]>
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> | |
a27bb332 | 12 | #include <linux/aio.h> |
0eeca283 | 13 | #include <linux/fsnotify.h> |
1da177e4 | 14 | #include <linux/security.h> |
630d9c47 | 15 | #include <linux/export.h> |
1da177e4 | 16 | #include <linux/syscalls.h> |
e28cc715 | 17 | #include <linux/pagemap.h> |
d6b29d7c | 18 | #include <linux/splice.h> |
561c6731 | 19 | #include <linux/compat.h> |
06ae43f3 | 20 | #include "internal.h" |
1da177e4 LT |
21 | |
22 | #include <asm/uaccess.h> | |
23 | #include <asm/unistd.h> | |
24 | ||
c0bd14af AV |
25 | typedef ssize_t (*io_fn_t)(struct file *, char __user *, size_t, loff_t *); |
26 | typedef ssize_t (*iov_fn_t)(struct kiocb *, const struct iovec *, | |
27 | unsigned long, loff_t); | |
28 | ||
4b6f5d20 | 29 | const struct file_operations generic_ro_fops = { |
1da177e4 | 30 | .llseek = generic_file_llseek, |
543ade1f BP |
31 | .read = do_sync_read, |
32 | .aio_read = generic_file_aio_read, | |
1da177e4 | 33 | .mmap = generic_file_readonly_mmap, |
534f2aaa | 34 | .splice_read = generic_file_splice_read, |
1da177e4 LT |
35 | }; |
36 | ||
37 | EXPORT_SYMBOL(generic_ro_fops); | |
38 | ||
cccb5a1e | 39 | static inline int unsigned_offsets(struct file *file) |
4a3956c7 | 40 | { |
cccb5a1e | 41 | return file->f_mode & FMODE_UNSIGNED_OFFSET; |
4a3956c7 KH |
42 | } |
43 | ||
46a1c2c7 JL |
44 | /** |
45 | * vfs_setpos - update the file offset for lseek | |
46 | * @file: file structure in question | |
47 | * @offset: file offset to seek to | |
48 | * @maxsize: maximum file size | |
49 | * | |
50 | * This is a low-level filesystem helper for updating the file offset to | |
51 | * the value specified by @offset if the given offset is valid and it is | |
52 | * not equal to the current file offset. | |
53 | * | |
54 | * Return the specified offset on success and -EINVAL on invalid offset. | |
55 | */ | |
56 | loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize) | |
ef3d0fd2 AK |
57 | { |
58 | if (offset < 0 && !unsigned_offsets(file)) | |
59 | return -EINVAL; | |
60 | if (offset > maxsize) | |
61 | return -EINVAL; | |
62 | ||
63 | if (offset != file->f_pos) { | |
64 | file->f_pos = offset; | |
65 | file->f_version = 0; | |
66 | } | |
67 | return offset; | |
68 | } | |
46a1c2c7 | 69 | EXPORT_SYMBOL(vfs_setpos); |
ef3d0fd2 | 70 | |
3a8cff4f | 71 | /** |
5760495a | 72 | * generic_file_llseek_size - generic llseek implementation for regular files |
3a8cff4f CH |
73 | * @file: file structure to seek on |
74 | * @offset: file offset to seek to | |
965c8e59 | 75 | * @whence: type of seek |
e8b96eb5 ES |
76 | * @size: max size of this file in file system |
77 | * @eof: offset used for SEEK_END position | |
3a8cff4f | 78 | * |
5760495a | 79 | * This is a variant of generic_file_llseek that allows passing in a custom |
e8b96eb5 | 80 | * maximum file size and a custom EOF position, for e.g. hashed directories |
ef3d0fd2 AK |
81 | * |
82 | * Synchronization: | |
5760495a | 83 | * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms) |
ef3d0fd2 AK |
84 | * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes. |
85 | * read/writes behave like SEEK_SET against seeks. | |
3a8cff4f | 86 | */ |
9465efc9 | 87 | loff_t |
965c8e59 | 88 | generic_file_llseek_size(struct file *file, loff_t offset, int whence, |
e8b96eb5 | 89 | loff_t maxsize, loff_t eof) |
1da177e4 | 90 | { |
965c8e59 | 91 | switch (whence) { |
3a8cff4f | 92 | case SEEK_END: |
e8b96eb5 | 93 | offset += eof; |
3a8cff4f CH |
94 | break; |
95 | case SEEK_CUR: | |
5b6f1eb9 AK |
96 | /* |
97 | * Here we special-case the lseek(fd, 0, SEEK_CUR) | |
98 | * position-querying operation. Avoid rewriting the "same" | |
99 | * f_pos value back to the file because a concurrent read(), | |
100 | * write() or lseek() might have altered it | |
101 | */ | |
102 | if (offset == 0) | |
103 | return file->f_pos; | |
ef3d0fd2 AK |
104 | /* |
105 | * f_lock protects against read/modify/write race with other | |
106 | * SEEK_CURs. Note that parallel writes and reads behave | |
107 | * like SEEK_SET. | |
108 | */ | |
109 | spin_lock(&file->f_lock); | |
46a1c2c7 | 110 | offset = vfs_setpos(file, file->f_pos + offset, maxsize); |
ef3d0fd2 AK |
111 | spin_unlock(&file->f_lock); |
112 | return offset; | |
982d8165 JB |
113 | case SEEK_DATA: |
114 | /* | |
115 | * In the generic case the entire file is data, so as long as | |
116 | * offset isn't at the end of the file then the offset is data. | |
117 | */ | |
e8b96eb5 | 118 | if (offset >= eof) |
982d8165 JB |
119 | return -ENXIO; |
120 | break; | |
121 | case SEEK_HOLE: | |
122 | /* | |
123 | * There is a virtual hole at the end of the file, so as long as | |
124 | * offset isn't i_size or larger, return i_size. | |
125 | */ | |
e8b96eb5 | 126 | if (offset >= eof) |
982d8165 | 127 | return -ENXIO; |
e8b96eb5 | 128 | offset = eof; |
982d8165 | 129 | break; |
1da177e4 | 130 | } |
3a8cff4f | 131 | |
46a1c2c7 | 132 | return vfs_setpos(file, offset, maxsize); |
5760495a AK |
133 | } |
134 | EXPORT_SYMBOL(generic_file_llseek_size); | |
135 | ||
136 | /** | |
137 | * generic_file_llseek - generic llseek implementation for regular files | |
138 | * @file: file structure to seek on | |
139 | * @offset: file offset to seek to | |
965c8e59 | 140 | * @whence: type of seek |
5760495a AK |
141 | * |
142 | * This is a generic implemenation of ->llseek useable for all normal local | |
143 | * filesystems. It just updates the file offset to the value specified by | |
546ae2d2 | 144 | * @offset and @whence. |
5760495a | 145 | */ |
965c8e59 | 146 | loff_t generic_file_llseek(struct file *file, loff_t offset, int whence) |
5760495a AK |
147 | { |
148 | struct inode *inode = file->f_mapping->host; | |
149 | ||
965c8e59 | 150 | return generic_file_llseek_size(file, offset, whence, |
e8b96eb5 ES |
151 | inode->i_sb->s_maxbytes, |
152 | i_size_read(inode)); | |
1da177e4 | 153 | } |
9465efc9 | 154 | EXPORT_SYMBOL(generic_file_llseek); |
1da177e4 | 155 | |
1bf9d14d AV |
156 | /** |
157 | * fixed_size_llseek - llseek implementation for fixed-sized devices | |
158 | * @file: file structure to seek on | |
159 | * @offset: file offset to seek to | |
160 | * @whence: type of seek | |
161 | * @size: size of the file | |
162 | * | |
163 | */ | |
164 | loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size) | |
165 | { | |
166 | switch (whence) { | |
167 | case SEEK_SET: case SEEK_CUR: case SEEK_END: | |
168 | return generic_file_llseek_size(file, offset, whence, | |
169 | size, size); | |
170 | default: | |
171 | return -EINVAL; | |
172 | } | |
173 | } | |
174 | EXPORT_SYMBOL(fixed_size_llseek); | |
175 | ||
ae6afc3f B |
176 | /** |
177 | * noop_llseek - No Operation Performed llseek implementation | |
178 | * @file: file structure to seek on | |
179 | * @offset: file offset to seek to | |
965c8e59 | 180 | * @whence: type of seek |
ae6afc3f B |
181 | * |
182 | * This is an implementation of ->llseek useable for the rare special case when | |
183 | * userspace expects the seek to succeed but the (device) file is actually not | |
184 | * able to perform the seek. In this case you use noop_llseek() instead of | |
185 | * falling back to the default implementation of ->llseek. | |
186 | */ | |
965c8e59 | 187 | loff_t noop_llseek(struct file *file, loff_t offset, int whence) |
ae6afc3f B |
188 | { |
189 | return file->f_pos; | |
190 | } | |
191 | EXPORT_SYMBOL(noop_llseek); | |
192 | ||
965c8e59 | 193 | loff_t no_llseek(struct file *file, loff_t offset, int whence) |
1da177e4 LT |
194 | { |
195 | return -ESPIPE; | |
196 | } | |
197 | EXPORT_SYMBOL(no_llseek); | |
198 | ||
965c8e59 | 199 | loff_t default_llseek(struct file *file, loff_t offset, int whence) |
1da177e4 | 200 | { |
496ad9aa | 201 | struct inode *inode = file_inode(file); |
16abef0e | 202 | loff_t retval; |
1da177e4 | 203 | |
982d8165 | 204 | mutex_lock(&inode->i_mutex); |
965c8e59 | 205 | switch (whence) { |
7b8e8924 | 206 | case SEEK_END: |
982d8165 | 207 | offset += i_size_read(inode); |
1da177e4 | 208 | break; |
7b8e8924 | 209 | case SEEK_CUR: |
5b6f1eb9 AK |
210 | if (offset == 0) { |
211 | retval = file->f_pos; | |
212 | goto out; | |
213 | } | |
1da177e4 | 214 | offset += file->f_pos; |
982d8165 JB |
215 | break; |
216 | case SEEK_DATA: | |
217 | /* | |
218 | * In the generic case the entire file is data, so as | |
219 | * long as offset isn't at the end of the file then the | |
220 | * offset is data. | |
221 | */ | |
bacb2d81 DC |
222 | if (offset >= inode->i_size) { |
223 | retval = -ENXIO; | |
224 | goto out; | |
225 | } | |
982d8165 JB |
226 | break; |
227 | case SEEK_HOLE: | |
228 | /* | |
229 | * There is a virtual hole at the end of the file, so | |
230 | * as long as offset isn't i_size or larger, return | |
231 | * i_size. | |
232 | */ | |
bacb2d81 DC |
233 | if (offset >= inode->i_size) { |
234 | retval = -ENXIO; | |
235 | goto out; | |
236 | } | |
982d8165 JB |
237 | offset = inode->i_size; |
238 | break; | |
1da177e4 LT |
239 | } |
240 | retval = -EINVAL; | |
cccb5a1e | 241 | if (offset >= 0 || unsigned_offsets(file)) { |
1da177e4 LT |
242 | if (offset != file->f_pos) { |
243 | file->f_pos = offset; | |
244 | file->f_version = 0; | |
245 | } | |
246 | retval = offset; | |
247 | } | |
5b6f1eb9 | 248 | out: |
982d8165 | 249 | mutex_unlock(&inode->i_mutex); |
1da177e4 LT |
250 | return retval; |
251 | } | |
252 | EXPORT_SYMBOL(default_llseek); | |
253 | ||
965c8e59 | 254 | loff_t vfs_llseek(struct file *file, loff_t offset, int whence) |
1da177e4 LT |
255 | { |
256 | loff_t (*fn)(struct file *, loff_t, int); | |
257 | ||
258 | fn = no_llseek; | |
259 | if (file->f_mode & FMODE_LSEEK) { | |
72c2d531 | 260 | if (file->f_op->llseek) |
1da177e4 LT |
261 | fn = file->f_op->llseek; |
262 | } | |
965c8e59 | 263 | return fn(file, offset, whence); |
1da177e4 LT |
264 | } |
265 | EXPORT_SYMBOL(vfs_llseek); | |
266 | ||
9c225f26 LT |
267 | /* |
268 | * We only lock f_pos if we have threads or if the file might be | |
269 | * shared with another process. In both cases we'll have an elevated | |
270 | * file count (done either by fdget() or by fork()). | |
271 | */ | |
272 | static inline struct fd fdget_pos(int fd) | |
273 | { | |
274 | struct fd f = fdget(fd); | |
275 | struct file *file = f.file; | |
276 | ||
277 | if (file && (file->f_mode & FMODE_ATOMIC_POS)) { | |
278 | if (file_count(file) > 1) { | |
279 | f.flags |= FDPUT_POS_UNLOCK; | |
280 | mutex_lock(&file->f_pos_lock); | |
281 | } | |
282 | } | |
283 | return f; | |
284 | } | |
285 | ||
286 | static inline void fdput_pos(struct fd f) | |
287 | { | |
288 | if (f.flags & FDPUT_POS_UNLOCK) | |
289 | mutex_unlock(&f.file->f_pos_lock); | |
290 | fdput(f); | |
291 | } | |
292 | ||
965c8e59 | 293 | SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence) |
1da177e4 LT |
294 | { |
295 | off_t retval; | |
9c225f26 | 296 | struct fd f = fdget_pos(fd); |
2903ff01 AV |
297 | if (!f.file) |
298 | return -EBADF; | |
1da177e4 LT |
299 | |
300 | retval = -EINVAL; | |
965c8e59 AM |
301 | if (whence <= SEEK_MAX) { |
302 | loff_t res = vfs_llseek(f.file, offset, whence); | |
1da177e4 LT |
303 | retval = res; |
304 | if (res != (loff_t)retval) | |
305 | retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */ | |
306 | } | |
9c225f26 | 307 | fdput_pos(f); |
1da177e4 LT |
308 | return retval; |
309 | } | |
310 | ||
561c6731 AV |
311 | #ifdef CONFIG_COMPAT |
312 | COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence) | |
313 | { | |
314 | return sys_lseek(fd, offset, whence); | |
315 | } | |
316 | #endif | |
317 | ||
1da177e4 | 318 | #ifdef __ARCH_WANT_SYS_LLSEEK |
003d7ab4 HC |
319 | SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high, |
320 | unsigned long, offset_low, loff_t __user *, result, | |
965c8e59 | 321 | unsigned int, whence) |
1da177e4 LT |
322 | { |
323 | int retval; | |
2903ff01 | 324 | struct fd f = fdget(fd); |
1da177e4 | 325 | loff_t offset; |
1da177e4 | 326 | |
2903ff01 AV |
327 | if (!f.file) |
328 | return -EBADF; | |
1da177e4 LT |
329 | |
330 | retval = -EINVAL; | |
965c8e59 | 331 | if (whence > SEEK_MAX) |
1da177e4 LT |
332 | goto out_putf; |
333 | ||
2903ff01 | 334 | offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low, |
965c8e59 | 335 | whence); |
1da177e4 LT |
336 | |
337 | retval = (int)offset; | |
338 | if (offset >= 0) { | |
339 | retval = -EFAULT; | |
340 | if (!copy_to_user(result, &offset, sizeof(offset))) | |
341 | retval = 0; | |
342 | } | |
343 | out_putf: | |
2903ff01 | 344 | fdput(f); |
1da177e4 LT |
345 | return retval; |
346 | } | |
347 | #endif | |
348 | ||
e28cc715 LT |
349 | /* |
350 | * rw_verify_area doesn't like huge counts. We limit | |
351 | * them to something that fits in "int" so that others | |
352 | * won't have to do range checks all the time. | |
353 | */ | |
68d70d03 | 354 | int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count) |
1da177e4 LT |
355 | { |
356 | struct inode *inode; | |
357 | loff_t pos; | |
c43e259c | 358 | int retval = -EINVAL; |
1da177e4 | 359 | |
496ad9aa | 360 | inode = file_inode(file); |
e28cc715 | 361 | if (unlikely((ssize_t) count < 0)) |
c43e259c | 362 | return retval; |
1da177e4 | 363 | pos = *ppos; |
cccb5a1e AV |
364 | if (unlikely(pos < 0)) { |
365 | if (!unsigned_offsets(file)) | |
366 | return retval; | |
367 | if (count >= -pos) /* both values are in 0..LLONG_MAX */ | |
368 | return -EOVERFLOW; | |
369 | } else if (unlikely((loff_t) (pos + count) < 0)) { | |
370 | if (!unsigned_offsets(file)) | |
4a3956c7 KH |
371 | return retval; |
372 | } | |
1da177e4 | 373 | |
a16877ca | 374 | if (unlikely(inode->i_flock && mandatory_lock(inode))) { |
c43e259c | 375 | retval = locks_mandatory_area( |
e28cc715 LT |
376 | read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE, |
377 | inode, file, pos, count); | |
378 | if (retval < 0) | |
379 | return retval; | |
380 | } | |
c43e259c JM |
381 | retval = security_file_permission(file, |
382 | read_write == READ ? MAY_READ : MAY_WRITE); | |
383 | if (retval) | |
384 | return retval; | |
e28cc715 | 385 | return count > MAX_RW_COUNT ? MAX_RW_COUNT : count; |
1da177e4 LT |
386 | } |
387 | ||
388 | ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos) | |
389 | { | |
027445c3 | 390 | struct iovec iov = { .iov_base = buf, .iov_len = len }; |
1da177e4 LT |
391 | struct kiocb kiocb; |
392 | ssize_t ret; | |
393 | ||
394 | init_sync_kiocb(&kiocb, filp); | |
395 | kiocb.ki_pos = *ppos; | |
61964eba | 396 | kiocb.ki_nbytes = len; |
027445c3 | 397 | |
41003a7b | 398 | ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos); |
1da177e4 LT |
399 | if (-EIOCBQUEUED == ret) |
400 | ret = wait_on_sync_kiocb(&kiocb); | |
401 | *ppos = kiocb.ki_pos; | |
402 | return ret; | |
403 | } | |
404 | ||
405 | EXPORT_SYMBOL(do_sync_read); | |
406 | ||
407 | ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos) | |
408 | { | |
409 | ssize_t ret; | |
410 | ||
411 | if (!(file->f_mode & FMODE_READ)) | |
412 | return -EBADF; | |
72c2d531 | 413 | if (!file->f_op->read && !file->f_op->aio_read) |
1da177e4 LT |
414 | return -EINVAL; |
415 | if (unlikely(!access_ok(VERIFY_WRITE, buf, count))) | |
416 | return -EFAULT; | |
417 | ||
418 | ret = rw_verify_area(READ, file, pos, count); | |
e28cc715 LT |
419 | if (ret >= 0) { |
420 | count = ret; | |
c43e259c JM |
421 | if (file->f_op->read) |
422 | ret = file->f_op->read(file, buf, count, pos); | |
423 | else | |
424 | ret = do_sync_read(file, buf, count, pos); | |
425 | if (ret > 0) { | |
2a12a9d7 | 426 | fsnotify_access(file); |
c43e259c | 427 | add_rchar(current, ret); |
1da177e4 | 428 | } |
c43e259c | 429 | inc_syscr(current); |
1da177e4 LT |
430 | } |
431 | ||
432 | return ret; | |
433 | } | |
434 | ||
435 | EXPORT_SYMBOL(vfs_read); | |
436 | ||
437 | ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos) | |
438 | { | |
027445c3 | 439 | struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len }; |
1da177e4 LT |
440 | struct kiocb kiocb; |
441 | ssize_t ret; | |
442 | ||
443 | init_sync_kiocb(&kiocb, filp); | |
444 | kiocb.ki_pos = *ppos; | |
61964eba | 445 | kiocb.ki_nbytes = len; |
027445c3 | 446 | |
41003a7b | 447 | ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos); |
1da177e4 LT |
448 | if (-EIOCBQUEUED == ret) |
449 | ret = wait_on_sync_kiocb(&kiocb); | |
450 | *ppos = kiocb.ki_pos; | |
451 | return ret; | |
452 | } | |
453 | ||
454 | EXPORT_SYMBOL(do_sync_write); | |
455 | ||
06ae43f3 AV |
456 | ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos) |
457 | { | |
458 | mm_segment_t old_fs; | |
459 | const char __user *p; | |
460 | ssize_t ret; | |
461 | ||
72c2d531 | 462 | if (!file->f_op->write && !file->f_op->aio_write) |
3e84f48e AV |
463 | return -EINVAL; |
464 | ||
06ae43f3 AV |
465 | old_fs = get_fs(); |
466 | set_fs(get_ds()); | |
467 | p = (__force const char __user *)buf; | |
468 | if (count > MAX_RW_COUNT) | |
469 | count = MAX_RW_COUNT; | |
470 | if (file->f_op->write) | |
471 | ret = file->f_op->write(file, p, count, pos); | |
472 | else | |
473 | ret = do_sync_write(file, p, count, pos); | |
474 | set_fs(old_fs); | |
475 | if (ret > 0) { | |
476 | fsnotify_modify(file); | |
477 | add_wchar(current, ret); | |
478 | } | |
479 | inc_syscw(current); | |
480 | return ret; | |
481 | } | |
482 | ||
1da177e4 LT |
483 | ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos) |
484 | { | |
485 | ssize_t ret; | |
486 | ||
487 | if (!(file->f_mode & FMODE_WRITE)) | |
488 | return -EBADF; | |
72c2d531 | 489 | if (!file->f_op->write && !file->f_op->aio_write) |
1da177e4 LT |
490 | return -EINVAL; |
491 | if (unlikely(!access_ok(VERIFY_READ, buf, count))) | |
492 | return -EFAULT; | |
493 | ||
494 | ret = rw_verify_area(WRITE, file, pos, count); | |
e28cc715 LT |
495 | if (ret >= 0) { |
496 | count = ret; | |
03d95eb2 | 497 | file_start_write(file); |
c43e259c JM |
498 | if (file->f_op->write) |
499 | ret = file->f_op->write(file, buf, count, pos); | |
500 | else | |
501 | ret = do_sync_write(file, buf, count, pos); | |
502 | if (ret > 0) { | |
2a12a9d7 | 503 | fsnotify_modify(file); |
c43e259c | 504 | add_wchar(current, ret); |
1da177e4 | 505 | } |
c43e259c | 506 | inc_syscw(current); |
03d95eb2 | 507 | file_end_write(file); |
1da177e4 LT |
508 | } |
509 | ||
510 | return ret; | |
511 | } | |
512 | ||
513 | EXPORT_SYMBOL(vfs_write); | |
514 | ||
515 | static inline loff_t file_pos_read(struct file *file) | |
516 | { | |
517 | return file->f_pos; | |
518 | } | |
519 | ||
520 | static inline void file_pos_write(struct file *file, loff_t pos) | |
521 | { | |
522 | file->f_pos = pos; | |
523 | } | |
524 | ||
3cdad428 | 525 | SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count) |
1da177e4 | 526 | { |
9c225f26 | 527 | struct fd f = fdget_pos(fd); |
1da177e4 | 528 | ssize_t ret = -EBADF; |
1da177e4 | 529 | |
2903ff01 AV |
530 | if (f.file) { |
531 | loff_t pos = file_pos_read(f.file); | |
532 | ret = vfs_read(f.file, buf, count, &pos); | |
5faf153e AV |
533 | if (ret >= 0) |
534 | file_pos_write(f.file, pos); | |
9c225f26 | 535 | fdput_pos(f); |
1da177e4 | 536 | } |
1da177e4 LT |
537 | return ret; |
538 | } | |
1da177e4 | 539 | |
3cdad428 HC |
540 | SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf, |
541 | size_t, count) | |
1da177e4 | 542 | { |
9c225f26 | 543 | struct fd f = fdget_pos(fd); |
1da177e4 | 544 | ssize_t ret = -EBADF; |
1da177e4 | 545 | |
2903ff01 AV |
546 | if (f.file) { |
547 | loff_t pos = file_pos_read(f.file); | |
548 | ret = vfs_write(f.file, buf, count, &pos); | |
5faf153e AV |
549 | if (ret >= 0) |
550 | file_pos_write(f.file, pos); | |
9c225f26 | 551 | fdput_pos(f); |
1da177e4 LT |
552 | } |
553 | ||
554 | return ret; | |
555 | } | |
556 | ||
4a0fd5bf AV |
557 | SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf, |
558 | size_t, count, loff_t, pos) | |
1da177e4 | 559 | { |
2903ff01 | 560 | struct fd f; |
1da177e4 | 561 | ssize_t ret = -EBADF; |
1da177e4 LT |
562 | |
563 | if (pos < 0) | |
564 | return -EINVAL; | |
565 | ||
2903ff01 AV |
566 | f = fdget(fd); |
567 | if (f.file) { | |
1da177e4 | 568 | ret = -ESPIPE; |
2903ff01 AV |
569 | if (f.file->f_mode & FMODE_PREAD) |
570 | ret = vfs_read(f.file, buf, count, &pos); | |
571 | fdput(f); | |
1da177e4 LT |
572 | } |
573 | ||
574 | return ret; | |
575 | } | |
576 | ||
4a0fd5bf AV |
577 | SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf, |
578 | size_t, count, loff_t, pos) | |
1da177e4 | 579 | { |
2903ff01 | 580 | struct fd f; |
1da177e4 | 581 | ssize_t ret = -EBADF; |
1da177e4 LT |
582 | |
583 | if (pos < 0) | |
584 | return -EINVAL; | |
585 | ||
2903ff01 AV |
586 | f = fdget(fd); |
587 | if (f.file) { | |
1da177e4 | 588 | ret = -ESPIPE; |
2903ff01 AV |
589 | if (f.file->f_mode & FMODE_PWRITE) |
590 | ret = vfs_write(f.file, buf, count, &pos); | |
591 | fdput(f); | |
1da177e4 LT |
592 | } |
593 | ||
594 | return ret; | |
595 | } | |
596 | ||
597 | /* | |
598 | * Reduce an iovec's length in-place. Return the resulting number of segments | |
599 | */ | |
600 | unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to) | |
601 | { | |
602 | unsigned long seg = 0; | |
603 | size_t len = 0; | |
604 | ||
605 | while (seg < nr_segs) { | |
606 | seg++; | |
607 | if (len + iov->iov_len >= to) { | |
608 | iov->iov_len = to - len; | |
609 | break; | |
610 | } | |
611 | len += iov->iov_len; | |
612 | iov++; | |
613 | } | |
614 | return seg; | |
615 | } | |
19295529 | 616 | EXPORT_SYMBOL(iov_shorten); |
1da177e4 | 617 | |
72ec3516 | 618 | static ssize_t do_sync_readv_writev(struct file *filp, const struct iovec *iov, |
ee0b3e67 BP |
619 | unsigned long nr_segs, size_t len, loff_t *ppos, iov_fn_t fn) |
620 | { | |
621 | struct kiocb kiocb; | |
622 | ssize_t ret; | |
623 | ||
624 | init_sync_kiocb(&kiocb, filp); | |
625 | kiocb.ki_pos = *ppos; | |
ee0b3e67 BP |
626 | kiocb.ki_nbytes = len; |
627 | ||
41003a7b | 628 | ret = fn(&kiocb, iov, nr_segs, kiocb.ki_pos); |
ee0b3e67 BP |
629 | if (ret == -EIOCBQUEUED) |
630 | ret = wait_on_sync_kiocb(&kiocb); | |
631 | *ppos = kiocb.ki_pos; | |
632 | return ret; | |
633 | } | |
634 | ||
635 | /* Do it by hand, with file-ops */ | |
72ec3516 | 636 | static ssize_t do_loop_readv_writev(struct file *filp, struct iovec *iov, |
ee0b3e67 BP |
637 | unsigned long nr_segs, loff_t *ppos, io_fn_t fn) |
638 | { | |
639 | struct iovec *vector = iov; | |
640 | ssize_t ret = 0; | |
641 | ||
642 | while (nr_segs > 0) { | |
643 | void __user *base; | |
644 | size_t len; | |
645 | ssize_t nr; | |
646 | ||
647 | base = vector->iov_base; | |
648 | len = vector->iov_len; | |
649 | vector++; | |
650 | nr_segs--; | |
651 | ||
652 | nr = fn(filp, base, len, ppos); | |
653 | ||
654 | if (nr < 0) { | |
655 | if (!ret) | |
656 | ret = nr; | |
657 | break; | |
658 | } | |
659 | ret += nr; | |
660 | if (nr != len) | |
661 | break; | |
662 | } | |
663 | ||
664 | return ret; | |
665 | } | |
666 | ||
1da177e4 LT |
667 | /* A write operation does a read from user space and vice versa */ |
668 | #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ) | |
669 | ||
eed4e51f BP |
670 | ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector, |
671 | unsigned long nr_segs, unsigned long fast_segs, | |
672 | struct iovec *fast_pointer, | |
ac34ebb3 | 673 | struct iovec **ret_pointer) |
435f49a5 | 674 | { |
eed4e51f | 675 | unsigned long seg; |
435f49a5 | 676 | ssize_t ret; |
eed4e51f BP |
677 | struct iovec *iov = fast_pointer; |
678 | ||
435f49a5 LT |
679 | /* |
680 | * SuS says "The readv() function *may* fail if the iovcnt argument | |
681 | * was less than or equal to 0, or greater than {IOV_MAX}. Linux has | |
682 | * traditionally returned zero for zero segments, so... | |
683 | */ | |
eed4e51f BP |
684 | if (nr_segs == 0) { |
685 | ret = 0; | |
435f49a5 | 686 | goto out; |
eed4e51f BP |
687 | } |
688 | ||
435f49a5 LT |
689 | /* |
690 | * First get the "struct iovec" from user memory and | |
691 | * verify all the pointers | |
692 | */ | |
eed4e51f BP |
693 | if (nr_segs > UIO_MAXIOV) { |
694 | ret = -EINVAL; | |
435f49a5 | 695 | goto out; |
eed4e51f BP |
696 | } |
697 | if (nr_segs > fast_segs) { | |
435f49a5 | 698 | iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL); |
eed4e51f BP |
699 | if (iov == NULL) { |
700 | ret = -ENOMEM; | |
435f49a5 | 701 | goto out; |
eed4e51f | 702 | } |
435f49a5 | 703 | } |
eed4e51f BP |
704 | if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) { |
705 | ret = -EFAULT; | |
435f49a5 | 706 | goto out; |
eed4e51f BP |
707 | } |
708 | ||
435f49a5 | 709 | /* |
eed4e51f BP |
710 | * According to the Single Unix Specification we should return EINVAL |
711 | * if an element length is < 0 when cast to ssize_t or if the | |
712 | * total length would overflow the ssize_t return value of the | |
713 | * system call. | |
435f49a5 LT |
714 | * |
715 | * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the | |
716 | * overflow case. | |
717 | */ | |
eed4e51f | 718 | ret = 0; |
435f49a5 LT |
719 | for (seg = 0; seg < nr_segs; seg++) { |
720 | void __user *buf = iov[seg].iov_base; | |
721 | ssize_t len = (ssize_t)iov[seg].iov_len; | |
eed4e51f BP |
722 | |
723 | /* see if we we're about to use an invalid len or if | |
724 | * it's about to overflow ssize_t */ | |
435f49a5 | 725 | if (len < 0) { |
eed4e51f | 726 | ret = -EINVAL; |
435f49a5 | 727 | goto out; |
eed4e51f | 728 | } |
ac34ebb3 | 729 | if (type >= 0 |
fcf63409 | 730 | && unlikely(!access_ok(vrfy_dir(type), buf, len))) { |
eed4e51f | 731 | ret = -EFAULT; |
435f49a5 LT |
732 | goto out; |
733 | } | |
734 | if (len > MAX_RW_COUNT - ret) { | |
735 | len = MAX_RW_COUNT - ret; | |
736 | iov[seg].iov_len = len; | |
eed4e51f | 737 | } |
eed4e51f | 738 | ret += len; |
435f49a5 | 739 | } |
eed4e51f BP |
740 | out: |
741 | *ret_pointer = iov; | |
742 | return ret; | |
743 | } | |
744 | ||
1da177e4 LT |
745 | static ssize_t do_readv_writev(int type, struct file *file, |
746 | const struct iovec __user * uvector, | |
747 | unsigned long nr_segs, loff_t *pos) | |
748 | { | |
1da177e4 LT |
749 | size_t tot_len; |
750 | struct iovec iovstack[UIO_FASTIOV]; | |
ee0b3e67 | 751 | struct iovec *iov = iovstack; |
1da177e4 | 752 | ssize_t ret; |
1da177e4 LT |
753 | io_fn_t fn; |
754 | iov_fn_t fnv; | |
755 | ||
eed4e51f | 756 | ret = rw_copy_check_uvector(type, uvector, nr_segs, |
ac34ebb3 | 757 | ARRAY_SIZE(iovstack), iovstack, &iov); |
eed4e51f | 758 | if (ret <= 0) |
1da177e4 | 759 | goto out; |
1da177e4 | 760 | |
eed4e51f | 761 | tot_len = ret; |
1da177e4 | 762 | ret = rw_verify_area(type, file, pos, tot_len); |
e28cc715 | 763 | if (ret < 0) |
411b67b4 | 764 | goto out; |
1da177e4 LT |
765 | |
766 | fnv = NULL; | |
767 | if (type == READ) { | |
768 | fn = file->f_op->read; | |
ee0b3e67 | 769 | fnv = file->f_op->aio_read; |
1da177e4 LT |
770 | } else { |
771 | fn = (io_fn_t)file->f_op->write; | |
ee0b3e67 | 772 | fnv = file->f_op->aio_write; |
03d95eb2 | 773 | file_start_write(file); |
1da177e4 LT |
774 | } |
775 | ||
ee0b3e67 BP |
776 | if (fnv) |
777 | ret = do_sync_readv_writev(file, iov, nr_segs, tot_len, | |
778 | pos, fnv); | |
779 | else | |
780 | ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn); | |
1da177e4 | 781 | |
03d95eb2 AV |
782 | if (type != READ) |
783 | file_end_write(file); | |
784 | ||
1da177e4 LT |
785 | out: |
786 | if (iov != iovstack) | |
787 | kfree(iov); | |
0eeca283 RL |
788 | if ((ret + (type == READ)) > 0) { |
789 | if (type == READ) | |
2a12a9d7 | 790 | fsnotify_access(file); |
0eeca283 | 791 | else |
2a12a9d7 | 792 | fsnotify_modify(file); |
0eeca283 | 793 | } |
1da177e4 | 794 | return ret; |
1da177e4 LT |
795 | } |
796 | ||
797 | ssize_t vfs_readv(struct file *file, const struct iovec __user *vec, | |
798 | unsigned long vlen, loff_t *pos) | |
799 | { | |
800 | if (!(file->f_mode & FMODE_READ)) | |
801 | return -EBADF; | |
72c2d531 | 802 | if (!file->f_op->aio_read && !file->f_op->read) |
1da177e4 LT |
803 | return -EINVAL; |
804 | ||
805 | return do_readv_writev(READ, file, vec, vlen, pos); | |
806 | } | |
807 | ||
808 | EXPORT_SYMBOL(vfs_readv); | |
809 | ||
810 | ssize_t vfs_writev(struct file *file, const struct iovec __user *vec, | |
811 | unsigned long vlen, loff_t *pos) | |
812 | { | |
813 | if (!(file->f_mode & FMODE_WRITE)) | |
814 | return -EBADF; | |
72c2d531 | 815 | if (!file->f_op->aio_write && !file->f_op->write) |
1da177e4 LT |
816 | return -EINVAL; |
817 | ||
818 | return do_readv_writev(WRITE, file, vec, vlen, pos); | |
819 | } | |
820 | ||
821 | EXPORT_SYMBOL(vfs_writev); | |
822 | ||
3cdad428 HC |
823 | SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec, |
824 | unsigned long, vlen) | |
1da177e4 | 825 | { |
9c225f26 | 826 | struct fd f = fdget_pos(fd); |
1da177e4 | 827 | ssize_t ret = -EBADF; |
1da177e4 | 828 | |
2903ff01 AV |
829 | if (f.file) { |
830 | loff_t pos = file_pos_read(f.file); | |
831 | ret = vfs_readv(f.file, vec, vlen, &pos); | |
5faf153e AV |
832 | if (ret >= 0) |
833 | file_pos_write(f.file, pos); | |
9c225f26 | 834 | fdput_pos(f); |
1da177e4 LT |
835 | } |
836 | ||
837 | if (ret > 0) | |
4b98d11b AD |
838 | add_rchar(current, ret); |
839 | inc_syscr(current); | |
1da177e4 LT |
840 | return ret; |
841 | } | |
842 | ||
3cdad428 HC |
843 | SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec, |
844 | unsigned long, vlen) | |
1da177e4 | 845 | { |
9c225f26 | 846 | struct fd f = fdget_pos(fd); |
1da177e4 | 847 | ssize_t ret = -EBADF; |
1da177e4 | 848 | |
2903ff01 AV |
849 | if (f.file) { |
850 | loff_t pos = file_pos_read(f.file); | |
851 | ret = vfs_writev(f.file, vec, vlen, &pos); | |
5faf153e AV |
852 | if (ret >= 0) |
853 | file_pos_write(f.file, pos); | |
9c225f26 | 854 | fdput_pos(f); |
1da177e4 LT |
855 | } |
856 | ||
857 | if (ret > 0) | |
4b98d11b AD |
858 | add_wchar(current, ret); |
859 | inc_syscw(current); | |
1da177e4 LT |
860 | return ret; |
861 | } | |
862 | ||
601cc11d LT |
863 | static inline loff_t pos_from_hilo(unsigned long high, unsigned long low) |
864 | { | |
865 | #define HALF_LONG_BITS (BITS_PER_LONG / 2) | |
866 | return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low; | |
867 | } | |
868 | ||
f3554f4b | 869 | SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec, |
601cc11d | 870 | unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h) |
f3554f4b | 871 | { |
601cc11d | 872 | loff_t pos = pos_from_hilo(pos_h, pos_l); |
2903ff01 | 873 | struct fd f; |
f3554f4b | 874 | ssize_t ret = -EBADF; |
f3554f4b GH |
875 | |
876 | if (pos < 0) | |
877 | return -EINVAL; | |
878 | ||
2903ff01 AV |
879 | f = fdget(fd); |
880 | if (f.file) { | |
f3554f4b | 881 | ret = -ESPIPE; |
2903ff01 AV |
882 | if (f.file->f_mode & FMODE_PREAD) |
883 | ret = vfs_readv(f.file, vec, vlen, &pos); | |
884 | fdput(f); | |
f3554f4b GH |
885 | } |
886 | ||
887 | if (ret > 0) | |
888 | add_rchar(current, ret); | |
889 | inc_syscr(current); | |
890 | return ret; | |
891 | } | |
892 | ||
893 | SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec, | |
601cc11d | 894 | unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h) |
f3554f4b | 895 | { |
601cc11d | 896 | loff_t pos = pos_from_hilo(pos_h, pos_l); |
2903ff01 | 897 | struct fd f; |
f3554f4b | 898 | ssize_t ret = -EBADF; |
f3554f4b GH |
899 | |
900 | if (pos < 0) | |
901 | return -EINVAL; | |
902 | ||
2903ff01 AV |
903 | f = fdget(fd); |
904 | if (f.file) { | |
f3554f4b | 905 | ret = -ESPIPE; |
2903ff01 AV |
906 | if (f.file->f_mode & FMODE_PWRITE) |
907 | ret = vfs_writev(f.file, vec, vlen, &pos); | |
908 | fdput(f); | |
f3554f4b GH |
909 | } |
910 | ||
911 | if (ret > 0) | |
912 | add_wchar(current, ret); | |
913 | inc_syscw(current); | |
914 | return ret; | |
915 | } | |
916 | ||
72ec3516 AV |
917 | #ifdef CONFIG_COMPAT |
918 | ||
919 | static ssize_t compat_do_readv_writev(int type, struct file *file, | |
920 | const struct compat_iovec __user *uvector, | |
921 | unsigned long nr_segs, loff_t *pos) | |
922 | { | |
923 | compat_ssize_t tot_len; | |
924 | struct iovec iovstack[UIO_FASTIOV]; | |
925 | struct iovec *iov = iovstack; | |
926 | ssize_t ret; | |
927 | io_fn_t fn; | |
928 | iov_fn_t fnv; | |
929 | ||
72ec3516 AV |
930 | ret = compat_rw_copy_check_uvector(type, uvector, nr_segs, |
931 | UIO_FASTIOV, iovstack, &iov); | |
932 | if (ret <= 0) | |
933 | goto out; | |
934 | ||
935 | tot_len = ret; | |
936 | ret = rw_verify_area(type, file, pos, tot_len); | |
937 | if (ret < 0) | |
938 | goto out; | |
939 | ||
940 | fnv = NULL; | |
941 | if (type == READ) { | |
942 | fn = file->f_op->read; | |
943 | fnv = file->f_op->aio_read; | |
944 | } else { | |
945 | fn = (io_fn_t)file->f_op->write; | |
946 | fnv = file->f_op->aio_write; | |
03d95eb2 | 947 | file_start_write(file); |
72ec3516 AV |
948 | } |
949 | ||
03d95eb2 | 950 | if (fnv) |
72ec3516 AV |
951 | ret = do_sync_readv_writev(file, iov, nr_segs, tot_len, |
952 | pos, fnv); | |
03d95eb2 | 953 | else |
72ec3516 AV |
954 | ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn); |
955 | ||
03d95eb2 AV |
956 | if (type != READ) |
957 | file_end_write(file); | |
958 | ||
72ec3516 AV |
959 | out: |
960 | if (iov != iovstack) | |
961 | kfree(iov); | |
962 | if ((ret + (type == READ)) > 0) { | |
963 | if (type == READ) | |
964 | fsnotify_access(file); | |
965 | else | |
966 | fsnotify_modify(file); | |
967 | } | |
968 | return ret; | |
969 | } | |
970 | ||
971 | static size_t compat_readv(struct file *file, | |
972 | const struct compat_iovec __user *vec, | |
973 | unsigned long vlen, loff_t *pos) | |
974 | { | |
975 | ssize_t ret = -EBADF; | |
976 | ||
977 | if (!(file->f_mode & FMODE_READ)) | |
978 | goto out; | |
979 | ||
980 | ret = -EINVAL; | |
72c2d531 | 981 | if (!file->f_op->aio_read && !file->f_op->read) |
72ec3516 AV |
982 | goto out; |
983 | ||
984 | ret = compat_do_readv_writev(READ, file, vec, vlen, pos); | |
985 | ||
986 | out: | |
987 | if (ret > 0) | |
988 | add_rchar(current, ret); | |
989 | inc_syscr(current); | |
990 | return ret; | |
991 | } | |
992 | ||
dfd948e3 | 993 | COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd, |
72ec3516 | 994 | const struct compat_iovec __user *,vec, |
dfd948e3 | 995 | compat_ulong_t, vlen) |
72ec3516 | 996 | { |
9c225f26 | 997 | struct fd f = fdget_pos(fd); |
72ec3516 AV |
998 | ssize_t ret; |
999 | loff_t pos; | |
1000 | ||
1001 | if (!f.file) | |
1002 | return -EBADF; | |
1003 | pos = f.file->f_pos; | |
1004 | ret = compat_readv(f.file, vec, vlen, &pos); | |
5faf153e AV |
1005 | if (ret >= 0) |
1006 | f.file->f_pos = pos; | |
9c225f26 | 1007 | fdput_pos(f); |
72ec3516 AV |
1008 | return ret; |
1009 | } | |
1010 | ||
1011 | COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd, | |
1012 | const struct compat_iovec __user *,vec, | |
1013 | unsigned long, vlen, loff_t, pos) | |
1014 | { | |
1015 | struct fd f; | |
1016 | ssize_t ret; | |
1017 | ||
1018 | if (pos < 0) | |
1019 | return -EINVAL; | |
1020 | f = fdget(fd); | |
1021 | if (!f.file) | |
1022 | return -EBADF; | |
1023 | ret = -ESPIPE; | |
1024 | if (f.file->f_mode & FMODE_PREAD) | |
1025 | ret = compat_readv(f.file, vec, vlen, &pos); | |
1026 | fdput(f); | |
1027 | return ret; | |
1028 | } | |
1029 | ||
dfd948e3 | 1030 | COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd, |
72ec3516 | 1031 | const struct compat_iovec __user *,vec, |
dfd948e3 | 1032 | compat_ulong_t, vlen, u32, pos_low, u32, pos_high) |
72ec3516 AV |
1033 | { |
1034 | loff_t pos = ((loff_t)pos_high << 32) | pos_low; | |
1035 | return compat_sys_preadv64(fd, vec, vlen, pos); | |
1036 | } | |
1037 | ||
1038 | static size_t compat_writev(struct file *file, | |
1039 | const struct compat_iovec __user *vec, | |
1040 | unsigned long vlen, loff_t *pos) | |
1041 | { | |
1042 | ssize_t ret = -EBADF; | |
1043 | ||
1044 | if (!(file->f_mode & FMODE_WRITE)) | |
1045 | goto out; | |
1046 | ||
1047 | ret = -EINVAL; | |
72c2d531 | 1048 | if (!file->f_op->aio_write && !file->f_op->write) |
72ec3516 AV |
1049 | goto out; |
1050 | ||
1051 | ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos); | |
1052 | ||
1053 | out: | |
1054 | if (ret > 0) | |
1055 | add_wchar(current, ret); | |
1056 | inc_syscw(current); | |
1057 | return ret; | |
1058 | } | |
1059 | ||
dfd948e3 | 1060 | COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd, |
72ec3516 | 1061 | const struct compat_iovec __user *, vec, |
dfd948e3 | 1062 | compat_ulong_t, vlen) |
72ec3516 | 1063 | { |
9c225f26 | 1064 | struct fd f = fdget_pos(fd); |
72ec3516 AV |
1065 | ssize_t ret; |
1066 | loff_t pos; | |
1067 | ||
1068 | if (!f.file) | |
1069 | return -EBADF; | |
1070 | pos = f.file->f_pos; | |
1071 | ret = compat_writev(f.file, vec, vlen, &pos); | |
5faf153e AV |
1072 | if (ret >= 0) |
1073 | f.file->f_pos = pos; | |
9c225f26 | 1074 | fdput_pos(f); |
72ec3516 AV |
1075 | return ret; |
1076 | } | |
1077 | ||
1078 | COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd, | |
1079 | const struct compat_iovec __user *,vec, | |
1080 | unsigned long, vlen, loff_t, pos) | |
1081 | { | |
1082 | struct fd f; | |
1083 | ssize_t ret; | |
1084 | ||
1085 | if (pos < 0) | |
1086 | return -EINVAL; | |
1087 | f = fdget(fd); | |
1088 | if (!f.file) | |
1089 | return -EBADF; | |
1090 | ret = -ESPIPE; | |
1091 | if (f.file->f_mode & FMODE_PWRITE) | |
1092 | ret = compat_writev(f.file, vec, vlen, &pos); | |
1093 | fdput(f); | |
1094 | return ret; | |
1095 | } | |
1096 | ||
dfd948e3 | 1097 | COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd, |
72ec3516 | 1098 | const struct compat_iovec __user *,vec, |
dfd948e3 | 1099 | compat_ulong_t, vlen, u32, pos_low, u32, pos_high) |
72ec3516 AV |
1100 | { |
1101 | loff_t pos = ((loff_t)pos_high << 32) | pos_low; | |
1102 | return compat_sys_pwritev64(fd, vec, vlen, pos); | |
1103 | } | |
1104 | #endif | |
1105 | ||
19f4fc3a AV |
1106 | static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos, |
1107 | size_t count, loff_t max) | |
1da177e4 | 1108 | { |
2903ff01 AV |
1109 | struct fd in, out; |
1110 | struct inode *in_inode, *out_inode; | |
1da177e4 | 1111 | loff_t pos; |
7995bd28 | 1112 | loff_t out_pos; |
1da177e4 | 1113 | ssize_t retval; |
2903ff01 | 1114 | int fl; |
1da177e4 LT |
1115 | |
1116 | /* | |
1117 | * Get input file, and verify that it is ok.. | |
1118 | */ | |
1119 | retval = -EBADF; | |
2903ff01 AV |
1120 | in = fdget(in_fd); |
1121 | if (!in.file) | |
1da177e4 | 1122 | goto out; |
2903ff01 | 1123 | if (!(in.file->f_mode & FMODE_READ)) |
1da177e4 | 1124 | goto fput_in; |
1da177e4 | 1125 | retval = -ESPIPE; |
7995bd28 AV |
1126 | if (!ppos) { |
1127 | pos = in.file->f_pos; | |
1128 | } else { | |
1129 | pos = *ppos; | |
2903ff01 | 1130 | if (!(in.file->f_mode & FMODE_PREAD)) |
1da177e4 | 1131 | goto fput_in; |
7995bd28 AV |
1132 | } |
1133 | retval = rw_verify_area(READ, in.file, &pos, count); | |
e28cc715 | 1134 | if (retval < 0) |
1da177e4 | 1135 | goto fput_in; |
e28cc715 | 1136 | count = retval; |
1da177e4 | 1137 | |
1da177e4 LT |
1138 | /* |
1139 | * Get output file, and verify that it is ok.. | |
1140 | */ | |
1141 | retval = -EBADF; | |
2903ff01 AV |
1142 | out = fdget(out_fd); |
1143 | if (!out.file) | |
1da177e4 | 1144 | goto fput_in; |
2903ff01 | 1145 | if (!(out.file->f_mode & FMODE_WRITE)) |
1da177e4 LT |
1146 | goto fput_out; |
1147 | retval = -EINVAL; | |
496ad9aa AV |
1148 | in_inode = file_inode(in.file); |
1149 | out_inode = file_inode(out.file); | |
7995bd28 AV |
1150 | out_pos = out.file->f_pos; |
1151 | retval = rw_verify_area(WRITE, out.file, &out_pos, count); | |
e28cc715 | 1152 | if (retval < 0) |
1da177e4 | 1153 | goto fput_out; |
e28cc715 | 1154 | count = retval; |
1da177e4 | 1155 | |
1da177e4 LT |
1156 | if (!max) |
1157 | max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes); | |
1158 | ||
1da177e4 LT |
1159 | if (unlikely(pos + count > max)) { |
1160 | retval = -EOVERFLOW; | |
1161 | if (pos >= max) | |
1162 | goto fput_out; | |
1163 | count = max - pos; | |
1164 | } | |
1165 | ||
d96e6e71 | 1166 | fl = 0; |
534f2aaa | 1167 | #if 0 |
d96e6e71 JA |
1168 | /* |
1169 | * We need to debate whether we can enable this or not. The | |
1170 | * man page documents EAGAIN return for the output at least, | |
1171 | * and the application is arguably buggy if it doesn't expect | |
1172 | * EAGAIN on a non-blocking file descriptor. | |
1173 | */ | |
2903ff01 | 1174 | if (in.file->f_flags & O_NONBLOCK) |
d96e6e71 | 1175 | fl = SPLICE_F_NONBLOCK; |
534f2aaa | 1176 | #endif |
50cd2c57 | 1177 | file_start_write(out.file); |
7995bd28 | 1178 | retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl); |
50cd2c57 | 1179 | file_end_write(out.file); |
1da177e4 LT |
1180 | |
1181 | if (retval > 0) { | |
4b98d11b AD |
1182 | add_rchar(current, retval); |
1183 | add_wchar(current, retval); | |
a68c2f12 SW |
1184 | fsnotify_access(in.file); |
1185 | fsnotify_modify(out.file); | |
7995bd28 AV |
1186 | out.file->f_pos = out_pos; |
1187 | if (ppos) | |
1188 | *ppos = pos; | |
1189 | else | |
1190 | in.file->f_pos = pos; | |
1da177e4 | 1191 | } |
1da177e4 | 1192 | |
4b98d11b AD |
1193 | inc_syscr(current); |
1194 | inc_syscw(current); | |
7995bd28 | 1195 | if (pos > max) |
1da177e4 LT |
1196 | retval = -EOVERFLOW; |
1197 | ||
1198 | fput_out: | |
2903ff01 | 1199 | fdput(out); |
1da177e4 | 1200 | fput_in: |
2903ff01 | 1201 | fdput(in); |
1da177e4 LT |
1202 | out: |
1203 | return retval; | |
1204 | } | |
1205 | ||
002c8976 | 1206 | SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count) |
1da177e4 LT |
1207 | { |
1208 | loff_t pos; | |
1209 | off_t off; | |
1210 | ssize_t ret; | |
1211 | ||
1212 | if (offset) { | |
1213 | if (unlikely(get_user(off, offset))) | |
1214 | return -EFAULT; | |
1215 | pos = off; | |
1216 | ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS); | |
1217 | if (unlikely(put_user(pos, offset))) | |
1218 | return -EFAULT; | |
1219 | return ret; | |
1220 | } | |
1221 | ||
1222 | return do_sendfile(out_fd, in_fd, NULL, count, 0); | |
1223 | } | |
1224 | ||
002c8976 | 1225 | SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count) |
1da177e4 LT |
1226 | { |
1227 | loff_t pos; | |
1228 | ssize_t ret; | |
1229 | ||
1230 | if (offset) { | |
1231 | if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t)))) | |
1232 | return -EFAULT; | |
1233 | ret = do_sendfile(out_fd, in_fd, &pos, count, 0); | |
1234 | if (unlikely(put_user(pos, offset))) | |
1235 | return -EFAULT; | |
1236 | return ret; | |
1237 | } | |
1238 | ||
1239 | return do_sendfile(out_fd, in_fd, NULL, count, 0); | |
1240 | } | |
19f4fc3a AV |
1241 | |
1242 | #ifdef CONFIG_COMPAT | |
1243 | COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, | |
1244 | compat_off_t __user *, offset, compat_size_t, count) | |
1245 | { | |
1246 | loff_t pos; | |
1247 | off_t off; | |
1248 | ssize_t ret; | |
1249 | ||
1250 | if (offset) { | |
1251 | if (unlikely(get_user(off, offset))) | |
1252 | return -EFAULT; | |
1253 | pos = off; | |
1254 | ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS); | |
1255 | if (unlikely(put_user(pos, offset))) | |
1256 | return -EFAULT; | |
1257 | return ret; | |
1258 | } | |
1259 | ||
1260 | return do_sendfile(out_fd, in_fd, NULL, count, 0); | |
1261 | } | |
1262 | ||
1263 | COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, | |
1264 | compat_loff_t __user *, offset, compat_size_t, count) | |
1265 | { | |
1266 | loff_t pos; | |
1267 | ssize_t ret; | |
1268 | ||
1269 | if (offset) { | |
1270 | if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t)))) | |
1271 | return -EFAULT; | |
1272 | ret = do_sendfile(out_fd, in_fd, &pos, count, 0); | |
1273 | if (unlikely(put_user(pos, offset))) | |
1274 | return -EFAULT; | |
1275 | return ret; | |
1276 | } | |
1277 | ||
1278 | return do_sendfile(out_fd, in_fd, NULL, count, 0); | |
1279 | } | |
1280 | #endif |