1 /* 2 * linux/mm/filemap.c 3 * 4 * Copyright (C) 1994-1999 Linus Torvalds 5 */ 6 7 /* 8 * This file handles the generic file mmap semantics used by 9 * most "normal" filesystems (but you don't /have/ to use this: 10 * the NFS filesystem used to do this differently, for example) 11 */ 12 #include <linux/export.h> 13 #include <linux/compiler.h> 14 #include <linux/dax.h> 15 #include <linux/fs.h> 16 #include <linux/sched/signal.h> 17 #include <linux/uaccess.h> 18 #include <linux/capability.h> 19 #include <linux/kernel_stat.h> 20 #include <linux/gfp.h> 21 #include <linux/mm.h> 22 #include <linux/swap.h> 23 #include <linux/mman.h> 24 #include <linux/pagemap.h> 25 #include <linux/file.h> 26 #include <linux/uio.h> 27 #include <linux/hash.h> 28 #include <linux/writeback.h> 29 #include <linux/backing-dev.h> 30 #include <linux/pagevec.h> 31 #include <linux/blkdev.h> 32 #include <linux/security.h> 33 #include <linux/cpuset.h> 34 #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */ 35 #include <linux/hugetlb.h> 36 #include <linux/memcontrol.h> 37 #include <linux/cleancache.h> 38 #include <linux/shmem_fs.h> 39 #include <linux/rmap.h> 40 #include "internal.h" 41 42 #define CREATE_TRACE_POINTS 43 #include <trace/events/filemap.h> 44 45 /* 46 * FIXME: remove all knowledge of the buffer layer from the core VM 47 */ 48 #include <linux/buffer_head.h> /* for try_to_free_buffers */ 49 50 #include <asm/mman.h> 51 52 /* 53 * Shared mappings implemented 30.11.1994. It's not fully working yet, 54 * though. 55 * 56 * Shared mappings now work. 15.8.1995 Bruno. 57 * 58 * finished 'unifying' the page and buffer cache and SMP-threaded the 59 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> 60 * 61 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> 62 */ 63 64 /* 65 * Lock ordering: 66 * 67 * ->i_mmap_rwsem (truncate_pagecache) 68 * ->private_lock (__free_pte->__set_page_dirty_buffers) 69 * ->swap_lock (exclusive_swap_page, others) 70 * ->mapping->tree_lock 71 * 72 * ->i_mutex 73 * ->i_mmap_rwsem (truncate->unmap_mapping_range) 74 * 75 * ->mmap_sem 76 * ->i_mmap_rwsem 77 * ->page_table_lock or pte_lock (various, mainly in memory.c) 78 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock) 79 * 80 * ->mmap_sem 81 * ->lock_page (access_process_vm) 82 * 83 * ->i_mutex (generic_perform_write) 84 * ->mmap_sem (fault_in_pages_readable->do_page_fault) 85 * 86 * bdi->wb.list_lock 87 * sb_lock (fs/fs-writeback.c) 88 * ->mapping->tree_lock (__sync_single_inode) 89 * 90 * ->i_mmap_rwsem 91 * ->anon_vma.lock (vma_adjust) 92 * 93 * ->anon_vma.lock 94 * ->page_table_lock or pte_lock (anon_vma_prepare and various) 95 * 96 * ->page_table_lock or pte_lock 97 * ->swap_lock (try_to_unmap_one) 98 * ->private_lock (try_to_unmap_one) 99 * ->tree_lock (try_to_unmap_one) 100 * ->zone_lru_lock(zone) (follow_page->mark_page_accessed) 101 * ->zone_lru_lock(zone) (check_pte_range->isolate_lru_page) 102 * ->private_lock (page_remove_rmap->set_page_dirty) 103 * ->tree_lock (page_remove_rmap->set_page_dirty) 104 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) 105 * ->inode->i_lock (page_remove_rmap->set_page_dirty) 106 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg) 107 * bdi.wb->list_lock (zap_pte_range->set_page_dirty) 108 * ->inode->i_lock (zap_pte_range->set_page_dirty) 109 * ->private_lock (zap_pte_range->__set_page_dirty_buffers) 110 * 111 * ->i_mmap_rwsem 112 * ->tasklist_lock (memory_failure, collect_procs_ao) 113 */ 114 115 static int page_cache_tree_insert(struct address_space *mapping, 116 struct page *page, void **shadowp) 117 { 118 struct radix_tree_node *node; 119 void **slot; 120 int error; 121 122 error = __radix_tree_create(&mapping->page_tree, page->index, 0, 123 &node, &slot); 124 if (error) 125 return error; 126 if (*slot) { 127 void *p; 128 129 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock); 130 if (!radix_tree_exceptional_entry(p)) 131 return -EEXIST; 132 133 mapping->nrexceptional--; 134 if (shadowp) 135 *shadowp = p; 136 } 137 __radix_tree_replace(&mapping->page_tree, node, slot, page, 138 workingset_lookup_update(mapping)); 139 mapping->nrpages++; 140 return 0; 141 } 142 143 static void page_cache_tree_delete(struct address_space *mapping, 144 struct page *page, void *shadow) 145 { 146 int i, nr; 147 148 /* hugetlb pages are represented by one entry in the radix tree */ 149 nr = PageHuge(page) ? 1 : hpage_nr_pages(page); 150 151 VM_BUG_ON_PAGE(!PageLocked(page), page); 152 VM_BUG_ON_PAGE(PageTail(page), page); 153 VM_BUG_ON_PAGE(nr != 1 && shadow, page); 154 155 for (i = 0; i < nr; i++) { 156 struct radix_tree_node *node; 157 void **slot; 158 159 __radix_tree_lookup(&mapping->page_tree, page->index + i, 160 &node, &slot); 161 162 VM_BUG_ON_PAGE(!node && nr != 1, page); 163 164 radix_tree_clear_tags(&mapping->page_tree, node, slot); 165 __radix_tree_replace(&mapping->page_tree, node, slot, shadow, 166 workingset_lookup_update(mapping)); 167 } 168 169 page->mapping = NULL; 170 /* Leave page->index set: truncation lookup relies upon it */ 171 172 if (shadow) { 173 mapping->nrexceptional += nr; 174 /* 175 * Make sure the nrexceptional update is committed before 176 * the nrpages update so that final truncate racing 177 * with reclaim does not see both counters 0 at the 178 * same time and miss a shadow entry. 179 */ 180 smp_wmb(); 181 } 182 mapping->nrpages -= nr; 183 } 184 185 static void unaccount_page_cache_page(struct address_space *mapping, 186 struct page *page) 187 { 188 int nr; 189 190 /* 191 * if we're uptodate, flush out into the cleancache, otherwise 192 * invalidate any existing cleancache entries. We can't leave 193 * stale data around in the cleancache once our page is gone 194 */ 195 if (PageUptodate(page) && PageMappedToDisk(page)) 196 cleancache_put_page(page); 197 else 198 cleancache_invalidate_page(mapping, page); 199 200 VM_BUG_ON_PAGE(PageTail(page), page); 201 VM_BUG_ON_PAGE(page_mapped(page), page); 202 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) { 203 int mapcount; 204 205 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n", 206 current->comm, page_to_pfn(page)); 207 dump_page(page, "still mapped when deleted"); 208 dump_stack(); 209 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); 210 211 mapcount = page_mapcount(page); 212 if (mapping_exiting(mapping) && 213 page_count(page) >= mapcount + 2) { 214 /* 215 * All vmas have already been torn down, so it's 216 * a good bet that actually the page is unmapped, 217 * and we'd prefer not to leak it: if we're wrong, 218 * some other bad page check should catch it later. 219 */ 220 page_mapcount_reset(page); 221 page_ref_sub(page, mapcount); 222 } 223 } 224 225 /* hugetlb pages do not participate in page cache accounting. */ 226 if (PageHuge(page)) 227 return; 228 229 nr = hpage_nr_pages(page); 230 231 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr); 232 if (PageSwapBacked(page)) { 233 __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr); 234 if (PageTransHuge(page)) 235 __dec_node_page_state(page, NR_SHMEM_THPS); 236 } else { 237 VM_BUG_ON_PAGE(PageTransHuge(page), page); 238 } 239 240 /* 241 * At this point page must be either written or cleaned by 242 * truncate. Dirty page here signals a bug and loss of 243 * unwritten data. 244 * 245 * This fixes dirty accounting after removing the page entirely 246 * but leaves PageDirty set: it has no effect for truncated 247 * page and anyway will be cleared before returning page into 248 * buddy allocator. 249 */ 250 if (WARN_ON_ONCE(PageDirty(page))) 251 account_page_cleaned(page, mapping, inode_to_wb(mapping->host)); 252 } 253 254 /* 255 * Delete a page from the page cache and free it. Caller has to make 256 * sure the page is locked and that nobody else uses it - or that usage 257 * is safe. The caller must hold the mapping's tree_lock. 258 */ 259 void __delete_from_page_cache(struct page *page, void *shadow) 260 { 261 struct address_space *mapping = page->mapping; 262 263 trace_mm_filemap_delete_from_page_cache(page); 264 265 unaccount_page_cache_page(mapping, page); 266 page_cache_tree_delete(mapping, page, shadow); 267 } 268 269 static void page_cache_free_page(struct address_space *mapping, 270 struct page *page) 271 { 272 void (*freepage)(struct page *); 273 274 freepage = mapping->a_ops->freepage; 275 if (freepage) 276 freepage(page); 277 278 if (PageTransHuge(page) && !PageHuge(page)) { 279 page_ref_sub(page, HPAGE_PMD_NR); 280 VM_BUG_ON_PAGE(page_count(page) <= 0, page); 281 } else { 282 put_page(page); 283 } 284 } 285 286 /** 287 * delete_from_page_cache - delete page from page cache 288 * @page: the page which the kernel is trying to remove from page cache 289 * 290 * This must be called only on pages that have been verified to be in the page 291 * cache and locked. It will never put the page into the free list, the caller 292 * has a reference on the page. 293 */ 294 void delete_from_page_cache(struct page *page) 295 { 296 struct address_space *mapping = page_mapping(page); 297 unsigned long flags; 298 299 BUG_ON(!PageLocked(page)); 300 spin_lock_irqsave(&mapping->tree_lock, flags); 301 __delete_from_page_cache(page, NULL); 302 spin_unlock_irqrestore(&mapping->tree_lock, flags); 303 304 page_cache_free_page(mapping, page); 305 } 306 EXPORT_SYMBOL(delete_from_page_cache); 307 308 /* 309 * page_cache_tree_delete_batch - delete several pages from page cache 310 * @mapping: the mapping to which pages belong 311 * @pvec: pagevec with pages to delete 312 * 313 * The function walks over mapping->page_tree and removes pages passed in @pvec 314 * from the radix tree. The function expects @pvec to be sorted by page index. 315 * It tolerates holes in @pvec (radix tree entries at those indices are not 316 * modified). The function expects only THP head pages to be present in the 317 * @pvec and takes care to delete all corresponding tail pages from the radix 318 * tree as well. 319 * 320 * The function expects mapping->tree_lock to be held. 321 */ 322 static void 323 page_cache_tree_delete_batch(struct address_space *mapping, 324 struct pagevec *pvec) 325 { 326 struct radix_tree_iter iter; 327 void **slot; 328 int total_pages = 0; 329 int i = 0, tail_pages = 0; 330 struct page *page; 331 pgoff_t start; 332 333 start = pvec->pages[0]->index; 334 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { 335 if (i >= pagevec_count(pvec) && !tail_pages) 336 break; 337 page = radix_tree_deref_slot_protected(slot, 338 &mapping->tree_lock); 339 if (radix_tree_exceptional_entry(page)) 340 continue; 341 if (!tail_pages) { 342 /* 343 * Some page got inserted in our range? Skip it. We 344 * have our pages locked so they are protected from 345 * being removed. 346 */ 347 if (page != pvec->pages[i]) 348 continue; 349 WARN_ON_ONCE(!PageLocked(page)); 350 if (PageTransHuge(page) && !PageHuge(page)) 351 tail_pages = HPAGE_PMD_NR - 1; 352 page->mapping = NULL; 353 /* 354 * Leave page->index set: truncation lookup relies 355 * upon it 356 */ 357 i++; 358 } else { 359 tail_pages--; 360 } 361 radix_tree_clear_tags(&mapping->page_tree, iter.node, slot); 362 __radix_tree_replace(&mapping->page_tree, iter.node, slot, NULL, 363 workingset_lookup_update(mapping)); 364 total_pages++; 365 } 366 mapping->nrpages -= total_pages; 367 } 368 369 void delete_from_page_cache_batch(struct address_space *mapping, 370 struct pagevec *pvec) 371 { 372 int i; 373 unsigned long flags; 374 375 if (!pagevec_count(pvec)) 376 return; 377 378 spin_lock_irqsave(&mapping->tree_lock, flags); 379 for (i = 0; i < pagevec_count(pvec); i++) { 380 trace_mm_filemap_delete_from_page_cache(pvec->pages[i]); 381 382 unaccount_page_cache_page(mapping, pvec->pages[i]); 383 } 384 page_cache_tree_delete_batch(mapping, pvec); 385 spin_unlock_irqrestore(&mapping->tree_lock, flags); 386 387 for (i = 0; i < pagevec_count(pvec); i++) 388 page_cache_free_page(mapping, pvec->pages[i]); 389 } 390 391 int filemap_check_errors(struct address_space *mapping) 392 { 393 int ret = 0; 394 /* Check for outstanding write errors */ 395 if (test_bit(AS_ENOSPC, &mapping->flags) && 396 test_and_clear_bit(AS_ENOSPC, &mapping->flags)) 397 ret = -ENOSPC; 398 if (test_bit(AS_EIO, &mapping->flags) && 399 test_and_clear_bit(AS_EIO, &mapping->flags)) 400 ret = -EIO; 401 return ret; 402 } 403 EXPORT_SYMBOL(filemap_check_errors); 404 405 static int filemap_check_and_keep_errors(struct address_space *mapping) 406 { 407 /* Check for outstanding write errors */ 408 if (test_bit(AS_EIO, &mapping->flags)) 409 return -EIO; 410 if (test_bit(AS_ENOSPC, &mapping->flags)) 411 return -ENOSPC; 412 return 0; 413 } 414 415 /** 416 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range 417 * @mapping: address space structure to write 418 * @start: offset in bytes where the range starts 419 * @end: offset in bytes where the range ends (inclusive) 420 * @sync_mode: enable synchronous operation 421 * 422 * Start writeback against all of a mapping's dirty pages that lie 423 * within the byte offsets <start, end> inclusive. 424 * 425 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as 426 * opposed to a regular memory cleansing writeback. The difference between 427 * these two operations is that if a dirty page/buffer is encountered, it must 428 * be waited upon, and not just skipped over. 429 */ 430 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, 431 loff_t end, int sync_mode) 432 { 433 int ret; 434 struct writeback_control wbc = { 435 .sync_mode = sync_mode, 436 .nr_to_write = LONG_MAX, 437 .range_start = start, 438 .range_end = end, 439 }; 440 441 if (!mapping_cap_writeback_dirty(mapping)) 442 return 0; 443 444 wbc_attach_fdatawrite_inode(&wbc, mapping->host); 445 ret = do_writepages(mapping, &wbc); 446 wbc_detach_inode(&wbc); 447 return ret; 448 } 449 450 static inline int __filemap_fdatawrite(struct address_space *mapping, 451 int sync_mode) 452 { 453 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); 454 } 455 456 int filemap_fdatawrite(struct address_space *mapping) 457 { 458 return __filemap_fdatawrite(mapping, WB_SYNC_ALL); 459 } 460 EXPORT_SYMBOL(filemap_fdatawrite); 461 462 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, 463 loff_t end) 464 { 465 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); 466 } 467 EXPORT_SYMBOL(filemap_fdatawrite_range); 468 469 /** 470 * filemap_flush - mostly a non-blocking flush 471 * @mapping: target address_space 472 * 473 * This is a mostly non-blocking flush. Not suitable for data-integrity 474 * purposes - I/O may not be started against all dirty pages. 475 */ 476 int filemap_flush(struct address_space *mapping) 477 { 478 return __filemap_fdatawrite(mapping, WB_SYNC_NONE); 479 } 480 EXPORT_SYMBOL(filemap_flush); 481 482 /** 483 * filemap_range_has_page - check if a page exists in range. 484 * @mapping: address space within which to check 485 * @start_byte: offset in bytes where the range starts 486 * @end_byte: offset in bytes where the range ends (inclusive) 487 * 488 * Find at least one page in the range supplied, usually used to check if 489 * direct writing in this range will trigger a writeback. 490 */ 491 bool filemap_range_has_page(struct address_space *mapping, 492 loff_t start_byte, loff_t end_byte) 493 { 494 pgoff_t index = start_byte >> PAGE_SHIFT; 495 pgoff_t end = end_byte >> PAGE_SHIFT; 496 struct page *page; 497 498 if (end_byte < start_byte) 499 return false; 500 501 if (mapping->nrpages == 0) 502 return false; 503 504 if (!find_get_pages_range(mapping, &index, end, 1, &page)) 505 return false; 506 put_page(page); 507 return true; 508 } 509 EXPORT_SYMBOL(filemap_range_has_page); 510 511 static void __filemap_fdatawait_range(struct address_space *mapping, 512 loff_t start_byte, loff_t end_byte) 513 { 514 pgoff_t index = start_byte >> PAGE_SHIFT; 515 pgoff_t end = end_byte >> PAGE_SHIFT; 516 struct pagevec pvec; 517 int nr_pages; 518 519 if (end_byte < start_byte) 520 return; 521 522 pagevec_init(&pvec); 523 while (index <= end) { 524 unsigned i; 525 526 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, 527 end, PAGECACHE_TAG_WRITEBACK); 528 if (!nr_pages) 529 break; 530 531 for (i = 0; i < nr_pages; i++) { 532 struct page *page = pvec.pages[i]; 533 534 wait_on_page_writeback(page); 535 ClearPageError(page); 536 } 537 pagevec_release(&pvec); 538 cond_resched(); 539 } 540 } 541 542 /** 543 * filemap_fdatawait_range - wait for writeback to complete 544 * @mapping: address space structure to wait for 545 * @start_byte: offset in bytes where the range starts 546 * @end_byte: offset in bytes where the range ends (inclusive) 547 * 548 * Walk the list of under-writeback pages of the given address space 549 * in the given range and wait for all of them. Check error status of 550 * the address space and return it. 551 * 552 * Since the error status of the address space is cleared by this function, 553 * callers are responsible for checking the return value and handling and/or 554 * reporting the error. 555 */ 556 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, 557 loff_t end_byte) 558 { 559 __filemap_fdatawait_range(mapping, start_byte, end_byte); 560 return filemap_check_errors(mapping); 561 } 562 EXPORT_SYMBOL(filemap_fdatawait_range); 563 564 /** 565 * file_fdatawait_range - wait for writeback to complete 566 * @file: file pointing to address space structure to wait for 567 * @start_byte: offset in bytes where the range starts 568 * @end_byte: offset in bytes where the range ends (inclusive) 569 * 570 * Walk the list of under-writeback pages of the address space that file 571 * refers to, in the given range and wait for all of them. Check error 572 * status of the address space vs. the file->f_wb_err cursor and return it. 573 * 574 * Since the error status of the file is advanced by this function, 575 * callers are responsible for checking the return value and handling and/or 576 * reporting the error. 577 */ 578 int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte) 579 { 580 struct address_space *mapping = file->f_mapping; 581 582 __filemap_fdatawait_range(mapping, start_byte, end_byte); 583 return file_check_and_advance_wb_err(file); 584 } 585 EXPORT_SYMBOL(file_fdatawait_range); 586 587 /** 588 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors 589 * @mapping: address space structure to wait for 590 * 591 * Walk the list of under-writeback pages of the given address space 592 * and wait for all of them. Unlike filemap_fdatawait(), this function 593 * does not clear error status of the address space. 594 * 595 * Use this function if callers don't handle errors themselves. Expected 596 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), 597 * fsfreeze(8) 598 */ 599 int filemap_fdatawait_keep_errors(struct address_space *mapping) 600 { 601 __filemap_fdatawait_range(mapping, 0, LLONG_MAX); 602 return filemap_check_and_keep_errors(mapping); 603 } 604 EXPORT_SYMBOL(filemap_fdatawait_keep_errors); 605 606 static bool mapping_needs_writeback(struct address_space *mapping) 607 { 608 return (!dax_mapping(mapping) && mapping->nrpages) || 609 (dax_mapping(mapping) && mapping->nrexceptional); 610 } 611 612 int filemap_write_and_wait(struct address_space *mapping) 613 { 614 int err = 0; 615 616 if (mapping_needs_writeback(mapping)) { 617 err = filemap_fdatawrite(mapping); 618 /* 619 * Even if the above returned error, the pages may be 620 * written partially (e.g. -ENOSPC), so we wait for it. 621 * But the -EIO is special case, it may indicate the worst 622 * thing (e.g. bug) happened, so we avoid waiting for it. 623 */ 624 if (err != -EIO) { 625 int err2 = filemap_fdatawait(mapping); 626 if (!err) 627 err = err2; 628 } else { 629 /* Clear any previously stored errors */ 630 filemap_check_errors(mapping); 631 } 632 } else { 633 err = filemap_check_errors(mapping); 634 } 635 return err; 636 } 637 EXPORT_SYMBOL(filemap_write_and_wait); 638 639 /** 640 * filemap_write_and_wait_range - write out & wait on a file range 641 * @mapping: the address_space for the pages 642 * @lstart: offset in bytes where the range starts 643 * @lend: offset in bytes where the range ends (inclusive) 644 * 645 * Write out and wait upon file offsets lstart->lend, inclusive. 646 * 647 * Note that @lend is inclusive (describes the last byte to be written) so 648 * that this function can be used to write to the very end-of-file (end = -1). 649 */ 650 int filemap_write_and_wait_range(struct address_space *mapping, 651 loff_t lstart, loff_t lend) 652 { 653 int err = 0; 654 655 if (mapping_needs_writeback(mapping)) { 656 err = __filemap_fdatawrite_range(mapping, lstart, lend, 657 WB_SYNC_ALL); 658 /* See comment of filemap_write_and_wait() */ 659 if (err != -EIO) { 660 int err2 = filemap_fdatawait_range(mapping, 661 lstart, lend); 662 if (!err) 663 err = err2; 664 } else { 665 /* Clear any previously stored errors */ 666 filemap_check_errors(mapping); 667 } 668 } else { 669 err = filemap_check_errors(mapping); 670 } 671 return err; 672 } 673 EXPORT_SYMBOL(filemap_write_and_wait_range); 674 675 void __filemap_set_wb_err(struct address_space *mapping, int err) 676 { 677 errseq_t eseq = errseq_set(&mapping->wb_err, err); 678 679 trace_filemap_set_wb_err(mapping, eseq); 680 } 681 EXPORT_SYMBOL(__filemap_set_wb_err); 682 683 /** 684 * file_check_and_advance_wb_err - report wb error (if any) that was previously 685 * and advance wb_err to current one 686 * @file: struct file on which the error is being reported 687 * 688 * When userland calls fsync (or something like nfsd does the equivalent), we 689 * want to report any writeback errors that occurred since the last fsync (or 690 * since the file was opened if there haven't been any). 691 * 692 * Grab the wb_err from the mapping. If it matches what we have in the file, 693 * then just quickly return 0. The file is all caught up. 694 * 695 * If it doesn't match, then take the mapping value, set the "seen" flag in 696 * it and try to swap it into place. If it works, or another task beat us 697 * to it with the new value, then update the f_wb_err and return the error 698 * portion. The error at this point must be reported via proper channels 699 * (a'la fsync, or NFS COMMIT operation, etc.). 700 * 701 * While we handle mapping->wb_err with atomic operations, the f_wb_err 702 * value is protected by the f_lock since we must ensure that it reflects 703 * the latest value swapped in for this file descriptor. 704 */ 705 int file_check_and_advance_wb_err(struct file *file) 706 { 707 int err = 0; 708 errseq_t old = READ_ONCE(file->f_wb_err); 709 struct address_space *mapping = file->f_mapping; 710 711 /* Locklessly handle the common case where nothing has changed */ 712 if (errseq_check(&mapping->wb_err, old)) { 713 /* Something changed, must use slow path */ 714 spin_lock(&file->f_lock); 715 old = file->f_wb_err; 716 err = errseq_check_and_advance(&mapping->wb_err, 717 &file->f_wb_err); 718 trace_file_check_and_advance_wb_err(file, old); 719 spin_unlock(&file->f_lock); 720 } 721 722 /* 723 * We're mostly using this function as a drop in replacement for 724 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect 725 * that the legacy code would have had on these flags. 726 */ 727 clear_bit(AS_EIO, &mapping->flags); 728 clear_bit(AS_ENOSPC, &mapping->flags); 729 return err; 730 } 731 EXPORT_SYMBOL(file_check_and_advance_wb_err); 732 733 /** 734 * file_write_and_wait_range - write out & wait on a file range 735 * @file: file pointing to address_space with pages 736 * @lstart: offset in bytes where the range starts 737 * @lend: offset in bytes where the range ends (inclusive) 738 * 739 * Write out and wait upon file offsets lstart->lend, inclusive. 740 * 741 * Note that @lend is inclusive (describes the last byte to be written) so 742 * that this function can be used to write to the very end-of-file (end = -1). 743 * 744 * After writing out and waiting on the data, we check and advance the 745 * f_wb_err cursor to the latest value, and return any errors detected there. 746 */ 747 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend) 748 { 749 int err = 0, err2; 750 struct address_space *mapping = file->f_mapping; 751 752 if (mapping_needs_writeback(mapping)) { 753 err = __filemap_fdatawrite_range(mapping, lstart, lend, 754 WB_SYNC_ALL); 755 /* See comment of filemap_write_and_wait() */ 756 if (err != -EIO) 757 __filemap_fdatawait_range(mapping, lstart, lend); 758 } 759 err2 = file_check_and_advance_wb_err(file); 760 if (!err) 761 err = err2; 762 return err; 763 } 764 EXPORT_SYMBOL(file_write_and_wait_range); 765 766 /** 767 * replace_page_cache_page - replace a pagecache page with a new one 768 * @old: page to be replaced 769 * @new: page to replace with 770 * @gfp_mask: allocation mode 771 * 772 * This function replaces a page in the pagecache with a new one. On 773 * success it acquires the pagecache reference for the new page and 774 * drops it for the old page. Both the old and new pages must be 775 * locked. This function does not add the new page to the LRU, the 776 * caller must do that. 777 * 778 * The remove + add is atomic. The only way this function can fail is 779 * memory allocation failure. 780 */ 781 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask) 782 { 783 int error; 784 785 VM_BUG_ON_PAGE(!PageLocked(old), old); 786 VM_BUG_ON_PAGE(!PageLocked(new), new); 787 VM_BUG_ON_PAGE(new->mapping, new); 788 789 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); 790 if (!error) { 791 struct address_space *mapping = old->mapping; 792 void (*freepage)(struct page *); 793 unsigned long flags; 794 795 pgoff_t offset = old->index; 796 freepage = mapping->a_ops->freepage; 797 798 get_page(new); 799 new->mapping = mapping; 800 new->index = offset; 801 802 spin_lock_irqsave(&mapping->tree_lock, flags); 803 __delete_from_page_cache(old, NULL); 804 error = page_cache_tree_insert(mapping, new, NULL); 805 BUG_ON(error); 806 807 /* 808 * hugetlb pages do not participate in page cache accounting. 809 */ 810 if (!PageHuge(new)) 811 __inc_node_page_state(new, NR_FILE_PAGES); 812 if (PageSwapBacked(new)) 813 __inc_node_page_state(new, NR_SHMEM); 814 spin_unlock_irqrestore(&mapping->tree_lock, flags); 815 mem_cgroup_migrate(old, new); 816 radix_tree_preload_end(); 817 if (freepage) 818 freepage(old); 819 put_page(old); 820 } 821 822 return error; 823 } 824 EXPORT_SYMBOL_GPL(replace_page_cache_page); 825 826 static int __add_to_page_cache_locked(struct page *page, 827 struct address_space *mapping, 828 pgoff_t offset, gfp_t gfp_mask, 829 void **shadowp) 830 { 831 int huge = PageHuge(page); 832 struct mem_cgroup *memcg; 833 int error; 834 835 VM_BUG_ON_PAGE(!PageLocked(page), page); 836 VM_BUG_ON_PAGE(PageSwapBacked(page), page); 837 838 if (!huge) { 839 error = mem_cgroup_try_charge(page, current->mm, 840 gfp_mask, &memcg, false); 841 if (error) 842 return error; 843 } 844 845 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM); 846 if (error) { 847 if (!huge) 848 mem_cgroup_cancel_charge(page, memcg, false); 849 return error; 850 } 851 852 get_page(page); 853 page->mapping = mapping; 854 page->index = offset; 855 856 spin_lock_irq(&mapping->tree_lock); 857 error = page_cache_tree_insert(mapping, page, shadowp); 858 radix_tree_preload_end(); 859 if (unlikely(error)) 860 goto err_insert; 861 862 /* hugetlb pages do not participate in page cache accounting. */ 863 if (!huge) 864 __inc_node_page_state(page, NR_FILE_PAGES); 865 spin_unlock_irq(&mapping->tree_lock); 866 if (!huge) 867 mem_cgroup_commit_charge(page, memcg, false, false); 868 trace_mm_filemap_add_to_page_cache(page); 869 return 0; 870 err_insert: 871 page->mapping = NULL; 872 /* Leave page->index set: truncation relies upon it */ 873 spin_unlock_irq(&mapping->tree_lock); 874 if (!huge) 875 mem_cgroup_cancel_charge(page, memcg, false); 876 put_page(page); 877 return error; 878 } 879 880 /** 881 * add_to_page_cache_locked - add a locked page to the pagecache 882 * @page: page to add 883 * @mapping: the page's address_space 884 * @offset: page index 885 * @gfp_mask: page allocation mode 886 * 887 * This function is used to add a page to the pagecache. It must be locked. 888 * This function does not add the page to the LRU. The caller must do that. 889 */ 890 int add_to_page_cache_locked(struct page *page, struct address_space *mapping, 891 pgoff_t offset, gfp_t gfp_mask) 892 { 893 return __add_to_page_cache_locked(page, mapping, offset, 894 gfp_mask, NULL); 895 } 896 EXPORT_SYMBOL(add_to_page_cache_locked); 897 898 int add_to_page_cache_lru(struct page *page, struct address_space *mapping, 899 pgoff_t offset, gfp_t gfp_mask) 900 { 901 void *shadow = NULL; 902 int ret; 903 904 __SetPageLocked(page); 905 ret = __add_to_page_cache_locked(page, mapping, offset, 906 gfp_mask, &shadow); 907 if (unlikely(ret)) 908 __ClearPageLocked(page); 909 else { 910 /* 911 * The page might have been evicted from cache only 912 * recently, in which case it should be activated like 913 * any other repeatedly accessed page. 914 * The exception is pages getting rewritten; evicting other 915 * data from the working set, only to cache data that will 916 * get overwritten with something else, is a waste of memory. 917 */ 918 if (!(gfp_mask & __GFP_WRITE) && 919 shadow && workingset_refault(shadow)) { 920 SetPageActive(page); 921 workingset_activation(page); 922 } else 923 ClearPageActive(page); 924 lru_cache_add(page); 925 } 926 return ret; 927 } 928 EXPORT_SYMBOL_GPL(add_to_page_cache_lru); 929 930 #ifdef CONFIG_NUMA 931 struct page *__page_cache_alloc(gfp_t gfp) 932 { 933 int n; 934 struct page *page; 935 936 if (cpuset_do_page_mem_spread()) { 937 unsigned int cpuset_mems_cookie; 938 do { 939 cpuset_mems_cookie = read_mems_allowed_begin(); 940 n = cpuset_mem_spread_node(); 941 page = __alloc_pages_node(n, gfp, 0); 942 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie)); 943 944 return page; 945 } 946 return alloc_pages(gfp, 0); 947 } 948 EXPORT_SYMBOL(__page_cache_alloc); 949 #endif 950 951 /* 952 * In order to wait for pages to become available there must be 953 * waitqueues associated with pages. By using a hash table of 954 * waitqueues where the bucket discipline is to maintain all 955 * waiters on the same queue and wake all when any of the pages 956 * become available, and for the woken contexts to check to be 957 * sure the appropriate page became available, this saves space 958 * at a cost of "thundering herd" phenomena during rare hash 959 * collisions. 960 */ 961 #define PAGE_WAIT_TABLE_BITS 8 962 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS) 963 static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned; 964 965 static wait_queue_head_t *page_waitqueue(struct page *page) 966 { 967 return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)]; 968 } 969 970 void __init pagecache_init(void) 971 { 972 int i; 973 974 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++) 975 init_waitqueue_head(&page_wait_table[i]); 976 977 page_writeback_init(); 978 } 979 980 /* This has the same layout as wait_bit_key - see fs/cachefiles/rdwr.c */ 981 struct wait_page_key { 982 struct page *page; 983 int bit_nr; 984 int page_match; 985 }; 986 987 struct wait_page_queue { 988 struct page *page; 989 int bit_nr; 990 wait_queue_entry_t wait; 991 }; 992 993 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg) 994 { 995 struct wait_page_key *key = arg; 996 struct wait_page_queue *wait_page 997 = container_of(wait, struct wait_page_queue, wait); 998 999 if (wait_page->page != key->page) 1000 return 0; 1001 key->page_match = 1; 1002 1003 if (wait_page->bit_nr != key->bit_nr) 1004 return 0; 1005 1006 /* Stop walking if it's locked */ 1007 if (test_bit(key->bit_nr, &key->page->flags)) 1008 return -1; 1009 1010 return autoremove_wake_function(wait, mode, sync, key); 1011 } 1012 1013 static void wake_up_page_bit(struct page *page, int bit_nr) 1014 { 1015 wait_queue_head_t *q = page_waitqueue(page); 1016 struct wait_page_key key; 1017 unsigned long flags; 1018 wait_queue_entry_t bookmark; 1019 1020 key.page = page; 1021 key.bit_nr = bit_nr; 1022 key.page_match = 0; 1023 1024 bookmark.flags = 0; 1025 bookmark.private = NULL; 1026 bookmark.func = NULL; 1027 INIT_LIST_HEAD(&bookmark.entry); 1028 1029 spin_lock_irqsave(&q->lock, flags); 1030 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); 1031 1032 while (bookmark.flags & WQ_FLAG_BOOKMARK) { 1033 /* 1034 * Take a breather from holding the lock, 1035 * allow pages that finish wake up asynchronously 1036 * to acquire the lock and remove themselves 1037 * from wait queue 1038 */ 1039 spin_unlock_irqrestore(&q->lock, flags); 1040 cpu_relax(); 1041 spin_lock_irqsave(&q->lock, flags); 1042 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); 1043 } 1044 1045 /* 1046 * It is possible for other pages to have collided on the waitqueue 1047 * hash, so in that case check for a page match. That prevents a long- 1048 * term waiter 1049 * 1050 * It is still possible to miss a case here, when we woke page waiters 1051 * and removed them from the waitqueue, but there are still other 1052 * page waiters. 1053 */ 1054 if (!waitqueue_active(q) || !key.page_match) { 1055 ClearPageWaiters(page); 1056 /* 1057 * It's possible to miss clearing Waiters here, when we woke 1058 * our page waiters, but the hashed waitqueue has waiters for 1059 * other pages on it. 1060 * 1061 * That's okay, it's a rare case. The next waker will clear it. 1062 */ 1063 } 1064 spin_unlock_irqrestore(&q->lock, flags); 1065 } 1066 1067 static void wake_up_page(struct page *page, int bit) 1068 { 1069 if (!PageWaiters(page)) 1070 return; 1071 wake_up_page_bit(page, bit); 1072 } 1073 1074 static inline int wait_on_page_bit_common(wait_queue_head_t *q, 1075 struct page *page, int bit_nr, int state, bool lock) 1076 { 1077 struct wait_page_queue wait_page; 1078 wait_queue_entry_t *wait = &wait_page.wait; 1079 int ret = 0; 1080 1081 init_wait(wait); 1082 wait->flags = lock ? WQ_FLAG_EXCLUSIVE : 0; 1083 wait->func = wake_page_function; 1084 wait_page.page = page; 1085 wait_page.bit_nr = bit_nr; 1086 1087 for (;;) { 1088 spin_lock_irq(&q->lock); 1089 1090 if (likely(list_empty(&wait->entry))) { 1091 __add_wait_queue_entry_tail(q, wait); 1092 SetPageWaiters(page); 1093 } 1094 1095 set_current_state(state); 1096 1097 spin_unlock_irq(&q->lock); 1098 1099 if (likely(test_bit(bit_nr, &page->flags))) { 1100 io_schedule(); 1101 } 1102 1103 if (lock) { 1104 if (!test_and_set_bit_lock(bit_nr, &page->flags)) 1105 break; 1106 } else { 1107 if (!test_bit(bit_nr, &page->flags)) 1108 break; 1109 } 1110 1111 if (unlikely(signal_pending_state(state, current))) { 1112 ret = -EINTR; 1113 break; 1114 } 1115 } 1116 1117 finish_wait(q, wait); 1118 1119 /* 1120 * A signal could leave PageWaiters set. Clearing it here if 1121 * !waitqueue_active would be possible (by open-coding finish_wait), 1122 * but still fail to catch it in the case of wait hash collision. We 1123 * already can fail to clear wait hash collision cases, so don't 1124 * bother with signals either. 1125 */ 1126 1127 return ret; 1128 } 1129 1130 void wait_on_page_bit(struct page *page, int bit_nr) 1131 { 1132 wait_queue_head_t *q = page_waitqueue(page); 1133 wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, false); 1134 } 1135 EXPORT_SYMBOL(wait_on_page_bit); 1136 1137 int wait_on_page_bit_killable(struct page *page, int bit_nr) 1138 { 1139 wait_queue_head_t *q = page_waitqueue(page); 1140 return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, false); 1141 } 1142 EXPORT_SYMBOL(wait_on_page_bit_killable); 1143 1144 /** 1145 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue 1146 * @page: Page defining the wait queue of interest 1147 * @waiter: Waiter to add to the queue 1148 * 1149 * Add an arbitrary @waiter to the wait queue for the nominated @page. 1150 */ 1151 void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter) 1152 { 1153 wait_queue_head_t *q = page_waitqueue(page); 1154 unsigned long flags; 1155 1156 spin_lock_irqsave(&q->lock, flags); 1157 __add_wait_queue_entry_tail(q, waiter); 1158 SetPageWaiters(page); 1159 spin_unlock_irqrestore(&q->lock, flags); 1160 } 1161 EXPORT_SYMBOL_GPL(add_page_wait_queue); 1162 1163 #ifndef clear_bit_unlock_is_negative_byte 1164 1165 /* 1166 * PG_waiters is the high bit in the same byte as PG_lock. 1167 * 1168 * On x86 (and on many other architectures), we can clear PG_lock and 1169 * test the sign bit at the same time. But if the architecture does 1170 * not support that special operation, we just do this all by hand 1171 * instead. 1172 * 1173 * The read of PG_waiters has to be after (or concurrently with) PG_locked 1174 * being cleared, but a memory barrier should be unneccssary since it is 1175 * in the same byte as PG_locked. 1176 */ 1177 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem) 1178 { 1179 clear_bit_unlock(nr, mem); 1180 /* smp_mb__after_atomic(); */ 1181 return test_bit(PG_waiters, mem); 1182 } 1183 1184 #endif 1185 1186 /** 1187 * unlock_page - unlock a locked page 1188 * @page: the page 1189 * 1190 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). 1191 * Also wakes sleepers in wait_on_page_writeback() because the wakeup 1192 * mechanism between PageLocked pages and PageWriteback pages is shared. 1193 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. 1194 * 1195 * Note that this depends on PG_waiters being the sign bit in the byte 1196 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to 1197 * clear the PG_locked bit and test PG_waiters at the same time fairly 1198 * portably (architectures that do LL/SC can test any bit, while x86 can 1199 * test the sign bit). 1200 */ 1201 void unlock_page(struct page *page) 1202 { 1203 BUILD_BUG_ON(PG_waiters != 7); 1204 page = compound_head(page); 1205 VM_BUG_ON_PAGE(!PageLocked(page), page); 1206 if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags)) 1207 wake_up_page_bit(page, PG_locked); 1208 } 1209 EXPORT_SYMBOL(unlock_page); 1210 1211 /** 1212 * end_page_writeback - end writeback against a page 1213 * @page: the page 1214 */ 1215 void end_page_writeback(struct page *page) 1216 { 1217 /* 1218 * TestClearPageReclaim could be used here but it is an atomic 1219 * operation and overkill in this particular case. Failing to 1220 * shuffle a page marked for immediate reclaim is too mild to 1221 * justify taking an atomic operation penalty at the end of 1222 * ever page writeback. 1223 */ 1224 if (PageReclaim(page)) { 1225 ClearPageReclaim(page); 1226 rotate_reclaimable_page(page); 1227 } 1228 1229 if (!test_clear_page_writeback(page)) 1230 BUG(); 1231 1232 smp_mb__after_atomic(); 1233 wake_up_page(page, PG_writeback); 1234 } 1235 EXPORT_SYMBOL(end_page_writeback); 1236 1237 /* 1238 * After completing I/O on a page, call this routine to update the page 1239 * flags appropriately 1240 */ 1241 void page_endio(struct page *page, bool is_write, int err) 1242 { 1243 if (!is_write) { 1244 if (!err) { 1245 SetPageUptodate(page); 1246 } else { 1247 ClearPageUptodate(page); 1248 SetPageError(page); 1249 } 1250 unlock_page(page); 1251 } else { 1252 if (err) { 1253 struct address_space *mapping; 1254 1255 SetPageError(page); 1256 mapping = page_mapping(page); 1257 if (mapping) 1258 mapping_set_error(mapping, err); 1259 } 1260 end_page_writeback(page); 1261 } 1262 } 1263 EXPORT_SYMBOL_GPL(page_endio); 1264 1265 /** 1266 * __lock_page - get a lock on the page, assuming we need to sleep to get it 1267 * @__page: the page to lock 1268 */ 1269 void __lock_page(struct page *__page) 1270 { 1271 struct page *page = compound_head(__page); 1272 wait_queue_head_t *q = page_waitqueue(page); 1273 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, true); 1274 } 1275 EXPORT_SYMBOL(__lock_page); 1276 1277 int __lock_page_killable(struct page *__page) 1278 { 1279 struct page *page = compound_head(__page); 1280 wait_queue_head_t *q = page_waitqueue(page); 1281 return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE, true); 1282 } 1283 EXPORT_SYMBOL_GPL(__lock_page_killable); 1284 1285 /* 1286 * Return values: 1287 * 1 - page is locked; mmap_sem is still held. 1288 * 0 - page is not locked. 1289 * mmap_sem has been released (up_read()), unless flags had both 1290 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in 1291 * which case mmap_sem is still held. 1292 * 1293 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1 1294 * with the page locked and the mmap_sem unperturbed. 1295 */ 1296 int __lock_page_or_retry(struct page *page, struct mm_struct *mm, 1297 unsigned int flags) 1298 { 1299 if (flags & FAULT_FLAG_ALLOW_RETRY) { 1300 /* 1301 * CAUTION! In this case, mmap_sem is not released 1302 * even though return 0. 1303 */ 1304 if (flags & FAULT_FLAG_RETRY_NOWAIT) 1305 return 0; 1306 1307 up_read(&mm->mmap_sem); 1308 if (flags & FAULT_FLAG_KILLABLE) 1309 wait_on_page_locked_killable(page); 1310 else 1311 wait_on_page_locked(page); 1312 return 0; 1313 } else { 1314 if (flags & FAULT_FLAG_KILLABLE) { 1315 int ret; 1316 1317 ret = __lock_page_killable(page); 1318 if (ret) { 1319 up_read(&mm->mmap_sem); 1320 return 0; 1321 } 1322 } else 1323 __lock_page(page); 1324 return 1; 1325 } 1326 } 1327 1328 /** 1329 * page_cache_next_hole - find the next hole (not-present entry) 1330 * @mapping: mapping 1331 * @index: index 1332 * @max_scan: maximum range to search 1333 * 1334 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the 1335 * lowest indexed hole. 1336 * 1337 * Returns: the index of the hole if found, otherwise returns an index 1338 * outside of the set specified (in which case 'return - index >= 1339 * max_scan' will be true). In rare cases of index wrap-around, 0 will 1340 * be returned. 1341 * 1342 * page_cache_next_hole may be called under rcu_read_lock. However, 1343 * like radix_tree_gang_lookup, this will not atomically search a 1344 * snapshot of the tree at a single point in time. For example, if a 1345 * hole is created at index 5, then subsequently a hole is created at 1346 * index 10, page_cache_next_hole covering both indexes may return 10 1347 * if called under rcu_read_lock. 1348 */ 1349 pgoff_t page_cache_next_hole(struct address_space *mapping, 1350 pgoff_t index, unsigned long max_scan) 1351 { 1352 unsigned long i; 1353 1354 for (i = 0; i < max_scan; i++) { 1355 struct page *page; 1356 1357 page = radix_tree_lookup(&mapping->page_tree, index); 1358 if (!page || radix_tree_exceptional_entry(page)) 1359 break; 1360 index++; 1361 if (index == 0) 1362 break; 1363 } 1364 1365 return index; 1366 } 1367 EXPORT_SYMBOL(page_cache_next_hole); 1368 1369 /** 1370 * page_cache_prev_hole - find the prev hole (not-present entry) 1371 * @mapping: mapping 1372 * @index: index 1373 * @max_scan: maximum range to search 1374 * 1375 * Search backwards in the range [max(index-max_scan+1, 0), index] for 1376 * the first hole. 1377 * 1378 * Returns: the index of the hole if found, otherwise returns an index 1379 * outside of the set specified (in which case 'index - return >= 1380 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX 1381 * will be returned. 1382 * 1383 * page_cache_prev_hole may be called under rcu_read_lock. However, 1384 * like radix_tree_gang_lookup, this will not atomically search a 1385 * snapshot of the tree at a single point in time. For example, if a 1386 * hole is created at index 10, then subsequently a hole is created at 1387 * index 5, page_cache_prev_hole covering both indexes may return 5 if 1388 * called under rcu_read_lock. 1389 */ 1390 pgoff_t page_cache_prev_hole(struct address_space *mapping, 1391 pgoff_t index, unsigned long max_scan) 1392 { 1393 unsigned long i; 1394 1395 for (i = 0; i < max_scan; i++) { 1396 struct page *page; 1397 1398 page = radix_tree_lookup(&mapping->page_tree, index); 1399 if (!page || radix_tree_exceptional_entry(page)) 1400 break; 1401 index--; 1402 if (index == ULONG_MAX) 1403 break; 1404 } 1405 1406 return index; 1407 } 1408 EXPORT_SYMBOL(page_cache_prev_hole); 1409 1410 /** 1411 * find_get_entry - find and get a page cache entry 1412 * @mapping: the address_space to search 1413 * @offset: the page cache index 1414 * 1415 * Looks up the page cache slot at @mapping & @offset. If there is a 1416 * page cache page, it is returned with an increased refcount. 1417 * 1418 * If the slot holds a shadow entry of a previously evicted page, or a 1419 * swap entry from shmem/tmpfs, it is returned. 1420 * 1421 * Otherwise, %NULL is returned. 1422 */ 1423 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset) 1424 { 1425 void **pagep; 1426 struct page *head, *page; 1427 1428 rcu_read_lock(); 1429 repeat: 1430 page = NULL; 1431 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset); 1432 if (pagep) { 1433 page = radix_tree_deref_slot(pagep); 1434 if (unlikely(!page)) 1435 goto out; 1436 if (radix_tree_exception(page)) { 1437 if (radix_tree_deref_retry(page)) 1438 goto repeat; 1439 /* 1440 * A shadow entry of a recently evicted page, 1441 * or a swap entry from shmem/tmpfs. Return 1442 * it without attempting to raise page count. 1443 */ 1444 goto out; 1445 } 1446 1447 head = compound_head(page); 1448 if (!page_cache_get_speculative(head)) 1449 goto repeat; 1450 1451 /* The page was split under us? */ 1452 if (compound_head(page) != head) { 1453 put_page(head); 1454 goto repeat; 1455 } 1456 1457 /* 1458 * Has the page moved? 1459 * This is part of the lockless pagecache protocol. See 1460 * include/linux/pagemap.h for details. 1461 */ 1462 if (unlikely(page != *pagep)) { 1463 put_page(head); 1464 goto repeat; 1465 } 1466 } 1467 out: 1468 rcu_read_unlock(); 1469 1470 return page; 1471 } 1472 EXPORT_SYMBOL(find_get_entry); 1473 1474 /** 1475 * find_lock_entry - locate, pin and lock a page cache entry 1476 * @mapping: the address_space to search 1477 * @offset: the page cache index 1478 * 1479 * Looks up the page cache slot at @mapping & @offset. If there is a 1480 * page cache page, it is returned locked and with an increased 1481 * refcount. 1482 * 1483 * If the slot holds a shadow entry of a previously evicted page, or a 1484 * swap entry from shmem/tmpfs, it is returned. 1485 * 1486 * Otherwise, %NULL is returned. 1487 * 1488 * find_lock_entry() may sleep. 1489 */ 1490 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset) 1491 { 1492 struct page *page; 1493 1494 repeat: 1495 page = find_get_entry(mapping, offset); 1496 if (page && !radix_tree_exception(page)) { 1497 lock_page(page); 1498 /* Has the page been truncated? */ 1499 if (unlikely(page_mapping(page) != mapping)) { 1500 unlock_page(page); 1501 put_page(page); 1502 goto repeat; 1503 } 1504 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page); 1505 } 1506 return page; 1507 } 1508 EXPORT_SYMBOL(find_lock_entry); 1509 1510 /** 1511 * pagecache_get_page - find and get a page reference 1512 * @mapping: the address_space to search 1513 * @offset: the page index 1514 * @fgp_flags: PCG flags 1515 * @gfp_mask: gfp mask to use for the page cache data page allocation 1516 * 1517 * Looks up the page cache slot at @mapping & @offset. 1518 * 1519 * PCG flags modify how the page is returned. 1520 * 1521 * @fgp_flags can be: 1522 * 1523 * - FGP_ACCESSED: the page will be marked accessed 1524 * - FGP_LOCK: Page is return locked 1525 * - FGP_CREAT: If page is not present then a new page is allocated using 1526 * @gfp_mask and added to the page cache and the VM's LRU 1527 * list. The page is returned locked and with an increased 1528 * refcount. Otherwise, NULL is returned. 1529 * 1530 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even 1531 * if the GFP flags specified for FGP_CREAT are atomic. 1532 * 1533 * If there is a page cache page, it is returned with an increased refcount. 1534 */ 1535 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset, 1536 int fgp_flags, gfp_t gfp_mask) 1537 { 1538 struct page *page; 1539 1540 repeat: 1541 page = find_get_entry(mapping, offset); 1542 if (radix_tree_exceptional_entry(page)) 1543 page = NULL; 1544 if (!page) 1545 goto no_page; 1546 1547 if (fgp_flags & FGP_LOCK) { 1548 if (fgp_flags & FGP_NOWAIT) { 1549 if (!trylock_page(page)) { 1550 put_page(page); 1551 return NULL; 1552 } 1553 } else { 1554 lock_page(page); 1555 } 1556 1557 /* Has the page been truncated? */ 1558 if (unlikely(page->mapping != mapping)) { 1559 unlock_page(page); 1560 put_page(page); 1561 goto repeat; 1562 } 1563 VM_BUG_ON_PAGE(page->index != offset, page); 1564 } 1565 1566 if (page && (fgp_flags & FGP_ACCESSED)) 1567 mark_page_accessed(page); 1568 1569 no_page: 1570 if (!page && (fgp_flags & FGP_CREAT)) { 1571 int err; 1572 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping)) 1573 gfp_mask |= __GFP_WRITE; 1574 if (fgp_flags & FGP_NOFS) 1575 gfp_mask &= ~__GFP_FS; 1576 1577 page = __page_cache_alloc(gfp_mask); 1578 if (!page) 1579 return NULL; 1580 1581 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK))) 1582 fgp_flags |= FGP_LOCK; 1583 1584 /* Init accessed so avoid atomic mark_page_accessed later */ 1585 if (fgp_flags & FGP_ACCESSED) 1586 __SetPageReferenced(page); 1587 1588 err = add_to_page_cache_lru(page, mapping, offset, 1589 gfp_mask & GFP_RECLAIM_MASK); 1590 if (unlikely(err)) { 1591 put_page(page); 1592 page = NULL; 1593 if (err == -EEXIST) 1594 goto repeat; 1595 } 1596 } 1597 1598 return page; 1599 } 1600 EXPORT_SYMBOL(pagecache_get_page); 1601 1602 /** 1603 * find_get_entries - gang pagecache lookup 1604 * @mapping: The address_space to search 1605 * @start: The starting page cache index 1606 * @nr_entries: The maximum number of entries 1607 * @entries: Where the resulting entries are placed 1608 * @indices: The cache indices corresponding to the entries in @entries 1609 * 1610 * find_get_entries() will search for and return a group of up to 1611 * @nr_entries entries in the mapping. The entries are placed at 1612 * @entries. find_get_entries() takes a reference against any actual 1613 * pages it returns. 1614 * 1615 * The search returns a group of mapping-contiguous page cache entries 1616 * with ascending indexes. There may be holes in the indices due to 1617 * not-present pages. 1618 * 1619 * Any shadow entries of evicted pages, or swap entries from 1620 * shmem/tmpfs, are included in the returned array. 1621 * 1622 * find_get_entries() returns the number of pages and shadow entries 1623 * which were found. 1624 */ 1625 unsigned find_get_entries(struct address_space *mapping, 1626 pgoff_t start, unsigned int nr_entries, 1627 struct page **entries, pgoff_t *indices) 1628 { 1629 void **slot; 1630 unsigned int ret = 0; 1631 struct radix_tree_iter iter; 1632 1633 if (!nr_entries) 1634 return 0; 1635 1636 rcu_read_lock(); 1637 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { 1638 struct page *head, *page; 1639 repeat: 1640 page = radix_tree_deref_slot(slot); 1641 if (unlikely(!page)) 1642 continue; 1643 if (radix_tree_exception(page)) { 1644 if (radix_tree_deref_retry(page)) { 1645 slot = radix_tree_iter_retry(&iter); 1646 continue; 1647 } 1648 /* 1649 * A shadow entry of a recently evicted page, a swap 1650 * entry from shmem/tmpfs or a DAX entry. Return it 1651 * without attempting to raise page count. 1652 */ 1653 goto export; 1654 } 1655 1656 head = compound_head(page); 1657 if (!page_cache_get_speculative(head)) 1658 goto repeat; 1659 1660 /* The page was split under us? */ 1661 if (compound_head(page) != head) { 1662 put_page(head); 1663 goto repeat; 1664 } 1665 1666 /* Has the page moved? */ 1667 if (unlikely(page != *slot)) { 1668 put_page(head); 1669 goto repeat; 1670 } 1671 export: 1672 indices[ret] = iter.index; 1673 entries[ret] = page; 1674 if (++ret == nr_entries) 1675 break; 1676 } 1677 rcu_read_unlock(); 1678 return ret; 1679 } 1680 1681 /** 1682 * find_get_pages_range - gang pagecache lookup 1683 * @mapping: The address_space to search 1684 * @start: The starting page index 1685 * @end: The final page index (inclusive) 1686 * @nr_pages: The maximum number of pages 1687 * @pages: Where the resulting pages are placed 1688 * 1689 * find_get_pages_range() will search for and return a group of up to @nr_pages 1690 * pages in the mapping starting at index @start and up to index @end 1691 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes 1692 * a reference against the returned pages. 1693 * 1694 * The search returns a group of mapping-contiguous pages with ascending 1695 * indexes. There may be holes in the indices due to not-present pages. 1696 * We also update @start to index the next page for the traversal. 1697 * 1698 * find_get_pages_range() returns the number of pages which were found. If this 1699 * number is smaller than @nr_pages, the end of specified range has been 1700 * reached. 1701 */ 1702 unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start, 1703 pgoff_t end, unsigned int nr_pages, 1704 struct page **pages) 1705 { 1706 struct radix_tree_iter iter; 1707 void **slot; 1708 unsigned ret = 0; 1709 1710 if (unlikely(!nr_pages)) 1711 return 0; 1712 1713 rcu_read_lock(); 1714 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, *start) { 1715 struct page *head, *page; 1716 1717 if (iter.index > end) 1718 break; 1719 repeat: 1720 page = radix_tree_deref_slot(slot); 1721 if (unlikely(!page)) 1722 continue; 1723 1724 if (radix_tree_exception(page)) { 1725 if (radix_tree_deref_retry(page)) { 1726 slot = radix_tree_iter_retry(&iter); 1727 continue; 1728 } 1729 /* 1730 * A shadow entry of a recently evicted page, 1731 * or a swap entry from shmem/tmpfs. Skip 1732 * over it. 1733 */ 1734 continue; 1735 } 1736 1737 head = compound_head(page); 1738 if (!page_cache_get_speculative(head)) 1739 goto repeat; 1740 1741 /* The page was split under us? */ 1742 if (compound_head(page) != head) { 1743 put_page(head); 1744 goto repeat; 1745 } 1746 1747 /* Has the page moved? */ 1748 if (unlikely(page != *slot)) { 1749 put_page(head); 1750 goto repeat; 1751 } 1752 1753 pages[ret] = page; 1754 if (++ret == nr_pages) { 1755 *start = pages[ret - 1]->index + 1; 1756 goto out; 1757 } 1758 } 1759 1760 /* 1761 * We come here when there is no page beyond @end. We take care to not 1762 * overflow the index @start as it confuses some of the callers. This 1763 * breaks the iteration when there is page at index -1 but that is 1764 * already broken anyway. 1765 */ 1766 if (end == (pgoff_t)-1) 1767 *start = (pgoff_t)-1; 1768 else 1769 *start = end + 1; 1770 out: 1771 rcu_read_unlock(); 1772 1773 return ret; 1774 } 1775 1776 /** 1777 * find_get_pages_contig - gang contiguous pagecache lookup 1778 * @mapping: The address_space to search 1779 * @index: The starting page index 1780 * @nr_pages: The maximum number of pages 1781 * @pages: Where the resulting pages are placed 1782 * 1783 * find_get_pages_contig() works exactly like find_get_pages(), except 1784 * that the returned number of pages are guaranteed to be contiguous. 1785 * 1786 * find_get_pages_contig() returns the number of pages which were found. 1787 */ 1788 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, 1789 unsigned int nr_pages, struct page **pages) 1790 { 1791 struct radix_tree_iter iter; 1792 void **slot; 1793 unsigned int ret = 0; 1794 1795 if (unlikely(!nr_pages)) 1796 return 0; 1797 1798 rcu_read_lock(); 1799 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) { 1800 struct page *head, *page; 1801 repeat: 1802 page = radix_tree_deref_slot(slot); 1803 /* The hole, there no reason to continue */ 1804 if (unlikely(!page)) 1805 break; 1806 1807 if (radix_tree_exception(page)) { 1808 if (radix_tree_deref_retry(page)) { 1809 slot = radix_tree_iter_retry(&iter); 1810 continue; 1811 } 1812 /* 1813 * A shadow entry of a recently evicted page, 1814 * or a swap entry from shmem/tmpfs. Stop 1815 * looking for contiguous pages. 1816 */ 1817 break; 1818 } 1819 1820 head = compound_head(page); 1821 if (!page_cache_get_speculative(head)) 1822 goto repeat; 1823 1824 /* The page was split under us? */ 1825 if (compound_head(page) != head) { 1826 put_page(head); 1827 goto repeat; 1828 } 1829 1830 /* Has the page moved? */ 1831 if (unlikely(page != *slot)) { 1832 put_page(head); 1833 goto repeat; 1834 } 1835 1836 /* 1837 * must check mapping and index after taking the ref. 1838 * otherwise we can get both false positives and false 1839 * negatives, which is just confusing to the caller. 1840 */ 1841 if (page->mapping == NULL || page_to_pgoff(page) != iter.index) { 1842 put_page(page); 1843 break; 1844 } 1845 1846 pages[ret] = page; 1847 if (++ret == nr_pages) 1848 break; 1849 } 1850 rcu_read_unlock(); 1851 return ret; 1852 } 1853 EXPORT_SYMBOL(find_get_pages_contig); 1854 1855 /** 1856 * find_get_pages_range_tag - find and return pages in given range matching @tag 1857 * @mapping: the address_space to search 1858 * @index: the starting page index 1859 * @end: The final page index (inclusive) 1860 * @tag: the tag index 1861 * @nr_pages: the maximum number of pages 1862 * @pages: where the resulting pages are placed 1863 * 1864 * Like find_get_pages, except we only return pages which are tagged with 1865 * @tag. We update @index to index the next page for the traversal. 1866 */ 1867 unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index, 1868 pgoff_t end, int tag, unsigned int nr_pages, 1869 struct page **pages) 1870 { 1871 struct radix_tree_iter iter; 1872 void **slot; 1873 unsigned ret = 0; 1874 1875 if (unlikely(!nr_pages)) 1876 return 0; 1877 1878 rcu_read_lock(); 1879 radix_tree_for_each_tagged(slot, &mapping->page_tree, 1880 &iter, *index, tag) { 1881 struct page *head, *page; 1882 1883 if (iter.index > end) 1884 break; 1885 repeat: 1886 page = radix_tree_deref_slot(slot); 1887 if (unlikely(!page)) 1888 continue; 1889 1890 if (radix_tree_exception(page)) { 1891 if (radix_tree_deref_retry(page)) { 1892 slot = radix_tree_iter_retry(&iter); 1893 continue; 1894 } 1895 /* 1896 * A shadow entry of a recently evicted page. 1897 * 1898 * Those entries should never be tagged, but 1899 * this tree walk is lockless and the tags are 1900 * looked up in bulk, one radix tree node at a 1901 * time, so there is a sizable window for page 1902 * reclaim to evict a page we saw tagged. 1903 * 1904 * Skip over it. 1905 */ 1906 continue; 1907 } 1908 1909 head = compound_head(page); 1910 if (!page_cache_get_speculative(head)) 1911 goto repeat; 1912 1913 /* The page was split under us? */ 1914 if (compound_head(page) != head) { 1915 put_page(head); 1916 goto repeat; 1917 } 1918 1919 /* Has the page moved? */ 1920 if (unlikely(page != *slot)) { 1921 put_page(head); 1922 goto repeat; 1923 } 1924 1925 pages[ret] = page; 1926 if (++ret == nr_pages) { 1927 *index = pages[ret - 1]->index + 1; 1928 goto out; 1929 } 1930 } 1931 1932 /* 1933 * We come here when we got at @end. We take care to not overflow the 1934 * index @index as it confuses some of the callers. This breaks the 1935 * iteration when there is page at index -1 but that is already broken 1936 * anyway. 1937 */ 1938 if (end == (pgoff_t)-1) 1939 *index = (pgoff_t)-1; 1940 else 1941 *index = end + 1; 1942 out: 1943 rcu_read_unlock(); 1944 1945 return ret; 1946 } 1947 EXPORT_SYMBOL(find_get_pages_range_tag); 1948 1949 /** 1950 * find_get_entries_tag - find and return entries that match @tag 1951 * @mapping: the address_space to search 1952 * @start: the starting page cache index 1953 * @tag: the tag index 1954 * @nr_entries: the maximum number of entries 1955 * @entries: where the resulting entries are placed 1956 * @indices: the cache indices corresponding to the entries in @entries 1957 * 1958 * Like find_get_entries, except we only return entries which are tagged with 1959 * @tag. 1960 */ 1961 unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start, 1962 int tag, unsigned int nr_entries, 1963 struct page **entries, pgoff_t *indices) 1964 { 1965 void **slot; 1966 unsigned int ret = 0; 1967 struct radix_tree_iter iter; 1968 1969 if (!nr_entries) 1970 return 0; 1971 1972 rcu_read_lock(); 1973 radix_tree_for_each_tagged(slot, &mapping->page_tree, 1974 &iter, start, tag) { 1975 struct page *head, *page; 1976 repeat: 1977 page = radix_tree_deref_slot(slot); 1978 if (unlikely(!page)) 1979 continue; 1980 if (radix_tree_exception(page)) { 1981 if (radix_tree_deref_retry(page)) { 1982 slot = radix_tree_iter_retry(&iter); 1983 continue; 1984 } 1985 1986 /* 1987 * A shadow entry of a recently evicted page, a swap 1988 * entry from shmem/tmpfs or a DAX entry. Return it 1989 * without attempting to raise page count. 1990 */ 1991 goto export; 1992 } 1993 1994 head = compound_head(page); 1995 if (!page_cache_get_speculative(head)) 1996 goto repeat; 1997 1998 /* The page was split under us? */ 1999 if (compound_head(page) != head) { 2000 put_page(head); 2001 goto repeat; 2002 } 2003 2004 /* Has the page moved? */ 2005 if (unlikely(page != *slot)) { 2006 put_page(head); 2007 goto repeat; 2008 } 2009 export: 2010 indices[ret] = iter.index; 2011 entries[ret] = page; 2012 if (++ret == nr_entries) 2013 break; 2014 } 2015 rcu_read_unlock(); 2016 return ret; 2017 } 2018 EXPORT_SYMBOL(find_get_entries_tag); 2019 2020 /* 2021 * CD/DVDs are error prone. When a medium error occurs, the driver may fail 2022 * a _large_ part of the i/o request. Imagine the worst scenario: 2023 * 2024 * ---R__________________________________________B__________ 2025 * ^ reading here ^ bad block(assume 4k) 2026 * 2027 * read(R) => miss => readahead(R...B) => media error => frustrating retries 2028 * => failing the whole request => read(R) => read(R+1) => 2029 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => 2030 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => 2031 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... 2032 * 2033 * It is going insane. Fix it by quickly scaling down the readahead size. 2034 */ 2035 static void shrink_readahead_size_eio(struct file *filp, 2036 struct file_ra_state *ra) 2037 { 2038 ra->ra_pages /= 4; 2039 } 2040 2041 /** 2042 * generic_file_buffered_read - generic file read routine 2043 * @iocb: the iocb to read 2044 * @iter: data destination 2045 * @written: already copied 2046 * 2047 * This is a generic file read routine, and uses the 2048 * mapping->a_ops->readpage() function for the actual low-level stuff. 2049 * 2050 * This is really ugly. But the goto's actually try to clarify some 2051 * of the logic when it comes to error handling etc. 2052 */ 2053 static ssize_t generic_file_buffered_read(struct kiocb *iocb, 2054 struct iov_iter *iter, ssize_t written) 2055 { 2056 struct file *filp = iocb->ki_filp; 2057 struct address_space *mapping = filp->f_mapping; 2058 struct inode *inode = mapping->host; 2059 struct file_ra_state *ra = &filp->f_ra; 2060 loff_t *ppos = &iocb->ki_pos; 2061 pgoff_t index; 2062 pgoff_t last_index; 2063 pgoff_t prev_index; 2064 unsigned long offset; /* offset into pagecache page */ 2065 unsigned int prev_offset; 2066 int error = 0; 2067 2068 if (unlikely(*ppos >= inode->i_sb->s_maxbytes)) 2069 return 0; 2070 iov_iter_truncate(iter, inode->i_sb->s_maxbytes); 2071 2072 index = *ppos >> PAGE_SHIFT; 2073 prev_index = ra->prev_pos >> PAGE_SHIFT; 2074 prev_offset = ra->prev_pos & (PAGE_SIZE-1); 2075 last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT; 2076 offset = *ppos & ~PAGE_MASK; 2077 2078 for (;;) { 2079 struct page *page; 2080 pgoff_t end_index; 2081 loff_t isize; 2082 unsigned long nr, ret; 2083 2084 cond_resched(); 2085 find_page: 2086 if (fatal_signal_pending(current)) { 2087 error = -EINTR; 2088 goto out; 2089 } 2090 2091 page = find_get_page(mapping, index); 2092 if (!page) { 2093 if (iocb->ki_flags & IOCB_NOWAIT) 2094 goto would_block; 2095 page_cache_sync_readahead(mapping, 2096 ra, filp, 2097 index, last_index - index); 2098 page = find_get_page(mapping, index); 2099 if (unlikely(page == NULL)) 2100 goto no_cached_page; 2101 } 2102 if (PageReadahead(page)) { 2103 page_cache_async_readahead(mapping, 2104 ra, filp, page, 2105 index, last_index - index); 2106 } 2107 if (!PageUptodate(page)) { 2108 if (iocb->ki_flags & IOCB_NOWAIT) { 2109 put_page(page); 2110 goto would_block; 2111 } 2112 2113 /* 2114 * See comment in do_read_cache_page on why 2115 * wait_on_page_locked is used to avoid unnecessarily 2116 * serialisations and why it's safe. 2117 */ 2118 error = wait_on_page_locked_killable(page); 2119 if (unlikely(error)) 2120 goto readpage_error; 2121 if (PageUptodate(page)) 2122 goto page_ok; 2123 2124 if (inode->i_blkbits == PAGE_SHIFT || 2125 !mapping->a_ops->is_partially_uptodate) 2126 goto page_not_up_to_date; 2127 /* pipes can't handle partially uptodate pages */ 2128 if (unlikely(iter->type & ITER_PIPE)) 2129 goto page_not_up_to_date; 2130 if (!trylock_page(page)) 2131 goto page_not_up_to_date; 2132 /* Did it get truncated before we got the lock? */ 2133 if (!page->mapping) 2134 goto page_not_up_to_date_locked; 2135 if (!mapping->a_ops->is_partially_uptodate(page, 2136 offset, iter->count)) 2137 goto page_not_up_to_date_locked; 2138 unlock_page(page); 2139 } 2140 page_ok: 2141 /* 2142 * i_size must be checked after we know the page is Uptodate. 2143 * 2144 * Checking i_size after the check allows us to calculate 2145 * the correct value for "nr", which means the zero-filled 2146 * part of the page is not copied back to userspace (unless 2147 * another truncate extends the file - this is desired though). 2148 */ 2149 2150 isize = i_size_read(inode); 2151 end_index = (isize - 1) >> PAGE_SHIFT; 2152 if (unlikely(!isize || index > end_index)) { 2153 put_page(page); 2154 goto out; 2155 } 2156 2157 /* nr is the maximum number of bytes to copy from this page */ 2158 nr = PAGE_SIZE; 2159 if (index == end_index) { 2160 nr = ((isize - 1) & ~PAGE_MASK) + 1; 2161 if (nr <= offset) { 2162 put_page(page); 2163 goto out; 2164 } 2165 } 2166 nr = nr - offset; 2167 2168 /* If users can be writing to this page using arbitrary 2169 * virtual addresses, take care about potential aliasing 2170 * before reading the page on the kernel side. 2171 */ 2172 if (mapping_writably_mapped(mapping)) 2173 flush_dcache_page(page); 2174 2175 /* 2176 * When a sequential read accesses a page several times, 2177 * only mark it as accessed the first time. 2178 */ 2179 if (prev_index != index || offset != prev_offset) 2180 mark_page_accessed(page); 2181 prev_index = index; 2182 2183 /* 2184 * Ok, we have the page, and it's up-to-date, so 2185 * now we can copy it to user space... 2186 */ 2187 2188 ret = copy_page_to_iter(page, offset, nr, iter); 2189 offset += ret; 2190 index += offset >> PAGE_SHIFT; 2191 offset &= ~PAGE_MASK; 2192 prev_offset = offset; 2193 2194 put_page(page); 2195 written += ret; 2196 if (!iov_iter_count(iter)) 2197 goto out; 2198 if (ret < nr) { 2199 error = -EFAULT; 2200 goto out; 2201 } 2202 continue; 2203 2204 page_not_up_to_date: 2205 /* Get exclusive access to the page ... */ 2206 error = lock_page_killable(page); 2207 if (unlikely(error)) 2208 goto readpage_error; 2209 2210 page_not_up_to_date_locked: 2211 /* Did it get truncated before we got the lock? */ 2212 if (!page->mapping) { 2213 unlock_page(page); 2214 put_page(page); 2215 continue; 2216 } 2217 2218 /* Did somebody else fill it already? */ 2219 if (PageUptodate(page)) { 2220 unlock_page(page); 2221 goto page_ok; 2222 } 2223 2224 readpage: 2225 /* 2226 * A previous I/O error may have been due to temporary 2227 * failures, eg. multipath errors. 2228 * PG_error will be set again if readpage fails. 2229 */ 2230 ClearPageError(page); 2231 /* Start the actual read. The read will unlock the page. */ 2232 error = mapping->a_ops->readpage(filp, page); 2233 2234 if (unlikely(error)) { 2235 if (error == AOP_TRUNCATED_PAGE) { 2236 put_page(page); 2237 error = 0; 2238 goto find_page; 2239 } 2240 goto readpage_error; 2241 } 2242 2243 if (!PageUptodate(page)) { 2244 error = lock_page_killable(page); 2245 if (unlikely(error)) 2246 goto readpage_error; 2247 if (!PageUptodate(page)) { 2248 if (page->mapping == NULL) { 2249 /* 2250 * invalidate_mapping_pages got it 2251 */ 2252 unlock_page(page); 2253 put_page(page); 2254 goto find_page; 2255 } 2256 unlock_page(page); 2257 shrink_readahead_size_eio(filp, ra); 2258 error = -EIO; 2259 goto readpage_error; 2260 } 2261 unlock_page(page); 2262 } 2263 2264 goto page_ok; 2265 2266 readpage_error: 2267 /* UHHUH! A synchronous read error occurred. Report it */ 2268 put_page(page); 2269 goto out; 2270 2271 no_cached_page: 2272 /* 2273 * Ok, it wasn't cached, so we need to create a new 2274 * page.. 2275 */ 2276 page = page_cache_alloc(mapping); 2277 if (!page) { 2278 error = -ENOMEM; 2279 goto out; 2280 } 2281 error = add_to_page_cache_lru(page, mapping, index, 2282 mapping_gfp_constraint(mapping, GFP_KERNEL)); 2283 if (error) { 2284 put_page(page); 2285 if (error == -EEXIST) { 2286 error = 0; 2287 goto find_page; 2288 } 2289 goto out; 2290 } 2291 goto readpage; 2292 } 2293 2294 would_block: 2295 error = -EAGAIN; 2296 out: 2297 ra->prev_pos = prev_index; 2298 ra->prev_pos <<= PAGE_SHIFT; 2299 ra->prev_pos |= prev_offset; 2300 2301 *ppos = ((loff_t)index << PAGE_SHIFT) + offset; 2302 file_accessed(filp); 2303 return written ? written : error; 2304 } 2305 2306 /** 2307 * generic_file_read_iter - generic filesystem read routine 2308 * @iocb: kernel I/O control block 2309 * @iter: destination for the data read 2310 * 2311 * This is the "read_iter()" routine for all filesystems 2312 * that can use the page cache directly. 2313 */ 2314 ssize_t 2315 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) 2316 { 2317 size_t count = iov_iter_count(iter); 2318 ssize_t retval = 0; 2319 2320 if (!count) 2321 goto out; /* skip atime */ 2322 2323 if (iocb->ki_flags & IOCB_DIRECT) { 2324 struct file *file = iocb->ki_filp; 2325 struct address_space *mapping = file->f_mapping; 2326 struct inode *inode = mapping->host; 2327 loff_t size; 2328 2329 size = i_size_read(inode); 2330 if (iocb->ki_flags & IOCB_NOWAIT) { 2331 if (filemap_range_has_page(mapping, iocb->ki_pos, 2332 iocb->ki_pos + count - 1)) 2333 return -EAGAIN; 2334 } else { 2335 retval = filemap_write_and_wait_range(mapping, 2336 iocb->ki_pos, 2337 iocb->ki_pos + count - 1); 2338 if (retval < 0) 2339 goto out; 2340 } 2341 2342 file_accessed(file); 2343 2344 retval = mapping->a_ops->direct_IO(iocb, iter); 2345 if (retval >= 0) { 2346 iocb->ki_pos += retval; 2347 count -= retval; 2348 } 2349 iov_iter_revert(iter, count - iov_iter_count(iter)); 2350 2351 /* 2352 * Btrfs can have a short DIO read if we encounter 2353 * compressed extents, so if there was an error, or if 2354 * we've already read everything we wanted to, or if 2355 * there was a short read because we hit EOF, go ahead 2356 * and return. Otherwise fallthrough to buffered io for 2357 * the rest of the read. Buffered reads will not work for 2358 * DAX files, so don't bother trying. 2359 */ 2360 if (retval < 0 || !count || iocb->ki_pos >= size || 2361 IS_DAX(inode)) 2362 goto out; 2363 } 2364 2365 retval = generic_file_buffered_read(iocb, iter, retval); 2366 out: 2367 return retval; 2368 } 2369 EXPORT_SYMBOL(generic_file_read_iter); 2370 2371 #ifdef CONFIG_MMU 2372 /** 2373 * page_cache_read - adds requested page to the page cache if not already there 2374 * @file: file to read 2375 * @offset: page index 2376 * @gfp_mask: memory allocation flags 2377 * 2378 * This adds the requested page to the page cache if it isn't already there, 2379 * and schedules an I/O to read in its contents from disk. 2380 */ 2381 static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask) 2382 { 2383 struct address_space *mapping = file->f_mapping; 2384 struct page *page; 2385 int ret; 2386 2387 do { 2388 page = __page_cache_alloc(gfp_mask); 2389 if (!page) 2390 return -ENOMEM; 2391 2392 ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask & GFP_KERNEL); 2393 if (ret == 0) 2394 ret = mapping->a_ops->readpage(file, page); 2395 else if (ret == -EEXIST) 2396 ret = 0; /* losing race to add is OK */ 2397 2398 put_page(page); 2399 2400 } while (ret == AOP_TRUNCATED_PAGE); 2401 2402 return ret; 2403 } 2404 2405 #define MMAP_LOTSAMISS (100) 2406 2407 /* 2408 * Synchronous readahead happens when we don't even find 2409 * a page in the page cache at all. 2410 */ 2411 static void do_sync_mmap_readahead(struct vm_area_struct *vma, 2412 struct file_ra_state *ra, 2413 struct file *file, 2414 pgoff_t offset) 2415 { 2416 struct address_space *mapping = file->f_mapping; 2417 2418 /* If we don't want any read-ahead, don't bother */ 2419 if (vma->vm_flags & VM_RAND_READ) 2420 return; 2421 if (!ra->ra_pages) 2422 return; 2423 2424 if (vma->vm_flags & VM_SEQ_READ) { 2425 page_cache_sync_readahead(mapping, ra, file, offset, 2426 ra->ra_pages); 2427 return; 2428 } 2429 2430 /* Avoid banging the cache line if not needed */ 2431 if (ra->mmap_miss < MMAP_LOTSAMISS * 10) 2432 ra->mmap_miss++; 2433 2434 /* 2435 * Do we miss much more than hit in this file? If so, 2436 * stop bothering with read-ahead. It will only hurt. 2437 */ 2438 if (ra->mmap_miss > MMAP_LOTSAMISS) 2439 return; 2440 2441 /* 2442 * mmap read-around 2443 */ 2444 ra->start = max_t(long, 0, offset - ra->ra_pages / 2); 2445 ra->size = ra->ra_pages; 2446 ra->async_size = ra->ra_pages / 4; 2447 ra_submit(ra, mapping, file); 2448 } 2449 2450 /* 2451 * Asynchronous readahead happens when we find the page and PG_readahead, 2452 * so we want to possibly extend the readahead further.. 2453 */ 2454 static void do_async_mmap_readahead(struct vm_area_struct *vma, 2455 struct file_ra_state *ra, 2456 struct file *file, 2457 struct page *page, 2458 pgoff_t offset) 2459 { 2460 struct address_space *mapping = file->f_mapping; 2461 2462 /* If we don't want any read-ahead, don't bother */ 2463 if (vma->vm_flags & VM_RAND_READ) 2464 return; 2465 if (ra->mmap_miss > 0) 2466 ra->mmap_miss--; 2467 if (PageReadahead(page)) 2468 page_cache_async_readahead(mapping, ra, file, 2469 page, offset, ra->ra_pages); 2470 } 2471 2472 /** 2473 * filemap_fault - read in file data for page fault handling 2474 * @vmf: struct vm_fault containing details of the fault 2475 * 2476 * filemap_fault() is invoked via the vma operations vector for a 2477 * mapped memory region to read in file data during a page fault. 2478 * 2479 * The goto's are kind of ugly, but this streamlines the normal case of having 2480 * it in the page cache, and handles the special cases reasonably without 2481 * having a lot of duplicated code. 2482 * 2483 * vma->vm_mm->mmap_sem must be held on entry. 2484 * 2485 * If our return value has VM_FAULT_RETRY set, it's because 2486 * lock_page_or_retry() returned 0. 2487 * The mmap_sem has usually been released in this case. 2488 * See __lock_page_or_retry() for the exception. 2489 * 2490 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem 2491 * has not been released. 2492 * 2493 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set. 2494 */ 2495 int filemap_fault(struct vm_fault *vmf) 2496 { 2497 int error; 2498 struct file *file = vmf->vma->vm_file; 2499 struct address_space *mapping = file->f_mapping; 2500 struct file_ra_state *ra = &file->f_ra; 2501 struct inode *inode = mapping->host; 2502 pgoff_t offset = vmf->pgoff; 2503 pgoff_t max_off; 2504 struct page *page; 2505 int ret = 0; 2506 2507 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); 2508 if (unlikely(offset >= max_off)) 2509 return VM_FAULT_SIGBUS; 2510 2511 /* 2512 * Do we have something in the page cache already? 2513 */ 2514 page = find_get_page(mapping, offset); 2515 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) { 2516 /* 2517 * We found the page, so try async readahead before 2518 * waiting for the lock. 2519 */ 2520 do_async_mmap_readahead(vmf->vma, ra, file, page, offset); 2521 } else if (!page) { 2522 /* No page in the page cache at all */ 2523 do_sync_mmap_readahead(vmf->vma, ra, file, offset); 2524 count_vm_event(PGMAJFAULT); 2525 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT); 2526 ret = VM_FAULT_MAJOR; 2527 retry_find: 2528 page = find_get_page(mapping, offset); 2529 if (!page) 2530 goto no_cached_page; 2531 } 2532 2533 if (!lock_page_or_retry(page, vmf->vma->vm_mm, vmf->flags)) { 2534 put_page(page); 2535 return ret | VM_FAULT_RETRY; 2536 } 2537 2538 /* Did it get truncated? */ 2539 if (unlikely(page->mapping != mapping)) { 2540 unlock_page(page); 2541 put_page(page); 2542 goto retry_find; 2543 } 2544 VM_BUG_ON_PAGE(page->index != offset, page); 2545 2546 /* 2547 * We have a locked page in the page cache, now we need to check 2548 * that it's up-to-date. If not, it is going to be due to an error. 2549 */ 2550 if (unlikely(!PageUptodate(page))) 2551 goto page_not_uptodate; 2552 2553 /* 2554 * Found the page and have a reference on it. 2555 * We must recheck i_size under page lock. 2556 */ 2557 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); 2558 if (unlikely(offset >= max_off)) { 2559 unlock_page(page); 2560 put_page(page); 2561 return VM_FAULT_SIGBUS; 2562 } 2563 2564 vmf->page = page; 2565 return ret | VM_FAULT_LOCKED; 2566 2567 no_cached_page: 2568 /* 2569 * We're only likely to ever get here if MADV_RANDOM is in 2570 * effect. 2571 */ 2572 error = page_cache_read(file, offset, vmf->gfp_mask); 2573 2574 /* 2575 * The page we want has now been added to the page cache. 2576 * In the unlikely event that someone removed it in the 2577 * meantime, we'll just come back here and read it again. 2578 */ 2579 if (error >= 0) 2580 goto retry_find; 2581 2582 /* 2583 * An error return from page_cache_read can result if the 2584 * system is low on memory, or a problem occurs while trying 2585 * to schedule I/O. 2586 */ 2587 if (error == -ENOMEM) 2588 return VM_FAULT_OOM; 2589 return VM_FAULT_SIGBUS; 2590 2591 page_not_uptodate: 2592 /* 2593 * Umm, take care of errors if the page isn't up-to-date. 2594 * Try to re-read it _once_. We do this synchronously, 2595 * because there really aren't any performance issues here 2596 * and we need to check for errors. 2597 */ 2598 ClearPageError(page); 2599 error = mapping->a_ops->readpage(file, page); 2600 if (!error) { 2601 wait_on_page_locked(page); 2602 if (!PageUptodate(page)) 2603 error = -EIO; 2604 } 2605 put_page(page); 2606 2607 if (!error || error == AOP_TRUNCATED_PAGE) 2608 goto retry_find; 2609 2610 /* Things didn't work out. Return zero to tell the mm layer so. */ 2611 shrink_readahead_size_eio(file, ra); 2612 return VM_FAULT_SIGBUS; 2613 } 2614 EXPORT_SYMBOL(filemap_fault); 2615 2616 void filemap_map_pages(struct vm_fault *vmf, 2617 pgoff_t start_pgoff, pgoff_t end_pgoff) 2618 { 2619 struct radix_tree_iter iter; 2620 void **slot; 2621 struct file *file = vmf->vma->vm_file; 2622 struct address_space *mapping = file->f_mapping; 2623 pgoff_t last_pgoff = start_pgoff; 2624 unsigned long max_idx; 2625 struct page *head, *page; 2626 2627 rcu_read_lock(); 2628 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, 2629 start_pgoff) { 2630 if (iter.index > end_pgoff) 2631 break; 2632 repeat: 2633 page = radix_tree_deref_slot(slot); 2634 if (unlikely(!page)) 2635 goto next; 2636 if (radix_tree_exception(page)) { 2637 if (radix_tree_deref_retry(page)) { 2638 slot = radix_tree_iter_retry(&iter); 2639 continue; 2640 } 2641 goto next; 2642 } 2643 2644 head = compound_head(page); 2645 if (!page_cache_get_speculative(head)) 2646 goto repeat; 2647 2648 /* The page was split under us? */ 2649 if (compound_head(page) != head) { 2650 put_page(head); 2651 goto repeat; 2652 } 2653 2654 /* Has the page moved? */ 2655 if (unlikely(page != *slot)) { 2656 put_page(head); 2657 goto repeat; 2658 } 2659 2660 if (!PageUptodate(page) || 2661 PageReadahead(page) || 2662 PageHWPoison(page)) 2663 goto skip; 2664 if (!trylock_page(page)) 2665 goto skip; 2666 2667 if (page->mapping != mapping || !PageUptodate(page)) 2668 goto unlock; 2669 2670 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE); 2671 if (page->index >= max_idx) 2672 goto unlock; 2673 2674 if (file->f_ra.mmap_miss > 0) 2675 file->f_ra.mmap_miss--; 2676 2677 vmf->address += (iter.index - last_pgoff) << PAGE_SHIFT; 2678 if (vmf->pte) 2679 vmf->pte += iter.index - last_pgoff; 2680 last_pgoff = iter.index; 2681 if (alloc_set_pte(vmf, NULL, page)) 2682 goto unlock; 2683 unlock_page(page); 2684 goto next; 2685 unlock: 2686 unlock_page(page); 2687 skip: 2688 put_page(page); 2689 next: 2690 /* Huge page is mapped? No need to proceed. */ 2691 if (pmd_trans_huge(*vmf->pmd)) 2692 break; 2693 if (iter.index == end_pgoff) 2694 break; 2695 } 2696 rcu_read_unlock(); 2697 } 2698 EXPORT_SYMBOL(filemap_map_pages); 2699 2700 int filemap_page_mkwrite(struct vm_fault *vmf) 2701 { 2702 struct page *page = vmf->page; 2703 struct inode *inode = file_inode(vmf->vma->vm_file); 2704 int ret = VM_FAULT_LOCKED; 2705 2706 sb_start_pagefault(inode->i_sb); 2707 file_update_time(vmf->vma->vm_file); 2708 lock_page(page); 2709 if (page->mapping != inode->i_mapping) { 2710 unlock_page(page); 2711 ret = VM_FAULT_NOPAGE; 2712 goto out; 2713 } 2714 /* 2715 * We mark the page dirty already here so that when freeze is in 2716 * progress, we are guaranteed that writeback during freezing will 2717 * see the dirty page and writeprotect it again. 2718 */ 2719 set_page_dirty(page); 2720 wait_for_stable_page(page); 2721 out: 2722 sb_end_pagefault(inode->i_sb); 2723 return ret; 2724 } 2725 EXPORT_SYMBOL(filemap_page_mkwrite); 2726 2727 const struct vm_operations_struct generic_file_vm_ops = { 2728 .fault = filemap_fault, 2729 .map_pages = filemap_map_pages, 2730 .page_mkwrite = filemap_page_mkwrite, 2731 }; 2732 2733 /* This is used for a general mmap of a disk file */ 2734 2735 int generic_file_mmap(struct file * file, struct vm_area_struct * vma) 2736 { 2737 struct address_space *mapping = file->f_mapping; 2738 2739 if (!mapping->a_ops->readpage) 2740 return -ENOEXEC; 2741 file_accessed(file); 2742 vma->vm_ops = &generic_file_vm_ops; 2743 return 0; 2744 } 2745 2746 /* 2747 * This is for filesystems which do not implement ->writepage. 2748 */ 2749 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) 2750 { 2751 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) 2752 return -EINVAL; 2753 return generic_file_mmap(file, vma); 2754 } 2755 #else 2756 int generic_file_mmap(struct file * file, struct vm_area_struct * vma) 2757 { 2758 return -ENOSYS; 2759 } 2760 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) 2761 { 2762 return -ENOSYS; 2763 } 2764 #endif /* CONFIG_MMU */ 2765 2766 EXPORT_SYMBOL(generic_file_mmap); 2767 EXPORT_SYMBOL(generic_file_readonly_mmap); 2768 2769 static struct page *wait_on_page_read(struct page *page) 2770 { 2771 if (!IS_ERR(page)) { 2772 wait_on_page_locked(page); 2773 if (!PageUptodate(page)) { 2774 put_page(page); 2775 page = ERR_PTR(-EIO); 2776 } 2777 } 2778 return page; 2779 } 2780 2781 static struct page *do_read_cache_page(struct address_space *mapping, 2782 pgoff_t index, 2783 int (*filler)(void *, struct page *), 2784 void *data, 2785 gfp_t gfp) 2786 { 2787 struct page *page; 2788 int err; 2789 repeat: 2790 page = find_get_page(mapping, index); 2791 if (!page) { 2792 page = __page_cache_alloc(gfp); 2793 if (!page) 2794 return ERR_PTR(-ENOMEM); 2795 err = add_to_page_cache_lru(page, mapping, index, gfp); 2796 if (unlikely(err)) { 2797 put_page(page); 2798 if (err == -EEXIST) 2799 goto repeat; 2800 /* Presumably ENOMEM for radix tree node */ 2801 return ERR_PTR(err); 2802 } 2803 2804 filler: 2805 err = filler(data, page); 2806 if (err < 0) { 2807 put_page(page); 2808 return ERR_PTR(err); 2809 } 2810 2811 page = wait_on_page_read(page); 2812 if (IS_ERR(page)) 2813 return page; 2814 goto out; 2815 } 2816 if (PageUptodate(page)) 2817 goto out; 2818 2819 /* 2820 * Page is not up to date and may be locked due one of the following 2821 * case a: Page is being filled and the page lock is held 2822 * case b: Read/write error clearing the page uptodate status 2823 * case c: Truncation in progress (page locked) 2824 * case d: Reclaim in progress 2825 * 2826 * Case a, the page will be up to date when the page is unlocked. 2827 * There is no need to serialise on the page lock here as the page 2828 * is pinned so the lock gives no additional protection. Even if the 2829 * the page is truncated, the data is still valid if PageUptodate as 2830 * it's a race vs truncate race. 2831 * Case b, the page will not be up to date 2832 * Case c, the page may be truncated but in itself, the data may still 2833 * be valid after IO completes as it's a read vs truncate race. The 2834 * operation must restart if the page is not uptodate on unlock but 2835 * otherwise serialising on page lock to stabilise the mapping gives 2836 * no additional guarantees to the caller as the page lock is 2837 * released before return. 2838 * Case d, similar to truncation. If reclaim holds the page lock, it 2839 * will be a race with remove_mapping that determines if the mapping 2840 * is valid on unlock but otherwise the data is valid and there is 2841 * no need to serialise with page lock. 2842 * 2843 * As the page lock gives no additional guarantee, we optimistically 2844 * wait on the page to be unlocked and check if it's up to date and 2845 * use the page if it is. Otherwise, the page lock is required to 2846 * distinguish between the different cases. The motivation is that we 2847 * avoid spurious serialisations and wakeups when multiple processes 2848 * wait on the same page for IO to complete. 2849 */ 2850 wait_on_page_locked(page); 2851 if (PageUptodate(page)) 2852 goto out; 2853 2854 /* Distinguish between all the cases under the safety of the lock */ 2855 lock_page(page); 2856 2857 /* Case c or d, restart the operation */ 2858 if (!page->mapping) { 2859 unlock_page(page); 2860 put_page(page); 2861 goto repeat; 2862 } 2863 2864 /* Someone else locked and filled the page in a very small window */ 2865 if (PageUptodate(page)) { 2866 unlock_page(page); 2867 goto out; 2868 } 2869 goto filler; 2870 2871 out: 2872 mark_page_accessed(page); 2873 return page; 2874 } 2875 2876 /** 2877 * read_cache_page - read into page cache, fill it if needed 2878 * @mapping: the page's address_space 2879 * @index: the page index 2880 * @filler: function to perform the read 2881 * @data: first arg to filler(data, page) function, often left as NULL 2882 * 2883 * Read into the page cache. If a page already exists, and PageUptodate() is 2884 * not set, try to fill the page and wait for it to become unlocked. 2885 * 2886 * If the page does not get brought uptodate, return -EIO. 2887 */ 2888 struct page *read_cache_page(struct address_space *mapping, 2889 pgoff_t index, 2890 int (*filler)(void *, struct page *), 2891 void *data) 2892 { 2893 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping)); 2894 } 2895 EXPORT_SYMBOL(read_cache_page); 2896 2897 /** 2898 * read_cache_page_gfp - read into page cache, using specified page allocation flags. 2899 * @mapping: the page's address_space 2900 * @index: the page index 2901 * @gfp: the page allocator flags to use if allocating 2902 * 2903 * This is the same as "read_mapping_page(mapping, index, NULL)", but with 2904 * any new page allocations done using the specified allocation flags. 2905 * 2906 * If the page does not get brought uptodate, return -EIO. 2907 */ 2908 struct page *read_cache_page_gfp(struct address_space *mapping, 2909 pgoff_t index, 2910 gfp_t gfp) 2911 { 2912 filler_t *filler = (filler_t *)mapping->a_ops->readpage; 2913 2914 return do_read_cache_page(mapping, index, filler, NULL, gfp); 2915 } 2916 EXPORT_SYMBOL(read_cache_page_gfp); 2917 2918 /* 2919 * Performs necessary checks before doing a write 2920 * 2921 * Can adjust writing position or amount of bytes to write. 2922 * Returns appropriate error code that caller should return or 2923 * zero in case that write should be allowed. 2924 */ 2925 inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from) 2926 { 2927 struct file *file = iocb->ki_filp; 2928 struct inode *inode = file->f_mapping->host; 2929 unsigned long limit = rlimit(RLIMIT_FSIZE); 2930 loff_t pos; 2931 2932 if (!iov_iter_count(from)) 2933 return 0; 2934 2935 /* FIXME: this is for backwards compatibility with 2.4 */ 2936 if (iocb->ki_flags & IOCB_APPEND) 2937 iocb->ki_pos = i_size_read(inode); 2938 2939 pos = iocb->ki_pos; 2940 2941 if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT)) 2942 return -EINVAL; 2943 2944 if (limit != RLIM_INFINITY) { 2945 if (iocb->ki_pos >= limit) { 2946 send_sig(SIGXFSZ, current, 0); 2947 return -EFBIG; 2948 } 2949 iov_iter_truncate(from, limit - (unsigned long)pos); 2950 } 2951 2952 /* 2953 * LFS rule 2954 */ 2955 if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS && 2956 !(file->f_flags & O_LARGEFILE))) { 2957 if (pos >= MAX_NON_LFS) 2958 return -EFBIG; 2959 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos); 2960 } 2961 2962 /* 2963 * Are we about to exceed the fs block limit ? 2964 * 2965 * If we have written data it becomes a short write. If we have 2966 * exceeded without writing data we send a signal and return EFBIG. 2967 * Linus frestrict idea will clean these up nicely.. 2968 */ 2969 if (unlikely(pos >= inode->i_sb->s_maxbytes)) 2970 return -EFBIG; 2971 2972 iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos); 2973 return iov_iter_count(from); 2974 } 2975 EXPORT_SYMBOL(generic_write_checks); 2976 2977 int pagecache_write_begin(struct file *file, struct address_space *mapping, 2978 loff_t pos, unsigned len, unsigned flags, 2979 struct page **pagep, void **fsdata) 2980 { 2981 const struct address_space_operations *aops = mapping->a_ops; 2982 2983 return aops->write_begin(file, mapping, pos, len, flags, 2984 pagep, fsdata); 2985 } 2986 EXPORT_SYMBOL(pagecache_write_begin); 2987 2988 int pagecache_write_end(struct file *file, struct address_space *mapping, 2989 loff_t pos, unsigned len, unsigned copied, 2990 struct page *page, void *fsdata) 2991 { 2992 const struct address_space_operations *aops = mapping->a_ops; 2993 2994 return aops->write_end(file, mapping, pos, len, copied, page, fsdata); 2995 } 2996 EXPORT_SYMBOL(pagecache_write_end); 2997 2998 ssize_t 2999 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from) 3000 { 3001 struct file *file = iocb->ki_filp; 3002 struct address_space *mapping = file->f_mapping; 3003 struct inode *inode = mapping->host; 3004 loff_t pos = iocb->ki_pos; 3005 ssize_t written; 3006 size_t write_len; 3007 pgoff_t end; 3008 3009 write_len = iov_iter_count(from); 3010 end = (pos + write_len - 1) >> PAGE_SHIFT; 3011 3012 if (iocb->ki_flags & IOCB_NOWAIT) { 3013 /* If there are pages to writeback, return */ 3014 if (filemap_range_has_page(inode->i_mapping, pos, 3015 pos + iov_iter_count(from))) 3016 return -EAGAIN; 3017 } else { 3018 written = filemap_write_and_wait_range(mapping, pos, 3019 pos + write_len - 1); 3020 if (written) 3021 goto out; 3022 } 3023 3024 /* 3025 * After a write we want buffered reads to be sure to go to disk to get 3026 * the new data. We invalidate clean cached page from the region we're 3027 * about to write. We do this *before* the write so that we can return 3028 * without clobbering -EIOCBQUEUED from ->direct_IO(). 3029 */ 3030 written = invalidate_inode_pages2_range(mapping, 3031 pos >> PAGE_SHIFT, end); 3032 /* 3033 * If a page can not be invalidated, return 0 to fall back 3034 * to buffered write. 3035 */ 3036 if (written) { 3037 if (written == -EBUSY) 3038 return 0; 3039 goto out; 3040 } 3041 3042 written = mapping->a_ops->direct_IO(iocb, from); 3043 3044 /* 3045 * Finally, try again to invalidate clean pages which might have been 3046 * cached by non-direct readahead, or faulted in by get_user_pages() 3047 * if the source of the write was an mmap'ed region of the file 3048 * we're writing. Either one is a pretty crazy thing to do, 3049 * so we don't support it 100%. If this invalidation 3050 * fails, tough, the write still worked... 3051 * 3052 * Most of the time we do not need this since dio_complete() will do 3053 * the invalidation for us. However there are some file systems that 3054 * do not end up with dio_complete() being called, so let's not break 3055 * them by removing it completely 3056 */ 3057 if (mapping->nrpages) 3058 invalidate_inode_pages2_range(mapping, 3059 pos >> PAGE_SHIFT, end); 3060 3061 if (written > 0) { 3062 pos += written; 3063 write_len -= written; 3064 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { 3065 i_size_write(inode, pos); 3066 mark_inode_dirty(inode); 3067 } 3068 iocb->ki_pos = pos; 3069 } 3070 iov_iter_revert(from, write_len - iov_iter_count(from)); 3071 out: 3072 return written; 3073 } 3074 EXPORT_SYMBOL(generic_file_direct_write); 3075 3076 /* 3077 * Find or create a page at the given pagecache position. Return the locked 3078 * page. This function is specifically for buffered writes. 3079 */ 3080 struct page *grab_cache_page_write_begin(struct address_space *mapping, 3081 pgoff_t index, unsigned flags) 3082 { 3083 struct page *page; 3084 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT; 3085 3086 if (flags & AOP_FLAG_NOFS) 3087 fgp_flags |= FGP_NOFS; 3088 3089 page = pagecache_get_page(mapping, index, fgp_flags, 3090 mapping_gfp_mask(mapping)); 3091 if (page) 3092 wait_for_stable_page(page); 3093 3094 return page; 3095 } 3096 EXPORT_SYMBOL(grab_cache_page_write_begin); 3097 3098 ssize_t generic_perform_write(struct file *file, 3099 struct iov_iter *i, loff_t pos) 3100 { 3101 struct address_space *mapping = file->f_mapping; 3102 const struct address_space_operations *a_ops = mapping->a_ops; 3103 long status = 0; 3104 ssize_t written = 0; 3105 unsigned int flags = 0; 3106 3107 do { 3108 struct page *page; 3109 unsigned long offset; /* Offset into pagecache page */ 3110 unsigned long bytes; /* Bytes to write to page */ 3111 size_t copied; /* Bytes copied from user */ 3112 void *fsdata; 3113 3114 offset = (pos & (PAGE_SIZE - 1)); 3115 bytes = min_t(unsigned long, PAGE_SIZE - offset, 3116 iov_iter_count(i)); 3117 3118 again: 3119 /* 3120 * Bring in the user page that we will copy from _first_. 3121 * Otherwise there's a nasty deadlock on copying from the 3122 * same page as we're writing to, without it being marked 3123 * up-to-date. 3124 * 3125 * Not only is this an optimisation, but it is also required 3126 * to check that the address is actually valid, when atomic 3127 * usercopies are used, below. 3128 */ 3129 if (unlikely(iov_iter_fault_in_readable(i, bytes))) { 3130 status = -EFAULT; 3131 break; 3132 } 3133 3134 if (fatal_signal_pending(current)) { 3135 status = -EINTR; 3136 break; 3137 } 3138 3139 status = a_ops->write_begin(file, mapping, pos, bytes, flags, 3140 &page, &fsdata); 3141 if (unlikely(status < 0)) 3142 break; 3143 3144 if (mapping_writably_mapped(mapping)) 3145 flush_dcache_page(page); 3146 3147 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); 3148 flush_dcache_page(page); 3149 3150 status = a_ops->write_end(file, mapping, pos, bytes, copied, 3151 page, fsdata); 3152 if (unlikely(status < 0)) 3153 break; 3154 copied = status; 3155 3156 cond_resched(); 3157 3158 iov_iter_advance(i, copied); 3159 if (unlikely(copied == 0)) { 3160 /* 3161 * If we were unable to copy any data at all, we must 3162 * fall back to a single segment length write. 3163 * 3164 * If we didn't fallback here, we could livelock 3165 * because not all segments in the iov can be copied at 3166 * once without a pagefault. 3167 */ 3168 bytes = min_t(unsigned long, PAGE_SIZE - offset, 3169 iov_iter_single_seg_count(i)); 3170 goto again; 3171 } 3172 pos += copied; 3173 written += copied; 3174 3175 balance_dirty_pages_ratelimited(mapping); 3176 } while (iov_iter_count(i)); 3177 3178 return written ? written : status; 3179 } 3180 EXPORT_SYMBOL(generic_perform_write); 3181 3182 /** 3183 * __generic_file_write_iter - write data to a file 3184 * @iocb: IO state structure (file, offset, etc.) 3185 * @from: iov_iter with data to write 3186 * 3187 * This function does all the work needed for actually writing data to a 3188 * file. It does all basic checks, removes SUID from the file, updates 3189 * modification times and calls proper subroutines depending on whether we 3190 * do direct IO or a standard buffered write. 3191 * 3192 * It expects i_mutex to be grabbed unless we work on a block device or similar 3193 * object which does not need locking at all. 3194 * 3195 * This function does *not* take care of syncing data in case of O_SYNC write. 3196 * A caller has to handle it. This is mainly due to the fact that we want to 3197 * avoid syncing under i_mutex. 3198 */ 3199 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) 3200 { 3201 struct file *file = iocb->ki_filp; 3202 struct address_space * mapping = file->f_mapping; 3203 struct inode *inode = mapping->host; 3204 ssize_t written = 0; 3205 ssize_t err; 3206 ssize_t status; 3207 3208 /* We can write back this queue in page reclaim */ 3209 current->backing_dev_info = inode_to_bdi(inode); 3210 err = file_remove_privs(file); 3211 if (err) 3212 goto out; 3213 3214 err = file_update_time(file); 3215 if (err) 3216 goto out; 3217 3218 if (iocb->ki_flags & IOCB_DIRECT) { 3219 loff_t pos, endbyte; 3220 3221 written = generic_file_direct_write(iocb, from); 3222 /* 3223 * If the write stopped short of completing, fall back to 3224 * buffered writes. Some filesystems do this for writes to 3225 * holes, for example. For DAX files, a buffered write will 3226 * not succeed (even if it did, DAX does not handle dirty 3227 * page-cache pages correctly). 3228 */ 3229 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode)) 3230 goto out; 3231 3232 status = generic_perform_write(file, from, pos = iocb->ki_pos); 3233 /* 3234 * If generic_perform_write() returned a synchronous error 3235 * then we want to return the number of bytes which were 3236 * direct-written, or the error code if that was zero. Note 3237 * that this differs from normal direct-io semantics, which 3238 * will return -EFOO even if some bytes were written. 3239 */ 3240 if (unlikely(status < 0)) { 3241 err = status; 3242 goto out; 3243 } 3244 /* 3245 * We need to ensure that the page cache pages are written to 3246 * disk and invalidated to preserve the expected O_DIRECT 3247 * semantics. 3248 */ 3249 endbyte = pos + status - 1; 3250 err = filemap_write_and_wait_range(mapping, pos, endbyte); 3251 if (err == 0) { 3252 iocb->ki_pos = endbyte + 1; 3253 written += status; 3254 invalidate_mapping_pages(mapping, 3255 pos >> PAGE_SHIFT, 3256 endbyte >> PAGE_SHIFT); 3257 } else { 3258 /* 3259 * We don't know how much we wrote, so just return 3260 * the number of bytes which were direct-written 3261 */ 3262 } 3263 } else { 3264 written = generic_perform_write(file, from, iocb->ki_pos); 3265 if (likely(written > 0)) 3266 iocb->ki_pos += written; 3267 } 3268 out: 3269 current->backing_dev_info = NULL; 3270 return written ? written : err; 3271 } 3272 EXPORT_SYMBOL(__generic_file_write_iter); 3273 3274 /** 3275 * generic_file_write_iter - write data to a file 3276 * @iocb: IO state structure 3277 * @from: iov_iter with data to write 3278 * 3279 * This is a wrapper around __generic_file_write_iter() to be used by most 3280 * filesystems. It takes care of syncing the file in case of O_SYNC file 3281 * and acquires i_mutex as needed. 3282 */ 3283 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) 3284 { 3285 struct file *file = iocb->ki_filp; 3286 struct inode *inode = file->f_mapping->host; 3287 ssize_t ret; 3288 3289 inode_lock(inode); 3290 ret = generic_write_checks(iocb, from); 3291 if (ret > 0) 3292 ret = __generic_file_write_iter(iocb, from); 3293 inode_unlock(inode); 3294 3295 if (ret > 0) 3296 ret = generic_write_sync(iocb, ret); 3297 return ret; 3298 } 3299 EXPORT_SYMBOL(generic_file_write_iter); 3300 3301 /** 3302 * try_to_release_page() - release old fs-specific metadata on a page 3303 * 3304 * @page: the page which the kernel is trying to free 3305 * @gfp_mask: memory allocation flags (and I/O mode) 3306 * 3307 * The address_space is to try to release any data against the page 3308 * (presumably at page->private). If the release was successful, return '1'. 3309 * Otherwise return zero. 3310 * 3311 * This may also be called if PG_fscache is set on a page, indicating that the 3312 * page is known to the local caching routines. 3313 * 3314 * The @gfp_mask argument specifies whether I/O may be performed to release 3315 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS). 3316 * 3317 */ 3318 int try_to_release_page(struct page *page, gfp_t gfp_mask) 3319 { 3320 struct address_space * const mapping = page->mapping; 3321 3322 BUG_ON(!PageLocked(page)); 3323 if (PageWriteback(page)) 3324 return 0; 3325 3326 if (mapping && mapping->a_ops->releasepage) 3327 return mapping->a_ops->releasepage(page, gfp_mask); 3328 return try_to_free_buffers(page); 3329 } 3330 3331 EXPORT_SYMBOL(try_to_release_page); 3332