1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * linux/mm/page_io.c
4 *
5 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
6 *
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
12 */
13
14 #include <linux/mm.h>
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>
28
get_swap_bio(gfp_t gfp_flags,struct page * page,bio_end_io_t end_io)29 static struct bio *get_swap_bio(gfp_t gfp_flags,
30 struct page *page, bio_end_io_t end_io)
31 {
32 struct bio *bio;
33
34 bio = bio_alloc(gfp_flags, 1);
35 if (bio) {
36 struct block_device *bdev;
37
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;
42
43 bio_add_page(bio, page, PAGE_SIZE * hpage_nr_pages(page), 0);
44 }
45 return bio;
46 }
47
end_swap_bio_write(struct bio * bio)48 void end_swap_bio_write(struct bio *bio)
49 {
50 struct page *page = bio_first_page_all(bio);
51
52 if (bio->bi_status) {
53 SetPageError(page);
54 /*
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)
58 * very quickly.
59 *
60 * Also clear PG_reclaim to avoid rotate_reclaimable_page()
61 */
62 set_page_dirty(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);
67 }
68 end_page_writeback(page);
69 bio_put(bio);
70 }
71
swap_slot_free_notify(struct page * page)72 static void swap_slot_free_notify(struct page *page)
73 {
74 struct swap_info_struct *sis;
75 struct gendisk *disk;
76 swp_entry_t entry;
77
78 /*
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
82 * this optimization.
83 */
84 if (unlikely(!PageSwapCache(page)))
85 return;
86
87 sis = page_swap_info(page);
88 if (!(sis->flags & SWP_BLKDEV))
89 return;
90
91 /*
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
95 * isn't redirtied.
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.
106 */
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;
111
112 offset = swp_offset(entry);
113
114 SetPageDirty(page);
115 disk->fops->swap_slot_free_notify(sis->bdev,
116 offset);
117 }
118 }
119
end_swap_bio_read(struct bio * bio)120 static void end_swap_bio_read(struct bio *bio)
121 {
122 struct page *page = bio_first_page_all(bio);
123 struct task_struct *waiter = bio->bi_private;
124
125 if (bio->bi_status) {
126 SetPageError(page);
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);
131 goto out;
132 }
133
134 SetPageUptodate(page);
135 swap_slot_free_notify(page);
136 out:
137 unlock_page(page);
138 WRITE_ONCE(bio->bi_private, NULL);
139 bio_put(bio);
140 if (waiter) {
141 blk_wake_io_task(waiter);
142 put_task_struct(waiter);
143 }
144 }
145
generic_swapfile_activate(struct swap_info_struct * sis,struct file * swap_file,sector_t * span)146 int generic_swapfile_activate(struct swap_info_struct *sis,
147 struct file *swap_file,
148 sector_t *span)
149 {
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;
154 unsigned blkbits;
155 sector_t probe_block;
156 sector_t last_block;
157 sector_t lowest_block = -1;
158 sector_t highest_block = 0;
159 int nr_extents = 0;
160 int ret;
161
162 blkbits = inode->i_blkbits;
163 blocks_per_page = PAGE_SIZE >> blkbits;
164
165 /*
166 * Map all the blocks into the extent tree. This code doesn't try
167 * to be very smart.
168 */
169 probe_block = 0;
170 page_no = 0;
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;
176
177 cond_resched();
178
179 first_block = bmap(inode, probe_block);
180 if (first_block == 0)
181 goto bad_bmap;
182
183 /*
184 * It must be PAGE_SIZE aligned on-disk
185 */
186 if (first_block & (blocks_per_page - 1)) {
187 probe_block++;
188 goto reprobe;
189 }
190
191 for (block_in_page = 1; block_in_page < blocks_per_page;
192 block_in_page++) {
193 sector_t block;
194
195 block = bmap(inode, probe_block + block_in_page);
196 if (block == 0)
197 goto bad_bmap;
198 if (block != first_block + block_in_page) {
199 /* Discontiguity */
200 probe_block++;
201 goto reprobe;
202 }
203 }
204
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;
211 }
212
213 /*
214 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
215 */
216 ret = add_swap_extent(sis, page_no, 1, first_block);
217 if (ret < 0)
218 goto out;
219 nr_extents += ret;
220 page_no++;
221 probe_block += blocks_per_page;
222 reprobe:
223 continue;
224 }
225 ret = nr_extents;
226 *span = 1 + highest_block - lowest_block;
227 if (page_no == 0)
228 page_no = 1; /* force Empty message */
229 sis->max = page_no;
230 sis->pages = page_no - 1;
231 sis->highest_bit = page_no - 1;
232 out:
233 return ret;
234 bad_bmap:
235 pr_err("swapon: swapfile has holes\n");
236 ret = -EINVAL;
237 goto out;
238 }
239
240 /*
241 * We may have stale swap cache pages in memory: notice
242 * them here and get rid of the unnecessary final write.
243 */
swap_writepage(struct page * page,struct writeback_control * wbc)244 int swap_writepage(struct page *page, struct writeback_control *wbc)
245 {
246 int ret = 0;
247
248 if (try_to_free_swap(page)) {
249 unlock_page(page);
250 goto out;
251 }
252 if (frontswap_store(page) == 0) {
253 set_page_writeback(page);
254 unlock_page(page);
255 end_page_writeback(page);
256 goto out;
257 }
258 ret = __swap_writepage(page, wbc, end_swap_bio_write);
259 out:
260 return ret;
261 }
262
swap_page_sector(struct page * page)263 static sector_t swap_page_sector(struct page *page)
264 {
265 return (sector_t)__page_file_index(page) << (PAGE_SHIFT - 9);
266 }
267
count_swpout_vm_event(struct page * page)268 static inline void count_swpout_vm_event(struct page *page)
269 {
270 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
271 if (unlikely(PageTransHuge(page)))
272 count_vm_event(THP_SWPOUT);
273 #endif
274 count_vm_events(PSWPOUT, hpage_nr_pages(page));
275 }
276
__swap_writepage(struct page * page,struct writeback_control * wbc,bio_end_io_t end_write_func)277 int __swap_writepage(struct page *page, struct writeback_control *wbc,
278 bio_end_io_t end_write_func)
279 {
280 struct bio *bio;
281 int ret;
282 struct swap_info_struct *sis = page_swap_info(page);
283
284 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
285 if (sis->flags & SWP_FS) {
286 struct kiocb kiocb;
287 struct file *swap_file = sis->swap_file;
288 struct address_space *mapping = swap_file->f_mapping;
289 struct bio_vec bv = {
290 .bv_page = page,
291 .bv_len = PAGE_SIZE,
292 .bv_offset = 0
293 };
294 struct iov_iter from;
295
296 iov_iter_bvec(&from, WRITE, &bv, 1, PAGE_SIZE);
297 init_sync_kiocb(&kiocb, swap_file);
298 kiocb.ki_pos = page_file_offset(page);
299
300 set_page_writeback(page);
301 unlock_page(page);
302 ret = mapping->a_ops->direct_IO(&kiocb, &from);
303 if (ret == PAGE_SIZE) {
304 count_vm_event(PSWPOUT);
305 ret = 0;
306 } else {
307 /*
308 * In the case of swap-over-nfs, this can be a
309 * temporary failure if the system has limited
310 * memory for allocating transmit buffers.
311 * Mark the page dirty and avoid
312 * rotate_reclaimable_page but rate-limit the
313 * messages but do not flag PageError like
314 * the normal direct-to-bio case as it could
315 * be temporary.
316 */
317 set_page_dirty(page);
318 ClearPageReclaim(page);
319 pr_err_ratelimited("Write error on dio swapfile (%llu)\n",
320 page_file_offset(page));
321 }
322 end_page_writeback(page);
323 return ret;
324 }
325
326 ret = bdev_write_page(sis->bdev, swap_page_sector(page), page, wbc);
327 if (!ret) {
328 count_swpout_vm_event(page);
329 return 0;
330 }
331
332 ret = 0;
333 bio = get_swap_bio(GFP_NOIO, page, end_write_func);
334 if (bio == NULL) {
335 set_page_dirty(page);
336 unlock_page(page);
337 ret = -ENOMEM;
338 goto out;
339 }
340 bio->bi_opf = REQ_OP_WRITE | REQ_SWAP | wbc_to_write_flags(wbc);
341 bio_associate_blkg_from_page(bio, page);
342 count_swpout_vm_event(page);
343 set_page_writeback(page);
344 unlock_page(page);
345 submit_bio(bio);
346 out:
347 return ret;
348 }
349
swap_readpage(struct page * page,bool synchronous)350 int swap_readpage(struct page *page, bool synchronous)
351 {
352 struct bio *bio;
353 int ret = 0;
354 struct swap_info_struct *sis = page_swap_info(page);
355 blk_qc_t qc;
356 struct gendisk *disk;
357
358 VM_BUG_ON_PAGE(!PageSwapCache(page) && !synchronous, page);
359 VM_BUG_ON_PAGE(!PageLocked(page), page);
360 VM_BUG_ON_PAGE(PageUptodate(page), page);
361 if (frontswap_load(page) == 0) {
362 SetPageUptodate(page);
363 unlock_page(page);
364 goto out;
365 }
366
367 if (sis->flags & SWP_FS) {
368 struct file *swap_file = sis->swap_file;
369 struct address_space *mapping = swap_file->f_mapping;
370
371 ret = mapping->a_ops->readpage(swap_file, page);
372 if (!ret)
373 count_vm_event(PSWPIN);
374 return ret;
375 }
376
377 ret = bdev_read_page(sis->bdev, swap_page_sector(page), page);
378 if (!ret) {
379 if (trylock_page(page)) {
380 swap_slot_free_notify(page);
381 unlock_page(page);
382 }
383
384 count_vm_event(PSWPIN);
385 return 0;
386 }
387
388 ret = 0;
389 bio = get_swap_bio(GFP_KERNEL, page, end_swap_bio_read);
390 if (bio == NULL) {
391 unlock_page(page);
392 ret = -ENOMEM;
393 goto out;
394 }
395 disk = bio->bi_disk;
396 /*
397 * Keep this task valid during swap readpage because the oom killer may
398 * attempt to access it in the page fault retry time check.
399 */
400 bio_set_op_attrs(bio, REQ_OP_READ, 0);
401 if (synchronous) {
402 bio->bi_opf |= REQ_HIPRI;
403 get_task_struct(current);
404 bio->bi_private = current;
405 }
406 count_vm_event(PSWPIN);
407 bio_get(bio);
408 qc = submit_bio(bio);
409 while (synchronous) {
410 set_current_state(TASK_UNINTERRUPTIBLE);
411 if (!READ_ONCE(bio->bi_private))
412 break;
413
414 if (!blk_poll(disk->queue, qc, true))
415 io_schedule();
416 }
417 __set_current_state(TASK_RUNNING);
418 bio_put(bio);
419
420 out:
421 return ret;
422 }
423
swap_set_page_dirty(struct page * page)424 int swap_set_page_dirty(struct page *page)
425 {
426 struct swap_info_struct *sis = page_swap_info(page);
427
428 if (sis->flags & SWP_FS) {
429 struct address_space *mapping = sis->swap_file->f_mapping;
430
431 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
432 return mapping->a_ops->set_page_dirty(page);
433 } else {
434 return __set_page_dirty_no_writeback(page);
435 }
436 }
437
