fs/afs/write.c

   1 /* handling of writes to regular files and writing back to the server
   2  *
   3  * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
   4  * Written by David Howells (dhowells@redhat.com)
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License
   8  * as published by the Free Software Foundation; either version
   9  * 2 of the License, or (at your option) any later version.
  10  */
  11
  12 #include <linux/slab.h>
  13 #include <linux/fs.h>
  14 #include <linux/pagemap.h>
  15 #include <linux/writeback.h>
  16 #include <linux/pagevec.h>
  17 #include "internal.h"
  18
  19 static int afs_write_back_from_locked_page(struct afs_writeback *wb,
  20                                            struct page *page);
  21
  22 /*
  23  * mark a page as having been made dirty and thus needing writeback
  24  */
  25 int afs_set_page_dirty(struct page *page)
  26 {
  27         _enter("");
  28         return __set_page_dirty_nobuffers(page);
  29 }
  30
  31 /*
  32  * unlink a writeback record because its usage has reached zero
  33  * - must be called with the wb->vnode->writeback_lock held
  34  */
  35 static void afs_unlink_writeback(struct afs_writeback *wb)
  36 {
  37         struct afs_writeback *front;
  38         struct afs_vnode *vnode = wb->vnode;
  39
  40         list_del_init(&wb->link);
  41         if (!list_empty(&vnode->writebacks)) {
  42                 /* if an fsync rises to the front of the queue then wake it
  43                  * up */
  44                 front = list_entry(vnode->writebacks.next,
  45                                    struct afs_writeback, link);
  46                 if (front->state == AFS_WBACK_SYNCING) {
  47                         _debug("wake up sync");
  48                         front->state = AFS_WBACK_COMPLETE;
  49                         wake_up(&front->waitq);
  50                 }
  51         }
  52 }
  53
  54 /*
  55  * free a writeback record
  56  */
  57 static void afs_free_writeback(struct afs_writeback *wb)
  58 {
  59         _enter("");
  60         key_put(wb->key);
  61         kfree(wb);
  62 }
  63
  64 /*
  65  * dispose of a reference to a writeback record
  66  */
  67 void afs_put_writeback(struct afs_writeback *wb)
  68 {
  69         struct afs_vnode *vnode = wb->vnode;
  70
  71         _enter("{%d}", wb->usage);
  72
  73         spin_lock(&vnode->writeback_lock);
  74         if (--wb->usage == 0)
  75                 afs_unlink_writeback(wb);
  76         else
  77                 wb = NULL;
  78         spin_unlock(&vnode->writeback_lock);
  79         if (wb)
  80                 afs_free_writeback(wb);
  81 }
  82
  83 /*
  84  * partly or wholly fill a page that's under preparation for writing
  85  */
  86 static int afs_fill_page(struct afs_vnode *vnode, struct key *key,
  87                          unsigned start, unsigned len, struct page *page)
  88 {
  89         int ret;
  90
  91         _enter(",,%u,%u", start, len);
  92
  93         ASSERTCMP(start + len, <=, PAGE_SIZE);
  94
  95         ret = afs_vnode_fetch_data(vnode, key, start, len, page);
  96         if (ret < 0) {
  97                 if (ret == -ENOENT) {
  98                         _debug("got NOENT from server"
  99                                " - marking file deleted and stale");
 100                         set_bit(AFS_VNODE_DELETED, &vnode->flags);
 101                         ret = -ESTALE;
 102                 }
 103         }
 104
 105         _leave(" = %d", ret);
 106         return ret;
 107 }
 108
 109 /*
 110  * prepare a page for being written to
 111  */
 112 static int afs_prepare_page(struct afs_vnode *vnode, struct page *page,
 113                             struct key *key, unsigned offset, unsigned to)
 114 {
 115         unsigned eof, tail, start, stop, len;
 116         loff_t i_size, pos;
 117         void *p;
 118         int ret;
 119
 120         _enter("");
 121
 122         if (offset == 0 && to == PAGE_SIZE)
 123                 return 0;
 124
 125         p = kmap_atomic(page, KM_USER0);
 126
 127         i_size = i_size_read(&vnode->vfs_inode);
 128         pos = (loff_t) page->index << PAGE_SHIFT;
 129         if (pos >= i_size) {
 130                 /* partial write, page beyond EOF */
 131                 _debug("beyond");
 132                 if (offset > 0)
 133                         memset(p, 0, offset);
 134                 if (to < PAGE_SIZE)
 135                         memset(p + to, 0, PAGE_SIZE - to);
 136                 kunmap_atomic(p, KM_USER0);
 137                 return 0;
 138         }
 139
 140         if (i_size - pos >= PAGE_SIZE) {
 141                 /* partial write, page entirely before EOF */
 142                 _debug("before");
 143                 tail = eof = PAGE_SIZE;
 144         } else {
 145                 /* partial write, page overlaps EOF */
 146                 eof = i_size - pos;
 147                 _debug("overlap %u", eof);
 148                 tail = max(eof, to);
 149                 if (tail < PAGE_SIZE)
 150                         memset(p + tail, 0, PAGE_SIZE - tail);
 151                 if (offset > eof)
 152                         memset(p + eof, 0, PAGE_SIZE - eof);
 153         }
 154
 155         kunmap_atomic(p, KM_USER0);
 156
 157         ret = 0;
 158         if (offset > 0 || eof > to) {
 159                 /* need to fill one or two bits that aren't going to be written
 160                  * (cover both fillers in one read if there are two) */
 161                 start = (offset > 0) ? 0 : to;
 162                 stop = (eof > to) ? eof : offset;
 163                 len = stop - start;
 164                 _debug("wr=%u-%u av=0-%u rd=%u@%u",
 165                        offset, to, eof, start, len);
 166                 ret = afs_fill_page(vnode, key, start, len, page);
 167         }
 168
 169         _leave(" = %d", ret);
 170         return ret;
 171 }
 172
 173 /*
 174  * prepare to perform part of a write to a page
 175  * - the caller holds the page locked, preventing it from being written out or
 176  *   modified by anyone else
 177  */
 178 int afs_prepare_write(struct file *file, struct page *page,
 179                       unsigned offset, unsigned to)
 180 {
 181         struct afs_writeback *candidate, *wb;
 182         struct afs_vnode *vnode = AFS_FS_I(file->f_dentry->d_inode);
 183         struct key *key = file->private_data;
 184         pgoff_t index;
 185         int ret;
 186
 187         _enter("{%x:%u},{%lx},%u,%u",
 188                vnode->fid.vid, vnode->fid.vnode, page->index, offset, to);
 189
 190         candidate = kzalloc(sizeof(*candidate), GFP_KERNEL);
 191         if (!candidate)
 192                 return -ENOMEM;
 193         candidate->vnode = vnode;
 194         candidate->first = candidate->last = page->index;
 195         candidate->offset_first = offset;
 196         candidate->to_last = to;
 197         candidate->usage = 1;
 198         candidate->state = AFS_WBACK_PENDING;
 199         init_waitqueue_head(&candidate->waitq);
 200
 201         if (!PageUptodate(page)) {
 202                 _debug("not up to date");
 203                 ret = afs_prepare_page(vnode, page, key, offset, to);
 204                 if (ret < 0) {
 205                         kfree(candidate);
 206                         _leave(" = %d [prep]", ret);
 207                         return ret;
 208                 }
 209         }
 210
 211 try_again:
 212         index = page->index;
 213         spin_lock(&vnode->writeback_lock);
 214
 215         /* see if this page is already pending a writeback under a suitable key
 216          * - if so we can just join onto that one */
 217         wb = (struct afs_writeback *) page_private(page);
 218         if (wb) {
 219                 if (wb->key == key && wb->state == AFS_WBACK_PENDING)
 220                         goto subsume_in_current_wb;
 221                 goto flush_conflicting_wb;
 222         }
 223
 224         if (index > 0) {
 225                 /* see if we can find an already pending writeback that we can
 226                  * append this page to */
 227                 list_for_each_entry(wb, &vnode->writebacks, link) {
 228                         if (wb->last == index - 1 && wb->key == key &&
 229                             wb->state == AFS_WBACK_PENDING)
 230                                 goto append_to_previous_wb;
 231                 }
 232         }
 233
 234         list_add_tail(&candidate->link, &vnode->writebacks);
 235         candidate->key = key_get(key);
 236         spin_unlock(&vnode->writeback_lock);
 237         SetPagePrivate(page);
 238         set_page_private(page, (unsigned long) candidate);
 239         _leave(" = 0 [new]");
 240         return 0;
 241
 242 subsume_in_current_wb:
 243         _debug("subsume");
 244         ASSERTRANGE(wb->first, <=, index, <=, wb->last);
 245         if (index == wb->first && offset < wb->offset_first)
 246                 wb->offset_first = offset;
 247         if (index == wb->last && to > wb->to_last)
 248                 wb->to_last = to;
 249         spin_unlock(&vnode->writeback_lock);
 250         kfree(candidate);
 251         _leave(" = 0 [sub]");
 252         return 0;
 253
 254 append_to_previous_wb:
 255         _debug("append into %lx-%lx", wb->first, wb->last);
 256         wb->usage++;
 257         wb->last++;
 258         wb->to_last = to;
 259         spin_unlock(&vnode->writeback_lock);
 260         SetPagePrivate(page);
 261         set_page_private(page, (unsigned long) wb);
 262         kfree(candidate);
 263         _leave(" = 0 [app]");
 264         return 0;
 265
 266         /* the page is currently bound to another context, so if it's dirty we
 267          * need to flush it before we can use the new context */
 268 flush_conflicting_wb:
 269         _debug("flush conflict");
 270         if (wb->state == AFS_WBACK_PENDING)
 271                 wb->state = AFS_WBACK_CONFLICTING;
 272         spin_unlock(&vnode->writeback_lock);
 273         if (PageDirty(page)) {
 274                 ret = afs_write_back_from_locked_page(wb, page);
 275                 if (ret < 0) {
 276                         afs_put_writeback(candidate);
 277                         _leave(" = %d", ret);
 278                         return ret;
 279                 }
 280         }
 281
 282         /* the page holds a ref on the writeback record */
 283         afs_put_writeback(wb);
 284         set_page_private(page, 0);
 285         ClearPagePrivate(page);
 286         goto try_again;
 287 }
 288
 289 /*
 290  * finalise part of a write to a page
 291  */
 292 int afs_commit_write(struct file *file, struct page *page,
 293                      unsigned offset, unsigned to)
 294 {
 295         struct afs_vnode *vnode = AFS_FS_I(file->f_dentry->d_inode);
 296         loff_t i_size, maybe_i_size;
 297
 298         _enter("{%x:%u},{%lx},%u,%u",
 299                vnode->fid.vid, vnode->fid.vnode, page->index, offset, to);
 300
 301         maybe_i_size = (loff_t) page->index << PAGE_SHIFT;
 302         maybe_i_size += to;
 303
 304         i_size = i_size_read(&vnode->vfs_inode);
 305         if (maybe_i_size > i_size) {
 306                 spin_lock(&vnode->writeback_lock);
 307                 i_size = i_size_read(&vnode->vfs_inode);
 308                 if (maybe_i_size > i_size)
 309                         i_size_write(&vnode->vfs_inode, maybe_i_size);
 310                 spin_unlock(&vnode->writeback_lock);
 311         }
 312
 313         SetPageUptodate(page);
 314         set_page_dirty(page);
 315         if (PageDirty(page))
 316                 _debug("dirtied");
 317
 318         return 0;
 319 }
 320
 321 /*
 322  * kill all the pages in the given range
 323  */
 324 static void afs_kill_pages(struct afs_vnode *vnode, bool error,
 325                            pgoff_t first, pgoff_t last)
 326 {
 327         struct pagevec pv;
 328         unsigned count, loop;
 329
 330         _enter("{%x:%u},%lx-%lx",
 331                vnode->fid.vid, vnode->fid.vnode, first, last);
 332
 333         pagevec_init(&pv, 0);
 334
 335         do {
 336                 _debug("kill %lx-%lx", first, last);
 337
 338                 count = last - first + 1;
 339                 if (count > PAGEVEC_SIZE)
 340                         count = PAGEVEC_SIZE;
 341                 pv.nr = find_get_pages_contig(vnode->vfs_inode.i_mapping,
 342                                               first, count, pv.pages);
 343                 ASSERTCMP(pv.nr, ==, count);
 344
 345                 for (loop = 0; loop < count; loop++) {
 346                         ClearPageUptodate(pv.pages[loop]);
 347                         if (error)
 348                                 SetPageError(pv.pages[loop]);
 349                         end_page_writeback(pv.pages[loop]);
 350                 }
 351
 352                 __pagevec_release(&pv);
 353         } while (first < last);
 354
 355         _leave("");
 356 }
 357
 358 /*
 359  * synchronously write back the locked page and any subsequent non-locked dirty
 360  * pages also covered by the same writeback record
 361  */
 362 static int afs_write_back_from_locked_page(struct afs_writeback *wb,
 363                                            struct page *primary_page)
 364 {
 365         struct page *pages[8], *page;
 366         unsigned long count;
 367         unsigned n, offset, to;
 368         pgoff_t start, first, last;
 369         int loop, ret;
 370
 371         _enter(",%lx", primary_page->index);
 372
 373         count = 1;
 374         if (!clear_page_dirty_for_io(primary_page))
 375                 BUG();
 376         if (test_set_page_writeback(primary_page))
 377                 BUG();
 378
 379         /* find all consecutive lockable dirty pages, stopping when we find a
 380          * page that is not immediately lockable, is not dirty or is missing,
 381          * or we reach the end of the range */
 382         start = primary_page->index;
 383         if (start >= wb->last)
 384                 goto no_more;
 385         start++;
 386         do {
 387                 _debug("more %lx [%lx]", start, count);
 388                 n = wb->last - start + 1;
 389                 if (n > ARRAY_SIZE(pages))
 390                         n = ARRAY_SIZE(pages);
 391                 n = find_get_pages_contig(wb->vnode->vfs_inode.i_mapping,
 392                                           start, n, pages);
 393                 _debug("fgpc %u", n);
 394                 if (n == 0)
 395                         goto no_more;
 396                 if (pages[0]->index != start) {
 397                         do {
 398                                 put_page(pages[--n]);
 399                         } while (n > 0);
 400                         goto no_more;
 401                 }
 402
 403                 for (loop = 0; loop < n; loop++) {
 404                         page = pages[loop];
 405                         if (page->index > wb->last)
 406                                 break;
 407                         if (TestSetPageLocked(page))
 408                                 break;
 409                         if (!PageDirty(page) ||
 410                             page_private(page) != (unsigned long) wb) {
 411                                 unlock_page(page);
 412                                 break;
 413                         }
 414                         if (!clear_page_dirty_for_io(page))
 415                                 BUG();
 416                         if (test_set_page_writeback(page))
 417                                 BUG();
 418                         unlock_page(page);
 419                         put_page(page);
 420                 }
 421                 count += loop;
 422                 if (loop < n) {
 423                         for (; loop < n; loop++)
 424                                 put_page(pages[loop]);
 425                         goto no_more;
 426                 }
 427
 428                 start += loop;
 429         } while (start <= wb->last && count < 65536);
 430
 431 no_more:
 432         /* we now have a contiguous set of dirty pages, each with writeback set
 433          * and the dirty mark cleared; the first page is locked and must remain
 434          * so, all the rest are unlocked */
 435         first = primary_page->index;
 436         last = first + count - 1;
 437
 438         offset = (first == wb->first) ? wb->offset_first : 0;
 439         to = (last == wb->last) ? wb->to_last : PAGE_SIZE;
 440
 441         _debug("write back %lx[%u..] to %lx[..%u]", first, offset, last, to);
 442
 443         ret = afs_vnode_store_data(wb, first, last, offset, to);
 444         if (ret < 0) {
 445                 switch (ret) {
 446                 case -EDQUOT:
 447                 case -ENOSPC:
 448                         set_bit(AS_ENOSPC,
 449                                 &wb->vnode->vfs_inode.i_mapping->flags);
 450                         break;
 451                 case -EROFS:
 452                 case -EIO:
 453                 case -EREMOTEIO:
 454                 case -EFBIG:
 455                 case -ENOENT:
 456                 case -ENOMEDIUM:
 457                 case -ENXIO:
 458                         afs_kill_pages(wb->vnode, true, first, last);
 459                         set_bit(AS_EIO, &wb->vnode->vfs_inode.i_mapping->flags);
 460                         break;
 461                 case -EACCES:
 462                 case -EPERM:
 463                 case -ENOKEY:
 464                 case -EKEYEXPIRED:
 465                 case -EKEYREJECTED:
 466                 case -EKEYREVOKED:
 467                         afs_kill_pages(wb->vnode, false, first, last);
 468                         break;
 469                 default:
 470                         break;
 471                 }
 472         } else {
 473                 ret = count;
 474         }
 475
 476         _leave(" = %d", ret);
 477         return ret;
 478 }
 479
 480 /*
 481  * write a page back to the server
 482  * - the caller locked the page for us
 483  */
 484 int afs_writepage(struct page *page, struct writeback_control *wbc)
 485 {
 486         struct backing_dev_info *bdi = page->mapping->backing_dev_info;
 487         struct afs_writeback *wb;
 488         int ret;
 489
 490         _enter("{%lx},", page->index);
 491
 492         wb = (struct afs_writeback *) page_private(page);
 493         ASSERT(wb != NULL);
 494
 495         ret = afs_write_back_from_locked_page(wb, page);
 496         unlock_page(page);
 497         if (ret < 0) {
 498                 _leave(" = %d", ret);
 499                 return 0;
 500         }
 501
 502         wbc->nr_to_write -= ret;
 503         if (wbc->nonblocking && bdi_write_congested(bdi))
 504                 wbc->encountered_congestion = 1;
 505
 506         _leave(" = 0");
 507         return 0;
 508 }
 509
 510 /*
 511  * write a region of pages back to the server
 512  */
 513 int afs_writepages_region(struct address_space *mapping,
 514                           struct writeback_control *wbc,
 515                           pgoff_t index, pgoff_t end, pgoff_t *_next)
 516 {
 517         struct backing_dev_info *bdi = mapping->backing_dev_info;
 518         struct afs_writeback *wb;
 519         struct page *page;
 520         int ret, n;
 521
 522         _enter(",,%lx,%lx,", index, end);
 523
 524         do {
 525                 n = find_get_pages_tag(mapping, &index, PAGECACHE_TAG_DIRTY,
 526                                        1, &page);
 527                 if (!n)
 528                         break;
 529
 530                 _debug("wback %lx", page->index);
 531
 532                 if (page->index > end) {
 533                         *_next = index;
 534                         page_cache_release(page);
 535                         _leave(" = 0 [%lx]", *_next);
 536                         return 0;
 537                 }
 538
 539                 /* at this point we hold neither mapping->tree_lock nor lock on
 540                  * the page itself: the page may be truncated or invalidated
 541                  * (changing page->mapping to NULL), or even swizzled back from
 542                  * swapper_space to tmpfs file mapping
 543                  */
 544                 lock_page(page);
 545
 546                 if (page->mapping != mapping) {
 547                         unlock_page(page);
 548                         page_cache_release(page);
 549                         continue;
 550                 }
 551
 552                 if (wbc->sync_mode != WB_SYNC_NONE)
 553                         wait_on_page_writeback(page);
 554
 555                 if (PageWriteback(page) || !PageDirty(page)) {
 556                         unlock_page(page);
 557                         continue;
 558                 }
 559
 560                 wb = (struct afs_writeback *) page_private(page);
 561                 ASSERT(wb != NULL);
 562
 563                 spin_lock(&wb->vnode->writeback_lock);
 564                 wb->state = AFS_WBACK_WRITING;
 565                 spin_unlock(&wb->vnode->writeback_lock);
 566
 567                 ret = afs_write_back_from_locked_page(wb, page);
 568                 unlock_page(page);
 569                 page_cache_release(page);
 570                 if (ret < 0) {
 571                         _leave(" = %d", ret);
 572                         return ret;
 573                 }
 574
 575                 wbc->nr_to_write -= ret;
 576
 577                 if (wbc->nonblocking && bdi_write_congested(bdi)) {
 578                         wbc->encountered_congestion = 1;
 579                         break;
 580                 }
 581
 582                 cond_resched();
 583         } while (index < end && wbc->nr_to_write > 0);
 584
 585         *_next = index;
 586         _leave(" = 0 [%lx]", *_next);
 587         return 0;
 588 }
 589
 590 /*
 591  * write some of the pending data back to the server
 592  */
 593 int afs_writepages(struct address_space *mapping,
 594                    struct writeback_control *wbc)
 595 {
 596         struct backing_dev_info *bdi = mapping->backing_dev_info;
 597         pgoff_t start, end, next;
 598         int ret;
 599
 600         _enter("");
 601
 602         if (wbc->nonblocking && bdi_write_congested(bdi)) {
 603                 wbc->encountered_congestion = 1;
 604                 _leave(" = 0 [congest]");
 605                 return 0;
 606         }
 607
 608         if (wbc->range_cyclic) {
 609                 start = mapping->writeback_index;
 610                 end = -1;
 611                 ret = afs_writepages_region(mapping, wbc, start, end, &next);
 612                 if (start > 0 && wbc->nr_to_write > 0 && ret == 0 &&
 613                     !(wbc->nonblocking && wbc->encountered_congestion))
 614                         ret = afs_writepages_region(mapping, wbc, 0, start,
 615                                                     &next);
 616                 mapping->writeback_index = next;
 617         } else if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) {
 618                 end = (pgoff_t)(LLONG_MAX >> PAGE_CACHE_SHIFT);
 619                 ret = afs_writepages_region(mapping, wbc, 0, end, &next);
 620                 if (wbc->nr_to_write > 0)
 621                         mapping->writeback_index = next;
 622         } else {
 623                 start = wbc->range_start >> PAGE_CACHE_SHIFT;
 624                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
 625                 ret = afs_writepages_region(mapping, wbc, start, end, &next);
 626         }
 627
 628         _leave(" = %d", ret);
 629         return ret;
 630 }
 631
 632 /*
 633  * write an inode back
 634  */
 635 int afs_write_inode(struct inode *inode, int sync)
 636 {
 637         struct afs_vnode *vnode = AFS_FS_I(inode);
 638         int ret;
 639
 640         _enter("{%x:%u},", vnode->fid.vid, vnode->fid.vnode);
 641
 642         ret = 0;
 643         if (sync) {
 644                 ret = filemap_fdatawait(inode->i_mapping);
 645                 if (ret < 0)
 646                         __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
 647         }
 648
 649         _leave(" = %d", ret);
 650         return ret;
 651 }
 652
 653 /*
 654  * completion of write to server
 655  */
 656 void afs_pages_written_back(struct afs_vnode *vnode, struct afs_call *call)
 657 {
 658         struct afs_writeback *wb = call->wb;
 659         struct pagevec pv;
 660         unsigned count, loop;
 661         pgoff_t first = call->first, last = call->last;
 662         bool free_wb;
 663
 664         _enter("{%x:%u},{%lx-%lx}",
 665                vnode->fid.vid, vnode->fid.vnode, first, last);
 666
 667         ASSERT(wb != NULL);
 668
 669         pagevec_init(&pv, 0);
 670
 671         do {
 672                 _debug("done %lx-%lx", first, last);
 673
 674                 count = last - first + 1;
 675                 if (count > PAGEVEC_SIZE)
 676                         count = PAGEVEC_SIZE;
 677                 pv.nr = find_get_pages_contig(call->mapping, first, count,
 678                                               pv.pages);
 679                 ASSERTCMP(pv.nr, ==, count);
 680
 681                 spin_lock(&vnode->writeback_lock);
 682                 for (loop = 0; loop < count; loop++) {
 683                         struct page *page = pv.pages[loop];
 684                         end_page_writeback(page);
 685                         if (page_private(page) == (unsigned long) wb) {
 686                                 set_page_private(page, 0);
 687                                 ClearPagePrivate(page);
 688                                 wb->usage--;
 689                         }
 690                 }
 691                 free_wb = false;
 692                 if (wb->usage == 0) {
 693                         afs_unlink_writeback(wb);
 694                         free_wb = true;
 695                 }
 696                 spin_unlock(&vnode->writeback_lock);
 697                 first += count;
 698                 if (free_wb) {
 699                         afs_free_writeback(wb);
 700                         wb = NULL;
 701                 }
 702
 703                 __pagevec_release(&pv);
 704         } while (first <= last);
 705
 706         _leave("");
 707 }
 708
 709 /*
 710  * write to an AFS file
 711  */
 712 ssize_t afs_file_write(struct kiocb *iocb, const struct iovec *iov,
 713                        unsigned long nr_segs, loff_t pos)
 714 {
 715         struct dentry *dentry = iocb->ki_filp->f_path.dentry;
 716         struct afs_vnode *vnode = AFS_FS_I(dentry->d_inode);
 717         ssize_t result;
 718         size_t count = iov_length(iov, nr_segs);
 719         int ret;
 720
 721         _enter("{%x.%u},{%zu},%lu,",
 722                vnode->fid.vid, vnode->fid.vnode, count, nr_segs);
 723
 724         if (IS_SWAPFILE(&vnode->vfs_inode)) {
 725                 printk(KERN_INFO
 726                        "AFS: Attempt to write to active swap file!\n");
 727                 return -EBUSY;
 728         }
 729
 730         if (!count)
 731                 return 0;
 732
 733         result = generic_file_aio_write(iocb, iov, nr_segs, pos);
 734         if (IS_ERR_VALUE(result)) {
 735                 _leave(" = %zd", result);
 736                 return result;
 737         }
 738
 739         /* return error values for O_SYNC and IS_SYNC() */
 740         if (IS_SYNC(&vnode->vfs_inode) || iocb->ki_filp->f_flags & O_SYNC) {
 741                 ret = afs_fsync(iocb->ki_filp, dentry, 1);
 742                 if (ret < 0)
 743                         result = ret;
 744         }
 745
 746         _leave(" = %zd", result);
 747         return result;
 748 }
 749
 750 /*
 751  * flush the vnode to the fileserver
 752  */
 753 int afs_writeback_all(struct afs_vnode *vnode)
 754 {
 755         struct address_space *mapping = vnode->vfs_inode.i_mapping;
 756         struct writeback_control wbc = {
 757                 .bdi            = mapping->backing_dev_info,
 758                 .sync_mode      = WB_SYNC_ALL,
 759                 .nr_to_write    = LONG_MAX,
 760                 .for_writepages = 1,
 761                 .range_cyclic   = 1,
 762         };
 763         int ret;
 764
 765         _enter("");
 766
 767         ret = mapping->a_ops->writepages(mapping, &wbc);
 768         __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
 769
 770         _leave(" = %d", ret);
 771         return ret;
 772 }
 773
 774 /*
 775  * flush any dirty pages for this process, and check for write errors.
 776  * - the return status from this call provides a reliable indication of
 777  *   whether any write errors occurred for this process.
 778  */
 779 int afs_fsync(struct file *file, struct dentry *dentry, int datasync)
 780 {
 781         struct afs_writeback *wb, *xwb;
 782         struct afs_vnode *vnode = AFS_FS_I(dentry->d_inode);
 783         int ret;
 784
 785         _enter("{%x:%u},{n=%s},%d",
 786                vnode->fid.vid, vnode->fid.vnode, dentry->d_name.name,
 787                datasync);
 788
 789         /* use a writeback record as a marker in the queue - when this reaches
 790          * the front of the queue, all the outstanding writes are either
 791          * completed or rejected */
 792         wb = kzalloc(sizeof(*wb), GFP_KERNEL);
 793         if (!wb)
 794                 return -ENOMEM;
 795         wb->vnode = vnode;
 796         wb->first = 0;
 797         wb->last = -1;
 798         wb->offset_first = 0;
 799         wb->to_last = PAGE_SIZE;
 800         wb->usage = 1;
 801         wb->state = AFS_WBACK_SYNCING;
 802         init_waitqueue_head(&wb->waitq);
 803
 804         spin_lock(&vnode->writeback_lock);
 805         list_for_each_entry(xwb, &vnode->writebacks, link) {
 806                 if (xwb->state == AFS_WBACK_PENDING)
 807                         xwb->state = AFS_WBACK_CONFLICTING;
 808         }
 809         list_add_tail(&wb->link, &vnode->writebacks);
 810         spin_unlock(&vnode->writeback_lock);
 811
 812         /* push all the outstanding writebacks to the server */
 813         ret = afs_writeback_all(vnode);
 814         if (ret < 0) {
 815                 afs_put_writeback(wb);
 816                 _leave(" = %d [wb]", ret);
 817                 return ret;
 818         }
 819
 820         /* wait for the preceding writes to actually complete */
 821         ret = wait_event_interruptible(wb->waitq,
 822                                        wb->state == AFS_WBACK_COMPLETE ||
 823                                        vnode->writebacks.next == &wb->link);
 824         afs_put_writeback(wb);
 825         _leave(" = %d", ret);
 826         return ret;
 827 }