fs/gfs2/aops.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   4  * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
   5  */
   6
   7 #include <linux/sched.h>
   8 #include <linux/slab.h>
   9 #include <linux/spinlock.h>
  10 #include <linux/completion.h>
  11 #include <linux/buffer_head.h>
  12 #include <linux/pagemap.h>
  13 #include <linux/pagevec.h>
  14 #include <linux/mpage.h>
  15 #include <linux/fs.h>
  16 #include <linux/writeback.h>
  17 #include <linux/swap.h>
  18 #include <linux/gfs2_ondisk.h>
  19 #include <linux/backing-dev.h>
  20 #include <linux/uio.h>
  21 #include <trace/events/writeback.h>
  22 #include <linux/sched/signal.h>
  23
  24 #include "gfs2.h"
  25 #include "incore.h"
  26 #include "bmap.h"
  27 #include "glock.h"
  28 #include "inode.h"
  29 #include "log.h"
  30 #include "meta_io.h"
  31 #include "quota.h"
  32 #include "trans.h"
  33 #include "rgrp.h"
  34 #include "super.h"
  35 #include "util.h"
  36 #include "glops.h"
  37 #include "aops.h"
  38
  39
  40 void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page,
  41                             unsigned int from, unsigned int len)
  42 {
  43         struct buffer_head *head = page_buffers(page);
  44         unsigned int bsize = head->b_size;
  45         struct buffer_head *bh;
  46         unsigned int to = from + len;
  47         unsigned int start, end;
  48
  49         for (bh = head, start = 0; bh != head || !start;
  50              bh = bh->b_this_page, start = end) {
  51                 end = start + bsize;
  52                 if (end <= from)
  53                         continue;
  54                 if (start >= to)
  55                         break;
  56                 set_buffer_uptodate(bh);
  57                 gfs2_trans_add_data(ip->i_gl, bh);
  58         }
  59 }
  60
  61 /**
  62  * gfs2_get_block_noalloc - Fills in a buffer head with details about a block
  63  * @inode: The inode
  64  * @lblock: The block number to look up
  65  * @bh_result: The buffer head to return the result in
  66  * @create: Non-zero if we may add block to the file
  67  *
  68  * Returns: errno
  69  */
  70
  71 static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock,
  72                                   struct buffer_head *bh_result, int create)
  73 {
  74         int error;
  75
  76         error = gfs2_block_map(inode, lblock, bh_result, 0);
  77         if (error)
  78                 return error;
  79         if (!buffer_mapped(bh_result))
  80                 return -EIO;
  81         return 0;
  82 }
  83
  84 /**
  85  * gfs2_writepage - Write page for writeback mappings
  86  * @page: The page
  87  * @wbc: The writeback control
  88  */
  89 static int gfs2_writepage(struct page *page, struct writeback_control *wbc)
  90 {
  91         struct inode *inode = page->mapping->host;
  92         struct gfs2_inode *ip = GFS2_I(inode);
  93         struct gfs2_sbd *sdp = GFS2_SB(inode);
  94         loff_t i_size = i_size_read(inode);
  95         pgoff_t end_index = i_size >> PAGE_SHIFT;
  96         unsigned offset;
  97
  98         if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl)))
  99                 goto out;
 100         if (current->journal_info)
 101                 goto redirty;
 102         /* Is the page fully outside i_size? (truncate in progress) */
 103         offset = i_size & (PAGE_SIZE-1);
 104         if (page->index > end_index || (page->index == end_index && !offset)) {
 105                 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
 106                 goto out;
 107         }
 108
 109         return nobh_writepage(page, gfs2_get_block_noalloc, wbc);
 110
 111 redirty:
 112         redirty_page_for_writepage(wbc, page);
 113 out:
 114         unlock_page(page);
 115         return 0;
 116 }
 117
 118 /* This is the same as calling block_write_full_page, but it also
 119  * writes pages outside of i_size
 120  */
 121 static int gfs2_write_full_page(struct page *page, get_block_t *get_block,
 122                                 struct writeback_control *wbc)
 123 {
 124         struct inode * const inode = page->mapping->host;
 125         loff_t i_size = i_size_read(inode);
 126         const pgoff_t end_index = i_size >> PAGE_SHIFT;
 127         unsigned offset;
 128
 129         /*
 130          * The page straddles i_size.  It must be zeroed out on each and every
 131          * writepage invocation because it may be mmapped.  "A file is mapped
 132          * in multiples of the page size.  For a file that is not a multiple of
 133          * the  page size, the remaining memory is zeroed when mapped, and
 134          * writes to that region are not written out to the file."
 135          */
 136         offset = i_size & (PAGE_SIZE-1);
 137         if (page->index == end_index && offset)
 138                 zero_user_segment(page, offset, PAGE_SIZE);
 139
 140         return __block_write_full_page(inode, page, get_block, wbc,
 141                                        end_buffer_async_write);
 142 }
 143
 144 /**
 145  * __gfs2_jdata_writepage - The core of jdata writepage
 146  * @page: The page to write
 147  * @wbc: The writeback control
 148  *
 149  * This is shared between writepage and writepages and implements the
 150  * core of the writepage operation. If a transaction is required then
 151  * PageChecked will have been set and the transaction will have
 152  * already been started before this is called.
 153  */
 154
 155 static int __gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
 156 {
 157         struct inode *inode = page->mapping->host;
 158         struct gfs2_inode *ip = GFS2_I(inode);
 159
 160         if (PageChecked(page)) {
 161                 ClearPageChecked(page);
 162                 if (!page_has_buffers(page)) {
 163                         create_empty_buffers(page, inode->i_sb->s_blocksize,
 164                                              BIT(BH_Dirty)|BIT(BH_Uptodate));
 165                 }
 166                 gfs2_page_add_databufs(ip, page, 0, PAGE_SIZE);
 167         }
 168         return gfs2_write_full_page(page, gfs2_get_block_noalloc, wbc);
 169 }
 170
 171 /**
 172  * gfs2_jdata_writepage - Write complete page
 173  * @page: Page to write
 174  * @wbc: The writeback control
 175  *
 176  * Returns: errno
 177  *
 178  */
 179
 180 static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
 181 {
 182         struct inode *inode = page->mapping->host;
 183         struct gfs2_inode *ip = GFS2_I(inode);
 184         struct gfs2_sbd *sdp = GFS2_SB(inode);
 185         int ret;
 186
 187         if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl)))
 188                 goto out;
 189         if (PageChecked(page) || current->journal_info)
 190                 goto out_ignore;
 191         ret = __gfs2_jdata_writepage(page, wbc);
 192         return ret;
 193
 194 out_ignore:
 195         redirty_page_for_writepage(wbc, page);
 196 out:
 197         unlock_page(page);
 198         return 0;
 199 }
 200
 201 /**
 202  * gfs2_writepages - Write a bunch of dirty pages back to disk
 203  * @mapping: The mapping to write
 204  * @wbc: Write-back control
 205  *
 206  * Used for both ordered and writeback modes.
 207  */
 208 static int gfs2_writepages(struct address_space *mapping,
 209                            struct writeback_control *wbc)
 210 {
 211         struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
 212         int ret = mpage_writepages(mapping, wbc, gfs2_get_block_noalloc);
 213
 214         /*
 215          * Even if we didn't write any pages here, we might still be holding
 216          * dirty pages in the ail. We forcibly flush the ail because we don't
 217          * want balance_dirty_pages() to loop indefinitely trying to write out
 218          * pages held in the ail that it can't find.
 219          */
 220         if (ret == 0)
 221                 set_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags);
 222
 223         return ret;
 224 }
 225
 226 /**
 227  * gfs2_write_jdata_pagevec - Write back a pagevec's worth of pages
 228  * @mapping: The mapping
 229  * @wbc: The writeback control
 230  * @pvec: The vector of pages
 231  * @nr_pages: The number of pages to write
 232  * @done_index: Page index
 233  *
 234  * Returns: non-zero if loop should terminate, zero otherwise
 235  */
 236
 237 static int gfs2_write_jdata_pagevec(struct address_space *mapping,
 238                                     struct writeback_control *wbc,
 239                                     struct pagevec *pvec,
 240                                     int nr_pages,
 241                                     pgoff_t *done_index)
 242 {
 243         struct inode *inode = mapping->host;
 244         struct gfs2_sbd *sdp = GFS2_SB(inode);
 245         unsigned nrblocks = nr_pages * (PAGE_SIZE >> inode->i_blkbits);
 246         int i;
 247         int ret;
 248
 249         ret = gfs2_trans_begin(sdp, nrblocks, nrblocks);
 250         if (ret < 0)
 251                 return ret;
 252
 253         for(i = 0; i < nr_pages; i++) {
 254                 struct page *page = pvec->pages[i];
 255
 256                 *done_index = page->index;
 257
 258                 lock_page(page);
 259
 260                 if (unlikely(page->mapping != mapping)) {
 261 continue_unlock:
 262                         unlock_page(page);
 263                         continue;
 264                 }
 265
 266                 if (!PageDirty(page)) {
 267                         /* someone wrote it for us */
 268                         goto continue_unlock;
 269                 }
 270
 271                 if (PageWriteback(page)) {
 272                         if (wbc->sync_mode != WB_SYNC_NONE)
 273                                 wait_on_page_writeback(page);
 274                         else
 275                                 goto continue_unlock;
 276                 }
 277
 278                 BUG_ON(PageWriteback(page));
 279                 if (!clear_page_dirty_for_io(page))
 280                         goto continue_unlock;
 281
 282                 trace_wbc_writepage(wbc, inode_to_bdi(inode));
 283
 284                 ret = __gfs2_jdata_writepage(page, wbc);
 285                 if (unlikely(ret)) {
 286                         if (ret == AOP_WRITEPAGE_ACTIVATE) {
 287                                 unlock_page(page);
 288                                 ret = 0;
 289                         } else {
 290
 291                                 /*
 292                                  * done_index is set past this page,
 293                                  * so media errors will not choke
 294                                  * background writeout for the entire
 295                                  * file. This has consequences for
 296                                  * range_cyclic semantics (ie. it may
 297                                  * not be suitable for data integrity
 298                                  * writeout).
 299                                  */
 300                                 *done_index = page->index + 1;
 301                                 ret = 1;
 302                                 break;
 303                         }
 304                 }
 305
 306                 /*
 307                  * We stop writing back only if we are not doing
 308                  * integrity sync. In case of integrity sync we have to
 309                  * keep going until we have written all the pages
 310                  * we tagged for writeback prior to entering this loop.
 311                  */
 312                 if (--wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) {
 313                         ret = 1;
 314                         break;
 315                 }
 316
 317         }
 318         gfs2_trans_end(sdp);
 319         return ret;
 320 }
 321
 322 /**
 323  * gfs2_write_cache_jdata - Like write_cache_pages but different
 324  * @mapping: The mapping to write
 325  * @wbc: The writeback control
 326  *
 327  * The reason that we use our own function here is that we need to
 328  * start transactions before we grab page locks. This allows us
 329  * to get the ordering right.
 330  */
 331
 332 static int gfs2_write_cache_jdata(struct address_space *mapping,
 333                                   struct writeback_control *wbc)
 334 {
 335         int ret = 0;
 336         int done = 0;
 337         struct pagevec pvec;
 338         int nr_pages;
 339         pgoff_t writeback_index;
 340         pgoff_t index;
 341         pgoff_t end;
 342         pgoff_t done_index;
 343         int cycled;
 344         int range_whole = 0;
 345         xa_mark_t tag;
 346
 347         pagevec_init(&pvec);
 348         if (wbc->range_cyclic) {
 349                 writeback_index = mapping->writeback_index; /* prev offset */
 350                 index = writeback_index;
 351                 if (index == 0)
 352                         cycled = 1;
 353                 else
 354                         cycled = 0;
 355                 end = -1;
 356         } else {
 357                 index = wbc->range_start >> PAGE_SHIFT;
 358                 end = wbc->range_end >> PAGE_SHIFT;
 359                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
 360                         range_whole = 1;
 361                 cycled = 1; /* ignore range_cyclic tests */
 362         }
 363         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
 364                 tag = PAGECACHE_TAG_TOWRITE;
 365         else
 366                 tag = PAGECACHE_TAG_DIRTY;
 367
 368 retry:
 369         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
 370                 tag_pages_for_writeback(mapping, index, end);
 371         done_index = index;
 372         while (!done && (index <= end)) {
 373                 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
 374                                 tag);
 375                 if (nr_pages == 0)
 376                         break;
 377
 378                 ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, &done_index);
 379                 if (ret)
 380                         done = 1;
 381                 if (ret > 0)
 382                         ret = 0;
 383                 pagevec_release(&pvec);
 384                 cond_resched();
 385         }
 386
 387         if (!cycled && !done) {
 388                 /*
 389                  * range_cyclic:
 390                  * We hit the last page and there is more work to be done: wrap
 391                  * back to the start of the file
 392                  */
 393                 cycled = 1;
 394                 index = 0;
 395                 end = writeback_index - 1;
 396                 goto retry;
 397         }
 398
 399         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
 400                 mapping->writeback_index = done_index;
 401
 402         return ret;
 403 }
 404
 405
 406 /**
 407  * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
 408  * @mapping: The mapping to write
 409  * @wbc: The writeback control
 410  *
 411  */
 412
 413 static int gfs2_jdata_writepages(struct address_space *mapping,
 414                                  struct writeback_control *wbc)
 415 {
 416         struct gfs2_inode *ip = GFS2_I(mapping->host);
 417         struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
 418         int ret;
 419
 420         ret = gfs2_write_cache_jdata(mapping, wbc);
 421         if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) {
 422                 gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
 423                                GFS2_LFC_JDATA_WPAGES);
 424                 ret = gfs2_write_cache_jdata(mapping, wbc);
 425         }
 426         return ret;
 427 }
 428
 429 /**
 430  * stuffed_readpage - Fill in a Linux page with stuffed file data
 431  * @ip: the inode
 432  * @page: the page
 433  *
 434  * Returns: errno
 435  */
 436 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
 437 {
 438         struct buffer_head *dibh;
 439         u64 dsize = i_size_read(&ip->i_inode);
 440         void *kaddr;
 441         int error;
 442
 443         /*
 444          * Due to the order of unstuffing files and ->fault(), we can be
 445          * asked for a zero page in the case of a stuffed file being extended,
 446          * so we need to supply one here. It doesn't happen often.
 447          */
 448         if (unlikely(page->index)) {
 449                 zero_user(page, 0, PAGE_SIZE);
 450                 SetPageUptodate(page);
 451                 return 0;
 452         }
 453
 454         error = gfs2_meta_inode_buffer(ip, &dibh);
 455         if (error)
 456                 return error;
 457
 458         kaddr = kmap_atomic(page);
 459         memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
 460         memset(kaddr + dsize, 0, PAGE_SIZE - dsize);
 461         kunmap_atomic(kaddr);
 462         flush_dcache_page(page);
 463         brelse(dibh);
 464         SetPageUptodate(page);
 465
 466         return 0;
 467 }
 468
 469
 470 /**
 471  * __gfs2_readpage - readpage
 472  * @file: The file to read a page for
 473  * @page: The page to read
 474  *
 475  * This is the core of gfs2's readpage. It's used by the internal file
 476  * reading code as in that case we already hold the glock. Also it's
 477  * called by gfs2_readpage() once the required lock has been granted.
 478  */
 479
 480 static int __gfs2_readpage(void *file, struct page *page)
 481 {
 482         struct gfs2_inode *ip = GFS2_I(page->mapping->host);
 483         struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
 484
 485         int error;
 486
 487         if (i_blocksize(page->mapping->host) == PAGE_SIZE &&
 488             !page_has_buffers(page)) {
 489                 error = iomap_readpage(page, &gfs2_iomap_ops);
 490         } else if (gfs2_is_stuffed(ip)) {
 491                 error = stuffed_readpage(ip, page);
 492                 unlock_page(page);
 493         } else {
 494                 error = mpage_readpage(page, gfs2_block_map);
 495         }
 496
 497         if (unlikely(test_bit(SDF_WITHDRAWN, &sdp->sd_flags)))
 498                 return -EIO;
 499
 500         return error;
 501 }
 502
 503 /**
 504  * gfs2_readpage - read a page of a file
 505  * @file: The file to read
 506  * @page: The page of the file
 507  *
 508  * This deals with the locking required. We have to unlock and
 509  * relock the page in order to get the locking in the right
 510  * order.
 511  */
 512
 513 static int gfs2_readpage(struct file *file, struct page *page)
 514 {
 515         struct address_space *mapping = page->mapping;
 516         struct gfs2_inode *ip = GFS2_I(mapping->host);
 517         struct gfs2_holder gh;
 518         int error;
 519
 520         unlock_page(page);
 521         gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
 522         error = gfs2_glock_nq(&gh);
 523         if (unlikely(error))
 524                 goto out;
 525         error = AOP_TRUNCATED_PAGE;
 526         lock_page(page);
 527         if (page->mapping == mapping && !PageUptodate(page))
 528                 error = __gfs2_readpage(file, page);
 529         else
 530                 unlock_page(page);
 531         gfs2_glock_dq(&gh);
 532 out:
 533         gfs2_holder_uninit(&gh);
 534         if (error && error != AOP_TRUNCATED_PAGE)
 535                 lock_page(page);
 536         return error;
 537 }
 538
 539 /**
 540  * gfs2_internal_read - read an internal file
 541  * @ip: The gfs2 inode
 542  * @buf: The buffer to fill
 543  * @pos: The file position
 544  * @size: The amount to read
 545  *
 546  */
 547
 548 int gfs2_internal_read(struct gfs2_inode *ip, char *buf, loff_t *pos,
 549                        unsigned size)
 550 {
 551         struct address_space *mapping = ip->i_inode.i_mapping;
 552         unsigned long index = *pos >> PAGE_SHIFT;
 553         unsigned offset = *pos & (PAGE_SIZE - 1);
 554         unsigned copied = 0;
 555         unsigned amt;
 556         struct page *page;
 557         void *p;
 558
 559         do {
 560                 amt = size - copied;
 561                 if (offset + size > PAGE_SIZE)
 562                         amt = PAGE_SIZE - offset;
 563                 page = read_cache_page(mapping, index, __gfs2_readpage, NULL);
 564                 if (IS_ERR(page))
 565                         return PTR_ERR(page);
 566                 p = kmap_atomic(page);
 567                 memcpy(buf + copied, p + offset, amt);
 568                 kunmap_atomic(p);
 569                 put_page(page);
 570                 copied += amt;
 571                 index++;
 572                 offset = 0;
 573         } while(copied < size);
 574         (*pos) += size;
 575         return size;
 576 }
 577
 578 /**
 579  * gfs2_readpages - Read a bunch of pages at once
 580  * @file: The file to read from
 581  * @mapping: Address space info
 582  * @pages: List of pages to read
 583  * @nr_pages: Number of pages to read
 584  *
 585  * Some notes:
 586  * 1. This is only for readahead, so we can simply ignore any things
 587  *    which are slightly inconvenient (such as locking conflicts between
 588  *    the page lock and the glock) and return having done no I/O. Its
 589  *    obviously not something we'd want to do on too regular a basis.
 590  *    Any I/O we ignore at this time will be done via readpage later.
 591  * 2. We don't handle stuffed files here we let readpage do the honours.
 592  * 3. mpage_readpages() does most of the heavy lifting in the common case.
 593  * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
 594  */
 595
 596 static int gfs2_readpages(struct file *file, struct address_space *mapping,
 597                           struct list_head *pages, unsigned nr_pages)
 598 {
 599         struct inode *inode = mapping->host;
 600         struct gfs2_inode *ip = GFS2_I(inode);
 601         struct gfs2_sbd *sdp = GFS2_SB(inode);
 602         struct gfs2_holder gh;
 603         int ret;
 604
 605         gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
 606         ret = gfs2_glock_nq(&gh);
 607         if (unlikely(ret))
 608                 goto out_uninit;
 609         if (!gfs2_is_stuffed(ip))
 610                 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_block_map);
 611         gfs2_glock_dq(&gh);
 612 out_uninit:
 613         gfs2_holder_uninit(&gh);
 614         if (unlikely(test_bit(SDF_WITHDRAWN, &sdp->sd_flags)))
 615                 ret = -EIO;
 616         return ret;
 617 }
 618
 619 /**
 620  * adjust_fs_space - Adjusts the free space available due to gfs2_grow
 621  * @inode: the rindex inode
 622  */
 623 void adjust_fs_space(struct inode *inode)
 624 {
 625         struct gfs2_sbd *sdp = GFS2_SB(inode);
 626         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 627         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 628         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 629         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 630         struct buffer_head *m_bh, *l_bh;
 631         u64 fs_total, new_free;
 632
 633         if (gfs2_trans_begin(sdp, 2 * RES_STATFS, 0) != 0)
 634                 return;
 635
 636         /* Total up the file system space, according to the latest rindex. */
 637         fs_total = gfs2_ri_total(sdp);
 638         if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0)
 639                 goto out;
 640
 641         spin_lock(&sdp->sd_statfs_spin);
 642         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 643                               sizeof(struct gfs2_dinode));
 644         if (fs_total > (m_sc->sc_total + l_sc->sc_total))
 645                 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
 646         else
 647                 new_free = 0;
 648         spin_unlock(&sdp->sd_statfs_spin);
 649         fs_warn(sdp, "File system extended by %llu blocks.\n",
 650                 (unsigned long long)new_free);
 651         gfs2_statfs_change(sdp, new_free, new_free, 0);
 652
 653         if (gfs2_meta_inode_buffer(l_ip, &l_bh) != 0)
 654                 goto out2;
 655         update_statfs(sdp, m_bh, l_bh);
 656         brelse(l_bh);
 657 out2:
 658         brelse(m_bh);
 659 out:
 660         sdp->sd_rindex_uptodate = 0;
 661         gfs2_trans_end(sdp);
 662 }
 663
 664 /**
 665  * jdata_set_page_dirty - Page dirtying function
 666  * @page: The page to dirty
 667  *
 668  * Returns: 1 if it dirtyed the page, or 0 otherwise
 669  */
 670
 671 static int jdata_set_page_dirty(struct page *page)
 672 {
 673         SetPageChecked(page);
 674         return __set_page_dirty_buffers(page);
 675 }
 676
 677 /**
 678  * gfs2_bmap - Block map function
 679  * @mapping: Address space info
 680  * @lblock: The block to map
 681  *
 682  * Returns: The disk address for the block or 0 on hole or error
 683  */
 684
 685 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
 686 {
 687         struct gfs2_inode *ip = GFS2_I(mapping->host);
 688         struct gfs2_holder i_gh;
 689         sector_t dblock = 0;
 690         int error;
 691
 692         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
 693         if (error)
 694                 return 0;
 695
 696         if (!gfs2_is_stuffed(ip))
 697                 dblock = iomap_bmap(mapping, lblock, &gfs2_iomap_ops);
 698
 699         gfs2_glock_dq_uninit(&i_gh);
 700
 701         return dblock;
 702 }
 703
 704 static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
 705 {
 706         struct gfs2_bufdata *bd;
 707
 708         lock_buffer(bh);
 709         gfs2_log_lock(sdp);
 710         clear_buffer_dirty(bh);
 711         bd = bh->b_private;
 712         if (bd) {
 713                 if (!list_empty(&bd->bd_list) && !buffer_pinned(bh))
 714                         list_del_init(&bd->bd_list);
 715                 else
 716                         gfs2_remove_from_journal(bh, REMOVE_JDATA);
 717         }
 718         bh->b_bdev = NULL;
 719         clear_buffer_mapped(bh);
 720         clear_buffer_req(bh);
 721         clear_buffer_new(bh);
 722         gfs2_log_unlock(sdp);
 723         unlock_buffer(bh);
 724 }
 725
 726 static void gfs2_invalidatepage(struct page *page, unsigned int offset,
 727                                 unsigned int length)
 728 {
 729         struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
 730         unsigned int stop = offset + length;
 731         int partial_page = (offset || length < PAGE_SIZE);
 732         struct buffer_head *bh, *head;
 733         unsigned long pos = 0;
 734
 735         BUG_ON(!PageLocked(page));
 736         if (!partial_page)
 737                 ClearPageChecked(page);
 738         if (!page_has_buffers(page))
 739                 goto out;
 740
 741         bh = head = page_buffers(page);
 742         do {
 743                 if (pos + bh->b_size > stop)
 744                         return;
 745
 746                 if (offset <= pos)
 747                         gfs2_discard(sdp, bh);
 748                 pos += bh->b_size;
 749                 bh = bh->b_this_page;
 750         } while (bh != head);
 751 out:
 752         if (!partial_page)
 753                 try_to_release_page(page, 0);
 754 }
 755
 756 /**
 757  * gfs2_releasepage - free the metadata associated with a page
 758  * @page: the page that's being released
 759  * @gfp_mask: passed from Linux VFS, ignored by us
 760  *
 761  * Calls try_to_free_buffers() to free the buffers and put the page if the
 762  * buffers can be released.
 763  *
 764  * Returns: 1 if the page was put or else 0
 765  */
 766
 767 int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
 768 {
 769         struct address_space *mapping = page->mapping;
 770         struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
 771         struct buffer_head *bh, *head;
 772         struct gfs2_bufdata *bd;
 773
 774         if (!page_has_buffers(page))
 775                 return 0;
 776
 777         /*
 778          * From xfs_vm_releasepage: mm accommodates an old ext3 case where
 779          * clean pages might not have had the dirty bit cleared.  Thus, it can
 780          * send actual dirty pages to ->releasepage() via shrink_active_list().
 781          *
 782          * As a workaround, we skip pages that contain dirty buffers below.
 783          * Once ->releasepage isn't called on dirty pages anymore, we can warn
 784          * on dirty buffers like we used to here again.
 785          */
 786
 787         gfs2_log_lock(sdp);
 788         spin_lock(&sdp->sd_ail_lock);
 789         head = bh = page_buffers(page);
 790         do {
 791                 if (atomic_read(&bh->b_count))
 792                         goto cannot_release;
 793                 bd = bh->b_private;
 794                 if (bd && bd->bd_tr)
 795                         goto cannot_release;
 796                 if (buffer_dirty(bh) || WARN_ON(buffer_pinned(bh)))
 797                         goto cannot_release;
 798                 bh = bh->b_this_page;
 799         } while(bh != head);
 800         spin_unlock(&sdp->sd_ail_lock);
 801
 802         head = bh = page_buffers(page);
 803         do {
 804                 bd = bh->b_private;
 805                 if (bd) {
 806                         gfs2_assert_warn(sdp, bd->bd_bh == bh);
 807                         if (!list_empty(&bd->bd_list))
 808                                 list_del_init(&bd->bd_list);
 809                         bd->bd_bh = NULL;
 810                         bh->b_private = NULL;
 811                         kmem_cache_free(gfs2_bufdata_cachep, bd);
 812                 }
 813
 814                 bh = bh->b_this_page;
 815         } while (bh != head);
 816         gfs2_log_unlock(sdp);
 817
 818         return try_to_free_buffers(page);
 819
 820 cannot_release:
 821         spin_unlock(&sdp->sd_ail_lock);
 822         gfs2_log_unlock(sdp);
 823         return 0;
 824 }
 825
 826 static const struct address_space_operations gfs2_aops = {
 827         .writepage = gfs2_writepage,
 828         .writepages = gfs2_writepages,
 829         .readpage = gfs2_readpage,
 830         .readpages = gfs2_readpages,
 831         .bmap = gfs2_bmap,
 832         .invalidatepage = gfs2_invalidatepage,
 833         .releasepage = gfs2_releasepage,
 834         .direct_IO = noop_direct_IO,
 835         .migratepage = buffer_migrate_page,
 836         .is_partially_uptodate = block_is_partially_uptodate,
 837         .error_remove_page = generic_error_remove_page,
 838 };
 839
 840 static const struct address_space_operations gfs2_jdata_aops = {
 841         .writepage = gfs2_jdata_writepage,
 842         .writepages = gfs2_jdata_writepages,
 843         .readpage = gfs2_readpage,
 844         .readpages = gfs2_readpages,
 845         .set_page_dirty = jdata_set_page_dirty,
 846         .bmap = gfs2_bmap,
 847         .invalidatepage = gfs2_invalidatepage,
 848         .releasepage = gfs2_releasepage,
 849         .is_partially_uptodate = block_is_partially_uptodate,
 850         .error_remove_page = generic_error_remove_page,
 851 };
 852
 853 void gfs2_set_aops(struct inode *inode)
 854 {
 855         if (gfs2_is_jdata(GFS2_I(inode)))
 856                 inode->i_mapping->a_ops = &gfs2_jdata_aops;
 857         else
 858                 inode->i_mapping->a_ops = &gfs2_aops;
 859 }