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)
54 /* Pipe packetized mode is controlled by O_DIRECT flag */
55 if (!S_ISFIFO(inode->i_mode) && (arg & O_DIRECT)) {
56 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
57 !filp->f_mapping->a_ops->direct_IO)
61 if (filp->f_op->check_flags)
62 error = filp->f_op->check_flags(arg);
67 * ->fasync() is responsible for setting the FASYNC bit.
69 if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op->fasync) {
70 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
76 spin_lock(&filp->f_lock);
77 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
78 spin_unlock(&filp->f_lock);
84 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
87 write_lock_irq(&filp->f_owner.lock);
88 if (force || !filp->f_owner.pid) {
89 put_pid(filp->f_owner.pid);
90 filp->f_owner.pid = get_pid(pid);
91 filp->f_owner.pid_type = type;
94 const struct cred *cred = current_cred();
95 filp->f_owner.uid = cred->uid;
96 filp->f_owner.euid = cred->euid;
99 write_unlock_irq(&filp->f_owner.lock);
102 void __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
105 security_file_set_fowner(filp);
106 f_modown(filp, pid, type, force);
108 EXPORT_SYMBOL(__f_setown);
110 void f_setown(struct file *filp, unsigned long arg, int force)
121 pid = find_vpid(who);
122 __f_setown(filp, pid, type, force);
125 EXPORT_SYMBOL(f_setown);
127 void f_delown(struct file *filp)
129 f_modown(filp, NULL, PIDTYPE_PID, 1);
132 pid_t f_getown(struct file *filp)
135 read_lock(&filp->f_owner.lock);
136 pid = pid_vnr(filp->f_owner.pid);
137 if (filp->f_owner.pid_type == PIDTYPE_PGID)
139 read_unlock(&filp->f_owner.lock);
143 static int f_setown_ex(struct file *filp, unsigned long arg)
145 struct f_owner_ex __user *owner_p = (void __user *)arg;
146 struct f_owner_ex owner;
151 ret = copy_from_user(&owner, owner_p, sizeof(owner));
155 switch (owner.type) {
173 pid = find_vpid(owner.pid);
174 if (owner.pid && !pid)
177 __f_setown(filp, pid, type, 1);
183 static int f_getown_ex(struct file *filp, unsigned long arg)
185 struct f_owner_ex __user *owner_p = (void __user *)arg;
186 struct f_owner_ex owner;
189 read_lock(&filp->f_owner.lock);
190 owner.pid = pid_vnr(filp->f_owner.pid);
191 switch (filp->f_owner.pid_type) {
193 owner.type = F_OWNER_TID;
197 owner.type = F_OWNER_PID;
201 owner.type = F_OWNER_PGRP;
209 read_unlock(&filp->f_owner.lock);
212 ret = copy_to_user(owner_p, &owner, sizeof(owner));
219 #ifdef CONFIG_CHECKPOINT_RESTORE
220 static int f_getowner_uids(struct file *filp, unsigned long arg)
222 struct user_namespace *user_ns = current_user_ns();
223 uid_t __user *dst = (void __user *)arg;
227 read_lock(&filp->f_owner.lock);
228 src[0] = from_kuid(user_ns, filp->f_owner.uid);
229 src[1] = from_kuid(user_ns, filp->f_owner.euid);
230 read_unlock(&filp->f_owner.lock);
232 err = put_user(src[0], &dst[0]);
233 err |= put_user(src[1], &dst[1]);
238 static int f_getowner_uids(struct file *filp, unsigned long arg)
244 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
251 err = f_dupfd(arg, filp, 0);
253 case F_DUPFD_CLOEXEC:
254 err = f_dupfd(arg, filp, O_CLOEXEC);
257 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
261 set_close_on_exec(fd, arg & FD_CLOEXEC);
267 err = setfl(fd, filp, arg);
269 #if BITS_PER_LONG != 32
270 /* 32-bit arches must use fcntl64() */
274 err = fcntl_getlk(filp, cmd, (struct flock __user *) arg);
276 #if BITS_PER_LONG != 32
277 /* 32-bit arches must use fcntl64() */
284 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
288 * XXX If f_owner is a process group, the
289 * negative return value will get converted
290 * into an error. Oops. If we keep the
291 * current syscall conventions, the only way
292 * to fix this will be in libc.
294 err = f_getown(filp);
295 force_successful_syscall_return();
298 f_setown(filp, arg, 1);
302 err = f_getown_ex(filp, arg);
305 err = f_setown_ex(filp, arg);
307 case F_GETOWNER_UIDS:
308 err = f_getowner_uids(filp, arg);
311 err = filp->f_owner.signum;
314 /* arg == 0 restores default behaviour. */
315 if (!valid_signal(arg)) {
319 filp->f_owner.signum = arg;
322 err = fcntl_getlease(filp);
325 err = fcntl_setlease(fd, filp, arg);
328 err = fcntl_dirnotify(fd, filp, arg);
332 err = pipe_fcntl(filp, cmd, arg);
336 err = shmem_fcntl(filp, cmd, arg);
344 static int check_fcntl_cmd(unsigned cmd)
348 case F_DUPFD_CLOEXEC:
357 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
359 struct fd f = fdget_raw(fd);
365 if (unlikely(f.file->f_mode & FMODE_PATH)) {
366 if (!check_fcntl_cmd(cmd))
370 err = security_file_fcntl(f.file, cmd, arg);
372 err = do_fcntl(fd, cmd, arg, f.file);
380 #if BITS_PER_LONG == 32
381 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
384 struct fd f = fdget_raw(fd);
390 if (unlikely(f.file->f_mode & FMODE_PATH)) {
391 if (!check_fcntl_cmd(cmd))
395 err = security_file_fcntl(f.file, cmd, arg);
402 err = fcntl_getlk64(f.file, cmd, (struct flock64 __user *) arg);
408 err = fcntl_setlk64(fd, f.file, cmd,
409 (struct flock64 __user *) arg);
412 err = do_fcntl(fd, cmd, arg, f.file);
422 /* Table to convert sigio signal codes into poll band bitmaps */
424 static const long band_table[NSIGPOLL] = {
425 POLLIN | POLLRDNORM, /* POLL_IN */
426 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
427 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
428 POLLERR, /* POLL_ERR */
429 POLLPRI | POLLRDBAND, /* POLL_PRI */
430 POLLHUP | POLLERR /* POLL_HUP */
433 static inline int sigio_perm(struct task_struct *p,
434 struct fown_struct *fown, int sig)
436 const struct cred *cred;
440 cred = __task_cred(p);
441 ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
442 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
443 uid_eq(fown->uid, cred->suid) || uid_eq(fown->uid, cred->uid)) &&
444 !security_file_send_sigiotask(p, fown, sig));
449 static void send_sigio_to_task(struct task_struct *p,
450 struct fown_struct *fown,
451 int fd, int reason, int group)
454 * F_SETSIG can change ->signum lockless in parallel, make
455 * sure we read it once and use the same value throughout.
457 int signum = ACCESS_ONCE(fown->signum);
459 if (!sigio_perm(p, fown, signum))
465 /* Queue a rt signal with the appropriate fd as its
466 value. We use SI_SIGIO as the source, not
467 SI_KERNEL, since kernel signals always get
468 delivered even if we can't queue. Failure to
469 queue in this case _should_ be reported; we fall
470 back to SIGIO in that case. --sct */
471 si.si_signo = signum;
474 /* Make sure we are called with one of the POLL_*
475 reasons, otherwise we could leak kernel stack into
477 BUG_ON((reason & __SI_MASK) != __SI_POLL);
478 if (reason - POLL_IN >= NSIGPOLL)
481 si.si_band = band_table[reason - POLL_IN];
483 if (!do_send_sig_info(signum, &si, p, group))
485 /* fall-through: fall back on the old plain SIGIO signal */
487 do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
491 void send_sigio(struct fown_struct *fown, int fd, int band)
493 struct task_struct *p;
498 read_lock(&fown->lock);
500 type = fown->pid_type;
501 if (type == PIDTYPE_MAX) {
508 goto out_unlock_fown;
510 read_lock(&tasklist_lock);
511 do_each_pid_task(pid, type, p) {
512 send_sigio_to_task(p, fown, fd, band, group);
513 } while_each_pid_task(pid, type, p);
514 read_unlock(&tasklist_lock);
516 read_unlock(&fown->lock);
519 static void send_sigurg_to_task(struct task_struct *p,
520 struct fown_struct *fown, int group)
522 if (sigio_perm(p, fown, SIGURG))
523 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
526 int send_sigurg(struct fown_struct *fown)
528 struct task_struct *p;
534 read_lock(&fown->lock);
536 type = fown->pid_type;
537 if (type == PIDTYPE_MAX) {
544 goto out_unlock_fown;
548 read_lock(&tasklist_lock);
549 do_each_pid_task(pid, type, p) {
550 send_sigurg_to_task(p, fown, group);
551 } while_each_pid_task(pid, type, p);
552 read_unlock(&tasklist_lock);
554 read_unlock(&fown->lock);
558 static DEFINE_SPINLOCK(fasync_lock);
559 static struct kmem_cache *fasync_cache __read_mostly;
561 static void fasync_free_rcu(struct rcu_head *head)
563 kmem_cache_free(fasync_cache,
564 container_of(head, struct fasync_struct, fa_rcu));
568 * Remove a fasync entry. If successfully removed, return
569 * positive and clear the FASYNC flag. If no entry exists,
570 * do nothing and return 0.
572 * NOTE! It is very important that the FASYNC flag always
573 * match the state "is the filp on a fasync list".
576 int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
578 struct fasync_struct *fa, **fp;
581 spin_lock(&filp->f_lock);
582 spin_lock(&fasync_lock);
583 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
584 if (fa->fa_file != filp)
587 spin_lock_irq(&fa->fa_lock);
589 spin_unlock_irq(&fa->fa_lock);
592 call_rcu(&fa->fa_rcu, fasync_free_rcu);
593 filp->f_flags &= ~FASYNC;
597 spin_unlock(&fasync_lock);
598 spin_unlock(&filp->f_lock);
602 struct fasync_struct *fasync_alloc(void)
604 return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
608 * NOTE! This can be used only for unused fasync entries:
609 * entries that actually got inserted on the fasync list
610 * need to be released by rcu - see fasync_remove_entry.
612 void fasync_free(struct fasync_struct *new)
614 kmem_cache_free(fasync_cache, new);
618 * Insert a new entry into the fasync list. Return the pointer to the
619 * old one if we didn't use the new one.
621 * NOTE! It is very important that the FASYNC flag always
622 * match the state "is the filp on a fasync list".
624 struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
626 struct fasync_struct *fa, **fp;
628 spin_lock(&filp->f_lock);
629 spin_lock(&fasync_lock);
630 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
631 if (fa->fa_file != filp)
634 spin_lock_irq(&fa->fa_lock);
636 spin_unlock_irq(&fa->fa_lock);
640 spin_lock_init(&new->fa_lock);
641 new->magic = FASYNC_MAGIC;
644 new->fa_next = *fapp;
645 rcu_assign_pointer(*fapp, new);
646 filp->f_flags |= FASYNC;
649 spin_unlock(&fasync_lock);
650 spin_unlock(&filp->f_lock);
655 * Add a fasync entry. Return negative on error, positive if
656 * added, and zero if did nothing but change an existing one.
658 static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
660 struct fasync_struct *new;
662 new = fasync_alloc();
667 * fasync_insert_entry() returns the old (update) entry if
670 * So free the (unused) new entry and return 0 to let the
671 * caller know that we didn't add any new fasync entries.
673 if (fasync_insert_entry(fd, filp, fapp, new)) {
682 * fasync_helper() is used by almost all character device drivers
683 * to set up the fasync queue, and for regular files by the file
684 * lease code. It returns negative on error, 0 if it did no changes
685 * and positive if it added/deleted the entry.
687 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
690 return fasync_remove_entry(filp, fapp);
691 return fasync_add_entry(fd, filp, fapp);
694 EXPORT_SYMBOL(fasync_helper);
697 * rcu_read_lock() is held
699 static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
702 struct fown_struct *fown;
705 if (fa->magic != FASYNC_MAGIC) {
706 printk(KERN_ERR "kill_fasync: bad magic number in "
710 spin_lock_irqsave(&fa->fa_lock, flags);
712 fown = &fa->fa_file->f_owner;
713 /* Don't send SIGURG to processes which have not set a
714 queued signum: SIGURG has its own default signalling
716 if (!(sig == SIGURG && fown->signum == 0))
717 send_sigio(fown, fa->fa_fd, band);
719 spin_unlock_irqrestore(&fa->fa_lock, flags);
720 fa = rcu_dereference(fa->fa_next);
724 void kill_fasync(struct fasync_struct **fp, int sig, int band)
726 /* First a quick test without locking: usually
731 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
735 EXPORT_SYMBOL(kill_fasync);
737 static int __init fcntl_init(void)
740 * Please add new bits here to ensure allocation uniqueness.
741 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
742 * is defined as O_NONBLOCK on some platforms and not on others.
744 BUILD_BUG_ON(21 - 1 /* for O_RDONLY being 0 */ != HWEIGHT32(
745 O_RDONLY | O_WRONLY | O_RDWR |
746 O_CREAT | O_EXCL | O_NOCTTY |
747 O_TRUNC | O_APPEND | /* O_NONBLOCK | */
748 __O_SYNC | O_DSYNC | FASYNC |
749 O_DIRECT | O_LARGEFILE | O_DIRECTORY |
750 O_NOFOLLOW | O_NOATIME | O_CLOEXEC |
751 __FMODE_EXEC | O_PATH | __O_TMPFILE |
755 fasync_cache = kmem_cache_create("fasync_cache",
756 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
760 module_init(fcntl_init)