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