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>
29 #include <linux/poll.h>
30 #include <asm/siginfo.h>
31 #include <linux/uaccess.h>
33 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
35 static int setfl(int fd, struct file * filp, unsigned long arg)
37 struct inode * inode = file_inode(filp);
41 * O_APPEND cannot be cleared if the file is marked as append-only
42 * and the file is open for write.
44 if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
47 /* O_NOATIME can only be set by the owner or superuser */
48 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
49 if (!inode_owner_or_capable(inode))
52 /* required for strict SunOS emulation */
53 if (O_NONBLOCK != O_NDELAY)
57 /* Pipe packetized mode is controlled by O_DIRECT flag */
58 if (!S_ISFIFO(inode->i_mode) && (arg & O_DIRECT)) {
59 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
60 !filp->f_mapping->a_ops->direct_IO)
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 spin_unlock(&filp->f_lock);
87 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
90 write_lock_irq(&filp->f_owner.lock);
91 if (force || !filp->f_owner.pid) {
92 put_pid(filp->f_owner.pid);
93 filp->f_owner.pid = get_pid(pid);
94 filp->f_owner.pid_type = type;
97 const struct cred *cred = current_cred();
98 filp->f_owner.uid = cred->uid;
99 filp->f_owner.euid = cred->euid;
102 write_unlock_irq(&filp->f_owner.lock);
105 void __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
108 security_file_set_fowner(filp);
109 f_modown(filp, pid, type, force);
111 EXPORT_SYMBOL(__f_setown);
113 int f_setown(struct file *filp, unsigned long arg, int force)
116 struct pid *pid = NULL;
117 int who = arg, ret = 0;
121 /* avoid overflow below */
131 pid = find_vpid(who);
137 __f_setown(filp, pid, type, force);
142 EXPORT_SYMBOL(f_setown);
144 void f_delown(struct file *filp)
146 f_modown(filp, NULL, PIDTYPE_TGID, 1);
149 pid_t f_getown(struct file *filp)
152 read_lock(&filp->f_owner.lock);
153 pid = pid_vnr(filp->f_owner.pid);
154 if (filp->f_owner.pid_type == PIDTYPE_PGID)
156 read_unlock(&filp->f_owner.lock);
160 static int f_setown_ex(struct file *filp, unsigned long arg)
162 struct f_owner_ex __user *owner_p = (void __user *)arg;
163 struct f_owner_ex owner;
168 ret = copy_from_user(&owner, owner_p, sizeof(owner));
172 switch (owner.type) {
190 pid = find_vpid(owner.pid);
191 if (owner.pid && !pid)
194 __f_setown(filp, pid, type, 1);
200 static int f_getown_ex(struct file *filp, unsigned long arg)
202 struct f_owner_ex __user *owner_p = (void __user *)arg;
203 struct f_owner_ex owner;
206 read_lock(&filp->f_owner.lock);
207 owner.pid = pid_vnr(filp->f_owner.pid);
208 switch (filp->f_owner.pid_type) {
210 owner.type = F_OWNER_TID;
214 owner.type = F_OWNER_PID;
218 owner.type = F_OWNER_PGRP;
226 read_unlock(&filp->f_owner.lock);
229 ret = copy_to_user(owner_p, &owner, sizeof(owner));
236 #ifdef CONFIG_CHECKPOINT_RESTORE
237 static int f_getowner_uids(struct file *filp, unsigned long arg)
239 struct user_namespace *user_ns = current_user_ns();
240 uid_t __user *dst = (void __user *)arg;
244 read_lock(&filp->f_owner.lock);
245 src[0] = from_kuid(user_ns, filp->f_owner.uid);
246 src[1] = from_kuid(user_ns, filp->f_owner.euid);
247 read_unlock(&filp->f_owner.lock);
249 err = put_user(src[0], &dst[0]);
250 err |= put_user(src[1], &dst[1]);
255 static int f_getowner_uids(struct file *filp, unsigned long arg)
261 static bool rw_hint_valid(enum rw_hint hint)
264 case RWF_WRITE_LIFE_NOT_SET:
265 case RWH_WRITE_LIFE_NONE:
266 case RWH_WRITE_LIFE_SHORT:
267 case RWH_WRITE_LIFE_MEDIUM:
268 case RWH_WRITE_LIFE_LONG:
269 case RWH_WRITE_LIFE_EXTREME:
276 static long fcntl_rw_hint(struct file *file, unsigned int cmd,
279 struct inode *inode = file_inode(file);
280 u64 *argp = (u64 __user *)arg;
285 case F_GET_FILE_RW_HINT:
286 h = file_write_hint(file);
287 if (copy_to_user(argp, &h, sizeof(*argp)))
290 case F_SET_FILE_RW_HINT:
291 if (copy_from_user(&h, argp, sizeof(h)))
293 hint = (enum rw_hint) h;
294 if (!rw_hint_valid(hint))
297 spin_lock(&file->f_lock);
298 file->f_write_hint = hint;
299 spin_unlock(&file->f_lock);
302 h = inode->i_write_hint;
303 if (copy_to_user(argp, &h, sizeof(*argp)))
307 if (copy_from_user(&h, argp, sizeof(h)))
309 hint = (enum rw_hint) h;
310 if (!rw_hint_valid(hint))
314 inode->i_write_hint = hint;
322 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
325 void __user *argp = (void __user *)arg;
331 err = f_dupfd(arg, filp, 0);
333 case F_DUPFD_CLOEXEC:
334 err = f_dupfd(arg, filp, O_CLOEXEC);
337 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
341 set_close_on_exec(fd, arg & FD_CLOEXEC);
347 err = setfl(fd, filp, arg);
349 #if BITS_PER_LONG != 32
350 /* 32-bit arches must use fcntl64() */
354 if (copy_from_user(&flock, argp, sizeof(flock)))
356 err = fcntl_getlk(filp, cmd, &flock);
357 if (!err && copy_to_user(argp, &flock, sizeof(flock)))
360 #if BITS_PER_LONG != 32
361 /* 32-bit arches must use fcntl64() */
368 if (copy_from_user(&flock, argp, sizeof(flock)))
370 err = fcntl_setlk(fd, filp, cmd, &flock);
374 * XXX If f_owner is a process group, the
375 * negative return value will get converted
376 * into an error. Oops. If we keep the
377 * current syscall conventions, the only way
378 * to fix this will be in libc.
380 err = f_getown(filp);
381 force_successful_syscall_return();
384 err = f_setown(filp, arg, 1);
387 err = f_getown_ex(filp, arg);
390 err = f_setown_ex(filp, arg);
392 case F_GETOWNER_UIDS:
393 err = f_getowner_uids(filp, arg);
396 err = filp->f_owner.signum;
399 /* arg == 0 restores default behaviour. */
400 if (!valid_signal(arg)) {
404 filp->f_owner.signum = arg;
407 err = fcntl_getlease(filp);
410 err = fcntl_setlease(fd, filp, arg);
413 err = fcntl_dirnotify(fd, filp, arg);
417 err = pipe_fcntl(filp, cmd, arg);
421 err = memfd_fcntl(filp, cmd, arg);
425 case F_GET_FILE_RW_HINT:
426 case F_SET_FILE_RW_HINT:
427 err = fcntl_rw_hint(filp, cmd, arg);
435 static int check_fcntl_cmd(unsigned cmd)
439 case F_DUPFD_CLOEXEC:
448 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
450 struct fd f = fdget_raw(fd);
456 if (unlikely(f.file->f_mode & FMODE_PATH)) {
457 if (!check_fcntl_cmd(cmd))
461 err = security_file_fcntl(f.file, cmd, arg);
463 err = do_fcntl(fd, cmd, arg, f.file);
471 #if BITS_PER_LONG == 32
472 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
475 void __user *argp = (void __user *)arg;
476 struct fd f = fdget_raw(fd);
477 struct flock64 flock;
483 if (unlikely(f.file->f_mode & FMODE_PATH)) {
484 if (!check_fcntl_cmd(cmd))
488 err = security_file_fcntl(f.file, cmd, arg);
496 if (copy_from_user(&flock, argp, sizeof(flock)))
498 err = fcntl_getlk64(f.file, cmd, &flock);
499 if (!err && copy_to_user(argp, &flock, sizeof(flock)))
507 if (copy_from_user(&flock, argp, sizeof(flock)))
509 err = fcntl_setlk64(fd, f.file, cmd, &flock);
512 err = do_fcntl(fd, cmd, arg, f.file);
523 /* careful - don't use anywhere else */
524 #define copy_flock_fields(dst, src) \
525 (dst)->l_type = (src)->l_type; \
526 (dst)->l_whence = (src)->l_whence; \
527 (dst)->l_start = (src)->l_start; \
528 (dst)->l_len = (src)->l_len; \
529 (dst)->l_pid = (src)->l_pid;
531 static int get_compat_flock(struct flock *kfl, const struct compat_flock __user *ufl)
533 struct compat_flock fl;
535 if (copy_from_user(&fl, ufl, sizeof(struct compat_flock)))
537 copy_flock_fields(kfl, &fl);
541 static int get_compat_flock64(struct flock *kfl, const struct compat_flock64 __user *ufl)
543 struct compat_flock64 fl;
545 if (copy_from_user(&fl, ufl, sizeof(struct compat_flock64)))
547 copy_flock_fields(kfl, &fl);
551 static int put_compat_flock(const struct flock *kfl, struct compat_flock __user *ufl)
553 struct compat_flock fl;
555 memset(&fl, 0, sizeof(struct compat_flock));
556 copy_flock_fields(&fl, kfl);
557 if (copy_to_user(ufl, &fl, sizeof(struct compat_flock)))
562 static int put_compat_flock64(const struct flock *kfl, struct compat_flock64 __user *ufl)
564 struct compat_flock64 fl;
566 BUILD_BUG_ON(sizeof(kfl->l_start) > sizeof(ufl->l_start));
567 BUILD_BUG_ON(sizeof(kfl->l_len) > sizeof(ufl->l_len));
569 memset(&fl, 0, sizeof(struct compat_flock64));
570 copy_flock_fields(&fl, kfl);
571 if (copy_to_user(ufl, &fl, sizeof(struct compat_flock64)))
575 #undef copy_flock_fields
578 convert_fcntl_cmd(unsigned int cmd)
593 * GETLK was successful and we need to return the data, but it needs to fit in
594 * the compat structure.
595 * l_start shouldn't be too big, unless the original start + end is greater than
596 * COMPAT_OFF_T_MAX, in which case the app was asking for trouble, so we return
597 * -EOVERFLOW in that case. l_len could be too big, in which case we just
598 * truncate it, and only allow the app to see that part of the conflicting lock
599 * that might make sense to it anyway
601 static int fixup_compat_flock(struct flock *flock)
603 if (flock->l_start > COMPAT_OFF_T_MAX)
605 if (flock->l_len > COMPAT_OFF_T_MAX)
606 flock->l_len = COMPAT_OFF_T_MAX;
610 static long do_compat_fcntl64(unsigned int fd, unsigned int cmd,
613 struct fd f = fdget_raw(fd);
620 if (unlikely(f.file->f_mode & FMODE_PATH)) {
621 if (!check_fcntl_cmd(cmd))
625 err = security_file_fcntl(f.file, cmd, arg);
631 err = get_compat_flock(&flock, compat_ptr(arg));
634 err = fcntl_getlk(f.file, convert_fcntl_cmd(cmd), &flock);
637 err = fixup_compat_flock(&flock);
639 err = put_compat_flock(&flock, compat_ptr(arg));
643 err = get_compat_flock64(&flock, compat_ptr(arg));
646 err = fcntl_getlk(f.file, convert_fcntl_cmd(cmd), &flock);
648 err = put_compat_flock64(&flock, compat_ptr(arg));
652 err = get_compat_flock(&flock, compat_ptr(arg));
655 err = fcntl_setlk(fd, f.file, convert_fcntl_cmd(cmd), &flock);
661 err = get_compat_flock64(&flock, compat_ptr(arg));
664 err = fcntl_setlk(fd, f.file, convert_fcntl_cmd(cmd), &flock);
667 err = do_fcntl(fd, cmd, arg, f.file);
675 COMPAT_SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
678 return do_compat_fcntl64(fd, cmd, arg);
681 COMPAT_SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd,
693 return do_compat_fcntl64(fd, cmd, arg);
697 /* Table to convert sigio signal codes into poll band bitmaps */
699 static const __poll_t band_table[NSIGPOLL] = {
700 EPOLLIN | EPOLLRDNORM, /* POLL_IN */
701 EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND, /* POLL_OUT */
702 EPOLLIN | EPOLLRDNORM | EPOLLMSG, /* POLL_MSG */
703 EPOLLERR, /* POLL_ERR */
704 EPOLLPRI | EPOLLRDBAND, /* POLL_PRI */
705 EPOLLHUP | EPOLLERR /* POLL_HUP */
708 static inline int sigio_perm(struct task_struct *p,
709 struct fown_struct *fown, int sig)
711 const struct cred *cred;
715 cred = __task_cred(p);
716 ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
717 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
718 uid_eq(fown->uid, cred->suid) || uid_eq(fown->uid, cred->uid)) &&
719 !security_file_send_sigiotask(p, fown, sig));
724 static void send_sigio_to_task(struct task_struct *p,
725 struct fown_struct *fown,
726 int fd, int reason, enum pid_type type)
729 * F_SETSIG can change ->signum lockless in parallel, make
730 * sure we read it once and use the same value throughout.
732 int signum = READ_ONCE(fown->signum);
734 if (!sigio_perm(p, fown, signum))
740 /* Queue a rt signal with the appropriate fd as its
741 value. We use SI_SIGIO as the source, not
742 SI_KERNEL, since kernel signals always get
743 delivered even if we can't queue. Failure to
744 queue in this case _should_ be reported; we fall
745 back to SIGIO in that case. --sct */
747 si.si_signo = signum;
751 * Posix definies POLL_IN and friends to be signal
752 * specific si_codes for SIG_POLL. Linux extended
753 * these si_codes to other signals in a way that is
754 * ambiguous if other signals also have signal
755 * specific si_codes. In that case use SI_SIGIO instead
756 * to remove the ambiguity.
758 if ((signum != SIGPOLL) && sig_specific_sicodes(signum))
759 si.si_code = SI_SIGIO;
761 /* Make sure we are called with one of the POLL_*
762 reasons, otherwise we could leak kernel stack into
764 BUG_ON((reason < POLL_IN) || ((reason - POLL_IN) >= NSIGPOLL));
765 if (reason - POLL_IN >= NSIGPOLL)
768 si.si_band = mangle_poll(band_table[reason - POLL_IN]);
770 if (!do_send_sig_info(signum, &si, p, type))
772 /* fall-through - fall back on the old plain SIGIO signal */
774 do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, type);
778 void send_sigio(struct fown_struct *fown, int fd, int band)
780 struct task_struct *p;
784 read_lock(&fown->lock);
786 type = fown->pid_type;
789 goto out_unlock_fown;
791 if (type <= PIDTYPE_TGID) {
793 p = pid_task(pid, PIDTYPE_PID);
795 send_sigio_to_task(p, fown, fd, band, type);
798 read_lock(&tasklist_lock);
799 do_each_pid_task(pid, type, p) {
800 send_sigio_to_task(p, fown, fd, band, type);
801 } while_each_pid_task(pid, type, p);
802 read_unlock(&tasklist_lock);
805 read_unlock(&fown->lock);
808 static void send_sigurg_to_task(struct task_struct *p,
809 struct fown_struct *fown, enum pid_type type)
811 if (sigio_perm(p, fown, SIGURG))
812 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, type);
815 int send_sigurg(struct fown_struct *fown)
817 struct task_struct *p;
822 read_lock(&fown->lock);
824 type = fown->pid_type;
827 goto out_unlock_fown;
831 if (type <= PIDTYPE_TGID) {
833 p = pid_task(pid, PIDTYPE_PID);
835 send_sigurg_to_task(p, fown, type);
838 read_lock(&tasklist_lock);
839 do_each_pid_task(pid, type, p) {
840 send_sigurg_to_task(p, fown, type);
841 } while_each_pid_task(pid, type, p);
842 read_unlock(&tasklist_lock);
845 read_unlock(&fown->lock);
849 static DEFINE_SPINLOCK(fasync_lock);
850 static struct kmem_cache *fasync_cache __read_mostly;
852 static void fasync_free_rcu(struct rcu_head *head)
854 kmem_cache_free(fasync_cache,
855 container_of(head, struct fasync_struct, fa_rcu));
859 * Remove a fasync entry. If successfully removed, return
860 * positive and clear the FASYNC flag. If no entry exists,
861 * do nothing and return 0.
863 * NOTE! It is very important that the FASYNC flag always
864 * match the state "is the filp on a fasync list".
867 int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
869 struct fasync_struct *fa, **fp;
872 spin_lock(&filp->f_lock);
873 spin_lock(&fasync_lock);
874 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
875 if (fa->fa_file != filp)
878 write_lock_irq(&fa->fa_lock);
880 write_unlock_irq(&fa->fa_lock);
883 call_rcu(&fa->fa_rcu, fasync_free_rcu);
884 filp->f_flags &= ~FASYNC;
888 spin_unlock(&fasync_lock);
889 spin_unlock(&filp->f_lock);
893 struct fasync_struct *fasync_alloc(void)
895 return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
899 * NOTE! This can be used only for unused fasync entries:
900 * entries that actually got inserted on the fasync list
901 * need to be released by rcu - see fasync_remove_entry.
903 void fasync_free(struct fasync_struct *new)
905 kmem_cache_free(fasync_cache, new);
909 * Insert a new entry into the fasync list. Return the pointer to the
910 * old one if we didn't use the new one.
912 * NOTE! It is very important that the FASYNC flag always
913 * match the state "is the filp on a fasync list".
915 struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
917 struct fasync_struct *fa, **fp;
919 spin_lock(&filp->f_lock);
920 spin_lock(&fasync_lock);
921 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
922 if (fa->fa_file != filp)
925 write_lock_irq(&fa->fa_lock);
927 write_unlock_irq(&fa->fa_lock);
931 rwlock_init(&new->fa_lock);
932 new->magic = FASYNC_MAGIC;
935 new->fa_next = *fapp;
936 rcu_assign_pointer(*fapp, new);
937 filp->f_flags |= FASYNC;
940 spin_unlock(&fasync_lock);
941 spin_unlock(&filp->f_lock);
946 * Add a fasync entry. Return negative on error, positive if
947 * added, and zero if did nothing but change an existing one.
949 static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
951 struct fasync_struct *new;
953 new = fasync_alloc();
958 * fasync_insert_entry() returns the old (update) entry if
961 * So free the (unused) new entry and return 0 to let the
962 * caller know that we didn't add any new fasync entries.
964 if (fasync_insert_entry(fd, filp, fapp, new)) {
973 * fasync_helper() is used by almost all character device drivers
974 * to set up the fasync queue, and for regular files by the file
975 * lease code. It returns negative on error, 0 if it did no changes
976 * and positive if it added/deleted the entry.
978 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
981 return fasync_remove_entry(filp, fapp);
982 return fasync_add_entry(fd, filp, fapp);
985 EXPORT_SYMBOL(fasync_helper);
988 * rcu_read_lock() is held
990 static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
993 struct fown_struct *fown;
995 if (fa->magic != FASYNC_MAGIC) {
996 printk(KERN_ERR "kill_fasync: bad magic number in "
1000 read_lock(&fa->fa_lock);
1002 fown = &fa->fa_file->f_owner;
1003 /* Don't send SIGURG to processes which have not set a
1004 queued signum: SIGURG has its own default signalling
1006 if (!(sig == SIGURG && fown->signum == 0))
1007 send_sigio(fown, fa->fa_fd, band);
1009 read_unlock(&fa->fa_lock);
1010 fa = rcu_dereference(fa->fa_next);
1014 void kill_fasync(struct fasync_struct **fp, int sig, int band)
1016 /* First a quick test without locking: usually
1017 * the list is empty.
1021 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
1025 EXPORT_SYMBOL(kill_fasync);
1027 static int __init fcntl_init(void)
1030 * Please add new bits here to ensure allocation uniqueness.
1031 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
1032 * is defined as O_NONBLOCK on some platforms and not on others.
1034 BUILD_BUG_ON(21 - 1 /* for O_RDONLY being 0 */ !=
1036 (VALID_OPEN_FLAGS & ~(O_NONBLOCK | O_NDELAY)) |
1037 __FMODE_EXEC | __FMODE_NONOTIFY));
1039 fasync_cache = kmem_cache_create("fasync_cache",
1040 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
1044 module_init(fcntl_init)