fs/nfs/blocklayout/blocklayout.c

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