1 /*
2  *  linux/fs/fcntl.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6 
7 #include <linux/syscalls.h>
8 #include <linux/init.h>
9 #include <linux/mm.h>
10 #include <linux/fs.h>
11 #include <linux/file.h>
12 #include <linux/fdtable.h>
13 #include <linux/capability.h>
14 #include <linux/dnotify.h>
15 #include <linux/slab.h>
16 #include <linux/module.h>
17 #include <linux/pipe_fs_i.h>
18 #include <linux/security.h>
19 #include <linux/ptrace.h>
20 #include <linux/signal.h>
21 #include <linux/rcupdate.h>
22 #include <linux/pid_namespace.h>
23 
24 #include <asm/poll.h>
25 #include <asm/siginfo.h>
26 #include <asm/uaccess.h>
27 
set_close_on_exec(unsigned int fd,int flag)28 void set_close_on_exec(unsigned int fd, int flag)
29 {
30 	struct files_struct *files = current->files;
31 	struct fdtable *fdt;
32 	spin_lock(&files->file_lock);
33 	fdt = files_fdtable(files);
34 	if (flag)
35 		__set_close_on_exec(fd, fdt);
36 	else
37 		__clear_close_on_exec(fd, fdt);
38 	spin_unlock(&files->file_lock);
39 }
40 
get_close_on_exec(unsigned int fd)41 static bool get_close_on_exec(unsigned int fd)
42 {
43 	struct files_struct *files = current->files;
44 	struct fdtable *fdt;
45 	bool res;
46 	rcu_read_lock();
47 	fdt = files_fdtable(files);
48 	res = close_on_exec(fd, fdt);
49 	rcu_read_unlock();
50 	return res;
51 }
52 
SYSCALL_DEFINE3(dup3,unsigned int,oldfd,unsigned int,newfd,int,flags)53 SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
54 {
55 	int err = -EBADF;
56 	struct file * file, *tofree;
57 	struct files_struct * files = current->files;
58 	struct fdtable *fdt;
59 
60 	if ((flags & ~O_CLOEXEC) != 0)
61 		return -EINVAL;
62 
63 	if (unlikely(oldfd == newfd))
64 		return -EINVAL;
65 
66 	spin_lock(&files->file_lock);
67 	err = expand_files(files, newfd);
68 	file = fcheck(oldfd);
69 	if (unlikely(!file))
70 		goto Ebadf;
71 	if (unlikely(err < 0)) {
72 		if (err == -EMFILE)
73 			goto Ebadf;
74 		goto out_unlock;
75 	}
76 	/*
77 	 * We need to detect attempts to do dup2() over allocated but still
78 	 * not finished descriptor.  NB: OpenBSD avoids that at the price of
79 	 * extra work in their equivalent of fget() - they insert struct
80 	 * file immediately after grabbing descriptor, mark it larval if
81 	 * more work (e.g. actual opening) is needed and make sure that
82 	 * fget() treats larval files as absent.  Potentially interesting,
83 	 * but while extra work in fget() is trivial, locking implications
84 	 * and amount of surgery on open()-related paths in VFS are not.
85 	 * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
86 	 * deadlocks in rather amusing ways, AFAICS.  All of that is out of
87 	 * scope of POSIX or SUS, since neither considers shared descriptor
88 	 * tables and this condition does not arise without those.
89 	 */
90 	err = -EBUSY;
91 	fdt = files_fdtable(files);
92 	tofree = fdt->fd[newfd];
93 	if (!tofree && fd_is_open(newfd, fdt))
94 		goto out_unlock;
95 	get_file(file);
96 	rcu_assign_pointer(fdt->fd[newfd], file);
97 	__set_open_fd(newfd, fdt);
98 	if (flags & O_CLOEXEC)
99 		__set_close_on_exec(newfd, fdt);
100 	else
101 		__clear_close_on_exec(newfd, fdt);
102 	spin_unlock(&files->file_lock);
103 
104 	if (tofree)
105 		filp_close(tofree, files);
106 
107 	return newfd;
108 
109 Ebadf:
110 	err = -EBADF;
111 out_unlock:
112 	spin_unlock(&files->file_lock);
113 	return err;
114 }
115 
SYSCALL_DEFINE2(dup2,unsigned int,oldfd,unsigned int,newfd)116 SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
117 {
118 	if (unlikely(newfd == oldfd)) { /* corner case */
119 		struct files_struct *files = current->files;
120 		int retval = oldfd;
121 
122 		rcu_read_lock();
123 		if (!fcheck_files(files, oldfd))
124 			retval = -EBADF;
125 		rcu_read_unlock();
126 		return retval;
127 	}
128 	return sys_dup3(oldfd, newfd, 0);
129 }
130 
SYSCALL_DEFINE1(dup,unsigned int,fildes)131 SYSCALL_DEFINE1(dup, unsigned int, fildes)
132 {
133 	int ret = -EBADF;
134 	struct file *file = fget_raw(fildes);
135 
136 	if (file) {
137 		ret = get_unused_fd();
138 		if (ret >= 0)
139 			fd_install(ret, file);
140 		else
141 			fput(file);
142 	}
143 	return ret;
144 }
145 
146 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
147 
setfl(int fd,struct file * filp,unsigned long arg)148 static int setfl(int fd, struct file * filp, unsigned long arg)
149 {
150 	struct inode * inode = filp->f_path.dentry->d_inode;
151 	int error = 0;
152 
153 	/*
154 	 * O_APPEND cannot be cleared if the file is marked as append-only
155 	 * and the file is open for write.
156 	 */
157 	if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
158 		return -EPERM;
159 
160 	/* O_NOATIME can only be set by the owner or superuser */
161 	if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
162 		if (!inode_owner_or_capable(inode))
163 			return -EPERM;
164 
165 	/* required for strict SunOS emulation */
166 	if (O_NONBLOCK != O_NDELAY)
167 	       if (arg & O_NDELAY)
168 		   arg |= O_NONBLOCK;
169 
170 	if (arg & O_DIRECT) {
171 		if (!filp->f_mapping || !filp->f_mapping->a_ops ||
172 			!filp->f_mapping->a_ops->direct_IO)
173 				return -EINVAL;
174 	}
175 
176 	if (filp->f_op && filp->f_op->check_flags)
177 		error = filp->f_op->check_flags(arg);
178 	if (error)
179 		return error;
180 
181 	/*
182 	 * ->fasync() is responsible for setting the FASYNC bit.
183 	 */
184 	if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op &&
185 			filp->f_op->fasync) {
186 		error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
187 		if (error < 0)
188 			goto out;
189 		if (error > 0)
190 			error = 0;
191 	}
192 	spin_lock(&filp->f_lock);
193 	filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
194 	spin_unlock(&filp->f_lock);
195 
196  out:
197 	return error;
198 }
199 
f_modown(struct file * filp,struct pid * pid,enum pid_type type,int force)200 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
201                      int force)
202 {
203 	write_lock_irq(&filp->f_owner.lock);
204 	if (force || !filp->f_owner.pid) {
205 		put_pid(filp->f_owner.pid);
206 		filp->f_owner.pid = get_pid(pid);
207 		filp->f_owner.pid_type = type;
208 
209 		if (pid) {
210 			const struct cred *cred = current_cred();
211 			filp->f_owner.uid = cred->uid;
212 			filp->f_owner.euid = cred->euid;
213 		}
214 	}
215 	write_unlock_irq(&filp->f_owner.lock);
216 }
217 
__f_setown(struct file * filp,struct pid * pid,enum pid_type type,int force)218 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
219 		int force)
220 {
221 	int err;
222 
223 	err = security_file_set_fowner(filp);
224 	if (err)
225 		return err;
226 
227 	f_modown(filp, pid, type, force);
228 	return 0;
229 }
230 EXPORT_SYMBOL(__f_setown);
231 
f_setown(struct file * filp,unsigned long arg,int force)232 int f_setown(struct file *filp, unsigned long arg, int force)
233 {
234 	enum pid_type type;
235 	struct pid *pid;
236 	int who = arg;
237 	int result;
238 	type = PIDTYPE_PID;
239 	if (who < 0) {
240 		type = PIDTYPE_PGID;
241 		who = -who;
242 	}
243 	rcu_read_lock();
244 	pid = find_vpid(who);
245 	result = __f_setown(filp, pid, type, force);
246 	rcu_read_unlock();
247 	return result;
248 }
249 EXPORT_SYMBOL(f_setown);
250 
f_delown(struct file * filp)251 void f_delown(struct file *filp)
252 {
253 	f_modown(filp, NULL, PIDTYPE_PID, 1);
254 }
255 
f_getown(struct file * filp)256 pid_t f_getown(struct file *filp)
257 {
258 	pid_t pid;
259 	read_lock(&filp->f_owner.lock);
260 	pid = pid_vnr(filp->f_owner.pid);
261 	if (filp->f_owner.pid_type == PIDTYPE_PGID)
262 		pid = -pid;
263 	read_unlock(&filp->f_owner.lock);
264 	return pid;
265 }
266 
f_setown_ex(struct file * filp,unsigned long arg)267 static int f_setown_ex(struct file *filp, unsigned long arg)
268 {
269 	struct f_owner_ex * __user owner_p = (void * __user)arg;
270 	struct f_owner_ex owner;
271 	struct pid *pid;
272 	int type;
273 	int ret;
274 
275 	ret = copy_from_user(&owner, owner_p, sizeof(owner));
276 	if (ret)
277 		return -EFAULT;
278 
279 	switch (owner.type) {
280 	case F_OWNER_TID:
281 		type = PIDTYPE_MAX;
282 		break;
283 
284 	case F_OWNER_PID:
285 		type = PIDTYPE_PID;
286 		break;
287 
288 	case F_OWNER_PGRP:
289 		type = PIDTYPE_PGID;
290 		break;
291 
292 	default:
293 		return -EINVAL;
294 	}
295 
296 	rcu_read_lock();
297 	pid = find_vpid(owner.pid);
298 	if (owner.pid && !pid)
299 		ret = -ESRCH;
300 	else
301 		ret = __f_setown(filp, pid, type, 1);
302 	rcu_read_unlock();
303 
304 	return ret;
305 }
306 
f_getown_ex(struct file * filp,unsigned long arg)307 static int f_getown_ex(struct file *filp, unsigned long arg)
308 {
309 	struct f_owner_ex * __user owner_p = (void * __user)arg;
310 	struct f_owner_ex owner;
311 	int ret = 0;
312 
313 	read_lock(&filp->f_owner.lock);
314 	owner.pid = pid_vnr(filp->f_owner.pid);
315 	switch (filp->f_owner.pid_type) {
316 	case PIDTYPE_MAX:
317 		owner.type = F_OWNER_TID;
318 		break;
319 
320 	case PIDTYPE_PID:
321 		owner.type = F_OWNER_PID;
322 		break;
323 
324 	case PIDTYPE_PGID:
325 		owner.type = F_OWNER_PGRP;
326 		break;
327 
328 	default:
329 		WARN_ON(1);
330 		ret = -EINVAL;
331 		break;
332 	}
333 	read_unlock(&filp->f_owner.lock);
334 
335 	if (!ret) {
336 		ret = copy_to_user(owner_p, &owner, sizeof(owner));
337 		if (ret)
338 			ret = -EFAULT;
339 	}
340 	return ret;
341 }
342 
do_fcntl(int fd,unsigned int cmd,unsigned long arg,struct file * filp)343 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
344 		struct file *filp)
345 {
346 	long err = -EINVAL;
347 
348 	switch (cmd) {
349 	case F_DUPFD:
350 	case F_DUPFD_CLOEXEC:
351 		if (arg >= rlimit(RLIMIT_NOFILE))
352 			break;
353 		err = alloc_fd(arg, cmd == F_DUPFD_CLOEXEC ? O_CLOEXEC : 0);
354 		if (err >= 0) {
355 			get_file(filp);
356 			fd_install(err, filp);
357 		}
358 		break;
359 	case F_GETFD:
360 		err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
361 		break;
362 	case F_SETFD:
363 		err = 0;
364 		set_close_on_exec(fd, arg & FD_CLOEXEC);
365 		break;
366 	case F_GETFL:
367 		err = filp->f_flags;
368 		break;
369 	case F_SETFL:
370 		err = setfl(fd, filp, arg);
371 		break;
372 	case F_GETLK:
373 		err = fcntl_getlk(filp, (struct flock __user *) arg);
374 		break;
375 	case F_SETLK:
376 	case F_SETLKW:
377 		err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
378 		break;
379 	case F_GETOWN:
380 		/*
381 		 * XXX If f_owner is a process group, the
382 		 * negative return value will get converted
383 		 * into an error.  Oops.  If we keep the
384 		 * current syscall conventions, the only way
385 		 * to fix this will be in libc.
386 		 */
387 		err = f_getown(filp);
388 		force_successful_syscall_return();
389 		break;
390 	case F_SETOWN:
391 		err = f_setown(filp, arg, 1);
392 		break;
393 	case F_GETOWN_EX:
394 		err = f_getown_ex(filp, arg);
395 		break;
396 	case F_SETOWN_EX:
397 		err = f_setown_ex(filp, arg);
398 		break;
399 	case F_GETSIG:
400 		err = filp->f_owner.signum;
401 		break;
402 	case F_SETSIG:
403 		/* arg == 0 restores default behaviour. */
404 		if (!valid_signal(arg)) {
405 			break;
406 		}
407 		err = 0;
408 		filp->f_owner.signum = arg;
409 		break;
410 	case F_GETLEASE:
411 		err = fcntl_getlease(filp);
412 		break;
413 	case F_SETLEASE:
414 		err = fcntl_setlease(fd, filp, arg);
415 		break;
416 	case F_NOTIFY:
417 		err = fcntl_dirnotify(fd, filp, arg);
418 		break;
419 	case F_SETPIPE_SZ:
420 	case F_GETPIPE_SZ:
421 		err = pipe_fcntl(filp, cmd, arg);
422 		break;
423 	default:
424 		break;
425 	}
426 	return err;
427 }
428 
check_fcntl_cmd(unsigned cmd)429 static int check_fcntl_cmd(unsigned cmd)
430 {
431 	switch (cmd) {
432 	case F_DUPFD:
433 	case F_DUPFD_CLOEXEC:
434 	case F_GETFD:
435 	case F_SETFD:
436 	case F_GETFL:
437 		return 1;
438 	}
439 	return 0;
440 }
441 
SYSCALL_DEFINE3(fcntl,unsigned int,fd,unsigned int,cmd,unsigned long,arg)442 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
443 {
444 	struct file *filp;
445 	long err = -EBADF;
446 
447 	filp = fget_raw(fd);
448 	if (!filp)
449 		goto out;
450 
451 	if (unlikely(filp->f_mode & FMODE_PATH)) {
452 		if (!check_fcntl_cmd(cmd)) {
453 			fput(filp);
454 			goto out;
455 		}
456 	}
457 
458 	err = security_file_fcntl(filp, cmd, arg);
459 	if (err) {
460 		fput(filp);
461 		return err;
462 	}
463 
464 	err = do_fcntl(fd, cmd, arg, filp);
465 
466  	fput(filp);
467 out:
468 	return err;
469 }
470 
471 #if BITS_PER_LONG == 32
SYSCALL_DEFINE3(fcntl64,unsigned int,fd,unsigned int,cmd,unsigned long,arg)472 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
473 		unsigned long, arg)
474 {
475 	struct file * filp;
476 	long err;
477 
478 	err = -EBADF;
479 	filp = fget_raw(fd);
480 	if (!filp)
481 		goto out;
482 
483 	if (unlikely(filp->f_mode & FMODE_PATH)) {
484 		if (!check_fcntl_cmd(cmd)) {
485 			fput(filp);
486 			goto out;
487 		}
488 	}
489 
490 	err = security_file_fcntl(filp, cmd, arg);
491 	if (err) {
492 		fput(filp);
493 		return err;
494 	}
495 	err = -EBADF;
496 
497 	switch (cmd) {
498 		case F_GETLK64:
499 			err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
500 			break;
501 		case F_SETLK64:
502 		case F_SETLKW64:
503 			err = fcntl_setlk64(fd, filp, cmd,
504 					(struct flock64 __user *) arg);
505 			break;
506 		default:
507 			err = do_fcntl(fd, cmd, arg, filp);
508 			break;
509 	}
510 	fput(filp);
511 out:
512 	return err;
513 }
514 #endif
515 
516 /* Table to convert sigio signal codes into poll band bitmaps */
517 
518 static const long band_table[NSIGPOLL] = {
519 	POLLIN | POLLRDNORM,			/* POLL_IN */
520 	POLLOUT | POLLWRNORM | POLLWRBAND,	/* POLL_OUT */
521 	POLLIN | POLLRDNORM | POLLMSG,		/* POLL_MSG */
522 	POLLERR,				/* POLL_ERR */
523 	POLLPRI | POLLRDBAND,			/* POLL_PRI */
524 	POLLHUP | POLLERR			/* POLL_HUP */
525 };
526 
sigio_perm(struct task_struct * p,struct fown_struct * fown,int sig)527 static inline int sigio_perm(struct task_struct *p,
528                              struct fown_struct *fown, int sig)
529 {
530 	const struct cred *cred;
531 	int ret;
532 
533 	rcu_read_lock();
534 	cred = __task_cred(p);
535 	ret = ((fown->euid == 0 ||
536 		fown->euid == cred->suid || fown->euid == cred->uid ||
537 		fown->uid  == cred->suid || fown->uid  == cred->uid) &&
538 	       !security_file_send_sigiotask(p, fown, sig));
539 	rcu_read_unlock();
540 	return ret;
541 }
542 
send_sigio_to_task(struct task_struct * p,struct fown_struct * fown,int fd,int reason,int group)543 static void send_sigio_to_task(struct task_struct *p,
544 			       struct fown_struct *fown,
545 			       int fd, int reason, int group)
546 {
547 	/*
548 	 * F_SETSIG can change ->signum lockless in parallel, make
549 	 * sure we read it once and use the same value throughout.
550 	 */
551 	int signum = ACCESS_ONCE(fown->signum);
552 
553 	if (!sigio_perm(p, fown, signum))
554 		return;
555 
556 	switch (signum) {
557 		siginfo_t si;
558 		default:
559 			/* Queue a rt signal with the appropriate fd as its
560 			   value.  We use SI_SIGIO as the source, not
561 			   SI_KERNEL, since kernel signals always get
562 			   delivered even if we can't queue.  Failure to
563 			   queue in this case _should_ be reported; we fall
564 			   back to SIGIO in that case. --sct */
565 			si.si_signo = signum;
566 			si.si_errno = 0;
567 		        si.si_code  = reason;
568 			/* Make sure we are called with one of the POLL_*
569 			   reasons, otherwise we could leak kernel stack into
570 			   userspace.  */
571 			BUG_ON((reason & __SI_MASK) != __SI_POLL);
572 			if (reason - POLL_IN >= NSIGPOLL)
573 				si.si_band  = ~0L;
574 			else
575 				si.si_band = band_table[reason - POLL_IN];
576 			si.si_fd    = fd;
577 			if (!do_send_sig_info(signum, &si, p, group))
578 				break;
579 		/* fall-through: fall back on the old plain SIGIO signal */
580 		case 0:
581 			do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
582 	}
583 }
584 
send_sigio(struct fown_struct * fown,int fd,int band)585 void send_sigio(struct fown_struct *fown, int fd, int band)
586 {
587 	struct task_struct *p;
588 	enum pid_type type;
589 	struct pid *pid;
590 	int group = 1;
591 
592 	read_lock(&fown->lock);
593 
594 	type = fown->pid_type;
595 	if (type == PIDTYPE_MAX) {
596 		group = 0;
597 		type = PIDTYPE_PID;
598 	}
599 
600 	pid = fown->pid;
601 	if (!pid)
602 		goto out_unlock_fown;
603 
604 	read_lock(&tasklist_lock);
605 	do_each_pid_task(pid, type, p) {
606 		send_sigio_to_task(p, fown, fd, band, group);
607 	} while_each_pid_task(pid, type, p);
608 	read_unlock(&tasklist_lock);
609  out_unlock_fown:
610 	read_unlock(&fown->lock);
611 }
612 
send_sigurg_to_task(struct task_struct * p,struct fown_struct * fown,int group)613 static void send_sigurg_to_task(struct task_struct *p,
614 				struct fown_struct *fown, int group)
615 {
616 	if (sigio_perm(p, fown, SIGURG))
617 		do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
618 }
619 
send_sigurg(struct fown_struct * fown)620 int send_sigurg(struct fown_struct *fown)
621 {
622 	struct task_struct *p;
623 	enum pid_type type;
624 	struct pid *pid;
625 	int group = 1;
626 	int ret = 0;
627 
628 	read_lock(&fown->lock);
629 
630 	type = fown->pid_type;
631 	if (type == PIDTYPE_MAX) {
632 		group = 0;
633 		type = PIDTYPE_PID;
634 	}
635 
636 	pid = fown->pid;
637 	if (!pid)
638 		goto out_unlock_fown;
639 
640 	ret = 1;
641 
642 	read_lock(&tasklist_lock);
643 	do_each_pid_task(pid, type, p) {
644 		send_sigurg_to_task(p, fown, group);
645 	} while_each_pid_task(pid, type, p);
646 	read_unlock(&tasklist_lock);
647  out_unlock_fown:
648 	read_unlock(&fown->lock);
649 	return ret;
650 }
651 
652 static DEFINE_SPINLOCK(fasync_lock);
653 static struct kmem_cache *fasync_cache __read_mostly;
654 
fasync_free_rcu(struct rcu_head * head)655 static void fasync_free_rcu(struct rcu_head *head)
656 {
657 	kmem_cache_free(fasync_cache,
658 			container_of(head, struct fasync_struct, fa_rcu));
659 }
660 
661 /*
662  * Remove a fasync entry. If successfully removed, return
663  * positive and clear the FASYNC flag. If no entry exists,
664  * do nothing and return 0.
665  *
666  * NOTE! It is very important that the FASYNC flag always
667  * match the state "is the filp on a fasync list".
668  *
669  */
fasync_remove_entry(struct file * filp,struct fasync_struct ** fapp)670 int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
671 {
672 	struct fasync_struct *fa, **fp;
673 	int result = 0;
674 
675 	spin_lock(&filp->f_lock);
676 	spin_lock(&fasync_lock);
677 	for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
678 		if (fa->fa_file != filp)
679 			continue;
680 
681 		spin_lock_irq(&fa->fa_lock);
682 		fa->fa_file = NULL;
683 		spin_unlock_irq(&fa->fa_lock);
684 
685 		*fp = fa->fa_next;
686 		call_rcu(&fa->fa_rcu, fasync_free_rcu);
687 		filp->f_flags &= ~FASYNC;
688 		result = 1;
689 		break;
690 	}
691 	spin_unlock(&fasync_lock);
692 	spin_unlock(&filp->f_lock);
693 	return result;
694 }
695 
fasync_alloc(void)696 struct fasync_struct *fasync_alloc(void)
697 {
698 	return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
699 }
700 
701 /*
702  * NOTE! This can be used only for unused fasync entries:
703  * entries that actually got inserted on the fasync list
704  * need to be released by rcu - see fasync_remove_entry.
705  */
fasync_free(struct fasync_struct * new)706 void fasync_free(struct fasync_struct *new)
707 {
708 	kmem_cache_free(fasync_cache, new);
709 }
710 
711 /*
712  * Insert a new entry into the fasync list.  Return the pointer to the
713  * old one if we didn't use the new one.
714  *
715  * NOTE! It is very important that the FASYNC flag always
716  * match the state "is the filp on a fasync list".
717  */
fasync_insert_entry(int fd,struct file * filp,struct fasync_struct ** fapp,struct fasync_struct * new)718 struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
719 {
720         struct fasync_struct *fa, **fp;
721 
722 	spin_lock(&filp->f_lock);
723 	spin_lock(&fasync_lock);
724 	for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
725 		if (fa->fa_file != filp)
726 			continue;
727 
728 		spin_lock_irq(&fa->fa_lock);
729 		fa->fa_fd = fd;
730 		spin_unlock_irq(&fa->fa_lock);
731 		goto out;
732 	}
733 
734 	spin_lock_init(&new->fa_lock);
735 	new->magic = FASYNC_MAGIC;
736 	new->fa_file = filp;
737 	new->fa_fd = fd;
738 	new->fa_next = *fapp;
739 	rcu_assign_pointer(*fapp, new);
740 	filp->f_flags |= FASYNC;
741 
742 out:
743 	spin_unlock(&fasync_lock);
744 	spin_unlock(&filp->f_lock);
745 	return fa;
746 }
747 
748 /*
749  * Add a fasync entry. Return negative on error, positive if
750  * added, and zero if did nothing but change an existing one.
751  */
fasync_add_entry(int fd,struct file * filp,struct fasync_struct ** fapp)752 static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
753 {
754 	struct fasync_struct *new;
755 
756 	new = fasync_alloc();
757 	if (!new)
758 		return -ENOMEM;
759 
760 	/*
761 	 * fasync_insert_entry() returns the old (update) entry if
762 	 * it existed.
763 	 *
764 	 * So free the (unused) new entry and return 0 to let the
765 	 * caller know that we didn't add any new fasync entries.
766 	 */
767 	if (fasync_insert_entry(fd, filp, fapp, new)) {
768 		fasync_free(new);
769 		return 0;
770 	}
771 
772 	return 1;
773 }
774 
775 /*
776  * fasync_helper() is used by almost all character device drivers
777  * to set up the fasync queue, and for regular files by the file
778  * lease code. It returns negative on error, 0 if it did no changes
779  * and positive if it added/deleted the entry.
780  */
fasync_helper(int fd,struct file * filp,int on,struct fasync_struct ** fapp)781 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
782 {
783 	if (!on)
784 		return fasync_remove_entry(filp, fapp);
785 	return fasync_add_entry(fd, filp, fapp);
786 }
787 
788 EXPORT_SYMBOL(fasync_helper);
789 
790 /*
791  * rcu_read_lock() is held
792  */
kill_fasync_rcu(struct fasync_struct * fa,int sig,int band)793 static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
794 {
795 	while (fa) {
796 		struct fown_struct *fown;
797 		unsigned long flags;
798 
799 		if (fa->magic != FASYNC_MAGIC) {
800 			printk(KERN_ERR "kill_fasync: bad magic number in "
801 			       "fasync_struct!\n");
802 			return;
803 		}
804 		spin_lock_irqsave(&fa->fa_lock, flags);
805 		if (fa->fa_file) {
806 			fown = &fa->fa_file->f_owner;
807 			/* Don't send SIGURG to processes which have not set a
808 			   queued signum: SIGURG has its own default signalling
809 			   mechanism. */
810 			if (!(sig == SIGURG && fown->signum == 0))
811 				send_sigio(fown, fa->fa_fd, band);
812 		}
813 		spin_unlock_irqrestore(&fa->fa_lock, flags);
814 		fa = rcu_dereference(fa->fa_next);
815 	}
816 }
817 
kill_fasync(struct fasync_struct ** fp,int sig,int band)818 void kill_fasync(struct fasync_struct **fp, int sig, int band)
819 {
820 	/* First a quick test without locking: usually
821 	 * the list is empty.
822 	 */
823 	if (*fp) {
824 		rcu_read_lock();
825 		kill_fasync_rcu(rcu_dereference(*fp), sig, band);
826 		rcu_read_unlock();
827 	}
828 }
829 EXPORT_SYMBOL(kill_fasync);
830 
fcntl_init(void)831 static int __init fcntl_init(void)
832 {
833 	/*
834 	 * Please add new bits here to ensure allocation uniqueness.
835 	 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
836 	 * is defined as O_NONBLOCK on some platforms and not on others.
837 	 */
838 	BUILD_BUG_ON(19 - 1 /* for O_RDONLY being 0 */ != HWEIGHT32(
839 		O_RDONLY	| O_WRONLY	| O_RDWR	|
840 		O_CREAT		| O_EXCL	| O_NOCTTY	|
841 		O_TRUNC		| O_APPEND	| /* O_NONBLOCK	| */
842 		__O_SYNC	| O_DSYNC	| FASYNC	|
843 		O_DIRECT	| O_LARGEFILE	| O_DIRECTORY	|
844 		O_NOFOLLOW	| O_NOATIME	| O_CLOEXEC	|
845 		__FMODE_EXEC	| O_PATH
846 		));
847 
848 	fasync_cache = kmem_cache_create("fasync_cache",
849 		sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
850 	return 0;
851 }
852 
853 module_init(fcntl_init)
854