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 
ext4_init_pageio(void)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 
ext4_exit_pageio(void)48 void ext4_exit_pageio(void)
49 {
50 	kmem_cache_destroy(io_end_cachep);
51 	kmem_cache_destroy(io_page_cachep);
52 }
53 
ext4_ioend_wait(struct inode * inode)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 
put_io_page(struct ext4_io_page * io_page)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 
ext4_free_io_end(ext4_io_end_t * io)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  */
ext4_end_io_nolock(ext4_io_end_t * io)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  */
ext4_end_io_work(struct work_struct * work)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 
ext4_init_io_end(struct inode * inode,gfp_t flags)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  */
buffer_io_error(struct buffer_head * bh)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 
ext4_end_bio(struct bio * bio,int error)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 		int partial_write = 0;
207 
208 		head = page_buffers(page);
209 		if (error)
210 			SetPageError(page);
211 		BUG_ON(!head);
212 		if (head->b_size != PAGE_CACHE_SIZE) {
213 			loff_t offset;
214 			loff_t io_end_offset = io_end->offset + io_end->size;
215 
216 			offset = (sector_t) page->index << PAGE_CACHE_SHIFT;
217 			bh = head;
218 			do {
219 				if ((offset >= io_end->offset) &&
220 				    (offset+bh->b_size <= io_end_offset)) {
221 					if (error)
222 						buffer_io_error(bh);
223 
224 				}
225 				if (buffer_delay(bh))
226 					partial_write = 1;
227 				else if (!buffer_mapped(bh))
228 					clear_buffer_dirty(bh);
229 				else if (buffer_dirty(bh))
230 					partial_write = 1;
231 				offset += bh->b_size;
232 				bh = bh->b_this_page;
233 			} while (bh != head);
234 		}
235 
236 		/*
237 		 * If this is a partial write which happened to make
238 		 * all buffers uptodate then we can optimize away a
239 		 * bogus readpage() for the next read(). Here we
240 		 * 'discover' whether the page went uptodate as a
241 		 * result of this (potentially partial) write.
242 		 */
243 		if (!partial_write)
244 			SetPageUptodate(page);
245 
246 		put_io_page(io_end->pages[i]);
247 	}
248 	io_end->num_io_pages = 0;
249 	inode = io_end->inode;
250 
251 	if (error) {
252 		io_end->flag |= EXT4_IO_END_ERROR;
253 		ext4_warning(inode->i_sb, "I/O error writing to inode %lu "
254 			     "(offset %llu size %ld starting block %llu)",
255 			     inode->i_ino,
256 			     (unsigned long long) io_end->offset,
257 			     (long) io_end->size,
258 			     (unsigned long long)
259 			     bi_sector >> (inode->i_blkbits - 9));
260 	}
261 
262 	if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
263 		ext4_free_io_end(io_end);
264 		return;
265 	}
266 
267 	/* Add the io_end to per-inode completed io list*/
268 	spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
269 	list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
270 	spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
271 
272 	wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
273 	/* queue the work to convert unwritten extents to written */
274 	queue_work(wq, &io_end->work);
275 }
276 
ext4_io_submit(struct ext4_io_submit * io)277 void ext4_io_submit(struct ext4_io_submit *io)
278 {
279 	struct bio *bio = io->io_bio;
280 
281 	if (bio) {
282 		bio_get(io->io_bio);
283 		submit_bio(io->io_op, io->io_bio);
284 		BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP));
285 		bio_put(io->io_bio);
286 	}
287 	io->io_bio = NULL;
288 	io->io_op = 0;
289 	io->io_end = NULL;
290 }
291 
io_submit_init(struct ext4_io_submit * io,struct inode * inode,struct writeback_control * wbc,struct buffer_head * bh)292 static int io_submit_init(struct ext4_io_submit *io,
293 			  struct inode *inode,
294 			  struct writeback_control *wbc,
295 			  struct buffer_head *bh)
296 {
297 	ext4_io_end_t *io_end;
298 	struct page *page = bh->b_page;
299 	int nvecs = bio_get_nr_vecs(bh->b_bdev);
300 	struct bio *bio;
301 
302 	io_end = ext4_init_io_end(inode, GFP_NOFS);
303 	if (!io_end)
304 		return -ENOMEM;
305 	do {
306 		bio = bio_alloc(GFP_NOIO, nvecs);
307 		nvecs >>= 1;
308 	} while (bio == NULL);
309 
310 	bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
311 	bio->bi_bdev = bh->b_bdev;
312 	bio->bi_private = io->io_end = io_end;
313 	bio->bi_end_io = ext4_end_bio;
314 
315 	io_end->offset = (page->index << PAGE_CACHE_SHIFT) + bh_offset(bh);
316 
317 	io->io_bio = bio;
318 	io->io_op = (wbc->sync_mode == WB_SYNC_ALL ?  WRITE_SYNC : WRITE);
319 	io->io_next_block = bh->b_blocknr;
320 	return 0;
321 }
322 
io_submit_add_bh(struct ext4_io_submit * io,struct ext4_io_page * io_page,struct inode * inode,struct writeback_control * wbc,struct buffer_head * bh)323 static int io_submit_add_bh(struct ext4_io_submit *io,
324 			    struct ext4_io_page *io_page,
325 			    struct inode *inode,
326 			    struct writeback_control *wbc,
327 			    struct buffer_head *bh)
328 {
329 	ext4_io_end_t *io_end;
330 	int ret;
331 
332 	if (buffer_new(bh)) {
333 		clear_buffer_new(bh);
334 		unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
335 	}
336 
337 	if (!buffer_mapped(bh) || buffer_delay(bh)) {
338 		if (!buffer_mapped(bh))
339 			clear_buffer_dirty(bh);
340 		if (io->io_bio)
341 			ext4_io_submit(io);
342 		return 0;
343 	}
344 
345 	if (io->io_bio && bh->b_blocknr != io->io_next_block) {
346 submit_and_retry:
347 		ext4_io_submit(io);
348 	}
349 	if (io->io_bio == NULL) {
350 		ret = io_submit_init(io, inode, wbc, bh);
351 		if (ret)
352 			return ret;
353 	}
354 	io_end = io->io_end;
355 	if ((io_end->num_io_pages >= MAX_IO_PAGES) &&
356 	    (io_end->pages[io_end->num_io_pages-1] != io_page))
357 		goto submit_and_retry;
358 	if (buffer_uninit(bh))
359 		io->io_end->flag |= EXT4_IO_END_UNWRITTEN;
360 	io->io_end->size += bh->b_size;
361 	io->io_next_block++;
362 	ret = bio_add_page(io->io_bio, bh->b_page, bh->b_size, bh_offset(bh));
363 	if (ret != bh->b_size)
364 		goto submit_and_retry;
365 	if ((io_end->num_io_pages == 0) ||
366 	    (io_end->pages[io_end->num_io_pages-1] != io_page)) {
367 		io_end->pages[io_end->num_io_pages++] = io_page;
368 		atomic_inc(&io_page->p_count);
369 	}
370 	return 0;
371 }
372 
ext4_bio_write_page(struct ext4_io_submit * io,struct page * page,int len,struct writeback_control * wbc)373 int ext4_bio_write_page(struct ext4_io_submit *io,
374 			struct page *page,
375 			int len,
376 			struct writeback_control *wbc)
377 {
378 	struct inode *inode = page->mapping->host;
379 	unsigned block_start, block_end, blocksize;
380 	struct ext4_io_page *io_page;
381 	struct buffer_head *bh, *head;
382 	int ret = 0;
383 
384 	blocksize = 1 << inode->i_blkbits;
385 
386 	BUG_ON(!PageLocked(page));
387 	BUG_ON(PageWriteback(page));
388 
389 	io_page = kmem_cache_alloc(io_page_cachep, GFP_NOFS);
390 	if (!io_page) {
391 		set_page_dirty(page);
392 		unlock_page(page);
393 		return -ENOMEM;
394 	}
395 	io_page->p_page = page;
396 	atomic_set(&io_page->p_count, 1);
397 	get_page(page);
398 	set_page_writeback(page);
399 	ClearPageError(page);
400 
401 	for (bh = head = page_buffers(page), block_start = 0;
402 	     bh != head || !block_start;
403 	     block_start = block_end, bh = bh->b_this_page) {
404 
405 		block_end = block_start + blocksize;
406 		if (block_start >= len) {
407 			clear_buffer_dirty(bh);
408 			set_buffer_uptodate(bh);
409 			continue;
410 		}
411 		clear_buffer_dirty(bh);
412 		ret = io_submit_add_bh(io, io_page, inode, wbc, bh);
413 		if (ret) {
414 			/*
415 			 * We only get here on ENOMEM.  Not much else
416 			 * we can do but mark the page as dirty, and
417 			 * better luck next time.
418 			 */
419 			set_page_dirty(page);
420 			break;
421 		}
422 	}
423 	unlock_page(page);
424 	/*
425 	 * If the page was truncated before we could do the writeback,
426 	 * or we had a memory allocation error while trying to write
427 	 * the first buffer head, we won't have submitted any pages for
428 	 * I/O.  In that case we need to make sure we've cleared the
429 	 * PageWriteback bit from the page to prevent the system from
430 	 * wedging later on.
431 	 */
432 	put_io_page(io_page);
433 	return ret;
434 }
435