blob: a4a5260b0279b77b37738540b1e8c24fb446a3e5 [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>
Paul Jackson44110fe2006-03-24 03:16:04 -080032#include <linux/cpuset.h>
Nick Piggin2f718ff2007-10-16 01:24:59 -070033#include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080034#include <linux/memcontrol.h>
Dan Magenheimerc515e1f2011-05-26 10:01:43 -060035#include <linux/cleancache.h>
Nick Piggin0f8053a2006-03-22 00:08:33 -080036#include "internal.h"
37
Linus Torvalds1da177e2005-04-16 15:20:36 -070038/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070039 * FIXME: remove all knowledge of the buffer layer from the core VM
40 */
Jan Kara148f9482009-08-17 19:52:36 +020041#include <linux/buffer_head.h> /* for try_to_free_buffers */
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include <asm/mman.h>
44
45/*
46 * Shared mappings implemented 30.11.1994. It's not fully working yet,
47 * though.
48 *
49 * Shared mappings now work. 15.8.1995 Bruno.
50 *
51 * finished 'unifying' the page and buffer cache and SMP-threaded the
52 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
53 *
54 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
55 */
56
57/*
58 * Lock ordering:
59 *
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070060 * ->i_mmap_mutex (truncate_pagecache)
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 * ->private_lock (__free_pte->__set_page_dirty_buffers)
Hugh Dickins5d337b92005-09-03 15:54:41 -070062 * ->swap_lock (exclusive_swap_page, others)
63 * ->mapping->tree_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080065 * ->i_mutex
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070066 * ->i_mmap_mutex (truncate->unmap_mapping_range)
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 *
68 * ->mmap_sem
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070069 * ->i_mmap_mutex
Hugh Dickinsb8072f02005-10-29 18:16:41 -070070 * ->page_table_lock or pte_lock (various, mainly in memory.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
72 *
73 * ->mmap_sem
74 * ->lock_page (access_process_vm)
75 *
Nick Piggin82591e62006-10-19 23:29:10 -070076 * ->i_mutex (generic_file_buffered_write)
77 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 *
Christoph Hellwigf758eea2011-04-21 18:19:44 -060079 * bdi->wb.list_lock
Dave Chinnera66979a2011-03-22 22:23:41 +110080 * sb_lock (fs/fs-writeback.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 * ->mapping->tree_lock (__sync_single_inode)
82 *
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070083 * ->i_mmap_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 * ->anon_vma.lock (vma_adjust)
85 *
86 * ->anon_vma.lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070087 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -070089 * ->page_table_lock or pte_lock
Hugh Dickins5d337b92005-09-03 15:54:41 -070090 * ->swap_lock (try_to_unmap_one)
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 * ->private_lock (try_to_unmap_one)
92 * ->tree_lock (try_to_unmap_one)
93 * ->zone.lru_lock (follow_page->mark_page_accessed)
Nick Piggin053837f2006-01-18 17:42:27 -080094 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 * ->private_lock (page_remove_rmap->set_page_dirty)
96 * ->tree_lock (page_remove_rmap->set_page_dirty)
Christoph Hellwigf758eea2011-04-21 18:19:44 -060097 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +110098 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
Christoph Hellwigf758eea2011-04-21 18:19:44 -060099 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100100 * ->inode->i_lock (zap_pte_range->set_page_dirty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
102 *
Andi Kleen9a3c5312012-03-21 16:34:09 -0700103 * ->i_mmap_mutex
104 * ->tasklist_lock (memory_failure, collect_procs_ao)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 */
106
107/*
Minchan Kime64a7822011-03-22 16:32:44 -0700108 * Delete a page from the page cache and free it. Caller has to make
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 * sure the page is locked and that nobody else uses it - or that usage
Nick Piggin19fd6232008-07-25 19:45:32 -0700110 * is safe. The caller must hold the mapping's tree_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 */
Minchan Kime64a7822011-03-22 16:32:44 -0700112void __delete_from_page_cache(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113{
114 struct address_space *mapping = page->mapping;
115
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600116 /*
117 * if we're uptodate, flush out into the cleancache, otherwise
118 * invalidate any existing cleancache entries. We can't leave
119 * stale data around in the cleancache once our page is gone
120 */
121 if (PageUptodate(page) && PageMappedToDisk(page))
122 cleancache_put_page(page);
123 else
Dan Magenheimer31677602011-09-21 11:56:28 -0400124 cleancache_invalidate_page(mapping, page);
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600125
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 radix_tree_delete(&mapping->page_tree, page->index);
127 page->mapping = NULL;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700128 /* Leave page->index set: truncation lookup relies upon it */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 mapping->nrpages--;
Christoph Lameter347ce432006-06-30 01:55:35 -0700130 __dec_zone_page_state(page, NR_FILE_PAGES);
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700131 if (PageSwapBacked(page))
132 __dec_zone_page_state(page, NR_SHMEM);
Nick Piggin45426812007-07-15 23:38:12 -0700133 BUG_ON(page_mapped(page));
Linus Torvalds3a692792007-12-19 14:05:13 -0800134
135 /*
136 * Some filesystems seem to re-dirty the page even after
137 * the VM has canceled the dirty bit (eg ext3 journaling).
138 *
139 * Fix it up by doing a final dirty accounting check after
140 * having removed the page entirely.
141 */
142 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
143 dec_zone_page_state(page, NR_FILE_DIRTY);
144 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
145 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146}
147
Minchan Kim702cfbf2011-03-22 16:32:43 -0700148/**
149 * delete_from_page_cache - delete page from page cache
150 * @page: the page which the kernel is trying to remove from page cache
151 *
152 * This must be called only on pages that have been verified to be in the page
153 * cache and locked. It will never put the page into the free list, the caller
154 * has a reference on the page.
155 */
156void delete_from_page_cache(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157{
158 struct address_space *mapping = page->mapping;
Linus Torvalds6072d132010-12-01 13:35:19 -0500159 void (*freepage)(struct page *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160
Matt Mackallcd7619d2005-05-01 08:59:01 -0700161 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162
Linus Torvalds6072d132010-12-01 13:35:19 -0500163 freepage = mapping->a_ops->freepage;
Nick Piggin19fd6232008-07-25 19:45:32 -0700164 spin_lock_irq(&mapping->tree_lock);
Minchan Kime64a7822011-03-22 16:32:44 -0700165 __delete_from_page_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700166 spin_unlock_irq(&mapping->tree_lock);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700167 mem_cgroup_uncharge_cache_page(page);
Linus Torvalds6072d132010-12-01 13:35:19 -0500168
169 if (freepage)
170 freepage(page);
Minchan Kim97cecb52011-03-22 16:30:53 -0700171 page_cache_release(page);
172}
173EXPORT_SYMBOL(delete_from_page_cache);
174
Jens Axboe7eaceac2011-03-10 08:52:07 +0100175static int sleep_on_page(void *word)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 io_schedule();
178 return 0;
179}
180
Jens Axboe7eaceac2011-03-10 08:52:07 +0100181static int sleep_on_page_killable(void *word)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500182{
Jens Axboe7eaceac2011-03-10 08:52:07 +0100183 sleep_on_page(word);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500184 return fatal_signal_pending(current) ? -EINTR : 0;
185}
186
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700188 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700189 * @mapping: address space structure to write
190 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800191 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700192 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700194 * Start writeback against all of a mapping's dirty pages that lie
195 * within the byte offsets <start, end> inclusive.
196 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700198 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 * these two operations is that if a dirty page/buffer is encountered, it must
200 * be waited upon, and not just skipped over.
201 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800202int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
203 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204{
205 int ret;
206 struct writeback_control wbc = {
207 .sync_mode = sync_mode,
Nick Piggin05fe4782009-01-06 14:39:08 -0800208 .nr_to_write = LONG_MAX,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700209 .range_start = start,
210 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 };
212
213 if (!mapping_cap_writeback_dirty(mapping))
214 return 0;
215
216 ret = do_writepages(mapping, &wbc);
217 return ret;
218}
219
220static inline int __filemap_fdatawrite(struct address_space *mapping,
221 int sync_mode)
222{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700223 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224}
225
226int filemap_fdatawrite(struct address_space *mapping)
227{
228 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
229}
230EXPORT_SYMBOL(filemap_fdatawrite);
231
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400232int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800233 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234{
235 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
236}
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400237EXPORT_SYMBOL(filemap_fdatawrite_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238
Randy Dunlap485bb992006-06-23 02:03:49 -0700239/**
240 * filemap_flush - mostly a non-blocking flush
241 * @mapping: target address_space
242 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 * This is a mostly non-blocking flush. Not suitable for data-integrity
244 * purposes - I/O may not be started against all dirty pages.
245 */
246int filemap_flush(struct address_space *mapping)
247{
248 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
249}
250EXPORT_SYMBOL(filemap_flush);
251
Randy Dunlap485bb992006-06-23 02:03:49 -0700252/**
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200253 * filemap_fdatawait_range - wait for writeback to complete
254 * @mapping: address space structure to wait for
255 * @start_byte: offset in bytes where the range starts
256 * @end_byte: offset in bytes where the range ends (inclusive)
Randy Dunlap485bb992006-06-23 02:03:49 -0700257 *
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200258 * Walk the list of under-writeback pages of the given address space
259 * in the given range and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 */
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200261int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
262 loff_t end_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263{
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200264 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
265 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 struct pagevec pvec;
267 int nr_pages;
268 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200270 if (end_byte < start_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 return 0;
272
273 pagevec_init(&pvec, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 while ((index <= end) &&
275 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
276 PAGECACHE_TAG_WRITEBACK,
277 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
278 unsigned i;
279
280 for (i = 0; i < nr_pages; i++) {
281 struct page *page = pvec.pages[i];
282
283 /* until radix tree lookup accepts end_index */
284 if (page->index > end)
285 continue;
286
287 wait_on_page_writeback(page);
Rik van Riel212260a2011-01-13 15:46:06 -0800288 if (TestClearPageError(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 ret = -EIO;
290 }
291 pagevec_release(&pvec);
292 cond_resched();
293 }
294
295 /* Check for outstanding write errors */
296 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
297 ret = -ENOSPC;
298 if (test_and_clear_bit(AS_EIO, &mapping->flags))
299 ret = -EIO;
300
301 return ret;
302}
Jan Karad3bccb6f2009-08-17 19:30:27 +0200303EXPORT_SYMBOL(filemap_fdatawait_range);
304
305/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700306 * filemap_fdatawait - wait for all under-writeback pages to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 * @mapping: address space structure to wait for
Randy Dunlap485bb992006-06-23 02:03:49 -0700308 *
309 * Walk the list of under-writeback pages of the given address space
310 * and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 */
312int filemap_fdatawait(struct address_space *mapping)
313{
314 loff_t i_size = i_size_read(mapping->host);
315
316 if (i_size == 0)
317 return 0;
318
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200319 return filemap_fdatawait_range(mapping, 0, i_size - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320}
321EXPORT_SYMBOL(filemap_fdatawait);
322
323int filemap_write_and_wait(struct address_space *mapping)
324{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800325 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326
327 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800328 err = filemap_fdatawrite(mapping);
329 /*
330 * Even if the above returned error, the pages may be
331 * written partially (e.g. -ENOSPC), so we wait for it.
332 * But the -EIO is special case, it may indicate the worst
333 * thing (e.g. bug) happened, so we avoid waiting for it.
334 */
335 if (err != -EIO) {
336 int err2 = filemap_fdatawait(mapping);
337 if (!err)
338 err = err2;
339 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800341 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800343EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344
Randy Dunlap485bb992006-06-23 02:03:49 -0700345/**
346 * filemap_write_and_wait_range - write out & wait on a file range
347 * @mapping: the address_space for the pages
348 * @lstart: offset in bytes where the range starts
349 * @lend: offset in bytes where the range ends (inclusive)
350 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800351 * Write out and wait upon file offsets lstart->lend, inclusive.
352 *
353 * Note that `lend' is inclusive (describes the last byte to be written) so
354 * that this function can be used to write to the very end-of-file (end = -1).
355 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356int filemap_write_and_wait_range(struct address_space *mapping,
357 loff_t lstart, loff_t lend)
358{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800359 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360
361 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800362 err = __filemap_fdatawrite_range(mapping, lstart, lend,
363 WB_SYNC_ALL);
364 /* See comment of filemap_write_and_wait() */
365 if (err != -EIO) {
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200366 int err2 = filemap_fdatawait_range(mapping,
367 lstart, lend);
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800368 if (!err)
369 err = err2;
370 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800372 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373}
Chris Masonf6995582009-04-15 13:22:37 -0400374EXPORT_SYMBOL(filemap_write_and_wait_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375
Randy Dunlap485bb992006-06-23 02:03:49 -0700376/**
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700377 * replace_page_cache_page - replace a pagecache page with a new one
378 * @old: page to be replaced
379 * @new: page to replace with
380 * @gfp_mask: allocation mode
381 *
382 * This function replaces a page in the pagecache with a new one. On
383 * success it acquires the pagecache reference for the new page and
384 * drops it for the old page. Both the old and new pages must be
385 * locked. This function does not add the new page to the LRU, the
386 * caller must do that.
387 *
388 * The remove + add is atomic. The only way this function can fail is
389 * memory allocation failure.
390 */
391int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
392{
393 int error;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700394
395 VM_BUG_ON(!PageLocked(old));
396 VM_BUG_ON(!PageLocked(new));
397 VM_BUG_ON(new->mapping);
398
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700399 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
400 if (!error) {
401 struct address_space *mapping = old->mapping;
402 void (*freepage)(struct page *);
403
404 pgoff_t offset = old->index;
405 freepage = mapping->a_ops->freepage;
406
407 page_cache_get(new);
408 new->mapping = mapping;
409 new->index = offset;
410
411 spin_lock_irq(&mapping->tree_lock);
Minchan Kime64a7822011-03-22 16:32:44 -0700412 __delete_from_page_cache(old);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700413 error = radix_tree_insert(&mapping->page_tree, offset, new);
414 BUG_ON(error);
415 mapping->nrpages++;
416 __inc_zone_page_state(new, NR_FILE_PAGES);
417 if (PageSwapBacked(new))
418 __inc_zone_page_state(new, NR_SHMEM);
419 spin_unlock_irq(&mapping->tree_lock);
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -0800420 /* mem_cgroup codes must not be called under tree_lock */
421 mem_cgroup_replace_page_cache(old, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700422 radix_tree_preload_end();
423 if (freepage)
424 freepage(old);
425 page_cache_release(old);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700426 }
427
428 return error;
429}
430EXPORT_SYMBOL_GPL(replace_page_cache_page);
431
432/**
Nick Piggine2867812008-07-25 19:45:30 -0700433 * add_to_page_cache_locked - add a locked page to the pagecache
Randy Dunlap485bb992006-06-23 02:03:49 -0700434 * @page: page to add
435 * @mapping: the page's address_space
436 * @offset: page index
437 * @gfp_mask: page allocation mode
438 *
Nick Piggine2867812008-07-25 19:45:30 -0700439 * This function is used to add a page to the pagecache. It must be locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 * This function does not add the page to the LRU. The caller must do that.
441 */
Nick Piggine2867812008-07-25 19:45:30 -0700442int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400443 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444{
Nick Piggine2867812008-07-25 19:45:30 -0700445 int error;
446
447 VM_BUG_ON(!PageLocked(page));
Hugh Dickins31475dd2011-08-03 16:21:27 -0700448 VM_BUG_ON(PageSwapBacked(page));
Nick Piggine2867812008-07-25 19:45:30 -0700449
450 error = mem_cgroup_cache_charge(page, current->mm,
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -0800451 gfp_mask & GFP_RECLAIM_MASK);
Balbir Singh35c754d2008-02-07 00:14:05 -0800452 if (error)
453 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454
Balbir Singh35c754d2008-02-07 00:14:05 -0800455 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 if (error == 0) {
Nick Piggine2867812008-07-25 19:45:30 -0700457 page_cache_get(page);
458 page->mapping = mapping;
459 page->index = offset;
460
Nick Piggin19fd6232008-07-25 19:45:32 -0700461 spin_lock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 error = radix_tree_insert(&mapping->page_tree, offset, page);
Nick Piggine2867812008-07-25 19:45:30 -0700463 if (likely(!error)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 mapping->nrpages++;
Christoph Lameter347ce432006-06-30 01:55:35 -0700465 __inc_zone_page_state(page, NR_FILE_PAGES);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700466 spin_unlock_irq(&mapping->tree_lock);
Nick Piggine2867812008-07-25 19:45:30 -0700467 } else {
468 page->mapping = NULL;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700469 /* Leave page->index set: truncation relies upon it */
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700470 spin_unlock_irq(&mapping->tree_lock);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700471 mem_cgroup_uncharge_cache_page(page);
Nick Piggine2867812008-07-25 19:45:30 -0700472 page_cache_release(page);
473 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 radix_tree_preload_end();
Balbir Singh35c754d2008-02-07 00:14:05 -0800475 } else
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700476 mem_cgroup_uncharge_cache_page(page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800477out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 return error;
479}
Nick Piggine2867812008-07-25 19:45:30 -0700480EXPORT_SYMBOL(add_to_page_cache_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481
482int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400483 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484{
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700485 int ret;
486
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700487 ret = add_to_page_cache(page, mapping, offset, gfp_mask);
Hugh Dickins31475dd2011-08-03 16:21:27 -0700488 if (ret == 0)
489 lru_cache_add_file(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 return ret;
491}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300492EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493
Paul Jackson44110fe2006-03-24 03:16:04 -0800494#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700495struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800496{
Miao Xiec0ff7452010-05-24 14:32:08 -0700497 int n;
498 struct page *page;
499
Paul Jackson44110fe2006-03-24 03:16:04 -0800500 if (cpuset_do_page_mem_spread()) {
Mel Gormancc9a6c82012-03-21 16:34:11 -0700501 unsigned int cpuset_mems_cookie;
502 do {
503 cpuset_mems_cookie = get_mems_allowed();
504 n = cpuset_mem_spread_node();
505 page = alloc_pages_exact_node(n, gfp, 0);
506 } while (!put_mems_allowed(cpuset_mems_cookie) && !page);
507
Miao Xiec0ff7452010-05-24 14:32:08 -0700508 return page;
Paul Jackson44110fe2006-03-24 03:16:04 -0800509 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700510 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800511}
Nick Piggin2ae88142006-10-28 10:38:23 -0700512EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800513#endif
514
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515/*
516 * In order to wait for pages to become available there must be
517 * waitqueues associated with pages. By using a hash table of
518 * waitqueues where the bucket discipline is to maintain all
519 * waiters on the same queue and wake all when any of the pages
520 * become available, and for the woken contexts to check to be
521 * sure the appropriate page became available, this saves space
522 * at a cost of "thundering herd" phenomena during rare hash
523 * collisions.
524 */
525static wait_queue_head_t *page_waitqueue(struct page *page)
526{
527 const struct zone *zone = page_zone(page);
528
529 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
530}
531
532static inline void wake_up_page(struct page *page, int bit)
533{
534 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
535}
536
Harvey Harrison920c7a52008-02-04 22:29:26 -0800537void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538{
539 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
540
541 if (test_bit(bit_nr, &page->flags))
Jens Axboe7eaceac2011-03-10 08:52:07 +0100542 __wait_on_bit(page_waitqueue(page), &wait, sleep_on_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 TASK_UNINTERRUPTIBLE);
544}
545EXPORT_SYMBOL(wait_on_page_bit);
546
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700547int wait_on_page_bit_killable(struct page *page, int bit_nr)
548{
549 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
550
551 if (!test_bit(bit_nr, &page->flags))
552 return 0;
553
554 return __wait_on_bit(page_waitqueue(page), &wait,
555 sleep_on_page_killable, TASK_KILLABLE);
556}
557
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558/**
David Howells385e1ca5f2009-04-03 16:42:39 +0100559 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
Randy Dunlap697f6192009-04-13 14:39:54 -0700560 * @page: Page defining the wait queue of interest
561 * @waiter: Waiter to add to the queue
David Howells385e1ca5f2009-04-03 16:42:39 +0100562 *
563 * Add an arbitrary @waiter to the wait queue for the nominated @page.
564 */
565void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
566{
567 wait_queue_head_t *q = page_waitqueue(page);
568 unsigned long flags;
569
570 spin_lock_irqsave(&q->lock, flags);
571 __add_wait_queue(q, waiter);
572 spin_unlock_irqrestore(&q->lock, flags);
573}
574EXPORT_SYMBOL_GPL(add_page_wait_queue);
575
576/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700577 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 * @page: the page
579 *
580 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
581 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
582 * mechananism between PageLocked pages and PageWriteback pages is shared.
583 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
584 *
Nick Piggin8413ac92008-10-18 20:26:59 -0700585 * The mb is necessary to enforce ordering between the clear_bit and the read
586 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800588void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589{
Nick Piggin8413ac92008-10-18 20:26:59 -0700590 VM_BUG_ON(!PageLocked(page));
591 clear_bit_unlock(PG_locked, &page->flags);
592 smp_mb__after_clear_bit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 wake_up_page(page, PG_locked);
594}
595EXPORT_SYMBOL(unlock_page);
596
Randy Dunlap485bb992006-06-23 02:03:49 -0700597/**
598 * end_page_writeback - end writeback against a page
599 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 */
601void end_page_writeback(struct page *page)
602{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700603 if (TestClearPageReclaim(page))
604 rotate_reclaimable_page(page);
605
606 if (!test_clear_page_writeback(page))
607 BUG();
608
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 smp_mb__after_clear_bit();
610 wake_up_page(page, PG_writeback);
611}
612EXPORT_SYMBOL(end_page_writeback);
613
Randy Dunlap485bb992006-06-23 02:03:49 -0700614/**
615 * __lock_page - get a lock on the page, assuming we need to sleep to get it
616 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800618void __lock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619{
620 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
621
Jens Axboe7eaceac2011-03-10 08:52:07 +0100622 __wait_on_bit_lock(page_waitqueue(page), &wait, sleep_on_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 TASK_UNINTERRUPTIBLE);
624}
625EXPORT_SYMBOL(__lock_page);
626
Harvey Harrisonb5606c22008-02-13 15:03:16 -0800627int __lock_page_killable(struct page *page)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500628{
629 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
630
631 return __wait_on_bit_lock(page_waitqueue(page), &wait,
Jens Axboe7eaceac2011-03-10 08:52:07 +0100632 sleep_on_page_killable, TASK_KILLABLE);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500633}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300634EXPORT_SYMBOL_GPL(__lock_page_killable);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500635
Michel Lespinassed065bd82010-10-26 14:21:57 -0700636int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
637 unsigned int flags)
638{
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700639 if (flags & FAULT_FLAG_ALLOW_RETRY) {
640 /*
641 * CAUTION! In this case, mmap_sem is not released
642 * even though return 0.
643 */
644 if (flags & FAULT_FLAG_RETRY_NOWAIT)
645 return 0;
646
647 up_read(&mm->mmap_sem);
648 if (flags & FAULT_FLAG_KILLABLE)
649 wait_on_page_locked_killable(page);
650 else
Gleb Natapov318b2752011-03-22 16:30:51 -0700651 wait_on_page_locked(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -0700652 return 0;
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700653 } else {
654 if (flags & FAULT_FLAG_KILLABLE) {
655 int ret;
656
657 ret = __lock_page_killable(page);
658 if (ret) {
659 up_read(&mm->mmap_sem);
660 return 0;
661 }
662 } else
663 __lock_page(page);
664 return 1;
Michel Lespinassed065bd82010-10-26 14:21:57 -0700665 }
666}
667
Randy Dunlap485bb992006-06-23 02:03:49 -0700668/**
669 * find_get_page - find and get a page reference
670 * @mapping: the address_space to search
671 * @offset: the page index
672 *
Nick Pigginda6052f2006-09-25 23:31:35 -0700673 * Is there a pagecache struct page at the given (mapping, offset) tuple?
674 * If yes, increment its refcount and return it; if no, return NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 */
Nick Piggina60637c2008-07-25 19:45:31 -0700676struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677{
Nick Piggina60637c2008-07-25 19:45:31 -0700678 void **pagep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 struct page *page;
680
Nick Piggina60637c2008-07-25 19:45:31 -0700681 rcu_read_lock();
682repeat:
683 page = NULL;
684 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
685 if (pagep) {
686 page = radix_tree_deref_slot(pagep);
Nick Piggin27d20fd2010-11-11 14:05:19 -0800687 if (unlikely(!page))
688 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700689 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700690 if (radix_tree_deref_retry(page))
691 goto repeat;
692 /*
693 * Otherwise, shmem/tmpfs must be storing a swap entry
694 * here as an exceptional entry: so return it without
695 * attempting to raise page count.
696 */
697 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700698 }
Nick Piggina60637c2008-07-25 19:45:31 -0700699 if (!page_cache_get_speculative(page))
700 goto repeat;
701
702 /*
703 * Has the page moved?
704 * This is part of the lockless pagecache protocol. See
705 * include/linux/pagemap.h for details.
706 */
707 if (unlikely(page != *pagep)) {
708 page_cache_release(page);
709 goto repeat;
710 }
711 }
Nick Piggin27d20fd2010-11-11 14:05:19 -0800712out:
Nick Piggina60637c2008-07-25 19:45:31 -0700713 rcu_read_unlock();
714
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 return page;
716}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717EXPORT_SYMBOL(find_get_page);
718
Randy Dunlap485bb992006-06-23 02:03:49 -0700719/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 * find_lock_page - locate, pin and lock a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700721 * @mapping: the address_space to search
722 * @offset: the page index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 *
724 * Locates the desired pagecache page, locks it, increments its reference
725 * count and returns its address.
726 *
727 * Returns zero if the page was not present. find_lock_page() may sleep.
728 */
Nick Piggina60637c2008-07-25 19:45:31 -0700729struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730{
731 struct page *page;
732
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733repeat:
Nick Piggina60637c2008-07-25 19:45:31 -0700734 page = find_get_page(mapping, offset);
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700735 if (page && !radix_tree_exception(page)) {
Nick Piggina60637c2008-07-25 19:45:31 -0700736 lock_page(page);
737 /* Has the page been truncated? */
738 if (unlikely(page->mapping != mapping)) {
739 unlock_page(page);
740 page_cache_release(page);
741 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 }
Nick Piggina60637c2008-07-25 19:45:31 -0700743 VM_BUG_ON(page->index != offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 return page;
746}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747EXPORT_SYMBOL(find_lock_page);
748
749/**
750 * find_or_create_page - locate or add a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700751 * @mapping: the page's address_space
752 * @index: the page's index into the mapping
753 * @gfp_mask: page allocation mode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 *
755 * Locates a page in the pagecache. If the page is not present, a new page
756 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
757 * LRU list. The returned page is locked and has its reference count
758 * incremented.
759 *
760 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
761 * allocation!
762 *
763 * find_or_create_page() returns the desired page's address, or zero on
764 * memory exhaustion.
765 */
766struct page *find_or_create_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -0700767 pgoff_t index, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768{
Nick Piggineb2be182007-10-16 01:24:57 -0700769 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 int err;
771repeat:
772 page = find_lock_page(mapping, index);
773 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -0700774 page = __page_cache_alloc(gfp_mask);
775 if (!page)
776 return NULL;
Nick Piggin67d58ac2009-01-06 14:40:28 -0800777 /*
778 * We want a regular kernel memory (not highmem or DMA etc)
779 * allocation for the radix tree nodes, but we need to honour
780 * the context-specific requirements the caller has asked for.
781 * GFP_RECLAIM_MASK collects those requirements.
782 */
783 err = add_to_page_cache_lru(page, mapping, index,
784 (gfp_mask & GFP_RECLAIM_MASK));
Nick Piggineb2be182007-10-16 01:24:57 -0700785 if (unlikely(err)) {
786 page_cache_release(page);
787 page = NULL;
788 if (err == -EEXIST)
789 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 return page;
793}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794EXPORT_SYMBOL(find_or_create_page);
795
796/**
797 * find_get_pages - gang pagecache lookup
798 * @mapping: The address_space to search
799 * @start: The starting page index
800 * @nr_pages: The maximum number of pages
801 * @pages: Where the resulting pages are placed
802 *
803 * find_get_pages() will search for and return a group of up to
804 * @nr_pages pages in the mapping. The pages are placed at @pages.
805 * find_get_pages() takes a reference against the returned pages.
806 *
807 * The search returns a group of mapping-contiguous pages with ascending
808 * indexes. There may be holes in the indices due to not-present pages.
809 *
810 * find_get_pages() returns the number of pages which were found.
811 */
812unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
813 unsigned int nr_pages, struct page **pages)
814{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700815 struct radix_tree_iter iter;
816 void **slot;
817 unsigned ret = 0;
818
819 if (unlikely(!nr_pages))
820 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821
Nick Piggina60637c2008-07-25 19:45:31 -0700822 rcu_read_lock();
823restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700824 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
Nick Piggina60637c2008-07-25 19:45:31 -0700825 struct page *page;
826repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700827 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -0700828 if (unlikely(!page))
829 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -0700830
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700831 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700832 if (radix_tree_deref_retry(page)) {
833 /*
834 * Transient condition which can only trigger
835 * when entry at index 0 moves out of or back
836 * to root: none yet gotten, safe to restart.
837 */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700838 WARN_ON(iter.index);
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700839 goto restart;
840 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700841 /*
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700842 * Otherwise, shmem/tmpfs must be storing a swap entry
843 * here as an exceptional entry: so skip over it -
844 * we only reach this from invalidate_mapping_pages().
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700845 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700846 continue;
Nick Piggin27d20fd2010-11-11 14:05:19 -0800847 }
Nick Piggina60637c2008-07-25 19:45:31 -0700848
849 if (!page_cache_get_speculative(page))
850 goto repeat;
851
852 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700853 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -0700854 page_cache_release(page);
855 goto repeat;
856 }
857
858 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700859 if (++ret == nr_pages)
860 break;
Nick Piggina60637c2008-07-25 19:45:31 -0700861 }
Hugh Dickins5b280c02011-03-22 16:33:07 -0700862
Nick Piggina60637c2008-07-25 19:45:31 -0700863 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 return ret;
865}
866
Jens Axboeebf43502006-04-27 08:46:01 +0200867/**
868 * find_get_pages_contig - gang contiguous pagecache lookup
869 * @mapping: The address_space to search
870 * @index: The starting page index
871 * @nr_pages: The maximum number of pages
872 * @pages: Where the resulting pages are placed
873 *
874 * find_get_pages_contig() works exactly like find_get_pages(), except
875 * that the returned number of pages are guaranteed to be contiguous.
876 *
877 * find_get_pages_contig() returns the number of pages which were found.
878 */
879unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
880 unsigned int nr_pages, struct page **pages)
881{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700882 struct radix_tree_iter iter;
883 void **slot;
884 unsigned int ret = 0;
885
886 if (unlikely(!nr_pages))
887 return 0;
Jens Axboeebf43502006-04-27 08:46:01 +0200888
Nick Piggina60637c2008-07-25 19:45:31 -0700889 rcu_read_lock();
890restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700891 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
Nick Piggina60637c2008-07-25 19:45:31 -0700892 struct page *page;
893repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700894 page = radix_tree_deref_slot(slot);
895 /* The hole, there no reason to continue */
Nick Piggina60637c2008-07-25 19:45:31 -0700896 if (unlikely(!page))
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700897 break;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -0700898
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700899 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700900 if (radix_tree_deref_retry(page)) {
901 /*
902 * Transient condition which can only trigger
903 * when entry at index 0 moves out of or back
904 * to root: none yet gotten, safe to restart.
905 */
906 goto restart;
907 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700908 /*
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700909 * Otherwise, shmem/tmpfs must be storing a swap entry
910 * here as an exceptional entry: so stop looking for
911 * contiguous pages.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700912 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700913 break;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700914 }
Nick Piggina60637c2008-07-25 19:45:31 -0700915
Nick Piggina60637c2008-07-25 19:45:31 -0700916 if (!page_cache_get_speculative(page))
917 goto repeat;
918
919 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700920 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -0700921 page_cache_release(page);
922 goto repeat;
923 }
924
Nick Piggin9cbb4cb2011-01-13 15:45:51 -0800925 /*
926 * must check mapping and index after taking the ref.
927 * otherwise we can get both false positives and false
928 * negatives, which is just confusing to the caller.
929 */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700930 if (page->mapping == NULL || page->index != iter.index) {
Nick Piggin9cbb4cb2011-01-13 15:45:51 -0800931 page_cache_release(page);
932 break;
933 }
934
Nick Piggina60637c2008-07-25 19:45:31 -0700935 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700936 if (++ret == nr_pages)
937 break;
Jens Axboeebf43502006-04-27 08:46:01 +0200938 }
Nick Piggina60637c2008-07-25 19:45:31 -0700939 rcu_read_unlock();
940 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +0200941}
David Howellsef71c152007-05-09 02:33:44 -0700942EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +0200943
Randy Dunlap485bb992006-06-23 02:03:49 -0700944/**
945 * find_get_pages_tag - find and return pages that match @tag
946 * @mapping: the address_space to search
947 * @index: the starting page index
948 * @tag: the tag index
949 * @nr_pages: the maximum number of pages
950 * @pages: where the resulting pages are placed
951 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -0700953 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 */
955unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
956 int tag, unsigned int nr_pages, struct page **pages)
957{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700958 struct radix_tree_iter iter;
959 void **slot;
960 unsigned ret = 0;
961
962 if (unlikely(!nr_pages))
963 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964
Nick Piggina60637c2008-07-25 19:45:31 -0700965 rcu_read_lock();
966restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700967 radix_tree_for_each_tagged(slot, &mapping->page_tree,
968 &iter, *index, tag) {
Nick Piggina60637c2008-07-25 19:45:31 -0700969 struct page *page;
970repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700971 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -0700972 if (unlikely(!page))
973 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -0700974
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700975 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700976 if (radix_tree_deref_retry(page)) {
977 /*
978 * Transient condition which can only trigger
979 * when entry at index 0 moves out of or back
980 * to root: none yet gotten, safe to restart.
981 */
982 goto restart;
983 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700984 /*
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700985 * This function is never used on a shmem/tmpfs
986 * mapping, so a swap entry won't be found here.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700987 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700988 BUG();
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700989 }
Nick Piggina60637c2008-07-25 19:45:31 -0700990
991 if (!page_cache_get_speculative(page))
992 goto repeat;
993
994 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700995 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -0700996 page_cache_release(page);
997 goto repeat;
998 }
999
1000 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001001 if (++ret == nr_pages)
1002 break;
Nick Piggina60637c2008-07-25 19:45:31 -07001003 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001004
Nick Piggina60637c2008-07-25 19:45:31 -07001005 rcu_read_unlock();
1006
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 if (ret)
1008 *index = pages[ret - 1]->index + 1;
Nick Piggina60637c2008-07-25 19:45:31 -07001009
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 return ret;
1011}
David Howellsef71c152007-05-09 02:33:44 -07001012EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013
Randy Dunlap485bb992006-06-23 02:03:49 -07001014/**
1015 * grab_cache_page_nowait - returns locked page at given index in given cache
1016 * @mapping: target address_space
1017 * @index: the page index
1018 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08001019 * Same as grab_cache_page(), but do not wait if the page is unavailable.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 * This is intended for speculative data generators, where the data can
1021 * be regenerated if the page couldn't be grabbed. This routine should
1022 * be safe to call while holding the lock for another page.
1023 *
1024 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
1025 * and deadlock against the caller's locked page.
1026 */
1027struct page *
Fengguang Wu57f6b962007-10-16 01:24:37 -07001028grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029{
1030 struct page *page = find_get_page(mapping, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031
1032 if (page) {
Nick Piggin529ae9a2008-08-02 12:01:03 +02001033 if (trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 return page;
1035 page_cache_release(page);
1036 return NULL;
1037 }
Nick Piggin2ae88142006-10-28 10:38:23 -07001038 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
Nick Piggin67d58ac2009-01-06 14:40:28 -08001039 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 page_cache_release(page);
1041 page = NULL;
1042 }
1043 return page;
1044}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045EXPORT_SYMBOL(grab_cache_page_nowait);
1046
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001047/*
1048 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1049 * a _large_ part of the i/o request. Imagine the worst scenario:
1050 *
1051 * ---R__________________________________________B__________
1052 * ^ reading here ^ bad block(assume 4k)
1053 *
1054 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1055 * => failing the whole request => read(R) => read(R+1) =>
1056 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1057 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1058 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1059 *
1060 * It is going insane. Fix it by quickly scaling down the readahead size.
1061 */
1062static void shrink_readahead_size_eio(struct file *filp,
1063 struct file_ra_state *ra)
1064{
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001065 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001066}
1067
Randy Dunlap485bb992006-06-23 02:03:49 -07001068/**
Christoph Hellwig36e78912008-02-08 04:21:24 -08001069 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001070 * @filp: the file to read
1071 * @ppos: current file position
1072 * @desc: read_descriptor
1073 * @actor: read method
1074 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -07001076 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 *
1078 * This is really ugly. But the goto's actually try to clarify some
1079 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 */
Christoph Hellwig36e78912008-02-08 04:21:24 -08001081static void do_generic_file_read(struct file *filp, loff_t *ppos,
1082 read_descriptor_t *desc, read_actor_t actor)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083{
Christoph Hellwig36e78912008-02-08 04:21:24 -08001084 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -08001086 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001087 pgoff_t index;
1088 pgoff_t last_index;
1089 pgoff_t prev_index;
1090 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -07001091 unsigned int prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 index = *ppos >> PAGE_CACHE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -07001095 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1096 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1098 offset = *ppos & ~PAGE_CACHE_MASK;
1099
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 for (;;) {
1101 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001102 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001103 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 unsigned long nr, ret;
1105
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107find_page:
1108 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001109 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001110 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001111 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001112 index, last_index - index);
1113 page = find_get_page(mapping, index);
1114 if (unlikely(page == NULL))
1115 goto no_cached_page;
1116 }
1117 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001118 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001119 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001120 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001122 if (!PageUptodate(page)) {
1123 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1124 !mapping->a_ops->is_partially_uptodate)
1125 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001126 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001127 goto page_not_up_to_date;
Dave Hansen8d056cb2010-11-11 14:05:15 -08001128 /* Did it get truncated before we got the lock? */
1129 if (!page->mapping)
1130 goto page_not_up_to_date_locked;
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001131 if (!mapping->a_ops->is_partially_uptodate(page,
1132 desc, offset))
1133 goto page_not_up_to_date_locked;
1134 unlock_page(page);
1135 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001137 /*
1138 * i_size must be checked after we know the page is Uptodate.
1139 *
1140 * Checking i_size after the check allows us to calculate
1141 * the correct value for "nr", which means the zero-filled
1142 * part of the page is not copied back to userspace (unless
1143 * another truncate extends the file - this is desired though).
1144 */
1145
1146 isize = i_size_read(inode);
1147 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1148 if (unlikely(!isize || index > end_index)) {
1149 page_cache_release(page);
1150 goto out;
1151 }
1152
1153 /* nr is the maximum number of bytes to copy from this page */
1154 nr = PAGE_CACHE_SIZE;
1155 if (index == end_index) {
1156 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1157 if (nr <= offset) {
1158 page_cache_release(page);
1159 goto out;
1160 }
1161 }
1162 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163
1164 /* If users can be writing to this page using arbitrary
1165 * virtual addresses, take care about potential aliasing
1166 * before reading the page on the kernel side.
1167 */
1168 if (mapping_writably_mapped(mapping))
1169 flush_dcache_page(page);
1170
1171 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001172 * When a sequential read accesses a page several times,
1173 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 */
Jan Karaec0f1632007-05-06 14:49:25 -07001175 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 mark_page_accessed(page);
1177 prev_index = index;
1178
1179 /*
1180 * Ok, we have the page, and it's up-to-date, so
1181 * now we can copy it to user space...
1182 *
1183 * The actor routine returns how many bytes were actually used..
1184 * NOTE! This may not be the same as how much of a user buffer
1185 * we filled up (we may be padding etc), so we can only update
1186 * "pos" here (the actor routine has to update the user buffer
1187 * pointers and the remaining count).
1188 */
1189 ret = actor(desc, page, offset, nr);
1190 offset += ret;
1191 index += offset >> PAGE_CACHE_SHIFT;
1192 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001193 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194
1195 page_cache_release(page);
1196 if (ret == nr && desc->count)
1197 continue;
1198 goto out;
1199
1200page_not_up_to_date:
1201 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001202 error = lock_page_killable(page);
1203 if (unlikely(error))
1204 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001206page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001207 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 if (!page->mapping) {
1209 unlock_page(page);
1210 page_cache_release(page);
1211 continue;
1212 }
1213
1214 /* Did somebody else fill it already? */
1215 if (PageUptodate(page)) {
1216 unlock_page(page);
1217 goto page_ok;
1218 }
1219
1220readpage:
Jeff Moyer91803b42010-05-26 11:49:40 -04001221 /*
1222 * A previous I/O error may have been due to temporary
1223 * failures, eg. multipath errors.
1224 * PG_error will be set again if readpage fails.
1225 */
1226 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 /* Start the actual read. The read will unlock the page. */
1228 error = mapping->a_ops->readpage(filp, page);
1229
Zach Brown994fc28c2005-12-15 14:28:17 -08001230 if (unlikely(error)) {
1231 if (error == AOP_TRUNCATED_PAGE) {
1232 page_cache_release(page);
1233 goto find_page;
1234 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001236 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237
1238 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001239 error = lock_page_killable(page);
1240 if (unlikely(error))
1241 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 if (!PageUptodate(page)) {
1243 if (page->mapping == NULL) {
1244 /*
Christoph Hellwig2ecdc822010-01-26 17:27:20 +01001245 * invalidate_mapping_pages got it
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 */
1247 unlock_page(page);
1248 page_cache_release(page);
1249 goto find_page;
1250 }
1251 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001252 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001253 error = -EIO;
1254 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 }
1256 unlock_page(page);
1257 }
1258
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 goto page_ok;
1260
1261readpage_error:
1262 /* UHHUH! A synchronous read error occurred. Report it */
1263 desc->error = error;
1264 page_cache_release(page);
1265 goto out;
1266
1267no_cached_page:
1268 /*
1269 * Ok, it wasn't cached, so we need to create a new
1270 * page..
1271 */
Nick Piggineb2be182007-10-16 01:24:57 -07001272 page = page_cache_alloc_cold(mapping);
1273 if (!page) {
1274 desc->error = -ENOMEM;
1275 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 }
Nick Piggineb2be182007-10-16 01:24:57 -07001277 error = add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 index, GFP_KERNEL);
1279 if (error) {
Nick Piggineb2be182007-10-16 01:24:57 -07001280 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 if (error == -EEXIST)
1282 goto find_page;
1283 desc->error = error;
1284 goto out;
1285 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 goto readpage;
1287 }
1288
1289out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001290 ra->prev_pos = prev_index;
1291 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1292 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001294 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07001295 file_accessed(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297
1298int file_read_actor(read_descriptor_t *desc, struct page *page,
1299 unsigned long offset, unsigned long size)
1300{
1301 char *kaddr;
1302 unsigned long left, count = desc->count;
1303
1304 if (size > count)
1305 size = count;
1306
1307 /*
1308 * Faults on the destination of a read are common, so do it before
1309 * taking the kmap.
1310 */
1311 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08001312 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 left = __copy_to_user_inatomic(desc->arg.buf,
1314 kaddr + offset, size);
Cong Wang9b04c5f2011-11-25 23:14:39 +08001315 kunmap_atomic(kaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 if (left == 0)
1317 goto success;
1318 }
1319
1320 /* Do it the slow way */
1321 kaddr = kmap(page);
1322 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1323 kunmap(page);
1324
1325 if (left) {
1326 size -= left;
1327 desc->error = -EFAULT;
1328 }
1329success:
1330 desc->count = count - size;
1331 desc->written += size;
1332 desc->arg.buf += size;
1333 return size;
1334}
1335
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001336/*
1337 * Performs necessary checks before doing a write
1338 * @iov: io vector request
1339 * @nr_segs: number of segments in the iovec
1340 * @count: number of bytes to write
1341 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1342 *
1343 * Adjust number of segments and amount of bytes to write (nr_segs should be
1344 * properly initialized first). Returns appropriate error code that caller
1345 * should return or zero in case that write should be allowed.
1346 */
1347int generic_segment_checks(const struct iovec *iov,
1348 unsigned long *nr_segs, size_t *count, int access_flags)
1349{
1350 unsigned long seg;
1351 size_t cnt = 0;
1352 for (seg = 0; seg < *nr_segs; seg++) {
1353 const struct iovec *iv = &iov[seg];
1354
1355 /*
1356 * If any segment has a negative length, or the cumulative
1357 * length ever wraps negative then return -EINVAL.
1358 */
1359 cnt += iv->iov_len;
1360 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1361 return -EINVAL;
1362 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1363 continue;
1364 if (seg == 0)
1365 return -EFAULT;
1366 *nr_segs = seg;
1367 cnt -= iv->iov_len; /* This segment is no good */
1368 break;
1369 }
1370 *count = cnt;
1371 return 0;
1372}
1373EXPORT_SYMBOL(generic_segment_checks);
1374
Randy Dunlap485bb992006-06-23 02:03:49 -07001375/**
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001376 * generic_file_aio_read - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001377 * @iocb: kernel I/O control block
1378 * @iov: io vector request
1379 * @nr_segs: number of segments in the iovec
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001380 * @pos: current file position
Randy Dunlap485bb992006-06-23 02:03:49 -07001381 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 * This is the "read()" routine for all filesystems
1383 * that can use the page cache directly.
1384 */
1385ssize_t
Badari Pulavarty543ade12006-09-30 23:28:48 -07001386generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1387 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388{
1389 struct file *filp = iocb->ki_filp;
1390 ssize_t retval;
Josef Bacik66f998f2010-05-23 11:00:54 -04001391 unsigned long seg = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 size_t count;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001393 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394
1395 count = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001396 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1397 if (retval)
1398 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399
1400 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1401 if (filp->f_flags & O_DIRECT) {
Badari Pulavarty543ade12006-09-30 23:28:48 -07001402 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 struct address_space *mapping;
1404 struct inode *inode;
1405
1406 mapping = filp->f_mapping;
1407 inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 if (!count)
1409 goto out; /* skip atime */
1410 size = i_size_read(inode);
1411 if (pos < size) {
Nick Piggin48b47c52009-01-06 14:40:22 -08001412 retval = filemap_write_and_wait_range(mapping, pos,
1413 pos + iov_length(iov, nr_segs) - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07001414 if (!retval) {
Shaohua Li3deaa712012-02-03 15:37:17 -08001415 struct blk_plug plug;
1416
1417 blk_start_plug(&plug);
Christoph Hellwiga969e902008-07-23 21:27:04 -07001418 retval = mapping->a_ops->direct_IO(READ, iocb,
1419 iov, pos, nr_segs);
Shaohua Li3deaa712012-02-03 15:37:17 -08001420 blk_finish_plug(&plug);
Christoph Hellwiga969e902008-07-23 21:27:04 -07001421 }
Josef Bacik66f998f2010-05-23 11:00:54 -04001422 if (retval > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 *ppos = pos + retval;
Josef Bacik66f998f2010-05-23 11:00:54 -04001424 count -= retval;
1425 }
1426
1427 /*
1428 * Btrfs can have a short DIO read if we encounter
1429 * compressed extents, so if there was an error, or if
1430 * we've already read everything we wanted to, or if
1431 * there was a short read because we hit EOF, go ahead
1432 * and return. Otherwise fallthrough to buffered io for
1433 * the rest of the read.
1434 */
1435 if (retval < 0 || !count || *ppos >= size) {
Hugh Dickins11fa9772008-07-23 21:27:34 -07001436 file_accessed(filp);
1437 goto out;
1438 }
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001439 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 }
1441
Josef Bacik66f998f2010-05-23 11:00:54 -04001442 count = retval;
Hugh Dickins11fa9772008-07-23 21:27:34 -07001443 for (seg = 0; seg < nr_segs; seg++) {
1444 read_descriptor_t desc;
Josef Bacik66f998f2010-05-23 11:00:54 -04001445 loff_t offset = 0;
1446
1447 /*
1448 * If we did a short DIO read we need to skip the section of the
1449 * iov that we've already read data into.
1450 */
1451 if (count) {
1452 if (count > iov[seg].iov_len) {
1453 count -= iov[seg].iov_len;
1454 continue;
1455 }
1456 offset = count;
1457 count = 0;
1458 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459
Hugh Dickins11fa9772008-07-23 21:27:34 -07001460 desc.written = 0;
Josef Bacik66f998f2010-05-23 11:00:54 -04001461 desc.arg.buf = iov[seg].iov_base + offset;
1462 desc.count = iov[seg].iov_len - offset;
Hugh Dickins11fa9772008-07-23 21:27:34 -07001463 if (desc.count == 0)
1464 continue;
1465 desc.error = 0;
1466 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1467 retval += desc.written;
1468 if (desc.error) {
1469 retval = retval ?: desc.error;
1470 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 }
Hugh Dickins11fa9772008-07-23 21:27:34 -07001472 if (desc.count > 0)
1473 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 }
1475out:
1476 return retval;
1477}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478EXPORT_SYMBOL(generic_file_aio_read);
1479
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001481/**
1482 * page_cache_read - adds requested page to the page cache if not already there
1483 * @file: file to read
1484 * @offset: page index
1485 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 * This adds the requested page to the page cache if it isn't already there,
1487 * and schedules an I/O to read in its contents from disk.
1488 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001489static int page_cache_read(struct file *file, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490{
1491 struct address_space *mapping = file->f_mapping;
1492 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001493 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494
Zach Brown994fc28c2005-12-15 14:28:17 -08001495 do {
1496 page = page_cache_alloc_cold(mapping);
1497 if (!page)
1498 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499
Zach Brown994fc28c2005-12-15 14:28:17 -08001500 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1501 if (ret == 0)
1502 ret = mapping->a_ops->readpage(file, page);
1503 else if (ret == -EEXIST)
1504 ret = 0; /* losing race to add is OK */
1505
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507
Zach Brown994fc28c2005-12-15 14:28:17 -08001508 } while (ret == AOP_TRUNCATED_PAGE);
1509
1510 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511}
1512
1513#define MMAP_LOTSAMISS (100)
1514
Linus Torvaldsef00e082009-06-16 15:31:25 -07001515/*
1516 * Synchronous readahead happens when we don't even find
1517 * a page in the page cache at all.
1518 */
1519static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1520 struct file_ra_state *ra,
1521 struct file *file,
1522 pgoff_t offset)
1523{
1524 unsigned long ra_pages;
1525 struct address_space *mapping = file->f_mapping;
1526
1527 /* If we don't want any read-ahead, don't bother */
1528 if (VM_RandomReadHint(vma))
1529 return;
Wu Fengguang275b12b2011-05-24 17:12:28 -07001530 if (!ra->ra_pages)
1531 return;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001532
Wu Fengguang2cbea1d2011-05-24 17:12:30 -07001533 if (VM_SequentialReadHint(vma)) {
Wu Fengguang7ffc59b2009-06-16 15:31:38 -07001534 page_cache_sync_readahead(mapping, ra, file, offset,
1535 ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001536 return;
1537 }
1538
Andi Kleen207d04b2011-05-24 17:12:29 -07001539 /* Avoid banging the cache line if not needed */
1540 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001541 ra->mmap_miss++;
1542
1543 /*
1544 * Do we miss much more than hit in this file? If so,
1545 * stop bothering with read-ahead. It will only hurt.
1546 */
1547 if (ra->mmap_miss > MMAP_LOTSAMISS)
1548 return;
1549
Wu Fengguangd30a1102009-06-16 15:31:30 -07001550 /*
1551 * mmap read-around
1552 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001553 ra_pages = max_sane_readahead(ra->ra_pages);
Wu Fengguang275b12b2011-05-24 17:12:28 -07001554 ra->start = max_t(long, 0, offset - ra_pages / 2);
1555 ra->size = ra_pages;
Wu Fengguang2cbea1d2011-05-24 17:12:30 -07001556 ra->async_size = ra_pages / 4;
Wu Fengguang275b12b2011-05-24 17:12:28 -07001557 ra_submit(ra, mapping, file);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001558}
1559
1560/*
1561 * Asynchronous readahead happens when we find the page and PG_readahead,
1562 * so we want to possibly extend the readahead further..
1563 */
1564static void do_async_mmap_readahead(struct vm_area_struct *vma,
1565 struct file_ra_state *ra,
1566 struct file *file,
1567 struct page *page,
1568 pgoff_t offset)
1569{
1570 struct address_space *mapping = file->f_mapping;
1571
1572 /* If we don't want any read-ahead, don't bother */
1573 if (VM_RandomReadHint(vma))
1574 return;
1575 if (ra->mmap_miss > 0)
1576 ra->mmap_miss--;
1577 if (PageReadahead(page))
Wu Fengguang2fad6f52009-06-16 15:31:29 -07001578 page_cache_async_readahead(mapping, ra, file,
1579 page, offset, ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001580}
1581
Randy Dunlap485bb992006-06-23 02:03:49 -07001582/**
Nick Piggin54cb8822007-07-19 01:46:59 -07001583 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07001584 * @vma: vma in which the fault was taken
1585 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07001586 *
Nick Piggin54cb8822007-07-19 01:46:59 -07001587 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 * mapped memory region to read in file data during a page fault.
1589 *
1590 * The goto's are kind of ugly, but this streamlines the normal case of having
1591 * it in the page cache, and handles the special cases reasonably without
1592 * having a lot of duplicated code.
1593 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001594int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595{
1596 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07001597 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 struct address_space *mapping = file->f_mapping;
1599 struct file_ra_state *ra = &file->f_ra;
1600 struct inode *inode = mapping->host;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001601 pgoff_t offset = vmf->pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 struct page *page;
Jan Kara2004dc82008-02-08 04:20:11 -08001603 pgoff_t size;
Nick Piggin83c54072007-07-19 01:47:05 -07001604 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001607 if (offset >= size)
Linus Torvalds5307cc12007-10-31 09:19:46 -07001608 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 * Do we have something in the page cache already?
1612 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001613 page = find_get_page(mapping, offset);
1614 if (likely(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07001616 * We found the page, so try async readahead before
1617 * waiting for the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001619 do_async_mmap_readahead(vma, ra, file, page, offset);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001620 } else {
1621 /* No page in the page cache at all */
1622 do_sync_mmap_readahead(vma, ra, file, offset);
1623 count_vm_event(PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07001624 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001625 ret = VM_FAULT_MAJOR;
1626retry_find:
Michel Lespinasseb522c942010-10-26 14:21:56 -07001627 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 if (!page)
1629 goto no_cached_page;
1630 }
1631
Michel Lespinassed88c0922010-11-02 13:05:18 -07001632 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1633 page_cache_release(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -07001634 return ret | VM_FAULT_RETRY;
Michel Lespinassed88c0922010-11-02 13:05:18 -07001635 }
Michel Lespinasseb522c942010-10-26 14:21:56 -07001636
1637 /* Did it get truncated? */
1638 if (unlikely(page->mapping != mapping)) {
1639 unlock_page(page);
1640 put_page(page);
1641 goto retry_find;
1642 }
1643 VM_BUG_ON(page->index != offset);
1644
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 /*
Nick Piggind00806b2007-07-19 01:46:57 -07001646 * We have a locked page in the page cache, now we need to check
1647 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 */
Nick Piggind00806b2007-07-19 01:46:57 -07001649 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 goto page_not_uptodate;
1651
Linus Torvaldsef00e082009-06-16 15:31:25 -07001652 /*
1653 * Found the page and have a reference on it.
1654 * We must recheck i_size under page lock.
1655 */
Nick Piggind00806b2007-07-19 01:46:57 -07001656 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001657 if (unlikely(offset >= size)) {
Nick Piggind00806b2007-07-19 01:46:57 -07001658 unlock_page(page);
Yan Zheng745ad482007-10-08 10:08:37 -07001659 page_cache_release(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07001660 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001661 }
1662
Nick Piggind0217ac2007-07-19 01:47:03 -07001663 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07001664 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666no_cached_page:
1667 /*
1668 * We're only likely to ever get here if MADV_RANDOM is in
1669 * effect.
1670 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001671 error = page_cache_read(file, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672
1673 /*
1674 * The page we want has now been added to the page cache.
1675 * In the unlikely event that someone removed it in the
1676 * meantime, we'll just come back here and read it again.
1677 */
1678 if (error >= 0)
1679 goto retry_find;
1680
1681 /*
1682 * An error return from page_cache_read can result if the
1683 * system is low on memory, or a problem occurs while trying
1684 * to schedule I/O.
1685 */
1686 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07001687 return VM_FAULT_OOM;
1688 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689
1690page_not_uptodate:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 /*
1692 * Umm, take care of errors if the page isn't up-to-date.
1693 * Try to re-read it _once_. We do this synchronously,
1694 * because there really aren't any performance issues here
1695 * and we need to check for errors.
1696 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001698 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07001699 if (!error) {
1700 wait_on_page_locked(page);
1701 if (!PageUptodate(page))
1702 error = -EIO;
1703 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07001705
1706 if (!error || error == AOP_TRUNCATED_PAGE)
1707 goto retry_find;
1708
1709 /* Things didn't work out. Return zero to tell the mm layer so. */
1710 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07001711 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07001712}
1713EXPORT_SYMBOL(filemap_fault);
1714
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001715const struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07001716 .fault = filemap_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717};
1718
1719/* This is used for a general mmap of a disk file */
1720
1721int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1722{
1723 struct address_space *mapping = file->f_mapping;
1724
1725 if (!mapping->a_ops->readpage)
1726 return -ENOEXEC;
1727 file_accessed(file);
1728 vma->vm_ops = &generic_file_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001729 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 return 0;
1731}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732
1733/*
1734 * This is for filesystems which do not implement ->writepage.
1735 */
1736int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1737{
1738 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1739 return -EINVAL;
1740 return generic_file_mmap(file, vma);
1741}
1742#else
1743int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1744{
1745 return -ENOSYS;
1746}
1747int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1748{
1749 return -ENOSYS;
1750}
1751#endif /* CONFIG_MMU */
1752
1753EXPORT_SYMBOL(generic_file_mmap);
1754EXPORT_SYMBOL(generic_file_readonly_mmap);
1755
Nick Piggin6fe69002007-05-06 14:49:04 -07001756static struct page *__read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001757 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001758 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001759 void *data,
1760 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761{
Nick Piggineb2be182007-10-16 01:24:57 -07001762 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 int err;
1764repeat:
1765 page = find_get_page(mapping, index);
1766 if (!page) {
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001767 page = __page_cache_alloc(gfp | __GFP_COLD);
Nick Piggineb2be182007-10-16 01:24:57 -07001768 if (!page)
1769 return ERR_PTR(-ENOMEM);
Dave Kleikampe6f67b82011-12-21 11:05:48 -06001770 err = add_to_page_cache_lru(page, mapping, index, gfp);
Nick Piggineb2be182007-10-16 01:24:57 -07001771 if (unlikely(err)) {
1772 page_cache_release(page);
1773 if (err == -EEXIST)
1774 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 return ERR_PTR(err);
1777 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 err = filler(data, page);
1779 if (err < 0) {
1780 page_cache_release(page);
1781 page = ERR_PTR(err);
1782 }
1783 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 return page;
1785}
1786
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001787static struct page *do_read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001788 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001789 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001790 void *data,
1791 gfp_t gfp)
1792
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793{
1794 struct page *page;
1795 int err;
1796
1797retry:
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001798 page = __read_cache_page(mapping, index, filler, data, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01001800 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 if (PageUptodate(page))
1802 goto out;
1803
1804 lock_page(page);
1805 if (!page->mapping) {
1806 unlock_page(page);
1807 page_cache_release(page);
1808 goto retry;
1809 }
1810 if (PageUptodate(page)) {
1811 unlock_page(page);
1812 goto out;
1813 }
1814 err = filler(data, page);
1815 if (err < 0) {
1816 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01001817 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 }
David Howellsc855ff32007-05-09 13:42:20 +01001819out:
Nick Piggin6fe69002007-05-06 14:49:04 -07001820 mark_page_accessed(page);
1821 return page;
1822}
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001823
1824/**
1825 * read_cache_page_async - read into page cache, fill it if needed
1826 * @mapping: the page's address_space
1827 * @index: the page index
1828 * @filler: function to perform the read
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001829 * @data: first arg to filler(data, page) function, often left as NULL
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001830 *
1831 * Same as read_cache_page, but don't wait for page to become unlocked
1832 * after submitting it to the filler.
1833 *
1834 * Read into the page cache. If a page already exists, and PageUptodate() is
1835 * not set, try to fill the page but don't wait for it to become unlocked.
1836 *
1837 * If the page does not get brought uptodate, return -EIO.
1838 */
1839struct page *read_cache_page_async(struct address_space *mapping,
1840 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001841 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001842 void *data)
1843{
1844 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
1845}
Nick Piggin6fe69002007-05-06 14:49:04 -07001846EXPORT_SYMBOL(read_cache_page_async);
1847
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001848static struct page *wait_on_page_read(struct page *page)
1849{
1850 if (!IS_ERR(page)) {
1851 wait_on_page_locked(page);
1852 if (!PageUptodate(page)) {
1853 page_cache_release(page);
1854 page = ERR_PTR(-EIO);
1855 }
1856 }
1857 return page;
1858}
1859
1860/**
1861 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
1862 * @mapping: the page's address_space
1863 * @index: the page index
1864 * @gfp: the page allocator flags to use if allocating
1865 *
1866 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
Dave Kleikampe6f67b82011-12-21 11:05:48 -06001867 * any new page allocations done using the specified allocation flags.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001868 *
1869 * If the page does not get brought uptodate, return -EIO.
1870 */
1871struct page *read_cache_page_gfp(struct address_space *mapping,
1872 pgoff_t index,
1873 gfp_t gfp)
1874{
1875 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
1876
1877 return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp));
1878}
1879EXPORT_SYMBOL(read_cache_page_gfp);
1880
Nick Piggin6fe69002007-05-06 14:49:04 -07001881/**
1882 * read_cache_page - read into page cache, fill it if needed
1883 * @mapping: the page's address_space
1884 * @index: the page index
1885 * @filler: function to perform the read
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001886 * @data: first arg to filler(data, page) function, often left as NULL
Nick Piggin6fe69002007-05-06 14:49:04 -07001887 *
1888 * Read into the page cache. If a page already exists, and PageUptodate() is
1889 * not set, try to fill the page then wait for it to become unlocked.
1890 *
1891 * If the page does not get brought uptodate, return -EIO.
1892 */
1893struct page *read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001894 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001895 int (*filler)(void *, struct page *),
Nick Piggin6fe69002007-05-06 14:49:04 -07001896 void *data)
1897{
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001898 return wait_on_page_read(read_cache_page_async(mapping, index, filler, data));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900EXPORT_SYMBOL(read_cache_page);
1901
Nick Piggin2f718ff2007-10-16 01:24:59 -07001902static size_t __iovec_copy_from_user_inatomic(char *vaddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 const struct iovec *iov, size_t base, size_t bytes)
1904{
Ingo Molnarf1800532009-03-02 11:00:57 +01001905 size_t copied = 0, left = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906
1907 while (bytes) {
1908 char __user *buf = iov->iov_base + base;
1909 int copy = min(bytes, iov->iov_len - base);
1910
1911 base = 0;
Ingo Molnarf1800532009-03-02 11:00:57 +01001912 left = __copy_from_user_inatomic(vaddr, buf, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 copied += copy;
1914 bytes -= copy;
1915 vaddr += copy;
1916 iov++;
1917
NeilBrown01408c42006-06-25 05:47:58 -07001918 if (unlikely(left))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920 }
1921 return copied - left;
1922}
1923
1924/*
Nick Piggin2f718ff2007-10-16 01:24:59 -07001925 * Copy as much as we can into the page and return the number of bytes which
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02001926 * were successfully copied. If a fault is encountered then return the number of
Nick Piggin2f718ff2007-10-16 01:24:59 -07001927 * bytes which were copied.
1928 */
1929size_t iov_iter_copy_from_user_atomic(struct page *page,
1930 struct iov_iter *i, unsigned long offset, size_t bytes)
1931{
1932 char *kaddr;
1933 size_t copied;
1934
1935 BUG_ON(!in_atomic());
Cong Wang9b04c5f2011-11-25 23:14:39 +08001936 kaddr = kmap_atomic(page);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001937 if (likely(i->nr_segs == 1)) {
1938 int left;
1939 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01001940 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001941 copied = bytes - left;
1942 } else {
1943 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1944 i->iov, i->iov_offset, bytes);
1945 }
Cong Wang9b04c5f2011-11-25 23:14:39 +08001946 kunmap_atomic(kaddr);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001947
1948 return copied;
1949}
Nick Piggin89e10782007-10-16 01:25:07 -07001950EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001951
1952/*
1953 * This has the same sideeffects and return value as
1954 * iov_iter_copy_from_user_atomic().
1955 * The difference is that it attempts to resolve faults.
1956 * Page must not be locked.
1957 */
1958size_t iov_iter_copy_from_user(struct page *page,
1959 struct iov_iter *i, unsigned long offset, size_t bytes)
1960{
1961 char *kaddr;
1962 size_t copied;
1963
1964 kaddr = kmap(page);
1965 if (likely(i->nr_segs == 1)) {
1966 int left;
1967 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01001968 left = __copy_from_user(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001969 copied = bytes - left;
1970 } else {
1971 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1972 i->iov, i->iov_offset, bytes);
1973 }
1974 kunmap(page);
1975 return copied;
1976}
Nick Piggin89e10782007-10-16 01:25:07 -07001977EXPORT_SYMBOL(iov_iter_copy_from_user);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001978
Nick Pigginf7009262008-03-10 11:43:59 -07001979void iov_iter_advance(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07001980{
Nick Pigginf7009262008-03-10 11:43:59 -07001981 BUG_ON(i->count < bytes);
1982
Nick Piggin2f718ff2007-10-16 01:24:59 -07001983 if (likely(i->nr_segs == 1)) {
1984 i->iov_offset += bytes;
Nick Pigginf7009262008-03-10 11:43:59 -07001985 i->count -= bytes;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001986 } else {
1987 const struct iovec *iov = i->iov;
1988 size_t base = i->iov_offset;
Jeff Layton39be79c2011-10-27 23:53:08 +02001989 unsigned long nr_segs = i->nr_segs;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001990
Nick Piggin124d3b72008-02-02 15:01:17 +01001991 /*
1992 * The !iov->iov_len check ensures we skip over unlikely
Nick Pigginf7009262008-03-10 11:43:59 -07001993 * zero-length segments (without overruning the iovec).
Nick Piggin124d3b72008-02-02 15:01:17 +01001994 */
Linus Torvalds94ad3742008-07-30 14:45:12 -07001995 while (bytes || unlikely(i->count && !iov->iov_len)) {
Nick Pigginf7009262008-03-10 11:43:59 -07001996 int copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001997
Nick Pigginf7009262008-03-10 11:43:59 -07001998 copy = min(bytes, iov->iov_len - base);
1999 BUG_ON(!i->count || i->count < copy);
2000 i->count -= copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002001 bytes -= copy;
2002 base += copy;
2003 if (iov->iov_len == base) {
2004 iov++;
Jeff Layton39be79c2011-10-27 23:53:08 +02002005 nr_segs--;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002006 base = 0;
2007 }
2008 }
2009 i->iov = iov;
2010 i->iov_offset = base;
Jeff Layton39be79c2011-10-27 23:53:08 +02002011 i->nr_segs = nr_segs;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002012 }
2013}
Nick Piggin89e10782007-10-16 01:25:07 -07002014EXPORT_SYMBOL(iov_iter_advance);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002015
Nick Pigginafddba42007-10-16 01:25:01 -07002016/*
2017 * Fault in the first iovec of the given iov_iter, to a maximum length
2018 * of bytes. Returns 0 on success, or non-zero if the memory could not be
2019 * accessed (ie. because it is an invalid address).
2020 *
2021 * writev-intensive code may want this to prefault several iovecs -- that
2022 * would be possible (callers must not rely on the fact that _only_ the
2023 * first iovec will be faulted with the current implementation).
2024 */
2025int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07002026{
Nick Piggin2f718ff2007-10-16 01:24:59 -07002027 char __user *buf = i->iov->iov_base + i->iov_offset;
Nick Pigginafddba42007-10-16 01:25:01 -07002028 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
2029 return fault_in_pages_readable(buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002030}
Nick Piggin89e10782007-10-16 01:25:07 -07002031EXPORT_SYMBOL(iov_iter_fault_in_readable);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002032
2033/*
2034 * Return the count of just the current iov_iter segment.
2035 */
2036size_t iov_iter_single_seg_count(struct iov_iter *i)
2037{
2038 const struct iovec *iov = i->iov;
2039 if (i->nr_segs == 1)
2040 return i->count;
2041 else
2042 return min(i->count, iov->iov_len - i->iov_offset);
2043}
Nick Piggin89e10782007-10-16 01:25:07 -07002044EXPORT_SYMBOL(iov_iter_single_seg_count);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002045
2046/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 * Performs necessary checks before doing a write
2048 *
Randy Dunlap485bb992006-06-23 02:03:49 -07002049 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 * Returns appropriate error code that caller should return or
2051 * zero in case that write should be allowed.
2052 */
2053inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2054{
2055 struct inode *inode = file->f_mapping->host;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002056 unsigned long limit = rlimit(RLIMIT_FSIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057
2058 if (unlikely(*pos < 0))
2059 return -EINVAL;
2060
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 if (!isblk) {
2062 /* FIXME: this is for backwards compatibility with 2.4 */
2063 if (file->f_flags & O_APPEND)
2064 *pos = i_size_read(inode);
2065
2066 if (limit != RLIM_INFINITY) {
2067 if (*pos >= limit) {
2068 send_sig(SIGXFSZ, current, 0);
2069 return -EFBIG;
2070 }
2071 if (*count > limit - (typeof(limit))*pos) {
2072 *count = limit - (typeof(limit))*pos;
2073 }
2074 }
2075 }
2076
2077 /*
2078 * LFS rule
2079 */
2080 if (unlikely(*pos + *count > MAX_NON_LFS &&
2081 !(file->f_flags & O_LARGEFILE))) {
2082 if (*pos >= MAX_NON_LFS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 return -EFBIG;
2084 }
2085 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2086 *count = MAX_NON_LFS - (unsigned long)*pos;
2087 }
2088 }
2089
2090 /*
2091 * Are we about to exceed the fs block limit ?
2092 *
2093 * If we have written data it becomes a short write. If we have
2094 * exceeded without writing data we send a signal and return EFBIG.
2095 * Linus frestrict idea will clean these up nicely..
2096 */
2097 if (likely(!isblk)) {
2098 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2099 if (*count || *pos > inode->i_sb->s_maxbytes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 return -EFBIG;
2101 }
2102 /* zero-length writes at ->s_maxbytes are OK */
2103 }
2104
2105 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2106 *count = inode->i_sb->s_maxbytes - *pos;
2107 } else {
David Howells93614012006-09-30 20:45:40 +02002108#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 loff_t isize;
2110 if (bdev_read_only(I_BDEV(inode)))
2111 return -EPERM;
2112 isize = i_size_read(inode);
2113 if (*pos >= isize) {
2114 if (*count || *pos > isize)
2115 return -ENOSPC;
2116 }
2117
2118 if (*pos + *count > isize)
2119 *count = isize - *pos;
David Howells93614012006-09-30 20:45:40 +02002120#else
2121 return -EPERM;
2122#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 }
2124 return 0;
2125}
2126EXPORT_SYMBOL(generic_write_checks);
2127
Nick Pigginafddba42007-10-16 01:25:01 -07002128int pagecache_write_begin(struct file *file, struct address_space *mapping,
2129 loff_t pos, unsigned len, unsigned flags,
2130 struct page **pagep, void **fsdata)
2131{
2132 const struct address_space_operations *aops = mapping->a_ops;
2133
Nick Piggin4e02ed42008-10-29 14:00:55 -07002134 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002135 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002136}
2137EXPORT_SYMBOL(pagecache_write_begin);
2138
2139int pagecache_write_end(struct file *file, struct address_space *mapping,
2140 loff_t pos, unsigned len, unsigned copied,
2141 struct page *page, void *fsdata)
2142{
2143 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002144
Nick Piggin4e02ed42008-10-29 14:00:55 -07002145 mark_page_accessed(page);
2146 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002147}
2148EXPORT_SYMBOL(pagecache_write_end);
2149
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150ssize_t
2151generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2152 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2153 size_t count, size_t ocount)
2154{
2155 struct file *file = iocb->ki_filp;
2156 struct address_space *mapping = file->f_mapping;
2157 struct inode *inode = mapping->host;
2158 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002159 size_t write_len;
2160 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161
2162 if (count != ocount)
2163 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2164
Christoph Hellwiga969e902008-07-23 21:27:04 -07002165 write_len = iov_length(iov, *nr_segs);
2166 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002167
Nick Piggin48b47c52009-01-06 14:40:22 -08002168 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002169 if (written)
2170 goto out;
2171
2172 /*
2173 * After a write we want buffered reads to be sure to go to disk to get
2174 * the new data. We invalidate clean cached page from the region we're
2175 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002176 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002177 */
2178 if (mapping->nrpages) {
2179 written = invalidate_inode_pages2_range(mapping,
2180 pos >> PAGE_CACHE_SHIFT, end);
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002181 /*
2182 * If a page can not be invalidated, return 0 to fall back
2183 * to buffered write.
2184 */
2185 if (written) {
2186 if (written == -EBUSY)
2187 return 0;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002188 goto out;
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002189 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002190 }
2191
2192 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2193
2194 /*
2195 * Finally, try again to invalidate clean pages which might have been
2196 * cached by non-direct readahead, or faulted in by get_user_pages()
2197 * if the source of the write was an mmap'ed region of the file
2198 * we're writing. Either one is a pretty crazy thing to do,
2199 * so we don't support it 100%. If this invalidation
2200 * fails, tough, the write still worked...
2201 */
2202 if (mapping->nrpages) {
2203 invalidate_inode_pages2_range(mapping,
2204 pos >> PAGE_CACHE_SHIFT, end);
2205 }
2206
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 if (written > 0) {
Namhyung Kim01166512010-10-26 14:21:58 -07002208 pos += written;
2209 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2210 i_size_write(inode, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 mark_inode_dirty(inode);
2212 }
Namhyung Kim01166512010-10-26 14:21:58 -07002213 *ppos = pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002215out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216 return written;
2217}
2218EXPORT_SYMBOL(generic_file_direct_write);
2219
Nick Piggineb2be182007-10-16 01:24:57 -07002220/*
2221 * Find or create a page at the given pagecache position. Return the locked
2222 * page. This function is specifically for buffered writes.
2223 */
Nick Piggin54566b22009-01-04 12:00:53 -08002224struct page *grab_cache_page_write_begin(struct address_space *mapping,
2225 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07002226{
2227 int status;
Johannes Weiner0faa70c2012-01-10 15:07:53 -08002228 gfp_t gfp_mask;
Nick Piggineb2be182007-10-16 01:24:57 -07002229 struct page *page;
Nick Piggin54566b22009-01-04 12:00:53 -08002230 gfp_t gfp_notmask = 0;
Johannes Weiner0faa70c2012-01-10 15:07:53 -08002231
Fengguang Wu1010bb12012-03-21 16:34:08 -07002232 gfp_mask = mapping_gfp_mask(mapping);
2233 if (mapping_cap_account_dirty(mapping))
2234 gfp_mask |= __GFP_WRITE;
Nick Piggin54566b22009-01-04 12:00:53 -08002235 if (flags & AOP_FLAG_NOFS)
2236 gfp_notmask = __GFP_FS;
Nick Piggineb2be182007-10-16 01:24:57 -07002237repeat:
2238 page = find_lock_page(mapping, index);
Steven Rostedtc585a262011-01-13 15:46:18 -08002239 if (page)
Darrick J. Wong3d08bcc2011-05-27 12:23:34 -07002240 goto found;
Nick Piggineb2be182007-10-16 01:24:57 -07002241
Johannes Weiner0faa70c2012-01-10 15:07:53 -08002242 page = __page_cache_alloc(gfp_mask & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002243 if (!page)
2244 return NULL;
Nick Piggin54566b22009-01-04 12:00:53 -08002245 status = add_to_page_cache_lru(page, mapping, index,
2246 GFP_KERNEL & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002247 if (unlikely(status)) {
2248 page_cache_release(page);
2249 if (status == -EEXIST)
2250 goto repeat;
2251 return NULL;
2252 }
Darrick J. Wong3d08bcc2011-05-27 12:23:34 -07002253found:
2254 wait_on_page_writeback(page);
Nick Piggineb2be182007-10-16 01:24:57 -07002255 return page;
2256}
Nick Piggin54566b22009-01-04 12:00:53 -08002257EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07002258
Nick Pigginafddba42007-10-16 01:25:01 -07002259static ssize_t generic_perform_write(struct file *file,
2260 struct iov_iter *i, loff_t pos)
2261{
2262 struct address_space *mapping = file->f_mapping;
2263 const struct address_space_operations *a_ops = mapping->a_ops;
2264 long status = 0;
2265 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002266 unsigned int flags = 0;
2267
2268 /*
2269 * Copies from kernel address space cannot fail (NFSD is a big user).
2270 */
2271 if (segment_eq(get_fs(), KERNEL_DS))
2272 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002273
2274 do {
2275 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002276 unsigned long offset; /* Offset into pagecache page */
2277 unsigned long bytes; /* Bytes to write to page */
2278 size_t copied; /* Bytes copied from user */
2279 void *fsdata;
2280
2281 offset = (pos & (PAGE_CACHE_SIZE - 1));
Nick Pigginafddba42007-10-16 01:25:01 -07002282 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2283 iov_iter_count(i));
2284
2285again:
Nick Pigginafddba42007-10-16 01:25:01 -07002286 /*
2287 * Bring in the user page that we will copy from _first_.
2288 * Otherwise there's a nasty deadlock on copying from the
2289 * same page as we're writing to, without it being marked
2290 * up-to-date.
2291 *
2292 * Not only is this an optimisation, but it is also required
2293 * to check that the address is actually valid, when atomic
2294 * usercopies are used, below.
2295 */
2296 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2297 status = -EFAULT;
2298 break;
2299 }
2300
Nick Piggin674b8922007-10-16 01:25:03 -07002301 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002302 &page, &fsdata);
2303 if (unlikely(status))
2304 break;
2305
anfei zhou931e80e2010-02-02 13:44:02 -08002306 if (mapping_writably_mapped(mapping))
2307 flush_dcache_page(page);
2308
Nick Pigginafddba42007-10-16 01:25:01 -07002309 pagefault_disable();
2310 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2311 pagefault_enable();
2312 flush_dcache_page(page);
2313
Josef Bacikc8236db2009-07-05 12:08:18 -07002314 mark_page_accessed(page);
Nick Pigginafddba42007-10-16 01:25:01 -07002315 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2316 page, fsdata);
2317 if (unlikely(status < 0))
2318 break;
2319 copied = status;
2320
2321 cond_resched();
2322
Nick Piggin124d3b72008-02-02 15:01:17 +01002323 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002324 if (unlikely(copied == 0)) {
2325 /*
2326 * If we were unable to copy any data at all, we must
2327 * fall back to a single segment length write.
2328 *
2329 * If we didn't fallback here, we could livelock
2330 * because not all segments in the iov can be copied at
2331 * once without a pagefault.
2332 */
2333 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2334 iov_iter_single_seg_count(i));
2335 goto again;
2336 }
Nick Pigginafddba42007-10-16 01:25:01 -07002337 pos += copied;
2338 written += copied;
2339
2340 balance_dirty_pages_ratelimited(mapping);
Jan Karaa50527b2011-12-02 09:17:02 +08002341 if (fatal_signal_pending(current)) {
2342 status = -EINTR;
2343 break;
2344 }
Nick Pigginafddba42007-10-16 01:25:01 -07002345 } while (iov_iter_count(i));
2346
2347 return written ? written : status;
2348}
2349
2350ssize_t
2351generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2352 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2353 size_t count, ssize_t written)
2354{
2355 struct file *file = iocb->ki_filp;
Nick Pigginafddba42007-10-16 01:25:01 -07002356 ssize_t status;
2357 struct iov_iter i;
2358
2359 iov_iter_init(&i, iov, nr_segs, count, written);
Nick Piggin4e02ed42008-10-29 14:00:55 -07002360 status = generic_perform_write(file, &i, pos);
Nick Pigginafddba42007-10-16 01:25:01 -07002361
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362 if (likely(status >= 0)) {
Nick Pigginafddba42007-10-16 01:25:01 -07002363 written += status;
2364 *ppos = pos + status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365 }
2366
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367 return written ? written : status;
2368}
2369EXPORT_SYMBOL(generic_file_buffered_write);
2370
Jan Karae4dd9de2009-08-17 18:10:06 +02002371/**
2372 * __generic_file_aio_write - write data to a file
2373 * @iocb: IO state structure (file, offset, etc.)
2374 * @iov: vector with data to write
2375 * @nr_segs: number of segments in the vector
2376 * @ppos: position where to write
2377 *
2378 * This function does all the work needed for actually writing data to a
2379 * file. It does all basic checks, removes SUID from the file, updates
2380 * modification times and calls proper subroutines depending on whether we
2381 * do direct IO or a standard buffered write.
2382 *
2383 * It expects i_mutex to be grabbed unless we work on a block device or similar
2384 * object which does not need locking at all.
2385 *
2386 * This function does *not* take care of syncing data in case of O_SYNC write.
2387 * A caller has to handle it. This is mainly due to the fact that we want to
2388 * avoid syncing under i_mutex.
2389 */
2390ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2391 unsigned long nr_segs, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392{
2393 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002394 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395 size_t ocount; /* original count */
2396 size_t count; /* after file limit checks */
2397 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398 loff_t pos;
2399 ssize_t written;
2400 ssize_t err;
2401
2402 ocount = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07002403 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2404 if (err)
2405 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406
2407 count = ocount;
2408 pos = *ppos;
2409
2410 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2411
2412 /* We can write back this queue in page reclaim */
2413 current->backing_dev_info = mapping->backing_dev_info;
2414 written = 0;
2415
2416 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2417 if (err)
2418 goto out;
2419
2420 if (count == 0)
2421 goto out;
2422
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02002423 err = file_remove_suid(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 if (err)
2425 goto out;
2426
Josef Bacikc3b2da32012-03-26 09:59:21 -04002427 err = file_update_time(file);
2428 if (err)
2429 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430
2431 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2432 if (unlikely(file->f_flags & O_DIRECT)) {
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002433 loff_t endbyte;
2434 ssize_t written_buffered;
2435
2436 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2437 ppos, count, ocount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 if (written < 0 || written == count)
2439 goto out;
2440 /*
2441 * direct-io write to a hole: fall through to buffered I/O
2442 * for completing the rest of the request.
2443 */
2444 pos += written;
2445 count -= written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002446 written_buffered = generic_file_buffered_write(iocb, iov,
2447 nr_segs, pos, ppos, count,
2448 written);
2449 /*
2450 * If generic_file_buffered_write() retuned a synchronous error
2451 * then we want to return the number of bytes which were
2452 * direct-written, or the error code if that was zero. Note
2453 * that this differs from normal direct-io semantics, which
2454 * will return -EFOO even if some bytes were written.
2455 */
2456 if (written_buffered < 0) {
2457 err = written_buffered;
2458 goto out;
2459 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002461 /*
2462 * We need to ensure that the page cache pages are written to
2463 * disk and invalidated to preserve the expected O_DIRECT
2464 * semantics.
2465 */
2466 endbyte = pos + written_buffered - written - 1;
Christoph Hellwigc05c4ed2009-09-23 15:07:30 +02002467 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002468 if (err == 0) {
2469 written = written_buffered;
2470 invalidate_mapping_pages(mapping,
2471 pos >> PAGE_CACHE_SHIFT,
2472 endbyte >> PAGE_CACHE_SHIFT);
2473 } else {
2474 /*
2475 * We don't know how much we wrote, so just return
2476 * the number of bytes which were direct-written
2477 */
2478 }
2479 } else {
2480 written = generic_file_buffered_write(iocb, iov, nr_segs,
2481 pos, ppos, count, written);
2482 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483out:
2484 current->backing_dev_info = NULL;
2485 return written ? written : err;
2486}
Jan Karae4dd9de2009-08-17 18:10:06 +02002487EXPORT_SYMBOL(__generic_file_aio_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488
Jan Karae4dd9de2009-08-17 18:10:06 +02002489/**
2490 * generic_file_aio_write - write data to a file
2491 * @iocb: IO state structure
2492 * @iov: vector with data to write
2493 * @nr_segs: number of segments in the vector
2494 * @pos: position in file where to write
2495 *
2496 * This is a wrapper around __generic_file_aio_write() to be used by most
2497 * filesystems. It takes care of syncing the file in case of O_SYNC file
2498 * and acquires i_mutex as needed.
2499 */
Badari Pulavarty027445c2006-09-30 23:28:46 -07002500ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2501 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502{
2503 struct file *file = iocb->ki_filp;
Jan Kara148f9482009-08-17 19:52:36 +02002504 struct inode *inode = file->f_mapping->host;
Jens Axboe55602dd2010-06-24 15:05:37 +02002505 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507
2508 BUG_ON(iocb->ki_pos != pos);
2509
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002510 mutex_lock(&inode->i_mutex);
Jens Axboe55602dd2010-06-24 15:05:37 +02002511 blk_start_plug(&plug);
Jan Karae4dd9de2009-08-17 18:10:06 +02002512 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002513 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514
Jan Kara148f9482009-08-17 19:52:36 +02002515 if (ret > 0 || ret == -EIOCBQUEUED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516 ssize_t err;
2517
Jan Kara148f9482009-08-17 19:52:36 +02002518 err = generic_write_sync(file, pos, ret);
Jan Karac7b50db2009-08-18 16:18:20 +02002519 if (err < 0 && ret > 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520 ret = err;
2521 }
Jens Axboe55602dd2010-06-24 15:05:37 +02002522 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523 return ret;
2524}
2525EXPORT_SYMBOL(generic_file_aio_write);
2526
David Howellscf9a2ae2006-08-29 19:05:54 +01002527/**
2528 * try_to_release_page() - release old fs-specific metadata on a page
2529 *
2530 * @page: the page which the kernel is trying to free
2531 * @gfp_mask: memory allocation flags (and I/O mode)
2532 *
2533 * The address_space is to try to release any data against the page
2534 * (presumably at page->private). If the release was successful, return `1'.
2535 * Otherwise return zero.
2536 *
David Howells266cf652009-04-03 16:42:36 +01002537 * This may also be called if PG_fscache is set on a page, indicating that the
2538 * page is known to the local caching routines.
2539 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002540 * The @gfp_mask argument specifies whether I/O may be performed to release
Mingming Cao3f31fdd2008-07-25 01:46:22 -07002541 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01002542 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002543 */
2544int try_to_release_page(struct page *page, gfp_t gfp_mask)
2545{
2546 struct address_space * const mapping = page->mapping;
2547
2548 BUG_ON(!PageLocked(page));
2549 if (PageWriteback(page))
2550 return 0;
2551
2552 if (mapping && mapping->a_ops->releasepage)
2553 return mapping->a_ops->releasepage(page, gfp_mask);
2554 return try_to_free_buffers(page);
2555}
2556
2557EXPORT_SYMBOL(try_to_release_page);