4 * Copyright (C) 1991, 1992 Linus Torvalds
7 #include <linux/syscalls.h>
8 #include <linux/init.h>
10 #include <linux/sched/task.h>
12 #include <linux/file.h>
13 #include <linux/fdtable.h>
14 #include <linux/capability.h>
15 #include <linux/dnotify.h>
16 #include <linux/slab.h>
17 #include <linux/module.h>
18 #include <linux/pipe_fs_i.h>
19 #include <linux/security.h>
20 #include <linux/ptrace.h>
21 #include <linux/signal.h>
22 #include <linux/rcupdate.h>
23 #include <linux/pid_namespace.h>
24 #include <linux/user_namespace.h>
25 #include <linux/shmem_fs.h>
26 #include <linux/compat.h>
29 #include <asm/siginfo.h>
30 #include <linux/uaccess.h>
32 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
34 static int setfl(int fd, struct file * filp, unsigned long arg)
36 struct inode * inode = file_inode(filp);
40 * O_APPEND cannot be cleared if the file is marked as append-only
41 * and the file is open for write.
43 if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
46 /* O_NOATIME can only be set by the owner or superuser */
47 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
48 if (!inode_owner_or_capable(inode))
51 /* required for strict SunOS emulation */
52 if (O_NONBLOCK != O_NDELAY)
56 /* Pipe packetized mode is controlled by O_DIRECT flag */
57 if (!S_ISFIFO(inode->i_mode) && (arg & O_DIRECT)) {
58 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
59 !filp->f_mapping->a_ops->direct_IO)
63 if (filp->f_op->check_flags)
64 error = filp->f_op->check_flags(arg);
69 * ->fasync() is responsible for setting the FASYNC bit.
71 if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op->fasync) {
72 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
78 spin_lock(&filp->f_lock);
79 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
80 spin_unlock(&filp->f_lock);
86 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
89 write_lock_irq(&filp->f_owner.lock);
90 if (force || !filp->f_owner.pid) {
91 put_pid(filp->f_owner.pid);
92 filp->f_owner.pid = get_pid(pid);
93 filp->f_owner.pid_type = type;
96 const struct cred *cred = current_cred();
97 filp->f_owner.uid = cred->uid;
98 filp->f_owner.euid = cred->euid;
101 write_unlock_irq(&filp->f_owner.lock);
104 void __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
107 security_file_set_fowner(filp);
108 f_modown(filp, pid, type, force);
110 EXPORT_SYMBOL(__f_setown);
112 int f_setown(struct file *filp, unsigned long arg, int force)
119 /* avoid overflow below */
127 pid = find_vpid(who);
128 __f_setown(filp, pid, type, force);
133 EXPORT_SYMBOL(f_setown);
135 void f_delown(struct file *filp)
137 f_modown(filp, NULL, PIDTYPE_PID, 1);
140 pid_t f_getown(struct file *filp)
143 read_lock(&filp->f_owner.lock);
144 pid = pid_vnr(filp->f_owner.pid);
145 if (filp->f_owner.pid_type == PIDTYPE_PGID)
147 read_unlock(&filp->f_owner.lock);
151 static int f_setown_ex(struct file *filp, unsigned long arg)
153 struct f_owner_ex __user *owner_p = (void __user *)arg;
154 struct f_owner_ex owner;
159 ret = copy_from_user(&owner, owner_p, sizeof(owner));
163 switch (owner.type) {
181 pid = find_vpid(owner.pid);
182 if (owner.pid && !pid)
185 __f_setown(filp, pid, type, 1);
191 static int f_getown_ex(struct file *filp, unsigned long arg)
193 struct f_owner_ex __user *owner_p = (void __user *)arg;
194 struct f_owner_ex owner;
197 read_lock(&filp->f_owner.lock);
198 owner.pid = pid_vnr(filp->f_owner.pid);
199 switch (filp->f_owner.pid_type) {
201 owner.type = F_OWNER_TID;
205 owner.type = F_OWNER_PID;
209 owner.type = F_OWNER_PGRP;
217 read_unlock(&filp->f_owner.lock);
220 ret = copy_to_user(owner_p, &owner, sizeof(owner));
227 #ifdef CONFIG_CHECKPOINT_RESTORE
228 static int f_getowner_uids(struct file *filp, unsigned long arg)
230 struct user_namespace *user_ns = current_user_ns();
231 uid_t __user *dst = (void __user *)arg;
235 read_lock(&filp->f_owner.lock);
236 src[0] = from_kuid(user_ns, filp->f_owner.uid);
237 src[1] = from_kuid(user_ns, filp->f_owner.euid);
238 read_unlock(&filp->f_owner.lock);
240 err = put_user(src[0], &dst[0]);
241 err |= put_user(src[1], &dst[1]);
246 static int f_getowner_uids(struct file *filp, unsigned long arg)
252 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
255 void __user *argp = (void __user *)arg;
261 err = f_dupfd(arg, filp, 0);
263 case F_DUPFD_CLOEXEC:
264 err = f_dupfd(arg, filp, O_CLOEXEC);
267 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
271 set_close_on_exec(fd, arg & FD_CLOEXEC);
277 err = setfl(fd, filp, arg);
279 #if BITS_PER_LONG != 32
280 /* 32-bit arches must use fcntl64() */
284 if (copy_from_user(&flock, argp, sizeof(flock)))
286 err = fcntl_getlk(filp, cmd, &flock);
287 if (!err && copy_to_user(argp, &flock, sizeof(flock)))
290 #if BITS_PER_LONG != 32
291 /* 32-bit arches must use fcntl64() */
298 if (copy_from_user(&flock, argp, sizeof(flock)))
300 err = fcntl_setlk(fd, filp, cmd, &flock);
304 * XXX If f_owner is a process group, the
305 * negative return value will get converted
306 * into an error. Oops. If we keep the
307 * current syscall conventions, the only way
308 * to fix this will be in libc.
310 err = f_getown(filp);
311 force_successful_syscall_return();
314 err = f_setown(filp, arg, 1);
317 err = f_getown_ex(filp, arg);
320 err = f_setown_ex(filp, arg);
322 case F_GETOWNER_UIDS:
323 err = f_getowner_uids(filp, arg);
326 err = filp->f_owner.signum;
329 /* arg == 0 restores default behaviour. */
330 if (!valid_signal(arg)) {
334 filp->f_owner.signum = arg;
337 err = fcntl_getlease(filp);
340 err = fcntl_setlease(fd, filp, arg);
343 err = fcntl_dirnotify(fd, filp, arg);
347 err = pipe_fcntl(filp, cmd, arg);
351 err = shmem_fcntl(filp, cmd, arg);
359 static int check_fcntl_cmd(unsigned cmd)
363 case F_DUPFD_CLOEXEC:
372 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
374 struct fd f = fdget_raw(fd);
380 if (unlikely(f.file->f_mode & FMODE_PATH)) {
381 if (!check_fcntl_cmd(cmd))
385 err = security_file_fcntl(f.file, cmd, arg);
387 err = do_fcntl(fd, cmd, arg, f.file);
395 #if BITS_PER_LONG == 32
396 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
399 void __user *argp = (void __user *)arg;
400 struct fd f = fdget_raw(fd);
401 struct flock64 flock;
407 if (unlikely(f.file->f_mode & FMODE_PATH)) {
408 if (!check_fcntl_cmd(cmd))
412 err = security_file_fcntl(f.file, cmd, arg);
420 if (copy_from_user(&flock, argp, sizeof(flock)))
422 err = fcntl_getlk64(f.file, cmd, &flock);
423 if (!err && copy_to_user(argp, &flock, sizeof(flock)))
431 if (copy_from_user(&flock, argp, sizeof(flock)))
433 err = fcntl_setlk64(fd, f.file, cmd, &flock);
436 err = do_fcntl(fd, cmd, arg, f.file);
447 static int get_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
449 if (!access_ok(VERIFY_READ, ufl, sizeof(*ufl)) ||
450 __get_user(kfl->l_type, &ufl->l_type) ||
451 __get_user(kfl->l_whence, &ufl->l_whence) ||
452 __get_user(kfl->l_start, &ufl->l_start) ||
453 __get_user(kfl->l_len, &ufl->l_len) ||
454 __get_user(kfl->l_pid, &ufl->l_pid))
459 static int put_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
461 if (!access_ok(VERIFY_WRITE, ufl, sizeof(*ufl)) ||
462 __put_user(kfl->l_type, &ufl->l_type) ||
463 __put_user(kfl->l_whence, &ufl->l_whence) ||
464 __put_user(kfl->l_start, &ufl->l_start) ||
465 __put_user(kfl->l_len, &ufl->l_len) ||
466 __put_user(kfl->l_pid, &ufl->l_pid))
471 #ifndef HAVE_ARCH_GET_COMPAT_FLOCK64
472 static int get_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
474 if (!access_ok(VERIFY_READ, ufl, sizeof(*ufl)) ||
475 __get_user(kfl->l_type, &ufl->l_type) ||
476 __get_user(kfl->l_whence, &ufl->l_whence) ||
477 __get_user(kfl->l_start, &ufl->l_start) ||
478 __get_user(kfl->l_len, &ufl->l_len) ||
479 __get_user(kfl->l_pid, &ufl->l_pid))
485 #ifndef HAVE_ARCH_PUT_COMPAT_FLOCK64
486 static int put_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
488 if (!access_ok(VERIFY_WRITE, ufl, sizeof(*ufl)) ||
489 __put_user(kfl->l_type, &ufl->l_type) ||
490 __put_user(kfl->l_whence, &ufl->l_whence) ||
491 __put_user(kfl->l_start, &ufl->l_start) ||
492 __put_user(kfl->l_len, &ufl->l_len) ||
493 __put_user(kfl->l_pid, &ufl->l_pid))
500 convert_fcntl_cmd(unsigned int cmd)
515 * GETLK was successful and we need to return the data, but it needs to fit in
516 * the compat structure.
517 * l_start shouldn't be too big, unless the original start + end is greater than
518 * COMPAT_OFF_T_MAX, in which case the app was asking for trouble, so we return
519 * -EOVERFLOW in that case. l_len could be too big, in which case we just
520 * truncate it, and only allow the app to see that part of the conflicting lock
521 * that might make sense to it anyway
523 static int fixup_compat_flock(struct flock *flock)
525 if (flock->l_start > COMPAT_OFF_T_MAX)
527 if (flock->l_len > COMPAT_OFF_T_MAX)
528 flock->l_len = COMPAT_OFF_T_MAX;
532 COMPAT_SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
535 struct fd f = fdget_raw(fd);
542 if (unlikely(f.file->f_mode & FMODE_PATH)) {
543 if (!check_fcntl_cmd(cmd))
547 err = security_file_fcntl(f.file, cmd, arg);
553 err = get_compat_flock(&flock, compat_ptr(arg));
556 err = fcntl_getlk(f.file, convert_fcntl_cmd(cmd), &flock);
559 err = fixup_compat_flock(&flock);
562 err = put_compat_flock(&flock, compat_ptr(arg));
566 err = get_compat_flock64(&flock, compat_ptr(arg));
569 err = fcntl_getlk(f.file, convert_fcntl_cmd(cmd), &flock);
572 err = fixup_compat_flock(&flock);
575 err = put_compat_flock64(&flock, compat_ptr(arg));
579 err = get_compat_flock(&flock, compat_ptr(arg));
582 err = fcntl_setlk(fd, f.file, convert_fcntl_cmd(cmd), &flock);
588 err = get_compat_flock64(&flock, compat_ptr(arg));
591 err = fcntl_setlk(fd, f.file, convert_fcntl_cmd(cmd), &flock);
594 err = do_fcntl(fd, cmd, arg, f.file);
602 COMPAT_SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd,
614 return compat_sys_fcntl64(fd, cmd, arg);
618 /* Table to convert sigio signal codes into poll band bitmaps */
620 static const long band_table[NSIGPOLL] = {
621 POLLIN | POLLRDNORM, /* POLL_IN */
622 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
623 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
624 POLLERR, /* POLL_ERR */
625 POLLPRI | POLLRDBAND, /* POLL_PRI */
626 POLLHUP | POLLERR /* POLL_HUP */
629 static inline int sigio_perm(struct task_struct *p,
630 struct fown_struct *fown, int sig)
632 const struct cred *cred;
636 cred = __task_cred(p);
637 ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
638 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
639 uid_eq(fown->uid, cred->suid) || uid_eq(fown->uid, cred->uid)) &&
640 !security_file_send_sigiotask(p, fown, sig));
645 static void send_sigio_to_task(struct task_struct *p,
646 struct fown_struct *fown,
647 int fd, int reason, int group)
650 * F_SETSIG can change ->signum lockless in parallel, make
651 * sure we read it once and use the same value throughout.
653 int signum = ACCESS_ONCE(fown->signum);
655 if (!sigio_perm(p, fown, signum))
661 /* Queue a rt signal with the appropriate fd as its
662 value. We use SI_SIGIO as the source, not
663 SI_KERNEL, since kernel signals always get
664 delivered even if we can't queue. Failure to
665 queue in this case _should_ be reported; we fall
666 back to SIGIO in that case. --sct */
667 si.si_signo = signum;
670 /* Make sure we are called with one of the POLL_*
671 reasons, otherwise we could leak kernel stack into
673 BUG_ON((reason & __SI_MASK) != __SI_POLL);
674 if (reason - POLL_IN >= NSIGPOLL)
677 si.si_band = band_table[reason - POLL_IN];
679 if (!do_send_sig_info(signum, &si, p, group))
681 /* fall-through: fall back on the old plain SIGIO signal */
683 do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
687 void send_sigio(struct fown_struct *fown, int fd, int band)
689 struct task_struct *p;
694 read_lock(&fown->lock);
696 type = fown->pid_type;
697 if (type == PIDTYPE_MAX) {
704 goto out_unlock_fown;
706 read_lock(&tasklist_lock);
707 do_each_pid_task(pid, type, p) {
708 send_sigio_to_task(p, fown, fd, band, group);
709 } while_each_pid_task(pid, type, p);
710 read_unlock(&tasklist_lock);
712 read_unlock(&fown->lock);
715 static void send_sigurg_to_task(struct task_struct *p,
716 struct fown_struct *fown, int group)
718 if (sigio_perm(p, fown, SIGURG))
719 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
722 int send_sigurg(struct fown_struct *fown)
724 struct task_struct *p;
730 read_lock(&fown->lock);
732 type = fown->pid_type;
733 if (type == PIDTYPE_MAX) {
740 goto out_unlock_fown;
744 read_lock(&tasklist_lock);
745 do_each_pid_task(pid, type, p) {
746 send_sigurg_to_task(p, fown, group);
747 } while_each_pid_task(pid, type, p);
748 read_unlock(&tasklist_lock);
750 read_unlock(&fown->lock);
754 static DEFINE_SPINLOCK(fasync_lock);
755 static struct kmem_cache *fasync_cache __read_mostly;
757 static void fasync_free_rcu(struct rcu_head *head)
759 kmem_cache_free(fasync_cache,
760 container_of(head, struct fasync_struct, fa_rcu));
764 * Remove a fasync entry. If successfully removed, return
765 * positive and clear the FASYNC flag. If no entry exists,
766 * do nothing and return 0.
768 * NOTE! It is very important that the FASYNC flag always
769 * match the state "is the filp on a fasync list".
772 int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
774 struct fasync_struct *fa, **fp;
777 spin_lock(&filp->f_lock);
778 spin_lock(&fasync_lock);
779 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
780 if (fa->fa_file != filp)
783 spin_lock_irq(&fa->fa_lock);
785 spin_unlock_irq(&fa->fa_lock);
788 call_rcu(&fa->fa_rcu, fasync_free_rcu);
789 filp->f_flags &= ~FASYNC;
793 spin_unlock(&fasync_lock);
794 spin_unlock(&filp->f_lock);
798 struct fasync_struct *fasync_alloc(void)
800 return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
804 * NOTE! This can be used only for unused fasync entries:
805 * entries that actually got inserted on the fasync list
806 * need to be released by rcu - see fasync_remove_entry.
808 void fasync_free(struct fasync_struct *new)
810 kmem_cache_free(fasync_cache, new);
814 * Insert a new entry into the fasync list. Return the pointer to the
815 * old one if we didn't use the new one.
817 * NOTE! It is very important that the FASYNC flag always
818 * match the state "is the filp on a fasync list".
820 struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
822 struct fasync_struct *fa, **fp;
824 spin_lock(&filp->f_lock);
825 spin_lock(&fasync_lock);
826 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
827 if (fa->fa_file != filp)
830 spin_lock_irq(&fa->fa_lock);
832 spin_unlock_irq(&fa->fa_lock);
836 spin_lock_init(&new->fa_lock);
837 new->magic = FASYNC_MAGIC;
840 new->fa_next = *fapp;
841 rcu_assign_pointer(*fapp, new);
842 filp->f_flags |= FASYNC;
845 spin_unlock(&fasync_lock);
846 spin_unlock(&filp->f_lock);
851 * Add a fasync entry. Return negative on error, positive if
852 * added, and zero if did nothing but change an existing one.
854 static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
856 struct fasync_struct *new;
858 new = fasync_alloc();
863 * fasync_insert_entry() returns the old (update) entry if
866 * So free the (unused) new entry and return 0 to let the
867 * caller know that we didn't add any new fasync entries.
869 if (fasync_insert_entry(fd, filp, fapp, new)) {
878 * fasync_helper() is used by almost all character device drivers
879 * to set up the fasync queue, and for regular files by the file
880 * lease code. It returns negative on error, 0 if it did no changes
881 * and positive if it added/deleted the entry.
883 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
886 return fasync_remove_entry(filp, fapp);
887 return fasync_add_entry(fd, filp, fapp);
890 EXPORT_SYMBOL(fasync_helper);
893 * rcu_read_lock() is held
895 static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
898 struct fown_struct *fown;
901 if (fa->magic != FASYNC_MAGIC) {
902 printk(KERN_ERR "kill_fasync: bad magic number in "
906 spin_lock_irqsave(&fa->fa_lock, flags);
908 fown = &fa->fa_file->f_owner;
909 /* Don't send SIGURG to processes which have not set a
910 queued signum: SIGURG has its own default signalling
912 if (!(sig == SIGURG && fown->signum == 0))
913 send_sigio(fown, fa->fa_fd, band);
915 spin_unlock_irqrestore(&fa->fa_lock, flags);
916 fa = rcu_dereference(fa->fa_next);
920 void kill_fasync(struct fasync_struct **fp, int sig, int band)
922 /* First a quick test without locking: usually
927 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
931 EXPORT_SYMBOL(kill_fasync);
933 static int __init fcntl_init(void)
936 * Please add new bits here to ensure allocation uniqueness.
937 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
938 * is defined as O_NONBLOCK on some platforms and not on others.
940 BUILD_BUG_ON(21 - 1 /* for O_RDONLY being 0 */ !=
942 (VALID_OPEN_FLAGS & ~(O_NONBLOCK | O_NDELAY)) |
943 __FMODE_EXEC | __FMODE_NONOTIFY));
945 fasync_cache = kmem_cache_create("fasync_cache",
946 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
950 module_init(fcntl_init)