David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0-only |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2 | /* |
| 3 | * linux/mm/filemap.c |
| 4 | * |
| 5 | * Copyright (C) 1994-1999 Linus Torvalds |
| 6 | */ |
| 7 | |
| 8 | /* |
| 9 | * This file handles the generic file mmap semantics used by |
| 10 | * most "normal" filesystems (but you don't /have/ to use this: |
| 11 | * the NFS filesystem used to do this differently, for example) |
| 12 | */ |
| 13 | #include <linux/export.h> |
| 14 | #include <linux/compiler.h> |
| 15 | #include <linux/dax.h> |
| 16 | #include <linux/fs.h> |
| 17 | #include <linux/sched/signal.h> |
| 18 | #include <linux/uaccess.h> |
| 19 | #include <linux/capability.h> |
| 20 | #include <linux/kernel_stat.h> |
| 21 | #include <linux/gfp.h> |
| 22 | #include <linux/mm.h> |
| 23 | #include <linux/swap.h> |
| 24 | #include <linux/mman.h> |
| 25 | #include <linux/pagemap.h> |
| 26 | #include <linux/file.h> |
| 27 | #include <linux/uio.h> |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 28 | #include <linux/error-injection.h> |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 29 | #include <linux/hash.h> |
| 30 | #include <linux/writeback.h> |
| 31 | #include <linux/backing-dev.h> |
| 32 | #include <linux/pagevec.h> |
| 33 | #include <linux/blkdev.h> |
| 34 | #include <linux/security.h> |
| 35 | #include <linux/cpuset.h> |
| 36 | #include <linux/hugetlb.h> |
| 37 | #include <linux/memcontrol.h> |
| 38 | #include <linux/cleancache.h> |
| 39 | #include <linux/shmem_fs.h> |
| 40 | #include <linux/rmap.h> |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 41 | #include <linux/delayacct.h> |
| 42 | #include <linux/psi.h> |
| 43 | #include <linux/ramfs.h> |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 44 | #include "internal.h" |
| 45 | |
| 46 | #define CREATE_TRACE_POINTS |
| 47 | #include <trace/events/filemap.h> |
| 48 | |
| 49 | /* |
| 50 | * FIXME: remove all knowledge of the buffer layer from the core VM |
| 51 | */ |
| 52 | #include <linux/buffer_head.h> /* for try_to_free_buffers */ |
| 53 | |
| 54 | #include <asm/mman.h> |
| 55 | |
| 56 | /* |
| 57 | * Shared mappings implemented 30.11.1994. It's not fully working yet, |
| 58 | * though. |
| 59 | * |
| 60 | * Shared mappings now work. 15.8.1995 Bruno. |
| 61 | * |
| 62 | * finished 'unifying' the page and buffer cache and SMP-threaded the |
| 63 | * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> |
| 64 | * |
| 65 | * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> |
| 66 | */ |
| 67 | |
| 68 | /* |
| 69 | * Lock ordering: |
| 70 | * |
| 71 | * ->i_mmap_rwsem (truncate_pagecache) |
| 72 | * ->private_lock (__free_pte->__set_page_dirty_buffers) |
| 73 | * ->swap_lock (exclusive_swap_page, others) |
| 74 | * ->i_pages lock |
| 75 | * |
| 76 | * ->i_mutex |
| 77 | * ->i_mmap_rwsem (truncate->unmap_mapping_range) |
| 78 | * |
| 79 | * ->mmap_sem |
| 80 | * ->i_mmap_rwsem |
| 81 | * ->page_table_lock or pte_lock (various, mainly in memory.c) |
| 82 | * ->i_pages lock (arch-dependent flush_dcache_mmap_lock) |
| 83 | * |
| 84 | * ->mmap_sem |
| 85 | * ->lock_page (access_process_vm) |
| 86 | * |
| 87 | * ->i_mutex (generic_perform_write) |
| 88 | * ->mmap_sem (fault_in_pages_readable->do_page_fault) |
| 89 | * |
| 90 | * bdi->wb.list_lock |
| 91 | * sb_lock (fs/fs-writeback.c) |
| 92 | * ->i_pages lock (__sync_single_inode) |
| 93 | * |
| 94 | * ->i_mmap_rwsem |
| 95 | * ->anon_vma.lock (vma_adjust) |
| 96 | * |
| 97 | * ->anon_vma.lock |
| 98 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
| 99 | * |
| 100 | * ->page_table_lock or pte_lock |
| 101 | * ->swap_lock (try_to_unmap_one) |
| 102 | * ->private_lock (try_to_unmap_one) |
| 103 | * ->i_pages lock (try_to_unmap_one) |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 104 | * ->pgdat->lru_lock (follow_page->mark_page_accessed) |
| 105 | * ->pgdat->lru_lock (check_pte_range->isolate_lru_page) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 106 | * ->private_lock (page_remove_rmap->set_page_dirty) |
| 107 | * ->i_pages lock (page_remove_rmap->set_page_dirty) |
| 108 | * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) |
| 109 | * ->inode->i_lock (page_remove_rmap->set_page_dirty) |
| 110 | * ->memcg->move_lock (page_remove_rmap->lock_page_memcg) |
| 111 | * bdi.wb->list_lock (zap_pte_range->set_page_dirty) |
| 112 | * ->inode->i_lock (zap_pte_range->set_page_dirty) |
| 113 | * ->private_lock (zap_pte_range->__set_page_dirty_buffers) |
| 114 | * |
| 115 | * ->i_mmap_rwsem |
| 116 | * ->tasklist_lock (memory_failure, collect_procs_ao) |
| 117 | */ |
| 118 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 119 | static void page_cache_delete(struct address_space *mapping, |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 120 | struct page *page, void *shadow) |
| 121 | { |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 122 | XA_STATE(xas, &mapping->i_pages, page->index); |
| 123 | unsigned int nr = 1; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 124 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 125 | mapping_set_update(&xas, mapping); |
| 126 | |
| 127 | /* hugetlb pages are represented by a single entry in the xarray */ |
| 128 | if (!PageHuge(page)) { |
| 129 | xas_set_order(&xas, page->index, compound_order(page)); |
| 130 | nr = compound_nr(page); |
| 131 | } |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 132 | |
| 133 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
| 134 | VM_BUG_ON_PAGE(PageTail(page), page); |
| 135 | VM_BUG_ON_PAGE(nr != 1 && shadow, page); |
| 136 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 137 | xas_store(&xas, shadow); |
| 138 | xas_init_marks(&xas); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 139 | |
| 140 | page->mapping = NULL; |
| 141 | /* Leave page->index set: truncation lookup relies upon it */ |
| 142 | |
| 143 | if (shadow) { |
| 144 | mapping->nrexceptional += nr; |
| 145 | /* |
| 146 | * Make sure the nrexceptional update is committed before |
| 147 | * the nrpages update so that final truncate racing |
| 148 | * with reclaim does not see both counters 0 at the |
| 149 | * same time and miss a shadow entry. |
| 150 | */ |
| 151 | smp_wmb(); |
| 152 | } |
| 153 | mapping->nrpages -= nr; |
| 154 | } |
| 155 | |
| 156 | static void unaccount_page_cache_page(struct address_space *mapping, |
| 157 | struct page *page) |
| 158 | { |
| 159 | int nr; |
| 160 | |
| 161 | /* |
| 162 | * if we're uptodate, flush out into the cleancache, otherwise |
| 163 | * invalidate any existing cleancache entries. We can't leave |
| 164 | * stale data around in the cleancache once our page is gone |
| 165 | */ |
| 166 | if (PageUptodate(page) && PageMappedToDisk(page)) |
| 167 | cleancache_put_page(page); |
| 168 | else |
| 169 | cleancache_invalidate_page(mapping, page); |
| 170 | |
| 171 | VM_BUG_ON_PAGE(PageTail(page), page); |
| 172 | VM_BUG_ON_PAGE(page_mapped(page), page); |
| 173 | if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) { |
| 174 | int mapcount; |
| 175 | |
| 176 | pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n", |
| 177 | current->comm, page_to_pfn(page)); |
| 178 | dump_page(page, "still mapped when deleted"); |
| 179 | dump_stack(); |
| 180 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
| 181 | |
| 182 | mapcount = page_mapcount(page); |
| 183 | if (mapping_exiting(mapping) && |
| 184 | page_count(page) >= mapcount + 2) { |
| 185 | /* |
| 186 | * All vmas have already been torn down, so it's |
| 187 | * a good bet that actually the page is unmapped, |
| 188 | * and we'd prefer not to leak it: if we're wrong, |
| 189 | * some other bad page check should catch it later. |
| 190 | */ |
| 191 | page_mapcount_reset(page); |
| 192 | page_ref_sub(page, mapcount); |
| 193 | } |
| 194 | } |
| 195 | |
| 196 | /* hugetlb pages do not participate in page cache accounting. */ |
| 197 | if (PageHuge(page)) |
| 198 | return; |
| 199 | |
| 200 | nr = hpage_nr_pages(page); |
| 201 | |
| 202 | __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr); |
| 203 | if (PageSwapBacked(page)) { |
| 204 | __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr); |
| 205 | if (PageTransHuge(page)) |
| 206 | __dec_node_page_state(page, NR_SHMEM_THPS); |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 207 | } else if (PageTransHuge(page)) { |
| 208 | __dec_node_page_state(page, NR_FILE_THPS); |
| 209 | filemap_nr_thps_dec(mapping); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 210 | } |
| 211 | |
| 212 | /* |
| 213 | * At this point page must be either written or cleaned by |
| 214 | * truncate. Dirty page here signals a bug and loss of |
| 215 | * unwritten data. |
| 216 | * |
| 217 | * This fixes dirty accounting after removing the page entirely |
| 218 | * but leaves PageDirty set: it has no effect for truncated |
| 219 | * page and anyway will be cleared before returning page into |
| 220 | * buddy allocator. |
| 221 | */ |
| 222 | if (WARN_ON_ONCE(PageDirty(page))) |
| 223 | account_page_cleaned(page, mapping, inode_to_wb(mapping->host)); |
| 224 | } |
| 225 | |
| 226 | /* |
| 227 | * Delete a page from the page cache and free it. Caller has to make |
| 228 | * sure the page is locked and that nobody else uses it - or that usage |
| 229 | * is safe. The caller must hold the i_pages lock. |
| 230 | */ |
| 231 | void __delete_from_page_cache(struct page *page, void *shadow) |
| 232 | { |
| 233 | struct address_space *mapping = page->mapping; |
| 234 | |
| 235 | trace_mm_filemap_delete_from_page_cache(page); |
| 236 | |
| 237 | unaccount_page_cache_page(mapping, page); |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 238 | page_cache_delete(mapping, page, shadow); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 239 | } |
| 240 | |
| 241 | static void page_cache_free_page(struct address_space *mapping, |
| 242 | struct page *page) |
| 243 | { |
| 244 | void (*freepage)(struct page *); |
| 245 | |
| 246 | freepage = mapping->a_ops->freepage; |
| 247 | if (freepage) |
| 248 | freepage(page); |
| 249 | |
| 250 | if (PageTransHuge(page) && !PageHuge(page)) { |
| 251 | page_ref_sub(page, HPAGE_PMD_NR); |
| 252 | VM_BUG_ON_PAGE(page_count(page) <= 0, page); |
| 253 | } else { |
| 254 | put_page(page); |
| 255 | } |
| 256 | } |
| 257 | |
| 258 | /** |
| 259 | * delete_from_page_cache - delete page from page cache |
| 260 | * @page: the page which the kernel is trying to remove from page cache |
| 261 | * |
| 262 | * This must be called only on pages that have been verified to be in the page |
| 263 | * cache and locked. It will never put the page into the free list, the caller |
| 264 | * has a reference on the page. |
| 265 | */ |
| 266 | void delete_from_page_cache(struct page *page) |
| 267 | { |
| 268 | struct address_space *mapping = page_mapping(page); |
| 269 | unsigned long flags; |
| 270 | |
| 271 | BUG_ON(!PageLocked(page)); |
| 272 | xa_lock_irqsave(&mapping->i_pages, flags); |
| 273 | __delete_from_page_cache(page, NULL); |
| 274 | xa_unlock_irqrestore(&mapping->i_pages, flags); |
| 275 | |
| 276 | page_cache_free_page(mapping, page); |
| 277 | } |
| 278 | EXPORT_SYMBOL(delete_from_page_cache); |
| 279 | |
| 280 | /* |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 281 | * page_cache_delete_batch - delete several pages from page cache |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 282 | * @mapping: the mapping to which pages belong |
| 283 | * @pvec: pagevec with pages to delete |
| 284 | * |
| 285 | * The function walks over mapping->i_pages and removes pages passed in @pvec |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 286 | * from the mapping. The function expects @pvec to be sorted by page index |
| 287 | * and is optimised for it to be dense. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 288 | * It tolerates holes in @pvec (mapping entries at those indices are not |
| 289 | * modified). The function expects only THP head pages to be present in the |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 290 | * @pvec. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 291 | * |
| 292 | * The function expects the i_pages lock to be held. |
| 293 | */ |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 294 | static void page_cache_delete_batch(struct address_space *mapping, |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 295 | struct pagevec *pvec) |
| 296 | { |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 297 | XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 298 | int total_pages = 0; |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 299 | int i = 0; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 300 | struct page *page; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 301 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 302 | mapping_set_update(&xas, mapping); |
| 303 | xas_for_each(&xas, page, ULONG_MAX) { |
| 304 | if (i >= pagevec_count(pvec)) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 305 | break; |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 306 | |
| 307 | /* A swap/dax/shadow entry got inserted? Skip it. */ |
| 308 | if (xa_is_value(page)) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 309 | continue; |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 310 | /* |
| 311 | * A page got inserted in our range? Skip it. We have our |
| 312 | * pages locked so they are protected from being removed. |
| 313 | * If we see a page whose index is higher than ours, it |
| 314 | * means our page has been removed, which shouldn't be |
| 315 | * possible because we're holding the PageLock. |
| 316 | */ |
| 317 | if (page != pvec->pages[i]) { |
| 318 | VM_BUG_ON_PAGE(page->index > pvec->pages[i]->index, |
| 319 | page); |
| 320 | continue; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 321 | } |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 322 | |
| 323 | WARN_ON_ONCE(!PageLocked(page)); |
| 324 | |
| 325 | if (page->index == xas.xa_index) |
| 326 | page->mapping = NULL; |
| 327 | /* Leave page->index set: truncation lookup relies on it */ |
| 328 | |
| 329 | /* |
| 330 | * Move to the next page in the vector if this is a regular |
| 331 | * page or the index is of the last sub-page of this compound |
| 332 | * page. |
| 333 | */ |
| 334 | if (page->index + compound_nr(page) - 1 == xas.xa_index) |
| 335 | i++; |
| 336 | xas_store(&xas, NULL); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 337 | total_pages++; |
| 338 | } |
| 339 | mapping->nrpages -= total_pages; |
| 340 | } |
| 341 | |
| 342 | void delete_from_page_cache_batch(struct address_space *mapping, |
| 343 | struct pagevec *pvec) |
| 344 | { |
| 345 | int i; |
| 346 | unsigned long flags; |
| 347 | |
| 348 | if (!pagevec_count(pvec)) |
| 349 | return; |
| 350 | |
| 351 | xa_lock_irqsave(&mapping->i_pages, flags); |
| 352 | for (i = 0; i < pagevec_count(pvec); i++) { |
| 353 | trace_mm_filemap_delete_from_page_cache(pvec->pages[i]); |
| 354 | |
| 355 | unaccount_page_cache_page(mapping, pvec->pages[i]); |
| 356 | } |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 357 | page_cache_delete_batch(mapping, pvec); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 358 | xa_unlock_irqrestore(&mapping->i_pages, flags); |
| 359 | |
| 360 | for (i = 0; i < pagevec_count(pvec); i++) |
| 361 | page_cache_free_page(mapping, pvec->pages[i]); |
| 362 | } |
| 363 | |
| 364 | int filemap_check_errors(struct address_space *mapping) |
| 365 | { |
| 366 | int ret = 0; |
| 367 | /* Check for outstanding write errors */ |
| 368 | if (test_bit(AS_ENOSPC, &mapping->flags) && |
| 369 | test_and_clear_bit(AS_ENOSPC, &mapping->flags)) |
| 370 | ret = -ENOSPC; |
| 371 | if (test_bit(AS_EIO, &mapping->flags) && |
| 372 | test_and_clear_bit(AS_EIO, &mapping->flags)) |
| 373 | ret = -EIO; |
| 374 | return ret; |
| 375 | } |
| 376 | EXPORT_SYMBOL(filemap_check_errors); |
| 377 | |
| 378 | static int filemap_check_and_keep_errors(struct address_space *mapping) |
| 379 | { |
| 380 | /* Check for outstanding write errors */ |
| 381 | if (test_bit(AS_EIO, &mapping->flags)) |
| 382 | return -EIO; |
| 383 | if (test_bit(AS_ENOSPC, &mapping->flags)) |
| 384 | return -ENOSPC; |
| 385 | return 0; |
| 386 | } |
| 387 | |
| 388 | /** |
| 389 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
| 390 | * @mapping: address space structure to write |
| 391 | * @start: offset in bytes where the range starts |
| 392 | * @end: offset in bytes where the range ends (inclusive) |
| 393 | * @sync_mode: enable synchronous operation |
| 394 | * |
| 395 | * Start writeback against all of a mapping's dirty pages that lie |
| 396 | * within the byte offsets <start, end> inclusive. |
| 397 | * |
| 398 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
| 399 | * opposed to a regular memory cleansing writeback. The difference between |
| 400 | * these two operations is that if a dirty page/buffer is encountered, it must |
| 401 | * be waited upon, and not just skipped over. |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 402 | * |
| 403 | * Return: %0 on success, negative error code otherwise. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 404 | */ |
| 405 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
| 406 | loff_t end, int sync_mode) |
| 407 | { |
| 408 | int ret; |
| 409 | struct writeback_control wbc = { |
| 410 | .sync_mode = sync_mode, |
| 411 | .nr_to_write = LONG_MAX, |
| 412 | .range_start = start, |
| 413 | .range_end = end, |
| 414 | }; |
| 415 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 416 | if (!mapping_cap_writeback_dirty(mapping) || |
| 417 | !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 418 | return 0; |
| 419 | |
| 420 | wbc_attach_fdatawrite_inode(&wbc, mapping->host); |
| 421 | ret = do_writepages(mapping, &wbc); |
| 422 | wbc_detach_inode(&wbc); |
| 423 | return ret; |
| 424 | } |
| 425 | |
| 426 | static inline int __filemap_fdatawrite(struct address_space *mapping, |
| 427 | int sync_mode) |
| 428 | { |
| 429 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
| 430 | } |
| 431 | |
| 432 | int filemap_fdatawrite(struct address_space *mapping) |
| 433 | { |
| 434 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); |
| 435 | } |
| 436 | EXPORT_SYMBOL(filemap_fdatawrite); |
| 437 | |
| 438 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
| 439 | loff_t end) |
| 440 | { |
| 441 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); |
| 442 | } |
| 443 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
| 444 | |
| 445 | /** |
| 446 | * filemap_flush - mostly a non-blocking flush |
| 447 | * @mapping: target address_space |
| 448 | * |
| 449 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
| 450 | * purposes - I/O may not be started against all dirty pages. |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 451 | * |
| 452 | * Return: %0 on success, negative error code otherwise. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 453 | */ |
| 454 | int filemap_flush(struct address_space *mapping) |
| 455 | { |
| 456 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); |
| 457 | } |
| 458 | EXPORT_SYMBOL(filemap_flush); |
| 459 | |
| 460 | /** |
| 461 | * filemap_range_has_page - check if a page exists in range. |
| 462 | * @mapping: address space within which to check |
| 463 | * @start_byte: offset in bytes where the range starts |
| 464 | * @end_byte: offset in bytes where the range ends (inclusive) |
| 465 | * |
| 466 | * Find at least one page in the range supplied, usually used to check if |
| 467 | * direct writing in this range will trigger a writeback. |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 468 | * |
| 469 | * Return: %true if at least one page exists in the specified range, |
| 470 | * %false otherwise. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 471 | */ |
| 472 | bool filemap_range_has_page(struct address_space *mapping, |
| 473 | loff_t start_byte, loff_t end_byte) |
| 474 | { |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 475 | struct page *page; |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 476 | XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT); |
| 477 | pgoff_t max = end_byte >> PAGE_SHIFT; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 478 | |
| 479 | if (end_byte < start_byte) |
| 480 | return false; |
| 481 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 482 | rcu_read_lock(); |
| 483 | for (;;) { |
| 484 | page = xas_find(&xas, max); |
| 485 | if (xas_retry(&xas, page)) |
| 486 | continue; |
| 487 | /* Shadow entries don't count */ |
| 488 | if (xa_is_value(page)) |
| 489 | continue; |
| 490 | /* |
| 491 | * We don't need to try to pin this page; we're about to |
| 492 | * release the RCU lock anyway. It is enough to know that |
| 493 | * there was a page here recently. |
| 494 | */ |
| 495 | break; |
| 496 | } |
| 497 | rcu_read_unlock(); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 498 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 499 | return page != NULL; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 500 | } |
| 501 | EXPORT_SYMBOL(filemap_range_has_page); |
| 502 | |
| 503 | static void __filemap_fdatawait_range(struct address_space *mapping, |
| 504 | loff_t start_byte, loff_t end_byte) |
| 505 | { |
| 506 | pgoff_t index = start_byte >> PAGE_SHIFT; |
| 507 | pgoff_t end = end_byte >> PAGE_SHIFT; |
| 508 | struct pagevec pvec; |
| 509 | int nr_pages; |
| 510 | |
| 511 | if (end_byte < start_byte) |
| 512 | return; |
| 513 | |
| 514 | pagevec_init(&pvec); |
| 515 | while (index <= end) { |
| 516 | unsigned i; |
| 517 | |
| 518 | nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, |
| 519 | end, PAGECACHE_TAG_WRITEBACK); |
| 520 | if (!nr_pages) |
| 521 | break; |
| 522 | |
| 523 | for (i = 0; i < nr_pages; i++) { |
| 524 | struct page *page = pvec.pages[i]; |
| 525 | |
| 526 | wait_on_page_writeback(page); |
| 527 | ClearPageError(page); |
| 528 | } |
| 529 | pagevec_release(&pvec); |
| 530 | cond_resched(); |
| 531 | } |
| 532 | } |
| 533 | |
| 534 | /** |
| 535 | * filemap_fdatawait_range - wait for writeback to complete |
| 536 | * @mapping: address space structure to wait for |
| 537 | * @start_byte: offset in bytes where the range starts |
| 538 | * @end_byte: offset in bytes where the range ends (inclusive) |
| 539 | * |
| 540 | * Walk the list of under-writeback pages of the given address space |
| 541 | * in the given range and wait for all of them. Check error status of |
| 542 | * the address space and return it. |
| 543 | * |
| 544 | * Since the error status of the address space is cleared by this function, |
| 545 | * callers are responsible for checking the return value and handling and/or |
| 546 | * reporting the error. |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 547 | * |
| 548 | * Return: error status of the address space. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 549 | */ |
| 550 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, |
| 551 | loff_t end_byte) |
| 552 | { |
| 553 | __filemap_fdatawait_range(mapping, start_byte, end_byte); |
| 554 | return filemap_check_errors(mapping); |
| 555 | } |
| 556 | EXPORT_SYMBOL(filemap_fdatawait_range); |
| 557 | |
| 558 | /** |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 559 | * filemap_fdatawait_range_keep_errors - wait for writeback to complete |
| 560 | * @mapping: address space structure to wait for |
| 561 | * @start_byte: offset in bytes where the range starts |
| 562 | * @end_byte: offset in bytes where the range ends (inclusive) |
| 563 | * |
| 564 | * Walk the list of under-writeback pages of the given address space in the |
| 565 | * given range and wait for all of them. Unlike filemap_fdatawait_range(), |
| 566 | * this function does not clear error status of the address space. |
| 567 | * |
| 568 | * Use this function if callers don't handle errors themselves. Expected |
| 569 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), |
| 570 | * fsfreeze(8) |
| 571 | */ |
| 572 | int filemap_fdatawait_range_keep_errors(struct address_space *mapping, |
| 573 | loff_t start_byte, loff_t end_byte) |
| 574 | { |
| 575 | __filemap_fdatawait_range(mapping, start_byte, end_byte); |
| 576 | return filemap_check_and_keep_errors(mapping); |
| 577 | } |
| 578 | EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors); |
| 579 | |
| 580 | /** |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 581 | * file_fdatawait_range - wait for writeback to complete |
| 582 | * @file: file pointing to address space structure to wait for |
| 583 | * @start_byte: offset in bytes where the range starts |
| 584 | * @end_byte: offset in bytes where the range ends (inclusive) |
| 585 | * |
| 586 | * Walk the list of under-writeback pages of the address space that file |
| 587 | * refers to, in the given range and wait for all of them. Check error |
| 588 | * status of the address space vs. the file->f_wb_err cursor and return it. |
| 589 | * |
| 590 | * Since the error status of the file is advanced by this function, |
| 591 | * callers are responsible for checking the return value and handling and/or |
| 592 | * reporting the error. |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 593 | * |
| 594 | * Return: error status of the address space vs. the file->f_wb_err cursor. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 595 | */ |
| 596 | int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte) |
| 597 | { |
| 598 | struct address_space *mapping = file->f_mapping; |
| 599 | |
| 600 | __filemap_fdatawait_range(mapping, start_byte, end_byte); |
| 601 | return file_check_and_advance_wb_err(file); |
| 602 | } |
| 603 | EXPORT_SYMBOL(file_fdatawait_range); |
| 604 | |
| 605 | /** |
| 606 | * filemap_fdatawait_keep_errors - wait for writeback without clearing errors |
| 607 | * @mapping: address space structure to wait for |
| 608 | * |
| 609 | * Walk the list of under-writeback pages of the given address space |
| 610 | * and wait for all of them. Unlike filemap_fdatawait(), this function |
| 611 | * does not clear error status of the address space. |
| 612 | * |
| 613 | * Use this function if callers don't handle errors themselves. Expected |
| 614 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), |
| 615 | * fsfreeze(8) |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 616 | * |
| 617 | * Return: error status of the address space. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 618 | */ |
| 619 | int filemap_fdatawait_keep_errors(struct address_space *mapping) |
| 620 | { |
| 621 | __filemap_fdatawait_range(mapping, 0, LLONG_MAX); |
| 622 | return filemap_check_and_keep_errors(mapping); |
| 623 | } |
| 624 | EXPORT_SYMBOL(filemap_fdatawait_keep_errors); |
| 625 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 626 | /* Returns true if writeback might be needed or already in progress. */ |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 627 | static bool mapping_needs_writeback(struct address_space *mapping) |
| 628 | { |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 629 | if (dax_mapping(mapping)) |
| 630 | return mapping->nrexceptional; |
| 631 | |
| 632 | return mapping->nrpages; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 633 | } |
| 634 | |
| 635 | int filemap_write_and_wait(struct address_space *mapping) |
| 636 | { |
| 637 | int err = 0; |
| 638 | |
| 639 | if (mapping_needs_writeback(mapping)) { |
| 640 | err = filemap_fdatawrite(mapping); |
| 641 | /* |
| 642 | * Even if the above returned error, the pages may be |
| 643 | * written partially (e.g. -ENOSPC), so we wait for it. |
| 644 | * But the -EIO is special case, it may indicate the worst |
| 645 | * thing (e.g. bug) happened, so we avoid waiting for it. |
| 646 | */ |
| 647 | if (err != -EIO) { |
| 648 | int err2 = filemap_fdatawait(mapping); |
| 649 | if (!err) |
| 650 | err = err2; |
| 651 | } else { |
| 652 | /* Clear any previously stored errors */ |
| 653 | filemap_check_errors(mapping); |
| 654 | } |
| 655 | } else { |
| 656 | err = filemap_check_errors(mapping); |
| 657 | } |
| 658 | return err; |
| 659 | } |
| 660 | EXPORT_SYMBOL(filemap_write_and_wait); |
| 661 | |
| 662 | /** |
| 663 | * filemap_write_and_wait_range - write out & wait on a file range |
| 664 | * @mapping: the address_space for the pages |
| 665 | * @lstart: offset in bytes where the range starts |
| 666 | * @lend: offset in bytes where the range ends (inclusive) |
| 667 | * |
| 668 | * Write out and wait upon file offsets lstart->lend, inclusive. |
| 669 | * |
| 670 | * Note that @lend is inclusive (describes the last byte to be written) so |
| 671 | * that this function can be used to write to the very end-of-file (end = -1). |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 672 | * |
| 673 | * Return: error status of the address space. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 674 | */ |
| 675 | int filemap_write_and_wait_range(struct address_space *mapping, |
| 676 | loff_t lstart, loff_t lend) |
| 677 | { |
| 678 | int err = 0; |
| 679 | |
| 680 | if (mapping_needs_writeback(mapping)) { |
| 681 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
| 682 | WB_SYNC_ALL); |
| 683 | /* See comment of filemap_write_and_wait() */ |
| 684 | if (err != -EIO) { |
| 685 | int err2 = filemap_fdatawait_range(mapping, |
| 686 | lstart, lend); |
| 687 | if (!err) |
| 688 | err = err2; |
| 689 | } else { |
| 690 | /* Clear any previously stored errors */ |
| 691 | filemap_check_errors(mapping); |
| 692 | } |
| 693 | } else { |
| 694 | err = filemap_check_errors(mapping); |
| 695 | } |
| 696 | return err; |
| 697 | } |
| 698 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
| 699 | |
| 700 | void __filemap_set_wb_err(struct address_space *mapping, int err) |
| 701 | { |
| 702 | errseq_t eseq = errseq_set(&mapping->wb_err, err); |
| 703 | |
| 704 | trace_filemap_set_wb_err(mapping, eseq); |
| 705 | } |
| 706 | EXPORT_SYMBOL(__filemap_set_wb_err); |
| 707 | |
| 708 | /** |
| 709 | * file_check_and_advance_wb_err - report wb error (if any) that was previously |
| 710 | * and advance wb_err to current one |
| 711 | * @file: struct file on which the error is being reported |
| 712 | * |
| 713 | * When userland calls fsync (or something like nfsd does the equivalent), we |
| 714 | * want to report any writeback errors that occurred since the last fsync (or |
| 715 | * since the file was opened if there haven't been any). |
| 716 | * |
| 717 | * Grab the wb_err from the mapping. If it matches what we have in the file, |
| 718 | * then just quickly return 0. The file is all caught up. |
| 719 | * |
| 720 | * If it doesn't match, then take the mapping value, set the "seen" flag in |
| 721 | * it and try to swap it into place. If it works, or another task beat us |
| 722 | * to it with the new value, then update the f_wb_err and return the error |
| 723 | * portion. The error at this point must be reported via proper channels |
| 724 | * (a'la fsync, or NFS COMMIT operation, etc.). |
| 725 | * |
| 726 | * While we handle mapping->wb_err with atomic operations, the f_wb_err |
| 727 | * value is protected by the f_lock since we must ensure that it reflects |
| 728 | * the latest value swapped in for this file descriptor. |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 729 | * |
| 730 | * Return: %0 on success, negative error code otherwise. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 731 | */ |
| 732 | int file_check_and_advance_wb_err(struct file *file) |
| 733 | { |
| 734 | int err = 0; |
| 735 | errseq_t old = READ_ONCE(file->f_wb_err); |
| 736 | struct address_space *mapping = file->f_mapping; |
| 737 | |
| 738 | /* Locklessly handle the common case where nothing has changed */ |
| 739 | if (errseq_check(&mapping->wb_err, old)) { |
| 740 | /* Something changed, must use slow path */ |
| 741 | spin_lock(&file->f_lock); |
| 742 | old = file->f_wb_err; |
| 743 | err = errseq_check_and_advance(&mapping->wb_err, |
| 744 | &file->f_wb_err); |
| 745 | trace_file_check_and_advance_wb_err(file, old); |
| 746 | spin_unlock(&file->f_lock); |
| 747 | } |
| 748 | |
| 749 | /* |
| 750 | * We're mostly using this function as a drop in replacement for |
| 751 | * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect |
| 752 | * that the legacy code would have had on these flags. |
| 753 | */ |
| 754 | clear_bit(AS_EIO, &mapping->flags); |
| 755 | clear_bit(AS_ENOSPC, &mapping->flags); |
| 756 | return err; |
| 757 | } |
| 758 | EXPORT_SYMBOL(file_check_and_advance_wb_err); |
| 759 | |
| 760 | /** |
| 761 | * file_write_and_wait_range - write out & wait on a file range |
| 762 | * @file: file pointing to address_space with pages |
| 763 | * @lstart: offset in bytes where the range starts |
| 764 | * @lend: offset in bytes where the range ends (inclusive) |
| 765 | * |
| 766 | * Write out and wait upon file offsets lstart->lend, inclusive. |
| 767 | * |
| 768 | * Note that @lend is inclusive (describes the last byte to be written) so |
| 769 | * that this function can be used to write to the very end-of-file (end = -1). |
| 770 | * |
| 771 | * After writing out and waiting on the data, we check and advance the |
| 772 | * f_wb_err cursor to the latest value, and return any errors detected there. |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 773 | * |
| 774 | * Return: %0 on success, negative error code otherwise. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 775 | */ |
| 776 | int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend) |
| 777 | { |
| 778 | int err = 0, err2; |
| 779 | struct address_space *mapping = file->f_mapping; |
| 780 | |
| 781 | if (mapping_needs_writeback(mapping)) { |
| 782 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
| 783 | WB_SYNC_ALL); |
| 784 | /* See comment of filemap_write_and_wait() */ |
| 785 | if (err != -EIO) |
| 786 | __filemap_fdatawait_range(mapping, lstart, lend); |
| 787 | } |
| 788 | err2 = file_check_and_advance_wb_err(file); |
| 789 | if (!err) |
| 790 | err = err2; |
| 791 | return err; |
| 792 | } |
| 793 | EXPORT_SYMBOL(file_write_and_wait_range); |
| 794 | |
| 795 | /** |
| 796 | * replace_page_cache_page - replace a pagecache page with a new one |
| 797 | * @old: page to be replaced |
| 798 | * @new: page to replace with |
| 799 | * @gfp_mask: allocation mode |
| 800 | * |
| 801 | * This function replaces a page in the pagecache with a new one. On |
| 802 | * success it acquires the pagecache reference for the new page and |
| 803 | * drops it for the old page. Both the old and new pages must be |
| 804 | * locked. This function does not add the new page to the LRU, the |
| 805 | * caller must do that. |
| 806 | * |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 807 | * The remove + add is atomic. This function cannot fail. |
| 808 | * |
| 809 | * Return: %0 |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 810 | */ |
| 811 | int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask) |
| 812 | { |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 813 | struct address_space *mapping = old->mapping; |
| 814 | void (*freepage)(struct page *) = mapping->a_ops->freepage; |
| 815 | pgoff_t offset = old->index; |
| 816 | XA_STATE(xas, &mapping->i_pages, offset); |
| 817 | unsigned long flags; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 818 | |
| 819 | VM_BUG_ON_PAGE(!PageLocked(old), old); |
| 820 | VM_BUG_ON_PAGE(!PageLocked(new), new); |
| 821 | VM_BUG_ON_PAGE(new->mapping, new); |
| 822 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 823 | get_page(new); |
| 824 | new->mapping = mapping; |
| 825 | new->index = offset; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 826 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 827 | xas_lock_irqsave(&xas, flags); |
| 828 | xas_store(&xas, new); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 829 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 830 | old->mapping = NULL; |
| 831 | /* hugetlb pages do not participate in page cache accounting. */ |
| 832 | if (!PageHuge(old)) |
| 833 | __dec_node_page_state(new, NR_FILE_PAGES); |
| 834 | if (!PageHuge(new)) |
| 835 | __inc_node_page_state(new, NR_FILE_PAGES); |
| 836 | if (PageSwapBacked(old)) |
| 837 | __dec_node_page_state(new, NR_SHMEM); |
| 838 | if (PageSwapBacked(new)) |
| 839 | __inc_node_page_state(new, NR_SHMEM); |
| 840 | xas_unlock_irqrestore(&xas, flags); |
| 841 | mem_cgroup_migrate(old, new); |
| 842 | if (freepage) |
| 843 | freepage(old); |
| 844 | put_page(old); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 845 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 846 | return 0; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 847 | } |
| 848 | EXPORT_SYMBOL_GPL(replace_page_cache_page); |
| 849 | |
Olivier Deprez | 0e64123 | 2021-09-23 10:07:05 +0200 | [diff] [blame^] | 850 | noinline int __add_to_page_cache_locked(struct page *page, |
| 851 | struct address_space *mapping, |
| 852 | pgoff_t offset, gfp_t gfp_mask, |
| 853 | void **shadowp) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 854 | { |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 855 | XA_STATE(xas, &mapping->i_pages, offset); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 856 | int huge = PageHuge(page); |
| 857 | struct mem_cgroup *memcg; |
| 858 | int error; |
| 859 | |
| 860 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
| 861 | VM_BUG_ON_PAGE(PageSwapBacked(page), page); |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 862 | mapping_set_update(&xas, mapping); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 863 | |
| 864 | if (!huge) { |
| 865 | error = mem_cgroup_try_charge(page, current->mm, |
| 866 | gfp_mask, &memcg, false); |
| 867 | if (error) |
| 868 | return error; |
| 869 | } |
| 870 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 871 | get_page(page); |
| 872 | page->mapping = mapping; |
| 873 | page->index = offset; |
Olivier Deprez | 0e64123 | 2021-09-23 10:07:05 +0200 | [diff] [blame^] | 874 | gfp_mask &= GFP_RECLAIM_MASK; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 875 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 876 | do { |
Olivier Deprez | 0e64123 | 2021-09-23 10:07:05 +0200 | [diff] [blame^] | 877 | unsigned int order = xa_get_order(xas.xa, xas.xa_index); |
| 878 | void *entry, *old = NULL; |
| 879 | |
| 880 | if (order > thp_order(page)) |
| 881 | xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index), |
| 882 | order, gfp_mask); |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 883 | xas_lock_irq(&xas); |
Olivier Deprez | 0e64123 | 2021-09-23 10:07:05 +0200 | [diff] [blame^] | 884 | xas_for_each_conflict(&xas, entry) { |
| 885 | old = entry; |
| 886 | if (!xa_is_value(entry)) { |
| 887 | xas_set_err(&xas, -EEXIST); |
| 888 | goto unlock; |
| 889 | } |
| 890 | } |
| 891 | |
| 892 | if (old) { |
| 893 | if (shadowp) |
| 894 | *shadowp = old; |
| 895 | /* entry may have been split before we acquired lock */ |
| 896 | order = xa_get_order(xas.xa, xas.xa_index); |
| 897 | if (order > thp_order(page)) { |
| 898 | xas_split(&xas, old, order); |
| 899 | xas_reset(&xas); |
| 900 | } |
| 901 | } |
| 902 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 903 | xas_store(&xas, page); |
| 904 | if (xas_error(&xas)) |
| 905 | goto unlock; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 906 | |
Olivier Deprez | 0e64123 | 2021-09-23 10:07:05 +0200 | [diff] [blame^] | 907 | if (old) |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 908 | mapping->nrexceptional--; |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 909 | mapping->nrpages++; |
| 910 | |
| 911 | /* hugetlb pages do not participate in page cache accounting */ |
| 912 | if (!huge) |
| 913 | __inc_node_page_state(page, NR_FILE_PAGES); |
| 914 | unlock: |
| 915 | xas_unlock_irq(&xas); |
Olivier Deprez | 0e64123 | 2021-09-23 10:07:05 +0200 | [diff] [blame^] | 916 | } while (xas_nomem(&xas, gfp_mask)); |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 917 | |
| 918 | if (xas_error(&xas)) |
| 919 | goto error; |
| 920 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 921 | if (!huge) |
| 922 | mem_cgroup_commit_charge(page, memcg, false, false); |
| 923 | trace_mm_filemap_add_to_page_cache(page); |
| 924 | return 0; |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 925 | error: |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 926 | page->mapping = NULL; |
| 927 | /* Leave page->index set: truncation relies upon it */ |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 928 | if (!huge) |
| 929 | mem_cgroup_cancel_charge(page, memcg, false); |
| 930 | put_page(page); |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 931 | return xas_error(&xas); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 932 | } |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 933 | ALLOW_ERROR_INJECTION(__add_to_page_cache_locked, ERRNO); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 934 | |
| 935 | /** |
| 936 | * add_to_page_cache_locked - add a locked page to the pagecache |
| 937 | * @page: page to add |
| 938 | * @mapping: the page's address_space |
| 939 | * @offset: page index |
| 940 | * @gfp_mask: page allocation mode |
| 941 | * |
| 942 | * This function is used to add a page to the pagecache. It must be locked. |
| 943 | * This function does not add the page to the LRU. The caller must do that. |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 944 | * |
| 945 | * Return: %0 on success, negative error code otherwise. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 946 | */ |
| 947 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, |
| 948 | pgoff_t offset, gfp_t gfp_mask) |
| 949 | { |
| 950 | return __add_to_page_cache_locked(page, mapping, offset, |
| 951 | gfp_mask, NULL); |
| 952 | } |
| 953 | EXPORT_SYMBOL(add_to_page_cache_locked); |
| 954 | |
| 955 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, |
| 956 | pgoff_t offset, gfp_t gfp_mask) |
| 957 | { |
| 958 | void *shadow = NULL; |
| 959 | int ret; |
| 960 | |
| 961 | __SetPageLocked(page); |
| 962 | ret = __add_to_page_cache_locked(page, mapping, offset, |
| 963 | gfp_mask, &shadow); |
| 964 | if (unlikely(ret)) |
| 965 | __ClearPageLocked(page); |
| 966 | else { |
| 967 | /* |
| 968 | * The page might have been evicted from cache only |
| 969 | * recently, in which case it should be activated like |
| 970 | * any other repeatedly accessed page. |
| 971 | * The exception is pages getting rewritten; evicting other |
| 972 | * data from the working set, only to cache data that will |
| 973 | * get overwritten with something else, is a waste of memory. |
| 974 | */ |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 975 | WARN_ON_ONCE(PageActive(page)); |
| 976 | if (!(gfp_mask & __GFP_WRITE) && shadow) |
| 977 | workingset_refault(page, shadow); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 978 | lru_cache_add(page); |
| 979 | } |
| 980 | return ret; |
| 981 | } |
| 982 | EXPORT_SYMBOL_GPL(add_to_page_cache_lru); |
| 983 | |
| 984 | #ifdef CONFIG_NUMA |
| 985 | struct page *__page_cache_alloc(gfp_t gfp) |
| 986 | { |
| 987 | int n; |
| 988 | struct page *page; |
| 989 | |
| 990 | if (cpuset_do_page_mem_spread()) { |
| 991 | unsigned int cpuset_mems_cookie; |
| 992 | do { |
| 993 | cpuset_mems_cookie = read_mems_allowed_begin(); |
| 994 | n = cpuset_mem_spread_node(); |
| 995 | page = __alloc_pages_node(n, gfp, 0); |
| 996 | } while (!page && read_mems_allowed_retry(cpuset_mems_cookie)); |
| 997 | |
| 998 | return page; |
| 999 | } |
| 1000 | return alloc_pages(gfp, 0); |
| 1001 | } |
| 1002 | EXPORT_SYMBOL(__page_cache_alloc); |
| 1003 | #endif |
| 1004 | |
| 1005 | /* |
| 1006 | * In order to wait for pages to become available there must be |
| 1007 | * waitqueues associated with pages. By using a hash table of |
| 1008 | * waitqueues where the bucket discipline is to maintain all |
| 1009 | * waiters on the same queue and wake all when any of the pages |
| 1010 | * become available, and for the woken contexts to check to be |
| 1011 | * sure the appropriate page became available, this saves space |
| 1012 | * at a cost of "thundering herd" phenomena during rare hash |
| 1013 | * collisions. |
| 1014 | */ |
| 1015 | #define PAGE_WAIT_TABLE_BITS 8 |
| 1016 | #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS) |
| 1017 | static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned; |
| 1018 | |
| 1019 | static wait_queue_head_t *page_waitqueue(struct page *page) |
| 1020 | { |
| 1021 | return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)]; |
| 1022 | } |
| 1023 | |
| 1024 | void __init pagecache_init(void) |
| 1025 | { |
| 1026 | int i; |
| 1027 | |
| 1028 | for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++) |
| 1029 | init_waitqueue_head(&page_wait_table[i]); |
| 1030 | |
| 1031 | page_writeback_init(); |
| 1032 | } |
| 1033 | |
| 1034 | /* This has the same layout as wait_bit_key - see fs/cachefiles/rdwr.c */ |
| 1035 | struct wait_page_key { |
| 1036 | struct page *page; |
| 1037 | int bit_nr; |
| 1038 | int page_match; |
| 1039 | }; |
| 1040 | |
| 1041 | struct wait_page_queue { |
| 1042 | struct page *page; |
| 1043 | int bit_nr; |
| 1044 | wait_queue_entry_t wait; |
| 1045 | }; |
| 1046 | |
| 1047 | static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg) |
| 1048 | { |
| 1049 | struct wait_page_key *key = arg; |
| 1050 | struct wait_page_queue *wait_page |
| 1051 | = container_of(wait, struct wait_page_queue, wait); |
| 1052 | |
| 1053 | if (wait_page->page != key->page) |
| 1054 | return 0; |
| 1055 | key->page_match = 1; |
| 1056 | |
| 1057 | if (wait_page->bit_nr != key->bit_nr) |
| 1058 | return 0; |
| 1059 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1060 | /* |
| 1061 | * Stop walking if it's locked. |
| 1062 | * Is this safe if put_and_wait_on_page_locked() is in use? |
| 1063 | * Yes: the waker must hold a reference to this page, and if PG_locked |
| 1064 | * has now already been set by another task, that task must also hold |
| 1065 | * a reference to the *same usage* of this page; so there is no need |
| 1066 | * to walk on to wake even the put_and_wait_on_page_locked() callers. |
| 1067 | */ |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1068 | if (test_bit(key->bit_nr, &key->page->flags)) |
| 1069 | return -1; |
| 1070 | |
| 1071 | return autoremove_wake_function(wait, mode, sync, key); |
| 1072 | } |
| 1073 | |
| 1074 | static void wake_up_page_bit(struct page *page, int bit_nr) |
| 1075 | { |
| 1076 | wait_queue_head_t *q = page_waitqueue(page); |
| 1077 | struct wait_page_key key; |
| 1078 | unsigned long flags; |
| 1079 | wait_queue_entry_t bookmark; |
| 1080 | |
| 1081 | key.page = page; |
| 1082 | key.bit_nr = bit_nr; |
| 1083 | key.page_match = 0; |
| 1084 | |
| 1085 | bookmark.flags = 0; |
| 1086 | bookmark.private = NULL; |
| 1087 | bookmark.func = NULL; |
| 1088 | INIT_LIST_HEAD(&bookmark.entry); |
| 1089 | |
| 1090 | spin_lock_irqsave(&q->lock, flags); |
| 1091 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); |
| 1092 | |
| 1093 | while (bookmark.flags & WQ_FLAG_BOOKMARK) { |
| 1094 | /* |
| 1095 | * Take a breather from holding the lock, |
| 1096 | * allow pages that finish wake up asynchronously |
| 1097 | * to acquire the lock and remove themselves |
| 1098 | * from wait queue |
| 1099 | */ |
| 1100 | spin_unlock_irqrestore(&q->lock, flags); |
| 1101 | cpu_relax(); |
| 1102 | spin_lock_irqsave(&q->lock, flags); |
| 1103 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); |
| 1104 | } |
| 1105 | |
| 1106 | /* |
| 1107 | * It is possible for other pages to have collided on the waitqueue |
| 1108 | * hash, so in that case check for a page match. That prevents a long- |
| 1109 | * term waiter |
| 1110 | * |
| 1111 | * It is still possible to miss a case here, when we woke page waiters |
| 1112 | * and removed them from the waitqueue, but there are still other |
| 1113 | * page waiters. |
| 1114 | */ |
| 1115 | if (!waitqueue_active(q) || !key.page_match) { |
| 1116 | ClearPageWaiters(page); |
| 1117 | /* |
| 1118 | * It's possible to miss clearing Waiters here, when we woke |
| 1119 | * our page waiters, but the hashed waitqueue has waiters for |
| 1120 | * other pages on it. |
| 1121 | * |
| 1122 | * That's okay, it's a rare case. The next waker will clear it. |
| 1123 | */ |
| 1124 | } |
| 1125 | spin_unlock_irqrestore(&q->lock, flags); |
| 1126 | } |
| 1127 | |
| 1128 | static void wake_up_page(struct page *page, int bit) |
| 1129 | { |
| 1130 | if (!PageWaiters(page)) |
| 1131 | return; |
| 1132 | wake_up_page_bit(page, bit); |
| 1133 | } |
| 1134 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1135 | /* |
| 1136 | * A choice of three behaviors for wait_on_page_bit_common(): |
| 1137 | */ |
| 1138 | enum behavior { |
| 1139 | EXCLUSIVE, /* Hold ref to page and take the bit when woken, like |
| 1140 | * __lock_page() waiting on then setting PG_locked. |
| 1141 | */ |
| 1142 | SHARED, /* Hold ref to page and check the bit when woken, like |
| 1143 | * wait_on_page_writeback() waiting on PG_writeback. |
| 1144 | */ |
| 1145 | DROP, /* Drop ref to page before wait, no check when woken, |
| 1146 | * like put_and_wait_on_page_locked() on PG_locked. |
| 1147 | */ |
| 1148 | }; |
| 1149 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1150 | static inline int wait_on_page_bit_common(wait_queue_head_t *q, |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1151 | struct page *page, int bit_nr, int state, enum behavior behavior) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1152 | { |
| 1153 | struct wait_page_queue wait_page; |
| 1154 | wait_queue_entry_t *wait = &wait_page.wait; |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1155 | bool bit_is_set; |
| 1156 | bool thrashing = false; |
| 1157 | bool delayacct = false; |
| 1158 | unsigned long pflags; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1159 | int ret = 0; |
| 1160 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1161 | if (bit_nr == PG_locked && |
| 1162 | !PageUptodate(page) && PageWorkingset(page)) { |
| 1163 | if (!PageSwapBacked(page)) { |
| 1164 | delayacct_thrashing_start(); |
| 1165 | delayacct = true; |
| 1166 | } |
| 1167 | psi_memstall_enter(&pflags); |
| 1168 | thrashing = true; |
| 1169 | } |
| 1170 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1171 | init_wait(wait); |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1172 | wait->flags = behavior == EXCLUSIVE ? WQ_FLAG_EXCLUSIVE : 0; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1173 | wait->func = wake_page_function; |
| 1174 | wait_page.page = page; |
| 1175 | wait_page.bit_nr = bit_nr; |
| 1176 | |
| 1177 | for (;;) { |
| 1178 | spin_lock_irq(&q->lock); |
| 1179 | |
| 1180 | if (likely(list_empty(&wait->entry))) { |
| 1181 | __add_wait_queue_entry_tail(q, wait); |
| 1182 | SetPageWaiters(page); |
| 1183 | } |
| 1184 | |
| 1185 | set_current_state(state); |
| 1186 | |
| 1187 | spin_unlock_irq(&q->lock); |
| 1188 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1189 | bit_is_set = test_bit(bit_nr, &page->flags); |
| 1190 | if (behavior == DROP) |
| 1191 | put_page(page); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1192 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1193 | if (likely(bit_is_set)) |
| 1194 | io_schedule(); |
| 1195 | |
| 1196 | if (behavior == EXCLUSIVE) { |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1197 | if (!test_and_set_bit_lock(bit_nr, &page->flags)) |
| 1198 | break; |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1199 | } else if (behavior == SHARED) { |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1200 | if (!test_bit(bit_nr, &page->flags)) |
| 1201 | break; |
| 1202 | } |
| 1203 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1204 | if (signal_pending_state(state, current)) { |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1205 | ret = -EINTR; |
| 1206 | break; |
| 1207 | } |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1208 | |
| 1209 | if (behavior == DROP) { |
| 1210 | /* |
| 1211 | * We can no longer safely access page->flags: |
| 1212 | * even if CONFIG_MEMORY_HOTREMOVE is not enabled, |
| 1213 | * there is a risk of waiting forever on a page reused |
| 1214 | * for something that keeps it locked indefinitely. |
| 1215 | * But best check for -EINTR above before breaking. |
| 1216 | */ |
| 1217 | break; |
| 1218 | } |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1219 | } |
| 1220 | |
| 1221 | finish_wait(q, wait); |
| 1222 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1223 | if (thrashing) { |
| 1224 | if (delayacct) |
| 1225 | delayacct_thrashing_end(); |
| 1226 | psi_memstall_leave(&pflags); |
| 1227 | } |
| 1228 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1229 | /* |
| 1230 | * A signal could leave PageWaiters set. Clearing it here if |
| 1231 | * !waitqueue_active would be possible (by open-coding finish_wait), |
| 1232 | * but still fail to catch it in the case of wait hash collision. We |
| 1233 | * already can fail to clear wait hash collision cases, so don't |
| 1234 | * bother with signals either. |
| 1235 | */ |
| 1236 | |
| 1237 | return ret; |
| 1238 | } |
| 1239 | |
| 1240 | void wait_on_page_bit(struct page *page, int bit_nr) |
| 1241 | { |
| 1242 | wait_queue_head_t *q = page_waitqueue(page); |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1243 | wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, SHARED); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1244 | } |
| 1245 | EXPORT_SYMBOL(wait_on_page_bit); |
| 1246 | |
| 1247 | int wait_on_page_bit_killable(struct page *page, int bit_nr) |
| 1248 | { |
| 1249 | wait_queue_head_t *q = page_waitqueue(page); |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1250 | return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, SHARED); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1251 | } |
| 1252 | EXPORT_SYMBOL(wait_on_page_bit_killable); |
| 1253 | |
| 1254 | /** |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1255 | * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked |
| 1256 | * @page: The page to wait for. |
| 1257 | * |
| 1258 | * The caller should hold a reference on @page. They expect the page to |
| 1259 | * become unlocked relatively soon, but do not wish to hold up migration |
| 1260 | * (for example) by holding the reference while waiting for the page to |
| 1261 | * come unlocked. After this function returns, the caller should not |
| 1262 | * dereference @page. |
| 1263 | */ |
| 1264 | void put_and_wait_on_page_locked(struct page *page) |
| 1265 | { |
| 1266 | wait_queue_head_t *q; |
| 1267 | |
| 1268 | page = compound_head(page); |
| 1269 | q = page_waitqueue(page); |
| 1270 | wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, DROP); |
| 1271 | } |
| 1272 | |
| 1273 | /** |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1274 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue |
| 1275 | * @page: Page defining the wait queue of interest |
| 1276 | * @waiter: Waiter to add to the queue |
| 1277 | * |
| 1278 | * Add an arbitrary @waiter to the wait queue for the nominated @page. |
| 1279 | */ |
| 1280 | void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter) |
| 1281 | { |
| 1282 | wait_queue_head_t *q = page_waitqueue(page); |
| 1283 | unsigned long flags; |
| 1284 | |
| 1285 | spin_lock_irqsave(&q->lock, flags); |
| 1286 | __add_wait_queue_entry_tail(q, waiter); |
| 1287 | SetPageWaiters(page); |
| 1288 | spin_unlock_irqrestore(&q->lock, flags); |
| 1289 | } |
| 1290 | EXPORT_SYMBOL_GPL(add_page_wait_queue); |
| 1291 | |
| 1292 | #ifndef clear_bit_unlock_is_negative_byte |
| 1293 | |
| 1294 | /* |
| 1295 | * PG_waiters is the high bit in the same byte as PG_lock. |
| 1296 | * |
| 1297 | * On x86 (and on many other architectures), we can clear PG_lock and |
| 1298 | * test the sign bit at the same time. But if the architecture does |
| 1299 | * not support that special operation, we just do this all by hand |
| 1300 | * instead. |
| 1301 | * |
| 1302 | * The read of PG_waiters has to be after (or concurrently with) PG_locked |
| 1303 | * being cleared, but a memory barrier should be unneccssary since it is |
| 1304 | * in the same byte as PG_locked. |
| 1305 | */ |
| 1306 | static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem) |
| 1307 | { |
| 1308 | clear_bit_unlock(nr, mem); |
| 1309 | /* smp_mb__after_atomic(); */ |
| 1310 | return test_bit(PG_waiters, mem); |
| 1311 | } |
| 1312 | |
| 1313 | #endif |
| 1314 | |
| 1315 | /** |
| 1316 | * unlock_page - unlock a locked page |
| 1317 | * @page: the page |
| 1318 | * |
| 1319 | * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). |
| 1320 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup |
| 1321 | * mechanism between PageLocked pages and PageWriteback pages is shared. |
| 1322 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. |
| 1323 | * |
| 1324 | * Note that this depends on PG_waiters being the sign bit in the byte |
| 1325 | * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to |
| 1326 | * clear the PG_locked bit and test PG_waiters at the same time fairly |
| 1327 | * portably (architectures that do LL/SC can test any bit, while x86 can |
| 1328 | * test the sign bit). |
| 1329 | */ |
| 1330 | void unlock_page(struct page *page) |
| 1331 | { |
| 1332 | BUILD_BUG_ON(PG_waiters != 7); |
| 1333 | page = compound_head(page); |
| 1334 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
| 1335 | if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags)) |
| 1336 | wake_up_page_bit(page, PG_locked); |
| 1337 | } |
| 1338 | EXPORT_SYMBOL(unlock_page); |
| 1339 | |
| 1340 | /** |
| 1341 | * end_page_writeback - end writeback against a page |
| 1342 | * @page: the page |
| 1343 | */ |
| 1344 | void end_page_writeback(struct page *page) |
| 1345 | { |
| 1346 | /* |
| 1347 | * TestClearPageReclaim could be used here but it is an atomic |
| 1348 | * operation and overkill in this particular case. Failing to |
| 1349 | * shuffle a page marked for immediate reclaim is too mild to |
| 1350 | * justify taking an atomic operation penalty at the end of |
| 1351 | * ever page writeback. |
| 1352 | */ |
| 1353 | if (PageReclaim(page)) { |
| 1354 | ClearPageReclaim(page); |
| 1355 | rotate_reclaimable_page(page); |
| 1356 | } |
| 1357 | |
| 1358 | if (!test_clear_page_writeback(page)) |
| 1359 | BUG(); |
| 1360 | |
| 1361 | smp_mb__after_atomic(); |
| 1362 | wake_up_page(page, PG_writeback); |
| 1363 | } |
| 1364 | EXPORT_SYMBOL(end_page_writeback); |
| 1365 | |
| 1366 | /* |
| 1367 | * After completing I/O on a page, call this routine to update the page |
| 1368 | * flags appropriately |
| 1369 | */ |
| 1370 | void page_endio(struct page *page, bool is_write, int err) |
| 1371 | { |
| 1372 | if (!is_write) { |
| 1373 | if (!err) { |
| 1374 | SetPageUptodate(page); |
| 1375 | } else { |
| 1376 | ClearPageUptodate(page); |
| 1377 | SetPageError(page); |
| 1378 | } |
| 1379 | unlock_page(page); |
| 1380 | } else { |
| 1381 | if (err) { |
| 1382 | struct address_space *mapping; |
| 1383 | |
| 1384 | SetPageError(page); |
| 1385 | mapping = page_mapping(page); |
| 1386 | if (mapping) |
| 1387 | mapping_set_error(mapping, err); |
| 1388 | } |
| 1389 | end_page_writeback(page); |
| 1390 | } |
| 1391 | } |
| 1392 | EXPORT_SYMBOL_GPL(page_endio); |
| 1393 | |
| 1394 | /** |
| 1395 | * __lock_page - get a lock on the page, assuming we need to sleep to get it |
| 1396 | * @__page: the page to lock |
| 1397 | */ |
| 1398 | void __lock_page(struct page *__page) |
| 1399 | { |
| 1400 | struct page *page = compound_head(__page); |
| 1401 | wait_queue_head_t *q = page_waitqueue(page); |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1402 | wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, |
| 1403 | EXCLUSIVE); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1404 | } |
| 1405 | EXPORT_SYMBOL(__lock_page); |
| 1406 | |
| 1407 | int __lock_page_killable(struct page *__page) |
| 1408 | { |
| 1409 | struct page *page = compound_head(__page); |
| 1410 | wait_queue_head_t *q = page_waitqueue(page); |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1411 | return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE, |
| 1412 | EXCLUSIVE); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1413 | } |
| 1414 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
| 1415 | |
| 1416 | /* |
| 1417 | * Return values: |
| 1418 | * 1 - page is locked; mmap_sem is still held. |
| 1419 | * 0 - page is not locked. |
| 1420 | * mmap_sem has been released (up_read()), unless flags had both |
| 1421 | * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in |
| 1422 | * which case mmap_sem is still held. |
| 1423 | * |
| 1424 | * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1 |
| 1425 | * with the page locked and the mmap_sem unperturbed. |
| 1426 | */ |
| 1427 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
| 1428 | unsigned int flags) |
| 1429 | { |
| 1430 | if (flags & FAULT_FLAG_ALLOW_RETRY) { |
| 1431 | /* |
| 1432 | * CAUTION! In this case, mmap_sem is not released |
| 1433 | * even though return 0. |
| 1434 | */ |
| 1435 | if (flags & FAULT_FLAG_RETRY_NOWAIT) |
| 1436 | return 0; |
| 1437 | |
| 1438 | up_read(&mm->mmap_sem); |
| 1439 | if (flags & FAULT_FLAG_KILLABLE) |
| 1440 | wait_on_page_locked_killable(page); |
| 1441 | else |
| 1442 | wait_on_page_locked(page); |
| 1443 | return 0; |
| 1444 | } else { |
| 1445 | if (flags & FAULT_FLAG_KILLABLE) { |
| 1446 | int ret; |
| 1447 | |
| 1448 | ret = __lock_page_killable(page); |
| 1449 | if (ret) { |
| 1450 | up_read(&mm->mmap_sem); |
| 1451 | return 0; |
| 1452 | } |
| 1453 | } else |
| 1454 | __lock_page(page); |
| 1455 | return 1; |
| 1456 | } |
| 1457 | } |
| 1458 | |
| 1459 | /** |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1460 | * page_cache_next_miss() - Find the next gap in the page cache. |
| 1461 | * @mapping: Mapping. |
| 1462 | * @index: Index. |
| 1463 | * @max_scan: Maximum range to search. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1464 | * |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1465 | * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the |
| 1466 | * gap with the lowest index. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1467 | * |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1468 | * This function may be called under the rcu_read_lock. However, this will |
| 1469 | * not atomically search a snapshot of the cache at a single point in time. |
| 1470 | * For example, if a gap is created at index 5, then subsequently a gap is |
| 1471 | * created at index 10, page_cache_next_miss covering both indices may |
| 1472 | * return 10 if called under the rcu_read_lock. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1473 | * |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1474 | * Return: The index of the gap if found, otherwise an index outside the |
| 1475 | * range specified (in which case 'return - index >= max_scan' will be true). |
| 1476 | * In the rare case of index wrap-around, 0 will be returned. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1477 | */ |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1478 | pgoff_t page_cache_next_miss(struct address_space *mapping, |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1479 | pgoff_t index, unsigned long max_scan) |
| 1480 | { |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1481 | XA_STATE(xas, &mapping->i_pages, index); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1482 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1483 | while (max_scan--) { |
| 1484 | void *entry = xas_next(&xas); |
| 1485 | if (!entry || xa_is_value(entry)) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1486 | break; |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1487 | if (xas.xa_index == 0) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1488 | break; |
| 1489 | } |
| 1490 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1491 | return xas.xa_index; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1492 | } |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1493 | EXPORT_SYMBOL(page_cache_next_miss); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1494 | |
| 1495 | /** |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1496 | * page_cache_prev_miss() - Find the previous gap in the page cache. |
| 1497 | * @mapping: Mapping. |
| 1498 | * @index: Index. |
| 1499 | * @max_scan: Maximum range to search. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1500 | * |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1501 | * Search the range [max(index - max_scan + 1, 0), index] for the |
| 1502 | * gap with the highest index. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1503 | * |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1504 | * This function may be called under the rcu_read_lock. However, this will |
| 1505 | * not atomically search a snapshot of the cache at a single point in time. |
| 1506 | * For example, if a gap is created at index 10, then subsequently a gap is |
| 1507 | * created at index 5, page_cache_prev_miss() covering both indices may |
| 1508 | * return 5 if called under the rcu_read_lock. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1509 | * |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1510 | * Return: The index of the gap if found, otherwise an index outside the |
| 1511 | * range specified (in which case 'index - return >= max_scan' will be true). |
| 1512 | * In the rare case of wrap-around, ULONG_MAX will be returned. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1513 | */ |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1514 | pgoff_t page_cache_prev_miss(struct address_space *mapping, |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1515 | pgoff_t index, unsigned long max_scan) |
| 1516 | { |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1517 | XA_STATE(xas, &mapping->i_pages, index); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1518 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1519 | while (max_scan--) { |
| 1520 | void *entry = xas_prev(&xas); |
| 1521 | if (!entry || xa_is_value(entry)) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1522 | break; |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1523 | if (xas.xa_index == ULONG_MAX) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1524 | break; |
| 1525 | } |
| 1526 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1527 | return xas.xa_index; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1528 | } |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1529 | EXPORT_SYMBOL(page_cache_prev_miss); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1530 | |
| 1531 | /** |
| 1532 | * find_get_entry - find and get a page cache entry |
| 1533 | * @mapping: the address_space to search |
| 1534 | * @offset: the page cache index |
| 1535 | * |
| 1536 | * Looks up the page cache slot at @mapping & @offset. If there is a |
| 1537 | * page cache page, it is returned with an increased refcount. |
| 1538 | * |
| 1539 | * If the slot holds a shadow entry of a previously evicted page, or a |
| 1540 | * swap entry from shmem/tmpfs, it is returned. |
| 1541 | * |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1542 | * Return: the found page or shadow entry, %NULL if nothing is found. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1543 | */ |
| 1544 | struct page *find_get_entry(struct address_space *mapping, pgoff_t offset) |
| 1545 | { |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1546 | XA_STATE(xas, &mapping->i_pages, offset); |
| 1547 | struct page *page; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1548 | |
| 1549 | rcu_read_lock(); |
| 1550 | repeat: |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1551 | xas_reset(&xas); |
| 1552 | page = xas_load(&xas); |
| 1553 | if (xas_retry(&xas, page)) |
| 1554 | goto repeat; |
| 1555 | /* |
| 1556 | * A shadow entry of a recently evicted page, or a swap entry from |
| 1557 | * shmem/tmpfs. Return it without attempting to raise page count. |
| 1558 | */ |
| 1559 | if (!page || xa_is_value(page)) |
| 1560 | goto out; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1561 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1562 | if (!page_cache_get_speculative(page)) |
| 1563 | goto repeat; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1564 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1565 | /* |
| 1566 | * Has the page moved or been split? |
| 1567 | * This is part of the lockless pagecache protocol. See |
| 1568 | * include/linux/pagemap.h for details. |
| 1569 | */ |
| 1570 | if (unlikely(page != xas_reload(&xas))) { |
| 1571 | put_page(page); |
| 1572 | goto repeat; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1573 | } |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1574 | page = find_subpage(page, offset); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1575 | out: |
| 1576 | rcu_read_unlock(); |
| 1577 | |
| 1578 | return page; |
| 1579 | } |
| 1580 | EXPORT_SYMBOL(find_get_entry); |
| 1581 | |
| 1582 | /** |
| 1583 | * find_lock_entry - locate, pin and lock a page cache entry |
| 1584 | * @mapping: the address_space to search |
| 1585 | * @offset: the page cache index |
| 1586 | * |
| 1587 | * Looks up the page cache slot at @mapping & @offset. If there is a |
| 1588 | * page cache page, it is returned locked and with an increased |
| 1589 | * refcount. |
| 1590 | * |
| 1591 | * If the slot holds a shadow entry of a previously evicted page, or a |
| 1592 | * swap entry from shmem/tmpfs, it is returned. |
| 1593 | * |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1594 | * find_lock_entry() may sleep. |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1595 | * |
| 1596 | * Return: the found page or shadow entry, %NULL if nothing is found. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1597 | */ |
| 1598 | struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset) |
| 1599 | { |
| 1600 | struct page *page; |
| 1601 | |
| 1602 | repeat: |
| 1603 | page = find_get_entry(mapping, offset); |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1604 | if (page && !xa_is_value(page)) { |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1605 | lock_page(page); |
| 1606 | /* Has the page been truncated? */ |
| 1607 | if (unlikely(page_mapping(page) != mapping)) { |
| 1608 | unlock_page(page); |
| 1609 | put_page(page); |
| 1610 | goto repeat; |
| 1611 | } |
| 1612 | VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page); |
| 1613 | } |
| 1614 | return page; |
| 1615 | } |
| 1616 | EXPORT_SYMBOL(find_lock_entry); |
| 1617 | |
| 1618 | /** |
| 1619 | * pagecache_get_page - find and get a page reference |
| 1620 | * @mapping: the address_space to search |
| 1621 | * @offset: the page index |
| 1622 | * @fgp_flags: PCG flags |
| 1623 | * @gfp_mask: gfp mask to use for the page cache data page allocation |
| 1624 | * |
| 1625 | * Looks up the page cache slot at @mapping & @offset. |
| 1626 | * |
| 1627 | * PCG flags modify how the page is returned. |
| 1628 | * |
| 1629 | * @fgp_flags can be: |
| 1630 | * |
| 1631 | * - FGP_ACCESSED: the page will be marked accessed |
| 1632 | * - FGP_LOCK: Page is return locked |
| 1633 | * - FGP_CREAT: If page is not present then a new page is allocated using |
| 1634 | * @gfp_mask and added to the page cache and the VM's LRU |
| 1635 | * list. The page is returned locked and with an increased |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1636 | * refcount. |
| 1637 | * - FGP_FOR_MMAP: Similar to FGP_CREAT, only we want to allow the caller to do |
| 1638 | * its own locking dance if the page is already in cache, or unlock the page |
| 1639 | * before returning if we had to add the page to pagecache. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1640 | * |
| 1641 | * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even |
| 1642 | * if the GFP flags specified for FGP_CREAT are atomic. |
| 1643 | * |
| 1644 | * If there is a page cache page, it is returned with an increased refcount. |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1645 | * |
| 1646 | * Return: the found page or %NULL otherwise. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1647 | */ |
| 1648 | struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset, |
| 1649 | int fgp_flags, gfp_t gfp_mask) |
| 1650 | { |
| 1651 | struct page *page; |
| 1652 | |
| 1653 | repeat: |
| 1654 | page = find_get_entry(mapping, offset); |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1655 | if (xa_is_value(page)) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1656 | page = NULL; |
| 1657 | if (!page) |
| 1658 | goto no_page; |
| 1659 | |
| 1660 | if (fgp_flags & FGP_LOCK) { |
| 1661 | if (fgp_flags & FGP_NOWAIT) { |
| 1662 | if (!trylock_page(page)) { |
| 1663 | put_page(page); |
| 1664 | return NULL; |
| 1665 | } |
| 1666 | } else { |
| 1667 | lock_page(page); |
| 1668 | } |
| 1669 | |
| 1670 | /* Has the page been truncated? */ |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1671 | if (unlikely(compound_head(page)->mapping != mapping)) { |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1672 | unlock_page(page); |
| 1673 | put_page(page); |
| 1674 | goto repeat; |
| 1675 | } |
| 1676 | VM_BUG_ON_PAGE(page->index != offset, page); |
| 1677 | } |
| 1678 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1679 | if (fgp_flags & FGP_ACCESSED) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1680 | mark_page_accessed(page); |
| 1681 | |
| 1682 | no_page: |
| 1683 | if (!page && (fgp_flags & FGP_CREAT)) { |
| 1684 | int err; |
| 1685 | if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping)) |
| 1686 | gfp_mask |= __GFP_WRITE; |
| 1687 | if (fgp_flags & FGP_NOFS) |
| 1688 | gfp_mask &= ~__GFP_FS; |
| 1689 | |
| 1690 | page = __page_cache_alloc(gfp_mask); |
| 1691 | if (!page) |
| 1692 | return NULL; |
| 1693 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1694 | if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP)))) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1695 | fgp_flags |= FGP_LOCK; |
| 1696 | |
| 1697 | /* Init accessed so avoid atomic mark_page_accessed later */ |
| 1698 | if (fgp_flags & FGP_ACCESSED) |
| 1699 | __SetPageReferenced(page); |
| 1700 | |
| 1701 | err = add_to_page_cache_lru(page, mapping, offset, gfp_mask); |
| 1702 | if (unlikely(err)) { |
| 1703 | put_page(page); |
| 1704 | page = NULL; |
| 1705 | if (err == -EEXIST) |
| 1706 | goto repeat; |
| 1707 | } |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1708 | |
| 1709 | /* |
| 1710 | * add_to_page_cache_lru locks the page, and for mmap we expect |
| 1711 | * an unlocked page. |
| 1712 | */ |
| 1713 | if (page && (fgp_flags & FGP_FOR_MMAP)) |
| 1714 | unlock_page(page); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1715 | } |
| 1716 | |
| 1717 | return page; |
| 1718 | } |
| 1719 | EXPORT_SYMBOL(pagecache_get_page); |
| 1720 | |
| 1721 | /** |
| 1722 | * find_get_entries - gang pagecache lookup |
| 1723 | * @mapping: The address_space to search |
| 1724 | * @start: The starting page cache index |
| 1725 | * @nr_entries: The maximum number of entries |
| 1726 | * @entries: Where the resulting entries are placed |
| 1727 | * @indices: The cache indices corresponding to the entries in @entries |
| 1728 | * |
| 1729 | * find_get_entries() will search for and return a group of up to |
| 1730 | * @nr_entries entries in the mapping. The entries are placed at |
| 1731 | * @entries. find_get_entries() takes a reference against any actual |
| 1732 | * pages it returns. |
| 1733 | * |
| 1734 | * The search returns a group of mapping-contiguous page cache entries |
| 1735 | * with ascending indexes. There may be holes in the indices due to |
| 1736 | * not-present pages. |
| 1737 | * |
| 1738 | * Any shadow entries of evicted pages, or swap entries from |
| 1739 | * shmem/tmpfs, are included in the returned array. |
| 1740 | * |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1741 | * Return: the number of pages and shadow entries which were found. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1742 | */ |
| 1743 | unsigned find_get_entries(struct address_space *mapping, |
| 1744 | pgoff_t start, unsigned int nr_entries, |
| 1745 | struct page **entries, pgoff_t *indices) |
| 1746 | { |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1747 | XA_STATE(xas, &mapping->i_pages, start); |
| 1748 | struct page *page; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1749 | unsigned int ret = 0; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1750 | |
| 1751 | if (!nr_entries) |
| 1752 | return 0; |
| 1753 | |
| 1754 | rcu_read_lock(); |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1755 | xas_for_each(&xas, page, ULONG_MAX) { |
| 1756 | if (xas_retry(&xas, page)) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1757 | continue; |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1758 | /* |
| 1759 | * A shadow entry of a recently evicted page, a swap |
| 1760 | * entry from shmem/tmpfs or a DAX entry. Return it |
| 1761 | * without attempting to raise page count. |
| 1762 | */ |
| 1763 | if (xa_is_value(page)) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1764 | goto export; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1765 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1766 | if (!page_cache_get_speculative(page)) |
| 1767 | goto retry; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1768 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1769 | /* Has the page moved or been split? */ |
| 1770 | if (unlikely(page != xas_reload(&xas))) |
| 1771 | goto put_page; |
| 1772 | page = find_subpage(page, xas.xa_index); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1773 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1774 | export: |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1775 | indices[ret] = xas.xa_index; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1776 | entries[ret] = page; |
| 1777 | if (++ret == nr_entries) |
| 1778 | break; |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1779 | continue; |
| 1780 | put_page: |
| 1781 | put_page(page); |
| 1782 | retry: |
| 1783 | xas_reset(&xas); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1784 | } |
| 1785 | rcu_read_unlock(); |
| 1786 | return ret; |
| 1787 | } |
| 1788 | |
| 1789 | /** |
| 1790 | * find_get_pages_range - gang pagecache lookup |
| 1791 | * @mapping: The address_space to search |
| 1792 | * @start: The starting page index |
| 1793 | * @end: The final page index (inclusive) |
| 1794 | * @nr_pages: The maximum number of pages |
| 1795 | * @pages: Where the resulting pages are placed |
| 1796 | * |
| 1797 | * find_get_pages_range() will search for and return a group of up to @nr_pages |
| 1798 | * pages in the mapping starting at index @start and up to index @end |
| 1799 | * (inclusive). The pages are placed at @pages. find_get_pages_range() takes |
| 1800 | * a reference against the returned pages. |
| 1801 | * |
| 1802 | * The search returns a group of mapping-contiguous pages with ascending |
| 1803 | * indexes. There may be holes in the indices due to not-present pages. |
| 1804 | * We also update @start to index the next page for the traversal. |
| 1805 | * |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1806 | * Return: the number of pages which were found. If this number is |
| 1807 | * smaller than @nr_pages, the end of specified range has been |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1808 | * reached. |
| 1809 | */ |
| 1810 | unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start, |
| 1811 | pgoff_t end, unsigned int nr_pages, |
| 1812 | struct page **pages) |
| 1813 | { |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1814 | XA_STATE(xas, &mapping->i_pages, *start); |
| 1815 | struct page *page; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1816 | unsigned ret = 0; |
| 1817 | |
| 1818 | if (unlikely(!nr_pages)) |
| 1819 | return 0; |
| 1820 | |
| 1821 | rcu_read_lock(); |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1822 | xas_for_each(&xas, page, end) { |
| 1823 | if (xas_retry(&xas, page)) |
| 1824 | continue; |
| 1825 | /* Skip over shadow, swap and DAX entries */ |
| 1826 | if (xa_is_value(page)) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1827 | continue; |
| 1828 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1829 | if (!page_cache_get_speculative(page)) |
| 1830 | goto retry; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1831 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1832 | /* Has the page moved or been split? */ |
| 1833 | if (unlikely(page != xas_reload(&xas))) |
| 1834 | goto put_page; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1835 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1836 | pages[ret] = find_subpage(page, xas.xa_index); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1837 | if (++ret == nr_pages) { |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1838 | *start = xas.xa_index + 1; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1839 | goto out; |
| 1840 | } |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1841 | continue; |
| 1842 | put_page: |
| 1843 | put_page(page); |
| 1844 | retry: |
| 1845 | xas_reset(&xas); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1846 | } |
| 1847 | |
| 1848 | /* |
| 1849 | * We come here when there is no page beyond @end. We take care to not |
| 1850 | * overflow the index @start as it confuses some of the callers. This |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1851 | * breaks the iteration when there is a page at index -1 but that is |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1852 | * already broken anyway. |
| 1853 | */ |
| 1854 | if (end == (pgoff_t)-1) |
| 1855 | *start = (pgoff_t)-1; |
| 1856 | else |
| 1857 | *start = end + 1; |
| 1858 | out: |
| 1859 | rcu_read_unlock(); |
| 1860 | |
| 1861 | return ret; |
| 1862 | } |
| 1863 | |
| 1864 | /** |
| 1865 | * find_get_pages_contig - gang contiguous pagecache lookup |
| 1866 | * @mapping: The address_space to search |
| 1867 | * @index: The starting page index |
| 1868 | * @nr_pages: The maximum number of pages |
| 1869 | * @pages: Where the resulting pages are placed |
| 1870 | * |
| 1871 | * find_get_pages_contig() works exactly like find_get_pages(), except |
| 1872 | * that the returned number of pages are guaranteed to be contiguous. |
| 1873 | * |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1874 | * Return: the number of pages which were found. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1875 | */ |
| 1876 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, |
| 1877 | unsigned int nr_pages, struct page **pages) |
| 1878 | { |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1879 | XA_STATE(xas, &mapping->i_pages, index); |
| 1880 | struct page *page; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1881 | unsigned int ret = 0; |
| 1882 | |
| 1883 | if (unlikely(!nr_pages)) |
| 1884 | return 0; |
| 1885 | |
| 1886 | rcu_read_lock(); |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1887 | for (page = xas_load(&xas); page; page = xas_next(&xas)) { |
| 1888 | if (xas_retry(&xas, page)) |
| 1889 | continue; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1890 | /* |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1891 | * If the entry has been swapped out, we can stop looking. |
| 1892 | * No current caller is looking for DAX entries. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1893 | */ |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1894 | if (xa_is_value(page)) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1895 | break; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1896 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1897 | if (!page_cache_get_speculative(page)) |
| 1898 | goto retry; |
| 1899 | |
| 1900 | /* Has the page moved or been split? */ |
| 1901 | if (unlikely(page != xas_reload(&xas))) |
| 1902 | goto put_page; |
| 1903 | |
| 1904 | pages[ret] = find_subpage(page, xas.xa_index); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1905 | if (++ret == nr_pages) |
| 1906 | break; |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1907 | continue; |
| 1908 | put_page: |
| 1909 | put_page(page); |
| 1910 | retry: |
| 1911 | xas_reset(&xas); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1912 | } |
| 1913 | rcu_read_unlock(); |
| 1914 | return ret; |
| 1915 | } |
| 1916 | EXPORT_SYMBOL(find_get_pages_contig); |
| 1917 | |
| 1918 | /** |
| 1919 | * find_get_pages_range_tag - find and return pages in given range matching @tag |
| 1920 | * @mapping: the address_space to search |
| 1921 | * @index: the starting page index |
| 1922 | * @end: The final page index (inclusive) |
| 1923 | * @tag: the tag index |
| 1924 | * @nr_pages: the maximum number of pages |
| 1925 | * @pages: where the resulting pages are placed |
| 1926 | * |
| 1927 | * Like find_get_pages, except we only return pages which are tagged with |
| 1928 | * @tag. We update @index to index the next page for the traversal. |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1929 | * |
| 1930 | * Return: the number of pages which were found. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1931 | */ |
| 1932 | unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index, |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1933 | pgoff_t end, xa_mark_t tag, unsigned int nr_pages, |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1934 | struct page **pages) |
| 1935 | { |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1936 | XA_STATE(xas, &mapping->i_pages, *index); |
| 1937 | struct page *page; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1938 | unsigned ret = 0; |
| 1939 | |
| 1940 | if (unlikely(!nr_pages)) |
| 1941 | return 0; |
| 1942 | |
| 1943 | rcu_read_lock(); |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1944 | xas_for_each_marked(&xas, page, end, tag) { |
| 1945 | if (xas_retry(&xas, page)) |
| 1946 | continue; |
| 1947 | /* |
| 1948 | * Shadow entries should never be tagged, but this iteration |
| 1949 | * is lockless so there is a window for page reclaim to evict |
| 1950 | * a page we saw tagged. Skip over it. |
| 1951 | */ |
| 1952 | if (xa_is_value(page)) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1953 | continue; |
| 1954 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1955 | if (!page_cache_get_speculative(page)) |
| 1956 | goto retry; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1957 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1958 | /* Has the page moved or been split? */ |
| 1959 | if (unlikely(page != xas_reload(&xas))) |
| 1960 | goto put_page; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1961 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1962 | pages[ret] = find_subpage(page, xas.xa_index); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1963 | if (++ret == nr_pages) { |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1964 | *index = xas.xa_index + 1; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1965 | goto out; |
| 1966 | } |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1967 | continue; |
| 1968 | put_page: |
| 1969 | put_page(page); |
| 1970 | retry: |
| 1971 | xas_reset(&xas); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1972 | } |
| 1973 | |
| 1974 | /* |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1975 | * We come here when we got to @end. We take care to not overflow the |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1976 | * index @index as it confuses some of the callers. This breaks the |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1977 | * iteration when there is a page at index -1 but that is already |
| 1978 | * broken anyway. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1979 | */ |
| 1980 | if (end == (pgoff_t)-1) |
| 1981 | *index = (pgoff_t)-1; |
| 1982 | else |
| 1983 | *index = end + 1; |
| 1984 | out: |
| 1985 | rcu_read_unlock(); |
| 1986 | |
| 1987 | return ret; |
| 1988 | } |
| 1989 | EXPORT_SYMBOL(find_get_pages_range_tag); |
| 1990 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1991 | /* |
| 1992 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail |
| 1993 | * a _large_ part of the i/o request. Imagine the worst scenario: |
| 1994 | * |
| 1995 | * ---R__________________________________________B__________ |
| 1996 | * ^ reading here ^ bad block(assume 4k) |
| 1997 | * |
| 1998 | * read(R) => miss => readahead(R...B) => media error => frustrating retries |
| 1999 | * => failing the whole request => read(R) => read(R+1) => |
| 2000 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => |
| 2001 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => |
| 2002 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... |
| 2003 | * |
| 2004 | * It is going insane. Fix it by quickly scaling down the readahead size. |
| 2005 | */ |
| 2006 | static void shrink_readahead_size_eio(struct file *filp, |
| 2007 | struct file_ra_state *ra) |
| 2008 | { |
| 2009 | ra->ra_pages /= 4; |
| 2010 | } |
| 2011 | |
| 2012 | /** |
| 2013 | * generic_file_buffered_read - generic file read routine |
| 2014 | * @iocb: the iocb to read |
| 2015 | * @iter: data destination |
| 2016 | * @written: already copied |
| 2017 | * |
| 2018 | * This is a generic file read routine, and uses the |
| 2019 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
| 2020 | * |
| 2021 | * This is really ugly. But the goto's actually try to clarify some |
| 2022 | * of the logic when it comes to error handling etc. |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2023 | * |
| 2024 | * Return: |
| 2025 | * * total number of bytes copied, including those the were already @written |
| 2026 | * * negative error code if nothing was copied |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2027 | */ |
| 2028 | static ssize_t generic_file_buffered_read(struct kiocb *iocb, |
| 2029 | struct iov_iter *iter, ssize_t written) |
| 2030 | { |
| 2031 | struct file *filp = iocb->ki_filp; |
| 2032 | struct address_space *mapping = filp->f_mapping; |
| 2033 | struct inode *inode = mapping->host; |
| 2034 | struct file_ra_state *ra = &filp->f_ra; |
| 2035 | loff_t *ppos = &iocb->ki_pos; |
| 2036 | pgoff_t index; |
| 2037 | pgoff_t last_index; |
| 2038 | pgoff_t prev_index; |
| 2039 | unsigned long offset; /* offset into pagecache page */ |
| 2040 | unsigned int prev_offset; |
| 2041 | int error = 0; |
| 2042 | |
| 2043 | if (unlikely(*ppos >= inode->i_sb->s_maxbytes)) |
| 2044 | return 0; |
| 2045 | iov_iter_truncate(iter, inode->i_sb->s_maxbytes); |
| 2046 | |
| 2047 | index = *ppos >> PAGE_SHIFT; |
| 2048 | prev_index = ra->prev_pos >> PAGE_SHIFT; |
| 2049 | prev_offset = ra->prev_pos & (PAGE_SIZE-1); |
| 2050 | last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT; |
| 2051 | offset = *ppos & ~PAGE_MASK; |
| 2052 | |
| 2053 | for (;;) { |
| 2054 | struct page *page; |
| 2055 | pgoff_t end_index; |
| 2056 | loff_t isize; |
| 2057 | unsigned long nr, ret; |
| 2058 | |
| 2059 | cond_resched(); |
| 2060 | find_page: |
| 2061 | if (fatal_signal_pending(current)) { |
| 2062 | error = -EINTR; |
| 2063 | goto out; |
| 2064 | } |
| 2065 | |
| 2066 | page = find_get_page(mapping, index); |
| 2067 | if (!page) { |
| 2068 | if (iocb->ki_flags & IOCB_NOWAIT) |
| 2069 | goto would_block; |
| 2070 | page_cache_sync_readahead(mapping, |
| 2071 | ra, filp, |
| 2072 | index, last_index - index); |
| 2073 | page = find_get_page(mapping, index); |
| 2074 | if (unlikely(page == NULL)) |
| 2075 | goto no_cached_page; |
| 2076 | } |
| 2077 | if (PageReadahead(page)) { |
| 2078 | page_cache_async_readahead(mapping, |
| 2079 | ra, filp, page, |
| 2080 | index, last_index - index); |
| 2081 | } |
| 2082 | if (!PageUptodate(page)) { |
| 2083 | if (iocb->ki_flags & IOCB_NOWAIT) { |
| 2084 | put_page(page); |
| 2085 | goto would_block; |
| 2086 | } |
| 2087 | |
| 2088 | /* |
| 2089 | * See comment in do_read_cache_page on why |
| 2090 | * wait_on_page_locked is used to avoid unnecessarily |
| 2091 | * serialisations and why it's safe. |
| 2092 | */ |
| 2093 | error = wait_on_page_locked_killable(page); |
| 2094 | if (unlikely(error)) |
| 2095 | goto readpage_error; |
| 2096 | if (PageUptodate(page)) |
| 2097 | goto page_ok; |
| 2098 | |
| 2099 | if (inode->i_blkbits == PAGE_SHIFT || |
| 2100 | !mapping->a_ops->is_partially_uptodate) |
| 2101 | goto page_not_up_to_date; |
| 2102 | /* pipes can't handle partially uptodate pages */ |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2103 | if (unlikely(iov_iter_is_pipe(iter))) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2104 | goto page_not_up_to_date; |
| 2105 | if (!trylock_page(page)) |
| 2106 | goto page_not_up_to_date; |
| 2107 | /* Did it get truncated before we got the lock? */ |
| 2108 | if (!page->mapping) |
| 2109 | goto page_not_up_to_date_locked; |
| 2110 | if (!mapping->a_ops->is_partially_uptodate(page, |
| 2111 | offset, iter->count)) |
| 2112 | goto page_not_up_to_date_locked; |
| 2113 | unlock_page(page); |
| 2114 | } |
| 2115 | page_ok: |
| 2116 | /* |
| 2117 | * i_size must be checked after we know the page is Uptodate. |
| 2118 | * |
| 2119 | * Checking i_size after the check allows us to calculate |
| 2120 | * the correct value for "nr", which means the zero-filled |
| 2121 | * part of the page is not copied back to userspace (unless |
| 2122 | * another truncate extends the file - this is desired though). |
| 2123 | */ |
| 2124 | |
| 2125 | isize = i_size_read(inode); |
| 2126 | end_index = (isize - 1) >> PAGE_SHIFT; |
| 2127 | if (unlikely(!isize || index > end_index)) { |
| 2128 | put_page(page); |
| 2129 | goto out; |
| 2130 | } |
| 2131 | |
| 2132 | /* nr is the maximum number of bytes to copy from this page */ |
| 2133 | nr = PAGE_SIZE; |
| 2134 | if (index == end_index) { |
| 2135 | nr = ((isize - 1) & ~PAGE_MASK) + 1; |
| 2136 | if (nr <= offset) { |
| 2137 | put_page(page); |
| 2138 | goto out; |
| 2139 | } |
| 2140 | } |
| 2141 | nr = nr - offset; |
| 2142 | |
| 2143 | /* If users can be writing to this page using arbitrary |
| 2144 | * virtual addresses, take care about potential aliasing |
| 2145 | * before reading the page on the kernel side. |
| 2146 | */ |
| 2147 | if (mapping_writably_mapped(mapping)) |
| 2148 | flush_dcache_page(page); |
| 2149 | |
| 2150 | /* |
| 2151 | * When a sequential read accesses a page several times, |
| 2152 | * only mark it as accessed the first time. |
| 2153 | */ |
| 2154 | if (prev_index != index || offset != prev_offset) |
| 2155 | mark_page_accessed(page); |
| 2156 | prev_index = index; |
| 2157 | |
| 2158 | /* |
| 2159 | * Ok, we have the page, and it's up-to-date, so |
| 2160 | * now we can copy it to user space... |
| 2161 | */ |
| 2162 | |
| 2163 | ret = copy_page_to_iter(page, offset, nr, iter); |
| 2164 | offset += ret; |
| 2165 | index += offset >> PAGE_SHIFT; |
| 2166 | offset &= ~PAGE_MASK; |
| 2167 | prev_offset = offset; |
| 2168 | |
| 2169 | put_page(page); |
| 2170 | written += ret; |
| 2171 | if (!iov_iter_count(iter)) |
| 2172 | goto out; |
| 2173 | if (ret < nr) { |
| 2174 | error = -EFAULT; |
| 2175 | goto out; |
| 2176 | } |
| 2177 | continue; |
| 2178 | |
| 2179 | page_not_up_to_date: |
| 2180 | /* Get exclusive access to the page ... */ |
| 2181 | error = lock_page_killable(page); |
| 2182 | if (unlikely(error)) |
| 2183 | goto readpage_error; |
| 2184 | |
| 2185 | page_not_up_to_date_locked: |
| 2186 | /* Did it get truncated before we got the lock? */ |
| 2187 | if (!page->mapping) { |
| 2188 | unlock_page(page); |
| 2189 | put_page(page); |
| 2190 | continue; |
| 2191 | } |
| 2192 | |
| 2193 | /* Did somebody else fill it already? */ |
| 2194 | if (PageUptodate(page)) { |
| 2195 | unlock_page(page); |
| 2196 | goto page_ok; |
| 2197 | } |
| 2198 | |
| 2199 | readpage: |
| 2200 | /* |
| 2201 | * A previous I/O error may have been due to temporary |
| 2202 | * failures, eg. multipath errors. |
| 2203 | * PG_error will be set again if readpage fails. |
| 2204 | */ |
| 2205 | ClearPageError(page); |
| 2206 | /* Start the actual read. The read will unlock the page. */ |
| 2207 | error = mapping->a_ops->readpage(filp, page); |
| 2208 | |
| 2209 | if (unlikely(error)) { |
| 2210 | if (error == AOP_TRUNCATED_PAGE) { |
| 2211 | put_page(page); |
| 2212 | error = 0; |
| 2213 | goto find_page; |
| 2214 | } |
| 2215 | goto readpage_error; |
| 2216 | } |
| 2217 | |
| 2218 | if (!PageUptodate(page)) { |
| 2219 | error = lock_page_killable(page); |
| 2220 | if (unlikely(error)) |
| 2221 | goto readpage_error; |
| 2222 | if (!PageUptodate(page)) { |
| 2223 | if (page->mapping == NULL) { |
| 2224 | /* |
| 2225 | * invalidate_mapping_pages got it |
| 2226 | */ |
| 2227 | unlock_page(page); |
| 2228 | put_page(page); |
| 2229 | goto find_page; |
| 2230 | } |
| 2231 | unlock_page(page); |
| 2232 | shrink_readahead_size_eio(filp, ra); |
| 2233 | error = -EIO; |
| 2234 | goto readpage_error; |
| 2235 | } |
| 2236 | unlock_page(page); |
| 2237 | } |
| 2238 | |
| 2239 | goto page_ok; |
| 2240 | |
| 2241 | readpage_error: |
| 2242 | /* UHHUH! A synchronous read error occurred. Report it */ |
| 2243 | put_page(page); |
| 2244 | goto out; |
| 2245 | |
| 2246 | no_cached_page: |
| 2247 | /* |
| 2248 | * Ok, it wasn't cached, so we need to create a new |
| 2249 | * page.. |
| 2250 | */ |
| 2251 | page = page_cache_alloc(mapping); |
| 2252 | if (!page) { |
| 2253 | error = -ENOMEM; |
| 2254 | goto out; |
| 2255 | } |
| 2256 | error = add_to_page_cache_lru(page, mapping, index, |
| 2257 | mapping_gfp_constraint(mapping, GFP_KERNEL)); |
| 2258 | if (error) { |
| 2259 | put_page(page); |
| 2260 | if (error == -EEXIST) { |
| 2261 | error = 0; |
| 2262 | goto find_page; |
| 2263 | } |
| 2264 | goto out; |
| 2265 | } |
| 2266 | goto readpage; |
| 2267 | } |
| 2268 | |
| 2269 | would_block: |
| 2270 | error = -EAGAIN; |
| 2271 | out: |
| 2272 | ra->prev_pos = prev_index; |
| 2273 | ra->prev_pos <<= PAGE_SHIFT; |
| 2274 | ra->prev_pos |= prev_offset; |
| 2275 | |
| 2276 | *ppos = ((loff_t)index << PAGE_SHIFT) + offset; |
| 2277 | file_accessed(filp); |
| 2278 | return written ? written : error; |
| 2279 | } |
| 2280 | |
| 2281 | /** |
| 2282 | * generic_file_read_iter - generic filesystem read routine |
| 2283 | * @iocb: kernel I/O control block |
| 2284 | * @iter: destination for the data read |
| 2285 | * |
| 2286 | * This is the "read_iter()" routine for all filesystems |
| 2287 | * that can use the page cache directly. |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2288 | * Return: |
| 2289 | * * number of bytes copied, even for partial reads |
| 2290 | * * negative error code if nothing was read |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2291 | */ |
| 2292 | ssize_t |
| 2293 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
| 2294 | { |
| 2295 | size_t count = iov_iter_count(iter); |
| 2296 | ssize_t retval = 0; |
| 2297 | |
| 2298 | if (!count) |
| 2299 | goto out; /* skip atime */ |
| 2300 | |
| 2301 | if (iocb->ki_flags & IOCB_DIRECT) { |
| 2302 | struct file *file = iocb->ki_filp; |
| 2303 | struct address_space *mapping = file->f_mapping; |
| 2304 | struct inode *inode = mapping->host; |
| 2305 | loff_t size; |
| 2306 | |
| 2307 | size = i_size_read(inode); |
| 2308 | if (iocb->ki_flags & IOCB_NOWAIT) { |
| 2309 | if (filemap_range_has_page(mapping, iocb->ki_pos, |
| 2310 | iocb->ki_pos + count - 1)) |
| 2311 | return -EAGAIN; |
| 2312 | } else { |
| 2313 | retval = filemap_write_and_wait_range(mapping, |
| 2314 | iocb->ki_pos, |
| 2315 | iocb->ki_pos + count - 1); |
| 2316 | if (retval < 0) |
| 2317 | goto out; |
| 2318 | } |
| 2319 | |
| 2320 | file_accessed(file); |
| 2321 | |
| 2322 | retval = mapping->a_ops->direct_IO(iocb, iter); |
| 2323 | if (retval >= 0) { |
| 2324 | iocb->ki_pos += retval; |
| 2325 | count -= retval; |
| 2326 | } |
| 2327 | iov_iter_revert(iter, count - iov_iter_count(iter)); |
| 2328 | |
| 2329 | /* |
| 2330 | * Btrfs can have a short DIO read if we encounter |
| 2331 | * compressed extents, so if there was an error, or if |
| 2332 | * we've already read everything we wanted to, or if |
| 2333 | * there was a short read because we hit EOF, go ahead |
| 2334 | * and return. Otherwise fallthrough to buffered io for |
| 2335 | * the rest of the read. Buffered reads will not work for |
| 2336 | * DAX files, so don't bother trying. |
| 2337 | */ |
| 2338 | if (retval < 0 || !count || iocb->ki_pos >= size || |
| 2339 | IS_DAX(inode)) |
| 2340 | goto out; |
| 2341 | } |
| 2342 | |
| 2343 | retval = generic_file_buffered_read(iocb, iter, retval); |
| 2344 | out: |
| 2345 | return retval; |
| 2346 | } |
| 2347 | EXPORT_SYMBOL(generic_file_read_iter); |
| 2348 | |
| 2349 | #ifdef CONFIG_MMU |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2350 | #define MMAP_LOTSAMISS (100) |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2351 | /* |
| 2352 | * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_sem |
| 2353 | * @vmf - the vm_fault for this fault. |
| 2354 | * @page - the page to lock. |
| 2355 | * @fpin - the pointer to the file we may pin (or is already pinned). |
| 2356 | * |
| 2357 | * This works similar to lock_page_or_retry in that it can drop the mmap_sem. |
| 2358 | * It differs in that it actually returns the page locked if it returns 1 and 0 |
| 2359 | * if it couldn't lock the page. If we did have to drop the mmap_sem then fpin |
| 2360 | * will point to the pinned file and needs to be fput()'ed at a later point. |
| 2361 | */ |
| 2362 | static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page, |
| 2363 | struct file **fpin) |
| 2364 | { |
| 2365 | if (trylock_page(page)) |
| 2366 | return 1; |
| 2367 | |
| 2368 | /* |
| 2369 | * NOTE! This will make us return with VM_FAULT_RETRY, but with |
| 2370 | * the mmap_sem still held. That's how FAULT_FLAG_RETRY_NOWAIT |
| 2371 | * is supposed to work. We have way too many special cases.. |
| 2372 | */ |
| 2373 | if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT) |
| 2374 | return 0; |
| 2375 | |
| 2376 | *fpin = maybe_unlock_mmap_for_io(vmf, *fpin); |
| 2377 | if (vmf->flags & FAULT_FLAG_KILLABLE) { |
| 2378 | if (__lock_page_killable(page)) { |
| 2379 | /* |
| 2380 | * We didn't have the right flags to drop the mmap_sem, |
| 2381 | * but all fault_handlers only check for fatal signals |
| 2382 | * if we return VM_FAULT_RETRY, so we need to drop the |
| 2383 | * mmap_sem here and return 0 if we don't have a fpin. |
| 2384 | */ |
| 2385 | if (*fpin == NULL) |
| 2386 | up_read(&vmf->vma->vm_mm->mmap_sem); |
| 2387 | return 0; |
| 2388 | } |
| 2389 | } else |
| 2390 | __lock_page(page); |
| 2391 | return 1; |
| 2392 | } |
| 2393 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2394 | |
| 2395 | /* |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2396 | * Synchronous readahead happens when we don't even find a page in the page |
| 2397 | * cache at all. We don't want to perform IO under the mmap sem, so if we have |
| 2398 | * to drop the mmap sem we return the file that was pinned in order for us to do |
| 2399 | * that. If we didn't pin a file then we return NULL. The file that is |
| 2400 | * returned needs to be fput()'ed when we're done with it. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2401 | */ |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2402 | static struct file *do_sync_mmap_readahead(struct vm_fault *vmf) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2403 | { |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2404 | struct file *file = vmf->vma->vm_file; |
| 2405 | struct file_ra_state *ra = &file->f_ra; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2406 | struct address_space *mapping = file->f_mapping; |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2407 | struct file *fpin = NULL; |
| 2408 | pgoff_t offset = vmf->pgoff; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2409 | |
| 2410 | /* If we don't want any read-ahead, don't bother */ |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2411 | if (vmf->vma->vm_flags & VM_RAND_READ) |
| 2412 | return fpin; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2413 | if (!ra->ra_pages) |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2414 | return fpin; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2415 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2416 | if (vmf->vma->vm_flags & VM_SEQ_READ) { |
| 2417 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2418 | page_cache_sync_readahead(mapping, ra, file, offset, |
| 2419 | ra->ra_pages); |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2420 | return fpin; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2421 | } |
| 2422 | |
| 2423 | /* Avoid banging the cache line if not needed */ |
| 2424 | if (ra->mmap_miss < MMAP_LOTSAMISS * 10) |
| 2425 | ra->mmap_miss++; |
| 2426 | |
| 2427 | /* |
| 2428 | * Do we miss much more than hit in this file? If so, |
| 2429 | * stop bothering with read-ahead. It will only hurt. |
| 2430 | */ |
| 2431 | if (ra->mmap_miss > MMAP_LOTSAMISS) |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2432 | return fpin; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2433 | |
| 2434 | /* |
| 2435 | * mmap read-around |
| 2436 | */ |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2437 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2438 | ra->start = max_t(long, 0, offset - ra->ra_pages / 2); |
| 2439 | ra->size = ra->ra_pages; |
| 2440 | ra->async_size = ra->ra_pages / 4; |
| 2441 | ra_submit(ra, mapping, file); |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2442 | return fpin; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2443 | } |
| 2444 | |
| 2445 | /* |
| 2446 | * Asynchronous readahead happens when we find the page and PG_readahead, |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2447 | * so we want to possibly extend the readahead further. We return the file that |
| 2448 | * was pinned if we have to drop the mmap_sem in order to do IO. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2449 | */ |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2450 | static struct file *do_async_mmap_readahead(struct vm_fault *vmf, |
| 2451 | struct page *page) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2452 | { |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2453 | struct file *file = vmf->vma->vm_file; |
| 2454 | struct file_ra_state *ra = &file->f_ra; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2455 | struct address_space *mapping = file->f_mapping; |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2456 | struct file *fpin = NULL; |
| 2457 | pgoff_t offset = vmf->pgoff; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2458 | |
| 2459 | /* If we don't want any read-ahead, don't bother */ |
Olivier Deprez | 0e64123 | 2021-09-23 10:07:05 +0200 | [diff] [blame^] | 2460 | if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages) |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2461 | return fpin; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2462 | if (ra->mmap_miss > 0) |
| 2463 | ra->mmap_miss--; |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2464 | if (PageReadahead(page)) { |
| 2465 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2466 | page_cache_async_readahead(mapping, ra, file, |
| 2467 | page, offset, ra->ra_pages); |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2468 | } |
| 2469 | return fpin; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2470 | } |
| 2471 | |
| 2472 | /** |
| 2473 | * filemap_fault - read in file data for page fault handling |
| 2474 | * @vmf: struct vm_fault containing details of the fault |
| 2475 | * |
| 2476 | * filemap_fault() is invoked via the vma operations vector for a |
| 2477 | * mapped memory region to read in file data during a page fault. |
| 2478 | * |
| 2479 | * The goto's are kind of ugly, but this streamlines the normal case of having |
| 2480 | * it in the page cache, and handles the special cases reasonably without |
| 2481 | * having a lot of duplicated code. |
| 2482 | * |
| 2483 | * vma->vm_mm->mmap_sem must be held on entry. |
| 2484 | * |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2485 | * If our return value has VM_FAULT_RETRY set, it's because the mmap_sem |
| 2486 | * may be dropped before doing I/O or by lock_page_maybe_drop_mmap(). |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2487 | * |
| 2488 | * If our return value does not have VM_FAULT_RETRY set, the mmap_sem |
| 2489 | * has not been released. |
| 2490 | * |
| 2491 | * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set. |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2492 | * |
| 2493 | * Return: bitwise-OR of %VM_FAULT_ codes. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2494 | */ |
| 2495 | vm_fault_t filemap_fault(struct vm_fault *vmf) |
| 2496 | { |
| 2497 | int error; |
| 2498 | struct file *file = vmf->vma->vm_file; |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2499 | struct file *fpin = NULL; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2500 | struct address_space *mapping = file->f_mapping; |
| 2501 | struct file_ra_state *ra = &file->f_ra; |
| 2502 | struct inode *inode = mapping->host; |
| 2503 | pgoff_t offset = vmf->pgoff; |
| 2504 | pgoff_t max_off; |
| 2505 | struct page *page; |
| 2506 | vm_fault_t ret = 0; |
| 2507 | |
| 2508 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
| 2509 | if (unlikely(offset >= max_off)) |
| 2510 | return VM_FAULT_SIGBUS; |
| 2511 | |
| 2512 | /* |
| 2513 | * Do we have something in the page cache already? |
| 2514 | */ |
| 2515 | page = find_get_page(mapping, offset); |
| 2516 | if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) { |
| 2517 | /* |
| 2518 | * We found the page, so try async readahead before |
| 2519 | * waiting for the lock. |
| 2520 | */ |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2521 | fpin = do_async_mmap_readahead(vmf, page); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2522 | } else if (!page) { |
| 2523 | /* No page in the page cache at all */ |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2524 | count_vm_event(PGMAJFAULT); |
| 2525 | count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT); |
| 2526 | ret = VM_FAULT_MAJOR; |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2527 | fpin = do_sync_mmap_readahead(vmf); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2528 | retry_find: |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2529 | page = pagecache_get_page(mapping, offset, |
| 2530 | FGP_CREAT|FGP_FOR_MMAP, |
| 2531 | vmf->gfp_mask); |
| 2532 | if (!page) { |
| 2533 | if (fpin) |
| 2534 | goto out_retry; |
| 2535 | return vmf_error(-ENOMEM); |
| 2536 | } |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2537 | } |
| 2538 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2539 | if (!lock_page_maybe_drop_mmap(vmf, page, &fpin)) |
| 2540 | goto out_retry; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2541 | |
| 2542 | /* Did it get truncated? */ |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2543 | if (unlikely(compound_head(page)->mapping != mapping)) { |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2544 | unlock_page(page); |
| 2545 | put_page(page); |
| 2546 | goto retry_find; |
| 2547 | } |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2548 | VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2549 | |
| 2550 | /* |
| 2551 | * We have a locked page in the page cache, now we need to check |
| 2552 | * that it's up-to-date. If not, it is going to be due to an error. |
| 2553 | */ |
| 2554 | if (unlikely(!PageUptodate(page))) |
| 2555 | goto page_not_uptodate; |
| 2556 | |
| 2557 | /* |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2558 | * We've made it this far and we had to drop our mmap_sem, now is the |
| 2559 | * time to return to the upper layer and have it re-find the vma and |
| 2560 | * redo the fault. |
| 2561 | */ |
| 2562 | if (fpin) { |
| 2563 | unlock_page(page); |
| 2564 | goto out_retry; |
| 2565 | } |
| 2566 | |
| 2567 | /* |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2568 | * Found the page and have a reference on it. |
| 2569 | * We must recheck i_size under page lock. |
| 2570 | */ |
| 2571 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
| 2572 | if (unlikely(offset >= max_off)) { |
| 2573 | unlock_page(page); |
| 2574 | put_page(page); |
| 2575 | return VM_FAULT_SIGBUS; |
| 2576 | } |
| 2577 | |
| 2578 | vmf->page = page; |
| 2579 | return ret | VM_FAULT_LOCKED; |
| 2580 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2581 | page_not_uptodate: |
| 2582 | /* |
| 2583 | * Umm, take care of errors if the page isn't up-to-date. |
| 2584 | * Try to re-read it _once_. We do this synchronously, |
| 2585 | * because there really aren't any performance issues here |
| 2586 | * and we need to check for errors. |
| 2587 | */ |
| 2588 | ClearPageError(page); |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2589 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2590 | error = mapping->a_ops->readpage(file, page); |
| 2591 | if (!error) { |
| 2592 | wait_on_page_locked(page); |
| 2593 | if (!PageUptodate(page)) |
| 2594 | error = -EIO; |
| 2595 | } |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2596 | if (fpin) |
| 2597 | goto out_retry; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2598 | put_page(page); |
| 2599 | |
| 2600 | if (!error || error == AOP_TRUNCATED_PAGE) |
| 2601 | goto retry_find; |
| 2602 | |
| 2603 | /* Things didn't work out. Return zero to tell the mm layer so. */ |
| 2604 | shrink_readahead_size_eio(file, ra); |
| 2605 | return VM_FAULT_SIGBUS; |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2606 | |
| 2607 | out_retry: |
| 2608 | /* |
| 2609 | * We dropped the mmap_sem, we need to return to the fault handler to |
| 2610 | * re-find the vma and come back and find our hopefully still populated |
| 2611 | * page. |
| 2612 | */ |
| 2613 | if (page) |
| 2614 | put_page(page); |
| 2615 | if (fpin) |
| 2616 | fput(fpin); |
| 2617 | return ret | VM_FAULT_RETRY; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2618 | } |
| 2619 | EXPORT_SYMBOL(filemap_fault); |
| 2620 | |
| 2621 | void filemap_map_pages(struct vm_fault *vmf, |
| 2622 | pgoff_t start_pgoff, pgoff_t end_pgoff) |
| 2623 | { |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2624 | struct file *file = vmf->vma->vm_file; |
| 2625 | struct address_space *mapping = file->f_mapping; |
| 2626 | pgoff_t last_pgoff = start_pgoff; |
| 2627 | unsigned long max_idx; |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2628 | XA_STATE(xas, &mapping->i_pages, start_pgoff); |
| 2629 | struct page *page; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2630 | |
| 2631 | rcu_read_lock(); |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2632 | xas_for_each(&xas, page, end_pgoff) { |
| 2633 | if (xas_retry(&xas, page)) |
| 2634 | continue; |
| 2635 | if (xa_is_value(page)) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2636 | goto next; |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2637 | |
| 2638 | /* |
| 2639 | * Check for a locked page first, as a speculative |
| 2640 | * reference may adversely influence page migration. |
| 2641 | */ |
| 2642 | if (PageLocked(page)) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2643 | goto next; |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2644 | if (!page_cache_get_speculative(page)) |
| 2645 | goto next; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2646 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2647 | /* Has the page moved or been split? */ |
| 2648 | if (unlikely(page != xas_reload(&xas))) |
| 2649 | goto skip; |
| 2650 | page = find_subpage(page, xas.xa_index); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2651 | |
| 2652 | if (!PageUptodate(page) || |
| 2653 | PageReadahead(page) || |
| 2654 | PageHWPoison(page)) |
| 2655 | goto skip; |
| 2656 | if (!trylock_page(page)) |
| 2657 | goto skip; |
| 2658 | |
| 2659 | if (page->mapping != mapping || !PageUptodate(page)) |
| 2660 | goto unlock; |
| 2661 | |
| 2662 | max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE); |
| 2663 | if (page->index >= max_idx) |
| 2664 | goto unlock; |
| 2665 | |
| 2666 | if (file->f_ra.mmap_miss > 0) |
| 2667 | file->f_ra.mmap_miss--; |
| 2668 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2669 | vmf->address += (xas.xa_index - last_pgoff) << PAGE_SHIFT; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2670 | if (vmf->pte) |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2671 | vmf->pte += xas.xa_index - last_pgoff; |
| 2672 | last_pgoff = xas.xa_index; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2673 | if (alloc_set_pte(vmf, NULL, page)) |
| 2674 | goto unlock; |
| 2675 | unlock_page(page); |
| 2676 | goto next; |
| 2677 | unlock: |
| 2678 | unlock_page(page); |
| 2679 | skip: |
| 2680 | put_page(page); |
| 2681 | next: |
| 2682 | /* Huge page is mapped? No need to proceed. */ |
| 2683 | if (pmd_trans_huge(*vmf->pmd)) |
| 2684 | break; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2685 | } |
| 2686 | rcu_read_unlock(); |
| 2687 | } |
| 2688 | EXPORT_SYMBOL(filemap_map_pages); |
| 2689 | |
| 2690 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
| 2691 | { |
| 2692 | struct page *page = vmf->page; |
| 2693 | struct inode *inode = file_inode(vmf->vma->vm_file); |
| 2694 | vm_fault_t ret = VM_FAULT_LOCKED; |
| 2695 | |
| 2696 | sb_start_pagefault(inode->i_sb); |
| 2697 | file_update_time(vmf->vma->vm_file); |
| 2698 | lock_page(page); |
| 2699 | if (page->mapping != inode->i_mapping) { |
| 2700 | unlock_page(page); |
| 2701 | ret = VM_FAULT_NOPAGE; |
| 2702 | goto out; |
| 2703 | } |
| 2704 | /* |
| 2705 | * We mark the page dirty already here so that when freeze is in |
| 2706 | * progress, we are guaranteed that writeback during freezing will |
| 2707 | * see the dirty page and writeprotect it again. |
| 2708 | */ |
| 2709 | set_page_dirty(page); |
| 2710 | wait_for_stable_page(page); |
| 2711 | out: |
| 2712 | sb_end_pagefault(inode->i_sb); |
| 2713 | return ret; |
| 2714 | } |
| 2715 | |
| 2716 | const struct vm_operations_struct generic_file_vm_ops = { |
| 2717 | .fault = filemap_fault, |
| 2718 | .map_pages = filemap_map_pages, |
| 2719 | .page_mkwrite = filemap_page_mkwrite, |
| 2720 | }; |
| 2721 | |
| 2722 | /* This is used for a general mmap of a disk file */ |
| 2723 | |
| 2724 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) |
| 2725 | { |
| 2726 | struct address_space *mapping = file->f_mapping; |
| 2727 | |
| 2728 | if (!mapping->a_ops->readpage) |
| 2729 | return -ENOEXEC; |
| 2730 | file_accessed(file); |
| 2731 | vma->vm_ops = &generic_file_vm_ops; |
| 2732 | return 0; |
| 2733 | } |
| 2734 | |
| 2735 | /* |
| 2736 | * This is for filesystems which do not implement ->writepage. |
| 2737 | */ |
| 2738 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) |
| 2739 | { |
| 2740 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) |
| 2741 | return -EINVAL; |
| 2742 | return generic_file_mmap(file, vma); |
| 2743 | } |
| 2744 | #else |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2745 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2746 | { |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2747 | return VM_FAULT_SIGBUS; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2748 | } |
| 2749 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) |
| 2750 | { |
| 2751 | return -ENOSYS; |
| 2752 | } |
| 2753 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) |
| 2754 | { |
| 2755 | return -ENOSYS; |
| 2756 | } |
| 2757 | #endif /* CONFIG_MMU */ |
| 2758 | |
| 2759 | EXPORT_SYMBOL(filemap_page_mkwrite); |
| 2760 | EXPORT_SYMBOL(generic_file_mmap); |
| 2761 | EXPORT_SYMBOL(generic_file_readonly_mmap); |
| 2762 | |
| 2763 | static struct page *wait_on_page_read(struct page *page) |
| 2764 | { |
| 2765 | if (!IS_ERR(page)) { |
| 2766 | wait_on_page_locked(page); |
| 2767 | if (!PageUptodate(page)) { |
| 2768 | put_page(page); |
| 2769 | page = ERR_PTR(-EIO); |
| 2770 | } |
| 2771 | } |
| 2772 | return page; |
| 2773 | } |
| 2774 | |
| 2775 | static struct page *do_read_cache_page(struct address_space *mapping, |
| 2776 | pgoff_t index, |
| 2777 | int (*filler)(void *, struct page *), |
| 2778 | void *data, |
| 2779 | gfp_t gfp) |
| 2780 | { |
| 2781 | struct page *page; |
| 2782 | int err; |
| 2783 | repeat: |
| 2784 | page = find_get_page(mapping, index); |
| 2785 | if (!page) { |
| 2786 | page = __page_cache_alloc(gfp); |
| 2787 | if (!page) |
| 2788 | return ERR_PTR(-ENOMEM); |
| 2789 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
| 2790 | if (unlikely(err)) { |
| 2791 | put_page(page); |
| 2792 | if (err == -EEXIST) |
| 2793 | goto repeat; |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2794 | /* Presumably ENOMEM for xarray node */ |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2795 | return ERR_PTR(err); |
| 2796 | } |
| 2797 | |
| 2798 | filler: |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2799 | if (filler) |
| 2800 | err = filler(data, page); |
| 2801 | else |
| 2802 | err = mapping->a_ops->readpage(data, page); |
| 2803 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2804 | if (err < 0) { |
| 2805 | put_page(page); |
| 2806 | return ERR_PTR(err); |
| 2807 | } |
| 2808 | |
| 2809 | page = wait_on_page_read(page); |
| 2810 | if (IS_ERR(page)) |
| 2811 | return page; |
| 2812 | goto out; |
| 2813 | } |
| 2814 | if (PageUptodate(page)) |
| 2815 | goto out; |
| 2816 | |
| 2817 | /* |
| 2818 | * Page is not up to date and may be locked due one of the following |
| 2819 | * case a: Page is being filled and the page lock is held |
| 2820 | * case b: Read/write error clearing the page uptodate status |
| 2821 | * case c: Truncation in progress (page locked) |
| 2822 | * case d: Reclaim in progress |
| 2823 | * |
| 2824 | * Case a, the page will be up to date when the page is unlocked. |
| 2825 | * There is no need to serialise on the page lock here as the page |
| 2826 | * is pinned so the lock gives no additional protection. Even if the |
| 2827 | * the page is truncated, the data is still valid if PageUptodate as |
| 2828 | * it's a race vs truncate race. |
| 2829 | * Case b, the page will not be up to date |
| 2830 | * Case c, the page may be truncated but in itself, the data may still |
| 2831 | * be valid after IO completes as it's a read vs truncate race. The |
| 2832 | * operation must restart if the page is not uptodate on unlock but |
| 2833 | * otherwise serialising on page lock to stabilise the mapping gives |
| 2834 | * no additional guarantees to the caller as the page lock is |
| 2835 | * released before return. |
| 2836 | * Case d, similar to truncation. If reclaim holds the page lock, it |
| 2837 | * will be a race with remove_mapping that determines if the mapping |
| 2838 | * is valid on unlock but otherwise the data is valid and there is |
| 2839 | * no need to serialise with page lock. |
| 2840 | * |
| 2841 | * As the page lock gives no additional guarantee, we optimistically |
| 2842 | * wait on the page to be unlocked and check if it's up to date and |
| 2843 | * use the page if it is. Otherwise, the page lock is required to |
| 2844 | * distinguish between the different cases. The motivation is that we |
| 2845 | * avoid spurious serialisations and wakeups when multiple processes |
| 2846 | * wait on the same page for IO to complete. |
| 2847 | */ |
| 2848 | wait_on_page_locked(page); |
| 2849 | if (PageUptodate(page)) |
| 2850 | goto out; |
| 2851 | |
| 2852 | /* Distinguish between all the cases under the safety of the lock */ |
| 2853 | lock_page(page); |
| 2854 | |
| 2855 | /* Case c or d, restart the operation */ |
| 2856 | if (!page->mapping) { |
| 2857 | unlock_page(page); |
| 2858 | put_page(page); |
| 2859 | goto repeat; |
| 2860 | } |
| 2861 | |
| 2862 | /* Someone else locked and filled the page in a very small window */ |
| 2863 | if (PageUptodate(page)) { |
| 2864 | unlock_page(page); |
| 2865 | goto out; |
| 2866 | } |
Olivier Deprez | 0e64123 | 2021-09-23 10:07:05 +0200 | [diff] [blame^] | 2867 | |
| 2868 | /* |
| 2869 | * A previous I/O error may have been due to temporary |
| 2870 | * failures. |
| 2871 | * Clear page error before actual read, PG_error will be |
| 2872 | * set again if read page fails. |
| 2873 | */ |
| 2874 | ClearPageError(page); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2875 | goto filler; |
| 2876 | |
| 2877 | out: |
| 2878 | mark_page_accessed(page); |
| 2879 | return page; |
| 2880 | } |
| 2881 | |
| 2882 | /** |
| 2883 | * read_cache_page - read into page cache, fill it if needed |
| 2884 | * @mapping: the page's address_space |
| 2885 | * @index: the page index |
| 2886 | * @filler: function to perform the read |
| 2887 | * @data: first arg to filler(data, page) function, often left as NULL |
| 2888 | * |
| 2889 | * Read into the page cache. If a page already exists, and PageUptodate() is |
| 2890 | * not set, try to fill the page and wait for it to become unlocked. |
| 2891 | * |
| 2892 | * If the page does not get brought uptodate, return -EIO. |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2893 | * |
| 2894 | * Return: up to date page on success, ERR_PTR() on failure. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2895 | */ |
| 2896 | struct page *read_cache_page(struct address_space *mapping, |
| 2897 | pgoff_t index, |
| 2898 | int (*filler)(void *, struct page *), |
| 2899 | void *data) |
| 2900 | { |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2901 | return do_read_cache_page(mapping, index, filler, data, |
| 2902 | mapping_gfp_mask(mapping)); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2903 | } |
| 2904 | EXPORT_SYMBOL(read_cache_page); |
| 2905 | |
| 2906 | /** |
| 2907 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. |
| 2908 | * @mapping: the page's address_space |
| 2909 | * @index: the page index |
| 2910 | * @gfp: the page allocator flags to use if allocating |
| 2911 | * |
| 2912 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with |
| 2913 | * any new page allocations done using the specified allocation flags. |
| 2914 | * |
| 2915 | * If the page does not get brought uptodate, return -EIO. |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2916 | * |
| 2917 | * Return: up to date page on success, ERR_PTR() on failure. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2918 | */ |
| 2919 | struct page *read_cache_page_gfp(struct address_space *mapping, |
| 2920 | pgoff_t index, |
| 2921 | gfp_t gfp) |
| 2922 | { |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2923 | return do_read_cache_page(mapping, index, NULL, NULL, gfp); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2924 | } |
| 2925 | EXPORT_SYMBOL(read_cache_page_gfp); |
| 2926 | |
| 2927 | /* |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2928 | * Don't operate on ranges the page cache doesn't support, and don't exceed the |
| 2929 | * LFS limits. If pos is under the limit it becomes a short access. If it |
| 2930 | * exceeds the limit we return -EFBIG. |
| 2931 | */ |
| 2932 | static int generic_write_check_limits(struct file *file, loff_t pos, |
| 2933 | loff_t *count) |
| 2934 | { |
| 2935 | struct inode *inode = file->f_mapping->host; |
| 2936 | loff_t max_size = inode->i_sb->s_maxbytes; |
| 2937 | loff_t limit = rlimit(RLIMIT_FSIZE); |
| 2938 | |
| 2939 | if (limit != RLIM_INFINITY) { |
| 2940 | if (pos >= limit) { |
| 2941 | send_sig(SIGXFSZ, current, 0); |
| 2942 | return -EFBIG; |
| 2943 | } |
| 2944 | *count = min(*count, limit - pos); |
| 2945 | } |
| 2946 | |
| 2947 | if (!(file->f_flags & O_LARGEFILE)) |
| 2948 | max_size = MAX_NON_LFS; |
| 2949 | |
| 2950 | if (unlikely(pos >= max_size)) |
| 2951 | return -EFBIG; |
| 2952 | |
| 2953 | *count = min(*count, max_size - pos); |
| 2954 | |
| 2955 | return 0; |
| 2956 | } |
| 2957 | |
| 2958 | /* |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2959 | * Performs necessary checks before doing a write |
| 2960 | * |
| 2961 | * Can adjust writing position or amount of bytes to write. |
| 2962 | * Returns appropriate error code that caller should return or |
| 2963 | * zero in case that write should be allowed. |
| 2964 | */ |
| 2965 | inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from) |
| 2966 | { |
| 2967 | struct file *file = iocb->ki_filp; |
| 2968 | struct inode *inode = file->f_mapping->host; |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2969 | loff_t count; |
| 2970 | int ret; |
| 2971 | |
| 2972 | if (IS_SWAPFILE(inode)) |
| 2973 | return -ETXTBSY; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2974 | |
| 2975 | if (!iov_iter_count(from)) |
| 2976 | return 0; |
| 2977 | |
| 2978 | /* FIXME: this is for backwards compatibility with 2.4 */ |
| 2979 | if (iocb->ki_flags & IOCB_APPEND) |
| 2980 | iocb->ki_pos = i_size_read(inode); |
| 2981 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2982 | if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT)) |
| 2983 | return -EINVAL; |
| 2984 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2985 | count = iov_iter_count(from); |
| 2986 | ret = generic_write_check_limits(file, iocb->ki_pos, &count); |
| 2987 | if (ret) |
| 2988 | return ret; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2989 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2990 | iov_iter_truncate(from, count); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2991 | return iov_iter_count(from); |
| 2992 | } |
| 2993 | EXPORT_SYMBOL(generic_write_checks); |
| 2994 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 2995 | /* |
| 2996 | * Performs necessary checks before doing a clone. |
| 2997 | * |
| 2998 | * Can adjust amount of bytes to clone via @req_count argument. |
| 2999 | * Returns appropriate error code that caller should return or |
| 3000 | * zero in case the clone should be allowed. |
| 3001 | */ |
| 3002 | int generic_remap_checks(struct file *file_in, loff_t pos_in, |
| 3003 | struct file *file_out, loff_t pos_out, |
| 3004 | loff_t *req_count, unsigned int remap_flags) |
| 3005 | { |
| 3006 | struct inode *inode_in = file_in->f_mapping->host; |
| 3007 | struct inode *inode_out = file_out->f_mapping->host; |
| 3008 | uint64_t count = *req_count; |
| 3009 | uint64_t bcount; |
| 3010 | loff_t size_in, size_out; |
| 3011 | loff_t bs = inode_out->i_sb->s_blocksize; |
| 3012 | int ret; |
| 3013 | |
| 3014 | /* The start of both ranges must be aligned to an fs block. */ |
| 3015 | if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_out, bs)) |
| 3016 | return -EINVAL; |
| 3017 | |
| 3018 | /* Ensure offsets don't wrap. */ |
| 3019 | if (pos_in + count < pos_in || pos_out + count < pos_out) |
| 3020 | return -EINVAL; |
| 3021 | |
| 3022 | size_in = i_size_read(inode_in); |
| 3023 | size_out = i_size_read(inode_out); |
| 3024 | |
| 3025 | /* Dedupe requires both ranges to be within EOF. */ |
| 3026 | if ((remap_flags & REMAP_FILE_DEDUP) && |
| 3027 | (pos_in >= size_in || pos_in + count > size_in || |
| 3028 | pos_out >= size_out || pos_out + count > size_out)) |
| 3029 | return -EINVAL; |
| 3030 | |
| 3031 | /* Ensure the infile range is within the infile. */ |
| 3032 | if (pos_in >= size_in) |
| 3033 | return -EINVAL; |
| 3034 | count = min(count, size_in - (uint64_t)pos_in); |
| 3035 | |
| 3036 | ret = generic_write_check_limits(file_out, pos_out, &count); |
| 3037 | if (ret) |
| 3038 | return ret; |
| 3039 | |
| 3040 | /* |
| 3041 | * If the user wanted us to link to the infile's EOF, round up to the |
| 3042 | * next block boundary for this check. |
| 3043 | * |
| 3044 | * Otherwise, make sure the count is also block-aligned, having |
| 3045 | * already confirmed the starting offsets' block alignment. |
| 3046 | */ |
| 3047 | if (pos_in + count == size_in) { |
| 3048 | bcount = ALIGN(size_in, bs) - pos_in; |
| 3049 | } else { |
| 3050 | if (!IS_ALIGNED(count, bs)) |
| 3051 | count = ALIGN_DOWN(count, bs); |
| 3052 | bcount = count; |
| 3053 | } |
| 3054 | |
| 3055 | /* Don't allow overlapped cloning within the same file. */ |
| 3056 | if (inode_in == inode_out && |
| 3057 | pos_out + bcount > pos_in && |
| 3058 | pos_out < pos_in + bcount) |
| 3059 | return -EINVAL; |
| 3060 | |
| 3061 | /* |
| 3062 | * We shortened the request but the caller can't deal with that, so |
| 3063 | * bounce the request back to userspace. |
| 3064 | */ |
| 3065 | if (*req_count != count && !(remap_flags & REMAP_FILE_CAN_SHORTEN)) |
| 3066 | return -EINVAL; |
| 3067 | |
| 3068 | *req_count = count; |
| 3069 | return 0; |
| 3070 | } |
| 3071 | |
| 3072 | |
| 3073 | /* |
| 3074 | * Performs common checks before doing a file copy/clone |
| 3075 | * from @file_in to @file_out. |
| 3076 | */ |
| 3077 | int generic_file_rw_checks(struct file *file_in, struct file *file_out) |
| 3078 | { |
| 3079 | struct inode *inode_in = file_inode(file_in); |
| 3080 | struct inode *inode_out = file_inode(file_out); |
| 3081 | |
| 3082 | /* Don't copy dirs, pipes, sockets... */ |
| 3083 | if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode)) |
| 3084 | return -EISDIR; |
| 3085 | if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode)) |
| 3086 | return -EINVAL; |
| 3087 | |
| 3088 | if (!(file_in->f_mode & FMODE_READ) || |
| 3089 | !(file_out->f_mode & FMODE_WRITE) || |
| 3090 | (file_out->f_flags & O_APPEND)) |
| 3091 | return -EBADF; |
| 3092 | |
| 3093 | return 0; |
| 3094 | } |
| 3095 | |
| 3096 | /* |
| 3097 | * Performs necessary checks before doing a file copy |
| 3098 | * |
| 3099 | * Can adjust amount of bytes to copy via @req_count argument. |
| 3100 | * Returns appropriate error code that caller should return or |
| 3101 | * zero in case the copy should be allowed. |
| 3102 | */ |
| 3103 | int generic_copy_file_checks(struct file *file_in, loff_t pos_in, |
| 3104 | struct file *file_out, loff_t pos_out, |
| 3105 | size_t *req_count, unsigned int flags) |
| 3106 | { |
| 3107 | struct inode *inode_in = file_inode(file_in); |
| 3108 | struct inode *inode_out = file_inode(file_out); |
| 3109 | uint64_t count = *req_count; |
| 3110 | loff_t size_in; |
| 3111 | int ret; |
| 3112 | |
| 3113 | ret = generic_file_rw_checks(file_in, file_out); |
| 3114 | if (ret) |
| 3115 | return ret; |
| 3116 | |
| 3117 | /* Don't touch certain kinds of inodes */ |
| 3118 | if (IS_IMMUTABLE(inode_out)) |
| 3119 | return -EPERM; |
| 3120 | |
| 3121 | if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out)) |
| 3122 | return -ETXTBSY; |
| 3123 | |
| 3124 | /* Ensure offsets don't wrap. */ |
| 3125 | if (pos_in + count < pos_in || pos_out + count < pos_out) |
| 3126 | return -EOVERFLOW; |
| 3127 | |
| 3128 | /* Shorten the copy to EOF */ |
| 3129 | size_in = i_size_read(inode_in); |
| 3130 | if (pos_in >= size_in) |
| 3131 | count = 0; |
| 3132 | else |
| 3133 | count = min(count, size_in - (uint64_t)pos_in); |
| 3134 | |
| 3135 | ret = generic_write_check_limits(file_out, pos_out, &count); |
| 3136 | if (ret) |
| 3137 | return ret; |
| 3138 | |
| 3139 | /* Don't allow overlapped copying within the same file. */ |
| 3140 | if (inode_in == inode_out && |
| 3141 | pos_out + count > pos_in && |
| 3142 | pos_out < pos_in + count) |
| 3143 | return -EINVAL; |
| 3144 | |
| 3145 | *req_count = count; |
| 3146 | return 0; |
| 3147 | } |
| 3148 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 3149 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
| 3150 | loff_t pos, unsigned len, unsigned flags, |
| 3151 | struct page **pagep, void **fsdata) |
| 3152 | { |
| 3153 | const struct address_space_operations *aops = mapping->a_ops; |
| 3154 | |
| 3155 | return aops->write_begin(file, mapping, pos, len, flags, |
| 3156 | pagep, fsdata); |
| 3157 | } |
| 3158 | EXPORT_SYMBOL(pagecache_write_begin); |
| 3159 | |
| 3160 | int pagecache_write_end(struct file *file, struct address_space *mapping, |
| 3161 | loff_t pos, unsigned len, unsigned copied, |
| 3162 | struct page *page, void *fsdata) |
| 3163 | { |
| 3164 | const struct address_space_operations *aops = mapping->a_ops; |
| 3165 | |
| 3166 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); |
| 3167 | } |
| 3168 | EXPORT_SYMBOL(pagecache_write_end); |
| 3169 | |
| 3170 | ssize_t |
| 3171 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from) |
| 3172 | { |
| 3173 | struct file *file = iocb->ki_filp; |
| 3174 | struct address_space *mapping = file->f_mapping; |
| 3175 | struct inode *inode = mapping->host; |
| 3176 | loff_t pos = iocb->ki_pos; |
| 3177 | ssize_t written; |
| 3178 | size_t write_len; |
| 3179 | pgoff_t end; |
| 3180 | |
| 3181 | write_len = iov_iter_count(from); |
| 3182 | end = (pos + write_len - 1) >> PAGE_SHIFT; |
| 3183 | |
| 3184 | if (iocb->ki_flags & IOCB_NOWAIT) { |
| 3185 | /* If there are pages to writeback, return */ |
| 3186 | if (filemap_range_has_page(inode->i_mapping, pos, |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 3187 | pos + write_len - 1)) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 3188 | return -EAGAIN; |
| 3189 | } else { |
| 3190 | written = filemap_write_and_wait_range(mapping, pos, |
| 3191 | pos + write_len - 1); |
| 3192 | if (written) |
| 3193 | goto out; |
| 3194 | } |
| 3195 | |
| 3196 | /* |
| 3197 | * After a write we want buffered reads to be sure to go to disk to get |
| 3198 | * the new data. We invalidate clean cached page from the region we're |
| 3199 | * about to write. We do this *before* the write so that we can return |
| 3200 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
| 3201 | */ |
| 3202 | written = invalidate_inode_pages2_range(mapping, |
| 3203 | pos >> PAGE_SHIFT, end); |
| 3204 | /* |
| 3205 | * If a page can not be invalidated, return 0 to fall back |
| 3206 | * to buffered write. |
| 3207 | */ |
| 3208 | if (written) { |
| 3209 | if (written == -EBUSY) |
| 3210 | return 0; |
| 3211 | goto out; |
| 3212 | } |
| 3213 | |
| 3214 | written = mapping->a_ops->direct_IO(iocb, from); |
| 3215 | |
| 3216 | /* |
| 3217 | * Finally, try again to invalidate clean pages which might have been |
| 3218 | * cached by non-direct readahead, or faulted in by get_user_pages() |
| 3219 | * if the source of the write was an mmap'ed region of the file |
| 3220 | * we're writing. Either one is a pretty crazy thing to do, |
| 3221 | * so we don't support it 100%. If this invalidation |
| 3222 | * fails, tough, the write still worked... |
| 3223 | * |
| 3224 | * Most of the time we do not need this since dio_complete() will do |
| 3225 | * the invalidation for us. However there are some file systems that |
| 3226 | * do not end up with dio_complete() being called, so let's not break |
| 3227 | * them by removing it completely |
| 3228 | */ |
| 3229 | if (mapping->nrpages) |
| 3230 | invalidate_inode_pages2_range(mapping, |
| 3231 | pos >> PAGE_SHIFT, end); |
| 3232 | |
| 3233 | if (written > 0) { |
| 3234 | pos += written; |
| 3235 | write_len -= written; |
| 3236 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
| 3237 | i_size_write(inode, pos); |
| 3238 | mark_inode_dirty(inode); |
| 3239 | } |
| 3240 | iocb->ki_pos = pos; |
| 3241 | } |
| 3242 | iov_iter_revert(from, write_len - iov_iter_count(from)); |
| 3243 | out: |
| 3244 | return written; |
| 3245 | } |
| 3246 | EXPORT_SYMBOL(generic_file_direct_write); |
| 3247 | |
| 3248 | /* |
| 3249 | * Find or create a page at the given pagecache position. Return the locked |
| 3250 | * page. This function is specifically for buffered writes. |
| 3251 | */ |
| 3252 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
| 3253 | pgoff_t index, unsigned flags) |
| 3254 | { |
| 3255 | struct page *page; |
| 3256 | int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT; |
| 3257 | |
| 3258 | if (flags & AOP_FLAG_NOFS) |
| 3259 | fgp_flags |= FGP_NOFS; |
| 3260 | |
| 3261 | page = pagecache_get_page(mapping, index, fgp_flags, |
| 3262 | mapping_gfp_mask(mapping)); |
| 3263 | if (page) |
| 3264 | wait_for_stable_page(page); |
| 3265 | |
| 3266 | return page; |
| 3267 | } |
| 3268 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
| 3269 | |
| 3270 | ssize_t generic_perform_write(struct file *file, |
| 3271 | struct iov_iter *i, loff_t pos) |
| 3272 | { |
| 3273 | struct address_space *mapping = file->f_mapping; |
| 3274 | const struct address_space_operations *a_ops = mapping->a_ops; |
| 3275 | long status = 0; |
| 3276 | ssize_t written = 0; |
| 3277 | unsigned int flags = 0; |
| 3278 | |
| 3279 | do { |
| 3280 | struct page *page; |
| 3281 | unsigned long offset; /* Offset into pagecache page */ |
| 3282 | unsigned long bytes; /* Bytes to write to page */ |
| 3283 | size_t copied; /* Bytes copied from user */ |
| 3284 | void *fsdata; |
| 3285 | |
| 3286 | offset = (pos & (PAGE_SIZE - 1)); |
| 3287 | bytes = min_t(unsigned long, PAGE_SIZE - offset, |
| 3288 | iov_iter_count(i)); |
| 3289 | |
| 3290 | again: |
| 3291 | /* |
| 3292 | * Bring in the user page that we will copy from _first_. |
| 3293 | * Otherwise there's a nasty deadlock on copying from the |
| 3294 | * same page as we're writing to, without it being marked |
| 3295 | * up-to-date. |
| 3296 | * |
| 3297 | * Not only is this an optimisation, but it is also required |
| 3298 | * to check that the address is actually valid, when atomic |
| 3299 | * usercopies are used, below. |
| 3300 | */ |
| 3301 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { |
| 3302 | status = -EFAULT; |
| 3303 | break; |
| 3304 | } |
| 3305 | |
| 3306 | if (fatal_signal_pending(current)) { |
| 3307 | status = -EINTR; |
| 3308 | break; |
| 3309 | } |
| 3310 | |
| 3311 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
| 3312 | &page, &fsdata); |
| 3313 | if (unlikely(status < 0)) |
| 3314 | break; |
| 3315 | |
| 3316 | if (mapping_writably_mapped(mapping)) |
| 3317 | flush_dcache_page(page); |
| 3318 | |
| 3319 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); |
| 3320 | flush_dcache_page(page); |
| 3321 | |
| 3322 | status = a_ops->write_end(file, mapping, pos, bytes, copied, |
| 3323 | page, fsdata); |
| 3324 | if (unlikely(status < 0)) |
| 3325 | break; |
| 3326 | copied = status; |
| 3327 | |
| 3328 | cond_resched(); |
| 3329 | |
| 3330 | iov_iter_advance(i, copied); |
| 3331 | if (unlikely(copied == 0)) { |
| 3332 | /* |
| 3333 | * If we were unable to copy any data at all, we must |
| 3334 | * fall back to a single segment length write. |
| 3335 | * |
| 3336 | * If we didn't fallback here, we could livelock |
| 3337 | * because not all segments in the iov can be copied at |
| 3338 | * once without a pagefault. |
| 3339 | */ |
| 3340 | bytes = min_t(unsigned long, PAGE_SIZE - offset, |
| 3341 | iov_iter_single_seg_count(i)); |
| 3342 | goto again; |
| 3343 | } |
| 3344 | pos += copied; |
| 3345 | written += copied; |
| 3346 | |
| 3347 | balance_dirty_pages_ratelimited(mapping); |
| 3348 | } while (iov_iter_count(i)); |
| 3349 | |
| 3350 | return written ? written : status; |
| 3351 | } |
| 3352 | EXPORT_SYMBOL(generic_perform_write); |
| 3353 | |
| 3354 | /** |
| 3355 | * __generic_file_write_iter - write data to a file |
| 3356 | * @iocb: IO state structure (file, offset, etc.) |
| 3357 | * @from: iov_iter with data to write |
| 3358 | * |
| 3359 | * This function does all the work needed for actually writing data to a |
| 3360 | * file. It does all basic checks, removes SUID from the file, updates |
| 3361 | * modification times and calls proper subroutines depending on whether we |
| 3362 | * do direct IO or a standard buffered write. |
| 3363 | * |
| 3364 | * It expects i_mutex to be grabbed unless we work on a block device or similar |
| 3365 | * object which does not need locking at all. |
| 3366 | * |
| 3367 | * This function does *not* take care of syncing data in case of O_SYNC write. |
| 3368 | * A caller has to handle it. This is mainly due to the fact that we want to |
| 3369 | * avoid syncing under i_mutex. |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 3370 | * |
| 3371 | * Return: |
| 3372 | * * number of bytes written, even for truncated writes |
| 3373 | * * negative error code if no data has been written at all |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 3374 | */ |
| 3375 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
| 3376 | { |
| 3377 | struct file *file = iocb->ki_filp; |
| 3378 | struct address_space * mapping = file->f_mapping; |
| 3379 | struct inode *inode = mapping->host; |
| 3380 | ssize_t written = 0; |
| 3381 | ssize_t err; |
| 3382 | ssize_t status; |
| 3383 | |
| 3384 | /* We can write back this queue in page reclaim */ |
| 3385 | current->backing_dev_info = inode_to_bdi(inode); |
| 3386 | err = file_remove_privs(file); |
| 3387 | if (err) |
| 3388 | goto out; |
| 3389 | |
| 3390 | err = file_update_time(file); |
| 3391 | if (err) |
| 3392 | goto out; |
| 3393 | |
| 3394 | if (iocb->ki_flags & IOCB_DIRECT) { |
| 3395 | loff_t pos, endbyte; |
| 3396 | |
| 3397 | written = generic_file_direct_write(iocb, from); |
| 3398 | /* |
| 3399 | * If the write stopped short of completing, fall back to |
| 3400 | * buffered writes. Some filesystems do this for writes to |
| 3401 | * holes, for example. For DAX files, a buffered write will |
| 3402 | * not succeed (even if it did, DAX does not handle dirty |
| 3403 | * page-cache pages correctly). |
| 3404 | */ |
| 3405 | if (written < 0 || !iov_iter_count(from) || IS_DAX(inode)) |
| 3406 | goto out; |
| 3407 | |
| 3408 | status = generic_perform_write(file, from, pos = iocb->ki_pos); |
| 3409 | /* |
| 3410 | * If generic_perform_write() returned a synchronous error |
| 3411 | * then we want to return the number of bytes which were |
| 3412 | * direct-written, or the error code if that was zero. Note |
| 3413 | * that this differs from normal direct-io semantics, which |
| 3414 | * will return -EFOO even if some bytes were written. |
| 3415 | */ |
| 3416 | if (unlikely(status < 0)) { |
| 3417 | err = status; |
| 3418 | goto out; |
| 3419 | } |
| 3420 | /* |
| 3421 | * We need to ensure that the page cache pages are written to |
| 3422 | * disk and invalidated to preserve the expected O_DIRECT |
| 3423 | * semantics. |
| 3424 | */ |
| 3425 | endbyte = pos + status - 1; |
| 3426 | err = filemap_write_and_wait_range(mapping, pos, endbyte); |
| 3427 | if (err == 0) { |
| 3428 | iocb->ki_pos = endbyte + 1; |
| 3429 | written += status; |
| 3430 | invalidate_mapping_pages(mapping, |
| 3431 | pos >> PAGE_SHIFT, |
| 3432 | endbyte >> PAGE_SHIFT); |
| 3433 | } else { |
| 3434 | /* |
| 3435 | * We don't know how much we wrote, so just return |
| 3436 | * the number of bytes which were direct-written |
| 3437 | */ |
| 3438 | } |
| 3439 | } else { |
| 3440 | written = generic_perform_write(file, from, iocb->ki_pos); |
| 3441 | if (likely(written > 0)) |
| 3442 | iocb->ki_pos += written; |
| 3443 | } |
| 3444 | out: |
| 3445 | current->backing_dev_info = NULL; |
| 3446 | return written ? written : err; |
| 3447 | } |
| 3448 | EXPORT_SYMBOL(__generic_file_write_iter); |
| 3449 | |
| 3450 | /** |
| 3451 | * generic_file_write_iter - write data to a file |
| 3452 | * @iocb: IO state structure |
| 3453 | * @from: iov_iter with data to write |
| 3454 | * |
| 3455 | * This is a wrapper around __generic_file_write_iter() to be used by most |
| 3456 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
| 3457 | * and acquires i_mutex as needed. |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 3458 | * Return: |
| 3459 | * * negative error code if no data has been written at all of |
| 3460 | * vfs_fsync_range() failed for a synchronous write |
| 3461 | * * number of bytes written, even for truncated writes |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 3462 | */ |
| 3463 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
| 3464 | { |
| 3465 | struct file *file = iocb->ki_filp; |
| 3466 | struct inode *inode = file->f_mapping->host; |
| 3467 | ssize_t ret; |
| 3468 | |
| 3469 | inode_lock(inode); |
| 3470 | ret = generic_write_checks(iocb, from); |
| 3471 | if (ret > 0) |
| 3472 | ret = __generic_file_write_iter(iocb, from); |
| 3473 | inode_unlock(inode); |
| 3474 | |
| 3475 | if (ret > 0) |
| 3476 | ret = generic_write_sync(iocb, ret); |
| 3477 | return ret; |
| 3478 | } |
| 3479 | EXPORT_SYMBOL(generic_file_write_iter); |
| 3480 | |
| 3481 | /** |
| 3482 | * try_to_release_page() - release old fs-specific metadata on a page |
| 3483 | * |
| 3484 | * @page: the page which the kernel is trying to free |
| 3485 | * @gfp_mask: memory allocation flags (and I/O mode) |
| 3486 | * |
| 3487 | * The address_space is to try to release any data against the page |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 3488 | * (presumably at page->private). |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 3489 | * |
| 3490 | * This may also be called if PG_fscache is set on a page, indicating that the |
| 3491 | * page is known to the local caching routines. |
| 3492 | * |
| 3493 | * The @gfp_mask argument specifies whether I/O may be performed to release |
| 3494 | * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS). |
| 3495 | * |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 3496 | * Return: %1 if the release was successful, otherwise return zero. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 3497 | */ |
| 3498 | int try_to_release_page(struct page *page, gfp_t gfp_mask) |
| 3499 | { |
| 3500 | struct address_space * const mapping = page->mapping; |
| 3501 | |
| 3502 | BUG_ON(!PageLocked(page)); |
| 3503 | if (PageWriteback(page)) |
| 3504 | return 0; |
| 3505 | |
| 3506 | if (mapping && mapping->a_ops->releasepage) |
| 3507 | return mapping->a_ops->releasepage(page, gfp_mask); |
| 3508 | return try_to_free_buffers(page); |
| 3509 | } |
| 3510 | |
| 3511 | EXPORT_SYMBOL(try_to_release_page); |