x86/mm/pat: Don't report PAT on CPUs that don't support it
[linux/fpc-iii.git] / fs / fcntl.c
blobde1b16bb6a29c02ee5c2856e5c5411f978716b4f
1 /*
2 * linux/fs/fcntl.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
7 #include <linux/syscalls.h>
8 #include <linux/init.h>
9 #include <linux/mm.h>
10 #include <linux/sched/task.h>
11 #include <linux/fs.h>
12 #include <linux/file.h>
13 #include <linux/fdtable.h>
14 #include <linux/capability.h>
15 #include <linux/dnotify.h>
16 #include <linux/slab.h>
17 #include <linux/module.h>
18 #include <linux/pipe_fs_i.h>
19 #include <linux/security.h>
20 #include <linux/ptrace.h>
21 #include <linux/signal.h>
22 #include <linux/rcupdate.h>
23 #include <linux/pid_namespace.h>
24 #include <linux/user_namespace.h>
25 #include <linux/shmem_fs.h>
27 #include <asm/poll.h>
28 #include <asm/siginfo.h>
29 #include <linux/uaccess.h>
31 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
33 static int setfl(int fd, struct file * filp, unsigned long arg)
35 struct inode * inode = file_inode(filp);
36 int error = 0;
39 * O_APPEND cannot be cleared if the file is marked as append-only
40 * and the file is open for write.
42 if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
43 return -EPERM;
45 /* O_NOATIME can only be set by the owner or superuser */
46 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
47 if (!inode_owner_or_capable(inode))
48 return -EPERM;
50 /* required for strict SunOS emulation */
51 if (O_NONBLOCK != O_NDELAY)
52 if (arg & O_NDELAY)
53 arg |= O_NONBLOCK;
55 /* Pipe packetized mode is controlled by O_DIRECT flag */
56 if (!S_ISFIFO(inode->i_mode) && (arg & O_DIRECT)) {
57 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
58 !filp->f_mapping->a_ops->direct_IO)
59 return -EINVAL;
62 if (filp->f_op->check_flags)
63 error = filp->f_op->check_flags(arg);
64 if (error)
65 return error;
68 * ->fasync() is responsible for setting the FASYNC bit.
70 if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op->fasync) {
71 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
72 if (error < 0)
73 goto out;
74 if (error > 0)
75 error = 0;
77 spin_lock(&filp->f_lock);
78 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
79 spin_unlock(&filp->f_lock);
81 out:
82 return error;
85 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
86 int force)
88 write_lock_irq(&filp->f_owner.lock);
89 if (force || !filp->f_owner.pid) {
90 put_pid(filp->f_owner.pid);
91 filp->f_owner.pid = get_pid(pid);
92 filp->f_owner.pid_type = type;
94 if (pid) {
95 const struct cred *cred = current_cred();
96 filp->f_owner.uid = cred->uid;
97 filp->f_owner.euid = cred->euid;
100 write_unlock_irq(&filp->f_owner.lock);
103 void __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
104 int force)
106 security_file_set_fowner(filp);
107 f_modown(filp, pid, type, force);
109 EXPORT_SYMBOL(__f_setown);
111 void f_setown(struct file *filp, unsigned long arg, int force)
113 enum pid_type type;
114 struct pid *pid;
115 int who = arg;
116 type = PIDTYPE_PID;
117 if (who < 0) {
118 type = PIDTYPE_PGID;
119 who = -who;
121 rcu_read_lock();
122 pid = find_vpid(who);
123 __f_setown(filp, pid, type, force);
124 rcu_read_unlock();
126 EXPORT_SYMBOL(f_setown);
128 void f_delown(struct file *filp)
130 f_modown(filp, NULL, PIDTYPE_PID, 1);
133 pid_t f_getown(struct file *filp)
135 pid_t pid;
136 read_lock(&filp->f_owner.lock);
137 pid = pid_vnr(filp->f_owner.pid);
138 if (filp->f_owner.pid_type == PIDTYPE_PGID)
139 pid = -pid;
140 read_unlock(&filp->f_owner.lock);
141 return pid;
144 static int f_setown_ex(struct file *filp, unsigned long arg)
146 struct f_owner_ex __user *owner_p = (void __user *)arg;
147 struct f_owner_ex owner;
148 struct pid *pid;
149 int type;
150 int ret;
152 ret = copy_from_user(&owner, owner_p, sizeof(owner));
153 if (ret)
154 return -EFAULT;
156 switch (owner.type) {
157 case F_OWNER_TID:
158 type = PIDTYPE_MAX;
159 break;
161 case F_OWNER_PID:
162 type = PIDTYPE_PID;
163 break;
165 case F_OWNER_PGRP:
166 type = PIDTYPE_PGID;
167 break;
169 default:
170 return -EINVAL;
173 rcu_read_lock();
174 pid = find_vpid(owner.pid);
175 if (owner.pid && !pid)
176 ret = -ESRCH;
177 else
178 __f_setown(filp, pid, type, 1);
179 rcu_read_unlock();
181 return ret;
184 static int f_getown_ex(struct file *filp, unsigned long arg)
186 struct f_owner_ex __user *owner_p = (void __user *)arg;
187 struct f_owner_ex owner;
188 int ret = 0;
190 read_lock(&filp->f_owner.lock);
191 owner.pid = pid_vnr(filp->f_owner.pid);
192 switch (filp->f_owner.pid_type) {
193 case PIDTYPE_MAX:
194 owner.type = F_OWNER_TID;
195 break;
197 case PIDTYPE_PID:
198 owner.type = F_OWNER_PID;
199 break;
201 case PIDTYPE_PGID:
202 owner.type = F_OWNER_PGRP;
203 break;
205 default:
206 WARN_ON(1);
207 ret = -EINVAL;
208 break;
210 read_unlock(&filp->f_owner.lock);
212 if (!ret) {
213 ret = copy_to_user(owner_p, &owner, sizeof(owner));
214 if (ret)
215 ret = -EFAULT;
217 return ret;
220 #ifdef CONFIG_CHECKPOINT_RESTORE
221 static int f_getowner_uids(struct file *filp, unsigned long arg)
223 struct user_namespace *user_ns = current_user_ns();
224 uid_t __user *dst = (void __user *)arg;
225 uid_t src[2];
226 int err;
228 read_lock(&filp->f_owner.lock);
229 src[0] = from_kuid(user_ns, filp->f_owner.uid);
230 src[1] = from_kuid(user_ns, filp->f_owner.euid);
231 read_unlock(&filp->f_owner.lock);
233 err = put_user(src[0], &dst[0]);
234 err |= put_user(src[1], &dst[1]);
236 return err;
238 #else
239 static int f_getowner_uids(struct file *filp, unsigned long arg)
241 return -EINVAL;
243 #endif
245 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
246 struct file *filp)
248 long err = -EINVAL;
250 switch (cmd) {
251 case F_DUPFD:
252 err = f_dupfd(arg, filp, 0);
253 break;
254 case F_DUPFD_CLOEXEC:
255 err = f_dupfd(arg, filp, O_CLOEXEC);
256 break;
257 case F_GETFD:
258 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
259 break;
260 case F_SETFD:
261 err = 0;
262 set_close_on_exec(fd, arg & FD_CLOEXEC);
263 break;
264 case F_GETFL:
265 err = filp->f_flags;
266 break;
267 case F_SETFL:
268 err = setfl(fd, filp, arg);
269 break;
270 #if BITS_PER_LONG != 32
271 /* 32-bit arches must use fcntl64() */
272 case F_OFD_GETLK:
273 #endif
274 case F_GETLK:
275 err = fcntl_getlk(filp, cmd, (struct flock __user *) arg);
276 break;
277 #if BITS_PER_LONG != 32
278 /* 32-bit arches must use fcntl64() */
279 case F_OFD_SETLK:
280 case F_OFD_SETLKW:
281 #endif
282 /* Fallthrough */
283 case F_SETLK:
284 case F_SETLKW:
285 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
286 break;
287 case F_GETOWN:
289 * XXX If f_owner is a process group, the
290 * negative return value will get converted
291 * into an error. Oops. If we keep the
292 * current syscall conventions, the only way
293 * to fix this will be in libc.
295 err = f_getown(filp);
296 force_successful_syscall_return();
297 break;
298 case F_SETOWN:
299 f_setown(filp, arg, 1);
300 err = 0;
301 break;
302 case F_GETOWN_EX:
303 err = f_getown_ex(filp, arg);
304 break;
305 case F_SETOWN_EX:
306 err = f_setown_ex(filp, arg);
307 break;
308 case F_GETOWNER_UIDS:
309 err = f_getowner_uids(filp, arg);
310 break;
311 case F_GETSIG:
312 err = filp->f_owner.signum;
313 break;
314 case F_SETSIG:
315 /* arg == 0 restores default behaviour. */
316 if (!valid_signal(arg)) {
317 break;
319 err = 0;
320 filp->f_owner.signum = arg;
321 break;
322 case F_GETLEASE:
323 err = fcntl_getlease(filp);
324 break;
325 case F_SETLEASE:
326 err = fcntl_setlease(fd, filp, arg);
327 break;
328 case F_NOTIFY:
329 err = fcntl_dirnotify(fd, filp, arg);
330 break;
331 case F_SETPIPE_SZ:
332 case F_GETPIPE_SZ:
333 err = pipe_fcntl(filp, cmd, arg);
334 break;
335 case F_ADD_SEALS:
336 case F_GET_SEALS:
337 err = shmem_fcntl(filp, cmd, arg);
338 break;
339 default:
340 break;
342 return err;
345 static int check_fcntl_cmd(unsigned cmd)
347 switch (cmd) {
348 case F_DUPFD:
349 case F_DUPFD_CLOEXEC:
350 case F_GETFD:
351 case F_SETFD:
352 case F_GETFL:
353 return 1;
355 return 0;
358 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
360 struct fd f = fdget_raw(fd);
361 long err = -EBADF;
363 if (!f.file)
364 goto out;
366 if (unlikely(f.file->f_mode & FMODE_PATH)) {
367 if (!check_fcntl_cmd(cmd))
368 goto out1;
371 err = security_file_fcntl(f.file, cmd, arg);
372 if (!err)
373 err = do_fcntl(fd, cmd, arg, f.file);
375 out1:
376 fdput(f);
377 out:
378 return err;
381 #if BITS_PER_LONG == 32
382 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
383 unsigned long, arg)
385 struct fd f = fdget_raw(fd);
386 long err = -EBADF;
388 if (!f.file)
389 goto out;
391 if (unlikely(f.file->f_mode & FMODE_PATH)) {
392 if (!check_fcntl_cmd(cmd))
393 goto out1;
396 err = security_file_fcntl(f.file, cmd, arg);
397 if (err)
398 goto out1;
400 switch (cmd) {
401 case F_GETLK64:
402 case F_OFD_GETLK:
403 err = fcntl_getlk64(f.file, cmd, (struct flock64 __user *) arg);
404 break;
405 case F_SETLK64:
406 case F_SETLKW64:
407 case F_OFD_SETLK:
408 case F_OFD_SETLKW:
409 err = fcntl_setlk64(fd, f.file, cmd,
410 (struct flock64 __user *) arg);
411 break;
412 default:
413 err = do_fcntl(fd, cmd, arg, f.file);
414 break;
416 out1:
417 fdput(f);
418 out:
419 return err;
421 #endif
423 /* Table to convert sigio signal codes into poll band bitmaps */
425 static const long band_table[NSIGPOLL] = {
426 POLLIN | POLLRDNORM, /* POLL_IN */
427 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
428 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
429 POLLERR, /* POLL_ERR */
430 POLLPRI | POLLRDBAND, /* POLL_PRI */
431 POLLHUP | POLLERR /* POLL_HUP */
434 static inline int sigio_perm(struct task_struct *p,
435 struct fown_struct *fown, int sig)
437 const struct cred *cred;
438 int ret;
440 rcu_read_lock();
441 cred = __task_cred(p);
442 ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
443 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
444 uid_eq(fown->uid, cred->suid) || uid_eq(fown->uid, cred->uid)) &&
445 !security_file_send_sigiotask(p, fown, sig));
446 rcu_read_unlock();
447 return ret;
450 static void send_sigio_to_task(struct task_struct *p,
451 struct fown_struct *fown,
452 int fd, int reason, int group)
455 * F_SETSIG can change ->signum lockless in parallel, make
456 * sure we read it once and use the same value throughout.
458 int signum = ACCESS_ONCE(fown->signum);
460 if (!sigio_perm(p, fown, signum))
461 return;
463 switch (signum) {
464 siginfo_t si;
465 default:
466 /* Queue a rt signal with the appropriate fd as its
467 value. We use SI_SIGIO as the source, not
468 SI_KERNEL, since kernel signals always get
469 delivered even if we can't queue. Failure to
470 queue in this case _should_ be reported; we fall
471 back to SIGIO in that case. --sct */
472 si.si_signo = signum;
473 si.si_errno = 0;
474 si.si_code = reason;
475 /* Make sure we are called with one of the POLL_*
476 reasons, otherwise we could leak kernel stack into
477 userspace. */
478 BUG_ON((reason & __SI_MASK) != __SI_POLL);
479 if (reason - POLL_IN >= NSIGPOLL)
480 si.si_band = ~0L;
481 else
482 si.si_band = band_table[reason - POLL_IN];
483 si.si_fd = fd;
484 if (!do_send_sig_info(signum, &si, p, group))
485 break;
486 /* fall-through: fall back on the old plain SIGIO signal */
487 case 0:
488 do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
492 void send_sigio(struct fown_struct *fown, int fd, int band)
494 struct task_struct *p;
495 enum pid_type type;
496 struct pid *pid;
497 int group = 1;
499 read_lock(&fown->lock);
501 type = fown->pid_type;
502 if (type == PIDTYPE_MAX) {
503 group = 0;
504 type = PIDTYPE_PID;
507 pid = fown->pid;
508 if (!pid)
509 goto out_unlock_fown;
511 read_lock(&tasklist_lock);
512 do_each_pid_task(pid, type, p) {
513 send_sigio_to_task(p, fown, fd, band, group);
514 } while_each_pid_task(pid, type, p);
515 read_unlock(&tasklist_lock);
516 out_unlock_fown:
517 read_unlock(&fown->lock);
520 static void send_sigurg_to_task(struct task_struct *p,
521 struct fown_struct *fown, int group)
523 if (sigio_perm(p, fown, SIGURG))
524 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
527 int send_sigurg(struct fown_struct *fown)
529 struct task_struct *p;
530 enum pid_type type;
531 struct pid *pid;
532 int group = 1;
533 int ret = 0;
535 read_lock(&fown->lock);
537 type = fown->pid_type;
538 if (type == PIDTYPE_MAX) {
539 group = 0;
540 type = PIDTYPE_PID;
543 pid = fown->pid;
544 if (!pid)
545 goto out_unlock_fown;
547 ret = 1;
549 read_lock(&tasklist_lock);
550 do_each_pid_task(pid, type, p) {
551 send_sigurg_to_task(p, fown, group);
552 } while_each_pid_task(pid, type, p);
553 read_unlock(&tasklist_lock);
554 out_unlock_fown:
555 read_unlock(&fown->lock);
556 return ret;
559 static DEFINE_SPINLOCK(fasync_lock);
560 static struct kmem_cache *fasync_cache __read_mostly;
562 static void fasync_free_rcu(struct rcu_head *head)
564 kmem_cache_free(fasync_cache,
565 container_of(head, struct fasync_struct, fa_rcu));
569 * Remove a fasync entry. If successfully removed, return
570 * positive and clear the FASYNC flag. If no entry exists,
571 * do nothing and return 0.
573 * NOTE! It is very important that the FASYNC flag always
574 * match the state "is the filp on a fasync list".
577 int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
579 struct fasync_struct *fa, **fp;
580 int result = 0;
582 spin_lock(&filp->f_lock);
583 spin_lock(&fasync_lock);
584 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
585 if (fa->fa_file != filp)
586 continue;
588 spin_lock_irq(&fa->fa_lock);
589 fa->fa_file = NULL;
590 spin_unlock_irq(&fa->fa_lock);
592 *fp = fa->fa_next;
593 call_rcu(&fa->fa_rcu, fasync_free_rcu);
594 filp->f_flags &= ~FASYNC;
595 result = 1;
596 break;
598 spin_unlock(&fasync_lock);
599 spin_unlock(&filp->f_lock);
600 return result;
603 struct fasync_struct *fasync_alloc(void)
605 return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
609 * NOTE! This can be used only for unused fasync entries:
610 * entries that actually got inserted on the fasync list
611 * need to be released by rcu - see fasync_remove_entry.
613 void fasync_free(struct fasync_struct *new)
615 kmem_cache_free(fasync_cache, new);
619 * Insert a new entry into the fasync list. Return the pointer to the
620 * old one if we didn't use the new one.
622 * NOTE! It is very important that the FASYNC flag always
623 * match the state "is the filp on a fasync list".
625 struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
627 struct fasync_struct *fa, **fp;
629 spin_lock(&filp->f_lock);
630 spin_lock(&fasync_lock);
631 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
632 if (fa->fa_file != filp)
633 continue;
635 spin_lock_irq(&fa->fa_lock);
636 fa->fa_fd = fd;
637 spin_unlock_irq(&fa->fa_lock);
638 goto out;
641 spin_lock_init(&new->fa_lock);
642 new->magic = FASYNC_MAGIC;
643 new->fa_file = filp;
644 new->fa_fd = fd;
645 new->fa_next = *fapp;
646 rcu_assign_pointer(*fapp, new);
647 filp->f_flags |= FASYNC;
649 out:
650 spin_unlock(&fasync_lock);
651 spin_unlock(&filp->f_lock);
652 return fa;
656 * Add a fasync entry. Return negative on error, positive if
657 * added, and zero if did nothing but change an existing one.
659 static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
661 struct fasync_struct *new;
663 new = fasync_alloc();
664 if (!new)
665 return -ENOMEM;
668 * fasync_insert_entry() returns the old (update) entry if
669 * it existed.
671 * So free the (unused) new entry and return 0 to let the
672 * caller know that we didn't add any new fasync entries.
674 if (fasync_insert_entry(fd, filp, fapp, new)) {
675 fasync_free(new);
676 return 0;
679 return 1;
683 * fasync_helper() is used by almost all character device drivers
684 * to set up the fasync queue, and for regular files by the file
685 * lease code. It returns negative on error, 0 if it did no changes
686 * and positive if it added/deleted the entry.
688 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
690 if (!on)
691 return fasync_remove_entry(filp, fapp);
692 return fasync_add_entry(fd, filp, fapp);
695 EXPORT_SYMBOL(fasync_helper);
698 * rcu_read_lock() is held
700 static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
702 while (fa) {
703 struct fown_struct *fown;
704 unsigned long flags;
706 if (fa->magic != FASYNC_MAGIC) {
707 printk(KERN_ERR "kill_fasync: bad magic number in "
708 "fasync_struct!\n");
709 return;
711 spin_lock_irqsave(&fa->fa_lock, flags);
712 if (fa->fa_file) {
713 fown = &fa->fa_file->f_owner;
714 /* Don't send SIGURG to processes which have not set a
715 queued signum: SIGURG has its own default signalling
716 mechanism. */
717 if (!(sig == SIGURG && fown->signum == 0))
718 send_sigio(fown, fa->fa_fd, band);
720 spin_unlock_irqrestore(&fa->fa_lock, flags);
721 fa = rcu_dereference(fa->fa_next);
725 void kill_fasync(struct fasync_struct **fp, int sig, int band)
727 /* First a quick test without locking: usually
728 * the list is empty.
730 if (*fp) {
731 rcu_read_lock();
732 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
733 rcu_read_unlock();
736 EXPORT_SYMBOL(kill_fasync);
738 static int __init fcntl_init(void)
741 * Please add new bits here to ensure allocation uniqueness.
742 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
743 * is defined as O_NONBLOCK on some platforms and not on others.
745 BUILD_BUG_ON(21 - 1 /* for O_RDONLY being 0 */ !=
746 HWEIGHT32(
747 (VALID_OPEN_FLAGS & ~(O_NONBLOCK | O_NDELAY)) |
748 __FMODE_EXEC | __FMODE_NONOTIFY));
750 fasync_cache = kmem_cache_create("fasync_cache",
751 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
752 return 0;
755 module_init(fcntl_init)