4 * Copyright (C) 1991, 1992 Linus Torvalds
7 #include <linux/syscalls.h>
8 #include <linux/init.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>
24 #include <linux/shmem_fs.h>
27 #include <asm/siginfo.h>
28 #include <asm/uaccess.h>
30 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
32 static int setfl(int fd, struct file * filp, unsigned long arg)
34 struct inode * inode = file_inode(filp);
38 * O_APPEND cannot be cleared if the file is marked as append-only
39 * and the file is open for write.
41 if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
44 /* O_NOATIME can only be set by the owner or superuser */
45 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
46 if (!inode_owner_or_capable(inode))
49 /* required for strict SunOS emulation */
50 if (O_NONBLOCK != O_NDELAY)
55 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
56 !filp->f_mapping->a_ops->direct_IO)
60 if (filp->f_op->check_flags)
61 error = filp->f_op->check_flags(arg);
66 * ->fasync() is responsible for setting the FASYNC bit.
68 if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op->fasync) {
69 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 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);
83 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
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;
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 void __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
104 security_file_set_fowner(filp);
105 f_modown(filp, pid, type, force);
107 EXPORT_SYMBOL(__f_setown);
109 void f_setown(struct file *filp, unsigned long arg, int force)
116 /* avoid overflow below */
124 pid = find_vpid(who);
125 __f_setown(filp, pid, type, force);
128 EXPORT_SYMBOL(f_setown);
130 void f_delown(struct file *filp)
132 f_modown(filp, NULL, PIDTYPE_PID, 1);
135 pid_t f_getown(struct file *filp)
138 read_lock(&filp->f_owner.lock);
139 pid = pid_vnr(filp->f_owner.pid);
140 if (filp->f_owner.pid_type == PIDTYPE_PGID)
142 read_unlock(&filp->f_owner.lock);
146 static int f_setown_ex(struct file *filp, unsigned long arg)
148 struct f_owner_ex __user *owner_p = (void __user *)arg;
149 struct f_owner_ex owner;
154 ret = copy_from_user(&owner, owner_p, sizeof(owner));
158 switch (owner.type) {
176 pid = find_vpid(owner.pid);
177 if (owner.pid && !pid)
180 __f_setown(filp, pid, type, 1);
186 static int f_getown_ex(struct file *filp, unsigned long arg)
188 struct f_owner_ex __user *owner_p = (void __user *)arg;
189 struct f_owner_ex owner;
192 read_lock(&filp->f_owner.lock);
193 owner.pid = pid_vnr(filp->f_owner.pid);
194 switch (filp->f_owner.pid_type) {
196 owner.type = F_OWNER_TID;
200 owner.type = F_OWNER_PID;
204 owner.type = F_OWNER_PGRP;
212 read_unlock(&filp->f_owner.lock);
215 ret = copy_to_user(owner_p, &owner, sizeof(owner));
222 #ifdef CONFIG_CHECKPOINT_RESTORE
223 static int f_getowner_uids(struct file *filp, unsigned long arg)
225 struct user_namespace *user_ns = current_user_ns();
226 uid_t __user *dst = (void __user *)arg;
230 read_lock(&filp->f_owner.lock);
231 src[0] = from_kuid(user_ns, filp->f_owner.uid);
232 src[1] = from_kuid(user_ns, filp->f_owner.euid);
233 read_unlock(&filp->f_owner.lock);
235 err = put_user(src[0], &dst[0]);
236 err |= put_user(src[1], &dst[1]);
241 static int f_getowner_uids(struct file *filp, unsigned long arg)
247 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
254 err = f_dupfd(arg, filp, 0);
256 case F_DUPFD_CLOEXEC:
257 err = f_dupfd(arg, filp, O_CLOEXEC);
260 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
264 set_close_on_exec(fd, arg & FD_CLOEXEC);
270 err = setfl(fd, filp, arg);
272 #if BITS_PER_LONG != 32
273 /* 32-bit arches must use fcntl64() */
277 err = fcntl_getlk(filp, cmd, (struct flock __user *) arg);
279 #if BITS_PER_LONG != 32
280 /* 32-bit arches must use fcntl64() */
287 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
291 * XXX If f_owner is a process group, the
292 * negative return value will get converted
293 * into an error. Oops. If we keep the
294 * current syscall conventions, the only way
295 * to fix this will be in libc.
297 err = f_getown(filp);
298 force_successful_syscall_return();
301 f_setown(filp, arg, 1);
305 err = f_getown_ex(filp, arg);
308 err = f_setown_ex(filp, arg);
310 case F_GETOWNER_UIDS:
311 err = f_getowner_uids(filp, arg);
314 err = filp->f_owner.signum;
317 /* arg == 0 restores default behaviour. */
318 if (!valid_signal(arg)) {
322 filp->f_owner.signum = arg;
325 err = fcntl_getlease(filp);
328 err = fcntl_setlease(fd, filp, arg);
331 err = fcntl_dirnotify(fd, filp, arg);
335 err = pipe_fcntl(filp, cmd, arg);
339 err = shmem_fcntl(filp, cmd, arg);
347 static int check_fcntl_cmd(unsigned cmd)
351 case F_DUPFD_CLOEXEC:
360 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
362 struct fd f = fdget_raw(fd);
368 if (unlikely(f.file->f_mode & FMODE_PATH)) {
369 if (!check_fcntl_cmd(cmd))
373 err = security_file_fcntl(f.file, cmd, arg);
375 err = do_fcntl(fd, cmd, arg, f.file);
383 #if BITS_PER_LONG == 32
384 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
387 struct fd f = fdget_raw(fd);
393 if (unlikely(f.file->f_mode & FMODE_PATH)) {
394 if (!check_fcntl_cmd(cmd))
398 err = security_file_fcntl(f.file, cmd, arg);
405 err = fcntl_getlk64(f.file, cmd, (struct flock64 __user *) arg);
411 err = fcntl_setlk64(fd, f.file, cmd,
412 (struct flock64 __user *) arg);
415 err = do_fcntl(fd, cmd, arg, f.file);
425 /* Table to convert sigio signal codes into poll band bitmaps */
427 static const long band_table[NSIGPOLL] = {
428 POLLIN | POLLRDNORM, /* POLL_IN */
429 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
430 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
431 POLLERR, /* POLL_ERR */
432 POLLPRI | POLLRDBAND, /* POLL_PRI */
433 POLLHUP | POLLERR /* POLL_HUP */
436 static inline int sigio_perm(struct task_struct *p,
437 struct fown_struct *fown, int sig)
439 const struct cred *cred;
443 cred = __task_cred(p);
444 ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
445 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
446 uid_eq(fown->uid, cred->suid) || uid_eq(fown->uid, cred->uid)) &&
447 !security_file_send_sigiotask(p, fown, sig));
452 static void send_sigio_to_task(struct task_struct *p,
453 struct fown_struct *fown,
454 int fd, int reason, int group)
457 * F_SETSIG can change ->signum lockless in parallel, make
458 * sure we read it once and use the same value throughout.
460 int signum = ACCESS_ONCE(fown->signum);
462 if (!sigio_perm(p, fown, signum))
468 /* Queue a rt signal with the appropriate fd as its
469 value. We use SI_SIGIO as the source, not
470 SI_KERNEL, since kernel signals always get
471 delivered even if we can't queue. Failure to
472 queue in this case _should_ be reported; we fall
473 back to SIGIO in that case. --sct */
474 si.si_signo = signum;
477 /* Make sure we are called with one of the POLL_*
478 reasons, otherwise we could leak kernel stack into
480 BUG_ON((reason & __SI_MASK) != __SI_POLL);
481 if (reason - POLL_IN >= NSIGPOLL)
484 si.si_band = band_table[reason - POLL_IN];
486 if (!do_send_sig_info(signum, &si, p, group))
488 /* fall-through: fall back on the old plain SIGIO signal */
490 do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
494 void send_sigio(struct fown_struct *fown, int fd, int band)
496 struct task_struct *p;
501 read_lock(&fown->lock);
503 type = fown->pid_type;
504 if (type == PIDTYPE_MAX) {
511 goto out_unlock_fown;
513 read_lock(&tasklist_lock);
514 do_each_pid_task(pid, type, p) {
515 send_sigio_to_task(p, fown, fd, band, group);
516 } while_each_pid_task(pid, type, p);
517 read_unlock(&tasklist_lock);
519 read_unlock(&fown->lock);
522 static void send_sigurg_to_task(struct task_struct *p,
523 struct fown_struct *fown, int group)
525 if (sigio_perm(p, fown, SIGURG))
526 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
529 int send_sigurg(struct fown_struct *fown)
531 struct task_struct *p;
537 read_lock(&fown->lock);
539 type = fown->pid_type;
540 if (type == PIDTYPE_MAX) {
547 goto out_unlock_fown;
551 read_lock(&tasklist_lock);
552 do_each_pid_task(pid, type, p) {
553 send_sigurg_to_task(p, fown, group);
554 } while_each_pid_task(pid, type, p);
555 read_unlock(&tasklist_lock);
557 read_unlock(&fown->lock);
561 static DEFINE_SPINLOCK(fasync_lock);
562 static struct kmem_cache *fasync_cache __read_mostly;
564 static void fasync_free_rcu(struct rcu_head *head)
566 kmem_cache_free(fasync_cache,
567 container_of(head, struct fasync_struct, fa_rcu));
571 * Remove a fasync entry. If successfully removed, return
572 * positive and clear the FASYNC flag. If no entry exists,
573 * do nothing and return 0.
575 * NOTE! It is very important that the FASYNC flag always
576 * match the state "is the filp on a fasync list".
579 int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
581 struct fasync_struct *fa, **fp;
584 spin_lock(&filp->f_lock);
585 spin_lock(&fasync_lock);
586 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
587 if (fa->fa_file != filp)
590 spin_lock_irq(&fa->fa_lock);
592 spin_unlock_irq(&fa->fa_lock);
595 call_rcu(&fa->fa_rcu, fasync_free_rcu);
596 filp->f_flags &= ~FASYNC;
600 spin_unlock(&fasync_lock);
601 spin_unlock(&filp->f_lock);
605 struct fasync_struct *fasync_alloc(void)
607 return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
611 * NOTE! This can be used only for unused fasync entries:
612 * entries that actually got inserted on the fasync list
613 * need to be released by rcu - see fasync_remove_entry.
615 void fasync_free(struct fasync_struct *new)
617 kmem_cache_free(fasync_cache, new);
621 * Insert a new entry into the fasync list. Return the pointer to the
622 * old one if we didn't use the new one.
624 * NOTE! It is very important that the FASYNC flag always
625 * match the state "is the filp on a fasync list".
627 struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
629 struct fasync_struct *fa, **fp;
631 spin_lock(&filp->f_lock);
632 spin_lock(&fasync_lock);
633 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
634 if (fa->fa_file != filp)
637 spin_lock_irq(&fa->fa_lock);
639 spin_unlock_irq(&fa->fa_lock);
643 spin_lock_init(&new->fa_lock);
644 new->magic = FASYNC_MAGIC;
647 new->fa_next = *fapp;
648 rcu_assign_pointer(*fapp, new);
649 filp->f_flags |= FASYNC;
652 spin_unlock(&fasync_lock);
653 spin_unlock(&filp->f_lock);
658 * Add a fasync entry. Return negative on error, positive if
659 * added, and zero if did nothing but change an existing one.
661 static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
663 struct fasync_struct *new;
665 new = fasync_alloc();
670 * fasync_insert_entry() returns the old (update) entry if
673 * So free the (unused) new entry and return 0 to let the
674 * caller know that we didn't add any new fasync entries.
676 if (fasync_insert_entry(fd, filp, fapp, new)) {
685 * fasync_helper() is used by almost all character device drivers
686 * to set up the fasync queue, and for regular files by the file
687 * lease code. It returns negative on error, 0 if it did no changes
688 * and positive if it added/deleted the entry.
690 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
693 return fasync_remove_entry(filp, fapp);
694 return fasync_add_entry(fd, filp, fapp);
697 EXPORT_SYMBOL(fasync_helper);
700 * rcu_read_lock() is held
702 static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
705 struct fown_struct *fown;
708 if (fa->magic != FASYNC_MAGIC) {
709 printk(KERN_ERR "kill_fasync: bad magic number in "
713 spin_lock_irqsave(&fa->fa_lock, flags);
715 fown = &fa->fa_file->f_owner;
716 /* Don't send SIGURG to processes which have not set a
717 queued signum: SIGURG has its own default signalling
719 if (!(sig == SIGURG && fown->signum == 0))
720 send_sigio(fown, fa->fa_fd, band);
722 spin_unlock_irqrestore(&fa->fa_lock, flags);
723 fa = rcu_dereference(fa->fa_next);
727 void kill_fasync(struct fasync_struct **fp, int sig, int band)
729 /* First a quick test without locking: usually
734 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
738 EXPORT_SYMBOL(kill_fasync);
740 static int __init fcntl_init(void)
743 * Please add new bits here to ensure allocation uniqueness.
744 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
745 * is defined as O_NONBLOCK on some platforms and not on others.
747 BUILD_BUG_ON(21 - 1 /* for O_RDONLY being 0 */ !=
749 (VALID_OPEN_FLAGS & ~(O_NONBLOCK | O_NDELAY)) |
750 __FMODE_EXEC | __FMODE_NONOTIFY));
752 fasync_cache = kmem_cache_create("fasync_cache",
753 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
757 module_init(fcntl_init)