xref: /linux/fs/ext4/file.c (revision 9e9f60108423f18a99c9cc93ef7f23490ecc709b)
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 = file->f_flags & O_DIRECT;
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 static const struct vm_operations_struct ext4_file_vm_ops = {
195 	.fault		= filemap_fault,
196 	.map_pages	= filemap_map_pages,
197 	.page_mkwrite   = ext4_page_mkwrite,
198 	.remap_pages	= generic_file_remap_pages,
199 };
200 
201 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
202 {
203 	file_accessed(file);
204 	vma->vm_ops = &ext4_file_vm_ops;
205 	return 0;
206 }
207 
208 static int ext4_file_open(struct inode * inode, struct file * filp)
209 {
210 	struct super_block *sb = inode->i_sb;
211 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
212 	struct vfsmount *mnt = filp->f_path.mnt;
213 	struct path path;
214 	char buf[64], *cp;
215 
216 	if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
217 		     !(sb->s_flags & MS_RDONLY))) {
218 		sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
219 		/*
220 		 * Sample where the filesystem has been mounted and
221 		 * store it in the superblock for sysadmin convenience
222 		 * when trying to sort through large numbers of block
223 		 * devices or filesystem images.
224 		 */
225 		memset(buf, 0, sizeof(buf));
226 		path.mnt = mnt;
227 		path.dentry = mnt->mnt_root;
228 		cp = d_path(&path, buf, sizeof(buf));
229 		if (!IS_ERR(cp)) {
230 			handle_t *handle;
231 			int err;
232 
233 			handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
234 			if (IS_ERR(handle))
235 				return PTR_ERR(handle);
236 			BUFFER_TRACE(sbi->s_sbh, "get_write_access");
237 			err = ext4_journal_get_write_access(handle, sbi->s_sbh);
238 			if (err) {
239 				ext4_journal_stop(handle);
240 				return err;
241 			}
242 			strlcpy(sbi->s_es->s_last_mounted, cp,
243 				sizeof(sbi->s_es->s_last_mounted));
244 			ext4_handle_dirty_super(handle, sb);
245 			ext4_journal_stop(handle);
246 		}
247 	}
248 	/*
249 	 * Set up the jbd2_inode if we are opening the inode for
250 	 * writing and the journal is present
251 	 */
252 	if (filp->f_mode & FMODE_WRITE) {
253 		int ret = ext4_inode_attach_jinode(inode);
254 		if (ret < 0)
255 			return ret;
256 	}
257 	return dquot_file_open(inode, filp);
258 }
259 
260 /*
261  * Here we use ext4_map_blocks() to get a block mapping for a extent-based
262  * file rather than ext4_ext_walk_space() because we can introduce
263  * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
264  * function.  When extent status tree has been fully implemented, it will
265  * track all extent status for a file and we can directly use it to
266  * retrieve the offset for SEEK_DATA/SEEK_HOLE.
267  */
268 
269 /*
270  * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
271  * lookup page cache to check whether or not there has some data between
272  * [startoff, endoff] because, if this range contains an unwritten extent,
273  * we determine this extent as a data or a hole according to whether the
274  * page cache has data or not.
275  */
276 static int ext4_find_unwritten_pgoff(struct inode *inode, int whence,
277 				     loff_t endoff, loff_t *offset)
278 {
279 	struct pagevec pvec;
280 	pgoff_t index;
281 	pgoff_t end;
282 	loff_t startoff;
283 	loff_t lastoff;
284 	int found = 0;
285 
286 	startoff = *offset;
287 	lastoff = startoff;
288 
289 
290 	index = startoff >> PAGE_CACHE_SHIFT;
291 	end = endoff >> PAGE_CACHE_SHIFT;
292 
293 	pagevec_init(&pvec, 0);
294 	do {
295 		int i, num;
296 		unsigned long nr_pages;
297 
298 		num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
299 		nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
300 					  (pgoff_t)num);
301 		if (nr_pages == 0) {
302 			if (whence == SEEK_DATA)
303 				break;
304 
305 			BUG_ON(whence != SEEK_HOLE);
306 			/*
307 			 * If this is the first time to go into the loop and
308 			 * offset is not beyond the end offset, it will be a
309 			 * hole at this offset
310 			 */
311 			if (lastoff == startoff || lastoff < endoff)
312 				found = 1;
313 			break;
314 		}
315 
316 		/*
317 		 * If this is the first time to go into the loop and
318 		 * offset is smaller than the first page offset, it will be a
319 		 * hole at this offset.
320 		 */
321 		if (lastoff == startoff && whence == SEEK_HOLE &&
322 		    lastoff < page_offset(pvec.pages[0])) {
323 			found = 1;
324 			break;
325 		}
326 
327 		for (i = 0; i < nr_pages; i++) {
328 			struct page *page = pvec.pages[i];
329 			struct buffer_head *bh, *head;
330 
331 			/*
332 			 * If the current offset is not beyond the end of given
333 			 * range, it will be a hole.
334 			 */
335 			if (lastoff < endoff && whence == SEEK_HOLE &&
336 			    page->index > end) {
337 				found = 1;
338 				*offset = lastoff;
339 				goto out;
340 			}
341 
342 			lock_page(page);
343 
344 			if (unlikely(page->mapping != inode->i_mapping)) {
345 				unlock_page(page);
346 				continue;
347 			}
348 
349 			if (!page_has_buffers(page)) {
350 				unlock_page(page);
351 				continue;
352 			}
353 
354 			if (page_has_buffers(page)) {
355 				lastoff = page_offset(page);
356 				bh = head = page_buffers(page);
357 				do {
358 					if (buffer_uptodate(bh) ||
359 					    buffer_unwritten(bh)) {
360 						if (whence == SEEK_DATA)
361 							found = 1;
362 					} else {
363 						if (whence == SEEK_HOLE)
364 							found = 1;
365 					}
366 					if (found) {
367 						*offset = max_t(loff_t,
368 							startoff, lastoff);
369 						unlock_page(page);
370 						goto out;
371 					}
372 					lastoff += bh->b_size;
373 					bh = bh->b_this_page;
374 				} while (bh != head);
375 			}
376 
377 			lastoff = page_offset(page) + PAGE_SIZE;
378 			unlock_page(page);
379 		}
380 
381 		/*
382 		 * The no. of pages is less than our desired, that would be a
383 		 * hole in there.
384 		 */
385 		if (nr_pages < num && whence == SEEK_HOLE) {
386 			found = 1;
387 			*offset = lastoff;
388 			break;
389 		}
390 
391 		index = pvec.pages[i - 1]->index + 1;
392 		pagevec_release(&pvec);
393 	} while (index <= end);
394 
395 out:
396 	pagevec_release(&pvec);
397 	return found;
398 }
399 
400 /*
401  * ext4_seek_data() retrieves the offset for SEEK_DATA.
402  */
403 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
404 {
405 	struct inode *inode = file->f_mapping->host;
406 	struct fiemap_extent_info fie;
407 	struct fiemap_extent ext[2];
408 	loff_t next;
409 	int i, ret = 0;
410 
411 	mutex_lock(&inode->i_mutex);
412 	if (offset >= inode->i_size) {
413 		mutex_unlock(&inode->i_mutex);
414 		return -ENXIO;
415 	}
416 	fie.fi_flags = 0;
417 	fie.fi_extents_max = 2;
418 	fie.fi_extents_start = (struct fiemap_extent __user *) &ext;
419 	while (1) {
420 		mm_segment_t old_fs = get_fs();
421 
422 		fie.fi_extents_mapped = 0;
423 		memset(ext, 0, sizeof(*ext) * fie.fi_extents_max);
424 
425 		set_fs(get_ds());
426 		ret = ext4_fiemap(inode, &fie, offset, maxsize - offset);
427 		set_fs(old_fs);
428 		if (ret)
429 			break;
430 
431 		/* No extents found, EOF */
432 		if (!fie.fi_extents_mapped) {
433 			ret = -ENXIO;
434 			break;
435 		}
436 		for (i = 0; i < fie.fi_extents_mapped; i++) {
437 			next = (loff_t)(ext[i].fe_length + ext[i].fe_logical);
438 
439 			if (offset < (loff_t)ext[i].fe_logical)
440 				offset = (loff_t)ext[i].fe_logical;
441 			/*
442 			 * If extent is not unwritten, then it contains valid
443 			 * data, mapped or delayed.
444 			 */
445 			if (!(ext[i].fe_flags & FIEMAP_EXTENT_UNWRITTEN))
446 				goto out;
447 
448 			/*
449 			 * If there is a unwritten extent at this offset,
450 			 * it will be as a data or a hole according to page
451 			 * cache that has data or not.
452 			 */
453 			if (ext4_find_unwritten_pgoff(inode, SEEK_DATA,
454 						      next, &offset))
455 				goto out;
456 
457 			if (ext[i].fe_flags & FIEMAP_EXTENT_LAST) {
458 				ret = -ENXIO;
459 				goto out;
460 			}
461 			offset = next;
462 		}
463 	}
464 	if (offset > inode->i_size)
465 		offset = inode->i_size;
466 out:
467 	mutex_unlock(&inode->i_mutex);
468 	if (ret)
469 		return ret;
470 
471 	return vfs_setpos(file, offset, maxsize);
472 }
473 
474 /*
475  * ext4_seek_hole() retrieves the offset for SEEK_HOLE
476  */
477 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
478 {
479 	struct inode *inode = file->f_mapping->host;
480 	struct fiemap_extent_info fie;
481 	struct fiemap_extent ext[2];
482 	loff_t next;
483 	int i, ret = 0;
484 
485 	mutex_lock(&inode->i_mutex);
486 	if (offset >= inode->i_size) {
487 		mutex_unlock(&inode->i_mutex);
488 		return -ENXIO;
489 	}
490 
491 	fie.fi_flags = 0;
492 	fie.fi_extents_max = 2;
493 	fie.fi_extents_start = (struct fiemap_extent __user *)&ext;
494 	while (1) {
495 		mm_segment_t old_fs = get_fs();
496 
497 		fie.fi_extents_mapped = 0;
498 		memset(ext, 0, sizeof(*ext));
499 
500 		set_fs(get_ds());
501 		ret = ext4_fiemap(inode, &fie, offset, maxsize - offset);
502 		set_fs(old_fs);
503 		if (ret)
504 			break;
505 
506 		/* No extents found */
507 		if (!fie.fi_extents_mapped)
508 			break;
509 
510 		for (i = 0; i < fie.fi_extents_mapped; i++) {
511 			next = (loff_t)(ext[i].fe_logical + ext[i].fe_length);
512 			/*
513 			 * If extent is not unwritten, then it contains valid
514 			 * data, mapped or delayed.
515 			 */
516 			if (!(ext[i].fe_flags & FIEMAP_EXTENT_UNWRITTEN)) {
517 				if (offset < (loff_t)ext[i].fe_logical)
518 					goto out;
519 				offset = next;
520 				continue;
521 			}
522 			/*
523 			 * If there is a unwritten extent at this offset,
524 			 * it will be as a data or a hole according to page
525 			 * cache that has data or not.
526 			 */
527 			if (ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
528 						      next, &offset))
529 				goto out;
530 
531 			offset = next;
532 			if (ext[i].fe_flags & FIEMAP_EXTENT_LAST)
533 				goto out;
534 		}
535 	}
536 	if (offset > inode->i_size)
537 		offset = inode->i_size;
538 out:
539 	mutex_unlock(&inode->i_mutex);
540 	if (ret)
541 		return ret;
542 
543 	return vfs_setpos(file, offset, maxsize);
544 }
545 
546 /*
547  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
548  * by calling generic_file_llseek_size() with the appropriate maxbytes
549  * value for each.
550  */
551 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
552 {
553 	struct inode *inode = file->f_mapping->host;
554 	loff_t maxbytes;
555 
556 	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
557 		maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
558 	else
559 		maxbytes = inode->i_sb->s_maxbytes;
560 
561 	switch (whence) {
562 	case SEEK_SET:
563 	case SEEK_CUR:
564 	case SEEK_END:
565 		return generic_file_llseek_size(file, offset, whence,
566 						maxbytes, i_size_read(inode));
567 	case SEEK_DATA:
568 		return ext4_seek_data(file, offset, maxbytes);
569 	case SEEK_HOLE:
570 		return ext4_seek_hole(file, offset, maxbytes);
571 	}
572 
573 	return -EINVAL;
574 }
575 
576 const struct file_operations ext4_file_operations = {
577 	.llseek		= ext4_llseek,
578 	.read		= new_sync_read,
579 	.write		= new_sync_write,
580 	.read_iter	= generic_file_read_iter,
581 	.write_iter	= ext4_file_write_iter,
582 	.unlocked_ioctl = ext4_ioctl,
583 #ifdef CONFIG_COMPAT
584 	.compat_ioctl	= ext4_compat_ioctl,
585 #endif
586 	.mmap		= ext4_file_mmap,
587 	.open		= ext4_file_open,
588 	.release	= ext4_release_file,
589 	.fsync		= ext4_sync_file,
590 	.splice_read	= generic_file_splice_read,
591 	.splice_write	= iter_file_splice_write,
592 	.fallocate	= ext4_fallocate,
593 };
594 
595 const struct inode_operations ext4_file_inode_operations = {
596 	.setattr	= ext4_setattr,
597 	.getattr	= ext4_getattr,
598 	.setxattr	= generic_setxattr,
599 	.getxattr	= generic_getxattr,
600 	.listxattr	= ext4_listxattr,
601 	.removexattr	= generic_removexattr,
602 	.get_acl	= ext4_get_acl,
603 	.set_acl	= ext4_set_acl,
604 	.fiemap		= ext4_fiemap,
605 };
606 
607