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/dax.h>
26 #include <linux/quotaops.h>
27 #include <linux/pagevec.h>
28 #include <linux/uio.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  */
ext4_release_file(struct inode * inode,struct file * filp)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 
ext4_unwritten_wait(struct inode * inode)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
ext4_unaligned_aio(struct inode * inode,struct iov_iter * from,loff_t pos)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
ext4_file_write_iter(struct kiocb * iocb,struct iov_iter * from)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 = iocb->ki_flags & IOCB_DIRECT;
99 	int overwrite = 0;
100 	ssize_t ret;
101 
102 	/*
103 	 * Unaligned direct AIO must be serialized; see comment above
104 	 * In the case of O_APPEND, assume that we must always serialize
105 	 */
106 	if (o_direct &&
107 	    ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
108 	    !is_sync_kiocb(iocb) &&
109 	    (iocb->ki_flags & IOCB_APPEND ||
110 	     ext4_unaligned_aio(inode, from, iocb->ki_pos))) {
111 		aio_mutex = ext4_aio_mutex(inode);
112 		mutex_lock(aio_mutex);
113 		ext4_unwritten_wait(inode);
114 	}
115 
116 	mutex_lock(&inode->i_mutex);
117 	ret = generic_write_checks(iocb, from);
118 	if (ret <= 0)
119 		goto out;
120 
121 	/*
122 	 * If we have encountered a bitmap-format file, the size limit
123 	 * is smaller than s_maxbytes, which is for extent-mapped files.
124 	 */
125 	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
126 		struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
127 
128 		if (iocb->ki_pos >= sbi->s_bitmap_maxbytes) {
129 			ret = -EFBIG;
130 			goto out;
131 		}
132 		iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos);
133 	}
134 
135 	iocb->private = &overwrite;
136 	if (o_direct) {
137 		size_t length = iov_iter_count(from);
138 		loff_t pos = iocb->ki_pos;
139 		blk_start_plug(&plug);
140 
141 		/* check whether we do a DIO overwrite or not */
142 		if (ext4_should_dioread_nolock(inode) && !aio_mutex &&
143 		    !file->f_mapping->nrpages && pos + length <= i_size_read(inode)) {
144 			struct ext4_map_blocks map;
145 			unsigned int blkbits = inode->i_blkbits;
146 			int err, len;
147 
148 			map.m_lblk = pos >> blkbits;
149 			map.m_len = (EXT4_BLOCK_ALIGN(pos + length, blkbits) >> blkbits)
150 				- map.m_lblk;
151 			len = map.m_len;
152 
153 			err = ext4_map_blocks(NULL, inode, &map, 0);
154 			/*
155 			 * 'err==len' means that all of blocks has
156 			 * been preallocated no matter they are
157 			 * initialized or not.  For excluding
158 			 * unwritten extents, we need to check
159 			 * m_flags.  There are two conditions that
160 			 * indicate for initialized extents.  1) If we
161 			 * hit extent cache, EXT4_MAP_MAPPED flag is
162 			 * returned; 2) If we do a real lookup,
163 			 * non-flags are returned.  So we should check
164 			 * these two conditions.
165 			 */
166 			if (err == len && (map.m_flags & EXT4_MAP_MAPPED))
167 				overwrite = 1;
168 		}
169 	}
170 
171 	ret = __generic_file_write_iter(iocb, from);
172 	mutex_unlock(&inode->i_mutex);
173 
174 	if (ret > 0) {
175 		ssize_t err;
176 
177 		err = generic_write_sync(file, iocb->ki_pos - ret, ret);
178 		if (err < 0)
179 			ret = err;
180 	}
181 	if (o_direct)
182 		blk_finish_plug(&plug);
183 
184 	if (aio_mutex)
185 		mutex_unlock(aio_mutex);
186 	return ret;
187 
188 out:
189 	mutex_unlock(&inode->i_mutex);
190 	if (aio_mutex)
191 		mutex_unlock(aio_mutex);
192 	return ret;
193 }
194 
195 #ifdef CONFIG_FS_DAX
ext4_end_io_unwritten(struct buffer_head * bh,int uptodate)196 static void ext4_end_io_unwritten(struct buffer_head *bh, int uptodate)
197 {
198 	struct inode *inode = bh->b_assoc_map->host;
199 	/* XXX: breaks on 32-bit > 16TB. Is that even supported? */
200 	loff_t offset = (loff_t)(uintptr_t)bh->b_private << inode->i_blkbits;
201 	int err;
202 	if (!uptodate)
203 		return;
204 	WARN_ON(!buffer_unwritten(bh));
205 	err = ext4_convert_unwritten_extents(NULL, inode, offset, bh->b_size);
206 }
207 
ext4_dax_fault(struct vm_area_struct * vma,struct vm_fault * vmf)208 static int ext4_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
209 {
210 	int result;
211 	handle_t *handle = NULL;
212 	struct inode *inode = file_inode(vma->vm_file);
213 	struct super_block *sb = inode->i_sb;
214 	bool write = vmf->flags & FAULT_FLAG_WRITE;
215 
216 	if (write) {
217 		sb_start_pagefault(sb);
218 		file_update_time(vma->vm_file);
219 		down_read(&EXT4_I(inode)->i_mmap_sem);
220 		handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
221 						EXT4_DATA_TRANS_BLOCKS(sb));
222 	} else
223 		down_read(&EXT4_I(inode)->i_mmap_sem);
224 
225 	if (IS_ERR(handle))
226 		result = VM_FAULT_SIGBUS;
227 	else
228 		result = __dax_fault(vma, vmf, ext4_get_block_dax,
229 						ext4_end_io_unwritten);
230 
231 	if (write) {
232 		if (!IS_ERR(handle))
233 			ext4_journal_stop(handle);
234 		up_read(&EXT4_I(inode)->i_mmap_sem);
235 		sb_end_pagefault(sb);
236 	} else
237 		up_read(&EXT4_I(inode)->i_mmap_sem);
238 
239 	return result;
240 }
241 
ext4_dax_pmd_fault(struct vm_area_struct * vma,unsigned long addr,pmd_t * pmd,unsigned int flags)242 static int ext4_dax_pmd_fault(struct vm_area_struct *vma, unsigned long addr,
243 						pmd_t *pmd, unsigned int flags)
244 {
245 	int result;
246 	handle_t *handle = NULL;
247 	struct inode *inode = file_inode(vma->vm_file);
248 	struct super_block *sb = inode->i_sb;
249 	bool write = flags & FAULT_FLAG_WRITE;
250 
251 	if (write) {
252 		sb_start_pagefault(sb);
253 		file_update_time(vma->vm_file);
254 		down_read(&EXT4_I(inode)->i_mmap_sem);
255 		handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
256 				ext4_chunk_trans_blocks(inode,
257 							PMD_SIZE / PAGE_SIZE));
258 	} else
259 		down_read(&EXT4_I(inode)->i_mmap_sem);
260 
261 	if (IS_ERR(handle))
262 		result = VM_FAULT_SIGBUS;
263 	else
264 		result = __dax_pmd_fault(vma, addr, pmd, flags,
265 				ext4_get_block_dax, ext4_end_io_unwritten);
266 
267 	if (write) {
268 		if (!IS_ERR(handle))
269 			ext4_journal_stop(handle);
270 		up_read(&EXT4_I(inode)->i_mmap_sem);
271 		sb_end_pagefault(sb);
272 	} else
273 		up_read(&EXT4_I(inode)->i_mmap_sem);
274 
275 	return result;
276 }
277 
ext4_dax_mkwrite(struct vm_area_struct * vma,struct vm_fault * vmf)278 static int ext4_dax_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
279 {
280 	int err;
281 	struct inode *inode = file_inode(vma->vm_file);
282 
283 	sb_start_pagefault(inode->i_sb);
284 	file_update_time(vma->vm_file);
285 	down_read(&EXT4_I(inode)->i_mmap_sem);
286 	err = __dax_mkwrite(vma, vmf, ext4_get_block_dax,
287 			    ext4_end_io_unwritten);
288 	up_read(&EXT4_I(inode)->i_mmap_sem);
289 	sb_end_pagefault(inode->i_sb);
290 
291 	return err;
292 }
293 
294 /*
295  * Handle write fault for VM_MIXEDMAP mappings. Similarly to ext4_dax_mkwrite()
296  * handler we check for races agaist truncate. Note that since we cycle through
297  * i_mmap_sem, we are sure that also any hole punching that began before we
298  * were called is finished by now and so if it included part of the file we
299  * are working on, our pte will get unmapped and the check for pte_same() in
300  * wp_pfn_shared() fails. Thus fault gets retried and things work out as
301  * desired.
302  */
ext4_dax_pfn_mkwrite(struct vm_area_struct * vma,struct vm_fault * vmf)303 static int ext4_dax_pfn_mkwrite(struct vm_area_struct *vma,
304 				struct vm_fault *vmf)
305 {
306 	struct inode *inode = file_inode(vma->vm_file);
307 	struct super_block *sb = inode->i_sb;
308 	int ret = VM_FAULT_NOPAGE;
309 	loff_t size;
310 
311 	sb_start_pagefault(sb);
312 	file_update_time(vma->vm_file);
313 	down_read(&EXT4_I(inode)->i_mmap_sem);
314 	size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
315 	if (vmf->pgoff >= size)
316 		ret = VM_FAULT_SIGBUS;
317 	up_read(&EXT4_I(inode)->i_mmap_sem);
318 	sb_end_pagefault(sb);
319 
320 	return ret;
321 }
322 
323 static const struct vm_operations_struct ext4_dax_vm_ops = {
324 	.fault		= ext4_dax_fault,
325 	.pmd_fault	= ext4_dax_pmd_fault,
326 	.page_mkwrite	= ext4_dax_mkwrite,
327 	.pfn_mkwrite	= ext4_dax_pfn_mkwrite,
328 };
329 #else
330 #define ext4_dax_vm_ops	ext4_file_vm_ops
331 #endif
332 
333 static const struct vm_operations_struct ext4_file_vm_ops = {
334 	.fault		= ext4_filemap_fault,
335 	.map_pages	= filemap_map_pages,
336 	.page_mkwrite   = ext4_page_mkwrite,
337 };
338 
ext4_file_mmap(struct file * file,struct vm_area_struct * vma)339 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
340 {
341 	struct inode *inode = file->f_mapping->host;
342 
343 	if (ext4_encrypted_inode(inode)) {
344 		int err = ext4_get_encryption_info(inode);
345 		if (err)
346 			return 0;
347 		if (ext4_encryption_info(inode) == NULL)
348 			return -ENOKEY;
349 	}
350 	file_accessed(file);
351 	if (IS_DAX(file_inode(file))) {
352 		vma->vm_ops = &ext4_dax_vm_ops;
353 		vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
354 	} else {
355 		vma->vm_ops = &ext4_file_vm_ops;
356 	}
357 	return 0;
358 }
359 
ext4_file_open(struct inode * inode,struct file * filp)360 static int ext4_file_open(struct inode * inode, struct file * filp)
361 {
362 	struct super_block *sb = inode->i_sb;
363 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
364 	struct vfsmount *mnt = filp->f_path.mnt;
365 	struct path path;
366 	char buf[64], *cp;
367 	int ret;
368 
369 	if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
370 		     !(sb->s_flags & MS_RDONLY))) {
371 		sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
372 		/*
373 		 * Sample where the filesystem has been mounted and
374 		 * store it in the superblock for sysadmin convenience
375 		 * when trying to sort through large numbers of block
376 		 * devices or filesystem images.
377 		 */
378 		memset(buf, 0, sizeof(buf));
379 		path.mnt = mnt;
380 		path.dentry = mnt->mnt_root;
381 		cp = d_path(&path, buf, sizeof(buf));
382 		if (!IS_ERR(cp)) {
383 			handle_t *handle;
384 			int err;
385 
386 			handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
387 			if (IS_ERR(handle))
388 				return PTR_ERR(handle);
389 			BUFFER_TRACE(sbi->s_sbh, "get_write_access");
390 			err = ext4_journal_get_write_access(handle, sbi->s_sbh);
391 			if (err) {
392 				ext4_journal_stop(handle);
393 				return err;
394 			}
395 			strlcpy(sbi->s_es->s_last_mounted, cp,
396 				sizeof(sbi->s_es->s_last_mounted));
397 			ext4_handle_dirty_super(handle, sb);
398 			ext4_journal_stop(handle);
399 		}
400 	}
401 	if (ext4_encrypted_inode(inode)) {
402 		ret = ext4_get_encryption_info(inode);
403 		if (ret)
404 			return -EACCES;
405 		if (ext4_encryption_info(inode) == NULL)
406 			return -ENOKEY;
407 	}
408 	/*
409 	 * Set up the jbd2_inode if we are opening the inode for
410 	 * writing and the journal is present
411 	 */
412 	if (filp->f_mode & FMODE_WRITE) {
413 		ret = ext4_inode_attach_jinode(inode);
414 		if (ret < 0)
415 			return ret;
416 	}
417 	return dquot_file_open(inode, filp);
418 }
419 
420 /*
421  * Here we use ext4_map_blocks() to get a block mapping for a extent-based
422  * file rather than ext4_ext_walk_space() because we can introduce
423  * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
424  * function.  When extent status tree has been fully implemented, it will
425  * track all extent status for a file and we can directly use it to
426  * retrieve the offset for SEEK_DATA/SEEK_HOLE.
427  */
428 
429 /*
430  * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
431  * lookup page cache to check whether or not there has some data between
432  * [startoff, endoff] because, if this range contains an unwritten extent,
433  * we determine this extent as a data or a hole according to whether the
434  * page cache has data or not.
435  */
ext4_find_unwritten_pgoff(struct inode * inode,int whence,struct ext4_map_blocks * map,loff_t * offset)436 static int ext4_find_unwritten_pgoff(struct inode *inode,
437 				     int whence,
438 				     struct ext4_map_blocks *map,
439 				     loff_t *offset)
440 {
441 	struct pagevec pvec;
442 	unsigned int blkbits;
443 	pgoff_t index;
444 	pgoff_t end;
445 	loff_t endoff;
446 	loff_t startoff;
447 	loff_t lastoff;
448 	int found = 0;
449 
450 	blkbits = inode->i_sb->s_blocksize_bits;
451 	startoff = *offset;
452 	lastoff = startoff;
453 	endoff = (loff_t)(map->m_lblk + map->m_len) << blkbits;
454 
455 	index = startoff >> PAGE_CACHE_SHIFT;
456 	end = endoff >> PAGE_CACHE_SHIFT;
457 
458 	pagevec_init(&pvec, 0);
459 	do {
460 		int i, num;
461 		unsigned long nr_pages;
462 
463 		num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
464 		nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
465 					  (pgoff_t)num);
466 		if (nr_pages == 0) {
467 			if (whence == SEEK_DATA)
468 				break;
469 
470 			BUG_ON(whence != SEEK_HOLE);
471 			/*
472 			 * If this is the first time to go into the loop and
473 			 * offset is not beyond the end offset, it will be a
474 			 * hole at this offset
475 			 */
476 			if (lastoff == startoff || lastoff < endoff)
477 				found = 1;
478 			break;
479 		}
480 
481 		/*
482 		 * If this is the first time to go into the loop and
483 		 * offset is smaller than the first page offset, it will be a
484 		 * hole at this offset.
485 		 */
486 		if (lastoff == startoff && whence == SEEK_HOLE &&
487 		    lastoff < page_offset(pvec.pages[0])) {
488 			found = 1;
489 			break;
490 		}
491 
492 		for (i = 0; i < nr_pages; i++) {
493 			struct page *page = pvec.pages[i];
494 			struct buffer_head *bh, *head;
495 
496 			/*
497 			 * If the current offset is not beyond the end of given
498 			 * range, it will be a hole.
499 			 */
500 			if (lastoff < endoff && whence == SEEK_HOLE &&
501 			    page->index > end) {
502 				found = 1;
503 				*offset = lastoff;
504 				goto out;
505 			}
506 
507 			lock_page(page);
508 
509 			if (unlikely(page->mapping != inode->i_mapping)) {
510 				unlock_page(page);
511 				continue;
512 			}
513 
514 			if (!page_has_buffers(page)) {
515 				unlock_page(page);
516 				continue;
517 			}
518 
519 			if (page_has_buffers(page)) {
520 				lastoff = page_offset(page);
521 				bh = head = page_buffers(page);
522 				do {
523 					if (buffer_uptodate(bh) ||
524 					    buffer_unwritten(bh)) {
525 						if (whence == SEEK_DATA)
526 							found = 1;
527 					} else {
528 						if (whence == SEEK_HOLE)
529 							found = 1;
530 					}
531 					if (found) {
532 						*offset = max_t(loff_t,
533 							startoff, lastoff);
534 						unlock_page(page);
535 						goto out;
536 					}
537 					lastoff += bh->b_size;
538 					bh = bh->b_this_page;
539 				} while (bh != head);
540 			}
541 
542 			lastoff = page_offset(page) + PAGE_SIZE;
543 			unlock_page(page);
544 		}
545 
546 		/*
547 		 * The no. of pages is less than our desired, that would be a
548 		 * hole in there.
549 		 */
550 		if (nr_pages < num && whence == SEEK_HOLE) {
551 			found = 1;
552 			*offset = lastoff;
553 			break;
554 		}
555 
556 		index = pvec.pages[i - 1]->index + 1;
557 		pagevec_release(&pvec);
558 	} while (index <= end);
559 
560 out:
561 	pagevec_release(&pvec);
562 	return found;
563 }
564 
565 /*
566  * ext4_seek_data() retrieves the offset for SEEK_DATA.
567  */
ext4_seek_data(struct file * file,loff_t offset,loff_t maxsize)568 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
569 {
570 	struct inode *inode = file->f_mapping->host;
571 	struct ext4_map_blocks map;
572 	struct extent_status es;
573 	ext4_lblk_t start, last, end;
574 	loff_t dataoff, isize;
575 	int blkbits;
576 	int ret = 0;
577 
578 	mutex_lock(&inode->i_mutex);
579 
580 	isize = i_size_read(inode);
581 	if (offset >= isize) {
582 		mutex_unlock(&inode->i_mutex);
583 		return -ENXIO;
584 	}
585 
586 	blkbits = inode->i_sb->s_blocksize_bits;
587 	start = offset >> blkbits;
588 	last = start;
589 	end = isize >> blkbits;
590 	dataoff = offset;
591 
592 	do {
593 		map.m_lblk = last;
594 		map.m_len = end - last + 1;
595 		ret = ext4_map_blocks(NULL, inode, &map, 0);
596 		if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
597 			if (last != start)
598 				dataoff = (loff_t)last << blkbits;
599 			break;
600 		}
601 
602 		/*
603 		 * If there is a delay extent at this offset,
604 		 * it will be as a data.
605 		 */
606 		ext4_es_find_delayed_extent_range(inode, last, last, &es);
607 		if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
608 			if (last != start)
609 				dataoff = (loff_t)last << blkbits;
610 			break;
611 		}
612 
613 		/*
614 		 * If there is a unwritten extent at this offset,
615 		 * it will be as a data or a hole according to page
616 		 * cache that has data or not.
617 		 */
618 		if (map.m_flags & EXT4_MAP_UNWRITTEN) {
619 			int unwritten;
620 			unwritten = ext4_find_unwritten_pgoff(inode, SEEK_DATA,
621 							      &map, &dataoff);
622 			if (unwritten)
623 				break;
624 		}
625 
626 		last++;
627 		dataoff = (loff_t)last << blkbits;
628 	} while (last <= end);
629 
630 	mutex_unlock(&inode->i_mutex);
631 
632 	if (dataoff > isize)
633 		return -ENXIO;
634 
635 	return vfs_setpos(file, dataoff, maxsize);
636 }
637 
638 /*
639  * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
640  */
ext4_seek_hole(struct file * file,loff_t offset,loff_t maxsize)641 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
642 {
643 	struct inode *inode = file->f_mapping->host;
644 	struct ext4_map_blocks map;
645 	struct extent_status es;
646 	ext4_lblk_t start, last, end;
647 	loff_t holeoff, isize;
648 	int blkbits;
649 	int ret = 0;
650 
651 	mutex_lock(&inode->i_mutex);
652 
653 	isize = i_size_read(inode);
654 	if (offset >= isize) {
655 		mutex_unlock(&inode->i_mutex);
656 		return -ENXIO;
657 	}
658 
659 	blkbits = inode->i_sb->s_blocksize_bits;
660 	start = offset >> blkbits;
661 	last = start;
662 	end = isize >> blkbits;
663 	holeoff = offset;
664 
665 	do {
666 		map.m_lblk = last;
667 		map.m_len = end - last + 1;
668 		ret = ext4_map_blocks(NULL, inode, &map, 0);
669 		if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
670 			last += ret;
671 			holeoff = (loff_t)last << blkbits;
672 			continue;
673 		}
674 
675 		/*
676 		 * If there is a delay extent at this offset,
677 		 * we will skip this extent.
678 		 */
679 		ext4_es_find_delayed_extent_range(inode, last, last, &es);
680 		if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
681 			last = es.es_lblk + es.es_len;
682 			holeoff = (loff_t)last << blkbits;
683 			continue;
684 		}
685 
686 		/*
687 		 * If there is a unwritten extent at this offset,
688 		 * it will be as a data or a hole according to page
689 		 * cache that has data or not.
690 		 */
691 		if (map.m_flags & EXT4_MAP_UNWRITTEN) {
692 			int unwritten;
693 			unwritten = ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
694 							      &map, &holeoff);
695 			if (!unwritten) {
696 				last += ret;
697 				holeoff = (loff_t)last << blkbits;
698 				continue;
699 			}
700 		}
701 
702 		/* find a hole */
703 		break;
704 	} while (last <= end);
705 
706 	mutex_unlock(&inode->i_mutex);
707 
708 	if (holeoff > isize)
709 		holeoff = isize;
710 
711 	return vfs_setpos(file, holeoff, maxsize);
712 }
713 
714 /*
715  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
716  * by calling generic_file_llseek_size() with the appropriate maxbytes
717  * value for each.
718  */
ext4_llseek(struct file * file,loff_t offset,int whence)719 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
720 {
721 	struct inode *inode = file->f_mapping->host;
722 	loff_t maxbytes;
723 
724 	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
725 		maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
726 	else
727 		maxbytes = inode->i_sb->s_maxbytes;
728 
729 	switch (whence) {
730 	case SEEK_SET:
731 	case SEEK_CUR:
732 	case SEEK_END:
733 		return generic_file_llseek_size(file, offset, whence,
734 						maxbytes, i_size_read(inode));
735 	case SEEK_DATA:
736 		return ext4_seek_data(file, offset, maxbytes);
737 	case SEEK_HOLE:
738 		return ext4_seek_hole(file, offset, maxbytes);
739 	}
740 
741 	return -EINVAL;
742 }
743 
744 const struct file_operations ext4_file_operations = {
745 	.llseek		= ext4_llseek,
746 	.read_iter	= generic_file_read_iter,
747 	.write_iter	= ext4_file_write_iter,
748 	.unlocked_ioctl = ext4_ioctl,
749 #ifdef CONFIG_COMPAT
750 	.compat_ioctl	= ext4_compat_ioctl,
751 #endif
752 	.mmap		= ext4_file_mmap,
753 	.open		= ext4_file_open,
754 	.release	= ext4_release_file,
755 	.fsync		= ext4_sync_file,
756 	.splice_read	= generic_file_splice_read,
757 	.splice_write	= iter_file_splice_write,
758 	.fallocate	= ext4_fallocate,
759 };
760 
761 const struct inode_operations ext4_file_inode_operations = {
762 	.setattr	= ext4_setattr,
763 	.getattr	= ext4_getattr,
764 	.setxattr	= generic_setxattr,
765 	.getxattr	= generic_getxattr,
766 	.listxattr	= ext4_listxattr,
767 	.removexattr	= generic_removexattr,
768 	.get_acl	= ext4_get_acl,
769 	.set_acl	= ext4_set_acl,
770 	.fiemap		= ext4_fiemap,
771 };
772 
773