]>
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 | ||
364443cb JK |
34 | #ifdef CONFIG_FS_DAX |
35 | static ssize_t ext4_dax_read_iter(struct kiocb *iocb, struct iov_iter *to) | |
36 | { | |
37 | struct inode *inode = file_inode(iocb->ki_filp); | |
38 | ssize_t ret; | |
39 | ||
40 | inode_lock_shared(inode); | |
41 | /* | |
42 | * Recheck under inode lock - at this point we are sure it cannot | |
43 | * change anymore | |
44 | */ | |
45 | if (!IS_DAX(inode)) { | |
46 | inode_unlock_shared(inode); | |
47 | /* Fallback to buffered IO in case we cannot support DAX */ | |
48 | return generic_file_read_iter(iocb, to); | |
49 | } | |
50 | ret = dax_iomap_rw(iocb, to, &ext4_iomap_ops); | |
51 | inode_unlock_shared(inode); | |
52 | ||
53 | file_accessed(iocb->ki_filp); | |
54 | return ret; | |
55 | } | |
56 | #endif | |
57 | ||
58 | static ssize_t ext4_file_read_iter(struct kiocb *iocb, struct iov_iter *to) | |
59 | { | |
0db1ff22 TT |
60 | if (unlikely(ext4_forced_shutdown(EXT4_SB(file_inode(iocb->ki_filp)->i_sb)))) |
61 | return -EIO; | |
62 | ||
364443cb JK |
63 | if (!iov_iter_count(to)) |
64 | return 0; /* skip atime */ | |
65 | ||
66 | #ifdef CONFIG_FS_DAX | |
67 | if (IS_DAX(file_inode(iocb->ki_filp))) | |
68 | return ext4_dax_read_iter(iocb, to); | |
69 | #endif | |
70 | return generic_file_read_iter(iocb, to); | |
71 | } | |
72 | ||
ac27a0ec DK |
73 | /* |
74 | * Called when an inode is released. Note that this is different | |
617ba13b | 75 | * from ext4_file_open: open gets called at every open, but release |
ac27a0ec DK |
76 | * gets called only when /all/ the files are closed. |
77 | */ | |
af5bc92d | 78 | static int ext4_release_file(struct inode *inode, struct file *filp) |
ac27a0ec | 79 | { |
19f5fb7a | 80 | if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) { |
7d8f9f7d | 81 | ext4_alloc_da_blocks(inode); |
19f5fb7a | 82 | ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 83 | } |
ac27a0ec DK |
84 | /* if we are the last writer on the inode, drop the block reservation */ |
85 | if ((filp->f_mode & FMODE_WRITE) && | |
d6014301 AK |
86 | (atomic_read(&inode->i_writecount) == 1) && |
87 | !EXT4_I(inode)->i_reserved_data_blocks) | |
ac27a0ec | 88 | { |
0e855ac8 | 89 | down_write(&EXT4_I(inode)->i_data_sem); |
c2ea3fde | 90 | ext4_discard_preallocations(inode); |
0e855ac8 | 91 | up_write(&EXT4_I(inode)->i_data_sem); |
ac27a0ec DK |
92 | } |
93 | if (is_dx(inode) && filp->private_data) | |
617ba13b | 94 | ext4_htree_free_dir_info(filp->private_data); |
ac27a0ec DK |
95 | |
96 | return 0; | |
97 | } | |
98 | ||
c197855e | 99 | static void ext4_unwritten_wait(struct inode *inode) |
e9e3bcec ES |
100 | { |
101 | wait_queue_head_t *wq = ext4_ioend_wq(inode); | |
102 | ||
e27f41e1 | 103 | wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_unwritten) == 0)); |
e9e3bcec ES |
104 | } |
105 | ||
106 | /* | |
107 | * This tests whether the IO in question is block-aligned or not. | |
108 | * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they | |
109 | * are converted to written only after the IO is complete. Until they are | |
110 | * mapped, these blocks appear as holes, so dio_zero_block() will assume that | |
111 | * it needs to zero out portions of the start and/or end block. If 2 AIO | |
112 | * threads are at work on the same unwritten block, they must be synchronized | |
113 | * or one thread will zero the other's data, causing corruption. | |
114 | */ | |
115 | static int | |
9b884164 | 116 | ext4_unaligned_aio(struct inode *inode, struct iov_iter *from, loff_t pos) |
e9e3bcec ES |
117 | { |
118 | struct super_block *sb = inode->i_sb; | |
119 | int blockmask = sb->s_blocksize - 1; | |
e9e3bcec | 120 | |
6e6358fc | 121 | if (pos >= i_size_read(inode)) |
e9e3bcec ES |
122 | return 0; |
123 | ||
9b884164 | 124 | if ((pos | iov_iter_alignment(from)) & blockmask) |
e9e3bcec ES |
125 | return 1; |
126 | ||
127 | return 0; | |
128 | } | |
129 | ||
213bcd9c JK |
130 | /* Is IO overwriting allocated and initialized blocks? */ |
131 | static bool ext4_overwrite_io(struct inode *inode, loff_t pos, loff_t len) | |
132 | { | |
133 | struct ext4_map_blocks map; | |
134 | unsigned int blkbits = inode->i_blkbits; | |
135 | int err, blklen; | |
136 | ||
137 | if (pos + len > i_size_read(inode)) | |
138 | return false; | |
139 | ||
140 | map.m_lblk = pos >> blkbits; | |
141 | map.m_len = EXT4_MAX_BLOCKS(len, pos, blkbits); | |
142 | blklen = map.m_len; | |
143 | ||
144 | err = ext4_map_blocks(NULL, inode, &map, 0); | |
145 | /* | |
146 | * 'err==len' means that all of the blocks have been preallocated, | |
147 | * regardless of whether they have been initialized or not. To exclude | |
148 | * unwritten extents, we need to check m_flags. | |
149 | */ | |
150 | return err == blklen && (map.m_flags & EXT4_MAP_MAPPED); | |
151 | } | |
152 | ||
153 | static ssize_t ext4_write_checks(struct kiocb *iocb, struct iov_iter *from) | |
154 | { | |
155 | struct inode *inode = file_inode(iocb->ki_filp); | |
156 | ssize_t ret; | |
157 | ||
158 | ret = generic_write_checks(iocb, from); | |
159 | if (ret <= 0) | |
160 | return ret; | |
161 | /* | |
162 | * If we have encountered a bitmap-format file, the size limit | |
163 | * is smaller than s_maxbytes, which is for extent-mapped files. | |
164 | */ | |
165 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { | |
166 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
167 | ||
168 | if (iocb->ki_pos >= sbi->s_bitmap_maxbytes) | |
169 | return -EFBIG; | |
170 | iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos); | |
171 | } | |
172 | return iov_iter_count(from); | |
173 | } | |
174 | ||
776722e8 JK |
175 | #ifdef CONFIG_FS_DAX |
176 | static ssize_t | |
177 | ext4_dax_write_iter(struct kiocb *iocb, struct iov_iter *from) | |
178 | { | |
179 | struct inode *inode = file_inode(iocb->ki_filp); | |
180 | ssize_t ret; | |
776722e8 JK |
181 | |
182 | inode_lock(inode); | |
183 | ret = ext4_write_checks(iocb, from); | |
184 | if (ret <= 0) | |
185 | goto out; | |
186 | ret = file_remove_privs(iocb->ki_filp); | |
187 | if (ret) | |
188 | goto out; | |
189 | ret = file_update_time(iocb->ki_filp); | |
190 | if (ret) | |
191 | goto out; | |
192 | ||
776722e8 JK |
193 | ret = dax_iomap_rw(iocb, from, &ext4_iomap_ops); |
194 | out: | |
ff5462e3 | 195 | inode_unlock(inode); |
776722e8 JK |
196 | if (ret > 0) |
197 | ret = generic_write_sync(iocb, ret); | |
198 | return ret; | |
199 | } | |
200 | #endif | |
201 | ||
ac27a0ec | 202 | static ssize_t |
9b884164 | 203 | ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
ac27a0ec | 204 | { |
8ad2850f | 205 | struct inode *inode = file_inode(iocb->ki_filp); |
2ba48ce5 | 206 | int o_direct = iocb->ki_flags & IOCB_DIRECT; |
e142d052 | 207 | int unaligned_aio = 0; |
4bd809db | 208 | int overwrite = 0; |
8563000d | 209 | ssize_t ret; |
7608e610 | 210 | |
0db1ff22 TT |
211 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
212 | return -EIO; | |
213 | ||
776722e8 JK |
214 | #ifdef CONFIG_FS_DAX |
215 | if (IS_DAX(inode)) | |
216 | return ext4_dax_write_iter(iocb, from); | |
217 | #endif | |
218 | ||
e142d052 | 219 | inode_lock(inode); |
213bcd9c | 220 | ret = ext4_write_checks(iocb, from); |
e142d052 JK |
221 | if (ret <= 0) |
222 | goto out; | |
223 | ||
f5ccfe1d | 224 | /* |
e142d052 JK |
225 | * Unaligned direct AIO must be serialized among each other as zeroing |
226 | * of partial blocks of two competing unaligned AIOs can result in data | |
227 | * corruption. | |
f5ccfe1d | 228 | */ |
e142d052 | 229 | if (o_direct && ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) && |
f5ccfe1d | 230 | !is_sync_kiocb(iocb) && |
e142d052 JK |
231 | ext4_unaligned_aio(inode, from, iocb->ki_pos)) { |
232 | unaligned_aio = 1; | |
f5ccfe1d TT |
233 | ext4_unwritten_wait(inode); |
234 | } | |
235 | ||
a41537e6 | 236 | iocb->private = &overwrite; |
213bcd9c JK |
237 | /* Check whether we do a DIO overwrite or not */ |
238 | if (o_direct && ext4_should_dioread_nolock(inode) && !unaligned_aio && | |
239 | ext4_overwrite_io(inode, iocb->ki_pos, iov_iter_count(from))) | |
240 | overwrite = 1; | |
7608e610 | 241 | |
9b884164 | 242 | ret = __generic_file_write_iter(iocb, from); |
5955102c | 243 | inode_unlock(inode); |
7608e610 | 244 | |
e2592217 CH |
245 | if (ret > 0) |
246 | ret = generic_write_sync(iocb, ret); | |
e9e3bcec | 247 | |
e768d7ff AV |
248 | return ret; |
249 | ||
250 | out: | |
5955102c | 251 | inode_unlock(inode); |
e9e3bcec | 252 | return ret; |
ac27a0ec DK |
253 | } |
254 | ||
923ae0ff RZ |
255 | #ifdef CONFIG_FS_DAX |
256 | static int ext4_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf) | |
257 | { | |
01a33b4a | 258 | int result; |
ea3d7209 JK |
259 | struct inode *inode = file_inode(vma->vm_file); |
260 | struct super_block *sb = inode->i_sb; | |
01a33b4a MW |
261 | bool write = vmf->flags & FAULT_FLAG_WRITE; |
262 | ||
263 | if (write) { | |
264 | sb_start_pagefault(sb); | |
265 | file_update_time(vma->vm_file); | |
1db17542 JK |
266 | } |
267 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
268 | result = dax_iomap_fault(vma, vmf, &ext4_iomap_ops); | |
269 | up_read(&EXT4_I(inode)->i_mmap_sem); | |
270 | if (write) | |
01a33b4a | 271 | sb_end_pagefault(sb); |
01a33b4a MW |
272 | |
273 | return result; | |
923ae0ff RZ |
274 | } |
275 | ||
d8a849e1 | 276 | static int |
f4200391 | 277 | ext4_dax_pmd_fault(struct vm_fault *vmf) |
11bd1a9e | 278 | { |
01a33b4a | 279 | int result; |
f4200391 | 280 | struct inode *inode = file_inode(vmf->vma->vm_file); |
01a33b4a | 281 | struct super_block *sb = inode->i_sb; |
d8a849e1 | 282 | bool write = vmf->flags & FAULT_FLAG_WRITE; |
01a33b4a MW |
283 | |
284 | if (write) { | |
285 | sb_start_pagefault(sb); | |
f4200391 | 286 | file_update_time(vmf->vma->vm_file); |
e2ae766c | 287 | } |
1db17542 | 288 | down_read(&EXT4_I(inode)->i_mmap_sem); |
f4200391 | 289 | result = dax_iomap_pmd_fault(vmf, &ext4_iomap_ops); |
1db17542 JK |
290 | up_read(&EXT4_I(inode)->i_mmap_sem); |
291 | if (write) | |
01a33b4a | 292 | sb_end_pagefault(sb); |
01a33b4a MW |
293 | |
294 | return result; | |
11bd1a9e MW |
295 | } |
296 | ||
ea3d7209 | 297 | /* |
1e9d180b | 298 | * Handle write fault for VM_MIXEDMAP mappings. Similarly to ext4_dax_fault() |
ea3d7209 JK |
299 | * handler we check for races agaist truncate. Note that since we cycle through |
300 | * i_mmap_sem, we are sure that also any hole punching that began before we | |
301 | * were called is finished by now and so if it included part of the file we | |
302 | * are working on, our pte will get unmapped and the check for pte_same() in | |
303 | * wp_pfn_shared() fails. Thus fault gets retried and things work out as | |
304 | * desired. | |
305 | */ | |
306 | static int ext4_dax_pfn_mkwrite(struct vm_area_struct *vma, | |
307 | struct vm_fault *vmf) | |
308 | { | |
309 | struct inode *inode = file_inode(vma->vm_file); | |
310 | struct super_block *sb = inode->i_sb; | |
ea3d7209 | 311 | loff_t size; |
d5be7a03 | 312 | int ret; |
ea3d7209 JK |
313 | |
314 | sb_start_pagefault(sb); | |
315 | file_update_time(vma->vm_file); | |
316 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
317 | size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
318 | if (vmf->pgoff >= size) | |
319 | ret = VM_FAULT_SIGBUS; | |
d5be7a03 RZ |
320 | else |
321 | ret = dax_pfn_mkwrite(vma, vmf); | |
ea3d7209 JK |
322 | up_read(&EXT4_I(inode)->i_mmap_sem); |
323 | sb_end_pagefault(sb); | |
324 | ||
325 | return ret; | |
923ae0ff RZ |
326 | } |
327 | ||
328 | static const struct vm_operations_struct ext4_dax_vm_ops = { | |
329 | .fault = ext4_dax_fault, | |
11bd1a9e | 330 | .pmd_fault = ext4_dax_pmd_fault, |
1e9d180b | 331 | .page_mkwrite = ext4_dax_fault, |
ea3d7209 | 332 | .pfn_mkwrite = ext4_dax_pfn_mkwrite, |
923ae0ff RZ |
333 | }; |
334 | #else | |
335 | #define ext4_dax_vm_ops ext4_file_vm_ops | |
336 | #endif | |
337 | ||
f0f37e2f | 338 | static const struct vm_operations_struct ext4_file_vm_ops = { |
ea3d7209 | 339 | .fault = ext4_filemap_fault, |
f1820361 | 340 | .map_pages = filemap_map_pages, |
2e9ee850 AK |
341 | .page_mkwrite = ext4_page_mkwrite, |
342 | }; | |
343 | ||
344 | static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma) | |
345 | { | |
c9c7429c MH |
346 | struct inode *inode = file->f_mapping->host; |
347 | ||
0db1ff22 TT |
348 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
349 | return -EIO; | |
350 | ||
c9c7429c | 351 | if (ext4_encrypted_inode(inode)) { |
a7550b30 | 352 | int err = fscrypt_get_encryption_info(inode); |
c9c7429c MH |
353 | if (err) |
354 | return 0; | |
a7550b30 | 355 | if (!fscrypt_has_encryption_key(inode)) |
abdd438b | 356 | return -ENOKEY; |
c9c7429c | 357 | } |
2e9ee850 | 358 | file_accessed(file); |
923ae0ff RZ |
359 | if (IS_DAX(file_inode(file))) { |
360 | vma->vm_ops = &ext4_dax_vm_ops; | |
11bd1a9e | 361 | vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE; |
923ae0ff RZ |
362 | } else { |
363 | vma->vm_ops = &ext4_file_vm_ops; | |
364 | } | |
2e9ee850 AK |
365 | return 0; |
366 | } | |
367 | ||
bc0b0d6d TT |
368 | static int ext4_file_open(struct inode * inode, struct file * filp) |
369 | { | |
370 | struct super_block *sb = inode->i_sb; | |
371 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
372 | struct vfsmount *mnt = filp->f_path.mnt; | |
9dd78d8c | 373 | struct dentry *dir; |
bc0b0d6d TT |
374 | struct path path; |
375 | char buf[64], *cp; | |
c9c7429c | 376 | int ret; |
bc0b0d6d | 377 | |
0db1ff22 TT |
378 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
379 | return -EIO; | |
380 | ||
bc0b0d6d TT |
381 | if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) && |
382 | !(sb->s_flags & MS_RDONLY))) { | |
383 | sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED; | |
384 | /* | |
385 | * Sample where the filesystem has been mounted and | |
386 | * store it in the superblock for sysadmin convenience | |
387 | * when trying to sort through large numbers of block | |
388 | * devices or filesystem images. | |
389 | */ | |
390 | memset(buf, 0, sizeof(buf)); | |
3899167d AV |
391 | path.mnt = mnt; |
392 | path.dentry = mnt->mnt_root; | |
bc0b0d6d | 393 | cp = d_path(&path, buf, sizeof(buf)); |
bc0b0d6d | 394 | if (!IS_ERR(cp)) { |
044ce47f JK |
395 | handle_t *handle; |
396 | int err; | |
397 | ||
9924a92a | 398 | handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1); |
044ce47f JK |
399 | if (IS_ERR(handle)) |
400 | return PTR_ERR(handle); | |
5d601255 | 401 | BUFFER_TRACE(sbi->s_sbh, "get_write_access"); |
044ce47f JK |
402 | err = ext4_journal_get_write_access(handle, sbi->s_sbh); |
403 | if (err) { | |
404 | ext4_journal_stop(handle); | |
405 | return err; | |
406 | } | |
cf803903 DW |
407 | strlcpy(sbi->s_es->s_last_mounted, cp, |
408 | sizeof(sbi->s_es->s_last_mounted)); | |
044ce47f JK |
409 | ext4_handle_dirty_super(handle, sb); |
410 | ext4_journal_stop(handle); | |
bc0b0d6d TT |
411 | } |
412 | } | |
abdd438b | 413 | if (ext4_encrypted_inode(inode)) { |
a7550b30 | 414 | ret = fscrypt_get_encryption_info(inode); |
abdd438b TT |
415 | if (ret) |
416 | return -EACCES; | |
a7550b30 | 417 | if (!fscrypt_has_encryption_key(inode)) |
abdd438b TT |
418 | return -ENOKEY; |
419 | } | |
9dd78d8c | 420 | |
c0a37d48 | 421 | dir = dget_parent(file_dentry(filp)); |
9dd78d8c | 422 | if (ext4_encrypted_inode(d_inode(dir)) && |
a7550b30 | 423 | !fscrypt_has_permitted_context(d_inode(dir), inode)) { |
ff978b09 | 424 | ext4_warning(inode->i_sb, |
8d2ae1cb | 425 | "Inconsistent encryption contexts: %lu/%lu", |
9dd78d8c | 426 | (unsigned long) d_inode(dir)->i_ino, |
ff978b09 | 427 | (unsigned long) inode->i_ino); |
9dd78d8c | 428 | dput(dir); |
ff978b09 TT |
429 | return -EPERM; |
430 | } | |
9dd78d8c | 431 | dput(dir); |
8aefcd55 TT |
432 | /* |
433 | * Set up the jbd2_inode if we are opening the inode for | |
434 | * writing and the journal is present | |
435 | */ | |
a361293f | 436 | if (filp->f_mode & FMODE_WRITE) { |
c9c7429c | 437 | ret = ext4_inode_attach_jinode(inode); |
a361293f JK |
438 | if (ret < 0) |
439 | return ret; | |
8aefcd55 | 440 | } |
abdd438b | 441 | return dquot_file_open(inode, filp); |
bc0b0d6d TT |
442 | } |
443 | ||
c8c0df24 ZL |
444 | /* |
445 | * Here we use ext4_map_blocks() to get a block mapping for a extent-based | |
446 | * file rather than ext4_ext_walk_space() because we can introduce | |
447 | * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same | |
448 | * function. When extent status tree has been fully implemented, it will | |
449 | * track all extent status for a file and we can directly use it to | |
450 | * retrieve the offset for SEEK_DATA/SEEK_HOLE. | |
451 | */ | |
452 | ||
453 | /* | |
454 | * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to | |
455 | * lookup page cache to check whether or not there has some data between | |
456 | * [startoff, endoff] because, if this range contains an unwritten extent, | |
457 | * we determine this extent as a data or a hole according to whether the | |
458 | * page cache has data or not. | |
459 | */ | |
ad7fefb1 TT |
460 | static int ext4_find_unwritten_pgoff(struct inode *inode, |
461 | int whence, | |
2d90c160 | 462 | ext4_lblk_t end_blk, |
ad7fefb1 | 463 | loff_t *offset) |
c8c0df24 ZL |
464 | { |
465 | struct pagevec pvec; | |
ad7fefb1 | 466 | unsigned int blkbits; |
c8c0df24 ZL |
467 | pgoff_t index; |
468 | pgoff_t end; | |
ad7fefb1 | 469 | loff_t endoff; |
c8c0df24 ZL |
470 | loff_t startoff; |
471 | loff_t lastoff; | |
472 | int found = 0; | |
473 | ||
ad7fefb1 | 474 | blkbits = inode->i_sb->s_blocksize_bits; |
c8c0df24 ZL |
475 | startoff = *offset; |
476 | lastoff = startoff; | |
2d90c160 | 477 | endoff = (loff_t)end_blk << blkbits; |
c8c0df24 | 478 | |
09cbfeaf KS |
479 | index = startoff >> PAGE_SHIFT; |
480 | end = endoff >> PAGE_SHIFT; | |
c8c0df24 ZL |
481 | |
482 | pagevec_init(&pvec, 0); | |
483 | do { | |
484 | int i, num; | |
485 | unsigned long nr_pages; | |
486 | ||
487 | num = min_t(pgoff_t, end - index, PAGEVEC_SIZE); | |
488 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index, | |
489 | (pgoff_t)num); | |
490 | if (nr_pages == 0) { | |
965c8e59 | 491 | if (whence == SEEK_DATA) |
c8c0df24 ZL |
492 | break; |
493 | ||
965c8e59 | 494 | BUG_ON(whence != SEEK_HOLE); |
c8c0df24 ZL |
495 | /* |
496 | * If this is the first time to go into the loop and | |
497 | * offset is not beyond the end offset, it will be a | |
498 | * hole at this offset | |
499 | */ | |
500 | if (lastoff == startoff || lastoff < endoff) | |
501 | found = 1; | |
502 | break; | |
503 | } | |
504 | ||
505 | /* | |
506 | * If this is the first time to go into the loop and | |
507 | * offset is smaller than the first page offset, it will be a | |
508 | * hole at this offset. | |
509 | */ | |
965c8e59 | 510 | if (lastoff == startoff && whence == SEEK_HOLE && |
c8c0df24 ZL |
511 | lastoff < page_offset(pvec.pages[0])) { |
512 | found = 1; | |
513 | break; | |
514 | } | |
515 | ||
516 | for (i = 0; i < nr_pages; i++) { | |
517 | struct page *page = pvec.pages[i]; | |
518 | struct buffer_head *bh, *head; | |
519 | ||
520 | /* | |
521 | * If the current offset is not beyond the end of given | |
522 | * range, it will be a hole. | |
523 | */ | |
965c8e59 | 524 | if (lastoff < endoff && whence == SEEK_HOLE && |
c8c0df24 ZL |
525 | page->index > end) { |
526 | found = 1; | |
527 | *offset = lastoff; | |
528 | goto out; | |
529 | } | |
530 | ||
531 | lock_page(page); | |
532 | ||
533 | if (unlikely(page->mapping != inode->i_mapping)) { | |
534 | unlock_page(page); | |
535 | continue; | |
536 | } | |
537 | ||
538 | if (!page_has_buffers(page)) { | |
539 | unlock_page(page); | |
540 | continue; | |
541 | } | |
542 | ||
543 | if (page_has_buffers(page)) { | |
544 | lastoff = page_offset(page); | |
545 | bh = head = page_buffers(page); | |
546 | do { | |
547 | if (buffer_uptodate(bh) || | |
548 | buffer_unwritten(bh)) { | |
965c8e59 | 549 | if (whence == SEEK_DATA) |
c8c0df24 ZL |
550 | found = 1; |
551 | } else { | |
965c8e59 | 552 | if (whence == SEEK_HOLE) |
c8c0df24 ZL |
553 | found = 1; |
554 | } | |
555 | if (found) { | |
556 | *offset = max_t(loff_t, | |
557 | startoff, lastoff); | |
558 | unlock_page(page); | |
559 | goto out; | |
560 | } | |
561 | lastoff += bh->b_size; | |
562 | bh = bh->b_this_page; | |
563 | } while (bh != head); | |
564 | } | |
565 | ||
566 | lastoff = page_offset(page) + PAGE_SIZE; | |
567 | unlock_page(page); | |
568 | } | |
569 | ||
570 | /* | |
571 | * The no. of pages is less than our desired, that would be a | |
572 | * hole in there. | |
573 | */ | |
965c8e59 | 574 | if (nr_pages < num && whence == SEEK_HOLE) { |
c8c0df24 ZL |
575 | found = 1; |
576 | *offset = lastoff; | |
577 | break; | |
578 | } | |
579 | ||
580 | index = pvec.pages[i - 1]->index + 1; | |
581 | pagevec_release(&pvec); | |
582 | } while (index <= end); | |
583 | ||
584 | out: | |
585 | pagevec_release(&pvec); | |
586 | return found; | |
587 | } | |
588 | ||
589 | /* | |
590 | * ext4_seek_data() retrieves the offset for SEEK_DATA. | |
591 | */ | |
592 | static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize) | |
593 | { | |
594 | struct inode *inode = file->f_mapping->host; | |
ad7fefb1 TT |
595 | struct extent_status es; |
596 | ext4_lblk_t start, last, end; | |
597 | loff_t dataoff, isize; | |
598 | int blkbits; | |
2d90c160 | 599 | int ret; |
c8c0df24 | 600 | |
5955102c | 601 | inode_lock(inode); |
ad7fefb1 TT |
602 | |
603 | isize = i_size_read(inode); | |
604 | if (offset >= isize) { | |
5955102c | 605 | inode_unlock(inode); |
c8c0df24 ZL |
606 | return -ENXIO; |
607 | } | |
ad7fefb1 TT |
608 | |
609 | blkbits = inode->i_sb->s_blocksize_bits; | |
610 | start = offset >> blkbits; | |
611 | last = start; | |
612 | end = isize >> blkbits; | |
613 | dataoff = offset; | |
614 | ||
615 | do { | |
2d90c160 JK |
616 | ret = ext4_get_next_extent(inode, last, end - last + 1, &es); |
617 | if (ret <= 0) { | |
618 | /* No extent found -> no data */ | |
619 | if (ret == 0) | |
620 | ret = -ENXIO; | |
621 | inode_unlock(inode); | |
622 | return ret; | |
ad7fefb1 | 623 | } |
c8c0df24 | 624 | |
2d90c160 JK |
625 | last = es.es_lblk; |
626 | if (last != start) | |
627 | dataoff = (loff_t)last << blkbits; | |
628 | if (!ext4_es_is_unwritten(&es)) | |
c8c0df24 | 629 | break; |
c8c0df24 | 630 | |
ad7fefb1 TT |
631 | /* |
632 | * If there is a unwritten extent at this offset, | |
633 | * it will be as a data or a hole according to page | |
634 | * cache that has data or not. | |
635 | */ | |
2d90c160 JK |
636 | if (ext4_find_unwritten_pgoff(inode, SEEK_DATA, |
637 | es.es_lblk + es.es_len, &dataoff)) | |
638 | break; | |
639 | last += es.es_len; | |
ad7fefb1 | 640 | dataoff = (loff_t)last << blkbits; |
2d90c160 | 641 | cond_resched(); |
ad7fefb1 | 642 | } while (last <= end); |
c8c0df24 | 643 | |
5955102c | 644 | inode_unlock(inode); |
c8c0df24 | 645 | |
ad7fefb1 TT |
646 | if (dataoff > isize) |
647 | return -ENXIO; | |
648 | ||
649 | return vfs_setpos(file, dataoff, maxsize); | |
c8c0df24 ZL |
650 | } |
651 | ||
652 | /* | |
ad7fefb1 | 653 | * ext4_seek_hole() retrieves the offset for SEEK_HOLE. |
c8c0df24 ZL |
654 | */ |
655 | static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize) | |
656 | { | |
657 | struct inode *inode = file->f_mapping->host; | |
ad7fefb1 TT |
658 | struct extent_status es; |
659 | ext4_lblk_t start, last, end; | |
660 | loff_t holeoff, isize; | |
661 | int blkbits; | |
2d90c160 | 662 | int ret; |
c8c0df24 | 663 | |
5955102c | 664 | inode_lock(inode); |
ad7fefb1 TT |
665 | |
666 | isize = i_size_read(inode); | |
667 | if (offset >= isize) { | |
5955102c | 668 | inode_unlock(inode); |
c8c0df24 ZL |
669 | return -ENXIO; |
670 | } | |
671 | ||
ad7fefb1 TT |
672 | blkbits = inode->i_sb->s_blocksize_bits; |
673 | start = offset >> blkbits; | |
674 | last = start; | |
675 | end = isize >> blkbits; | |
676 | holeoff = offset; | |
c8c0df24 | 677 | |
ad7fefb1 | 678 | do { |
2d90c160 JK |
679 | ret = ext4_get_next_extent(inode, last, end - last + 1, &es); |
680 | if (ret < 0) { | |
681 | inode_unlock(inode); | |
682 | return ret; | |
ad7fefb1 | 683 | } |
2d90c160 JK |
684 | /* Found a hole? */ |
685 | if (ret == 0 || es.es_lblk > last) { | |
686 | if (last != start) | |
687 | holeoff = (loff_t)last << blkbits; | |
688 | break; | |
ad7fefb1 | 689 | } |
ad7fefb1 TT |
690 | /* |
691 | * If there is a unwritten extent at this offset, | |
692 | * it will be as a data or a hole according to page | |
693 | * cache that has data or not. | |
694 | */ | |
2d90c160 JK |
695 | if (ext4_es_is_unwritten(&es) && |
696 | ext4_find_unwritten_pgoff(inode, SEEK_HOLE, | |
697 | last + es.es_len, &holeoff)) | |
698 | break; | |
ad7fefb1 | 699 | |
2d90c160 JK |
700 | last += es.es_len; |
701 | holeoff = (loff_t)last << blkbits; | |
702 | cond_resched(); | |
ad7fefb1 TT |
703 | } while (last <= end); |
704 | ||
5955102c | 705 | inode_unlock(inode); |
c8c0df24 | 706 | |
ad7fefb1 TT |
707 | if (holeoff > isize) |
708 | holeoff = isize; | |
709 | ||
710 | return vfs_setpos(file, holeoff, maxsize); | |
c8c0df24 ZL |
711 | } |
712 | ||
e0d10bfa | 713 | /* |
ec7268ce ES |
714 | * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values |
715 | * by calling generic_file_llseek_size() with the appropriate maxbytes | |
716 | * value for each. | |
e0d10bfa | 717 | */ |
965c8e59 | 718 | loff_t ext4_llseek(struct file *file, loff_t offset, int whence) |
e0d10bfa TO |
719 | { |
720 | struct inode *inode = file->f_mapping->host; | |
721 | loff_t maxbytes; | |
722 | ||
723 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) | |
724 | maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes; | |
725 | else | |
726 | maxbytes = inode->i_sb->s_maxbytes; | |
e0d10bfa | 727 | |
965c8e59 | 728 | switch (whence) { |
c8c0df24 ZL |
729 | case SEEK_SET: |
730 | case SEEK_CUR: | |
731 | case SEEK_END: | |
965c8e59 | 732 | return generic_file_llseek_size(file, offset, whence, |
c8c0df24 ZL |
733 | maxbytes, i_size_read(inode)); |
734 | case SEEK_DATA: | |
735 | return ext4_seek_data(file, offset, maxbytes); | |
736 | case SEEK_HOLE: | |
737 | return ext4_seek_hole(file, offset, maxbytes); | |
738 | } | |
739 | ||
740 | return -EINVAL; | |
e0d10bfa TO |
741 | } |
742 | ||
617ba13b | 743 | const struct file_operations ext4_file_operations = { |
e0d10bfa | 744 | .llseek = ext4_llseek, |
364443cb | 745 | .read_iter = ext4_file_read_iter, |
9b884164 | 746 | .write_iter = ext4_file_write_iter, |
5cdd7b2d | 747 | .unlocked_ioctl = ext4_ioctl, |
ac27a0ec | 748 | #ifdef CONFIG_COMPAT |
617ba13b | 749 | .compat_ioctl = ext4_compat_ioctl, |
ac27a0ec | 750 | #endif |
2e9ee850 | 751 | .mmap = ext4_file_mmap, |
bc0b0d6d | 752 | .open = ext4_file_open, |
617ba13b MC |
753 | .release = ext4_release_file, |
754 | .fsync = ext4_sync_file, | |
dbe6ec81 | 755 | .get_unmapped_area = thp_get_unmapped_area, |
ac27a0ec | 756 | .splice_read = generic_file_splice_read, |
8d020765 | 757 | .splice_write = iter_file_splice_write, |
2fe17c10 | 758 | .fallocate = ext4_fallocate, |
ac27a0ec DK |
759 | }; |
760 | ||
754661f1 | 761 | const struct inode_operations ext4_file_inode_operations = { |
617ba13b | 762 | .setattr = ext4_setattr, |
3e3398a0 | 763 | .getattr = ext4_getattr, |
617ba13b | 764 | .listxattr = ext4_listxattr, |
4e34e719 | 765 | .get_acl = ext4_get_acl, |
64e178a7 | 766 | .set_acl = ext4_set_acl, |
6873fa0d | 767 | .fiemap = ext4_fiemap, |
ac27a0ec DK |
768 | }; |
769 |