fs/nfs/blocklayout/blocklayout.c

   1 /*
   2  *  linux/fs/nfs/blocklayout/blocklayout.c
   3  *
   4  *  Module for the NFSv4.1 pNFS block layout driver.
   5  *
   6  *  Copyright (c) 2006 The Regents of the University of Michigan.
   7  *  All rights reserved.
   8  *
   9  *  Andy Adamson <andros@citi.umich.edu>
  10  *  Fred Isaman <iisaman@umich.edu>
  11  *
  12  * permission is granted to use, copy, create derivative works and
  13  * redistribute this software and such derivative works for any purpose,
  14  * so long as the name of the university of michigan is not used in
  15  * any advertising or publicity pertaining to the use or distribution
  16  * of this software without specific, written prior authorization.  if
  17  * the above copyright notice or any other identification of the
  18  * university of michigan is included in any copy of any portion of
  19  * this software, then the disclaimer below must also be included.
  20  *
  21  * this software is provided as is, without representation from the
  22  * university of michigan as to its fitness for any purpose, and without
  23  * warranty by the university of michigan of any kind, either express
  24  * or implied, including without limitation the implied warranties of
  25  * merchantability and fitness for a particular purpose.  the regents
  26  * of the university of michigan shall not be liable for any damages,
  27  * including special, indirect, incidental, or consequential damages,
  28  * with respect to any claim arising out or in connection with the use
  29  * of the software, even if it has been or is hereafter advised of the
  30  * possibility of such damages.
  31  */
  32
  33 #include <linux/module.h>
  34 #include <linux/init.h>
  35 #include <linux/mount.h>
  36 #include <linux/namei.h>
  37 #include <linux/bio.h>          /* struct bio */
  38 #include <linux/buffer_head.h>  /* various write calls */
  39 #include <linux/prefetch.h>
  40
  41 #include "blocklayout.h"
  42
  43 #define NFSDBG_FACILITY NFSDBG_PNFS_LD
  44
  45 MODULE_LICENSE("GPL");
  46 MODULE_AUTHOR("Andy Adamson <andros@citi.umich.edu>");
  47 MODULE_DESCRIPTION("The NFSv4.1 pNFS Block layout driver");
  48
  49 static void print_page(struct page *page)
  50 {
  51         dprintk("PRINTPAGE page %p\n", page);
  52         dprintk("       PagePrivate %d\n", PagePrivate(page));
  53         dprintk("       PageUptodate %d\n", PageUptodate(page));
  54         dprintk("       PageError %d\n", PageError(page));
  55         dprintk("       PageDirty %d\n", PageDirty(page));
  56         dprintk("       PageReferenced %d\n", PageReferenced(page));
  57         dprintk("       PageLocked %d\n", PageLocked(page));
  58         dprintk("       PageWriteback %d\n", PageWriteback(page));
  59         dprintk("       PageMappedToDisk %d\n", PageMappedToDisk(page));
  60         dprintk("\n");
  61 }
  62
  63 /* Given the be associated with isect, determine if page data needs to be
  64  * initialized.
  65  */
  66 static int is_hole(struct pnfs_block_extent *be, sector_t isect)
  67 {
  68         if (be->be_state == PNFS_BLOCK_NONE_DATA)
  69                 return 1;
  70         else if (be->be_state != PNFS_BLOCK_INVALID_DATA)
  71                 return 0;
  72         else
  73                 return !bl_is_sector_init(be->be_inval, isect);
  74 }
  75
  76 /* Given the be associated with isect, determine if page data can be
  77  * written to disk.
  78  */
  79 static int is_writable(struct pnfs_block_extent *be, sector_t isect)
  80 {
  81         return (be->be_state == PNFS_BLOCK_READWRITE_DATA ||
  82                 be->be_state == PNFS_BLOCK_INVALID_DATA);
  83 }
  84
  85 /* The data we are handed might be spread across several bios.  We need
  86  * to track when the last one is finished.
  87  */
  88 struct parallel_io {
  89         struct kref refcnt;
  90         void (*pnfs_callback) (void *data, int num_se);
  91         void *data;
  92         int bse_count;
  93 };
  94
  95 static inline struct parallel_io *alloc_parallel(void *data)
  96 {
  97         struct parallel_io *rv;
  98
  99         rv  = kmalloc(sizeof(*rv), GFP_NOFS);
 100         if (rv) {
 101                 rv->data = data;
 102                 kref_init(&rv->refcnt);
 103                 rv->bse_count = 0;
 104         }
 105         return rv;
 106 }
 107
 108 static inline void get_parallel(struct parallel_io *p)
 109 {
 110         kref_get(&p->refcnt);
 111 }
 112
 113 static void destroy_parallel(struct kref *kref)
 114 {
 115         struct parallel_io *p = container_of(kref, struct parallel_io, refcnt);
 116
 117         dprintk("%s enter\n", __func__);
 118         p->pnfs_callback(p->data, p->bse_count);
 119         kfree(p);
 120 }
 121
 122 static inline void put_parallel(struct parallel_io *p)
 123 {
 124         kref_put(&p->refcnt, destroy_parallel);
 125 }
 126
 127 static struct bio *
 128 bl_submit_bio(int rw, struct bio *bio)
 129 {
 130         if (bio) {
 131                 get_parallel(bio->bi_private);
 132                 dprintk("%s submitting %s bio %u@%llu\n", __func__,
 133                         rw == READ ? "read" : "write",
 134                         bio->bi_size, (unsigned long long)bio->bi_sector);
 135                 submit_bio(rw, bio);
 136         }
 137         return NULL;
 138 }
 139
 140 static struct bio *bl_alloc_init_bio(int npg, sector_t isect,
 141                                      struct pnfs_block_extent *be,
 142                                      void (*end_io)(struct bio *, int err),
 143                                      struct parallel_io *par)
 144 {
 145         struct bio *bio;
 146
 147         npg = min(npg, BIO_MAX_PAGES);
 148         bio = bio_alloc(GFP_NOIO, npg);
 149         if (!bio && (current->flags & PF_MEMALLOC)) {
 150                 while (!bio && (npg /= 2))
 151                         bio = bio_alloc(GFP_NOIO, npg);
 152         }
 153
 154         if (bio) {
 155                 bio->bi_sector = isect - be->be_f_offset + be->be_v_offset;
 156                 bio->bi_bdev = be->be_mdev;
 157                 bio->bi_end_io = end_io;
 158                 bio->bi_private = par;
 159         }
 160         return bio;
 161 }
 162
 163 static struct bio *bl_add_page_to_bio(struct bio *bio, int npg, int rw,
 164                                       sector_t isect, struct page *page,
 165                                       struct pnfs_block_extent *be,
 166                                       void (*end_io)(struct bio *, int err),
 167                                       struct parallel_io *par)
 168 {
 169 retry:
 170         if (!bio) {
 171                 bio = bl_alloc_init_bio(npg, isect, be, end_io, par);
 172                 if (!bio)
 173                         return ERR_PTR(-ENOMEM);
 174         }
 175         if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
 176                 bio = bl_submit_bio(rw, bio);
 177                 goto retry;
 178         }
 179         return bio;
 180 }
 181
 182 /* This is basically copied from mpage_end_io_read */
 183 static void bl_end_io_read(struct bio *bio, int err)
 184 {
 185         struct parallel_io *par = bio->bi_private;
 186         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 187         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
 188         struct nfs_read_data *rdata = (struct nfs_read_data *)par->data;
 189
 190         do {
 191                 struct page *page = bvec->bv_page;
 192
 193                 if (--bvec >= bio->bi_io_vec)
 194                         prefetchw(&bvec->bv_page->flags);
 195                 if (uptodate)
 196                         SetPageUptodate(page);
 197         } while (bvec >= bio->bi_io_vec);
 198         if (!uptodate) {
 199                 if (!rdata->pnfs_error)
 200                         rdata->pnfs_error = -EIO;
 201                 pnfs_set_lo_fail(rdata->lseg);
 202         }
 203         bio_put(bio);
 204         put_parallel(par);
 205 }
 206
 207 static void bl_read_cleanup(struct work_struct *work)
 208 {
 209         struct rpc_task *task;
 210         struct nfs_read_data *rdata;
 211         dprintk("%s enter\n", __func__);
 212         task = container_of(work, struct rpc_task, u.tk_work);
 213         rdata = container_of(task, struct nfs_read_data, task);
 214         pnfs_ld_read_done(rdata);
 215 }
 216
 217 static void
 218 bl_end_par_io_read(void *data, int unused)
 219 {
 220         struct nfs_read_data *rdata = data;
 221
 222         rdata->task.tk_status = rdata->pnfs_error;
 223         INIT_WORK(&rdata->task.u.tk_work, bl_read_cleanup);
 224         schedule_work(&rdata->task.u.tk_work);
 225 }
 226
 227 static enum pnfs_try_status
 228 bl_read_pagelist(struct nfs_read_data *rdata)
 229 {
 230         int i, hole;
 231         struct bio *bio = NULL;
 232         struct pnfs_block_extent *be = NULL, *cow_read = NULL;
 233         sector_t isect, extent_length = 0;
 234         struct parallel_io *par;
 235         loff_t f_offset = rdata->args.offset;
 236         struct page **pages = rdata->args.pages;
 237         int pg_index = rdata->args.pgbase >> PAGE_CACHE_SHIFT;
 238
 239         dprintk("%s enter nr_pages %u offset %lld count %u\n", __func__,
 240                rdata->npages, f_offset, (unsigned int)rdata->args.count);
 241
 242         par = alloc_parallel(rdata);
 243         if (!par)
 244                 goto use_mds;
 245         par->pnfs_callback = bl_end_par_io_read;
 246         /* At this point, we can no longer jump to use_mds */
 247
 248         isect = (sector_t) (f_offset >> SECTOR_SHIFT);
 249         /* Code assumes extents are page-aligned */
 250         for (i = pg_index; i < rdata->npages; i++) {
 251                 if (!extent_length) {
 252                         /* We've used up the previous extent */
 253                         bl_put_extent(be);
 254                         bl_put_extent(cow_read);
 255                         bio = bl_submit_bio(READ, bio);
 256                         /* Get the next one */
 257                         be = bl_find_get_extent(BLK_LSEG2EXT(rdata->lseg),
 258                                              isect, &cow_read);
 259                         if (!be) {
 260                                 rdata->pnfs_error = -EIO;
 261                                 goto out;
 262                         }
 263                         extent_length = be->be_length -
 264                                 (isect - be->be_f_offset);
 265                         if (cow_read) {
 266                                 sector_t cow_length = cow_read->be_length -
 267                                         (isect - cow_read->be_f_offset);
 268                                 extent_length = min(extent_length, cow_length);
 269                         }
 270                 }
 271                 hole = is_hole(be, isect);
 272                 if (hole && !cow_read) {
 273                         bio = bl_submit_bio(READ, bio);
 274                         /* Fill hole w/ zeroes w/o accessing device */
 275                         dprintk("%s Zeroing page for hole\n", __func__);
 276                         zero_user_segment(pages[i], 0, PAGE_CACHE_SIZE);
 277                         print_page(pages[i]);
 278                         SetPageUptodate(pages[i]);
 279                 } else {
 280                         struct pnfs_block_extent *be_read;
 281
 282                         be_read = (hole && cow_read) ? cow_read : be;
 283                         bio = bl_add_page_to_bio(bio, rdata->npages - i, READ,
 284                                                  isect, pages[i], be_read,
 285                                                  bl_end_io_read, par);
 286                         if (IS_ERR(bio)) {
 287                                 rdata->pnfs_error = PTR_ERR(bio);
 288                                 bio = NULL;
 289                                 goto out;
 290                         }
 291                 }
 292                 isect += PAGE_CACHE_SECTORS;
 293                 extent_length -= PAGE_CACHE_SECTORS;
 294         }
 295         if ((isect << SECTOR_SHIFT) >= rdata->inode->i_size) {
 296                 rdata->res.eof = 1;
 297                 rdata->res.count = rdata->inode->i_size - f_offset;
 298         } else {
 299                 rdata->res.count = (isect << SECTOR_SHIFT) - f_offset;
 300         }
 301 out:
 302         bl_put_extent(be);
 303         bl_put_extent(cow_read);
 304         bl_submit_bio(READ, bio);
 305         put_parallel(par);
 306         return PNFS_ATTEMPTED;
 307
 308  use_mds:
 309         dprintk("Giving up and using normal NFS\n");
 310         return PNFS_NOT_ATTEMPTED;
 311 }
 312
 313 static void mark_extents_written(struct pnfs_block_layout *bl,
 314                                  __u64 offset, __u32 count)
 315 {
 316         sector_t isect, end;
 317         struct pnfs_block_extent *be;
 318         struct pnfs_block_short_extent *se;
 319
 320         dprintk("%s(%llu, %u)\n", __func__, offset, count);
 321         if (count == 0)
 322                 return;
 323         isect = (offset & (long)(PAGE_CACHE_MASK)) >> SECTOR_SHIFT;
 324         end = (offset + count + PAGE_CACHE_SIZE - 1) & (long)(PAGE_CACHE_MASK);
 325         end >>= SECTOR_SHIFT;
 326         while (isect < end) {
 327                 sector_t len;
 328                 be = bl_find_get_extent(bl, isect, NULL);
 329                 BUG_ON(!be); /* FIXME */
 330                 len = min(end, be->be_f_offset + be->be_length) - isect;
 331                 if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
 332                         se = bl_pop_one_short_extent(be->be_inval);
 333                         BUG_ON(!se);
 334                         bl_mark_for_commit(be, isect, len, se);
 335                 }
 336                 isect += len;
 337                 bl_put_extent(be);
 338         }
 339 }
 340
 341 static void bl_end_io_write_zero(struct bio *bio, int err)
 342 {
 343         struct parallel_io *par = bio->bi_private;
 344         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 345         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
 346         struct nfs_write_data *wdata = (struct nfs_write_data *)par->data;
 347
 348         do {
 349                 struct page *page = bvec->bv_page;
 350
 351                 if (--bvec >= bio->bi_io_vec)
 352                         prefetchw(&bvec->bv_page->flags);
 353                 /* This is the zeroing page we added */
 354                 end_page_writeback(page);
 355                 page_cache_release(page);
 356         } while (bvec >= bio->bi_io_vec);
 357
 358         if (unlikely(!uptodate)) {
 359                 if (!wdata->pnfs_error)
 360                         wdata->pnfs_error = -EIO;
 361                 pnfs_set_lo_fail(wdata->lseg);
 362         }
 363         bio_put(bio);
 364         put_parallel(par);
 365 }
 366
 367 static void bl_end_io_write(struct bio *bio, int err)
 368 {
 369         struct parallel_io *par = bio->bi_private;
 370         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 371         struct nfs_write_data *wdata = (struct nfs_write_data *)par->data;
 372
 373         if (!uptodate) {
 374                 if (!wdata->pnfs_error)
 375                         wdata->pnfs_error = -EIO;
 376                 pnfs_set_lo_fail(wdata->lseg);
 377         }
 378         bio_put(bio);
 379         put_parallel(par);
 380 }
 381
 382 /* Function scheduled for call during bl_end_par_io_write,
 383  * it marks sectors as written and extends the commitlist.
 384  */
 385 static void bl_write_cleanup(struct work_struct *work)
 386 {
 387         struct rpc_task *task;
 388         struct nfs_write_data *wdata;
 389         dprintk("%s enter\n", __func__);
 390         task = container_of(work, struct rpc_task, u.tk_work);
 391         wdata = container_of(task, struct nfs_write_data, task);
 392         if (likely(!wdata->pnfs_error)) {
 393                 /* Marks for LAYOUTCOMMIT */
 394                 mark_extents_written(BLK_LSEG2EXT(wdata->lseg),
 395                                      wdata->args.offset, wdata->args.count);
 396         }
 397         pnfs_ld_write_done(wdata);
 398 }
 399
 400 /* Called when last of bios associated with a bl_write_pagelist call finishes */
 401 static void bl_end_par_io_write(void *data, int num_se)
 402 {
 403         struct nfs_write_data *wdata = data;
 404
 405         if (unlikely(wdata->pnfs_error)) {
 406                 bl_free_short_extents(&BLK_LSEG2EXT(wdata->lseg)->bl_inval,
 407                                         num_se);
 408         }
 409
 410         wdata->task.tk_status = wdata->pnfs_error;
 411         wdata->verf.committed = NFS_FILE_SYNC;
 412         INIT_WORK(&wdata->task.u.tk_work, bl_write_cleanup);
 413         schedule_work(&wdata->task.u.tk_work);
 414 }
 415
 416 /* FIXME STUB - mark intersection of layout and page as bad, so is not
 417  * used again.
 418  */
 419 static void mark_bad_read(void)
 420 {
 421         return;
 422 }
 423
 424 /*
 425  * map_block:  map a requested I/0 block (isect) into an offset in the LVM
 426  * block_device
 427  */
 428 static void
 429 map_block(struct buffer_head *bh, sector_t isect, struct pnfs_block_extent *be)
 430 {
 431         dprintk("%s enter be=%p\n", __func__, be);
 432
 433         set_buffer_mapped(bh);
 434         bh->b_bdev = be->be_mdev;
 435         bh->b_blocknr = (isect - be->be_f_offset + be->be_v_offset) >>
 436             (be->be_mdev->bd_inode->i_blkbits - SECTOR_SHIFT);
 437
 438         dprintk("%s isect %llu, bh->b_blocknr %ld, using bsize %Zd\n",
 439                 __func__, (unsigned long long)isect, (long)bh->b_blocknr,
 440                 bh->b_size);
 441         return;
 442 }
 443
 444 /* Given an unmapped page, zero it or read in page for COW, page is locked
 445  * by caller.
 446  */
 447 static int
 448 init_page_for_write(struct page *page, struct pnfs_block_extent *cow_read)
 449 {
 450         struct buffer_head *bh = NULL;
 451         int ret = 0;
 452         sector_t isect;
 453
 454         dprintk("%s enter, %p\n", __func__, page);
 455         BUG_ON(PageUptodate(page));
 456         if (!cow_read) {
 457                 zero_user_segment(page, 0, PAGE_SIZE);
 458                 SetPageUptodate(page);
 459                 goto cleanup;
 460         }
 461
 462         bh = alloc_page_buffers(page, PAGE_CACHE_SIZE, 0);
 463         if (!bh) {
 464                 ret = -ENOMEM;
 465                 goto cleanup;
 466         }
 467
 468         isect = (sector_t) page->index << PAGE_CACHE_SECTOR_SHIFT;
 469         map_block(bh, isect, cow_read);
 470         if (!bh_uptodate_or_lock(bh))
 471                 ret = bh_submit_read(bh);
 472         if (ret)
 473                 goto cleanup;
 474         SetPageUptodate(page);
 475
 476 cleanup:
 477         bl_put_extent(cow_read);
 478         if (bh)
 479                 free_buffer_head(bh);
 480         if (ret) {
 481                 /* Need to mark layout with bad read...should now
 482                  * just use nfs4 for reads and writes.
 483                  */
 484                 mark_bad_read();
 485         }
 486         return ret;
 487 }
 488
 489 /* Find or create a zeroing page marked being writeback.
 490  * Return ERR_PTR on error, NULL to indicate skip this page and page itself
 491  * to indicate write out.
 492  */
 493 static struct page *
 494 bl_find_get_zeroing_page(struct inode *inode, pgoff_t index,
 495                         struct pnfs_block_extent *cow_read)
 496 {
 497         struct page *page;
 498         int locked = 0;
 499         page = find_get_page(inode->i_mapping, index);
 500         if (page)
 501                 goto check_page;
 502
 503         page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
 504         if (unlikely(!page)) {
 505                 dprintk("%s oom\n", __func__);
 506                 return ERR_PTR(-ENOMEM);
 507         }
 508         locked = 1;
 509
 510 check_page:
 511         /* PageDirty: Other will write this out
 512          * PageWriteback: Other is writing this out
 513          * PageUptodate: It was read before
 514          */
 515         if (PageDirty(page) || PageWriteback(page)) {
 516                 print_page(page);
 517                 if (locked)
 518                         unlock_page(page);
 519                 page_cache_release(page);
 520                 return NULL;
 521         }
 522
 523         if (!locked) {
 524                 lock_page(page);
 525                 locked = 1;
 526                 goto check_page;
 527         }
 528         if (!PageUptodate(page)) {
 529                 /* New page, readin or zero it */
 530                 init_page_for_write(page, cow_read);
 531         }
 532         set_page_writeback(page);
 533         unlock_page(page);
 534
 535         return page;
 536 }
 537
 538 static enum pnfs_try_status
 539 bl_write_pagelist(struct nfs_write_data *wdata, int sync)
 540 {
 541         int i, ret, npg_zero, pg_index, last = 0;
 542         struct bio *bio = NULL;
 543         struct pnfs_block_extent *be = NULL, *cow_read = NULL;
 544         sector_t isect, last_isect = 0, extent_length = 0;
 545         struct parallel_io *par;
 546         loff_t offset = wdata->args.offset;
 547         size_t count = wdata->args.count;
 548         struct page **pages = wdata->args.pages;
 549         struct page *page;
 550         pgoff_t index;
 551         u64 temp;
 552         int npg_per_block =
 553             NFS_SERVER(wdata->inode)->pnfs_blksize >> PAGE_CACHE_SHIFT;
 554
 555         dprintk("%s enter, %Zu@%lld\n", __func__, count, offset);
 556         /* At this point, wdata->pages is a (sequential) list of nfs_pages.
 557          * We want to write each, and if there is an error set pnfs_error
 558          * to have it redone using nfs.
 559          */
 560         par = alloc_parallel(wdata);
 561         if (!par)
 562                 goto out_mds;
 563         par->pnfs_callback = bl_end_par_io_write;
 564         /* At this point, have to be more careful with error handling */
 565
 566         isect = (sector_t) ((offset & (long)PAGE_CACHE_MASK) >> SECTOR_SHIFT);
 567         be = bl_find_get_extent(BLK_LSEG2EXT(wdata->lseg), isect, &cow_read);
 568         if (!be || !is_writable(be, isect)) {
 569                 dprintk("%s no matching extents!\n", __func__);
 570                 goto out_mds;
 571         }
 572
 573         /* First page inside INVALID extent */
 574         if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
 575                 if (likely(!bl_push_one_short_extent(be->be_inval)))
 576                         par->bse_count++;
 577                 else
 578                         goto out_mds;
 579                 temp = offset >> PAGE_CACHE_SHIFT;
 580                 npg_zero = do_div(temp, npg_per_block);
 581                 isect = (sector_t) (((offset - npg_zero * PAGE_CACHE_SIZE) &
 582                                      (long)PAGE_CACHE_MASK) >> SECTOR_SHIFT);
 583                 extent_length = be->be_length - (isect - be->be_f_offset);
 584
 585 fill_invalid_ext:
 586                 dprintk("%s need to zero %d pages\n", __func__, npg_zero);
 587                 for (;npg_zero > 0; npg_zero--) {
 588                         if (bl_is_sector_init(be->be_inval, isect)) {
 589                                 dprintk("isect %llu already init\n",
 590                                         (unsigned long long)isect);
 591                                 goto next_page;
 592                         }
 593                         /* page ref released in bl_end_io_write_zero */
 594                         index = isect >> PAGE_CACHE_SECTOR_SHIFT;
 595                         dprintk("%s zero %dth page: index %lu isect %llu\n",
 596                                 __func__, npg_zero, index,
 597                                 (unsigned long long)isect);
 598                         page = bl_find_get_zeroing_page(wdata->inode, index,
 599                                                         cow_read);
 600                         if (unlikely(IS_ERR(page))) {
 601                                 wdata->pnfs_error = PTR_ERR(page);
 602                                 goto out;
 603                         } else if (page == NULL)
 604                                 goto next_page;
 605
 606                         ret = bl_mark_sectors_init(be->be_inval, isect,
 607                                                        PAGE_CACHE_SECTORS);
 608                         if (unlikely(ret)) {
 609                                 dprintk("%s bl_mark_sectors_init fail %d\n",
 610                                         __func__, ret);
 611                                 end_page_writeback(page);
 612                                 page_cache_release(page);
 613                                 wdata->pnfs_error = ret;
 614                                 goto out;
 615                         }
 616                         if (likely(!bl_push_one_short_extent(be->be_inval)))
 617                                 par->bse_count++;
 618                         else {
 619                                 end_page_writeback(page);
 620                                 page_cache_release(page);
 621                                 wdata->pnfs_error = -ENOMEM;
 622                                 goto out;
 623                         }
 624                         /* FIXME: This should be done in bi_end_io */
 625                         mark_extents_written(BLK_LSEG2EXT(wdata->lseg),
 626                                              page->index << PAGE_CACHE_SHIFT,
 627                                              PAGE_CACHE_SIZE);
 628
 629                         bio = bl_add_page_to_bio(bio, npg_zero, WRITE,
 630                                                  isect, page, be,
 631                                                  bl_end_io_write_zero, par);
 632                         if (IS_ERR(bio)) {
 633                                 wdata->pnfs_error = PTR_ERR(bio);
 634                                 bio = NULL;
 635                                 goto out;
 636                         }
 637 next_page:
 638                         isect += PAGE_CACHE_SECTORS;
 639                         extent_length -= PAGE_CACHE_SECTORS;
 640                 }
 641                 if (last)
 642                         goto write_done;
 643         }
 644         bio = bl_submit_bio(WRITE, bio);
 645
 646         /* Middle pages */
 647         pg_index = wdata->args.pgbase >> PAGE_CACHE_SHIFT;
 648         for (i = pg_index; i < wdata->npages; i++) {
 649                 if (!extent_length) {
 650                         /* We've used up the previous extent */
 651                         bl_put_extent(be);
 652                         bio = bl_submit_bio(WRITE, bio);
 653                         /* Get the next one */
 654                         be = bl_find_get_extent(BLK_LSEG2EXT(wdata->lseg),
 655                                              isect, NULL);
 656                         if (!be || !is_writable(be, isect)) {
 657                                 wdata->pnfs_error = -EINVAL;
 658                                 goto out;
 659                         }
 660                         if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
 661                                 if (likely(!bl_push_one_short_extent(
 662                                                                 be->be_inval)))
 663                                         par->bse_count++;
 664                                 else {
 665                                         wdata->pnfs_error = -ENOMEM;
 666                                         goto out;
 667                                 }
 668                         }
 669                         extent_length = be->be_length -
 670                             (isect - be->be_f_offset);
 671                 }
 672                 if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
 673                         ret = bl_mark_sectors_init(be->be_inval, isect,
 674                                                        PAGE_CACHE_SECTORS);
 675                         if (unlikely(ret)) {
 676                                 dprintk("%s bl_mark_sectors_init fail %d\n",
 677                                         __func__, ret);
 678                                 wdata->pnfs_error = ret;
 679                                 goto out;
 680                         }
 681                 }
 682                 bio = bl_add_page_to_bio(bio, wdata->npages - i, WRITE,
 683                                          isect, pages[i], be,
 684                                          bl_end_io_write, par);
 685                 if (IS_ERR(bio)) {
 686                         wdata->pnfs_error = PTR_ERR(bio);
 687                         bio = NULL;
 688                         goto out;
 689                 }
 690                 isect += PAGE_CACHE_SECTORS;
 691                 last_isect = isect;
 692                 extent_length -= PAGE_CACHE_SECTORS;
 693         }
 694
 695         /* Last page inside INVALID extent */
 696         if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
 697                 bio = bl_submit_bio(WRITE, bio);
 698                 temp = last_isect >> PAGE_CACHE_SECTOR_SHIFT;
 699                 npg_zero = npg_per_block - do_div(temp, npg_per_block);
 700                 if (npg_zero < npg_per_block) {
 701                         last = 1;
 702                         goto fill_invalid_ext;
 703                 }
 704         }
 705
 706 write_done:
 707         wdata->res.count = (last_isect << SECTOR_SHIFT) - (offset);
 708         if (count < wdata->res.count) {
 709                 wdata->res.count = count;
 710         }
 711 out:
 712         bl_put_extent(be);
 713         bl_submit_bio(WRITE, bio);
 714         put_parallel(par);
 715         return PNFS_ATTEMPTED;
 716 out_mds:
 717         bl_put_extent(be);
 718         kfree(par);
 719         return PNFS_NOT_ATTEMPTED;
 720 }
 721
 722 /* FIXME - range ignored */
 723 static void
 724 release_extents(struct pnfs_block_layout *bl, struct pnfs_layout_range *range)
 725 {
 726         int i;
 727         struct pnfs_block_extent *be;
 728
 729         spin_lock(&bl->bl_ext_lock);
 730         for (i = 0; i < EXTENT_LISTS; i++) {
 731                 while (!list_empty(&bl->bl_extents[i])) {
 732                         be = list_first_entry(&bl->bl_extents[i],
 733                                               struct pnfs_block_extent,
 734                                               be_node);
 735                         list_del(&be->be_node);
 736                         bl_put_extent(be);
 737                 }
 738         }
 739         spin_unlock(&bl->bl_ext_lock);
 740 }
 741
 742 static void
 743 release_inval_marks(struct pnfs_inval_markings *marks)
 744 {
 745         struct pnfs_inval_tracking *pos, *temp;
 746         struct pnfs_block_short_extent *se, *stemp;
 747
 748         list_for_each_entry_safe(pos, temp, &marks->im_tree.mtt_stub, it_link) {
 749                 list_del(&pos->it_link);
 750                 kfree(pos);
 751         }
 752
 753         list_for_each_entry_safe(se, stemp, &marks->im_extents, bse_node) {
 754                 list_del(&se->bse_node);
 755                 kfree(se);
 756         }
 757         return;
 758 }
 759
 760 static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo)
 761 {
 762         struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
 763
 764         dprintk("%s enter\n", __func__);
 765         release_extents(bl, NULL);
 766         release_inval_marks(&bl->bl_inval);
 767         kfree(bl);
 768 }
 769
 770 static struct pnfs_layout_hdr *bl_alloc_layout_hdr(struct inode *inode,
 771                                                    gfp_t gfp_flags)
 772 {
 773         struct pnfs_block_layout *bl;
 774
 775         dprintk("%s enter\n", __func__);
 776         bl = kzalloc(sizeof(*bl), gfp_flags);
 777         if (!bl)
 778                 return NULL;
 779         spin_lock_init(&bl->bl_ext_lock);
 780         INIT_LIST_HEAD(&bl->bl_extents[0]);
 781         INIT_LIST_HEAD(&bl->bl_extents[1]);
 782         INIT_LIST_HEAD(&bl->bl_commit);
 783         INIT_LIST_HEAD(&bl->bl_committing);
 784         bl->bl_count = 0;
 785         bl->bl_blocksize = NFS_SERVER(inode)->pnfs_blksize >> SECTOR_SHIFT;
 786         BL_INIT_INVAL_MARKS(&bl->bl_inval, bl->bl_blocksize);
 787         return &bl->bl_layout;
 788 }
 789
 790 static void bl_free_lseg(struct pnfs_layout_segment *lseg)
 791 {
 792         dprintk("%s enter\n", __func__);
 793         kfree(lseg);
 794 }
 795
 796 /* We pretty much ignore lseg, and store all data layout wide, so we
 797  * can correctly merge.
 798  */
 799 static struct pnfs_layout_segment *bl_alloc_lseg(struct pnfs_layout_hdr *lo,
 800                                                  struct nfs4_layoutget_res *lgr,
 801                                                  gfp_t gfp_flags)
 802 {
 803         struct pnfs_layout_segment *lseg;
 804         int status;
 805
 806         dprintk("%s enter\n", __func__);
 807         lseg = kzalloc(sizeof(*lseg), gfp_flags);
 808         if (!lseg)
 809                 return ERR_PTR(-ENOMEM);
 810         status = nfs4_blk_process_layoutget(lo, lgr, gfp_flags);
 811         if (status) {
 812                 /* We don't want to call the full-blown bl_free_lseg,
 813                  * since on error extents were not touched.
 814                  */
 815                 kfree(lseg);
 816                 return ERR_PTR(status);
 817         }
 818         return lseg;
 819 }
 820
 821 static void
 822 bl_encode_layoutcommit(struct pnfs_layout_hdr *lo, struct xdr_stream *xdr,
 823                        const struct nfs4_layoutcommit_args *arg)
 824 {
 825         dprintk("%s enter\n", __func__);
 826         encode_pnfs_block_layoutupdate(BLK_LO2EXT(lo), xdr, arg);
 827 }
 828
 829 static void
 830 bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata)
 831 {
 832         struct pnfs_layout_hdr *lo = NFS_I(lcdata->args.inode)->layout;
 833
 834         dprintk("%s enter\n", __func__);
 835         clean_pnfs_block_layoutupdate(BLK_LO2EXT(lo), &lcdata->args, lcdata->res.status);
 836 }
 837
 838 static void free_blk_mountid(struct block_mount_id *mid)
 839 {
 840         if (mid) {
 841                 struct pnfs_block_dev *dev, *tmp;
 842
 843                 /* No need to take bm_lock as we are last user freeing bm_devlist */
 844                 list_for_each_entry_safe(dev, tmp, &mid->bm_devlist, bm_node) {
 845                         list_del(&dev->bm_node);
 846                         bl_free_block_dev(dev);
 847                 }
 848                 kfree(mid);
 849         }
 850 }
 851
 852 /* This is mostly copied from the filelayout's get_device_info function.
 853  * It seems much of this should be at the generic pnfs level.
 854  */
 855 static struct pnfs_block_dev *
 856 nfs4_blk_get_deviceinfo(struct nfs_server *server, const struct nfs_fh *fh,
 857                         struct nfs4_deviceid *d_id)
 858 {
 859         struct pnfs_device *dev;
 860         struct pnfs_block_dev *rv;
 861         u32 max_resp_sz;
 862         int max_pages;
 863         struct page **pages = NULL;
 864         int i, rc;
 865
 866         /*
 867          * Use the session max response size as the basis for setting
 868          * GETDEVICEINFO's maxcount
 869          */
 870         max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz;
 871         max_pages = max_resp_sz >> PAGE_SHIFT;
 872         dprintk("%s max_resp_sz %u max_pages %d\n",
 873                 __func__, max_resp_sz, max_pages);
 874
 875         dev = kmalloc(sizeof(*dev), GFP_NOFS);
 876         if (!dev) {
 877                 dprintk("%s kmalloc failed\n", __func__);
 878                 return ERR_PTR(-ENOMEM);
 879         }
 880
 881         pages = kzalloc(max_pages * sizeof(struct page *), GFP_NOFS);
 882         if (pages == NULL) {
 883                 kfree(dev);
 884                 return ERR_PTR(-ENOMEM);
 885         }
 886         for (i = 0; i < max_pages; i++) {
 887                 pages[i] = alloc_page(GFP_NOFS);
 888                 if (!pages[i]) {
 889                         rv = ERR_PTR(-ENOMEM);
 890                         goto out_free;
 891                 }
 892         }
 893
 894         memcpy(&dev->dev_id, d_id, sizeof(*d_id));
 895         dev->layout_type = LAYOUT_BLOCK_VOLUME;
 896         dev->pages = pages;
 897         dev->pgbase = 0;
 898         dev->pglen = PAGE_SIZE * max_pages;
 899         dev->mincount = 0;
 900
 901         dprintk("%s: dev_id: %s\n", __func__, dev->dev_id.data);
 902         rc = nfs4_proc_getdeviceinfo(server, dev);
 903         dprintk("%s getdevice info returns %d\n", __func__, rc);
 904         if (rc) {
 905                 rv = ERR_PTR(rc);
 906                 goto out_free;
 907         }
 908
 909         rv = nfs4_blk_decode_device(server, dev);
 910  out_free:
 911         for (i = 0; i < max_pages; i++)
 912                 __free_page(pages[i]);
 913         kfree(pages);
 914         kfree(dev);
 915         return rv;
 916 }
 917
 918 static int
 919 bl_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *fh)
 920 {
 921         struct block_mount_id *b_mt_id = NULL;
 922         struct pnfs_devicelist *dlist = NULL;
 923         struct pnfs_block_dev *bdev;
 924         LIST_HEAD(block_disklist);
 925         int status, i;
 926
 927         dprintk("%s enter\n", __func__);
 928
 929         if (server->pnfs_blksize == 0) {
 930                 dprintk("%s Server did not return blksize\n", __func__);
 931                 return -EINVAL;
 932         }
 933         b_mt_id = kzalloc(sizeof(struct block_mount_id), GFP_NOFS);
 934         if (!b_mt_id) {
 935                 status = -ENOMEM;
 936                 goto out_error;
 937         }
 938         /* Initialize nfs4 block layout mount id */
 939         spin_lock_init(&b_mt_id->bm_lock);
 940         INIT_LIST_HEAD(&b_mt_id->bm_devlist);
 941
 942         dlist = kmalloc(sizeof(struct pnfs_devicelist), GFP_NOFS);
 943         if (!dlist) {
 944                 status = -ENOMEM;
 945                 goto out_error;
 946         }
 947         dlist->eof = 0;
 948         while (!dlist->eof) {
 949                 status = nfs4_proc_getdevicelist(server, fh, dlist);
 950                 if (status)
 951                         goto out_error;
 952                 dprintk("%s GETDEVICELIST numdevs=%i, eof=%i\n",
 953                         __func__, dlist->num_devs, dlist->eof);
 954                 for (i = 0; i < dlist->num_devs; i++) {
 955                         bdev = nfs4_blk_get_deviceinfo(server, fh,
 956                                                        &dlist->dev_id[i]);
 957                         if (IS_ERR(bdev)) {
 958                                 status = PTR_ERR(bdev);
 959                                 goto out_error;
 960                         }
 961                         spin_lock(&b_mt_id->bm_lock);
 962                         list_add(&bdev->bm_node, &b_mt_id->bm_devlist);
 963                         spin_unlock(&b_mt_id->bm_lock);
 964                 }
 965         }
 966         dprintk("%s SUCCESS\n", __func__);
 967         server->pnfs_ld_data = b_mt_id;
 968
 969  out_return:
 970         kfree(dlist);
 971         return status;
 972
 973  out_error:
 974         free_blk_mountid(b_mt_id);
 975         goto out_return;
 976 }
 977
 978 static int
 979 bl_clear_layoutdriver(struct nfs_server *server)
 980 {
 981         struct block_mount_id *b_mt_id = server->pnfs_ld_data;
 982
 983         dprintk("%s enter\n", __func__);
 984         free_blk_mountid(b_mt_id);
 985         dprintk("%s RETURNS\n", __func__);
 986         return 0;
 987 }
 988
 989 static const struct nfs_pageio_ops bl_pg_read_ops = {
 990         .pg_init = pnfs_generic_pg_init_read,
 991         .pg_test = pnfs_generic_pg_test,
 992         .pg_doio = pnfs_generic_pg_readpages,
 993 };
 994
 995 static const struct nfs_pageio_ops bl_pg_write_ops = {
 996         .pg_init = pnfs_generic_pg_init_write,
 997         .pg_test = pnfs_generic_pg_test,
 998         .pg_doio = pnfs_generic_pg_writepages,
 999 };
1000
1001 static struct pnfs_layoutdriver_type blocklayout_type = {
1002         .id                             = LAYOUT_BLOCK_VOLUME,
1003         .name                           = "LAYOUT_BLOCK_VOLUME",
1004         .read_pagelist                  = bl_read_pagelist,
1005         .write_pagelist                 = bl_write_pagelist,
1006         .alloc_layout_hdr               = bl_alloc_layout_hdr,
1007         .free_layout_hdr                = bl_free_layout_hdr,
1008         .alloc_lseg                     = bl_alloc_lseg,
1009         .free_lseg                      = bl_free_lseg,
1010         .encode_layoutcommit            = bl_encode_layoutcommit,
1011         .cleanup_layoutcommit           = bl_cleanup_layoutcommit,
1012         .set_layoutdriver               = bl_set_layoutdriver,
1013         .clear_layoutdriver             = bl_clear_layoutdriver,
1014         .pg_read_ops                    = &bl_pg_read_ops,
1015         .pg_write_ops                   = &bl_pg_write_ops,
1016 };
1017
1018 static const struct rpc_pipe_ops bl_upcall_ops = {
1019         .upcall         = rpc_pipe_generic_upcall,
1020         .downcall       = bl_pipe_downcall,
1021         .destroy_msg    = bl_pipe_destroy_msg,
1022 };
1023
1024 static struct dentry *nfs4blocklayout_register_sb(struct super_block *sb,
1025                                             struct rpc_pipe *pipe)
1026 {
1027         struct dentry *dir, *dentry;
1028
1029         dir = rpc_d_lookup_sb(sb, NFS_PIPE_DIRNAME);
1030         if (dir == NULL)
1031                 return ERR_PTR(-ENOENT);
1032         dentry = rpc_mkpipe_dentry(dir, "blocklayout", NULL, pipe);
1033         dput(dir);
1034         return dentry;
1035 }
1036
1037 static void nfs4blocklayout_unregister_sb(struct super_block *sb,
1038                                           struct rpc_pipe *pipe)
1039 {
1040         if (pipe->dentry)
1041                 rpc_unlink(pipe->dentry);
1042 }
1043
1044 static int rpc_pipefs_event(struct notifier_block *nb, unsigned long event,
1045                            void *ptr)
1046 {
1047         struct super_block *sb = ptr;
1048         struct net *net = sb->s_fs_info;
1049         struct nfs_net *nn = net_generic(net, nfs_net_id);
1050         struct dentry *dentry;
1051         int ret = 0;
1052
1053         if (!try_module_get(THIS_MODULE))
1054                 return 0;
1055
1056         if (nn->bl_device_pipe == NULL) {
1057                 module_put(THIS_MODULE);
1058                 return 0;
1059         }
1060
1061         switch (event) {
1062         case RPC_PIPEFS_MOUNT:
1063                 dentry = nfs4blocklayout_register_sb(sb, nn->bl_device_pipe);
1064                 if (IS_ERR(dentry)) {
1065                         ret = PTR_ERR(dentry);
1066                         break;
1067                 }
1068                 nn->bl_device_pipe->dentry = dentry;
1069                 break;
1070         case RPC_PIPEFS_UMOUNT:
1071                 if (nn->bl_device_pipe->dentry)
1072                         nfs4blocklayout_unregister_sb(sb, nn->bl_device_pipe);
1073                 break;
1074         default:
1075                 ret = -ENOTSUPP;
1076                 break;
1077         }
1078         module_put(THIS_MODULE);
1079         return ret;
1080 }
1081
1082 static struct notifier_block nfs4blocklayout_block = {
1083         .notifier_call = rpc_pipefs_event,
1084 };
1085
1086 static struct dentry *nfs4blocklayout_register_net(struct net *net,
1087                                                    struct rpc_pipe *pipe)
1088 {
1089         struct super_block *pipefs_sb;
1090         struct dentry *dentry;
1091
1092         pipefs_sb = rpc_get_sb_net(net);
1093         if (!pipefs_sb)
1094                 return NULL;
1095         dentry = nfs4blocklayout_register_sb(pipefs_sb, pipe);
1096         rpc_put_sb_net(net);
1097         return dentry;
1098 }
1099
1100 static void nfs4blocklayout_unregister_net(struct net *net,
1101                                            struct rpc_pipe *pipe)
1102 {
1103         struct super_block *pipefs_sb;
1104
1105         pipefs_sb = rpc_get_sb_net(net);
1106         if (pipefs_sb) {
1107                 nfs4blocklayout_unregister_sb(pipefs_sb, pipe);
1108                 rpc_put_sb_net(net);
1109         }
1110 }
1111
1112 static int nfs4blocklayout_net_init(struct net *net)
1113 {
1114         struct nfs_net *nn = net_generic(net, nfs_net_id);
1115         struct dentry *dentry;
1116
1117         init_waitqueue_head(&nn->bl_wq);
1118         nn->bl_device_pipe = rpc_mkpipe_data(&bl_upcall_ops, 0);
1119         if (IS_ERR(nn->bl_device_pipe))
1120                 return PTR_ERR(nn->bl_device_pipe);
1121         dentry = nfs4blocklayout_register_net(net, nn->bl_device_pipe);
1122         if (IS_ERR(dentry)) {
1123                 rpc_destroy_pipe_data(nn->bl_device_pipe);
1124                 return PTR_ERR(dentry);
1125         }
1126         nn->bl_device_pipe->dentry = dentry;
1127         return 0;
1128 }
1129
1130 static void nfs4blocklayout_net_exit(struct net *net)
1131 {
1132         struct nfs_net *nn = net_generic(net, nfs_net_id);
1133
1134         nfs4blocklayout_unregister_net(net, nn->bl_device_pipe);
1135         rpc_destroy_pipe_data(nn->bl_device_pipe);
1136         nn->bl_device_pipe = NULL;
1137 }
1138
1139 static struct pernet_operations nfs4blocklayout_net_ops = {
1140         .init = nfs4blocklayout_net_init,
1141         .exit = nfs4blocklayout_net_exit,
1142 };
1143
1144 static int __init nfs4blocklayout_init(void)
1145 {
1146         int ret;
1147
1148         dprintk("%s: NFSv4 Block Layout Driver Registering...\n", __func__);
1149
1150         ret = pnfs_register_layoutdriver(&blocklayout_type);
1151         if (ret)
1152                 goto out;
1153
1154         ret = rpc_pipefs_notifier_register(&nfs4blocklayout_block);
1155         if (ret)
1156                 goto out_remove;
1157         ret = register_pernet_subsys(&nfs4blocklayout_net_ops);
1158         if (ret)
1159                 goto out_notifier;
1160 out:
1161         return ret;
1162
1163 out_notifier:
1164         rpc_pipefs_notifier_unregister(&nfs4blocklayout_block);
1165 out_remove:
1166         pnfs_unregister_layoutdriver(&blocklayout_type);
1167         return ret;
1168 }
1169
1170 static void __exit nfs4blocklayout_exit(void)
1171 {
1172         dprintk("%s: NFSv4 Block Layout Driver Unregistering...\n",
1173                __func__);
1174
1175         rpc_pipefs_notifier_unregister(&nfs4blocklayout_block);
1176         unregister_pernet_subsys(&nfs4blocklayout_net_ops);
1177         pnfs_unregister_layoutdriver(&blocklayout_type);
1178 }
1179
1180 MODULE_ALIAS("nfs-layouttype4-3");
1181
1182 module_init(nfs4blocklayout_init);
1183 module_exit(nfs4blocklayout_exit);