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_end_io = end_io;
37 bio_add_page(bio, page, PAGE_SIZE, 0);
38 BUG_ON(bio->bi_iter.bi_size != PAGE_SIZE);
43 void end_swap_bio_write(struct bio *bio)
45 struct page *page = bio->bi_io_vec[0].bv_page;
50 * We failed to write the page out to swap-space.
51 * Re-dirty the page in order to avoid it being reclaimed.
52 * Also print a dire warning that things will go BAD (tm)
55 * Also clear PG_reclaim to avoid rotate_reclaimable_page()
58 pr_alert("Write-error on swap-device (%u:%u:%llu)\n",
59 imajor(bio->bi_bdev->bd_inode),
60 iminor(bio->bi_bdev->bd_inode),
61 (unsigned long long)bio->bi_iter.bi_sector);
62 ClearPageReclaim(page);
64 end_page_writeback(page);
68 static void swap_slot_free_notify(struct page *page)
70 struct swap_info_struct *sis;
74 * There is no guarantee that the page is in swap cache - the software
75 * suspend code (at least) uses end_swap_bio_read() against a non-
76 * swapcache page. So we must check PG_swapcache before proceeding with
79 if (unlikely(!PageSwapCache(page)))
82 sis = page_swap_info(page);
83 if (!(sis->flags & SWP_BLKDEV))
87 * The swap subsystem performs lazy swap slot freeing,
88 * expecting that the page will be swapped out again.
89 * So we can avoid an unnecessary write if the page
91 * This is good for real swap storage because we can
92 * reduce unnecessary I/O and enhance wear-leveling
93 * if an SSD is used as the as swap device.
94 * But if in-memory swap device (eg zram) is used,
95 * this causes a duplicated copy between uncompressed
96 * data in VM-owned memory and compressed data in
97 * zram-owned memory. So let's free zram-owned memory
98 * and make the VM-owned decompressed page *dirty*,
99 * so the page should be swapped out somewhere again if
100 * we again wish to reclaim it.
102 disk = sis->bdev->bd_disk;
103 if (disk->fops->swap_slot_free_notify) {
105 unsigned long offset;
107 entry.val = page_private(page);
108 offset = swp_offset(entry);
111 disk->fops->swap_slot_free_notify(sis->bdev,
116 static void end_swap_bio_read(struct bio *bio)
118 struct page *page = bio->bi_io_vec[0].bv_page;
122 ClearPageUptodate(page);
123 pr_alert("Read-error on swap-device (%u:%u:%llu)\n",
124 imajor(bio->bi_bdev->bd_inode),
125 iminor(bio->bi_bdev->bd_inode),
126 (unsigned long long)bio->bi_iter.bi_sector);
130 SetPageUptodate(page);
131 swap_slot_free_notify(page);
137 int generic_swapfile_activate(struct swap_info_struct *sis,
138 struct file *swap_file,
141 struct address_space *mapping = swap_file->f_mapping;
142 struct inode *inode = mapping->host;
143 unsigned blocks_per_page;
144 unsigned long page_no;
146 sector_t probe_block;
148 sector_t lowest_block = -1;
149 sector_t highest_block = 0;
153 blkbits = inode->i_blkbits;
154 blocks_per_page = PAGE_SIZE >> blkbits;
157 * Map all the blocks into the extent list. This code doesn't try
162 last_block = i_size_read(inode) >> blkbits;
163 while ((probe_block + blocks_per_page) <= last_block &&
164 page_no < sis->max) {
165 unsigned block_in_page;
166 sector_t first_block;
170 first_block = bmap(inode, probe_block);
171 if (first_block == 0)
175 * It must be PAGE_SIZE aligned on-disk
177 if (first_block & (blocks_per_page - 1)) {
182 for (block_in_page = 1; block_in_page < blocks_per_page;
186 block = bmap(inode, probe_block + block_in_page);
189 if (block != first_block + block_in_page) {
196 first_block >>= (PAGE_SHIFT - blkbits);
197 if (page_no) { /* exclude the header page */
198 if (first_block < lowest_block)
199 lowest_block = first_block;
200 if (first_block > highest_block)
201 highest_block = first_block;
205 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
207 ret = add_swap_extent(sis, page_no, 1, first_block);
212 probe_block += blocks_per_page;
217 *span = 1 + highest_block - lowest_block;
219 page_no = 1; /* force Empty message */
221 sis->pages = page_no - 1;
222 sis->highest_bit = page_no - 1;
226 pr_err("swapon: swapfile has holes\n");
232 * We may have stale swap cache pages in memory: notice
233 * them here and get rid of the unnecessary final write.
235 int swap_writepage(struct page *page, struct writeback_control *wbc)
239 if (try_to_free_swap(page)) {
243 if (frontswap_store(page) == 0) {
244 set_page_writeback(page);
246 end_page_writeback(page);
249 ret = __swap_writepage(page, wbc, end_swap_bio_write);
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 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
262 if (sis->flags & SWP_FILE) {
264 struct file *swap_file = sis->swap_file;
265 struct address_space *mapping = swap_file->f_mapping;
266 struct bio_vec bv = {
271 struct iov_iter from;
273 iov_iter_bvec(&from, ITER_BVEC | WRITE, &bv, 1, PAGE_SIZE);
274 init_sync_kiocb(&kiocb, swap_file);
275 kiocb.ki_pos = page_file_offset(page);
277 set_page_writeback(page);
279 ret = mapping->a_ops->direct_IO(&kiocb, &from);
280 if (ret == PAGE_SIZE) {
281 count_vm_event(PSWPOUT);
285 * In the case of swap-over-nfs, this can be a
286 * temporary failure if the system has limited
287 * memory for allocating transmit buffers.
288 * Mark the page dirty and avoid
289 * rotate_reclaimable_page but rate-limit the
290 * messages but do not flag PageError like
291 * the normal direct-to-bio case as it could
294 set_page_dirty(page);
295 ClearPageReclaim(page);
296 pr_err_ratelimited("Write error on dio swapfile (%llu)\n",
297 page_file_offset(page));
299 end_page_writeback(page);
303 ret = bdev_write_page(sis->bdev, map_swap_page(page, &sis->bdev),
306 count_vm_event(PSWPOUT);
311 bio = get_swap_bio(GFP_NOIO, page, end_write_func);
313 set_page_dirty(page);
318 if (wbc->sync_mode == WB_SYNC_ALL)
319 bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC);
321 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
322 count_vm_event(PSWPOUT);
323 set_page_writeback(page);
330 int swap_readpage(struct page *page)
334 struct swap_info_struct *sis = page_swap_info(page);
336 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
337 VM_BUG_ON_PAGE(!PageLocked(page), page);
338 VM_BUG_ON_PAGE(PageUptodate(page), page);
339 if (frontswap_load(page) == 0) {
340 SetPageUptodate(page);
345 if (sis->flags & SWP_FILE) {
346 struct file *swap_file = sis->swap_file;
347 struct address_space *mapping = swap_file->f_mapping;
349 ret = mapping->a_ops->readpage(swap_file, page);
351 count_vm_event(PSWPIN);
355 ret = bdev_read_page(sis->bdev, map_swap_page(page, &sis->bdev), page);
357 if (trylock_page(page)) {
358 swap_slot_free_notify(page);
362 count_vm_event(PSWPIN);
367 bio = get_swap_bio(GFP_KERNEL, page, end_swap_bio_read);
373 bio_set_op_attrs(bio, REQ_OP_READ, 0);
374 count_vm_event(PSWPIN);
380 int swap_set_page_dirty(struct page *page)
382 struct swap_info_struct *sis = page_swap_info(page);
384 if (sis->flags & SWP_FILE) {
385 struct address_space *mapping = sis->swap_file->f_mapping;
387 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
388 return mapping->a_ops->set_page_dirty(page);
390 return __set_page_dirty_no_writeback(page);