| 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, ip->i_gl->gl_state == LM_ST_EXCLUSIVE)) |
| 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 enough 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 && wbc->nr_to_write > 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 | size_t size = 0; |
| 218 | int nr_folios = folio_batch_count(fbatch); |
| 219 | |
| 220 | for (i = 0; i < nr_folios; i++) |
| 221 | size += folio_size(fbatch->folios[i]); |
| 222 | nrblocks = 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_next_index(folio); |
| 276 | ret = 1; |
| 277 | break; |
| 278 | } |
| 279 | } |
| 280 | |
| 281 | /* |
| 282 | * We stop writing back only if we are not doing |
| 283 | * integrity sync. In case of integrity sync we have to |
| 284 | * keep going until we have written all the pages |
| 285 | * we tagged for writeback prior to entering this loop. |
| 286 | */ |
| 287 | if (--wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) { |
| 288 | ret = 1; |
| 289 | break; |
| 290 | } |
| 291 | |
| 292 | } |
| 293 | gfs2_trans_end(sdp); |
| 294 | return ret; |
| 295 | } |
| 296 | |
| 297 | /** |
| 298 | * gfs2_write_cache_jdata - Like write_cache_pages but different |
| 299 | * @mapping: The mapping to write |
| 300 | * @wbc: The writeback control |
| 301 | * |
| 302 | * The reason that we use our own function here is that we need to |
| 303 | * start transactions before we grab page locks. This allows us |
| 304 | * to get the ordering right. |
| 305 | */ |
| 306 | |
| 307 | static int gfs2_write_cache_jdata(struct address_space *mapping, |
| 308 | struct writeback_control *wbc) |
| 309 | { |
| 310 | int ret = 0; |
| 311 | int done = 0; |
| 312 | struct folio_batch fbatch; |
| 313 | int nr_folios; |
| 314 | pgoff_t writeback_index; |
| 315 | pgoff_t index; |
| 316 | pgoff_t end; |
| 317 | pgoff_t done_index; |
| 318 | int cycled; |
| 319 | int range_whole = 0; |
| 320 | xa_mark_t tag; |
| 321 | |
| 322 | folio_batch_init(&fbatch); |
| 323 | if (wbc->range_cyclic) { |
| 324 | writeback_index = mapping->writeback_index; /* prev offset */ |
| 325 | index = writeback_index; |
| 326 | if (index == 0) |
| 327 | cycled = 1; |
| 328 | else |
| 329 | cycled = 0; |
| 330 | end = -1; |
| 331 | } else { |
| 332 | index = wbc->range_start >> PAGE_SHIFT; |
| 333 | end = wbc->range_end >> PAGE_SHIFT; |
| 334 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
| 335 | range_whole = 1; |
| 336 | cycled = 1; /* ignore range_cyclic tests */ |
| 337 | } |
| 338 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
| 339 | tag = PAGECACHE_TAG_TOWRITE; |
| 340 | else |
| 341 | tag = PAGECACHE_TAG_DIRTY; |
| 342 | |
| 343 | retry: |
| 344 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
| 345 | tag_pages_for_writeback(mapping, index, end); |
| 346 | done_index = index; |
| 347 | while (!done && (index <= end)) { |
| 348 | nr_folios = filemap_get_folios_tag(mapping, &index, end, |
| 349 | tag, &fbatch); |
| 350 | if (nr_folios == 0) |
| 351 | break; |
| 352 | |
| 353 | ret = gfs2_write_jdata_batch(mapping, wbc, &fbatch, |
| 354 | &done_index); |
| 355 | if (ret) |
| 356 | done = 1; |
| 357 | if (ret > 0) |
| 358 | ret = 0; |
| 359 | folio_batch_release(&fbatch); |
| 360 | cond_resched(); |
| 361 | } |
| 362 | |
| 363 | if (!cycled && !done) { |
| 364 | /* |
| 365 | * range_cyclic: |
| 366 | * We hit the last page and there is more work to be done: wrap |
| 367 | * back to the start of the file |
| 368 | */ |
| 369 | cycled = 1; |
| 370 | index = 0; |
| 371 | end = writeback_index - 1; |
| 372 | goto retry; |
| 373 | } |
| 374 | |
| 375 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
| 376 | mapping->writeback_index = done_index; |
| 377 | |
| 378 | return ret; |
| 379 | } |
| 380 | |
| 381 | |
| 382 | /** |
| 383 | * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk |
| 384 | * @mapping: The mapping to write |
| 385 | * @wbc: The writeback control |
| 386 | * |
| 387 | */ |
| 388 | |
| 389 | static int gfs2_jdata_writepages(struct address_space *mapping, |
| 390 | struct writeback_control *wbc) |
| 391 | { |
| 392 | struct gfs2_inode *ip = GFS2_I(mapping->host); |
| 393 | struct gfs2_sbd *sdp = GFS2_SB(mapping->host); |
| 394 | int ret; |
| 395 | |
| 396 | ret = gfs2_write_cache_jdata(mapping, wbc); |
| 397 | if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) { |
| 398 | gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL | |
| 399 | GFS2_LFC_JDATA_WPAGES); |
| 400 | ret = gfs2_write_cache_jdata(mapping, wbc); |
| 401 | } |
| 402 | return ret; |
| 403 | } |
| 404 | |
| 405 | /** |
| 406 | * stuffed_readpage - Fill in a Linux page with stuffed file data |
| 407 | * @ip: the inode |
| 408 | * @page: the page |
| 409 | * |
| 410 | * Returns: errno |
| 411 | */ |
| 412 | static int stuffed_readpage(struct gfs2_inode *ip, struct page *page) |
| 413 | { |
| 414 | struct buffer_head *dibh; |
| 415 | u64 dsize = i_size_read(&ip->i_inode); |
| 416 | void *kaddr; |
| 417 | int error; |
| 418 | |
| 419 | /* |
| 420 | * Due to the order of unstuffing files and ->fault(), we can be |
| 421 | * asked for a zero page in the case of a stuffed file being extended, |
| 422 | * so we need to supply one here. It doesn't happen often. |
| 423 | */ |
| 424 | if (unlikely(page->index)) { |
| 425 | zero_user(page, 0, PAGE_SIZE); |
| 426 | SetPageUptodate(page); |
| 427 | return 0; |
| 428 | } |
| 429 | |
| 430 | error = gfs2_meta_inode_buffer(ip, &dibh); |
| 431 | if (error) |
| 432 | return error; |
| 433 | |
| 434 | kaddr = kmap_local_page(page); |
| 435 | memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize); |
| 436 | memset(kaddr + dsize, 0, PAGE_SIZE - dsize); |
| 437 | kunmap_local(kaddr); |
| 438 | flush_dcache_page(page); |
| 439 | brelse(dibh); |
| 440 | SetPageUptodate(page); |
| 441 | |
| 442 | return 0; |
| 443 | } |
| 444 | |
| 445 | /** |
| 446 | * gfs2_read_folio - read a folio from a file |
| 447 | * @file: The file to read |
| 448 | * @folio: The folio in the file |
| 449 | */ |
| 450 | static int gfs2_read_folio(struct file *file, struct folio *folio) |
| 451 | { |
| 452 | struct inode *inode = folio->mapping->host; |
| 453 | struct gfs2_inode *ip = GFS2_I(inode); |
| 454 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
| 455 | int error; |
| 456 | |
| 457 | if (!gfs2_is_jdata(ip) || |
| 458 | (i_blocksize(inode) == PAGE_SIZE && !folio_buffers(folio))) { |
| 459 | error = iomap_read_folio(folio, &gfs2_iomap_ops); |
| 460 | } else if (gfs2_is_stuffed(ip)) { |
| 461 | error = stuffed_readpage(ip, &folio->page); |
| 462 | folio_unlock(folio); |
| 463 | } else { |
| 464 | error = mpage_read_folio(folio, gfs2_block_map); |
| 465 | } |
| 466 | |
| 467 | if (unlikely(gfs2_withdrawn(sdp))) |
| 468 | return -EIO; |
| 469 | |
| 470 | return error; |
| 471 | } |
| 472 | |
| 473 | /** |
| 474 | * gfs2_internal_read - read an internal file |
| 475 | * @ip: The gfs2 inode |
| 476 | * @buf: The buffer to fill |
| 477 | * @pos: The file position |
| 478 | * @size: The amount to read |
| 479 | * |
| 480 | */ |
| 481 | |
| 482 | int gfs2_internal_read(struct gfs2_inode *ip, char *buf, loff_t *pos, |
| 483 | unsigned size) |
| 484 | { |
| 485 | struct address_space *mapping = ip->i_inode.i_mapping; |
| 486 | unsigned long index = *pos >> PAGE_SHIFT; |
| 487 | unsigned offset = *pos & (PAGE_SIZE - 1); |
| 488 | unsigned copied = 0; |
| 489 | unsigned amt; |
| 490 | struct page *page; |
| 491 | |
| 492 | do { |
| 493 | page = read_cache_page(mapping, index, gfs2_read_folio, NULL); |
| 494 | if (IS_ERR(page)) { |
| 495 | if (PTR_ERR(page) == -EINTR) |
| 496 | continue; |
| 497 | return PTR_ERR(page); |
| 498 | } |
| 499 | amt = size - copied; |
| 500 | if (offset + size > PAGE_SIZE) |
| 501 | amt = PAGE_SIZE - offset; |
| 502 | memcpy_from_page(buf + copied, page, offset, amt); |
| 503 | put_page(page); |
| 504 | copied += amt; |
| 505 | index++; |
| 506 | offset = 0; |
| 507 | } while(copied < size); |
| 508 | (*pos) += size; |
| 509 | return size; |
| 510 | } |
| 511 | |
| 512 | /** |
| 513 | * gfs2_readahead - Read a bunch of pages at once |
| 514 | * @rac: Read-ahead control structure |
| 515 | * |
| 516 | * Some notes: |
| 517 | * 1. This is only for readahead, so we can simply ignore any things |
| 518 | * which are slightly inconvenient (such as locking conflicts between |
| 519 | * the page lock and the glock) and return having done no I/O. Its |
| 520 | * obviously not something we'd want to do on too regular a basis. |
| 521 | * Any I/O we ignore at this time will be done via readpage later. |
| 522 | * 2. We don't handle stuffed files here we let readpage do the honours. |
| 523 | * 3. mpage_readahead() does most of the heavy lifting in the common case. |
| 524 | * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places. |
| 525 | */ |
| 526 | |
| 527 | static void gfs2_readahead(struct readahead_control *rac) |
| 528 | { |
| 529 | struct inode *inode = rac->mapping->host; |
| 530 | struct gfs2_inode *ip = GFS2_I(inode); |
| 531 | |
| 532 | if (gfs2_is_stuffed(ip)) |
| 533 | ; |
| 534 | else if (gfs2_is_jdata(ip)) |
| 535 | mpage_readahead(rac, gfs2_block_map); |
| 536 | else |
| 537 | iomap_readahead(rac, &gfs2_iomap_ops); |
| 538 | } |
| 539 | |
| 540 | /** |
| 541 | * adjust_fs_space - Adjusts the free space available due to gfs2_grow |
| 542 | * @inode: the rindex inode |
| 543 | */ |
| 544 | void adjust_fs_space(struct inode *inode) |
| 545 | { |
| 546 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
| 547 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); |
| 548 | struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; |
| 549 | struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; |
| 550 | struct buffer_head *m_bh; |
| 551 | u64 fs_total, new_free; |
| 552 | |
| 553 | if (gfs2_trans_begin(sdp, 2 * RES_STATFS, 0) != 0) |
| 554 | return; |
| 555 | |
| 556 | /* Total up the file system space, according to the latest rindex. */ |
| 557 | fs_total = gfs2_ri_total(sdp); |
| 558 | if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0) |
| 559 | goto out; |
| 560 | |
| 561 | spin_lock(&sdp->sd_statfs_spin); |
| 562 | gfs2_statfs_change_in(m_sc, m_bh->b_data + |
| 563 | sizeof(struct gfs2_dinode)); |
| 564 | if (fs_total > (m_sc->sc_total + l_sc->sc_total)) |
| 565 | new_free = fs_total - (m_sc->sc_total + l_sc->sc_total); |
| 566 | else |
| 567 | new_free = 0; |
| 568 | spin_unlock(&sdp->sd_statfs_spin); |
| 569 | fs_warn(sdp, "File system extended by %llu blocks.\n", |
| 570 | (unsigned long long)new_free); |
| 571 | gfs2_statfs_change(sdp, new_free, new_free, 0); |
| 572 | |
| 573 | update_statfs(sdp, m_bh); |
| 574 | brelse(m_bh); |
| 575 | out: |
| 576 | sdp->sd_rindex_uptodate = 0; |
| 577 | gfs2_trans_end(sdp); |
| 578 | } |
| 579 | |
| 580 | static bool jdata_dirty_folio(struct address_space *mapping, |
| 581 | struct folio *folio) |
| 582 | { |
| 583 | if (current->journal_info) |
| 584 | folio_set_checked(folio); |
| 585 | return block_dirty_folio(mapping, folio); |
| 586 | } |
| 587 | |
| 588 | /** |
| 589 | * gfs2_bmap - Block map function |
| 590 | * @mapping: Address space info |
| 591 | * @lblock: The block to map |
| 592 | * |
| 593 | * Returns: The disk address for the block or 0 on hole or error |
| 594 | */ |
| 595 | |
| 596 | static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock) |
| 597 | { |
| 598 | struct gfs2_inode *ip = GFS2_I(mapping->host); |
| 599 | struct gfs2_holder i_gh; |
| 600 | sector_t dblock = 0; |
| 601 | int error; |
| 602 | |
| 603 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh); |
| 604 | if (error) |
| 605 | return 0; |
| 606 | |
| 607 | if (!gfs2_is_stuffed(ip)) |
| 608 | dblock = iomap_bmap(mapping, lblock, &gfs2_iomap_ops); |
| 609 | |
| 610 | gfs2_glock_dq_uninit(&i_gh); |
| 611 | |
| 612 | return dblock; |
| 613 | } |
| 614 | |
| 615 | static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh) |
| 616 | { |
| 617 | struct gfs2_bufdata *bd; |
| 618 | |
| 619 | lock_buffer(bh); |
| 620 | gfs2_log_lock(sdp); |
| 621 | clear_buffer_dirty(bh); |
| 622 | bd = bh->b_private; |
| 623 | if (bd) { |
| 624 | if (!list_empty(&bd->bd_list) && !buffer_pinned(bh)) |
| 625 | list_del_init(&bd->bd_list); |
| 626 | else { |
| 627 | spin_lock(&sdp->sd_ail_lock); |
| 628 | gfs2_remove_from_journal(bh, REMOVE_JDATA); |
| 629 | spin_unlock(&sdp->sd_ail_lock); |
| 630 | } |
| 631 | } |
| 632 | bh->b_bdev = NULL; |
| 633 | clear_buffer_mapped(bh); |
| 634 | clear_buffer_req(bh); |
| 635 | clear_buffer_new(bh); |
| 636 | gfs2_log_unlock(sdp); |
| 637 | unlock_buffer(bh); |
| 638 | } |
| 639 | |
| 640 | static void gfs2_invalidate_folio(struct folio *folio, size_t offset, |
| 641 | size_t length) |
| 642 | { |
| 643 | struct gfs2_sbd *sdp = GFS2_SB(folio->mapping->host); |
| 644 | size_t stop = offset + length; |
| 645 | int partial_page = (offset || length < folio_size(folio)); |
| 646 | struct buffer_head *bh, *head; |
| 647 | unsigned long pos = 0; |
| 648 | |
| 649 | BUG_ON(!folio_test_locked(folio)); |
| 650 | if (!partial_page) |
| 651 | folio_clear_checked(folio); |
| 652 | head = folio_buffers(folio); |
| 653 | if (!head) |
| 654 | goto out; |
| 655 | |
| 656 | bh = head; |
| 657 | do { |
| 658 | if (pos + bh->b_size > stop) |
| 659 | return; |
| 660 | |
| 661 | if (offset <= pos) |
| 662 | gfs2_discard(sdp, bh); |
| 663 | pos += bh->b_size; |
| 664 | bh = bh->b_this_page; |
| 665 | } while (bh != head); |
| 666 | out: |
| 667 | if (!partial_page) |
| 668 | filemap_release_folio(folio, 0); |
| 669 | } |
| 670 | |
| 671 | /** |
| 672 | * gfs2_release_folio - free the metadata associated with a folio |
| 673 | * @folio: the folio that's being released |
| 674 | * @gfp_mask: passed from Linux VFS, ignored by us |
| 675 | * |
| 676 | * Calls try_to_free_buffers() to free the buffers and put the folio if the |
| 677 | * buffers can be released. |
| 678 | * |
| 679 | * Returns: true if the folio was put or else false |
| 680 | */ |
| 681 | |
| 682 | bool gfs2_release_folio(struct folio *folio, gfp_t gfp_mask) |
| 683 | { |
| 684 | struct address_space *mapping = folio->mapping; |
| 685 | struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping); |
| 686 | struct buffer_head *bh, *head; |
| 687 | struct gfs2_bufdata *bd; |
| 688 | |
| 689 | head = folio_buffers(folio); |
| 690 | if (!head) |
| 691 | return false; |
| 692 | |
| 693 | /* |
| 694 | * mm accommodates an old ext3 case where clean folios might |
| 695 | * not have had the dirty bit cleared. Thus, it can send actual |
| 696 | * dirty folios to ->release_folio() via shrink_active_list(). |
| 697 | * |
| 698 | * As a workaround, we skip folios that contain dirty buffers |
| 699 | * below. Once ->release_folio isn't called on dirty folios |
| 700 | * anymore, we can warn on dirty buffers like we used to here |
| 701 | * again. |
| 702 | */ |
| 703 | |
| 704 | gfs2_log_lock(sdp); |
| 705 | bh = head; |
| 706 | do { |
| 707 | if (atomic_read(&bh->b_count)) |
| 708 | goto cannot_release; |
| 709 | bd = bh->b_private; |
| 710 | if (bd && bd->bd_tr) |
| 711 | goto cannot_release; |
| 712 | if (buffer_dirty(bh) || WARN_ON(buffer_pinned(bh))) |
| 713 | goto cannot_release; |
| 714 | bh = bh->b_this_page; |
| 715 | } while (bh != head); |
| 716 | |
| 717 | bh = head; |
| 718 | do { |
| 719 | bd = bh->b_private; |
| 720 | if (bd) { |
| 721 | gfs2_assert_warn(sdp, bd->bd_bh == bh); |
| 722 | bd->bd_bh = NULL; |
| 723 | bh->b_private = NULL; |
| 724 | /* |
| 725 | * The bd may still be queued as a revoke, in which |
| 726 | * case we must not dequeue nor free it. |
| 727 | */ |
| 728 | if (!bd->bd_blkno && !list_empty(&bd->bd_list)) |
| 729 | list_del_init(&bd->bd_list); |
| 730 | if (list_empty(&bd->bd_list)) |
| 731 | kmem_cache_free(gfs2_bufdata_cachep, bd); |
| 732 | } |
| 733 | |
| 734 | bh = bh->b_this_page; |
| 735 | } while (bh != head); |
| 736 | gfs2_log_unlock(sdp); |
| 737 | |
| 738 | return try_to_free_buffers(folio); |
| 739 | |
| 740 | cannot_release: |
| 741 | gfs2_log_unlock(sdp); |
| 742 | return false; |
| 743 | } |
| 744 | |
| 745 | static const struct address_space_operations gfs2_aops = { |
| 746 | .writepages = gfs2_writepages, |
| 747 | .read_folio = gfs2_read_folio, |
| 748 | .readahead = gfs2_readahead, |
| 749 | .dirty_folio = iomap_dirty_folio, |
| 750 | .release_folio = iomap_release_folio, |
| 751 | .invalidate_folio = iomap_invalidate_folio, |
| 752 | .bmap = gfs2_bmap, |
| 753 | .migrate_folio = filemap_migrate_folio, |
| 754 | .is_partially_uptodate = iomap_is_partially_uptodate, |
| 755 | .error_remove_page = generic_error_remove_page, |
| 756 | }; |
| 757 | |
| 758 | static const struct address_space_operations gfs2_jdata_aops = { |
| 759 | .writepage = gfs2_jdata_writepage, |
| 760 | .writepages = gfs2_jdata_writepages, |
| 761 | .read_folio = gfs2_read_folio, |
| 762 | .readahead = gfs2_readahead, |
| 763 | .dirty_folio = jdata_dirty_folio, |
| 764 | .bmap = gfs2_bmap, |
| 765 | .invalidate_folio = gfs2_invalidate_folio, |
| 766 | .release_folio = gfs2_release_folio, |
| 767 | .is_partially_uptodate = block_is_partially_uptodate, |
| 768 | .error_remove_page = generic_error_remove_page, |
| 769 | }; |
| 770 | |
| 771 | void gfs2_set_aops(struct inode *inode) |
| 772 | { |
| 773 | if (gfs2_is_jdata(GFS2_I(inode))) |
| 774 | inode->i_mapping->a_ops = &gfs2_jdata_aops; |
| 775 | else |
| 776 | inode->i_mapping->a_ops = &gfs2_aops; |
| 777 | } |