fcntl.c revision 2f488f69
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#include <linux/mount.h>
29
30#include <linux/poll.h>
31#include <asm/siginfo.h>
32#include <linux/uaccess.h>
33
34#define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
35
36static int setfl(int fd, struct file * filp, unsigned long arg)
37{
38	struct inode * inode = file_inode(filp);
39	int error = 0;
40
41	/*
42	 * O_APPEND cannot be cleared if the file is marked as append-only
43	 * and the file is open for write.
44	 */
45	if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
46		return -EPERM;
47
48	/* O_NOATIME can only be set by the owner or superuser */
49	if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
50		if (!inode_owner_or_capable(file_mnt_user_ns(filp), inode))
51			return -EPERM;
52
53	/* required for strict SunOS emulation */
54	if (O_NONBLOCK != O_NDELAY)
55	       if (arg & O_NDELAY)
56		   arg |= O_NONBLOCK;
57
58	/* Pipe packetized mode is controlled by O_DIRECT flag */
59	if (!S_ISFIFO(inode->i_mode) && (arg & O_DIRECT)) {
60		if (!filp->f_mapping || !filp->f_mapping->a_ops ||
61			!filp->f_mapping->a_ops->direct_IO)
62				return -EINVAL;
63	}
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	spin_unlock(&filp->f_lock);
83
84 out:
85	return error;
86}
87
88static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
89                     int force)
90{
91	write_lock_irq(&filp->f_owner.lock);
92	if (force || !filp->f_owner.pid) {
93		put_pid(filp->f_owner.pid);
94		filp->f_owner.pid = get_pid(pid);
95		filp->f_owner.pid_type = type;
96
97		if (pid) {
98			const struct cred *cred = current_cred();
99			filp->f_owner.uid = cred->uid;
100			filp->f_owner.euid = cred->euid;
101		}
102	}
103	write_unlock_irq(&filp->f_owner.lock);
104}
105
106void __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
107		int force)
108{
109	security_file_set_fowner(filp);
110	f_modown(filp, pid, type, force);
111}
112EXPORT_SYMBOL(__f_setown);
113
114int f_setown(struct file *filp, unsigned long arg, int force)
115{
116	enum pid_type type;
117	struct pid *pid = NULL;
118	int who = arg, ret = 0;
119
120	type = PIDTYPE_TGID;
121	if (who < 0) {
122		/* avoid overflow below */
123		if (who == INT_MIN)
124			return -EINVAL;
125
126		type = PIDTYPE_PGID;
127		who = -who;
128	}
129
130	rcu_read_lock();
131	if (who) {
132		pid = find_vpid(who);
133		if (!pid)
134			ret = -ESRCH;
135	}
136
137	if (!ret)
138		__f_setown(filp, pid, type, force);
139	rcu_read_unlock();
140
141	return ret;
142}
143EXPORT_SYMBOL(f_setown);
144
145void f_delown(struct file *filp)
146{
147	f_modown(filp, NULL, PIDTYPE_TGID, 1);
148}
149
150pid_t f_getown(struct file *filp)
151{
152	pid_t pid = 0;
153
154	read_lock_irq(&filp->f_owner.lock);
155	rcu_read_lock();
156	if (pid_task(filp->f_owner.pid, filp->f_owner.pid_type)) {
157		pid = pid_vnr(filp->f_owner.pid);
158		if (filp->f_owner.pid_type == PIDTYPE_PGID)
159			pid = -pid;
160	}
161	rcu_read_unlock();
162	read_unlock_irq(&filp->f_owner.lock);
163	return pid;
164}
165
166static int f_setown_ex(struct file *filp, unsigned long arg)
167{
168	struct f_owner_ex __user *owner_p = (void __user *)arg;
169	struct f_owner_ex owner;
170	struct pid *pid;
171	int type;
172	int ret;
173
174	ret = copy_from_user(&owner, owner_p, sizeof(owner));
175	if (ret)
176		return -EFAULT;
177
178	switch (owner.type) {
179	case F_OWNER_TID:
180		type = PIDTYPE_PID;
181		break;
182
183	case F_OWNER_PID:
184		type = PIDTYPE_TGID;
185		break;
186
187	case F_OWNER_PGRP:
188		type = PIDTYPE_PGID;
189		break;
190
191	default:
192		return -EINVAL;
193	}
194
195	rcu_read_lock();
196	pid = find_vpid(owner.pid);
197	if (owner.pid && !pid)
198		ret = -ESRCH;
199	else
200		 __f_setown(filp, pid, type, 1);
201	rcu_read_unlock();
202
203	return ret;
204}
205
206static int f_getown_ex(struct file *filp, unsigned long arg)
207{
208	struct f_owner_ex __user *owner_p = (void __user *)arg;
209	struct f_owner_ex owner = {};
210	int ret = 0;
211
212	read_lock_irq(&filp->f_owner.lock);
213	rcu_read_lock();
214	if (pid_task(filp->f_owner.pid, filp->f_owner.pid_type))
215		owner.pid = pid_vnr(filp->f_owner.pid);
216	rcu_read_unlock();
217	switch (filp->f_owner.pid_type) {
218	case PIDTYPE_PID:
219		owner.type = F_OWNER_TID;
220		break;
221
222	case PIDTYPE_TGID:
223		owner.type = F_OWNER_PID;
224		break;
225
226	case PIDTYPE_PGID:
227		owner.type = F_OWNER_PGRP;
228		break;
229
230	default:
231		WARN_ON(1);
232		ret = -EINVAL;
233		break;
234	}
235	read_unlock_irq(&filp->f_owner.lock);
236
237	if (!ret) {
238		ret = copy_to_user(owner_p, &owner, sizeof(owner));
239		if (ret)
240			ret = -EFAULT;
241	}
242	return ret;
243}
244
245#ifdef CONFIG_CHECKPOINT_RESTORE
246static int f_getowner_uids(struct file *filp, unsigned long arg)
247{
248	struct user_namespace *user_ns = current_user_ns();
249	uid_t __user *dst = (void __user *)arg;
250	uid_t src[2];
251	int err;
252
253	read_lock_irq(&filp->f_owner.lock);
254	src[0] = from_kuid(user_ns, filp->f_owner.uid);
255	src[1] = from_kuid(user_ns, filp->f_owner.euid);
256	read_unlock_irq(&filp->f_owner.lock);
257
258	err  = put_user(src[0], &dst[0]);
259	err |= put_user(src[1], &dst[1]);
260
261	return err;
262}
263#else
264static int f_getowner_uids(struct file *filp, unsigned long arg)
265{
266	return -EINVAL;
267}
268#endif
269
270static bool rw_hint_valid(enum rw_hint hint)
271{
272	switch (hint) {
273	case RWH_WRITE_LIFE_NOT_SET:
274	case RWH_WRITE_LIFE_NONE:
275	case RWH_WRITE_LIFE_SHORT:
276	case RWH_WRITE_LIFE_MEDIUM:
277	case RWH_WRITE_LIFE_LONG:
278	case RWH_WRITE_LIFE_EXTREME:
279		return true;
280	default:
281		return false;
282	}
283}
284
285static long fcntl_rw_hint(struct file *file, unsigned int cmd,
286			  unsigned long arg)
287{
288	struct inode *inode = file_inode(file);
289	u64 __user *argp = (u64 __user *)arg;
290	enum rw_hint hint;
291	u64 h;
292
293	switch (cmd) {
294	case F_GET_FILE_RW_HINT:
295		h = file_write_hint(file);
296		if (copy_to_user(argp, &h, sizeof(*argp)))
297			return -EFAULT;
298		return 0;
299	case F_SET_FILE_RW_HINT:
300		if (copy_from_user(&h, argp, sizeof(h)))
301			return -EFAULT;
302		hint = (enum rw_hint) h;
303		if (!rw_hint_valid(hint))
304			return -EINVAL;
305
306		spin_lock(&file->f_lock);
307		file->f_write_hint = hint;
308		spin_unlock(&file->f_lock);
309		return 0;
310	case F_GET_RW_HINT:
311		h = inode->i_write_hint;
312		if (copy_to_user(argp, &h, sizeof(*argp)))
313			return -EFAULT;
314		return 0;
315	case F_SET_RW_HINT:
316		if (copy_from_user(&h, argp, sizeof(h)))
317			return -EFAULT;
318		hint = (enum rw_hint) h;
319		if (!rw_hint_valid(hint))
320			return -EINVAL;
321
322		inode_lock(inode);
323		inode->i_write_hint = hint;
324		inode_unlock(inode);
325		return 0;
326	default:
327		return -EINVAL;
328	}
329}
330
331static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
332		struct file *filp)
333{
334	void __user *argp = (void __user *)arg;
335	struct flock flock;
336	long err = -EINVAL;
337
338	switch (cmd) {
339	case F_DUPFD:
340		err = f_dupfd(arg, filp, 0);
341		break;
342	case F_DUPFD_CLOEXEC:
343		err = f_dupfd(arg, 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, arg & 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);
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(&flock, argp, sizeof(flock)))
364			return -EFAULT;
365		err = fcntl_getlk(filp, cmd, &flock);
366		if (!err && copy_to_user(argp, &flock, 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(&flock, argp, 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, arg, 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(arg)) {
410			break;
411		}
412		err = 0;
413		filp->f_owner.signum = arg;
414		break;
415	case F_GETLEASE:
416		err = fcntl_getlease(filp);
417		break;
418	case F_SETLEASE:
419		err = fcntl_setlease(fd, filp, arg);
420		break;
421	case F_NOTIFY:
422		err = fcntl_dirnotify(fd, filp, arg);
423		break;
424	case F_SETPIPE_SZ:
425	case F_GETPIPE_SZ:
426		err = pipe_fcntl(filp, cmd, arg);
427		break;
428	case F_ADD_SEALS:
429	case F_GET_SEALS:
430		err = memfd_fcntl(filp, cmd, arg);
431		break;
432	case F_GET_RW_HINT:
433	case F_SET_RW_HINT:
434	case F_GET_FILE_RW_HINT:
435	case F_SET_FILE_RW_HINT:
436		err = fcntl_rw_hint(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(f.file, cmd, arg);
471	if (!err)
472		err = do_fcntl(fd, cmd, arg, f.file);
473
474out1:
475 	fdput(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(&fl, ufl, 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(&fl, ufl, 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(ufl, &fl, 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(ufl, &fl, 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(f.file, cmd, arg);
635	if (err)
636		goto out_put;
637
638	switch (cmd) {
639	case F_GETLK:
640		err = get_compat_flock(&flock, compat_ptr(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);
647		if (!err)
648			err = put_compat_flock(&flock, compat_ptr(arg));
649		break;
650	case F_GETLK64:
651	case F_OFD_GETLK:
652		err = get_compat_flock64(&flock, compat_ptr(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(&flock, compat_ptr(arg));
658		break;
659	case F_SETLK:
660	case F_SETLKW:
661		err = get_compat_flock(&flock, compat_ptr(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(&flock, compat_ptr(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, f.file);
677		break;
678	}
679out_put:
680	fdput(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(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, 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(&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(band_table[reason - POLL_IN]);
779			si.si_fd    = fd;
780			if (!do_send_sig_info(signum, &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, 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, 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 __read_mostly;
864
865static void fasync_free_rcu(struct rcu_head *head)
866{
867	kmem_cache_free(fasync_cache,
868			container_of(head, struct fasync_struct, fa_rcu));
869}
870
871/*
872 * Remove a fasync entry. If successfully removed, return
873 * positive and clear the FASYNC flag. If no entry exists,
874 * do nothing and return 0.
875 *
876 * NOTE! It is very important that the FASYNC flag always
877 * match the state "is the filp on a fasync list".
878 *
879 */
880int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
881{
882	struct fasync_struct *fa, **fp;
883	int result = 0;
884
885	spin_lock(&filp->f_lock);
886	spin_lock(&fasync_lock);
887	for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
888		if (fa->fa_file != filp)
889			continue;
890
891		write_lock_irq(&fa->fa_lock);
892		fa->fa_file = NULL;
893		write_unlock_irq(&fa->fa_lock);
894
895		*fp = fa->fa_next;
896		call_rcu(&fa->fa_rcu, fasync_free_rcu);
897		filp->f_flags &= ~FASYNC;
898		result = 1;
899		break;
900	}
901	spin_unlock(&fasync_lock);
902	spin_unlock(&filp->f_lock);
903	return result;
904}
905
906struct fasync_struct *fasync_alloc(void)
907{
908	return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
909}
910
911/*
912 * NOTE! This can be used only for unused fasync entries:
913 * entries that actually got inserted on the fasync list
914 * need to be released by rcu - see fasync_remove_entry.
915 */
916void fasync_free(struct fasync_struct *new)
917{
918	kmem_cache_free(fasync_cache, new);
919}
920
921/*
922 * Insert a new entry into the fasync list.  Return the pointer to the
923 * old one if we didn't use the new one.
924 *
925 * NOTE! It is very important that the FASYNC flag always
926 * match the state "is the filp on a fasync list".
927 */
928struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
929{
930        struct fasync_struct *fa, **fp;
931
932	spin_lock(&filp->f_lock);
933	spin_lock(&fasync_lock);
934	for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
935		if (fa->fa_file != filp)
936			continue;
937
938		write_lock_irq(&fa->fa_lock);
939		fa->fa_fd = fd;
940		write_unlock_irq(&fa->fa_lock);
941		goto out;
942	}
943
944	rwlock_init(&new->fa_lock);
945	new->magic = FASYNC_MAGIC;
946	new->fa_file = filp;
947	new->fa_fd = fd;
948	new->fa_next = *fapp;
949	rcu_assign_pointer(*fapp, new);
950	filp->f_flags |= FASYNC;
951
952out:
953	spin_unlock(&fasync_lock);
954	spin_unlock(&filp->f_lock);
955	return fa;
956}
957
958/*
959 * Add a fasync entry. Return negative on error, positive if
960 * added, and zero if did nothing but change an existing one.
961 */
962static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
963{
964	struct fasync_struct *new;
965
966	new = fasync_alloc();
967	if (!new)
968		return -ENOMEM;
969
970	/*
971	 * fasync_insert_entry() returns the old (update) entry if
972	 * it existed.
973	 *
974	 * So free the (unused) new entry and return 0 to let the
975	 * caller know that we didn't add any new fasync entries.
976	 */
977	if (fasync_insert_entry(fd, filp, fapp, new)) {
978		fasync_free(new);
979		return 0;
980	}
981
982	return 1;
983}
984
985/*
986 * fasync_helper() is used by almost all character device drivers
987 * to set up the fasync queue, and for regular files by the file
988 * lease code. It returns negative on error, 0 if it did no changes
989 * and positive if it added/deleted the entry.
990 */
991int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
992{
993	if (!on)
994		return fasync_remove_entry(filp, fapp);
995	return fasync_add_entry(fd, filp, fapp);
996}
997
998EXPORT_SYMBOL(fasync_helper);
999
1000/*
1001 * rcu_read_lock() is held
1002 */
1003static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
1004{
1005	while (fa) {
1006		struct fown_struct *fown;
1007		unsigned long flags;
1008
1009		if (fa->magic != FASYNC_MAGIC) {
1010			printk(KERN_ERR "kill_fasync: bad magic number in "
1011			       "fasync_struct!\n");
1012			return;
1013		}
1014		read_lock_irqsave(&fa->fa_lock, flags);
1015		if (fa->fa_file) {
1016			fown = &fa->fa_file->f_owner;
1017			/* Don't send SIGURG to processes which have not set a
1018			   queued signum: SIGURG has its own default signalling
1019			   mechanism. */
1020			if (!(sig == SIGURG && fown->signum == 0))
1021				send_sigio(fown, fa->fa_fd, band);
1022		}
1023		read_unlock_irqrestore(&fa->fa_lock, flags);
1024		fa = rcu_dereference(fa->fa_next);
1025	}
1026}
1027
1028void kill_fasync(struct fasync_struct **fp, int sig, int band)
1029{
1030	/* First a quick test without locking: usually
1031	 * the list is empty.
1032	 */
1033	if (*fp) {
1034		rcu_read_lock();
1035		kill_fasync_rcu(rcu_dereference(*fp), sig, band);
1036		rcu_read_unlock();
1037	}
1038}
1039EXPORT_SYMBOL(kill_fasync);
1040
1041static int __init fcntl_init(void)
1042{
1043	/*
1044	 * Please add new bits here to ensure allocation uniqueness.
1045	 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
1046	 * is defined as O_NONBLOCK on some platforms and not on others.
1047	 */
1048	BUILD_BUG_ON(21 - 1 /* for O_RDONLY being 0 */ !=
1049		HWEIGHT32(
1050			(VALID_OPEN_FLAGS & ~(O_NONBLOCK | O_NDELAY)) |
1051			__FMODE_EXEC | __FMODE_NONOTIFY));
1052
1053	fasync_cache = kmem_cache_create("fasync_cache",
1054		sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
1055	return 0;
1056}
1057
1058module_init(fcntl_init)
1059