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	struct bio_vec *bvec;
65
66	bio_for_each_segment_all(bvec, bio, i) {
67		struct page *page = bvec->bv_page;
68#ifdef CONFIG_EXT4_FS_ENCRYPTION
69		struct page *data_page = NULL;
70		struct ext4_crypto_ctx *ctx = 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			ctx = (struct ext4_crypto_ctx *)page_private(data_page);
86			page = ctx->w.control_page;
87		}
88#endif
89
90		if (bio->bi_error) {
91			SetPageError(page);
92			set_bit(AS_EIO, &page->mapping->flags);
93		}
94		bh = head = page_buffers(page);
95		/*
96		 * We check all buffers in the page under BH_Uptodate_Lock
97		 * to avoid races with other end io clearing async_write flags
98		 */
99		local_irq_save(flags);
100		bit_spin_lock(BH_Uptodate_Lock, &head->b_state);
101		do {
102			if (bh_offset(bh) < bio_start ||
103			    bh_offset(bh) + bh->b_size > bio_end) {
104				if (buffer_async_write(bh))
105					under_io++;
106				continue;
107			}
108			clear_buffer_async_write(bh);
109			if (bio->bi_error)
110				buffer_io_error(bh);
111		} while ((bh = bh->b_this_page) != head);
112		bit_spin_unlock(BH_Uptodate_Lock, &head->b_state);
113		local_irq_restore(flags);
114		if (!under_io) {
115#ifdef CONFIG_EXT4_FS_ENCRYPTION
116			if (ctx)
117				ext4_restore_control_page(data_page);
118#endif
119			end_page_writeback(page);
120		}
121	}
122}
123
124static void ext4_release_io_end(ext4_io_end_t *io_end)
125{
126	struct bio *bio, *next_bio;
127
128	BUG_ON(!list_empty(&io_end->list));
129	BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
130	WARN_ON(io_end->handle);
131
132	if (atomic_dec_and_test(&EXT4_I(io_end->inode)->i_ioend_count))
133		wake_up_all(ext4_ioend_wq(io_end->inode));
134
135	for (bio = io_end->bio; bio; bio = next_bio) {
136		next_bio = bio->bi_private;
137		ext4_finish_bio(bio);
138		bio_put(bio);
139	}
140	kmem_cache_free(io_end_cachep, io_end);
141}
142
143static void ext4_clear_io_unwritten_flag(ext4_io_end_t *io_end)
144{
145	struct inode *inode = io_end->inode;
146
147	io_end->flag &= ~EXT4_IO_END_UNWRITTEN;
148	/* Wake up anyone waiting on unwritten extent conversion */
149	if (atomic_dec_and_test(&EXT4_I(inode)->i_unwritten))
150		wake_up_all(ext4_ioend_wq(inode));
151}
152
153/*
154 * Check a range of space and convert unwritten extents to written. Note that
155 * we are protected from truncate touching same part of extent tree by the
156 * fact that truncate code waits for all DIO to finish (thus exclusion from
157 * direct IO is achieved) and also waits for PageWriteback bits. Thus we
158 * cannot get to ext4_ext_truncate() before all IOs overlapping that range are
159 * completed (happens from ext4_free_ioend()).
160 */
161static int ext4_end_io(ext4_io_end_t *io)
162{
163	struct inode *inode = io->inode;
164	loff_t offset = io->offset;
165	ssize_t size = io->size;
166	handle_t *handle = io->handle;
167	int ret = 0;
168
169	ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
170		   "list->prev 0x%p\n",
171		   io, inode->i_ino, io->list.next, io->list.prev);
172
173	io->handle = NULL;	/* Following call will use up the handle */
174	ret = ext4_convert_unwritten_extents(handle, inode, offset, size);
175	if (ret < 0) {
176		ext4_msg(inode->i_sb, KERN_EMERG,
177			 "failed to convert unwritten extents to written "
178			 "extents -- potential data loss!  "
179			 "(inode %lu, offset %llu, size %zd, error %d)",
180			 inode->i_ino, offset, size, ret);
181	}
182	ext4_clear_io_unwritten_flag(io);
183	ext4_release_io_end(io);
184	return ret;
185}
186
187static void dump_completed_IO(struct inode *inode, struct list_head *head)
188{
189#ifdef	EXT4FS_DEBUG
190	struct list_head *cur, *before, *after;
191	ext4_io_end_t *io, *io0, *io1;
192
193	if (list_empty(head))
194		return;
195
196	ext4_debug("Dump inode %lu completed io list\n", inode->i_ino);
197	list_for_each_entry(io, head, list) {
198		cur = &io->list;
199		before = cur->prev;
200		io0 = container_of(before, ext4_io_end_t, list);
201		after = cur->next;
202		io1 = container_of(after, ext4_io_end_t, list);
203
204		ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
205			    io, inode->i_ino, io0, io1);
206	}
207#endif
208}
209
210/* Add the io_end to per-inode completed end_io list. */
211static void ext4_add_complete_io(ext4_io_end_t *io_end)
212{
213	struct ext4_inode_info *ei = EXT4_I(io_end->inode);
214	struct ext4_sb_info *sbi = EXT4_SB(io_end->inode->i_sb);
215	struct workqueue_struct *wq;
216	unsigned long flags;
217
218	/* Only reserved conversions from writeback should enter here */
219	WARN_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
220	WARN_ON(!io_end->handle && sbi->s_journal);
221	spin_lock_irqsave(&ei->i_completed_io_lock, flags);
222	wq = sbi->rsv_conversion_wq;
223	if (list_empty(&ei->i_rsv_conversion_list))
224		queue_work(wq, &ei->i_rsv_conversion_work);
225	list_add_tail(&io_end->list, &ei->i_rsv_conversion_list);
226	spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
227}
228
229static int ext4_do_flush_completed_IO(struct inode *inode,
230				      struct list_head *head)
231{
232	ext4_io_end_t *io;
233	struct list_head unwritten;
234	unsigned long flags;
235	struct ext4_inode_info *ei = EXT4_I(inode);
236	int err, ret = 0;
237
238	spin_lock_irqsave(&ei->i_completed_io_lock, flags);
239	dump_completed_IO(inode, head);
240	list_replace_init(head, &unwritten);
241	spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
242
243	while (!list_empty(&unwritten)) {
244		io = list_entry(unwritten.next, ext4_io_end_t, list);
245		BUG_ON(!(io->flag & EXT4_IO_END_UNWRITTEN));
246		list_del_init(&io->list);
247
248		err = ext4_end_io(io);
249		if (unlikely(!ret && err))
250			ret = err;
251	}
252	return ret;
253}
254
255/*
256 * work on completed IO, to convert unwritten extents to extents
257 */
258void ext4_end_io_rsv_work(struct work_struct *work)
259{
260	struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
261						  i_rsv_conversion_work);
262	ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_rsv_conversion_list);
263}
264
265ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
266{
267	ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags);
268	if (io) {
269		atomic_inc(&EXT4_I(inode)->i_ioend_count);
270		io->inode = inode;
271		INIT_LIST_HEAD(&io->list);
272		atomic_set(&io->count, 1);
273	}
274	return io;
275}
276
277void ext4_put_io_end_defer(ext4_io_end_t *io_end)
278{
279	if (atomic_dec_and_test(&io_end->count)) {
280		if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) || !io_end->size) {
281			ext4_release_io_end(io_end);
282			return;
283		}
284		ext4_add_complete_io(io_end);
285	}
286}
287
288int ext4_put_io_end(ext4_io_end_t *io_end)
289{
290	int err = 0;
291
292	if (atomic_dec_and_test(&io_end->count)) {
293		if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
294			err = ext4_convert_unwritten_extents(io_end->handle,
295						io_end->inode, io_end->offset,
296						io_end->size);
297			io_end->handle = NULL;
298			ext4_clear_io_unwritten_flag(io_end);
299		}
300		ext4_release_io_end(io_end);
301	}
302	return err;
303}
304
305ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end)
306{
307	atomic_inc(&io_end->count);
308	return io_end;
309}
310
311/* BIO completion function for page writeback */
312static void ext4_end_bio(struct bio *bio)
313{
314	ext4_io_end_t *io_end = bio->bi_private;
315	sector_t bi_sector = bio->bi_iter.bi_sector;
316
317	BUG_ON(!io_end);
318	bio->bi_end_io = NULL;
319
320	if (bio->bi_error) {
321		struct inode *inode = io_end->inode;
322
323		ext4_warning(inode->i_sb, "I/O error %d writing to inode %lu "
324			     "(offset %llu size %ld starting block %llu)",
325			     bio->bi_error, inode->i_ino,
326			     (unsigned long long) io_end->offset,
327			     (long) io_end->size,
328			     (unsigned long long)
329			     bi_sector >> (inode->i_blkbits - 9));
330		mapping_set_error(inode->i_mapping, bio->bi_error);
331	}
332
333	if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
334		/*
335		 * Link bio into list hanging from io_end. We have to do it
336		 * atomically as bio completions can be racing against each
337		 * other.
338		 */
339		bio->bi_private = xchg(&io_end->bio, bio);
340		ext4_put_io_end_defer(io_end);
341	} else {
342		/*
343		 * Drop io_end reference early. Inode can get freed once
344		 * we finish the bio.
345		 */
346		ext4_put_io_end_defer(io_end);
347		ext4_finish_bio(bio);
348		bio_put(bio);
349	}
350}
351
352void ext4_io_submit(struct ext4_io_submit *io)
353{
354	struct bio *bio = io->io_bio;
355
356	if (bio) {
357		int io_op = io->io_wbc->sync_mode == WB_SYNC_ALL ?
358			    WRITE_SYNC : WRITE;
359		bio_get(io->io_bio);
360		submit_bio(io_op, io->io_bio);
361		bio_put(io->io_bio);
362	}
363	io->io_bio = NULL;
364}
365
366void ext4_io_submit_init(struct ext4_io_submit *io,
367			 struct writeback_control *wbc)
368{
369	io->io_wbc = wbc;
370	io->io_bio = NULL;
371	io->io_end = NULL;
372}
373
374static int io_submit_init_bio(struct ext4_io_submit *io,
375			      struct buffer_head *bh)
376{
377	struct bio *bio;
378
379	bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
380	if (!bio)
381		return -ENOMEM;
382	wbc_init_bio(io->io_wbc, bio);
383	bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
384	bio->bi_bdev = bh->b_bdev;
385	bio->bi_end_io = ext4_end_bio;
386	bio->bi_private = ext4_get_io_end(io->io_end);
387	io->io_bio = bio;
388	io->io_next_block = bh->b_blocknr;
389	return 0;
390}
391
392static int io_submit_add_bh(struct ext4_io_submit *io,
393			    struct inode *inode,
394			    struct page *page,
395			    struct buffer_head *bh)
396{
397	int ret;
398
399	if (io->io_bio && bh->b_blocknr != io->io_next_block) {
400submit_and_retry:
401		ext4_io_submit(io);
402	}
403	if (io->io_bio == NULL) {
404		ret = io_submit_init_bio(io, bh);
405		if (ret)
406			return ret;
407	}
408	ret = bio_add_page(io->io_bio, page, bh->b_size, bh_offset(bh));
409	if (ret != bh->b_size)
410		goto submit_and_retry;
411	wbc_account_io(io->io_wbc, page, bh->b_size);
412	io->io_next_block++;
413	return 0;
414}
415
416int ext4_bio_write_page(struct ext4_io_submit *io,
417			struct page *page,
418			int len,
419			struct writeback_control *wbc,
420			bool keep_towrite)
421{
422	struct page *data_page = NULL;
423	struct inode *inode = page->mapping->host;
424	unsigned block_start, blocksize;
425	struct buffer_head *bh, *head;
426	int ret = 0;
427	int nr_submitted = 0;
428	int nr_to_submit = 0;
429
430	blocksize = 1 << inode->i_blkbits;
431
432	BUG_ON(!PageLocked(page));
433	BUG_ON(PageWriteback(page));
434
435	if (keep_towrite)
436		set_page_writeback_keepwrite(page);
437	else
438		set_page_writeback(page);
439	ClearPageError(page);
440
441	/*
442	 * Comments copied from block_write_full_page:
443	 *
444	 * The page straddles i_size.  It must be zeroed out on each and every
445	 * writepage invocation because it may be mmapped.  "A file is mapped
446	 * in multiples of the page size.  For a file that is not a multiple of
447	 * the page size, the remaining memory is zeroed when mapped, and
448	 * writes to that region are not written out to the file."
449	 */
450	if (len < PAGE_CACHE_SIZE)
451		zero_user_segment(page, len, PAGE_CACHE_SIZE);
452	/*
453	 * In the first loop we prepare and mark buffers to submit. We have to
454	 * mark all buffers in the page before submitting so that
455	 * end_page_writeback() cannot be called from ext4_bio_end_io() when IO
456	 * on the first buffer finishes and we are still working on submitting
457	 * the second buffer.
458	 */
459	bh = head = page_buffers(page);
460	do {
461		block_start = bh_offset(bh);
462		if (block_start >= len) {
463			clear_buffer_dirty(bh);
464			set_buffer_uptodate(bh);
465			continue;
466		}
467		if (!buffer_dirty(bh) || buffer_delay(bh) ||
468		    !buffer_mapped(bh) || buffer_unwritten(bh)) {
469			/* A hole? We can safely clear the dirty bit */
470			if (!buffer_mapped(bh))
471				clear_buffer_dirty(bh);
472			if (io->io_bio)
473				ext4_io_submit(io);
474			continue;
475		}
476		if (buffer_new(bh)) {
477			clear_buffer_new(bh);
478			unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
479		}
480		set_buffer_async_write(bh);
481		nr_to_submit++;
482	} while ((bh = bh->b_this_page) != head);
483
484	bh = head = page_buffers(page);
485
486	if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode) &&
487	    nr_to_submit) {
488		data_page = ext4_encrypt(inode, page);
489		if (IS_ERR(data_page)) {
490			ret = PTR_ERR(data_page);
491			data_page = NULL;
492			goto out;
493		}
494	}
495
496	/* Now submit buffers to write */
497	do {
498		if (!buffer_async_write(bh))
499			continue;
500		ret = io_submit_add_bh(io, inode,
501				       data_page ? data_page : page, bh);
502		if (ret) {
503			/*
504			 * We only get here on ENOMEM.  Not much else
505			 * we can do but mark the page as dirty, and
506			 * better luck next time.
507			 */
508			break;
509		}
510		nr_submitted++;
511		clear_buffer_dirty(bh);
512	} while ((bh = bh->b_this_page) != head);
513
514	/* Error stopped previous loop? Clean up buffers... */
515	if (ret) {
516	out:
517		if (data_page)
518			ext4_restore_control_page(data_page);
519		printk_ratelimited(KERN_ERR "%s: ret = %d\n", __func__, ret);
520		redirty_page_for_writepage(wbc, page);
521		do {
522			clear_buffer_async_write(bh);
523			bh = bh->b_this_page;
524		} while (bh != head);
525	}
526	unlock_page(page);
527	/* Nothing submitted - we have to end page writeback */
528	if (!nr_submitted)
529		end_page_writeback(page);
530	return ret;
531}
532