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#include <linux/user_namespace.h>
24#include <linux/shmem_fs.h>
25
26#include <asm/poll.h>
27#include <asm/siginfo.h>
28#include <asm/uaccess.h>
29
30#define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
31
32static int setfl(int fd, struct file * filp, unsigned long arg)
33{
34	struct inode * inode = file_inode(filp);
35	int error = 0;
36
37	/*
38	 * O_APPEND cannot be cleared if the file is marked as append-only
39	 * and the file is open for write.
40	 */
41	if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
42		return -EPERM;
43
44	/* O_NOATIME can only be set by the owner or superuser */
45	if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
46		if (!inode_owner_or_capable(inode))
47			return -EPERM;
48
49	/* required for strict SunOS emulation */
50	if (O_NONBLOCK != O_NDELAY)
51	       if (arg & O_NDELAY)
52		   arg |= O_NONBLOCK;
53
54	if (arg & O_DIRECT) {
55		if (!filp->f_mapping || !filp->f_mapping->a_ops ||
56			!filp->f_mapping->a_ops->direct_IO)
57				return -EINVAL;
58	}
59
60	if (filp->f_op->check_flags)
61		error = filp->f_op->check_flags(arg);
62	if (error)
63		return error;
64
65	/*
66	 * ->fasync() is responsible for setting the FASYNC bit.
67	 */
68	if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op->fasync) {
69		error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
70		if (error < 0)
71			goto out;
72		if (error > 0)
73			error = 0;
74	}
75	spin_lock(&filp->f_lock);
76	filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
77	spin_unlock(&filp->f_lock);
78
79 out:
80	return error;
81}
82
83static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
84                     int force)
85{
86	write_lock_irq(&filp->f_owner.lock);
87	if (force || !filp->f_owner.pid) {
88		put_pid(filp->f_owner.pid);
89		filp->f_owner.pid = get_pid(pid);
90		filp->f_owner.pid_type = type;
91
92		if (pid) {
93			const struct cred *cred = current_cred();
94			filp->f_owner.uid = cred->uid;
95			filp->f_owner.euid = cred->euid;
96		}
97	}
98	write_unlock_irq(&filp->f_owner.lock);
99}
100
101void __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
102		int force)
103{
104	security_file_set_fowner(filp);
105	f_modown(filp, pid, type, force);
106}
107EXPORT_SYMBOL(__f_setown);
108
109void f_setown(struct file *filp, unsigned long arg, int force)
110{
111	enum pid_type type;
112	struct pid *pid;
113	int who = arg;
114	type = PIDTYPE_PID;
115	if (who < 0) {
116		type = PIDTYPE_PGID;
117		who = -who;
118	}
119	rcu_read_lock();
120	pid = find_vpid(who);
121	__f_setown(filp, pid, type, force);
122	rcu_read_unlock();
123}
124EXPORT_SYMBOL(f_setown);
125
126void f_delown(struct file *filp)
127{
128	f_modown(filp, NULL, PIDTYPE_PID, 1);
129}
130
131pid_t f_getown(struct file *filp)
132{
133	pid_t pid;
134	read_lock(&filp->f_owner.lock);
135	pid = pid_vnr(filp->f_owner.pid);
136	if (filp->f_owner.pid_type == PIDTYPE_PGID)
137		pid = -pid;
138	read_unlock(&filp->f_owner.lock);
139	return pid;
140}
141
142static int f_setown_ex(struct file *filp, unsigned long arg)
143{
144	struct f_owner_ex __user *owner_p = (void __user *)arg;
145	struct f_owner_ex owner;
146	struct pid *pid;
147	int type;
148	int ret;
149
150	ret = copy_from_user(&owner, owner_p, sizeof(owner));
151	if (ret)
152		return -EFAULT;
153
154	switch (owner.type) {
155	case F_OWNER_TID:
156		type = PIDTYPE_MAX;
157		break;
158
159	case F_OWNER_PID:
160		type = PIDTYPE_PID;
161		break;
162
163	case F_OWNER_PGRP:
164		type = PIDTYPE_PGID;
165		break;
166
167	default:
168		return -EINVAL;
169	}
170
171	rcu_read_lock();
172	pid = find_vpid(owner.pid);
173	if (owner.pid && !pid)
174		ret = -ESRCH;
175	else
176		 __f_setown(filp, pid, type, 1);
177	rcu_read_unlock();
178
179	return ret;
180}
181
182static int f_getown_ex(struct file *filp, unsigned long arg)
183{
184	struct f_owner_ex __user *owner_p = (void __user *)arg;
185	struct f_owner_ex owner;
186	int ret = 0;
187
188	read_lock(&filp->f_owner.lock);
189	owner.pid = pid_vnr(filp->f_owner.pid);
190	switch (filp->f_owner.pid_type) {
191	case PIDTYPE_MAX:
192		owner.type = F_OWNER_TID;
193		break;
194
195	case PIDTYPE_PID:
196		owner.type = F_OWNER_PID;
197		break;
198
199	case PIDTYPE_PGID:
200		owner.type = F_OWNER_PGRP;
201		break;
202
203	default:
204		WARN_ON(1);
205		ret = -EINVAL;
206		break;
207	}
208	read_unlock(&filp->f_owner.lock);
209
210	if (!ret) {
211		ret = copy_to_user(owner_p, &owner, sizeof(owner));
212		if (ret)
213			ret = -EFAULT;
214	}
215	return ret;
216}
217
218#ifdef CONFIG_CHECKPOINT_RESTORE
219static int f_getowner_uids(struct file *filp, unsigned long arg)
220{
221	struct user_namespace *user_ns = current_user_ns();
222	uid_t __user *dst = (void __user *)arg;
223	uid_t src[2];
224	int err;
225
226	read_lock(&filp->f_owner.lock);
227	src[0] = from_kuid(user_ns, filp->f_owner.uid);
228	src[1] = from_kuid(user_ns, filp->f_owner.euid);
229	read_unlock(&filp->f_owner.lock);
230
231	err  = put_user(src[0], &dst[0]);
232	err |= put_user(src[1], &dst[1]);
233
234	return err;
235}
236#else
237static int f_getowner_uids(struct file *filp, unsigned long arg)
238{
239	return -EINVAL;
240}
241#endif
242
243static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
244		struct file *filp)
245{
246	long err = -EINVAL;
247
248	switch (cmd) {
249	case F_DUPFD:
250		err = f_dupfd(arg, filp, 0);
251		break;
252	case F_DUPFD_CLOEXEC:
253		err = f_dupfd(arg, filp, O_CLOEXEC);
254		break;
255	case F_GETFD:
256		err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
257		break;
258	case F_SETFD:
259		err = 0;
260		set_close_on_exec(fd, arg & FD_CLOEXEC);
261		break;
262	case F_GETFL:
263		err = filp->f_flags;
264		break;
265	case F_SETFL:
266		err = setfl(fd, filp, arg);
267		break;
268#if BITS_PER_LONG != 32
269	/* 32-bit arches must use fcntl64() */
270	case F_OFD_GETLK:
271#endif
272	case F_GETLK:
273		err = fcntl_getlk(filp, cmd, (struct flock __user *) arg);
274		break;
275#if BITS_PER_LONG != 32
276	/* 32-bit arches must use fcntl64() */
277	case F_OFD_SETLK:
278	case F_OFD_SETLKW:
279#endif
280		/* Fallthrough */
281	case F_SETLK:
282	case F_SETLKW:
283		err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
284		break;
285	case F_GETOWN:
286		/*
287		 * XXX If f_owner is a process group, the
288		 * negative return value will get converted
289		 * into an error.  Oops.  If we keep the
290		 * current syscall conventions, the only way
291		 * to fix this will be in libc.
292		 */
293		err = f_getown(filp);
294		force_successful_syscall_return();
295		break;
296	case F_SETOWN:
297		f_setown(filp, arg, 1);
298		err = 0;
299		break;
300	case F_GETOWN_EX:
301		err = f_getown_ex(filp, arg);
302		break;
303	case F_SETOWN_EX:
304		err = f_setown_ex(filp, arg);
305		break;
306	case F_GETOWNER_UIDS:
307		err = f_getowner_uids(filp, arg);
308		break;
309	case F_GETSIG:
310		err = filp->f_owner.signum;
311		break;
312	case F_SETSIG:
313		/* arg == 0 restores default behaviour. */
314		if (!valid_signal(arg)) {
315			break;
316		}
317		err = 0;
318		filp->f_owner.signum = arg;
319		break;
320	case F_GETLEASE:
321		err = fcntl_getlease(filp);
322		break;
323	case F_SETLEASE:
324		err = fcntl_setlease(fd, filp, arg);
325		break;
326	case F_NOTIFY:
327		err = fcntl_dirnotify(fd, filp, arg);
328		break;
329	case F_SETPIPE_SZ:
330	case F_GETPIPE_SZ:
331		err = pipe_fcntl(filp, cmd, arg);
332		break;
333	case F_ADD_SEALS:
334	case F_GET_SEALS:
335		err = shmem_fcntl(filp, cmd, arg);
336		break;
337	default:
338		break;
339	}
340	return err;
341}
342
343static int check_fcntl_cmd(unsigned cmd)
344{
345	switch (cmd) {
346	case F_DUPFD:
347	case F_DUPFD_CLOEXEC:
348	case F_GETFD:
349	case F_SETFD:
350	case F_GETFL:
351		return 1;
352	}
353	return 0;
354}
355
356SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
357{
358	struct fd f = fdget_raw(fd);
359	long err = -EBADF;
360
361	if (!f.file)
362		goto out;
363
364	if (unlikely(f.file->f_mode & FMODE_PATH)) {
365		if (!check_fcntl_cmd(cmd))
366			goto out1;
367	}
368
369	err = security_file_fcntl(f.file, cmd, arg);
370	if (!err)
371		err = do_fcntl(fd, cmd, arg, f.file);
372
373out1:
374 	fdput(f);
375out:
376	return err;
377}
378
379#if BITS_PER_LONG == 32
380SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
381		unsigned long, arg)
382{
383	struct fd f = fdget_raw(fd);
384	long err = -EBADF;
385
386	if (!f.file)
387		goto out;
388
389	if (unlikely(f.file->f_mode & FMODE_PATH)) {
390		if (!check_fcntl_cmd(cmd))
391			goto out1;
392	}
393
394	err = security_file_fcntl(f.file, cmd, arg);
395	if (err)
396		goto out1;
397
398	switch (cmd) {
399	case F_GETLK64:
400	case F_OFD_GETLK:
401		err = fcntl_getlk64(f.file, cmd, (struct flock64 __user *) arg);
402		break;
403	case F_SETLK64:
404	case F_SETLKW64:
405	case F_OFD_SETLK:
406	case F_OFD_SETLKW:
407		err = fcntl_setlk64(fd, f.file, cmd,
408				(struct flock64 __user *) arg);
409		break;
410	default:
411		err = do_fcntl(fd, cmd, arg, f.file);
412		break;
413	}
414out1:
415	fdput(f);
416out:
417	return err;
418}
419#endif
420
421/* Table to convert sigio signal codes into poll band bitmaps */
422
423static const long band_table[NSIGPOLL] = {
424	POLLIN | POLLRDNORM,			/* POLL_IN */
425	POLLOUT | POLLWRNORM | POLLWRBAND,	/* POLL_OUT */
426	POLLIN | POLLRDNORM | POLLMSG,		/* POLL_MSG */
427	POLLERR,				/* POLL_ERR */
428	POLLPRI | POLLRDBAND,			/* POLL_PRI */
429	POLLHUP | POLLERR			/* POLL_HUP */
430};
431
432static inline int sigio_perm(struct task_struct *p,
433                             struct fown_struct *fown, int sig)
434{
435	const struct cred *cred;
436	int ret;
437
438	rcu_read_lock();
439	cred = __task_cred(p);
440	ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
441		uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
442		uid_eq(fown->uid,  cred->suid) || uid_eq(fown->uid,  cred->uid)) &&
443	       !security_file_send_sigiotask(p, fown, sig));
444	rcu_read_unlock();
445	return ret;
446}
447
448static void send_sigio_to_task(struct task_struct *p,
449			       struct fown_struct *fown,
450			       int fd, int reason, int group)
451{
452	/*
453	 * F_SETSIG can change ->signum lockless in parallel, make
454	 * sure we read it once and use the same value throughout.
455	 */
456	int signum = ACCESS_ONCE(fown->signum);
457
458	if (!sigio_perm(p, fown, signum))
459		return;
460
461	switch (signum) {
462		siginfo_t si;
463		default:
464			/* Queue a rt signal with the appropriate fd as its
465			   value.  We use SI_SIGIO as the source, not
466			   SI_KERNEL, since kernel signals always get
467			   delivered even if we can't queue.  Failure to
468			   queue in this case _should_ be reported; we fall
469			   back to SIGIO in that case. --sct */
470			si.si_signo = signum;
471			si.si_errno = 0;
472		        si.si_code  = reason;
473			/* Make sure we are called with one of the POLL_*
474			   reasons, otherwise we could leak kernel stack into
475			   userspace.  */
476			BUG_ON((reason & __SI_MASK) != __SI_POLL);
477			if (reason - POLL_IN >= NSIGPOLL)
478				si.si_band  = ~0L;
479			else
480				si.si_band = band_table[reason - POLL_IN];
481			si.si_fd    = fd;
482			if (!do_send_sig_info(signum, &si, p, group))
483				break;
484		/* fall-through: fall back on the old plain SIGIO signal */
485		case 0:
486			do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
487	}
488}
489
490void send_sigio(struct fown_struct *fown, int fd, int band)
491{
492	struct task_struct *p;
493	enum pid_type type;
494	struct pid *pid;
495	int group = 1;
496
497	read_lock(&fown->lock);
498
499	type = fown->pid_type;
500	if (type == PIDTYPE_MAX) {
501		group = 0;
502		type = PIDTYPE_PID;
503	}
504
505	pid = fown->pid;
506	if (!pid)
507		goto out_unlock_fown;
508
509	read_lock(&tasklist_lock);
510	do_each_pid_task(pid, type, p) {
511		send_sigio_to_task(p, fown, fd, band, group);
512	} while_each_pid_task(pid, type, p);
513	read_unlock(&tasklist_lock);
514 out_unlock_fown:
515	read_unlock(&fown->lock);
516}
517
518static void send_sigurg_to_task(struct task_struct *p,
519				struct fown_struct *fown, int group)
520{
521	if (sigio_perm(p, fown, SIGURG))
522		do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
523}
524
525int send_sigurg(struct fown_struct *fown)
526{
527	struct task_struct *p;
528	enum pid_type type;
529	struct pid *pid;
530	int group = 1;
531	int ret = 0;
532
533	read_lock(&fown->lock);
534
535	type = fown->pid_type;
536	if (type == PIDTYPE_MAX) {
537		group = 0;
538		type = PIDTYPE_PID;
539	}
540
541	pid = fown->pid;
542	if (!pid)
543		goto out_unlock_fown;
544
545	ret = 1;
546
547	read_lock(&tasklist_lock);
548	do_each_pid_task(pid, type, p) {
549		send_sigurg_to_task(p, fown, group);
550	} while_each_pid_task(pid, type, p);
551	read_unlock(&tasklist_lock);
552 out_unlock_fown:
553	read_unlock(&fown->lock);
554	return ret;
555}
556
557static DEFINE_SPINLOCK(fasync_lock);
558static struct kmem_cache *fasync_cache __read_mostly;
559
560static void fasync_free_rcu(struct rcu_head *head)
561{
562	kmem_cache_free(fasync_cache,
563			container_of(head, struct fasync_struct, fa_rcu));
564}
565
566/*
567 * Remove a fasync entry. If successfully removed, return
568 * positive and clear the FASYNC flag. If no entry exists,
569 * do nothing and return 0.
570 *
571 * NOTE! It is very important that the FASYNC flag always
572 * match the state "is the filp on a fasync list".
573 *
574 */
575int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
576{
577	struct fasync_struct *fa, **fp;
578	int result = 0;
579
580	spin_lock(&filp->f_lock);
581	spin_lock(&fasync_lock);
582	for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
583		if (fa->fa_file != filp)
584			continue;
585
586		spin_lock_irq(&fa->fa_lock);
587		fa->fa_file = NULL;
588		spin_unlock_irq(&fa->fa_lock);
589
590		*fp = fa->fa_next;
591		call_rcu(&fa->fa_rcu, fasync_free_rcu);
592		filp->f_flags &= ~FASYNC;
593		result = 1;
594		break;
595	}
596	spin_unlock(&fasync_lock);
597	spin_unlock(&filp->f_lock);
598	return result;
599}
600
601struct fasync_struct *fasync_alloc(void)
602{
603	return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
604}
605
606/*
607 * NOTE! This can be used only for unused fasync entries:
608 * entries that actually got inserted on the fasync list
609 * need to be released by rcu - see fasync_remove_entry.
610 */
611void fasync_free(struct fasync_struct *new)
612{
613	kmem_cache_free(fasync_cache, new);
614}
615
616/*
617 * Insert a new entry into the fasync list.  Return the pointer to the
618 * old one if we didn't use the new one.
619 *
620 * NOTE! It is very important that the FASYNC flag always
621 * match the state "is the filp on a fasync list".
622 */
623struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
624{
625        struct fasync_struct *fa, **fp;
626
627	spin_lock(&filp->f_lock);
628	spin_lock(&fasync_lock);
629	for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
630		if (fa->fa_file != filp)
631			continue;
632
633		spin_lock_irq(&fa->fa_lock);
634		fa->fa_fd = fd;
635		spin_unlock_irq(&fa->fa_lock);
636		goto out;
637	}
638
639	spin_lock_init(&new->fa_lock);
640	new->magic = FASYNC_MAGIC;
641	new->fa_file = filp;
642	new->fa_fd = fd;
643	new->fa_next = *fapp;
644	rcu_assign_pointer(*fapp, new);
645	filp->f_flags |= FASYNC;
646
647out:
648	spin_unlock(&fasync_lock);
649	spin_unlock(&filp->f_lock);
650	return fa;
651}
652
653/*
654 * Add a fasync entry. Return negative on error, positive if
655 * added, and zero if did nothing but change an existing one.
656 */
657static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
658{
659	struct fasync_struct *new;
660
661	new = fasync_alloc();
662	if (!new)
663		return -ENOMEM;
664
665	/*
666	 * fasync_insert_entry() returns the old (update) entry if
667	 * it existed.
668	 *
669	 * So free the (unused) new entry and return 0 to let the
670	 * caller know that we didn't add any new fasync entries.
671	 */
672	if (fasync_insert_entry(fd, filp, fapp, new)) {
673		fasync_free(new);
674		return 0;
675	}
676
677	return 1;
678}
679
680/*
681 * fasync_helper() is used by almost all character device drivers
682 * to set up the fasync queue, and for regular files by the file
683 * lease code. It returns negative on error, 0 if it did no changes
684 * and positive if it added/deleted the entry.
685 */
686int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
687{
688	if (!on)
689		return fasync_remove_entry(filp, fapp);
690	return fasync_add_entry(fd, filp, fapp);
691}
692
693EXPORT_SYMBOL(fasync_helper);
694
695/*
696 * rcu_read_lock() is held
697 */
698static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
699{
700	while (fa) {
701		struct fown_struct *fown;
702		unsigned long flags;
703
704		if (fa->magic != FASYNC_MAGIC) {
705			printk(KERN_ERR "kill_fasync: bad magic number in "
706			       "fasync_struct!\n");
707			return;
708		}
709		spin_lock_irqsave(&fa->fa_lock, flags);
710		if (fa->fa_file) {
711			fown = &fa->fa_file->f_owner;
712			/* Don't send SIGURG to processes which have not set a
713			   queued signum: SIGURG has its own default signalling
714			   mechanism. */
715			if (!(sig == SIGURG && fown->signum == 0))
716				send_sigio(fown, fa->fa_fd, band);
717		}
718		spin_unlock_irqrestore(&fa->fa_lock, flags);
719		fa = rcu_dereference(fa->fa_next);
720	}
721}
722
723void kill_fasync(struct fasync_struct **fp, int sig, int band)
724{
725	/* First a quick test without locking: usually
726	 * the list is empty.
727	 */
728	if (*fp) {
729		rcu_read_lock();
730		kill_fasync_rcu(rcu_dereference(*fp), sig, band);
731		rcu_read_unlock();
732	}
733}
734EXPORT_SYMBOL(kill_fasync);
735
736static int __init fcntl_init(void)
737{
738	/*
739	 * Please add new bits here to ensure allocation uniqueness.
740	 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
741	 * is defined as O_NONBLOCK on some platforms and not on others.
742	 */
743	BUILD_BUG_ON(21 - 1 /* for O_RDONLY being 0 */ != HWEIGHT32(
744		O_RDONLY	| O_WRONLY	| O_RDWR	|
745		O_CREAT		| O_EXCL	| O_NOCTTY	|
746		O_TRUNC		| O_APPEND	| /* O_NONBLOCK	| */
747		__O_SYNC	| O_DSYNC	| FASYNC	|
748		O_DIRECT	| O_LARGEFILE	| O_DIRECTORY	|
749		O_NOFOLLOW	| O_NOATIME	| O_CLOEXEC	|
750		__FMODE_EXEC	| O_PATH	| __O_TMPFILE	|
751		__FMODE_NONOTIFY
752		));
753
754	fasync_cache = kmem_cache_create("fasync_cache",
755		sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
756	return 0;
757}
758
759module_init(fcntl_init)
760