blob: a8a3ef45fed753b68ac1cc4a94c9260979a37879 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/page_io.c
3 *
4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 *
6 * Swap reorganised 29.12.95,
7 * Asynchronous swapping added 30.12.95. Stephen Tweedie
8 * Removed race in async swapping. 14.4.1996. Bruno Haible
9 * Add swap of shared pages through the page cache. 20.2.1998. Stephen Tweedie
10 * Always use brw_page, life becomes simpler. 12 May 1998 Eric Biederman
11 */
12
13#include <linux/mm.h>
14#include <linux/kernel_stat.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090015#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/pagemap.h>
17#include <linux/swap.h>
18#include <linux/bio.h>
19#include <linux/swapops.h>
Mel Gorman62c230b2012-07-31 16:44:55 -070020#include <linux/buffer_head.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/writeback.h>
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060022#include <linux/frontswap.h>
Kent Overstreeta27bb332013-05-07 16:19:08 -070023#include <linux/aio.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <asm/pgtable.h>
25
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -080026static struct bio *get_swap_bio(gfp_t gfp_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -070027 struct page *page, bio_end_io_t end_io)
28{
29 struct bio *bio;
30
31 bio = bio_alloc(gfp_flags, 1);
32 if (bio) {
Lee Schermerhornd4906e12009-12-14 17:58:49 -080033 bio->bi_sector = map_swap_page(page, &bio->bi_bdev);
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -080034 bio->bi_sector <<= PAGE_SHIFT - 9;
Linus Torvalds1da177e2005-04-16 15:20:36 -070035 bio->bi_io_vec[0].bv_page = page;
36 bio->bi_io_vec[0].bv_len = PAGE_SIZE;
37 bio->bi_io_vec[0].bv_offset = 0;
38 bio->bi_vcnt = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070039 bio->bi_size = PAGE_SIZE;
40 bio->bi_end_io = end_io;
41 }
42 return bio;
43}
44
Seth Jennings1eec6702013-04-29 15:08:35 -070045void end_swap_bio_write(struct bio *bio, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -070046{
47 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
48 struct page *page = bio->bi_io_vec[0].bv_page;
49
Peter Zijlstra6ddab3b2006-09-25 23:31:26 -070050 if (!uptodate) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070051 SetPageError(page);
Peter Zijlstra6ddab3b2006-09-25 23:31:26 -070052 /*
53 * We failed to write the page out to swap-space.
54 * Re-dirty the page in order to avoid it being reclaimed.
55 * Also print a dire warning that things will go BAD (tm)
56 * very quickly.
57 *
58 * Also clear PG_reclaim to avoid rotate_reclaimable_page()
59 */
60 set_page_dirty(page);
61 printk(KERN_ALERT "Write-error on swap-device (%u:%u:%Lu)\n",
62 imajor(bio->bi_bdev->bd_inode),
63 iminor(bio->bi_bdev->bd_inode),
64 (unsigned long long)bio->bi_sector);
65 ClearPageReclaim(page);
66 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 end_page_writeback(page);
68 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070069}
70
NeilBrown6712ecf2007-09-27 12:47:43 +020071void end_swap_bio_read(struct bio *bio, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -070072{
73 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
74 struct page *page = bio->bi_io_vec[0].bv_page;
75
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 if (!uptodate) {
77 SetPageError(page);
78 ClearPageUptodate(page);
Peter Zijlstra6ddab3b2006-09-25 23:31:26 -070079 printk(KERN_ALERT "Read-error on swap-device (%u:%u:%Lu)\n",
80 imajor(bio->bi_bdev->bd_inode),
81 iminor(bio->bi_bdev->bd_inode),
82 (unsigned long long)bio->bi_sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 } else {
84 SetPageUptodate(page);
85 }
86 unlock_page(page);
87 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088}
89
Mel Gormana509bc12012-07-31 16:44:57 -070090int generic_swapfile_activate(struct swap_info_struct *sis,
91 struct file *swap_file,
92 sector_t *span)
93{
94 struct address_space *mapping = swap_file->f_mapping;
95 struct inode *inode = mapping->host;
96 unsigned blocks_per_page;
97 unsigned long page_no;
98 unsigned blkbits;
99 sector_t probe_block;
100 sector_t last_block;
101 sector_t lowest_block = -1;
102 sector_t highest_block = 0;
103 int nr_extents = 0;
104 int ret;
105
106 blkbits = inode->i_blkbits;
107 blocks_per_page = PAGE_SIZE >> blkbits;
108
109 /*
110 * Map all the blocks into the extent list. This code doesn't try
111 * to be very smart.
112 */
113 probe_block = 0;
114 page_no = 0;
115 last_block = i_size_read(inode) >> blkbits;
116 while ((probe_block + blocks_per_page) <= last_block &&
117 page_no < sis->max) {
118 unsigned block_in_page;
119 sector_t first_block;
120
121 first_block = bmap(inode, probe_block);
122 if (first_block == 0)
123 goto bad_bmap;
124
125 /*
126 * It must be PAGE_SIZE aligned on-disk
127 */
128 if (first_block & (blocks_per_page - 1)) {
129 probe_block++;
130 goto reprobe;
131 }
132
133 for (block_in_page = 1; block_in_page < blocks_per_page;
134 block_in_page++) {
135 sector_t block;
136
137 block = bmap(inode, probe_block + block_in_page);
138 if (block == 0)
139 goto bad_bmap;
140 if (block != first_block + block_in_page) {
141 /* Discontiguity */
142 probe_block++;
143 goto reprobe;
144 }
145 }
146
147 first_block >>= (PAGE_SHIFT - blkbits);
148 if (page_no) { /* exclude the header page */
149 if (first_block < lowest_block)
150 lowest_block = first_block;
151 if (first_block > highest_block)
152 highest_block = first_block;
153 }
154
155 /*
156 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
157 */
158 ret = add_swap_extent(sis, page_no, 1, first_block);
159 if (ret < 0)
160 goto out;
161 nr_extents += ret;
162 page_no++;
163 probe_block += blocks_per_page;
164reprobe:
165 continue;
166 }
167 ret = nr_extents;
168 *span = 1 + highest_block - lowest_block;
169 if (page_no == 0)
170 page_no = 1; /* force Empty message */
171 sis->max = page_no;
172 sis->pages = page_no - 1;
173 sis->highest_bit = page_no - 1;
174out:
175 return ret;
176bad_bmap:
177 printk(KERN_ERR "swapon: swapfile has holes\n");
178 ret = -EINVAL;
179 goto out;
180}
181
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182/*
183 * We may have stale swap cache pages in memory: notice
184 * them here and get rid of the unnecessary final write.
185 */
186int swap_writepage(struct page *page, struct writeback_control *wbc)
187{
Seth Jennings2f772e62013-04-29 15:08:34 -0700188 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800190 if (try_to_free_swap(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 unlock_page(page);
192 goto out;
193 }
Konrad Rzeszutek Wilk165c8ae2012-05-15 11:32:15 -0400194 if (frontswap_store(page) == 0) {
Dan Magenheimer38b5faf2012-04-09 17:08:06 -0600195 set_page_writeback(page);
196 unlock_page(page);
197 end_page_writeback(page);
198 goto out;
199 }
Seth Jennings1eec6702013-04-29 15:08:35 -0700200 ret = __swap_writepage(page, wbc, end_swap_bio_write);
Seth Jennings2f772e62013-04-29 15:08:34 -0700201out:
202 return ret;
203}
204
Seth Jennings1eec6702013-04-29 15:08:35 -0700205int __swap_writepage(struct page *page, struct writeback_control *wbc,
206 void (*end_write_func)(struct bio *, int))
Seth Jennings2f772e62013-04-29 15:08:34 -0700207{
208 struct bio *bio;
209 int ret = 0, rw = WRITE;
210 struct swap_info_struct *sis = page_swap_info(page);
Mel Gorman62c230b2012-07-31 16:44:55 -0700211
212 if (sis->flags & SWP_FILE) {
213 struct kiocb kiocb;
214 struct file *swap_file = sis->swap_file;
215 struct address_space *mapping = swap_file->f_mapping;
216 struct iovec iov = {
Mel Gorman5a178112012-07-31 16:45:02 -0700217 .iov_base = kmap(page),
Mel Gorman62c230b2012-07-31 16:44:55 -0700218 .iov_len = PAGE_SIZE,
219 };
220
221 init_sync_kiocb(&kiocb, swap_file);
222 kiocb.ki_pos = page_file_offset(page);
223 kiocb.ki_left = PAGE_SIZE;
224 kiocb.ki_nbytes = PAGE_SIZE;
225
Mel Gorman0cdc4442013-04-29 15:08:48 -0700226 set_page_writeback(page);
Mel Gorman62c230b2012-07-31 16:44:55 -0700227 unlock_page(page);
228 ret = mapping->a_ops->direct_IO(KERNEL_WRITE,
229 &kiocb, &iov,
230 kiocb.ki_pos, 1);
Mel Gorman5a178112012-07-31 16:45:02 -0700231 kunmap(page);
Mel Gorman62c230b2012-07-31 16:44:55 -0700232 if (ret == PAGE_SIZE) {
233 count_vm_event(PSWPOUT);
234 ret = 0;
Jerome Marchand2d30d312013-04-29 15:08:47 -0700235 } else {
Mel Gorman0cdc4442013-04-29 15:08:48 -0700236 /*
237 * In the case of swap-over-nfs, this can be a
238 * temporary failure if the system has limited
239 * memory for allocating transmit buffers.
240 * Mark the page dirty and avoid
241 * rotate_reclaimable_page but rate-limit the
242 * messages but do not flag PageError like
243 * the normal direct-to-bio case as it could
244 * be temporary.
245 */
Jerome Marchand2d30d312013-04-29 15:08:47 -0700246 set_page_dirty(page);
Mel Gorman0cdc4442013-04-29 15:08:48 -0700247 ClearPageReclaim(page);
248 pr_err_ratelimited("Write error on dio swapfile (%Lu)\n",
249 page_file_offset(page));
Mel Gorman62c230b2012-07-31 16:44:55 -0700250 }
Mel Gorman0cdc4442013-04-29 15:08:48 -0700251 end_page_writeback(page);
Mel Gorman62c230b2012-07-31 16:44:55 -0700252 return ret;
253 }
254
Seth Jennings1eec6702013-04-29 15:08:35 -0700255 bio = get_swap_bio(GFP_NOIO, page, end_write_func);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 if (bio == NULL) {
257 set_page_dirty(page);
258 unlock_page(page);
259 ret = -ENOMEM;
260 goto out;
261 }
262 if (wbc->sync_mode == WB_SYNC_ALL)
Jens Axboe721a9602011-03-09 11:56:30 +0100263 rw |= REQ_SYNC;
Christoph Lameterf8891e52006-06-30 01:55:45 -0700264 count_vm_event(PSWPOUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 set_page_writeback(page);
266 unlock_page(page);
267 submit_bio(rw, bio);
268out:
269 return ret;
270}
271
Minchan Kimaca8bf32009-06-16 15:33:02 -0700272int swap_readpage(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273{
274 struct bio *bio;
275 int ret = 0;
Mel Gorman62c230b2012-07-31 16:44:55 -0700276 struct swap_info_struct *sis = page_swap_info(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277
Hugh Dickins51726b12009-01-06 14:39:25 -0800278 VM_BUG_ON(!PageLocked(page));
279 VM_BUG_ON(PageUptodate(page));
Konrad Rzeszutek Wilk165c8ae2012-05-15 11:32:15 -0400280 if (frontswap_load(page) == 0) {
Dan Magenheimer38b5faf2012-04-09 17:08:06 -0600281 SetPageUptodate(page);
282 unlock_page(page);
283 goto out;
284 }
Mel Gorman62c230b2012-07-31 16:44:55 -0700285
286 if (sis->flags & SWP_FILE) {
287 struct file *swap_file = sis->swap_file;
288 struct address_space *mapping = swap_file->f_mapping;
289
290 ret = mapping->a_ops->readpage(swap_file, page);
291 if (!ret)
292 count_vm_event(PSWPIN);
293 return ret;
294 }
295
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -0800296 bio = get_swap_bio(GFP_KERNEL, page, end_swap_bio_read);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 if (bio == NULL) {
298 unlock_page(page);
299 ret = -ENOMEM;
300 goto out;
301 }
Christoph Lameterf8891e52006-06-30 01:55:45 -0700302 count_vm_event(PSWPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 submit_bio(READ, bio);
304out:
305 return ret;
306}
Mel Gorman62c230b2012-07-31 16:44:55 -0700307
308int swap_set_page_dirty(struct page *page)
309{
310 struct swap_info_struct *sis = page_swap_info(page);
311
312 if (sis->flags & SWP_FILE) {
313 struct address_space *mapping = sis->swap_file->f_mapping;
314 return mapping->a_ops->set_page_dirty(page);
315 } else {
316 return __set_page_dirty_no_writeback(page);
317 }
318}