blob: 18aac7819cc97b4f2b9fe834600a30614d3e8b33 [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>
Minchan Kimb430e9d2013-07-03 15:01:24 -070023#include <linux/blkdev.h>
Christoph Hellwige2e40f22015-02-22 08:58:50 -080024#include <linux/uio.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_end_io = end_io;
Kent Overstreet6cf66b42014-12-22 12:48:42 +010037
38 bio_add_page(bio, page, PAGE_SIZE, 0);
39 BUG_ON(bio->bi_iter.bi_size != PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -070040 }
41 return bio;
42}
43
Christoph Hellwig4246a0b2015-07-20 15:29:37 +020044void end_swap_bio_write(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070045{
Linus Torvalds1da177e2005-04-16 15:20:36 -070046 struct page *page = bio->bi_io_vec[0].bv_page;
47
Christoph Hellwig4246a0b2015-07-20 15:29:37 +020048 if (bio->bi_error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070049 SetPageError(page);
Peter Zijlstra6ddab3b2006-09-25 23:31:26 -070050 /*
51 * We failed to write the page out to swap-space.
52 * Re-dirty the page in order to avoid it being reclaimed.
53 * Also print a dire warning that things will go BAD (tm)
54 * very quickly.
55 *
56 * Also clear PG_reclaim to avoid rotate_reclaimable_page()
57 */
58 set_page_dirty(page);
Joe Perches11705322016-03-17 14:19:50 -070059 pr_alert("Write-error on swap-device (%u:%u:%llu)\n",
60 imajor(bio->bi_bdev->bd_inode),
61 iminor(bio->bi_bdev->bd_inode),
62 (unsigned long long)bio->bi_iter.bi_sector);
Peter Zijlstra6ddab3b2006-09-25 23:31:26 -070063 ClearPageReclaim(page);
64 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 end_page_writeback(page);
66 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070067}
68
Minchan Kim3f2b1a02016-03-22 14:24:36 -070069static void swap_slot_free_notify(struct page *page)
70{
71 struct swap_info_struct *sis;
72 struct gendisk *disk;
73
74 /*
75 * There is no guarantee that the page is in swap cache - the software
76 * suspend code (at least) uses end_swap_bio_read() against a non-
77 * swapcache page. So we must check PG_swapcache before proceeding with
78 * this optimization.
79 */
80 if (unlikely(!PageSwapCache(page)))
81 return;
82
83 sis = page_swap_info(page);
84 if (!(sis->flags & SWP_BLKDEV))
85 return;
86
87 /*
88 * The swap subsystem performs lazy swap slot freeing,
89 * expecting that the page will be swapped out again.
90 * So we can avoid an unnecessary write if the page
91 * isn't redirtied.
92 * This is good for real swap storage because we can
93 * reduce unnecessary I/O and enhance wear-leveling
94 * if an SSD is used as the as swap device.
95 * But if in-memory swap device (eg zram) is used,
96 * this causes a duplicated copy between uncompressed
97 * data in VM-owned memory and compressed data in
98 * zram-owned memory. So let's free zram-owned memory
99 * and make the VM-owned decompressed page *dirty*,
100 * so the page should be swapped out somewhere again if
101 * we again wish to reclaim it.
102 */
103 disk = sis->bdev->bd_disk;
104 if (disk->fops->swap_slot_free_notify) {
105 swp_entry_t entry;
106 unsigned long offset;
107
108 entry.val = page_private(page);
109 offset = swp_offset(entry);
110
111 SetPageDirty(page);
112 disk->fops->swap_slot_free_notify(sis->bdev,
113 offset);
114 }
115}
116
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200117static void end_swap_bio_read(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 struct page *page = bio->bi_io_vec[0].bv_page;
120
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200121 if (bio->bi_error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 SetPageError(page);
123 ClearPageUptodate(page);
Joe Perches11705322016-03-17 14:19:50 -0700124 pr_alert("Read-error on swap-device (%u:%u:%llu)\n",
125 imajor(bio->bi_bdev->bd_inode),
126 iminor(bio->bi_bdev->bd_inode),
127 (unsigned long long)bio->bi_iter.bi_sector);
Minchan Kimb430e9d2013-07-03 15:01:24 -0700128 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 }
Minchan Kimb430e9d2013-07-03 15:01:24 -0700130
131 SetPageUptodate(page);
Minchan Kim3f2b1a02016-03-22 14:24:36 -0700132 swap_slot_free_notify(page);
Minchan Kimb430e9d2013-07-03 15:01:24 -0700133out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 unlock_page(page);
135 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136}
137
Mel Gormana509bc12012-07-31 16:44:57 -0700138int generic_swapfile_activate(struct swap_info_struct *sis,
139 struct file *swap_file,
140 sector_t *span)
141{
142 struct address_space *mapping = swap_file->f_mapping;
143 struct inode *inode = mapping->host;
144 unsigned blocks_per_page;
145 unsigned long page_no;
146 unsigned blkbits;
147 sector_t probe_block;
148 sector_t last_block;
149 sector_t lowest_block = -1;
150 sector_t highest_block = 0;
151 int nr_extents = 0;
152 int ret;
153
154 blkbits = inode->i_blkbits;
155 blocks_per_page = PAGE_SIZE >> blkbits;
156
157 /*
158 * Map all the blocks into the extent list. This code doesn't try
159 * to be very smart.
160 */
161 probe_block = 0;
162 page_no = 0;
163 last_block = i_size_read(inode) >> blkbits;
164 while ((probe_block + blocks_per_page) <= last_block &&
165 page_no < sis->max) {
166 unsigned block_in_page;
167 sector_t first_block;
168
169 first_block = bmap(inode, probe_block);
170 if (first_block == 0)
171 goto bad_bmap;
172
173 /*
174 * It must be PAGE_SIZE aligned on-disk
175 */
176 if (first_block & (blocks_per_page - 1)) {
177 probe_block++;
178 goto reprobe;
179 }
180
181 for (block_in_page = 1; block_in_page < blocks_per_page;
182 block_in_page++) {
183 sector_t block;
184
185 block = bmap(inode, probe_block + block_in_page);
186 if (block == 0)
187 goto bad_bmap;
188 if (block != first_block + block_in_page) {
189 /* Discontiguity */
190 probe_block++;
191 goto reprobe;
192 }
193 }
194
195 first_block >>= (PAGE_SHIFT - blkbits);
196 if (page_no) { /* exclude the header page */
197 if (first_block < lowest_block)
198 lowest_block = first_block;
199 if (first_block > highest_block)
200 highest_block = first_block;
201 }
202
203 /*
204 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
205 */
206 ret = add_swap_extent(sis, page_no, 1, first_block);
207 if (ret < 0)
208 goto out;
209 nr_extents += ret;
210 page_no++;
211 probe_block += blocks_per_page;
212reprobe:
213 continue;
214 }
215 ret = nr_extents;
216 *span = 1 + highest_block - lowest_block;
217 if (page_no == 0)
218 page_no = 1; /* force Empty message */
219 sis->max = page_no;
220 sis->pages = page_no - 1;
221 sis->highest_bit = page_no - 1;
222out:
223 return ret;
224bad_bmap:
Joe Perches11705322016-03-17 14:19:50 -0700225 pr_err("swapon: swapfile has holes\n");
Mel Gormana509bc12012-07-31 16:44:57 -0700226 ret = -EINVAL;
227 goto out;
228}
229
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230/*
231 * We may have stale swap cache pages in memory: notice
232 * them here and get rid of the unnecessary final write.
233 */
234int swap_writepage(struct page *page, struct writeback_control *wbc)
235{
Seth Jennings2f772e62013-04-29 15:08:34 -0700236 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800238 if (try_to_free_swap(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 unlock_page(page);
240 goto out;
241 }
Konrad Rzeszutek Wilk165c8ae2012-05-15 11:32:15 -0400242 if (frontswap_store(page) == 0) {
Dan Magenheimer38b5faf2012-04-09 17:08:06 -0600243 set_page_writeback(page);
244 unlock_page(page);
245 end_page_writeback(page);
246 goto out;
247 }
Seth Jennings1eec6702013-04-29 15:08:35 -0700248 ret = __swap_writepage(page, wbc, end_swap_bio_write);
Seth Jennings2f772e62013-04-29 15:08:34 -0700249out:
250 return ret;
251}
252
Matthew Wilcoxdd6bd0d2014-06-04 16:07:48 -0700253static sector_t swap_page_sector(struct page *page)
254{
255 return (sector_t)__page_file_index(page) << (PAGE_CACHE_SHIFT - 9);
256}
257
Seth Jennings1eec6702013-04-29 15:08:35 -0700258int __swap_writepage(struct page *page, struct writeback_control *wbc,
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200259 bio_end_io_t end_write_func)
Seth Jennings2f772e62013-04-29 15:08:34 -0700260{
261 struct bio *bio;
Matthew Wilcoxdd6bd0d2014-06-04 16:07:48 -0700262 int ret, rw = WRITE;
Seth Jennings2f772e62013-04-29 15:08:34 -0700263 struct swap_info_struct *sis = page_swap_info(page);
Mel Gorman62c230b2012-07-31 16:44:55 -0700264
265 if (sis->flags & SWP_FILE) {
266 struct kiocb kiocb;
267 struct file *swap_file = sis->swap_file;
268 struct address_space *mapping = swap_file->f_mapping;
Al Viro62a80672014-04-04 23:12:29 -0400269 struct bio_vec bv = {
270 .bv_page = page,
271 .bv_len = PAGE_SIZE,
272 .bv_offset = 0
Mel Gorman62c230b2012-07-31 16:44:55 -0700273 };
Al Viro05afcb72015-01-23 01:08:07 -0500274 struct iov_iter from;
Mel Gorman62c230b2012-07-31 16:44:55 -0700275
Al Viro05afcb72015-01-23 01:08:07 -0500276 iov_iter_bvec(&from, ITER_BVEC | WRITE, &bv, 1, PAGE_SIZE);
Mel Gorman62c230b2012-07-31 16:44:55 -0700277 init_sync_kiocb(&kiocb, swap_file);
278 kiocb.ki_pos = page_file_offset(page);
Mel Gorman62c230b2012-07-31 16:44:55 -0700279
Mel Gorman0cdc4442013-04-29 15:08:48 -0700280 set_page_writeback(page);
Mel Gorman62c230b2012-07-31 16:44:55 -0700281 unlock_page(page);
Omar Sandoval22c61862015-03-16 04:33:53 -0700282 ret = mapping->a_ops->direct_IO(&kiocb, &from, kiocb.ki_pos);
Mel Gorman62c230b2012-07-31 16:44:55 -0700283 if (ret == PAGE_SIZE) {
284 count_vm_event(PSWPOUT);
285 ret = 0;
Jerome Marchand2d30d312013-04-29 15:08:47 -0700286 } else {
Mel Gorman0cdc4442013-04-29 15:08:48 -0700287 /*
288 * In the case of swap-over-nfs, this can be a
289 * temporary failure if the system has limited
290 * memory for allocating transmit buffers.
291 * Mark the page dirty and avoid
292 * rotate_reclaimable_page but rate-limit the
293 * messages but do not flag PageError like
294 * the normal direct-to-bio case as it could
295 * be temporary.
296 */
Jerome Marchand2d30d312013-04-29 15:08:47 -0700297 set_page_dirty(page);
Mel Gorman0cdc4442013-04-29 15:08:48 -0700298 ClearPageReclaim(page);
Joe Perches11705322016-03-17 14:19:50 -0700299 pr_err_ratelimited("Write error on dio swapfile (%llu)\n",
300 page_file_offset(page));
Mel Gorman62c230b2012-07-31 16:44:55 -0700301 }
Mel Gorman0cdc4442013-04-29 15:08:48 -0700302 end_page_writeback(page);
Mel Gorman62c230b2012-07-31 16:44:55 -0700303 return ret;
304 }
305
Matthew Wilcoxdd6bd0d2014-06-04 16:07:48 -0700306 ret = bdev_write_page(sis->bdev, swap_page_sector(page), page, wbc);
307 if (!ret) {
308 count_vm_event(PSWPOUT);
309 return 0;
310 }
311
312 ret = 0;
Seth Jennings1eec6702013-04-29 15:08:35 -0700313 bio = get_swap_bio(GFP_NOIO, page, end_write_func);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 if (bio == NULL) {
315 set_page_dirty(page);
316 unlock_page(page);
317 ret = -ENOMEM;
318 goto out;
319 }
320 if (wbc->sync_mode == WB_SYNC_ALL)
Jens Axboe721a9602011-03-09 11:56:30 +0100321 rw |= REQ_SYNC;
Christoph Lameterf8891e52006-06-30 01:55:45 -0700322 count_vm_event(PSWPOUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 set_page_writeback(page);
324 unlock_page(page);
325 submit_bio(rw, bio);
326out:
327 return ret;
328}
329
Minchan Kimaca8bf32009-06-16 15:33:02 -0700330int swap_readpage(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331{
332 struct bio *bio;
333 int ret = 0;
Mel Gorman62c230b2012-07-31 16:44:55 -0700334 struct swap_info_struct *sis = page_swap_info(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335
Sasha Levin309381fea2014-01-23 15:52:54 -0800336 VM_BUG_ON_PAGE(!PageLocked(page), page);
337 VM_BUG_ON_PAGE(PageUptodate(page), page);
Konrad Rzeszutek Wilk165c8ae2012-05-15 11:32:15 -0400338 if (frontswap_load(page) == 0) {
Dan Magenheimer38b5faf2012-04-09 17:08:06 -0600339 SetPageUptodate(page);
340 unlock_page(page);
341 goto out;
342 }
Mel Gorman62c230b2012-07-31 16:44:55 -0700343
344 if (sis->flags & SWP_FILE) {
345 struct file *swap_file = sis->swap_file;
346 struct address_space *mapping = swap_file->f_mapping;
347
348 ret = mapping->a_ops->readpage(swap_file, page);
349 if (!ret)
350 count_vm_event(PSWPIN);
351 return ret;
352 }
353
Matthew Wilcoxdd6bd0d2014-06-04 16:07:48 -0700354 ret = bdev_read_page(sis->bdev, swap_page_sector(page), page);
355 if (!ret) {
Minchan Kim3f2b1a02016-03-22 14:24:36 -0700356 swap_slot_free_notify(page);
Matthew Wilcoxdd6bd0d2014-06-04 16:07:48 -0700357 count_vm_event(PSWPIN);
358 return 0;
359 }
360
361 ret = 0;
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -0800362 bio = get_swap_bio(GFP_KERNEL, page, end_swap_bio_read);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 if (bio == NULL) {
364 unlock_page(page);
365 ret = -ENOMEM;
366 goto out;
367 }
Christoph Lameterf8891e52006-06-30 01:55:45 -0700368 count_vm_event(PSWPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 submit_bio(READ, bio);
370out:
371 return ret;
372}
Mel Gorman62c230b2012-07-31 16:44:55 -0700373
374int swap_set_page_dirty(struct page *page)
375{
376 struct swap_info_struct *sis = page_swap_info(page);
377
378 if (sis->flags & SWP_FILE) {
379 struct address_space *mapping = sis->swap_file->f_mapping;
380 return mapping->a_ops->set_page_dirty(page);
381 } else {
382 return __set_page_dirty_no_writeback(page);
383 }
384}