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
27#include "ext4_jbd2.h"
28#include "xattr.h"
29#include "acl.h"
30
31static struct kmem_cache *io_end_cachep;
32
33int __init ext4_init_pageio(void)
34{
35	io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT);
36	if (io_end_cachep == NULL)
37		return -ENOMEM;
38	return 0;
39}
40
41void ext4_exit_pageio(void)
42{
43	kmem_cache_destroy(io_end_cachep);
44}
45
46/*
47 * Print an buffer I/O error compatible with the fs/buffer.c.  This
48 * provides compatibility with dmesg scrapers that look for a specific
49 * buffer I/O error message.  We really need a unified error reporting
50 * structure to userspace ala Digital Unix's uerf system, but it's
51 * probably not going to happen in my lifetime, due to LKML politics...
52 */
53static void buffer_io_error(struct buffer_head *bh)
54{
55	char b[BDEVNAME_SIZE];
56	printk_ratelimited(KERN_ERR "Buffer I/O error on device %s, logical block %llu\n",
57			bdevname(bh->b_bdev, b),
58			(unsigned long long)bh->b_blocknr);
59}
60
61static void ext4_finish_bio(struct bio *bio)
62{
63	int i;
64	int error = !test_bit(BIO_UPTODATE, &bio->bi_flags);
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		struct ext4_crypto_ctx *ctx = NULL;
72#endif
73		struct buffer_head *bh, *head;
74		unsigned bio_start = bvec->bv_offset;
75		unsigned bio_end = bio_start + bvec->bv_len;
76		unsigned under_io = 0;
77		unsigned long flags;
78
79		if (!page)
80			continue;
81
82#ifdef CONFIG_EXT4_FS_ENCRYPTION
83		if (!page->mapping) {
84			/* The bounce data pages are unmapped. */
85			data_page = page;
86			ctx = (struct ext4_crypto_ctx *)page_private(data_page);
87			page = ctx->control_page;
88		}
89#endif
90
91		if (error) {
92			SetPageError(page);
93			set_bit(AS_EIO, &page->mapping->flags);
94		}
95		bh = head = page_buffers(page);
96		/*
97		 * We check all buffers in the page under BH_Uptodate_Lock
98		 * to avoid races with other end io clearing async_write flags
99		 */
100		local_irq_save(flags);
101		bit_spin_lock(BH_Uptodate_Lock, &head->b_state);
102		do {
103			if (bh_offset(bh) < bio_start ||
104			    bh_offset(bh) + bh->b_size > bio_end) {
105				if (buffer_async_write(bh))
106					under_io++;
107				continue;
108			}
109			clear_buffer_async_write(bh);
110			if (error)
111				buffer_io_error(bh);
112		} while ((bh = bh->b_this_page) != head);
113		bit_spin_unlock(BH_Uptodate_Lock, &head->b_state);
114		local_irq_restore(flags);
115		if (!under_io) {
116#ifdef CONFIG_EXT4_FS_ENCRYPTION
117			if (ctx)
118				ext4_restore_control_page(data_page);
119#endif
120			end_page_writeback(page);
121		}
122	}
123}
124
125static void ext4_release_io_end(ext4_io_end_t *io_end)
126{
127	struct bio *bio, *next_bio;
128
129	BUG_ON(!list_empty(&io_end->list));
130	BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
131	WARN_ON(io_end->handle);
132
133	if (atomic_dec_and_test(&EXT4_I(io_end->inode)->i_ioend_count))
134		wake_up_all(ext4_ioend_wq(io_end->inode));
135
136	for (bio = io_end->bio; bio; bio = next_bio) {
137		next_bio = bio->bi_private;
138		ext4_finish_bio(bio);
139		bio_put(bio);
140	}
141	kmem_cache_free(io_end_cachep, io_end);
142}
143
144static void ext4_clear_io_unwritten_flag(ext4_io_end_t *io_end)
145{
146	struct inode *inode = io_end->inode;
147
148	io_end->flag &= ~EXT4_IO_END_UNWRITTEN;
149	/* Wake up anyone waiting on unwritten extent conversion */
150	if (atomic_dec_and_test(&EXT4_I(inode)->i_unwritten))
151		wake_up_all(ext4_ioend_wq(inode));
152}
153
154/*
155 * Check a range of space and convert unwritten extents to written. Note that
156 * we are protected from truncate touching same part of extent tree by the
157 * fact that truncate code waits for all DIO to finish (thus exclusion from
158 * direct IO is achieved) and also waits for PageWriteback bits. Thus we
159 * cannot get to ext4_ext_truncate() before all IOs overlapping that range are
160 * completed (happens from ext4_free_ioend()).
161 */
162static int ext4_end_io(ext4_io_end_t *io)
163{
164	struct inode *inode = io->inode;
165	loff_t offset = io->offset;
166	ssize_t size = io->size;
167	handle_t *handle = io->handle;
168	int ret = 0;
169
170	ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
171		   "list->prev 0x%p\n",
172		   io, inode->i_ino, io->list.next, io->list.prev);
173
174	io->handle = NULL;	/* Following call will use up the handle */
175	ret = ext4_convert_unwritten_extents(handle, inode, offset, size);
176	if (ret < 0) {
177		ext4_msg(inode->i_sb, KERN_EMERG,
178			 "failed to convert unwritten extents to written "
179			 "extents -- potential data loss!  "
180			 "(inode %lu, offset %llu, size %zd, error %d)",
181			 inode->i_ino, offset, size, ret);
182	}
183	ext4_clear_io_unwritten_flag(io);
184	ext4_release_io_end(io);
185	return ret;
186}
187
188static void dump_completed_IO(struct inode *inode, struct list_head *head)
189{
190#ifdef	EXT4FS_DEBUG
191	struct list_head *cur, *before, *after;
192	ext4_io_end_t *io, *io0, *io1;
193
194	if (list_empty(head))
195		return;
196
197	ext4_debug("Dump inode %lu completed io list\n", inode->i_ino);
198	list_for_each_entry(io, head, list) {
199		cur = &io->list;
200		before = cur->prev;
201		io0 = container_of(before, ext4_io_end_t, list);
202		after = cur->next;
203		io1 = container_of(after, ext4_io_end_t, list);
204
205		ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
206			    io, inode->i_ino, io0, io1);
207	}
208#endif
209}
210
211/* Add the io_end to per-inode completed end_io list. */
212static void ext4_add_complete_io(ext4_io_end_t *io_end)
213{
214	struct ext4_inode_info *ei = EXT4_I(io_end->inode);
215	struct ext4_sb_info *sbi = EXT4_SB(io_end->inode->i_sb);
216	struct workqueue_struct *wq;
217	unsigned long flags;
218
219	/* Only reserved conversions from writeback should enter here */
220	WARN_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
221	WARN_ON(!io_end->handle && sbi->s_journal);
222	spin_lock_irqsave(&ei->i_completed_io_lock, flags);
223	wq = sbi->rsv_conversion_wq;
224	if (list_empty(&ei->i_rsv_conversion_list))
225		queue_work(wq, &ei->i_rsv_conversion_work);
226	list_add_tail(&io_end->list, &ei->i_rsv_conversion_list);
227	spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
228}
229
230static int ext4_do_flush_completed_IO(struct inode *inode,
231				      struct list_head *head)
232{
233	ext4_io_end_t *io;
234	struct list_head unwritten;
235	unsigned long flags;
236	struct ext4_inode_info *ei = EXT4_I(inode);
237	int err, ret = 0;
238
239	spin_lock_irqsave(&ei->i_completed_io_lock, flags);
240	dump_completed_IO(inode, head);
241	list_replace_init(head, &unwritten);
242	spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
243
244	while (!list_empty(&unwritten)) {
245		io = list_entry(unwritten.next, ext4_io_end_t, list);
246		BUG_ON(!(io->flag & EXT4_IO_END_UNWRITTEN));
247		list_del_init(&io->list);
248
249		err = ext4_end_io(io);
250		if (unlikely(!ret && err))
251			ret = err;
252	}
253	return ret;
254}
255
256/*
257 * work on completed IO, to convert unwritten extents to extents
258 */
259void ext4_end_io_rsv_work(struct work_struct *work)
260{
261	struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
262						  i_rsv_conversion_work);
263	ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_rsv_conversion_list);
264}
265
266ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
267{
268	ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags);
269	if (io) {
270		atomic_inc(&EXT4_I(inode)->i_ioend_count);
271		io->inode = inode;
272		INIT_LIST_HEAD(&io->list);
273		atomic_set(&io->count, 1);
274	}
275	return io;
276}
277
278void ext4_put_io_end_defer(ext4_io_end_t *io_end)
279{
280	if (atomic_dec_and_test(&io_end->count)) {
281		if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) || !io_end->size) {
282			ext4_release_io_end(io_end);
283			return;
284		}
285		ext4_add_complete_io(io_end);
286	}
287}
288
289int ext4_put_io_end(ext4_io_end_t *io_end)
290{
291	int err = 0;
292
293	if (atomic_dec_and_test(&io_end->count)) {
294		if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
295			err = ext4_convert_unwritten_extents(io_end->handle,
296						io_end->inode, io_end->offset,
297						io_end->size);
298			io_end->handle = NULL;
299			ext4_clear_io_unwritten_flag(io_end);
300		}
301		ext4_release_io_end(io_end);
302	}
303	return err;
304}
305
306ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end)
307{
308	atomic_inc(&io_end->count);
309	return io_end;
310}
311
312/* BIO completion function for page writeback */
313static void ext4_end_bio(struct bio *bio, int error)
314{
315	ext4_io_end_t *io_end = bio->bi_private;
316	sector_t bi_sector = bio->bi_iter.bi_sector;
317
318	BUG_ON(!io_end);
319	bio->bi_end_io = NULL;
320	if (test_bit(BIO_UPTODATE, &bio->bi_flags))
321		error = 0;
322
323	if (error) {
324		struct inode *inode = io_end->inode;
325
326		ext4_warning(inode->i_sb, "I/O error %d writing to inode %lu "
327			     "(offset %llu size %ld starting block %llu)",
328			     error, inode->i_ino,
329			     (unsigned long long) io_end->offset,
330			     (long) io_end->size,
331			     (unsigned long long)
332			     bi_sector >> (inode->i_blkbits - 9));
333		mapping_set_error(inode->i_mapping, error);
334	}
335
336	if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
337		/*
338		 * Link bio into list hanging from io_end. We have to do it
339		 * atomically as bio completions can be racing against each
340		 * other.
341		 */
342		bio->bi_private = xchg(&io_end->bio, bio);
343		ext4_put_io_end_defer(io_end);
344	} else {
345		/*
346		 * Drop io_end reference early. Inode can get freed once
347		 * we finish the bio.
348		 */
349		ext4_put_io_end_defer(io_end);
350		ext4_finish_bio(bio);
351		bio_put(bio);
352	}
353}
354
355void ext4_io_submit(struct ext4_io_submit *io)
356{
357	struct bio *bio = io->io_bio;
358
359	if (bio) {
360		bio_get(io->io_bio);
361		submit_bio(io->io_op, io->io_bio);
362		BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP));
363		bio_put(io->io_bio);
364	}
365	io->io_bio = NULL;
366}
367
368void ext4_io_submit_init(struct ext4_io_submit *io,
369			 struct writeback_control *wbc)
370{
371	io->io_op = (wbc->sync_mode == WB_SYNC_ALL ?  WRITE_SYNC : WRITE);
372	io->io_bio = NULL;
373	io->io_end = NULL;
374}
375
376static int io_submit_init_bio(struct ext4_io_submit *io,
377			      struct buffer_head *bh)
378{
379	int nvecs = bio_get_nr_vecs(bh->b_bdev);
380	struct bio *bio;
381
382	bio = bio_alloc(GFP_NOIO, min(nvecs, BIO_MAX_PAGES));
383	if (!bio)
384		return -ENOMEM;
385	bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
386	bio->bi_bdev = bh->b_bdev;
387	bio->bi_end_io = ext4_end_bio;
388	bio->bi_private = ext4_get_io_end(io->io_end);
389	io->io_bio = bio;
390	io->io_next_block = bh->b_blocknr;
391	return 0;
392}
393
394static int io_submit_add_bh(struct ext4_io_submit *io,
395			    struct inode *inode,
396			    struct page *page,
397			    struct buffer_head *bh)
398{
399	int ret;
400
401	if (io->io_bio && bh->b_blocknr != io->io_next_block) {
402submit_and_retry:
403		ext4_io_submit(io);
404	}
405	if (io->io_bio == NULL) {
406		ret = io_submit_init_bio(io, bh);
407		if (ret)
408			return ret;
409	}
410	ret = bio_add_page(io->io_bio, page, bh->b_size, bh_offset(bh));
411	if (ret != bh->b_size)
412		goto submit_and_retry;
413	io->io_next_block++;
414	return 0;
415}
416
417int ext4_bio_write_page(struct ext4_io_submit *io,
418			struct page *page,
419			int len,
420			struct writeback_control *wbc,
421			bool keep_towrite)
422{
423	struct page *data_page = NULL;
424	struct inode *inode = page->mapping->host;
425	unsigned block_start, blocksize;
426	struct buffer_head *bh, *head;
427	int ret = 0;
428	int nr_submitted = 0;
429	int nr_to_submit = 0;
430
431	blocksize = 1 << inode->i_blkbits;
432
433	BUG_ON(!PageLocked(page));
434	BUG_ON(PageWriteback(page));
435
436	if (keep_towrite)
437		set_page_writeback_keepwrite(page);
438	else
439		set_page_writeback(page);
440	ClearPageError(page);
441
442	/*
443	 * Comments copied from block_write_full_page:
444	 *
445	 * The page straddles i_size.  It must be zeroed out on each and every
446	 * writepage invocation because it may be mmapped.  "A file is mapped
447	 * in multiples of the page size.  For a file that is not a multiple of
448	 * the page size, the remaining memory is zeroed when mapped, and
449	 * writes to that region are not written out to the file."
450	 */
451	if (len < PAGE_CACHE_SIZE)
452		zero_user_segment(page, len, PAGE_CACHE_SIZE);
453	/*
454	 * In the first loop we prepare and mark buffers to submit. We have to
455	 * mark all buffers in the page before submitting so that
456	 * end_page_writeback() cannot be called from ext4_bio_end_io() when IO
457	 * on the first buffer finishes and we are still working on submitting
458	 * the second buffer.
459	 */
460	bh = head = page_buffers(page);
461	do {
462		block_start = bh_offset(bh);
463		if (block_start >= len) {
464			clear_buffer_dirty(bh);
465			set_buffer_uptodate(bh);
466			continue;
467		}
468		if (!buffer_dirty(bh) || buffer_delay(bh) ||
469		    !buffer_mapped(bh) || buffer_unwritten(bh)) {
470			/* A hole? We can safely clear the dirty bit */
471			if (!buffer_mapped(bh))
472				clear_buffer_dirty(bh);
473			if (io->io_bio)
474				ext4_io_submit(io);
475			continue;
476		}
477		if (buffer_new(bh)) {
478			clear_buffer_new(bh);
479			unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
480		}
481		set_buffer_async_write(bh);
482		nr_to_submit++;
483	} while ((bh = bh->b_this_page) != head);
484
485	bh = head = page_buffers(page);
486
487	if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode) &&
488	    nr_to_submit) {
489		data_page = ext4_encrypt(inode, page);
490		if (IS_ERR(data_page)) {
491			ret = PTR_ERR(data_page);
492			data_page = NULL;
493			goto out;
494		}
495	}
496
497	/* Now submit buffers to write */
498	do {
499		if (!buffer_async_write(bh))
500			continue;
501		ret = io_submit_add_bh(io, inode,
502				       data_page ? data_page : page, bh);
503		if (ret) {
504			/*
505			 * We only get here on ENOMEM.  Not much else
506			 * we can do but mark the page as dirty, and
507			 * better luck next time.
508			 */
509			break;
510		}
511		nr_submitted++;
512		clear_buffer_dirty(bh);
513	} while ((bh = bh->b_this_page) != head);
514
515	/* Error stopped previous loop? Clean up buffers... */
516	if (ret) {
517	out:
518		if (data_page)
519			ext4_restore_control_page(data_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}
533