projects/linux·blob master · refs

1186 lines · 26.5 KiB

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_obj(struct fown_struct);
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 long arg)
359{
360 struct inode *inode = file_inode(file);
361 u64 __user *argp = (u64 __user *)arg;
362 u64 hint = READ_ONCE(inode->i_write_hint);
363
364 if (copy_to_user(argp, &hint, sizeof(*argp)))
365 return -EFAULT;
366 return 0;
367}
368
369static long fcntl_set_rw_hint(struct file *file, unsigned long arg)
370{
371 struct inode *inode = file_inode(file);
372 u64 __user *argp = (u64 __user *)arg;
373 u64 hint;
374
375 /* nothing is ever written to it */
376 if (IS_IMMUTABLE(inode))
377 return -EPERM;
378 if (!inode_owner_or_capable(file_mnt_idmap(file), inode))
379 return -EPERM;
380
381 if (copy_from_user(&hint, argp, sizeof(hint)))
382 return -EFAULT;
383 if (!rw_hint_valid(hint))
384 return -EINVAL;
385
386 WRITE_ONCE(inode->i_write_hint, hint);
387
388 /*
389 * file->f_mapping->host may differ from inode. As an example,
390 * blkdev_open() modifies file->f_mapping.
391 */
392 if (file->f_mapping->host != inode)
393 WRITE_ONCE(file->f_mapping->host->i_write_hint, hint);
394
395 return 0;
396}
397
398/* Is the file descriptor a dup of the file? */
399static long f_dupfd_query(int fd, struct file *filp)
400{
401 CLASS(fd_raw, f)(fd);
402
403 if (fd_empty(f))
404 return -EBADF;
405
406 /*
407 * We can do the 'fdput()' immediately, as the only thing that
408 * matters is the pointer value which isn't changed by the fdput.
409 *
410 * Technically we didn't need a ref at all, and 'fdget()' was
411 * overkill, but given our lockless file pointer lookup, the
412 * alternatives are complicated.
413 */
414 return fd_file(f) == filp;
415}
416
417/* Let the caller figure out whether a given file was just created. */
418static long f_created_query(const struct file *filp)
419{
420 return !!(filp->f_mode & FMODE_CREATED);
421}
422
423static int f_owner_sig(struct file *filp, int signum, bool setsig)
424{
425 int ret = 0;
426 struct fown_struct *f_owner;
427
428 might_sleep();
429
430 if (setsig) {
431 if (!valid_signal(signum))
432 return -EINVAL;
433
434 ret = file_f_owner_allocate(filp);
435 if (ret)
436 return ret;
437 }
438
439 f_owner = file_f_owner(filp);
440 if (setsig)
441 f_owner->signum = signum;
442 else if (f_owner)
443 ret = f_owner->signum;
444 return ret;
445}
446
447static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
448 struct file *filp)
449{
450 void __user *argp = (void __user *)arg;
451 struct delegation deleg;
452 int argi = (int)arg;
453 struct flock flock;
454 long err = -EINVAL;
455
456 switch (cmd) {
457 case F_CREATED_QUERY:
458 err = f_created_query(filp);
459 break;
460 case F_DUPFD:
461 err = f_dupfd(argi, filp, 0);
462 break;
463 case F_DUPFD_CLOEXEC:
464 err = f_dupfd(argi, filp, O_CLOEXEC);
465 break;
466 case F_DUPFD_QUERY:
467 err = f_dupfd_query(argi, filp);
468 break;
469 case F_GETFD:
470 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
471 break;
472 case F_SETFD:
473 err = 0;
474 set_close_on_exec(fd, argi & FD_CLOEXEC);
475 break;
476 case F_GETFL:
477 err = filp->f_flags;
478 break;
479 case F_SETFL:
480 err = setfl(fd, filp, argi);
481 break;
482#if BITS_PER_LONG != 32
483 /* 32-bit arches must use fcntl64() */
484 case F_OFD_GETLK:
485#endif
486 case F_GETLK:
487 if (copy_from_user(&flock, argp, sizeof(flock)))
488 return -EFAULT;
489 err = fcntl_getlk(filp, cmd, &flock);
490 if (!err && copy_to_user(argp, &flock, sizeof(flock)))
491 return -EFAULT;
492 break;
493#if BITS_PER_LONG != 32
494 /* 32-bit arches must use fcntl64() */
495 case F_OFD_SETLK:
496 case F_OFD_SETLKW:
497 fallthrough;
498#endif
499 case F_SETLK:
500 case F_SETLKW:
501 if (copy_from_user(&flock, argp, sizeof(flock)))
502 return -EFAULT;
503 err = fcntl_setlk(fd, filp, cmd, &flock);
504 break;
505 case F_GETOWN:
506 /*
507 * XXX If f_owner is a process group, the
508 * negative return value will get converted
509 * into an error. Oops. If we keep the
510 * current syscall conventions, the only way
511 * to fix this will be in libc.
512 */
513 err = f_getown(filp);
514 force_successful_syscall_return();
515 break;
516 case F_SETOWN:
517 err = f_setown(filp, argi, 1);
518 break;
519 case F_GETOWN_EX:
520 err = f_getown_ex(filp, arg);
521 break;
522 case F_SETOWN_EX:
523 err = f_setown_ex(filp, arg);
524 break;
525 case F_GETOWNER_UIDS:
526 err = f_getowner_uids(filp, arg);
527 break;
528 case F_GETSIG:
529 err = f_owner_sig(filp, 0, false);
530 break;
531 case F_SETSIG:
532 err = f_owner_sig(filp, argi, true);
533 break;
534 case F_GETLEASE:
535 err = fcntl_getlease(filp);
536 break;
537 case F_SETLEASE:
538 err = fcntl_setlease(fd, filp, argi);
539 break;
540 case F_NOTIFY:
541 err = fcntl_dirnotify(fd, filp, argi);
542 break;
543 case F_SETPIPE_SZ:
544 case F_GETPIPE_SZ:
545 err = pipe_fcntl(filp, cmd, argi);
546 break;
547 case F_ADD_SEALS:
548 case F_GET_SEALS:
549 err = memfd_fcntl(filp, cmd, argi);
550 break;
551 case F_GET_RW_HINT:
552 err = fcntl_get_rw_hint(filp, arg);
553 break;
554 case F_SET_RW_HINT:
555 err = fcntl_set_rw_hint(filp, arg);
556 break;
557 case F_GETDELEG:
558 if (copy_from_user(&deleg, argp, sizeof(deleg)))
559 return -EFAULT;
560 err = fcntl_getdeleg(filp, &deleg);
561 if (!err && copy_to_user(argp, &deleg, sizeof(deleg)))
562 return -EFAULT;
563 break;
564 case F_SETDELEG:
565 if (copy_from_user(&deleg, argp, sizeof(deleg)))
566 return -EFAULT;
567 err = fcntl_setdeleg(fd, filp, &deleg);
568 break;
569 default:
570 break;
571 }
572 return err;
573}
574
575static int check_fcntl_cmd(unsigned cmd)
576{
577 switch (cmd) {
578 case F_CREATED_QUERY:
579 case F_DUPFD:
580 case F_DUPFD_CLOEXEC:
581 case F_DUPFD_QUERY:
582 case F_GETFD:
583 case F_SETFD:
584 case F_GETFL:
585 return 1;
586 }
587 return 0;
588}
589
590SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
591{
592 CLASS(fd_raw, f)(fd);
593 long err;
594
595 if (fd_empty(f))
596 return -EBADF;
597
598 if (unlikely(fd_file(f)->f_mode & FMODE_PATH)) {
599 if (!check_fcntl_cmd(cmd))
600 return -EBADF;
601 }
602
603 err = security_file_fcntl(fd_file(f), cmd, arg);
604 if (!err)
605 err = do_fcntl(fd, cmd, arg, fd_file(f));
606
607 return err;
608}
609
610#if BITS_PER_LONG == 32
611SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
612 unsigned long, arg)
613{
614 void __user *argp = (void __user *)arg;
615 CLASS(fd_raw, f)(fd);
616 struct flock64 flock;
617 long err;
618
619 if (fd_empty(f))
620 return -EBADF;
621
622 if (unlikely(fd_file(f)->f_mode & FMODE_PATH)) {
623 if (!check_fcntl_cmd(cmd))
624 return -EBADF;
625 }
626
627 err = security_file_fcntl(fd_file(f), cmd, arg);
628 if (err)
629 return err;
630
631 switch (cmd) {
632 case F_GETLK64:
633 case F_OFD_GETLK:
634 err = -EFAULT;
635 if (copy_from_user(&flock, argp, sizeof(flock)))
636 break;
637 err = fcntl_getlk64(fd_file(f), cmd, &flock);
638 if (!err && copy_to_user(argp, &flock, sizeof(flock)))
639 err = -EFAULT;
640 break;
641 case F_SETLK64:
642 case F_SETLKW64:
643 case F_OFD_SETLK:
644 case F_OFD_SETLKW:
645 err = -EFAULT;
646 if (copy_from_user(&flock, argp, sizeof(flock)))
647 break;
648 err = fcntl_setlk64(fd, fd_file(f), cmd, &flock);
649 break;
650 default:
651 err = do_fcntl(fd, cmd, arg, fd_file(f));
652 break;
653 }
654 return err;
655}
656#endif
657
658#ifdef CONFIG_COMPAT
659/* careful - don't use anywhere else */
660#define copy_flock_fields(dst, src) \
661 (dst)->l_type = (src)->l_type; \
662 (dst)->l_whence = (src)->l_whence; \
663 (dst)->l_start = (src)->l_start; \
664 (dst)->l_len = (src)->l_len; \
665 (dst)->l_pid = (src)->l_pid;
666
667static int get_compat_flock(struct flock *kfl, const struct compat_flock __user *ufl)
668{
669 struct compat_flock fl;
670
671 if (copy_from_user(&fl, ufl, sizeof(struct compat_flock)))
672 return -EFAULT;
673 copy_flock_fields(kfl, &fl);
674 return 0;
675}
676
677static int get_compat_flock64(struct flock *kfl, const struct compat_flock64 __user *ufl)
678{
679 struct compat_flock64 fl;
680
681 if (copy_from_user(&fl, ufl, sizeof(struct compat_flock64)))
682 return -EFAULT;
683 copy_flock_fields(kfl, &fl);
684 return 0;
685}
686
687static int put_compat_flock(const struct flock *kfl, struct compat_flock __user *ufl)
688{
689 struct compat_flock fl;
690
691 memset(&fl, 0, sizeof(struct compat_flock));
692 copy_flock_fields(&fl, kfl);
693 if (copy_to_user(ufl, &fl, sizeof(struct compat_flock)))
694 return -EFAULT;
695 return 0;
696}
697
698static int put_compat_flock64(const struct flock *kfl, struct compat_flock64 __user *ufl)
699{
700 struct compat_flock64 fl;
701
702 BUILD_BUG_ON(sizeof(kfl->l_start) > sizeof(ufl->l_start));
703 BUILD_BUG_ON(sizeof(kfl->l_len) > sizeof(ufl->l_len));
704
705 memset(&fl, 0, sizeof(struct compat_flock64));
706 copy_flock_fields(&fl, kfl);
707 if (copy_to_user(ufl, &fl, sizeof(struct compat_flock64)))
708 return -EFAULT;
709 return 0;
710}
711#undef copy_flock_fields
712
713static unsigned int
714convert_fcntl_cmd(unsigned int cmd)
715{
716 switch (cmd) {
717 case F_GETLK64:
718 return F_GETLK;
719 case F_SETLK64:
720 return F_SETLK;
721 case F_SETLKW64:
722 return F_SETLKW;
723 }
724
725 return cmd;
726}
727
728/*
729 * GETLK was successful and we need to return the data, but it needs to fit in
730 * the compat structure.
731 * l_start shouldn't be too big, unless the original start + end is greater than
732 * COMPAT_OFF_T_MAX, in which case the app was asking for trouble, so we return
733 * -EOVERFLOW in that case. l_len could be too big, in which case we just
734 * truncate it, and only allow the app to see that part of the conflicting lock
735 * that might make sense to it anyway
736 */
737static int fixup_compat_flock(struct flock *flock)
738{
739 if (flock->l_start > COMPAT_OFF_T_MAX)
740 return -EOVERFLOW;
741 if (flock->l_len > COMPAT_OFF_T_MAX)
742 flock->l_len = COMPAT_OFF_T_MAX;
743 return 0;
744}
745
746static long do_compat_fcntl64(unsigned int fd, unsigned int cmd,
747 compat_ulong_t arg)
748{
749 CLASS(fd_raw, f)(fd);
750 struct flock flock;
751 long err;
752
753 if (fd_empty(f))
754 return -EBADF;
755
756 if (unlikely(fd_file(f)->f_mode & FMODE_PATH)) {
757 if (!check_fcntl_cmd(cmd))
758 return -EBADF;
759 }
760
761 err = security_file_fcntl(fd_file(f), cmd, arg);
762 if (err)
763 return err;
764
765 switch (cmd) {
766 case F_GETLK:
767 err = get_compat_flock(&flock, compat_ptr(arg));
768 if (err)
769 break;
770 err = fcntl_getlk(fd_file(f), convert_fcntl_cmd(cmd), &flock);
771 if (err)
772 break;
773 err = fixup_compat_flock(&flock);
774 if (!err)
775 err = put_compat_flock(&flock, compat_ptr(arg));
776 break;
777 case F_GETLK64:
778 case F_OFD_GETLK:
779 err = get_compat_flock64(&flock, compat_ptr(arg));
780 if (err)
781 break;
782 err = fcntl_getlk(fd_file(f), convert_fcntl_cmd(cmd), &flock);
783 if (!err)
784 err = put_compat_flock64(&flock, compat_ptr(arg));
785 break;
786 case F_SETLK:
787 case F_SETLKW:
788 err = get_compat_flock(&flock, compat_ptr(arg));
789 if (err)
790 break;
791 err = fcntl_setlk(fd, fd_file(f), convert_fcntl_cmd(cmd), &flock);
792 break;
793 case F_SETLK64:
794 case F_SETLKW64:
795 case F_OFD_SETLK:
796 case F_OFD_SETLKW:
797 err = get_compat_flock64(&flock, compat_ptr(arg));
798 if (err)
799 break;
800 err = fcntl_setlk(fd, fd_file(f), convert_fcntl_cmd(cmd), &flock);
801 break;
802 default:
803 err = do_fcntl(fd, cmd, arg, fd_file(f));
804 break;
805 }
806 return err;
807}
808
809COMPAT_SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
810 compat_ulong_t, arg)
811{
812 return do_compat_fcntl64(fd, cmd, arg);
813}
814
815COMPAT_SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd,
816 compat_ulong_t, arg)
817{
818 switch (cmd) {
819 case F_GETLK64:
820 case F_SETLK64:
821 case F_SETLKW64:
822 case F_OFD_GETLK:
823 case F_OFD_SETLK:
824 case F_OFD_SETLKW:
825 return -EINVAL;
826 }
827 return do_compat_fcntl64(fd, cmd, arg);
828}
829#endif
830
831/* Table to convert sigio signal codes into poll band bitmaps */
832
833static const __poll_t band_table[NSIGPOLL] = {
834 EPOLLIN | EPOLLRDNORM, /* POLL_IN */
835 EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND, /* POLL_OUT */
836 EPOLLIN | EPOLLRDNORM | EPOLLMSG, /* POLL_MSG */
837 EPOLLERR, /* POLL_ERR */
838 EPOLLPRI | EPOLLRDBAND, /* POLL_PRI */
839 EPOLLHUP | EPOLLERR /* POLL_HUP */
840};
841
842static inline int sigio_perm(struct task_struct *p,
843 struct fown_struct *fown, int sig)
844{
845 const struct cred *cred;
846 int ret;
847
848 rcu_read_lock();
849 cred = __task_cred(p);
850 ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
851 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
852 uid_eq(fown->uid, cred->suid) || uid_eq(fown->uid, cred->uid)) &&
853 !security_file_send_sigiotask(p, fown, sig));
854 rcu_read_unlock();
855 return ret;
856}
857
858static void send_sigio_to_task(struct task_struct *p,
859 struct fown_struct *fown,
860 int fd, int reason, enum pid_type type)
861{
862 /*
863 * F_SETSIG can change ->signum lockless in parallel, make
864 * sure we read it once and use the same value throughout.
865 */
866 int signum = READ_ONCE(fown->signum);
867
868 if (!sigio_perm(p, fown, signum))
869 return;
870
871 switch (signum) {
872 default: {
873 kernel_siginfo_t si;
874
875 /* Queue a rt signal with the appropriate fd as its
876 value. We use SI_SIGIO as the source, not
877 SI_KERNEL, since kernel signals always get
878 delivered even if we can't queue. Failure to
879 queue in this case _should_ be reported; we fall
880 back to SIGIO in that case. --sct */
881 clear_siginfo(&si);
882 si.si_signo = signum;
883 si.si_errno = 0;
884 si.si_code = reason;
885 /*
886 * Posix definies POLL_IN and friends to be signal
887 * specific si_codes for SIG_POLL. Linux extended
888 * these si_codes to other signals in a way that is
889 * ambiguous if other signals also have signal
890 * specific si_codes. In that case use SI_SIGIO instead
891 * to remove the ambiguity.
892 */
893 if ((signum != SIGPOLL) && sig_specific_sicodes(signum))
894 si.si_code = SI_SIGIO;
895
896 /* Make sure we are called with one of the POLL_*
897 reasons, otherwise we could leak kernel stack into
898 userspace. */
899 BUG_ON((reason < POLL_IN) || ((reason - POLL_IN) >= NSIGPOLL));
900 if (reason - POLL_IN >= NSIGPOLL)
901 si.si_band = ~0L;
902 else
903 si.si_band = mangle_poll(band_table[reason - POLL_IN]);
904 si.si_fd = fd;
905 if (!do_send_sig_info(signum, &si, p, type))
906 break;
907 }
908 fallthrough; /* fall back on the old plain SIGIO signal */
909 case 0:
910 do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, type);
911 }
912}
913
914void send_sigio(struct fown_struct *fown, int fd, int band)
915{
916 struct task_struct *p;
917 enum pid_type type;
918 unsigned long flags;
919 struct pid *pid;
920
921 read_lock_irqsave(&fown->lock, flags);
922
923 type = fown->pid_type;
924 pid = fown->pid;
925 if (!pid)
926 goto out_unlock_fown;
927
928 if (type <= PIDTYPE_TGID) {
929 rcu_read_lock();
930 p = pid_task(pid, PIDTYPE_PID);
931 if (p)
932 send_sigio_to_task(p, fown, fd, band, type);
933 rcu_read_unlock();
934 } else {
935 rcu_read_lock();
936 do_each_pid_task(pid, type, p) {
937 send_sigio_to_task(p, fown, fd, band, type);
938 } while_each_pid_task(pid, type, p);
939 rcu_read_unlock();
940 }
941 out_unlock_fown:
942 read_unlock_irqrestore(&fown->lock, flags);
943}
944
945static void send_sigurg_to_task(struct task_struct *p,
946 struct fown_struct *fown, enum pid_type type)
947{
948 if (sigio_perm(p, fown, SIGURG))
949 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, type);
950}
951
952int send_sigurg(struct file *file)
953{
954 struct fown_struct *fown;
955 struct task_struct *p;
956 enum pid_type type;
957 struct pid *pid;
958 unsigned long flags;
959 int ret = 0;
960
961 fown = file_f_owner(file);
962 if (!fown)
963 return 0;
964
965 read_lock_irqsave(&fown->lock, flags);
966
967 type = fown->pid_type;
968 pid = fown->pid;
969 if (!pid)
970 goto out_unlock_fown;
971
972 ret = 1;
973
974 if (type <= PIDTYPE_TGID) {
975 rcu_read_lock();
976 p = pid_task(pid, PIDTYPE_PID);
977 if (p)
978 send_sigurg_to_task(p, fown, type);
979 rcu_read_unlock();
980 } else {
981 rcu_read_lock();
982 do_each_pid_task(pid, type, p) {
983 send_sigurg_to_task(p, fown, type);
984 } while_each_pid_task(pid, type, p);
985 rcu_read_unlock();
986 }
987 out_unlock_fown:
988 read_unlock_irqrestore(&fown->lock, flags);
989 return ret;
990}
991
992static DEFINE_SPINLOCK(fasync_lock);
993static struct kmem_cache *fasync_cache __ro_after_init;
994
995/*
996 * Remove a fasync entry. If successfully removed, return
997 * positive and clear the FASYNC flag. If no entry exists,
998 * do nothing and return 0.
999 *
1000 * NOTE! It is very important that the FASYNC flag always
1001 * match the state "is the filp on a fasync list".
1002 *
1003 */
1004int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
1005{
1006 struct fasync_struct *fa, **fp;
1007 int result = 0;
1008
1009 spin_lock(&filp->f_lock);
1010 spin_lock(&fasync_lock);
1011 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
1012 if (fa->fa_file != filp)
1013 continue;
1014
1015 write_lock_irq(&fa->fa_lock);
1016 fa->fa_file = NULL;
1017 write_unlock_irq(&fa->fa_lock);
1018
1019 *fp = fa->fa_next;
1020 kfree_rcu(fa, fa_rcu);
1021 filp->f_flags &= ~FASYNC;
1022 result = 1;
1023 break;
1024 }
1025 spin_unlock(&fasync_lock);
1026 spin_unlock(&filp->f_lock);
1027 return result;
1028}
1029
1030struct fasync_struct *fasync_alloc(void)
1031{
1032 return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
1033}
1034
1035/*
1036 * NOTE! This can be used only for unused fasync entries:
1037 * entries that actually got inserted on the fasync list
1038 * need to be released by rcu - see fasync_remove_entry.
1039 */
1040void fasync_free(struct fasync_struct *new)
1041{
1042 kmem_cache_free(fasync_cache, new);
1043}
1044
1045/*
1046 * Insert a new entry into the fasync list. Return the pointer to the
1047 * old one if we didn't use the new one.
1048 *
1049 * NOTE! It is very important that the FASYNC flag always
1050 * match the state "is the filp on a fasync list".
1051 */
1052struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
1053{
1054 struct fasync_struct *fa, **fp;
1055
1056 spin_lock(&filp->f_lock);
1057 spin_lock(&fasync_lock);
1058 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
1059 if (fa->fa_file != filp)
1060 continue;
1061
1062 write_lock_irq(&fa->fa_lock);
1063 fa->fa_fd = fd;
1064 write_unlock_irq(&fa->fa_lock);
1065 goto out;
1066 }
1067
1068 rwlock_init(&new->fa_lock);
1069 new->magic = FASYNC_MAGIC;
1070 new->fa_file = filp;
1071 new->fa_fd = fd;
1072 new->fa_next = *fapp;
1073 rcu_assign_pointer(*fapp, new);
1074 filp->f_flags |= FASYNC;
1075
1076out:
1077 spin_unlock(&fasync_lock);
1078 spin_unlock(&filp->f_lock);
1079 return fa;
1080}
1081
1082/*
1083 * Add a fasync entry. Return negative on error, positive if
1084 * added, and zero if did nothing but change an existing one.
1085 */
1086static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
1087{
1088 struct fasync_struct *new;
1089
1090 new = fasync_alloc();
1091 if (!new)
1092 return -ENOMEM;
1093
1094 /*
1095 * fasync_insert_entry() returns the old (update) entry if
1096 * it existed.
1097 *
1098 * So free the (unused) new entry and return 0 to let the
1099 * caller know that we didn't add any new fasync entries.
1100 */
1101 if (fasync_insert_entry(fd, filp, fapp, new)) {
1102 fasync_free(new);
1103 return 0;
1104 }
1105
1106 return 1;
1107}
1108
1109/*
1110 * fasync_helper() is used by almost all character device drivers
1111 * to set up the fasync queue, and for regular files by the file
1112 * lease code. It returns negative on error, 0 if it did no changes
1113 * and positive if it added/deleted the entry.
1114 */
1115int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
1116{
1117 if (!on)
1118 return fasync_remove_entry(filp, fapp);
1119 return fasync_add_entry(fd, filp, fapp);
1120}
1121
1122EXPORT_SYMBOL(fasync_helper);
1123
1124/*
1125 * rcu_read_lock() is held
1126 */
1127static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
1128{
1129 while (fa) {
1130 struct fown_struct *fown;
1131 unsigned long flags;
1132
1133 if (fa->magic != FASYNC_MAGIC) {
1134 printk(KERN_ERR "kill_fasync: bad magic number in "
1135 "fasync_struct!\n");
1136 return;
1137 }
1138 read_lock_irqsave(&fa->fa_lock, flags);
1139 if (fa->fa_file) {
1140 fown = file_f_owner(fa->fa_file);
1141 if (!fown)
1142 goto next;
1143 /* Don't send SIGURG to processes which have not set a
1144 queued signum: SIGURG has its own default signalling
1145 mechanism. */
1146 if (!(sig == SIGURG && fown->signum == 0))
1147 send_sigio(fown, fa->fa_fd, band);
1148 }
1149next:
1150 read_unlock_irqrestore(&fa->fa_lock, flags);
1151 fa = rcu_dereference(fa->fa_next);
1152 }
1153}
1154
1155void kill_fasync(struct fasync_struct **fp, int sig, int band)
1156{
1157 /* First a quick test without locking: usually
1158 * the list is empty.
1159 */
1160 if (*fp) {
1161 rcu_read_lock();
1162 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
1163 rcu_read_unlock();
1164 }
1165}
1166EXPORT_SYMBOL(kill_fasync);
1167
1168static int __init fcntl_init(void)
1169{
1170 /*
1171 * Please add new bits here to ensure allocation uniqueness.
1172 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
1173 * is defined as O_NONBLOCK on some platforms and not on others.
1174 */
1175 BUILD_BUG_ON(22 - 1 /* for O_RDONLY being 0 */ !=
1176 HWEIGHT32(
1177 (VALID_OPEN_FLAGS & ~(O_NONBLOCK | O_NDELAY)) |
1178 __FMODE_EXEC | __O_REGULAR));
1179
1180 fasync_cache = kmem_cache_create("fasync_cache",
1181 sizeof(struct fasync_struct), 0,
1182 SLAB_PANIC | SLAB_ACCOUNT, NULL);
1183 return 0;
1184}
1185
1186module_init(fcntl_init)