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 -ENODATA;
  81         return 0;
  82 }
  83
  84 /**
  85  * gfs2_write_jdata_page - gfs2 jdata-specific version of block_write_full_page
  86  * @page: The page to write
  87  * @wbc: The writeback control
  88  *
  89  * This is the same as calling block_write_full_page, but it also
  90  * writes pages outside of i_size
  91  */
  92 static int gfs2_write_jdata_page(struct page *page,
  93                                  struct writeback_control *wbc)
  94 {
  95         struct inode * const inode = page->mapping->host;
  96         loff_t i_size = i_size_read(inode);
  97         const pgoff_t end_index = i_size >> PAGE_SHIFT;
  98         unsigned offset;
  99
 100         /*
 101          * The page straddles i_size.  It must be zeroed out on each and every
 102          * writepage invocation because it may be mmapped.  "A file is mapped
 103          * in multiples of the page size.  For a file that is not a multiple of
 104          * the  page size, the remaining memory is zeroed when mapped, and
 105          * writes to that region are not written out to the file."
 106          */
 107         offset = i_size & (PAGE_SIZE - 1);
 108         if (page->index == end_index && offset)
 109                 zero_user_segment(page, offset, PAGE_SIZE);
 110
 111         return __block_write_full_page(inode, page, gfs2_get_block_noalloc, wbc,
 112                                        end_buffer_async_write);
 113 }
 114
 115 /**
 116  * __gfs2_jdata_writepage - The core of jdata writepage
 117  * @page: The page to write
 118  * @wbc: The writeback control
 119  *
 120  * This is shared between writepage and writepages and implements the
 121  * core of the writepage operation. If a transaction is required then
 122  * PageChecked will have been set and the transaction will have
 123  * already been started before this is called.
 124  */
 125
 126 static int __gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
 127 {
 128         struct inode *inode = page->mapping->host;
 129         struct gfs2_inode *ip = GFS2_I(inode);
 130
 131         if (PageChecked(page)) {
 132                 ClearPageChecked(page);
 133                 if (!page_has_buffers(page)) {
 134                         create_empty_buffers(page, inode->i_sb->s_blocksize,
 135                                              BIT(BH_Dirty)|BIT(BH_Uptodate));
 136                 }
 137                 gfs2_page_add_databufs(ip, page, 0, PAGE_SIZE);
 138         }
 139         return gfs2_write_jdata_page(page, wbc);
 140 }
 141
 142 /**
 143  * gfs2_jdata_writepage - Write complete page
 144  * @page: Page to write
 145  * @wbc: The writeback control
 146  *
 147  * Returns: errno
 148  *
 149  */
 150
 151 static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
 152 {
 153         struct inode *inode = page->mapping->host;
 154         struct gfs2_inode *ip = GFS2_I(inode);
 155         struct gfs2_sbd *sdp = GFS2_SB(inode);
 156
 157         if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl)))
 158                 goto out;
 159         if (PageChecked(page) || current->journal_info)
 160                 goto out_ignore;
 161         return __gfs2_jdata_writepage(page, wbc);
 162
 163 out_ignore:
 164         redirty_page_for_writepage(wbc, page);
 165 out:
 166         unlock_page(page);
 167         return 0;
 168 }
 169
 170 /**
 171  * gfs2_writepages - Write a bunch of dirty pages back to disk
 172  * @mapping: The mapping to write
 173  * @wbc: Write-back control
 174  *
 175  * Used for both ordered and writeback modes.
 176  */
 177 static int gfs2_writepages(struct address_space *mapping,
 178                            struct writeback_control *wbc)
 179 {
 180         struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
 181         struct iomap_writepage_ctx wpc = { };
 182         int ret;
 183
 184         /*
 185          * Even if we didn't write enough pages here, we might still be holding
 186          * dirty pages in the ail. We forcibly flush the ail because we don't
 187          * want balance_dirty_pages() to loop indefinitely trying to write out
 188          * pages held in the ail that it can't find.
 189          */
 190         ret = iomap_writepages(mapping, wbc, &wpc, &gfs2_writeback_ops);
 191         if (ret == 0 && wbc->nr_to_write > 0)
 192                 set_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags);
 193         return ret;
 194 }
 195
 196 /**
 197  * gfs2_write_jdata_pagevec - Write back a pagevec's worth of pages
 198  * @mapping: The mapping
 199  * @wbc: The writeback control
 200  * @pvec: The vector of pages
 201  * @nr_pages: The number of pages to write
 202  * @done_index: Page index
 203  *
 204  * Returns: non-zero if loop should terminate, zero otherwise
 205  */
 206
 207 static int gfs2_write_jdata_pagevec(struct address_space *mapping,
 208                                     struct writeback_control *wbc,
 209                                     struct pagevec *pvec,
 210                                     int nr_pages,
 211                                     pgoff_t *done_index)
 212 {
 213         struct inode *inode = mapping->host;
 214         struct gfs2_sbd *sdp = GFS2_SB(inode);
 215         unsigned nrblocks = nr_pages * (PAGE_SIZE >> inode->i_blkbits);
 216         int i;
 217         int ret;
 218
 219         ret = gfs2_trans_begin(sdp, nrblocks, nrblocks);
 220         if (ret < 0)
 221                 return ret;
 222
 223         for(i = 0; i < nr_pages; i++) {
 224                 struct page *page = pvec->pages[i];
 225
 226                 *done_index = page->index;
 227
 228                 lock_page(page);
 229
 230                 if (unlikely(page->mapping != mapping)) {
 231 continue_unlock:
 232                         unlock_page(page);
 233                         continue;
 234                 }
 235
 236                 if (!PageDirty(page)) {
 237                         /* someone wrote it for us */
 238                         goto continue_unlock;
 239                 }
 240
 241                 if (PageWriteback(page)) {
 242                         if (wbc->sync_mode != WB_SYNC_NONE)
 243                                 wait_on_page_writeback(page);
 244                         else
 245                                 goto continue_unlock;
 246                 }
 247
 248                 BUG_ON(PageWriteback(page));
 249                 if (!clear_page_dirty_for_io(page))
 250                         goto continue_unlock;
 251
 252                 trace_wbc_writepage(wbc, inode_to_bdi(inode));
 253
 254                 ret = __gfs2_jdata_writepage(page, wbc);
 255                 if (unlikely(ret)) {
 256                         if (ret == AOP_WRITEPAGE_ACTIVATE) {
 257                                 unlock_page(page);
 258                                 ret = 0;
 259                         } else {
 260
 261                                 /*
 262                                  * done_index is set past this page,
 263                                  * so media errors will not choke
 264                                  * background writeout for the entire
 265                                  * file. This has consequences for
 266                                  * range_cyclic semantics (ie. it may
 267                                  * not be suitable for data integrity
 268                                  * writeout).
 269                                  */
 270                                 *done_index = page->index + 1;
 271                                 ret = 1;
 272                                 break;
 273                         }
 274                 }
 275
 276                 /*
 277                  * We stop writing back only if we are not doing
 278                  * integrity sync. In case of integrity sync we have to
 279                  * keep going until we have written all the pages
 280                  * we tagged for writeback prior to entering this loop.
 281                  */
 282                 if (--wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) {
 283                         ret = 1;
 284                         break;
 285                 }
 286
 287         }
 288         gfs2_trans_end(sdp);
 289         return ret;
 290 }
 291
 292 /**
 293  * gfs2_write_cache_jdata - Like write_cache_pages but different
 294  * @mapping: The mapping to write
 295  * @wbc: The writeback control
 296  *
 297  * The reason that we use our own function here is that we need to
 298  * start transactions before we grab page locks. This allows us
 299  * to get the ordering right.
 300  */
 301
 302 static int gfs2_write_cache_jdata(struct address_space *mapping,
 303                                   struct writeback_control *wbc)
 304 {
 305         int ret = 0;
 306         int done = 0;
 307         struct pagevec pvec;
 308         int nr_pages;
 309         pgoff_t writeback_index;
 310         pgoff_t index;
 311         pgoff_t end;
 312         pgoff_t done_index;
 313         int cycled;
 314         int range_whole = 0;
 315         xa_mark_t tag;
 316
 317         pagevec_init(&pvec);
 318         if (wbc->range_cyclic) {
 319                 writeback_index = mapping->writeback_index; /* prev offset */
 320                 index = writeback_index;
 321                 if (index == 0)
 322                         cycled = 1;
 323                 else
 324                         cycled = 0;
 325                 end = -1;
 326         } else {
 327                 index = wbc->range_start >> PAGE_SHIFT;
 328                 end = wbc->range_end >> PAGE_SHIFT;
 329                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
 330                         range_whole = 1;
 331                 cycled = 1; /* ignore range_cyclic tests */
 332         }
 333         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
 334                 tag = PAGECACHE_TAG_TOWRITE;
 335         else
 336                 tag = PAGECACHE_TAG_DIRTY;
 337
 338 retry:
 339         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
 340                 tag_pages_for_writeback(mapping, index, end);
 341         done_index = index;
 342         while (!done && (index <= end)) {
 343                 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
 344                                 tag);
 345                 if (nr_pages == 0)
 346                         break;
 347
 348                 ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, &done_index);
 349                 if (ret)
 350                         done = 1;
 351                 if (ret > 0)
 352                         ret = 0;
 353                 pagevec_release(&pvec);
 354                 cond_resched();
 355         }
 356
 357         if (!cycled && !done) {
 358                 /*
 359                  * range_cyclic:
 360                  * We hit the last page and there is more work to be done: wrap
 361                  * back to the start of the file
 362                  */
 363                 cycled = 1;
 364                 index = 0;
 365                 end = writeback_index - 1;
 366                 goto retry;
 367         }
 368
 369         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
 370                 mapping->writeback_index = done_index;
 371
 372         return ret;
 373 }
 374
 375
 376 /**
 377  * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
 378  * @mapping: The mapping to write
 379  * @wbc: The writeback control
 380  *
 381  */
 382
 383 static int gfs2_jdata_writepages(struct address_space *mapping,
 384                                  struct writeback_control *wbc)
 385 {
 386         struct gfs2_inode *ip = GFS2_I(mapping->host);
 387         struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
 388         int ret;
 389
 390         ret = gfs2_write_cache_jdata(mapping, wbc);
 391         if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) {
 392                 gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
 393                                GFS2_LFC_JDATA_WPAGES);
 394                 ret = gfs2_write_cache_jdata(mapping, wbc);
 395         }
 396         return ret;
 397 }
 398
 399 /**
 400  * stuffed_readpage - Fill in a Linux page with stuffed file data
 401  * @ip: the inode
 402  * @page: the page
 403  *
 404  * Returns: errno
 405  */
 406 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
 407 {
 408         struct buffer_head *dibh;
 409         u64 dsize = i_size_read(&ip->i_inode);
 410         void *kaddr;
 411         int error;
 412
 413         /*
 414          * Due to the order of unstuffing files and ->fault(), we can be
 415          * asked for a zero page in the case of a stuffed file being extended,
 416          * so we need to supply one here. It doesn't happen often.
 417          */
 418         if (unlikely(page->index)) {
 419                 zero_user(page, 0, PAGE_SIZE);
 420                 SetPageUptodate(page);
 421                 return 0;
 422         }
 423
 424         error = gfs2_meta_inode_buffer(ip, &dibh);
 425         if (error)
 426                 return error;
 427
 428         kaddr = kmap_atomic(page);
 429         memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
 430         memset(kaddr + dsize, 0, PAGE_SIZE - dsize);
 431         kunmap_atomic(kaddr);
 432         flush_dcache_page(page);
 433         brelse(dibh);
 434         SetPageUptodate(page);
 435
 436         return 0;
 437 }
 438
 439 /**
 440  * gfs2_read_folio - read a folio from a file
 441  * @file: The file to read
 442  * @folio: The folio in the file
 443  */
 444 static int gfs2_read_folio(struct file *file, struct folio *folio)
 445 {
 446         struct inode *inode = folio->mapping->host;
 447         struct gfs2_inode *ip = GFS2_I(inode);
 448         struct gfs2_sbd *sdp = GFS2_SB(inode);
 449         int error;
 450
 451         if (!gfs2_is_jdata(ip) ||
 452             (i_blocksize(inode) == PAGE_SIZE && !folio_buffers(folio))) {
 453                 error = iomap_read_folio(folio, &gfs2_iomap_ops);
 454         } else if (gfs2_is_stuffed(ip)) {
 455                 error = stuffed_readpage(ip, &folio->page);
 456                 folio_unlock(folio);
 457         } else {
 458                 error = mpage_read_folio(folio, gfs2_block_map);
 459         }
 460
 461         if (unlikely(gfs2_withdrawn(sdp)))
 462                 return -EIO;
 463
 464         return error;
 465 }
 466
 467 /**
 468  * gfs2_internal_read - read an internal file
 469  * @ip: The gfs2 inode
 470  * @buf: The buffer to fill
 471  * @pos: The file position
 472  * @size: The amount to read
 473  *
 474  */
 475
 476 int gfs2_internal_read(struct gfs2_inode *ip, char *buf, loff_t *pos,
 477                        unsigned size)
 478 {
 479         struct address_space *mapping = ip->i_inode.i_mapping;
 480         unsigned long index = *pos >> PAGE_SHIFT;
 481         unsigned offset = *pos & (PAGE_SIZE - 1);
 482         unsigned copied = 0;
 483         unsigned amt;
 484         struct page *page;
 485         void *p;
 486
 487         do {
 488                 amt = size - copied;
 489                 if (offset + size > PAGE_SIZE)
 490                         amt = PAGE_SIZE - offset;
 491                 page = read_cache_page(mapping, index, gfs2_read_folio, NULL);
 492                 if (IS_ERR(page))
 493                         return PTR_ERR(page);
 494                 p = kmap_atomic(page);
 495                 memcpy(buf + copied, p + offset, amt);
 496                 kunmap_atomic(p);
 497                 put_page(page);
 498                 copied += amt;
 499                 index++;
 500                 offset = 0;
 501         } while(copied < size);
 502         (*pos) += size;
 503         return size;
 504 }
 505
 506 /**
 507  * gfs2_readahead - Read a bunch of pages at once
 508  * @rac: Read-ahead control structure
 509  *
 510  * Some notes:
 511  * 1. This is only for readahead, so we can simply ignore any things
 512  *    which are slightly inconvenient (such as locking conflicts between
 513  *    the page lock and the glock) and return having done no I/O. Its
 514  *    obviously not something we'd want to do on too regular a basis.
 515  *    Any I/O we ignore at this time will be done via readpage later.
 516  * 2. We don't handle stuffed files here we let readpage do the honours.
 517  * 3. mpage_readahead() does most of the heavy lifting in the common case.
 518  * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
 519  */
 520
 521 static void gfs2_readahead(struct readahead_control *rac)
 522 {
 523         struct inode *inode = rac->mapping->host;
 524         struct gfs2_inode *ip = GFS2_I(inode);
 525
 526         if (gfs2_is_stuffed(ip))
 527                 ;
 528         else if (gfs2_is_jdata(ip))
 529                 mpage_readahead(rac, gfs2_block_map);
 530         else
 531                 iomap_readahead(rac, &gfs2_iomap_ops);
 532 }
 533
 534 /**
 535  * adjust_fs_space - Adjusts the free space available due to gfs2_grow
 536  * @inode: the rindex inode
 537  */
 538 void adjust_fs_space(struct inode *inode)
 539 {
 540         struct gfs2_sbd *sdp = GFS2_SB(inode);
 541         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 542         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 543         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 544         struct buffer_head *m_bh;
 545         u64 fs_total, new_free;
 546
 547         if (gfs2_trans_begin(sdp, 2 * RES_STATFS, 0) != 0)
 548                 return;
 549
 550         /* Total up the file system space, according to the latest rindex. */
 551         fs_total = gfs2_ri_total(sdp);
 552         if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0)
 553                 goto out;
 554
 555         spin_lock(&sdp->sd_statfs_spin);
 556         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 557                               sizeof(struct gfs2_dinode));
 558         if (fs_total > (m_sc->sc_total + l_sc->sc_total))
 559                 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
 560         else
 561                 new_free = 0;
 562         spin_unlock(&sdp->sd_statfs_spin);
 563         fs_warn(sdp, "File system extended by %llu blocks.\n",
 564                 (unsigned long long)new_free);
 565         gfs2_statfs_change(sdp, new_free, new_free, 0);
 566
 567         update_statfs(sdp, m_bh);
 568         brelse(m_bh);
 569 out:
 570         sdp->sd_rindex_uptodate = 0;
 571         gfs2_trans_end(sdp);
 572 }
 573
 574 static bool jdata_dirty_folio(struct address_space *mapping,
 575                 struct folio *folio)
 576 {
 577         if (current->journal_info)
 578                 folio_set_checked(folio);
 579         return block_dirty_folio(mapping, folio);
 580 }
 581
 582 /**
 583  * gfs2_bmap - Block map function
 584  * @mapping: Address space info
 585  * @lblock: The block to map
 586  *
 587  * Returns: The disk address for the block or 0 on hole or error
 588  */
 589
 590 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
 591 {
 592         struct gfs2_inode *ip = GFS2_I(mapping->host);
 593         struct gfs2_holder i_gh;
 594         sector_t dblock = 0;
 595         int error;
 596
 597         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
 598         if (error)
 599                 return 0;
 600
 601         if (!gfs2_is_stuffed(ip))
 602                 dblock = iomap_bmap(mapping, lblock, &gfs2_iomap_ops);
 603
 604         gfs2_glock_dq_uninit(&i_gh);
 605
 606         return dblock;
 607 }
 608
 609 static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
 610 {
 611         struct gfs2_bufdata *bd;
 612
 613         lock_buffer(bh);
 614         gfs2_log_lock(sdp);
 615         clear_buffer_dirty(bh);
 616         bd = bh->b_private;
 617         if (bd) {
 618                 if (!list_empty(&bd->bd_list) && !buffer_pinned(bh))
 619                         list_del_init(&bd->bd_list);
 620                 else {
 621                         spin_lock(&sdp->sd_ail_lock);
 622                         gfs2_remove_from_journal(bh, REMOVE_JDATA);
 623                         spin_unlock(&sdp->sd_ail_lock);
 624                 }
 625         }
 626         bh->b_bdev = NULL;
 627         clear_buffer_mapped(bh);
 628         clear_buffer_req(bh);
 629         clear_buffer_new(bh);
 630         gfs2_log_unlock(sdp);
 631         unlock_buffer(bh);
 632 }
 633
 634 static void gfs2_invalidate_folio(struct folio *folio, size_t offset,
 635                                 size_t length)
 636 {
 637         struct gfs2_sbd *sdp = GFS2_SB(folio->mapping->host);
 638         size_t stop = offset + length;
 639         int partial_page = (offset || length < folio_size(folio));
 640         struct buffer_head *bh, *head;
 641         unsigned long pos = 0;
 642
 643         BUG_ON(!folio_test_locked(folio));
 644         if (!partial_page)
 645                 folio_clear_checked(folio);
 646         head = folio_buffers(folio);
 647         if (!head)
 648                 goto out;
 649
 650         bh = head;
 651         do {
 652                 if (pos + bh->b_size > stop)
 653                         return;
 654
 655                 if (offset <= pos)
 656                         gfs2_discard(sdp, bh);
 657                 pos += bh->b_size;
 658                 bh = bh->b_this_page;
 659         } while (bh != head);
 660 out:
 661         if (!partial_page)
 662                 filemap_release_folio(folio, 0);
 663 }
 664
 665 /**
 666  * gfs2_release_folio - free the metadata associated with a folio
 667  * @folio: the folio that's being released
 668  * @gfp_mask: passed from Linux VFS, ignored by us
 669  *
 670  * Calls try_to_free_buffers() to free the buffers and put the folio if the
 671  * buffers can be released.
 672  *
 673  * Returns: true if the folio was put or else false
 674  */
 675
 676 bool gfs2_release_folio(struct folio *folio, gfp_t gfp_mask)
 677 {
 678         struct address_space *mapping = folio->mapping;
 679         struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
 680         struct buffer_head *bh, *head;
 681         struct gfs2_bufdata *bd;
 682
 683         head = folio_buffers(folio);
 684         if (!head)
 685                 return false;
 686
 687         /*
 688          * mm accommodates an old ext3 case where clean folios might
 689          * not have had the dirty bit cleared.  Thus, it can send actual
 690          * dirty folios to ->release_folio() via shrink_active_list().
 691          *
 692          * As a workaround, we skip folios that contain dirty buffers
 693          * below.  Once ->release_folio isn't called on dirty folios
 694          * anymore, we can warn on dirty buffers like we used to here
 695          * again.
 696          */
 697
 698         gfs2_log_lock(sdp);
 699         bh = head;
 700         do {
 701                 if (atomic_read(&bh->b_count))
 702                         goto cannot_release;
 703                 bd = bh->b_private;
 704                 if (bd && bd->bd_tr)
 705                         goto cannot_release;
 706                 if (buffer_dirty(bh) || WARN_ON(buffer_pinned(bh)))
 707                         goto cannot_release;
 708                 bh = bh->b_this_page;
 709         } while (bh != head);
 710
 711         bh = head;
 712         do {
 713                 bd = bh->b_private;
 714                 if (bd) {
 715                         gfs2_assert_warn(sdp, bd->bd_bh == bh);
 716                         bd->bd_bh = NULL;
 717                         bh->b_private = NULL;
 718                         /*
 719                          * The bd may still be queued as a revoke, in which
 720                          * case we must not dequeue nor free it.
 721                          */
 722                         if (!bd->bd_blkno && !list_empty(&bd->bd_list))
 723                                 list_del_init(&bd->bd_list);
 724                         if (list_empty(&bd->bd_list))
 725                                 kmem_cache_free(gfs2_bufdata_cachep, bd);
 726                 }
 727
 728                 bh = bh->b_this_page;
 729         } while (bh != head);
 730         gfs2_log_unlock(sdp);
 731
 732         return try_to_free_buffers(folio);
 733
 734 cannot_release:
 735         gfs2_log_unlock(sdp);
 736         return false;
 737 }
 738
 739 static const struct address_space_operations gfs2_aops = {
 740         .writepages = gfs2_writepages,
 741         .read_folio = gfs2_read_folio,
 742         .readahead = gfs2_readahead,
 743         .dirty_folio = filemap_dirty_folio,
 744         .release_folio = iomap_release_folio,
 745         .invalidate_folio = iomap_invalidate_folio,
 746         .bmap = gfs2_bmap,
 747         .direct_IO = noop_direct_IO,
 748         .migrate_folio = filemap_migrate_folio,
 749         .is_partially_uptodate = iomap_is_partially_uptodate,
 750         .error_remove_page = generic_error_remove_page,
 751 };
 752
 753 static const struct address_space_operations gfs2_jdata_aops = {
 754         .writepage = gfs2_jdata_writepage,
 755         .writepages = gfs2_jdata_writepages,
 756         .read_folio = gfs2_read_folio,
 757         .readahead = gfs2_readahead,
 758         .dirty_folio = jdata_dirty_folio,
 759         .bmap = gfs2_bmap,
 760         .invalidate_folio = gfs2_invalidate_folio,
 761         .release_folio = gfs2_release_folio,
 762         .is_partially_uptodate = block_is_partially_uptodate,
 763         .error_remove_page = generic_error_remove_page,
 764 };
 765
 766 void gfs2_set_aops(struct inode *inode)
 767 {
 768         if (gfs2_is_jdata(GFS2_I(inode)))
 769                 inode->i_mapping->a_ops = &gfs2_jdata_aops;
 770         else
 771                 inode->i_mapping->a_ops = &gfs2_aops;
 772 }