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/psi.h>
26 #include <linux/uio.h>
27 #include <linux/sched/task.h>
28
end_swap_bio_write(struct bio * bio)29 void end_swap_bio_write(struct bio *bio)
30 {
31 struct page *page = bio_first_page_all(bio);
32
33 if (bio->bi_status) {
34 SetPageError(page);
35 /*
36 * We failed to write the page out to swap-space.
37 * Re-dirty the page in order to avoid it being reclaimed.
38 * Also print a dire warning that things will go BAD (tm)
39 * very quickly.
40 *
41 * Also clear PG_reclaim to avoid rotate_reclaimable_page()
42 */
43 set_page_dirty(page);
44 pr_alert_ratelimited("Write-error on swap-device (%u:%u:%llu)\n",
45 MAJOR(bio_dev(bio)), MINOR(bio_dev(bio)),
46 (unsigned long long)bio->bi_iter.bi_sector);
47 ClearPageReclaim(page);
48 }
49 end_page_writeback(page);
50 bio_put(bio);
51 }
52
end_swap_bio_read(struct bio * bio)53 static void end_swap_bio_read(struct bio *bio)
54 {
55 struct page *page = bio_first_page_all(bio);
56 struct task_struct *waiter = bio->bi_private;
57
58 if (bio->bi_status) {
59 SetPageError(page);
60 ClearPageUptodate(page);
61 pr_alert_ratelimited("Read-error on swap-device (%u:%u:%llu)\n",
62 MAJOR(bio_dev(bio)), MINOR(bio_dev(bio)),
63 (unsigned long long)bio->bi_iter.bi_sector);
64 goto out;
65 }
66
67 SetPageUptodate(page);
68 out:
69 unlock_page(page);
70 WRITE_ONCE(bio->bi_private, NULL);
71 bio_put(bio);
72 if (waiter) {
73 blk_wake_io_task(waiter);
74 put_task_struct(waiter);
75 }
76 }
77
generic_swapfile_activate(struct swap_info_struct * sis,struct file * swap_file,sector_t * span)78 int generic_swapfile_activate(struct swap_info_struct *sis,
79 struct file *swap_file,
80 sector_t *span)
81 {
82 struct address_space *mapping = swap_file->f_mapping;
83 struct inode *inode = mapping->host;
84 unsigned blocks_per_page;
85 unsigned long page_no;
86 unsigned blkbits;
87 sector_t probe_block;
88 sector_t last_block;
89 sector_t lowest_block = -1;
90 sector_t highest_block = 0;
91 int nr_extents = 0;
92 int ret;
93
94 blkbits = inode->i_blkbits;
95 blocks_per_page = PAGE_SIZE >> blkbits;
96
97 /*
98 * Map all the blocks into the extent tree. This code doesn't try
99 * to be very smart.
100 */
101 probe_block = 0;
102 page_no = 0;
103 last_block = i_size_read(inode) >> blkbits;
104 while ((probe_block + blocks_per_page) <= last_block &&
105 page_no < sis->max) {
106 unsigned block_in_page;
107 sector_t first_block;
108
109 cond_resched();
110
111 first_block = probe_block;
112 ret = bmap(inode, &first_block);
113 if (ret || !first_block)
114 goto bad_bmap;
115
116 /*
117 * It must be PAGE_SIZE aligned on-disk
118 */
119 if (first_block & (blocks_per_page - 1)) {
120 probe_block++;
121 goto reprobe;
122 }
123
124 for (block_in_page = 1; block_in_page < blocks_per_page;
125 block_in_page++) {
126 sector_t block;
127
128 block = probe_block + block_in_page;
129 ret = bmap(inode, &block);
130 if (ret || !block)
131 goto bad_bmap;
132
133 if (block != first_block + block_in_page) {
134 /* Discontiguity */
135 probe_block++;
136 goto reprobe;
137 }
138 }
139
140 first_block >>= (PAGE_SHIFT - blkbits);
141 if (page_no) { /* exclude the header page */
142 if (first_block < lowest_block)
143 lowest_block = first_block;
144 if (first_block > highest_block)
145 highest_block = first_block;
146 }
147
148 /*
149 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
150 */
151 ret = add_swap_extent(sis, page_no, 1, first_block);
152 if (ret < 0)
153 goto out;
154 nr_extents += ret;
155 page_no++;
156 probe_block += blocks_per_page;
157 reprobe:
158 continue;
159 }
160 ret = nr_extents;
161 *span = 1 + highest_block - lowest_block;
162 if (page_no == 0)
163 page_no = 1; /* force Empty message */
164 sis->max = page_no;
165 sis->pages = page_no - 1;
166 sis->highest_bit = page_no - 1;
167 out:
168 return ret;
169 bad_bmap:
170 pr_err("swapon: swapfile has holes\n");
171 ret = -EINVAL;
172 goto out;
173 }
174
175 /*
176 * We may have stale swap cache pages in memory: notice
177 * them here and get rid of the unnecessary final write.
178 */
swap_writepage(struct page * page,struct writeback_control * wbc)179 int swap_writepage(struct page *page, struct writeback_control *wbc)
180 {
181 int ret = 0;
182
183 if (try_to_free_swap(page)) {
184 unlock_page(page);
185 goto out;
186 }
187 /*
188 * Arch code may have to preserve more data than just the page
189 * contents, e.g. memory tags.
190 */
191 ret = arch_prepare_to_swap(page);
192 if (ret) {
193 set_page_dirty(page);
194 unlock_page(page);
195 goto out;
196 }
197 if (frontswap_store(page) == 0) {
198 set_page_writeback(page);
199 unlock_page(page);
200 end_page_writeback(page);
201 goto out;
202 }
203 ret = __swap_writepage(page, wbc, end_swap_bio_write);
204 out:
205 return ret;
206 }
207
count_swpout_vm_event(struct page * page)208 static inline void count_swpout_vm_event(struct page *page)
209 {
210 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
211 if (unlikely(PageTransHuge(page)))
212 count_vm_event(THP_SWPOUT);
213 #endif
214 count_vm_events(PSWPOUT, thp_nr_pages(page));
215 }
216
217 #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
bio_associate_blkg_from_page(struct bio * bio,struct page * page)218 static void bio_associate_blkg_from_page(struct bio *bio, struct page *page)
219 {
220 struct cgroup_subsys_state *css;
221 struct mem_cgroup *memcg;
222
223 memcg = page_memcg(page);
224 if (!memcg)
225 return;
226
227 rcu_read_lock();
228 css = cgroup_e_css(memcg->css.cgroup, &io_cgrp_subsys);
229 bio_associate_blkg_from_css(bio, css);
230 rcu_read_unlock();
231 }
232 #else
233 #define bio_associate_blkg_from_page(bio, page) do { } while (0)
234 #endif /* CONFIG_MEMCG && CONFIG_BLK_CGROUP */
235
__swap_writepage(struct page * page,struct writeback_control * wbc,bio_end_io_t end_write_func)236 int __swap_writepage(struct page *page, struct writeback_control *wbc,
237 bio_end_io_t end_write_func)
238 {
239 struct bio *bio;
240 int ret;
241 struct swap_info_struct *sis = page_swap_info(page);
242
243 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
244 if (data_race(sis->flags & SWP_FS_OPS)) {
245 struct kiocb kiocb;
246 struct file *swap_file = sis->swap_file;
247 struct address_space *mapping = swap_file->f_mapping;
248 struct bio_vec bv = {
249 .bv_page = page,
250 .bv_len = PAGE_SIZE,
251 .bv_offset = 0
252 };
253 struct iov_iter from;
254
255 iov_iter_bvec(&from, WRITE, &bv, 1, PAGE_SIZE);
256 init_sync_kiocb(&kiocb, swap_file);
257 kiocb.ki_pos = page_file_offset(page);
258
259 set_page_writeback(page);
260 unlock_page(page);
261 ret = mapping->a_ops->direct_IO(&kiocb, &from);
262 if (ret == PAGE_SIZE) {
263 count_vm_event(PSWPOUT);
264 ret = 0;
265 } else {
266 /*
267 * In the case of swap-over-nfs, this can be a
268 * temporary failure if the system has limited
269 * memory for allocating transmit buffers.
270 * Mark the page dirty and avoid
271 * rotate_reclaimable_page but rate-limit the
272 * messages but do not flag PageError like
273 * the normal direct-to-bio case as it could
274 * be temporary.
275 */
276 set_page_dirty(page);
277 ClearPageReclaim(page);
278 pr_err_ratelimited("Write error on dio swapfile (%llu)\n",
279 page_file_offset(page));
280 }
281 end_page_writeback(page);
282 return ret;
283 }
284
285 ret = bdev_write_page(sis->bdev, swap_page_sector(page), page, wbc);
286 if (!ret) {
287 count_swpout_vm_event(page);
288 return 0;
289 }
290
291 bio = bio_alloc(GFP_NOIO, 1);
292 bio_set_dev(bio, sis->bdev);
293 bio->bi_iter.bi_sector = swap_page_sector(page);
294 bio->bi_opf = REQ_OP_WRITE | REQ_SWAP | wbc_to_write_flags(wbc);
295 bio->bi_end_io = end_write_func;
296 bio_add_page(bio, page, thp_size(page), 0);
297
298 bio_associate_blkg_from_page(bio, page);
299 count_swpout_vm_event(page);
300 set_page_writeback(page);
301 unlock_page(page);
302 submit_bio(bio);
303
304 return 0;
305 }
306
swap_readpage(struct page * page,bool synchronous)307 int swap_readpage(struct page *page, bool synchronous)
308 {
309 struct bio *bio;
310 int ret = 0;
311 struct swap_info_struct *sis = page_swap_info(page);
312 blk_qc_t qc;
313 struct gendisk *disk;
314 unsigned long pflags;
315
316 VM_BUG_ON_PAGE(!PageSwapCache(page) && !synchronous, page);
317 VM_BUG_ON_PAGE(!PageLocked(page), page);
318 VM_BUG_ON_PAGE(PageUptodate(page), page);
319
320 /*
321 * Count submission time as memory stall. When the device is congested,
322 * or the submitting cgroup IO-throttled, submission can be a
323 * significant part of overall IO time.
324 */
325 psi_memstall_enter(&pflags);
326
327 if (frontswap_load(page) == 0) {
328 SetPageUptodate(page);
329 unlock_page(page);
330 goto out;
331 }
332
333 if (data_race(sis->flags & SWP_FS_OPS)) {
334 struct file *swap_file = sis->swap_file;
335 struct address_space *mapping = swap_file->f_mapping;
336
337 ret = mapping->a_ops->readpage(swap_file, page);
338 if (!ret)
339 count_vm_event(PSWPIN);
340 goto out;
341 }
342
343 if (sis->flags & SWP_SYNCHRONOUS_IO) {
344 ret = bdev_read_page(sis->bdev, swap_page_sector(page), page);
345 if (!ret) {
346 count_vm_event(PSWPIN);
347 goto out;
348 }
349 }
350
351 ret = 0;
352 bio = bio_alloc(GFP_KERNEL, 1);
353 bio_set_dev(bio, sis->bdev);
354 bio->bi_opf = REQ_OP_READ;
355 bio->bi_iter.bi_sector = swap_page_sector(page);
356 bio->bi_end_io = end_swap_bio_read;
357 bio_add_page(bio, page, thp_size(page), 0);
358
359 disk = bio->bi_bdev->bd_disk;
360 /*
361 * Keep this task valid during swap readpage because the oom killer may
362 * attempt to access it in the page fault retry time check.
363 */
364 if (synchronous) {
365 bio->bi_opf |= REQ_HIPRI;
366 get_task_struct(current);
367 bio->bi_private = current;
368 }
369 count_vm_event(PSWPIN);
370 bio_get(bio);
371 qc = submit_bio(bio);
372 while (synchronous) {
373 set_current_state(TASK_UNINTERRUPTIBLE);
374 if (!READ_ONCE(bio->bi_private))
375 break;
376
377 if (!blk_poll(disk->queue, qc, true))
378 blk_io_schedule();
379 }
380 __set_current_state(TASK_RUNNING);
381 bio_put(bio);
382
383 out:
384 psi_memstall_leave(&pflags);
385 return ret;
386 }
387
swap_set_page_dirty(struct page * page)388 int swap_set_page_dirty(struct page *page)
389 {
390 struct swap_info_struct *sis = page_swap_info(page);
391
392 if (data_race(sis->flags & SWP_FS_OPS)) {
393 struct address_space *mapping = sis->swap_file->f_mapping;
394
395 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
396 return mapping->a_ops->set_page_dirty(page);
397 } else {
398 return __set_page_dirty_no_writeback(page);
399 }
400 }
401