SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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                 SetPageUptodate(page);
 210         }
 211
 212 try_again:
 213         index = page->index;
 214         spin_lock(&vnode->writeback_lock);
 215
 216         /* see if this page is already pending a writeback under a suitable key
 217          * - if so we can just join onto that one */
 218         wb = (struct afs_writeback *) page_private(page);
 219         if (wb) {
 220                 if (wb->key == key && wb->state == AFS_WBACK_PENDING)
 221                         goto subsume_in_current_wb;
 222                 goto flush_conflicting_wb;
 223         }
 224
 225         if (index > 0) {
 226                 /* see if we can find an already pending writeback that we can
 227                  * append this page to */
 228                 list_for_each_entry(wb, &vnode->writebacks, link) {
 229                         if (wb->last == index - 1 && wb->key == key &&
 230                             wb->state == AFS_WBACK_PENDING)
 231                                 goto append_to_previous_wb;
 232                 }
 233         }
 234
 235         list_add_tail(&candidate->link, &vnode->writebacks);
 236         candidate->key = key_get(key);
 237         spin_unlock(&vnode->writeback_lock);
 238         SetPagePrivate(page);
 239         set_page_private(page, (unsigned long) candidate);
 240         _leave(" = 0 [new]");
 241         return 0;
 242
 243 subsume_in_current_wb:
 244         _debug("subsume");
 245         ASSERTRANGE(wb->first, <=, index, <=, wb->last);
 246         if (index == wb->first && offset < wb->offset_first)
 247                 wb->offset_first = offset;
 248         if (index == wb->last && to > wb->to_last)
 249                 wb->to_last = to;
 250         spin_unlock(&vnode->writeback_lock);
 251         kfree(candidate);
 252         _leave(" = 0 [sub]");
 253         return 0;
 254
 255 append_to_previous_wb:
 256         _debug("append into %lx-%lx", wb->first, wb->last);
 257         wb->usage++;
 258         wb->last++;
 259         wb->to_last = to;
 260         spin_unlock(&vnode->writeback_lock);
 261         SetPagePrivate(page);
 262         set_page_private(page, (unsigned long) wb);
 263         kfree(candidate);
 264         _leave(" = 0 [app]");
 265         return 0;
 266
 267         /* the page is currently bound to another context, so if it's dirty we
 268          * need to flush it before we can use the new context */
 269 flush_conflicting_wb:
 270         _debug("flush conflict");
 271         if (wb->state == AFS_WBACK_PENDING)
 272                 wb->state = AFS_WBACK_CONFLICTING;
 273         spin_unlock(&vnode->writeback_lock);
 274         if (PageDirty(page)) {
 275                 ret = afs_write_back_from_locked_page(wb, page);
 276                 if (ret < 0) {
 277                         afs_put_writeback(candidate);
 278                         _leave(" = %d", ret);
 279                         return ret;
 280                 }
 281         }
 282
 283         /* the page holds a ref on the writeback record */
 284         afs_put_writeback(wb);
 285         set_page_private(page, 0);
 286         ClearPagePrivate(page);
 287         goto try_again;
 288 }
 289
 290 /*
 291  * finalise part of a write to a page
 292  */
 293 int afs_commit_write(struct file *file, struct page *page,
 294                      unsigned offset, unsigned to)
 295 {
 296         struct afs_vnode *vnode = AFS_FS_I(file->f_dentry->d_inode);
 297         loff_t i_size, maybe_i_size;
 298
 299         _enter("{%x:%u},{%lx},%u,%u",
 300                vnode->fid.vid, vnode->fid.vnode, page->index, offset, to);
 301
 302         maybe_i_size = (loff_t) page->index << PAGE_SHIFT;
 303         maybe_i_size += to;
 304
 305         i_size = i_size_read(&vnode->vfs_inode);
 306         if (maybe_i_size > i_size) {
 307                 spin_lock(&vnode->writeback_lock);
 308                 i_size = i_size_read(&vnode->vfs_inode);
 309                 if (maybe_i_size > i_size)
 310                         i_size_write(&vnode->vfs_inode, maybe_i_size);
 311                 spin_unlock(&vnode->writeback_lock);
 312         }
 313
 314         set_page_dirty(page);
 315
 316         if (PageDirty(page))
 317                 _debug("dirtied");
 318
 319         return 0;
 320 }
 321
 322 /*
 323  * kill all the pages in the given range
 324  */
 325 static void afs_kill_pages(struct afs_vnode *vnode, bool error,
 326                            pgoff_t first, pgoff_t last)
 327 {
 328         struct pagevec pv;
 329         unsigned count, loop;
 330
 331         _enter("{%x:%u},%lx-%lx",
 332                vnode->fid.vid, vnode->fid.vnode, first, last);
 333
 334         pagevec_init(&pv, 0);
 335
 336         do {
 337                 _debug("kill %lx-%lx", first, last);
 338
 339                 count = last - first + 1;
 340                 if (count > PAGEVEC_SIZE)
 341                         count = PAGEVEC_SIZE;
 342                 pv.nr = find_get_pages_contig(vnode->vfs_inode.i_mapping,
 343                                               first, count, pv.pages);
 344                 ASSERTCMP(pv.nr, ==, count);
 345
 346                 for (loop = 0; loop < count; loop++) {
 347                         ClearPageUptodate(pv.pages[loop]);
 348                         if (error)
 349                                 SetPageError(pv.pages[loop]);
 350                         end_page_writeback(pv.pages[loop]);
 351                 }
 352
 353                 __pagevec_release(&pv);
 354         } while (first < last);
 355
 356         _leave("");
 357 }
 358
 359 /*
 360  * synchronously write back the locked page and any subsequent non-locked dirty
 361  * pages also covered by the same writeback record
 362  */
 363 static int afs_write_back_from_locked_page(struct afs_writeback *wb,
 364                                            struct page *primary_page)
 365 {
 366         struct page *pages[8], *page;
 367         unsigned long count;
 368         unsigned n, offset, to;
 369         pgoff_t start, first, last;
 370         int loop, ret;
 371
 372         _enter(",%lx", primary_page->index);
 373
 374         count = 1;
 375         if (!clear_page_dirty_for_io(primary_page))
 376                 BUG();
 377         if (test_set_page_writeback(primary_page))
 378                 BUG();
 379
 380         /* find all consecutive lockable dirty pages, stopping when we find a
 381          * page that is not immediately lockable, is not dirty or is missing,
 382          * or we reach the end of the range */
 383         start = primary_page->index;
 384         if (start >= wb->last)
 385                 goto no_more;
 386         start++;
 387         do {
 388                 _debug("more %lx [%lx]", start, count);
 389                 n = wb->last - start + 1;
 390                 if (n > ARRAY_SIZE(pages))
 391                         n = ARRAY_SIZE(pages);
 392                 n = find_get_pages_contig(wb->vnode->vfs_inode.i_mapping,
 393                                           start, n, pages);
 394                 _debug("fgpc %u", n);
 395                 if (n == 0)
 396                         goto no_more;
 397                 if (pages[0]->index != start) {
 398                         do {
 399                                 put_page(pages[--n]);
 400                         } while (n > 0);
 401                         goto no_more;
 402                 }
 403
 404                 for (loop = 0; loop < n; loop++) {
 405                         page = pages[loop];
 406                         if (page->index > wb->last)
 407                                 break;
 408                         if (TestSetPageLocked(page))
 409                                 break;
 410                         if (!PageDirty(page) ||
 411                             page_private(page) != (unsigned long) wb) {
 412                                 unlock_page(page);
 413                                 break;
 414                         }
 415                         if (!clear_page_dirty_for_io(page))
 416                                 BUG();
 417                         if (test_set_page_writeback(page))
 418                                 BUG();
 419                         unlock_page(page);
 420                         put_page(page);
 421                 }
 422                 count += loop;
 423                 if (loop < n) {
 424                         for (; loop < n; loop++)
 425                                 put_page(pages[loop]);
 426                         goto no_more;
 427                 }
 428
 429                 start += loop;
 430         } while (start <= wb->last && count < 65536);
 431
 432 no_more:
 433         /* we now have a contiguous set of dirty pages, each with writeback set
 434          * and the dirty mark cleared; the first page is locked and must remain
 435          * so, all the rest are unlocked */
 436         first = primary_page->index;
 437         last = first + count - 1;
 438
 439         offset = (first == wb->first) ? wb->offset_first : 0;
 440         to = (last == wb->last) ? wb->to_last : PAGE_SIZE;
 441
 442         _debug("write back %lx[%u..] to %lx[..%u]", first, offset, last, to);
 443
 444         ret = afs_vnode_store_data(wb, first, last, offset, to);
 445         if (ret < 0) {
 446                 switch (ret) {
 447                 case -EDQUOT:
 448                 case -ENOSPC:
 449                         set_bit(AS_ENOSPC,
 450                                 &wb->vnode->vfs_inode.i_mapping->flags);
 451                         break;
 452                 case -EROFS:
 453                 case -EIO:
 454                 case -EREMOTEIO:
 455                 case -EFBIG:
 456                 case -ENOENT:
 457                 case -ENOMEDIUM:
 458                 case -ENXIO:
 459                         afs_kill_pages(wb->vnode, true, first, last);
 460                         set_bit(AS_EIO, &wb->vnode->vfs_inode.i_mapping->flags);
 461                         break;
 462                 case -EACCES:
 463                 case -EPERM:
 464                 case -ENOKEY:
 465                 case -EKEYEXPIRED:
 466                 case -EKEYREJECTED:
 467                 case -EKEYREVOKED:
 468                         afs_kill_pages(wb->vnode, false, first, last);
 469                         break;
 470                 default:
 471                         break;
 472                 }
 473         } else {
 474                 ret = count;
 475         }
 476
 477         _leave(" = %d", ret);
 478         return ret;
 479 }
 480
 481 /*
 482  * write a page back to the server
 483  * - the caller locked the page for us
 484  */
 485 int afs_writepage(struct page *page, struct writeback_control *wbc)
 486 {
 487         struct backing_dev_info *bdi = page->mapping->backing_dev_info;
 488         struct afs_writeback *wb;
 489         int ret;
 490
 491         _enter("{%lx},", page->index);
 492
 493         wb = (struct afs_writeback *) page_private(page);
 494         ASSERT(wb != NULL);
 495
 496         ret = afs_write_back_from_locked_page(wb, page);
 497         unlock_page(page);
 498         if (ret < 0) {
 499                 _leave(" = %d", ret);
 500                 return 0;
 501         }
 502
 503         wbc->nr_to_write -= ret;
 504         if (wbc->nonblocking && bdi_write_congested(bdi))
 505                 wbc->encountered_congestion = 1;
 506
 507         _leave(" = 0");
 508         return 0;
 509 }
 510
 511 /*
 512  * write a region of pages back to the server
 513  */
 514 int afs_writepages_region(struct address_space *mapping,
 515                           struct writeback_control *wbc,
 516                           pgoff_t index, pgoff_t end, pgoff_t *_next)
 517 {
 518         struct backing_dev_info *bdi = mapping->backing_dev_info;
 519         struct afs_writeback *wb;
 520         struct page *page;
 521         int ret, n;
 522
 523         _enter(",,%lx,%lx,", index, end);
 524
 525         do {
 526                 n = find_get_pages_tag(mapping, &index, PAGECACHE_TAG_DIRTY,
 527                                        1, &page);
 528                 if (!n)
 529                         break;
 530
 531                 _debug("wback %lx", page->index);
 532
 533                 if (page->index > end) {
 534                         *_next = index;
 535                         page_cache_release(page);
 536                         _leave(" = 0 [%lx]", *_next);
 537                         return 0;
 538                 }
 539
 540                 /* at this point we hold neither mapping->tree_lock nor lock on
 541                  * the page itself: the page may be truncated or invalidated
 542                  * (changing page->mapping to NULL), or even swizzled back from
 543                  * swapper_space to tmpfs file mapping
 544                  */
 545                 lock_page(page);
 546
 547                 if (page->mapping != mapping) {
 548                         unlock_page(page);
 549                         page_cache_release(page);
 550                         continue;
 551                 }
 552
 553                 if (wbc->sync_mode != WB_SYNC_NONE)
 554                         wait_on_page_writeback(page);
 555
 556                 if (PageWriteback(page) || !PageDirty(page)) {
 557                         unlock_page(page);
 558                         continue;
 559                 }
 560
 561                 wb = (struct afs_writeback *) page_private(page);
 562                 ASSERT(wb != NULL);
 563
 564                 spin_lock(&wb->vnode->writeback_lock);
 565                 wb->state = AFS_WBACK_WRITING;
 566                 spin_unlock(&wb->vnode->writeback_lock);
 567
 568                 ret = afs_write_back_from_locked_page(wb, page);
 569                 unlock_page(page);
 570                 page_cache_release(page);
 571                 if (ret < 0) {
 572                         _leave(" = %d", ret);
 573                         return ret;
 574                 }
 575
 576                 wbc->nr_to_write -= ret;
 577
 578                 if (wbc->nonblocking && bdi_write_congested(bdi)) {
 579                         wbc->encountered_congestion = 1;
 580                         break;
 581                 }
 582
 583                 cond_resched();
 584         } while (index < end && wbc->nr_to_write > 0);
 585
 586         *_next = index;
 587         _leave(" = 0 [%lx]", *_next);
 588         return 0;
 589 }
 590
 591 /*
 592  * write some of the pending data back to the server
 593  */
 594 int afs_writepages(struct address_space *mapping,
 595                    struct writeback_control *wbc)
 596 {
 597         struct backing_dev_info *bdi = mapping->backing_dev_info;
 598         pgoff_t start, end, next;
 599         int ret;
 600
 601         _enter("");
 602
 603         if (wbc->nonblocking && bdi_write_congested(bdi)) {
 604                 wbc->encountered_congestion = 1;
 605                 _leave(" = 0 [congest]");
 606                 return 0;
 607         }
 608
 609         if (wbc->range_cyclic) {
 610                 start = mapping->writeback_index;
 611                 end = -1;
 612                 ret = afs_writepages_region(mapping, wbc, start, end, &next);
 613                 if (start > 0 && wbc->nr_to_write > 0 && ret == 0 &&
 614                     !(wbc->nonblocking && wbc->encountered_congestion))
 615                         ret = afs_writepages_region(mapping, wbc, 0, start,
 616                                                     &next);
 617                 mapping->writeback_index = next;
 618         } else if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) {
 619                 end = (pgoff_t)(LLONG_MAX >> PAGE_CACHE_SHIFT);
 620                 ret = afs_writepages_region(mapping, wbc, 0, end, &next);
 621                 if (wbc->nr_to_write > 0)
 622                         mapping->writeback_index = next;
 623         } else {
 624                 start = wbc->range_start >> PAGE_CACHE_SHIFT;
 625                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
 626                 ret = afs_writepages_region(mapping, wbc, start, end, &next);
 627         }
 628
 629         _leave(" = %d", ret);
 630         return ret;
 631 }
 632
 633 /*
 634  * write an inode back
 635  */
 636 int afs_write_inode(struct inode *inode, int sync)
 637 {
 638         struct afs_vnode *vnode = AFS_FS_I(inode);
 639         int ret;
 640
 641         _enter("{%x:%u},", vnode->fid.vid, vnode->fid.vnode);
 642
 643         ret = 0;
 644         if (sync) {
 645                 ret = filemap_fdatawait(inode->i_mapping);
 646                 if (ret < 0)
 647                         __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
 648         }
 649
 650         _leave(" = %d", ret);
 651         return ret;
 652 }
 653
 654 /*
 655  * completion of write to server
 656  */
 657 void afs_pages_written_back(struct afs_vnode *vnode, struct afs_call *call)
 658 {
 659         struct afs_writeback *wb = call->wb;
 660         struct pagevec pv;
 661         unsigned count, loop;
 662         pgoff_t first = call->first, last = call->last;
 663         bool free_wb;
 664
 665         _enter("{%x:%u},{%lx-%lx}",
 666                vnode->fid.vid, vnode->fid.vnode, first, last);
 667
 668         ASSERT(wb != NULL);
 669
 670         pagevec_init(&pv, 0);
 671
 672         do {
 673                 _debug("done %lx-%lx", first, last);
 674
 675                 count = last - first + 1;
 676                 if (count > PAGEVEC_SIZE)
 677                         count = PAGEVEC_SIZE;
 678                 pv.nr = find_get_pages_contig(call->mapping, first, count,
 679                                               pv.pages);
 680                 ASSERTCMP(pv.nr, ==, count);
 681
 682                 spin_lock(&vnode->writeback_lock);
 683                 for (loop = 0; loop < count; loop++) {
 684                         struct page *page = pv.pages[loop];
 685                         end_page_writeback(page);
 686                         if (page_private(page) == (unsigned long) wb) {
 687                                 set_page_private(page, 0);
 688                                 ClearPagePrivate(page);
 689                                 wb->usage--;
 690                         }
 691                 }
 692                 free_wb = false;
 693                 if (wb->usage == 0) {
 694                         afs_unlink_writeback(wb);
 695                         free_wb = true;
 696                 }
 697                 spin_unlock(&vnode->writeback_lock);
 698                 first += count;
 699                 if (free_wb) {
 700                         afs_free_writeback(wb);
 701                         wb = NULL;
 702                 }
 703
 704                 __pagevec_release(&pv);
 705         } while (first <= last);
 706
 707         _leave("");
 708 }
 709
 710 /*
 711  * write to an AFS file
 712  */
 713 ssize_t afs_file_write(struct kiocb *iocb, const struct iovec *iov,
 714                        unsigned long nr_segs, loff_t pos)
 715 {
 716         struct dentry *dentry = iocb->ki_filp->f_path.dentry;
 717         struct afs_vnode *vnode = AFS_FS_I(dentry->d_inode);
 718         ssize_t result;
 719         size_t count = iov_length(iov, nr_segs);
 720         int ret;
 721
 722         _enter("{%x.%u},{%zu},%lu,",
 723                vnode->fid.vid, vnode->fid.vnode, count, nr_segs);
 724
 725         if (IS_SWAPFILE(&vnode->vfs_inode)) {
 726                 printk(KERN_INFO
 727                        "AFS: Attempt to write to active swap file!\n");
 728                 return -EBUSY;
 729         }
 730
 731         if (!count)
 732                 return 0;
 733
 734         result = generic_file_aio_write(iocb, iov, nr_segs, pos);
 735         if (IS_ERR_VALUE(result)) {
 736                 _leave(" = %zd", result);
 737                 return result;
 738         }
 739
 740         /* return error values for O_SYNC and IS_SYNC() */
 741         if (IS_SYNC(&vnode->vfs_inode) || iocb->ki_filp->f_flags & O_SYNC) {
 742                 ret = afs_fsync(iocb->ki_filp, dentry, 1);
 743                 if (ret < 0)
 744                         result = ret;
 745         }
 746
 747         _leave(" = %zd", result);
 748         return result;
 749 }
 750
 751 /*
 752  * flush the vnode to the fileserver
 753  */
 754 int afs_writeback_all(struct afs_vnode *vnode)
 755 {
 756         struct address_space *mapping = vnode->vfs_inode.i_mapping;
 757         struct writeback_control wbc = {
 758                 .bdi            = mapping->backing_dev_info,
 759                 .sync_mode      = WB_SYNC_ALL,
 760                 .nr_to_write    = LONG_MAX,
 761                 .for_writepages = 1,
 762                 .range_cyclic   = 1,
 763         };
 764         int ret;
 765
 766         _enter("");
 767
 768         ret = mapping->a_ops->writepages(mapping, &wbc);
 769         __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
 770
 771         _leave(" = %d", ret);
 772         return ret;
 773 }
 774
 775 /*
 776  * flush any dirty pages for this process, and check for write errors.
 777  * - the return status from this call provides a reliable indication of
 778  *   whether any write errors occurred for this process.
 779  */
 780 int afs_fsync(struct file *file, struct dentry *dentry, int datasync)
 781 {
 782         struct afs_writeback *wb, *xwb;
 783         struct afs_vnode *vnode = AFS_FS_I(dentry->d_inode);
 784         int ret;
 785
 786         _enter("{%x:%u},{n=%s},%d",
 787                vnode->fid.vid, vnode->fid.vnode, dentry->d_name.name,
 788                datasync);
 789
 790         /* use a writeback record as a marker in the queue - when this reaches
 791          * the front of the queue, all the outstanding writes are either
 792          * completed or rejected */
 793         wb = kzalloc(sizeof(*wb), GFP_KERNEL);
 794         if (!wb)
 795                 return -ENOMEM;
 796         wb->vnode = vnode;
 797         wb->first = 0;
 798         wb->last = -1;
 799         wb->offset_first = 0;
 800         wb->to_last = PAGE_SIZE;
 801         wb->usage = 1;
 802         wb->state = AFS_WBACK_SYNCING;
 803         init_waitqueue_head(&wb->waitq);
 804
 805         spin_lock(&vnode->writeback_lock);
 806         list_for_each_entry(xwb, &vnode->writebacks, link) {
 807                 if (xwb->state == AFS_WBACK_PENDING)
 808                         xwb->state = AFS_WBACK_CONFLICTING;
 809         }
 810         list_add_tail(&wb->link, &vnode->writebacks);
 811         spin_unlock(&vnode->writeback_lock);
 812
 813         /* push all the outstanding writebacks to the server */
 814         ret = afs_writeback_all(vnode);
 815         if (ret < 0) {
 816                 afs_put_writeback(wb);
 817                 _leave(" = %d [wb]", ret);
 818                 return ret;
 819         }
 820
 821         /* wait for the preceding writes to actually complete */
 822         ret = wait_event_interruptible(wb->waitq,
 823                                        wb->state == AFS_WBACK_COMPLETE ||
 824                                        vnode->writebacks.next == &wb->link);
 825         afs_put_writeback(wb);
 826         _leave(" = %d", ret);
 827         return ret;
 828 }