Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/read_write.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992 Linus Torvalds | |
5 | */ | |
6 | ||
7 | #include <linux/slab.h> | |
8 | #include <linux/stat.h> | |
9 | #include <linux/fcntl.h> | |
10 | #include <linux/file.h> | |
11 | #include <linux/uio.h> | |
0eeca283 | 12 | #include <linux/fsnotify.h> |
1da177e4 LT |
13 | #include <linux/security.h> |
14 | #include <linux/module.h> | |
15 | #include <linux/syscalls.h> | |
e28cc715 | 16 | #include <linux/pagemap.h> |
d6b29d7c | 17 | #include <linux/splice.h> |
ee0b3e67 | 18 | #include "read_write.h" |
1da177e4 LT |
19 | |
20 | #include <asm/uaccess.h> | |
21 | #include <asm/unistd.h> | |
22 | ||
4b6f5d20 | 23 | const struct file_operations generic_ro_fops = { |
1da177e4 | 24 | .llseek = generic_file_llseek, |
543ade1f BP |
25 | .read = do_sync_read, |
26 | .aio_read = generic_file_aio_read, | |
1da177e4 | 27 | .mmap = generic_file_readonly_mmap, |
534f2aaa | 28 | .splice_read = generic_file_splice_read, |
1da177e4 LT |
29 | }; |
30 | ||
31 | EXPORT_SYMBOL(generic_ro_fops); | |
32 | ||
cccb5a1e | 33 | static inline int unsigned_offsets(struct file *file) |
4a3956c7 | 34 | { |
cccb5a1e | 35 | return file->f_mode & FMODE_UNSIGNED_OFFSET; |
4a3956c7 KH |
36 | } |
37 | ||
3a8cff4f CH |
38 | /** |
39 | * generic_file_llseek_unlocked - lockless generic llseek implementation | |
40 | * @file: file structure to seek on | |
41 | * @offset: file offset to seek to | |
42 | * @origin: type of seek | |
43 | * | |
44 | * Updates the file offset to the value specified by @offset and @origin. | |
45 | * Locking must be provided by the caller. | |
46 | */ | |
9465efc9 AK |
47 | loff_t |
48 | generic_file_llseek_unlocked(struct file *file, loff_t offset, int origin) | |
1da177e4 | 49 | { |
1da177e4 LT |
50 | struct inode *inode = file->f_mapping->host; |
51 | ||
1da177e4 | 52 | switch (origin) { |
3a8cff4f CH |
53 | case SEEK_END: |
54 | offset += inode->i_size; | |
55 | break; | |
56 | case SEEK_CUR: | |
5b6f1eb9 AK |
57 | /* |
58 | * Here we special-case the lseek(fd, 0, SEEK_CUR) | |
59 | * position-querying operation. Avoid rewriting the "same" | |
60 | * f_pos value back to the file because a concurrent read(), | |
61 | * write() or lseek() might have altered it | |
62 | */ | |
63 | if (offset == 0) | |
64 | return file->f_pos; | |
3a8cff4f CH |
65 | offset += file->f_pos; |
66 | break; | |
1da177e4 | 67 | } |
3a8cff4f | 68 | |
cccb5a1e | 69 | if (offset < 0 && !unsigned_offsets(file)) |
4a3956c7 KH |
70 | return -EINVAL; |
71 | if (offset > inode->i_sb->s_maxbytes) | |
3a8cff4f CH |
72 | return -EINVAL; |
73 | ||
74 | /* Special lock needed here? */ | |
75 | if (offset != file->f_pos) { | |
76 | file->f_pos = offset; | |
77 | file->f_version = 0; | |
1da177e4 | 78 | } |
3a8cff4f CH |
79 | |
80 | return offset; | |
1da177e4 | 81 | } |
9465efc9 | 82 | EXPORT_SYMBOL(generic_file_llseek_unlocked); |
1da177e4 | 83 | |
3a8cff4f CH |
84 | /** |
85 | * generic_file_llseek - generic llseek implementation for regular files | |
86 | * @file: file structure to seek on | |
87 | * @offset: file offset to seek to | |
88 | * @origin: type of seek | |
89 | * | |
90 | * This is a generic implemenation of ->llseek useable for all normal local | |
91 | * filesystems. It just updates the file offset to the value specified by | |
92 | * @offset and @origin under i_mutex. | |
93 | */ | |
9465efc9 | 94 | loff_t generic_file_llseek(struct file *file, loff_t offset, int origin) |
1da177e4 | 95 | { |
3a8cff4f CH |
96 | loff_t rval; |
97 | ||
9465efc9 | 98 | mutex_lock(&file->f_dentry->d_inode->i_mutex); |
3a8cff4f | 99 | rval = generic_file_llseek_unlocked(file, offset, origin); |
9465efc9 | 100 | mutex_unlock(&file->f_dentry->d_inode->i_mutex); |
3a8cff4f CH |
101 | |
102 | return rval; | |
1da177e4 | 103 | } |
9465efc9 | 104 | EXPORT_SYMBOL(generic_file_llseek); |
1da177e4 | 105 | |
ae6afc3f B |
106 | /** |
107 | * noop_llseek - No Operation Performed llseek implementation | |
108 | * @file: file structure to seek on | |
109 | * @offset: file offset to seek to | |
110 | * @origin: type of seek | |
111 | * | |
112 | * This is an implementation of ->llseek useable for the rare special case when | |
113 | * userspace expects the seek to succeed but the (device) file is actually not | |
114 | * able to perform the seek. In this case you use noop_llseek() instead of | |
115 | * falling back to the default implementation of ->llseek. | |
116 | */ | |
117 | loff_t noop_llseek(struct file *file, loff_t offset, int origin) | |
118 | { | |
119 | return file->f_pos; | |
120 | } | |
121 | EXPORT_SYMBOL(noop_llseek); | |
122 | ||
1da177e4 LT |
123 | loff_t no_llseek(struct file *file, loff_t offset, int origin) |
124 | { | |
125 | return -ESPIPE; | |
126 | } | |
127 | EXPORT_SYMBOL(no_llseek); | |
128 | ||
129 | loff_t default_llseek(struct file *file, loff_t offset, int origin) | |
130 | { | |
16abef0e | 131 | loff_t retval; |
1da177e4 | 132 | |
ab91261f | 133 | mutex_lock(&file->f_dentry->d_inode->i_mutex); |
1da177e4 | 134 | switch (origin) { |
7b8e8924 | 135 | case SEEK_END: |
0f7fc9e4 | 136 | offset += i_size_read(file->f_path.dentry->d_inode); |
1da177e4 | 137 | break; |
7b8e8924 | 138 | case SEEK_CUR: |
5b6f1eb9 AK |
139 | if (offset == 0) { |
140 | retval = file->f_pos; | |
141 | goto out; | |
142 | } | |
1da177e4 LT |
143 | offset += file->f_pos; |
144 | } | |
145 | retval = -EINVAL; | |
cccb5a1e | 146 | if (offset >= 0 || unsigned_offsets(file)) { |
1da177e4 LT |
147 | if (offset != file->f_pos) { |
148 | file->f_pos = offset; | |
149 | file->f_version = 0; | |
150 | } | |
151 | retval = offset; | |
152 | } | |
5b6f1eb9 | 153 | out: |
ab91261f | 154 | mutex_unlock(&file->f_dentry->d_inode->i_mutex); |
1da177e4 LT |
155 | return retval; |
156 | } | |
157 | EXPORT_SYMBOL(default_llseek); | |
158 | ||
159 | loff_t vfs_llseek(struct file *file, loff_t offset, int origin) | |
160 | { | |
161 | loff_t (*fn)(struct file *, loff_t, int); | |
162 | ||
163 | fn = no_llseek; | |
164 | if (file->f_mode & FMODE_LSEEK) { | |
1da177e4 LT |
165 | if (file->f_op && file->f_op->llseek) |
166 | fn = file->f_op->llseek; | |
167 | } | |
168 | return fn(file, offset, origin); | |
169 | } | |
170 | EXPORT_SYMBOL(vfs_llseek); | |
171 | ||
003d7ab4 | 172 | SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, origin) |
1da177e4 LT |
173 | { |
174 | off_t retval; | |
175 | struct file * file; | |
176 | int fput_needed; | |
177 | ||
178 | retval = -EBADF; | |
179 | file = fget_light(fd, &fput_needed); | |
180 | if (!file) | |
181 | goto bad; | |
182 | ||
183 | retval = -EINVAL; | |
1ae7075b | 184 | if (origin <= SEEK_MAX) { |
1da177e4 LT |
185 | loff_t res = vfs_llseek(file, offset, origin); |
186 | retval = res; | |
187 | if (res != (loff_t)retval) | |
188 | retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */ | |
189 | } | |
190 | fput_light(file, fput_needed); | |
191 | bad: | |
192 | return retval; | |
193 | } | |
194 | ||
195 | #ifdef __ARCH_WANT_SYS_LLSEEK | |
003d7ab4 HC |
196 | SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high, |
197 | unsigned long, offset_low, loff_t __user *, result, | |
198 | unsigned int, origin) | |
1da177e4 LT |
199 | { |
200 | int retval; | |
201 | struct file * file; | |
202 | loff_t offset; | |
203 | int fput_needed; | |
204 | ||
205 | retval = -EBADF; | |
206 | file = fget_light(fd, &fput_needed); | |
207 | if (!file) | |
208 | goto bad; | |
209 | ||
210 | retval = -EINVAL; | |
1ae7075b | 211 | if (origin > SEEK_MAX) |
1da177e4 LT |
212 | goto out_putf; |
213 | ||
214 | offset = vfs_llseek(file, ((loff_t) offset_high << 32) | offset_low, | |
215 | origin); | |
216 | ||
217 | retval = (int)offset; | |
218 | if (offset >= 0) { | |
219 | retval = -EFAULT; | |
220 | if (!copy_to_user(result, &offset, sizeof(offset))) | |
221 | retval = 0; | |
222 | } | |
223 | out_putf: | |
224 | fput_light(file, fput_needed); | |
225 | bad: | |
226 | return retval; | |
227 | } | |
228 | #endif | |
229 | ||
4a3956c7 | 230 | |
e28cc715 LT |
231 | /* |
232 | * rw_verify_area doesn't like huge counts. We limit | |
233 | * them to something that fits in "int" so that others | |
234 | * won't have to do range checks all the time. | |
235 | */ | |
1da177e4 LT |
236 | int rw_verify_area(int read_write, struct file *file, loff_t *ppos, size_t count) |
237 | { | |
238 | struct inode *inode; | |
239 | loff_t pos; | |
c43e259c | 240 | int retval = -EINVAL; |
1da177e4 | 241 | |
163da958 | 242 | inode = file->f_path.dentry->d_inode; |
e28cc715 | 243 | if (unlikely((ssize_t) count < 0)) |
c43e259c | 244 | return retval; |
1da177e4 | 245 | pos = *ppos; |
cccb5a1e AV |
246 | if (unlikely(pos < 0)) { |
247 | if (!unsigned_offsets(file)) | |
248 | return retval; | |
249 | if (count >= -pos) /* both values are in 0..LLONG_MAX */ | |
250 | return -EOVERFLOW; | |
251 | } else if (unlikely((loff_t) (pos + count) < 0)) { | |
252 | if (!unsigned_offsets(file)) | |
4a3956c7 KH |
253 | return retval; |
254 | } | |
1da177e4 | 255 | |
a16877ca | 256 | if (unlikely(inode->i_flock && mandatory_lock(inode))) { |
c43e259c | 257 | retval = locks_mandatory_area( |
e28cc715 LT |
258 | read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE, |
259 | inode, file, pos, count); | |
260 | if (retval < 0) | |
261 | return retval; | |
262 | } | |
c43e259c JM |
263 | retval = security_file_permission(file, |
264 | read_write == READ ? MAY_READ : MAY_WRITE); | |
265 | if (retval) | |
266 | return retval; | |
e28cc715 | 267 | return count > MAX_RW_COUNT ? MAX_RW_COUNT : count; |
1da177e4 LT |
268 | } |
269 | ||
63e68809 BL |
270 | static void wait_on_retry_sync_kiocb(struct kiocb *iocb) |
271 | { | |
272 | set_current_state(TASK_UNINTERRUPTIBLE); | |
273 | if (!kiocbIsKicked(iocb)) | |
274 | schedule(); | |
275 | else | |
276 | kiocbClearKicked(iocb); | |
277 | __set_current_state(TASK_RUNNING); | |
278 | } | |
279 | ||
1da177e4 LT |
280 | ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos) |
281 | { | |
027445c3 | 282 | struct iovec iov = { .iov_base = buf, .iov_len = len }; |
1da177e4 LT |
283 | struct kiocb kiocb; |
284 | ssize_t ret; | |
285 | ||
286 | init_sync_kiocb(&kiocb, filp); | |
287 | kiocb.ki_pos = *ppos; | |
027445c3 | 288 | kiocb.ki_left = len; |
61964eba | 289 | kiocb.ki_nbytes = len; |
027445c3 BP |
290 | |
291 | for (;;) { | |
292 | ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos); | |
293 | if (ret != -EIOCBRETRY) | |
294 | break; | |
63e68809 | 295 | wait_on_retry_sync_kiocb(&kiocb); |
027445c3 | 296 | } |
63e68809 | 297 | |
1da177e4 LT |
298 | if (-EIOCBQUEUED == ret) |
299 | ret = wait_on_sync_kiocb(&kiocb); | |
300 | *ppos = kiocb.ki_pos; | |
301 | return ret; | |
302 | } | |
303 | ||
304 | EXPORT_SYMBOL(do_sync_read); | |
305 | ||
306 | ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos) | |
307 | { | |
308 | ssize_t ret; | |
309 | ||
310 | if (!(file->f_mode & FMODE_READ)) | |
311 | return -EBADF; | |
312 | if (!file->f_op || (!file->f_op->read && !file->f_op->aio_read)) | |
313 | return -EINVAL; | |
314 | if (unlikely(!access_ok(VERIFY_WRITE, buf, count))) | |
315 | return -EFAULT; | |
316 | ||
317 | ret = rw_verify_area(READ, file, pos, count); | |
e28cc715 LT |
318 | if (ret >= 0) { |
319 | count = ret; | |
c43e259c JM |
320 | if (file->f_op->read) |
321 | ret = file->f_op->read(file, buf, count, pos); | |
322 | else | |
323 | ret = do_sync_read(file, buf, count, pos); | |
324 | if (ret > 0) { | |
2a12a9d7 | 325 | fsnotify_access(file); |
c43e259c | 326 | add_rchar(current, ret); |
1da177e4 | 327 | } |
c43e259c | 328 | inc_syscr(current); |
1da177e4 LT |
329 | } |
330 | ||
331 | return ret; | |
332 | } | |
333 | ||
334 | EXPORT_SYMBOL(vfs_read); | |
335 | ||
336 | ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos) | |
337 | { | |
027445c3 | 338 | struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len }; |
1da177e4 LT |
339 | struct kiocb kiocb; |
340 | ssize_t ret; | |
341 | ||
342 | init_sync_kiocb(&kiocb, filp); | |
343 | kiocb.ki_pos = *ppos; | |
027445c3 | 344 | kiocb.ki_left = len; |
61964eba | 345 | kiocb.ki_nbytes = len; |
027445c3 BP |
346 | |
347 | for (;;) { | |
348 | ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos); | |
349 | if (ret != -EIOCBRETRY) | |
350 | break; | |
63e68809 | 351 | wait_on_retry_sync_kiocb(&kiocb); |
027445c3 | 352 | } |
63e68809 | 353 | |
1da177e4 LT |
354 | if (-EIOCBQUEUED == ret) |
355 | ret = wait_on_sync_kiocb(&kiocb); | |
356 | *ppos = kiocb.ki_pos; | |
357 | return ret; | |
358 | } | |
359 | ||
360 | EXPORT_SYMBOL(do_sync_write); | |
361 | ||
362 | ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos) | |
363 | { | |
364 | ssize_t ret; | |
365 | ||
366 | if (!(file->f_mode & FMODE_WRITE)) | |
367 | return -EBADF; | |
368 | if (!file->f_op || (!file->f_op->write && !file->f_op->aio_write)) | |
369 | return -EINVAL; | |
370 | if (unlikely(!access_ok(VERIFY_READ, buf, count))) | |
371 | return -EFAULT; | |
372 | ||
373 | ret = rw_verify_area(WRITE, file, pos, count); | |
e28cc715 LT |
374 | if (ret >= 0) { |
375 | count = ret; | |
c43e259c JM |
376 | if (file->f_op->write) |
377 | ret = file->f_op->write(file, buf, count, pos); | |
378 | else | |
379 | ret = do_sync_write(file, buf, count, pos); | |
380 | if (ret > 0) { | |
2a12a9d7 | 381 | fsnotify_modify(file); |
c43e259c | 382 | add_wchar(current, ret); |
1da177e4 | 383 | } |
c43e259c | 384 | inc_syscw(current); |
1da177e4 LT |
385 | } |
386 | ||
387 | return ret; | |
388 | } | |
389 | ||
390 | EXPORT_SYMBOL(vfs_write); | |
391 | ||
392 | static inline loff_t file_pos_read(struct file *file) | |
393 | { | |
394 | return file->f_pos; | |
395 | } | |
396 | ||
397 | static inline void file_pos_write(struct file *file, loff_t pos) | |
398 | { | |
399 | file->f_pos = pos; | |
400 | } | |
401 | ||
3cdad428 | 402 | SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count) |
1da177e4 LT |
403 | { |
404 | struct file *file; | |
405 | ssize_t ret = -EBADF; | |
406 | int fput_needed; | |
407 | ||
408 | file = fget_light(fd, &fput_needed); | |
409 | if (file) { | |
410 | loff_t pos = file_pos_read(file); | |
411 | ret = vfs_read(file, buf, count, &pos); | |
412 | file_pos_write(file, pos); | |
413 | fput_light(file, fput_needed); | |
414 | } | |
415 | ||
416 | return ret; | |
417 | } | |
1da177e4 | 418 | |
3cdad428 HC |
419 | SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf, |
420 | size_t, count) | |
1da177e4 LT |
421 | { |
422 | struct file *file; | |
423 | ssize_t ret = -EBADF; | |
424 | int fput_needed; | |
425 | ||
426 | file = fget_light(fd, &fput_needed); | |
427 | if (file) { | |
428 | loff_t pos = file_pos_read(file); | |
429 | ret = vfs_write(file, buf, count, &pos); | |
430 | file_pos_write(file, pos); | |
431 | fput_light(file, fput_needed); | |
432 | } | |
433 | ||
434 | return ret; | |
435 | } | |
436 | ||
6673e0c3 HC |
437 | SYSCALL_DEFINE(pread64)(unsigned int fd, char __user *buf, |
438 | size_t count, loff_t pos) | |
1da177e4 LT |
439 | { |
440 | struct file *file; | |
441 | ssize_t ret = -EBADF; | |
442 | int fput_needed; | |
443 | ||
444 | if (pos < 0) | |
445 | return -EINVAL; | |
446 | ||
447 | file = fget_light(fd, &fput_needed); | |
448 | if (file) { | |
449 | ret = -ESPIPE; | |
450 | if (file->f_mode & FMODE_PREAD) | |
451 | ret = vfs_read(file, buf, count, &pos); | |
452 | fput_light(file, fput_needed); | |
453 | } | |
454 | ||
455 | return ret; | |
456 | } | |
6673e0c3 HC |
457 | #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS |
458 | asmlinkage long SyS_pread64(long fd, long buf, long count, loff_t pos) | |
459 | { | |
460 | return SYSC_pread64((unsigned int) fd, (char __user *) buf, | |
461 | (size_t) count, pos); | |
462 | } | |
463 | SYSCALL_ALIAS(sys_pread64, SyS_pread64); | |
464 | #endif | |
1da177e4 | 465 | |
6673e0c3 HC |
466 | SYSCALL_DEFINE(pwrite64)(unsigned int fd, const char __user *buf, |
467 | size_t count, loff_t pos) | |
1da177e4 LT |
468 | { |
469 | struct file *file; | |
470 | ssize_t ret = -EBADF; | |
471 | int fput_needed; | |
472 | ||
473 | if (pos < 0) | |
474 | return -EINVAL; | |
475 | ||
476 | file = fget_light(fd, &fput_needed); | |
477 | if (file) { | |
478 | ret = -ESPIPE; | |
479 | if (file->f_mode & FMODE_PWRITE) | |
480 | ret = vfs_write(file, buf, count, &pos); | |
481 | fput_light(file, fput_needed); | |
482 | } | |
483 | ||
484 | return ret; | |
485 | } | |
6673e0c3 HC |
486 | #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS |
487 | asmlinkage long SyS_pwrite64(long fd, long buf, long count, loff_t pos) | |
488 | { | |
489 | return SYSC_pwrite64((unsigned int) fd, (const char __user *) buf, | |
490 | (size_t) count, pos); | |
491 | } | |
492 | SYSCALL_ALIAS(sys_pwrite64, SyS_pwrite64); | |
493 | #endif | |
1da177e4 LT |
494 | |
495 | /* | |
496 | * Reduce an iovec's length in-place. Return the resulting number of segments | |
497 | */ | |
498 | unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to) | |
499 | { | |
500 | unsigned long seg = 0; | |
501 | size_t len = 0; | |
502 | ||
503 | while (seg < nr_segs) { | |
504 | seg++; | |
505 | if (len + iov->iov_len >= to) { | |
506 | iov->iov_len = to - len; | |
507 | break; | |
508 | } | |
509 | len += iov->iov_len; | |
510 | iov++; | |
511 | } | |
512 | return seg; | |
513 | } | |
19295529 | 514 | EXPORT_SYMBOL(iov_shorten); |
1da177e4 | 515 | |
ee0b3e67 BP |
516 | ssize_t do_sync_readv_writev(struct file *filp, const struct iovec *iov, |
517 | unsigned long nr_segs, size_t len, loff_t *ppos, iov_fn_t fn) | |
518 | { | |
519 | struct kiocb kiocb; | |
520 | ssize_t ret; | |
521 | ||
522 | init_sync_kiocb(&kiocb, filp); | |
523 | kiocb.ki_pos = *ppos; | |
524 | kiocb.ki_left = len; | |
525 | kiocb.ki_nbytes = len; | |
526 | ||
527 | for (;;) { | |
528 | ret = fn(&kiocb, iov, nr_segs, kiocb.ki_pos); | |
529 | if (ret != -EIOCBRETRY) | |
530 | break; | |
531 | wait_on_retry_sync_kiocb(&kiocb); | |
532 | } | |
533 | ||
534 | if (ret == -EIOCBQUEUED) | |
535 | ret = wait_on_sync_kiocb(&kiocb); | |
536 | *ppos = kiocb.ki_pos; | |
537 | return ret; | |
538 | } | |
539 | ||
540 | /* Do it by hand, with file-ops */ | |
541 | ssize_t do_loop_readv_writev(struct file *filp, struct iovec *iov, | |
542 | unsigned long nr_segs, loff_t *ppos, io_fn_t fn) | |
543 | { | |
544 | struct iovec *vector = iov; | |
545 | ssize_t ret = 0; | |
546 | ||
547 | while (nr_segs > 0) { | |
548 | void __user *base; | |
549 | size_t len; | |
550 | ssize_t nr; | |
551 | ||
552 | base = vector->iov_base; | |
553 | len = vector->iov_len; | |
554 | vector++; | |
555 | nr_segs--; | |
556 | ||
557 | nr = fn(filp, base, len, ppos); | |
558 | ||
559 | if (nr < 0) { | |
560 | if (!ret) | |
561 | ret = nr; | |
562 | break; | |
563 | } | |
564 | ret += nr; | |
565 | if (nr != len) | |
566 | break; | |
567 | } | |
568 | ||
569 | return ret; | |
570 | } | |
571 | ||
1da177e4 LT |
572 | /* A write operation does a read from user space and vice versa */ |
573 | #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ) | |
574 | ||
eed4e51f BP |
575 | ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector, |
576 | unsigned long nr_segs, unsigned long fast_segs, | |
577 | struct iovec *fast_pointer, | |
578 | struct iovec **ret_pointer) | |
435f49a5 | 579 | { |
eed4e51f | 580 | unsigned long seg; |
435f49a5 | 581 | ssize_t ret; |
eed4e51f BP |
582 | struct iovec *iov = fast_pointer; |
583 | ||
435f49a5 LT |
584 | /* |
585 | * SuS says "The readv() function *may* fail if the iovcnt argument | |
586 | * was less than or equal to 0, or greater than {IOV_MAX}. Linux has | |
587 | * traditionally returned zero for zero segments, so... | |
588 | */ | |
eed4e51f BP |
589 | if (nr_segs == 0) { |
590 | ret = 0; | |
435f49a5 | 591 | goto out; |
eed4e51f BP |
592 | } |
593 | ||
435f49a5 LT |
594 | /* |
595 | * First get the "struct iovec" from user memory and | |
596 | * verify all the pointers | |
597 | */ | |
eed4e51f BP |
598 | if (nr_segs > UIO_MAXIOV) { |
599 | ret = -EINVAL; | |
435f49a5 | 600 | goto out; |
eed4e51f BP |
601 | } |
602 | if (nr_segs > fast_segs) { | |
435f49a5 | 603 | iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL); |
eed4e51f BP |
604 | if (iov == NULL) { |
605 | ret = -ENOMEM; | |
435f49a5 | 606 | goto out; |
eed4e51f | 607 | } |
435f49a5 | 608 | } |
eed4e51f BP |
609 | if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) { |
610 | ret = -EFAULT; | |
435f49a5 | 611 | goto out; |
eed4e51f BP |
612 | } |
613 | ||
435f49a5 | 614 | /* |
eed4e51f BP |
615 | * According to the Single Unix Specification we should return EINVAL |
616 | * if an element length is < 0 when cast to ssize_t or if the | |
617 | * total length would overflow the ssize_t return value of the | |
618 | * system call. | |
435f49a5 LT |
619 | * |
620 | * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the | |
621 | * overflow case. | |
622 | */ | |
eed4e51f | 623 | ret = 0; |
435f49a5 LT |
624 | for (seg = 0; seg < nr_segs; seg++) { |
625 | void __user *buf = iov[seg].iov_base; | |
626 | ssize_t len = (ssize_t)iov[seg].iov_len; | |
eed4e51f BP |
627 | |
628 | /* see if we we're about to use an invalid len or if | |
629 | * it's about to overflow ssize_t */ | |
435f49a5 | 630 | if (len < 0) { |
eed4e51f | 631 | ret = -EINVAL; |
435f49a5 | 632 | goto out; |
eed4e51f BP |
633 | } |
634 | if (unlikely(!access_ok(vrfy_dir(type), buf, len))) { | |
635 | ret = -EFAULT; | |
435f49a5 LT |
636 | goto out; |
637 | } | |
638 | if (len > MAX_RW_COUNT - ret) { | |
639 | len = MAX_RW_COUNT - ret; | |
640 | iov[seg].iov_len = len; | |
eed4e51f | 641 | } |
eed4e51f | 642 | ret += len; |
435f49a5 | 643 | } |
eed4e51f BP |
644 | out: |
645 | *ret_pointer = iov; | |
646 | return ret; | |
647 | } | |
648 | ||
1da177e4 LT |
649 | static ssize_t do_readv_writev(int type, struct file *file, |
650 | const struct iovec __user * uvector, | |
651 | unsigned long nr_segs, loff_t *pos) | |
652 | { | |
1da177e4 LT |
653 | size_t tot_len; |
654 | struct iovec iovstack[UIO_FASTIOV]; | |
ee0b3e67 | 655 | struct iovec *iov = iovstack; |
1da177e4 | 656 | ssize_t ret; |
1da177e4 LT |
657 | io_fn_t fn; |
658 | iov_fn_t fnv; | |
659 | ||
eed4e51f BP |
660 | if (!file->f_op) { |
661 | ret = -EINVAL; | |
1da177e4 | 662 | goto out; |
1da177e4 | 663 | } |
1da177e4 | 664 | |
eed4e51f BP |
665 | ret = rw_copy_check_uvector(type, uvector, nr_segs, |
666 | ARRAY_SIZE(iovstack), iovstack, &iov); | |
667 | if (ret <= 0) | |
1da177e4 | 668 | goto out; |
1da177e4 | 669 | |
eed4e51f | 670 | tot_len = ret; |
1da177e4 | 671 | ret = rw_verify_area(type, file, pos, tot_len); |
e28cc715 | 672 | if (ret < 0) |
411b67b4 | 673 | goto out; |
1da177e4 LT |
674 | |
675 | fnv = NULL; | |
676 | if (type == READ) { | |
677 | fn = file->f_op->read; | |
ee0b3e67 | 678 | fnv = file->f_op->aio_read; |
1da177e4 LT |
679 | } else { |
680 | fn = (io_fn_t)file->f_op->write; | |
ee0b3e67 | 681 | fnv = file->f_op->aio_write; |
1da177e4 LT |
682 | } |
683 | ||
ee0b3e67 BP |
684 | if (fnv) |
685 | ret = do_sync_readv_writev(file, iov, nr_segs, tot_len, | |
686 | pos, fnv); | |
687 | else | |
688 | ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn); | |
1da177e4 | 689 | |
1da177e4 LT |
690 | out: |
691 | if (iov != iovstack) | |
692 | kfree(iov); | |
0eeca283 RL |
693 | if ((ret + (type == READ)) > 0) { |
694 | if (type == READ) | |
2a12a9d7 | 695 | fsnotify_access(file); |
0eeca283 | 696 | else |
2a12a9d7 | 697 | fsnotify_modify(file); |
0eeca283 | 698 | } |
1da177e4 | 699 | return ret; |
1da177e4 LT |
700 | } |
701 | ||
702 | ssize_t vfs_readv(struct file *file, const struct iovec __user *vec, | |
703 | unsigned long vlen, loff_t *pos) | |
704 | { | |
705 | if (!(file->f_mode & FMODE_READ)) | |
706 | return -EBADF; | |
ee0b3e67 | 707 | if (!file->f_op || (!file->f_op->aio_read && !file->f_op->read)) |
1da177e4 LT |
708 | return -EINVAL; |
709 | ||
710 | return do_readv_writev(READ, file, vec, vlen, pos); | |
711 | } | |
712 | ||
713 | EXPORT_SYMBOL(vfs_readv); | |
714 | ||
715 | ssize_t vfs_writev(struct file *file, const struct iovec __user *vec, | |
716 | unsigned long vlen, loff_t *pos) | |
717 | { | |
718 | if (!(file->f_mode & FMODE_WRITE)) | |
719 | return -EBADF; | |
ee0b3e67 | 720 | if (!file->f_op || (!file->f_op->aio_write && !file->f_op->write)) |
1da177e4 LT |
721 | return -EINVAL; |
722 | ||
723 | return do_readv_writev(WRITE, file, vec, vlen, pos); | |
724 | } | |
725 | ||
726 | EXPORT_SYMBOL(vfs_writev); | |
727 | ||
3cdad428 HC |
728 | SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec, |
729 | unsigned long, vlen) | |
1da177e4 LT |
730 | { |
731 | struct file *file; | |
732 | ssize_t ret = -EBADF; | |
733 | int fput_needed; | |
734 | ||
735 | file = fget_light(fd, &fput_needed); | |
736 | if (file) { | |
737 | loff_t pos = file_pos_read(file); | |
738 | ret = vfs_readv(file, vec, vlen, &pos); | |
739 | file_pos_write(file, pos); | |
740 | fput_light(file, fput_needed); | |
741 | } | |
742 | ||
743 | if (ret > 0) | |
4b98d11b AD |
744 | add_rchar(current, ret); |
745 | inc_syscr(current); | |
1da177e4 LT |
746 | return ret; |
747 | } | |
748 | ||
3cdad428 HC |
749 | SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec, |
750 | unsigned long, vlen) | |
1da177e4 LT |
751 | { |
752 | struct file *file; | |
753 | ssize_t ret = -EBADF; | |
754 | int fput_needed; | |
755 | ||
756 | file = fget_light(fd, &fput_needed); | |
757 | if (file) { | |
758 | loff_t pos = file_pos_read(file); | |
759 | ret = vfs_writev(file, vec, vlen, &pos); | |
760 | file_pos_write(file, pos); | |
761 | fput_light(file, fput_needed); | |
762 | } | |
763 | ||
764 | if (ret > 0) | |
4b98d11b AD |
765 | add_wchar(current, ret); |
766 | inc_syscw(current); | |
1da177e4 LT |
767 | return ret; |
768 | } | |
769 | ||
601cc11d LT |
770 | static inline loff_t pos_from_hilo(unsigned long high, unsigned long low) |
771 | { | |
772 | #define HALF_LONG_BITS (BITS_PER_LONG / 2) | |
773 | return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low; | |
774 | } | |
775 | ||
f3554f4b | 776 | SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec, |
601cc11d | 777 | unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h) |
f3554f4b | 778 | { |
601cc11d | 779 | loff_t pos = pos_from_hilo(pos_h, pos_l); |
f3554f4b GH |
780 | struct file *file; |
781 | ssize_t ret = -EBADF; | |
782 | int fput_needed; | |
783 | ||
784 | if (pos < 0) | |
785 | return -EINVAL; | |
786 | ||
787 | file = fget_light(fd, &fput_needed); | |
788 | if (file) { | |
789 | ret = -ESPIPE; | |
790 | if (file->f_mode & FMODE_PREAD) | |
791 | ret = vfs_readv(file, vec, vlen, &pos); | |
792 | fput_light(file, fput_needed); | |
793 | } | |
794 | ||
795 | if (ret > 0) | |
796 | add_rchar(current, ret); | |
797 | inc_syscr(current); | |
798 | return ret; | |
799 | } | |
800 | ||
801 | SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec, | |
601cc11d | 802 | unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h) |
f3554f4b | 803 | { |
601cc11d | 804 | loff_t pos = pos_from_hilo(pos_h, pos_l); |
f3554f4b GH |
805 | struct file *file; |
806 | ssize_t ret = -EBADF; | |
807 | int fput_needed; | |
808 | ||
809 | if (pos < 0) | |
810 | return -EINVAL; | |
811 | ||
812 | file = fget_light(fd, &fput_needed); | |
813 | if (file) { | |
814 | ret = -ESPIPE; | |
815 | if (file->f_mode & FMODE_PWRITE) | |
816 | ret = vfs_writev(file, vec, vlen, &pos); | |
817 | fput_light(file, fput_needed); | |
818 | } | |
819 | ||
820 | if (ret > 0) | |
821 | add_wchar(current, ret); | |
822 | inc_syscw(current); | |
823 | return ret; | |
824 | } | |
825 | ||
1da177e4 LT |
826 | static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos, |
827 | size_t count, loff_t max) | |
828 | { | |
829 | struct file * in_file, * out_file; | |
830 | struct inode * in_inode, * out_inode; | |
831 | loff_t pos; | |
832 | ssize_t retval; | |
534f2aaa | 833 | int fput_needed_in, fput_needed_out, fl; |
1da177e4 LT |
834 | |
835 | /* | |
836 | * Get input file, and verify that it is ok.. | |
837 | */ | |
838 | retval = -EBADF; | |
839 | in_file = fget_light(in_fd, &fput_needed_in); | |
840 | if (!in_file) | |
841 | goto out; | |
842 | if (!(in_file->f_mode & FMODE_READ)) | |
843 | goto fput_in; | |
1da177e4 LT |
844 | retval = -ESPIPE; |
845 | if (!ppos) | |
846 | ppos = &in_file->f_pos; | |
847 | else | |
848 | if (!(in_file->f_mode & FMODE_PREAD)) | |
849 | goto fput_in; | |
850 | retval = rw_verify_area(READ, in_file, ppos, count); | |
e28cc715 | 851 | if (retval < 0) |
1da177e4 | 852 | goto fput_in; |
e28cc715 | 853 | count = retval; |
1da177e4 | 854 | |
1da177e4 LT |
855 | /* |
856 | * Get output file, and verify that it is ok.. | |
857 | */ | |
858 | retval = -EBADF; | |
859 | out_file = fget_light(out_fd, &fput_needed_out); | |
860 | if (!out_file) | |
861 | goto fput_in; | |
862 | if (!(out_file->f_mode & FMODE_WRITE)) | |
863 | goto fput_out; | |
864 | retval = -EINVAL; | |
6818173b | 865 | in_inode = in_file->f_path.dentry->d_inode; |
0f7fc9e4 | 866 | out_inode = out_file->f_path.dentry->d_inode; |
1da177e4 | 867 | retval = rw_verify_area(WRITE, out_file, &out_file->f_pos, count); |
e28cc715 | 868 | if (retval < 0) |
1da177e4 | 869 | goto fput_out; |
e28cc715 | 870 | count = retval; |
1da177e4 | 871 | |
1da177e4 LT |
872 | if (!max) |
873 | max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes); | |
874 | ||
875 | pos = *ppos; | |
1da177e4 LT |
876 | if (unlikely(pos + count > max)) { |
877 | retval = -EOVERFLOW; | |
878 | if (pos >= max) | |
879 | goto fput_out; | |
880 | count = max - pos; | |
881 | } | |
882 | ||
d96e6e71 | 883 | fl = 0; |
534f2aaa | 884 | #if 0 |
d96e6e71 JA |
885 | /* |
886 | * We need to debate whether we can enable this or not. The | |
887 | * man page documents EAGAIN return for the output at least, | |
888 | * and the application is arguably buggy if it doesn't expect | |
889 | * EAGAIN on a non-blocking file descriptor. | |
890 | */ | |
891 | if (in_file->f_flags & O_NONBLOCK) | |
892 | fl = SPLICE_F_NONBLOCK; | |
534f2aaa | 893 | #endif |
d96e6e71 | 894 | retval = do_splice_direct(in_file, ppos, out_file, count, fl); |
1da177e4 LT |
895 | |
896 | if (retval > 0) { | |
4b98d11b AD |
897 | add_rchar(current, retval); |
898 | add_wchar(current, retval); | |
1da177e4 | 899 | } |
1da177e4 | 900 | |
4b98d11b AD |
901 | inc_syscr(current); |
902 | inc_syscw(current); | |
1da177e4 LT |
903 | if (*ppos > max) |
904 | retval = -EOVERFLOW; | |
905 | ||
906 | fput_out: | |
907 | fput_light(out_file, fput_needed_out); | |
908 | fput_in: | |
909 | fput_light(in_file, fput_needed_in); | |
910 | out: | |
911 | return retval; | |
912 | } | |
913 | ||
002c8976 | 914 | SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count) |
1da177e4 LT |
915 | { |
916 | loff_t pos; | |
917 | off_t off; | |
918 | ssize_t ret; | |
919 | ||
920 | if (offset) { | |
921 | if (unlikely(get_user(off, offset))) | |
922 | return -EFAULT; | |
923 | pos = off; | |
924 | ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS); | |
925 | if (unlikely(put_user(pos, offset))) | |
926 | return -EFAULT; | |
927 | return ret; | |
928 | } | |
929 | ||
930 | return do_sendfile(out_fd, in_fd, NULL, count, 0); | |
931 | } | |
932 | ||
002c8976 | 933 | SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count) |
1da177e4 LT |
934 | { |
935 | loff_t pos; | |
936 | ssize_t ret; | |
937 | ||
938 | if (offset) { | |
939 | if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t)))) | |
940 | return -EFAULT; | |
941 | ret = do_sendfile(out_fd, in_fd, &pos, count, 0); | |
942 | if (unlikely(put_user(pos, offset))) | |
943 | return -EFAULT; | |
944 | return ret; | |
945 | } | |
946 | ||
947 | return do_sendfile(out_fd, in_fd, NULL, count, 0); | |
948 | } |