blob: efe6fd67cb0ee53b80ee557dc488d2fba22dd128 [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);
Vinayak Menon5450ff12018-03-23 11:25:43 +053059 pr_alert_ratelimited("Write-error on swap-device (%u:%u:%llu)\n",
Joe Perches11705322016-03-17 14:19:50 -070060 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
Mikulas Patocka7e4411b2016-07-28 15:48:47 -0700169 cond_resched();
170
Mel Gormana509bc12012-07-31 16:44:57 -0700171 first_block = bmap(inode, probe_block);
172 if (first_block == 0)
173 goto bad_bmap;
174
175 /*
176 * It must be PAGE_SIZE aligned on-disk
177 */
178 if (first_block & (blocks_per_page - 1)) {
179 probe_block++;
180 goto reprobe;
181 }
182
183 for (block_in_page = 1; block_in_page < blocks_per_page;
184 block_in_page++) {
185 sector_t block;
186
187 block = bmap(inode, probe_block + block_in_page);
188 if (block == 0)
189 goto bad_bmap;
190 if (block != first_block + block_in_page) {
191 /* Discontiguity */
192 probe_block++;
193 goto reprobe;
194 }
195 }
196
197 first_block >>= (PAGE_SHIFT - blkbits);
198 if (page_no) { /* exclude the header page */
199 if (first_block < lowest_block)
200 lowest_block = first_block;
201 if (first_block > highest_block)
202 highest_block = first_block;
203 }
204
205 /*
206 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
207 */
208 ret = add_swap_extent(sis, page_no, 1, first_block);
209 if (ret < 0)
210 goto out;
211 nr_extents += ret;
212 page_no++;
213 probe_block += blocks_per_page;
214reprobe:
215 continue;
216 }
217 ret = nr_extents;
218 *span = 1 + highest_block - lowest_block;
219 if (page_no == 0)
220 page_no = 1; /* force Empty message */
221 sis->max = page_no;
222 sis->pages = page_no - 1;
223 sis->highest_bit = page_no - 1;
224out:
225 return ret;
226bad_bmap:
Joe Perches11705322016-03-17 14:19:50 -0700227 pr_err("swapon: swapfile has holes\n");
Mel Gormana509bc12012-07-31 16:44:57 -0700228 ret = -EINVAL;
229 goto out;
230}
231
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232/*
233 * We may have stale swap cache pages in memory: notice
234 * them here and get rid of the unnecessary final write.
235 */
236int swap_writepage(struct page *page, struct writeback_control *wbc)
237{
Seth Jennings2f772e62013-04-29 15:08:34 -0700238 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800240 if (try_to_free_swap(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 unlock_page(page);
242 goto out;
243 }
Konrad Rzeszutek Wilk165c8ae2012-05-15 11:32:15 -0400244 if (frontswap_store(page) == 0) {
Dan Magenheimer38b5faf2012-04-09 17:08:06 -0600245 set_page_writeback(page);
246 unlock_page(page);
247 end_page_writeback(page);
248 goto out;
249 }
Seth Jennings1eec6702013-04-29 15:08:35 -0700250 ret = __swap_writepage(page, wbc, end_swap_bio_write);
Seth Jennings2f772e62013-04-29 15:08:34 -0700251out:
252 return ret;
253}
254
Matthew Wilcoxdd6bd0d2014-06-04 16:07:48 -0700255static sector_t swap_page_sector(struct page *page)
256{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300257 return (sector_t)__page_file_index(page) << (PAGE_SHIFT - 9);
Matthew Wilcoxdd6bd0d2014-06-04 16:07:48 -0700258}
259
Seth Jennings1eec6702013-04-29 15:08:35 -0700260int __swap_writepage(struct page *page, struct writeback_control *wbc,
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200261 bio_end_io_t end_write_func)
Seth Jennings2f772e62013-04-29 15:08:34 -0700262{
263 struct bio *bio;
Mike Christie4e49ea42016-06-05 14:31:41 -0500264 int ret;
Seth Jennings2f772e62013-04-29 15:08:34 -0700265 struct swap_info_struct *sis = page_swap_info(page);
Mel Gorman62c230b2012-07-31 16:44:55 -0700266
Andrew Mortoncc30c5d2016-10-07 17:00:52 -0700267 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
Mel Gorman62c230b2012-07-31 16:44:55 -0700268 if (sis->flags & SWP_FILE) {
269 struct kiocb kiocb;
270 struct file *swap_file = sis->swap_file;
271 struct address_space *mapping = swap_file->f_mapping;
Al Viro62a80672014-04-04 23:12:29 -0400272 struct bio_vec bv = {
273 .bv_page = page,
274 .bv_len = PAGE_SIZE,
275 .bv_offset = 0
Mel Gorman62c230b2012-07-31 16:44:55 -0700276 };
Al Viro05afcb72015-01-23 01:08:07 -0500277 struct iov_iter from;
Mel Gorman62c230b2012-07-31 16:44:55 -0700278
Al Viro05afcb72015-01-23 01:08:07 -0500279 iov_iter_bvec(&from, ITER_BVEC | WRITE, &bv, 1, PAGE_SIZE);
Mel Gorman62c230b2012-07-31 16:44:55 -0700280 init_sync_kiocb(&kiocb, swap_file);
281 kiocb.ki_pos = page_file_offset(page);
Mel Gorman62c230b2012-07-31 16:44:55 -0700282
Mel Gorman0cdc4442013-04-29 15:08:48 -0700283 set_page_writeback(page);
Mel Gorman62c230b2012-07-31 16:44:55 -0700284 unlock_page(page);
Christoph Hellwigc8b8e322016-04-07 08:51:58 -0700285 ret = mapping->a_ops->direct_IO(&kiocb, &from);
Mel Gorman62c230b2012-07-31 16:44:55 -0700286 if (ret == PAGE_SIZE) {
287 count_vm_event(PSWPOUT);
288 ret = 0;
Jerome Marchand2d30d312013-04-29 15:08:47 -0700289 } else {
Mel Gorman0cdc4442013-04-29 15:08:48 -0700290 /*
291 * In the case of swap-over-nfs, this can be a
292 * temporary failure if the system has limited
293 * memory for allocating transmit buffers.
294 * Mark the page dirty and avoid
295 * rotate_reclaimable_page but rate-limit the
296 * messages but do not flag PageError like
297 * the normal direct-to-bio case as it could
298 * be temporary.
299 */
Jerome Marchand2d30d312013-04-29 15:08:47 -0700300 set_page_dirty(page);
Mel Gorman0cdc4442013-04-29 15:08:48 -0700301 ClearPageReclaim(page);
Joe Perches11705322016-03-17 14:19:50 -0700302 pr_err_ratelimited("Write error on dio swapfile (%llu)\n",
303 page_file_offset(page));
Mel Gorman62c230b2012-07-31 16:44:55 -0700304 }
Mel Gorman0cdc4442013-04-29 15:08:48 -0700305 end_page_writeback(page);
Mel Gorman62c230b2012-07-31 16:44:55 -0700306 return ret;
307 }
308
Matthew Wilcoxdd6bd0d2014-06-04 16:07:48 -0700309 ret = bdev_write_page(sis->bdev, swap_page_sector(page), page, wbc);
310 if (!ret) {
311 count_vm_event(PSWPOUT);
312 return 0;
313 }
314
315 ret = 0;
Seth Jennings1eec6702013-04-29 15:08:35 -0700316 bio = get_swap_bio(GFP_NOIO, page, end_write_func);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 if (bio == NULL) {
318 set_page_dirty(page);
319 unlock_page(page);
320 ret = -ENOMEM;
321 goto out;
322 }
323 if (wbc->sync_mode == WB_SYNC_ALL)
Jens Axboeba13e832016-08-01 09:38:44 -0600324 bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC);
325 else
326 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700327 count_vm_event(PSWPOUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 set_page_writeback(page);
329 unlock_page(page);
Mike Christie4e49ea42016-06-05 14:31:41 -0500330 submit_bio(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331out:
332 return ret;
333}
334
Minchan Kimaca8bf32009-06-16 15:33:02 -0700335int swap_readpage(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336{
337 struct bio *bio;
338 int ret = 0;
Mel Gorman62c230b2012-07-31 16:44:55 -0700339 struct swap_info_struct *sis = page_swap_info(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340
Andrew Mortoncc30c5d2016-10-07 17:00:52 -0700341 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
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) {
Minchan Kimb06bad12016-04-28 16:18:41 -0700362 if (trylock_page(page)) {
363 swap_slot_free_notify(page);
364 unlock_page(page);
365 }
366
Matthew Wilcoxdd6bd0d2014-06-04 16:07:48 -0700367 count_vm_event(PSWPIN);
368 return 0;
369 }
370
371 ret = 0;
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -0800372 bio = get_swap_bio(GFP_KERNEL, page, end_swap_bio_read);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 if (bio == NULL) {
374 unlock_page(page);
375 ret = -ENOMEM;
376 goto out;
377 }
Mike Christie95fe6c12016-06-05 14:31:48 -0500378 bio_set_op_attrs(bio, REQ_OP_READ, 0);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700379 count_vm_event(PSWPIN);
Mike Christie4e49ea42016-06-05 14:31:41 -0500380 submit_bio(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381out:
382 return ret;
383}
Mel Gorman62c230b2012-07-31 16:44:55 -0700384
385int swap_set_page_dirty(struct page *page)
386{
387 struct swap_info_struct *sis = page_swap_info(page);
388
389 if (sis->flags & SWP_FILE) {
390 struct address_space *mapping = sis->swap_file->f_mapping;
Andrew Mortoncc30c5d2016-10-07 17:00:52 -0700391
392 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
Mel Gorman62c230b2012-07-31 16:44:55 -0700393 return mapping->a_ops->set_page_dirty(page);
394 } else {
395 return __set_page_dirty_no_writeback(page);
396 }
397}