1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 1991, 1992 Linus Torvalds
8 #include <linux/syscalls.h>
9 #include <linux/init.h>
11 #include <linux/sched/task.h>
13 #include <linux/file.h>
14 #include <linux/fdtable.h>
15 #include <linux/capability.h>
16 #include <linux/dnotify.h>
17 #include <linux/slab.h>
18 #include <linux/module.h>
19 #include <linux/pipe_fs_i.h>
20 #include <linux/security.h>
21 #include <linux/ptrace.h>
22 #include <linux/signal.h>
23 #include <linux/rcupdate.h>
24 #include <linux/pid_namespace.h>
25 #include <linux/user_namespace.h>
26 #include <linux/memfd.h>
27 #include <linux/compat.h>
28 #include <linux/mount.h>
30 #include <linux/poll.h>
31 #include <asm/siginfo.h>
32 #include <linux/uaccess.h>
34 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
36 static int setfl(int fd, struct file * filp, unsigned long arg)
38 struct inode * inode = file_inode(filp);
42 * O_APPEND cannot be cleared if the file is marked as append-only
43 * and the file is open for write.
45 if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
48 /* O_NOATIME can only be set by the owner or superuser */
49 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
50 if (!inode_owner_or_capable(file_mnt_user_ns(filp), inode))
53 /* required for strict SunOS emulation */
54 if (O_NONBLOCK != O_NDELAY)
58 /* Pipe packetized mode is controlled by O_DIRECT flag */
59 if (!S_ISFIFO(inode->i_mode) &&
61 !(filp->f_mode & FMODE_CAN_ODIRECT))
64 if (filp->f_op->check_flags)
65 error = filp->f_op->check_flags(arg);
70 * ->fasync() is responsible for setting the FASYNC bit.
72 if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op->fasync) {
73 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
79 spin_lock(&filp->f_lock);
80 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
81 filp->f_iocb_flags = iocb_flags(filp);
82 spin_unlock(&filp->f_lock);
88 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
91 write_lock_irq(&filp->f_owner.lock);
92 if (force || !filp->f_owner.pid) {
93 put_pid(filp->f_owner.pid);
94 filp->f_owner.pid = get_pid(pid);
95 filp->f_owner.pid_type = type;
98 const struct cred *cred = current_cred();
99 filp->f_owner.uid = cred->uid;
100 filp->f_owner.euid = cred->euid;
103 write_unlock_irq(&filp->f_owner.lock);
106 void __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
109 security_file_set_fowner(filp);
110 f_modown(filp, pid, type, force);
112 EXPORT_SYMBOL(__f_setown);
114 int f_setown(struct file *filp, unsigned long arg, int force)
117 struct pid *pid = NULL;
118 int who = arg, ret = 0;
122 /* avoid overflow below */
132 pid = find_vpid(who);
138 __f_setown(filp, pid, type, force);
143 EXPORT_SYMBOL(f_setown);
145 void f_delown(struct file *filp)
147 f_modown(filp, NULL, PIDTYPE_TGID, 1);
150 pid_t f_getown(struct file *filp)
154 read_lock_irq(&filp->f_owner.lock);
156 if (pid_task(filp->f_owner.pid, filp->f_owner.pid_type)) {
157 pid = pid_vnr(filp->f_owner.pid);
158 if (filp->f_owner.pid_type == PIDTYPE_PGID)
162 read_unlock_irq(&filp->f_owner.lock);
166 static int f_setown_ex(struct file *filp, unsigned long arg)
168 struct f_owner_ex __user *owner_p = (void __user *)arg;
169 struct f_owner_ex owner;
174 ret = copy_from_user(&owner, owner_p, sizeof(owner));
178 switch (owner.type) {
196 pid = find_vpid(owner.pid);
197 if (owner.pid && !pid)
200 __f_setown(filp, pid, type, 1);
206 static int f_getown_ex(struct file *filp, unsigned long arg)
208 struct f_owner_ex __user *owner_p = (void __user *)arg;
209 struct f_owner_ex owner = {};
212 read_lock_irq(&filp->f_owner.lock);
214 if (pid_task(filp->f_owner.pid, filp->f_owner.pid_type))
215 owner.pid = pid_vnr(filp->f_owner.pid);
217 switch (filp->f_owner.pid_type) {
219 owner.type = F_OWNER_TID;
223 owner.type = F_OWNER_PID;
227 owner.type = F_OWNER_PGRP;
235 read_unlock_irq(&filp->f_owner.lock);
238 ret = copy_to_user(owner_p, &owner, sizeof(owner));
245 #ifdef CONFIG_CHECKPOINT_RESTORE
246 static int f_getowner_uids(struct file *filp, unsigned long arg)
248 struct user_namespace *user_ns = current_user_ns();
249 uid_t __user *dst = (void __user *)arg;
253 read_lock_irq(&filp->f_owner.lock);
254 src[0] = from_kuid(user_ns, filp->f_owner.uid);
255 src[1] = from_kuid(user_ns, filp->f_owner.euid);
256 read_unlock_irq(&filp->f_owner.lock);
258 err = put_user(src[0], &dst[0]);
259 err |= put_user(src[1], &dst[1]);
264 static int f_getowner_uids(struct file *filp, unsigned long arg)
270 static bool rw_hint_valid(enum rw_hint hint)
273 case RWH_WRITE_LIFE_NOT_SET:
274 case RWH_WRITE_LIFE_NONE:
275 case RWH_WRITE_LIFE_SHORT:
276 case RWH_WRITE_LIFE_MEDIUM:
277 case RWH_WRITE_LIFE_LONG:
278 case RWH_WRITE_LIFE_EXTREME:
285 static long fcntl_rw_hint(struct file *file, unsigned int cmd,
288 struct inode *inode = file_inode(file);
289 u64 __user *argp = (u64 __user *)arg;
295 h = inode->i_write_hint;
296 if (copy_to_user(argp, &h, sizeof(*argp)))
300 if (copy_from_user(&h, argp, sizeof(h)))
302 hint = (enum rw_hint) h;
303 if (!rw_hint_valid(hint))
307 inode->i_write_hint = hint;
315 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
318 void __user *argp = (void __user *)arg;
324 err = f_dupfd(arg, filp, 0);
326 case F_DUPFD_CLOEXEC:
327 err = f_dupfd(arg, filp, O_CLOEXEC);
330 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
334 set_close_on_exec(fd, arg & FD_CLOEXEC);
340 err = setfl(fd, filp, arg);
342 #if BITS_PER_LONG != 32
343 /* 32-bit arches must use fcntl64() */
347 if (copy_from_user(&flock, argp, sizeof(flock)))
349 err = fcntl_getlk(filp, cmd, &flock);
350 if (!err && copy_to_user(argp, &flock, sizeof(flock)))
353 #if BITS_PER_LONG != 32
354 /* 32-bit arches must use fcntl64() */
361 if (copy_from_user(&flock, argp, sizeof(flock)))
363 err = fcntl_setlk(fd, filp, cmd, &flock);
367 * XXX If f_owner is a process group, the
368 * negative return value will get converted
369 * into an error. Oops. If we keep the
370 * current syscall conventions, the only way
371 * to fix this will be in libc.
373 err = f_getown(filp);
374 force_successful_syscall_return();
377 err = f_setown(filp, arg, 1);
380 err = f_getown_ex(filp, arg);
383 err = f_setown_ex(filp, arg);
385 case F_GETOWNER_UIDS:
386 err = f_getowner_uids(filp, arg);
389 err = filp->f_owner.signum;
392 /* arg == 0 restores default behaviour. */
393 if (!valid_signal(arg)) {
397 filp->f_owner.signum = arg;
400 err = fcntl_getlease(filp);
403 err = fcntl_setlease(fd, filp, arg);
406 err = fcntl_dirnotify(fd, filp, arg);
410 err = pipe_fcntl(filp, cmd, arg);
414 err = memfd_fcntl(filp, cmd, arg);
418 err = fcntl_rw_hint(filp, cmd, arg);
426 static int check_fcntl_cmd(unsigned cmd)
430 case F_DUPFD_CLOEXEC:
439 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
441 struct fd f = fdget_raw(fd);
447 if (unlikely(f.file->f_mode & FMODE_PATH)) {
448 if (!check_fcntl_cmd(cmd))
452 err = security_file_fcntl(f.file, cmd, arg);
454 err = do_fcntl(fd, cmd, arg, f.file);
462 #if BITS_PER_LONG == 32
463 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
466 void __user *argp = (void __user *)arg;
467 struct fd f = fdget_raw(fd);
468 struct flock64 flock;
474 if (unlikely(f.file->f_mode & FMODE_PATH)) {
475 if (!check_fcntl_cmd(cmd))
479 err = security_file_fcntl(f.file, cmd, arg);
487 if (copy_from_user(&flock, argp, sizeof(flock)))
489 err = fcntl_getlk64(f.file, cmd, &flock);
490 if (!err && copy_to_user(argp, &flock, sizeof(flock)))
498 if (copy_from_user(&flock, argp, sizeof(flock)))
500 err = fcntl_setlk64(fd, f.file, cmd, &flock);
503 err = do_fcntl(fd, cmd, arg, f.file);
514 /* careful - don't use anywhere else */
515 #define copy_flock_fields(dst, src) \
516 (dst)->l_type = (src)->l_type; \
517 (dst)->l_whence = (src)->l_whence; \
518 (dst)->l_start = (src)->l_start; \
519 (dst)->l_len = (src)->l_len; \
520 (dst)->l_pid = (src)->l_pid;
522 static int get_compat_flock(struct flock *kfl, const struct compat_flock __user *ufl)
524 struct compat_flock fl;
526 if (copy_from_user(&fl, ufl, sizeof(struct compat_flock)))
528 copy_flock_fields(kfl, &fl);
532 static int get_compat_flock64(struct flock *kfl, const struct compat_flock64 __user *ufl)
534 struct compat_flock64 fl;
536 if (copy_from_user(&fl, ufl, sizeof(struct compat_flock64)))
538 copy_flock_fields(kfl, &fl);
542 static int put_compat_flock(const struct flock *kfl, struct compat_flock __user *ufl)
544 struct compat_flock fl;
546 memset(&fl, 0, sizeof(struct compat_flock));
547 copy_flock_fields(&fl, kfl);
548 if (copy_to_user(ufl, &fl, sizeof(struct compat_flock)))
553 static int put_compat_flock64(const struct flock *kfl, struct compat_flock64 __user *ufl)
555 struct compat_flock64 fl;
557 BUILD_BUG_ON(sizeof(kfl->l_start) > sizeof(ufl->l_start));
558 BUILD_BUG_ON(sizeof(kfl->l_len) > sizeof(ufl->l_len));
560 memset(&fl, 0, sizeof(struct compat_flock64));
561 copy_flock_fields(&fl, kfl);
562 if (copy_to_user(ufl, &fl, sizeof(struct compat_flock64)))
566 #undef copy_flock_fields
569 convert_fcntl_cmd(unsigned int cmd)
584 * GETLK was successful and we need to return the data, but it needs to fit in
585 * the compat structure.
586 * l_start shouldn't be too big, unless the original start + end is greater than
587 * COMPAT_OFF_T_MAX, in which case the app was asking for trouble, so we return
588 * -EOVERFLOW in that case. l_len could be too big, in which case we just
589 * truncate it, and only allow the app to see that part of the conflicting lock
590 * that might make sense to it anyway
592 static int fixup_compat_flock(struct flock *flock)
594 if (flock->l_start > COMPAT_OFF_T_MAX)
596 if (flock->l_len > COMPAT_OFF_T_MAX)
597 flock->l_len = COMPAT_OFF_T_MAX;
601 static long do_compat_fcntl64(unsigned int fd, unsigned int cmd,
604 struct fd f = fdget_raw(fd);
611 if (unlikely(f.file->f_mode & FMODE_PATH)) {
612 if (!check_fcntl_cmd(cmd))
616 err = security_file_fcntl(f.file, cmd, arg);
622 err = get_compat_flock(&flock, compat_ptr(arg));
625 err = fcntl_getlk(f.file, convert_fcntl_cmd(cmd), &flock);
628 err = fixup_compat_flock(&flock);
630 err = put_compat_flock(&flock, compat_ptr(arg));
634 err = get_compat_flock64(&flock, compat_ptr(arg));
637 err = fcntl_getlk(f.file, convert_fcntl_cmd(cmd), &flock);
639 err = put_compat_flock64(&flock, compat_ptr(arg));
643 err = get_compat_flock(&flock, compat_ptr(arg));
646 err = fcntl_setlk(fd, f.file, convert_fcntl_cmd(cmd), &flock);
652 err = get_compat_flock64(&flock, compat_ptr(arg));
655 err = fcntl_setlk(fd, f.file, convert_fcntl_cmd(cmd), &flock);
658 err = do_fcntl(fd, cmd, arg, f.file);
666 COMPAT_SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
669 return do_compat_fcntl64(fd, cmd, arg);
672 COMPAT_SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd,
684 return do_compat_fcntl64(fd, cmd, arg);
688 /* Table to convert sigio signal codes into poll band bitmaps */
690 static const __poll_t band_table[NSIGPOLL] = {
691 EPOLLIN | EPOLLRDNORM, /* POLL_IN */
692 EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND, /* POLL_OUT */
693 EPOLLIN | EPOLLRDNORM | EPOLLMSG, /* POLL_MSG */
694 EPOLLERR, /* POLL_ERR */
695 EPOLLPRI | EPOLLRDBAND, /* POLL_PRI */
696 EPOLLHUP | EPOLLERR /* POLL_HUP */
699 static inline int sigio_perm(struct task_struct *p,
700 struct fown_struct *fown, int sig)
702 const struct cred *cred;
706 cred = __task_cred(p);
707 ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
708 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
709 uid_eq(fown->uid, cred->suid) || uid_eq(fown->uid, cred->uid)) &&
710 !security_file_send_sigiotask(p, fown, sig));
715 static void send_sigio_to_task(struct task_struct *p,
716 struct fown_struct *fown,
717 int fd, int reason, enum pid_type type)
720 * F_SETSIG can change ->signum lockless in parallel, make
721 * sure we read it once and use the same value throughout.
723 int signum = READ_ONCE(fown->signum);
725 if (!sigio_perm(p, fown, signum))
732 /* Queue a rt signal with the appropriate fd as its
733 value. We use SI_SIGIO as the source, not
734 SI_KERNEL, since kernel signals always get
735 delivered even if we can't queue. Failure to
736 queue in this case _should_ be reported; we fall
737 back to SIGIO in that case. --sct */
739 si.si_signo = signum;
743 * Posix definies POLL_IN and friends to be signal
744 * specific si_codes for SIG_POLL. Linux extended
745 * these si_codes to other signals in a way that is
746 * ambiguous if other signals also have signal
747 * specific si_codes. In that case use SI_SIGIO instead
748 * to remove the ambiguity.
750 if ((signum != SIGPOLL) && sig_specific_sicodes(signum))
751 si.si_code = SI_SIGIO;
753 /* Make sure we are called with one of the POLL_*
754 reasons, otherwise we could leak kernel stack into
756 BUG_ON((reason < POLL_IN) || ((reason - POLL_IN) >= NSIGPOLL));
757 if (reason - POLL_IN >= NSIGPOLL)
760 si.si_band = mangle_poll(band_table[reason - POLL_IN]);
762 if (!do_send_sig_info(signum, &si, p, type))
765 fallthrough; /* fall back on the old plain SIGIO signal */
767 do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, type);
771 void send_sigio(struct fown_struct *fown, int fd, int band)
773 struct task_struct *p;
778 read_lock_irqsave(&fown->lock, flags);
780 type = fown->pid_type;
783 goto out_unlock_fown;
785 if (type <= PIDTYPE_TGID) {
787 p = pid_task(pid, PIDTYPE_PID);
789 send_sigio_to_task(p, fown, fd, band, type);
792 read_lock(&tasklist_lock);
793 do_each_pid_task(pid, type, p) {
794 send_sigio_to_task(p, fown, fd, band, type);
795 } while_each_pid_task(pid, type, p);
796 read_unlock(&tasklist_lock);
799 read_unlock_irqrestore(&fown->lock, flags);
802 static void send_sigurg_to_task(struct task_struct *p,
803 struct fown_struct *fown, enum pid_type type)
805 if (sigio_perm(p, fown, SIGURG))
806 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, type);
809 int send_sigurg(struct fown_struct *fown)
811 struct task_struct *p;
817 read_lock_irqsave(&fown->lock, flags);
819 type = fown->pid_type;
822 goto out_unlock_fown;
826 if (type <= PIDTYPE_TGID) {
828 p = pid_task(pid, PIDTYPE_PID);
830 send_sigurg_to_task(p, fown, type);
833 read_lock(&tasklist_lock);
834 do_each_pid_task(pid, type, p) {
835 send_sigurg_to_task(p, fown, type);
836 } while_each_pid_task(pid, type, p);
837 read_unlock(&tasklist_lock);
840 read_unlock_irqrestore(&fown->lock, flags);
844 static DEFINE_SPINLOCK(fasync_lock);
845 static struct kmem_cache *fasync_cache __read_mostly;
847 static void fasync_free_rcu(struct rcu_head *head)
849 kmem_cache_free(fasync_cache,
850 container_of(head, struct fasync_struct, fa_rcu));
854 * Remove a fasync entry. If successfully removed, return
855 * positive and clear the FASYNC flag. If no entry exists,
856 * do nothing and return 0.
858 * NOTE! It is very important that the FASYNC flag always
859 * match the state "is the filp on a fasync list".
862 int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
864 struct fasync_struct *fa, **fp;
867 spin_lock(&filp->f_lock);
868 spin_lock(&fasync_lock);
869 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
870 if (fa->fa_file != filp)
873 write_lock_irq(&fa->fa_lock);
875 write_unlock_irq(&fa->fa_lock);
878 call_rcu(&fa->fa_rcu, fasync_free_rcu);
879 filp->f_flags &= ~FASYNC;
883 spin_unlock(&fasync_lock);
884 spin_unlock(&filp->f_lock);
888 struct fasync_struct *fasync_alloc(void)
890 return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
894 * NOTE! This can be used only for unused fasync entries:
895 * entries that actually got inserted on the fasync list
896 * need to be released by rcu - see fasync_remove_entry.
898 void fasync_free(struct fasync_struct *new)
900 kmem_cache_free(fasync_cache, new);
904 * Insert a new entry into the fasync list. Return the pointer to the
905 * old one if we didn't use the new one.
907 * NOTE! It is very important that the FASYNC flag always
908 * match the state "is the filp on a fasync list".
910 struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
912 struct fasync_struct *fa, **fp;
914 spin_lock(&filp->f_lock);
915 spin_lock(&fasync_lock);
916 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
917 if (fa->fa_file != filp)
920 write_lock_irq(&fa->fa_lock);
922 write_unlock_irq(&fa->fa_lock);
926 rwlock_init(&new->fa_lock);
927 new->magic = FASYNC_MAGIC;
930 new->fa_next = *fapp;
931 rcu_assign_pointer(*fapp, new);
932 filp->f_flags |= FASYNC;
935 spin_unlock(&fasync_lock);
936 spin_unlock(&filp->f_lock);
941 * Add a fasync entry. Return negative on error, positive if
942 * added, and zero if did nothing but change an existing one.
944 static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
946 struct fasync_struct *new;
948 new = fasync_alloc();
953 * fasync_insert_entry() returns the old (update) entry if
956 * So free the (unused) new entry and return 0 to let the
957 * caller know that we didn't add any new fasync entries.
959 if (fasync_insert_entry(fd, filp, fapp, new)) {
968 * fasync_helper() is used by almost all character device drivers
969 * to set up the fasync queue, and for regular files by the file
970 * lease code. It returns negative on error, 0 if it did no changes
971 * and positive if it added/deleted the entry.
973 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
976 return fasync_remove_entry(filp, fapp);
977 return fasync_add_entry(fd, filp, fapp);
980 EXPORT_SYMBOL(fasync_helper);
983 * rcu_read_lock() is held
985 static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
988 struct fown_struct *fown;
991 if (fa->magic != FASYNC_MAGIC) {
992 printk(KERN_ERR "kill_fasync: bad magic number in "
996 read_lock_irqsave(&fa->fa_lock, flags);
998 fown = &fa->fa_file->f_owner;
999 /* Don't send SIGURG to processes which have not set a
1000 queued signum: SIGURG has its own default signalling
1002 if (!(sig == SIGURG && fown->signum == 0))
1003 send_sigio(fown, fa->fa_fd, band);
1005 read_unlock_irqrestore(&fa->fa_lock, flags);
1006 fa = rcu_dereference(fa->fa_next);
1010 void kill_fasync(struct fasync_struct **fp, int sig, int band)
1012 /* First a quick test without locking: usually
1013 * the list is empty.
1017 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
1021 EXPORT_SYMBOL(kill_fasync);
1023 static int __init fcntl_init(void)
1026 * Please add new bits here to ensure allocation uniqueness.
1027 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
1028 * is defined as O_NONBLOCK on some platforms and not on others.
1030 BUILD_BUG_ON(21 - 1 /* for O_RDONLY being 0 */ !=
1032 (VALID_OPEN_FLAGS & ~(O_NONBLOCK | O_NDELAY)) |
1033 __FMODE_EXEC | __FMODE_NONOTIFY));
1035 fasync_cache = kmem_cache_create("fasync_cache",
1036 sizeof(struct fasync_struct), 0,
1037 SLAB_PANIC | SLAB_ACCOUNT, NULL);
1041 module_init(fcntl_init)