1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
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
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>
29 void end_swap_bio_write(struct bio
*bio
)
31 struct page
*page
= bio_first_page_all(bio
);
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)
41 * Also clear PG_reclaim to avoid rotate_reclaimable_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
);
49 end_page_writeback(page
);
53 static void swap_slot_free_notify(struct page
*page
)
55 struct swap_info_struct
*sis
;
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
65 if (unlikely(!PageSwapCache(page
)))
68 sis
= page_swap_info(page
);
69 if (data_race(!(sis
->flags
& SWP_BLKDEV
)))
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
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.
88 disk
= sis
->bdev
->bd_disk
;
89 entry
.val
= page_private(page
);
90 if (disk
->fops
->swap_slot_free_notify
&& __swap_count(entry
) == 1) {
93 offset
= swp_offset(entry
);
96 disk
->fops
->swap_slot_free_notify(sis
->bdev
,
101 static void end_swap_bio_read(struct bio
*bio
)
103 struct page
*page
= bio_first_page_all(bio
);
104 struct task_struct
*waiter
= bio
->bi_private
;
106 if (bio
->bi_status
) {
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
);
115 SetPageUptodate(page
);
116 swap_slot_free_notify(page
);
119 WRITE_ONCE(bio
->bi_private
, NULL
);
122 blk_wake_io_task(waiter
);
123 put_task_struct(waiter
);
127 int generic_swapfile_activate(struct swap_info_struct
*sis
,
128 struct file
*swap_file
,
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
;
136 sector_t probe_block
;
138 sector_t lowest_block
= -1;
139 sector_t highest_block
= 0;
143 blkbits
= inode
->i_blkbits
;
144 blocks_per_page
= PAGE_SIZE
>> blkbits
;
147 * Map all the blocks into the extent tree. This code doesn't try
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
;
160 first_block
= probe_block
;
161 ret
= bmap(inode
, &first_block
);
162 if (ret
|| !first_block
)
166 * It must be PAGE_SIZE aligned on-disk
168 if (first_block
& (blocks_per_page
- 1)) {
173 for (block_in_page
= 1; block_in_page
< blocks_per_page
;
177 block
= probe_block
+ block_in_page
;
178 ret
= bmap(inode
, &block
);
182 if (block
!= first_block
+ block_in_page
) {
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
;
198 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
200 ret
= add_swap_extent(sis
, page_no
, 1, first_block
);
205 probe_block
+= blocks_per_page
;
210 *span
= 1 + highest_block
- lowest_block
;
212 page_no
= 1; /* force Empty message */
214 sis
->pages
= page_no
- 1;
215 sis
->highest_bit
= page_no
- 1;
219 pr_err("swapon: swapfile has holes\n");
225 * We may have stale swap cache pages in memory: notice
226 * them here and get rid of the unnecessary final write.
228 int swap_writepage(struct page
*page
, struct writeback_control
*wbc
)
232 if (try_to_free_swap(page
)) {
237 * Arch code may have to preserve more data than just the page
238 * contents, e.g. memory tags.
240 ret
= arch_prepare_to_swap(page
);
242 set_page_dirty(page
);
246 if (frontswap_store(page
) == 0) {
247 set_page_writeback(page
);
249 end_page_writeback(page
);
252 ret
= __swap_writepage(page
, wbc
, end_swap_bio_write
);
257 static inline void count_swpout_vm_event(struct page
*page
)
259 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
260 if (unlikely(PageTransHuge(page
)))
261 count_vm_event(THP_SWPOUT
);
263 count_vm_events(PSWPOUT
, thp_nr_pages(page
));
266 #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
267 static void bio_associate_blkg_from_page(struct bio
*bio
, struct page
*page
)
269 struct cgroup_subsys_state
*css
;
270 struct mem_cgroup
*memcg
;
272 memcg
= page_memcg(page
);
277 css
= cgroup_e_css(memcg
->css
.cgroup
, &io_cgrp_subsys
);
278 bio_associate_blkg_from_css(bio
, css
);
282 #define bio_associate_blkg_from_page(bio, page) do { } while (0)
283 #endif /* CONFIG_MEMCG && CONFIG_BLK_CGROUP */
285 int __swap_writepage(struct page
*page
, struct writeback_control
*wbc
,
286 bio_end_io_t end_write_func
)
290 struct swap_info_struct
*sis
= page_swap_info(page
);
292 VM_BUG_ON_PAGE(!PageSwapCache(page
), page
);
293 if (data_race(sis
->flags
& SWP_FS_OPS
)) {
295 struct file
*swap_file
= sis
->swap_file
;
296 struct address_space
*mapping
= swap_file
->f_mapping
;
297 struct bio_vec bv
= {
302 struct iov_iter from
;
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
);
308 set_page_writeback(page
);
310 ret
= mapping
->a_ops
->direct_IO(&kiocb
, &from
);
311 if (ret
== PAGE_SIZE
) {
312 count_vm_event(PSWPOUT
);
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
325 set_page_dirty(page
);
326 ClearPageReclaim(page
);
327 pr_err_ratelimited("Write error on dio swapfile (%llu)\n",
328 page_file_offset(page
));
330 end_page_writeback(page
);
334 ret
= bdev_write_page(sis
->bdev
, swap_page_sector(page
), page
, wbc
);
336 count_swpout_vm_event(page
);
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);
347 bio_associate_blkg_from_page(bio
, page
);
348 count_swpout_vm_event(page
);
349 set_page_writeback(page
);
356 int swap_readpage(struct page
*page
, bool synchronous
)
360 struct swap_info_struct
*sis
= page_swap_info(page
);
362 struct gendisk
*disk
;
363 unsigned long pflags
;
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
);
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.
374 psi_memstall_enter(&pflags
);
376 if (frontswap_load(page
) == 0) {
377 SetPageUptodate(page
);
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
;
386 ret
= mapping
->a_ops
->readpage(swap_file
, page
);
388 count_vm_event(PSWPIN
);
392 if (sis
->flags
& SWP_SYNCHRONOUS_IO
) {
393 ret
= bdev_read_page(sis
->bdev
, swap_page_sector(page
), page
);
395 if (trylock_page(page
)) {
396 swap_slot_free_notify(page
);
400 count_vm_event(PSWPIN
);
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);
413 disk
= bio
->bi_bdev
->bd_disk
;
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.
419 bio
->bi_opf
|= REQ_HIPRI
;
420 get_task_struct(current
);
421 bio
->bi_private
= current
;
423 count_vm_event(PSWPIN
);
425 qc
= submit_bio(bio
);
426 while (synchronous
) {
427 set_current_state(TASK_UNINTERRUPTIBLE
);
428 if (!READ_ONCE(bio
->bi_private
))
431 if (!blk_poll(disk
->queue
, qc
, true))
434 __set_current_state(TASK_RUNNING
);
438 psi_memstall_leave(&pflags
);
442 int swap_set_page_dirty(struct page
*page
)
444 struct swap_info_struct
*sis
= page_swap_info(page
);
446 if (data_race(sis
->flags
& SWP_FS_OPS
)) {
447 struct address_space
*mapping
= sis
->swap_file
->f_mapping
;
449 VM_BUG_ON_PAGE(!PageSwapCache(page
), page
);
450 return mapping
->a_ops
->set_page_dirty(page
);
452 return __set_page_dirty_no_writeback(page
);