blob: 22d1c3df61acfa61ae957ede4aedfa01a3e9ab7a [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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>
Al Viro9f3acc32008-04-24 07:44:08 -040012#include <linux/fdtable.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080013#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/dnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/slab.h>
16#include <linux/module.h>
Jens Axboe35f3d142010-05-20 10:43:18 +020017#include <linux/pipe_fs_i.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/security.h>
19#include <linux/ptrace.h>
Jesper Juhl7ed20e12005-05-01 08:59:14 -070020#include <linux/signal.h>
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070021#include <linux/rcupdate.h>
Pavel Emelyanovb4888932007-10-18 23:40:14 -070022#include <linux/pid_namespace.h>
Cyrill Gorcunov1d151c32012-07-30 14:43:00 -070023#include <linux/user_namespace.h>
David Herrmann40e041a2014-08-08 14:25:27 -070024#include <linux/shmem_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025
26#include <asm/poll.h>
27#include <asm/siginfo.h>
28#include <asm/uaccess.h>
29
Jonathan Corbet76398422009-02-01 14:26:59 -070030#define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
32static int setfl(int fd, struct file * filp, unsigned long arg)
33{
Al Viro496ad9a2013-01-23 17:07:38 -050034 struct inode * inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -070035 int error = 0;
36
dean gaudet7d95c8f2006-02-03 03:04:30 -080037 /*
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))
Linus Torvalds1da177e2005-04-16 15:20:36 -070042 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))
Serge E. Hallyn2e149672011-03-23 16:43:26 -070046 if (!inode_owner_or_capable(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -070047 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
Al Viro72c2d532013-09-22 16:27:52 -040060 if (filp->f_op->check_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 error = filp->f_op->check_flags(arg);
62 if (error)
63 return error;
64
Jonathan Corbet218d11a2008-12-05 16:12:48 -070065 /*
Jonathan Corbet76398422009-02-01 14:26:59 -070066 * ->fasync() is responsible for setting the FASYNC bit.
Jonathan Corbet218d11a2008-12-05 16:12:48 -070067 */
Al Viro72c2d532013-09-22 16:27:52 -040068 if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op->fasync) {
Jonathan Corbet76398422009-02-01 14:26:59 -070069 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
70 if (error < 0)
71 goto out;
Jonathan Corbet60aa4922009-02-01 14:52:56 -070072 if (error > 0)
73 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070074 }
Jonathan Corbetdb1dd4d2009-02-06 15:25:24 -070075 spin_lock(&filp->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
Jonathan Corbetdb1dd4d2009-02-06 15:25:24 -070077 spin_unlock(&filp->f_lock);
Jonathan Corbet76398422009-02-01 14:26:59 -070078
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 return error;
81}
82
Eric W. Biederman609d7fa2006-10-02 02:17:15 -070083static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
Oleg Nesterov2f38d702009-06-16 22:07:46 +020084 int force)
Linus Torvalds1da177e2005-04-16 15:20:36 -070085{
Linus Torvalds80e1e822010-02-07 10:11:23 -080086 write_lock_irq(&filp->f_owner.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 if (force || !filp->f_owner.pid) {
Eric W. Biederman609d7fa2006-10-02 02:17:15 -070088 put_pid(filp->f_owner.pid);
89 filp->f_owner.pid = get_pid(pid);
90 filp->f_owner.pid_type = type;
Oleg Nesterov2f38d702009-06-16 22:07:46 +020091
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 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 }
Linus Torvalds80e1e822010-02-07 10:11:23 -080098 write_unlock_irq(&filp->f_owner.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070099}
100
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700101int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
102 int force)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103{
104 int err;
Oleg Nesterov2f38d702009-06-16 22:07:46 +0200105
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 err = security_file_set_fowner(filp);
107 if (err)
108 return err;
109
Oleg Nesterov2f38d702009-06-16 22:07:46 +0200110 f_modown(filp, pid, type, force);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 return 0;
112}
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700113EXPORT_SYMBOL(__f_setown);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700115int f_setown(struct file *filp, unsigned long arg, int force)
116{
117 enum pid_type type;
118 struct pid *pid;
119 int who = arg;
120 int result;
121 type = PIDTYPE_PID;
122 if (who < 0) {
123 type = PIDTYPE_PGID;
124 who = -who;
125 }
126 rcu_read_lock();
Pavel Emelyanovb4888932007-10-18 23:40:14 -0700127 pid = find_vpid(who);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700128 result = __f_setown(filp, pid, type, force);
129 rcu_read_unlock();
130 return result;
131}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132EXPORT_SYMBOL(f_setown);
133
134void f_delown(struct file *filp)
135{
Oleg Nesterov2f38d702009-06-16 22:07:46 +0200136 f_modown(filp, NULL, PIDTYPE_PID, 1);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700137}
138
139pid_t f_getown(struct file *filp)
140{
141 pid_t pid;
Eric W. Biederman43fa1ad2006-10-02 02:17:27 -0700142 read_lock(&filp->f_owner.lock);
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -0800143 pid = pid_vnr(filp->f_owner.pid);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700144 if (filp->f_owner.pid_type == PIDTYPE_PGID)
145 pid = -pid;
Eric W. Biederman43fa1ad2006-10-02 02:17:27 -0700146 read_unlock(&filp->f_owner.lock);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700147 return pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148}
149
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700150static int f_setown_ex(struct file *filp, unsigned long arg)
151{
Al Viro63784dd2012-09-26 21:43:05 -0400152 struct f_owner_ex __user *owner_p = (void __user *)arg;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700153 struct f_owner_ex owner;
154 struct pid *pid;
155 int type;
156 int ret;
157
158 ret = copy_from_user(&owner, owner_p, sizeof(owner));
159 if (ret)
Dan Carpenter5b544702010-06-03 12:35:42 +0200160 return -EFAULT;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700161
162 switch (owner.type) {
163 case F_OWNER_TID:
164 type = PIDTYPE_MAX;
165 break;
166
167 case F_OWNER_PID:
168 type = PIDTYPE_PID;
169 break;
170
Peter Zijlstra978b4052009-11-17 14:06:24 -0800171 case F_OWNER_PGRP:
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700172 type = PIDTYPE_PGID;
173 break;
174
175 default:
176 return -EINVAL;
177 }
178
179 rcu_read_lock();
180 pid = find_vpid(owner.pid);
181 if (owner.pid && !pid)
182 ret = -ESRCH;
183 else
184 ret = __f_setown(filp, pid, type, 1);
185 rcu_read_unlock();
186
187 return ret;
188}
189
190static int f_getown_ex(struct file *filp, unsigned long arg)
191{
Al Viro63784dd2012-09-26 21:43:05 -0400192 struct f_owner_ex __user *owner_p = (void __user *)arg;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700193 struct f_owner_ex owner;
194 int ret = 0;
195
196 read_lock(&filp->f_owner.lock);
197 owner.pid = pid_vnr(filp->f_owner.pid);
198 switch (filp->f_owner.pid_type) {
199 case PIDTYPE_MAX:
200 owner.type = F_OWNER_TID;
201 break;
202
203 case PIDTYPE_PID:
204 owner.type = F_OWNER_PID;
205 break;
206
207 case PIDTYPE_PGID:
Peter Zijlstra978b4052009-11-17 14:06:24 -0800208 owner.type = F_OWNER_PGRP;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700209 break;
210
211 default:
212 WARN_ON(1);
213 ret = -EINVAL;
214 break;
215 }
216 read_unlock(&filp->f_owner.lock);
217
Dan Carpenter5b544702010-06-03 12:35:42 +0200218 if (!ret) {
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700219 ret = copy_to_user(owner_p, &owner, sizeof(owner));
Dan Carpenter5b544702010-06-03 12:35:42 +0200220 if (ret)
221 ret = -EFAULT;
222 }
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700223 return ret;
224}
225
Cyrill Gorcunov1d151c32012-07-30 14:43:00 -0700226#ifdef CONFIG_CHECKPOINT_RESTORE
227static int f_getowner_uids(struct file *filp, unsigned long arg)
228{
229 struct user_namespace *user_ns = current_user_ns();
Al Viro63784dd2012-09-26 21:43:05 -0400230 uid_t __user *dst = (void __user *)arg;
Cyrill Gorcunov1d151c32012-07-30 14:43:00 -0700231 uid_t src[2];
232 int err;
233
234 read_lock(&filp->f_owner.lock);
235 src[0] = from_kuid(user_ns, filp->f_owner.uid);
236 src[1] = from_kuid(user_ns, filp->f_owner.euid);
237 read_unlock(&filp->f_owner.lock);
238
239 err = put_user(src[0], &dst[0]);
240 err |= put_user(src[1], &dst[1]);
241
242 return err;
243}
244#else
245static int f_getowner_uids(struct file *filp, unsigned long arg)
246{
247 return -EINVAL;
248}
249#endif
250
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
252 struct file *filp)
253{
254 long err = -EINVAL;
255
256 switch (cmd) {
257 case F_DUPFD:
Al Virofe17f222012-08-21 11:48:11 -0400258 err = f_dupfd(arg, filp, 0);
259 break;
Ulrich Drepper22d2b352007-10-16 23:30:26 -0700260 case F_DUPFD_CLOEXEC:
Al Viro12197712012-10-08 23:21:58 +0100261 err = f_dupfd(arg, filp, O_CLOEXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 break;
263 case F_GETFD:
264 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
265 break;
266 case F_SETFD:
267 err = 0;
268 set_close_on_exec(fd, arg & FD_CLOEXEC);
269 break;
270 case F_GETFL:
271 err = filp->f_flags;
272 break;
273 case F_SETFL:
274 err = setfl(fd, filp, arg);
275 break;
Jeff Layton5d50ffd2014-02-03 12:13:10 -0500276#if BITS_PER_LONG != 32
277 /* 32-bit arches must use fcntl64() */
Jeff Layton0d3f7a22014-04-22 08:23:58 -0400278 case F_OFD_GETLK:
Jeff Layton5d50ffd2014-02-03 12:13:10 -0500279#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 case F_GETLK:
Jeff Laytonc1e62b82014-02-03 12:13:09 -0500281 err = fcntl_getlk(filp, cmd, (struct flock __user *) arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 break;
Jeff Layton5d50ffd2014-02-03 12:13:10 -0500283#if BITS_PER_LONG != 32
284 /* 32-bit arches must use fcntl64() */
Jeff Layton0d3f7a22014-04-22 08:23:58 -0400285 case F_OFD_SETLK:
286 case F_OFD_SETLKW:
Jeff Layton5d50ffd2014-02-03 12:13:10 -0500287#endif
288 /* Fallthrough */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 case F_SETLK:
290 case F_SETLKW:
Peter Staubachc2936212005-07-27 11:45:09 -0700291 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 break;
293 case F_GETOWN:
294 /*
295 * XXX If f_owner is a process group, the
296 * negative return value will get converted
297 * into an error. Oops. If we keep the
298 * current syscall conventions, the only way
299 * to fix this will be in libc.
300 */
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700301 err = f_getown(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 force_successful_syscall_return();
303 break;
304 case F_SETOWN:
305 err = f_setown(filp, arg, 1);
306 break;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700307 case F_GETOWN_EX:
308 err = f_getown_ex(filp, arg);
309 break;
310 case F_SETOWN_EX:
311 err = f_setown_ex(filp, arg);
312 break;
Cyrill Gorcunov1d151c32012-07-30 14:43:00 -0700313 case F_GETOWNER_UIDS:
314 err = f_getowner_uids(filp, arg);
315 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 case F_GETSIG:
317 err = filp->f_owner.signum;
318 break;
319 case F_SETSIG:
320 /* arg == 0 restores default behaviour. */
Jesper Juhl7ed20e12005-05-01 08:59:14 -0700321 if (!valid_signal(arg)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 break;
323 }
324 err = 0;
325 filp->f_owner.signum = arg;
326 break;
327 case F_GETLEASE:
328 err = fcntl_getlease(filp);
329 break;
330 case F_SETLEASE:
331 err = fcntl_setlease(fd, filp, arg);
332 break;
333 case F_NOTIFY:
334 err = fcntl_dirnotify(fd, filp, arg);
335 break;
Jens Axboe35f3d142010-05-20 10:43:18 +0200336 case F_SETPIPE_SZ:
337 case F_GETPIPE_SZ:
338 err = pipe_fcntl(filp, cmd, arg);
339 break;
David Herrmann40e041a2014-08-08 14:25:27 -0700340 case F_ADD_SEALS:
341 case F_GET_SEALS:
342 err = shmem_fcntl(filp, cmd, arg);
343 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 default:
345 break;
346 }
347 return err;
348}
349
Al Viro1abf0c72011-03-13 03:51:11 -0400350static int check_fcntl_cmd(unsigned cmd)
351{
352 switch (cmd) {
353 case F_DUPFD:
354 case F_DUPFD_CLOEXEC:
355 case F_GETFD:
356 case F_SETFD:
357 case F_GETFL:
358 return 1;
359 }
360 return 0;
361}
362
Heiko Carstensa26eab22009-01-14 14:14:17 +0100363SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364{
Al Viro2903ff02012-08-28 12:52:22 -0400365 struct fd f = fdget_raw(fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 long err = -EBADF;
367
Al Viro2903ff02012-08-28 12:52:22 -0400368 if (!f.file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 goto out;
370
Al Viro2903ff02012-08-28 12:52:22 -0400371 if (unlikely(f.file->f_mode & FMODE_PATH)) {
Al Viro545ec2c2012-04-21 18:42:19 -0400372 if (!check_fcntl_cmd(cmd))
373 goto out1;
Al Viro1abf0c72011-03-13 03:51:11 -0400374 }
375
Al Viro2903ff02012-08-28 12:52:22 -0400376 err = security_file_fcntl(f.file, cmd, arg);
Al Viro545ec2c2012-04-21 18:42:19 -0400377 if (!err)
Al Viro2903ff02012-08-28 12:52:22 -0400378 err = do_fcntl(fd, cmd, arg, f.file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379
Al Viro545ec2c2012-04-21 18:42:19 -0400380out1:
Al Viro2903ff02012-08-28 12:52:22 -0400381 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382out:
383 return err;
384}
385
386#if BITS_PER_LONG == 32
Heiko Carstensa26eab22009-01-14 14:14:17 +0100387SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
388 unsigned long, arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389{
Al Viro2903ff02012-08-28 12:52:22 -0400390 struct fd f = fdget_raw(fd);
Al Viro545ec2c2012-04-21 18:42:19 -0400391 long err = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392
Al Viro2903ff02012-08-28 12:52:22 -0400393 if (!f.file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 goto out;
395
Al Viro2903ff02012-08-28 12:52:22 -0400396 if (unlikely(f.file->f_mode & FMODE_PATH)) {
Al Viro545ec2c2012-04-21 18:42:19 -0400397 if (!check_fcntl_cmd(cmd))
398 goto out1;
Al Viro1abf0c72011-03-13 03:51:11 -0400399 }
400
Al Viro2903ff02012-08-28 12:52:22 -0400401 err = security_file_fcntl(f.file, cmd, arg);
Al Viro545ec2c2012-04-21 18:42:19 -0400402 if (err)
403 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404
405 switch (cmd) {
Jeff Layton5d50ffd2014-02-03 12:13:10 -0500406 case F_GETLK64:
Jeff Layton0d3f7a22014-04-22 08:23:58 -0400407 case F_OFD_GETLK:
Jeff Layton5d50ffd2014-02-03 12:13:10 -0500408 err = fcntl_getlk64(f.file, cmd, (struct flock64 __user *) arg);
409 break;
410 case F_SETLK64:
411 case F_SETLKW64:
Jeff Layton0d3f7a22014-04-22 08:23:58 -0400412 case F_OFD_SETLK:
413 case F_OFD_SETLKW:
Jeff Layton5d50ffd2014-02-03 12:13:10 -0500414 err = fcntl_setlk64(fd, f.file, cmd,
415 (struct flock64 __user *) arg);
416 break;
417 default:
418 err = do_fcntl(fd, cmd, arg, f.file);
419 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 }
Al Viro545ec2c2012-04-21 18:42:19 -0400421out1:
Al Viro2903ff02012-08-28 12:52:22 -0400422 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423out:
424 return err;
425}
426#endif
427
428/* Table to convert sigio signal codes into poll band bitmaps */
429
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800430static const long band_table[NSIGPOLL] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 POLLIN | POLLRDNORM, /* POLL_IN */
432 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
433 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
434 POLLERR, /* POLL_ERR */
435 POLLPRI | POLLRDBAND, /* POLL_PRI */
436 POLLHUP | POLLERR /* POLL_HUP */
437};
438
439static inline int sigio_perm(struct task_struct *p,
440 struct fown_struct *fown, int sig)
441{
David Howellsc69e8d92008-11-14 10:39:19 +1100442 const struct cred *cred;
443 int ret;
444
445 rcu_read_lock();
446 cred = __task_cred(p);
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -0800447 ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
448 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
449 uid_eq(fown->uid, cred->suid) || uid_eq(fown->uid, cred->uid)) &&
David Howellsc69e8d92008-11-14 10:39:19 +1100450 !security_file_send_sigiotask(p, fown, sig));
451 rcu_read_unlock();
452 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453}
454
455static void send_sigio_to_task(struct task_struct *p,
Oleg Nesterov8eeee4e2009-06-17 00:27:10 +0200456 struct fown_struct *fown,
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700457 int fd, int reason, int group)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458{
Oleg Nesterov8eeee4e2009-06-17 00:27:10 +0200459 /*
460 * F_SETSIG can change ->signum lockless in parallel, make
461 * sure we read it once and use the same value throughout.
462 */
463 int signum = ACCESS_ONCE(fown->signum);
464
465 if (!sigio_perm(p, fown, signum))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 return;
467
Oleg Nesterov8eeee4e2009-06-17 00:27:10 +0200468 switch (signum) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 siginfo_t si;
470 default:
471 /* Queue a rt signal with the appropriate fd as its
472 value. We use SI_SIGIO as the source, not
473 SI_KERNEL, since kernel signals always get
474 delivered even if we can't queue. Failure to
475 queue in this case _should_ be reported; we fall
476 back to SIGIO in that case. --sct */
Oleg Nesterov8eeee4e2009-06-17 00:27:10 +0200477 si.si_signo = signum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 si.si_errno = 0;
479 si.si_code = reason;
480 /* Make sure we are called with one of the POLL_*
481 reasons, otherwise we could leak kernel stack into
482 userspace. */
Eric Sesterhennf6298aa2006-04-02 13:37:19 +0200483 BUG_ON((reason & __SI_MASK) != __SI_POLL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 if (reason - POLL_IN >= NSIGPOLL)
485 si.si_band = ~0L;
486 else
487 si.si_band = band_table[reason - POLL_IN];
488 si.si_fd = fd;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700489 if (!do_send_sig_info(signum, &si, p, group))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 break;
491 /* fall-through: fall back on the old plain SIGIO signal */
492 case 0:
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700493 do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 }
495}
496
497void send_sigio(struct fown_struct *fown, int fd, int band)
498{
499 struct task_struct *p;
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700500 enum pid_type type;
501 struct pid *pid;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700502 int group = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503
504 read_lock(&fown->lock);
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700505
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700506 type = fown->pid_type;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700507 if (type == PIDTYPE_MAX) {
508 group = 0;
509 type = PIDTYPE_PID;
510 }
511
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 pid = fown->pid;
513 if (!pid)
514 goto out_unlock_fown;
515
516 read_lock(&tasklist_lock);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700517 do_each_pid_task(pid, type, p) {
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700518 send_sigio_to_task(p, fown, fd, band, group);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700519 } while_each_pid_task(pid, type, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 read_unlock(&tasklist_lock);
521 out_unlock_fown:
522 read_unlock(&fown->lock);
523}
524
525static void send_sigurg_to_task(struct task_struct *p,
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700526 struct fown_struct *fown, int group)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527{
528 if (sigio_perm(p, fown, SIGURG))
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700529 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530}
531
532int send_sigurg(struct fown_struct *fown)
533{
534 struct task_struct *p;
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700535 enum pid_type type;
536 struct pid *pid;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700537 int group = 1;
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700538 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539
540 read_lock(&fown->lock);
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700541
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700542 type = fown->pid_type;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700543 if (type == PIDTYPE_MAX) {
544 group = 0;
545 type = PIDTYPE_PID;
546 }
547
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 pid = fown->pid;
549 if (!pid)
550 goto out_unlock_fown;
551
552 ret = 1;
553
554 read_lock(&tasklist_lock);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700555 do_each_pid_task(pid, type, p) {
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700556 send_sigurg_to_task(p, fown, group);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700557 } while_each_pid_task(pid, type, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 read_unlock(&tasklist_lock);
559 out_unlock_fown:
560 read_unlock(&fown->lock);
561 return ret;
562}
563
Eric Dumazet989a2972010-04-14 09:55:35 +0000564static DEFINE_SPINLOCK(fasync_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800565static struct kmem_cache *fasync_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566
Eric Dumazet989a2972010-04-14 09:55:35 +0000567static void fasync_free_rcu(struct rcu_head *head)
568{
569 kmem_cache_free(fasync_cache,
570 container_of(head, struct fasync_struct, fa_rcu));
571}
572
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573/*
Linus Torvalds53281b62009-12-16 08:23:37 -0800574 * Remove a fasync entry. If successfully removed, return
575 * positive and clear the FASYNC flag. If no entry exists,
576 * do nothing and return 0.
577 *
578 * NOTE! It is very important that the FASYNC flag always
579 * match the state "is the filp on a fasync list".
580 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 */
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400582int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583{
584 struct fasync_struct *fa, **fp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 int result = 0;
586
Jonathan Corbet4a6a4492009-03-27 12:24:31 -0600587 spin_lock(&filp->f_lock);
Eric Dumazet989a2972010-04-14 09:55:35 +0000588 spin_lock(&fasync_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
Linus Torvalds53281b62009-12-16 08:23:37 -0800590 if (fa->fa_file != filp)
591 continue;
Eric Dumazet989a2972010-04-14 09:55:35 +0000592
593 spin_lock_irq(&fa->fa_lock);
594 fa->fa_file = NULL;
595 spin_unlock_irq(&fa->fa_lock);
596
Linus Torvalds53281b62009-12-16 08:23:37 -0800597 *fp = fa->fa_next;
Eric Dumazet989a2972010-04-14 09:55:35 +0000598 call_rcu(&fa->fa_rcu, fasync_free_rcu);
Jonathan Corbet76398422009-02-01 14:26:59 -0700599 filp->f_flags &= ~FASYNC;
Linus Torvalds53281b62009-12-16 08:23:37 -0800600 result = 1;
601 break;
602 }
Eric Dumazet989a2972010-04-14 09:55:35 +0000603 spin_unlock(&fasync_lock);
Jonathan Corbet4a6a4492009-03-27 12:24:31 -0600604 spin_unlock(&filp->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 return result;
606}
607
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400608struct fasync_struct *fasync_alloc(void)
Linus Torvalds53281b62009-12-16 08:23:37 -0800609{
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400610 return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
611}
Linus Torvalds53281b62009-12-16 08:23:37 -0800612
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400613/*
614 * NOTE! This can be used only for unused fasync entries:
615 * entries that actually got inserted on the fasync list
616 * need to be released by rcu - see fasync_remove_entry.
617 */
618void fasync_free(struct fasync_struct *new)
619{
620 kmem_cache_free(fasync_cache, new);
621}
622
623/*
624 * Insert a new entry into the fasync list. Return the pointer to the
625 * old one if we didn't use the new one.
Linus Torvalds55f335a2010-10-27 18:17:02 -0700626 *
627 * NOTE! It is very important that the FASYNC flag always
628 * match the state "is the filp on a fasync list".
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400629 */
630struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
631{
632 struct fasync_struct *fa, **fp;
Linus Torvalds53281b62009-12-16 08:23:37 -0800633
634 spin_lock(&filp->f_lock);
Eric Dumazet989a2972010-04-14 09:55:35 +0000635 spin_lock(&fasync_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800636 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
637 if (fa->fa_file != filp)
638 continue;
Eric Dumazet989a2972010-04-14 09:55:35 +0000639
640 spin_lock_irq(&fa->fa_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800641 fa->fa_fd = fd;
Eric Dumazet989a2972010-04-14 09:55:35 +0000642 spin_unlock_irq(&fa->fa_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800643 goto out;
644 }
645
Eric Dumazet989a2972010-04-14 09:55:35 +0000646 spin_lock_init(&new->fa_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800647 new->magic = FASYNC_MAGIC;
648 new->fa_file = filp;
649 new->fa_fd = fd;
650 new->fa_next = *fapp;
Eric Dumazet989a2972010-04-14 09:55:35 +0000651 rcu_assign_pointer(*fapp, new);
Linus Torvalds53281b62009-12-16 08:23:37 -0800652 filp->f_flags |= FASYNC;
653
654out:
Eric Dumazet989a2972010-04-14 09:55:35 +0000655 spin_unlock(&fasync_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800656 spin_unlock(&filp->f_lock);
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400657 return fa;
658}
659
660/*
661 * Add a fasync entry. Return negative on error, positive if
662 * added, and zero if did nothing but change an existing one.
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400663 */
664static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
665{
666 struct fasync_struct *new;
667
668 new = fasync_alloc();
669 if (!new)
670 return -ENOMEM;
671
672 /*
673 * fasync_insert_entry() returns the old (update) entry if
674 * it existed.
675 *
676 * So free the (unused) new entry and return 0 to let the
677 * caller know that we didn't add any new fasync entries.
678 */
679 if (fasync_insert_entry(fd, filp, fapp, new)) {
680 fasync_free(new);
681 return 0;
682 }
683
684 return 1;
Linus Torvalds53281b62009-12-16 08:23:37 -0800685}
686
687/*
688 * fasync_helper() is used by almost all character device drivers
689 * to set up the fasync queue, and for regular files by the file
690 * lease code. It returns negative on error, 0 if it did no changes
691 * and positive if it added/deleted the entry.
692 */
693int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
694{
695 if (!on)
696 return fasync_remove_entry(filp, fapp);
697 return fasync_add_entry(fd, filp, fapp);
698}
699
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700EXPORT_SYMBOL(fasync_helper);
701
Eric Dumazet989a2972010-04-14 09:55:35 +0000702/*
703 * rcu_read_lock() is held
704 */
705static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706{
707 while (fa) {
Eric Dumazet989a2972010-04-14 09:55:35 +0000708 struct fown_struct *fown;
Andrew Mortonf4985dc2010-06-29 15:05:42 -0700709 unsigned long flags;
710
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 if (fa->magic != FASYNC_MAGIC) {
712 printk(KERN_ERR "kill_fasync: bad magic number in "
713 "fasync_struct!\n");
714 return;
715 }
Andrew Mortonf4985dc2010-06-29 15:05:42 -0700716 spin_lock_irqsave(&fa->fa_lock, flags);
Eric Dumazet989a2972010-04-14 09:55:35 +0000717 if (fa->fa_file) {
718 fown = &fa->fa_file->f_owner;
719 /* Don't send SIGURG to processes which have not set a
720 queued signum: SIGURG has its own default signalling
721 mechanism. */
722 if (!(sig == SIGURG && fown->signum == 0))
723 send_sigio(fown, fa->fa_fd, band);
724 }
Andrew Mortonf4985dc2010-06-29 15:05:42 -0700725 spin_unlock_irqrestore(&fa->fa_lock, flags);
Eric Dumazet989a2972010-04-14 09:55:35 +0000726 fa = rcu_dereference(fa->fa_next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 }
728}
729
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730void kill_fasync(struct fasync_struct **fp, int sig, int band)
731{
732 /* First a quick test without locking: usually
733 * the list is empty.
734 */
735 if (*fp) {
Eric Dumazet989a2972010-04-14 09:55:35 +0000736 rcu_read_lock();
737 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
738 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 }
740}
741EXPORT_SYMBOL(kill_fasync);
742
Wu Fengguang454eedb2010-08-10 18:01:29 -0700743static int __init fcntl_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744{
James Bottomley3ab04d52010-09-09 16:38:12 -0700745 /*
746 * Please add new bits here to ensure allocation uniqueness.
747 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
748 * is defined as O_NONBLOCK on some platforms and not on others.
749 */
Zheng Liu3d62c452013-07-25 08:13:19 +0800750 BUILD_BUG_ON(20 - 1 /* for O_RDONLY being 0 */ != HWEIGHT32(
Wu Fengguang454eedb2010-08-10 18:01:29 -0700751 O_RDONLY | O_WRONLY | O_RDWR |
752 O_CREAT | O_EXCL | O_NOCTTY |
James Bottomley3ab04d52010-09-09 16:38:12 -0700753 O_TRUNC | O_APPEND | /* O_NONBLOCK | */
Wu Fengguang454eedb2010-08-10 18:01:29 -0700754 __O_SYNC | O_DSYNC | FASYNC |
755 O_DIRECT | O_LARGEFILE | O_DIRECTORY |
756 O_NOFOLLOW | O_NOATIME | O_CLOEXEC |
Zheng Liu3d62c452013-07-25 08:13:19 +0800757 __FMODE_EXEC | O_PATH | __O_TMPFILE
Wu Fengguang454eedb2010-08-10 18:01:29 -0700758 ));
759
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 fasync_cache = kmem_cache_create("fasync_cache",
Paul Mundt20c2df82007-07-20 10:11:58 +0900761 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 return 0;
763}
764
Wu Fengguang454eedb2010-08-10 18:01:29 -0700765module_init(fcntl_init)