]>
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/aio.h> | |
13 | #include <linux/fsnotify.h> | |
14 | #include <linux/security.h> | |
15 | #include <linux/export.h> | |
16 | #include <linux/syscalls.h> | |
17 | #include <linux/pagemap.h> | |
18 | #include <linux/splice.h> | |
19 | #include <linux/compat.h> | |
20 | #include "internal.h" | |
21 | ||
22 | #include <asm/uaccess.h> | |
23 | #include <asm/unistd.h> | |
24 | ||
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 | ||
29 | const struct file_operations generic_ro_fops = { | |
30 | .llseek = generic_file_llseek, | |
31 | .read = do_sync_read, | |
32 | .aio_read = generic_file_aio_read, | |
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 | 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 | ||
59 | /** | |
60 | * generic_file_llseek_size - generic llseek implementation for regular files | |
61 | * @file: file structure to seek on | |
62 | * @offset: file offset to seek to | |
63 | * @whence: type of seek | |
64 | * @size: max size of this file in file system | |
65 | * @eof: offset used for SEEK_END position | |
66 | * | |
67 | * This is a variant of generic_file_llseek that allows passing in a custom | |
68 | * maximum file size and a custom EOF position, for e.g. hashed directories | |
69 | * | |
70 | * Synchronization: | |
71 | * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms) | |
72 | * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes. | |
73 | * read/writes behave like SEEK_SET against seeks. | |
74 | */ | |
75 | loff_t | |
76 | generic_file_llseek_size(struct file *file, loff_t offset, int whence, | |
77 | loff_t maxsize, loff_t eof) | |
78 | { | |
79 | struct inode *inode = file->f_mapping->host; | |
80 | ||
81 | switch (whence) { | |
82 | case SEEK_END: | |
83 | offset += eof; | |
84 | break; | |
85 | case SEEK_CUR: | |
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; | |
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, | |
101 | maxsize); | |
102 | spin_unlock(&file->f_lock); | |
103 | return offset; | |
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 | */ | |
109 | if (offset >= eof) | |
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 | */ | |
117 | if (offset >= eof) | |
118 | return -ENXIO; | |
119 | offset = eof; | |
120 | break; | |
121 | } | |
122 | ||
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 | |
131 | * @whence: type of seek | |
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 | |
135 | * @offset and @whence. | |
136 | */ | |
137 | loff_t generic_file_llseek(struct file *file, loff_t offset, int whence) | |
138 | { | |
139 | struct inode *inode = file->f_mapping->host; | |
140 | ||
141 | return generic_file_llseek_size(file, offset, whence, | |
142 | inode->i_sb->s_maxbytes, | |
143 | i_size_read(inode)); | |
144 | } | |
145 | EXPORT_SYMBOL(generic_file_llseek); | |
146 | ||
147 | /** | |
148 | * noop_llseek - No Operation Performed llseek implementation | |
149 | * @file: file structure to seek on | |
150 | * @offset: file offset to seek to | |
151 | * @whence: type of seek | |
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 | */ | |
158 | loff_t noop_llseek(struct file *file, loff_t offset, int whence) | |
159 | { | |
160 | return file->f_pos; | |
161 | } | |
162 | EXPORT_SYMBOL(noop_llseek); | |
163 | ||
164 | loff_t no_llseek(struct file *file, loff_t offset, int whence) | |
165 | { | |
166 | return -ESPIPE; | |
167 | } | |
168 | EXPORT_SYMBOL(no_llseek); | |
169 | ||
170 | loff_t default_llseek(struct file *file, loff_t offset, int whence) | |
171 | { | |
172 | struct inode *inode = file_inode(file); | |
173 | loff_t retval; | |
174 | ||
175 | mutex_lock(&inode->i_mutex); | |
176 | switch (whence) { | |
177 | case SEEK_END: | |
178 | offset += i_size_read(inode); | |
179 | break; | |
180 | case SEEK_CUR: | |
181 | if (offset == 0) { | |
182 | retval = file->f_pos; | |
183 | goto out; | |
184 | } | |
185 | offset += file->f_pos; | |
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 | */ | |
193 | if (offset >= inode->i_size) { | |
194 | retval = -ENXIO; | |
195 | goto out; | |
196 | } | |
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 | */ | |
204 | if (offset >= inode->i_size) { | |
205 | retval = -ENXIO; | |
206 | goto out; | |
207 | } | |
208 | offset = inode->i_size; | |
209 | break; | |
210 | } | |
211 | retval = -EINVAL; | |
212 | if (offset >= 0 || unsigned_offsets(file)) { | |
213 | if (offset != file->f_pos) { | |
214 | file->f_pos = offset; | |
215 | file->f_version = 0; | |
216 | } | |
217 | retval = offset; | |
218 | } | |
219 | out: | |
220 | mutex_unlock(&inode->i_mutex); | |
221 | return retval; | |
222 | } | |
223 | EXPORT_SYMBOL(default_llseek); | |
224 | ||
225 | loff_t vfs_llseek(struct file *file, loff_t offset, int whence) | |
226 | { | |
227 | loff_t (*fn)(struct file *, loff_t, int); | |
228 | ||
229 | fn = no_llseek; | |
230 | if (file->f_mode & FMODE_LSEEK) { | |
231 | if (file->f_op && file->f_op->llseek) | |
232 | fn = file->f_op->llseek; | |
233 | } | |
234 | return fn(file, offset, whence); | |
235 | } | |
236 | EXPORT_SYMBOL(vfs_llseek); | |
237 | ||
238 | SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence) | |
239 | { | |
240 | off_t retval; | |
241 | struct fd f = fdget(fd); | |
242 | if (!f.file) | |
243 | return -EBADF; | |
244 | ||
245 | retval = -EINVAL; | |
246 | if (whence <= SEEK_MAX) { | |
247 | loff_t res = vfs_llseek(f.file, offset, whence); | |
248 | retval = res; | |
249 | if (res != (loff_t)retval) | |
250 | retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */ | |
251 | } | |
252 | fdput(f); | |
253 | return retval; | |
254 | } | |
255 | ||
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 | ||
263 | #ifdef __ARCH_WANT_SYS_LLSEEK | |
264 | SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high, | |
265 | unsigned long, offset_low, loff_t __user *, result, | |
266 | unsigned int, whence) | |
267 | { | |
268 | int retval; | |
269 | struct fd f = fdget(fd); | |
270 | loff_t offset; | |
271 | ||
272 | if (!f.file) | |
273 | return -EBADF; | |
274 | ||
275 | retval = -EINVAL; | |
276 | if (whence > SEEK_MAX) | |
277 | goto out_putf; | |
278 | ||
279 | offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low, | |
280 | whence); | |
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: | |
289 | fdput(f); | |
290 | return retval; | |
291 | } | |
292 | #endif | |
293 | ||
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 | */ | |
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; | |
303 | int retval = -EINVAL; | |
304 | ||
305 | inode = file_inode(file); | |
306 | if (unlikely((ssize_t) count < 0)) | |
307 | return retval; | |
308 | pos = *ppos; | |
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)) | |
316 | return retval; | |
317 | } | |
318 | ||
319 | if (unlikely(inode->i_flock && mandatory_lock(inode))) { | |
320 | retval = locks_mandatory_area( | |
321 | read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE, | |
322 | inode, file, pos, count); | |
323 | if (retval < 0) | |
324 | return retval; | |
325 | } | |
326 | retval = security_file_permission(file, | |
327 | read_write == READ ? MAY_READ : MAY_WRITE); | |
328 | if (retval) | |
329 | return retval; | |
330 | return count > MAX_RW_COUNT ? MAX_RW_COUNT : count; | |
331 | } | |
332 | ||
333 | ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos) | |
334 | { | |
335 | struct iovec iov = { .iov_base = buf, .iov_len = len }; | |
336 | struct kiocb kiocb; | |
337 | ssize_t ret; | |
338 | ||
339 | init_sync_kiocb(&kiocb, filp); | |
340 | kiocb.ki_pos = *ppos; | |
341 | kiocb.ki_left = len; | |
342 | kiocb.ki_nbytes = len; | |
343 | ||
344 | ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos); | |
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); | |
365 | if (ret >= 0) { | |
366 | count = ret; | |
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) { | |
372 | fsnotify_access(file); | |
373 | add_rchar(current, ret); | |
374 | } | |
375 | inc_syscr(current); | |
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 | { | |
385 | struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len }; | |
386 | struct kiocb kiocb; | |
387 | ssize_t ret; | |
388 | ||
389 | init_sync_kiocb(&kiocb, filp); | |
390 | kiocb.ki_pos = *ppos; | |
391 | kiocb.ki_left = len; | |
392 | kiocb.ki_nbytes = len; | |
393 | ||
394 | ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos); | |
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 | ||
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 | ||
409 | if (!file->f_op || (!file->f_op->write && !file->f_op->aio_write)) | |
410 | return -EINVAL; | |
411 | ||
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 | ||
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); | |
442 | if (ret >= 0) { | |
443 | count = ret; | |
444 | file_start_write(file); | |
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) { | |
450 | fsnotify_modify(file); | |
451 | add_wchar(current, ret); | |
452 | } | |
453 | inc_syscw(current); | |
454 | file_end_write(file); | |
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 | ||
472 | SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count) | |
473 | { | |
474 | struct fd f = fdget(fd); | |
475 | ssize_t ret = -EBADF; | |
476 | ||
477 | if (f.file) { | |
478 | loff_t pos = file_pos_read(f.file); | |
479 | ret = vfs_read(f.file, buf, count, &pos); | |
480 | file_pos_write(f.file, pos); | |
481 | fdput(f); | |
482 | } | |
483 | return ret; | |
484 | } | |
485 | ||
486 | SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf, | |
487 | size_t, count) | |
488 | { | |
489 | struct fd f = fdget(fd); | |
490 | ssize_t ret = -EBADF; | |
491 | ||
492 | if (f.file) { | |
493 | loff_t pos = file_pos_read(f.file); | |
494 | ret = vfs_write(f.file, buf, count, &pos); | |
495 | file_pos_write(f.file, pos); | |
496 | fdput(f); | |
497 | } | |
498 | ||
499 | return ret; | |
500 | } | |
501 | ||
502 | SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf, | |
503 | size_t, count, loff_t, pos) | |
504 | { | |
505 | struct fd f; | |
506 | ssize_t ret = -EBADF; | |
507 | ||
508 | if (pos < 0) | |
509 | return -EINVAL; | |
510 | ||
511 | f = fdget(fd); | |
512 | if (f.file) { | |
513 | ret = -ESPIPE; | |
514 | if (f.file->f_mode & FMODE_PREAD) | |
515 | ret = vfs_read(f.file, buf, count, &pos); | |
516 | fdput(f); | |
517 | } | |
518 | ||
519 | return ret; | |
520 | } | |
521 | ||
522 | SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf, | |
523 | size_t, count, loff_t, pos) | |
524 | { | |
525 | struct fd f; | |
526 | ssize_t ret = -EBADF; | |
527 | ||
528 | if (pos < 0) | |
529 | return -EINVAL; | |
530 | ||
531 | f = fdget(fd); | |
532 | if (f.file) { | |
533 | ret = -ESPIPE; | |
534 | if (f.file->f_mode & FMODE_PWRITE) | |
535 | ret = vfs_write(f.file, buf, count, &pos); | |
536 | fdput(f); | |
537 | } | |
538 | ||
539 | return ret; | |
540 | } | |
541 | ||
542 | /* | |
543 | * Reduce an iovec's length in-place. Return the resulting number of segments | |
544 | */ | |
545 | unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to) | |
546 | { | |
547 | unsigned long seg = 0; | |
548 | size_t len = 0; | |
549 | ||
550 | while (seg < nr_segs) { | |
551 | seg++; | |
552 | if (len + iov->iov_len >= to) { | |
553 | iov->iov_len = to - len; | |
554 | break; | |
555 | } | |
556 | len += iov->iov_len; | |
557 | iov++; | |
558 | } | |
559 | return seg; | |
560 | } | |
561 | EXPORT_SYMBOL(iov_shorten); | |
562 | ||
563 | static ssize_t do_sync_readv_writev(struct file *filp, const struct iovec *iov, | |
564 | unsigned long nr_segs, size_t len, loff_t *ppos, iov_fn_t fn) | |
565 | { | |
566 | struct kiocb kiocb; | |
567 | ssize_t ret; | |
568 | ||
569 | init_sync_kiocb(&kiocb, filp); | |
570 | kiocb.ki_pos = *ppos; | |
571 | kiocb.ki_left = len; | |
572 | kiocb.ki_nbytes = len; | |
573 | ||
574 | ret = fn(&kiocb, iov, nr_segs, kiocb.ki_pos); | |
575 | if (ret == -EIOCBQUEUED) | |
576 | ret = wait_on_sync_kiocb(&kiocb); | |
577 | *ppos = kiocb.ki_pos; | |
578 | return ret; | |
579 | } | |
580 | ||
581 | /* Do it by hand, with file-ops */ | |
582 | static ssize_t do_loop_readv_writev(struct file *filp, struct iovec *iov, | |
583 | unsigned long nr_segs, loff_t *ppos, io_fn_t fn) | |
584 | { | |
585 | struct iovec *vector = iov; | |
586 | ssize_t ret = 0; | |
587 | ||
588 | while (nr_segs > 0) { | |
589 | void __user *base; | |
590 | size_t len; | |
591 | ssize_t nr; | |
592 | ||
593 | base = vector->iov_base; | |
594 | len = vector->iov_len; | |
595 | vector++; | |
596 | nr_segs--; | |
597 | ||
598 | nr = fn(filp, base, len, ppos); | |
599 | ||
600 | if (nr < 0) { | |
601 | if (!ret) | |
602 | ret = nr; | |
603 | break; | |
604 | } | |
605 | ret += nr; | |
606 | if (nr != len) | |
607 | break; | |
608 | } | |
609 | ||
610 | return ret; | |
611 | } | |
612 | ||
613 | /* A write operation does a read from user space and vice versa */ | |
614 | #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ) | |
615 | ||
616 | ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector, | |
617 | unsigned long nr_segs, unsigned long fast_segs, | |
618 | struct iovec *fast_pointer, | |
619 | struct iovec **ret_pointer) | |
620 | { | |
621 | unsigned long seg; | |
622 | ssize_t ret; | |
623 | struct iovec *iov = fast_pointer; | |
624 | ||
625 | /* | |
626 | * SuS says "The readv() function *may* fail if the iovcnt argument | |
627 | * was less than or equal to 0, or greater than {IOV_MAX}. Linux has | |
628 | * traditionally returned zero for zero segments, so... | |
629 | */ | |
630 | if (nr_segs == 0) { | |
631 | ret = 0; | |
632 | goto out; | |
633 | } | |
634 | ||
635 | /* | |
636 | * First get the "struct iovec" from user memory and | |
637 | * verify all the pointers | |
638 | */ | |
639 | if (nr_segs > UIO_MAXIOV) { | |
640 | ret = -EINVAL; | |
641 | goto out; | |
642 | } | |
643 | if (nr_segs > fast_segs) { | |
644 | iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL); | |
645 | if (iov == NULL) { | |
646 | ret = -ENOMEM; | |
647 | goto out; | |
648 | } | |
649 | } | |
650 | if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) { | |
651 | ret = -EFAULT; | |
652 | goto out; | |
653 | } | |
654 | ||
655 | /* | |
656 | * According to the Single Unix Specification we should return EINVAL | |
657 | * if an element length is < 0 when cast to ssize_t or if the | |
658 | * total length would overflow the ssize_t return value of the | |
659 | * system call. | |
660 | * | |
661 | * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the | |
662 | * overflow case. | |
663 | */ | |
664 | ret = 0; | |
665 | for (seg = 0; seg < nr_segs; seg++) { | |
666 | void __user *buf = iov[seg].iov_base; | |
667 | ssize_t len = (ssize_t)iov[seg].iov_len; | |
668 | ||
669 | /* see if we we're about to use an invalid len or if | |
670 | * it's about to overflow ssize_t */ | |
671 | if (len < 0) { | |
672 | ret = -EINVAL; | |
673 | goto out; | |
674 | } | |
675 | if (type >= 0 | |
676 | && unlikely(!access_ok(vrfy_dir(type), buf, len))) { | |
677 | ret = -EFAULT; | |
678 | goto out; | |
679 | } | |
680 | if (len > MAX_RW_COUNT - ret) { | |
681 | len = MAX_RW_COUNT - ret; | |
682 | iov[seg].iov_len = len; | |
683 | } | |
684 | ret += len; | |
685 | } | |
686 | out: | |
687 | *ret_pointer = iov; | |
688 | return ret; | |
689 | } | |
690 | ||
691 | static ssize_t do_readv_writev(int type, struct file *file, | |
692 | const struct iovec __user * uvector, | |
693 | unsigned long nr_segs, loff_t *pos) | |
694 | { | |
695 | size_t tot_len; | |
696 | struct iovec iovstack[UIO_FASTIOV]; | |
697 | struct iovec *iov = iovstack; | |
698 | ssize_t ret; | |
699 | io_fn_t fn; | |
700 | iov_fn_t fnv; | |
701 | ||
702 | if (!file->f_op) { | |
703 | ret = -EINVAL; | |
704 | goto out; | |
705 | } | |
706 | ||
707 | ret = rw_copy_check_uvector(type, uvector, nr_segs, | |
708 | ARRAY_SIZE(iovstack), iovstack, &iov); | |
709 | if (ret <= 0) | |
710 | goto out; | |
711 | ||
712 | tot_len = ret; | |
713 | ret = rw_verify_area(type, file, pos, tot_len); | |
714 | if (ret < 0) | |
715 | goto out; | |
716 | ||
717 | fnv = NULL; | |
718 | if (type == READ) { | |
719 | fn = file->f_op->read; | |
720 | fnv = file->f_op->aio_read; | |
721 | } else { | |
722 | fn = (io_fn_t)file->f_op->write; | |
723 | fnv = file->f_op->aio_write; | |
724 | file_start_write(file); | |
725 | } | |
726 | ||
727 | if (fnv) | |
728 | ret = do_sync_readv_writev(file, iov, nr_segs, tot_len, | |
729 | pos, fnv); | |
730 | else | |
731 | ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn); | |
732 | ||
733 | if (type != READ) | |
734 | file_end_write(file); | |
735 | ||
736 | out: | |
737 | if (iov != iovstack) | |
738 | kfree(iov); | |
739 | if ((ret + (type == READ)) > 0) { | |
740 | if (type == READ) | |
741 | fsnotify_access(file); | |
742 | else | |
743 | fsnotify_modify(file); | |
744 | } | |
745 | return ret; | |
746 | } | |
747 | ||
748 | ssize_t vfs_readv(struct file *file, const struct iovec __user *vec, | |
749 | unsigned long vlen, loff_t *pos) | |
750 | { | |
751 | if (!(file->f_mode & FMODE_READ)) | |
752 | return -EBADF; | |
753 | if (!file->f_op || (!file->f_op->aio_read && !file->f_op->read)) | |
754 | return -EINVAL; | |
755 | ||
756 | return do_readv_writev(READ, file, vec, vlen, pos); | |
757 | } | |
758 | ||
759 | EXPORT_SYMBOL(vfs_readv); | |
760 | ||
761 | ssize_t vfs_writev(struct file *file, const struct iovec __user *vec, | |
762 | unsigned long vlen, loff_t *pos) | |
763 | { | |
764 | if (!(file->f_mode & FMODE_WRITE)) | |
765 | return -EBADF; | |
766 | if (!file->f_op || (!file->f_op->aio_write && !file->f_op->write)) | |
767 | return -EINVAL; | |
768 | ||
769 | return do_readv_writev(WRITE, file, vec, vlen, pos); | |
770 | } | |
771 | ||
772 | EXPORT_SYMBOL(vfs_writev); | |
773 | ||
774 | SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec, | |
775 | unsigned long, vlen) | |
776 | { | |
777 | struct fd f = fdget(fd); | |
778 | ssize_t ret = -EBADF; | |
779 | ||
780 | if (f.file) { | |
781 | loff_t pos = file_pos_read(f.file); | |
782 | ret = vfs_readv(f.file, vec, vlen, &pos); | |
783 | file_pos_write(f.file, pos); | |
784 | fdput(f); | |
785 | } | |
786 | ||
787 | if (ret > 0) | |
788 | add_rchar(current, ret); | |
789 | inc_syscr(current); | |
790 | return ret; | |
791 | } | |
792 | ||
793 | SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec, | |
794 | unsigned long, vlen) | |
795 | { | |
796 | struct fd f = fdget(fd); | |
797 | ssize_t ret = -EBADF; | |
798 | ||
799 | if (f.file) { | |
800 | loff_t pos = file_pos_read(f.file); | |
801 | ret = vfs_writev(f.file, vec, vlen, &pos); | |
802 | file_pos_write(f.file, pos); | |
803 | fdput(f); | |
804 | } | |
805 | ||
806 | if (ret > 0) | |
807 | add_wchar(current, ret); | |
808 | inc_syscw(current); | |
809 | return ret; | |
810 | } | |
811 | ||
812 | static inline loff_t pos_from_hilo(unsigned long high, unsigned long low) | |
813 | { | |
814 | #define HALF_LONG_BITS (BITS_PER_LONG / 2) | |
815 | return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low; | |
816 | } | |
817 | ||
818 | SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec, | |
819 | unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h) | |
820 | { | |
821 | loff_t pos = pos_from_hilo(pos_h, pos_l); | |
822 | struct fd f; | |
823 | ssize_t ret = -EBADF; | |
824 | ||
825 | if (pos < 0) | |
826 | return -EINVAL; | |
827 | ||
828 | f = fdget(fd); | |
829 | if (f.file) { | |
830 | ret = -ESPIPE; | |
831 | if (f.file->f_mode & FMODE_PREAD) | |
832 | ret = vfs_readv(f.file, vec, vlen, &pos); | |
833 | fdput(f); | |
834 | } | |
835 | ||
836 | if (ret > 0) | |
837 | add_rchar(current, ret); | |
838 | inc_syscr(current); | |
839 | return ret; | |
840 | } | |
841 | ||
842 | SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec, | |
843 | unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h) | |
844 | { | |
845 | loff_t pos = pos_from_hilo(pos_h, pos_l); | |
846 | struct fd f; | |
847 | ssize_t ret = -EBADF; | |
848 | ||
849 | if (pos < 0) | |
850 | return -EINVAL; | |
851 | ||
852 | f = fdget(fd); | |
853 | if (f.file) { | |
854 | ret = -ESPIPE; | |
855 | if (f.file->f_mode & FMODE_PWRITE) | |
856 | ret = vfs_writev(f.file, vec, vlen, &pos); | |
857 | fdput(f); | |
858 | } | |
859 | ||
860 | if (ret > 0) | |
861 | add_wchar(current, ret); | |
862 | inc_syscw(current); | |
863 | return ret; | |
864 | } | |
865 | ||
866 | #ifdef CONFIG_COMPAT | |
867 | ||
868 | static ssize_t compat_do_readv_writev(int type, struct file *file, | |
869 | const struct compat_iovec __user *uvector, | |
870 | unsigned long nr_segs, loff_t *pos) | |
871 | { | |
872 | compat_ssize_t tot_len; | |
873 | struct iovec iovstack[UIO_FASTIOV]; | |
874 | struct iovec *iov = iovstack; | |
875 | ssize_t ret; | |
876 | io_fn_t fn; | |
877 | iov_fn_t fnv; | |
878 | ||
879 | ret = -EINVAL; | |
880 | if (!file->f_op) | |
881 | goto out; | |
882 | ||
883 | ret = -EFAULT; | |
884 | if (!access_ok(VERIFY_READ, uvector, nr_segs*sizeof(*uvector))) | |
885 | goto out; | |
886 | ||
887 | ret = compat_rw_copy_check_uvector(type, uvector, nr_segs, | |
888 | UIO_FASTIOV, iovstack, &iov); | |
889 | if (ret <= 0) | |
890 | goto out; | |
891 | ||
892 | tot_len = ret; | |
893 | ret = rw_verify_area(type, file, pos, tot_len); | |
894 | if (ret < 0) | |
895 | goto out; | |
896 | ||
897 | fnv = NULL; | |
898 | if (type == READ) { | |
899 | fn = file->f_op->read; | |
900 | fnv = file->f_op->aio_read; | |
901 | } else { | |
902 | fn = (io_fn_t)file->f_op->write; | |
903 | fnv = file->f_op->aio_write; | |
904 | file_start_write(file); | |
905 | } | |
906 | ||
907 | if (fnv) | |
908 | ret = do_sync_readv_writev(file, iov, nr_segs, tot_len, | |
909 | pos, fnv); | |
910 | else | |
911 | ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn); | |
912 | ||
913 | if (type != READ) | |
914 | file_end_write(file); | |
915 | ||
916 | out: | |
917 | if (iov != iovstack) | |
918 | kfree(iov); | |
919 | if ((ret + (type == READ)) > 0) { | |
920 | if (type == READ) | |
921 | fsnotify_access(file); | |
922 | else | |
923 | fsnotify_modify(file); | |
924 | } | |
925 | return ret; | |
926 | } | |
927 | ||
928 | static size_t compat_readv(struct file *file, | |
929 | const struct compat_iovec __user *vec, | |
930 | unsigned long vlen, loff_t *pos) | |
931 | { | |
932 | ssize_t ret = -EBADF; | |
933 | ||
934 | if (!(file->f_mode & FMODE_READ)) | |
935 | goto out; | |
936 | ||
937 | ret = -EINVAL; | |
938 | if (!file->f_op || (!file->f_op->aio_read && !file->f_op->read)) | |
939 | goto out; | |
940 | ||
941 | ret = compat_do_readv_writev(READ, file, vec, vlen, pos); | |
942 | ||
943 | out: | |
944 | if (ret > 0) | |
945 | add_rchar(current, ret); | |
946 | inc_syscr(current); | |
947 | return ret; | |
948 | } | |
949 | ||
950 | COMPAT_SYSCALL_DEFINE3(readv, unsigned long, fd, | |
951 | const struct compat_iovec __user *,vec, | |
952 | unsigned long, vlen) | |
953 | { | |
954 | struct fd f = fdget(fd); | |
955 | ssize_t ret; | |
956 | loff_t pos; | |
957 | ||
958 | if (!f.file) | |
959 | return -EBADF; | |
960 | pos = f.file->f_pos; | |
961 | ret = compat_readv(f.file, vec, vlen, &pos); | |
962 | f.file->f_pos = pos; | |
963 | fdput(f); | |
964 | return ret; | |
965 | } | |
966 | ||
967 | COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd, | |
968 | const struct compat_iovec __user *,vec, | |
969 | unsigned long, vlen, loff_t, pos) | |
970 | { | |
971 | struct fd f; | |
972 | ssize_t ret; | |
973 | ||
974 | if (pos < 0) | |
975 | return -EINVAL; | |
976 | f = fdget(fd); | |
977 | if (!f.file) | |
978 | return -EBADF; | |
979 | ret = -ESPIPE; | |
980 | if (f.file->f_mode & FMODE_PREAD) | |
981 | ret = compat_readv(f.file, vec, vlen, &pos); | |
982 | fdput(f); | |
983 | return ret; | |
984 | } | |
985 | ||
986 | COMPAT_SYSCALL_DEFINE5(preadv, unsigned long, fd, | |
987 | const struct compat_iovec __user *,vec, | |
988 | unsigned long, vlen, u32, pos_low, u32, pos_high) | |
989 | { | |
990 | loff_t pos = ((loff_t)pos_high << 32) | pos_low; | |
991 | return compat_sys_preadv64(fd, vec, vlen, pos); | |
992 | } | |
993 | ||
994 | static size_t compat_writev(struct file *file, | |
995 | const struct compat_iovec __user *vec, | |
996 | unsigned long vlen, loff_t *pos) | |
997 | { | |
998 | ssize_t ret = -EBADF; | |
999 | ||
1000 | if (!(file->f_mode & FMODE_WRITE)) | |
1001 | goto out; | |
1002 | ||
1003 | ret = -EINVAL; | |
1004 | if (!file->f_op || (!file->f_op->aio_write && !file->f_op->write)) | |
1005 | goto out; | |
1006 | ||
1007 | ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos); | |
1008 | ||
1009 | out: | |
1010 | if (ret > 0) | |
1011 | add_wchar(current, ret); | |
1012 | inc_syscw(current); | |
1013 | return ret; | |
1014 | } | |
1015 | ||
1016 | COMPAT_SYSCALL_DEFINE3(writev, unsigned long, fd, | |
1017 | const struct compat_iovec __user *, vec, | |
1018 | unsigned long, vlen) | |
1019 | { | |
1020 | struct fd f = fdget(fd); | |
1021 | ssize_t ret; | |
1022 | loff_t pos; | |
1023 | ||
1024 | if (!f.file) | |
1025 | return -EBADF; | |
1026 | pos = f.file->f_pos; | |
1027 | ret = compat_writev(f.file, vec, vlen, &pos); | |
1028 | f.file->f_pos = pos; | |
1029 | fdput(f); | |
1030 | return ret; | |
1031 | } | |
1032 | ||
1033 | COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd, | |
1034 | const struct compat_iovec __user *,vec, | |
1035 | unsigned long, vlen, loff_t, pos) | |
1036 | { | |
1037 | struct fd f; | |
1038 | ssize_t ret; | |
1039 | ||
1040 | if (pos < 0) | |
1041 | return -EINVAL; | |
1042 | f = fdget(fd); | |
1043 | if (!f.file) | |
1044 | return -EBADF; | |
1045 | ret = -ESPIPE; | |
1046 | if (f.file->f_mode & FMODE_PWRITE) | |
1047 | ret = compat_writev(f.file, vec, vlen, &pos); | |
1048 | fdput(f); | |
1049 | return ret; | |
1050 | } | |
1051 | ||
1052 | COMPAT_SYSCALL_DEFINE5(pwritev, unsigned long, fd, | |
1053 | const struct compat_iovec __user *,vec, | |
1054 | unsigned long, vlen, u32, pos_low, u32, pos_high) | |
1055 | { | |
1056 | loff_t pos = ((loff_t)pos_high << 32) | pos_low; | |
1057 | return compat_sys_pwritev64(fd, vec, vlen, pos); | |
1058 | } | |
1059 | #endif | |
1060 | ||
1061 | static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos, | |
1062 | size_t count, loff_t max) | |
1063 | { | |
1064 | struct fd in, out; | |
1065 | struct inode *in_inode, *out_inode; | |
1066 | loff_t pos; | |
1067 | loff_t out_pos; | |
1068 | ssize_t retval; | |
1069 | int fl; | |
1070 | ||
1071 | /* | |
1072 | * Get input file, and verify that it is ok.. | |
1073 | */ | |
1074 | retval = -EBADF; | |
1075 | in = fdget(in_fd); | |
1076 | if (!in.file) | |
1077 | goto out; | |
1078 | if (!(in.file->f_mode & FMODE_READ)) | |
1079 | goto fput_in; | |
1080 | retval = -ESPIPE; | |
1081 | if (!ppos) { | |
1082 | pos = in.file->f_pos; | |
1083 | } else { | |
1084 | pos = *ppos; | |
1085 | if (!(in.file->f_mode & FMODE_PREAD)) | |
1086 | goto fput_in; | |
1087 | } | |
1088 | retval = rw_verify_area(READ, in.file, &pos, count); | |
1089 | if (retval < 0) | |
1090 | goto fput_in; | |
1091 | count = retval; | |
1092 | ||
1093 | /* | |
1094 | * Get output file, and verify that it is ok.. | |
1095 | */ | |
1096 | retval = -EBADF; | |
1097 | out = fdget(out_fd); | |
1098 | if (!out.file) | |
1099 | goto fput_in; | |
1100 | if (!(out.file->f_mode & FMODE_WRITE)) | |
1101 | goto fput_out; | |
1102 | retval = -EINVAL; | |
1103 | in_inode = file_inode(in.file); | |
1104 | out_inode = file_inode(out.file); | |
1105 | out_pos = out.file->f_pos; | |
1106 | retval = rw_verify_area(WRITE, out.file, &out_pos, count); | |
1107 | if (retval < 0) | |
1108 | goto fput_out; | |
1109 | count = retval; | |
1110 | ||
1111 | if (!max) | |
1112 | max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes); | |
1113 | ||
1114 | if (unlikely(pos + count > max)) { | |
1115 | retval = -EOVERFLOW; | |
1116 | if (pos >= max) | |
1117 | goto fput_out; | |
1118 | count = max - pos; | |
1119 | } | |
1120 | ||
1121 | fl = 0; | |
1122 | #if 0 | |
1123 | /* | |
1124 | * We need to debate whether we can enable this or not. The | |
1125 | * man page documents EAGAIN return for the output at least, | |
1126 | * and the application is arguably buggy if it doesn't expect | |
1127 | * EAGAIN on a non-blocking file descriptor. | |
1128 | */ | |
1129 | if (in.file->f_flags & O_NONBLOCK) | |
1130 | fl = SPLICE_F_NONBLOCK; | |
1131 | #endif | |
1132 | file_start_write(out.file); | |
1133 | retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl); | |
1134 | file_end_write(out.file); | |
1135 | ||
1136 | if (retval > 0) { | |
1137 | add_rchar(current, retval); | |
1138 | add_wchar(current, retval); | |
1139 | fsnotify_access(in.file); | |
1140 | fsnotify_modify(out.file); | |
1141 | out.file->f_pos = out_pos; | |
1142 | if (ppos) | |
1143 | *ppos = pos; | |
1144 | else | |
1145 | in.file->f_pos = pos; | |
1146 | } | |
1147 | ||
1148 | inc_syscr(current); | |
1149 | inc_syscw(current); | |
1150 | if (pos > max) | |
1151 | retval = -EOVERFLOW; | |
1152 | ||
1153 | fput_out: | |
1154 | fdput(out); | |
1155 | fput_in: | |
1156 | fdput(in); | |
1157 | out: | |
1158 | return retval; | |
1159 | } | |
1160 | ||
1161 | SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count) | |
1162 | { | |
1163 | loff_t pos; | |
1164 | off_t off; | |
1165 | ssize_t ret; | |
1166 | ||
1167 | if (offset) { | |
1168 | if (unlikely(get_user(off, offset))) | |
1169 | return -EFAULT; | |
1170 | pos = off; | |
1171 | ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS); | |
1172 | if (unlikely(put_user(pos, offset))) | |
1173 | return -EFAULT; | |
1174 | return ret; | |
1175 | } | |
1176 | ||
1177 | return do_sendfile(out_fd, in_fd, NULL, count, 0); | |
1178 | } | |
1179 | ||
1180 | SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count) | |
1181 | { | |
1182 | loff_t pos; | |
1183 | ssize_t ret; | |
1184 | ||
1185 | if (offset) { | |
1186 | if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t)))) | |
1187 | return -EFAULT; | |
1188 | ret = do_sendfile(out_fd, in_fd, &pos, count, 0); | |
1189 | if (unlikely(put_user(pos, offset))) | |
1190 | return -EFAULT; | |
1191 | return ret; | |
1192 | } | |
1193 | ||
1194 | return do_sendfile(out_fd, in_fd, NULL, count, 0); | |
1195 | } | |
1196 | ||
1197 | #ifdef CONFIG_COMPAT | |
1198 | COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, | |
1199 | compat_off_t __user *, offset, compat_size_t, count) | |
1200 | { | |
1201 | loff_t pos; | |
1202 | off_t off; | |
1203 | ssize_t ret; | |
1204 | ||
1205 | if (offset) { | |
1206 | if (unlikely(get_user(off, offset))) | |
1207 | return -EFAULT; | |
1208 | pos = off; | |
1209 | ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS); | |
1210 | if (unlikely(put_user(pos, offset))) | |
1211 | return -EFAULT; | |
1212 | return ret; | |
1213 | } | |
1214 | ||
1215 | return do_sendfile(out_fd, in_fd, NULL, count, 0); | |
1216 | } | |
1217 | ||
1218 | COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, | |
1219 | compat_loff_t __user *, offset, compat_size_t, count) | |
1220 | { | |
1221 | loff_t pos; | |
1222 | ssize_t ret; | |
1223 | ||
1224 | if (offset) { | |
1225 | if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t)))) | |
1226 | return -EFAULT; | |
1227 | ret = do_sendfile(out_fd, in_fd, &pos, count, 0); | |
1228 | if (unlikely(put_user(pos, offset))) | |
1229 | return -EFAULT; | |
1230 | return ret; | |
1231 | } | |
1232 | ||
1233 | return do_sendfile(out_fd, in_fd, NULL, count, 0); | |
1234 | } | |
1235 | #endif |