blob: 867d40222ec798ce99a8332b991e8d7c4bdfb8ec [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>
Rik van Riel4f98a2f2008-10-18 20:26:32 -070036#include <linux/mm_inline.h> /* for page_is_file_cache() */
Dan Magenheimerc515e1f2011-05-26 10:01:43 -060037#include <linux/cleancache.h>
Nick Piggin0f8053a2006-03-22 00:08:33 -080038#include "internal.h"
39
Linus Torvalds1da177e2005-04-16 15:20:36 -070040/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070041 * FIXME: remove all knowledge of the buffer layer from the core VM
42 */
Jan Kara148f9482009-08-17 19:52:36 +020043#include <linux/buffer_head.h> /* for try_to_free_buffers */
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
Linus Torvalds1da177e2005-04-16 15:20:36 -070045#include <asm/mman.h>
46
47/*
48 * Shared mappings implemented 30.11.1994. It's not fully working yet,
49 * though.
50 *
51 * Shared mappings now work. 15.8.1995 Bruno.
52 *
53 * finished 'unifying' the page and buffer cache and SMP-threaded the
54 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
55 *
56 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
57 */
58
59/*
60 * Lock ordering:
61 *
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070062 * ->i_mmap_mutex (truncate_pagecache)
Linus Torvalds1da177e2005-04-16 15:20:36 -070063 * ->private_lock (__free_pte->__set_page_dirty_buffers)
Hugh Dickins5d337b92005-09-03 15:54:41 -070064 * ->swap_lock (exclusive_swap_page, others)
65 * ->mapping->tree_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080067 * ->i_mutex
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070068 * ->i_mmap_mutex (truncate->unmap_mapping_range)
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 *
70 * ->mmap_sem
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070071 * ->i_mmap_mutex
Hugh Dickinsb8072f02005-10-29 18:16:41 -070072 * ->page_table_lock or pte_lock (various, mainly in memory.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
74 *
75 * ->mmap_sem
76 * ->lock_page (access_process_vm)
77 *
Nick Piggin82591e62006-10-19 23:29:10 -070078 * ->i_mutex (generic_file_buffered_write)
79 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 *
Christoph Hellwigf758eea2011-04-21 18:19:44 -060081 * bdi->wb.list_lock
Dave Chinnera66979a2011-03-22 22:23:41 +110082 * sb_lock (fs/fs-writeback.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 * ->mapping->tree_lock (__sync_single_inode)
84 *
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070085 * ->i_mmap_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 * ->anon_vma.lock (vma_adjust)
87 *
88 * ->anon_vma.lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070089 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
Linus Torvalds1da177e2005-04-16 15:20:36 -070090 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -070091 * ->page_table_lock or pte_lock
Hugh Dickins5d337b92005-09-03 15:54:41 -070092 * ->swap_lock (try_to_unmap_one)
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 * ->private_lock (try_to_unmap_one)
94 * ->tree_lock (try_to_unmap_one)
95 * ->zone.lru_lock (follow_page->mark_page_accessed)
Nick Piggin053837f2006-01-18 17:42:27 -080096 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 * ->private_lock (page_remove_rmap->set_page_dirty)
98 * ->tree_lock (page_remove_rmap->set_page_dirty)
Christoph Hellwigf758eea2011-04-21 18:19:44 -060099 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100100 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600101 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100102 * ->inode->i_lock (zap_pte_range->set_page_dirty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
104 *
Andi Kleen6a460792009-09-16 11:50:15 +0200105 * (code doesn't rely on that order, so you could switch it around)
106 * ->tasklist_lock (memory_failure, collect_procs_ao)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700107 * ->i_mmap_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 */
109
110/*
Minchan Kime64a7822011-03-22 16:32:44 -0700111 * Delete a page from the page cache and free it. Caller has to make
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 * sure the page is locked and that nobody else uses it - or that usage
Nick Piggin19fd6232008-07-25 19:45:32 -0700113 * is safe. The caller must hold the mapping's tree_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 */
Minchan Kime64a7822011-03-22 16:32:44 -0700115void __delete_from_page_cache(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116{
117 struct address_space *mapping = page->mapping;
118
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600119 /*
120 * if we're uptodate, flush out into the cleancache, otherwise
121 * invalidate any existing cleancache entries. We can't leave
122 * stale data around in the cleancache once our page is gone
123 */
124 if (PageUptodate(page) && PageMappedToDisk(page))
125 cleancache_put_page(page);
126 else
127 cleancache_flush_page(mapping, page);
128
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 radix_tree_delete(&mapping->page_tree, page->index);
130 page->mapping = NULL;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700131 /* Leave page->index set: truncation lookup relies upon it */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 mapping->nrpages--;
Christoph Lameter347ce432006-06-30 01:55:35 -0700133 __dec_zone_page_state(page, NR_FILE_PAGES);
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700134 if (PageSwapBacked(page))
135 __dec_zone_page_state(page, NR_SHMEM);
Nick Piggin45426812007-07-15 23:38:12 -0700136 BUG_ON(page_mapped(page));
Linus Torvalds3a692792007-12-19 14:05:13 -0800137
138 /*
139 * Some filesystems seem to re-dirty the page even after
140 * the VM has canceled the dirty bit (eg ext3 journaling).
141 *
142 * Fix it up by doing a final dirty accounting check after
143 * having removed the page entirely.
144 */
145 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
146 dec_zone_page_state(page, NR_FILE_DIRTY);
147 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
148 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149}
150
Minchan Kim702cfbf2011-03-22 16:32:43 -0700151/**
152 * delete_from_page_cache - delete page from page cache
153 * @page: the page which the kernel is trying to remove from page cache
154 *
155 * This must be called only on pages that have been verified to be in the page
156 * cache and locked. It will never put the page into the free list, the caller
157 * has a reference on the page.
158 */
159void delete_from_page_cache(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160{
161 struct address_space *mapping = page->mapping;
Linus Torvalds6072d132010-12-01 13:35:19 -0500162 void (*freepage)(struct page *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163
Matt Mackallcd7619d2005-05-01 08:59:01 -0700164 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165
Linus Torvalds6072d132010-12-01 13:35:19 -0500166 freepage = mapping->a_ops->freepage;
Nick Piggin19fd6232008-07-25 19:45:32 -0700167 spin_lock_irq(&mapping->tree_lock);
Minchan Kime64a7822011-03-22 16:32:44 -0700168 __delete_from_page_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700169 spin_unlock_irq(&mapping->tree_lock);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700170 mem_cgroup_uncharge_cache_page(page);
Linus Torvalds6072d132010-12-01 13:35:19 -0500171
172 if (freepage)
173 freepage(page);
Minchan Kim97cecb52011-03-22 16:30:53 -0700174 page_cache_release(page);
175}
176EXPORT_SYMBOL(delete_from_page_cache);
177
Jens Axboe7eaceac2011-03-10 08:52:07 +0100178static int sleep_on_page(void *word)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 io_schedule();
181 return 0;
182}
183
Jens Axboe7eaceac2011-03-10 08:52:07 +0100184static int sleep_on_page_killable(void *word)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500185{
Jens Axboe7eaceac2011-03-10 08:52:07 +0100186 sleep_on_page(word);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500187 return fatal_signal_pending(current) ? -EINTR : 0;
188}
189
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700191 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700192 * @mapping: address space structure to write
193 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800194 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700195 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700197 * Start writeback against all of a mapping's dirty pages that lie
198 * within the byte offsets <start, end> inclusive.
199 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700201 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 * these two operations is that if a dirty page/buffer is encountered, it must
203 * be waited upon, and not just skipped over.
204 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800205int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
206 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207{
208 int ret;
209 struct writeback_control wbc = {
210 .sync_mode = sync_mode,
Nick Piggin05fe4782009-01-06 14:39:08 -0800211 .nr_to_write = LONG_MAX,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700212 .range_start = start,
213 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 };
215
216 if (!mapping_cap_writeback_dirty(mapping))
217 return 0;
218
219 ret = do_writepages(mapping, &wbc);
220 return ret;
221}
222
223static inline int __filemap_fdatawrite(struct address_space *mapping,
224 int sync_mode)
225{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700226 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227}
228
229int filemap_fdatawrite(struct address_space *mapping)
230{
231 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
232}
233EXPORT_SYMBOL(filemap_fdatawrite);
234
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400235int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800236 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237{
238 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
239}
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400240EXPORT_SYMBOL(filemap_fdatawrite_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241
Randy Dunlap485bb992006-06-23 02:03:49 -0700242/**
243 * filemap_flush - mostly a non-blocking flush
244 * @mapping: target address_space
245 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 * This is a mostly non-blocking flush. Not suitable for data-integrity
247 * purposes - I/O may not be started against all dirty pages.
248 */
249int filemap_flush(struct address_space *mapping)
250{
251 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
252}
253EXPORT_SYMBOL(filemap_flush);
254
Randy Dunlap485bb992006-06-23 02:03:49 -0700255/**
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200256 * filemap_fdatawait_range - wait for writeback to complete
257 * @mapping: address space structure to wait for
258 * @start_byte: offset in bytes where the range starts
259 * @end_byte: offset in bytes where the range ends (inclusive)
Randy Dunlap485bb992006-06-23 02:03:49 -0700260 *
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200261 * Walk the list of under-writeback pages of the given address space
262 * in the given range and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 */
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200264int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
265 loff_t end_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266{
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200267 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
268 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 struct pagevec pvec;
270 int nr_pages;
271 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200273 if (end_byte < start_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 return 0;
275
276 pagevec_init(&pvec, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 while ((index <= end) &&
278 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
279 PAGECACHE_TAG_WRITEBACK,
280 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
281 unsigned i;
282
283 for (i = 0; i < nr_pages; i++) {
284 struct page *page = pvec.pages[i];
285
286 /* until radix tree lookup accepts end_index */
287 if (page->index > end)
288 continue;
289
290 wait_on_page_writeback(page);
Rik van Riel212260a2011-01-13 15:46:06 -0800291 if (TestClearPageError(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 ret = -EIO;
293 }
294 pagevec_release(&pvec);
295 cond_resched();
296 }
297
298 /* Check for outstanding write errors */
299 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
300 ret = -ENOSPC;
301 if (test_and_clear_bit(AS_EIO, &mapping->flags))
302 ret = -EIO;
303
304 return ret;
305}
Jan Karad3bccb6f2009-08-17 19:30:27 +0200306EXPORT_SYMBOL(filemap_fdatawait_range);
307
308/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700309 * filemap_fdatawait - wait for all under-writeback pages to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 * @mapping: address space structure to wait for
Randy Dunlap485bb992006-06-23 02:03:49 -0700311 *
312 * Walk the list of under-writeback pages of the given address space
313 * and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 */
315int filemap_fdatawait(struct address_space *mapping)
316{
317 loff_t i_size = i_size_read(mapping->host);
318
319 if (i_size == 0)
320 return 0;
321
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200322 return filemap_fdatawait_range(mapping, 0, i_size - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323}
324EXPORT_SYMBOL(filemap_fdatawait);
325
326int filemap_write_and_wait(struct address_space *mapping)
327{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800328 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329
330 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800331 err = filemap_fdatawrite(mapping);
332 /*
333 * Even if the above returned error, the pages may be
334 * written partially (e.g. -ENOSPC), so we wait for it.
335 * But the -EIO is special case, it may indicate the worst
336 * thing (e.g. bug) happened, so we avoid waiting for it.
337 */
338 if (err != -EIO) {
339 int err2 = filemap_fdatawait(mapping);
340 if (!err)
341 err = err2;
342 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800344 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800346EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347
Randy Dunlap485bb992006-06-23 02:03:49 -0700348/**
349 * filemap_write_and_wait_range - write out & wait on a file range
350 * @mapping: the address_space for the pages
351 * @lstart: offset in bytes where the range starts
352 * @lend: offset in bytes where the range ends (inclusive)
353 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800354 * Write out and wait upon file offsets lstart->lend, inclusive.
355 *
356 * Note that `lend' is inclusive (describes the last byte to be written) so
357 * that this function can be used to write to the very end-of-file (end = -1).
358 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359int filemap_write_and_wait_range(struct address_space *mapping,
360 loff_t lstart, loff_t lend)
361{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800362 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363
364 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800365 err = __filemap_fdatawrite_range(mapping, lstart, lend,
366 WB_SYNC_ALL);
367 /* See comment of filemap_write_and_wait() */
368 if (err != -EIO) {
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200369 int err2 = filemap_fdatawait_range(mapping,
370 lstart, lend);
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800371 if (!err)
372 err = err2;
373 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800375 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376}
Chris Masonf6995582009-04-15 13:22:37 -0400377EXPORT_SYMBOL(filemap_write_and_wait_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378
Randy Dunlap485bb992006-06-23 02:03:49 -0700379/**
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700380 * replace_page_cache_page - replace a pagecache page with a new one
381 * @old: page to be replaced
382 * @new: page to replace with
383 * @gfp_mask: allocation mode
384 *
385 * This function replaces a page in the pagecache with a new one. On
386 * success it acquires the pagecache reference for the new page and
387 * drops it for the old page. Both the old and new pages must be
388 * locked. This function does not add the new page to the LRU, the
389 * caller must do that.
390 *
391 * The remove + add is atomic. The only way this function can fail is
392 * memory allocation failure.
393 */
394int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
395{
396 int error;
397 struct mem_cgroup *memcg = NULL;
398
399 VM_BUG_ON(!PageLocked(old));
400 VM_BUG_ON(!PageLocked(new));
401 VM_BUG_ON(new->mapping);
402
403 /*
404 * This is not page migration, but prepare_migration and
405 * end_migration does enough work for charge replacement.
406 *
407 * In the longer term we probably want a specialized function
408 * for moving the charge from old to new in a more efficient
409 * manner.
410 */
411 error = mem_cgroup_prepare_migration(old, new, &memcg, gfp_mask);
412 if (error)
413 return error;
414
415 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
416 if (!error) {
417 struct address_space *mapping = old->mapping;
418 void (*freepage)(struct page *);
419
420 pgoff_t offset = old->index;
421 freepage = mapping->a_ops->freepage;
422
423 page_cache_get(new);
424 new->mapping = mapping;
425 new->index = offset;
426
427 spin_lock_irq(&mapping->tree_lock);
Minchan Kime64a7822011-03-22 16:32:44 -0700428 __delete_from_page_cache(old);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700429 error = radix_tree_insert(&mapping->page_tree, offset, new);
430 BUG_ON(error);
431 mapping->nrpages++;
432 __inc_zone_page_state(new, NR_FILE_PAGES);
433 if (PageSwapBacked(new))
434 __inc_zone_page_state(new, NR_SHMEM);
435 spin_unlock_irq(&mapping->tree_lock);
436 radix_tree_preload_end();
437 if (freepage)
438 freepage(old);
439 page_cache_release(old);
440 mem_cgroup_end_migration(memcg, old, new, true);
441 } else {
442 mem_cgroup_end_migration(memcg, old, new, false);
443 }
444
445 return error;
446}
447EXPORT_SYMBOL_GPL(replace_page_cache_page);
448
449/**
Nick Piggine2867812008-07-25 19:45:30 -0700450 * add_to_page_cache_locked - add a locked page to the pagecache
Randy Dunlap485bb992006-06-23 02:03:49 -0700451 * @page: page to add
452 * @mapping: the page's address_space
453 * @offset: page index
454 * @gfp_mask: page allocation mode
455 *
Nick Piggine2867812008-07-25 19:45:30 -0700456 * This function is used to add a page to the pagecache. It must be locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 * This function does not add the page to the LRU. The caller must do that.
458 */
Nick Piggine2867812008-07-25 19:45:30 -0700459int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400460 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461{
Nick Piggine2867812008-07-25 19:45:30 -0700462 int error;
463
464 VM_BUG_ON(!PageLocked(page));
465
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);
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700482 if (PageSwapBacked(page))
483 __inc_zone_page_state(page, NR_SHMEM);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700484 spin_unlock_irq(&mapping->tree_lock);
Nick Piggine2867812008-07-25 19:45:30 -0700485 } else {
486 page->mapping = NULL;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700487 /* Leave page->index set: truncation relies upon it */
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700488 spin_unlock_irq(&mapping->tree_lock);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700489 mem_cgroup_uncharge_cache_page(page);
Nick Piggine2867812008-07-25 19:45:30 -0700490 page_cache_release(page);
491 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 radix_tree_preload_end();
Balbir Singh35c754d2008-02-07 00:14:05 -0800493 } else
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700494 mem_cgroup_uncharge_cache_page(page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800495out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 return error;
497}
Nick Piggine2867812008-07-25 19:45:30 -0700498EXPORT_SYMBOL(add_to_page_cache_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499
500int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400501 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502{
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700503 int ret;
504
505 /*
506 * Splice_read and readahead add shmem/tmpfs pages into the page cache
507 * before shmem_readpage has a chance to mark them as SwapBacked: they
KOSAKI Motohiroe9d6c152010-05-24 14:31:48 -0700508 * need to go on the anon lru below, and mem_cgroup_cache_charge
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700509 * (called in add_to_page_cache) needs to know where they're going too.
510 */
511 if (mapping_cap_swap_backed(mapping))
512 SetPageSwapBacked(page);
513
514 ret = add_to_page_cache(page, mapping, offset, gfp_mask);
515 if (ret == 0) {
516 if (page_is_file_cache(page))
517 lru_cache_add_file(page);
518 else
KOSAKI Motohiroe9d6c152010-05-24 14:31:48 -0700519 lru_cache_add_anon(page);
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700520 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 return ret;
522}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300523EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524
Paul Jackson44110fe2006-03-24 03:16:04 -0800525#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700526struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800527{
Miao Xiec0ff7452010-05-24 14:32:08 -0700528 int n;
529 struct page *page;
530
Paul Jackson44110fe2006-03-24 03:16:04 -0800531 if (cpuset_do_page_mem_spread()) {
Miao Xiec0ff7452010-05-24 14:32:08 -0700532 get_mems_allowed();
533 n = cpuset_mem_spread_node();
534 page = alloc_pages_exact_node(n, gfp, 0);
535 put_mems_allowed();
536 return page;
Paul Jackson44110fe2006-03-24 03:16:04 -0800537 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700538 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800539}
Nick Piggin2ae88142006-10-28 10:38:23 -0700540EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800541#endif
542
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543/*
544 * In order to wait for pages to become available there must be
545 * waitqueues associated with pages. By using a hash table of
546 * waitqueues where the bucket discipline is to maintain all
547 * waiters on the same queue and wake all when any of the pages
548 * become available, and for the woken contexts to check to be
549 * sure the appropriate page became available, this saves space
550 * at a cost of "thundering herd" phenomena during rare hash
551 * collisions.
552 */
553static wait_queue_head_t *page_waitqueue(struct page *page)
554{
555 const struct zone *zone = page_zone(page);
556
557 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
558}
559
560static inline void wake_up_page(struct page *page, int bit)
561{
562 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
563}
564
Harvey Harrison920c7a52008-02-04 22:29:26 -0800565void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566{
567 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
568
569 if (test_bit(bit_nr, &page->flags))
Jens Axboe7eaceac2011-03-10 08:52:07 +0100570 __wait_on_bit(page_waitqueue(page), &wait, sleep_on_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 TASK_UNINTERRUPTIBLE);
572}
573EXPORT_SYMBOL(wait_on_page_bit);
574
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700575int wait_on_page_bit_killable(struct page *page, int bit_nr)
576{
577 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
578
579 if (!test_bit(bit_nr, &page->flags))
580 return 0;
581
582 return __wait_on_bit(page_waitqueue(page), &wait,
583 sleep_on_page_killable, TASK_KILLABLE);
584}
585
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586/**
David Howells385e1ca5f2009-04-03 16:42:39 +0100587 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
Randy Dunlap697f6192009-04-13 14:39:54 -0700588 * @page: Page defining the wait queue of interest
589 * @waiter: Waiter to add to the queue
David Howells385e1ca5f2009-04-03 16:42:39 +0100590 *
591 * Add an arbitrary @waiter to the wait queue for the nominated @page.
592 */
593void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
594{
595 wait_queue_head_t *q = page_waitqueue(page);
596 unsigned long flags;
597
598 spin_lock_irqsave(&q->lock, flags);
599 __add_wait_queue(q, waiter);
600 spin_unlock_irqrestore(&q->lock, flags);
601}
602EXPORT_SYMBOL_GPL(add_page_wait_queue);
603
604/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700605 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 * @page: the page
607 *
608 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
609 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
610 * mechananism between PageLocked pages and PageWriteback pages is shared.
611 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
612 *
Nick Piggin8413ac92008-10-18 20:26:59 -0700613 * The mb is necessary to enforce ordering between the clear_bit and the read
614 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800616void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617{
Nick Piggin8413ac92008-10-18 20:26:59 -0700618 VM_BUG_ON(!PageLocked(page));
619 clear_bit_unlock(PG_locked, &page->flags);
620 smp_mb__after_clear_bit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 wake_up_page(page, PG_locked);
622}
623EXPORT_SYMBOL(unlock_page);
624
Randy Dunlap485bb992006-06-23 02:03:49 -0700625/**
626 * end_page_writeback - end writeback against a page
627 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 */
629void end_page_writeback(struct page *page)
630{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700631 if (TestClearPageReclaim(page))
632 rotate_reclaimable_page(page);
633
634 if (!test_clear_page_writeback(page))
635 BUG();
636
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 smp_mb__after_clear_bit();
638 wake_up_page(page, PG_writeback);
639}
640EXPORT_SYMBOL(end_page_writeback);
641
Randy Dunlap485bb992006-06-23 02:03:49 -0700642/**
643 * __lock_page - get a lock on the page, assuming we need to sleep to get it
644 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800646void __lock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647{
648 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
649
Jens Axboe7eaceac2011-03-10 08:52:07 +0100650 __wait_on_bit_lock(page_waitqueue(page), &wait, sleep_on_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 TASK_UNINTERRUPTIBLE);
652}
653EXPORT_SYMBOL(__lock_page);
654
Harvey Harrisonb5606c22008-02-13 15:03:16 -0800655int __lock_page_killable(struct page *page)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500656{
657 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
658
659 return __wait_on_bit_lock(page_waitqueue(page), &wait,
Jens Axboe7eaceac2011-03-10 08:52:07 +0100660 sleep_on_page_killable, TASK_KILLABLE);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500661}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300662EXPORT_SYMBOL_GPL(__lock_page_killable);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500663
Michel Lespinassed065bd82010-10-26 14:21:57 -0700664int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
665 unsigned int flags)
666{
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700667 if (flags & FAULT_FLAG_ALLOW_RETRY) {
668 /*
669 * CAUTION! In this case, mmap_sem is not released
670 * even though return 0.
671 */
672 if (flags & FAULT_FLAG_RETRY_NOWAIT)
673 return 0;
674
675 up_read(&mm->mmap_sem);
676 if (flags & FAULT_FLAG_KILLABLE)
677 wait_on_page_locked_killable(page);
678 else
Gleb Natapov318b2752011-03-22 16:30:51 -0700679 wait_on_page_locked(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -0700680 return 0;
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700681 } else {
682 if (flags & FAULT_FLAG_KILLABLE) {
683 int ret;
684
685 ret = __lock_page_killable(page);
686 if (ret) {
687 up_read(&mm->mmap_sem);
688 return 0;
689 }
690 } else
691 __lock_page(page);
692 return 1;
Michel Lespinassed065bd82010-10-26 14:21:57 -0700693 }
694}
695
Randy Dunlap485bb992006-06-23 02:03:49 -0700696/**
697 * find_get_page - find and get a page reference
698 * @mapping: the address_space to search
699 * @offset: the page index
700 *
Nick Pigginda6052f2006-09-25 23:31:35 -0700701 * Is there a pagecache struct page at the given (mapping, offset) tuple?
702 * If yes, increment its refcount and return it; if no, return NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 */
Nick Piggina60637c2008-07-25 19:45:31 -0700704struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705{
Nick Piggina60637c2008-07-25 19:45:31 -0700706 void **pagep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 struct page *page;
708
Nick Piggina60637c2008-07-25 19:45:31 -0700709 rcu_read_lock();
710repeat:
711 page = NULL;
712 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
713 if (pagep) {
714 page = radix_tree_deref_slot(pagep);
Nick Piggin27d20fd2010-11-11 14:05:19 -0800715 if (unlikely(!page))
716 goto out;
717 if (radix_tree_deref_retry(page))
Nick Piggina60637c2008-07-25 19:45:31 -0700718 goto repeat;
719
720 if (!page_cache_get_speculative(page))
721 goto repeat;
722
723 /*
724 * Has the page moved?
725 * This is part of the lockless pagecache protocol. See
726 * include/linux/pagemap.h for details.
727 */
728 if (unlikely(page != *pagep)) {
729 page_cache_release(page);
730 goto repeat;
731 }
732 }
Nick Piggin27d20fd2010-11-11 14:05:19 -0800733out:
Nick Piggina60637c2008-07-25 19:45:31 -0700734 rcu_read_unlock();
735
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 return page;
737}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738EXPORT_SYMBOL(find_get_page);
739
Randy Dunlap485bb992006-06-23 02:03:49 -0700740/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 * find_lock_page - locate, pin and lock a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700742 * @mapping: the address_space to search
743 * @offset: the page index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 *
745 * Locates the desired pagecache page, locks it, increments its reference
746 * count and returns its address.
747 *
748 * Returns zero if the page was not present. find_lock_page() may sleep.
749 */
Nick Piggina60637c2008-07-25 19:45:31 -0700750struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751{
752 struct page *page;
753
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754repeat:
Nick Piggina60637c2008-07-25 19:45:31 -0700755 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 if (page) {
Nick Piggina60637c2008-07-25 19:45:31 -0700757 lock_page(page);
758 /* Has the page been truncated? */
759 if (unlikely(page->mapping != mapping)) {
760 unlock_page(page);
761 page_cache_release(page);
762 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 }
Nick Piggina60637c2008-07-25 19:45:31 -0700764 VM_BUG_ON(page->index != offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 return page;
767}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768EXPORT_SYMBOL(find_lock_page);
769
770/**
771 * find_or_create_page - locate or add a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700772 * @mapping: the page's address_space
773 * @index: the page's index into the mapping
774 * @gfp_mask: page allocation mode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 *
776 * Locates a page in the pagecache. If the page is not present, a new page
777 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
778 * LRU list. The returned page is locked and has its reference count
779 * incremented.
780 *
781 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
782 * allocation!
783 *
784 * find_or_create_page() returns the desired page's address, or zero on
785 * memory exhaustion.
786 */
787struct page *find_or_create_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -0700788 pgoff_t index, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789{
Nick Piggineb2be182007-10-16 01:24:57 -0700790 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 int err;
792repeat:
793 page = find_lock_page(mapping, index);
794 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -0700795 page = __page_cache_alloc(gfp_mask);
796 if (!page)
797 return NULL;
Nick Piggin67d58ac2009-01-06 14:40:28 -0800798 /*
799 * We want a regular kernel memory (not highmem or DMA etc)
800 * allocation for the radix tree nodes, but we need to honour
801 * the context-specific requirements the caller has asked for.
802 * GFP_RECLAIM_MASK collects those requirements.
803 */
804 err = add_to_page_cache_lru(page, mapping, index,
805 (gfp_mask & GFP_RECLAIM_MASK));
Nick Piggineb2be182007-10-16 01:24:57 -0700806 if (unlikely(err)) {
807 page_cache_release(page);
808 page = NULL;
809 if (err == -EEXIST)
810 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 return page;
814}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815EXPORT_SYMBOL(find_or_create_page);
816
817/**
818 * find_get_pages - gang pagecache lookup
819 * @mapping: The address_space to search
820 * @start: The starting page index
821 * @nr_pages: The maximum number of pages
822 * @pages: Where the resulting pages are placed
823 *
824 * find_get_pages() will search for and return a group of up to
825 * @nr_pages pages in the mapping. The pages are placed at @pages.
826 * find_get_pages() takes a reference against the returned pages.
827 *
828 * The search returns a group of mapping-contiguous pages with ascending
829 * indexes. There may be holes in the indices due to not-present pages.
830 *
831 * find_get_pages() returns the number of pages which were found.
832 */
833unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
834 unsigned int nr_pages, struct page **pages)
835{
836 unsigned int i;
837 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700838 unsigned int nr_found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839
Nick Piggina60637c2008-07-25 19:45:31 -0700840 rcu_read_lock();
841restart:
842 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
843 (void ***)pages, start, nr_pages);
844 ret = 0;
845 for (i = 0; i < nr_found; i++) {
846 struct page *page;
847repeat:
848 page = radix_tree_deref_slot((void **)pages[i]);
849 if (unlikely(!page))
850 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -0700851
852 /*
853 * This can only trigger when the entry at index 0 moves out
854 * of or back to the root: none yet gotten, safe to restart.
855 */
Nick Piggin27d20fd2010-11-11 14:05:19 -0800856 if (radix_tree_deref_retry(page)) {
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -0700857 WARN_ON(start | i);
Nick Piggina60637c2008-07-25 19:45:31 -0700858 goto restart;
Nick Piggin27d20fd2010-11-11 14:05:19 -0800859 }
Nick Piggina60637c2008-07-25 19:45:31 -0700860
861 if (!page_cache_get_speculative(page))
862 goto repeat;
863
864 /* Has the page moved? */
865 if (unlikely(page != *((void **)pages[i]))) {
866 page_cache_release(page);
867 goto repeat;
868 }
869
870 pages[ret] = page;
871 ret++;
872 }
Hugh Dickins5b280c02011-03-22 16:33:07 -0700873
874 /*
875 * If all entries were removed before we could secure them,
876 * try again, because callers stop trying once 0 is returned.
877 */
878 if (unlikely(!ret && nr_found))
879 goto restart;
Nick Piggina60637c2008-07-25 19:45:31 -0700880 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 return ret;
882}
883
Jens Axboeebf43502006-04-27 08:46:01 +0200884/**
885 * find_get_pages_contig - gang contiguous pagecache lookup
886 * @mapping: The address_space to search
887 * @index: The starting page index
888 * @nr_pages: The maximum number of pages
889 * @pages: Where the resulting pages are placed
890 *
891 * find_get_pages_contig() works exactly like find_get_pages(), except
892 * that the returned number of pages are guaranteed to be contiguous.
893 *
894 * find_get_pages_contig() returns the number of pages which were found.
895 */
896unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
897 unsigned int nr_pages, struct page **pages)
898{
899 unsigned int i;
900 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700901 unsigned int nr_found;
Jens Axboeebf43502006-04-27 08:46:01 +0200902
Nick Piggina60637c2008-07-25 19:45:31 -0700903 rcu_read_lock();
904restart:
905 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
906 (void ***)pages, index, nr_pages);
907 ret = 0;
908 for (i = 0; i < nr_found; i++) {
909 struct page *page;
910repeat:
911 page = radix_tree_deref_slot((void **)pages[i]);
912 if (unlikely(!page))
913 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -0700914
915 /*
916 * This can only trigger when the entry at index 0 moves out
917 * of or back to the root: none yet gotten, safe to restart.
918 */
Nick Piggin27d20fd2010-11-11 14:05:19 -0800919 if (radix_tree_deref_retry(page))
Nick Piggina60637c2008-07-25 19:45:31 -0700920 goto restart;
921
Nick Piggina60637c2008-07-25 19:45:31 -0700922 if (!page_cache_get_speculative(page))
923 goto repeat;
924
925 /* Has the page moved? */
926 if (unlikely(page != *((void **)pages[i]))) {
927 page_cache_release(page);
928 goto repeat;
929 }
930
Nick Piggin9cbb4cb2011-01-13 15:45:51 -0800931 /*
932 * must check mapping and index after taking the ref.
933 * otherwise we can get both false positives and false
934 * negatives, which is just confusing to the caller.
935 */
936 if (page->mapping == NULL || page->index != index) {
937 page_cache_release(page);
938 break;
939 }
940
Nick Piggina60637c2008-07-25 19:45:31 -0700941 pages[ret] = page;
942 ret++;
Jens Axboeebf43502006-04-27 08:46:01 +0200943 index++;
944 }
Nick Piggina60637c2008-07-25 19:45:31 -0700945 rcu_read_unlock();
946 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +0200947}
David Howellsef71c152007-05-09 02:33:44 -0700948EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +0200949
Randy Dunlap485bb992006-06-23 02:03:49 -0700950/**
951 * find_get_pages_tag - find and return pages that match @tag
952 * @mapping: the address_space to search
953 * @index: the starting page index
954 * @tag: the tag index
955 * @nr_pages: the maximum number of pages
956 * @pages: where the resulting pages are placed
957 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -0700959 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 */
961unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
962 int tag, unsigned int nr_pages, struct page **pages)
963{
964 unsigned int i;
965 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700966 unsigned int nr_found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967
Nick Piggina60637c2008-07-25 19:45:31 -0700968 rcu_read_lock();
969restart:
970 nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree,
971 (void ***)pages, *index, nr_pages, tag);
972 ret = 0;
973 for (i = 0; i < nr_found; i++) {
974 struct page *page;
975repeat:
976 page = radix_tree_deref_slot((void **)pages[i]);
977 if (unlikely(!page))
978 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -0700979
980 /*
981 * This can only trigger when the entry at index 0 moves out
982 * of or back to the root: none yet gotten, safe to restart.
983 */
Nick Piggin27d20fd2010-11-11 14:05:19 -0800984 if (radix_tree_deref_retry(page))
Nick Piggina60637c2008-07-25 19:45:31 -0700985 goto restart;
986
987 if (!page_cache_get_speculative(page))
988 goto repeat;
989
990 /* Has the page moved? */
991 if (unlikely(page != *((void **)pages[i]))) {
992 page_cache_release(page);
993 goto repeat;
994 }
995
996 pages[ret] = page;
997 ret++;
998 }
Hugh Dickins5b280c02011-03-22 16:33:07 -0700999
1000 /*
1001 * If all entries were removed before we could secure them,
1002 * try again, because callers stop trying once 0 is returned.
1003 */
1004 if (unlikely(!ret && nr_found))
1005 goto restart;
Nick Piggina60637c2008-07-25 19:45:31 -07001006 rcu_read_unlock();
1007
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 if (ret)
1009 *index = pages[ret - 1]->index + 1;
Nick Piggina60637c2008-07-25 19:45:31 -07001010
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 return ret;
1012}
David Howellsef71c152007-05-09 02:33:44 -07001013EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014
Randy Dunlap485bb992006-06-23 02:03:49 -07001015/**
1016 * grab_cache_page_nowait - returns locked page at given index in given cache
1017 * @mapping: target address_space
1018 * @index: the page index
1019 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08001020 * Same as grab_cache_page(), but do not wait if the page is unavailable.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 * This is intended for speculative data generators, where the data can
1022 * be regenerated if the page couldn't be grabbed. This routine should
1023 * be safe to call while holding the lock for another page.
1024 *
1025 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
1026 * and deadlock against the caller's locked page.
1027 */
1028struct page *
Fengguang Wu57f6b962007-10-16 01:24:37 -07001029grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030{
1031 struct page *page = find_get_page(mapping, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032
1033 if (page) {
Nick Piggin529ae9a2008-08-02 12:01:03 +02001034 if (trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 return page;
1036 page_cache_release(page);
1037 return NULL;
1038 }
Nick Piggin2ae88142006-10-28 10:38:23 -07001039 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
Nick Piggin67d58ac2009-01-06 14:40:28 -08001040 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 page_cache_release(page);
1042 page = NULL;
1043 }
1044 return page;
1045}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046EXPORT_SYMBOL(grab_cache_page_nowait);
1047
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001048/*
1049 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1050 * a _large_ part of the i/o request. Imagine the worst scenario:
1051 *
1052 * ---R__________________________________________B__________
1053 * ^ reading here ^ bad block(assume 4k)
1054 *
1055 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1056 * => failing the whole request => read(R) => read(R+1) =>
1057 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1058 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1059 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1060 *
1061 * It is going insane. Fix it by quickly scaling down the readahead size.
1062 */
1063static void shrink_readahead_size_eio(struct file *filp,
1064 struct file_ra_state *ra)
1065{
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001066 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001067}
1068
Randy Dunlap485bb992006-06-23 02:03:49 -07001069/**
Christoph Hellwig36e78912008-02-08 04:21:24 -08001070 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001071 * @filp: the file to read
1072 * @ppos: current file position
1073 * @desc: read_descriptor
1074 * @actor: read method
1075 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -07001077 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 *
1079 * This is really ugly. But the goto's actually try to clarify some
1080 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 */
Christoph Hellwig36e78912008-02-08 04:21:24 -08001082static void do_generic_file_read(struct file *filp, loff_t *ppos,
1083 read_descriptor_t *desc, read_actor_t actor)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084{
Christoph Hellwig36e78912008-02-08 04:21:24 -08001085 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -08001087 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001088 pgoff_t index;
1089 pgoff_t last_index;
1090 pgoff_t prev_index;
1091 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -07001092 unsigned int prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 index = *ppos >> PAGE_CACHE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -07001096 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1097 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1099 offset = *ppos & ~PAGE_CACHE_MASK;
1100
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 for (;;) {
1102 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001103 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001104 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 unsigned long nr, ret;
1106
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108find_page:
1109 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001110 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001111 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001112 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001113 index, last_index - index);
1114 page = find_get_page(mapping, index);
1115 if (unlikely(page == NULL))
1116 goto no_cached_page;
1117 }
1118 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001119 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001120 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001121 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001123 if (!PageUptodate(page)) {
1124 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1125 !mapping->a_ops->is_partially_uptodate)
1126 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001127 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001128 goto page_not_up_to_date;
Dave Hansen8d056cb2010-11-11 14:05:15 -08001129 /* Did it get truncated before we got the lock? */
1130 if (!page->mapping)
1131 goto page_not_up_to_date_locked;
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001132 if (!mapping->a_ops->is_partially_uptodate(page,
1133 desc, offset))
1134 goto page_not_up_to_date_locked;
1135 unlock_page(page);
1136 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001138 /*
1139 * i_size must be checked after we know the page is Uptodate.
1140 *
1141 * Checking i_size after the check allows us to calculate
1142 * the correct value for "nr", which means the zero-filled
1143 * part of the page is not copied back to userspace (unless
1144 * another truncate extends the file - this is desired though).
1145 */
1146
1147 isize = i_size_read(inode);
1148 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1149 if (unlikely(!isize || index > end_index)) {
1150 page_cache_release(page);
1151 goto out;
1152 }
1153
1154 /* nr is the maximum number of bytes to copy from this page */
1155 nr = PAGE_CACHE_SIZE;
1156 if (index == end_index) {
1157 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1158 if (nr <= offset) {
1159 page_cache_release(page);
1160 goto out;
1161 }
1162 }
1163 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164
1165 /* If users can be writing to this page using arbitrary
1166 * virtual addresses, take care about potential aliasing
1167 * before reading the page on the kernel side.
1168 */
1169 if (mapping_writably_mapped(mapping))
1170 flush_dcache_page(page);
1171
1172 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001173 * When a sequential read accesses a page several times,
1174 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 */
Jan Karaec0f1632007-05-06 14:49:25 -07001176 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 mark_page_accessed(page);
1178 prev_index = index;
1179
1180 /*
1181 * Ok, we have the page, and it's up-to-date, so
1182 * now we can copy it to user space...
1183 *
1184 * The actor routine returns how many bytes were actually used..
1185 * NOTE! This may not be the same as how much of a user buffer
1186 * we filled up (we may be padding etc), so we can only update
1187 * "pos" here (the actor routine has to update the user buffer
1188 * pointers and the remaining count).
1189 */
1190 ret = actor(desc, page, offset, nr);
1191 offset += ret;
1192 index += offset >> PAGE_CACHE_SHIFT;
1193 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001194 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195
1196 page_cache_release(page);
1197 if (ret == nr && desc->count)
1198 continue;
1199 goto out;
1200
1201page_not_up_to_date:
1202 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001203 error = lock_page_killable(page);
1204 if (unlikely(error))
1205 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001207page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001208 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 if (!page->mapping) {
1210 unlock_page(page);
1211 page_cache_release(page);
1212 continue;
1213 }
1214
1215 /* Did somebody else fill it already? */
1216 if (PageUptodate(page)) {
1217 unlock_page(page);
1218 goto page_ok;
1219 }
1220
1221readpage:
Jeff Moyer91803b42010-05-26 11:49:40 -04001222 /*
1223 * A previous I/O error may have been due to temporary
1224 * failures, eg. multipath errors.
1225 * PG_error will be set again if readpage fails.
1226 */
1227 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 /* Start the actual read. The read will unlock the page. */
1229 error = mapping->a_ops->readpage(filp, page);
1230
Zach Brown994fc28c2005-12-15 14:28:17 -08001231 if (unlikely(error)) {
1232 if (error == AOP_TRUNCATED_PAGE) {
1233 page_cache_release(page);
1234 goto find_page;
1235 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001237 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238
1239 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001240 error = lock_page_killable(page);
1241 if (unlikely(error))
1242 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 if (!PageUptodate(page)) {
1244 if (page->mapping == NULL) {
1245 /*
Christoph Hellwig2ecdc822010-01-26 17:27:20 +01001246 * invalidate_mapping_pages got it
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 */
1248 unlock_page(page);
1249 page_cache_release(page);
1250 goto find_page;
1251 }
1252 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001253 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001254 error = -EIO;
1255 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 }
1257 unlock_page(page);
1258 }
1259
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 goto page_ok;
1261
1262readpage_error:
1263 /* UHHUH! A synchronous read error occurred. Report it */
1264 desc->error = error;
1265 page_cache_release(page);
1266 goto out;
1267
1268no_cached_page:
1269 /*
1270 * Ok, it wasn't cached, so we need to create a new
1271 * page..
1272 */
Nick Piggineb2be182007-10-16 01:24:57 -07001273 page = page_cache_alloc_cold(mapping);
1274 if (!page) {
1275 desc->error = -ENOMEM;
1276 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 }
Nick Piggineb2be182007-10-16 01:24:57 -07001278 error = add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 index, GFP_KERNEL);
1280 if (error) {
Nick Piggineb2be182007-10-16 01:24:57 -07001281 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 if (error == -EEXIST)
1283 goto find_page;
1284 desc->error = error;
1285 goto out;
1286 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 goto readpage;
1288 }
1289
1290out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001291 ra->prev_pos = prev_index;
1292 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1293 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001295 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07001296 file_accessed(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298
1299int file_read_actor(read_descriptor_t *desc, struct page *page,
1300 unsigned long offset, unsigned long size)
1301{
1302 char *kaddr;
1303 unsigned long left, count = desc->count;
1304
1305 if (size > count)
1306 size = count;
1307
1308 /*
1309 * Faults on the destination of a read are common, so do it before
1310 * taking the kmap.
1311 */
1312 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1313 kaddr = kmap_atomic(page, KM_USER0);
1314 left = __copy_to_user_inatomic(desc->arg.buf,
1315 kaddr + offset, size);
1316 kunmap_atomic(kaddr, KM_USER0);
1317 if (left == 0)
1318 goto success;
1319 }
1320
1321 /* Do it the slow way */
1322 kaddr = kmap(page);
1323 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1324 kunmap(page);
1325
1326 if (left) {
1327 size -= left;
1328 desc->error = -EFAULT;
1329 }
1330success:
1331 desc->count = count - size;
1332 desc->written += size;
1333 desc->arg.buf += size;
1334 return size;
1335}
1336
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001337/*
1338 * Performs necessary checks before doing a write
1339 * @iov: io vector request
1340 * @nr_segs: number of segments in the iovec
1341 * @count: number of bytes to write
1342 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1343 *
1344 * Adjust number of segments and amount of bytes to write (nr_segs should be
1345 * properly initialized first). Returns appropriate error code that caller
1346 * should return or zero in case that write should be allowed.
1347 */
1348int generic_segment_checks(const struct iovec *iov,
1349 unsigned long *nr_segs, size_t *count, int access_flags)
1350{
1351 unsigned long seg;
1352 size_t cnt = 0;
1353 for (seg = 0; seg < *nr_segs; seg++) {
1354 const struct iovec *iv = &iov[seg];
1355
1356 /*
1357 * If any segment has a negative length, or the cumulative
1358 * length ever wraps negative then return -EINVAL.
1359 */
1360 cnt += iv->iov_len;
1361 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1362 return -EINVAL;
1363 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1364 continue;
1365 if (seg == 0)
1366 return -EFAULT;
1367 *nr_segs = seg;
1368 cnt -= iv->iov_len; /* This segment is no good */
1369 break;
1370 }
1371 *count = cnt;
1372 return 0;
1373}
1374EXPORT_SYMBOL(generic_segment_checks);
1375
Randy Dunlap485bb992006-06-23 02:03:49 -07001376/**
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001377 * generic_file_aio_read - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001378 * @iocb: kernel I/O control block
1379 * @iov: io vector request
1380 * @nr_segs: number of segments in the iovec
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001381 * @pos: current file position
Randy Dunlap485bb992006-06-23 02:03:49 -07001382 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 * This is the "read()" routine for all filesystems
1384 * that can use the page cache directly.
1385 */
1386ssize_t
Badari Pulavarty543ade12006-09-30 23:28:48 -07001387generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1388 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389{
1390 struct file *filp = iocb->ki_filp;
1391 ssize_t retval;
Josef Bacik66f998f2010-05-23 11:00:54 -04001392 unsigned long seg = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 size_t count;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001394 loff_t *ppos = &iocb->ki_pos;
Jens Axboe55602dd2010-06-24 15:05:37 +02001395 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396
1397 count = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001398 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1399 if (retval)
1400 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401
Jens Axboe55602dd2010-06-24 15:05:37 +02001402 blk_start_plug(&plug);
1403
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1405 if (filp->f_flags & O_DIRECT) {
Badari Pulavarty543ade12006-09-30 23:28:48 -07001406 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 struct address_space *mapping;
1408 struct inode *inode;
1409
1410 mapping = filp->f_mapping;
1411 inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 if (!count)
1413 goto out; /* skip atime */
1414 size = i_size_read(inode);
1415 if (pos < size) {
Nick Piggin48b47c52009-01-06 14:40:22 -08001416 retval = filemap_write_and_wait_range(mapping, pos,
1417 pos + iov_length(iov, nr_segs) - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07001418 if (!retval) {
1419 retval = mapping->a_ops->direct_IO(READ, iocb,
1420 iov, pos, nr_segs);
1421 }
Josef Bacik66f998f2010-05-23 11:00:54 -04001422 if (retval > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 *ppos = pos + retval;
Josef Bacik66f998f2010-05-23 11:00:54 -04001424 count -= retval;
1425 }
1426
1427 /*
1428 * Btrfs can have a short DIO read if we encounter
1429 * compressed extents, so if there was an error, or if
1430 * we've already read everything we wanted to, or if
1431 * there was a short read because we hit EOF, go ahead
1432 * and return. Otherwise fallthrough to buffered io for
1433 * the rest of the read.
1434 */
1435 if (retval < 0 || !count || *ppos >= size) {
Hugh Dickins11fa9772008-07-23 21:27:34 -07001436 file_accessed(filp);
1437 goto out;
1438 }
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001439 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 }
1441
Josef Bacik66f998f2010-05-23 11:00:54 -04001442 count = retval;
Hugh Dickins11fa9772008-07-23 21:27:34 -07001443 for (seg = 0; seg < nr_segs; seg++) {
1444 read_descriptor_t desc;
Josef Bacik66f998f2010-05-23 11:00:54 -04001445 loff_t offset = 0;
1446
1447 /*
1448 * If we did a short DIO read we need to skip the section of the
1449 * iov that we've already read data into.
1450 */
1451 if (count) {
1452 if (count > iov[seg].iov_len) {
1453 count -= iov[seg].iov_len;
1454 continue;
1455 }
1456 offset = count;
1457 count = 0;
1458 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459
Hugh Dickins11fa9772008-07-23 21:27:34 -07001460 desc.written = 0;
Josef Bacik66f998f2010-05-23 11:00:54 -04001461 desc.arg.buf = iov[seg].iov_base + offset;
1462 desc.count = iov[seg].iov_len - offset;
Hugh Dickins11fa9772008-07-23 21:27:34 -07001463 if (desc.count == 0)
1464 continue;
1465 desc.error = 0;
1466 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1467 retval += desc.written;
1468 if (desc.error) {
1469 retval = retval ?: desc.error;
1470 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 }
Hugh Dickins11fa9772008-07-23 21:27:34 -07001472 if (desc.count > 0)
1473 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 }
1475out:
Jens Axboe55602dd2010-06-24 15:05:37 +02001476 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 return retval;
1478}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479EXPORT_SYMBOL(generic_file_aio_read);
1480
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481static ssize_t
1482do_readahead(struct address_space *mapping, struct file *filp,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001483 pgoff_t index, unsigned long nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484{
1485 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1486 return -EINVAL;
1487
Wu Fengguangf7e839d2009-06-16 15:31:20 -07001488 force_page_cache_readahead(mapping, filp, index, nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 return 0;
1490}
1491
Heiko Carstens6673e0c2009-01-14 14:14:02 +01001492SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493{
1494 ssize_t ret;
1495 struct file *file;
1496
1497 ret = -EBADF;
1498 file = fget(fd);
1499 if (file) {
1500 if (file->f_mode & FMODE_READ) {
1501 struct address_space *mapping = file->f_mapping;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001502 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1503 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 unsigned long len = end - start + 1;
1505 ret = do_readahead(mapping, file, start, len);
1506 }
1507 fput(file);
1508 }
1509 return ret;
1510}
Heiko Carstens6673e0c2009-01-14 14:14:02 +01001511#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1512asmlinkage long SyS_readahead(long fd, loff_t offset, long count)
1513{
1514 return SYSC_readahead((int) fd, offset, (size_t) count);
1515}
1516SYSCALL_ALIAS(sys_readahead, SyS_readahead);
1517#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518
1519#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001520/**
1521 * page_cache_read - adds requested page to the page cache if not already there
1522 * @file: file to read
1523 * @offset: page index
1524 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 * This adds the requested page to the page cache if it isn't already there,
1526 * and schedules an I/O to read in its contents from disk.
1527 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001528static int page_cache_read(struct file *file, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529{
1530 struct address_space *mapping = file->f_mapping;
1531 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001532 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533
Zach Brown994fc28c2005-12-15 14:28:17 -08001534 do {
1535 page = page_cache_alloc_cold(mapping);
1536 if (!page)
1537 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538
Zach Brown994fc28c2005-12-15 14:28:17 -08001539 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1540 if (ret == 0)
1541 ret = mapping->a_ops->readpage(file, page);
1542 else if (ret == -EEXIST)
1543 ret = 0; /* losing race to add is OK */
1544
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546
Zach Brown994fc28c2005-12-15 14:28:17 -08001547 } while (ret == AOP_TRUNCATED_PAGE);
1548
1549 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550}
1551
1552#define MMAP_LOTSAMISS (100)
1553
Linus Torvaldsef00e082009-06-16 15:31:25 -07001554/*
1555 * Synchronous readahead happens when we don't even find
1556 * a page in the page cache at all.
1557 */
1558static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1559 struct file_ra_state *ra,
1560 struct file *file,
1561 pgoff_t offset)
1562{
1563 unsigned long ra_pages;
1564 struct address_space *mapping = file->f_mapping;
1565
1566 /* If we don't want any read-ahead, don't bother */
1567 if (VM_RandomReadHint(vma))
1568 return;
Wu Fengguang275b12b2011-05-24 17:12:28 -07001569 if (!ra->ra_pages)
1570 return;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001571
Wu Fengguang2cbea1d2011-05-24 17:12:30 -07001572 if (VM_SequentialReadHint(vma)) {
Wu Fengguang7ffc59b2009-06-16 15:31:38 -07001573 page_cache_sync_readahead(mapping, ra, file, offset,
1574 ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001575 return;
1576 }
1577
Andi Kleen207d04b2011-05-24 17:12:29 -07001578 /* Avoid banging the cache line if not needed */
1579 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001580 ra->mmap_miss++;
1581
1582 /*
1583 * Do we miss much more than hit in this file? If so,
1584 * stop bothering with read-ahead. It will only hurt.
1585 */
1586 if (ra->mmap_miss > MMAP_LOTSAMISS)
1587 return;
1588
Wu Fengguangd30a1102009-06-16 15:31:30 -07001589 /*
1590 * mmap read-around
1591 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001592 ra_pages = max_sane_readahead(ra->ra_pages);
Wu Fengguang275b12b2011-05-24 17:12:28 -07001593 ra->start = max_t(long, 0, offset - ra_pages / 2);
1594 ra->size = ra_pages;
Wu Fengguang2cbea1d2011-05-24 17:12:30 -07001595 ra->async_size = ra_pages / 4;
Wu Fengguang275b12b2011-05-24 17:12:28 -07001596 ra_submit(ra, mapping, file);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001597}
1598
1599/*
1600 * Asynchronous readahead happens when we find the page and PG_readahead,
1601 * so we want to possibly extend the readahead further..
1602 */
1603static void do_async_mmap_readahead(struct vm_area_struct *vma,
1604 struct file_ra_state *ra,
1605 struct file *file,
1606 struct page *page,
1607 pgoff_t offset)
1608{
1609 struct address_space *mapping = file->f_mapping;
1610
1611 /* If we don't want any read-ahead, don't bother */
1612 if (VM_RandomReadHint(vma))
1613 return;
1614 if (ra->mmap_miss > 0)
1615 ra->mmap_miss--;
1616 if (PageReadahead(page))
Wu Fengguang2fad6f52009-06-16 15:31:29 -07001617 page_cache_async_readahead(mapping, ra, file,
1618 page, offset, ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001619}
1620
Randy Dunlap485bb992006-06-23 02:03:49 -07001621/**
Nick Piggin54cb8822007-07-19 01:46:59 -07001622 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07001623 * @vma: vma in which the fault was taken
1624 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07001625 *
Nick Piggin54cb8822007-07-19 01:46:59 -07001626 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 * mapped memory region to read in file data during a page fault.
1628 *
1629 * The goto's are kind of ugly, but this streamlines the normal case of having
1630 * it in the page cache, and handles the special cases reasonably without
1631 * having a lot of duplicated code.
1632 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001633int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634{
1635 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07001636 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 struct address_space *mapping = file->f_mapping;
1638 struct file_ra_state *ra = &file->f_ra;
1639 struct inode *inode = mapping->host;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001640 pgoff_t offset = vmf->pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 struct page *page;
Jan Kara2004dc82008-02-08 04:20:11 -08001642 pgoff_t size;
Nick Piggin83c54072007-07-19 01:47:05 -07001643 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001646 if (offset >= size)
Linus Torvalds5307cc12007-10-31 09:19:46 -07001647 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 * Do we have something in the page cache already?
1651 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001652 page = find_get_page(mapping, offset);
1653 if (likely(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07001655 * We found the page, so try async readahead before
1656 * waiting for the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001658 do_async_mmap_readahead(vma, ra, file, page, offset);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001659 } else {
1660 /* No page in the page cache at all */
1661 do_sync_mmap_readahead(vma, ra, file, offset);
1662 count_vm_event(PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07001663 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001664 ret = VM_FAULT_MAJOR;
1665retry_find:
Michel Lespinasseb522c942010-10-26 14:21:56 -07001666 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 if (!page)
1668 goto no_cached_page;
1669 }
1670
Michel Lespinassed88c0922010-11-02 13:05:18 -07001671 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1672 page_cache_release(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -07001673 return ret | VM_FAULT_RETRY;
Michel Lespinassed88c0922010-11-02 13:05:18 -07001674 }
Michel Lespinasseb522c942010-10-26 14:21:56 -07001675
1676 /* Did it get truncated? */
1677 if (unlikely(page->mapping != mapping)) {
1678 unlock_page(page);
1679 put_page(page);
1680 goto retry_find;
1681 }
1682 VM_BUG_ON(page->index != offset);
1683
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 /*
Nick Piggind00806b2007-07-19 01:46:57 -07001685 * We have a locked page in the page cache, now we need to check
1686 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 */
Nick Piggind00806b2007-07-19 01:46:57 -07001688 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 goto page_not_uptodate;
1690
Linus Torvaldsef00e082009-06-16 15:31:25 -07001691 /*
1692 * Found the page and have a reference on it.
1693 * We must recheck i_size under page lock.
1694 */
Nick Piggind00806b2007-07-19 01:46:57 -07001695 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001696 if (unlikely(offset >= size)) {
Nick Piggind00806b2007-07-19 01:46:57 -07001697 unlock_page(page);
Yan Zheng745ad482007-10-08 10:08:37 -07001698 page_cache_release(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07001699 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001700 }
1701
Nick Piggind0217ac2007-07-19 01:47:03 -07001702 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07001703 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705no_cached_page:
1706 /*
1707 * We're only likely to ever get here if MADV_RANDOM is in
1708 * effect.
1709 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001710 error = page_cache_read(file, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711
1712 /*
1713 * The page we want has now been added to the page cache.
1714 * In the unlikely event that someone removed it in the
1715 * meantime, we'll just come back here and read it again.
1716 */
1717 if (error >= 0)
1718 goto retry_find;
1719
1720 /*
1721 * An error return from page_cache_read can result if the
1722 * system is low on memory, or a problem occurs while trying
1723 * to schedule I/O.
1724 */
1725 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07001726 return VM_FAULT_OOM;
1727 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728
1729page_not_uptodate:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 /*
1731 * Umm, take care of errors if the page isn't up-to-date.
1732 * Try to re-read it _once_. We do this synchronously,
1733 * because there really aren't any performance issues here
1734 * and we need to check for errors.
1735 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001737 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07001738 if (!error) {
1739 wait_on_page_locked(page);
1740 if (!PageUptodate(page))
1741 error = -EIO;
1742 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07001744
1745 if (!error || error == AOP_TRUNCATED_PAGE)
1746 goto retry_find;
1747
1748 /* Things didn't work out. Return zero to tell the mm layer so. */
1749 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07001750 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07001751}
1752EXPORT_SYMBOL(filemap_fault);
1753
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001754const struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07001755 .fault = filemap_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756};
1757
1758/* This is used for a general mmap of a disk file */
1759
1760int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1761{
1762 struct address_space *mapping = file->f_mapping;
1763
1764 if (!mapping->a_ops->readpage)
1765 return -ENOEXEC;
1766 file_accessed(file);
1767 vma->vm_ops = &generic_file_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001768 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 return 0;
1770}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771
1772/*
1773 * This is for filesystems which do not implement ->writepage.
1774 */
1775int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1776{
1777 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1778 return -EINVAL;
1779 return generic_file_mmap(file, vma);
1780}
1781#else
1782int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1783{
1784 return -ENOSYS;
1785}
1786int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1787{
1788 return -ENOSYS;
1789}
1790#endif /* CONFIG_MMU */
1791
1792EXPORT_SYMBOL(generic_file_mmap);
1793EXPORT_SYMBOL(generic_file_readonly_mmap);
1794
Nick Piggin6fe69002007-05-06 14:49:04 -07001795static struct page *__read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001796 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001797 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001798 void *data,
1799 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800{
Nick Piggineb2be182007-10-16 01:24:57 -07001801 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 int err;
1803repeat:
1804 page = find_get_page(mapping, index);
1805 if (!page) {
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001806 page = __page_cache_alloc(gfp | __GFP_COLD);
Nick Piggineb2be182007-10-16 01:24:57 -07001807 if (!page)
1808 return ERR_PTR(-ENOMEM);
1809 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
1810 if (unlikely(err)) {
1811 page_cache_release(page);
1812 if (err == -EEXIST)
1813 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 return ERR_PTR(err);
1816 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 err = filler(data, page);
1818 if (err < 0) {
1819 page_cache_release(page);
1820 page = ERR_PTR(err);
1821 }
1822 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823 return page;
1824}
1825
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001826static struct page *do_read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001827 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001828 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001829 void *data,
1830 gfp_t gfp)
1831
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832{
1833 struct page *page;
1834 int err;
1835
1836retry:
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001837 page = __read_cache_page(mapping, index, filler, data, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01001839 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 if (PageUptodate(page))
1841 goto out;
1842
1843 lock_page(page);
1844 if (!page->mapping) {
1845 unlock_page(page);
1846 page_cache_release(page);
1847 goto retry;
1848 }
1849 if (PageUptodate(page)) {
1850 unlock_page(page);
1851 goto out;
1852 }
1853 err = filler(data, page);
1854 if (err < 0) {
1855 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01001856 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 }
David Howellsc855ff32007-05-09 13:42:20 +01001858out:
Nick Piggin6fe69002007-05-06 14:49:04 -07001859 mark_page_accessed(page);
1860 return page;
1861}
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001862
1863/**
1864 * read_cache_page_async - read into page cache, fill it if needed
1865 * @mapping: the page's address_space
1866 * @index: the page index
1867 * @filler: function to perform the read
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001868 * @data: first arg to filler(data, page) function, often left as NULL
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001869 *
1870 * Same as read_cache_page, but don't wait for page to become unlocked
1871 * after submitting it to the filler.
1872 *
1873 * Read into the page cache. If a page already exists, and PageUptodate() is
1874 * not set, try to fill the page but don't wait for it to become unlocked.
1875 *
1876 * If the page does not get brought uptodate, return -EIO.
1877 */
1878struct page *read_cache_page_async(struct address_space *mapping,
1879 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001880 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001881 void *data)
1882{
1883 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
1884}
Nick Piggin6fe69002007-05-06 14:49:04 -07001885EXPORT_SYMBOL(read_cache_page_async);
1886
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001887static struct page *wait_on_page_read(struct page *page)
1888{
1889 if (!IS_ERR(page)) {
1890 wait_on_page_locked(page);
1891 if (!PageUptodate(page)) {
1892 page_cache_release(page);
1893 page = ERR_PTR(-EIO);
1894 }
1895 }
1896 return page;
1897}
1898
1899/**
1900 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
1901 * @mapping: the page's address_space
1902 * @index: the page index
1903 * @gfp: the page allocator flags to use if allocating
1904 *
1905 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
1906 * any new page allocations done using the specified allocation flags. Note
1907 * that the Radix tree operations will still use GFP_KERNEL, so you can't
1908 * expect to do this atomically or anything like that - but you can pass in
1909 * other page requirements.
1910 *
1911 * If the page does not get brought uptodate, return -EIO.
1912 */
1913struct page *read_cache_page_gfp(struct address_space *mapping,
1914 pgoff_t index,
1915 gfp_t gfp)
1916{
1917 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
1918
1919 return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp));
1920}
1921EXPORT_SYMBOL(read_cache_page_gfp);
1922
Nick Piggin6fe69002007-05-06 14:49:04 -07001923/**
1924 * read_cache_page - read into page cache, fill it if needed
1925 * @mapping: the page's address_space
1926 * @index: the page index
1927 * @filler: function to perform the read
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001928 * @data: first arg to filler(data, page) function, often left as NULL
Nick Piggin6fe69002007-05-06 14:49:04 -07001929 *
1930 * Read into the page cache. If a page already exists, and PageUptodate() is
1931 * not set, try to fill the page then wait for it to become unlocked.
1932 *
1933 * If the page does not get brought uptodate, return -EIO.
1934 */
1935struct page *read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001936 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001937 int (*filler)(void *, struct page *),
Nick Piggin6fe69002007-05-06 14:49:04 -07001938 void *data)
1939{
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001940 return wait_on_page_read(read_cache_page_async(mapping, index, filler, data));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942EXPORT_SYMBOL(read_cache_page);
1943
1944/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 * The logic we want is
1946 *
1947 * if suid or (sgid and xgrp)
1948 * remove privs
1949 */
Jens Axboe01de85e2006-10-17 19:50:36 +02001950int should_remove_suid(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951{
1952 mode_t mode = dentry->d_inode->i_mode;
1953 int kill = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954
1955 /* suid always must be killed */
1956 if (unlikely(mode & S_ISUID))
1957 kill = ATTR_KILL_SUID;
1958
1959 /*
1960 * sgid without any exec bits is just a mandatory locking mark; leave
1961 * it alone. If some exec bits are set, it's a real sgid; kill it.
1962 */
1963 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1964 kill |= ATTR_KILL_SGID;
1965
Dmitri Monakhov7f5ff762008-12-01 14:34:56 -08001966 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
Jens Axboe01de85e2006-10-17 19:50:36 +02001967 return kill;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968
Jens Axboe01de85e2006-10-17 19:50:36 +02001969 return 0;
1970}
Mark Fashehd23a1472006-10-17 17:05:18 -07001971EXPORT_SYMBOL(should_remove_suid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001972
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +02001973static int __remove_suid(struct dentry *dentry, int kill)
Jens Axboe01de85e2006-10-17 19:50:36 +02001974{
1975 struct iattr newattrs;
1976
1977 newattrs.ia_valid = ATTR_FORCE | kill;
1978 return notify_change(dentry, &newattrs);
1979}
1980
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001981int file_remove_suid(struct file *file)
Jens Axboe01de85e2006-10-17 19:50:36 +02001982{
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001983 struct dentry *dentry = file->f_path.dentry;
Andi Kleen69b45732011-05-28 08:25:51 -07001984 struct inode *inode = dentry->d_inode;
1985 int killsuid;
1986 int killpriv;
Serge E. Hallynb5376772007-10-16 23:31:36 -07001987 int error = 0;
Jens Axboe01de85e2006-10-17 19:50:36 +02001988
Andi Kleen69b45732011-05-28 08:25:51 -07001989 /* Fast path for nothing security related */
1990 if (IS_NOSEC(inode))
1991 return 0;
1992
1993 killsuid = should_remove_suid(dentry);
1994 killpriv = security_inode_need_killpriv(dentry);
1995
Serge E. Hallynb5376772007-10-16 23:31:36 -07001996 if (killpriv < 0)
1997 return killpriv;
1998 if (killpriv)
1999 error = security_inode_killpriv(dentry);
2000 if (!error && killsuid)
2001 error = __remove_suid(dentry, killsuid);
Al Viro9e1f1de2011-06-03 18:24:58 -04002002 if (!error && (inode->i_sb->s_flags & MS_NOSEC))
Andi Kleen69b45732011-05-28 08:25:51 -07002003 inode->i_flags |= S_NOSEC;
Jens Axboe01de85e2006-10-17 19:50:36 +02002004
Serge E. Hallynb5376772007-10-16 23:31:36 -07002005 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006}
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02002007EXPORT_SYMBOL(file_remove_suid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008
Nick Piggin2f718ff2007-10-16 01:24:59 -07002009static size_t __iovec_copy_from_user_inatomic(char *vaddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010 const struct iovec *iov, size_t base, size_t bytes)
2011{
Ingo Molnarf1800532009-03-02 11:00:57 +01002012 size_t copied = 0, left = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013
2014 while (bytes) {
2015 char __user *buf = iov->iov_base + base;
2016 int copy = min(bytes, iov->iov_len - base);
2017
2018 base = 0;
Ingo Molnarf1800532009-03-02 11:00:57 +01002019 left = __copy_from_user_inatomic(vaddr, buf, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 copied += copy;
2021 bytes -= copy;
2022 vaddr += copy;
2023 iov++;
2024
NeilBrown01408c42006-06-25 05:47:58 -07002025 if (unlikely(left))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 }
2028 return copied - left;
2029}
2030
2031/*
Nick Piggin2f718ff2007-10-16 01:24:59 -07002032 * Copy as much as we can into the page and return the number of bytes which
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02002033 * were successfully copied. If a fault is encountered then return the number of
Nick Piggin2f718ff2007-10-16 01:24:59 -07002034 * bytes which were copied.
2035 */
2036size_t iov_iter_copy_from_user_atomic(struct page *page,
2037 struct iov_iter *i, unsigned long offset, size_t bytes)
2038{
2039 char *kaddr;
2040 size_t copied;
2041
2042 BUG_ON(!in_atomic());
2043 kaddr = kmap_atomic(page, KM_USER0);
2044 if (likely(i->nr_segs == 1)) {
2045 int left;
2046 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01002047 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002048 copied = bytes - left;
2049 } else {
2050 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
2051 i->iov, i->iov_offset, bytes);
2052 }
2053 kunmap_atomic(kaddr, KM_USER0);
2054
2055 return copied;
2056}
Nick Piggin89e10782007-10-16 01:25:07 -07002057EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002058
2059/*
2060 * This has the same sideeffects and return value as
2061 * iov_iter_copy_from_user_atomic().
2062 * The difference is that it attempts to resolve faults.
2063 * Page must not be locked.
2064 */
2065size_t iov_iter_copy_from_user(struct page *page,
2066 struct iov_iter *i, unsigned long offset, size_t bytes)
2067{
2068 char *kaddr;
2069 size_t copied;
2070
2071 kaddr = kmap(page);
2072 if (likely(i->nr_segs == 1)) {
2073 int left;
2074 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01002075 left = __copy_from_user(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002076 copied = bytes - left;
2077 } else {
2078 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
2079 i->iov, i->iov_offset, bytes);
2080 }
2081 kunmap(page);
2082 return copied;
2083}
Nick Piggin89e10782007-10-16 01:25:07 -07002084EXPORT_SYMBOL(iov_iter_copy_from_user);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002085
Nick Pigginf7009262008-03-10 11:43:59 -07002086void iov_iter_advance(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07002087{
Nick Pigginf7009262008-03-10 11:43:59 -07002088 BUG_ON(i->count < bytes);
2089
Nick Piggin2f718ff2007-10-16 01:24:59 -07002090 if (likely(i->nr_segs == 1)) {
2091 i->iov_offset += bytes;
Nick Pigginf7009262008-03-10 11:43:59 -07002092 i->count -= bytes;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002093 } else {
2094 const struct iovec *iov = i->iov;
2095 size_t base = i->iov_offset;
2096
Nick Piggin124d3b72008-02-02 15:01:17 +01002097 /*
2098 * The !iov->iov_len check ensures we skip over unlikely
Nick Pigginf7009262008-03-10 11:43:59 -07002099 * zero-length segments (without overruning the iovec).
Nick Piggin124d3b72008-02-02 15:01:17 +01002100 */
Linus Torvalds94ad3742008-07-30 14:45:12 -07002101 while (bytes || unlikely(i->count && !iov->iov_len)) {
Nick Pigginf7009262008-03-10 11:43:59 -07002102 int copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002103
Nick Pigginf7009262008-03-10 11:43:59 -07002104 copy = min(bytes, iov->iov_len - base);
2105 BUG_ON(!i->count || i->count < copy);
2106 i->count -= copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002107 bytes -= copy;
2108 base += copy;
2109 if (iov->iov_len == base) {
2110 iov++;
2111 base = 0;
2112 }
2113 }
2114 i->iov = iov;
2115 i->iov_offset = base;
2116 }
2117}
Nick Piggin89e10782007-10-16 01:25:07 -07002118EXPORT_SYMBOL(iov_iter_advance);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002119
Nick Pigginafddba42007-10-16 01:25:01 -07002120/*
2121 * Fault in the first iovec of the given iov_iter, to a maximum length
2122 * of bytes. Returns 0 on success, or non-zero if the memory could not be
2123 * accessed (ie. because it is an invalid address).
2124 *
2125 * writev-intensive code may want this to prefault several iovecs -- that
2126 * would be possible (callers must not rely on the fact that _only_ the
2127 * first iovec will be faulted with the current implementation).
2128 */
2129int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07002130{
Nick Piggin2f718ff2007-10-16 01:24:59 -07002131 char __user *buf = i->iov->iov_base + i->iov_offset;
Nick Pigginafddba42007-10-16 01:25:01 -07002132 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
2133 return fault_in_pages_readable(buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002134}
Nick Piggin89e10782007-10-16 01:25:07 -07002135EXPORT_SYMBOL(iov_iter_fault_in_readable);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002136
2137/*
2138 * Return the count of just the current iov_iter segment.
2139 */
2140size_t iov_iter_single_seg_count(struct iov_iter *i)
2141{
2142 const struct iovec *iov = i->iov;
2143 if (i->nr_segs == 1)
2144 return i->count;
2145 else
2146 return min(i->count, iov->iov_len - i->iov_offset);
2147}
Nick Piggin89e10782007-10-16 01:25:07 -07002148EXPORT_SYMBOL(iov_iter_single_seg_count);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002149
2150/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 * Performs necessary checks before doing a write
2152 *
Randy Dunlap485bb992006-06-23 02:03:49 -07002153 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154 * Returns appropriate error code that caller should return or
2155 * zero in case that write should be allowed.
2156 */
2157inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2158{
2159 struct inode *inode = file->f_mapping->host;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002160 unsigned long limit = rlimit(RLIMIT_FSIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161
2162 if (unlikely(*pos < 0))
2163 return -EINVAL;
2164
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 if (!isblk) {
2166 /* FIXME: this is for backwards compatibility with 2.4 */
2167 if (file->f_flags & O_APPEND)
2168 *pos = i_size_read(inode);
2169
2170 if (limit != RLIM_INFINITY) {
2171 if (*pos >= limit) {
2172 send_sig(SIGXFSZ, current, 0);
2173 return -EFBIG;
2174 }
2175 if (*count > limit - (typeof(limit))*pos) {
2176 *count = limit - (typeof(limit))*pos;
2177 }
2178 }
2179 }
2180
2181 /*
2182 * LFS rule
2183 */
2184 if (unlikely(*pos + *count > MAX_NON_LFS &&
2185 !(file->f_flags & O_LARGEFILE))) {
2186 if (*pos >= MAX_NON_LFS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 return -EFBIG;
2188 }
2189 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2190 *count = MAX_NON_LFS - (unsigned long)*pos;
2191 }
2192 }
2193
2194 /*
2195 * Are we about to exceed the fs block limit ?
2196 *
2197 * If we have written data it becomes a short write. If we have
2198 * exceeded without writing data we send a signal and return EFBIG.
2199 * Linus frestrict idea will clean these up nicely..
2200 */
2201 if (likely(!isblk)) {
2202 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2203 if (*count || *pos > inode->i_sb->s_maxbytes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 return -EFBIG;
2205 }
2206 /* zero-length writes at ->s_maxbytes are OK */
2207 }
2208
2209 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2210 *count = inode->i_sb->s_maxbytes - *pos;
2211 } else {
David Howells93614012006-09-30 20:45:40 +02002212#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213 loff_t isize;
2214 if (bdev_read_only(I_BDEV(inode)))
2215 return -EPERM;
2216 isize = i_size_read(inode);
2217 if (*pos >= isize) {
2218 if (*count || *pos > isize)
2219 return -ENOSPC;
2220 }
2221
2222 if (*pos + *count > isize)
2223 *count = isize - *pos;
David Howells93614012006-09-30 20:45:40 +02002224#else
2225 return -EPERM;
2226#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227 }
2228 return 0;
2229}
2230EXPORT_SYMBOL(generic_write_checks);
2231
Nick Pigginafddba42007-10-16 01:25:01 -07002232int pagecache_write_begin(struct file *file, struct address_space *mapping,
2233 loff_t pos, unsigned len, unsigned flags,
2234 struct page **pagep, void **fsdata)
2235{
2236 const struct address_space_operations *aops = mapping->a_ops;
2237
Nick Piggin4e02ed42008-10-29 14:00:55 -07002238 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002239 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002240}
2241EXPORT_SYMBOL(pagecache_write_begin);
2242
2243int pagecache_write_end(struct file *file, struct address_space *mapping,
2244 loff_t pos, unsigned len, unsigned copied,
2245 struct page *page, void *fsdata)
2246{
2247 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002248
Nick Piggin4e02ed42008-10-29 14:00:55 -07002249 mark_page_accessed(page);
2250 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002251}
2252EXPORT_SYMBOL(pagecache_write_end);
2253
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254ssize_t
2255generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2256 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2257 size_t count, size_t ocount)
2258{
2259 struct file *file = iocb->ki_filp;
2260 struct address_space *mapping = file->f_mapping;
2261 struct inode *inode = mapping->host;
2262 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002263 size_t write_len;
2264 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265
2266 if (count != ocount)
2267 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2268
Christoph Hellwiga969e902008-07-23 21:27:04 -07002269 write_len = iov_length(iov, *nr_segs);
2270 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002271
Nick Piggin48b47c52009-01-06 14:40:22 -08002272 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002273 if (written)
2274 goto out;
2275
2276 /*
2277 * After a write we want buffered reads to be sure to go to disk to get
2278 * the new data. We invalidate clean cached page from the region we're
2279 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002280 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002281 */
2282 if (mapping->nrpages) {
2283 written = invalidate_inode_pages2_range(mapping,
2284 pos >> PAGE_CACHE_SHIFT, end);
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002285 /*
2286 * If a page can not be invalidated, return 0 to fall back
2287 * to buffered write.
2288 */
2289 if (written) {
2290 if (written == -EBUSY)
2291 return 0;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002292 goto out;
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002293 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002294 }
2295
2296 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2297
2298 /*
2299 * Finally, try again to invalidate clean pages which might have been
2300 * cached by non-direct readahead, or faulted in by get_user_pages()
2301 * if the source of the write was an mmap'ed region of the file
2302 * we're writing. Either one is a pretty crazy thing to do,
2303 * so we don't support it 100%. If this invalidation
2304 * fails, tough, the write still worked...
2305 */
2306 if (mapping->nrpages) {
2307 invalidate_inode_pages2_range(mapping,
2308 pos >> PAGE_CACHE_SHIFT, end);
2309 }
2310
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 if (written > 0) {
Namhyung Kim01166512010-10-26 14:21:58 -07002312 pos += written;
2313 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2314 i_size_write(inode, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 mark_inode_dirty(inode);
2316 }
Namhyung Kim01166512010-10-26 14:21:58 -07002317 *ppos = pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002319out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 return written;
2321}
2322EXPORT_SYMBOL(generic_file_direct_write);
2323
Nick Piggineb2be182007-10-16 01:24:57 -07002324/*
2325 * Find or create a page at the given pagecache position. Return the locked
2326 * page. This function is specifically for buffered writes.
2327 */
Nick Piggin54566b22009-01-04 12:00:53 -08002328struct page *grab_cache_page_write_begin(struct address_space *mapping,
2329 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07002330{
2331 int status;
2332 struct page *page;
Nick Piggin54566b22009-01-04 12:00:53 -08002333 gfp_t gfp_notmask = 0;
2334 if (flags & AOP_FLAG_NOFS)
2335 gfp_notmask = __GFP_FS;
Nick Piggineb2be182007-10-16 01:24:57 -07002336repeat:
2337 page = find_lock_page(mapping, index);
Steven Rostedtc585a262011-01-13 15:46:18 -08002338 if (page)
Darrick J. Wong3d08bcc2011-05-27 12:23:34 -07002339 goto found;
Nick Piggineb2be182007-10-16 01:24:57 -07002340
Nick Piggin54566b22009-01-04 12:00:53 -08002341 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002342 if (!page)
2343 return NULL;
Nick Piggin54566b22009-01-04 12:00:53 -08002344 status = add_to_page_cache_lru(page, mapping, index,
2345 GFP_KERNEL & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002346 if (unlikely(status)) {
2347 page_cache_release(page);
2348 if (status == -EEXIST)
2349 goto repeat;
2350 return NULL;
2351 }
Darrick J. Wong3d08bcc2011-05-27 12:23:34 -07002352found:
2353 wait_on_page_writeback(page);
Nick Piggineb2be182007-10-16 01:24:57 -07002354 return page;
2355}
Nick Piggin54566b22009-01-04 12:00:53 -08002356EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07002357
Nick Pigginafddba42007-10-16 01:25:01 -07002358static ssize_t generic_perform_write(struct file *file,
2359 struct iov_iter *i, loff_t pos)
2360{
2361 struct address_space *mapping = file->f_mapping;
2362 const struct address_space_operations *a_ops = mapping->a_ops;
2363 long status = 0;
2364 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002365 unsigned int flags = 0;
2366
2367 /*
2368 * Copies from kernel address space cannot fail (NFSD is a big user).
2369 */
2370 if (segment_eq(get_fs(), KERNEL_DS))
2371 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002372
2373 do {
2374 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002375 unsigned long offset; /* Offset into pagecache page */
2376 unsigned long bytes; /* Bytes to write to page */
2377 size_t copied; /* Bytes copied from user */
2378 void *fsdata;
2379
2380 offset = (pos & (PAGE_CACHE_SIZE - 1));
Nick Pigginafddba42007-10-16 01:25:01 -07002381 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2382 iov_iter_count(i));
2383
2384again:
2385
2386 /*
2387 * Bring in the user page that we will copy from _first_.
2388 * Otherwise there's a nasty deadlock on copying from the
2389 * same page as we're writing to, without it being marked
2390 * up-to-date.
2391 *
2392 * Not only is this an optimisation, but it is also required
2393 * to check that the address is actually valid, when atomic
2394 * usercopies are used, below.
2395 */
2396 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2397 status = -EFAULT;
2398 break;
2399 }
2400
Nick Piggin674b8922007-10-16 01:25:03 -07002401 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002402 &page, &fsdata);
2403 if (unlikely(status))
2404 break;
2405
anfei zhou931e80e2010-02-02 13:44:02 -08002406 if (mapping_writably_mapped(mapping))
2407 flush_dcache_page(page);
2408
Nick Pigginafddba42007-10-16 01:25:01 -07002409 pagefault_disable();
2410 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2411 pagefault_enable();
2412 flush_dcache_page(page);
2413
Josef Bacikc8236db2009-07-05 12:08:18 -07002414 mark_page_accessed(page);
Nick Pigginafddba42007-10-16 01:25:01 -07002415 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2416 page, fsdata);
2417 if (unlikely(status < 0))
2418 break;
2419 copied = status;
2420
2421 cond_resched();
2422
Nick Piggin124d3b72008-02-02 15:01:17 +01002423 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002424 if (unlikely(copied == 0)) {
2425 /*
2426 * If we were unable to copy any data at all, we must
2427 * fall back to a single segment length write.
2428 *
2429 * If we didn't fallback here, we could livelock
2430 * because not all segments in the iov can be copied at
2431 * once without a pagefault.
2432 */
2433 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2434 iov_iter_single_seg_count(i));
2435 goto again;
2436 }
Nick Pigginafddba42007-10-16 01:25:01 -07002437 pos += copied;
2438 written += copied;
2439
2440 balance_dirty_pages_ratelimited(mapping);
2441
2442 } while (iov_iter_count(i));
2443
2444 return written ? written : status;
2445}
2446
2447ssize_t
2448generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2449 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2450 size_t count, ssize_t written)
2451{
2452 struct file *file = iocb->ki_filp;
Nick Pigginafddba42007-10-16 01:25:01 -07002453 ssize_t status;
2454 struct iov_iter i;
2455
2456 iov_iter_init(&i, iov, nr_segs, count, written);
Nick Piggin4e02ed42008-10-29 14:00:55 -07002457 status = generic_perform_write(file, &i, pos);
Nick Pigginafddba42007-10-16 01:25:01 -07002458
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459 if (likely(status >= 0)) {
Nick Pigginafddba42007-10-16 01:25:01 -07002460 written += status;
2461 *ppos = pos + status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462 }
2463
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464 return written ? written : status;
2465}
2466EXPORT_SYMBOL(generic_file_buffered_write);
2467
Jan Karae4dd9de2009-08-17 18:10:06 +02002468/**
2469 * __generic_file_aio_write - write data to a file
2470 * @iocb: IO state structure (file, offset, etc.)
2471 * @iov: vector with data to write
2472 * @nr_segs: number of segments in the vector
2473 * @ppos: position where to write
2474 *
2475 * This function does all the work needed for actually writing data to a
2476 * file. It does all basic checks, removes SUID from the file, updates
2477 * modification times and calls proper subroutines depending on whether we
2478 * do direct IO or a standard buffered write.
2479 *
2480 * It expects i_mutex to be grabbed unless we work on a block device or similar
2481 * object which does not need locking at all.
2482 *
2483 * This function does *not* take care of syncing data in case of O_SYNC write.
2484 * A caller has to handle it. This is mainly due to the fact that we want to
2485 * avoid syncing under i_mutex.
2486 */
2487ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2488 unsigned long nr_segs, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489{
2490 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002491 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492 size_t ocount; /* original count */
2493 size_t count; /* after file limit checks */
2494 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495 loff_t pos;
2496 ssize_t written;
2497 ssize_t err;
2498
2499 ocount = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07002500 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2501 if (err)
2502 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503
2504 count = ocount;
2505 pos = *ppos;
2506
2507 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2508
2509 /* We can write back this queue in page reclaim */
2510 current->backing_dev_info = mapping->backing_dev_info;
2511 written = 0;
2512
2513 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2514 if (err)
2515 goto out;
2516
2517 if (count == 0)
2518 goto out;
2519
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02002520 err = file_remove_suid(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521 if (err)
2522 goto out;
2523
Christoph Hellwig870f4812006-01-09 20:52:01 -08002524 file_update_time(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525
2526 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2527 if (unlikely(file->f_flags & O_DIRECT)) {
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002528 loff_t endbyte;
2529 ssize_t written_buffered;
2530
2531 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2532 ppos, count, ocount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533 if (written < 0 || written == count)
2534 goto out;
2535 /*
2536 * direct-io write to a hole: fall through to buffered I/O
2537 * for completing the rest of the request.
2538 */
2539 pos += written;
2540 count -= written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002541 written_buffered = generic_file_buffered_write(iocb, iov,
2542 nr_segs, pos, ppos, count,
2543 written);
2544 /*
2545 * If generic_file_buffered_write() retuned a synchronous error
2546 * then we want to return the number of bytes which were
2547 * direct-written, or the error code if that was zero. Note
2548 * that this differs from normal direct-io semantics, which
2549 * will return -EFOO even if some bytes were written.
2550 */
2551 if (written_buffered < 0) {
2552 err = written_buffered;
2553 goto out;
2554 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002556 /*
2557 * We need to ensure that the page cache pages are written to
2558 * disk and invalidated to preserve the expected O_DIRECT
2559 * semantics.
2560 */
2561 endbyte = pos + written_buffered - written - 1;
Christoph Hellwigc05c4ed2009-09-23 15:07:30 +02002562 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002563 if (err == 0) {
2564 written = written_buffered;
2565 invalidate_mapping_pages(mapping,
2566 pos >> PAGE_CACHE_SHIFT,
2567 endbyte >> PAGE_CACHE_SHIFT);
2568 } else {
2569 /*
2570 * We don't know how much we wrote, so just return
2571 * the number of bytes which were direct-written
2572 */
2573 }
2574 } else {
2575 written = generic_file_buffered_write(iocb, iov, nr_segs,
2576 pos, ppos, count, written);
2577 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578out:
2579 current->backing_dev_info = NULL;
2580 return written ? written : err;
2581}
Jan Karae4dd9de2009-08-17 18:10:06 +02002582EXPORT_SYMBOL(__generic_file_aio_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583
Jan Karae4dd9de2009-08-17 18:10:06 +02002584/**
2585 * generic_file_aio_write - write data to a file
2586 * @iocb: IO state structure
2587 * @iov: vector with data to write
2588 * @nr_segs: number of segments in the vector
2589 * @pos: position in file where to write
2590 *
2591 * This is a wrapper around __generic_file_aio_write() to be used by most
2592 * filesystems. It takes care of syncing the file in case of O_SYNC file
2593 * and acquires i_mutex as needed.
2594 */
Badari Pulavarty027445c2006-09-30 23:28:46 -07002595ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2596 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597{
2598 struct file *file = iocb->ki_filp;
Jan Kara148f9482009-08-17 19:52:36 +02002599 struct inode *inode = file->f_mapping->host;
Jens Axboe55602dd2010-06-24 15:05:37 +02002600 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602
2603 BUG_ON(iocb->ki_pos != pos);
2604
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002605 mutex_lock(&inode->i_mutex);
Jens Axboe55602dd2010-06-24 15:05:37 +02002606 blk_start_plug(&plug);
Jan Karae4dd9de2009-08-17 18:10:06 +02002607 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002608 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609
Jan Kara148f9482009-08-17 19:52:36 +02002610 if (ret > 0 || ret == -EIOCBQUEUED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611 ssize_t err;
2612
Jan Kara148f9482009-08-17 19:52:36 +02002613 err = generic_write_sync(file, pos, ret);
Jan Karac7b50db2009-08-18 16:18:20 +02002614 if (err < 0 && ret > 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615 ret = err;
2616 }
Jens Axboe55602dd2010-06-24 15:05:37 +02002617 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618 return ret;
2619}
2620EXPORT_SYMBOL(generic_file_aio_write);
2621
David Howellscf9a2ae2006-08-29 19:05:54 +01002622/**
2623 * try_to_release_page() - release old fs-specific metadata on a page
2624 *
2625 * @page: the page which the kernel is trying to free
2626 * @gfp_mask: memory allocation flags (and I/O mode)
2627 *
2628 * The address_space is to try to release any data against the page
2629 * (presumably at page->private). If the release was successful, return `1'.
2630 * Otherwise return zero.
2631 *
David Howells266cf652009-04-03 16:42:36 +01002632 * This may also be called if PG_fscache is set on a page, indicating that the
2633 * page is known to the local caching routines.
2634 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002635 * The @gfp_mask argument specifies whether I/O may be performed to release
Mingming Cao3f31fdd2008-07-25 01:46:22 -07002636 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01002637 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002638 */
2639int try_to_release_page(struct page *page, gfp_t gfp_mask)
2640{
2641 struct address_space * const mapping = page->mapping;
2642
2643 BUG_ON(!PageLocked(page));
2644 if (PageWriteback(page))
2645 return 0;
2646
2647 if (mapping && mapping->a_ops->releasepage)
2648 return mapping->a_ops->releasepage(page, gfp_mask);
2649 return try_to_free_buffers(page);
2650}
2651
2652EXPORT_SYMBOL(try_to_release_page);