| 1 | |
| 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/folio_batch.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 | |
| 41 | * gfs2_get_block_noalloc - Fills in a buffer head with details about a block
|
| 42 | * @inode: The inode
|
| 43 | * @lblock: The block number to look up
|
| 44 | * @bh_result: The buffer head to return the result in
|
| 45 | * @create: Non-zero if we may add block to the file
|
| 46 | *
|
| 47 | * Returns: errno
|
| 48 |
|
| 49 |
|
| 50 | static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock,
|
| 51 | struct buffer_head *bh_result, int create)
|
| 52 | {
|
| 53 | int error;
|
| 54 |
|
| 55 | error = gfs2_block_map(inode, lblock, bh_result, 0);
|
| 56 | if (error)
|
| 57 | return error;
|
| 58 | if (!buffer_mapped(bh_result))
|
| 59 | return -ENODATA;
|
| 60 | return 0;
|
| 61 | }
|
| 62 |
|
| 63 | |
| 64 | * gfs2_write_jdata_folio - gfs2 jdata-specific version of block_write_full_folio
|
| 65 | * @folio: The folio to write
|
| 66 | * @wbc: The writeback control
|
| 67 | *
|
| 68 | * This is the same as calling block_write_full_folio, but it also
|
| 69 | * writes pages outside of i_size
|
| 70 |
|
| 71 | static int gfs2_write_jdata_folio(struct folio *folio,
|
| 72 | struct writeback_control *wbc)
|
| 73 | {
|
| 74 | struct inode * const inode = folio->mapping->host;
|
| 75 | loff_t i_size = i_size_read(inode);
|
| 76 |
|
| 77 | |
| 78 | * The folio straddles i_size. It must be zeroed out on each and every
|
| 79 | * writepage invocation because it may be mmapped. "A file is mapped
|
| 80 | * in multiples of the page size. For a file that is not a multiple of
|
| 81 | * the page size, the remaining memory is zeroed when mapped, and
|
| 82 | * writes to that region are not written out to the file."
|
| 83 |
|
| 84 | if (folio_pos(folio) < i_size && i_size < folio_next_pos(folio))
|
| 85 | folio_zero_segment(folio, offset_in_folio(folio, i_size),
|
| 86 | folio_size(folio));
|
| 87 |
|
| 88 | return __block_write_full_folio(inode, folio, gfs2_get_block_noalloc,
|
| 89 | wbc);
|
| 90 | }
|
| 91 |
|
| 92 | |
| 93 | * __gfs2_jdata_write_folio - The core of jdata writepage
|
| 94 | * @folio: The folio to write
|
| 95 | * @wbc: The writeback control
|
| 96 | *
|
| 97 | * Implements the core of write back. If a transaction is required then
|
| 98 | * the checked flag will have been set and the transaction will have
|
| 99 | * already been started before this is called.
|
| 100 |
|
| 101 | static int __gfs2_jdata_write_folio(struct folio *folio,
|
| 102 | struct writeback_control *wbc)
|
| 103 | {
|
| 104 | struct inode *inode = folio->mapping->host;
|
| 105 | struct gfs2_glock *gl = gfs2_inode_glock(inode);
|
| 106 |
|
| 107 | if (folio_test_checked(folio)) {
|
| 108 | folio_clear_checked(folio);
|
| 109 | if (!folio_buffers(folio)) {
|
| 110 | create_empty_buffers(folio,
|
| 111 | inode->i_sb->s_blocksize,
|
| 112 | BIT(BH_Dirty)|BIT(BH_Uptodate));
|
| 113 | }
|
| 114 | gfs2_trans_add_databufs(gl, folio, 0, folio_size(folio));
|
| 115 | }
|
| 116 | return gfs2_write_jdata_folio(folio, wbc);
|
| 117 | }
|
| 118 |
|
| 119 | |
| 120 | * gfs2_jdata_writeback - Write jdata folios to the log
|
| 121 | * @mapping: The mapping to write
|
| 122 | * @wbc: The writeback control
|
| 123 | *
|
| 124 | * Returns: errno
|
| 125 |
|
| 126 | int gfs2_jdata_writeback(struct address_space *mapping, struct writeback_control *wbc)
|
| 127 | {
|
| 128 | struct inode *inode = mapping->host;
|
| 129 | struct gfs2_glock *gl = gfs2_inode_glock(inode);
|
| 130 | struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
|
| 131 | struct folio *folio = NULL;
|
| 132 | int error;
|
| 133 |
|
| 134 | BUG_ON(current->journal_info);
|
| 135 | if (gfs2_assert_withdraw(sdp, gl->gl_state == LM_ST_EXCLUSIVE))
|
| 136 | return 0;
|
| 137 |
|
| 138 | while ((folio = writeback_iter(mapping, wbc, folio, &error))) {
|
| 139 | if (folio_test_checked(folio)) {
|
| 140 | folio_redirty_for_writepage(wbc, folio);
|
| 141 | folio_unlock(folio);
|
| 142 | continue;
|
| 143 | }
|
| 144 | error = __gfs2_jdata_write_folio(folio, wbc);
|
| 145 | }
|
| 146 |
|
| 147 | return error;
|
| 148 | }
|
| 149 |
|
| 150 | |
| 151 | * gfs2_writepages - Write a bunch of dirty pages back to disk
|
| 152 | * @mapping: The mapping to write
|
| 153 | * @wbc: Write-back control
|
| 154 | *
|
| 155 | * Used for both ordered and writeback modes.
|
| 156 |
|
| 157 | static int gfs2_writepages(struct address_space *mapping,
|
| 158 | struct writeback_control *wbc)
|
| 159 | {
|
| 160 | struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
|
| 161 | long initial_nr_to_write = wbc->nr_to_write;
|
| 162 | struct iomap_writepage_ctx wpc = {
|
| 163 | .inode = mapping->host,
|
| 164 | .wbc = wbc,
|
| 165 | .ops = &gfs2_writeback_ops,
|
| 166 | };
|
| 167 | int ret;
|
| 168 |
|
| 169 | |
| 170 | * Even if we didn't write any pages here, we might still be holding
|
| 171 | * dirty pages in the ail. We forcibly flush the ail because we don't
|
| 172 | * want balance_dirty_pages() to loop indefinitely trying to write out
|
| 173 | * pages held in the ail that it can't find.
|
| 174 |
|
| 175 | ret = iomap_writepages(&wpc);
|
| 176 | if (ret == 0 && wbc->nr_to_write == initial_nr_to_write)
|
| 177 | set_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags);
|
| 178 | return ret;
|
| 179 | }
|
| 180 |
|
| 181 | |
| 182 | * gfs2_write_jdata_batch - Write back a folio batch's worth of folios
|
| 183 | * @mapping: The mapping
|
| 184 | * @wbc: The writeback control
|
| 185 | * @fbatch: The batch of folios
|
| 186 | * @done_index: Page index
|
| 187 | *
|
| 188 | * Returns: non-zero if loop should terminate, zero otherwise
|
| 189 |
|
| 190 |
|
| 191 | static int gfs2_write_jdata_batch(struct address_space *mapping,
|
| 192 | struct writeback_control *wbc,
|
| 193 | struct folio_batch *fbatch,
|
| 194 | pgoff_t *done_index)
|
| 195 | {
|
| 196 | struct inode *inode = mapping->host;
|
| 197 | struct gfs2_sbd *sdp = GFS2_SB(inode);
|
| 198 | unsigned nrblocks;
|
| 199 | int i;
|
| 200 | int ret;
|
| 201 | size_t size = 0;
|
| 202 | int nr_folios = folio_batch_count(fbatch);
|
| 203 |
|
| 204 | for (i = 0; i < nr_folios; i++)
|
| 205 | size += folio_size(fbatch->folios[i]);
|
| 206 | nrblocks = size >> inode->i_blkbits;
|
| 207 |
|
| 208 | ret = gfs2_trans_begin(sdp, nrblocks, nrblocks);
|
| 209 | if (ret < 0)
|
| 210 | return ret;
|
| 211 |
|
| 212 | for (i = 0; i < nr_folios; i++) {
|
| 213 | struct folio *folio = fbatch->folios[i];
|
| 214 |
|
| 215 | *done_index = folio->index;
|
| 216 |
|
| 217 | folio_lock(folio);
|
| 218 |
|
| 219 | if (unlikely(folio->mapping != mapping)) {
|
| 220 | continue_unlock:
|
| 221 | folio_unlock(folio);
|
| 222 | continue;
|
| 223 | }
|
| 224 |
|
| 225 | if (!folio_test_dirty(folio)) {
|
| 226 |
|
| 227 | goto continue_unlock;
|
| 228 | }
|
| 229 |
|
| 230 | if (folio_test_writeback(folio)) {
|
| 231 | if (wbc->sync_mode != WB_SYNC_NONE)
|
| 232 | folio_wait_writeback(folio);
|
| 233 | else
|
| 234 | goto continue_unlock;
|
| 235 | }
|
| 236 |
|
| 237 | BUG_ON(folio_test_writeback(folio));
|
| 238 | if (!folio_clear_dirty_for_io(folio))
|
| 239 | goto continue_unlock;
|
| 240 |
|
| 241 | trace_wbc_writepage(wbc, inode_to_bdi(inode));
|
| 242 |
|
| 243 | ret = __gfs2_jdata_write_folio(folio, wbc);
|
| 244 | if (unlikely(ret)) {
|
| 245 | |
| 246 | * done_index is set past this page, so media errors
|
| 247 | * will not choke background writeout for the entire
|
| 248 | * file. This has consequences for range_cyclic
|
| 249 | * semantics (ie. it may not be suitable for data
|
| 250 | * integrity writeout).
|
| 251 |
|
| 252 | *done_index = folio_next_index(folio);
|
| 253 | ret = 1;
|
| 254 | break;
|
| 255 | }
|
| 256 |
|
| 257 | |
| 258 | * We stop writing back only if we are not doing
|
| 259 | * integrity sync. In case of integrity sync we have to
|
| 260 | * keep going until we have written all the pages
|
| 261 | * we tagged for writeback prior to entering this loop.
|
| 262 |
|
| 263 | if (--wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) {
|
| 264 | ret = 1;
|
| 265 | break;
|
| 266 | }
|
| 267 |
|
| 268 | }
|
| 269 | gfs2_trans_end(sdp);
|
| 270 | return ret;
|
| 271 | }
|
| 272 |
|
| 273 | |
| 274 | * gfs2_write_cache_jdata - Like write_cache_pages but different
|
| 275 | * @mapping: The mapping to write
|
| 276 | * @wbc: The writeback control
|
| 277 | *
|
| 278 | * The reason that we use our own function here is that we need to
|
| 279 | * start transactions before we grab page locks. This allows us
|
| 280 | * to get the ordering right.
|
| 281 |
|
| 282 |
|
| 283 | static int gfs2_write_cache_jdata(struct address_space *mapping,
|
| 284 | struct writeback_control *wbc)
|
| 285 | {
|
| 286 | int ret = 0;
|
| 287 | int done = 0;
|
| 288 | struct folio_batch fbatch;
|
| 289 | int nr_folios;
|
| 290 | pgoff_t writeback_index;
|
| 291 | pgoff_t index;
|
| 292 | pgoff_t end;
|
| 293 | pgoff_t done_index;
|
| 294 | int cycled;
|
| 295 | int range_whole = 0;
|
| 296 | xa_mark_t tag;
|
| 297 |
|
| 298 | folio_batch_init(&fbatch);
|
| 299 | if (wbc->range_cyclic) {
|
| 300 | writeback_index = mapping->writeback_index;
|
| 301 | index = writeback_index;
|
| 302 | if (index == 0)
|
| 303 | cycled = 1;
|
| 304 | else
|
| 305 | cycled = 0;
|
| 306 | end = -1;
|
| 307 | } else {
|
| 308 | index = wbc->range_start >> PAGE_SHIFT;
|
| 309 | end = wbc->range_end >> PAGE_SHIFT;
|
| 310 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
|
| 311 | range_whole = 1;
|
| 312 | cycled = 1;
|
| 313 | }
|
| 314 | tag = wbc_to_tag(wbc);
|
| 315 |
|
| 316 | retry:
|
| 317 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
|
| 318 | tag_pages_for_writeback(mapping, index, end);
|
| 319 | done_index = index;
|
| 320 | while (!done && (index <= end)) {
|
| 321 | nr_folios = filemap_get_folios_tag(mapping, &index, end,
|
| 322 | tag, &fbatch);
|
| 323 | if (nr_folios == 0)
|
| 324 | break;
|
| 325 |
|
| 326 | ret = gfs2_write_jdata_batch(mapping, wbc, &fbatch,
|
| 327 | &done_index);
|
| 328 | if (ret)
|
| 329 | done = 1;
|
| 330 | if (ret > 0)
|
| 331 | ret = 0;
|
| 332 | folio_batch_release(&fbatch);
|
| 333 | cond_resched();
|
| 334 | }
|
| 335 |
|
| 336 | if (!cycled && !done) {
|
| 337 | |
| 338 | * range_cyclic:
|
| 339 | * We hit the last page and there is more work to be done: wrap
|
| 340 | * back to the start of the file
|
| 341 |
|
| 342 | cycled = 1;
|
| 343 | index = 0;
|
| 344 | end = writeback_index - 1;
|
| 345 | goto retry;
|
| 346 | }
|
| 347 |
|
| 348 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
|
| 349 | mapping->writeback_index = done_index;
|
| 350 |
|
| 351 | return ret;
|
| 352 | }
|
| 353 |
|
| 354 |
|
| 355 | |
| 356 | * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
|
| 357 | * @mapping: The mapping to write
|
| 358 | * @wbc: The writeback control
|
| 359 | *
|
| 360 |
|
| 361 |
|
| 362 | static int gfs2_jdata_writepages(struct address_space *mapping,
|
| 363 | struct writeback_control *wbc)
|
| 364 | {
|
| 365 | struct gfs2_glock *gl = gfs2_inode_glock(mapping->host);
|
| 366 | struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
|
| 367 | int ret;
|
| 368 |
|
| 369 | ret = gfs2_write_cache_jdata(mapping, wbc);
|
| 370 | if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) {
|
| 371 | gfs2_log_flush(sdp, gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
|
| 372 | GFS2_LFC_JDATA_WPAGES);
|
| 373 | ret = gfs2_write_cache_jdata(mapping, wbc);
|
| 374 | }
|
| 375 | return ret;
|
| 376 | }
|
| 377 |
|
| 378 | |
| 379 | * stuffed_read_folio - Fill in a Linux folio with stuffed file data
|
| 380 | * @ip: the inode
|
| 381 | * @folio: the folio
|
| 382 | *
|
| 383 | * Returns: errno
|
| 384 |
|
| 385 | static int stuffed_read_folio(struct gfs2_inode *ip, struct folio *folio)
|
| 386 | {
|
| 387 | struct buffer_head *dibh = NULL;
|
| 388 | size_t dsize = i_size_read(&ip->i_inode);
|
| 389 | void *from = NULL;
|
| 390 | int error = 0;
|
| 391 |
|
| 392 | |
| 393 | * Due to the order of unstuffing files and ->fault(), we can be
|
| 394 | * asked for a zero folio in the case of a stuffed file being extended,
|
| 395 | * so we need to supply one here. It doesn't happen often.
|
| 396 |
|
| 397 | if (unlikely(folio->index)) {
|
| 398 | dsize = 0;
|
| 399 | } else {
|
| 400 | error = gfs2_meta_inode_buffer(ip, &dibh);
|
| 401 | if (error)
|
| 402 | goto out;
|
| 403 | from = dibh->b_data + sizeof(struct gfs2_dinode);
|
| 404 | }
|
| 405 |
|
| 406 | folio_fill_tail(folio, 0, from, dsize);
|
| 407 | brelse(dibh);
|
| 408 | out:
|
| 409 | folio_end_read(folio, error == 0);
|
| 410 |
|
| 411 | return error;
|
| 412 | }
|
| 413 |
|
| 414 | |
| 415 | * gfs2_read_folio - read a folio from a file
|
| 416 | * @file: The file to read
|
| 417 | * @folio: The folio in the file
|
| 418 |
|
| 419 | static int gfs2_read_folio(struct file *file, struct folio *folio)
|
| 420 | {
|
| 421 | struct inode *inode = folio->mapping->host;
|
| 422 | struct gfs2_inode *ip = GFS2_I(inode);
|
| 423 | struct gfs2_sbd *sdp = GFS2_SB(inode);
|
| 424 | int error = 0;
|
| 425 |
|
| 426 | if (!gfs2_is_jdata(ip) ||
|
| 427 | (i_blocksize(inode) == PAGE_SIZE && !folio_buffers(folio))) {
|
| 428 | iomap_bio_read_folio(folio, &gfs2_iomap_ops);
|
| 429 | } else if (gfs2_is_stuffed(ip)) {
|
| 430 | error = stuffed_read_folio(ip, folio);
|
| 431 | } else {
|
| 432 | error = mpage_read_folio(folio, gfs2_block_map);
|
| 433 | }
|
| 434 |
|
| 435 | if (gfs2_withdrawn(sdp))
|
| 436 | return -EIO;
|
| 437 |
|
| 438 | return error;
|
| 439 | }
|
| 440 |
|
| 441 | |
| 442 | * gfs2_internal_read - read an internal file
|
| 443 | * @ip: The gfs2 inode
|
| 444 | * @buf: The buffer to fill
|
| 445 | * @pos: The file position
|
| 446 | * @size: The amount to read
|
| 447 | *
|
| 448 |
|
| 449 |
|
| 450 | ssize_t gfs2_internal_read(struct gfs2_inode *ip, char *buf, loff_t *pos,
|
| 451 | size_t size)
|
| 452 | {
|
| 453 | struct address_space *mapping = ip->i_inode.i_mapping;
|
| 454 | unsigned long index = *pos >> PAGE_SHIFT;
|
| 455 | size_t copied = 0;
|
| 456 |
|
| 457 | do {
|
| 458 | size_t offset, chunk;
|
| 459 | struct folio *folio;
|
| 460 |
|
| 461 | folio = read_cache_folio(mapping, index, gfs2_read_folio, NULL);
|
| 462 | if (IS_ERR(folio)) {
|
| 463 | if (PTR_ERR(folio) == -EINTR)
|
| 464 | continue;
|
| 465 | return PTR_ERR(folio);
|
| 466 | }
|
| 467 | offset = *pos + copied - folio_pos(folio);
|
| 468 | chunk = min(size - copied, folio_size(folio) - offset);
|
| 469 | memcpy_from_folio(buf + copied, folio, offset, chunk);
|
| 470 | index = folio_next_index(folio);
|
| 471 | folio_put(folio);
|
| 472 | copied += chunk;
|
| 473 | } while(copied < size);
|
| 474 | (*pos) += size;
|
| 475 | return size;
|
| 476 | }
|
| 477 |
|
| 478 | |
| 479 | * gfs2_readahead - Read a bunch of pages at once
|
| 480 | * @rac: Read-ahead control structure
|
| 481 | *
|
| 482 | * Some notes:
|
| 483 | * 1. This is only for readahead, so we can simply ignore any things
|
| 484 | * which are slightly inconvenient (such as locking conflicts between
|
| 485 | * the page lock and the glock) and return having done no I/O. Its
|
| 486 | * obviously not something we'd want to do on too regular a basis.
|
| 487 | * Any I/O we ignore at this time will be done via readpage later.
|
| 488 | * 2. We don't handle stuffed files here we let readpage do the honours.
|
| 489 | * 3. mpage_readahead() does most of the heavy lifting in the common case.
|
| 490 | * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
|
| 491 |
|
| 492 |
|
| 493 | static void gfs2_readahead(struct readahead_control *rac)
|
| 494 | {
|
| 495 | struct inode *inode = rac->mapping->host;
|
| 496 | struct gfs2_inode *ip = GFS2_I(inode);
|
| 497 |
|
| 498 | if (gfs2_is_stuffed(ip))
|
| 499 | ;
|
| 500 | else if (gfs2_is_jdata(ip))
|
| 501 | mpage_readahead(rac, gfs2_block_map);
|
| 502 | else
|
| 503 | iomap_bio_readahead(rac, &gfs2_iomap_ops);
|
| 504 | }
|
| 505 |
|
| 506 | |
| 507 | * adjust_fs_space - Adjusts the free space available due to gfs2_grow
|
| 508 | * @inode: the rindex inode
|
| 509 |
|
| 510 | void adjust_fs_space(struct inode *inode)
|
| 511 | {
|
| 512 | struct gfs2_sbd *sdp = GFS2_SB(inode);
|
| 513 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
|
| 514 | struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
|
| 515 | struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
|
| 516 | struct buffer_head *m_bh;
|
| 517 | u64 fs_total, new_free;
|
| 518 |
|
| 519 | if (gfs2_trans_begin(sdp, 2 * RES_STATFS, 0) != 0)
|
| 520 | return;
|
| 521 |
|
| 522 |
|
| 523 | fs_total = gfs2_ri_total(sdp);
|
| 524 | if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0)
|
| 525 | goto out;
|
| 526 |
|
| 527 | spin_lock(&sdp->sd_statfs_spin);
|
| 528 | gfs2_statfs_change_in(m_sc, m_bh->b_data +
|
| 529 | sizeof(struct gfs2_dinode));
|
| 530 | if (fs_total > (m_sc->sc_total + l_sc->sc_total))
|
| 531 | new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
|
| 532 | else
|
| 533 | new_free = 0;
|
| 534 | spin_unlock(&sdp->sd_statfs_spin);
|
| 535 | fs_warn(sdp, "File system extended by %llu blocks.\n",
|
| 536 | (unsigned long long)new_free);
|
| 537 | gfs2_statfs_change(sdp, new_free, new_free, 0);
|
| 538 |
|
| 539 | update_statfs(sdp, m_bh);
|
| 540 | brelse(m_bh);
|
| 541 | out:
|
| 542 | sdp->sd_rindex_uptodate = 0;
|
| 543 | gfs2_trans_end(sdp);
|
| 544 | }
|
| 545 |
|
| 546 | static bool gfs2_jdata_dirty_folio(struct address_space *mapping,
|
| 547 | struct folio *folio)
|
| 548 | {
|
| 549 | if (current->journal_info)
|
| 550 | folio_set_checked(folio);
|
| 551 | return block_dirty_folio(mapping, folio);
|
| 552 | }
|
| 553 |
|
| 554 | |
| 555 | * gfs2_bmap - Block map function
|
| 556 | * @mapping: Address space info
|
| 557 | * @lblock: The block to map
|
| 558 | *
|
| 559 | * Returns: The disk address for the block or 0 on hole or error
|
| 560 |
|
| 561 |
|
| 562 | static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
|
| 563 | {
|
| 564 | struct gfs2_glock *gl = gfs2_inode_glock(mapping->host);
|
| 565 | struct gfs2_inode *ip = GFS2_I(mapping->host);
|
| 566 | struct gfs2_holder i_gh;
|
| 567 | sector_t dblock = 0;
|
| 568 | int error;
|
| 569 |
|
| 570 | error = gfs2_glock_nq_init(gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
|
| 571 | if (error)
|
| 572 | return 0;
|
| 573 |
|
| 574 | if (!gfs2_is_stuffed(ip))
|
| 575 | dblock = iomap_bmap(mapping, lblock, &gfs2_iomap_ops);
|
| 576 |
|
| 577 | gfs2_glock_dq_uninit(&i_gh);
|
| 578 |
|
| 579 | return dblock;
|
| 580 | }
|
| 581 |
|
| 582 | static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
|
| 583 | {
|
| 584 | struct gfs2_bufdata *bd;
|
| 585 |
|
| 586 | lock_buffer(bh);
|
| 587 | spin_lock(&sdp->sd_log_lock);
|
| 588 | clear_buffer_dirty(bh);
|
| 589 | bd = bh->b_private;
|
| 590 | if (bd) {
|
| 591 | if (!list_empty(&bd->bd_list) && !buffer_pinned(bh))
|
| 592 | list_del_init(&bd->bd_list);
|
| 593 | else {
|
| 594 | spin_lock(&sdp->sd_ail_lock);
|
| 595 | gfs2_remove_from_journal(bh, REMOVE_JDATA);
|
| 596 | spin_unlock(&sdp->sd_ail_lock);
|
| 597 | }
|
| 598 | }
|
| 599 | bh->b_bdev = NULL;
|
| 600 | clear_buffer_mapped(bh);
|
| 601 | clear_buffer_req(bh);
|
| 602 | clear_buffer_new(bh);
|
| 603 | spin_unlock(&sdp->sd_log_lock);
|
| 604 | unlock_buffer(bh);
|
| 605 | }
|
| 606 |
|
| 607 | static void gfs2_invalidate_folio(struct folio *folio, size_t offset,
|
| 608 | size_t length)
|
| 609 | {
|
| 610 | struct gfs2_sbd *sdp = GFS2_SB(folio->mapping->host);
|
| 611 | size_t stop = offset + length;
|
| 612 | int partial_page = (offset || length < folio_size(folio));
|
| 613 | struct buffer_head *bh, *head;
|
| 614 | unsigned long pos = 0;
|
| 615 |
|
| 616 | BUG_ON(!folio_test_locked(folio));
|
| 617 | if (!partial_page)
|
| 618 | folio_clear_checked(folio);
|
| 619 | head = folio_buffers(folio);
|
| 620 | if (!head)
|
| 621 | goto out;
|
| 622 |
|
| 623 | bh = head;
|
| 624 | do {
|
| 625 | if (pos + bh->b_size > stop)
|
| 626 | return;
|
| 627 |
|
| 628 | if (offset <= pos)
|
| 629 | gfs2_discard(sdp, bh);
|
| 630 | pos += bh->b_size;
|
| 631 | bh = bh->b_this_page;
|
| 632 | } while (bh != head);
|
| 633 | out:
|
| 634 | if (!partial_page)
|
| 635 | filemap_release_folio(folio, 0);
|
| 636 | }
|
| 637 |
|
| 638 | |
| 639 | * gfs2_release_folio - free the metadata associated with a folio
|
| 640 | * @folio: the folio that's being released
|
| 641 | * @gfp_mask: passed from Linux VFS, ignored by us
|
| 642 | *
|
| 643 | * Calls try_to_free_buffers() to free the buffers and put the folio if the
|
| 644 | * buffers can be released.
|
| 645 | *
|
| 646 | * Returns: true if the folio was put or else false
|
| 647 |
|
| 648 |
|
| 649 | bool gfs2_release_folio(struct folio *folio, gfp_t gfp_mask)
|
| 650 | {
|
| 651 | struct address_space *mapping = folio->mapping;
|
| 652 | struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
|
| 653 | struct buffer_head *bh, *head;
|
| 654 | struct gfs2_bufdata *bd;
|
| 655 |
|
| 656 | head = folio_buffers(folio);
|
| 657 | if (!head)
|
| 658 | return false;
|
| 659 |
|
| 660 | |
| 661 | * mm accommodates an old ext3 case where clean folios might
|
| 662 | * not have had the dirty bit cleared. Thus, it can send actual
|
| 663 | * dirty folios to ->release_folio() via shrink_active_list().
|
| 664 | *
|
| 665 | * As a workaround, we skip folios that contain dirty buffers
|
| 666 | * below. Once ->release_folio isn't called on dirty folios
|
| 667 | * anymore, we can warn on dirty buffers like we used to here
|
| 668 | * again.
|
| 669 |
|
| 670 |
|
| 671 | spin_lock(&sdp->sd_log_lock);
|
| 672 | bh = head;
|
| 673 | do {
|
| 674 | if (atomic_read(&bh->b_count))
|
| 675 | goto cannot_release;
|
| 676 | bd = bh->b_private;
|
| 677 | if (bd && bd->bd_tr)
|
| 678 | goto cannot_release;
|
| 679 | if (buffer_dirty(bh) || WARN_ON(buffer_pinned(bh)))
|
| 680 | goto cannot_release;
|
| 681 | bh = bh->b_this_page;
|
| 682 | } while (bh != head);
|
| 683 |
|
| 684 | bh = head;
|
| 685 | do {
|
| 686 | bd = bh->b_private;
|
| 687 | if (bd) {
|
| 688 | gfs2_assert_warn(sdp, bd->bd_bh == bh);
|
| 689 | bd->bd_bh = NULL;
|
| 690 | bh->b_private = NULL;
|
| 691 | |
| 692 | * The bd may still be queued as a revoke, in which
|
| 693 | * case we must not dequeue nor free it.
|
| 694 |
|
| 695 | if (!bd->bd_blkno && !list_empty(&bd->bd_list))
|
| 696 | list_del_init(&bd->bd_list);
|
| 697 | if (list_empty(&bd->bd_list))
|
| 698 | kmem_cache_free(gfs2_bufdata_cachep, bd);
|
| 699 | }
|
| 700 |
|
| 701 | bh = bh->b_this_page;
|
| 702 | } while (bh != head);
|
| 703 | spin_unlock(&sdp->sd_log_lock);
|
| 704 |
|
| 705 | return try_to_free_buffers(folio);
|
| 706 |
|
| 707 | cannot_release:
|
| 708 | spin_unlock(&sdp->sd_log_lock);
|
| 709 | return false;
|
| 710 | }
|
| 711 |
|
| 712 | static const struct address_space_operations gfs2_aops = {
|
| 713 | .writepages = gfs2_writepages,
|
| 714 | .read_folio = gfs2_read_folio,
|
| 715 | .readahead = gfs2_readahead,
|
| 716 | .dirty_folio = iomap_dirty_folio,
|
| 717 | .release_folio = iomap_release_folio,
|
| 718 | .invalidate_folio = iomap_invalidate_folio,
|
| 719 | .bmap = gfs2_bmap,
|
| 720 | .migrate_folio = filemap_migrate_folio,
|
| 721 | .is_partially_uptodate = iomap_is_partially_uptodate,
|
| 722 | .error_remove_folio = generic_error_remove_folio,
|
| 723 | };
|
| 724 |
|
| 725 | static const struct address_space_operations gfs2_jdata_aops = {
|
| 726 | .writepages = gfs2_jdata_writepages,
|
| 727 | .read_folio = gfs2_read_folio,
|
| 728 | .readahead = gfs2_readahead,
|
| 729 | .dirty_folio = gfs2_jdata_dirty_folio,
|
| 730 | .bmap = gfs2_bmap,
|
| 731 | .migrate_folio = buffer_migrate_folio,
|
| 732 | .invalidate_folio = gfs2_invalidate_folio,
|
| 733 | .release_folio = gfs2_release_folio,
|
| 734 | .is_partially_uptodate = block_is_partially_uptodate,
|
| 735 | .error_remove_folio = generic_error_remove_folio,
|
| 736 | };
|
| 737 |
|
| 738 | void gfs2_set_aops(struct inode *inode)
|
| 739 | {
|
| 740 | if (gfs2_is_jdata(GFS2_I(inode)))
|
| 741 | inode->i_mapping->a_ops = &gfs2_jdata_aops;
|
| 742 | else
|
| 743 | inode->i_mapping->a_ops = &gfs2_aops;
|
| 744 | }
|