1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * linux/mm/filemap.c 4 * 5 * Copyright (C) 1994-1999 Linus Torvalds 6 */ 7 8 /* 9 * This file handles the generic file mmap semantics used by 10 * most "normal" filesystems (but you don't /have/ to use this: 11 * the NFS filesystem used to do this differently, for example) 12 */ 13 #include <linux/export.h> 14 #include <linux/compiler.h> 15 #include <linux/dax.h> 16 #include <linux/fs.h> 17 #include <linux/sched/signal.h> 18 #include <linux/uaccess.h> 19 #include <linux/capability.h> 20 #include <linux/kernel_stat.h> 21 #include <linux/gfp.h> 22 #include <linux/mm.h> 23 #include <linux/swap.h> 24 #include <linux/swapops.h> 25 #include <linux/mman.h> 26 #include <linux/pagemap.h> 27 #include <linux/file.h> 28 #include <linux/uio.h> 29 #include <linux/error-injection.h> 30 #include <linux/hash.h> 31 #include <linux/writeback.h> 32 #include <linux/backing-dev.h> 33 #include <linux/pagevec.h> 34 #include <linux/security.h> 35 #include <linux/cpuset.h> 36 #include <linux/hugetlb.h> 37 #include <linux/memcontrol.h> 38 #include <linux/shmem_fs.h> 39 #include <linux/rmap.h> 40 #include <linux/delayacct.h> 41 #include <linux/psi.h> 42 #include <linux/ramfs.h> 43 #include <linux/page_idle.h> 44 #include <linux/migrate.h> 45 #include <asm/pgalloc.h> 46 #include <asm/tlbflush.h> 47 #include "internal.h" 48 49 #define CREATE_TRACE_POINTS 50 #include <trace/events/filemap.h> 51 52 /* 53 * FIXME: remove all knowledge of the buffer layer from the core VM 54 */ 55 #include <linux/buffer_head.h> /* for try_to_free_buffers */ 56 57 #include <asm/mman.h> 58 59 /* 60 * Shared mappings implemented 30.11.1994. It's not fully working yet, 61 * though. 62 * 63 * Shared mappings now work. 15.8.1995 Bruno. 64 * 65 * finished 'unifying' the page and buffer cache and SMP-threaded the 66 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> 67 * 68 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> 69 */ 70 71 /* 72 * Lock ordering: 73 * 74 * ->i_mmap_rwsem (truncate_pagecache) 75 * ->private_lock (__free_pte->block_dirty_folio) 76 * ->swap_lock (exclusive_swap_page, others) 77 * ->i_pages lock 78 * 79 * ->i_rwsem 80 * ->invalidate_lock (acquired by fs in truncate path) 81 * ->i_mmap_rwsem (truncate->unmap_mapping_range) 82 * 83 * ->mmap_lock 84 * ->i_mmap_rwsem 85 * ->page_table_lock or pte_lock (various, mainly in memory.c) 86 * ->i_pages lock (arch-dependent flush_dcache_mmap_lock) 87 * 88 * ->mmap_lock 89 * ->invalidate_lock (filemap_fault) 90 * ->lock_page (filemap_fault, access_process_vm) 91 * 92 * ->i_rwsem (generic_perform_write) 93 * ->mmap_lock (fault_in_readable->do_page_fault) 94 * 95 * bdi->wb.list_lock 96 * sb_lock (fs/fs-writeback.c) 97 * ->i_pages lock (__sync_single_inode) 98 * 99 * ->i_mmap_rwsem 100 * ->anon_vma.lock (vma_adjust) 101 * 102 * ->anon_vma.lock 103 * ->page_table_lock or pte_lock (anon_vma_prepare and various) 104 * 105 * ->page_table_lock or pte_lock 106 * ->swap_lock (try_to_unmap_one) 107 * ->private_lock (try_to_unmap_one) 108 * ->i_pages lock (try_to_unmap_one) 109 * ->lruvec->lru_lock (follow_page->mark_page_accessed) 110 * ->lruvec->lru_lock (check_pte_range->isolate_lru_page) 111 * ->private_lock (page_remove_rmap->set_page_dirty) 112 * ->i_pages lock (page_remove_rmap->set_page_dirty) 113 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) 114 * ->inode->i_lock (page_remove_rmap->set_page_dirty) 115 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg) 116 * bdi.wb->list_lock (zap_pte_range->set_page_dirty) 117 * ->inode->i_lock (zap_pte_range->set_page_dirty) 118 * ->private_lock (zap_pte_range->block_dirty_folio) 119 * 120 * ->i_mmap_rwsem 121 * ->tasklist_lock (memory_failure, collect_procs_ao) 122 */ 123 124 static void page_cache_delete(struct address_space *mapping, 125 struct folio *folio, void *shadow) 126 { 127 XA_STATE(xas, &mapping->i_pages, folio->index); 128 long nr = 1; 129 130 mapping_set_update(&xas, mapping); 131 132 /* hugetlb pages are represented by a single entry in the xarray */ 133 if (!folio_test_hugetlb(folio)) { 134 xas_set_order(&xas, folio->index, folio_order(folio)); 135 nr = folio_nr_pages(folio); 136 } 137 138 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); 139 140 xas_store(&xas, shadow); 141 xas_init_marks(&xas); 142 143 folio->mapping = NULL; 144 /* Leave page->index set: truncation lookup relies upon it */ 145 mapping->nrpages -= nr; 146 } 147 148 static void filemap_unaccount_folio(struct address_space *mapping, 149 struct folio *folio) 150 { 151 long nr; 152 153 VM_BUG_ON_FOLIO(folio_mapped(folio), folio); 154 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(folio_mapped(folio))) { 155 int mapcount; 156 157 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n", 158 current->comm, folio_pfn(folio)); 159 dump_page(&folio->page, "still mapped when deleted"); 160 dump_stack(); 161 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); 162 163 mapcount = page_mapcount(&folio->page); 164 if (mapping_exiting(mapping) && 165 folio_ref_count(folio) >= mapcount + 2) { 166 /* 167 * All vmas have already been torn down, so it's 168 * a good bet that actually the folio is unmapped, 169 * and we'd prefer not to leak it: if we're wrong, 170 * some other bad page check should catch it later. 171 */ 172 page_mapcount_reset(&folio->page); 173 folio_ref_sub(folio, mapcount); 174 } 175 } 176 177 /* hugetlb folios do not participate in page cache accounting. */ 178 if (folio_test_hugetlb(folio)) 179 return; 180 181 nr = folio_nr_pages(folio); 182 183 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr); 184 if (folio_test_swapbacked(folio)) { 185 __lruvec_stat_mod_folio(folio, NR_SHMEM, -nr); 186 if (folio_test_pmd_mappable(folio)) 187 __lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, -nr); 188 } else if (folio_test_pmd_mappable(folio)) { 189 __lruvec_stat_mod_folio(folio, NR_FILE_THPS, -nr); 190 filemap_nr_thps_dec(mapping); 191 } 192 193 /* 194 * At this point folio must be either written or cleaned by 195 * truncate. Dirty folio here signals a bug and loss of 196 * unwritten data. 197 * 198 * This fixes dirty accounting after removing the folio entirely 199 * but leaves the dirty flag set: it has no effect for truncated 200 * folio and anyway will be cleared before returning folio to 201 * buddy allocator. 202 */ 203 if (WARN_ON_ONCE(folio_test_dirty(folio))) 204 folio_account_cleaned(folio, mapping, 205 inode_to_wb(mapping->host)); 206 } 207 208 /* 209 * Delete a page from the page cache and free it. Caller has to make 210 * sure the page is locked and that nobody else uses it - or that usage 211 * is safe. The caller must hold the i_pages lock. 212 */ 213 void __filemap_remove_folio(struct folio *folio, void *shadow) 214 { 215 struct address_space *mapping = folio->mapping; 216 217 trace_mm_filemap_delete_from_page_cache(folio); 218 filemap_unaccount_folio(mapping, folio); 219 page_cache_delete(mapping, folio, shadow); 220 } 221 222 void filemap_free_folio(struct address_space *mapping, struct folio *folio) 223 { 224 void (*freepage)(struct page *); 225 int refs = 1; 226 227 freepage = mapping->a_ops->freepage; 228 if (freepage) 229 freepage(&folio->page); 230 231 if (folio_test_large(folio) && !folio_test_hugetlb(folio)) 232 refs = folio_nr_pages(folio); 233 folio_put_refs(folio, refs); 234 } 235 236 /** 237 * filemap_remove_folio - Remove folio from page cache. 238 * @folio: The folio. 239 * 240 * This must be called only on folios that are locked and have been 241 * verified to be in the page cache. It will never put the folio into 242 * the free list because the caller has a reference on the page. 243 */ 244 void filemap_remove_folio(struct folio *folio) 245 { 246 struct address_space *mapping = folio->mapping; 247 248 BUG_ON(!folio_test_locked(folio)); 249 spin_lock(&mapping->host->i_lock); 250 xa_lock_irq(&mapping->i_pages); 251 __filemap_remove_folio(folio, NULL); 252 xa_unlock_irq(&mapping->i_pages); 253 if (mapping_shrinkable(mapping)) 254 inode_add_lru(mapping->host); 255 spin_unlock(&mapping->host->i_lock); 256 257 filemap_free_folio(mapping, folio); 258 } 259 260 /* 261 * page_cache_delete_batch - delete several folios from page cache 262 * @mapping: the mapping to which folios belong 263 * @fbatch: batch of folios to delete 264 * 265 * The function walks over mapping->i_pages and removes folios passed in 266 * @fbatch from the mapping. The function expects @fbatch to be sorted 267 * by page index and is optimised for it to be dense. 268 * It tolerates holes in @fbatch (mapping entries at those indices are not 269 * modified). 270 * 271 * The function expects the i_pages lock to be held. 272 */ 273 static void page_cache_delete_batch(struct address_space *mapping, 274 struct folio_batch *fbatch) 275 { 276 XA_STATE(xas, &mapping->i_pages, fbatch->folios[0]->index); 277 long total_pages = 0; 278 int i = 0; 279 struct folio *folio; 280 281 mapping_set_update(&xas, mapping); 282 xas_for_each(&xas, folio, ULONG_MAX) { 283 if (i >= folio_batch_count(fbatch)) 284 break; 285 286 /* A swap/dax/shadow entry got inserted? Skip it. */ 287 if (xa_is_value(folio)) 288 continue; 289 /* 290 * A page got inserted in our range? Skip it. We have our 291 * pages locked so they are protected from being removed. 292 * If we see a page whose index is higher than ours, it 293 * means our page has been removed, which shouldn't be 294 * possible because we're holding the PageLock. 295 */ 296 if (folio != fbatch->folios[i]) { 297 VM_BUG_ON_FOLIO(folio->index > 298 fbatch->folios[i]->index, folio); 299 continue; 300 } 301 302 WARN_ON_ONCE(!folio_test_locked(folio)); 303 304 folio->mapping = NULL; 305 /* Leave folio->index set: truncation lookup relies on it */ 306 307 i++; 308 xas_store(&xas, NULL); 309 total_pages += folio_nr_pages(folio); 310 } 311 mapping->nrpages -= total_pages; 312 } 313 314 void delete_from_page_cache_batch(struct address_space *mapping, 315 struct folio_batch *fbatch) 316 { 317 int i; 318 319 if (!folio_batch_count(fbatch)) 320 return; 321 322 spin_lock(&mapping->host->i_lock); 323 xa_lock_irq(&mapping->i_pages); 324 for (i = 0; i < folio_batch_count(fbatch); i++) { 325 struct folio *folio = fbatch->folios[i]; 326 327 trace_mm_filemap_delete_from_page_cache(folio); 328 filemap_unaccount_folio(mapping, folio); 329 } 330 page_cache_delete_batch(mapping, fbatch); 331 xa_unlock_irq(&mapping->i_pages); 332 if (mapping_shrinkable(mapping)) 333 inode_add_lru(mapping->host); 334 spin_unlock(&mapping->host->i_lock); 335 336 for (i = 0; i < folio_batch_count(fbatch); i++) 337 filemap_free_folio(mapping, fbatch->folios[i]); 338 } 339 340 int filemap_check_errors(struct address_space *mapping) 341 { 342 int ret = 0; 343 /* Check for outstanding write errors */ 344 if (test_bit(AS_ENOSPC, &mapping->flags) && 345 test_and_clear_bit(AS_ENOSPC, &mapping->flags)) 346 ret = -ENOSPC; 347 if (test_bit(AS_EIO, &mapping->flags) && 348 test_and_clear_bit(AS_EIO, &mapping->flags)) 349 ret = -EIO; 350 return ret; 351 } 352 EXPORT_SYMBOL(filemap_check_errors); 353 354 static int filemap_check_and_keep_errors(struct address_space *mapping) 355 { 356 /* Check for outstanding write errors */ 357 if (test_bit(AS_EIO, &mapping->flags)) 358 return -EIO; 359 if (test_bit(AS_ENOSPC, &mapping->flags)) 360 return -ENOSPC; 361 return 0; 362 } 363 364 /** 365 * filemap_fdatawrite_wbc - start writeback on mapping dirty pages in range 366 * @mapping: address space structure to write 367 * @wbc: the writeback_control controlling the writeout 368 * 369 * Call writepages on the mapping using the provided wbc to control the 370 * writeout. 371 * 372 * Return: %0 on success, negative error code otherwise. 373 */ 374 int filemap_fdatawrite_wbc(struct address_space *mapping, 375 struct writeback_control *wbc) 376 { 377 int ret; 378 379 if (!mapping_can_writeback(mapping) || 380 !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 381 return 0; 382 383 wbc_attach_fdatawrite_inode(wbc, mapping->host); 384 ret = do_writepages(mapping, wbc); 385 wbc_detach_inode(wbc); 386 return ret; 387 } 388 EXPORT_SYMBOL(filemap_fdatawrite_wbc); 389 390 /** 391 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range 392 * @mapping: address space structure to write 393 * @start: offset in bytes where the range starts 394 * @end: offset in bytes where the range ends (inclusive) 395 * @sync_mode: enable synchronous operation 396 * 397 * Start writeback against all of a mapping's dirty pages that lie 398 * within the byte offsets <start, end> inclusive. 399 * 400 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as 401 * opposed to a regular memory cleansing writeback. The difference between 402 * these two operations is that if a dirty page/buffer is encountered, it must 403 * be waited upon, and not just skipped over. 404 * 405 * Return: %0 on success, negative error code otherwise. 406 */ 407 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, 408 loff_t end, int sync_mode) 409 { 410 struct writeback_control wbc = { 411 .sync_mode = sync_mode, 412 .nr_to_write = LONG_MAX, 413 .range_start = start, 414 .range_end = end, 415 }; 416 417 return filemap_fdatawrite_wbc(mapping, &wbc); 418 } 419 420 static inline int __filemap_fdatawrite(struct address_space *mapping, 421 int sync_mode) 422 { 423 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); 424 } 425 426 int filemap_fdatawrite(struct address_space *mapping) 427 { 428 return __filemap_fdatawrite(mapping, WB_SYNC_ALL); 429 } 430 EXPORT_SYMBOL(filemap_fdatawrite); 431 432 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, 433 loff_t end) 434 { 435 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); 436 } 437 EXPORT_SYMBOL(filemap_fdatawrite_range); 438 439 /** 440 * filemap_flush - mostly a non-blocking flush 441 * @mapping: target address_space 442 * 443 * This is a mostly non-blocking flush. Not suitable for data-integrity 444 * purposes - I/O may not be started against all dirty pages. 445 * 446 * Return: %0 on success, negative error code otherwise. 447 */ 448 int filemap_flush(struct address_space *mapping) 449 { 450 return __filemap_fdatawrite(mapping, WB_SYNC_NONE); 451 } 452 EXPORT_SYMBOL(filemap_flush); 453 454 /** 455 * filemap_range_has_page - check if a page exists in range. 456 * @mapping: address space within which to check 457 * @start_byte: offset in bytes where the range starts 458 * @end_byte: offset in bytes where the range ends (inclusive) 459 * 460 * Find at least one page in the range supplied, usually used to check if 461 * direct writing in this range will trigger a writeback. 462 * 463 * Return: %true if at least one page exists in the specified range, 464 * %false otherwise. 465 */ 466 bool filemap_range_has_page(struct address_space *mapping, 467 loff_t start_byte, loff_t end_byte) 468 { 469 struct page *page; 470 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT); 471 pgoff_t max = end_byte >> PAGE_SHIFT; 472 473 if (end_byte < start_byte) 474 return false; 475 476 rcu_read_lock(); 477 for (;;) { 478 page = xas_find(&xas, max); 479 if (xas_retry(&xas, page)) 480 continue; 481 /* Shadow entries don't count */ 482 if (xa_is_value(page)) 483 continue; 484 /* 485 * We don't need to try to pin this page; we're about to 486 * release the RCU lock anyway. It is enough to know that 487 * there was a page here recently. 488 */ 489 break; 490 } 491 rcu_read_unlock(); 492 493 return page != NULL; 494 } 495 EXPORT_SYMBOL(filemap_range_has_page); 496 497 static void __filemap_fdatawait_range(struct address_space *mapping, 498 loff_t start_byte, loff_t end_byte) 499 { 500 pgoff_t index = start_byte >> PAGE_SHIFT; 501 pgoff_t end = end_byte >> PAGE_SHIFT; 502 struct pagevec pvec; 503 int nr_pages; 504 505 if (end_byte < start_byte) 506 return; 507 508 pagevec_init(&pvec); 509 while (index <= end) { 510 unsigned i; 511 512 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, 513 end, PAGECACHE_TAG_WRITEBACK); 514 if (!nr_pages) 515 break; 516 517 for (i = 0; i < nr_pages; i++) { 518 struct page *page = pvec.pages[i]; 519 520 wait_on_page_writeback(page); 521 ClearPageError(page); 522 } 523 pagevec_release(&pvec); 524 cond_resched(); 525 } 526 } 527 528 /** 529 * filemap_fdatawait_range - wait for writeback to complete 530 * @mapping: address space structure to wait for 531 * @start_byte: offset in bytes where the range starts 532 * @end_byte: offset in bytes where the range ends (inclusive) 533 * 534 * Walk the list of under-writeback pages of the given address space 535 * in the given range and wait for all of them. Check error status of 536 * the address space and return it. 537 * 538 * Since the error status of the address space is cleared by this function, 539 * callers are responsible for checking the return value and handling and/or 540 * reporting the error. 541 * 542 * Return: error status of the address space. 543 */ 544 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, 545 loff_t end_byte) 546 { 547 __filemap_fdatawait_range(mapping, start_byte, end_byte); 548 return filemap_check_errors(mapping); 549 } 550 EXPORT_SYMBOL(filemap_fdatawait_range); 551 552 /** 553 * filemap_fdatawait_range_keep_errors - wait for writeback to complete 554 * @mapping: address space structure to wait for 555 * @start_byte: offset in bytes where the range starts 556 * @end_byte: offset in bytes where the range ends (inclusive) 557 * 558 * Walk the list of under-writeback pages of the given address space in the 559 * given range and wait for all of them. Unlike filemap_fdatawait_range(), 560 * this function does not clear error status of the address space. 561 * 562 * Use this function if callers don't handle errors themselves. Expected 563 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), 564 * fsfreeze(8) 565 */ 566 int filemap_fdatawait_range_keep_errors(struct address_space *mapping, 567 loff_t start_byte, loff_t end_byte) 568 { 569 __filemap_fdatawait_range(mapping, start_byte, end_byte); 570 return filemap_check_and_keep_errors(mapping); 571 } 572 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors); 573 574 /** 575 * file_fdatawait_range - wait for writeback to complete 576 * @file: file pointing to address space structure to wait for 577 * @start_byte: offset in bytes where the range starts 578 * @end_byte: offset in bytes where the range ends (inclusive) 579 * 580 * Walk the list of under-writeback pages of the address space that file 581 * refers to, in the given range and wait for all of them. Check error 582 * status of the address space vs. the file->f_wb_err cursor and return it. 583 * 584 * Since the error status of the file is advanced by this function, 585 * callers are responsible for checking the return value and handling and/or 586 * reporting the error. 587 * 588 * Return: error status of the address space vs. the file->f_wb_err cursor. 589 */ 590 int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte) 591 { 592 struct address_space *mapping = file->f_mapping; 593 594 __filemap_fdatawait_range(mapping, start_byte, end_byte); 595 return file_check_and_advance_wb_err(file); 596 } 597 EXPORT_SYMBOL(file_fdatawait_range); 598 599 /** 600 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors 601 * @mapping: address space structure to wait for 602 * 603 * Walk the list of under-writeback pages of the given address space 604 * and wait for all of them. Unlike filemap_fdatawait(), this function 605 * does not clear error status of the address space. 606 * 607 * Use this function if callers don't handle errors themselves. Expected 608 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), 609 * fsfreeze(8) 610 * 611 * Return: error status of the address space. 612 */ 613 int filemap_fdatawait_keep_errors(struct address_space *mapping) 614 { 615 __filemap_fdatawait_range(mapping, 0, LLONG_MAX); 616 return filemap_check_and_keep_errors(mapping); 617 } 618 EXPORT_SYMBOL(filemap_fdatawait_keep_errors); 619 620 /* Returns true if writeback might be needed or already in progress. */ 621 static bool mapping_needs_writeback(struct address_space *mapping) 622 { 623 return mapping->nrpages; 624 } 625 626 bool filemap_range_has_writeback(struct address_space *mapping, 627 loff_t start_byte, loff_t end_byte) 628 { 629 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT); 630 pgoff_t max = end_byte >> PAGE_SHIFT; 631 struct page *page; 632 633 if (end_byte < start_byte) 634 return false; 635 636 rcu_read_lock(); 637 xas_for_each(&xas, page, max) { 638 if (xas_retry(&xas, page)) 639 continue; 640 if (xa_is_value(page)) 641 continue; 642 if (PageDirty(page) || PageLocked(page) || PageWriteback(page)) 643 break; 644 } 645 rcu_read_unlock(); 646 return page != NULL; 647 } 648 EXPORT_SYMBOL_GPL(filemap_range_has_writeback); 649 650 /** 651 * filemap_write_and_wait_range - write out & wait on a file range 652 * @mapping: the address_space for the pages 653 * @lstart: offset in bytes where the range starts 654 * @lend: offset in bytes where the range ends (inclusive) 655 * 656 * Write out and wait upon file offsets lstart->lend, inclusive. 657 * 658 * Note that @lend is inclusive (describes the last byte to be written) so 659 * that this function can be used to write to the very end-of-file (end = -1). 660 * 661 * Return: error status of the address space. 662 */ 663 int filemap_write_and_wait_range(struct address_space *mapping, 664 loff_t lstart, loff_t lend) 665 { 666 int err = 0; 667 668 if (mapping_needs_writeback(mapping)) { 669 err = __filemap_fdatawrite_range(mapping, lstart, lend, 670 WB_SYNC_ALL); 671 /* 672 * Even if the above returned error, the pages may be 673 * written partially (e.g. -ENOSPC), so we wait for it. 674 * But the -EIO is special case, it may indicate the worst 675 * thing (e.g. bug) happened, so we avoid waiting for it. 676 */ 677 if (err != -EIO) { 678 int err2 = filemap_fdatawait_range(mapping, 679 lstart, lend); 680 if (!err) 681 err = err2; 682 } else { 683 /* Clear any previously stored errors */ 684 filemap_check_errors(mapping); 685 } 686 } else { 687 err = filemap_check_errors(mapping); 688 } 689 return err; 690 } 691 EXPORT_SYMBOL(filemap_write_and_wait_range); 692 693 void __filemap_set_wb_err(struct address_space *mapping, int err) 694 { 695 errseq_t eseq = errseq_set(&mapping->wb_err, err); 696 697 trace_filemap_set_wb_err(mapping, eseq); 698 } 699 EXPORT_SYMBOL(__filemap_set_wb_err); 700 701 /** 702 * file_check_and_advance_wb_err - report wb error (if any) that was previously 703 * and advance wb_err to current one 704 * @file: struct file on which the error is being reported 705 * 706 * When userland calls fsync (or something like nfsd does the equivalent), we 707 * want to report any writeback errors that occurred since the last fsync (or 708 * since the file was opened if there haven't been any). 709 * 710 * Grab the wb_err from the mapping. If it matches what we have in the file, 711 * then just quickly return 0. The file is all caught up. 712 * 713 * If it doesn't match, then take the mapping value, set the "seen" flag in 714 * it and try to swap it into place. If it works, or another task beat us 715 * to it with the new value, then update the f_wb_err and return the error 716 * portion. The error at this point must be reported via proper channels 717 * (a'la fsync, or NFS COMMIT operation, etc.). 718 * 719 * While we handle mapping->wb_err with atomic operations, the f_wb_err 720 * value is protected by the f_lock since we must ensure that it reflects 721 * the latest value swapped in for this file descriptor. 722 * 723 * Return: %0 on success, negative error code otherwise. 724 */ 725 int file_check_and_advance_wb_err(struct file *file) 726 { 727 int err = 0; 728 errseq_t old = READ_ONCE(file->f_wb_err); 729 struct address_space *mapping = file->f_mapping; 730 731 /* Locklessly handle the common case where nothing has changed */ 732 if (errseq_check(&mapping->wb_err, old)) { 733 /* Something changed, must use slow path */ 734 spin_lock(&file->f_lock); 735 old = file->f_wb_err; 736 err = errseq_check_and_advance(&mapping->wb_err, 737 &file->f_wb_err); 738 trace_file_check_and_advance_wb_err(file, old); 739 spin_unlock(&file->f_lock); 740 } 741 742 /* 743 * We're mostly using this function as a drop in replacement for 744 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect 745 * that the legacy code would have had on these flags. 746 */ 747 clear_bit(AS_EIO, &mapping->flags); 748 clear_bit(AS_ENOSPC, &mapping->flags); 749 return err; 750 } 751 EXPORT_SYMBOL(file_check_and_advance_wb_err); 752 753 /** 754 * file_write_and_wait_range - write out & wait on a file range 755 * @file: file pointing to address_space with pages 756 * @lstart: offset in bytes where the range starts 757 * @lend: offset in bytes where the range ends (inclusive) 758 * 759 * Write out and wait upon file offsets lstart->lend, inclusive. 760 * 761 * Note that @lend is inclusive (describes the last byte to be written) so 762 * that this function can be used to write to the very end-of-file (end = -1). 763 * 764 * After writing out and waiting on the data, we check and advance the 765 * f_wb_err cursor to the latest value, and return any errors detected there. 766 * 767 * Return: %0 on success, negative error code otherwise. 768 */ 769 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend) 770 { 771 int err = 0, err2; 772 struct address_space *mapping = file->f_mapping; 773 774 if (mapping_needs_writeback(mapping)) { 775 err = __filemap_fdatawrite_range(mapping, lstart, lend, 776 WB_SYNC_ALL); 777 /* See comment of filemap_write_and_wait() */ 778 if (err != -EIO) 779 __filemap_fdatawait_range(mapping, lstart, lend); 780 } 781 err2 = file_check_and_advance_wb_err(file); 782 if (!err) 783 err = err2; 784 return err; 785 } 786 EXPORT_SYMBOL(file_write_and_wait_range); 787 788 /** 789 * replace_page_cache_page - replace a pagecache page with a new one 790 * @old: page to be replaced 791 * @new: page to replace with 792 * 793 * This function replaces a page in the pagecache with a new one. On 794 * success it acquires the pagecache reference for the new page and 795 * drops it for the old page. Both the old and new pages must be 796 * locked. This function does not add the new page to the LRU, the 797 * caller must do that. 798 * 799 * The remove + add is atomic. This function cannot fail. 800 */ 801 void replace_page_cache_page(struct page *old, struct page *new) 802 { 803 struct folio *fold = page_folio(old); 804 struct folio *fnew = page_folio(new); 805 struct address_space *mapping = old->mapping; 806 void (*freepage)(struct page *) = mapping->a_ops->freepage; 807 pgoff_t offset = old->index; 808 XA_STATE(xas, &mapping->i_pages, offset); 809 810 VM_BUG_ON_PAGE(!PageLocked(old), old); 811 VM_BUG_ON_PAGE(!PageLocked(new), new); 812 VM_BUG_ON_PAGE(new->mapping, new); 813 814 get_page(new); 815 new->mapping = mapping; 816 new->index = offset; 817 818 mem_cgroup_migrate(fold, fnew); 819 820 xas_lock_irq(&xas); 821 xas_store(&xas, new); 822 823 old->mapping = NULL; 824 /* hugetlb pages do not participate in page cache accounting. */ 825 if (!PageHuge(old)) 826 __dec_lruvec_page_state(old, NR_FILE_PAGES); 827 if (!PageHuge(new)) 828 __inc_lruvec_page_state(new, NR_FILE_PAGES); 829 if (PageSwapBacked(old)) 830 __dec_lruvec_page_state(old, NR_SHMEM); 831 if (PageSwapBacked(new)) 832 __inc_lruvec_page_state(new, NR_SHMEM); 833 xas_unlock_irq(&xas); 834 if (freepage) 835 freepage(old); 836 put_page(old); 837 } 838 EXPORT_SYMBOL_GPL(replace_page_cache_page); 839 840 noinline int __filemap_add_folio(struct address_space *mapping, 841 struct folio *folio, pgoff_t index, gfp_t gfp, void **shadowp) 842 { 843 XA_STATE(xas, &mapping->i_pages, index); 844 int huge = folio_test_hugetlb(folio); 845 bool charged = false; 846 long nr = 1; 847 848 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); 849 VM_BUG_ON_FOLIO(folio_test_swapbacked(folio), folio); 850 mapping_set_update(&xas, mapping); 851 852 if (!huge) { 853 int error = mem_cgroup_charge(folio, NULL, gfp); 854 VM_BUG_ON_FOLIO(index & (folio_nr_pages(folio) - 1), folio); 855 if (error) 856 return error; 857 charged = true; 858 xas_set_order(&xas, index, folio_order(folio)); 859 nr = folio_nr_pages(folio); 860 } 861 862 gfp &= GFP_RECLAIM_MASK; 863 folio_ref_add(folio, nr); 864 folio->mapping = mapping; 865 folio->index = xas.xa_index; 866 867 do { 868 unsigned int order = xa_get_order(xas.xa, xas.xa_index); 869 void *entry, *old = NULL; 870 871 if (order > folio_order(folio)) 872 xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index), 873 order, gfp); 874 xas_lock_irq(&xas); 875 xas_for_each_conflict(&xas, entry) { 876 old = entry; 877 if (!xa_is_value(entry)) { 878 xas_set_err(&xas, -EEXIST); 879 goto unlock; 880 } 881 } 882 883 if (old) { 884 if (shadowp) 885 *shadowp = old; 886 /* entry may have been split before we acquired lock */ 887 order = xa_get_order(xas.xa, xas.xa_index); 888 if (order > folio_order(folio)) { 889 /* How to handle large swap entries? */ 890 BUG_ON(shmem_mapping(mapping)); 891 xas_split(&xas, old, order); 892 xas_reset(&xas); 893 } 894 } 895 896 xas_store(&xas, folio); 897 if (xas_error(&xas)) 898 goto unlock; 899 900 mapping->nrpages += nr; 901 902 /* hugetlb pages do not participate in page cache accounting */ 903 if (!huge) { 904 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr); 905 if (folio_test_pmd_mappable(folio)) 906 __lruvec_stat_mod_folio(folio, 907 NR_FILE_THPS, nr); 908 } 909 unlock: 910 xas_unlock_irq(&xas); 911 } while (xas_nomem(&xas, gfp)); 912 913 if (xas_error(&xas)) 914 goto error; 915 916 trace_mm_filemap_add_to_page_cache(folio); 917 return 0; 918 error: 919 if (charged) 920 mem_cgroup_uncharge(folio); 921 folio->mapping = NULL; 922 /* Leave page->index set: truncation relies upon it */ 923 folio_put_refs(folio, nr); 924 return xas_error(&xas); 925 } 926 ALLOW_ERROR_INJECTION(__filemap_add_folio, ERRNO); 927 928 /** 929 * add_to_page_cache_locked - add a locked page to the pagecache 930 * @page: page to add 931 * @mapping: the page's address_space 932 * @offset: page index 933 * @gfp_mask: page allocation mode 934 * 935 * This function is used to add a page to the pagecache. It must be locked. 936 * This function does not add the page to the LRU. The caller must do that. 937 * 938 * Return: %0 on success, negative error code otherwise. 939 */ 940 int add_to_page_cache_locked(struct page *page, struct address_space *mapping, 941 pgoff_t offset, gfp_t gfp_mask) 942 { 943 return __filemap_add_folio(mapping, page_folio(page), offset, 944 gfp_mask, NULL); 945 } 946 EXPORT_SYMBOL(add_to_page_cache_locked); 947 948 int filemap_add_folio(struct address_space *mapping, struct folio *folio, 949 pgoff_t index, gfp_t gfp) 950 { 951 void *shadow = NULL; 952 int ret; 953 954 __folio_set_locked(folio); 955 ret = __filemap_add_folio(mapping, folio, index, gfp, &shadow); 956 if (unlikely(ret)) 957 __folio_clear_locked(folio); 958 else { 959 /* 960 * The folio might have been evicted from cache only 961 * recently, in which case it should be activated like 962 * any other repeatedly accessed folio. 963 * The exception is folios getting rewritten; evicting other 964 * data from the working set, only to cache data that will 965 * get overwritten with something else, is a waste of memory. 966 */ 967 WARN_ON_ONCE(folio_test_active(folio)); 968 if (!(gfp & __GFP_WRITE) && shadow) 969 workingset_refault(folio, shadow); 970 folio_add_lru(folio); 971 } 972 return ret; 973 } 974 EXPORT_SYMBOL_GPL(filemap_add_folio); 975 976 #ifdef CONFIG_NUMA 977 struct folio *filemap_alloc_folio(gfp_t gfp, unsigned int order) 978 { 979 int n; 980 struct folio *folio; 981 982 if (cpuset_do_page_mem_spread()) { 983 unsigned int cpuset_mems_cookie; 984 do { 985 cpuset_mems_cookie = read_mems_allowed_begin(); 986 n = cpuset_mem_spread_node(); 987 folio = __folio_alloc_node(gfp, order, n); 988 } while (!folio && read_mems_allowed_retry(cpuset_mems_cookie)); 989 990 return folio; 991 } 992 return folio_alloc(gfp, order); 993 } 994 EXPORT_SYMBOL(filemap_alloc_folio); 995 #endif 996 997 /* 998 * filemap_invalidate_lock_two - lock invalidate_lock for two mappings 999 * 1000 * Lock exclusively invalidate_lock of any passed mapping that is not NULL. 1001 * 1002 * @mapping1: the first mapping to lock 1003 * @mapping2: the second mapping to lock 1004 */ 1005 void filemap_invalidate_lock_two(struct address_space *mapping1, 1006 struct address_space *mapping2) 1007 { 1008 if (mapping1 > mapping2) 1009 swap(mapping1, mapping2); 1010 if (mapping1) 1011 down_write(&mapping1->invalidate_lock); 1012 if (mapping2 && mapping1 != mapping2) 1013 down_write_nested(&mapping2->invalidate_lock, 1); 1014 } 1015 EXPORT_SYMBOL(filemap_invalidate_lock_two); 1016 1017 /* 1018 * filemap_invalidate_unlock_two - unlock invalidate_lock for two mappings 1019 * 1020 * Unlock exclusive invalidate_lock of any passed mapping that is not NULL. 1021 * 1022 * @mapping1: the first mapping to unlock 1023 * @mapping2: the second mapping to unlock 1024 */ 1025 void filemap_invalidate_unlock_two(struct address_space *mapping1, 1026 struct address_space *mapping2) 1027 { 1028 if (mapping1) 1029 up_write(&mapping1->invalidate_lock); 1030 if (mapping2 && mapping1 != mapping2) 1031 up_write(&mapping2->invalidate_lock); 1032 } 1033 EXPORT_SYMBOL(filemap_invalidate_unlock_two); 1034 1035 /* 1036 * In order to wait for pages to become available there must be 1037 * waitqueues associated with pages. By using a hash table of 1038 * waitqueues where the bucket discipline is to maintain all 1039 * waiters on the same queue and wake all when any of the pages 1040 * become available, and for the woken contexts to check to be 1041 * sure the appropriate page became available, this saves space 1042 * at a cost of "thundering herd" phenomena during rare hash 1043 * collisions. 1044 */ 1045 #define PAGE_WAIT_TABLE_BITS 8 1046 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS) 1047 static wait_queue_head_t folio_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned; 1048 1049 static wait_queue_head_t *folio_waitqueue(struct folio *folio) 1050 { 1051 return &folio_wait_table[hash_ptr(folio, PAGE_WAIT_TABLE_BITS)]; 1052 } 1053 1054 void __init pagecache_init(void) 1055 { 1056 int i; 1057 1058 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++) 1059 init_waitqueue_head(&folio_wait_table[i]); 1060 1061 page_writeback_init(); 1062 1063 /* 1064 * tmpfs uses the ZERO_PAGE for reading holes: it is up-to-date, 1065 * and splice's page_cache_pipe_buf_confirm() needs to see that. 1066 */ 1067 SetPageUptodate(ZERO_PAGE(0)); 1068 } 1069 1070 /* 1071 * The page wait code treats the "wait->flags" somewhat unusually, because 1072 * we have multiple different kinds of waits, not just the usual "exclusive" 1073 * one. 1074 * 1075 * We have: 1076 * 1077 * (a) no special bits set: 1078 * 1079 * We're just waiting for the bit to be released, and when a waker 1080 * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up, 1081 * and remove it from the wait queue. 1082 * 1083 * Simple and straightforward. 1084 * 1085 * (b) WQ_FLAG_EXCLUSIVE: 1086 * 1087 * The waiter is waiting to get the lock, and only one waiter should 1088 * be woken up to avoid any thundering herd behavior. We'll set the 1089 * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue. 1090 * 1091 * This is the traditional exclusive wait. 1092 * 1093 * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM: 1094 * 1095 * The waiter is waiting to get the bit, and additionally wants the 1096 * lock to be transferred to it for fair lock behavior. If the lock 1097 * cannot be taken, we stop walking the wait queue without waking 1098 * the waiter. 1099 * 1100 * This is the "fair lock handoff" case, and in addition to setting 1101 * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see 1102 * that it now has the lock. 1103 */ 1104 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg) 1105 { 1106 unsigned int flags; 1107 struct wait_page_key *key = arg; 1108 struct wait_page_queue *wait_page 1109 = container_of(wait, struct wait_page_queue, wait); 1110 1111 if (!wake_page_match(wait_page, key)) 1112 return 0; 1113 1114 /* 1115 * If it's a lock handoff wait, we get the bit for it, and 1116 * stop walking (and do not wake it up) if we can't. 1117 */ 1118 flags = wait->flags; 1119 if (flags & WQ_FLAG_EXCLUSIVE) { 1120 if (test_bit(key->bit_nr, &key->folio->flags)) 1121 return -1; 1122 if (flags & WQ_FLAG_CUSTOM) { 1123 if (test_and_set_bit(key->bit_nr, &key->folio->flags)) 1124 return -1; 1125 flags |= WQ_FLAG_DONE; 1126 } 1127 } 1128 1129 /* 1130 * We are holding the wait-queue lock, but the waiter that 1131 * is waiting for this will be checking the flags without 1132 * any locking. 1133 * 1134 * So update the flags atomically, and wake up the waiter 1135 * afterwards to avoid any races. This store-release pairs 1136 * with the load-acquire in folio_wait_bit_common(). 1137 */ 1138 smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN); 1139 wake_up_state(wait->private, mode); 1140 1141 /* 1142 * Ok, we have successfully done what we're waiting for, 1143 * and we can unconditionally remove the wait entry. 1144 * 1145 * Note that this pairs with the "finish_wait()" in the 1146 * waiter, and has to be the absolute last thing we do. 1147 * After this list_del_init(&wait->entry) the wait entry 1148 * might be de-allocated and the process might even have 1149 * exited. 1150 */ 1151 list_del_init_careful(&wait->entry); 1152 return (flags & WQ_FLAG_EXCLUSIVE) != 0; 1153 } 1154 1155 static void folio_wake_bit(struct folio *folio, int bit_nr) 1156 { 1157 wait_queue_head_t *q = folio_waitqueue(folio); 1158 struct wait_page_key key; 1159 unsigned long flags; 1160 wait_queue_entry_t bookmark; 1161 1162 key.folio = folio; 1163 key.bit_nr = bit_nr; 1164 key.page_match = 0; 1165 1166 bookmark.flags = 0; 1167 bookmark.private = NULL; 1168 bookmark.func = NULL; 1169 INIT_LIST_HEAD(&bookmark.entry); 1170 1171 spin_lock_irqsave(&q->lock, flags); 1172 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); 1173 1174 while (bookmark.flags & WQ_FLAG_BOOKMARK) { 1175 /* 1176 * Take a breather from holding the lock, 1177 * allow pages that finish wake up asynchronously 1178 * to acquire the lock and remove themselves 1179 * from wait queue 1180 */ 1181 spin_unlock_irqrestore(&q->lock, flags); 1182 cpu_relax(); 1183 spin_lock_irqsave(&q->lock, flags); 1184 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); 1185 } 1186 1187 /* 1188 * It is possible for other pages to have collided on the waitqueue 1189 * hash, so in that case check for a page match. That prevents a long- 1190 * term waiter 1191 * 1192 * It is still possible to miss a case here, when we woke page waiters 1193 * and removed them from the waitqueue, but there are still other 1194 * page waiters. 1195 */ 1196 if (!waitqueue_active(q) || !key.page_match) { 1197 folio_clear_waiters(folio); 1198 /* 1199 * It's possible to miss clearing Waiters here, when we woke 1200 * our page waiters, but the hashed waitqueue has waiters for 1201 * other pages on it. 1202 * 1203 * That's okay, it's a rare case. The next waker will clear it. 1204 */ 1205 } 1206 spin_unlock_irqrestore(&q->lock, flags); 1207 } 1208 1209 static void folio_wake(struct folio *folio, int bit) 1210 { 1211 if (!folio_test_waiters(folio)) 1212 return; 1213 folio_wake_bit(folio, bit); 1214 } 1215 1216 /* 1217 * A choice of three behaviors for folio_wait_bit_common(): 1218 */ 1219 enum behavior { 1220 EXCLUSIVE, /* Hold ref to page and take the bit when woken, like 1221 * __folio_lock() waiting on then setting PG_locked. 1222 */ 1223 SHARED, /* Hold ref to page and check the bit when woken, like 1224 * folio_wait_writeback() waiting on PG_writeback. 1225 */ 1226 DROP, /* Drop ref to page before wait, no check when woken, 1227 * like folio_put_wait_locked() on PG_locked. 1228 */ 1229 }; 1230 1231 /* 1232 * Attempt to check (or get) the folio flag, and mark us done 1233 * if successful. 1234 */ 1235 static inline bool folio_trylock_flag(struct folio *folio, int bit_nr, 1236 struct wait_queue_entry *wait) 1237 { 1238 if (wait->flags & WQ_FLAG_EXCLUSIVE) { 1239 if (test_and_set_bit(bit_nr, &folio->flags)) 1240 return false; 1241 } else if (test_bit(bit_nr, &folio->flags)) 1242 return false; 1243 1244 wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE; 1245 return true; 1246 } 1247 1248 /* How many times do we accept lock stealing from under a waiter? */ 1249 int sysctl_page_lock_unfairness = 5; 1250 1251 static inline int folio_wait_bit_common(struct folio *folio, int bit_nr, 1252 int state, enum behavior behavior) 1253 { 1254 wait_queue_head_t *q = folio_waitqueue(folio); 1255 int unfairness = sysctl_page_lock_unfairness; 1256 struct wait_page_queue wait_page; 1257 wait_queue_entry_t *wait = &wait_page.wait; 1258 bool thrashing = false; 1259 bool delayacct = false; 1260 unsigned long pflags; 1261 1262 if (bit_nr == PG_locked && 1263 !folio_test_uptodate(folio) && folio_test_workingset(folio)) { 1264 if (!folio_test_swapbacked(folio)) { 1265 delayacct_thrashing_start(); 1266 delayacct = true; 1267 } 1268 psi_memstall_enter(&pflags); 1269 thrashing = true; 1270 } 1271 1272 init_wait(wait); 1273 wait->func = wake_page_function; 1274 wait_page.folio = folio; 1275 wait_page.bit_nr = bit_nr; 1276 1277 repeat: 1278 wait->flags = 0; 1279 if (behavior == EXCLUSIVE) { 1280 wait->flags = WQ_FLAG_EXCLUSIVE; 1281 if (--unfairness < 0) 1282 wait->flags |= WQ_FLAG_CUSTOM; 1283 } 1284 1285 /* 1286 * Do one last check whether we can get the 1287 * page bit synchronously. 1288 * 1289 * Do the folio_set_waiters() marking before that 1290 * to let any waker we _just_ missed know they 1291 * need to wake us up (otherwise they'll never 1292 * even go to the slow case that looks at the 1293 * page queue), and add ourselves to the wait 1294 * queue if we need to sleep. 1295 * 1296 * This part needs to be done under the queue 1297 * lock to avoid races. 1298 */ 1299 spin_lock_irq(&q->lock); 1300 folio_set_waiters(folio); 1301 if (!folio_trylock_flag(folio, bit_nr, wait)) 1302 __add_wait_queue_entry_tail(q, wait); 1303 spin_unlock_irq(&q->lock); 1304 1305 /* 1306 * From now on, all the logic will be based on 1307 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to 1308 * see whether the page bit testing has already 1309 * been done by the wake function. 1310 * 1311 * We can drop our reference to the folio. 1312 */ 1313 if (behavior == DROP) 1314 folio_put(folio); 1315 1316 /* 1317 * Note that until the "finish_wait()", or until 1318 * we see the WQ_FLAG_WOKEN flag, we need to 1319 * be very careful with the 'wait->flags', because 1320 * we may race with a waker that sets them. 1321 */ 1322 for (;;) { 1323 unsigned int flags; 1324 1325 set_current_state(state); 1326 1327 /* Loop until we've been woken or interrupted */ 1328 flags = smp_load_acquire(&wait->flags); 1329 if (!(flags & WQ_FLAG_WOKEN)) { 1330 if (signal_pending_state(state, current)) 1331 break; 1332 1333 io_schedule(); 1334 continue; 1335 } 1336 1337 /* If we were non-exclusive, we're done */ 1338 if (behavior != EXCLUSIVE) 1339 break; 1340 1341 /* If the waker got the lock for us, we're done */ 1342 if (flags & WQ_FLAG_DONE) 1343 break; 1344 1345 /* 1346 * Otherwise, if we're getting the lock, we need to 1347 * try to get it ourselves. 1348 * 1349 * And if that fails, we'll have to retry this all. 1350 */ 1351 if (unlikely(test_and_set_bit(bit_nr, folio_flags(folio, 0)))) 1352 goto repeat; 1353 1354 wait->flags |= WQ_FLAG_DONE; 1355 break; 1356 } 1357 1358 /* 1359 * If a signal happened, this 'finish_wait()' may remove the last 1360 * waiter from the wait-queues, but the folio waiters bit will remain 1361 * set. That's ok. The next wakeup will take care of it, and trying 1362 * to do it here would be difficult and prone to races. 1363 */ 1364 finish_wait(q, wait); 1365 1366 if (thrashing) { 1367 if (delayacct) 1368 delayacct_thrashing_end(); 1369 psi_memstall_leave(&pflags); 1370 } 1371 1372 /* 1373 * NOTE! The wait->flags weren't stable until we've done the 1374 * 'finish_wait()', and we could have exited the loop above due 1375 * to a signal, and had a wakeup event happen after the signal 1376 * test but before the 'finish_wait()'. 1377 * 1378 * So only after the finish_wait() can we reliably determine 1379 * if we got woken up or not, so we can now figure out the final 1380 * return value based on that state without races. 1381 * 1382 * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive 1383 * waiter, but an exclusive one requires WQ_FLAG_DONE. 1384 */ 1385 if (behavior == EXCLUSIVE) 1386 return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR; 1387 1388 return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR; 1389 } 1390 1391 #ifdef CONFIG_MIGRATION 1392 /** 1393 * migration_entry_wait_on_locked - Wait for a migration entry to be removed 1394 * @entry: migration swap entry. 1395 * @ptep: mapped pte pointer. Will return with the ptep unmapped. Only required 1396 * for pte entries, pass NULL for pmd entries. 1397 * @ptl: already locked ptl. This function will drop the lock. 1398 * 1399 * Wait for a migration entry referencing the given page to be removed. This is 1400 * equivalent to put_and_wait_on_page_locked(page, TASK_UNINTERRUPTIBLE) except 1401 * this can be called without taking a reference on the page. Instead this 1402 * should be called while holding the ptl for the migration entry referencing 1403 * the page. 1404 * 1405 * Returns after unmapping and unlocking the pte/ptl with pte_unmap_unlock(). 1406 * 1407 * This follows the same logic as folio_wait_bit_common() so see the comments 1408 * there. 1409 */ 1410 void migration_entry_wait_on_locked(swp_entry_t entry, pte_t *ptep, 1411 spinlock_t *ptl) 1412 { 1413 struct wait_page_queue wait_page; 1414 wait_queue_entry_t *wait = &wait_page.wait; 1415 bool thrashing = false; 1416 bool delayacct = false; 1417 unsigned long pflags; 1418 wait_queue_head_t *q; 1419 struct folio *folio = page_folio(pfn_swap_entry_to_page(entry)); 1420 1421 q = folio_waitqueue(folio); 1422 if (!folio_test_uptodate(folio) && folio_test_workingset(folio)) { 1423 if (!folio_test_swapbacked(folio)) { 1424 delayacct_thrashing_start(); 1425 delayacct = true; 1426 } 1427 psi_memstall_enter(&pflags); 1428 thrashing = true; 1429 } 1430 1431 init_wait(wait); 1432 wait->func = wake_page_function; 1433 wait_page.folio = folio; 1434 wait_page.bit_nr = PG_locked; 1435 wait->flags = 0; 1436 1437 spin_lock_irq(&q->lock); 1438 folio_set_waiters(folio); 1439 if (!folio_trylock_flag(folio, PG_locked, wait)) 1440 __add_wait_queue_entry_tail(q, wait); 1441 spin_unlock_irq(&q->lock); 1442 1443 /* 1444 * If a migration entry exists for the page the migration path must hold 1445 * a valid reference to the page, and it must take the ptl to remove the 1446 * migration entry. So the page is valid until the ptl is dropped. 1447 */ 1448 if (ptep) 1449 pte_unmap_unlock(ptep, ptl); 1450 else 1451 spin_unlock(ptl); 1452 1453 for (;;) { 1454 unsigned int flags; 1455 1456 set_current_state(TASK_UNINTERRUPTIBLE); 1457 1458 /* Loop until we've been woken or interrupted */ 1459 flags = smp_load_acquire(&wait->flags); 1460 if (!(flags & WQ_FLAG_WOKEN)) { 1461 if (signal_pending_state(TASK_UNINTERRUPTIBLE, current)) 1462 break; 1463 1464 io_schedule(); 1465 continue; 1466 } 1467 break; 1468 } 1469 1470 finish_wait(q, wait); 1471 1472 if (thrashing) { 1473 if (delayacct) 1474 delayacct_thrashing_end(); 1475 psi_memstall_leave(&pflags); 1476 } 1477 } 1478 #endif 1479 1480 void folio_wait_bit(struct folio *folio, int bit_nr) 1481 { 1482 folio_wait_bit_common(folio, bit_nr, TASK_UNINTERRUPTIBLE, SHARED); 1483 } 1484 EXPORT_SYMBOL(folio_wait_bit); 1485 1486 int folio_wait_bit_killable(struct folio *folio, int bit_nr) 1487 { 1488 return folio_wait_bit_common(folio, bit_nr, TASK_KILLABLE, SHARED); 1489 } 1490 EXPORT_SYMBOL(folio_wait_bit_killable); 1491 1492 /** 1493 * folio_put_wait_locked - Drop a reference and wait for it to be unlocked 1494 * @folio: The folio to wait for. 1495 * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc). 1496 * 1497 * The caller should hold a reference on @folio. They expect the page to 1498 * become unlocked relatively soon, but do not wish to hold up migration 1499 * (for example) by holding the reference while waiting for the folio to 1500 * come unlocked. After this function returns, the caller should not 1501 * dereference @folio. 1502 * 1503 * Return: 0 if the folio was unlocked or -EINTR if interrupted by a signal. 1504 */ 1505 int folio_put_wait_locked(struct folio *folio, int state) 1506 { 1507 return folio_wait_bit_common(folio, PG_locked, state, DROP); 1508 } 1509 1510 /** 1511 * folio_add_wait_queue - Add an arbitrary waiter to a folio's wait queue 1512 * @folio: Folio defining the wait queue of interest 1513 * @waiter: Waiter to add to the queue 1514 * 1515 * Add an arbitrary @waiter to the wait queue for the nominated @folio. 1516 */ 1517 void folio_add_wait_queue(struct folio *folio, wait_queue_entry_t *waiter) 1518 { 1519 wait_queue_head_t *q = folio_waitqueue(folio); 1520 unsigned long flags; 1521 1522 spin_lock_irqsave(&q->lock, flags); 1523 __add_wait_queue_entry_tail(q, waiter); 1524 folio_set_waiters(folio); 1525 spin_unlock_irqrestore(&q->lock, flags); 1526 } 1527 EXPORT_SYMBOL_GPL(folio_add_wait_queue); 1528 1529 #ifndef clear_bit_unlock_is_negative_byte 1530 1531 /* 1532 * PG_waiters is the high bit in the same byte as PG_lock. 1533 * 1534 * On x86 (and on many other architectures), we can clear PG_lock and 1535 * test the sign bit at the same time. But if the architecture does 1536 * not support that special operation, we just do this all by hand 1537 * instead. 1538 * 1539 * The read of PG_waiters has to be after (or concurrently with) PG_locked 1540 * being cleared, but a memory barrier should be unnecessary since it is 1541 * in the same byte as PG_locked. 1542 */ 1543 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem) 1544 { 1545 clear_bit_unlock(nr, mem); 1546 /* smp_mb__after_atomic(); */ 1547 return test_bit(PG_waiters, mem); 1548 } 1549 1550 #endif 1551 1552 /** 1553 * folio_unlock - Unlock a locked folio. 1554 * @folio: The folio. 1555 * 1556 * Unlocks the folio and wakes up any thread sleeping on the page lock. 1557 * 1558 * Context: May be called from interrupt or process context. May not be 1559 * called from NMI context. 1560 */ 1561 void folio_unlock(struct folio *folio) 1562 { 1563 /* Bit 7 allows x86 to check the byte's sign bit */ 1564 BUILD_BUG_ON(PG_waiters != 7); 1565 BUILD_BUG_ON(PG_locked > 7); 1566 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); 1567 if (clear_bit_unlock_is_negative_byte(PG_locked, folio_flags(folio, 0))) 1568 folio_wake_bit(folio, PG_locked); 1569 } 1570 EXPORT_SYMBOL(folio_unlock); 1571 1572 /** 1573 * folio_end_private_2 - Clear PG_private_2 and wake any waiters. 1574 * @folio: The folio. 1575 * 1576 * Clear the PG_private_2 bit on a folio and wake up any sleepers waiting for 1577 * it. The folio reference held for PG_private_2 being set is released. 1578 * 1579 * This is, for example, used when a netfs folio is being written to a local 1580 * disk cache, thereby allowing writes to the cache for the same folio to be 1581 * serialised. 1582 */ 1583 void folio_end_private_2(struct folio *folio) 1584 { 1585 VM_BUG_ON_FOLIO(!folio_test_private_2(folio), folio); 1586 clear_bit_unlock(PG_private_2, folio_flags(folio, 0)); 1587 folio_wake_bit(folio, PG_private_2); 1588 folio_put(folio); 1589 } 1590 EXPORT_SYMBOL(folio_end_private_2); 1591 1592 /** 1593 * folio_wait_private_2 - Wait for PG_private_2 to be cleared on a folio. 1594 * @folio: The folio to wait on. 1595 * 1596 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio. 1597 */ 1598 void folio_wait_private_2(struct folio *folio) 1599 { 1600 while (folio_test_private_2(folio)) 1601 folio_wait_bit(folio, PG_private_2); 1602 } 1603 EXPORT_SYMBOL(folio_wait_private_2); 1604 1605 /** 1606 * folio_wait_private_2_killable - Wait for PG_private_2 to be cleared on a folio. 1607 * @folio: The folio to wait on. 1608 * 1609 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio or until a 1610 * fatal signal is received by the calling task. 1611 * 1612 * Return: 1613 * - 0 if successful. 1614 * - -EINTR if a fatal signal was encountered. 1615 */ 1616 int folio_wait_private_2_killable(struct folio *folio) 1617 { 1618 int ret = 0; 1619 1620 while (folio_test_private_2(folio)) { 1621 ret = folio_wait_bit_killable(folio, PG_private_2); 1622 if (ret < 0) 1623 break; 1624 } 1625 1626 return ret; 1627 } 1628 EXPORT_SYMBOL(folio_wait_private_2_killable); 1629 1630 /** 1631 * folio_end_writeback - End writeback against a folio. 1632 * @folio: The folio. 1633 */ 1634 void folio_end_writeback(struct folio *folio) 1635 { 1636 /* 1637 * folio_test_clear_reclaim() could be used here but it is an 1638 * atomic operation and overkill in this particular case. Failing 1639 * to shuffle a folio marked for immediate reclaim is too mild 1640 * a gain to justify taking an atomic operation penalty at the 1641 * end of every folio writeback. 1642 */ 1643 if (folio_test_reclaim(folio)) { 1644 folio_clear_reclaim(folio); 1645 folio_rotate_reclaimable(folio); 1646 } 1647 1648 /* 1649 * Writeback does not hold a folio reference of its own, relying 1650 * on truncation to wait for the clearing of PG_writeback. 1651 * But here we must make sure that the folio is not freed and 1652 * reused before the folio_wake(). 1653 */ 1654 folio_get(folio); 1655 if (!__folio_end_writeback(folio)) 1656 BUG(); 1657 1658 smp_mb__after_atomic(); 1659 folio_wake(folio, PG_writeback); 1660 acct_reclaim_writeback(folio); 1661 folio_put(folio); 1662 } 1663 EXPORT_SYMBOL(folio_end_writeback); 1664 1665 /* 1666 * After completing I/O on a page, call this routine to update the page 1667 * flags appropriately 1668 */ 1669 void page_endio(struct page *page, bool is_write, int err) 1670 { 1671 if (!is_write) { 1672 if (!err) { 1673 SetPageUptodate(page); 1674 } else { 1675 ClearPageUptodate(page); 1676 SetPageError(page); 1677 } 1678 unlock_page(page); 1679 } else { 1680 if (err) { 1681 struct address_space *mapping; 1682 1683 SetPageError(page); 1684 mapping = page_mapping(page); 1685 if (mapping) 1686 mapping_set_error(mapping, err); 1687 } 1688 end_page_writeback(page); 1689 } 1690 } 1691 EXPORT_SYMBOL_GPL(page_endio); 1692 1693 /** 1694 * __folio_lock - Get a lock on the folio, assuming we need to sleep to get it. 1695 * @folio: The folio to lock 1696 */ 1697 void __folio_lock(struct folio *folio) 1698 { 1699 folio_wait_bit_common(folio, PG_locked, TASK_UNINTERRUPTIBLE, 1700 EXCLUSIVE); 1701 } 1702 EXPORT_SYMBOL(__folio_lock); 1703 1704 int __folio_lock_killable(struct folio *folio) 1705 { 1706 return folio_wait_bit_common(folio, PG_locked, TASK_KILLABLE, 1707 EXCLUSIVE); 1708 } 1709 EXPORT_SYMBOL_GPL(__folio_lock_killable); 1710 1711 static int __folio_lock_async(struct folio *folio, struct wait_page_queue *wait) 1712 { 1713 struct wait_queue_head *q = folio_waitqueue(folio); 1714 int ret = 0; 1715 1716 wait->folio = folio; 1717 wait->bit_nr = PG_locked; 1718 1719 spin_lock_irq(&q->lock); 1720 __add_wait_queue_entry_tail(q, &wait->wait); 1721 folio_set_waiters(folio); 1722 ret = !folio_trylock(folio); 1723 /* 1724 * If we were successful now, we know we're still on the 1725 * waitqueue as we're still under the lock. This means it's 1726 * safe to remove and return success, we know the callback 1727 * isn't going to trigger. 1728 */ 1729 if (!ret) 1730 __remove_wait_queue(q, &wait->wait); 1731 else 1732 ret = -EIOCBQUEUED; 1733 spin_unlock_irq(&q->lock); 1734 return ret; 1735 } 1736 1737 /* 1738 * Return values: 1739 * true - folio is locked; mmap_lock is still held. 1740 * false - folio is not locked. 1741 * mmap_lock has been released (mmap_read_unlock(), unless flags had both 1742 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in 1743 * which case mmap_lock is still held. 1744 * 1745 * If neither ALLOW_RETRY nor KILLABLE are set, will always return true 1746 * with the folio locked and the mmap_lock unperturbed. 1747 */ 1748 bool __folio_lock_or_retry(struct folio *folio, struct mm_struct *mm, 1749 unsigned int flags) 1750 { 1751 if (fault_flag_allow_retry_first(flags)) { 1752 /* 1753 * CAUTION! In this case, mmap_lock is not released 1754 * even though return 0. 1755 */ 1756 if (flags & FAULT_FLAG_RETRY_NOWAIT) 1757 return false; 1758 1759 mmap_read_unlock(mm); 1760 if (flags & FAULT_FLAG_KILLABLE) 1761 folio_wait_locked_killable(folio); 1762 else 1763 folio_wait_locked(folio); 1764 return false; 1765 } 1766 if (flags & FAULT_FLAG_KILLABLE) { 1767 bool ret; 1768 1769 ret = __folio_lock_killable(folio); 1770 if (ret) { 1771 mmap_read_unlock(mm); 1772 return false; 1773 } 1774 } else { 1775 __folio_lock(folio); 1776 } 1777 1778 return true; 1779 } 1780 1781 /** 1782 * page_cache_next_miss() - Find the next gap in the page cache. 1783 * @mapping: Mapping. 1784 * @index: Index. 1785 * @max_scan: Maximum range to search. 1786 * 1787 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the 1788 * gap with the lowest index. 1789 * 1790 * This function may be called under the rcu_read_lock. However, this will 1791 * not atomically search a snapshot of the cache at a single point in time. 1792 * For example, if a gap is created at index 5, then subsequently a gap is 1793 * created at index 10, page_cache_next_miss covering both indices may 1794 * return 10 if called under the rcu_read_lock. 1795 * 1796 * Return: The index of the gap if found, otherwise an index outside the 1797 * range specified (in which case 'return - index >= max_scan' will be true). 1798 * In the rare case of index wrap-around, 0 will be returned. 1799 */ 1800 pgoff_t page_cache_next_miss(struct address_space *mapping, 1801 pgoff_t index, unsigned long max_scan) 1802 { 1803 XA_STATE(xas, &mapping->i_pages, index); 1804 1805 while (max_scan--) { 1806 void *entry = xas_next(&xas); 1807 if (!entry || xa_is_value(entry)) 1808 break; 1809 if (xas.xa_index == 0) 1810 break; 1811 } 1812 1813 return xas.xa_index; 1814 } 1815 EXPORT_SYMBOL(page_cache_next_miss); 1816 1817 /** 1818 * page_cache_prev_miss() - Find the previous gap in the page cache. 1819 * @mapping: Mapping. 1820 * @index: Index. 1821 * @max_scan: Maximum range to search. 1822 * 1823 * Search the range [max(index - max_scan + 1, 0), index] for the 1824 * gap with the highest index. 1825 * 1826 * This function may be called under the rcu_read_lock. However, this will 1827 * not atomically search a snapshot of the cache at a single point in time. 1828 * For example, if a gap is created at index 10, then subsequently a gap is 1829 * created at index 5, page_cache_prev_miss() covering both indices may 1830 * return 5 if called under the rcu_read_lock. 1831 * 1832 * Return: The index of the gap if found, otherwise an index outside the 1833 * range specified (in which case 'index - return >= max_scan' will be true). 1834 * In the rare case of wrap-around, ULONG_MAX will be returned. 1835 */ 1836 pgoff_t page_cache_prev_miss(struct address_space *mapping, 1837 pgoff_t index, unsigned long max_scan) 1838 { 1839 XA_STATE(xas, &mapping->i_pages, index); 1840 1841 while (max_scan--) { 1842 void *entry = xas_prev(&xas); 1843 if (!entry || xa_is_value(entry)) 1844 break; 1845 if (xas.xa_index == ULONG_MAX) 1846 break; 1847 } 1848 1849 return xas.xa_index; 1850 } 1851 EXPORT_SYMBOL(page_cache_prev_miss); 1852 1853 /* 1854 * Lockless page cache protocol: 1855 * On the lookup side: 1856 * 1. Load the folio from i_pages 1857 * 2. Increment the refcount if it's not zero 1858 * 3. If the folio is not found by xas_reload(), put the refcount and retry 1859 * 1860 * On the removal side: 1861 * A. Freeze the page (by zeroing the refcount if nobody else has a reference) 1862 * B. Remove the page from i_pages 1863 * C. Return the page to the page allocator 1864 * 1865 * This means that any page may have its reference count temporarily 1866 * increased by a speculative page cache (or fast GUP) lookup as it can 1867 * be allocated by another user before the RCU grace period expires. 1868 * Because the refcount temporarily acquired here may end up being the 1869 * last refcount on the page, any page allocation must be freeable by 1870 * folio_put(). 1871 */ 1872 1873 /* 1874 * mapping_get_entry - Get a page cache entry. 1875 * @mapping: the address_space to search 1876 * @index: The page cache index. 1877 * 1878 * Looks up the page cache entry at @mapping & @index. If it is a folio, 1879 * it is returned with an increased refcount. If it is a shadow entry 1880 * of a previously evicted folio, or a swap entry from shmem/tmpfs, 1881 * it is returned without further action. 1882 * 1883 * Return: The folio, swap or shadow entry, %NULL if nothing is found. 1884 */ 1885 static void *mapping_get_entry(struct address_space *mapping, pgoff_t index) 1886 { 1887 XA_STATE(xas, &mapping->i_pages, index); 1888 struct folio *folio; 1889 1890 rcu_read_lock(); 1891 repeat: 1892 xas_reset(&xas); 1893 folio = xas_load(&xas); 1894 if (xas_retry(&xas, folio)) 1895 goto repeat; 1896 /* 1897 * A shadow entry of a recently evicted page, or a swap entry from 1898 * shmem/tmpfs. Return it without attempting to raise page count. 1899 */ 1900 if (!folio || xa_is_value(folio)) 1901 goto out; 1902 1903 if (!folio_try_get_rcu(folio)) 1904 goto repeat; 1905 1906 if (unlikely(folio != xas_reload(&xas))) { 1907 folio_put(folio); 1908 goto repeat; 1909 } 1910 out: 1911 rcu_read_unlock(); 1912 1913 return folio; 1914 } 1915 1916 /** 1917 * __filemap_get_folio - Find and get a reference to a folio. 1918 * @mapping: The address_space to search. 1919 * @index: The page index. 1920 * @fgp_flags: %FGP flags modify how the folio is returned. 1921 * @gfp: Memory allocation flags to use if %FGP_CREAT is specified. 1922 * 1923 * Looks up the page cache entry at @mapping & @index. 1924 * 1925 * @fgp_flags can be zero or more of these flags: 1926 * 1927 * * %FGP_ACCESSED - The folio will be marked accessed. 1928 * * %FGP_LOCK - The folio is returned locked. 1929 * * %FGP_ENTRY - If there is a shadow / swap / DAX entry, return it 1930 * instead of allocating a new folio to replace it. 1931 * * %FGP_CREAT - If no page is present then a new page is allocated using 1932 * @gfp and added to the page cache and the VM's LRU list. 1933 * The page is returned locked and with an increased refcount. 1934 * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the 1935 * page is already in cache. If the page was allocated, unlock it before 1936 * returning so the caller can do the same dance. 1937 * * %FGP_WRITE - The page will be written to by the caller. 1938 * * %FGP_NOFS - __GFP_FS will get cleared in gfp. 1939 * * %FGP_NOWAIT - Don't get blocked by page lock. 1940 * * %FGP_STABLE - Wait for the folio to be stable (finished writeback) 1941 * 1942 * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even 1943 * if the %GFP flags specified for %FGP_CREAT are atomic. 1944 * 1945 * If there is a page cache page, it is returned with an increased refcount. 1946 * 1947 * Return: The found folio or %NULL otherwise. 1948 */ 1949 struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index, 1950 int fgp_flags, gfp_t gfp) 1951 { 1952 struct folio *folio; 1953 1954 repeat: 1955 folio = mapping_get_entry(mapping, index); 1956 if (xa_is_value(folio)) { 1957 if (fgp_flags & FGP_ENTRY) 1958 return folio; 1959 folio = NULL; 1960 } 1961 if (!folio) 1962 goto no_page; 1963 1964 if (fgp_flags & FGP_LOCK) { 1965 if (fgp_flags & FGP_NOWAIT) { 1966 if (!folio_trylock(folio)) { 1967 folio_put(folio); 1968 return NULL; 1969 } 1970 } else { 1971 folio_lock(folio); 1972 } 1973 1974 /* Has the page been truncated? */ 1975 if (unlikely(folio->mapping != mapping)) { 1976 folio_unlock(folio); 1977 folio_put(folio); 1978 goto repeat; 1979 } 1980 VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio); 1981 } 1982 1983 if (fgp_flags & FGP_ACCESSED) 1984 folio_mark_accessed(folio); 1985 else if (fgp_flags & FGP_WRITE) { 1986 /* Clear idle flag for buffer write */ 1987 if (folio_test_idle(folio)) 1988 folio_clear_idle(folio); 1989 } 1990 1991 if (fgp_flags & FGP_STABLE) 1992 folio_wait_stable(folio); 1993 no_page: 1994 if (!folio && (fgp_flags & FGP_CREAT)) { 1995 int err; 1996 if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping)) 1997 gfp |= __GFP_WRITE; 1998 if (fgp_flags & FGP_NOFS) 1999 gfp &= ~__GFP_FS; 2000 2001 folio = filemap_alloc_folio(gfp, 0); 2002 if (!folio) 2003 return NULL; 2004 2005 if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP)))) 2006 fgp_flags |= FGP_LOCK; 2007 2008 /* Init accessed so avoid atomic mark_page_accessed later */ 2009 if (fgp_flags & FGP_ACCESSED) 2010 __folio_set_referenced(folio); 2011 2012 err = filemap_add_folio(mapping, folio, index, gfp); 2013 if (unlikely(err)) { 2014 folio_put(folio); 2015 folio = NULL; 2016 if (err == -EEXIST) 2017 goto repeat; 2018 } 2019 2020 /* 2021 * filemap_add_folio locks the page, and for mmap 2022 * we expect an unlocked page. 2023 */ 2024 if (folio && (fgp_flags & FGP_FOR_MMAP)) 2025 folio_unlock(folio); 2026 } 2027 2028 return folio; 2029 } 2030 EXPORT_SYMBOL(__filemap_get_folio); 2031 2032 static inline struct folio *find_get_entry(struct xa_state *xas, pgoff_t max, 2033 xa_mark_t mark) 2034 { 2035 struct folio *folio; 2036 2037 retry: 2038 if (mark == XA_PRESENT) 2039 folio = xas_find(xas, max); 2040 else 2041 folio = xas_find_marked(xas, max, mark); 2042 2043 if (xas_retry(xas, folio)) 2044 goto retry; 2045 /* 2046 * A shadow entry of a recently evicted page, a swap 2047 * entry from shmem/tmpfs or a DAX entry. Return it 2048 * without attempting to raise page count. 2049 */ 2050 if (!folio || xa_is_value(folio)) 2051 return folio; 2052 2053 if (!folio_try_get_rcu(folio)) 2054 goto reset; 2055 2056 if (unlikely(folio != xas_reload(xas))) { 2057 folio_put(folio); 2058 goto reset; 2059 } 2060 2061 return folio; 2062 reset: 2063 xas_reset(xas); 2064 goto retry; 2065 } 2066 2067 /** 2068 * find_get_entries - gang pagecache lookup 2069 * @mapping: The address_space to search 2070 * @start: The starting page cache index 2071 * @end: The final page index (inclusive). 2072 * @fbatch: Where the resulting entries are placed. 2073 * @indices: The cache indices corresponding to the entries in @entries 2074 * 2075 * find_get_entries() will search for and return a batch of entries in 2076 * the mapping. The entries are placed in @fbatch. find_get_entries() 2077 * takes a reference on any actual folios it returns. 2078 * 2079 * The entries have ascending indexes. The indices may not be consecutive 2080 * due to not-present entries or large folios. 2081 * 2082 * Any shadow entries of evicted folios, or swap entries from 2083 * shmem/tmpfs, are included in the returned array. 2084 * 2085 * Return: The number of entries which were found. 2086 */ 2087 unsigned find_get_entries(struct address_space *mapping, pgoff_t start, 2088 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices) 2089 { 2090 XA_STATE(xas, &mapping->i_pages, start); 2091 struct folio *folio; 2092 2093 rcu_read_lock(); 2094 while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) { 2095 indices[fbatch->nr] = xas.xa_index; 2096 if (!folio_batch_add(fbatch, folio)) 2097 break; 2098 } 2099 rcu_read_unlock(); 2100 2101 return folio_batch_count(fbatch); 2102 } 2103 2104 /** 2105 * find_lock_entries - Find a batch of pagecache entries. 2106 * @mapping: The address_space to search. 2107 * @start: The starting page cache index. 2108 * @end: The final page index (inclusive). 2109 * @fbatch: Where the resulting entries are placed. 2110 * @indices: The cache indices of the entries in @fbatch. 2111 * 2112 * find_lock_entries() will return a batch of entries from @mapping. 2113 * Swap, shadow and DAX entries are included. Folios are returned 2114 * locked and with an incremented refcount. Folios which are locked 2115 * by somebody else or under writeback are skipped. Folios which are 2116 * partially outside the range are not returned. 2117 * 2118 * The entries have ascending indexes. The indices may not be consecutive 2119 * due to not-present entries, large folios, folios which could not be 2120 * locked or folios under writeback. 2121 * 2122 * Return: The number of entries which were found. 2123 */ 2124 unsigned find_lock_entries(struct address_space *mapping, pgoff_t start, 2125 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices) 2126 { 2127 XA_STATE(xas, &mapping->i_pages, start); 2128 struct folio *folio; 2129 2130 rcu_read_lock(); 2131 while ((folio = find_get_entry(&xas, end, XA_PRESENT))) { 2132 if (!xa_is_value(folio)) { 2133 if (folio->index < start) 2134 goto put; 2135 if (folio->index + folio_nr_pages(folio) - 1 > end) 2136 goto put; 2137 if (!folio_trylock(folio)) 2138 goto put; 2139 if (folio->mapping != mapping || 2140 folio_test_writeback(folio)) 2141 goto unlock; 2142 VM_BUG_ON_FOLIO(!folio_contains(folio, xas.xa_index), 2143 folio); 2144 } 2145 indices[fbatch->nr] = xas.xa_index; 2146 if (!folio_batch_add(fbatch, folio)) 2147 break; 2148 continue; 2149 unlock: 2150 folio_unlock(folio); 2151 put: 2152 folio_put(folio); 2153 } 2154 rcu_read_unlock(); 2155 2156 return folio_batch_count(fbatch); 2157 } 2158 2159 static inline 2160 bool folio_more_pages(struct folio *folio, pgoff_t index, pgoff_t max) 2161 { 2162 if (!folio_test_large(folio) || folio_test_hugetlb(folio)) 2163 return false; 2164 if (index >= max) 2165 return false; 2166 return index < folio->index + folio_nr_pages(folio) - 1; 2167 } 2168 2169 /** 2170 * find_get_pages_range - gang pagecache lookup 2171 * @mapping: The address_space to search 2172 * @start: The starting page index 2173 * @end: The final page index (inclusive) 2174 * @nr_pages: The maximum number of pages 2175 * @pages: Where the resulting pages are placed 2176 * 2177 * find_get_pages_range() will search for and return a group of up to @nr_pages 2178 * pages in the mapping starting at index @start and up to index @end 2179 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes 2180 * a reference against the returned pages. 2181 * 2182 * The search returns a group of mapping-contiguous pages with ascending 2183 * indexes. There may be holes in the indices due to not-present pages. 2184 * We also update @start to index the next page for the traversal. 2185 * 2186 * Return: the number of pages which were found. If this number is 2187 * smaller than @nr_pages, the end of specified range has been 2188 * reached. 2189 */ 2190 unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start, 2191 pgoff_t end, unsigned int nr_pages, 2192 struct page **pages) 2193 { 2194 XA_STATE(xas, &mapping->i_pages, *start); 2195 struct folio *folio; 2196 unsigned ret = 0; 2197 2198 if (unlikely(!nr_pages)) 2199 return 0; 2200 2201 rcu_read_lock(); 2202 while ((folio = find_get_entry(&xas, end, XA_PRESENT))) { 2203 /* Skip over shadow, swap and DAX entries */ 2204 if (xa_is_value(folio)) 2205 continue; 2206 2207 again: 2208 pages[ret] = folio_file_page(folio, xas.xa_index); 2209 if (++ret == nr_pages) { 2210 *start = xas.xa_index + 1; 2211 goto out; 2212 } 2213 if (folio_more_pages(folio, xas.xa_index, end)) { 2214 xas.xa_index++; 2215 folio_ref_inc(folio); 2216 goto again; 2217 } 2218 } 2219 2220 /* 2221 * We come here when there is no page beyond @end. We take care to not 2222 * overflow the index @start as it confuses some of the callers. This 2223 * breaks the iteration when there is a page at index -1 but that is 2224 * already broken anyway. 2225 */ 2226 if (end == (pgoff_t)-1) 2227 *start = (pgoff_t)-1; 2228 else 2229 *start = end + 1; 2230 out: 2231 rcu_read_unlock(); 2232 2233 return ret; 2234 } 2235 2236 /** 2237 * find_get_pages_contig - gang contiguous pagecache lookup 2238 * @mapping: The address_space to search 2239 * @index: The starting page index 2240 * @nr_pages: The maximum number of pages 2241 * @pages: Where the resulting pages are placed 2242 * 2243 * find_get_pages_contig() works exactly like find_get_pages_range(), 2244 * except that the returned number of pages are guaranteed to be 2245 * contiguous. 2246 * 2247 * Return: the number of pages which were found. 2248 */ 2249 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, 2250 unsigned int nr_pages, struct page **pages) 2251 { 2252 XA_STATE(xas, &mapping->i_pages, index); 2253 struct folio *folio; 2254 unsigned int ret = 0; 2255 2256 if (unlikely(!nr_pages)) 2257 return 0; 2258 2259 rcu_read_lock(); 2260 for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) { 2261 if (xas_retry(&xas, folio)) 2262 continue; 2263 /* 2264 * If the entry has been swapped out, we can stop looking. 2265 * No current caller is looking for DAX entries. 2266 */ 2267 if (xa_is_value(folio)) 2268 break; 2269 2270 if (!folio_try_get_rcu(folio)) 2271 goto retry; 2272 2273 if (unlikely(folio != xas_reload(&xas))) 2274 goto put_page; 2275 2276 again: 2277 pages[ret] = folio_file_page(folio, xas.xa_index); 2278 if (++ret == nr_pages) 2279 break; 2280 if (folio_more_pages(folio, xas.xa_index, ULONG_MAX)) { 2281 xas.xa_index++; 2282 folio_ref_inc(folio); 2283 goto again; 2284 } 2285 continue; 2286 put_page: 2287 folio_put(folio); 2288 retry: 2289 xas_reset(&xas); 2290 } 2291 rcu_read_unlock(); 2292 return ret; 2293 } 2294 EXPORT_SYMBOL(find_get_pages_contig); 2295 2296 /** 2297 * find_get_pages_range_tag - Find and return head pages matching @tag. 2298 * @mapping: the address_space to search 2299 * @index: the starting page index 2300 * @end: The final page index (inclusive) 2301 * @tag: the tag index 2302 * @nr_pages: the maximum number of pages 2303 * @pages: where the resulting pages are placed 2304 * 2305 * Like find_get_pages_range(), except we only return head pages which are 2306 * tagged with @tag. @index is updated to the index immediately after the 2307 * last page we return, ready for the next iteration. 2308 * 2309 * Return: the number of pages which were found. 2310 */ 2311 unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index, 2312 pgoff_t end, xa_mark_t tag, unsigned int nr_pages, 2313 struct page **pages) 2314 { 2315 XA_STATE(xas, &mapping->i_pages, *index); 2316 struct folio *folio; 2317 unsigned ret = 0; 2318 2319 if (unlikely(!nr_pages)) 2320 return 0; 2321 2322 rcu_read_lock(); 2323 while ((folio = find_get_entry(&xas, end, tag))) { 2324 /* 2325 * Shadow entries should never be tagged, but this iteration 2326 * is lockless so there is a window for page reclaim to evict 2327 * a page we saw tagged. Skip over it. 2328 */ 2329 if (xa_is_value(folio)) 2330 continue; 2331 2332 pages[ret] = &folio->page; 2333 if (++ret == nr_pages) { 2334 *index = folio->index + folio_nr_pages(folio); 2335 goto out; 2336 } 2337 } 2338 2339 /* 2340 * We come here when we got to @end. We take care to not overflow the 2341 * index @index as it confuses some of the callers. This breaks the 2342 * iteration when there is a page at index -1 but that is already 2343 * broken anyway. 2344 */ 2345 if (end == (pgoff_t)-1) 2346 *index = (pgoff_t)-1; 2347 else 2348 *index = end + 1; 2349 out: 2350 rcu_read_unlock(); 2351 2352 return ret; 2353 } 2354 EXPORT_SYMBOL(find_get_pages_range_tag); 2355 2356 /* 2357 * CD/DVDs are error prone. When a medium error occurs, the driver may fail 2358 * a _large_ part of the i/o request. Imagine the worst scenario: 2359 * 2360 * ---R__________________________________________B__________ 2361 * ^ reading here ^ bad block(assume 4k) 2362 * 2363 * read(R) => miss => readahead(R...B) => media error => frustrating retries 2364 * => failing the whole request => read(R) => read(R+1) => 2365 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => 2366 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => 2367 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... 2368 * 2369 * It is going insane. Fix it by quickly scaling down the readahead size. 2370 */ 2371 static void shrink_readahead_size_eio(struct file_ra_state *ra) 2372 { 2373 ra->ra_pages /= 4; 2374 } 2375 2376 /* 2377 * filemap_get_read_batch - Get a batch of folios for read 2378 * 2379 * Get a batch of folios which represent a contiguous range of bytes in 2380 * the file. No exceptional entries will be returned. If @index is in 2381 * the middle of a folio, the entire folio will be returned. The last 2382 * folio in the batch may have the readahead flag set or the uptodate flag 2383 * clear so that the caller can take the appropriate action. 2384 */ 2385 static void filemap_get_read_batch(struct address_space *mapping, 2386 pgoff_t index, pgoff_t max, struct folio_batch *fbatch) 2387 { 2388 XA_STATE(xas, &mapping->i_pages, index); 2389 struct folio *folio; 2390 2391 rcu_read_lock(); 2392 for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) { 2393 if (xas_retry(&xas, folio)) 2394 continue; 2395 if (xas.xa_index > max || xa_is_value(folio)) 2396 break; 2397 if (!folio_try_get_rcu(folio)) 2398 goto retry; 2399 2400 if (unlikely(folio != xas_reload(&xas))) 2401 goto put_folio; 2402 2403 if (!folio_batch_add(fbatch, folio)) 2404 break; 2405 if (!folio_test_uptodate(folio)) 2406 break; 2407 if (folio_test_readahead(folio)) 2408 break; 2409 xas_advance(&xas, folio->index + folio_nr_pages(folio) - 1); 2410 continue; 2411 put_folio: 2412 folio_put(folio); 2413 retry: 2414 xas_reset(&xas); 2415 } 2416 rcu_read_unlock(); 2417 } 2418 2419 static int filemap_read_folio(struct file *file, struct address_space *mapping, 2420 struct folio *folio) 2421 { 2422 int error; 2423 2424 /* 2425 * A previous I/O error may have been due to temporary failures, 2426 * eg. multipath errors. PG_error will be set again if readpage 2427 * fails. 2428 */ 2429 folio_clear_error(folio); 2430 /* Start the actual read. The read will unlock the page. */ 2431 error = mapping->a_ops->readpage(file, &folio->page); 2432 if (error) 2433 return error; 2434 2435 error = folio_wait_locked_killable(folio); 2436 if (error) 2437 return error; 2438 if (folio_test_uptodate(folio)) 2439 return 0; 2440 shrink_readahead_size_eio(&file->f_ra); 2441 return -EIO; 2442 } 2443 2444 static bool filemap_range_uptodate(struct address_space *mapping, 2445 loff_t pos, struct iov_iter *iter, struct folio *folio) 2446 { 2447 int count; 2448 2449 if (folio_test_uptodate(folio)) 2450 return true; 2451 /* pipes can't handle partially uptodate pages */ 2452 if (iov_iter_is_pipe(iter)) 2453 return false; 2454 if (!mapping->a_ops->is_partially_uptodate) 2455 return false; 2456 if (mapping->host->i_blkbits >= folio_shift(folio)) 2457 return false; 2458 2459 count = iter->count; 2460 if (folio_pos(folio) > pos) { 2461 count -= folio_pos(folio) - pos; 2462 pos = 0; 2463 } else { 2464 pos -= folio_pos(folio); 2465 } 2466 2467 return mapping->a_ops->is_partially_uptodate(folio, pos, count); 2468 } 2469 2470 static int filemap_update_page(struct kiocb *iocb, 2471 struct address_space *mapping, struct iov_iter *iter, 2472 struct folio *folio) 2473 { 2474 int error; 2475 2476 if (iocb->ki_flags & IOCB_NOWAIT) { 2477 if (!filemap_invalidate_trylock_shared(mapping)) 2478 return -EAGAIN; 2479 } else { 2480 filemap_invalidate_lock_shared(mapping); 2481 } 2482 2483 if (!folio_trylock(folio)) { 2484 error = -EAGAIN; 2485 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO)) 2486 goto unlock_mapping; 2487 if (!(iocb->ki_flags & IOCB_WAITQ)) { 2488 filemap_invalidate_unlock_shared(mapping); 2489 /* 2490 * This is where we usually end up waiting for a 2491 * previously submitted readahead to finish. 2492 */ 2493 folio_put_wait_locked(folio, TASK_KILLABLE); 2494 return AOP_TRUNCATED_PAGE; 2495 } 2496 error = __folio_lock_async(folio, iocb->ki_waitq); 2497 if (error) 2498 goto unlock_mapping; 2499 } 2500 2501 error = AOP_TRUNCATED_PAGE; 2502 if (!folio->mapping) 2503 goto unlock; 2504 2505 error = 0; 2506 if (filemap_range_uptodate(mapping, iocb->ki_pos, iter, folio)) 2507 goto unlock; 2508 2509 error = -EAGAIN; 2510 if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ)) 2511 goto unlock; 2512 2513 error = filemap_read_folio(iocb->ki_filp, mapping, folio); 2514 goto unlock_mapping; 2515 unlock: 2516 folio_unlock(folio); 2517 unlock_mapping: 2518 filemap_invalidate_unlock_shared(mapping); 2519 if (error == AOP_TRUNCATED_PAGE) 2520 folio_put(folio); 2521 return error; 2522 } 2523 2524 static int filemap_create_folio(struct file *file, 2525 struct address_space *mapping, pgoff_t index, 2526 struct folio_batch *fbatch) 2527 { 2528 struct folio *folio; 2529 int error; 2530 2531 folio = filemap_alloc_folio(mapping_gfp_mask(mapping), 0); 2532 if (!folio) 2533 return -ENOMEM; 2534 2535 /* 2536 * Protect against truncate / hole punch. Grabbing invalidate_lock 2537 * here assures we cannot instantiate and bring uptodate new 2538 * pagecache folios after evicting page cache during truncate 2539 * and before actually freeing blocks. Note that we could 2540 * release invalidate_lock after inserting the folio into 2541 * the page cache as the locked folio would then be enough to 2542 * synchronize with hole punching. But there are code paths 2543 * such as filemap_update_page() filling in partially uptodate 2544 * pages or ->readpages() that need to hold invalidate_lock 2545 * while mapping blocks for IO so let's hold the lock here as 2546 * well to keep locking rules simple. 2547 */ 2548 filemap_invalidate_lock_shared(mapping); 2549 error = filemap_add_folio(mapping, folio, index, 2550 mapping_gfp_constraint(mapping, GFP_KERNEL)); 2551 if (error == -EEXIST) 2552 error = AOP_TRUNCATED_PAGE; 2553 if (error) 2554 goto error; 2555 2556 error = filemap_read_folio(file, mapping, folio); 2557 if (error) 2558 goto error; 2559 2560 filemap_invalidate_unlock_shared(mapping); 2561 folio_batch_add(fbatch, folio); 2562 return 0; 2563 error: 2564 filemap_invalidate_unlock_shared(mapping); 2565 folio_put(folio); 2566 return error; 2567 } 2568 2569 static int filemap_readahead(struct kiocb *iocb, struct file *file, 2570 struct address_space *mapping, struct folio *folio, 2571 pgoff_t last_index) 2572 { 2573 DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, folio->index); 2574 2575 if (iocb->ki_flags & IOCB_NOIO) 2576 return -EAGAIN; 2577 page_cache_async_ra(&ractl, folio, last_index - folio->index); 2578 return 0; 2579 } 2580 2581 static int filemap_get_pages(struct kiocb *iocb, struct iov_iter *iter, 2582 struct folio_batch *fbatch) 2583 { 2584 struct file *filp = iocb->ki_filp; 2585 struct address_space *mapping = filp->f_mapping; 2586 struct file_ra_state *ra = &filp->f_ra; 2587 pgoff_t index = iocb->ki_pos >> PAGE_SHIFT; 2588 pgoff_t last_index; 2589 struct folio *folio; 2590 int err = 0; 2591 2592 last_index = DIV_ROUND_UP(iocb->ki_pos + iter->count, PAGE_SIZE); 2593 retry: 2594 if (fatal_signal_pending(current)) 2595 return -EINTR; 2596 2597 filemap_get_read_batch(mapping, index, last_index, fbatch); 2598 if (!folio_batch_count(fbatch)) { 2599 if (iocb->ki_flags & IOCB_NOIO) 2600 return -EAGAIN; 2601 page_cache_sync_readahead(mapping, ra, filp, index, 2602 last_index - index); 2603 filemap_get_read_batch(mapping, index, last_index, fbatch); 2604 } 2605 if (!folio_batch_count(fbatch)) { 2606 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ)) 2607 return -EAGAIN; 2608 err = filemap_create_folio(filp, mapping, 2609 iocb->ki_pos >> PAGE_SHIFT, fbatch); 2610 if (err == AOP_TRUNCATED_PAGE) 2611 goto retry; 2612 return err; 2613 } 2614 2615 folio = fbatch->folios[folio_batch_count(fbatch) - 1]; 2616 if (folio_test_readahead(folio)) { 2617 err = filemap_readahead(iocb, filp, mapping, folio, last_index); 2618 if (err) 2619 goto err; 2620 } 2621 if (!folio_test_uptodate(folio)) { 2622 if ((iocb->ki_flags & IOCB_WAITQ) && 2623 folio_batch_count(fbatch) > 1) 2624 iocb->ki_flags |= IOCB_NOWAIT; 2625 err = filemap_update_page(iocb, mapping, iter, folio); 2626 if (err) 2627 goto err; 2628 } 2629 2630 return 0; 2631 err: 2632 if (err < 0) 2633 folio_put(folio); 2634 if (likely(--fbatch->nr)) 2635 return 0; 2636 if (err == AOP_TRUNCATED_PAGE) 2637 goto retry; 2638 return err; 2639 } 2640 2641 /** 2642 * filemap_read - Read data from the page cache. 2643 * @iocb: The iocb to read. 2644 * @iter: Destination for the data. 2645 * @already_read: Number of bytes already read by the caller. 2646 * 2647 * Copies data from the page cache. If the data is not currently present, 2648 * uses the readahead and readpage address_space operations to fetch it. 2649 * 2650 * Return: Total number of bytes copied, including those already read by 2651 * the caller. If an error happens before any bytes are copied, returns 2652 * a negative error number. 2653 */ 2654 ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter, 2655 ssize_t already_read) 2656 { 2657 struct file *filp = iocb->ki_filp; 2658 struct file_ra_state *ra = &filp->f_ra; 2659 struct address_space *mapping = filp->f_mapping; 2660 struct inode *inode = mapping->host; 2661 struct folio_batch fbatch; 2662 int i, error = 0; 2663 bool writably_mapped; 2664 loff_t isize, end_offset; 2665 2666 if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes)) 2667 return 0; 2668 if (unlikely(!iov_iter_count(iter))) 2669 return 0; 2670 2671 iov_iter_truncate(iter, inode->i_sb->s_maxbytes); 2672 folio_batch_init(&fbatch); 2673 2674 do { 2675 cond_resched(); 2676 2677 /* 2678 * If we've already successfully copied some data, then we 2679 * can no longer safely return -EIOCBQUEUED. Hence mark 2680 * an async read NOWAIT at that point. 2681 */ 2682 if ((iocb->ki_flags & IOCB_WAITQ) && already_read) 2683 iocb->ki_flags |= IOCB_NOWAIT; 2684 2685 if (unlikely(iocb->ki_pos >= i_size_read(inode))) 2686 break; 2687 2688 error = filemap_get_pages(iocb, iter, &fbatch); 2689 if (error < 0) 2690 break; 2691 2692 /* 2693 * i_size must be checked after we know the pages are Uptodate. 2694 * 2695 * Checking i_size after the check allows us to calculate 2696 * the correct value for "nr", which means the zero-filled 2697 * part of the page is not copied back to userspace (unless 2698 * another truncate extends the file - this is desired though). 2699 */ 2700 isize = i_size_read(inode); 2701 if (unlikely(iocb->ki_pos >= isize)) 2702 goto put_folios; 2703 end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count); 2704 2705 /* 2706 * Once we start copying data, we don't want to be touching any 2707 * cachelines that might be contended: 2708 */ 2709 writably_mapped = mapping_writably_mapped(mapping); 2710 2711 /* 2712 * When a sequential read accesses a page several times, only 2713 * mark it as accessed the first time. 2714 */ 2715 if (iocb->ki_pos >> PAGE_SHIFT != 2716 ra->prev_pos >> PAGE_SHIFT) 2717 folio_mark_accessed(fbatch.folios[0]); 2718 2719 for (i = 0; i < folio_batch_count(&fbatch); i++) { 2720 struct folio *folio = fbatch.folios[i]; 2721 size_t fsize = folio_size(folio); 2722 size_t offset = iocb->ki_pos & (fsize - 1); 2723 size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos, 2724 fsize - offset); 2725 size_t copied; 2726 2727 if (end_offset < folio_pos(folio)) 2728 break; 2729 if (i > 0) 2730 folio_mark_accessed(folio); 2731 /* 2732 * If users can be writing to this folio using arbitrary 2733 * virtual addresses, take care of potential aliasing 2734 * before reading the folio on the kernel side. 2735 */ 2736 if (writably_mapped) 2737 flush_dcache_folio(folio); 2738 2739 copied = copy_folio_to_iter(folio, offset, bytes, iter); 2740 2741 already_read += copied; 2742 iocb->ki_pos += copied; 2743 ra->prev_pos = iocb->ki_pos; 2744 2745 if (copied < bytes) { 2746 error = -EFAULT; 2747 break; 2748 } 2749 } 2750 put_folios: 2751 for (i = 0; i < folio_batch_count(&fbatch); i++) 2752 folio_put(fbatch.folios[i]); 2753 folio_batch_init(&fbatch); 2754 } while (iov_iter_count(iter) && iocb->ki_pos < isize && !error); 2755 2756 file_accessed(filp); 2757 2758 return already_read ? already_read : error; 2759 } 2760 EXPORT_SYMBOL_GPL(filemap_read); 2761 2762 /** 2763 * generic_file_read_iter - generic filesystem read routine 2764 * @iocb: kernel I/O control block 2765 * @iter: destination for the data read 2766 * 2767 * This is the "read_iter()" routine for all filesystems 2768 * that can use the page cache directly. 2769 * 2770 * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall 2771 * be returned when no data can be read without waiting for I/O requests 2772 * to complete; it doesn't prevent readahead. 2773 * 2774 * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O 2775 * requests shall be made for the read or for readahead. When no data 2776 * can be read, -EAGAIN shall be returned. When readahead would be 2777 * triggered, a partial, possibly empty read shall be returned. 2778 * 2779 * Return: 2780 * * number of bytes copied, even for partial reads 2781 * * negative error code (or 0 if IOCB_NOIO) if nothing was read 2782 */ 2783 ssize_t 2784 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) 2785 { 2786 size_t count = iov_iter_count(iter); 2787 ssize_t retval = 0; 2788 2789 if (!count) 2790 return 0; /* skip atime */ 2791 2792 if (iocb->ki_flags & IOCB_DIRECT) { 2793 struct file *file = iocb->ki_filp; 2794 struct address_space *mapping = file->f_mapping; 2795 struct inode *inode = mapping->host; 2796 2797 if (iocb->ki_flags & IOCB_NOWAIT) { 2798 if (filemap_range_needs_writeback(mapping, iocb->ki_pos, 2799 iocb->ki_pos + count - 1)) 2800 return -EAGAIN; 2801 } else { 2802 retval = filemap_write_and_wait_range(mapping, 2803 iocb->ki_pos, 2804 iocb->ki_pos + count - 1); 2805 if (retval < 0) 2806 return retval; 2807 } 2808 2809 file_accessed(file); 2810 2811 retval = mapping->a_ops->direct_IO(iocb, iter); 2812 if (retval >= 0) { 2813 iocb->ki_pos += retval; 2814 count -= retval; 2815 } 2816 if (retval != -EIOCBQUEUED) 2817 iov_iter_revert(iter, count - iov_iter_count(iter)); 2818 2819 /* 2820 * Btrfs can have a short DIO read if we encounter 2821 * compressed extents, so if there was an error, or if 2822 * we've already read everything we wanted to, or if 2823 * there was a short read because we hit EOF, go ahead 2824 * and return. Otherwise fallthrough to buffered io for 2825 * the rest of the read. Buffered reads will not work for 2826 * DAX files, so don't bother trying. 2827 */ 2828 if (retval < 0 || !count || IS_DAX(inode)) 2829 return retval; 2830 if (iocb->ki_pos >= i_size_read(inode)) 2831 return retval; 2832 } 2833 2834 return filemap_read(iocb, iter, retval); 2835 } 2836 EXPORT_SYMBOL(generic_file_read_iter); 2837 2838 static inline loff_t folio_seek_hole_data(struct xa_state *xas, 2839 struct address_space *mapping, struct folio *folio, 2840 loff_t start, loff_t end, bool seek_data) 2841 { 2842 const struct address_space_operations *ops = mapping->a_ops; 2843 size_t offset, bsz = i_blocksize(mapping->host); 2844 2845 if (xa_is_value(folio) || folio_test_uptodate(folio)) 2846 return seek_data ? start : end; 2847 if (!ops->is_partially_uptodate) 2848 return seek_data ? end : start; 2849 2850 xas_pause(xas); 2851 rcu_read_unlock(); 2852 folio_lock(folio); 2853 if (unlikely(folio->mapping != mapping)) 2854 goto unlock; 2855 2856 offset = offset_in_folio(folio, start) & ~(bsz - 1); 2857 2858 do { 2859 if (ops->is_partially_uptodate(folio, offset, bsz) == 2860 seek_data) 2861 break; 2862 start = (start + bsz) & ~(bsz - 1); 2863 offset += bsz; 2864 } while (offset < folio_size(folio)); 2865 unlock: 2866 folio_unlock(folio); 2867 rcu_read_lock(); 2868 return start; 2869 } 2870 2871 static inline size_t seek_folio_size(struct xa_state *xas, struct folio *folio) 2872 { 2873 if (xa_is_value(folio)) 2874 return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index); 2875 return folio_size(folio); 2876 } 2877 2878 /** 2879 * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache. 2880 * @mapping: Address space to search. 2881 * @start: First byte to consider. 2882 * @end: Limit of search (exclusive). 2883 * @whence: Either SEEK_HOLE or SEEK_DATA. 2884 * 2885 * If the page cache knows which blocks contain holes and which blocks 2886 * contain data, your filesystem can use this function to implement 2887 * SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are 2888 * entirely memory-based such as tmpfs, and filesystems which support 2889 * unwritten extents. 2890 * 2891 * Return: The requested offset on success, or -ENXIO if @whence specifies 2892 * SEEK_DATA and there is no data after @start. There is an implicit hole 2893 * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start 2894 * and @end contain data. 2895 */ 2896 loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start, 2897 loff_t end, int whence) 2898 { 2899 XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT); 2900 pgoff_t max = (end - 1) >> PAGE_SHIFT; 2901 bool seek_data = (whence == SEEK_DATA); 2902 struct folio *folio; 2903 2904 if (end <= start) 2905 return -ENXIO; 2906 2907 rcu_read_lock(); 2908 while ((folio = find_get_entry(&xas, max, XA_PRESENT))) { 2909 loff_t pos = (u64)xas.xa_index << PAGE_SHIFT; 2910 size_t seek_size; 2911 2912 if (start < pos) { 2913 if (!seek_data) 2914 goto unlock; 2915 start = pos; 2916 } 2917 2918 seek_size = seek_folio_size(&xas, folio); 2919 pos = round_up((u64)pos + 1, seek_size); 2920 start = folio_seek_hole_data(&xas, mapping, folio, start, pos, 2921 seek_data); 2922 if (start < pos) 2923 goto unlock; 2924 if (start >= end) 2925 break; 2926 if (seek_size > PAGE_SIZE) 2927 xas_set(&xas, pos >> PAGE_SHIFT); 2928 if (!xa_is_value(folio)) 2929 folio_put(folio); 2930 } 2931 if (seek_data) 2932 start = -ENXIO; 2933 unlock: 2934 rcu_read_unlock(); 2935 if (folio && !xa_is_value(folio)) 2936 folio_put(folio); 2937 if (start > end) 2938 return end; 2939 return start; 2940 } 2941 2942 #ifdef CONFIG_MMU 2943 #define MMAP_LOTSAMISS (100) 2944 /* 2945 * lock_folio_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock 2946 * @vmf - the vm_fault for this fault. 2947 * @folio - the folio to lock. 2948 * @fpin - the pointer to the file we may pin (or is already pinned). 2949 * 2950 * This works similar to lock_folio_or_retry in that it can drop the 2951 * mmap_lock. It differs in that it actually returns the folio locked 2952 * if it returns 1 and 0 if it couldn't lock the folio. If we did have 2953 * to drop the mmap_lock then fpin will point to the pinned file and 2954 * needs to be fput()'ed at a later point. 2955 */ 2956 static int lock_folio_maybe_drop_mmap(struct vm_fault *vmf, struct folio *folio, 2957 struct file **fpin) 2958 { 2959 if (folio_trylock(folio)) 2960 return 1; 2961 2962 /* 2963 * NOTE! This will make us return with VM_FAULT_RETRY, but with 2964 * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT 2965 * is supposed to work. We have way too many special cases.. 2966 */ 2967 if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT) 2968 return 0; 2969 2970 *fpin = maybe_unlock_mmap_for_io(vmf, *fpin); 2971 if (vmf->flags & FAULT_FLAG_KILLABLE) { 2972 if (__folio_lock_killable(folio)) { 2973 /* 2974 * We didn't have the right flags to drop the mmap_lock, 2975 * but all fault_handlers only check for fatal signals 2976 * if we return VM_FAULT_RETRY, so we need to drop the 2977 * mmap_lock here and return 0 if we don't have a fpin. 2978 */ 2979 if (*fpin == NULL) 2980 mmap_read_unlock(vmf->vma->vm_mm); 2981 return 0; 2982 } 2983 } else 2984 __folio_lock(folio); 2985 2986 return 1; 2987 } 2988 2989 /* 2990 * Synchronous readahead happens when we don't even find a page in the page 2991 * cache at all. We don't want to perform IO under the mmap sem, so if we have 2992 * to drop the mmap sem we return the file that was pinned in order for us to do 2993 * that. If we didn't pin a file then we return NULL. The file that is 2994 * returned needs to be fput()'ed when we're done with it. 2995 */ 2996 static struct file *do_sync_mmap_readahead(struct vm_fault *vmf) 2997 { 2998 struct file *file = vmf->vma->vm_file; 2999 struct file_ra_state *ra = &file->f_ra; 3000 struct address_space *mapping = file->f_mapping; 3001 DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff); 3002 struct file *fpin = NULL; 3003 unsigned int mmap_miss; 3004 3005 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 3006 /* Use the readahead code, even if readahead is disabled */ 3007 if (vmf->vma->vm_flags & VM_HUGEPAGE) { 3008 fpin = maybe_unlock_mmap_for_io(vmf, fpin); 3009 ractl._index &= ~((unsigned long)HPAGE_PMD_NR - 1); 3010 ra->size = HPAGE_PMD_NR; 3011 /* 3012 * Fetch two PMD folios, so we get the chance to actually 3013 * readahead, unless we've been told not to. 3014 */ 3015 if (!(vmf->vma->vm_flags & VM_RAND_READ)) 3016 ra->size *= 2; 3017 ra->async_size = HPAGE_PMD_NR; 3018 page_cache_ra_order(&ractl, ra, HPAGE_PMD_ORDER); 3019 return fpin; 3020 } 3021 #endif 3022 3023 /* If we don't want any read-ahead, don't bother */ 3024 if (vmf->vma->vm_flags & VM_RAND_READ) 3025 return fpin; 3026 if (!ra->ra_pages) 3027 return fpin; 3028 3029 if (vmf->vma->vm_flags & VM_SEQ_READ) { 3030 fpin = maybe_unlock_mmap_for_io(vmf, fpin); 3031 page_cache_sync_ra(&ractl, ra->ra_pages); 3032 return fpin; 3033 } 3034 3035 /* Avoid banging the cache line if not needed */ 3036 mmap_miss = READ_ONCE(ra->mmap_miss); 3037 if (mmap_miss < MMAP_LOTSAMISS * 10) 3038 WRITE_ONCE(ra->mmap_miss, ++mmap_miss); 3039 3040 /* 3041 * Do we miss much more than hit in this file? If so, 3042 * stop bothering with read-ahead. It will only hurt. 3043 */ 3044 if (mmap_miss > MMAP_LOTSAMISS) 3045 return fpin; 3046 3047 /* 3048 * mmap read-around 3049 */ 3050 fpin = maybe_unlock_mmap_for_io(vmf, fpin); 3051 ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2); 3052 ra->size = ra->ra_pages; 3053 ra->async_size = ra->ra_pages / 4; 3054 ractl._index = ra->start; 3055 page_cache_ra_order(&ractl, ra, 0); 3056 return fpin; 3057 } 3058 3059 /* 3060 * Asynchronous readahead happens when we find the page and PG_readahead, 3061 * so we want to possibly extend the readahead further. We return the file that 3062 * was pinned if we have to drop the mmap_lock in order to do IO. 3063 */ 3064 static struct file *do_async_mmap_readahead(struct vm_fault *vmf, 3065 struct folio *folio) 3066 { 3067 struct file *file = vmf->vma->vm_file; 3068 struct file_ra_state *ra = &file->f_ra; 3069 DEFINE_READAHEAD(ractl, file, ra, file->f_mapping, vmf->pgoff); 3070 struct file *fpin = NULL; 3071 unsigned int mmap_miss; 3072 3073 /* If we don't want any read-ahead, don't bother */ 3074 if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages) 3075 return fpin; 3076 3077 mmap_miss = READ_ONCE(ra->mmap_miss); 3078 if (mmap_miss) 3079 WRITE_ONCE(ra->mmap_miss, --mmap_miss); 3080 3081 if (folio_test_readahead(folio)) { 3082 fpin = maybe_unlock_mmap_for_io(vmf, fpin); 3083 page_cache_async_ra(&ractl, folio, ra->ra_pages); 3084 } 3085 return fpin; 3086 } 3087 3088 /** 3089 * filemap_fault - read in file data for page fault handling 3090 * @vmf: struct vm_fault containing details of the fault 3091 * 3092 * filemap_fault() is invoked via the vma operations vector for a 3093 * mapped memory region to read in file data during a page fault. 3094 * 3095 * The goto's are kind of ugly, but this streamlines the normal case of having 3096 * it in the page cache, and handles the special cases reasonably without 3097 * having a lot of duplicated code. 3098 * 3099 * vma->vm_mm->mmap_lock must be held on entry. 3100 * 3101 * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock 3102 * may be dropped before doing I/O or by lock_folio_maybe_drop_mmap(). 3103 * 3104 * If our return value does not have VM_FAULT_RETRY set, the mmap_lock 3105 * has not been released. 3106 * 3107 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set. 3108 * 3109 * Return: bitwise-OR of %VM_FAULT_ codes. 3110 */ 3111 vm_fault_t filemap_fault(struct vm_fault *vmf) 3112 { 3113 int error; 3114 struct file *file = vmf->vma->vm_file; 3115 struct file *fpin = NULL; 3116 struct address_space *mapping = file->f_mapping; 3117 struct inode *inode = mapping->host; 3118 pgoff_t max_idx, index = vmf->pgoff; 3119 struct folio *folio; 3120 vm_fault_t ret = 0; 3121 bool mapping_locked = false; 3122 3123 max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); 3124 if (unlikely(index >= max_idx)) 3125 return VM_FAULT_SIGBUS; 3126 3127 /* 3128 * Do we have something in the page cache already? 3129 */ 3130 folio = filemap_get_folio(mapping, index); 3131 if (likely(folio)) { 3132 /* 3133 * We found the page, so try async readahead before waiting for 3134 * the lock. 3135 */ 3136 if (!(vmf->flags & FAULT_FLAG_TRIED)) 3137 fpin = do_async_mmap_readahead(vmf, folio); 3138 if (unlikely(!folio_test_uptodate(folio))) { 3139 filemap_invalidate_lock_shared(mapping); 3140 mapping_locked = true; 3141 } 3142 } else { 3143 /* No page in the page cache at all */ 3144 count_vm_event(PGMAJFAULT); 3145 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT); 3146 ret = VM_FAULT_MAJOR; 3147 fpin = do_sync_mmap_readahead(vmf); 3148 retry_find: 3149 /* 3150 * See comment in filemap_create_folio() why we need 3151 * invalidate_lock 3152 */ 3153 if (!mapping_locked) { 3154 filemap_invalidate_lock_shared(mapping); 3155 mapping_locked = true; 3156 } 3157 folio = __filemap_get_folio(mapping, index, 3158 FGP_CREAT|FGP_FOR_MMAP, 3159 vmf->gfp_mask); 3160 if (!folio) { 3161 if (fpin) 3162 goto out_retry; 3163 filemap_invalidate_unlock_shared(mapping); 3164 return VM_FAULT_OOM; 3165 } 3166 } 3167 3168 if (!lock_folio_maybe_drop_mmap(vmf, folio, &fpin)) 3169 goto out_retry; 3170 3171 /* Did it get truncated? */ 3172 if (unlikely(folio->mapping != mapping)) { 3173 folio_unlock(folio); 3174 folio_put(folio); 3175 goto retry_find; 3176 } 3177 VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio); 3178 3179 /* 3180 * We have a locked page in the page cache, now we need to check 3181 * that it's up-to-date. If not, it is going to be due to an error. 3182 */ 3183 if (unlikely(!folio_test_uptodate(folio))) { 3184 /* 3185 * The page was in cache and uptodate and now it is not. 3186 * Strange but possible since we didn't hold the page lock all 3187 * the time. Let's drop everything get the invalidate lock and 3188 * try again. 3189 */ 3190 if (!mapping_locked) { 3191 folio_unlock(folio); 3192 folio_put(folio); 3193 goto retry_find; 3194 } 3195 goto page_not_uptodate; 3196 } 3197 3198 /* 3199 * We've made it this far and we had to drop our mmap_lock, now is the 3200 * time to return to the upper layer and have it re-find the vma and 3201 * redo the fault. 3202 */ 3203 if (fpin) { 3204 folio_unlock(folio); 3205 goto out_retry; 3206 } 3207 if (mapping_locked) 3208 filemap_invalidate_unlock_shared(mapping); 3209 3210 /* 3211 * Found the page and have a reference on it. 3212 * We must recheck i_size under page lock. 3213 */ 3214 max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); 3215 if (unlikely(index >= max_idx)) { 3216 folio_unlock(folio); 3217 folio_put(folio); 3218 return VM_FAULT_SIGBUS; 3219 } 3220 3221 vmf->page = folio_file_page(folio, index); 3222 return ret | VM_FAULT_LOCKED; 3223 3224 page_not_uptodate: 3225 /* 3226 * Umm, take care of errors if the page isn't up-to-date. 3227 * Try to re-read it _once_. We do this synchronously, 3228 * because there really aren't any performance issues here 3229 * and we need to check for errors. 3230 */ 3231 fpin = maybe_unlock_mmap_for_io(vmf, fpin); 3232 error = filemap_read_folio(file, mapping, folio); 3233 if (fpin) 3234 goto out_retry; 3235 folio_put(folio); 3236 3237 if (!error || error == AOP_TRUNCATED_PAGE) 3238 goto retry_find; 3239 filemap_invalidate_unlock_shared(mapping); 3240 3241 return VM_FAULT_SIGBUS; 3242 3243 out_retry: 3244 /* 3245 * We dropped the mmap_lock, we need to return to the fault handler to 3246 * re-find the vma and come back and find our hopefully still populated 3247 * page. 3248 */ 3249 if (folio) 3250 folio_put(folio); 3251 if (mapping_locked) 3252 filemap_invalidate_unlock_shared(mapping); 3253 if (fpin) 3254 fput(fpin); 3255 return ret | VM_FAULT_RETRY; 3256 } 3257 EXPORT_SYMBOL(filemap_fault); 3258 3259 static bool filemap_map_pmd(struct vm_fault *vmf, struct page *page) 3260 { 3261 struct mm_struct *mm = vmf->vma->vm_mm; 3262 3263 /* Huge page is mapped? No need to proceed. */ 3264 if (pmd_trans_huge(*vmf->pmd)) { 3265 unlock_page(page); 3266 put_page(page); 3267 return true; 3268 } 3269 3270 if (pmd_none(*vmf->pmd) && PageTransHuge(page)) { 3271 vm_fault_t ret = do_set_pmd(vmf, page); 3272 if (!ret) { 3273 /* The page is mapped successfully, reference consumed. */ 3274 unlock_page(page); 3275 return true; 3276 } 3277 } 3278 3279 if (pmd_none(*vmf->pmd)) 3280 pmd_install(mm, vmf->pmd, &vmf->prealloc_pte); 3281 3282 /* See comment in handle_pte_fault() */ 3283 if (pmd_devmap_trans_unstable(vmf->pmd)) { 3284 unlock_page(page); 3285 put_page(page); 3286 return true; 3287 } 3288 3289 return false; 3290 } 3291 3292 static struct folio *next_uptodate_page(struct folio *folio, 3293 struct address_space *mapping, 3294 struct xa_state *xas, pgoff_t end_pgoff) 3295 { 3296 unsigned long max_idx; 3297 3298 do { 3299 if (!folio) 3300 return NULL; 3301 if (xas_retry(xas, folio)) 3302 continue; 3303 if (xa_is_value(folio)) 3304 continue; 3305 if (folio_test_locked(folio)) 3306 continue; 3307 if (!folio_try_get_rcu(folio)) 3308 continue; 3309 /* Has the page moved or been split? */ 3310 if (unlikely(folio != xas_reload(xas))) 3311 goto skip; 3312 if (!folio_test_uptodate(folio) || folio_test_readahead(folio)) 3313 goto skip; 3314 if (!folio_trylock(folio)) 3315 goto skip; 3316 if (folio->mapping != mapping) 3317 goto unlock; 3318 if (!folio_test_uptodate(folio)) 3319 goto unlock; 3320 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE); 3321 if (xas->xa_index >= max_idx) 3322 goto unlock; 3323 return folio; 3324 unlock: 3325 folio_unlock(folio); 3326 skip: 3327 folio_put(folio); 3328 } while ((folio = xas_next_entry(xas, end_pgoff)) != NULL); 3329 3330 return NULL; 3331 } 3332 3333 static inline struct folio *first_map_page(struct address_space *mapping, 3334 struct xa_state *xas, 3335 pgoff_t end_pgoff) 3336 { 3337 return next_uptodate_page(xas_find(xas, end_pgoff), 3338 mapping, xas, end_pgoff); 3339 } 3340 3341 static inline struct folio *next_map_page(struct address_space *mapping, 3342 struct xa_state *xas, 3343 pgoff_t end_pgoff) 3344 { 3345 return next_uptodate_page(xas_next_entry(xas, end_pgoff), 3346 mapping, xas, end_pgoff); 3347 } 3348 3349 vm_fault_t filemap_map_pages(struct vm_fault *vmf, 3350 pgoff_t start_pgoff, pgoff_t end_pgoff) 3351 { 3352 struct vm_area_struct *vma = vmf->vma; 3353 struct file *file = vma->vm_file; 3354 struct address_space *mapping = file->f_mapping; 3355 pgoff_t last_pgoff = start_pgoff; 3356 unsigned long addr; 3357 XA_STATE(xas, &mapping->i_pages, start_pgoff); 3358 struct folio *folio; 3359 struct page *page; 3360 unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss); 3361 vm_fault_t ret = 0; 3362 3363 rcu_read_lock(); 3364 folio = first_map_page(mapping, &xas, end_pgoff); 3365 if (!folio) 3366 goto out; 3367 3368 if (filemap_map_pmd(vmf, &folio->page)) { 3369 ret = VM_FAULT_NOPAGE; 3370 goto out; 3371 } 3372 3373 addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT); 3374 vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl); 3375 do { 3376 again: 3377 page = folio_file_page(folio, xas.xa_index); 3378 if (PageHWPoison(page)) 3379 goto unlock; 3380 3381 if (mmap_miss > 0) 3382 mmap_miss--; 3383 3384 addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT; 3385 vmf->pte += xas.xa_index - last_pgoff; 3386 last_pgoff = xas.xa_index; 3387 3388 if (!pte_none(*vmf->pte)) 3389 goto unlock; 3390 3391 /* We're about to handle the fault */ 3392 if (vmf->address == addr) 3393 ret = VM_FAULT_NOPAGE; 3394 3395 do_set_pte(vmf, page, addr); 3396 /* no need to invalidate: a not-present page won't be cached */ 3397 update_mmu_cache(vma, addr, vmf->pte); 3398 if (folio_more_pages(folio, xas.xa_index, end_pgoff)) { 3399 xas.xa_index++; 3400 folio_ref_inc(folio); 3401 goto again; 3402 } 3403 folio_unlock(folio); 3404 continue; 3405 unlock: 3406 if (folio_more_pages(folio, xas.xa_index, end_pgoff)) { 3407 xas.xa_index++; 3408 goto again; 3409 } 3410 folio_unlock(folio); 3411 folio_put(folio); 3412 } while ((folio = next_map_page(mapping, &xas, end_pgoff)) != NULL); 3413 pte_unmap_unlock(vmf->pte, vmf->ptl); 3414 out: 3415 rcu_read_unlock(); 3416 WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss); 3417 return ret; 3418 } 3419 EXPORT_SYMBOL(filemap_map_pages); 3420 3421 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) 3422 { 3423 struct address_space *mapping = vmf->vma->vm_file->f_mapping; 3424 struct folio *folio = page_folio(vmf->page); 3425 vm_fault_t ret = VM_FAULT_LOCKED; 3426 3427 sb_start_pagefault(mapping->host->i_sb); 3428 file_update_time(vmf->vma->vm_file); 3429 folio_lock(folio); 3430 if (folio->mapping != mapping) { 3431 folio_unlock(folio); 3432 ret = VM_FAULT_NOPAGE; 3433 goto out; 3434 } 3435 /* 3436 * We mark the folio dirty already here so that when freeze is in 3437 * progress, we are guaranteed that writeback during freezing will 3438 * see the dirty folio and writeprotect it again. 3439 */ 3440 folio_mark_dirty(folio); 3441 folio_wait_stable(folio); 3442 out: 3443 sb_end_pagefault(mapping->host->i_sb); 3444 return ret; 3445 } 3446 3447 const struct vm_operations_struct generic_file_vm_ops = { 3448 .fault = filemap_fault, 3449 .map_pages = filemap_map_pages, 3450 .page_mkwrite = filemap_page_mkwrite, 3451 }; 3452 3453 /* This is used for a general mmap of a disk file */ 3454 3455 int generic_file_mmap(struct file *file, struct vm_area_struct *vma) 3456 { 3457 struct address_space *mapping = file->f_mapping; 3458 3459 if (!mapping->a_ops->readpage) 3460 return -ENOEXEC; 3461 file_accessed(file); 3462 vma->vm_ops = &generic_file_vm_ops; 3463 return 0; 3464 } 3465 3466 /* 3467 * This is for filesystems which do not implement ->writepage. 3468 */ 3469 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) 3470 { 3471 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) 3472 return -EINVAL; 3473 return generic_file_mmap(file, vma); 3474 } 3475 #else 3476 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) 3477 { 3478 return VM_FAULT_SIGBUS; 3479 } 3480 int generic_file_mmap(struct file *file, struct vm_area_struct *vma) 3481 { 3482 return -ENOSYS; 3483 } 3484 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) 3485 { 3486 return -ENOSYS; 3487 } 3488 #endif /* CONFIG_MMU */ 3489 3490 EXPORT_SYMBOL(filemap_page_mkwrite); 3491 EXPORT_SYMBOL(generic_file_mmap); 3492 EXPORT_SYMBOL(generic_file_readonly_mmap); 3493 3494 static struct folio *do_read_cache_folio(struct address_space *mapping, 3495 pgoff_t index, filler_t filler, void *data, gfp_t gfp) 3496 { 3497 struct folio *folio; 3498 int err; 3499 repeat: 3500 folio = filemap_get_folio(mapping, index); 3501 if (!folio) { 3502 folio = filemap_alloc_folio(gfp, 0); 3503 if (!folio) 3504 return ERR_PTR(-ENOMEM); 3505 err = filemap_add_folio(mapping, folio, index, gfp); 3506 if (unlikely(err)) { 3507 folio_put(folio); 3508 if (err == -EEXIST) 3509 goto repeat; 3510 /* Presumably ENOMEM for xarray node */ 3511 return ERR_PTR(err); 3512 } 3513 3514 filler: 3515 if (filler) 3516 err = filler(data, &folio->page); 3517 else 3518 err = mapping->a_ops->readpage(data, &folio->page); 3519 3520 if (err < 0) { 3521 folio_put(folio); 3522 return ERR_PTR(err); 3523 } 3524 3525 folio_wait_locked(folio); 3526 if (!folio_test_uptodate(folio)) { 3527 folio_put(folio); 3528 return ERR_PTR(-EIO); 3529 } 3530 3531 goto out; 3532 } 3533 if (folio_test_uptodate(folio)) 3534 goto out; 3535 3536 if (!folio_trylock(folio)) { 3537 folio_put_wait_locked(folio, TASK_UNINTERRUPTIBLE); 3538 goto repeat; 3539 } 3540 3541 /* Folio was truncated from mapping */ 3542 if (!folio->mapping) { 3543 folio_unlock(folio); 3544 folio_put(folio); 3545 goto repeat; 3546 } 3547 3548 /* Someone else locked and filled the page in a very small window */ 3549 if (folio_test_uptodate(folio)) { 3550 folio_unlock(folio); 3551 goto out; 3552 } 3553 3554 /* 3555 * A previous I/O error may have been due to temporary 3556 * failures. 3557 * Clear page error before actual read, PG_error will be 3558 * set again if read page fails. 3559 */ 3560 folio_clear_error(folio); 3561 goto filler; 3562 3563 out: 3564 folio_mark_accessed(folio); 3565 return folio; 3566 } 3567 3568 /** 3569 * read_cache_folio - read into page cache, fill it if needed 3570 * @mapping: the page's address_space 3571 * @index: the page index 3572 * @filler: function to perform the read 3573 * @data: first arg to filler(data, page) function, often left as NULL 3574 * 3575 * Read into the page cache. If a page already exists, and PageUptodate() is 3576 * not set, try to fill the page and wait for it to become unlocked. 3577 * 3578 * If the page does not get brought uptodate, return -EIO. 3579 * 3580 * The function expects mapping->invalidate_lock to be already held. 3581 * 3582 * Return: up to date page on success, ERR_PTR() on failure. 3583 */ 3584 struct folio *read_cache_folio(struct address_space *mapping, pgoff_t index, 3585 filler_t filler, void *data) 3586 { 3587 return do_read_cache_folio(mapping, index, filler, data, 3588 mapping_gfp_mask(mapping)); 3589 } 3590 EXPORT_SYMBOL(read_cache_folio); 3591 3592 static struct page *do_read_cache_page(struct address_space *mapping, 3593 pgoff_t index, filler_t *filler, void *data, gfp_t gfp) 3594 { 3595 struct folio *folio; 3596 3597 folio = do_read_cache_folio(mapping, index, filler, data, gfp); 3598 if (IS_ERR(folio)) 3599 return &folio->page; 3600 return folio_file_page(folio, index); 3601 } 3602 3603 struct page *read_cache_page(struct address_space *mapping, 3604 pgoff_t index, filler_t *filler, void *data) 3605 { 3606 return do_read_cache_page(mapping, index, filler, data, 3607 mapping_gfp_mask(mapping)); 3608 } 3609 EXPORT_SYMBOL(read_cache_page); 3610 3611 /** 3612 * read_cache_page_gfp - read into page cache, using specified page allocation flags. 3613 * @mapping: the page's address_space 3614 * @index: the page index 3615 * @gfp: the page allocator flags to use if allocating 3616 * 3617 * This is the same as "read_mapping_page(mapping, index, NULL)", but with 3618 * any new page allocations done using the specified allocation flags. 3619 * 3620 * If the page does not get brought uptodate, return -EIO. 3621 * 3622 * The function expects mapping->invalidate_lock to be already held. 3623 * 3624 * Return: up to date page on success, ERR_PTR() on failure. 3625 */ 3626 struct page *read_cache_page_gfp(struct address_space *mapping, 3627 pgoff_t index, 3628 gfp_t gfp) 3629 { 3630 return do_read_cache_page(mapping, index, NULL, NULL, gfp); 3631 } 3632 EXPORT_SYMBOL(read_cache_page_gfp); 3633 3634 int pagecache_write_begin(struct file *file, struct address_space *mapping, 3635 loff_t pos, unsigned len, unsigned flags, 3636 struct page **pagep, void **fsdata) 3637 { 3638 const struct address_space_operations *aops = mapping->a_ops; 3639 3640 return aops->write_begin(file, mapping, pos, len, flags, 3641 pagep, fsdata); 3642 } 3643 EXPORT_SYMBOL(pagecache_write_begin); 3644 3645 int pagecache_write_end(struct file *file, struct address_space *mapping, 3646 loff_t pos, unsigned len, unsigned copied, 3647 struct page *page, void *fsdata) 3648 { 3649 const struct address_space_operations *aops = mapping->a_ops; 3650 3651 return aops->write_end(file, mapping, pos, len, copied, page, fsdata); 3652 } 3653 EXPORT_SYMBOL(pagecache_write_end); 3654 3655 /* 3656 * Warn about a page cache invalidation failure during a direct I/O write. 3657 */ 3658 void dio_warn_stale_pagecache(struct file *filp) 3659 { 3660 static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST); 3661 char pathname[128]; 3662 char *path; 3663 3664 errseq_set(&filp->f_mapping->wb_err, -EIO); 3665 if (__ratelimit(&_rs)) { 3666 path = file_path(filp, pathname, sizeof(pathname)); 3667 if (IS_ERR(path)) 3668 path = "(unknown)"; 3669 pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n"); 3670 pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid, 3671 current->comm); 3672 } 3673 } 3674 3675 ssize_t 3676 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from) 3677 { 3678 struct file *file = iocb->ki_filp; 3679 struct address_space *mapping = file->f_mapping; 3680 struct inode *inode = mapping->host; 3681 loff_t pos = iocb->ki_pos; 3682 ssize_t written; 3683 size_t write_len; 3684 pgoff_t end; 3685 3686 write_len = iov_iter_count(from); 3687 end = (pos + write_len - 1) >> PAGE_SHIFT; 3688 3689 if (iocb->ki_flags & IOCB_NOWAIT) { 3690 /* If there are pages to writeback, return */ 3691 if (filemap_range_has_page(file->f_mapping, pos, 3692 pos + write_len - 1)) 3693 return -EAGAIN; 3694 } else { 3695 written = filemap_write_and_wait_range(mapping, pos, 3696 pos + write_len - 1); 3697 if (written) 3698 goto out; 3699 } 3700 3701 /* 3702 * After a write we want buffered reads to be sure to go to disk to get 3703 * the new data. We invalidate clean cached page from the region we're 3704 * about to write. We do this *before* the write so that we can return 3705 * without clobbering -EIOCBQUEUED from ->direct_IO(). 3706 */ 3707 written = invalidate_inode_pages2_range(mapping, 3708 pos >> PAGE_SHIFT, end); 3709 /* 3710 * If a page can not be invalidated, return 0 to fall back 3711 * to buffered write. 3712 */ 3713 if (written) { 3714 if (written == -EBUSY) 3715 return 0; 3716 goto out; 3717 } 3718 3719 written = mapping->a_ops->direct_IO(iocb, from); 3720 3721 /* 3722 * Finally, try again to invalidate clean pages which might have been 3723 * cached by non-direct readahead, or faulted in by get_user_pages() 3724 * if the source of the write was an mmap'ed region of the file 3725 * we're writing. Either one is a pretty crazy thing to do, 3726 * so we don't support it 100%. If this invalidation 3727 * fails, tough, the write still worked... 3728 * 3729 * Most of the time we do not need this since dio_complete() will do 3730 * the invalidation for us. However there are some file systems that 3731 * do not end up with dio_complete() being called, so let's not break 3732 * them by removing it completely. 3733 * 3734 * Noticeable example is a blkdev_direct_IO(). 3735 * 3736 * Skip invalidation for async writes or if mapping has no pages. 3737 */ 3738 if (written > 0 && mapping->nrpages && 3739 invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end)) 3740 dio_warn_stale_pagecache(file); 3741 3742 if (written > 0) { 3743 pos += written; 3744 write_len -= written; 3745 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { 3746 i_size_write(inode, pos); 3747 mark_inode_dirty(inode); 3748 } 3749 iocb->ki_pos = pos; 3750 } 3751 if (written != -EIOCBQUEUED) 3752 iov_iter_revert(from, write_len - iov_iter_count(from)); 3753 out: 3754 return written; 3755 } 3756 EXPORT_SYMBOL(generic_file_direct_write); 3757 3758 ssize_t generic_perform_write(struct file *file, 3759 struct iov_iter *i, loff_t pos) 3760 { 3761 struct address_space *mapping = file->f_mapping; 3762 const struct address_space_operations *a_ops = mapping->a_ops; 3763 long status = 0; 3764 ssize_t written = 0; 3765 unsigned int flags = 0; 3766 3767 do { 3768 struct page *page; 3769 unsigned long offset; /* Offset into pagecache page */ 3770 unsigned long bytes; /* Bytes to write to page */ 3771 size_t copied; /* Bytes copied from user */ 3772 void *fsdata; 3773 3774 offset = (pos & (PAGE_SIZE - 1)); 3775 bytes = min_t(unsigned long, PAGE_SIZE - offset, 3776 iov_iter_count(i)); 3777 3778 again: 3779 /* 3780 * Bring in the user page that we will copy from _first_. 3781 * Otherwise there's a nasty deadlock on copying from the 3782 * same page as we're writing to, without it being marked 3783 * up-to-date. 3784 */ 3785 if (unlikely(fault_in_iov_iter_readable(i, bytes))) { 3786 status = -EFAULT; 3787 break; 3788 } 3789 3790 if (fatal_signal_pending(current)) { 3791 status = -EINTR; 3792 break; 3793 } 3794 3795 status = a_ops->write_begin(file, mapping, pos, bytes, flags, 3796 &page, &fsdata); 3797 if (unlikely(status < 0)) 3798 break; 3799 3800 if (mapping_writably_mapped(mapping)) 3801 flush_dcache_page(page); 3802 3803 copied = copy_page_from_iter_atomic(page, offset, bytes, i); 3804 flush_dcache_page(page); 3805 3806 status = a_ops->write_end(file, mapping, pos, bytes, copied, 3807 page, fsdata); 3808 if (unlikely(status != copied)) { 3809 iov_iter_revert(i, copied - max(status, 0L)); 3810 if (unlikely(status < 0)) 3811 break; 3812 } 3813 cond_resched(); 3814 3815 if (unlikely(status == 0)) { 3816 /* 3817 * A short copy made ->write_end() reject the 3818 * thing entirely. Might be memory poisoning 3819 * halfway through, might be a race with munmap, 3820 * might be severe memory pressure. 3821 */ 3822 if (copied) 3823 bytes = copied; 3824 goto again; 3825 } 3826 pos += status; 3827 written += status; 3828 3829 balance_dirty_pages_ratelimited(mapping); 3830 } while (iov_iter_count(i)); 3831 3832 return written ? written : status; 3833 } 3834 EXPORT_SYMBOL(generic_perform_write); 3835 3836 /** 3837 * __generic_file_write_iter - write data to a file 3838 * @iocb: IO state structure (file, offset, etc.) 3839 * @from: iov_iter with data to write 3840 * 3841 * This function does all the work needed for actually writing data to a 3842 * file. It does all basic checks, removes SUID from the file, updates 3843 * modification times and calls proper subroutines depending on whether we 3844 * do direct IO or a standard buffered write. 3845 * 3846 * It expects i_rwsem to be grabbed unless we work on a block device or similar 3847 * object which does not need locking at all. 3848 * 3849 * This function does *not* take care of syncing data in case of O_SYNC write. 3850 * A caller has to handle it. This is mainly due to the fact that we want to 3851 * avoid syncing under i_rwsem. 3852 * 3853 * Return: 3854 * * number of bytes written, even for truncated writes 3855 * * negative error code if no data has been written at all 3856 */ 3857 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) 3858 { 3859 struct file *file = iocb->ki_filp; 3860 struct address_space *mapping = file->f_mapping; 3861 struct inode *inode = mapping->host; 3862 ssize_t written = 0; 3863 ssize_t err; 3864 ssize_t status; 3865 3866 /* We can write back this queue in page reclaim */ 3867 current->backing_dev_info = inode_to_bdi(inode); 3868 err = file_remove_privs(file); 3869 if (err) 3870 goto out; 3871 3872 err = file_update_time(file); 3873 if (err) 3874 goto out; 3875 3876 if (iocb->ki_flags & IOCB_DIRECT) { 3877 loff_t pos, endbyte; 3878 3879 written = generic_file_direct_write(iocb, from); 3880 /* 3881 * If the write stopped short of completing, fall back to 3882 * buffered writes. Some filesystems do this for writes to 3883 * holes, for example. For DAX files, a buffered write will 3884 * not succeed (even if it did, DAX does not handle dirty 3885 * page-cache pages correctly). 3886 */ 3887 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode)) 3888 goto out; 3889 3890 status = generic_perform_write(file, from, pos = iocb->ki_pos); 3891 /* 3892 * If generic_perform_write() returned a synchronous error 3893 * then we want to return the number of bytes which were 3894 * direct-written, or the error code if that was zero. Note 3895 * that this differs from normal direct-io semantics, which 3896 * will return -EFOO even if some bytes were written. 3897 */ 3898 if (unlikely(status < 0)) { 3899 err = status; 3900 goto out; 3901 } 3902 /* 3903 * We need to ensure that the page cache pages are written to 3904 * disk and invalidated to preserve the expected O_DIRECT 3905 * semantics. 3906 */ 3907 endbyte = pos + status - 1; 3908 err = filemap_write_and_wait_range(mapping, pos, endbyte); 3909 if (err == 0) { 3910 iocb->ki_pos = endbyte + 1; 3911 written += status; 3912 invalidate_mapping_pages(mapping, 3913 pos >> PAGE_SHIFT, 3914 endbyte >> PAGE_SHIFT); 3915 } else { 3916 /* 3917 * We don't know how much we wrote, so just return 3918 * the number of bytes which were direct-written 3919 */ 3920 } 3921 } else { 3922 written = generic_perform_write(file, from, iocb->ki_pos); 3923 if (likely(written > 0)) 3924 iocb->ki_pos += written; 3925 } 3926 out: 3927 current->backing_dev_info = NULL; 3928 return written ? written : err; 3929 } 3930 EXPORT_SYMBOL(__generic_file_write_iter); 3931 3932 /** 3933 * generic_file_write_iter - write data to a file 3934 * @iocb: IO state structure 3935 * @from: iov_iter with data to write 3936 * 3937 * This is a wrapper around __generic_file_write_iter() to be used by most 3938 * filesystems. It takes care of syncing the file in case of O_SYNC file 3939 * and acquires i_rwsem as needed. 3940 * Return: 3941 * * negative error code if no data has been written at all of 3942 * vfs_fsync_range() failed for a synchronous write 3943 * * number of bytes written, even for truncated writes 3944 */ 3945 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) 3946 { 3947 struct file *file = iocb->ki_filp; 3948 struct inode *inode = file->f_mapping->host; 3949 ssize_t ret; 3950 3951 inode_lock(inode); 3952 ret = generic_write_checks(iocb, from); 3953 if (ret > 0) 3954 ret = __generic_file_write_iter(iocb, from); 3955 inode_unlock(inode); 3956 3957 if (ret > 0) 3958 ret = generic_write_sync(iocb, ret); 3959 return ret; 3960 } 3961 EXPORT_SYMBOL(generic_file_write_iter); 3962 3963 /** 3964 * filemap_release_folio() - Release fs-specific metadata on a folio. 3965 * @folio: The folio which the kernel is trying to free. 3966 * @gfp: Memory allocation flags (and I/O mode). 3967 * 3968 * The address_space is trying to release any data attached to a folio 3969 * (presumably at folio->private). 3970 * 3971 * This will also be called if the private_2 flag is set on a page, 3972 * indicating that the folio has other metadata associated with it. 3973 * 3974 * The @gfp argument specifies whether I/O may be performed to release 3975 * this page (__GFP_IO), and whether the call may block 3976 * (__GFP_RECLAIM & __GFP_FS). 3977 * 3978 * Return: %true if the release was successful, otherwise %false. 3979 */ 3980 bool filemap_release_folio(struct folio *folio, gfp_t gfp) 3981 { 3982 struct address_space * const mapping = folio->mapping; 3983 3984 BUG_ON(!folio_test_locked(folio)); 3985 if (folio_test_writeback(folio)) 3986 return false; 3987 3988 if (mapping && mapping->a_ops->releasepage) 3989 return mapping->a_ops->releasepage(&folio->page, gfp); 3990 return try_to_free_buffers(&folio->page); 3991 } 3992 EXPORT_SYMBOL(filemap_release_folio); 3993