xref: /linux/fs/ext4/file.c (revision 0d456bad36d42d16022be045c8a53ddbb59ee478)
1 /*
2  *  linux/fs/ext4/file.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/file.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  ext4 fs regular file handling primitives
16  *
17  *  64-bit file support on 64-bit platforms by Jakub Jelinek
18  *	(jj@sunsite.ms.mff.cuni.cz)
19  */
20 
21 #include <linux/time.h>
22 #include <linux/fs.h>
23 #include <linux/jbd2.h>
24 #include <linux/mount.h>
25 #include <linux/path.h>
26 #include <linux/quotaops.h>
27 #include <linux/pagevec.h>
28 #include "ext4.h"
29 #include "ext4_jbd2.h"
30 #include "xattr.h"
31 #include "acl.h"
32 
33 /*
34  * Called when an inode is released. Note that this is different
35  * from ext4_file_open: open gets called at every open, but release
36  * gets called only when /all/ the files are closed.
37  */
38 static int ext4_release_file(struct inode *inode, struct file *filp)
39 {
40 	if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
41 		ext4_alloc_da_blocks(inode);
42 		ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
43 	}
44 	/* if we are the last writer on the inode, drop the block reservation */
45 	if ((filp->f_mode & FMODE_WRITE) &&
46 			(atomic_read(&inode->i_writecount) == 1) &&
47 		        !EXT4_I(inode)->i_reserved_data_blocks)
48 	{
49 		down_write(&EXT4_I(inode)->i_data_sem);
50 		ext4_discard_preallocations(inode);
51 		up_write(&EXT4_I(inode)->i_data_sem);
52 	}
53 	if (is_dx(inode) && filp->private_data)
54 		ext4_htree_free_dir_info(filp->private_data);
55 
56 	return 0;
57 }
58 
59 void ext4_unwritten_wait(struct inode *inode)
60 {
61 	wait_queue_head_t *wq = ext4_ioend_wq(inode);
62 
63 	wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_unwritten) == 0));
64 }
65 
66 /*
67  * This tests whether the IO in question is block-aligned or not.
68  * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
69  * are converted to written only after the IO is complete.  Until they are
70  * mapped, these blocks appear as holes, so dio_zero_block() will assume that
71  * it needs to zero out portions of the start and/or end block.  If 2 AIO
72  * threads are at work on the same unwritten block, they must be synchronized
73  * or one thread will zero the other's data, causing corruption.
74  */
75 static int
76 ext4_unaligned_aio(struct inode *inode, const struct iovec *iov,
77 		   unsigned long nr_segs, loff_t pos)
78 {
79 	struct super_block *sb = inode->i_sb;
80 	int blockmask = sb->s_blocksize - 1;
81 	size_t count = iov_length(iov, nr_segs);
82 	loff_t final_size = pos + count;
83 
84 	if (pos >= inode->i_size)
85 		return 0;
86 
87 	if ((pos & blockmask) || (final_size & blockmask))
88 		return 1;
89 
90 	return 0;
91 }
92 
93 static ssize_t
94 ext4_file_dio_write(struct kiocb *iocb, const struct iovec *iov,
95 		    unsigned long nr_segs, loff_t pos)
96 {
97 	struct file *file = iocb->ki_filp;
98 	struct inode *inode = file->f_mapping->host;
99 	struct blk_plug plug;
100 	int unaligned_aio = 0;
101 	ssize_t ret;
102 	int overwrite = 0;
103 	size_t length = iov_length(iov, nr_segs);
104 
105 	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
106 	    !is_sync_kiocb(iocb))
107 		unaligned_aio = ext4_unaligned_aio(inode, iov, nr_segs, pos);
108 
109 	/* Unaligned direct AIO must be serialized; see comment above */
110 	if (unaligned_aio) {
111 		static unsigned long unaligned_warn_time;
112 
113 		/* Warn about this once per day */
114 		if (printk_timed_ratelimit(&unaligned_warn_time, 60*60*24*HZ))
115 			ext4_msg(inode->i_sb, KERN_WARNING,
116 				 "Unaligned AIO/DIO on inode %ld by %s; "
117 				 "performance will be poor.",
118 				 inode->i_ino, current->comm);
119 		mutex_lock(ext4_aio_mutex(inode));
120 		ext4_unwritten_wait(inode);
121 	}
122 
123 	BUG_ON(iocb->ki_pos != pos);
124 
125 	mutex_lock(&inode->i_mutex);
126 	blk_start_plug(&plug);
127 
128 	iocb->private = &overwrite;
129 
130 	/* check whether we do a DIO overwrite or not */
131 	if (ext4_should_dioread_nolock(inode) && !unaligned_aio &&
132 	    !file->f_mapping->nrpages && pos + length <= i_size_read(inode)) {
133 		struct ext4_map_blocks map;
134 		unsigned int blkbits = inode->i_blkbits;
135 		int err, len;
136 
137 		map.m_lblk = pos >> blkbits;
138 		map.m_len = (EXT4_BLOCK_ALIGN(pos + length, blkbits) >> blkbits)
139 			- map.m_lblk;
140 		len = map.m_len;
141 
142 		err = ext4_map_blocks(NULL, inode, &map, 0);
143 		/*
144 		 * 'err==len' means that all of blocks has been preallocated no
145 		 * matter they are initialized or not.  For excluding
146 		 * uninitialized extents, we need to check m_flags.  There are
147 		 * two conditions that indicate for initialized extents.
148 		 * 1) If we hit extent cache, EXT4_MAP_MAPPED flag is returned;
149 		 * 2) If we do a real lookup, non-flags are returned.
150 		 * So we should check these two conditions.
151 		 */
152 		if (err == len && (map.m_flags & EXT4_MAP_MAPPED))
153 			overwrite = 1;
154 	}
155 
156 	ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
157 	mutex_unlock(&inode->i_mutex);
158 
159 	if (ret > 0 || ret == -EIOCBQUEUED) {
160 		ssize_t err;
161 
162 		err = generic_write_sync(file, pos, ret);
163 		if (err < 0 && ret > 0)
164 			ret = err;
165 	}
166 	blk_finish_plug(&plug);
167 
168 	if (unaligned_aio)
169 		mutex_unlock(ext4_aio_mutex(inode));
170 
171 	return ret;
172 }
173 
174 static ssize_t
175 ext4_file_write(struct kiocb *iocb, const struct iovec *iov,
176 		unsigned long nr_segs, loff_t pos)
177 {
178 	struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
179 	ssize_t ret;
180 
181 	/*
182 	 * If we have encountered a bitmap-format file, the size limit
183 	 * is smaller than s_maxbytes, which is for extent-mapped files.
184 	 */
185 
186 	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
187 		struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
188 		size_t length = iov_length(iov, nr_segs);
189 
190 		if ((pos > sbi->s_bitmap_maxbytes ||
191 		    (pos == sbi->s_bitmap_maxbytes && length > 0)))
192 			return -EFBIG;
193 
194 		if (pos + length > sbi->s_bitmap_maxbytes) {
195 			nr_segs = iov_shorten((struct iovec *)iov, nr_segs,
196 					      sbi->s_bitmap_maxbytes - pos);
197 		}
198 	}
199 
200 	if (unlikely(iocb->ki_filp->f_flags & O_DIRECT))
201 		ret = ext4_file_dio_write(iocb, iov, nr_segs, pos);
202 	else
203 		ret = generic_file_aio_write(iocb, iov, nr_segs, pos);
204 
205 	return ret;
206 }
207 
208 static const struct vm_operations_struct ext4_file_vm_ops = {
209 	.fault		= filemap_fault,
210 	.page_mkwrite   = ext4_page_mkwrite,
211 	.remap_pages	= generic_file_remap_pages,
212 };
213 
214 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
215 {
216 	struct address_space *mapping = file->f_mapping;
217 
218 	if (!mapping->a_ops->readpage)
219 		return -ENOEXEC;
220 	file_accessed(file);
221 	vma->vm_ops = &ext4_file_vm_ops;
222 	return 0;
223 }
224 
225 static int ext4_file_open(struct inode * inode, struct file * filp)
226 {
227 	struct super_block *sb = inode->i_sb;
228 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
229 	struct ext4_inode_info *ei = EXT4_I(inode);
230 	struct vfsmount *mnt = filp->f_path.mnt;
231 	struct path path;
232 	char buf[64], *cp;
233 
234 	if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
235 		     !(sb->s_flags & MS_RDONLY))) {
236 		sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
237 		/*
238 		 * Sample where the filesystem has been mounted and
239 		 * store it in the superblock for sysadmin convenience
240 		 * when trying to sort through large numbers of block
241 		 * devices or filesystem images.
242 		 */
243 		memset(buf, 0, sizeof(buf));
244 		path.mnt = mnt;
245 		path.dentry = mnt->mnt_root;
246 		cp = d_path(&path, buf, sizeof(buf));
247 		if (!IS_ERR(cp)) {
248 			handle_t *handle;
249 			int err;
250 
251 			handle = ext4_journal_start_sb(sb, 1);
252 			if (IS_ERR(handle))
253 				return PTR_ERR(handle);
254 			err = ext4_journal_get_write_access(handle, sbi->s_sbh);
255 			if (err) {
256 				ext4_journal_stop(handle);
257 				return err;
258 			}
259 			strlcpy(sbi->s_es->s_last_mounted, cp,
260 				sizeof(sbi->s_es->s_last_mounted));
261 			ext4_handle_dirty_super(handle, sb);
262 			ext4_journal_stop(handle);
263 		}
264 	}
265 	/*
266 	 * Set up the jbd2_inode if we are opening the inode for
267 	 * writing and the journal is present
268 	 */
269 	if (sbi->s_journal && !ei->jinode && (filp->f_mode & FMODE_WRITE)) {
270 		struct jbd2_inode *jinode = jbd2_alloc_inode(GFP_KERNEL);
271 
272 		spin_lock(&inode->i_lock);
273 		if (!ei->jinode) {
274 			if (!jinode) {
275 				spin_unlock(&inode->i_lock);
276 				return -ENOMEM;
277 			}
278 			ei->jinode = jinode;
279 			jbd2_journal_init_jbd_inode(ei->jinode, inode);
280 			jinode = NULL;
281 		}
282 		spin_unlock(&inode->i_lock);
283 		if (unlikely(jinode != NULL))
284 			jbd2_free_inode(jinode);
285 	}
286 	return dquot_file_open(inode, filp);
287 }
288 
289 /*
290  * Here we use ext4_map_blocks() to get a block mapping for a extent-based
291  * file rather than ext4_ext_walk_space() because we can introduce
292  * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
293  * function.  When extent status tree has been fully implemented, it will
294  * track all extent status for a file and we can directly use it to
295  * retrieve the offset for SEEK_DATA/SEEK_HOLE.
296  */
297 
298 /*
299  * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
300  * lookup page cache to check whether or not there has some data between
301  * [startoff, endoff] because, if this range contains an unwritten extent,
302  * we determine this extent as a data or a hole according to whether the
303  * page cache has data or not.
304  */
305 static int ext4_find_unwritten_pgoff(struct inode *inode,
306 				     int whence,
307 				     struct ext4_map_blocks *map,
308 				     loff_t *offset)
309 {
310 	struct pagevec pvec;
311 	unsigned int blkbits;
312 	pgoff_t index;
313 	pgoff_t end;
314 	loff_t endoff;
315 	loff_t startoff;
316 	loff_t lastoff;
317 	int found = 0;
318 
319 	blkbits = inode->i_sb->s_blocksize_bits;
320 	startoff = *offset;
321 	lastoff = startoff;
322 	endoff = (map->m_lblk + map->m_len) << blkbits;
323 
324 	index = startoff >> PAGE_CACHE_SHIFT;
325 	end = endoff >> PAGE_CACHE_SHIFT;
326 
327 	pagevec_init(&pvec, 0);
328 	do {
329 		int i, num;
330 		unsigned long nr_pages;
331 
332 		num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
333 		nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
334 					  (pgoff_t)num);
335 		if (nr_pages == 0) {
336 			if (whence == SEEK_DATA)
337 				break;
338 
339 			BUG_ON(whence != SEEK_HOLE);
340 			/*
341 			 * If this is the first time to go into the loop and
342 			 * offset is not beyond the end offset, it will be a
343 			 * hole at this offset
344 			 */
345 			if (lastoff == startoff || lastoff < endoff)
346 				found = 1;
347 			break;
348 		}
349 
350 		/*
351 		 * If this is the first time to go into the loop and
352 		 * offset is smaller than the first page offset, it will be a
353 		 * hole at this offset.
354 		 */
355 		if (lastoff == startoff && whence == SEEK_HOLE &&
356 		    lastoff < page_offset(pvec.pages[0])) {
357 			found = 1;
358 			break;
359 		}
360 
361 		for (i = 0; i < nr_pages; i++) {
362 			struct page *page = pvec.pages[i];
363 			struct buffer_head *bh, *head;
364 
365 			/*
366 			 * If the current offset is not beyond the end of given
367 			 * range, it will be a hole.
368 			 */
369 			if (lastoff < endoff && whence == SEEK_HOLE &&
370 			    page->index > end) {
371 				found = 1;
372 				*offset = lastoff;
373 				goto out;
374 			}
375 
376 			lock_page(page);
377 
378 			if (unlikely(page->mapping != inode->i_mapping)) {
379 				unlock_page(page);
380 				continue;
381 			}
382 
383 			if (!page_has_buffers(page)) {
384 				unlock_page(page);
385 				continue;
386 			}
387 
388 			if (page_has_buffers(page)) {
389 				lastoff = page_offset(page);
390 				bh = head = page_buffers(page);
391 				do {
392 					if (buffer_uptodate(bh) ||
393 					    buffer_unwritten(bh)) {
394 						if (whence == SEEK_DATA)
395 							found = 1;
396 					} else {
397 						if (whence == SEEK_HOLE)
398 							found = 1;
399 					}
400 					if (found) {
401 						*offset = max_t(loff_t,
402 							startoff, lastoff);
403 						unlock_page(page);
404 						goto out;
405 					}
406 					lastoff += bh->b_size;
407 					bh = bh->b_this_page;
408 				} while (bh != head);
409 			}
410 
411 			lastoff = page_offset(page) + PAGE_SIZE;
412 			unlock_page(page);
413 		}
414 
415 		/*
416 		 * The no. of pages is less than our desired, that would be a
417 		 * hole in there.
418 		 */
419 		if (nr_pages < num && whence == SEEK_HOLE) {
420 			found = 1;
421 			*offset = lastoff;
422 			break;
423 		}
424 
425 		index = pvec.pages[i - 1]->index + 1;
426 		pagevec_release(&pvec);
427 	} while (index <= end);
428 
429 out:
430 	pagevec_release(&pvec);
431 	return found;
432 }
433 
434 /*
435  * ext4_seek_data() retrieves the offset for SEEK_DATA.
436  */
437 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
438 {
439 	struct inode *inode = file->f_mapping->host;
440 	struct ext4_map_blocks map;
441 	struct extent_status es;
442 	ext4_lblk_t start, last, end;
443 	loff_t dataoff, isize;
444 	int blkbits;
445 	int ret = 0;
446 
447 	mutex_lock(&inode->i_mutex);
448 
449 	isize = i_size_read(inode);
450 	if (offset >= isize) {
451 		mutex_unlock(&inode->i_mutex);
452 		return -ENXIO;
453 	}
454 
455 	blkbits = inode->i_sb->s_blocksize_bits;
456 	start = offset >> blkbits;
457 	last = start;
458 	end = isize >> blkbits;
459 	dataoff = offset;
460 
461 	do {
462 		map.m_lblk = last;
463 		map.m_len = end - last + 1;
464 		ret = ext4_map_blocks(NULL, inode, &map, 0);
465 		if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
466 			if (last != start)
467 				dataoff = last << blkbits;
468 			break;
469 		}
470 
471 		/*
472 		 * If there is a delay extent at this offset,
473 		 * it will be as a data.
474 		 */
475 		es.start = last;
476 		(void)ext4_es_find_extent(inode, &es);
477 		if (last >= es.start &&
478 		    last < es.start + es.len) {
479 			if (last != start)
480 				dataoff = last << blkbits;
481 			break;
482 		}
483 
484 		/*
485 		 * If there is a unwritten extent at this offset,
486 		 * it will be as a data or a hole according to page
487 		 * cache that has data or not.
488 		 */
489 		if (map.m_flags & EXT4_MAP_UNWRITTEN) {
490 			int unwritten;
491 			unwritten = ext4_find_unwritten_pgoff(inode, SEEK_DATA,
492 							      &map, &dataoff);
493 			if (unwritten)
494 				break;
495 		}
496 
497 		last++;
498 		dataoff = last << blkbits;
499 	} while (last <= end);
500 
501 	mutex_unlock(&inode->i_mutex);
502 
503 	if (dataoff > isize)
504 		return -ENXIO;
505 
506 	if (dataoff < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET))
507 		return -EINVAL;
508 	if (dataoff > maxsize)
509 		return -EINVAL;
510 
511 	if (dataoff != file->f_pos) {
512 		file->f_pos = dataoff;
513 		file->f_version = 0;
514 	}
515 
516 	return dataoff;
517 }
518 
519 /*
520  * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
521  */
522 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
523 {
524 	struct inode *inode = file->f_mapping->host;
525 	struct ext4_map_blocks map;
526 	struct extent_status es;
527 	ext4_lblk_t start, last, end;
528 	loff_t holeoff, isize;
529 	int blkbits;
530 	int ret = 0;
531 
532 	mutex_lock(&inode->i_mutex);
533 
534 	isize = i_size_read(inode);
535 	if (offset >= isize) {
536 		mutex_unlock(&inode->i_mutex);
537 		return -ENXIO;
538 	}
539 
540 	blkbits = inode->i_sb->s_blocksize_bits;
541 	start = offset >> blkbits;
542 	last = start;
543 	end = isize >> blkbits;
544 	holeoff = offset;
545 
546 	do {
547 		map.m_lblk = last;
548 		map.m_len = end - last + 1;
549 		ret = ext4_map_blocks(NULL, inode, &map, 0);
550 		if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
551 			last += ret;
552 			holeoff = last << blkbits;
553 			continue;
554 		}
555 
556 		/*
557 		 * If there is a delay extent at this offset,
558 		 * we will skip this extent.
559 		 */
560 		es.start = last;
561 		(void)ext4_es_find_extent(inode, &es);
562 		if (last >= es.start &&
563 		    last < es.start + es.len) {
564 			last = es.start + es.len;
565 			holeoff = last << blkbits;
566 			continue;
567 		}
568 
569 		/*
570 		 * If there is a unwritten extent at this offset,
571 		 * it will be as a data or a hole according to page
572 		 * cache that has data or not.
573 		 */
574 		if (map.m_flags & EXT4_MAP_UNWRITTEN) {
575 			int unwritten;
576 			unwritten = ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
577 							      &map, &holeoff);
578 			if (!unwritten) {
579 				last += ret;
580 				holeoff = last << blkbits;
581 				continue;
582 			}
583 		}
584 
585 		/* find a hole */
586 		break;
587 	} while (last <= end);
588 
589 	mutex_unlock(&inode->i_mutex);
590 
591 	if (holeoff > isize)
592 		holeoff = isize;
593 
594 	if (holeoff < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET))
595 		return -EINVAL;
596 	if (holeoff > maxsize)
597 		return -EINVAL;
598 
599 	if (holeoff != file->f_pos) {
600 		file->f_pos = holeoff;
601 		file->f_version = 0;
602 	}
603 
604 	return holeoff;
605 }
606 
607 /*
608  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
609  * by calling generic_file_llseek_size() with the appropriate maxbytes
610  * value for each.
611  */
612 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
613 {
614 	struct inode *inode = file->f_mapping->host;
615 	loff_t maxbytes;
616 
617 	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
618 		maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
619 	else
620 		maxbytes = inode->i_sb->s_maxbytes;
621 
622 	switch (whence) {
623 	case SEEK_SET:
624 	case SEEK_CUR:
625 	case SEEK_END:
626 		return generic_file_llseek_size(file, offset, whence,
627 						maxbytes, i_size_read(inode));
628 	case SEEK_DATA:
629 		return ext4_seek_data(file, offset, maxbytes);
630 	case SEEK_HOLE:
631 		return ext4_seek_hole(file, offset, maxbytes);
632 	}
633 
634 	return -EINVAL;
635 }
636 
637 const struct file_operations ext4_file_operations = {
638 	.llseek		= ext4_llseek,
639 	.read		= do_sync_read,
640 	.write		= do_sync_write,
641 	.aio_read	= generic_file_aio_read,
642 	.aio_write	= ext4_file_write,
643 	.unlocked_ioctl = ext4_ioctl,
644 #ifdef CONFIG_COMPAT
645 	.compat_ioctl	= ext4_compat_ioctl,
646 #endif
647 	.mmap		= ext4_file_mmap,
648 	.open		= ext4_file_open,
649 	.release	= ext4_release_file,
650 	.fsync		= ext4_sync_file,
651 	.splice_read	= generic_file_splice_read,
652 	.splice_write	= generic_file_splice_write,
653 	.fallocate	= ext4_fallocate,
654 };
655 
656 const struct inode_operations ext4_file_inode_operations = {
657 	.setattr	= ext4_setattr,
658 	.getattr	= ext4_getattr,
659 	.setxattr	= generic_setxattr,
660 	.getxattr	= generic_getxattr,
661 	.listxattr	= ext4_listxattr,
662 	.removexattr	= generic_removexattr,
663 	.get_acl	= ext4_get_acl,
664 	.fiemap		= ext4_fiemap,
665 };
666 
667