blob: 955db8b0d4970aa38b0ada56ac5f221618ed74aa [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>
Minchan Kimb430e9d2013-07-03 15:01:24 -070024#include <linux/blkdev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <asm/pgtable.h>
26
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -080027static struct bio *get_swap_bio(gfp_t gfp_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -070028 struct page *page, bio_end_io_t end_io)
29{
30 struct bio *bio;
31
32 bio = bio_alloc(gfp_flags, 1);
33 if (bio) {
Kent Overstreet4f024f32013-10-11 15:44:27 -070034 bio->bi_iter.bi_sector = map_swap_page(page, &bio->bi_bdev);
35 bio->bi_iter.bi_sector <<= PAGE_SHIFT - 9;
Linus Torvalds1da177e2005-04-16 15:20:36 -070036 bio->bi_io_vec[0].bv_page = page;
37 bio->bi_io_vec[0].bv_len = PAGE_SIZE;
38 bio->bi_io_vec[0].bv_offset = 0;
39 bio->bi_vcnt = 1;
Kent Overstreet4f024f32013-10-11 15:44:27 -070040 bio->bi_iter.bi_size = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -070041 bio->bi_end_io = end_io;
42 }
43 return bio;
44}
45
Seth Jennings1eec6702013-04-29 15:08:35 -070046void end_swap_bio_write(struct bio *bio, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -070047{
48 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
49 struct page *page = bio->bi_io_vec[0].bv_page;
50
Peter Zijlstra6ddab3b2006-09-25 23:31:26 -070051 if (!uptodate) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070052 SetPageError(page);
Peter Zijlstra6ddab3b2006-09-25 23:31:26 -070053 /*
54 * We failed to write the page out to swap-space.
55 * Re-dirty the page in order to avoid it being reclaimed.
56 * Also print a dire warning that things will go BAD (tm)
57 * very quickly.
58 *
59 * Also clear PG_reclaim to avoid rotate_reclaimable_page()
60 */
61 set_page_dirty(page);
62 printk(KERN_ALERT "Write-error on swap-device (%u:%u:%Lu)\n",
63 imajor(bio->bi_bdev->bd_inode),
64 iminor(bio->bi_bdev->bd_inode),
Kent Overstreet4f024f32013-10-11 15:44:27 -070065 (unsigned long long)bio->bi_iter.bi_sector);
Peter Zijlstra6ddab3b2006-09-25 23:31:26 -070066 ClearPageReclaim(page);
67 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 end_page_writeback(page);
69 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070070}
71
NeilBrown6712ecf2007-09-27 12:47:43 +020072void end_swap_bio_read(struct bio *bio, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -070073{
74 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
75 struct page *page = bio->bi_io_vec[0].bv_page;
76
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 if (!uptodate) {
78 SetPageError(page);
79 ClearPageUptodate(page);
Peter Zijlstra6ddab3b2006-09-25 23:31:26 -070080 printk(KERN_ALERT "Read-error on swap-device (%u:%u:%Lu)\n",
81 imajor(bio->bi_bdev->bd_inode),
82 iminor(bio->bi_bdev->bd_inode),
Kent Overstreet4f024f32013-10-11 15:44:27 -070083 (unsigned long long)bio->bi_iter.bi_sector);
Minchan Kimb430e9d2013-07-03 15:01:24 -070084 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -070085 }
Minchan Kimb430e9d2013-07-03 15:01:24 -070086
87 SetPageUptodate(page);
88
89 /*
90 * There is no guarantee that the page is in swap cache - the software
91 * suspend code (at least) uses end_swap_bio_read() against a non-
92 * swapcache page. So we must check PG_swapcache before proceeding with
93 * this optimization.
94 */
95 if (likely(PageSwapCache(page))) {
96 struct swap_info_struct *sis;
97
98 sis = page_swap_info(page);
99 if (sis->flags & SWP_BLKDEV) {
100 /*
101 * The swap subsystem performs lazy swap slot freeing,
102 * expecting that the page will be swapped out again.
103 * So we can avoid an unnecessary write if the page
104 * isn't redirtied.
105 * This is good for real swap storage because we can
106 * reduce unnecessary I/O and enhance wear-leveling
107 * if an SSD is used as the as swap device.
108 * But if in-memory swap device (eg zram) is used,
109 * this causes a duplicated copy between uncompressed
110 * data in VM-owned memory and compressed data in
111 * zram-owned memory. So let's free zram-owned memory
112 * and make the VM-owned decompressed page *dirty*,
113 * so the page should be swapped out somewhere again if
114 * we again wish to reclaim it.
115 */
116 struct gendisk *disk = sis->bdev->bd_disk;
117 if (disk->fops->swap_slot_free_notify) {
118 swp_entry_t entry;
119 unsigned long offset;
120
121 entry.val = page_private(page);
122 offset = swp_offset(entry);
123
124 SetPageDirty(page);
125 disk->fops->swap_slot_free_notify(sis->bdev,
126 offset);
127 }
128 }
129 }
130
131out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 unlock_page(page);
133 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134}
135
Mel Gormana509bc12012-07-31 16:44:57 -0700136int generic_swapfile_activate(struct swap_info_struct *sis,
137 struct file *swap_file,
138 sector_t *span)
139{
140 struct address_space *mapping = swap_file->f_mapping;
141 struct inode *inode = mapping->host;
142 unsigned blocks_per_page;
143 unsigned long page_no;
144 unsigned blkbits;
145 sector_t probe_block;
146 sector_t last_block;
147 sector_t lowest_block = -1;
148 sector_t highest_block = 0;
149 int nr_extents = 0;
150 int ret;
151
152 blkbits = inode->i_blkbits;
153 blocks_per_page = PAGE_SIZE >> blkbits;
154
155 /*
156 * Map all the blocks into the extent list. This code doesn't try
157 * to be very smart.
158 */
159 probe_block = 0;
160 page_no = 0;
161 last_block = i_size_read(inode) >> blkbits;
162 while ((probe_block + blocks_per_page) <= last_block &&
163 page_no < sis->max) {
164 unsigned block_in_page;
165 sector_t first_block;
166
167 first_block = bmap(inode, probe_block);
168 if (first_block == 0)
169 goto bad_bmap;
170
171 /*
172 * It must be PAGE_SIZE aligned on-disk
173 */
174 if (first_block & (blocks_per_page - 1)) {
175 probe_block++;
176 goto reprobe;
177 }
178
179 for (block_in_page = 1; block_in_page < blocks_per_page;
180 block_in_page++) {
181 sector_t block;
182
183 block = bmap(inode, probe_block + block_in_page);
184 if (block == 0)
185 goto bad_bmap;
186 if (block != first_block + block_in_page) {
187 /* Discontiguity */
188 probe_block++;
189 goto reprobe;
190 }
191 }
192
193 first_block >>= (PAGE_SHIFT - blkbits);
194 if (page_no) { /* exclude the header page */
195 if (first_block < lowest_block)
196 lowest_block = first_block;
197 if (first_block > highest_block)
198 highest_block = first_block;
199 }
200
201 /*
202 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
203 */
204 ret = add_swap_extent(sis, page_no, 1, first_block);
205 if (ret < 0)
206 goto out;
207 nr_extents += ret;
208 page_no++;
209 probe_block += blocks_per_page;
210reprobe:
211 continue;
212 }
213 ret = nr_extents;
214 *span = 1 + highest_block - lowest_block;
215 if (page_no == 0)
216 page_no = 1; /* force Empty message */
217 sis->max = page_no;
218 sis->pages = page_no - 1;
219 sis->highest_bit = page_no - 1;
220out:
221 return ret;
222bad_bmap:
223 printk(KERN_ERR "swapon: swapfile has holes\n");
224 ret = -EINVAL;
225 goto out;
226}
227
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228/*
229 * We may have stale swap cache pages in memory: notice
230 * them here and get rid of the unnecessary final write.
231 */
232int swap_writepage(struct page *page, struct writeback_control *wbc)
233{
Seth Jennings2f772e62013-04-29 15:08:34 -0700234 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800236 if (try_to_free_swap(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 unlock_page(page);
238 goto out;
239 }
Konrad Rzeszutek Wilk165c8ae2012-05-15 11:32:15 -0400240 if (frontswap_store(page) == 0) {
Dan Magenheimer38b5faf2012-04-09 17:08:06 -0600241 set_page_writeback(page);
242 unlock_page(page);
243 end_page_writeback(page);
244 goto out;
245 }
Seth Jennings1eec6702013-04-29 15:08:35 -0700246 ret = __swap_writepage(page, wbc, end_swap_bio_write);
Seth Jennings2f772e62013-04-29 15:08:34 -0700247out:
248 return ret;
249}
250
Matthew Wilcoxdd6bd0d2014-06-04 16:07:48 -0700251static sector_t swap_page_sector(struct page *page)
252{
253 return (sector_t)__page_file_index(page) << (PAGE_CACHE_SHIFT - 9);
254}
255
Seth Jennings1eec6702013-04-29 15:08:35 -0700256int __swap_writepage(struct page *page, struct writeback_control *wbc,
257 void (*end_write_func)(struct bio *, int))
Seth Jennings2f772e62013-04-29 15:08:34 -0700258{
259 struct bio *bio;
Matthew Wilcoxdd6bd0d2014-06-04 16:07:48 -0700260 int ret, rw = WRITE;
Seth Jennings2f772e62013-04-29 15:08:34 -0700261 struct swap_info_struct *sis = page_swap_info(page);
Mel Gorman62c230b2012-07-31 16:44:55 -0700262
263 if (sis->flags & SWP_FILE) {
264 struct kiocb kiocb;
265 struct file *swap_file = sis->swap_file;
266 struct address_space *mapping = swap_file->f_mapping;
Al Viro62a80672014-04-04 23:12:29 -0400267 struct bio_vec bv = {
268 .bv_page = page,
269 .bv_len = PAGE_SIZE,
270 .bv_offset = 0
Mel Gorman62c230b2012-07-31 16:44:55 -0700271 };
Al Viro62a80672014-04-04 23:12:29 -0400272 struct iov_iter from = {
273 .type = ITER_BVEC | WRITE,
274 .count = PAGE_SIZE,
275 .iov_offset = 0,
276 .nr_segs = 1,
Mel Gorman62c230b2012-07-31 16:44:55 -0700277 };
Al Viro05064082014-06-14 07:12:41 +0100278 from.bvec = &bv; /* older gcc versions are broken */
Mel Gorman62c230b2012-07-31 16:44:55 -0700279
280 init_sync_kiocb(&kiocb, swap_file);
281 kiocb.ki_pos = page_file_offset(page);
Mel Gorman62c230b2012-07-31 16:44:55 -0700282 kiocb.ki_nbytes = PAGE_SIZE;
283
Mel Gorman0cdc4442013-04-29 15:08:48 -0700284 set_page_writeback(page);
Mel Gorman62c230b2012-07-31 16:44:55 -0700285 unlock_page(page);
Al Viro62a80672014-04-04 23:12:29 -0400286 ret = mapping->a_ops->direct_IO(ITER_BVEC | WRITE,
Al Virod8d3d942014-03-04 21:27:34 -0500287 &kiocb, &from,
288 kiocb.ki_pos);
Mel Gorman62c230b2012-07-31 16:44:55 -0700289 if (ret == PAGE_SIZE) {
290 count_vm_event(PSWPOUT);
291 ret = 0;
Jerome Marchand2d30d312013-04-29 15:08:47 -0700292 } else {
Mel Gorman0cdc4442013-04-29 15:08:48 -0700293 /*
294 * In the case of swap-over-nfs, this can be a
295 * temporary failure if the system has limited
296 * memory for allocating transmit buffers.
297 * Mark the page dirty and avoid
298 * rotate_reclaimable_page but rate-limit the
299 * messages but do not flag PageError like
300 * the normal direct-to-bio case as it could
301 * be temporary.
302 */
Jerome Marchand2d30d312013-04-29 15:08:47 -0700303 set_page_dirty(page);
Mel Gorman0cdc4442013-04-29 15:08:48 -0700304 ClearPageReclaim(page);
305 pr_err_ratelimited("Write error on dio swapfile (%Lu)\n",
306 page_file_offset(page));
Mel Gorman62c230b2012-07-31 16:44:55 -0700307 }
Mel Gorman0cdc4442013-04-29 15:08:48 -0700308 end_page_writeback(page);
Mel Gorman62c230b2012-07-31 16:44:55 -0700309 return ret;
310 }
311
Matthew Wilcoxdd6bd0d2014-06-04 16:07:48 -0700312 ret = bdev_write_page(sis->bdev, swap_page_sector(page), page, wbc);
313 if (!ret) {
314 count_vm_event(PSWPOUT);
315 return 0;
316 }
317
318 ret = 0;
Seth Jennings1eec6702013-04-29 15:08:35 -0700319 bio = get_swap_bio(GFP_NOIO, page, end_write_func);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 if (bio == NULL) {
321 set_page_dirty(page);
322 unlock_page(page);
323 ret = -ENOMEM;
324 goto out;
325 }
326 if (wbc->sync_mode == WB_SYNC_ALL)
Jens Axboe721a9602011-03-09 11:56:30 +0100327 rw |= REQ_SYNC;
Christoph Lameterf8891e52006-06-30 01:55:45 -0700328 count_vm_event(PSWPOUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 set_page_writeback(page);
330 unlock_page(page);
331 submit_bio(rw, bio);
332out:
333 return ret;
334}
335
Minchan Kimaca8bf32009-06-16 15:33:02 -0700336int swap_readpage(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337{
338 struct bio *bio;
339 int ret = 0;
Mel Gorman62c230b2012-07-31 16:44:55 -0700340 struct swap_info_struct *sis = page_swap_info(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341
Sasha Levin309381fea2014-01-23 15:52:54 -0800342 VM_BUG_ON_PAGE(!PageLocked(page), page);
343 VM_BUG_ON_PAGE(PageUptodate(page), page);
Konrad Rzeszutek Wilk165c8ae2012-05-15 11:32:15 -0400344 if (frontswap_load(page) == 0) {
Dan Magenheimer38b5faf2012-04-09 17:08:06 -0600345 SetPageUptodate(page);
346 unlock_page(page);
347 goto out;
348 }
Mel Gorman62c230b2012-07-31 16:44:55 -0700349
350 if (sis->flags & SWP_FILE) {
351 struct file *swap_file = sis->swap_file;
352 struct address_space *mapping = swap_file->f_mapping;
353
354 ret = mapping->a_ops->readpage(swap_file, page);
355 if (!ret)
356 count_vm_event(PSWPIN);
357 return ret;
358 }
359
Matthew Wilcoxdd6bd0d2014-06-04 16:07:48 -0700360 ret = bdev_read_page(sis->bdev, swap_page_sector(page), page);
361 if (!ret) {
362 count_vm_event(PSWPIN);
363 return 0;
364 }
365
366 ret = 0;
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -0800367 bio = get_swap_bio(GFP_KERNEL, page, end_swap_bio_read);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 if (bio == NULL) {
369 unlock_page(page);
370 ret = -ENOMEM;
371 goto out;
372 }
Christoph Lameterf8891e52006-06-30 01:55:45 -0700373 count_vm_event(PSWPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 submit_bio(READ, bio);
375out:
376 return ret;
377}
Mel Gorman62c230b2012-07-31 16:44:55 -0700378
379int swap_set_page_dirty(struct page *page)
380{
381 struct swap_info_struct *sis = page_swap_info(page);
382
383 if (sis->flags & SWP_FILE) {
384 struct address_space *mapping = sis->swap_file->f_mapping;
385 return mapping->a_ops->set_page_dirty(page);
386 } else {
387 return __set_page_dirty_no_writeback(page);
388 }
389}