ipv6: add support of equal cost multipath (ECMP)
[linux-2.6.git] / fs / fcntl.c
blob71a600a19f06078f235d8dbba6388cee2594365a
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 = filp->f_path.dentry->d_inode;
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 && 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 &&
68 filp->f_op->fasync) {
69 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
70 if (error < 0)
71 goto out;
72 if (error > 0)
73 error = 0;
75 spin_lock(&filp->f_lock);
76 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
77 spin_unlock(&filp->f_lock);
79 out:
80 return error;
83 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
84 int force)
86 write_lock_irq(&filp->f_owner.lock);
87 if (force || !filp->f_owner.pid) {
88 put_pid(filp->f_owner.pid);
89 filp->f_owner.pid = get_pid(pid);
90 filp->f_owner.pid_type = type;
92 if (pid) {
93 const struct cred *cred = current_cred();
94 filp->f_owner.uid = cred->uid;
95 filp->f_owner.euid = cred->euid;
98 write_unlock_irq(&filp->f_owner.lock);
101 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
102 int force)
104 int err;
106 err = security_file_set_fowner(filp);
107 if (err)
108 return err;
110 f_modown(filp, pid, type, force);
111 return 0;
113 EXPORT_SYMBOL(__f_setown);
115 int f_setown(struct file *filp, unsigned long arg, int force)
117 enum pid_type type;
118 struct pid *pid;
119 int who = arg;
120 int result;
121 type = PIDTYPE_PID;
122 if (who < 0) {
123 type = PIDTYPE_PGID;
124 who = -who;
126 rcu_read_lock();
127 pid = find_vpid(who);
128 result = __f_setown(filp, pid, type, force);
129 rcu_read_unlock();
130 return result;
132 EXPORT_SYMBOL(f_setown);
134 void f_delown(struct file *filp)
136 f_modown(filp, NULL, PIDTYPE_PID, 1);
139 pid_t f_getown(struct file *filp)
141 pid_t pid;
142 read_lock(&filp->f_owner.lock);
143 pid = pid_vnr(filp->f_owner.pid);
144 if (filp->f_owner.pid_type == PIDTYPE_PGID)
145 pid = -pid;
146 read_unlock(&filp->f_owner.lock);
147 return pid;
150 static int f_setown_ex(struct file *filp, unsigned long arg)
152 struct f_owner_ex __user *owner_p = (void __user *)arg;
153 struct f_owner_ex owner;
154 struct pid *pid;
155 int type;
156 int ret;
158 ret = copy_from_user(&owner, owner_p, sizeof(owner));
159 if (ret)
160 return -EFAULT;
162 switch (owner.type) {
163 case F_OWNER_TID:
164 type = PIDTYPE_MAX;
165 break;
167 case F_OWNER_PID:
168 type = PIDTYPE_PID;
169 break;
171 case F_OWNER_PGRP:
172 type = PIDTYPE_PGID;
173 break;
175 default:
176 return -EINVAL;
179 rcu_read_lock();
180 pid = find_vpid(owner.pid);
181 if (owner.pid && !pid)
182 ret = -ESRCH;
183 else
184 ret = __f_setown(filp, pid, type, 1);
185 rcu_read_unlock();
187 return ret;
190 static int f_getown_ex(struct file *filp, unsigned long arg)
192 struct f_owner_ex __user *owner_p = (void __user *)arg;
193 struct f_owner_ex owner;
194 int ret = 0;
196 read_lock(&filp->f_owner.lock);
197 owner.pid = pid_vnr(filp->f_owner.pid);
198 switch (filp->f_owner.pid_type) {
199 case PIDTYPE_MAX:
200 owner.type = F_OWNER_TID;
201 break;
203 case PIDTYPE_PID:
204 owner.type = F_OWNER_PID;
205 break;
207 case PIDTYPE_PGID:
208 owner.type = F_OWNER_PGRP;
209 break;
211 default:
212 WARN_ON(1);
213 ret = -EINVAL;
214 break;
216 read_unlock(&filp->f_owner.lock);
218 if (!ret) {
219 ret = copy_to_user(owner_p, &owner, sizeof(owner));
220 if (ret)
221 ret = -EFAULT;
223 return ret;
226 #ifdef CONFIG_CHECKPOINT_RESTORE
227 static int f_getowner_uids(struct file *filp, unsigned long arg)
229 struct user_namespace *user_ns = current_user_ns();
230 uid_t __user *dst = (void __user *)arg;
231 uid_t src[2];
232 int err;
234 read_lock(&filp->f_owner.lock);
235 src[0] = from_kuid(user_ns, filp->f_owner.uid);
236 src[1] = from_kuid(user_ns, filp->f_owner.euid);
237 read_unlock(&filp->f_owner.lock);
239 err = put_user(src[0], &dst[0]);
240 err |= put_user(src[1], &dst[1]);
242 return err;
244 #else
245 static int f_getowner_uids(struct file *filp, unsigned long arg)
247 return -EINVAL;
249 #endif
251 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
252 struct file *filp)
254 long err = -EINVAL;
256 switch (cmd) {
257 case F_DUPFD:
258 err = f_dupfd(arg, filp, 0);
259 break;
260 case F_DUPFD_CLOEXEC:
261 err = f_dupfd(arg, filp, O_CLOEXEC);
262 break;
263 case F_GETFD:
264 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
265 break;
266 case F_SETFD:
267 err = 0;
268 set_close_on_exec(fd, arg & FD_CLOEXEC);
269 break;
270 case F_GETFL:
271 err = filp->f_flags;
272 break;
273 case F_SETFL:
274 err = setfl(fd, filp, arg);
275 break;
276 case F_GETLK:
277 err = fcntl_getlk(filp, (struct flock __user *) arg);
278 break;
279 case F_SETLK:
280 case F_SETLKW:
281 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
282 break;
283 case F_GETOWN:
285 * XXX If f_owner is a process group, the
286 * negative return value will get converted
287 * into an error. Oops. If we keep the
288 * current syscall conventions, the only way
289 * to fix this will be in libc.
291 err = f_getown(filp);
292 force_successful_syscall_return();
293 break;
294 case F_SETOWN:
295 err = f_setown(filp, arg, 1);
296 break;
297 case F_GETOWN_EX:
298 err = f_getown_ex(filp, arg);
299 break;
300 case F_SETOWN_EX:
301 err = f_setown_ex(filp, arg);
302 break;
303 case F_GETOWNER_UIDS:
304 err = f_getowner_uids(filp, arg);
305 break;
306 case F_GETSIG:
307 err = filp->f_owner.signum;
308 break;
309 case F_SETSIG:
310 /* arg == 0 restores default behaviour. */
311 if (!valid_signal(arg)) {
312 break;
314 err = 0;
315 filp->f_owner.signum = arg;
316 break;
317 case F_GETLEASE:
318 err = fcntl_getlease(filp);
319 break;
320 case F_SETLEASE:
321 err = fcntl_setlease(fd, filp, arg);
322 break;
323 case F_NOTIFY:
324 err = fcntl_dirnotify(fd, filp, arg);
325 break;
326 case F_SETPIPE_SZ:
327 case F_GETPIPE_SZ:
328 err = pipe_fcntl(filp, cmd, arg);
329 break;
330 default:
331 break;
333 return err;
336 static int check_fcntl_cmd(unsigned cmd)
338 switch (cmd) {
339 case F_DUPFD:
340 case F_DUPFD_CLOEXEC:
341 case F_GETFD:
342 case F_SETFD:
343 case F_GETFL:
344 return 1;
346 return 0;
349 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
351 struct fd f = fdget_raw(fd);
352 long err = -EBADF;
354 if (!f.file)
355 goto out;
357 if (unlikely(f.file->f_mode & FMODE_PATH)) {
358 if (!check_fcntl_cmd(cmd))
359 goto out1;
362 err = security_file_fcntl(f.file, cmd, arg);
363 if (!err)
364 err = do_fcntl(fd, cmd, arg, f.file);
366 out1:
367 fdput(f);
368 out:
369 return err;
372 #if BITS_PER_LONG == 32
373 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
374 unsigned long, arg)
376 struct fd f = fdget_raw(fd);
377 long err = -EBADF;
379 if (!f.file)
380 goto out;
382 if (unlikely(f.file->f_mode & FMODE_PATH)) {
383 if (!check_fcntl_cmd(cmd))
384 goto out1;
387 err = security_file_fcntl(f.file, cmd, arg);
388 if (err)
389 goto out1;
391 switch (cmd) {
392 case F_GETLK64:
393 err = fcntl_getlk64(f.file, (struct flock64 __user *) arg);
394 break;
395 case F_SETLK64:
396 case F_SETLKW64:
397 err = fcntl_setlk64(fd, f.file, cmd,
398 (struct flock64 __user *) arg);
399 break;
400 default:
401 err = do_fcntl(fd, cmd, arg, f.file);
402 break;
404 out1:
405 fdput(f);
406 out:
407 return err;
409 #endif
411 /* Table to convert sigio signal codes into poll band bitmaps */
413 static const long band_table[NSIGPOLL] = {
414 POLLIN | POLLRDNORM, /* POLL_IN */
415 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
416 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
417 POLLERR, /* POLL_ERR */
418 POLLPRI | POLLRDBAND, /* POLL_PRI */
419 POLLHUP | POLLERR /* POLL_HUP */
422 static inline int sigio_perm(struct task_struct *p,
423 struct fown_struct *fown, int sig)
425 const struct cred *cred;
426 int ret;
428 rcu_read_lock();
429 cred = __task_cred(p);
430 ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
431 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
432 uid_eq(fown->uid, cred->suid) || uid_eq(fown->uid, cred->uid)) &&
433 !security_file_send_sigiotask(p, fown, sig));
434 rcu_read_unlock();
435 return ret;
438 static void send_sigio_to_task(struct task_struct *p,
439 struct fown_struct *fown,
440 int fd, int reason, int group)
443 * F_SETSIG can change ->signum lockless in parallel, make
444 * sure we read it once and use the same value throughout.
446 int signum = ACCESS_ONCE(fown->signum);
448 if (!sigio_perm(p, fown, signum))
449 return;
451 switch (signum) {
452 siginfo_t si;
453 default:
454 /* Queue a rt signal with the appropriate fd as its
455 value. We use SI_SIGIO as the source, not
456 SI_KERNEL, since kernel signals always get
457 delivered even if we can't queue. Failure to
458 queue in this case _should_ be reported; we fall
459 back to SIGIO in that case. --sct */
460 si.si_signo = signum;
461 si.si_errno = 0;
462 si.si_code = reason;
463 /* Make sure we are called with one of the POLL_*
464 reasons, otherwise we could leak kernel stack into
465 userspace. */
466 BUG_ON((reason & __SI_MASK) != __SI_POLL);
467 if (reason - POLL_IN >= NSIGPOLL)
468 si.si_band = ~0L;
469 else
470 si.si_band = band_table[reason - POLL_IN];
471 si.si_fd = fd;
472 if (!do_send_sig_info(signum, &si, p, group))
473 break;
474 /* fall-through: fall back on the old plain SIGIO signal */
475 case 0:
476 do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
480 void send_sigio(struct fown_struct *fown, int fd, int band)
482 struct task_struct *p;
483 enum pid_type type;
484 struct pid *pid;
485 int group = 1;
487 read_lock(&fown->lock);
489 type = fown->pid_type;
490 if (type == PIDTYPE_MAX) {
491 group = 0;
492 type = PIDTYPE_PID;
495 pid = fown->pid;
496 if (!pid)
497 goto out_unlock_fown;
499 read_lock(&tasklist_lock);
500 do_each_pid_task(pid, type, p) {
501 send_sigio_to_task(p, fown, fd, band, group);
502 } while_each_pid_task(pid, type, p);
503 read_unlock(&tasklist_lock);
504 out_unlock_fown:
505 read_unlock(&fown->lock);
508 static void send_sigurg_to_task(struct task_struct *p,
509 struct fown_struct *fown, int group)
511 if (sigio_perm(p, fown, SIGURG))
512 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
515 int send_sigurg(struct fown_struct *fown)
517 struct task_struct *p;
518 enum pid_type type;
519 struct pid *pid;
520 int group = 1;
521 int ret = 0;
523 read_lock(&fown->lock);
525 type = fown->pid_type;
526 if (type == PIDTYPE_MAX) {
527 group = 0;
528 type = PIDTYPE_PID;
531 pid = fown->pid;
532 if (!pid)
533 goto out_unlock_fown;
535 ret = 1;
537 read_lock(&tasklist_lock);
538 do_each_pid_task(pid, type, p) {
539 send_sigurg_to_task(p, fown, group);
540 } while_each_pid_task(pid, type, p);
541 read_unlock(&tasklist_lock);
542 out_unlock_fown:
543 read_unlock(&fown->lock);
544 return ret;
547 static DEFINE_SPINLOCK(fasync_lock);
548 static struct kmem_cache *fasync_cache __read_mostly;
550 static void fasync_free_rcu(struct rcu_head *head)
552 kmem_cache_free(fasync_cache,
553 container_of(head, struct fasync_struct, fa_rcu));
557 * Remove a fasync entry. If successfully removed, return
558 * positive and clear the FASYNC flag. If no entry exists,
559 * do nothing and return 0.
561 * NOTE! It is very important that the FASYNC flag always
562 * match the state "is the filp on a fasync list".
565 int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
567 struct fasync_struct *fa, **fp;
568 int result = 0;
570 spin_lock(&filp->f_lock);
571 spin_lock(&fasync_lock);
572 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
573 if (fa->fa_file != filp)
574 continue;
576 spin_lock_irq(&fa->fa_lock);
577 fa->fa_file = NULL;
578 spin_unlock_irq(&fa->fa_lock);
580 *fp = fa->fa_next;
581 call_rcu(&fa->fa_rcu, fasync_free_rcu);
582 filp->f_flags &= ~FASYNC;
583 result = 1;
584 break;
586 spin_unlock(&fasync_lock);
587 spin_unlock(&filp->f_lock);
588 return result;
591 struct fasync_struct *fasync_alloc(void)
593 return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
597 * NOTE! This can be used only for unused fasync entries:
598 * entries that actually got inserted on the fasync list
599 * need to be released by rcu - see fasync_remove_entry.
601 void fasync_free(struct fasync_struct *new)
603 kmem_cache_free(fasync_cache, new);
607 * Insert a new entry into the fasync list. Return the pointer to the
608 * old one if we didn't use the new one.
610 * NOTE! It is very important that the FASYNC flag always
611 * match the state "is the filp on a fasync list".
613 struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
615 struct fasync_struct *fa, **fp;
617 spin_lock(&filp->f_lock);
618 spin_lock(&fasync_lock);
619 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
620 if (fa->fa_file != filp)
621 continue;
623 spin_lock_irq(&fa->fa_lock);
624 fa->fa_fd = fd;
625 spin_unlock_irq(&fa->fa_lock);
626 goto out;
629 spin_lock_init(&new->fa_lock);
630 new->magic = FASYNC_MAGIC;
631 new->fa_file = filp;
632 new->fa_fd = fd;
633 new->fa_next = *fapp;
634 rcu_assign_pointer(*fapp, new);
635 filp->f_flags |= FASYNC;
637 out:
638 spin_unlock(&fasync_lock);
639 spin_unlock(&filp->f_lock);
640 return fa;
644 * Add a fasync entry. Return negative on error, positive if
645 * added, and zero if did nothing but change an existing one.
647 static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
649 struct fasync_struct *new;
651 new = fasync_alloc();
652 if (!new)
653 return -ENOMEM;
656 * fasync_insert_entry() returns the old (update) entry if
657 * it existed.
659 * So free the (unused) new entry and return 0 to let the
660 * caller know that we didn't add any new fasync entries.
662 if (fasync_insert_entry(fd, filp, fapp, new)) {
663 fasync_free(new);
664 return 0;
667 return 1;
671 * fasync_helper() is used by almost all character device drivers
672 * to set up the fasync queue, and for regular files by the file
673 * lease code. It returns negative on error, 0 if it did no changes
674 * and positive if it added/deleted the entry.
676 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
678 if (!on)
679 return fasync_remove_entry(filp, fapp);
680 return fasync_add_entry(fd, filp, fapp);
683 EXPORT_SYMBOL(fasync_helper);
686 * rcu_read_lock() is held
688 static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
690 while (fa) {
691 struct fown_struct *fown;
692 unsigned long flags;
694 if (fa->magic != FASYNC_MAGIC) {
695 printk(KERN_ERR "kill_fasync: bad magic number in "
696 "fasync_struct!\n");
697 return;
699 spin_lock_irqsave(&fa->fa_lock, flags);
700 if (fa->fa_file) {
701 fown = &fa->fa_file->f_owner;
702 /* Don't send SIGURG to processes which have not set a
703 queued signum: SIGURG has its own default signalling
704 mechanism. */
705 if (!(sig == SIGURG && fown->signum == 0))
706 send_sigio(fown, fa->fa_fd, band);
708 spin_unlock_irqrestore(&fa->fa_lock, flags);
709 fa = rcu_dereference(fa->fa_next);
713 void kill_fasync(struct fasync_struct **fp, int sig, int band)
715 /* First a quick test without locking: usually
716 * the list is empty.
718 if (*fp) {
719 rcu_read_lock();
720 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
721 rcu_read_unlock();
724 EXPORT_SYMBOL(kill_fasync);
726 static int __init fcntl_init(void)
729 * Please add new bits here to ensure allocation uniqueness.
730 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
731 * is defined as O_NONBLOCK on some platforms and not on others.
733 BUILD_BUG_ON(19 - 1 /* for O_RDONLY being 0 */ != HWEIGHT32(
734 O_RDONLY | O_WRONLY | O_RDWR |
735 O_CREAT | O_EXCL | O_NOCTTY |
736 O_TRUNC | O_APPEND | /* O_NONBLOCK | */
737 __O_SYNC | O_DSYNC | FASYNC |
738 O_DIRECT | O_LARGEFILE | O_DIRECTORY |
739 O_NOFOLLOW | O_NOATIME | O_CLOEXEC |
740 __FMODE_EXEC | O_PATH
743 fasync_cache = kmem_cache_create("fasync_cache",
744 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
745 return 0;
748 module_init(fcntl_init)