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>
25 #include <asm/siginfo.h>
26 #include <asm/uaccess.h>
28 void set_close_on_exec(unsigned int fd, int flag)
30 struct files_struct *files = current->files;
32 spin_lock(&files->file_lock);
33 fdt = files_fdtable(files);
35 FD_SET(fd, fdt->close_on_exec);
37 FD_CLR(fd, fdt->close_on_exec);
38 spin_unlock(&files->file_lock);
41 static int get_close_on_exec(unsigned int fd)
43 struct files_struct *files = current->files;
47 fdt = files_fdtable(files);
48 res = FD_ISSET(fd, fdt->close_on_exec);
53 SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
56 struct file * file, *tofree;
57 struct files_struct * files = current->files;
60 if ((flags & ~O_CLOEXEC) != 0)
63 if (unlikely(oldfd == newfd))
66 spin_lock(&files->file_lock);
67 err = expand_files(files, newfd);
71 if (unlikely(err < 0)) {
77 * We need to detect attempts to do dup2() over allocated but still
78 * not finished descriptor. NB: OpenBSD avoids that at the price of
79 * extra work in their equivalent of fget() - they insert struct
80 * file immediately after grabbing descriptor, mark it larval if
81 * more work (e.g. actual opening) is needed and make sure that
82 * fget() treats larval files as absent. Potentially interesting,
83 * but while extra work in fget() is trivial, locking implications
84 * and amount of surgery on open()-related paths in VFS are not.
85 * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
86 * deadlocks in rather amusing ways, AFAICS. All of that is out of
87 * scope of POSIX or SUS, since neither considers shared descriptor
88 * tables and this condition does not arise without those.
91 fdt = files_fdtable(files);
92 tofree = fdt->fd[newfd];
93 if (!tofree && FD_ISSET(newfd, fdt->open_fds))
96 rcu_assign_pointer(fdt->fd[newfd], file);
97 FD_SET(newfd, fdt->open_fds);
98 if (flags & O_CLOEXEC)
99 FD_SET(newfd, fdt->close_on_exec);
101 FD_CLR(newfd, fdt->close_on_exec);
102 spin_unlock(&files->file_lock);
105 filp_close(tofree, files);
112 spin_unlock(&files->file_lock);
116 SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
118 if (unlikely(newfd == oldfd)) { /* corner case */
119 struct files_struct *files = current->files;
123 if (!fcheck_files(files, oldfd))
128 return sys_dup3(oldfd, newfd, 0);
131 SYSCALL_DEFINE1(dup, unsigned int, fildes)
134 struct file *file = fget_raw(fildes);
137 ret = get_unused_fd();
139 fd_install(ret, file);
146 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
148 static int setfl(int fd, struct file * filp, unsigned long arg)
150 struct inode * inode = filp->f_path.dentry->d_inode;
154 * O_APPEND cannot be cleared if the file is marked as append-only
155 * and the file is open for write.
157 if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
160 /* O_NOATIME can only be set by the owner or superuser */
161 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
162 if (!inode_owner_or_capable(inode))
165 /* required for strict SunOS emulation */
166 if (O_NONBLOCK != O_NDELAY)
170 if (arg & O_DIRECT) {
171 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
172 !filp->f_mapping->a_ops->direct_IO)
176 if (filp->f_op && filp->f_op->check_flags)
177 error = filp->f_op->check_flags(arg);
182 * ->fasync() is responsible for setting the FASYNC bit.
184 if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op &&
185 filp->f_op->fasync) {
186 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
192 spin_lock(&filp->f_lock);
193 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
194 spin_unlock(&filp->f_lock);
200 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
203 write_lock_irq(&filp->f_owner.lock);
204 if (force || !filp->f_owner.pid) {
205 put_pid(filp->f_owner.pid);
206 filp->f_owner.pid = get_pid(pid);
207 filp->f_owner.pid_type = type;
210 const struct cred *cred = current_cred();
211 filp->f_owner.uid = cred->uid;
212 filp->f_owner.euid = cred->euid;
215 write_unlock_irq(&filp->f_owner.lock);
218 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
223 err = security_file_set_fowner(filp);
227 f_modown(filp, pid, type, force);
230 EXPORT_SYMBOL(__f_setown);
232 int f_setown(struct file *filp, unsigned long arg, int force)
244 pid = find_vpid(who);
245 result = __f_setown(filp, pid, type, force);
249 EXPORT_SYMBOL(f_setown);
251 void f_delown(struct file *filp)
253 f_modown(filp, NULL, PIDTYPE_PID, 1);
256 pid_t f_getown(struct file *filp)
259 read_lock(&filp->f_owner.lock);
260 pid = pid_vnr(filp->f_owner.pid);
261 if (filp->f_owner.pid_type == PIDTYPE_PGID)
263 read_unlock(&filp->f_owner.lock);
267 static int f_setown_ex(struct file *filp, unsigned long arg)
269 struct f_owner_ex * __user owner_p = (void * __user)arg;
270 struct f_owner_ex owner;
275 ret = copy_from_user(&owner, owner_p, sizeof(owner));
279 switch (owner.type) {
297 pid = find_vpid(owner.pid);
298 if (owner.pid && !pid)
301 ret = __f_setown(filp, pid, type, 1);
307 static int f_getown_ex(struct file *filp, unsigned long arg)
309 struct f_owner_ex * __user owner_p = (void * __user)arg;
310 struct f_owner_ex owner;
313 read_lock(&filp->f_owner.lock);
314 owner.pid = pid_vnr(filp->f_owner.pid);
315 switch (filp->f_owner.pid_type) {
317 owner.type = F_OWNER_TID;
321 owner.type = F_OWNER_PID;
325 owner.type = F_OWNER_PGRP;
333 read_unlock(&filp->f_owner.lock);
336 ret = copy_to_user(owner_p, &owner, sizeof(owner));
343 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
350 case F_DUPFD_CLOEXEC:
351 if (arg >= rlimit(RLIMIT_NOFILE))
353 err = alloc_fd(arg, cmd == F_DUPFD_CLOEXEC ? O_CLOEXEC : 0);
356 fd_install(err, filp);
360 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
364 set_close_on_exec(fd, arg & FD_CLOEXEC);
370 err = setfl(fd, filp, arg);
373 err = fcntl_getlk(filp, (struct flock __user *) arg);
377 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
381 * XXX If f_owner is a process group, the
382 * negative return value will get converted
383 * into an error. Oops. If we keep the
384 * current syscall conventions, the only way
385 * to fix this will be in libc.
387 err = f_getown(filp);
388 force_successful_syscall_return();
391 err = f_setown(filp, arg, 1);
394 err = f_getown_ex(filp, arg);
397 err = f_setown_ex(filp, arg);
400 err = filp->f_owner.signum;
403 /* arg == 0 restores default behaviour. */
404 if (!valid_signal(arg)) {
408 filp->f_owner.signum = arg;
411 err = fcntl_getlease(filp);
414 err = fcntl_setlease(fd, filp, arg);
417 err = fcntl_dirnotify(fd, filp, arg);
421 err = pipe_fcntl(filp, cmd, arg);
429 static int check_fcntl_cmd(unsigned cmd)
433 case F_DUPFD_CLOEXEC:
442 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
451 if (unlikely(filp->f_mode & FMODE_PATH)) {
452 if (!check_fcntl_cmd(cmd)) {
458 err = security_file_fcntl(filp, cmd, arg);
464 err = do_fcntl(fd, cmd, arg, filp);
471 #if BITS_PER_LONG == 32
472 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
483 if (unlikely(filp->f_mode & FMODE_PATH)) {
484 if (!check_fcntl_cmd(cmd)) {
490 err = security_file_fcntl(filp, cmd, arg);
499 err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
503 err = fcntl_setlk64(fd, filp, cmd,
504 (struct flock64 __user *) arg);
507 err = do_fcntl(fd, cmd, arg, filp);
516 /* Table to convert sigio signal codes into poll band bitmaps */
518 static const long band_table[NSIGPOLL] = {
519 POLLIN | POLLRDNORM, /* POLL_IN */
520 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
521 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
522 POLLERR, /* POLL_ERR */
523 POLLPRI | POLLRDBAND, /* POLL_PRI */
524 POLLHUP | POLLERR /* POLL_HUP */
527 static inline int sigio_perm(struct task_struct *p,
528 struct fown_struct *fown, int sig)
530 const struct cred *cred;
534 cred = __task_cred(p);
535 ret = ((fown->euid == 0 ||
536 fown->euid == cred->suid || fown->euid == cred->uid ||
537 fown->uid == cred->suid || fown->uid == cred->uid) &&
538 !security_file_send_sigiotask(p, fown, sig));
543 static void send_sigio_to_task(struct task_struct *p,
544 struct fown_struct *fown,
545 int fd, int reason, int group)
548 * F_SETSIG can change ->signum lockless in parallel, make
549 * sure we read it once and use the same value throughout.
551 int signum = ACCESS_ONCE(fown->signum);
553 if (!sigio_perm(p, fown, signum))
559 /* Queue a rt signal with the appropriate fd as its
560 value. We use SI_SIGIO as the source, not
561 SI_KERNEL, since kernel signals always get
562 delivered even if we can't queue. Failure to
563 queue in this case _should_ be reported; we fall
564 back to SIGIO in that case. --sct */
565 si.si_signo = signum;
569 * Posix definies POLL_IN and friends to be signal
570 * specific si_codes for SIG_POLL. Linux extended
571 * these si_codes to other signals in a way that is
572 * ambiguous if other signals also have signal
573 * specific si_codes. In that case use SI_SIGIO instead
574 * to remove the ambiguity.
576 if (sig_specific_sicodes(signum))
577 si.si_code = SI_SIGIO;
579 /* Make sure we are called with one of the POLL_*
580 reasons, otherwise we could leak kernel stack into
582 BUG_ON((reason < POLL_IN) || ((reason - POLL_IN) >= NSIGPOLL));
583 if (reason - POLL_IN >= NSIGPOLL)
586 si.si_band = band_table[reason - POLL_IN];
588 if (!do_send_sig_info(signum, &si, p, group))
590 /* fall-through: fall back on the old plain SIGIO signal */
592 do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
596 void send_sigio(struct fown_struct *fown, int fd, int band)
598 struct task_struct *p;
603 read_lock(&fown->lock);
605 type = fown->pid_type;
606 if (type == PIDTYPE_MAX) {
613 goto out_unlock_fown;
615 read_lock(&tasklist_lock);
616 do_each_pid_task(pid, type, p) {
617 send_sigio_to_task(p, fown, fd, band, group);
618 } while_each_pid_task(pid, type, p);
619 read_unlock(&tasklist_lock);
621 read_unlock(&fown->lock);
624 static void send_sigurg_to_task(struct task_struct *p,
625 struct fown_struct *fown, int group)
627 if (sigio_perm(p, fown, SIGURG))
628 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
631 int send_sigurg(struct fown_struct *fown)
633 struct task_struct *p;
639 read_lock(&fown->lock);
641 type = fown->pid_type;
642 if (type == PIDTYPE_MAX) {
649 goto out_unlock_fown;
653 read_lock(&tasklist_lock);
654 do_each_pid_task(pid, type, p) {
655 send_sigurg_to_task(p, fown, group);
656 } while_each_pid_task(pid, type, p);
657 read_unlock(&tasklist_lock);
659 read_unlock(&fown->lock);
663 static DEFINE_SPINLOCK(fasync_lock);
664 static struct kmem_cache *fasync_cache __read_mostly;
666 static void fasync_free_rcu(struct rcu_head *head)
668 kmem_cache_free(fasync_cache,
669 container_of(head, struct fasync_struct, fa_rcu));
673 * Remove a fasync entry. If successfully removed, return
674 * positive and clear the FASYNC flag. If no entry exists,
675 * do nothing and return 0.
677 * NOTE! It is very important that the FASYNC flag always
678 * match the state "is the filp on a fasync list".
681 int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
683 struct fasync_struct *fa, **fp;
686 spin_lock(&filp->f_lock);
687 spin_lock(&fasync_lock);
688 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
689 if (fa->fa_file != filp)
692 spin_lock_irq(&fa->fa_lock);
694 spin_unlock_irq(&fa->fa_lock);
697 call_rcu(&fa->fa_rcu, fasync_free_rcu);
698 filp->f_flags &= ~FASYNC;
702 spin_unlock(&fasync_lock);
703 spin_unlock(&filp->f_lock);
707 struct fasync_struct *fasync_alloc(void)
709 return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
713 * NOTE! This can be used only for unused fasync entries:
714 * entries that actually got inserted on the fasync list
715 * need to be released by rcu - see fasync_remove_entry.
717 void fasync_free(struct fasync_struct *new)
719 kmem_cache_free(fasync_cache, new);
723 * Insert a new entry into the fasync list. Return the pointer to the
724 * old one if we didn't use the new one.
726 * NOTE! It is very important that the FASYNC flag always
727 * match the state "is the filp on a fasync list".
729 struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
731 struct fasync_struct *fa, **fp;
733 spin_lock(&filp->f_lock);
734 spin_lock(&fasync_lock);
735 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
736 if (fa->fa_file != filp)
739 spin_lock_irq(&fa->fa_lock);
741 spin_unlock_irq(&fa->fa_lock);
745 spin_lock_init(&new->fa_lock);
746 new->magic = FASYNC_MAGIC;
749 new->fa_next = *fapp;
750 rcu_assign_pointer(*fapp, new);
751 filp->f_flags |= FASYNC;
754 spin_unlock(&fasync_lock);
755 spin_unlock(&filp->f_lock);
760 * Add a fasync entry. Return negative on error, positive if
761 * added, and zero if did nothing but change an existing one.
763 static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
765 struct fasync_struct *new;
767 new = fasync_alloc();
772 * fasync_insert_entry() returns the old (update) entry if
775 * So free the (unused) new entry and return 0 to let the
776 * caller know that we didn't add any new fasync entries.
778 if (fasync_insert_entry(fd, filp, fapp, new)) {
787 * fasync_helper() is used by almost all character device drivers
788 * to set up the fasync queue, and for regular files by the file
789 * lease code. It returns negative on error, 0 if it did no changes
790 * and positive if it added/deleted the entry.
792 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
795 return fasync_remove_entry(filp, fapp);
796 return fasync_add_entry(fd, filp, fapp);
799 EXPORT_SYMBOL(fasync_helper);
802 * rcu_read_lock() is held
804 static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
807 struct fown_struct *fown;
810 if (fa->magic != FASYNC_MAGIC) {
811 printk(KERN_ERR "kill_fasync: bad magic number in "
815 spin_lock_irqsave(&fa->fa_lock, flags);
817 fown = &fa->fa_file->f_owner;
818 /* Don't send SIGURG to processes which have not set a
819 queued signum: SIGURG has its own default signalling
821 if (!(sig == SIGURG && fown->signum == 0))
822 send_sigio(fown, fa->fa_fd, band);
824 spin_unlock_irqrestore(&fa->fa_lock, flags);
825 fa = rcu_dereference(fa->fa_next);
829 void kill_fasync(struct fasync_struct **fp, int sig, int band)
831 /* First a quick test without locking: usually
836 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
840 EXPORT_SYMBOL(kill_fasync);
842 static int __init fcntl_init(void)
845 * Please add new bits here to ensure allocation uniqueness.
846 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
847 * is defined as O_NONBLOCK on some platforms and not on others.
849 BUILD_BUG_ON(19 - 1 /* for O_RDONLY being 0 */ != HWEIGHT32(
850 O_RDONLY | O_WRONLY | O_RDWR |
851 O_CREAT | O_EXCL | O_NOCTTY |
852 O_TRUNC | O_APPEND | /* O_NONBLOCK | */
853 __O_SYNC | O_DSYNC | FASYNC |
854 O_DIRECT | O_LARGEFILE | O_DIRECTORY |
855 O_NOFOLLOW | O_NOATIME | O_CLOEXEC |
856 __FMODE_EXEC | O_PATH
859 fasync_cache = kmem_cache_create("fasync_cache",
860 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
864 module_init(fcntl_init)