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>
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;
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))
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))
48 /* required for strict SunOS emulation */
49 if (O_NONBLOCK != O_NDELAY)
54 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
55 !filp->f_mapping->a_ops->direct_IO)
59 if (filp->f_op && filp->f_op->check_flags)
60 error = filp->f_op->check_flags(arg);
65 * ->fasync() is responsible for setting the FASYNC bit.
67 if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op &&
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 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
106 err = security_file_set_fowner(filp);
110 f_modown(filp, pid, type, force);
113 EXPORT_SYMBOL(__f_setown);
115 int f_setown(struct file *filp, unsigned long arg, int force)
127 pid = find_vpid(who);
128 result = __f_setown(filp, pid, type, force);
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)
142 read_lock(&filp->f_owner.lock);
143 pid = pid_vnr(filp->f_owner.pid);
144 if (filp->f_owner.pid_type == PIDTYPE_PGID)
146 read_unlock(&filp->f_owner.lock);
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;
158 ret = copy_from_user(&owner, owner_p, sizeof(owner));
162 switch (owner.type) {
180 pid = find_vpid(owner.pid);
181 if (owner.pid && !pid)
184 ret = __f_setown(filp, pid, type, 1);
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;
196 read_lock(&filp->f_owner.lock);
197 owner.pid = pid_vnr(filp->f_owner.pid);
198 switch (filp->f_owner.pid_type) {
200 owner.type = F_OWNER_TID;
204 owner.type = F_OWNER_PID;
208 owner.type = F_OWNER_PGRP;
216 read_unlock(&filp->f_owner.lock);
219 ret = copy_to_user(owner_p, &owner, sizeof(owner));
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;
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]);
245 static int f_getowner_uids(struct file *filp, unsigned long arg)
251 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
258 err = f_dupfd(arg, filp, 0);
260 case F_DUPFD_CLOEXEC:
261 err = f_dupfd(arg, filp, FD_CLOEXEC);
264 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
268 set_close_on_exec(fd, arg & FD_CLOEXEC);
274 err = setfl(fd, filp, arg);
277 err = fcntl_getlk(filp, (struct flock __user *) arg);
281 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
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();
295 err = f_setown(filp, arg, 1);
298 err = f_getown_ex(filp, arg);
301 err = f_setown_ex(filp, arg);
303 case F_GETOWNER_UIDS:
304 err = f_getowner_uids(filp, arg);
307 err = filp->f_owner.signum;
310 /* arg == 0 restores default behaviour. */
311 if (!valid_signal(arg)) {
315 filp->f_owner.signum = arg;
318 err = fcntl_getlease(filp);
321 err = fcntl_setlease(fd, filp, arg);
324 err = fcntl_dirnotify(fd, filp, arg);
328 err = pipe_fcntl(filp, cmd, arg);
336 static int check_fcntl_cmd(unsigned cmd)
340 case F_DUPFD_CLOEXEC:
349 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
355 filp = fget_raw_light(fd, &fput_needed);
359 if (unlikely(filp->f_mode & FMODE_PATH)) {
360 if (!check_fcntl_cmd(cmd))
364 err = security_file_fcntl(filp, cmd, arg);
366 err = do_fcntl(fd, cmd, arg, filp);
369 fput_light(filp, fput_needed);
374 #if BITS_PER_LONG == 32
375 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
382 filp = fget_raw_light(fd, &fput_needed);
386 if (unlikely(filp->f_mode & FMODE_PATH)) {
387 if (!check_fcntl_cmd(cmd))
391 err = security_file_fcntl(filp, cmd, arg);
397 err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
401 err = fcntl_setlk64(fd, filp, cmd,
402 (struct flock64 __user *) arg);
405 err = do_fcntl(fd, cmd, arg, filp);
409 fput_light(filp, fput_needed);
415 /* Table to convert sigio signal codes into poll band bitmaps */
417 static const long band_table[NSIGPOLL] = {
418 POLLIN | POLLRDNORM, /* POLL_IN */
419 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
420 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
421 POLLERR, /* POLL_ERR */
422 POLLPRI | POLLRDBAND, /* POLL_PRI */
423 POLLHUP | POLLERR /* POLL_HUP */
426 static inline int sigio_perm(struct task_struct *p,
427 struct fown_struct *fown, int sig)
429 const struct cred *cred;
433 cred = __task_cred(p);
434 ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
435 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
436 uid_eq(fown->uid, cred->suid) || uid_eq(fown->uid, cred->uid)) &&
437 !security_file_send_sigiotask(p, fown, sig));
442 static void send_sigio_to_task(struct task_struct *p,
443 struct fown_struct *fown,
444 int fd, int reason, int group)
447 * F_SETSIG can change ->signum lockless in parallel, make
448 * sure we read it once and use the same value throughout.
450 int signum = ACCESS_ONCE(fown->signum);
452 if (!sigio_perm(p, fown, signum))
458 /* Queue a rt signal with the appropriate fd as its
459 value. We use SI_SIGIO as the source, not
460 SI_KERNEL, since kernel signals always get
461 delivered even if we can't queue. Failure to
462 queue in this case _should_ be reported; we fall
463 back to SIGIO in that case. --sct */
464 si.si_signo = signum;
467 /* Make sure we are called with one of the POLL_*
468 reasons, otherwise we could leak kernel stack into
470 BUG_ON((reason & __SI_MASK) != __SI_POLL);
471 if (reason - POLL_IN >= NSIGPOLL)
474 si.si_band = band_table[reason - POLL_IN];
476 if (!do_send_sig_info(signum, &si, p, group))
478 /* fall-through: fall back on the old plain SIGIO signal */
480 do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
484 void send_sigio(struct fown_struct *fown, int fd, int band)
486 struct task_struct *p;
491 read_lock(&fown->lock);
493 type = fown->pid_type;
494 if (type == PIDTYPE_MAX) {
501 goto out_unlock_fown;
503 read_lock(&tasklist_lock);
504 do_each_pid_task(pid, type, p) {
505 send_sigio_to_task(p, fown, fd, band, group);
506 } while_each_pid_task(pid, type, p);
507 read_unlock(&tasklist_lock);
509 read_unlock(&fown->lock);
512 static void send_sigurg_to_task(struct task_struct *p,
513 struct fown_struct *fown, int group)
515 if (sigio_perm(p, fown, SIGURG))
516 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
519 int send_sigurg(struct fown_struct *fown)
521 struct task_struct *p;
527 read_lock(&fown->lock);
529 type = fown->pid_type;
530 if (type == PIDTYPE_MAX) {
537 goto out_unlock_fown;
541 read_lock(&tasklist_lock);
542 do_each_pid_task(pid, type, p) {
543 send_sigurg_to_task(p, fown, group);
544 } while_each_pid_task(pid, type, p);
545 read_unlock(&tasklist_lock);
547 read_unlock(&fown->lock);
551 static DEFINE_SPINLOCK(fasync_lock);
552 static struct kmem_cache *fasync_cache __read_mostly;
554 static void fasync_free_rcu(struct rcu_head *head)
556 kmem_cache_free(fasync_cache,
557 container_of(head, struct fasync_struct, fa_rcu));
561 * Remove a fasync entry. If successfully removed, return
562 * positive and clear the FASYNC flag. If no entry exists,
563 * do nothing and return 0.
565 * NOTE! It is very important that the FASYNC flag always
566 * match the state "is the filp on a fasync list".
569 int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
571 struct fasync_struct *fa, **fp;
574 spin_lock(&filp->f_lock);
575 spin_lock(&fasync_lock);
576 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
577 if (fa->fa_file != filp)
580 spin_lock_irq(&fa->fa_lock);
582 spin_unlock_irq(&fa->fa_lock);
585 call_rcu(&fa->fa_rcu, fasync_free_rcu);
586 filp->f_flags &= ~FASYNC;
590 spin_unlock(&fasync_lock);
591 spin_unlock(&filp->f_lock);
595 struct fasync_struct *fasync_alloc(void)
597 return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
601 * NOTE! This can be used only for unused fasync entries:
602 * entries that actually got inserted on the fasync list
603 * need to be released by rcu - see fasync_remove_entry.
605 void fasync_free(struct fasync_struct *new)
607 kmem_cache_free(fasync_cache, new);
611 * Insert a new entry into the fasync list. Return the pointer to the
612 * old one if we didn't use the new one.
614 * NOTE! It is very important that the FASYNC flag always
615 * match the state "is the filp on a fasync list".
617 struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
619 struct fasync_struct *fa, **fp;
621 spin_lock(&filp->f_lock);
622 spin_lock(&fasync_lock);
623 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
624 if (fa->fa_file != filp)
627 spin_lock_irq(&fa->fa_lock);
629 spin_unlock_irq(&fa->fa_lock);
633 spin_lock_init(&new->fa_lock);
634 new->magic = FASYNC_MAGIC;
637 new->fa_next = *fapp;
638 rcu_assign_pointer(*fapp, new);
639 filp->f_flags |= FASYNC;
642 spin_unlock(&fasync_lock);
643 spin_unlock(&filp->f_lock);
648 * Add a fasync entry. Return negative on error, positive if
649 * added, and zero if did nothing but change an existing one.
651 static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
653 struct fasync_struct *new;
655 new = fasync_alloc();
660 * fasync_insert_entry() returns the old (update) entry if
663 * So free the (unused) new entry and return 0 to let the
664 * caller know that we didn't add any new fasync entries.
666 if (fasync_insert_entry(fd, filp, fapp, new)) {
675 * fasync_helper() is used by almost all character device drivers
676 * to set up the fasync queue, and for regular files by the file
677 * lease code. It returns negative on error, 0 if it did no changes
678 * and positive if it added/deleted the entry.
680 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
683 return fasync_remove_entry(filp, fapp);
684 return fasync_add_entry(fd, filp, fapp);
687 EXPORT_SYMBOL(fasync_helper);
690 * rcu_read_lock() is held
692 static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
695 struct fown_struct *fown;
698 if (fa->magic != FASYNC_MAGIC) {
699 printk(KERN_ERR "kill_fasync: bad magic number in "
703 spin_lock_irqsave(&fa->fa_lock, flags);
705 fown = &fa->fa_file->f_owner;
706 /* Don't send SIGURG to processes which have not set a
707 queued signum: SIGURG has its own default signalling
709 if (!(sig == SIGURG && fown->signum == 0))
710 send_sigio(fown, fa->fa_fd, band);
712 spin_unlock_irqrestore(&fa->fa_lock, flags);
713 fa = rcu_dereference(fa->fa_next);
717 void kill_fasync(struct fasync_struct **fp, int sig, int band)
719 /* First a quick test without locking: usually
724 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
728 EXPORT_SYMBOL(kill_fasync);
730 static int __init fcntl_init(void)
733 * Please add new bits here to ensure allocation uniqueness.
734 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
735 * is defined as O_NONBLOCK on some platforms and not on others.
737 BUILD_BUG_ON(19 - 1 /* for O_RDONLY being 0 */ != HWEIGHT32(
738 O_RDONLY | O_WRONLY | O_RDWR |
739 O_CREAT | O_EXCL | O_NOCTTY |
740 O_TRUNC | O_APPEND | /* O_NONBLOCK | */
741 __O_SYNC | O_DSYNC | FASYNC |
742 O_DIRECT | O_LARGEFILE | O_DIRECTORY |
743 O_NOFOLLOW | O_NOATIME | O_CLOEXEC |
744 __FMODE_EXEC | O_PATH
747 fasync_cache = kmem_cache_create("fasync_cache",
748 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
752 module_init(fcntl_init)