blob: 350a2c8cfd28f3f1339613b31564da1c2422e917 [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
Stanislav Kinsburskiy0dbf5f22015-12-15 19:41:31 +040054 /* Pipe packetized mode is controlled by O_DIRECT flag */
55 if (!S_ISFIFO(filp->f_inode->i_mode) && (arg & O_DIRECT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070056 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
57 !filp->f_mapping->a_ops->direct_IO)
58 return -EINVAL;
59 }
60
Al Viro72c2d532013-09-22 16:27:52 -040061 if (filp->f_op->check_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 error = filp->f_op->check_flags(arg);
63 if (error)
64 return error;
65
Jonathan Corbet218d11a2008-12-05 16:12:48 -070066 /*
Jonathan Corbet76398422009-02-01 14:26:59 -070067 * ->fasync() is responsible for setting the FASYNC bit.
Jonathan Corbet218d11a2008-12-05 16:12:48 -070068 */
Al Viro72c2d532013-09-22 16:27:52 -040069 if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op->fasync) {
Jonathan Corbet76398422009-02-01 14:26:59 -070070 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
71 if (error < 0)
72 goto out;
Jonathan Corbet60aa4922009-02-01 14:52:56 -070073 if (error > 0)
74 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 }
Jonathan Corbetdb1dd4d2009-02-06 15:25:24 -070076 spin_lock(&filp->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
Jonathan Corbetdb1dd4d2009-02-06 15:25:24 -070078 spin_unlock(&filp->f_lock);
Jonathan Corbet76398422009-02-01 14:26:59 -070079
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 return error;
82}
83
Eric W. Biederman609d7fa2006-10-02 02:17:15 -070084static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
Oleg Nesterov2f38d702009-06-16 22:07:46 +020085 int force)
Linus Torvalds1da177e2005-04-16 15:20:36 -070086{
Linus Torvalds80e1e822010-02-07 10:11:23 -080087 write_lock_irq(&filp->f_owner.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 if (force || !filp->f_owner.pid) {
Eric W. Biederman609d7fa2006-10-02 02:17:15 -070089 put_pid(filp->f_owner.pid);
90 filp->f_owner.pid = get_pid(pid);
91 filp->f_owner.pid_type = type;
Oleg Nesterov2f38d702009-06-16 22:07:46 +020092
93 if (pid) {
94 const struct cred *cred = current_cred();
95 filp->f_owner.uid = cred->uid;
96 filp->f_owner.euid = cred->euid;
97 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 }
Linus Torvalds80e1e822010-02-07 10:11:23 -080099 write_unlock_irq(&filp->f_owner.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100}
101
Jeff Laytone0b93ed2014-08-22 11:27:32 -0400102void __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700103 int force)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104{
Jeff Laytone0b93ed2014-08-22 11:27:32 -0400105 security_file_set_fowner(filp);
Oleg Nesterov2f38d702009-06-16 22:07:46 +0200106 f_modown(filp, pid, type, force);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107}
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700108EXPORT_SYMBOL(__f_setown);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109
Jeff Laytone0b93ed2014-08-22 11:27:32 -0400110void f_setown(struct file *filp, unsigned long arg, int force)
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700111{
112 enum pid_type type;
113 struct pid *pid;
114 int who = arg;
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700115 type = PIDTYPE_PID;
116 if (who < 0) {
117 type = PIDTYPE_PGID;
118 who = -who;
119 }
120 rcu_read_lock();
Pavel Emelyanovb4888932007-10-18 23:40:14 -0700121 pid = find_vpid(who);
Jeff Laytone0b93ed2014-08-22 11:27:32 -0400122 __f_setown(filp, pid, type, force);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700123 rcu_read_unlock();
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700124}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125EXPORT_SYMBOL(f_setown);
126
127void f_delown(struct file *filp)
128{
Oleg Nesterov2f38d702009-06-16 22:07:46 +0200129 f_modown(filp, NULL, PIDTYPE_PID, 1);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700130}
131
132pid_t f_getown(struct file *filp)
133{
134 pid_t pid;
Eric W. Biederman43fa1ad2006-10-02 02:17:27 -0700135 read_lock(&filp->f_owner.lock);
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -0800136 pid = pid_vnr(filp->f_owner.pid);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700137 if (filp->f_owner.pid_type == PIDTYPE_PGID)
138 pid = -pid;
Eric W. Biederman43fa1ad2006-10-02 02:17:27 -0700139 read_unlock(&filp->f_owner.lock);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700140 return pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141}
142
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700143static int f_setown_ex(struct file *filp, unsigned long arg)
144{
Al Viro63784dd2012-09-26 21:43:05 -0400145 struct f_owner_ex __user *owner_p = (void __user *)arg;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700146 struct f_owner_ex owner;
147 struct pid *pid;
148 int type;
149 int ret;
150
151 ret = copy_from_user(&owner, owner_p, sizeof(owner));
152 if (ret)
Dan Carpenter5b544702010-06-03 12:35:42 +0200153 return -EFAULT;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700154
155 switch (owner.type) {
156 case F_OWNER_TID:
157 type = PIDTYPE_MAX;
158 break;
159
160 case F_OWNER_PID:
161 type = PIDTYPE_PID;
162 break;
163
Peter Zijlstra978b4052009-11-17 14:06:24 -0800164 case F_OWNER_PGRP:
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700165 type = PIDTYPE_PGID;
166 break;
167
168 default:
169 return -EINVAL;
170 }
171
172 rcu_read_lock();
173 pid = find_vpid(owner.pid);
174 if (owner.pid && !pid)
175 ret = -ESRCH;
176 else
Jeff Laytone0b93ed2014-08-22 11:27:32 -0400177 __f_setown(filp, pid, type, 1);
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700178 rcu_read_unlock();
179
180 return ret;
181}
182
183static int f_getown_ex(struct file *filp, unsigned long arg)
184{
Al Viro63784dd2012-09-26 21:43:05 -0400185 struct f_owner_ex __user *owner_p = (void __user *)arg;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700186 struct f_owner_ex owner;
187 int ret = 0;
188
189 read_lock(&filp->f_owner.lock);
190 owner.pid = pid_vnr(filp->f_owner.pid);
191 switch (filp->f_owner.pid_type) {
192 case PIDTYPE_MAX:
193 owner.type = F_OWNER_TID;
194 break;
195
196 case PIDTYPE_PID:
197 owner.type = F_OWNER_PID;
198 break;
199
200 case PIDTYPE_PGID:
Peter Zijlstra978b4052009-11-17 14:06:24 -0800201 owner.type = F_OWNER_PGRP;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700202 break;
203
204 default:
205 WARN_ON(1);
206 ret = -EINVAL;
207 break;
208 }
209 read_unlock(&filp->f_owner.lock);
210
Dan Carpenter5b544702010-06-03 12:35:42 +0200211 if (!ret) {
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700212 ret = copy_to_user(owner_p, &owner, sizeof(owner));
Dan Carpenter5b544702010-06-03 12:35:42 +0200213 if (ret)
214 ret = -EFAULT;
215 }
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700216 return ret;
217}
218
Cyrill Gorcunov1d151c32012-07-30 14:43:00 -0700219#ifdef CONFIG_CHECKPOINT_RESTORE
220static int f_getowner_uids(struct file *filp, unsigned long arg)
221{
222 struct user_namespace *user_ns = current_user_ns();
Al Viro63784dd2012-09-26 21:43:05 -0400223 uid_t __user *dst = (void __user *)arg;
Cyrill Gorcunov1d151c32012-07-30 14:43:00 -0700224 uid_t src[2];
225 int err;
226
227 read_lock(&filp->f_owner.lock);
228 src[0] = from_kuid(user_ns, filp->f_owner.uid);
229 src[1] = from_kuid(user_ns, filp->f_owner.euid);
230 read_unlock(&filp->f_owner.lock);
231
232 err = put_user(src[0], &dst[0]);
233 err |= put_user(src[1], &dst[1]);
234
235 return err;
236}
237#else
238static int f_getowner_uids(struct file *filp, unsigned long arg)
239{
240 return -EINVAL;
241}
242#endif
243
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
245 struct file *filp)
246{
247 long err = -EINVAL;
248
249 switch (cmd) {
250 case F_DUPFD:
Al Virofe17f222012-08-21 11:48:11 -0400251 err = f_dupfd(arg, filp, 0);
252 break;
Ulrich Drepper22d2b352007-10-16 23:30:26 -0700253 case F_DUPFD_CLOEXEC:
Al Viro12197712012-10-08 23:21:58 +0100254 err = f_dupfd(arg, filp, O_CLOEXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 break;
256 case F_GETFD:
257 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
258 break;
259 case F_SETFD:
260 err = 0;
261 set_close_on_exec(fd, arg & FD_CLOEXEC);
262 break;
263 case F_GETFL:
264 err = filp->f_flags;
265 break;
266 case F_SETFL:
267 err = setfl(fd, filp, arg);
268 break;
Jeff Layton5d50ffd2014-02-03 12:13:10 -0500269#if BITS_PER_LONG != 32
270 /* 32-bit arches must use fcntl64() */
Jeff Layton0d3f7a22014-04-22 08:23:58 -0400271 case F_OFD_GETLK:
Jeff Layton5d50ffd2014-02-03 12:13:10 -0500272#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 case F_GETLK:
Jeff Laytonc1e62b82014-02-03 12:13:09 -0500274 err = fcntl_getlk(filp, cmd, (struct flock __user *) arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 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_SETLK:
279 case F_OFD_SETLKW:
Jeff Layton5d50ffd2014-02-03 12:13:10 -0500280#endif
281 /* Fallthrough */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 case F_SETLK:
283 case F_SETLKW:
Peter Staubachc2936212005-07-27 11:45:09 -0700284 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 break;
286 case F_GETOWN:
287 /*
288 * XXX If f_owner is a process group, the
289 * negative return value will get converted
290 * into an error. Oops. If we keep the
291 * current syscall conventions, the only way
292 * to fix this will be in libc.
293 */
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700294 err = f_getown(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 force_successful_syscall_return();
296 break;
297 case F_SETOWN:
Jeff Laytone0b93ed2014-08-22 11:27:32 -0400298 f_setown(filp, arg, 1);
299 err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 break;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700301 case F_GETOWN_EX:
302 err = f_getown_ex(filp, arg);
303 break;
304 case F_SETOWN_EX:
305 err = f_setown_ex(filp, arg);
306 break;
Cyrill Gorcunov1d151c32012-07-30 14:43:00 -0700307 case F_GETOWNER_UIDS:
308 err = f_getowner_uids(filp, arg);
309 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 case F_GETSIG:
311 err = filp->f_owner.signum;
312 break;
313 case F_SETSIG:
314 /* arg == 0 restores default behaviour. */
Jesper Juhl7ed20e12005-05-01 08:59:14 -0700315 if (!valid_signal(arg)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 break;
317 }
318 err = 0;
319 filp->f_owner.signum = arg;
320 break;
321 case F_GETLEASE:
322 err = fcntl_getlease(filp);
323 break;
324 case F_SETLEASE:
325 err = fcntl_setlease(fd, filp, arg);
326 break;
327 case F_NOTIFY:
328 err = fcntl_dirnotify(fd, filp, arg);
329 break;
Jens Axboe35f3d142010-05-20 10:43:18 +0200330 case F_SETPIPE_SZ:
331 case F_GETPIPE_SZ:
332 err = pipe_fcntl(filp, cmd, arg);
333 break;
David Herrmann40e041a2014-08-08 14:25:27 -0700334 case F_ADD_SEALS:
335 case F_GET_SEALS:
336 err = shmem_fcntl(filp, cmd, arg);
337 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 default:
339 break;
340 }
341 return err;
342}
343
Al Viro1abf0c72011-03-13 03:51:11 -0400344static int check_fcntl_cmd(unsigned cmd)
345{
346 switch (cmd) {
347 case F_DUPFD:
348 case F_DUPFD_CLOEXEC:
349 case F_GETFD:
350 case F_SETFD:
351 case F_GETFL:
352 return 1;
353 }
354 return 0;
355}
356
Heiko Carstensa26eab22009-01-14 14:14:17 +0100357SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358{
Al Viro2903ff02012-08-28 12:52:22 -0400359 struct fd f = fdget_raw(fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 long err = -EBADF;
361
Al Viro2903ff02012-08-28 12:52:22 -0400362 if (!f.file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 goto out;
364
Al Viro2903ff02012-08-28 12:52:22 -0400365 if (unlikely(f.file->f_mode & FMODE_PATH)) {
Al Viro545ec2c2012-04-21 18:42:19 -0400366 if (!check_fcntl_cmd(cmd))
367 goto out1;
Al Viro1abf0c72011-03-13 03:51:11 -0400368 }
369
Al Viro2903ff02012-08-28 12:52:22 -0400370 err = security_file_fcntl(f.file, cmd, arg);
Al Viro545ec2c2012-04-21 18:42:19 -0400371 if (!err)
Al Viro2903ff02012-08-28 12:52:22 -0400372 err = do_fcntl(fd, cmd, arg, f.file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373
Al Viro545ec2c2012-04-21 18:42:19 -0400374out1:
Al Viro2903ff02012-08-28 12:52:22 -0400375 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376out:
377 return err;
378}
379
380#if BITS_PER_LONG == 32
Heiko Carstensa26eab22009-01-14 14:14:17 +0100381SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
382 unsigned long, arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383{
Al Viro2903ff02012-08-28 12:52:22 -0400384 struct fd f = fdget_raw(fd);
Al Viro545ec2c2012-04-21 18:42:19 -0400385 long err = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386
Al Viro2903ff02012-08-28 12:52:22 -0400387 if (!f.file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 goto out;
389
Al Viro2903ff02012-08-28 12:52:22 -0400390 if (unlikely(f.file->f_mode & FMODE_PATH)) {
Al Viro545ec2c2012-04-21 18:42:19 -0400391 if (!check_fcntl_cmd(cmd))
392 goto out1;
Al Viro1abf0c72011-03-13 03:51:11 -0400393 }
394
Al Viro2903ff02012-08-28 12:52:22 -0400395 err = security_file_fcntl(f.file, cmd, arg);
Al Viro545ec2c2012-04-21 18:42:19 -0400396 if (err)
397 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398
399 switch (cmd) {
Jeff Layton5d50ffd2014-02-03 12:13:10 -0500400 case F_GETLK64:
Jeff Layton0d3f7a22014-04-22 08:23:58 -0400401 case F_OFD_GETLK:
Jeff Layton5d50ffd2014-02-03 12:13:10 -0500402 err = fcntl_getlk64(f.file, cmd, (struct flock64 __user *) arg);
403 break;
404 case F_SETLK64:
405 case F_SETLKW64:
Jeff Layton0d3f7a22014-04-22 08:23:58 -0400406 case F_OFD_SETLK:
407 case F_OFD_SETLKW:
Jeff Layton5d50ffd2014-02-03 12:13:10 -0500408 err = fcntl_setlk64(fd, f.file, cmd,
409 (struct flock64 __user *) arg);
410 break;
411 default:
412 err = do_fcntl(fd, cmd, arg, f.file);
413 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 }
Al Viro545ec2c2012-04-21 18:42:19 -0400415out1:
Al Viro2903ff02012-08-28 12:52:22 -0400416 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417out:
418 return err;
419}
420#endif
421
422/* Table to convert sigio signal codes into poll band bitmaps */
423
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800424static const long band_table[NSIGPOLL] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 POLLIN | POLLRDNORM, /* POLL_IN */
426 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
427 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
428 POLLERR, /* POLL_ERR */
429 POLLPRI | POLLRDBAND, /* POLL_PRI */
430 POLLHUP | POLLERR /* POLL_HUP */
431};
432
433static inline int sigio_perm(struct task_struct *p,
434 struct fown_struct *fown, int sig)
435{
David Howellsc69e8d92008-11-14 10:39:19 +1100436 const struct cred *cred;
437 int ret;
438
439 rcu_read_lock();
440 cred = __task_cred(p);
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -0800441 ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
442 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
443 uid_eq(fown->uid, cred->suid) || uid_eq(fown->uid, cred->uid)) &&
David Howellsc69e8d92008-11-14 10:39:19 +1100444 !security_file_send_sigiotask(p, fown, sig));
445 rcu_read_unlock();
446 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447}
448
449static void send_sigio_to_task(struct task_struct *p,
Oleg Nesterov8eeee4e2009-06-17 00:27:10 +0200450 struct fown_struct *fown,
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700451 int fd, int reason, int group)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452{
Oleg Nesterov8eeee4e2009-06-17 00:27:10 +0200453 /*
454 * F_SETSIG can change ->signum lockless in parallel, make
455 * sure we read it once and use the same value throughout.
456 */
457 int signum = ACCESS_ONCE(fown->signum);
458
459 if (!sigio_perm(p, fown, signum))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 return;
461
Oleg Nesterov8eeee4e2009-06-17 00:27:10 +0200462 switch (signum) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 siginfo_t si;
464 default:
465 /* Queue a rt signal with the appropriate fd as its
466 value. We use SI_SIGIO as the source, not
467 SI_KERNEL, since kernel signals always get
468 delivered even if we can't queue. Failure to
469 queue in this case _should_ be reported; we fall
470 back to SIGIO in that case. --sct */
Oleg Nesterov8eeee4e2009-06-17 00:27:10 +0200471 si.si_signo = signum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 si.si_errno = 0;
473 si.si_code = reason;
474 /* Make sure we are called with one of the POLL_*
475 reasons, otherwise we could leak kernel stack into
476 userspace. */
Eric Sesterhennf6298aa2006-04-02 13:37:19 +0200477 BUG_ON((reason & __SI_MASK) != __SI_POLL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 if (reason - POLL_IN >= NSIGPOLL)
479 si.si_band = ~0L;
480 else
481 si.si_band = band_table[reason - POLL_IN];
482 si.si_fd = fd;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700483 if (!do_send_sig_info(signum, &si, p, group))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 break;
485 /* fall-through: fall back on the old plain SIGIO signal */
486 case 0:
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700487 do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 }
489}
490
491void send_sigio(struct fown_struct *fown, int fd, int band)
492{
493 struct task_struct *p;
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700494 enum pid_type type;
495 struct pid *pid;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700496 int group = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497
498 read_lock(&fown->lock);
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700499
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700500 type = fown->pid_type;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700501 if (type == PIDTYPE_MAX) {
502 group = 0;
503 type = PIDTYPE_PID;
504 }
505
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 pid = fown->pid;
507 if (!pid)
508 goto out_unlock_fown;
509
510 read_lock(&tasklist_lock);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700511 do_each_pid_task(pid, type, p) {
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700512 send_sigio_to_task(p, fown, fd, band, group);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700513 } while_each_pid_task(pid, type, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 read_unlock(&tasklist_lock);
515 out_unlock_fown:
516 read_unlock(&fown->lock);
517}
518
519static void send_sigurg_to_task(struct task_struct *p,
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700520 struct fown_struct *fown, int group)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521{
522 if (sigio_perm(p, fown, SIGURG))
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700523 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524}
525
526int send_sigurg(struct fown_struct *fown)
527{
528 struct task_struct *p;
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700529 enum pid_type type;
530 struct pid *pid;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700531 int group = 1;
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700532 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533
534 read_lock(&fown->lock);
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700535
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700536 type = fown->pid_type;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700537 if (type == PIDTYPE_MAX) {
538 group = 0;
539 type = PIDTYPE_PID;
540 }
541
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 pid = fown->pid;
543 if (!pid)
544 goto out_unlock_fown;
545
546 ret = 1;
547
548 read_lock(&tasklist_lock);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700549 do_each_pid_task(pid, type, p) {
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700550 send_sigurg_to_task(p, fown, group);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700551 } while_each_pid_task(pid, type, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 read_unlock(&tasklist_lock);
553 out_unlock_fown:
554 read_unlock(&fown->lock);
555 return ret;
556}
557
Eric Dumazet989a2972010-04-14 09:55:35 +0000558static DEFINE_SPINLOCK(fasync_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800559static struct kmem_cache *fasync_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560
Eric Dumazet989a2972010-04-14 09:55:35 +0000561static void fasync_free_rcu(struct rcu_head *head)
562{
563 kmem_cache_free(fasync_cache,
564 container_of(head, struct fasync_struct, fa_rcu));
565}
566
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567/*
Linus Torvalds53281b62009-12-16 08:23:37 -0800568 * Remove a fasync entry. If successfully removed, return
569 * positive and clear the FASYNC flag. If no entry exists,
570 * do nothing and return 0.
571 *
572 * NOTE! It is very important that the FASYNC flag always
573 * match the state "is the filp on a fasync list".
574 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 */
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400576int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577{
578 struct fasync_struct *fa, **fp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 int result = 0;
580
Jonathan Corbet4a6a4492009-03-27 12:24:31 -0600581 spin_lock(&filp->f_lock);
Eric Dumazet989a2972010-04-14 09:55:35 +0000582 spin_lock(&fasync_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
Linus Torvalds53281b62009-12-16 08:23:37 -0800584 if (fa->fa_file != filp)
585 continue;
Eric Dumazet989a2972010-04-14 09:55:35 +0000586
587 spin_lock_irq(&fa->fa_lock);
588 fa->fa_file = NULL;
589 spin_unlock_irq(&fa->fa_lock);
590
Linus Torvalds53281b62009-12-16 08:23:37 -0800591 *fp = fa->fa_next;
Eric Dumazet989a2972010-04-14 09:55:35 +0000592 call_rcu(&fa->fa_rcu, fasync_free_rcu);
Jonathan Corbet76398422009-02-01 14:26:59 -0700593 filp->f_flags &= ~FASYNC;
Linus Torvalds53281b62009-12-16 08:23:37 -0800594 result = 1;
595 break;
596 }
Eric Dumazet989a2972010-04-14 09:55:35 +0000597 spin_unlock(&fasync_lock);
Jonathan Corbet4a6a4492009-03-27 12:24:31 -0600598 spin_unlock(&filp->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 return result;
600}
601
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400602struct fasync_struct *fasync_alloc(void)
Linus Torvalds53281b62009-12-16 08:23:37 -0800603{
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400604 return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
605}
Linus Torvalds53281b62009-12-16 08:23:37 -0800606
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400607/*
608 * NOTE! This can be used only for unused fasync entries:
609 * entries that actually got inserted on the fasync list
610 * need to be released by rcu - see fasync_remove_entry.
611 */
612void fasync_free(struct fasync_struct *new)
613{
614 kmem_cache_free(fasync_cache, new);
615}
616
617/*
618 * Insert a new entry into the fasync list. Return the pointer to the
619 * old one if we didn't use the new one.
Linus Torvalds55f335a2010-10-27 18:17:02 -0700620 *
621 * NOTE! It is very important that the FASYNC flag always
622 * match the state "is the filp on a fasync list".
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400623 */
624struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
625{
626 struct fasync_struct *fa, **fp;
Linus Torvalds53281b62009-12-16 08:23:37 -0800627
628 spin_lock(&filp->f_lock);
Eric Dumazet989a2972010-04-14 09:55:35 +0000629 spin_lock(&fasync_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800630 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
631 if (fa->fa_file != filp)
632 continue;
Eric Dumazet989a2972010-04-14 09:55:35 +0000633
634 spin_lock_irq(&fa->fa_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800635 fa->fa_fd = fd;
Eric Dumazet989a2972010-04-14 09:55:35 +0000636 spin_unlock_irq(&fa->fa_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800637 goto out;
638 }
639
Eric Dumazet989a2972010-04-14 09:55:35 +0000640 spin_lock_init(&new->fa_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800641 new->magic = FASYNC_MAGIC;
642 new->fa_file = filp;
643 new->fa_fd = fd;
644 new->fa_next = *fapp;
Eric Dumazet989a2972010-04-14 09:55:35 +0000645 rcu_assign_pointer(*fapp, new);
Linus Torvalds53281b62009-12-16 08:23:37 -0800646 filp->f_flags |= FASYNC;
647
648out:
Eric Dumazet989a2972010-04-14 09:55:35 +0000649 spin_unlock(&fasync_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800650 spin_unlock(&filp->f_lock);
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400651 return fa;
652}
653
654/*
655 * Add a fasync entry. Return negative on error, positive if
656 * added, and zero if did nothing but change an existing one.
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400657 */
658static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
659{
660 struct fasync_struct *new;
661
662 new = fasync_alloc();
663 if (!new)
664 return -ENOMEM;
665
666 /*
667 * fasync_insert_entry() returns the old (update) entry if
668 * it existed.
669 *
670 * So free the (unused) new entry and return 0 to let the
671 * caller know that we didn't add any new fasync entries.
672 */
673 if (fasync_insert_entry(fd, filp, fapp, new)) {
674 fasync_free(new);
675 return 0;
676 }
677
678 return 1;
Linus Torvalds53281b62009-12-16 08:23:37 -0800679}
680
681/*
682 * fasync_helper() is used by almost all character device drivers
683 * to set up the fasync queue, and for regular files by the file
684 * lease code. It returns negative on error, 0 if it did no changes
685 * and positive if it added/deleted the entry.
686 */
687int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
688{
689 if (!on)
690 return fasync_remove_entry(filp, fapp);
691 return fasync_add_entry(fd, filp, fapp);
692}
693
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694EXPORT_SYMBOL(fasync_helper);
695
Eric Dumazet989a2972010-04-14 09:55:35 +0000696/*
697 * rcu_read_lock() is held
698 */
699static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700{
701 while (fa) {
Eric Dumazet989a2972010-04-14 09:55:35 +0000702 struct fown_struct *fown;
Andrew Mortonf4985dc2010-06-29 15:05:42 -0700703 unsigned long flags;
704
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 if (fa->magic != FASYNC_MAGIC) {
706 printk(KERN_ERR "kill_fasync: bad magic number in "
707 "fasync_struct!\n");
708 return;
709 }
Andrew Mortonf4985dc2010-06-29 15:05:42 -0700710 spin_lock_irqsave(&fa->fa_lock, flags);
Eric Dumazet989a2972010-04-14 09:55:35 +0000711 if (fa->fa_file) {
712 fown = &fa->fa_file->f_owner;
713 /* Don't send SIGURG to processes which have not set a
714 queued signum: SIGURG has its own default signalling
715 mechanism. */
716 if (!(sig == SIGURG && fown->signum == 0))
717 send_sigio(fown, fa->fa_fd, band);
718 }
Andrew Mortonf4985dc2010-06-29 15:05:42 -0700719 spin_unlock_irqrestore(&fa->fa_lock, flags);
Eric Dumazet989a2972010-04-14 09:55:35 +0000720 fa = rcu_dereference(fa->fa_next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 }
722}
723
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724void kill_fasync(struct fasync_struct **fp, int sig, int band)
725{
726 /* First a quick test without locking: usually
727 * the list is empty.
728 */
729 if (*fp) {
Eric Dumazet989a2972010-04-14 09:55:35 +0000730 rcu_read_lock();
731 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
732 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 }
734}
735EXPORT_SYMBOL(kill_fasync);
736
Wu Fengguang454eedb2010-08-10 18:01:29 -0700737static int __init fcntl_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738{
James Bottomley3ab04d52010-09-09 16:38:12 -0700739 /*
740 * Please add new bits here to ensure allocation uniqueness.
741 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
742 * is defined as O_NONBLOCK on some platforms and not on others.
743 */
David Drysdale75069f22015-01-08 14:32:29 -0800744 BUILD_BUG_ON(21 - 1 /* for O_RDONLY being 0 */ != HWEIGHT32(
Wu Fengguang454eedb2010-08-10 18:01:29 -0700745 O_RDONLY | O_WRONLY | O_RDWR |
746 O_CREAT | O_EXCL | O_NOCTTY |
James Bottomley3ab04d52010-09-09 16:38:12 -0700747 O_TRUNC | O_APPEND | /* O_NONBLOCK | */
Wu Fengguang454eedb2010-08-10 18:01:29 -0700748 __O_SYNC | O_DSYNC | FASYNC |
749 O_DIRECT | O_LARGEFILE | O_DIRECTORY |
750 O_NOFOLLOW | O_NOATIME | O_CLOEXEC |
David Drysdale75069f22015-01-08 14:32:29 -0800751 __FMODE_EXEC | O_PATH | __O_TMPFILE |
752 __FMODE_NONOTIFY
Wu Fengguang454eedb2010-08-10 18:01:29 -0700753 ));
754
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 fasync_cache = kmem_cache_create("fasync_cache",
Paul Mundt20c2df82007-07-20 10:11:58 +0900756 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 return 0;
758}
759
Wu Fengguang454eedb2010-08-10 18:01:29 -0700760module_init(fcntl_init)