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/uio.h>
26 #include <linux/sched/task.h>
27 #include <asm/pgtable.h>
29 static struct bio *get_swap_bio(gfp_t gfp_flags,
30 struct page *page, bio_end_io_t end_io)
32 int i, nr = hpage_nr_pages(page);
35 bio = bio_alloc(gfp_flags, nr);
37 struct block_device *bdev;
39 bio->bi_iter.bi_sector = map_swap_page(page, &bdev);
40 bio_set_dev(bio, bdev);
41 bio->bi_end_io = end_io;
43 for (i = 0; i < nr; i++)
44 bio_add_page(bio, page + i, PAGE_SIZE, 0);
45 VM_BUG_ON(bio->bi_iter.bi_size != PAGE_SIZE * nr);
50 void end_swap_bio_write(struct bio *bio)
52 struct page *page = bio->bi_io_vec[0].bv_page;
57 * We failed to write the page out to swap-space.
58 * Re-dirty the page in order to avoid it being reclaimed.
59 * Also print a dire warning that things will go BAD (tm)
62 * Also clear PG_reclaim to avoid rotate_reclaimable_page()
65 pr_alert("Write-error on swap-device (%u:%u:%llu)\n",
66 MAJOR(bio_dev(bio)), MINOR(bio_dev(bio)),
67 (unsigned long long)bio->bi_iter.bi_sector);
68 ClearPageReclaim(page);
70 end_page_writeback(page);
74 static void swap_slot_free_notify(struct page *page)
76 struct swap_info_struct *sis;
80 * There is no guarantee that the page is in swap cache - the software
81 * suspend code (at least) uses end_swap_bio_read() against a non-
82 * swapcache page. So we must check PG_swapcache before proceeding with
85 if (unlikely(!PageSwapCache(page)))
88 sis = page_swap_info(page);
89 if (!(sis->flags & SWP_BLKDEV))
93 * The swap subsystem performs lazy swap slot freeing,
94 * expecting that the page will be swapped out again.
95 * So we can avoid an unnecessary write if the page
97 * This is good for real swap storage because we can
98 * reduce unnecessary I/O and enhance wear-leveling
99 * if an SSD is used as the as swap device.
100 * But if in-memory swap device (eg zram) is used,
101 * this causes a duplicated copy between uncompressed
102 * data in VM-owned memory and compressed data in
103 * zram-owned memory. So let's free zram-owned memory
104 * and make the VM-owned decompressed page *dirty*,
105 * so the page should be swapped out somewhere again if
106 * we again wish to reclaim it.
108 disk = sis->bdev->bd_disk;
109 if (disk->fops->swap_slot_free_notify) {
111 unsigned long offset;
113 entry.val = page_private(page);
114 offset = swp_offset(entry);
117 disk->fops->swap_slot_free_notify(sis->bdev,
122 static void end_swap_bio_read(struct bio *bio)
124 struct page *page = bio->bi_io_vec[0].bv_page;
125 struct task_struct *waiter = bio->bi_private;
127 if (bio->bi_status) {
129 ClearPageUptodate(page);
130 pr_alert("Read-error on swap-device (%u:%u:%llu)\n",
131 MAJOR(bio_dev(bio)), MINOR(bio_dev(bio)),
132 (unsigned long long)bio->bi_iter.bi_sector);
136 SetPageUptodate(page);
137 swap_slot_free_notify(page);
140 WRITE_ONCE(bio->bi_private, NULL);
142 wake_up_process(waiter);
143 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 list. 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 = bmap(inode, probe_block);
180 if (first_block == 0)
184 * It must be PAGE_SIZE aligned on-disk
186 if (first_block & (blocks_per_page - 1)) {
191 for (block_in_page = 1; block_in_page < blocks_per_page;
195 block = bmap(inode, probe_block + block_in_page);
198 if (block != first_block + block_in_page) {
205 first_block >>= (PAGE_SHIFT - blkbits);
206 if (page_no) { /* exclude the header page */
207 if (first_block < lowest_block)
208 lowest_block = first_block;
209 if (first_block > highest_block)
210 highest_block = first_block;
214 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
216 ret = add_swap_extent(sis, page_no, 1, first_block);
221 probe_block += blocks_per_page;
226 *span = 1 + highest_block - lowest_block;
228 page_no = 1; /* force Empty message */
230 sis->pages = page_no - 1;
231 sis->highest_bit = page_no - 1;
235 pr_err("swapon: swapfile has holes\n");
241 * We may have stale swap cache pages in memory: notice
242 * them here and get rid of the unnecessary final write.
244 int swap_writepage(struct page *page, struct writeback_control *wbc)
248 if (try_to_free_swap(page)) {
252 if (frontswap_store(page) == 0) {
253 set_page_writeback(page);
255 end_page_writeback(page);
258 ret = __swap_writepage(page, wbc, end_swap_bio_write);
263 static inline void count_swpout_vm_event(struct page *page)
265 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
266 if (unlikely(PageTransHuge(page)))
267 count_vm_event(THP_SWPOUT);
269 count_vm_events(PSWPOUT, hpage_nr_pages(page));
272 int __swap_writepage(struct page *page, struct writeback_control *wbc,
273 bio_end_io_t end_write_func)
277 struct swap_info_struct *sis = page_swap_info(page);
279 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
280 if (sis->flags & SWP_FILE) {
282 struct file *swap_file = sis->swap_file;
283 struct address_space *mapping = swap_file->f_mapping;
284 struct bio_vec bv = {
289 struct iov_iter from;
291 iov_iter_bvec(&from, ITER_BVEC | WRITE, &bv, 1, PAGE_SIZE);
292 init_sync_kiocb(&kiocb, swap_file);
293 kiocb.ki_pos = page_file_offset(page);
295 set_page_writeback(page);
297 ret = mapping->a_ops->direct_IO(&kiocb, &from);
298 if (ret == PAGE_SIZE) {
299 count_vm_event(PSWPOUT);
303 * In the case of swap-over-nfs, this can be a
304 * temporary failure if the system has limited
305 * memory for allocating transmit buffers.
306 * Mark the page dirty and avoid
307 * rotate_reclaimable_page but rate-limit the
308 * messages but do not flag PageError like
309 * the normal direct-to-bio case as it could
312 set_page_dirty(page);
313 ClearPageReclaim(page);
314 pr_err_ratelimited("Write error on dio swapfile (%llu)\n",
315 page_file_offset(page));
317 end_page_writeback(page);
321 ret = bdev_write_page(sis->bdev, map_swap_page(page, &sis->bdev),
324 count_swpout_vm_event(page);
329 bio = get_swap_bio(GFP_NOIO, page, end_write_func);
331 set_page_dirty(page);
336 bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc);
337 count_swpout_vm_event(page);
338 set_page_writeback(page);
345 int swap_readpage(struct page *page, bool do_poll)
349 struct swap_info_struct *sis = page_swap_info(page);
351 struct gendisk *disk;
353 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
354 VM_BUG_ON_PAGE(!PageLocked(page), page);
355 VM_BUG_ON_PAGE(PageUptodate(page), page);
356 if (frontswap_load(page) == 0) {
357 SetPageUptodate(page);
362 if (sis->flags & SWP_FILE) {
363 struct file *swap_file = sis->swap_file;
364 struct address_space *mapping = swap_file->f_mapping;
366 ret = mapping->a_ops->readpage(swap_file, page);
368 count_vm_event(PSWPIN);
372 ret = bdev_read_page(sis->bdev, map_swap_page(page, &sis->bdev), page);
374 if (trylock_page(page)) {
375 swap_slot_free_notify(page);
379 count_vm_event(PSWPIN);
384 bio = get_swap_bio(GFP_KERNEL, page, end_swap_bio_read);
392 * Keep this task valid during swap readpage because the oom killer may
393 * attempt to access it in the page fault retry time check.
395 get_task_struct(current);
396 bio->bi_private = current;
397 bio_set_op_attrs(bio, REQ_OP_READ, 0);
398 count_vm_event(PSWPIN);
400 qc = submit_bio(bio);
402 set_current_state(TASK_UNINTERRUPTIBLE);
403 if (!READ_ONCE(bio->bi_private))
406 if (!blk_mq_poll(disk->queue, qc))
409 __set_current_state(TASK_RUNNING);
416 int swap_set_page_dirty(struct page *page)
418 struct swap_info_struct *sis = page_swap_info(page);
420 if (sis->flags & SWP_FILE) {
421 struct address_space *mapping = sis->swap_file->f_mapping;
423 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
424 return mapping->a_ops->set_page_dirty(page);
426 return __set_page_dirty_no_writeback(page);