4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
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
14 #include <linux/kernel_stat.h>
15 #include <linux/gfp.h>
16 #include <linux/pagemap.h>
17 #include <linux/swap.h>
18 #include <linux/bio.h>
19 #include <linux/swapops.h>
20 #include <linux/buffer_head.h>
21 #include <linux/writeback.h>
22 #include <linux/frontswap.h>
23 #include <linux/aio.h>
24 #include <linux/blkdev.h>
25 #include <asm/pgtable.h>
27 static struct bio *get_swap_bio(gfp_t gfp_flags,
28 struct page *page, bio_end_io_t end_io)
32 bio = bio_alloc(gfp_flags, 1);
34 bio->bi_iter.bi_sector = map_swap_page(page, &bio->bi_bdev);
35 bio->bi_iter.bi_sector <<= PAGE_SHIFT - 9;
36 bio->bi_io_vec[0].bv_page = page;
37 bio->bi_io_vec[0].bv_len = PAGE_SIZE;
38 bio->bi_io_vec[0].bv_offset = 0;
40 bio->bi_iter.bi_size = PAGE_SIZE;
41 bio->bi_end_io = end_io;
46 void end_swap_bio_write(struct bio *bio, int err)
48 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
49 struct page *page = bio->bi_io_vec[0].bv_page;
54 * We failed to write the page out to swap-space.
55 * Re-dirty the page in order to avoid it being reclaimed.
56 * Also print a dire warning that things will go BAD (tm)
59 * Also clear PG_reclaim to avoid rotate_reclaimable_page()
62 printk(KERN_ALERT "Write-error on swap-device (%u:%u:%Lu)\n",
63 imajor(bio->bi_bdev->bd_inode),
64 iminor(bio->bi_bdev->bd_inode),
65 (unsigned long long)bio->bi_iter.bi_sector);
66 ClearPageReclaim(page);
68 end_page_writeback(page);
72 void end_swap_bio_read(struct bio *bio, int err)
74 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
75 struct page *page = bio->bi_io_vec[0].bv_page;
79 ClearPageUptodate(page);
80 printk(KERN_ALERT "Read-error on swap-device (%u:%u:%Lu)\n",
81 imajor(bio->bi_bdev->bd_inode),
82 iminor(bio->bi_bdev->bd_inode),
83 (unsigned long long)bio->bi_iter.bi_sector);
87 SetPageUptodate(page);
90 * There is no guarantee that the page is in swap cache - the software
91 * suspend code (at least) uses end_swap_bio_read() against a non-
92 * swapcache page. So we must check PG_swapcache before proceeding with
95 if (likely(PageSwapCache(page))) {
96 struct swap_info_struct *sis;
98 sis = page_swap_info(page);
99 if (sis->flags & SWP_BLKDEV) {
101 * The swap subsystem performs lazy swap slot freeing,
102 * expecting that the page will be swapped out again.
103 * So we can avoid an unnecessary write if the page
105 * This is good for real swap storage because we can
106 * reduce unnecessary I/O and enhance wear-leveling
107 * if an SSD is used as the as swap device.
108 * But if in-memory swap device (eg zram) is used,
109 * this causes a duplicated copy between uncompressed
110 * data in VM-owned memory and compressed data in
111 * zram-owned memory. So let's free zram-owned memory
112 * and make the VM-owned decompressed page *dirty*,
113 * so the page should be swapped out somewhere again if
114 * we again wish to reclaim it.
116 struct gendisk *disk = sis->bdev->bd_disk;
117 if (disk->fops->swap_slot_free_notify) {
119 unsigned long offset;
121 entry.val = page_private(page);
122 offset = swp_offset(entry);
125 disk->fops->swap_slot_free_notify(sis->bdev,
136 int generic_swapfile_activate(struct swap_info_struct *sis,
137 struct file *swap_file,
140 struct address_space *mapping = swap_file->f_mapping;
141 struct inode *inode = mapping->host;
142 unsigned blocks_per_page;
143 unsigned long page_no;
145 sector_t probe_block;
147 sector_t lowest_block = -1;
148 sector_t highest_block = 0;
152 blkbits = inode->i_blkbits;
153 blocks_per_page = PAGE_SIZE >> blkbits;
156 * Map all the blocks into the extent list. This code doesn't try
161 last_block = i_size_read(inode) >> blkbits;
162 while ((probe_block + blocks_per_page) <= last_block &&
163 page_no < sis->max) {
164 unsigned block_in_page;
165 sector_t first_block;
167 first_block = bmap(inode, probe_block);
168 if (first_block == 0)
172 * It must be PAGE_SIZE aligned on-disk
174 if (first_block & (blocks_per_page - 1)) {
179 for (block_in_page = 1; block_in_page < blocks_per_page;
183 block = bmap(inode, probe_block + block_in_page);
186 if (block != first_block + block_in_page) {
193 first_block >>= (PAGE_SHIFT - blkbits);
194 if (page_no) { /* exclude the header page */
195 if (first_block < lowest_block)
196 lowest_block = first_block;
197 if (first_block > highest_block)
198 highest_block = first_block;
202 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
204 ret = add_swap_extent(sis, page_no, 1, first_block);
209 probe_block += blocks_per_page;
214 *span = 1 + highest_block - lowest_block;
216 page_no = 1; /* force Empty message */
218 sis->pages = page_no - 1;
219 sis->highest_bit = page_no - 1;
223 printk(KERN_ERR "swapon: swapfile has holes\n");
229 * We may have stale swap cache pages in memory: notice
230 * them here and get rid of the unnecessary final write.
232 int swap_writepage(struct page *page, struct writeback_control *wbc)
236 if (try_to_free_swap(page)) {
240 if (frontswap_store(page) == 0) {
241 set_page_writeback(page);
243 end_page_writeback(page);
246 ret = __swap_writepage(page, wbc, end_swap_bio_write);
251 int __swap_writepage(struct page *page, struct writeback_control *wbc,
252 void (*end_write_func)(struct bio *, int))
255 int ret = 0, rw = WRITE;
256 struct swap_info_struct *sis = page_swap_info(page);
258 if (sis->flags & SWP_FILE) {
260 struct file *swap_file = sis->swap_file;
261 struct address_space *mapping = swap_file->f_mapping;
263 .iov_base = kmap(page),
264 .iov_len = PAGE_SIZE,
266 struct iov_iter from;
268 init_sync_kiocb(&kiocb, swap_file);
269 kiocb.ki_pos = page_file_offset(page);
270 kiocb.ki_nbytes = PAGE_SIZE;
271 iov_iter_init(&from, &iov, 1, PAGE_SIZE, 0);
273 set_page_writeback(page);
275 ret = mapping->a_ops->direct_IO(KERNEL_WRITE,
279 if (ret == PAGE_SIZE) {
280 count_vm_event(PSWPOUT);
284 * In the case of swap-over-nfs, this can be a
285 * temporary failure if the system has limited
286 * memory for allocating transmit buffers.
287 * Mark the page dirty and avoid
288 * rotate_reclaimable_page but rate-limit the
289 * messages but do not flag PageError like
290 * the normal direct-to-bio case as it could
293 set_page_dirty(page);
294 ClearPageReclaim(page);
295 pr_err_ratelimited("Write error on dio swapfile (%Lu)\n",
296 page_file_offset(page));
298 end_page_writeback(page);
302 bio = get_swap_bio(GFP_NOIO, page, end_write_func);
304 set_page_dirty(page);
309 if (wbc->sync_mode == WB_SYNC_ALL)
311 count_vm_event(PSWPOUT);
312 set_page_writeback(page);
319 int swap_readpage(struct page *page)
323 struct swap_info_struct *sis = page_swap_info(page);
325 VM_BUG_ON_PAGE(!PageLocked(page), page);
326 VM_BUG_ON_PAGE(PageUptodate(page), page);
327 if (frontswap_load(page) == 0) {
328 SetPageUptodate(page);
333 if (sis->flags & SWP_FILE) {
334 struct file *swap_file = sis->swap_file;
335 struct address_space *mapping = swap_file->f_mapping;
337 ret = mapping->a_ops->readpage(swap_file, page);
339 count_vm_event(PSWPIN);
343 bio = get_swap_bio(GFP_KERNEL, page, end_swap_bio_read);
349 count_vm_event(PSWPIN);
350 submit_bio(READ, bio);
355 int swap_set_page_dirty(struct page *page)
357 struct swap_info_struct *sis = page_swap_info(page);
359 if (sis->flags & SWP_FILE) {
360 struct address_space *mapping = sis->swap_file->f_mapping;
361 return mapping->a_ops->set_page_dirty(page);
363 return __set_page_dirty_no_writeback(page);