Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/mm/page_io.c
4 *
5 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
6 *
7 * Swap reorganised 29.12.95,
8 * Asynchronous swapping added 30.12.95. Stephen Tweedie
9 * Removed race in async swapping. 14.4.1996. Bruno Haible
10 * Add swap of shared pages through the page cache. 20.2.1998. Stephen Tweedie
11 * Always use brw_page, life becomes simpler. 12 May 1998 Eric Biederman
12 */
13
14#include <linux/mm.h>
15#include <linux/kernel_stat.h>
16#include <linux/gfp.h>
17#include <linux/pagemap.h>
18#include <linux/swap.h>
19#include <linux/bio.h>
20#include <linux/swapops.h>
21#include <linux/buffer_head.h>
22#include <linux/writeback.h>
23#include <linux/frontswap.h>
24#include <linux/blkdev.h>
25#include <linux/psi.h>
26#include <linux/uio.h>
27#include <linux/sched/task.h>
28
29void end_swap_bio_write(struct bio *bio)
30{
31 struct page *page = bio_first_page_all(bio);
32
33 if (bio->bi_status) {
34 SetPageError(page);
35 /*
36 * We failed to write the page out to swap-space.
37 * Re-dirty the page in order to avoid it being reclaimed.
38 * Also print a dire warning that things will go BAD (tm)
39 * very quickly.
40 *
41 * Also clear PG_reclaim to avoid rotate_reclaimable_page()
42 */
43 set_page_dirty(page);
44 pr_alert_ratelimited("Write-error on swap-device (%u:%u:%llu)\n",
45 MAJOR(bio_dev(bio)), MINOR(bio_dev(bio)),
46 (unsigned long long)bio->bi_iter.bi_sector);
47 ClearPageReclaim(page);
48 }
49 end_page_writeback(page);
50 bio_put(bio);
51}
52
53static void swap_slot_free_notify(struct page *page)
54{
55 struct swap_info_struct *sis;
56 struct gendisk *disk;
57 swp_entry_t entry;
58
59 /*
60 * There is no guarantee that the page is in swap cache - the software
61 * suspend code (at least) uses end_swap_bio_read() against a non-
62 * swapcache page. So we must check PG_swapcache before proceeding with
63 * this optimization.
64 */
65 if (unlikely(!PageSwapCache(page)))
66 return;
67
68 sis = page_swap_info(page);
69 if (data_race(!(sis->flags & SWP_BLKDEV)))
70 return;
71
72 /*
73 * The swap subsystem performs lazy swap slot freeing,
74 * expecting that the page will be swapped out again.
75 * So we can avoid an unnecessary write if the page
76 * isn't redirtied.
77 * This is good for real swap storage because we can
78 * reduce unnecessary I/O and enhance wear-leveling
79 * if an SSD is used as the as swap device.
80 * But if in-memory swap device (eg zram) is used,
81 * this causes a duplicated copy between uncompressed
82 * data in VM-owned memory and compressed data in
83 * zram-owned memory. So let's free zram-owned memory
84 * and make the VM-owned decompressed page *dirty*,
85 * so the page should be swapped out somewhere again if
86 * we again wish to reclaim it.
87 */
88 disk = sis->bdev->bd_disk;
89 entry.val = page_private(page);
90 if (disk->fops->swap_slot_free_notify && __swap_count(entry) == 1) {
91 unsigned long offset;
92
93 offset = swp_offset(entry);
94
95 SetPageDirty(page);
96 disk->fops->swap_slot_free_notify(sis->bdev,
97 offset);
98 }
99}
100
101static void end_swap_bio_read(struct bio *bio)
102{
103 struct page *page = bio_first_page_all(bio);
104 struct task_struct *waiter = bio->bi_private;
105
106 if (bio->bi_status) {
107 SetPageError(page);
108 ClearPageUptodate(page);
109 pr_alert_ratelimited("Read-error on swap-device (%u:%u:%llu)\n",
110 MAJOR(bio_dev(bio)), MINOR(bio_dev(bio)),
111 (unsigned long long)bio->bi_iter.bi_sector);
112 goto out;
113 }
114
115 SetPageUptodate(page);
116 swap_slot_free_notify(page);
117out:
118 unlock_page(page);
119 WRITE_ONCE(bio->bi_private, NULL);
120 bio_put(bio);
121 if (waiter) {
122 blk_wake_io_task(waiter);
123 put_task_struct(waiter);
124 }
125}
126
127int generic_swapfile_activate(struct swap_info_struct *sis,
128 struct file *swap_file,
129 sector_t *span)
130{
131 struct address_space *mapping = swap_file->f_mapping;
132 struct inode *inode = mapping->host;
133 unsigned blocks_per_page;
134 unsigned long page_no;
135 unsigned blkbits;
136 sector_t probe_block;
137 sector_t last_block;
138 sector_t lowest_block = -1;
139 sector_t highest_block = 0;
140 int nr_extents = 0;
141 int ret;
142
143 blkbits = inode->i_blkbits;
144 blocks_per_page = PAGE_SIZE >> blkbits;
145
146 /*
147 * Map all the blocks into the extent tree. This code doesn't try
148 * to be very smart.
149 */
150 probe_block = 0;
151 page_no = 0;
152 last_block = i_size_read(inode) >> blkbits;
153 while ((probe_block + blocks_per_page) <= last_block &&
154 page_no < sis->max) {
155 unsigned block_in_page;
156 sector_t first_block;
157
158 cond_resched();
159
160 first_block = probe_block;
161 ret = bmap(inode, &first_block);
162 if (ret || !first_block)
163 goto bad_bmap;
164
165 /*
166 * It must be PAGE_SIZE aligned on-disk
167 */
168 if (first_block & (blocks_per_page - 1)) {
169 probe_block++;
170 goto reprobe;
171 }
172
173 for (block_in_page = 1; block_in_page < blocks_per_page;
174 block_in_page++) {
175 sector_t block;
176
177 block = probe_block + block_in_page;
178 ret = bmap(inode, &block);
179 if (ret || !block)
180 goto bad_bmap;
181
182 if (block != first_block + block_in_page) {
183 /* Discontiguity */
184 probe_block++;
185 goto reprobe;
186 }
187 }
188
189 first_block >>= (PAGE_SHIFT - blkbits);
190 if (page_no) { /* exclude the header page */
191 if (first_block < lowest_block)
192 lowest_block = first_block;
193 if (first_block > highest_block)
194 highest_block = first_block;
195 }
196
197 /*
198 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
199 */
200 ret = add_swap_extent(sis, page_no, 1, first_block);
201 if (ret < 0)
202 goto out;
203 nr_extents += ret;
204 page_no++;
205 probe_block += blocks_per_page;
206reprobe:
207 continue;
208 }
209 ret = nr_extents;
210 *span = 1 + highest_block - lowest_block;
211 if (page_no == 0)
212 page_no = 1; /* force Empty message */
213 sis->max = page_no;
214 sis->pages = page_no - 1;
215 sis->highest_bit = page_no - 1;
216out:
217 return ret;
218bad_bmap:
219 pr_err("swapon: swapfile has holes\n");
220 ret = -EINVAL;
221 goto out;
222}
223
224/*
225 * We may have stale swap cache pages in memory: notice
226 * them here and get rid of the unnecessary final write.
227 */
228int swap_writepage(struct page *page, struct writeback_control *wbc)
229{
230 int ret = 0;
231
232 if (try_to_free_swap(page)) {
233 unlock_page(page);
234 goto out;
235 }
236 /*
237 * Arch code may have to preserve more data than just the page
238 * contents, e.g. memory tags.
239 */
240 ret = arch_prepare_to_swap(page);
241 if (ret) {
242 set_page_dirty(page);
243 unlock_page(page);
244 goto out;
245 }
246 if (frontswap_store(page) == 0) {
247 set_page_writeback(page);
248 unlock_page(page);
249 end_page_writeback(page);
250 goto out;
251 }
252 ret = __swap_writepage(page, wbc, end_swap_bio_write);
253out:
254 return ret;
255}
256
257static inline void count_swpout_vm_event(struct page *page)
258{
259#ifdef CONFIG_TRANSPARENT_HUGEPAGE
260 if (unlikely(PageTransHuge(page)))
261 count_vm_event(THP_SWPOUT);
262#endif
263 count_vm_events(PSWPOUT, thp_nr_pages(page));
264}
265
266#if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
267static void bio_associate_blkg_from_page(struct bio *bio, struct page *page)
268{
269 struct cgroup_subsys_state *css;
270 struct mem_cgroup *memcg;
271
272 memcg = page_memcg(page);
273 if (!memcg)
274 return;
275
276 rcu_read_lock();
277 css = cgroup_e_css(memcg->css.cgroup, &io_cgrp_subsys);
278 bio_associate_blkg_from_css(bio, css);
279 rcu_read_unlock();
280}
281#else
282#define bio_associate_blkg_from_page(bio, page) do { } while (0)
283#endif /* CONFIG_MEMCG && CONFIG_BLK_CGROUP */
284
285int __swap_writepage(struct page *page, struct writeback_control *wbc,
286 bio_end_io_t end_write_func)
287{
288 struct bio *bio;
289 int ret;
290 struct swap_info_struct *sis = page_swap_info(page);
291
292 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
293 if (data_race(sis->flags & SWP_FS_OPS)) {
294 struct kiocb kiocb;
295 struct file *swap_file = sis->swap_file;
296 struct address_space *mapping = swap_file->f_mapping;
297 struct bio_vec bv = {
298 .bv_page = page,
299 .bv_len = PAGE_SIZE,
300 .bv_offset = 0
301 };
302 struct iov_iter from;
303
304 iov_iter_bvec(&from, WRITE, &bv, 1, PAGE_SIZE);
305 init_sync_kiocb(&kiocb, swap_file);
306 kiocb.ki_pos = page_file_offset(page);
307
308 set_page_writeback(page);
309 unlock_page(page);
310 ret = mapping->a_ops->direct_IO(&kiocb, &from);
311 if (ret == PAGE_SIZE) {
312 count_vm_event(PSWPOUT);
313 ret = 0;
314 } else {
315 /*
316 * In the case of swap-over-nfs, this can be a
317 * temporary failure if the system has limited
318 * memory for allocating transmit buffers.
319 * Mark the page dirty and avoid
320 * rotate_reclaimable_page but rate-limit the
321 * messages but do not flag PageError like
322 * the normal direct-to-bio case as it could
323 * be temporary.
324 */
325 set_page_dirty(page);
326 ClearPageReclaim(page);
327 pr_err_ratelimited("Write error on dio swapfile (%llu)\n",
328 page_file_offset(page));
329 }
330 end_page_writeback(page);
331 return ret;
332 }
333
334 ret = bdev_write_page(sis->bdev, swap_page_sector(page), page, wbc);
335 if (!ret) {
336 count_swpout_vm_event(page);
337 return 0;
338 }
339
340 bio = bio_alloc(GFP_NOIO, 1);
341 bio_set_dev(bio, sis->bdev);
342 bio->bi_iter.bi_sector = swap_page_sector(page);
343 bio->bi_opf = REQ_OP_WRITE | REQ_SWAP | wbc_to_write_flags(wbc);
344 bio->bi_end_io = end_write_func;
345 bio_add_page(bio, page, thp_size(page), 0);
346
347 bio_associate_blkg_from_page(bio, page);
348 count_swpout_vm_event(page);
349 set_page_writeback(page);
350 unlock_page(page);
351 submit_bio(bio);
352
353 return 0;
354}
355
356int swap_readpage(struct page *page, bool synchronous)
357{
358 struct bio *bio;
359 int ret = 0;
360 struct swap_info_struct *sis = page_swap_info(page);
361 blk_qc_t qc;
362 struct gendisk *disk;
363 unsigned long pflags;
364
365 VM_BUG_ON_PAGE(!PageSwapCache(page) && !synchronous, page);
366 VM_BUG_ON_PAGE(!PageLocked(page), page);
367 VM_BUG_ON_PAGE(PageUptodate(page), page);
368
369 /*
370 * Count submission time as memory stall. When the device is congested,
371 * or the submitting cgroup IO-throttled, submission can be a
372 * significant part of overall IO time.
373 */
374 psi_memstall_enter(&pflags);
375
376 if (frontswap_load(page) == 0) {
377 SetPageUptodate(page);
378 unlock_page(page);
379 goto out;
380 }
381
382 if (data_race(sis->flags & SWP_FS_OPS)) {
383 struct file *swap_file = sis->swap_file;
384 struct address_space *mapping = swap_file->f_mapping;
385
386 ret = mapping->a_ops->readpage(swap_file, page);
387 if (!ret)
388 count_vm_event(PSWPIN);
389 goto out;
390 }
391
392 if (sis->flags & SWP_SYNCHRONOUS_IO) {
393 ret = bdev_read_page(sis->bdev, swap_page_sector(page), page);
394 if (!ret) {
395 if (trylock_page(page)) {
396 swap_slot_free_notify(page);
397 unlock_page(page);
398 }
399
400 count_vm_event(PSWPIN);
401 goto out;
402 }
403 }
404
405 ret = 0;
406 bio = bio_alloc(GFP_KERNEL, 1);
407 bio_set_dev(bio, sis->bdev);
408 bio->bi_opf = REQ_OP_READ;
409 bio->bi_iter.bi_sector = swap_page_sector(page);
410 bio->bi_end_io = end_swap_bio_read;
411 bio_add_page(bio, page, thp_size(page), 0);
412
413 disk = bio->bi_bdev->bd_disk;
414 /*
415 * Keep this task valid during swap readpage because the oom killer may
416 * attempt to access it in the page fault retry time check.
417 */
418 if (synchronous) {
419 bio->bi_opf |= REQ_HIPRI;
420 get_task_struct(current);
421 bio->bi_private = current;
422 }
423 count_vm_event(PSWPIN);
424 bio_get(bio);
425 qc = submit_bio(bio);
426 while (synchronous) {
427 set_current_state(TASK_UNINTERRUPTIBLE);
428 if (!READ_ONCE(bio->bi_private))
429 break;
430
431 if (!blk_poll(disk->queue, qc, true))
432 blk_io_schedule();
433 }
434 __set_current_state(TASK_RUNNING);
435 bio_put(bio);
436
437out:
438 psi_memstall_leave(&pflags);
439 return ret;
440}
441
442int swap_set_page_dirty(struct page *page)
443{
444 struct swap_info_struct *sis = page_swap_info(page);
445
446 if (data_race(sis->flags & SWP_FS_OPS)) {
447 struct address_space *mapping = sis->swap_file->f_mapping;
448
449 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
450 return mapping->a_ops->set_page_dirty(page);
451 } else {
452 return __set_page_dirty_no_writeback(page);
453 }
454}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/mm/page_io.c
4 *
5 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
6 *
7 * Swap reorganised 29.12.95,
8 * Asynchronous swapping added 30.12.95. Stephen Tweedie
9 * Removed race in async swapping. 14.4.1996. Bruno Haible
10 * Add swap of shared pages through the page cache. 20.2.1998. Stephen Tweedie
11 * Always use brw_page, life becomes simpler. 12 May 1998 Eric Biederman
12 */
13
14#include <linux/mm.h>
15#include <linux/kernel_stat.h>
16#include <linux/gfp.h>
17#include <linux/pagemap.h>
18#include <linux/swap.h>
19#include <linux/bio.h>
20#include <linux/swapops.h>
21#include <linux/buffer_head.h>
22#include <linux/writeback.h>
23#include <linux/frontswap.h>
24#include <linux/blkdev.h>
25#include <linux/psi.h>
26#include <linux/uio.h>
27#include <linux/sched/task.h>
28
29static struct bio *get_swap_bio(gfp_t gfp_flags,
30 struct page *page, bio_end_io_t end_io)
31{
32 struct bio *bio;
33
34 bio = bio_alloc(gfp_flags, 1);
35 if (bio) {
36 struct block_device *bdev;
37
38 bio->bi_iter.bi_sector = map_swap_page(page, &bdev);
39 bio_set_dev(bio, bdev);
40 bio->bi_iter.bi_sector <<= PAGE_SHIFT - 9;
41 bio->bi_end_io = end_io;
42
43 bio_add_page(bio, page, thp_size(page), 0);
44 }
45 return bio;
46}
47
48void end_swap_bio_write(struct bio *bio)
49{
50 struct page *page = bio_first_page_all(bio);
51
52 if (bio->bi_status) {
53 SetPageError(page);
54 /*
55 * We failed to write the page out to swap-space.
56 * Re-dirty the page in order to avoid it being reclaimed.
57 * Also print a dire warning that things will go BAD (tm)
58 * very quickly.
59 *
60 * Also clear PG_reclaim to avoid rotate_reclaimable_page()
61 */
62 set_page_dirty(page);
63 pr_alert("Write-error on swap-device (%u:%u:%llu)\n",
64 MAJOR(bio_dev(bio)), MINOR(bio_dev(bio)),
65 (unsigned long long)bio->bi_iter.bi_sector);
66 ClearPageReclaim(page);
67 }
68 end_page_writeback(page);
69 bio_put(bio);
70}
71
72static void swap_slot_free_notify(struct page *page)
73{
74 struct swap_info_struct *sis;
75 struct gendisk *disk;
76 swp_entry_t entry;
77
78 /*
79 * There is no guarantee that the page is in swap cache - the software
80 * suspend code (at least) uses end_swap_bio_read() against a non-
81 * swapcache page. So we must check PG_swapcache before proceeding with
82 * this optimization.
83 */
84 if (unlikely(!PageSwapCache(page)))
85 return;
86
87 sis = page_swap_info(page);
88 if (data_race(!(sis->flags & SWP_BLKDEV)))
89 return;
90
91 /*
92 * The swap subsystem performs lazy swap slot freeing,
93 * expecting that the page will be swapped out again.
94 * So we can avoid an unnecessary write if the page
95 * isn't redirtied.
96 * This is good for real swap storage because we can
97 * reduce unnecessary I/O and enhance wear-leveling
98 * if an SSD is used as the as swap device.
99 * But if in-memory swap device (eg zram) is used,
100 * this causes a duplicated copy between uncompressed
101 * data in VM-owned memory and compressed data in
102 * zram-owned memory. So let's free zram-owned memory
103 * and make the VM-owned decompressed page *dirty*,
104 * so the page should be swapped out somewhere again if
105 * we again wish to reclaim it.
106 */
107 disk = sis->bdev->bd_disk;
108 entry.val = page_private(page);
109 if (disk->fops->swap_slot_free_notify && __swap_count(entry) == 1) {
110 unsigned long offset;
111
112 offset = swp_offset(entry);
113
114 SetPageDirty(page);
115 disk->fops->swap_slot_free_notify(sis->bdev,
116 offset);
117 }
118}
119
120static void end_swap_bio_read(struct bio *bio)
121{
122 struct page *page = bio_first_page_all(bio);
123 struct task_struct *waiter = bio->bi_private;
124
125 if (bio->bi_status) {
126 SetPageError(page);
127 ClearPageUptodate(page);
128 pr_alert("Read-error on swap-device (%u:%u:%llu)\n",
129 MAJOR(bio_dev(bio)), MINOR(bio_dev(bio)),
130 (unsigned long long)bio->bi_iter.bi_sector);
131 goto out;
132 }
133
134 SetPageUptodate(page);
135 swap_slot_free_notify(page);
136out:
137 unlock_page(page);
138 WRITE_ONCE(bio->bi_private, NULL);
139 bio_put(bio);
140 if (waiter) {
141 blk_wake_io_task(waiter);
142 put_task_struct(waiter);
143 }
144}
145
146int generic_swapfile_activate(struct swap_info_struct *sis,
147 struct file *swap_file,
148 sector_t *span)
149{
150 struct address_space *mapping = swap_file->f_mapping;
151 struct inode *inode = mapping->host;
152 unsigned blocks_per_page;
153 unsigned long page_no;
154 unsigned blkbits;
155 sector_t probe_block;
156 sector_t last_block;
157 sector_t lowest_block = -1;
158 sector_t highest_block = 0;
159 int nr_extents = 0;
160 int ret;
161
162 blkbits = inode->i_blkbits;
163 blocks_per_page = PAGE_SIZE >> blkbits;
164
165 /*
166 * Map all the blocks into the extent tree. This code doesn't try
167 * to be very smart.
168 */
169 probe_block = 0;
170 page_no = 0;
171 last_block = i_size_read(inode) >> blkbits;
172 while ((probe_block + blocks_per_page) <= last_block &&
173 page_no < sis->max) {
174 unsigned block_in_page;
175 sector_t first_block;
176
177 cond_resched();
178
179 first_block = probe_block;
180 ret = bmap(inode, &first_block);
181 if (ret || !first_block)
182 goto bad_bmap;
183
184 /*
185 * It must be PAGE_SIZE aligned on-disk
186 */
187 if (first_block & (blocks_per_page - 1)) {
188 probe_block++;
189 goto reprobe;
190 }
191
192 for (block_in_page = 1; block_in_page < blocks_per_page;
193 block_in_page++) {
194 sector_t block;
195
196 block = probe_block + block_in_page;
197 ret = bmap(inode, &block);
198 if (ret || !block)
199 goto bad_bmap;
200
201 if (block != first_block + block_in_page) {
202 /* Discontiguity */
203 probe_block++;
204 goto reprobe;
205 }
206 }
207
208 first_block >>= (PAGE_SHIFT - blkbits);
209 if (page_no) { /* exclude the header page */
210 if (first_block < lowest_block)
211 lowest_block = first_block;
212 if (first_block > highest_block)
213 highest_block = first_block;
214 }
215
216 /*
217 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
218 */
219 ret = add_swap_extent(sis, page_no, 1, first_block);
220 if (ret < 0)
221 goto out;
222 nr_extents += ret;
223 page_no++;
224 probe_block += blocks_per_page;
225reprobe:
226 continue;
227 }
228 ret = nr_extents;
229 *span = 1 + highest_block - lowest_block;
230 if (page_no == 0)
231 page_no = 1; /* force Empty message */
232 sis->max = page_no;
233 sis->pages = page_no - 1;
234 sis->highest_bit = page_no - 1;
235out:
236 return ret;
237bad_bmap:
238 pr_err("swapon: swapfile has holes\n");
239 ret = -EINVAL;
240 goto out;
241}
242
243/*
244 * We may have stale swap cache pages in memory: notice
245 * them here and get rid of the unnecessary final write.
246 */
247int swap_writepage(struct page *page, struct writeback_control *wbc)
248{
249 int ret = 0;
250
251 if (try_to_free_swap(page)) {
252 unlock_page(page);
253 goto out;
254 }
255 if (frontswap_store(page) == 0) {
256 set_page_writeback(page);
257 unlock_page(page);
258 end_page_writeback(page);
259 goto out;
260 }
261 ret = __swap_writepage(page, wbc, end_swap_bio_write);
262out:
263 return ret;
264}
265
266static sector_t swap_page_sector(struct page *page)
267{
268 return (sector_t)__page_file_index(page) << (PAGE_SHIFT - 9);
269}
270
271static inline void count_swpout_vm_event(struct page *page)
272{
273#ifdef CONFIG_TRANSPARENT_HUGEPAGE
274 if (unlikely(PageTransHuge(page)))
275 count_vm_event(THP_SWPOUT);
276#endif
277 count_vm_events(PSWPOUT, thp_nr_pages(page));
278}
279
280#if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
281static void bio_associate_blkg_from_page(struct bio *bio, struct page *page)
282{
283 struct cgroup_subsys_state *css;
284
285 if (!page->mem_cgroup)
286 return;
287
288 rcu_read_lock();
289 css = cgroup_e_css(page->mem_cgroup->css.cgroup, &io_cgrp_subsys);
290 bio_associate_blkg_from_css(bio, css);
291 rcu_read_unlock();
292}
293#else
294#define bio_associate_blkg_from_page(bio, page) do { } while (0)
295#endif /* CONFIG_MEMCG && CONFIG_BLK_CGROUP */
296
297int __swap_writepage(struct page *page, struct writeback_control *wbc,
298 bio_end_io_t end_write_func)
299{
300 struct bio *bio;
301 int ret;
302 struct swap_info_struct *sis = page_swap_info(page);
303
304 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
305 if (data_race(sis->flags & SWP_FS)) {
306 struct kiocb kiocb;
307 struct file *swap_file = sis->swap_file;
308 struct address_space *mapping = swap_file->f_mapping;
309 struct bio_vec bv = {
310 .bv_page = page,
311 .bv_len = PAGE_SIZE,
312 .bv_offset = 0
313 };
314 struct iov_iter from;
315
316 iov_iter_bvec(&from, WRITE, &bv, 1, PAGE_SIZE);
317 init_sync_kiocb(&kiocb, swap_file);
318 kiocb.ki_pos = page_file_offset(page);
319
320 set_page_writeback(page);
321 unlock_page(page);
322 ret = mapping->a_ops->direct_IO(&kiocb, &from);
323 if (ret == PAGE_SIZE) {
324 count_vm_event(PSWPOUT);
325 ret = 0;
326 } else {
327 /*
328 * In the case of swap-over-nfs, this can be a
329 * temporary failure if the system has limited
330 * memory for allocating transmit buffers.
331 * Mark the page dirty and avoid
332 * rotate_reclaimable_page but rate-limit the
333 * messages but do not flag PageError like
334 * the normal direct-to-bio case as it could
335 * be temporary.
336 */
337 set_page_dirty(page);
338 ClearPageReclaim(page);
339 pr_err_ratelimited("Write error on dio swapfile (%llu)\n",
340 page_file_offset(page));
341 }
342 end_page_writeback(page);
343 return ret;
344 }
345
346 ret = bdev_write_page(sis->bdev, swap_page_sector(page), page, wbc);
347 if (!ret) {
348 count_swpout_vm_event(page);
349 return 0;
350 }
351
352 ret = 0;
353 bio = get_swap_bio(GFP_NOIO, page, end_write_func);
354 if (bio == NULL) {
355 set_page_dirty(page);
356 unlock_page(page);
357 ret = -ENOMEM;
358 goto out;
359 }
360 bio->bi_opf = REQ_OP_WRITE | REQ_SWAP | wbc_to_write_flags(wbc);
361 bio_associate_blkg_from_page(bio, page);
362 count_swpout_vm_event(page);
363 set_page_writeback(page);
364 unlock_page(page);
365 submit_bio(bio);
366out:
367 return ret;
368}
369
370int swap_readpage(struct page *page, bool synchronous)
371{
372 struct bio *bio;
373 int ret = 0;
374 struct swap_info_struct *sis = page_swap_info(page);
375 blk_qc_t qc;
376 struct gendisk *disk;
377 unsigned long pflags;
378
379 VM_BUG_ON_PAGE(!PageSwapCache(page) && !synchronous, page);
380 VM_BUG_ON_PAGE(!PageLocked(page), page);
381 VM_BUG_ON_PAGE(PageUptodate(page), page);
382
383 /*
384 * Count submission time as memory stall. When the device is congested,
385 * or the submitting cgroup IO-throttled, submission can be a
386 * significant part of overall IO time.
387 */
388 psi_memstall_enter(&pflags);
389
390 if (frontswap_load(page) == 0) {
391 SetPageUptodate(page);
392 unlock_page(page);
393 goto out;
394 }
395
396 if (data_race(sis->flags & SWP_FS)) {
397 struct file *swap_file = sis->swap_file;
398 struct address_space *mapping = swap_file->f_mapping;
399
400 ret = mapping->a_ops->readpage(swap_file, page);
401 if (!ret)
402 count_vm_event(PSWPIN);
403 goto out;
404 }
405
406 ret = bdev_read_page(sis->bdev, swap_page_sector(page), page);
407 if (!ret) {
408 if (trylock_page(page)) {
409 swap_slot_free_notify(page);
410 unlock_page(page);
411 }
412
413 count_vm_event(PSWPIN);
414 goto out;
415 }
416
417 ret = 0;
418 bio = get_swap_bio(GFP_KERNEL, page, end_swap_bio_read);
419 if (bio == NULL) {
420 unlock_page(page);
421 ret = -ENOMEM;
422 goto out;
423 }
424 disk = bio->bi_disk;
425 /*
426 * Keep this task valid during swap readpage because the oom killer may
427 * attempt to access it in the page fault retry time check.
428 */
429 bio_set_op_attrs(bio, REQ_OP_READ, 0);
430 if (synchronous) {
431 bio->bi_opf |= REQ_HIPRI;
432 get_task_struct(current);
433 bio->bi_private = current;
434 }
435 count_vm_event(PSWPIN);
436 bio_get(bio);
437 qc = submit_bio(bio);
438 while (synchronous) {
439 set_current_state(TASK_UNINTERRUPTIBLE);
440 if (!READ_ONCE(bio->bi_private))
441 break;
442
443 if (!blk_poll(disk->queue, qc, true))
444 blk_io_schedule();
445 }
446 __set_current_state(TASK_RUNNING);
447 bio_put(bio);
448
449out:
450 psi_memstall_leave(&pflags);
451 return ret;
452}
453
454int swap_set_page_dirty(struct page *page)
455{
456 struct swap_info_struct *sis = page_swap_info(page);
457
458 if (data_race(sis->flags & SWP_FS)) {
459 struct address_space *mapping = sis->swap_file->f_mapping;
460
461 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
462 return mapping->a_ops->set_page_dirty(page);
463 } else {
464 return __set_page_dirty_no_writeback(page);
465 }
466}