| 1 | |
| 2 | |
| 3 | * linux/mm/filemap.c
|
| 4 | *
|
| 5 | * Copyright (C) 1994-1999 Linus Torvalds
|
| 6 |
|
| 7 |
|
| 8 | |
| 9 | * This file handles the generic file mmap semantics used by
|
| 10 | * most "normal" filesystems (but you don't /have/ to use this:
|
| 11 | * the NFS filesystem used to do this differently, for example)
|
| 12 |
|
| 13 | #include <linux/export.h>
|
| 14 | #include <linux/compiler.h>
|
| 15 | #include <linux/dax.h>
|
| 16 | #include <linux/fs.h>
|
| 17 | #include <linux/sched/signal.h>
|
| 18 | #include <linux/uaccess.h>
|
| 19 | #include <linux/capability.h>
|
| 20 | #include <linux/kernel_stat.h>
|
| 21 | #include <linux/gfp.h>
|
| 22 | #include <linux/mm.h>
|
| 23 | #include <linux/swap.h>
|
| 24 | #include <linux/leafops.h>
|
| 25 | #include <linux/syscalls.h>
|
| 26 | #include <linux/mman.h>
|
| 27 | #include <linux/pagemap.h>
|
| 28 | #include <linux/file.h>
|
| 29 | #include <linux/uio.h>
|
| 30 | #include <linux/error-injection.h>
|
| 31 | #include <linux/hash.h>
|
| 32 | #include <linux/writeback.h>
|
| 33 | #include <linux/backing-dev.h>
|
| 34 | #include <linux/folio_batch.h>
|
| 35 | #include <linux/security.h>
|
| 36 | #include <linux/cpuset.h>
|
| 37 | #include <linux/hugetlb.h>
|
| 38 | #include <linux/memcontrol.h>
|
| 39 | #include <linux/shmem_fs.h>
|
| 40 | #include <linux/rmap.h>
|
| 41 | #include <linux/delayacct.h>
|
| 42 | #include <linux/psi.h>
|
| 43 | #include <linux/ramfs.h>
|
| 44 | #include <linux/page_idle.h>
|
| 45 | #include <linux/migrate.h>
|
| 46 | #include <linux/pipe_fs_i.h>
|
| 47 | #include <linux/splice.h>
|
| 48 | #include <linux/rcupdate_wait.h>
|
| 49 | #include <linux/sched/mm.h>
|
| 50 | #include <linux/sysctl.h>
|
| 51 | #include <linux/pgalloc.h>
|
| 52 |
|
| 53 | #include <asm/tlbflush.h>
|
| 54 | #include "internal.h"
|
| 55 |
|
| 56 | #define CREATE_TRACE_POINTS
|
| 57 | #include <trace/events/filemap.h>
|
| 58 |
|
| 59 | |
| 60 | * FIXME: remove all knowledge of the buffer layer from the core VM
|
| 61 |
|
| 62 | #include <linux/buffer_head.h>
|
| 63 |
|
| 64 | #include <asm/mman.h>
|
| 65 |
|
| 66 | #include "swap.h"
|
| 67 |
|
| 68 | |
| 69 | * Shared mappings implemented 30.11.1994. It's not fully working yet,
|
| 70 | * though.
|
| 71 | *
|
| 72 | * Shared mappings now work. 15.8.1995 Bruno.
|
| 73 | *
|
| 74 | * finished 'unifying' the page and buffer cache and SMP-threaded the
|
| 75 | * page-cache, 21.05.1999, Ingo Molnar <[email protected]>
|
| 76 | *
|
| 77 | * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <[email protected]>
|
| 78 |
|
| 79 |
|
| 80 | |
| 81 | * Lock ordering:
|
| 82 | *
|
| 83 | * ->i_mmap_rwsem (truncate_pagecache)
|
| 84 | * ->private_lock (__free_pte->block_dirty_folio)
|
| 85 | * ->swap_lock (exclusive_swap_page, others)
|
| 86 | * ->i_pages lock
|
| 87 | *
|
| 88 | * ->i_rwsem
|
| 89 | * ->invalidate_lock (acquired by fs in truncate path)
|
| 90 | * ->i_mmap_rwsem (truncate->unmap_mapping_range)
|
| 91 | *
|
| 92 | * ->mmap_lock
|
| 93 | * ->i_mmap_rwsem
|
| 94 | * ->page_table_lock or pte_lock (various, mainly in memory.c)
|
| 95 | * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
|
| 96 | *
|
| 97 | * ->mmap_lock
|
| 98 | * ->invalidate_lock (filemap_fault)
|
| 99 | * ->lock_page (filemap_fault, access_process_vm)
|
| 100 | *
|
| 101 | * ->i_rwsem (generic_perform_write)
|
| 102 | * ->mmap_lock (fault_in_readable->do_page_fault)
|
| 103 | *
|
| 104 | * bdi->wb.list_lock
|
| 105 | * sb_lock (fs/fs-writeback.c)
|
| 106 | * ->i_pages lock (__sync_single_inode)
|
| 107 | *
|
| 108 | * ->i_mmap_rwsem
|
| 109 | * ->anon_vma.lock (vma_merge)
|
| 110 | *
|
| 111 | * ->anon_vma.lock
|
| 112 | * ->page_table_lock or pte_lock (anon_vma_prepare and various)
|
| 113 | *
|
| 114 | * ->page_table_lock or pte_lock
|
| 115 | * ->swap_lock (try_to_unmap_one)
|
| 116 | * ->private_lock (try_to_unmap_one)
|
| 117 | * ->i_pages lock (try_to_unmap_one)
|
| 118 | * ->lruvec->lru_lock (follow_page_mask->mark_page_accessed)
|
| 119 | * ->lruvec->lru_lock (check_pte_range->folio_isolate_lru)
|
| 120 | * ->private_lock (folio_remove_rmap_pte->set_page_dirty)
|
| 121 | * ->i_pages lock (folio_remove_rmap_pte->set_page_dirty)
|
| 122 | * bdi.wb->list_lock (folio_remove_rmap_pte->set_page_dirty)
|
| 123 | * ->inode->i_lock (folio_remove_rmap_pte->set_page_dirty)
|
| 124 | * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
|
| 125 | * ->inode->i_lock (zap_pte_range->set_page_dirty)
|
| 126 | * ->private_lock (zap_pte_range->block_dirty_folio)
|
| 127 |
|
| 128 |
|
| 129 | static void page_cache_delete(struct address_space *mapping,
|
| 130 | struct folio *folio, void *shadow)
|
| 131 | {
|
| 132 | XA_STATE(xas, &mapping->i_pages, folio->index);
|
| 133 | long nr = 1;
|
| 134 |
|
| 135 | mapping_set_update(&xas, mapping);
|
| 136 |
|
| 137 | xas_set_order(&xas, folio->index, folio_order(folio));
|
| 138 | nr = folio_nr_pages(folio);
|
| 139 |
|
| 140 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
|
| 141 |
|
| 142 | xas_store(&xas, shadow);
|
| 143 | xas_init_marks(&xas);
|
| 144 |
|
| 145 | folio->mapping = NULL;
|
| 146 |
|
| 147 | mapping->nrpages -= nr;
|
| 148 | }
|
| 149 |
|
| 150 | static void filemap_unaccount_folio(struct address_space *mapping,
|
| 151 | struct folio *folio)
|
| 152 | {
|
| 153 | long nr;
|
| 154 |
|
| 155 | VM_BUG_ON_FOLIO(folio_mapped(folio), folio);
|
| 156 | if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(folio_mapped(folio))) {
|
| 157 | pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
|
| 158 | current->comm, folio_pfn(folio));
|
| 159 | dump_page(&folio->page, "still mapped when deleted");
|
| 160 | dump_stack();
|
| 161 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
|
| 162 |
|
| 163 | if (mapping_exiting(mapping) && !folio_test_large(folio)) {
|
| 164 | int mapcount = folio_mapcount(folio);
|
| 165 |
|
| 166 | if (folio_ref_count(folio) >= mapcount + 2) {
|
| 167 | |
| 168 | * All vmas have already been torn down, so it's
|
| 169 | * a good bet that actually the page is unmapped
|
| 170 | * and we'd rather not leak it: if we're wrong,
|
| 171 | * another bad page check should catch it later.
|
| 172 |
|
| 173 | atomic_set(&folio->_mapcount, -1);
|
| 174 | folio_ref_sub(folio, mapcount);
|
| 175 | }
|
| 176 | }
|
| 177 | }
|
| 178 |
|
| 179 |
|
| 180 | if (folio_test_hugetlb(folio))
|
| 181 | return;
|
| 182 |
|
| 183 | nr = folio_nr_pages(folio);
|
| 184 |
|
| 185 | lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr);
|
| 186 | if (folio_test_swapbacked(folio)) {
|
| 187 | lruvec_stat_mod_folio(folio, NR_SHMEM, -nr);
|
| 188 | if (folio_test_pmd_mappable(folio))
|
| 189 | lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, -nr);
|
| 190 | } else if (folio_test_pmd_mappable(folio)) {
|
| 191 | lruvec_stat_mod_folio(folio, NR_FILE_THPS, -nr);
|
| 192 | }
|
| 193 | if (test_bit(AS_KERNEL_FILE, &folio->mapping->flags))
|
| 194 | mod_node_page_state(folio_pgdat(folio),
|
| 195 | NR_KERNEL_FILE_PAGES, -nr);
|
| 196 |
|
| 197 | |
| 198 | * At this point folio must be either written or cleaned by
|
| 199 | * truncate. Dirty folio here signals a bug and loss of
|
| 200 | * unwritten data - on ordinary filesystems.
|
| 201 | *
|
| 202 | * But it's harmless on in-memory filesystems like tmpfs; and can
|
| 203 | * occur when a driver which did get_user_pages() sets page dirty
|
| 204 | * before putting it, while the inode is being finally evicted.
|
| 205 | *
|
| 206 | * Below fixes dirty accounting after removing the folio entirely
|
| 207 | * but leaves the dirty flag set: it has no effect for truncated
|
| 208 | * folio and anyway will be cleared before returning folio to
|
| 209 | * buddy allocator.
|
| 210 |
|
| 211 | if (WARN_ON_ONCE(folio_test_dirty(folio) &&
|
| 212 | mapping_can_writeback(mapping)))
|
| 213 | folio_account_cleaned(folio, inode_to_wb(mapping->host));
|
| 214 | }
|
| 215 |
|
| 216 | |
| 217 | * Delete a page from the page cache and free it. Caller has to make
|
| 218 | * sure the page is locked and that nobody else uses it - or that usage
|
| 219 | * is safe. The caller must hold the i_pages lock.
|
| 220 |
|
| 221 | void __filemap_remove_folio(struct folio *folio, void *shadow)
|
| 222 | {
|
| 223 | struct address_space *mapping = folio->mapping;
|
| 224 |
|
| 225 | trace_mm_filemap_delete_from_page_cache(folio);
|
| 226 | filemap_unaccount_folio(mapping, folio);
|
| 227 | page_cache_delete(mapping, folio, shadow);
|
| 228 | }
|
| 229 |
|
| 230 | static void filemap_free_folio(const struct address_space *mapping,
|
| 231 | struct folio *folio)
|
| 232 | {
|
| 233 | void (*free_folio)(struct folio *);
|
| 234 |
|
| 235 | free_folio = mapping->a_ops->free_folio;
|
| 236 | if (free_folio)
|
| 237 | free_folio(folio);
|
| 238 |
|
| 239 | folio_put_refs(folio, folio_nr_pages(folio));
|
| 240 | }
|
| 241 |
|
| 242 | |
| 243 | * filemap_remove_folio - Remove folio from page cache.
|
| 244 | * @folio: The folio.
|
| 245 | *
|
| 246 | * This must be called only on folios that are locked and have been
|
| 247 | * verified to be in the page cache. It will never put the folio into
|
| 248 | * the free list because the caller has a reference on the page.
|
| 249 |
|
| 250 | void filemap_remove_folio(struct folio *folio)
|
| 251 | {
|
| 252 | struct address_space *mapping = folio->mapping;
|
| 253 |
|
| 254 | BUG_ON(!folio_test_locked(folio));
|
| 255 | spin_lock(&mapping->host->i_lock);
|
| 256 | xa_lock_irq(&mapping->i_pages);
|
| 257 | __filemap_remove_folio(folio, NULL);
|
| 258 | xa_unlock_irq(&mapping->i_pages);
|
| 259 | if (mapping_shrinkable(mapping))
|
| 260 | inode_lru_list_add(mapping->host);
|
| 261 | spin_unlock(&mapping->host->i_lock);
|
| 262 |
|
| 263 | filemap_free_folio(mapping, folio);
|
| 264 | }
|
| 265 |
|
| 266 | |
| 267 | * page_cache_delete_batch - delete several folios from page cache
|
| 268 | * @mapping: the mapping to which folios belong
|
| 269 | * @fbatch: batch of folios to delete
|
| 270 | *
|
| 271 | * The function walks over mapping->i_pages and removes folios passed in
|
| 272 | * @fbatch from the mapping. The function expects @fbatch to be sorted
|
| 273 | * by page index and is optimised for it to be dense.
|
| 274 | * It tolerates holes in @fbatch (mapping entries at those indices are not
|
| 275 | * modified).
|
| 276 | *
|
| 277 | * The function expects the i_pages lock to be held.
|
| 278 |
|
| 279 | static void page_cache_delete_batch(struct address_space *mapping,
|
| 280 | struct folio_batch *fbatch)
|
| 281 | {
|
| 282 | XA_STATE(xas, &mapping->i_pages, fbatch->folios[0]->index);
|
| 283 | long total_pages = 0;
|
| 284 | int i = 0;
|
| 285 | struct folio *folio;
|
| 286 |
|
| 287 | mapping_set_update(&xas, mapping);
|
| 288 | xas_for_each(&xas, folio, ULONG_MAX) {
|
| 289 | if (i >= folio_batch_count(fbatch))
|
| 290 | break;
|
| 291 |
|
| 292 |
|
| 293 | if (xa_is_value(folio))
|
| 294 | continue;
|
| 295 | |
| 296 | * A page got inserted in our range? Skip it. We have our
|
| 297 | * pages locked so they are protected from being removed.
|
| 298 | * If we see a page whose index is higher than ours, it
|
| 299 | * means our page has been removed, which shouldn't be
|
| 300 | * possible because we're holding the PageLock.
|
| 301 |
|
| 302 | if (folio != fbatch->folios[i]) {
|
| 303 | VM_BUG_ON_FOLIO(folio->index >
|
| 304 | fbatch->folios[i]->index, folio);
|
| 305 | continue;
|
| 306 | }
|
| 307 |
|
| 308 | WARN_ON_ONCE(!folio_test_locked(folio));
|
| 309 |
|
| 310 | folio->mapping = NULL;
|
| 311 |
|
| 312 |
|
| 313 | i++;
|
| 314 | xas_store(&xas, NULL);
|
| 315 | total_pages += folio_nr_pages(folio);
|
| 316 | }
|
| 317 | mapping->nrpages -= total_pages;
|
| 318 | }
|
| 319 |
|
| 320 | void delete_from_page_cache_batch(struct address_space *mapping,
|
| 321 | struct folio_batch *fbatch)
|
| 322 | {
|
| 323 | int i;
|
| 324 |
|
| 325 | if (!folio_batch_count(fbatch))
|
| 326 | return;
|
| 327 |
|
| 328 | spin_lock(&mapping->host->i_lock);
|
| 329 | xa_lock_irq(&mapping->i_pages);
|
| 330 | for (i = 0; i < folio_batch_count(fbatch); i++) {
|
| 331 | struct folio *folio = fbatch->folios[i];
|
| 332 |
|
| 333 | trace_mm_filemap_delete_from_page_cache(folio);
|
| 334 | filemap_unaccount_folio(mapping, folio);
|
| 335 | }
|
| 336 | page_cache_delete_batch(mapping, fbatch);
|
| 337 | xa_unlock_irq(&mapping->i_pages);
|
| 338 | if (mapping_shrinkable(mapping))
|
| 339 | inode_lru_list_add(mapping->host);
|
| 340 | spin_unlock(&mapping->host->i_lock);
|
| 341 |
|
| 342 | for (i = 0; i < folio_batch_count(fbatch); i++)
|
| 343 | filemap_free_folio(mapping, fbatch->folios[i]);
|
| 344 | }
|
| 345 |
|
| 346 | int filemap_check_errors(struct address_space *mapping)
|
| 347 | {
|
| 348 | int ret = 0;
|
| 349 |
|
| 350 | if (test_bit(AS_ENOSPC, &mapping->flags) &&
|
| 351 | test_and_clear_bit(AS_ENOSPC, &mapping->flags))
|
| 352 | ret = -ENOSPC;
|
| 353 | if (test_bit(AS_EIO, &mapping->flags) &&
|
| 354 | test_and_clear_bit(AS_EIO, &mapping->flags))
|
| 355 | ret = -EIO;
|
| 356 | return ret;
|
| 357 | }
|
| 358 | EXPORT_SYMBOL(filemap_check_errors);
|
| 359 |
|
| 360 | static int filemap_check_and_keep_errors(struct address_space *mapping)
|
| 361 | {
|
| 362 |
|
| 363 | if (test_bit(AS_EIO, &mapping->flags))
|
| 364 | return -EIO;
|
| 365 | if (test_bit(AS_ENOSPC, &mapping->flags))
|
| 366 | return -ENOSPC;
|
| 367 | return 0;
|
| 368 | }
|
| 369 |
|
| 370 | static int filemap_writeback(struct address_space *mapping, loff_t start,
|
| 371 | loff_t end, enum writeback_sync_modes sync_mode,
|
| 372 | long *nr_to_write)
|
| 373 | {
|
| 374 | struct writeback_control wbc = {
|
| 375 | .sync_mode = sync_mode,
|
| 376 | .nr_to_write = nr_to_write ? *nr_to_write : LONG_MAX,
|
| 377 | .range_start = start,
|
| 378 | .range_end = end,
|
| 379 | };
|
| 380 | int ret;
|
| 381 |
|
| 382 | if (!mapping_can_writeback(mapping) ||
|
| 383 | !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
|
| 384 | return 0;
|
| 385 |
|
| 386 | wbc_attach_fdatawrite_inode(&wbc, mapping->host);
|
| 387 | ret = do_writepages(mapping, &wbc);
|
| 388 | wbc_detach_inode(&wbc);
|
| 389 |
|
| 390 | if (!ret && nr_to_write)
|
| 391 | *nr_to_write = wbc.nr_to_write;
|
| 392 | return ret;
|
| 393 | }
|
| 394 |
|
| 395 | |
| 396 | * filemap_fdatawrite_range - start writeback on mapping dirty pages in range
|
| 397 | * @mapping: address space structure to write
|
| 398 | * @start: offset in bytes where the range starts
|
| 399 | * @end: offset in bytes where the range ends (inclusive)
|
| 400 | *
|
| 401 | * Start writeback against all of a mapping's dirty pages that lie
|
| 402 | * within the byte offsets <start, end> inclusive.
|
| 403 | *
|
| 404 | * This is a data integrity operation that waits upon dirty or in writeback
|
| 405 | * pages.
|
| 406 | *
|
| 407 | * Return: %0 on success, negative error code otherwise.
|
| 408 |
|
| 409 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
|
| 410 | loff_t end)
|
| 411 | {
|
| 412 | return filemap_writeback(mapping, start, end, WB_SYNC_ALL, NULL);
|
| 413 | }
|
| 414 | EXPORT_SYMBOL(filemap_fdatawrite_range);
|
| 415 |
|
| 416 | int filemap_fdatawrite(struct address_space *mapping)
|
| 417 | {
|
| 418 | return filemap_fdatawrite_range(mapping, 0, LLONG_MAX);
|
| 419 | }
|
| 420 | EXPORT_SYMBOL(filemap_fdatawrite);
|
| 421 |
|
| 422 | |
| 423 | * filemap_flush_range - start writeback on a range
|
| 424 | * @mapping: target address_space
|
| 425 | * @start: index to start writeback on
|
| 426 | * @end: last (inclusive) index for writeback
|
| 427 | *
|
| 428 | * This is a non-integrity writeback helper, to start writing back folios
|
| 429 | * for the indicated range.
|
| 430 | *
|
| 431 | * Return: %0 on success, negative error code otherwise.
|
| 432 |
|
| 433 | int filemap_flush_range(struct address_space *mapping, loff_t start,
|
| 434 | loff_t end)
|
| 435 | {
|
| 436 | return filemap_writeback(mapping, start, end, WB_SYNC_NONE, NULL);
|
| 437 | }
|
| 438 | EXPORT_SYMBOL_GPL(filemap_flush_range);
|
| 439 |
|
| 440 | |
| 441 | * filemap_flush - mostly a non-blocking flush
|
| 442 | * @mapping: target address_space
|
| 443 | *
|
| 444 | * This is a mostly non-blocking flush. Not suitable for data-integrity
|
| 445 | * purposes - I/O may not be started against all dirty pages.
|
| 446 | *
|
| 447 | * Return: %0 on success, negative error code otherwise.
|
| 448 |
|
| 449 | int filemap_flush(struct address_space *mapping)
|
| 450 | {
|
| 451 | return filemap_flush_range(mapping, 0, LLONG_MAX);
|
| 452 | }
|
| 453 | EXPORT_SYMBOL(filemap_flush);
|
| 454 |
|
| 455 | |
| 456 | * Start writeback on @nr_to_write pages from @mapping. No one but the existing
|
| 457 | * btrfs caller should be using this. Talk to linux-mm if you think adding a
|
| 458 | * new caller is a good idea.
|
| 459 |
|
| 460 | int filemap_flush_nr(struct address_space *mapping, long *nr_to_write)
|
| 461 | {
|
| 462 | return filemap_writeback(mapping, 0, LLONG_MAX, WB_SYNC_NONE,
|
| 463 | nr_to_write);
|
| 464 | }
|
| 465 | EXPORT_SYMBOL_FOR_MODULES(filemap_flush_nr, "btrfs");
|
| 466 |
|
| 467 | |
| 468 | * filemap_range_has_page - check if a page exists in range.
|
| 469 | * @mapping: address space within which to check
|
| 470 | * @start_byte: offset in bytes where the range starts
|
| 471 | * @end_byte: offset in bytes where the range ends (inclusive)
|
| 472 | *
|
| 473 | * Find at least one page in the range supplied, usually used to check if
|
| 474 | * direct writing in this range will trigger a writeback.
|
| 475 | *
|
| 476 | * Return: %true if at least one page exists in the specified range,
|
| 477 | * %false otherwise.
|
| 478 |
|
| 479 | bool filemap_range_has_page(struct address_space *mapping,
|
| 480 | loff_t start_byte, loff_t end_byte)
|
| 481 | {
|
| 482 | struct folio *folio;
|
| 483 | XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
|
| 484 | pgoff_t max = end_byte >> PAGE_SHIFT;
|
| 485 |
|
| 486 | if (end_byte < start_byte)
|
| 487 | return false;
|
| 488 |
|
| 489 | rcu_read_lock();
|
| 490 | for (;;) {
|
| 491 | folio = xas_find(&xas, max);
|
| 492 | if (xas_retry(&xas, folio))
|
| 493 | continue;
|
| 494 |
|
| 495 | if (xa_is_value(folio))
|
| 496 | continue;
|
| 497 | |
| 498 | * We don't need to try to pin this page; we're about to
|
| 499 | * release the RCU lock anyway. It is enough to know that
|
| 500 | * there was a page here recently.
|
| 501 |
|
| 502 | break;
|
| 503 | }
|
| 504 | rcu_read_unlock();
|
| 505 |
|
| 506 | return folio != NULL;
|
| 507 | }
|
| 508 | EXPORT_SYMBOL(filemap_range_has_page);
|
| 509 |
|
| 510 | static void __filemap_fdatawait_range(struct address_space *mapping,
|
| 511 | loff_t start_byte, loff_t end_byte)
|
| 512 | {
|
| 513 | pgoff_t index = start_byte >> PAGE_SHIFT;
|
| 514 | pgoff_t end = end_byte >> PAGE_SHIFT;
|
| 515 | struct folio_batch fbatch;
|
| 516 | unsigned nr_folios;
|
| 517 |
|
| 518 | folio_batch_init(&fbatch);
|
| 519 |
|
| 520 | while (index <= end) {
|
| 521 | unsigned i;
|
| 522 |
|
| 523 | nr_folios = filemap_get_folios_tag(mapping, &index, end,
|
| 524 | PAGECACHE_TAG_WRITEBACK, &fbatch);
|
| 525 |
|
| 526 | if (!nr_folios)
|
| 527 | break;
|
| 528 |
|
| 529 | for (i = 0; i < nr_folios; i++) {
|
| 530 | struct folio *folio = fbatch.folios[i];
|
| 531 |
|
| 532 | folio_wait_writeback(folio);
|
| 533 | }
|
| 534 | folio_batch_release(&fbatch);
|
| 535 | cond_resched();
|
| 536 | }
|
| 537 | }
|
| 538 |
|
| 539 | |
| 540 | * filemap_fdatawait_range - wait for writeback to complete
|
| 541 | * @mapping: address space structure to wait for
|
| 542 | * @start_byte: offset in bytes where the range starts
|
| 543 | * @end_byte: offset in bytes where the range ends (inclusive)
|
| 544 | *
|
| 545 | * Walk the list of under-writeback pages of the given address space
|
| 546 | * in the given range and wait for all of them. Check error status of
|
| 547 | * the address space and return it.
|
| 548 | *
|
| 549 | * Since the error status of the address space is cleared by this function,
|
| 550 | * callers are responsible for checking the return value and handling and/or
|
| 551 | * reporting the error.
|
| 552 | *
|
| 553 | * Return: error status of the address space.
|
| 554 |
|
| 555 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
|
| 556 | loff_t end_byte)
|
| 557 | {
|
| 558 | __filemap_fdatawait_range(mapping, start_byte, end_byte);
|
| 559 | return filemap_check_errors(mapping);
|
| 560 | }
|
| 561 | EXPORT_SYMBOL(filemap_fdatawait_range);
|
| 562 |
|
| 563 | |
| 564 | * filemap_fdatawait_range_keep_errors - wait for writeback to complete
|
| 565 | * @mapping: address space structure to wait for
|
| 566 | * @start_byte: offset in bytes where the range starts
|
| 567 | * @end_byte: offset in bytes where the range ends (inclusive)
|
| 568 | *
|
| 569 | * Walk the list of under-writeback pages of the given address space in the
|
| 570 | * given range and wait for all of them. Unlike filemap_fdatawait_range(),
|
| 571 | * this function does not clear error status of the address space.
|
| 572 | *
|
| 573 | * Use this function if callers don't handle errors themselves. Expected
|
| 574 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
|
| 575 | * fsfreeze(8)
|
| 576 |
|
| 577 | int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
|
| 578 | loff_t start_byte, loff_t end_byte)
|
| 579 | {
|
| 580 | __filemap_fdatawait_range(mapping, start_byte, end_byte);
|
| 581 | return filemap_check_and_keep_errors(mapping);
|
| 582 | }
|
| 583 | EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
|
| 584 |
|
| 585 | |
| 586 | * file_fdatawait_range - wait for writeback to complete
|
| 587 | * @file: file pointing to address space structure to wait for
|
| 588 | * @start_byte: offset in bytes where the range starts
|
| 589 | * @end_byte: offset in bytes where the range ends (inclusive)
|
| 590 | *
|
| 591 | * Walk the list of under-writeback pages of the address space that file
|
| 592 | * refers to, in the given range and wait for all of them. Check error
|
| 593 | * status of the address space vs. the file->f_wb_err cursor and return it.
|
| 594 | *
|
| 595 | * Since the error status of the file is advanced by this function,
|
| 596 | * callers are responsible for checking the return value and handling and/or
|
| 597 | * reporting the error.
|
| 598 | *
|
| 599 | * Return: error status of the address space vs. the file->f_wb_err cursor.
|
| 600 |
|
| 601 | int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
|
| 602 | {
|
| 603 | struct address_space *mapping = file->f_mapping;
|
| 604 |
|
| 605 | __filemap_fdatawait_range(mapping, start_byte, end_byte);
|
| 606 | return file_check_and_advance_wb_err(file);
|
| 607 | }
|
| 608 | EXPORT_SYMBOL(file_fdatawait_range);
|
| 609 |
|
| 610 | |
| 611 | * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
|
| 612 | * @mapping: address space structure to wait for
|
| 613 | *
|
| 614 | * Walk the list of under-writeback pages of the given address space
|
| 615 | * and wait for all of them. Unlike filemap_fdatawait(), this function
|
| 616 | * does not clear error status of the address space.
|
| 617 | *
|
| 618 | * Use this function if callers don't handle errors themselves. Expected
|
| 619 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
|
| 620 | * fsfreeze(8)
|
| 621 | *
|
| 622 | * Return: error status of the address space.
|
| 623 |
|
| 624 | int filemap_fdatawait_keep_errors(struct address_space *mapping)
|
| 625 | {
|
| 626 | __filemap_fdatawait_range(mapping, 0, LLONG_MAX);
|
| 627 | return filemap_check_and_keep_errors(mapping);
|
| 628 | }
|
| 629 | EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
|
| 630 |
|
| 631 |
|
| 632 | static bool mapping_needs_writeback(struct address_space *mapping)
|
| 633 | {
|
| 634 | return mapping->nrpages;
|
| 635 | }
|
| 636 |
|
| 637 | bool filemap_range_has_writeback(struct address_space *mapping,
|
| 638 | loff_t start_byte, loff_t end_byte)
|
| 639 | {
|
| 640 | XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
|
| 641 | pgoff_t max = end_byte >> PAGE_SHIFT;
|
| 642 | struct folio *folio;
|
| 643 |
|
| 644 | if (end_byte < start_byte)
|
| 645 | return false;
|
| 646 |
|
| 647 | rcu_read_lock();
|
| 648 | xas_for_each(&xas, folio, max) {
|
| 649 | if (xas_retry(&xas, folio))
|
| 650 | continue;
|
| 651 | if (xa_is_value(folio))
|
| 652 | continue;
|
| 653 | if (folio_test_dirty(folio) || folio_test_locked(folio) ||
|
| 654 | folio_test_writeback(folio))
|
| 655 | break;
|
| 656 | }
|
| 657 | rcu_read_unlock();
|
| 658 | return folio != NULL;
|
| 659 | }
|
| 660 | EXPORT_SYMBOL_GPL(filemap_range_has_writeback);
|
| 661 |
|
| 662 | |
| 663 | * filemap_write_and_wait_range - write out & wait on a file range
|
| 664 | * @mapping: the address_space for the pages
|
| 665 | * @lstart: offset in bytes where the range starts
|
| 666 | * @lend: offset in bytes where the range ends (inclusive)
|
| 667 | *
|
| 668 | * Write out and wait upon file offsets lstart->lend, inclusive.
|
| 669 | *
|
| 670 | * Note that @lend is inclusive (describes the last byte to be written) so
|
| 671 | * that this function can be used to write to the very end-of-file (end = -1).
|
| 672 | *
|
| 673 | * Return: error status of the address space.
|
| 674 |
|
| 675 | int filemap_write_and_wait_range(struct address_space *mapping,
|
| 676 | loff_t lstart, loff_t lend)
|
| 677 | {
|
| 678 | int err = 0, err2;
|
| 679 |
|
| 680 | if (lend < lstart)
|
| 681 | return 0;
|
| 682 |
|
| 683 | if (mapping_needs_writeback(mapping)) {
|
| 684 | err = filemap_fdatawrite_range(mapping, lstart, lend);
|
| 685 | |
| 686 | * Even if the above returned error, the pages may be
|
| 687 | * written partially (e.g. -ENOSPC), so we wait for it.
|
| 688 | * But the -EIO is special case, it may indicate the worst
|
| 689 | * thing (e.g. bug) happened, so we avoid waiting for it.
|
| 690 |
|
| 691 | if (err != -EIO)
|
| 692 | __filemap_fdatawait_range(mapping, lstart, lend);
|
| 693 | }
|
| 694 | err2 = filemap_check_errors(mapping);
|
| 695 | if (!err)
|
| 696 | err = err2;
|
| 697 | return err;
|
| 698 | }
|
| 699 | EXPORT_SYMBOL(filemap_write_and_wait_range);
|
| 700 |
|
| 701 | void __filemap_set_wb_err(struct address_space *mapping, int err)
|
| 702 | {
|
| 703 | errseq_t eseq = errseq_set(&mapping->wb_err, err);
|
| 704 |
|
| 705 | trace_filemap_set_wb_err(mapping, eseq);
|
| 706 | }
|
| 707 | EXPORT_SYMBOL(__filemap_set_wb_err);
|
| 708 |
|
| 709 | |
| 710 | * file_check_and_advance_wb_err - report wb error (if any) that was previously
|
| 711 | * and advance wb_err to current one
|
| 712 | * @file: struct file on which the error is being reported
|
| 713 | *
|
| 714 | * When userland calls fsync (or something like nfsd does the equivalent), we
|
| 715 | * want to report any writeback errors that occurred since the last fsync (or
|
| 716 | * since the file was opened if there haven't been any).
|
| 717 | *
|
| 718 | * Grab the wb_err from the mapping. If it matches what we have in the file,
|
| 719 | * then just quickly return 0. The file is all caught up.
|
| 720 | *
|
| 721 | * If it doesn't match, then take the mapping value, set the "seen" flag in
|
| 722 | * it and try to swap it into place. If it works, or another task beat us
|
| 723 | * to it with the new value, then update the f_wb_err and return the error
|
| 724 | * portion. The error at this point must be reported via proper channels
|
| 725 | * (a'la fsync, or NFS COMMIT operation, etc.).
|
| 726 | *
|
| 727 | * While we handle mapping->wb_err with atomic operations, the f_wb_err
|
| 728 | * value is protected by the f_lock since we must ensure that it reflects
|
| 729 | * the latest value swapped in for this file descriptor.
|
| 730 | *
|
| 731 | * Return: %0 on success, negative error code otherwise.
|
| 732 |
|
| 733 | int file_check_and_advance_wb_err(struct file *file)
|
| 734 | {
|
| 735 | int err = 0;
|
| 736 | errseq_t old = READ_ONCE(file->f_wb_err);
|
| 737 | struct address_space *mapping = file->f_mapping;
|
| 738 |
|
| 739 |
|
| 740 | if (errseq_check(&mapping->wb_err, old)) {
|
| 741 |
|
| 742 | spin_lock(&file->f_lock);
|
| 743 | old = file->f_wb_err;
|
| 744 | err = errseq_check_and_advance(&mapping->wb_err,
|
| 745 | &file->f_wb_err);
|
| 746 | trace_file_check_and_advance_wb_err(file, old);
|
| 747 | spin_unlock(&file->f_lock);
|
| 748 | }
|
| 749 |
|
| 750 | |
| 751 | * We're mostly using this function as a drop in replacement for
|
| 752 | * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
|
| 753 | * that the legacy code would have had on these flags.
|
| 754 |
|
| 755 | clear_bit(AS_EIO, &mapping->flags);
|
| 756 | clear_bit(AS_ENOSPC, &mapping->flags);
|
| 757 | return err;
|
| 758 | }
|
| 759 | EXPORT_SYMBOL(file_check_and_advance_wb_err);
|
| 760 |
|
| 761 | |
| 762 | * file_write_and_wait_range - write out & wait on a file range
|
| 763 | * @file: file pointing to address_space with pages
|
| 764 | * @lstart: offset in bytes where the range starts
|
| 765 | * @lend: offset in bytes where the range ends (inclusive)
|
| 766 | *
|
| 767 | * Write out and wait upon file offsets lstart->lend, inclusive.
|
| 768 | *
|
| 769 | * Note that @lend is inclusive (describes the last byte to be written) so
|
| 770 | * that this function can be used to write to the very end-of-file (end = -1).
|
| 771 | *
|
| 772 | * After writing out and waiting on the data, we check and advance the
|
| 773 | * f_wb_err cursor to the latest value, and return any errors detected there.
|
| 774 | *
|
| 775 | * Return: %0 on success, negative error code otherwise.
|
| 776 |
|
| 777 | int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
|
| 778 | {
|
| 779 | int err = 0, err2;
|
| 780 | struct address_space *mapping = file->f_mapping;
|
| 781 |
|
| 782 | if (lend < lstart)
|
| 783 | return 0;
|
| 784 |
|
| 785 | if (mapping_needs_writeback(mapping)) {
|
| 786 | err = filemap_fdatawrite_range(mapping, lstart, lend);
|
| 787 |
|
| 788 | if (err != -EIO)
|
| 789 | __filemap_fdatawait_range(mapping, lstart, lend);
|
| 790 | }
|
| 791 | err2 = file_check_and_advance_wb_err(file);
|
| 792 | if (!err)
|
| 793 | err = err2;
|
| 794 | return err;
|
| 795 | }
|
| 796 | EXPORT_SYMBOL(file_write_and_wait_range);
|
| 797 |
|
| 798 | |
| 799 | * replace_page_cache_folio - replace a pagecache folio with a new one
|
| 800 | * @old: folio to be replaced
|
| 801 | * @new: folio to replace with
|
| 802 | *
|
| 803 | * This function replaces a folio in the pagecache with a new one. On
|
| 804 | * success it acquires the pagecache reference for the new folio and
|
| 805 | * drops it for the old folio. Both the old and new folios must be
|
| 806 | * locked. This function does not add the new folio to the LRU, the
|
| 807 | * caller must do that.
|
| 808 | *
|
| 809 | * The remove + add is atomic. This function cannot fail.
|
| 810 |
|
| 811 | void replace_page_cache_folio(struct folio *old, struct folio *new)
|
| 812 | {
|
| 813 | struct address_space *mapping = old->mapping;
|
| 814 | void (*free_folio)(struct folio *) = mapping->a_ops->free_folio;
|
| 815 | pgoff_t offset = old->index;
|
| 816 | XA_STATE(xas, &mapping->i_pages, offset);
|
| 817 |
|
| 818 | VM_BUG_ON_FOLIO(!folio_test_locked(old), old);
|
| 819 | VM_BUG_ON_FOLIO(!folio_test_locked(new), new);
|
| 820 | VM_BUG_ON_FOLIO(new->mapping, new);
|
| 821 |
|
| 822 | folio_get(new);
|
| 823 | new->mapping = mapping;
|
| 824 | new->index = offset;
|
| 825 |
|
| 826 | mem_cgroup_replace_folio(old, new);
|
| 827 |
|
| 828 | xas_lock_irq(&xas);
|
| 829 | xas_store(&xas, new);
|
| 830 |
|
| 831 | old->mapping = NULL;
|
| 832 |
|
| 833 | if (!folio_test_hugetlb(old))
|
| 834 | lruvec_stat_sub_folio(old, NR_FILE_PAGES);
|
| 835 | if (!folio_test_hugetlb(new))
|
| 836 | lruvec_stat_add_folio(new, NR_FILE_PAGES);
|
| 837 | if (folio_test_swapbacked(old))
|
| 838 | lruvec_stat_sub_folio(old, NR_SHMEM);
|
| 839 | if (folio_test_swapbacked(new))
|
| 840 | lruvec_stat_add_folio(new, NR_SHMEM);
|
| 841 | xas_unlock_irq(&xas);
|
| 842 | if (free_folio)
|
| 843 | free_folio(old);
|
| 844 | folio_put(old);
|
| 845 | }
|
| 846 | EXPORT_SYMBOL_GPL(replace_page_cache_folio);
|
| 847 |
|
| 848 | noinline int __filemap_add_folio(struct address_space *mapping,
|
| 849 | struct folio *folio, pgoff_t index, gfp_t gfp, void **shadowp)
|
| 850 | {
|
| 851 | XA_STATE_ORDER(xas, &mapping->i_pages, index, folio_order(folio));
|
| 852 | bool huge;
|
| 853 | long nr;
|
| 854 | unsigned int forder = folio_order(folio);
|
| 855 |
|
| 856 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
|
| 857 | VM_BUG_ON_FOLIO(folio_test_swapbacked(folio), folio);
|
| 858 | VM_BUG_ON_FOLIO(folio_order(folio) < mapping_min_folio_order(mapping),
|
| 859 | folio);
|
| 860 | mapping_set_update(&xas, mapping);
|
| 861 |
|
| 862 | VM_BUG_ON_FOLIO(index & (folio_nr_pages(folio) - 1), folio);
|
| 863 | huge = folio_test_hugetlb(folio);
|
| 864 | nr = folio_nr_pages(folio);
|
| 865 |
|
| 866 | gfp &= GFP_RECLAIM_MASK;
|
| 867 | folio_ref_add(folio, nr);
|
| 868 | folio->mapping = mapping;
|
| 869 | folio->index = xas.xa_index;
|
| 870 |
|
| 871 | for (;;) {
|
| 872 | int order = -1;
|
| 873 | void *entry, *old = NULL;
|
| 874 |
|
| 875 | xas_lock_irq(&xas);
|
| 876 | xas_for_each_conflict(&xas, entry) {
|
| 877 | old = entry;
|
| 878 | if (!xa_is_value(entry)) {
|
| 879 | xas_set_err(&xas, -EEXIST);
|
| 880 | goto unlock;
|
| 881 | }
|
| 882 | |
| 883 | * If a larger entry exists,
|
| 884 | * it will be the first and only entry iterated.
|
| 885 |
|
| 886 | if (order == -1)
|
| 887 | order = xas_get_order(&xas);
|
| 888 | }
|
| 889 |
|
| 890 | if (old) {
|
| 891 | if (order > 0 && order > forder) {
|
| 892 | unsigned int split_order = max(forder,
|
| 893 | xas_try_split_min_order(order));
|
| 894 |
|
| 895 |
|
| 896 | BUG_ON(shmem_mapping(mapping));
|
| 897 |
|
| 898 | while (order > forder) {
|
| 899 | xas_set_order(&xas, index, split_order);
|
| 900 | xas_try_split(&xas, old, order);
|
| 901 | if (xas_error(&xas))
|
| 902 | goto unlock;
|
| 903 | order = split_order;
|
| 904 | split_order =
|
| 905 | max(xas_try_split_min_order(
|
| 906 | split_order),
|
| 907 | forder);
|
| 908 | }
|
| 909 | xas_reset(&xas);
|
| 910 | }
|
| 911 | if (shadowp)
|
| 912 | *shadowp = old;
|
| 913 | }
|
| 914 |
|
| 915 | xas_store(&xas, folio);
|
| 916 | if (xas_error(&xas))
|
| 917 | goto unlock;
|
| 918 |
|
| 919 | mapping->nrpages += nr;
|
| 920 |
|
| 921 |
|
| 922 | if (!huge) {
|
| 923 | lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr);
|
| 924 | if (folio_test_pmd_mappable(folio))
|
| 925 | lruvec_stat_mod_folio(folio,
|
| 926 | NR_FILE_THPS, nr);
|
| 927 | }
|
| 928 |
|
| 929 | unlock:
|
| 930 | xas_unlock_irq(&xas);
|
| 931 |
|
| 932 | if (!xas_nomem(&xas, gfp))
|
| 933 | break;
|
| 934 |
|
| 935 | |
| 936 | * Lock has been dropped: start again with the original index
|
| 937 | * and order (but now with the memory reserved by xas_nomem()).
|
| 938 |
|
| 939 | xas_set_order(&xas, index, forder);
|
| 940 | }
|
| 941 |
|
| 942 | if (xas_error(&xas))
|
| 943 | goto error;
|
| 944 |
|
| 945 | trace_mm_filemap_add_to_page_cache(folio);
|
| 946 | return 0;
|
| 947 | error:
|
| 948 | folio->mapping = NULL;
|
| 949 |
|
| 950 | folio_put_refs(folio, nr);
|
| 951 | return xas_error(&xas);
|
| 952 | }
|
| 953 | ALLOW_ERROR_INJECTION(__filemap_add_folio, ERRNO);
|
| 954 |
|
| 955 | int filemap_add_folio(struct address_space *mapping, struct folio *folio,
|
| 956 | pgoff_t index, gfp_t gfp)
|
| 957 | {
|
| 958 | void *shadow = NULL;
|
| 959 | int ret;
|
| 960 | struct mem_cgroup *tmp;
|
| 961 | bool kernel_file = test_bit(AS_KERNEL_FILE, &mapping->flags);
|
| 962 |
|
| 963 | if (kernel_file)
|
| 964 | tmp = set_active_memcg(root_mem_cgroup);
|
| 965 | ret = mem_cgroup_charge(folio, NULL, gfp);
|
| 966 | if (kernel_file)
|
| 967 | set_active_memcg(tmp);
|
| 968 | if (ret)
|
| 969 | return ret;
|
| 970 |
|
| 971 | __folio_set_locked(folio);
|
| 972 | ret = __filemap_add_folio(mapping, folio, index, gfp, &shadow);
|
| 973 | if (unlikely(ret)) {
|
| 974 | mem_cgroup_uncharge(folio);
|
| 975 | __folio_clear_locked(folio);
|
| 976 | } else {
|
| 977 | |
| 978 | * The folio might have been evicted from cache only
|
| 979 | * recently, in which case it should be activated like
|
| 980 | * any other repeatedly accessed folio.
|
| 981 | * The exception is folios getting rewritten; evicting other
|
| 982 | * data from the working set, only to cache data that will
|
| 983 | * get overwritten with something else, is a waste of memory.
|
| 984 |
|
| 985 | WARN_ON_ONCE(folio_test_active(folio));
|
| 986 | if (!(gfp & __GFP_WRITE) && shadow)
|
| 987 | workingset_refault(folio, shadow);
|
| 988 | folio_add_lru(folio);
|
| 989 | if (kernel_file)
|
| 990 | mod_node_page_state(folio_pgdat(folio),
|
| 991 | NR_KERNEL_FILE_PAGES,
|
| 992 | folio_nr_pages(folio));
|
| 993 | }
|
| 994 | return ret;
|
| 995 | }
|
| 996 | EXPORT_SYMBOL_GPL(filemap_add_folio);
|
| 997 |
|
| 998 | #ifdef CONFIG_NUMA
|
| 999 | struct folio *filemap_alloc_folio_noprof(gfp_t gfp, unsigned int order,
|
| 1000 | struct mempolicy *policy)
|
| 1001 | {
|
| 1002 | int n;
|
| 1003 | struct folio *folio;
|
| 1004 |
|
| 1005 | if (policy)
|
| 1006 | return folio_alloc_mpol_noprof(gfp, order, policy,
|
| 1007 | NO_INTERLEAVE_INDEX, numa_node_id());
|
| 1008 |
|
| 1009 | if (cpuset_do_page_mem_spread()) {
|
| 1010 | unsigned int cpuset_mems_cookie;
|
| 1011 | do {
|
| 1012 | cpuset_mems_cookie = read_mems_allowed_begin();
|
| 1013 | n = cpuset_mem_spread_node();
|
| 1014 | folio = __folio_alloc_node_noprof(gfp, order, n);
|
| 1015 | } while (!folio && read_mems_allowed_retry(cpuset_mems_cookie));
|
| 1016 |
|
| 1017 | return folio;
|
| 1018 | }
|
| 1019 | return folio_alloc_noprof(gfp, order);
|
| 1020 | }
|
| 1021 | EXPORT_SYMBOL(filemap_alloc_folio_noprof);
|
| 1022 | #endif
|
| 1023 |
|
| 1024 | |
| 1025 | * filemap_invalidate_lock_two - lock invalidate_lock for two mappings
|
| 1026 | *
|
| 1027 | * Lock exclusively invalidate_lock of any passed mapping that is not NULL.
|
| 1028 | *
|
| 1029 | * @mapping1: the first mapping to lock
|
| 1030 | * @mapping2: the second mapping to lock
|
| 1031 |
|
| 1032 | void filemap_invalidate_lock_two(struct address_space *mapping1,
|
| 1033 | struct address_space *mapping2)
|
| 1034 | {
|
| 1035 | if (mapping1 > mapping2)
|
| 1036 | swap(mapping1, mapping2);
|
| 1037 | if (mapping1)
|
| 1038 | down_write(&mapping1->invalidate_lock);
|
| 1039 | if (mapping2 && mapping1 != mapping2)
|
| 1040 | down_write_nested(&mapping2->invalidate_lock, 1);
|
| 1041 | }
|
| 1042 | EXPORT_SYMBOL(filemap_invalidate_lock_two);
|
| 1043 |
|
| 1044 | |
| 1045 | * filemap_invalidate_unlock_two - unlock invalidate_lock for two mappings
|
| 1046 | *
|
| 1047 | * Unlock exclusive invalidate_lock of any passed mapping that is not NULL.
|
| 1048 | *
|
| 1049 | * @mapping1: the first mapping to unlock
|
| 1050 | * @mapping2: the second mapping to unlock
|
| 1051 |
|
| 1052 | void filemap_invalidate_unlock_two(struct address_space *mapping1,
|
| 1053 | struct address_space *mapping2)
|
| 1054 | {
|
| 1055 | if (mapping1)
|
| 1056 | up_write(&mapping1->invalidate_lock);
|
| 1057 | if (mapping2 && mapping1 != mapping2)
|
| 1058 | up_write(&mapping2->invalidate_lock);
|
| 1059 | }
|
| 1060 | EXPORT_SYMBOL(filemap_invalidate_unlock_two);
|
| 1061 |
|
| 1062 | |
| 1063 | * In order to wait for pages to become available there must be
|
| 1064 | * waitqueues associated with pages. By using a hash table of
|
| 1065 | * waitqueues where the bucket discipline is to maintain all
|
| 1066 | * waiters on the same queue and wake all when any of the pages
|
| 1067 | * become available, and for the woken contexts to check to be
|
| 1068 | * sure the appropriate page became available, this saves space
|
| 1069 | * at a cost of "thundering herd" phenomena during rare hash
|
| 1070 | * collisions.
|
| 1071 |
|
| 1072 | #define PAGE_WAIT_TABLE_BITS 8
|
| 1073 | #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
|
| 1074 | static wait_queue_head_t folio_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
|
| 1075 |
|
| 1076 | static wait_queue_head_t *folio_waitqueue(struct folio *folio)
|
| 1077 | {
|
| 1078 | return &folio_wait_table[hash_ptr(folio, PAGE_WAIT_TABLE_BITS)];
|
| 1079 | }
|
| 1080 |
|
| 1081 |
|
| 1082 | static int sysctl_page_lock_unfairness = 5;
|
| 1083 | static const struct ctl_table filemap_sysctl_table[] = {
|
| 1084 | {
|
| 1085 | .procname = "page_lock_unfairness",
|
| 1086 | .data = &sysctl_page_lock_unfairness,
|
| 1087 | .maxlen = sizeof(sysctl_page_lock_unfairness),
|
| 1088 | .mode = 0644,
|
| 1089 | .proc_handler = proc_dointvec_minmax,
|
| 1090 | .extra1 = SYSCTL_ZERO,
|
| 1091 | }
|
| 1092 | };
|
| 1093 |
|
| 1094 | void __init pagecache_init(void)
|
| 1095 | {
|
| 1096 | int i;
|
| 1097 |
|
| 1098 | for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
|
| 1099 | init_waitqueue_head(&folio_wait_table[i]);
|
| 1100 |
|
| 1101 | page_writeback_init();
|
| 1102 | register_sysctl_init("vm", filemap_sysctl_table);
|
| 1103 | }
|
| 1104 |
|
| 1105 | |
| 1106 | * The page wait code treats the "wait->flags" somewhat unusually, because
|
| 1107 | * we have multiple different kinds of waits, not just the usual "exclusive"
|
| 1108 | * one.
|
| 1109 | *
|
| 1110 | * We have:
|
| 1111 | *
|
| 1112 | * (a) no special bits set:
|
| 1113 | *
|
| 1114 | * We're just waiting for the bit to be released, and when a waker
|
| 1115 | * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
|
| 1116 | * and remove it from the wait queue.
|
| 1117 | *
|
| 1118 | * Simple and straightforward.
|
| 1119 | *
|
| 1120 | * (b) WQ_FLAG_EXCLUSIVE:
|
| 1121 | *
|
| 1122 | * The waiter is waiting to get the lock, and only one waiter should
|
| 1123 | * be woken up to avoid any thundering herd behavior. We'll set the
|
| 1124 | * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
|
| 1125 | *
|
| 1126 | * This is the traditional exclusive wait.
|
| 1127 | *
|
| 1128 | * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
|
| 1129 | *
|
| 1130 | * The waiter is waiting to get the bit, and additionally wants the
|
| 1131 | * lock to be transferred to it for fair lock behavior. If the lock
|
| 1132 | * cannot be taken, we stop walking the wait queue without waking
|
| 1133 | * the waiter.
|
| 1134 | *
|
| 1135 | * This is the "fair lock handoff" case, and in addition to setting
|
| 1136 | * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
|
| 1137 | * that it now has the lock.
|
| 1138 |
|
| 1139 | static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
|
| 1140 | {
|
| 1141 | unsigned int flags;
|
| 1142 | struct wait_page_key *key = arg;
|
| 1143 | struct wait_page_queue *wait_page
|
| 1144 | = container_of(wait, struct wait_page_queue, wait);
|
| 1145 |
|
| 1146 | if (!wake_page_match(wait_page, key))
|
| 1147 | return 0;
|
| 1148 |
|
| 1149 | |
| 1150 | * If it's a lock handoff wait, we get the bit for it, and
|
| 1151 | * stop walking (and do not wake it up) if we can't.
|
| 1152 |
|
| 1153 | flags = wait->flags;
|
| 1154 | if (flags & WQ_FLAG_EXCLUSIVE) {
|
| 1155 | if (test_bit(key->bit_nr, &key->folio->flags.f))
|
| 1156 | return -1;
|
| 1157 | if (flags & WQ_FLAG_CUSTOM) {
|
| 1158 | if (test_and_set_bit(key->bit_nr, &key->folio->flags.f))
|
| 1159 | return -1;
|
| 1160 | flags |= WQ_FLAG_DONE;
|
| 1161 | }
|
| 1162 | }
|
| 1163 |
|
| 1164 | |
| 1165 | * We are holding the wait-queue lock, but the waiter that
|
| 1166 | * is waiting for this will be checking the flags without
|
| 1167 | * any locking.
|
| 1168 | *
|
| 1169 | * So update the flags atomically, and wake up the waiter
|
| 1170 | * afterwards to avoid any races. This store-release pairs
|
| 1171 | * with the load-acquire in folio_wait_bit_common().
|
| 1172 |
|
| 1173 | smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN);
|
| 1174 | wake_up_state(wait->private, mode);
|
| 1175 |
|
| 1176 | |
| 1177 | * Ok, we have successfully done what we're waiting for,
|
| 1178 | * and we can unconditionally remove the wait entry.
|
| 1179 | *
|
| 1180 | * Note that this pairs with the "finish_wait()" in the
|
| 1181 | * waiter, and has to be the absolute last thing we do.
|
| 1182 | * After this list_del_init(&wait->entry) the wait entry
|
| 1183 | * might be de-allocated and the process might even have
|
| 1184 | * exited.
|
| 1185 |
|
| 1186 | list_del_init_careful(&wait->entry);
|
| 1187 | return (flags & WQ_FLAG_EXCLUSIVE) != 0;
|
| 1188 | }
|
| 1189 |
|
| 1190 | static void folio_wake_bit(struct folio *folio, int bit_nr)
|
| 1191 | {
|
| 1192 | wait_queue_head_t *q = folio_waitqueue(folio);
|
| 1193 | struct wait_page_key key;
|
| 1194 | unsigned long flags;
|
| 1195 |
|
| 1196 | key.folio = folio;
|
| 1197 | key.bit_nr = bit_nr;
|
| 1198 | key.page_match = 0;
|
| 1199 |
|
| 1200 | spin_lock_irqsave(&q->lock, flags);
|
| 1201 | __wake_up_locked_key(q, TASK_NORMAL, &key);
|
| 1202 |
|
| 1203 | |
| 1204 | * It's possible to miss clearing waiters here, when we woke our page
|
| 1205 | * waiters, but the hashed waitqueue has waiters for other pages on it.
|
| 1206 | * That's okay, it's a rare case. The next waker will clear it.
|
| 1207 | *
|
| 1208 | * Note that, depending on the page pool (buddy, hugetlb, ZONE_DEVICE,
|
| 1209 | * other), the flag may be cleared in the course of freeing the page;
|
| 1210 | * but that is not required for correctness.
|
| 1211 |
|
| 1212 | if (!waitqueue_active(q) || !key.page_match)
|
| 1213 | folio_clear_waiters(folio);
|
| 1214 |
|
| 1215 | spin_unlock_irqrestore(&q->lock, flags);
|
| 1216 | }
|
| 1217 |
|
| 1218 | |
| 1219 | * A choice of three behaviors for folio_wait_bit_common():
|
| 1220 |
|
| 1221 | enum behavior {
|
| 1222 | EXCLUSIVE, |
| 1223 | * __folio_lock() waiting on then setting PG_locked.
|
| 1224 |
|
| 1225 | SHARED, |
| 1226 | * folio_wait_writeback() waiting on PG_writeback.
|
| 1227 |
|
| 1228 | DROP, |
| 1229 | * like folio_put_wait_locked() on PG_locked.
|
| 1230 |
|
| 1231 | };
|
| 1232 |
|
| 1233 | |
| 1234 | * Attempt to check (or get) the folio flag, and mark us done
|
| 1235 | * if successful.
|
| 1236 |
|
| 1237 | static inline bool folio_trylock_flag(struct folio *folio, int bit_nr,
|
| 1238 | struct wait_queue_entry *wait)
|
| 1239 | {
|
| 1240 | if (wait->flags & WQ_FLAG_EXCLUSIVE) {
|
| 1241 | if (test_and_set_bit(bit_nr, &folio->flags.f))
|
| 1242 | return false;
|
| 1243 | } else if (test_bit(bit_nr, &folio->flags.f))
|
| 1244 | return false;
|
| 1245 |
|
| 1246 | wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE;
|
| 1247 | return true;
|
| 1248 | }
|
| 1249 |
|
| 1250 | static inline int folio_wait_bit_common(struct folio *folio, int bit_nr,
|
| 1251 | int state, enum behavior behavior)
|
| 1252 | {
|
| 1253 | wait_queue_head_t *q = folio_waitqueue(folio);
|
| 1254 | int unfairness = sysctl_page_lock_unfairness;
|
| 1255 | struct wait_page_queue wait_page;
|
| 1256 | wait_queue_entry_t *wait = &wait_page.wait;
|
| 1257 | bool thrashing = false;
|
| 1258 | unsigned long pflags;
|
| 1259 | bool in_thrashing;
|
| 1260 |
|
| 1261 | if (bit_nr == PG_locked &&
|
| 1262 | !folio_test_uptodate(folio) && folio_test_workingset(folio)) {
|
| 1263 | delayacct_thrashing_start(&in_thrashing);
|
| 1264 | psi_memstall_enter(&pflags);
|
| 1265 | thrashing = true;
|
| 1266 | }
|
| 1267 |
|
| 1268 | init_wait(wait);
|
| 1269 | wait->func = wake_page_function;
|
| 1270 | wait_page.folio = folio;
|
| 1271 | wait_page.bit_nr = bit_nr;
|
| 1272 |
|
| 1273 | repeat:
|
| 1274 | wait->flags = 0;
|
| 1275 | if (behavior == EXCLUSIVE) {
|
| 1276 | wait->flags = WQ_FLAG_EXCLUSIVE;
|
| 1277 | if (--unfairness < 0)
|
| 1278 | wait->flags |= WQ_FLAG_CUSTOM;
|
| 1279 | }
|
| 1280 |
|
| 1281 | |
| 1282 | * Do one last check whether we can get the
|
| 1283 | * page bit synchronously.
|
| 1284 | *
|
| 1285 | * Do the folio_set_waiters() marking before that
|
| 1286 | * to let any waker we _just_ missed know they
|
| 1287 | * need to wake us up (otherwise they'll never
|
| 1288 | * even go to the slow case that looks at the
|
| 1289 | * page queue), and add ourselves to the wait
|
| 1290 | * queue if we need to sleep.
|
| 1291 | *
|
| 1292 | * This part needs to be done under the queue
|
| 1293 | * lock to avoid races.
|
| 1294 |
|
| 1295 | spin_lock_irq(&q->lock);
|
| 1296 | folio_set_waiters(folio);
|
| 1297 | if (!folio_trylock_flag(folio, bit_nr, wait))
|
| 1298 | __add_wait_queue_entry_tail(q, wait);
|
| 1299 | spin_unlock_irq(&q->lock);
|
| 1300 |
|
| 1301 | |
| 1302 | * From now on, all the logic will be based on
|
| 1303 | * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
|
| 1304 | * see whether the page bit testing has already
|
| 1305 | * been done by the wake function.
|
| 1306 | *
|
| 1307 | * We can drop our reference to the folio.
|
| 1308 |
|
| 1309 | if (behavior == DROP)
|
| 1310 | folio_put(folio);
|
| 1311 |
|
| 1312 | |
| 1313 | * Note that until the "finish_wait()", or until
|
| 1314 | * we see the WQ_FLAG_WOKEN flag, we need to
|
| 1315 | * be very careful with the 'wait->flags', because
|
| 1316 | * we may race with a waker that sets them.
|
| 1317 |
|
| 1318 | for (;;) {
|
| 1319 | unsigned int flags;
|
| 1320 |
|
| 1321 | set_current_state(state);
|
| 1322 |
|
| 1323 |
|
| 1324 | flags = smp_load_acquire(&wait->flags);
|
| 1325 | if (!(flags & WQ_FLAG_WOKEN)) {
|
| 1326 | if (signal_pending_state(state, current))
|
| 1327 | break;
|
| 1328 |
|
| 1329 | io_schedule();
|
| 1330 | continue;
|
| 1331 | }
|
| 1332 |
|
| 1333 |
|
| 1334 | if (behavior != EXCLUSIVE)
|
| 1335 | break;
|
| 1336 |
|
| 1337 |
|
| 1338 | if (flags & WQ_FLAG_DONE)
|
| 1339 | break;
|
| 1340 |
|
| 1341 | |
| 1342 | * Otherwise, if we're getting the lock, we need to
|
| 1343 | * try to get it ourselves.
|
| 1344 | *
|
| 1345 | * And if that fails, we'll have to retry this all.
|
| 1346 |
|
| 1347 | if (unlikely(test_and_set_bit(bit_nr, folio_flags(folio, 0))))
|
| 1348 | goto repeat;
|
| 1349 |
|
| 1350 | wait->flags |= WQ_FLAG_DONE;
|
| 1351 | break;
|
| 1352 | }
|
| 1353 |
|
| 1354 | |
| 1355 | * If a signal happened, this 'finish_wait()' may remove the last
|
| 1356 | * waiter from the wait-queues, but the folio waiters bit will remain
|
| 1357 | * set. That's ok. The next wakeup will take care of it, and trying
|
| 1358 | * to do it here would be difficult and prone to races.
|
| 1359 |
|
| 1360 | finish_wait(q, wait);
|
| 1361 |
|
| 1362 | if (thrashing) {
|
| 1363 | delayacct_thrashing_end(&in_thrashing);
|
| 1364 | psi_memstall_leave(&pflags);
|
| 1365 | }
|
| 1366 |
|
| 1367 | |
| 1368 | * NOTE! The wait->flags weren't stable until we've done the
|
| 1369 | * 'finish_wait()', and we could have exited the loop above due
|
| 1370 | * to a signal, and had a wakeup event happen after the signal
|
| 1371 | * test but before the 'finish_wait()'.
|
| 1372 | *
|
| 1373 | * So only after the finish_wait() can we reliably determine
|
| 1374 | * if we got woken up or not, so we can now figure out the final
|
| 1375 | * return value based on that state without races.
|
| 1376 | *
|
| 1377 | * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
|
| 1378 | * waiter, but an exclusive one requires WQ_FLAG_DONE.
|
| 1379 |
|
| 1380 | if (behavior == EXCLUSIVE)
|
| 1381 | return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR;
|
| 1382 |
|
| 1383 | return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
|
| 1384 | }
|
| 1385 |
|
| 1386 | #ifdef CONFIG_MIGRATION
|
| 1387 | |
| 1388 | * softleaf_entry_wait_on_locked - Wait for a migration entry or
|
| 1389 | * device_private entry to be removed.
|
| 1390 | * @entry: migration or device_private swap entry.
|
| 1391 | * @ptl: already locked ptl. This function will drop the lock.
|
| 1392 | *
|
| 1393 | * Wait for a migration entry referencing the given page, or device_private
|
| 1394 | * entry referencing a dvice_private page to be unlocked. This is
|
| 1395 | * equivalent to folio_put_wait_locked(folio, TASK_UNINTERRUPTIBLE) except
|
| 1396 | * this can be called without taking a reference on the page. Instead this
|
| 1397 | * should be called while holding the ptl for @entry referencing
|
| 1398 | * the page.
|
| 1399 | *
|
| 1400 | * Returns after unlocking the ptl.
|
| 1401 | *
|
| 1402 | * This follows the same logic as folio_wait_bit_common() so see the comments
|
| 1403 | * there.
|
| 1404 |
|
| 1405 | void softleaf_entry_wait_on_locked(softleaf_t entry, spinlock_t *ptl)
|
| 1406 | __releases(ptl)
|
| 1407 | {
|
| 1408 | struct wait_page_queue wait_page;
|
| 1409 | wait_queue_entry_t *wait = &wait_page.wait;
|
| 1410 | bool thrashing = false;
|
| 1411 | unsigned long pflags;
|
| 1412 | bool in_thrashing;
|
| 1413 | wait_queue_head_t *q;
|
| 1414 | struct folio *folio = softleaf_to_folio(entry);
|
| 1415 |
|
| 1416 | q = folio_waitqueue(folio);
|
| 1417 | if (!folio_test_uptodate(folio) && folio_test_workingset(folio)) {
|
| 1418 | delayacct_thrashing_start(&in_thrashing);
|
| 1419 | psi_memstall_enter(&pflags);
|
| 1420 | thrashing = true;
|
| 1421 | }
|
| 1422 |
|
| 1423 | init_wait(wait);
|
| 1424 | wait->func = wake_page_function;
|
| 1425 | wait_page.folio = folio;
|
| 1426 | wait_page.bit_nr = PG_locked;
|
| 1427 | wait->flags = 0;
|
| 1428 |
|
| 1429 | spin_lock_irq(&q->lock);
|
| 1430 | folio_set_waiters(folio);
|
| 1431 | if (!folio_trylock_flag(folio, PG_locked, wait))
|
| 1432 | __add_wait_queue_entry_tail(q, wait);
|
| 1433 | spin_unlock_irq(&q->lock);
|
| 1434 |
|
| 1435 | |
| 1436 | * If a migration entry exists for the page the migration path must hold
|
| 1437 | * a valid reference to the page, and it must take the ptl to remove the
|
| 1438 | * migration entry. So the page is valid until the ptl is dropped.
|
| 1439 | * Similarly any path attempting to drop the last reference to a
|
| 1440 | * device-private page needs to grab the ptl to remove the device-private
|
| 1441 | * entry.
|
| 1442 |
|
| 1443 | spin_unlock(ptl);
|
| 1444 |
|
| 1445 | for (;;) {
|
| 1446 | unsigned int flags;
|
| 1447 |
|
| 1448 | set_current_state(TASK_UNINTERRUPTIBLE);
|
| 1449 |
|
| 1450 |
|
| 1451 | flags = smp_load_acquire(&wait->flags);
|
| 1452 | if (!(flags & WQ_FLAG_WOKEN)) {
|
| 1453 | if (signal_pending_state(TASK_UNINTERRUPTIBLE, current))
|
| 1454 | break;
|
| 1455 |
|
| 1456 | io_schedule();
|
| 1457 | continue;
|
| 1458 | }
|
| 1459 | break;
|
| 1460 | }
|
| 1461 |
|
| 1462 | finish_wait(q, wait);
|
| 1463 |
|
| 1464 | if (thrashing) {
|
| 1465 | delayacct_thrashing_end(&in_thrashing);
|
| 1466 | psi_memstall_leave(&pflags);
|
| 1467 | }
|
| 1468 | }
|
| 1469 | #endif
|
| 1470 |
|
| 1471 | void folio_wait_bit(struct folio *folio, int bit_nr)
|
| 1472 | {
|
| 1473 | folio_wait_bit_common(folio, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
|
| 1474 | }
|
| 1475 | EXPORT_SYMBOL(folio_wait_bit);
|
| 1476 |
|
| 1477 | int folio_wait_bit_killable(struct folio *folio, int bit_nr)
|
| 1478 | {
|
| 1479 | return folio_wait_bit_common(folio, bit_nr, TASK_KILLABLE, SHARED);
|
| 1480 | }
|
| 1481 | EXPORT_SYMBOL(folio_wait_bit_killable);
|
| 1482 |
|
| 1483 | |
| 1484 | * folio_put_wait_locked - Drop a reference and wait for it to be unlocked
|
| 1485 | * @folio: The folio to wait for.
|
| 1486 | * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc).
|
| 1487 | *
|
| 1488 | * The caller should hold a reference on @folio. They expect the page to
|
| 1489 | * become unlocked relatively soon, but do not wish to hold up migration
|
| 1490 | * (for example) by holding the reference while waiting for the folio to
|
| 1491 | * come unlocked. After this function returns, the caller should not
|
| 1492 | * dereference @folio.
|
| 1493 | *
|
| 1494 | * Return: 0 if the folio was unlocked or -EINTR if interrupted by a signal.
|
| 1495 |
|
| 1496 | static int folio_put_wait_locked(struct folio *folio, int state)
|
| 1497 | {
|
| 1498 | return folio_wait_bit_common(folio, PG_locked, state, DROP);
|
| 1499 | }
|
| 1500 |
|
| 1501 | |
| 1502 | * folio_unlock - Unlock a locked folio.
|
| 1503 | * @folio: The folio.
|
| 1504 | *
|
| 1505 | * Unlocks the folio and wakes up any thread sleeping on the page lock.
|
| 1506 | *
|
| 1507 | * Context: May be called from interrupt or process context. May not be
|
| 1508 | * called from NMI context.
|
| 1509 |
|
| 1510 | void folio_unlock(struct folio *folio)
|
| 1511 | {
|
| 1512 |
|
| 1513 | BUILD_BUG_ON(PG_waiters != 7);
|
| 1514 | BUILD_BUG_ON(PG_locked > 7);
|
| 1515 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
|
| 1516 | if (folio_xor_flags_has_waiters(folio, 1 << PG_locked))
|
| 1517 | folio_wake_bit(folio, PG_locked);
|
| 1518 | }
|
| 1519 | EXPORT_SYMBOL(folio_unlock);
|
| 1520 |
|
| 1521 | |
| 1522 | * folio_end_read - End read on a folio.
|
| 1523 | * @folio: The folio.
|
| 1524 | * @success: True if all reads completed successfully.
|
| 1525 | *
|
| 1526 | * When all reads against a folio have completed, filesystems should
|
| 1527 | * call this function to let the pagecache know that no more reads
|
| 1528 | * are outstanding. This will unlock the folio and wake up any thread
|
| 1529 | * sleeping on the lock. The folio will also be marked uptodate if all
|
| 1530 | * reads succeeded.
|
| 1531 | *
|
| 1532 | * Context: May be called from interrupt or process context. May not be
|
| 1533 | * called from NMI context.
|
| 1534 |
|
| 1535 | void folio_end_read(struct folio *folio, bool success)
|
| 1536 | {
|
| 1537 | unsigned long mask = 1 << PG_locked;
|
| 1538 |
|
| 1539 |
|
| 1540 | BUILD_BUG_ON(PG_uptodate > 7);
|
| 1541 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
|
| 1542 | VM_BUG_ON_FOLIO(success && folio_test_uptodate(folio), folio);
|
| 1543 |
|
| 1544 | if (likely(success))
|
| 1545 | mask |= 1 << PG_uptodate;
|
| 1546 | if (folio_xor_flags_has_waiters(folio, mask))
|
| 1547 | folio_wake_bit(folio, PG_locked);
|
| 1548 | }
|
| 1549 | EXPORT_SYMBOL(folio_end_read);
|
| 1550 |
|
| 1551 | |
| 1552 | * folio_end_private_2 - Clear PG_private_2 and wake any waiters.
|
| 1553 | * @folio: The folio.
|
| 1554 | *
|
| 1555 | * Clear the PG_private_2 bit on a folio and wake up any sleepers waiting for
|
| 1556 | * it. The folio reference held for PG_private_2 being set is released.
|
| 1557 | *
|
| 1558 | * This is, for example, used when a netfs folio is being written to a local
|
| 1559 | * disk cache, thereby allowing writes to the cache for the same folio to be
|
| 1560 | * serialised.
|
| 1561 |
|
| 1562 | void folio_end_private_2(struct folio *folio)
|
| 1563 | {
|
| 1564 | VM_BUG_ON_FOLIO(!folio_test_private_2(folio), folio);
|
| 1565 | clear_bit_unlock(PG_private_2, folio_flags(folio, 0));
|
| 1566 | folio_wake_bit(folio, PG_private_2);
|
| 1567 | folio_put(folio);
|
| 1568 | }
|
| 1569 | EXPORT_SYMBOL(folio_end_private_2);
|
| 1570 |
|
| 1571 | |
| 1572 | * folio_wait_private_2 - Wait for PG_private_2 to be cleared on a folio.
|
| 1573 | * @folio: The folio to wait on.
|
| 1574 | *
|
| 1575 | * Wait for PG_private_2 to be cleared on a folio.
|
| 1576 |
|
| 1577 | void folio_wait_private_2(struct folio *folio)
|
| 1578 | {
|
| 1579 | while (folio_test_private_2(folio))
|
| 1580 | folio_wait_bit(folio, PG_private_2);
|
| 1581 | }
|
| 1582 | EXPORT_SYMBOL(folio_wait_private_2);
|
| 1583 |
|
| 1584 | |
| 1585 | * folio_wait_private_2_killable - Wait for PG_private_2 to be cleared on a folio.
|
| 1586 | * @folio: The folio to wait on.
|
| 1587 | *
|
| 1588 | * Wait for PG_private_2 to be cleared on a folio or until a fatal signal is
|
| 1589 | * received by the calling task.
|
| 1590 | *
|
| 1591 | * Return:
|
| 1592 | * - 0 if successful.
|
| 1593 | * - -EINTR if a fatal signal was encountered.
|
| 1594 |
|
| 1595 | int folio_wait_private_2_killable(struct folio *folio)
|
| 1596 | {
|
| 1597 | int ret = 0;
|
| 1598 |
|
| 1599 | while (folio_test_private_2(folio)) {
|
| 1600 | ret = folio_wait_bit_killable(folio, PG_private_2);
|
| 1601 | if (ret < 0)
|
| 1602 | break;
|
| 1603 | }
|
| 1604 |
|
| 1605 | return ret;
|
| 1606 | }
|
| 1607 | EXPORT_SYMBOL(folio_wait_private_2_killable);
|
| 1608 |
|
| 1609 | static void filemap_end_dropbehind(struct folio *folio)
|
| 1610 | {
|
| 1611 | struct address_space *mapping = folio->mapping;
|
| 1612 |
|
| 1613 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
|
| 1614 |
|
| 1615 | if (folio_test_writeback(folio) || folio_test_dirty(folio))
|
| 1616 | return;
|
| 1617 | if (!folio_test_clear_dropbehind(folio))
|
| 1618 | return;
|
| 1619 | if (mapping && !folio_mapped(folio))
|
| 1620 | folio_unmap_invalidate(mapping, folio, 0);
|
| 1621 | }
|
| 1622 |
|
| 1623 | |
| 1624 | * If folio was marked as dropbehind, then pages should be dropped when writeback
|
| 1625 | * completes. Do that now. If we fail, it's likely because of a big folio -
|
| 1626 | * just reset dropbehind for that case and latter completions should invalidate.
|
| 1627 |
|
| 1628 | void folio_end_dropbehind(struct folio *folio)
|
| 1629 | {
|
| 1630 | if (!folio_test_dropbehind(folio))
|
| 1631 | return;
|
| 1632 |
|
| 1633 | |
| 1634 | * Hitting !in_task() should not happen off RWF_DONTCACHE writeback,
|
| 1635 | * but can happen if normal writeback just happens to find dirty folios
|
| 1636 | * that were created as part of uncached writeback, and that writeback
|
| 1637 | * would otherwise not need non-IRQ handling. Just skip the
|
| 1638 | * invalidation in that case.
|
| 1639 |
|
| 1640 | if (in_task() && folio_trylock(folio)) {
|
| 1641 | filemap_end_dropbehind(folio);
|
| 1642 | folio_unlock(folio);
|
| 1643 | }
|
| 1644 | }
|
| 1645 | EXPORT_SYMBOL_GPL(folio_end_dropbehind);
|
| 1646 |
|
| 1647 | |
| 1648 | * folio_end_writeback_no_dropbehind - End writeback against a folio.
|
| 1649 | * @folio: The folio.
|
| 1650 | *
|
| 1651 | * The folio must actually be under writeback.
|
| 1652 | * This call is intended for filesystems that need to defer dropbehind.
|
| 1653 | *
|
| 1654 | * Context: May be called from process or interrupt context.
|
| 1655 |
|
| 1656 | void folio_end_writeback_no_dropbehind(struct folio *folio)
|
| 1657 | {
|
| 1658 | VM_BUG_ON_FOLIO(!folio_test_writeback(folio), folio);
|
| 1659 |
|
| 1660 | |
| 1661 | * folio_test_clear_reclaim() could be used here but it is an
|
| 1662 | * atomic operation and overkill in this particular case. Failing
|
| 1663 | * to shuffle a folio marked for immediate reclaim is too mild
|
| 1664 | * a gain to justify taking an atomic operation penalty at the
|
| 1665 | * end of every folio writeback.
|
| 1666 |
|
| 1667 | if (folio_test_reclaim(folio)) {
|
| 1668 | folio_clear_reclaim(folio);
|
| 1669 | folio_rotate_reclaimable(folio);
|
| 1670 | }
|
| 1671 |
|
| 1672 | if (__folio_end_writeback(folio))
|
| 1673 | folio_wake_bit(folio, PG_writeback);
|
| 1674 |
|
| 1675 | acct_reclaim_writeback(folio);
|
| 1676 | }
|
| 1677 | EXPORT_SYMBOL_GPL(folio_end_writeback_no_dropbehind);
|
| 1678 |
|
| 1679 | |
| 1680 | * folio_end_writeback - End writeback against a folio.
|
| 1681 | * @folio: The folio.
|
| 1682 | *
|
| 1683 | * The folio must actually be under writeback.
|
| 1684 | *
|
| 1685 | * Context: May be called from process or interrupt context.
|
| 1686 |
|
| 1687 | void folio_end_writeback(struct folio *folio)
|
| 1688 | {
|
| 1689 | VM_BUG_ON_FOLIO(!folio_test_writeback(folio), folio);
|
| 1690 |
|
| 1691 | |
| 1692 | * Writeback does not hold a folio reference of its own, relying
|
| 1693 | * on truncation to wait for the clearing of PG_writeback.
|
| 1694 | * But here we must make sure that the folio is not freed and
|
| 1695 | * reused before the folio_wake_bit().
|
| 1696 |
|
| 1697 | folio_get(folio);
|
| 1698 | folio_end_writeback_no_dropbehind(folio);
|
| 1699 | folio_end_dropbehind(folio);
|
| 1700 | folio_put(folio);
|
| 1701 | }
|
| 1702 | EXPORT_SYMBOL(folio_end_writeback);
|
| 1703 |
|
| 1704 | |
| 1705 | * __folio_lock - Get a lock on the folio, assuming we need to sleep to get it.
|
| 1706 | * @folio: The folio to lock
|
| 1707 |
|
| 1708 | void __folio_lock(struct folio *folio)
|
| 1709 | {
|
| 1710 | folio_wait_bit_common(folio, PG_locked, TASK_UNINTERRUPTIBLE,
|
| 1711 | EXCLUSIVE);
|
| 1712 | }
|
| 1713 | EXPORT_SYMBOL(__folio_lock);
|
| 1714 |
|
| 1715 | int __folio_lock_killable(struct folio *folio)
|
| 1716 | {
|
| 1717 | return folio_wait_bit_common(folio, PG_locked, TASK_KILLABLE,
|
| 1718 | EXCLUSIVE);
|
| 1719 | }
|
| 1720 | EXPORT_SYMBOL_GPL(__folio_lock_killable);
|
| 1721 |
|
| 1722 | static int __folio_lock_async(struct folio *folio, struct wait_page_queue *wait)
|
| 1723 | {
|
| 1724 | struct wait_queue_head *q = folio_waitqueue(folio);
|
| 1725 | int ret;
|
| 1726 |
|
| 1727 | wait->folio = folio;
|
| 1728 | wait->bit_nr = PG_locked;
|
| 1729 |
|
| 1730 | spin_lock_irq(&q->lock);
|
| 1731 | __add_wait_queue_entry_tail(q, &wait->wait);
|
| 1732 | folio_set_waiters(folio);
|
| 1733 | ret = !folio_trylock(folio);
|
| 1734 | |
| 1735 | * If we were successful now, we know we're still on the
|
| 1736 | * waitqueue as we're still under the lock. This means it's
|
| 1737 | * safe to remove and return success, we know the callback
|
| 1738 | * isn't going to trigger.
|
| 1739 |
|
| 1740 | if (!ret)
|
| 1741 | __remove_wait_queue(q, &wait->wait);
|
| 1742 | else
|
| 1743 | ret = -EIOCBQUEUED;
|
| 1744 | spin_unlock_irq(&q->lock);
|
| 1745 | return ret;
|
| 1746 | }
|
| 1747 |
|
| 1748 | |
| 1749 | * Return values:
|
| 1750 | * 0 - folio is locked.
|
| 1751 | * non-zero - folio is not locked.
|
| 1752 | * mmap_lock or per-VMA lock has been released (mmap_read_unlock() or
|
| 1753 | * vma_end_read()), unless flags had both FAULT_FLAG_ALLOW_RETRY and
|
| 1754 | * FAULT_FLAG_RETRY_NOWAIT set, in which case the lock is still held.
|
| 1755 | *
|
| 1756 | * If neither ALLOW_RETRY nor KILLABLE are set, will always return 0
|
| 1757 | * with the folio locked and the mmap_lock/per-VMA lock is left unperturbed.
|
| 1758 |
|
| 1759 | vm_fault_t __folio_lock_or_retry(struct folio *folio, struct vm_fault *vmf)
|
| 1760 | {
|
| 1761 | unsigned int flags = vmf->flags;
|
| 1762 |
|
| 1763 | if (fault_flag_allow_retry_first(flags)) {
|
| 1764 | |
| 1765 | * CAUTION! In this case, mmap_lock/per-VMA lock is not
|
| 1766 | * released even though returning VM_FAULT_RETRY.
|
| 1767 |
|
| 1768 | if (flags & FAULT_FLAG_RETRY_NOWAIT)
|
| 1769 | return VM_FAULT_RETRY;
|
| 1770 |
|
| 1771 | release_fault_lock(vmf);
|
| 1772 | if (flags & FAULT_FLAG_KILLABLE)
|
| 1773 | folio_wait_locked_killable(folio);
|
| 1774 | else
|
| 1775 | folio_wait_locked(folio);
|
| 1776 | return VM_FAULT_RETRY;
|
| 1777 | }
|
| 1778 | if (flags & FAULT_FLAG_KILLABLE) {
|
| 1779 | bool ret;
|
| 1780 |
|
| 1781 | ret = __folio_lock_killable(folio);
|
| 1782 | if (ret) {
|
| 1783 | release_fault_lock(vmf);
|
| 1784 | return VM_FAULT_RETRY;
|
| 1785 | }
|
| 1786 | } else {
|
| 1787 | __folio_lock(folio);
|
| 1788 | }
|
| 1789 |
|
| 1790 | return 0;
|
| 1791 | }
|
| 1792 |
|
| 1793 | |
| 1794 | * page_cache_next_miss() - Find the next gap in the page cache.
|
| 1795 | * @mapping: Mapping.
|
| 1796 | * @index: Index.
|
| 1797 | * @max_scan: Maximum range to search.
|
| 1798 | *
|
| 1799 | * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
|
| 1800 | * gap with the lowest index.
|
| 1801 | *
|
| 1802 | * This function may be called under the rcu_read_lock. However, this will
|
| 1803 | * not atomically search a snapshot of the cache at a single point in time.
|
| 1804 | * For example, if a gap is created at index 5, then subsequently a gap is
|
| 1805 | * created at index 10, page_cache_next_miss covering both indices may
|
| 1806 | * return 10 if called under the rcu_read_lock.
|
| 1807 | *
|
| 1808 | * Return: The index of the gap if found, otherwise an index outside the
|
| 1809 | * range specified (in which case 'return - index >= max_scan' will be true).
|
| 1810 | * In the rare case of index wrap-around, 0 will be returned.
|
| 1811 |
|
| 1812 | pgoff_t page_cache_next_miss(struct address_space *mapping,
|
| 1813 | pgoff_t index, unsigned long max_scan)
|
| 1814 | {
|
| 1815 | XA_STATE(xas, &mapping->i_pages, index);
|
| 1816 |
|
| 1817 | while (max_scan--) {
|
| 1818 | void *entry = xas_next(&xas);
|
| 1819 | if (!entry || xa_is_value(entry))
|
| 1820 | return xas.xa_index;
|
| 1821 | if (xas.xa_index == 0)
|
| 1822 | return 0;
|
| 1823 | }
|
| 1824 |
|
| 1825 |
|
| 1826 | return xas.xa_index + 1;
|
| 1827 | }
|
| 1828 | EXPORT_SYMBOL(page_cache_next_miss);
|
| 1829 |
|
| 1830 | |
| 1831 | * page_cache_prev_miss() - Find the previous gap in the page cache.
|
| 1832 | * @mapping: Mapping.
|
| 1833 | * @index: Index.
|
| 1834 | * @max_scan: Maximum range to search.
|
| 1835 | *
|
| 1836 | * Search the range [max(index - max_scan + 1, 0), index] for the
|
| 1837 | * gap with the highest index.
|
| 1838 | *
|
| 1839 | * This function may be called under the rcu_read_lock. However, this will
|
| 1840 | * not atomically search a snapshot of the cache at a single point in time.
|
| 1841 | * For example, if a gap is created at index 10, then subsequently a gap is
|
| 1842 | * created at index 5, page_cache_prev_miss() covering both indices may
|
| 1843 | * return 5 if called under the rcu_read_lock.
|
| 1844 | *
|
| 1845 | * Return: The index of the gap if found, otherwise an index outside the
|
| 1846 | * range specified (in which case 'index - return >= max_scan' will be true).
|
| 1847 | * In the rare case of wrap-around, ULONG_MAX will be returned.
|
| 1848 |
|
| 1849 | pgoff_t page_cache_prev_miss(struct address_space *mapping,
|
| 1850 | pgoff_t index, unsigned long max_scan)
|
| 1851 | {
|
| 1852 | XA_STATE(xas, &mapping->i_pages, index);
|
| 1853 |
|
| 1854 | while (max_scan--) {
|
| 1855 | void *entry = xas_prev(&xas);
|
| 1856 | if (!entry || xa_is_value(entry))
|
| 1857 | return xas.xa_index;
|
| 1858 | if (xas.xa_index == ULONG_MAX)
|
| 1859 | return ULONG_MAX;
|
| 1860 | }
|
| 1861 |
|
| 1862 |
|
| 1863 | return xas.xa_index - 1;
|
| 1864 | }
|
| 1865 | EXPORT_SYMBOL(page_cache_prev_miss);
|
| 1866 |
|
| 1867 | |
| 1868 | * Lockless page cache protocol:
|
| 1869 | * On the lookup side:
|
| 1870 | * 1. Load the folio from i_pages
|
| 1871 | * 2. Increment the refcount if it's not zero
|
| 1872 | * 3. If the folio is not found by xas_reload(), put the refcount and retry
|
| 1873 | *
|
| 1874 | * On the removal side:
|
| 1875 | * A. Freeze the page (by zeroing the refcount if nobody else has a reference)
|
| 1876 | * B. Remove the page from i_pages
|
| 1877 | * C. Return the page to the page allocator
|
| 1878 | *
|
| 1879 | * This means that any page may have its reference count temporarily
|
| 1880 | * increased by a speculative page cache (or GUP-fast) lookup as it can
|
| 1881 | * be allocated by another user before the RCU grace period expires.
|
| 1882 | * Because the refcount temporarily acquired here may end up being the
|
| 1883 | * last refcount on the page, any page allocation must be freeable by
|
| 1884 | * folio_put().
|
| 1885 |
|
| 1886 |
|
| 1887 | |
| 1888 | * filemap_get_entry - Get a page cache entry.
|
| 1889 | * @mapping: the address_space to search
|
| 1890 | * @index: The page cache index.
|
| 1891 | *
|
| 1892 | * Looks up the page cache entry at @mapping & @index. If it is a folio,
|
| 1893 | * it is returned with an increased refcount. If it is a shadow entry
|
| 1894 | * of a previously evicted folio, or a swap entry from shmem/tmpfs,
|
| 1895 | * it is returned without further action.
|
| 1896 | *
|
| 1897 | * Return: The folio, swap or shadow entry, %NULL if nothing is found.
|
| 1898 |
|
| 1899 | void *filemap_get_entry(struct address_space *mapping, pgoff_t index)
|
| 1900 | {
|
| 1901 | XA_STATE(xas, &mapping->i_pages, index);
|
| 1902 | struct folio *folio;
|
| 1903 |
|
| 1904 | rcu_read_lock();
|
| 1905 | repeat:
|
| 1906 | xas_reset(&xas);
|
| 1907 | folio = xas_load(&xas);
|
| 1908 | if (xas_retry(&xas, folio))
|
| 1909 | goto repeat;
|
| 1910 | |
| 1911 | * A shadow entry of a recently evicted page, or a swap entry from
|
| 1912 | * shmem/tmpfs. Return it without attempting to raise page count.
|
| 1913 |
|
| 1914 | if (!folio || xa_is_value(folio))
|
| 1915 | goto out;
|
| 1916 |
|
| 1917 | if (!folio_try_get(folio))
|
| 1918 | goto repeat;
|
| 1919 |
|
| 1920 | if (unlikely(folio != xas_reload(&xas))) {
|
| 1921 | folio_put(folio);
|
| 1922 | goto repeat;
|
| 1923 | }
|
| 1924 | out:
|
| 1925 | rcu_read_unlock();
|
| 1926 |
|
| 1927 | return folio;
|
| 1928 | }
|
| 1929 |
|
| 1930 | |
| 1931 | * __filemap_get_folio_mpol - Find and get a reference to a folio.
|
| 1932 | * @mapping: The address_space to search.
|
| 1933 | * @index: The page index.
|
| 1934 | * @fgp_flags: %FGP flags modify how the folio is returned.
|
| 1935 | * @gfp: Memory allocation flags to use if %FGP_CREAT is specified.
|
| 1936 | * @policy: NUMA memory allocation policy to follow.
|
| 1937 | *
|
| 1938 | * Looks up the page cache entry at @mapping & @index.
|
| 1939 | *
|
| 1940 | * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
|
| 1941 | * if the %GFP flags specified for %FGP_CREAT are atomic.
|
| 1942 | *
|
| 1943 | * If this function returns a folio, it is returned with an increased refcount.
|
| 1944 | *
|
| 1945 | * Return: The found folio or an ERR_PTR() otherwise.
|
| 1946 |
|
| 1947 | struct folio *__filemap_get_folio_mpol(struct address_space *mapping,
|
| 1948 | pgoff_t index, fgf_t fgp_flags, gfp_t gfp, struct mempolicy *policy)
|
| 1949 | {
|
| 1950 | struct folio *folio;
|
| 1951 |
|
| 1952 | repeat:
|
| 1953 | folio = filemap_get_entry(mapping, index);
|
| 1954 | if (xa_is_value(folio))
|
| 1955 | folio = NULL;
|
| 1956 | if (!folio)
|
| 1957 | goto no_page;
|
| 1958 |
|
| 1959 | if (fgp_flags & FGP_LOCK) {
|
| 1960 | if (fgp_flags & FGP_NOWAIT) {
|
| 1961 | if (!folio_trylock(folio)) {
|
| 1962 | folio_put(folio);
|
| 1963 | return ERR_PTR(-EAGAIN);
|
| 1964 | }
|
| 1965 | } else {
|
| 1966 | folio_lock(folio);
|
| 1967 | }
|
| 1968 |
|
| 1969 |
|
| 1970 | if (unlikely(folio->mapping != mapping)) {
|
| 1971 | folio_unlock(folio);
|
| 1972 | folio_put(folio);
|
| 1973 | goto repeat;
|
| 1974 | }
|
| 1975 | VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
|
| 1976 | }
|
| 1977 |
|
| 1978 | if (fgp_flags & FGP_ACCESSED)
|
| 1979 | folio_mark_accessed(folio);
|
| 1980 | else if (fgp_flags & FGP_WRITE) {
|
| 1981 |
|
| 1982 | if (folio_test_idle(folio))
|
| 1983 | folio_clear_idle(folio);
|
| 1984 | }
|
| 1985 |
|
| 1986 | if (fgp_flags & FGP_STABLE)
|
| 1987 | folio_wait_stable(folio);
|
| 1988 | no_page:
|
| 1989 | if (!folio && (fgp_flags & FGP_CREAT)) {
|
| 1990 | unsigned int min_order = mapping_min_folio_order(mapping);
|
| 1991 | unsigned int order = max(min_order, FGF_GET_ORDER(fgp_flags));
|
| 1992 | int err;
|
| 1993 | index = mapping_align_index(mapping, index);
|
| 1994 |
|
| 1995 | if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping))
|
| 1996 | gfp |= __GFP_WRITE;
|
| 1997 | if (fgp_flags & FGP_NOFS)
|
| 1998 | gfp &= ~__GFP_FS;
|
| 1999 | if (fgp_flags & FGP_NOWAIT) {
|
| 2000 | gfp &= ~GFP_KERNEL;
|
| 2001 | gfp |= GFP_NOWAIT;
|
| 2002 | }
|
| 2003 | if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
|
| 2004 | fgp_flags |= FGP_LOCK;
|
| 2005 |
|
| 2006 | if (order > mapping_max_folio_order(mapping))
|
| 2007 | order = mapping_max_folio_order(mapping);
|
| 2008 |
|
| 2009 | if (index & ((1UL << order) - 1))
|
| 2010 | order = __ffs(index);
|
| 2011 |
|
| 2012 | do {
|
| 2013 | gfp_t alloc_gfp = gfp;
|
| 2014 |
|
| 2015 | err = -ENOMEM;
|
| 2016 | if (order > min_order)
|
| 2017 | alloc_gfp |= __GFP_NORETRY | __GFP_NOWARN;
|
| 2018 | folio = filemap_alloc_folio(alloc_gfp, order, policy);
|
| 2019 | if (!folio)
|
| 2020 | continue;
|
| 2021 |
|
| 2022 |
|
| 2023 | if (fgp_flags & FGP_ACCESSED)
|
| 2024 | __folio_set_referenced(folio);
|
| 2025 | if (fgp_flags & FGP_DONTCACHE)
|
| 2026 | __folio_set_dropbehind(folio);
|
| 2027 |
|
| 2028 | err = filemap_add_folio(mapping, folio, index, gfp);
|
| 2029 | if (!err)
|
| 2030 | break;
|
| 2031 | folio_put(folio);
|
| 2032 | folio = NULL;
|
| 2033 | } while (order-- > min_order);
|
| 2034 |
|
| 2035 | if (err == -EEXIST)
|
| 2036 | goto repeat;
|
| 2037 | if (err) {
|
| 2038 | |
| 2039 | * When NOWAIT I/O fails to allocate folios this could
|
| 2040 | * be due to a nonblocking memory allocation and not
|
| 2041 | * because the system actually is out of memory.
|
| 2042 | * Return -EAGAIN so that there caller retries in a
|
| 2043 | * blocking fashion instead of propagating -ENOMEM
|
| 2044 | * to the application.
|
| 2045 |
|
| 2046 | if ((fgp_flags & FGP_NOWAIT) && err == -ENOMEM)
|
| 2047 | err = -EAGAIN;
|
| 2048 | return ERR_PTR(err);
|
| 2049 | }
|
| 2050 | |
| 2051 | * filemap_add_folio locks the page, and for mmap
|
| 2052 | * we expect an unlocked page.
|
| 2053 |
|
| 2054 | if (folio && (fgp_flags & FGP_FOR_MMAP))
|
| 2055 | folio_unlock(folio);
|
| 2056 | }
|
| 2057 |
|
| 2058 | if (!folio)
|
| 2059 | return ERR_PTR(-ENOENT);
|
| 2060 |
|
| 2061 | if (!(fgp_flags & FGP_DONTCACHE) && folio_test_clear_dropbehind(folio)) {
|
| 2062 | if (folio_test_dirty(folio) &&
|
| 2063 | mapping_can_writeback(mapping)) {
|
| 2064 | struct inode *inode = mapping->host;
|
| 2065 | struct bdi_writeback *wb;
|
| 2066 | struct wb_lock_cookie cookie = {};
|
| 2067 | long nr = folio_nr_pages(folio);
|
| 2068 |
|
| 2069 | wb = unlocked_inode_to_wb_begin(inode, &cookie);
|
| 2070 | wb_stat_mod(wb, WB_DONTCACHE_DIRTY, -nr);
|
| 2071 | unlocked_inode_to_wb_end(inode, &cookie);
|
| 2072 | }
|
| 2073 | }
|
| 2074 | return folio;
|
| 2075 | }
|
| 2076 | EXPORT_SYMBOL(__filemap_get_folio_mpol);
|
| 2077 |
|
| 2078 | static inline struct folio *find_get_entry(struct xa_state *xas, pgoff_t max,
|
| 2079 | xa_mark_t mark)
|
| 2080 | {
|
| 2081 | struct folio *folio;
|
| 2082 |
|
| 2083 | retry:
|
| 2084 | if (mark == XA_PRESENT)
|
| 2085 | folio = xas_find(xas, max);
|
| 2086 | else
|
| 2087 | folio = xas_find_marked(xas, max, mark);
|
| 2088 |
|
| 2089 | if (xas_retry(xas, folio))
|
| 2090 | goto retry;
|
| 2091 | |
| 2092 | * A shadow entry of a recently evicted page, a swap
|
| 2093 | * entry from shmem/tmpfs or a DAX entry. Return it
|
| 2094 | * without attempting to raise page count.
|
| 2095 |
|
| 2096 | if (!folio || xa_is_value(folio))
|
| 2097 | return folio;
|
| 2098 |
|
| 2099 | if (!folio_try_get(folio))
|
| 2100 | goto reset;
|
| 2101 |
|
| 2102 | if (unlikely(folio != xas_reload(xas))) {
|
| 2103 | folio_put(folio);
|
| 2104 | goto reset;
|
| 2105 | }
|
| 2106 |
|
| 2107 | return folio;
|
| 2108 | reset:
|
| 2109 | xas_reset(xas);
|
| 2110 | goto retry;
|
| 2111 | }
|
| 2112 |
|
| 2113 | |
| 2114 | * find_get_entries - gang pagecache lookup
|
| 2115 | * @mapping: The address_space to search
|
| 2116 | * @start: The starting page cache index
|
| 2117 | * @end: The final page index (inclusive).
|
| 2118 | * @fbatch: Where the resulting entries are placed.
|
| 2119 | * @indices: The cache indices corresponding to the entries in @entries
|
| 2120 | *
|
| 2121 | * find_get_entries() will search for and return a batch of entries in
|
| 2122 | * the mapping. The entries are placed in @fbatch. find_get_entries()
|
| 2123 | * takes a reference on any actual folios it returns.
|
| 2124 | *
|
| 2125 | * The entries have ascending indexes. The indices may not be consecutive
|
| 2126 | * due to not-present entries or large folios.
|
| 2127 | *
|
| 2128 | * Any shadow entries of evicted folios, or swap entries from
|
| 2129 | * shmem/tmpfs, are included in the returned array.
|
| 2130 | *
|
| 2131 | * Return: The number of entries which were found.
|
| 2132 |
|
| 2133 | unsigned find_get_entries(struct address_space *mapping, pgoff_t *start,
|
| 2134 | pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
|
| 2135 | {
|
| 2136 | XA_STATE(xas, &mapping->i_pages, *start);
|
| 2137 | struct folio *folio;
|
| 2138 |
|
| 2139 | rcu_read_lock();
|
| 2140 | while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) {
|
| 2141 | indices[fbatch->nr] = xas.xa_index;
|
| 2142 | if (!folio_batch_add(fbatch, folio))
|
| 2143 | break;
|
| 2144 | }
|
| 2145 |
|
| 2146 | if (folio_batch_count(fbatch)) {
|
| 2147 | unsigned long nr;
|
| 2148 | int idx = folio_batch_count(fbatch) - 1;
|
| 2149 |
|
| 2150 | folio = fbatch->folios[idx];
|
| 2151 | if (!xa_is_value(folio))
|
| 2152 | nr = folio_nr_pages(folio);
|
| 2153 | else
|
| 2154 | nr = 1 << xa_get_order(&mapping->i_pages, indices[idx]);
|
| 2155 | *start = round_down(indices[idx] + nr, nr);
|
| 2156 | }
|
| 2157 | rcu_read_unlock();
|
| 2158 |
|
| 2159 | return folio_batch_count(fbatch);
|
| 2160 | }
|
| 2161 |
|
| 2162 | |
| 2163 | * find_lock_entries - Find a batch of pagecache entries.
|
| 2164 | * @mapping: The address_space to search.
|
| 2165 | * @start: The starting page cache index.
|
| 2166 | * @end: The final page index (inclusive).
|
| 2167 | * @fbatch: Where the resulting entries are placed.
|
| 2168 | * @indices: The cache indices of the entries in @fbatch.
|
| 2169 | *
|
| 2170 | * find_lock_entries() will return a batch of entries from @mapping.
|
| 2171 | * Swap, shadow and DAX entries are included. Folios are returned
|
| 2172 | * locked and with an incremented refcount. Folios which are locked
|
| 2173 | * by somebody else or under writeback are skipped. Folios which are
|
| 2174 | * partially outside the range are not returned.
|
| 2175 | *
|
| 2176 | * The entries have ascending indexes. The indices may not be consecutive
|
| 2177 | * due to not-present entries, large folios, folios which could not be
|
| 2178 | * locked or folios under writeback.
|
| 2179 | *
|
| 2180 | * Return: The number of entries which were found.
|
| 2181 |
|
| 2182 | unsigned find_lock_entries(struct address_space *mapping, pgoff_t *start,
|
| 2183 | pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
|
| 2184 | {
|
| 2185 | XA_STATE(xas, &mapping->i_pages, *start);
|
| 2186 | struct folio *folio;
|
| 2187 |
|
| 2188 | rcu_read_lock();
|
| 2189 | while ((folio = find_get_entry(&xas, end, XA_PRESENT))) {
|
| 2190 | unsigned long base;
|
| 2191 | unsigned long nr;
|
| 2192 |
|
| 2193 | if (!xa_is_value(folio)) {
|
| 2194 | nr = folio_nr_pages(folio);
|
| 2195 | base = folio->index;
|
| 2196 |
|
| 2197 | if (base < *start)
|
| 2198 | goto put;
|
| 2199 |
|
| 2200 | if (base + nr - 1 > end)
|
| 2201 | goto put;
|
| 2202 | if (!folio_trylock(folio))
|
| 2203 | goto put;
|
| 2204 | if (folio->mapping != mapping ||
|
| 2205 | folio_test_writeback(folio))
|
| 2206 | goto unlock;
|
| 2207 | VM_BUG_ON_FOLIO(!folio_contains(folio, xas.xa_index),
|
| 2208 | folio);
|
| 2209 | } else {
|
| 2210 | nr = 1 << xas_get_order(&xas);
|
| 2211 | base = xas.xa_index & ~(nr - 1);
|
| 2212 |
|
| 2213 | if (base < *start)
|
| 2214 | continue;
|
| 2215 |
|
| 2216 | if (base + nr - 1 > end)
|
| 2217 | break;
|
| 2218 | }
|
| 2219 |
|
| 2220 |
|
| 2221 | *start = base + nr;
|
| 2222 | indices[fbatch->nr] = xas.xa_index;
|
| 2223 | if (!folio_batch_add(fbatch, folio))
|
| 2224 | break;
|
| 2225 | continue;
|
| 2226 | unlock:
|
| 2227 | folio_unlock(folio);
|
| 2228 | put:
|
| 2229 | folio_put(folio);
|
| 2230 | }
|
| 2231 | rcu_read_unlock();
|
| 2232 |
|
| 2233 | return folio_batch_count(fbatch);
|
| 2234 | }
|
| 2235 |
|
| 2236 | |
| 2237 | * filemap_get_folios - Get a batch of folios
|
| 2238 | * @mapping: The address_space to search
|
| 2239 | * @start: The starting page index
|
| 2240 | * @end: The final page index (inclusive)
|
| 2241 | * @fbatch: The batch to fill.
|
| 2242 | *
|
| 2243 | * Search for and return a batch of folios in the mapping starting at
|
| 2244 | * index @start and up to index @end (inclusive). The folios are returned
|
| 2245 | * in @fbatch with an elevated reference count.
|
| 2246 | *
|
| 2247 | * Return: The number of folios which were found.
|
| 2248 | * We also update @start to index the next folio for the traversal.
|
| 2249 |
|
| 2250 | unsigned filemap_get_folios(struct address_space *mapping, pgoff_t *start,
|
| 2251 | pgoff_t end, struct folio_batch *fbatch)
|
| 2252 | {
|
| 2253 | return filemap_get_folios_tag(mapping, start, end, XA_PRESENT, fbatch);
|
| 2254 | }
|
| 2255 | EXPORT_SYMBOL(filemap_get_folios);
|
| 2256 |
|
| 2257 | |
| 2258 | * filemap_get_folios_contig - Get a batch of contiguous folios
|
| 2259 | * @mapping: The address_space to search
|
| 2260 | * @start: The starting page index
|
| 2261 | * @end: The final page index (inclusive)
|
| 2262 | * @fbatch: The batch to fill
|
| 2263 | *
|
| 2264 | * filemap_get_folios_contig() works exactly like filemap_get_folios(),
|
| 2265 | * except the returned folios are guaranteed to be contiguous. This may
|
| 2266 | * not return all contiguous folios if the batch gets filled up.
|
| 2267 | *
|
| 2268 | * Return: The number of folios found.
|
| 2269 | * Also update @start to be positioned for traversal of the next folio.
|
| 2270 |
|
| 2271 |
|
| 2272 | unsigned filemap_get_folios_contig(struct address_space *mapping,
|
| 2273 | pgoff_t *start, pgoff_t end, struct folio_batch *fbatch)
|
| 2274 | {
|
| 2275 | XA_STATE(xas, &mapping->i_pages, *start);
|
| 2276 | unsigned long nr;
|
| 2277 | struct folio *folio;
|
| 2278 |
|
| 2279 | if (*start > end)
|
| 2280 | return 0;
|
| 2281 |
|
| 2282 | rcu_read_lock();
|
| 2283 | for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) {
|
| 2284 | if (xas_retry(&xas, folio))
|
| 2285 | continue;
|
| 2286 | |
| 2287 | * If the entry has been swapped out, we can stop looking.
|
| 2288 | * No current caller is looking for DAX entries.
|
| 2289 |
|
| 2290 | if (xa_is_value(folio))
|
| 2291 | break;
|
| 2292 |
|
| 2293 |
|
| 2294 | if (xa_is_sibling(folio))
|
| 2295 | break;
|
| 2296 |
|
| 2297 | if (!folio_try_get(folio))
|
| 2298 | goto retry;
|
| 2299 |
|
| 2300 | if (unlikely(folio != xas_reload(&xas)))
|
| 2301 | goto put_folio;
|
| 2302 |
|
| 2303 | if (!folio_batch_add(fbatch, folio))
|
| 2304 | break;
|
| 2305 |
|
| 2306 | xas_advance(&xas, folio_next_index(folio) - 1);
|
| 2307 | if (xas.xa_index >= end)
|
| 2308 | break;
|
| 2309 | continue;
|
| 2310 |
|
| 2311 | put_folio:
|
| 2312 | folio_put(folio);
|
| 2313 | retry:
|
| 2314 | xas_reset(&xas);
|
| 2315 | }
|
| 2316 | rcu_read_unlock();
|
| 2317 |
|
| 2318 | nr = folio_batch_count(fbatch);
|
| 2319 | if (nr) {
|
| 2320 | folio = fbatch->folios[nr - 1];
|
| 2321 | *start = folio_next_index(folio);
|
| 2322 | }
|
| 2323 | return nr;
|
| 2324 | }
|
| 2325 | EXPORT_SYMBOL(filemap_get_folios_contig);
|
| 2326 |
|
| 2327 | |
| 2328 | * filemap_get_folios_tag - Get a batch of folios matching @tag
|
| 2329 | * @mapping: The address_space to search
|
| 2330 | * @start: The starting page index
|
| 2331 | * @end: The final page index (inclusive)
|
| 2332 | * @tag: The tag index
|
| 2333 | * @fbatch: The batch to fill
|
| 2334 | *
|
| 2335 | * The first folio may start before @start; if it does, it will contain
|
| 2336 | * @start. The final folio may extend beyond @end; if it does, it will
|
| 2337 | * contain @end. The folios have ascending indices. There may be gaps
|
| 2338 | * between the folios if there are indices which have no folio in the
|
| 2339 | * page cache. If folios are added to or removed from the page cache
|
| 2340 | * while this is running, they may or may not be found by this call.
|
| 2341 | * Only returns folios that are tagged with @tag.
|
| 2342 | *
|
| 2343 | * Return: The number of folios found.
|
| 2344 | * Also update @start to index the next folio for traversal.
|
| 2345 |
|
| 2346 | unsigned filemap_get_folios_tag(struct address_space *mapping, pgoff_t *start,
|
| 2347 | pgoff_t end, xa_mark_t tag, struct folio_batch *fbatch)
|
| 2348 | {
|
| 2349 | XA_STATE(xas, &mapping->i_pages, *start);
|
| 2350 | struct folio *folio;
|
| 2351 |
|
| 2352 | rcu_read_lock();
|
| 2353 | while ((folio = find_get_entry(&xas, end, tag)) != NULL) {
|
| 2354 | |
| 2355 | * Shadow entries should never be tagged, but this iteration
|
| 2356 | * is lockless so there is a window for page reclaim to evict
|
| 2357 | * a page we saw tagged. Skip over it.
|
| 2358 |
|
| 2359 | if (xa_is_value(folio))
|
| 2360 | continue;
|
| 2361 | if (!folio_batch_add(fbatch, folio)) {
|
| 2362 | *start = folio_next_index(folio);
|
| 2363 | goto out;
|
| 2364 | }
|
| 2365 | }
|
| 2366 | |
| 2367 | * We come here when there is no page beyond @end. We take care to not
|
| 2368 | * overflow the index @start as it confuses some of the callers. This
|
| 2369 | * breaks the iteration when there is a page at index -1 but that is
|
| 2370 | * already broke anyway.
|
| 2371 |
|
| 2372 | if (end == (pgoff_t)-1)
|
| 2373 | *start = (pgoff_t)-1;
|
| 2374 | else
|
| 2375 | *start = end + 1;
|
| 2376 | out:
|
| 2377 | rcu_read_unlock();
|
| 2378 |
|
| 2379 | return folio_batch_count(fbatch);
|
| 2380 | }
|
| 2381 | EXPORT_SYMBOL(filemap_get_folios_tag);
|
| 2382 |
|
| 2383 | |
| 2384 | * filemap_get_folios_dirty - Get a batch of dirty folios
|
| 2385 | * @mapping: The address_space to search
|
| 2386 | * @start: The starting folio index
|
| 2387 | * @end: The final folio index (inclusive)
|
| 2388 | * @fbatch: The batch to fill
|
| 2389 | *
|
| 2390 | * filemap_get_folios_dirty() works exactly like filemap_get_folios(), except
|
| 2391 | * the returned folios are presumed to be dirty or undergoing writeback. Dirty
|
| 2392 | * state is presumed because we don't block on folio lock nor want to miss
|
| 2393 | * folios. Callers that need to can recheck state upon locking the folio.
|
| 2394 | *
|
| 2395 | * This may not return all dirty folios if the batch gets filled up.
|
| 2396 | *
|
| 2397 | * Return: The number of folios found.
|
| 2398 | * Also update @start to be positioned for traversal of the next folio.
|
| 2399 |
|
| 2400 | unsigned filemap_get_folios_dirty(struct address_space *mapping, pgoff_t *start,
|
| 2401 | pgoff_t end, struct folio_batch *fbatch)
|
| 2402 | {
|
| 2403 | XA_STATE(xas, &mapping->i_pages, *start);
|
| 2404 | struct folio *folio;
|
| 2405 |
|
| 2406 | rcu_read_lock();
|
| 2407 | while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) {
|
| 2408 | if (xa_is_value(folio))
|
| 2409 | continue;
|
| 2410 | if (folio_trylock(folio)) {
|
| 2411 | bool clean = !folio_test_dirty(folio) &&
|
| 2412 | !folio_test_writeback(folio);
|
| 2413 | folio_unlock(folio);
|
| 2414 | if (clean) {
|
| 2415 | folio_put(folio);
|
| 2416 | continue;
|
| 2417 | }
|
| 2418 | }
|
| 2419 | if (!folio_batch_add(fbatch, folio)) {
|
| 2420 | *start = folio_next_index(folio);
|
| 2421 | goto out;
|
| 2422 | }
|
| 2423 | }
|
| 2424 | |
| 2425 | * We come here when there is no folio beyond @end. We take care to not
|
| 2426 | * overflow the index @start as it confuses some of the callers. This
|
| 2427 | * breaks the iteration when there is a folio at index -1 but that is
|
| 2428 | * already broke anyway.
|
| 2429 |
|
| 2430 | if (end == (pgoff_t)-1)
|
| 2431 | *start = (pgoff_t)-1;
|
| 2432 | else
|
| 2433 | *start = end + 1;
|
| 2434 | out:
|
| 2435 | rcu_read_unlock();
|
| 2436 |
|
| 2437 | return folio_batch_count(fbatch);
|
| 2438 | }
|
| 2439 |
|
| 2440 | |
| 2441 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail
|
| 2442 | * a _large_ part of the i/o request. Imagine the worst scenario:
|
| 2443 | *
|
| 2444 | * ---R__________________________________________B__________
|
| 2445 | * ^ reading here ^ bad block(assume 4k)
|
| 2446 | *
|
| 2447 | * read(R) => miss => readahead(R...B) => media error => frustrating retries
|
| 2448 | * => failing the whole request => read(R) => read(R+1) =>
|
| 2449 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
|
| 2450 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
|
| 2451 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
|
| 2452 | *
|
| 2453 | * It is going insane. Fix it by quickly scaling down the readahead size.
|
| 2454 |
|
| 2455 | static void shrink_readahead_size_eio(struct file_ra_state *ra)
|
| 2456 | {
|
| 2457 | ra->ra_pages /= 4;
|
| 2458 | }
|
| 2459 |
|
| 2460 | |
| 2461 | * filemap_get_read_batch - Get a batch of folios for read
|
| 2462 | *
|
| 2463 | * Get a batch of folios which represent a contiguous range of bytes in
|
| 2464 | * the file. No exceptional entries will be returned. If @index is in
|
| 2465 | * the middle of a folio, the entire folio will be returned. The last
|
| 2466 | * folio in the batch may have the readahead flag set or the uptodate flag
|
| 2467 | * clear so that the caller can take the appropriate action.
|
| 2468 |
|
| 2469 | static void filemap_get_read_batch(struct address_space *mapping,
|
| 2470 | pgoff_t index, pgoff_t max, struct folio_batch *fbatch)
|
| 2471 | {
|
| 2472 | XA_STATE(xas, &mapping->i_pages, index);
|
| 2473 | struct folio *folio;
|
| 2474 |
|
| 2475 | if (index > max)
|
| 2476 | return;
|
| 2477 |
|
| 2478 | rcu_read_lock();
|
| 2479 | for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) {
|
| 2480 | if (xas_retry(&xas, folio))
|
| 2481 | continue;
|
| 2482 | if (xa_is_value(folio))
|
| 2483 | break;
|
| 2484 | if (xa_is_sibling(folio))
|
| 2485 | break;
|
| 2486 | if (!folio_try_get(folio))
|
| 2487 | goto retry;
|
| 2488 |
|
| 2489 | if (unlikely(folio != xas_reload(&xas)))
|
| 2490 | goto put_folio;
|
| 2491 |
|
| 2492 | if (!folio_batch_add(fbatch, folio))
|
| 2493 | break;
|
| 2494 | if (!folio_test_uptodate(folio))
|
| 2495 | break;
|
| 2496 | if (folio_test_readahead(folio))
|
| 2497 | break;
|
| 2498 | xas_advance(&xas, folio_next_index(folio) - 1);
|
| 2499 | if (xas.xa_index >= max)
|
| 2500 | break;
|
| 2501 | continue;
|
| 2502 | put_folio:
|
| 2503 | folio_put(folio);
|
| 2504 | retry:
|
| 2505 | xas_reset(&xas);
|
| 2506 | }
|
| 2507 | rcu_read_unlock();
|
| 2508 | }
|
| 2509 |
|
| 2510 | static int filemap_read_folio(struct file *file, filler_t filler,
|
| 2511 | struct folio *folio)
|
| 2512 | {
|
| 2513 | bool workingset = folio_test_workingset(folio);
|
| 2514 | unsigned long pflags;
|
| 2515 | int error;
|
| 2516 |
|
| 2517 |
|
| 2518 | if (unlikely(workingset))
|
| 2519 | psi_memstall_enter(&pflags);
|
| 2520 | error = filler(file, folio);
|
| 2521 | if (unlikely(workingset))
|
| 2522 | psi_memstall_leave(&pflags);
|
| 2523 | if (error)
|
| 2524 | return error;
|
| 2525 |
|
| 2526 | error = folio_wait_locked_killable(folio);
|
| 2527 | if (error)
|
| 2528 | return error;
|
| 2529 | if (folio_test_uptodate(folio))
|
| 2530 | return 0;
|
| 2531 | if (file)
|
| 2532 | shrink_readahead_size_eio(&file->f_ra);
|
| 2533 | return -EIO;
|
| 2534 | }
|
| 2535 |
|
| 2536 | static bool filemap_range_uptodate(struct address_space *mapping,
|
| 2537 | loff_t pos, size_t count, struct folio *folio,
|
| 2538 | bool need_uptodate)
|
| 2539 | {
|
| 2540 | if (folio_test_uptodate(folio))
|
| 2541 | return true;
|
| 2542 |
|
| 2543 | if (need_uptodate)
|
| 2544 | return false;
|
| 2545 | if (!mapping->a_ops->is_partially_uptodate)
|
| 2546 | return false;
|
| 2547 | if (mapping->host->i_blkbits >= folio_shift(folio))
|
| 2548 | return false;
|
| 2549 |
|
| 2550 | if (folio_pos(folio) > pos) {
|
| 2551 | count -= folio_pos(folio) - pos;
|
| 2552 | pos = 0;
|
| 2553 | } else {
|
| 2554 | pos -= folio_pos(folio);
|
| 2555 | }
|
| 2556 |
|
| 2557 | if (pos == 0 && count >= folio_size(folio))
|
| 2558 | return false;
|
| 2559 |
|
| 2560 | return mapping->a_ops->is_partially_uptodate(folio, pos, count);
|
| 2561 | }
|
| 2562 |
|
| 2563 | static int filemap_update_page(struct kiocb *iocb,
|
| 2564 | struct address_space *mapping, size_t count,
|
| 2565 | struct folio *folio, bool need_uptodate)
|
| 2566 | {
|
| 2567 | int error;
|
| 2568 |
|
| 2569 | if (iocb->ki_flags & IOCB_NOWAIT) {
|
| 2570 | if (!filemap_invalidate_trylock_shared(mapping))
|
| 2571 | return -EAGAIN;
|
| 2572 | } else {
|
| 2573 | filemap_invalidate_lock_shared(mapping);
|
| 2574 | }
|
| 2575 |
|
| 2576 | if (!folio_trylock(folio)) {
|
| 2577 | error = -EAGAIN;
|
| 2578 | if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO))
|
| 2579 | goto unlock_mapping;
|
| 2580 | if (!(iocb->ki_flags & IOCB_WAITQ)) {
|
| 2581 | filemap_invalidate_unlock_shared(mapping);
|
| 2582 | |
| 2583 | * This is where we usually end up waiting for a
|
| 2584 | * previously submitted readahead to finish.
|
| 2585 |
|
| 2586 | folio_put_wait_locked(folio, TASK_KILLABLE);
|
| 2587 | return AOP_TRUNCATED_PAGE;
|
| 2588 | }
|
| 2589 | error = __folio_lock_async(folio, iocb->ki_waitq);
|
| 2590 | if (error)
|
| 2591 | goto unlock_mapping;
|
| 2592 | }
|
| 2593 |
|
| 2594 | error = AOP_TRUNCATED_PAGE;
|
| 2595 | if (!folio->mapping)
|
| 2596 | goto unlock;
|
| 2597 |
|
| 2598 | error = 0;
|
| 2599 | if (filemap_range_uptodate(mapping, iocb->ki_pos, count, folio,
|
| 2600 | need_uptodate))
|
| 2601 | goto unlock;
|
| 2602 |
|
| 2603 | error = -EAGAIN;
|
| 2604 | if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ))
|
| 2605 | goto unlock;
|
| 2606 |
|
| 2607 | error = filemap_read_folio(iocb->ki_filp, mapping->a_ops->read_folio,
|
| 2608 | folio);
|
| 2609 | goto unlock_mapping;
|
| 2610 | unlock:
|
| 2611 | folio_unlock(folio);
|
| 2612 | unlock_mapping:
|
| 2613 | filemap_invalidate_unlock_shared(mapping);
|
| 2614 | if (error == AOP_TRUNCATED_PAGE)
|
| 2615 | folio_put(folio);
|
| 2616 | return error;
|
| 2617 | }
|
| 2618 |
|
| 2619 | static int filemap_create_folio(struct kiocb *iocb, struct folio_batch *fbatch)
|
| 2620 | {
|
| 2621 | struct address_space *mapping = iocb->ki_filp->f_mapping;
|
| 2622 | struct folio *folio;
|
| 2623 | int error;
|
| 2624 | unsigned int min_order = mapping_min_folio_order(mapping);
|
| 2625 | pgoff_t index;
|
| 2626 |
|
| 2627 | if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ))
|
| 2628 | return -EAGAIN;
|
| 2629 |
|
| 2630 | folio = filemap_alloc_folio(mapping_gfp_mask(mapping), min_order, NULL);
|
| 2631 | if (!folio)
|
| 2632 | return -ENOMEM;
|
| 2633 | if (iocb->ki_flags & IOCB_DONTCACHE)
|
| 2634 | __folio_set_dropbehind(folio);
|
| 2635 |
|
| 2636 | |
| 2637 | * Protect against truncate / hole punch. Grabbing invalidate_lock
|
| 2638 | * here assures we cannot instantiate and bring uptodate new
|
| 2639 | * pagecache folios after evicting page cache during truncate
|
| 2640 | * and before actually freeing blocks. Note that we could
|
| 2641 | * release invalidate_lock after inserting the folio into
|
| 2642 | * the page cache as the locked folio would then be enough to
|
| 2643 | * synchronize with hole punching. But there are code paths
|
| 2644 | * such as filemap_update_page() filling in partially uptodate
|
| 2645 | * pages or ->readahead() that need to hold invalidate_lock
|
| 2646 | * while mapping blocks for IO so let's hold the lock here as
|
| 2647 | * well to keep locking rules simple.
|
| 2648 |
|
| 2649 | filemap_invalidate_lock_shared(mapping);
|
| 2650 | index = (iocb->ki_pos >> (PAGE_SHIFT + min_order)) << min_order;
|
| 2651 | error = filemap_add_folio(mapping, folio, index,
|
| 2652 | mapping_gfp_constraint(mapping, GFP_KERNEL));
|
| 2653 | if (error == -EEXIST)
|
| 2654 | error = AOP_TRUNCATED_PAGE;
|
| 2655 | if (error)
|
| 2656 | goto error;
|
| 2657 |
|
| 2658 | error = filemap_read_folio(iocb->ki_filp, mapping->a_ops->read_folio,
|
| 2659 | folio);
|
| 2660 | if (error)
|
| 2661 | goto error;
|
| 2662 |
|
| 2663 | filemap_invalidate_unlock_shared(mapping);
|
| 2664 | folio_batch_add(fbatch, folio);
|
| 2665 | return 0;
|
| 2666 | error:
|
| 2667 | filemap_invalidate_unlock_shared(mapping);
|
| 2668 | folio_put(folio);
|
| 2669 | return error;
|
| 2670 | }
|
| 2671 |
|
| 2672 | static int filemap_readahead(struct kiocb *iocb, struct file *file,
|
| 2673 | struct address_space *mapping, struct folio *folio,
|
| 2674 | pgoff_t last_index)
|
| 2675 | {
|
| 2676 | DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, folio->index);
|
| 2677 |
|
| 2678 | if (iocb->ki_flags & IOCB_NOIO)
|
| 2679 | return -EAGAIN;
|
| 2680 | if (iocb->ki_flags & IOCB_DONTCACHE)
|
| 2681 | ractl.dropbehind = 1;
|
| 2682 | page_cache_async_ra(&ractl, folio, last_index - folio->index);
|
| 2683 | return 0;
|
| 2684 | }
|
| 2685 |
|
| 2686 | static int filemap_get_pages(struct kiocb *iocb, size_t count,
|
| 2687 | struct folio_batch *fbatch, bool need_uptodate)
|
| 2688 | {
|
| 2689 | struct file *filp = iocb->ki_filp;
|
| 2690 | struct address_space *mapping = filp->f_mapping;
|
| 2691 | pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
|
| 2692 | pgoff_t last_index;
|
| 2693 | struct folio *folio;
|
| 2694 | unsigned int flags;
|
| 2695 | int err = 0;
|
| 2696 |
|
| 2697 |
|
| 2698 | last_index = round_up(iocb->ki_pos + count,
|
| 2699 | mapping_min_folio_nrbytes(mapping)) >> PAGE_SHIFT;
|
| 2700 | retry:
|
| 2701 | if (fatal_signal_pending(current))
|
| 2702 | return -EINTR;
|
| 2703 |
|
| 2704 | filemap_get_read_batch(mapping, index, last_index - 1, fbatch);
|
| 2705 | if (!folio_batch_count(fbatch)) {
|
| 2706 | DEFINE_READAHEAD(ractl, filp, &filp->f_ra, mapping, index);
|
| 2707 |
|
| 2708 | if (iocb->ki_flags & IOCB_NOIO)
|
| 2709 | return -EAGAIN;
|
| 2710 | if (iocb->ki_flags & IOCB_NOWAIT)
|
| 2711 | flags = memalloc_noio_save();
|
| 2712 | if (iocb->ki_flags & IOCB_DONTCACHE)
|
| 2713 | ractl.dropbehind = 1;
|
| 2714 | page_cache_sync_ra(&ractl, last_index - index);
|
| 2715 | if (iocb->ki_flags & IOCB_NOWAIT)
|
| 2716 | memalloc_noio_restore(flags);
|
| 2717 | filemap_get_read_batch(mapping, index, last_index - 1, fbatch);
|
| 2718 | }
|
| 2719 | if (!folio_batch_count(fbatch)) {
|
| 2720 | err = filemap_create_folio(iocb, fbatch);
|
| 2721 | if (err == AOP_TRUNCATED_PAGE)
|
| 2722 | goto retry;
|
| 2723 | return err;
|
| 2724 | }
|
| 2725 |
|
| 2726 | folio = fbatch->folios[folio_batch_count(fbatch) - 1];
|
| 2727 | if (folio_test_readahead(folio)) {
|
| 2728 | err = filemap_readahead(iocb, filp, mapping, folio, last_index);
|
| 2729 | if (err)
|
| 2730 | goto err;
|
| 2731 | }
|
| 2732 | if (!folio_test_uptodate(folio)) {
|
| 2733 | if (folio_batch_count(fbatch) > 1) {
|
| 2734 | err = -EAGAIN;
|
| 2735 | goto err;
|
| 2736 | }
|
| 2737 | err = filemap_update_page(iocb, mapping, count, folio,
|
| 2738 | need_uptodate);
|
| 2739 | if (err)
|
| 2740 | goto err;
|
| 2741 | }
|
| 2742 |
|
| 2743 | trace_mm_filemap_get_pages(mapping, index, last_index - 1);
|
| 2744 | return 0;
|
| 2745 | err:
|
| 2746 | if (err < 0)
|
| 2747 | folio_put(folio);
|
| 2748 | if (likely(--fbatch->nr))
|
| 2749 | return 0;
|
| 2750 | if (err == AOP_TRUNCATED_PAGE)
|
| 2751 | goto retry;
|
| 2752 | return err;
|
| 2753 | }
|
| 2754 |
|
| 2755 | static inline bool pos_same_folio(loff_t pos1, loff_t pos2, struct folio *folio)
|
| 2756 | {
|
| 2757 | unsigned int shift = folio_shift(folio);
|
| 2758 |
|
| 2759 | return (pos1 >> shift == pos2 >> shift);
|
| 2760 | }
|
| 2761 |
|
| 2762 | static void filemap_end_dropbehind_read(struct folio *folio)
|
| 2763 | {
|
| 2764 | if (!folio_test_dropbehind(folio))
|
| 2765 | return;
|
| 2766 | if (folio_test_writeback(folio) || folio_test_dirty(folio))
|
| 2767 | return;
|
| 2768 | if (folio_trylock(folio)) {
|
| 2769 | filemap_end_dropbehind(folio);
|
| 2770 | folio_unlock(folio);
|
| 2771 | }
|
| 2772 | }
|
| 2773 |
|
| 2774 | |
| 2775 | * filemap_read - Read data from the page cache.
|
| 2776 | * @iocb: The iocb to read.
|
| 2777 | * @iter: Destination for the data.
|
| 2778 | * @already_read: Number of bytes already read by the caller.
|
| 2779 | *
|
| 2780 | * Copies data from the page cache. If the data is not currently present,
|
| 2781 | * uses the readahead and read_folio address_space operations to fetch it.
|
| 2782 | *
|
| 2783 | * Return: Total number of bytes copied, including those already read by
|
| 2784 | * the caller. If an error happens before any bytes are copied, returns
|
| 2785 | * a negative error number.
|
| 2786 |
|
| 2787 | ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
|
| 2788 | ssize_t already_read)
|
| 2789 | {
|
| 2790 | struct file *filp = iocb->ki_filp;
|
| 2791 | struct file_ra_state *ra = &filp->f_ra;
|
| 2792 | struct address_space *mapping = filp->f_mapping;
|
| 2793 | struct inode *inode = mapping->host;
|
| 2794 | struct folio_batch fbatch;
|
| 2795 | int i, error = 0;
|
| 2796 | bool writably_mapped;
|
| 2797 | loff_t isize, end_offset;
|
| 2798 | loff_t last_pos = ra->prev_pos;
|
| 2799 |
|
| 2800 | if (unlikely(iocb->ki_pos < 0))
|
| 2801 | return -EINVAL;
|
| 2802 | if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
|
| 2803 | return 0;
|
| 2804 | if (unlikely(!iov_iter_count(iter)))
|
| 2805 | return 0;
|
| 2806 |
|
| 2807 | iov_iter_truncate(iter, inode->i_sb->s_maxbytes - iocb->ki_pos);
|
| 2808 | folio_batch_init(&fbatch);
|
| 2809 |
|
| 2810 | do {
|
| 2811 | cond_resched();
|
| 2812 |
|
| 2813 | |
| 2814 | * If we've already successfully copied some data, then we
|
| 2815 | * can no longer safely return -EIOCBQUEUED. Hence mark
|
| 2816 | * an async read NOWAIT at that point.
|
| 2817 |
|
| 2818 | if ((iocb->ki_flags & IOCB_WAITQ) && already_read)
|
| 2819 | iocb->ki_flags |= IOCB_NOWAIT;
|
| 2820 |
|
| 2821 | if (unlikely(iocb->ki_pos >= i_size_read(inode)))
|
| 2822 | break;
|
| 2823 |
|
| 2824 | error = filemap_get_pages(iocb, iter->count, &fbatch, false);
|
| 2825 | if (error < 0)
|
| 2826 | break;
|
| 2827 |
|
| 2828 | |
| 2829 | * i_size must be checked after we know the pages are Uptodate.
|
| 2830 | *
|
| 2831 | * Checking i_size after the check allows us to calculate
|
| 2832 | * the correct value for "nr", which means the zero-filled
|
| 2833 | * part of the page is not copied back to userspace (unless
|
| 2834 | * another truncate extends the file - this is desired though).
|
| 2835 |
|
| 2836 | isize = i_size_read(inode);
|
| 2837 | if (unlikely(iocb->ki_pos >= isize))
|
| 2838 | goto put_folios;
|
| 2839 | end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
|
| 2840 |
|
| 2841 | |
| 2842 | * Once we start copying data, we don't want to be touching any
|
| 2843 | * cachelines that might be contended:
|
| 2844 |
|
| 2845 | writably_mapped = mapping_writably_mapped(mapping);
|
| 2846 |
|
| 2847 | |
| 2848 | * When a read accesses the same folio several times, only
|
| 2849 | * mark it as accessed the first time.
|
| 2850 |
|
| 2851 | if (!pos_same_folio(iocb->ki_pos, last_pos - 1,
|
| 2852 | fbatch.folios[0]))
|
| 2853 | folio_mark_accessed(fbatch.folios[0]);
|
| 2854 |
|
| 2855 | for (i = 0; i < folio_batch_count(&fbatch); i++) {
|
| 2856 | struct folio *folio = fbatch.folios[i];
|
| 2857 | size_t fsize = folio_size(folio);
|
| 2858 | size_t offset = iocb->ki_pos & (fsize - 1);
|
| 2859 | size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos,
|
| 2860 | fsize - offset);
|
| 2861 | size_t copied;
|
| 2862 |
|
| 2863 | if (end_offset < folio_pos(folio))
|
| 2864 | break;
|
| 2865 | if (i > 0)
|
| 2866 | folio_mark_accessed(folio);
|
| 2867 | |
| 2868 | * If users can be writing to this folio using arbitrary
|
| 2869 | * virtual addresses, take care of potential aliasing
|
| 2870 | * before reading the folio on the kernel side.
|
| 2871 |
|
| 2872 | if (writably_mapped)
|
| 2873 | flush_dcache_folio(folio);
|
| 2874 |
|
| 2875 | copied = copy_folio_to_iter(folio, offset, bytes, iter);
|
| 2876 |
|
| 2877 | already_read += copied;
|
| 2878 | iocb->ki_pos += copied;
|
| 2879 | last_pos = iocb->ki_pos;
|
| 2880 |
|
| 2881 | if (copied < bytes) {
|
| 2882 | error = -EFAULT;
|
| 2883 | break;
|
| 2884 | }
|
| 2885 | }
|
| 2886 | put_folios:
|
| 2887 | for (i = 0; i < folio_batch_count(&fbatch); i++) {
|
| 2888 | struct folio *folio = fbatch.folios[i];
|
| 2889 |
|
| 2890 | filemap_end_dropbehind_read(folio);
|
| 2891 | folio_put(folio);
|
| 2892 | }
|
| 2893 | folio_batch_init(&fbatch);
|
| 2894 | } while (iov_iter_count(iter) && iocb->ki_pos < isize && !error);
|
| 2895 |
|
| 2896 | file_accessed(filp);
|
| 2897 | ra->prev_pos = last_pos;
|
| 2898 | return already_read ? already_read : error;
|
| 2899 | }
|
| 2900 | EXPORT_SYMBOL_GPL(filemap_read);
|
| 2901 |
|
| 2902 | int kiocb_write_and_wait(struct kiocb *iocb, size_t count)
|
| 2903 | {
|
| 2904 | struct address_space *mapping = iocb->ki_filp->f_mapping;
|
| 2905 | loff_t pos = iocb->ki_pos;
|
| 2906 | loff_t end = pos + count - 1;
|
| 2907 |
|
| 2908 | if (iocb->ki_flags & IOCB_NOWAIT) {
|
| 2909 | if (filemap_range_needs_writeback(mapping, pos, end))
|
| 2910 | return -EAGAIN;
|
| 2911 | return 0;
|
| 2912 | }
|
| 2913 |
|
| 2914 | return filemap_write_and_wait_range(mapping, pos, end);
|
| 2915 | }
|
| 2916 | EXPORT_SYMBOL_GPL(kiocb_write_and_wait);
|
| 2917 |
|
| 2918 | int filemap_invalidate_pages(struct address_space *mapping,
|
| 2919 | loff_t pos, loff_t end, bool nowait)
|
| 2920 | {
|
| 2921 | int ret;
|
| 2922 |
|
| 2923 | if (nowait) {
|
| 2924 |
|
| 2925 | if (filemap_range_has_page(mapping, pos, end))
|
| 2926 | return -EAGAIN;
|
| 2927 | } else {
|
| 2928 | ret = filemap_write_and_wait_range(mapping, pos, end);
|
| 2929 | if (ret)
|
| 2930 | return ret;
|
| 2931 | }
|
| 2932 |
|
| 2933 | |
| 2934 | * After a write we want buffered reads to be sure to go to disk to get
|
| 2935 | * the new data. We invalidate clean cached page from the region we're
|
| 2936 | * about to write. We do this *before* the write so that we can return
|
| 2937 | * without clobbering -EIOCBQUEUED from ->direct_IO().
|
| 2938 |
|
| 2939 | return invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT,
|
| 2940 | end >> PAGE_SHIFT);
|
| 2941 | }
|
| 2942 |
|
| 2943 | int kiocb_invalidate_pages(struct kiocb *iocb, size_t count)
|
| 2944 | {
|
| 2945 | struct address_space *mapping = iocb->ki_filp->f_mapping;
|
| 2946 |
|
| 2947 | return filemap_invalidate_pages(mapping, iocb->ki_pos,
|
| 2948 | iocb->ki_pos + count - 1,
|
| 2949 | iocb->ki_flags & IOCB_NOWAIT);
|
| 2950 | }
|
| 2951 | EXPORT_SYMBOL_GPL(kiocb_invalidate_pages);
|
| 2952 |
|
| 2953 | |
| 2954 | * generic_file_read_iter - generic filesystem read routine
|
| 2955 | * @iocb: kernel I/O control block
|
| 2956 | * @iter: destination for the data read
|
| 2957 | *
|
| 2958 | * This is the "read_iter()" routine for all filesystems
|
| 2959 | * that can use the page cache directly.
|
| 2960 | *
|
| 2961 | * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
|
| 2962 | * be returned when no data can be read without waiting for I/O requests
|
| 2963 | * to complete; it doesn't prevent readahead.
|
| 2964 | *
|
| 2965 | * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
|
| 2966 | * requests shall be made for the read or for readahead. When no data
|
| 2967 | * can be read, -EAGAIN shall be returned. When readahead would be
|
| 2968 | * triggered, a partial, possibly empty read shall be returned.
|
| 2969 | *
|
| 2970 | * Return:
|
| 2971 | * * number of bytes copied, even for partial reads
|
| 2972 | * * negative error code (or 0 if IOCB_NOIO) if nothing was read
|
| 2973 |
|
| 2974 | ssize_t
|
| 2975 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
|
| 2976 | {
|
| 2977 | size_t count = iov_iter_count(iter);
|
| 2978 | ssize_t retval = 0;
|
| 2979 |
|
| 2980 | if (!count)
|
| 2981 | return 0;
|
| 2982 |
|
| 2983 | if (iocb->ki_flags & IOCB_DIRECT) {
|
| 2984 | struct file *file = iocb->ki_filp;
|
| 2985 | struct address_space *mapping = file->f_mapping;
|
| 2986 | struct inode *inode = mapping->host;
|
| 2987 |
|
| 2988 | retval = kiocb_write_and_wait(iocb, count);
|
| 2989 | if (retval < 0)
|
| 2990 | return retval;
|
| 2991 | file_accessed(file);
|
| 2992 |
|
| 2993 | retval = mapping->a_ops->direct_IO(iocb, iter);
|
| 2994 | if (retval >= 0) {
|
| 2995 | iocb->ki_pos += retval;
|
| 2996 | count -= retval;
|
| 2997 | }
|
| 2998 | if (retval != -EIOCBQUEUED)
|
| 2999 | iov_iter_revert(iter, count - iov_iter_count(iter));
|
| 3000 |
|
| 3001 | |
| 3002 | * Btrfs can have a short DIO read if we encounter
|
| 3003 | * compressed extents, so if there was an error, or if
|
| 3004 | * we've already read everything we wanted to, or if
|
| 3005 | * there was a short read because we hit EOF, go ahead
|
| 3006 | * and return. Otherwise fallthrough to buffered io for
|
| 3007 | * the rest of the read. Buffered reads will not work for
|
| 3008 | * DAX files, so don't bother trying.
|
| 3009 |
|
| 3010 | if (retval < 0 || !count || IS_DAX(inode))
|
| 3011 | return retval;
|
| 3012 | if (iocb->ki_pos >= i_size_read(inode))
|
| 3013 | return retval;
|
| 3014 | }
|
| 3015 |
|
| 3016 | return filemap_read(iocb, iter, retval);
|
| 3017 | }
|
| 3018 | EXPORT_SYMBOL(generic_file_read_iter);
|
| 3019 |
|
| 3020 | |
| 3021 | * Splice subpages from a folio into a pipe.
|
| 3022 |
|
| 3023 | size_t splice_folio_into_pipe(struct pipe_inode_info *pipe,
|
| 3024 | struct folio *folio, loff_t fpos, size_t size)
|
| 3025 | {
|
| 3026 | struct page *page;
|
| 3027 | size_t spliced = 0, offset = offset_in_folio(folio, fpos);
|
| 3028 |
|
| 3029 | page = folio_page(folio, offset / PAGE_SIZE);
|
| 3030 | size = min(size, folio_size(folio) - offset);
|
| 3031 | offset %= PAGE_SIZE;
|
| 3032 |
|
| 3033 | while (spliced < size && !pipe_is_full(pipe)) {
|
| 3034 | struct pipe_buffer *buf = pipe_head_buf(pipe);
|
| 3035 | size_t part = min_t(size_t, PAGE_SIZE - offset, size - spliced);
|
| 3036 |
|
| 3037 | *buf = (struct pipe_buffer) {
|
| 3038 | .ops = &page_cache_pipe_buf_ops,
|
| 3039 | .page = page,
|
| 3040 | .offset = offset,
|
| 3041 | .len = part,
|
| 3042 | };
|
| 3043 | folio_get(folio);
|
| 3044 | pipe->head++;
|
| 3045 | page++;
|
| 3046 | spliced += part;
|
| 3047 | offset = 0;
|
| 3048 | }
|
| 3049 |
|
| 3050 | return spliced;
|
| 3051 | }
|
| 3052 |
|
| 3053 | |
| 3054 | * filemap_splice_read - Splice data from a file's pagecache into a pipe
|
| 3055 | * @in: The file to read from
|
| 3056 | * @ppos: Pointer to the file position to read from
|
| 3057 | * @pipe: The pipe to splice into
|
| 3058 | * @len: The amount to splice
|
| 3059 | * @flags: The SPLICE_F_* flags
|
| 3060 | *
|
| 3061 | * This function gets folios from a file's pagecache and splices them into the
|
| 3062 | * pipe. Readahead will be called as necessary to fill more folios. This may
|
| 3063 | * be used for blockdevs also.
|
| 3064 | *
|
| 3065 | * Return: On success, the number of bytes read will be returned and *@ppos
|
| 3066 | * will be updated if appropriate; 0 will be returned if there is no more data
|
| 3067 | * to be read; -EAGAIN will be returned if the pipe had no space, and some
|
| 3068 | * other negative error code will be returned on error. A short read may occur
|
| 3069 | * if the pipe has insufficient space, we reach the end of the data or we hit a
|
| 3070 | * hole.
|
| 3071 |
|
| 3072 | ssize_t filemap_splice_read(struct file *in, loff_t *ppos,
|
| 3073 | struct pipe_inode_info *pipe,
|
| 3074 | size_t len, unsigned int flags)
|
| 3075 | {
|
| 3076 | struct folio_batch fbatch;
|
| 3077 | struct kiocb iocb;
|
| 3078 | size_t total_spliced = 0, used, npages;
|
| 3079 | loff_t isize, end_offset;
|
| 3080 | bool writably_mapped;
|
| 3081 | int i, error = 0;
|
| 3082 |
|
| 3083 | if (unlikely(*ppos >= in->f_mapping->host->i_sb->s_maxbytes))
|
| 3084 | return 0;
|
| 3085 |
|
| 3086 | init_sync_kiocb(&iocb, in);
|
| 3087 | iocb.ki_pos = *ppos;
|
| 3088 |
|
| 3089 |
|
| 3090 | used = pipe_buf_usage(pipe);
|
| 3091 | npages = max_t(ssize_t, pipe->max_usage - used, 0);
|
| 3092 | len = min_t(size_t, len, npages * PAGE_SIZE);
|
| 3093 |
|
| 3094 | folio_batch_init(&fbatch);
|
| 3095 |
|
| 3096 | do {
|
| 3097 | cond_resched();
|
| 3098 |
|
| 3099 | if (*ppos >= i_size_read(in->f_mapping->host))
|
| 3100 | break;
|
| 3101 |
|
| 3102 | iocb.ki_pos = *ppos;
|
| 3103 | error = filemap_get_pages(&iocb, len, &fbatch, true);
|
| 3104 | if (error < 0)
|
| 3105 | break;
|
| 3106 |
|
| 3107 | |
| 3108 | * i_size must be checked after we know the pages are Uptodate.
|
| 3109 | *
|
| 3110 | * Checking i_size after the check allows us to calculate
|
| 3111 | * the correct value for "nr", which means the zero-filled
|
| 3112 | * part of the page is not copied back to userspace (unless
|
| 3113 | * another truncate extends the file - this is desired though).
|
| 3114 |
|
| 3115 | isize = i_size_read(in->f_mapping->host);
|
| 3116 | if (unlikely(*ppos >= isize))
|
| 3117 | break;
|
| 3118 | end_offset = min_t(loff_t, isize, *ppos + len);
|
| 3119 |
|
| 3120 | |
| 3121 | * Once we start copying data, we don't want to be touching any
|
| 3122 | * cachelines that might be contended:
|
| 3123 |
|
| 3124 | writably_mapped = mapping_writably_mapped(in->f_mapping);
|
| 3125 |
|
| 3126 | for (i = 0; i < folio_batch_count(&fbatch); i++) {
|
| 3127 | struct folio *folio = fbatch.folios[i];
|
| 3128 | size_t n;
|
| 3129 |
|
| 3130 | if (folio_pos(folio) >= end_offset)
|
| 3131 | goto out;
|
| 3132 | folio_mark_accessed(folio);
|
| 3133 |
|
| 3134 | |
| 3135 | * If users can be writing to this folio using arbitrary
|
| 3136 | * virtual addresses, take care of potential aliasing
|
| 3137 | * before reading the folio on the kernel side.
|
| 3138 |
|
| 3139 | if (writably_mapped)
|
| 3140 | flush_dcache_folio(folio);
|
| 3141 |
|
| 3142 | n = min_t(loff_t, len, isize - *ppos);
|
| 3143 | n = splice_folio_into_pipe(pipe, folio, *ppos, n);
|
| 3144 | if (!n)
|
| 3145 | goto out;
|
| 3146 | len -= n;
|
| 3147 | total_spliced += n;
|
| 3148 | *ppos += n;
|
| 3149 | in->f_ra.prev_pos = *ppos;
|
| 3150 | if (pipe_is_full(pipe))
|
| 3151 | goto out;
|
| 3152 | }
|
| 3153 |
|
| 3154 | folio_batch_release(&fbatch);
|
| 3155 | } while (len);
|
| 3156 |
|
| 3157 | out:
|
| 3158 | folio_batch_release(&fbatch);
|
| 3159 | file_accessed(in);
|
| 3160 |
|
| 3161 | return total_spliced ? total_spliced : error;
|
| 3162 | }
|
| 3163 | EXPORT_SYMBOL(filemap_splice_read);
|
| 3164 |
|
| 3165 | static inline loff_t folio_seek_hole_data(struct xa_state *xas,
|
| 3166 | struct address_space *mapping, struct folio *folio,
|
| 3167 | loff_t start, loff_t end, bool seek_data)
|
| 3168 | {
|
| 3169 | const struct address_space_operations *ops = mapping->a_ops;
|
| 3170 | size_t offset, bsz = i_blocksize(mapping->host);
|
| 3171 |
|
| 3172 | if (xa_is_value(folio) || folio_test_uptodate(folio))
|
| 3173 | return seek_data ? start : end;
|
| 3174 | if (!ops->is_partially_uptodate)
|
| 3175 | return seek_data ? end : start;
|
| 3176 |
|
| 3177 | xas_pause(xas);
|
| 3178 | rcu_read_unlock();
|
| 3179 | folio_lock(folio);
|
| 3180 | if (unlikely(folio->mapping != mapping))
|
| 3181 | goto unlock;
|
| 3182 |
|
| 3183 | offset = offset_in_folio(folio, start) & ~(bsz - 1);
|
| 3184 |
|
| 3185 | do {
|
| 3186 | if (ops->is_partially_uptodate(folio, offset, bsz) ==
|
| 3187 | seek_data)
|
| 3188 | break;
|
| 3189 | start = (start + bsz) & ~((u64)bsz - 1);
|
| 3190 | offset += bsz;
|
| 3191 | } while (offset < folio_size(folio));
|
| 3192 | unlock:
|
| 3193 | folio_unlock(folio);
|
| 3194 | rcu_read_lock();
|
| 3195 | return start;
|
| 3196 | }
|
| 3197 |
|
| 3198 | static inline size_t seek_folio_size(struct xa_state *xas, struct folio *folio)
|
| 3199 | {
|
| 3200 | if (xa_is_value(folio))
|
| 3201 | return PAGE_SIZE << xas_get_order(xas);
|
| 3202 | return folio_size(folio);
|
| 3203 | }
|
| 3204 |
|
| 3205 | |
| 3206 | * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache.
|
| 3207 | * @mapping: Address space to search.
|
| 3208 | * @start: First byte to consider.
|
| 3209 | * @end: Limit of search (exclusive).
|
| 3210 | * @whence: Either SEEK_HOLE or SEEK_DATA.
|
| 3211 | *
|
| 3212 | * If the page cache knows which blocks contain holes and which blocks
|
| 3213 | * contain data, your filesystem can use this function to implement
|
| 3214 | * SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are
|
| 3215 | * entirely memory-based such as tmpfs, and filesystems which support
|
| 3216 | * unwritten extents.
|
| 3217 | *
|
| 3218 | * Return: The requested offset on success, or -ENXIO if @whence specifies
|
| 3219 | * SEEK_DATA and there is no data after @start. There is an implicit hole
|
| 3220 | * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start
|
| 3221 | * and @end contain data.
|
| 3222 |
|
| 3223 | loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start,
|
| 3224 | loff_t end, int whence)
|
| 3225 | {
|
| 3226 | XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT);
|
| 3227 | pgoff_t max = (end - 1) >> PAGE_SHIFT;
|
| 3228 | bool seek_data = (whence == SEEK_DATA);
|
| 3229 | struct folio *folio;
|
| 3230 |
|
| 3231 | if (end <= start)
|
| 3232 | return -ENXIO;
|
| 3233 |
|
| 3234 | rcu_read_lock();
|
| 3235 | while ((folio = find_get_entry(&xas, max, XA_PRESENT))) {
|
| 3236 | loff_t pos = (u64)xas.xa_index << PAGE_SHIFT;
|
| 3237 | size_t seek_size;
|
| 3238 | u64 next;
|
| 3239 |
|
| 3240 | if (start < pos) {
|
| 3241 | if (!seek_data)
|
| 3242 | goto unlock;
|
| 3243 | start = pos;
|
| 3244 | }
|
| 3245 |
|
| 3246 | seek_size = seek_folio_size(&xas, folio);
|
| 3247 | next = round_up((u64)pos + 1, seek_size);
|
| 3248 | if (next > (u64)end)
|
| 3249 | pos = end;
|
| 3250 | else
|
| 3251 | pos = next;
|
| 3252 | start = folio_seek_hole_data(&xas, mapping, folio, start, pos,
|
| 3253 | seek_data);
|
| 3254 | if (start < pos)
|
| 3255 | goto unlock;
|
| 3256 | if (start >= end)
|
| 3257 | break;
|
| 3258 | if (seek_size > PAGE_SIZE)
|
| 3259 | xas_set(&xas, pos >> PAGE_SHIFT);
|
| 3260 | if (!xa_is_value(folio))
|
| 3261 | folio_put(folio);
|
| 3262 | }
|
| 3263 | if (seek_data)
|
| 3264 | start = -ENXIO;
|
| 3265 | unlock:
|
| 3266 | rcu_read_unlock();
|
| 3267 | if (folio && !xa_is_value(folio))
|
| 3268 | folio_put(folio);
|
| 3269 | if (start > end)
|
| 3270 | return end;
|
| 3271 | return start;
|
| 3272 | }
|
| 3273 |
|
| 3274 | #ifdef CONFIG_MMU
|
| 3275 | #define MMAP_LOTSAMISS (100)
|
| 3276 | |
| 3277 | * lock_folio_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
|
| 3278 | * @vmf - the vm_fault for this fault.
|
| 3279 | * @folio - the folio to lock.
|
| 3280 | * @fpin - the pointer to the file we may pin (or is already pinned).
|
| 3281 | *
|
| 3282 | * This works similar to lock_folio_or_retry in that it can drop the
|
| 3283 | * mmap_lock. It differs in that it actually returns the folio locked
|
| 3284 | * if it returns 1 and 0 if it couldn't lock the folio. If we did have
|
| 3285 | * to drop the mmap_lock then fpin will point to the pinned file and
|
| 3286 | * needs to be fput()'ed at a later point.
|
| 3287 |
|
| 3288 | static int lock_folio_maybe_drop_mmap(struct vm_fault *vmf, struct folio *folio,
|
| 3289 | struct file **fpin)
|
| 3290 | {
|
| 3291 | if (folio_trylock(folio))
|
| 3292 | return 1;
|
| 3293 |
|
| 3294 | |
| 3295 | * NOTE! This will make us return with VM_FAULT_RETRY, but with
|
| 3296 | * the fault lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
|
| 3297 | * is supposed to work. We have way too many special cases..
|
| 3298 |
|
| 3299 | if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
|
| 3300 | return 0;
|
| 3301 |
|
| 3302 | *fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
|
| 3303 | if (vmf->flags & FAULT_FLAG_KILLABLE) {
|
| 3304 | if (__folio_lock_killable(folio)) {
|
| 3305 | |
| 3306 | * We didn't have the right flags to drop the
|
| 3307 | * fault lock, but all fault_handlers only check
|
| 3308 | * for fatal signals if we return VM_FAULT_RETRY,
|
| 3309 | * so we need to drop the fault lock here and
|
| 3310 | * return 0 if we don't have a fpin.
|
| 3311 |
|
| 3312 | if (*fpin == NULL)
|
| 3313 | release_fault_lock(vmf);
|
| 3314 | return 0;
|
| 3315 | }
|
| 3316 | } else
|
| 3317 | __folio_lock(folio);
|
| 3318 |
|
| 3319 | return 1;
|
| 3320 | }
|
| 3321 |
|
| 3322 | |
| 3323 | * Synchronous readahead happens when we don't even find a page in the page
|
| 3324 | * cache at all. We don't want to perform IO under the mmap sem, so if we have
|
| 3325 | * to drop the mmap sem we return the file that was pinned in order for us to do
|
| 3326 | * that. If we didn't pin a file then we return NULL. The file that is
|
| 3327 | * returned needs to be fput()'ed when we're done with it.
|
| 3328 |
|
| 3329 | static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
|
| 3330 | {
|
| 3331 | struct file *file = vmf->vma->vm_file;
|
| 3332 | struct file_ra_state *ra = &file->f_ra;
|
| 3333 | struct address_space *mapping = file->f_mapping;
|
| 3334 | DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff);
|
| 3335 | struct file *fpin = NULL;
|
| 3336 | vm_flags_t vm_flags = vmf->vma->vm_flags;
|
| 3337 | bool force_thp_readahead = false;
|
| 3338 | unsigned int thp_order = 0;
|
| 3339 | unsigned short mmap_miss;
|
| 3340 |
|
| 3341 |
|
| 3342 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && (vm_flags & VM_HUGEPAGE)) {
|
| 3343 | |
| 3344 | * Cap max THP order at 2MB: this is the common PMD-sized
|
| 3345 | * hugepage size, and it avoids memory pressure from very
|
| 3346 | * large forced readahead when mapping_max_folio_order() is
|
| 3347 | * high (for example, 128MB with 64K base pages on arm64).
|
| 3348 |
|
| 3349 | if (mapping_large_folio_support(mapping)) {
|
| 3350 | force_thp_readahead = true;
|
| 3351 | thp_order = min_t(unsigned int,
|
| 3352 | mapping_max_folio_order(mapping),
|
| 3353 | get_order(SZ_2M));
|
| 3354 | }
|
| 3355 | }
|
| 3356 |
|
| 3357 | if (!force_thp_readahead) {
|
| 3358 | |
| 3359 | * If we don't want any read-ahead, don't bother.
|
| 3360 | * VM_EXEC case below is already intended for random access.
|
| 3361 |
|
| 3362 | if ((vm_flags & (VM_RAND_READ | VM_EXEC)) == VM_RAND_READ)
|
| 3363 | return fpin;
|
| 3364 |
|
| 3365 | if (!ra->ra_pages)
|
| 3366 | return fpin;
|
| 3367 |
|
| 3368 | if (vm_flags & VM_SEQ_READ) {
|
| 3369 | fpin = maybe_unlock_mmap_for_io(vmf, fpin);
|
| 3370 | page_cache_sync_ra(&ractl, ra->ra_pages);
|
| 3371 | return fpin;
|
| 3372 | }
|
| 3373 | }
|
| 3374 |
|
| 3375 | if (!(vm_flags & (VM_SEQ_READ | VM_EXEC))) {
|
| 3376 |
|
| 3377 | mmap_miss = READ_ONCE(ra->mmap_miss);
|
| 3378 | if (mmap_miss < MMAP_LOTSAMISS * 10)
|
| 3379 | WRITE_ONCE(ra->mmap_miss, ++mmap_miss);
|
| 3380 |
|
| 3381 | |
| 3382 | * Do we miss much more than hit in this file? If so,
|
| 3383 | * stop bothering with read-ahead. It will only hurt.
|
| 3384 |
|
| 3385 | if (mmap_miss > MMAP_LOTSAMISS)
|
| 3386 | return fpin;
|
| 3387 | }
|
| 3388 |
|
| 3389 | if (force_thp_readahead) {
|
| 3390 | unsigned long folio_nr_pages = 1UL << thp_order;
|
| 3391 |
|
| 3392 | fpin = maybe_unlock_mmap_for_io(vmf, fpin);
|
| 3393 | ractl._index &= ~(folio_nr_pages - 1);
|
| 3394 | ra->size = folio_nr_pages;
|
| 3395 | |
| 3396 | * Fetch two folios so we get the chance to actually
|
| 3397 | * readahead, unless we've been told not to.
|
| 3398 |
|
| 3399 | if (!(vm_flags & VM_RAND_READ))
|
| 3400 | ra->size *= 2;
|
| 3401 | ra->async_size = folio_nr_pages;
|
| 3402 | ra->order = thp_order;
|
| 3403 | page_cache_ra_order(&ractl, ra);
|
| 3404 | return fpin;
|
| 3405 | }
|
| 3406 |
|
| 3407 | if (vm_flags & VM_EXEC) {
|
| 3408 | |
| 3409 | * Allow arch to request a preferred minimum folio order for
|
| 3410 | * executable memory. This can often be beneficial to
|
| 3411 | * performance if (e.g.) arm64 can contpte-map the folio.
|
| 3412 | * Executable memory rarely benefits from readahead, due to its
|
| 3413 | * random access nature, so set async_size to 0.
|
| 3414 | *
|
| 3415 | * Limit to the boundaries of the VMA to avoid reading in any
|
| 3416 | * pad that might exist between sections, which would be a waste
|
| 3417 | * of memory.
|
| 3418 |
|
| 3419 | struct vm_area_struct *vma = vmf->vma;
|
| 3420 | const unsigned long start = vma_start_pgoff(vma);
|
| 3421 | const unsigned long end = vma_end_pgoff(vma);
|
| 3422 | unsigned long ra_end;
|
| 3423 |
|
| 3424 | ra->order = exec_folio_order();
|
| 3425 | ra->start = round_down(vmf->pgoff, 1UL << ra->order);
|
| 3426 | ra->start = max(ra->start, start);
|
| 3427 | ra_end = round_up(ra->start + ra->ra_pages, 1UL << ra->order);
|
| 3428 | ra_end = min(ra_end, end);
|
| 3429 | ra->size = ra_end - ra->start;
|
| 3430 | ra->async_size = 0;
|
| 3431 | } else {
|
| 3432 | |
| 3433 | * mmap read-around
|
| 3434 |
|
| 3435 | ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2);
|
| 3436 | ra->size = ra->ra_pages;
|
| 3437 | ra->async_size = ra->ra_pages / 4;
|
| 3438 | ra->order = 0;
|
| 3439 | }
|
| 3440 |
|
| 3441 | fpin = maybe_unlock_mmap_for_io(vmf, fpin);
|
| 3442 | ractl._index = ra->start;
|
| 3443 | page_cache_ra_order(&ractl, ra);
|
| 3444 | return fpin;
|
| 3445 | }
|
| 3446 |
|
| 3447 | |
| 3448 | * Asynchronous readahead happens when we find the page and PG_readahead,
|
| 3449 | * so we want to possibly extend the readahead further. We return the file that
|
| 3450 | * was pinned if we have to drop the mmap_lock in order to do IO.
|
| 3451 |
|
| 3452 | static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
|
| 3453 | struct folio *folio)
|
| 3454 | {
|
| 3455 | struct file *file = vmf->vma->vm_file;
|
| 3456 | struct file_ra_state *ra = &file->f_ra;
|
| 3457 | DEFINE_READAHEAD(ractl, file, ra, file->f_mapping, vmf->pgoff);
|
| 3458 | struct file *fpin = NULL;
|
| 3459 | unsigned short mmap_miss;
|
| 3460 |
|
| 3461 |
|
| 3462 | if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages)
|
| 3463 | return fpin;
|
| 3464 |
|
| 3465 | |
| 3466 | * If the folio is locked, we're likely racing against another fault.
|
| 3467 | * Don't touch the mmap_miss counter to avoid decreasing it multiple
|
| 3468 | * times for a single folio and break the balance with mmap_miss
|
| 3469 | * increase in do_sync_mmap_readahead().
|
| 3470 | *
|
| 3471 | * VM_SEQ_READ and VM_EXEC mappings skip the mmap_miss increment in
|
| 3472 | * do_sync_mmap_readahead(), so skip the decrement here as well to
|
| 3473 | * keep the counter symmetric.
|
| 3474 |
|
| 3475 | if (likely(!folio_test_locked(folio)) &&
|
| 3476 | !(vmf->vma->vm_flags & (VM_SEQ_READ | VM_EXEC))) {
|
| 3477 | mmap_miss = READ_ONCE(ra->mmap_miss);
|
| 3478 | if (mmap_miss)
|
| 3479 | WRITE_ONCE(ra->mmap_miss, --mmap_miss);
|
| 3480 | }
|
| 3481 |
|
| 3482 | if (folio_test_readahead(folio)) {
|
| 3483 | fpin = maybe_unlock_mmap_for_io(vmf, fpin);
|
| 3484 | page_cache_async_ra(&ractl, folio, ra->ra_pages);
|
| 3485 | }
|
| 3486 | return fpin;
|
| 3487 | }
|
| 3488 |
|
| 3489 | static vm_fault_t filemap_fault_recheck_pte_none(struct vm_fault *vmf)
|
| 3490 | {
|
| 3491 | struct vm_area_struct *vma = vmf->vma;
|
| 3492 | vm_fault_t ret = 0;
|
| 3493 | pte_t *ptep;
|
| 3494 |
|
| 3495 | |
| 3496 | * We might have COW'ed a pagecache folio and might now have an mlocked
|
| 3497 | * anon folio mapped. The original pagecache folio is not mlocked and
|
| 3498 | * might have been evicted. During a read+clear/modify/write update of
|
| 3499 | * the PTE, such as done in do_numa_page()/change_pte_range(), we
|
| 3500 | * temporarily clear the PTE under PT lock and might detect it here as
|
| 3501 | * "none" when not holding the PT lock.
|
| 3502 | *
|
| 3503 | * Not rechecking the PTE under PT lock could result in an unexpected
|
| 3504 | * major fault in an mlock'ed region. Recheck only for this special
|
| 3505 | * scenario while holding the PT lock, to not degrade non-mlocked
|
| 3506 | * scenarios. Recheck the PTE without PT lock firstly, thereby reducing
|
| 3507 | * the number of times we hold PT lock.
|
| 3508 |
|
| 3509 | if (!(vma->vm_flags & VM_LOCKED))
|
| 3510 | return 0;
|
| 3511 |
|
| 3512 | if (!(vmf->flags & FAULT_FLAG_ORIG_PTE_VALID))
|
| 3513 | return 0;
|
| 3514 |
|
| 3515 | ptep = pte_offset_map_ro_nolock(vma->vm_mm, vmf->pmd, vmf->address,
|
| 3516 | &vmf->ptl);
|
| 3517 | if (unlikely(!ptep))
|
| 3518 | return VM_FAULT_NOPAGE;
|
| 3519 |
|
| 3520 | if (unlikely(!pte_none(ptep_get_lockless(ptep)))) {
|
| 3521 | ret = VM_FAULT_NOPAGE;
|
| 3522 | } else {
|
| 3523 | spin_lock(vmf->ptl);
|
| 3524 | if (unlikely(!pte_none(ptep_get(ptep))))
|
| 3525 | ret = VM_FAULT_NOPAGE;
|
| 3526 | spin_unlock(vmf->ptl);
|
| 3527 | }
|
| 3528 | pte_unmap(ptep);
|
| 3529 | return ret;
|
| 3530 | }
|
| 3531 |
|
| 3532 | |
| 3533 | * filemap_fault - read in file data for page fault handling
|
| 3534 | * @vmf: struct vm_fault containing details of the fault
|
| 3535 | *
|
| 3536 | * filemap_fault() is invoked via the vma operations vector for a
|
| 3537 | * mapped memory region to read in file data during a page fault.
|
| 3538 | *
|
| 3539 | * The goto's are kind of ugly, but this streamlines the normal case of having
|
| 3540 | * it in the page cache, and handles the special cases reasonably without
|
| 3541 | * having a lot of duplicated code.
|
| 3542 | *
|
| 3543 | * vma->vm_mm->mmap_lock must be held on entry.
|
| 3544 | *
|
| 3545 | * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
|
| 3546 | * may be dropped before doing I/O or by lock_folio_maybe_drop_mmap().
|
| 3547 | *
|
| 3548 | * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
|
| 3549 | * has not been released.
|
| 3550 | *
|
| 3551 | * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
|
| 3552 | *
|
| 3553 | * Return: bitwise-OR of %VM_FAULT_ codes.
|
| 3554 |
|
| 3555 | vm_fault_t filemap_fault(struct vm_fault *vmf)
|
| 3556 | {
|
| 3557 | int error;
|
| 3558 | struct file *file = vmf->vma->vm_file;
|
| 3559 | struct file *fpin = NULL;
|
| 3560 | struct address_space *mapping = file->f_mapping;
|
| 3561 | struct inode *inode = mapping->host;
|
| 3562 | pgoff_t max_idx, index = vmf->pgoff;
|
| 3563 | struct folio *folio;
|
| 3564 | vm_fault_t ret = 0;
|
| 3565 | bool mapping_locked = false;
|
| 3566 |
|
| 3567 | max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
|
| 3568 | if (unlikely(index >= max_idx))
|
| 3569 | return VM_FAULT_SIGBUS;
|
| 3570 |
|
| 3571 | trace_mm_filemap_fault(mapping, index);
|
| 3572 |
|
| 3573 | |
| 3574 | * Do we have something in the page cache already?
|
| 3575 |
|
| 3576 | folio = filemap_get_folio(mapping, index);
|
| 3577 | if (likely(!IS_ERR(folio))) {
|
| 3578 | |
| 3579 | * We found the page, so try async readahead before waiting for
|
| 3580 | * the lock.
|
| 3581 |
|
| 3582 | if (!(vmf->flags & FAULT_FLAG_TRIED))
|
| 3583 | fpin = do_async_mmap_readahead(vmf, folio);
|
| 3584 | if (unlikely(!folio_test_uptodate(folio))) {
|
| 3585 | filemap_invalidate_lock_shared(mapping);
|
| 3586 | mapping_locked = true;
|
| 3587 | }
|
| 3588 | } else {
|
| 3589 | ret = filemap_fault_recheck_pte_none(vmf);
|
| 3590 | if (unlikely(ret))
|
| 3591 | return ret;
|
| 3592 |
|
| 3593 |
|
| 3594 | count_vm_event(PGMAJFAULT);
|
| 3595 | count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
|
| 3596 | ret = VM_FAULT_MAJOR;
|
| 3597 | fpin = do_sync_mmap_readahead(vmf);
|
| 3598 | retry_find:
|
| 3599 | |
| 3600 | * See comment in filemap_create_folio() why we need
|
| 3601 | * invalidate_lock
|
| 3602 |
|
| 3603 | if (!mapping_locked) {
|
| 3604 | filemap_invalidate_lock_shared(mapping);
|
| 3605 | mapping_locked = true;
|
| 3606 | }
|
| 3607 | folio = __filemap_get_folio(mapping, index,
|
| 3608 | FGP_CREAT|FGP_FOR_MMAP,
|
| 3609 | vmf->gfp_mask);
|
| 3610 | if (IS_ERR(folio)) {
|
| 3611 | if (fpin)
|
| 3612 | goto out_retry;
|
| 3613 | filemap_invalidate_unlock_shared(mapping);
|
| 3614 | return VM_FAULT_OOM;
|
| 3615 | }
|
| 3616 | }
|
| 3617 |
|
| 3618 | if (!lock_folio_maybe_drop_mmap(vmf, folio, &fpin))
|
| 3619 | goto out_retry;
|
| 3620 |
|
| 3621 |
|
| 3622 | if (unlikely(folio->mapping != mapping)) {
|
| 3623 | folio_unlock(folio);
|
| 3624 | folio_put(folio);
|
| 3625 | goto retry_find;
|
| 3626 | }
|
| 3627 | VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
|
| 3628 |
|
| 3629 | |
| 3630 | * We have a locked folio in the page cache, now we need to check
|
| 3631 | * that it's up-to-date. If not, it is going to be due to an error,
|
| 3632 | * or because readahead was otherwise unable to retrieve it.
|
| 3633 |
|
| 3634 | if (unlikely(!folio_test_uptodate(folio))) {
|
| 3635 | |
| 3636 | * If the invalidate lock is not held, the folio was in cache
|
| 3637 | * and uptodate and now it is not. Strange but possible since we
|
| 3638 | * didn't hold the page lock all the time. Let's drop
|
| 3639 | * everything, get the invalidate lock and try again.
|
| 3640 |
|
| 3641 | if (!mapping_locked) {
|
| 3642 | folio_unlock(folio);
|
| 3643 | folio_put(folio);
|
| 3644 | goto retry_find;
|
| 3645 | }
|
| 3646 |
|
| 3647 | |
| 3648 | * OK, the folio is really not uptodate. This can be because the
|
| 3649 | * VMA has the VM_RAND_READ flag set, or because an error
|
| 3650 | * arose. Let's read it in directly.
|
| 3651 |
|
| 3652 | goto page_not_uptodate;
|
| 3653 | }
|
| 3654 |
|
| 3655 | |
| 3656 | * We've made it this far and we had to drop our mmap_lock, now is the
|
| 3657 | * time to return to the upper layer and have it re-find the vma and
|
| 3658 | * redo the fault.
|
| 3659 |
|
| 3660 | if (fpin) {
|
| 3661 | folio_unlock(folio);
|
| 3662 | goto out_retry;
|
| 3663 | }
|
| 3664 | if (mapping_locked)
|
| 3665 | filemap_invalidate_unlock_shared(mapping);
|
| 3666 |
|
| 3667 | |
| 3668 | * Found the page and have a reference on it.
|
| 3669 | * We must recheck i_size under page lock.
|
| 3670 |
|
| 3671 | max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
|
| 3672 | if (unlikely(index >= max_idx)) {
|
| 3673 | folio_unlock(folio);
|
| 3674 | folio_put(folio);
|
| 3675 | return VM_FAULT_SIGBUS;
|
| 3676 | }
|
| 3677 |
|
| 3678 | vmf->page = folio_file_page(folio, index);
|
| 3679 | return ret | VM_FAULT_LOCKED;
|
| 3680 |
|
| 3681 | page_not_uptodate:
|
| 3682 | |
| 3683 | * Umm, take care of errors if the page isn't up-to-date.
|
| 3684 | * Try to re-read it _once_. We do this synchronously,
|
| 3685 | * because there really aren't any performance issues here
|
| 3686 | * and we need to check for errors.
|
| 3687 |
|
| 3688 | fpin = maybe_unlock_mmap_for_io(vmf, fpin);
|
| 3689 | error = filemap_read_folio(file, mapping->a_ops->read_folio, folio);
|
| 3690 | if (fpin)
|
| 3691 | goto out_retry;
|
| 3692 | folio_put(folio);
|
| 3693 |
|
| 3694 | if (!error || error == AOP_TRUNCATED_PAGE)
|
| 3695 | goto retry_find;
|
| 3696 | filemap_invalidate_unlock_shared(mapping);
|
| 3697 |
|
| 3698 | return VM_FAULT_SIGBUS;
|
| 3699 |
|
| 3700 | out_retry:
|
| 3701 | |
| 3702 | * We dropped the mmap_lock, we need to return to the fault handler to
|
| 3703 | * re-find the vma and come back and find our hopefully still populated
|
| 3704 | * page.
|
| 3705 |
|
| 3706 | if (!IS_ERR(folio))
|
| 3707 | folio_put(folio);
|
| 3708 | if (mapping_locked)
|
| 3709 | filemap_invalidate_unlock_shared(mapping);
|
| 3710 | if (fpin)
|
| 3711 | fput(fpin);
|
| 3712 | return ret | VM_FAULT_RETRY;
|
| 3713 | }
|
| 3714 | EXPORT_SYMBOL(filemap_fault);
|
| 3715 |
|
| 3716 | static bool filemap_map_pmd(struct vm_fault *vmf, struct folio *folio,
|
| 3717 | pgoff_t start)
|
| 3718 | {
|
| 3719 | struct mm_struct *mm = vmf->vma->vm_mm;
|
| 3720 |
|
| 3721 |
|
| 3722 | if (pmd_trans_huge(*vmf->pmd)) {
|
| 3723 | folio_unlock(folio);
|
| 3724 | folio_put(folio);
|
| 3725 | return true;
|
| 3726 | }
|
| 3727 |
|
| 3728 | if (pmd_none(*vmf->pmd) && folio_test_pmd_mappable(folio)) {
|
| 3729 | struct page *page = folio_file_page(folio, start);
|
| 3730 | vm_fault_t ret = do_set_pmd(vmf, folio, page);
|
| 3731 | if (!ret) {
|
| 3732 |
|
| 3733 | folio_unlock(folio);
|
| 3734 | return true;
|
| 3735 | }
|
| 3736 | }
|
| 3737 |
|
| 3738 | if (pmd_none(*vmf->pmd) && vmf->prealloc_pte)
|
| 3739 | pmd_install(mm, vmf->pmd, &vmf->prealloc_pte);
|
| 3740 |
|
| 3741 | return false;
|
| 3742 | }
|
| 3743 |
|
| 3744 | static struct folio *next_uptodate_folio(struct xa_state *xas,
|
| 3745 | struct address_space *mapping, pgoff_t end_pgoff)
|
| 3746 | {
|
| 3747 | struct folio *folio = xas_next_entry(xas, end_pgoff);
|
| 3748 | unsigned long max_idx;
|
| 3749 |
|
| 3750 | do {
|
| 3751 | if (!folio)
|
| 3752 | return NULL;
|
| 3753 | if (xas_retry(xas, folio))
|
| 3754 | continue;
|
| 3755 | if (xa_is_value(folio))
|
| 3756 | continue;
|
| 3757 | if (!folio_try_get(folio))
|
| 3758 | continue;
|
| 3759 | if (folio_test_locked(folio))
|
| 3760 | goto skip;
|
| 3761 |
|
| 3762 | if (unlikely(folio != xas_reload(xas)))
|
| 3763 | goto skip;
|
| 3764 | if (!folio_test_uptodate(folio) || folio_test_readahead(folio))
|
| 3765 | goto skip;
|
| 3766 | if (!folio_trylock(folio))
|
| 3767 | goto skip;
|
| 3768 | if (folio->mapping != mapping)
|
| 3769 | goto unlock;
|
| 3770 | if (!folio_test_uptodate(folio))
|
| 3771 | goto unlock;
|
| 3772 | max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
|
| 3773 | if (xas->xa_index >= max_idx)
|
| 3774 | goto unlock;
|
| 3775 | return folio;
|
| 3776 | unlock:
|
| 3777 | folio_unlock(folio);
|
| 3778 | skip:
|
| 3779 | folio_put(folio);
|
| 3780 | } while ((folio = xas_next_entry(xas, end_pgoff)) != NULL);
|
| 3781 |
|
| 3782 | return NULL;
|
| 3783 | }
|
| 3784 |
|
| 3785 | |
| 3786 | * Map page range [start_page, start_page + nr_pages) of folio.
|
| 3787 | * start_page is gotten from start by folio_page(folio, start)
|
| 3788 |
|
| 3789 | static vm_fault_t filemap_map_folio_range(struct vm_fault *vmf,
|
| 3790 | struct folio *folio, unsigned long start,
|
| 3791 | unsigned long addr, unsigned int nr_pages,
|
| 3792 | unsigned long *rss, pgoff_t file_end)
|
| 3793 | {
|
| 3794 | struct address_space *mapping = folio->mapping;
|
| 3795 | unsigned int ref_from_caller = 1;
|
| 3796 | vm_fault_t ret = 0;
|
| 3797 | struct page *page = folio_page(folio, start);
|
| 3798 | unsigned int count = 0;
|
| 3799 | pte_t *old_ptep = vmf->pte;
|
| 3800 | unsigned long addr0;
|
| 3801 |
|
| 3802 | |
| 3803 | * Map the large folio fully where possible:
|
| 3804 | *
|
| 3805 | * - The folio is fully within size of the file or belong
|
| 3806 | * to shmem/tmpfs;
|
| 3807 | * - The folio doesn't cross VMA boundary;
|
| 3808 | * - The folio doesn't cross page table boundary;
|
| 3809 |
|
| 3810 | addr0 = addr - start * PAGE_SIZE;
|
| 3811 | if ((file_end >= folio_next_index(folio) || shmem_mapping(mapping)) &&
|
| 3812 | folio_within_vma(folio, vmf->vma) &&
|
| 3813 | (addr0 & PMD_MASK) == ((addr0 + folio_size(folio) - 1) & PMD_MASK)) {
|
| 3814 | vmf->pte -= start;
|
| 3815 | page -= start;
|
| 3816 | addr = addr0;
|
| 3817 | nr_pages = folio_nr_pages(folio);
|
| 3818 | }
|
| 3819 |
|
| 3820 | do {
|
| 3821 | if (PageHWPoison(page + count))
|
| 3822 | goto skip;
|
| 3823 |
|
| 3824 | |
| 3825 | * NOTE: If there're PTE markers, we'll leave them to be
|
| 3826 | * handled in the specific fault path, and it'll prohibit the
|
| 3827 | * fault-around logic.
|
| 3828 |
|
| 3829 | if (!pte_none(ptep_get(&vmf->pte[count])))
|
| 3830 | goto skip;
|
| 3831 |
|
| 3832 | count++;
|
| 3833 | continue;
|
| 3834 | skip:
|
| 3835 | if (count) {
|
| 3836 | set_pte_range(vmf, folio, page, count, addr);
|
| 3837 | *rss += count;
|
| 3838 | folio_ref_add(folio, count - ref_from_caller);
|
| 3839 | ref_from_caller = 0;
|
| 3840 | if (in_range(vmf->address, addr, count * PAGE_SIZE))
|
| 3841 | ret = VM_FAULT_NOPAGE;
|
| 3842 | }
|
| 3843 |
|
| 3844 | count++;
|
| 3845 | page += count;
|
| 3846 | vmf->pte += count;
|
| 3847 | addr += count * PAGE_SIZE;
|
| 3848 | count = 0;
|
| 3849 | } while (--nr_pages > 0);
|
| 3850 |
|
| 3851 | if (count) {
|
| 3852 | set_pte_range(vmf, folio, page, count, addr);
|
| 3853 | *rss += count;
|
| 3854 | folio_ref_add(folio, count - ref_from_caller);
|
| 3855 | ref_from_caller = 0;
|
| 3856 | if (in_range(vmf->address, addr, count * PAGE_SIZE))
|
| 3857 | ret = VM_FAULT_NOPAGE;
|
| 3858 | }
|
| 3859 |
|
| 3860 | vmf->pte = old_ptep;
|
| 3861 | if (ref_from_caller)
|
| 3862 |
|
| 3863 | folio_ref_dec(folio);
|
| 3864 |
|
| 3865 | return ret;
|
| 3866 | }
|
| 3867 |
|
| 3868 | static vm_fault_t filemap_map_order0_folio(struct vm_fault *vmf,
|
| 3869 | struct folio *folio, unsigned long addr,
|
| 3870 | unsigned long *rss)
|
| 3871 | {
|
| 3872 | vm_fault_t ret = 0;
|
| 3873 | struct page *page = &folio->page;
|
| 3874 |
|
| 3875 | if (PageHWPoison(page))
|
| 3876 | goto out;
|
| 3877 |
|
| 3878 | |
| 3879 | * NOTE: If there're PTE markers, we'll leave them to be
|
| 3880 | * handled in the specific fault path, and it'll prohibit
|
| 3881 | * the fault-around logic.
|
| 3882 |
|
| 3883 | if (!pte_none(ptep_get(vmf->pte)))
|
| 3884 | goto out;
|
| 3885 |
|
| 3886 | if (vmf->address == addr)
|
| 3887 | ret = VM_FAULT_NOPAGE;
|
| 3888 |
|
| 3889 | set_pte_range(vmf, folio, page, 1, addr);
|
| 3890 | (*rss)++;
|
| 3891 | return ret;
|
| 3892 |
|
| 3893 | out:
|
| 3894 |
|
| 3895 | folio_ref_dec(folio);
|
| 3896 | return ret;
|
| 3897 | }
|
| 3898 |
|
| 3899 | vm_fault_t filemap_map_pages(struct vm_fault *vmf,
|
| 3900 | pgoff_t start_pgoff, pgoff_t end_pgoff)
|
| 3901 | {
|
| 3902 | struct vm_area_struct *vma = vmf->vma;
|
| 3903 | struct file *file = vma->vm_file;
|
| 3904 | struct address_space *mapping = file->f_mapping;
|
| 3905 | pgoff_t file_end, last_pgoff = start_pgoff;
|
| 3906 | unsigned long addr;
|
| 3907 | XA_STATE(xas, &mapping->i_pages, start_pgoff);
|
| 3908 | struct folio *folio;
|
| 3909 | vm_fault_t ret = 0;
|
| 3910 | unsigned long rss = 0;
|
| 3911 | unsigned int nr_pages = 0, folio_type;
|
| 3912 |
|
| 3913 | |
| 3914 | * Recalculate end_pgoff based on file_end before calling
|
| 3915 | * next_uptodate_folio() to avoid races with concurrent
|
| 3916 | * truncation.
|
| 3917 |
|
| 3918 | file_end = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE) - 1;
|
| 3919 | end_pgoff = min(end_pgoff, file_end);
|
| 3920 |
|
| 3921 | rcu_read_lock();
|
| 3922 | folio = next_uptodate_folio(&xas, mapping, end_pgoff);
|
| 3923 | if (!folio)
|
| 3924 | goto out;
|
| 3925 |
|
| 3926 | |
| 3927 | * Do not allow to map with PMD across i_size to preserve
|
| 3928 | * SIGBUS semantics.
|
| 3929 | *
|
| 3930 | * Make an exception for shmem/tmpfs that for long time
|
| 3931 | * intentionally mapped with PMDs across i_size.
|
| 3932 |
|
| 3933 | if ((file_end >= folio_next_index(folio) || shmem_mapping(mapping)) &&
|
| 3934 | filemap_map_pmd(vmf, folio, start_pgoff)) {
|
| 3935 | ret = VM_FAULT_NOPAGE;
|
| 3936 | goto out;
|
| 3937 | }
|
| 3938 |
|
| 3939 | addr = vma->vm_start +
|
| 3940 | ((start_pgoff - vma_start_pgoff(vma)) << PAGE_SHIFT);
|
| 3941 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl);
|
| 3942 | if (!vmf->pte) {
|
| 3943 | folio_unlock(folio);
|
| 3944 | folio_put(folio);
|
| 3945 | goto out;
|
| 3946 | }
|
| 3947 |
|
| 3948 | folio_type = mm_counter_file(folio);
|
| 3949 | do {
|
| 3950 | unsigned long end;
|
| 3951 | vm_fault_t map_ret;
|
| 3952 |
|
| 3953 | addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
|
| 3954 | vmf->pte += xas.xa_index - last_pgoff;
|
| 3955 | last_pgoff = xas.xa_index;
|
| 3956 | end = folio_next_index(folio) - 1;
|
| 3957 | nr_pages = min(end, end_pgoff) - xas.xa_index + 1;
|
| 3958 |
|
| 3959 | if (!folio_test_large(folio)) {
|
| 3960 | map_ret = filemap_map_order0_folio(vmf, folio, addr,
|
| 3961 | &rss);
|
| 3962 | } else {
|
| 3963 | unsigned long start = xas.xa_index - folio->index;
|
| 3964 |
|
| 3965 | map_ret = filemap_map_folio_range(vmf, folio, start,
|
| 3966 | addr, nr_pages, &rss,
|
| 3967 | file_end);
|
| 3968 | }
|
| 3969 | ret |= map_ret;
|
| 3970 |
|
| 3971 | |
| 3972 | * If there are too many folios that are recently evicted
|
| 3973 | * in a file, they will probably continue to be evicted.
|
| 3974 | * In such situation, read-ahead is only a waste of IO.
|
| 3975 | * Don't decrease mmap_miss in this scenario to make sure
|
| 3976 | * we can stop read-ahead.
|
| 3977 | *
|
| 3978 | * VM_SEQ_READ and VM_EXEC mappings skip the mmap_miss
|
| 3979 | * increment in do_sync_mmap_readahead(), so skip the
|
| 3980 | * decrement here as well to keep the counter symmetric.
|
| 3981 |
|
| 3982 | if ((map_ret & VM_FAULT_NOPAGE) &&
|
| 3983 | !(vmf->flags & FAULT_FLAG_TRIED) &&
|
| 3984 | !folio_test_workingset(folio) &&
|
| 3985 | !(vma->vm_flags & (VM_SEQ_READ | VM_EXEC))) {
|
| 3986 | unsigned short mmap_miss;
|
| 3987 |
|
| 3988 | mmap_miss = READ_ONCE(file->f_ra.mmap_miss);
|
| 3989 | if (mmap_miss)
|
| 3990 | WRITE_ONCE(file->f_ra.mmap_miss,
|
| 3991 | mmap_miss - 1);
|
| 3992 | }
|
| 3993 |
|
| 3994 | folio_unlock(folio);
|
| 3995 | } while ((folio = next_uptodate_folio(&xas, mapping, end_pgoff)) != NULL);
|
| 3996 | add_mm_counter(vma->vm_mm, folio_type, rss);
|
| 3997 | pte_unmap_unlock(vmf->pte, vmf->ptl);
|
| 3998 | trace_mm_filemap_map_pages(mapping, start_pgoff, end_pgoff);
|
| 3999 | out:
|
| 4000 | rcu_read_unlock();
|
| 4001 |
|
| 4002 | return ret;
|
| 4003 | }
|
| 4004 | EXPORT_SYMBOL(filemap_map_pages);
|
| 4005 |
|
| 4006 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
|
| 4007 | {
|
| 4008 | struct address_space *mapping = vmf->vma->vm_file->f_mapping;
|
| 4009 | struct folio *folio = page_folio(vmf->page);
|
| 4010 | vm_fault_t ret = VM_FAULT_LOCKED;
|
| 4011 |
|
| 4012 | sb_start_pagefault(mapping->host->i_sb);
|
| 4013 | file_update_time(vmf->vma->vm_file);
|
| 4014 | folio_lock(folio);
|
| 4015 | if (folio->mapping != mapping) {
|
| 4016 | folio_unlock(folio);
|
| 4017 | ret = VM_FAULT_NOPAGE;
|
| 4018 | goto out;
|
| 4019 | }
|
| 4020 | |
| 4021 | * We mark the folio dirty already here so that when freeze is in
|
| 4022 | * progress, we are guaranteed that writeback during freezing will
|
| 4023 | * see the dirty folio and writeprotect it again.
|
| 4024 |
|
| 4025 | folio_mark_dirty(folio);
|
| 4026 | folio_wait_stable(folio);
|
| 4027 | out:
|
| 4028 | sb_end_pagefault(mapping->host->i_sb);
|
| 4029 | return ret;
|
| 4030 | }
|
| 4031 |
|
| 4032 | const struct vm_operations_struct generic_file_vm_ops = {
|
| 4033 | .fault = filemap_fault,
|
| 4034 | .map_pages = filemap_map_pages,
|
| 4035 | .page_mkwrite = filemap_page_mkwrite,
|
| 4036 | };
|
| 4037 |
|
| 4038 |
|
| 4039 |
|
| 4040 | int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
|
| 4041 | {
|
| 4042 | struct address_space *mapping = file->f_mapping;
|
| 4043 |
|
| 4044 | if (!mapping->a_ops->read_folio)
|
| 4045 | return -ENOEXEC;
|
| 4046 | file_accessed(file);
|
| 4047 | vma->vm_ops = &generic_file_vm_ops;
|
| 4048 | return 0;
|
| 4049 | }
|
| 4050 |
|
| 4051 | int generic_file_mmap_prepare(struct vm_area_desc *desc)
|
| 4052 | {
|
| 4053 | struct file *file = desc->file;
|
| 4054 | struct address_space *mapping = file->f_mapping;
|
| 4055 |
|
| 4056 | if (!mapping->a_ops->read_folio)
|
| 4057 | return -ENOEXEC;
|
| 4058 | file_accessed(file);
|
| 4059 | desc->vm_ops = &generic_file_vm_ops;
|
| 4060 | return 0;
|
| 4061 | }
|
| 4062 |
|
| 4063 | |
| 4064 | * This is for filesystems which do not implement ->writepage.
|
| 4065 |
|
| 4066 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
|
| 4067 | {
|
| 4068 | if (vma_is_shared_maywrite(vma))
|
| 4069 | return -EINVAL;
|
| 4070 | return generic_file_mmap(file, vma);
|
| 4071 | }
|
| 4072 |
|
| 4073 | int generic_file_readonly_mmap_prepare(struct vm_area_desc *desc)
|
| 4074 | {
|
| 4075 | if (is_shared_maywrite(&desc->vma_flags))
|
| 4076 | return -EINVAL;
|
| 4077 | return generic_file_mmap_prepare(desc);
|
| 4078 | }
|
| 4079 | #else
|
| 4080 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
|
| 4081 | {
|
| 4082 | return VM_FAULT_SIGBUS;
|
| 4083 | }
|
| 4084 | int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
|
| 4085 | {
|
| 4086 | return -ENOSYS;
|
| 4087 | }
|
| 4088 | int generic_file_mmap_prepare(struct vm_area_desc *desc)
|
| 4089 | {
|
| 4090 | return -ENOSYS;
|
| 4091 | }
|
| 4092 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
|
| 4093 | {
|
| 4094 | return -ENOSYS;
|
| 4095 | }
|
| 4096 | int generic_file_readonly_mmap_prepare(struct vm_area_desc *desc)
|
| 4097 | {
|
| 4098 | return -ENOSYS;
|
| 4099 | }
|
| 4100 | #endif
|
| 4101 |
|
| 4102 | EXPORT_SYMBOL(filemap_page_mkwrite);
|
| 4103 | EXPORT_SYMBOL(generic_file_mmap);
|
| 4104 | EXPORT_SYMBOL(generic_file_mmap_prepare);
|
| 4105 | EXPORT_SYMBOL(generic_file_readonly_mmap);
|
| 4106 | EXPORT_SYMBOL(generic_file_readonly_mmap_prepare);
|
| 4107 |
|
| 4108 | static struct folio *do_read_cache_folio(struct address_space *mapping,
|
| 4109 | pgoff_t index, filler_t filler, struct file *file, gfp_t gfp)
|
| 4110 | {
|
| 4111 | struct folio *folio;
|
| 4112 | int err;
|
| 4113 |
|
| 4114 | if (!filler)
|
| 4115 | filler = mapping->a_ops->read_folio;
|
| 4116 | repeat:
|
| 4117 | folio = filemap_get_folio(mapping, index);
|
| 4118 | if (IS_ERR(folio)) {
|
| 4119 | folio = filemap_alloc_folio(gfp, mapping_min_folio_order(mapping), NULL);
|
| 4120 | if (!folio)
|
| 4121 | return ERR_PTR(-ENOMEM);
|
| 4122 | index = mapping_align_index(mapping, index);
|
| 4123 | err = filemap_add_folio(mapping, folio, index, gfp);
|
| 4124 | if (unlikely(err)) {
|
| 4125 | folio_put(folio);
|
| 4126 | if (err == -EEXIST)
|
| 4127 | goto repeat;
|
| 4128 |
|
| 4129 | return ERR_PTR(err);
|
| 4130 | }
|
| 4131 |
|
| 4132 | goto filler;
|
| 4133 | }
|
| 4134 | if (folio_test_uptodate(folio))
|
| 4135 | goto out;
|
| 4136 |
|
| 4137 | if (!folio_trylock(folio)) {
|
| 4138 | folio_put_wait_locked(folio, TASK_UNINTERRUPTIBLE);
|
| 4139 | goto repeat;
|
| 4140 | }
|
| 4141 |
|
| 4142 |
|
| 4143 | if (!folio->mapping) {
|
| 4144 | folio_unlock(folio);
|
| 4145 | folio_put(folio);
|
| 4146 | goto repeat;
|
| 4147 | }
|
| 4148 |
|
| 4149 |
|
| 4150 | if (folio_test_uptodate(folio)) {
|
| 4151 | folio_unlock(folio);
|
| 4152 | goto out;
|
| 4153 | }
|
| 4154 |
|
| 4155 | filler:
|
| 4156 | err = filemap_read_folio(file, filler, folio);
|
| 4157 | if (err) {
|
| 4158 | folio_put(folio);
|
| 4159 | if (err == AOP_TRUNCATED_PAGE)
|
| 4160 | goto repeat;
|
| 4161 | return ERR_PTR(err);
|
| 4162 | }
|
| 4163 |
|
| 4164 | out:
|
| 4165 | folio_mark_accessed(folio);
|
| 4166 | return folio;
|
| 4167 | }
|
| 4168 |
|
| 4169 | |
| 4170 | * read_cache_folio - Read into page cache, fill it if needed.
|
| 4171 | * @mapping: The address_space to read from.
|
| 4172 | * @index: The index to read.
|
| 4173 | * @filler: Function to perform the read, or NULL to use aops->read_folio().
|
| 4174 | * @file: Passed to filler function, may be NULL if not required.
|
| 4175 | *
|
| 4176 | * Read one page into the page cache. If it succeeds, the folio returned
|
| 4177 | * will contain @index, but it may not be the first page of the folio.
|
| 4178 | *
|
| 4179 | * If the filler function returns an error, it will be returned to the
|
| 4180 | * caller.
|
| 4181 | *
|
| 4182 | * Context: May sleep. Expects mapping->invalidate_lock to be held.
|
| 4183 | * Return: An uptodate folio on success, ERR_PTR() on failure.
|
| 4184 |
|
| 4185 | struct folio *read_cache_folio(struct address_space *mapping, pgoff_t index,
|
| 4186 | filler_t filler, struct file *file)
|
| 4187 | {
|
| 4188 | return do_read_cache_folio(mapping, index, filler, file,
|
| 4189 | mapping_gfp_mask(mapping));
|
| 4190 | }
|
| 4191 | EXPORT_SYMBOL(read_cache_folio);
|
| 4192 |
|
| 4193 | |
| 4194 | * mapping_read_folio_gfp - Read into page cache, using specified allocation flags.
|
| 4195 | * @mapping: The address_space for the folio.
|
| 4196 | * @index: The index that the allocated folio will contain.
|
| 4197 | * @gfp: The page allocator flags to use if allocating.
|
| 4198 | *
|
| 4199 | * This is the same as "read_cache_folio(mapping, index, NULL, NULL)", but with
|
| 4200 | * any new memory allocations done using the specified allocation flags.
|
| 4201 | *
|
| 4202 | * The most likely error from this function is EIO, but ENOMEM is
|
| 4203 | * possible and so is EINTR. If ->read_folio returns another error,
|
| 4204 | * that will be returned to the caller.
|
| 4205 | *
|
| 4206 | * The function expects mapping->invalidate_lock to be already held.
|
| 4207 | *
|
| 4208 | * Return: Uptodate folio on success, ERR_PTR() on failure.
|
| 4209 |
|
| 4210 | struct folio *mapping_read_folio_gfp(struct address_space *mapping,
|
| 4211 | pgoff_t index, gfp_t gfp)
|
| 4212 | {
|
| 4213 | return do_read_cache_folio(mapping, index, NULL, NULL, gfp);
|
| 4214 | }
|
| 4215 | EXPORT_SYMBOL(mapping_read_folio_gfp);
|
| 4216 |
|
| 4217 | static struct page *do_read_cache_page(struct address_space *mapping,
|
| 4218 | pgoff_t index, filler_t *filler, struct file *file, gfp_t gfp)
|
| 4219 | {
|
| 4220 | struct folio *folio;
|
| 4221 |
|
| 4222 | folio = do_read_cache_folio(mapping, index, filler, file, gfp);
|
| 4223 | if (IS_ERR(folio))
|
| 4224 | return &folio->page;
|
| 4225 | return folio_file_page(folio, index);
|
| 4226 | }
|
| 4227 |
|
| 4228 | struct page *read_cache_page(struct address_space *mapping,
|
| 4229 | pgoff_t index, filler_t *filler, struct file *file)
|
| 4230 | {
|
| 4231 | return do_read_cache_page(mapping, index, filler, file,
|
| 4232 | mapping_gfp_mask(mapping));
|
| 4233 | }
|
| 4234 | EXPORT_SYMBOL(read_cache_page);
|
| 4235 |
|
| 4236 | |
| 4237 | * read_cache_page_gfp - read into page cache, using specified page allocation flags.
|
| 4238 | * @mapping: the page's address_space
|
| 4239 | * @index: the page index
|
| 4240 | * @gfp: the page allocator flags to use if allocating
|
| 4241 | *
|
| 4242 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with
|
| 4243 | * any new page allocations done using the specified allocation flags.
|
| 4244 | *
|
| 4245 | * If the page does not get brought uptodate, return -EIO.
|
| 4246 | *
|
| 4247 | * The function expects mapping->invalidate_lock to be already held.
|
| 4248 | *
|
| 4249 | * Return: up to date page on success, ERR_PTR() on failure.
|
| 4250 |
|
| 4251 | struct page *read_cache_page_gfp(struct address_space *mapping,
|
| 4252 | pgoff_t index,
|
| 4253 | gfp_t gfp)
|
| 4254 | {
|
| 4255 | return do_read_cache_page(mapping, index, NULL, NULL, gfp);
|
| 4256 | }
|
| 4257 | EXPORT_SYMBOL(read_cache_page_gfp);
|
| 4258 |
|
| 4259 | |
| 4260 | * Warn about a page cache invalidation failure during a direct I/O write.
|
| 4261 |
|
| 4262 | static void dio_warn_stale_pagecache(struct file *filp)
|
| 4263 | {
|
| 4264 | static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
|
| 4265 | char pathname[128];
|
| 4266 | char *path;
|
| 4267 |
|
| 4268 | errseq_set(&filp->f_mapping->wb_err, -EIO);
|
| 4269 | if (__ratelimit(&_rs)) {
|
| 4270 | path = file_path(filp, pathname, sizeof(pathname));
|
| 4271 | if (IS_ERR(path))
|
| 4272 | path = "(unknown)";
|
| 4273 | pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n");
|
| 4274 | pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
|
| 4275 | current->comm);
|
| 4276 | }
|
| 4277 | }
|
| 4278 |
|
| 4279 | void kiocb_invalidate_post_direct_write(struct kiocb *iocb, size_t count)
|
| 4280 | {
|
| 4281 | struct address_space *mapping = iocb->ki_filp->f_mapping;
|
| 4282 |
|
| 4283 | if (mapping->nrpages &&
|
| 4284 | invalidate_inode_pages2_range(mapping,
|
| 4285 | iocb->ki_pos >> PAGE_SHIFT,
|
| 4286 | (iocb->ki_pos + count - 1) >> PAGE_SHIFT))
|
| 4287 | dio_warn_stale_pagecache(iocb->ki_filp);
|
| 4288 | }
|
| 4289 |
|
| 4290 | ssize_t
|
| 4291 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
|
| 4292 | {
|
| 4293 | struct address_space *mapping = iocb->ki_filp->f_mapping;
|
| 4294 | size_t write_len = iov_iter_count(from);
|
| 4295 | ssize_t written;
|
| 4296 |
|
| 4297 | |
| 4298 | * If a page can not be invalidated, return 0 to fall back
|
| 4299 | * to buffered write.
|
| 4300 |
|
| 4301 | written = kiocb_invalidate_pages(iocb, write_len);
|
| 4302 | if (written) {
|
| 4303 | if (written == -EBUSY)
|
| 4304 | return 0;
|
| 4305 | return written;
|
| 4306 | }
|
| 4307 |
|
| 4308 | written = mapping->a_ops->direct_IO(iocb, from);
|
| 4309 |
|
| 4310 | |
| 4311 | * Finally, try again to invalidate clean pages which might have been
|
| 4312 | * cached by non-direct readahead, or faulted in by get_user_pages()
|
| 4313 | * if the source of the write was an mmap'ed region of the file
|
| 4314 | * we're writing. Either one is a pretty crazy thing to do,
|
| 4315 | * so we don't support it 100%. If this invalidation
|
| 4316 | * fails, tough, the write still worked...
|
| 4317 | *
|
| 4318 | * Most of the time we do not need this since dio_complete() will do
|
| 4319 | * the invalidation for us. However there are some file systems that
|
| 4320 | * do not end up with dio_complete() being called, so let's not break
|
| 4321 | * them by removing it completely.
|
| 4322 | *
|
| 4323 | * Noticeable example is a blkdev_direct_IO().
|
| 4324 | *
|
| 4325 | * Skip invalidation for async writes or if mapping has no pages.
|
| 4326 |
|
| 4327 | if (written > 0) {
|
| 4328 | struct inode *inode = mapping->host;
|
| 4329 | loff_t pos = iocb->ki_pos;
|
| 4330 |
|
| 4331 | kiocb_invalidate_post_direct_write(iocb, written);
|
| 4332 | pos += written;
|
| 4333 | write_len -= written;
|
| 4334 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
|
| 4335 | i_size_write(inode, pos);
|
| 4336 | mark_inode_dirty(inode);
|
| 4337 | }
|
| 4338 | iocb->ki_pos = pos;
|
| 4339 | }
|
| 4340 | if (written != -EIOCBQUEUED)
|
| 4341 | iov_iter_revert(from, write_len - iov_iter_count(from));
|
| 4342 | return written;
|
| 4343 | }
|
| 4344 | EXPORT_SYMBOL(generic_file_direct_write);
|
| 4345 |
|
| 4346 | ssize_t generic_perform_write(struct kiocb *iocb, struct iov_iter *i)
|
| 4347 | {
|
| 4348 | struct file *file = iocb->ki_filp;
|
| 4349 | loff_t pos = iocb->ki_pos;
|
| 4350 | struct address_space *mapping = file->f_mapping;
|
| 4351 | const struct address_space_operations *a_ops = mapping->a_ops;
|
| 4352 | size_t chunk = mapping_max_folio_size(mapping);
|
| 4353 | long status = 0;
|
| 4354 | ssize_t written = 0;
|
| 4355 |
|
| 4356 | do {
|
| 4357 | struct folio *folio;
|
| 4358 | size_t offset;
|
| 4359 | size_t bytes;
|
| 4360 | size_t copied;
|
| 4361 | void *fsdata = NULL;
|
| 4362 |
|
| 4363 | bytes = iov_iter_count(i);
|
| 4364 | retry:
|
| 4365 | offset = pos & (chunk - 1);
|
| 4366 | bytes = min(chunk - offset, bytes);
|
| 4367 | balance_dirty_pages_ratelimited(mapping);
|
| 4368 |
|
| 4369 | if (fatal_signal_pending(current)) {
|
| 4370 | status = -EINTR;
|
| 4371 | break;
|
| 4372 | }
|
| 4373 |
|
| 4374 | status = a_ops->write_begin(iocb, mapping, pos, bytes,
|
| 4375 | &folio, &fsdata);
|
| 4376 | if (unlikely(status < 0))
|
| 4377 | break;
|
| 4378 |
|
| 4379 | offset = offset_in_folio(folio, pos);
|
| 4380 | if (bytes > folio_size(folio) - offset)
|
| 4381 | bytes = folio_size(folio) - offset;
|
| 4382 |
|
| 4383 | if (mapping_writably_mapped(mapping))
|
| 4384 | flush_dcache_folio(folio);
|
| 4385 |
|
| 4386 | |
| 4387 | * Faults here on mmap()s can recurse into arbitrary
|
| 4388 | * filesystem code. Lots of locks are held that can
|
| 4389 | * deadlock. Use an atomic copy to avoid deadlocking
|
| 4390 | * in page fault handling.
|
| 4391 |
|
| 4392 | copied = copy_folio_from_iter_atomic(folio, offset, bytes, i);
|
| 4393 | flush_dcache_folio(folio);
|
| 4394 |
|
| 4395 | status = a_ops->write_end(iocb, mapping, pos, bytes, copied,
|
| 4396 | folio, fsdata);
|
| 4397 | if (unlikely(status != copied)) {
|
| 4398 | iov_iter_revert(i, copied - max(status, 0L));
|
| 4399 | if (unlikely(status < 0))
|
| 4400 | break;
|
| 4401 | }
|
| 4402 | cond_resched();
|
| 4403 |
|
| 4404 | if (unlikely(status == 0)) {
|
| 4405 | |
| 4406 | * A short copy made ->write_end() reject the
|
| 4407 | * thing entirely. Might be memory poisoning
|
| 4408 | * halfway through, might be a race with munmap,
|
| 4409 | * might be severe memory pressure.
|
| 4410 |
|
| 4411 | if (chunk > PAGE_SIZE)
|
| 4412 | chunk /= 2;
|
| 4413 | if (copied) {
|
| 4414 | bytes = copied;
|
| 4415 | goto retry;
|
| 4416 | }
|
| 4417 |
|
| 4418 | |
| 4419 | * 'folio' is now unlocked and faults on it can be
|
| 4420 | * handled. Ensure forward progress by trying to
|
| 4421 | * fault it in now.
|
| 4422 |
|
| 4423 | if (fault_in_iov_iter_readable(i, bytes) == bytes) {
|
| 4424 | status = -EFAULT;
|
| 4425 | break;
|
| 4426 | }
|
| 4427 | } else {
|
| 4428 | pos += status;
|
| 4429 | written += status;
|
| 4430 | }
|
| 4431 | } while (iov_iter_count(i));
|
| 4432 |
|
| 4433 | if (!written)
|
| 4434 | return status;
|
| 4435 | iocb->ki_pos += written;
|
| 4436 | return written;
|
| 4437 | }
|
| 4438 | EXPORT_SYMBOL(generic_perform_write);
|
| 4439 |
|
| 4440 | |
| 4441 | * __generic_file_write_iter - write data to a file
|
| 4442 | * @iocb: IO state structure (file, offset, etc.)
|
| 4443 | * @from: iov_iter with data to write
|
| 4444 | *
|
| 4445 | * This function does all the work needed for actually writing data to a
|
| 4446 | * file. It does all basic checks, removes SUID from the file, updates
|
| 4447 | * modification times and calls proper subroutines depending on whether we
|
| 4448 | * do direct IO or a standard buffered write.
|
| 4449 | *
|
| 4450 | * It expects i_rwsem to be grabbed unless we work on a block device or similar
|
| 4451 | * object which does not need locking at all.
|
| 4452 | *
|
| 4453 | * This function does *not* take care of syncing data in case of O_SYNC write.
|
| 4454 | * A caller has to handle it. This is mainly due to the fact that we want to
|
| 4455 | * avoid syncing under i_rwsem.
|
| 4456 | *
|
| 4457 | * Return:
|
| 4458 | * * number of bytes written, even for truncated writes
|
| 4459 | * * negative error code if no data has been written at all
|
| 4460 |
|
| 4461 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
|
| 4462 | {
|
| 4463 | struct file *file = iocb->ki_filp;
|
| 4464 | struct address_space *mapping = file->f_mapping;
|
| 4465 | struct inode *inode = mapping->host;
|
| 4466 | ssize_t ret;
|
| 4467 |
|
| 4468 | ret = file_remove_privs(file);
|
| 4469 | if (ret)
|
| 4470 | return ret;
|
| 4471 |
|
| 4472 | ret = file_update_time(file);
|
| 4473 | if (ret)
|
| 4474 | return ret;
|
| 4475 |
|
| 4476 | if (iocb->ki_flags & IOCB_DIRECT) {
|
| 4477 | ret = generic_file_direct_write(iocb, from);
|
| 4478 | |
| 4479 | * If the write stopped short of completing, fall back to
|
| 4480 | * buffered writes. Some filesystems do this for writes to
|
| 4481 | * holes, for example. For DAX files, a buffered write will
|
| 4482 | * not succeed (even if it did, DAX does not handle dirty
|
| 4483 | * page-cache pages correctly).
|
| 4484 |
|
| 4485 | if (ret < 0 || !iov_iter_count(from) || IS_DAX(inode))
|
| 4486 | return ret;
|
| 4487 | return direct_write_fallback(iocb, from, ret,
|
| 4488 | generic_perform_write(iocb, from));
|
| 4489 | }
|
| 4490 |
|
| 4491 | return generic_perform_write(iocb, from);
|
| 4492 | }
|
| 4493 | EXPORT_SYMBOL(__generic_file_write_iter);
|
| 4494 |
|
| 4495 | |
| 4496 | * generic_file_write_iter - write data to a file
|
| 4497 | * @iocb: IO state structure
|
| 4498 | * @from: iov_iter with data to write
|
| 4499 | *
|
| 4500 | * This is a wrapper around __generic_file_write_iter() to be used by most
|
| 4501 | * filesystems. It takes care of syncing the file in case of O_SYNC file
|
| 4502 | * and acquires i_rwsem as needed.
|
| 4503 | * Return:
|
| 4504 | * * negative error code if no data has been written at all of
|
| 4505 | * vfs_fsync_range() failed for a synchronous write
|
| 4506 | * * number of bytes written, even for truncated writes
|
| 4507 |
|
| 4508 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
|
| 4509 | {
|
| 4510 | struct file *file = iocb->ki_filp;
|
| 4511 | struct inode *inode = file->f_mapping->host;
|
| 4512 | ssize_t ret;
|
| 4513 |
|
| 4514 | inode_lock(inode);
|
| 4515 | ret = generic_write_checks(iocb, from);
|
| 4516 | if (ret > 0)
|
| 4517 | ret = __generic_file_write_iter(iocb, from);
|
| 4518 | inode_unlock(inode);
|
| 4519 |
|
| 4520 | if (ret > 0)
|
| 4521 | ret = generic_write_sync(iocb, ret);
|
| 4522 | return ret;
|
| 4523 | }
|
| 4524 | EXPORT_SYMBOL(generic_file_write_iter);
|
| 4525 |
|
| 4526 | |
| 4527 | * filemap_release_folio() - Release fs-specific metadata on a folio.
|
| 4528 | * @folio: The folio which the kernel is trying to free.
|
| 4529 | * @gfp: Memory allocation flags (and I/O mode).
|
| 4530 | *
|
| 4531 | * The address_space is trying to release any data attached to a folio
|
| 4532 | * (presumably at folio->private).
|
| 4533 | *
|
| 4534 | * This will also be called if the private_2 flag is set on a page,
|
| 4535 | * indicating that the folio has other metadata associated with it.
|
| 4536 | *
|
| 4537 | * The @gfp argument specifies whether I/O may be performed to release
|
| 4538 | * this page (__GFP_IO), and whether the call may block
|
| 4539 | * (__GFP_RECLAIM & __GFP_FS).
|
| 4540 | *
|
| 4541 | * Return: %true if the release was successful, otherwise %false.
|
| 4542 |
|
| 4543 | bool filemap_release_folio(struct folio *folio, gfp_t gfp)
|
| 4544 | {
|
| 4545 | struct address_space * const mapping = folio->mapping;
|
| 4546 |
|
| 4547 | BUG_ON(!folio_test_locked(folio));
|
| 4548 | if (!folio_needs_release(folio))
|
| 4549 | return true;
|
| 4550 | if (folio_test_writeback(folio))
|
| 4551 | return false;
|
| 4552 |
|
| 4553 | if (mapping && mapping->a_ops->release_folio)
|
| 4554 | return mapping->a_ops->release_folio(folio, gfp);
|
| 4555 | return try_to_free_buffers(folio);
|
| 4556 | }
|
| 4557 | EXPORT_SYMBOL(filemap_release_folio);
|
| 4558 |
|
| 4559 | |
| 4560 | * filemap_invalidate_inode - Invalidate/forcibly write back a range of an inode's pagecache
|
| 4561 | * @inode: The inode to flush
|
| 4562 | * @flush: Set to write back rather than simply invalidate.
|
| 4563 | * @start: First byte to in range.
|
| 4564 | * @end: Last byte in range (inclusive), or LLONG_MAX for everything from start
|
| 4565 | * onwards.
|
| 4566 | *
|
| 4567 | * Invalidate all the folios on an inode that contribute to the specified
|
| 4568 | * range, possibly writing them back first. Whilst the operation is
|
| 4569 | * undertaken, the invalidate lock is held to prevent new folios from being
|
| 4570 | * installed.
|
| 4571 |
|
| 4572 | int filemap_invalidate_inode(struct inode *inode, bool flush,
|
| 4573 | loff_t start, loff_t end)
|
| 4574 | {
|
| 4575 | struct address_space *mapping = inode->i_mapping;
|
| 4576 | pgoff_t first = start >> PAGE_SHIFT;
|
| 4577 | pgoff_t last = end >> PAGE_SHIFT;
|
| 4578 | pgoff_t nr = end == LLONG_MAX ? ULONG_MAX : last - first + 1;
|
| 4579 |
|
| 4580 | if (!mapping || !mapping->nrpages || end < start)
|
| 4581 | goto out;
|
| 4582 |
|
| 4583 |
|
| 4584 | filemap_invalidate_lock(mapping);
|
| 4585 |
|
| 4586 | if (!mapping->nrpages)
|
| 4587 | goto unlock;
|
| 4588 |
|
| 4589 | unmap_mapping_pages(mapping, first, nr, false);
|
| 4590 |
|
| 4591 |
|
| 4592 | if (flush)
|
| 4593 | filemap_fdatawrite_range(mapping, start, end);
|
| 4594 |
|
| 4595 |
|
| 4596 | invalidate_inode_pages2_range(mapping, start / PAGE_SIZE, end / PAGE_SIZE);
|
| 4597 |
|
| 4598 | unlock:
|
| 4599 | filemap_invalidate_unlock(mapping);
|
| 4600 | out:
|
| 4601 | return filemap_check_errors(mapping);
|
| 4602 | }
|
| 4603 | EXPORT_SYMBOL_GPL(filemap_invalidate_inode);
|
| 4604 |
|
| 4605 | #ifdef CONFIG_CACHESTAT_SYSCALL
|
| 4606 | |
| 4607 | * filemap_cachestat() - compute the page cache statistics of a mapping
|
| 4608 | * @mapping: The mapping to compute the statistics for.
|
| 4609 | * @first_index: The starting page cache index.
|
| 4610 | * @last_index: The final page index (inclusive).
|
| 4611 | * @cs: the cachestat struct to write the result to.
|
| 4612 | *
|
| 4613 | * This will query the page cache statistics of a mapping in the
|
| 4614 | * page range of [first_index, last_index] (inclusive). The statistics
|
| 4615 | * queried include: number of dirty pages, number of pages marked for
|
| 4616 | * writeback, and the number of (recently) evicted pages.
|
| 4617 |
|
| 4618 | static void filemap_cachestat(struct address_space *mapping,
|
| 4619 | pgoff_t first_index, pgoff_t last_index, struct cachestat *cs)
|
| 4620 | {
|
| 4621 | XA_STATE(xas, &mapping->i_pages, first_index);
|
| 4622 | struct folio *folio;
|
| 4623 |
|
| 4624 |
|
| 4625 | mem_cgroup_flush_stats_ratelimited(NULL);
|
| 4626 |
|
| 4627 | rcu_read_lock();
|
| 4628 | xas_for_each(&xas, folio, last_index) {
|
| 4629 | int order;
|
| 4630 | unsigned long nr_pages;
|
| 4631 | pgoff_t folio_first_index, folio_last_index;
|
| 4632 |
|
| 4633 | |
| 4634 | * Don't deref the folio. It is not pinned, and might
|
| 4635 | * get freed (and reused) underneath us.
|
| 4636 | *
|
| 4637 | * We *could* pin it, but that would be expensive for
|
| 4638 | * what should be a fast and lightweight syscall.
|
| 4639 | *
|
| 4640 | * Instead, derive all information of interest from
|
| 4641 | * the rcu-protected xarray.
|
| 4642 |
|
| 4643 |
|
| 4644 | if (xas_retry(&xas, folio))
|
| 4645 | continue;
|
| 4646 |
|
| 4647 | order = xas_get_order(&xas);
|
| 4648 | nr_pages = 1 << order;
|
| 4649 | folio_first_index = round_down(xas.xa_index, 1 << order);
|
| 4650 | folio_last_index = folio_first_index + nr_pages - 1;
|
| 4651 |
|
| 4652 |
|
| 4653 | if (folio_first_index < first_index)
|
| 4654 | nr_pages -= first_index - folio_first_index;
|
| 4655 |
|
| 4656 | if (folio_last_index > last_index)
|
| 4657 | nr_pages -= folio_last_index - last_index;
|
| 4658 |
|
| 4659 | if (xa_is_value(folio)) {
|
| 4660 |
|
| 4661 | void *shadow = (void *)folio;
|
| 4662 | bool workingset;
|
| 4663 |
|
| 4664 | cs->nr_evicted += nr_pages;
|
| 4665 |
|
| 4666 | #ifdef CONFIG_SWAP
|
| 4667 | if (shmem_mapping(mapping)) {
|
| 4668 |
|
| 4669 | swp_entry_t swp = radix_to_swp_entry(folio);
|
| 4670 |
|
| 4671 |
|
| 4672 | if (!softleaf_is_swap(swp))
|
| 4673 | goto resched;
|
| 4674 |
|
| 4675 | |
| 4676 | * Getting a swap entry from the shmem
|
| 4677 | * inode means we beat
|
| 4678 | * shmem_unuse(). rcu_read_lock()
|
| 4679 | * ensures swapoff waits for us before
|
| 4680 | * freeing the swapper space. However,
|
| 4681 | * we can race with swapping and
|
| 4682 | * invalidation, so there might not be
|
| 4683 | * a shadow in the swapcache (yet).
|
| 4684 |
|
| 4685 | shadow = swap_cache_get_shadow(swp);
|
| 4686 | if (!shadow)
|
| 4687 | goto resched;
|
| 4688 | }
|
| 4689 | #endif
|
| 4690 | if (workingset_test_recent(shadow, true, &workingset, false))
|
| 4691 | cs->nr_recently_evicted += nr_pages;
|
| 4692 |
|
| 4693 | goto resched;
|
| 4694 | }
|
| 4695 |
|
| 4696 |
|
| 4697 | cs->nr_cache += nr_pages;
|
| 4698 |
|
| 4699 | if (xas_get_mark(&xas, PAGECACHE_TAG_DIRTY))
|
| 4700 | cs->nr_dirty += nr_pages;
|
| 4701 |
|
| 4702 | if (xas_get_mark(&xas, PAGECACHE_TAG_WRITEBACK))
|
| 4703 | cs->nr_writeback += nr_pages;
|
| 4704 |
|
| 4705 | resched:
|
| 4706 | if (need_resched()) {
|
| 4707 | xas_pause(&xas);
|
| 4708 | cond_resched_rcu();
|
| 4709 | }
|
| 4710 | }
|
| 4711 | rcu_read_unlock();
|
| 4712 | }
|
| 4713 |
|
| 4714 | |
| 4715 | * See mincore: reveal pagecache information only for files
|
| 4716 | * that the calling process has write access to, or could (if
|
| 4717 | * tried) open for writing.
|
| 4718 |
|
| 4719 | static inline bool can_do_cachestat(struct file *f)
|
| 4720 | {
|
| 4721 | if (f->f_mode & FMODE_WRITE)
|
| 4722 | return true;
|
| 4723 | if (file_owner_or_capable(f))
|
| 4724 | return true;
|
| 4725 | return file_permission(f, MAY_WRITE) == 0;
|
| 4726 | }
|
| 4727 |
|
| 4728 | |
| 4729 | * The cachestat(2) system call.
|
| 4730 | *
|
| 4731 | * cachestat() returns the page cache statistics of a file in the
|
| 4732 | * bytes range specified by `off` and `len`: number of cached pages,
|
| 4733 | * number of dirty pages, number of pages marked for writeback,
|
| 4734 | * number of evicted pages, and number of recently evicted pages.
|
| 4735 | *
|
| 4736 | * An evicted page is a page that is previously in the page cache
|
| 4737 | * but has been evicted since. A page is recently evicted if its last
|
| 4738 | * eviction was recent enough that its reentry to the cache would
|
| 4739 | * indicate that it is actively being used by the system, and that
|
| 4740 | * there is memory pressure on the system.
|
| 4741 | *
|
| 4742 | * `off` and `len` must be non-negative integers. If `len` > 0,
|
| 4743 | * the queried range is [`off`, `off` + `len`]. If `len` == 0,
|
| 4744 | * we will query in the range from `off` to the end of the file.
|
| 4745 | *
|
| 4746 | * The `flags` argument is unused for now, but is included for future
|
| 4747 | * extensibility. User should pass 0 (i.e no flag specified).
|
| 4748 | *
|
| 4749 | * Currently, hugetlbfs is not supported.
|
| 4750 | *
|
| 4751 | * Because the status of a page can change after cachestat() checks it
|
| 4752 | * but before it returns to the application, the returned values may
|
| 4753 | * contain stale information.
|
| 4754 | *
|
| 4755 | * return values:
|
| 4756 | * zero - success
|
| 4757 | * -EFAULT - cstat or cstat_range points to an illegal address
|
| 4758 | * -EINVAL - invalid flags
|
| 4759 | * -EBADF - invalid file descriptor
|
| 4760 | * -EOPNOTSUPP - file descriptor is of a hugetlbfs file
|
| 4761 |
|
| 4762 | SYSCALL_DEFINE4(cachestat, unsigned int, fd,
|
| 4763 | struct cachestat_range __user *, cstat_range,
|
| 4764 | struct cachestat __user *, cstat, unsigned int, flags)
|
| 4765 | {
|
| 4766 | CLASS(fd, f)(fd);
|
| 4767 | struct address_space *mapping;
|
| 4768 | struct cachestat_range csr;
|
| 4769 | struct cachestat cs;
|
| 4770 | pgoff_t first_index, last_index;
|
| 4771 |
|
| 4772 | if (fd_empty(f))
|
| 4773 | return -EBADF;
|
| 4774 |
|
| 4775 | if (copy_from_user(&csr, cstat_range,
|
| 4776 | sizeof(struct cachestat_range)))
|
| 4777 | return -EFAULT;
|
| 4778 |
|
| 4779 |
|
| 4780 | if (is_file_hugepages(fd_file(f)))
|
| 4781 | return -EOPNOTSUPP;
|
| 4782 |
|
| 4783 | if (!can_do_cachestat(fd_file(f)))
|
| 4784 | return -EPERM;
|
| 4785 |
|
| 4786 | if (flags != 0)
|
| 4787 | return -EINVAL;
|
| 4788 |
|
| 4789 | first_index = csr.off >> PAGE_SHIFT;
|
| 4790 | last_index =
|
| 4791 | csr.len == 0 ? ULONG_MAX : (csr.off + csr.len - 1) >> PAGE_SHIFT;
|
| 4792 | memset(&cs, 0, sizeof(struct cachestat));
|
| 4793 | mapping = fd_file(f)->f_mapping;
|
| 4794 | filemap_cachestat(mapping, first_index, last_index, &cs);
|
| 4795 |
|
| 4796 | if (copy_to_user(cstat, &cs, sizeof(struct cachestat)))
|
| 4797 | return -EFAULT;
|
| 4798 |
|
| 4799 | return 0;
|
| 4800 | }
|
| 4801 | #endif
|