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 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, O_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);
276 #if BITS_PER_LONG != 32
277 /* 32-bit arches must use fcntl64() */
281 err = fcntl_getlk(filp, cmd, (struct flock __user *) arg);
283 #if BITS_PER_LONG != 32
284 /* 32-bit arches must use fcntl64() */
291 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
295 * XXX If f_owner is a process group, the
296 * negative return value will get converted
297 * into an error. Oops. If we keep the
298 * current syscall conventions, the only way
299 * to fix this will be in libc.
301 err = f_getown(filp);
302 force_successful_syscall_return();
305 err = f_setown(filp, arg, 1);
308 err = f_getown_ex(filp, arg);
311 err = f_setown_ex(filp, arg);
313 case F_GETOWNER_UIDS:
314 err = f_getowner_uids(filp, arg);
317 err = filp->f_owner.signum;
320 /* arg == 0 restores default behaviour. */
321 if (!valid_signal(arg)) {
325 filp->f_owner.signum = arg;
328 err = fcntl_getlease(filp);
331 err = fcntl_setlease(fd, filp, arg);
334 err = fcntl_dirnotify(fd, filp, arg);
338 err = pipe_fcntl(filp, cmd, arg);
342 err = shmem_fcntl(filp, cmd, arg);
350 static int check_fcntl_cmd(unsigned cmd)
354 case F_DUPFD_CLOEXEC:
363 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
365 struct fd f = fdget_raw(fd);
371 if (unlikely(f.file->f_mode & FMODE_PATH)) {
372 if (!check_fcntl_cmd(cmd))
376 err = security_file_fcntl(f.file, cmd, arg);
378 err = do_fcntl(fd, cmd, arg, f.file);
386 #if BITS_PER_LONG == 32
387 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
390 struct fd f = fdget_raw(fd);
396 if (unlikely(f.file->f_mode & FMODE_PATH)) {
397 if (!check_fcntl_cmd(cmd))
401 err = security_file_fcntl(f.file, cmd, arg);
408 err = fcntl_getlk64(f.file, cmd, (struct flock64 __user *) arg);
414 err = fcntl_setlk64(fd, f.file, cmd,
415 (struct flock64 __user *) arg);
418 err = do_fcntl(fd, cmd, arg, f.file);
428 /* Table to convert sigio signal codes into poll band bitmaps */
430 static const long band_table[NSIGPOLL] = {
431 POLLIN | POLLRDNORM, /* POLL_IN */
432 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
433 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
434 POLLERR, /* POLL_ERR */
435 POLLPRI | POLLRDBAND, /* POLL_PRI */
436 POLLHUP | POLLERR /* POLL_HUP */
439 static inline int sigio_perm(struct task_struct *p,
440 struct fown_struct *fown, int sig)
442 const struct cred *cred;
446 cred = __task_cred(p);
447 ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
448 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
449 uid_eq(fown->uid, cred->suid) || uid_eq(fown->uid, cred->uid)) &&
450 !security_file_send_sigiotask(p, fown, sig));
455 static void send_sigio_to_task(struct task_struct *p,
456 struct fown_struct *fown,
457 int fd, int reason, int group)
460 * F_SETSIG can change ->signum lockless in parallel, make
461 * sure we read it once and use the same value throughout.
463 int signum = ACCESS_ONCE(fown->signum);
465 if (!sigio_perm(p, fown, signum))
471 /* Queue a rt signal with the appropriate fd as its
472 value. We use SI_SIGIO as the source, not
473 SI_KERNEL, since kernel signals always get
474 delivered even if we can't queue. Failure to
475 queue in this case _should_ be reported; we fall
476 back to SIGIO in that case. --sct */
477 si.si_signo = signum;
480 /* Make sure we are called with one of the POLL_*
481 reasons, otherwise we could leak kernel stack into
483 BUG_ON((reason & __SI_MASK) != __SI_POLL);
484 if (reason - POLL_IN >= NSIGPOLL)
487 si.si_band = band_table[reason - POLL_IN];
489 if (!do_send_sig_info(signum, &si, p, group))
491 /* fall-through: fall back on the old plain SIGIO signal */
493 do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
497 void send_sigio(struct fown_struct *fown, int fd, int band)
499 struct task_struct *p;
504 read_lock(&fown->lock);
506 type = fown->pid_type;
507 if (type == PIDTYPE_MAX) {
514 goto out_unlock_fown;
516 read_lock(&tasklist_lock);
517 do_each_pid_task(pid, type, p) {
518 send_sigio_to_task(p, fown, fd, band, group);
519 } while_each_pid_task(pid, type, p);
520 read_unlock(&tasklist_lock);
522 read_unlock(&fown->lock);
525 static void send_sigurg_to_task(struct task_struct *p,
526 struct fown_struct *fown, int group)
528 if (sigio_perm(p, fown, SIGURG))
529 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
532 int send_sigurg(struct fown_struct *fown)
534 struct task_struct *p;
540 read_lock(&fown->lock);
542 type = fown->pid_type;
543 if (type == PIDTYPE_MAX) {
550 goto out_unlock_fown;
554 read_lock(&tasklist_lock);
555 do_each_pid_task(pid, type, p) {
556 send_sigurg_to_task(p, fown, group);
557 } while_each_pid_task(pid, type, p);
558 read_unlock(&tasklist_lock);
560 read_unlock(&fown->lock);
564 static DEFINE_SPINLOCK(fasync_lock);
565 static struct kmem_cache *fasync_cache __read_mostly;
567 static void fasync_free_rcu(struct rcu_head *head)
569 kmem_cache_free(fasync_cache,
570 container_of(head, struct fasync_struct, fa_rcu));
574 * Remove a fasync entry. If successfully removed, return
575 * positive and clear the FASYNC flag. If no entry exists,
576 * do nothing and return 0.
578 * NOTE! It is very important that the FASYNC flag always
579 * match the state "is the filp on a fasync list".
582 int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
584 struct fasync_struct *fa, **fp;
587 spin_lock(&filp->f_lock);
588 spin_lock(&fasync_lock);
589 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
590 if (fa->fa_file != filp)
593 spin_lock_irq(&fa->fa_lock);
595 spin_unlock_irq(&fa->fa_lock);
598 call_rcu(&fa->fa_rcu, fasync_free_rcu);
599 filp->f_flags &= ~FASYNC;
603 spin_unlock(&fasync_lock);
604 spin_unlock(&filp->f_lock);
608 struct fasync_struct *fasync_alloc(void)
610 return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
614 * NOTE! This can be used only for unused fasync entries:
615 * entries that actually got inserted on the fasync list
616 * need to be released by rcu - see fasync_remove_entry.
618 void fasync_free(struct fasync_struct *new)
620 kmem_cache_free(fasync_cache, new);
624 * Insert a new entry into the fasync list. Return the pointer to the
625 * old one if we didn't use the new one.
627 * NOTE! It is very important that the FASYNC flag always
628 * match the state "is the filp on a fasync list".
630 struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
632 struct fasync_struct *fa, **fp;
634 spin_lock(&filp->f_lock);
635 spin_lock(&fasync_lock);
636 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
637 if (fa->fa_file != filp)
640 spin_lock_irq(&fa->fa_lock);
642 spin_unlock_irq(&fa->fa_lock);
646 spin_lock_init(&new->fa_lock);
647 new->magic = FASYNC_MAGIC;
650 new->fa_next = *fapp;
651 rcu_assign_pointer(*fapp, new);
652 filp->f_flags |= FASYNC;
655 spin_unlock(&fasync_lock);
656 spin_unlock(&filp->f_lock);
661 * Add a fasync entry. Return negative on error, positive if
662 * added, and zero if did nothing but change an existing one.
664 static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
666 struct fasync_struct *new;
668 new = fasync_alloc();
673 * fasync_insert_entry() returns the old (update) entry if
676 * So free the (unused) new entry and return 0 to let the
677 * caller know that we didn't add any new fasync entries.
679 if (fasync_insert_entry(fd, filp, fapp, new)) {
688 * fasync_helper() is used by almost all character device drivers
689 * to set up the fasync queue, and for regular files by the file
690 * lease code. It returns negative on error, 0 if it did no changes
691 * and positive if it added/deleted the entry.
693 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
696 return fasync_remove_entry(filp, fapp);
697 return fasync_add_entry(fd, filp, fapp);
700 EXPORT_SYMBOL(fasync_helper);
703 * rcu_read_lock() is held
705 static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
708 struct fown_struct *fown;
711 if (fa->magic != FASYNC_MAGIC) {
712 printk(KERN_ERR "kill_fasync: bad magic number in "
716 spin_lock_irqsave(&fa->fa_lock, flags);
718 fown = &fa->fa_file->f_owner;
719 /* Don't send SIGURG to processes which have not set a
720 queued signum: SIGURG has its own default signalling
722 if (!(sig == SIGURG && fown->signum == 0))
723 send_sigio(fown, fa->fa_fd, band);
725 spin_unlock_irqrestore(&fa->fa_lock, flags);
726 fa = rcu_dereference(fa->fa_next);
730 void kill_fasync(struct fasync_struct **fp, int sig, int band)
732 /* First a quick test without locking: usually
737 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
741 EXPORT_SYMBOL(kill_fasync);
743 static int __init fcntl_init(void)
746 * Please add new bits here to ensure allocation uniqueness.
747 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
748 * is defined as O_NONBLOCK on some platforms and not on others.
750 BUILD_BUG_ON(20 - 1 /* for O_RDONLY being 0 */ != HWEIGHT32(
751 O_RDONLY | O_WRONLY | O_RDWR |
752 O_CREAT | O_EXCL | O_NOCTTY |
753 O_TRUNC | O_APPEND | /* O_NONBLOCK | */
754 __O_SYNC | O_DSYNC | FASYNC |
755 O_DIRECT | O_LARGEFILE | O_DIRECTORY |
756 O_NOFOLLOW | O_NOATIME | O_CLOEXEC |
757 __FMODE_EXEC | O_PATH | __O_TMPFILE
760 fasync_cache = kmem_cache_create("fasync_cache",
761 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
765 module_init(fcntl_init)