]>
Commit | Line | Data |
---|---|---|
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> | |
12 | #include <linux/fsnotify.h> | |
13 | #include <linux/security.h> | |
14 | #include <linux/export.h> | |
15 | #include <linux/syscalls.h> | |
16 | #include <linux/pagemap.h> | |
17 | #include <linux/splice.h> | |
18 | #include <linux/compat.h> | |
19 | #include "internal.h" | |
20 | ||
21 | #include <asm/uaccess.h> | |
22 | #include <asm/unistd.h> | |
23 | ||
24 | typedef ssize_t (*io_fn_t)(struct file *, char __user *, size_t, loff_t *); | |
25 | typedef ssize_t (*iov_fn_t)(struct kiocb *, const struct iovec *, | |
26 | unsigned long, loff_t); | |
27 | typedef ssize_t (*iter_fn_t)(struct kiocb *, struct iov_iter *); | |
28 | ||
29 | const struct file_operations generic_ro_fops = { | |
30 | .llseek = generic_file_llseek, | |
31 | .read = new_sync_read, | |
32 | .read_iter = generic_file_read_iter, | |
33 | .mmap = generic_file_readonly_mmap, | |
34 | .splice_read = generic_file_splice_read, | |
35 | }; | |
36 | ||
37 | EXPORT_SYMBOL(generic_ro_fops); | |
38 | ||
39 | static inline int unsigned_offsets(struct file *file) | |
40 | { | |
41 | return file->f_mode & FMODE_UNSIGNED_OFFSET; | |
42 | } | |
43 | ||
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) | |
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 | } | |
69 | EXPORT_SYMBOL(vfs_setpos); | |
70 | ||
71 | /** | |
72 | * generic_file_llseek_size - generic llseek implementation for regular files | |
73 | * @file: file structure to seek on | |
74 | * @offset: file offset to seek to | |
75 | * @whence: type of seek | |
76 | * @size: max size of this file in file system | |
77 | * @eof: offset used for SEEK_END position | |
78 | * | |
79 | * This is a variant of generic_file_llseek that allows passing in a custom | |
80 | * maximum file size and a custom EOF position, for e.g. hashed directories | |
81 | * | |
82 | * Synchronization: | |
83 | * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms) | |
84 | * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes. | |
85 | * read/writes behave like SEEK_SET against seeks. | |
86 | */ | |
87 | loff_t | |
88 | generic_file_llseek_size(struct file *file, loff_t offset, int whence, | |
89 | loff_t maxsize, loff_t eof) | |
90 | { | |
91 | switch (whence) { | |
92 | case SEEK_END: | |
93 | offset += eof; | |
94 | break; | |
95 | case SEEK_CUR: | |
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; | |
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); | |
110 | offset = vfs_setpos(file, file->f_pos + offset, maxsize); | |
111 | spin_unlock(&file->f_lock); | |
112 | return offset; | |
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 | */ | |
118 | if (offset >= eof) | |
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 | */ | |
126 | if (offset >= eof) | |
127 | return -ENXIO; | |
128 | offset = eof; | |
129 | break; | |
130 | } | |
131 | ||
132 | return vfs_setpos(file, offset, maxsize); | |
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 | |
140 | * @whence: type of seek | |
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 | |
144 | * @offset and @whence. | |
145 | */ | |
146 | loff_t generic_file_llseek(struct file *file, loff_t offset, int whence) | |
147 | { | |
148 | struct inode *inode = file->f_mapping->host; | |
149 | ||
150 | return generic_file_llseek_size(file, offset, whence, | |
151 | inode->i_sb->s_maxbytes, | |
152 | i_size_read(inode)); | |
153 | } | |
154 | EXPORT_SYMBOL(generic_file_llseek); | |
155 | ||
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 | ||
176 | /** | |
177 | * noop_llseek - No Operation Performed llseek implementation | |
178 | * @file: file structure to seek on | |
179 | * @offset: file offset to seek to | |
180 | * @whence: type of seek | |
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 | */ | |
187 | loff_t noop_llseek(struct file *file, loff_t offset, int whence) | |
188 | { | |
189 | return file->f_pos; | |
190 | } | |
191 | EXPORT_SYMBOL(noop_llseek); | |
192 | ||
193 | loff_t no_llseek(struct file *file, loff_t offset, int whence) | |
194 | { | |
195 | return -ESPIPE; | |
196 | } | |
197 | EXPORT_SYMBOL(no_llseek); | |
198 | ||
199 | loff_t default_llseek(struct file *file, loff_t offset, int whence) | |
200 | { | |
201 | struct inode *inode = file_inode(file); | |
202 | loff_t retval; | |
203 | ||
204 | mutex_lock(&inode->i_mutex); | |
205 | switch (whence) { | |
206 | case SEEK_END: | |
207 | offset += i_size_read(inode); | |
208 | break; | |
209 | case SEEK_CUR: | |
210 | if (offset == 0) { | |
211 | retval = file->f_pos; | |
212 | goto out; | |
213 | } | |
214 | offset += file->f_pos; | |
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 | */ | |
222 | if (offset >= inode->i_size) { | |
223 | retval = -ENXIO; | |
224 | goto out; | |
225 | } | |
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 | */ | |
233 | if (offset >= inode->i_size) { | |
234 | retval = -ENXIO; | |
235 | goto out; | |
236 | } | |
237 | offset = inode->i_size; | |
238 | break; | |
239 | } | |
240 | retval = -EINVAL; | |
241 | if (offset >= 0 || unsigned_offsets(file)) { | |
242 | if (offset != file->f_pos) { | |
243 | file->f_pos = offset; | |
244 | file->f_version = 0; | |
245 | } | |
246 | retval = offset; | |
247 | } | |
248 | out: | |
249 | mutex_unlock(&inode->i_mutex); | |
250 | return retval; | |
251 | } | |
252 | EXPORT_SYMBOL(default_llseek); | |
253 | ||
254 | loff_t vfs_llseek(struct file *file, loff_t offset, int whence) | |
255 | { | |
256 | loff_t (*fn)(struct file *, loff_t, int); | |
257 | ||
258 | fn = no_llseek; | |
259 | if (file->f_mode & FMODE_LSEEK) { | |
260 | if (file->f_op->llseek) | |
261 | fn = file->f_op->llseek; | |
262 | } | |
263 | return fn(file, offset, whence); | |
264 | } | |
265 | EXPORT_SYMBOL(vfs_llseek); | |
266 | ||
267 | static inline struct fd fdget_pos(int fd) | |
268 | { | |
269 | return __to_fd(__fdget_pos(fd)); | |
270 | } | |
271 | ||
272 | static inline void fdput_pos(struct fd f) | |
273 | { | |
274 | if (f.flags & FDPUT_POS_UNLOCK) | |
275 | mutex_unlock(&f.file->f_pos_lock); | |
276 | fdput(f); | |
277 | } | |
278 | ||
279 | SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence) | |
280 | { | |
281 | off_t retval; | |
282 | struct fd f = fdget_pos(fd); | |
283 | if (!f.file) | |
284 | return -EBADF; | |
285 | ||
286 | retval = -EINVAL; | |
287 | if (whence <= SEEK_MAX) { | |
288 | loff_t res = vfs_llseek(f.file, offset, whence); | |
289 | retval = res; | |
290 | if (res != (loff_t)retval) | |
291 | retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */ | |
292 | } | |
293 | fdput_pos(f); | |
294 | return retval; | |
295 | } | |
296 | ||
297 | #ifdef CONFIG_COMPAT | |
298 | COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence) | |
299 | { | |
300 | return sys_lseek(fd, offset, whence); | |
301 | } | |
302 | #endif | |
303 | ||
304 | #ifdef __ARCH_WANT_SYS_LLSEEK | |
305 | SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high, | |
306 | unsigned long, offset_low, loff_t __user *, result, | |
307 | unsigned int, whence) | |
308 | { | |
309 | int retval; | |
310 | struct fd f = fdget_pos(fd); | |
311 | loff_t offset; | |
312 | ||
313 | if (!f.file) | |
314 | return -EBADF; | |
315 | ||
316 | retval = -EINVAL; | |
317 | if (whence > SEEK_MAX) | |
318 | goto out_putf; | |
319 | ||
320 | offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low, | |
321 | whence); | |
322 | ||
323 | retval = (int)offset; | |
324 | if (offset >= 0) { | |
325 | retval = -EFAULT; | |
326 | if (!copy_to_user(result, &offset, sizeof(offset))) | |
327 | retval = 0; | |
328 | } | |
329 | out_putf: | |
330 | fdput_pos(f); | |
331 | return retval; | |
332 | } | |
333 | #endif | |
334 | ||
335 | ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos) | |
336 | { | |
337 | struct kiocb kiocb; | |
338 | ssize_t ret; | |
339 | ||
340 | if (!file->f_op->read_iter) | |
341 | return -EINVAL; | |
342 | ||
343 | init_sync_kiocb(&kiocb, file); | |
344 | kiocb.ki_pos = *ppos; | |
345 | ||
346 | iter->type |= READ; | |
347 | ret = file->f_op->read_iter(&kiocb, iter); | |
348 | BUG_ON(ret == -EIOCBQUEUED); | |
349 | if (ret > 0) | |
350 | *ppos = kiocb.ki_pos; | |
351 | return ret; | |
352 | } | |
353 | EXPORT_SYMBOL(vfs_iter_read); | |
354 | ||
355 | ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos) | |
356 | { | |
357 | struct kiocb kiocb; | |
358 | ssize_t ret; | |
359 | ||
360 | if (!file->f_op->write_iter) | |
361 | return -EINVAL; | |
362 | ||
363 | init_sync_kiocb(&kiocb, file); | |
364 | kiocb.ki_pos = *ppos; | |
365 | ||
366 | iter->type |= WRITE; | |
367 | ret = file->f_op->write_iter(&kiocb, iter); | |
368 | BUG_ON(ret == -EIOCBQUEUED); | |
369 | if (ret > 0) | |
370 | *ppos = kiocb.ki_pos; | |
371 | return ret; | |
372 | } | |
373 | EXPORT_SYMBOL(vfs_iter_write); | |
374 | ||
375 | /* | |
376 | * rw_verify_area doesn't like huge counts. We limit | |
377 | * them to something that fits in "int" so that others | |
378 | * won't have to do range checks all the time. | |
379 | */ | |
380 | int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count) | |
381 | { | |
382 | struct inode *inode; | |
383 | loff_t pos; | |
384 | int retval = -EINVAL; | |
385 | ||
386 | inode = file_inode(file); | |
387 | if (unlikely((ssize_t) count < 0)) | |
388 | return retval; | |
389 | pos = *ppos; | |
390 | if (unlikely(pos < 0)) { | |
391 | if (!unsigned_offsets(file)) | |
392 | return retval; | |
393 | if (count >= -pos) /* both values are in 0..LLONG_MAX */ | |
394 | return -EOVERFLOW; | |
395 | } else if (unlikely((loff_t) (pos + count) < 0)) { | |
396 | if (!unsigned_offsets(file)) | |
397 | return retval; | |
398 | } | |
399 | ||
400 | if (unlikely(inode->i_flctx && mandatory_lock(inode))) { | |
401 | retval = locks_mandatory_area( | |
402 | read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE, | |
403 | inode, file, pos, count); | |
404 | if (retval < 0) | |
405 | return retval; | |
406 | } | |
407 | retval = security_file_permission(file, | |
408 | read_write == READ ? MAY_READ : MAY_WRITE); | |
409 | if (retval) | |
410 | return retval; | |
411 | return count > MAX_RW_COUNT ? MAX_RW_COUNT : count; | |
412 | } | |
413 | ||
414 | ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos) | |
415 | { | |
416 | struct iovec iov = { .iov_base = buf, .iov_len = len }; | |
417 | struct kiocb kiocb; | |
418 | ssize_t ret; | |
419 | ||
420 | init_sync_kiocb(&kiocb, filp); | |
421 | kiocb.ki_pos = *ppos; | |
422 | ||
423 | ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos); | |
424 | BUG_ON(ret == -EIOCBQUEUED); | |
425 | *ppos = kiocb.ki_pos; | |
426 | return ret; | |
427 | } | |
428 | ||
429 | EXPORT_SYMBOL(do_sync_read); | |
430 | ||
431 | ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos) | |
432 | { | |
433 | struct iovec iov = { .iov_base = buf, .iov_len = len }; | |
434 | struct kiocb kiocb; | |
435 | struct iov_iter iter; | |
436 | ssize_t ret; | |
437 | ||
438 | init_sync_kiocb(&kiocb, filp); | |
439 | kiocb.ki_pos = *ppos; | |
440 | iov_iter_init(&iter, READ, &iov, 1, len); | |
441 | ||
442 | ret = filp->f_op->read_iter(&kiocb, &iter); | |
443 | BUG_ON(ret == -EIOCBQUEUED); | |
444 | *ppos = kiocb.ki_pos; | |
445 | return ret; | |
446 | } | |
447 | ||
448 | EXPORT_SYMBOL(new_sync_read); | |
449 | ||
450 | ssize_t __vfs_read(struct file *file, char __user *buf, size_t count, | |
451 | loff_t *pos) | |
452 | { | |
453 | ssize_t ret; | |
454 | ||
455 | if (file->f_op->read) | |
456 | ret = file->f_op->read(file, buf, count, pos); | |
457 | else if (file->f_op->aio_read) | |
458 | ret = do_sync_read(file, buf, count, pos); | |
459 | else if (file->f_op->read_iter) | |
460 | ret = new_sync_read(file, buf, count, pos); | |
461 | else | |
462 | ret = -EINVAL; | |
463 | ||
464 | return ret; | |
465 | } | |
466 | ||
467 | ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos) | |
468 | { | |
469 | ssize_t ret; | |
470 | ||
471 | if (!(file->f_mode & FMODE_READ)) | |
472 | return -EBADF; | |
473 | if (!(file->f_mode & FMODE_CAN_READ)) | |
474 | return -EINVAL; | |
475 | if (unlikely(!access_ok(VERIFY_WRITE, buf, count))) | |
476 | return -EFAULT; | |
477 | ||
478 | ret = rw_verify_area(READ, file, pos, count); | |
479 | if (ret >= 0) { | |
480 | count = ret; | |
481 | ret = __vfs_read(file, buf, count, pos); | |
482 | if (ret > 0) { | |
483 | fsnotify_access(file); | |
484 | add_rchar(current, ret); | |
485 | } | |
486 | inc_syscr(current); | |
487 | } | |
488 | ||
489 | return ret; | |
490 | } | |
491 | ||
492 | EXPORT_SYMBOL(vfs_read); | |
493 | ||
494 | ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos) | |
495 | { | |
496 | struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len }; | |
497 | struct kiocb kiocb; | |
498 | ssize_t ret; | |
499 | ||
500 | init_sync_kiocb(&kiocb, filp); | |
501 | kiocb.ki_pos = *ppos; | |
502 | ||
503 | ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos); | |
504 | BUG_ON(ret == -EIOCBQUEUED); | |
505 | *ppos = kiocb.ki_pos; | |
506 | return ret; | |
507 | } | |
508 | ||
509 | EXPORT_SYMBOL(do_sync_write); | |
510 | ||
511 | ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos) | |
512 | { | |
513 | struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len }; | |
514 | struct kiocb kiocb; | |
515 | struct iov_iter iter; | |
516 | ssize_t ret; | |
517 | ||
518 | init_sync_kiocb(&kiocb, filp); | |
519 | kiocb.ki_pos = *ppos; | |
520 | iov_iter_init(&iter, WRITE, &iov, 1, len); | |
521 | ||
522 | ret = filp->f_op->write_iter(&kiocb, &iter); | |
523 | BUG_ON(ret == -EIOCBQUEUED); | |
524 | *ppos = kiocb.ki_pos; | |
525 | return ret; | |
526 | } | |
527 | ||
528 | EXPORT_SYMBOL(new_sync_write); | |
529 | ||
530 | ssize_t __vfs_write(struct file *file, const char __user *p, size_t count, | |
531 | loff_t *pos) | |
532 | { | |
533 | if (file->f_op->write) | |
534 | return file->f_op->write(file, p, count, pos); | |
535 | else if (file->f_op->aio_write) | |
536 | return do_sync_write(file, p, count, pos); | |
537 | else if (file->f_op->write_iter) | |
538 | return new_sync_write(file, p, count, pos); | |
539 | else | |
540 | return -EINVAL; | |
541 | } | |
542 | EXPORT_SYMBOL(__vfs_write); | |
543 | ||
544 | ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos) | |
545 | { | |
546 | mm_segment_t old_fs; | |
547 | const char __user *p; | |
548 | ssize_t ret; | |
549 | ||
550 | if (!(file->f_mode & FMODE_CAN_WRITE)) | |
551 | return -EINVAL; | |
552 | ||
553 | old_fs = get_fs(); | |
554 | set_fs(get_ds()); | |
555 | p = (__force const char __user *)buf; | |
556 | if (count > MAX_RW_COUNT) | |
557 | count = MAX_RW_COUNT; | |
558 | ret = __vfs_write(file, p, count, pos); | |
559 | set_fs(old_fs); | |
560 | if (ret > 0) { | |
561 | fsnotify_modify(file); | |
562 | add_wchar(current, ret); | |
563 | } | |
564 | inc_syscw(current); | |
565 | return ret; | |
566 | } | |
567 | ||
568 | EXPORT_SYMBOL(__kernel_write); | |
569 | ||
570 | ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos) | |
571 | { | |
572 | ssize_t ret; | |
573 | ||
574 | if (!(file->f_mode & FMODE_WRITE)) | |
575 | return -EBADF; | |
576 | if (!(file->f_mode & FMODE_CAN_WRITE)) | |
577 | return -EINVAL; | |
578 | if (unlikely(!access_ok(VERIFY_READ, buf, count))) | |
579 | return -EFAULT; | |
580 | ||
581 | ret = rw_verify_area(WRITE, file, pos, count); | |
582 | if (ret >= 0) { | |
583 | count = ret; | |
584 | file_start_write(file); | |
585 | ret = __vfs_write(file, buf, count, pos); | |
586 | if (ret > 0) { | |
587 | fsnotify_modify(file); | |
588 | add_wchar(current, ret); | |
589 | } | |
590 | inc_syscw(current); | |
591 | file_end_write(file); | |
592 | } | |
593 | ||
594 | return ret; | |
595 | } | |
596 | ||
597 | EXPORT_SYMBOL(vfs_write); | |
598 | ||
599 | static inline loff_t file_pos_read(struct file *file) | |
600 | { | |
601 | return file->f_pos; | |
602 | } | |
603 | ||
604 | static inline void file_pos_write(struct file *file, loff_t pos) | |
605 | { | |
606 | file->f_pos = pos; | |
607 | } | |
608 | ||
609 | SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count) | |
610 | { | |
611 | struct fd f = fdget_pos(fd); | |
612 | ssize_t ret = -EBADF; | |
613 | ||
614 | if (f.file) { | |
615 | loff_t pos = file_pos_read(f.file); | |
616 | ret = vfs_read(f.file, buf, count, &pos); | |
617 | if (ret >= 0) | |
618 | file_pos_write(f.file, pos); | |
619 | fdput_pos(f); | |
620 | } | |
621 | return ret; | |
622 | } | |
623 | ||
624 | SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf, | |
625 | size_t, count) | |
626 | { | |
627 | struct fd f = fdget_pos(fd); | |
628 | ssize_t ret = -EBADF; | |
629 | ||
630 | if (f.file) { | |
631 | loff_t pos = file_pos_read(f.file); | |
632 | ret = vfs_write(f.file, buf, count, &pos); | |
633 | if (ret >= 0) | |
634 | file_pos_write(f.file, pos); | |
635 | fdput_pos(f); | |
636 | } | |
637 | ||
638 | return ret; | |
639 | } | |
640 | ||
641 | SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf, | |
642 | size_t, count, loff_t, pos) | |
643 | { | |
644 | struct fd f; | |
645 | ssize_t ret = -EBADF; | |
646 | ||
647 | if (pos < 0) | |
648 | return -EINVAL; | |
649 | ||
650 | f = fdget(fd); | |
651 | if (f.file) { | |
652 | ret = -ESPIPE; | |
653 | if (f.file->f_mode & FMODE_PREAD) | |
654 | ret = vfs_read(f.file, buf, count, &pos); | |
655 | fdput(f); | |
656 | } | |
657 | ||
658 | return ret; | |
659 | } | |
660 | ||
661 | SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf, | |
662 | size_t, count, loff_t, pos) | |
663 | { | |
664 | struct fd f; | |
665 | ssize_t ret = -EBADF; | |
666 | ||
667 | if (pos < 0) | |
668 | return -EINVAL; | |
669 | ||
670 | f = fdget(fd); | |
671 | if (f.file) { | |
672 | ret = -ESPIPE; | |
673 | if (f.file->f_mode & FMODE_PWRITE) | |
674 | ret = vfs_write(f.file, buf, count, &pos); | |
675 | fdput(f); | |
676 | } | |
677 | ||
678 | return ret; | |
679 | } | |
680 | ||
681 | /* | |
682 | * Reduce an iovec's length in-place. Return the resulting number of segments | |
683 | */ | |
684 | unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to) | |
685 | { | |
686 | unsigned long seg = 0; | |
687 | size_t len = 0; | |
688 | ||
689 | while (seg < nr_segs) { | |
690 | seg++; | |
691 | if (len + iov->iov_len >= to) { | |
692 | iov->iov_len = to - len; | |
693 | break; | |
694 | } | |
695 | len += iov->iov_len; | |
696 | iov++; | |
697 | } | |
698 | return seg; | |
699 | } | |
700 | EXPORT_SYMBOL(iov_shorten); | |
701 | ||
702 | static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter, | |
703 | loff_t *ppos, iter_fn_t fn) | |
704 | { | |
705 | struct kiocb kiocb; | |
706 | ssize_t ret; | |
707 | ||
708 | init_sync_kiocb(&kiocb, filp); | |
709 | kiocb.ki_pos = *ppos; | |
710 | ||
711 | ret = fn(&kiocb, iter); | |
712 | BUG_ON(ret == -EIOCBQUEUED); | |
713 | *ppos = kiocb.ki_pos; | |
714 | return ret; | |
715 | } | |
716 | ||
717 | static ssize_t do_sync_readv_writev(struct file *filp, struct iov_iter *iter, | |
718 | loff_t *ppos, iov_fn_t fn) | |
719 | { | |
720 | struct kiocb kiocb; | |
721 | ssize_t ret; | |
722 | ||
723 | init_sync_kiocb(&kiocb, filp); | |
724 | kiocb.ki_pos = *ppos; | |
725 | ||
726 | ret = fn(&kiocb, iter->iov, iter->nr_segs, kiocb.ki_pos); | |
727 | BUG_ON(ret == -EIOCBQUEUED); | |
728 | *ppos = kiocb.ki_pos; | |
729 | return ret; | |
730 | } | |
731 | ||
732 | /* Do it by hand, with file-ops */ | |
733 | static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter, | |
734 | loff_t *ppos, io_fn_t fn) | |
735 | { | |
736 | ssize_t ret = 0; | |
737 | ||
738 | while (iov_iter_count(iter)) { | |
739 | struct iovec iovec = iov_iter_iovec(iter); | |
740 | ssize_t nr; | |
741 | ||
742 | nr = fn(filp, iovec.iov_base, iovec.iov_len, ppos); | |
743 | ||
744 | if (nr < 0) { | |
745 | if (!ret) | |
746 | ret = nr; | |
747 | break; | |
748 | } | |
749 | ret += nr; | |
750 | if (nr != iovec.iov_len) | |
751 | break; | |
752 | iov_iter_advance(iter, nr); | |
753 | } | |
754 | ||
755 | return ret; | |
756 | } | |
757 | ||
758 | /* A write operation does a read from user space and vice versa */ | |
759 | #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ) | |
760 | ||
761 | ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector, | |
762 | unsigned long nr_segs, unsigned long fast_segs, | |
763 | struct iovec *fast_pointer, | |
764 | struct iovec **ret_pointer) | |
765 | { | |
766 | unsigned long seg; | |
767 | ssize_t ret; | |
768 | struct iovec *iov = fast_pointer; | |
769 | ||
770 | /* | |
771 | * SuS says "The readv() function *may* fail if the iovcnt argument | |
772 | * was less than or equal to 0, or greater than {IOV_MAX}. Linux has | |
773 | * traditionally returned zero for zero segments, so... | |
774 | */ | |
775 | if (nr_segs == 0) { | |
776 | ret = 0; | |
777 | goto out; | |
778 | } | |
779 | ||
780 | /* | |
781 | * First get the "struct iovec" from user memory and | |
782 | * verify all the pointers | |
783 | */ | |
784 | if (nr_segs > UIO_MAXIOV) { | |
785 | ret = -EINVAL; | |
786 | goto out; | |
787 | } | |
788 | if (nr_segs > fast_segs) { | |
789 | iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL); | |
790 | if (iov == NULL) { | |
791 | ret = -ENOMEM; | |
792 | goto out; | |
793 | } | |
794 | } | |
795 | if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) { | |
796 | ret = -EFAULT; | |
797 | goto out; | |
798 | } | |
799 | ||
800 | /* | |
801 | * According to the Single Unix Specification we should return EINVAL | |
802 | * if an element length is < 0 when cast to ssize_t or if the | |
803 | * total length would overflow the ssize_t return value of the | |
804 | * system call. | |
805 | * | |
806 | * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the | |
807 | * overflow case. | |
808 | */ | |
809 | ret = 0; | |
810 | for (seg = 0; seg < nr_segs; seg++) { | |
811 | void __user *buf = iov[seg].iov_base; | |
812 | ssize_t len = (ssize_t)iov[seg].iov_len; | |
813 | ||
814 | /* see if we we're about to use an invalid len or if | |
815 | * it's about to overflow ssize_t */ | |
816 | if (len < 0) { | |
817 | ret = -EINVAL; | |
818 | goto out; | |
819 | } | |
820 | if (type >= 0 | |
821 | && unlikely(!access_ok(vrfy_dir(type), buf, len))) { | |
822 | ret = -EFAULT; | |
823 | goto out; | |
824 | } | |
825 | if (len > MAX_RW_COUNT - ret) { | |
826 | len = MAX_RW_COUNT - ret; | |
827 | iov[seg].iov_len = len; | |
828 | } | |
829 | ret += len; | |
830 | } | |
831 | out: | |
832 | *ret_pointer = iov; | |
833 | return ret; | |
834 | } | |
835 | ||
836 | static ssize_t do_readv_writev(int type, struct file *file, | |
837 | const struct iovec __user * uvector, | |
838 | unsigned long nr_segs, loff_t *pos) | |
839 | { | |
840 | size_t tot_len; | |
841 | struct iovec iovstack[UIO_FASTIOV]; | |
842 | struct iovec *iov = iovstack; | |
843 | struct iov_iter iter; | |
844 | ssize_t ret; | |
845 | io_fn_t fn; | |
846 | iov_fn_t fnv; | |
847 | iter_fn_t iter_fn; | |
848 | ||
849 | ret = import_iovec(type, uvector, nr_segs, | |
850 | ARRAY_SIZE(iovstack), &iov, &iter); | |
851 | if (ret < 0) | |
852 | return ret; | |
853 | ||
854 | tot_len = iov_iter_count(&iter); | |
855 | if (!tot_len) | |
856 | goto out; | |
857 | ret = rw_verify_area(type, file, pos, tot_len); | |
858 | if (ret < 0) | |
859 | goto out; | |
860 | ||
861 | fnv = NULL; | |
862 | if (type == READ) { | |
863 | fn = file->f_op->read; | |
864 | fnv = file->f_op->aio_read; | |
865 | iter_fn = file->f_op->read_iter; | |
866 | } else { | |
867 | fn = (io_fn_t)file->f_op->write; | |
868 | fnv = file->f_op->aio_write; | |
869 | iter_fn = file->f_op->write_iter; | |
870 | file_start_write(file); | |
871 | } | |
872 | ||
873 | if (iter_fn) | |
874 | ret = do_iter_readv_writev(file, &iter, pos, iter_fn); | |
875 | else if (fnv) | |
876 | ret = do_sync_readv_writev(file, &iter, pos, fnv); | |
877 | else | |
878 | ret = do_loop_readv_writev(file, &iter, pos, fn); | |
879 | ||
880 | if (type != READ) | |
881 | file_end_write(file); | |
882 | ||
883 | out: | |
884 | kfree(iov); | |
885 | if ((ret + (type == READ)) > 0) { | |
886 | if (type == READ) | |
887 | fsnotify_access(file); | |
888 | else | |
889 | fsnotify_modify(file); | |
890 | } | |
891 | return ret; | |
892 | } | |
893 | ||
894 | ssize_t vfs_readv(struct file *file, const struct iovec __user *vec, | |
895 | unsigned long vlen, loff_t *pos) | |
896 | { | |
897 | if (!(file->f_mode & FMODE_READ)) | |
898 | return -EBADF; | |
899 | if (!(file->f_mode & FMODE_CAN_READ)) | |
900 | return -EINVAL; | |
901 | ||
902 | return do_readv_writev(READ, file, vec, vlen, pos); | |
903 | } | |
904 | ||
905 | EXPORT_SYMBOL(vfs_readv); | |
906 | ||
907 | ssize_t vfs_writev(struct file *file, const struct iovec __user *vec, | |
908 | unsigned long vlen, loff_t *pos) | |
909 | { | |
910 | if (!(file->f_mode & FMODE_WRITE)) | |
911 | return -EBADF; | |
912 | if (!(file->f_mode & FMODE_CAN_WRITE)) | |
913 | return -EINVAL; | |
914 | ||
915 | return do_readv_writev(WRITE, file, vec, vlen, pos); | |
916 | } | |
917 | ||
918 | EXPORT_SYMBOL(vfs_writev); | |
919 | ||
920 | SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec, | |
921 | unsigned long, vlen) | |
922 | { | |
923 | struct fd f = fdget_pos(fd); | |
924 | ssize_t ret = -EBADF; | |
925 | ||
926 | if (f.file) { | |
927 | loff_t pos = file_pos_read(f.file); | |
928 | ret = vfs_readv(f.file, vec, vlen, &pos); | |
929 | if (ret >= 0) | |
930 | file_pos_write(f.file, pos); | |
931 | fdput_pos(f); | |
932 | } | |
933 | ||
934 | if (ret > 0) | |
935 | add_rchar(current, ret); | |
936 | inc_syscr(current); | |
937 | return ret; | |
938 | } | |
939 | ||
940 | SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec, | |
941 | unsigned long, vlen) | |
942 | { | |
943 | struct fd f = fdget_pos(fd); | |
944 | ssize_t ret = -EBADF; | |
945 | ||
946 | if (f.file) { | |
947 | loff_t pos = file_pos_read(f.file); | |
948 | ret = vfs_writev(f.file, vec, vlen, &pos); | |
949 | if (ret >= 0) | |
950 | file_pos_write(f.file, pos); | |
951 | fdput_pos(f); | |
952 | } | |
953 | ||
954 | if (ret > 0) | |
955 | add_wchar(current, ret); | |
956 | inc_syscw(current); | |
957 | return ret; | |
958 | } | |
959 | ||
960 | static inline loff_t pos_from_hilo(unsigned long high, unsigned long low) | |
961 | { | |
962 | #define HALF_LONG_BITS (BITS_PER_LONG / 2) | |
963 | return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low; | |
964 | } | |
965 | ||
966 | SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec, | |
967 | unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h) | |
968 | { | |
969 | loff_t pos = pos_from_hilo(pos_h, pos_l); | |
970 | struct fd f; | |
971 | ssize_t ret = -EBADF; | |
972 | ||
973 | if (pos < 0) | |
974 | return -EINVAL; | |
975 | ||
976 | f = fdget(fd); | |
977 | if (f.file) { | |
978 | ret = -ESPIPE; | |
979 | if (f.file->f_mode & FMODE_PREAD) | |
980 | ret = vfs_readv(f.file, vec, vlen, &pos); | |
981 | fdput(f); | |
982 | } | |
983 | ||
984 | if (ret > 0) | |
985 | add_rchar(current, ret); | |
986 | inc_syscr(current); | |
987 | return ret; | |
988 | } | |
989 | ||
990 | SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec, | |
991 | unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h) | |
992 | { | |
993 | loff_t pos = pos_from_hilo(pos_h, pos_l); | |
994 | struct fd f; | |
995 | ssize_t ret = -EBADF; | |
996 | ||
997 | if (pos < 0) | |
998 | return -EINVAL; | |
999 | ||
1000 | f = fdget(fd); | |
1001 | if (f.file) { | |
1002 | ret = -ESPIPE; | |
1003 | if (f.file->f_mode & FMODE_PWRITE) | |
1004 | ret = vfs_writev(f.file, vec, vlen, &pos); | |
1005 | fdput(f); | |
1006 | } | |
1007 | ||
1008 | if (ret > 0) | |
1009 | add_wchar(current, ret); | |
1010 | inc_syscw(current); | |
1011 | return ret; | |
1012 | } | |
1013 | ||
1014 | #ifdef CONFIG_COMPAT | |
1015 | ||
1016 | static ssize_t compat_do_readv_writev(int type, struct file *file, | |
1017 | const struct compat_iovec __user *uvector, | |
1018 | unsigned long nr_segs, loff_t *pos) | |
1019 | { | |
1020 | compat_ssize_t tot_len; | |
1021 | struct iovec iovstack[UIO_FASTIOV]; | |
1022 | struct iovec *iov = iovstack; | |
1023 | struct iov_iter iter; | |
1024 | ssize_t ret; | |
1025 | io_fn_t fn; | |
1026 | iov_fn_t fnv; | |
1027 | iter_fn_t iter_fn; | |
1028 | ||
1029 | ret = compat_import_iovec(type, uvector, nr_segs, | |
1030 | UIO_FASTIOV, &iov, &iter); | |
1031 | if (ret < 0) | |
1032 | return ret; | |
1033 | ||
1034 | tot_len = iov_iter_count(&iter); | |
1035 | if (!tot_len) | |
1036 | goto out; | |
1037 | ret = rw_verify_area(type, file, pos, tot_len); | |
1038 | if (ret < 0) | |
1039 | goto out; | |
1040 | ||
1041 | fnv = NULL; | |
1042 | if (type == READ) { | |
1043 | fn = file->f_op->read; | |
1044 | fnv = file->f_op->aio_read; | |
1045 | iter_fn = file->f_op->read_iter; | |
1046 | } else { | |
1047 | fn = (io_fn_t)file->f_op->write; | |
1048 | fnv = file->f_op->aio_write; | |
1049 | iter_fn = file->f_op->write_iter; | |
1050 | file_start_write(file); | |
1051 | } | |
1052 | ||
1053 | if (iter_fn) | |
1054 | ret = do_iter_readv_writev(file, &iter, pos, iter_fn); | |
1055 | else if (fnv) | |
1056 | ret = do_sync_readv_writev(file, &iter, pos, fnv); | |
1057 | else | |
1058 | ret = do_loop_readv_writev(file, &iter, pos, fn); | |
1059 | ||
1060 | if (type != READ) | |
1061 | file_end_write(file); | |
1062 | ||
1063 | out: | |
1064 | kfree(iov); | |
1065 | if ((ret + (type == READ)) > 0) { | |
1066 | if (type == READ) | |
1067 | fsnotify_access(file); | |
1068 | else | |
1069 | fsnotify_modify(file); | |
1070 | } | |
1071 | return ret; | |
1072 | } | |
1073 | ||
1074 | static size_t compat_readv(struct file *file, | |
1075 | const struct compat_iovec __user *vec, | |
1076 | unsigned long vlen, loff_t *pos) | |
1077 | { | |
1078 | ssize_t ret = -EBADF; | |
1079 | ||
1080 | if (!(file->f_mode & FMODE_READ)) | |
1081 | goto out; | |
1082 | ||
1083 | ret = -EINVAL; | |
1084 | if (!(file->f_mode & FMODE_CAN_READ)) | |
1085 | goto out; | |
1086 | ||
1087 | ret = compat_do_readv_writev(READ, file, vec, vlen, pos); | |
1088 | ||
1089 | out: | |
1090 | if (ret > 0) | |
1091 | add_rchar(current, ret); | |
1092 | inc_syscr(current); | |
1093 | return ret; | |
1094 | } | |
1095 | ||
1096 | COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd, | |
1097 | const struct compat_iovec __user *,vec, | |
1098 | compat_ulong_t, vlen) | |
1099 | { | |
1100 | struct fd f = fdget_pos(fd); | |
1101 | ssize_t ret; | |
1102 | loff_t pos; | |
1103 | ||
1104 | if (!f.file) | |
1105 | return -EBADF; | |
1106 | pos = f.file->f_pos; | |
1107 | ret = compat_readv(f.file, vec, vlen, &pos); | |
1108 | if (ret >= 0) | |
1109 | f.file->f_pos = pos; | |
1110 | fdput_pos(f); | |
1111 | return ret; | |
1112 | } | |
1113 | ||
1114 | static long __compat_sys_preadv64(unsigned long fd, | |
1115 | const struct compat_iovec __user *vec, | |
1116 | unsigned long vlen, loff_t pos) | |
1117 | { | |
1118 | struct fd f; | |
1119 | ssize_t ret; | |
1120 | ||
1121 | if (pos < 0) | |
1122 | return -EINVAL; | |
1123 | f = fdget(fd); | |
1124 | if (!f.file) | |
1125 | return -EBADF; | |
1126 | ret = -ESPIPE; | |
1127 | if (f.file->f_mode & FMODE_PREAD) | |
1128 | ret = compat_readv(f.file, vec, vlen, &pos); | |
1129 | fdput(f); | |
1130 | return ret; | |
1131 | } | |
1132 | ||
1133 | #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64 | |
1134 | COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd, | |
1135 | const struct compat_iovec __user *,vec, | |
1136 | unsigned long, vlen, loff_t, pos) | |
1137 | { | |
1138 | return __compat_sys_preadv64(fd, vec, vlen, pos); | |
1139 | } | |
1140 | #endif | |
1141 | ||
1142 | COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd, | |
1143 | const struct compat_iovec __user *,vec, | |
1144 | compat_ulong_t, vlen, u32, pos_low, u32, pos_high) | |
1145 | { | |
1146 | loff_t pos = ((loff_t)pos_high << 32) | pos_low; | |
1147 | ||
1148 | return __compat_sys_preadv64(fd, vec, vlen, pos); | |
1149 | } | |
1150 | ||
1151 | static size_t compat_writev(struct file *file, | |
1152 | const struct compat_iovec __user *vec, | |
1153 | unsigned long vlen, loff_t *pos) | |
1154 | { | |
1155 | ssize_t ret = -EBADF; | |
1156 | ||
1157 | if (!(file->f_mode & FMODE_WRITE)) | |
1158 | goto out; | |
1159 | ||
1160 | ret = -EINVAL; | |
1161 | if (!(file->f_mode & FMODE_CAN_WRITE)) | |
1162 | goto out; | |
1163 | ||
1164 | ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos); | |
1165 | ||
1166 | out: | |
1167 | if (ret > 0) | |
1168 | add_wchar(current, ret); | |
1169 | inc_syscw(current); | |
1170 | return ret; | |
1171 | } | |
1172 | ||
1173 | COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd, | |
1174 | const struct compat_iovec __user *, vec, | |
1175 | compat_ulong_t, vlen) | |
1176 | { | |
1177 | struct fd f = fdget_pos(fd); | |
1178 | ssize_t ret; | |
1179 | loff_t pos; | |
1180 | ||
1181 | if (!f.file) | |
1182 | return -EBADF; | |
1183 | pos = f.file->f_pos; | |
1184 | ret = compat_writev(f.file, vec, vlen, &pos); | |
1185 | if (ret >= 0) | |
1186 | f.file->f_pos = pos; | |
1187 | fdput_pos(f); | |
1188 | return ret; | |
1189 | } | |
1190 | ||
1191 | static long __compat_sys_pwritev64(unsigned long fd, | |
1192 | const struct compat_iovec __user *vec, | |
1193 | unsigned long vlen, loff_t pos) | |
1194 | { | |
1195 | struct fd f; | |
1196 | ssize_t ret; | |
1197 | ||
1198 | if (pos < 0) | |
1199 | return -EINVAL; | |
1200 | f = fdget(fd); | |
1201 | if (!f.file) | |
1202 | return -EBADF; | |
1203 | ret = -ESPIPE; | |
1204 | if (f.file->f_mode & FMODE_PWRITE) | |
1205 | ret = compat_writev(f.file, vec, vlen, &pos); | |
1206 | fdput(f); | |
1207 | return ret; | |
1208 | } | |
1209 | ||
1210 | #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64 | |
1211 | COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd, | |
1212 | const struct compat_iovec __user *,vec, | |
1213 | unsigned long, vlen, loff_t, pos) | |
1214 | { | |
1215 | return __compat_sys_pwritev64(fd, vec, vlen, pos); | |
1216 | } | |
1217 | #endif | |
1218 | ||
1219 | COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd, | |
1220 | const struct compat_iovec __user *,vec, | |
1221 | compat_ulong_t, vlen, u32, pos_low, u32, pos_high) | |
1222 | { | |
1223 | loff_t pos = ((loff_t)pos_high << 32) | pos_low; | |
1224 | ||
1225 | return __compat_sys_pwritev64(fd, vec, vlen, pos); | |
1226 | } | |
1227 | #endif | |
1228 | ||
1229 | static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos, | |
1230 | size_t count, loff_t max) | |
1231 | { | |
1232 | struct fd in, out; | |
1233 | struct inode *in_inode, *out_inode; | |
1234 | loff_t pos; | |
1235 | loff_t out_pos; | |
1236 | ssize_t retval; | |
1237 | int fl; | |
1238 | ||
1239 | /* | |
1240 | * Get input file, and verify that it is ok.. | |
1241 | */ | |
1242 | retval = -EBADF; | |
1243 | in = fdget(in_fd); | |
1244 | if (!in.file) | |
1245 | goto out; | |
1246 | if (!(in.file->f_mode & FMODE_READ)) | |
1247 | goto fput_in; | |
1248 | retval = -ESPIPE; | |
1249 | if (!ppos) { | |
1250 | pos = in.file->f_pos; | |
1251 | } else { | |
1252 | pos = *ppos; | |
1253 | if (!(in.file->f_mode & FMODE_PREAD)) | |
1254 | goto fput_in; | |
1255 | } | |
1256 | retval = rw_verify_area(READ, in.file, &pos, count); | |
1257 | if (retval < 0) | |
1258 | goto fput_in; | |
1259 | count = retval; | |
1260 | ||
1261 | /* | |
1262 | * Get output file, and verify that it is ok.. | |
1263 | */ | |
1264 | retval = -EBADF; | |
1265 | out = fdget(out_fd); | |
1266 | if (!out.file) | |
1267 | goto fput_in; | |
1268 | if (!(out.file->f_mode & FMODE_WRITE)) | |
1269 | goto fput_out; | |
1270 | retval = -EINVAL; | |
1271 | in_inode = file_inode(in.file); | |
1272 | out_inode = file_inode(out.file); | |
1273 | out_pos = out.file->f_pos; | |
1274 | retval = rw_verify_area(WRITE, out.file, &out_pos, count); | |
1275 | if (retval < 0) | |
1276 | goto fput_out; | |
1277 | count = retval; | |
1278 | ||
1279 | if (!max) | |
1280 | max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes); | |
1281 | ||
1282 | if (unlikely(pos + count > max)) { | |
1283 | retval = -EOVERFLOW; | |
1284 | if (pos >= max) | |
1285 | goto fput_out; | |
1286 | count = max - pos; | |
1287 | } | |
1288 | ||
1289 | fl = 0; | |
1290 | #if 0 | |
1291 | /* | |
1292 | * We need to debate whether we can enable this or not. The | |
1293 | * man page documents EAGAIN return for the output at least, | |
1294 | * and the application is arguably buggy if it doesn't expect | |
1295 | * EAGAIN on a non-blocking file descriptor. | |
1296 | */ | |
1297 | if (in.file->f_flags & O_NONBLOCK) | |
1298 | fl = SPLICE_F_NONBLOCK; | |
1299 | #endif | |
1300 | file_start_write(out.file); | |
1301 | retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl); | |
1302 | file_end_write(out.file); | |
1303 | ||
1304 | if (retval > 0) { | |
1305 | add_rchar(current, retval); | |
1306 | add_wchar(current, retval); | |
1307 | fsnotify_access(in.file); | |
1308 | fsnotify_modify(out.file); | |
1309 | out.file->f_pos = out_pos; | |
1310 | if (ppos) | |
1311 | *ppos = pos; | |
1312 | else | |
1313 | in.file->f_pos = pos; | |
1314 | } | |
1315 | ||
1316 | inc_syscr(current); | |
1317 | inc_syscw(current); | |
1318 | if (pos > max) | |
1319 | retval = -EOVERFLOW; | |
1320 | ||
1321 | fput_out: | |
1322 | fdput(out); | |
1323 | fput_in: | |
1324 | fdput(in); | |
1325 | out: | |
1326 | return retval; | |
1327 | } | |
1328 | ||
1329 | SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count) | |
1330 | { | |
1331 | loff_t pos; | |
1332 | off_t off; | |
1333 | ssize_t ret; | |
1334 | ||
1335 | if (offset) { | |
1336 | if (unlikely(get_user(off, offset))) | |
1337 | return -EFAULT; | |
1338 | pos = off; | |
1339 | ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS); | |
1340 | if (unlikely(put_user(pos, offset))) | |
1341 | return -EFAULT; | |
1342 | return ret; | |
1343 | } | |
1344 | ||
1345 | return do_sendfile(out_fd, in_fd, NULL, count, 0); | |
1346 | } | |
1347 | ||
1348 | SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count) | |
1349 | { | |
1350 | loff_t pos; | |
1351 | ssize_t ret; | |
1352 | ||
1353 | if (offset) { | |
1354 | if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t)))) | |
1355 | return -EFAULT; | |
1356 | ret = do_sendfile(out_fd, in_fd, &pos, count, 0); | |
1357 | if (unlikely(put_user(pos, offset))) | |
1358 | return -EFAULT; | |
1359 | return ret; | |
1360 | } | |
1361 | ||
1362 | return do_sendfile(out_fd, in_fd, NULL, count, 0); | |
1363 | } | |
1364 | ||
1365 | #ifdef CONFIG_COMPAT | |
1366 | COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, | |
1367 | compat_off_t __user *, offset, compat_size_t, count) | |
1368 | { | |
1369 | loff_t pos; | |
1370 | off_t off; | |
1371 | ssize_t ret; | |
1372 | ||
1373 | if (offset) { | |
1374 | if (unlikely(get_user(off, offset))) | |
1375 | return -EFAULT; | |
1376 | pos = off; | |
1377 | ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS); | |
1378 | if (unlikely(put_user(pos, offset))) | |
1379 | return -EFAULT; | |
1380 | return ret; | |
1381 | } | |
1382 | ||
1383 | return do_sendfile(out_fd, in_fd, NULL, count, 0); | |
1384 | } | |
1385 | ||
1386 | COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, | |
1387 | compat_loff_t __user *, offset, compat_size_t, count) | |
1388 | { | |
1389 | loff_t pos; | |
1390 | ssize_t ret; | |
1391 | ||
1392 | if (offset) { | |
1393 | if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t)))) | |
1394 | return -EFAULT; | |
1395 | ret = do_sendfile(out_fd, in_fd, &pos, count, 0); | |
1396 | if (unlikely(put_user(pos, offset))) | |
1397 | return -EFAULT; | |
1398 | return ret; | |
1399 | } | |
1400 | ||
1401 | return do_sendfile(out_fd, in_fd, NULL, count, 0); | |
1402 | } | |
1403 | #endif |