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/blkdev.h>
24 #include <linux/uio.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)
48 struct page *page = bio->bi_io_vec[0].bv_page;
53 * We failed to write the page out to swap-space.
54 * Re-dirty the page in order to avoid it being reclaimed.
55 * Also print a dire warning that things will go BAD (tm)
58 * Also clear PG_reclaim to avoid rotate_reclaimable_page()
61 printk(KERN_ALERT "Write-error on swap-device (%u:%u:%Lu)\n",
62 imajor(bio->bi_bdev->bd_inode),
63 iminor(bio->bi_bdev->bd_inode),
64 (unsigned long long)bio->bi_iter.bi_sector);
65 ClearPageReclaim(page);
67 end_page_writeback(page);
71 static void end_swap_bio_read(struct bio *bio)
73 struct page *page = bio->bi_io_vec[0].bv_page;
77 ClearPageUptodate(page);
78 printk(KERN_ALERT "Read-error on swap-device (%u:%u:%Lu)\n",
79 imajor(bio->bi_bdev->bd_inode),
80 iminor(bio->bi_bdev->bd_inode),
81 (unsigned long long)bio->bi_iter.bi_sector);
85 SetPageUptodate(page);
88 * There is no guarantee that the page is in swap cache - the software
89 * suspend code (at least) uses end_swap_bio_read() against a non-
90 * swapcache page. So we must check PG_swapcache before proceeding with
93 if (likely(PageSwapCache(page))) {
94 struct swap_info_struct *sis;
96 sis = page_swap_info(page);
97 if (sis->flags & SWP_BLKDEV) {
99 * The swap subsystem performs lazy swap slot freeing,
100 * expecting that the page will be swapped out again.
101 * So we can avoid an unnecessary write if the page
103 * This is good for real swap storage because we can
104 * reduce unnecessary I/O and enhance wear-leveling
105 * if an SSD is used as the as swap device.
106 * But if in-memory swap device (eg zram) is used,
107 * this causes a duplicated copy between uncompressed
108 * data in VM-owned memory and compressed data in
109 * zram-owned memory. So let's free zram-owned memory
110 * and make the VM-owned decompressed page *dirty*,
111 * so the page should be swapped out somewhere again if
112 * we again wish to reclaim it.
114 struct gendisk *disk = sis->bdev->bd_disk;
115 if (disk->fops->swap_slot_free_notify) {
117 unsigned long offset;
119 entry.val = page_private(page);
120 offset = swp_offset(entry);
123 disk->fops->swap_slot_free_notify(sis->bdev,
134 int generic_swapfile_activate(struct swap_info_struct *sis,
135 struct file *swap_file,
138 struct address_space *mapping = swap_file->f_mapping;
139 struct inode *inode = mapping->host;
140 unsigned blocks_per_page;
141 unsigned long page_no;
143 sector_t probe_block;
145 sector_t lowest_block = -1;
146 sector_t highest_block = 0;
150 blkbits = inode->i_blkbits;
151 blocks_per_page = PAGE_SIZE >> blkbits;
154 * Map all the blocks into the extent list. This code doesn't try
159 last_block = i_size_read(inode) >> blkbits;
160 while ((probe_block + blocks_per_page) <= last_block &&
161 page_no < sis->max) {
162 unsigned block_in_page;
163 sector_t first_block;
165 first_block = bmap(inode, probe_block);
166 if (first_block == 0)
170 * It must be PAGE_SIZE aligned on-disk
172 if (first_block & (blocks_per_page - 1)) {
177 for (block_in_page = 1; block_in_page < blocks_per_page;
181 block = bmap(inode, probe_block + block_in_page);
184 if (block != first_block + block_in_page) {
191 first_block >>= (PAGE_SHIFT - blkbits);
192 if (page_no) { /* exclude the header page */
193 if (first_block < lowest_block)
194 lowest_block = first_block;
195 if (first_block > highest_block)
196 highest_block = first_block;
200 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
202 ret = add_swap_extent(sis, page_no, 1, first_block);
207 probe_block += blocks_per_page;
212 *span = 1 + highest_block - lowest_block;
214 page_no = 1; /* force Empty message */
216 sis->pages = page_no - 1;
217 sis->highest_bit = page_no - 1;
221 printk(KERN_ERR "swapon: swapfile has holes\n");
227 * We may have stale swap cache pages in memory: notice
228 * them here and get rid of the unnecessary final write.
230 int swap_writepage(struct page *page, struct writeback_control *wbc)
234 if (try_to_free_swap(page)) {
238 if (frontswap_store(page) == 0) {
239 set_page_writeback(page);
241 end_page_writeback(page);
244 ret = __swap_writepage(page, wbc, end_swap_bio_write);
249 static sector_t swap_page_sector(struct page *page)
251 return (sector_t)__page_file_index(page) << (PAGE_CACHE_SHIFT - 9);
254 int __swap_writepage(struct page *page, struct writeback_control *wbc,
255 bio_end_io_t end_write_func)
259 struct swap_info_struct *sis = page_swap_info(page);
261 if (sis->flags & SWP_FILE) {
263 struct file *swap_file = sis->swap_file;
264 struct address_space *mapping = swap_file->f_mapping;
265 struct bio_vec bv = {
270 struct iov_iter from;
272 iov_iter_bvec(&from, ITER_BVEC | WRITE, &bv, 1, PAGE_SIZE);
273 init_sync_kiocb(&kiocb, swap_file);
274 kiocb.ki_pos = page_file_offset(page);
276 set_page_writeback(page);
278 ret = mapping->a_ops->direct_IO(&kiocb, &from, kiocb.ki_pos);
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 ret = bdev_write_page(sis->bdev, swap_page_sector(page), page, wbc);
304 count_vm_event(PSWPOUT);
309 bio = get_swap_bio(GFP_NOIO, page, end_write_func);
311 set_page_dirty(page);
316 if (wbc->sync_mode == WB_SYNC_ALL)
318 count_vm_event(PSWPOUT);
319 set_page_writeback(page);
326 int swap_readpage(struct page *page)
330 struct swap_info_struct *sis = page_swap_info(page);
332 VM_BUG_ON_PAGE(!PageLocked(page), page);
333 VM_BUG_ON_PAGE(PageUptodate(page), page);
334 if (frontswap_load(page) == 0) {
335 SetPageUptodate(page);
340 if (sis->flags & SWP_FILE) {
341 struct file *swap_file = sis->swap_file;
342 struct address_space *mapping = swap_file->f_mapping;
344 ret = mapping->a_ops->readpage(swap_file, page);
346 count_vm_event(PSWPIN);
350 ret = bdev_read_page(sis->bdev, swap_page_sector(page), page);
352 count_vm_event(PSWPIN);
357 bio = get_swap_bio(GFP_KERNEL, page, end_swap_bio_read);
363 count_vm_event(PSWPIN);
364 submit_bio(READ, bio);
369 int swap_set_page_dirty(struct page *page)
371 struct swap_info_struct *sis = page_swap_info(page);
373 if (sis->flags & SWP_FILE) {
374 struct address_space *mapping = sis->swap_file->f_mapping;
375 return mapping->a_ops->set_page_dirty(page);
377 return __set_page_dirty_no_writeback(page);