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

source code of linux/fs/fcntl.c