Loading...
1/*
2 * linux/fs/fcntl.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7#include <linux/syscalls.h>
8#include <linux/init.h>
9#include <linux/mm.h>
10#include <linux/fs.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
24#include <asm/poll.h>
25#include <asm/siginfo.h>
26#include <asm/uaccess.h>
27
28void set_close_on_exec(unsigned int fd, int flag)
29{
30 struct files_struct *files = current->files;
31 struct fdtable *fdt;
32 spin_lock(&files->file_lock);
33 fdt = files_fdtable(files);
34 if (flag)
35 __set_close_on_exec(fd, fdt);
36 else
37 __clear_close_on_exec(fd, fdt);
38 spin_unlock(&files->file_lock);
39}
40
41static bool get_close_on_exec(unsigned int fd)
42{
43 struct files_struct *files = current->files;
44 struct fdtable *fdt;
45 bool res;
46 rcu_read_lock();
47 fdt = files_fdtable(files);
48 res = close_on_exec(fd, fdt);
49 rcu_read_unlock();
50 return res;
51}
52
53SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
54{
55 int err = -EBADF;
56 struct file * file, *tofree;
57 struct files_struct * files = current->files;
58 struct fdtable *fdt;
59
60 if ((flags & ~O_CLOEXEC) != 0)
61 return -EINVAL;
62
63 if (unlikely(oldfd == newfd))
64 return -EINVAL;
65
66 spin_lock(&files->file_lock);
67 err = expand_files(files, newfd);
68 file = fcheck(oldfd);
69 if (unlikely(!file))
70 goto Ebadf;
71 if (unlikely(err < 0)) {
72 if (err == -EMFILE)
73 goto Ebadf;
74 goto out_unlock;
75 }
76 /*
77 * We need to detect attempts to do dup2() over allocated but still
78 * not finished descriptor. NB: OpenBSD avoids that at the price of
79 * extra work in their equivalent of fget() - they insert struct
80 * file immediately after grabbing descriptor, mark it larval if
81 * more work (e.g. actual opening) is needed and make sure that
82 * fget() treats larval files as absent. Potentially interesting,
83 * but while extra work in fget() is trivial, locking implications
84 * and amount of surgery on open()-related paths in VFS are not.
85 * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
86 * deadlocks in rather amusing ways, AFAICS. All of that is out of
87 * scope of POSIX or SUS, since neither considers shared descriptor
88 * tables and this condition does not arise without those.
89 */
90 err = -EBUSY;
91 fdt = files_fdtable(files);
92 tofree = fdt->fd[newfd];
93 if (!tofree && fd_is_open(newfd, fdt))
94 goto out_unlock;
95 get_file(file);
96 rcu_assign_pointer(fdt->fd[newfd], file);
97 __set_open_fd(newfd, fdt);
98 if (flags & O_CLOEXEC)
99 __set_close_on_exec(newfd, fdt);
100 else
101 __clear_close_on_exec(newfd, fdt);
102 spin_unlock(&files->file_lock);
103
104 if (tofree)
105 filp_close(tofree, files);
106
107 return newfd;
108
109Ebadf:
110 err = -EBADF;
111out_unlock:
112 spin_unlock(&files->file_lock);
113 return err;
114}
115
116SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
117{
118 if (unlikely(newfd == oldfd)) { /* corner case */
119 struct files_struct *files = current->files;
120 int retval = oldfd;
121
122 rcu_read_lock();
123 if (!fcheck_files(files, oldfd))
124 retval = -EBADF;
125 rcu_read_unlock();
126 return retval;
127 }
128 return sys_dup3(oldfd, newfd, 0);
129}
130
131SYSCALL_DEFINE1(dup, unsigned int, fildes)
132{
133 int ret = -EBADF;
134 struct file *file = fget_raw(fildes);
135
136 if (file) {
137 ret = get_unused_fd();
138 if (ret >= 0)
139 fd_install(ret, file);
140 else
141 fput(file);
142 }
143 return ret;
144}
145
146#define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
147
148static int setfl(int fd, struct file * filp, unsigned long arg)
149{
150 struct inode * inode = filp->f_path.dentry->d_inode;
151 int error = 0;
152
153 /*
154 * O_APPEND cannot be cleared if the file is marked as append-only
155 * and the file is open for write.
156 */
157 if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
158 return -EPERM;
159
160 /* O_NOATIME can only be set by the owner or superuser */
161 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
162 if (!inode_owner_or_capable(inode))
163 return -EPERM;
164
165 /* required for strict SunOS emulation */
166 if (O_NONBLOCK != O_NDELAY)
167 if (arg & O_NDELAY)
168 arg |= O_NONBLOCK;
169
170 if (arg & O_DIRECT) {
171 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
172 !filp->f_mapping->a_ops->direct_IO)
173 return -EINVAL;
174 }
175
176 if (filp->f_op && filp->f_op->check_flags)
177 error = filp->f_op->check_flags(arg);
178 if (error)
179 return error;
180
181 /*
182 * ->fasync() is responsible for setting the FASYNC bit.
183 */
184 if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op &&
185 filp->f_op->fasync) {
186 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
187 if (error < 0)
188 goto out;
189 if (error > 0)
190 error = 0;
191 }
192 spin_lock(&filp->f_lock);
193 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
194 spin_unlock(&filp->f_lock);
195
196 out:
197 return error;
198}
199
200static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
201 int force)
202{
203 write_lock_irq(&filp->f_owner.lock);
204 if (force || !filp->f_owner.pid) {
205 put_pid(filp->f_owner.pid);
206 filp->f_owner.pid = get_pid(pid);
207 filp->f_owner.pid_type = type;
208
209 if (pid) {
210 const struct cred *cred = current_cred();
211 filp->f_owner.uid = cred->uid;
212 filp->f_owner.euid = cred->euid;
213 }
214 }
215 write_unlock_irq(&filp->f_owner.lock);
216}
217
218int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
219 int force)
220{
221 int err;
222
223 err = security_file_set_fowner(filp);
224 if (err)
225 return err;
226
227 f_modown(filp, pid, type, force);
228 return 0;
229}
230EXPORT_SYMBOL(__f_setown);
231
232int f_setown(struct file *filp, unsigned long arg, int force)
233{
234 enum pid_type type;
235 struct pid *pid;
236 int who = arg;
237 int result;
238 type = PIDTYPE_PID;
239 if (who < 0) {
240 type = PIDTYPE_PGID;
241 who = -who;
242 }
243 rcu_read_lock();
244 pid = find_vpid(who);
245 result = __f_setown(filp, pid, type, force);
246 rcu_read_unlock();
247 return result;
248}
249EXPORT_SYMBOL(f_setown);
250
251void f_delown(struct file *filp)
252{
253 f_modown(filp, NULL, PIDTYPE_PID, 1);
254}
255
256pid_t f_getown(struct file *filp)
257{
258 pid_t pid;
259 read_lock(&filp->f_owner.lock);
260 pid = pid_vnr(filp->f_owner.pid);
261 if (filp->f_owner.pid_type == PIDTYPE_PGID)
262 pid = -pid;
263 read_unlock(&filp->f_owner.lock);
264 return pid;
265}
266
267static int f_setown_ex(struct file *filp, unsigned long arg)
268{
269 struct f_owner_ex * __user owner_p = (void * __user)arg;
270 struct f_owner_ex owner;
271 struct pid *pid;
272 int type;
273 int ret;
274
275 ret = copy_from_user(&owner, owner_p, sizeof(owner));
276 if (ret)
277 return -EFAULT;
278
279 switch (owner.type) {
280 case F_OWNER_TID:
281 type = PIDTYPE_MAX;
282 break;
283
284 case F_OWNER_PID:
285 type = PIDTYPE_PID;
286 break;
287
288 case F_OWNER_PGRP:
289 type = PIDTYPE_PGID;
290 break;
291
292 default:
293 return -EINVAL;
294 }
295
296 rcu_read_lock();
297 pid = find_vpid(owner.pid);
298 if (owner.pid && !pid)
299 ret = -ESRCH;
300 else
301 ret = __f_setown(filp, pid, type, 1);
302 rcu_read_unlock();
303
304 return ret;
305}
306
307static int f_getown_ex(struct file *filp, unsigned long arg)
308{
309 struct f_owner_ex * __user owner_p = (void * __user)arg;
310 struct f_owner_ex owner;
311 int ret = 0;
312
313 read_lock(&filp->f_owner.lock);
314 owner.pid = pid_vnr(filp->f_owner.pid);
315 switch (filp->f_owner.pid_type) {
316 case PIDTYPE_MAX:
317 owner.type = F_OWNER_TID;
318 break;
319
320 case PIDTYPE_PID:
321 owner.type = F_OWNER_PID;
322 break;
323
324 case PIDTYPE_PGID:
325 owner.type = F_OWNER_PGRP;
326 break;
327
328 default:
329 WARN_ON(1);
330 ret = -EINVAL;
331 break;
332 }
333 read_unlock(&filp->f_owner.lock);
334
335 if (!ret) {
336 ret = copy_to_user(owner_p, &owner, sizeof(owner));
337 if (ret)
338 ret = -EFAULT;
339 }
340 return ret;
341}
342
343static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
344 struct file *filp)
345{
346 long err = -EINVAL;
347
348 switch (cmd) {
349 case F_DUPFD:
350 case F_DUPFD_CLOEXEC:
351 if (arg >= rlimit(RLIMIT_NOFILE))
352 break;
353 err = alloc_fd(arg, cmd == F_DUPFD_CLOEXEC ? O_CLOEXEC : 0);
354 if (err >= 0) {
355 get_file(filp);
356 fd_install(err, filp);
357 }
358 break;
359 case F_GETFD:
360 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
361 break;
362 case F_SETFD:
363 err = 0;
364 set_close_on_exec(fd, arg & FD_CLOEXEC);
365 break;
366 case F_GETFL:
367 err = filp->f_flags;
368 break;
369 case F_SETFL:
370 err = setfl(fd, filp, arg);
371 break;
372 case F_GETLK:
373 err = fcntl_getlk(filp, (struct flock __user *) arg);
374 break;
375 case F_SETLK:
376 case F_SETLKW:
377 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
378 break;
379 case F_GETOWN:
380 /*
381 * XXX If f_owner is a process group, the
382 * negative return value will get converted
383 * into an error. Oops. If we keep the
384 * current syscall conventions, the only way
385 * to fix this will be in libc.
386 */
387 err = f_getown(filp);
388 force_successful_syscall_return();
389 break;
390 case F_SETOWN:
391 err = f_setown(filp, arg, 1);
392 break;
393 case F_GETOWN_EX:
394 err = f_getown_ex(filp, arg);
395 break;
396 case F_SETOWN_EX:
397 err = f_setown_ex(filp, arg);
398 break;
399 case F_GETSIG:
400 err = filp->f_owner.signum;
401 break;
402 case F_SETSIG:
403 /* arg == 0 restores default behaviour. */
404 if (!valid_signal(arg)) {
405 break;
406 }
407 err = 0;
408 filp->f_owner.signum = arg;
409 break;
410 case F_GETLEASE:
411 err = fcntl_getlease(filp);
412 break;
413 case F_SETLEASE:
414 err = fcntl_setlease(fd, filp, arg);
415 break;
416 case F_NOTIFY:
417 err = fcntl_dirnotify(fd, filp, arg);
418 break;
419 case F_SETPIPE_SZ:
420 case F_GETPIPE_SZ:
421 err = pipe_fcntl(filp, cmd, arg);
422 break;
423 default:
424 break;
425 }
426 return err;
427}
428
429static int check_fcntl_cmd(unsigned cmd)
430{
431 switch (cmd) {
432 case F_DUPFD:
433 case F_DUPFD_CLOEXEC:
434 case F_GETFD:
435 case F_SETFD:
436 case F_GETFL:
437 return 1;
438 }
439 return 0;
440}
441
442SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
443{
444 struct file *filp;
445 int fput_needed;
446 long err = -EBADF;
447
448 filp = fget_raw_light(fd, &fput_needed);
449 if (!filp)
450 goto out;
451
452 if (unlikely(filp->f_mode & FMODE_PATH)) {
453 if (!check_fcntl_cmd(cmd))
454 goto out1;
455 }
456
457 err = security_file_fcntl(filp, cmd, arg);
458 if (!err)
459 err = do_fcntl(fd, cmd, arg, filp);
460
461out1:
462 fput_light(filp, fput_needed);
463out:
464 return err;
465}
466
467#if BITS_PER_LONG == 32
468SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
469 unsigned long, arg)
470{
471 struct file * filp;
472 long err = -EBADF;
473 int fput_needed;
474
475 filp = fget_raw_light(fd, &fput_needed);
476 if (!filp)
477 goto out;
478
479 if (unlikely(filp->f_mode & FMODE_PATH)) {
480 if (!check_fcntl_cmd(cmd))
481 goto out1;
482 }
483
484 err = security_file_fcntl(filp, cmd, arg);
485 if (err)
486 goto out1;
487
488 switch (cmd) {
489 case F_GETLK64:
490 err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
491 break;
492 case F_SETLK64:
493 case F_SETLKW64:
494 err = fcntl_setlk64(fd, filp, cmd,
495 (struct flock64 __user *) arg);
496 break;
497 default:
498 err = do_fcntl(fd, cmd, arg, filp);
499 break;
500 }
501out1:
502 fput_light(filp, fput_needed);
503out:
504 return err;
505}
506#endif
507
508/* Table to convert sigio signal codes into poll band bitmaps */
509
510static const long band_table[NSIGPOLL] = {
511 POLLIN | POLLRDNORM, /* POLL_IN */
512 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
513 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
514 POLLERR, /* POLL_ERR */
515 POLLPRI | POLLRDBAND, /* POLL_PRI */
516 POLLHUP | POLLERR /* POLL_HUP */
517};
518
519static inline int sigio_perm(struct task_struct *p,
520 struct fown_struct *fown, int sig)
521{
522 const struct cred *cred;
523 int ret;
524
525 rcu_read_lock();
526 cred = __task_cred(p);
527 ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
528 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
529 uid_eq(fown->uid, cred->suid) || uid_eq(fown->uid, cred->uid)) &&
530 !security_file_send_sigiotask(p, fown, sig));
531 rcu_read_unlock();
532 return ret;
533}
534
535static void send_sigio_to_task(struct task_struct *p,
536 struct fown_struct *fown,
537 int fd, int reason, int group)
538{
539 /*
540 * F_SETSIG can change ->signum lockless in parallel, make
541 * sure we read it once and use the same value throughout.
542 */
543 int signum = ACCESS_ONCE(fown->signum);
544
545 if (!sigio_perm(p, fown, signum))
546 return;
547
548 switch (signum) {
549 siginfo_t si;
550 default:
551 /* Queue a rt signal with the appropriate fd as its
552 value. We use SI_SIGIO as the source, not
553 SI_KERNEL, since kernel signals always get
554 delivered even if we can't queue. Failure to
555 queue in this case _should_ be reported; we fall
556 back to SIGIO in that case. --sct */
557 si.si_signo = signum;
558 si.si_errno = 0;
559 si.si_code = reason;
560 /* Make sure we are called with one of the POLL_*
561 reasons, otherwise we could leak kernel stack into
562 userspace. */
563 BUG_ON((reason & __SI_MASK) != __SI_POLL);
564 if (reason - POLL_IN >= NSIGPOLL)
565 si.si_band = ~0L;
566 else
567 si.si_band = band_table[reason - POLL_IN];
568 si.si_fd = fd;
569 if (!do_send_sig_info(signum, &si, p, group))
570 break;
571 /* fall-through: fall back on the old plain SIGIO signal */
572 case 0:
573 do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
574 }
575}
576
577void send_sigio(struct fown_struct *fown, int fd, int band)
578{
579 struct task_struct *p;
580 enum pid_type type;
581 struct pid *pid;
582 int group = 1;
583
584 read_lock(&fown->lock);
585
586 type = fown->pid_type;
587 if (type == PIDTYPE_MAX) {
588 group = 0;
589 type = PIDTYPE_PID;
590 }
591
592 pid = fown->pid;
593 if (!pid)
594 goto out_unlock_fown;
595
596 read_lock(&tasklist_lock);
597 do_each_pid_task(pid, type, p) {
598 send_sigio_to_task(p, fown, fd, band, group);
599 } while_each_pid_task(pid, type, p);
600 read_unlock(&tasklist_lock);
601 out_unlock_fown:
602 read_unlock(&fown->lock);
603}
604
605static void send_sigurg_to_task(struct task_struct *p,
606 struct fown_struct *fown, int group)
607{
608 if (sigio_perm(p, fown, SIGURG))
609 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
610}
611
612int send_sigurg(struct fown_struct *fown)
613{
614 struct task_struct *p;
615 enum pid_type type;
616 struct pid *pid;
617 int group = 1;
618 int ret = 0;
619
620 read_lock(&fown->lock);
621
622 type = fown->pid_type;
623 if (type == PIDTYPE_MAX) {
624 group = 0;
625 type = PIDTYPE_PID;
626 }
627
628 pid = fown->pid;
629 if (!pid)
630 goto out_unlock_fown;
631
632 ret = 1;
633
634 read_lock(&tasklist_lock);
635 do_each_pid_task(pid, type, p) {
636 send_sigurg_to_task(p, fown, group);
637 } while_each_pid_task(pid, type, p);
638 read_unlock(&tasklist_lock);
639 out_unlock_fown:
640 read_unlock(&fown->lock);
641 return ret;
642}
643
644static DEFINE_SPINLOCK(fasync_lock);
645static struct kmem_cache *fasync_cache __read_mostly;
646
647static void fasync_free_rcu(struct rcu_head *head)
648{
649 kmem_cache_free(fasync_cache,
650 container_of(head, struct fasync_struct, fa_rcu));
651}
652
653/*
654 * Remove a fasync entry. If successfully removed, return
655 * positive and clear the FASYNC flag. If no entry exists,
656 * do nothing and return 0.
657 *
658 * NOTE! It is very important that the FASYNC flag always
659 * match the state "is the filp on a fasync list".
660 *
661 */
662int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
663{
664 struct fasync_struct *fa, **fp;
665 int result = 0;
666
667 spin_lock(&filp->f_lock);
668 spin_lock(&fasync_lock);
669 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
670 if (fa->fa_file != filp)
671 continue;
672
673 spin_lock_irq(&fa->fa_lock);
674 fa->fa_file = NULL;
675 spin_unlock_irq(&fa->fa_lock);
676
677 *fp = fa->fa_next;
678 call_rcu(&fa->fa_rcu, fasync_free_rcu);
679 filp->f_flags &= ~FASYNC;
680 result = 1;
681 break;
682 }
683 spin_unlock(&fasync_lock);
684 spin_unlock(&filp->f_lock);
685 return result;
686}
687
688struct fasync_struct *fasync_alloc(void)
689{
690 return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
691}
692
693/*
694 * NOTE! This can be used only for unused fasync entries:
695 * entries that actually got inserted on the fasync list
696 * need to be released by rcu - see fasync_remove_entry.
697 */
698void fasync_free(struct fasync_struct *new)
699{
700 kmem_cache_free(fasync_cache, new);
701}
702
703/*
704 * Insert a new entry into the fasync list. Return the pointer to the
705 * old one if we didn't use the new one.
706 *
707 * NOTE! It is very important that the FASYNC flag always
708 * match the state "is the filp on a fasync list".
709 */
710struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
711{
712 struct fasync_struct *fa, **fp;
713
714 spin_lock(&filp->f_lock);
715 spin_lock(&fasync_lock);
716 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
717 if (fa->fa_file != filp)
718 continue;
719
720 spin_lock_irq(&fa->fa_lock);
721 fa->fa_fd = fd;
722 spin_unlock_irq(&fa->fa_lock);
723 goto out;
724 }
725
726 spin_lock_init(&new->fa_lock);
727 new->magic = FASYNC_MAGIC;
728 new->fa_file = filp;
729 new->fa_fd = fd;
730 new->fa_next = *fapp;
731 rcu_assign_pointer(*fapp, new);
732 filp->f_flags |= FASYNC;
733
734out:
735 spin_unlock(&fasync_lock);
736 spin_unlock(&filp->f_lock);
737 return fa;
738}
739
740/*
741 * Add a fasync entry. Return negative on error, positive if
742 * added, and zero if did nothing but change an existing one.
743 */
744static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
745{
746 struct fasync_struct *new;
747
748 new = fasync_alloc();
749 if (!new)
750 return -ENOMEM;
751
752 /*
753 * fasync_insert_entry() returns the old (update) entry if
754 * it existed.
755 *
756 * So free the (unused) new entry and return 0 to let the
757 * caller know that we didn't add any new fasync entries.
758 */
759 if (fasync_insert_entry(fd, filp, fapp, new)) {
760 fasync_free(new);
761 return 0;
762 }
763
764 return 1;
765}
766
767/*
768 * fasync_helper() is used by almost all character device drivers
769 * to set up the fasync queue, and for regular files by the file
770 * lease code. It returns negative on error, 0 if it did no changes
771 * and positive if it added/deleted the entry.
772 */
773int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
774{
775 if (!on)
776 return fasync_remove_entry(filp, fapp);
777 return fasync_add_entry(fd, filp, fapp);
778}
779
780EXPORT_SYMBOL(fasync_helper);
781
782/*
783 * rcu_read_lock() is held
784 */
785static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
786{
787 while (fa) {
788 struct fown_struct *fown;
789 unsigned long flags;
790
791 if (fa->magic != FASYNC_MAGIC) {
792 printk(KERN_ERR "kill_fasync: bad magic number in "
793 "fasync_struct!\n");
794 return;
795 }
796 spin_lock_irqsave(&fa->fa_lock, flags);
797 if (fa->fa_file) {
798 fown = &fa->fa_file->f_owner;
799 /* Don't send SIGURG to processes which have not set a
800 queued signum: SIGURG has its own default signalling
801 mechanism. */
802 if (!(sig == SIGURG && fown->signum == 0))
803 send_sigio(fown, fa->fa_fd, band);
804 }
805 spin_unlock_irqrestore(&fa->fa_lock, flags);
806 fa = rcu_dereference(fa->fa_next);
807 }
808}
809
810void kill_fasync(struct fasync_struct **fp, int sig, int band)
811{
812 /* First a quick test without locking: usually
813 * the list is empty.
814 */
815 if (*fp) {
816 rcu_read_lock();
817 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
818 rcu_read_unlock();
819 }
820}
821EXPORT_SYMBOL(kill_fasync);
822
823static int __init fcntl_init(void)
824{
825 /*
826 * Please add new bits here to ensure allocation uniqueness.
827 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
828 * is defined as O_NONBLOCK on some platforms and not on others.
829 */
830 BUILD_BUG_ON(19 - 1 /* for O_RDONLY being 0 */ != HWEIGHT32(
831 O_RDONLY | O_WRONLY | O_RDWR |
832 O_CREAT | O_EXCL | O_NOCTTY |
833 O_TRUNC | O_APPEND | /* O_NONBLOCK | */
834 __O_SYNC | O_DSYNC | FASYNC |
835 O_DIRECT | O_LARGEFILE | O_DIRECTORY |
836 O_NOFOLLOW | O_NOATIME | O_CLOEXEC |
837 __FMODE_EXEC | O_PATH
838 ));
839
840 fasync_cache = kmem_cache_create("fasync_cache",
841 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
842 return 0;
843}
844
845module_init(fcntl_init)
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/fs/fcntl.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8#include <linux/syscalls.h>
9#include <linux/init.h>
10#include <linux/mm.h>
11#include <linux/sched/task.h>
12#include <linux/fs.h>
13#include <linux/filelock.h>
14#include <linux/file.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>
29#include <linux/rw_hint.h>
30
31#include <linux/poll.h>
32#include <asm/siginfo.h>
33#include <linux/uaccess.h>
34
35#include "internal.h"
36
37#define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
38
39static int setfl(int fd, struct file * filp, unsigned int arg)
40{
41 struct inode * inode = file_inode(filp);
42 int error = 0;
43
44 /*
45 * O_APPEND cannot be cleared if the file is marked as append-only
46 * and the file is open for write.
47 */
48 if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
49 return -EPERM;
50
51 /* O_NOATIME can only be set by the owner or superuser */
52 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
53 if (!inode_owner_or_capable(file_mnt_idmap(filp), inode))
54 return -EPERM;
55
56 /* required for strict SunOS emulation */
57 if (O_NONBLOCK != O_NDELAY)
58 if (arg & O_NDELAY)
59 arg |= O_NONBLOCK;
60
61 /* Pipe packetized mode is controlled by O_DIRECT flag */
62 if (!S_ISFIFO(inode->i_mode) &&
63 (arg & O_DIRECT) &&
64 !(filp->f_mode & FMODE_CAN_ODIRECT))
65 return -EINVAL;
66
67 if (filp->f_op->check_flags)
68 error = filp->f_op->check_flags(arg);
69 if (error)
70 return error;
71
72 /*
73 * ->fasync() is responsible for setting the FASYNC bit.
74 */
75 if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op->fasync) {
76 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
77 if (error < 0)
78 goto out;
79 if (error > 0)
80 error = 0;
81 }
82 spin_lock(&filp->f_lock);
83 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
84 filp->f_iocb_flags = iocb_flags(filp);
85 spin_unlock(&filp->f_lock);
86
87 out:
88 return error;
89}
90
91/*
92 * Allocate an file->f_owner struct if it doesn't exist, handling racing
93 * allocations correctly.
94 */
95int file_f_owner_allocate(struct file *file)
96{
97 struct fown_struct *f_owner;
98
99 f_owner = file_f_owner(file);
100 if (f_owner)
101 return 0;
102
103 f_owner = kzalloc(sizeof(struct fown_struct), GFP_KERNEL);
104 if (!f_owner)
105 return -ENOMEM;
106
107 rwlock_init(&f_owner->lock);
108 f_owner->file = file;
109 /* If someone else raced us, drop our allocation. */
110 if (unlikely(cmpxchg(&file->f_owner, NULL, f_owner)))
111 kfree(f_owner);
112 return 0;
113}
114EXPORT_SYMBOL(file_f_owner_allocate);
115
116void file_f_owner_release(struct file *file)
117{
118 struct fown_struct *f_owner;
119
120 f_owner = file_f_owner(file);
121 if (f_owner) {
122 put_pid(f_owner->pid);
123 kfree(f_owner);
124 }
125}
126
127void __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
128 int force)
129{
130 struct fown_struct *f_owner;
131
132 f_owner = file_f_owner(filp);
133 if (WARN_ON_ONCE(!f_owner))
134 return;
135
136 write_lock_irq(&f_owner->lock);
137 if (force || !f_owner->pid) {
138 put_pid(f_owner->pid);
139 f_owner->pid = get_pid(pid);
140 f_owner->pid_type = type;
141
142 if (pid) {
143 const struct cred *cred = current_cred();
144 security_file_set_fowner(filp);
145 f_owner->uid = cred->uid;
146 f_owner->euid = cred->euid;
147 }
148 }
149 write_unlock_irq(&f_owner->lock);
150}
151EXPORT_SYMBOL(__f_setown);
152
153int f_setown(struct file *filp, int who, int force)
154{
155 enum pid_type type;
156 struct pid *pid = NULL;
157 int ret = 0;
158
159 might_sleep();
160
161 type = PIDTYPE_TGID;
162 if (who < 0) {
163 /* avoid overflow below */
164 if (who == INT_MIN)
165 return -EINVAL;
166
167 type = PIDTYPE_PGID;
168 who = -who;
169 }
170
171 ret = file_f_owner_allocate(filp);
172 if (ret)
173 return ret;
174
175 rcu_read_lock();
176 if (who) {
177 pid = find_vpid(who);
178 if (!pid)
179 ret = -ESRCH;
180 }
181
182 if (!ret)
183 __f_setown(filp, pid, type, force);
184 rcu_read_unlock();
185
186 return ret;
187}
188EXPORT_SYMBOL(f_setown);
189
190void f_delown(struct file *filp)
191{
192 __f_setown(filp, NULL, PIDTYPE_TGID, 1);
193}
194
195pid_t f_getown(struct file *filp)
196{
197 pid_t pid = 0;
198 struct fown_struct *f_owner;
199
200 f_owner = file_f_owner(filp);
201 if (!f_owner)
202 return pid;
203
204 read_lock_irq(&f_owner->lock);
205 rcu_read_lock();
206 if (pid_task(f_owner->pid, f_owner->pid_type)) {
207 pid = pid_vnr(f_owner->pid);
208 if (f_owner->pid_type == PIDTYPE_PGID)
209 pid = -pid;
210 }
211 rcu_read_unlock();
212 read_unlock_irq(&f_owner->lock);
213 return pid;
214}
215
216static int f_setown_ex(struct file *filp, unsigned long arg)
217{
218 struct f_owner_ex __user *owner_p = (void __user *)arg;
219 struct f_owner_ex owner;
220 struct pid *pid;
221 int type;
222 int ret;
223
224 ret = copy_from_user(&owner, owner_p, sizeof(owner));
225 if (ret)
226 return -EFAULT;
227
228 switch (owner.type) {
229 case F_OWNER_TID:
230 type = PIDTYPE_PID;
231 break;
232
233 case F_OWNER_PID:
234 type = PIDTYPE_TGID;
235 break;
236
237 case F_OWNER_PGRP:
238 type = PIDTYPE_PGID;
239 break;
240
241 default:
242 return -EINVAL;
243 }
244
245 ret = file_f_owner_allocate(filp);
246 if (ret)
247 return ret;
248
249 rcu_read_lock();
250 pid = find_vpid(owner.pid);
251 if (owner.pid && !pid)
252 ret = -ESRCH;
253 else
254 __f_setown(filp, pid, type, 1);
255 rcu_read_unlock();
256
257 return ret;
258}
259
260static int f_getown_ex(struct file *filp, unsigned long arg)
261{
262 struct f_owner_ex __user *owner_p = (void __user *)arg;
263 struct f_owner_ex owner = {};
264 int ret = 0;
265 struct fown_struct *f_owner;
266 enum pid_type pid_type = PIDTYPE_PID;
267
268 f_owner = file_f_owner(filp);
269 if (f_owner) {
270 read_lock_irq(&f_owner->lock);
271 rcu_read_lock();
272 if (pid_task(f_owner->pid, f_owner->pid_type))
273 owner.pid = pid_vnr(f_owner->pid);
274 rcu_read_unlock();
275 pid_type = f_owner->pid_type;
276 }
277
278 switch (pid_type) {
279 case PIDTYPE_PID:
280 owner.type = F_OWNER_TID;
281 break;
282
283 case PIDTYPE_TGID:
284 owner.type = F_OWNER_PID;
285 break;
286
287 case PIDTYPE_PGID:
288 owner.type = F_OWNER_PGRP;
289 break;
290
291 default:
292 WARN_ON(1);
293 ret = -EINVAL;
294 break;
295 }
296 if (f_owner)
297 read_unlock_irq(&f_owner->lock);
298
299 if (!ret) {
300 ret = copy_to_user(owner_p, &owner, sizeof(owner));
301 if (ret)
302 ret = -EFAULT;
303 }
304 return ret;
305}
306
307#ifdef CONFIG_CHECKPOINT_RESTORE
308static int f_getowner_uids(struct file *filp, unsigned long arg)
309{
310 struct user_namespace *user_ns = current_user_ns();
311 struct fown_struct *f_owner;
312 uid_t __user *dst = (void __user *)arg;
313 uid_t src[2] = {0, 0};
314 int err;
315
316 f_owner = file_f_owner(filp);
317 if (f_owner) {
318 read_lock_irq(&f_owner->lock);
319 src[0] = from_kuid(user_ns, f_owner->uid);
320 src[1] = from_kuid(user_ns, f_owner->euid);
321 read_unlock_irq(&f_owner->lock);
322 }
323
324 err = put_user(src[0], &dst[0]);
325 err |= put_user(src[1], &dst[1]);
326
327 return err;
328}
329#else
330static int f_getowner_uids(struct file *filp, unsigned long arg)
331{
332 return -EINVAL;
333}
334#endif
335
336static bool rw_hint_valid(u64 hint)
337{
338 BUILD_BUG_ON(WRITE_LIFE_NOT_SET != RWH_WRITE_LIFE_NOT_SET);
339 BUILD_BUG_ON(WRITE_LIFE_NONE != RWH_WRITE_LIFE_NONE);
340 BUILD_BUG_ON(WRITE_LIFE_SHORT != RWH_WRITE_LIFE_SHORT);
341 BUILD_BUG_ON(WRITE_LIFE_MEDIUM != RWH_WRITE_LIFE_MEDIUM);
342 BUILD_BUG_ON(WRITE_LIFE_LONG != RWH_WRITE_LIFE_LONG);
343 BUILD_BUG_ON(WRITE_LIFE_EXTREME != RWH_WRITE_LIFE_EXTREME);
344
345 switch (hint) {
346 case RWH_WRITE_LIFE_NOT_SET:
347 case RWH_WRITE_LIFE_NONE:
348 case RWH_WRITE_LIFE_SHORT:
349 case RWH_WRITE_LIFE_MEDIUM:
350 case RWH_WRITE_LIFE_LONG:
351 case RWH_WRITE_LIFE_EXTREME:
352 return true;
353 default:
354 return false;
355 }
356}
357
358static long fcntl_get_rw_hint(struct file *file, unsigned int cmd,
359 unsigned long arg)
360{
361 struct inode *inode = file_inode(file);
362 u64 __user *argp = (u64 __user *)arg;
363 u64 hint = READ_ONCE(inode->i_write_hint);
364
365 if (copy_to_user(argp, &hint, sizeof(*argp)))
366 return -EFAULT;
367 return 0;
368}
369
370static long fcntl_set_rw_hint(struct file *file, unsigned int cmd,
371 unsigned long arg)
372{
373 struct inode *inode = file_inode(file);
374 u64 __user *argp = (u64 __user *)arg;
375 u64 hint;
376
377 if (!inode_owner_or_capable(file_mnt_idmap(file), inode))
378 return -EPERM;
379
380 if (copy_from_user(&hint, argp, sizeof(hint)))
381 return -EFAULT;
382 if (!rw_hint_valid(hint))
383 return -EINVAL;
384
385 WRITE_ONCE(inode->i_write_hint, hint);
386
387 /*
388 * file->f_mapping->host may differ from inode. As an example,
389 * blkdev_open() modifies file->f_mapping.
390 */
391 if (file->f_mapping->host != inode)
392 WRITE_ONCE(file->f_mapping->host->i_write_hint, hint);
393
394 return 0;
395}
396
397/* Is the file descriptor a dup of the file? */
398static long f_dupfd_query(int fd, struct file *filp)
399{
400 CLASS(fd_raw, f)(fd);
401
402 if (fd_empty(f))
403 return -EBADF;
404
405 /*
406 * We can do the 'fdput()' immediately, as the only thing that
407 * matters is the pointer value which isn't changed by the fdput.
408 *
409 * Technically we didn't need a ref at all, and 'fdget()' was
410 * overkill, but given our lockless file pointer lookup, the
411 * alternatives are complicated.
412 */
413 return fd_file(f) == filp;
414}
415
416/* Let the caller figure out whether a given file was just created. */
417static long f_created_query(const struct file *filp)
418{
419 return !!(filp->f_mode & FMODE_CREATED);
420}
421
422static int f_owner_sig(struct file *filp, int signum, bool setsig)
423{
424 int ret = 0;
425 struct fown_struct *f_owner;
426
427 might_sleep();
428
429 if (setsig) {
430 if (!valid_signal(signum))
431 return -EINVAL;
432
433 ret = file_f_owner_allocate(filp);
434 if (ret)
435 return ret;
436 }
437
438 f_owner = file_f_owner(filp);
439 if (setsig)
440 f_owner->signum = signum;
441 else if (f_owner)
442 ret = f_owner->signum;
443 return ret;
444}
445
446static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
447 struct file *filp)
448{
449 void __user *argp = (void __user *)arg;
450 int argi = (int)arg;
451 struct flock flock;
452 long err = -EINVAL;
453
454 switch (cmd) {
455 case F_CREATED_QUERY:
456 err = f_created_query(filp);
457 break;
458 case F_DUPFD:
459 err = f_dupfd(argi, filp, 0);
460 break;
461 case F_DUPFD_CLOEXEC:
462 err = f_dupfd(argi, filp, O_CLOEXEC);
463 break;
464 case F_DUPFD_QUERY:
465 err = f_dupfd_query(argi, filp);
466 break;
467 case F_GETFD:
468 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
469 break;
470 case F_SETFD:
471 err = 0;
472 set_close_on_exec(fd, argi & FD_CLOEXEC);
473 break;
474 case F_GETFL:
475 err = filp->f_flags;
476 break;
477 case F_SETFL:
478 err = setfl(fd, filp, argi);
479 break;
480#if BITS_PER_LONG != 32
481 /* 32-bit arches must use fcntl64() */
482 case F_OFD_GETLK:
483#endif
484 case F_GETLK:
485 if (copy_from_user(&flock, argp, sizeof(flock)))
486 return -EFAULT;
487 err = fcntl_getlk(filp, cmd, &flock);
488 if (!err && copy_to_user(argp, &flock, sizeof(flock)))
489 return -EFAULT;
490 break;
491#if BITS_PER_LONG != 32
492 /* 32-bit arches must use fcntl64() */
493 case F_OFD_SETLK:
494 case F_OFD_SETLKW:
495 fallthrough;
496#endif
497 case F_SETLK:
498 case F_SETLKW:
499 if (copy_from_user(&flock, argp, sizeof(flock)))
500 return -EFAULT;
501 err = fcntl_setlk(fd, filp, cmd, &flock);
502 break;
503 case F_GETOWN:
504 /*
505 * XXX If f_owner is a process group, the
506 * negative return value will get converted
507 * into an error. Oops. If we keep the
508 * current syscall conventions, the only way
509 * to fix this will be in libc.
510 */
511 err = f_getown(filp);
512 force_successful_syscall_return();
513 break;
514 case F_SETOWN:
515 err = f_setown(filp, argi, 1);
516 break;
517 case F_GETOWN_EX:
518 err = f_getown_ex(filp, arg);
519 break;
520 case F_SETOWN_EX:
521 err = f_setown_ex(filp, arg);
522 break;
523 case F_GETOWNER_UIDS:
524 err = f_getowner_uids(filp, arg);
525 break;
526 case F_GETSIG:
527 err = f_owner_sig(filp, 0, false);
528 break;
529 case F_SETSIG:
530 err = f_owner_sig(filp, argi, true);
531 break;
532 case F_GETLEASE:
533 err = fcntl_getlease(filp);
534 break;
535 case F_SETLEASE:
536 err = fcntl_setlease(fd, filp, argi);
537 break;
538 case F_NOTIFY:
539 err = fcntl_dirnotify(fd, filp, argi);
540 break;
541 case F_SETPIPE_SZ:
542 case F_GETPIPE_SZ:
543 err = pipe_fcntl(filp, cmd, argi);
544 break;
545 case F_ADD_SEALS:
546 case F_GET_SEALS:
547 err = memfd_fcntl(filp, cmd, argi);
548 break;
549 case F_GET_RW_HINT:
550 err = fcntl_get_rw_hint(filp, cmd, arg);
551 break;
552 case F_SET_RW_HINT:
553 err = fcntl_set_rw_hint(filp, cmd, arg);
554 break;
555 default:
556 break;
557 }
558 return err;
559}
560
561static int check_fcntl_cmd(unsigned cmd)
562{
563 switch (cmd) {
564 case F_CREATED_QUERY:
565 case F_DUPFD:
566 case F_DUPFD_CLOEXEC:
567 case F_DUPFD_QUERY:
568 case F_GETFD:
569 case F_SETFD:
570 case F_GETFL:
571 return 1;
572 }
573 return 0;
574}
575
576SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
577{
578 CLASS(fd_raw, f)(fd);
579 long err;
580
581 if (fd_empty(f))
582 return -EBADF;
583
584 if (unlikely(fd_file(f)->f_mode & FMODE_PATH)) {
585 if (!check_fcntl_cmd(cmd))
586 return -EBADF;
587 }
588
589 err = security_file_fcntl(fd_file(f), cmd, arg);
590 if (!err)
591 err = do_fcntl(fd, cmd, arg, fd_file(f));
592
593 return err;
594}
595
596#if BITS_PER_LONG == 32
597SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
598 unsigned long, arg)
599{
600 void __user *argp = (void __user *)arg;
601 CLASS(fd_raw, f)(fd);
602 struct flock64 flock;
603 long err;
604
605 if (fd_empty(f))
606 return -EBADF;
607
608 if (unlikely(fd_file(f)->f_mode & FMODE_PATH)) {
609 if (!check_fcntl_cmd(cmd))
610 return -EBADF;
611 }
612
613 err = security_file_fcntl(fd_file(f), cmd, arg);
614 if (err)
615 return err;
616
617 switch (cmd) {
618 case F_GETLK64:
619 case F_OFD_GETLK:
620 err = -EFAULT;
621 if (copy_from_user(&flock, argp, sizeof(flock)))
622 break;
623 err = fcntl_getlk64(fd_file(f), cmd, &flock);
624 if (!err && copy_to_user(argp, &flock, sizeof(flock)))
625 err = -EFAULT;
626 break;
627 case F_SETLK64:
628 case F_SETLKW64:
629 case F_OFD_SETLK:
630 case F_OFD_SETLKW:
631 err = -EFAULT;
632 if (copy_from_user(&flock, argp, sizeof(flock)))
633 break;
634 err = fcntl_setlk64(fd, fd_file(f), cmd, &flock);
635 break;
636 default:
637 err = do_fcntl(fd, cmd, arg, fd_file(f));
638 break;
639 }
640 return err;
641}
642#endif
643
644#ifdef CONFIG_COMPAT
645/* careful - don't use anywhere else */
646#define copy_flock_fields(dst, src) \
647 (dst)->l_type = (src)->l_type; \
648 (dst)->l_whence = (src)->l_whence; \
649 (dst)->l_start = (src)->l_start; \
650 (dst)->l_len = (src)->l_len; \
651 (dst)->l_pid = (src)->l_pid;
652
653static int get_compat_flock(struct flock *kfl, const struct compat_flock __user *ufl)
654{
655 struct compat_flock fl;
656
657 if (copy_from_user(&fl, ufl, sizeof(struct compat_flock)))
658 return -EFAULT;
659 copy_flock_fields(kfl, &fl);
660 return 0;
661}
662
663static int get_compat_flock64(struct flock *kfl, const struct compat_flock64 __user *ufl)
664{
665 struct compat_flock64 fl;
666
667 if (copy_from_user(&fl, ufl, sizeof(struct compat_flock64)))
668 return -EFAULT;
669 copy_flock_fields(kfl, &fl);
670 return 0;
671}
672
673static int put_compat_flock(const struct flock *kfl, struct compat_flock __user *ufl)
674{
675 struct compat_flock fl;
676
677 memset(&fl, 0, sizeof(struct compat_flock));
678 copy_flock_fields(&fl, kfl);
679 if (copy_to_user(ufl, &fl, sizeof(struct compat_flock)))
680 return -EFAULT;
681 return 0;
682}
683
684static int put_compat_flock64(const struct flock *kfl, struct compat_flock64 __user *ufl)
685{
686 struct compat_flock64 fl;
687
688 BUILD_BUG_ON(sizeof(kfl->l_start) > sizeof(ufl->l_start));
689 BUILD_BUG_ON(sizeof(kfl->l_len) > sizeof(ufl->l_len));
690
691 memset(&fl, 0, sizeof(struct compat_flock64));
692 copy_flock_fields(&fl, kfl);
693 if (copy_to_user(ufl, &fl, sizeof(struct compat_flock64)))
694 return -EFAULT;
695 return 0;
696}
697#undef copy_flock_fields
698
699static unsigned int
700convert_fcntl_cmd(unsigned int cmd)
701{
702 switch (cmd) {
703 case F_GETLK64:
704 return F_GETLK;
705 case F_SETLK64:
706 return F_SETLK;
707 case F_SETLKW64:
708 return F_SETLKW;
709 }
710
711 return cmd;
712}
713
714/*
715 * GETLK was successful and we need to return the data, but it needs to fit in
716 * the compat structure.
717 * l_start shouldn't be too big, unless the original start + end is greater than
718 * COMPAT_OFF_T_MAX, in which case the app was asking for trouble, so we return
719 * -EOVERFLOW in that case. l_len could be too big, in which case we just
720 * truncate it, and only allow the app to see that part of the conflicting lock
721 * that might make sense to it anyway
722 */
723static int fixup_compat_flock(struct flock *flock)
724{
725 if (flock->l_start > COMPAT_OFF_T_MAX)
726 return -EOVERFLOW;
727 if (flock->l_len > COMPAT_OFF_T_MAX)
728 flock->l_len = COMPAT_OFF_T_MAX;
729 return 0;
730}
731
732static long do_compat_fcntl64(unsigned int fd, unsigned int cmd,
733 compat_ulong_t arg)
734{
735 CLASS(fd_raw, f)(fd);
736 struct flock flock;
737 long err;
738
739 if (fd_empty(f))
740 return -EBADF;
741
742 if (unlikely(fd_file(f)->f_mode & FMODE_PATH)) {
743 if (!check_fcntl_cmd(cmd))
744 return -EBADF;
745 }
746
747 err = security_file_fcntl(fd_file(f), cmd, arg);
748 if (err)
749 return err;
750
751 switch (cmd) {
752 case F_GETLK:
753 err = get_compat_flock(&flock, compat_ptr(arg));
754 if (err)
755 break;
756 err = fcntl_getlk(fd_file(f), convert_fcntl_cmd(cmd), &flock);
757 if (err)
758 break;
759 err = fixup_compat_flock(&flock);
760 if (!err)
761 err = put_compat_flock(&flock, compat_ptr(arg));
762 break;
763 case F_GETLK64:
764 case F_OFD_GETLK:
765 err = get_compat_flock64(&flock, compat_ptr(arg));
766 if (err)
767 break;
768 err = fcntl_getlk(fd_file(f), convert_fcntl_cmd(cmd), &flock);
769 if (!err)
770 err = put_compat_flock64(&flock, compat_ptr(arg));
771 break;
772 case F_SETLK:
773 case F_SETLKW:
774 err = get_compat_flock(&flock, compat_ptr(arg));
775 if (err)
776 break;
777 err = fcntl_setlk(fd, fd_file(f), convert_fcntl_cmd(cmd), &flock);
778 break;
779 case F_SETLK64:
780 case F_SETLKW64:
781 case F_OFD_SETLK:
782 case F_OFD_SETLKW:
783 err = get_compat_flock64(&flock, compat_ptr(arg));
784 if (err)
785 break;
786 err = fcntl_setlk(fd, fd_file(f), convert_fcntl_cmd(cmd), &flock);
787 break;
788 default:
789 err = do_fcntl(fd, cmd, arg, fd_file(f));
790 break;
791 }
792 return err;
793}
794
795COMPAT_SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
796 compat_ulong_t, arg)
797{
798 return do_compat_fcntl64(fd, cmd, arg);
799}
800
801COMPAT_SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd,
802 compat_ulong_t, arg)
803{
804 switch (cmd) {
805 case F_GETLK64:
806 case F_SETLK64:
807 case F_SETLKW64:
808 case F_OFD_GETLK:
809 case F_OFD_SETLK:
810 case F_OFD_SETLKW:
811 return -EINVAL;
812 }
813 return do_compat_fcntl64(fd, cmd, arg);
814}
815#endif
816
817/* Table to convert sigio signal codes into poll band bitmaps */
818
819static const __poll_t band_table[NSIGPOLL] = {
820 EPOLLIN | EPOLLRDNORM, /* POLL_IN */
821 EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND, /* POLL_OUT */
822 EPOLLIN | EPOLLRDNORM | EPOLLMSG, /* POLL_MSG */
823 EPOLLERR, /* POLL_ERR */
824 EPOLLPRI | EPOLLRDBAND, /* POLL_PRI */
825 EPOLLHUP | EPOLLERR /* POLL_HUP */
826};
827
828static inline int sigio_perm(struct task_struct *p,
829 struct fown_struct *fown, int sig)
830{
831 const struct cred *cred;
832 int ret;
833
834 rcu_read_lock();
835 cred = __task_cred(p);
836 ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
837 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
838 uid_eq(fown->uid, cred->suid) || uid_eq(fown->uid, cred->uid)) &&
839 !security_file_send_sigiotask(p, fown, sig));
840 rcu_read_unlock();
841 return ret;
842}
843
844static void send_sigio_to_task(struct task_struct *p,
845 struct fown_struct *fown,
846 int fd, int reason, enum pid_type type)
847{
848 /*
849 * F_SETSIG can change ->signum lockless in parallel, make
850 * sure we read it once and use the same value throughout.
851 */
852 int signum = READ_ONCE(fown->signum);
853
854 if (!sigio_perm(p, fown, signum))
855 return;
856
857 switch (signum) {
858 default: {
859 kernel_siginfo_t si;
860
861 /* Queue a rt signal with the appropriate fd as its
862 value. We use SI_SIGIO as the source, not
863 SI_KERNEL, since kernel signals always get
864 delivered even if we can't queue. Failure to
865 queue in this case _should_ be reported; we fall
866 back to SIGIO in that case. --sct */
867 clear_siginfo(&si);
868 si.si_signo = signum;
869 si.si_errno = 0;
870 si.si_code = reason;
871 /*
872 * Posix definies POLL_IN and friends to be signal
873 * specific si_codes for SIG_POLL. Linux extended
874 * these si_codes to other signals in a way that is
875 * ambiguous if other signals also have signal
876 * specific si_codes. In that case use SI_SIGIO instead
877 * to remove the ambiguity.
878 */
879 if ((signum != SIGPOLL) && sig_specific_sicodes(signum))
880 si.si_code = SI_SIGIO;
881
882 /* Make sure we are called with one of the POLL_*
883 reasons, otherwise we could leak kernel stack into
884 userspace. */
885 BUG_ON((reason < POLL_IN) || ((reason - POLL_IN) >= NSIGPOLL));
886 if (reason - POLL_IN >= NSIGPOLL)
887 si.si_band = ~0L;
888 else
889 si.si_band = mangle_poll(band_table[reason - POLL_IN]);
890 si.si_fd = fd;
891 if (!do_send_sig_info(signum, &si, p, type))
892 break;
893 }
894 fallthrough; /* fall back on the old plain SIGIO signal */
895 case 0:
896 do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, type);
897 }
898}
899
900void send_sigio(struct fown_struct *fown, int fd, int band)
901{
902 struct task_struct *p;
903 enum pid_type type;
904 unsigned long flags;
905 struct pid *pid;
906
907 read_lock_irqsave(&fown->lock, flags);
908
909 type = fown->pid_type;
910 pid = fown->pid;
911 if (!pid)
912 goto out_unlock_fown;
913
914 if (type <= PIDTYPE_TGID) {
915 rcu_read_lock();
916 p = pid_task(pid, PIDTYPE_PID);
917 if (p)
918 send_sigio_to_task(p, fown, fd, band, type);
919 rcu_read_unlock();
920 } else {
921 read_lock(&tasklist_lock);
922 do_each_pid_task(pid, type, p) {
923 send_sigio_to_task(p, fown, fd, band, type);
924 } while_each_pid_task(pid, type, p);
925 read_unlock(&tasklist_lock);
926 }
927 out_unlock_fown:
928 read_unlock_irqrestore(&fown->lock, flags);
929}
930
931static void send_sigurg_to_task(struct task_struct *p,
932 struct fown_struct *fown, enum pid_type type)
933{
934 if (sigio_perm(p, fown, SIGURG))
935 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, type);
936}
937
938int send_sigurg(struct file *file)
939{
940 struct fown_struct *fown;
941 struct task_struct *p;
942 enum pid_type type;
943 struct pid *pid;
944 unsigned long flags;
945 int ret = 0;
946
947 fown = file_f_owner(file);
948 if (!fown)
949 return 0;
950
951 read_lock_irqsave(&fown->lock, flags);
952
953 type = fown->pid_type;
954 pid = fown->pid;
955 if (!pid)
956 goto out_unlock_fown;
957
958 ret = 1;
959
960 if (type <= PIDTYPE_TGID) {
961 rcu_read_lock();
962 p = pid_task(pid, PIDTYPE_PID);
963 if (p)
964 send_sigurg_to_task(p, fown, type);
965 rcu_read_unlock();
966 } else {
967 read_lock(&tasklist_lock);
968 do_each_pid_task(pid, type, p) {
969 send_sigurg_to_task(p, fown, type);
970 } while_each_pid_task(pid, type, p);
971 read_unlock(&tasklist_lock);
972 }
973 out_unlock_fown:
974 read_unlock_irqrestore(&fown->lock, flags);
975 return ret;
976}
977
978static DEFINE_SPINLOCK(fasync_lock);
979static struct kmem_cache *fasync_cache __ro_after_init;
980
981/*
982 * Remove a fasync entry. If successfully removed, return
983 * positive and clear the FASYNC flag. If no entry exists,
984 * do nothing and return 0.
985 *
986 * NOTE! It is very important that the FASYNC flag always
987 * match the state "is the filp on a fasync list".
988 *
989 */
990int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
991{
992 struct fasync_struct *fa, **fp;
993 int result = 0;
994
995 spin_lock(&filp->f_lock);
996 spin_lock(&fasync_lock);
997 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
998 if (fa->fa_file != filp)
999 continue;
1000
1001 write_lock_irq(&fa->fa_lock);
1002 fa->fa_file = NULL;
1003 write_unlock_irq(&fa->fa_lock);
1004
1005 *fp = fa->fa_next;
1006 kfree_rcu(fa, fa_rcu);
1007 filp->f_flags &= ~FASYNC;
1008 result = 1;
1009 break;
1010 }
1011 spin_unlock(&fasync_lock);
1012 spin_unlock(&filp->f_lock);
1013 return result;
1014}
1015
1016struct fasync_struct *fasync_alloc(void)
1017{
1018 return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
1019}
1020
1021/*
1022 * NOTE! This can be used only for unused fasync entries:
1023 * entries that actually got inserted on the fasync list
1024 * need to be released by rcu - see fasync_remove_entry.
1025 */
1026void fasync_free(struct fasync_struct *new)
1027{
1028 kmem_cache_free(fasync_cache, new);
1029}
1030
1031/*
1032 * Insert a new entry into the fasync list. Return the pointer to the
1033 * old one if we didn't use the new one.
1034 *
1035 * NOTE! It is very important that the FASYNC flag always
1036 * match the state "is the filp on a fasync list".
1037 */
1038struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
1039{
1040 struct fasync_struct *fa, **fp;
1041
1042 spin_lock(&filp->f_lock);
1043 spin_lock(&fasync_lock);
1044 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
1045 if (fa->fa_file != filp)
1046 continue;
1047
1048 write_lock_irq(&fa->fa_lock);
1049 fa->fa_fd = fd;
1050 write_unlock_irq(&fa->fa_lock);
1051 goto out;
1052 }
1053
1054 rwlock_init(&new->fa_lock);
1055 new->magic = FASYNC_MAGIC;
1056 new->fa_file = filp;
1057 new->fa_fd = fd;
1058 new->fa_next = *fapp;
1059 rcu_assign_pointer(*fapp, new);
1060 filp->f_flags |= FASYNC;
1061
1062out:
1063 spin_unlock(&fasync_lock);
1064 spin_unlock(&filp->f_lock);
1065 return fa;
1066}
1067
1068/*
1069 * Add a fasync entry. Return negative on error, positive if
1070 * added, and zero if did nothing but change an existing one.
1071 */
1072static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
1073{
1074 struct fasync_struct *new;
1075
1076 new = fasync_alloc();
1077 if (!new)
1078 return -ENOMEM;
1079
1080 /*
1081 * fasync_insert_entry() returns the old (update) entry if
1082 * it existed.
1083 *
1084 * So free the (unused) new entry and return 0 to let the
1085 * caller know that we didn't add any new fasync entries.
1086 */
1087 if (fasync_insert_entry(fd, filp, fapp, new)) {
1088 fasync_free(new);
1089 return 0;
1090 }
1091
1092 return 1;
1093}
1094
1095/*
1096 * fasync_helper() is used by almost all character device drivers
1097 * to set up the fasync queue, and for regular files by the file
1098 * lease code. It returns negative on error, 0 if it did no changes
1099 * and positive if it added/deleted the entry.
1100 */
1101int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
1102{
1103 if (!on)
1104 return fasync_remove_entry(filp, fapp);
1105 return fasync_add_entry(fd, filp, fapp);
1106}
1107
1108EXPORT_SYMBOL(fasync_helper);
1109
1110/*
1111 * rcu_read_lock() is held
1112 */
1113static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
1114{
1115 while (fa) {
1116 struct fown_struct *fown;
1117 unsigned long flags;
1118
1119 if (fa->magic != FASYNC_MAGIC) {
1120 printk(KERN_ERR "kill_fasync: bad magic number in "
1121 "fasync_struct!\n");
1122 return;
1123 }
1124 read_lock_irqsave(&fa->fa_lock, flags);
1125 if (fa->fa_file) {
1126 fown = file_f_owner(fa->fa_file);
1127 if (!fown)
1128 goto next;
1129 /* Don't send SIGURG to processes which have not set a
1130 queued signum: SIGURG has its own default signalling
1131 mechanism. */
1132 if (!(sig == SIGURG && fown->signum == 0))
1133 send_sigio(fown, fa->fa_fd, band);
1134 }
1135next:
1136 read_unlock_irqrestore(&fa->fa_lock, flags);
1137 fa = rcu_dereference(fa->fa_next);
1138 }
1139}
1140
1141void kill_fasync(struct fasync_struct **fp, int sig, int band)
1142{
1143 /* First a quick test without locking: usually
1144 * the list is empty.
1145 */
1146 if (*fp) {
1147 rcu_read_lock();
1148 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
1149 rcu_read_unlock();
1150 }
1151}
1152EXPORT_SYMBOL(kill_fasync);
1153
1154static int __init fcntl_init(void)
1155{
1156 /*
1157 * Please add new bits here to ensure allocation uniqueness.
1158 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
1159 * is defined as O_NONBLOCK on some platforms and not on others.
1160 */
1161 BUILD_BUG_ON(21 - 1 /* for O_RDONLY being 0 */ !=
1162 HWEIGHT32(
1163 (VALID_OPEN_FLAGS & ~(O_NONBLOCK | O_NDELAY)) |
1164 __FMODE_EXEC | __FMODE_NONOTIFY));
1165
1166 fasync_cache = kmem_cache_create("fasync_cache",
1167 sizeof(struct fasync_struct), 0,
1168 SLAB_PANIC | SLAB_ACCOUNT, NULL);
1169 return 0;
1170}
1171
1172module_init(fcntl_init)