xref: /openbmc/linux/fs/ext4/page-io.c (revision dba8b469) 
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/module.h>
10 #include <linux/fs.h>
11 #include <linux/time.h>
12 #include <linux/jbd2.h>
13 #include <linux/highuid.h>
14 #include <linux/pagemap.h>
15 #include <linux/quotaops.h>
16 #include <linux/string.h>
17 #include <linux/buffer_head.h>
18 #include <linux/writeback.h>
19 #include <linux/pagevec.h>
20 #include <linux/mpage.h>
21 #include <linux/namei.h>
22 #include <linux/uio.h>
23 #include <linux/bio.h>
24 #include <linux/workqueue.h>
25 #include <linux/kernel.h>
26 #include <linux/slab.h>
27 
28 #include "ext4_jbd2.h"
29 #include "xattr.h"
30 #include "acl.h"
31 #include "ext4_extents.h"
32 
33 static struct kmem_cache *io_page_cachep, *io_end_cachep;
34 
35 int __init ext4_init_pageio(void)
36 {
37 	io_page_cachep = KMEM_CACHE(ext4_io_page, SLAB_RECLAIM_ACCOUNT);
38 	if (io_page_cachep == NULL)
39 		return -ENOMEM;
40 	io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT);
41 	if (io_end_cachep == NULL) {
42 		kmem_cache_destroy(io_page_cachep);
43 		return -ENOMEM;
44 	}
45 	return 0;
46 }
47 
48 void ext4_exit_pageio(void)
49 {
50 	kmem_cache_destroy(io_end_cachep);
51 	kmem_cache_destroy(io_page_cachep);
52 }
53 
54 void ext4_ioend_wait(struct inode *inode)
55 {
56 	wait_queue_head_t *wq = ext4_ioend_wq(inode);
57 
58 	wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_ioend_count) == 0));
59 }
60 
61 static void put_io_page(struct ext4_io_page *io_page)
62 {
63 	if (atomic_dec_and_test(&io_page->p_count)) {
64 		end_page_writeback(io_page->p_page);
65 		put_page(io_page->p_page);
66 		kmem_cache_free(io_page_cachep, io_page);
67 	}
68 }
69 
70 void ext4_free_io_end(ext4_io_end_t *io)
71 {
72 	int i;
73 	wait_queue_head_t *wq;
74 
75 	BUG_ON(!io);
76 	if (io->page)
77 		put_page(io->page);
78 	for (i = 0; i < io->num_io_pages; i++)
79 		put_io_page(io->pages[i]);
80 	io->num_io_pages = 0;
81 	wq = ext4_ioend_wq(io->inode);
82 	if (atomic_dec_and_test(&EXT4_I(io->inode)->i_ioend_count) &&
83 	    waitqueue_active(wq))
84 		wake_up_all(wq);
85 	kmem_cache_free(io_end_cachep, io);
86 }
87 
88 /*
89  * check a range of space and convert unwritten extents to written.
90  */
91 int ext4_end_io_nolock(ext4_io_end_t *io)
92 {
93 	struct inode *inode = io->inode;
94 	loff_t offset = io->offset;
95 	ssize_t size = io->size;
96 	wait_queue_head_t *wq;
97 	int ret = 0;
98 
99 	ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
100 		   "list->prev 0x%p\n",
101 		   io, inode->i_ino, io->list.next, io->list.prev);
102 
103 	if (list_empty(&io->list))
104 		return ret;
105 
106 	if (!(io->flag & EXT4_IO_END_UNWRITTEN))
107 		return ret;
108 
109 	ret = ext4_convert_unwritten_extents(inode, offset, size);
110 	if (ret < 0) {
111 		printk(KERN_EMERG "%s: failed to convert unwritten "
112 			"extents to written extents, error is %d "
113 			"io is still on inode %lu aio dio list\n",
114 		       __func__, ret, inode->i_ino);
115 		return ret;
116 	}
117 
118 	if (io->iocb)
119 		aio_complete(io->iocb, io->result, 0);
120 	/* clear the DIO AIO unwritten flag */
121 	if (io->flag & EXT4_IO_END_UNWRITTEN) {
122 		io->flag &= ~EXT4_IO_END_UNWRITTEN;
123 		/* Wake up anyone waiting on unwritten extent conversion */
124 		wq = ext4_ioend_wq(io->inode);
125 		if (atomic_dec_and_test(&EXT4_I(inode)->i_aiodio_unwritten) &&
126 		    waitqueue_active(wq)) {
127 			wake_up_all(wq);
128 		}
129 	}
130 
131 	return ret;
132 }
133 
134 /*
135  * work on completed aio dio IO, to convert unwritten extents to extents
136  */
137 static void ext4_end_io_work(struct work_struct *work)
138 {
139 	ext4_io_end_t		*io = container_of(work, ext4_io_end_t, work);
140 	struct inode		*inode = io->inode;
141 	struct ext4_inode_info	*ei = EXT4_I(inode);
142 	unsigned long		flags;
143 	int			ret;
144 
145 	mutex_lock(&inode->i_mutex);
146 	ret = ext4_end_io_nolock(io);
147 	if (ret < 0) {
148 		mutex_unlock(&inode->i_mutex);
149 		return;
150 	}
151 
152 	spin_lock_irqsave(&ei->i_completed_io_lock, flags);
153 	if (!list_empty(&io->list))
154 		list_del_init(&io->list);
155 	spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
156 	mutex_unlock(&inode->i_mutex);
157 	ext4_free_io_end(io);
158 }
159 
160 ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
161 {
162 	ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags);
163 	if (io) {
164 		atomic_inc(&EXT4_I(inode)->i_ioend_count);
165 		io->inode = inode;
166 		INIT_WORK(&io->work, ext4_end_io_work);
167 		INIT_LIST_HEAD(&io->list);
168 	}
169 	return io;
170 }
171 
172 /*
173  * Print an buffer I/O error compatible with the fs/buffer.c.  This
174  * provides compatibility with dmesg scrapers that look for a specific
175  * buffer I/O error message.  We really need a unified error reporting
176  * structure to userspace ala Digital Unix's uerf system, but it's
177  * probably not going to happen in my lifetime, due to LKML politics...
178  */
179 static void buffer_io_error(struct buffer_head *bh)
180 {
181 	char b[BDEVNAME_SIZE];
182 	printk(KERN_ERR "Buffer I/O error on device %s, logical block %llu\n",
183 			bdevname(bh->b_bdev, b),
184 			(unsigned long long)bh->b_blocknr);
185 }
186 
187 static void ext4_end_bio(struct bio *bio, int error)
188 {
189 	ext4_io_end_t *io_end = bio->bi_private;
190 	struct workqueue_struct *wq;
191 	struct inode *inode;
192 	unsigned long flags;
193 	int i;
194 	sector_t bi_sector = bio->bi_sector;
195 
196 	BUG_ON(!io_end);
197 	bio->bi_private = NULL;
198 	bio->bi_end_io = NULL;
199 	if (test_bit(BIO_UPTODATE, &bio->bi_flags))
200 		error = 0;
201 	bio_put(bio);
202 
203 	for (i = 0; i < io_end->num_io_pages; i++) {
204 		struct page *page = io_end->pages[i]->p_page;
205 		struct buffer_head *bh, *head;
206 		loff_t offset;
207 		loff_t io_end_offset;
208 
209 		if (error) {
210 			SetPageError(page);
211 			set_bit(AS_EIO, &page->mapping->flags);
212 			head = page_buffers(page);
213 			BUG_ON(!head);
214 
215 			io_end_offset = io_end->offset + io_end->size;
216 
217 			offset = (sector_t) page->index << PAGE_CACHE_SHIFT;
218 			bh = head;
219 			do {
220 				if ((offset >= io_end->offset) &&
221 				    (offset+bh->b_size <= io_end_offset))
222 					buffer_io_error(bh);
223 
224 				offset += bh->b_size;
225 				bh = bh->b_this_page;
226 			} while (bh != head);
227 		}
228 
229 		put_io_page(io_end->pages[i]);
230 	}
231 	io_end->num_io_pages = 0;
232 	inode = io_end->inode;
233 
234 	if (error) {
235 		io_end->flag |= EXT4_IO_END_ERROR;
236 		ext4_warning(inode->i_sb, "I/O error writing to inode %lu "
237 			     "(offset %llu size %ld starting block %llu)",
238 			     inode->i_ino,
239 			     (unsigned long long) io_end->offset,
240 			     (long) io_end->size,
241 			     (unsigned long long)
242 			     bi_sector >> (inode->i_blkbits - 9));
243 	}
244 
245 	if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
246 		ext4_free_io_end(io_end);
247 		return;
248 	}
249 
250 	/* Add the io_end to per-inode completed io list*/
251 	spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
252 	list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
253 	spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
254 
255 	wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
256 	/* queue the work to convert unwritten extents to written */
257 	queue_work(wq, &io_end->work);
258 }
259 
260 void ext4_io_submit(struct ext4_io_submit *io)
261 {
262 	struct bio *bio = io->io_bio;
263 
264 	if (bio) {
265 		bio_get(io->io_bio);
266 		submit_bio(io->io_op, io->io_bio);
267 		BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP));
268 		bio_put(io->io_bio);
269 	}
270 	io->io_bio = NULL;
271 	io->io_op = 0;
272 	io->io_end = NULL;
273 }
274 
275 static int io_submit_init(struct ext4_io_submit *io,
276 			  struct inode *inode,
277 			  struct writeback_control *wbc,
278 			  struct buffer_head *bh)
279 {
280 	ext4_io_end_t *io_end;
281 	struct page *page = bh->b_page;
282 	int nvecs = bio_get_nr_vecs(bh->b_bdev);
283 	struct bio *bio;
284 
285 	io_end = ext4_init_io_end(inode, GFP_NOFS);
286 	if (!io_end)
287 		return -ENOMEM;
288 	bio = bio_alloc(GFP_NOIO, min(nvecs, BIO_MAX_PAGES));
289 	bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
290 	bio->bi_bdev = bh->b_bdev;
291 	bio->bi_private = io->io_end = io_end;
292 	bio->bi_end_io = ext4_end_bio;
293 
294 	io_end->offset = (page->index << PAGE_CACHE_SHIFT) + bh_offset(bh);
295 
296 	io->io_bio = bio;
297 	io->io_op = (wbc->sync_mode == WB_SYNC_ALL ?  WRITE_SYNC : WRITE);
298 	io->io_next_block = bh->b_blocknr;
299 	return 0;
300 }
301 
302 static int io_submit_add_bh(struct ext4_io_submit *io,
303 			    struct ext4_io_page *io_page,
304 			    struct inode *inode,
305 			    struct writeback_control *wbc,
306 			    struct buffer_head *bh)
307 {
308 	ext4_io_end_t *io_end;
309 	int ret;
310 
311 	if (buffer_new(bh)) {
312 		clear_buffer_new(bh);
313 		unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
314 	}
315 
316 	if (!buffer_mapped(bh) || buffer_delay(bh)) {
317 		if (!buffer_mapped(bh))
318 			clear_buffer_dirty(bh);
319 		if (io->io_bio)
320 			ext4_io_submit(io);
321 		return 0;
322 	}
323 
324 	if (io->io_bio && bh->b_blocknr != io->io_next_block) {
325 submit_and_retry:
326 		ext4_io_submit(io);
327 	}
328 	if (io->io_bio == NULL) {
329 		ret = io_submit_init(io, inode, wbc, bh);
330 		if (ret)
331 			return ret;
332 	}
333 	io_end = io->io_end;
334 	if ((io_end->num_io_pages >= MAX_IO_PAGES) &&
335 	    (io_end->pages[io_end->num_io_pages-1] != io_page))
336 		goto submit_and_retry;
337 	if (buffer_uninit(bh))
338 		io->io_end->flag |= EXT4_IO_END_UNWRITTEN;
339 	io->io_end->size += bh->b_size;
340 	io->io_next_block++;
341 	ret = bio_add_page(io->io_bio, bh->b_page, bh->b_size, bh_offset(bh));
342 	if (ret != bh->b_size)
343 		goto submit_and_retry;
344 	if ((io_end->num_io_pages == 0) ||
345 	    (io_end->pages[io_end->num_io_pages-1] != io_page)) {
346 		io_end->pages[io_end->num_io_pages++] = io_page;
347 		atomic_inc(&io_page->p_count);
348 	}
349 	return 0;
350 }
351 
352 int ext4_bio_write_page(struct ext4_io_submit *io,
353 			struct page *page,
354 			int len,
355 			struct writeback_control *wbc)
356 {
357 	struct inode *inode = page->mapping->host;
358 	unsigned block_start, block_end, blocksize;
359 	struct ext4_io_page *io_page;
360 	struct buffer_head *bh, *head;
361 	int ret = 0;
362 
363 	blocksize = 1 << inode->i_blkbits;
364 
365 	BUG_ON(!PageLocked(page));
366 	BUG_ON(PageWriteback(page));
367 
368 	io_page = kmem_cache_alloc(io_page_cachep, GFP_NOFS);
369 	if (!io_page) {
370 		set_page_dirty(page);
371 		unlock_page(page);
372 		return -ENOMEM;
373 	}
374 	io_page->p_page = page;
375 	atomic_set(&io_page->p_count, 1);
376 	get_page(page);
377 	set_page_writeback(page);
378 	ClearPageError(page);
379 
380 	for (bh = head = page_buffers(page), block_start = 0;
381 	     bh != head || !block_start;
382 	     block_start = block_end, bh = bh->b_this_page) {
383 
384 		block_end = block_start + blocksize;
385 		if (block_start >= len) {
386 			clear_buffer_dirty(bh);
387 			set_buffer_uptodate(bh);
388 			continue;
389 		}
390 		clear_buffer_dirty(bh);
391 		ret = io_submit_add_bh(io, io_page, inode, wbc, bh);
392 		if (ret) {
393 			/*
394 			 * We only get here on ENOMEM.  Not much else
395 			 * we can do but mark the page as dirty, and
396 			 * better luck next time.
397 			 */
398 			set_page_dirty(page);
399 			break;
400 		}
401 	}
402 	unlock_page(page);
403 	/*
404 	 * If the page was truncated before we could do the writeback,
405 	 * or we had a memory allocation error while trying to write
406 	 * the first buffer head, we won't have submitted any pages for
407 	 * I/O.  In that case we need to make sure we've cleared the
408 	 * PageWriteback bit from the page to prevent the system from
409 	 * wedging later on.
410 	 */
411 	put_io_page(io_page);
412 	return ret;
413 }
414