]> git.ipfire.org Git - people/arne_f/kernel.git/blame - mm/filemap.c
ARM: 8839/1: kprobe: make patch_lock a raw_spinlock_t
[people/arne_f/kernel.git] / mm / filemap.c
CommitLineData
1da177e4
LT
1/*
2 * linux/mm/filemap.c
3 *
4 * Copyright (C) 1994-1999 Linus Torvalds
5 */
6
7/*
8 * This file handles the generic file mmap semantics used by
9 * most "normal" filesystems (but you don't /have/ to use this:
10 * the NFS filesystem used to do this differently, for example)
11 */
b95f1b31 12#include <linux/export.h>
1da177e4 13#include <linux/compiler.h>
f9fe48be 14#include <linux/dax.h>
1da177e4 15#include <linux/fs.h>
c22ce143 16#include <linux/uaccess.h>
c59ede7b 17#include <linux/capability.h>
1da177e4 18#include <linux/kernel_stat.h>
5a0e3ad6 19#include <linux/gfp.h>
1da177e4
LT
20#include <linux/mm.h>
21#include <linux/swap.h>
22#include <linux/mman.h>
23#include <linux/pagemap.h>
24#include <linux/file.h>
25#include <linux/uio.h>
26#include <linux/hash.h>
27#include <linux/writeback.h>
53253383 28#include <linux/backing-dev.h>
1da177e4
LT
29#include <linux/pagevec.h>
30#include <linux/blkdev.h>
31#include <linux/security.h>
44110fe3 32#include <linux/cpuset.h>
2f718ffc 33#include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
00501b53 34#include <linux/hugetlb.h>
8a9f3ccd 35#include <linux/memcontrol.h>
c515e1fd 36#include <linux/cleancache.h>
f1820361 37#include <linux/rmap.h>
0f8053a5
NP
38#include "internal.h"
39
fe0bfaaf
RJ
40#define CREATE_TRACE_POINTS
41#include <trace/events/filemap.h>
42
1da177e4 43/*
1da177e4
LT
44 * FIXME: remove all knowledge of the buffer layer from the core VM
45 */
148f948b 46#include <linux/buffer_head.h> /* for try_to_free_buffers */
1da177e4 47
1da177e4
LT
48#include <asm/mman.h>
49
50/*
51 * Shared mappings implemented 30.11.1994. It's not fully working yet,
52 * though.
53 *
54 * Shared mappings now work. 15.8.1995 Bruno.
55 *
56 * finished 'unifying' the page and buffer cache and SMP-threaded the
57 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
58 *
59 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
60 */
61
62/*
63 * Lock ordering:
64 *
c8c06efa 65 * ->i_mmap_rwsem (truncate_pagecache)
1da177e4 66 * ->private_lock (__free_pte->__set_page_dirty_buffers)
5d337b91
HD
67 * ->swap_lock (exclusive_swap_page, others)
68 * ->mapping->tree_lock
1da177e4 69 *
1b1dcc1b 70 * ->i_mutex
c8c06efa 71 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
1da177e4
LT
72 *
73 * ->mmap_sem
c8c06efa 74 * ->i_mmap_rwsem
b8072f09 75 * ->page_table_lock or pte_lock (various, mainly in memory.c)
1da177e4
LT
76 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
77 *
78 * ->mmap_sem
79 * ->lock_page (access_process_vm)
80 *
ccad2365 81 * ->i_mutex (generic_perform_write)
82591e6e 82 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
1da177e4 83 *
f758eeab 84 * bdi->wb.list_lock
a66979ab 85 * sb_lock (fs/fs-writeback.c)
1da177e4
LT
86 * ->mapping->tree_lock (__sync_single_inode)
87 *
c8c06efa 88 * ->i_mmap_rwsem
1da177e4
LT
89 * ->anon_vma.lock (vma_adjust)
90 *
91 * ->anon_vma.lock
b8072f09 92 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
1da177e4 93 *
b8072f09 94 * ->page_table_lock or pte_lock
5d337b91 95 * ->swap_lock (try_to_unmap_one)
1da177e4
LT
96 * ->private_lock (try_to_unmap_one)
97 * ->tree_lock (try_to_unmap_one)
a52633d8
MG
98 * ->zone_lru_lock(zone) (follow_page->mark_page_accessed)
99 * ->zone_lru_lock(zone) (check_pte_range->isolate_lru_page)
1da177e4
LT
100 * ->private_lock (page_remove_rmap->set_page_dirty)
101 * ->tree_lock (page_remove_rmap->set_page_dirty)
f758eeab 102 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
250df6ed 103 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
81f8c3a4 104 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
f758eeab 105 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
250df6ed 106 * ->inode->i_lock (zap_pte_range->set_page_dirty)
1da177e4
LT
107 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
108 *
c8c06efa 109 * ->i_mmap_rwsem
9a3c531d 110 * ->tasklist_lock (memory_failure, collect_procs_ao)
1da177e4
LT
111 */
112
22f2ac51
JW
113static int page_cache_tree_insert(struct address_space *mapping,
114 struct page *page, void **shadowp)
115{
116 struct radix_tree_node *node;
117 void **slot;
118 int error;
119
120 error = __radix_tree_create(&mapping->page_tree, page->index, 0,
121 &node, &slot);
122 if (error)
123 return error;
124 if (*slot) {
125 void *p;
126
127 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
128 if (!radix_tree_exceptional_entry(p))
129 return -EEXIST;
130
131 mapping->nrexceptional--;
132 if (!dax_mapping(mapping)) {
133 if (shadowp)
134 *shadowp = p;
135 if (node)
136 workingset_node_shadows_dec(node);
137 } else {
138 /* DAX can replace empty locked entry with a hole */
139 WARN_ON_ONCE(p !=
140 (void *)(RADIX_TREE_EXCEPTIONAL_ENTRY |
141 RADIX_DAX_ENTRY_LOCK));
142 /* DAX accounts exceptional entries as normal pages */
143 if (node)
144 workingset_node_pages_dec(node);
145 /* Wakeup waiters for exceptional entry lock */
146 dax_wake_mapping_entry_waiter(mapping, page->index,
87fa6f37 147 true);
22f2ac51
JW
148 }
149 }
150 radix_tree_replace_slot(slot, page);
151 mapping->nrpages++;
152 if (node) {
153 workingset_node_pages_inc(node);
154 /*
155 * Don't track node that contains actual pages.
156 *
157 * Avoid acquiring the list_lru lock if already
158 * untracked. The list_empty() test is safe as
159 * node->private_list is protected by
160 * mapping->tree_lock.
161 */
162 if (!list_empty(&node->private_list))
163 list_lru_del(&workingset_shadow_nodes,
164 &node->private_list);
165 }
166 return 0;
167}
168
91b0abe3
JW
169static void page_cache_tree_delete(struct address_space *mapping,
170 struct page *page, void *shadow)
171{
83929372 172 int i, nr = PageHuge(page) ? 1 : hpage_nr_pages(page);
91b0abe3 173
83929372
KS
174 VM_BUG_ON_PAGE(!PageLocked(page), page);
175 VM_BUG_ON_PAGE(PageTail(page), page);
176 VM_BUG_ON_PAGE(nr != 1 && shadow, page);
449dd698 177
83929372 178 for (i = 0; i < nr; i++) {
d3798ae8
JW
179 struct radix_tree_node *node;
180 void **slot;
181
182 __radix_tree_lookup(&mapping->page_tree, page->index + i,
183 &node, &slot);
184
185 radix_tree_clear_tags(&mapping->page_tree, node, slot);
186
83929372
KS
187 if (!node) {
188 VM_BUG_ON_PAGE(nr != 1, page);
d3798ae8
JW
189 /*
190 * We need a node to properly account shadow
191 * entries. Don't plant any without. XXX
192 */
193 shadow = NULL;
83929372 194 }
449dd698 195
d3798ae8
JW
196 radix_tree_replace_slot(slot, shadow);
197
198 if (!node)
199 break;
200
83929372
KS
201 workingset_node_pages_dec(node);
202 if (shadow)
203 workingset_node_shadows_inc(node);
204 else
205 if (__radix_tree_delete_node(&mapping->page_tree, node))
206 continue;
207
208 /*
209 * Track node that only contains shadow entries. DAX mappings
210 * contain no shadow entries and may contain other exceptional
211 * entries so skip those.
212 *
213 * Avoid acquiring the list_lru lock if already tracked.
214 * The list_empty() test is safe as node->private_list is
215 * protected by mapping->tree_lock.
216 */
217 if (!dax_mapping(mapping) && !workingset_node_pages(node) &&
218 list_empty(&node->private_list)) {
219 node->private_data = mapping;
220 list_lru_add(&workingset_shadow_nodes,
221 &node->private_list);
222 }
449dd698 223 }
d3798ae8
JW
224
225 if (shadow) {
226 mapping->nrexceptional += nr;
227 /*
228 * Make sure the nrexceptional update is committed before
229 * the nrpages update so that final truncate racing
230 * with reclaim does not see both counters 0 at the
231 * same time and miss a shadow entry.
232 */
233 smp_wmb();
234 }
235 mapping->nrpages -= nr;
91b0abe3
JW
236}
237
1da177e4 238/*
e64a782f 239 * Delete a page from the page cache and free it. Caller has to make
1da177e4 240 * sure the page is locked and that nobody else uses it - or that usage
fdf1cdb9 241 * is safe. The caller must hold the mapping's tree_lock.
1da177e4 242 */
62cccb8c 243void __delete_from_page_cache(struct page *page, void *shadow)
1da177e4
LT
244{
245 struct address_space *mapping = page->mapping;
83929372 246 int nr = hpage_nr_pages(page);
1da177e4 247
fe0bfaaf 248 trace_mm_filemap_delete_from_page_cache(page);
c515e1fd
DM
249 /*
250 * if we're uptodate, flush out into the cleancache, otherwise
251 * invalidate any existing cleancache entries. We can't leave
252 * stale data around in the cleancache once our page is gone
253 */
254 if (PageUptodate(page) && PageMappedToDisk(page))
255 cleancache_put_page(page);
256 else
3167760f 257 cleancache_invalidate_page(mapping, page);
c515e1fd 258
83929372 259 VM_BUG_ON_PAGE(PageTail(page), page);
06b241f3
HD
260 VM_BUG_ON_PAGE(page_mapped(page), page);
261 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
262 int mapcount;
263
264 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
265 current->comm, page_to_pfn(page));
266 dump_page(page, "still mapped when deleted");
267 dump_stack();
268 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
269
270 mapcount = page_mapcount(page);
271 if (mapping_exiting(mapping) &&
272 page_count(page) >= mapcount + 2) {
273 /*
274 * All vmas have already been torn down, so it's
275 * a good bet that actually the page is unmapped,
276 * and we'd prefer not to leak it: if we're wrong,
277 * some other bad page check should catch it later.
278 */
279 page_mapcount_reset(page);
6d061f9f 280 page_ref_sub(page, mapcount);
06b241f3
HD
281 }
282 }
283
91b0abe3
JW
284 page_cache_tree_delete(mapping, page, shadow);
285
1da177e4 286 page->mapping = NULL;
b85e0eff 287 /* Leave page->index set: truncation lookup relies upon it */
91b0abe3 288
4165b9b4
MH
289 /* hugetlb pages do not participate in page cache accounting. */
290 if (!PageHuge(page))
11fb9989 291 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
800d8c63 292 if (PageSwapBacked(page)) {
11fb9989 293 __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr);
800d8c63 294 if (PageTransHuge(page))
11fb9989 295 __dec_node_page_state(page, NR_SHMEM_THPS);
800d8c63
KS
296 } else {
297 VM_BUG_ON_PAGE(PageTransHuge(page) && !PageHuge(page), page);
298 }
3a692790
LT
299
300 /*
b9ea2515
KK
301 * At this point page must be either written or cleaned by truncate.
302 * Dirty page here signals a bug and loss of unwritten data.
3a692790 303 *
b9ea2515
KK
304 * This fixes dirty accounting after removing the page entirely but
305 * leaves PageDirty set: it has no effect for truncated page and
306 * anyway will be cleared before returning page into buddy allocator.
3a692790 307 */
b9ea2515 308 if (WARN_ON_ONCE(PageDirty(page)))
62cccb8c 309 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
1da177e4
LT
310}
311
702cfbf9
MK
312/**
313 * delete_from_page_cache - delete page from page cache
314 * @page: the page which the kernel is trying to remove from page cache
315 *
316 * This must be called only on pages that have been verified to be in the page
317 * cache and locked. It will never put the page into the free list, the caller
318 * has a reference on the page.
319 */
320void delete_from_page_cache(struct page *page)
1da177e4 321{
83929372 322 struct address_space *mapping = page_mapping(page);
c4843a75 323 unsigned long flags;
6072d13c 324 void (*freepage)(struct page *);
1da177e4 325
cd7619d6 326 BUG_ON(!PageLocked(page));
1da177e4 327
6072d13c 328 freepage = mapping->a_ops->freepage;
c4843a75 329
c4843a75 330 spin_lock_irqsave(&mapping->tree_lock, flags);
62cccb8c 331 __delete_from_page_cache(page, NULL);
c4843a75 332 spin_unlock_irqrestore(&mapping->tree_lock, flags);
6072d13c
LT
333
334 if (freepage)
335 freepage(page);
83929372
KS
336
337 if (PageTransHuge(page) && !PageHuge(page)) {
338 page_ref_sub(page, HPAGE_PMD_NR);
339 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
340 } else {
341 put_page(page);
342 }
97cecb5a
MK
343}
344EXPORT_SYMBOL(delete_from_page_cache);
345
d72d9e2a 346int filemap_check_errors(struct address_space *mapping)
865ffef3
DM
347{
348 int ret = 0;
349 /* Check for outstanding write errors */
7fcbbaf1
JA
350 if (test_bit(AS_ENOSPC, &mapping->flags) &&
351 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
865ffef3 352 ret = -ENOSPC;
7fcbbaf1
JA
353 if (test_bit(AS_EIO, &mapping->flags) &&
354 test_and_clear_bit(AS_EIO, &mapping->flags))
865ffef3
DM
355 ret = -EIO;
356 return ret;
357}
d72d9e2a 358EXPORT_SYMBOL(filemap_check_errors);
865ffef3 359
1da177e4 360/**
485bb99b 361 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
67be2dd1
MW
362 * @mapping: address space structure to write
363 * @start: offset in bytes where the range starts
469eb4d0 364 * @end: offset in bytes where the range ends (inclusive)
67be2dd1 365 * @sync_mode: enable synchronous operation
1da177e4 366 *
485bb99b
RD
367 * Start writeback against all of a mapping's dirty pages that lie
368 * within the byte offsets <start, end> inclusive.
369 *
1da177e4 370 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
485bb99b 371 * opposed to a regular memory cleansing writeback. The difference between
1da177e4
LT
372 * these two operations is that if a dirty page/buffer is encountered, it must
373 * be waited upon, and not just skipped over.
374 */
ebcf28e1
AM
375int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
376 loff_t end, int sync_mode)
1da177e4
LT
377{
378 int ret;
379 struct writeback_control wbc = {
380 .sync_mode = sync_mode,
05fe478d 381 .nr_to_write = LONG_MAX,
111ebb6e
OH
382 .range_start = start,
383 .range_end = end,
1da177e4
LT
384 };
385
386 if (!mapping_cap_writeback_dirty(mapping))
387 return 0;
388
b16b1deb 389 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
1da177e4 390 ret = do_writepages(mapping, &wbc);
b16b1deb 391 wbc_detach_inode(&wbc);
1da177e4
LT
392 return ret;
393}
394
395static inline int __filemap_fdatawrite(struct address_space *mapping,
396 int sync_mode)
397{
111ebb6e 398 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
1da177e4
LT
399}
400
401int filemap_fdatawrite(struct address_space *mapping)
402{
403 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
404}
405EXPORT_SYMBOL(filemap_fdatawrite);
406
f4c0a0fd 407int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
ebcf28e1 408 loff_t end)
1da177e4
LT
409{
410 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
411}
f4c0a0fd 412EXPORT_SYMBOL(filemap_fdatawrite_range);
1da177e4 413
485bb99b
RD
414/**
415 * filemap_flush - mostly a non-blocking flush
416 * @mapping: target address_space
417 *
1da177e4
LT
418 * This is a mostly non-blocking flush. Not suitable for data-integrity
419 * purposes - I/O may not be started against all dirty pages.
420 */
421int filemap_flush(struct address_space *mapping)
422{
423 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
424}
425EXPORT_SYMBOL(filemap_flush);
426
aa750fd7
JN
427static int __filemap_fdatawait_range(struct address_space *mapping,
428 loff_t start_byte, loff_t end_byte)
1da177e4 429{
09cbfeaf
KS
430 pgoff_t index = start_byte >> PAGE_SHIFT;
431 pgoff_t end = end_byte >> PAGE_SHIFT;
1da177e4
LT
432 struct pagevec pvec;
433 int nr_pages;
aa750fd7 434 int ret = 0;
1da177e4 435
94004ed7 436 if (end_byte < start_byte)
865ffef3 437 goto out;
1da177e4
LT
438
439 pagevec_init(&pvec, 0);
1da177e4
LT
440 while ((index <= end) &&
441 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
442 PAGECACHE_TAG_WRITEBACK,
443 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
444 unsigned i;
445
446 for (i = 0; i < nr_pages; i++) {
447 struct page *page = pvec.pages[i];
448
449 /* until radix tree lookup accepts end_index */
450 if (page->index > end)
451 continue;
452
453 wait_on_page_writeback(page);
212260aa 454 if (TestClearPageError(page))
1da177e4
LT
455 ret = -EIO;
456 }
457 pagevec_release(&pvec);
458 cond_resched();
459 }
865ffef3 460out:
aa750fd7
JN
461 return ret;
462}
463
464/**
465 * filemap_fdatawait_range - wait for writeback to complete
466 * @mapping: address space structure to wait for
467 * @start_byte: offset in bytes where the range starts
468 * @end_byte: offset in bytes where the range ends (inclusive)
469 *
470 * Walk the list of under-writeback pages of the given address space
471 * in the given range and wait for all of them. Check error status of
472 * the address space and return it.
473 *
474 * Since the error status of the address space is cleared by this function,
475 * callers are responsible for checking the return value and handling and/or
476 * reporting the error.
477 */
478int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
479 loff_t end_byte)
480{
481 int ret, ret2;
482
483 ret = __filemap_fdatawait_range(mapping, start_byte, end_byte);
865ffef3
DM
484 ret2 = filemap_check_errors(mapping);
485 if (!ret)
486 ret = ret2;
1da177e4
LT
487
488 return ret;
489}
d3bccb6f
JK
490EXPORT_SYMBOL(filemap_fdatawait_range);
491
aa750fd7
JN
492/**
493 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
494 * @mapping: address space structure to wait for
495 *
496 * Walk the list of under-writeback pages of the given address space
497 * and wait for all of them. Unlike filemap_fdatawait(), this function
498 * does not clear error status of the address space.
499 *
500 * Use this function if callers don't handle errors themselves. Expected
501 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
502 * fsfreeze(8)
503 */
504void filemap_fdatawait_keep_errors(struct address_space *mapping)
505{
506 loff_t i_size = i_size_read(mapping->host);
507
508 if (i_size == 0)
509 return;
510
511 __filemap_fdatawait_range(mapping, 0, i_size - 1);
512}
513
1da177e4 514/**
485bb99b 515 * filemap_fdatawait - wait for all under-writeback pages to complete
1da177e4 516 * @mapping: address space structure to wait for
485bb99b
RD
517 *
518 * Walk the list of under-writeback pages of the given address space
aa750fd7
JN
519 * and wait for all of them. Check error status of the address space
520 * and return it.
521 *
522 * Since the error status of the address space is cleared by this function,
523 * callers are responsible for checking the return value and handling and/or
524 * reporting the error.
1da177e4
LT
525 */
526int filemap_fdatawait(struct address_space *mapping)
527{
528 loff_t i_size = i_size_read(mapping->host);
529
530 if (i_size == 0)
531 return 0;
532
94004ed7 533 return filemap_fdatawait_range(mapping, 0, i_size - 1);
1da177e4
LT
534}
535EXPORT_SYMBOL(filemap_fdatawait);
536
537int filemap_write_and_wait(struct address_space *mapping)
538{
28fd1298 539 int err = 0;
1da177e4 540
7f6d5b52
RZ
541 if ((!dax_mapping(mapping) && mapping->nrpages) ||
542 (dax_mapping(mapping) && mapping->nrexceptional)) {
28fd1298
OH
543 err = filemap_fdatawrite(mapping);
544 /*
545 * Even if the above returned error, the pages may be
546 * written partially (e.g. -ENOSPC), so we wait for it.
547 * But the -EIO is special case, it may indicate the worst
548 * thing (e.g. bug) happened, so we avoid waiting for it.
549 */
550 if (err != -EIO) {
551 int err2 = filemap_fdatawait(mapping);
552 if (!err)
553 err = err2;
554 }
865ffef3
DM
555 } else {
556 err = filemap_check_errors(mapping);
1da177e4 557 }
28fd1298 558 return err;
1da177e4 559}
28fd1298 560EXPORT_SYMBOL(filemap_write_and_wait);
1da177e4 561
485bb99b
RD
562/**
563 * filemap_write_and_wait_range - write out & wait on a file range
564 * @mapping: the address_space for the pages
565 * @lstart: offset in bytes where the range starts
566 * @lend: offset in bytes where the range ends (inclusive)
567 *
469eb4d0
AM
568 * Write out and wait upon file offsets lstart->lend, inclusive.
569 *
570 * Note that `lend' is inclusive (describes the last byte to be written) so
571 * that this function can be used to write to the very end-of-file (end = -1).
572 */
1da177e4
LT
573int filemap_write_and_wait_range(struct address_space *mapping,
574 loff_t lstart, loff_t lend)
575{
28fd1298 576 int err = 0;
1da177e4 577
7f6d5b52
RZ
578 if ((!dax_mapping(mapping) && mapping->nrpages) ||
579 (dax_mapping(mapping) && mapping->nrexceptional)) {
28fd1298
OH
580 err = __filemap_fdatawrite_range(mapping, lstart, lend,
581 WB_SYNC_ALL);
582 /* See comment of filemap_write_and_wait() */
583 if (err != -EIO) {
94004ed7
CH
584 int err2 = filemap_fdatawait_range(mapping,
585 lstart, lend);
28fd1298
OH
586 if (!err)
587 err = err2;
588 }
865ffef3
DM
589 } else {
590 err = filemap_check_errors(mapping);
1da177e4 591 }
28fd1298 592 return err;
1da177e4 593}
f6995585 594EXPORT_SYMBOL(filemap_write_and_wait_range);
1da177e4 595
ef6a3c63
MS
596/**
597 * replace_page_cache_page - replace a pagecache page with a new one
598 * @old: page to be replaced
599 * @new: page to replace with
600 * @gfp_mask: allocation mode
601 *
602 * This function replaces a page in the pagecache with a new one. On
603 * success it acquires the pagecache reference for the new page and
604 * drops it for the old page. Both the old and new pages must be
605 * locked. This function does not add the new page to the LRU, the
606 * caller must do that.
607 *
608 * The remove + add is atomic. The only way this function can fail is
609 * memory allocation failure.
610 */
611int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
612{
613 int error;
ef6a3c63 614
309381fe
SL
615 VM_BUG_ON_PAGE(!PageLocked(old), old);
616 VM_BUG_ON_PAGE(!PageLocked(new), new);
617 VM_BUG_ON_PAGE(new->mapping, new);
ef6a3c63 618
f4c86fa0 619 error = radix_tree_preload(gfp_mask & GFP_RECLAIM_MASK);
ef6a3c63
MS
620 if (!error) {
621 struct address_space *mapping = old->mapping;
622 void (*freepage)(struct page *);
c4843a75 623 unsigned long flags;
ef6a3c63
MS
624
625 pgoff_t offset = old->index;
626 freepage = mapping->a_ops->freepage;
627
09cbfeaf 628 get_page(new);
ef6a3c63
MS
629 new->mapping = mapping;
630 new->index = offset;
631
c4843a75 632 spin_lock_irqsave(&mapping->tree_lock, flags);
62cccb8c 633 __delete_from_page_cache(old, NULL);
22f2ac51 634 error = page_cache_tree_insert(mapping, new, NULL);
ef6a3c63 635 BUG_ON(error);
4165b9b4
MH
636
637 /*
638 * hugetlb pages do not participate in page cache accounting.
639 */
640 if (!PageHuge(new))
11fb9989 641 __inc_node_page_state(new, NR_FILE_PAGES);
ef6a3c63 642 if (PageSwapBacked(new))
11fb9989 643 __inc_node_page_state(new, NR_SHMEM);
c4843a75 644 spin_unlock_irqrestore(&mapping->tree_lock, flags);
6a93ca8f 645 mem_cgroup_migrate(old, new);
ef6a3c63
MS
646 radix_tree_preload_end();
647 if (freepage)
648 freepage(old);
09cbfeaf 649 put_page(old);
ef6a3c63
MS
650 }
651
652 return error;
653}
654EXPORT_SYMBOL_GPL(replace_page_cache_page);
655
a528910e
JW
656static int __add_to_page_cache_locked(struct page *page,
657 struct address_space *mapping,
658 pgoff_t offset, gfp_t gfp_mask,
659 void **shadowp)
1da177e4 660{
00501b53
JW
661 int huge = PageHuge(page);
662 struct mem_cgroup *memcg;
e286781d
NP
663 int error;
664
309381fe
SL
665 VM_BUG_ON_PAGE(!PageLocked(page), page);
666 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
e286781d 667
00501b53
JW
668 if (!huge) {
669 error = mem_cgroup_try_charge(page, current->mm,
f627c2f5 670 gfp_mask, &memcg, false);
00501b53
JW
671 if (error)
672 return error;
673 }
1da177e4 674
f4c86fa0 675 error = radix_tree_maybe_preload(gfp_mask & GFP_RECLAIM_MASK);
66a0c8ee 676 if (error) {
00501b53 677 if (!huge)
f627c2f5 678 mem_cgroup_cancel_charge(page, memcg, false);
66a0c8ee
KS
679 return error;
680 }
681
09cbfeaf 682 get_page(page);
66a0c8ee
KS
683 page->mapping = mapping;
684 page->index = offset;
685
686 spin_lock_irq(&mapping->tree_lock);
a528910e 687 error = page_cache_tree_insert(mapping, page, shadowp);
66a0c8ee
KS
688 radix_tree_preload_end();
689 if (unlikely(error))
690 goto err_insert;
4165b9b4
MH
691
692 /* hugetlb pages do not participate in page cache accounting. */
693 if (!huge)
11fb9989 694 __inc_node_page_state(page, NR_FILE_PAGES);
66a0c8ee 695 spin_unlock_irq(&mapping->tree_lock);
00501b53 696 if (!huge)
f627c2f5 697 mem_cgroup_commit_charge(page, memcg, false, false);
66a0c8ee
KS
698 trace_mm_filemap_add_to_page_cache(page);
699 return 0;
700err_insert:
701 page->mapping = NULL;
702 /* Leave page->index set: truncation relies upon it */
703 spin_unlock_irq(&mapping->tree_lock);
00501b53 704 if (!huge)
f627c2f5 705 mem_cgroup_cancel_charge(page, memcg, false);
09cbfeaf 706 put_page(page);
1da177e4
LT
707 return error;
708}
a528910e
JW
709
710/**
711 * add_to_page_cache_locked - add a locked page to the pagecache
712 * @page: page to add
713 * @mapping: the page's address_space
714 * @offset: page index
715 * @gfp_mask: page allocation mode
716 *
717 * This function is used to add a page to the pagecache. It must be locked.
718 * This function does not add the page to the LRU. The caller must do that.
719 */
720int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
721 pgoff_t offset, gfp_t gfp_mask)
722{
723 return __add_to_page_cache_locked(page, mapping, offset,
724 gfp_mask, NULL);
725}
e286781d 726EXPORT_SYMBOL(add_to_page_cache_locked);
1da177e4
LT
727
728int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
6daa0e28 729 pgoff_t offset, gfp_t gfp_mask)
1da177e4 730{
a528910e 731 void *shadow = NULL;
4f98a2fe
RR
732 int ret;
733
48c935ad 734 __SetPageLocked(page);
a528910e
JW
735 ret = __add_to_page_cache_locked(page, mapping, offset,
736 gfp_mask, &shadow);
737 if (unlikely(ret))
48c935ad 738 __ClearPageLocked(page);
a528910e
JW
739 else {
740 /*
741 * The page might have been evicted from cache only
742 * recently, in which case it should be activated like
743 * any other repeatedly accessed page.
f0281a00
RR
744 * The exception is pages getting rewritten; evicting other
745 * data from the working set, only to cache data that will
746 * get overwritten with something else, is a waste of memory.
a528910e 747 */
f0281a00
RR
748 if (!(gfp_mask & __GFP_WRITE) &&
749 shadow && workingset_refault(shadow)) {
a528910e
JW
750 SetPageActive(page);
751 workingset_activation(page);
752 } else
753 ClearPageActive(page);
754 lru_cache_add(page);
755 }
1da177e4
LT
756 return ret;
757}
18bc0bbd 758EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
1da177e4 759
44110fe3 760#ifdef CONFIG_NUMA
2ae88149 761struct page *__page_cache_alloc(gfp_t gfp)
44110fe3 762{
c0ff7453
MX
763 int n;
764 struct page *page;
765
44110fe3 766 if (cpuset_do_page_mem_spread()) {
cc9a6c87
MG
767 unsigned int cpuset_mems_cookie;
768 do {
d26914d1 769 cpuset_mems_cookie = read_mems_allowed_begin();
cc9a6c87 770 n = cpuset_mem_spread_node();
96db800f 771 page = __alloc_pages_node(n, gfp, 0);
d26914d1 772 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
cc9a6c87 773
c0ff7453 774 return page;
44110fe3 775 }
2ae88149 776 return alloc_pages(gfp, 0);
44110fe3 777}
2ae88149 778EXPORT_SYMBOL(__page_cache_alloc);
44110fe3
PJ
779#endif
780
1da177e4
LT
781/*
782 * In order to wait for pages to become available there must be
783 * waitqueues associated with pages. By using a hash table of
784 * waitqueues where the bucket discipline is to maintain all
785 * waiters on the same queue and wake all when any of the pages
786 * become available, and for the woken contexts to check to be
787 * sure the appropriate page became available, this saves space
788 * at a cost of "thundering herd" phenomena during rare hash
789 * collisions.
790 */
a4796e37 791wait_queue_head_t *page_waitqueue(struct page *page)
1da177e4 792{
9dcb8b68 793 return bit_waitqueue(page, 0);
1da177e4 794}
a4796e37 795EXPORT_SYMBOL(page_waitqueue);
1da177e4 796
920c7a5d 797void wait_on_page_bit(struct page *page, int bit_nr)
1da177e4
LT
798{
799 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
800
801 if (test_bit(bit_nr, &page->flags))
74316201 802 __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io,
1da177e4
LT
803 TASK_UNINTERRUPTIBLE);
804}
805EXPORT_SYMBOL(wait_on_page_bit);
806
f62e00cc
KM
807int wait_on_page_bit_killable(struct page *page, int bit_nr)
808{
809 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
810
811 if (!test_bit(bit_nr, &page->flags))
812 return 0;
813
814 return __wait_on_bit(page_waitqueue(page), &wait,
74316201 815 bit_wait_io, TASK_KILLABLE);
f62e00cc
KM
816}
817
cbbce822
N
818int wait_on_page_bit_killable_timeout(struct page *page,
819 int bit_nr, unsigned long timeout)
820{
821 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
822
823 wait.key.timeout = jiffies + timeout;
824 if (!test_bit(bit_nr, &page->flags))
825 return 0;
826 return __wait_on_bit(page_waitqueue(page), &wait,
827 bit_wait_io_timeout, TASK_KILLABLE);
828}
829EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout);
830
385e1ca5
DH
831/**
832 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
697f619f
RD
833 * @page: Page defining the wait queue of interest
834 * @waiter: Waiter to add to the queue
385e1ca5
DH
835 *
836 * Add an arbitrary @waiter to the wait queue for the nominated @page.
837 */
838void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
839{
840 wait_queue_head_t *q = page_waitqueue(page);
841 unsigned long flags;
842
843 spin_lock_irqsave(&q->lock, flags);
844 __add_wait_queue(q, waiter);
845 spin_unlock_irqrestore(&q->lock, flags);
846}
847EXPORT_SYMBOL_GPL(add_page_wait_queue);
848
1da177e4 849/**
485bb99b 850 * unlock_page - unlock a locked page
1da177e4
LT
851 * @page: the page
852 *
853 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
854 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
da3dae54 855 * mechanism between PageLocked pages and PageWriteback pages is shared.
1da177e4
LT
856 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
857 *
8413ac9d
NP
858 * The mb is necessary to enforce ordering between the clear_bit and the read
859 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
1da177e4 860 */
920c7a5d 861void unlock_page(struct page *page)
1da177e4 862{
48c935ad 863 page = compound_head(page);
309381fe 864 VM_BUG_ON_PAGE(!PageLocked(page), page);
8413ac9d 865 clear_bit_unlock(PG_locked, &page->flags);
4e857c58 866 smp_mb__after_atomic();
1da177e4
LT
867 wake_up_page(page, PG_locked);
868}
869EXPORT_SYMBOL(unlock_page);
870
485bb99b
RD
871/**
872 * end_page_writeback - end writeback against a page
873 * @page: the page
1da177e4
LT
874 */
875void end_page_writeback(struct page *page)
876{
888cf2db
MG
877 /*
878 * TestClearPageReclaim could be used here but it is an atomic
879 * operation and overkill in this particular case. Failing to
880 * shuffle a page marked for immediate reclaim is too mild to
881 * justify taking an atomic operation penalty at the end of
882 * ever page writeback.
883 */
884 if (PageReclaim(page)) {
885 ClearPageReclaim(page);
ac6aadb2 886 rotate_reclaimable_page(page);
888cf2db 887 }
ac6aadb2
MS
888
889 if (!test_clear_page_writeback(page))
890 BUG();
891
4e857c58 892 smp_mb__after_atomic();
1da177e4
LT
893 wake_up_page(page, PG_writeback);
894}
895EXPORT_SYMBOL(end_page_writeback);
896
57d99845
MW
897/*
898 * After completing I/O on a page, call this routine to update the page
899 * flags appropriately
900 */
c11f0c0b 901void page_endio(struct page *page, bool is_write, int err)
57d99845 902{
c11f0c0b 903 if (!is_write) {
57d99845
MW
904 if (!err) {
905 SetPageUptodate(page);
906 } else {
907 ClearPageUptodate(page);
908 SetPageError(page);
909 }
910 unlock_page(page);
abf54548 911 } else {
57d99845 912 if (err) {
2c290eed
MK
913 struct address_space *mapping;
914
57d99845 915 SetPageError(page);
2c290eed
MK
916 mapping = page_mapping(page);
917 if (mapping)
918 mapping_set_error(mapping, err);
57d99845
MW
919 }
920 end_page_writeback(page);
921 }
922}
923EXPORT_SYMBOL_GPL(page_endio);
924
485bb99b
RD
925/**
926 * __lock_page - get a lock on the page, assuming we need to sleep to get it
927 * @page: the page to lock
1da177e4 928 */
920c7a5d 929void __lock_page(struct page *page)
1da177e4 930{
48c935ad
KS
931 struct page *page_head = compound_head(page);
932 DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked);
1da177e4 933
48c935ad 934 __wait_on_bit_lock(page_waitqueue(page_head), &wait, bit_wait_io,
1da177e4
LT
935 TASK_UNINTERRUPTIBLE);
936}
937EXPORT_SYMBOL(__lock_page);
938
b5606c2d 939int __lock_page_killable(struct page *page)
2687a356 940{
48c935ad
KS
941 struct page *page_head = compound_head(page);
942 DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked);
2687a356 943
48c935ad 944 return __wait_on_bit_lock(page_waitqueue(page_head), &wait,
74316201 945 bit_wait_io, TASK_KILLABLE);
2687a356 946}
18bc0bbd 947EXPORT_SYMBOL_GPL(__lock_page_killable);
2687a356 948
9a95f3cf
PC
949/*
950 * Return values:
951 * 1 - page is locked; mmap_sem is still held.
952 * 0 - page is not locked.
953 * mmap_sem has been released (up_read()), unless flags had both
954 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
955 * which case mmap_sem is still held.
956 *
957 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
958 * with the page locked and the mmap_sem unperturbed.
959 */
d065bd81
ML
960int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
961 unsigned int flags)
962{
37b23e05
KM
963 if (flags & FAULT_FLAG_ALLOW_RETRY) {
964 /*
965 * CAUTION! In this case, mmap_sem is not released
966 * even though return 0.
967 */
968 if (flags & FAULT_FLAG_RETRY_NOWAIT)
969 return 0;
970
971 up_read(&mm->mmap_sem);
972 if (flags & FAULT_FLAG_KILLABLE)
973 wait_on_page_locked_killable(page);
974 else
318b275f 975 wait_on_page_locked(page);
d065bd81 976 return 0;
37b23e05
KM
977 } else {
978 if (flags & FAULT_FLAG_KILLABLE) {
979 int ret;
980
981 ret = __lock_page_killable(page);
982 if (ret) {
983 up_read(&mm->mmap_sem);
984 return 0;
985 }
986 } else
987 __lock_page(page);
988 return 1;
d065bd81
ML
989 }
990}
991
e7b563bb
JW
992/**
993 * page_cache_next_hole - find the next hole (not-present entry)
994 * @mapping: mapping
995 * @index: index
996 * @max_scan: maximum range to search
997 *
998 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
999 * lowest indexed hole.
1000 *
1001 * Returns: the index of the hole if found, otherwise returns an index
1002 * outside of the set specified (in which case 'return - index >=
1003 * max_scan' will be true). In rare cases of index wrap-around, 0 will
1004 * be returned.
1005 *
1006 * page_cache_next_hole may be called under rcu_read_lock. However,
1007 * like radix_tree_gang_lookup, this will not atomically search a
1008 * snapshot of the tree at a single point in time. For example, if a
1009 * hole is created at index 5, then subsequently a hole is created at
1010 * index 10, page_cache_next_hole covering both indexes may return 10
1011 * if called under rcu_read_lock.
1012 */
1013pgoff_t page_cache_next_hole(struct address_space *mapping,
1014 pgoff_t index, unsigned long max_scan)
1015{
1016 unsigned long i;
1017
1018 for (i = 0; i < max_scan; i++) {
0cd6144a
JW
1019 struct page *page;
1020
1021 page = radix_tree_lookup(&mapping->page_tree, index);
1022 if (!page || radix_tree_exceptional_entry(page))
e7b563bb
JW
1023 break;
1024 index++;
1025 if (index == 0)
1026 break;
1027 }
1028
1029 return index;
1030}
1031EXPORT_SYMBOL(page_cache_next_hole);
1032
1033/**
1034 * page_cache_prev_hole - find the prev hole (not-present entry)
1035 * @mapping: mapping
1036 * @index: index
1037 * @max_scan: maximum range to search
1038 *
1039 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1040 * the first hole.
1041 *
1042 * Returns: the index of the hole if found, otherwise returns an index
1043 * outside of the set specified (in which case 'index - return >=
1044 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1045 * will be returned.
1046 *
1047 * page_cache_prev_hole may be called under rcu_read_lock. However,
1048 * like radix_tree_gang_lookup, this will not atomically search a
1049 * snapshot of the tree at a single point in time. For example, if a
1050 * hole is created at index 10, then subsequently a hole is created at
1051 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1052 * called under rcu_read_lock.
1053 */
1054pgoff_t page_cache_prev_hole(struct address_space *mapping,
1055 pgoff_t index, unsigned long max_scan)
1056{
1057 unsigned long i;
1058
1059 for (i = 0; i < max_scan; i++) {
0cd6144a
JW
1060 struct page *page;
1061
1062 page = radix_tree_lookup(&mapping->page_tree, index);
1063 if (!page || radix_tree_exceptional_entry(page))
e7b563bb
JW
1064 break;
1065 index--;
1066 if (index == ULONG_MAX)
1067 break;
1068 }
1069
1070 return index;
1071}
1072EXPORT_SYMBOL(page_cache_prev_hole);
1073
485bb99b 1074/**
0cd6144a 1075 * find_get_entry - find and get a page cache entry
485bb99b 1076 * @mapping: the address_space to search
0cd6144a
JW
1077 * @offset: the page cache index
1078 *
1079 * Looks up the page cache slot at @mapping & @offset. If there is a
1080 * page cache page, it is returned with an increased refcount.
485bb99b 1081 *
139b6a6f
JW
1082 * If the slot holds a shadow entry of a previously evicted page, or a
1083 * swap entry from shmem/tmpfs, it is returned.
0cd6144a
JW
1084 *
1085 * Otherwise, %NULL is returned.
1da177e4 1086 */
0cd6144a 1087struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1da177e4 1088{
a60637c8 1089 void **pagep;
83929372 1090 struct page *head, *page;
1da177e4 1091
a60637c8
NP
1092 rcu_read_lock();
1093repeat:
1094 page = NULL;
1095 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
1096 if (pagep) {
1097 page = radix_tree_deref_slot(pagep);
27d20fdd
NP
1098 if (unlikely(!page))
1099 goto out;
a2c16d6c 1100 if (radix_tree_exception(page)) {
8079b1c8
HD
1101 if (radix_tree_deref_retry(page))
1102 goto repeat;
1103 /*
139b6a6f
JW
1104 * A shadow entry of a recently evicted page,
1105 * or a swap entry from shmem/tmpfs. Return
1106 * it without attempting to raise page count.
8079b1c8
HD
1107 */
1108 goto out;
a2c16d6c 1109 }
83929372
KS
1110
1111 head = compound_head(page);
1112 if (!page_cache_get_speculative(head))
1113 goto repeat;
1114
1115 /* The page was split under us? */
1116 if (compound_head(page) != head) {
1117 put_page(head);
a60637c8 1118 goto repeat;
83929372 1119 }
a60637c8
NP
1120
1121 /*
1122 * Has the page moved?
1123 * This is part of the lockless pagecache protocol. See
1124 * include/linux/pagemap.h for details.
1125 */
1126 if (unlikely(page != *pagep)) {
83929372 1127 put_page(head);
a60637c8
NP
1128 goto repeat;
1129 }
1130 }
27d20fdd 1131out:
a60637c8
NP
1132 rcu_read_unlock();
1133
1da177e4
LT
1134 return page;
1135}
0cd6144a 1136EXPORT_SYMBOL(find_get_entry);
1da177e4 1137
0cd6144a
JW
1138/**
1139 * find_lock_entry - locate, pin and lock a page cache entry
1140 * @mapping: the address_space to search
1141 * @offset: the page cache index
1142 *
1143 * Looks up the page cache slot at @mapping & @offset. If there is a
1144 * page cache page, it is returned locked and with an increased
1145 * refcount.
1146 *
139b6a6f
JW
1147 * If the slot holds a shadow entry of a previously evicted page, or a
1148 * swap entry from shmem/tmpfs, it is returned.
0cd6144a
JW
1149 *
1150 * Otherwise, %NULL is returned.
1151 *
1152 * find_lock_entry() may sleep.
1153 */
1154struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1da177e4
LT
1155{
1156 struct page *page;
1157
1da177e4 1158repeat:
0cd6144a 1159 page = find_get_entry(mapping, offset);
a2c16d6c 1160 if (page && !radix_tree_exception(page)) {
a60637c8
NP
1161 lock_page(page);
1162 /* Has the page been truncated? */
83929372 1163 if (unlikely(page_mapping(page) != mapping)) {
a60637c8 1164 unlock_page(page);
09cbfeaf 1165 put_page(page);
a60637c8 1166 goto repeat;
1da177e4 1167 }
83929372 1168 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
1da177e4 1169 }
1da177e4
LT
1170 return page;
1171}
0cd6144a
JW
1172EXPORT_SYMBOL(find_lock_entry);
1173
1174/**
2457aec6 1175 * pagecache_get_page - find and get a page reference
0cd6144a
JW
1176 * @mapping: the address_space to search
1177 * @offset: the page index
2457aec6 1178 * @fgp_flags: PCG flags
45f87de5 1179 * @gfp_mask: gfp mask to use for the page cache data page allocation
0cd6144a 1180 *
2457aec6 1181 * Looks up the page cache slot at @mapping & @offset.
1da177e4 1182 *
75325189 1183 * PCG flags modify how the page is returned.
0cd6144a 1184 *
2457aec6
MG
1185 * FGP_ACCESSED: the page will be marked accessed
1186 * FGP_LOCK: Page is return locked
1187 * FGP_CREAT: If page is not present then a new page is allocated using
45f87de5
MH
1188 * @gfp_mask and added to the page cache and the VM's LRU
1189 * list. The page is returned locked and with an increased
1190 * refcount. Otherwise, %NULL is returned.
1da177e4 1191 *
2457aec6
MG
1192 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1193 * if the GFP flags specified for FGP_CREAT are atomic.
1da177e4 1194 *
2457aec6 1195 * If there is a page cache page, it is returned with an increased refcount.
1da177e4 1196 */
2457aec6 1197struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
45f87de5 1198 int fgp_flags, gfp_t gfp_mask)
1da177e4 1199{
eb2be189 1200 struct page *page;
2457aec6 1201
1da177e4 1202repeat:
2457aec6
MG
1203 page = find_get_entry(mapping, offset);
1204 if (radix_tree_exceptional_entry(page))
1205 page = NULL;
1206 if (!page)
1207 goto no_page;
1208
1209 if (fgp_flags & FGP_LOCK) {
1210 if (fgp_flags & FGP_NOWAIT) {
1211 if (!trylock_page(page)) {
09cbfeaf 1212 put_page(page);
2457aec6
MG
1213 return NULL;
1214 }
1215 } else {
1216 lock_page(page);
1217 }
1218
1219 /* Has the page been truncated? */
1220 if (unlikely(page->mapping != mapping)) {
1221 unlock_page(page);
09cbfeaf 1222 put_page(page);
2457aec6
MG
1223 goto repeat;
1224 }
1225 VM_BUG_ON_PAGE(page->index != offset, page);
1226 }
1227
1228 if (page && (fgp_flags & FGP_ACCESSED))
1229 mark_page_accessed(page);
1230
1231no_page:
1232 if (!page && (fgp_flags & FGP_CREAT)) {
1233 int err;
1234 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
45f87de5
MH
1235 gfp_mask |= __GFP_WRITE;
1236 if (fgp_flags & FGP_NOFS)
1237 gfp_mask &= ~__GFP_FS;
2457aec6 1238
45f87de5 1239 page = __page_cache_alloc(gfp_mask);
eb2be189
NP
1240 if (!page)
1241 return NULL;
2457aec6
MG
1242
1243 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1244 fgp_flags |= FGP_LOCK;
1245
eb39d618 1246 /* Init accessed so avoid atomic mark_page_accessed later */
2457aec6 1247 if (fgp_flags & FGP_ACCESSED)
eb39d618 1248 __SetPageReferenced(page);
2457aec6 1249
f4c86fa0 1250 err = add_to_page_cache_lru(page, mapping, offset, gfp_mask);
eb2be189 1251 if (unlikely(err)) {
09cbfeaf 1252 put_page(page);
eb2be189
NP
1253 page = NULL;
1254 if (err == -EEXIST)
1255 goto repeat;
1da177e4 1256 }
1da177e4 1257 }
2457aec6 1258
1da177e4
LT
1259 return page;
1260}
2457aec6 1261EXPORT_SYMBOL(pagecache_get_page);
1da177e4 1262
0cd6144a
JW
1263/**
1264 * find_get_entries - gang pagecache lookup
1265 * @mapping: The address_space to search
1266 * @start: The starting page cache index
1267 * @nr_entries: The maximum number of entries
1268 * @entries: Where the resulting entries are placed
1269 * @indices: The cache indices corresponding to the entries in @entries
1270 *
1271 * find_get_entries() will search for and return a group of up to
1272 * @nr_entries entries in the mapping. The entries are placed at
1273 * @entries. find_get_entries() takes a reference against any actual
1274 * pages it returns.
1275 *
1276 * The search returns a group of mapping-contiguous page cache entries
1277 * with ascending indexes. There may be holes in the indices due to
1278 * not-present pages.
1279 *
139b6a6f
JW
1280 * Any shadow entries of evicted pages, or swap entries from
1281 * shmem/tmpfs, are included in the returned array.
0cd6144a
JW
1282 *
1283 * find_get_entries() returns the number of pages and shadow entries
1284 * which were found.
1285 */
1286unsigned find_get_entries(struct address_space *mapping,
1287 pgoff_t start, unsigned int nr_entries,
1288 struct page **entries, pgoff_t *indices)
1289{
1290 void **slot;
1291 unsigned int ret = 0;
1292 struct radix_tree_iter iter;
1293
1294 if (!nr_entries)
1295 return 0;
1296
1297 rcu_read_lock();
0cd6144a 1298 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
83929372 1299 struct page *head, *page;
0cd6144a
JW
1300repeat:
1301 page = radix_tree_deref_slot(slot);
1302 if (unlikely(!page))
1303 continue;
1304 if (radix_tree_exception(page)) {
2cf938aa
MW
1305 if (radix_tree_deref_retry(page)) {
1306 slot = radix_tree_iter_retry(&iter);
1307 continue;
1308 }
0cd6144a 1309 /*
f9fe48be
RZ
1310 * A shadow entry of a recently evicted page, a swap
1311 * entry from shmem/tmpfs or a DAX entry. Return it
1312 * without attempting to raise page count.
0cd6144a
JW
1313 */
1314 goto export;
1315 }
83929372
KS
1316
1317 head = compound_head(page);
1318 if (!page_cache_get_speculative(head))
1319 goto repeat;
1320
1321 /* The page was split under us? */
1322 if (compound_head(page) != head) {
1323 put_page(head);
0cd6144a 1324 goto repeat;
83929372 1325 }
0cd6144a
JW
1326
1327 /* Has the page moved? */
1328 if (unlikely(page != *slot)) {
83929372 1329 put_page(head);
0cd6144a
JW
1330 goto repeat;
1331 }
1332export:
1333 indices[ret] = iter.index;
1334 entries[ret] = page;
1335 if (++ret == nr_entries)
1336 break;
1337 }
1338 rcu_read_unlock();
1339 return ret;
1340}
1341
1da177e4
LT
1342/**
1343 * find_get_pages - gang pagecache lookup
1344 * @mapping: The address_space to search
1345 * @start: The starting page index
1346 * @nr_pages: The maximum number of pages
1347 * @pages: Where the resulting pages are placed
1348 *
1349 * find_get_pages() will search for and return a group of up to
1350 * @nr_pages pages in the mapping. The pages are placed at @pages.
1351 * find_get_pages() takes a reference against the returned pages.
1352 *
1353 * The search returns a group of mapping-contiguous pages with ascending
1354 * indexes. There may be holes in the indices due to not-present pages.
1355 *
1356 * find_get_pages() returns the number of pages which were found.
1357 */
1358unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1359 unsigned int nr_pages, struct page **pages)
1360{
0fc9d104
KK
1361 struct radix_tree_iter iter;
1362 void **slot;
1363 unsigned ret = 0;
1364
1365 if (unlikely(!nr_pages))
1366 return 0;
a60637c8
NP
1367
1368 rcu_read_lock();
0fc9d104 1369 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
83929372 1370 struct page *head, *page;
a60637c8 1371repeat:
0fc9d104 1372 page = radix_tree_deref_slot(slot);
a60637c8
NP
1373 if (unlikely(!page))
1374 continue;
9d8aa4ea 1375
a2c16d6c 1376 if (radix_tree_exception(page)) {
8079b1c8 1377 if (radix_tree_deref_retry(page)) {
2cf938aa
MW
1378 slot = radix_tree_iter_retry(&iter);
1379 continue;
8079b1c8 1380 }
a2c16d6c 1381 /*
139b6a6f
JW
1382 * A shadow entry of a recently evicted page,
1383 * or a swap entry from shmem/tmpfs. Skip
1384 * over it.
a2c16d6c 1385 */
8079b1c8 1386 continue;
27d20fdd 1387 }
a60637c8 1388
83929372
KS
1389 head = compound_head(page);
1390 if (!page_cache_get_speculative(head))
1391 goto repeat;
1392
1393 /* The page was split under us? */
1394 if (compound_head(page) != head) {
1395 put_page(head);
a60637c8 1396 goto repeat;
83929372 1397 }
a60637c8
NP
1398
1399 /* Has the page moved? */
0fc9d104 1400 if (unlikely(page != *slot)) {
83929372 1401 put_page(head);
a60637c8
NP
1402 goto repeat;
1403 }
1da177e4 1404
a60637c8 1405 pages[ret] = page;
0fc9d104
KK
1406 if (++ret == nr_pages)
1407 break;
a60637c8 1408 }
5b280c0c 1409
a60637c8 1410 rcu_read_unlock();
1da177e4
LT
1411 return ret;
1412}
1413
ebf43500
JA
1414/**
1415 * find_get_pages_contig - gang contiguous pagecache lookup
1416 * @mapping: The address_space to search
1417 * @index: The starting page index
1418 * @nr_pages: The maximum number of pages
1419 * @pages: Where the resulting pages are placed
1420 *
1421 * find_get_pages_contig() works exactly like find_get_pages(), except
1422 * that the returned number of pages are guaranteed to be contiguous.
1423 *
1424 * find_get_pages_contig() returns the number of pages which were found.
1425 */
1426unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1427 unsigned int nr_pages, struct page **pages)
1428{
0fc9d104
KK
1429 struct radix_tree_iter iter;
1430 void **slot;
1431 unsigned int ret = 0;
1432
1433 if (unlikely(!nr_pages))
1434 return 0;
a60637c8
NP
1435
1436 rcu_read_lock();
0fc9d104 1437 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
83929372 1438 struct page *head, *page;
a60637c8 1439repeat:
0fc9d104
KK
1440 page = radix_tree_deref_slot(slot);
1441 /* The hole, there no reason to continue */
a60637c8 1442 if (unlikely(!page))
0fc9d104 1443 break;
9d8aa4ea 1444
a2c16d6c 1445 if (radix_tree_exception(page)) {
8079b1c8 1446 if (radix_tree_deref_retry(page)) {
2cf938aa
MW
1447 slot = radix_tree_iter_retry(&iter);
1448 continue;
8079b1c8 1449 }
a2c16d6c 1450 /*
139b6a6f
JW
1451 * A shadow entry of a recently evicted page,
1452 * or a swap entry from shmem/tmpfs. Stop
1453 * looking for contiguous pages.
a2c16d6c 1454 */
8079b1c8 1455 break;
a2c16d6c 1456 }
ebf43500 1457
83929372
KS
1458 head = compound_head(page);
1459 if (!page_cache_get_speculative(head))
1460 goto repeat;
1461
1462 /* The page was split under us? */
1463 if (compound_head(page) != head) {
1464 put_page(head);
a60637c8 1465 goto repeat;
83929372 1466 }
a60637c8
NP
1467
1468 /* Has the page moved? */
0fc9d104 1469 if (unlikely(page != *slot)) {
83929372 1470 put_page(head);
a60637c8
NP
1471 goto repeat;
1472 }
1473
9cbb4cb2
NP
1474 /*
1475 * must check mapping and index after taking the ref.
1476 * otherwise we can get both false positives and false
1477 * negatives, which is just confusing to the caller.
1478 */
83929372 1479 if (page->mapping == NULL || page_to_pgoff(page) != iter.index) {
09cbfeaf 1480 put_page(page);
9cbb4cb2
NP
1481 break;
1482 }
1483
a60637c8 1484 pages[ret] = page;
0fc9d104
KK
1485 if (++ret == nr_pages)
1486 break;
ebf43500 1487 }
a60637c8
NP
1488 rcu_read_unlock();
1489 return ret;
ebf43500 1490}
ef71c15c 1491EXPORT_SYMBOL(find_get_pages_contig);
ebf43500 1492
485bb99b
RD
1493/**
1494 * find_get_pages_tag - find and return pages that match @tag
1495 * @mapping: the address_space to search
1496 * @index: the starting page index
1497 * @tag: the tag index
1498 * @nr_pages: the maximum number of pages
1499 * @pages: where the resulting pages are placed
1500 *
1da177e4 1501 * Like find_get_pages, except we only return pages which are tagged with
485bb99b 1502 * @tag. We update @index to index the next page for the traversal.
1da177e4
LT
1503 */
1504unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1505 int tag, unsigned int nr_pages, struct page **pages)
1506{
0fc9d104
KK
1507 struct radix_tree_iter iter;
1508 void **slot;
1509 unsigned ret = 0;
1510
1511 if (unlikely(!nr_pages))
1512 return 0;
a60637c8
NP
1513
1514 rcu_read_lock();
0fc9d104
KK
1515 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1516 &iter, *index, tag) {
83929372 1517 struct page *head, *page;
a60637c8 1518repeat:
0fc9d104 1519 page = radix_tree_deref_slot(slot);
a60637c8
NP
1520 if (unlikely(!page))
1521 continue;
9d8aa4ea 1522
a2c16d6c 1523 if (radix_tree_exception(page)) {
8079b1c8 1524 if (radix_tree_deref_retry(page)) {
2cf938aa
MW
1525 slot = radix_tree_iter_retry(&iter);
1526 continue;
8079b1c8 1527 }
a2c16d6c 1528 /*
139b6a6f
JW
1529 * A shadow entry of a recently evicted page.
1530 *
1531 * Those entries should never be tagged, but
1532 * this tree walk is lockless and the tags are
1533 * looked up in bulk, one radix tree node at a
1534 * time, so there is a sizable window for page
1535 * reclaim to evict a page we saw tagged.
1536 *
1537 * Skip over it.
a2c16d6c 1538 */
139b6a6f 1539 continue;
a2c16d6c 1540 }
a60637c8 1541
83929372
KS
1542 head = compound_head(page);
1543 if (!page_cache_get_speculative(head))
a60637c8
NP
1544 goto repeat;
1545
83929372
KS
1546 /* The page was split under us? */
1547 if (compound_head(page) != head) {
1548 put_page(head);
1549 goto repeat;
1550 }
1551
a60637c8 1552 /* Has the page moved? */
0fc9d104 1553 if (unlikely(page != *slot)) {
83929372 1554 put_page(head);
a60637c8
NP
1555 goto repeat;
1556 }
1557
1558 pages[ret] = page;
0fc9d104
KK
1559 if (++ret == nr_pages)
1560 break;
a60637c8 1561 }
5b280c0c 1562
a60637c8 1563 rcu_read_unlock();
1da177e4 1564
1da177e4
LT
1565 if (ret)
1566 *index = pages[ret - 1]->index + 1;
a60637c8 1567
1da177e4
LT
1568 return ret;
1569}
ef71c15c 1570EXPORT_SYMBOL(find_get_pages_tag);
1da177e4 1571
7e7f7749
RZ
1572/**
1573 * find_get_entries_tag - find and return entries that match @tag
1574 * @mapping: the address_space to search
1575 * @start: the starting page cache index
1576 * @tag: the tag index
1577 * @nr_entries: the maximum number of entries
1578 * @entries: where the resulting entries are placed
1579 * @indices: the cache indices corresponding to the entries in @entries
1580 *
1581 * Like find_get_entries, except we only return entries which are tagged with
1582 * @tag.
1583 */
1584unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
1585 int tag, unsigned int nr_entries,
1586 struct page **entries, pgoff_t *indices)
1587{
1588 void **slot;
1589 unsigned int ret = 0;
1590 struct radix_tree_iter iter;
1591
1592 if (!nr_entries)
1593 return 0;
1594
1595 rcu_read_lock();
7e7f7749
RZ
1596 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1597 &iter, start, tag) {
83929372 1598 struct page *head, *page;
7e7f7749
RZ
1599repeat:
1600 page = radix_tree_deref_slot(slot);
1601 if (unlikely(!page))
1602 continue;
1603 if (radix_tree_exception(page)) {
1604 if (radix_tree_deref_retry(page)) {
2cf938aa
MW
1605 slot = radix_tree_iter_retry(&iter);
1606 continue;
7e7f7749
RZ
1607 }
1608
1609 /*
1610 * A shadow entry of a recently evicted page, a swap
1611 * entry from shmem/tmpfs or a DAX entry. Return it
1612 * without attempting to raise page count.
1613 */
1614 goto export;
1615 }
83929372
KS
1616
1617 head = compound_head(page);
1618 if (!page_cache_get_speculative(head))
7e7f7749
RZ
1619 goto repeat;
1620
83929372
KS
1621 /* The page was split under us? */
1622 if (compound_head(page) != head) {
1623 put_page(head);
1624 goto repeat;
1625 }
1626
7e7f7749
RZ
1627 /* Has the page moved? */
1628 if (unlikely(page != *slot)) {
83929372 1629 put_page(head);
7e7f7749
RZ
1630 goto repeat;
1631 }
1632export:
1633 indices[ret] = iter.index;
1634 entries[ret] = page;
1635 if (++ret == nr_entries)
1636 break;
1637 }
1638 rcu_read_unlock();
1639 return ret;
1640}
1641EXPORT_SYMBOL(find_get_entries_tag);
1642
76d42bd9
WF
1643/*
1644 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1645 * a _large_ part of the i/o request. Imagine the worst scenario:
1646 *
1647 * ---R__________________________________________B__________
1648 * ^ reading here ^ bad block(assume 4k)
1649 *
1650 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1651 * => failing the whole request => read(R) => read(R+1) =>
1652 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1653 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1654 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1655 *
1656 * It is going insane. Fix it by quickly scaling down the readahead size.
1657 */
1658static void shrink_readahead_size_eio(struct file *filp,
1659 struct file_ra_state *ra)
1660{
76d42bd9 1661 ra->ra_pages /= 4;
76d42bd9
WF
1662}
1663
485bb99b 1664/**
36e78914 1665 * do_generic_file_read - generic file read routine
485bb99b
RD
1666 * @filp: the file to read
1667 * @ppos: current file position
6e58e79d
AV
1668 * @iter: data destination
1669 * @written: already copied
485bb99b 1670 *
1da177e4 1671 * This is a generic file read routine, and uses the
485bb99b 1672 * mapping->a_ops->readpage() function for the actual low-level stuff.
1da177e4
LT
1673 *
1674 * This is really ugly. But the goto's actually try to clarify some
1675 * of the logic when it comes to error handling etc.
1da177e4 1676 */
6e58e79d
AV
1677static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1678 struct iov_iter *iter, ssize_t written)
1da177e4 1679{
36e78914 1680 struct address_space *mapping = filp->f_mapping;
1da177e4 1681 struct inode *inode = mapping->host;
36e78914 1682 struct file_ra_state *ra = &filp->f_ra;
57f6b96c
FW
1683 pgoff_t index;
1684 pgoff_t last_index;
1685 pgoff_t prev_index;
1686 unsigned long offset; /* offset into pagecache page */
ec0f1637 1687 unsigned int prev_offset;
6e58e79d 1688 int error = 0;
1da177e4 1689
c2a9737f 1690 if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
48466c47 1691 return 0;
c2a9737f
WF
1692 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
1693
09cbfeaf
KS
1694 index = *ppos >> PAGE_SHIFT;
1695 prev_index = ra->prev_pos >> PAGE_SHIFT;
1696 prev_offset = ra->prev_pos & (PAGE_SIZE-1);
1697 last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
1698 offset = *ppos & ~PAGE_MASK;
1da177e4 1699
1da177e4
LT
1700 for (;;) {
1701 struct page *page;
57f6b96c 1702 pgoff_t end_index;
a32ea1e1 1703 loff_t isize;
1da177e4
LT
1704 unsigned long nr, ret;
1705
1da177e4 1706 cond_resched();
1da177e4 1707find_page:
b67c7d39
MH
1708 if (fatal_signal_pending(current)) {
1709 error = -EINTR;
1710 goto out;
1711 }
1712
1da177e4 1713 page = find_get_page(mapping, index);
3ea89ee8 1714 if (!page) {
cf914a7d 1715 page_cache_sync_readahead(mapping,
7ff81078 1716 ra, filp,
3ea89ee8
FW
1717 index, last_index - index);
1718 page = find_get_page(mapping, index);
1719 if (unlikely(page == NULL))
1720 goto no_cached_page;
1721 }
1722 if (PageReadahead(page)) {
cf914a7d 1723 page_cache_async_readahead(mapping,
7ff81078 1724 ra, filp, page,
3ea89ee8 1725 index, last_index - index);
1da177e4 1726 }
8ab22b9a 1727 if (!PageUptodate(page)) {
ebded027
MG
1728 /*
1729 * See comment in do_read_cache_page on why
1730 * wait_on_page_locked is used to avoid unnecessarily
1731 * serialisations and why it's safe.
1732 */
c4b209a4
BVA
1733 error = wait_on_page_locked_killable(page);
1734 if (unlikely(error))
1735 goto readpage_error;
ebded027
MG
1736 if (PageUptodate(page))
1737 goto page_ok;
1738
09cbfeaf 1739 if (inode->i_blkbits == PAGE_SHIFT ||
8ab22b9a
HH
1740 !mapping->a_ops->is_partially_uptodate)
1741 goto page_not_up_to_date;
6d6d36bc
EG
1742 /* pipes can't handle partially uptodate pages */
1743 if (unlikely(iter->type & ITER_PIPE))
1744 goto page_not_up_to_date;
529ae9aa 1745 if (!trylock_page(page))
8ab22b9a 1746 goto page_not_up_to_date;
8d056cb9
DH
1747 /* Did it get truncated before we got the lock? */
1748 if (!page->mapping)
1749 goto page_not_up_to_date_locked;
8ab22b9a 1750 if (!mapping->a_ops->is_partially_uptodate(page,
6e58e79d 1751 offset, iter->count))
8ab22b9a
HH
1752 goto page_not_up_to_date_locked;
1753 unlock_page(page);
1754 }
1da177e4 1755page_ok:
a32ea1e1
N
1756 /*
1757 * i_size must be checked after we know the page is Uptodate.
1758 *
1759 * Checking i_size after the check allows us to calculate
1760 * the correct value for "nr", which means the zero-filled
1761 * part of the page is not copied back to userspace (unless
1762 * another truncate extends the file - this is desired though).
1763 */
1764
1765 isize = i_size_read(inode);
09cbfeaf 1766 end_index = (isize - 1) >> PAGE_SHIFT;
a32ea1e1 1767 if (unlikely(!isize || index > end_index)) {
09cbfeaf 1768 put_page(page);
a32ea1e1
N
1769 goto out;
1770 }
1771
1772 /* nr is the maximum number of bytes to copy from this page */
09cbfeaf 1773 nr = PAGE_SIZE;
a32ea1e1 1774 if (index == end_index) {
09cbfeaf 1775 nr = ((isize - 1) & ~PAGE_MASK) + 1;
a32ea1e1 1776 if (nr <= offset) {
09cbfeaf 1777 put_page(page);
a32ea1e1
N
1778 goto out;
1779 }
1780 }
1781 nr = nr - offset;
1da177e4
LT
1782
1783 /* If users can be writing to this page using arbitrary
1784 * virtual addresses, take care about potential aliasing
1785 * before reading the page on the kernel side.
1786 */
1787 if (mapping_writably_mapped(mapping))
1788 flush_dcache_page(page);
1789
1790 /*
ec0f1637
JK
1791 * When a sequential read accesses a page several times,
1792 * only mark it as accessed the first time.
1da177e4 1793 */
ec0f1637 1794 if (prev_index != index || offset != prev_offset)
1da177e4
LT
1795 mark_page_accessed(page);
1796 prev_index = index;
1797
1798 /*
1799 * Ok, we have the page, and it's up-to-date, so
1800 * now we can copy it to user space...
1da177e4 1801 */
6e58e79d
AV
1802
1803 ret = copy_page_to_iter(page, offset, nr, iter);
1da177e4 1804 offset += ret;
09cbfeaf
KS
1805 index += offset >> PAGE_SHIFT;
1806 offset &= ~PAGE_MASK;
6ce745ed 1807 prev_offset = offset;
1da177e4 1808
09cbfeaf 1809 put_page(page);
6e58e79d
AV
1810 written += ret;
1811 if (!iov_iter_count(iter))
1812 goto out;
1813 if (ret < nr) {
1814 error = -EFAULT;
1815 goto out;
1816 }
1817 continue;
1da177e4
LT
1818
1819page_not_up_to_date:
1820 /* Get exclusive access to the page ... */
85462323
ON
1821 error = lock_page_killable(page);
1822 if (unlikely(error))
1823 goto readpage_error;
1da177e4 1824
8ab22b9a 1825page_not_up_to_date_locked:
da6052f7 1826 /* Did it get truncated before we got the lock? */
1da177e4
LT
1827 if (!page->mapping) {
1828 unlock_page(page);
09cbfeaf 1829 put_page(page);
1da177e4
LT
1830 continue;
1831 }
1832
1833 /* Did somebody else fill it already? */
1834 if (PageUptodate(page)) {
1835 unlock_page(page);
1836 goto page_ok;
1837 }
1838
1839readpage:
91803b49
JM
1840 /*
1841 * A previous I/O error may have been due to temporary
1842 * failures, eg. multipath errors.
1843 * PG_error will be set again if readpage fails.
1844 */
1845 ClearPageError(page);
1da177e4
LT
1846 /* Start the actual read. The read will unlock the page. */
1847 error = mapping->a_ops->readpage(filp, page);
1848
994fc28c
ZB
1849 if (unlikely(error)) {
1850 if (error == AOP_TRUNCATED_PAGE) {
09cbfeaf 1851 put_page(page);
6e58e79d 1852 error = 0;
994fc28c
ZB
1853 goto find_page;
1854 }
1da177e4 1855 goto readpage_error;
994fc28c 1856 }
1da177e4
LT
1857
1858 if (!PageUptodate(page)) {
85462323
ON
1859 error = lock_page_killable(page);
1860 if (unlikely(error))
1861 goto readpage_error;
1da177e4
LT
1862 if (!PageUptodate(page)) {
1863 if (page->mapping == NULL) {
1864 /*
2ecdc82e 1865 * invalidate_mapping_pages got it
1da177e4
LT
1866 */
1867 unlock_page(page);
09cbfeaf 1868 put_page(page);
1da177e4
LT
1869 goto find_page;
1870 }
1871 unlock_page(page);
7ff81078 1872 shrink_readahead_size_eio(filp, ra);
85462323
ON
1873 error = -EIO;
1874 goto readpage_error;
1da177e4
LT
1875 }
1876 unlock_page(page);
1877 }
1878
1da177e4
LT
1879 goto page_ok;
1880
1881readpage_error:
1882 /* UHHUH! A synchronous read error occurred. Report it */
09cbfeaf 1883 put_page(page);
1da177e4
LT
1884 goto out;
1885
1886no_cached_page:
1887 /*
1888 * Ok, it wasn't cached, so we need to create a new
1889 * page..
1890 */
eb2be189
NP
1891 page = page_cache_alloc_cold(mapping);
1892 if (!page) {
6e58e79d 1893 error = -ENOMEM;
eb2be189 1894 goto out;
1da177e4 1895 }
6afdb859 1896 error = add_to_page_cache_lru(page, mapping, index,
c62d2555 1897 mapping_gfp_constraint(mapping, GFP_KERNEL));
1da177e4 1898 if (error) {
09cbfeaf 1899 put_page(page);
6e58e79d
AV
1900 if (error == -EEXIST) {
1901 error = 0;
1da177e4 1902 goto find_page;
6e58e79d 1903 }
1da177e4
LT
1904 goto out;
1905 }
1da177e4
LT
1906 goto readpage;
1907 }
1908
1909out:
7ff81078 1910 ra->prev_pos = prev_index;
09cbfeaf 1911 ra->prev_pos <<= PAGE_SHIFT;
7ff81078 1912 ra->prev_pos |= prev_offset;
1da177e4 1913
09cbfeaf 1914 *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
0c6aa263 1915 file_accessed(filp);
6e58e79d 1916 return written ? written : error;
1da177e4
LT
1917}
1918
485bb99b 1919/**
6abd2322 1920 * generic_file_read_iter - generic filesystem read routine
485bb99b 1921 * @iocb: kernel I/O control block
6abd2322 1922 * @iter: destination for the data read
485bb99b 1923 *
6abd2322 1924 * This is the "read_iter()" routine for all filesystems
1da177e4
LT
1925 * that can use the page cache directly.
1926 */
1927ssize_t
ed978a81 1928generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
1da177e4 1929{
ed978a81 1930 struct file *file = iocb->ki_filp;
cb66a7a1 1931 ssize_t retval = 0;
e7080a43
NS
1932 size_t count = iov_iter_count(iter);
1933
1934 if (!count)
1935 goto out; /* skip atime */
1da177e4 1936
2ba48ce5 1937 if (iocb->ki_flags & IOCB_DIRECT) {
ed978a81
AV
1938 struct address_space *mapping = file->f_mapping;
1939 struct inode *inode = mapping->host;
0d5b0cf2 1940 struct iov_iter data = *iter;
543ade1f 1941 loff_t size;
1da177e4 1942
1da177e4 1943 size = i_size_read(inode);
c64fb5c7
CH
1944 retval = filemap_write_and_wait_range(mapping, iocb->ki_pos,
1945 iocb->ki_pos + count - 1);
0d5b0cf2
CH
1946 if (retval < 0)
1947 goto out;
d8d3d94b 1948
0d5b0cf2
CH
1949 file_accessed(file);
1950
1951 retval = mapping->a_ops->direct_IO(iocb, &data);
c3a69024 1952 if (retval >= 0) {
c64fb5c7 1953 iocb->ki_pos += retval;
ed978a81 1954 iov_iter_advance(iter, retval);
9fe55eea 1955 }
66f998f6 1956
9fe55eea
SW
1957 /*
1958 * Btrfs can have a short DIO read if we encounter
1959 * compressed extents, so if there was an error, or if
1960 * we've already read everything we wanted to, or if
1961 * there was a short read because we hit EOF, go ahead
1962 * and return. Otherwise fallthrough to buffered io for
fbbbad4b
MW
1963 * the rest of the read. Buffered reads will not work for
1964 * DAX files, so don't bother trying.
9fe55eea 1965 */
c64fb5c7 1966 if (retval < 0 || !iov_iter_count(iter) || iocb->ki_pos >= size ||
0d5b0cf2 1967 IS_DAX(inode))
9fe55eea 1968 goto out;
1da177e4
LT
1969 }
1970
c64fb5c7 1971 retval = do_generic_file_read(file, &iocb->ki_pos, iter, retval);
1da177e4
LT
1972out:
1973 return retval;
1974}
ed978a81 1975EXPORT_SYMBOL(generic_file_read_iter);
1da177e4 1976
1da177e4 1977#ifdef CONFIG_MMU
485bb99b
RD
1978/**
1979 * page_cache_read - adds requested page to the page cache if not already there
1980 * @file: file to read
1981 * @offset: page index
62eb320a 1982 * @gfp_mask: memory allocation flags
485bb99b 1983 *
1da177e4
LT
1984 * This adds the requested page to the page cache if it isn't already there,
1985 * and schedules an I/O to read in its contents from disk.
1986 */
c20cd45e 1987static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask)
1da177e4
LT
1988{
1989 struct address_space *mapping = file->f_mapping;
99dadfdd 1990 struct page *page;
994fc28c 1991 int ret;
1da177e4 1992
994fc28c 1993 do {
c20cd45e 1994 page = __page_cache_alloc(gfp_mask|__GFP_COLD);
994fc28c
ZB
1995 if (!page)
1996 return -ENOMEM;
1997
f4c86fa0 1998 ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask);
994fc28c
ZB
1999 if (ret == 0)
2000 ret = mapping->a_ops->readpage(file, page);
2001 else if (ret == -EEXIST)
2002 ret = 0; /* losing race to add is OK */
1da177e4 2003
09cbfeaf 2004 put_page(page);
1da177e4 2005
994fc28c 2006 } while (ret == AOP_TRUNCATED_PAGE);
99dadfdd 2007
994fc28c 2008 return ret;
1da177e4
LT
2009}
2010
2011#define MMAP_LOTSAMISS (100)
2012
ef00e08e
LT
2013/*
2014 * Synchronous readahead happens when we don't even find
2015 * a page in the page cache at all.
2016 */
2017static void do_sync_mmap_readahead(struct vm_area_struct *vma,
2018 struct file_ra_state *ra,
2019 struct file *file,
2020 pgoff_t offset)
2021{
ef00e08e
LT
2022 struct address_space *mapping = file->f_mapping;
2023
2024 /* If we don't want any read-ahead, don't bother */
64363aad 2025 if (vma->vm_flags & VM_RAND_READ)
ef00e08e 2026 return;
275b12bf
WF
2027 if (!ra->ra_pages)
2028 return;
ef00e08e 2029
64363aad 2030 if (vma->vm_flags & VM_SEQ_READ) {
7ffc59b4
WF
2031 page_cache_sync_readahead(mapping, ra, file, offset,
2032 ra->ra_pages);
ef00e08e
LT
2033 return;
2034 }
2035
207d04ba
AK
2036 /* Avoid banging the cache line if not needed */
2037 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
ef00e08e
LT
2038 ra->mmap_miss++;
2039
2040 /*
2041 * Do we miss much more than hit in this file? If so,
2042 * stop bothering with read-ahead. It will only hurt.
2043 */
2044 if (ra->mmap_miss > MMAP_LOTSAMISS)
2045 return;
2046
d30a1100
WF
2047 /*
2048 * mmap read-around
2049 */
600e19af
RG
2050 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
2051 ra->size = ra->ra_pages;
2052 ra->async_size = ra->ra_pages / 4;
275b12bf 2053 ra_submit(ra, mapping, file);
ef00e08e
LT
2054}
2055
2056/*
2057 * Asynchronous readahead happens when we find the page and PG_readahead,
2058 * so we want to possibly extend the readahead further..
2059 */
2060static void do_async_mmap_readahead(struct vm_area_struct *vma,
2061 struct file_ra_state *ra,
2062 struct file *file,
2063 struct page *page,
2064 pgoff_t offset)
2065{
2066 struct address_space *mapping = file->f_mapping;
2067
2068 /* If we don't want any read-ahead, don't bother */
64363aad 2069 if (vma->vm_flags & VM_RAND_READ)
ef00e08e
LT
2070 return;
2071 if (ra->mmap_miss > 0)
2072 ra->mmap_miss--;
2073 if (PageReadahead(page))
2fad6f5d
WF
2074 page_cache_async_readahead(mapping, ra, file,
2075 page, offset, ra->ra_pages);
ef00e08e
LT
2076}
2077
485bb99b 2078/**
54cb8821 2079 * filemap_fault - read in file data for page fault handling
d0217ac0
NP
2080 * @vma: vma in which the fault was taken
2081 * @vmf: struct vm_fault containing details of the fault
485bb99b 2082 *
54cb8821 2083 * filemap_fault() is invoked via the vma operations vector for a
1da177e4
LT
2084 * mapped memory region to read in file data during a page fault.
2085 *
2086 * The goto's are kind of ugly, but this streamlines the normal case of having
2087 * it in the page cache, and handles the special cases reasonably without
2088 * having a lot of duplicated code.
9a95f3cf
PC
2089 *
2090 * vma->vm_mm->mmap_sem must be held on entry.
2091 *
2092 * If our return value has VM_FAULT_RETRY set, it's because
2093 * lock_page_or_retry() returned 0.
2094 * The mmap_sem has usually been released in this case.
2095 * See __lock_page_or_retry() for the exception.
2096 *
2097 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2098 * has not been released.
2099 *
2100 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
1da177e4 2101 */
d0217ac0 2102int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1da177e4
LT
2103{
2104 int error;
54cb8821 2105 struct file *file = vma->vm_file;
1da177e4
LT
2106 struct address_space *mapping = file->f_mapping;
2107 struct file_ra_state *ra = &file->f_ra;
2108 struct inode *inode = mapping->host;
ef00e08e 2109 pgoff_t offset = vmf->pgoff;
1da177e4 2110 struct page *page;
99e3e53f 2111 loff_t size;
83c54070 2112 int ret = 0;
1da177e4 2113
09cbfeaf
KS
2114 size = round_up(i_size_read(inode), PAGE_SIZE);
2115 if (offset >= size >> PAGE_SHIFT)
5307cc1a 2116 return VM_FAULT_SIGBUS;
1da177e4 2117
1da177e4 2118 /*
49426420 2119 * Do we have something in the page cache already?
1da177e4 2120 */
ef00e08e 2121 page = find_get_page(mapping, offset);
45cac65b 2122 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
1da177e4 2123 /*
ef00e08e
LT
2124 * We found the page, so try async readahead before
2125 * waiting for the lock.
1da177e4 2126 */
ef00e08e 2127 do_async_mmap_readahead(vma, ra, file, page, offset);
45cac65b 2128 } else if (!page) {
ef00e08e
LT
2129 /* No page in the page cache at all */
2130 do_sync_mmap_readahead(vma, ra, file, offset);
2131 count_vm_event(PGMAJFAULT);
456f998e 2132 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
ef00e08e
LT
2133 ret = VM_FAULT_MAJOR;
2134retry_find:
b522c94d 2135 page = find_get_page(mapping, offset);
1da177e4
LT
2136 if (!page)
2137 goto no_cached_page;
2138 }
2139
d88c0922 2140 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
09cbfeaf 2141 put_page(page);
d065bd81 2142 return ret | VM_FAULT_RETRY;
d88c0922 2143 }
b522c94d
ML
2144
2145 /* Did it get truncated? */
2146 if (unlikely(page->mapping != mapping)) {
2147 unlock_page(page);
2148 put_page(page);
2149 goto retry_find;
2150 }
309381fe 2151 VM_BUG_ON_PAGE(page->index != offset, page);
b522c94d 2152
1da177e4 2153 /*
d00806b1
NP
2154 * We have a locked page in the page cache, now we need to check
2155 * that it's up-to-date. If not, it is going to be due to an error.
1da177e4 2156 */
d00806b1 2157 if (unlikely(!PageUptodate(page)))
1da177e4
LT
2158 goto page_not_uptodate;
2159
ef00e08e
LT
2160 /*
2161 * Found the page and have a reference on it.
2162 * We must recheck i_size under page lock.
2163 */
09cbfeaf
KS
2164 size = round_up(i_size_read(inode), PAGE_SIZE);
2165 if (unlikely(offset >= size >> PAGE_SHIFT)) {
d00806b1 2166 unlock_page(page);
09cbfeaf 2167 put_page(page);
5307cc1a 2168 return VM_FAULT_SIGBUS;
d00806b1
NP
2169 }
2170
d0217ac0 2171 vmf->page = page;
83c54070 2172 return ret | VM_FAULT_LOCKED;
1da177e4 2173
1da177e4
LT
2174no_cached_page:
2175 /*
2176 * We're only likely to ever get here if MADV_RANDOM is in
2177 * effect.
2178 */
c20cd45e 2179 error = page_cache_read(file, offset, vmf->gfp_mask);
1da177e4
LT
2180
2181 /*
2182 * The page we want has now been added to the page cache.
2183 * In the unlikely event that someone removed it in the
2184 * meantime, we'll just come back here and read it again.
2185 */
2186 if (error >= 0)
2187 goto retry_find;
2188
2189 /*
2190 * An error return from page_cache_read can result if the
2191 * system is low on memory, or a problem occurs while trying
2192 * to schedule I/O.
2193 */
2194 if (error == -ENOMEM)
d0217ac0
NP
2195 return VM_FAULT_OOM;
2196 return VM_FAULT_SIGBUS;
1da177e4
LT
2197
2198page_not_uptodate:
1da177e4
LT
2199 /*
2200 * Umm, take care of errors if the page isn't up-to-date.
2201 * Try to re-read it _once_. We do this synchronously,
2202 * because there really aren't any performance issues here
2203 * and we need to check for errors.
2204 */
1da177e4 2205 ClearPageError(page);
994fc28c 2206 error = mapping->a_ops->readpage(file, page);
3ef0f720
MS
2207 if (!error) {
2208 wait_on_page_locked(page);
2209 if (!PageUptodate(page))
2210 error = -EIO;
2211 }
09cbfeaf 2212 put_page(page);
d00806b1
NP
2213
2214 if (!error || error == AOP_TRUNCATED_PAGE)
994fc28c 2215 goto retry_find;
1da177e4 2216
d00806b1 2217 /* Things didn't work out. Return zero to tell the mm layer so. */
76d42bd9 2218 shrink_readahead_size_eio(file, ra);
d0217ac0 2219 return VM_FAULT_SIGBUS;
54cb8821
NP
2220}
2221EXPORT_SYMBOL(filemap_fault);
2222
bae473a4
KS
2223void filemap_map_pages(struct fault_env *fe,
2224 pgoff_t start_pgoff, pgoff_t end_pgoff)
f1820361
KS
2225{
2226 struct radix_tree_iter iter;
2227 void **slot;
bae473a4 2228 struct file *file = fe->vma->vm_file;
f1820361 2229 struct address_space *mapping = file->f_mapping;
bae473a4 2230 pgoff_t last_pgoff = start_pgoff;
f1820361 2231 loff_t size;
83929372 2232 struct page *head, *page;
f1820361
KS
2233
2234 rcu_read_lock();
bae473a4
KS
2235 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter,
2236 start_pgoff) {
2237 if (iter.index > end_pgoff)
f1820361
KS
2238 break;
2239repeat:
2240 page = radix_tree_deref_slot(slot);
2241 if (unlikely(!page))
2242 goto next;
2243 if (radix_tree_exception(page)) {
2cf938aa
MW
2244 if (radix_tree_deref_retry(page)) {
2245 slot = radix_tree_iter_retry(&iter);
2246 continue;
2247 }
2248 goto next;
f1820361
KS
2249 }
2250
83929372
KS
2251 head = compound_head(page);
2252 if (!page_cache_get_speculative(head))
f1820361
KS
2253 goto repeat;
2254
83929372
KS
2255 /* The page was split under us? */
2256 if (compound_head(page) != head) {
2257 put_page(head);
2258 goto repeat;
2259 }
2260
f1820361
KS
2261 /* Has the page moved? */
2262 if (unlikely(page != *slot)) {
83929372 2263 put_page(head);
f1820361
KS
2264 goto repeat;
2265 }
2266
2267 if (!PageUptodate(page) ||
2268 PageReadahead(page) ||
2269 PageHWPoison(page))
2270 goto skip;
2271 if (!trylock_page(page))
2272 goto skip;
2273
2274 if (page->mapping != mapping || !PageUptodate(page))
2275 goto unlock;
2276
09cbfeaf
KS
2277 size = round_up(i_size_read(mapping->host), PAGE_SIZE);
2278 if (page->index >= size >> PAGE_SHIFT)
f1820361
KS
2279 goto unlock;
2280
f1820361
KS
2281 if (file->f_ra.mmap_miss > 0)
2282 file->f_ra.mmap_miss--;
7267ec00
KS
2283
2284 fe->address += (iter.index - last_pgoff) << PAGE_SHIFT;
2285 if (fe->pte)
2286 fe->pte += iter.index - last_pgoff;
2287 last_pgoff = iter.index;
2288 if (alloc_set_pte(fe, NULL, page))
2289 goto unlock;
f1820361
KS
2290 unlock_page(page);
2291 goto next;
2292unlock:
2293 unlock_page(page);
2294skip:
09cbfeaf 2295 put_page(page);
f1820361 2296next:
7267ec00
KS
2297 /* Huge page is mapped? No need to proceed. */
2298 if (pmd_trans_huge(*fe->pmd))
2299 break;
bae473a4 2300 if (iter.index == end_pgoff)
f1820361
KS
2301 break;
2302 }
2303 rcu_read_unlock();
2304}
2305EXPORT_SYMBOL(filemap_map_pages);
2306
4fcf1c62
JK
2307int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2308{
2309 struct page *page = vmf->page;
496ad9aa 2310 struct inode *inode = file_inode(vma->vm_file);
4fcf1c62
JK
2311 int ret = VM_FAULT_LOCKED;
2312
14da9200 2313 sb_start_pagefault(inode->i_sb);
4fcf1c62
JK
2314 file_update_time(vma->vm_file);
2315 lock_page(page);
2316 if (page->mapping != inode->i_mapping) {
2317 unlock_page(page);
2318 ret = VM_FAULT_NOPAGE;
2319 goto out;
2320 }
14da9200
JK
2321 /*
2322 * We mark the page dirty already here so that when freeze is in
2323 * progress, we are guaranteed that writeback during freezing will
2324 * see the dirty page and writeprotect it again.
2325 */
2326 set_page_dirty(page);
1d1d1a76 2327 wait_for_stable_page(page);
4fcf1c62 2328out:
14da9200 2329 sb_end_pagefault(inode->i_sb);
4fcf1c62
JK
2330 return ret;
2331}
2332EXPORT_SYMBOL(filemap_page_mkwrite);
2333
f0f37e2f 2334const struct vm_operations_struct generic_file_vm_ops = {
54cb8821 2335 .fault = filemap_fault,
f1820361 2336 .map_pages = filemap_map_pages,
4fcf1c62 2337 .page_mkwrite = filemap_page_mkwrite,
1da177e4
LT
2338};
2339
2340/* This is used for a general mmap of a disk file */
2341
2342int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2343{
2344 struct address_space *mapping = file->f_mapping;
2345
2346 if (!mapping->a_ops->readpage)
2347 return -ENOEXEC;
2348 file_accessed(file);
2349 vma->vm_ops = &generic_file_vm_ops;
2350 return 0;
2351}
1da177e4
LT
2352
2353/*
2354 * This is for filesystems which do not implement ->writepage.
2355 */
2356int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2357{
2358 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2359 return -EINVAL;
2360 return generic_file_mmap(file, vma);
2361}
2362#else
2363int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2364{
2365 return -ENOSYS;
2366}
2367int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2368{
2369 return -ENOSYS;
2370}
2371#endif /* CONFIG_MMU */
2372
2373EXPORT_SYMBOL(generic_file_mmap);
2374EXPORT_SYMBOL(generic_file_readonly_mmap);
2375
67f9fd91
SL
2376static struct page *wait_on_page_read(struct page *page)
2377{
2378 if (!IS_ERR(page)) {
2379 wait_on_page_locked(page);
2380 if (!PageUptodate(page)) {
09cbfeaf 2381 put_page(page);
67f9fd91
SL
2382 page = ERR_PTR(-EIO);
2383 }
2384 }
2385 return page;
2386}
2387
32b63529 2388static struct page *do_read_cache_page(struct address_space *mapping,
57f6b96c 2389 pgoff_t index,
5e5358e7 2390 int (*filler)(void *, struct page *),
0531b2aa
LT
2391 void *data,
2392 gfp_t gfp)
1da177e4 2393{
eb2be189 2394 struct page *page;
1da177e4
LT
2395 int err;
2396repeat:
2397 page = find_get_page(mapping, index);
2398 if (!page) {
0531b2aa 2399 page = __page_cache_alloc(gfp | __GFP_COLD);
eb2be189
NP
2400 if (!page)
2401 return ERR_PTR(-ENOMEM);
e6f67b8c 2402 err = add_to_page_cache_lru(page, mapping, index, gfp);
eb2be189 2403 if (unlikely(err)) {
09cbfeaf 2404 put_page(page);
eb2be189
NP
2405 if (err == -EEXIST)
2406 goto repeat;
1da177e4 2407 /* Presumably ENOMEM for radix tree node */
1da177e4
LT
2408 return ERR_PTR(err);
2409 }
32b63529
MG
2410
2411filler:
1da177e4
LT
2412 err = filler(data, page);
2413 if (err < 0) {
09cbfeaf 2414 put_page(page);
32b63529 2415 return ERR_PTR(err);
1da177e4 2416 }
1da177e4 2417
32b63529
MG
2418 page = wait_on_page_read(page);
2419 if (IS_ERR(page))
2420 return page;
2421 goto out;
2422 }
1da177e4
LT
2423 if (PageUptodate(page))
2424 goto out;
2425
ebded027
MG
2426 /*
2427 * Page is not up to date and may be locked due one of the following
2428 * case a: Page is being filled and the page lock is held
2429 * case b: Read/write error clearing the page uptodate status
2430 * case c: Truncation in progress (page locked)
2431 * case d: Reclaim in progress
2432 *
2433 * Case a, the page will be up to date when the page is unlocked.
2434 * There is no need to serialise on the page lock here as the page
2435 * is pinned so the lock gives no additional protection. Even if the
2436 * the page is truncated, the data is still valid if PageUptodate as
2437 * it's a race vs truncate race.
2438 * Case b, the page will not be up to date
2439 * Case c, the page may be truncated but in itself, the data may still
2440 * be valid after IO completes as it's a read vs truncate race. The
2441 * operation must restart if the page is not uptodate on unlock but
2442 * otherwise serialising on page lock to stabilise the mapping gives
2443 * no additional guarantees to the caller as the page lock is
2444 * released before return.
2445 * Case d, similar to truncation. If reclaim holds the page lock, it
2446 * will be a race with remove_mapping that determines if the mapping
2447 * is valid on unlock but otherwise the data is valid and there is
2448 * no need to serialise with page lock.
2449 *
2450 * As the page lock gives no additional guarantee, we optimistically
2451 * wait on the page to be unlocked and check if it's up to date and
2452 * use the page if it is. Otherwise, the page lock is required to
2453 * distinguish between the different cases. The motivation is that we
2454 * avoid spurious serialisations and wakeups when multiple processes
2455 * wait on the same page for IO to complete.
2456 */
2457 wait_on_page_locked(page);
2458 if (PageUptodate(page))
2459 goto out;
2460
2461 /* Distinguish between all the cases under the safety of the lock */
1da177e4 2462 lock_page(page);
ebded027
MG
2463
2464 /* Case c or d, restart the operation */
1da177e4
LT
2465 if (!page->mapping) {
2466 unlock_page(page);
09cbfeaf 2467 put_page(page);
32b63529 2468 goto repeat;
1da177e4 2469 }
ebded027
MG
2470
2471 /* Someone else locked and filled the page in a very small window */
1da177e4
LT
2472 if (PageUptodate(page)) {
2473 unlock_page(page);
2474 goto out;
2475 }
32b63529
MG
2476 goto filler;
2477
c855ff37 2478out:
6fe6900e
NP
2479 mark_page_accessed(page);
2480 return page;
2481}
0531b2aa
LT
2482
2483/**
67f9fd91 2484 * read_cache_page - read into page cache, fill it if needed
0531b2aa
LT
2485 * @mapping: the page's address_space
2486 * @index: the page index
2487 * @filler: function to perform the read
5e5358e7 2488 * @data: first arg to filler(data, page) function, often left as NULL
0531b2aa 2489 *
0531b2aa 2490 * Read into the page cache. If a page already exists, and PageUptodate() is
67f9fd91 2491 * not set, try to fill the page and wait for it to become unlocked.
0531b2aa
LT
2492 *
2493 * If the page does not get brought uptodate, return -EIO.
2494 */
67f9fd91 2495struct page *read_cache_page(struct address_space *mapping,
0531b2aa 2496 pgoff_t index,
5e5358e7 2497 int (*filler)(void *, struct page *),
0531b2aa
LT
2498 void *data)
2499{
2500 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2501}
67f9fd91 2502EXPORT_SYMBOL(read_cache_page);
0531b2aa
LT
2503
2504/**
2505 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2506 * @mapping: the page's address_space
2507 * @index: the page index
2508 * @gfp: the page allocator flags to use if allocating
2509 *
2510 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
e6f67b8c 2511 * any new page allocations done using the specified allocation flags.
0531b2aa
LT
2512 *
2513 * If the page does not get brought uptodate, return -EIO.
2514 */
2515struct page *read_cache_page_gfp(struct address_space *mapping,
2516 pgoff_t index,
2517 gfp_t gfp)
2518{
2519 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2520
67f9fd91 2521 return do_read_cache_page(mapping, index, filler, NULL, gfp);
0531b2aa
LT
2522}
2523EXPORT_SYMBOL(read_cache_page_gfp);
2524
1da177e4
LT
2525/*
2526 * Performs necessary checks before doing a write
2527 *
485bb99b 2528 * Can adjust writing position or amount of bytes to write.
1da177e4
LT
2529 * Returns appropriate error code that caller should return or
2530 * zero in case that write should be allowed.
2531 */
3309dd04 2532inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
1da177e4 2533{
3309dd04 2534 struct file *file = iocb->ki_filp;
1da177e4 2535 struct inode *inode = file->f_mapping->host;
59e99e5b 2536 unsigned long limit = rlimit(RLIMIT_FSIZE);
3309dd04 2537 loff_t pos;
1da177e4 2538
3309dd04
AV
2539 if (!iov_iter_count(from))
2540 return 0;
1da177e4 2541
0fa6b005 2542 /* FIXME: this is for backwards compatibility with 2.4 */
2ba48ce5 2543 if (iocb->ki_flags & IOCB_APPEND)
3309dd04 2544 iocb->ki_pos = i_size_read(inode);
1da177e4 2545
3309dd04 2546 pos = iocb->ki_pos;
1da177e4 2547
0fa6b005 2548 if (limit != RLIM_INFINITY) {
3309dd04 2549 if (iocb->ki_pos >= limit) {
0fa6b005
AV
2550 send_sig(SIGXFSZ, current, 0);
2551 return -EFBIG;
1da177e4 2552 }
3309dd04 2553 iov_iter_truncate(from, limit - (unsigned long)pos);
1da177e4
LT
2554 }
2555
2556 /*
2557 * LFS rule
2558 */
3309dd04 2559 if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
1da177e4 2560 !(file->f_flags & O_LARGEFILE))) {
3309dd04 2561 if (pos >= MAX_NON_LFS)
1da177e4 2562 return -EFBIG;
3309dd04 2563 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
1da177e4
LT
2564 }
2565
2566 /*
2567 * Are we about to exceed the fs block limit ?
2568 *
2569 * If we have written data it becomes a short write. If we have
2570 * exceeded without writing data we send a signal and return EFBIG.
2571 * Linus frestrict idea will clean these up nicely..
2572 */
3309dd04
AV
2573 if (unlikely(pos >= inode->i_sb->s_maxbytes))
2574 return -EFBIG;
1da177e4 2575
3309dd04
AV
2576 iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2577 return iov_iter_count(from);
1da177e4
LT
2578}
2579EXPORT_SYMBOL(generic_write_checks);
2580
afddba49
NP
2581int pagecache_write_begin(struct file *file, struct address_space *mapping,
2582 loff_t pos, unsigned len, unsigned flags,
2583 struct page **pagep, void **fsdata)
2584{
2585 const struct address_space_operations *aops = mapping->a_ops;
2586
4e02ed4b 2587 return aops->write_begin(file, mapping, pos, len, flags,
afddba49 2588 pagep, fsdata);
afddba49
NP
2589}
2590EXPORT_SYMBOL(pagecache_write_begin);
2591
2592int pagecache_write_end(struct file *file, struct address_space *mapping,
2593 loff_t pos, unsigned len, unsigned copied,
2594 struct page *page, void *fsdata)
2595{
2596 const struct address_space_operations *aops = mapping->a_ops;
afddba49 2597
4e02ed4b 2598 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
afddba49
NP
2599}
2600EXPORT_SYMBOL(pagecache_write_end);
2601
1da177e4 2602ssize_t
1af5bb49 2603generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
1da177e4
LT
2604{
2605 struct file *file = iocb->ki_filp;
2606 struct address_space *mapping = file->f_mapping;
2607 struct inode *inode = mapping->host;
1af5bb49 2608 loff_t pos = iocb->ki_pos;
1da177e4 2609 ssize_t written;
a969e903
CH
2610 size_t write_len;
2611 pgoff_t end;
26978b8b 2612 struct iov_iter data;
1da177e4 2613
0c949334 2614 write_len = iov_iter_count(from);
09cbfeaf 2615 end = (pos + write_len - 1) >> PAGE_SHIFT;
a969e903 2616
48b47c56 2617 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
a969e903
CH
2618 if (written)
2619 goto out;
2620
2621 /*
2622 * After a write we want buffered reads to be sure to go to disk to get
2623 * the new data. We invalidate clean cached page from the region we're
2624 * about to write. We do this *before* the write so that we can return
6ccfa806 2625 * without clobbering -EIOCBQUEUED from ->direct_IO().
a969e903
CH
2626 */
2627 if (mapping->nrpages) {
2628 written = invalidate_inode_pages2_range(mapping,
09cbfeaf 2629 pos >> PAGE_SHIFT, end);
6ccfa806
HH
2630 /*
2631 * If a page can not be invalidated, return 0 to fall back
2632 * to buffered write.
2633 */
2634 if (written) {
2635 if (written == -EBUSY)
2636 return 0;
a969e903 2637 goto out;
6ccfa806 2638 }
a969e903
CH
2639 }
2640
26978b8b 2641 data = *from;
c8b8e32d 2642 written = mapping->a_ops->direct_IO(iocb, &data);
a969e903
CH
2643
2644 /*
2645 * Finally, try again to invalidate clean pages which might have been
2646 * cached by non-direct readahead, or faulted in by get_user_pages()
2647 * if the source of the write was an mmap'ed region of the file
2648 * we're writing. Either one is a pretty crazy thing to do,
2649 * so we don't support it 100%. If this invalidation
2650 * fails, tough, the write still worked...
2651 */
2652 if (mapping->nrpages) {
2653 invalidate_inode_pages2_range(mapping,
09cbfeaf 2654 pos >> PAGE_SHIFT, end);
a969e903
CH
2655 }
2656
1da177e4 2657 if (written > 0) {
0116651c 2658 pos += written;
f8579f86 2659 iov_iter_advance(from, written);
0116651c
NK
2660 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2661 i_size_write(inode, pos);
1da177e4
LT
2662 mark_inode_dirty(inode);
2663 }
5cb6c6c7 2664 iocb->ki_pos = pos;
1da177e4 2665 }
a969e903 2666out:
1da177e4
LT
2667 return written;
2668}
2669EXPORT_SYMBOL(generic_file_direct_write);
2670
eb2be189
NP
2671/*
2672 * Find or create a page at the given pagecache position. Return the locked
2673 * page. This function is specifically for buffered writes.
2674 */
54566b2c
NP
2675struct page *grab_cache_page_write_begin(struct address_space *mapping,
2676 pgoff_t index, unsigned flags)
eb2be189 2677{
eb2be189 2678 struct page *page;
bbddabe2 2679 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
0faa70cb 2680
54566b2c 2681 if (flags & AOP_FLAG_NOFS)
2457aec6
MG
2682 fgp_flags |= FGP_NOFS;
2683
2684 page = pagecache_get_page(mapping, index, fgp_flags,
45f87de5 2685 mapping_gfp_mask(mapping));
c585a267 2686 if (page)
2457aec6 2687 wait_for_stable_page(page);
eb2be189 2688
eb2be189
NP
2689 return page;
2690}
54566b2c 2691EXPORT_SYMBOL(grab_cache_page_write_begin);
eb2be189 2692
3b93f911 2693ssize_t generic_perform_write(struct file *file,
afddba49
NP
2694 struct iov_iter *i, loff_t pos)
2695{
2696 struct address_space *mapping = file->f_mapping;
2697 const struct address_space_operations *a_ops = mapping->a_ops;
2698 long status = 0;
2699 ssize_t written = 0;
674b892e
NP
2700 unsigned int flags = 0;
2701
2702 /*
2703 * Copies from kernel address space cannot fail (NFSD is a big user).
2704 */
777eda2c 2705 if (!iter_is_iovec(i))
674b892e 2706 flags |= AOP_FLAG_UNINTERRUPTIBLE;
afddba49
NP
2707
2708 do {
2709 struct page *page;
afddba49
NP
2710 unsigned long offset; /* Offset into pagecache page */
2711 unsigned long bytes; /* Bytes to write to page */
2712 size_t copied; /* Bytes copied from user */
2713 void *fsdata;
2714
09cbfeaf
KS
2715 offset = (pos & (PAGE_SIZE - 1));
2716 bytes = min_t(unsigned long, PAGE_SIZE - offset,
afddba49
NP
2717 iov_iter_count(i));
2718
2719again:
00a3d660
LT
2720 /*
2721 * Bring in the user page that we will copy from _first_.
2722 * Otherwise there's a nasty deadlock on copying from the
2723 * same page as we're writing to, without it being marked
2724 * up-to-date.
2725 *
2726 * Not only is this an optimisation, but it is also required
2727 * to check that the address is actually valid, when atomic
2728 * usercopies are used, below.
2729 */
2730 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2731 status = -EFAULT;
2732 break;
2733 }
2734
296291cd
JK
2735 if (fatal_signal_pending(current)) {
2736 status = -EINTR;
2737 break;
2738 }
2739
674b892e 2740 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
afddba49 2741 &page, &fsdata);
2457aec6 2742 if (unlikely(status < 0))
afddba49
NP
2743 break;
2744
931e80e4 2745 if (mapping_writably_mapped(mapping))
2746 flush_dcache_page(page);
00a3d660 2747
afddba49 2748 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
afddba49
NP
2749 flush_dcache_page(page);
2750
2751 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2752 page, fsdata);
2753 if (unlikely(status < 0))
2754 break;
2755 copied = status;
2756
2757 cond_resched();
2758
124d3b70 2759 iov_iter_advance(i, copied);
afddba49
NP
2760 if (unlikely(copied == 0)) {
2761 /*
2762 * If we were unable to copy any data at all, we must
2763 * fall back to a single segment length write.
2764 *
2765 * If we didn't fallback here, we could livelock
2766 * because not all segments in the iov can be copied at
2767 * once without a pagefault.
2768 */
09cbfeaf 2769 bytes = min_t(unsigned long, PAGE_SIZE - offset,
afddba49
NP
2770 iov_iter_single_seg_count(i));
2771 goto again;
2772 }
afddba49
NP
2773 pos += copied;
2774 written += copied;
2775
2776 balance_dirty_pages_ratelimited(mapping);
afddba49
NP
2777 } while (iov_iter_count(i));
2778
2779 return written ? written : status;
2780}
3b93f911 2781EXPORT_SYMBOL(generic_perform_write);
1da177e4 2782
e4dd9de3 2783/**
8174202b 2784 * __generic_file_write_iter - write data to a file
e4dd9de3 2785 * @iocb: IO state structure (file, offset, etc.)
8174202b 2786 * @from: iov_iter with data to write
e4dd9de3
JK
2787 *
2788 * This function does all the work needed for actually writing data to a
2789 * file. It does all basic checks, removes SUID from the file, updates
2790 * modification times and calls proper subroutines depending on whether we
2791 * do direct IO or a standard buffered write.
2792 *
2793 * It expects i_mutex to be grabbed unless we work on a block device or similar
2794 * object which does not need locking at all.
2795 *
2796 * This function does *not* take care of syncing data in case of O_SYNC write.
2797 * A caller has to handle it. This is mainly due to the fact that we want to
2798 * avoid syncing under i_mutex.
2799 */
8174202b 2800ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1da177e4
LT
2801{
2802 struct file *file = iocb->ki_filp;
fb5527e6 2803 struct address_space * mapping = file->f_mapping;
1da177e4 2804 struct inode *inode = mapping->host;
3b93f911 2805 ssize_t written = 0;
1da177e4 2806 ssize_t err;
3b93f911 2807 ssize_t status;
1da177e4 2808
1da177e4 2809 /* We can write back this queue in page reclaim */
de1414a6 2810 current->backing_dev_info = inode_to_bdi(inode);
5fa8e0a1 2811 err = file_remove_privs(file);
1da177e4
LT
2812 if (err)
2813 goto out;
2814
c3b2da31
JB
2815 err = file_update_time(file);
2816 if (err)
2817 goto out;
1da177e4 2818
2ba48ce5 2819 if (iocb->ki_flags & IOCB_DIRECT) {
0b8def9d 2820 loff_t pos, endbyte;
fb5527e6 2821
1af5bb49 2822 written = generic_file_direct_write(iocb, from);
1da177e4 2823 /*
fbbbad4b
MW
2824 * If the write stopped short of completing, fall back to
2825 * buffered writes. Some filesystems do this for writes to
2826 * holes, for example. For DAX files, a buffered write will
2827 * not succeed (even if it did, DAX does not handle dirty
2828 * page-cache pages correctly).
1da177e4 2829 */
0b8def9d 2830 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
fbbbad4b
MW
2831 goto out;
2832
0b8def9d 2833 status = generic_perform_write(file, from, pos = iocb->ki_pos);
fb5527e6 2834 /*
3b93f911 2835 * If generic_perform_write() returned a synchronous error
fb5527e6
JM
2836 * then we want to return the number of bytes which were
2837 * direct-written, or the error code if that was zero. Note
2838 * that this differs from normal direct-io semantics, which
2839 * will return -EFOO even if some bytes were written.
2840 */
60bb4529 2841 if (unlikely(status < 0)) {
3b93f911 2842 err = status;
fb5527e6
JM
2843 goto out;
2844 }
fb5527e6
JM
2845 /*
2846 * We need to ensure that the page cache pages are written to
2847 * disk and invalidated to preserve the expected O_DIRECT
2848 * semantics.
2849 */
3b93f911 2850 endbyte = pos + status - 1;
0b8def9d 2851 err = filemap_write_and_wait_range(mapping, pos, endbyte);
fb5527e6 2852 if (err == 0) {
0b8def9d 2853 iocb->ki_pos = endbyte + 1;
3b93f911 2854 written += status;
fb5527e6 2855 invalidate_mapping_pages(mapping,
09cbfeaf
KS
2856 pos >> PAGE_SHIFT,
2857 endbyte >> PAGE_SHIFT);
fb5527e6
JM
2858 } else {
2859 /*
2860 * We don't know how much we wrote, so just return
2861 * the number of bytes which were direct-written
2862 */
2863 }
2864 } else {
0b8def9d
AV
2865 written = generic_perform_write(file, from, iocb->ki_pos);
2866 if (likely(written > 0))
2867 iocb->ki_pos += written;
fb5527e6 2868 }
1da177e4
LT
2869out:
2870 current->backing_dev_info = NULL;
2871 return written ? written : err;
2872}
8174202b 2873EXPORT_SYMBOL(__generic_file_write_iter);
e4dd9de3 2874
e4dd9de3 2875/**
8174202b 2876 * generic_file_write_iter - write data to a file
e4dd9de3 2877 * @iocb: IO state structure
8174202b 2878 * @from: iov_iter with data to write
e4dd9de3 2879 *
8174202b 2880 * This is a wrapper around __generic_file_write_iter() to be used by most
e4dd9de3
JK
2881 * filesystems. It takes care of syncing the file in case of O_SYNC file
2882 * and acquires i_mutex as needed.
2883 */
8174202b 2884ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1da177e4
LT
2885{
2886 struct file *file = iocb->ki_filp;
148f948b 2887 struct inode *inode = file->f_mapping->host;
1da177e4 2888 ssize_t ret;
1da177e4 2889
5955102c 2890 inode_lock(inode);
3309dd04
AV
2891 ret = generic_write_checks(iocb, from);
2892 if (ret > 0)
5f380c7f 2893 ret = __generic_file_write_iter(iocb, from);
5955102c 2894 inode_unlock(inode);
1da177e4 2895
e2592217
CH
2896 if (ret > 0)
2897 ret = generic_write_sync(iocb, ret);
1da177e4
LT
2898 return ret;
2899}
8174202b 2900EXPORT_SYMBOL(generic_file_write_iter);
1da177e4 2901
cf9a2ae8
DH
2902/**
2903 * try_to_release_page() - release old fs-specific metadata on a page
2904 *
2905 * @page: the page which the kernel is trying to free
2906 * @gfp_mask: memory allocation flags (and I/O mode)
2907 *
2908 * The address_space is to try to release any data against the page
2909 * (presumably at page->private). If the release was successful, return `1'.
2910 * Otherwise return zero.
2911 *
266cf658
DH
2912 * This may also be called if PG_fscache is set on a page, indicating that the
2913 * page is known to the local caching routines.
2914 *
cf9a2ae8 2915 * The @gfp_mask argument specifies whether I/O may be performed to release
71baba4b 2916 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
cf9a2ae8 2917 *
cf9a2ae8
DH
2918 */
2919int try_to_release_page(struct page *page, gfp_t gfp_mask)
2920{
2921 struct address_space * const mapping = page->mapping;
2922
2923 BUG_ON(!PageLocked(page));
2924 if (PageWriteback(page))
2925 return 0;
2926
2927 if (mapping && mapping->a_ops->releasepage)
2928 return mapping->a_ops->releasepage(page, gfp_mask);
2929 return try_to_free_buffers(page);
2930}
2931
2932EXPORT_SYMBOL(try_to_release_page);