Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
3 | * linux/fs/read_write.c | |
4 | * | |
5 | * Copyright (C) 1991, 1992 Linus Torvalds | |
6 | */ | |
7 | ||
b12fb7f4 | 8 | #include <linux/slab.h> |
1da177e4 | 9 | #include <linux/stat.h> |
b12fb7f4 | 10 | #include <linux/sched/xacct.h> |
1da177e4 LT |
11 | #include <linux/fcntl.h> |
12 | #include <linux/file.h> | |
13 | #include <linux/uio.h> | |
0eeca283 | 14 | #include <linux/fsnotify.h> |
1da177e4 | 15 | #include <linux/security.h> |
630d9c47 | 16 | #include <linux/export.h> |
1da177e4 | 17 | #include <linux/syscalls.h> |
e28cc715 | 18 | #include <linux/pagemap.h> |
d6b29d7c | 19 | #include <linux/splice.h> |
561c6731 | 20 | #include <linux/compat.h> |
29732938 | 21 | #include <linux/mount.h> |
2feb55f8 | 22 | #include <linux/fs.h> |
06ae43f3 | 23 | #include "internal.h" |
1da177e4 | 24 | |
7c0f6ba6 | 25 | #include <linux/uaccess.h> |
1da177e4 LT |
26 | #include <asm/unistd.h> |
27 | ||
4b6f5d20 | 28 | const struct file_operations generic_ro_fops = { |
1da177e4 | 29 | .llseek = generic_file_llseek, |
aad4f8bb | 30 | .read_iter = generic_file_read_iter, |
1da177e4 | 31 | .mmap = generic_file_readonly_mmap, |
534f2aaa | 32 | .splice_read = generic_file_splice_read, |
1da177e4 LT |
33 | }; |
34 | ||
35 | EXPORT_SYMBOL(generic_ro_fops); | |
36 | ||
ddef7ed2 | 37 | static inline bool unsigned_offsets(struct file *file) |
4a3956c7 | 38 | { |
cccb5a1e | 39 | return file->f_mode & FMODE_UNSIGNED_OFFSET; |
4a3956c7 KH |
40 | } |
41 | ||
46a1c2c7 JL |
42 | /** |
43 | * vfs_setpos - update the file offset for lseek | |
44 | * @file: file structure in question | |
45 | * @offset: file offset to seek to | |
46 | * @maxsize: maximum file size | |
47 | * | |
48 | * This is a low-level filesystem helper for updating the file offset to | |
49 | * the value specified by @offset if the given offset is valid and it is | |
50 | * not equal to the current file offset. | |
51 | * | |
52 | * Return the specified offset on success and -EINVAL on invalid offset. | |
53 | */ | |
54 | loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize) | |
ef3d0fd2 AK |
55 | { |
56 | if (offset < 0 && !unsigned_offsets(file)) | |
57 | return -EINVAL; | |
58 | if (offset > maxsize) | |
59 | return -EINVAL; | |
60 | ||
61 | if (offset != file->f_pos) { | |
62 | file->f_pos = offset; | |
63 | file->f_version = 0; | |
64 | } | |
65 | return offset; | |
66 | } | |
46a1c2c7 | 67 | EXPORT_SYMBOL(vfs_setpos); |
ef3d0fd2 | 68 | |
3a8cff4f | 69 | /** |
5760495a | 70 | * generic_file_llseek_size - generic llseek implementation for regular files |
3a8cff4f CH |
71 | * @file: file structure to seek on |
72 | * @offset: file offset to seek to | |
965c8e59 | 73 | * @whence: type of seek |
e8b96eb5 ES |
74 | * @size: max size of this file in file system |
75 | * @eof: offset used for SEEK_END position | |
3a8cff4f | 76 | * |
5760495a | 77 | * This is a variant of generic_file_llseek that allows passing in a custom |
e8b96eb5 | 78 | * maximum file size and a custom EOF position, for e.g. hashed directories |
ef3d0fd2 AK |
79 | * |
80 | * Synchronization: | |
5760495a | 81 | * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms) |
ef3d0fd2 AK |
82 | * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes. |
83 | * read/writes behave like SEEK_SET against seeks. | |
3a8cff4f | 84 | */ |
9465efc9 | 85 | loff_t |
965c8e59 | 86 | generic_file_llseek_size(struct file *file, loff_t offset, int whence, |
e8b96eb5 | 87 | loff_t maxsize, loff_t eof) |
1da177e4 | 88 | { |
965c8e59 | 89 | switch (whence) { |
3a8cff4f | 90 | case SEEK_END: |
e8b96eb5 | 91 | offset += eof; |
3a8cff4f CH |
92 | break; |
93 | case SEEK_CUR: | |
5b6f1eb9 AK |
94 | /* |
95 | * Here we special-case the lseek(fd, 0, SEEK_CUR) | |
96 | * position-querying operation. Avoid rewriting the "same" | |
97 | * f_pos value back to the file because a concurrent read(), | |
98 | * write() or lseek() might have altered it | |
99 | */ | |
100 | if (offset == 0) | |
101 | return file->f_pos; | |
ef3d0fd2 AK |
102 | /* |
103 | * f_lock protects against read/modify/write race with other | |
104 | * SEEK_CURs. Note that parallel writes and reads behave | |
105 | * like SEEK_SET. | |
106 | */ | |
107 | spin_lock(&file->f_lock); | |
46a1c2c7 | 108 | offset = vfs_setpos(file, file->f_pos + offset, maxsize); |
ef3d0fd2 AK |
109 | spin_unlock(&file->f_lock); |
110 | return offset; | |
982d8165 JB |
111 | case SEEK_DATA: |
112 | /* | |
113 | * In the generic case the entire file is data, so as long as | |
114 | * offset isn't at the end of the file then the offset is data. | |
115 | */ | |
fc46820b | 116 | if ((unsigned long long)offset >= eof) |
982d8165 JB |
117 | return -ENXIO; |
118 | break; | |
119 | case SEEK_HOLE: | |
120 | /* | |
121 | * There is a virtual hole at the end of the file, so as long as | |
122 | * offset isn't i_size or larger, return i_size. | |
123 | */ | |
fc46820b | 124 | if ((unsigned long long)offset >= eof) |
982d8165 | 125 | return -ENXIO; |
e8b96eb5 | 126 | offset = eof; |
982d8165 | 127 | break; |
1da177e4 | 128 | } |
3a8cff4f | 129 | |
46a1c2c7 | 130 | return vfs_setpos(file, offset, maxsize); |
5760495a AK |
131 | } |
132 | EXPORT_SYMBOL(generic_file_llseek_size); | |
133 | ||
134 | /** | |
135 | * generic_file_llseek - generic llseek implementation for regular files | |
136 | * @file: file structure to seek on | |
137 | * @offset: file offset to seek to | |
965c8e59 | 138 | * @whence: type of seek |
5760495a AK |
139 | * |
140 | * This is a generic implemenation of ->llseek useable for all normal local | |
141 | * filesystems. It just updates the file offset to the value specified by | |
546ae2d2 | 142 | * @offset and @whence. |
5760495a | 143 | */ |
965c8e59 | 144 | loff_t generic_file_llseek(struct file *file, loff_t offset, int whence) |
5760495a AK |
145 | { |
146 | struct inode *inode = file->f_mapping->host; | |
147 | ||
965c8e59 | 148 | return generic_file_llseek_size(file, offset, whence, |
e8b96eb5 ES |
149 | inode->i_sb->s_maxbytes, |
150 | i_size_read(inode)); | |
1da177e4 | 151 | } |
9465efc9 | 152 | EXPORT_SYMBOL(generic_file_llseek); |
1da177e4 | 153 | |
1bf9d14d AV |
154 | /** |
155 | * fixed_size_llseek - llseek implementation for fixed-sized devices | |
156 | * @file: file structure to seek on | |
157 | * @offset: file offset to seek to | |
158 | * @whence: type of seek | |
159 | * @size: size of the file | |
160 | * | |
161 | */ | |
162 | loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size) | |
163 | { | |
164 | switch (whence) { | |
165 | case SEEK_SET: case SEEK_CUR: case SEEK_END: | |
166 | return generic_file_llseek_size(file, offset, whence, | |
167 | size, size); | |
168 | default: | |
169 | return -EINVAL; | |
170 | } | |
171 | } | |
172 | EXPORT_SYMBOL(fixed_size_llseek); | |
173 | ||
b25472f9 AV |
174 | /** |
175 | * no_seek_end_llseek - llseek implementation for fixed-sized devices | |
176 | * @file: file structure to seek on | |
177 | * @offset: file offset to seek to | |
178 | * @whence: type of seek | |
179 | * | |
180 | */ | |
181 | loff_t no_seek_end_llseek(struct file *file, loff_t offset, int whence) | |
182 | { | |
183 | switch (whence) { | |
184 | case SEEK_SET: case SEEK_CUR: | |
185 | return generic_file_llseek_size(file, offset, whence, | |
2feb55f8 | 186 | OFFSET_MAX, 0); |
b25472f9 AV |
187 | default: |
188 | return -EINVAL; | |
189 | } | |
190 | } | |
191 | EXPORT_SYMBOL(no_seek_end_llseek); | |
192 | ||
193 | /** | |
194 | * no_seek_end_llseek_size - llseek implementation for fixed-sized devices | |
195 | * @file: file structure to seek on | |
196 | * @offset: file offset to seek to | |
197 | * @whence: type of seek | |
198 | * @size: maximal offset allowed | |
199 | * | |
200 | */ | |
201 | loff_t no_seek_end_llseek_size(struct file *file, loff_t offset, int whence, loff_t size) | |
202 | { | |
203 | switch (whence) { | |
204 | case SEEK_SET: case SEEK_CUR: | |
205 | return generic_file_llseek_size(file, offset, whence, | |
206 | size, 0); | |
207 | default: | |
208 | return -EINVAL; | |
209 | } | |
210 | } | |
211 | EXPORT_SYMBOL(no_seek_end_llseek_size); | |
212 | ||
ae6afc3f B |
213 | /** |
214 | * noop_llseek - No Operation Performed llseek implementation | |
215 | * @file: file structure to seek on | |
216 | * @offset: file offset to seek to | |
965c8e59 | 217 | * @whence: type of seek |
ae6afc3f B |
218 | * |
219 | * This is an implementation of ->llseek useable for the rare special case when | |
220 | * userspace expects the seek to succeed but the (device) file is actually not | |
221 | * able to perform the seek. In this case you use noop_llseek() instead of | |
222 | * falling back to the default implementation of ->llseek. | |
223 | */ | |
965c8e59 | 224 | loff_t noop_llseek(struct file *file, loff_t offset, int whence) |
ae6afc3f B |
225 | { |
226 | return file->f_pos; | |
227 | } | |
228 | EXPORT_SYMBOL(noop_llseek); | |
229 | ||
965c8e59 | 230 | loff_t no_llseek(struct file *file, loff_t offset, int whence) |
1da177e4 LT |
231 | { |
232 | return -ESPIPE; | |
233 | } | |
234 | EXPORT_SYMBOL(no_llseek); | |
235 | ||
965c8e59 | 236 | loff_t default_llseek(struct file *file, loff_t offset, int whence) |
1da177e4 | 237 | { |
496ad9aa | 238 | struct inode *inode = file_inode(file); |
16abef0e | 239 | loff_t retval; |
1da177e4 | 240 | |
5955102c | 241 | inode_lock(inode); |
965c8e59 | 242 | switch (whence) { |
7b8e8924 | 243 | case SEEK_END: |
982d8165 | 244 | offset += i_size_read(inode); |
1da177e4 | 245 | break; |
7b8e8924 | 246 | case SEEK_CUR: |
5b6f1eb9 AK |
247 | if (offset == 0) { |
248 | retval = file->f_pos; | |
249 | goto out; | |
250 | } | |
1da177e4 | 251 | offset += file->f_pos; |
982d8165 JB |
252 | break; |
253 | case SEEK_DATA: | |
254 | /* | |
255 | * In the generic case the entire file is data, so as | |
256 | * long as offset isn't at the end of the file then the | |
257 | * offset is data. | |
258 | */ | |
bacb2d81 DC |
259 | if (offset >= inode->i_size) { |
260 | retval = -ENXIO; | |
261 | goto out; | |
262 | } | |
982d8165 JB |
263 | break; |
264 | case SEEK_HOLE: | |
265 | /* | |
266 | * There is a virtual hole at the end of the file, so | |
267 | * as long as offset isn't i_size or larger, return | |
268 | * i_size. | |
269 | */ | |
bacb2d81 DC |
270 | if (offset >= inode->i_size) { |
271 | retval = -ENXIO; | |
272 | goto out; | |
273 | } | |
982d8165 JB |
274 | offset = inode->i_size; |
275 | break; | |
1da177e4 LT |
276 | } |
277 | retval = -EINVAL; | |
cccb5a1e | 278 | if (offset >= 0 || unsigned_offsets(file)) { |
1da177e4 LT |
279 | if (offset != file->f_pos) { |
280 | file->f_pos = offset; | |
281 | file->f_version = 0; | |
282 | } | |
283 | retval = offset; | |
284 | } | |
5b6f1eb9 | 285 | out: |
5955102c | 286 | inode_unlock(inode); |
1da177e4 LT |
287 | return retval; |
288 | } | |
289 | EXPORT_SYMBOL(default_llseek); | |
290 | ||
965c8e59 | 291 | loff_t vfs_llseek(struct file *file, loff_t offset, int whence) |
1da177e4 LT |
292 | { |
293 | loff_t (*fn)(struct file *, loff_t, int); | |
294 | ||
295 | fn = no_llseek; | |
296 | if (file->f_mode & FMODE_LSEEK) { | |
72c2d531 | 297 | if (file->f_op->llseek) |
1da177e4 LT |
298 | fn = file->f_op->llseek; |
299 | } | |
965c8e59 | 300 | return fn(file, offset, whence); |
1da177e4 LT |
301 | } |
302 | EXPORT_SYMBOL(vfs_llseek); | |
303 | ||
bef17329 | 304 | static off_t ksys_lseek(unsigned int fd, off_t offset, unsigned int whence) |
1da177e4 LT |
305 | { |
306 | off_t retval; | |
9c225f26 | 307 | struct fd f = fdget_pos(fd); |
2903ff01 AV |
308 | if (!f.file) |
309 | return -EBADF; | |
1da177e4 LT |
310 | |
311 | retval = -EINVAL; | |
965c8e59 AM |
312 | if (whence <= SEEK_MAX) { |
313 | loff_t res = vfs_llseek(f.file, offset, whence); | |
1da177e4 LT |
314 | retval = res; |
315 | if (res != (loff_t)retval) | |
316 | retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */ | |
317 | } | |
9c225f26 | 318 | fdput_pos(f); |
1da177e4 LT |
319 | return retval; |
320 | } | |
321 | ||
76847e43 DB |
322 | SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence) |
323 | { | |
324 | return ksys_lseek(fd, offset, whence); | |
325 | } | |
326 | ||
561c6731 AV |
327 | #ifdef CONFIG_COMPAT |
328 | COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence) | |
329 | { | |
76847e43 | 330 | return ksys_lseek(fd, offset, whence); |
561c6731 AV |
331 | } |
332 | #endif | |
333 | ||
9e62ccec MS |
334 | #if !defined(CONFIG_64BIT) || defined(CONFIG_COMPAT) || \ |
335 | defined(__ARCH_WANT_SYS_LLSEEK) | |
003d7ab4 HC |
336 | SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high, |
337 | unsigned long, offset_low, loff_t __user *, result, | |
965c8e59 | 338 | unsigned int, whence) |
1da177e4 LT |
339 | { |
340 | int retval; | |
d7a15f8d | 341 | struct fd f = fdget_pos(fd); |
1da177e4 | 342 | loff_t offset; |
1da177e4 | 343 | |
2903ff01 AV |
344 | if (!f.file) |
345 | return -EBADF; | |
1da177e4 LT |
346 | |
347 | retval = -EINVAL; | |
965c8e59 | 348 | if (whence > SEEK_MAX) |
1da177e4 LT |
349 | goto out_putf; |
350 | ||
2903ff01 | 351 | offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low, |
965c8e59 | 352 | whence); |
1da177e4 LT |
353 | |
354 | retval = (int)offset; | |
355 | if (offset >= 0) { | |
356 | retval = -EFAULT; | |
357 | if (!copy_to_user(result, &offset, sizeof(offset))) | |
358 | retval = 0; | |
359 | } | |
360 | out_putf: | |
d7a15f8d | 361 | fdput_pos(f); |
1da177e4 LT |
362 | return retval; |
363 | } | |
364 | #endif | |
365 | ||
68d70d03 | 366 | int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count) |
1da177e4 | 367 | { |
e28cc715 | 368 | if (unlikely((ssize_t) count < 0)) |
2949e842 | 369 | return -EINVAL; |
1da177e4 | 370 | |
438ab720 KS |
371 | if (ppos) { |
372 | loff_t pos = *ppos; | |
373 | ||
374 | if (unlikely(pos < 0)) { | |
375 | if (!unsigned_offsets(file)) | |
2949e842 | 376 | return -EINVAL; |
438ab720 KS |
377 | if (count >= -pos) /* both values are in 0..LLONG_MAX */ |
378 | return -EOVERFLOW; | |
379 | } else if (unlikely((loff_t) (pos + count) < 0)) { | |
380 | if (!unsigned_offsets(file)) | |
2949e842 | 381 | return -EINVAL; |
438ab720 | 382 | } |
e28cc715 | 383 | } |
438ab720 | 384 | |
bc61384d | 385 | return security_file_permission(file, |
c43e259c | 386 | read_write == READ ? MAY_READ : MAY_WRITE); |
1da177e4 LT |
387 | } |
388 | ||
5d5d5689 | 389 | static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos) |
293bc982 AV |
390 | { |
391 | struct iovec iov = { .iov_base = buf, .iov_len = len }; | |
392 | struct kiocb kiocb; | |
393 | struct iov_iter iter; | |
394 | ssize_t ret; | |
395 | ||
396 | init_sync_kiocb(&kiocb, filp); | |
438ab720 | 397 | kiocb.ki_pos = (ppos ? *ppos : 0); |
293bc982 AV |
398 | iov_iter_init(&iter, READ, &iov, 1, len); |
399 | ||
bb7462b6 | 400 | ret = call_read_iter(filp, &kiocb, &iter); |
599bd19b | 401 | BUG_ON(ret == -EIOCBQUEUED); |
438ab720 KS |
402 | if (ppos) |
403 | *ppos = kiocb.ki_pos; | |
293bc982 AV |
404 | return ret; |
405 | } | |
406 | ||
4d03e3cc CH |
407 | static int warn_unsupported(struct file *file, const char *op) |
408 | { | |
409 | pr_warn_ratelimited( | |
410 | "kernel %s not supported for file %pD4 (pid: %d comm: %.20s)\n", | |
411 | op, file, current->pid, current->comm); | |
412 | return -EINVAL; | |
413 | } | |
414 | ||
61a707c5 CH |
415 | ssize_t __kernel_read(struct file *file, void *buf, size_t count, loff_t *pos) |
416 | { | |
4d03e3cc CH |
417 | struct kvec iov = { |
418 | .iov_base = buf, | |
419 | .iov_len = min_t(size_t, count, MAX_RW_COUNT), | |
420 | }; | |
421 | struct kiocb kiocb; | |
422 | struct iov_iter iter; | |
61a707c5 CH |
423 | ssize_t ret; |
424 | ||
425 | if (WARN_ON_ONCE(!(file->f_mode & FMODE_READ))) | |
426 | return -EINVAL; | |
427 | if (!(file->f_mode & FMODE_CAN_READ)) | |
428 | return -EINVAL; | |
4d03e3cc CH |
429 | /* |
430 | * Also fail if ->read_iter and ->read are both wired up as that | |
431 | * implies very convoluted semantics. | |
432 | */ | |
433 | if (unlikely(!file->f_op->read_iter || file->f_op->read)) | |
434 | return warn_unsupported(file, "read"); | |
61a707c5 | 435 | |
4d03e3cc | 436 | init_sync_kiocb(&kiocb, file); |
7b84b665 | 437 | kiocb.ki_pos = pos ? *pos : 0; |
4d03e3cc CH |
438 | iov_iter_kvec(&iter, READ, &iov, 1, iov.iov_len); |
439 | ret = file->f_op->read_iter(&kiocb, &iter); | |
61a707c5 | 440 | if (ret > 0) { |
7b84b665 MWO |
441 | if (pos) |
442 | *pos = kiocb.ki_pos; | |
61a707c5 CH |
443 | fsnotify_access(file); |
444 | add_rchar(current, ret); | |
445 | } | |
446 | inc_syscr(current); | |
447 | return ret; | |
448 | } | |
449 | ||
bdd1d2d3 | 450 | ssize_t kernel_read(struct file *file, void *buf, size_t count, loff_t *pos) |
c41fbad0 | 451 | { |
6209dd91 | 452 | ssize_t ret; |
c41fbad0 | 453 | |
6209dd91 CH |
454 | ret = rw_verify_area(READ, file, pos, count); |
455 | if (ret) | |
456 | return ret; | |
457 | return __kernel_read(file, buf, count, pos); | |
c41fbad0 CH |
458 | } |
459 | EXPORT_SYMBOL(kernel_read); | |
6fb5032e | 460 | |
1da177e4 LT |
461 | ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos) |
462 | { | |
463 | ssize_t ret; | |
464 | ||
465 | if (!(file->f_mode & FMODE_READ)) | |
466 | return -EBADF; | |
7f7f25e8 | 467 | if (!(file->f_mode & FMODE_CAN_READ)) |
1da177e4 | 468 | return -EINVAL; |
96d4f267 | 469 | if (unlikely(!access_ok(buf, count))) |
1da177e4 LT |
470 | return -EFAULT; |
471 | ||
472 | ret = rw_verify_area(READ, file, pos, count); | |
775802c0 CH |
473 | if (ret) |
474 | return ret; | |
475 | if (count > MAX_RW_COUNT) | |
476 | count = MAX_RW_COUNT; | |
1da177e4 | 477 | |
775802c0 CH |
478 | if (file->f_op->read) |
479 | ret = file->f_op->read(file, buf, count, pos); | |
480 | else if (file->f_op->read_iter) | |
481 | ret = new_sync_read(file, buf, count, pos); | |
482 | else | |
483 | ret = -EINVAL; | |
484 | if (ret > 0) { | |
485 | fsnotify_access(file); | |
486 | add_rchar(current, ret); | |
487 | } | |
488 | inc_syscr(current); | |
1da177e4 LT |
489 | return ret; |
490 | } | |
491 | ||
5d5d5689 | 492 | static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos) |
293bc982 AV |
493 | { |
494 | struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len }; | |
495 | struct kiocb kiocb; | |
496 | struct iov_iter iter; | |
497 | ssize_t ret; | |
498 | ||
499 | init_sync_kiocb(&kiocb, filp); | |
438ab720 | 500 | kiocb.ki_pos = (ppos ? *ppos : 0); |
293bc982 AV |
501 | iov_iter_init(&iter, WRITE, &iov, 1, len); |
502 | ||
bb7462b6 | 503 | ret = call_write_iter(filp, &kiocb, &iter); |
599bd19b | 504 | BUG_ON(ret == -EIOCBQUEUED); |
438ab720 | 505 | if (ret > 0 && ppos) |
f765b134 | 506 | *ppos = kiocb.ki_pos; |
293bc982 AV |
507 | return ret; |
508 | } | |
509 | ||
81238b2c | 510 | /* caller is responsible for file_start_write/file_end_write */ |
73e18f7c | 511 | ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t *pos) |
06ae43f3 | 512 | { |
4d03e3cc CH |
513 | struct kvec iov = { |
514 | .iov_base = (void *)buf, | |
515 | .iov_len = min_t(size_t, count, MAX_RW_COUNT), | |
516 | }; | |
517 | struct kiocb kiocb; | |
518 | struct iov_iter iter; | |
06ae43f3 AV |
519 | ssize_t ret; |
520 | ||
a01ac27b CH |
521 | if (WARN_ON_ONCE(!(file->f_mode & FMODE_WRITE))) |
522 | return -EBADF; | |
7f7f25e8 | 523 | if (!(file->f_mode & FMODE_CAN_WRITE)) |
3e84f48e | 524 | return -EINVAL; |
4d03e3cc CH |
525 | /* |
526 | * Also fail if ->write_iter and ->write are both wired up as that | |
527 | * implies very convoluted semantics. | |
528 | */ | |
529 | if (unlikely(!file->f_op->write_iter || file->f_op->write)) | |
530 | return warn_unsupported(file, "write"); | |
3e84f48e | 531 | |
4d03e3cc | 532 | init_sync_kiocb(&kiocb, file); |
4c207ef4 | 533 | kiocb.ki_pos = pos ? *pos : 0; |
4d03e3cc CH |
534 | iov_iter_kvec(&iter, WRITE, &iov, 1, iov.iov_len); |
535 | ret = file->f_op->write_iter(&kiocb, &iter); | |
06ae43f3 | 536 | if (ret > 0) { |
4c207ef4 MWO |
537 | if (pos) |
538 | *pos = kiocb.ki_pos; | |
06ae43f3 AV |
539 | fsnotify_modify(file); |
540 | add_wchar(current, ret); | |
541 | } | |
542 | inc_syscw(current); | |
543 | return ret; | |
544 | } | |
90fb7027 LT |
545 | /* |
546 | * This "EXPORT_SYMBOL_GPL()" is more of a "EXPORT_SYMBOL_DONTUSE()", | |
547 | * but autofs is one of the few internal kernel users that actually | |
548 | * wants this _and_ can be built as a module. So we need to export | |
549 | * this symbol for autofs, even though it really isn't appropriate | |
550 | * for any other kernel modules. | |
551 | */ | |
552 | EXPORT_SYMBOL_GPL(__kernel_write); | |
2ec3a12a | 553 | |
e13ec939 CH |
554 | ssize_t kernel_write(struct file *file, const void *buf, size_t count, |
555 | loff_t *pos) | |
ac452aca | 556 | { |
81238b2c | 557 | ssize_t ret; |
ac452aca | 558 | |
81238b2c CH |
559 | ret = rw_verify_area(WRITE, file, pos, count); |
560 | if (ret) | |
561 | return ret; | |
ac452aca | 562 | |
81238b2c CH |
563 | file_start_write(file); |
564 | ret = __kernel_write(file, buf, count, pos); | |
565 | file_end_write(file); | |
566 | return ret; | |
ac452aca CH |
567 | } |
568 | EXPORT_SYMBOL(kernel_write); | |
569 | ||
1da177e4 LT |
570 | ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos) |
571 | { | |
572 | ssize_t ret; | |
573 | ||
574 | if (!(file->f_mode & FMODE_WRITE)) | |
575 | return -EBADF; | |
7f7f25e8 | 576 | if (!(file->f_mode & FMODE_CAN_WRITE)) |
1da177e4 | 577 | return -EINVAL; |
96d4f267 | 578 | if (unlikely(!access_ok(buf, count))) |
1da177e4 LT |
579 | return -EFAULT; |
580 | ||
581 | ret = rw_verify_area(WRITE, file, pos, count); | |
53ad8626 CH |
582 | if (ret) |
583 | return ret; | |
584 | if (count > MAX_RW_COUNT) | |
585 | count = MAX_RW_COUNT; | |
586 | file_start_write(file); | |
587 | if (file->f_op->write) | |
588 | ret = file->f_op->write(file, buf, count, pos); | |
589 | else if (file->f_op->write_iter) | |
590 | ret = new_sync_write(file, buf, count, pos); | |
591 | else | |
592 | ret = -EINVAL; | |
593 | if (ret > 0) { | |
594 | fsnotify_modify(file); | |
595 | add_wchar(current, ret); | |
1da177e4 | 596 | } |
53ad8626 CH |
597 | inc_syscw(current); |
598 | file_end_write(file); | |
1da177e4 LT |
599 | return ret; |
600 | } | |
601 | ||
438ab720 KS |
602 | /* file_ppos returns &file->f_pos or NULL if file is stream */ |
603 | static inline loff_t *file_ppos(struct file *file) | |
1da177e4 | 604 | { |
438ab720 | 605 | return file->f_mode & FMODE_STREAM ? NULL : &file->f_pos; |
1da177e4 LT |
606 | } |
607 | ||
3ce4a7bf | 608 | ssize_t ksys_read(unsigned int fd, char __user *buf, size_t count) |
1da177e4 | 609 | { |
9c225f26 | 610 | struct fd f = fdget_pos(fd); |
1da177e4 | 611 | ssize_t ret = -EBADF; |
1da177e4 | 612 | |
2903ff01 | 613 | if (f.file) { |
438ab720 KS |
614 | loff_t pos, *ppos = file_ppos(f.file); |
615 | if (ppos) { | |
616 | pos = *ppos; | |
617 | ppos = &pos; | |
618 | } | |
619 | ret = vfs_read(f.file, buf, count, ppos); | |
620 | if (ret >= 0 && ppos) | |
621 | f.file->f_pos = pos; | |
9c225f26 | 622 | fdput_pos(f); |
1da177e4 | 623 | } |
1da177e4 LT |
624 | return ret; |
625 | } | |
1da177e4 | 626 | |
3ce4a7bf DB |
627 | SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count) |
628 | { | |
629 | return ksys_read(fd, buf, count); | |
630 | } | |
631 | ||
e7a3e8b2 | 632 | ssize_t ksys_write(unsigned int fd, const char __user *buf, size_t count) |
1da177e4 | 633 | { |
9c225f26 | 634 | struct fd f = fdget_pos(fd); |
1da177e4 | 635 | ssize_t ret = -EBADF; |
1da177e4 | 636 | |
2903ff01 | 637 | if (f.file) { |
438ab720 KS |
638 | loff_t pos, *ppos = file_ppos(f.file); |
639 | if (ppos) { | |
640 | pos = *ppos; | |
641 | ppos = &pos; | |
642 | } | |
643 | ret = vfs_write(f.file, buf, count, ppos); | |
644 | if (ret >= 0 && ppos) | |
645 | f.file->f_pos = pos; | |
9c225f26 | 646 | fdput_pos(f); |
1da177e4 LT |
647 | } |
648 | ||
649 | return ret; | |
650 | } | |
651 | ||
e7a3e8b2 DB |
652 | SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf, |
653 | size_t, count) | |
654 | { | |
655 | return ksys_write(fd, buf, count); | |
656 | } | |
657 | ||
36028d5d DB |
658 | ssize_t ksys_pread64(unsigned int fd, char __user *buf, size_t count, |
659 | loff_t pos) | |
1da177e4 | 660 | { |
2903ff01 | 661 | struct fd f; |
1da177e4 | 662 | ssize_t ret = -EBADF; |
1da177e4 LT |
663 | |
664 | if (pos < 0) | |
665 | return -EINVAL; | |
666 | ||
2903ff01 AV |
667 | f = fdget(fd); |
668 | if (f.file) { | |
1da177e4 | 669 | ret = -ESPIPE; |
2903ff01 AV |
670 | if (f.file->f_mode & FMODE_PREAD) |
671 | ret = vfs_read(f.file, buf, count, &pos); | |
672 | fdput(f); | |
1da177e4 LT |
673 | } |
674 | ||
675 | return ret; | |
676 | } | |
677 | ||
36028d5d DB |
678 | SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf, |
679 | size_t, count, loff_t, pos) | |
680 | { | |
681 | return ksys_pread64(fd, buf, count, pos); | |
682 | } | |
683 | ||
684 | ssize_t ksys_pwrite64(unsigned int fd, const char __user *buf, | |
685 | size_t count, loff_t pos) | |
1da177e4 | 686 | { |
2903ff01 | 687 | struct fd f; |
1da177e4 | 688 | ssize_t ret = -EBADF; |
1da177e4 LT |
689 | |
690 | if (pos < 0) | |
691 | return -EINVAL; | |
692 | ||
2903ff01 AV |
693 | f = fdget(fd); |
694 | if (f.file) { | |
1da177e4 | 695 | ret = -ESPIPE; |
2903ff01 AV |
696 | if (f.file->f_mode & FMODE_PWRITE) |
697 | ret = vfs_write(f.file, buf, count, &pos); | |
698 | fdput(f); | |
1da177e4 LT |
699 | } |
700 | ||
701 | return ret; | |
702 | } | |
703 | ||
36028d5d DB |
704 | SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf, |
705 | size_t, count, loff_t, pos) | |
706 | { | |
707 | return ksys_pwrite64(fd, buf, count, pos); | |
708 | } | |
709 | ||
ac15ac06 | 710 | static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter, |
ddef7ed2 | 711 | loff_t *ppos, int type, rwf_t flags) |
293bc982 AV |
712 | { |
713 | struct kiocb kiocb; | |
293bc982 AV |
714 | ssize_t ret; |
715 | ||
716 | init_sync_kiocb(&kiocb, filp); | |
fdd2f5b7 GR |
717 | ret = kiocb_set_rw_flags(&kiocb, flags); |
718 | if (ret) | |
719 | return ret; | |
438ab720 | 720 | kiocb.ki_pos = (ppos ? *ppos : 0); |
293bc982 | 721 | |
0f78d06a | 722 | if (type == READ) |
bb7462b6 | 723 | ret = call_read_iter(filp, &kiocb, iter); |
0f78d06a | 724 | else |
bb7462b6 | 725 | ret = call_write_iter(filp, &kiocb, iter); |
599bd19b | 726 | BUG_ON(ret == -EIOCBQUEUED); |
438ab720 KS |
727 | if (ppos) |
728 | *ppos = kiocb.ki_pos; | |
293bc982 AV |
729 | return ret; |
730 | } | |
731 | ||
ee0b3e67 | 732 | /* Do it by hand, with file-ops */ |
ac15ac06 | 733 | static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter, |
ddef7ed2 | 734 | loff_t *ppos, int type, rwf_t flags) |
ee0b3e67 | 735 | { |
ee0b3e67 BP |
736 | ssize_t ret = 0; |
737 | ||
97be7ebe | 738 | if (flags & ~RWF_HIPRI) |
793b80ef CH |
739 | return -EOPNOTSUPP; |
740 | ||
ac15ac06 AV |
741 | while (iov_iter_count(iter)) { |
742 | struct iovec iovec = iov_iter_iovec(iter); | |
ee0b3e67 BP |
743 | ssize_t nr; |
744 | ||
0f78d06a MS |
745 | if (type == READ) { |
746 | nr = filp->f_op->read(filp, iovec.iov_base, | |
747 | iovec.iov_len, ppos); | |
748 | } else { | |
749 | nr = filp->f_op->write(filp, iovec.iov_base, | |
750 | iovec.iov_len, ppos); | |
751 | } | |
ee0b3e67 BP |
752 | |
753 | if (nr < 0) { | |
754 | if (!ret) | |
755 | ret = nr; | |
756 | break; | |
757 | } | |
758 | ret += nr; | |
ac15ac06 | 759 | if (nr != iovec.iov_len) |
ee0b3e67 | 760 | break; |
ac15ac06 | 761 | iov_iter_advance(iter, nr); |
ee0b3e67 BP |
762 | } |
763 | ||
764 | return ret; | |
765 | } | |
766 | ||
19c73586 | 767 | static ssize_t do_iter_read(struct file *file, struct iov_iter *iter, |
ddef7ed2 | 768 | loff_t *pos, rwf_t flags) |
1da177e4 | 769 | { |
1da177e4 | 770 | size_t tot_len; |
7687a7a4 | 771 | ssize_t ret = 0; |
1da177e4 | 772 | |
edab5fe3 CH |
773 | if (!(file->f_mode & FMODE_READ)) |
774 | return -EBADF; | |
775 | if (!(file->f_mode & FMODE_CAN_READ)) | |
776 | return -EINVAL; | |
777 | ||
7687a7a4 | 778 | tot_len = iov_iter_count(iter); |
0504c074 AV |
779 | if (!tot_len) |
780 | goto out; | |
19c73586 | 781 | ret = rw_verify_area(READ, file, pos, tot_len); |
e28cc715 | 782 | if (ret < 0) |
19c73586 | 783 | return ret; |
1da177e4 | 784 | |
19c73586 CH |
785 | if (file->f_op->read_iter) |
786 | ret = do_iter_readv_writev(file, iter, pos, READ, flags); | |
ee0b3e67 | 787 | else |
19c73586 | 788 | ret = do_loop_readv_writev(file, iter, pos, READ, flags); |
1da177e4 | 789 | out: |
19c73586 CH |
790 | if (ret >= 0) |
791 | fsnotify_access(file); | |
1da177e4 | 792 | return ret; |
1da177e4 LT |
793 | } |
794 | ||
5dcdc43e JX |
795 | ssize_t vfs_iocb_iter_read(struct file *file, struct kiocb *iocb, |
796 | struct iov_iter *iter) | |
797 | { | |
798 | size_t tot_len; | |
799 | ssize_t ret = 0; | |
800 | ||
801 | if (!file->f_op->read_iter) | |
802 | return -EINVAL; | |
803 | if (!(file->f_mode & FMODE_READ)) | |
804 | return -EBADF; | |
805 | if (!(file->f_mode & FMODE_CAN_READ)) | |
806 | return -EINVAL; | |
807 | ||
808 | tot_len = iov_iter_count(iter); | |
809 | if (!tot_len) | |
810 | goto out; | |
811 | ret = rw_verify_area(READ, file, &iocb->ki_pos, tot_len); | |
812 | if (ret < 0) | |
813 | return ret; | |
814 | ||
815 | ret = call_read_iter(file, iocb, iter); | |
816 | out: | |
817 | if (ret >= 0) | |
818 | fsnotify_access(file); | |
819 | return ret; | |
820 | } | |
821 | EXPORT_SYMBOL(vfs_iocb_iter_read); | |
822 | ||
18e9710e | 823 | ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos, |
ddef7ed2 | 824 | rwf_t flags) |
7687a7a4 | 825 | { |
18e9710e CH |
826 | if (!file->f_op->read_iter) |
827 | return -EINVAL; | |
828 | return do_iter_read(file, iter, ppos, flags); | |
829 | } | |
830 | EXPORT_SYMBOL(vfs_iter_read); | |
7687a7a4 | 831 | |
19c73586 | 832 | static ssize_t do_iter_write(struct file *file, struct iov_iter *iter, |
ddef7ed2 | 833 | loff_t *pos, rwf_t flags) |
19c73586 CH |
834 | { |
835 | size_t tot_len; | |
836 | ssize_t ret = 0; | |
03d95eb2 | 837 | |
edab5fe3 CH |
838 | if (!(file->f_mode & FMODE_WRITE)) |
839 | return -EBADF; | |
840 | if (!(file->f_mode & FMODE_CAN_WRITE)) | |
841 | return -EINVAL; | |
842 | ||
19c73586 CH |
843 | tot_len = iov_iter_count(iter); |
844 | if (!tot_len) | |
845 | return 0; | |
846 | ret = rw_verify_area(WRITE, file, pos, tot_len); | |
7687a7a4 MS |
847 | if (ret < 0) |
848 | return ret; | |
849 | ||
19c73586 CH |
850 | if (file->f_op->write_iter) |
851 | ret = do_iter_readv_writev(file, iter, pos, WRITE, flags); | |
852 | else | |
853 | ret = do_loop_readv_writev(file, iter, pos, WRITE, flags); | |
19c73586 CH |
854 | if (ret > 0) |
855 | fsnotify_modify(file); | |
7687a7a4 MS |
856 | return ret; |
857 | } | |
858 | ||
5dcdc43e JX |
859 | ssize_t vfs_iocb_iter_write(struct file *file, struct kiocb *iocb, |
860 | struct iov_iter *iter) | |
861 | { | |
862 | size_t tot_len; | |
863 | ssize_t ret = 0; | |
864 | ||
865 | if (!file->f_op->write_iter) | |
866 | return -EINVAL; | |
867 | if (!(file->f_mode & FMODE_WRITE)) | |
868 | return -EBADF; | |
869 | if (!(file->f_mode & FMODE_CAN_WRITE)) | |
870 | return -EINVAL; | |
871 | ||
872 | tot_len = iov_iter_count(iter); | |
873 | if (!tot_len) | |
874 | return 0; | |
875 | ret = rw_verify_area(WRITE, file, &iocb->ki_pos, tot_len); | |
876 | if (ret < 0) | |
877 | return ret; | |
878 | ||
879 | ret = call_write_iter(file, iocb, iter); | |
880 | if (ret > 0) | |
881 | fsnotify_modify(file); | |
882 | ||
883 | return ret; | |
884 | } | |
885 | EXPORT_SYMBOL(vfs_iocb_iter_write); | |
886 | ||
abbb6589 | 887 | ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos, |
ddef7ed2 | 888 | rwf_t flags) |
abbb6589 CH |
889 | { |
890 | if (!file->f_op->write_iter) | |
891 | return -EINVAL; | |
892 | return do_iter_write(file, iter, ppos, flags); | |
893 | } | |
894 | EXPORT_SYMBOL(vfs_iter_write); | |
895 | ||
36e2c742 | 896 | static ssize_t vfs_readv(struct file *file, const struct iovec __user *vec, |
ddef7ed2 | 897 | unsigned long vlen, loff_t *pos, rwf_t flags) |
1da177e4 | 898 | { |
7687a7a4 MS |
899 | struct iovec iovstack[UIO_FASTIOV]; |
900 | struct iovec *iov = iovstack; | |
901 | struct iov_iter iter; | |
902 | ssize_t ret; | |
1da177e4 | 903 | |
251b42a1 | 904 | ret = import_iovec(READ, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter); |
edab5fe3 CH |
905 | if (ret >= 0) { |
906 | ret = do_iter_read(file, &iter, pos, flags); | |
907 | kfree(iov); | |
908 | } | |
1da177e4 | 909 | |
251b42a1 CH |
910 | return ret; |
911 | } | |
1da177e4 | 912 | |
9725d4ce | 913 | static ssize_t vfs_writev(struct file *file, const struct iovec __user *vec, |
ddef7ed2 | 914 | unsigned long vlen, loff_t *pos, rwf_t flags) |
1da177e4 | 915 | { |
251b42a1 CH |
916 | struct iovec iovstack[UIO_FASTIOV]; |
917 | struct iovec *iov = iovstack; | |
918 | struct iov_iter iter; | |
919 | ssize_t ret; | |
1da177e4 | 920 | |
251b42a1 | 921 | ret = import_iovec(WRITE, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter); |
edab5fe3 | 922 | if (ret >= 0) { |
62473a2d | 923 | file_start_write(file); |
edab5fe3 | 924 | ret = do_iter_write(file, &iter, pos, flags); |
62473a2d | 925 | file_end_write(file); |
edab5fe3 CH |
926 | kfree(iov); |
927 | } | |
251b42a1 | 928 | return ret; |
1da177e4 | 929 | } |
1da177e4 | 930 | |
f17d8b35 | 931 | static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec, |
ddef7ed2 | 932 | unsigned long vlen, rwf_t flags) |
1da177e4 | 933 | { |
9c225f26 | 934 | struct fd f = fdget_pos(fd); |
1da177e4 | 935 | ssize_t ret = -EBADF; |
1da177e4 | 936 | |
2903ff01 | 937 | if (f.file) { |
438ab720 KS |
938 | loff_t pos, *ppos = file_ppos(f.file); |
939 | if (ppos) { | |
940 | pos = *ppos; | |
941 | ppos = &pos; | |
942 | } | |
943 | ret = vfs_readv(f.file, vec, vlen, ppos, flags); | |
944 | if (ret >= 0 && ppos) | |
945 | f.file->f_pos = pos; | |
9c225f26 | 946 | fdput_pos(f); |
1da177e4 LT |
947 | } |
948 | ||
949 | if (ret > 0) | |
4b98d11b AD |
950 | add_rchar(current, ret); |
951 | inc_syscr(current); | |
1da177e4 LT |
952 | return ret; |
953 | } | |
954 | ||
f17d8b35 | 955 | static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec, |
ddef7ed2 | 956 | unsigned long vlen, rwf_t flags) |
1da177e4 | 957 | { |
9c225f26 | 958 | struct fd f = fdget_pos(fd); |
1da177e4 | 959 | ssize_t ret = -EBADF; |
1da177e4 | 960 | |
2903ff01 | 961 | if (f.file) { |
438ab720 KS |
962 | loff_t pos, *ppos = file_ppos(f.file); |
963 | if (ppos) { | |
964 | pos = *ppos; | |
965 | ppos = &pos; | |
966 | } | |
967 | ret = vfs_writev(f.file, vec, vlen, ppos, flags); | |
968 | if (ret >= 0 && ppos) | |
969 | f.file->f_pos = pos; | |
9c225f26 | 970 | fdput_pos(f); |
1da177e4 LT |
971 | } |
972 | ||
973 | if (ret > 0) | |
4b98d11b AD |
974 | add_wchar(current, ret); |
975 | inc_syscw(current); | |
1da177e4 LT |
976 | return ret; |
977 | } | |
978 | ||
601cc11d LT |
979 | static inline loff_t pos_from_hilo(unsigned long high, unsigned long low) |
980 | { | |
981 | #define HALF_LONG_BITS (BITS_PER_LONG / 2) | |
982 | return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low; | |
983 | } | |
984 | ||
f17d8b35 | 985 | static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec, |
ddef7ed2 | 986 | unsigned long vlen, loff_t pos, rwf_t flags) |
f3554f4b | 987 | { |
2903ff01 | 988 | struct fd f; |
f3554f4b | 989 | ssize_t ret = -EBADF; |
f3554f4b GH |
990 | |
991 | if (pos < 0) | |
992 | return -EINVAL; | |
993 | ||
2903ff01 AV |
994 | f = fdget(fd); |
995 | if (f.file) { | |
f3554f4b | 996 | ret = -ESPIPE; |
2903ff01 | 997 | if (f.file->f_mode & FMODE_PREAD) |
f17d8b35 | 998 | ret = vfs_readv(f.file, vec, vlen, &pos, flags); |
2903ff01 | 999 | fdput(f); |
f3554f4b GH |
1000 | } |
1001 | ||
1002 | if (ret > 0) | |
1003 | add_rchar(current, ret); | |
1004 | inc_syscr(current); | |
1005 | return ret; | |
1006 | } | |
1007 | ||
f17d8b35 | 1008 | static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec, |
ddef7ed2 | 1009 | unsigned long vlen, loff_t pos, rwf_t flags) |
f3554f4b | 1010 | { |
2903ff01 | 1011 | struct fd f; |
f3554f4b | 1012 | ssize_t ret = -EBADF; |
f3554f4b GH |
1013 | |
1014 | if (pos < 0) | |
1015 | return -EINVAL; | |
1016 | ||
2903ff01 AV |
1017 | f = fdget(fd); |
1018 | if (f.file) { | |
f3554f4b | 1019 | ret = -ESPIPE; |
2903ff01 | 1020 | if (f.file->f_mode & FMODE_PWRITE) |
f17d8b35 | 1021 | ret = vfs_writev(f.file, vec, vlen, &pos, flags); |
2903ff01 | 1022 | fdput(f); |
f3554f4b GH |
1023 | } |
1024 | ||
1025 | if (ret > 0) | |
1026 | add_wchar(current, ret); | |
1027 | inc_syscw(current); | |
1028 | return ret; | |
1029 | } | |
1030 | ||
f17d8b35 MT |
1031 | SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec, |
1032 | unsigned long, vlen) | |
1033 | { | |
1034 | return do_readv(fd, vec, vlen, 0); | |
1035 | } | |
1036 | ||
1037 | SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec, | |
1038 | unsigned long, vlen) | |
1039 | { | |
1040 | return do_writev(fd, vec, vlen, 0); | |
1041 | } | |
1042 | ||
1043 | SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec, | |
1044 | unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h) | |
1045 | { | |
1046 | loff_t pos = pos_from_hilo(pos_h, pos_l); | |
1047 | ||
1048 | return do_preadv(fd, vec, vlen, pos, 0); | |
1049 | } | |
1050 | ||
1051 | SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec, | |
1052 | unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h, | |
ddef7ed2 | 1053 | rwf_t, flags) |
f17d8b35 MT |
1054 | { |
1055 | loff_t pos = pos_from_hilo(pos_h, pos_l); | |
1056 | ||
1057 | if (pos == -1) | |
1058 | return do_readv(fd, vec, vlen, flags); | |
1059 | ||
1060 | return do_preadv(fd, vec, vlen, pos, flags); | |
1061 | } | |
1062 | ||
1063 | SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec, | |
1064 | unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h) | |
1065 | { | |
1066 | loff_t pos = pos_from_hilo(pos_h, pos_l); | |
1067 | ||
1068 | return do_pwritev(fd, vec, vlen, pos, 0); | |
1069 | } | |
1070 | ||
1071 | SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec, | |
1072 | unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h, | |
ddef7ed2 | 1073 | rwf_t, flags) |
f17d8b35 MT |
1074 | { |
1075 | loff_t pos = pos_from_hilo(pos_h, pos_l); | |
1076 | ||
1077 | if (pos == -1) | |
1078 | return do_writev(fd, vec, vlen, flags); | |
1079 | ||
1080 | return do_pwritev(fd, vec, vlen, pos, flags); | |
1081 | } | |
1082 | ||
3523a9d4 CH |
1083 | /* |
1084 | * Various compat syscalls. Note that they all pretend to take a native | |
1085 | * iovec - import_iovec will properly treat those as compat_iovecs based on | |
1086 | * in_compat_syscall(). | |
1087 | */ | |
72ec3516 | 1088 | #ifdef CONFIG_COMPAT |
378a10f3 HC |
1089 | #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64 |
1090 | COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd, | |
3523a9d4 | 1091 | const struct iovec __user *, vec, |
378a10f3 HC |
1092 | unsigned long, vlen, loff_t, pos) |
1093 | { | |
3523a9d4 | 1094 | return do_preadv(fd, vec, vlen, pos, 0); |
378a10f3 HC |
1095 | } |
1096 | #endif | |
1097 | ||
dfd948e3 | 1098 | COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd, |
3523a9d4 | 1099 | const struct iovec __user *, vec, |
dfd948e3 | 1100 | compat_ulong_t, vlen, u32, pos_low, u32, pos_high) |
72ec3516 AV |
1101 | { |
1102 | loff_t pos = ((loff_t)pos_high << 32) | pos_low; | |
378a10f3 | 1103 | |
3523a9d4 | 1104 | return do_preadv(fd, vec, vlen, pos, 0); |
f17d8b35 MT |
1105 | } |
1106 | ||
3ebfd81f L |
1107 | #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2 |
1108 | COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd, | |
3523a9d4 | 1109 | const struct iovec __user *, vec, |
ddef7ed2 | 1110 | unsigned long, vlen, loff_t, pos, rwf_t, flags) |
3ebfd81f | 1111 | { |
cc4b1242 | 1112 | if (pos == -1) |
3523a9d4 CH |
1113 | return do_readv(fd, vec, vlen, flags); |
1114 | return do_preadv(fd, vec, vlen, pos, flags); | |
3ebfd81f L |
1115 | } |
1116 | #endif | |
1117 | ||
f17d8b35 | 1118 | COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd, |
3523a9d4 | 1119 | const struct iovec __user *, vec, |
f17d8b35 | 1120 | compat_ulong_t, vlen, u32, pos_low, u32, pos_high, |
ddef7ed2 | 1121 | rwf_t, flags) |
f17d8b35 MT |
1122 | { |
1123 | loff_t pos = ((loff_t)pos_high << 32) | pos_low; | |
1124 | ||
1125 | if (pos == -1) | |
3523a9d4 CH |
1126 | return do_readv(fd, vec, vlen, flags); |
1127 | return do_preadv(fd, vec, vlen, pos, flags); | |
72ec3516 AV |
1128 | } |
1129 | ||
378a10f3 HC |
1130 | #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64 |
1131 | COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd, | |
3523a9d4 | 1132 | const struct iovec __user *, vec, |
378a10f3 HC |
1133 | unsigned long, vlen, loff_t, pos) |
1134 | { | |
3523a9d4 | 1135 | return do_pwritev(fd, vec, vlen, pos, 0); |
378a10f3 HC |
1136 | } |
1137 | #endif | |
1138 | ||
dfd948e3 | 1139 | COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd, |
3523a9d4 | 1140 | const struct iovec __user *,vec, |
dfd948e3 | 1141 | compat_ulong_t, vlen, u32, pos_low, u32, pos_high) |
72ec3516 AV |
1142 | { |
1143 | loff_t pos = ((loff_t)pos_high << 32) | pos_low; | |
378a10f3 | 1144 | |
3523a9d4 | 1145 | return do_pwritev(fd, vec, vlen, pos, 0); |
72ec3516 | 1146 | } |
f17d8b35 | 1147 | |
3ebfd81f L |
1148 | #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2 |
1149 | COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd, | |
3523a9d4 | 1150 | const struct iovec __user *, vec, |
ddef7ed2 | 1151 | unsigned long, vlen, loff_t, pos, rwf_t, flags) |
3ebfd81f | 1152 | { |
cc4b1242 | 1153 | if (pos == -1) |
3523a9d4 CH |
1154 | return do_writev(fd, vec, vlen, flags); |
1155 | return do_pwritev(fd, vec, vlen, pos, flags); | |
3ebfd81f L |
1156 | } |
1157 | #endif | |
1158 | ||
f17d8b35 | 1159 | COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd, |
3523a9d4 | 1160 | const struct iovec __user *,vec, |
ddef7ed2 | 1161 | compat_ulong_t, vlen, u32, pos_low, u32, pos_high, rwf_t, flags) |
f17d8b35 MT |
1162 | { |
1163 | loff_t pos = ((loff_t)pos_high << 32) | pos_low; | |
1164 | ||
1165 | if (pos == -1) | |
3523a9d4 CH |
1166 | return do_writev(fd, vec, vlen, flags); |
1167 | return do_pwritev(fd, vec, vlen, pos, flags); | |
72ec3516 | 1168 | } |
3523a9d4 | 1169 | #endif /* CONFIG_COMPAT */ |
72ec3516 | 1170 | |
19f4fc3a AV |
1171 | static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos, |
1172 | size_t count, loff_t max) | |
1da177e4 | 1173 | { |
2903ff01 AV |
1174 | struct fd in, out; |
1175 | struct inode *in_inode, *out_inode; | |
b964bf53 | 1176 | struct pipe_inode_info *opipe; |
1da177e4 | 1177 | loff_t pos; |
7995bd28 | 1178 | loff_t out_pos; |
1da177e4 | 1179 | ssize_t retval; |
2903ff01 | 1180 | int fl; |
1da177e4 LT |
1181 | |
1182 | /* | |
1183 | * Get input file, and verify that it is ok.. | |
1184 | */ | |
1185 | retval = -EBADF; | |
2903ff01 AV |
1186 | in = fdget(in_fd); |
1187 | if (!in.file) | |
1da177e4 | 1188 | goto out; |
2903ff01 | 1189 | if (!(in.file->f_mode & FMODE_READ)) |
1da177e4 | 1190 | goto fput_in; |
1da177e4 | 1191 | retval = -ESPIPE; |
7995bd28 AV |
1192 | if (!ppos) { |
1193 | pos = in.file->f_pos; | |
1194 | } else { | |
1195 | pos = *ppos; | |
2903ff01 | 1196 | if (!(in.file->f_mode & FMODE_PREAD)) |
1da177e4 | 1197 | goto fput_in; |
7995bd28 AV |
1198 | } |
1199 | retval = rw_verify_area(READ, in.file, &pos, count); | |
e28cc715 | 1200 | if (retval < 0) |
1da177e4 | 1201 | goto fput_in; |
bc61384d AV |
1202 | if (count > MAX_RW_COUNT) |
1203 | count = MAX_RW_COUNT; | |
1da177e4 | 1204 | |
1da177e4 LT |
1205 | /* |
1206 | * Get output file, and verify that it is ok.. | |
1207 | */ | |
1208 | retval = -EBADF; | |
2903ff01 AV |
1209 | out = fdget(out_fd); |
1210 | if (!out.file) | |
1da177e4 | 1211 | goto fput_in; |
2903ff01 | 1212 | if (!(out.file->f_mode & FMODE_WRITE)) |
1da177e4 | 1213 | goto fput_out; |
496ad9aa AV |
1214 | in_inode = file_inode(in.file); |
1215 | out_inode = file_inode(out.file); | |
7995bd28 | 1216 | out_pos = out.file->f_pos; |
1da177e4 | 1217 | |
1da177e4 LT |
1218 | if (!max) |
1219 | max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes); | |
1220 | ||
1da177e4 LT |
1221 | if (unlikely(pos + count > max)) { |
1222 | retval = -EOVERFLOW; | |
1223 | if (pos >= max) | |
1224 | goto fput_out; | |
1225 | count = max - pos; | |
1226 | } | |
1227 | ||
d96e6e71 | 1228 | fl = 0; |
534f2aaa | 1229 | #if 0 |
d96e6e71 JA |
1230 | /* |
1231 | * We need to debate whether we can enable this or not. The | |
1232 | * man page documents EAGAIN return for the output at least, | |
1233 | * and the application is arguably buggy if it doesn't expect | |
1234 | * EAGAIN on a non-blocking file descriptor. | |
1235 | */ | |
2903ff01 | 1236 | if (in.file->f_flags & O_NONBLOCK) |
d96e6e71 | 1237 | fl = SPLICE_F_NONBLOCK; |
534f2aaa | 1238 | #endif |
b964bf53 AV |
1239 | opipe = get_pipe_info(out.file, true); |
1240 | if (!opipe) { | |
1241 | retval = rw_verify_area(WRITE, out.file, &out_pos, count); | |
1242 | if (retval < 0) | |
1243 | goto fput_out; | |
1244 | file_start_write(out.file); | |
1245 | retval = do_splice_direct(in.file, &pos, out.file, &out_pos, | |
1246 | count, fl); | |
1247 | file_end_write(out.file); | |
1248 | } else { | |
1249 | retval = splice_file_to_pipe(in.file, opipe, &pos, count, fl); | |
1250 | } | |
1da177e4 LT |
1251 | |
1252 | if (retval > 0) { | |
4b98d11b AD |
1253 | add_rchar(current, retval); |
1254 | add_wchar(current, retval); | |
a68c2f12 SW |
1255 | fsnotify_access(in.file); |
1256 | fsnotify_modify(out.file); | |
7995bd28 AV |
1257 | out.file->f_pos = out_pos; |
1258 | if (ppos) | |
1259 | *ppos = pos; | |
1260 | else | |
1261 | in.file->f_pos = pos; | |
1da177e4 | 1262 | } |
1da177e4 | 1263 | |
4b98d11b AD |
1264 | inc_syscr(current); |
1265 | inc_syscw(current); | |
7995bd28 | 1266 | if (pos > max) |
1da177e4 LT |
1267 | retval = -EOVERFLOW; |
1268 | ||
1269 | fput_out: | |
2903ff01 | 1270 | fdput(out); |
1da177e4 | 1271 | fput_in: |
2903ff01 | 1272 | fdput(in); |
1da177e4 LT |
1273 | out: |
1274 | return retval; | |
1275 | } | |
1276 | ||
002c8976 | 1277 | SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count) |
1da177e4 LT |
1278 | { |
1279 | loff_t pos; | |
1280 | off_t off; | |
1281 | ssize_t ret; | |
1282 | ||
1283 | if (offset) { | |
1284 | if (unlikely(get_user(off, offset))) | |
1285 | return -EFAULT; | |
1286 | pos = off; | |
1287 | ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS); | |
1288 | if (unlikely(put_user(pos, offset))) | |
1289 | return -EFAULT; | |
1290 | return ret; | |
1291 | } | |
1292 | ||
1293 | return do_sendfile(out_fd, in_fd, NULL, count, 0); | |
1294 | } | |
1295 | ||
002c8976 | 1296 | SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count) |
1da177e4 LT |
1297 | { |
1298 | loff_t pos; | |
1299 | ssize_t ret; | |
1300 | ||
1301 | if (offset) { | |
1302 | if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t)))) | |
1303 | return -EFAULT; | |
1304 | ret = do_sendfile(out_fd, in_fd, &pos, count, 0); | |
1305 | if (unlikely(put_user(pos, offset))) | |
1306 | return -EFAULT; | |
1307 | return ret; | |
1308 | } | |
1309 | ||
1310 | return do_sendfile(out_fd, in_fd, NULL, count, 0); | |
1311 | } | |
19f4fc3a AV |
1312 | |
1313 | #ifdef CONFIG_COMPAT | |
1314 | COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, | |
1315 | compat_off_t __user *, offset, compat_size_t, count) | |
1316 | { | |
1317 | loff_t pos; | |
1318 | off_t off; | |
1319 | ssize_t ret; | |
1320 | ||
1321 | if (offset) { | |
1322 | if (unlikely(get_user(off, offset))) | |
1323 | return -EFAULT; | |
1324 | pos = off; | |
1325 | ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS); | |
1326 | if (unlikely(put_user(pos, offset))) | |
1327 | return -EFAULT; | |
1328 | return ret; | |
1329 | } | |
1330 | ||
1331 | return do_sendfile(out_fd, in_fd, NULL, count, 0); | |
1332 | } | |
1333 | ||
1334 | COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, | |
1335 | compat_loff_t __user *, offset, compat_size_t, count) | |
1336 | { | |
1337 | loff_t pos; | |
1338 | ssize_t ret; | |
1339 | ||
1340 | if (offset) { | |
1341 | if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t)))) | |
1342 | return -EFAULT; | |
1343 | ret = do_sendfile(out_fd, in_fd, &pos, count, 0); | |
1344 | if (unlikely(put_user(pos, offset))) | |
1345 | return -EFAULT; | |
1346 | return ret; | |
1347 | } | |
1348 | ||
1349 | return do_sendfile(out_fd, in_fd, NULL, count, 0); | |
1350 | } | |
1351 | #endif | |
29732938 | 1352 | |
f16acc9d DC |
1353 | /** |
1354 | * generic_copy_file_range - copy data between two files | |
1355 | * @file_in: file structure to read from | |
1356 | * @pos_in: file offset to read from | |
1357 | * @file_out: file structure to write data to | |
1358 | * @pos_out: file offset to write data to | |
1359 | * @len: amount of data to copy | |
1360 | * @flags: copy flags | |
1361 | * | |
1362 | * This is a generic filesystem helper to copy data from one file to another. | |
1363 | * It has no constraints on the source or destination file owners - the files | |
1364 | * can belong to different superblocks and different filesystem types. Short | |
1365 | * copies are allowed. | |
1366 | * | |
1367 | * This should be called from the @file_out filesystem, as per the | |
1368 | * ->copy_file_range() method. | |
1369 | * | |
1370 | * Returns the number of bytes copied or a negative error indicating the | |
1371 | * failure. | |
1372 | */ | |
1373 | ||
1374 | ssize_t generic_copy_file_range(struct file *file_in, loff_t pos_in, | |
1375 | struct file *file_out, loff_t pos_out, | |
1376 | size_t len, unsigned int flags) | |
1377 | { | |
1378 | return do_splice_direct(file_in, &pos_in, file_out, &pos_out, | |
1379 | len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0); | |
1380 | } | |
1381 | EXPORT_SYMBOL(generic_copy_file_range); | |
1382 | ||
64bf5ff5 DC |
1383 | static ssize_t do_copy_file_range(struct file *file_in, loff_t pos_in, |
1384 | struct file *file_out, loff_t pos_out, | |
1385 | size_t len, unsigned int flags) | |
1386 | { | |
5dae222a AG |
1387 | /* |
1388 | * Although we now allow filesystems to handle cross sb copy, passing | |
1389 | * a file of the wrong filesystem type to filesystem driver can result | |
1390 | * in an attempt to dereference the wrong type of ->private_data, so | |
1391 | * avoid doing that until we really have a good reason. NFS defines | |
1392 | * several different file_system_type structures, but they all end up | |
1393 | * using the same ->copy_file_range() function pointer. | |
1394 | */ | |
1395 | if (file_out->f_op->copy_file_range && | |
1396 | file_out->f_op->copy_file_range == file_in->f_op->copy_file_range) | |
64bf5ff5 DC |
1397 | return file_out->f_op->copy_file_range(file_in, pos_in, |
1398 | file_out, pos_out, | |
1399 | len, flags); | |
1400 | ||
1401 | return generic_copy_file_range(file_in, pos_in, file_out, pos_out, len, | |
1402 | flags); | |
1403 | } | |
1404 | ||
407e9c63 DW |
1405 | /* |
1406 | * Performs necessary checks before doing a file copy | |
1407 | * | |
1408 | * Can adjust amount of bytes to copy via @req_count argument. | |
1409 | * Returns appropriate error code that caller should return or | |
1410 | * zero in case the copy should be allowed. | |
1411 | */ | |
1412 | static int generic_copy_file_checks(struct file *file_in, loff_t pos_in, | |
1413 | struct file *file_out, loff_t pos_out, | |
1414 | size_t *req_count, unsigned int flags) | |
1415 | { | |
1416 | struct inode *inode_in = file_inode(file_in); | |
1417 | struct inode *inode_out = file_inode(file_out); | |
1418 | uint64_t count = *req_count; | |
1419 | loff_t size_in; | |
1420 | int ret; | |
1421 | ||
1422 | ret = generic_file_rw_checks(file_in, file_out); | |
1423 | if (ret) | |
1424 | return ret; | |
1425 | ||
1426 | /* Don't touch certain kinds of inodes */ | |
1427 | if (IS_IMMUTABLE(inode_out)) | |
1428 | return -EPERM; | |
1429 | ||
1430 | if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out)) | |
1431 | return -ETXTBSY; | |
1432 | ||
1433 | /* Ensure offsets don't wrap. */ | |
1434 | if (pos_in + count < pos_in || pos_out + count < pos_out) | |
1435 | return -EOVERFLOW; | |
1436 | ||
1437 | /* Shorten the copy to EOF */ | |
1438 | size_in = i_size_read(inode_in); | |
1439 | if (pos_in >= size_in) | |
1440 | count = 0; | |
1441 | else | |
1442 | count = min(count, size_in - (uint64_t)pos_in); | |
1443 | ||
1444 | ret = generic_write_check_limits(file_out, pos_out, &count); | |
1445 | if (ret) | |
1446 | return ret; | |
1447 | ||
1448 | /* Don't allow overlapped copying within the same file. */ | |
1449 | if (inode_in == inode_out && | |
1450 | pos_out + count > pos_in && | |
1451 | pos_out < pos_in + count) | |
1452 | return -EINVAL; | |
1453 | ||
1454 | *req_count = count; | |
1455 | return 0; | |
1456 | } | |
1457 | ||
29732938 ZB |
1458 | /* |
1459 | * copy_file_range() differs from regular file read and write in that it | |
1460 | * specifically allows return partial success. When it does so is up to | |
1461 | * the copy_file_range method. | |
1462 | */ | |
1463 | ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in, | |
1464 | struct file *file_out, loff_t pos_out, | |
1465 | size_t len, unsigned int flags) | |
1466 | { | |
29732938 ZB |
1467 | ssize_t ret; |
1468 | ||
1469 | if (flags != 0) | |
1470 | return -EINVAL; | |
1471 | ||
96e6e8f4 AG |
1472 | ret = generic_copy_file_checks(file_in, pos_in, file_out, pos_out, &len, |
1473 | flags); | |
a3171351 AG |
1474 | if (unlikely(ret)) |
1475 | return ret; | |
11cbfb10 | 1476 | |
29732938 | 1477 | ret = rw_verify_area(READ, file_in, &pos_in, len); |
bc61384d AV |
1478 | if (unlikely(ret)) |
1479 | return ret; | |
1480 | ||
1481 | ret = rw_verify_area(WRITE, file_out, &pos_out, len); | |
1482 | if (unlikely(ret)) | |
29732938 ZB |
1483 | return ret; |
1484 | ||
29732938 ZB |
1485 | if (len == 0) |
1486 | return 0; | |
1487 | ||
bfe219d3 | 1488 | file_start_write(file_out); |
29732938 | 1489 | |
a76b5b04 CH |
1490 | /* |
1491 | * Try cloning first, this is supported by more file systems, and | |
1492 | * more efficient if both clone and copy are supported (e.g. NFS). | |
1493 | */ | |
5dae222a AG |
1494 | if (file_in->f_op->remap_file_range && |
1495 | file_inode(file_in)->i_sb == file_inode(file_out)->i_sb) { | |
42ec3d4c DW |
1496 | loff_t cloned; |
1497 | ||
1498 | cloned = file_in->f_op->remap_file_range(file_in, pos_in, | |
1499 | file_out, pos_out, | |
eca3654e DW |
1500 | min_t(loff_t, MAX_RW_COUNT, len), |
1501 | REMAP_FILE_CAN_SHORTEN); | |
42ec3d4c DW |
1502 | if (cloned > 0) { |
1503 | ret = cloned; | |
a76b5b04 CH |
1504 | goto done; |
1505 | } | |
1506 | } | |
1507 | ||
64bf5ff5 DC |
1508 | ret = do_copy_file_range(file_in, pos_in, file_out, pos_out, len, |
1509 | flags); | |
1510 | WARN_ON_ONCE(ret == -EOPNOTSUPP); | |
a76b5b04 | 1511 | done: |
29732938 ZB |
1512 | if (ret > 0) { |
1513 | fsnotify_access(file_in); | |
1514 | add_rchar(current, ret); | |
1515 | fsnotify_modify(file_out); | |
1516 | add_wchar(current, ret); | |
1517 | } | |
a76b5b04 | 1518 | |
29732938 ZB |
1519 | inc_syscr(current); |
1520 | inc_syscw(current); | |
1521 | ||
bfe219d3 | 1522 | file_end_write(file_out); |
29732938 ZB |
1523 | |
1524 | return ret; | |
1525 | } | |
1526 | EXPORT_SYMBOL(vfs_copy_file_range); | |
1527 | ||
1528 | SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in, | |
1529 | int, fd_out, loff_t __user *, off_out, | |
1530 | size_t, len, unsigned int, flags) | |
1531 | { | |
1532 | loff_t pos_in; | |
1533 | loff_t pos_out; | |
1534 | struct fd f_in; | |
1535 | struct fd f_out; | |
1536 | ssize_t ret = -EBADF; | |
1537 | ||
1538 | f_in = fdget(fd_in); | |
1539 | if (!f_in.file) | |
1540 | goto out2; | |
1541 | ||
1542 | f_out = fdget(fd_out); | |
1543 | if (!f_out.file) | |
1544 | goto out1; | |
1545 | ||
1546 | ret = -EFAULT; | |
1547 | if (off_in) { | |
1548 | if (copy_from_user(&pos_in, off_in, sizeof(loff_t))) | |
1549 | goto out; | |
1550 | } else { | |
1551 | pos_in = f_in.file->f_pos; | |
1552 | } | |
1553 | ||
1554 | if (off_out) { | |
1555 | if (copy_from_user(&pos_out, off_out, sizeof(loff_t))) | |
1556 | goto out; | |
1557 | } else { | |
1558 | pos_out = f_out.file->f_pos; | |
1559 | } | |
1560 | ||
1561 | ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len, | |
1562 | flags); | |
1563 | if (ret > 0) { | |
1564 | pos_in += ret; | |
1565 | pos_out += ret; | |
1566 | ||
1567 | if (off_in) { | |
1568 | if (copy_to_user(off_in, &pos_in, sizeof(loff_t))) | |
1569 | ret = -EFAULT; | |
1570 | } else { | |
1571 | f_in.file->f_pos = pos_in; | |
1572 | } | |
1573 | ||
1574 | if (off_out) { | |
1575 | if (copy_to_user(off_out, &pos_out, sizeof(loff_t))) | |
1576 | ret = -EFAULT; | |
1577 | } else { | |
1578 | f_out.file->f_pos = pos_out; | |
1579 | } | |
1580 | } | |
1581 | ||
1582 | out: | |
1583 | fdput(f_out); | |
1584 | out1: | |
1585 | fdput(f_in); | |
1586 | out2: | |
1587 | return ret; | |
1588 | } | |
04b38d60 | 1589 | |
edc58dd0 | 1590 | /* |
407e9c63 DW |
1591 | * Don't operate on ranges the page cache doesn't support, and don't exceed the |
1592 | * LFS limits. If pos is under the limit it becomes a short access. If it | |
1593 | * exceeds the limit we return -EFBIG. | |
edc58dd0 | 1594 | */ |
407e9c63 | 1595 | int generic_write_check_limits(struct file *file, loff_t pos, loff_t *count) |
edc58dd0 | 1596 | { |
407e9c63 DW |
1597 | struct inode *inode = file->f_mapping->host; |
1598 | loff_t max_size = inode->i_sb->s_maxbytes; | |
1599 | loff_t limit = rlimit(RLIMIT_FSIZE); | |
edc58dd0 | 1600 | |
407e9c63 DW |
1601 | if (limit != RLIM_INFINITY) { |
1602 | if (pos >= limit) { | |
1603 | send_sig(SIGXFSZ, current, 0); | |
1604 | return -EFBIG; | |
edc58dd0 | 1605 | } |
407e9c63 DW |
1606 | *count = min(*count, limit - pos); |
1607 | } | |
edc58dd0 | 1608 | |
407e9c63 DW |
1609 | if (!(file->f_flags & O_LARGEFILE)) |
1610 | max_size = MAX_NON_LFS; | |
c32e5f39 | 1611 | |
407e9c63 DW |
1612 | if (unlikely(pos >= max_size)) |
1613 | return -EFBIG; | |
c32e5f39 | 1614 | |
407e9c63 | 1615 | *count = min(*count, max_size - pos); |
c32e5f39 | 1616 | |
c32e5f39 | 1617 | return 0; |
c32e5f39 | 1618 | } |
04b38d60 | 1619 | |
876bec6f | 1620 | /* |
407e9c63 | 1621 | * Performs necessary checks before doing a write |
22725ce4 | 1622 | * |
407e9c63 DW |
1623 | * Can adjust writing position or amount of bytes to write. |
1624 | * Returns appropriate error code that caller should return or | |
1625 | * zero in case that write should be allowed. | |
876bec6f | 1626 | */ |
407e9c63 | 1627 | ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from) |
876bec6f | 1628 | { |
407e9c63 DW |
1629 | struct file *file = iocb->ki_filp; |
1630 | struct inode *inode = file->f_mapping->host; | |
1631 | loff_t count; | |
876bec6f DW |
1632 | int ret; |
1633 | ||
407e9c63 | 1634 | if (IS_SWAPFILE(inode)) |
876bec6f DW |
1635 | return -ETXTBSY; |
1636 | ||
407e9c63 DW |
1637 | if (!iov_iter_count(from)) |
1638 | return 0; | |
5de4480a | 1639 | |
407e9c63 DW |
1640 | /* FIXME: this is for backwards compatibility with 2.4 */ |
1641 | if (iocb->ki_flags & IOCB_APPEND) | |
1642 | iocb->ki_pos = i_size_read(inode); | |
1b4f42a1 | 1643 | |
407e9c63 DW |
1644 | if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT)) |
1645 | return -EINVAL; | |
1b4f42a1 | 1646 | |
407e9c63 DW |
1647 | count = iov_iter_count(from); |
1648 | ret = generic_write_check_limits(file, iocb->ki_pos, &count); | |
1b4f42a1 MS |
1649 | if (ret) |
1650 | return ret; | |
1651 | ||
407e9c63 DW |
1652 | iov_iter_truncate(from, count); |
1653 | return iov_iter_count(from); | |
1b4f42a1 | 1654 | } |
407e9c63 | 1655 | EXPORT_SYMBOL(generic_write_checks); |
1b4f42a1 | 1656 | |
407e9c63 DW |
1657 | /* |
1658 | * Performs common checks before doing a file copy/clone | |
1659 | * from @file_in to @file_out. | |
1660 | */ | |
1661 | int generic_file_rw_checks(struct file *file_in, struct file *file_out) | |
54dbc151 | 1662 | { |
407e9c63 DW |
1663 | struct inode *inode_in = file_inode(file_in); |
1664 | struct inode *inode_out = file_inode(file_out); | |
54dbc151 | 1665 | |
407e9c63 DW |
1666 | /* Don't copy dirs, pipes, sockets... */ |
1667 | if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode)) | |
494633fa | 1668 | return -EISDIR; |
407e9c63 | 1669 | if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode)) |
22725ce4 DW |
1670 | return -EINVAL; |
1671 | ||
407e9c63 DW |
1672 | if (!(file_in->f_mode & FMODE_READ) || |
1673 | !(file_out->f_mode & FMODE_WRITE) || | |
1674 | (file_out->f_flags & O_APPEND)) | |
1675 | return -EBADF; | |
1b4f42a1 | 1676 | |
407e9c63 | 1677 | return 0; |
54dbc151 | 1678 | } |