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