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