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