blob: 5cf820a7c8ec2c1c0e8744eb03750e35b9255ef9 [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 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/compiler.h>
14#include <linux/fs.h>
Hiro Yoshiokac22ce142006-06-23 02:04:16 -070015#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/aio.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080017#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/kernel_stat.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090019#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/mm.h>
21#include <linux/swap.h>
22#include <linux/mman.h>
23#include <linux/pagemap.h>
24#include <linux/file.h>
25#include <linux/uio.h>
26#include <linux/hash.h>
27#include <linux/writeback.h>
Linus Torvalds53253382007-10-18 14:47:32 -070028#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/pagevec.h>
30#include <linux/blkdev.h>
31#include <linux/security.h>
32#include <linux/syscalls.h>
Paul Jackson44110fe2006-03-24 03:16:04 -080033#include <linux/cpuset.h>
Nick Piggin2f718ff2007-10-16 01:24:59 -070034#include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080035#include <linux/memcontrol.h>
Dan Magenheimerc515e1f2011-05-26 10:01:43 -060036#include <linux/cleancache.h>
Nick Piggin0f8053a2006-03-22 00:08:33 -080037#include "internal.h"
38
Linus Torvalds1da177e2005-04-16 15:20:36 -070039/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070040 * FIXME: remove all knowledge of the buffer layer from the core VM
41 */
Jan Kara148f9482009-08-17 19:52:36 +020042#include <linux/buffer_head.h> /* for try_to_free_buffers */
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include <asm/mman.h>
45
46/*
47 * Shared mappings implemented 30.11.1994. It's not fully working yet,
48 * though.
49 *
50 * Shared mappings now work. 15.8.1995 Bruno.
51 *
52 * finished 'unifying' the page and buffer cache and SMP-threaded the
53 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
54 *
55 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
56 */
57
58/*
59 * Lock ordering:
60 *
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070061 * ->i_mmap_mutex (truncate_pagecache)
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 * ->private_lock (__free_pte->__set_page_dirty_buffers)
Hugh Dickins5d337b92005-09-03 15:54:41 -070063 * ->swap_lock (exclusive_swap_page, others)
64 * ->mapping->tree_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080066 * ->i_mutex
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070067 * ->i_mmap_mutex (truncate->unmap_mapping_range)
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 *
69 * ->mmap_sem
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070070 * ->i_mmap_mutex
Hugh Dickinsb8072f02005-10-29 18:16:41 -070071 * ->page_table_lock or pte_lock (various, mainly in memory.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
73 *
74 * ->mmap_sem
75 * ->lock_page (access_process_vm)
76 *
Nick Piggin82591e62006-10-19 23:29:10 -070077 * ->i_mutex (generic_file_buffered_write)
78 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 *
Christoph Hellwigf758eea2011-04-21 18:19:44 -060080 * bdi->wb.list_lock
Dave Chinnera66979a2011-03-22 22:23:41 +110081 * sb_lock (fs/fs-writeback.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 * ->mapping->tree_lock (__sync_single_inode)
83 *
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070084 * ->i_mmap_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -070085 * ->anon_vma.lock (vma_adjust)
86 *
87 * ->anon_vma.lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070088 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -070090 * ->page_table_lock or pte_lock
Hugh Dickins5d337b92005-09-03 15:54:41 -070091 * ->swap_lock (try_to_unmap_one)
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 * ->private_lock (try_to_unmap_one)
93 * ->tree_lock (try_to_unmap_one)
94 * ->zone.lru_lock (follow_page->mark_page_accessed)
Nick Piggin053837f2006-01-18 17:42:27 -080095 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 * ->private_lock (page_remove_rmap->set_page_dirty)
97 * ->tree_lock (page_remove_rmap->set_page_dirty)
Christoph Hellwigf758eea2011-04-21 18:19:44 -060098 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +110099 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600100 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100101 * ->inode->i_lock (zap_pte_range->set_page_dirty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
103 *
Andi Kleen6a460792009-09-16 11:50:15 +0200104 * (code doesn't rely on that order, so you could switch it around)
105 * ->tasklist_lock (memory_failure, collect_procs_ao)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700106 * ->i_mmap_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 */
108
109/*
Minchan Kime64a7822011-03-22 16:32:44 -0700110 * Delete a page from the page cache and free it. Caller has to make
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 * sure the page is locked and that nobody else uses it - or that usage
Nick Piggin19fd6232008-07-25 19:45:32 -0700112 * is safe. The caller must hold the mapping's tree_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 */
Minchan Kime64a7822011-03-22 16:32:44 -0700114void __delete_from_page_cache(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115{
116 struct address_space *mapping = page->mapping;
117
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600118 /*
119 * if we're uptodate, flush out into the cleancache, otherwise
120 * invalidate any existing cleancache entries. We can't leave
121 * stale data around in the cleancache once our page is gone
122 */
123 if (PageUptodate(page) && PageMappedToDisk(page))
124 cleancache_put_page(page);
125 else
126 cleancache_flush_page(mapping, page);
127
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 radix_tree_delete(&mapping->page_tree, page->index);
129 page->mapping = NULL;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700130 /* Leave page->index set: truncation lookup relies upon it */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 mapping->nrpages--;
Christoph Lameter347ce432006-06-30 01:55:35 -0700132 __dec_zone_page_state(page, NR_FILE_PAGES);
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700133 if (PageSwapBacked(page))
134 __dec_zone_page_state(page, NR_SHMEM);
Nick Piggin45426812007-07-15 23:38:12 -0700135 BUG_ON(page_mapped(page));
Linus Torvalds3a692792007-12-19 14:05:13 -0800136
137 /*
138 * Some filesystems seem to re-dirty the page even after
139 * the VM has canceled the dirty bit (eg ext3 journaling).
140 *
141 * Fix it up by doing a final dirty accounting check after
142 * having removed the page entirely.
143 */
144 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
145 dec_zone_page_state(page, NR_FILE_DIRTY);
146 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
147 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148}
149
Minchan Kim702cfbf2011-03-22 16:32:43 -0700150/**
151 * delete_from_page_cache - delete page from page cache
152 * @page: the page which the kernel is trying to remove from page cache
153 *
154 * This must be called only on pages that have been verified to be in the page
155 * cache and locked. It will never put the page into the free list, the caller
156 * has a reference on the page.
157 */
158void delete_from_page_cache(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159{
160 struct address_space *mapping = page->mapping;
Linus Torvalds6072d132010-12-01 13:35:19 -0500161 void (*freepage)(struct page *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162
Matt Mackallcd7619d2005-05-01 08:59:01 -0700163 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164
Linus Torvalds6072d132010-12-01 13:35:19 -0500165 freepage = mapping->a_ops->freepage;
Nick Piggin19fd6232008-07-25 19:45:32 -0700166 spin_lock_irq(&mapping->tree_lock);
Minchan Kime64a7822011-03-22 16:32:44 -0700167 __delete_from_page_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700168 spin_unlock_irq(&mapping->tree_lock);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700169 mem_cgroup_uncharge_cache_page(page);
Linus Torvalds6072d132010-12-01 13:35:19 -0500170
171 if (freepage)
172 freepage(page);
Minchan Kim97cecb52011-03-22 16:30:53 -0700173 page_cache_release(page);
174}
175EXPORT_SYMBOL(delete_from_page_cache);
176
Jens Axboe7eaceac2011-03-10 08:52:07 +0100177static int sleep_on_page(void *word)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 io_schedule();
180 return 0;
181}
182
Jens Axboe7eaceac2011-03-10 08:52:07 +0100183static int sleep_on_page_killable(void *word)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500184{
Jens Axboe7eaceac2011-03-10 08:52:07 +0100185 sleep_on_page(word);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500186 return fatal_signal_pending(current) ? -EINTR : 0;
187}
188
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700190 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700191 * @mapping: address space structure to write
192 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800193 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700194 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700196 * Start writeback against all of a mapping's dirty pages that lie
197 * within the byte offsets <start, end> inclusive.
198 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700200 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 * these two operations is that if a dirty page/buffer is encountered, it must
202 * be waited upon, and not just skipped over.
203 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800204int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
205 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206{
207 int ret;
208 struct writeback_control wbc = {
209 .sync_mode = sync_mode,
Nick Piggin05fe4782009-01-06 14:39:08 -0800210 .nr_to_write = LONG_MAX,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700211 .range_start = start,
212 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 };
214
215 if (!mapping_cap_writeback_dirty(mapping))
216 return 0;
217
218 ret = do_writepages(mapping, &wbc);
219 return ret;
220}
221
222static inline int __filemap_fdatawrite(struct address_space *mapping,
223 int sync_mode)
224{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700225 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226}
227
228int filemap_fdatawrite(struct address_space *mapping)
229{
230 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
231}
232EXPORT_SYMBOL(filemap_fdatawrite);
233
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400234int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800235 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236{
237 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
238}
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400239EXPORT_SYMBOL(filemap_fdatawrite_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240
Randy Dunlap485bb992006-06-23 02:03:49 -0700241/**
242 * filemap_flush - mostly a non-blocking flush
243 * @mapping: target address_space
244 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 * This is a mostly non-blocking flush. Not suitable for data-integrity
246 * purposes - I/O may not be started against all dirty pages.
247 */
248int filemap_flush(struct address_space *mapping)
249{
250 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
251}
252EXPORT_SYMBOL(filemap_flush);
253
Randy Dunlap485bb992006-06-23 02:03:49 -0700254/**
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200255 * filemap_fdatawait_range - wait for writeback to complete
256 * @mapping: address space structure to wait for
257 * @start_byte: offset in bytes where the range starts
258 * @end_byte: offset in bytes where the range ends (inclusive)
Randy Dunlap485bb992006-06-23 02:03:49 -0700259 *
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200260 * Walk the list of under-writeback pages of the given address space
261 * in the given range and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 */
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200263int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
264 loff_t end_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265{
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200266 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
267 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 struct pagevec pvec;
269 int nr_pages;
270 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200272 if (end_byte < start_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 return 0;
274
275 pagevec_init(&pvec, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 while ((index <= end) &&
277 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
278 PAGECACHE_TAG_WRITEBACK,
279 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
280 unsigned i;
281
282 for (i = 0; i < nr_pages; i++) {
283 struct page *page = pvec.pages[i];
284
285 /* until radix tree lookup accepts end_index */
286 if (page->index > end)
287 continue;
288
289 wait_on_page_writeback(page);
Rik van Riel212260a2011-01-13 15:46:06 -0800290 if (TestClearPageError(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 ret = -EIO;
292 }
293 pagevec_release(&pvec);
294 cond_resched();
295 }
296
297 /* Check for outstanding write errors */
298 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
299 ret = -ENOSPC;
300 if (test_and_clear_bit(AS_EIO, &mapping->flags))
301 ret = -EIO;
302
303 return ret;
304}
Jan Karad3bccb6f2009-08-17 19:30:27 +0200305EXPORT_SYMBOL(filemap_fdatawait_range);
306
307/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700308 * filemap_fdatawait - wait for all under-writeback pages to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 * @mapping: address space structure to wait for
Randy Dunlap485bb992006-06-23 02:03:49 -0700310 *
311 * Walk the list of under-writeback pages of the given address space
312 * and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 */
314int filemap_fdatawait(struct address_space *mapping)
315{
316 loff_t i_size = i_size_read(mapping->host);
317
318 if (i_size == 0)
319 return 0;
320
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200321 return filemap_fdatawait_range(mapping, 0, i_size - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322}
323EXPORT_SYMBOL(filemap_fdatawait);
324
325int filemap_write_and_wait(struct address_space *mapping)
326{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800327 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328
329 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800330 err = filemap_fdatawrite(mapping);
331 /*
332 * Even if the above returned error, the pages may be
333 * written partially (e.g. -ENOSPC), so we wait for it.
334 * But the -EIO is special case, it may indicate the worst
335 * thing (e.g. bug) happened, so we avoid waiting for it.
336 */
337 if (err != -EIO) {
338 int err2 = filemap_fdatawait(mapping);
339 if (!err)
340 err = err2;
341 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800343 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800345EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346
Randy Dunlap485bb992006-06-23 02:03:49 -0700347/**
348 * filemap_write_and_wait_range - write out & wait on a file range
349 * @mapping: the address_space for the pages
350 * @lstart: offset in bytes where the range starts
351 * @lend: offset in bytes where the range ends (inclusive)
352 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800353 * Write out and wait upon file offsets lstart->lend, inclusive.
354 *
355 * Note that `lend' is inclusive (describes the last byte to be written) so
356 * that this function can be used to write to the very end-of-file (end = -1).
357 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358int filemap_write_and_wait_range(struct address_space *mapping,
359 loff_t lstart, loff_t lend)
360{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800361 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362
363 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800364 err = __filemap_fdatawrite_range(mapping, lstart, lend,
365 WB_SYNC_ALL);
366 /* See comment of filemap_write_and_wait() */
367 if (err != -EIO) {
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200368 int err2 = filemap_fdatawait_range(mapping,
369 lstart, lend);
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800370 if (!err)
371 err = err2;
372 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800374 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375}
Chris Masonf6995582009-04-15 13:22:37 -0400376EXPORT_SYMBOL(filemap_write_and_wait_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377
Randy Dunlap485bb992006-06-23 02:03:49 -0700378/**
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700379 * replace_page_cache_page - replace a pagecache page with a new one
380 * @old: page to be replaced
381 * @new: page to replace with
382 * @gfp_mask: allocation mode
383 *
384 * This function replaces a page in the pagecache with a new one. On
385 * success it acquires the pagecache reference for the new page and
386 * drops it for the old page. Both the old and new pages must be
387 * locked. This function does not add the new page to the LRU, the
388 * caller must do that.
389 *
390 * The remove + add is atomic. The only way this function can fail is
391 * memory allocation failure.
392 */
393int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
394{
395 int error;
396 struct mem_cgroup *memcg = NULL;
397
398 VM_BUG_ON(!PageLocked(old));
399 VM_BUG_ON(!PageLocked(new));
400 VM_BUG_ON(new->mapping);
401
402 /*
403 * This is not page migration, but prepare_migration and
404 * end_migration does enough work for charge replacement.
405 *
406 * In the longer term we probably want a specialized function
407 * for moving the charge from old to new in a more efficient
408 * manner.
409 */
410 error = mem_cgroup_prepare_migration(old, new, &memcg, gfp_mask);
411 if (error)
412 return error;
413
414 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
415 if (!error) {
416 struct address_space *mapping = old->mapping;
417 void (*freepage)(struct page *);
418
419 pgoff_t offset = old->index;
420 freepage = mapping->a_ops->freepage;
421
422 page_cache_get(new);
423 new->mapping = mapping;
424 new->index = offset;
425
426 spin_lock_irq(&mapping->tree_lock);
Minchan Kime64a7822011-03-22 16:32:44 -0700427 __delete_from_page_cache(old);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700428 error = radix_tree_insert(&mapping->page_tree, offset, new);
429 BUG_ON(error);
430 mapping->nrpages++;
431 __inc_zone_page_state(new, NR_FILE_PAGES);
432 if (PageSwapBacked(new))
433 __inc_zone_page_state(new, NR_SHMEM);
434 spin_unlock_irq(&mapping->tree_lock);
435 radix_tree_preload_end();
436 if (freepage)
437 freepage(old);
438 page_cache_release(old);
439 mem_cgroup_end_migration(memcg, old, new, true);
440 } else {
441 mem_cgroup_end_migration(memcg, old, new, false);
442 }
443
444 return error;
445}
446EXPORT_SYMBOL_GPL(replace_page_cache_page);
447
448/**
Nick Piggine2867812008-07-25 19:45:30 -0700449 * add_to_page_cache_locked - add a locked page to the pagecache
Randy Dunlap485bb992006-06-23 02:03:49 -0700450 * @page: page to add
451 * @mapping: the page's address_space
452 * @offset: page index
453 * @gfp_mask: page allocation mode
454 *
Nick Piggine2867812008-07-25 19:45:30 -0700455 * This function is used to add a page to the pagecache. It must be locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 * This function does not add the page to the LRU. The caller must do that.
457 */
Nick Piggine2867812008-07-25 19:45:30 -0700458int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400459 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460{
Nick Piggine2867812008-07-25 19:45:30 -0700461 int error;
462
463 VM_BUG_ON(!PageLocked(page));
Hugh Dickins31475dd2011-08-03 16:21:27 -0700464 VM_BUG_ON(PageSwapBacked(page));
Nick Piggine2867812008-07-25 19:45:30 -0700465
466 error = mem_cgroup_cache_charge(page, current->mm,
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -0800467 gfp_mask & GFP_RECLAIM_MASK);
Balbir Singh35c754d2008-02-07 00:14:05 -0800468 if (error)
469 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470
Balbir Singh35c754d2008-02-07 00:14:05 -0800471 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 if (error == 0) {
Nick Piggine2867812008-07-25 19:45:30 -0700473 page_cache_get(page);
474 page->mapping = mapping;
475 page->index = offset;
476
Nick Piggin19fd6232008-07-25 19:45:32 -0700477 spin_lock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 error = radix_tree_insert(&mapping->page_tree, offset, page);
Nick Piggine2867812008-07-25 19:45:30 -0700479 if (likely(!error)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 mapping->nrpages++;
Christoph Lameter347ce432006-06-30 01:55:35 -0700481 __inc_zone_page_state(page, NR_FILE_PAGES);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700482 spin_unlock_irq(&mapping->tree_lock);
Nick Piggine2867812008-07-25 19:45:30 -0700483 } else {
484 page->mapping = NULL;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700485 /* Leave page->index set: truncation relies upon it */
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700486 spin_unlock_irq(&mapping->tree_lock);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700487 mem_cgroup_uncharge_cache_page(page);
Nick Piggine2867812008-07-25 19:45:30 -0700488 page_cache_release(page);
489 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 radix_tree_preload_end();
Balbir Singh35c754d2008-02-07 00:14:05 -0800491 } else
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700492 mem_cgroup_uncharge_cache_page(page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800493out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 return error;
495}
Nick Piggine2867812008-07-25 19:45:30 -0700496EXPORT_SYMBOL(add_to_page_cache_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497
498int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400499 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500{
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700501 int ret;
502
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700503 ret = add_to_page_cache(page, mapping, offset, gfp_mask);
Hugh Dickins31475dd2011-08-03 16:21:27 -0700504 if (ret == 0)
505 lru_cache_add_file(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 return ret;
507}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300508EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509
Paul Jackson44110fe2006-03-24 03:16:04 -0800510#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700511struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800512{
Miao Xiec0ff7452010-05-24 14:32:08 -0700513 int n;
514 struct page *page;
515
Paul Jackson44110fe2006-03-24 03:16:04 -0800516 if (cpuset_do_page_mem_spread()) {
Miao Xiec0ff7452010-05-24 14:32:08 -0700517 get_mems_allowed();
518 n = cpuset_mem_spread_node();
519 page = alloc_pages_exact_node(n, gfp, 0);
520 put_mems_allowed();
521 return page;
Paul Jackson44110fe2006-03-24 03:16:04 -0800522 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700523 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800524}
Nick Piggin2ae88142006-10-28 10:38:23 -0700525EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800526#endif
527
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528/*
529 * In order to wait for pages to become available there must be
530 * waitqueues associated with pages. By using a hash table of
531 * waitqueues where the bucket discipline is to maintain all
532 * waiters on the same queue and wake all when any of the pages
533 * become available, and for the woken contexts to check to be
534 * sure the appropriate page became available, this saves space
535 * at a cost of "thundering herd" phenomena during rare hash
536 * collisions.
537 */
538static wait_queue_head_t *page_waitqueue(struct page *page)
539{
540 const struct zone *zone = page_zone(page);
541
542 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
543}
544
545static inline void wake_up_page(struct page *page, int bit)
546{
547 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
548}
549
Harvey Harrison920c7a52008-02-04 22:29:26 -0800550void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551{
552 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
553
554 if (test_bit(bit_nr, &page->flags))
Jens Axboe7eaceac2011-03-10 08:52:07 +0100555 __wait_on_bit(page_waitqueue(page), &wait, sleep_on_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 TASK_UNINTERRUPTIBLE);
557}
558EXPORT_SYMBOL(wait_on_page_bit);
559
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700560int wait_on_page_bit_killable(struct page *page, int bit_nr)
561{
562 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
563
564 if (!test_bit(bit_nr, &page->flags))
565 return 0;
566
567 return __wait_on_bit(page_waitqueue(page), &wait,
568 sleep_on_page_killable, TASK_KILLABLE);
569}
570
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571/**
David Howells385e1ca5f2009-04-03 16:42:39 +0100572 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
Randy Dunlap697f6192009-04-13 14:39:54 -0700573 * @page: Page defining the wait queue of interest
574 * @waiter: Waiter to add to the queue
David Howells385e1ca5f2009-04-03 16:42:39 +0100575 *
576 * Add an arbitrary @waiter to the wait queue for the nominated @page.
577 */
578void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
579{
580 wait_queue_head_t *q = page_waitqueue(page);
581 unsigned long flags;
582
583 spin_lock_irqsave(&q->lock, flags);
584 __add_wait_queue(q, waiter);
585 spin_unlock_irqrestore(&q->lock, flags);
586}
587EXPORT_SYMBOL_GPL(add_page_wait_queue);
588
589/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700590 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 * @page: the page
592 *
593 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
594 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
595 * mechananism between PageLocked pages and PageWriteback pages is shared.
596 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
597 *
Nick Piggin8413ac92008-10-18 20:26:59 -0700598 * The mb is necessary to enforce ordering between the clear_bit and the read
599 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800601void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602{
Nick Piggin8413ac92008-10-18 20:26:59 -0700603 VM_BUG_ON(!PageLocked(page));
604 clear_bit_unlock(PG_locked, &page->flags);
605 smp_mb__after_clear_bit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 wake_up_page(page, PG_locked);
607}
608EXPORT_SYMBOL(unlock_page);
609
Randy Dunlap485bb992006-06-23 02:03:49 -0700610/**
611 * end_page_writeback - end writeback against a page
612 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 */
614void end_page_writeback(struct page *page)
615{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700616 if (TestClearPageReclaim(page))
617 rotate_reclaimable_page(page);
618
619 if (!test_clear_page_writeback(page))
620 BUG();
621
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 smp_mb__after_clear_bit();
623 wake_up_page(page, PG_writeback);
624}
625EXPORT_SYMBOL(end_page_writeback);
626
Randy Dunlap485bb992006-06-23 02:03:49 -0700627/**
628 * __lock_page - get a lock on the page, assuming we need to sleep to get it
629 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800631void __lock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632{
633 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
634
Jens Axboe7eaceac2011-03-10 08:52:07 +0100635 __wait_on_bit_lock(page_waitqueue(page), &wait, sleep_on_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 TASK_UNINTERRUPTIBLE);
637}
638EXPORT_SYMBOL(__lock_page);
639
Harvey Harrisonb5606c22008-02-13 15:03:16 -0800640int __lock_page_killable(struct page *page)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500641{
642 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
643
644 return __wait_on_bit_lock(page_waitqueue(page), &wait,
Jens Axboe7eaceac2011-03-10 08:52:07 +0100645 sleep_on_page_killable, TASK_KILLABLE);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500646}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300647EXPORT_SYMBOL_GPL(__lock_page_killable);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500648
Michel Lespinassed065bd82010-10-26 14:21:57 -0700649int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
650 unsigned int flags)
651{
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700652 if (flags & FAULT_FLAG_ALLOW_RETRY) {
653 /*
654 * CAUTION! In this case, mmap_sem is not released
655 * even though return 0.
656 */
657 if (flags & FAULT_FLAG_RETRY_NOWAIT)
658 return 0;
659
660 up_read(&mm->mmap_sem);
661 if (flags & FAULT_FLAG_KILLABLE)
662 wait_on_page_locked_killable(page);
663 else
Gleb Natapov318b2752011-03-22 16:30:51 -0700664 wait_on_page_locked(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -0700665 return 0;
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700666 } else {
667 if (flags & FAULT_FLAG_KILLABLE) {
668 int ret;
669
670 ret = __lock_page_killable(page);
671 if (ret) {
672 up_read(&mm->mmap_sem);
673 return 0;
674 }
675 } else
676 __lock_page(page);
677 return 1;
Michel Lespinassed065bd82010-10-26 14:21:57 -0700678 }
679}
680
Randy Dunlap485bb992006-06-23 02:03:49 -0700681/**
682 * find_get_page - find and get a page reference
683 * @mapping: the address_space to search
684 * @offset: the page index
685 *
Nick Pigginda6052f2006-09-25 23:31:35 -0700686 * Is there a pagecache struct page at the given (mapping, offset) tuple?
687 * If yes, increment its refcount and return it; if no, return NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 */
Nick Piggina60637c2008-07-25 19:45:31 -0700689struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690{
Nick Piggina60637c2008-07-25 19:45:31 -0700691 void **pagep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 struct page *page;
693
Nick Piggina60637c2008-07-25 19:45:31 -0700694 rcu_read_lock();
695repeat:
696 page = NULL;
697 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
698 if (pagep) {
699 page = radix_tree_deref_slot(pagep);
Nick Piggin27d20fd2010-11-11 14:05:19 -0800700 if (unlikely(!page))
701 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700702 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700703 if (radix_tree_deref_retry(page))
704 goto repeat;
705 /*
706 * Otherwise, shmem/tmpfs must be storing a swap entry
707 * here as an exceptional entry: so return it without
708 * attempting to raise page count.
709 */
710 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700711 }
Nick Piggina60637c2008-07-25 19:45:31 -0700712 if (!page_cache_get_speculative(page))
713 goto repeat;
714
715 /*
716 * Has the page moved?
717 * This is part of the lockless pagecache protocol. See
718 * include/linux/pagemap.h for details.
719 */
720 if (unlikely(page != *pagep)) {
721 page_cache_release(page);
722 goto repeat;
723 }
724 }
Nick Piggin27d20fd2010-11-11 14:05:19 -0800725out:
Nick Piggina60637c2008-07-25 19:45:31 -0700726 rcu_read_unlock();
727
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 return page;
729}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730EXPORT_SYMBOL(find_get_page);
731
Randy Dunlap485bb992006-06-23 02:03:49 -0700732/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 * find_lock_page - locate, pin and lock a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700734 * @mapping: the address_space to search
735 * @offset: the page index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 *
737 * Locates the desired pagecache page, locks it, increments its reference
738 * count and returns its address.
739 *
740 * Returns zero if the page was not present. find_lock_page() may sleep.
741 */
Nick Piggina60637c2008-07-25 19:45:31 -0700742struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743{
744 struct page *page;
745
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746repeat:
Nick Piggina60637c2008-07-25 19:45:31 -0700747 page = find_get_page(mapping, offset);
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700748 if (page && !radix_tree_exception(page)) {
Nick Piggina60637c2008-07-25 19:45:31 -0700749 lock_page(page);
750 /* Has the page been truncated? */
751 if (unlikely(page->mapping != mapping)) {
752 unlock_page(page);
753 page_cache_release(page);
754 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 }
Nick Piggina60637c2008-07-25 19:45:31 -0700756 VM_BUG_ON(page->index != offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 return page;
759}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760EXPORT_SYMBOL(find_lock_page);
761
762/**
763 * find_or_create_page - locate or add a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700764 * @mapping: the page's address_space
765 * @index: the page's index into the mapping
766 * @gfp_mask: page allocation mode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 *
768 * Locates a page in the pagecache. If the page is not present, a new page
769 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
770 * LRU list. The returned page is locked and has its reference count
771 * incremented.
772 *
773 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
774 * allocation!
775 *
776 * find_or_create_page() returns the desired page's address, or zero on
777 * memory exhaustion.
778 */
779struct page *find_or_create_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -0700780 pgoff_t index, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781{
Nick Piggineb2be182007-10-16 01:24:57 -0700782 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 int err;
784repeat:
785 page = find_lock_page(mapping, index);
786 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -0700787 page = __page_cache_alloc(gfp_mask);
788 if (!page)
789 return NULL;
Nick Piggin67d58ac2009-01-06 14:40:28 -0800790 /*
791 * We want a regular kernel memory (not highmem or DMA etc)
792 * allocation for the radix tree nodes, but we need to honour
793 * the context-specific requirements the caller has asked for.
794 * GFP_RECLAIM_MASK collects those requirements.
795 */
796 err = add_to_page_cache_lru(page, mapping, index,
797 (gfp_mask & GFP_RECLAIM_MASK));
Nick Piggineb2be182007-10-16 01:24:57 -0700798 if (unlikely(err)) {
799 page_cache_release(page);
800 page = NULL;
801 if (err == -EEXIST)
802 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 return page;
806}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807EXPORT_SYMBOL(find_or_create_page);
808
809/**
810 * find_get_pages - gang pagecache lookup
811 * @mapping: The address_space to search
812 * @start: The starting page index
813 * @nr_pages: The maximum number of pages
814 * @pages: Where the resulting pages are placed
815 *
816 * find_get_pages() will search for and return a group of up to
817 * @nr_pages pages in the mapping. The pages are placed at @pages.
818 * find_get_pages() takes a reference against the returned pages.
819 *
820 * The search returns a group of mapping-contiguous pages with ascending
821 * indexes. There may be holes in the indices due to not-present pages.
822 *
823 * find_get_pages() returns the number of pages which were found.
824 */
825unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
826 unsigned int nr_pages, struct page **pages)
827{
828 unsigned int i;
829 unsigned int ret;
Shaohua Licc39c6a2011-09-15 08:45:19 +0800830 unsigned int nr_found, nr_skip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831
Nick Piggina60637c2008-07-25 19:45:31 -0700832 rcu_read_lock();
833restart:
834 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
Hugh Dickins63286502011-08-03 16:21:18 -0700835 (void ***)pages, NULL, start, nr_pages);
Nick Piggina60637c2008-07-25 19:45:31 -0700836 ret = 0;
Shaohua Licc39c6a2011-09-15 08:45:19 +0800837 nr_skip = 0;
Nick Piggina60637c2008-07-25 19:45:31 -0700838 for (i = 0; i < nr_found; i++) {
839 struct page *page;
840repeat:
841 page = radix_tree_deref_slot((void **)pages[i]);
842 if (unlikely(!page))
843 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -0700844
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700845 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700846 if (radix_tree_deref_retry(page)) {
847 /*
848 * Transient condition which can only trigger
849 * when entry at index 0 moves out of or back
850 * to root: none yet gotten, safe to restart.
851 */
852 WARN_ON(start | i);
853 goto restart;
854 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700855 /*
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700856 * Otherwise, shmem/tmpfs must be storing a swap entry
857 * here as an exceptional entry: so skip over it -
858 * we only reach this from invalidate_mapping_pages().
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700859 */
Shaohua Licc39c6a2011-09-15 08:45:19 +0800860 nr_skip++;
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700861 continue;
Nick Piggin27d20fd2010-11-11 14:05:19 -0800862 }
Nick Piggina60637c2008-07-25 19:45:31 -0700863
864 if (!page_cache_get_speculative(page))
865 goto repeat;
866
867 /* Has the page moved? */
868 if (unlikely(page != *((void **)pages[i]))) {
869 page_cache_release(page);
870 goto repeat;
871 }
872
873 pages[ret] = page;
874 ret++;
875 }
Hugh Dickins5b280c02011-03-22 16:33:07 -0700876
877 /*
878 * If all entries were removed before we could secure them,
879 * try again, because callers stop trying once 0 is returned.
880 */
Shaohua Licc39c6a2011-09-15 08:45:19 +0800881 if (unlikely(!ret && nr_found > nr_skip))
Hugh Dickins5b280c02011-03-22 16:33:07 -0700882 goto restart;
Nick Piggina60637c2008-07-25 19:45:31 -0700883 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 return ret;
885}
886
Jens Axboeebf43502006-04-27 08:46:01 +0200887/**
888 * find_get_pages_contig - gang contiguous pagecache lookup
889 * @mapping: The address_space to search
890 * @index: The starting page index
891 * @nr_pages: The maximum number of pages
892 * @pages: Where the resulting pages are placed
893 *
894 * find_get_pages_contig() works exactly like find_get_pages(), except
895 * that the returned number of pages are guaranteed to be contiguous.
896 *
897 * find_get_pages_contig() returns the number of pages which were found.
898 */
899unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
900 unsigned int nr_pages, struct page **pages)
901{
902 unsigned int i;
903 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700904 unsigned int nr_found;
Jens Axboeebf43502006-04-27 08:46:01 +0200905
Nick Piggina60637c2008-07-25 19:45:31 -0700906 rcu_read_lock();
907restart:
908 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
Hugh Dickins63286502011-08-03 16:21:18 -0700909 (void ***)pages, NULL, index, nr_pages);
Nick Piggina60637c2008-07-25 19:45:31 -0700910 ret = 0;
911 for (i = 0; i < nr_found; i++) {
912 struct page *page;
913repeat:
914 page = radix_tree_deref_slot((void **)pages[i]);
915 if (unlikely(!page))
916 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -0700917
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700918 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700919 if (radix_tree_deref_retry(page)) {
920 /*
921 * Transient condition which can only trigger
922 * when entry at index 0 moves out of or back
923 * to root: none yet gotten, safe to restart.
924 */
925 goto restart;
926 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700927 /*
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700928 * Otherwise, shmem/tmpfs must be storing a swap entry
929 * here as an exceptional entry: so stop looking for
930 * contiguous pages.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700931 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700932 break;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700933 }
Nick Piggina60637c2008-07-25 19:45:31 -0700934
Nick Piggina60637c2008-07-25 19:45:31 -0700935 if (!page_cache_get_speculative(page))
936 goto repeat;
937
938 /* Has the page moved? */
939 if (unlikely(page != *((void **)pages[i]))) {
940 page_cache_release(page);
941 goto repeat;
942 }
943
Nick Piggin9cbb4cb2011-01-13 15:45:51 -0800944 /*
945 * must check mapping and index after taking the ref.
946 * otherwise we can get both false positives and false
947 * negatives, which is just confusing to the caller.
948 */
949 if (page->mapping == NULL || page->index != index) {
950 page_cache_release(page);
951 break;
952 }
953
Nick Piggina60637c2008-07-25 19:45:31 -0700954 pages[ret] = page;
955 ret++;
Jens Axboeebf43502006-04-27 08:46:01 +0200956 index++;
957 }
Nick Piggina60637c2008-07-25 19:45:31 -0700958 rcu_read_unlock();
959 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +0200960}
David Howellsef71c152007-05-09 02:33:44 -0700961EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +0200962
Randy Dunlap485bb992006-06-23 02:03:49 -0700963/**
964 * find_get_pages_tag - find and return pages that match @tag
965 * @mapping: the address_space to search
966 * @index: the starting page index
967 * @tag: the tag index
968 * @nr_pages: the maximum number of pages
969 * @pages: where the resulting pages are placed
970 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -0700972 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 */
974unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
975 int tag, unsigned int nr_pages, struct page **pages)
976{
977 unsigned int i;
978 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700979 unsigned int nr_found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980
Nick Piggina60637c2008-07-25 19:45:31 -0700981 rcu_read_lock();
982restart:
983 nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree,
984 (void ***)pages, *index, nr_pages, tag);
985 ret = 0;
986 for (i = 0; i < nr_found; i++) {
987 struct page *page;
988repeat:
989 page = radix_tree_deref_slot((void **)pages[i]);
990 if (unlikely(!page))
991 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -0700992
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700993 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700994 if (radix_tree_deref_retry(page)) {
995 /*
996 * Transient condition which can only trigger
997 * when entry at index 0 moves out of or back
998 * to root: none yet gotten, safe to restart.
999 */
1000 goto restart;
1001 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001002 /*
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001003 * This function is never used on a shmem/tmpfs
1004 * mapping, so a swap entry won't be found here.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001005 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001006 BUG();
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001007 }
Nick Piggina60637c2008-07-25 19:45:31 -07001008
1009 if (!page_cache_get_speculative(page))
1010 goto repeat;
1011
1012 /* Has the page moved? */
1013 if (unlikely(page != *((void **)pages[i]))) {
1014 page_cache_release(page);
1015 goto repeat;
1016 }
1017
1018 pages[ret] = page;
1019 ret++;
1020 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001021
1022 /*
1023 * If all entries were removed before we could secure them,
1024 * try again, because callers stop trying once 0 is returned.
1025 */
1026 if (unlikely(!ret && nr_found))
1027 goto restart;
Nick Piggina60637c2008-07-25 19:45:31 -07001028 rcu_read_unlock();
1029
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 if (ret)
1031 *index = pages[ret - 1]->index + 1;
Nick Piggina60637c2008-07-25 19:45:31 -07001032
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 return ret;
1034}
David Howellsef71c152007-05-09 02:33:44 -07001035EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036
Randy Dunlap485bb992006-06-23 02:03:49 -07001037/**
1038 * grab_cache_page_nowait - returns locked page at given index in given cache
1039 * @mapping: target address_space
1040 * @index: the page index
1041 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08001042 * Same as grab_cache_page(), but do not wait if the page is unavailable.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 * This is intended for speculative data generators, where the data can
1044 * be regenerated if the page couldn't be grabbed. This routine should
1045 * be safe to call while holding the lock for another page.
1046 *
1047 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
1048 * and deadlock against the caller's locked page.
1049 */
1050struct page *
Fengguang Wu57f6b962007-10-16 01:24:37 -07001051grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052{
1053 struct page *page = find_get_page(mapping, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054
1055 if (page) {
Nick Piggin529ae9a2008-08-02 12:01:03 +02001056 if (trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 return page;
1058 page_cache_release(page);
1059 return NULL;
1060 }
Nick Piggin2ae88142006-10-28 10:38:23 -07001061 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
Nick Piggin67d58ac2009-01-06 14:40:28 -08001062 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 page_cache_release(page);
1064 page = NULL;
1065 }
1066 return page;
1067}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068EXPORT_SYMBOL(grab_cache_page_nowait);
1069
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001070/*
1071 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1072 * a _large_ part of the i/o request. Imagine the worst scenario:
1073 *
1074 * ---R__________________________________________B__________
1075 * ^ reading here ^ bad block(assume 4k)
1076 *
1077 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1078 * => failing the whole request => read(R) => read(R+1) =>
1079 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1080 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1081 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1082 *
1083 * It is going insane. Fix it by quickly scaling down the readahead size.
1084 */
1085static void shrink_readahead_size_eio(struct file *filp,
1086 struct file_ra_state *ra)
1087{
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001088 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001089}
1090
Randy Dunlap485bb992006-06-23 02:03:49 -07001091/**
Christoph Hellwig36e78912008-02-08 04:21:24 -08001092 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001093 * @filp: the file to read
1094 * @ppos: current file position
1095 * @desc: read_descriptor
1096 * @actor: read method
1097 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -07001099 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 *
1101 * This is really ugly. But the goto's actually try to clarify some
1102 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 */
Christoph Hellwig36e78912008-02-08 04:21:24 -08001104static void do_generic_file_read(struct file *filp, loff_t *ppos,
1105 read_descriptor_t *desc, read_actor_t actor)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106{
Christoph Hellwig36e78912008-02-08 04:21:24 -08001107 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -08001109 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001110 pgoff_t index;
1111 pgoff_t last_index;
1112 pgoff_t prev_index;
1113 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -07001114 unsigned int prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 index = *ppos >> PAGE_CACHE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -07001118 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1119 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1121 offset = *ppos & ~PAGE_CACHE_MASK;
1122
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 for (;;) {
1124 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001125 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001126 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 unsigned long nr, ret;
1128
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130find_page:
1131 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001132 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001133 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001134 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001135 index, last_index - index);
1136 page = find_get_page(mapping, index);
1137 if (unlikely(page == NULL))
1138 goto no_cached_page;
1139 }
1140 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001141 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001142 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001143 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001145 if (!PageUptodate(page)) {
1146 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1147 !mapping->a_ops->is_partially_uptodate)
1148 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001149 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001150 goto page_not_up_to_date;
Dave Hansen8d056cb2010-11-11 14:05:15 -08001151 /* Did it get truncated before we got the lock? */
1152 if (!page->mapping)
1153 goto page_not_up_to_date_locked;
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001154 if (!mapping->a_ops->is_partially_uptodate(page,
1155 desc, offset))
1156 goto page_not_up_to_date_locked;
1157 unlock_page(page);
1158 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001160 /*
1161 * i_size must be checked after we know the page is Uptodate.
1162 *
1163 * Checking i_size after the check allows us to calculate
1164 * the correct value for "nr", which means the zero-filled
1165 * part of the page is not copied back to userspace (unless
1166 * another truncate extends the file - this is desired though).
1167 */
1168
1169 isize = i_size_read(inode);
1170 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1171 if (unlikely(!isize || index > end_index)) {
1172 page_cache_release(page);
1173 goto out;
1174 }
1175
1176 /* nr is the maximum number of bytes to copy from this page */
1177 nr = PAGE_CACHE_SIZE;
1178 if (index == end_index) {
1179 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1180 if (nr <= offset) {
1181 page_cache_release(page);
1182 goto out;
1183 }
1184 }
1185 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186
1187 /* If users can be writing to this page using arbitrary
1188 * virtual addresses, take care about potential aliasing
1189 * before reading the page on the kernel side.
1190 */
1191 if (mapping_writably_mapped(mapping))
1192 flush_dcache_page(page);
1193
1194 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001195 * When a sequential read accesses a page several times,
1196 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 */
Jan Karaec0f1632007-05-06 14:49:25 -07001198 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 mark_page_accessed(page);
1200 prev_index = index;
1201
1202 /*
1203 * Ok, we have the page, and it's up-to-date, so
1204 * now we can copy it to user space...
1205 *
1206 * The actor routine returns how many bytes were actually used..
1207 * NOTE! This may not be the same as how much of a user buffer
1208 * we filled up (we may be padding etc), so we can only update
1209 * "pos" here (the actor routine has to update the user buffer
1210 * pointers and the remaining count).
1211 */
1212 ret = actor(desc, page, offset, nr);
1213 offset += ret;
1214 index += offset >> PAGE_CACHE_SHIFT;
1215 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001216 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217
1218 page_cache_release(page);
1219 if (ret == nr && desc->count)
1220 continue;
1221 goto out;
1222
1223page_not_up_to_date:
1224 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001225 error = lock_page_killable(page);
1226 if (unlikely(error))
1227 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001229page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001230 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 if (!page->mapping) {
1232 unlock_page(page);
1233 page_cache_release(page);
1234 continue;
1235 }
1236
1237 /* Did somebody else fill it already? */
1238 if (PageUptodate(page)) {
1239 unlock_page(page);
1240 goto page_ok;
1241 }
1242
1243readpage:
Jeff Moyer91803b42010-05-26 11:49:40 -04001244 /*
1245 * A previous I/O error may have been due to temporary
1246 * failures, eg. multipath errors.
1247 * PG_error will be set again if readpage fails.
1248 */
1249 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 /* Start the actual read. The read will unlock the page. */
1251 error = mapping->a_ops->readpage(filp, page);
1252
Zach Brown994fc28c2005-12-15 14:28:17 -08001253 if (unlikely(error)) {
1254 if (error == AOP_TRUNCATED_PAGE) {
1255 page_cache_release(page);
1256 goto find_page;
1257 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001259 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260
1261 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001262 error = lock_page_killable(page);
1263 if (unlikely(error))
1264 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 if (!PageUptodate(page)) {
1266 if (page->mapping == NULL) {
1267 /*
Christoph Hellwig2ecdc822010-01-26 17:27:20 +01001268 * invalidate_mapping_pages got it
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 */
1270 unlock_page(page);
1271 page_cache_release(page);
1272 goto find_page;
1273 }
1274 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001275 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001276 error = -EIO;
1277 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 }
1279 unlock_page(page);
1280 }
1281
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 goto page_ok;
1283
1284readpage_error:
1285 /* UHHUH! A synchronous read error occurred. Report it */
1286 desc->error = error;
1287 page_cache_release(page);
1288 goto out;
1289
1290no_cached_page:
1291 /*
1292 * Ok, it wasn't cached, so we need to create a new
1293 * page..
1294 */
Nick Piggineb2be182007-10-16 01:24:57 -07001295 page = page_cache_alloc_cold(mapping);
1296 if (!page) {
1297 desc->error = -ENOMEM;
1298 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 }
Nick Piggineb2be182007-10-16 01:24:57 -07001300 error = add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 index, GFP_KERNEL);
1302 if (error) {
Nick Piggineb2be182007-10-16 01:24:57 -07001303 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 if (error == -EEXIST)
1305 goto find_page;
1306 desc->error = error;
1307 goto out;
1308 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 goto readpage;
1310 }
1311
1312out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001313 ra->prev_pos = prev_index;
1314 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1315 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001317 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07001318 file_accessed(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320
1321int file_read_actor(read_descriptor_t *desc, struct page *page,
1322 unsigned long offset, unsigned long size)
1323{
1324 char *kaddr;
1325 unsigned long left, count = desc->count;
1326
1327 if (size > count)
1328 size = count;
1329
1330 /*
1331 * Faults on the destination of a read are common, so do it before
1332 * taking the kmap.
1333 */
1334 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1335 kaddr = kmap_atomic(page, KM_USER0);
1336 left = __copy_to_user_inatomic(desc->arg.buf,
1337 kaddr + offset, size);
1338 kunmap_atomic(kaddr, KM_USER0);
1339 if (left == 0)
1340 goto success;
1341 }
1342
1343 /* Do it the slow way */
1344 kaddr = kmap(page);
1345 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1346 kunmap(page);
1347
1348 if (left) {
1349 size -= left;
1350 desc->error = -EFAULT;
1351 }
1352success:
1353 desc->count = count - size;
1354 desc->written += size;
1355 desc->arg.buf += size;
1356 return size;
1357}
1358
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001359/*
1360 * Performs necessary checks before doing a write
1361 * @iov: io vector request
1362 * @nr_segs: number of segments in the iovec
1363 * @count: number of bytes to write
1364 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1365 *
1366 * Adjust number of segments and amount of bytes to write (nr_segs should be
1367 * properly initialized first). Returns appropriate error code that caller
1368 * should return or zero in case that write should be allowed.
1369 */
1370int generic_segment_checks(const struct iovec *iov,
1371 unsigned long *nr_segs, size_t *count, int access_flags)
1372{
1373 unsigned long seg;
1374 size_t cnt = 0;
1375 for (seg = 0; seg < *nr_segs; seg++) {
1376 const struct iovec *iv = &iov[seg];
1377
1378 /*
1379 * If any segment has a negative length, or the cumulative
1380 * length ever wraps negative then return -EINVAL.
1381 */
1382 cnt += iv->iov_len;
1383 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1384 return -EINVAL;
1385 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1386 continue;
1387 if (seg == 0)
1388 return -EFAULT;
1389 *nr_segs = seg;
1390 cnt -= iv->iov_len; /* This segment is no good */
1391 break;
1392 }
1393 *count = cnt;
1394 return 0;
1395}
1396EXPORT_SYMBOL(generic_segment_checks);
1397
Randy Dunlap485bb992006-06-23 02:03:49 -07001398/**
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001399 * generic_file_aio_read - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001400 * @iocb: kernel I/O control block
1401 * @iov: io vector request
1402 * @nr_segs: number of segments in the iovec
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001403 * @pos: current file position
Randy Dunlap485bb992006-06-23 02:03:49 -07001404 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 * This is the "read()" routine for all filesystems
1406 * that can use the page cache directly.
1407 */
1408ssize_t
Badari Pulavarty543ade12006-09-30 23:28:48 -07001409generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1410 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411{
1412 struct file *filp = iocb->ki_filp;
1413 ssize_t retval;
Josef Bacik66f998f2010-05-23 11:00:54 -04001414 unsigned long seg = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 size_t count;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001416 loff_t *ppos = &iocb->ki_pos;
Jens Axboe55602dd2010-06-24 15:05:37 +02001417 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418
1419 count = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001420 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1421 if (retval)
1422 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423
Jens Axboe55602dd2010-06-24 15:05:37 +02001424 blk_start_plug(&plug);
1425
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1427 if (filp->f_flags & O_DIRECT) {
Badari Pulavarty543ade12006-09-30 23:28:48 -07001428 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 struct address_space *mapping;
1430 struct inode *inode;
1431
1432 mapping = filp->f_mapping;
1433 inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 if (!count)
1435 goto out; /* skip atime */
1436 size = i_size_read(inode);
1437 if (pos < size) {
Nick Piggin48b47c52009-01-06 14:40:22 -08001438 retval = filemap_write_and_wait_range(mapping, pos,
1439 pos + iov_length(iov, nr_segs) - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07001440 if (!retval) {
1441 retval = mapping->a_ops->direct_IO(READ, iocb,
1442 iov, pos, nr_segs);
1443 }
Josef Bacik66f998f2010-05-23 11:00:54 -04001444 if (retval > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 *ppos = pos + retval;
Josef Bacik66f998f2010-05-23 11:00:54 -04001446 count -= retval;
1447 }
1448
1449 /*
1450 * Btrfs can have a short DIO read if we encounter
1451 * compressed extents, so if there was an error, or if
1452 * we've already read everything we wanted to, or if
1453 * there was a short read because we hit EOF, go ahead
1454 * and return. Otherwise fallthrough to buffered io for
1455 * the rest of the read.
1456 */
1457 if (retval < 0 || !count || *ppos >= size) {
Hugh Dickins11fa9772008-07-23 21:27:34 -07001458 file_accessed(filp);
1459 goto out;
1460 }
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001461 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 }
1463
Josef Bacik66f998f2010-05-23 11:00:54 -04001464 count = retval;
Hugh Dickins11fa9772008-07-23 21:27:34 -07001465 for (seg = 0; seg < nr_segs; seg++) {
1466 read_descriptor_t desc;
Josef Bacik66f998f2010-05-23 11:00:54 -04001467 loff_t offset = 0;
1468
1469 /*
1470 * If we did a short DIO read we need to skip the section of the
1471 * iov that we've already read data into.
1472 */
1473 if (count) {
1474 if (count > iov[seg].iov_len) {
1475 count -= iov[seg].iov_len;
1476 continue;
1477 }
1478 offset = count;
1479 count = 0;
1480 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481
Hugh Dickins11fa9772008-07-23 21:27:34 -07001482 desc.written = 0;
Josef Bacik66f998f2010-05-23 11:00:54 -04001483 desc.arg.buf = iov[seg].iov_base + offset;
1484 desc.count = iov[seg].iov_len - offset;
Hugh Dickins11fa9772008-07-23 21:27:34 -07001485 if (desc.count == 0)
1486 continue;
1487 desc.error = 0;
1488 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1489 retval += desc.written;
1490 if (desc.error) {
1491 retval = retval ?: desc.error;
1492 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 }
Hugh Dickins11fa9772008-07-23 21:27:34 -07001494 if (desc.count > 0)
1495 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 }
1497out:
Jens Axboe55602dd2010-06-24 15:05:37 +02001498 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 return retval;
1500}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501EXPORT_SYMBOL(generic_file_aio_read);
1502
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503static ssize_t
1504do_readahead(struct address_space *mapping, struct file *filp,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001505 pgoff_t index, unsigned long nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506{
1507 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1508 return -EINVAL;
1509
Wu Fengguangf7e839d2009-06-16 15:31:20 -07001510 force_page_cache_readahead(mapping, filp, index, nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 return 0;
1512}
1513
Heiko Carstens6673e0c2009-01-14 14:14:02 +01001514SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515{
1516 ssize_t ret;
1517 struct file *file;
1518
1519 ret = -EBADF;
1520 file = fget(fd);
1521 if (file) {
1522 if (file->f_mode & FMODE_READ) {
1523 struct address_space *mapping = file->f_mapping;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001524 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1525 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 unsigned long len = end - start + 1;
1527 ret = do_readahead(mapping, file, start, len);
1528 }
1529 fput(file);
1530 }
1531 return ret;
1532}
Heiko Carstens6673e0c2009-01-14 14:14:02 +01001533#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1534asmlinkage long SyS_readahead(long fd, loff_t offset, long count)
1535{
1536 return SYSC_readahead((int) fd, offset, (size_t) count);
1537}
1538SYSCALL_ALIAS(sys_readahead, SyS_readahead);
1539#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540
1541#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001542/**
1543 * page_cache_read - adds requested page to the page cache if not already there
1544 * @file: file to read
1545 * @offset: page index
1546 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 * This adds the requested page to the page cache if it isn't already there,
1548 * and schedules an I/O to read in its contents from disk.
1549 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001550static int page_cache_read(struct file *file, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551{
1552 struct address_space *mapping = file->f_mapping;
1553 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001554 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555
Zach Brown994fc28c2005-12-15 14:28:17 -08001556 do {
1557 page = page_cache_alloc_cold(mapping);
1558 if (!page)
1559 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560
Zach Brown994fc28c2005-12-15 14:28:17 -08001561 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1562 if (ret == 0)
1563 ret = mapping->a_ops->readpage(file, page);
1564 else if (ret == -EEXIST)
1565 ret = 0; /* losing race to add is OK */
1566
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568
Zach Brown994fc28c2005-12-15 14:28:17 -08001569 } while (ret == AOP_TRUNCATED_PAGE);
1570
1571 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572}
1573
1574#define MMAP_LOTSAMISS (100)
1575
Linus Torvaldsef00e082009-06-16 15:31:25 -07001576/*
1577 * Synchronous readahead happens when we don't even find
1578 * a page in the page cache at all.
1579 */
1580static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1581 struct file_ra_state *ra,
1582 struct file *file,
1583 pgoff_t offset)
1584{
1585 unsigned long ra_pages;
1586 struct address_space *mapping = file->f_mapping;
1587
1588 /* If we don't want any read-ahead, don't bother */
1589 if (VM_RandomReadHint(vma))
1590 return;
Wu Fengguang275b12b2011-05-24 17:12:28 -07001591 if (!ra->ra_pages)
1592 return;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001593
Wu Fengguang2cbea1d2011-05-24 17:12:30 -07001594 if (VM_SequentialReadHint(vma)) {
Wu Fengguang7ffc59b2009-06-16 15:31:38 -07001595 page_cache_sync_readahead(mapping, ra, file, offset,
1596 ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001597 return;
1598 }
1599
Andi Kleen207d04b2011-05-24 17:12:29 -07001600 /* Avoid banging the cache line if not needed */
1601 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001602 ra->mmap_miss++;
1603
1604 /*
1605 * Do we miss much more than hit in this file? If so,
1606 * stop bothering with read-ahead. It will only hurt.
1607 */
1608 if (ra->mmap_miss > MMAP_LOTSAMISS)
1609 return;
1610
Wu Fengguangd30a1102009-06-16 15:31:30 -07001611 /*
1612 * mmap read-around
1613 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001614 ra_pages = max_sane_readahead(ra->ra_pages);
Wu Fengguang275b12b2011-05-24 17:12:28 -07001615 ra->start = max_t(long, 0, offset - ra_pages / 2);
1616 ra->size = ra_pages;
Wu Fengguang2cbea1d2011-05-24 17:12:30 -07001617 ra->async_size = ra_pages / 4;
Wu Fengguang275b12b2011-05-24 17:12:28 -07001618 ra_submit(ra, mapping, file);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001619}
1620
1621/*
1622 * Asynchronous readahead happens when we find the page and PG_readahead,
1623 * so we want to possibly extend the readahead further..
1624 */
1625static void do_async_mmap_readahead(struct vm_area_struct *vma,
1626 struct file_ra_state *ra,
1627 struct file *file,
1628 struct page *page,
1629 pgoff_t offset)
1630{
1631 struct address_space *mapping = file->f_mapping;
1632
1633 /* If we don't want any read-ahead, don't bother */
1634 if (VM_RandomReadHint(vma))
1635 return;
1636 if (ra->mmap_miss > 0)
1637 ra->mmap_miss--;
1638 if (PageReadahead(page))
Wu Fengguang2fad6f52009-06-16 15:31:29 -07001639 page_cache_async_readahead(mapping, ra, file,
1640 page, offset, ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001641}
1642
Randy Dunlap485bb992006-06-23 02:03:49 -07001643/**
Nick Piggin54cb8822007-07-19 01:46:59 -07001644 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07001645 * @vma: vma in which the fault was taken
1646 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07001647 *
Nick Piggin54cb8822007-07-19 01:46:59 -07001648 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 * mapped memory region to read in file data during a page fault.
1650 *
1651 * The goto's are kind of ugly, but this streamlines the normal case of having
1652 * it in the page cache, and handles the special cases reasonably without
1653 * having a lot of duplicated code.
1654 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001655int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656{
1657 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07001658 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 struct address_space *mapping = file->f_mapping;
1660 struct file_ra_state *ra = &file->f_ra;
1661 struct inode *inode = mapping->host;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001662 pgoff_t offset = vmf->pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 struct page *page;
Jan Kara2004dc82008-02-08 04:20:11 -08001664 pgoff_t size;
Nick Piggin83c54072007-07-19 01:47:05 -07001665 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001668 if (offset >= size)
Linus Torvalds5307cc12007-10-31 09:19:46 -07001669 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 * Do we have something in the page cache already?
1673 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001674 page = find_get_page(mapping, offset);
1675 if (likely(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07001677 * We found the page, so try async readahead before
1678 * waiting for the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001680 do_async_mmap_readahead(vma, ra, file, page, offset);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001681 } else {
1682 /* No page in the page cache at all */
1683 do_sync_mmap_readahead(vma, ra, file, offset);
1684 count_vm_event(PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07001685 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001686 ret = VM_FAULT_MAJOR;
1687retry_find:
Michel Lespinasseb522c942010-10-26 14:21:56 -07001688 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 if (!page)
1690 goto no_cached_page;
1691 }
1692
Michel Lespinassed88c0922010-11-02 13:05:18 -07001693 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1694 page_cache_release(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -07001695 return ret | VM_FAULT_RETRY;
Michel Lespinassed88c0922010-11-02 13:05:18 -07001696 }
Michel Lespinasseb522c942010-10-26 14:21:56 -07001697
1698 /* Did it get truncated? */
1699 if (unlikely(page->mapping != mapping)) {
1700 unlock_page(page);
1701 put_page(page);
1702 goto retry_find;
1703 }
1704 VM_BUG_ON(page->index != offset);
1705
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 /*
Nick Piggind00806b2007-07-19 01:46:57 -07001707 * We have a locked page in the page cache, now we need to check
1708 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 */
Nick Piggind00806b2007-07-19 01:46:57 -07001710 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 goto page_not_uptodate;
1712
Linus Torvaldsef00e082009-06-16 15:31:25 -07001713 /*
1714 * Found the page and have a reference on it.
1715 * We must recheck i_size under page lock.
1716 */
Nick Piggind00806b2007-07-19 01:46:57 -07001717 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001718 if (unlikely(offset >= size)) {
Nick Piggind00806b2007-07-19 01:46:57 -07001719 unlock_page(page);
Yan Zheng745ad482007-10-08 10:08:37 -07001720 page_cache_release(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07001721 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001722 }
1723
Nick Piggind0217ac2007-07-19 01:47:03 -07001724 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07001725 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727no_cached_page:
1728 /*
1729 * We're only likely to ever get here if MADV_RANDOM is in
1730 * effect.
1731 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001732 error = page_cache_read(file, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733
1734 /*
1735 * The page we want has now been added to the page cache.
1736 * In the unlikely event that someone removed it in the
1737 * meantime, we'll just come back here and read it again.
1738 */
1739 if (error >= 0)
1740 goto retry_find;
1741
1742 /*
1743 * An error return from page_cache_read can result if the
1744 * system is low on memory, or a problem occurs while trying
1745 * to schedule I/O.
1746 */
1747 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07001748 return VM_FAULT_OOM;
1749 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750
1751page_not_uptodate:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 /*
1753 * Umm, take care of errors if the page isn't up-to-date.
1754 * Try to re-read it _once_. We do this synchronously,
1755 * because there really aren't any performance issues here
1756 * and we need to check for errors.
1757 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001759 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07001760 if (!error) {
1761 wait_on_page_locked(page);
1762 if (!PageUptodate(page))
1763 error = -EIO;
1764 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07001766
1767 if (!error || error == AOP_TRUNCATED_PAGE)
1768 goto retry_find;
1769
1770 /* Things didn't work out. Return zero to tell the mm layer so. */
1771 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07001772 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07001773}
1774EXPORT_SYMBOL(filemap_fault);
1775
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001776const struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07001777 .fault = filemap_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778};
1779
1780/* This is used for a general mmap of a disk file */
1781
1782int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1783{
1784 struct address_space *mapping = file->f_mapping;
1785
1786 if (!mapping->a_ops->readpage)
1787 return -ENOEXEC;
1788 file_accessed(file);
1789 vma->vm_ops = &generic_file_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001790 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 return 0;
1792}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793
1794/*
1795 * This is for filesystems which do not implement ->writepage.
1796 */
1797int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1798{
1799 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1800 return -EINVAL;
1801 return generic_file_mmap(file, vma);
1802}
1803#else
1804int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1805{
1806 return -ENOSYS;
1807}
1808int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1809{
1810 return -ENOSYS;
1811}
1812#endif /* CONFIG_MMU */
1813
1814EXPORT_SYMBOL(generic_file_mmap);
1815EXPORT_SYMBOL(generic_file_readonly_mmap);
1816
Nick Piggin6fe69002007-05-06 14:49:04 -07001817static struct page *__read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001818 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001819 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001820 void *data,
1821 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822{
Nick Piggineb2be182007-10-16 01:24:57 -07001823 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 int err;
1825repeat:
1826 page = find_get_page(mapping, index);
1827 if (!page) {
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001828 page = __page_cache_alloc(gfp | __GFP_COLD);
Nick Piggineb2be182007-10-16 01:24:57 -07001829 if (!page)
1830 return ERR_PTR(-ENOMEM);
1831 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
1832 if (unlikely(err)) {
1833 page_cache_release(page);
1834 if (err == -EEXIST)
1835 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 return ERR_PTR(err);
1838 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 err = filler(data, page);
1840 if (err < 0) {
1841 page_cache_release(page);
1842 page = ERR_PTR(err);
1843 }
1844 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 return page;
1846}
1847
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001848static struct page *do_read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001849 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001850 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001851 void *data,
1852 gfp_t gfp)
1853
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854{
1855 struct page *page;
1856 int err;
1857
1858retry:
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001859 page = __read_cache_page(mapping, index, filler, data, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01001861 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862 if (PageUptodate(page))
1863 goto out;
1864
1865 lock_page(page);
1866 if (!page->mapping) {
1867 unlock_page(page);
1868 page_cache_release(page);
1869 goto retry;
1870 }
1871 if (PageUptodate(page)) {
1872 unlock_page(page);
1873 goto out;
1874 }
1875 err = filler(data, page);
1876 if (err < 0) {
1877 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01001878 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 }
David Howellsc855ff32007-05-09 13:42:20 +01001880out:
Nick Piggin6fe69002007-05-06 14:49:04 -07001881 mark_page_accessed(page);
1882 return page;
1883}
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001884
1885/**
1886 * read_cache_page_async - read into page cache, fill it if needed
1887 * @mapping: the page's address_space
1888 * @index: the page index
1889 * @filler: function to perform the read
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001890 * @data: first arg to filler(data, page) function, often left as NULL
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001891 *
1892 * Same as read_cache_page, but don't wait for page to become unlocked
1893 * after submitting it to the filler.
1894 *
1895 * Read into the page cache. If a page already exists, and PageUptodate() is
1896 * not set, try to fill the page but don't wait for it to become unlocked.
1897 *
1898 * If the page does not get brought uptodate, return -EIO.
1899 */
1900struct page *read_cache_page_async(struct address_space *mapping,
1901 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001902 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001903 void *data)
1904{
1905 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
1906}
Nick Piggin6fe69002007-05-06 14:49:04 -07001907EXPORT_SYMBOL(read_cache_page_async);
1908
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001909static struct page *wait_on_page_read(struct page *page)
1910{
1911 if (!IS_ERR(page)) {
1912 wait_on_page_locked(page);
1913 if (!PageUptodate(page)) {
1914 page_cache_release(page);
1915 page = ERR_PTR(-EIO);
1916 }
1917 }
1918 return page;
1919}
1920
1921/**
1922 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
1923 * @mapping: the page's address_space
1924 * @index: the page index
1925 * @gfp: the page allocator flags to use if allocating
1926 *
1927 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
1928 * any new page allocations done using the specified allocation flags. Note
1929 * that the Radix tree operations will still use GFP_KERNEL, so you can't
1930 * expect to do this atomically or anything like that - but you can pass in
1931 * other page requirements.
1932 *
1933 * If the page does not get brought uptodate, return -EIO.
1934 */
1935struct page *read_cache_page_gfp(struct address_space *mapping,
1936 pgoff_t index,
1937 gfp_t gfp)
1938{
1939 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
1940
1941 return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp));
1942}
1943EXPORT_SYMBOL(read_cache_page_gfp);
1944
Nick Piggin6fe69002007-05-06 14:49:04 -07001945/**
1946 * read_cache_page - read into page cache, fill it if needed
1947 * @mapping: the page's address_space
1948 * @index: the page index
1949 * @filler: function to perform the read
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001950 * @data: first arg to filler(data, page) function, often left as NULL
Nick Piggin6fe69002007-05-06 14:49:04 -07001951 *
1952 * Read into the page cache. If a page already exists, and PageUptodate() is
1953 * not set, try to fill the page then wait for it to become unlocked.
1954 *
1955 * If the page does not get brought uptodate, return -EIO.
1956 */
1957struct page *read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001958 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001959 int (*filler)(void *, struct page *),
Nick Piggin6fe69002007-05-06 14:49:04 -07001960 void *data)
1961{
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001962 return wait_on_page_read(read_cache_page_async(mapping, index, filler, data));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964EXPORT_SYMBOL(read_cache_page);
1965
1966/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 * The logic we want is
1968 *
1969 * if suid or (sgid and xgrp)
1970 * remove privs
1971 */
Jens Axboe01de85e2006-10-17 19:50:36 +02001972int should_remove_suid(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973{
1974 mode_t mode = dentry->d_inode->i_mode;
1975 int kill = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976
1977 /* suid always must be killed */
1978 if (unlikely(mode & S_ISUID))
1979 kill = ATTR_KILL_SUID;
1980
1981 /*
1982 * sgid without any exec bits is just a mandatory locking mark; leave
1983 * it alone. If some exec bits are set, it's a real sgid; kill it.
1984 */
1985 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1986 kill |= ATTR_KILL_SGID;
1987
Dmitri Monakhov7f5ff762008-12-01 14:34:56 -08001988 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
Jens Axboe01de85e2006-10-17 19:50:36 +02001989 return kill;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990
Jens Axboe01de85e2006-10-17 19:50:36 +02001991 return 0;
1992}
Mark Fashehd23a1472006-10-17 17:05:18 -07001993EXPORT_SYMBOL(should_remove_suid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001994
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +02001995static int __remove_suid(struct dentry *dentry, int kill)
Jens Axboe01de85e2006-10-17 19:50:36 +02001996{
1997 struct iattr newattrs;
1998
1999 newattrs.ia_valid = ATTR_FORCE | kill;
2000 return notify_change(dentry, &newattrs);
2001}
2002
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02002003int file_remove_suid(struct file *file)
Jens Axboe01de85e2006-10-17 19:50:36 +02002004{
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02002005 struct dentry *dentry = file->f_path.dentry;
Andi Kleen69b45732011-05-28 08:25:51 -07002006 struct inode *inode = dentry->d_inode;
2007 int killsuid;
2008 int killpriv;
Serge E. Hallynb5376772007-10-16 23:31:36 -07002009 int error = 0;
Jens Axboe01de85e2006-10-17 19:50:36 +02002010
Andi Kleen69b45732011-05-28 08:25:51 -07002011 /* Fast path for nothing security related */
2012 if (IS_NOSEC(inode))
2013 return 0;
2014
2015 killsuid = should_remove_suid(dentry);
2016 killpriv = security_inode_need_killpriv(dentry);
2017
Serge E. Hallynb5376772007-10-16 23:31:36 -07002018 if (killpriv < 0)
2019 return killpriv;
2020 if (killpriv)
2021 error = security_inode_killpriv(dentry);
2022 if (!error && killsuid)
2023 error = __remove_suid(dentry, killsuid);
Al Viro9e1f1de2011-06-03 18:24:58 -04002024 if (!error && (inode->i_sb->s_flags & MS_NOSEC))
Andi Kleen69b45732011-05-28 08:25:51 -07002025 inode->i_flags |= S_NOSEC;
Jens Axboe01de85e2006-10-17 19:50:36 +02002026
Serge E. Hallynb5376772007-10-16 23:31:36 -07002027 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028}
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02002029EXPORT_SYMBOL(file_remove_suid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030
Nick Piggin2f718ff2007-10-16 01:24:59 -07002031static size_t __iovec_copy_from_user_inatomic(char *vaddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 const struct iovec *iov, size_t base, size_t bytes)
2033{
Ingo Molnarf1800532009-03-02 11:00:57 +01002034 size_t copied = 0, left = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035
2036 while (bytes) {
2037 char __user *buf = iov->iov_base + base;
2038 int copy = min(bytes, iov->iov_len - base);
2039
2040 base = 0;
Ingo Molnarf1800532009-03-02 11:00:57 +01002041 left = __copy_from_user_inatomic(vaddr, buf, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 copied += copy;
2043 bytes -= copy;
2044 vaddr += copy;
2045 iov++;
2046
NeilBrown01408c42006-06-25 05:47:58 -07002047 if (unlikely(left))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 }
2050 return copied - left;
2051}
2052
2053/*
Nick Piggin2f718ff2007-10-16 01:24:59 -07002054 * Copy as much as we can into the page and return the number of bytes which
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02002055 * were successfully copied. If a fault is encountered then return the number of
Nick Piggin2f718ff2007-10-16 01:24:59 -07002056 * bytes which were copied.
2057 */
2058size_t iov_iter_copy_from_user_atomic(struct page *page,
2059 struct iov_iter *i, unsigned long offset, size_t bytes)
2060{
2061 char *kaddr;
2062 size_t copied;
2063
2064 BUG_ON(!in_atomic());
2065 kaddr = kmap_atomic(page, KM_USER0);
2066 if (likely(i->nr_segs == 1)) {
2067 int left;
2068 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01002069 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002070 copied = bytes - left;
2071 } else {
2072 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
2073 i->iov, i->iov_offset, bytes);
2074 }
2075 kunmap_atomic(kaddr, KM_USER0);
2076
2077 return copied;
2078}
Nick Piggin89e10782007-10-16 01:25:07 -07002079EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002080
2081/*
2082 * This has the same sideeffects and return value as
2083 * iov_iter_copy_from_user_atomic().
2084 * The difference is that it attempts to resolve faults.
2085 * Page must not be locked.
2086 */
2087size_t iov_iter_copy_from_user(struct page *page,
2088 struct iov_iter *i, unsigned long offset, size_t bytes)
2089{
2090 char *kaddr;
2091 size_t copied;
2092
2093 kaddr = kmap(page);
2094 if (likely(i->nr_segs == 1)) {
2095 int left;
2096 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01002097 left = __copy_from_user(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002098 copied = bytes - left;
2099 } else {
2100 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
2101 i->iov, i->iov_offset, bytes);
2102 }
2103 kunmap(page);
2104 return copied;
2105}
Nick Piggin89e10782007-10-16 01:25:07 -07002106EXPORT_SYMBOL(iov_iter_copy_from_user);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002107
Nick Pigginf7009262008-03-10 11:43:59 -07002108void iov_iter_advance(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07002109{
Nick Pigginf7009262008-03-10 11:43:59 -07002110 BUG_ON(i->count < bytes);
2111
Nick Piggin2f718ff2007-10-16 01:24:59 -07002112 if (likely(i->nr_segs == 1)) {
2113 i->iov_offset += bytes;
Nick Pigginf7009262008-03-10 11:43:59 -07002114 i->count -= bytes;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002115 } else {
2116 const struct iovec *iov = i->iov;
2117 size_t base = i->iov_offset;
Jeff Layton39be79c2011-10-27 23:53:08 +02002118 unsigned long nr_segs = i->nr_segs;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002119
Nick Piggin124d3b72008-02-02 15:01:17 +01002120 /*
2121 * The !iov->iov_len check ensures we skip over unlikely
Nick Pigginf7009262008-03-10 11:43:59 -07002122 * zero-length segments (without overruning the iovec).
Nick Piggin124d3b72008-02-02 15:01:17 +01002123 */
Linus Torvalds94ad3742008-07-30 14:45:12 -07002124 while (bytes || unlikely(i->count && !iov->iov_len)) {
Nick Pigginf7009262008-03-10 11:43:59 -07002125 int copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002126
Nick Pigginf7009262008-03-10 11:43:59 -07002127 copy = min(bytes, iov->iov_len - base);
2128 BUG_ON(!i->count || i->count < copy);
2129 i->count -= copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002130 bytes -= copy;
2131 base += copy;
2132 if (iov->iov_len == base) {
2133 iov++;
Jeff Layton39be79c2011-10-27 23:53:08 +02002134 nr_segs--;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002135 base = 0;
2136 }
2137 }
2138 i->iov = iov;
2139 i->iov_offset = base;
Jeff Layton39be79c2011-10-27 23:53:08 +02002140 i->nr_segs = nr_segs;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002141 }
2142}
Nick Piggin89e10782007-10-16 01:25:07 -07002143EXPORT_SYMBOL(iov_iter_advance);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002144
Nick Pigginafddba42007-10-16 01:25:01 -07002145/*
2146 * Fault in the first iovec of the given iov_iter, to a maximum length
2147 * of bytes. Returns 0 on success, or non-zero if the memory could not be
2148 * accessed (ie. because it is an invalid address).
2149 *
2150 * writev-intensive code may want this to prefault several iovecs -- that
2151 * would be possible (callers must not rely on the fact that _only_ the
2152 * first iovec will be faulted with the current implementation).
2153 */
2154int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07002155{
Nick Piggin2f718ff2007-10-16 01:24:59 -07002156 char __user *buf = i->iov->iov_base + i->iov_offset;
Nick Pigginafddba42007-10-16 01:25:01 -07002157 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
2158 return fault_in_pages_readable(buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002159}
Nick Piggin89e10782007-10-16 01:25:07 -07002160EXPORT_SYMBOL(iov_iter_fault_in_readable);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002161
2162/*
2163 * Return the count of just the current iov_iter segment.
2164 */
2165size_t iov_iter_single_seg_count(struct iov_iter *i)
2166{
2167 const struct iovec *iov = i->iov;
2168 if (i->nr_segs == 1)
2169 return i->count;
2170 else
2171 return min(i->count, iov->iov_len - i->iov_offset);
2172}
Nick Piggin89e10782007-10-16 01:25:07 -07002173EXPORT_SYMBOL(iov_iter_single_seg_count);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002174
2175/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 * Performs necessary checks before doing a write
2177 *
Randy Dunlap485bb992006-06-23 02:03:49 -07002178 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 * Returns appropriate error code that caller should return or
2180 * zero in case that write should be allowed.
2181 */
2182inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2183{
2184 struct inode *inode = file->f_mapping->host;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002185 unsigned long limit = rlimit(RLIMIT_FSIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186
2187 if (unlikely(*pos < 0))
2188 return -EINVAL;
2189
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190 if (!isblk) {
2191 /* FIXME: this is for backwards compatibility with 2.4 */
2192 if (file->f_flags & O_APPEND)
2193 *pos = i_size_read(inode);
2194
2195 if (limit != RLIM_INFINITY) {
2196 if (*pos >= limit) {
2197 send_sig(SIGXFSZ, current, 0);
2198 return -EFBIG;
2199 }
2200 if (*count > limit - (typeof(limit))*pos) {
2201 *count = limit - (typeof(limit))*pos;
2202 }
2203 }
2204 }
2205
2206 /*
2207 * LFS rule
2208 */
2209 if (unlikely(*pos + *count > MAX_NON_LFS &&
2210 !(file->f_flags & O_LARGEFILE))) {
2211 if (*pos >= MAX_NON_LFS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 return -EFBIG;
2213 }
2214 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2215 *count = MAX_NON_LFS - (unsigned long)*pos;
2216 }
2217 }
2218
2219 /*
2220 * Are we about to exceed the fs block limit ?
2221 *
2222 * If we have written data it becomes a short write. If we have
2223 * exceeded without writing data we send a signal and return EFBIG.
2224 * Linus frestrict idea will clean these up nicely..
2225 */
2226 if (likely(!isblk)) {
2227 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2228 if (*count || *pos > inode->i_sb->s_maxbytes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229 return -EFBIG;
2230 }
2231 /* zero-length writes at ->s_maxbytes are OK */
2232 }
2233
2234 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2235 *count = inode->i_sb->s_maxbytes - *pos;
2236 } else {
David Howells93614012006-09-30 20:45:40 +02002237#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 loff_t isize;
2239 if (bdev_read_only(I_BDEV(inode)))
2240 return -EPERM;
2241 isize = i_size_read(inode);
2242 if (*pos >= isize) {
2243 if (*count || *pos > isize)
2244 return -ENOSPC;
2245 }
2246
2247 if (*pos + *count > isize)
2248 *count = isize - *pos;
David Howells93614012006-09-30 20:45:40 +02002249#else
2250 return -EPERM;
2251#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 }
2253 return 0;
2254}
2255EXPORT_SYMBOL(generic_write_checks);
2256
Nick Pigginafddba42007-10-16 01:25:01 -07002257int pagecache_write_begin(struct file *file, struct address_space *mapping,
2258 loff_t pos, unsigned len, unsigned flags,
2259 struct page **pagep, void **fsdata)
2260{
2261 const struct address_space_operations *aops = mapping->a_ops;
2262
Nick Piggin4e02ed42008-10-29 14:00:55 -07002263 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002264 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002265}
2266EXPORT_SYMBOL(pagecache_write_begin);
2267
2268int pagecache_write_end(struct file *file, struct address_space *mapping,
2269 loff_t pos, unsigned len, unsigned copied,
2270 struct page *page, void *fsdata)
2271{
2272 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002273
Nick Piggin4e02ed42008-10-29 14:00:55 -07002274 mark_page_accessed(page);
2275 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002276}
2277EXPORT_SYMBOL(pagecache_write_end);
2278
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279ssize_t
2280generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2281 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2282 size_t count, size_t ocount)
2283{
2284 struct file *file = iocb->ki_filp;
2285 struct address_space *mapping = file->f_mapping;
2286 struct inode *inode = mapping->host;
2287 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002288 size_t write_len;
2289 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290
2291 if (count != ocount)
2292 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2293
Christoph Hellwiga969e902008-07-23 21:27:04 -07002294 write_len = iov_length(iov, *nr_segs);
2295 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002296
Nick Piggin48b47c52009-01-06 14:40:22 -08002297 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002298 if (written)
2299 goto out;
2300
2301 /*
2302 * After a write we want buffered reads to be sure to go to disk to get
2303 * the new data. We invalidate clean cached page from the region we're
2304 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002305 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002306 */
2307 if (mapping->nrpages) {
2308 written = invalidate_inode_pages2_range(mapping,
2309 pos >> PAGE_CACHE_SHIFT, end);
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002310 /*
2311 * If a page can not be invalidated, return 0 to fall back
2312 * to buffered write.
2313 */
2314 if (written) {
2315 if (written == -EBUSY)
2316 return 0;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002317 goto out;
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002318 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002319 }
2320
2321 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2322
2323 /*
2324 * Finally, try again to invalidate clean pages which might have been
2325 * cached by non-direct readahead, or faulted in by get_user_pages()
2326 * if the source of the write was an mmap'ed region of the file
2327 * we're writing. Either one is a pretty crazy thing to do,
2328 * so we don't support it 100%. If this invalidation
2329 * fails, tough, the write still worked...
2330 */
2331 if (mapping->nrpages) {
2332 invalidate_inode_pages2_range(mapping,
2333 pos >> PAGE_CACHE_SHIFT, end);
2334 }
2335
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 if (written > 0) {
Namhyung Kim01166512010-10-26 14:21:58 -07002337 pos += written;
2338 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2339 i_size_write(inode, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340 mark_inode_dirty(inode);
2341 }
Namhyung Kim01166512010-10-26 14:21:58 -07002342 *ppos = pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002344out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 return written;
2346}
2347EXPORT_SYMBOL(generic_file_direct_write);
2348
Nick Piggineb2be182007-10-16 01:24:57 -07002349/*
2350 * Find or create a page at the given pagecache position. Return the locked
2351 * page. This function is specifically for buffered writes.
2352 */
Nick Piggin54566b22009-01-04 12:00:53 -08002353struct page *grab_cache_page_write_begin(struct address_space *mapping,
2354 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07002355{
2356 int status;
2357 struct page *page;
Nick Piggin54566b22009-01-04 12:00:53 -08002358 gfp_t gfp_notmask = 0;
2359 if (flags & AOP_FLAG_NOFS)
2360 gfp_notmask = __GFP_FS;
Nick Piggineb2be182007-10-16 01:24:57 -07002361repeat:
2362 page = find_lock_page(mapping, index);
Steven Rostedtc585a262011-01-13 15:46:18 -08002363 if (page)
Darrick J. Wong3d08bcc2011-05-27 12:23:34 -07002364 goto found;
Nick Piggineb2be182007-10-16 01:24:57 -07002365
Nick Piggin54566b22009-01-04 12:00:53 -08002366 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002367 if (!page)
2368 return NULL;
Nick Piggin54566b22009-01-04 12:00:53 -08002369 status = add_to_page_cache_lru(page, mapping, index,
2370 GFP_KERNEL & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002371 if (unlikely(status)) {
2372 page_cache_release(page);
2373 if (status == -EEXIST)
2374 goto repeat;
2375 return NULL;
2376 }
Darrick J. Wong3d08bcc2011-05-27 12:23:34 -07002377found:
2378 wait_on_page_writeback(page);
Nick Piggineb2be182007-10-16 01:24:57 -07002379 return page;
2380}
Nick Piggin54566b22009-01-04 12:00:53 -08002381EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07002382
Nick Pigginafddba42007-10-16 01:25:01 -07002383static ssize_t generic_perform_write(struct file *file,
2384 struct iov_iter *i, loff_t pos)
2385{
2386 struct address_space *mapping = file->f_mapping;
2387 const struct address_space_operations *a_ops = mapping->a_ops;
2388 long status = 0;
2389 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002390 unsigned int flags = 0;
2391
2392 /*
2393 * Copies from kernel address space cannot fail (NFSD is a big user).
2394 */
2395 if (segment_eq(get_fs(), KERNEL_DS))
2396 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002397
2398 do {
2399 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002400 unsigned long offset; /* Offset into pagecache page */
2401 unsigned long bytes; /* Bytes to write to page */
2402 size_t copied; /* Bytes copied from user */
2403 void *fsdata;
2404
2405 offset = (pos & (PAGE_CACHE_SIZE - 1));
Nick Pigginafddba42007-10-16 01:25:01 -07002406 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2407 iov_iter_count(i));
2408
2409again:
2410
2411 /*
2412 * Bring in the user page that we will copy from _first_.
2413 * Otherwise there's a nasty deadlock on copying from the
2414 * same page as we're writing to, without it being marked
2415 * up-to-date.
2416 *
2417 * Not only is this an optimisation, but it is also required
2418 * to check that the address is actually valid, when atomic
2419 * usercopies are used, below.
2420 */
2421 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2422 status = -EFAULT;
2423 break;
2424 }
2425
Nick Piggin674b8922007-10-16 01:25:03 -07002426 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002427 &page, &fsdata);
2428 if (unlikely(status))
2429 break;
2430
anfei zhou931e80e2010-02-02 13:44:02 -08002431 if (mapping_writably_mapped(mapping))
2432 flush_dcache_page(page);
2433
Nick Pigginafddba42007-10-16 01:25:01 -07002434 pagefault_disable();
2435 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2436 pagefault_enable();
2437 flush_dcache_page(page);
2438
Josef Bacikc8236db2009-07-05 12:08:18 -07002439 mark_page_accessed(page);
Nick Pigginafddba42007-10-16 01:25:01 -07002440 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2441 page, fsdata);
2442 if (unlikely(status < 0))
2443 break;
2444 copied = status;
2445
2446 cond_resched();
2447
Nick Piggin124d3b72008-02-02 15:01:17 +01002448 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002449 if (unlikely(copied == 0)) {
2450 /*
2451 * If we were unable to copy any data at all, we must
2452 * fall back to a single segment length write.
2453 *
2454 * If we didn't fallback here, we could livelock
2455 * because not all segments in the iov can be copied at
2456 * once without a pagefault.
2457 */
2458 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2459 iov_iter_single_seg_count(i));
2460 goto again;
2461 }
Nick Pigginafddba42007-10-16 01:25:01 -07002462 pos += copied;
2463 written += copied;
2464
2465 balance_dirty_pages_ratelimited(mapping);
2466
2467 } while (iov_iter_count(i));
2468
2469 return written ? written : status;
2470}
2471
2472ssize_t
2473generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2474 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2475 size_t count, ssize_t written)
2476{
2477 struct file *file = iocb->ki_filp;
Nick Pigginafddba42007-10-16 01:25:01 -07002478 ssize_t status;
2479 struct iov_iter i;
2480
2481 iov_iter_init(&i, iov, nr_segs, count, written);
Nick Piggin4e02ed42008-10-29 14:00:55 -07002482 status = generic_perform_write(file, &i, pos);
Nick Pigginafddba42007-10-16 01:25:01 -07002483
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484 if (likely(status >= 0)) {
Nick Pigginafddba42007-10-16 01:25:01 -07002485 written += status;
2486 *ppos = pos + status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487 }
2488
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489 return written ? written : status;
2490}
2491EXPORT_SYMBOL(generic_file_buffered_write);
2492
Jan Karae4dd9de2009-08-17 18:10:06 +02002493/**
2494 * __generic_file_aio_write - write data to a file
2495 * @iocb: IO state structure (file, offset, etc.)
2496 * @iov: vector with data to write
2497 * @nr_segs: number of segments in the vector
2498 * @ppos: position where to write
2499 *
2500 * This function does all the work needed for actually writing data to a
2501 * file. It does all basic checks, removes SUID from the file, updates
2502 * modification times and calls proper subroutines depending on whether we
2503 * do direct IO or a standard buffered write.
2504 *
2505 * It expects i_mutex to be grabbed unless we work on a block device or similar
2506 * object which does not need locking at all.
2507 *
2508 * This function does *not* take care of syncing data in case of O_SYNC write.
2509 * A caller has to handle it. This is mainly due to the fact that we want to
2510 * avoid syncing under i_mutex.
2511 */
2512ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2513 unsigned long nr_segs, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514{
2515 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002516 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517 size_t ocount; /* original count */
2518 size_t count; /* after file limit checks */
2519 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520 loff_t pos;
2521 ssize_t written;
2522 ssize_t err;
2523
2524 ocount = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07002525 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2526 if (err)
2527 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528
2529 count = ocount;
2530 pos = *ppos;
2531
2532 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2533
2534 /* We can write back this queue in page reclaim */
2535 current->backing_dev_info = mapping->backing_dev_info;
2536 written = 0;
2537
2538 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2539 if (err)
2540 goto out;
2541
2542 if (count == 0)
2543 goto out;
2544
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02002545 err = file_remove_suid(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546 if (err)
2547 goto out;
2548
Christoph Hellwig870f4812006-01-09 20:52:01 -08002549 file_update_time(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550
2551 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2552 if (unlikely(file->f_flags & O_DIRECT)) {
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002553 loff_t endbyte;
2554 ssize_t written_buffered;
2555
2556 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2557 ppos, count, ocount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 if (written < 0 || written == count)
2559 goto out;
2560 /*
2561 * direct-io write to a hole: fall through to buffered I/O
2562 * for completing the rest of the request.
2563 */
2564 pos += written;
2565 count -= written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002566 written_buffered = generic_file_buffered_write(iocb, iov,
2567 nr_segs, pos, ppos, count,
2568 written);
2569 /*
2570 * If generic_file_buffered_write() retuned a synchronous error
2571 * then we want to return the number of bytes which were
2572 * direct-written, or the error code if that was zero. Note
2573 * that this differs from normal direct-io semantics, which
2574 * will return -EFOO even if some bytes were written.
2575 */
2576 if (written_buffered < 0) {
2577 err = written_buffered;
2578 goto out;
2579 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002581 /*
2582 * We need to ensure that the page cache pages are written to
2583 * disk and invalidated to preserve the expected O_DIRECT
2584 * semantics.
2585 */
2586 endbyte = pos + written_buffered - written - 1;
Christoph Hellwigc05c4ed2009-09-23 15:07:30 +02002587 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002588 if (err == 0) {
2589 written = written_buffered;
2590 invalidate_mapping_pages(mapping,
2591 pos >> PAGE_CACHE_SHIFT,
2592 endbyte >> PAGE_CACHE_SHIFT);
2593 } else {
2594 /*
2595 * We don't know how much we wrote, so just return
2596 * the number of bytes which were direct-written
2597 */
2598 }
2599 } else {
2600 written = generic_file_buffered_write(iocb, iov, nr_segs,
2601 pos, ppos, count, written);
2602 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603out:
2604 current->backing_dev_info = NULL;
2605 return written ? written : err;
2606}
Jan Karae4dd9de2009-08-17 18:10:06 +02002607EXPORT_SYMBOL(__generic_file_aio_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608
Jan Karae4dd9de2009-08-17 18:10:06 +02002609/**
2610 * generic_file_aio_write - write data to a file
2611 * @iocb: IO state structure
2612 * @iov: vector with data to write
2613 * @nr_segs: number of segments in the vector
2614 * @pos: position in file where to write
2615 *
2616 * This is a wrapper around __generic_file_aio_write() to be used by most
2617 * filesystems. It takes care of syncing the file in case of O_SYNC file
2618 * and acquires i_mutex as needed.
2619 */
Badari Pulavarty027445c2006-09-30 23:28:46 -07002620ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2621 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622{
2623 struct file *file = iocb->ki_filp;
Jan Kara148f9482009-08-17 19:52:36 +02002624 struct inode *inode = file->f_mapping->host;
Jens Axboe55602dd2010-06-24 15:05:37 +02002625 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627
2628 BUG_ON(iocb->ki_pos != pos);
2629
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002630 mutex_lock(&inode->i_mutex);
Jens Axboe55602dd2010-06-24 15:05:37 +02002631 blk_start_plug(&plug);
Jan Karae4dd9de2009-08-17 18:10:06 +02002632 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002633 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634
Jan Kara148f9482009-08-17 19:52:36 +02002635 if (ret > 0 || ret == -EIOCBQUEUED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636 ssize_t err;
2637
Jan Kara148f9482009-08-17 19:52:36 +02002638 err = generic_write_sync(file, pos, ret);
Jan Karac7b50db2009-08-18 16:18:20 +02002639 if (err < 0 && ret > 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640 ret = err;
2641 }
Jens Axboe55602dd2010-06-24 15:05:37 +02002642 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643 return ret;
2644}
2645EXPORT_SYMBOL(generic_file_aio_write);
2646
David Howellscf9a2ae2006-08-29 19:05:54 +01002647/**
2648 * try_to_release_page() - release old fs-specific metadata on a page
2649 *
2650 * @page: the page which the kernel is trying to free
2651 * @gfp_mask: memory allocation flags (and I/O mode)
2652 *
2653 * The address_space is to try to release any data against the page
2654 * (presumably at page->private). If the release was successful, return `1'.
2655 * Otherwise return zero.
2656 *
David Howells266cf652009-04-03 16:42:36 +01002657 * This may also be called if PG_fscache is set on a page, indicating that the
2658 * page is known to the local caching routines.
2659 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002660 * The @gfp_mask argument specifies whether I/O may be performed to release
Mingming Cao3f31fdd2008-07-25 01:46:22 -07002661 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01002662 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002663 */
2664int try_to_release_page(struct page *page, gfp_t gfp_mask)
2665{
2666 struct address_space * const mapping = page->mapping;
2667
2668 BUG_ON(!PageLocked(page));
2669 if (PageWriteback(page))
2670 return 0;
2671
2672 if (mapping && mapping->a_ops->releasepage)
2673 return mapping->a_ops->releasepage(page, gfp_mask);
2674 return try_to_free_buffers(page);
2675}
2676
2677EXPORT_SYMBOL(try_to_release_page);