| 1 | // SPDX-License-Identifier: GPL-2.0+ |
| 2 | /* |
| 3 | * Buffer/page management specific to NILFS |
| 4 | * |
| 5 | * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation. |
| 6 | * |
| 7 | * Written by Ryusuke Konishi and Seiji Kihara. |
| 8 | */ |
| 9 | |
| 10 | #include <linux/pagemap.h> |
| 11 | #include <linux/writeback.h> |
| 12 | #include <linux/swap.h> |
| 13 | #include <linux/bitops.h> |
| 14 | #include <linux/page-flags.h> |
| 15 | #include <linux/list.h> |
| 16 | #include <linux/highmem.h> |
| 17 | #include <linux/pagevec.h> |
| 18 | #include <linux/gfp.h> |
| 19 | #include "nilfs.h" |
| 20 | #include "page.h" |
| 21 | #include "mdt.h" |
| 22 | |
| 23 | |
| 24 | #define NILFS_BUFFER_INHERENT_BITS \ |
| 25 | (BIT(BH_Uptodate) | BIT(BH_Mapped) | BIT(BH_NILFS_Node) | \ |
| 26 | BIT(BH_NILFS_Volatile) | BIT(BH_NILFS_Checked)) |
| 27 | |
| 28 | static struct buffer_head *__nilfs_get_folio_block(struct folio *folio, |
| 29 | unsigned long block, pgoff_t index, int blkbits, |
| 30 | unsigned long b_state) |
| 31 | |
| 32 | { |
| 33 | unsigned long first_block; |
| 34 | struct buffer_head *bh = folio_buffers(folio); |
| 35 | |
| 36 | if (!bh) |
| 37 | bh = create_empty_buffers(folio, 1 << blkbits, b_state); |
| 38 | |
| 39 | first_block = (unsigned long)index << (PAGE_SHIFT - blkbits); |
| 40 | bh = get_nth_bh(bh, block - first_block); |
| 41 | |
| 42 | touch_buffer(bh); |
| 43 | wait_on_buffer(bh); |
| 44 | return bh; |
| 45 | } |
| 46 | |
| 47 | struct buffer_head *nilfs_grab_buffer(struct inode *inode, |
| 48 | struct address_space *mapping, |
| 49 | unsigned long blkoff, |
| 50 | unsigned long b_state) |
| 51 | { |
| 52 | int blkbits = inode->i_blkbits; |
| 53 | pgoff_t index = blkoff >> (PAGE_SHIFT - blkbits); |
| 54 | struct folio *folio; |
| 55 | struct buffer_head *bh; |
| 56 | |
| 57 | folio = filemap_grab_folio(mapping, index); |
| 58 | if (IS_ERR(folio)) |
| 59 | return NULL; |
| 60 | |
| 61 | bh = __nilfs_get_folio_block(folio, blkoff, index, blkbits, b_state); |
| 62 | if (unlikely(!bh)) { |
| 63 | folio_unlock(folio); |
| 64 | folio_put(folio); |
| 65 | return NULL; |
| 66 | } |
| 67 | return bh; |
| 68 | } |
| 69 | |
| 70 | /** |
| 71 | * nilfs_forget_buffer - discard dirty state |
| 72 | * @bh: buffer head of the buffer to be discarded |
| 73 | */ |
| 74 | void nilfs_forget_buffer(struct buffer_head *bh) |
| 75 | { |
| 76 | struct folio *folio = bh->b_folio; |
| 77 | const unsigned long clear_bits = |
| 78 | (BIT(BH_Uptodate) | BIT(BH_Dirty) | BIT(BH_Mapped) | |
| 79 | BIT(BH_Async_Write) | BIT(BH_NILFS_Volatile) | |
| 80 | BIT(BH_NILFS_Checked) | BIT(BH_NILFS_Redirected)); |
| 81 | |
| 82 | lock_buffer(bh); |
| 83 | set_mask_bits(&bh->b_state, clear_bits, 0); |
| 84 | if (nilfs_folio_buffers_clean(folio)) |
| 85 | __nilfs_clear_folio_dirty(folio); |
| 86 | |
| 87 | bh->b_blocknr = -1; |
| 88 | folio_clear_uptodate(folio); |
| 89 | folio_clear_mappedtodisk(folio); |
| 90 | unlock_buffer(bh); |
| 91 | brelse(bh); |
| 92 | } |
| 93 | |
| 94 | /** |
| 95 | * nilfs_copy_buffer -- copy buffer data and flags |
| 96 | * @dbh: destination buffer |
| 97 | * @sbh: source buffer |
| 98 | */ |
| 99 | void nilfs_copy_buffer(struct buffer_head *dbh, struct buffer_head *sbh) |
| 100 | { |
| 101 | void *kaddr0, *kaddr1; |
| 102 | unsigned long bits; |
| 103 | struct page *spage = sbh->b_page, *dpage = dbh->b_page; |
| 104 | struct buffer_head *bh; |
| 105 | |
| 106 | kaddr0 = kmap_local_page(spage); |
| 107 | kaddr1 = kmap_local_page(dpage); |
| 108 | memcpy(kaddr1 + bh_offset(dbh), kaddr0 + bh_offset(sbh), sbh->b_size); |
| 109 | kunmap_local(kaddr1); |
| 110 | kunmap_local(kaddr0); |
| 111 | |
| 112 | dbh->b_state = sbh->b_state & NILFS_BUFFER_INHERENT_BITS; |
| 113 | dbh->b_blocknr = sbh->b_blocknr; |
| 114 | dbh->b_bdev = sbh->b_bdev; |
| 115 | |
| 116 | bh = dbh; |
| 117 | bits = sbh->b_state & (BIT(BH_Uptodate) | BIT(BH_Mapped)); |
| 118 | while ((bh = bh->b_this_page) != dbh) { |
| 119 | lock_buffer(bh); |
| 120 | bits &= bh->b_state; |
| 121 | unlock_buffer(bh); |
| 122 | } |
| 123 | if (bits & BIT(BH_Uptodate)) |
| 124 | SetPageUptodate(dpage); |
| 125 | else |
| 126 | ClearPageUptodate(dpage); |
| 127 | if (bits & BIT(BH_Mapped)) |
| 128 | SetPageMappedToDisk(dpage); |
| 129 | else |
| 130 | ClearPageMappedToDisk(dpage); |
| 131 | } |
| 132 | |
| 133 | /** |
| 134 | * nilfs_folio_buffers_clean - Check if a folio has dirty buffers or not. |
| 135 | * @folio: Folio to be checked. |
| 136 | * |
| 137 | * nilfs_folio_buffers_clean() returns false if the folio has dirty buffers. |
| 138 | * Otherwise, it returns true. |
| 139 | */ |
| 140 | bool nilfs_folio_buffers_clean(struct folio *folio) |
| 141 | { |
| 142 | struct buffer_head *bh, *head; |
| 143 | |
| 144 | bh = head = folio_buffers(folio); |
| 145 | do { |
| 146 | if (buffer_dirty(bh)) |
| 147 | return false; |
| 148 | bh = bh->b_this_page; |
| 149 | } while (bh != head); |
| 150 | return true; |
| 151 | } |
| 152 | |
| 153 | void nilfs_folio_bug(struct folio *folio) |
| 154 | { |
| 155 | struct buffer_head *bh, *head; |
| 156 | struct address_space *m; |
| 157 | unsigned long ino; |
| 158 | |
| 159 | if (unlikely(!folio)) { |
| 160 | printk(KERN_CRIT "NILFS_FOLIO_BUG(NULL)\n"); |
| 161 | return; |
| 162 | } |
| 163 | |
| 164 | m = folio->mapping; |
| 165 | ino = m ? m->host->i_ino : 0; |
| 166 | |
| 167 | printk(KERN_CRIT "NILFS_FOLIO_BUG(%p): cnt=%d index#=%llu flags=0x%lx " |
| 168 | "mapping=%p ino=%lu\n", |
| 169 | folio, folio_ref_count(folio), |
| 170 | (unsigned long long)folio->index, folio->flags, m, ino); |
| 171 | |
| 172 | head = folio_buffers(folio); |
| 173 | if (head) { |
| 174 | int i = 0; |
| 175 | |
| 176 | bh = head; |
| 177 | do { |
| 178 | printk(KERN_CRIT |
| 179 | " BH[%d] %p: cnt=%d block#=%llu state=0x%lx\n", |
| 180 | i++, bh, atomic_read(&bh->b_count), |
| 181 | (unsigned long long)bh->b_blocknr, bh->b_state); |
| 182 | bh = bh->b_this_page; |
| 183 | } while (bh != head); |
| 184 | } |
| 185 | } |
| 186 | |
| 187 | /** |
| 188 | * nilfs_copy_folio -- copy the folio with buffers |
| 189 | * @dst: destination folio |
| 190 | * @src: source folio |
| 191 | * @copy_dirty: flag whether to copy dirty states on the folio's buffer heads. |
| 192 | * |
| 193 | * This function is for both data folios and btnode folios. The dirty flag |
| 194 | * should be treated by caller. The folio must not be under i/o. |
| 195 | * Both src and dst folio must be locked |
| 196 | */ |
| 197 | static void nilfs_copy_folio(struct folio *dst, struct folio *src, |
| 198 | bool copy_dirty) |
| 199 | { |
| 200 | struct buffer_head *dbh, *dbufs, *sbh; |
| 201 | unsigned long mask = NILFS_BUFFER_INHERENT_BITS; |
| 202 | |
| 203 | BUG_ON(folio_test_writeback(dst)); |
| 204 | |
| 205 | sbh = folio_buffers(src); |
| 206 | dbh = folio_buffers(dst); |
| 207 | if (!dbh) |
| 208 | dbh = create_empty_buffers(dst, sbh->b_size, 0); |
| 209 | |
| 210 | if (copy_dirty) |
| 211 | mask |= BIT(BH_Dirty); |
| 212 | |
| 213 | dbufs = dbh; |
| 214 | do { |
| 215 | lock_buffer(sbh); |
| 216 | lock_buffer(dbh); |
| 217 | dbh->b_state = sbh->b_state & mask; |
| 218 | dbh->b_blocknr = sbh->b_blocknr; |
| 219 | dbh->b_bdev = sbh->b_bdev; |
| 220 | sbh = sbh->b_this_page; |
| 221 | dbh = dbh->b_this_page; |
| 222 | } while (dbh != dbufs); |
| 223 | |
| 224 | folio_copy(dst, src); |
| 225 | |
| 226 | if (folio_test_uptodate(src) && !folio_test_uptodate(dst)) |
| 227 | folio_mark_uptodate(dst); |
| 228 | else if (!folio_test_uptodate(src) && folio_test_uptodate(dst)) |
| 229 | folio_clear_uptodate(dst); |
| 230 | if (folio_test_mappedtodisk(src) && !folio_test_mappedtodisk(dst)) |
| 231 | folio_set_mappedtodisk(dst); |
| 232 | else if (!folio_test_mappedtodisk(src) && folio_test_mappedtodisk(dst)) |
| 233 | folio_clear_mappedtodisk(dst); |
| 234 | |
| 235 | do { |
| 236 | unlock_buffer(sbh); |
| 237 | unlock_buffer(dbh); |
| 238 | sbh = sbh->b_this_page; |
| 239 | dbh = dbh->b_this_page; |
| 240 | } while (dbh != dbufs); |
| 241 | } |
| 242 | |
| 243 | int nilfs_copy_dirty_pages(struct address_space *dmap, |
| 244 | struct address_space *smap) |
| 245 | { |
| 246 | struct folio_batch fbatch; |
| 247 | unsigned int i; |
| 248 | pgoff_t index = 0; |
| 249 | int err = 0; |
| 250 | |
| 251 | folio_batch_init(&fbatch); |
| 252 | repeat: |
| 253 | if (!filemap_get_folios_tag(smap, &index, (pgoff_t)-1, |
| 254 | PAGECACHE_TAG_DIRTY, &fbatch)) |
| 255 | return 0; |
| 256 | |
| 257 | for (i = 0; i < folio_batch_count(&fbatch); i++) { |
| 258 | struct folio *folio = fbatch.folios[i], *dfolio; |
| 259 | |
| 260 | folio_lock(folio); |
| 261 | if (unlikely(!folio_test_dirty(folio))) |
| 262 | NILFS_FOLIO_BUG(folio, "inconsistent dirty state"); |
| 263 | |
| 264 | dfolio = filemap_grab_folio(dmap, folio->index); |
| 265 | if (IS_ERR(dfolio)) { |
| 266 | /* No empty page is added to the page cache */ |
| 267 | folio_unlock(folio); |
| 268 | err = PTR_ERR(dfolio); |
| 269 | break; |
| 270 | } |
| 271 | if (unlikely(!folio_buffers(folio))) |
| 272 | NILFS_FOLIO_BUG(folio, |
| 273 | "found empty page in dat page cache"); |
| 274 | |
| 275 | nilfs_copy_folio(dfolio, folio, true); |
| 276 | filemap_dirty_folio(folio_mapping(dfolio), dfolio); |
| 277 | |
| 278 | folio_unlock(dfolio); |
| 279 | folio_put(dfolio); |
| 280 | folio_unlock(folio); |
| 281 | } |
| 282 | folio_batch_release(&fbatch); |
| 283 | cond_resched(); |
| 284 | |
| 285 | if (likely(!err)) |
| 286 | goto repeat; |
| 287 | return err; |
| 288 | } |
| 289 | |
| 290 | /** |
| 291 | * nilfs_copy_back_pages -- copy back pages to original cache from shadow cache |
| 292 | * @dmap: destination page cache |
| 293 | * @smap: source page cache |
| 294 | * |
| 295 | * No pages must be added to the cache during this process. |
| 296 | * This must be ensured by the caller. |
| 297 | */ |
| 298 | void nilfs_copy_back_pages(struct address_space *dmap, |
| 299 | struct address_space *smap) |
| 300 | { |
| 301 | struct folio_batch fbatch; |
| 302 | unsigned int i, n; |
| 303 | pgoff_t start = 0; |
| 304 | |
| 305 | folio_batch_init(&fbatch); |
| 306 | repeat: |
| 307 | n = filemap_get_folios(smap, &start, ~0UL, &fbatch); |
| 308 | if (!n) |
| 309 | return; |
| 310 | |
| 311 | for (i = 0; i < folio_batch_count(&fbatch); i++) { |
| 312 | struct folio *folio = fbatch.folios[i], *dfolio; |
| 313 | pgoff_t index = folio->index; |
| 314 | |
| 315 | folio_lock(folio); |
| 316 | dfolio = filemap_lock_folio(dmap, index); |
| 317 | if (!IS_ERR(dfolio)) { |
| 318 | /* overwrite existing folio in the destination cache */ |
| 319 | WARN_ON(folio_test_dirty(dfolio)); |
| 320 | nilfs_copy_folio(dfolio, folio, false); |
| 321 | folio_unlock(dfolio); |
| 322 | folio_put(dfolio); |
| 323 | /* Do we not need to remove folio from smap here? */ |
| 324 | } else { |
| 325 | struct folio *f; |
| 326 | |
| 327 | /* move the folio to the destination cache */ |
| 328 | xa_lock_irq(&smap->i_pages); |
| 329 | f = __xa_erase(&smap->i_pages, index); |
| 330 | WARN_ON(folio != f); |
| 331 | smap->nrpages--; |
| 332 | xa_unlock_irq(&smap->i_pages); |
| 333 | |
| 334 | xa_lock_irq(&dmap->i_pages); |
| 335 | f = __xa_store(&dmap->i_pages, index, folio, GFP_NOFS); |
| 336 | if (unlikely(f)) { |
| 337 | /* Probably -ENOMEM */ |
| 338 | folio->mapping = NULL; |
| 339 | folio_put(folio); |
| 340 | } else { |
| 341 | folio->mapping = dmap; |
| 342 | dmap->nrpages++; |
| 343 | if (folio_test_dirty(folio)) |
| 344 | __xa_set_mark(&dmap->i_pages, index, |
| 345 | PAGECACHE_TAG_DIRTY); |
| 346 | } |
| 347 | xa_unlock_irq(&dmap->i_pages); |
| 348 | } |
| 349 | folio_unlock(folio); |
| 350 | } |
| 351 | folio_batch_release(&fbatch); |
| 352 | cond_resched(); |
| 353 | |
| 354 | goto repeat; |
| 355 | } |
| 356 | |
| 357 | /** |
| 358 | * nilfs_clear_dirty_pages - discard dirty pages in address space |
| 359 | * @mapping: address space with dirty pages for discarding |
| 360 | */ |
| 361 | void nilfs_clear_dirty_pages(struct address_space *mapping) |
| 362 | { |
| 363 | struct folio_batch fbatch; |
| 364 | unsigned int i; |
| 365 | pgoff_t index = 0; |
| 366 | |
| 367 | folio_batch_init(&fbatch); |
| 368 | |
| 369 | while (filemap_get_folios_tag(mapping, &index, (pgoff_t)-1, |
| 370 | PAGECACHE_TAG_DIRTY, &fbatch)) { |
| 371 | for (i = 0; i < folio_batch_count(&fbatch); i++) { |
| 372 | struct folio *folio = fbatch.folios[i]; |
| 373 | |
| 374 | folio_lock(folio); |
| 375 | |
| 376 | /* |
| 377 | * This folio may have been removed from the address |
| 378 | * space by truncation or invalidation when the lock |
| 379 | * was acquired. Skip processing in that case. |
| 380 | */ |
| 381 | if (likely(folio->mapping == mapping)) |
| 382 | nilfs_clear_folio_dirty(folio); |
| 383 | |
| 384 | folio_unlock(folio); |
| 385 | } |
| 386 | folio_batch_release(&fbatch); |
| 387 | cond_resched(); |
| 388 | } |
| 389 | } |
| 390 | |
| 391 | /** |
| 392 | * nilfs_clear_folio_dirty - discard dirty folio |
| 393 | * @folio: dirty folio that will be discarded |
| 394 | */ |
| 395 | void nilfs_clear_folio_dirty(struct folio *folio) |
| 396 | { |
| 397 | struct buffer_head *bh, *head; |
| 398 | |
| 399 | BUG_ON(!folio_test_locked(folio)); |
| 400 | |
| 401 | folio_clear_uptodate(folio); |
| 402 | folio_clear_mappedtodisk(folio); |
| 403 | |
| 404 | head = folio_buffers(folio); |
| 405 | if (head) { |
| 406 | const unsigned long clear_bits = |
| 407 | (BIT(BH_Uptodate) | BIT(BH_Dirty) | BIT(BH_Mapped) | |
| 408 | BIT(BH_Async_Write) | BIT(BH_NILFS_Volatile) | |
| 409 | BIT(BH_NILFS_Checked) | BIT(BH_NILFS_Redirected)); |
| 410 | |
| 411 | bh = head; |
| 412 | do { |
| 413 | lock_buffer(bh); |
| 414 | set_mask_bits(&bh->b_state, clear_bits, 0); |
| 415 | unlock_buffer(bh); |
| 416 | } while (bh = bh->b_this_page, bh != head); |
| 417 | } |
| 418 | |
| 419 | __nilfs_clear_folio_dirty(folio); |
| 420 | } |
| 421 | |
| 422 | unsigned int nilfs_page_count_clean_buffers(struct page *page, |
| 423 | unsigned int from, unsigned int to) |
| 424 | { |
| 425 | unsigned int block_start, block_end; |
| 426 | struct buffer_head *bh, *head; |
| 427 | unsigned int nc = 0; |
| 428 | |
| 429 | for (bh = head = page_buffers(page), block_start = 0; |
| 430 | bh != head || !block_start; |
| 431 | block_start = block_end, bh = bh->b_this_page) { |
| 432 | block_end = block_start + bh->b_size; |
| 433 | if (block_end > from && block_start < to && !buffer_dirty(bh)) |
| 434 | nc++; |
| 435 | } |
| 436 | return nc; |
| 437 | } |
| 438 | |
| 439 | /* |
| 440 | * NILFS2 needs clear_page_dirty() in the following two cases: |
| 441 | * |
| 442 | * 1) For B-tree node pages and data pages of DAT file, NILFS2 clears dirty |
| 443 | * flag of pages when it copies back pages from shadow cache to the |
| 444 | * original cache. |
| 445 | * |
| 446 | * 2) Some B-tree operations like insertion or deletion may dispose buffers |
| 447 | * in dirty state, and this needs to cancel the dirty state of their pages. |
| 448 | */ |
| 449 | void __nilfs_clear_folio_dirty(struct folio *folio) |
| 450 | { |
| 451 | struct address_space *mapping = folio->mapping; |
| 452 | |
| 453 | if (mapping) { |
| 454 | xa_lock_irq(&mapping->i_pages); |
| 455 | if (folio_test_dirty(folio)) { |
| 456 | __xa_clear_mark(&mapping->i_pages, folio->index, |
| 457 | PAGECACHE_TAG_DIRTY); |
| 458 | xa_unlock_irq(&mapping->i_pages); |
| 459 | folio_clear_dirty_for_io(folio); |
| 460 | return; |
| 461 | } |
| 462 | xa_unlock_irq(&mapping->i_pages); |
| 463 | return; |
| 464 | } |
| 465 | folio_clear_dirty(folio); |
| 466 | } |
| 467 | |
| 468 | /** |
| 469 | * nilfs_find_uncommitted_extent - find extent of uncommitted data |
| 470 | * @inode: inode |
| 471 | * @start_blk: start block offset (in) |
| 472 | * @blkoff: start offset of the found extent (out) |
| 473 | * |
| 474 | * This function searches an extent of buffers marked "delayed" which |
| 475 | * starts from a block offset equal to or larger than @start_blk. If |
| 476 | * such an extent was found, this will store the start offset in |
| 477 | * @blkoff and return its length in blocks. Otherwise, zero is |
| 478 | * returned. |
| 479 | */ |
| 480 | unsigned long nilfs_find_uncommitted_extent(struct inode *inode, |
| 481 | sector_t start_blk, |
| 482 | sector_t *blkoff) |
| 483 | { |
| 484 | unsigned int i, nr_folios; |
| 485 | pgoff_t index; |
| 486 | unsigned long length = 0; |
| 487 | struct folio_batch fbatch; |
| 488 | struct folio *folio; |
| 489 | |
| 490 | if (inode->i_mapping->nrpages == 0) |
| 491 | return 0; |
| 492 | |
| 493 | index = start_blk >> (PAGE_SHIFT - inode->i_blkbits); |
| 494 | |
| 495 | folio_batch_init(&fbatch); |
| 496 | |
| 497 | repeat: |
| 498 | nr_folios = filemap_get_folios_contig(inode->i_mapping, &index, ULONG_MAX, |
| 499 | &fbatch); |
| 500 | if (nr_folios == 0) |
| 501 | return length; |
| 502 | |
| 503 | i = 0; |
| 504 | do { |
| 505 | folio = fbatch.folios[i]; |
| 506 | |
| 507 | folio_lock(folio); |
| 508 | if (folio_buffers(folio)) { |
| 509 | struct buffer_head *bh, *head; |
| 510 | sector_t b; |
| 511 | |
| 512 | b = folio->index << (PAGE_SHIFT - inode->i_blkbits); |
| 513 | bh = head = folio_buffers(folio); |
| 514 | do { |
| 515 | if (b < start_blk) |
| 516 | continue; |
| 517 | if (buffer_delay(bh)) { |
| 518 | if (length == 0) |
| 519 | *blkoff = b; |
| 520 | length++; |
| 521 | } else if (length > 0) { |
| 522 | goto out_locked; |
| 523 | } |
| 524 | } while (++b, bh = bh->b_this_page, bh != head); |
| 525 | } else { |
| 526 | if (length > 0) |
| 527 | goto out_locked; |
| 528 | } |
| 529 | folio_unlock(folio); |
| 530 | |
| 531 | } while (++i < nr_folios); |
| 532 | |
| 533 | folio_batch_release(&fbatch); |
| 534 | cond_resched(); |
| 535 | goto repeat; |
| 536 | |
| 537 | out_locked: |
| 538 | folio_unlock(folio); |
| 539 | folio_batch_release(&fbatch); |
| 540 | return length; |
| 541 | } |