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_trans_add_databufs(struct gfs2_inode *ip, struct folio *folio,
  41                              size_t from, size_t len)
  42 {
  43         struct buffer_head *head = folio_buffers(folio);
  44         unsigned int bsize = head->b_size;
  45         struct buffer_head *bh;
  46         size_t to = from + len;
  47         size_t 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_folio - gfs2 jdata-specific version of block_write_full_page
  86  * @folio: The folio 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_folio(struct folio *folio,
  93                                  struct writeback_control *wbc)
  94 {
  95         struct inode * const inode = folio->mapping->host;
  96         loff_t i_size = i_size_read(inode);
  97
  98         /*
  99          * The folio straddles i_size.  It must be zeroed out on each and every
 100          * writepage invocation because it may be mmapped.  "A file is mapped
 101          * in multiples of the page size.  For a file that is not a multiple of
 102          * the page size, the remaining memory is zeroed when mapped, and
 103          * writes to that region are not written out to the file."
 104          */
 105         if (folio_pos(folio) < i_size &&
 106             i_size < folio_pos(folio) + folio_size(folio))
 107                 folio_zero_segment(folio, offset_in_folio(folio, i_size),
 108                                 folio_size(folio));
 109
 110         return __block_write_full_folio(inode, folio, gfs2_get_block_noalloc,
 111                         wbc, end_buffer_async_write);
 112 }
 113
 114 /**
 115  * __gfs2_jdata_write_folio - The core of jdata writepage
 116  * @folio: The folio to write
 117  * @wbc: The writeback control
 118  *
 119  * This is shared between writepage and writepages and implements the
 120  * core of the writepage operation. If a transaction is required then
 121  * the checked flag will have been set and the transaction will have
 122  * already been started before this is called.
 123  */
 124 static int __gfs2_jdata_write_folio(struct folio *folio,
 125                 struct writeback_control *wbc)
 126 {
 127         struct inode *inode = folio->mapping->host;
 128         struct gfs2_inode *ip = GFS2_I(inode);
 129
 130         if (folio_test_checked(folio)) {
 131                 folio_clear_checked(folio);
 132                 if (!folio_buffers(folio)) {
 133                         folio_create_empty_buffers(folio,
 134                                         inode->i_sb->s_blocksize,
 135                                         BIT(BH_Dirty)|BIT(BH_Uptodate));
 136                 }
 137                 gfs2_trans_add_databufs(ip, folio, 0, folio_size(folio));
 138         }
 139         return gfs2_write_jdata_folio(folio, 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 folio *folio = page_folio(page);
 154         struct inode *inode = page->mapping->host;
 155         struct gfs2_inode *ip = GFS2_I(inode);
 156         struct gfs2_sbd *sdp = GFS2_SB(inode);
 157
 158         if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl)))
 159                 goto out;
 160         if (folio_test_checked(folio) || current->journal_info)
 161                 goto out_ignore;
 162         return __gfs2_jdata_write_folio(folio, wbc);
 163
 164 out_ignore:
 165         folio_redirty_for_writepage(wbc, folio);
 166 out:
 167         folio_unlock(folio);
 168         return 0;
 169 }
 170
 171 /**
 172  * gfs2_writepages - Write a bunch of dirty pages back to disk
 173  * @mapping: The mapping to write
 174  * @wbc: Write-back control
 175  *
 176  * Used for both ordered and writeback modes.
 177  */
 178 static int gfs2_writepages(struct address_space *mapping,
 179                            struct writeback_control *wbc)
 180 {
 181         struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
 182         struct iomap_writepage_ctx wpc = { };
 183         int ret;
 184
 185         /*
 186          * Even if we didn't write any pages here, we might still be holding
 187          * dirty pages in the ail. We forcibly flush the ail because we don't
 188          * want balance_dirty_pages() to loop indefinitely trying to write out
 189          * pages held in the ail that it can't find.
 190          */
 191         ret = iomap_writepages(mapping, wbc, &wpc, &gfs2_writeback_ops);
 192         if (ret == 0)
 193                 set_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags);
 194         return ret;
 195 }
 196
 197 /**
 198  * gfs2_write_jdata_batch - Write back a folio batch's worth of folios
 199  * @mapping: The mapping
 200  * @wbc: The writeback control
 201  * @fbatch: The batch of folios
 202  * @done_index: Page index
 203  *
 204  * Returns: non-zero if loop should terminate, zero otherwise
 205  */
 206
 207 static int gfs2_write_jdata_batch(struct address_space *mapping,
 208                                     struct writeback_control *wbc,
 209                                     struct folio_batch *fbatch,
 210                                     pgoff_t *done_index)
 211 {
 212         struct inode *inode = mapping->host;
 213         struct gfs2_sbd *sdp = GFS2_SB(inode);
 214         unsigned nrblocks;
 215         int i;
 216         int ret;
 217         int nr_pages = 0;
 218         int nr_folios = folio_batch_count(fbatch);
 219
 220         for (i = 0; i < nr_folios; i++)
 221                 nr_pages += folio_nr_pages(fbatch->folios[i]);
 222         nrblocks = nr_pages * (PAGE_SIZE >> inode->i_blkbits);
 223
 224         ret = gfs2_trans_begin(sdp, nrblocks, nrblocks);
 225         if (ret < 0)
 226                 return ret;
 227
 228         for (i = 0; i < nr_folios; i++) {
 229                 struct folio *folio = fbatch->folios[i];
 230
 231                 *done_index = folio->index;
 232
 233                 folio_lock(folio);
 234
 235                 if (unlikely(folio->mapping != mapping)) {
 236 continue_unlock:
 237                         folio_unlock(folio);
 238                         continue;
 239                 }
 240
 241                 if (!folio_test_dirty(folio)) {
 242                         /* someone wrote it for us */
 243                         goto continue_unlock;
 244                 }
 245
 246                 if (folio_test_writeback(folio)) {
 247                         if (wbc->sync_mode != WB_SYNC_NONE)
 248                                 folio_wait_writeback(folio);
 249                         else
 250                                 goto continue_unlock;
 251                 }
 252
 253                 BUG_ON(folio_test_writeback(folio));
 254                 if (!folio_clear_dirty_for_io(folio))
 255                         goto continue_unlock;
 256
 257                 trace_wbc_writepage(wbc, inode_to_bdi(inode));
 258
 259                 ret = __gfs2_jdata_write_folio(folio, wbc);
 260                 if (unlikely(ret)) {
 261                         if (ret == AOP_WRITEPAGE_ACTIVATE) {
 262                                 folio_unlock(folio);
 263                                 ret = 0;
 264                         } else {
 265
 266                                 /*
 267                                  * done_index is set past this page,
 268                                  * so media errors will not choke
 269                                  * background writeout for the entire
 270                                  * file. This has consequences for
 271                                  * range_cyclic semantics (ie. it may
 272                                  * not be suitable for data integrity
 273                                  * writeout).
 274                                  */
 275                                 *done_index = folio->index +
 276                                         folio_nr_pages(folio);
 277                                 ret = 1;
 278                                 break;
 279                         }
 280                 }
 281
 282                 /*
 283                  * We stop writing back only if we are not doing
 284                  * integrity sync. In case of integrity sync we have to
 285                  * keep going until we have written all the pages
 286                  * we tagged for writeback prior to entering this loop.
 287                  */
 288                 if (--wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) {
 289                         ret = 1;
 290                         break;
 291                 }
 292
 293         }
 294         gfs2_trans_end(sdp);
 295         return ret;
 296 }
 297
 298 /**
 299  * gfs2_write_cache_jdata - Like write_cache_pages but different
 300  * @mapping: The mapping to write
 301  * @wbc: The writeback control
 302  *
 303  * The reason that we use our own function here is that we need to
 304  * start transactions before we grab page locks. This allows us
 305  * to get the ordering right.
 306  */
 307
 308 static int gfs2_write_cache_jdata(struct address_space *mapping,
 309                                   struct writeback_control *wbc)
 310 {
 311         int ret = 0;
 312         int done = 0;
 313         struct folio_batch fbatch;
 314         int nr_folios;
 315         pgoff_t writeback_index;
 316         pgoff_t index;
 317         pgoff_t end;
 318         pgoff_t done_index;
 319         int cycled;
 320         int range_whole = 0;
 321         xa_mark_t tag;
 322
 323         folio_batch_init(&fbatch);
 324         if (wbc->range_cyclic) {
 325                 writeback_index = mapping->writeback_index; /* prev offset */
 326                 index = writeback_index;
 327                 if (index == 0)
 328                         cycled = 1;
 329                 else
 330                         cycled = 0;
 331                 end = -1;
 332         } else {
 333                 index = wbc->range_start >> PAGE_SHIFT;
 334                 end = wbc->range_end >> PAGE_SHIFT;
 335                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
 336                         range_whole = 1;
 337                 cycled = 1; /* ignore range_cyclic tests */
 338         }
 339         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
 340                 tag = PAGECACHE_TAG_TOWRITE;
 341         else
 342                 tag = PAGECACHE_TAG_DIRTY;
 343
 344 retry:
 345         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
 346                 tag_pages_for_writeback(mapping, index, end);
 347         done_index = index;
 348         while (!done && (index <= end)) {
 349                 nr_folios = filemap_get_folios_tag(mapping, &index, end,
 350                                 tag, &fbatch);
 351                 if (nr_folios == 0)
 352                         break;
 353
 354                 ret = gfs2_write_jdata_batch(mapping, wbc, &fbatch,
 355                                 &done_index);
 356                 if (ret)
 357                         done = 1;
 358                 if (ret > 0)
 359                         ret = 0;
 360                 folio_batch_release(&fbatch);
 361                 cond_resched();
 362         }
 363
 364         if (!cycled && !done) {
 365                 /*
 366                  * range_cyclic:
 367                  * We hit the last page and there is more work to be done: wrap
 368                  * back to the start of the file
 369                  */
 370                 cycled = 1;
 371                 index = 0;
 372                 end = writeback_index - 1;
 373                 goto retry;
 374         }
 375
 376         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
 377                 mapping->writeback_index = done_index;
 378
 379         return ret;
 380 }
 381
 382
 383 /**
 384  * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
 385  * @mapping: The mapping to write
 386  * @wbc: The writeback control
 387  *
 388  */
 389
 390 static int gfs2_jdata_writepages(struct address_space *mapping,
 391                                  struct writeback_control *wbc)
 392 {
 393         struct gfs2_inode *ip = GFS2_I(mapping->host);
 394         struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
 395         int ret;
 396
 397         ret = gfs2_write_cache_jdata(mapping, wbc);
 398         if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) {
 399                 gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
 400                                GFS2_LFC_JDATA_WPAGES);
 401                 ret = gfs2_write_cache_jdata(mapping, wbc);
 402         }
 403         return ret;
 404 }
 405
 406 /**
 407  * stuffed_readpage - Fill in a Linux page with stuffed file data
 408  * @ip: the inode
 409  * @page: the page
 410  *
 411  * Returns: errno
 412  */
 413 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
 414 {
 415         struct buffer_head *dibh;
 416         u64 dsize = i_size_read(&ip->i_inode);
 417         void *kaddr;
 418         int error;
 419
 420         /*
 421          * Due to the order of unstuffing files and ->fault(), we can be
 422          * asked for a zero page in the case of a stuffed file being extended,
 423          * so we need to supply one here. It doesn't happen often.
 424          */
 425         if (unlikely(page->index)) {
 426                 zero_user(page, 0, PAGE_SIZE);
 427                 SetPageUptodate(page);
 428                 return 0;
 429         }
 430
 431         error = gfs2_meta_inode_buffer(ip, &dibh);
 432         if (error)
 433                 return error;
 434
 435         kaddr = kmap_local_page(page);
 436         memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
 437         memset(kaddr + dsize, 0, PAGE_SIZE - dsize);
 438         kunmap_local(kaddr);
 439         flush_dcache_page(page);
 440         brelse(dibh);
 441         SetPageUptodate(page);
 442
 443         return 0;
 444 }
 445
 446 /**
 447  * gfs2_read_folio - read a folio from a file
 448  * @file: The file to read
 449  * @folio: The folio in the file
 450  */
 451 static int gfs2_read_folio(struct file *file, struct folio *folio)
 452 {
 453         struct inode *inode = folio->mapping->host;
 454         struct gfs2_inode *ip = GFS2_I(inode);
 455         struct gfs2_sbd *sdp = GFS2_SB(inode);
 456         int error;
 457
 458         if (!gfs2_is_jdata(ip) ||
 459             (i_blocksize(inode) == PAGE_SIZE && !folio_buffers(folio))) {
 460                 error = iomap_read_folio(folio, &gfs2_iomap_ops);
 461         } else if (gfs2_is_stuffed(ip)) {
 462                 error = stuffed_readpage(ip, &folio->page);
 463                 folio_unlock(folio);
 464         } else {
 465                 error = mpage_read_folio(folio, gfs2_block_map);
 466         }
 467
 468         if (unlikely(gfs2_withdrawn(sdp)))
 469                 return -EIO;
 470
 471         return error;
 472 }
 473
 474 /**
 475  * gfs2_internal_read - read an internal file
 476  * @ip: The gfs2 inode
 477  * @buf: The buffer to fill
 478  * @pos: The file position
 479  * @size: The amount to read
 480  *
 481  */
 482
 483 int gfs2_internal_read(struct gfs2_inode *ip, char *buf, loff_t *pos,
 484                        unsigned size)
 485 {
 486         struct address_space *mapping = ip->i_inode.i_mapping;
 487         unsigned long index = *pos >> PAGE_SHIFT;
 488         unsigned offset = *pos & (PAGE_SIZE - 1);
 489         unsigned copied = 0;
 490         unsigned amt;
 491         struct page *page;
 492
 493         do {
 494                 page = read_cache_page(mapping, index, gfs2_read_folio, NULL);
 495                 if (IS_ERR(page)) {
 496                         if (PTR_ERR(page) == -EINTR)
 497                                 continue;
 498                         return PTR_ERR(page);
 499                 }
 500                 amt = size - copied;
 501                 if (offset + size > PAGE_SIZE)
 502                         amt = PAGE_SIZE - offset;
 503                 memcpy_from_page(buf + copied, page, offset, amt);
 504                 put_page(page);
 505                 copied += amt;
 506                 index++;
 507                 offset = 0;
 508         } while(copied < size);
 509         (*pos) += size;
 510         return size;
 511 }
 512
 513 /**
 514  * gfs2_readahead - Read a bunch of pages at once
 515  * @rac: Read-ahead control structure
 516  *
 517  * Some notes:
 518  * 1. This is only for readahead, so we can simply ignore any things
 519  *    which are slightly inconvenient (such as locking conflicts between
 520  *    the page lock and the glock) and return having done no I/O. Its
 521  *    obviously not something we'd want to do on too regular a basis.
 522  *    Any I/O we ignore at this time will be done via readpage later.
 523  * 2. We don't handle stuffed files here we let readpage do the honours.
 524  * 3. mpage_readahead() does most of the heavy lifting in the common case.
 525  * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
 526  */
 527
 528 static void gfs2_readahead(struct readahead_control *rac)
 529 {
 530         struct inode *inode = rac->mapping->host;
 531         struct gfs2_inode *ip = GFS2_I(inode);
 532
 533         if (gfs2_is_stuffed(ip))
 534                 ;
 535         else if (gfs2_is_jdata(ip))
 536                 mpage_readahead(rac, gfs2_block_map);
 537         else
 538                 iomap_readahead(rac, &gfs2_iomap_ops);
 539 }
 540
 541 /**
 542  * adjust_fs_space - Adjusts the free space available due to gfs2_grow
 543  * @inode: the rindex inode
 544  */
 545 void adjust_fs_space(struct inode *inode)
 546 {
 547         struct gfs2_sbd *sdp = GFS2_SB(inode);
 548         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 549         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 550         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 551         struct buffer_head *m_bh;
 552         u64 fs_total, new_free;
 553
 554         if (gfs2_trans_begin(sdp, 2 * RES_STATFS, 0) != 0)
 555                 return;
 556
 557         /* Total up the file system space, according to the latest rindex. */
 558         fs_total = gfs2_ri_total(sdp);
 559         if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0)
 560                 goto out;
 561
 562         spin_lock(&sdp->sd_statfs_spin);
 563         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 564                               sizeof(struct gfs2_dinode));
 565         if (fs_total > (m_sc->sc_total + l_sc->sc_total))
 566                 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
 567         else
 568                 new_free = 0;
 569         spin_unlock(&sdp->sd_statfs_spin);
 570         fs_warn(sdp, "File system extended by %llu blocks.\n",
 571                 (unsigned long long)new_free);
 572         gfs2_statfs_change(sdp, new_free, new_free, 0);
 573
 574         update_statfs(sdp, m_bh);
 575         brelse(m_bh);
 576 out:
 577         sdp->sd_rindex_uptodate = 0;
 578         gfs2_trans_end(sdp);
 579 }
 580
 581 static bool jdata_dirty_folio(struct address_space *mapping,
 582                 struct folio *folio)
 583 {
 584         if (current->journal_info)
 585                 folio_set_checked(folio);
 586         return block_dirty_folio(mapping, folio);
 587 }
 588
 589 /**
 590  * gfs2_bmap - Block map function
 591  * @mapping: Address space info
 592  * @lblock: The block to map
 593  *
 594  * Returns: The disk address for the block or 0 on hole or error
 595  */
 596
 597 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
 598 {
 599         struct gfs2_inode *ip = GFS2_I(mapping->host);
 600         struct gfs2_holder i_gh;
 601         sector_t dblock = 0;
 602         int error;
 603
 604         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
 605         if (error)
 606                 return 0;
 607
 608         if (!gfs2_is_stuffed(ip))
 609                 dblock = iomap_bmap(mapping, lblock, &gfs2_iomap_ops);
 610
 611         gfs2_glock_dq_uninit(&i_gh);
 612
 613         return dblock;
 614 }
 615
 616 static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
 617 {
 618         struct gfs2_bufdata *bd;
 619
 620         lock_buffer(bh);
 621         gfs2_log_lock(sdp);
 622         clear_buffer_dirty(bh);
 623         bd = bh->b_private;
 624         if (bd) {
 625                 if (!list_empty(&bd->bd_list) && !buffer_pinned(bh))
 626                         list_del_init(&bd->bd_list);
 627                 else {
 628                         spin_lock(&sdp->sd_ail_lock);
 629                         gfs2_remove_from_journal(bh, REMOVE_JDATA);
 630                         spin_unlock(&sdp->sd_ail_lock);
 631                 }
 632         }
 633         bh->b_bdev = NULL;
 634         clear_buffer_mapped(bh);
 635         clear_buffer_req(bh);
 636         clear_buffer_new(bh);
 637         gfs2_log_unlock(sdp);
 638         unlock_buffer(bh);
 639 }
 640
 641 static void gfs2_invalidate_folio(struct folio *folio, size_t offset,
 642                                 size_t length)
 643 {
 644         struct gfs2_sbd *sdp = GFS2_SB(folio->mapping->host);
 645         size_t stop = offset + length;
 646         int partial_page = (offset || length < folio_size(folio));
 647         struct buffer_head *bh, *head;
 648         unsigned long pos = 0;
 649
 650         BUG_ON(!folio_test_locked(folio));
 651         if (!partial_page)
 652                 folio_clear_checked(folio);
 653         head = folio_buffers(folio);
 654         if (!head)
 655                 goto out;
 656
 657         bh = head;
 658         do {
 659                 if (pos + bh->b_size > stop)
 660                         return;
 661
 662                 if (offset <= pos)
 663                         gfs2_discard(sdp, bh);
 664                 pos += bh->b_size;
 665                 bh = bh->b_this_page;
 666         } while (bh != head);
 667 out:
 668         if (!partial_page)
 669                 filemap_release_folio(folio, 0);
 670 }
 671
 672 /**
 673  * gfs2_release_folio - free the metadata associated with a folio
 674  * @folio: the folio that's being released
 675  * @gfp_mask: passed from Linux VFS, ignored by us
 676  *
 677  * Calls try_to_free_buffers() to free the buffers and put the folio if the
 678  * buffers can be released.
 679  *
 680  * Returns: true if the folio was put or else false
 681  */
 682
 683 bool gfs2_release_folio(struct folio *folio, gfp_t gfp_mask)
 684 {
 685         struct address_space *mapping = folio->mapping;
 686         struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
 687         struct buffer_head *bh, *head;
 688         struct gfs2_bufdata *bd;
 689
 690         head = folio_buffers(folio);
 691         if (!head)
 692                 return false;
 693
 694         /*
 695          * mm accommodates an old ext3 case where clean folios might
 696          * not have had the dirty bit cleared.  Thus, it can send actual
 697          * dirty folios to ->release_folio() via shrink_active_list().
 698          *
 699          * As a workaround, we skip folios that contain dirty buffers
 700          * below.  Once ->release_folio isn't called on dirty folios
 701          * anymore, we can warn on dirty buffers like we used to here
 702          * again.
 703          */
 704
 705         gfs2_log_lock(sdp);
 706         bh = head;
 707         do {
 708                 if (atomic_read(&bh->b_count))
 709                         goto cannot_release;
 710                 bd = bh->b_private;
 711                 if (bd && bd->bd_tr)
 712                         goto cannot_release;
 713                 if (buffer_dirty(bh) || WARN_ON(buffer_pinned(bh)))
 714                         goto cannot_release;
 715                 bh = bh->b_this_page;
 716         } while (bh != head);
 717
 718         bh = head;
 719         do {
 720                 bd = bh->b_private;
 721                 if (bd) {
 722                         gfs2_assert_warn(sdp, bd->bd_bh == bh);
 723                         bd->bd_bh = NULL;
 724                         bh->b_private = NULL;
 725                         /*
 726                          * The bd may still be queued as a revoke, in which
 727                          * case we must not dequeue nor free it.
 728                          */
 729                         if (!bd->bd_blkno && !list_empty(&bd->bd_list))
 730                                 list_del_init(&bd->bd_list);
 731                         if (list_empty(&bd->bd_list))
 732                                 kmem_cache_free(gfs2_bufdata_cachep, bd);
 733                 }
 734
 735                 bh = bh->b_this_page;
 736         } while (bh != head);
 737         gfs2_log_unlock(sdp);
 738
 739         return try_to_free_buffers(folio);
 740
 741 cannot_release:
 742         gfs2_log_unlock(sdp);
 743         return false;
 744 }
 745
 746 static const struct address_space_operations gfs2_aops = {
 747         .writepages = gfs2_writepages,
 748         .read_folio = gfs2_read_folio,
 749         .readahead = gfs2_readahead,
 750         .dirty_folio = filemap_dirty_folio,
 751         .release_folio = iomap_release_folio,
 752         .invalidate_folio = iomap_invalidate_folio,
 753         .bmap = gfs2_bmap,
 754         .migrate_folio = filemap_migrate_folio,
 755         .is_partially_uptodate = iomap_is_partially_uptodate,
 756         .error_remove_page = generic_error_remove_page,
 757 };
 758
 759 static const struct address_space_operations gfs2_jdata_aops = {
 760         .writepage = gfs2_jdata_writepage,
 761         .writepages = gfs2_jdata_writepages,
 762         .read_folio = gfs2_read_folio,
 763         .readahead = gfs2_readahead,
 764         .dirty_folio = jdata_dirty_folio,
 765         .bmap = gfs2_bmap,
 766         .invalidate_folio = gfs2_invalidate_folio,
 767         .release_folio = gfs2_release_folio,
 768         .is_partially_uptodate = block_is_partially_uptodate,
 769         .error_remove_page = generic_error_remove_page,
 770 };
 771
 772 void gfs2_set_aops(struct inode *inode)
 773 {
 774         if (gfs2_is_jdata(GFS2_I(inode)))
 775                 inode->i_mapping->a_ops = &gfs2_jdata_aops;
 776         else
 777                 inode->i_mapping->a_ops = &gfs2_aops;
 778 }