xref: /openbmc/linux/fs/ext4/file.c (revision 034f90b3)
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/aio.h>
27 #include <linux/quotaops.h>
28 #include <linux/pagevec.h>
29 #include "ext4.h"
30 #include "ext4_jbd2.h"
31 #include "xattr.h"
32 #include "acl.h"
33 
34 /*
35  * Called when an inode is released. Note that this is different
36  * from ext4_file_open: open gets called at every open, but release
37  * gets called only when /all/ the files are closed.
38  */
39 static int ext4_release_file(struct inode *inode, struct file *filp)
40 {
41 	if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
42 		ext4_alloc_da_blocks(inode);
43 		ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
44 	}
45 	/* if we are the last writer on the inode, drop the block reservation */
46 	if ((filp->f_mode & FMODE_WRITE) &&
47 			(atomic_read(&inode->i_writecount) == 1) &&
48 		        !EXT4_I(inode)->i_reserved_data_blocks)
49 	{
50 		down_write(&EXT4_I(inode)->i_data_sem);
51 		ext4_discard_preallocations(inode);
52 		up_write(&EXT4_I(inode)->i_data_sem);
53 	}
54 	if (is_dx(inode) && filp->private_data)
55 		ext4_htree_free_dir_info(filp->private_data);
56 
57 	return 0;
58 }
59 
60 static void ext4_unwritten_wait(struct inode *inode)
61 {
62 	wait_queue_head_t *wq = ext4_ioend_wq(inode);
63 
64 	wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_unwritten) == 0));
65 }
66 
67 /*
68  * This tests whether the IO in question is block-aligned or not.
69  * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
70  * are converted to written only after the IO is complete.  Until they are
71  * mapped, these blocks appear as holes, so dio_zero_block() will assume that
72  * it needs to zero out portions of the start and/or end block.  If 2 AIO
73  * threads are at work on the same unwritten block, they must be synchronized
74  * or one thread will zero the other's data, causing corruption.
75  */
76 static int
77 ext4_unaligned_aio(struct inode *inode, struct iov_iter *from, loff_t pos)
78 {
79 	struct super_block *sb = inode->i_sb;
80 	int blockmask = sb->s_blocksize - 1;
81 
82 	if (pos >= i_size_read(inode))
83 		return 0;
84 
85 	if ((pos | iov_iter_alignment(from)) & blockmask)
86 		return 1;
87 
88 	return 0;
89 }
90 
91 static ssize_t
92 ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
93 {
94 	struct file *file = iocb->ki_filp;
95 	struct inode *inode = file_inode(iocb->ki_filp);
96 	struct mutex *aio_mutex = NULL;
97 	struct blk_plug plug;
98 	int o_direct = io_is_direct(file);
99 	int overwrite = 0;
100 	size_t length = iov_iter_count(from);
101 	ssize_t ret;
102 	loff_t pos = iocb->ki_pos;
103 
104 	/*
105 	 * Unaligned direct AIO must be serialized; see comment above
106 	 * In the case of O_APPEND, assume that we must always serialize
107 	 */
108 	if (o_direct &&
109 	    ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
110 	    !is_sync_kiocb(iocb) &&
111 	    (file->f_flags & O_APPEND ||
112 	     ext4_unaligned_aio(inode, from, pos))) {
113 		aio_mutex = ext4_aio_mutex(inode);
114 		mutex_lock(aio_mutex);
115 		ext4_unwritten_wait(inode);
116 	}
117 
118 	mutex_lock(&inode->i_mutex);
119 	if (file->f_flags & O_APPEND)
120 		iocb->ki_pos = pos = i_size_read(inode);
121 
122 	/*
123 	 * If we have encountered a bitmap-format file, the size limit
124 	 * is smaller than s_maxbytes, which is for extent-mapped files.
125 	 */
126 	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
127 		struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
128 
129 		if ((pos > sbi->s_bitmap_maxbytes) ||
130 		    (pos == sbi->s_bitmap_maxbytes && length > 0)) {
131 			mutex_unlock(&inode->i_mutex);
132 			ret = -EFBIG;
133 			goto errout;
134 		}
135 
136 		if (pos + length > sbi->s_bitmap_maxbytes)
137 			iov_iter_truncate(from, sbi->s_bitmap_maxbytes - pos);
138 	}
139 
140 	iocb->private = &overwrite;
141 	if (o_direct) {
142 		blk_start_plug(&plug);
143 
144 
145 		/* check whether we do a DIO overwrite or not */
146 		if (ext4_should_dioread_nolock(inode) && !aio_mutex &&
147 		    !file->f_mapping->nrpages && pos + length <= i_size_read(inode)) {
148 			struct ext4_map_blocks map;
149 			unsigned int blkbits = inode->i_blkbits;
150 			int err, len;
151 
152 			map.m_lblk = pos >> blkbits;
153 			map.m_len = (EXT4_BLOCK_ALIGN(pos + length, blkbits) >> blkbits)
154 				- map.m_lblk;
155 			len = map.m_len;
156 
157 			err = ext4_map_blocks(NULL, inode, &map, 0);
158 			/*
159 			 * 'err==len' means that all of blocks has
160 			 * been preallocated no matter they are
161 			 * initialized or not.  For excluding
162 			 * unwritten extents, we need to check
163 			 * m_flags.  There are two conditions that
164 			 * indicate for initialized extents.  1) If we
165 			 * hit extent cache, EXT4_MAP_MAPPED flag is
166 			 * returned; 2) If we do a real lookup,
167 			 * non-flags are returned.  So we should check
168 			 * these two conditions.
169 			 */
170 			if (err == len && (map.m_flags & EXT4_MAP_MAPPED))
171 				overwrite = 1;
172 		}
173 	}
174 
175 	ret = __generic_file_write_iter(iocb, from);
176 	mutex_unlock(&inode->i_mutex);
177 
178 	if (ret > 0) {
179 		ssize_t err;
180 
181 		err = generic_write_sync(file, iocb->ki_pos - ret, ret);
182 		if (err < 0)
183 			ret = err;
184 	}
185 	if (o_direct)
186 		blk_finish_plug(&plug);
187 
188 errout:
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 };
210 #else
211 #define ext4_dax_vm_ops	ext4_file_vm_ops
212 #endif
213 
214 static const struct vm_operations_struct ext4_file_vm_ops = {
215 	.fault		= filemap_fault,
216 	.map_pages	= filemap_map_pages,
217 	.page_mkwrite   = ext4_page_mkwrite,
218 };
219 
220 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
221 {
222 	file_accessed(file);
223 	if (IS_DAX(file_inode(file))) {
224 		vma->vm_ops = &ext4_dax_vm_ops;
225 		vma->vm_flags |= VM_MIXEDMAP;
226 	} else {
227 		vma->vm_ops = &ext4_file_vm_ops;
228 	}
229 	return 0;
230 }
231 
232 static int ext4_file_open(struct inode * inode, struct file * filp)
233 {
234 	struct super_block *sb = inode->i_sb;
235 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
236 	struct vfsmount *mnt = filp->f_path.mnt;
237 	struct path path;
238 	char buf[64], *cp;
239 
240 	if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
241 		     !(sb->s_flags & MS_RDONLY))) {
242 		sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
243 		/*
244 		 * Sample where the filesystem has been mounted and
245 		 * store it in the superblock for sysadmin convenience
246 		 * when trying to sort through large numbers of block
247 		 * devices or filesystem images.
248 		 */
249 		memset(buf, 0, sizeof(buf));
250 		path.mnt = mnt;
251 		path.dentry = mnt->mnt_root;
252 		cp = d_path(&path, buf, sizeof(buf));
253 		if (!IS_ERR(cp)) {
254 			handle_t *handle;
255 			int err;
256 
257 			handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
258 			if (IS_ERR(handle))
259 				return PTR_ERR(handle);
260 			BUFFER_TRACE(sbi->s_sbh, "get_write_access");
261 			err = ext4_journal_get_write_access(handle, sbi->s_sbh);
262 			if (err) {
263 				ext4_journal_stop(handle);
264 				return err;
265 			}
266 			strlcpy(sbi->s_es->s_last_mounted, cp,
267 				sizeof(sbi->s_es->s_last_mounted));
268 			ext4_handle_dirty_super(handle, sb);
269 			ext4_journal_stop(handle);
270 		}
271 	}
272 	/*
273 	 * Set up the jbd2_inode if we are opening the inode for
274 	 * writing and the journal is present
275 	 */
276 	if (filp->f_mode & FMODE_WRITE) {
277 		int ret = ext4_inode_attach_jinode(inode);
278 		if (ret < 0)
279 			return ret;
280 	}
281 	return dquot_file_open(inode, filp);
282 }
283 
284 /*
285  * Here we use ext4_map_blocks() to get a block mapping for a extent-based
286  * file rather than ext4_ext_walk_space() because we can introduce
287  * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
288  * function.  When extent status tree has been fully implemented, it will
289  * track all extent status for a file and we can directly use it to
290  * retrieve the offset for SEEK_DATA/SEEK_HOLE.
291  */
292 
293 /*
294  * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
295  * lookup page cache to check whether or not there has some data between
296  * [startoff, endoff] because, if this range contains an unwritten extent,
297  * we determine this extent as a data or a hole according to whether the
298  * page cache has data or not.
299  */
300 static int ext4_find_unwritten_pgoff(struct inode *inode,
301 				     int whence,
302 				     struct ext4_map_blocks *map,
303 				     loff_t *offset)
304 {
305 	struct pagevec pvec;
306 	unsigned int blkbits;
307 	pgoff_t index;
308 	pgoff_t end;
309 	loff_t endoff;
310 	loff_t startoff;
311 	loff_t lastoff;
312 	int found = 0;
313 
314 	blkbits = inode->i_sb->s_blocksize_bits;
315 	startoff = *offset;
316 	lastoff = startoff;
317 	endoff = (loff_t)(map->m_lblk + map->m_len) << blkbits;
318 
319 	index = startoff >> PAGE_CACHE_SHIFT;
320 	end = endoff >> PAGE_CACHE_SHIFT;
321 
322 	pagevec_init(&pvec, 0);
323 	do {
324 		int i, num;
325 		unsigned long nr_pages;
326 
327 		num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
328 		nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
329 					  (pgoff_t)num);
330 		if (nr_pages == 0) {
331 			if (whence == SEEK_DATA)
332 				break;
333 
334 			BUG_ON(whence != SEEK_HOLE);
335 			/*
336 			 * If this is the first time to go into the loop and
337 			 * offset is not beyond the end offset, it will be a
338 			 * hole at this offset
339 			 */
340 			if (lastoff == startoff || lastoff < endoff)
341 				found = 1;
342 			break;
343 		}
344 
345 		/*
346 		 * If this is the first time to go into the loop and
347 		 * offset is smaller than the first page offset, it will be a
348 		 * hole at this offset.
349 		 */
350 		if (lastoff == startoff && whence == SEEK_HOLE &&
351 		    lastoff < page_offset(pvec.pages[0])) {
352 			found = 1;
353 			break;
354 		}
355 
356 		for (i = 0; i < nr_pages; i++) {
357 			struct page *page = pvec.pages[i];
358 			struct buffer_head *bh, *head;
359 
360 			/*
361 			 * If the current offset is not beyond the end of given
362 			 * range, it will be a hole.
363 			 */
364 			if (lastoff < endoff && whence == SEEK_HOLE &&
365 			    page->index > end) {
366 				found = 1;
367 				*offset = lastoff;
368 				goto out;
369 			}
370 
371 			lock_page(page);
372 
373 			if (unlikely(page->mapping != inode->i_mapping)) {
374 				unlock_page(page);
375 				continue;
376 			}
377 
378 			if (!page_has_buffers(page)) {
379 				unlock_page(page);
380 				continue;
381 			}
382 
383 			if (page_has_buffers(page)) {
384 				lastoff = page_offset(page);
385 				bh = head = page_buffers(page);
386 				do {
387 					if (buffer_uptodate(bh) ||
388 					    buffer_unwritten(bh)) {
389 						if (whence == SEEK_DATA)
390 							found = 1;
391 					} else {
392 						if (whence == SEEK_HOLE)
393 							found = 1;
394 					}
395 					if (found) {
396 						*offset = max_t(loff_t,
397 							startoff, lastoff);
398 						unlock_page(page);
399 						goto out;
400 					}
401 					lastoff += bh->b_size;
402 					bh = bh->b_this_page;
403 				} while (bh != head);
404 			}
405 
406 			lastoff = page_offset(page) + PAGE_SIZE;
407 			unlock_page(page);
408 		}
409 
410 		/*
411 		 * The no. of pages is less than our desired, that would be a
412 		 * hole in there.
413 		 */
414 		if (nr_pages < num && whence == SEEK_HOLE) {
415 			found = 1;
416 			*offset = lastoff;
417 			break;
418 		}
419 
420 		index = pvec.pages[i - 1]->index + 1;
421 		pagevec_release(&pvec);
422 	} while (index <= end);
423 
424 out:
425 	pagevec_release(&pvec);
426 	return found;
427 }
428 
429 /*
430  * ext4_seek_data() retrieves the offset for SEEK_DATA.
431  */
432 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
433 {
434 	struct inode *inode = file->f_mapping->host;
435 	struct ext4_map_blocks map;
436 	struct extent_status es;
437 	ext4_lblk_t start, last, end;
438 	loff_t dataoff, isize;
439 	int blkbits;
440 	int ret = 0;
441 
442 	mutex_lock(&inode->i_mutex);
443 
444 	isize = i_size_read(inode);
445 	if (offset >= isize) {
446 		mutex_unlock(&inode->i_mutex);
447 		return -ENXIO;
448 	}
449 
450 	blkbits = inode->i_sb->s_blocksize_bits;
451 	start = offset >> blkbits;
452 	last = start;
453 	end = isize >> blkbits;
454 	dataoff = offset;
455 
456 	do {
457 		map.m_lblk = last;
458 		map.m_len = end - last + 1;
459 		ret = ext4_map_blocks(NULL, inode, &map, 0);
460 		if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
461 			if (last != start)
462 				dataoff = (loff_t)last << blkbits;
463 			break;
464 		}
465 
466 		/*
467 		 * If there is a delay extent at this offset,
468 		 * it will be as a data.
469 		 */
470 		ext4_es_find_delayed_extent_range(inode, last, last, &es);
471 		if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
472 			if (last != start)
473 				dataoff = (loff_t)last << blkbits;
474 			break;
475 		}
476 
477 		/*
478 		 * If there is a unwritten extent at this offset,
479 		 * it will be as a data or a hole according to page
480 		 * cache that has data or not.
481 		 */
482 		if (map.m_flags & EXT4_MAP_UNWRITTEN) {
483 			int unwritten;
484 			unwritten = ext4_find_unwritten_pgoff(inode, SEEK_DATA,
485 							      &map, &dataoff);
486 			if (unwritten)
487 				break;
488 		}
489 
490 		last++;
491 		dataoff = (loff_t)last << blkbits;
492 	} while (last <= end);
493 
494 	mutex_unlock(&inode->i_mutex);
495 
496 	if (dataoff > isize)
497 		return -ENXIO;
498 
499 	return vfs_setpos(file, dataoff, maxsize);
500 }
501 
502 /*
503  * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
504  */
505 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
506 {
507 	struct inode *inode = file->f_mapping->host;
508 	struct ext4_map_blocks map;
509 	struct extent_status es;
510 	ext4_lblk_t start, last, end;
511 	loff_t holeoff, isize;
512 	int blkbits;
513 	int ret = 0;
514 
515 	mutex_lock(&inode->i_mutex);
516 
517 	isize = i_size_read(inode);
518 	if (offset >= isize) {
519 		mutex_unlock(&inode->i_mutex);
520 		return -ENXIO;
521 	}
522 
523 	blkbits = inode->i_sb->s_blocksize_bits;
524 	start = offset >> blkbits;
525 	last = start;
526 	end = isize >> blkbits;
527 	holeoff = offset;
528 
529 	do {
530 		map.m_lblk = last;
531 		map.m_len = end - last + 1;
532 		ret = ext4_map_blocks(NULL, inode, &map, 0);
533 		if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
534 			last += ret;
535 			holeoff = (loff_t)last << blkbits;
536 			continue;
537 		}
538 
539 		/*
540 		 * If there is a delay extent at this offset,
541 		 * we will skip this extent.
542 		 */
543 		ext4_es_find_delayed_extent_range(inode, last, last, &es);
544 		if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
545 			last = es.es_lblk + es.es_len;
546 			holeoff = (loff_t)last << blkbits;
547 			continue;
548 		}
549 
550 		/*
551 		 * If there is a unwritten extent at this offset,
552 		 * it will be as a data or a hole according to page
553 		 * cache that has data or not.
554 		 */
555 		if (map.m_flags & EXT4_MAP_UNWRITTEN) {
556 			int unwritten;
557 			unwritten = ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
558 							      &map, &holeoff);
559 			if (!unwritten) {
560 				last += ret;
561 				holeoff = (loff_t)last << blkbits;
562 				continue;
563 			}
564 		}
565 
566 		/* find a hole */
567 		break;
568 	} while (last <= end);
569 
570 	mutex_unlock(&inode->i_mutex);
571 
572 	if (holeoff > isize)
573 		holeoff = isize;
574 
575 	return vfs_setpos(file, holeoff, maxsize);
576 }
577 
578 /*
579  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
580  * by calling generic_file_llseek_size() with the appropriate maxbytes
581  * value for each.
582  */
583 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
584 {
585 	struct inode *inode = file->f_mapping->host;
586 	loff_t maxbytes;
587 
588 	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
589 		maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
590 	else
591 		maxbytes = inode->i_sb->s_maxbytes;
592 
593 	switch (whence) {
594 	case SEEK_SET:
595 	case SEEK_CUR:
596 	case SEEK_END:
597 		return generic_file_llseek_size(file, offset, whence,
598 						maxbytes, i_size_read(inode));
599 	case SEEK_DATA:
600 		return ext4_seek_data(file, offset, maxbytes);
601 	case SEEK_HOLE:
602 		return ext4_seek_hole(file, offset, maxbytes);
603 	}
604 
605 	return -EINVAL;
606 }
607 
608 const struct file_operations ext4_file_operations = {
609 	.llseek		= ext4_llseek,
610 	.read		= new_sync_read,
611 	.write		= new_sync_write,
612 	.read_iter	= generic_file_read_iter,
613 	.write_iter	= ext4_file_write_iter,
614 	.unlocked_ioctl = ext4_ioctl,
615 #ifdef CONFIG_COMPAT
616 	.compat_ioctl	= ext4_compat_ioctl,
617 #endif
618 	.mmap		= ext4_file_mmap,
619 	.open		= ext4_file_open,
620 	.release	= ext4_release_file,
621 	.fsync		= ext4_sync_file,
622 	.splice_read	= generic_file_splice_read,
623 	.splice_write	= iter_file_splice_write,
624 	.fallocate	= ext4_fallocate,
625 };
626 
627 #ifdef CONFIG_FS_DAX
628 const struct file_operations ext4_dax_file_operations = {
629 	.llseek		= ext4_llseek,
630 	.read		= new_sync_read,
631 	.write		= new_sync_write,
632 	.read_iter	= generic_file_read_iter,
633 	.write_iter	= ext4_file_write_iter,
634 	.unlocked_ioctl = ext4_ioctl,
635 #ifdef CONFIG_COMPAT
636 	.compat_ioctl	= ext4_compat_ioctl,
637 #endif
638 	.mmap		= ext4_file_mmap,
639 	.open		= ext4_file_open,
640 	.release	= ext4_release_file,
641 	.fsync		= ext4_sync_file,
642 	/* Splice not yet supported with DAX */
643 	.fallocate	= ext4_fallocate,
644 };
645 #endif
646 
647 const struct inode_operations ext4_file_inode_operations = {
648 	.setattr	= ext4_setattr,
649 	.getattr	= ext4_getattr,
650 	.setxattr	= generic_setxattr,
651 	.getxattr	= generic_getxattr,
652 	.listxattr	= ext4_listxattr,
653 	.removexattr	= generic_removexattr,
654 	.get_acl	= ext4_get_acl,
655 	.set_acl	= ext4_set_acl,
656 	.fiemap		= ext4_fiemap,
657 };
658 
659