Linux Audio

Check our new training course

Linux kernel drivers training

May 6-19, 2025
Register
Loading...
v4.10.11
 
  1/*
  2 * linux/fs/ext4/page-io.c
  3 *
  4 * This contains the new page_io functions for ext4
  5 *
  6 * Written by Theodore Ts'o, 2010.
  7 */
  8
  9#include <linux/fs.h>
 10#include <linux/time.h>
 11#include <linux/highuid.h>
 12#include <linux/pagemap.h>
 13#include <linux/quotaops.h>
 14#include <linux/string.h>
 15#include <linux/buffer_head.h>
 16#include <linux/writeback.h>
 17#include <linux/pagevec.h>
 18#include <linux/mpage.h>
 19#include <linux/namei.h>
 20#include <linux/uio.h>
 21#include <linux/bio.h>
 22#include <linux/workqueue.h>
 23#include <linux/kernel.h>
 24#include <linux/slab.h>
 25#include <linux/mm.h>
 26#include <linux/backing-dev.h>
 27#include <linux/fscrypto.h>
 28
 29#include "ext4_jbd2.h"
 30#include "xattr.h"
 31#include "acl.h"
 32
 33static struct kmem_cache *io_end_cachep;
 34
 35int __init ext4_init_pageio(void)
 36{
 37	io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT);
 38	if (io_end_cachep == NULL)
 39		return -ENOMEM;
 40	return 0;
 41}
 42
 43void ext4_exit_pageio(void)
 44{
 45	kmem_cache_destroy(io_end_cachep);
 46}
 47
 48/*
 49 * Print an buffer I/O error compatible with the fs/buffer.c.  This
 50 * provides compatibility with dmesg scrapers that look for a specific
 51 * buffer I/O error message.  We really need a unified error reporting
 52 * structure to userspace ala Digital Unix's uerf system, but it's
 53 * probably not going to happen in my lifetime, due to LKML politics...
 54 */
 55static void buffer_io_error(struct buffer_head *bh)
 56{
 57	printk_ratelimited(KERN_ERR "Buffer I/O error on device %pg, logical block %llu\n",
 58		       bh->b_bdev,
 59			(unsigned long long)bh->b_blocknr);
 60}
 61
 62static void ext4_finish_bio(struct bio *bio)
 63{
 64	int i;
 65	struct bio_vec *bvec;
 
 66
 67	bio_for_each_segment_all(bvec, bio, i) {
 68		struct page *page = bvec->bv_page;
 69#ifdef CONFIG_EXT4_FS_ENCRYPTION
 70		struct page *data_page = NULL;
 71#endif
 72		struct buffer_head *bh, *head;
 73		unsigned bio_start = bvec->bv_offset;
 74		unsigned bio_end = bio_start + bvec->bv_len;
 75		unsigned under_io = 0;
 76		unsigned long flags;
 77
 78		if (!page)
 79			continue;
 80
 81#ifdef CONFIG_EXT4_FS_ENCRYPTION
 82		if (!page->mapping) {
 83			/* The bounce data pages are unmapped. */
 84			data_page = page;
 85			fscrypt_pullback_bio_page(&page, false);
 86		}
 87#endif
 88
 89		if (bio->bi_error) {
 90			SetPageError(page);
 91			mapping_set_error(page->mapping, -EIO);
 92		}
 93		bh = head = page_buffers(page);
 94		/*
 95		 * We check all buffers in the page under BH_Uptodate_Lock
 96		 * to avoid races with other end io clearing async_write flags
 97		 */
 98		local_irq_save(flags);
 99		bit_spin_lock(BH_Uptodate_Lock, &head->b_state);
100		do {
101			if (bh_offset(bh) < bio_start ||
102			    bh_offset(bh) + bh->b_size > bio_end) {
103				if (buffer_async_write(bh))
104					under_io++;
105				continue;
106			}
107			clear_buffer_async_write(bh);
108			if (bio->bi_error)
109				buffer_io_error(bh);
110		} while ((bh = bh->b_this_page) != head);
111		bit_spin_unlock(BH_Uptodate_Lock, &head->b_state);
112		local_irq_restore(flags);
113		if (!under_io) {
114#ifdef CONFIG_EXT4_FS_ENCRYPTION
115			if (data_page)
116				fscrypt_restore_control_page(data_page);
117#endif
118			end_page_writeback(page);
119		}
120	}
121}
122
123static void ext4_release_io_end(ext4_io_end_t *io_end)
124{
125	struct bio *bio, *next_bio;
126
127	BUG_ON(!list_empty(&io_end->list));
128	BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
129	WARN_ON(io_end->handle);
130
131	for (bio = io_end->bio; bio; bio = next_bio) {
132		next_bio = bio->bi_private;
133		ext4_finish_bio(bio);
134		bio_put(bio);
135	}
136	kmem_cache_free(io_end_cachep, io_end);
137}
138
139/*
140 * Check a range of space and convert unwritten extents to written. Note that
141 * we are protected from truncate touching same part of extent tree by the
142 * fact that truncate code waits for all DIO to finish (thus exclusion from
143 * direct IO is achieved) and also waits for PageWriteback bits. Thus we
144 * cannot get to ext4_ext_truncate() before all IOs overlapping that range are
145 * completed (happens from ext4_free_ioend()).
146 */
147static int ext4_end_io(ext4_io_end_t *io)
148{
149	struct inode *inode = io->inode;
150	loff_t offset = io->offset;
151	ssize_t size = io->size;
152	handle_t *handle = io->handle;
153	int ret = 0;
154
155	ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
156		   "list->prev 0x%p\n",
157		   io, inode->i_ino, io->list.next, io->list.prev);
158
159	io->handle = NULL;	/* Following call will use up the handle */
160	ret = ext4_convert_unwritten_extents(handle, inode, offset, size);
161	if (ret < 0) {
162		ext4_msg(inode->i_sb, KERN_EMERG,
163			 "failed to convert unwritten extents to written "
164			 "extents -- potential data loss!  "
165			 "(inode %lu, offset %llu, size %zd, error %d)",
166			 inode->i_ino, offset, size, ret);
167	}
168	ext4_clear_io_unwritten_flag(io);
169	ext4_release_io_end(io);
170	return ret;
171}
172
173static void dump_completed_IO(struct inode *inode, struct list_head *head)
174{
175#ifdef	EXT4FS_DEBUG
176	struct list_head *cur, *before, *after;
177	ext4_io_end_t *io, *io0, *io1;
178
179	if (list_empty(head))
180		return;
181
182	ext4_debug("Dump inode %lu completed io list\n", inode->i_ino);
183	list_for_each_entry(io, head, list) {
184		cur = &io->list;
185		before = cur->prev;
186		io0 = container_of(before, ext4_io_end_t, list);
187		after = cur->next;
188		io1 = container_of(after, ext4_io_end_t, list);
189
190		ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
191			    io, inode->i_ino, io0, io1);
192	}
193#endif
194}
195
196/* Add the io_end to per-inode completed end_io list. */
197static void ext4_add_complete_io(ext4_io_end_t *io_end)
198{
199	struct ext4_inode_info *ei = EXT4_I(io_end->inode);
200	struct ext4_sb_info *sbi = EXT4_SB(io_end->inode->i_sb);
201	struct workqueue_struct *wq;
202	unsigned long flags;
203
204	/* Only reserved conversions from writeback should enter here */
205	WARN_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
206	WARN_ON(!io_end->handle && sbi->s_journal);
207	spin_lock_irqsave(&ei->i_completed_io_lock, flags);
208	wq = sbi->rsv_conversion_wq;
209	if (list_empty(&ei->i_rsv_conversion_list))
210		queue_work(wq, &ei->i_rsv_conversion_work);
211	list_add_tail(&io_end->list, &ei->i_rsv_conversion_list);
212	spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
213}
214
215static int ext4_do_flush_completed_IO(struct inode *inode,
216				      struct list_head *head)
217{
218	ext4_io_end_t *io;
219	struct list_head unwritten;
220	unsigned long flags;
221	struct ext4_inode_info *ei = EXT4_I(inode);
222	int err, ret = 0;
223
224	spin_lock_irqsave(&ei->i_completed_io_lock, flags);
225	dump_completed_IO(inode, head);
226	list_replace_init(head, &unwritten);
227	spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
228
229	while (!list_empty(&unwritten)) {
230		io = list_entry(unwritten.next, ext4_io_end_t, list);
231		BUG_ON(!(io->flag & EXT4_IO_END_UNWRITTEN));
232		list_del_init(&io->list);
233
234		err = ext4_end_io(io);
235		if (unlikely(!ret && err))
236			ret = err;
237	}
238	return ret;
239}
240
241/*
242 * work on completed IO, to convert unwritten extents to extents
243 */
244void ext4_end_io_rsv_work(struct work_struct *work)
245{
246	struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
247						  i_rsv_conversion_work);
248	ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_rsv_conversion_list);
249}
250
251ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
252{
253	ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags);
254	if (io) {
255		io->inode = inode;
256		INIT_LIST_HEAD(&io->list);
257		atomic_set(&io->count, 1);
258	}
259	return io;
260}
261
262void ext4_put_io_end_defer(ext4_io_end_t *io_end)
263{
264	if (atomic_dec_and_test(&io_end->count)) {
265		if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) || !io_end->size) {
266			ext4_release_io_end(io_end);
267			return;
268		}
269		ext4_add_complete_io(io_end);
270	}
271}
272
273int ext4_put_io_end(ext4_io_end_t *io_end)
274{
275	int err = 0;
276
277	if (atomic_dec_and_test(&io_end->count)) {
278		if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
279			err = ext4_convert_unwritten_extents(io_end->handle,
280						io_end->inode, io_end->offset,
281						io_end->size);
282			io_end->handle = NULL;
283			ext4_clear_io_unwritten_flag(io_end);
284		}
285		ext4_release_io_end(io_end);
286	}
287	return err;
288}
289
290ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end)
291{
292	atomic_inc(&io_end->count);
293	return io_end;
294}
295
296/* BIO completion function for page writeback */
297static void ext4_end_bio(struct bio *bio)
298{
299	ext4_io_end_t *io_end = bio->bi_private;
300	sector_t bi_sector = bio->bi_iter.bi_sector;
 
301
302	BUG_ON(!io_end);
 
 
 
 
 
 
 
 
303	bio->bi_end_io = NULL;
304
305	if (bio->bi_error) {
306		struct inode *inode = io_end->inode;
307
308		ext4_warning(inode->i_sb, "I/O error %d writing to inode %lu "
309			     "(offset %llu size %ld starting block %llu)",
310			     bio->bi_error, inode->i_ino,
311			     (unsigned long long) io_end->offset,
312			     (long) io_end->size,
313			     (unsigned long long)
314			     bi_sector >> (inode->i_blkbits - 9));
315		mapping_set_error(inode->i_mapping, bio->bi_error);
 
316	}
317
318	if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
319		/*
320		 * Link bio into list hanging from io_end. We have to do it
321		 * atomically as bio completions can be racing against each
322		 * other.
323		 */
324		bio->bi_private = xchg(&io_end->bio, bio);
325		ext4_put_io_end_defer(io_end);
326	} else {
327		/*
328		 * Drop io_end reference early. Inode can get freed once
329		 * we finish the bio.
330		 */
331		ext4_put_io_end_defer(io_end);
332		ext4_finish_bio(bio);
333		bio_put(bio);
334	}
335}
336
337void ext4_io_submit(struct ext4_io_submit *io)
338{
339	struct bio *bio = io->io_bio;
340
341	if (bio) {
342		int io_op_flags = io->io_wbc->sync_mode == WB_SYNC_ALL ?
343				  REQ_SYNC : 0;
 
344		bio_set_op_attrs(io->io_bio, REQ_OP_WRITE, io_op_flags);
345		submit_bio(io->io_bio);
346	}
347	io->io_bio = NULL;
348}
349
350void ext4_io_submit_init(struct ext4_io_submit *io,
351			 struct writeback_control *wbc)
352{
353	io->io_wbc = wbc;
354	io->io_bio = NULL;
355	io->io_end = NULL;
356}
357
358static int io_submit_init_bio(struct ext4_io_submit *io,
359			      struct buffer_head *bh)
360{
361	struct bio *bio;
362
363	bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
364	if (!bio)
365		return -ENOMEM;
366	wbc_init_bio(io->io_wbc, bio);
367	bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
368	bio->bi_bdev = bh->b_bdev;
369	bio->bi_end_io = ext4_end_bio;
370	bio->bi_private = ext4_get_io_end(io->io_end);
371	io->io_bio = bio;
372	io->io_next_block = bh->b_blocknr;
 
373	return 0;
374}
375
376static int io_submit_add_bh(struct ext4_io_submit *io,
377			    struct inode *inode,
378			    struct page *page,
379			    struct buffer_head *bh)
380{
381	int ret;
382
383	if (io->io_bio && bh->b_blocknr != io->io_next_block) {
384submit_and_retry:
385		ext4_io_submit(io);
386	}
387	if (io->io_bio == NULL) {
388		ret = io_submit_init_bio(io, bh);
389		if (ret)
390			return ret;
 
391	}
392	ret = bio_add_page(io->io_bio, page, bh->b_size, bh_offset(bh));
393	if (ret != bh->b_size)
394		goto submit_and_retry;
395	wbc_account_io(io->io_wbc, page, bh->b_size);
396	io->io_next_block++;
397	return 0;
398}
399
400int ext4_bio_write_page(struct ext4_io_submit *io,
401			struct page *page,
402			int len,
403			struct writeback_control *wbc,
404			bool keep_towrite)
405{
406	struct page *data_page = NULL;
407	struct inode *inode = page->mapping->host;
408	unsigned block_start;
409	struct buffer_head *bh, *head;
410	int ret = 0;
411	int nr_submitted = 0;
412	int nr_to_submit = 0;
413
414	BUG_ON(!PageLocked(page));
415	BUG_ON(PageWriteback(page));
416
417	if (keep_towrite)
418		set_page_writeback_keepwrite(page);
419	else
420		set_page_writeback(page);
421	ClearPageError(page);
422
423	/*
424	 * Comments copied from block_write_full_page:
425	 *
426	 * The page straddles i_size.  It must be zeroed out on each and every
427	 * writepage invocation because it may be mmapped.  "A file is mapped
428	 * in multiples of the page size.  For a file that is not a multiple of
429	 * the page size, the remaining memory is zeroed when mapped, and
430	 * writes to that region are not written out to the file."
431	 */
432	if (len < PAGE_SIZE)
433		zero_user_segment(page, len, PAGE_SIZE);
434	/*
435	 * In the first loop we prepare and mark buffers to submit. We have to
436	 * mark all buffers in the page before submitting so that
437	 * end_page_writeback() cannot be called from ext4_bio_end_io() when IO
438	 * on the first buffer finishes and we are still working on submitting
439	 * the second buffer.
440	 */
441	bh = head = page_buffers(page);
442	do {
443		block_start = bh_offset(bh);
444		if (block_start >= len) {
445			clear_buffer_dirty(bh);
446			set_buffer_uptodate(bh);
447			continue;
448		}
449		if (!buffer_dirty(bh) || buffer_delay(bh) ||
450		    !buffer_mapped(bh) || buffer_unwritten(bh)) {
451			/* A hole? We can safely clear the dirty bit */
452			if (!buffer_mapped(bh))
453				clear_buffer_dirty(bh);
454			if (io->io_bio)
455				ext4_io_submit(io);
456			continue;
457		}
458		if (buffer_new(bh)) {
459			clear_buffer_new(bh);
460			clean_bdev_bh_alias(bh);
461		}
462		set_buffer_async_write(bh);
463		nr_to_submit++;
464	} while ((bh = bh->b_this_page) != head);
465
466	bh = head = page_buffers(page);
467
468	if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode) &&
469	    nr_to_submit) {
 
 
 
 
 
 
470		gfp_t gfp_flags = GFP_NOFS;
 
471
472	retry_encrypt:
473		data_page = fscrypt_encrypt_page(inode, page, PAGE_SIZE, 0,
474						page->index, gfp_flags);
475		if (IS_ERR(data_page)) {
476			ret = PTR_ERR(data_page);
477			if (ret == -ENOMEM && wbc->sync_mode == WB_SYNC_ALL) {
478				if (io->io_bio) {
479					ext4_io_submit(io);
480					congestion_wait(BLK_RW_ASYNC, HZ/50);
481				}
482				gfp_flags |= __GFP_NOFAIL;
483				goto retry_encrypt;
484			}
485			data_page = NULL;
486			goto out;
487		}
488	}
489
490	/* Now submit buffers to write */
491	do {
492		if (!buffer_async_write(bh))
493			continue;
494		ret = io_submit_add_bh(io, inode,
495				       data_page ? data_page : page, bh);
496		if (ret) {
497			/*
498			 * We only get here on ENOMEM.  Not much else
499			 * we can do but mark the page as dirty, and
500			 * better luck next time.
501			 */
502			break;
503		}
504		nr_submitted++;
505		clear_buffer_dirty(bh);
506	} while ((bh = bh->b_this_page) != head);
507
508	/* Error stopped previous loop? Clean up buffers... */
509	if (ret) {
510	out:
511		if (data_page)
512			fscrypt_restore_control_page(data_page);
513		printk_ratelimited(KERN_ERR "%s: ret = %d\n", __func__, ret);
514		redirty_page_for_writepage(wbc, page);
515		do {
516			clear_buffer_async_write(bh);
517			bh = bh->b_this_page;
518		} while (bh != head);
519	}
520	unlock_page(page);
521	/* Nothing submitted - we have to end page writeback */
522	if (!nr_submitted)
523		end_page_writeback(page);
524	return ret;
525}
v5.4
  1// SPDX-License-Identifier: GPL-2.0
  2/*
  3 * linux/fs/ext4/page-io.c
  4 *
  5 * This contains the new page_io functions for ext4
  6 *
  7 * Written by Theodore Ts'o, 2010.
  8 */
  9
 10#include <linux/fs.h>
 11#include <linux/time.h>
 12#include <linux/highuid.h>
 13#include <linux/pagemap.h>
 14#include <linux/quotaops.h>
 15#include <linux/string.h>
 16#include <linux/buffer_head.h>
 17#include <linux/writeback.h>
 18#include <linux/pagevec.h>
 19#include <linux/mpage.h>
 20#include <linux/namei.h>
 21#include <linux/uio.h>
 22#include <linux/bio.h>
 23#include <linux/workqueue.h>
 24#include <linux/kernel.h>
 25#include <linux/slab.h>
 26#include <linux/mm.h>
 27#include <linux/backing-dev.h>
 
 28
 29#include "ext4_jbd2.h"
 30#include "xattr.h"
 31#include "acl.h"
 32
 33static struct kmem_cache *io_end_cachep;
 34
 35int __init ext4_init_pageio(void)
 36{
 37	io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT);
 38	if (io_end_cachep == NULL)
 39		return -ENOMEM;
 40	return 0;
 41}
 42
 43void ext4_exit_pageio(void)
 44{
 45	kmem_cache_destroy(io_end_cachep);
 46}
 47
 48/*
 49 * Print an buffer I/O error compatible with the fs/buffer.c.  This
 50 * provides compatibility with dmesg scrapers that look for a specific
 51 * buffer I/O error message.  We really need a unified error reporting
 52 * structure to userspace ala Digital Unix's uerf system, but it's
 53 * probably not going to happen in my lifetime, due to LKML politics...
 54 */
 55static void buffer_io_error(struct buffer_head *bh)
 56{
 57	printk_ratelimited(KERN_ERR "Buffer I/O error on device %pg, logical block %llu\n",
 58		       bh->b_bdev,
 59			(unsigned long long)bh->b_blocknr);
 60}
 61
 62static void ext4_finish_bio(struct bio *bio)
 63{
 
 64	struct bio_vec *bvec;
 65	struct bvec_iter_all iter_all;
 66
 67	bio_for_each_segment_all(bvec, bio, iter_all) {
 68		struct page *page = bvec->bv_page;
 69		struct page *bounce_page = NULL;
 
 
 70		struct buffer_head *bh, *head;
 71		unsigned bio_start = bvec->bv_offset;
 72		unsigned bio_end = bio_start + bvec->bv_len;
 73		unsigned under_io = 0;
 74		unsigned long flags;
 75
 76		if (!page)
 77			continue;
 78
 79		if (fscrypt_is_bounce_page(page)) {
 80			bounce_page = page;
 81			page = fscrypt_pagecache_page(bounce_page);
 
 
 82		}
 
 83
 84		if (bio->bi_status) {
 85			SetPageError(page);
 86			mapping_set_error(page->mapping, -EIO);
 87		}
 88		bh = head = page_buffers(page);
 89		/*
 90		 * We check all buffers in the page under BH_Uptodate_Lock
 91		 * to avoid races with other end io clearing async_write flags
 92		 */
 93		local_irq_save(flags);
 94		bit_spin_lock(BH_Uptodate_Lock, &head->b_state);
 95		do {
 96			if (bh_offset(bh) < bio_start ||
 97			    bh_offset(bh) + bh->b_size > bio_end) {
 98				if (buffer_async_write(bh))
 99					under_io++;
100				continue;
101			}
102			clear_buffer_async_write(bh);
103			if (bio->bi_status)
104				buffer_io_error(bh);
105		} while ((bh = bh->b_this_page) != head);
106		bit_spin_unlock(BH_Uptodate_Lock, &head->b_state);
107		local_irq_restore(flags);
108		if (!under_io) {
109			fscrypt_free_bounce_page(bounce_page);
 
 
 
110			end_page_writeback(page);
111		}
112	}
113}
114
115static void ext4_release_io_end(ext4_io_end_t *io_end)
116{
117	struct bio *bio, *next_bio;
118
119	BUG_ON(!list_empty(&io_end->list));
120	BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
121	WARN_ON(io_end->handle);
122
123	for (bio = io_end->bio; bio; bio = next_bio) {
124		next_bio = bio->bi_private;
125		ext4_finish_bio(bio);
126		bio_put(bio);
127	}
128	kmem_cache_free(io_end_cachep, io_end);
129}
130
131/*
132 * Check a range of space and convert unwritten extents to written. Note that
133 * we are protected from truncate touching same part of extent tree by the
134 * fact that truncate code waits for all DIO to finish (thus exclusion from
135 * direct IO is achieved) and also waits for PageWriteback bits. Thus we
136 * cannot get to ext4_ext_truncate() before all IOs overlapping that range are
137 * completed (happens from ext4_free_ioend()).
138 */
139static int ext4_end_io(ext4_io_end_t *io)
140{
141	struct inode *inode = io->inode;
142	loff_t offset = io->offset;
143	ssize_t size = io->size;
144	handle_t *handle = io->handle;
145	int ret = 0;
146
147	ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
148		   "list->prev 0x%p\n",
149		   io, inode->i_ino, io->list.next, io->list.prev);
150
151	io->handle = NULL;	/* Following call will use up the handle */
152	ret = ext4_convert_unwritten_extents(handle, inode, offset, size);
153	if (ret < 0 && !ext4_forced_shutdown(EXT4_SB(inode->i_sb))) {
154		ext4_msg(inode->i_sb, KERN_EMERG,
155			 "failed to convert unwritten extents to written "
156			 "extents -- potential data loss!  "
157			 "(inode %lu, offset %llu, size %zd, error %d)",
158			 inode->i_ino, offset, size, ret);
159	}
160	ext4_clear_io_unwritten_flag(io);
161	ext4_release_io_end(io);
162	return ret;
163}
164
165static void dump_completed_IO(struct inode *inode, struct list_head *head)
166{
167#ifdef	EXT4FS_DEBUG
168	struct list_head *cur, *before, *after;
169	ext4_io_end_t *io, *io0, *io1;
170
171	if (list_empty(head))
172		return;
173
174	ext4_debug("Dump inode %lu completed io list\n", inode->i_ino);
175	list_for_each_entry(io, head, list) {
176		cur = &io->list;
177		before = cur->prev;
178		io0 = container_of(before, ext4_io_end_t, list);
179		after = cur->next;
180		io1 = container_of(after, ext4_io_end_t, list);
181
182		ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
183			    io, inode->i_ino, io0, io1);
184	}
185#endif
186}
187
188/* Add the io_end to per-inode completed end_io list. */
189static void ext4_add_complete_io(ext4_io_end_t *io_end)
190{
191	struct ext4_inode_info *ei = EXT4_I(io_end->inode);
192	struct ext4_sb_info *sbi = EXT4_SB(io_end->inode->i_sb);
193	struct workqueue_struct *wq;
194	unsigned long flags;
195
196	/* Only reserved conversions from writeback should enter here */
197	WARN_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
198	WARN_ON(!io_end->handle && sbi->s_journal);
199	spin_lock_irqsave(&ei->i_completed_io_lock, flags);
200	wq = sbi->rsv_conversion_wq;
201	if (list_empty(&ei->i_rsv_conversion_list))
202		queue_work(wq, &ei->i_rsv_conversion_work);
203	list_add_tail(&io_end->list, &ei->i_rsv_conversion_list);
204	spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
205}
206
207static int ext4_do_flush_completed_IO(struct inode *inode,
208				      struct list_head *head)
209{
210	ext4_io_end_t *io;
211	struct list_head unwritten;
212	unsigned long flags;
213	struct ext4_inode_info *ei = EXT4_I(inode);
214	int err, ret = 0;
215
216	spin_lock_irqsave(&ei->i_completed_io_lock, flags);
217	dump_completed_IO(inode, head);
218	list_replace_init(head, &unwritten);
219	spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
220
221	while (!list_empty(&unwritten)) {
222		io = list_entry(unwritten.next, ext4_io_end_t, list);
223		BUG_ON(!(io->flag & EXT4_IO_END_UNWRITTEN));
224		list_del_init(&io->list);
225
226		err = ext4_end_io(io);
227		if (unlikely(!ret && err))
228			ret = err;
229	}
230	return ret;
231}
232
233/*
234 * work on completed IO, to convert unwritten extents to extents
235 */
236void ext4_end_io_rsv_work(struct work_struct *work)
237{
238	struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
239						  i_rsv_conversion_work);
240	ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_rsv_conversion_list);
241}
242
243ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
244{
245	ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags);
246	if (io) {
247		io->inode = inode;
248		INIT_LIST_HEAD(&io->list);
249		atomic_set(&io->count, 1);
250	}
251	return io;
252}
253
254void ext4_put_io_end_defer(ext4_io_end_t *io_end)
255{
256	if (atomic_dec_and_test(&io_end->count)) {
257		if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) || !io_end->size) {
258			ext4_release_io_end(io_end);
259			return;
260		}
261		ext4_add_complete_io(io_end);
262	}
263}
264
265int ext4_put_io_end(ext4_io_end_t *io_end)
266{
267	int err = 0;
268
269	if (atomic_dec_and_test(&io_end->count)) {
270		if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
271			err = ext4_convert_unwritten_extents(io_end->handle,
272						io_end->inode, io_end->offset,
273						io_end->size);
274			io_end->handle = NULL;
275			ext4_clear_io_unwritten_flag(io_end);
276		}
277		ext4_release_io_end(io_end);
278	}
279	return err;
280}
281
282ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end)
283{
284	atomic_inc(&io_end->count);
285	return io_end;
286}
287
288/* BIO completion function for page writeback */
289static void ext4_end_bio(struct bio *bio)
290{
291	ext4_io_end_t *io_end = bio->bi_private;
292	sector_t bi_sector = bio->bi_iter.bi_sector;
293	char b[BDEVNAME_SIZE];
294
295	if (WARN_ONCE(!io_end, "io_end is NULL: %s: sector %Lu len %u err %d\n",
296		      bio_devname(bio, b),
297		      (long long) bio->bi_iter.bi_sector,
298		      (unsigned) bio_sectors(bio),
299		      bio->bi_status)) {
300		ext4_finish_bio(bio);
301		bio_put(bio);
302		return;
303	}
304	bio->bi_end_io = NULL;
305
306	if (bio->bi_status) {
307		struct inode *inode = io_end->inode;
308
309		ext4_warning(inode->i_sb, "I/O error %d writing to inode %lu "
310			     "(offset %llu size %ld starting block %llu)",
311			     bio->bi_status, inode->i_ino,
312			     (unsigned long long) io_end->offset,
313			     (long) io_end->size,
314			     (unsigned long long)
315			     bi_sector >> (inode->i_blkbits - 9));
316		mapping_set_error(inode->i_mapping,
317				blk_status_to_errno(bio->bi_status));
318	}
319
320	if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
321		/*
322		 * Link bio into list hanging from io_end. We have to do it
323		 * atomically as bio completions can be racing against each
324		 * other.
325		 */
326		bio->bi_private = xchg(&io_end->bio, bio);
327		ext4_put_io_end_defer(io_end);
328	} else {
329		/*
330		 * Drop io_end reference early. Inode can get freed once
331		 * we finish the bio.
332		 */
333		ext4_put_io_end_defer(io_end);
334		ext4_finish_bio(bio);
335		bio_put(bio);
336	}
337}
338
339void ext4_io_submit(struct ext4_io_submit *io)
340{
341	struct bio *bio = io->io_bio;
342
343	if (bio) {
344		int io_op_flags = io->io_wbc->sync_mode == WB_SYNC_ALL ?
345				  REQ_SYNC : 0;
346		io->io_bio->bi_write_hint = io->io_end->inode->i_write_hint;
347		bio_set_op_attrs(io->io_bio, REQ_OP_WRITE, io_op_flags);
348		submit_bio(io->io_bio);
349	}
350	io->io_bio = NULL;
351}
352
353void ext4_io_submit_init(struct ext4_io_submit *io,
354			 struct writeback_control *wbc)
355{
356	io->io_wbc = wbc;
357	io->io_bio = NULL;
358	io->io_end = NULL;
359}
360
361static int io_submit_init_bio(struct ext4_io_submit *io,
362			      struct buffer_head *bh)
363{
364	struct bio *bio;
365
366	bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
367	if (!bio)
368		return -ENOMEM;
 
369	bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
370	bio_set_dev(bio, bh->b_bdev);
371	bio->bi_end_io = ext4_end_bio;
372	bio->bi_private = ext4_get_io_end(io->io_end);
373	io->io_bio = bio;
374	io->io_next_block = bh->b_blocknr;
375	wbc_init_bio(io->io_wbc, bio);
376	return 0;
377}
378
379static int io_submit_add_bh(struct ext4_io_submit *io,
380			    struct inode *inode,
381			    struct page *page,
382			    struct buffer_head *bh)
383{
384	int ret;
385
386	if (io->io_bio && bh->b_blocknr != io->io_next_block) {
387submit_and_retry:
388		ext4_io_submit(io);
389	}
390	if (io->io_bio == NULL) {
391		ret = io_submit_init_bio(io, bh);
392		if (ret)
393			return ret;
394		io->io_bio->bi_write_hint = inode->i_write_hint;
395	}
396	ret = bio_add_page(io->io_bio, page, bh->b_size, bh_offset(bh));
397	if (ret != bh->b_size)
398		goto submit_and_retry;
399	wbc_account_cgroup_owner(io->io_wbc, page, bh->b_size);
400	io->io_next_block++;
401	return 0;
402}
403
404int ext4_bio_write_page(struct ext4_io_submit *io,
405			struct page *page,
406			int len,
407			struct writeback_control *wbc,
408			bool keep_towrite)
409{
410	struct page *bounce_page = NULL;
411	struct inode *inode = page->mapping->host;
412	unsigned block_start;
413	struct buffer_head *bh, *head;
414	int ret = 0;
415	int nr_submitted = 0;
416	int nr_to_submit = 0;
417
418	BUG_ON(!PageLocked(page));
419	BUG_ON(PageWriteback(page));
420
421	if (keep_towrite)
422		set_page_writeback_keepwrite(page);
423	else
424		set_page_writeback(page);
425	ClearPageError(page);
426
427	/*
428	 * Comments copied from block_write_full_page:
429	 *
430	 * The page straddles i_size.  It must be zeroed out on each and every
431	 * writepage invocation because it may be mmapped.  "A file is mapped
432	 * in multiples of the page size.  For a file that is not a multiple of
433	 * the page size, the remaining memory is zeroed when mapped, and
434	 * writes to that region are not written out to the file."
435	 */
436	if (len < PAGE_SIZE)
437		zero_user_segment(page, len, PAGE_SIZE);
438	/*
439	 * In the first loop we prepare and mark buffers to submit. We have to
440	 * mark all buffers in the page before submitting so that
441	 * end_page_writeback() cannot be called from ext4_bio_end_io() when IO
442	 * on the first buffer finishes and we are still working on submitting
443	 * the second buffer.
444	 */
445	bh = head = page_buffers(page);
446	do {
447		block_start = bh_offset(bh);
448		if (block_start >= len) {
449			clear_buffer_dirty(bh);
450			set_buffer_uptodate(bh);
451			continue;
452		}
453		if (!buffer_dirty(bh) || buffer_delay(bh) ||
454		    !buffer_mapped(bh) || buffer_unwritten(bh)) {
455			/* A hole? We can safely clear the dirty bit */
456			if (!buffer_mapped(bh))
457				clear_buffer_dirty(bh);
458			if (io->io_bio)
459				ext4_io_submit(io);
460			continue;
461		}
462		if (buffer_new(bh))
463			clear_buffer_new(bh);
 
 
464		set_buffer_async_write(bh);
465		nr_to_submit++;
466	} while ((bh = bh->b_this_page) != head);
467
468	bh = head = page_buffers(page);
469
470	/*
471	 * If any blocks are being written to an encrypted file, encrypt them
472	 * into a bounce page.  For simplicity, just encrypt until the last
473	 * block which might be needed.  This may cause some unneeded blocks
474	 * (e.g. holes) to be unnecessarily encrypted, but this is rare and
475	 * can't happen in the common case of blocksize == PAGE_SIZE.
476	 */
477	if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode) && nr_to_submit) {
478		gfp_t gfp_flags = GFP_NOFS;
479		unsigned int enc_bytes = round_up(len, i_blocksize(inode));
480
481	retry_encrypt:
482		bounce_page = fscrypt_encrypt_pagecache_blocks(page, enc_bytes,
483							       0, gfp_flags);
484		if (IS_ERR(bounce_page)) {
485			ret = PTR_ERR(bounce_page);
486			if (ret == -ENOMEM && wbc->sync_mode == WB_SYNC_ALL) {
487				if (io->io_bio) {
488					ext4_io_submit(io);
489					congestion_wait(BLK_RW_ASYNC, HZ/50);
490				}
491				gfp_flags |= __GFP_NOFAIL;
492				goto retry_encrypt;
493			}
494			bounce_page = NULL;
495			goto out;
496		}
497	}
498
499	/* Now submit buffers to write */
500	do {
501		if (!buffer_async_write(bh))
502			continue;
503		ret = io_submit_add_bh(io, inode, bounce_page ?: page, bh);
 
504		if (ret) {
505			/*
506			 * We only get here on ENOMEM.  Not much else
507			 * we can do but mark the page as dirty, and
508			 * better luck next time.
509			 */
510			break;
511		}
512		nr_submitted++;
513		clear_buffer_dirty(bh);
514	} while ((bh = bh->b_this_page) != head);
515
516	/* Error stopped previous loop? Clean up buffers... */
517	if (ret) {
518	out:
519		fscrypt_free_bounce_page(bounce_page);
 
520		printk_ratelimited(KERN_ERR "%s: ret = %d\n", __func__, ret);
521		redirty_page_for_writepage(wbc, page);
522		do {
523			clear_buffer_async_write(bh);
524			bh = bh->b_this_page;
525		} while (bh != head);
526	}
527	unlock_page(page);
528	/* Nothing submitted - we have to end page writeback */
529	if (!nr_submitted)
530		end_page_writeback(page);
531	return ret;
532}