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 #include <linux/pagevec.h>
  41
  42 #include "../pnfs.h"
  43 #include "../nfs4session.h"
  44 #include "../internal.h"
  45 #include "blocklayout.h"
  46
  47 #define NFSDBG_FACILITY NFSDBG_PNFS_LD
  48
  49 MODULE_LICENSE("GPL");
  50 MODULE_AUTHOR("Andy Adamson <andros@citi.umich.edu>");
  51 MODULE_DESCRIPTION("The NFSv4.1 pNFS Block layout driver");
  52
  53 static void print_page(struct page *page)
  54 {
  55         dprintk("PRINTPAGE page %p\n", page);
  56         dprintk("       PagePrivate %d\n", PagePrivate(page));
  57         dprintk("       PageUptodate %d\n", PageUptodate(page));
  58         dprintk("       PageError %d\n", PageError(page));
  59         dprintk("       PageDirty %d\n", PageDirty(page));
  60         dprintk("       PageReferenced %d\n", PageReferenced(page));
  61         dprintk("       PageLocked %d\n", PageLocked(page));
  62         dprintk("       PageWriteback %d\n", PageWriteback(page));
  63         dprintk("       PageMappedToDisk %d\n", PageMappedToDisk(page));
  64         dprintk("\n");
  65 }
  66
  67 /* Given the be associated with isect, determine if page data needs to be
  68  * initialized.
  69  */
  70 static int is_hole(struct pnfs_block_extent *be, sector_t isect)
  71 {
  72         if (be->be_state == PNFS_BLOCK_NONE_DATA)
  73                 return 1;
  74         else if (be->be_state != PNFS_BLOCK_INVALID_DATA)
  75                 return 0;
  76         else
  77                 return !bl_is_sector_init(be->be_inval, isect);
  78 }
  79
  80 /* Given the be associated with isect, determine if page data can be
  81  * written to disk.
  82  */
  83 static int is_writable(struct pnfs_block_extent *be, sector_t isect)
  84 {
  85         return (be->be_state == PNFS_BLOCK_READWRITE_DATA ||
  86                 be->be_state == PNFS_BLOCK_INVALID_DATA);
  87 }
  88
  89 /* The data we are handed might be spread across several bios.  We need
  90  * to track when the last one is finished.
  91  */
  92 struct parallel_io {
  93         struct kref refcnt;
  94         void (*pnfs_callback) (void *data, int num_se);
  95         void *data;
  96         int bse_count;
  97 };
  98
  99 static inline struct parallel_io *alloc_parallel(void *data)
 100 {
 101         struct parallel_io *rv;
 102
 103         rv  = kmalloc(sizeof(*rv), GFP_NOFS);
 104         if (rv) {
 105                 rv->data = data;
 106                 kref_init(&rv->refcnt);
 107                 rv->bse_count = 0;
 108         }
 109         return rv;
 110 }
 111
 112 static inline void get_parallel(struct parallel_io *p)
 113 {
 114         kref_get(&p->refcnt);
 115 }
 116
 117 static void destroy_parallel(struct kref *kref)
 118 {
 119         struct parallel_io *p = container_of(kref, struct parallel_io, refcnt);
 120
 121         dprintk("%s enter\n", __func__);
 122         p->pnfs_callback(p->data, p->bse_count);
 123         kfree(p);
 124 }
 125
 126 static inline void put_parallel(struct parallel_io *p)
 127 {
 128         kref_put(&p->refcnt, destroy_parallel);
 129 }
 130
 131 static struct bio *
 132 bl_submit_bio(int rw, struct bio *bio)
 133 {
 134         if (bio) {
 135                 get_parallel(bio->bi_private);
 136                 dprintk("%s submitting %s bio %u@%llu\n", __func__,
 137                         rw == READ ? "read" : "write", bio->bi_iter.bi_size,
 138                         (unsigned long long)bio->bi_iter.bi_sector);
 139                 submit_bio(rw, bio);
 140         }
 141         return NULL;
 142 }
 143
 144 static struct bio *bl_alloc_init_bio(int npg, sector_t isect,
 145                                      struct pnfs_block_extent *be,
 146                                      void (*end_io)(struct bio *, int err),
 147                                      struct parallel_io *par)
 148 {
 149         struct bio *bio;
 150
 151         npg = min(npg, BIO_MAX_PAGES);
 152         bio = bio_alloc(GFP_NOIO, npg);
 153         if (!bio && (current->flags & PF_MEMALLOC)) {
 154                 while (!bio && (npg /= 2))
 155                         bio = bio_alloc(GFP_NOIO, npg);
 156         }
 157
 158         if (bio) {
 159                 bio->bi_iter.bi_sector = isect - be->be_f_offset +
 160                         be->be_v_offset;
 161                 bio->bi_bdev = be->be_mdev;
 162                 bio->bi_end_io = end_io;
 163                 bio->bi_private = par;
 164         }
 165         return bio;
 166 }
 167
 168 static struct bio *do_add_page_to_bio(struct bio *bio, int npg, int rw,
 169                                       sector_t isect, struct page *page,
 170                                       struct pnfs_block_extent *be,
 171                                       void (*end_io)(struct bio *, int err),
 172                                       struct parallel_io *par,
 173                                       unsigned int offset, int len)
 174 {
 175         isect = isect + (offset >> SECTOR_SHIFT);
 176         dprintk("%s: npg %d rw %d isect %llu offset %u len %d\n", __func__,
 177                 npg, rw, (unsigned long long)isect, offset, len);
 178 retry:
 179         if (!bio) {
 180                 bio = bl_alloc_init_bio(npg, isect, be, end_io, par);
 181                 if (!bio)
 182                         return ERR_PTR(-ENOMEM);
 183         }
 184         if (bio_add_page(bio, page, len, offset) < len) {
 185                 bio = bl_submit_bio(rw, bio);
 186                 goto retry;
 187         }
 188         return bio;
 189 }
 190
 191 static struct bio *bl_add_page_to_bio(struct bio *bio, int npg, int rw,
 192                                       sector_t isect, struct page *page,
 193                                       struct pnfs_block_extent *be,
 194                                       void (*end_io)(struct bio *, int err),
 195                                       struct parallel_io *par)
 196 {
 197         return do_add_page_to_bio(bio, npg, rw, isect, page, be,
 198                                   end_io, par, 0, PAGE_CACHE_SIZE);
 199 }
 200
 201 /* This is basically copied from mpage_end_io_read */
 202 static void bl_end_io_read(struct bio *bio, int err)
 203 {
 204         struct parallel_io *par = bio->bi_private;
 205         struct bio_vec *bvec;
 206         int i;
 207
 208         if (!err)
 209                 bio_for_each_segment_all(bvec, bio, i)
 210                         SetPageUptodate(bvec->bv_page);
 211
 212         if (err) {
 213                 struct nfs_pgio_header *header = par->data;
 214
 215                 if (!header->pnfs_error)
 216                         header->pnfs_error = -EIO;
 217                 pnfs_set_lo_fail(header->lseg);
 218         }
 219         bio_put(bio);
 220         put_parallel(par);
 221 }
 222
 223 static void bl_read_cleanup(struct work_struct *work)
 224 {
 225         struct rpc_task *task;
 226         struct nfs_pgio_header *hdr;
 227         dprintk("%s enter\n", __func__);
 228         task = container_of(work, struct rpc_task, u.tk_work);
 229         hdr = container_of(task, struct nfs_pgio_header, task);
 230         pnfs_ld_read_done(hdr);
 231 }
 232
 233 static void
 234 bl_end_par_io_read(void *data, int unused)
 235 {
 236         struct nfs_pgio_header *hdr = data;
 237
 238         hdr->task.tk_status = hdr->pnfs_error;
 239         INIT_WORK(&hdr->task.u.tk_work, bl_read_cleanup);
 240         schedule_work(&hdr->task.u.tk_work);
 241 }
 242
 243 static enum pnfs_try_status
 244 bl_read_pagelist(struct nfs_pgio_header *hdr)
 245 {
 246         struct nfs_pgio_header *header = hdr;
 247         int i, hole;
 248         struct bio *bio = NULL;
 249         struct pnfs_block_extent *be = NULL, *cow_read = NULL;
 250         sector_t isect, extent_length = 0;
 251         struct parallel_io *par;
 252         loff_t f_offset = hdr->args.offset;
 253         size_t bytes_left = hdr->args.count;
 254         unsigned int pg_offset, pg_len;
 255         struct page **pages = hdr->args.pages;
 256         int pg_index = hdr->args.pgbase >> PAGE_CACHE_SHIFT;
 257         const bool is_dio = (header->dreq != NULL);
 258
 259         dprintk("%s enter nr_pages %u offset %lld count %u\n", __func__,
 260                 hdr->page_array.npages, f_offset,
 261                 (unsigned int)hdr->args.count);
 262
 263         par = alloc_parallel(hdr);
 264         if (!par)
 265                 goto use_mds;
 266         par->pnfs_callback = bl_end_par_io_read;
 267         /* At this point, we can no longer jump to use_mds */
 268
 269         isect = (sector_t) (f_offset >> SECTOR_SHIFT);
 270         /* Code assumes extents are page-aligned */
 271         for (i = pg_index; i < hdr->page_array.npages; i++) {
 272                 if (!extent_length) {
 273                         /* We've used up the previous extent */
 274                         bl_put_extent(be);
 275                         bl_put_extent(cow_read);
 276                         bio = bl_submit_bio(READ, bio);
 277                         /* Get the next one */
 278                         be = bl_find_get_extent(BLK_LSEG2EXT(header->lseg),
 279                                              isect, &cow_read);
 280                         if (!be) {
 281                                 header->pnfs_error = -EIO;
 282                                 goto out;
 283                         }
 284                         extent_length = be->be_length -
 285                                 (isect - be->be_f_offset);
 286                         if (cow_read) {
 287                                 sector_t cow_length = cow_read->be_length -
 288                                         (isect - cow_read->be_f_offset);
 289                                 extent_length = min(extent_length, cow_length);
 290                         }
 291                 }
 292
 293                 if (is_dio) {
 294                         pg_offset = f_offset & ~PAGE_CACHE_MASK;
 295                         if (pg_offset + bytes_left > PAGE_CACHE_SIZE)
 296                                 pg_len = PAGE_CACHE_SIZE - pg_offset;
 297                         else
 298                                 pg_len = bytes_left;
 299
 300                         f_offset += pg_len;
 301                         bytes_left -= pg_len;
 302                         isect += (pg_offset >> SECTOR_SHIFT);
 303                 } else {
 304                         pg_offset = 0;
 305                         pg_len = PAGE_CACHE_SIZE;
 306                 }
 307
 308                 hole = is_hole(be, isect);
 309                 if (hole && !cow_read) {
 310                         bio = bl_submit_bio(READ, bio);
 311                         /* Fill hole w/ zeroes w/o accessing device */
 312                         dprintk("%s Zeroing page for hole\n", __func__);
 313                         zero_user_segment(pages[i], pg_offset, pg_len);
 314                         print_page(pages[i]);
 315                         SetPageUptodate(pages[i]);
 316                 } else {
 317                         struct pnfs_block_extent *be_read;
 318
 319                         be_read = (hole && cow_read) ? cow_read : be;
 320                         bio = do_add_page_to_bio(bio,
 321                                                  hdr->page_array.npages - i,
 322                                                  READ,
 323                                                  isect, pages[i], be_read,
 324                                                  bl_end_io_read, par,
 325                                                  pg_offset, pg_len);
 326                         if (IS_ERR(bio)) {
 327                                 header->pnfs_error = PTR_ERR(bio);
 328                                 bio = NULL;
 329                                 goto out;
 330                         }
 331                 }
 332                 isect += (pg_len >> SECTOR_SHIFT);
 333                 extent_length -= PAGE_CACHE_SECTORS;
 334         }
 335         if ((isect << SECTOR_SHIFT) >= header->inode->i_size) {
 336                 hdr->res.eof = 1;
 337                 hdr->res.count = header->inode->i_size - hdr->args.offset;
 338         } else {
 339                 hdr->res.count = (isect << SECTOR_SHIFT) - hdr->args.offset;
 340         }
 341 out:
 342         bl_put_extent(be);
 343         bl_put_extent(cow_read);
 344         bl_submit_bio(READ, bio);
 345         put_parallel(par);
 346         return PNFS_ATTEMPTED;
 347
 348  use_mds:
 349         dprintk("Giving up and using normal NFS\n");
 350         return PNFS_NOT_ATTEMPTED;
 351 }
 352
 353 static void mark_extents_written(struct pnfs_block_layout *bl,
 354                                  __u64 offset, __u32 count)
 355 {
 356         sector_t isect, end;
 357         struct pnfs_block_extent *be;
 358         struct pnfs_block_short_extent *se;
 359
 360         dprintk("%s(%llu, %u)\n", __func__, offset, count);
 361         if (count == 0)
 362                 return;
 363         isect = (offset & (long)(PAGE_CACHE_MASK)) >> SECTOR_SHIFT;
 364         end = (offset + count + PAGE_CACHE_SIZE - 1) & (long)(PAGE_CACHE_MASK);
 365         end >>= SECTOR_SHIFT;
 366         while (isect < end) {
 367                 sector_t len;
 368                 be = bl_find_get_extent(bl, isect, NULL);
 369                 BUG_ON(!be); /* FIXME */
 370                 len = min(end, be->be_f_offset + be->be_length) - isect;
 371                 if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
 372                         se = bl_pop_one_short_extent(be->be_inval);
 373                         BUG_ON(!se);
 374                         bl_mark_for_commit(be, isect, len, se);
 375                 }
 376                 isect += len;
 377                 bl_put_extent(be);
 378         }
 379 }
 380
 381 static void bl_end_io_write_zero(struct bio *bio, int err)
 382 {
 383         struct parallel_io *par = bio->bi_private;
 384         struct bio_vec *bvec;
 385         int i;
 386
 387         bio_for_each_segment_all(bvec, bio, i) {
 388                 /* This is the zeroing page we added */
 389                 end_page_writeback(bvec->bv_page);
 390                 page_cache_release(bvec->bv_page);
 391         }
 392
 393         if (unlikely(err)) {
 394                 struct nfs_pgio_header *header = par->data;
 395
 396                 if (!header->pnfs_error)
 397                         header->pnfs_error = -EIO;
 398                 pnfs_set_lo_fail(header->lseg);
 399         }
 400         bio_put(bio);
 401         put_parallel(par);
 402 }
 403
 404 static void bl_end_io_write(struct bio *bio, int err)
 405 {
 406         struct parallel_io *par = bio->bi_private;
 407         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 408         struct nfs_pgio_header *header = par->data;
 409
 410         if (!uptodate) {
 411                 if (!header->pnfs_error)
 412                         header->pnfs_error = -EIO;
 413                 pnfs_set_lo_fail(header->lseg);
 414         }
 415         bio_put(bio);
 416         put_parallel(par);
 417 }
 418
 419 /* Function scheduled for call during bl_end_par_io_write,
 420  * it marks sectors as written and extends the commitlist.
 421  */
 422 static void bl_write_cleanup(struct work_struct *work)
 423 {
 424         struct rpc_task *task;
 425         struct nfs_pgio_header *hdr;
 426         dprintk("%s enter\n", __func__);
 427         task = container_of(work, struct rpc_task, u.tk_work);
 428         hdr = container_of(task, struct nfs_pgio_header, task);
 429         if (likely(!hdr->pnfs_error)) {
 430                 /* Marks for LAYOUTCOMMIT */
 431                 mark_extents_written(BLK_LSEG2EXT(hdr->lseg),
 432                                      hdr->args.offset, hdr->args.count);
 433         }
 434         pnfs_ld_write_done(hdr);
 435 }
 436
 437 /* Called when last of bios associated with a bl_write_pagelist call finishes */
 438 static void bl_end_par_io_write(void *data, int num_se)
 439 {
 440         struct nfs_pgio_header *hdr = data;
 441
 442         if (unlikely(hdr->pnfs_error)) {
 443                 bl_free_short_extents(&BLK_LSEG2EXT(hdr->lseg)->bl_inval,
 444                                         num_se);
 445         }
 446
 447         hdr->task.tk_status = hdr->pnfs_error;
 448         hdr->verf.committed = NFS_FILE_SYNC;
 449         INIT_WORK(&hdr->task.u.tk_work, bl_write_cleanup);
 450         schedule_work(&hdr->task.u.tk_work);
 451 }
 452
 453 /* FIXME STUB - mark intersection of layout and page as bad, so is not
 454  * used again.
 455  */
 456 static void mark_bad_read(void)
 457 {
 458         return;
 459 }
 460
 461 /*
 462  * map_block:  map a requested I/0 block (isect) into an offset in the LVM
 463  * block_device
 464  */
 465 static void
 466 map_block(struct buffer_head *bh, sector_t isect, struct pnfs_block_extent *be)
 467 {
 468         dprintk("%s enter be=%p\n", __func__, be);
 469
 470         set_buffer_mapped(bh);
 471         bh->b_bdev = be->be_mdev;
 472         bh->b_blocknr = (isect - be->be_f_offset + be->be_v_offset) >>
 473             (be->be_mdev->bd_inode->i_blkbits - SECTOR_SHIFT);
 474
 475         dprintk("%s isect %llu, bh->b_blocknr %ld, using bsize %Zd\n",
 476                 __func__, (unsigned long long)isect, (long)bh->b_blocknr,
 477                 bh->b_size);
 478         return;
 479 }
 480
 481 static void
 482 bl_read_single_end_io(struct bio *bio, int error)
 483 {
 484         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
 485         struct page *page = bvec->bv_page;
 486
 487         /* Only one page in bvec */
 488         unlock_page(page);
 489 }
 490
 491 static int
 492 bl_do_readpage_sync(struct page *page, struct pnfs_block_extent *be,
 493                     unsigned int offset, unsigned int len)
 494 {
 495         struct bio *bio;
 496         struct page *shadow_page;
 497         sector_t isect;
 498         char *kaddr, *kshadow_addr;
 499         int ret = 0;
 500
 501         dprintk("%s: offset %u len %u\n", __func__, offset, len);
 502
 503         shadow_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
 504         if (shadow_page == NULL)
 505                 return -ENOMEM;
 506
 507         bio = bio_alloc(GFP_NOIO, 1);
 508         if (bio == NULL)
 509                 return -ENOMEM;
 510
 511         isect = (page->index << PAGE_CACHE_SECTOR_SHIFT) +
 512                 (offset / SECTOR_SIZE);
 513
 514         bio->bi_iter.bi_sector = isect - be->be_f_offset + be->be_v_offset;
 515         bio->bi_bdev = be->be_mdev;
 516         bio->bi_end_io = bl_read_single_end_io;
 517
 518         lock_page(shadow_page);
 519         if (bio_add_page(bio, shadow_page,
 520                          SECTOR_SIZE, round_down(offset, SECTOR_SIZE)) == 0) {
 521                 unlock_page(shadow_page);
 522                 bio_put(bio);
 523                 return -EIO;
 524         }
 525
 526         submit_bio(READ, bio);
 527         wait_on_page_locked(shadow_page);
 528         if (unlikely(!test_bit(BIO_UPTODATE, &bio->bi_flags))) {
 529                 ret = -EIO;
 530         } else {
 531                 kaddr = kmap_atomic(page);
 532                 kshadow_addr = kmap_atomic(shadow_page);
 533                 memcpy(kaddr + offset, kshadow_addr + offset, len);
 534                 kunmap_atomic(kshadow_addr);
 535                 kunmap_atomic(kaddr);
 536         }
 537         __free_page(shadow_page);
 538         bio_put(bio);
 539
 540         return ret;
 541 }
 542
 543 static int
 544 bl_read_partial_page_sync(struct page *page, struct pnfs_block_extent *be,
 545                           unsigned int dirty_offset, unsigned int dirty_len,
 546                           bool full_page)
 547 {
 548         int ret = 0;
 549         unsigned int start, end;
 550
 551         if (full_page) {
 552                 start = 0;
 553                 end = PAGE_CACHE_SIZE;
 554         } else {
 555                 start = round_down(dirty_offset, SECTOR_SIZE);
 556                 end = round_up(dirty_offset + dirty_len, SECTOR_SIZE);
 557         }
 558
 559         dprintk("%s: offset %u len %d\n", __func__, dirty_offset, dirty_len);
 560         if (!be) {
 561                 zero_user_segments(page, start, dirty_offset,
 562                                    dirty_offset + dirty_len, end);
 563                 if (start == 0 && end == PAGE_CACHE_SIZE &&
 564                     trylock_page(page)) {
 565                         SetPageUptodate(page);
 566                         unlock_page(page);
 567                 }
 568                 return ret;
 569         }
 570
 571         if (start != dirty_offset)
 572                 ret = bl_do_readpage_sync(page, be, start, dirty_offset - start);
 573
 574         if (!ret && (dirty_offset + dirty_len < end))
 575                 ret = bl_do_readpage_sync(page, be, dirty_offset + dirty_len,
 576                                           end - dirty_offset - dirty_len);
 577
 578         return ret;
 579 }
 580
 581 /* Given an unmapped page, zero it or read in page for COW, page is locked
 582  * by caller.
 583  */
 584 static int
 585 init_page_for_write(struct page *page, struct pnfs_block_extent *cow_read)
 586 {
 587         struct buffer_head *bh = NULL;
 588         int ret = 0;
 589         sector_t isect;
 590
 591         dprintk("%s enter, %p\n", __func__, page);
 592         BUG_ON(PageUptodate(page));
 593         if (!cow_read) {
 594                 zero_user_segment(page, 0, PAGE_SIZE);
 595                 SetPageUptodate(page);
 596                 goto cleanup;
 597         }
 598
 599         bh = alloc_page_buffers(page, PAGE_CACHE_SIZE, 0);
 600         if (!bh) {
 601                 ret = -ENOMEM;
 602                 goto cleanup;
 603         }
 604
 605         isect = (sector_t) page->index << PAGE_CACHE_SECTOR_SHIFT;
 606         map_block(bh, isect, cow_read);
 607         if (!bh_uptodate_or_lock(bh))
 608                 ret = bh_submit_read(bh);
 609         if (ret)
 610                 goto cleanup;
 611         SetPageUptodate(page);
 612
 613 cleanup:
 614         if (bh)
 615                 free_buffer_head(bh);
 616         if (ret) {
 617                 /* Need to mark layout with bad read...should now
 618                  * just use nfs4 for reads and writes.
 619                  */
 620                 mark_bad_read();
 621         }
 622         return ret;
 623 }
 624
 625 /* Find or create a zeroing page marked being writeback.
 626  * Return ERR_PTR on error, NULL to indicate skip this page and page itself
 627  * to indicate write out.
 628  */
 629 static struct page *
 630 bl_find_get_zeroing_page(struct inode *inode, pgoff_t index,
 631                         struct pnfs_block_extent *cow_read)
 632 {
 633         struct page *page;
 634         int locked = 0;
 635         page = find_get_page(inode->i_mapping, index);
 636         if (page)
 637                 goto check_page;
 638
 639         page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
 640         if (unlikely(!page)) {
 641                 dprintk("%s oom\n", __func__);
 642                 return ERR_PTR(-ENOMEM);
 643         }
 644         locked = 1;
 645
 646 check_page:
 647         /* PageDirty: Other will write this out
 648          * PageWriteback: Other is writing this out
 649          * PageUptodate: It was read before
 650          */
 651         if (PageDirty(page) || PageWriteback(page)) {
 652                 print_page(page);
 653                 if (locked)
 654                         unlock_page(page);
 655                 page_cache_release(page);
 656                 return NULL;
 657         }
 658
 659         if (!locked) {
 660                 lock_page(page);
 661                 locked = 1;
 662                 goto check_page;
 663         }
 664         if (!PageUptodate(page)) {
 665                 /* New page, readin or zero it */
 666                 init_page_for_write(page, cow_read);
 667         }
 668         set_page_writeback(page);
 669         unlock_page(page);
 670
 671         return page;
 672 }
 673
 674 static enum pnfs_try_status
 675 bl_write_pagelist(struct nfs_pgio_header *header, int sync)
 676 {
 677         int i, ret, npg_zero, pg_index, last = 0;
 678         struct bio *bio = NULL;
 679         struct pnfs_block_extent *be = NULL, *cow_read = NULL;
 680         sector_t isect, last_isect = 0, extent_length = 0;
 681         struct parallel_io *par = NULL;
 682         loff_t offset = header->args.offset;
 683         size_t count = header->args.count;
 684         unsigned int pg_offset, pg_len, saved_len;
 685         struct page **pages = header->args.pages;
 686         struct page *page;
 687         pgoff_t index;
 688         u64 temp;
 689         int npg_per_block =
 690             NFS_SERVER(header->inode)->pnfs_blksize >> PAGE_CACHE_SHIFT;
 691
 692         dprintk("%s enter, %Zu@%lld\n", __func__, count, offset);
 693
 694         if (header->dreq != NULL &&
 695             (!IS_ALIGNED(offset, NFS_SERVER(header->inode)->pnfs_blksize) ||
 696              !IS_ALIGNED(count, NFS_SERVER(header->inode)->pnfs_blksize))) {
 697                 dprintk("pnfsblock nonblock aligned DIO writes. Resend MDS\n");
 698                 goto out_mds;
 699         }
 700         /* At this point, header->page_aray is a (sequential) list of nfs_pages.
 701          * We want to write each, and if there is an error set pnfs_error
 702          * to have it redone using nfs.
 703          */
 704         par = alloc_parallel(header);
 705         if (!par)
 706                 goto out_mds;
 707         par->pnfs_callback = bl_end_par_io_write;
 708         /* At this point, have to be more careful with error handling */
 709
 710         isect = (sector_t) ((offset & (long)PAGE_CACHE_MASK) >> SECTOR_SHIFT);
 711         be = bl_find_get_extent(BLK_LSEG2EXT(header->lseg), isect, &cow_read);
 712         if (!be || !is_writable(be, isect)) {
 713                 dprintk("%s no matching extents!\n", __func__);
 714                 goto out_mds;
 715         }
 716
 717         /* First page inside INVALID extent */
 718         if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
 719                 if (likely(!bl_push_one_short_extent(be->be_inval)))
 720                         par->bse_count++;
 721                 else
 722                         goto out_mds;
 723                 temp = offset >> PAGE_CACHE_SHIFT;
 724                 npg_zero = do_div(temp, npg_per_block);
 725                 isect = (sector_t) (((offset - npg_zero * PAGE_CACHE_SIZE) &
 726                                      (long)PAGE_CACHE_MASK) >> SECTOR_SHIFT);
 727                 extent_length = be->be_length - (isect - be->be_f_offset);
 728
 729 fill_invalid_ext:
 730                 dprintk("%s need to zero %d pages\n", __func__, npg_zero);
 731                 for (;npg_zero > 0; npg_zero--) {
 732                         if (bl_is_sector_init(be->be_inval, isect)) {
 733                                 dprintk("isect %llu already init\n",
 734                                         (unsigned long long)isect);
 735                                 goto next_page;
 736                         }
 737                         /* page ref released in bl_end_io_write_zero */
 738                         index = isect >> PAGE_CACHE_SECTOR_SHIFT;
 739                         dprintk("%s zero %dth page: index %lu isect %llu\n",
 740                                 __func__, npg_zero, index,
 741                                 (unsigned long long)isect);
 742                         page = bl_find_get_zeroing_page(header->inode, index,
 743                                                         cow_read);
 744                         if (unlikely(IS_ERR(page))) {
 745                                 header->pnfs_error = PTR_ERR(page);
 746                                 goto out;
 747                         } else if (page == NULL)
 748                                 goto next_page;
 749
 750                         ret = bl_mark_sectors_init(be->be_inval, isect,
 751                                                        PAGE_CACHE_SECTORS);
 752                         if (unlikely(ret)) {
 753                                 dprintk("%s bl_mark_sectors_init fail %d\n",
 754                                         __func__, ret);
 755                                 end_page_writeback(page);
 756                                 page_cache_release(page);
 757                                 header->pnfs_error = ret;
 758                                 goto out;
 759                         }
 760                         if (likely(!bl_push_one_short_extent(be->be_inval)))
 761                                 par->bse_count++;
 762                         else {
 763                                 end_page_writeback(page);
 764                                 page_cache_release(page);
 765                                 header->pnfs_error = -ENOMEM;
 766                                 goto out;
 767                         }
 768                         /* FIXME: This should be done in bi_end_io */
 769                         mark_extents_written(BLK_LSEG2EXT(header->lseg),
 770                                              page->index << PAGE_CACHE_SHIFT,
 771                                              PAGE_CACHE_SIZE);
 772
 773                         bio = bl_add_page_to_bio(bio, npg_zero, WRITE,
 774                                                  isect, page, be,
 775                                                  bl_end_io_write_zero, par);
 776                         if (IS_ERR(bio)) {
 777                                 header->pnfs_error = PTR_ERR(bio);
 778                                 bio = NULL;
 779                                 goto out;
 780                         }
 781 next_page:
 782                         isect += PAGE_CACHE_SECTORS;
 783                         extent_length -= PAGE_CACHE_SECTORS;
 784                 }
 785                 if (last)
 786                         goto write_done;
 787         }
 788         bio = bl_submit_bio(WRITE, bio);
 789
 790         /* Middle pages */
 791         pg_index = header->args.pgbase >> PAGE_CACHE_SHIFT;
 792         for (i = pg_index; i < header->page_array.npages; i++) {
 793                 if (!extent_length) {
 794                         /* We've used up the previous extent */
 795                         bl_put_extent(be);
 796                         bl_put_extent(cow_read);
 797                         bio = bl_submit_bio(WRITE, bio);
 798                         /* Get the next one */
 799                         be = bl_find_get_extent(BLK_LSEG2EXT(header->lseg),
 800                                              isect, &cow_read);
 801                         if (!be || !is_writable(be, isect)) {
 802                                 header->pnfs_error = -EINVAL;
 803                                 goto out;
 804                         }
 805                         if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
 806                                 if (likely(!bl_push_one_short_extent(
 807                                                                 be->be_inval)))
 808                                         par->bse_count++;
 809                                 else {
 810                                         header->pnfs_error = -ENOMEM;
 811                                         goto out;
 812                                 }
 813                         }
 814                         extent_length = be->be_length -
 815                             (isect - be->be_f_offset);
 816                 }
 817
 818                 dprintk("%s offset %lld count %Zu\n", __func__, offset, count);
 819                 pg_offset = offset & ~PAGE_CACHE_MASK;
 820                 if (pg_offset + count > PAGE_CACHE_SIZE)
 821                         pg_len = PAGE_CACHE_SIZE - pg_offset;
 822                 else
 823                         pg_len = count;
 824
 825                 saved_len = pg_len;
 826                 if (be->be_state == PNFS_BLOCK_INVALID_DATA &&
 827                     !bl_is_sector_init(be->be_inval, isect)) {
 828                         ret = bl_read_partial_page_sync(pages[i], cow_read,
 829                                                         pg_offset, pg_len, true);
 830                         if (ret) {
 831                                 dprintk("%s bl_read_partial_page_sync fail %d\n",
 832                                         __func__, ret);
 833                                 header->pnfs_error = ret;
 834                                 goto out;
 835                         }
 836
 837                         ret = bl_mark_sectors_init(be->be_inval, isect,
 838                                                        PAGE_CACHE_SECTORS);
 839                         if (unlikely(ret)) {
 840                                 dprintk("%s bl_mark_sectors_init fail %d\n",
 841                                         __func__, ret);
 842                                 header->pnfs_error = ret;
 843                                 goto out;
 844                         }
 845
 846                         /* Expand to full page write */
 847                         pg_offset = 0;
 848                         pg_len = PAGE_CACHE_SIZE;
 849                 } else if  ((pg_offset & (SECTOR_SIZE - 1)) ||
 850                             (pg_len & (SECTOR_SIZE - 1))){
 851                         /* ahh, nasty case. We have to do sync full sector
 852                          * read-modify-write cycles.
 853                          */
 854                         unsigned int saved_offset = pg_offset;
 855                         ret = bl_read_partial_page_sync(pages[i], be, pg_offset,
 856                                                         pg_len, false);
 857                         pg_offset = round_down(pg_offset, SECTOR_SIZE);
 858                         pg_len = round_up(saved_offset + pg_len, SECTOR_SIZE)
 859                                  - pg_offset;
 860                 }
 861
 862
 863                 bio = do_add_page_to_bio(bio, header->page_array.npages - i,
 864                                          WRITE,
 865                                          isect, pages[i], be,
 866                                          bl_end_io_write, par,
 867                                          pg_offset, pg_len);
 868                 if (IS_ERR(bio)) {
 869                         header->pnfs_error = PTR_ERR(bio);
 870                         bio = NULL;
 871                         goto out;
 872                 }
 873                 offset += saved_len;
 874                 count -= saved_len;
 875                 isect += PAGE_CACHE_SECTORS;
 876                 last_isect = isect;
 877                 extent_length -= PAGE_CACHE_SECTORS;
 878         }
 879
 880         /* Last page inside INVALID extent */
 881         if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
 882                 bio = bl_submit_bio(WRITE, bio);
 883                 temp = last_isect >> PAGE_CACHE_SECTOR_SHIFT;
 884                 npg_zero = npg_per_block - do_div(temp, npg_per_block);
 885                 if (npg_zero < npg_per_block) {
 886                         last = 1;
 887                         goto fill_invalid_ext;
 888                 }
 889         }
 890
 891 write_done:
 892         header->res.count = header->args.count;
 893 out:
 894         bl_put_extent(be);
 895         bl_put_extent(cow_read);
 896         bl_submit_bio(WRITE, bio);
 897         put_parallel(par);
 898         return PNFS_ATTEMPTED;
 899 out_mds:
 900         bl_put_extent(be);
 901         bl_put_extent(cow_read);
 902         kfree(par);
 903         return PNFS_NOT_ATTEMPTED;
 904 }
 905
 906 /* FIXME - range ignored */
 907 static void
 908 release_extents(struct pnfs_block_layout *bl, struct pnfs_layout_range *range)
 909 {
 910         int i;
 911         struct pnfs_block_extent *be;
 912
 913         spin_lock(&bl->bl_ext_lock);
 914         for (i = 0; i < EXTENT_LISTS; i++) {
 915                 while (!list_empty(&bl->bl_extents[i])) {
 916                         be = list_first_entry(&bl->bl_extents[i],
 917                                               struct pnfs_block_extent,
 918                                               be_node);
 919                         list_del(&be->be_node);
 920                         bl_put_extent(be);
 921                 }
 922         }
 923         spin_unlock(&bl->bl_ext_lock);
 924 }
 925
 926 static void
 927 release_inval_marks(struct pnfs_inval_markings *marks)
 928 {
 929         struct pnfs_inval_tracking *pos, *temp;
 930         struct pnfs_block_short_extent *se, *stemp;
 931
 932         list_for_each_entry_safe(pos, temp, &marks->im_tree.mtt_stub, it_link) {
 933                 list_del(&pos->it_link);
 934                 kfree(pos);
 935         }
 936
 937         list_for_each_entry_safe(se, stemp, &marks->im_extents, bse_node) {
 938                 list_del(&se->bse_node);
 939                 kfree(se);
 940         }
 941         return;
 942 }
 943
 944 static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo)
 945 {
 946         struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
 947
 948         dprintk("%s enter\n", __func__);
 949         release_extents(bl, NULL);
 950         release_inval_marks(&bl->bl_inval);
 951         kfree(bl);
 952 }
 953
 954 static struct pnfs_layout_hdr *bl_alloc_layout_hdr(struct inode *inode,
 955                                                    gfp_t gfp_flags)
 956 {
 957         struct pnfs_block_layout *bl;
 958
 959         dprintk("%s enter\n", __func__);
 960         bl = kzalloc(sizeof(*bl), gfp_flags);
 961         if (!bl)
 962                 return NULL;
 963         spin_lock_init(&bl->bl_ext_lock);
 964         INIT_LIST_HEAD(&bl->bl_extents[0]);
 965         INIT_LIST_HEAD(&bl->bl_extents[1]);
 966         INIT_LIST_HEAD(&bl->bl_commit);
 967         INIT_LIST_HEAD(&bl->bl_committing);
 968         bl->bl_count = 0;
 969         bl->bl_blocksize = NFS_SERVER(inode)->pnfs_blksize >> SECTOR_SHIFT;
 970         BL_INIT_INVAL_MARKS(&bl->bl_inval, bl->bl_blocksize);
 971         return &bl->bl_layout;
 972 }
 973
 974 static void bl_free_lseg(struct pnfs_layout_segment *lseg)
 975 {
 976         dprintk("%s enter\n", __func__);
 977         kfree(lseg);
 978 }
 979
 980 /* We pretty much ignore lseg, and store all data layout wide, so we
 981  * can correctly merge.
 982  */
 983 static struct pnfs_layout_segment *bl_alloc_lseg(struct pnfs_layout_hdr *lo,
 984                                                  struct nfs4_layoutget_res *lgr,
 985                                                  gfp_t gfp_flags)
 986 {
 987         struct pnfs_layout_segment *lseg;
 988         int status;
 989
 990         dprintk("%s enter\n", __func__);
 991         lseg = kzalloc(sizeof(*lseg), gfp_flags);
 992         if (!lseg)
 993                 return ERR_PTR(-ENOMEM);
 994         status = nfs4_blk_process_layoutget(lo, lgr, gfp_flags);
 995         if (status) {
 996                 /* We don't want to call the full-blown bl_free_lseg,
 997                  * since on error extents were not touched.
 998                  */
 999                 kfree(lseg);
1000                 return ERR_PTR(status);
1001         }
1002         return lseg;
1003 }
1004
1005 static void
1006 bl_encode_layoutcommit(struct pnfs_layout_hdr *lo, struct xdr_stream *xdr,
1007                        const struct nfs4_layoutcommit_args *arg)
1008 {
1009         dprintk("%s enter\n", __func__);
1010         encode_pnfs_block_layoutupdate(BLK_LO2EXT(lo), xdr, arg);
1011 }
1012
1013 static void
1014 bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata)
1015 {
1016         struct pnfs_layout_hdr *lo = NFS_I(lcdata->args.inode)->layout;
1017
1018         dprintk("%s enter\n", __func__);
1019         clean_pnfs_block_layoutupdate(BLK_LO2EXT(lo), &lcdata->args, lcdata->res.status);
1020 }
1021
1022 static void free_blk_mountid(struct block_mount_id *mid)
1023 {
1024         if (mid) {
1025                 struct pnfs_block_dev *dev, *tmp;
1026
1027                 /* No need to take bm_lock as we are last user freeing bm_devlist */
1028                 list_for_each_entry_safe(dev, tmp, &mid->bm_devlist, bm_node) {
1029                         list_del(&dev->bm_node);
1030                         bl_free_block_dev(dev);
1031                 }
1032                 kfree(mid);
1033         }
1034 }
1035
1036 /* This is mostly copied from the filelayout_get_device_info function.
1037  * It seems much of this should be at the generic pnfs level.
1038  */
1039 static struct pnfs_block_dev *
1040 nfs4_blk_get_deviceinfo(struct nfs_server *server, const struct nfs_fh *fh,
1041                         struct nfs4_deviceid *d_id)
1042 {
1043         struct pnfs_device *dev;
1044         struct pnfs_block_dev *rv;
1045         u32 max_resp_sz;
1046         int max_pages;
1047         struct page **pages = NULL;
1048         int i, rc;
1049
1050         /*
1051          * Use the session max response size as the basis for setting
1052          * GETDEVICEINFO's maxcount
1053          */
1054         max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz;
1055         max_pages = nfs_page_array_len(0, max_resp_sz);
1056         dprintk("%s max_resp_sz %u max_pages %d\n",
1057                 __func__, max_resp_sz, max_pages);
1058
1059         dev = kmalloc(sizeof(*dev), GFP_NOFS);
1060         if (!dev) {
1061                 dprintk("%s kmalloc failed\n", __func__);
1062                 return ERR_PTR(-ENOMEM);
1063         }
1064
1065         pages = kcalloc(max_pages, sizeof(struct page *), GFP_NOFS);
1066         if (pages == NULL) {
1067                 kfree(dev);
1068                 return ERR_PTR(-ENOMEM);
1069         }
1070         for (i = 0; i < max_pages; i++) {
1071                 pages[i] = alloc_page(GFP_NOFS);
1072                 if (!pages[i]) {
1073                         rv = ERR_PTR(-ENOMEM);
1074                         goto out_free;
1075                 }
1076         }
1077
1078         memcpy(&dev->dev_id, d_id, sizeof(*d_id));
1079         dev->layout_type = LAYOUT_BLOCK_VOLUME;
1080         dev->pages = pages;
1081         dev->pgbase = 0;
1082         dev->pglen = PAGE_SIZE * max_pages;
1083         dev->mincount = 0;
1084         dev->maxcount = max_resp_sz - nfs41_maxgetdevinfo_overhead;
1085
1086         dprintk("%s: dev_id: %s\n", __func__, dev->dev_id.data);
1087         rc = nfs4_proc_getdeviceinfo(server, dev, NULL);
1088         dprintk("%s getdevice info returns %d\n", __func__, rc);
1089         if (rc) {
1090                 rv = ERR_PTR(rc);
1091                 goto out_free;
1092         }
1093
1094         rv = nfs4_blk_decode_device(server, dev);
1095  out_free:
1096         for (i = 0; i < max_pages; i++)
1097                 __free_page(pages[i]);
1098         kfree(pages);
1099         kfree(dev);
1100         return rv;
1101 }
1102
1103 static int
1104 bl_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *fh)
1105 {
1106         struct block_mount_id *b_mt_id = NULL;
1107         struct pnfs_devicelist *dlist = NULL;
1108         struct pnfs_block_dev *bdev;
1109         LIST_HEAD(block_disklist);
1110         int status, i;
1111
1112         dprintk("%s enter\n", __func__);
1113
1114         if (server->pnfs_blksize == 0) {
1115                 dprintk("%s Server did not return blksize\n", __func__);
1116                 return -EINVAL;
1117         }
1118         if (server->pnfs_blksize > PAGE_SIZE) {
1119                 printk(KERN_ERR "%s: pNFS blksize %d not supported.\n",
1120                         __func__, server->pnfs_blksize);
1121                 return -EINVAL;
1122         }
1123
1124         b_mt_id = kzalloc(sizeof(struct block_mount_id), GFP_NOFS);
1125         if (!b_mt_id) {
1126                 status = -ENOMEM;
1127                 goto out_error;
1128         }
1129         /* Initialize nfs4 block layout mount id */
1130         spin_lock_init(&b_mt_id->bm_lock);
1131         INIT_LIST_HEAD(&b_mt_id->bm_devlist);
1132
1133         dlist = kmalloc(sizeof(struct pnfs_devicelist), GFP_NOFS);
1134         if (!dlist) {
1135                 status = -ENOMEM;
1136                 goto out_error;
1137         }
1138         dlist->eof = 0;
1139         while (!dlist->eof) {
1140                 status = nfs4_proc_getdevicelist(server, fh, dlist);
1141                 if (status)
1142                         goto out_error;
1143                 dprintk("%s GETDEVICELIST numdevs=%i, eof=%i\n",
1144                         __func__, dlist->num_devs, dlist->eof);
1145                 for (i = 0; i < dlist->num_devs; i++) {
1146                         bdev = nfs4_blk_get_deviceinfo(server, fh,
1147                                                        &dlist->dev_id[i]);
1148                         if (IS_ERR(bdev)) {
1149                                 status = PTR_ERR(bdev);
1150                                 goto out_error;
1151                         }
1152                         spin_lock(&b_mt_id->bm_lock);
1153                         list_add(&bdev->bm_node, &b_mt_id->bm_devlist);
1154                         spin_unlock(&b_mt_id->bm_lock);
1155                 }
1156         }
1157         dprintk("%s SUCCESS\n", __func__);
1158         server->pnfs_ld_data = b_mt_id;
1159
1160  out_return:
1161         kfree(dlist);
1162         return status;
1163
1164  out_error:
1165         free_blk_mountid(b_mt_id);
1166         goto out_return;
1167 }
1168
1169 static int
1170 bl_clear_layoutdriver(struct nfs_server *server)
1171 {
1172         struct block_mount_id *b_mt_id = server->pnfs_ld_data;
1173
1174         dprintk("%s enter\n", __func__);
1175         free_blk_mountid(b_mt_id);
1176         dprintk("%s RETURNS\n", __func__);
1177         return 0;
1178 }
1179
1180 static bool
1181 is_aligned_req(struct nfs_page *req, unsigned int alignment)
1182 {
1183         return IS_ALIGNED(req->wb_offset, alignment) &&
1184                IS_ALIGNED(req->wb_bytes, alignment);
1185 }
1186
1187 static void
1188 bl_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1189 {
1190         if (pgio->pg_dreq != NULL &&
1191             !is_aligned_req(req, SECTOR_SIZE))
1192                 nfs_pageio_reset_read_mds(pgio);
1193         else
1194                 pnfs_generic_pg_init_read(pgio, req);
1195 }
1196
1197 /*
1198  * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
1199  * of bytes (maximum @req->wb_bytes) that can be coalesced.
1200  */
1201 static size_t
1202 bl_pg_test_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
1203                 struct nfs_page *req)
1204 {
1205         if (pgio->pg_dreq != NULL &&
1206             !is_aligned_req(req, SECTOR_SIZE))
1207                 return 0;
1208
1209         return pnfs_generic_pg_test(pgio, prev, req);
1210 }
1211
1212 /*
1213  * Return the number of contiguous bytes for a given inode
1214  * starting at page frame idx.
1215  */
1216 static u64 pnfs_num_cont_bytes(struct inode *inode, pgoff_t idx)
1217 {
1218         struct address_space *mapping = inode->i_mapping;
1219         pgoff_t end;
1220
1221         /* Optimize common case that writes from 0 to end of file */
1222         end = DIV_ROUND_UP(i_size_read(inode), PAGE_CACHE_SIZE);
1223         if (end != NFS_I(inode)->npages) {
1224                 rcu_read_lock();
1225                 end = page_cache_next_hole(mapping, idx + 1, ULONG_MAX);
1226                 rcu_read_unlock();
1227         }
1228
1229         if (!end)
1230                 return i_size_read(inode) - (idx << PAGE_CACHE_SHIFT);
1231         else
1232                 return (end - idx) << PAGE_CACHE_SHIFT;
1233 }
1234
1235 static void
1236 bl_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1237 {
1238         if (pgio->pg_dreq != NULL &&
1239             !is_aligned_req(req, PAGE_CACHE_SIZE)) {
1240                 nfs_pageio_reset_write_mds(pgio);
1241         } else {
1242                 u64 wb_size;
1243                 if (pgio->pg_dreq == NULL)
1244                         wb_size = pnfs_num_cont_bytes(pgio->pg_inode,
1245                                                       req->wb_index);
1246                 else
1247                         wb_size = nfs_dreq_bytes_left(pgio->pg_dreq);
1248
1249                 pnfs_generic_pg_init_write(pgio, req, wb_size);
1250         }
1251 }
1252
1253 /*
1254  * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
1255  * of bytes (maximum @req->wb_bytes) that can be coalesced.
1256  */
1257 static size_t
1258 bl_pg_test_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
1259                  struct nfs_page *req)
1260 {
1261         if (pgio->pg_dreq != NULL &&
1262             !is_aligned_req(req, PAGE_CACHE_SIZE))
1263                 return 0;
1264
1265         return pnfs_generic_pg_test(pgio, prev, req);
1266 }
1267
1268 static const struct nfs_pageio_ops bl_pg_read_ops = {
1269         .pg_init = bl_pg_init_read,
1270         .pg_test = bl_pg_test_read,
1271         .pg_doio = pnfs_generic_pg_readpages,
1272 };
1273
1274 static const struct nfs_pageio_ops bl_pg_write_ops = {
1275         .pg_init = bl_pg_init_write,
1276         .pg_test = bl_pg_test_write,
1277         .pg_doio = pnfs_generic_pg_writepages,
1278 };
1279
1280 static struct pnfs_layoutdriver_type blocklayout_type = {
1281         .id                             = LAYOUT_BLOCK_VOLUME,
1282         .name                           = "LAYOUT_BLOCK_VOLUME",
1283         .owner                          = THIS_MODULE,
1284         .read_pagelist                  = bl_read_pagelist,
1285         .write_pagelist                 = bl_write_pagelist,
1286         .alloc_layout_hdr               = bl_alloc_layout_hdr,
1287         .free_layout_hdr                = bl_free_layout_hdr,
1288         .alloc_lseg                     = bl_alloc_lseg,
1289         .free_lseg                      = bl_free_lseg,
1290         .encode_layoutcommit            = bl_encode_layoutcommit,
1291         .cleanup_layoutcommit           = bl_cleanup_layoutcommit,
1292         .set_layoutdriver               = bl_set_layoutdriver,
1293         .clear_layoutdriver             = bl_clear_layoutdriver,
1294         .pg_read_ops                    = &bl_pg_read_ops,
1295         .pg_write_ops                   = &bl_pg_write_ops,
1296 };
1297
1298 static const struct rpc_pipe_ops bl_upcall_ops = {
1299         .upcall         = rpc_pipe_generic_upcall,
1300         .downcall       = bl_pipe_downcall,
1301         .destroy_msg    = bl_pipe_destroy_msg,
1302 };
1303
1304 static struct dentry *nfs4blocklayout_register_sb(struct super_block *sb,
1305                                             struct rpc_pipe *pipe)
1306 {
1307         struct dentry *dir, *dentry;
1308
1309         dir = rpc_d_lookup_sb(sb, NFS_PIPE_DIRNAME);
1310         if (dir == NULL)
1311                 return ERR_PTR(-ENOENT);
1312         dentry = rpc_mkpipe_dentry(dir, "blocklayout", NULL, pipe);
1313         dput(dir);
1314         return dentry;
1315 }
1316
1317 static void nfs4blocklayout_unregister_sb(struct super_block *sb,
1318                                           struct rpc_pipe *pipe)
1319 {
1320         if (pipe->dentry)
1321                 rpc_unlink(pipe->dentry);
1322 }
1323
1324 static int rpc_pipefs_event(struct notifier_block *nb, unsigned long event,
1325                            void *ptr)
1326 {
1327         struct super_block *sb = ptr;
1328         struct net *net = sb->s_fs_info;
1329         struct nfs_net *nn = net_generic(net, nfs_net_id);
1330         struct dentry *dentry;
1331         int ret = 0;
1332
1333         if (!try_module_get(THIS_MODULE))
1334                 return 0;
1335
1336         if (nn->bl_device_pipe == NULL) {
1337                 module_put(THIS_MODULE);
1338                 return 0;
1339         }
1340
1341         switch (event) {
1342         case RPC_PIPEFS_MOUNT:
1343                 dentry = nfs4blocklayout_register_sb(sb, nn->bl_device_pipe);
1344                 if (IS_ERR(dentry)) {
1345                         ret = PTR_ERR(dentry);
1346                         break;
1347                 }
1348                 nn->bl_device_pipe->dentry = dentry;
1349                 break;
1350         case RPC_PIPEFS_UMOUNT:
1351                 if (nn->bl_device_pipe->dentry)
1352                         nfs4blocklayout_unregister_sb(sb, nn->bl_device_pipe);
1353                 break;
1354         default:
1355                 ret = -ENOTSUPP;
1356                 break;
1357         }
1358         module_put(THIS_MODULE);
1359         return ret;
1360 }
1361
1362 static struct notifier_block nfs4blocklayout_block = {
1363         .notifier_call = rpc_pipefs_event,
1364 };
1365
1366 static struct dentry *nfs4blocklayout_register_net(struct net *net,
1367                                                    struct rpc_pipe *pipe)
1368 {
1369         struct super_block *pipefs_sb;
1370         struct dentry *dentry;
1371
1372         pipefs_sb = rpc_get_sb_net(net);
1373         if (!pipefs_sb)
1374                 return NULL;
1375         dentry = nfs4blocklayout_register_sb(pipefs_sb, pipe);
1376         rpc_put_sb_net(net);
1377         return dentry;
1378 }
1379
1380 static void nfs4blocklayout_unregister_net(struct net *net,
1381                                            struct rpc_pipe *pipe)
1382 {
1383         struct super_block *pipefs_sb;
1384
1385         pipefs_sb = rpc_get_sb_net(net);
1386         if (pipefs_sb) {
1387                 nfs4blocklayout_unregister_sb(pipefs_sb, pipe);
1388                 rpc_put_sb_net(net);
1389         }
1390 }
1391
1392 static int nfs4blocklayout_net_init(struct net *net)
1393 {
1394         struct nfs_net *nn = net_generic(net, nfs_net_id);
1395         struct dentry *dentry;
1396
1397         init_waitqueue_head(&nn->bl_wq);
1398         nn->bl_device_pipe = rpc_mkpipe_data(&bl_upcall_ops, 0);
1399         if (IS_ERR(nn->bl_device_pipe))
1400                 return PTR_ERR(nn->bl_device_pipe);
1401         dentry = nfs4blocklayout_register_net(net, nn->bl_device_pipe);
1402         if (IS_ERR(dentry)) {
1403                 rpc_destroy_pipe_data(nn->bl_device_pipe);
1404                 return PTR_ERR(dentry);
1405         }
1406         nn->bl_device_pipe->dentry = dentry;
1407         return 0;
1408 }
1409
1410 static void nfs4blocklayout_net_exit(struct net *net)
1411 {
1412         struct nfs_net *nn = net_generic(net, nfs_net_id);
1413
1414         nfs4blocklayout_unregister_net(net, nn->bl_device_pipe);
1415         rpc_destroy_pipe_data(nn->bl_device_pipe);
1416         nn->bl_device_pipe = NULL;
1417 }
1418
1419 static struct pernet_operations nfs4blocklayout_net_ops = {
1420         .init = nfs4blocklayout_net_init,
1421         .exit = nfs4blocklayout_net_exit,
1422 };
1423
1424 static int __init nfs4blocklayout_init(void)
1425 {
1426         int ret;
1427
1428         dprintk("%s: NFSv4 Block Layout Driver Registering...\n", __func__);
1429
1430         ret = pnfs_register_layoutdriver(&blocklayout_type);
1431         if (ret)
1432                 goto out;
1433
1434         ret = rpc_pipefs_notifier_register(&nfs4blocklayout_block);
1435         if (ret)
1436                 goto out_remove;
1437         ret = register_pernet_subsys(&nfs4blocklayout_net_ops);
1438         if (ret)
1439                 goto out_notifier;
1440 out:
1441         return ret;
1442
1443 out_notifier:
1444         rpc_pipefs_notifier_unregister(&nfs4blocklayout_block);
1445 out_remove:
1446         pnfs_unregister_layoutdriver(&blocklayout_type);
1447         return ret;
1448 }
1449
1450 static void __exit nfs4blocklayout_exit(void)
1451 {
1452         dprintk("%s: NFSv4 Block Layout Driver Unregistering...\n",
1453                __func__);
1454
1455         rpc_pipefs_notifier_unregister(&nfs4blocklayout_block);
1456         unregister_pernet_subsys(&nfs4blocklayout_net_ops);
1457         pnfs_unregister_layoutdriver(&blocklayout_type);
1458 }
1459
1460 MODULE_ALIAS("nfs-layouttype4-3");
1461
1462 module_init(nfs4blocklayout_init);
1463 module_exit(nfs4blocklayout_exit);