xref: /linux/fs/ext4/file.c (revision 827634added7f38b7d724cab1dccdb2b004c13c3)
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/mount.h>
24 #include <linux/path.h>
25 #include <linux/quotaops.h>
26 #include <linux/pagevec.h>
27 #include <linux/uio.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 static 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, struct iov_iter *from, loff_t pos)
77 {
78 	struct super_block *sb = inode->i_sb;
79 	int blockmask = sb->s_blocksize - 1;
80 
81 	if (pos >= i_size_read(inode))
82 		return 0;
83 
84 	if ((pos | iov_iter_alignment(from)) & blockmask)
85 		return 1;
86 
87 	return 0;
88 }
89 
90 static ssize_t
91 ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
92 {
93 	struct file *file = iocb->ki_filp;
94 	struct inode *inode = file_inode(iocb->ki_filp);
95 	struct mutex *aio_mutex = NULL;
96 	struct blk_plug plug;
97 	int o_direct = iocb->ki_flags & IOCB_DIRECT;
98 	int overwrite = 0;
99 	ssize_t ret;
100 
101 	/*
102 	 * Unaligned direct AIO must be serialized; see comment above
103 	 * In the case of O_APPEND, assume that we must always serialize
104 	 */
105 	if (o_direct &&
106 	    ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
107 	    !is_sync_kiocb(iocb) &&
108 	    (iocb->ki_flags & IOCB_APPEND ||
109 	     ext4_unaligned_aio(inode, from, iocb->ki_pos))) {
110 		aio_mutex = ext4_aio_mutex(inode);
111 		mutex_lock(aio_mutex);
112 		ext4_unwritten_wait(inode);
113 	}
114 
115 	mutex_lock(&inode->i_mutex);
116 	ret = generic_write_checks(iocb, from);
117 	if (ret <= 0)
118 		goto out;
119 
120 	/*
121 	 * If we have encountered a bitmap-format file, the size limit
122 	 * is smaller than s_maxbytes, which is for extent-mapped files.
123 	 */
124 	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
125 		struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
126 
127 		if (iocb->ki_pos >= sbi->s_bitmap_maxbytes) {
128 			ret = -EFBIG;
129 			goto out;
130 		}
131 		iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos);
132 	}
133 
134 	iocb->private = &overwrite;
135 	if (o_direct) {
136 		size_t length = iov_iter_count(from);
137 		loff_t pos = iocb->ki_pos;
138 		blk_start_plug(&plug);
139 
140 		/* check whether we do a DIO overwrite or not */
141 		if (ext4_should_dioread_nolock(inode) && !aio_mutex &&
142 		    !file->f_mapping->nrpages && pos + length <= i_size_read(inode)) {
143 			struct ext4_map_blocks map;
144 			unsigned int blkbits = inode->i_blkbits;
145 			int err, len;
146 
147 			map.m_lblk = pos >> blkbits;
148 			map.m_len = (EXT4_BLOCK_ALIGN(pos + length, blkbits) >> blkbits)
149 				- map.m_lblk;
150 			len = map.m_len;
151 
152 			err = ext4_map_blocks(NULL, inode, &map, 0);
153 			/*
154 			 * 'err==len' means that all of blocks has
155 			 * been preallocated no matter they are
156 			 * initialized or not.  For excluding
157 			 * unwritten extents, we need to check
158 			 * m_flags.  There are two conditions that
159 			 * indicate for initialized extents.  1) If we
160 			 * hit extent cache, EXT4_MAP_MAPPED flag is
161 			 * returned; 2) If we do a real lookup,
162 			 * non-flags are returned.  So we should check
163 			 * these two conditions.
164 			 */
165 			if (err == len && (map.m_flags & EXT4_MAP_MAPPED))
166 				overwrite = 1;
167 		}
168 	}
169 
170 	ret = __generic_file_write_iter(iocb, from);
171 	mutex_unlock(&inode->i_mutex);
172 
173 	if (ret > 0) {
174 		ssize_t err;
175 
176 		err = generic_write_sync(file, iocb->ki_pos - ret, ret);
177 		if (err < 0)
178 			ret = err;
179 	}
180 	if (o_direct)
181 		blk_finish_plug(&plug);
182 
183 	if (aio_mutex)
184 		mutex_unlock(aio_mutex);
185 	return ret;
186 
187 out:
188 	mutex_unlock(&inode->i_mutex);
189 	if (aio_mutex)
190 		mutex_unlock(aio_mutex);
191 	return ret;
192 }
193 
194 #ifdef CONFIG_FS_DAX
195 static int ext4_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
196 {
197 	return dax_fault(vma, vmf, ext4_get_block);
198 					/* Is this the right get_block? */
199 }
200 
201 static int ext4_dax_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
202 {
203 	return dax_mkwrite(vma, vmf, ext4_get_block);
204 }
205 
206 static const struct vm_operations_struct ext4_dax_vm_ops = {
207 	.fault		= ext4_dax_fault,
208 	.page_mkwrite	= ext4_dax_mkwrite,
209 	.pfn_mkwrite	= dax_pfn_mkwrite,
210 };
211 #else
212 #define ext4_dax_vm_ops	ext4_file_vm_ops
213 #endif
214 
215 static const struct vm_operations_struct ext4_file_vm_ops = {
216 	.fault		= filemap_fault,
217 	.map_pages	= filemap_map_pages,
218 	.page_mkwrite   = ext4_page_mkwrite,
219 };
220 
221 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
222 {
223 	struct inode *inode = file->f_mapping->host;
224 
225 	if (ext4_encrypted_inode(inode)) {
226 		int err = ext4_generate_encryption_key(inode);
227 		if (err)
228 			return 0;
229 	}
230 	file_accessed(file);
231 	if (IS_DAX(file_inode(file))) {
232 		vma->vm_ops = &ext4_dax_vm_ops;
233 		vma->vm_flags |= VM_MIXEDMAP;
234 	} else {
235 		vma->vm_ops = &ext4_file_vm_ops;
236 	}
237 	return 0;
238 }
239 
240 static int ext4_file_open(struct inode * inode, struct file * filp)
241 {
242 	struct super_block *sb = inode->i_sb;
243 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
244 	struct vfsmount *mnt = filp->f_path.mnt;
245 	struct path path;
246 	char buf[64], *cp;
247 	int ret;
248 
249 	if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
250 		     !(sb->s_flags & MS_RDONLY))) {
251 		sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
252 		/*
253 		 * Sample where the filesystem has been mounted and
254 		 * store it in the superblock for sysadmin convenience
255 		 * when trying to sort through large numbers of block
256 		 * devices or filesystem images.
257 		 */
258 		memset(buf, 0, sizeof(buf));
259 		path.mnt = mnt;
260 		path.dentry = mnt->mnt_root;
261 		cp = d_path(&path, buf, sizeof(buf));
262 		if (!IS_ERR(cp)) {
263 			handle_t *handle;
264 			int err;
265 
266 			handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
267 			if (IS_ERR(handle))
268 				return PTR_ERR(handle);
269 			BUFFER_TRACE(sbi->s_sbh, "get_write_access");
270 			err = ext4_journal_get_write_access(handle, sbi->s_sbh);
271 			if (err) {
272 				ext4_journal_stop(handle);
273 				return err;
274 			}
275 			strlcpy(sbi->s_es->s_last_mounted, cp,
276 				sizeof(sbi->s_es->s_last_mounted));
277 			ext4_handle_dirty_super(handle, sb);
278 			ext4_journal_stop(handle);
279 		}
280 	}
281 	/*
282 	 * Set up the jbd2_inode if we are opening the inode for
283 	 * writing and the journal is present
284 	 */
285 	if (filp->f_mode & FMODE_WRITE) {
286 		ret = ext4_inode_attach_jinode(inode);
287 		if (ret < 0)
288 			return ret;
289 	}
290 	ret = dquot_file_open(inode, filp);
291 	if (!ret && ext4_encrypted_inode(inode)) {
292 		ret = ext4_generate_encryption_key(inode);
293 		if (ret)
294 			ret = -EACCES;
295 	}
296 	return ret;
297 }
298 
299 /*
300  * Here we use ext4_map_blocks() to get a block mapping for a extent-based
301  * file rather than ext4_ext_walk_space() because we can introduce
302  * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
303  * function.  When extent status tree has been fully implemented, it will
304  * track all extent status for a file and we can directly use it to
305  * retrieve the offset for SEEK_DATA/SEEK_HOLE.
306  */
307 
308 /*
309  * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
310  * lookup page cache to check whether or not there has some data between
311  * [startoff, endoff] because, if this range contains an unwritten extent,
312  * we determine this extent as a data or a hole according to whether the
313  * page cache has data or not.
314  */
315 static int ext4_find_unwritten_pgoff(struct inode *inode,
316 				     int whence,
317 				     struct ext4_map_blocks *map,
318 				     loff_t *offset)
319 {
320 	struct pagevec pvec;
321 	unsigned int blkbits;
322 	pgoff_t index;
323 	pgoff_t end;
324 	loff_t endoff;
325 	loff_t startoff;
326 	loff_t lastoff;
327 	int found = 0;
328 
329 	blkbits = inode->i_sb->s_blocksize_bits;
330 	startoff = *offset;
331 	lastoff = startoff;
332 	endoff = (loff_t)(map->m_lblk + map->m_len) << blkbits;
333 
334 	index = startoff >> PAGE_CACHE_SHIFT;
335 	end = endoff >> PAGE_CACHE_SHIFT;
336 
337 	pagevec_init(&pvec, 0);
338 	do {
339 		int i, num;
340 		unsigned long nr_pages;
341 
342 		num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
343 		nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
344 					  (pgoff_t)num);
345 		if (nr_pages == 0) {
346 			if (whence == SEEK_DATA)
347 				break;
348 
349 			BUG_ON(whence != SEEK_HOLE);
350 			/*
351 			 * If this is the first time to go into the loop and
352 			 * offset is not beyond the end offset, it will be a
353 			 * hole at this offset
354 			 */
355 			if (lastoff == startoff || lastoff < endoff)
356 				found = 1;
357 			break;
358 		}
359 
360 		/*
361 		 * If this is the first time to go into the loop and
362 		 * offset is smaller than the first page offset, it will be a
363 		 * hole at this offset.
364 		 */
365 		if (lastoff == startoff && whence == SEEK_HOLE &&
366 		    lastoff < page_offset(pvec.pages[0])) {
367 			found = 1;
368 			break;
369 		}
370 
371 		for (i = 0; i < nr_pages; i++) {
372 			struct page *page = pvec.pages[i];
373 			struct buffer_head *bh, *head;
374 
375 			/*
376 			 * If the current offset is not beyond the end of given
377 			 * range, it will be a hole.
378 			 */
379 			if (lastoff < endoff && whence == SEEK_HOLE &&
380 			    page->index > end) {
381 				found = 1;
382 				*offset = lastoff;
383 				goto out;
384 			}
385 
386 			lock_page(page);
387 
388 			if (unlikely(page->mapping != inode->i_mapping)) {
389 				unlock_page(page);
390 				continue;
391 			}
392 
393 			if (!page_has_buffers(page)) {
394 				unlock_page(page);
395 				continue;
396 			}
397 
398 			if (page_has_buffers(page)) {
399 				lastoff = page_offset(page);
400 				bh = head = page_buffers(page);
401 				do {
402 					if (buffer_uptodate(bh) ||
403 					    buffer_unwritten(bh)) {
404 						if (whence == SEEK_DATA)
405 							found = 1;
406 					} else {
407 						if (whence == SEEK_HOLE)
408 							found = 1;
409 					}
410 					if (found) {
411 						*offset = max_t(loff_t,
412 							startoff, lastoff);
413 						unlock_page(page);
414 						goto out;
415 					}
416 					lastoff += bh->b_size;
417 					bh = bh->b_this_page;
418 				} while (bh != head);
419 			}
420 
421 			lastoff = page_offset(page) + PAGE_SIZE;
422 			unlock_page(page);
423 		}
424 
425 		/*
426 		 * The no. of pages is less than our desired, that would be a
427 		 * hole in there.
428 		 */
429 		if (nr_pages < num && whence == SEEK_HOLE) {
430 			found = 1;
431 			*offset = lastoff;
432 			break;
433 		}
434 
435 		index = pvec.pages[i - 1]->index + 1;
436 		pagevec_release(&pvec);
437 	} while (index <= end);
438 
439 out:
440 	pagevec_release(&pvec);
441 	return found;
442 }
443 
444 /*
445  * ext4_seek_data() retrieves the offset for SEEK_DATA.
446  */
447 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
448 {
449 	struct inode *inode = file->f_mapping->host;
450 	struct ext4_map_blocks map;
451 	struct extent_status es;
452 	ext4_lblk_t start, last, end;
453 	loff_t dataoff, isize;
454 	int blkbits;
455 	int ret = 0;
456 
457 	mutex_lock(&inode->i_mutex);
458 
459 	isize = i_size_read(inode);
460 	if (offset >= isize) {
461 		mutex_unlock(&inode->i_mutex);
462 		return -ENXIO;
463 	}
464 
465 	blkbits = inode->i_sb->s_blocksize_bits;
466 	start = offset >> blkbits;
467 	last = start;
468 	end = isize >> blkbits;
469 	dataoff = offset;
470 
471 	do {
472 		map.m_lblk = last;
473 		map.m_len = end - last + 1;
474 		ret = ext4_map_blocks(NULL, inode, &map, 0);
475 		if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
476 			if (last != start)
477 				dataoff = (loff_t)last << blkbits;
478 			break;
479 		}
480 
481 		/*
482 		 * If there is a delay extent at this offset,
483 		 * it will be as a data.
484 		 */
485 		ext4_es_find_delayed_extent_range(inode, last, last, &es);
486 		if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
487 			if (last != start)
488 				dataoff = (loff_t)last << blkbits;
489 			break;
490 		}
491 
492 		/*
493 		 * If there is a unwritten extent at this offset,
494 		 * it will be as a data or a hole according to page
495 		 * cache that has data or not.
496 		 */
497 		if (map.m_flags & EXT4_MAP_UNWRITTEN) {
498 			int unwritten;
499 			unwritten = ext4_find_unwritten_pgoff(inode, SEEK_DATA,
500 							      &map, &dataoff);
501 			if (unwritten)
502 				break;
503 		}
504 
505 		last++;
506 		dataoff = (loff_t)last << blkbits;
507 	} while (last <= end);
508 
509 	mutex_unlock(&inode->i_mutex);
510 
511 	if (dataoff > isize)
512 		return -ENXIO;
513 
514 	return vfs_setpos(file, dataoff, maxsize);
515 }
516 
517 /*
518  * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
519  */
520 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
521 {
522 	struct inode *inode = file->f_mapping->host;
523 	struct ext4_map_blocks map;
524 	struct extent_status es;
525 	ext4_lblk_t start, last, end;
526 	loff_t holeoff, isize;
527 	int blkbits;
528 	int ret = 0;
529 
530 	mutex_lock(&inode->i_mutex);
531 
532 	isize = i_size_read(inode);
533 	if (offset >= isize) {
534 		mutex_unlock(&inode->i_mutex);
535 		return -ENXIO;
536 	}
537 
538 	blkbits = inode->i_sb->s_blocksize_bits;
539 	start = offset >> blkbits;
540 	last = start;
541 	end = isize >> blkbits;
542 	holeoff = offset;
543 
544 	do {
545 		map.m_lblk = last;
546 		map.m_len = end - last + 1;
547 		ret = ext4_map_blocks(NULL, inode, &map, 0);
548 		if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
549 			last += ret;
550 			holeoff = (loff_t)last << blkbits;
551 			continue;
552 		}
553 
554 		/*
555 		 * If there is a delay extent at this offset,
556 		 * we will skip this extent.
557 		 */
558 		ext4_es_find_delayed_extent_range(inode, last, last, &es);
559 		if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
560 			last = es.es_lblk + es.es_len;
561 			holeoff = (loff_t)last << blkbits;
562 			continue;
563 		}
564 
565 		/*
566 		 * If there is a unwritten extent at this offset,
567 		 * it will be as a data or a hole according to page
568 		 * cache that has data or not.
569 		 */
570 		if (map.m_flags & EXT4_MAP_UNWRITTEN) {
571 			int unwritten;
572 			unwritten = ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
573 							      &map, &holeoff);
574 			if (!unwritten) {
575 				last += ret;
576 				holeoff = (loff_t)last << blkbits;
577 				continue;
578 			}
579 		}
580 
581 		/* find a hole */
582 		break;
583 	} while (last <= end);
584 
585 	mutex_unlock(&inode->i_mutex);
586 
587 	if (holeoff > isize)
588 		holeoff = isize;
589 
590 	return vfs_setpos(file, holeoff, maxsize);
591 }
592 
593 /*
594  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
595  * by calling generic_file_llseek_size() with the appropriate maxbytes
596  * value for each.
597  */
598 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
599 {
600 	struct inode *inode = file->f_mapping->host;
601 	loff_t maxbytes;
602 
603 	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
604 		maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
605 	else
606 		maxbytes = inode->i_sb->s_maxbytes;
607 
608 	switch (whence) {
609 	case SEEK_SET:
610 	case SEEK_CUR:
611 	case SEEK_END:
612 		return generic_file_llseek_size(file, offset, whence,
613 						maxbytes, i_size_read(inode));
614 	case SEEK_DATA:
615 		return ext4_seek_data(file, offset, maxbytes);
616 	case SEEK_HOLE:
617 		return ext4_seek_hole(file, offset, maxbytes);
618 	}
619 
620 	return -EINVAL;
621 }
622 
623 const struct file_operations ext4_file_operations = {
624 	.llseek		= ext4_llseek,
625 	.read_iter	= generic_file_read_iter,
626 	.write_iter	= ext4_file_write_iter,
627 	.unlocked_ioctl = ext4_ioctl,
628 #ifdef CONFIG_COMPAT
629 	.compat_ioctl	= ext4_compat_ioctl,
630 #endif
631 	.mmap		= ext4_file_mmap,
632 	.open		= ext4_file_open,
633 	.release	= ext4_release_file,
634 	.fsync		= ext4_sync_file,
635 	.splice_read	= generic_file_splice_read,
636 	.splice_write	= iter_file_splice_write,
637 	.fallocate	= ext4_fallocate,
638 };
639 
640 const struct inode_operations ext4_file_inode_operations = {
641 	.setattr	= ext4_setattr,
642 	.getattr	= ext4_getattr,
643 	.setxattr	= generic_setxattr,
644 	.getxattr	= generic_getxattr,
645 	.listxattr	= ext4_listxattr,
646 	.removexattr	= generic_removexattr,
647 	.get_acl	= ext4_get_acl,
648 	.set_acl	= ext4_set_acl,
649 	.fiemap		= ext4_fiemap,
650 };
651 
652