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 static struct bio *get_swap_bio(gfp_t gfp_flags,
30 struct page *page, bio_end_io_t end_io)
34 bio = bio_alloc(gfp_flags, 1);
36 struct block_device *bdev;
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;
43 bio_add_page(bio, page, thp_size(page), 0);
48 void end_swap_bio_write(struct bio *bio)
50 struct page *page = bio_first_page_all(bio);
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)
60 * Also clear PG_reclaim to avoid rotate_reclaimable_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);
68 end_page_writeback(page);
72 static void swap_slot_free_notify(struct page *page)
74 struct swap_info_struct *sis;
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
84 if (unlikely(!PageSwapCache(page)))
87 sis = page_swap_info(page);
88 if (data_race(!(sis->flags & SWP_BLKDEV)))
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
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.
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;
112 offset = swp_offset(entry);
115 disk->fops->swap_slot_free_notify(sis->bdev,
120 static void end_swap_bio_read(struct bio *bio)
122 struct page *page = bio_first_page_all(bio);
123 struct task_struct *waiter = bio->bi_private;
125 if (bio->bi_status) {
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);
134 SetPageUptodate(page);
135 swap_slot_free_notify(page);
138 WRITE_ONCE(bio->bi_private, NULL);
141 blk_wake_io_task(waiter);
142 put_task_struct(waiter);
146 int generic_swapfile_activate(struct swap_info_struct *sis,
147 struct file *swap_file,
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;
155 sector_t probe_block;
157 sector_t lowest_block = -1;
158 sector_t highest_block = 0;
162 blkbits = inode->i_blkbits;
163 blocks_per_page = PAGE_SIZE >> blkbits;
166 * Map all the blocks into the extent tree. This code doesn't try
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;
179 first_block = probe_block;
180 ret = bmap(inode, &first_block);
181 if (ret || !first_block)
185 * It must be PAGE_SIZE aligned on-disk
187 if (first_block & (blocks_per_page - 1)) {
192 for (block_in_page = 1; block_in_page < blocks_per_page;
196 block = probe_block + block_in_page;
197 ret = bmap(inode, &block);
201 if (block != first_block + block_in_page) {
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;
217 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
219 ret = add_swap_extent(sis, page_no, 1, first_block);
224 probe_block += blocks_per_page;
229 *span = 1 + highest_block - lowest_block;
231 page_no = 1; /* force Empty message */
233 sis->pages = page_no - 1;
234 sis->highest_bit = page_no - 1;
238 pr_err("swapon: swapfile has holes\n");
244 * We may have stale swap cache pages in memory: notice
245 * them here and get rid of the unnecessary final write.
247 int swap_writepage(struct page *page, struct writeback_control *wbc)
251 if (try_to_free_swap(page)) {
255 if (frontswap_store(page) == 0) {
256 set_page_writeback(page);
258 end_page_writeback(page);
261 ret = __swap_writepage(page, wbc, end_swap_bio_write);
266 static sector_t swap_page_sector(struct page *page)
268 return (sector_t)__page_file_index(page) << (PAGE_SHIFT - 9);
271 static inline void count_swpout_vm_event(struct page *page)
273 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
274 if (unlikely(PageTransHuge(page)))
275 count_vm_event(THP_SWPOUT);
277 count_vm_events(PSWPOUT, thp_nr_pages(page));
280 #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
281 static void bio_associate_blkg_from_page(struct bio *bio, struct page *page)
283 struct cgroup_subsys_state *css;
285 if (!page->mem_cgroup)
289 css = cgroup_e_css(page->mem_cgroup->css.cgroup, &io_cgrp_subsys);
290 bio_associate_blkg_from_css(bio, css);
294 #define bio_associate_blkg_from_page(bio, page) do { } while (0)
295 #endif /* CONFIG_MEMCG && CONFIG_BLK_CGROUP */
297 int __swap_writepage(struct page *page, struct writeback_control *wbc,
298 bio_end_io_t end_write_func)
302 struct swap_info_struct *sis = page_swap_info(page);
304 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
305 if (data_race(sis->flags & SWP_FS)) {
307 struct file *swap_file = sis->swap_file;
308 struct address_space *mapping = swap_file->f_mapping;
309 struct bio_vec bv = {
314 struct iov_iter from;
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);
320 set_page_writeback(page);
322 ret = mapping->a_ops->direct_IO(&kiocb, &from);
323 if (ret == PAGE_SIZE) {
324 count_vm_event(PSWPOUT);
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
337 set_page_dirty(page);
338 ClearPageReclaim(page);
339 pr_err_ratelimited("Write error on dio swapfile (%llu)\n",
340 page_file_offset(page));
342 end_page_writeback(page);
346 ret = bdev_write_page(sis->bdev, swap_page_sector(page), page, wbc);
348 count_swpout_vm_event(page);
353 bio = get_swap_bio(GFP_NOIO, page, end_write_func);
355 set_page_dirty(page);
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);
370 int swap_readpage(struct page *page, bool synchronous)
374 struct swap_info_struct *sis = page_swap_info(page);
376 struct gendisk *disk;
377 unsigned long pflags;
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);
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.
388 psi_memstall_enter(&pflags);
390 if (frontswap_load(page) == 0) {
391 SetPageUptodate(page);
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;
400 ret = mapping->a_ops->readpage(swap_file, page);
402 count_vm_event(PSWPIN);
406 ret = bdev_read_page(sis->bdev, swap_page_sector(page), page);
408 if (trylock_page(page)) {
409 swap_slot_free_notify(page);
413 count_vm_event(PSWPIN);
418 bio = get_swap_bio(GFP_KERNEL, page, end_swap_bio_read);
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.
429 bio_set_op_attrs(bio, REQ_OP_READ, 0);
431 bio->bi_opf |= REQ_HIPRI;
432 get_task_struct(current);
433 bio->bi_private = current;
435 count_vm_event(PSWPIN);
437 qc = submit_bio(bio);
438 while (synchronous) {
439 set_current_state(TASK_UNINTERRUPTIBLE);
440 if (!READ_ONCE(bio->bi_private))
443 if (!blk_poll(disk->queue, qc, true))
446 __set_current_state(TASK_RUNNING);
450 psi_memstall_leave(&pflags);
454 int swap_set_page_dirty(struct page *page)
456 struct swap_info_struct *sis = page_swap_info(page);
458 if (data_race(sis->flags & SWP_FS)) {
459 struct address_space *mapping = sis->swap_file->f_mapping;
461 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
462 return mapping->a_ops->set_page_dirty(page);
464 return __set_page_dirty_no_writeback(page);