);
 926                 return(error);
 927         }
 928
 929         /*
 930          * Commit the transaction that added the attr name so that
 931          * later routines can manage their own transactions.
 932          */
 933         if ((error = xfs_attr_rolltrans(&args->trans, dp)))
 934                 return (error);
 935
 936         /*
 937          * If there was an out-of-line value, allocate the blocks we
 938          * identified for its storage and copy the value.  This is done
 939          * after we create the attribute so that we don't overflow the
 940          * maximum size of a transaction and/or hit a deadlock.
 941          */
 942         if (args->rmtblkno > 0) {
 943                 error = xfs_attr_rmtval_set(args);
 944                 if (error)
 945                         return(error);
 946         }
 947
 948         /*
 949          * If this is an atomic rename operation, we must "flip" the
 950          * incomplete flags on the "new" and "old" attribute/value pairs
 951          * so that one disappears and one appears atomically.  Then we
 952          * must remove the "old" attribute/value pair.
 953          */
 954         if (args->rename) {
 955                 /*
 956                  * In a separate transaction, set the incomplete flag on the
 957                  * "old" attr and clear the incomplete flag on the "new" attr.
 958                  */
 959                 error = xfs_attr_leaf_flipflags(args);
 960                 if (error)
 961                         return(error);
 962
 963                 /*
 964                  * Dismantle the "old" attribute/value pair by removing
 965                  * a "remote" value (if it exists).
 966                  */
 967                 args->index = args->index2;
 968                 args->blkno = args->blkno2;
 969                 args->rmtblkno = args->rmtblkno2;
 970                 args->rmtblkcnt = args->rmtblkcnt2;
 971                 if (args->rmtblkno) {
 972                         error = xfs_attr_rmtval_remove(args);
 973                         if (error)
 974                                 return(error);
 975                 }
 976
 977                 /*
 978                  * Read in the block containing the "old" attr, then
 979                  * remove the "old" attr from that block (neat, huh!)
 980                  */
 981                 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1,
 982                                                      &bp, XFS_ATTR_FORK);
 983                 if (error)
 984                         return(error);
 985                 ASSERT(bp != NULL);
 986                 (void)xfs_attr_leaf_remove(bp, args);
 987
 988                 /*
 989                  * If the result is small enough, shrink it all into the inode.
 990                  */
 991                 if (xfs_attr_shortform_allfit(bp, dp)) {
 992                         XFS_BMAP_INIT(args->flist, args->firstblock);
 993                         error = xfs_attr_leaf_to_shortform(bp, args);
 994                         /* bp is gone due to xfs_da_shrink_inode */
 995                         if (!error) {
 996                                 error = xfs_bmap_finish(&args->trans,
 997                                                         args->flist,
 998                                                         *args->firstblock,
 999                                                         &committed);
1000                         }
1001                         if (error) {
1002                                 ASSERT(committed);
1003                                 args->trans = NULL;
1004                                 xfs_bmap_cancel(args->flist);
1005                                 return(error);
1006                         }
1007
1008                         /*
1009                          * bmap_finish() may have committed the last trans
1010                          * and started a new one.  We need the inode to be
1011                          * in all transactions.
1012                          */
1013                         if (committed) {
1014                                 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1015                                 xfs_trans_ihold(args->trans, dp);
1016                         }
1017                 } else
1018                         xfs_da_buf_done(bp);
1019
1020                 /*
1021                  * Commit the remove and start the next trans in series.
1022                  */
1023                 error = xfs_attr_rolltrans(&args->trans, dp);
1024
1025         } else if (args->rmtblkno > 0) {
1026                 /*
1027                  * Added a "remote" value, just clear the incomplete flag.
1028                  */
1029                 error = xfs_attr_leaf_clearflag(args);
1030         }
1031         return(error);
1032 }
1033
1034 /*
1035  * Remove a name from the leaf attribute list structure
1036  *
1037  * This leaf block cannot have a "remote" value, we only call this routine
1038  * if bmap_one_block() says there is only one block (ie: no remote blks).
1039  */
1040 STATIC int
1041 xfs_attr_leaf_removename(xfs_da_args_t *args)
1042 {
1043         xfs_inode_t *dp;
1044         xfs_dabuf_t *bp;
1045         int committed;
1046         int error;
1047
1048         /*
1049          * Remove the attribute.
1050          */
1051         dp = args->dp;
1052         args->blkno = 0;
1053         error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
1054                                              XFS_ATTR_FORK);
1055         if (error) {
1056                 return(error);
1057         }
1058
1059         ASSERT(bp != NULL);
1060         error = xfs_attr_leaf_lookup_int(bp, args);
1061         if (error == ENOATTR) {
1062                 xfs_da_brelse(args->trans, bp);
1063                 return(error);
1064         }
1065
1066         (void)xfs_attr_leaf_remove(bp, args);
1067
1068         /*
1069          * If the result is small enough, shrink it all into the inode.
1070          */
1071         if (xfs_attr_shortform_allfit(bp, dp)) {
1072                 XFS_BMAP_INIT(args->flist, args->firstblock);
1073                 error = xfs_attr_leaf_to_shortform(bp, args);
1074                 /* bp is gone due to xfs_da_shrink_inode */
1075                 if (!error) {
1076                         error = xfs_bmap_finish(&args->trans, args->flist,
1077                                                 *args->firstblock, &committed);
1078                 }
1079                 if (error) {
1080                         ASSERT(committed);
1081                         args->trans = NULL;
1082                         xfs_bmap_cancel(args->flist);
1083                         return(error);
1084                 }
1085
1086                 /*
1087                  * bmap_finish() may have committed the last trans and started
1088                  * a new one.  We need the inode to be in all transactions.
1089                  */
1090                 if (committed) {
1091                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1092                         xfs_trans_ihold(args->trans, dp);
1093                 }
1094         } else
1095                 xfs_da_buf_done(bp);
1096         return(0);
1097 }
1098
1099 /*
1100  * Look up a name in a leaf attribute list structure.
1101  *
1102  * This leaf block cannot have a "remote" value, we only call this routine
1103  * if bmap_one_block() says there is only one block (ie: no remote blks).
1104  */
1105 int
1106 xfs_attr_leaf_get(xfs_da_args_t *args)
1107 {
1108         xfs_dabuf_t *bp;
1109         int error;
1110
1111         args->blkno = 0;
1112         error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
1113                                              XFS_ATTR_FORK);
1114         if (error)
1115                 return(error);
1116         ASSERT(bp != NULL);
1117
1118         error = xfs_attr_leaf_lookup_int(bp, args);
1119         if (error != EEXIST)  {
1120                 xfs_da_brelse(args->trans, bp);
1121                 return(error);
1122         }
1123         error = xfs_attr_leaf_getvalue(bp, args);
1124         xfs_da_brelse(args->trans, bp);
1125         if (!error && (args->rmtblkno > 0) && !(args->flags & ATTR_KERNOVAL)) {
1126                 error = xfs_attr_rmtval_get(args);
1127         }
1128         return(error);
1129 }
1130
1131 /*
1132  * Copy out attribute entries for attr_list(), for leaf attribute lists.
1133  */
1134 STATIC int
1135 xfs_attr_leaf_list(xfs_attr_list_context_t *context)
1136 {
1137         xfs_attr_leafblock_t *leaf;
1138         int error;
1139         xfs_dabuf_t *bp;
1140
1141         context->cursor->blkno = 0;
1142         error = xfs_da_read_buf(NULL, context->dp, 0, -1, &bp, XFS_ATTR_FORK);
1143         if (error)
1144                 return(error);
1145         ASSERT(bp != NULL);
1146         leaf = bp->data;
1147         if (unlikely(INT_GET(leaf->hdr.info.magic, ARCH_CONVERT)
1148                                                 != XFS_ATTR_LEAF_MAGIC)) {
1149                 XFS_CORRUPTION_ERROR("xfs_attr_leaf_list", XFS_ERRLEVEL_LOW,
1150                                      context->dp->i_mount, leaf);
1151                 xfs_da_brelse(NULL, bp);
1152                 return(XFS_ERROR(EFSCORRUPTED));
1153         }
1154
1155         (void)xfs_attr_leaf_list_int(bp, context);
1156         xfs_da_brelse(NULL, bp);
1157         return(0);
1158 }
1159
1160
1161 /*========================================================================
1162  * External routines when attribute list size > XFS_LBSIZE(mp).
1163  *========================================================================*/
1164
1165 /*
1166  * Add a name to a Btree-format attribute list.
1167  *
1168  * This will involve walking down the Btree, and may involve splitting
1169  * leaf nodes and even splitting intermediate nodes up to and including
1170  * the root node (a special case of an intermediate node).
1171  *
1172  * "Remote" attribute values confuse the issue and atomic rename operations
1173  * add a whole extra layer of confusion on top of that.
1174  */
1175 STATIC int
1176 xfs_attr_node_addname(xfs_da_args_t *args)
1177 {
1178         xfs_da_state_t *state;
1179         xfs_da_state_blk_t *blk;
1180         xfs_inode_t *dp;
1181         xfs_mount_t *mp;
1182         int committed, retval, error;
1183
1184         /*
1185          * Fill in bucket of arguments/results/context to carry around.
1186          */
1187         dp = args->dp;
1188         mp = dp->i_mount;
1189 restart:
1190         state = xfs_da_state_alloc();
1191         state->args = args;
1192         state->mp = mp;
1193         state->blocksize = state->mp->m_sb.sb_blocksize;
1194         state->node_ents = state->mp->m_attr_node_ents;
1195
1196         /*
1197          * Search to see if name already exists, and get back a pointer
1198          * to where it should go.
1199          */
1200         error = xfs_da_node_lookup_int(state, &retval);
1201         if (error)
1202                 goto out;
1203         blk = &state->path.blk[ state->path.active-1 ];
1204         ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1205         if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
1206                 goto out;
1207         } else if (retval == EEXIST) {
1208                 if (args->flags & ATTR_CREATE)
1209                         goto out;
1210                 args->rename = 1;                       /* atomic rename op */
1211                 args->blkno2 = args->blkno;             /* set 2nd entry info*/
1212                 args->index2 = args->index;
1213                 args->rmtblkno2 = args->rmtblkno;
1214                 args->rmtblkcnt2 = args->rmtblkcnt;
1215                 args->rmtblkno = 0;
1216                 args->rmtblkcnt = 0;
1217         }
1218
1219         retval = xfs_attr_leaf_add(blk->bp, state->args);
1220         if (retval == ENOSPC) {
1221                 if (state->path.active == 1) {
1222                         /*
1223                          * Its really a single leaf node, but it had
1224                          * out-of-line values so it looked like it *might*
1225                          * have been a b-tree.
1226                          */
1227                         xfs_da_state_free(state);
1228                         XFS_BMAP_INIT(args->flist, args->firstblock);
1229                         error = xfs_attr_leaf_to_node(args);
1230                         if (!error) {
1231                                 error = xfs_bmap_finish(&args->trans,
1232                                                         args->flist,
1233                                                         *args->firstblock,
1234                                                         &committed);
1235                         }
1236                         if (error) {
1237                                 ASSERT(committed);
1238                                 args->trans = NULL;
1239                                 xfs_bmap_cancel(args->flist);
1240                                 goto out;
1241                         }
1242
1243                         /*
1244                          * bmap_finish() may have committed the last trans
1245                          * and started a new one.  We need the inode to be
1246                          * in all transactions.
1247                          */
1248                         if (committed) {
1249                                 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1250                                 xfs_trans_ihold(args->trans, dp);
1251                         }
1252
1253                         /*
1254                          * Commit the node conversion and start the next
1255                          * trans in the chain.
1256                          */
1257                         if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1258                                 goto out;
1259
1260                         goto restart;
1261                 }
1262
1263                 /*
1264                  * Split as many Btree elements as required.
1265                  * This code tracks the new and old attr's location
1266                  * in the index/blkno/rmtblkno/rmtblkcnt fields and
1267                  * in the index2/blkno2/rmtblkno2/rmtblkcnt2 fields.
1268                  */
1269                 XFS_BMAP_INIT(args->flist, args->firstblock);
1270                 error = xfs_da_split(state);
1271                 if (!error) {
1272                         error = xfs_bmap_finish(&args->trans, args->flist,
1273                                                 *args->firstblock, &committed);
1274                 }
1275                 if (error) {
1276                         ASSERT(committed);
1277                         args->trans = NULL;
1278                         xfs_bmap_cancel(args->flist);
1279                         goto out;
1280                 }
1281
1282                 /*
1283                  * bmap_finish() may have committed the last trans and started
1284                  * a new one.  We need the inode to be in all transactions.
1285                  */
1286                 if (committed) {
1287                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1288                         xfs_trans_ihold(args->trans, dp);
1289                 }
1290         } else {
1291                 /*
1292                  * Addition succeeded, update Btree hashvals.
1293                  */
1294                 xfs_da_fixhashpath(state, &state->path);
1295         }
1296
1297         /*
1298          * Kill the state structure, we're done with it and need to
1299          * allow the buffers to come back later.
1300          */
1301         xfs_da_state_free(state);
1302         state = NULL;
1303
1304         /*
1305          * Commit the leaf addition or btree split and start the next
1306          * trans in the chain.
1307          */
1308         if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1309                 goto out;
1310
1311         /*
1312          * If there was an out-of-line value, allocate the blocks we
1313          * identified for its storage and copy the value.  This is done
1314          * after we create the attribute so that we don't overflow the
1315          * maximum size of a transaction and/or hit a deadlock.
1316          */
1317         if (args->rmtblkno > 0) {
1318                 error = xfs_attr_rmtval_set(args);
1319                 if (error)
1320                         return(error);
1321         }
1322
1323         /*
1324          * If this is an atomic rename operation, we must "flip" the
1325          * incomplete flags on the "new" and "old" attribute/value pairs
1326          * so that one disappears and one appears atomically.  Then we
1327          * must remove the "old" attribute/value pair.
1328          */
1329         if (args->rename) {
1330                 /*
1331                  * In a separate transaction, set the incomplete flag on the
1332                  * "old" attr and clear the incomplete flag on the "new" attr.
1333                  */
1334                 error = xfs_attr_leaf_flipflags(args);
1335                 if (error)
1336                         goto out;
1337
1338                 /*
1339                  * Dismantle the "old" attribute/value pair by removing
1340                  * a "remote" value (if it exists).
1341                  */
1342                 args->index = args->index2;
1343                 args->blkno = args->blkno2;
1344                 args->rmtblkno = args->rmtblkno2;
1345                 args->rmtblkcnt = args->rmtblkcnt2;
1346                 if (args->rmtblkno) {
1347                         error = xfs_attr_rmtval_remove(args);
1348                         if (error)
1349                                 return(error);
1350                 }
1351
1352                 /*
1353                  * Re-find the "old" attribute entry after any split ops.
1354                  * The INCOMPLETE flag means that we will find the "old"
1355                  * attr, not the "new" one.
1356                  */
1357                 args->flags |= XFS_ATTR_INCOMPLETE;
1358                 state = xfs_da_state_alloc();
1359                 state->args = args;
1360                 state->mp = mp;
1361                 state->blocksize = state->mp->m_sb.sb_blocksize;
1362                 state->node_ents = state->mp->m_attr_node_ents;
1363                 state->inleaf = 0;
1364                 error = xfs_da_node_lookup_int(state, &retval);
1365                 if (error)
1366                         goto out;
1367
1368                 /*
1369                  * Remove the name and update the hashvals in the tree.
1370                  */
1371                 blk = &state->path.blk[ state->path.active-1 ];
1372                 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1373                 error = xfs_attr_leaf_remove(blk->bp, args);
1374                 xfs_da_fixhashpath(state, &state->path);
1375
1376                 /*
1377                  * Check to see if the tree needs to be collapsed.
1378                  */
1379                 if (retval && (state->path.active > 1)) {
1380                         XFS_BMAP_INIT(args->flist, args->firstblock);
1381                         error = xfs_da_join(state);
1382                         if (!error) {
1383                                 error = xfs_bmap_finish(&args->trans,
1384                                                         args->flist,
1385                                                         *args->firstblock,
1386                                                         &committed);
1387                         }
1388                         if (error) {
1389                                 ASSERT(committed);
1390                                 args->trans = NULL;
1391                                 xfs_bmap_cancel(args->flist);
1392                                 goto out;
1393                         }
1394
1395                         /*
1396                          * bmap_finish() may have committed the last trans
1397                          * and started a new one.  We need the inode to be
1398                          * in all transactions.
1399                          */
1400                         if (committed) {
1401                                 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1402                                 xfs_trans_ihold(args->trans, dp);
1403                         }
1404                 }
1405
1406                 /*
1407                  * Commit and start the next trans in the chain.
1408                  */
1409                 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1410                         goto out;
1411
1412         } else if (args->rmtblkno > 0) {
1413                 /*
1414                  * Added a "remote" value, just clear the incomplete flag.
1415                  */
1416                 error = xfs_attr_leaf_clearflag(args);
1417                 if (error)
1418                         goto out;
1419         }
1420         retval = error = 0;
1421
1422 out:
1423         if (state)
1424                 xfs_da_state_free(state);
1425         if (error)
1426                 return(error);
1427         return(retval);
1428 }
1429
1430 /*
1431  * Remove a name from a B-tree attribute list.
1432  *
1433  * This will involve walking down the Btree, and may involve joining
1434  * leaf nodes and even joining intermediate nodes up to and including
1435  * the root node (a special case of an intermediate node).
1436  */
1437 STATIC int
1438 xfs_attr_node_removename(xfs_da_args_t *args)
1439 {
1440         xfs_da_state_t *state;
1441         xfs_da_state_blk_t *blk;
1442         xfs_inode_t *dp;
1443         xfs_dabuf_t *bp;
1444         int retval, error, committed;
1445
1446         /*
1447          * Tie a string around our finger to remind us where we are.
1448          */
1449         dp = args->dp;
1450         state = xfs_da_state_alloc();
1451         state->args = args;
1452         state->mp = dp->i_mount;
1453         state->blocksize = state->mp->m_sb.sb_blocksize;
1454         state->node_ents = state->mp->m_attr_node_ents;
1455
1456         /*
1457          * Search to see if name exists, and get back a pointer to it.
1458          */
1459         error = xfs_da_node_lookup_int(state, &retval);
1460         if (error || (retval != EEXIST)) {
1461                 if (error == 0)
1462                         error = retval;
1463                 goto out;
1464         }
1465
1466         /*
1467          * If there is an out-of-line value, de-allocate the blocks.
1468          * This is done before we remove the attribute so that we don't
1469          * overflow the maximum size of a transaction and/or hit a deadlock.
1470          */
1471         blk = &state->path.blk[ state->path.active-1 ];
1472         ASSERT(blk->bp != NULL);
1473         ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1474         if (args->rmtblkno > 0) {
1475                 /*
1476                  * Fill in disk block numbers in the state structure
1477                  * so that we can get the buffers back after we commit
1478                  * several transactions in the following calls.
1479                  */
1480                 error = xfs_attr_fillstate(state);
1481                 if (error)
1482                         goto out;
1483
1484                 /*
1485                  * Mark the attribute as INCOMPLETE, then bunmapi() the
1486                  * remote value.
1487                  */
1488                 error = xfs_attr_leaf_setflag(args);
1489                 if (error)
1490                         goto out;
1491                 error = xfs_attr_rmtval_remove(args);
1492                 if (error)
1493                         goto out;
1494
1495                 /*
1496                  * Refill the state structure with buffers, the prior calls
1497                  * released our buffers.
1498                  */
1499                 error = xfs_attr_refillstate(state);
1500                 if (error)
1501                         goto out;
1502         }
1503
1504         /*
1505          * Remove the name and update the hashvals in the tree.
1506          */
1507         blk = &state->path.blk[ state->path.active-1 ];
1508         ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1509         retval = xfs_attr_leaf_remove(blk->bp, args);
1510         xfs_da_fixhashpath(state, &state->path);
1511
1512         /*
1513          * Check to see if the tree needs to be collapsed.
1514          */
1515         if (retval && (state->path.active > 1)) {
1516                 XFS_BMAP_INIT(args->flist, args->firstblock);
1517                 error = xfs_da_join(state);
1518                 if (!error) {
1519                         error = xfs_bmap_finish(&args->trans, args->flist,
1520                                                 *args->firstblock, &committed);
1521                 }
1522                 if (error) {
1523                         ASSERT(committed);
1524                         args->trans = NULL;
1525                         xfs_bmap_cancel(args->flist);
1526                         goto out;
1527                 }
1528
1529                 /*
1530                  * bmap_finish() may have committed the last trans and started
1531                  * a new one.  We need the inode to be in all transactions.
1532                  */
1533                 if (committed) {
1534                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1535                         xfs_trans_ihold(args->trans, dp);
1536                 }
1537
1538                 /*
1539                  * Commit the Btree join operation and start a new trans.
1540                  */
1541                 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1542                         goto out;
1543         }
1544
1545         /*
1546          * If the result is small enough, push it all into the inode.
1547          */
1548         if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
1549                 /*
1550                  * Have to get rid of the copy of this dabuf in the state.
1551                  */
1552                 ASSERT(state->path.active == 1);
1553                 ASSERT(state->path.blk[0].bp);
1554                 xfs_da_buf_done(state->path.blk[0].bp);
1555                 state->path.blk[0].bp = NULL;
1556
1557                 error = xfs_da_read_buf(args->trans, args->dp, 0, -1, &bp,
1558                                                      XFS_ATTR_FORK);
1559                 if (error)
1560                         goto out;
1561                 ASSERT(INT_GET(((xfs_attr_leafblock_t *)
1562                                       bp->data)->hdr.info.magic, ARCH_CONVERT)
1563                                                        == XFS_ATTR_LEAF_MAGIC);
1564
1565                 if (xfs_attr_shortform_allfit(bp, dp)) {
1566                         XFS_BMAP_INIT(args->flist, args->firstblock);
1567                         error = xfs_attr_leaf_to_shortform(bp, args);
1568                         /* bp is gone due to xfs_da_shrink_inode */
1569                         if (!error) {
1570                                 error = xfs_bmap_finish(&args->trans,
1571                                                         args->flist,
1572                                                         *args->firstblock,
1573                                                         &committed);
1574                         }
1575                         if (error) {
1576                                 ASSERT(committed);
1577                                 args->trans = NULL;
1578                                 xfs_bmap_cancel(args->flist);
1579                                 goto out;
1580                         }
1581
1582                         /*
1583                          * bmap_finish() may have committed the last trans
1584                          * and started a new one.  We need the inode to be
1585                          * in all transactions.
1586                          */
1587                         if (committed) {
1588                                 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1589                                 xfs_trans_ihold(args->trans, dp);
1590                         }
1591                 } else
1592                         xfs_da_brelse(args->trans, bp);
1593         }
1594         error = 0;
1595
1596 out:
1597         xfs_da_state_free(state);
1598         return(error);
1599 }
1600
1601 /*
1602  * Fill in the disk block numbers in the state structure for the buffers
1603  * that are attached to the state structure.
1604  * This is done so that we can quickly reattach ourselves to those buffers
1605  * after some set of transaction commit's has released these buffers.
1606  */
1607 STATIC int
1608 xfs_attr_fillstate(xfs_da_state_t *state)
1609 {
1610         xfs_da_state_path_t *path;
1611         xfs_da_state_blk_t *blk;
1612         int level;
1613
1614         /*
1615          * Roll down the "path" in the state structure, storing the on-disk
1616          * block number for those buffers in the "path".
1617          */
1618         path = &state->path;
1619         ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1620         for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1621                 if (blk->bp) {
1622                         blk->disk_blkno = xfs_da_blkno(blk->bp);
1623                         xfs_da_buf_done(blk->bp);
1624                         blk->bp = NULL;
1625                 } else {
1626                         blk->disk_blkno = 0;
1627                 }
1628         }
1629
1630         /*
1631          * Roll down the "altpath" in the state structure, storing the on-disk
1632          * block number for those buffers in the "altpath".
1633          */
1634         path = &state->altpath;
1635         ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1636         for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1637                 if (blk->bp) {
1638                         blk->disk_blkno = xfs_da_blkno(blk->bp);
1639                         xfs_da_buf_done(blk->bp);
1640                         blk->bp = NULL;
1641                 } else {
1642                         blk->disk_blkno = 0;
1643                 }
1644         }
1645
1646         return(0);
1647 }
1648
1649 /*
1650  * Reattach the buffers to the state structure based on the disk block
1651  * numbers stored in the state structure.
1652  * This is done after some set of transaction commit's has released those
1653  * buffers from our grip.
1654  */
1655 STATIC int
1656 xfs_attr_refillstate(xfs_da_state_t *state)
1657 {
1658         xfs_da_state_path_t *path;
1659         xfs_da_state_blk_t *blk;
1660         int level, error;
1661
1662         /*
1663          * Roll down the "path" in the state structure, storing the on-disk
1664          * block number for those buffers in the "path".
1665          */
1666         path = &state->path;
1667         ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1668         for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1669                 if (blk->disk_blkno) {
1670                         error = xfs_da_read_buf(state->args->trans,
1671                                                 state->args->dp,
1672                                                 blk->blkno, blk->disk_blkno,
1673                                                 &blk->bp, XFS_ATTR_FORK);
1674                         if (error)
1675                                 return(error);
1676                 } else {
1677                         blk->bp = NULL;
1678                 }
1679         }
1680
1681         /*
1682          * Roll down the "altpath" in the state structure, storing the on-disk
1683          * block number for those buffers in the "altpath".
1684          */
1685         path = &state->altpath;
1686         ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1687         for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1688                 if (blk->disk_blkno) {
1689                         error = xfs_da_read_buf(state->args->trans,
1690                                                 state->args->dp,
1691                                                 blk->blkno, blk->disk_blkno,
1692                                                 &blk->bp, XFS_ATTR_FORK);
1693                         if (error)
1694                                 return(error);
1695                 } else {
1696                         blk->bp = NULL;
1697                 }
1698         }
1699
1700         return(0);
1701 }
1702
1703 /*
1704  * Look up a filename in a node attribute list.
1705  *
1706  * This routine gets called for any attribute fork that has more than one
1707  * block, ie: both true Btree attr lists and for single-leaf-blocks with
1708  * "remote" values taking up more blocks.
1709  */
1710 int
1711 xfs_attr_node_get(xfs_da_args_t *args)
1712 {
1713         xfs_da_state_t *state;
1714         xfs_da_state_blk_t *blk;
1715         int error, retval;
1716         int i;
1717
1718         state = xfs_da_state_alloc();
1719         state->args = args;
1720         state->mp = args->dp->i_mount;
1721         state->blocksize = state->mp->m_sb.sb_blocksize;
1722         state->node_ents = state->mp->m_attr_node_ents;
1723
1724         /*
1725          * Search to see if name exists, and get back a pointer to it.
1726          */
1727         error = xfs_da_node_lookup_int(state, &retval);
1728         if (error) {
1729                 retval = error;
1730         } else if (retval == EEXIST) {
1731                 blk = &state->path.blk[ state->path.active-1 ];
1732                 ASSERT(blk->bp != NULL);
1733                 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1734
1735                 /*
1736                  * Get the value, local or "remote"
1737                  */
1738                 retval = xfs_attr_leaf_getvalue(blk->bp, args);
1739                 if (!retval && (args->rmtblkno > 0)
1740                     && !(args->flags & ATTR_KERNOVAL)) {
1741                         retval = xfs_attr_rmtval_get(args);
1742                 }
1743         }
1744
1745         /*
1746          * If not in a transaction, we have to release all the buffers.
1747          */
1748         for (i = 0; i < state->path.active; i++) {
1749                 xfs_da_brelse(args->trans, state->path.blk[i].bp);
1750                 state->path.blk[i].bp = NULL;
1751         }
1752
1753         xfs_da_state_free(state);
1754         return(retval);
1755 }
1756
1757 STATIC int                                                      /* error */
1758 xfs_attr_node_list(xfs_attr_list_context_t *context)
1759 {
1760         attrlist_cursor_kern_t *cursor;
1761         xfs_attr_leafblock_t *leaf;
1762         xfs_da_intnode_t *node;
1763         xfs_da_node_entry_t *btree;
1764         int error, i;
1765         xfs_dabuf_t *bp;
1766
1767         cursor = context->cursor;
1768         cursor->initted = 1;
1769
1770         /*
1771          * Do all sorts of validation on the passed-in cursor structure.
1772          * If anything is amiss, ignore the cursor and look up the hashval
1773          * starting from the btree root.
1774          */
1775         bp = NULL;
1776         if (cursor->blkno > 0) {
1777                 error = xfs_da_read_buf(NULL, context->dp, cursor->blkno, -1,
1778                                               &bp, XFS_ATTR_FORK);
1779                 if ((error != 0) && (error != EFSCORRUPTED))
1780                         return(error);
1781                 if (bp) {
1782                         node = bp->data;
1783                         switch (INT_GET(node->hdr.info.magic, ARCH_CONVERT)) {
1784                         case XFS_DA_NODE_MAGIC:
1785                                 xfs_attr_trace_l_cn("wrong blk", context, node);
1786                                 xfs_da_brelse(NULL, bp);
1787                                 bp = NULL;
1788                                 break;
1789                         case XFS_ATTR_LEAF_MAGIC:
1790                                 leaf = bp->data;
1791                                 if (cursor->hashval >
1792                                     INT_GET(leaf->entries[
1793                                          INT_GET(leaf->hdr.count,
1794                                                 ARCH_CONVERT)-1].hashval,
1795                                                         ARCH_CONVERT)) {
1796                                         xfs_attr_trace_l_cl("wrong blk",
1797                                                            context, leaf);
1798                                         xfs_da_brelse(NULL, bp);
1799                                         bp = NULL;
1800                                 } else if (cursor->hashval <=
1801                                              INT_GET(leaf->entries[0].hashval,
1802                                                         ARCH_CONVERT)) {
1803                                         xfs_attr_trace_l_cl("maybe wrong blk",
1804                                                            context, leaf);
1805                                         xfs_da_brelse(NULL, bp);
1806                                         bp = NULL;
1807                                 }
1808                                 break;
1809                         default:
1810                                 xfs_attr_trace_l_c("wrong blk - ??", context);
1811                                 xfs_da_brelse(NULL, bp);
1812                                 bp = NULL;
1813                         }
1814                 }
1815         }
1816
1817         /*
1818          * We did not find what we expected given the cursor's contents,
1819          * so we start from the top and work down based on the hash value.
1820          * Note that start of node block is same as start of leaf block.
1821          */
1822         if (bp == NULL) {
1823                 cursor->blkno = 0;
1824                 for (;;) {
1825                         error = xfs_da_read_buf(NULL, context->dp,
1826                                                       cursor->blkno, -1, &bp,
1827                                                       XFS_ATTR_FORK);
1828                         if (error)
1829                                 return(error);
1830                         if (unlikely(bp == NULL)) {
1831                                 XFS_ERROR_REPORT("xfs_attr_node_list(2)",
1832                                                  XFS_ERRLEVEL_LOW,
1833                                                  context->dp->i_mount);
1834                                 return(XFS_ERROR(EFSCORRUPTED));
1835                         }
1836                         node = bp->data;
1837                         if (INT_GET(node->hdr.info.magic, ARCH_CONVERT)
1838                                                         == XFS_ATTR_LEAF_MAGIC)
1839                                 break;
1840                         if (unlikely(INT_GET(node->hdr.info.magic, ARCH_CONVERT)
1841                                                         != XFS_DA_NODE_MAGIC)) {
1842                                 XFS_CORRUPTION_ERROR("xfs_attr_node_list(3)",
1843                                                      XFS_ERRLEVEL_LOW,
1844                                                      context->dp->i_mount,
1845                                                      node);
1846                                 xfs_da_brelse(NULL, bp);
1847                                 return(XFS_ERROR(EFSCORRUPTED));
1848                         }
1849                         btree = node->btree;
1850                         for (i = 0;
1851                                 i < INT_GET(node->hdr.count, ARCH_CONVERT);
1852                                                                 btree++, i++) {
1853                                 if (cursor->hashval
1854                                                 <= INT_GET(btree->hashval,
1855                                                             ARCH_CONVERT)) {
1856                                         cursor->blkno = INT_GET(btree->before, ARCH_CONVERT);
1857                                         xfs_attr_trace_l_cb("descending",
1858                                                             context, btree);
1859                                         break;
1860                                 }
1861                         }
1862                         if (i == INT_GET(node->hdr.count, ARCH_CONVERT)) {
1863                                 xfs_da_brelse(NULL, bp);
1864                                 return(0);
1865                         }
1866                         xfs_da_brelse(NULL, bp);
1867                 }
1868         }
1869         ASSERT(bp != NULL);
1870
1871         /*
1872          * Roll upward through the blocks, processing each leaf block in
1873          * order.  As long as there is space in the result buffer, keep
1874          * adding the information.
1875          */
1876         for (;;) {
1877                 leaf = bp->data;
1878                 if (unlikely(INT_GET(leaf->hdr.info.magic, ARCH_CONVERT)
1879                                                 != XFS_ATTR_LEAF_MAGIC)) {
1880                         XFS_CORRUPTION_ERROR("xfs_attr_node_list(4)",
1881                                              XFS_ERRLEVEL_LOW,
1882                                              context->dp->i_mount, leaf);
1883                         xfs_da_brelse(NULL, bp);
1884                         return(XFS_ERROR(EFSCORRUPTED));
1885                 }
1886                 error = xfs_attr_leaf_list_int(bp, context);
1887                 if (error || !leaf->hdr.info.forw)
1888                         break;  /* not really an error, buffer full or EOF */
1889                 cursor->blkno = INT_GET(leaf->hdr.info.forw, ARCH_CONVERT);
1890                 xfs_da_brelse(NULL, bp);
1891                 error = xfs_da_read_buf(NULL, context->dp, cursor->blkno, -1,
1892                                               &bp, XFS_ATTR_FORK);
1893                 if (error)
1894                         return(error);
1895                 if (unlikely((bp == NULL))) {
1896                         XFS_ERROR_REPORT("xfs_attr_node_list(5)",
1897                                          XFS_ERRLEVEL_LOW,
1898                                          context->dp->i_mount);
1899                         return(XFS_ERROR(EFSCORRUPTED));
1900                 }
1901         }
1902         xfs_da_brelse(NULL, bp);
1903         return(0);
1904 }
1905
1906
1907 /*========================================================================
1908  * External routines for manipulating out-of-line attribute values.
1909  *========================================================================*/
1910
1911 /*
1912  * Read the value associated with an attribute from the out-of-line buffer
1913  * that we stored it in.
1914  */
1915 STATIC int
1916 xfs_attr_rmtval_get(xfs_da_args_t *args)
1917 {
1918         xfs_bmbt_irec_t map[ATTR_RMTVALUE_MAPSIZE];
1919         xfs_mount_t *mp;
1920         xfs_daddr_t dblkno;
1921         xfs_caddr_t dst;
1922         xfs_buf_t *bp;
1923         int nmap, error, tmp, valuelen, blkcnt, i;
1924         xfs_dablk_t lblkno;
1925
1926         ASSERT(!(args->flags & ATTR_KERNOVAL));
1927
1928         mp = args->dp->i_mount;
1929         dst = args->value;
1930         valuelen = args->valuelen;
1931         lblkno = args->rmtblkno;
1932         while (valuelen > 0) {
1933                 nmap = ATTR_RMTVALUE_MAPSIZE;
1934                 error = xfs_bmapi(args->trans, args->dp, (xfs_fileoff_t)lblkno,
1935                                   args->rmtblkcnt,
1936                                   XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
1937                                   NULL, 0, map, &nmap, NULL);
1938                 if (error)
1939                         return(error);
1940                 ASSERT(nmap >= 1);
1941
1942                 for (i = 0; (i < nmap) && (valuelen > 0); i++) {
1943                         ASSERT((map[i].br_startblock != DELAYSTARTBLOCK) &&
1944                                (map[i].br_startblock != HOLESTARTBLOCK));
1945                         dblkno = XFS_FSB_TO_DADDR(mp, map[i].br_startblock);
1946                         blkcnt = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
1947                         error = xfs_read_buf(mp, mp->m_ddev_targp, dblkno,
1948                                              blkcnt, XFS_BUF_LOCK, &bp);
1949                         if (error)
1950                                 return(error);
1951
1952                         tmp = (valuelen < XFS_BUF_SIZE(bp))
1953                                 ? valuelen : XFS_BUF_SIZE(bp);
1954                         xfs_biomove(bp, 0, tmp, dst, XFS_B_READ);
1955                         xfs_buf_relse(bp);
1956                         dst += tmp;
1957                         valuelen -= tmp;
1958
1959                         lblkno += map[i].br_blockcount;
1960                 }
1961         }
1962         ASSERT(valuelen == 0);
1963         return(0);
1964 }
1965
1966 /*
1967  * Write the value associated with an attribute into the out-of-line buffer
1968  * that we have defined for it.
1969  */
1970 STATIC int
1971 xfs_attr_rmtval_set(xfs_da_args_t *args)
1972 {
1973         xfs_mount_t *mp;
1974         xfs_fileoff_t lfileoff;
1975         xfs_inode_t *dp;
1976         xfs_bmbt_irec_t map;
1977         xfs_daddr_t dblkno;
1978         xfs_caddr_t src;
1979         xfs_buf_t *bp;
1980         xfs_dablk_t lblkno;
1981         int blkcnt, valuelen, nmap, error, tmp, committed;
1982
1983         dp = args->dp;
1984         mp = dp->i_mount;
1985         src = args->value;
1986
1987         /*
1988          * Find a "hole" in the attribute address space large enough for
1989          * us to drop the new attribute's value into.
1990          */
1991         blkcnt = XFS_B_TO_FSB(mp, args->valuelen);
1992         lfileoff = 0;
1993         error = xfs_bmap_first_unused(args->trans, args->dp, blkcnt, &lfileoff,
1994                                                    XFS_ATTR_FORK);
1995         if (error) {
1996                 return(error);
1997         }
1998         args->rmtblkno = lblkno = (xfs_dablk_t)lfileoff;
1999         args->rmtblkcnt = blkcnt;
2000
2001         /*
2002          * Roll through the "value", allocating blocks on disk as required.
2003          */
2004         while (blkcnt > 0) {
2005                 /*
2006                  * Allocate a single extent, up to the size of the value.
2007                  */
2008                 XFS_BMAP_INIT(args->flist, args->firstblock);
2009                 nmap = 1;
2010                 error = xfs_bmapi(args->trans, dp, (xfs_fileoff_t)lblkno,
2011                                   blkcnt,
2012                                   XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA |
2013                                                         XFS_BMAPI_WRITE,
2014                                   args->firstblock, args->total, &map, &nmap,
2015                                   args->flist);
2016                 if (!error) {
2017                         error = xfs_bmap_finish(&args->trans, args->flist,
2018                                                 *args->firstblock, &committed);
2019                 }
2020                 if (error) {
2021                         ASSERT(committed);
2022                         args->trans = NULL;
2023                         xfs_bmap_cancel(args->flist);
2024                         return(error);
2025                 }
2026
2027                 /*
2028                  * bmap_finish() may have committed the last trans and started
2029                  * a new one.  We need the inode to be in all transactions.
2030                  */
2031                 if (committed) {
2032                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
2033                         xfs_trans_ihold(args->trans, dp);
2034                 }
2035
2036                 ASSERT(nmap == 1);
2037                 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2038                        (map.br_startblock != HOLESTARTBLOCK));
2039                 lblkno += map.br_blockcount;
2040                 blkcnt -= map.br_blockcount;
2041
2042                 /*
2043                  * Start the next trans in the chain.
2044                  */
2045                 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
2046                         return (error);
2047         }
2048
2049         /*
2050          * Roll through the "value", copying the attribute value to the
2051          * already-allocated blocks.  Blocks are written synchronously
2052          * so that we can know they are all on disk before we turn off
2053          * the INCOMPLETE flag.
2054          */
2055         lblkno = args->rmtblkno;
2056         valuelen = args->valuelen;
2057         while (valuelen > 0) {
2058                 /*
2059                  * Try to remember where we decided to put the value.
2060                  */
2061                 XFS_BMAP_INIT(args->flist, args->firstblock);
2062                 nmap = 1;
2063                 error = xfs_bmapi(NULL, dp, (xfs_fileoff_t)lblkno,
2064                                   args->rmtblkcnt,
2065                                   XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2066                                   args->firstblock, 0, &map, &nmap, NULL);
2067                 if (error) {
2068                         return(error);
2069                 }
2070                 ASSERT(nmap == 1);
2071                 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2072                        (map.br_startblock != HOLESTARTBLOCK));
2073
2074                 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
2075                 blkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
2076
2077                 bp = xfs_buf_get_flags(mp->m_ddev_targp, dblkno,
2078                                                         blkcnt, XFS_BUF_LOCK);
2079                 ASSERT(bp);
2080                 ASSERT(!XFS_BUF_GETERROR(bp));
2081
2082                 tmp = (valuelen < XFS_BUF_SIZE(bp)) ? valuelen :
2083                                                         XFS_BUF_SIZE(bp);
2084                 xfs_biomove(bp, 0, tmp, src, XFS_B_WRITE);
2085                 if (tmp < XFS_BUF_SIZE(bp))
2086                         xfs_biozero(bp, tmp, XFS_BUF_SIZE(bp) - tmp);
2087                 if ((error = xfs_bwrite(mp, bp))) {/* GROT: NOTE: synchronous write */
2088                         return (error);
2089                 }
2090                 src += tmp;
2091                 valuelen -= tmp;
2092
2093                 lblkno += map.br_blockcount;
2094         }
2095         ASSERT(valuelen == 0);
2096         return(0);
2097 }
2098
2099 /*
2100  * Remove the value associated with an attribute by deleting the
2101  * out-of-line buffer that it is stored on.
2102  */
2103 STATIC int
2104 xfs_attr_rmtval_remove(xfs_da_args_t *args)
2105 {
2106         xfs_mount_t *mp;
2107         xfs_bmbt_irec_t map;
2108         xfs_buf_t *bp;
2109         xfs_daddr_t dblkno;
2110         xfs_dablk_t lblkno;
2111         int valuelen, blkcnt, nmap, error, done, committed;
2112
2113         mp = args->dp->i_mount;
2114
2115         /*
2116          * Roll through the "value", invalidating the attribute value's
2117          * blocks.
2118          */
2119         lblkno = args->rmtblkno;
2120         valuelen = args->rmtblkcnt;
2121         while (valuelen > 0) {
2122                 /*
2123                  * Try to remember where we decided to put the value.
2124                  */
2125                 XFS_BMAP_INIT(args->flist, args->firstblock);
2126                 nmap = 1;
2127                 error = xfs_bmapi(NULL, args->dp, (xfs_fileoff_t)lblkno,
2128                                         args->rmtblkcnt,
2129                                         XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2130                                         args->firstblock, 0, &map, &nmap,
2131                                         args->flist);
2132                 if (error) {
2133                         return(error);
2134                 }
2135                 ASSERT(nmap == 1);
2136                 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2137                        (map.br_startblock != HOLESTARTBLOCK));
2138
2139                 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
2140                 blkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
2141
2142                 /*
2143                  * If the "remote" value is in the cache, remove it.
2144                  */
2145                 bp = xfs_incore(mp->m_ddev_targp, dblkno, blkcnt,
2146                                 XFS_INCORE_TRYLOCK);
2147                 if (bp) {
2148                         XFS_BUF_STALE(bp);
2149                         XFS_BUF_UNDELAYWRITE(bp);
2150                         xfs_buf_relse(bp);
2151                         bp = NULL;
2152                 }
2153
2154                 valuelen -= map.br_blockcount;
2155
2156                 lblkno += map.br_blockcount;
2157         }
2158
2159         /*
2160          * Keep de-allocating extents until the remote-value region is gone.
2161          */
2162         lblkno = args->rmtblkno;
2163         blkcnt = args->rmtblkcnt;
2164         done = 0;
2165         while (!done) {
2166                 XFS_BMAP_INIT(args->flist, args->firstblock);
2167                 error = xfs_bunmapi(args->trans, args->dp, lblkno, blkcnt,
2168                                     XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2169                                     1, args->firstblock, args->flist, &done);
2170                 if (!error) {
2171                         error = xfs_bmap_finish(&args->trans, args->flist,
2172                                                 *args->firstblock, &committed);
2173                 }
2174                 if (error) {
2175                         ASSERT(committed);
2176                         args->trans = NULL;
2177                         xfs_bmap_cancel(args->flist);
2178                         return(error);
2179                 }
2180
2181                 /*
2182                  * bmap_finish() may have committed the last trans and started
2183                  * a new one.  We need the inode to be in all transactions.
2184                  */
2185                 if (committed) {
2186                         xfs_trans_ijoin(args->trans, args->dp, XFS_ILOCK_EXCL);
2187                         xfs_trans_ihold(args->trans, args->dp);
2188                 }
2189
2190                 /*
2191                  * Close out trans and start the next one in the chain.
2192                  */
2193                 if ((error = xfs_attr_rolltrans(&args->trans, args->dp)))
2194                         return (error);
2195         }
2196         return(0);
2197 }
2198
2199 #if defined(XFS_ATTR_TRACE)
2200 /*
2201  * Add a trace buffer entry for an attr_list context structure.
2202  */
2203 void
2204 xfs_attr_trace_l_c(char *where, struct xfs_attr_list_context *context)
2205 {
2206         xfs_attr_trace_enter(XFS_ATTR_KTRACE_L_C, where,
2207                 (__psunsigned_t)context->dp,
2208                 (__psunsigned_t)context->cursor->hashval,
2209                 (__psunsigned_t)context->cursor->blkno,
2210                 (__psunsigned_t)context->cursor->offset,
2211                 (__psunsigned_t)context->alist,
2212                 (__psunsigned_t)context->bufsize,
2213                 (__psunsigned_t)context->count,
2214                 (__psunsigned_t)context->firstu,
2215                 (__psunsigned_t)
2216                         ((context->count > 0) &&
2217                         !(context->flags & (ATTR_KERNAMELS|ATTR_KERNOVAL)))
2218                                 ? (ATTR_ENTRY(context->alist,
2219                                               context->count-1)->a_valuelen)
2220                                 : 0,
2221                 (__psunsigned_t)context->dupcnt,
2222                 (__psunsigned_t)context->flags,
2223                 (__psunsigned_t)NULL,
2224                 (__psunsigned_t)NULL,
2225                 (__psunsigned_t)NULL);
2226 }
2227
2228 /*
2229  * Add a trace buffer entry for a context structure and a Btree node.
2230  */
2231 void
2232 xfs_attr_trace_l_cn(char *where, struct xfs_attr_list_context *context,
2233                          struct xfs_da_intnode *node)
2234 {
2235         xfs_attr_trace_enter(XFS_ATTR_KTRACE_L_CN, where,
2236                 (__psunsigned_t)context->dp,
2237                 (__psunsigned_t)context->cursor->hashval,
2238                 (__psunsigned_t)context->cursor->blkno,
2239                 (__psunsigned_t)context->cursor->offset,
2240                 (__psunsigned_t)context->alist,
2241                 (__psunsigned_t)context->bufsize,
2242                 (__psunsigned_t)context->count,
2243                 (__psunsigned_t)context->firstu,
2244                 (__psunsigned_t)
2245                         ((context->count > 0) &&
2246                         !(context->flags & (ATTR_KERNAMELS|ATTR_KERNOVAL)))
2247                                 ? (ATTR_ENTRY(context->alist,
2248                                               context->count-1)->a_valuelen)
2249                                 : 0,
2250                 (__psunsigned_t)context->dupcnt,
2251                 (__psunsigned_t)context->flags,
2252                 (__psunsigned_t)INT_GET(node->hdr.count, ARCH_CONVERT),
2253                 (__psunsigned_t)INT_GET(node->btree[0].hashval, ARCH_CONVERT),
2254                 (__psunsigned_t)INT_GET(node->btree[INT_GET(node->hdr.count, ARCH_CONVERT)-1].hashval, ARCH_CONVERT));
2255 }
2256
2257 /*
2258  * Add a trace buffer entry for a context structure and a Btree element.
2259  */
2260 void
2261 xfs_attr_trace_l_cb(char *where, struct xfs_attr_list_context *context,
2262                           struct xfs_da_node_entry *btree)
2263 {
2264         xfs_attr_trace_enter(XFS_ATTR_KTRACE_L_CB, where,
2265                 (__psunsigned_t)context->dp,
2266                 (__psunsigned_t)context->cursor->hashval,
2267                 (__psunsigned_t)context->cursor->blkno,
2268                 (__psunsigned_t)context->cursor->offset,
2269                 (__psunsigned_t)context->alist,
2270                 (__psunsigned_t)context->bufsize,
2271                 (__psunsigned_t)context->count,
2272                 (__psunsigned_t)context->firstu,
2273                 (__psunsigned_t)
2274                         ((context->count > 0) &&
2275                         !(context->flags & (ATTR_KERNAMELS|ATTR_KERNOVAL)))
2276                                 ? (ATTR_ENTRY(context->alist,
2277                                               context->count-1)->a_valuelen)
2278                                 : 0,
2279                 (__psunsigned_t)context->dupcnt,
2280                 (__psunsigned_t)context->flags,
2281                 (__psunsigned_t)INT_GET(btree->hashval, ARCH_CONVERT),
2282                 (__psunsigned_t)INT_GET(btree->before, ARCH_CONVERT),
2283                 (__psunsigned_t)NULL);
2284 }
2285
2286 /*
2287  * Add a trace buffer entry for a context structure and a leaf block.
2288  */
2289 void
2290 xfs_attr_trace_l_cl(char *where, struct xfs_attr_list_context *context,
2291                               struct xfs_attr_leafblock *leaf)
2292 {
2293         xfs_attr_trace_enter(XFS_ATTR_KTRACE_L_CL, where,
2294                 (__psunsigned_t)context->dp,
2295                 (__psunsigned_t)context->cursor->hashval,
2296                 (__psunsigned_t)context->cursor->blkno,
2297                 (__psunsigned_t)context->cursor->offset,
2298                 (__psunsigned_t)context->alist,
2299                 (__psunsigned_t)context->bufsize,
2300                 (__psunsigned_t)context->count,
2301                 (__psunsigned_t)context->firstu,
2302                 (__psunsigned_t)
2303                         ((context->count > 0) &&
2304                         !(context->flags & (ATTR_KERNAMELS|ATTR_KERNOVAL)))
2305                                 ? (ATTR_ENTRY(context->alist,
2306                                               context->count-1)->a_valuelen)
2307                                 : 0,
2308                 (__psunsigned_t)context->dupcnt,
2309                 (__psunsigned_t)context->flags,
2310                 (__psunsigned_t)INT_GET(leaf->hdr.count, ARCH_CONVERT),
2311                 (__psunsigned_t)INT_GET(leaf->entries[0].hashval, ARCH_CONVERT),
2312                 (__psunsigned_t)INT_GET(leaf->entries[INT_GET(leaf->hdr.count, ARCH_CONVERT)-1].hashval, ARCH_CONVERT));
2313 }
2314
2315 /*
2316  * Add a trace buffer entry for the arguments given to the routine,
2317  * generic form.
2318  */
2319 void
2320 xfs_attr_trace_enter(int type, char *where,
2321                          __psunsigned_t a2, __psunsigned_t a3,
2322                          __psunsigned_t a4, __psunsigned_t a5,
2323                          __psunsigned_t a6, __psunsigned_t a7,
2324                          __psunsigned_t a8, __psunsigned_t a9,
2325                          __psunsigned_t a10, __psunsigned_t a11,
2326                          __psunsigned_t a12, __psunsigned_t a13,
2327                          __psunsigned_t a14, __psunsigned_t a15)
2328 {
2329         ASSERT(xfs_attr_trace_buf);
2330         ktrace_enter(xfs_attr_trace_buf, (void *)((__psunsigned_t)type),
2331                                          (void *)where,
2332                                          (void *)a2,  (void *)a3,  (void *)a4,
2333                                          (void *)a5,  (void *)a6,  (void *)a7,
2334                                          (void *)a8,  (void *)a9,  (void *)a10,
2335                                          (void *)a11, (void *)a12, (void *)a13,
2336                                          (void *)a14, (void *)a15);
2337 }
2338 #endif  /* XFS_ATTR_TRACE */
2339
2340
2341 /*========================================================================
2342  * System (pseudo) namespace attribute interface routines.
2343  *========================================================================*/
2344
2345 STATIC int
2346 posix_acl_access_set(
2347         vnode_t *vp, char *name, void *data, size_t size, int xflags)
2348 {
2349         return xfs_acl_vset(vp, data, size, _ACL_TYPE_ACCESS);
2350 }
2351
2352 STATIC int
2353 posix_acl_access_remove(
2354         struct vnode *vp, char *name, int xflags)
2355 {
2356         return xfs_acl_vremove(vp, _ACL_TYPE_ACCESS);
2357 }
2358
2359 STATIC int
2360 posix_acl_access_get(
2361         vnode_t *vp, char *name, void *data, size_t size, int xflags)
2362 {
2363         return xfs_acl_vget(vp, data, size, _ACL_TYPE_ACCESS);
2364 }
2365
2366 STATIC int
2367 posix_acl_access_exists(
2368         vnode_t *vp)
2369 {
2370         return xfs_acl_vhasacl_access(vp);
2371 }
2372
2373 STATIC int
2374 posix_acl_default_set(
2375         vnode_t *vp, char *name, void *data, size_t size, int xflags)
2376 {
2377         return xfs_acl_vset(vp, data, size, _ACL_TYPE_DEFAULT);
2378 }
2379
2380 STATIC int
2381 posix_acl_default_get(
2382         vnode_t *vp, char *name, void *data, size_t size, int xflags)
2383 {
2384         return xfs_acl_vget(vp, data, size, _ACL_TYPE_DEFAULT);
2385 }
2386
2387 STATIC int
2388 posix_acl_default_remove(
2389         struct vnode *vp, char *name, int xflags)
2390 {
2391         return xfs_acl_vremove(vp, _ACL_TYPE_DEFAULT);
2392 }
2393
2394 STATIC int
2395 posix_acl_default_exists(
2396         vnode_t *vp)
2397 {
2398         return xfs_acl_vhasacl_default(vp);
2399 }
2400
2401 struct attrnames posix_acl_access = {
2402         .attr_name      = "posix_acl_access",
2403         .attr_namelen   = sizeof("posix_acl_access") - 1,
2404         .attr_get       = posix_acl_access_get,
2405         .attr_set       = posix_acl_access_set,
2406         .attr_remove    = posix_acl_access_remove,
2407         .attr_exists    = posix_acl_access_exists,
2408 };
2409
2410 struct attrnames posix_acl_default = {
2411         .attr_name      = "posix_acl_default",
2412         .attr_namelen   = sizeof("posix_acl_default") - 1,
2413         .attr_get       = posix_acl_default_get,
2414         .attr_set       = posix_acl_default_set,
2415         .attr_remove    = posix_acl_default_remove,
2416         .attr_exists    = posix_acl_default_exists,
2417 };
2418
2419 struct attrnames *attr_system_names[] =
2420         { &posix_acl_access, &posix_acl_default };
2421
2422
2423 /*========================================================================
2424  * Namespace-prefix-style attribute name interface routines.
2425  *========================================================================*/
2426
2427 STATIC int
2428 attr_generic_set(
2429         struct vnode *vp, char *name, void *data, size_t size, int xflags)
2430 {
2431         int     error;
2432
2433         VOP_ATTR_SET(vp, name, data, size, xflags, NULL, error);
2434         return -error;
2435 }
2436
2437 STATIC int
2438 attr_generic_get(
2439         struct vnode *vp, char *name, void *data, size_t size, int xflags)
2440 {
2441         int     error, asize = size;
2442
2443         VOP_ATTR_GET(vp, name, data, &asize, xflags, NULL, error);
2444         if (!error)
2445                 return asize;
2446         return -error;
2447 }
2448
2449 STATIC int
2450 attr_generic_remove(
2451         struct vnode *vp, char *name, int xflags)
2452 {
2453         int     error;
2454
2455         VOP_ATTR_REMOVE(vp, name, xflags, NULL, error);
2456         return -error;
2457 }
2458
2459 STATIC int
2460 attr_generic_listadd(
2461         attrnames_t             *prefix,
2462         attrnames_t             *namesp,
2463         void                    *data,
2464         size_t                  size,
2465         ssize_t                 *result)
2466 {
2467         char                    *p = data + *result;
2468
2469         *result += prefix->attr_namelen;
2470         *result += namesp->attr_namelen + 1;
2471         if (!size)
2472                 return 0;
2473         if (*result > size)
2474                 return -ERANGE;
2475         strcpy(p, prefix->attr_name);
2476         p += prefix->attr_namelen;
2477         strcpy(p, namesp->attr_name);
2478         p += namesp->attr_namelen + 1;
2479         return 0;
2480 }
2481
2482 STATIC int
2483 attr_system_list(
2484         struct vnode            *vp,
2485         void                    *data,
2486         size_t                  size,
2487         ssize_t                 *result)
2488 {
2489         attrnames_t             *namesp;
2490         int                     i, error = 0;
2491
2492         for (i = 0; i < ATTR_SYSCOUNT; i++) {
2493                 namesp = attr_system_names[i];
2494                 if (!namesp->attr_exists || !namesp->attr_exists(vp))
2495                         continue;
2496                 error = attr_generic_listadd(&attr_system, namesp,
2497                                                 data, size, result);
2498                 if (error)
2499                         break;
2500         }
2501         return error;
2502 }
2503
2504 int
2505 attr_generic_list(
2506         struct vnode *vp, void *data, size_t size, int xflags, ssize_t *result)
2507 {
2508         attrlist_cursor_kern_t  cursor = { 0 };
2509         int                     error;
2510
2511         VOP_ATTR_LIST(vp, data, size, xflags, &cursor, NULL, error);
2512         if (error > 0)
2513                 return -error;
2514         *result = -error;
2515         return attr_system_list(vp, data, size, result);
2516 }
2517
2518 attrnames_t *
2519 attr_lookup_namespace(
2520         char                    *name,
2521         struct attrnames        **names,
2522         int                     nnames)
2523 {
2524         int                     i;
2525
2526         for (i = 0; i < nnames; i++)
2527                 if (!strncmp(name, names[i]->attr_name, names[i]->attr_namelen))
2528                         return names[i];
2529         return NULL;
2530 }
2531
2532 /*
2533  * Some checks to prevent people abusing EAs to get over quota:
2534  * - Don't allow modifying user EAs on devices/symlinks;
2535  * - Don't allow modifying user EAs if sticky bit set;
2536  */
2537 STATIC int
2538 attr_user_capable(
2539         struct vnode    *vp,
2540         cred_t          *cred)
2541 {
2542         struct inode    *inode = LINVFS_GET_IP(vp);
2543
2544         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
2545                 return -EPERM;
2546         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode) &&
2547             !capable(CAP_SYS_ADMIN))
2548                 return -EPERM;
2549         if (S_ISDIR(inode->i_mode) && (inode->i_mode & S_ISVTX) &&
2550             (current_fsuid(cred) != inode->i_uid) && !capable(CAP_FOWNER))
2551                 return -EPERM;
2552         return 0;
2553 }
2554
2555 STATIC int
2556 attr_trusted_capable(
2557         struct vnode    *vp,
2558         cred_t          *cred)
2559 {
2560         struct inode    *inode = LINVFS_GET_IP(vp);
2561
2562         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
2563                 return -EPERM;
2564         if (!capable(CAP_SYS_ADMIN))
2565                 return -EPERM;
2566         return 0;
2567 }
2568
2569 STATIC int
2570 attr_secure_capable(
2571         struct vnode    *vp,
2572         cred_t          *cred)
2573 {
2574         return -ENOSECURITY;
2575 }
2576
2577 STATIC int
2578 attr_system_set(
2579         struct vnode *vp, char *name, void *data, size_t size, int xflags)
2580 {
2581         attrnames_t     *namesp;
2582         int             error;
2583
2584         if (xflags & ATTR_CREATE)
2585                 return -EINVAL;
2586
2587         namesp = attr_lookup_namespace(name, attr_system_names, ATTR_SYSCOUNT);
2588         if (!namesp)
2589                 return -EOPNOTSUPP;
2590         error = namesp->attr_set(vp, name, data, size, xflags);
2591         if (!error)
2592                 error = vn_revalidate(vp);
2593         return error;
2594 }
2595
2596 STATIC int
2597 attr_system_get(
2598         struct vnode *vp, char *name, void *data, size_t size, int xflags)
2599 {
2600         attrnames_t     *namesp;
2601
2602         namesp = attr_lookup_namespace(name, attr_system_names, ATTR_SYSCOUNT);
2603         if (!namesp)
2604                 return -EOPNOTSUPP;
2605         return namesp->attr_get(vp, name, data, size, xflags);
2606 }
2607
2608 STATIC int
2609 attr_system_remove(
2610         struct vnode *vp, char *name, int xflags)
2611 {
2612         attrnames_t     *namesp;
2613
2614         namesp = attr_lookup_namespace(name, attr_system_names, ATTR_SYSCOUNT);
2615         if (!namesp)
2616                 return -EOPNOTSUPP;
2617         return namesp->attr_remove(vp, name, xflags);
2618 }
2619
2620 struct attrnames attr_system = {
2621         .attr_name      = "system.",
2622         .attr_namelen   = sizeof("system.") - 1,
2623         .attr_flag      = ATTR_SYSTEM,
2624         .attr_get       = attr_system_get,
2625         .attr_set       = attr_system_set,
2626         .attr_remove    = attr_system_remove,
2627         .attr_capable   = (attrcapable_t)fs_noerr,
2628 };
2629
2630 struct attrnames attr_trusted = {
2631         .attr_name      = "trusted.",
2632         .attr_namelen   = sizeof("trusted.") - 1,
2633         .attr_flag      = ATTR_ROOT,
2634         .attr_get       = attr_generic_get,
2635         .attr_set       = attr_generic_set,
2636         .attr_remove    = attr_generic_remove,
2637         .attr_capable   = attr_trusted_capable,
2638 };
2639
2640 struct attrnames attr_secure = {
2641         .attr_name      = "security.",
2642         .attr_namelen   = sizeof("security.") - 1,
2643         .attr_flag      = ATTR_SECURE,
2644         .attr_get       = attr_generic_get,
2645         .attr_set       = attr_generic_set,
2646         .attr_remove    = attr_generic_remove,
2647         .attr_capable   = attr_secure_capable,
2648 };
2649
2650 struct attrnames attr_user = {
2651         .attr_name      = "user.",
2652         .attr_namelen   = sizeof("user.") - 1,
2653         .attr_get       = attr_generic_get,
2654         .attr_set       = attr_generic_set,
2655         .attr_remove    = attr_generic_remove,
2656         .attr_capable   = attr_user_capable,
2657 };
2658
2659 struct attrnames *attr_namespaces[] =
2660         { &attr_system, &attr_trusted, &attr_secure, &attr_user };