| 1 | /* |
| 2 | * linux/mm/filemap.c |
| 3 | * |
| 4 | * Copyright (C) 1994-1999 Linus Torvalds |
| 5 | */ |
| 6 | |
| 7 | /* |
| 8 | * This file handles the generic file mmap semantics used by |
| 9 | * most "normal" filesystems (but you don't /have/ to use this: |
| 10 | * the NFS filesystem used to do this differently, for example) |
| 11 | */ |
| 12 | #include <linux/export.h> |
| 13 | #include <linux/compiler.h> |
| 14 | #include <linux/fs.h> |
| 15 | #include <linux/uaccess.h> |
| 16 | #include <linux/aio.h> |
| 17 | #include <linux/capability.h> |
| 18 | #include <linux/kernel_stat.h> |
| 19 | #include <linux/gfp.h> |
| 20 | #include <linux/mm.h> |
| 21 | #include <linux/swap.h> |
| 22 | #include <linux/mman.h> |
| 23 | #include <linux/pagemap.h> |
| 24 | #include <linux/file.h> |
| 25 | #include <linux/uio.h> |
| 26 | #include <linux/hash.h> |
| 27 | #include <linux/writeback.h> |
| 28 | #include <linux/backing-dev.h> |
| 29 | #include <linux/pagevec.h> |
| 30 | #include <linux/blkdev.h> |
| 31 | #include <linux/security.h> |
| 32 | #include <linux/cpuset.h> |
| 33 | #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */ |
| 34 | #include <linux/memcontrol.h> |
| 35 | #include <linux/cleancache.h> |
| 36 | #include "internal.h" |
| 37 | |
| 38 | #ifdef CONFIG_SDP |
| 39 | #include <sdp/cache_cleanup.h> |
| 40 | #endif |
| 41 | |
| 42 | #define CREATE_TRACE_POINTS |
| 43 | #include <trace/events/filemap.h> |
| 44 | |
| 45 | /* |
| 46 | * FIXME: remove all knowledge of the buffer layer from the core VM |
| 47 | */ |
| 48 | #include <linux/buffer_head.h> /* for try_to_free_buffers */ |
| 49 | |
| 50 | #include <asm/mman.h> |
| 51 | |
| 52 | /* |
| 53 | * Shared mappings implemented 30.11.1994. It's not fully working yet, |
| 54 | * though. |
| 55 | * |
| 56 | * Shared mappings now work. 15.8.1995 Bruno. |
| 57 | * |
| 58 | * finished 'unifying' the page and buffer cache and SMP-threaded the |
| 59 | * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> |
| 60 | * |
| 61 | * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> |
| 62 | */ |
| 63 | |
| 64 | /* |
| 65 | * Lock ordering: |
| 66 | * |
| 67 | * ->i_mmap_mutex (truncate_pagecache) |
| 68 | * ->private_lock (__free_pte->__set_page_dirty_buffers) |
| 69 | * ->swap_lock (exclusive_swap_page, others) |
| 70 | * ->mapping->tree_lock |
| 71 | * |
| 72 | * ->i_mutex |
| 73 | * ->i_mmap_mutex (truncate->unmap_mapping_range) |
| 74 | * |
| 75 | * ->mmap_sem |
| 76 | * ->i_mmap_mutex |
| 77 | * ->page_table_lock or pte_lock (various, mainly in memory.c) |
| 78 | * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock) |
| 79 | * |
| 80 | * ->mmap_sem |
| 81 | * ->lock_page (access_process_vm) |
| 82 | * |
| 83 | * ->i_mutex (generic_file_buffered_write) |
| 84 | * ->mmap_sem (fault_in_pages_readable->do_page_fault) |
| 85 | * |
| 86 | * bdi->wb.list_lock |
| 87 | * sb_lock (fs/fs-writeback.c) |
| 88 | * ->mapping->tree_lock (__sync_single_inode) |
| 89 | * |
| 90 | * ->i_mmap_mutex |
| 91 | * ->anon_vma.lock (vma_adjust) |
| 92 | * |
| 93 | * ->anon_vma.lock |
| 94 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
| 95 | * |
| 96 | * ->page_table_lock or pte_lock |
| 97 | * ->swap_lock (try_to_unmap_one) |
| 98 | * ->private_lock (try_to_unmap_one) |
| 99 | * ->tree_lock (try_to_unmap_one) |
| 100 | * ->zone.lru_lock (follow_page->mark_page_accessed) |
| 101 | * ->zone.lru_lock (check_pte_range->isolate_lru_page) |
| 102 | * ->private_lock (page_remove_rmap->set_page_dirty) |
| 103 | * ->tree_lock (page_remove_rmap->set_page_dirty) |
| 104 | * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) |
| 105 | * ->inode->i_lock (page_remove_rmap->set_page_dirty) |
| 106 | * bdi.wb->list_lock (zap_pte_range->set_page_dirty) |
| 107 | * ->inode->i_lock (zap_pte_range->set_page_dirty) |
| 108 | * ->private_lock (zap_pte_range->__set_page_dirty_buffers) |
| 109 | * |
| 110 | * ->i_mmap_mutex |
| 111 | * ->tasklist_lock (memory_failure, collect_procs_ao) |
| 112 | */ |
| 113 | |
| 114 | static void page_cache_tree_delete(struct address_space *mapping, |
| 115 | struct page *page, void *shadow) |
| 116 | { |
| 117 | if (shadow) { |
| 118 | void **slot; |
| 119 | |
| 120 | slot = radix_tree_lookup_slot(&mapping->page_tree, page->index); |
| 121 | radix_tree_replace_slot(slot, shadow); |
| 122 | mapping->nrshadows++; |
| 123 | /* |
| 124 | * Make sure the nrshadows update is committed before |
| 125 | * the nrpages update so that final truncate racing |
| 126 | * with reclaim does not see both counters 0 at the |
| 127 | * same time and miss a shadow entry. |
| 128 | */ |
| 129 | smp_wmb(); |
| 130 | } else |
| 131 | radix_tree_delete(&mapping->page_tree, page->index); |
| 132 | mapping->nrpages--; |
| 133 | } |
| 134 | |
| 135 | /* |
| 136 | * Delete a page from the page cache and free it. Caller has to make |
| 137 | * sure the page is locked and that nobody else uses it - or that usage |
| 138 | * is safe. The caller must hold the mapping's tree_lock. |
| 139 | */ |
| 140 | void __delete_from_page_cache(struct page *page, void *shadow) |
| 141 | { |
| 142 | struct address_space *mapping = page->mapping; |
| 143 | |
| 144 | #ifdef CONFIG_SDP |
| 145 | if(mapping_sensitive(mapping)) |
| 146 | sdp_page_cleanup(page); |
| 147 | #endif |
| 148 | |
| 149 | trace_mm_filemap_delete_from_page_cache(page); |
| 150 | /* |
| 151 | * if we're uptodate, flush out into the cleancache, otherwise |
| 152 | * invalidate any existing cleancache entries. We can't leave |
| 153 | * stale data around in the cleancache once our page is gone |
| 154 | */ |
| 155 | if (PageUptodate(page) && PageMappedToDisk(page)) |
| 156 | cleancache_put_page(page); |
| 157 | else |
| 158 | cleancache_invalidate_page(mapping, page); |
| 159 | |
| 160 | page_cache_tree_delete(mapping, page, shadow); |
| 161 | |
| 162 | page->mapping = NULL; |
| 163 | /* Leave page->index set: truncation lookup relies upon it */ |
| 164 | |
| 165 | __dec_zone_page_state(page, NR_FILE_PAGES); |
| 166 | if (PageSwapBacked(page)) |
| 167 | __dec_zone_page_state(page, NR_SHMEM); |
| 168 | BUG_ON(page_mapped(page)); |
| 169 | |
| 170 | /* |
| 171 | * Some filesystems seem to re-dirty the page even after |
| 172 | * the VM has canceled the dirty bit (eg ext3 journaling). |
| 173 | * |
| 174 | * Fix it up by doing a final dirty accounting check after |
| 175 | * having removed the page entirely. |
| 176 | */ |
| 177 | if (PageDirty(page) && mapping_cap_account_dirty(mapping)) { |
| 178 | dec_zone_page_state(page, NR_FILE_DIRTY); |
| 179 | dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE); |
| 180 | } |
| 181 | } |
| 182 | |
| 183 | /** |
| 184 | * delete_from_page_cache - delete page from page cache |
| 185 | * @page: the page which the kernel is trying to remove from page cache |
| 186 | * |
| 187 | * This must be called only on pages that have been verified to be in the page |
| 188 | * cache and locked. It will never put the page into the free list, the caller |
| 189 | * has a reference on the page. |
| 190 | */ |
| 191 | void delete_from_page_cache(struct page *page) |
| 192 | { |
| 193 | struct address_space *mapping = page->mapping; |
| 194 | void (*freepage)(struct page *); |
| 195 | |
| 196 | BUG_ON(!PageLocked(page)); |
| 197 | |
| 198 | freepage = mapping->a_ops->freepage; |
| 199 | spin_lock_irq(&mapping->tree_lock); |
| 200 | __delete_from_page_cache(page, NULL); |
| 201 | spin_unlock_irq(&mapping->tree_lock); |
| 202 | mem_cgroup_uncharge_cache_page(page); |
| 203 | |
| 204 | if (freepage) |
| 205 | freepage(page); |
| 206 | page_cache_release(page); |
| 207 | } |
| 208 | EXPORT_SYMBOL(delete_from_page_cache); |
| 209 | |
| 210 | static int sleep_on_page(void *word) |
| 211 | { |
| 212 | io_schedule(); |
| 213 | return 0; |
| 214 | } |
| 215 | |
| 216 | static int sleep_on_page_killable(void *word) |
| 217 | { |
| 218 | sleep_on_page(word); |
| 219 | return fatal_signal_pending(current) ? -EINTR : 0; |
| 220 | } |
| 221 | |
| 222 | static int filemap_check_errors(struct address_space *mapping) |
| 223 | { |
| 224 | int ret = 0; |
| 225 | /* Check for outstanding write errors */ |
| 226 | if (test_and_clear_bit(AS_ENOSPC, &mapping->flags)) |
| 227 | ret = -ENOSPC; |
| 228 | if (test_and_clear_bit(AS_EIO, &mapping->flags)) |
| 229 | ret = -EIO; |
| 230 | return ret; |
| 231 | } |
| 232 | |
| 233 | /** |
| 234 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
| 235 | * @mapping: address space structure to write |
| 236 | * @start: offset in bytes where the range starts |
| 237 | * @end: offset in bytes where the range ends (inclusive) |
| 238 | * @sync_mode: enable synchronous operation |
| 239 | * |
| 240 | * Start writeback against all of a mapping's dirty pages that lie |
| 241 | * within the byte offsets <start, end> inclusive. |
| 242 | * |
| 243 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
| 244 | * opposed to a regular memory cleansing writeback. The difference between |
| 245 | * these two operations is that if a dirty page/buffer is encountered, it must |
| 246 | * be waited upon, and not just skipped over. |
| 247 | */ |
| 248 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
| 249 | loff_t end, int sync_mode) |
| 250 | { |
| 251 | int ret; |
| 252 | struct writeback_control wbc = { |
| 253 | .sync_mode = sync_mode, |
| 254 | .nr_to_write = LONG_MAX, |
| 255 | .range_start = start, |
| 256 | .range_end = end, |
| 257 | }; |
| 258 | |
| 259 | if (!mapping_cap_writeback_dirty(mapping)) |
| 260 | return 0; |
| 261 | |
| 262 | ret = do_writepages(mapping, &wbc); |
| 263 | return ret; |
| 264 | } |
| 265 | |
| 266 | static inline int __filemap_fdatawrite(struct address_space *mapping, |
| 267 | int sync_mode) |
| 268 | { |
| 269 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
| 270 | } |
| 271 | |
| 272 | int filemap_fdatawrite(struct address_space *mapping) |
| 273 | { |
| 274 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); |
| 275 | } |
| 276 | EXPORT_SYMBOL(filemap_fdatawrite); |
| 277 | |
| 278 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
| 279 | loff_t end) |
| 280 | { |
| 281 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); |
| 282 | } |
| 283 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
| 284 | |
| 285 | /** |
| 286 | * filemap_flush - mostly a non-blocking flush |
| 287 | * @mapping: target address_space |
| 288 | * |
| 289 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
| 290 | * purposes - I/O may not be started against all dirty pages. |
| 291 | */ |
| 292 | int filemap_flush(struct address_space *mapping) |
| 293 | { |
| 294 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); |
| 295 | } |
| 296 | EXPORT_SYMBOL(filemap_flush); |
| 297 | |
| 298 | /** |
| 299 | * filemap_fdatawait_range - wait for writeback to complete |
| 300 | * @mapping: address space structure to wait for |
| 301 | * @start_byte: offset in bytes where the range starts |
| 302 | * @end_byte: offset in bytes where the range ends (inclusive) |
| 303 | * |
| 304 | * Walk the list of under-writeback pages of the given address space |
| 305 | * in the given range and wait for all of them. |
| 306 | */ |
| 307 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, |
| 308 | loff_t end_byte) |
| 309 | { |
| 310 | pgoff_t index = start_byte >> PAGE_CACHE_SHIFT; |
| 311 | pgoff_t end = end_byte >> PAGE_CACHE_SHIFT; |
| 312 | struct pagevec pvec; |
| 313 | int nr_pages; |
| 314 | int ret2, ret = 0; |
| 315 | |
| 316 | if (end_byte < start_byte) |
| 317 | goto out; |
| 318 | |
| 319 | pagevec_init(&pvec, 0); |
| 320 | while ((index <= end) && |
| 321 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, |
| 322 | PAGECACHE_TAG_WRITEBACK, |
| 323 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) { |
| 324 | unsigned i; |
| 325 | |
| 326 | for (i = 0; i < nr_pages; i++) { |
| 327 | struct page *page = pvec.pages[i]; |
| 328 | |
| 329 | /* until radix tree lookup accepts end_index */ |
| 330 | if (page->index > end) |
| 331 | continue; |
| 332 | |
| 333 | wait_on_page_writeback(page); |
| 334 | if (TestClearPageError(page)) |
| 335 | ret = -EIO; |
| 336 | } |
| 337 | pagevec_release(&pvec); |
| 338 | cond_resched(); |
| 339 | } |
| 340 | out: |
| 341 | ret2 = filemap_check_errors(mapping); |
| 342 | if (!ret) |
| 343 | ret = ret2; |
| 344 | |
| 345 | return ret; |
| 346 | } |
| 347 | EXPORT_SYMBOL(filemap_fdatawait_range); |
| 348 | |
| 349 | /** |
| 350 | * filemap_fdatawait - wait for all under-writeback pages to complete |
| 351 | * @mapping: address space structure to wait for |
| 352 | * |
| 353 | * Walk the list of under-writeback pages of the given address space |
| 354 | * and wait for all of them. |
| 355 | */ |
| 356 | int filemap_fdatawait(struct address_space *mapping) |
| 357 | { |
| 358 | loff_t i_size = i_size_read(mapping->host); |
| 359 | |
| 360 | if (i_size == 0) |
| 361 | return 0; |
| 362 | |
| 363 | return filemap_fdatawait_range(mapping, 0, i_size - 1); |
| 364 | } |
| 365 | EXPORT_SYMBOL(filemap_fdatawait); |
| 366 | |
| 367 | int filemap_write_and_wait(struct address_space *mapping) |
| 368 | { |
| 369 | int err = 0; |
| 370 | |
| 371 | if (mapping->nrpages) { |
| 372 | err = filemap_fdatawrite(mapping); |
| 373 | /* |
| 374 | * Even if the above returned error, the pages may be |
| 375 | * written partially (e.g. -ENOSPC), so we wait for it. |
| 376 | * But the -EIO is special case, it may indicate the worst |
| 377 | * thing (e.g. bug) happened, so we avoid waiting for it. |
| 378 | */ |
| 379 | if (err != -EIO) { |
| 380 | int err2 = filemap_fdatawait(mapping); |
| 381 | if (!err) |
| 382 | err = err2; |
| 383 | } |
| 384 | } else { |
| 385 | err = filemap_check_errors(mapping); |
| 386 | } |
| 387 | return err; |
| 388 | } |
| 389 | EXPORT_SYMBOL(filemap_write_and_wait); |
| 390 | |
| 391 | /** |
| 392 | * filemap_write_and_wait_range - write out & wait on a file range |
| 393 | * @mapping: the address_space for the pages |
| 394 | * @lstart: offset in bytes where the range starts |
| 395 | * @lend: offset in bytes where the range ends (inclusive) |
| 396 | * |
| 397 | * Write out and wait upon file offsets lstart->lend, inclusive. |
| 398 | * |
| 399 | * Note that `lend' is inclusive (describes the last byte to be written) so |
| 400 | * that this function can be used to write to the very end-of-file (end = -1). |
| 401 | */ |
| 402 | int filemap_write_and_wait_range(struct address_space *mapping, |
| 403 | loff_t lstart, loff_t lend) |
| 404 | { |
| 405 | int err = 0; |
| 406 | |
| 407 | if (mapping->nrpages) { |
| 408 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
| 409 | WB_SYNC_ALL); |
| 410 | /* See comment of filemap_write_and_wait() */ |
| 411 | if (err != -EIO) { |
| 412 | int err2 = filemap_fdatawait_range(mapping, |
| 413 | lstart, lend); |
| 414 | if (!err) |
| 415 | err = err2; |
| 416 | } |
| 417 | } else { |
| 418 | err = filemap_check_errors(mapping); |
| 419 | } |
| 420 | return err; |
| 421 | } |
| 422 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
| 423 | |
| 424 | /** |
| 425 | * replace_page_cache_page - replace a pagecache page with a new one |
| 426 | * @old: page to be replaced |
| 427 | * @new: page to replace with |
| 428 | * @gfp_mask: allocation mode |
| 429 | * |
| 430 | * This function replaces a page in the pagecache with a new one. On |
| 431 | * success it acquires the pagecache reference for the new page and |
| 432 | * drops it for the old page. Both the old and new pages must be |
| 433 | * locked. This function does not add the new page to the LRU, the |
| 434 | * caller must do that. |
| 435 | * |
| 436 | * The remove + add is atomic. The only way this function can fail is |
| 437 | * memory allocation failure. |
| 438 | */ |
| 439 | int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask) |
| 440 | { |
| 441 | int error; |
| 442 | |
| 443 | VM_BUG_ON(!PageLocked(old)); |
| 444 | VM_BUG_ON(!PageLocked(new)); |
| 445 | VM_BUG_ON(new->mapping); |
| 446 | |
| 447 | error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); |
| 448 | if (!error) { |
| 449 | struct address_space *mapping = old->mapping; |
| 450 | void (*freepage)(struct page *); |
| 451 | |
| 452 | pgoff_t offset = old->index; |
| 453 | freepage = mapping->a_ops->freepage; |
| 454 | |
| 455 | page_cache_get(new); |
| 456 | new->mapping = mapping; |
| 457 | new->index = offset; |
| 458 | |
| 459 | spin_lock_irq(&mapping->tree_lock); |
| 460 | __delete_from_page_cache(old, NULL); |
| 461 | error = radix_tree_insert(&mapping->page_tree, offset, new); |
| 462 | BUG_ON(error); |
| 463 | mapping->nrpages++; |
| 464 | __inc_zone_page_state(new, NR_FILE_PAGES); |
| 465 | if (PageSwapBacked(new)) |
| 466 | __inc_zone_page_state(new, NR_SHMEM); |
| 467 | spin_unlock_irq(&mapping->tree_lock); |
| 468 | /* mem_cgroup codes must not be called under tree_lock */ |
| 469 | mem_cgroup_replace_page_cache(old, new); |
| 470 | radix_tree_preload_end(); |
| 471 | if (freepage) |
| 472 | freepage(old); |
| 473 | page_cache_release(old); |
| 474 | } |
| 475 | |
| 476 | return error; |
| 477 | } |
| 478 | EXPORT_SYMBOL_GPL(replace_page_cache_page); |
| 479 | |
| 480 | /** |
| 481 | * add_to_page_cache_locked - add a locked page to the pagecache |
| 482 | * @page: page to add |
| 483 | * @mapping: the page's address_space |
| 484 | * @offset: page index |
| 485 | * @gfp_mask: page allocation mode |
| 486 | * |
| 487 | * This function is used to add a page to the pagecache. It must be locked. |
| 488 | * This function does not add the page to the LRU. The caller must do that. |
| 489 | */ |
| 490 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, |
| 491 | pgoff_t offset, gfp_t gfp_mask) |
| 492 | { |
| 493 | int error; |
| 494 | |
| 495 | VM_BUG_ON(!PageLocked(page)); |
| 496 | VM_BUG_ON(PageSwapBacked(page)); |
| 497 | |
| 498 | error = mem_cgroup_cache_charge(page, current->mm, |
| 499 | gfp_mask & GFP_RECLAIM_MASK); |
| 500 | if (error) |
| 501 | goto out; |
| 502 | |
| 503 | error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); |
| 504 | if (error == 0) { |
| 505 | page_cache_get(page); |
| 506 | page->mapping = mapping; |
| 507 | page->index = offset; |
| 508 | |
| 509 | spin_lock_irq(&mapping->tree_lock); |
| 510 | error = radix_tree_insert(&mapping->page_tree, offset, page); |
| 511 | if (likely(!error)) { |
| 512 | mapping->nrpages++; |
| 513 | __inc_zone_page_state(page, NR_FILE_PAGES); |
| 514 | spin_unlock_irq(&mapping->tree_lock); |
| 515 | trace_mm_filemap_add_to_page_cache(page); |
| 516 | } else { |
| 517 | page->mapping = NULL; |
| 518 | /* Leave page->index set: truncation relies upon it */ |
| 519 | spin_unlock_irq(&mapping->tree_lock); |
| 520 | mem_cgroup_uncharge_cache_page(page); |
| 521 | page_cache_release(page); |
| 522 | } |
| 523 | radix_tree_preload_end(); |
| 524 | } else |
| 525 | mem_cgroup_uncharge_cache_page(page); |
| 526 | out: |
| 527 | return error; |
| 528 | } |
| 529 | EXPORT_SYMBOL(add_to_page_cache_locked); |
| 530 | |
| 531 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, |
| 532 | pgoff_t offset, gfp_t gfp_mask) |
| 533 | { |
| 534 | int ret; |
| 535 | |
| 536 | ret = add_to_page_cache(page, mapping, offset, gfp_mask); |
| 537 | if (ret == 0) |
| 538 | lru_cache_add_file(page); |
| 539 | return ret; |
| 540 | } |
| 541 | EXPORT_SYMBOL_GPL(add_to_page_cache_lru); |
| 542 | |
| 543 | #ifdef CONFIG_NUMA |
| 544 | struct page *__page_cache_alloc(gfp_t gfp) |
| 545 | { |
| 546 | int n; |
| 547 | struct page *page; |
| 548 | |
| 549 | if (cpuset_do_page_mem_spread()) { |
| 550 | unsigned int cpuset_mems_cookie; |
| 551 | do { |
| 552 | cpuset_mems_cookie = get_mems_allowed(); |
| 553 | n = cpuset_mem_spread_node(); |
| 554 | page = alloc_pages_exact_node(n, gfp, 0); |
| 555 | } while (!put_mems_allowed(cpuset_mems_cookie) && !page); |
| 556 | |
| 557 | return page; |
| 558 | } |
| 559 | return alloc_pages(gfp, 0); |
| 560 | } |
| 561 | EXPORT_SYMBOL(__page_cache_alloc); |
| 562 | #endif |
| 563 | |
| 564 | /* |
| 565 | * In order to wait for pages to become available there must be |
| 566 | * waitqueues associated with pages. By using a hash table of |
| 567 | * waitqueues where the bucket discipline is to maintain all |
| 568 | * waiters on the same queue and wake all when any of the pages |
| 569 | * become available, and for the woken contexts to check to be |
| 570 | * sure the appropriate page became available, this saves space |
| 571 | * at a cost of "thundering herd" phenomena during rare hash |
| 572 | * collisions. |
| 573 | */ |
| 574 | static wait_queue_head_t *page_waitqueue(struct page *page) |
| 575 | { |
| 576 | const struct zone *zone = page_zone(page); |
| 577 | |
| 578 | return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)]; |
| 579 | } |
| 580 | |
| 581 | static inline void wake_up_page(struct page *page, int bit) |
| 582 | { |
| 583 | __wake_up_bit(page_waitqueue(page), &page->flags, bit); |
| 584 | } |
| 585 | |
| 586 | void wait_on_page_bit(struct page *page, int bit_nr) |
| 587 | { |
| 588 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); |
| 589 | |
| 590 | if (test_bit(bit_nr, &page->flags)) |
| 591 | __wait_on_bit(page_waitqueue(page), &wait, sleep_on_page, |
| 592 | TASK_UNINTERRUPTIBLE); |
| 593 | } |
| 594 | EXPORT_SYMBOL(wait_on_page_bit); |
| 595 | |
| 596 | int wait_on_page_bit_killable(struct page *page, int bit_nr) |
| 597 | { |
| 598 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); |
| 599 | |
| 600 | if (!test_bit(bit_nr, &page->flags)) |
| 601 | return 0; |
| 602 | |
| 603 | return __wait_on_bit(page_waitqueue(page), &wait, |
| 604 | sleep_on_page_killable, TASK_KILLABLE); |
| 605 | } |
| 606 | |
| 607 | /** |
| 608 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue |
| 609 | * @page: Page defining the wait queue of interest |
| 610 | * @waiter: Waiter to add to the queue |
| 611 | * |
| 612 | * Add an arbitrary @waiter to the wait queue for the nominated @page. |
| 613 | */ |
| 614 | void add_page_wait_queue(struct page *page, wait_queue_t *waiter) |
| 615 | { |
| 616 | wait_queue_head_t *q = page_waitqueue(page); |
| 617 | unsigned long flags; |
| 618 | |
| 619 | spin_lock_irqsave(&q->lock, flags); |
| 620 | __add_wait_queue(q, waiter); |
| 621 | spin_unlock_irqrestore(&q->lock, flags); |
| 622 | } |
| 623 | EXPORT_SYMBOL_GPL(add_page_wait_queue); |
| 624 | |
| 625 | /** |
| 626 | * unlock_page - unlock a locked page |
| 627 | * @page: the page |
| 628 | * |
| 629 | * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). |
| 630 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup |
| 631 | * mechananism between PageLocked pages and PageWriteback pages is shared. |
| 632 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. |
| 633 | * |
| 634 | * The mb is necessary to enforce ordering between the clear_bit and the read |
| 635 | * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()). |
| 636 | */ |
| 637 | void unlock_page(struct page *page) |
| 638 | { |
| 639 | VM_BUG_ON(!PageLocked(page)); |
| 640 | clear_bit_unlock(PG_locked, &page->flags); |
| 641 | smp_mb__after_atomic(); |
| 642 | wake_up_page(page, PG_locked); |
| 643 | } |
| 644 | EXPORT_SYMBOL(unlock_page); |
| 645 | |
| 646 | /** |
| 647 | * end_page_writeback - end writeback against a page |
| 648 | * @page: the page |
| 649 | */ |
| 650 | void end_page_writeback(struct page *page) |
| 651 | { |
| 652 | if (TestClearPageReclaim(page)) |
| 653 | rotate_reclaimable_page(page); |
| 654 | |
| 655 | if (!test_clear_page_writeback(page)) |
| 656 | BUG(); |
| 657 | |
| 658 | smp_mb__after_atomic(); |
| 659 | wake_up_page(page, PG_writeback); |
| 660 | } |
| 661 | EXPORT_SYMBOL(end_page_writeback); |
| 662 | |
| 663 | /** |
| 664 | * __lock_page - get a lock on the page, assuming we need to sleep to get it |
| 665 | * @page: the page to lock |
| 666 | */ |
| 667 | void __lock_page(struct page *page) |
| 668 | { |
| 669 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); |
| 670 | |
| 671 | __wait_on_bit_lock(page_waitqueue(page), &wait, sleep_on_page, |
| 672 | TASK_UNINTERRUPTIBLE); |
| 673 | } |
| 674 | EXPORT_SYMBOL(__lock_page); |
| 675 | |
| 676 | int __lock_page_killable(struct page *page) |
| 677 | { |
| 678 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); |
| 679 | |
| 680 | return __wait_on_bit_lock(page_waitqueue(page), &wait, |
| 681 | sleep_on_page_killable, TASK_KILLABLE); |
| 682 | } |
| 683 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
| 684 | |
| 685 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
| 686 | unsigned int flags) |
| 687 | { |
| 688 | if (flags & FAULT_FLAG_ALLOW_RETRY) { |
| 689 | /* |
| 690 | * CAUTION! In this case, mmap_sem is not released |
| 691 | * even though return 0. |
| 692 | */ |
| 693 | if (flags & FAULT_FLAG_RETRY_NOWAIT) |
| 694 | return 0; |
| 695 | |
| 696 | up_read(&mm->mmap_sem); |
| 697 | if (flags & FAULT_FLAG_KILLABLE) |
| 698 | wait_on_page_locked_killable(page); |
| 699 | else |
| 700 | wait_on_page_locked(page); |
| 701 | return 0; |
| 702 | } else { |
| 703 | if (flags & FAULT_FLAG_KILLABLE) { |
| 704 | int ret; |
| 705 | |
| 706 | ret = __lock_page_killable(page); |
| 707 | if (ret) { |
| 708 | up_read(&mm->mmap_sem); |
| 709 | return 0; |
| 710 | } |
| 711 | } else |
| 712 | __lock_page(page); |
| 713 | return 1; |
| 714 | } |
| 715 | } |
| 716 | |
| 717 | /** |
| 718 | * find_get_page - find and get a page reference |
| 719 | * @mapping: the address_space to search |
| 720 | * @offset: the page index |
| 721 | * |
| 722 | * Is there a pagecache struct page at the given (mapping, offset) tuple? |
| 723 | * If yes, increment its refcount and return it; if no, return NULL. |
| 724 | */ |
| 725 | struct page *find_get_page(struct address_space *mapping, pgoff_t offset) |
| 726 | { |
| 727 | void **pagep; |
| 728 | struct page *page; |
| 729 | |
| 730 | rcu_read_lock(); |
| 731 | repeat: |
| 732 | page = NULL; |
| 733 | pagep = radix_tree_lookup_slot(&mapping->page_tree, offset); |
| 734 | if (pagep) { |
| 735 | page = radix_tree_deref_slot(pagep); |
| 736 | if (unlikely(!page)) |
| 737 | goto out; |
| 738 | if (radix_tree_exception(page)) { |
| 739 | if (radix_tree_deref_retry(page)) |
| 740 | goto repeat; |
| 741 | /* |
| 742 | * Otherwise, shmem/tmpfs must be storing a swap entry |
| 743 | * here as an exceptional entry: so return it without |
| 744 | * attempting to raise page count. |
| 745 | */ |
| 746 | goto out; |
| 747 | } |
| 748 | if (!page_cache_get_speculative(page)) |
| 749 | goto repeat; |
| 750 | |
| 751 | /* |
| 752 | * Has the page moved? |
| 753 | * This is part of the lockless pagecache protocol. See |
| 754 | * include/linux/pagemap.h for details. |
| 755 | */ |
| 756 | if (unlikely(page != *pagep)) { |
| 757 | page_cache_release(page); |
| 758 | goto repeat; |
| 759 | } |
| 760 | } |
| 761 | out: |
| 762 | rcu_read_unlock(); |
| 763 | |
| 764 | return page; |
| 765 | } |
| 766 | EXPORT_SYMBOL(find_get_page); |
| 767 | |
| 768 | /** |
| 769 | * find_lock_page - locate, pin and lock a pagecache page |
| 770 | * @mapping: the address_space to search |
| 771 | * @offset: the page index |
| 772 | * |
| 773 | * Locates the desired pagecache page, locks it, increments its reference |
| 774 | * count and returns its address. |
| 775 | * |
| 776 | * Returns zero if the page was not present. find_lock_page() may sleep. |
| 777 | */ |
| 778 | struct page *find_lock_page(struct address_space *mapping, pgoff_t offset) |
| 779 | { |
| 780 | struct page *page; |
| 781 | |
| 782 | repeat: |
| 783 | page = find_get_page(mapping, offset); |
| 784 | if (page && !radix_tree_exception(page)) { |
| 785 | lock_page(page); |
| 786 | /* Has the page been truncated? */ |
| 787 | if (unlikely(page->mapping != mapping)) { |
| 788 | unlock_page(page); |
| 789 | page_cache_release(page); |
| 790 | goto repeat; |
| 791 | } |
| 792 | VM_BUG_ON(page->index != offset); |
| 793 | } |
| 794 | return page; |
| 795 | } |
| 796 | EXPORT_SYMBOL(find_lock_page); |
| 797 | |
| 798 | /** |
| 799 | * find_or_create_page - locate or add a pagecache page |
| 800 | * @mapping: the page's address_space |
| 801 | * @index: the page's index into the mapping |
| 802 | * @gfp_mask: page allocation mode |
| 803 | * |
| 804 | * Locates a page in the pagecache. If the page is not present, a new page |
| 805 | * is allocated using @gfp_mask and is added to the pagecache and to the VM's |
| 806 | * LRU list. The returned page is locked and has its reference count |
| 807 | * incremented. |
| 808 | * |
| 809 | * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic |
| 810 | * allocation! |
| 811 | * |
| 812 | * find_or_create_page() returns the desired page's address, or zero on |
| 813 | * memory exhaustion. |
| 814 | */ |
| 815 | struct page *find_or_create_page(struct address_space *mapping, |
| 816 | pgoff_t index, gfp_t gfp_mask) |
| 817 | { |
| 818 | struct page *page; |
| 819 | int err; |
| 820 | repeat: |
| 821 | page = find_lock_page(mapping, index); |
| 822 | if (!page) { |
| 823 | page = __page_cache_alloc(gfp_mask); |
| 824 | if (!page) |
| 825 | return NULL; |
| 826 | /* |
| 827 | * We want a regular kernel memory (not highmem or DMA etc) |
| 828 | * allocation for the radix tree nodes, but we need to honour |
| 829 | * the context-specific requirements the caller has asked for. |
| 830 | * GFP_RECLAIM_MASK collects those requirements. |
| 831 | */ |
| 832 | err = add_to_page_cache_lru(page, mapping, index, |
| 833 | (gfp_mask & GFP_RECLAIM_MASK)); |
| 834 | if (unlikely(err)) { |
| 835 | page_cache_release(page); |
| 836 | page = NULL; |
| 837 | if (err == -EEXIST) |
| 838 | goto repeat; |
| 839 | } |
| 840 | } |
| 841 | return page; |
| 842 | } |
| 843 | EXPORT_SYMBOL(find_or_create_page); |
| 844 | |
| 845 | /** |
| 846 | * find_get_pages - gang pagecache lookup |
| 847 | * @mapping: The address_space to search |
| 848 | * @start: The starting page index |
| 849 | * @nr_pages: The maximum number of pages |
| 850 | * @pages: Where the resulting pages are placed |
| 851 | * |
| 852 | * find_get_pages() will search for and return a group of up to |
| 853 | * @nr_pages pages in the mapping. The pages are placed at @pages. |
| 854 | * find_get_pages() takes a reference against the returned pages. |
| 855 | * |
| 856 | * The search returns a group of mapping-contiguous pages with ascending |
| 857 | * indexes. There may be holes in the indices due to not-present pages. |
| 858 | * |
| 859 | * find_get_pages() returns the number of pages which were found. |
| 860 | */ |
| 861 | unsigned find_get_pages(struct address_space *mapping, pgoff_t start, |
| 862 | unsigned int nr_pages, struct page **pages) |
| 863 | { |
| 864 | struct radix_tree_iter iter; |
| 865 | void **slot; |
| 866 | unsigned ret = 0; |
| 867 | |
| 868 | if (unlikely(!nr_pages)) |
| 869 | return 0; |
| 870 | |
| 871 | rcu_read_lock(); |
| 872 | restart: |
| 873 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
| 874 | struct page *page; |
| 875 | repeat: |
| 876 | page = radix_tree_deref_slot(slot); |
| 877 | if (unlikely(!page)) |
| 878 | continue; |
| 879 | |
| 880 | if (radix_tree_exception(page)) { |
| 881 | if (radix_tree_deref_retry(page)) { |
| 882 | /* |
| 883 | * Transient condition which can only trigger |
| 884 | * when entry at index 0 moves out of or back |
| 885 | * to root: none yet gotten, safe to restart. |
| 886 | */ |
| 887 | WARN_ON(iter.index); |
| 888 | goto restart; |
| 889 | } |
| 890 | /* |
| 891 | * Otherwise, shmem/tmpfs must be storing a swap entry |
| 892 | * here as an exceptional entry: so skip over it - |
| 893 | * we only reach this from invalidate_mapping_pages(). |
| 894 | */ |
| 895 | continue; |
| 896 | } |
| 897 | |
| 898 | if (!page_cache_get_speculative(page)) |
| 899 | goto repeat; |
| 900 | |
| 901 | /* Has the page moved? */ |
| 902 | if (unlikely(page != *slot)) { |
| 903 | page_cache_release(page); |
| 904 | goto repeat; |
| 905 | } |
| 906 | |
| 907 | pages[ret] = page; |
| 908 | if (++ret == nr_pages) |
| 909 | break; |
| 910 | } |
| 911 | |
| 912 | rcu_read_unlock(); |
| 913 | return ret; |
| 914 | } |
| 915 | |
| 916 | /** |
| 917 | * find_get_pages_contig - gang contiguous pagecache lookup |
| 918 | * @mapping: The address_space to search |
| 919 | * @index: The starting page index |
| 920 | * @nr_pages: The maximum number of pages |
| 921 | * @pages: Where the resulting pages are placed |
| 922 | * |
| 923 | * find_get_pages_contig() works exactly like find_get_pages(), except |
| 924 | * that the returned number of pages are guaranteed to be contiguous. |
| 925 | * |
| 926 | * find_get_pages_contig() returns the number of pages which were found. |
| 927 | */ |
| 928 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, |
| 929 | unsigned int nr_pages, struct page **pages) |
| 930 | { |
| 931 | struct radix_tree_iter iter; |
| 932 | void **slot; |
| 933 | unsigned int ret = 0; |
| 934 | |
| 935 | if (unlikely(!nr_pages)) |
| 936 | return 0; |
| 937 | |
| 938 | rcu_read_lock(); |
| 939 | restart: |
| 940 | radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) { |
| 941 | struct page *page; |
| 942 | repeat: |
| 943 | page = radix_tree_deref_slot(slot); |
| 944 | /* The hole, there no reason to continue */ |
| 945 | if (unlikely(!page)) |
| 946 | break; |
| 947 | |
| 948 | if (radix_tree_exception(page)) { |
| 949 | if (radix_tree_deref_retry(page)) { |
| 950 | /* |
| 951 | * Transient condition which can only trigger |
| 952 | * when entry at index 0 moves out of or back |
| 953 | * to root: none yet gotten, safe to restart. |
| 954 | */ |
| 955 | goto restart; |
| 956 | } |
| 957 | /* |
| 958 | * Otherwise, shmem/tmpfs must be storing a swap entry |
| 959 | * here as an exceptional entry: so stop looking for |
| 960 | * contiguous pages. |
| 961 | */ |
| 962 | break; |
| 963 | } |
| 964 | |
| 965 | if (!page_cache_get_speculative(page)) |
| 966 | goto repeat; |
| 967 | |
| 968 | /* Has the page moved? */ |
| 969 | if (unlikely(page != *slot)) { |
| 970 | page_cache_release(page); |
| 971 | goto repeat; |
| 972 | } |
| 973 | |
| 974 | /* |
| 975 | * must check mapping and index after taking the ref. |
| 976 | * otherwise we can get both false positives and false |
| 977 | * negatives, which is just confusing to the caller. |
| 978 | */ |
| 979 | if (page->mapping == NULL || page->index != iter.index) { |
| 980 | page_cache_release(page); |
| 981 | break; |
| 982 | } |
| 983 | |
| 984 | pages[ret] = page; |
| 985 | if (++ret == nr_pages) |
| 986 | break; |
| 987 | } |
| 988 | rcu_read_unlock(); |
| 989 | return ret; |
| 990 | } |
| 991 | EXPORT_SYMBOL(find_get_pages_contig); |
| 992 | |
| 993 | /** |
| 994 | * find_get_pages_tag - find and return pages that match @tag |
| 995 | * @mapping: the address_space to search |
| 996 | * @index: the starting page index |
| 997 | * @tag: the tag index |
| 998 | * @nr_pages: the maximum number of pages |
| 999 | * @pages: where the resulting pages are placed |
| 1000 | * |
| 1001 | * Like find_get_pages, except we only return pages which are tagged with |
| 1002 | * @tag. We update @index to index the next page for the traversal. |
| 1003 | */ |
| 1004 | unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index, |
| 1005 | int tag, unsigned int nr_pages, struct page **pages) |
| 1006 | { |
| 1007 | struct radix_tree_iter iter; |
| 1008 | void **slot; |
| 1009 | unsigned ret = 0; |
| 1010 | |
| 1011 | if (unlikely(!nr_pages)) |
| 1012 | return 0; |
| 1013 | |
| 1014 | rcu_read_lock(); |
| 1015 | restart: |
| 1016 | radix_tree_for_each_tagged(slot, &mapping->page_tree, |
| 1017 | &iter, *index, tag) { |
| 1018 | struct page *page; |
| 1019 | repeat: |
| 1020 | page = radix_tree_deref_slot(slot); |
| 1021 | if (unlikely(!page)) |
| 1022 | continue; |
| 1023 | |
| 1024 | if (radix_tree_exception(page)) { |
| 1025 | if (radix_tree_deref_retry(page)) { |
| 1026 | /* |
| 1027 | * Transient condition which can only trigger |
| 1028 | * when entry at index 0 moves out of or back |
| 1029 | * to root: none yet gotten, safe to restart. |
| 1030 | */ |
| 1031 | goto restart; |
| 1032 | } |
| 1033 | /* |
| 1034 | * This function is never used on a shmem/tmpfs |
| 1035 | * mapping, so a swap entry won't be found here. |
| 1036 | */ |
| 1037 | BUG(); |
| 1038 | } |
| 1039 | |
| 1040 | if (!page_cache_get_speculative(page)) |
| 1041 | goto repeat; |
| 1042 | |
| 1043 | /* Has the page moved? */ |
| 1044 | if (unlikely(page != *slot)) { |
| 1045 | page_cache_release(page); |
| 1046 | goto repeat; |
| 1047 | } |
| 1048 | |
| 1049 | pages[ret] = page; |
| 1050 | if (++ret == nr_pages) |
| 1051 | break; |
| 1052 | } |
| 1053 | |
| 1054 | rcu_read_unlock(); |
| 1055 | |
| 1056 | if (ret) |
| 1057 | *index = pages[ret - 1]->index + 1; |
| 1058 | |
| 1059 | return ret; |
| 1060 | } |
| 1061 | EXPORT_SYMBOL(find_get_pages_tag); |
| 1062 | |
| 1063 | /** |
| 1064 | * grab_cache_page_nowait - returns locked page at given index in given cache |
| 1065 | * @mapping: target address_space |
| 1066 | * @index: the page index |
| 1067 | * |
| 1068 | * Same as grab_cache_page(), but do not wait if the page is unavailable. |
| 1069 | * This is intended for speculative data generators, where the data can |
| 1070 | * be regenerated if the page couldn't be grabbed. This routine should |
| 1071 | * be safe to call while holding the lock for another page. |
| 1072 | * |
| 1073 | * Clear __GFP_FS when allocating the page to avoid recursion into the fs |
| 1074 | * and deadlock against the caller's locked page. |
| 1075 | */ |
| 1076 | struct page * |
| 1077 | grab_cache_page_nowait(struct address_space *mapping, pgoff_t index) |
| 1078 | { |
| 1079 | struct page *page = find_get_page(mapping, index); |
| 1080 | |
| 1081 | if (page) { |
| 1082 | if (trylock_page(page)) |
| 1083 | return page; |
| 1084 | page_cache_release(page); |
| 1085 | return NULL; |
| 1086 | } |
| 1087 | page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS); |
| 1088 | if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) { |
| 1089 | page_cache_release(page); |
| 1090 | page = NULL; |
| 1091 | } |
| 1092 | return page; |
| 1093 | } |
| 1094 | EXPORT_SYMBOL(grab_cache_page_nowait); |
| 1095 | |
| 1096 | /* |
| 1097 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail |
| 1098 | * a _large_ part of the i/o request. Imagine the worst scenario: |
| 1099 | * |
| 1100 | * ---R__________________________________________B__________ |
| 1101 | * ^ reading here ^ bad block(assume 4k) |
| 1102 | * |
| 1103 | * read(R) => miss => readahead(R...B) => media error => frustrating retries |
| 1104 | * => failing the whole request => read(R) => read(R+1) => |
| 1105 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => |
| 1106 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => |
| 1107 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... |
| 1108 | * |
| 1109 | * It is going insane. Fix it by quickly scaling down the readahead size. |
| 1110 | */ |
| 1111 | static void shrink_readahead_size_eio(struct file *filp, |
| 1112 | struct file_ra_state *ra) |
| 1113 | { |
| 1114 | ra->ra_pages /= 4; |
| 1115 | } |
| 1116 | |
| 1117 | /** |
| 1118 | * do_generic_file_read - generic file read routine |
| 1119 | * @filp: the file to read |
| 1120 | * @ppos: current file position |
| 1121 | * @desc: read_descriptor |
| 1122 | * @actor: read method |
| 1123 | * |
| 1124 | * This is a generic file read routine, and uses the |
| 1125 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
| 1126 | * |
| 1127 | * This is really ugly. But the goto's actually try to clarify some |
| 1128 | * of the logic when it comes to error handling etc. |
| 1129 | */ |
| 1130 | static void do_generic_file_read(struct file *filp, loff_t *ppos, |
| 1131 | read_descriptor_t *desc, read_actor_t actor) |
| 1132 | { |
| 1133 | struct address_space *mapping = filp->f_mapping; |
| 1134 | struct inode *inode = mapping->host; |
| 1135 | struct file_ra_state *ra = &filp->f_ra; |
| 1136 | pgoff_t index; |
| 1137 | pgoff_t last_index; |
| 1138 | pgoff_t prev_index; |
| 1139 | unsigned long offset; /* offset into pagecache page */ |
| 1140 | unsigned int prev_offset; |
| 1141 | int error; |
| 1142 | |
| 1143 | index = *ppos >> PAGE_CACHE_SHIFT; |
| 1144 | prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT; |
| 1145 | prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1); |
| 1146 | last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; |
| 1147 | offset = *ppos & ~PAGE_CACHE_MASK; |
| 1148 | |
| 1149 | for (;;) { |
| 1150 | struct page *page; |
| 1151 | pgoff_t end_index; |
| 1152 | loff_t isize; |
| 1153 | unsigned long nr, ret; |
| 1154 | |
| 1155 | cond_resched(); |
| 1156 | find_page: |
| 1157 | if (fatal_signal_pending(current)) { |
| 1158 | error = -EINTR; |
| 1159 | goto out; |
| 1160 | } |
| 1161 | |
| 1162 | page = find_get_page(mapping, index); |
| 1163 | if (!page) { |
| 1164 | page_cache_sync_readahead(mapping, |
| 1165 | ra, filp, |
| 1166 | index, last_index - index); |
| 1167 | page = find_get_page(mapping, index); |
| 1168 | if (unlikely(page == NULL)) |
| 1169 | goto no_cached_page; |
| 1170 | } |
| 1171 | if (PageReadahead(page)) { |
| 1172 | page_cache_async_readahead(mapping, |
| 1173 | ra, filp, page, |
| 1174 | index, last_index - index); |
| 1175 | } |
| 1176 | if (!PageUptodate(page)) { |
| 1177 | if (inode->i_blkbits == PAGE_CACHE_SHIFT || |
| 1178 | !mapping->a_ops->is_partially_uptodate) |
| 1179 | goto page_not_up_to_date; |
| 1180 | if (!trylock_page(page)) |
| 1181 | goto page_not_up_to_date; |
| 1182 | /* Did it get truncated before we got the lock? */ |
| 1183 | if (!page->mapping) |
| 1184 | goto page_not_up_to_date_locked; |
| 1185 | if (!mapping->a_ops->is_partially_uptodate(page, |
| 1186 | desc, offset)) |
| 1187 | goto page_not_up_to_date_locked; |
| 1188 | unlock_page(page); |
| 1189 | } |
| 1190 | page_ok: |
| 1191 | /* |
| 1192 | * i_size must be checked after we know the page is Uptodate. |
| 1193 | * |
| 1194 | * Checking i_size after the check allows us to calculate |
| 1195 | * the correct value for "nr", which means the zero-filled |
| 1196 | * part of the page is not copied back to userspace (unless |
| 1197 | * another truncate extends the file - this is desired though). |
| 1198 | */ |
| 1199 | |
| 1200 | isize = i_size_read(inode); |
| 1201 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; |
| 1202 | if (unlikely(!isize || index > end_index)) { |
| 1203 | page_cache_release(page); |
| 1204 | goto out; |
| 1205 | } |
| 1206 | |
| 1207 | /* nr is the maximum number of bytes to copy from this page */ |
| 1208 | nr = PAGE_CACHE_SIZE; |
| 1209 | if (index == end_index) { |
| 1210 | nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; |
| 1211 | if (nr <= offset) { |
| 1212 | page_cache_release(page); |
| 1213 | goto out; |
| 1214 | } |
| 1215 | } |
| 1216 | nr = nr - offset; |
| 1217 | |
| 1218 | /* If users can be writing to this page using arbitrary |
| 1219 | * virtual addresses, take care about potential aliasing |
| 1220 | * before reading the page on the kernel side. |
| 1221 | */ |
| 1222 | if (mapping_writably_mapped(mapping)) |
| 1223 | flush_dcache_page(page); |
| 1224 | |
| 1225 | /* |
| 1226 | * When a sequential read accesses a page several times, |
| 1227 | * only mark it as accessed the first time. |
| 1228 | */ |
| 1229 | if (prev_index != index || offset != prev_offset) |
| 1230 | mark_page_accessed(page); |
| 1231 | prev_index = index; |
| 1232 | |
| 1233 | /* |
| 1234 | * Ok, we have the page, and it's up-to-date, so |
| 1235 | * now we can copy it to user space... |
| 1236 | * |
| 1237 | * The actor routine returns how many bytes were actually used.. |
| 1238 | * NOTE! This may not be the same as how much of a user buffer |
| 1239 | * we filled up (we may be padding etc), so we can only update |
| 1240 | * "pos" here (the actor routine has to update the user buffer |
| 1241 | * pointers and the remaining count). |
| 1242 | */ |
| 1243 | ret = actor(desc, page, offset, nr); |
| 1244 | offset += ret; |
| 1245 | index += offset >> PAGE_CACHE_SHIFT; |
| 1246 | offset &= ~PAGE_CACHE_MASK; |
| 1247 | prev_offset = offset; |
| 1248 | |
| 1249 | page_cache_release(page); |
| 1250 | if (ret == nr && desc->count) |
| 1251 | continue; |
| 1252 | goto out; |
| 1253 | |
| 1254 | page_not_up_to_date: |
| 1255 | /* Get exclusive access to the page ... */ |
| 1256 | error = lock_page_killable(page); |
| 1257 | if (unlikely(error)) |
| 1258 | goto readpage_error; |
| 1259 | |
| 1260 | page_not_up_to_date_locked: |
| 1261 | /* Did it get truncated before we got the lock? */ |
| 1262 | if (!page->mapping) { |
| 1263 | unlock_page(page); |
| 1264 | page_cache_release(page); |
| 1265 | continue; |
| 1266 | } |
| 1267 | |
| 1268 | /* Did somebody else fill it already? */ |
| 1269 | if (PageUptodate(page)) { |
| 1270 | unlock_page(page); |
| 1271 | goto page_ok; |
| 1272 | } |
| 1273 | |
| 1274 | readpage: |
| 1275 | /* |
| 1276 | * A previous I/O error may have been due to temporary |
| 1277 | * failures, eg. multipath errors. |
| 1278 | * PG_error will be set again if readpage fails. |
| 1279 | */ |
| 1280 | ClearPageError(page); |
| 1281 | /* Start the actual read. The read will unlock the page. */ |
| 1282 | error = mapping->a_ops->readpage(filp, page); |
| 1283 | |
| 1284 | if (unlikely(error)) { |
| 1285 | if (error == AOP_TRUNCATED_PAGE) { |
| 1286 | page_cache_release(page); |
| 1287 | goto find_page; |
| 1288 | } |
| 1289 | goto readpage_error; |
| 1290 | } |
| 1291 | |
| 1292 | if (!PageUptodate(page)) { |
| 1293 | error = lock_page_killable(page); |
| 1294 | if (unlikely(error)) |
| 1295 | goto readpage_error; |
| 1296 | if (!PageUptodate(page)) { |
| 1297 | if (page->mapping == NULL) { |
| 1298 | /* |
| 1299 | * invalidate_mapping_pages got it |
| 1300 | */ |
| 1301 | unlock_page(page); |
| 1302 | page_cache_release(page); |
| 1303 | goto find_page; |
| 1304 | } |
| 1305 | unlock_page(page); |
| 1306 | shrink_readahead_size_eio(filp, ra); |
| 1307 | error = -EIO; |
| 1308 | goto readpage_error; |
| 1309 | } |
| 1310 | unlock_page(page); |
| 1311 | } |
| 1312 | |
| 1313 | goto page_ok; |
| 1314 | |
| 1315 | readpage_error: |
| 1316 | /* UHHUH! A synchronous read error occurred. Report it */ |
| 1317 | desc->error = error; |
| 1318 | page_cache_release(page); |
| 1319 | goto out; |
| 1320 | |
| 1321 | no_cached_page: |
| 1322 | /* |
| 1323 | * Ok, it wasn't cached, so we need to create a new |
| 1324 | * page.. |
| 1325 | */ |
| 1326 | page = page_cache_alloc_cold(mapping); |
| 1327 | if (!page) { |
| 1328 | desc->error = -ENOMEM; |
| 1329 | goto out; |
| 1330 | } |
| 1331 | error = add_to_page_cache_lru(page, mapping, |
| 1332 | index, GFP_KERNEL); |
| 1333 | if (error) { |
| 1334 | page_cache_release(page); |
| 1335 | if (error == -EEXIST) |
| 1336 | goto find_page; |
| 1337 | desc->error = error; |
| 1338 | goto out; |
| 1339 | } |
| 1340 | goto readpage; |
| 1341 | } |
| 1342 | |
| 1343 | out: |
| 1344 | ra->prev_pos = prev_index; |
| 1345 | ra->prev_pos <<= PAGE_CACHE_SHIFT; |
| 1346 | ra->prev_pos |= prev_offset; |
| 1347 | |
| 1348 | *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset; |
| 1349 | file_accessed(filp); |
| 1350 | } |
| 1351 | |
| 1352 | int file_read_actor(read_descriptor_t *desc, struct page *page, |
| 1353 | unsigned long offset, unsigned long size) |
| 1354 | { |
| 1355 | char *kaddr; |
| 1356 | unsigned long left, count = desc->count; |
| 1357 | |
| 1358 | if (size > count) |
| 1359 | size = count; |
| 1360 | |
| 1361 | /* |
| 1362 | * Faults on the destination of a read are common, so do it before |
| 1363 | * taking the kmap. |
| 1364 | */ |
| 1365 | if (!fault_in_pages_writeable(desc->arg.buf, size)) { |
| 1366 | kaddr = kmap_atomic(page); |
| 1367 | left = __copy_to_user_inatomic(desc->arg.buf, |
| 1368 | kaddr + offset, size); |
| 1369 | kunmap_atomic(kaddr); |
| 1370 | if (left == 0) |
| 1371 | goto success; |
| 1372 | } |
| 1373 | |
| 1374 | /* Do it the slow way */ |
| 1375 | kaddr = kmap(page); |
| 1376 | left = __copy_to_user(desc->arg.buf, kaddr + offset, size); |
| 1377 | kunmap(page); |
| 1378 | |
| 1379 | if (left) { |
| 1380 | size -= left; |
| 1381 | desc->error = -EFAULT; |
| 1382 | } |
| 1383 | success: |
| 1384 | desc->count = count - size; |
| 1385 | desc->written += size; |
| 1386 | desc->arg.buf += size; |
| 1387 | return size; |
| 1388 | } |
| 1389 | |
| 1390 | /* |
| 1391 | * Performs necessary checks before doing a write |
| 1392 | * @iov: io vector request |
| 1393 | * @nr_segs: number of segments in the iovec |
| 1394 | * @count: number of bytes to write |
| 1395 | * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE |
| 1396 | * |
| 1397 | * Adjust number of segments and amount of bytes to write (nr_segs should be |
| 1398 | * properly initialized first). Returns appropriate error code that caller |
| 1399 | * should return or zero in case that write should be allowed. |
| 1400 | */ |
| 1401 | int generic_segment_checks(const struct iovec *iov, |
| 1402 | unsigned long *nr_segs, size_t *count, int access_flags) |
| 1403 | { |
| 1404 | unsigned long seg; |
| 1405 | size_t cnt = 0; |
| 1406 | for (seg = 0; seg < *nr_segs; seg++) { |
| 1407 | const struct iovec *iv = &iov[seg]; |
| 1408 | |
| 1409 | /* |
| 1410 | * If any segment has a negative length, or the cumulative |
| 1411 | * length ever wraps negative then return -EINVAL. |
| 1412 | */ |
| 1413 | cnt += iv->iov_len; |
| 1414 | if (unlikely((ssize_t)(cnt|iv->iov_len) < 0)) |
| 1415 | return -EINVAL; |
| 1416 | if (access_ok(access_flags, iv->iov_base, iv->iov_len)) |
| 1417 | continue; |
| 1418 | if (seg == 0) |
| 1419 | return -EFAULT; |
| 1420 | *nr_segs = seg; |
| 1421 | cnt -= iv->iov_len; /* This segment is no good */ |
| 1422 | break; |
| 1423 | } |
| 1424 | *count = cnt; |
| 1425 | return 0; |
| 1426 | } |
| 1427 | EXPORT_SYMBOL(generic_segment_checks); |
| 1428 | |
| 1429 | /** |
| 1430 | * generic_file_aio_read - generic filesystem read routine |
| 1431 | * @iocb: kernel I/O control block |
| 1432 | * @iov: io vector request |
| 1433 | * @nr_segs: number of segments in the iovec |
| 1434 | * @pos: current file position |
| 1435 | * |
| 1436 | * This is the "read()" routine for all filesystems |
| 1437 | * that can use the page cache directly. |
| 1438 | */ |
| 1439 | ssize_t |
| 1440 | generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov, |
| 1441 | unsigned long nr_segs, loff_t pos) |
| 1442 | { |
| 1443 | struct file *filp = iocb->ki_filp; |
| 1444 | ssize_t retval; |
| 1445 | unsigned long seg = 0; |
| 1446 | size_t count; |
| 1447 | loff_t *ppos = &iocb->ki_pos; |
| 1448 | |
| 1449 | count = 0; |
| 1450 | retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE); |
| 1451 | if (retval) |
| 1452 | return retval; |
| 1453 | |
| 1454 | /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ |
| 1455 | if (filp->f_flags & O_DIRECT) { |
| 1456 | loff_t size; |
| 1457 | struct address_space *mapping; |
| 1458 | struct inode *inode; |
| 1459 | |
| 1460 | mapping = filp->f_mapping; |
| 1461 | inode = mapping->host; |
| 1462 | if (!count) |
| 1463 | goto out; /* skip atime */ |
| 1464 | size = i_size_read(inode); |
| 1465 | if (pos < size) { |
| 1466 | retval = filemap_write_and_wait_range(mapping, pos, |
| 1467 | pos + iov_length(iov, nr_segs) - 1); |
| 1468 | if (!retval) { |
| 1469 | retval = mapping->a_ops->direct_IO(READ, iocb, |
| 1470 | iov, pos, nr_segs); |
| 1471 | } |
| 1472 | if (retval > 0) { |
| 1473 | *ppos = pos + retval; |
| 1474 | count -= retval; |
| 1475 | } |
| 1476 | |
| 1477 | /* |
| 1478 | * Btrfs can have a short DIO read if we encounter |
| 1479 | * compressed extents, so if there was an error, or if |
| 1480 | * we've already read everything we wanted to, or if |
| 1481 | * there was a short read because we hit EOF, go ahead |
| 1482 | * and return. Otherwise fallthrough to buffered io for |
| 1483 | * the rest of the read. |
| 1484 | */ |
| 1485 | if (retval < 0 || !count || *ppos >= size) { |
| 1486 | file_accessed(filp); |
| 1487 | goto out; |
| 1488 | } |
| 1489 | } |
| 1490 | } |
| 1491 | |
| 1492 | count = retval; |
| 1493 | for (seg = 0; seg < nr_segs; seg++) { |
| 1494 | read_descriptor_t desc; |
| 1495 | loff_t offset = 0; |
| 1496 | |
| 1497 | /* |
| 1498 | * If we did a short DIO read we need to skip the section of the |
| 1499 | * iov that we've already read data into. |
| 1500 | */ |
| 1501 | if (count) { |
| 1502 | if (count > iov[seg].iov_len) { |
| 1503 | count -= iov[seg].iov_len; |
| 1504 | continue; |
| 1505 | } |
| 1506 | offset = count; |
| 1507 | count = 0; |
| 1508 | } |
| 1509 | |
| 1510 | desc.written = 0; |
| 1511 | desc.arg.buf = iov[seg].iov_base + offset; |
| 1512 | desc.count = iov[seg].iov_len - offset; |
| 1513 | if (desc.count == 0) |
| 1514 | continue; |
| 1515 | desc.error = 0; |
| 1516 | do_generic_file_read(filp, ppos, &desc, file_read_actor); |
| 1517 | retval += desc.written; |
| 1518 | if (desc.error) { |
| 1519 | retval = retval ?: desc.error; |
| 1520 | break; |
| 1521 | } |
| 1522 | if (desc.count > 0) |
| 1523 | break; |
| 1524 | } |
| 1525 | out: |
| 1526 | return retval; |
| 1527 | } |
| 1528 | EXPORT_SYMBOL(generic_file_aio_read); |
| 1529 | |
| 1530 | #ifdef CONFIG_MMU |
| 1531 | /** |
| 1532 | * page_cache_read - adds requested page to the page cache if not already there |
| 1533 | * @file: file to read |
| 1534 | * @offset: page index |
| 1535 | * |
| 1536 | * This adds the requested page to the page cache if it isn't already there, |
| 1537 | * and schedules an I/O to read in its contents from disk. |
| 1538 | */ |
| 1539 | static int page_cache_read(struct file *file, pgoff_t offset) |
| 1540 | { |
| 1541 | struct address_space *mapping = file->f_mapping; |
| 1542 | struct page *page; |
| 1543 | int ret; |
| 1544 | |
| 1545 | do { |
| 1546 | page = page_cache_alloc_cold(mapping); |
| 1547 | if (!page) |
| 1548 | return -ENOMEM; |
| 1549 | |
| 1550 | ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL); |
| 1551 | if (ret == 0) |
| 1552 | ret = mapping->a_ops->readpage(file, page); |
| 1553 | else if (ret == -EEXIST) |
| 1554 | ret = 0; /* losing race to add is OK */ |
| 1555 | |
| 1556 | page_cache_release(page); |
| 1557 | |
| 1558 | } while (ret == AOP_TRUNCATED_PAGE); |
| 1559 | |
| 1560 | return ret; |
| 1561 | } |
| 1562 | |
| 1563 | #define MMAP_LOTSAMISS (100) |
| 1564 | |
| 1565 | /* |
| 1566 | * Synchronous readahead happens when we don't even find |
| 1567 | * a page in the page cache at all. |
| 1568 | */ |
| 1569 | static void do_sync_mmap_readahead(struct vm_area_struct *vma, |
| 1570 | struct file_ra_state *ra, |
| 1571 | struct file *file, |
| 1572 | pgoff_t offset) |
| 1573 | { |
| 1574 | unsigned long ra_pages; |
| 1575 | struct address_space *mapping = file->f_mapping; |
| 1576 | |
| 1577 | /* If we don't want any read-ahead, don't bother */ |
| 1578 | if (VM_RandomReadHint(vma)) |
| 1579 | return; |
| 1580 | if (!ra->ra_pages) |
| 1581 | return; |
| 1582 | |
| 1583 | if (VM_SequentialReadHint(vma)) { |
| 1584 | page_cache_sync_readahead(mapping, ra, file, offset, |
| 1585 | ra->ra_pages); |
| 1586 | return; |
| 1587 | } |
| 1588 | |
| 1589 | /* Avoid banging the cache line if not needed */ |
| 1590 | if (ra->mmap_miss < MMAP_LOTSAMISS * 10) |
| 1591 | ra->mmap_miss++; |
| 1592 | |
| 1593 | /* |
| 1594 | * Do we miss much more than hit in this file? If so, |
| 1595 | * stop bothering with read-ahead. It will only hurt. |
| 1596 | */ |
| 1597 | if (ra->mmap_miss > MMAP_LOTSAMISS) |
| 1598 | return; |
| 1599 | |
| 1600 | /* |
| 1601 | * mmap read-around |
| 1602 | */ |
| 1603 | |
| 1604 | #if CONFIG_MMAP_READAROUND_LIMIT == 0 |
| 1605 | ra_pages = max_sane_readahead(ra->ra_pages); |
| 1606 | #else |
| 1607 | if (ra->ra_pages > CONFIG_MMAP_READAROUND_LIMIT) |
| 1608 | ra_pages = max_sane_readahead(CONFIG_MMAP_READAROUND_LIMIT); |
| 1609 | else |
| 1610 | ra_pages = max_sane_readahead(ra->ra_pages); |
| 1611 | #endif |
| 1612 | |
| 1613 | ra->start = max_t(long, 0, offset - ra_pages / 2); |
| 1614 | ra->size = ra_pages; |
| 1615 | ra->async_size = ra_pages / 4; |
| 1616 | ra_submit(ra, mapping, file); |
| 1617 | } |
| 1618 | |
| 1619 | /* |
| 1620 | * Asynchronous readahead happens when we find the page and PG_readahead, |
| 1621 | * so we want to possibly extend the readahead further.. |
| 1622 | */ |
| 1623 | static void do_async_mmap_readahead(struct vm_area_struct *vma, |
| 1624 | struct file_ra_state *ra, |
| 1625 | struct file *file, |
| 1626 | struct page *page, |
| 1627 | pgoff_t offset) |
| 1628 | { |
| 1629 | struct address_space *mapping = file->f_mapping; |
| 1630 | |
| 1631 | /* If we don't want any read-ahead, don't bother */ |
| 1632 | if (VM_RandomReadHint(vma)) |
| 1633 | return; |
| 1634 | if (ra->mmap_miss > 0) |
| 1635 | ra->mmap_miss--; |
| 1636 | if (PageReadahead(page)) |
| 1637 | page_cache_async_readahead(mapping, ra, file, |
| 1638 | page, offset, ra->ra_pages); |
| 1639 | } |
| 1640 | |
| 1641 | /** |
| 1642 | * filemap_fault - read in file data for page fault handling |
| 1643 | * @vma: vma in which the fault was taken |
| 1644 | * @vmf: struct vm_fault containing details of the fault |
| 1645 | * |
| 1646 | * filemap_fault() is invoked via the vma operations vector for a |
| 1647 | * mapped memory region to read in file data during a page fault. |
| 1648 | * |
| 1649 | * The goto's are kind of ugly, but this streamlines the normal case of having |
| 1650 | * it in the page cache, and handles the special cases reasonably without |
| 1651 | * having a lot of duplicated code. |
| 1652 | */ |
| 1653 | int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
| 1654 | { |
| 1655 | int error; |
| 1656 | struct file *file = vma->vm_file; |
| 1657 | struct address_space *mapping = file->f_mapping; |
| 1658 | struct file_ra_state *ra = &file->f_ra; |
| 1659 | struct inode *inode = mapping->host; |
| 1660 | pgoff_t offset = vmf->pgoff; |
| 1661 | struct page *page; |
| 1662 | pgoff_t size; |
| 1663 | int ret = 0; |
| 1664 | |
| 1665 | size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
| 1666 | if (offset >= size) |
| 1667 | return VM_FAULT_SIGBUS; |
| 1668 | |
| 1669 | /* |
| 1670 | * Do we have something in the page cache already? |
| 1671 | */ |
| 1672 | page = find_get_page(mapping, offset); |
| 1673 | if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) { |
| 1674 | /* |
| 1675 | * We found the page, so try async readahead before |
| 1676 | * waiting for the lock. |
| 1677 | */ |
| 1678 | do_async_mmap_readahead(vma, ra, file, page, offset); |
| 1679 | } else if (!page) { |
| 1680 | /* No page in the page cache at all */ |
| 1681 | do_sync_mmap_readahead(vma, ra, file, offset); |
| 1682 | count_vm_event(PGMAJFAULT); |
| 1683 | mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT); |
| 1684 | ret = VM_FAULT_MAJOR; |
| 1685 | retry_find: |
| 1686 | page = find_get_page(mapping, offset); |
| 1687 | if (!page) |
| 1688 | goto no_cached_page; |
| 1689 | } |
| 1690 | |
| 1691 | if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) { |
| 1692 | page_cache_release(page); |
| 1693 | return ret | VM_FAULT_RETRY; |
| 1694 | } |
| 1695 | |
| 1696 | /* Did it get truncated? */ |
| 1697 | if (unlikely(page->mapping != mapping)) { |
| 1698 | unlock_page(page); |
| 1699 | put_page(page); |
| 1700 | goto retry_find; |
| 1701 | } |
| 1702 | VM_BUG_ON(page->index != offset); |
| 1703 | |
| 1704 | /* |
| 1705 | * We have a locked page in the page cache, now we need to check |
| 1706 | * that it's up-to-date. If not, it is going to be due to an error. |
| 1707 | */ |
| 1708 | if (unlikely(!PageUptodate(page))) |
| 1709 | goto page_not_uptodate; |
| 1710 | |
| 1711 | /* |
| 1712 | * Found the page and have a reference on it. |
| 1713 | * We must recheck i_size under page lock. |
| 1714 | */ |
| 1715 | size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
| 1716 | if (unlikely(offset >= size)) { |
| 1717 | unlock_page(page); |
| 1718 | page_cache_release(page); |
| 1719 | return VM_FAULT_SIGBUS; |
| 1720 | } |
| 1721 | |
| 1722 | vmf->page = page; |
| 1723 | return ret | VM_FAULT_LOCKED; |
| 1724 | |
| 1725 | no_cached_page: |
| 1726 | /* |
| 1727 | * We're only likely to ever get here if MADV_RANDOM is in |
| 1728 | * effect. |
| 1729 | */ |
| 1730 | error = page_cache_read(file, offset); |
| 1731 | |
| 1732 | /* |
| 1733 | * The page we want has now been added to the page cache. |
| 1734 | * In the unlikely event that someone removed it in the |
| 1735 | * meantime, we'll just come back here and read it again. |
| 1736 | */ |
| 1737 | if (error >= 0) |
| 1738 | goto retry_find; |
| 1739 | |
| 1740 | /* |
| 1741 | * An error return from page_cache_read can result if the |
| 1742 | * system is low on memory, or a problem occurs while trying |
| 1743 | * to schedule I/O. |
| 1744 | */ |
| 1745 | if (error == -ENOMEM) |
| 1746 | return VM_FAULT_OOM; |
| 1747 | return VM_FAULT_SIGBUS; |
| 1748 | |
| 1749 | page_not_uptodate: |
| 1750 | /* |
| 1751 | * Umm, take care of errors if the page isn't up-to-date. |
| 1752 | * Try to re-read it _once_. We do this synchronously, |
| 1753 | * because there really aren't any performance issues here |
| 1754 | * and we need to check for errors. |
| 1755 | */ |
| 1756 | ClearPageError(page); |
| 1757 | error = mapping->a_ops->readpage(file, page); |
| 1758 | if (!error) { |
| 1759 | wait_on_page_locked(page); |
| 1760 | if (!PageUptodate(page)) |
| 1761 | error = -EIO; |
| 1762 | } |
| 1763 | page_cache_release(page); |
| 1764 | |
| 1765 | if (!error || error == AOP_TRUNCATED_PAGE) |
| 1766 | goto retry_find; |
| 1767 | |
| 1768 | /* Things didn't work out. Return zero to tell the mm layer so. */ |
| 1769 | shrink_readahead_size_eio(file, ra); |
| 1770 | return VM_FAULT_SIGBUS; |
| 1771 | } |
| 1772 | EXPORT_SYMBOL(filemap_fault); |
| 1773 | |
| 1774 | int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
| 1775 | { |
| 1776 | struct page *page = vmf->page; |
| 1777 | struct inode *inode = file_inode(vma->vm_file); |
| 1778 | int ret = VM_FAULT_LOCKED; |
| 1779 | |
| 1780 | sb_start_pagefault(inode->i_sb); |
| 1781 | file_update_time(vma->vm_file); |
| 1782 | lock_page(page); |
| 1783 | if (page->mapping != inode->i_mapping) { |
| 1784 | unlock_page(page); |
| 1785 | ret = VM_FAULT_NOPAGE; |
| 1786 | goto out; |
| 1787 | } |
| 1788 | /* |
| 1789 | * We mark the page dirty already here so that when freeze is in |
| 1790 | * progress, we are guaranteed that writeback during freezing will |
| 1791 | * see the dirty page and writeprotect it again. |
| 1792 | */ |
| 1793 | set_page_dirty(page); |
| 1794 | wait_for_stable_page(page); |
| 1795 | out: |
| 1796 | sb_end_pagefault(inode->i_sb); |
| 1797 | return ret; |
| 1798 | } |
| 1799 | EXPORT_SYMBOL(filemap_page_mkwrite); |
| 1800 | |
| 1801 | const struct vm_operations_struct generic_file_vm_ops = { |
| 1802 | .fault = filemap_fault, |
| 1803 | .page_mkwrite = filemap_page_mkwrite, |
| 1804 | .remap_pages = generic_file_remap_pages, |
| 1805 | }; |
| 1806 | |
| 1807 | /* This is used for a general mmap of a disk file */ |
| 1808 | |
| 1809 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) |
| 1810 | { |
| 1811 | struct address_space *mapping = file->f_mapping; |
| 1812 | |
| 1813 | if (!mapping->a_ops->readpage) |
| 1814 | return -ENOEXEC; |
| 1815 | file_accessed(file); |
| 1816 | vma->vm_ops = &generic_file_vm_ops; |
| 1817 | return 0; |
| 1818 | } |
| 1819 | |
| 1820 | /* |
| 1821 | * This is for filesystems which do not implement ->writepage. |
| 1822 | */ |
| 1823 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) |
| 1824 | { |
| 1825 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) |
| 1826 | return -EINVAL; |
| 1827 | return generic_file_mmap(file, vma); |
| 1828 | } |
| 1829 | #else |
| 1830 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) |
| 1831 | { |
| 1832 | return -ENOSYS; |
| 1833 | } |
| 1834 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) |
| 1835 | { |
| 1836 | return -ENOSYS; |
| 1837 | } |
| 1838 | #endif /* CONFIG_MMU */ |
| 1839 | |
| 1840 | EXPORT_SYMBOL(generic_file_mmap); |
| 1841 | EXPORT_SYMBOL(generic_file_readonly_mmap); |
| 1842 | |
| 1843 | static struct page *__read_cache_page(struct address_space *mapping, |
| 1844 | pgoff_t index, |
| 1845 | int (*filler)(void *, struct page *), |
| 1846 | void *data, |
| 1847 | gfp_t gfp) |
| 1848 | { |
| 1849 | struct page *page; |
| 1850 | int err; |
| 1851 | repeat: |
| 1852 | page = find_get_page(mapping, index); |
| 1853 | if (!page) { |
| 1854 | page = __page_cache_alloc(gfp | __GFP_COLD); |
| 1855 | if (!page) |
| 1856 | return ERR_PTR(-ENOMEM); |
| 1857 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
| 1858 | if (unlikely(err)) { |
| 1859 | page_cache_release(page); |
| 1860 | if (err == -EEXIST) |
| 1861 | goto repeat; |
| 1862 | /* Presumably ENOMEM for radix tree node */ |
| 1863 | return ERR_PTR(err); |
| 1864 | } |
| 1865 | err = filler(data, page); |
| 1866 | if (err < 0) { |
| 1867 | page_cache_release(page); |
| 1868 | page = ERR_PTR(err); |
| 1869 | } |
| 1870 | } |
| 1871 | return page; |
| 1872 | } |
| 1873 | |
| 1874 | static struct page *do_read_cache_page(struct address_space *mapping, |
| 1875 | pgoff_t index, |
| 1876 | int (*filler)(void *, struct page *), |
| 1877 | void *data, |
| 1878 | gfp_t gfp) |
| 1879 | |
| 1880 | { |
| 1881 | struct page *page; |
| 1882 | int err; |
| 1883 | |
| 1884 | retry: |
| 1885 | page = __read_cache_page(mapping, index, filler, data, gfp); |
| 1886 | if (IS_ERR(page)) |
| 1887 | return page; |
| 1888 | if (PageUptodate(page)) |
| 1889 | goto out; |
| 1890 | |
| 1891 | lock_page(page); |
| 1892 | if (!page->mapping) { |
| 1893 | unlock_page(page); |
| 1894 | page_cache_release(page); |
| 1895 | goto retry; |
| 1896 | } |
| 1897 | if (PageUptodate(page)) { |
| 1898 | unlock_page(page); |
| 1899 | goto out; |
| 1900 | } |
| 1901 | err = filler(data, page); |
| 1902 | if (err < 0) { |
| 1903 | page_cache_release(page); |
| 1904 | return ERR_PTR(err); |
| 1905 | } |
| 1906 | out: |
| 1907 | mark_page_accessed(page); |
| 1908 | return page; |
| 1909 | } |
| 1910 | |
| 1911 | /** |
| 1912 | * read_cache_page_async - read into page cache, fill it if needed |
| 1913 | * @mapping: the page's address_space |
| 1914 | * @index: the page index |
| 1915 | * @filler: function to perform the read |
| 1916 | * @data: first arg to filler(data, page) function, often left as NULL |
| 1917 | * |
| 1918 | * Same as read_cache_page, but don't wait for page to become unlocked |
| 1919 | * after submitting it to the filler. |
| 1920 | * |
| 1921 | * Read into the page cache. If a page already exists, and PageUptodate() is |
| 1922 | * not set, try to fill the page but don't wait for it to become unlocked. |
| 1923 | * |
| 1924 | * If the page does not get brought uptodate, return -EIO. |
| 1925 | */ |
| 1926 | struct page *read_cache_page_async(struct address_space *mapping, |
| 1927 | pgoff_t index, |
| 1928 | int (*filler)(void *, struct page *), |
| 1929 | void *data) |
| 1930 | { |
| 1931 | return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping)); |
| 1932 | } |
| 1933 | EXPORT_SYMBOL(read_cache_page_async); |
| 1934 | |
| 1935 | static struct page *wait_on_page_read(struct page *page) |
| 1936 | { |
| 1937 | if (!IS_ERR(page)) { |
| 1938 | wait_on_page_locked(page); |
| 1939 | if (!PageUptodate(page)) { |
| 1940 | page_cache_release(page); |
| 1941 | page = ERR_PTR(-EIO); |
| 1942 | } |
| 1943 | } |
| 1944 | return page; |
| 1945 | } |
| 1946 | |
| 1947 | /** |
| 1948 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. |
| 1949 | * @mapping: the page's address_space |
| 1950 | * @index: the page index |
| 1951 | * @gfp: the page allocator flags to use if allocating |
| 1952 | * |
| 1953 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with |
| 1954 | * any new page allocations done using the specified allocation flags. |
| 1955 | * |
| 1956 | * If the page does not get brought uptodate, return -EIO. |
| 1957 | */ |
| 1958 | struct page *read_cache_page_gfp(struct address_space *mapping, |
| 1959 | pgoff_t index, |
| 1960 | gfp_t gfp) |
| 1961 | { |
| 1962 | filler_t *filler = (filler_t *)mapping->a_ops->readpage; |
| 1963 | |
| 1964 | return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp)); |
| 1965 | } |
| 1966 | EXPORT_SYMBOL(read_cache_page_gfp); |
| 1967 | |
| 1968 | /** |
| 1969 | * read_cache_page - read into page cache, fill it if needed |
| 1970 | * @mapping: the page's address_space |
| 1971 | * @index: the page index |
| 1972 | * @filler: function to perform the read |
| 1973 | * @data: first arg to filler(data, page) function, often left as NULL |
| 1974 | * |
| 1975 | * Read into the page cache. If a page already exists, and PageUptodate() is |
| 1976 | * not set, try to fill the page then wait for it to become unlocked. |
| 1977 | * |
| 1978 | * If the page does not get brought uptodate, return -EIO. |
| 1979 | */ |
| 1980 | struct page *read_cache_page(struct address_space *mapping, |
| 1981 | pgoff_t index, |
| 1982 | int (*filler)(void *, struct page *), |
| 1983 | void *data) |
| 1984 | { |
| 1985 | return wait_on_page_read(read_cache_page_async(mapping, index, filler, data)); |
| 1986 | } |
| 1987 | EXPORT_SYMBOL(read_cache_page); |
| 1988 | |
| 1989 | static size_t __iovec_copy_from_user_inatomic(char *vaddr, |
| 1990 | const struct iovec *iov, size_t base, size_t bytes) |
| 1991 | { |
| 1992 | size_t copied = 0, left = 0; |
| 1993 | |
| 1994 | while (bytes) { |
| 1995 | char __user *buf = iov->iov_base + base; |
| 1996 | int copy = min(bytes, iov->iov_len - base); |
| 1997 | |
| 1998 | base = 0; |
| 1999 | left = __copy_from_user_inatomic(vaddr, buf, copy); |
| 2000 | copied += copy; |
| 2001 | bytes -= copy; |
| 2002 | vaddr += copy; |
| 2003 | iov++; |
| 2004 | |
| 2005 | if (unlikely(left)) |
| 2006 | break; |
| 2007 | } |
| 2008 | return copied - left; |
| 2009 | } |
| 2010 | |
| 2011 | /* |
| 2012 | * Copy as much as we can into the page and return the number of bytes which |
| 2013 | * were successfully copied. If a fault is encountered then return the number of |
| 2014 | * bytes which were copied. |
| 2015 | */ |
| 2016 | size_t iov_iter_copy_from_user_atomic(struct page *page, |
| 2017 | struct iov_iter *i, unsigned long offset, size_t bytes) |
| 2018 | { |
| 2019 | char *kaddr; |
| 2020 | size_t copied; |
| 2021 | |
| 2022 | BUG_ON(!in_atomic()); |
| 2023 | kaddr = kmap_atomic(page); |
| 2024 | if (likely(i->nr_segs == 1)) { |
| 2025 | int left; |
| 2026 | char __user *buf = i->iov->iov_base + i->iov_offset; |
| 2027 | left = __copy_from_user_inatomic(kaddr + offset, buf, bytes); |
| 2028 | copied = bytes - left; |
| 2029 | } else { |
| 2030 | copied = __iovec_copy_from_user_inatomic(kaddr + offset, |
| 2031 | i->iov, i->iov_offset, bytes); |
| 2032 | } |
| 2033 | kunmap_atomic(kaddr); |
| 2034 | |
| 2035 | return copied; |
| 2036 | } |
| 2037 | EXPORT_SYMBOL(iov_iter_copy_from_user_atomic); |
| 2038 | |
| 2039 | /* |
| 2040 | * This has the same sideeffects and return value as |
| 2041 | * iov_iter_copy_from_user_atomic(). |
| 2042 | * The difference is that it attempts to resolve faults. |
| 2043 | * Page must not be locked. |
| 2044 | */ |
| 2045 | size_t iov_iter_copy_from_user(struct page *page, |
| 2046 | struct iov_iter *i, unsigned long offset, size_t bytes) |
| 2047 | { |
| 2048 | char *kaddr; |
| 2049 | size_t copied; |
| 2050 | |
| 2051 | kaddr = kmap(page); |
| 2052 | if (likely(i->nr_segs == 1)) { |
| 2053 | int left; |
| 2054 | char __user *buf = i->iov->iov_base + i->iov_offset; |
| 2055 | left = __copy_from_user(kaddr + offset, buf, bytes); |
| 2056 | copied = bytes - left; |
| 2057 | } else { |
| 2058 | copied = __iovec_copy_from_user_inatomic(kaddr + offset, |
| 2059 | i->iov, i->iov_offset, bytes); |
| 2060 | } |
| 2061 | kunmap(page); |
| 2062 | return copied; |
| 2063 | } |
| 2064 | EXPORT_SYMBOL(iov_iter_copy_from_user); |
| 2065 | |
| 2066 | void iov_iter_advance(struct iov_iter *i, size_t bytes) |
| 2067 | { |
| 2068 | BUG_ON(i->count < bytes); |
| 2069 | |
| 2070 | if (likely(i->nr_segs == 1)) { |
| 2071 | i->iov_offset += bytes; |
| 2072 | i->count -= bytes; |
| 2073 | } else { |
| 2074 | const struct iovec *iov = i->iov; |
| 2075 | size_t base = i->iov_offset; |
| 2076 | unsigned long nr_segs = i->nr_segs; |
| 2077 | |
| 2078 | /* |
| 2079 | * The !iov->iov_len check ensures we skip over unlikely |
| 2080 | * zero-length segments (without overruning the iovec). |
| 2081 | */ |
| 2082 | while (bytes || unlikely(i->count && !iov->iov_len)) { |
| 2083 | int copy; |
| 2084 | |
| 2085 | copy = min(bytes, iov->iov_len - base); |
| 2086 | BUG_ON(!i->count || i->count < copy); |
| 2087 | i->count -= copy; |
| 2088 | bytes -= copy; |
| 2089 | base += copy; |
| 2090 | if (iov->iov_len == base) { |
| 2091 | iov++; |
| 2092 | nr_segs--; |
| 2093 | base = 0; |
| 2094 | } |
| 2095 | } |
| 2096 | i->iov = iov; |
| 2097 | i->iov_offset = base; |
| 2098 | i->nr_segs = nr_segs; |
| 2099 | } |
| 2100 | } |
| 2101 | EXPORT_SYMBOL(iov_iter_advance); |
| 2102 | |
| 2103 | /* |
| 2104 | * Fault in the first iovec of the given iov_iter, to a maximum length |
| 2105 | * of bytes. Returns 0 on success, or non-zero if the memory could not be |
| 2106 | * accessed (ie. because it is an invalid address). |
| 2107 | * |
| 2108 | * writev-intensive code may want this to prefault several iovecs -- that |
| 2109 | * would be possible (callers must not rely on the fact that _only_ the |
| 2110 | * first iovec will be faulted with the current implementation). |
| 2111 | */ |
| 2112 | int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes) |
| 2113 | { |
| 2114 | char __user *buf = i->iov->iov_base + i->iov_offset; |
| 2115 | bytes = min(bytes, i->iov->iov_len - i->iov_offset); |
| 2116 | return fault_in_pages_readable(buf, bytes); |
| 2117 | } |
| 2118 | EXPORT_SYMBOL(iov_iter_fault_in_readable); |
| 2119 | |
| 2120 | /* |
| 2121 | * Return the count of just the current iov_iter segment. |
| 2122 | */ |
| 2123 | size_t iov_iter_single_seg_count(const struct iov_iter *i) |
| 2124 | { |
| 2125 | const struct iovec *iov = i->iov; |
| 2126 | if (i->nr_segs == 1) |
| 2127 | return i->count; |
| 2128 | else |
| 2129 | return min(i->count, iov->iov_len - i->iov_offset); |
| 2130 | } |
| 2131 | EXPORT_SYMBOL(iov_iter_single_seg_count); |
| 2132 | |
| 2133 | /* |
| 2134 | * Performs necessary checks before doing a write |
| 2135 | * |
| 2136 | * Can adjust writing position or amount of bytes to write. |
| 2137 | * Returns appropriate error code that caller should return or |
| 2138 | * zero in case that write should be allowed. |
| 2139 | */ |
| 2140 | inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk) |
| 2141 | { |
| 2142 | struct inode *inode = file->f_mapping->host; |
| 2143 | unsigned long limit = rlimit(RLIMIT_FSIZE); |
| 2144 | |
| 2145 | if (unlikely(*pos < 0)) |
| 2146 | return -EINVAL; |
| 2147 | |
| 2148 | if (!isblk) { |
| 2149 | /* FIXME: this is for backwards compatibility with 2.4 */ |
| 2150 | if (file->f_flags & O_APPEND) |
| 2151 | *pos = i_size_read(inode); |
| 2152 | |
| 2153 | if (limit != RLIM_INFINITY) { |
| 2154 | if (*pos >= limit) { |
| 2155 | send_sig(SIGXFSZ, current, 0); |
| 2156 | return -EFBIG; |
| 2157 | } |
| 2158 | if (*count > limit - (typeof(limit))*pos) { |
| 2159 | *count = limit - (typeof(limit))*pos; |
| 2160 | } |
| 2161 | } |
| 2162 | } |
| 2163 | |
| 2164 | /* |
| 2165 | * LFS rule |
| 2166 | */ |
| 2167 | if (unlikely(*pos + *count > MAX_NON_LFS && |
| 2168 | !(file->f_flags & O_LARGEFILE))) { |
| 2169 | if (*pos >= MAX_NON_LFS) { |
| 2170 | return -EFBIG; |
| 2171 | } |
| 2172 | if (*count > MAX_NON_LFS - (unsigned long)*pos) { |
| 2173 | *count = MAX_NON_LFS - (unsigned long)*pos; |
| 2174 | } |
| 2175 | } |
| 2176 | |
| 2177 | /* |
| 2178 | * Are we about to exceed the fs block limit ? |
| 2179 | * |
| 2180 | * If we have written data it becomes a short write. If we have |
| 2181 | * exceeded without writing data we send a signal and return EFBIG. |
| 2182 | * Linus frestrict idea will clean these up nicely.. |
| 2183 | */ |
| 2184 | if (likely(!isblk)) { |
| 2185 | if (unlikely(*pos >= inode->i_sb->s_maxbytes)) { |
| 2186 | if (*count || *pos > inode->i_sb->s_maxbytes) { |
| 2187 | return -EFBIG; |
| 2188 | } |
| 2189 | /* zero-length writes at ->s_maxbytes are OK */ |
| 2190 | } |
| 2191 | |
| 2192 | if (unlikely(*pos + *count > inode->i_sb->s_maxbytes)) |
| 2193 | *count = inode->i_sb->s_maxbytes - *pos; |
| 2194 | } else { |
| 2195 | #ifdef CONFIG_BLOCK |
| 2196 | loff_t isize; |
| 2197 | if (bdev_read_only(I_BDEV(inode))) |
| 2198 | return -EPERM; |
| 2199 | isize = i_size_read(inode); |
| 2200 | if (*pos >= isize) { |
| 2201 | if (*count || *pos > isize) |
| 2202 | return -ENOSPC; |
| 2203 | } |
| 2204 | |
| 2205 | if (*pos + *count > isize) |
| 2206 | *count = isize - *pos; |
| 2207 | #else |
| 2208 | return -EPERM; |
| 2209 | #endif |
| 2210 | } |
| 2211 | return 0; |
| 2212 | } |
| 2213 | EXPORT_SYMBOL(generic_write_checks); |
| 2214 | |
| 2215 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
| 2216 | loff_t pos, unsigned len, unsigned flags, |
| 2217 | struct page **pagep, void **fsdata) |
| 2218 | { |
| 2219 | const struct address_space_operations *aops = mapping->a_ops; |
| 2220 | |
| 2221 | return aops->write_begin(file, mapping, pos, len, flags, |
| 2222 | pagep, fsdata); |
| 2223 | } |
| 2224 | EXPORT_SYMBOL(pagecache_write_begin); |
| 2225 | |
| 2226 | int pagecache_write_end(struct file *file, struct address_space *mapping, |
| 2227 | loff_t pos, unsigned len, unsigned copied, |
| 2228 | struct page *page, void *fsdata) |
| 2229 | { |
| 2230 | const struct address_space_operations *aops = mapping->a_ops; |
| 2231 | |
| 2232 | mark_page_accessed(page); |
| 2233 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); |
| 2234 | } |
| 2235 | EXPORT_SYMBOL(pagecache_write_end); |
| 2236 | |
| 2237 | ssize_t |
| 2238 | generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov, |
| 2239 | unsigned long *nr_segs, loff_t pos, loff_t *ppos, |
| 2240 | size_t count, size_t ocount) |
| 2241 | { |
| 2242 | struct file *file = iocb->ki_filp; |
| 2243 | struct address_space *mapping = file->f_mapping; |
| 2244 | struct inode *inode = mapping->host; |
| 2245 | ssize_t written; |
| 2246 | size_t write_len; |
| 2247 | pgoff_t end; |
| 2248 | |
| 2249 | if (count != ocount) |
| 2250 | *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count); |
| 2251 | |
| 2252 | write_len = iov_length(iov, *nr_segs); |
| 2253 | end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT; |
| 2254 | |
| 2255 | written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1); |
| 2256 | if (written) |
| 2257 | goto out; |
| 2258 | |
| 2259 | /* |
| 2260 | * After a write we want buffered reads to be sure to go to disk to get |
| 2261 | * the new data. We invalidate clean cached page from the region we're |
| 2262 | * about to write. We do this *before* the write so that we can return |
| 2263 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
| 2264 | */ |
| 2265 | if (mapping->nrpages) { |
| 2266 | written = invalidate_inode_pages2_range(mapping, |
| 2267 | pos >> PAGE_CACHE_SHIFT, end); |
| 2268 | /* |
| 2269 | * If a page can not be invalidated, return 0 to fall back |
| 2270 | * to buffered write. |
| 2271 | */ |
| 2272 | if (written) { |
| 2273 | if (written == -EBUSY) |
| 2274 | return 0; |
| 2275 | goto out; |
| 2276 | } |
| 2277 | } |
| 2278 | |
| 2279 | written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs); |
| 2280 | |
| 2281 | /* |
| 2282 | * Finally, try again to invalidate clean pages which might have been |
| 2283 | * cached by non-direct readahead, or faulted in by get_user_pages() |
| 2284 | * if the source of the write was an mmap'ed region of the file |
| 2285 | * we're writing. Either one is a pretty crazy thing to do, |
| 2286 | * so we don't support it 100%. If this invalidation |
| 2287 | * fails, tough, the write still worked... |
| 2288 | */ |
| 2289 | if (mapping->nrpages) { |
| 2290 | invalidate_inode_pages2_range(mapping, |
| 2291 | pos >> PAGE_CACHE_SHIFT, end); |
| 2292 | } |
| 2293 | |
| 2294 | if (written > 0) { |
| 2295 | pos += written; |
| 2296 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
| 2297 | i_size_write(inode, pos); |
| 2298 | mark_inode_dirty(inode); |
| 2299 | } |
| 2300 | *ppos = pos; |
| 2301 | } |
| 2302 | out: |
| 2303 | return written; |
| 2304 | } |
| 2305 | EXPORT_SYMBOL(generic_file_direct_write); |
| 2306 | |
| 2307 | /* |
| 2308 | * Find or create a page at the given pagecache position. Return the locked |
| 2309 | * page. This function is specifically for buffered writes. |
| 2310 | */ |
| 2311 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
| 2312 | pgoff_t index, unsigned flags) |
| 2313 | { |
| 2314 | int status; |
| 2315 | gfp_t gfp_mask; |
| 2316 | struct page *page; |
| 2317 | gfp_t gfp_notmask = 0; |
| 2318 | |
| 2319 | gfp_mask = mapping_gfp_mask(mapping); |
| 2320 | if (mapping_cap_account_dirty(mapping)) |
| 2321 | gfp_mask |= __GFP_WRITE; |
| 2322 | if (flags & AOP_FLAG_NOFS) |
| 2323 | gfp_notmask = __GFP_FS; |
| 2324 | repeat: |
| 2325 | page = find_lock_page(mapping, index); |
| 2326 | if (page) |
| 2327 | goto found; |
| 2328 | |
| 2329 | retry: |
| 2330 | page = __page_cache_alloc(gfp_mask & ~gfp_notmask); |
| 2331 | if (!page) |
| 2332 | return NULL; |
| 2333 | |
| 2334 | if (is_cma_pageblock(page)) { |
| 2335 | __free_page(page); |
| 2336 | gfp_notmask |= __GFP_MOVABLE; |
| 2337 | goto retry; |
| 2338 | } |
| 2339 | |
| 2340 | status = add_to_page_cache_lru(page, mapping, index, |
| 2341 | GFP_KERNEL & ~gfp_notmask); |
| 2342 | if (unlikely(status)) { |
| 2343 | page_cache_release(page); |
| 2344 | if (status == -EEXIST) |
| 2345 | goto repeat; |
| 2346 | return NULL; |
| 2347 | } |
| 2348 | found: |
| 2349 | wait_for_stable_page(page); |
| 2350 | return page; |
| 2351 | } |
| 2352 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
| 2353 | |
| 2354 | static ssize_t generic_perform_write(struct file *file, |
| 2355 | struct iov_iter *i, loff_t pos) |
| 2356 | { |
| 2357 | struct address_space *mapping = file->f_mapping; |
| 2358 | const struct address_space_operations *a_ops = mapping->a_ops; |
| 2359 | long status = 0; |
| 2360 | ssize_t written = 0; |
| 2361 | unsigned int flags = 0; |
| 2362 | |
| 2363 | /* |
| 2364 | * Copies from kernel address space cannot fail (NFSD is a big user). |
| 2365 | */ |
| 2366 | if (segment_eq(get_fs(), KERNEL_DS)) |
| 2367 | flags |= AOP_FLAG_UNINTERRUPTIBLE; |
| 2368 | |
| 2369 | do { |
| 2370 | struct page *page; |
| 2371 | unsigned long offset; /* Offset into pagecache page */ |
| 2372 | unsigned long bytes; /* Bytes to write to page */ |
| 2373 | size_t copied; /* Bytes copied from user */ |
| 2374 | void *fsdata; |
| 2375 | |
| 2376 | offset = (pos & (PAGE_CACHE_SIZE - 1)); |
| 2377 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, |
| 2378 | iov_iter_count(i)); |
| 2379 | |
| 2380 | again: |
| 2381 | /* |
| 2382 | * Bring in the user page that we will copy from _first_. |
| 2383 | * Otherwise there's a nasty deadlock on copying from the |
| 2384 | * same page as we're writing to, without it being marked |
| 2385 | * up-to-date. |
| 2386 | * |
| 2387 | * Not only is this an optimisation, but it is also required |
| 2388 | * to check that the address is actually valid, when atomic |
| 2389 | * usercopies are used, below. |
| 2390 | */ |
| 2391 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { |
| 2392 | status = -EFAULT; |
| 2393 | break; |
| 2394 | } |
| 2395 | |
| 2396 | if (fatal_signal_pending(current)) { |
| 2397 | status = -EINTR; |
| 2398 | break; |
| 2399 | } |
| 2400 | |
| 2401 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
| 2402 | &page, &fsdata); |
| 2403 | if (unlikely(status)) |
| 2404 | break; |
| 2405 | |
| 2406 | if (mapping_writably_mapped(mapping)) |
| 2407 | flush_dcache_page(page); |
| 2408 | |
| 2409 | pagefault_disable(); |
| 2410 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); |
| 2411 | pagefault_enable(); |
| 2412 | flush_dcache_page(page); |
| 2413 | |
| 2414 | mark_page_accessed(page); |
| 2415 | status = a_ops->write_end(file, mapping, pos, bytes, copied, |
| 2416 | page, fsdata); |
| 2417 | if (unlikely(status < 0)) |
| 2418 | break; |
| 2419 | copied = status; |
| 2420 | |
| 2421 | cond_resched(); |
| 2422 | |
| 2423 | iov_iter_advance(i, copied); |
| 2424 | if (unlikely(copied == 0)) { |
| 2425 | /* |
| 2426 | * If we were unable to copy any data at all, we must |
| 2427 | * fall back to a single segment length write. |
| 2428 | * |
| 2429 | * If we didn't fallback here, we could livelock |
| 2430 | * because not all segments in the iov can be copied at |
| 2431 | * once without a pagefault. |
| 2432 | */ |
| 2433 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, |
| 2434 | iov_iter_single_seg_count(i)); |
| 2435 | goto again; |
| 2436 | } |
| 2437 | pos += copied; |
| 2438 | written += copied; |
| 2439 | |
| 2440 | balance_dirty_pages_ratelimited(mapping); |
| 2441 | } while (iov_iter_count(i)); |
| 2442 | |
| 2443 | return written ? written : status; |
| 2444 | } |
| 2445 | |
| 2446 | ssize_t |
| 2447 | generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov, |
| 2448 | unsigned long nr_segs, loff_t pos, loff_t *ppos, |
| 2449 | size_t count, ssize_t written) |
| 2450 | { |
| 2451 | struct file *file = iocb->ki_filp; |
| 2452 | ssize_t status; |
| 2453 | struct iov_iter i; |
| 2454 | |
| 2455 | iov_iter_init(&i, iov, nr_segs, count, written); |
| 2456 | status = generic_perform_write(file, &i, pos); |
| 2457 | |
| 2458 | if (likely(status >= 0)) { |
| 2459 | written += status; |
| 2460 | *ppos = pos + status; |
| 2461 | } |
| 2462 | |
| 2463 | return written ? written : status; |
| 2464 | } |
| 2465 | EXPORT_SYMBOL(generic_file_buffered_write); |
| 2466 | |
| 2467 | /** |
| 2468 | * __generic_file_aio_write - write data to a file |
| 2469 | * @iocb: IO state structure (file, offset, etc.) |
| 2470 | * @iov: vector with data to write |
| 2471 | * @nr_segs: number of segments in the vector |
| 2472 | * @ppos: position where to write |
| 2473 | * |
| 2474 | * This function does all the work needed for actually writing data to a |
| 2475 | * file. It does all basic checks, removes SUID from the file, updates |
| 2476 | * modification times and calls proper subroutines depending on whether we |
| 2477 | * do direct IO or a standard buffered write. |
| 2478 | * |
| 2479 | * It expects i_mutex to be grabbed unless we work on a block device or similar |
| 2480 | * object which does not need locking at all. |
| 2481 | * |
| 2482 | * This function does *not* take care of syncing data in case of O_SYNC write. |
| 2483 | * A caller has to handle it. This is mainly due to the fact that we want to |
| 2484 | * avoid syncing under i_mutex. |
| 2485 | */ |
| 2486 | ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov, |
| 2487 | unsigned long nr_segs, loff_t *ppos) |
| 2488 | { |
| 2489 | struct file *file = iocb->ki_filp; |
| 2490 | struct address_space * mapping = file->f_mapping; |
| 2491 | size_t ocount; /* original count */ |
| 2492 | size_t count; /* after file limit checks */ |
| 2493 | struct inode *inode = mapping->host; |
| 2494 | loff_t pos; |
| 2495 | ssize_t written; |
| 2496 | ssize_t err; |
| 2497 | |
| 2498 | ocount = 0; |
| 2499 | err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ); |
| 2500 | if (err) |
| 2501 | return err; |
| 2502 | |
| 2503 | count = ocount; |
| 2504 | pos = *ppos; |
| 2505 | |
| 2506 | /* We can write back this queue in page reclaim */ |
| 2507 | current->backing_dev_info = mapping->backing_dev_info; |
| 2508 | written = 0; |
| 2509 | |
| 2510 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); |
| 2511 | if (err) |
| 2512 | goto out; |
| 2513 | |
| 2514 | if (count == 0) |
| 2515 | goto out; |
| 2516 | |
| 2517 | err = file_remove_suid(file); |
| 2518 | if (err) |
| 2519 | goto out; |
| 2520 | |
| 2521 | err = file_update_time(file); |
| 2522 | if (err) |
| 2523 | goto out; |
| 2524 | |
| 2525 | /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ |
| 2526 | if (unlikely(file->f_flags & O_DIRECT)) { |
| 2527 | loff_t endbyte; |
| 2528 | ssize_t written_buffered; |
| 2529 | |
| 2530 | written = generic_file_direct_write(iocb, iov, &nr_segs, pos, |
| 2531 | ppos, count, ocount); |
| 2532 | if (written < 0 || written == count) |
| 2533 | goto out; |
| 2534 | /* |
| 2535 | * direct-io write to a hole: fall through to buffered I/O |
| 2536 | * for completing the rest of the request. |
| 2537 | */ |
| 2538 | pos += written; |
| 2539 | count -= written; |
| 2540 | written_buffered = generic_file_buffered_write(iocb, iov, |
| 2541 | nr_segs, pos, ppos, count, |
| 2542 | written); |
| 2543 | /* |
| 2544 | * If generic_file_buffered_write() retuned a synchronous error |
| 2545 | * then we want to return the number of bytes which were |
| 2546 | * direct-written, or the error code if that was zero. Note |
| 2547 | * that this differs from normal direct-io semantics, which |
| 2548 | * will return -EFOO even if some bytes were written. |
| 2549 | */ |
| 2550 | if (written_buffered < 0) { |
| 2551 | err = written_buffered; |
| 2552 | goto out; |
| 2553 | } |
| 2554 | |
| 2555 | /* |
| 2556 | * We need to ensure that the page cache pages are written to |
| 2557 | * disk and invalidated to preserve the expected O_DIRECT |
| 2558 | * semantics. |
| 2559 | */ |
| 2560 | endbyte = pos + written_buffered - written - 1; |
| 2561 | err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte); |
| 2562 | if (err == 0) { |
| 2563 | written = written_buffered; |
| 2564 | invalidate_mapping_pages(mapping, |
| 2565 | pos >> PAGE_CACHE_SHIFT, |
| 2566 | endbyte >> PAGE_CACHE_SHIFT); |
| 2567 | } else { |
| 2568 | /* |
| 2569 | * We don't know how much we wrote, so just return |
| 2570 | * the number of bytes which were direct-written |
| 2571 | */ |
| 2572 | } |
| 2573 | } else { |
| 2574 | written = generic_file_buffered_write(iocb, iov, nr_segs, |
| 2575 | pos, ppos, count, written); |
| 2576 | } |
| 2577 | out: |
| 2578 | current->backing_dev_info = NULL; |
| 2579 | return written ? written : err; |
| 2580 | } |
| 2581 | EXPORT_SYMBOL(__generic_file_aio_write); |
| 2582 | |
| 2583 | /** |
| 2584 | * generic_file_aio_write - write data to a file |
| 2585 | * @iocb: IO state structure |
| 2586 | * @iov: vector with data to write |
| 2587 | * @nr_segs: number of segments in the vector |
| 2588 | * @pos: position in file where to write |
| 2589 | * |
| 2590 | * This is a wrapper around __generic_file_aio_write() to be used by most |
| 2591 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
| 2592 | * and acquires i_mutex as needed. |
| 2593 | */ |
| 2594 | ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov, |
| 2595 | unsigned long nr_segs, loff_t pos) |
| 2596 | { |
| 2597 | struct file *file = iocb->ki_filp; |
| 2598 | struct inode *inode = file->f_mapping->host; |
| 2599 | ssize_t ret; |
| 2600 | |
| 2601 | BUG_ON(iocb->ki_pos != pos); |
| 2602 | |
| 2603 | mutex_lock(&inode->i_mutex); |
| 2604 | ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos); |
| 2605 | mutex_unlock(&inode->i_mutex); |
| 2606 | |
| 2607 | if (ret > 0 || ret == -EIOCBQUEUED) { |
| 2608 | ssize_t err; |
| 2609 | |
| 2610 | err = generic_write_sync(file, pos, ret); |
| 2611 | if (err < 0 && ret > 0) |
| 2612 | ret = err; |
| 2613 | } |
| 2614 | return ret; |
| 2615 | } |
| 2616 | EXPORT_SYMBOL(generic_file_aio_write); |
| 2617 | |
| 2618 | /** |
| 2619 | * try_to_release_page() - release old fs-specific metadata on a page |
| 2620 | * |
| 2621 | * @page: the page which the kernel is trying to free |
| 2622 | * @gfp_mask: memory allocation flags (and I/O mode) |
| 2623 | * |
| 2624 | * The address_space is to try to release any data against the page |
| 2625 | * (presumably at page->private). If the release was successful, return `1'. |
| 2626 | * Otherwise return zero. |
| 2627 | * |
| 2628 | * This may also be called if PG_fscache is set on a page, indicating that the |
| 2629 | * page is known to the local caching routines. |
| 2630 | * |
| 2631 | * The @gfp_mask argument specifies whether I/O may be performed to release |
| 2632 | * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS). |
| 2633 | * |
| 2634 | */ |
| 2635 | int try_to_release_page(struct page *page, gfp_t gfp_mask) |
| 2636 | { |
| 2637 | struct address_space * const mapping = page->mapping; |
| 2638 | |
| 2639 | BUG_ON(!PageLocked(page)); |
| 2640 | if (PageWriteback(page)) |
| 2641 | return 0; |
| 2642 | |
| 2643 | if (mapping && mapping->a_ops->releasepage) |
| 2644 | return mapping->a_ops->releasepage(page, gfp_mask); |
| 2645 | return try_to_free_buffers(page); |
| 2646 | } |
| 2647 | |
| 2648 | EXPORT_SYMBOL(try_to_release_page); |