fs/nfs/blocklayout/blocklayout.c

   1 /*
   2  *  linux/fs/nfs/blocklayout/blocklayout.c
   3  *
   4  *  Module for the NFSv4.1 pNFS block layout driver.
   5  *
   6  *  Copyright (c) 2006 The Regents of the University of Michigan.
   7  *  All rights reserved.
   8  *
   9  *  Andy Adamson <andros@citi.umich.edu>
  10  *  Fred Isaman <iisaman@umich.edu>
  11  *
  12  * permission is granted to use, copy, create derivative works and
  13  * redistribute this software and such derivative works for any purpose,
  14  * so long as the name of the university of michigan is not used in
  15  * any advertising or publicity pertaining to the use or distribution
  16  * of this software without specific, written prior authorization.  if
  17  * the above copyright notice or any other identification of the
  18  * university of michigan is included in any copy of any portion of
  19  * this software, then the disclaimer below must also be included.
  20  *
  21  * this software is provided as is, without representation from the
  22  * university of michigan as to its fitness for any purpose, and without
  23  * warranty by the university of michigan of any kind, either express
  24  * or implied, including without limitation the implied warranties of
  25  * merchantability and fitness for a particular purpose.  the regents
  26  * of the university of michigan shall not be liable for any damages,
  27  * including special, indirect, incidental, or consequential damages,
  28  * with respect to any claim arising out or in connection with the use
  29  * of the software, even if it has been or is hereafter advised of the
  30  * possibility of such damages.
  31  */
  32
  33 #include <linux/module.h>
  34 #include <linux/init.h>
  35 #include <linux/mount.h>
  36 #include <linux/namei.h>
  37 #include <linux/bio.h>          /* struct bio */
  38 #include <linux/buffer_head.h>  /* various write calls */
  39 #include <linux/prefetch.h>
  40
  41 #include "../pnfs.h"
  42 #include "../internal.h"
  43 #include "blocklayout.h"
  44
  45 #define NFSDBG_FACILITY NFSDBG_PNFS_LD
  46
  47 MODULE_LICENSE("GPL");
  48 MODULE_AUTHOR("Andy Adamson <andros@citi.umich.edu>");
  49 MODULE_DESCRIPTION("The NFSv4.1 pNFS Block layout driver");
  50
  51 static void print_page(struct page *page)
  52 {
  53         dprintk("PRINTPAGE page %p\n", page);
  54         dprintk("       PagePrivate %d\n", PagePrivate(page));
  55         dprintk("       PageUptodate %d\n", PageUptodate(page));
  56         dprintk("       PageError %d\n", PageError(page));
  57         dprintk("       PageDirty %d\n", PageDirty(page));
  58         dprintk("       PageReferenced %d\n", PageReferenced(page));
  59         dprintk("       PageLocked %d\n", PageLocked(page));
  60         dprintk("       PageWriteback %d\n", PageWriteback(page));
  61         dprintk("       PageMappedToDisk %d\n", PageMappedToDisk(page));
  62         dprintk("\n");
  63 }
  64
  65 /* Given the be associated with isect, determine if page data needs to be
  66  * initialized.
  67  */
  68 static int is_hole(struct pnfs_block_extent *be, sector_t isect)
  69 {
  70         if (be->be_state == PNFS_BLOCK_NONE_DATA)
  71                 return 1;
  72         else if (be->be_state != PNFS_BLOCK_INVALID_DATA)
  73                 return 0;
  74         else
  75                 return !bl_is_sector_init(be->be_inval, isect);
  76 }
  77
  78 /* Given the be associated with isect, determine if page data can be
  79  * written to disk.
  80  */
  81 static int is_writable(struct pnfs_block_extent *be, sector_t isect)
  82 {
  83         return (be->be_state == PNFS_BLOCK_READWRITE_DATA ||
  84                 be->be_state == PNFS_BLOCK_INVALID_DATA);
  85 }
  86
  87 /* The data we are handed might be spread across several bios.  We need
  88  * to track when the last one is finished.
  89  */
  90 struct parallel_io {
  91         struct kref refcnt;
  92         void (*pnfs_callback) (void *data, int num_se);
  93         void *data;
  94         int bse_count;
  95 };
  96
  97 static inline struct parallel_io *alloc_parallel(void *data)
  98 {
  99         struct parallel_io *rv;
 100
 101         rv  = kmalloc(sizeof(*rv), GFP_NOFS);
 102         if (rv) {
 103                 rv->data = data;
 104                 kref_init(&rv->refcnt);
 105                 rv->bse_count = 0;
 106         }
 107         return rv;
 108 }
 109
 110 static inline void get_parallel(struct parallel_io *p)
 111 {
 112         kref_get(&p->refcnt);
 113 }
 114
 115 static void destroy_parallel(struct kref *kref)
 116 {
 117         struct parallel_io *p = container_of(kref, struct parallel_io, refcnt);
 118
 119         dprintk("%s enter\n", __func__);
 120         p->pnfs_callback(p->data, p->bse_count);
 121         kfree(p);
 122 }
 123
 124 static inline void put_parallel(struct parallel_io *p)
 125 {
 126         kref_put(&p->refcnt, destroy_parallel);
 127 }
 128
 129 static struct bio *
 130 bl_submit_bio(int rw, struct bio *bio)
 131 {
 132         if (bio) {
 133                 get_parallel(bio->bi_private);
 134                 dprintk("%s submitting %s bio %u@%llu\n", __func__,
 135                         rw == READ ? "read" : "write",
 136                         bio->bi_size, (unsigned long long)bio->bi_sector);
 137                 submit_bio(rw, bio);
 138         }
 139         return NULL;
 140 }
 141
 142 static struct bio *bl_alloc_init_bio(int npg, sector_t isect,
 143                                      struct pnfs_block_extent *be,
 144                                      void (*end_io)(struct bio *, int err),
 145                                      struct parallel_io *par)
 146 {
 147         struct bio *bio;
 148
 149         npg = min(npg, BIO_MAX_PAGES);
 150         bio = bio_alloc(GFP_NOIO, npg);
 151         if (!bio && (current->flags & PF_MEMALLOC)) {
 152                 while (!bio && (npg /= 2))
 153                         bio = bio_alloc(GFP_NOIO, npg);
 154         }
 155
 156         if (bio) {
 157                 bio->bi_sector = isect - be->be_f_offset + be->be_v_offset;
 158                 bio->bi_bdev = be->be_mdev;
 159                 bio->bi_end_io = end_io;
 160                 bio->bi_private = par;
 161         }
 162         return bio;
 163 }
 164
 165 static struct bio *do_add_page_to_bio(struct bio *bio, int npg, int rw,
 166                                       sector_t isect, struct page *page,
 167                                       struct pnfs_block_extent *be,
 168                                       void (*end_io)(struct bio *, int err),
 169                                       struct parallel_io *par,
 170                                       unsigned int offset, int len)
 171 {
 172         isect = isect + (offset >> SECTOR_SHIFT);
 173         dprintk("%s: npg %d rw %d isect %llu offset %u len %d\n", __func__,
 174                 npg, rw, (unsigned long long)isect, offset, len);
 175 retry:
 176         if (!bio) {
 177                 bio = bl_alloc_init_bio(npg, isect, be, end_io, par);
 178                 if (!bio)
 179                         return ERR_PTR(-ENOMEM);
 180         }
 181         if (bio_add_page(bio, page, len, offset) < len) {
 182                 bio = bl_submit_bio(rw, bio);
 183                 goto retry;
 184         }
 185         return bio;
 186 }
 187
 188 static struct bio *bl_add_page_to_bio(struct bio *bio, int npg, int rw,
 189                                       sector_t isect, struct page *page,
 190                                       struct pnfs_block_extent *be,
 191                                       void (*end_io)(struct bio *, int err),
 192                                       struct parallel_io *par)
 193 {
 194         return do_add_page_to_bio(bio, npg, rw, isect, page, be,
 195                                   end_io, par, 0, PAGE_CACHE_SIZE);
 196 }
 197
 198 /* This is basically copied from mpage_end_io_read */
 199 static void bl_end_io_read(struct bio *bio, int err)
 200 {
 201         struct parallel_io *par = bio->bi_private;
 202         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 203         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
 204
 205         do {
 206                 struct page *page = bvec->bv_page;
 207
 208                 if (--bvec >= bio->bi_io_vec)
 209                         prefetchw(&bvec->bv_page->flags);
 210                 if (uptodate)
 211                         SetPageUptodate(page);
 212         } while (bvec >= bio->bi_io_vec);
 213         if (!uptodate) {
 214                 struct nfs_read_data *rdata = par->data;
 215                 struct nfs_pgio_header *header = rdata->header;
 216
 217                 if (!header->pnfs_error)
 218                         header->pnfs_error = -EIO;
 219                 pnfs_set_lo_fail(header->lseg);
 220         }
 221         bio_put(bio);
 222         put_parallel(par);
 223 }
 224
 225 static void bl_read_cleanup(struct work_struct *work)
 226 {
 227         struct rpc_task *task;
 228         struct nfs_read_data *rdata;
 229         dprintk("%s enter\n", __func__);
 230         task = container_of(work, struct rpc_task, u.tk_work);
 231         rdata = container_of(task, struct nfs_read_data, task);
 232         pnfs_ld_read_done(rdata);
 233 }
 234
 235 static void
 236 bl_end_par_io_read(void *data, int unused)
 237 {
 238         struct nfs_read_data *rdata = data;
 239
 240         rdata->task.tk_status = rdata->header->pnfs_error;
 241         INIT_WORK(&rdata->task.u.tk_work, bl_read_cleanup);
 242         schedule_work(&rdata->task.u.tk_work);
 243 }
 244
 245 static enum pnfs_try_status
 246 bl_read_pagelist(struct nfs_read_data *rdata)
 247 {
 248         struct nfs_pgio_header *header = rdata->header;
 249         int i, hole;
 250         struct bio *bio = NULL;
 251         struct pnfs_block_extent *be = NULL, *cow_read = NULL;
 252         sector_t isect, extent_length = 0;
 253         struct parallel_io *par;
 254         loff_t f_offset = rdata->args.offset;
 255         size_t bytes_left = rdata->args.count;
 256         unsigned int pg_offset, pg_len;
 257         struct page **pages = rdata->args.pages;
 258         int pg_index = rdata->args.pgbase >> PAGE_CACHE_SHIFT;
 259         const bool is_dio = (header->dreq != NULL);
 260
 261         dprintk("%s enter nr_pages %u offset %lld count %u\n", __func__,
 262                rdata->pages.npages, f_offset, (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->pages.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, rdata->pages.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                 rdata->res.eof = 1;
 337                 rdata->res.count = header->inode->i_size - rdata->args.offset;
 338         } else {
 339                 rdata->res.count = (isect << SECTOR_SHIFT) - rdata->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         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 385         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
 386
 387         do {
 388                 struct page *page = bvec->bv_page;
 389
 390                 if (--bvec >= bio->bi_io_vec)
 391                         prefetchw(&bvec->bv_page->flags);
 392                 /* This is the zeroing page we added */
 393                 end_page_writeback(page);
 394                 page_cache_release(page);
 395         } while (bvec >= bio->bi_io_vec);
 396
 397         if (unlikely(!uptodate)) {
 398                 struct nfs_write_data *data = par->data;
 399                 struct nfs_pgio_header *header = data->header;
 400
 401                 if (!header->pnfs_error)
 402                         header->pnfs_error = -EIO;
 403                 pnfs_set_lo_fail(header->lseg);
 404         }
 405         bio_put(bio);
 406         put_parallel(par);
 407 }
 408
 409 static void bl_end_io_write(struct bio *bio, int err)
 410 {
 411         struct parallel_io *par = bio->bi_private;
 412         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 413         struct nfs_write_data *data = par->data;
 414         struct nfs_pgio_header *header = data->header;
 415
 416         if (!uptodate) {
 417                 if (!header->pnfs_error)
 418                         header->pnfs_error = -EIO;
 419                 pnfs_set_lo_fail(header->lseg);
 420         }
 421         bio_put(bio);
 422         put_parallel(par);
 423 }
 424
 425 /* Function scheduled for call during bl_end_par_io_write,
 426  * it marks sectors as written and extends the commitlist.
 427  */
 428 static void bl_write_cleanup(struct work_struct *work)
 429 {
 430         struct rpc_task *task;
 431         struct nfs_write_data *wdata;
 432         dprintk("%s enter\n", __func__);
 433         task = container_of(work, struct rpc_task, u.tk_work);
 434         wdata = container_of(task, struct nfs_write_data, task);
 435         if (likely(!wdata->header->pnfs_error)) {
 436                 /* Marks for LAYOUTCOMMIT */
 437                 mark_extents_written(BLK_LSEG2EXT(wdata->header->lseg),
 438                                      wdata->args.offset, wdata->args.count);
 439         }
 440         pnfs_ld_write_done(wdata);
 441 }
 442
 443 /* Called when last of bios associated with a bl_write_pagelist call finishes */
 444 static void bl_end_par_io_write(void *data, int num_se)
 445 {
 446         struct nfs_write_data *wdata = data;
 447
 448         if (unlikely(wdata->header->pnfs_error)) {
 449                 bl_free_short_extents(&BLK_LSEG2EXT(wdata->header->lseg)->bl_inval,
 450                                         num_se);
 451         }
 452
 453         wdata->task.tk_status = wdata->header->pnfs_error;
 454         wdata->verf.committed = NFS_FILE_SYNC;
 455         INIT_WORK(&wdata->task.u.tk_work, bl_write_cleanup);
 456         schedule_work(&wdata->task.u.tk_work);
 457 }
 458
 459 /* FIXME STUB - mark intersection of layout and page as bad, so is not
 460  * used again.
 461  */
 462 static void mark_bad_read(void)
 463 {
 464         return;
 465 }
 466
 467 /*
 468  * map_block:  map a requested I/0 block (isect) into an offset in the LVM
 469  * block_device
 470  */
 471 static void
 472 map_block(struct buffer_head *bh, sector_t isect, struct pnfs_block_extent *be)
 473 {
 474         dprintk("%s enter be=%p\n", __func__, be);
 475
 476         set_buffer_mapped(bh);
 477         bh->b_bdev = be->be_mdev;
 478         bh->b_blocknr = (isect - be->be_f_offset + be->be_v_offset) >>
 479             (be->be_mdev->bd_inode->i_blkbits - SECTOR_SHIFT);
 480
 481         dprintk("%s isect %llu, bh->b_blocknr %ld, using bsize %Zd\n",
 482                 __func__, (unsigned long long)isect, (long)bh->b_blocknr,
 483                 bh->b_size);
 484         return;
 485 }
 486
 487 static void
 488 bl_read_single_end_io(struct bio *bio, int error)
 489 {
 490         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
 491         struct page *page = bvec->bv_page;
 492
 493         /* Only one page in bvec */
 494         unlock_page(page);
 495 }
 496
 497 static int
 498 bl_do_readpage_sync(struct page *page, struct pnfs_block_extent *be,
 499                     unsigned int offset, unsigned int len)
 500 {
 501         struct bio *bio;
 502         struct page *shadow_page;
 503         sector_t isect;
 504         char *kaddr, *kshadow_addr;
 505         int ret = 0;
 506
 507         dprintk("%s: offset %u len %u\n", __func__, offset, len);
 508
 509         shadow_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
 510         if (shadow_page == NULL)
 511                 return -ENOMEM;
 512
 513         bio = bio_alloc(GFP_NOIO, 1);
 514         if (bio == NULL)
 515                 return -ENOMEM;
 516
 517         isect = (page->index << PAGE_CACHE_SECTOR_SHIFT) +
 518                 (offset / SECTOR_SIZE);
 519
 520         bio->bi_sector = isect - be->be_f_offset + be->be_v_offset;
 521         bio->bi_bdev = be->be_mdev;
 522         bio->bi_end_io = bl_read_single_end_io;
 523
 524         lock_page(shadow_page);
 525         if (bio_add_page(bio, shadow_page,
 526                          SECTOR_SIZE, round_down(offset, SECTOR_SIZE)) == 0) {
 527                 unlock_page(shadow_page);
 528                 bio_put(bio);
 529                 return -EIO;
 530         }
 531
 532         submit_bio(READ, bio);
 533         wait_on_page_locked(shadow_page);
 534         if (unlikely(!test_bit(BIO_UPTODATE, &bio->bi_flags))) {
 535                 ret = -EIO;
 536         } else {
 537                 kaddr = kmap_atomic(page);
 538                 kshadow_addr = kmap_atomic(shadow_page);
 539                 memcpy(kaddr + offset, kshadow_addr + offset, len);
 540                 kunmap_atomic(kshadow_addr);
 541                 kunmap_atomic(kaddr);
 542         }
 543         __free_page(shadow_page);
 544         bio_put(bio);
 545
 546         return ret;
 547 }
 548
 549 static int
 550 bl_read_partial_page_sync(struct page *page, struct pnfs_block_extent *be,
 551                           unsigned int dirty_offset, unsigned int dirty_len,
 552                           bool full_page)
 553 {
 554         int ret = 0;
 555         unsigned int start, end;
 556
 557         if (full_page) {
 558                 start = 0;
 559                 end = PAGE_CACHE_SIZE;
 560         } else {
 561                 start = round_down(dirty_offset, SECTOR_SIZE);
 562                 end = round_up(dirty_offset + dirty_len, SECTOR_SIZE);
 563         }
 564
 565         dprintk("%s: offset %u len %d\n", __func__, dirty_offset, dirty_len);
 566         if (!be) {
 567                 zero_user_segments(page, start, dirty_offset,
 568                                    dirty_offset + dirty_len, end);
 569                 if (start == 0 && end == PAGE_CACHE_SIZE &&
 570                     trylock_page(page)) {
 571                         SetPageUptodate(page);
 572                         unlock_page(page);
 573                 }
 574                 return ret;
 575         }
 576
 577         if (start != dirty_offset)
 578                 ret = bl_do_readpage_sync(page, be, start, dirty_offset - start);
 579
 580         if (!ret && (dirty_offset + dirty_len < end))
 581                 ret = bl_do_readpage_sync(page, be, dirty_offset + dirty_len,
 582                                           end - dirty_offset - dirty_len);
 583
 584         return ret;
 585 }
 586
 587 /* Given an unmapped page, zero it or read in page for COW, page is locked
 588  * by caller.
 589  */
 590 static int
 591 init_page_for_write(struct page *page, struct pnfs_block_extent *cow_read)
 592 {
 593         struct buffer_head *bh = NULL;
 594         int ret = 0;
 595         sector_t isect;
 596
 597         dprintk("%s enter, %p\n", __func__, page);
 598         BUG_ON(PageUptodate(page));
 599         if (!cow_read) {
 600                 zero_user_segment(page, 0, PAGE_SIZE);
 601                 SetPageUptodate(page);
 602                 goto cleanup;
 603         }
 604
 605         bh = alloc_page_buffers(page, PAGE_CACHE_SIZE, 0);
 606         if (!bh) {
 607                 ret = -ENOMEM;
 608                 goto cleanup;
 609         }
 610
 611         isect = (sector_t) page->index << PAGE_CACHE_SECTOR_SHIFT;
 612         map_block(bh, isect, cow_read);
 613         if (!bh_uptodate_or_lock(bh))
 614                 ret = bh_submit_read(bh);
 615         if (ret)
 616                 goto cleanup;
 617         SetPageUptodate(page);
 618
 619 cleanup:
 620         if (bh)
 621                 free_buffer_head(bh);
 622         if (ret) {
 623                 /* Need to mark layout with bad read...should now
 624                  * just use nfs4 for reads and writes.
 625                  */
 626                 mark_bad_read();
 627         }
 628         return ret;
 629 }
 630
 631 /* Find or create a zeroing page marked being writeback.
 632  * Return ERR_PTR on error, NULL to indicate skip this page and page itself
 633  * to indicate write out.
 634  */
 635 static struct page *
 636 bl_find_get_zeroing_page(struct inode *inode, pgoff_t index,
 637                         struct pnfs_block_extent *cow_read)
 638 {
 639         struct page *page;
 640         int locked = 0;
 641         page = find_get_page(inode->i_mapping, index);
 642         if (page)
 643                 goto check_page;
 644
 645         page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
 646         if (unlikely(!page)) {
 647                 dprintk("%s oom\n", __func__);
 648                 return ERR_PTR(-ENOMEM);
 649         }
 650         locked = 1;
 651
 652 check_page:
 653         /* PageDirty: Other will write this out
 654          * PageWriteback: Other is writing this out
 655          * PageUptodate: It was read before
 656          */
 657         if (PageDirty(page) || PageWriteback(page)) {
 658                 print_page(page);
 659                 if (locked)
 660                         unlock_page(page);
 661                 page_cache_release(page);
 662                 return NULL;
 663         }
 664
 665         if (!locked) {
 666                 lock_page(page);
 667                 locked = 1;
 668                 goto check_page;
 669         }
 670         if (!PageUptodate(page)) {
 671                 /* New page, readin or zero it */
 672                 init_page_for_write(page, cow_read);
 673         }
 674         set_page_writeback(page);
 675         unlock_page(page);
 676
 677         return page;
 678 }
 679
 680 static enum pnfs_try_status
 681 bl_write_pagelist(struct nfs_write_data *wdata, int sync)
 682 {
 683         struct nfs_pgio_header *header = wdata->header;
 684         int i, ret, npg_zero, pg_index, last = 0;
 685         struct bio *bio = NULL;
 686         struct pnfs_block_extent *be = NULL, *cow_read = NULL;
 687         sector_t isect, last_isect = 0, extent_length = 0;
 688         struct parallel_io *par = NULL;
 689         loff_t offset = wdata->args.offset;
 690         size_t count = wdata->args.count;
 691         unsigned int pg_offset, pg_len, saved_len;
 692         struct page **pages = wdata->args.pages;
 693         struct page *page;
 694         pgoff_t index;
 695         u64 temp;
 696         int npg_per_block =
 697             NFS_SERVER(header->inode)->pnfs_blksize >> PAGE_CACHE_SHIFT;
 698
 699         dprintk("%s enter, %Zu@%lld\n", __func__, count, offset);
 700
 701         if (header->dreq != NULL &&
 702             (!IS_ALIGNED(offset, NFS_SERVER(header->inode)->pnfs_blksize) ||
 703              !IS_ALIGNED(count, NFS_SERVER(header->inode)->pnfs_blksize))) {
 704                 dprintk("pnfsblock nonblock aligned DIO writes. Resend MDS\n");
 705                 goto out_mds;
 706         }
 707         /* At this point, wdata->pages is a (sequential) list of nfs_pages.
 708          * We want to write each, and if there is an error set pnfs_error
 709          * to have it redone using nfs.
 710          */
 711         par = alloc_parallel(wdata);
 712         if (!par)
 713                 goto out_mds;
 714         par->pnfs_callback = bl_end_par_io_write;
 715         /* At this point, have to be more careful with error handling */
 716
 717         isect = (sector_t) ((offset & (long)PAGE_CACHE_MASK) >> SECTOR_SHIFT);
 718         be = bl_find_get_extent(BLK_LSEG2EXT(header->lseg), isect, &cow_read);
 719         if (!be || !is_writable(be, isect)) {
 720                 dprintk("%s no matching extents!\n", __func__);
 721                 goto out_mds;
 722         }
 723
 724         /* First page inside INVALID extent */
 725         if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
 726                 if (likely(!bl_push_one_short_extent(be->be_inval)))
 727                         par->bse_count++;
 728                 else
 729                         goto out_mds;
 730                 temp = offset >> PAGE_CACHE_SHIFT;
 731                 npg_zero = do_div(temp, npg_per_block);
 732                 isect = (sector_t) (((offset - npg_zero * PAGE_CACHE_SIZE) &
 733                                      (long)PAGE_CACHE_MASK) >> SECTOR_SHIFT);
 734                 extent_length = be->be_length - (isect - be->be_f_offset);
 735
 736 fill_invalid_ext:
 737                 dprintk("%s need to zero %d pages\n", __func__, npg_zero);
 738                 for (;npg_zero > 0; npg_zero--) {
 739                         if (bl_is_sector_init(be->be_inval, isect)) {
 740                                 dprintk("isect %llu already init\n",
 741                                         (unsigned long long)isect);
 742                                 goto next_page;
 743                         }
 744                         /* page ref released in bl_end_io_write_zero */
 745                         index = isect >> PAGE_CACHE_SECTOR_SHIFT;
 746                         dprintk("%s zero %dth page: index %lu isect %llu\n",
 747                                 __func__, npg_zero, index,
 748                                 (unsigned long long)isect);
 749                         page = bl_find_get_zeroing_page(header->inode, index,
 750                                                         cow_read);
 751                         if (unlikely(IS_ERR(page))) {
 752                                 header->pnfs_error = PTR_ERR(page);
 753                                 goto out;
 754                         } else if (page == NULL)
 755                                 goto next_page;
 756
 757                         ret = bl_mark_sectors_init(be->be_inval, isect,
 758                                                        PAGE_CACHE_SECTORS);
 759                         if (unlikely(ret)) {
 760                                 dprintk("%s bl_mark_sectors_init fail %d\n",
 761                                         __func__, ret);
 762                                 end_page_writeback(page);
 763                                 page_cache_release(page);
 764                                 header->pnfs_error = ret;
 765                                 goto out;
 766                         }
 767                         if (likely(!bl_push_one_short_extent(be->be_inval)))
 768                                 par->bse_count++;
 769                         else {
 770                                 end_page_writeback(page);
 771                                 page_cache_release(page);
 772                                 header->pnfs_error = -ENOMEM;
 773                                 goto out;
 774                         }
 775                         /* FIXME: This should be done in bi_end_io */
 776                         mark_extents_written(BLK_LSEG2EXT(header->lseg),
 777                                              page->index << PAGE_CACHE_SHIFT,
 778                                              PAGE_CACHE_SIZE);
 779
 780                         bio = bl_add_page_to_bio(bio, npg_zero, WRITE,
 781                                                  isect, page, be,
 782                                                  bl_end_io_write_zero, par);
 783                         if (IS_ERR(bio)) {
 784                                 header->pnfs_error = PTR_ERR(bio);
 785                                 bio = NULL;
 786                                 goto out;
 787                         }
 788 next_page:
 789                         isect += PAGE_CACHE_SECTORS;
 790                         extent_length -= PAGE_CACHE_SECTORS;
 791                 }
 792                 if (last)
 793                         goto write_done;
 794         }
 795         bio = bl_submit_bio(WRITE, bio);
 796
 797         /* Middle pages */
 798         pg_index = wdata->args.pgbase >> PAGE_CACHE_SHIFT;
 799         for (i = pg_index; i < wdata->pages.npages; i++) {
 800                 if (!extent_length) {
 801                         /* We've used up the previous extent */
 802                         bl_put_extent(be);
 803                         bl_put_extent(cow_read);
 804                         bio = bl_submit_bio(WRITE, bio);
 805                         /* Get the next one */
 806                         be = bl_find_get_extent(BLK_LSEG2EXT(header->lseg),
 807                                              isect, &cow_read);
 808                         if (!be || !is_writable(be, isect)) {
 809                                 header->pnfs_error = -EINVAL;
 810                                 goto out;
 811                         }
 812                         if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
 813                                 if (likely(!bl_push_one_short_extent(
 814                                                                 be->be_inval)))
 815                                         par->bse_count++;
 816                                 else {
 817                                         header->pnfs_error = -ENOMEM;
 818                                         goto out;
 819                                 }
 820                         }
 821                         extent_length = be->be_length -
 822                             (isect - be->be_f_offset);
 823                 }
 824
 825                 dprintk("%s offset %lld count %Zu\n", __func__, offset, count);
 826                 pg_offset = offset & ~PAGE_CACHE_MASK;
 827                 if (pg_offset + count > PAGE_CACHE_SIZE)
 828                         pg_len = PAGE_CACHE_SIZE - pg_offset;
 829                 else
 830                         pg_len = count;
 831
 832                 saved_len = pg_len;
 833                 if (be->be_state == PNFS_BLOCK_INVALID_DATA &&
 834                     !bl_is_sector_init(be->be_inval, isect)) {
 835                         ret = bl_read_partial_page_sync(pages[i], cow_read,
 836                                                         pg_offset, pg_len, true);
 837                         if (ret) {
 838                                 dprintk("%s bl_read_partial_page_sync fail %d\n",
 839                                         __func__, ret);
 840                                 header->pnfs_error = ret;
 841                                 goto out;
 842                         }
 843
 844                         ret = bl_mark_sectors_init(be->be_inval, isect,
 845                                                        PAGE_CACHE_SECTORS);
 846                         if (unlikely(ret)) {
 847                                 dprintk("%s bl_mark_sectors_init fail %d\n",
 848                                         __func__, ret);
 849                                 header->pnfs_error = ret;
 850                                 goto out;
 851                         }
 852
 853                         /* Expand to full page write */
 854                         pg_offset = 0;
 855                         pg_len = PAGE_CACHE_SIZE;
 856                 } else if  ((pg_offset & (SECTOR_SIZE - 1)) ||
 857                             (pg_len & (SECTOR_SIZE - 1))){
 858                         /* ahh, nasty case. We have to do sync full sector
 859                          * read-modify-write cycles.
 860                          */
 861                         unsigned int saved_offset = pg_offset;
 862                         ret = bl_read_partial_page_sync(pages[i], be, pg_offset,
 863                                                         pg_len, false);
 864                         pg_offset = round_down(pg_offset, SECTOR_SIZE);
 865                         pg_len = round_up(saved_offset + pg_len, SECTOR_SIZE)
 866                                  - pg_offset;
 867                 }
 868
 869
 870                 bio = do_add_page_to_bio(bio, wdata->pages.npages - i, WRITE,
 871                                          isect, pages[i], be,
 872                                          bl_end_io_write, par,
 873                                          pg_offset, pg_len);
 874                 if (IS_ERR(bio)) {
 875                         header->pnfs_error = PTR_ERR(bio);
 876                         bio = NULL;
 877                         goto out;
 878                 }
 879                 offset += saved_len;
 880                 count -= saved_len;
 881                 isect += PAGE_CACHE_SECTORS;
 882                 last_isect = isect;
 883                 extent_length -= PAGE_CACHE_SECTORS;
 884         }
 885
 886         /* Last page inside INVALID extent */
 887         if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
 888                 bio = bl_submit_bio(WRITE, bio);
 889                 temp = last_isect >> PAGE_CACHE_SECTOR_SHIFT;
 890                 npg_zero = npg_per_block - do_div(temp, npg_per_block);
 891                 if (npg_zero < npg_per_block) {
 892                         last = 1;
 893                         goto fill_invalid_ext;
 894                 }
 895         }
 896
 897 write_done:
 898         wdata->res.count = wdata->args.count;
 899 out:
 900         bl_put_extent(be);
 901         bl_put_extent(cow_read);
 902         bl_submit_bio(WRITE, bio);
 903         put_parallel(par);
 904         return PNFS_ATTEMPTED;
 905 out_mds:
 906         bl_put_extent(be);
 907         bl_put_extent(cow_read);
 908         kfree(par);
 909         return PNFS_NOT_ATTEMPTED;
 910 }
 911
 912 /* FIXME - range ignored */
 913 static void
 914 release_extents(struct pnfs_block_layout *bl, struct pnfs_layout_range *range)
 915 {
 916         int i;
 917         struct pnfs_block_extent *be;
 918
 919         spin_lock(&bl->bl_ext_lock);
 920         for (i = 0; i < EXTENT_LISTS; i++) {
 921                 while (!list_empty(&bl->bl_extents[i])) {
 922                         be = list_first_entry(&bl->bl_extents[i],
 923                                               struct pnfs_block_extent,
 924                                               be_node);
 925                         list_del(&be->be_node);
 926                         bl_put_extent(be);
 927                 }
 928         }
 929         spin_unlock(&bl->bl_ext_lock);
 930 }
 931
 932 static void
 933 release_inval_marks(struct pnfs_inval_markings *marks)
 934 {
 935         struct pnfs_inval_tracking *pos, *temp;
 936         struct pnfs_block_short_extent *se, *stemp;
 937
 938         list_for_each_entry_safe(pos, temp, &marks->im_tree.mtt_stub, it_link) {
 939                 list_del(&pos->it_link);
 940                 kfree(pos);
 941         }
 942
 943         list_for_each_entry_safe(se, stemp, &marks->im_extents, bse_node) {
 944                 list_del(&se->bse_node);
 945                 kfree(se);
 946         }
 947         return;
 948 }
 949
 950 static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo)
 951 {
 952         struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
 953
 954         dprintk("%s enter\n", __func__);
 955         release_extents(bl, NULL);
 956         release_inval_marks(&bl->bl_inval);
 957         kfree(bl);
 958 }
 959
 960 static struct pnfs_layout_hdr *bl_alloc_layout_hdr(struct inode *inode,
 961                                                    gfp_t gfp_flags)
 962 {
 963         struct pnfs_block_layout *bl;
 964
 965         dprintk("%s enter\n", __func__);
 966         bl = kzalloc(sizeof(*bl), gfp_flags);
 967         if (!bl)
 968                 return NULL;
 969         spin_lock_init(&bl->bl_ext_lock);
 970         INIT_LIST_HEAD(&bl->bl_extents[0]);
 971         INIT_LIST_HEAD(&bl->bl_extents[1]);
 972         INIT_LIST_HEAD(&bl->bl_commit);
 973         INIT_LIST_HEAD(&bl->bl_committing);
 974         bl->bl_count = 0;
 975         bl->bl_blocksize = NFS_SERVER(inode)->pnfs_blksize >> SECTOR_SHIFT;
 976         BL_INIT_INVAL_MARKS(&bl->bl_inval, bl->bl_blocksize);
 977         return &bl->bl_layout;
 978 }
 979
 980 static void bl_free_lseg(struct pnfs_layout_segment *lseg)
 981 {
 982         dprintk("%s enter\n", __func__);
 983         kfree(lseg);
 984 }
 985
 986 /* We pretty much ignore lseg, and store all data layout wide, so we
 987  * can correctly merge.
 988  */
 989 static struct pnfs_layout_segment *bl_alloc_lseg(struct pnfs_layout_hdr *lo,
 990                                                  struct nfs4_layoutget_res *lgr,
 991                                                  gfp_t gfp_flags)
 992 {
 993         struct pnfs_layout_segment *lseg;
 994         int status;
 995
 996         dprintk("%s enter\n", __func__);
 997         lseg = kzalloc(sizeof(*lseg), gfp_flags);
 998         if (!lseg)
 999                 return ERR_PTR(-ENOMEM);
1000         status = nfs4_blk_process_layoutget(lo, lgr, gfp_flags);
1001         if (status) {
1002                 /* We don't want to call the full-blown bl_free_lseg,
1003                  * since on error extents were not touched.
1004                  */
1005                 kfree(lseg);
1006                 return ERR_PTR(status);
1007         }
1008         return lseg;
1009 }
1010
1011 static void
1012 bl_encode_layoutcommit(struct pnfs_layout_hdr *lo, struct xdr_stream *xdr,
1013                        const struct nfs4_layoutcommit_args *arg)
1014 {
1015         dprintk("%s enter\n", __func__);
1016         encode_pnfs_block_layoutupdate(BLK_LO2EXT(lo), xdr, arg);
1017 }
1018
1019 static void
1020 bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata)
1021 {
1022         struct pnfs_layout_hdr *lo = NFS_I(lcdata->args.inode)->layout;
1023
1024         dprintk("%s enter\n", __func__);
1025         clean_pnfs_block_layoutupdate(BLK_LO2EXT(lo), &lcdata->args, lcdata->res.status);
1026 }
1027
1028 static void free_blk_mountid(struct block_mount_id *mid)
1029 {
1030         if (mid) {
1031                 struct pnfs_block_dev *dev, *tmp;
1032
1033                 /* No need to take bm_lock as we are last user freeing bm_devlist */
1034                 list_for_each_entry_safe(dev, tmp, &mid->bm_devlist, bm_node) {
1035                         list_del(&dev->bm_node);
1036                         bl_free_block_dev(dev);
1037                 }
1038                 kfree(mid);
1039         }
1040 }
1041
1042 /* This is mostly copied from the filelayout_get_device_info function.
1043  * It seems much of this should be at the generic pnfs level.
1044  */
1045 static struct pnfs_block_dev *
1046 nfs4_blk_get_deviceinfo(struct nfs_server *server, const struct nfs_fh *fh,
1047                         struct nfs4_deviceid *d_id)
1048 {
1049         struct pnfs_device *dev;
1050         struct pnfs_block_dev *rv;
1051         u32 max_resp_sz;
1052         int max_pages;
1053         struct page **pages = NULL;
1054         int i, rc;
1055
1056         /*
1057          * Use the session max response size as the basis for setting
1058          * GETDEVICEINFO's maxcount
1059          */
1060         max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz;
1061         max_pages = nfs_page_array_len(0, max_resp_sz);
1062         dprintk("%s max_resp_sz %u max_pages %d\n",
1063                 __func__, max_resp_sz, max_pages);
1064
1065         dev = kmalloc(sizeof(*dev), GFP_NOFS);
1066         if (!dev) {
1067                 dprintk("%s kmalloc failed\n", __func__);
1068                 return ERR_PTR(-ENOMEM);
1069         }
1070
1071         pages = kzalloc(max_pages * sizeof(struct page *), GFP_NOFS);
1072         if (pages == NULL) {
1073                 kfree(dev);
1074                 return ERR_PTR(-ENOMEM);
1075         }
1076         for (i = 0; i < max_pages; i++) {
1077                 pages[i] = alloc_page(GFP_NOFS);
1078                 if (!pages[i]) {
1079                         rv = ERR_PTR(-ENOMEM);
1080                         goto out_free;
1081                 }
1082         }
1083
1084         memcpy(&dev->dev_id, d_id, sizeof(*d_id));
1085         dev->layout_type = LAYOUT_BLOCK_VOLUME;
1086         dev->pages = pages;
1087         dev->pgbase = 0;
1088         dev->pglen = PAGE_SIZE * max_pages;
1089         dev->mincount = 0;
1090
1091         dprintk("%s: dev_id: %s\n", __func__, dev->dev_id.data);
1092         rc = nfs4_proc_getdeviceinfo(server, dev);
1093         dprintk("%s getdevice info returns %d\n", __func__, rc);
1094         if (rc) {
1095                 rv = ERR_PTR(rc);
1096                 goto out_free;
1097         }
1098
1099         rv = nfs4_blk_decode_device(server, dev);
1100  out_free:
1101         for (i = 0; i < max_pages; i++)
1102                 __free_page(pages[i]);
1103         kfree(pages);
1104         kfree(dev);
1105         return rv;
1106 }
1107
1108 static int
1109 bl_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *fh)
1110 {
1111         struct block_mount_id *b_mt_id = NULL;
1112         struct pnfs_devicelist *dlist = NULL;
1113         struct pnfs_block_dev *bdev;
1114         LIST_HEAD(block_disklist);
1115         int status, i;
1116
1117         dprintk("%s enter\n", __func__);
1118
1119         if (server->pnfs_blksize == 0) {
1120                 dprintk("%s Server did not return blksize\n", __func__);
1121                 return -EINVAL;
1122         }
1123         b_mt_id = kzalloc(sizeof(struct block_mount_id), GFP_NOFS);
1124         if (!b_mt_id) {
1125                 status = -ENOMEM;
1126                 goto out_error;
1127         }
1128         /* Initialize nfs4 block layout mount id */
1129         spin_lock_init(&b_mt_id->bm_lock);
1130         INIT_LIST_HEAD(&b_mt_id->bm_devlist);
1131
1132         dlist = kmalloc(sizeof(struct pnfs_devicelist), GFP_NOFS);
1133         if (!dlist) {
1134                 status = -ENOMEM;
1135                 goto out_error;
1136         }
1137         dlist->eof = 0;
1138         while (!dlist->eof) {
1139                 status = nfs4_proc_getdevicelist(server, fh, dlist);
1140                 if (status)
1141                         goto out_error;
1142                 dprintk("%s GETDEVICELIST numdevs=%i, eof=%i\n",
1143                         __func__, dlist->num_devs, dlist->eof);
1144                 for (i = 0; i < dlist->num_devs; i++) {
1145                         bdev = nfs4_blk_get_deviceinfo(server, fh,
1146                                                        &dlist->dev_id[i]);
1147                         if (IS_ERR(bdev)) {
1148                                 status = PTR_ERR(bdev);
1149                                 goto out_error;
1150                         }
1151                         spin_lock(&b_mt_id->bm_lock);
1152                         list_add(&bdev->bm_node, &b_mt_id->bm_devlist);
1153                         spin_unlock(&b_mt_id->bm_lock);
1154                 }
1155         }
1156         dprintk("%s SUCCESS\n", __func__);
1157         server->pnfs_ld_data = b_mt_id;
1158
1159  out_return:
1160         kfree(dlist);
1161         return status;
1162
1163  out_error:
1164         free_blk_mountid(b_mt_id);
1165         goto out_return;
1166 }
1167
1168 static int
1169 bl_clear_layoutdriver(struct nfs_server *server)
1170 {
1171         struct block_mount_id *b_mt_id = server->pnfs_ld_data;
1172
1173         dprintk("%s enter\n", __func__);
1174         free_blk_mountid(b_mt_id);
1175         dprintk("%s RETURNS\n", __func__);
1176         return 0;
1177 }
1178
1179 static bool
1180 is_aligned_req(struct nfs_page *req, unsigned int alignment)
1181 {
1182         return IS_ALIGNED(req->wb_offset, alignment) &&
1183                IS_ALIGNED(req->wb_bytes, alignment);
1184 }
1185
1186 static void
1187 bl_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1188 {
1189         if (pgio->pg_dreq != NULL &&
1190             !is_aligned_req(req, SECTOR_SIZE))
1191                 nfs_pageio_reset_read_mds(pgio);
1192         else
1193                 pnfs_generic_pg_init_read(pgio, req);
1194 }
1195
1196 static bool
1197 bl_pg_test_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
1198                 struct nfs_page *req)
1199 {
1200         if (pgio->pg_dreq != NULL &&
1201             !is_aligned_req(req, SECTOR_SIZE))
1202                 return false;
1203
1204         return pnfs_generic_pg_test(pgio, prev, req);
1205 }
1206
1207 static void
1208 bl_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1209 {
1210         if (pgio->pg_dreq != NULL &&
1211             !is_aligned_req(req, PAGE_CACHE_SIZE))
1212                 nfs_pageio_reset_write_mds(pgio);
1213         else
1214                 pnfs_generic_pg_init_write(pgio, req);
1215 }
1216
1217 static bool
1218 bl_pg_test_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
1219                  struct nfs_page *req)
1220 {
1221         if (pgio->pg_dreq != NULL &&
1222             !is_aligned_req(req, PAGE_CACHE_SIZE))
1223                 return false;
1224
1225         return pnfs_generic_pg_test(pgio, prev, req);
1226 }
1227
1228 static const struct nfs_pageio_ops bl_pg_read_ops = {
1229         .pg_init = bl_pg_init_read,
1230         .pg_test = bl_pg_test_read,
1231         .pg_doio = pnfs_generic_pg_readpages,
1232 };
1233
1234 static const struct nfs_pageio_ops bl_pg_write_ops = {
1235         .pg_init = bl_pg_init_write,
1236         .pg_test = bl_pg_test_write,
1237         .pg_doio = pnfs_generic_pg_writepages,
1238 };
1239
1240 static struct pnfs_layoutdriver_type blocklayout_type = {
1241         .id                             = LAYOUT_BLOCK_VOLUME,
1242         .name                           = "LAYOUT_BLOCK_VOLUME",
1243         .read_pagelist                  = bl_read_pagelist,
1244         .write_pagelist                 = bl_write_pagelist,
1245         .alloc_layout_hdr               = bl_alloc_layout_hdr,
1246         .free_layout_hdr                = bl_free_layout_hdr,
1247         .alloc_lseg                     = bl_alloc_lseg,
1248         .free_lseg                      = bl_free_lseg,
1249         .encode_layoutcommit            = bl_encode_layoutcommit,
1250         .cleanup_layoutcommit           = bl_cleanup_layoutcommit,
1251         .set_layoutdriver               = bl_set_layoutdriver,
1252         .clear_layoutdriver             = bl_clear_layoutdriver,
1253         .pg_read_ops                    = &bl_pg_read_ops,
1254         .pg_write_ops                   = &bl_pg_write_ops,
1255 };
1256
1257 static const struct rpc_pipe_ops bl_upcall_ops = {
1258         .upcall         = rpc_pipe_generic_upcall,
1259         .downcall       = bl_pipe_downcall,
1260         .destroy_msg    = bl_pipe_destroy_msg,
1261 };
1262
1263 static struct dentry *nfs4blocklayout_register_sb(struct super_block *sb,
1264                                             struct rpc_pipe *pipe)
1265 {
1266         struct dentry *dir, *dentry;
1267
1268         dir = rpc_d_lookup_sb(sb, NFS_PIPE_DIRNAME);
1269         if (dir == NULL)
1270                 return ERR_PTR(-ENOENT);
1271         dentry = rpc_mkpipe_dentry(dir, "blocklayout", NULL, pipe);
1272         dput(dir);
1273         return dentry;
1274 }
1275
1276 static void nfs4blocklayout_unregister_sb(struct super_block *sb,
1277                                           struct rpc_pipe *pipe)
1278 {
1279         if (pipe->dentry)
1280                 rpc_unlink(pipe->dentry);
1281 }
1282
1283 static int rpc_pipefs_event(struct notifier_block *nb, unsigned long event,
1284                            void *ptr)
1285 {
1286         struct super_block *sb = ptr;
1287         struct net *net = sb->s_fs_info;
1288         struct nfs_net *nn = net_generic(net, nfs_net_id);
1289         struct dentry *dentry;
1290         int ret = 0;
1291
1292         if (!try_module_get(THIS_MODULE))
1293                 return 0;
1294
1295         if (nn->bl_device_pipe == NULL) {
1296                 module_put(THIS_MODULE);
1297                 return 0;
1298         }
1299
1300         switch (event) {
1301         case RPC_PIPEFS_MOUNT:
1302                 dentry = nfs4blocklayout_register_sb(sb, nn->bl_device_pipe);
1303                 if (IS_ERR(dentry)) {
1304                         ret = PTR_ERR(dentry);
1305                         break;
1306                 }
1307                 nn->bl_device_pipe->dentry = dentry;
1308                 break;
1309         case RPC_PIPEFS_UMOUNT:
1310                 if (nn->bl_device_pipe->dentry)
1311                         nfs4blocklayout_unregister_sb(sb, nn->bl_device_pipe);
1312                 break;
1313         default:
1314                 ret = -ENOTSUPP;
1315                 break;
1316         }
1317         module_put(THIS_MODULE);
1318         return ret;
1319 }
1320
1321 static struct notifier_block nfs4blocklayout_block = {
1322         .notifier_call = rpc_pipefs_event,
1323 };
1324
1325 static struct dentry *nfs4blocklayout_register_net(struct net *net,
1326                                                    struct rpc_pipe *pipe)
1327 {
1328         struct super_block *pipefs_sb;
1329         struct dentry *dentry;
1330
1331         pipefs_sb = rpc_get_sb_net(net);
1332         if (!pipefs_sb)
1333                 return NULL;
1334         dentry = nfs4blocklayout_register_sb(pipefs_sb, pipe);
1335         rpc_put_sb_net(net);
1336         return dentry;
1337 }
1338
1339 static void nfs4blocklayout_unregister_net(struct net *net,
1340                                            struct rpc_pipe *pipe)
1341 {
1342         struct super_block *pipefs_sb;
1343
1344         pipefs_sb = rpc_get_sb_net(net);
1345         if (pipefs_sb) {
1346                 nfs4blocklayout_unregister_sb(pipefs_sb, pipe);
1347                 rpc_put_sb_net(net);
1348         }
1349 }
1350
1351 static int nfs4blocklayout_net_init(struct net *net)
1352 {
1353         struct nfs_net *nn = net_generic(net, nfs_net_id);
1354         struct dentry *dentry;
1355
1356         init_waitqueue_head(&nn->bl_wq);
1357         nn->bl_device_pipe = rpc_mkpipe_data(&bl_upcall_ops, 0);
1358         if (IS_ERR(nn->bl_device_pipe))
1359                 return PTR_ERR(nn->bl_device_pipe);
1360         dentry = nfs4blocklayout_register_net(net, nn->bl_device_pipe);
1361         if (IS_ERR(dentry)) {
1362                 rpc_destroy_pipe_data(nn->bl_device_pipe);
1363                 return PTR_ERR(dentry);
1364         }
1365         nn->bl_device_pipe->dentry = dentry;
1366         return 0;
1367 }
1368
1369 static void nfs4blocklayout_net_exit(struct net *net)
1370 {
1371         struct nfs_net *nn = net_generic(net, nfs_net_id);
1372
1373         nfs4blocklayout_unregister_net(net, nn->bl_device_pipe);
1374         rpc_destroy_pipe_data(nn->bl_device_pipe);
1375         nn->bl_device_pipe = NULL;
1376 }
1377
1378 static struct pernet_operations nfs4blocklayout_net_ops = {
1379         .init = nfs4blocklayout_net_init,
1380         .exit = nfs4blocklayout_net_exit,
1381 };
1382
1383 static int __init nfs4blocklayout_init(void)
1384 {
1385         int ret;
1386
1387         dprintk("%s: NFSv4 Block Layout Driver Registering...\n", __func__);
1388
1389         ret = pnfs_register_layoutdriver(&blocklayout_type);
1390         if (ret)
1391                 goto out;
1392
1393         ret = rpc_pipefs_notifier_register(&nfs4blocklayout_block);
1394         if (ret)
1395                 goto out_remove;
1396         ret = register_pernet_subsys(&nfs4blocklayout_net_ops);
1397         if (ret)
1398                 goto out_notifier;
1399 out:
1400         return ret;
1401
1402 out_notifier:
1403         rpc_pipefs_notifier_unregister(&nfs4blocklayout_block);
1404 out_remove:
1405         pnfs_unregister_layoutdriver(&blocklayout_type);
1406         return ret;
1407 }
1408
1409 static void __exit nfs4blocklayout_exit(void)
1410 {
1411         dprintk("%s: NFSv4 Block Layout Driver Unregistering...\n",
1412                __func__);
1413
1414         rpc_pipefs_notifier_unregister(&nfs4blocklayout_block);
1415         unregister_pernet_subsys(&nfs4blocklayout_net_ops);
1416         pnfs_unregister_layoutdriver(&blocklayout_type);
1417 }
1418
1419 MODULE_ALIAS("nfs-layouttype4-3");
1420
1421 module_init(nfs4blocklayout_init);
1422 module_exit(nfs4blocklayout_exit);