]>
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); |
4bd809db | 96 | struct blk_plug plug; |
2ba48ce5 | 97 | int o_direct = iocb->ki_flags & IOCB_DIRECT; |
e142d052 | 98 | int unaligned_aio = 0; |
4bd809db | 99 | int overwrite = 0; |
8563000d | 100 | ssize_t ret; |
7608e610 | 101 | |
e142d052 JK |
102 | inode_lock(inode); |
103 | ret = generic_write_checks(iocb, from); | |
104 | if (ret <= 0) | |
105 | goto out; | |
106 | ||
f5ccfe1d | 107 | /* |
e142d052 JK |
108 | * Unaligned direct AIO must be serialized among each other as zeroing |
109 | * of partial blocks of two competing unaligned AIOs can result in data | |
110 | * corruption. | |
f5ccfe1d | 111 | */ |
e142d052 | 112 | if (o_direct && ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) && |
f5ccfe1d | 113 | !is_sync_kiocb(iocb) && |
e142d052 JK |
114 | ext4_unaligned_aio(inode, from, iocb->ki_pos)) { |
115 | unaligned_aio = 1; | |
f5ccfe1d TT |
116 | ext4_unwritten_wait(inode); |
117 | } | |
118 | ||
e2b46574 ES |
119 | /* |
120 | * If we have encountered a bitmap-format file, the size limit | |
121 | * is smaller than s_maxbytes, which is for extent-mapped files. | |
122 | */ | |
12e9b892 | 123 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 | 124 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
ac27a0ec | 125 | |
3309dd04 | 126 | if (iocb->ki_pos >= sbi->s_bitmap_maxbytes) { |
f5ccfe1d | 127 | ret = -EFBIG; |
e768d7ff | 128 | goto out; |
f5ccfe1d | 129 | } |
3309dd04 | 130 | iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos); |
e2b46574 ES |
131 | } |
132 | ||
a41537e6 | 133 | iocb->private = &overwrite; |
7ed07ba8 | 134 | if (o_direct) { |
3309dd04 AV |
135 | size_t length = iov_iter_count(from); |
136 | loff_t pos = iocb->ki_pos; | |
8ad2850f TT |
137 | blk_start_plug(&plug); |
138 | ||
8ad2850f | 139 | /* check whether we do a DIO overwrite or not */ |
e142d052 | 140 | if (ext4_should_dioread_nolock(inode) && !unaligned_aio && |
8ad2850f TT |
141 | !file->f_mapping->nrpages && pos + length <= i_size_read(inode)) { |
142 | struct ext4_map_blocks map; | |
143 | unsigned int blkbits = inode->i_blkbits; | |
144 | int err, len; | |
145 | ||
146 | map.m_lblk = pos >> blkbits; | |
147 | map.m_len = (EXT4_BLOCK_ALIGN(pos + length, blkbits) >> blkbits) | |
148 | - map.m_lblk; | |
149 | len = map.m_len; | |
150 | ||
151 | err = ext4_map_blocks(NULL, inode, &map, 0); | |
152 | /* | |
153 | * 'err==len' means that all of blocks has | |
154 | * been preallocated no matter they are | |
155 | * initialized or not. For excluding | |
156 | * unwritten extents, we need to check | |
157 | * m_flags. There are two conditions that | |
158 | * indicate for initialized extents. 1) If we | |
159 | * hit extent cache, EXT4_MAP_MAPPED flag is | |
160 | * returned; 2) If we do a real lookup, | |
161 | * non-flags are returned. So we should check | |
162 | * these two conditions. | |
163 | */ | |
164 | if (err == len && (map.m_flags & EXT4_MAP_MAPPED)) | |
165 | overwrite = 1; | |
166 | } | |
f5ccfe1d | 167 | } |
7608e610 | 168 | |
9b884164 | 169 | ret = __generic_file_write_iter(iocb, from); |
5955102c | 170 | inode_unlock(inode); |
7608e610 | 171 | |
7ed07ba8 TT |
172 | if (ret > 0) { |
173 | ssize_t err; | |
174 | ||
175 | err = generic_write_sync(file, iocb->ki_pos - ret, ret); | |
176 | if (err < 0) | |
177 | ret = err; | |
7608e610 | 178 | } |
7ed07ba8 TT |
179 | if (o_direct) |
180 | blk_finish_plug(&plug); | |
e9e3bcec | 181 | |
e768d7ff AV |
182 | return ret; |
183 | ||
184 | out: | |
5955102c | 185 | inode_unlock(inode); |
e9e3bcec | 186 | return ret; |
ac27a0ec DK |
187 | } |
188 | ||
923ae0ff RZ |
189 | #ifdef CONFIG_FS_DAX |
190 | static int ext4_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf) | |
191 | { | |
01a33b4a MW |
192 | int result; |
193 | handle_t *handle = NULL; | |
ea3d7209 JK |
194 | struct inode *inode = file_inode(vma->vm_file); |
195 | struct super_block *sb = inode->i_sb; | |
01a33b4a MW |
196 | bool write = vmf->flags & FAULT_FLAG_WRITE; |
197 | ||
198 | if (write) { | |
199 | sb_start_pagefault(sb); | |
200 | file_update_time(vma->vm_file); | |
ea3d7209 | 201 | down_read(&EXT4_I(inode)->i_mmap_sem); |
01a33b4a MW |
202 | handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE, |
203 | EXT4_DATA_TRANS_BLOCKS(sb)); | |
ea3d7209 JK |
204 | } else |
205 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
01a33b4a MW |
206 | |
207 | if (IS_ERR(handle)) | |
208 | result = VM_FAULT_SIGBUS; | |
209 | else | |
ba5843f5 | 210 | result = __dax_fault(vma, vmf, ext4_dax_mmap_get_block, NULL); |
01a33b4a MW |
211 | |
212 | if (write) { | |
213 | if (!IS_ERR(handle)) | |
214 | ext4_journal_stop(handle); | |
ea3d7209 | 215 | up_read(&EXT4_I(inode)->i_mmap_sem); |
01a33b4a | 216 | sb_end_pagefault(sb); |
ea3d7209 JK |
217 | } else |
218 | up_read(&EXT4_I(inode)->i_mmap_sem); | |
01a33b4a MW |
219 | |
220 | return result; | |
923ae0ff RZ |
221 | } |
222 | ||
11bd1a9e MW |
223 | static int ext4_dax_pmd_fault(struct vm_area_struct *vma, unsigned long addr, |
224 | pmd_t *pmd, unsigned int flags) | |
225 | { | |
01a33b4a MW |
226 | int result; |
227 | handle_t *handle = NULL; | |
228 | struct inode *inode = file_inode(vma->vm_file); | |
229 | struct super_block *sb = inode->i_sb; | |
230 | bool write = flags & FAULT_FLAG_WRITE; | |
231 | ||
232 | if (write) { | |
233 | sb_start_pagefault(sb); | |
234 | file_update_time(vma->vm_file); | |
ea3d7209 | 235 | down_read(&EXT4_I(inode)->i_mmap_sem); |
01a33b4a MW |
236 | handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE, |
237 | ext4_chunk_trans_blocks(inode, | |
238 | PMD_SIZE / PAGE_SIZE)); | |
ea3d7209 JK |
239 | } else |
240 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
01a33b4a MW |
241 | |
242 | if (IS_ERR(handle)) | |
243 | result = VM_FAULT_SIGBUS; | |
244 | else | |
245 | result = __dax_pmd_fault(vma, addr, pmd, flags, | |
ba5843f5 | 246 | ext4_dax_mmap_get_block, NULL); |
01a33b4a MW |
247 | |
248 | if (write) { | |
249 | if (!IS_ERR(handle)) | |
250 | ext4_journal_stop(handle); | |
ea3d7209 | 251 | up_read(&EXT4_I(inode)->i_mmap_sem); |
01a33b4a | 252 | sb_end_pagefault(sb); |
ea3d7209 JK |
253 | } else |
254 | up_read(&EXT4_I(inode)->i_mmap_sem); | |
01a33b4a MW |
255 | |
256 | return result; | |
11bd1a9e MW |
257 | } |
258 | ||
923ae0ff RZ |
259 | static int ext4_dax_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
260 | { | |
ea3d7209 JK |
261 | int err; |
262 | struct inode *inode = file_inode(vma->vm_file); | |
263 | ||
264 | sb_start_pagefault(inode->i_sb); | |
265 | file_update_time(vma->vm_file); | |
266 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
ba5843f5 | 267 | err = __dax_mkwrite(vma, vmf, ext4_dax_mmap_get_block, NULL); |
ea3d7209 JK |
268 | up_read(&EXT4_I(inode)->i_mmap_sem); |
269 | sb_end_pagefault(inode->i_sb); | |
270 | ||
271 | return err; | |
272 | } | |
273 | ||
274 | /* | |
275 | * Handle write fault for VM_MIXEDMAP mappings. Similarly to ext4_dax_mkwrite() | |
276 | * handler we check for races agaist truncate. Note that since we cycle through | |
277 | * i_mmap_sem, we are sure that also any hole punching that began before we | |
278 | * were called is finished by now and so if it included part of the file we | |
279 | * are working on, our pte will get unmapped and the check for pte_same() in | |
280 | * wp_pfn_shared() fails. Thus fault gets retried and things work out as | |
281 | * desired. | |
282 | */ | |
283 | static int ext4_dax_pfn_mkwrite(struct vm_area_struct *vma, | |
284 | struct vm_fault *vmf) | |
285 | { | |
286 | struct inode *inode = file_inode(vma->vm_file); | |
287 | struct super_block *sb = inode->i_sb; | |
ea3d7209 | 288 | loff_t size; |
d5be7a03 | 289 | int ret; |
ea3d7209 JK |
290 | |
291 | sb_start_pagefault(sb); | |
292 | file_update_time(vma->vm_file); | |
293 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
294 | size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
295 | if (vmf->pgoff >= size) | |
296 | ret = VM_FAULT_SIGBUS; | |
d5be7a03 RZ |
297 | else |
298 | ret = dax_pfn_mkwrite(vma, vmf); | |
ea3d7209 JK |
299 | up_read(&EXT4_I(inode)->i_mmap_sem); |
300 | sb_end_pagefault(sb); | |
301 | ||
302 | return ret; | |
923ae0ff RZ |
303 | } |
304 | ||
305 | static const struct vm_operations_struct ext4_dax_vm_ops = { | |
306 | .fault = ext4_dax_fault, | |
11bd1a9e | 307 | .pmd_fault = ext4_dax_pmd_fault, |
923ae0ff | 308 | .page_mkwrite = ext4_dax_mkwrite, |
ea3d7209 | 309 | .pfn_mkwrite = ext4_dax_pfn_mkwrite, |
923ae0ff RZ |
310 | }; |
311 | #else | |
312 | #define ext4_dax_vm_ops ext4_file_vm_ops | |
313 | #endif | |
314 | ||
f0f37e2f | 315 | static const struct vm_operations_struct ext4_file_vm_ops = { |
ea3d7209 | 316 | .fault = ext4_filemap_fault, |
f1820361 | 317 | .map_pages = filemap_map_pages, |
2e9ee850 AK |
318 | .page_mkwrite = ext4_page_mkwrite, |
319 | }; | |
320 | ||
321 | static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma) | |
322 | { | |
c9c7429c MH |
323 | struct inode *inode = file->f_mapping->host; |
324 | ||
325 | if (ext4_encrypted_inode(inode)) { | |
b7236e21 | 326 | int err = ext4_get_encryption_info(inode); |
c9c7429c MH |
327 | if (err) |
328 | return 0; | |
abdd438b TT |
329 | if (ext4_encryption_info(inode) == NULL) |
330 | return -ENOKEY; | |
c9c7429c | 331 | } |
2e9ee850 | 332 | file_accessed(file); |
923ae0ff RZ |
333 | if (IS_DAX(file_inode(file))) { |
334 | vma->vm_ops = &ext4_dax_vm_ops; | |
11bd1a9e | 335 | vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE; |
923ae0ff RZ |
336 | } else { |
337 | vma->vm_ops = &ext4_file_vm_ops; | |
338 | } | |
2e9ee850 AK |
339 | return 0; |
340 | } | |
341 | ||
bc0b0d6d TT |
342 | static int ext4_file_open(struct inode * inode, struct file * filp) |
343 | { | |
344 | struct super_block *sb = inode->i_sb; | |
345 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
346 | struct vfsmount *mnt = filp->f_path.mnt; | |
ff978b09 | 347 | struct inode *dir = filp->f_path.dentry->d_parent->d_inode; |
bc0b0d6d TT |
348 | struct path path; |
349 | char buf[64], *cp; | |
c9c7429c | 350 | int ret; |
bc0b0d6d TT |
351 | |
352 | if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) && | |
353 | !(sb->s_flags & MS_RDONLY))) { | |
354 | sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED; | |
355 | /* | |
356 | * Sample where the filesystem has been mounted and | |
357 | * store it in the superblock for sysadmin convenience | |
358 | * when trying to sort through large numbers of block | |
359 | * devices or filesystem images. | |
360 | */ | |
361 | memset(buf, 0, sizeof(buf)); | |
3899167d AV |
362 | path.mnt = mnt; |
363 | path.dentry = mnt->mnt_root; | |
bc0b0d6d | 364 | cp = d_path(&path, buf, sizeof(buf)); |
bc0b0d6d | 365 | if (!IS_ERR(cp)) { |
044ce47f JK |
366 | handle_t *handle; |
367 | int err; | |
368 | ||
9924a92a | 369 | handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1); |
044ce47f JK |
370 | if (IS_ERR(handle)) |
371 | return PTR_ERR(handle); | |
5d601255 | 372 | BUFFER_TRACE(sbi->s_sbh, "get_write_access"); |
044ce47f JK |
373 | err = ext4_journal_get_write_access(handle, sbi->s_sbh); |
374 | if (err) { | |
375 | ext4_journal_stop(handle); | |
376 | return err; | |
377 | } | |
cf803903 DW |
378 | strlcpy(sbi->s_es->s_last_mounted, cp, |
379 | sizeof(sbi->s_es->s_last_mounted)); | |
044ce47f JK |
380 | ext4_handle_dirty_super(handle, sb); |
381 | ext4_journal_stop(handle); | |
bc0b0d6d TT |
382 | } |
383 | } | |
abdd438b TT |
384 | if (ext4_encrypted_inode(inode)) { |
385 | ret = ext4_get_encryption_info(inode); | |
386 | if (ret) | |
387 | return -EACCES; | |
388 | if (ext4_encryption_info(inode) == NULL) | |
389 | return -ENOKEY; | |
390 | } | |
ff978b09 TT |
391 | if (ext4_encrypted_inode(dir) && |
392 | !ext4_is_child_context_consistent_with_parent(dir, inode)) { | |
393 | ext4_warning(inode->i_sb, | |
394 | "Inconsistent encryption contexts: %lu/%lu\n", | |
395 | (unsigned long) dir->i_ino, | |
396 | (unsigned long) inode->i_ino); | |
397 | return -EPERM; | |
398 | } | |
8aefcd55 TT |
399 | /* |
400 | * Set up the jbd2_inode if we are opening the inode for | |
401 | * writing and the journal is present | |
402 | */ | |
a361293f | 403 | if (filp->f_mode & FMODE_WRITE) { |
c9c7429c | 404 | ret = ext4_inode_attach_jinode(inode); |
a361293f JK |
405 | if (ret < 0) |
406 | return ret; | |
8aefcd55 | 407 | } |
abdd438b | 408 | return dquot_file_open(inode, filp); |
bc0b0d6d TT |
409 | } |
410 | ||
c8c0df24 ZL |
411 | /* |
412 | * Here we use ext4_map_blocks() to get a block mapping for a extent-based | |
413 | * file rather than ext4_ext_walk_space() because we can introduce | |
414 | * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same | |
415 | * function. When extent status tree has been fully implemented, it will | |
416 | * track all extent status for a file and we can directly use it to | |
417 | * retrieve the offset for SEEK_DATA/SEEK_HOLE. | |
418 | */ | |
419 | ||
420 | /* | |
421 | * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to | |
422 | * lookup page cache to check whether or not there has some data between | |
423 | * [startoff, endoff] because, if this range contains an unwritten extent, | |
424 | * we determine this extent as a data or a hole according to whether the | |
425 | * page cache has data or not. | |
426 | */ | |
ad7fefb1 TT |
427 | static int ext4_find_unwritten_pgoff(struct inode *inode, |
428 | int whence, | |
429 | struct ext4_map_blocks *map, | |
430 | loff_t *offset) | |
c8c0df24 ZL |
431 | { |
432 | struct pagevec pvec; | |
ad7fefb1 | 433 | unsigned int blkbits; |
c8c0df24 ZL |
434 | pgoff_t index; |
435 | pgoff_t end; | |
ad7fefb1 | 436 | loff_t endoff; |
c8c0df24 ZL |
437 | loff_t startoff; |
438 | loff_t lastoff; | |
439 | int found = 0; | |
440 | ||
ad7fefb1 | 441 | blkbits = inode->i_sb->s_blocksize_bits; |
c8c0df24 ZL |
442 | startoff = *offset; |
443 | lastoff = startoff; | |
ad7fefb1 | 444 | endoff = (loff_t)(map->m_lblk + map->m_len) << blkbits; |
c8c0df24 ZL |
445 | |
446 | index = startoff >> PAGE_CACHE_SHIFT; | |
447 | end = endoff >> PAGE_CACHE_SHIFT; | |
448 | ||
449 | pagevec_init(&pvec, 0); | |
450 | do { | |
451 | int i, num; | |
452 | unsigned long nr_pages; | |
453 | ||
454 | num = min_t(pgoff_t, end - index, PAGEVEC_SIZE); | |
455 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index, | |
456 | (pgoff_t)num); | |
457 | if (nr_pages == 0) { | |
965c8e59 | 458 | if (whence == SEEK_DATA) |
c8c0df24 ZL |
459 | break; |
460 | ||
965c8e59 | 461 | BUG_ON(whence != SEEK_HOLE); |
c8c0df24 ZL |
462 | /* |
463 | * If this is the first time to go into the loop and | |
464 | * offset is not beyond the end offset, it will be a | |
465 | * hole at this offset | |
466 | */ | |
467 | if (lastoff == startoff || lastoff < endoff) | |
468 | found = 1; | |
469 | break; | |
470 | } | |
471 | ||
472 | /* | |
473 | * If this is the first time to go into the loop and | |
474 | * offset is smaller than the first page offset, it will be a | |
475 | * hole at this offset. | |
476 | */ | |
965c8e59 | 477 | if (lastoff == startoff && whence == SEEK_HOLE && |
c8c0df24 ZL |
478 | lastoff < page_offset(pvec.pages[0])) { |
479 | found = 1; | |
480 | break; | |
481 | } | |
482 | ||
483 | for (i = 0; i < nr_pages; i++) { | |
484 | struct page *page = pvec.pages[i]; | |
485 | struct buffer_head *bh, *head; | |
486 | ||
487 | /* | |
488 | * If the current offset is not beyond the end of given | |
489 | * range, it will be a hole. | |
490 | */ | |
965c8e59 | 491 | if (lastoff < endoff && whence == SEEK_HOLE && |
c8c0df24 ZL |
492 | page->index > end) { |
493 | found = 1; | |
494 | *offset = lastoff; | |
495 | goto out; | |
496 | } | |
497 | ||
498 | lock_page(page); | |
499 | ||
500 | if (unlikely(page->mapping != inode->i_mapping)) { | |
501 | unlock_page(page); | |
502 | continue; | |
503 | } | |
504 | ||
505 | if (!page_has_buffers(page)) { | |
506 | unlock_page(page); | |
507 | continue; | |
508 | } | |
509 | ||
510 | if (page_has_buffers(page)) { | |
511 | lastoff = page_offset(page); | |
512 | bh = head = page_buffers(page); | |
513 | do { | |
514 | if (buffer_uptodate(bh) || | |
515 | buffer_unwritten(bh)) { | |
965c8e59 | 516 | if (whence == SEEK_DATA) |
c8c0df24 ZL |
517 | found = 1; |
518 | } else { | |
965c8e59 | 519 | if (whence == SEEK_HOLE) |
c8c0df24 ZL |
520 | found = 1; |
521 | } | |
522 | if (found) { | |
523 | *offset = max_t(loff_t, | |
524 | startoff, lastoff); | |
525 | unlock_page(page); | |
526 | goto out; | |
527 | } | |
528 | lastoff += bh->b_size; | |
529 | bh = bh->b_this_page; | |
530 | } while (bh != head); | |
531 | } | |
532 | ||
533 | lastoff = page_offset(page) + PAGE_SIZE; | |
534 | unlock_page(page); | |
535 | } | |
536 | ||
537 | /* | |
538 | * The no. of pages is less than our desired, that would be a | |
539 | * hole in there. | |
540 | */ | |
965c8e59 | 541 | if (nr_pages < num && whence == SEEK_HOLE) { |
c8c0df24 ZL |
542 | found = 1; |
543 | *offset = lastoff; | |
544 | break; | |
545 | } | |
546 | ||
547 | index = pvec.pages[i - 1]->index + 1; | |
548 | pagevec_release(&pvec); | |
549 | } while (index <= end); | |
550 | ||
551 | out: | |
552 | pagevec_release(&pvec); | |
553 | return found; | |
554 | } | |
555 | ||
556 | /* | |
557 | * ext4_seek_data() retrieves the offset for SEEK_DATA. | |
558 | */ | |
559 | static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize) | |
560 | { | |
561 | struct inode *inode = file->f_mapping->host; | |
ad7fefb1 TT |
562 | struct ext4_map_blocks map; |
563 | struct extent_status es; | |
564 | ext4_lblk_t start, last, end; | |
565 | loff_t dataoff, isize; | |
566 | int blkbits; | |
567 | int ret = 0; | |
c8c0df24 | 568 | |
5955102c | 569 | inode_lock(inode); |
ad7fefb1 TT |
570 | |
571 | isize = i_size_read(inode); | |
572 | if (offset >= isize) { | |
5955102c | 573 | inode_unlock(inode); |
c8c0df24 ZL |
574 | return -ENXIO; |
575 | } | |
ad7fefb1 TT |
576 | |
577 | blkbits = inode->i_sb->s_blocksize_bits; | |
578 | start = offset >> blkbits; | |
579 | last = start; | |
580 | end = isize >> blkbits; | |
581 | dataoff = offset; | |
582 | ||
583 | do { | |
584 | map.m_lblk = last; | |
585 | map.m_len = end - last + 1; | |
586 | ret = ext4_map_blocks(NULL, inode, &map, 0); | |
587 | if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) { | |
588 | if (last != start) | |
589 | dataoff = (loff_t)last << blkbits; | |
c8c0df24 | 590 | break; |
ad7fefb1 | 591 | } |
c8c0df24 | 592 | |
ad7fefb1 TT |
593 | /* |
594 | * If there is a delay extent at this offset, | |
595 | * it will be as a data. | |
596 | */ | |
597 | ext4_es_find_delayed_extent_range(inode, last, last, &es); | |
598 | if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) { | |
599 | if (last != start) | |
600 | dataoff = (loff_t)last << blkbits; | |
c8c0df24 ZL |
601 | break; |
602 | } | |
603 | ||
ad7fefb1 TT |
604 | /* |
605 | * If there is a unwritten extent at this offset, | |
606 | * it will be as a data or a hole according to page | |
607 | * cache that has data or not. | |
608 | */ | |
609 | if (map.m_flags & EXT4_MAP_UNWRITTEN) { | |
610 | int unwritten; | |
611 | unwritten = ext4_find_unwritten_pgoff(inode, SEEK_DATA, | |
612 | &map, &dataoff); | |
613 | if (unwritten) | |
614 | break; | |
615 | } | |
c8c0df24 | 616 | |
ad7fefb1 TT |
617 | last++; |
618 | dataoff = (loff_t)last << blkbits; | |
619 | } while (last <= end); | |
c8c0df24 | 620 | |
5955102c | 621 | inode_unlock(inode); |
c8c0df24 | 622 | |
ad7fefb1 TT |
623 | if (dataoff > isize) |
624 | return -ENXIO; | |
625 | ||
626 | return vfs_setpos(file, dataoff, maxsize); | |
c8c0df24 ZL |
627 | } |
628 | ||
629 | /* | |
ad7fefb1 | 630 | * ext4_seek_hole() retrieves the offset for SEEK_HOLE. |
c8c0df24 ZL |
631 | */ |
632 | static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize) | |
633 | { | |
634 | struct inode *inode = file->f_mapping->host; | |
ad7fefb1 TT |
635 | struct ext4_map_blocks map; |
636 | struct extent_status es; | |
637 | ext4_lblk_t start, last, end; | |
638 | loff_t holeoff, isize; | |
639 | int blkbits; | |
640 | int ret = 0; | |
c8c0df24 | 641 | |
5955102c | 642 | inode_lock(inode); |
ad7fefb1 TT |
643 | |
644 | isize = i_size_read(inode); | |
645 | if (offset >= isize) { | |
5955102c | 646 | inode_unlock(inode); |
c8c0df24 ZL |
647 | return -ENXIO; |
648 | } | |
649 | ||
ad7fefb1 TT |
650 | blkbits = inode->i_sb->s_blocksize_bits; |
651 | start = offset >> blkbits; | |
652 | last = start; | |
653 | end = isize >> blkbits; | |
654 | holeoff = offset; | |
c8c0df24 | 655 | |
ad7fefb1 TT |
656 | do { |
657 | map.m_lblk = last; | |
658 | map.m_len = end - last + 1; | |
659 | ret = ext4_map_blocks(NULL, inode, &map, 0); | |
660 | if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) { | |
661 | last += ret; | |
662 | holeoff = (loff_t)last << blkbits; | |
663 | continue; | |
664 | } | |
c8c0df24 | 665 | |
ad7fefb1 TT |
666 | /* |
667 | * If there is a delay extent at this offset, | |
668 | * we will skip this extent. | |
669 | */ | |
670 | ext4_es_find_delayed_extent_range(inode, last, last, &es); | |
671 | if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) { | |
672 | last = es.es_lblk + es.es_len; | |
673 | holeoff = (loff_t)last << blkbits; | |
674 | continue; | |
675 | } | |
14516bb7 | 676 | |
ad7fefb1 TT |
677 | /* |
678 | * If there is a unwritten extent at this offset, | |
679 | * it will be as a data or a hole according to page | |
680 | * cache that has data or not. | |
681 | */ | |
682 | if (map.m_flags & EXT4_MAP_UNWRITTEN) { | |
683 | int unwritten; | |
684 | unwritten = ext4_find_unwritten_pgoff(inode, SEEK_HOLE, | |
685 | &map, &holeoff); | |
686 | if (!unwritten) { | |
687 | last += ret; | |
688 | holeoff = (loff_t)last << blkbits; | |
c8c0df24 ZL |
689 | continue; |
690 | } | |
14516bb7 | 691 | } |
ad7fefb1 TT |
692 | |
693 | /* find a hole */ | |
694 | break; | |
695 | } while (last <= end); | |
696 | ||
5955102c | 697 | inode_unlock(inode); |
c8c0df24 | 698 | |
ad7fefb1 TT |
699 | if (holeoff > isize) |
700 | holeoff = isize; | |
701 | ||
702 | return vfs_setpos(file, holeoff, maxsize); | |
c8c0df24 ZL |
703 | } |
704 | ||
e0d10bfa | 705 | /* |
ec7268ce ES |
706 | * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values |
707 | * by calling generic_file_llseek_size() with the appropriate maxbytes | |
708 | * value for each. | |
e0d10bfa | 709 | */ |
965c8e59 | 710 | loff_t ext4_llseek(struct file *file, loff_t offset, int whence) |
e0d10bfa TO |
711 | { |
712 | struct inode *inode = file->f_mapping->host; | |
713 | loff_t maxbytes; | |
714 | ||
715 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) | |
716 | maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes; | |
717 | else | |
718 | maxbytes = inode->i_sb->s_maxbytes; | |
e0d10bfa | 719 | |
965c8e59 | 720 | switch (whence) { |
c8c0df24 ZL |
721 | case SEEK_SET: |
722 | case SEEK_CUR: | |
723 | case SEEK_END: | |
965c8e59 | 724 | return generic_file_llseek_size(file, offset, whence, |
c8c0df24 ZL |
725 | maxbytes, i_size_read(inode)); |
726 | case SEEK_DATA: | |
727 | return ext4_seek_data(file, offset, maxbytes); | |
728 | case SEEK_HOLE: | |
729 | return ext4_seek_hole(file, offset, maxbytes); | |
730 | } | |
731 | ||
732 | return -EINVAL; | |
e0d10bfa TO |
733 | } |
734 | ||
617ba13b | 735 | const struct file_operations ext4_file_operations = { |
e0d10bfa | 736 | .llseek = ext4_llseek, |
aad4f8bb | 737 | .read_iter = generic_file_read_iter, |
9b884164 | 738 | .write_iter = ext4_file_write_iter, |
5cdd7b2d | 739 | .unlocked_ioctl = ext4_ioctl, |
ac27a0ec | 740 | #ifdef CONFIG_COMPAT |
617ba13b | 741 | .compat_ioctl = ext4_compat_ioctl, |
ac27a0ec | 742 | #endif |
2e9ee850 | 743 | .mmap = ext4_file_mmap, |
bc0b0d6d | 744 | .open = ext4_file_open, |
617ba13b MC |
745 | .release = ext4_release_file, |
746 | .fsync = ext4_sync_file, | |
ac27a0ec | 747 | .splice_read = generic_file_splice_read, |
8d020765 | 748 | .splice_write = iter_file_splice_write, |
2fe17c10 | 749 | .fallocate = ext4_fallocate, |
ac27a0ec DK |
750 | }; |
751 | ||
754661f1 | 752 | const struct inode_operations ext4_file_inode_operations = { |
617ba13b | 753 | .setattr = ext4_setattr, |
3e3398a0 | 754 | .getattr = ext4_getattr, |
ac27a0ec DK |
755 | .setxattr = generic_setxattr, |
756 | .getxattr = generic_getxattr, | |
617ba13b | 757 | .listxattr = ext4_listxattr, |
ac27a0ec | 758 | .removexattr = generic_removexattr, |
4e34e719 | 759 | .get_acl = ext4_get_acl, |
64e178a7 | 760 | .set_acl = ext4_set_acl, |
6873fa0d | 761 | .fiemap = ext4_fiemap, |
ac27a0ec DK |
762 | }; |
763 |