Linux Audio

Check our new training course

Loading...
v5.4
  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
 29static 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
 48void 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
 72static 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
120static 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);
136out:
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
146int 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;
222reprobe:
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;
232out:
233	return ret;
234bad_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 */
244int 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);
259out:
260	return ret;
261}
262
263static sector_t swap_page_sector(struct page *page)
264{
265	return (sector_t)__page_file_index(page) << (PAGE_SHIFT - 9);
266}
267
268static 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
277int __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);
346out:
347	return ret;
348}
349
350int 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
420out:
 
421	return ret;
422}
423
424int 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}
v5.14.15
  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
 29void 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
 53static void swap_slot_free_notify(struct page *page)
 54{
 55	struct swap_info_struct *sis;
 56	struct gendisk *disk;
 57	swp_entry_t entry;
 58
 59	/*
 60	 * There is no guarantee that the page is in swap cache - the software
 61	 * suspend code (at least) uses end_swap_bio_read() against a non-
 62	 * swapcache page.  So we must check PG_swapcache before proceeding with
 63	 * this optimization.
 64	 */
 65	if (unlikely(!PageSwapCache(page)))
 66		return;
 67
 68	sis = page_swap_info(page);
 69	if (data_race(!(sis->flags & SWP_BLKDEV)))
 70		return;
 71
 72	/*
 73	 * The swap subsystem performs lazy swap slot freeing,
 74	 * expecting that the page will be swapped out again.
 75	 * So we can avoid an unnecessary write if the page
 76	 * isn't redirtied.
 77	 * This is good for real swap storage because we can
 78	 * reduce unnecessary I/O and enhance wear-leveling
 79	 * if an SSD is used as the as swap device.
 80	 * But if in-memory swap device (eg zram) is used,
 81	 * this causes a duplicated copy between uncompressed
 82	 * data in VM-owned memory and compressed data in
 83	 * zram-owned memory.  So let's free zram-owned memory
 84	 * and make the VM-owned decompressed page *dirty*,
 85	 * so the page should be swapped out somewhere again if
 86	 * we again wish to reclaim it.
 87	 */
 88	disk = sis->bdev->bd_disk;
 89	entry.val = page_private(page);
 90	if (disk->fops->swap_slot_free_notify && __swap_count(entry) == 1) {
 91		unsigned long offset;
 92
 93		offset = swp_offset(entry);
 94
 95		SetPageDirty(page);
 96		disk->fops->swap_slot_free_notify(sis->bdev,
 97				offset);
 98	}
 99}
100
101static void end_swap_bio_read(struct bio *bio)
102{
103	struct page *page = bio_first_page_all(bio);
104	struct task_struct *waiter = bio->bi_private;
105
106	if (bio->bi_status) {
107		SetPageError(page);
108		ClearPageUptodate(page);
109		pr_alert_ratelimited("Read-error on swap-device (%u:%u:%llu)\n",
110				     MAJOR(bio_dev(bio)), MINOR(bio_dev(bio)),
111				     (unsigned long long)bio->bi_iter.bi_sector);
112		goto out;
113	}
114
115	SetPageUptodate(page);
116	swap_slot_free_notify(page);
117out:
118	unlock_page(page);
119	WRITE_ONCE(bio->bi_private, NULL);
120	bio_put(bio);
121	if (waiter) {
122		blk_wake_io_task(waiter);
123		put_task_struct(waiter);
124	}
125}
126
127int generic_swapfile_activate(struct swap_info_struct *sis,
128				struct file *swap_file,
129				sector_t *span)
130{
131	struct address_space *mapping = swap_file->f_mapping;
132	struct inode *inode = mapping->host;
133	unsigned blocks_per_page;
134	unsigned long page_no;
135	unsigned blkbits;
136	sector_t probe_block;
137	sector_t last_block;
138	sector_t lowest_block = -1;
139	sector_t highest_block = 0;
140	int nr_extents = 0;
141	int ret;
142
143	blkbits = inode->i_blkbits;
144	blocks_per_page = PAGE_SIZE >> blkbits;
145
146	/*
147	 * Map all the blocks into the extent tree.  This code doesn't try
148	 * to be very smart.
149	 */
150	probe_block = 0;
151	page_no = 0;
152	last_block = i_size_read(inode) >> blkbits;
153	while ((probe_block + blocks_per_page) <= last_block &&
154			page_no < sis->max) {
155		unsigned block_in_page;
156		sector_t first_block;
157
158		cond_resched();
159
160		first_block = probe_block;
161		ret = bmap(inode, &first_block);
162		if (ret || !first_block)
163			goto bad_bmap;
164
165		/*
166		 * It must be PAGE_SIZE aligned on-disk
167		 */
168		if (first_block & (blocks_per_page - 1)) {
169			probe_block++;
170			goto reprobe;
171		}
172
173		for (block_in_page = 1; block_in_page < blocks_per_page;
174					block_in_page++) {
175			sector_t block;
176
177			block = probe_block + block_in_page;
178			ret = bmap(inode, &block);
179			if (ret || !block)
180				goto bad_bmap;
181
182			if (block != first_block + block_in_page) {
183				/* Discontiguity */
184				probe_block++;
185				goto reprobe;
186			}
187		}
188
189		first_block >>= (PAGE_SHIFT - blkbits);
190		if (page_no) {	/* exclude the header page */
191			if (first_block < lowest_block)
192				lowest_block = first_block;
193			if (first_block > highest_block)
194				highest_block = first_block;
195		}
196
197		/*
198		 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
199		 */
200		ret = add_swap_extent(sis, page_no, 1, first_block);
201		if (ret < 0)
202			goto out;
203		nr_extents += ret;
204		page_no++;
205		probe_block += blocks_per_page;
206reprobe:
207		continue;
208	}
209	ret = nr_extents;
210	*span = 1 + highest_block - lowest_block;
211	if (page_no == 0)
212		page_no = 1;	/* force Empty message */
213	sis->max = page_no;
214	sis->pages = page_no - 1;
215	sis->highest_bit = page_no - 1;
216out:
217	return ret;
218bad_bmap:
219	pr_err("swapon: swapfile has holes\n");
220	ret = -EINVAL;
221	goto out;
222}
223
224/*
225 * We may have stale swap cache pages in memory: notice
226 * them here and get rid of the unnecessary final write.
227 */
228int swap_writepage(struct page *page, struct writeback_control *wbc)
229{
230	int ret = 0;
231
232	if (try_to_free_swap(page)) {
233		unlock_page(page);
234		goto out;
235	}
236	/*
237	 * Arch code may have to preserve more data than just the page
238	 * contents, e.g. memory tags.
239	 */
240	ret = arch_prepare_to_swap(page);
241	if (ret) {
242		set_page_dirty(page);
243		unlock_page(page);
244		goto out;
245	}
246	if (frontswap_store(page) == 0) {
247		set_page_writeback(page);
248		unlock_page(page);
249		end_page_writeback(page);
250		goto out;
251	}
252	ret = __swap_writepage(page, wbc, end_swap_bio_write);
253out:
254	return ret;
255}
256
 
 
 
 
 
257static inline void count_swpout_vm_event(struct page *page)
258{
259#ifdef CONFIG_TRANSPARENT_HUGEPAGE
260	if (unlikely(PageTransHuge(page)))
261		count_vm_event(THP_SWPOUT);
262#endif
263	count_vm_events(PSWPOUT, thp_nr_pages(page));
264}
265
266#if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
267static void bio_associate_blkg_from_page(struct bio *bio, struct page *page)
268{
269	struct cgroup_subsys_state *css;
270	struct mem_cgroup *memcg;
271
272	memcg = page_memcg(page);
273	if (!memcg)
274		return;
275
276	rcu_read_lock();
277	css = cgroup_e_css(memcg->css.cgroup, &io_cgrp_subsys);
278	bio_associate_blkg_from_css(bio, css);
279	rcu_read_unlock();
280}
281#else
282#define bio_associate_blkg_from_page(bio, page)		do { } while (0)
283#endif /* CONFIG_MEMCG && CONFIG_BLK_CGROUP */
284
285int __swap_writepage(struct page *page, struct writeback_control *wbc,
286		bio_end_io_t end_write_func)
287{
288	struct bio *bio;
289	int ret;
290	struct swap_info_struct *sis = page_swap_info(page);
291
292	VM_BUG_ON_PAGE(!PageSwapCache(page), page);
293	if (data_race(sis->flags & SWP_FS_OPS)) {
294		struct kiocb kiocb;
295		struct file *swap_file = sis->swap_file;
296		struct address_space *mapping = swap_file->f_mapping;
297		struct bio_vec bv = {
298			.bv_page = page,
299			.bv_len  = PAGE_SIZE,
300			.bv_offset = 0
301		};
302		struct iov_iter from;
303
304		iov_iter_bvec(&from, WRITE, &bv, 1, PAGE_SIZE);
305		init_sync_kiocb(&kiocb, swap_file);
306		kiocb.ki_pos = page_file_offset(page);
307
308		set_page_writeback(page);
309		unlock_page(page);
310		ret = mapping->a_ops->direct_IO(&kiocb, &from);
311		if (ret == PAGE_SIZE) {
312			count_vm_event(PSWPOUT);
313			ret = 0;
314		} else {
315			/*
316			 * In the case of swap-over-nfs, this can be a
317			 * temporary failure if the system has limited
318			 * memory for allocating transmit buffers.
319			 * Mark the page dirty and avoid
320			 * rotate_reclaimable_page but rate-limit the
321			 * messages but do not flag PageError like
322			 * the normal direct-to-bio case as it could
323			 * be temporary.
324			 */
325			set_page_dirty(page);
326			ClearPageReclaim(page);
327			pr_err_ratelimited("Write error on dio swapfile (%llu)\n",
328					   page_file_offset(page));
329		}
330		end_page_writeback(page);
331		return ret;
332	}
333
334	ret = bdev_write_page(sis->bdev, swap_page_sector(page), page, wbc);
335	if (!ret) {
336		count_swpout_vm_event(page);
337		return 0;
338	}
339
340	bio = bio_alloc(GFP_NOIO, 1);
341	bio_set_dev(bio, sis->bdev);
342	bio->bi_iter.bi_sector = swap_page_sector(page);
 
 
 
 
 
343	bio->bi_opf = REQ_OP_WRITE | REQ_SWAP | wbc_to_write_flags(wbc);
344	bio->bi_end_io = end_write_func;
345	bio_add_page(bio, page, thp_size(page), 0);
346
347	bio_associate_blkg_from_page(bio, page);
348	count_swpout_vm_event(page);
349	set_page_writeback(page);
350	unlock_page(page);
351	submit_bio(bio);
352
353	return 0;
354}
355
356int swap_readpage(struct page *page, bool synchronous)
357{
358	struct bio *bio;
359	int ret = 0;
360	struct swap_info_struct *sis = page_swap_info(page);
361	blk_qc_t qc;
362	struct gendisk *disk;
363	unsigned long pflags;
364
365	VM_BUG_ON_PAGE(!PageSwapCache(page) && !synchronous, page);
366	VM_BUG_ON_PAGE(!PageLocked(page), page);
367	VM_BUG_ON_PAGE(PageUptodate(page), page);
368
369	/*
370	 * Count submission time as memory stall. When the device is congested,
371	 * or the submitting cgroup IO-throttled, submission can be a
372	 * significant part of overall IO time.
373	 */
374	psi_memstall_enter(&pflags);
375
376	if (frontswap_load(page) == 0) {
377		SetPageUptodate(page);
378		unlock_page(page);
379		goto out;
380	}
381
382	if (data_race(sis->flags & SWP_FS_OPS)) {
383		struct file *swap_file = sis->swap_file;
384		struct address_space *mapping = swap_file->f_mapping;
385
386		ret = mapping->a_ops->readpage(swap_file, page);
387		if (!ret)
388			count_vm_event(PSWPIN);
389		goto out;
390	}
391
392	if (sis->flags & SWP_SYNCHRONOUS_IO) {
393		ret = bdev_read_page(sis->bdev, swap_page_sector(page), page);
394		if (!ret) {
395			if (trylock_page(page)) {
396				swap_slot_free_notify(page);
397				unlock_page(page);
398			}
399
400			count_vm_event(PSWPIN);
401			goto out;
402		}
403	}
404
405	ret = 0;
406	bio = bio_alloc(GFP_KERNEL, 1);
407	bio_set_dev(bio, sis->bdev);
408	bio->bi_opf = REQ_OP_READ;
409	bio->bi_iter.bi_sector = swap_page_sector(page);
410	bio->bi_end_io = end_swap_bio_read;
411	bio_add_page(bio, page, thp_size(page), 0);
412
413	disk = bio->bi_bdev->bd_disk;
414	/*
415	 * Keep this task valid during swap readpage because the oom killer may
416	 * attempt to access it in the page fault retry time check.
417	 */
 
418	if (synchronous) {
419		bio->bi_opf |= REQ_HIPRI;
420		get_task_struct(current);
421		bio->bi_private = current;
422	}
423	count_vm_event(PSWPIN);
424	bio_get(bio);
425	qc = submit_bio(bio);
426	while (synchronous) {
427		set_current_state(TASK_UNINTERRUPTIBLE);
428		if (!READ_ONCE(bio->bi_private))
429			break;
430
431		if (!blk_poll(disk->queue, qc, true))
432			blk_io_schedule();
433	}
434	__set_current_state(TASK_RUNNING);
435	bio_put(bio);
436
437out:
438	psi_memstall_leave(&pflags);
439	return ret;
440}
441
442int swap_set_page_dirty(struct page *page)
443{
444	struct swap_info_struct *sis = page_swap_info(page);
445
446	if (data_race(sis->flags & SWP_FS_OPS)) {
447		struct address_space *mapping = sis->swap_file->f_mapping;
448
449		VM_BUG_ON_PAGE(!PageSwapCache(page), page);
450		return mapping->a_ops->set_page_dirty(page);
451	} else {
452		return __set_page_dirty_no_writeback(page);
453	}
454}