• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 >= ALIGN(i_size_read(inode), sb->s_blocksize))
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 - 1) + 1;
464 		nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
465 					  (pgoff_t)num);
466 		if (nr_pages == 0)
467 			break;
468 
469 		for (i = 0; i < nr_pages; i++) {
470 			struct page *page = pvec.pages[i];
471 			struct buffer_head *bh, *head;
472 
473 			/*
474 			 * If current offset is smaller than the page offset,
475 			 * there is a hole at this offset.
476 			 */
477 			if (whence == SEEK_HOLE && lastoff < endoff &&
478 			    lastoff < page_offset(pvec.pages[i])) {
479 				found = 1;
480 				*offset = lastoff;
481 				goto out;
482 			}
483 
484 			if (page->index > end)
485 				goto out;
486 
487 			lock_page(page);
488 
489 			if (unlikely(page->mapping != inode->i_mapping)) {
490 				unlock_page(page);
491 				continue;
492 			}
493 
494 			if (!page_has_buffers(page)) {
495 				unlock_page(page);
496 				continue;
497 			}
498 
499 			if (page_has_buffers(page)) {
500 				lastoff = page_offset(page);
501 				bh = head = page_buffers(page);
502 				do {
503 					if (lastoff + bh->b_size <= startoff)
504 						goto next;
505 					if (buffer_uptodate(bh) ||
506 					    buffer_unwritten(bh)) {
507 						if (whence == SEEK_DATA)
508 							found = 1;
509 					} else {
510 						if (whence == SEEK_HOLE)
511 							found = 1;
512 					}
513 					if (found) {
514 						*offset = max_t(loff_t,
515 							startoff, lastoff);
516 						unlock_page(page);
517 						goto out;
518 					}
519 next:
520 					lastoff += bh->b_size;
521 					bh = bh->b_this_page;
522 				} while (bh != head);
523 			}
524 
525 			lastoff = page_offset(page) + PAGE_SIZE;
526 			unlock_page(page);
527 		}
528 
529 		/* The no. of pages is less than our desired, we are done. */
530 		if (nr_pages < num)
531 			break;
532 
533 		index = pvec.pages[i - 1]->index + 1;
534 		pagevec_release(&pvec);
535 	} while (index <= end);
536 
537 	if (whence == SEEK_HOLE && lastoff < endoff) {
538 		found = 1;
539 		*offset = lastoff;
540 	}
541 out:
542 	pagevec_release(&pvec);
543 	return found;
544 }
545 
546 /*
547  * ext4_seek_data() retrieves the offset for SEEK_DATA.
548  */
ext4_seek_data(struct file * file,loff_t offset,loff_t maxsize)549 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
550 {
551 	struct inode *inode = file->f_mapping->host;
552 	struct ext4_map_blocks map;
553 	struct extent_status es;
554 	ext4_lblk_t start, last, end;
555 	loff_t dataoff, isize;
556 	int blkbits;
557 	int ret = 0;
558 
559 	mutex_lock(&inode->i_mutex);
560 
561 	isize = i_size_read(inode);
562 	if (offset < 0 || offset >= isize) {
563 		mutex_unlock(&inode->i_mutex);
564 		return -ENXIO;
565 	}
566 
567 	blkbits = inode->i_sb->s_blocksize_bits;
568 	start = offset >> blkbits;
569 	last = start;
570 	end = isize >> blkbits;
571 	dataoff = offset;
572 
573 	do {
574 		map.m_lblk = last;
575 		map.m_len = end - last + 1;
576 		ret = ext4_map_blocks(NULL, inode, &map, 0);
577 		if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
578 			if (last != start)
579 				dataoff = (loff_t)last << blkbits;
580 			break;
581 		}
582 
583 		/*
584 		 * If there is a delay extent at this offset,
585 		 * it will be as a data.
586 		 */
587 		ext4_es_find_delayed_extent_range(inode, last, last, &es);
588 		if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
589 			if (last != start)
590 				dataoff = (loff_t)last << blkbits;
591 			break;
592 		}
593 
594 		/*
595 		 * If there is a unwritten extent at this offset,
596 		 * it will be as a data or a hole according to page
597 		 * cache that has data or not.
598 		 */
599 		if (map.m_flags & EXT4_MAP_UNWRITTEN) {
600 			int unwritten;
601 			unwritten = ext4_find_unwritten_pgoff(inode, SEEK_DATA,
602 							      &map, &dataoff);
603 			if (unwritten)
604 				break;
605 		}
606 
607 		last++;
608 		dataoff = (loff_t)last << blkbits;
609 	} while (last <= end);
610 
611 	mutex_unlock(&inode->i_mutex);
612 
613 	if (dataoff > isize)
614 		return -ENXIO;
615 
616 	return vfs_setpos(file, dataoff, maxsize);
617 }
618 
619 /*
620  * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
621  */
ext4_seek_hole(struct file * file,loff_t offset,loff_t maxsize)622 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
623 {
624 	struct inode *inode = file->f_mapping->host;
625 	struct ext4_map_blocks map;
626 	struct extent_status es;
627 	ext4_lblk_t start, last, end;
628 	loff_t holeoff, isize;
629 	int blkbits;
630 	int ret = 0;
631 
632 	mutex_lock(&inode->i_mutex);
633 
634 	isize = i_size_read(inode);
635 	if (offset < 0 || offset >= isize) {
636 		mutex_unlock(&inode->i_mutex);
637 		return -ENXIO;
638 	}
639 
640 	blkbits = inode->i_sb->s_blocksize_bits;
641 	start = offset >> blkbits;
642 	last = start;
643 	end = isize >> blkbits;
644 	holeoff = offset;
645 
646 	do {
647 		map.m_lblk = last;
648 		map.m_len = end - last + 1;
649 		ret = ext4_map_blocks(NULL, inode, &map, 0);
650 		if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
651 			last += ret;
652 			holeoff = (loff_t)last << blkbits;
653 			continue;
654 		}
655 
656 		/*
657 		 * If there is a delay extent at this offset,
658 		 * we will skip this extent.
659 		 */
660 		ext4_es_find_delayed_extent_range(inode, last, last, &es);
661 		if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
662 			last = es.es_lblk + es.es_len;
663 			holeoff = (loff_t)last << blkbits;
664 			continue;
665 		}
666 
667 		/*
668 		 * If there is a unwritten extent at this offset,
669 		 * it will be as a data or a hole according to page
670 		 * cache that has data or not.
671 		 */
672 		if (map.m_flags & EXT4_MAP_UNWRITTEN) {
673 			int unwritten;
674 			unwritten = ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
675 							      &map, &holeoff);
676 			if (!unwritten) {
677 				last += ret;
678 				holeoff = (loff_t)last << blkbits;
679 				continue;
680 			}
681 		}
682 
683 		/* find a hole */
684 		break;
685 	} while (last <= end);
686 
687 	mutex_unlock(&inode->i_mutex);
688 
689 	if (holeoff > isize)
690 		holeoff = isize;
691 
692 	return vfs_setpos(file, holeoff, maxsize);
693 }
694 
695 /*
696  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
697  * by calling generic_file_llseek_size() with the appropriate maxbytes
698  * value for each.
699  */
ext4_llseek(struct file * file,loff_t offset,int whence)700 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
701 {
702 	struct inode *inode = file->f_mapping->host;
703 	loff_t maxbytes;
704 
705 	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
706 		maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
707 	else
708 		maxbytes = inode->i_sb->s_maxbytes;
709 
710 	switch (whence) {
711 	case SEEK_SET:
712 	case SEEK_CUR:
713 	case SEEK_END:
714 		return generic_file_llseek_size(file, offset, whence,
715 						maxbytes, i_size_read(inode));
716 	case SEEK_DATA:
717 		return ext4_seek_data(file, offset, maxbytes);
718 	case SEEK_HOLE:
719 		return ext4_seek_hole(file, offset, maxbytes);
720 	}
721 
722 	return -EINVAL;
723 }
724 
725 const struct file_operations ext4_file_operations = {
726 	.llseek		= ext4_llseek,
727 	.read_iter	= generic_file_read_iter,
728 	.write_iter	= ext4_file_write_iter,
729 	.unlocked_ioctl = ext4_ioctl,
730 #ifdef CONFIG_COMPAT
731 	.compat_ioctl	= ext4_compat_ioctl,
732 #endif
733 	.mmap		= ext4_file_mmap,
734 	.open		= ext4_file_open,
735 	.release	= ext4_release_file,
736 	.fsync		= ext4_sync_file,
737 	.splice_read	= generic_file_splice_read,
738 	.splice_write	= iter_file_splice_write,
739 	.fallocate	= ext4_fallocate,
740 };
741 
742 const struct inode_operations ext4_file_inode_operations = {
743 	.setattr	= ext4_setattr,
744 	.getattr	= ext4_getattr,
745 	.setxattr	= generic_setxattr,
746 	.getxattr	= generic_getxattr,
747 	.listxattr	= ext4_listxattr,
748 	.removexattr	= generic_removexattr,
749 	.get_acl	= ext4_get_acl,
750 	.set_acl	= ext4_set_acl,
751 	.fiemap		= ext4_fiemap,
752 };
753 
754