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 = file_inode(filp);
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->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->fasync) {
68 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
74 spin_lock(&filp->f_lock);
75 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
76 spin_unlock(&filp->f_lock);
82 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
85 write_lock_irq(&filp->f_owner.lock);
86 if (force || !filp->f_owner.pid) {
87 put_pid(filp->f_owner.pid);
88 filp->f_owner.pid = get_pid(pid);
89 filp->f_owner.pid_type = type;
92 const struct cred *cred = current_cred();
93 filp->f_owner.uid = cred->uid;
94 filp->f_owner.euid = cred->euid;
97 write_unlock_irq(&filp->f_owner.lock);
100 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
105 err = security_file_set_fowner(filp);
109 f_modown(filp, pid, type, force);
112 EXPORT_SYMBOL(__f_setown);
114 int f_setown(struct file *filp, unsigned long arg, int force)
126 pid = find_vpid(who);
127 result = __f_setown(filp, pid, type, force);
131 EXPORT_SYMBOL(f_setown);
133 void f_delown(struct file *filp)
135 f_modown(filp, NULL, PIDTYPE_PID, 1);
138 pid_t f_getown(struct file *filp)
141 read_lock(&filp->f_owner.lock);
142 pid = pid_vnr(filp->f_owner.pid);
143 if (filp->f_owner.pid_type == PIDTYPE_PGID)
145 read_unlock(&filp->f_owner.lock);
149 static int f_setown_ex(struct file *filp, unsigned long arg)
151 struct f_owner_ex __user *owner_p = (void __user *)arg;
152 struct f_owner_ex owner;
157 ret = copy_from_user(&owner, owner_p, sizeof(owner));
161 switch (owner.type) {
179 pid = find_vpid(owner.pid);
180 if (owner.pid && !pid)
183 ret = __f_setown(filp, pid, type, 1);
189 static int f_getown_ex(struct file *filp, unsigned long arg)
191 struct f_owner_ex __user *owner_p = (void __user *)arg;
192 struct f_owner_ex owner;
195 read_lock(&filp->f_owner.lock);
196 owner.pid = pid_vnr(filp->f_owner.pid);
197 switch (filp->f_owner.pid_type) {
199 owner.type = F_OWNER_TID;
203 owner.type = F_OWNER_PID;
207 owner.type = F_OWNER_PGRP;
215 read_unlock(&filp->f_owner.lock);
218 ret = copy_to_user(owner_p, &owner, sizeof(owner));
225 #ifdef CONFIG_CHECKPOINT_RESTORE
226 static int f_getowner_uids(struct file *filp, unsigned long arg)
228 struct user_namespace *user_ns = current_user_ns();
229 uid_t __user *dst = (void __user *)arg;
233 read_lock(&filp->f_owner.lock);
234 src[0] = from_kuid(user_ns, filp->f_owner.uid);
235 src[1] = from_kuid(user_ns, filp->f_owner.euid);
236 read_unlock(&filp->f_owner.lock);
238 err = put_user(src[0], &dst[0]);
239 err |= put_user(src[1], &dst[1]);
244 static int f_getowner_uids(struct file *filp, unsigned long arg)
250 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
257 err = f_dupfd(arg, filp, 0);
259 case F_DUPFD_CLOEXEC:
260 err = f_dupfd(arg, filp, O_CLOEXEC);
263 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
267 set_close_on_exec(fd, arg & FD_CLOEXEC);
273 err = setfl(fd, filp, arg);
276 err = fcntl_getlk(filp, (struct flock __user *) arg);
280 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
284 * XXX If f_owner is a process group, the
285 * negative return value will get converted
286 * into an error. Oops. If we keep the
287 * current syscall conventions, the only way
288 * to fix this will be in libc.
290 err = f_getown(filp);
291 force_successful_syscall_return();
294 err = f_setown(filp, arg, 1);
297 err = f_getown_ex(filp, arg);
300 err = f_setown_ex(filp, arg);
302 case F_GETOWNER_UIDS:
303 err = f_getowner_uids(filp, arg);
306 err = filp->f_owner.signum;
309 /* arg == 0 restores default behaviour. */
310 if (!valid_signal(arg)) {
314 filp->f_owner.signum = arg;
317 err = fcntl_getlease(filp);
320 err = fcntl_setlease(fd, filp, arg);
323 err = fcntl_dirnotify(fd, filp, arg);
327 err = pipe_fcntl(filp, cmd, arg);
335 static int check_fcntl_cmd(unsigned cmd)
339 case F_DUPFD_CLOEXEC:
348 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
350 struct fd f = fdget_raw(fd);
356 if (unlikely(f.file->f_mode & FMODE_PATH)) {
357 if (!check_fcntl_cmd(cmd))
361 err = security_file_fcntl(f.file, cmd, arg);
363 err = do_fcntl(fd, cmd, arg, f.file);
371 #if BITS_PER_LONG == 32
372 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
375 struct fd f = fdget_raw(fd);
381 if (unlikely(f.file->f_mode & FMODE_PATH)) {
382 if (!check_fcntl_cmd(cmd))
386 err = security_file_fcntl(f.file, cmd, arg);
392 err = fcntl_getlk64(f.file, (struct flock64 __user *) arg);
396 err = fcntl_setlk64(fd, f.file, cmd,
397 (struct flock64 __user *) arg);
400 err = do_fcntl(fd, cmd, arg, f.file);
410 /* Table to convert sigio signal codes into poll band bitmaps */
412 static const long band_table[NSIGPOLL] = {
413 POLLIN | POLLRDNORM, /* POLL_IN */
414 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
415 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
416 POLLERR, /* POLL_ERR */
417 POLLPRI | POLLRDBAND, /* POLL_PRI */
418 POLLHUP | POLLERR /* POLL_HUP */
421 static inline int sigio_perm(struct task_struct *p,
422 struct fown_struct *fown, int sig)
424 const struct cred *cred;
428 cred = __task_cred(p);
429 ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
430 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
431 uid_eq(fown->uid, cred->suid) || uid_eq(fown->uid, cred->uid)) &&
432 !security_file_send_sigiotask(p, fown, sig));
437 static void send_sigio_to_task(struct task_struct *p,
438 struct fown_struct *fown,
439 int fd, int reason, int group)
442 * F_SETSIG can change ->signum lockless in parallel, make
443 * sure we read it once and use the same value throughout.
445 int signum = ACCESS_ONCE(fown->signum);
447 if (!sigio_perm(p, fown, signum))
453 /* Queue a rt signal with the appropriate fd as its
454 value. We use SI_SIGIO as the source, not
455 SI_KERNEL, since kernel signals always get
456 delivered even if we can't queue. Failure to
457 queue in this case _should_ be reported; we fall
458 back to SIGIO in that case. --sct */
459 si.si_signo = signum;
462 /* Make sure we are called with one of the POLL_*
463 reasons, otherwise we could leak kernel stack into
465 BUG_ON((reason & __SI_MASK) != __SI_POLL);
466 if (reason - POLL_IN >= NSIGPOLL)
469 si.si_band = band_table[reason - POLL_IN];
471 if (!do_send_sig_info(signum, &si, p, group))
473 /* fall-through: fall back on the old plain SIGIO signal */
475 do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
479 void send_sigio(struct fown_struct *fown, int fd, int band)
481 struct task_struct *p;
486 read_lock(&fown->lock);
488 type = fown->pid_type;
489 if (type == PIDTYPE_MAX) {
496 goto out_unlock_fown;
498 read_lock(&tasklist_lock);
499 do_each_pid_task(pid, type, p) {
500 send_sigio_to_task(p, fown, fd, band, group);
501 } while_each_pid_task(pid, type, p);
502 read_unlock(&tasklist_lock);
504 read_unlock(&fown->lock);
507 static void send_sigurg_to_task(struct task_struct *p,
508 struct fown_struct *fown, int group)
510 if (sigio_perm(p, fown, SIGURG))
511 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
514 int send_sigurg(struct fown_struct *fown)
516 struct task_struct *p;
522 read_lock(&fown->lock);
524 type = fown->pid_type;
525 if (type == PIDTYPE_MAX) {
532 goto out_unlock_fown;
536 read_lock(&tasklist_lock);
537 do_each_pid_task(pid, type, p) {
538 send_sigurg_to_task(p, fown, group);
539 } while_each_pid_task(pid, type, p);
540 read_unlock(&tasklist_lock);
542 read_unlock(&fown->lock);
546 static DEFINE_SPINLOCK(fasync_lock);
547 static struct kmem_cache *fasync_cache __read_mostly;
549 static void fasync_free_rcu(struct rcu_head *head)
551 kmem_cache_free(fasync_cache,
552 container_of(head, struct fasync_struct, fa_rcu));
556 * Remove a fasync entry. If successfully removed, return
557 * positive and clear the FASYNC flag. If no entry exists,
558 * do nothing and return 0.
560 * NOTE! It is very important that the FASYNC flag always
561 * match the state "is the filp on a fasync list".
564 int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
566 struct fasync_struct *fa, **fp;
569 spin_lock(&filp->f_lock);
570 spin_lock(&fasync_lock);
571 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
572 if (fa->fa_file != filp)
575 spin_lock_irq(&fa->fa_lock);
577 spin_unlock_irq(&fa->fa_lock);
580 call_rcu(&fa->fa_rcu, fasync_free_rcu);
581 filp->f_flags &= ~FASYNC;
585 spin_unlock(&fasync_lock);
586 spin_unlock(&filp->f_lock);
590 struct fasync_struct *fasync_alloc(void)
592 return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
596 * NOTE! This can be used only for unused fasync entries:
597 * entries that actually got inserted on the fasync list
598 * need to be released by rcu - see fasync_remove_entry.
600 void fasync_free(struct fasync_struct *new)
602 kmem_cache_free(fasync_cache, new);
606 * Insert a new entry into the fasync list. Return the pointer to the
607 * old one if we didn't use the new one.
609 * NOTE! It is very important that the FASYNC flag always
610 * match the state "is the filp on a fasync list".
612 struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
614 struct fasync_struct *fa, **fp;
616 spin_lock(&filp->f_lock);
617 spin_lock(&fasync_lock);
618 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
619 if (fa->fa_file != filp)
622 spin_lock_irq(&fa->fa_lock);
624 spin_unlock_irq(&fa->fa_lock);
628 spin_lock_init(&new->fa_lock);
629 new->magic = FASYNC_MAGIC;
632 new->fa_next = *fapp;
633 rcu_assign_pointer(*fapp, new);
634 filp->f_flags |= FASYNC;
637 spin_unlock(&fasync_lock);
638 spin_unlock(&filp->f_lock);
643 * Add a fasync entry. Return negative on error, positive if
644 * added, and zero if did nothing but change an existing one.
646 static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
648 struct fasync_struct *new;
650 new = fasync_alloc();
655 * fasync_insert_entry() returns the old (update) entry if
658 * So free the (unused) new entry and return 0 to let the
659 * caller know that we didn't add any new fasync entries.
661 if (fasync_insert_entry(fd, filp, fapp, new)) {
670 * fasync_helper() is used by almost all character device drivers
671 * to set up the fasync queue, and for regular files by the file
672 * lease code. It returns negative on error, 0 if it did no changes
673 * and positive if it added/deleted the entry.
675 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
678 return fasync_remove_entry(filp, fapp);
679 return fasync_add_entry(fd, filp, fapp);
682 EXPORT_SYMBOL(fasync_helper);
685 * rcu_read_lock() is held
687 static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
690 struct fown_struct *fown;
693 if (fa->magic != FASYNC_MAGIC) {
694 printk(KERN_ERR "kill_fasync: bad magic number in "
698 spin_lock_irqsave(&fa->fa_lock, flags);
700 fown = &fa->fa_file->f_owner;
701 /* Don't send SIGURG to processes which have not set a
702 queued signum: SIGURG has its own default signalling
704 if (!(sig == SIGURG && fown->signum == 0))
705 send_sigio(fown, fa->fa_fd, band);
707 spin_unlock_irqrestore(&fa->fa_lock, flags);
708 fa = rcu_dereference(fa->fa_next);
712 void kill_fasync(struct fasync_struct **fp, int sig, int band)
714 /* First a quick test without locking: usually
719 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
723 EXPORT_SYMBOL(kill_fasync);
725 static int __init fcntl_init(void)
728 * Please add new bits here to ensure allocation uniqueness.
729 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
730 * is defined as O_NONBLOCK on some platforms and not on others.
732 BUILD_BUG_ON(20 - 1 /* for O_RDONLY being 0 */ != HWEIGHT32(
733 O_RDONLY | O_WRONLY | O_RDWR |
734 O_CREAT | O_EXCL | O_NOCTTY |
735 O_TRUNC | O_APPEND | /* O_NONBLOCK | */
736 __O_SYNC | O_DSYNC | FASYNC |
737 O_DIRECT | O_LARGEFILE | O_DIRECTORY |
738 O_NOFOLLOW | O_NOATIME | O_CLOEXEC |
739 __FMODE_EXEC | O_PATH | __O_TMPFILE
742 fasync_cache = kmem_cache_create("fasync_cache",
743 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
747 module_init(fcntl_init)