]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * linux/fs/open.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992 Linus Torvalds | |
5 | */ | |
6 | ||
7 | #include <linux/string.h> | |
8 | #include <linux/mm.h> | |
9 | #include <linux/file.h> | |
10 | #include <linux/fdtable.h> | |
11 | #include <linux/fsnotify.h> | |
12 | #include <linux/module.h> | |
13 | #include <linux/tty.h> | |
14 | #include <linux/namei.h> | |
15 | #include <linux/backing-dev.h> | |
16 | #include <linux/capability.h> | |
17 | #include <linux/securebits.h> | |
18 | #include <linux/security.h> | |
19 | #include <linux/mount.h> | |
20 | #include <linux/fcntl.h> | |
21 | #include <linux/slab.h> | |
22 | #include <asm/uaccess.h> | |
23 | #include <linux/fs.h> | |
24 | #include <linux/personality.h> | |
25 | #include <linux/pagemap.h> | |
26 | #include <linux/syscalls.h> | |
27 | #include <linux/rcupdate.h> | |
28 | #include <linux/audit.h> | |
29 | #include <linux/falloc.h> | |
30 | #include <linux/fs_struct.h> | |
31 | #include <linux/ima.h> | |
32 | #include <linux/dnotify.h> | |
33 | #include <linux/compat.h> | |
34 | ||
35 | #include "internal.h" | |
36 | ||
37 | int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs, | |
38 | struct file *filp) | |
39 | { | |
40 | int ret; | |
41 | struct iattr newattrs; | |
42 | ||
43 | /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */ | |
44 | if (length < 0) | |
45 | return -EINVAL; | |
46 | ||
47 | newattrs.ia_size = length; | |
48 | newattrs.ia_valid = ATTR_SIZE | time_attrs; | |
49 | if (filp) { | |
50 | newattrs.ia_file = filp; | |
51 | newattrs.ia_valid |= ATTR_FILE; | |
52 | } | |
53 | ||
54 | /* Remove suid/sgid on truncate too */ | |
55 | ret = should_remove_suid(dentry); | |
56 | if (ret) | |
57 | newattrs.ia_valid |= ret | ATTR_FORCE; | |
58 | ||
59 | mutex_lock(&dentry->d_inode->i_mutex); | |
60 | /* Note any delegations or leases have already been broken: */ | |
61 | ret = notify_change(dentry, &newattrs, NULL); | |
62 | mutex_unlock(&dentry->d_inode->i_mutex); | |
63 | return ret; | |
64 | } | |
65 | ||
66 | long vfs_truncate(struct path *path, loff_t length) | |
67 | { | |
68 | struct inode *inode; | |
69 | long error; | |
70 | ||
71 | inode = path->dentry->d_inode; | |
72 | ||
73 | /* For directories it's -EISDIR, for other non-regulars - -EINVAL */ | |
74 | if (S_ISDIR(inode->i_mode)) | |
75 | return -EISDIR; | |
76 | if (!S_ISREG(inode->i_mode)) | |
77 | return -EINVAL; | |
78 | ||
79 | error = mnt_want_write(path->mnt); | |
80 | if (error) | |
81 | goto out; | |
82 | ||
83 | error = inode_permission(inode, MAY_WRITE); | |
84 | if (error) | |
85 | goto mnt_drop_write_and_out; | |
86 | ||
87 | error = -EPERM; | |
88 | if (IS_APPEND(inode)) | |
89 | goto mnt_drop_write_and_out; | |
90 | ||
91 | error = get_write_access(inode); | |
92 | if (error) | |
93 | goto mnt_drop_write_and_out; | |
94 | ||
95 | /* | |
96 | * Make sure that there are no leases. get_write_access() protects | |
97 | * against the truncate racing with a lease-granting setlease(). | |
98 | */ | |
99 | error = break_lease(inode, O_WRONLY); | |
100 | if (error) | |
101 | goto put_write_and_out; | |
102 | ||
103 | error = locks_verify_truncate(inode, NULL, length); | |
104 | if (!error) | |
105 | error = security_path_truncate(path); | |
106 | if (!error) | |
107 | error = do_truncate(path->dentry, length, 0, NULL); | |
108 | ||
109 | put_write_and_out: | |
110 | put_write_access(inode); | |
111 | mnt_drop_write_and_out: | |
112 | mnt_drop_write(path->mnt); | |
113 | out: | |
114 | return error; | |
115 | } | |
116 | EXPORT_SYMBOL_GPL(vfs_truncate); | |
117 | ||
118 | static long do_sys_truncate(const char __user *pathname, loff_t length) | |
119 | { | |
120 | unsigned int lookup_flags = LOOKUP_FOLLOW; | |
121 | struct path path; | |
122 | int error; | |
123 | ||
124 | if (length < 0) /* sorry, but loff_t says... */ | |
125 | return -EINVAL; | |
126 | ||
127 | retry: | |
128 | error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path); | |
129 | if (!error) { | |
130 | error = vfs_truncate(&path, length); | |
131 | path_put(&path); | |
132 | } | |
133 | if (retry_estale(error, lookup_flags)) { | |
134 | lookup_flags |= LOOKUP_REVAL; | |
135 | goto retry; | |
136 | } | |
137 | return error; | |
138 | } | |
139 | ||
140 | SYSCALL_DEFINE2(truncate, const char __user *, path, long, length) | |
141 | { | |
142 | return do_sys_truncate(path, length); | |
143 | } | |
144 | ||
145 | #ifdef CONFIG_COMPAT | |
146 | COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length) | |
147 | { | |
148 | return do_sys_truncate(path, length); | |
149 | } | |
150 | #endif | |
151 | ||
152 | static long do_sys_ftruncate(unsigned int fd, loff_t length, int small) | |
153 | { | |
154 | struct inode *inode; | |
155 | struct dentry *dentry; | |
156 | struct fd f; | |
157 | int error; | |
158 | ||
159 | error = -EINVAL; | |
160 | if (length < 0) | |
161 | goto out; | |
162 | error = -EBADF; | |
163 | f = fdget(fd); | |
164 | if (!f.file) | |
165 | goto out; | |
166 | ||
167 | /* explicitly opened as large or we are on 64-bit box */ | |
168 | if (f.file->f_flags & O_LARGEFILE) | |
169 | small = 0; | |
170 | ||
171 | dentry = f.file->f_path.dentry; | |
172 | inode = dentry->d_inode; | |
173 | error = -EINVAL; | |
174 | if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE)) | |
175 | goto out_putf; | |
176 | ||
177 | error = -EINVAL; | |
178 | /* Cannot ftruncate over 2^31 bytes without large file support */ | |
179 | if (small && length > MAX_NON_LFS) | |
180 | goto out_putf; | |
181 | ||
182 | error = -EPERM; | |
183 | if (IS_APPEND(inode)) | |
184 | goto out_putf; | |
185 | ||
186 | sb_start_write(inode->i_sb); | |
187 | error = locks_verify_truncate(inode, f.file, length); | |
188 | if (!error) | |
189 | error = security_path_truncate(&f.file->f_path); | |
190 | if (!error) | |
191 | error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file); | |
192 | sb_end_write(inode->i_sb); | |
193 | out_putf: | |
194 | fdput(f); | |
195 | out: | |
196 | return error; | |
197 | } | |
198 | ||
199 | SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length) | |
200 | { | |
201 | return do_sys_ftruncate(fd, length, 1); | |
202 | } | |
203 | ||
204 | #ifdef CONFIG_COMPAT | |
205 | COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length) | |
206 | { | |
207 | return do_sys_ftruncate(fd, length, 1); | |
208 | } | |
209 | #endif | |
210 | ||
211 | /* LFS versions of truncate are only needed on 32 bit machines */ | |
212 | #if BITS_PER_LONG == 32 | |
213 | SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length) | |
214 | { | |
215 | return do_sys_truncate(path, length); | |
216 | } | |
217 | ||
218 | SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length) | |
219 | { | |
220 | return do_sys_ftruncate(fd, length, 0); | |
221 | } | |
222 | #endif /* BITS_PER_LONG == 32 */ | |
223 | ||
224 | ||
225 | int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len) | |
226 | { | |
227 | struct inode *inode = file_inode(file); | |
228 | long ret; | |
229 | ||
230 | if (offset < 0 || len <= 0) | |
231 | return -EINVAL; | |
232 | ||
233 | /* Return error if mode is not supported */ | |
234 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | | |
235 | FALLOC_FL_COLLAPSE_RANGE)) | |
236 | return -EOPNOTSUPP; | |
237 | ||
238 | /* Punch hole must have keep size set */ | |
239 | if ((mode & FALLOC_FL_PUNCH_HOLE) && | |
240 | !(mode & FALLOC_FL_KEEP_SIZE)) | |
241 | return -EOPNOTSUPP; | |
242 | ||
243 | /* Collapse range should only be used exclusively. */ | |
244 | if ((mode & FALLOC_FL_COLLAPSE_RANGE) && | |
245 | (mode & ~FALLOC_FL_COLLAPSE_RANGE)) | |
246 | return -EINVAL; | |
247 | ||
248 | if (!(file->f_mode & FMODE_WRITE)) | |
249 | return -EBADF; | |
250 | ||
251 | /* | |
252 | * It's not possible to punch hole or perform collapse range | |
253 | * on append only file | |
254 | */ | |
255 | if (mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_COLLAPSE_RANGE) | |
256 | && IS_APPEND(inode)) | |
257 | return -EPERM; | |
258 | ||
259 | if (IS_IMMUTABLE(inode)) | |
260 | return -EPERM; | |
261 | ||
262 | /* | |
263 | * Revalidate the write permissions, in case security policy has | |
264 | * changed since the files were opened. | |
265 | */ | |
266 | ret = security_file_permission(file, MAY_WRITE); | |
267 | if (ret) | |
268 | return ret; | |
269 | ||
270 | if (S_ISFIFO(inode->i_mode)) | |
271 | return -ESPIPE; | |
272 | ||
273 | /* | |
274 | * Let individual file system decide if it supports preallocation | |
275 | * for directories or not. | |
276 | */ | |
277 | if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode)) | |
278 | return -ENODEV; | |
279 | ||
280 | /* Check for wrap through zero too */ | |
281 | if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0)) | |
282 | return -EFBIG; | |
283 | ||
284 | /* | |
285 | * There is no need to overlap collapse range with EOF, in which case | |
286 | * it is effectively a truncate operation | |
287 | */ | |
288 | if ((mode & FALLOC_FL_COLLAPSE_RANGE) && | |
289 | (offset + len >= i_size_read(inode))) | |
290 | return -EINVAL; | |
291 | ||
292 | if (!file->f_op->fallocate) | |
293 | return -EOPNOTSUPP; | |
294 | ||
295 | sb_start_write(inode->i_sb); | |
296 | ret = file->f_op->fallocate(file, mode, offset, len); | |
297 | sb_end_write(inode->i_sb); | |
298 | return ret; | |
299 | } | |
300 | ||
301 | SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len) | |
302 | { | |
303 | struct fd f = fdget(fd); | |
304 | int error = -EBADF; | |
305 | ||
306 | if (f.file) { | |
307 | error = do_fallocate(f.file, mode, offset, len); | |
308 | fdput(f); | |
309 | } | |
310 | return error; | |
311 | } | |
312 | ||
313 | /* | |
314 | * access() needs to use the real uid/gid, not the effective uid/gid. | |
315 | * We do this by temporarily clearing all FS-related capabilities and | |
316 | * switching the fsuid/fsgid around to the real ones. | |
317 | */ | |
318 | SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode) | |
319 | { | |
320 | const struct cred *old_cred; | |
321 | struct cred *override_cred; | |
322 | struct path path; | |
323 | struct inode *inode; | |
324 | int res; | |
325 | unsigned int lookup_flags = LOOKUP_FOLLOW; | |
326 | ||
327 | if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */ | |
328 | return -EINVAL; | |
329 | ||
330 | override_cred = prepare_creds(); | |
331 | if (!override_cred) | |
332 | return -ENOMEM; | |
333 | ||
334 | override_cred->fsuid = override_cred->uid; | |
335 | override_cred->fsgid = override_cred->gid; | |
336 | ||
337 | if (!issecure(SECURE_NO_SETUID_FIXUP)) { | |
338 | /* Clear the capabilities if we switch to a non-root user */ | |
339 | kuid_t root_uid = make_kuid(override_cred->user_ns, 0); | |
340 | if (!uid_eq(override_cred->uid, root_uid)) | |
341 | cap_clear(override_cred->cap_effective); | |
342 | else | |
343 | override_cred->cap_effective = | |
344 | override_cred->cap_permitted; | |
345 | } | |
346 | ||
347 | old_cred = override_creds(override_cred); | |
348 | retry: | |
349 | res = user_path_at(dfd, filename, lookup_flags, &path); | |
350 | if (res) | |
351 | goto out; | |
352 | ||
353 | inode = path.dentry->d_inode; | |
354 | ||
355 | if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) { | |
356 | /* | |
357 | * MAY_EXEC on regular files is denied if the fs is mounted | |
358 | * with the "noexec" flag. | |
359 | */ | |
360 | res = -EACCES; | |
361 | if (path.mnt->mnt_flags & MNT_NOEXEC) | |
362 | goto out_path_release; | |
363 | } | |
364 | ||
365 | res = inode_permission(inode, mode | MAY_ACCESS); | |
366 | /* SuS v2 requires we report a read only fs too */ | |
367 | if (res || !(mode & S_IWOTH) || special_file(inode->i_mode)) | |
368 | goto out_path_release; | |
369 | /* | |
370 | * This is a rare case where using __mnt_is_readonly() | |
371 | * is OK without a mnt_want/drop_write() pair. Since | |
372 | * no actual write to the fs is performed here, we do | |
373 | * not need to telegraph to that to anyone. | |
374 | * | |
375 | * By doing this, we accept that this access is | |
376 | * inherently racy and know that the fs may change | |
377 | * state before we even see this result. | |
378 | */ | |
379 | if (__mnt_is_readonly(path.mnt)) | |
380 | res = -EROFS; | |
381 | ||
382 | out_path_release: | |
383 | path_put(&path); | |
384 | if (retry_estale(res, lookup_flags)) { | |
385 | lookup_flags |= LOOKUP_REVAL; | |
386 | goto retry; | |
387 | } | |
388 | out: | |
389 | revert_creds(old_cred); | |
390 | put_cred(override_cred); | |
391 | return res; | |
392 | } | |
393 | ||
394 | SYSCALL_DEFINE2(access, const char __user *, filename, int, mode) | |
395 | { | |
396 | return sys_faccessat(AT_FDCWD, filename, mode); | |
397 | } | |
398 | ||
399 | SYSCALL_DEFINE1(chdir, const char __user *, filename) | |
400 | { | |
401 | struct path path; | |
402 | int error; | |
403 | unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
404 | retry: | |
405 | error = user_path_at(AT_FDCWD, filename, lookup_flags, &path); | |
406 | if (error) | |
407 | goto out; | |
408 | ||
409 | error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR); | |
410 | if (error) | |
411 | goto dput_and_out; | |
412 | ||
413 | set_fs_pwd(current->fs, &path); | |
414 | ||
415 | dput_and_out: | |
416 | path_put(&path); | |
417 | if (retry_estale(error, lookup_flags)) { | |
418 | lookup_flags |= LOOKUP_REVAL; | |
419 | goto retry; | |
420 | } | |
421 | out: | |
422 | return error; | |
423 | } | |
424 | ||
425 | SYSCALL_DEFINE1(fchdir, unsigned int, fd) | |
426 | { | |
427 | struct fd f = fdget_raw(fd); | |
428 | struct inode *inode; | |
429 | int error = -EBADF; | |
430 | ||
431 | error = -EBADF; | |
432 | if (!f.file) | |
433 | goto out; | |
434 | ||
435 | inode = file_inode(f.file); | |
436 | ||
437 | error = -ENOTDIR; | |
438 | if (!S_ISDIR(inode->i_mode)) | |
439 | goto out_putf; | |
440 | ||
441 | error = inode_permission(inode, MAY_EXEC | MAY_CHDIR); | |
442 | if (!error) | |
443 | set_fs_pwd(current->fs, &f.file->f_path); | |
444 | out_putf: | |
445 | fdput(f); | |
446 | out: | |
447 | return error; | |
448 | } | |
449 | ||
450 | SYSCALL_DEFINE1(chroot, const char __user *, filename) | |
451 | { | |
452 | struct path path; | |
453 | int error; | |
454 | unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
455 | retry: | |
456 | error = user_path_at(AT_FDCWD, filename, lookup_flags, &path); | |
457 | if (error) | |
458 | goto out; | |
459 | ||
460 | error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR); | |
461 | if (error) | |
462 | goto dput_and_out; | |
463 | ||
464 | error = -EPERM; | |
465 | if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT)) | |
466 | goto dput_and_out; | |
467 | error = security_path_chroot(&path); | |
468 | if (error) | |
469 | goto dput_and_out; | |
470 | ||
471 | set_fs_root(current->fs, &path); | |
472 | error = 0; | |
473 | dput_and_out: | |
474 | path_put(&path); | |
475 | if (retry_estale(error, lookup_flags)) { | |
476 | lookup_flags |= LOOKUP_REVAL; | |
477 | goto retry; | |
478 | } | |
479 | out: | |
480 | return error; | |
481 | } | |
482 | ||
483 | static int chmod_common(struct path *path, umode_t mode) | |
484 | { | |
485 | struct inode *inode = path->dentry->d_inode; | |
486 | struct inode *delegated_inode = NULL; | |
487 | struct iattr newattrs; | |
488 | int error; | |
489 | ||
490 | error = mnt_want_write(path->mnt); | |
491 | if (error) | |
492 | return error; | |
493 | retry_deleg: | |
494 | mutex_lock(&inode->i_mutex); | |
495 | error = security_path_chmod(path, mode); | |
496 | if (error) | |
497 | goto out_unlock; | |
498 | newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO); | |
499 | newattrs.ia_valid = ATTR_MODE | ATTR_CTIME; | |
500 | error = notify_change(path->dentry, &newattrs, &delegated_inode); | |
501 | out_unlock: | |
502 | mutex_unlock(&inode->i_mutex); | |
503 | if (delegated_inode) { | |
504 | error = break_deleg_wait(&delegated_inode); | |
505 | if (!error) | |
506 | goto retry_deleg; | |
507 | } | |
508 | mnt_drop_write(path->mnt); | |
509 | return error; | |
510 | } | |
511 | ||
512 | SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode) | |
513 | { | |
514 | struct fd f = fdget(fd); | |
515 | int err = -EBADF; | |
516 | ||
517 | if (f.file) { | |
518 | audit_inode(NULL, f.file->f_path.dentry, 0); | |
519 | err = chmod_common(&f.file->f_path, mode); | |
520 | fdput(f); | |
521 | } | |
522 | return err; | |
523 | } | |
524 | ||
525 | SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode) | |
526 | { | |
527 | struct path path; | |
528 | int error; | |
529 | unsigned int lookup_flags = LOOKUP_FOLLOW; | |
530 | retry: | |
531 | error = user_path_at(dfd, filename, lookup_flags, &path); | |
532 | if (!error) { | |
533 | error = chmod_common(&path, mode); | |
534 | path_put(&path); | |
535 | if (retry_estale(error, lookup_flags)) { | |
536 | lookup_flags |= LOOKUP_REVAL; | |
537 | goto retry; | |
538 | } | |
539 | } | |
540 | return error; | |
541 | } | |
542 | ||
543 | SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode) | |
544 | { | |
545 | return sys_fchmodat(AT_FDCWD, filename, mode); | |
546 | } | |
547 | ||
548 | static int chown_common(struct path *path, uid_t user, gid_t group) | |
549 | { | |
550 | struct inode *inode = path->dentry->d_inode; | |
551 | struct inode *delegated_inode = NULL; | |
552 | int error; | |
553 | struct iattr newattrs; | |
554 | kuid_t uid; | |
555 | kgid_t gid; | |
556 | ||
557 | uid = make_kuid(current_user_ns(), user); | |
558 | gid = make_kgid(current_user_ns(), group); | |
559 | ||
560 | newattrs.ia_valid = ATTR_CTIME; | |
561 | if (user != (uid_t) -1) { | |
562 | if (!uid_valid(uid)) | |
563 | return -EINVAL; | |
564 | newattrs.ia_valid |= ATTR_UID; | |
565 | newattrs.ia_uid = uid; | |
566 | } | |
567 | if (group != (gid_t) -1) { | |
568 | if (!gid_valid(gid)) | |
569 | return -EINVAL; | |
570 | newattrs.ia_valid |= ATTR_GID; | |
571 | newattrs.ia_gid = gid; | |
572 | } | |
573 | if (!S_ISDIR(inode->i_mode)) | |
574 | newattrs.ia_valid |= | |
575 | ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV; | |
576 | retry_deleg: | |
577 | mutex_lock(&inode->i_mutex); | |
578 | error = security_path_chown(path, uid, gid); | |
579 | if (!error) | |
580 | error = notify_change(path->dentry, &newattrs, &delegated_inode); | |
581 | mutex_unlock(&inode->i_mutex); | |
582 | if (delegated_inode) { | |
583 | error = break_deleg_wait(&delegated_inode); | |
584 | if (!error) | |
585 | goto retry_deleg; | |
586 | } | |
587 | return error; | |
588 | } | |
589 | ||
590 | SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user, | |
591 | gid_t, group, int, flag) | |
592 | { | |
593 | struct path path; | |
594 | int error = -EINVAL; | |
595 | int lookup_flags; | |
596 | ||
597 | if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0) | |
598 | goto out; | |
599 | ||
600 | lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW; | |
601 | if (flag & AT_EMPTY_PATH) | |
602 | lookup_flags |= LOOKUP_EMPTY; | |
603 | retry: | |
604 | error = user_path_at(dfd, filename, lookup_flags, &path); | |
605 | if (error) | |
606 | goto out; | |
607 | error = mnt_want_write(path.mnt); | |
608 | if (error) | |
609 | goto out_release; | |
610 | error = chown_common(&path, user, group); | |
611 | mnt_drop_write(path.mnt); | |
612 | out_release: | |
613 | path_put(&path); | |
614 | if (retry_estale(error, lookup_flags)) { | |
615 | lookup_flags |= LOOKUP_REVAL; | |
616 | goto retry; | |
617 | } | |
618 | out: | |
619 | return error; | |
620 | } | |
621 | ||
622 | SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group) | |
623 | { | |
624 | return sys_fchownat(AT_FDCWD, filename, user, group, 0); | |
625 | } | |
626 | ||
627 | SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group) | |
628 | { | |
629 | return sys_fchownat(AT_FDCWD, filename, user, group, | |
630 | AT_SYMLINK_NOFOLLOW); | |
631 | } | |
632 | ||
633 | SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group) | |
634 | { | |
635 | struct fd f = fdget(fd); | |
636 | int error = -EBADF; | |
637 | ||
638 | if (!f.file) | |
639 | goto out; | |
640 | ||
641 | error = mnt_want_write_file(f.file); | |
642 | if (error) | |
643 | goto out_fput; | |
644 | audit_inode(NULL, f.file->f_path.dentry, 0); | |
645 | error = chown_common(&f.file->f_path, user, group); | |
646 | mnt_drop_write_file(f.file); | |
647 | out_fput: | |
648 | fdput(f); | |
649 | out: | |
650 | return error; | |
651 | } | |
652 | ||
653 | /* | |
654 | * You have to be very careful that these write | |
655 | * counts get cleaned up in error cases and | |
656 | * upon __fput(). This should probably never | |
657 | * be called outside of __dentry_open(). | |
658 | */ | |
659 | static inline int __get_file_write_access(struct inode *inode, | |
660 | struct vfsmount *mnt) | |
661 | { | |
662 | int error; | |
663 | error = get_write_access(inode); | |
664 | if (error) | |
665 | return error; | |
666 | /* | |
667 | * Do not take mount writer counts on | |
668 | * special files since no writes to | |
669 | * the mount itself will occur. | |
670 | */ | |
671 | if (!special_file(inode->i_mode)) { | |
672 | /* | |
673 | * Balanced in __fput() | |
674 | */ | |
675 | error = __mnt_want_write(mnt); | |
676 | if (error) | |
677 | put_write_access(inode); | |
678 | } | |
679 | return error; | |
680 | } | |
681 | ||
682 | int open_check_o_direct(struct file *f) | |
683 | { | |
684 | /* NB: we're sure to have correct a_ops only after f_op->open */ | |
685 | if (f->f_flags & O_DIRECT) { | |
686 | if (!f->f_mapping->a_ops || | |
687 | ((!f->f_mapping->a_ops->direct_IO) && | |
688 | (!f->f_mapping->a_ops->get_xip_mem))) { | |
689 | return -EINVAL; | |
690 | } | |
691 | } | |
692 | return 0; | |
693 | } | |
694 | ||
695 | static int do_dentry_open(struct file *f, | |
696 | int (*open)(struct inode *, struct file *), | |
697 | const struct cred *cred) | |
698 | { | |
699 | static const struct file_operations empty_fops = {}; | |
700 | struct inode *inode; | |
701 | int error; | |
702 | ||
703 | f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK | | |
704 | FMODE_PREAD | FMODE_PWRITE; | |
705 | ||
706 | if (unlikely(f->f_flags & O_PATH)) | |
707 | f->f_mode = FMODE_PATH; | |
708 | ||
709 | path_get(&f->f_path); | |
710 | inode = f->f_inode = f->f_path.dentry->d_inode; | |
711 | if (f->f_mode & FMODE_WRITE) { | |
712 | error = __get_file_write_access(inode, f->f_path.mnt); | |
713 | if (error) | |
714 | goto cleanup_file; | |
715 | if (!special_file(inode->i_mode)) | |
716 | file_take_write(f); | |
717 | } | |
718 | ||
719 | f->f_mapping = inode->i_mapping; | |
720 | ||
721 | if (unlikely(f->f_mode & FMODE_PATH)) { | |
722 | f->f_op = &empty_fops; | |
723 | return 0; | |
724 | } | |
725 | ||
726 | f->f_op = fops_get(inode->i_fop); | |
727 | if (unlikely(WARN_ON(!f->f_op))) { | |
728 | error = -ENODEV; | |
729 | goto cleanup_all; | |
730 | } | |
731 | ||
732 | error = security_file_open(f, cred); | |
733 | if (error) | |
734 | goto cleanup_all; | |
735 | ||
736 | error = break_lease(inode, f->f_flags); | |
737 | if (error) | |
738 | goto cleanup_all; | |
739 | ||
740 | if (!open) | |
741 | open = f->f_op->open; | |
742 | if (open) { | |
743 | error = open(inode, f); | |
744 | if (error) | |
745 | goto cleanup_all; | |
746 | } | |
747 | if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) | |
748 | i_readcount_inc(inode); | |
749 | ||
750 | f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC); | |
751 | ||
752 | file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping); | |
753 | ||
754 | return 0; | |
755 | ||
756 | cleanup_all: | |
757 | fops_put(f->f_op); | |
758 | if (f->f_mode & FMODE_WRITE) { | |
759 | put_write_access(inode); | |
760 | if (!special_file(inode->i_mode)) { | |
761 | /* | |
762 | * We don't consider this a real | |
763 | * mnt_want/drop_write() pair | |
764 | * because it all happenend right | |
765 | * here, so just reset the state. | |
766 | */ | |
767 | file_reset_write(f); | |
768 | __mnt_drop_write(f->f_path.mnt); | |
769 | } | |
770 | } | |
771 | cleanup_file: | |
772 | path_put(&f->f_path); | |
773 | f->f_path.mnt = NULL; | |
774 | f->f_path.dentry = NULL; | |
775 | f->f_inode = NULL; | |
776 | return error; | |
777 | } | |
778 | ||
779 | /** | |
780 | * finish_open - finish opening a file | |
781 | * @file: file pointer | |
782 | * @dentry: pointer to dentry | |
783 | * @open: open callback | |
784 | * @opened: state of open | |
785 | * | |
786 | * This can be used to finish opening a file passed to i_op->atomic_open(). | |
787 | * | |
788 | * If the open callback is set to NULL, then the standard f_op->open() | |
789 | * filesystem callback is substituted. | |
790 | * | |
791 | * NB: the dentry reference is _not_ consumed. If, for example, the dentry is | |
792 | * the return value of d_splice_alias(), then the caller needs to perform dput() | |
793 | * on it after finish_open(). | |
794 | * | |
795 | * On successful return @file is a fully instantiated open file. After this, if | |
796 | * an error occurs in ->atomic_open(), it needs to clean up with fput(). | |
797 | * | |
798 | * Returns zero on success or -errno if the open failed. | |
799 | */ | |
800 | int finish_open(struct file *file, struct dentry *dentry, | |
801 | int (*open)(struct inode *, struct file *), | |
802 | int *opened) | |
803 | { | |
804 | int error; | |
805 | BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ | |
806 | ||
807 | file->f_path.dentry = dentry; | |
808 | error = do_dentry_open(file, open, current_cred()); | |
809 | if (!error) | |
810 | *opened |= FILE_OPENED; | |
811 | ||
812 | return error; | |
813 | } | |
814 | EXPORT_SYMBOL(finish_open); | |
815 | ||
816 | /** | |
817 | * finish_no_open - finish ->atomic_open() without opening the file | |
818 | * | |
819 | * @file: file pointer | |
820 | * @dentry: dentry or NULL (as returned from ->lookup()) | |
821 | * | |
822 | * This can be used to set the result of a successful lookup in ->atomic_open(). | |
823 | * | |
824 | * NB: unlike finish_open() this function does consume the dentry reference and | |
825 | * the caller need not dput() it. | |
826 | * | |
827 | * Returns "1" which must be the return value of ->atomic_open() after having | |
828 | * called this function. | |
829 | */ | |
830 | int finish_no_open(struct file *file, struct dentry *dentry) | |
831 | { | |
832 | file->f_path.dentry = dentry; | |
833 | return 1; | |
834 | } | |
835 | EXPORT_SYMBOL(finish_no_open); | |
836 | ||
837 | struct file *dentry_open(const struct path *path, int flags, | |
838 | const struct cred *cred) | |
839 | { | |
840 | int error; | |
841 | struct file *f; | |
842 | ||
843 | validate_creds(cred); | |
844 | ||
845 | /* We must always pass in a valid mount pointer. */ | |
846 | BUG_ON(!path->mnt); | |
847 | ||
848 | f = get_empty_filp(); | |
849 | if (!IS_ERR(f)) { | |
850 | f->f_flags = flags; | |
851 | f->f_path = *path; | |
852 | error = do_dentry_open(f, NULL, cred); | |
853 | if (!error) { | |
854 | /* from now on we need fput() to dispose of f */ | |
855 | error = open_check_o_direct(f); | |
856 | if (error) { | |
857 | fput(f); | |
858 | f = ERR_PTR(error); | |
859 | } | |
860 | } else { | |
861 | put_filp(f); | |
862 | f = ERR_PTR(error); | |
863 | } | |
864 | } | |
865 | return f; | |
866 | } | |
867 | EXPORT_SYMBOL(dentry_open); | |
868 | ||
869 | static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op) | |
870 | { | |
871 | int lookup_flags = 0; | |
872 | int acc_mode; | |
873 | ||
874 | if (flags & (O_CREAT | __O_TMPFILE)) | |
875 | op->mode = (mode & S_IALLUGO) | S_IFREG; | |
876 | else | |
877 | op->mode = 0; | |
878 | ||
879 | /* Must never be set by userspace */ | |
880 | flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC; | |
881 | ||
882 | /* | |
883 | * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only | |
884 | * check for O_DSYNC if the need any syncing at all we enforce it's | |
885 | * always set instead of having to deal with possibly weird behaviour | |
886 | * for malicious applications setting only __O_SYNC. | |
887 | */ | |
888 | if (flags & __O_SYNC) | |
889 | flags |= O_DSYNC; | |
890 | ||
891 | if (flags & __O_TMPFILE) { | |
892 | if ((flags & O_TMPFILE_MASK) != O_TMPFILE) | |
893 | return -EINVAL; | |
894 | acc_mode = MAY_OPEN | ACC_MODE(flags); | |
895 | if (!(acc_mode & MAY_WRITE)) | |
896 | return -EINVAL; | |
897 | } else if (flags & O_PATH) { | |
898 | /* | |
899 | * If we have O_PATH in the open flag. Then we | |
900 | * cannot have anything other than the below set of flags | |
901 | */ | |
902 | flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH; | |
903 | acc_mode = 0; | |
904 | } else { | |
905 | acc_mode = MAY_OPEN | ACC_MODE(flags); | |
906 | } | |
907 | ||
908 | op->open_flag = flags; | |
909 | ||
910 | /* O_TRUNC implies we need access checks for write permissions */ | |
911 | if (flags & O_TRUNC) | |
912 | acc_mode |= MAY_WRITE; | |
913 | ||
914 | /* Allow the LSM permission hook to distinguish append | |
915 | access from general write access. */ | |
916 | if (flags & O_APPEND) | |
917 | acc_mode |= MAY_APPEND; | |
918 | ||
919 | op->acc_mode = acc_mode; | |
920 | ||
921 | op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN; | |
922 | ||
923 | if (flags & O_CREAT) { | |
924 | op->intent |= LOOKUP_CREATE; | |
925 | if (flags & O_EXCL) | |
926 | op->intent |= LOOKUP_EXCL; | |
927 | } | |
928 | ||
929 | if (flags & O_DIRECTORY) | |
930 | lookup_flags |= LOOKUP_DIRECTORY; | |
931 | if (!(flags & O_NOFOLLOW)) | |
932 | lookup_flags |= LOOKUP_FOLLOW; | |
933 | op->lookup_flags = lookup_flags; | |
934 | return 0; | |
935 | } | |
936 | ||
937 | /** | |
938 | * file_open_name - open file and return file pointer | |
939 | * | |
940 | * @name: struct filename containing path to open | |
941 | * @flags: open flags as per the open(2) second argument | |
942 | * @mode: mode for the new file if O_CREAT is set, else ignored | |
943 | * | |
944 | * This is the helper to open a file from kernelspace if you really | |
945 | * have to. But in generally you should not do this, so please move | |
946 | * along, nothing to see here.. | |
947 | */ | |
948 | struct file *file_open_name(struct filename *name, int flags, umode_t mode) | |
949 | { | |
950 | struct open_flags op; | |
951 | int err = build_open_flags(flags, mode, &op); | |
952 | return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op); | |
953 | } | |
954 | ||
955 | /** | |
956 | * filp_open - open file and return file pointer | |
957 | * | |
958 | * @filename: path to open | |
959 | * @flags: open flags as per the open(2) second argument | |
960 | * @mode: mode for the new file if O_CREAT is set, else ignored | |
961 | * | |
962 | * This is the helper to open a file from kernelspace if you really | |
963 | * have to. But in generally you should not do this, so please move | |
964 | * along, nothing to see here.. | |
965 | */ | |
966 | struct file *filp_open(const char *filename, int flags, umode_t mode) | |
967 | { | |
968 | struct filename name = {.name = filename}; | |
969 | return file_open_name(&name, flags, mode); | |
970 | } | |
971 | EXPORT_SYMBOL(filp_open); | |
972 | ||
973 | struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt, | |
974 | const char *filename, int flags) | |
975 | { | |
976 | struct open_flags op; | |
977 | int err = build_open_flags(flags, 0, &op); | |
978 | if (err) | |
979 | return ERR_PTR(err); | |
980 | if (flags & O_CREAT) | |
981 | return ERR_PTR(-EINVAL); | |
982 | if (!filename && (flags & O_DIRECTORY)) | |
983 | if (!dentry->d_inode->i_op->lookup) | |
984 | return ERR_PTR(-ENOTDIR); | |
985 | return do_file_open_root(dentry, mnt, filename, &op); | |
986 | } | |
987 | EXPORT_SYMBOL(file_open_root); | |
988 | ||
989 | long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode) | |
990 | { | |
991 | struct open_flags op; | |
992 | int fd = build_open_flags(flags, mode, &op); | |
993 | struct filename *tmp; | |
994 | ||
995 | if (fd) | |
996 | return fd; | |
997 | ||
998 | tmp = getname(filename); | |
999 | if (IS_ERR(tmp)) | |
1000 | return PTR_ERR(tmp); | |
1001 | ||
1002 | fd = get_unused_fd_flags(flags); | |
1003 | if (fd >= 0) { | |
1004 | struct file *f = do_filp_open(dfd, tmp, &op); | |
1005 | if (IS_ERR(f)) { | |
1006 | put_unused_fd(fd); | |
1007 | fd = PTR_ERR(f); | |
1008 | } else { | |
1009 | fsnotify_open(f); | |
1010 | fd_install(fd, f); | |
1011 | } | |
1012 | } | |
1013 | putname(tmp); | |
1014 | return fd; | |
1015 | } | |
1016 | ||
1017 | SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode) | |
1018 | { | |
1019 | if (force_o_largefile()) | |
1020 | flags |= O_LARGEFILE; | |
1021 | ||
1022 | return do_sys_open(AT_FDCWD, filename, flags, mode); | |
1023 | } | |
1024 | ||
1025 | SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, | |
1026 | umode_t, mode) | |
1027 | { | |
1028 | if (force_o_largefile()) | |
1029 | flags |= O_LARGEFILE; | |
1030 | ||
1031 | return do_sys_open(dfd, filename, flags, mode); | |
1032 | } | |
1033 | ||
1034 | #ifndef __alpha__ | |
1035 | ||
1036 | /* | |
1037 | * For backward compatibility? Maybe this should be moved | |
1038 | * into arch/i386 instead? | |
1039 | */ | |
1040 | SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode) | |
1041 | { | |
1042 | return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode); | |
1043 | } | |
1044 | ||
1045 | #endif | |
1046 | ||
1047 | /* | |
1048 | * "id" is the POSIX thread ID. We use the | |
1049 | * files pointer for this.. | |
1050 | */ | |
1051 | int filp_close(struct file *filp, fl_owner_t id) | |
1052 | { | |
1053 | int retval = 0; | |
1054 | ||
1055 | if (!file_count(filp)) { | |
1056 | printk(KERN_ERR "VFS: Close: file count is 0\n"); | |
1057 | return 0; | |
1058 | } | |
1059 | ||
1060 | if (filp->f_op->flush) | |
1061 | retval = filp->f_op->flush(filp, id); | |
1062 | ||
1063 | if (likely(!(filp->f_mode & FMODE_PATH))) { | |
1064 | dnotify_flush(filp, id); | |
1065 | locks_remove_posix(filp, id); | |
1066 | } | |
1067 | fput(filp); | |
1068 | return retval; | |
1069 | } | |
1070 | ||
1071 | EXPORT_SYMBOL(filp_close); | |
1072 | ||
1073 | /* | |
1074 | * Careful here! We test whether the file pointer is NULL before | |
1075 | * releasing the fd. This ensures that one clone task can't release | |
1076 | * an fd while another clone is opening it. | |
1077 | */ | |
1078 | SYSCALL_DEFINE1(close, unsigned int, fd) | |
1079 | { | |
1080 | int retval = __close_fd(current->files, fd); | |
1081 | ||
1082 | /* can't restart close syscall because file table entry was cleared */ | |
1083 | if (unlikely(retval == -ERESTARTSYS || | |
1084 | retval == -ERESTARTNOINTR || | |
1085 | retval == -ERESTARTNOHAND || | |
1086 | retval == -ERESTART_RESTARTBLOCK)) | |
1087 | retval = -EINTR; | |
1088 | ||
1089 | return retval; | |
1090 | } | |
1091 | EXPORT_SYMBOL(sys_close); | |
1092 | ||
1093 | /* | |
1094 | * This routine simulates a hangup on the tty, to arrange that users | |
1095 | * are given clean terminals at login time. | |
1096 | */ | |
1097 | SYSCALL_DEFINE0(vhangup) | |
1098 | { | |
1099 | if (capable(CAP_SYS_TTY_CONFIG)) { | |
1100 | tty_vhangup_self(); | |
1101 | return 0; | |
1102 | } | |
1103 | return -EPERM; | |
1104 | } | |
1105 | ||
1106 | /* | |
1107 | * Called when an inode is about to be open. | |
1108 | * We use this to disallow opening large files on 32bit systems if | |
1109 | * the caller didn't specify O_LARGEFILE. On 64bit systems we force | |
1110 | * on this flag in sys_open. | |
1111 | */ | |
1112 | int generic_file_open(struct inode * inode, struct file * filp) | |
1113 | { | |
1114 | if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS) | |
1115 | return -EOVERFLOW; | |
1116 | return 0; | |
1117 | } | |
1118 | ||
1119 | EXPORT_SYMBOL(generic_file_open); | |
1120 | ||
1121 | /* | |
1122 | * This is used by subsystems that don't want seekable | |
1123 | * file descriptors. The function is not supposed to ever fail, the only | |
1124 | * reason it returns an 'int' and not 'void' is so that it can be plugged | |
1125 | * directly into file_operations structure. | |
1126 | */ | |
1127 | int nonseekable_open(struct inode *inode, struct file *filp) | |
1128 | { | |
1129 | filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE); | |
1130 | return 0; | |
1131 | } | |
1132 | ||
1133 | EXPORT_SYMBOL(nonseekable_open); |