blob: c3811bc6b9e33ae88d2575d0387c810b53b95fb8 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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 */
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040012#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/compiler.h>
14#include <linux/fs.h>
Hiro Yoshiokac22ce142006-06-23 02:04:16 -070015#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/aio.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080017#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/kernel_stat.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090019#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#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>
Linus Torvalds53253382007-10-18 14:47:32 -070028#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/pagevec.h>
30#include <linux/blkdev.h>
31#include <linux/security.h>
32#include <linux/syscalls.h>
Paul Jackson44110fe2006-03-24 03:16:04 -080033#include <linux/cpuset.h>
Nick Piggin2f718ff2007-10-16 01:24:59 -070034#include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080035#include <linux/memcontrol.h>
Dan Magenheimerc515e1f2011-05-26 10:01:43 -060036#include <linux/cleancache.h>
Nick Piggin0f8053a2006-03-22 00:08:33 -080037#include "internal.h"
38
Linus Torvalds1da177e2005-04-16 15:20:36 -070039/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070040 * FIXME: remove all knowledge of the buffer layer from the core VM
41 */
Jan Kara148f9482009-08-17 19:52:36 +020042#include <linux/buffer_head.h> /* for try_to_free_buffers */
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include <asm/mman.h>
45
46/*
47 * Shared mappings implemented 30.11.1994. It's not fully working yet,
48 * though.
49 *
50 * Shared mappings now work. 15.8.1995 Bruno.
51 *
52 * finished 'unifying' the page and buffer cache and SMP-threaded the
53 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
54 *
55 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
56 */
57
58/*
59 * Lock ordering:
60 *
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070061 * ->i_mmap_mutex (truncate_pagecache)
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 * ->private_lock (__free_pte->__set_page_dirty_buffers)
Hugh Dickins5d337b92005-09-03 15:54:41 -070063 * ->swap_lock (exclusive_swap_page, others)
64 * ->mapping->tree_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080066 * ->i_mutex
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070067 * ->i_mmap_mutex (truncate->unmap_mapping_range)
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 *
69 * ->mmap_sem
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070070 * ->i_mmap_mutex
Hugh Dickinsb8072f02005-10-29 18:16:41 -070071 * ->page_table_lock or pte_lock (various, mainly in memory.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
73 *
74 * ->mmap_sem
75 * ->lock_page (access_process_vm)
76 *
Nick Piggin82591e62006-10-19 23:29:10 -070077 * ->i_mutex (generic_file_buffered_write)
78 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 *
Christoph Hellwigf758eea2011-04-21 18:19:44 -060080 * bdi->wb.list_lock
Dave Chinnera66979a2011-03-22 22:23:41 +110081 * sb_lock (fs/fs-writeback.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 * ->mapping->tree_lock (__sync_single_inode)
83 *
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070084 * ->i_mmap_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -070085 * ->anon_vma.lock (vma_adjust)
86 *
87 * ->anon_vma.lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070088 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -070090 * ->page_table_lock or pte_lock
Hugh Dickins5d337b92005-09-03 15:54:41 -070091 * ->swap_lock (try_to_unmap_one)
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 * ->private_lock (try_to_unmap_one)
93 * ->tree_lock (try_to_unmap_one)
94 * ->zone.lru_lock (follow_page->mark_page_accessed)
Nick Piggin053837f2006-01-18 17:42:27 -080095 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 * ->private_lock (page_remove_rmap->set_page_dirty)
97 * ->tree_lock (page_remove_rmap->set_page_dirty)
Christoph Hellwigf758eea2011-04-21 18:19:44 -060098 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +110099 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600100 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100101 * ->inode->i_lock (zap_pte_range->set_page_dirty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
103 *
Andi Kleen9a3c5312012-03-21 16:34:09 -0700104 * ->i_mmap_mutex
105 * ->tasklist_lock (memory_failure, collect_procs_ao)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 */
107
108/*
Minchan Kime64a7822011-03-22 16:32:44 -0700109 * Delete a page from the page cache and free it. Caller has to make
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 * sure the page is locked and that nobody else uses it - or that usage
Nick Piggin19fd6232008-07-25 19:45:32 -0700111 * is safe. The caller must hold the mapping's tree_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 */
Minchan Kime64a7822011-03-22 16:32:44 -0700113void __delete_from_page_cache(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114{
115 struct address_space *mapping = page->mapping;
116
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600117 /*
118 * if we're uptodate, flush out into the cleancache, otherwise
119 * invalidate any existing cleancache entries. We can't leave
120 * stale data around in the cleancache once our page is gone
121 */
122 if (PageUptodate(page) && PageMappedToDisk(page))
123 cleancache_put_page(page);
124 else
Dan Magenheimer31677602011-09-21 11:56:28 -0400125 cleancache_invalidate_page(mapping, page);
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600126
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 radix_tree_delete(&mapping->page_tree, page->index);
128 page->mapping = NULL;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700129 /* Leave page->index set: truncation lookup relies upon it */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 mapping->nrpages--;
Christoph Lameter347ce432006-06-30 01:55:35 -0700131 __dec_zone_page_state(page, NR_FILE_PAGES);
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700132 if (PageSwapBacked(page))
133 __dec_zone_page_state(page, NR_SHMEM);
Nick Piggin45426812007-07-15 23:38:12 -0700134 BUG_ON(page_mapped(page));
Linus Torvalds3a692792007-12-19 14:05:13 -0800135
136 /*
137 * Some filesystems seem to re-dirty the page even after
138 * the VM has canceled the dirty bit (eg ext3 journaling).
139 *
140 * Fix it up by doing a final dirty accounting check after
141 * having removed the page entirely.
142 */
143 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
144 dec_zone_page_state(page, NR_FILE_DIRTY);
145 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
146 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147}
148
Minchan Kim702cfbf2011-03-22 16:32:43 -0700149/**
150 * delete_from_page_cache - delete page from page cache
151 * @page: the page which the kernel is trying to remove from page cache
152 *
153 * This must be called only on pages that have been verified to be in the page
154 * cache and locked. It will never put the page into the free list, the caller
155 * has a reference on the page.
156 */
157void delete_from_page_cache(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158{
159 struct address_space *mapping = page->mapping;
Linus Torvalds6072d132010-12-01 13:35:19 -0500160 void (*freepage)(struct page *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161
Matt Mackallcd7619d2005-05-01 08:59:01 -0700162 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163
Linus Torvalds6072d132010-12-01 13:35:19 -0500164 freepage = mapping->a_ops->freepage;
Nick Piggin19fd6232008-07-25 19:45:32 -0700165 spin_lock_irq(&mapping->tree_lock);
Minchan Kime64a7822011-03-22 16:32:44 -0700166 __delete_from_page_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700167 spin_unlock_irq(&mapping->tree_lock);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700168 mem_cgroup_uncharge_cache_page(page);
Linus Torvalds6072d132010-12-01 13:35:19 -0500169
170 if (freepage)
171 freepage(page);
Minchan Kim97cecb52011-03-22 16:30:53 -0700172 page_cache_release(page);
173}
174EXPORT_SYMBOL(delete_from_page_cache);
175
Jens Axboe7eaceac2011-03-10 08:52:07 +0100176static int sleep_on_page(void *word)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 io_schedule();
179 return 0;
180}
181
Jens Axboe7eaceac2011-03-10 08:52:07 +0100182static int sleep_on_page_killable(void *word)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500183{
Jens Axboe7eaceac2011-03-10 08:52:07 +0100184 sleep_on_page(word);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500185 return fatal_signal_pending(current) ? -EINTR : 0;
186}
187
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700189 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700190 * @mapping: address space structure to write
191 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800192 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700193 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700195 * Start writeback against all of a mapping's dirty pages that lie
196 * within the byte offsets <start, end> inclusive.
197 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700199 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 * these two operations is that if a dirty page/buffer is encountered, it must
201 * be waited upon, and not just skipped over.
202 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800203int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
204 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205{
206 int ret;
207 struct writeback_control wbc = {
208 .sync_mode = sync_mode,
Nick Piggin05fe4782009-01-06 14:39:08 -0800209 .nr_to_write = LONG_MAX,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700210 .range_start = start,
211 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 };
213
214 if (!mapping_cap_writeback_dirty(mapping))
215 return 0;
216
217 ret = do_writepages(mapping, &wbc);
218 return ret;
219}
220
221static inline int __filemap_fdatawrite(struct address_space *mapping,
222 int sync_mode)
223{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700224 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225}
226
227int filemap_fdatawrite(struct address_space *mapping)
228{
229 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
230}
231EXPORT_SYMBOL(filemap_fdatawrite);
232
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400233int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800234 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235{
236 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
237}
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400238EXPORT_SYMBOL(filemap_fdatawrite_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239
Randy Dunlap485bb992006-06-23 02:03:49 -0700240/**
241 * filemap_flush - mostly a non-blocking flush
242 * @mapping: target address_space
243 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 * This is a mostly non-blocking flush. Not suitable for data-integrity
245 * purposes - I/O may not be started against all dirty pages.
246 */
247int filemap_flush(struct address_space *mapping)
248{
249 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
250}
251EXPORT_SYMBOL(filemap_flush);
252
Randy Dunlap485bb992006-06-23 02:03:49 -0700253/**
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200254 * filemap_fdatawait_range - wait for writeback to complete
255 * @mapping: address space structure to wait for
256 * @start_byte: offset in bytes where the range starts
257 * @end_byte: offset in bytes where the range ends (inclusive)
Randy Dunlap485bb992006-06-23 02:03:49 -0700258 *
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200259 * Walk the list of under-writeback pages of the given address space
260 * in the given range and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 */
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200262int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
263 loff_t end_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264{
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200265 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
266 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 struct pagevec pvec;
268 int nr_pages;
269 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200271 if (end_byte < start_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 return 0;
273
274 pagevec_init(&pvec, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 while ((index <= end) &&
276 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
277 PAGECACHE_TAG_WRITEBACK,
278 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
279 unsigned i;
280
281 for (i = 0; i < nr_pages; i++) {
282 struct page *page = pvec.pages[i];
283
284 /* until radix tree lookup accepts end_index */
285 if (page->index > end)
286 continue;
287
288 wait_on_page_writeback(page);
Rik van Riel212260a2011-01-13 15:46:06 -0800289 if (TestClearPageError(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 ret = -EIO;
291 }
292 pagevec_release(&pvec);
293 cond_resched();
294 }
295
296 /* Check for outstanding write errors */
297 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
298 ret = -ENOSPC;
299 if (test_and_clear_bit(AS_EIO, &mapping->flags))
300 ret = -EIO;
301
302 return ret;
303}
Jan Karad3bccb6f2009-08-17 19:30:27 +0200304EXPORT_SYMBOL(filemap_fdatawait_range);
305
306/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700307 * filemap_fdatawait - wait for all under-writeback pages to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 * @mapping: address space structure to wait for
Randy Dunlap485bb992006-06-23 02:03:49 -0700309 *
310 * Walk the list of under-writeback pages of the given address space
311 * and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 */
313int filemap_fdatawait(struct address_space *mapping)
314{
315 loff_t i_size = i_size_read(mapping->host);
316
317 if (i_size == 0)
318 return 0;
319
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200320 return filemap_fdatawait_range(mapping, 0, i_size - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321}
322EXPORT_SYMBOL(filemap_fdatawait);
323
324int filemap_write_and_wait(struct address_space *mapping)
325{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800326 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327
328 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800329 err = filemap_fdatawrite(mapping);
330 /*
331 * Even if the above returned error, the pages may be
332 * written partially (e.g. -ENOSPC), so we wait for it.
333 * But the -EIO is special case, it may indicate the worst
334 * thing (e.g. bug) happened, so we avoid waiting for it.
335 */
336 if (err != -EIO) {
337 int err2 = filemap_fdatawait(mapping);
338 if (!err)
339 err = err2;
340 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800342 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800344EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345
Randy Dunlap485bb992006-06-23 02:03:49 -0700346/**
347 * filemap_write_and_wait_range - write out & wait on a file range
348 * @mapping: the address_space for the pages
349 * @lstart: offset in bytes where the range starts
350 * @lend: offset in bytes where the range ends (inclusive)
351 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800352 * Write out and wait upon file offsets lstart->lend, inclusive.
353 *
354 * Note that `lend' is inclusive (describes the last byte to be written) so
355 * that this function can be used to write to the very end-of-file (end = -1).
356 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357int filemap_write_and_wait_range(struct address_space *mapping,
358 loff_t lstart, loff_t lend)
359{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800360 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361
362 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800363 err = __filemap_fdatawrite_range(mapping, lstart, lend,
364 WB_SYNC_ALL);
365 /* See comment of filemap_write_and_wait() */
366 if (err != -EIO) {
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200367 int err2 = filemap_fdatawait_range(mapping,
368 lstart, lend);
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800369 if (!err)
370 err = err2;
371 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800373 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374}
Chris Masonf6995582009-04-15 13:22:37 -0400375EXPORT_SYMBOL(filemap_write_and_wait_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376
Randy Dunlap485bb992006-06-23 02:03:49 -0700377/**
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700378 * replace_page_cache_page - replace a pagecache page with a new one
379 * @old: page to be replaced
380 * @new: page to replace with
381 * @gfp_mask: allocation mode
382 *
383 * This function replaces a page in the pagecache with a new one. On
384 * success it acquires the pagecache reference for the new page and
385 * drops it for the old page. Both the old and new pages must be
386 * locked. This function does not add the new page to the LRU, the
387 * caller must do that.
388 *
389 * The remove + add is atomic. The only way this function can fail is
390 * memory allocation failure.
391 */
392int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
393{
394 int error;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700395
396 VM_BUG_ON(!PageLocked(old));
397 VM_BUG_ON(!PageLocked(new));
398 VM_BUG_ON(new->mapping);
399
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700400 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
401 if (!error) {
402 struct address_space *mapping = old->mapping;
403 void (*freepage)(struct page *);
404
405 pgoff_t offset = old->index;
406 freepage = mapping->a_ops->freepage;
407
408 page_cache_get(new);
409 new->mapping = mapping;
410 new->index = offset;
411
412 spin_lock_irq(&mapping->tree_lock);
Minchan Kime64a7822011-03-22 16:32:44 -0700413 __delete_from_page_cache(old);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700414 error = radix_tree_insert(&mapping->page_tree, offset, new);
415 BUG_ON(error);
416 mapping->nrpages++;
417 __inc_zone_page_state(new, NR_FILE_PAGES);
418 if (PageSwapBacked(new))
419 __inc_zone_page_state(new, NR_SHMEM);
420 spin_unlock_irq(&mapping->tree_lock);
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -0800421 /* mem_cgroup codes must not be called under tree_lock */
422 mem_cgroup_replace_page_cache(old, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700423 radix_tree_preload_end();
424 if (freepage)
425 freepage(old);
426 page_cache_release(old);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700427 }
428
429 return error;
430}
431EXPORT_SYMBOL_GPL(replace_page_cache_page);
432
433/**
Nick Piggine2867812008-07-25 19:45:30 -0700434 * add_to_page_cache_locked - add a locked page to the pagecache
Randy Dunlap485bb992006-06-23 02:03:49 -0700435 * @page: page to add
436 * @mapping: the page's address_space
437 * @offset: page index
438 * @gfp_mask: page allocation mode
439 *
Nick Piggine2867812008-07-25 19:45:30 -0700440 * This function is used to add a page to the pagecache. It must be locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 * This function does not add the page to the LRU. The caller must do that.
442 */
Nick Piggine2867812008-07-25 19:45:30 -0700443int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400444 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445{
Nick Piggine2867812008-07-25 19:45:30 -0700446 int error;
447
448 VM_BUG_ON(!PageLocked(page));
Hugh Dickins31475dd2011-08-03 16:21:27 -0700449 VM_BUG_ON(PageSwapBacked(page));
Nick Piggine2867812008-07-25 19:45:30 -0700450
451 error = mem_cgroup_cache_charge(page, current->mm,
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -0800452 gfp_mask & GFP_RECLAIM_MASK);
Balbir Singh35c754d2008-02-07 00:14:05 -0800453 if (error)
454 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455
Balbir Singh35c754d2008-02-07 00:14:05 -0800456 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 if (error == 0) {
Nick Piggine2867812008-07-25 19:45:30 -0700458 page_cache_get(page);
459 page->mapping = mapping;
460 page->index = offset;
461
Nick Piggin19fd6232008-07-25 19:45:32 -0700462 spin_lock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 error = radix_tree_insert(&mapping->page_tree, offset, page);
Nick Piggine2867812008-07-25 19:45:30 -0700464 if (likely(!error)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 mapping->nrpages++;
Christoph Lameter347ce432006-06-30 01:55:35 -0700466 __inc_zone_page_state(page, NR_FILE_PAGES);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700467 spin_unlock_irq(&mapping->tree_lock);
Nick Piggine2867812008-07-25 19:45:30 -0700468 } else {
469 page->mapping = NULL;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700470 /* Leave page->index set: truncation relies upon it */
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700471 spin_unlock_irq(&mapping->tree_lock);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700472 mem_cgroup_uncharge_cache_page(page);
Nick Piggine2867812008-07-25 19:45:30 -0700473 page_cache_release(page);
474 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 radix_tree_preload_end();
Balbir Singh35c754d2008-02-07 00:14:05 -0800476 } else
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700477 mem_cgroup_uncharge_cache_page(page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800478out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 return error;
480}
Nick Piggine2867812008-07-25 19:45:30 -0700481EXPORT_SYMBOL(add_to_page_cache_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482
483int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400484 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485{
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700486 int ret;
487
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700488 ret = add_to_page_cache(page, mapping, offset, gfp_mask);
Hugh Dickins31475dd2011-08-03 16:21:27 -0700489 if (ret == 0)
490 lru_cache_add_file(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 return ret;
492}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300493EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494
Paul Jackson44110fe2006-03-24 03:16:04 -0800495#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700496struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800497{
Miao Xiec0ff7452010-05-24 14:32:08 -0700498 int n;
499 struct page *page;
500
Paul Jackson44110fe2006-03-24 03:16:04 -0800501 if (cpuset_do_page_mem_spread()) {
Mel Gormancc9a6c82012-03-21 16:34:11 -0700502 unsigned int cpuset_mems_cookie;
503 do {
504 cpuset_mems_cookie = get_mems_allowed();
505 n = cpuset_mem_spread_node();
506 page = alloc_pages_exact_node(n, gfp, 0);
507 } while (!put_mems_allowed(cpuset_mems_cookie) && !page);
508
Miao Xiec0ff7452010-05-24 14:32:08 -0700509 return page;
Paul Jackson44110fe2006-03-24 03:16:04 -0800510 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700511 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800512}
Nick Piggin2ae88142006-10-28 10:38:23 -0700513EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800514#endif
515
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516/*
517 * In order to wait for pages to become available there must be
518 * waitqueues associated with pages. By using a hash table of
519 * waitqueues where the bucket discipline is to maintain all
520 * waiters on the same queue and wake all when any of the pages
521 * become available, and for the woken contexts to check to be
522 * sure the appropriate page became available, this saves space
523 * at a cost of "thundering herd" phenomena during rare hash
524 * collisions.
525 */
526static wait_queue_head_t *page_waitqueue(struct page *page)
527{
528 const struct zone *zone = page_zone(page);
529
530 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
531}
532
533static inline void wake_up_page(struct page *page, int bit)
534{
535 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
536}
537
Harvey Harrison920c7a52008-02-04 22:29:26 -0800538void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539{
540 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
541
542 if (test_bit(bit_nr, &page->flags))
Jens Axboe7eaceac2011-03-10 08:52:07 +0100543 __wait_on_bit(page_waitqueue(page), &wait, sleep_on_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 TASK_UNINTERRUPTIBLE);
545}
546EXPORT_SYMBOL(wait_on_page_bit);
547
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700548int wait_on_page_bit_killable(struct page *page, int bit_nr)
549{
550 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
551
552 if (!test_bit(bit_nr, &page->flags))
553 return 0;
554
555 return __wait_on_bit(page_waitqueue(page), &wait,
556 sleep_on_page_killable, TASK_KILLABLE);
557}
558
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559/**
David Howells385e1ca5f2009-04-03 16:42:39 +0100560 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
Randy Dunlap697f6192009-04-13 14:39:54 -0700561 * @page: Page defining the wait queue of interest
562 * @waiter: Waiter to add to the queue
David Howells385e1ca5f2009-04-03 16:42:39 +0100563 *
564 * Add an arbitrary @waiter to the wait queue for the nominated @page.
565 */
566void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
567{
568 wait_queue_head_t *q = page_waitqueue(page);
569 unsigned long flags;
570
571 spin_lock_irqsave(&q->lock, flags);
572 __add_wait_queue(q, waiter);
573 spin_unlock_irqrestore(&q->lock, flags);
574}
575EXPORT_SYMBOL_GPL(add_page_wait_queue);
576
577/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700578 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 * @page: the page
580 *
581 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
582 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
583 * mechananism between PageLocked pages and PageWriteback pages is shared.
584 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
585 *
Nick Piggin8413ac92008-10-18 20:26:59 -0700586 * The mb is necessary to enforce ordering between the clear_bit and the read
587 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800589void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590{
Nick Piggin8413ac92008-10-18 20:26:59 -0700591 VM_BUG_ON(!PageLocked(page));
592 clear_bit_unlock(PG_locked, &page->flags);
593 smp_mb__after_clear_bit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 wake_up_page(page, PG_locked);
595}
596EXPORT_SYMBOL(unlock_page);
597
Randy Dunlap485bb992006-06-23 02:03:49 -0700598/**
599 * end_page_writeback - end writeback against a page
600 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 */
602void end_page_writeback(struct page *page)
603{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700604 if (TestClearPageReclaim(page))
605 rotate_reclaimable_page(page);
606
607 if (!test_clear_page_writeback(page))
608 BUG();
609
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 smp_mb__after_clear_bit();
611 wake_up_page(page, PG_writeback);
612}
613EXPORT_SYMBOL(end_page_writeback);
614
Randy Dunlap485bb992006-06-23 02:03:49 -0700615/**
616 * __lock_page - get a lock on the page, assuming we need to sleep to get it
617 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800619void __lock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620{
621 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
622
Jens Axboe7eaceac2011-03-10 08:52:07 +0100623 __wait_on_bit_lock(page_waitqueue(page), &wait, sleep_on_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 TASK_UNINTERRUPTIBLE);
625}
626EXPORT_SYMBOL(__lock_page);
627
Harvey Harrisonb5606c22008-02-13 15:03:16 -0800628int __lock_page_killable(struct page *page)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500629{
630 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
631
632 return __wait_on_bit_lock(page_waitqueue(page), &wait,
Jens Axboe7eaceac2011-03-10 08:52:07 +0100633 sleep_on_page_killable, TASK_KILLABLE);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500634}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300635EXPORT_SYMBOL_GPL(__lock_page_killable);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500636
Michel Lespinassed065bd82010-10-26 14:21:57 -0700637int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
638 unsigned int flags)
639{
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700640 if (flags & FAULT_FLAG_ALLOW_RETRY) {
641 /*
642 * CAUTION! In this case, mmap_sem is not released
643 * even though return 0.
644 */
645 if (flags & FAULT_FLAG_RETRY_NOWAIT)
646 return 0;
647
648 up_read(&mm->mmap_sem);
649 if (flags & FAULT_FLAG_KILLABLE)
650 wait_on_page_locked_killable(page);
651 else
Gleb Natapov318b2752011-03-22 16:30:51 -0700652 wait_on_page_locked(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -0700653 return 0;
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700654 } else {
655 if (flags & FAULT_FLAG_KILLABLE) {
656 int ret;
657
658 ret = __lock_page_killable(page);
659 if (ret) {
660 up_read(&mm->mmap_sem);
661 return 0;
662 }
663 } else
664 __lock_page(page);
665 return 1;
Michel Lespinassed065bd82010-10-26 14:21:57 -0700666 }
667}
668
Randy Dunlap485bb992006-06-23 02:03:49 -0700669/**
670 * find_get_page - find and get a page reference
671 * @mapping: the address_space to search
672 * @offset: the page index
673 *
Nick Pigginda6052f2006-09-25 23:31:35 -0700674 * Is there a pagecache struct page at the given (mapping, offset) tuple?
675 * If yes, increment its refcount and return it; if no, return NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 */
Nick Piggina60637c2008-07-25 19:45:31 -0700677struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678{
Nick Piggina60637c2008-07-25 19:45:31 -0700679 void **pagep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 struct page *page;
681
Nick Piggina60637c2008-07-25 19:45:31 -0700682 rcu_read_lock();
683repeat:
684 page = NULL;
685 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
686 if (pagep) {
687 page = radix_tree_deref_slot(pagep);
Nick Piggin27d20fd2010-11-11 14:05:19 -0800688 if (unlikely(!page))
689 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700690 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700691 if (radix_tree_deref_retry(page))
692 goto repeat;
693 /*
694 * Otherwise, shmem/tmpfs must be storing a swap entry
695 * here as an exceptional entry: so return it without
696 * attempting to raise page count.
697 */
698 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700699 }
Nick Piggina60637c2008-07-25 19:45:31 -0700700 if (!page_cache_get_speculative(page))
701 goto repeat;
702
703 /*
704 * Has the page moved?
705 * This is part of the lockless pagecache protocol. See
706 * include/linux/pagemap.h for details.
707 */
708 if (unlikely(page != *pagep)) {
709 page_cache_release(page);
710 goto repeat;
711 }
712 }
Nick Piggin27d20fd2010-11-11 14:05:19 -0800713out:
Nick Piggina60637c2008-07-25 19:45:31 -0700714 rcu_read_unlock();
715
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 return page;
717}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718EXPORT_SYMBOL(find_get_page);
719
Randy Dunlap485bb992006-06-23 02:03:49 -0700720/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 * find_lock_page - locate, pin and lock a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700722 * @mapping: the address_space to search
723 * @offset: the page index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 *
725 * Locates the desired pagecache page, locks it, increments its reference
726 * count and returns its address.
727 *
728 * Returns zero if the page was not present. find_lock_page() may sleep.
729 */
Nick Piggina60637c2008-07-25 19:45:31 -0700730struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731{
732 struct page *page;
733
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734repeat:
Nick Piggina60637c2008-07-25 19:45:31 -0700735 page = find_get_page(mapping, offset);
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700736 if (page && !radix_tree_exception(page)) {
Nick Piggina60637c2008-07-25 19:45:31 -0700737 lock_page(page);
738 /* Has the page been truncated? */
739 if (unlikely(page->mapping != mapping)) {
740 unlock_page(page);
741 page_cache_release(page);
742 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 }
Nick Piggina60637c2008-07-25 19:45:31 -0700744 VM_BUG_ON(page->index != offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 return page;
747}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748EXPORT_SYMBOL(find_lock_page);
749
750/**
751 * find_or_create_page - locate or add a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700752 * @mapping: the page's address_space
753 * @index: the page's index into the mapping
754 * @gfp_mask: page allocation mode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 *
756 * Locates a page in the pagecache. If the page is not present, a new page
757 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
758 * LRU list. The returned page is locked and has its reference count
759 * incremented.
760 *
761 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
762 * allocation!
763 *
764 * find_or_create_page() returns the desired page's address, or zero on
765 * memory exhaustion.
766 */
767struct page *find_or_create_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -0700768 pgoff_t index, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769{
Nick Piggineb2be182007-10-16 01:24:57 -0700770 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 int err;
772repeat:
773 page = find_lock_page(mapping, index);
774 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -0700775 page = __page_cache_alloc(gfp_mask);
776 if (!page)
777 return NULL;
Nick Piggin67d58ac2009-01-06 14:40:28 -0800778 /*
779 * We want a regular kernel memory (not highmem or DMA etc)
780 * allocation for the radix tree nodes, but we need to honour
781 * the context-specific requirements the caller has asked for.
782 * GFP_RECLAIM_MASK collects those requirements.
783 */
784 err = add_to_page_cache_lru(page, mapping, index,
785 (gfp_mask & GFP_RECLAIM_MASK));
Nick Piggineb2be182007-10-16 01:24:57 -0700786 if (unlikely(err)) {
787 page_cache_release(page);
788 page = NULL;
789 if (err == -EEXIST)
790 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 return page;
794}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795EXPORT_SYMBOL(find_or_create_page);
796
797/**
798 * find_get_pages - gang pagecache lookup
799 * @mapping: The address_space to search
800 * @start: The starting page index
801 * @nr_pages: The maximum number of pages
802 * @pages: Where the resulting pages are placed
803 *
804 * find_get_pages() will search for and return a group of up to
805 * @nr_pages pages in the mapping. The pages are placed at @pages.
806 * find_get_pages() takes a reference against the returned pages.
807 *
808 * The search returns a group of mapping-contiguous pages with ascending
809 * indexes. There may be holes in the indices due to not-present pages.
810 *
811 * find_get_pages() returns the number of pages which were found.
812 */
813unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
814 unsigned int nr_pages, struct page **pages)
815{
816 unsigned int i;
817 unsigned int ret;
Shaohua Licc39c6a2011-09-15 08:45:19 +0800818 unsigned int nr_found, nr_skip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819
Nick Piggina60637c2008-07-25 19:45:31 -0700820 rcu_read_lock();
821restart:
822 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
Hugh Dickins63286502011-08-03 16:21:18 -0700823 (void ***)pages, NULL, start, nr_pages);
Nick Piggina60637c2008-07-25 19:45:31 -0700824 ret = 0;
Shaohua Licc39c6a2011-09-15 08:45:19 +0800825 nr_skip = 0;
Nick Piggina60637c2008-07-25 19:45:31 -0700826 for (i = 0; i < nr_found; i++) {
827 struct page *page;
828repeat:
829 page = radix_tree_deref_slot((void **)pages[i]);
830 if (unlikely(!page))
831 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -0700832
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700833 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700834 if (radix_tree_deref_retry(page)) {
835 /*
836 * Transient condition which can only trigger
837 * when entry at index 0 moves out of or back
838 * to root: none yet gotten, safe to restart.
839 */
840 WARN_ON(start | i);
841 goto restart;
842 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700843 /*
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700844 * Otherwise, shmem/tmpfs must be storing a swap entry
845 * here as an exceptional entry: so skip over it -
846 * we only reach this from invalidate_mapping_pages().
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700847 */
Shaohua Licc39c6a2011-09-15 08:45:19 +0800848 nr_skip++;
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700849 continue;
Nick Piggin27d20fd2010-11-11 14:05:19 -0800850 }
Nick Piggina60637c2008-07-25 19:45:31 -0700851
852 if (!page_cache_get_speculative(page))
853 goto repeat;
854
855 /* Has the page moved? */
856 if (unlikely(page != *((void **)pages[i]))) {
857 page_cache_release(page);
858 goto repeat;
859 }
860
861 pages[ret] = page;
862 ret++;
863 }
Hugh Dickins5b280c02011-03-22 16:33:07 -0700864
865 /*
866 * If all entries were removed before we could secure them,
867 * try again, because callers stop trying once 0 is returned.
868 */
Shaohua Licc39c6a2011-09-15 08:45:19 +0800869 if (unlikely(!ret && nr_found > nr_skip))
Hugh Dickins5b280c02011-03-22 16:33:07 -0700870 goto restart;
Nick Piggina60637c2008-07-25 19:45:31 -0700871 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 return ret;
873}
874
Jens Axboeebf43502006-04-27 08:46:01 +0200875/**
876 * find_get_pages_contig - gang contiguous pagecache lookup
877 * @mapping: The address_space to search
878 * @index: The starting page index
879 * @nr_pages: The maximum number of pages
880 * @pages: Where the resulting pages are placed
881 *
882 * find_get_pages_contig() works exactly like find_get_pages(), except
883 * that the returned number of pages are guaranteed to be contiguous.
884 *
885 * find_get_pages_contig() returns the number of pages which were found.
886 */
887unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
888 unsigned int nr_pages, struct page **pages)
889{
890 unsigned int i;
891 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700892 unsigned int nr_found;
Jens Axboeebf43502006-04-27 08:46:01 +0200893
Nick Piggina60637c2008-07-25 19:45:31 -0700894 rcu_read_lock();
895restart:
896 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
Hugh Dickins63286502011-08-03 16:21:18 -0700897 (void ***)pages, NULL, index, nr_pages);
Nick Piggina60637c2008-07-25 19:45:31 -0700898 ret = 0;
899 for (i = 0; i < nr_found; i++) {
900 struct page *page;
901repeat:
902 page = radix_tree_deref_slot((void **)pages[i]);
903 if (unlikely(!page))
904 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -0700905
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700906 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700907 if (radix_tree_deref_retry(page)) {
908 /*
909 * Transient condition which can only trigger
910 * when entry at index 0 moves out of or back
911 * to root: none yet gotten, safe to restart.
912 */
913 goto restart;
914 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700915 /*
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700916 * Otherwise, shmem/tmpfs must be storing a swap entry
917 * here as an exceptional entry: so stop looking for
918 * contiguous pages.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700919 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700920 break;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700921 }
Nick Piggina60637c2008-07-25 19:45:31 -0700922
Nick Piggina60637c2008-07-25 19:45:31 -0700923 if (!page_cache_get_speculative(page))
924 goto repeat;
925
926 /* Has the page moved? */
927 if (unlikely(page != *((void **)pages[i]))) {
928 page_cache_release(page);
929 goto repeat;
930 }
931
Nick Piggin9cbb4cb2011-01-13 15:45:51 -0800932 /*
933 * must check mapping and index after taking the ref.
934 * otherwise we can get both false positives and false
935 * negatives, which is just confusing to the caller.
936 */
937 if (page->mapping == NULL || page->index != index) {
938 page_cache_release(page);
939 break;
940 }
941
Nick Piggina60637c2008-07-25 19:45:31 -0700942 pages[ret] = page;
943 ret++;
Jens Axboeebf43502006-04-27 08:46:01 +0200944 index++;
945 }
Nick Piggina60637c2008-07-25 19:45:31 -0700946 rcu_read_unlock();
947 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +0200948}
David Howellsef71c152007-05-09 02:33:44 -0700949EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +0200950
Randy Dunlap485bb992006-06-23 02:03:49 -0700951/**
952 * find_get_pages_tag - find and return pages that match @tag
953 * @mapping: the address_space to search
954 * @index: the starting page index
955 * @tag: the tag index
956 * @nr_pages: the maximum number of pages
957 * @pages: where the resulting pages are placed
958 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -0700960 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 */
962unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
963 int tag, unsigned int nr_pages, struct page **pages)
964{
965 unsigned int i;
966 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700967 unsigned int nr_found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968
Nick Piggina60637c2008-07-25 19:45:31 -0700969 rcu_read_lock();
970restart:
971 nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree,
972 (void ***)pages, *index, nr_pages, tag);
973 ret = 0;
974 for (i = 0; i < nr_found; i++) {
975 struct page *page;
976repeat:
977 page = radix_tree_deref_slot((void **)pages[i]);
978 if (unlikely(!page))
979 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -0700980
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700981 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700982 if (radix_tree_deref_retry(page)) {
983 /*
984 * Transient condition which can only trigger
985 * when entry at index 0 moves out of or back
986 * to root: none yet gotten, safe to restart.
987 */
988 goto restart;
989 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700990 /*
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700991 * This function is never used on a shmem/tmpfs
992 * mapping, so a swap entry won't be found here.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700993 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700994 BUG();
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700995 }
Nick Piggina60637c2008-07-25 19:45:31 -0700996
997 if (!page_cache_get_speculative(page))
998 goto repeat;
999
1000 /* Has the page moved? */
1001 if (unlikely(page != *((void **)pages[i]))) {
1002 page_cache_release(page);
1003 goto repeat;
1004 }
1005
1006 pages[ret] = page;
1007 ret++;
1008 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001009
1010 /*
1011 * If all entries were removed before we could secure them,
1012 * try again, because callers stop trying once 0 is returned.
1013 */
1014 if (unlikely(!ret && nr_found))
1015 goto restart;
Nick Piggina60637c2008-07-25 19:45:31 -07001016 rcu_read_unlock();
1017
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 if (ret)
1019 *index = pages[ret - 1]->index + 1;
Nick Piggina60637c2008-07-25 19:45:31 -07001020
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 return ret;
1022}
David Howellsef71c152007-05-09 02:33:44 -07001023EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024
Randy Dunlap485bb992006-06-23 02:03:49 -07001025/**
1026 * grab_cache_page_nowait - returns locked page at given index in given cache
1027 * @mapping: target address_space
1028 * @index: the page index
1029 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08001030 * Same as grab_cache_page(), but do not wait if the page is unavailable.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 * This is intended for speculative data generators, where the data can
1032 * be regenerated if the page couldn't be grabbed. This routine should
1033 * be safe to call while holding the lock for another page.
1034 *
1035 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
1036 * and deadlock against the caller's locked page.
1037 */
1038struct page *
Fengguang Wu57f6b962007-10-16 01:24:37 -07001039grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040{
1041 struct page *page = find_get_page(mapping, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042
1043 if (page) {
Nick Piggin529ae9a2008-08-02 12:01:03 +02001044 if (trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 return page;
1046 page_cache_release(page);
1047 return NULL;
1048 }
Nick Piggin2ae88142006-10-28 10:38:23 -07001049 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
Nick Piggin67d58ac2009-01-06 14:40:28 -08001050 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 page_cache_release(page);
1052 page = NULL;
1053 }
1054 return page;
1055}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056EXPORT_SYMBOL(grab_cache_page_nowait);
1057
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001058/*
1059 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1060 * a _large_ part of the i/o request. Imagine the worst scenario:
1061 *
1062 * ---R__________________________________________B__________
1063 * ^ reading here ^ bad block(assume 4k)
1064 *
1065 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1066 * => failing the whole request => read(R) => read(R+1) =>
1067 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1068 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1069 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1070 *
1071 * It is going insane. Fix it by quickly scaling down the readahead size.
1072 */
1073static void shrink_readahead_size_eio(struct file *filp,
1074 struct file_ra_state *ra)
1075{
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001076 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001077}
1078
Randy Dunlap485bb992006-06-23 02:03:49 -07001079/**
Christoph Hellwig36e78912008-02-08 04:21:24 -08001080 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001081 * @filp: the file to read
1082 * @ppos: current file position
1083 * @desc: read_descriptor
1084 * @actor: read method
1085 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -07001087 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 *
1089 * This is really ugly. But the goto's actually try to clarify some
1090 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 */
Christoph Hellwig36e78912008-02-08 04:21:24 -08001092static void do_generic_file_read(struct file *filp, loff_t *ppos,
1093 read_descriptor_t *desc, read_actor_t actor)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094{
Christoph Hellwig36e78912008-02-08 04:21:24 -08001095 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -08001097 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001098 pgoff_t index;
1099 pgoff_t last_index;
1100 pgoff_t prev_index;
1101 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -07001102 unsigned int prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 index = *ppos >> PAGE_CACHE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -07001106 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1107 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1109 offset = *ppos & ~PAGE_CACHE_MASK;
1110
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 for (;;) {
1112 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001113 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001114 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 unsigned long nr, ret;
1116
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118find_page:
1119 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001120 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001121 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001122 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001123 index, last_index - index);
1124 page = find_get_page(mapping, index);
1125 if (unlikely(page == NULL))
1126 goto no_cached_page;
1127 }
1128 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001129 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001130 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001131 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001133 if (!PageUptodate(page)) {
1134 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1135 !mapping->a_ops->is_partially_uptodate)
1136 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001137 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001138 goto page_not_up_to_date;
Dave Hansen8d056cb2010-11-11 14:05:15 -08001139 /* Did it get truncated before we got the lock? */
1140 if (!page->mapping)
1141 goto page_not_up_to_date_locked;
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001142 if (!mapping->a_ops->is_partially_uptodate(page,
1143 desc, offset))
1144 goto page_not_up_to_date_locked;
1145 unlock_page(page);
1146 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001148 /*
1149 * i_size must be checked after we know the page is Uptodate.
1150 *
1151 * Checking i_size after the check allows us to calculate
1152 * the correct value for "nr", which means the zero-filled
1153 * part of the page is not copied back to userspace (unless
1154 * another truncate extends the file - this is desired though).
1155 */
1156
1157 isize = i_size_read(inode);
1158 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1159 if (unlikely(!isize || index > end_index)) {
1160 page_cache_release(page);
1161 goto out;
1162 }
1163
1164 /* nr is the maximum number of bytes to copy from this page */
1165 nr = PAGE_CACHE_SIZE;
1166 if (index == end_index) {
1167 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1168 if (nr <= offset) {
1169 page_cache_release(page);
1170 goto out;
1171 }
1172 }
1173 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174
1175 /* If users can be writing to this page using arbitrary
1176 * virtual addresses, take care about potential aliasing
1177 * before reading the page on the kernel side.
1178 */
1179 if (mapping_writably_mapped(mapping))
1180 flush_dcache_page(page);
1181
1182 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001183 * When a sequential read accesses a page several times,
1184 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 */
Jan Karaec0f1632007-05-06 14:49:25 -07001186 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 mark_page_accessed(page);
1188 prev_index = index;
1189
1190 /*
1191 * Ok, we have the page, and it's up-to-date, so
1192 * now we can copy it to user space...
1193 *
1194 * The actor routine returns how many bytes were actually used..
1195 * NOTE! This may not be the same as how much of a user buffer
1196 * we filled up (we may be padding etc), so we can only update
1197 * "pos" here (the actor routine has to update the user buffer
1198 * pointers and the remaining count).
1199 */
1200 ret = actor(desc, page, offset, nr);
1201 offset += ret;
1202 index += offset >> PAGE_CACHE_SHIFT;
1203 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001204 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205
1206 page_cache_release(page);
1207 if (ret == nr && desc->count)
1208 continue;
1209 goto out;
1210
1211page_not_up_to_date:
1212 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001213 error = lock_page_killable(page);
1214 if (unlikely(error))
1215 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001217page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001218 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 if (!page->mapping) {
1220 unlock_page(page);
1221 page_cache_release(page);
1222 continue;
1223 }
1224
1225 /* Did somebody else fill it already? */
1226 if (PageUptodate(page)) {
1227 unlock_page(page);
1228 goto page_ok;
1229 }
1230
1231readpage:
Jeff Moyer91803b42010-05-26 11:49:40 -04001232 /*
1233 * A previous I/O error may have been due to temporary
1234 * failures, eg. multipath errors.
1235 * PG_error will be set again if readpage fails.
1236 */
1237 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 /* Start the actual read. The read will unlock the page. */
1239 error = mapping->a_ops->readpage(filp, page);
1240
Zach Brown994fc28c2005-12-15 14:28:17 -08001241 if (unlikely(error)) {
1242 if (error == AOP_TRUNCATED_PAGE) {
1243 page_cache_release(page);
1244 goto find_page;
1245 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001247 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248
1249 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001250 error = lock_page_killable(page);
1251 if (unlikely(error))
1252 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 if (!PageUptodate(page)) {
1254 if (page->mapping == NULL) {
1255 /*
Christoph Hellwig2ecdc822010-01-26 17:27:20 +01001256 * invalidate_mapping_pages got it
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 */
1258 unlock_page(page);
1259 page_cache_release(page);
1260 goto find_page;
1261 }
1262 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001263 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001264 error = -EIO;
1265 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 }
1267 unlock_page(page);
1268 }
1269
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 goto page_ok;
1271
1272readpage_error:
1273 /* UHHUH! A synchronous read error occurred. Report it */
1274 desc->error = error;
1275 page_cache_release(page);
1276 goto out;
1277
1278no_cached_page:
1279 /*
1280 * Ok, it wasn't cached, so we need to create a new
1281 * page..
1282 */
Nick Piggineb2be182007-10-16 01:24:57 -07001283 page = page_cache_alloc_cold(mapping);
1284 if (!page) {
1285 desc->error = -ENOMEM;
1286 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 }
Nick Piggineb2be182007-10-16 01:24:57 -07001288 error = add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 index, GFP_KERNEL);
1290 if (error) {
Nick Piggineb2be182007-10-16 01:24:57 -07001291 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 if (error == -EEXIST)
1293 goto find_page;
1294 desc->error = error;
1295 goto out;
1296 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 goto readpage;
1298 }
1299
1300out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001301 ra->prev_pos = prev_index;
1302 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1303 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001305 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07001306 file_accessed(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308
1309int file_read_actor(read_descriptor_t *desc, struct page *page,
1310 unsigned long offset, unsigned long size)
1311{
1312 char *kaddr;
1313 unsigned long left, count = desc->count;
1314
1315 if (size > count)
1316 size = count;
1317
1318 /*
1319 * Faults on the destination of a read are common, so do it before
1320 * taking the kmap.
1321 */
1322 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08001323 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 left = __copy_to_user_inatomic(desc->arg.buf,
1325 kaddr + offset, size);
Cong Wang9b04c5f2011-11-25 23:14:39 +08001326 kunmap_atomic(kaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 if (left == 0)
1328 goto success;
1329 }
1330
1331 /* Do it the slow way */
1332 kaddr = kmap(page);
1333 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1334 kunmap(page);
1335
1336 if (left) {
1337 size -= left;
1338 desc->error = -EFAULT;
1339 }
1340success:
1341 desc->count = count - size;
1342 desc->written += size;
1343 desc->arg.buf += size;
1344 return size;
1345}
1346
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001347/*
1348 * Performs necessary checks before doing a write
1349 * @iov: io vector request
1350 * @nr_segs: number of segments in the iovec
1351 * @count: number of bytes to write
1352 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1353 *
1354 * Adjust number of segments and amount of bytes to write (nr_segs should be
1355 * properly initialized first). Returns appropriate error code that caller
1356 * should return or zero in case that write should be allowed.
1357 */
1358int generic_segment_checks(const struct iovec *iov,
1359 unsigned long *nr_segs, size_t *count, int access_flags)
1360{
1361 unsigned long seg;
1362 size_t cnt = 0;
1363 for (seg = 0; seg < *nr_segs; seg++) {
1364 const struct iovec *iv = &iov[seg];
1365
1366 /*
1367 * If any segment has a negative length, or the cumulative
1368 * length ever wraps negative then return -EINVAL.
1369 */
1370 cnt += iv->iov_len;
1371 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1372 return -EINVAL;
1373 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1374 continue;
1375 if (seg == 0)
1376 return -EFAULT;
1377 *nr_segs = seg;
1378 cnt -= iv->iov_len; /* This segment is no good */
1379 break;
1380 }
1381 *count = cnt;
1382 return 0;
1383}
1384EXPORT_SYMBOL(generic_segment_checks);
1385
Randy Dunlap485bb992006-06-23 02:03:49 -07001386/**
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001387 * generic_file_aio_read - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001388 * @iocb: kernel I/O control block
1389 * @iov: io vector request
1390 * @nr_segs: number of segments in the iovec
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001391 * @pos: current file position
Randy Dunlap485bb992006-06-23 02:03:49 -07001392 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 * This is the "read()" routine for all filesystems
1394 * that can use the page cache directly.
1395 */
1396ssize_t
Badari Pulavarty543ade12006-09-30 23:28:48 -07001397generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1398 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399{
1400 struct file *filp = iocb->ki_filp;
1401 ssize_t retval;
Josef Bacik66f998f2010-05-23 11:00:54 -04001402 unsigned long seg = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 size_t count;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001404 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405
1406 count = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001407 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1408 if (retval)
1409 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410
1411 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1412 if (filp->f_flags & O_DIRECT) {
Badari Pulavarty543ade12006-09-30 23:28:48 -07001413 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 struct address_space *mapping;
1415 struct inode *inode;
1416
1417 mapping = filp->f_mapping;
1418 inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 if (!count)
1420 goto out; /* skip atime */
1421 size = i_size_read(inode);
1422 if (pos < size) {
Nick Piggin48b47c52009-01-06 14:40:22 -08001423 retval = filemap_write_and_wait_range(mapping, pos,
1424 pos + iov_length(iov, nr_segs) - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07001425 if (!retval) {
Shaohua Li3deaa712012-02-03 15:37:17 -08001426 struct blk_plug plug;
1427
1428 blk_start_plug(&plug);
Christoph Hellwiga969e902008-07-23 21:27:04 -07001429 retval = mapping->a_ops->direct_IO(READ, iocb,
1430 iov, pos, nr_segs);
Shaohua Li3deaa712012-02-03 15:37:17 -08001431 blk_finish_plug(&plug);
Christoph Hellwiga969e902008-07-23 21:27:04 -07001432 }
Josef Bacik66f998f2010-05-23 11:00:54 -04001433 if (retval > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 *ppos = pos + retval;
Josef Bacik66f998f2010-05-23 11:00:54 -04001435 count -= retval;
1436 }
1437
1438 /*
1439 * Btrfs can have a short DIO read if we encounter
1440 * compressed extents, so if there was an error, or if
1441 * we've already read everything we wanted to, or if
1442 * there was a short read because we hit EOF, go ahead
1443 * and return. Otherwise fallthrough to buffered io for
1444 * the rest of the read.
1445 */
1446 if (retval < 0 || !count || *ppos >= size) {
Hugh Dickins11fa9772008-07-23 21:27:34 -07001447 file_accessed(filp);
1448 goto out;
1449 }
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001450 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 }
1452
Josef Bacik66f998f2010-05-23 11:00:54 -04001453 count = retval;
Hugh Dickins11fa9772008-07-23 21:27:34 -07001454 for (seg = 0; seg < nr_segs; seg++) {
1455 read_descriptor_t desc;
Josef Bacik66f998f2010-05-23 11:00:54 -04001456 loff_t offset = 0;
1457
1458 /*
1459 * If we did a short DIO read we need to skip the section of the
1460 * iov that we've already read data into.
1461 */
1462 if (count) {
1463 if (count > iov[seg].iov_len) {
1464 count -= iov[seg].iov_len;
1465 continue;
1466 }
1467 offset = count;
1468 count = 0;
1469 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470
Hugh Dickins11fa9772008-07-23 21:27:34 -07001471 desc.written = 0;
Josef Bacik66f998f2010-05-23 11:00:54 -04001472 desc.arg.buf = iov[seg].iov_base + offset;
1473 desc.count = iov[seg].iov_len - offset;
Hugh Dickins11fa9772008-07-23 21:27:34 -07001474 if (desc.count == 0)
1475 continue;
1476 desc.error = 0;
1477 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1478 retval += desc.written;
1479 if (desc.error) {
1480 retval = retval ?: desc.error;
1481 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 }
Hugh Dickins11fa9772008-07-23 21:27:34 -07001483 if (desc.count > 0)
1484 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 }
1486out:
1487 return retval;
1488}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489EXPORT_SYMBOL(generic_file_aio_read);
1490
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491static ssize_t
1492do_readahead(struct address_space *mapping, struct file *filp,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001493 pgoff_t index, unsigned long nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494{
1495 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1496 return -EINVAL;
1497
Wu Fengguangf7e839d2009-06-16 15:31:20 -07001498 force_page_cache_readahead(mapping, filp, index, nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 return 0;
1500}
1501
Heiko Carstens6673e0c2009-01-14 14:14:02 +01001502SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503{
1504 ssize_t ret;
1505 struct file *file;
1506
1507 ret = -EBADF;
1508 file = fget(fd);
1509 if (file) {
1510 if (file->f_mode & FMODE_READ) {
1511 struct address_space *mapping = file->f_mapping;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001512 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1513 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 unsigned long len = end - start + 1;
1515 ret = do_readahead(mapping, file, start, len);
1516 }
1517 fput(file);
1518 }
1519 return ret;
1520}
Heiko Carstens6673e0c2009-01-14 14:14:02 +01001521#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1522asmlinkage long SyS_readahead(long fd, loff_t offset, long count)
1523{
1524 return SYSC_readahead((int) fd, offset, (size_t) count);
1525}
1526SYSCALL_ALIAS(sys_readahead, SyS_readahead);
1527#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528
1529#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001530/**
1531 * page_cache_read - adds requested page to the page cache if not already there
1532 * @file: file to read
1533 * @offset: page index
1534 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 * This adds the requested page to the page cache if it isn't already there,
1536 * and schedules an I/O to read in its contents from disk.
1537 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001538static int page_cache_read(struct file *file, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539{
1540 struct address_space *mapping = file->f_mapping;
1541 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001542 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543
Zach Brown994fc28c2005-12-15 14:28:17 -08001544 do {
1545 page = page_cache_alloc_cold(mapping);
1546 if (!page)
1547 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548
Zach Brown994fc28c2005-12-15 14:28:17 -08001549 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1550 if (ret == 0)
1551 ret = mapping->a_ops->readpage(file, page);
1552 else if (ret == -EEXIST)
1553 ret = 0; /* losing race to add is OK */
1554
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556
Zach Brown994fc28c2005-12-15 14:28:17 -08001557 } while (ret == AOP_TRUNCATED_PAGE);
1558
1559 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560}
1561
1562#define MMAP_LOTSAMISS (100)
1563
Linus Torvaldsef00e082009-06-16 15:31:25 -07001564/*
1565 * Synchronous readahead happens when we don't even find
1566 * a page in the page cache at all.
1567 */
1568static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1569 struct file_ra_state *ra,
1570 struct file *file,
1571 pgoff_t offset)
1572{
1573 unsigned long ra_pages;
1574 struct address_space *mapping = file->f_mapping;
1575
1576 /* If we don't want any read-ahead, don't bother */
1577 if (VM_RandomReadHint(vma))
1578 return;
Wu Fengguang275b12b2011-05-24 17:12:28 -07001579 if (!ra->ra_pages)
1580 return;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001581
Wu Fengguang2cbea1d2011-05-24 17:12:30 -07001582 if (VM_SequentialReadHint(vma)) {
Wu Fengguang7ffc59b2009-06-16 15:31:38 -07001583 page_cache_sync_readahead(mapping, ra, file, offset,
1584 ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001585 return;
1586 }
1587
Andi Kleen207d04b2011-05-24 17:12:29 -07001588 /* Avoid banging the cache line if not needed */
1589 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001590 ra->mmap_miss++;
1591
1592 /*
1593 * Do we miss much more than hit in this file? If so,
1594 * stop bothering with read-ahead. It will only hurt.
1595 */
1596 if (ra->mmap_miss > MMAP_LOTSAMISS)
1597 return;
1598
Wu Fengguangd30a1102009-06-16 15:31:30 -07001599 /*
1600 * mmap read-around
1601 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001602 ra_pages = max_sane_readahead(ra->ra_pages);
Wu Fengguang275b12b2011-05-24 17:12:28 -07001603 ra->start = max_t(long, 0, offset - ra_pages / 2);
1604 ra->size = ra_pages;
Wu Fengguang2cbea1d2011-05-24 17:12:30 -07001605 ra->async_size = ra_pages / 4;
Wu Fengguang275b12b2011-05-24 17:12:28 -07001606 ra_submit(ra, mapping, file);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001607}
1608
1609/*
1610 * Asynchronous readahead happens when we find the page and PG_readahead,
1611 * so we want to possibly extend the readahead further..
1612 */
1613static void do_async_mmap_readahead(struct vm_area_struct *vma,
1614 struct file_ra_state *ra,
1615 struct file *file,
1616 struct page *page,
1617 pgoff_t offset)
1618{
1619 struct address_space *mapping = file->f_mapping;
1620
1621 /* If we don't want any read-ahead, don't bother */
1622 if (VM_RandomReadHint(vma))
1623 return;
1624 if (ra->mmap_miss > 0)
1625 ra->mmap_miss--;
1626 if (PageReadahead(page))
Wu Fengguang2fad6f52009-06-16 15:31:29 -07001627 page_cache_async_readahead(mapping, ra, file,
1628 page, offset, ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001629}
1630
Randy Dunlap485bb992006-06-23 02:03:49 -07001631/**
Nick Piggin54cb8822007-07-19 01:46:59 -07001632 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07001633 * @vma: vma in which the fault was taken
1634 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07001635 *
Nick Piggin54cb8822007-07-19 01:46:59 -07001636 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 * mapped memory region to read in file data during a page fault.
1638 *
1639 * The goto's are kind of ugly, but this streamlines the normal case of having
1640 * it in the page cache, and handles the special cases reasonably without
1641 * having a lot of duplicated code.
1642 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001643int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644{
1645 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07001646 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 struct address_space *mapping = file->f_mapping;
1648 struct file_ra_state *ra = &file->f_ra;
1649 struct inode *inode = mapping->host;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001650 pgoff_t offset = vmf->pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 struct page *page;
Jan Kara2004dc82008-02-08 04:20:11 -08001652 pgoff_t size;
Nick Piggin83c54072007-07-19 01:47:05 -07001653 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001656 if (offset >= size)
Linus Torvalds5307cc12007-10-31 09:19:46 -07001657 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 * Do we have something in the page cache already?
1661 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001662 page = find_get_page(mapping, offset);
1663 if (likely(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07001665 * We found the page, so try async readahead before
1666 * waiting for the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001668 do_async_mmap_readahead(vma, ra, file, page, offset);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001669 } else {
1670 /* No page in the page cache at all */
1671 do_sync_mmap_readahead(vma, ra, file, offset);
1672 count_vm_event(PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07001673 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001674 ret = VM_FAULT_MAJOR;
1675retry_find:
Michel Lespinasseb522c942010-10-26 14:21:56 -07001676 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 if (!page)
1678 goto no_cached_page;
1679 }
1680
Michel Lespinassed88c0922010-11-02 13:05:18 -07001681 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1682 page_cache_release(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -07001683 return ret | VM_FAULT_RETRY;
Michel Lespinassed88c0922010-11-02 13:05:18 -07001684 }
Michel Lespinasseb522c942010-10-26 14:21:56 -07001685
1686 /* Did it get truncated? */
1687 if (unlikely(page->mapping != mapping)) {
1688 unlock_page(page);
1689 put_page(page);
1690 goto retry_find;
1691 }
1692 VM_BUG_ON(page->index != offset);
1693
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 /*
Nick Piggind00806b2007-07-19 01:46:57 -07001695 * We have a locked page in the page cache, now we need to check
1696 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 */
Nick Piggind00806b2007-07-19 01:46:57 -07001698 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 goto page_not_uptodate;
1700
Linus Torvaldsef00e082009-06-16 15:31:25 -07001701 /*
1702 * Found the page and have a reference on it.
1703 * We must recheck i_size under page lock.
1704 */
Nick Piggind00806b2007-07-19 01:46:57 -07001705 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001706 if (unlikely(offset >= size)) {
Nick Piggind00806b2007-07-19 01:46:57 -07001707 unlock_page(page);
Yan Zheng745ad482007-10-08 10:08:37 -07001708 page_cache_release(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07001709 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001710 }
1711
Nick Piggind0217ac2007-07-19 01:47:03 -07001712 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07001713 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715no_cached_page:
1716 /*
1717 * We're only likely to ever get here if MADV_RANDOM is in
1718 * effect.
1719 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001720 error = page_cache_read(file, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721
1722 /*
1723 * The page we want has now been added to the page cache.
1724 * In the unlikely event that someone removed it in the
1725 * meantime, we'll just come back here and read it again.
1726 */
1727 if (error >= 0)
1728 goto retry_find;
1729
1730 /*
1731 * An error return from page_cache_read can result if the
1732 * system is low on memory, or a problem occurs while trying
1733 * to schedule I/O.
1734 */
1735 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07001736 return VM_FAULT_OOM;
1737 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738
1739page_not_uptodate:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 /*
1741 * Umm, take care of errors if the page isn't up-to-date.
1742 * Try to re-read it _once_. We do this synchronously,
1743 * because there really aren't any performance issues here
1744 * and we need to check for errors.
1745 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001747 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07001748 if (!error) {
1749 wait_on_page_locked(page);
1750 if (!PageUptodate(page))
1751 error = -EIO;
1752 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07001754
1755 if (!error || error == AOP_TRUNCATED_PAGE)
1756 goto retry_find;
1757
1758 /* Things didn't work out. Return zero to tell the mm layer so. */
1759 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07001760 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07001761}
1762EXPORT_SYMBOL(filemap_fault);
1763
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001764const struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07001765 .fault = filemap_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766};
1767
1768/* This is used for a general mmap of a disk file */
1769
1770int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1771{
1772 struct address_space *mapping = file->f_mapping;
1773
1774 if (!mapping->a_ops->readpage)
1775 return -ENOEXEC;
1776 file_accessed(file);
1777 vma->vm_ops = &generic_file_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001778 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 return 0;
1780}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781
1782/*
1783 * This is for filesystems which do not implement ->writepage.
1784 */
1785int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1786{
1787 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1788 return -EINVAL;
1789 return generic_file_mmap(file, vma);
1790}
1791#else
1792int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1793{
1794 return -ENOSYS;
1795}
1796int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1797{
1798 return -ENOSYS;
1799}
1800#endif /* CONFIG_MMU */
1801
1802EXPORT_SYMBOL(generic_file_mmap);
1803EXPORT_SYMBOL(generic_file_readonly_mmap);
1804
Nick Piggin6fe69002007-05-06 14:49:04 -07001805static struct page *__read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001806 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001807 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001808 void *data,
1809 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810{
Nick Piggineb2be182007-10-16 01:24:57 -07001811 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 int err;
1813repeat:
1814 page = find_get_page(mapping, index);
1815 if (!page) {
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001816 page = __page_cache_alloc(gfp | __GFP_COLD);
Nick Piggineb2be182007-10-16 01:24:57 -07001817 if (!page)
1818 return ERR_PTR(-ENOMEM);
Dave Kleikampe6f67b82011-12-21 11:05:48 -06001819 err = add_to_page_cache_lru(page, mapping, index, gfp);
Nick Piggineb2be182007-10-16 01:24:57 -07001820 if (unlikely(err)) {
1821 page_cache_release(page);
1822 if (err == -EEXIST)
1823 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 return ERR_PTR(err);
1826 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 err = filler(data, page);
1828 if (err < 0) {
1829 page_cache_release(page);
1830 page = ERR_PTR(err);
1831 }
1832 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833 return page;
1834}
1835
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001836static struct page *do_read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001837 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001838 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001839 void *data,
1840 gfp_t gfp)
1841
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842{
1843 struct page *page;
1844 int err;
1845
1846retry:
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001847 page = __read_cache_page(mapping, index, filler, data, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01001849 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 if (PageUptodate(page))
1851 goto out;
1852
1853 lock_page(page);
1854 if (!page->mapping) {
1855 unlock_page(page);
1856 page_cache_release(page);
1857 goto retry;
1858 }
1859 if (PageUptodate(page)) {
1860 unlock_page(page);
1861 goto out;
1862 }
1863 err = filler(data, page);
1864 if (err < 0) {
1865 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01001866 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 }
David Howellsc855ff32007-05-09 13:42:20 +01001868out:
Nick Piggin6fe69002007-05-06 14:49:04 -07001869 mark_page_accessed(page);
1870 return page;
1871}
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001872
1873/**
1874 * read_cache_page_async - read into page cache, fill it if needed
1875 * @mapping: the page's address_space
1876 * @index: the page index
1877 * @filler: function to perform the read
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001878 * @data: first arg to filler(data, page) function, often left as NULL
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001879 *
1880 * Same as read_cache_page, but don't wait for page to become unlocked
1881 * after submitting it to the filler.
1882 *
1883 * Read into the page cache. If a page already exists, and PageUptodate() is
1884 * not set, try to fill the page but don't wait for it to become unlocked.
1885 *
1886 * If the page does not get brought uptodate, return -EIO.
1887 */
1888struct page *read_cache_page_async(struct address_space *mapping,
1889 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001890 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001891 void *data)
1892{
1893 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
1894}
Nick Piggin6fe69002007-05-06 14:49:04 -07001895EXPORT_SYMBOL(read_cache_page_async);
1896
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001897static struct page *wait_on_page_read(struct page *page)
1898{
1899 if (!IS_ERR(page)) {
1900 wait_on_page_locked(page);
1901 if (!PageUptodate(page)) {
1902 page_cache_release(page);
1903 page = ERR_PTR(-EIO);
1904 }
1905 }
1906 return page;
1907}
1908
1909/**
1910 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
1911 * @mapping: the page's address_space
1912 * @index: the page index
1913 * @gfp: the page allocator flags to use if allocating
1914 *
1915 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
Dave Kleikampe6f67b82011-12-21 11:05:48 -06001916 * any new page allocations done using the specified allocation flags.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001917 *
1918 * If the page does not get brought uptodate, return -EIO.
1919 */
1920struct page *read_cache_page_gfp(struct address_space *mapping,
1921 pgoff_t index,
1922 gfp_t gfp)
1923{
1924 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
1925
1926 return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp));
1927}
1928EXPORT_SYMBOL(read_cache_page_gfp);
1929
Nick Piggin6fe69002007-05-06 14:49:04 -07001930/**
1931 * read_cache_page - read into page cache, fill it if needed
1932 * @mapping: the page's address_space
1933 * @index: the page index
1934 * @filler: function to perform the read
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001935 * @data: first arg to filler(data, page) function, often left as NULL
Nick Piggin6fe69002007-05-06 14:49:04 -07001936 *
1937 * Read into the page cache. If a page already exists, and PageUptodate() is
1938 * not set, try to fill the page then wait for it to become unlocked.
1939 *
1940 * If the page does not get brought uptodate, return -EIO.
1941 */
1942struct page *read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001943 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001944 int (*filler)(void *, struct page *),
Nick Piggin6fe69002007-05-06 14:49:04 -07001945 void *data)
1946{
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001947 return wait_on_page_read(read_cache_page_async(mapping, index, filler, data));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949EXPORT_SYMBOL(read_cache_page);
1950
1951/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 * The logic we want is
1953 *
1954 * if suid or (sgid and xgrp)
1955 * remove privs
1956 */
Jens Axboe01de85e2006-10-17 19:50:36 +02001957int should_remove_suid(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958{
Al Viro649fc7b2011-07-26 04:31:05 -04001959 umode_t mode = dentry->d_inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 int kill = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961
1962 /* suid always must be killed */
1963 if (unlikely(mode & S_ISUID))
1964 kill = ATTR_KILL_SUID;
1965
1966 /*
1967 * sgid without any exec bits is just a mandatory locking mark; leave
1968 * it alone. If some exec bits are set, it's a real sgid; kill it.
1969 */
1970 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1971 kill |= ATTR_KILL_SGID;
1972
Dmitri Monakhov7f5ff762008-12-01 14:34:56 -08001973 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
Jens Axboe01de85e2006-10-17 19:50:36 +02001974 return kill;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975
Jens Axboe01de85e2006-10-17 19:50:36 +02001976 return 0;
1977}
Mark Fashehd23a1472006-10-17 17:05:18 -07001978EXPORT_SYMBOL(should_remove_suid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001979
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +02001980static int __remove_suid(struct dentry *dentry, int kill)
Jens Axboe01de85e2006-10-17 19:50:36 +02001981{
1982 struct iattr newattrs;
1983
1984 newattrs.ia_valid = ATTR_FORCE | kill;
1985 return notify_change(dentry, &newattrs);
1986}
1987
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001988int file_remove_suid(struct file *file)
Jens Axboe01de85e2006-10-17 19:50:36 +02001989{
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001990 struct dentry *dentry = file->f_path.dentry;
Andi Kleen69b45732011-05-28 08:25:51 -07001991 struct inode *inode = dentry->d_inode;
1992 int killsuid;
1993 int killpriv;
Serge E. Hallynb5376772007-10-16 23:31:36 -07001994 int error = 0;
Jens Axboe01de85e2006-10-17 19:50:36 +02001995
Andi Kleen69b45732011-05-28 08:25:51 -07001996 /* Fast path for nothing security related */
1997 if (IS_NOSEC(inode))
1998 return 0;
1999
2000 killsuid = should_remove_suid(dentry);
2001 killpriv = security_inode_need_killpriv(dentry);
2002
Serge E. Hallynb5376772007-10-16 23:31:36 -07002003 if (killpriv < 0)
2004 return killpriv;
2005 if (killpriv)
2006 error = security_inode_killpriv(dentry);
2007 if (!error && killsuid)
2008 error = __remove_suid(dentry, killsuid);
Al Viro9e1f1de2011-06-03 18:24:58 -04002009 if (!error && (inode->i_sb->s_flags & MS_NOSEC))
Andi Kleen69b45732011-05-28 08:25:51 -07002010 inode->i_flags |= S_NOSEC;
Jens Axboe01de85e2006-10-17 19:50:36 +02002011
Serge E. Hallynb5376772007-10-16 23:31:36 -07002012 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013}
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02002014EXPORT_SYMBOL(file_remove_suid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015
Nick Piggin2f718ff2007-10-16 01:24:59 -07002016static size_t __iovec_copy_from_user_inatomic(char *vaddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 const struct iovec *iov, size_t base, size_t bytes)
2018{
Ingo Molnarf1800532009-03-02 11:00:57 +01002019 size_t copied = 0, left = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020
2021 while (bytes) {
2022 char __user *buf = iov->iov_base + base;
2023 int copy = min(bytes, iov->iov_len - base);
2024
2025 base = 0;
Ingo Molnarf1800532009-03-02 11:00:57 +01002026 left = __copy_from_user_inatomic(vaddr, buf, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 copied += copy;
2028 bytes -= copy;
2029 vaddr += copy;
2030 iov++;
2031
NeilBrown01408c42006-06-25 05:47:58 -07002032 if (unlikely(left))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 }
2035 return copied - left;
2036}
2037
2038/*
Nick Piggin2f718ff2007-10-16 01:24:59 -07002039 * Copy as much as we can into the page and return the number of bytes which
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02002040 * were successfully copied. If a fault is encountered then return the number of
Nick Piggin2f718ff2007-10-16 01:24:59 -07002041 * bytes which were copied.
2042 */
2043size_t iov_iter_copy_from_user_atomic(struct page *page,
2044 struct iov_iter *i, unsigned long offset, size_t bytes)
2045{
2046 char *kaddr;
2047 size_t copied;
2048
2049 BUG_ON(!in_atomic());
Cong Wang9b04c5f2011-11-25 23:14:39 +08002050 kaddr = kmap_atomic(page);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002051 if (likely(i->nr_segs == 1)) {
2052 int left;
2053 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01002054 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002055 copied = bytes - left;
2056 } else {
2057 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
2058 i->iov, i->iov_offset, bytes);
2059 }
Cong Wang9b04c5f2011-11-25 23:14:39 +08002060 kunmap_atomic(kaddr);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002061
2062 return copied;
2063}
Nick Piggin89e10782007-10-16 01:25:07 -07002064EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002065
2066/*
2067 * This has the same sideeffects and return value as
2068 * iov_iter_copy_from_user_atomic().
2069 * The difference is that it attempts to resolve faults.
2070 * Page must not be locked.
2071 */
2072size_t iov_iter_copy_from_user(struct page *page,
2073 struct iov_iter *i, unsigned long offset, size_t bytes)
2074{
2075 char *kaddr;
2076 size_t copied;
2077
2078 kaddr = kmap(page);
2079 if (likely(i->nr_segs == 1)) {
2080 int left;
2081 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01002082 left = __copy_from_user(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002083 copied = bytes - left;
2084 } else {
2085 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
2086 i->iov, i->iov_offset, bytes);
2087 }
2088 kunmap(page);
2089 return copied;
2090}
Nick Piggin89e10782007-10-16 01:25:07 -07002091EXPORT_SYMBOL(iov_iter_copy_from_user);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002092
Nick Pigginf7009262008-03-10 11:43:59 -07002093void iov_iter_advance(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07002094{
Nick Pigginf7009262008-03-10 11:43:59 -07002095 BUG_ON(i->count < bytes);
2096
Nick Piggin2f718ff2007-10-16 01:24:59 -07002097 if (likely(i->nr_segs == 1)) {
2098 i->iov_offset += bytes;
Nick Pigginf7009262008-03-10 11:43:59 -07002099 i->count -= bytes;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002100 } else {
2101 const struct iovec *iov = i->iov;
2102 size_t base = i->iov_offset;
Jeff Layton39be79c2011-10-27 23:53:08 +02002103 unsigned long nr_segs = i->nr_segs;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002104
Nick Piggin124d3b72008-02-02 15:01:17 +01002105 /*
2106 * The !iov->iov_len check ensures we skip over unlikely
Nick Pigginf7009262008-03-10 11:43:59 -07002107 * zero-length segments (without overruning the iovec).
Nick Piggin124d3b72008-02-02 15:01:17 +01002108 */
Linus Torvalds94ad3742008-07-30 14:45:12 -07002109 while (bytes || unlikely(i->count && !iov->iov_len)) {
Nick Pigginf7009262008-03-10 11:43:59 -07002110 int copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002111
Nick Pigginf7009262008-03-10 11:43:59 -07002112 copy = min(bytes, iov->iov_len - base);
2113 BUG_ON(!i->count || i->count < copy);
2114 i->count -= copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002115 bytes -= copy;
2116 base += copy;
2117 if (iov->iov_len == base) {
2118 iov++;
Jeff Layton39be79c2011-10-27 23:53:08 +02002119 nr_segs--;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002120 base = 0;
2121 }
2122 }
2123 i->iov = iov;
2124 i->iov_offset = base;
Jeff Layton39be79c2011-10-27 23:53:08 +02002125 i->nr_segs = nr_segs;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002126 }
2127}
Nick Piggin89e10782007-10-16 01:25:07 -07002128EXPORT_SYMBOL(iov_iter_advance);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002129
Nick Pigginafddba42007-10-16 01:25:01 -07002130/*
2131 * Fault in the first iovec of the given iov_iter, to a maximum length
2132 * of bytes. Returns 0 on success, or non-zero if the memory could not be
2133 * accessed (ie. because it is an invalid address).
2134 *
2135 * writev-intensive code may want this to prefault several iovecs -- that
2136 * would be possible (callers must not rely on the fact that _only_ the
2137 * first iovec will be faulted with the current implementation).
2138 */
2139int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07002140{
Nick Piggin2f718ff2007-10-16 01:24:59 -07002141 char __user *buf = i->iov->iov_base + i->iov_offset;
Nick Pigginafddba42007-10-16 01:25:01 -07002142 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
2143 return fault_in_pages_readable(buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002144}
Nick Piggin89e10782007-10-16 01:25:07 -07002145EXPORT_SYMBOL(iov_iter_fault_in_readable);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002146
2147/*
2148 * Return the count of just the current iov_iter segment.
2149 */
2150size_t iov_iter_single_seg_count(struct iov_iter *i)
2151{
2152 const struct iovec *iov = i->iov;
2153 if (i->nr_segs == 1)
2154 return i->count;
2155 else
2156 return min(i->count, iov->iov_len - i->iov_offset);
2157}
Nick Piggin89e10782007-10-16 01:25:07 -07002158EXPORT_SYMBOL(iov_iter_single_seg_count);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002159
2160/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 * Performs necessary checks before doing a write
2162 *
Randy Dunlap485bb992006-06-23 02:03:49 -07002163 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 * Returns appropriate error code that caller should return or
2165 * zero in case that write should be allowed.
2166 */
2167inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2168{
2169 struct inode *inode = file->f_mapping->host;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002170 unsigned long limit = rlimit(RLIMIT_FSIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171
2172 if (unlikely(*pos < 0))
2173 return -EINVAL;
2174
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 if (!isblk) {
2176 /* FIXME: this is for backwards compatibility with 2.4 */
2177 if (file->f_flags & O_APPEND)
2178 *pos = i_size_read(inode);
2179
2180 if (limit != RLIM_INFINITY) {
2181 if (*pos >= limit) {
2182 send_sig(SIGXFSZ, current, 0);
2183 return -EFBIG;
2184 }
2185 if (*count > limit - (typeof(limit))*pos) {
2186 *count = limit - (typeof(limit))*pos;
2187 }
2188 }
2189 }
2190
2191 /*
2192 * LFS rule
2193 */
2194 if (unlikely(*pos + *count > MAX_NON_LFS &&
2195 !(file->f_flags & O_LARGEFILE))) {
2196 if (*pos >= MAX_NON_LFS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197 return -EFBIG;
2198 }
2199 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2200 *count = MAX_NON_LFS - (unsigned long)*pos;
2201 }
2202 }
2203
2204 /*
2205 * Are we about to exceed the fs block limit ?
2206 *
2207 * If we have written data it becomes a short write. If we have
2208 * exceeded without writing data we send a signal and return EFBIG.
2209 * Linus frestrict idea will clean these up nicely..
2210 */
2211 if (likely(!isblk)) {
2212 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2213 if (*count || *pos > inode->i_sb->s_maxbytes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 return -EFBIG;
2215 }
2216 /* zero-length writes at ->s_maxbytes are OK */
2217 }
2218
2219 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2220 *count = inode->i_sb->s_maxbytes - *pos;
2221 } else {
David Howells93614012006-09-30 20:45:40 +02002222#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223 loff_t isize;
2224 if (bdev_read_only(I_BDEV(inode)))
2225 return -EPERM;
2226 isize = i_size_read(inode);
2227 if (*pos >= isize) {
2228 if (*count || *pos > isize)
2229 return -ENOSPC;
2230 }
2231
2232 if (*pos + *count > isize)
2233 *count = isize - *pos;
David Howells93614012006-09-30 20:45:40 +02002234#else
2235 return -EPERM;
2236#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 }
2238 return 0;
2239}
2240EXPORT_SYMBOL(generic_write_checks);
2241
Nick Pigginafddba42007-10-16 01:25:01 -07002242int pagecache_write_begin(struct file *file, struct address_space *mapping,
2243 loff_t pos, unsigned len, unsigned flags,
2244 struct page **pagep, void **fsdata)
2245{
2246 const struct address_space_operations *aops = mapping->a_ops;
2247
Nick Piggin4e02ed42008-10-29 14:00:55 -07002248 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002249 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002250}
2251EXPORT_SYMBOL(pagecache_write_begin);
2252
2253int pagecache_write_end(struct file *file, struct address_space *mapping,
2254 loff_t pos, unsigned len, unsigned copied,
2255 struct page *page, void *fsdata)
2256{
2257 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002258
Nick Piggin4e02ed42008-10-29 14:00:55 -07002259 mark_page_accessed(page);
2260 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002261}
2262EXPORT_SYMBOL(pagecache_write_end);
2263
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264ssize_t
2265generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2266 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2267 size_t count, size_t ocount)
2268{
2269 struct file *file = iocb->ki_filp;
2270 struct address_space *mapping = file->f_mapping;
2271 struct inode *inode = mapping->host;
2272 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002273 size_t write_len;
2274 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275
2276 if (count != ocount)
2277 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2278
Christoph Hellwiga969e902008-07-23 21:27:04 -07002279 write_len = iov_length(iov, *nr_segs);
2280 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002281
Nick Piggin48b47c52009-01-06 14:40:22 -08002282 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002283 if (written)
2284 goto out;
2285
2286 /*
2287 * After a write we want buffered reads to be sure to go to disk to get
2288 * the new data. We invalidate clean cached page from the region we're
2289 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002290 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002291 */
2292 if (mapping->nrpages) {
2293 written = invalidate_inode_pages2_range(mapping,
2294 pos >> PAGE_CACHE_SHIFT, end);
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002295 /*
2296 * If a page can not be invalidated, return 0 to fall back
2297 * to buffered write.
2298 */
2299 if (written) {
2300 if (written == -EBUSY)
2301 return 0;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002302 goto out;
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002303 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002304 }
2305
2306 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2307
2308 /*
2309 * Finally, try again to invalidate clean pages which might have been
2310 * cached by non-direct readahead, or faulted in by get_user_pages()
2311 * if the source of the write was an mmap'ed region of the file
2312 * we're writing. Either one is a pretty crazy thing to do,
2313 * so we don't support it 100%. If this invalidation
2314 * fails, tough, the write still worked...
2315 */
2316 if (mapping->nrpages) {
2317 invalidate_inode_pages2_range(mapping,
2318 pos >> PAGE_CACHE_SHIFT, end);
2319 }
2320
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321 if (written > 0) {
Namhyung Kim01166512010-10-26 14:21:58 -07002322 pos += written;
2323 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2324 i_size_write(inode, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 mark_inode_dirty(inode);
2326 }
Namhyung Kim01166512010-10-26 14:21:58 -07002327 *ppos = pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002329out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 return written;
2331}
2332EXPORT_SYMBOL(generic_file_direct_write);
2333
Nick Piggineb2be182007-10-16 01:24:57 -07002334/*
2335 * Find or create a page at the given pagecache position. Return the locked
2336 * page. This function is specifically for buffered writes.
2337 */
Nick Piggin54566b22009-01-04 12:00:53 -08002338struct page *grab_cache_page_write_begin(struct address_space *mapping,
2339 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07002340{
2341 int status;
Johannes Weiner0faa70c2012-01-10 15:07:53 -08002342 gfp_t gfp_mask;
Nick Piggineb2be182007-10-16 01:24:57 -07002343 struct page *page;
Nick Piggin54566b22009-01-04 12:00:53 -08002344 gfp_t gfp_notmask = 0;
Johannes Weiner0faa70c2012-01-10 15:07:53 -08002345
Fengguang Wu1010bb12012-03-21 16:34:08 -07002346 gfp_mask = mapping_gfp_mask(mapping);
2347 if (mapping_cap_account_dirty(mapping))
2348 gfp_mask |= __GFP_WRITE;
Nick Piggin54566b22009-01-04 12:00:53 -08002349 if (flags & AOP_FLAG_NOFS)
2350 gfp_notmask = __GFP_FS;
Nick Piggineb2be182007-10-16 01:24:57 -07002351repeat:
2352 page = find_lock_page(mapping, index);
Steven Rostedtc585a262011-01-13 15:46:18 -08002353 if (page)
Darrick J. Wong3d08bcc2011-05-27 12:23:34 -07002354 goto found;
Nick Piggineb2be182007-10-16 01:24:57 -07002355
Johannes Weiner0faa70c2012-01-10 15:07:53 -08002356 page = __page_cache_alloc(gfp_mask & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002357 if (!page)
2358 return NULL;
Nick Piggin54566b22009-01-04 12:00:53 -08002359 status = add_to_page_cache_lru(page, mapping, index,
2360 GFP_KERNEL & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002361 if (unlikely(status)) {
2362 page_cache_release(page);
2363 if (status == -EEXIST)
2364 goto repeat;
2365 return NULL;
2366 }
Darrick J. Wong3d08bcc2011-05-27 12:23:34 -07002367found:
2368 wait_on_page_writeback(page);
Nick Piggineb2be182007-10-16 01:24:57 -07002369 return page;
2370}
Nick Piggin54566b22009-01-04 12:00:53 -08002371EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07002372
Nick Pigginafddba42007-10-16 01:25:01 -07002373static ssize_t generic_perform_write(struct file *file,
2374 struct iov_iter *i, loff_t pos)
2375{
2376 struct address_space *mapping = file->f_mapping;
2377 const struct address_space_operations *a_ops = mapping->a_ops;
2378 long status = 0;
2379 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002380 unsigned int flags = 0;
2381
2382 /*
2383 * Copies from kernel address space cannot fail (NFSD is a big user).
2384 */
2385 if (segment_eq(get_fs(), KERNEL_DS))
2386 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002387
2388 do {
2389 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002390 unsigned long offset; /* Offset into pagecache page */
2391 unsigned long bytes; /* Bytes to write to page */
2392 size_t copied; /* Bytes copied from user */
2393 void *fsdata;
2394
2395 offset = (pos & (PAGE_CACHE_SIZE - 1));
Nick Pigginafddba42007-10-16 01:25:01 -07002396 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2397 iov_iter_count(i));
2398
2399again:
Nick Pigginafddba42007-10-16 01:25:01 -07002400 /*
2401 * Bring in the user page that we will copy from _first_.
2402 * Otherwise there's a nasty deadlock on copying from the
2403 * same page as we're writing to, without it being marked
2404 * up-to-date.
2405 *
2406 * Not only is this an optimisation, but it is also required
2407 * to check that the address is actually valid, when atomic
2408 * usercopies are used, below.
2409 */
2410 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2411 status = -EFAULT;
2412 break;
2413 }
2414
Nick Piggin674b8922007-10-16 01:25:03 -07002415 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002416 &page, &fsdata);
2417 if (unlikely(status))
2418 break;
2419
anfei zhou931e80e2010-02-02 13:44:02 -08002420 if (mapping_writably_mapped(mapping))
2421 flush_dcache_page(page);
2422
Nick Pigginafddba42007-10-16 01:25:01 -07002423 pagefault_disable();
2424 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2425 pagefault_enable();
2426 flush_dcache_page(page);
2427
Josef Bacikc8236db2009-07-05 12:08:18 -07002428 mark_page_accessed(page);
Nick Pigginafddba42007-10-16 01:25:01 -07002429 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2430 page, fsdata);
2431 if (unlikely(status < 0))
2432 break;
2433 copied = status;
2434
2435 cond_resched();
2436
Nick Piggin124d3b72008-02-02 15:01:17 +01002437 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002438 if (unlikely(copied == 0)) {
2439 /*
2440 * If we were unable to copy any data at all, we must
2441 * fall back to a single segment length write.
2442 *
2443 * If we didn't fallback here, we could livelock
2444 * because not all segments in the iov can be copied at
2445 * once without a pagefault.
2446 */
2447 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2448 iov_iter_single_seg_count(i));
2449 goto again;
2450 }
Nick Pigginafddba42007-10-16 01:25:01 -07002451 pos += copied;
2452 written += copied;
2453
2454 balance_dirty_pages_ratelimited(mapping);
Jan Karaa50527b2011-12-02 09:17:02 +08002455 if (fatal_signal_pending(current)) {
2456 status = -EINTR;
2457 break;
2458 }
Nick Pigginafddba42007-10-16 01:25:01 -07002459 } while (iov_iter_count(i));
2460
2461 return written ? written : status;
2462}
2463
2464ssize_t
2465generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2466 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2467 size_t count, ssize_t written)
2468{
2469 struct file *file = iocb->ki_filp;
Nick Pigginafddba42007-10-16 01:25:01 -07002470 ssize_t status;
2471 struct iov_iter i;
2472
2473 iov_iter_init(&i, iov, nr_segs, count, written);
Nick Piggin4e02ed42008-10-29 14:00:55 -07002474 status = generic_perform_write(file, &i, pos);
Nick Pigginafddba42007-10-16 01:25:01 -07002475
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476 if (likely(status >= 0)) {
Nick Pigginafddba42007-10-16 01:25:01 -07002477 written += status;
2478 *ppos = pos + status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479 }
2480
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481 return written ? written : status;
2482}
2483EXPORT_SYMBOL(generic_file_buffered_write);
2484
Jan Karae4dd9de2009-08-17 18:10:06 +02002485/**
2486 * __generic_file_aio_write - write data to a file
2487 * @iocb: IO state structure (file, offset, etc.)
2488 * @iov: vector with data to write
2489 * @nr_segs: number of segments in the vector
2490 * @ppos: position where to write
2491 *
2492 * This function does all the work needed for actually writing data to a
2493 * file. It does all basic checks, removes SUID from the file, updates
2494 * modification times and calls proper subroutines depending on whether we
2495 * do direct IO or a standard buffered write.
2496 *
2497 * It expects i_mutex to be grabbed unless we work on a block device or similar
2498 * object which does not need locking at all.
2499 *
2500 * This function does *not* take care of syncing data in case of O_SYNC write.
2501 * A caller has to handle it. This is mainly due to the fact that we want to
2502 * avoid syncing under i_mutex.
2503 */
2504ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2505 unsigned long nr_segs, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506{
2507 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002508 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509 size_t ocount; /* original count */
2510 size_t count; /* after file limit checks */
2511 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002512 loff_t pos;
2513 ssize_t written;
2514 ssize_t err;
2515
2516 ocount = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07002517 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2518 if (err)
2519 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520
2521 count = ocount;
2522 pos = *ppos;
2523
2524 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2525
2526 /* We can write back this queue in page reclaim */
2527 current->backing_dev_info = mapping->backing_dev_info;
2528 written = 0;
2529
2530 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2531 if (err)
2532 goto out;
2533
2534 if (count == 0)
2535 goto out;
2536
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02002537 err = file_remove_suid(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 if (err)
2539 goto out;
2540
Christoph Hellwig870f4812006-01-09 20:52:01 -08002541 file_update_time(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542
2543 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2544 if (unlikely(file->f_flags & O_DIRECT)) {
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002545 loff_t endbyte;
2546 ssize_t written_buffered;
2547
2548 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2549 ppos, count, ocount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550 if (written < 0 || written == count)
2551 goto out;
2552 /*
2553 * direct-io write to a hole: fall through to buffered I/O
2554 * for completing the rest of the request.
2555 */
2556 pos += written;
2557 count -= written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002558 written_buffered = generic_file_buffered_write(iocb, iov,
2559 nr_segs, pos, ppos, count,
2560 written);
2561 /*
2562 * If generic_file_buffered_write() retuned a synchronous error
2563 * then we want to return the number of bytes which were
2564 * direct-written, or the error code if that was zero. Note
2565 * that this differs from normal direct-io semantics, which
2566 * will return -EFOO even if some bytes were written.
2567 */
2568 if (written_buffered < 0) {
2569 err = written_buffered;
2570 goto out;
2571 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002573 /*
2574 * We need to ensure that the page cache pages are written to
2575 * disk and invalidated to preserve the expected O_DIRECT
2576 * semantics.
2577 */
2578 endbyte = pos + written_buffered - written - 1;
Christoph Hellwigc05c4ed2009-09-23 15:07:30 +02002579 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002580 if (err == 0) {
2581 written = written_buffered;
2582 invalidate_mapping_pages(mapping,
2583 pos >> PAGE_CACHE_SHIFT,
2584 endbyte >> PAGE_CACHE_SHIFT);
2585 } else {
2586 /*
2587 * We don't know how much we wrote, so just return
2588 * the number of bytes which were direct-written
2589 */
2590 }
2591 } else {
2592 written = generic_file_buffered_write(iocb, iov, nr_segs,
2593 pos, ppos, count, written);
2594 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595out:
2596 current->backing_dev_info = NULL;
2597 return written ? written : err;
2598}
Jan Karae4dd9de2009-08-17 18:10:06 +02002599EXPORT_SYMBOL(__generic_file_aio_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600
Jan Karae4dd9de2009-08-17 18:10:06 +02002601/**
2602 * generic_file_aio_write - write data to a file
2603 * @iocb: IO state structure
2604 * @iov: vector with data to write
2605 * @nr_segs: number of segments in the vector
2606 * @pos: position in file where to write
2607 *
2608 * This is a wrapper around __generic_file_aio_write() to be used by most
2609 * filesystems. It takes care of syncing the file in case of O_SYNC file
2610 * and acquires i_mutex as needed.
2611 */
Badari Pulavarty027445c2006-09-30 23:28:46 -07002612ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2613 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002614{
2615 struct file *file = iocb->ki_filp;
Jan Kara148f9482009-08-17 19:52:36 +02002616 struct inode *inode = file->f_mapping->host;
Jens Axboe55602dd2010-06-24 15:05:37 +02002617 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619
2620 BUG_ON(iocb->ki_pos != pos);
2621
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002622 mutex_lock(&inode->i_mutex);
Jens Axboe55602dd2010-06-24 15:05:37 +02002623 blk_start_plug(&plug);
Jan Karae4dd9de2009-08-17 18:10:06 +02002624 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002625 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626
Jan Kara148f9482009-08-17 19:52:36 +02002627 if (ret > 0 || ret == -EIOCBQUEUED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628 ssize_t err;
2629
Jan Kara148f9482009-08-17 19:52:36 +02002630 err = generic_write_sync(file, pos, ret);
Jan Karac7b50db2009-08-18 16:18:20 +02002631 if (err < 0 && ret > 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632 ret = err;
2633 }
Jens Axboe55602dd2010-06-24 15:05:37 +02002634 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635 return ret;
2636}
2637EXPORT_SYMBOL(generic_file_aio_write);
2638
David Howellscf9a2ae2006-08-29 19:05:54 +01002639/**
2640 * try_to_release_page() - release old fs-specific metadata on a page
2641 *
2642 * @page: the page which the kernel is trying to free
2643 * @gfp_mask: memory allocation flags (and I/O mode)
2644 *
2645 * The address_space is to try to release any data against the page
2646 * (presumably at page->private). If the release was successful, return `1'.
2647 * Otherwise return zero.
2648 *
David Howells266cf652009-04-03 16:42:36 +01002649 * This may also be called if PG_fscache is set on a page, indicating that the
2650 * page is known to the local caching routines.
2651 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002652 * The @gfp_mask argument specifies whether I/O may be performed to release
Mingming Cao3f31fdd2008-07-25 01:46:22 -07002653 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01002654 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002655 */
2656int try_to_release_page(struct page *page, gfp_t gfp_mask)
2657{
2658 struct address_space * const mapping = page->mapping;
2659
2660 BUG_ON(!PageLocked(page));
2661 if (PageWriteback(page))
2662 return 0;
2663
2664 if (mapping && mapping->a_ops->releasepage)
2665 return mapping->a_ops->releasepage(page, gfp_mask);
2666 return try_to_free_buffers(page);
2667}
2668
2669EXPORT_SYMBOL(try_to_release_page);