]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/read_write.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992 Linus Torvalds | |
5 | */ | |
6 | ||
7 | #include <linux/slab.h> | |
8 | #include <linux/stat.h> | |
9 | #include <linux/fcntl.h> | |
10 | #include <linux/file.h> | |
11 | #include <linux/uio.h> | |
0eeca283 | 12 | #include <linux/fsnotify.h> |
1da177e4 | 13 | #include <linux/security.h> |
630d9c47 | 14 | #include <linux/export.h> |
1da177e4 | 15 | #include <linux/syscalls.h> |
e28cc715 | 16 | #include <linux/pagemap.h> |
d6b29d7c | 17 | #include <linux/splice.h> |
561c6731 | 18 | #include <linux/compat.h> |
29732938 | 19 | #include <linux/mount.h> |
2feb55f8 | 20 | #include <linux/fs.h> |
06ae43f3 | 21 | #include "internal.h" |
1da177e4 LT |
22 | |
23 | #include <asm/uaccess.h> | |
24 | #include <asm/unistd.h> | |
25 | ||
c0bd14af | 26 | typedef ssize_t (*io_fn_t)(struct file *, char __user *, size_t, loff_t *); |
293bc982 | 27 | typedef ssize_t (*iter_fn_t)(struct kiocb *, struct iov_iter *); |
c0bd14af | 28 | |
4b6f5d20 | 29 | const struct file_operations generic_ro_fops = { |
1da177e4 | 30 | .llseek = generic_file_llseek, |
aad4f8bb | 31 | .read_iter = generic_file_read_iter, |
1da177e4 | 32 | .mmap = generic_file_readonly_mmap, |
534f2aaa | 33 | .splice_read = generic_file_splice_read, |
1da177e4 LT |
34 | }; |
35 | ||
36 | EXPORT_SYMBOL(generic_ro_fops); | |
37 | ||
cccb5a1e | 38 | static inline int unsigned_offsets(struct file *file) |
4a3956c7 | 39 | { |
cccb5a1e | 40 | return file->f_mode & FMODE_UNSIGNED_OFFSET; |
4a3956c7 KH |
41 | } |
42 | ||
46a1c2c7 JL |
43 | /** |
44 | * vfs_setpos - update the file offset for lseek | |
45 | * @file: file structure in question | |
46 | * @offset: file offset to seek to | |
47 | * @maxsize: maximum file size | |
48 | * | |
49 | * This is a low-level filesystem helper for updating the file offset to | |
50 | * the value specified by @offset if the given offset is valid and it is | |
51 | * not equal to the current file offset. | |
52 | * | |
53 | * Return the specified offset on success and -EINVAL on invalid offset. | |
54 | */ | |
55 | loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize) | |
ef3d0fd2 AK |
56 | { |
57 | if (offset < 0 && !unsigned_offsets(file)) | |
58 | return -EINVAL; | |
59 | if (offset > maxsize) | |
60 | return -EINVAL; | |
61 | ||
62 | if (offset != file->f_pos) { | |
63 | file->f_pos = offset; | |
64 | file->f_version = 0; | |
65 | } | |
66 | return offset; | |
67 | } | |
46a1c2c7 | 68 | EXPORT_SYMBOL(vfs_setpos); |
ef3d0fd2 | 69 | |
3a8cff4f | 70 | /** |
5760495a | 71 | * generic_file_llseek_size - generic llseek implementation for regular files |
3a8cff4f CH |
72 | * @file: file structure to seek on |
73 | * @offset: file offset to seek to | |
965c8e59 | 74 | * @whence: type of seek |
e8b96eb5 ES |
75 | * @size: max size of this file in file system |
76 | * @eof: offset used for SEEK_END position | |
3a8cff4f | 77 | * |
5760495a | 78 | * This is a variant of generic_file_llseek that allows passing in a custom |
e8b96eb5 | 79 | * maximum file size and a custom EOF position, for e.g. hashed directories |
ef3d0fd2 AK |
80 | * |
81 | * Synchronization: | |
5760495a | 82 | * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms) |
ef3d0fd2 AK |
83 | * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes. |
84 | * read/writes behave like SEEK_SET against seeks. | |
3a8cff4f | 85 | */ |
9465efc9 | 86 | loff_t |
965c8e59 | 87 | generic_file_llseek_size(struct file *file, loff_t offset, int whence, |
e8b96eb5 | 88 | loff_t maxsize, loff_t eof) |
1da177e4 | 89 | { |
965c8e59 | 90 | switch (whence) { |
3a8cff4f | 91 | case SEEK_END: |
e8b96eb5 | 92 | offset += eof; |
3a8cff4f CH |
93 | break; |
94 | case SEEK_CUR: | |
5b6f1eb9 AK |
95 | /* |
96 | * Here we special-case the lseek(fd, 0, SEEK_CUR) | |
97 | * position-querying operation. Avoid rewriting the "same" | |
98 | * f_pos value back to the file because a concurrent read(), | |
99 | * write() or lseek() might have altered it | |
100 | */ | |
101 | if (offset == 0) | |
102 | return file->f_pos; | |
ef3d0fd2 AK |
103 | /* |
104 | * f_lock protects against read/modify/write race with other | |
105 | * SEEK_CURs. Note that parallel writes and reads behave | |
106 | * like SEEK_SET. | |
107 | */ | |
108 | spin_lock(&file->f_lock); | |
46a1c2c7 | 109 | offset = vfs_setpos(file, file->f_pos + offset, maxsize); |
ef3d0fd2 AK |
110 | spin_unlock(&file->f_lock); |
111 | return offset; | |
982d8165 JB |
112 | case SEEK_DATA: |
113 | /* | |
114 | * In the generic case the entire file is data, so as long as | |
115 | * offset isn't at the end of the file then the offset is data. | |
116 | */ | |
e8b96eb5 | 117 | if (offset >= eof) |
982d8165 JB |
118 | return -ENXIO; |
119 | break; | |
120 | case SEEK_HOLE: | |
121 | /* | |
122 | * There is a virtual hole at the end of the file, so as long as | |
123 | * offset isn't i_size or larger, return i_size. | |
124 | */ | |
e8b96eb5 | 125 | if (offset >= eof) |
982d8165 | 126 | return -ENXIO; |
e8b96eb5 | 127 | offset = eof; |
982d8165 | 128 | break; |
1da177e4 | 129 | } |
3a8cff4f | 130 | |
46a1c2c7 | 131 | return vfs_setpos(file, offset, maxsize); |
5760495a AK |
132 | } |
133 | EXPORT_SYMBOL(generic_file_llseek_size); | |
134 | ||
135 | /** | |
136 | * generic_file_llseek - generic llseek implementation for regular files | |
137 | * @file: file structure to seek on | |
138 | * @offset: file offset to seek to | |
965c8e59 | 139 | * @whence: type of seek |
5760495a AK |
140 | * |
141 | * This is a generic implemenation of ->llseek useable for all normal local | |
142 | * filesystems. It just updates the file offset to the value specified by | |
546ae2d2 | 143 | * @offset and @whence. |
5760495a | 144 | */ |
965c8e59 | 145 | loff_t generic_file_llseek(struct file *file, loff_t offset, int whence) |
5760495a AK |
146 | { |
147 | struct inode *inode = file->f_mapping->host; | |
148 | ||
965c8e59 | 149 | return generic_file_llseek_size(file, offset, whence, |
e8b96eb5 ES |
150 | inode->i_sb->s_maxbytes, |
151 | i_size_read(inode)); | |
1da177e4 | 152 | } |
9465efc9 | 153 | EXPORT_SYMBOL(generic_file_llseek); |
1da177e4 | 154 | |
1bf9d14d AV |
155 | /** |
156 | * fixed_size_llseek - llseek implementation for fixed-sized devices | |
157 | * @file: file structure to seek on | |
158 | * @offset: file offset to seek to | |
159 | * @whence: type of seek | |
160 | * @size: size of the file | |
161 | * | |
162 | */ | |
163 | loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size) | |
164 | { | |
165 | switch (whence) { | |
166 | case SEEK_SET: case SEEK_CUR: case SEEK_END: | |
167 | return generic_file_llseek_size(file, offset, whence, | |
168 | size, size); | |
169 | default: | |
170 | return -EINVAL; | |
171 | } | |
172 | } | |
173 | EXPORT_SYMBOL(fixed_size_llseek); | |
174 | ||
b25472f9 AV |
175 | /** |
176 | * no_seek_end_llseek - llseek implementation for fixed-sized devices | |
177 | * @file: file structure to seek on | |
178 | * @offset: file offset to seek to | |
179 | * @whence: type of seek | |
180 | * | |
181 | */ | |
182 | loff_t no_seek_end_llseek(struct file *file, loff_t offset, int whence) | |
183 | { | |
184 | switch (whence) { | |
185 | case SEEK_SET: case SEEK_CUR: | |
186 | return generic_file_llseek_size(file, offset, whence, | |
2feb55f8 | 187 | OFFSET_MAX, 0); |
b25472f9 AV |
188 | default: |
189 | return -EINVAL; | |
190 | } | |
191 | } | |
192 | EXPORT_SYMBOL(no_seek_end_llseek); | |
193 | ||
194 | /** | |
195 | * no_seek_end_llseek_size - llseek implementation for fixed-sized devices | |
196 | * @file: file structure to seek on | |
197 | * @offset: file offset to seek to | |
198 | * @whence: type of seek | |
199 | * @size: maximal offset allowed | |
200 | * | |
201 | */ | |
202 | loff_t no_seek_end_llseek_size(struct file *file, loff_t offset, int whence, loff_t size) | |
203 | { | |
204 | switch (whence) { | |
205 | case SEEK_SET: case SEEK_CUR: | |
206 | return generic_file_llseek_size(file, offset, whence, | |
207 | size, 0); | |
208 | default: | |
209 | return -EINVAL; | |
210 | } | |
211 | } | |
212 | EXPORT_SYMBOL(no_seek_end_llseek_size); | |
213 | ||
ae6afc3f B |
214 | /** |
215 | * noop_llseek - No Operation Performed llseek implementation | |
216 | * @file: file structure to seek on | |
217 | * @offset: file offset to seek to | |
965c8e59 | 218 | * @whence: type of seek |
ae6afc3f B |
219 | * |
220 | * This is an implementation of ->llseek useable for the rare special case when | |
221 | * userspace expects the seek to succeed but the (device) file is actually not | |
222 | * able to perform the seek. In this case you use noop_llseek() instead of | |
223 | * falling back to the default implementation of ->llseek. | |
224 | */ | |
965c8e59 | 225 | loff_t noop_llseek(struct file *file, loff_t offset, int whence) |
ae6afc3f B |
226 | { |
227 | return file->f_pos; | |
228 | } | |
229 | EXPORT_SYMBOL(noop_llseek); | |
230 | ||
965c8e59 | 231 | loff_t no_llseek(struct file *file, loff_t offset, int whence) |
1da177e4 LT |
232 | { |
233 | return -ESPIPE; | |
234 | } | |
235 | EXPORT_SYMBOL(no_llseek); | |
236 | ||
965c8e59 | 237 | loff_t default_llseek(struct file *file, loff_t offset, int whence) |
1da177e4 | 238 | { |
496ad9aa | 239 | struct inode *inode = file_inode(file); |
16abef0e | 240 | loff_t retval; |
1da177e4 | 241 | |
5955102c | 242 | inode_lock(inode); |
965c8e59 | 243 | switch (whence) { |
7b8e8924 | 244 | case SEEK_END: |
982d8165 | 245 | offset += i_size_read(inode); |
1da177e4 | 246 | break; |
7b8e8924 | 247 | case SEEK_CUR: |
5b6f1eb9 AK |
248 | if (offset == 0) { |
249 | retval = file->f_pos; | |
250 | goto out; | |
251 | } | |
1da177e4 | 252 | offset += file->f_pos; |
982d8165 JB |
253 | break; |
254 | case SEEK_DATA: | |
255 | /* | |
256 | * In the generic case the entire file is data, so as | |
257 | * long as offset isn't at the end of the file then the | |
258 | * offset is data. | |
259 | */ | |
bacb2d81 DC |
260 | if (offset >= inode->i_size) { |
261 | retval = -ENXIO; | |
262 | goto out; | |
263 | } | |
982d8165 JB |
264 | break; |
265 | case SEEK_HOLE: | |
266 | /* | |
267 | * There is a virtual hole at the end of the file, so | |
268 | * as long as offset isn't i_size or larger, return | |
269 | * i_size. | |
270 | */ | |
bacb2d81 DC |
271 | if (offset >= inode->i_size) { |
272 | retval = -ENXIO; | |
273 | goto out; | |
274 | } | |
982d8165 JB |
275 | offset = inode->i_size; |
276 | break; | |
1da177e4 LT |
277 | } |
278 | retval = -EINVAL; | |
cccb5a1e | 279 | if (offset >= 0 || unsigned_offsets(file)) { |
1da177e4 LT |
280 | if (offset != file->f_pos) { |
281 | file->f_pos = offset; | |
282 | file->f_version = 0; | |
283 | } | |
284 | retval = offset; | |
285 | } | |
5b6f1eb9 | 286 | out: |
5955102c | 287 | inode_unlock(inode); |
1da177e4 LT |
288 | return retval; |
289 | } | |
290 | EXPORT_SYMBOL(default_llseek); | |
291 | ||
965c8e59 | 292 | loff_t vfs_llseek(struct file *file, loff_t offset, int whence) |
1da177e4 LT |
293 | { |
294 | loff_t (*fn)(struct file *, loff_t, int); | |
295 | ||
296 | fn = no_llseek; | |
297 | if (file->f_mode & FMODE_LSEEK) { | |
72c2d531 | 298 | if (file->f_op->llseek) |
1da177e4 LT |
299 | fn = file->f_op->llseek; |
300 | } | |
965c8e59 | 301 | return fn(file, offset, whence); |
1da177e4 LT |
302 | } |
303 | EXPORT_SYMBOL(vfs_llseek); | |
304 | ||
965c8e59 | 305 | SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence) |
1da177e4 LT |
306 | { |
307 | off_t retval; | |
9c225f26 | 308 | struct fd f = fdget_pos(fd); |
2903ff01 AV |
309 | if (!f.file) |
310 | return -EBADF; | |
1da177e4 LT |
311 | |
312 | retval = -EINVAL; | |
965c8e59 AM |
313 | if (whence <= SEEK_MAX) { |
314 | loff_t res = vfs_llseek(f.file, offset, whence); | |
1da177e4 LT |
315 | retval = res; |
316 | if (res != (loff_t)retval) | |
317 | retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */ | |
318 | } | |
9c225f26 | 319 | fdput_pos(f); |
1da177e4 LT |
320 | return retval; |
321 | } | |
322 | ||
561c6731 AV |
323 | #ifdef CONFIG_COMPAT |
324 | COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence) | |
325 | { | |
326 | return sys_lseek(fd, offset, whence); | |
327 | } | |
328 | #endif | |
329 | ||
1da177e4 | 330 | #ifdef __ARCH_WANT_SYS_LLSEEK |
003d7ab4 HC |
331 | SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high, |
332 | unsigned long, offset_low, loff_t __user *, result, | |
965c8e59 | 333 | unsigned int, whence) |
1da177e4 LT |
334 | { |
335 | int retval; | |
d7a15f8d | 336 | struct fd f = fdget_pos(fd); |
1da177e4 | 337 | loff_t offset; |
1da177e4 | 338 | |
2903ff01 AV |
339 | if (!f.file) |
340 | return -EBADF; | |
1da177e4 LT |
341 | |
342 | retval = -EINVAL; | |
965c8e59 | 343 | if (whence > SEEK_MAX) |
1da177e4 LT |
344 | goto out_putf; |
345 | ||
2903ff01 | 346 | offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low, |
965c8e59 | 347 | whence); |
1da177e4 LT |
348 | |
349 | retval = (int)offset; | |
350 | if (offset >= 0) { | |
351 | retval = -EFAULT; | |
352 | if (!copy_to_user(result, &offset, sizeof(offset))) | |
353 | retval = 0; | |
354 | } | |
355 | out_putf: | |
d7a15f8d | 356 | fdput_pos(f); |
1da177e4 LT |
357 | return retval; |
358 | } | |
359 | #endif | |
360 | ||
dbe4e192 CH |
361 | ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos) |
362 | { | |
363 | struct kiocb kiocb; | |
364 | ssize_t ret; | |
365 | ||
366 | if (!file->f_op->read_iter) | |
367 | return -EINVAL; | |
368 | ||
369 | init_sync_kiocb(&kiocb, file); | |
370 | kiocb.ki_pos = *ppos; | |
dbe4e192 CH |
371 | |
372 | iter->type |= READ; | |
373 | ret = file->f_op->read_iter(&kiocb, iter); | |
599bd19b | 374 | BUG_ON(ret == -EIOCBQUEUED); |
dbe4e192 CH |
375 | if (ret > 0) |
376 | *ppos = kiocb.ki_pos; | |
377 | return ret; | |
378 | } | |
379 | EXPORT_SYMBOL(vfs_iter_read); | |
380 | ||
381 | ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos) | |
382 | { | |
383 | struct kiocb kiocb; | |
384 | ssize_t ret; | |
385 | ||
386 | if (!file->f_op->write_iter) | |
387 | return -EINVAL; | |
388 | ||
389 | init_sync_kiocb(&kiocb, file); | |
390 | kiocb.ki_pos = *ppos; | |
dbe4e192 CH |
391 | |
392 | iter->type |= WRITE; | |
393 | ret = file->f_op->write_iter(&kiocb, iter); | |
599bd19b | 394 | BUG_ON(ret == -EIOCBQUEUED); |
dbe4e192 CH |
395 | if (ret > 0) |
396 | *ppos = kiocb.ki_pos; | |
397 | return ret; | |
398 | } | |
399 | EXPORT_SYMBOL(vfs_iter_write); | |
400 | ||
e28cc715 LT |
401 | /* |
402 | * rw_verify_area doesn't like huge counts. We limit | |
403 | * them to something that fits in "int" so that others | |
404 | * won't have to do range checks all the time. | |
405 | */ | |
68d70d03 | 406 | int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count) |
1da177e4 LT |
407 | { |
408 | struct inode *inode; | |
409 | loff_t pos; | |
c43e259c | 410 | int retval = -EINVAL; |
1da177e4 | 411 | |
496ad9aa | 412 | inode = file_inode(file); |
e28cc715 | 413 | if (unlikely((ssize_t) count < 0)) |
c43e259c | 414 | return retval; |
1da177e4 | 415 | pos = *ppos; |
cccb5a1e AV |
416 | if (unlikely(pos < 0)) { |
417 | if (!unsigned_offsets(file)) | |
418 | return retval; | |
419 | if (count >= -pos) /* both values are in 0..LLONG_MAX */ | |
420 | return -EOVERFLOW; | |
421 | } else if (unlikely((loff_t) (pos + count) < 0)) { | |
422 | if (!unsigned_offsets(file)) | |
4a3956c7 KH |
423 | return retval; |
424 | } | |
1da177e4 | 425 | |
bd61e0a9 | 426 | if (unlikely(inode->i_flctx && mandatory_lock(inode))) { |
acc15575 CH |
427 | retval = locks_mandatory_area(inode, file, pos, pos + count - 1, |
428 | read_write == READ ? F_RDLCK : F_WRLCK); | |
e28cc715 LT |
429 | if (retval < 0) |
430 | return retval; | |
431 | } | |
c43e259c JM |
432 | retval = security_file_permission(file, |
433 | read_write == READ ? MAY_READ : MAY_WRITE); | |
434 | if (retval) | |
435 | return retval; | |
e28cc715 | 436 | return count > MAX_RW_COUNT ? MAX_RW_COUNT : count; |
1da177e4 LT |
437 | } |
438 | ||
5d5d5689 | 439 | static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos) |
293bc982 AV |
440 | { |
441 | struct iovec iov = { .iov_base = buf, .iov_len = len }; | |
442 | struct kiocb kiocb; | |
443 | struct iov_iter iter; | |
444 | ssize_t ret; | |
445 | ||
446 | init_sync_kiocb(&kiocb, filp); | |
447 | kiocb.ki_pos = *ppos; | |
293bc982 AV |
448 | iov_iter_init(&iter, READ, &iov, 1, len); |
449 | ||
450 | ret = filp->f_op->read_iter(&kiocb, &iter); | |
599bd19b | 451 | BUG_ON(ret == -EIOCBQUEUED); |
293bc982 AV |
452 | *ppos = kiocb.ki_pos; |
453 | return ret; | |
454 | } | |
455 | ||
6fb5032e DK |
456 | ssize_t __vfs_read(struct file *file, char __user *buf, size_t count, |
457 | loff_t *pos) | |
458 | { | |
6fb5032e | 459 | if (file->f_op->read) |
3d04c8a1 | 460 | return file->f_op->read(file, buf, count, pos); |
6fb5032e | 461 | else if (file->f_op->read_iter) |
3d04c8a1 | 462 | return new_sync_read(file, buf, count, pos); |
6fb5032e | 463 | else |
3d04c8a1 | 464 | return -EINVAL; |
6fb5032e | 465 | } |
3d04c8a1 | 466 | EXPORT_SYMBOL(__vfs_read); |
6fb5032e | 467 | |
1da177e4 LT |
468 | ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos) |
469 | { | |
470 | ssize_t ret; | |
471 | ||
472 | if (!(file->f_mode & FMODE_READ)) | |
473 | return -EBADF; | |
7f7f25e8 | 474 | if (!(file->f_mode & FMODE_CAN_READ)) |
1da177e4 LT |
475 | return -EINVAL; |
476 | if (unlikely(!access_ok(VERIFY_WRITE, buf, count))) | |
477 | return -EFAULT; | |
478 | ||
479 | ret = rw_verify_area(READ, file, pos, count); | |
e28cc715 LT |
480 | if (ret >= 0) { |
481 | count = ret; | |
6fb5032e | 482 | ret = __vfs_read(file, buf, count, pos); |
c43e259c | 483 | if (ret > 0) { |
2a12a9d7 | 484 | fsnotify_access(file); |
c43e259c | 485 | add_rchar(current, ret); |
1da177e4 | 486 | } |
c43e259c | 487 | inc_syscr(current); |
1da177e4 LT |
488 | } |
489 | ||
490 | return ret; | |
491 | } | |
492 | ||
493 | EXPORT_SYMBOL(vfs_read); | |
494 | ||
5d5d5689 | 495 | static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos) |
293bc982 AV |
496 | { |
497 | struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len }; | |
498 | struct kiocb kiocb; | |
499 | struct iov_iter iter; | |
500 | ssize_t ret; | |
501 | ||
502 | init_sync_kiocb(&kiocb, filp); | |
503 | kiocb.ki_pos = *ppos; | |
293bc982 AV |
504 | iov_iter_init(&iter, WRITE, &iov, 1, len); |
505 | ||
506 | ret = filp->f_op->write_iter(&kiocb, &iter); | |
599bd19b | 507 | BUG_ON(ret == -EIOCBQUEUED); |
f765b134 AV |
508 | if (ret > 0) |
509 | *ppos = kiocb.ki_pos; | |
293bc982 AV |
510 | return ret; |
511 | } | |
512 | ||
493c84c0 AV |
513 | ssize_t __vfs_write(struct file *file, const char __user *p, size_t count, |
514 | loff_t *pos) | |
515 | { | |
516 | if (file->f_op->write) | |
517 | return file->f_op->write(file, p, count, pos); | |
493c84c0 AV |
518 | else if (file->f_op->write_iter) |
519 | return new_sync_write(file, p, count, pos); | |
520 | else | |
521 | return -EINVAL; | |
522 | } | |
523 | EXPORT_SYMBOL(__vfs_write); | |
524 | ||
06ae43f3 AV |
525 | ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos) |
526 | { | |
527 | mm_segment_t old_fs; | |
528 | const char __user *p; | |
529 | ssize_t ret; | |
530 | ||
7f7f25e8 | 531 | if (!(file->f_mode & FMODE_CAN_WRITE)) |
3e84f48e AV |
532 | return -EINVAL; |
533 | ||
06ae43f3 AV |
534 | old_fs = get_fs(); |
535 | set_fs(get_ds()); | |
536 | p = (__force const char __user *)buf; | |
537 | if (count > MAX_RW_COUNT) | |
538 | count = MAX_RW_COUNT; | |
493c84c0 | 539 | ret = __vfs_write(file, p, count, pos); |
06ae43f3 AV |
540 | set_fs(old_fs); |
541 | if (ret > 0) { | |
542 | fsnotify_modify(file); | |
543 | add_wchar(current, ret); | |
544 | } | |
545 | inc_syscw(current); | |
546 | return ret; | |
547 | } | |
548 | ||
2ec3a12a AV |
549 | EXPORT_SYMBOL(__kernel_write); |
550 | ||
1da177e4 LT |
551 | ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos) |
552 | { | |
553 | ssize_t ret; | |
554 | ||
555 | if (!(file->f_mode & FMODE_WRITE)) | |
556 | return -EBADF; | |
7f7f25e8 | 557 | if (!(file->f_mode & FMODE_CAN_WRITE)) |
1da177e4 LT |
558 | return -EINVAL; |
559 | if (unlikely(!access_ok(VERIFY_READ, buf, count))) | |
560 | return -EFAULT; | |
561 | ||
562 | ret = rw_verify_area(WRITE, file, pos, count); | |
e28cc715 LT |
563 | if (ret >= 0) { |
564 | count = ret; | |
03d95eb2 | 565 | file_start_write(file); |
493c84c0 | 566 | ret = __vfs_write(file, buf, count, pos); |
c43e259c | 567 | if (ret > 0) { |
2a12a9d7 | 568 | fsnotify_modify(file); |
c43e259c | 569 | add_wchar(current, ret); |
1da177e4 | 570 | } |
c43e259c | 571 | inc_syscw(current); |
03d95eb2 | 572 | file_end_write(file); |
1da177e4 LT |
573 | } |
574 | ||
575 | return ret; | |
576 | } | |
577 | ||
578 | EXPORT_SYMBOL(vfs_write); | |
579 | ||
580 | static inline loff_t file_pos_read(struct file *file) | |
581 | { | |
582 | return file->f_pos; | |
583 | } | |
584 | ||
585 | static inline void file_pos_write(struct file *file, loff_t pos) | |
586 | { | |
587 | file->f_pos = pos; | |
588 | } | |
589 | ||
3cdad428 | 590 | SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count) |
1da177e4 | 591 | { |
9c225f26 | 592 | struct fd f = fdget_pos(fd); |
1da177e4 | 593 | ssize_t ret = -EBADF; |
1da177e4 | 594 | |
2903ff01 AV |
595 | if (f.file) { |
596 | loff_t pos = file_pos_read(f.file); | |
597 | ret = vfs_read(f.file, buf, count, &pos); | |
5faf153e AV |
598 | if (ret >= 0) |
599 | file_pos_write(f.file, pos); | |
9c225f26 | 600 | fdput_pos(f); |
1da177e4 | 601 | } |
1da177e4 LT |
602 | return ret; |
603 | } | |
1da177e4 | 604 | |
3cdad428 HC |
605 | SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf, |
606 | size_t, count) | |
1da177e4 | 607 | { |
9c225f26 | 608 | struct fd f = fdget_pos(fd); |
1da177e4 | 609 | ssize_t ret = -EBADF; |
1da177e4 | 610 | |
2903ff01 AV |
611 | if (f.file) { |
612 | loff_t pos = file_pos_read(f.file); | |
613 | ret = vfs_write(f.file, buf, count, &pos); | |
5faf153e AV |
614 | if (ret >= 0) |
615 | file_pos_write(f.file, pos); | |
9c225f26 | 616 | fdput_pos(f); |
1da177e4 LT |
617 | } |
618 | ||
619 | return ret; | |
620 | } | |
621 | ||
4a0fd5bf AV |
622 | SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf, |
623 | size_t, count, loff_t, pos) | |
1da177e4 | 624 | { |
2903ff01 | 625 | struct fd f; |
1da177e4 | 626 | ssize_t ret = -EBADF; |
1da177e4 LT |
627 | |
628 | if (pos < 0) | |
629 | return -EINVAL; | |
630 | ||
2903ff01 AV |
631 | f = fdget(fd); |
632 | if (f.file) { | |
1da177e4 | 633 | ret = -ESPIPE; |
2903ff01 AV |
634 | if (f.file->f_mode & FMODE_PREAD) |
635 | ret = vfs_read(f.file, buf, count, &pos); | |
636 | fdput(f); | |
1da177e4 LT |
637 | } |
638 | ||
639 | return ret; | |
640 | } | |
641 | ||
4a0fd5bf AV |
642 | SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf, |
643 | size_t, count, loff_t, pos) | |
1da177e4 | 644 | { |
2903ff01 | 645 | struct fd f; |
1da177e4 | 646 | ssize_t ret = -EBADF; |
1da177e4 LT |
647 | |
648 | if (pos < 0) | |
649 | return -EINVAL; | |
650 | ||
2903ff01 AV |
651 | f = fdget(fd); |
652 | if (f.file) { | |
1da177e4 | 653 | ret = -ESPIPE; |
2903ff01 AV |
654 | if (f.file->f_mode & FMODE_PWRITE) |
655 | ret = vfs_write(f.file, buf, count, &pos); | |
656 | fdput(f); | |
1da177e4 LT |
657 | } |
658 | ||
659 | return ret; | |
660 | } | |
661 | ||
662 | /* | |
663 | * Reduce an iovec's length in-place. Return the resulting number of segments | |
664 | */ | |
665 | unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to) | |
666 | { | |
667 | unsigned long seg = 0; | |
668 | size_t len = 0; | |
669 | ||
670 | while (seg < nr_segs) { | |
671 | seg++; | |
672 | if (len + iov->iov_len >= to) { | |
673 | iov->iov_len = to - len; | |
674 | break; | |
675 | } | |
676 | len += iov->iov_len; | |
677 | iov++; | |
678 | } | |
679 | return seg; | |
680 | } | |
19295529 | 681 | EXPORT_SYMBOL(iov_shorten); |
1da177e4 | 682 | |
ac15ac06 | 683 | static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter, |
793b80ef | 684 | loff_t *ppos, iter_fn_t fn, int flags) |
293bc982 AV |
685 | { |
686 | struct kiocb kiocb; | |
293bc982 AV |
687 | ssize_t ret; |
688 | ||
97be7ebe | 689 | if (flags & ~RWF_HIPRI) |
793b80ef CH |
690 | return -EOPNOTSUPP; |
691 | ||
293bc982 | 692 | init_sync_kiocb(&kiocb, filp); |
97be7ebe CH |
693 | if (flags & RWF_HIPRI) |
694 | kiocb.ki_flags |= IOCB_HIPRI; | |
293bc982 | 695 | kiocb.ki_pos = *ppos; |
293bc982 | 696 | |
ac15ac06 | 697 | ret = fn(&kiocb, iter); |
599bd19b | 698 | BUG_ON(ret == -EIOCBQUEUED); |
293bc982 AV |
699 | *ppos = kiocb.ki_pos; |
700 | return ret; | |
701 | } | |
702 | ||
ee0b3e67 | 703 | /* Do it by hand, with file-ops */ |
ac15ac06 | 704 | static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter, |
793b80ef | 705 | loff_t *ppos, io_fn_t fn, int flags) |
ee0b3e67 | 706 | { |
ee0b3e67 BP |
707 | ssize_t ret = 0; |
708 | ||
97be7ebe | 709 | if (flags & ~RWF_HIPRI) |
793b80ef CH |
710 | return -EOPNOTSUPP; |
711 | ||
ac15ac06 AV |
712 | while (iov_iter_count(iter)) { |
713 | struct iovec iovec = iov_iter_iovec(iter); | |
ee0b3e67 BP |
714 | ssize_t nr; |
715 | ||
ac15ac06 | 716 | nr = fn(filp, iovec.iov_base, iovec.iov_len, ppos); |
ee0b3e67 BP |
717 | |
718 | if (nr < 0) { | |
719 | if (!ret) | |
720 | ret = nr; | |
721 | break; | |
722 | } | |
723 | ret += nr; | |
ac15ac06 | 724 | if (nr != iovec.iov_len) |
ee0b3e67 | 725 | break; |
ac15ac06 | 726 | iov_iter_advance(iter, nr); |
ee0b3e67 BP |
727 | } |
728 | ||
729 | return ret; | |
730 | } | |
731 | ||
1da177e4 LT |
732 | /* A write operation does a read from user space and vice versa */ |
733 | #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ) | |
734 | ||
eed4e51f BP |
735 | ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector, |
736 | unsigned long nr_segs, unsigned long fast_segs, | |
737 | struct iovec *fast_pointer, | |
ac34ebb3 | 738 | struct iovec **ret_pointer) |
435f49a5 | 739 | { |
eed4e51f | 740 | unsigned long seg; |
435f49a5 | 741 | ssize_t ret; |
eed4e51f BP |
742 | struct iovec *iov = fast_pointer; |
743 | ||
435f49a5 LT |
744 | /* |
745 | * SuS says "The readv() function *may* fail if the iovcnt argument | |
746 | * was less than or equal to 0, or greater than {IOV_MAX}. Linux has | |
747 | * traditionally returned zero for zero segments, so... | |
748 | */ | |
eed4e51f BP |
749 | if (nr_segs == 0) { |
750 | ret = 0; | |
435f49a5 | 751 | goto out; |
eed4e51f BP |
752 | } |
753 | ||
435f49a5 LT |
754 | /* |
755 | * First get the "struct iovec" from user memory and | |
756 | * verify all the pointers | |
757 | */ | |
eed4e51f BP |
758 | if (nr_segs > UIO_MAXIOV) { |
759 | ret = -EINVAL; | |
435f49a5 | 760 | goto out; |
eed4e51f BP |
761 | } |
762 | if (nr_segs > fast_segs) { | |
435f49a5 | 763 | iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL); |
eed4e51f BP |
764 | if (iov == NULL) { |
765 | ret = -ENOMEM; | |
435f49a5 | 766 | goto out; |
eed4e51f | 767 | } |
435f49a5 | 768 | } |
eed4e51f BP |
769 | if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) { |
770 | ret = -EFAULT; | |
435f49a5 | 771 | goto out; |
eed4e51f BP |
772 | } |
773 | ||
435f49a5 | 774 | /* |
eed4e51f BP |
775 | * According to the Single Unix Specification we should return EINVAL |
776 | * if an element length is < 0 when cast to ssize_t or if the | |
777 | * total length would overflow the ssize_t return value of the | |
778 | * system call. | |
435f49a5 LT |
779 | * |
780 | * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the | |
781 | * overflow case. | |
782 | */ | |
eed4e51f | 783 | ret = 0; |
435f49a5 LT |
784 | for (seg = 0; seg < nr_segs; seg++) { |
785 | void __user *buf = iov[seg].iov_base; | |
786 | ssize_t len = (ssize_t)iov[seg].iov_len; | |
eed4e51f BP |
787 | |
788 | /* see if we we're about to use an invalid len or if | |
789 | * it's about to overflow ssize_t */ | |
435f49a5 | 790 | if (len < 0) { |
eed4e51f | 791 | ret = -EINVAL; |
435f49a5 | 792 | goto out; |
eed4e51f | 793 | } |
ac34ebb3 | 794 | if (type >= 0 |
fcf63409 | 795 | && unlikely(!access_ok(vrfy_dir(type), buf, len))) { |
eed4e51f | 796 | ret = -EFAULT; |
435f49a5 LT |
797 | goto out; |
798 | } | |
799 | if (len > MAX_RW_COUNT - ret) { | |
800 | len = MAX_RW_COUNT - ret; | |
801 | iov[seg].iov_len = len; | |
eed4e51f | 802 | } |
eed4e51f | 803 | ret += len; |
435f49a5 | 804 | } |
eed4e51f BP |
805 | out: |
806 | *ret_pointer = iov; | |
807 | return ret; | |
808 | } | |
809 | ||
1da177e4 LT |
810 | static ssize_t do_readv_writev(int type, struct file *file, |
811 | const struct iovec __user * uvector, | |
793b80ef CH |
812 | unsigned long nr_segs, loff_t *pos, |
813 | int flags) | |
1da177e4 | 814 | { |
1da177e4 LT |
815 | size_t tot_len; |
816 | struct iovec iovstack[UIO_FASTIOV]; | |
ee0b3e67 | 817 | struct iovec *iov = iovstack; |
ac15ac06 | 818 | struct iov_iter iter; |
1da177e4 | 819 | ssize_t ret; |
1da177e4 | 820 | io_fn_t fn; |
293bc982 | 821 | iter_fn_t iter_fn; |
1da177e4 | 822 | |
0504c074 AV |
823 | ret = import_iovec(type, uvector, nr_segs, |
824 | ARRAY_SIZE(iovstack), &iov, &iter); | |
825 | if (ret < 0) | |
826 | return ret; | |
1da177e4 | 827 | |
0504c074 AV |
828 | tot_len = iov_iter_count(&iter); |
829 | if (!tot_len) | |
830 | goto out; | |
1da177e4 | 831 | ret = rw_verify_area(type, file, pos, tot_len); |
e28cc715 | 832 | if (ret < 0) |
411b67b4 | 833 | goto out; |
1da177e4 | 834 | |
1da177e4 LT |
835 | if (type == READ) { |
836 | fn = file->f_op->read; | |
293bc982 | 837 | iter_fn = file->f_op->read_iter; |
1da177e4 LT |
838 | } else { |
839 | fn = (io_fn_t)file->f_op->write; | |
293bc982 | 840 | iter_fn = file->f_op->write_iter; |
03d95eb2 | 841 | file_start_write(file); |
1da177e4 LT |
842 | } |
843 | ||
293bc982 | 844 | if (iter_fn) |
793b80ef | 845 | ret = do_iter_readv_writev(file, &iter, pos, iter_fn, flags); |
ee0b3e67 | 846 | else |
793b80ef | 847 | ret = do_loop_readv_writev(file, &iter, pos, fn, flags); |
1da177e4 | 848 | |
03d95eb2 AV |
849 | if (type != READ) |
850 | file_end_write(file); | |
851 | ||
1da177e4 | 852 | out: |
0504c074 | 853 | kfree(iov); |
0eeca283 RL |
854 | if ((ret + (type == READ)) > 0) { |
855 | if (type == READ) | |
2a12a9d7 | 856 | fsnotify_access(file); |
0eeca283 | 857 | else |
2a12a9d7 | 858 | fsnotify_modify(file); |
0eeca283 | 859 | } |
1da177e4 | 860 | return ret; |
1da177e4 LT |
861 | } |
862 | ||
863 | ssize_t vfs_readv(struct file *file, const struct iovec __user *vec, | |
793b80ef | 864 | unsigned long vlen, loff_t *pos, int flags) |
1da177e4 LT |
865 | { |
866 | if (!(file->f_mode & FMODE_READ)) | |
867 | return -EBADF; | |
7f7f25e8 | 868 | if (!(file->f_mode & FMODE_CAN_READ)) |
1da177e4 LT |
869 | return -EINVAL; |
870 | ||
793b80ef | 871 | return do_readv_writev(READ, file, vec, vlen, pos, flags); |
1da177e4 LT |
872 | } |
873 | ||
874 | EXPORT_SYMBOL(vfs_readv); | |
875 | ||
876 | ssize_t vfs_writev(struct file *file, const struct iovec __user *vec, | |
793b80ef | 877 | unsigned long vlen, loff_t *pos, int flags) |
1da177e4 LT |
878 | { |
879 | if (!(file->f_mode & FMODE_WRITE)) | |
880 | return -EBADF; | |
7f7f25e8 | 881 | if (!(file->f_mode & FMODE_CAN_WRITE)) |
1da177e4 LT |
882 | return -EINVAL; |
883 | ||
793b80ef | 884 | return do_readv_writev(WRITE, file, vec, vlen, pos, flags); |
1da177e4 LT |
885 | } |
886 | ||
887 | EXPORT_SYMBOL(vfs_writev); | |
888 | ||
f17d8b35 MT |
889 | static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec, |
890 | unsigned long vlen, int flags) | |
1da177e4 | 891 | { |
9c225f26 | 892 | struct fd f = fdget_pos(fd); |
1da177e4 | 893 | ssize_t ret = -EBADF; |
1da177e4 | 894 | |
2903ff01 AV |
895 | if (f.file) { |
896 | loff_t pos = file_pos_read(f.file); | |
f17d8b35 | 897 | ret = vfs_readv(f.file, vec, vlen, &pos, flags); |
5faf153e AV |
898 | if (ret >= 0) |
899 | file_pos_write(f.file, pos); | |
9c225f26 | 900 | fdput_pos(f); |
1da177e4 LT |
901 | } |
902 | ||
903 | if (ret > 0) | |
4b98d11b AD |
904 | add_rchar(current, ret); |
905 | inc_syscr(current); | |
1da177e4 LT |
906 | return ret; |
907 | } | |
908 | ||
f17d8b35 MT |
909 | static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec, |
910 | unsigned long vlen, int flags) | |
1da177e4 | 911 | { |
9c225f26 | 912 | struct fd f = fdget_pos(fd); |
1da177e4 | 913 | ssize_t ret = -EBADF; |
1da177e4 | 914 | |
2903ff01 AV |
915 | if (f.file) { |
916 | loff_t pos = file_pos_read(f.file); | |
f17d8b35 | 917 | ret = vfs_writev(f.file, vec, vlen, &pos, flags); |
5faf153e AV |
918 | if (ret >= 0) |
919 | file_pos_write(f.file, pos); | |
9c225f26 | 920 | fdput_pos(f); |
1da177e4 LT |
921 | } |
922 | ||
923 | if (ret > 0) | |
4b98d11b AD |
924 | add_wchar(current, ret); |
925 | inc_syscw(current); | |
1da177e4 LT |
926 | return ret; |
927 | } | |
928 | ||
601cc11d LT |
929 | static inline loff_t pos_from_hilo(unsigned long high, unsigned long low) |
930 | { | |
931 | #define HALF_LONG_BITS (BITS_PER_LONG / 2) | |
932 | return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low; | |
933 | } | |
934 | ||
f17d8b35 MT |
935 | static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec, |
936 | unsigned long vlen, loff_t pos, int flags) | |
f3554f4b | 937 | { |
2903ff01 | 938 | struct fd f; |
f3554f4b | 939 | ssize_t ret = -EBADF; |
f3554f4b GH |
940 | |
941 | if (pos < 0) | |
942 | return -EINVAL; | |
943 | ||
2903ff01 AV |
944 | f = fdget(fd); |
945 | if (f.file) { | |
f3554f4b | 946 | ret = -ESPIPE; |
2903ff01 | 947 | if (f.file->f_mode & FMODE_PREAD) |
f17d8b35 | 948 | ret = vfs_readv(f.file, vec, vlen, &pos, flags); |
2903ff01 | 949 | fdput(f); |
f3554f4b GH |
950 | } |
951 | ||
952 | if (ret > 0) | |
953 | add_rchar(current, ret); | |
954 | inc_syscr(current); | |
955 | return ret; | |
956 | } | |
957 | ||
f17d8b35 MT |
958 | static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec, |
959 | unsigned long vlen, loff_t pos, int flags) | |
f3554f4b | 960 | { |
2903ff01 | 961 | struct fd f; |
f3554f4b | 962 | ssize_t ret = -EBADF; |
f3554f4b GH |
963 | |
964 | if (pos < 0) | |
965 | return -EINVAL; | |
966 | ||
2903ff01 AV |
967 | f = fdget(fd); |
968 | if (f.file) { | |
f3554f4b | 969 | ret = -ESPIPE; |
2903ff01 | 970 | if (f.file->f_mode & FMODE_PWRITE) |
f17d8b35 | 971 | ret = vfs_writev(f.file, vec, vlen, &pos, flags); |
2903ff01 | 972 | fdput(f); |
f3554f4b GH |
973 | } |
974 | ||
975 | if (ret > 0) | |
976 | add_wchar(current, ret); | |
977 | inc_syscw(current); | |
978 | return ret; | |
979 | } | |
980 | ||
f17d8b35 MT |
981 | SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec, |
982 | unsigned long, vlen) | |
983 | { | |
984 | return do_readv(fd, vec, vlen, 0); | |
985 | } | |
986 | ||
987 | SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec, | |
988 | unsigned long, vlen) | |
989 | { | |
990 | return do_writev(fd, vec, vlen, 0); | |
991 | } | |
992 | ||
993 | SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec, | |
994 | unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h) | |
995 | { | |
996 | loff_t pos = pos_from_hilo(pos_h, pos_l); | |
997 | ||
998 | return do_preadv(fd, vec, vlen, pos, 0); | |
999 | } | |
1000 | ||
1001 | SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec, | |
1002 | unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h, | |
1003 | int, flags) | |
1004 | { | |
1005 | loff_t pos = pos_from_hilo(pos_h, pos_l); | |
1006 | ||
1007 | if (pos == -1) | |
1008 | return do_readv(fd, vec, vlen, flags); | |
1009 | ||
1010 | return do_preadv(fd, vec, vlen, pos, flags); | |
1011 | } | |
1012 | ||
1013 | SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec, | |
1014 | unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h) | |
1015 | { | |
1016 | loff_t pos = pos_from_hilo(pos_h, pos_l); | |
1017 | ||
1018 | return do_pwritev(fd, vec, vlen, pos, 0); | |
1019 | } | |
1020 | ||
1021 | SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec, | |
1022 | unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h, | |
1023 | int, flags) | |
1024 | { | |
1025 | loff_t pos = pos_from_hilo(pos_h, pos_l); | |
1026 | ||
1027 | if (pos == -1) | |
1028 | return do_writev(fd, vec, vlen, flags); | |
1029 | ||
1030 | return do_pwritev(fd, vec, vlen, pos, flags); | |
1031 | } | |
1032 | ||
72ec3516 AV |
1033 | #ifdef CONFIG_COMPAT |
1034 | ||
1035 | static ssize_t compat_do_readv_writev(int type, struct file *file, | |
1036 | const struct compat_iovec __user *uvector, | |
793b80ef CH |
1037 | unsigned long nr_segs, loff_t *pos, |
1038 | int flags) | |
72ec3516 AV |
1039 | { |
1040 | compat_ssize_t tot_len; | |
1041 | struct iovec iovstack[UIO_FASTIOV]; | |
1042 | struct iovec *iov = iovstack; | |
ac15ac06 | 1043 | struct iov_iter iter; |
72ec3516 AV |
1044 | ssize_t ret; |
1045 | io_fn_t fn; | |
293bc982 | 1046 | iter_fn_t iter_fn; |
72ec3516 | 1047 | |
0504c074 AV |
1048 | ret = compat_import_iovec(type, uvector, nr_segs, |
1049 | UIO_FASTIOV, &iov, &iter); | |
1050 | if (ret < 0) | |
1051 | return ret; | |
72ec3516 | 1052 | |
0504c074 AV |
1053 | tot_len = iov_iter_count(&iter); |
1054 | if (!tot_len) | |
1055 | goto out; | |
72ec3516 AV |
1056 | ret = rw_verify_area(type, file, pos, tot_len); |
1057 | if (ret < 0) | |
1058 | goto out; | |
1059 | ||
72ec3516 AV |
1060 | if (type == READ) { |
1061 | fn = file->f_op->read; | |
293bc982 | 1062 | iter_fn = file->f_op->read_iter; |
72ec3516 AV |
1063 | } else { |
1064 | fn = (io_fn_t)file->f_op->write; | |
293bc982 | 1065 | iter_fn = file->f_op->write_iter; |
03d95eb2 | 1066 | file_start_write(file); |
72ec3516 AV |
1067 | } |
1068 | ||
293bc982 | 1069 | if (iter_fn) |
793b80ef | 1070 | ret = do_iter_readv_writev(file, &iter, pos, iter_fn, flags); |
03d95eb2 | 1071 | else |
793b80ef | 1072 | ret = do_loop_readv_writev(file, &iter, pos, fn, flags); |
72ec3516 | 1073 | |
03d95eb2 AV |
1074 | if (type != READ) |
1075 | file_end_write(file); | |
1076 | ||
72ec3516 | 1077 | out: |
0504c074 | 1078 | kfree(iov); |
72ec3516 AV |
1079 | if ((ret + (type == READ)) > 0) { |
1080 | if (type == READ) | |
1081 | fsnotify_access(file); | |
1082 | else | |
1083 | fsnotify_modify(file); | |
1084 | } | |
1085 | return ret; | |
1086 | } | |
1087 | ||
1088 | static size_t compat_readv(struct file *file, | |
1089 | const struct compat_iovec __user *vec, | |
f17d8b35 | 1090 | unsigned long vlen, loff_t *pos, int flags) |
72ec3516 AV |
1091 | { |
1092 | ssize_t ret = -EBADF; | |
1093 | ||
1094 | if (!(file->f_mode & FMODE_READ)) | |
1095 | goto out; | |
1096 | ||
1097 | ret = -EINVAL; | |
7f7f25e8 | 1098 | if (!(file->f_mode & FMODE_CAN_READ)) |
72ec3516 AV |
1099 | goto out; |
1100 | ||
f17d8b35 | 1101 | ret = compat_do_readv_writev(READ, file, vec, vlen, pos, flags); |
72ec3516 AV |
1102 | |
1103 | out: | |
1104 | if (ret > 0) | |
1105 | add_rchar(current, ret); | |
1106 | inc_syscr(current); | |
1107 | return ret; | |
1108 | } | |
1109 | ||
f17d8b35 MT |
1110 | static size_t do_compat_readv(compat_ulong_t fd, |
1111 | const struct compat_iovec __user *vec, | |
1112 | compat_ulong_t vlen, int flags) | |
72ec3516 | 1113 | { |
9c225f26 | 1114 | struct fd f = fdget_pos(fd); |
72ec3516 AV |
1115 | ssize_t ret; |
1116 | loff_t pos; | |
1117 | ||
1118 | if (!f.file) | |
1119 | return -EBADF; | |
1120 | pos = f.file->f_pos; | |
f17d8b35 | 1121 | ret = compat_readv(f.file, vec, vlen, &pos, flags); |
5faf153e AV |
1122 | if (ret >= 0) |
1123 | f.file->f_pos = pos; | |
9c225f26 | 1124 | fdput_pos(f); |
72ec3516 | 1125 | return ret; |
f17d8b35 | 1126 | |
72ec3516 AV |
1127 | } |
1128 | ||
f17d8b35 MT |
1129 | COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd, |
1130 | const struct compat_iovec __user *,vec, | |
1131 | compat_ulong_t, vlen) | |
1132 | { | |
1133 | return do_compat_readv(fd, vec, vlen, 0); | |
1134 | } | |
1135 | ||
1136 | static long do_compat_preadv64(unsigned long fd, | |
378a10f3 | 1137 | const struct compat_iovec __user *vec, |
f17d8b35 | 1138 | unsigned long vlen, loff_t pos, int flags) |
72ec3516 AV |
1139 | { |
1140 | struct fd f; | |
1141 | ssize_t ret; | |
1142 | ||
1143 | if (pos < 0) | |
1144 | return -EINVAL; | |
1145 | f = fdget(fd); | |
1146 | if (!f.file) | |
1147 | return -EBADF; | |
1148 | ret = -ESPIPE; | |
1149 | if (f.file->f_mode & FMODE_PREAD) | |
f17d8b35 | 1150 | ret = compat_readv(f.file, vec, vlen, &pos, flags); |
72ec3516 AV |
1151 | fdput(f); |
1152 | return ret; | |
1153 | } | |
1154 | ||
378a10f3 HC |
1155 | #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64 |
1156 | COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd, | |
1157 | const struct compat_iovec __user *,vec, | |
1158 | unsigned long, vlen, loff_t, pos) | |
1159 | { | |
f17d8b35 | 1160 | return do_compat_preadv64(fd, vec, vlen, pos, 0); |
378a10f3 HC |
1161 | } |
1162 | #endif | |
1163 | ||
dfd948e3 | 1164 | COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd, |
72ec3516 | 1165 | const struct compat_iovec __user *,vec, |
dfd948e3 | 1166 | compat_ulong_t, vlen, u32, pos_low, u32, pos_high) |
72ec3516 AV |
1167 | { |
1168 | loff_t pos = ((loff_t)pos_high << 32) | pos_low; | |
378a10f3 | 1169 | |
f17d8b35 MT |
1170 | return do_compat_preadv64(fd, vec, vlen, pos, 0); |
1171 | } | |
1172 | ||
1173 | COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd, | |
1174 | const struct compat_iovec __user *,vec, | |
1175 | compat_ulong_t, vlen, u32, pos_low, u32, pos_high, | |
1176 | int, flags) | |
1177 | { | |
1178 | loff_t pos = ((loff_t)pos_high << 32) | pos_low; | |
1179 | ||
1180 | if (pos == -1) | |
1181 | return do_compat_readv(fd, vec, vlen, flags); | |
1182 | ||
1183 | return do_compat_preadv64(fd, vec, vlen, pos, flags); | |
72ec3516 AV |
1184 | } |
1185 | ||
1186 | static size_t compat_writev(struct file *file, | |
1187 | const struct compat_iovec __user *vec, | |
f17d8b35 | 1188 | unsigned long vlen, loff_t *pos, int flags) |
72ec3516 AV |
1189 | { |
1190 | ssize_t ret = -EBADF; | |
1191 | ||
1192 | if (!(file->f_mode & FMODE_WRITE)) | |
1193 | goto out; | |
1194 | ||
1195 | ret = -EINVAL; | |
7f7f25e8 | 1196 | if (!(file->f_mode & FMODE_CAN_WRITE)) |
72ec3516 AV |
1197 | goto out; |
1198 | ||
793b80ef | 1199 | ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos, 0); |
72ec3516 AV |
1200 | |
1201 | out: | |
1202 | if (ret > 0) | |
1203 | add_wchar(current, ret); | |
1204 | inc_syscw(current); | |
1205 | return ret; | |
1206 | } | |
1207 | ||
f17d8b35 MT |
1208 | static size_t do_compat_writev(compat_ulong_t fd, |
1209 | const struct compat_iovec __user* vec, | |
1210 | compat_ulong_t vlen, int flags) | |
72ec3516 | 1211 | { |
9c225f26 | 1212 | struct fd f = fdget_pos(fd); |
72ec3516 AV |
1213 | ssize_t ret; |
1214 | loff_t pos; | |
1215 | ||
1216 | if (!f.file) | |
1217 | return -EBADF; | |
1218 | pos = f.file->f_pos; | |
f17d8b35 | 1219 | ret = compat_writev(f.file, vec, vlen, &pos, flags); |
5faf153e AV |
1220 | if (ret >= 0) |
1221 | f.file->f_pos = pos; | |
9c225f26 | 1222 | fdput_pos(f); |
72ec3516 AV |
1223 | return ret; |
1224 | } | |
1225 | ||
f17d8b35 MT |
1226 | COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd, |
1227 | const struct compat_iovec __user *, vec, | |
1228 | compat_ulong_t, vlen) | |
1229 | { | |
1230 | return do_compat_writev(fd, vec, vlen, 0); | |
1231 | } | |
1232 | ||
1233 | static long do_compat_pwritev64(unsigned long fd, | |
378a10f3 | 1234 | const struct compat_iovec __user *vec, |
f17d8b35 | 1235 | unsigned long vlen, loff_t pos, int flags) |
72ec3516 AV |
1236 | { |
1237 | struct fd f; | |
1238 | ssize_t ret; | |
1239 | ||
1240 | if (pos < 0) | |
1241 | return -EINVAL; | |
1242 | f = fdget(fd); | |
1243 | if (!f.file) | |
1244 | return -EBADF; | |
1245 | ret = -ESPIPE; | |
1246 | if (f.file->f_mode & FMODE_PWRITE) | |
f17d8b35 | 1247 | ret = compat_writev(f.file, vec, vlen, &pos, flags); |
72ec3516 AV |
1248 | fdput(f); |
1249 | return ret; | |
1250 | } | |
1251 | ||
378a10f3 HC |
1252 | #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64 |
1253 | COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd, | |
1254 | const struct compat_iovec __user *,vec, | |
1255 | unsigned long, vlen, loff_t, pos) | |
1256 | { | |
f17d8b35 | 1257 | return do_compat_pwritev64(fd, vec, vlen, pos, 0); |
378a10f3 HC |
1258 | } |
1259 | #endif | |
1260 | ||
dfd948e3 | 1261 | COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd, |
72ec3516 | 1262 | const struct compat_iovec __user *,vec, |
dfd948e3 | 1263 | compat_ulong_t, vlen, u32, pos_low, u32, pos_high) |
72ec3516 AV |
1264 | { |
1265 | loff_t pos = ((loff_t)pos_high << 32) | pos_low; | |
378a10f3 | 1266 | |
f17d8b35 | 1267 | return do_compat_pwritev64(fd, vec, vlen, pos, 0); |
72ec3516 | 1268 | } |
f17d8b35 MT |
1269 | |
1270 | COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd, | |
1271 | const struct compat_iovec __user *,vec, | |
1272 | compat_ulong_t, vlen, u32, pos_low, u32, pos_high, int, flags) | |
1273 | { | |
1274 | loff_t pos = ((loff_t)pos_high << 32) | pos_low; | |
1275 | ||
1276 | if (pos == -1) | |
1277 | return do_compat_writev(fd, vec, vlen, flags); | |
1278 | ||
1279 | return do_compat_pwritev64(fd, vec, vlen, pos, flags); | |
72ec3516 | 1280 | } |
f17d8b35 | 1281 | |
72ec3516 AV |
1282 | #endif |
1283 | ||
19f4fc3a AV |
1284 | static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos, |
1285 | size_t count, loff_t max) | |
1da177e4 | 1286 | { |
2903ff01 AV |
1287 | struct fd in, out; |
1288 | struct inode *in_inode, *out_inode; | |
1da177e4 | 1289 | loff_t pos; |
7995bd28 | 1290 | loff_t out_pos; |
1da177e4 | 1291 | ssize_t retval; |
2903ff01 | 1292 | int fl; |
1da177e4 LT |
1293 | |
1294 | /* | |
1295 | * Get input file, and verify that it is ok.. | |
1296 | */ | |
1297 | retval = -EBADF; | |
2903ff01 AV |
1298 | in = fdget(in_fd); |
1299 | if (!in.file) | |
1da177e4 | 1300 | goto out; |
2903ff01 | 1301 | if (!(in.file->f_mode & FMODE_READ)) |
1da177e4 | 1302 | goto fput_in; |
1da177e4 | 1303 | retval = -ESPIPE; |
7995bd28 AV |
1304 | if (!ppos) { |
1305 | pos = in.file->f_pos; | |
1306 | } else { | |
1307 | pos = *ppos; | |
2903ff01 | 1308 | if (!(in.file->f_mode & FMODE_PREAD)) |
1da177e4 | 1309 | goto fput_in; |
7995bd28 AV |
1310 | } |
1311 | retval = rw_verify_area(READ, in.file, &pos, count); | |
e28cc715 | 1312 | if (retval < 0) |
1da177e4 | 1313 | goto fput_in; |
e28cc715 | 1314 | count = retval; |
1da177e4 | 1315 | |
1da177e4 LT |
1316 | /* |
1317 | * Get output file, and verify that it is ok.. | |
1318 | */ | |
1319 | retval = -EBADF; | |
2903ff01 AV |
1320 | out = fdget(out_fd); |
1321 | if (!out.file) | |
1da177e4 | 1322 | goto fput_in; |
2903ff01 | 1323 | if (!(out.file->f_mode & FMODE_WRITE)) |
1da177e4 LT |
1324 | goto fput_out; |
1325 | retval = -EINVAL; | |
496ad9aa AV |
1326 | in_inode = file_inode(in.file); |
1327 | out_inode = file_inode(out.file); | |
7995bd28 AV |
1328 | out_pos = out.file->f_pos; |
1329 | retval = rw_verify_area(WRITE, out.file, &out_pos, count); | |
e28cc715 | 1330 | if (retval < 0) |
1da177e4 | 1331 | goto fput_out; |
e28cc715 | 1332 | count = retval; |
1da177e4 | 1333 | |
1da177e4 LT |
1334 | if (!max) |
1335 | max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes); | |
1336 | ||
1da177e4 LT |
1337 | if (unlikely(pos + count > max)) { |
1338 | retval = -EOVERFLOW; | |
1339 | if (pos >= max) | |
1340 | goto fput_out; | |
1341 | count = max - pos; | |
1342 | } | |
1343 | ||
d96e6e71 | 1344 | fl = 0; |
534f2aaa | 1345 | #if 0 |
d96e6e71 JA |
1346 | /* |
1347 | * We need to debate whether we can enable this or not. The | |
1348 | * man page documents EAGAIN return for the output at least, | |
1349 | * and the application is arguably buggy if it doesn't expect | |
1350 | * EAGAIN on a non-blocking file descriptor. | |
1351 | */ | |
2903ff01 | 1352 | if (in.file->f_flags & O_NONBLOCK) |
d96e6e71 | 1353 | fl = SPLICE_F_NONBLOCK; |
534f2aaa | 1354 | #endif |
50cd2c57 | 1355 | file_start_write(out.file); |
7995bd28 | 1356 | retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl); |
50cd2c57 | 1357 | file_end_write(out.file); |
1da177e4 LT |
1358 | |
1359 | if (retval > 0) { | |
4b98d11b AD |
1360 | add_rchar(current, retval); |
1361 | add_wchar(current, retval); | |
a68c2f12 SW |
1362 | fsnotify_access(in.file); |
1363 | fsnotify_modify(out.file); | |
7995bd28 AV |
1364 | out.file->f_pos = out_pos; |
1365 | if (ppos) | |
1366 | *ppos = pos; | |
1367 | else | |
1368 | in.file->f_pos = pos; | |
1da177e4 | 1369 | } |
1da177e4 | 1370 | |
4b98d11b AD |
1371 | inc_syscr(current); |
1372 | inc_syscw(current); | |
7995bd28 | 1373 | if (pos > max) |
1da177e4 LT |
1374 | retval = -EOVERFLOW; |
1375 | ||
1376 | fput_out: | |
2903ff01 | 1377 | fdput(out); |
1da177e4 | 1378 | fput_in: |
2903ff01 | 1379 | fdput(in); |
1da177e4 LT |
1380 | out: |
1381 | return retval; | |
1382 | } | |
1383 | ||
002c8976 | 1384 | SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count) |
1da177e4 LT |
1385 | { |
1386 | loff_t pos; | |
1387 | off_t off; | |
1388 | ssize_t ret; | |
1389 | ||
1390 | if (offset) { | |
1391 | if (unlikely(get_user(off, offset))) | |
1392 | return -EFAULT; | |
1393 | pos = off; | |
1394 | ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS); | |
1395 | if (unlikely(put_user(pos, offset))) | |
1396 | return -EFAULT; | |
1397 | return ret; | |
1398 | } | |
1399 | ||
1400 | return do_sendfile(out_fd, in_fd, NULL, count, 0); | |
1401 | } | |
1402 | ||
002c8976 | 1403 | SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count) |
1da177e4 LT |
1404 | { |
1405 | loff_t pos; | |
1406 | ssize_t ret; | |
1407 | ||
1408 | if (offset) { | |
1409 | if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t)))) | |
1410 | return -EFAULT; | |
1411 | ret = do_sendfile(out_fd, in_fd, &pos, count, 0); | |
1412 | if (unlikely(put_user(pos, offset))) | |
1413 | return -EFAULT; | |
1414 | return ret; | |
1415 | } | |
1416 | ||
1417 | return do_sendfile(out_fd, in_fd, NULL, count, 0); | |
1418 | } | |
19f4fc3a AV |
1419 | |
1420 | #ifdef CONFIG_COMPAT | |
1421 | COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, | |
1422 | compat_off_t __user *, offset, compat_size_t, count) | |
1423 | { | |
1424 | loff_t pos; | |
1425 | off_t off; | |
1426 | ssize_t ret; | |
1427 | ||
1428 | if (offset) { | |
1429 | if (unlikely(get_user(off, offset))) | |
1430 | return -EFAULT; | |
1431 | pos = off; | |
1432 | ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS); | |
1433 | if (unlikely(put_user(pos, offset))) | |
1434 | return -EFAULT; | |
1435 | return ret; | |
1436 | } | |
1437 | ||
1438 | return do_sendfile(out_fd, in_fd, NULL, count, 0); | |
1439 | } | |
1440 | ||
1441 | COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, | |
1442 | compat_loff_t __user *, offset, compat_size_t, count) | |
1443 | { | |
1444 | loff_t pos; | |
1445 | ssize_t ret; | |
1446 | ||
1447 | if (offset) { | |
1448 | if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t)))) | |
1449 | return -EFAULT; | |
1450 | ret = do_sendfile(out_fd, in_fd, &pos, count, 0); | |
1451 | if (unlikely(put_user(pos, offset))) | |
1452 | return -EFAULT; | |
1453 | return ret; | |
1454 | } | |
1455 | ||
1456 | return do_sendfile(out_fd, in_fd, NULL, count, 0); | |
1457 | } | |
1458 | #endif | |
29732938 ZB |
1459 | |
1460 | /* | |
1461 | * copy_file_range() differs from regular file read and write in that it | |
1462 | * specifically allows return partial success. When it does so is up to | |
1463 | * the copy_file_range method. | |
1464 | */ | |
1465 | ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in, | |
1466 | struct file *file_out, loff_t pos_out, | |
1467 | size_t len, unsigned int flags) | |
1468 | { | |
1469 | struct inode *inode_in = file_inode(file_in); | |
1470 | struct inode *inode_out = file_inode(file_out); | |
1471 | ssize_t ret; | |
1472 | ||
1473 | if (flags != 0) | |
1474 | return -EINVAL; | |
1475 | ||
1476 | /* copy_file_range allows full ssize_t len, ignoring MAX_RW_COUNT */ | |
1477 | ret = rw_verify_area(READ, file_in, &pos_in, len); | |
1478 | if (ret >= 0) | |
1479 | ret = rw_verify_area(WRITE, file_out, &pos_out, len); | |
1480 | if (ret < 0) | |
1481 | return ret; | |
1482 | ||
1483 | if (!(file_in->f_mode & FMODE_READ) || | |
1484 | !(file_out->f_mode & FMODE_WRITE) || | |
eac70053 | 1485 | (file_out->f_flags & O_APPEND)) |
29732938 ZB |
1486 | return -EBADF; |
1487 | ||
1488 | /* this could be relaxed once a method supports cross-fs copies */ | |
1489 | if (inode_in->i_sb != inode_out->i_sb) | |
1490 | return -EXDEV; | |
1491 | ||
1492 | if (len == 0) | |
1493 | return 0; | |
1494 | ||
1495 | ret = mnt_want_write_file(file_out); | |
1496 | if (ret) | |
1497 | return ret; | |
1498 | ||
eac70053 AS |
1499 | ret = -EOPNOTSUPP; |
1500 | if (file_out->f_op->copy_file_range) | |
1501 | ret = file_out->f_op->copy_file_range(file_in, pos_in, file_out, | |
1502 | pos_out, len, flags); | |
1503 | if (ret == -EOPNOTSUPP) | |
1504 | ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out, | |
1505 | len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0); | |
1506 | ||
29732938 ZB |
1507 | if (ret > 0) { |
1508 | fsnotify_access(file_in); | |
1509 | add_rchar(current, ret); | |
1510 | fsnotify_modify(file_out); | |
1511 | add_wchar(current, ret); | |
1512 | } | |
1513 | inc_syscr(current); | |
1514 | inc_syscw(current); | |
1515 | ||
1516 | mnt_drop_write_file(file_out); | |
1517 | ||
1518 | return ret; | |
1519 | } | |
1520 | EXPORT_SYMBOL(vfs_copy_file_range); | |
1521 | ||
1522 | SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in, | |
1523 | int, fd_out, loff_t __user *, off_out, | |
1524 | size_t, len, unsigned int, flags) | |
1525 | { | |
1526 | loff_t pos_in; | |
1527 | loff_t pos_out; | |
1528 | struct fd f_in; | |
1529 | struct fd f_out; | |
1530 | ssize_t ret = -EBADF; | |
1531 | ||
1532 | f_in = fdget(fd_in); | |
1533 | if (!f_in.file) | |
1534 | goto out2; | |
1535 | ||
1536 | f_out = fdget(fd_out); | |
1537 | if (!f_out.file) | |
1538 | goto out1; | |
1539 | ||
1540 | ret = -EFAULT; | |
1541 | if (off_in) { | |
1542 | if (copy_from_user(&pos_in, off_in, sizeof(loff_t))) | |
1543 | goto out; | |
1544 | } else { | |
1545 | pos_in = f_in.file->f_pos; | |
1546 | } | |
1547 | ||
1548 | if (off_out) { | |
1549 | if (copy_from_user(&pos_out, off_out, sizeof(loff_t))) | |
1550 | goto out; | |
1551 | } else { | |
1552 | pos_out = f_out.file->f_pos; | |
1553 | } | |
1554 | ||
1555 | ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len, | |
1556 | flags); | |
1557 | if (ret > 0) { | |
1558 | pos_in += ret; | |
1559 | pos_out += ret; | |
1560 | ||
1561 | if (off_in) { | |
1562 | if (copy_to_user(off_in, &pos_in, sizeof(loff_t))) | |
1563 | ret = -EFAULT; | |
1564 | } else { | |
1565 | f_in.file->f_pos = pos_in; | |
1566 | } | |
1567 | ||
1568 | if (off_out) { | |
1569 | if (copy_to_user(off_out, &pos_out, sizeof(loff_t))) | |
1570 | ret = -EFAULT; | |
1571 | } else { | |
1572 | f_out.file->f_pos = pos_out; | |
1573 | } | |
1574 | } | |
1575 | ||
1576 | out: | |
1577 | fdput(f_out); | |
1578 | out1: | |
1579 | fdput(f_in); | |
1580 | out2: | |
1581 | return ret; | |
1582 | } | |
04b38d60 CH |
1583 | |
1584 | static int clone_verify_area(struct file *file, loff_t pos, u64 len, bool write) | |
1585 | { | |
1586 | struct inode *inode = file_inode(file); | |
1587 | ||
1588 | if (unlikely(pos < 0)) | |
1589 | return -EINVAL; | |
1590 | ||
1591 | if (unlikely((loff_t) (pos + len) < 0)) | |
1592 | return -EINVAL; | |
1593 | ||
1594 | if (unlikely(inode->i_flctx && mandatory_lock(inode))) { | |
1595 | loff_t end = len ? pos + len - 1 : OFFSET_MAX; | |
1596 | int retval; | |
1597 | ||
1598 | retval = locks_mandatory_area(inode, file, pos, end, | |
1599 | write ? F_WRLCK : F_RDLCK); | |
1600 | if (retval < 0) | |
1601 | return retval; | |
1602 | } | |
1603 | ||
1604 | return security_file_permission(file, write ? MAY_WRITE : MAY_READ); | |
1605 | } | |
1606 | ||
1607 | int vfs_clone_file_range(struct file *file_in, loff_t pos_in, | |
1608 | struct file *file_out, loff_t pos_out, u64 len) | |
1609 | { | |
1610 | struct inode *inode_in = file_inode(file_in); | |
1611 | struct inode *inode_out = file_inode(file_out); | |
1612 | int ret; | |
1613 | ||
1614 | if (inode_in->i_sb != inode_out->i_sb || | |
1615 | file_in->f_path.mnt != file_out->f_path.mnt) | |
1616 | return -EXDEV; | |
1617 | ||
1618 | if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode)) | |
1619 | return -EISDIR; | |
1620 | if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode)) | |
d79bdd52 | 1621 | return -EINVAL; |
04b38d60 CH |
1622 | |
1623 | if (!(file_in->f_mode & FMODE_READ) || | |
1624 | !(file_out->f_mode & FMODE_WRITE) || | |
0fcbf996 | 1625 | (file_out->f_flags & O_APPEND)) |
04b38d60 CH |
1626 | return -EBADF; |
1627 | ||
0fcbf996 CH |
1628 | if (!file_in->f_op->clone_file_range) |
1629 | return -EOPNOTSUPP; | |
1630 | ||
04b38d60 CH |
1631 | ret = clone_verify_area(file_in, pos_in, len, false); |
1632 | if (ret) | |
1633 | return ret; | |
1634 | ||
1635 | ret = clone_verify_area(file_out, pos_out, len, true); | |
1636 | if (ret) | |
1637 | return ret; | |
1638 | ||
1639 | if (pos_in + len > i_size_read(inode_in)) | |
1640 | return -EINVAL; | |
1641 | ||
1642 | ret = mnt_want_write_file(file_out); | |
1643 | if (ret) | |
1644 | return ret; | |
1645 | ||
1646 | ret = file_in->f_op->clone_file_range(file_in, pos_in, | |
1647 | file_out, pos_out, len); | |
1648 | if (!ret) { | |
1649 | fsnotify_access(file_in); | |
1650 | fsnotify_modify(file_out); | |
1651 | } | |
1652 | ||
1653 | mnt_drop_write_file(file_out); | |
1654 | return ret; | |
1655 | } | |
1656 | EXPORT_SYMBOL(vfs_clone_file_range); | |
54dbc151 DW |
1657 | |
1658 | int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same) | |
1659 | { | |
1660 | struct file_dedupe_range_info *info; | |
1661 | struct inode *src = file_inode(file); | |
1662 | u64 off; | |
1663 | u64 len; | |
1664 | int i; | |
1665 | int ret; | |
1666 | bool is_admin = capable(CAP_SYS_ADMIN); | |
1667 | u16 count = same->dest_count; | |
1668 | struct file *dst_file; | |
1669 | loff_t dst_off; | |
1670 | ssize_t deduped; | |
1671 | ||
1672 | if (!(file->f_mode & FMODE_READ)) | |
1673 | return -EINVAL; | |
1674 | ||
1675 | if (same->reserved1 || same->reserved2) | |
1676 | return -EINVAL; | |
1677 | ||
1678 | off = same->src_offset; | |
1679 | len = same->src_length; | |
1680 | ||
1681 | ret = -EISDIR; | |
1682 | if (S_ISDIR(src->i_mode)) | |
1683 | goto out; | |
1684 | ||
1685 | ret = -EINVAL; | |
1686 | if (!S_ISREG(src->i_mode)) | |
1687 | goto out; | |
1688 | ||
1689 | ret = clone_verify_area(file, off, len, false); | |
1690 | if (ret < 0) | |
1691 | goto out; | |
1692 | ret = 0; | |
1693 | ||
1694 | /* pre-format output fields to sane values */ | |
1695 | for (i = 0; i < count; i++) { | |
1696 | same->info[i].bytes_deduped = 0ULL; | |
1697 | same->info[i].status = FILE_DEDUPE_RANGE_SAME; | |
1698 | } | |
1699 | ||
1700 | for (i = 0, info = same->info; i < count; i++, info++) { | |
1701 | struct inode *dst; | |
1702 | struct fd dst_fd = fdget(info->dest_fd); | |
1703 | ||
1704 | dst_file = dst_fd.file; | |
1705 | if (!dst_file) { | |
1706 | info->status = -EBADF; | |
1707 | goto next_loop; | |
1708 | } | |
1709 | dst = file_inode(dst_file); | |
1710 | ||
1711 | ret = mnt_want_write_file(dst_file); | |
1712 | if (ret) { | |
1713 | info->status = ret; | |
1714 | goto next_loop; | |
1715 | } | |
1716 | ||
1717 | dst_off = info->dest_offset; | |
1718 | ret = clone_verify_area(dst_file, dst_off, len, true); | |
1719 | if (ret < 0) { | |
1720 | info->status = ret; | |
1721 | goto next_file; | |
1722 | } | |
1723 | ret = 0; | |
1724 | ||
1725 | if (info->reserved) { | |
1726 | info->status = -EINVAL; | |
1727 | } else if (!(is_admin || (dst_file->f_mode & FMODE_WRITE))) { | |
1728 | info->status = -EINVAL; | |
1729 | } else if (file->f_path.mnt != dst_file->f_path.mnt) { | |
1730 | info->status = -EXDEV; | |
1731 | } else if (S_ISDIR(dst->i_mode)) { | |
1732 | info->status = -EISDIR; | |
1733 | } else if (dst_file->f_op->dedupe_file_range == NULL) { | |
1734 | info->status = -EINVAL; | |
1735 | } else { | |
1736 | deduped = dst_file->f_op->dedupe_file_range(file, off, | |
1737 | len, dst_file, | |
1738 | info->dest_offset); | |
1739 | if (deduped == -EBADE) | |
1740 | info->status = FILE_DEDUPE_RANGE_DIFFERS; | |
1741 | else if (deduped < 0) | |
1742 | info->status = deduped; | |
1743 | else | |
1744 | info->bytes_deduped += deduped; | |
1745 | } | |
1746 | ||
1747 | next_file: | |
1748 | mnt_drop_write_file(dst_file); | |
1749 | next_loop: | |
1750 | fdput(dst_fd); | |
e62e560f DW |
1751 | |
1752 | if (fatal_signal_pending(current)) | |
1753 | goto out; | |
54dbc151 DW |
1754 | } |
1755 | ||
1756 | out: | |
1757 | return ret; | |
1758 | } | |
1759 | EXPORT_SYMBOL(vfs_dedupe_file_range); |