Commit | Line | Data |
---|---|---|
ac27a0ec | 1 | /* |
617ba13b | 2 | * linux/fs/ext4/file.c |
ac27a0ec DK |
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 | * | |
617ba13b | 15 | * ext4 fs regular file handling primitives |
ac27a0ec DK |
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> | |
bc0b0d6d TT |
23 | #include <linux/mount.h> |
24 | #include <linux/path.h> | |
c94c2acf | 25 | #include <linux/dax.h> |
871a2931 | 26 | #include <linux/quotaops.h> |
c8c0df24 | 27 | #include <linux/pagevec.h> |
e2e40f2c | 28 | #include <linux/uio.h> |
3dcf5451 CH |
29 | #include "ext4.h" |
30 | #include "ext4_jbd2.h" | |
ac27a0ec DK |
31 | #include "xattr.h" |
32 | #include "acl.h" | |
33 | ||
34 | /* | |
35 | * Called when an inode is released. Note that this is different | |
617ba13b | 36 | * from ext4_file_open: open gets called at every open, but release |
ac27a0ec DK |
37 | * gets called only when /all/ the files are closed. |
38 | */ | |
af5bc92d | 39 | static int ext4_release_file(struct inode *inode, struct file *filp) |
ac27a0ec | 40 | { |
19f5fb7a | 41 | if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) { |
7d8f9f7d | 42 | ext4_alloc_da_blocks(inode); |
19f5fb7a | 43 | ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 44 | } |
ac27a0ec DK |
45 | /* if we are the last writer on the inode, drop the block reservation */ |
46 | if ((filp->f_mode & FMODE_WRITE) && | |
d6014301 AK |
47 | (atomic_read(&inode->i_writecount) == 1) && |
48 | !EXT4_I(inode)->i_reserved_data_blocks) | |
ac27a0ec | 49 | { |
0e855ac8 | 50 | down_write(&EXT4_I(inode)->i_data_sem); |
c2ea3fde | 51 | ext4_discard_preallocations(inode); |
0e855ac8 | 52 | up_write(&EXT4_I(inode)->i_data_sem); |
ac27a0ec DK |
53 | } |
54 | if (is_dx(inode) && filp->private_data) | |
617ba13b | 55 | ext4_htree_free_dir_info(filp->private_data); |
ac27a0ec DK |
56 | |
57 | return 0; | |
58 | } | |
59 | ||
c197855e | 60 | static void ext4_unwritten_wait(struct inode *inode) |
e9e3bcec ES |
61 | { |
62 | wait_queue_head_t *wq = ext4_ioend_wq(inode); | |
63 | ||
e27f41e1 | 64 | wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_unwritten) == 0)); |
e9e3bcec ES |
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 | |
9b884164 | 77 | ext4_unaligned_aio(struct inode *inode, struct iov_iter *from, loff_t pos) |
e9e3bcec ES |
78 | { |
79 | struct super_block *sb = inode->i_sb; | |
80 | int blockmask = sb->s_blocksize - 1; | |
e9e3bcec | 81 | |
6e6358fc | 82 | if (pos >= i_size_read(inode)) |
e9e3bcec ES |
83 | return 0; |
84 | ||
9b884164 | 85 | if ((pos | iov_iter_alignment(from)) & blockmask) |
e9e3bcec ES |
86 | return 1; |
87 | ||
88 | return 0; | |
89 | } | |
90 | ||
ac27a0ec | 91 | static ssize_t |
9b884164 | 92 | ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
ac27a0ec | 93 | { |
4bd809db | 94 | struct file *file = iocb->ki_filp; |
8ad2850f | 95 | struct inode *inode = file_inode(iocb->ki_filp); |
7ed07ba8 | 96 | struct mutex *aio_mutex = NULL; |
4bd809db | 97 | struct blk_plug plug; |
2ba48ce5 | 98 | int o_direct = iocb->ki_flags & IOCB_DIRECT; |
4bd809db | 99 | int overwrite = 0; |
8563000d | 100 | ssize_t ret; |
7608e610 | 101 | |
f5ccfe1d TT |
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) && | |
2ba48ce5 | 109 | (iocb->ki_flags & IOCB_APPEND || |
e768d7ff | 110 | ext4_unaligned_aio(inode, from, iocb->ki_pos))) { |
f5ccfe1d TT |
111 | aio_mutex = ext4_aio_mutex(inode); |
112 | mutex_lock(aio_mutex); | |
113 | ext4_unwritten_wait(inode); | |
114 | } | |
115 | ||
5955102c | 116 | inode_lock(inode); |
3309dd04 AV |
117 | ret = generic_write_checks(iocb, from); |
118 | if (ret <= 0) | |
e768d7ff | 119 | goto out; |
f5ccfe1d | 120 | |
e2b46574 ES |
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 | */ | |
12e9b892 | 125 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 | 126 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
ac27a0ec | 127 | |
3309dd04 | 128 | if (iocb->ki_pos >= sbi->s_bitmap_maxbytes) { |
f5ccfe1d | 129 | ret = -EFBIG; |
e768d7ff | 130 | goto out; |
f5ccfe1d | 131 | } |
3309dd04 | 132 | iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos); |
e2b46574 ES |
133 | } |
134 | ||
a41537e6 | 135 | iocb->private = &overwrite; |
7ed07ba8 | 136 | if (o_direct) { |
3309dd04 AV |
137 | size_t length = iov_iter_count(from); |
138 | loff_t pos = iocb->ki_pos; | |
8ad2850f TT |
139 | blk_start_plug(&plug); |
140 | ||
8ad2850f | 141 | /* check whether we do a DIO overwrite or not */ |
7ed07ba8 | 142 | if (ext4_should_dioread_nolock(inode) && !aio_mutex && |
8ad2850f TT |
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 | } | |
f5ccfe1d | 169 | } |
7608e610 | 170 | |
9b884164 | 171 | ret = __generic_file_write_iter(iocb, from); |
5955102c | 172 | inode_unlock(inode); |
7608e610 | 173 | |
7ed07ba8 TT |
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; | |
7608e610 | 180 | } |
7ed07ba8 TT |
181 | if (o_direct) |
182 | blk_finish_plug(&plug); | |
e9e3bcec | 183 | |
e768d7ff AV |
184 | if (aio_mutex) |
185 | mutex_unlock(aio_mutex); | |
186 | return ret; | |
187 | ||
188 | out: | |
5955102c | 189 | inode_unlock(inode); |
7ed07ba8 TT |
190 | if (aio_mutex) |
191 | mutex_unlock(aio_mutex); | |
e9e3bcec | 192 | return ret; |
ac27a0ec DK |
193 | } |
194 | ||
923ae0ff RZ |
195 | #ifdef CONFIG_FS_DAX |
196 | static int ext4_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf) | |
197 | { | |
01a33b4a MW |
198 | int result; |
199 | handle_t *handle = NULL; | |
ea3d7209 JK |
200 | struct inode *inode = file_inode(vma->vm_file); |
201 | struct super_block *sb = inode->i_sb; | |
01a33b4a MW |
202 | bool write = vmf->flags & FAULT_FLAG_WRITE; |
203 | ||
204 | if (write) { | |
205 | sb_start_pagefault(sb); | |
206 | file_update_time(vma->vm_file); | |
ea3d7209 | 207 | down_read(&EXT4_I(inode)->i_mmap_sem); |
01a33b4a MW |
208 | handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE, |
209 | EXT4_DATA_TRANS_BLOCKS(sb)); | |
ea3d7209 JK |
210 | } else |
211 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
01a33b4a MW |
212 | |
213 | if (IS_ERR(handle)) | |
214 | result = VM_FAULT_SIGBUS; | |
215 | else | |
ba5843f5 | 216 | result = __dax_fault(vma, vmf, ext4_dax_mmap_get_block, NULL); |
01a33b4a MW |
217 | |
218 | if (write) { | |
219 | if (!IS_ERR(handle)) | |
220 | ext4_journal_stop(handle); | |
ea3d7209 | 221 | up_read(&EXT4_I(inode)->i_mmap_sem); |
01a33b4a | 222 | sb_end_pagefault(sb); |
ea3d7209 JK |
223 | } else |
224 | up_read(&EXT4_I(inode)->i_mmap_sem); | |
01a33b4a MW |
225 | |
226 | return result; | |
923ae0ff RZ |
227 | } |
228 | ||
11bd1a9e MW |
229 | static int ext4_dax_pmd_fault(struct vm_area_struct *vma, unsigned long addr, |
230 | pmd_t *pmd, unsigned int flags) | |
231 | { | |
01a33b4a MW |
232 | int result; |
233 | handle_t *handle = NULL; | |
234 | struct inode *inode = file_inode(vma->vm_file); | |
235 | struct super_block *sb = inode->i_sb; | |
236 | bool write = flags & FAULT_FLAG_WRITE; | |
237 | ||
238 | if (write) { | |
239 | sb_start_pagefault(sb); | |
240 | file_update_time(vma->vm_file); | |
ea3d7209 | 241 | down_read(&EXT4_I(inode)->i_mmap_sem); |
01a33b4a MW |
242 | handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE, |
243 | ext4_chunk_trans_blocks(inode, | |
244 | PMD_SIZE / PAGE_SIZE)); | |
ea3d7209 JK |
245 | } else |
246 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
01a33b4a MW |
247 | |
248 | if (IS_ERR(handle)) | |
249 | result = VM_FAULT_SIGBUS; | |
250 | else | |
251 | result = __dax_pmd_fault(vma, addr, pmd, flags, | |
ba5843f5 | 252 | ext4_dax_mmap_get_block, NULL); |
01a33b4a MW |
253 | |
254 | if (write) { | |
255 | if (!IS_ERR(handle)) | |
256 | ext4_journal_stop(handle); | |
ea3d7209 | 257 | up_read(&EXT4_I(inode)->i_mmap_sem); |
01a33b4a | 258 | sb_end_pagefault(sb); |
ea3d7209 JK |
259 | } else |
260 | up_read(&EXT4_I(inode)->i_mmap_sem); | |
01a33b4a MW |
261 | |
262 | return result; | |
11bd1a9e MW |
263 | } |
264 | ||
923ae0ff RZ |
265 | static int ext4_dax_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
266 | { | |
ea3d7209 JK |
267 | int err; |
268 | struct inode *inode = file_inode(vma->vm_file); | |
269 | ||
270 | sb_start_pagefault(inode->i_sb); | |
271 | file_update_time(vma->vm_file); | |
272 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
ba5843f5 | 273 | err = __dax_mkwrite(vma, vmf, ext4_dax_mmap_get_block, NULL); |
ea3d7209 JK |
274 | up_read(&EXT4_I(inode)->i_mmap_sem); |
275 | sb_end_pagefault(inode->i_sb); | |
276 | ||
277 | return err; | |
278 | } | |
279 | ||
280 | /* | |
281 | * Handle write fault for VM_MIXEDMAP mappings. Similarly to ext4_dax_mkwrite() | |
282 | * handler we check for races agaist truncate. Note that since we cycle through | |
283 | * i_mmap_sem, we are sure that also any hole punching that began before we | |
284 | * were called is finished by now and so if it included part of the file we | |
285 | * are working on, our pte will get unmapped and the check for pte_same() in | |
286 | * wp_pfn_shared() fails. Thus fault gets retried and things work out as | |
287 | * desired. | |
288 | */ | |
289 | static int ext4_dax_pfn_mkwrite(struct vm_area_struct *vma, | |
290 | struct vm_fault *vmf) | |
291 | { | |
292 | struct inode *inode = file_inode(vma->vm_file); | |
293 | struct super_block *sb = inode->i_sb; | |
ea3d7209 | 294 | loff_t size; |
d5be7a03 | 295 | int ret; |
ea3d7209 JK |
296 | |
297 | sb_start_pagefault(sb); | |
298 | file_update_time(vma->vm_file); | |
299 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
300 | size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
301 | if (vmf->pgoff >= size) | |
302 | ret = VM_FAULT_SIGBUS; | |
d5be7a03 RZ |
303 | else |
304 | ret = dax_pfn_mkwrite(vma, vmf); | |
ea3d7209 JK |
305 | up_read(&EXT4_I(inode)->i_mmap_sem); |
306 | sb_end_pagefault(sb); | |
307 | ||
308 | return ret; | |
923ae0ff RZ |
309 | } |
310 | ||
311 | static const struct vm_operations_struct ext4_dax_vm_ops = { | |
312 | .fault = ext4_dax_fault, | |
11bd1a9e | 313 | .pmd_fault = ext4_dax_pmd_fault, |
923ae0ff | 314 | .page_mkwrite = ext4_dax_mkwrite, |
ea3d7209 | 315 | .pfn_mkwrite = ext4_dax_pfn_mkwrite, |
923ae0ff RZ |
316 | }; |
317 | #else | |
318 | #define ext4_dax_vm_ops ext4_file_vm_ops | |
319 | #endif | |
320 | ||
f0f37e2f | 321 | static const struct vm_operations_struct ext4_file_vm_ops = { |
ea3d7209 | 322 | .fault = ext4_filemap_fault, |
f1820361 | 323 | .map_pages = filemap_map_pages, |
2e9ee850 AK |
324 | .page_mkwrite = ext4_page_mkwrite, |
325 | }; | |
326 | ||
327 | static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma) | |
328 | { | |
c9c7429c MH |
329 | struct inode *inode = file->f_mapping->host; |
330 | ||
331 | if (ext4_encrypted_inode(inode)) { | |
b7236e21 | 332 | int err = ext4_get_encryption_info(inode); |
c9c7429c MH |
333 | if (err) |
334 | return 0; | |
abdd438b TT |
335 | if (ext4_encryption_info(inode) == NULL) |
336 | return -ENOKEY; | |
c9c7429c | 337 | } |
2e9ee850 | 338 | file_accessed(file); |
923ae0ff RZ |
339 | if (IS_DAX(file_inode(file))) { |
340 | vma->vm_ops = &ext4_dax_vm_ops; | |
11bd1a9e | 341 | vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE; |
923ae0ff RZ |
342 | } else { |
343 | vma->vm_ops = &ext4_file_vm_ops; | |
344 | } | |
2e9ee850 AK |
345 | return 0; |
346 | } | |
347 | ||
bc0b0d6d TT |
348 | static int ext4_file_open(struct inode * inode, struct file * filp) |
349 | { | |
350 | struct super_block *sb = inode->i_sb; | |
351 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
352 | struct vfsmount *mnt = filp->f_path.mnt; | |
353 | struct path path; | |
354 | char buf[64], *cp; | |
c9c7429c | 355 | int ret; |
bc0b0d6d TT |
356 | |
357 | if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) && | |
358 | !(sb->s_flags & MS_RDONLY))) { | |
359 | sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED; | |
360 | /* | |
361 | * Sample where the filesystem has been mounted and | |
362 | * store it in the superblock for sysadmin convenience | |
363 | * when trying to sort through large numbers of block | |
364 | * devices or filesystem images. | |
365 | */ | |
366 | memset(buf, 0, sizeof(buf)); | |
3899167d AV |
367 | path.mnt = mnt; |
368 | path.dentry = mnt->mnt_root; | |
bc0b0d6d | 369 | cp = d_path(&path, buf, sizeof(buf)); |
bc0b0d6d | 370 | if (!IS_ERR(cp)) { |
044ce47f JK |
371 | handle_t *handle; |
372 | int err; | |
373 | ||
9924a92a | 374 | handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1); |
044ce47f JK |
375 | if (IS_ERR(handle)) |
376 | return PTR_ERR(handle); | |
5d601255 | 377 | BUFFER_TRACE(sbi->s_sbh, "get_write_access"); |
044ce47f JK |
378 | err = ext4_journal_get_write_access(handle, sbi->s_sbh); |
379 | if (err) { | |
380 | ext4_journal_stop(handle); | |
381 | return err; | |
382 | } | |
cf803903 DW |
383 | strlcpy(sbi->s_es->s_last_mounted, cp, |
384 | sizeof(sbi->s_es->s_last_mounted)); | |
044ce47f JK |
385 | ext4_handle_dirty_super(handle, sb); |
386 | ext4_journal_stop(handle); | |
bc0b0d6d TT |
387 | } |
388 | } | |
abdd438b TT |
389 | if (ext4_encrypted_inode(inode)) { |
390 | ret = ext4_get_encryption_info(inode); | |
391 | if (ret) | |
392 | return -EACCES; | |
393 | if (ext4_encryption_info(inode) == NULL) | |
394 | return -ENOKEY; | |
395 | } | |
8aefcd55 TT |
396 | /* |
397 | * Set up the jbd2_inode if we are opening the inode for | |
398 | * writing and the journal is present | |
399 | */ | |
a361293f | 400 | if (filp->f_mode & FMODE_WRITE) { |
c9c7429c | 401 | ret = ext4_inode_attach_jinode(inode); |
a361293f JK |
402 | if (ret < 0) |
403 | return ret; | |
8aefcd55 | 404 | } |
abdd438b | 405 | return dquot_file_open(inode, filp); |
bc0b0d6d TT |
406 | } |
407 | ||
c8c0df24 ZL |
408 | /* |
409 | * Here we use ext4_map_blocks() to get a block mapping for a extent-based | |
410 | * file rather than ext4_ext_walk_space() because we can introduce | |
411 | * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same | |
412 | * function. When extent status tree has been fully implemented, it will | |
413 | * track all extent status for a file and we can directly use it to | |
414 | * retrieve the offset for SEEK_DATA/SEEK_HOLE. | |
415 | */ | |
416 | ||
417 | /* | |
418 | * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to | |
419 | * lookup page cache to check whether or not there has some data between | |
420 | * [startoff, endoff] because, if this range contains an unwritten extent, | |
421 | * we determine this extent as a data or a hole according to whether the | |
422 | * page cache has data or not. | |
423 | */ | |
ad7fefb1 TT |
424 | static int ext4_find_unwritten_pgoff(struct inode *inode, |
425 | int whence, | |
426 | struct ext4_map_blocks *map, | |
427 | loff_t *offset) | |
c8c0df24 ZL |
428 | { |
429 | struct pagevec pvec; | |
ad7fefb1 | 430 | unsigned int blkbits; |
c8c0df24 ZL |
431 | pgoff_t index; |
432 | pgoff_t end; | |
ad7fefb1 | 433 | loff_t endoff; |
c8c0df24 ZL |
434 | loff_t startoff; |
435 | loff_t lastoff; | |
436 | int found = 0; | |
437 | ||
ad7fefb1 | 438 | blkbits = inode->i_sb->s_blocksize_bits; |
c8c0df24 ZL |
439 | startoff = *offset; |
440 | lastoff = startoff; | |
ad7fefb1 | 441 | endoff = (loff_t)(map->m_lblk + map->m_len) << blkbits; |
c8c0df24 ZL |
442 | |
443 | index = startoff >> PAGE_CACHE_SHIFT; | |
444 | end = endoff >> PAGE_CACHE_SHIFT; | |
445 | ||
446 | pagevec_init(&pvec, 0); | |
447 | do { | |
448 | int i, num; | |
449 | unsigned long nr_pages; | |
450 | ||
451 | num = min_t(pgoff_t, end - index, PAGEVEC_SIZE); | |
452 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index, | |
453 | (pgoff_t)num); | |
454 | if (nr_pages == 0) { | |
965c8e59 | 455 | if (whence == SEEK_DATA) |
c8c0df24 ZL |
456 | break; |
457 | ||
965c8e59 | 458 | BUG_ON(whence != SEEK_HOLE); |
c8c0df24 ZL |
459 | /* |
460 | * If this is the first time to go into the loop and | |
461 | * offset is not beyond the end offset, it will be a | |
462 | * hole at this offset | |
463 | */ | |
464 | if (lastoff == startoff || lastoff < endoff) | |
465 | found = 1; | |
466 | break; | |
467 | } | |
468 | ||
469 | /* | |
470 | * If this is the first time to go into the loop and | |
471 | * offset is smaller than the first page offset, it will be a | |
472 | * hole at this offset. | |
473 | */ | |
965c8e59 | 474 | if (lastoff == startoff && whence == SEEK_HOLE && |
c8c0df24 ZL |
475 | lastoff < page_offset(pvec.pages[0])) { |
476 | found = 1; | |
477 | break; | |
478 | } | |
479 | ||
480 | for (i = 0; i < nr_pages; i++) { | |
481 | struct page *page = pvec.pages[i]; | |
482 | struct buffer_head *bh, *head; | |
483 | ||
484 | /* | |
485 | * If the current offset is not beyond the end of given | |
486 | * range, it will be a hole. | |
487 | */ | |
965c8e59 | 488 | if (lastoff < endoff && whence == SEEK_HOLE && |
c8c0df24 ZL |
489 | page->index > end) { |
490 | found = 1; | |
491 | *offset = lastoff; | |
492 | goto out; | |
493 | } | |
494 | ||
495 | lock_page(page); | |
496 | ||
497 | if (unlikely(page->mapping != inode->i_mapping)) { | |
498 | unlock_page(page); | |
499 | continue; | |
500 | } | |
501 | ||
502 | if (!page_has_buffers(page)) { | |
503 | unlock_page(page); | |
504 | continue; | |
505 | } | |
506 | ||
507 | if (page_has_buffers(page)) { | |
508 | lastoff = page_offset(page); | |
509 | bh = head = page_buffers(page); | |
510 | do { | |
511 | if (buffer_uptodate(bh) || | |
512 | buffer_unwritten(bh)) { | |
965c8e59 | 513 | if (whence == SEEK_DATA) |
c8c0df24 ZL |
514 | found = 1; |
515 | } else { | |
965c8e59 | 516 | if (whence == SEEK_HOLE) |
c8c0df24 ZL |
517 | found = 1; |
518 | } | |
519 | if (found) { | |
520 | *offset = max_t(loff_t, | |
521 | startoff, lastoff); | |
522 | unlock_page(page); | |
523 | goto out; | |
524 | } | |
525 | lastoff += bh->b_size; | |
526 | bh = bh->b_this_page; | |
527 | } while (bh != head); | |
528 | } | |
529 | ||
530 | lastoff = page_offset(page) + PAGE_SIZE; | |
531 | unlock_page(page); | |
532 | } | |
533 | ||
534 | /* | |
535 | * The no. of pages is less than our desired, that would be a | |
536 | * hole in there. | |
537 | */ | |
965c8e59 | 538 | if (nr_pages < num && whence == SEEK_HOLE) { |
c8c0df24 ZL |
539 | found = 1; |
540 | *offset = lastoff; | |
541 | break; | |
542 | } | |
543 | ||
544 | index = pvec.pages[i - 1]->index + 1; | |
545 | pagevec_release(&pvec); | |
546 | } while (index <= end); | |
547 | ||
548 | out: | |
549 | pagevec_release(&pvec); | |
550 | return found; | |
551 | } | |
552 | ||
553 | /* | |
554 | * ext4_seek_data() retrieves the offset for SEEK_DATA. | |
555 | */ | |
556 | static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize) | |
557 | { | |
558 | struct inode *inode = file->f_mapping->host; | |
ad7fefb1 TT |
559 | struct ext4_map_blocks map; |
560 | struct extent_status es; | |
561 | ext4_lblk_t start, last, end; | |
562 | loff_t dataoff, isize; | |
563 | int blkbits; | |
564 | int ret = 0; | |
c8c0df24 | 565 | |
5955102c | 566 | inode_lock(inode); |
ad7fefb1 TT |
567 | |
568 | isize = i_size_read(inode); | |
569 | if (offset >= isize) { | |
5955102c | 570 | inode_unlock(inode); |
c8c0df24 ZL |
571 | return -ENXIO; |
572 | } | |
ad7fefb1 TT |
573 | |
574 | blkbits = inode->i_sb->s_blocksize_bits; | |
575 | start = offset >> blkbits; | |
576 | last = start; | |
577 | end = isize >> blkbits; | |
578 | dataoff = offset; | |
579 | ||
580 | do { | |
581 | map.m_lblk = last; | |
582 | map.m_len = end - last + 1; | |
583 | ret = ext4_map_blocks(NULL, inode, &map, 0); | |
584 | if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) { | |
585 | if (last != start) | |
586 | dataoff = (loff_t)last << blkbits; | |
c8c0df24 | 587 | break; |
ad7fefb1 | 588 | } |
c8c0df24 | 589 | |
ad7fefb1 TT |
590 | /* |
591 | * If there is a delay extent at this offset, | |
592 | * it will be as a data. | |
593 | */ | |
594 | ext4_es_find_delayed_extent_range(inode, last, last, &es); | |
595 | if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) { | |
596 | if (last != start) | |
597 | dataoff = (loff_t)last << blkbits; | |
c8c0df24 ZL |
598 | break; |
599 | } | |
600 | ||
ad7fefb1 TT |
601 | /* |
602 | * If there is a unwritten extent at this offset, | |
603 | * it will be as a data or a hole according to page | |
604 | * cache that has data or not. | |
605 | */ | |
606 | if (map.m_flags & EXT4_MAP_UNWRITTEN) { | |
607 | int unwritten; | |
608 | unwritten = ext4_find_unwritten_pgoff(inode, SEEK_DATA, | |
609 | &map, &dataoff); | |
610 | if (unwritten) | |
611 | break; | |
612 | } | |
c8c0df24 | 613 | |
ad7fefb1 TT |
614 | last++; |
615 | dataoff = (loff_t)last << blkbits; | |
616 | } while (last <= end); | |
c8c0df24 | 617 | |
5955102c | 618 | inode_unlock(inode); |
c8c0df24 | 619 | |
ad7fefb1 TT |
620 | if (dataoff > isize) |
621 | return -ENXIO; | |
622 | ||
623 | return vfs_setpos(file, dataoff, maxsize); | |
c8c0df24 ZL |
624 | } |
625 | ||
626 | /* | |
ad7fefb1 | 627 | * ext4_seek_hole() retrieves the offset for SEEK_HOLE. |
c8c0df24 ZL |
628 | */ |
629 | static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize) | |
630 | { | |
631 | struct inode *inode = file->f_mapping->host; | |
ad7fefb1 TT |
632 | struct ext4_map_blocks map; |
633 | struct extent_status es; | |
634 | ext4_lblk_t start, last, end; | |
635 | loff_t holeoff, isize; | |
636 | int blkbits; | |
637 | int ret = 0; | |
c8c0df24 | 638 | |
5955102c | 639 | inode_lock(inode); |
ad7fefb1 TT |
640 | |
641 | isize = i_size_read(inode); | |
642 | if (offset >= isize) { | |
5955102c | 643 | inode_unlock(inode); |
c8c0df24 ZL |
644 | return -ENXIO; |
645 | } | |
646 | ||
ad7fefb1 TT |
647 | blkbits = inode->i_sb->s_blocksize_bits; |
648 | start = offset >> blkbits; | |
649 | last = start; | |
650 | end = isize >> blkbits; | |
651 | holeoff = offset; | |
c8c0df24 | 652 | |
ad7fefb1 TT |
653 | do { |
654 | map.m_lblk = last; | |
655 | map.m_len = end - last + 1; | |
656 | ret = ext4_map_blocks(NULL, inode, &map, 0); | |
657 | if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) { | |
658 | last += ret; | |
659 | holeoff = (loff_t)last << blkbits; | |
660 | continue; | |
661 | } | |
c8c0df24 | 662 | |
ad7fefb1 TT |
663 | /* |
664 | * If there is a delay extent at this offset, | |
665 | * we will skip this extent. | |
666 | */ | |
667 | ext4_es_find_delayed_extent_range(inode, last, last, &es); | |
668 | if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) { | |
669 | last = es.es_lblk + es.es_len; | |
670 | holeoff = (loff_t)last << blkbits; | |
671 | continue; | |
672 | } | |
14516bb7 | 673 | |
ad7fefb1 TT |
674 | /* |
675 | * If there is a unwritten extent at this offset, | |
676 | * it will be as a data or a hole according to page | |
677 | * cache that has data or not. | |
678 | */ | |
679 | if (map.m_flags & EXT4_MAP_UNWRITTEN) { | |
680 | int unwritten; | |
681 | unwritten = ext4_find_unwritten_pgoff(inode, SEEK_HOLE, | |
682 | &map, &holeoff); | |
683 | if (!unwritten) { | |
684 | last += ret; | |
685 | holeoff = (loff_t)last << blkbits; | |
c8c0df24 ZL |
686 | continue; |
687 | } | |
14516bb7 | 688 | } |
ad7fefb1 TT |
689 | |
690 | /* find a hole */ | |
691 | break; | |
692 | } while (last <= end); | |
693 | ||
5955102c | 694 | inode_unlock(inode); |
c8c0df24 | 695 | |
ad7fefb1 TT |
696 | if (holeoff > isize) |
697 | holeoff = isize; | |
698 | ||
699 | return vfs_setpos(file, holeoff, maxsize); | |
c8c0df24 ZL |
700 | } |
701 | ||
e0d10bfa | 702 | /* |
ec7268ce ES |
703 | * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values |
704 | * by calling generic_file_llseek_size() with the appropriate maxbytes | |
705 | * value for each. | |
e0d10bfa | 706 | */ |
965c8e59 | 707 | loff_t ext4_llseek(struct file *file, loff_t offset, int whence) |
e0d10bfa TO |
708 | { |
709 | struct inode *inode = file->f_mapping->host; | |
710 | loff_t maxbytes; | |
711 | ||
712 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) | |
713 | maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes; | |
714 | else | |
715 | maxbytes = inode->i_sb->s_maxbytes; | |
e0d10bfa | 716 | |
965c8e59 | 717 | switch (whence) { |
c8c0df24 ZL |
718 | case SEEK_SET: |
719 | case SEEK_CUR: | |
720 | case SEEK_END: | |
965c8e59 | 721 | return generic_file_llseek_size(file, offset, whence, |
c8c0df24 ZL |
722 | maxbytes, i_size_read(inode)); |
723 | case SEEK_DATA: | |
724 | return ext4_seek_data(file, offset, maxbytes); | |
725 | case SEEK_HOLE: | |
726 | return ext4_seek_hole(file, offset, maxbytes); | |
727 | } | |
728 | ||
729 | return -EINVAL; | |
e0d10bfa TO |
730 | } |
731 | ||
617ba13b | 732 | const struct file_operations ext4_file_operations = { |
e0d10bfa | 733 | .llseek = ext4_llseek, |
aad4f8bb | 734 | .read_iter = generic_file_read_iter, |
9b884164 | 735 | .write_iter = ext4_file_write_iter, |
5cdd7b2d | 736 | .unlocked_ioctl = ext4_ioctl, |
ac27a0ec | 737 | #ifdef CONFIG_COMPAT |
617ba13b | 738 | .compat_ioctl = ext4_compat_ioctl, |
ac27a0ec | 739 | #endif |
2e9ee850 | 740 | .mmap = ext4_file_mmap, |
bc0b0d6d | 741 | .open = ext4_file_open, |
617ba13b MC |
742 | .release = ext4_release_file, |
743 | .fsync = ext4_sync_file, | |
ac27a0ec | 744 | .splice_read = generic_file_splice_read, |
8d020765 | 745 | .splice_write = iter_file_splice_write, |
2fe17c10 | 746 | .fallocate = ext4_fallocate, |
ac27a0ec DK |
747 | }; |
748 | ||
754661f1 | 749 | const struct inode_operations ext4_file_inode_operations = { |
617ba13b | 750 | .setattr = ext4_setattr, |
3e3398a0 | 751 | .getattr = ext4_getattr, |
ac27a0ec DK |
752 | .setxattr = generic_setxattr, |
753 | .getxattr = generic_getxattr, | |
617ba13b | 754 | .listxattr = ext4_listxattr, |
ac27a0ec | 755 | .removexattr = generic_removexattr, |
4e34e719 | 756 | .get_acl = ext4_get_acl, |
64e178a7 | 757 | .set_acl = ext4_set_acl, |
6873fa0d | 758 | .fiemap = ext4_fiemap, |
ac27a0ec DK |
759 | }; |
760 |