nfs: handle multiple reqs in nfs_page_async_flush
[linux/fpc-iii.git] / fs / fcntl.c
blob72c82f69b01b28594e56bb9518df6f211f0d51a9
1 /*
2 * linux/fs/fcntl.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
7 #include <linux/syscalls.h>
8 #include <linux/init.h>
9 #include <linux/mm.h>
10 #include <linux/fs.h>
11 #include <linux/file.h>
12 #include <linux/fdtable.h>
13 #include <linux/capability.h>
14 #include <linux/dnotify.h>
15 #include <linux/slab.h>
16 #include <linux/module.h>
17 #include <linux/pipe_fs_i.h>
18 #include <linux/security.h>
19 #include <linux/ptrace.h>
20 #include <linux/signal.h>
21 #include <linux/rcupdate.h>
22 #include <linux/pid_namespace.h>
23 #include <linux/user_namespace.h>
25 #include <asm/poll.h>
26 #include <asm/siginfo.h>
27 #include <asm/uaccess.h>
29 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
31 static int setfl(int fd, struct file * filp, unsigned long arg)
33 struct inode * inode = file_inode(filp);
34 int error = 0;
37 * O_APPEND cannot be cleared if the file is marked as append-only
38 * and the file is open for write.
40 if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
41 return -EPERM;
43 /* O_NOATIME can only be set by the owner or superuser */
44 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
45 if (!inode_owner_or_capable(inode))
46 return -EPERM;
48 /* required for strict SunOS emulation */
49 if (O_NONBLOCK != O_NDELAY)
50 if (arg & O_NDELAY)
51 arg |= O_NONBLOCK;
53 if (arg & O_DIRECT) {
54 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
55 !filp->f_mapping->a_ops->direct_IO)
56 return -EINVAL;
59 if (filp->f_op->check_flags)
60 error = filp->f_op->check_flags(arg);
61 if (error)
62 return error;
65 * ->fasync() is responsible for setting the FASYNC bit.
67 if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op->fasync) {
68 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
69 if (error < 0)
70 goto out;
71 if (error > 0)
72 error = 0;
74 spin_lock(&filp->f_lock);
75 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
76 spin_unlock(&filp->f_lock);
78 out:
79 return error;
82 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
83 int force)
85 write_lock_irq(&filp->f_owner.lock);
86 if (force || !filp->f_owner.pid) {
87 put_pid(filp->f_owner.pid);
88 filp->f_owner.pid = get_pid(pid);
89 filp->f_owner.pid_type = type;
91 if (pid) {
92 const struct cred *cred = current_cred();
93 filp->f_owner.uid = cred->uid;
94 filp->f_owner.euid = cred->euid;
97 write_unlock_irq(&filp->f_owner.lock);
100 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
101 int force)
103 int err;
105 err = security_file_set_fowner(filp);
106 if (err)
107 return err;
109 f_modown(filp, pid, type, force);
110 return 0;
112 EXPORT_SYMBOL(__f_setown);
114 int f_setown(struct file *filp, unsigned long arg, int force)
116 enum pid_type type;
117 struct pid *pid;
118 int who = arg;
119 int result;
120 type = PIDTYPE_PID;
121 if (who < 0) {
122 type = PIDTYPE_PGID;
123 who = -who;
125 rcu_read_lock();
126 pid = find_vpid(who);
127 result = __f_setown(filp, pid, type, force);
128 rcu_read_unlock();
129 return result;
131 EXPORT_SYMBOL(f_setown);
133 void f_delown(struct file *filp)
135 f_modown(filp, NULL, PIDTYPE_PID, 1);
138 pid_t f_getown(struct file *filp)
140 pid_t pid;
141 read_lock(&filp->f_owner.lock);
142 pid = pid_vnr(filp->f_owner.pid);
143 if (filp->f_owner.pid_type == PIDTYPE_PGID)
144 pid = -pid;
145 read_unlock(&filp->f_owner.lock);
146 return pid;
149 static int f_setown_ex(struct file *filp, unsigned long arg)
151 struct f_owner_ex __user *owner_p = (void __user *)arg;
152 struct f_owner_ex owner;
153 struct pid *pid;
154 int type;
155 int ret;
157 ret = copy_from_user(&owner, owner_p, sizeof(owner));
158 if (ret)
159 return -EFAULT;
161 switch (owner.type) {
162 case F_OWNER_TID:
163 type = PIDTYPE_MAX;
164 break;
166 case F_OWNER_PID:
167 type = PIDTYPE_PID;
168 break;
170 case F_OWNER_PGRP:
171 type = PIDTYPE_PGID;
172 break;
174 default:
175 return -EINVAL;
178 rcu_read_lock();
179 pid = find_vpid(owner.pid);
180 if (owner.pid && !pid)
181 ret = -ESRCH;
182 else
183 ret = __f_setown(filp, pid, type, 1);
184 rcu_read_unlock();
186 return ret;
189 static int f_getown_ex(struct file *filp, unsigned long arg)
191 struct f_owner_ex __user *owner_p = (void __user *)arg;
192 struct f_owner_ex owner;
193 int ret = 0;
195 read_lock(&filp->f_owner.lock);
196 owner.pid = pid_vnr(filp->f_owner.pid);
197 switch (filp->f_owner.pid_type) {
198 case PIDTYPE_MAX:
199 owner.type = F_OWNER_TID;
200 break;
202 case PIDTYPE_PID:
203 owner.type = F_OWNER_PID;
204 break;
206 case PIDTYPE_PGID:
207 owner.type = F_OWNER_PGRP;
208 break;
210 default:
211 WARN_ON(1);
212 ret = -EINVAL;
213 break;
215 read_unlock(&filp->f_owner.lock);
217 if (!ret) {
218 ret = copy_to_user(owner_p, &owner, sizeof(owner));
219 if (ret)
220 ret = -EFAULT;
222 return ret;
225 #ifdef CONFIG_CHECKPOINT_RESTORE
226 static int f_getowner_uids(struct file *filp, unsigned long arg)
228 struct user_namespace *user_ns = current_user_ns();
229 uid_t __user *dst = (void __user *)arg;
230 uid_t src[2];
231 int err;
233 read_lock(&filp->f_owner.lock);
234 src[0] = from_kuid(user_ns, filp->f_owner.uid);
235 src[1] = from_kuid(user_ns, filp->f_owner.euid);
236 read_unlock(&filp->f_owner.lock);
238 err = put_user(src[0], &dst[0]);
239 err |= put_user(src[1], &dst[1]);
241 return err;
243 #else
244 static int f_getowner_uids(struct file *filp, unsigned long arg)
246 return -EINVAL;
248 #endif
250 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
251 struct file *filp)
253 long err = -EINVAL;
255 switch (cmd) {
256 case F_DUPFD:
257 err = f_dupfd(arg, filp, 0);
258 break;
259 case F_DUPFD_CLOEXEC:
260 err = f_dupfd(arg, filp, O_CLOEXEC);
261 break;
262 case F_GETFD:
263 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
264 break;
265 case F_SETFD:
266 err = 0;
267 set_close_on_exec(fd, arg & FD_CLOEXEC);
268 break;
269 case F_GETFL:
270 err = filp->f_flags;
271 break;
272 case F_SETFL:
273 err = setfl(fd, filp, arg);
274 break;
275 #if BITS_PER_LONG != 32
276 /* 32-bit arches must use fcntl64() */
277 case F_OFD_GETLK:
278 #endif
279 case F_GETLK:
280 err = fcntl_getlk(filp, cmd, (struct flock __user *) arg);
281 break;
282 #if BITS_PER_LONG != 32
283 /* 32-bit arches must use fcntl64() */
284 case F_OFD_SETLK:
285 case F_OFD_SETLKW:
286 #endif
287 /* Fallthrough */
288 case F_SETLK:
289 case F_SETLKW:
290 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
291 break;
292 case F_GETOWN:
294 * XXX If f_owner is a process group, the
295 * negative return value will get converted
296 * into an error. Oops. If we keep the
297 * current syscall conventions, the only way
298 * to fix this will be in libc.
300 err = f_getown(filp);
301 force_successful_syscall_return();
302 break;
303 case F_SETOWN:
304 err = f_setown(filp, arg, 1);
305 break;
306 case F_GETOWN_EX:
307 err = f_getown_ex(filp, arg);
308 break;
309 case F_SETOWN_EX:
310 err = f_setown_ex(filp, arg);
311 break;
312 case F_GETOWNER_UIDS:
313 err = f_getowner_uids(filp, arg);
314 break;
315 case F_GETSIG:
316 err = filp->f_owner.signum;
317 break;
318 case F_SETSIG:
319 /* arg == 0 restores default behaviour. */
320 if (!valid_signal(arg)) {
321 break;
323 err = 0;
324 filp->f_owner.signum = arg;
325 break;
326 case F_GETLEASE:
327 err = fcntl_getlease(filp);
328 break;
329 case F_SETLEASE:
330 err = fcntl_setlease(fd, filp, arg);
331 break;
332 case F_NOTIFY:
333 err = fcntl_dirnotify(fd, filp, arg);
334 break;
335 case F_SETPIPE_SZ:
336 case F_GETPIPE_SZ:
337 err = pipe_fcntl(filp, cmd, arg);
338 break;
339 default:
340 break;
342 return err;
345 static int check_fcntl_cmd(unsigned cmd)
347 switch (cmd) {
348 case F_DUPFD:
349 case F_DUPFD_CLOEXEC:
350 case F_GETFD:
351 case F_SETFD:
352 case F_GETFL:
353 return 1;
355 return 0;
358 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
360 struct fd f = fdget_raw(fd);
361 long err = -EBADF;
363 if (!f.file)
364 goto out;
366 if (unlikely(f.file->f_mode & FMODE_PATH)) {
367 if (!check_fcntl_cmd(cmd))
368 goto out1;
371 err = security_file_fcntl(f.file, cmd, arg);
372 if (!err)
373 err = do_fcntl(fd, cmd, arg, f.file);
375 out1:
376 fdput(f);
377 out:
378 return err;
381 #if BITS_PER_LONG == 32
382 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
383 unsigned long, arg)
385 struct fd f = fdget_raw(fd);
386 long err = -EBADF;
388 if (!f.file)
389 goto out;
391 if (unlikely(f.file->f_mode & FMODE_PATH)) {
392 if (!check_fcntl_cmd(cmd))
393 goto out1;
396 err = security_file_fcntl(f.file, cmd, arg);
397 if (err)
398 goto out1;
400 switch (cmd) {
401 case F_GETLK64:
402 case F_OFD_GETLK:
403 err = fcntl_getlk64(f.file, cmd, (struct flock64 __user *) arg);
404 break;
405 case F_SETLK64:
406 case F_SETLKW64:
407 case F_OFD_SETLK:
408 case F_OFD_SETLKW:
409 err = fcntl_setlk64(fd, f.file, cmd,
410 (struct flock64 __user *) arg);
411 break;
412 default:
413 err = do_fcntl(fd, cmd, arg, f.file);
414 break;
416 out1:
417 fdput(f);
418 out:
419 return err;
421 #endif
423 /* Table to convert sigio signal codes into poll band bitmaps */
425 static const long band_table[NSIGPOLL] = {
426 POLLIN | POLLRDNORM, /* POLL_IN */
427 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
428 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
429 POLLERR, /* POLL_ERR */
430 POLLPRI | POLLRDBAND, /* POLL_PRI */
431 POLLHUP | POLLERR /* POLL_HUP */
434 static inline int sigio_perm(struct task_struct *p,
435 struct fown_struct *fown, int sig)
437 const struct cred *cred;
438 int ret;
440 rcu_read_lock();
441 cred = __task_cred(p);
442 ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
443 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
444 uid_eq(fown->uid, cred->suid) || uid_eq(fown->uid, cred->uid)) &&
445 !security_file_send_sigiotask(p, fown, sig));
446 rcu_read_unlock();
447 return ret;
450 static void send_sigio_to_task(struct task_struct *p,
451 struct fown_struct *fown,
452 int fd, int reason, int group)
455 * F_SETSIG can change ->signum lockless in parallel, make
456 * sure we read it once and use the same value throughout.
458 int signum = ACCESS_ONCE(fown->signum);
460 if (!sigio_perm(p, fown, signum))
461 return;
463 switch (signum) {
464 siginfo_t si;
465 default:
466 /* Queue a rt signal with the appropriate fd as its
467 value. We use SI_SIGIO as the source, not
468 SI_KERNEL, since kernel signals always get
469 delivered even if we can't queue. Failure to
470 queue in this case _should_ be reported; we fall
471 back to SIGIO in that case. --sct */
472 si.si_signo = signum;
473 si.si_errno = 0;
474 si.si_code = reason;
475 /* Make sure we are called with one of the POLL_*
476 reasons, otherwise we could leak kernel stack into
477 userspace. */
478 BUG_ON((reason & __SI_MASK) != __SI_POLL);
479 if (reason - POLL_IN >= NSIGPOLL)
480 si.si_band = ~0L;
481 else
482 si.si_band = band_table[reason - POLL_IN];
483 si.si_fd = fd;
484 if (!do_send_sig_info(signum, &si, p, group))
485 break;
486 /* fall-through: fall back on the old plain SIGIO signal */
487 case 0:
488 do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
492 void send_sigio(struct fown_struct *fown, int fd, int band)
494 struct task_struct *p;
495 enum pid_type type;
496 struct pid *pid;
497 int group = 1;
499 read_lock(&fown->lock);
501 type = fown->pid_type;
502 if (type == PIDTYPE_MAX) {
503 group = 0;
504 type = PIDTYPE_PID;
507 pid = fown->pid;
508 if (!pid)
509 goto out_unlock_fown;
511 read_lock(&tasklist_lock);
512 do_each_pid_task(pid, type, p) {
513 send_sigio_to_task(p, fown, fd, band, group);
514 } while_each_pid_task(pid, type, p);
515 read_unlock(&tasklist_lock);
516 out_unlock_fown:
517 read_unlock(&fown->lock);
520 static void send_sigurg_to_task(struct task_struct *p,
521 struct fown_struct *fown, int group)
523 if (sigio_perm(p, fown, SIGURG))
524 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
527 int send_sigurg(struct fown_struct *fown)
529 struct task_struct *p;
530 enum pid_type type;
531 struct pid *pid;
532 int group = 1;
533 int ret = 0;
535 read_lock(&fown->lock);
537 type = fown->pid_type;
538 if (type == PIDTYPE_MAX) {
539 group = 0;
540 type = PIDTYPE_PID;
543 pid = fown->pid;
544 if (!pid)
545 goto out_unlock_fown;
547 ret = 1;
549 read_lock(&tasklist_lock);
550 do_each_pid_task(pid, type, p) {
551 send_sigurg_to_task(p, fown, group);
552 } while_each_pid_task(pid, type, p);
553 read_unlock(&tasklist_lock);
554 out_unlock_fown:
555 read_unlock(&fown->lock);
556 return ret;
559 static DEFINE_SPINLOCK(fasync_lock);
560 static struct kmem_cache *fasync_cache __read_mostly;
562 static void fasync_free_rcu(struct rcu_head *head)
564 kmem_cache_free(fasync_cache,
565 container_of(head, struct fasync_struct, fa_rcu));
569 * Remove a fasync entry. If successfully removed, return
570 * positive and clear the FASYNC flag. If no entry exists,
571 * do nothing and return 0.
573 * NOTE! It is very important that the FASYNC flag always
574 * match the state "is the filp on a fasync list".
577 int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
579 struct fasync_struct *fa, **fp;
580 int result = 0;
582 spin_lock(&filp->f_lock);
583 spin_lock(&fasync_lock);
584 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
585 if (fa->fa_file != filp)
586 continue;
588 spin_lock_irq(&fa->fa_lock);
589 fa->fa_file = NULL;
590 spin_unlock_irq(&fa->fa_lock);
592 *fp = fa->fa_next;
593 call_rcu(&fa->fa_rcu, fasync_free_rcu);
594 filp->f_flags &= ~FASYNC;
595 result = 1;
596 break;
598 spin_unlock(&fasync_lock);
599 spin_unlock(&filp->f_lock);
600 return result;
603 struct fasync_struct *fasync_alloc(void)
605 return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
609 * NOTE! This can be used only for unused fasync entries:
610 * entries that actually got inserted on the fasync list
611 * need to be released by rcu - see fasync_remove_entry.
613 void fasync_free(struct fasync_struct *new)
615 kmem_cache_free(fasync_cache, new);
619 * Insert a new entry into the fasync list. Return the pointer to the
620 * old one if we didn't use the new one.
622 * NOTE! It is very important that the FASYNC flag always
623 * match the state "is the filp on a fasync list".
625 struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
627 struct fasync_struct *fa, **fp;
629 spin_lock(&filp->f_lock);
630 spin_lock(&fasync_lock);
631 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
632 if (fa->fa_file != filp)
633 continue;
635 spin_lock_irq(&fa->fa_lock);
636 fa->fa_fd = fd;
637 spin_unlock_irq(&fa->fa_lock);
638 goto out;
641 spin_lock_init(&new->fa_lock);
642 new->magic = FASYNC_MAGIC;
643 new->fa_file = filp;
644 new->fa_fd = fd;
645 new->fa_next = *fapp;
646 rcu_assign_pointer(*fapp, new);
647 filp->f_flags |= FASYNC;
649 out:
650 spin_unlock(&fasync_lock);
651 spin_unlock(&filp->f_lock);
652 return fa;
656 * Add a fasync entry. Return negative on error, positive if
657 * added, and zero if did nothing but change an existing one.
659 static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
661 struct fasync_struct *new;
663 new = fasync_alloc();
664 if (!new)
665 return -ENOMEM;
668 * fasync_insert_entry() returns the old (update) entry if
669 * it existed.
671 * So free the (unused) new entry and return 0 to let the
672 * caller know that we didn't add any new fasync entries.
674 if (fasync_insert_entry(fd, filp, fapp, new)) {
675 fasync_free(new);
676 return 0;
679 return 1;
683 * fasync_helper() is used by almost all character device drivers
684 * to set up the fasync queue, and for regular files by the file
685 * lease code. It returns negative on error, 0 if it did no changes
686 * and positive if it added/deleted the entry.
688 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
690 if (!on)
691 return fasync_remove_entry(filp, fapp);
692 return fasync_add_entry(fd, filp, fapp);
695 EXPORT_SYMBOL(fasync_helper);
698 * rcu_read_lock() is held
700 static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
702 while (fa) {
703 struct fown_struct *fown;
704 unsigned long flags;
706 if (fa->magic != FASYNC_MAGIC) {
707 printk(KERN_ERR "kill_fasync: bad magic number in "
708 "fasync_struct!\n");
709 return;
711 spin_lock_irqsave(&fa->fa_lock, flags);
712 if (fa->fa_file) {
713 fown = &fa->fa_file->f_owner;
714 /* Don't send SIGURG to processes which have not set a
715 queued signum: SIGURG has its own default signalling
716 mechanism. */
717 if (!(sig == SIGURG && fown->signum == 0))
718 send_sigio(fown, fa->fa_fd, band);
720 spin_unlock_irqrestore(&fa->fa_lock, flags);
721 fa = rcu_dereference(fa->fa_next);
725 void kill_fasync(struct fasync_struct **fp, int sig, int band)
727 /* First a quick test without locking: usually
728 * the list is empty.
730 if (*fp) {
731 rcu_read_lock();
732 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
733 rcu_read_unlock();
736 EXPORT_SYMBOL(kill_fasync);
738 static int __init fcntl_init(void)
741 * Please add new bits here to ensure allocation uniqueness.
742 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
743 * is defined as O_NONBLOCK on some platforms and not on others.
745 BUILD_BUG_ON(20 - 1 /* for O_RDONLY being 0 */ != HWEIGHT32(
746 O_RDONLY | O_WRONLY | O_RDWR |
747 O_CREAT | O_EXCL | O_NOCTTY |
748 O_TRUNC | O_APPEND | /* O_NONBLOCK | */
749 __O_SYNC | O_DSYNC | FASYNC |
750 O_DIRECT | O_LARGEFILE | O_DIRECTORY |
751 O_NOFOLLOW | O_NOATIME | O_CLOEXEC |
752 __FMODE_EXEC | O_PATH | __O_TMPFILE
755 fasync_cache = kmem_cache_create("fasync_cache",
756 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
757 return 0;
760 module_init(fcntl_init)