blob: 71a600a19f06078f235d8dbba6388cee2594365a [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024
25#include <asm/poll.h>
26#include <asm/siginfo.h>
27#include <asm/uaccess.h>
28
Jonathan Corbet76398422009-02-01 14:26:59 -070029#define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
Linus Torvalds1da177e2005-04-16 15:20:36 -070030
31static int setfl(int fd, struct file * filp, unsigned long arg)
32{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -080033 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -070034 int error = 0;
35
dean gaudet7d95c8f2006-02-03 03:04:30 -080036 /*
37 * O_APPEND cannot be cleared if the file is marked as append-only
38 * and the file is open for write.
39 */
40 if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -070041 return -EPERM;
42
43 /* O_NOATIME can only be set by the owner or superuser */
44 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
Serge E. Hallyn2e149672011-03-23 16:43:26 -070045 if (!inode_owner_or_capable(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -070046 return -EPERM;
47
48 /* required for strict SunOS emulation */
49 if (O_NONBLOCK != O_NDELAY)
50 if (arg & O_NDELAY)
51 arg |= O_NONBLOCK;
52
53 if (arg & O_DIRECT) {
54 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
55 !filp->f_mapping->a_ops->direct_IO)
56 return -EINVAL;
57 }
58
59 if (filp->f_op && filp->f_op->check_flags)
60 error = filp->f_op->check_flags(arg);
61 if (error)
62 return error;
63
Jonathan Corbet218d11a2008-12-05 16:12:48 -070064 /*
Jonathan Corbet76398422009-02-01 14:26:59 -070065 * ->fasync() is responsible for setting the FASYNC bit.
Jonathan Corbet218d11a2008-12-05 16:12:48 -070066 */
Jonathan Corbet76398422009-02-01 14:26:59 -070067 if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op &&
68 filp->f_op->fasync) {
69 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;
276 case F_GETLK:
277 err = fcntl_getlk(filp, (struct flock __user *) arg);
278 break;
279 case F_SETLK:
280 case F_SETLKW:
Peter Staubachc2936212005-07-27 11:45:09 -0700281 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 break;
283 case F_GETOWN:
284 /*
285 * XXX If f_owner is a process group, the
286 * negative return value will get converted
287 * into an error. Oops. If we keep the
288 * current syscall conventions, the only way
289 * to fix this will be in libc.
290 */
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700291 err = f_getown(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 force_successful_syscall_return();
293 break;
294 case F_SETOWN:
295 err = f_setown(filp, arg, 1);
296 break;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700297 case F_GETOWN_EX:
298 err = f_getown_ex(filp, arg);
299 break;
300 case F_SETOWN_EX:
301 err = f_setown_ex(filp, arg);
302 break;
Cyrill Gorcunov1d151c32012-07-30 14:43:00 -0700303 case F_GETOWNER_UIDS:
304 err = f_getowner_uids(filp, arg);
305 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 case F_GETSIG:
307 err = filp->f_owner.signum;
308 break;
309 case F_SETSIG:
310 /* arg == 0 restores default behaviour. */
Jesper Juhl7ed20e12005-05-01 08:59:14 -0700311 if (!valid_signal(arg)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 break;
313 }
314 err = 0;
315 filp->f_owner.signum = arg;
316 break;
317 case F_GETLEASE:
318 err = fcntl_getlease(filp);
319 break;
320 case F_SETLEASE:
321 err = fcntl_setlease(fd, filp, arg);
322 break;
323 case F_NOTIFY:
324 err = fcntl_dirnotify(fd, filp, arg);
325 break;
Jens Axboe35f3d142010-05-20 10:43:18 +0200326 case F_SETPIPE_SZ:
327 case F_GETPIPE_SZ:
328 err = pipe_fcntl(filp, cmd, arg);
329 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 default:
331 break;
332 }
333 return err;
334}
335
Al Viro1abf0c72011-03-13 03:51:11 -0400336static int check_fcntl_cmd(unsigned cmd)
337{
338 switch (cmd) {
339 case F_DUPFD:
340 case F_DUPFD_CLOEXEC:
341 case F_GETFD:
342 case F_SETFD:
343 case F_GETFL:
344 return 1;
345 }
346 return 0;
347}
348
Heiko Carstensa26eab22009-01-14 14:14:17 +0100349SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350{
Al Viro2903ff02012-08-28 12:52:22 -0400351 struct fd f = fdget_raw(fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 long err = -EBADF;
353
Al Viro2903ff02012-08-28 12:52:22 -0400354 if (!f.file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 goto out;
356
Al Viro2903ff02012-08-28 12:52:22 -0400357 if (unlikely(f.file->f_mode & FMODE_PATH)) {
Al Viro545ec2c2012-04-21 18:42:19 -0400358 if (!check_fcntl_cmd(cmd))
359 goto out1;
Al Viro1abf0c72011-03-13 03:51:11 -0400360 }
361
Al Viro2903ff02012-08-28 12:52:22 -0400362 err = security_file_fcntl(f.file, cmd, arg);
Al Viro545ec2c2012-04-21 18:42:19 -0400363 if (!err)
Al Viro2903ff02012-08-28 12:52:22 -0400364 err = do_fcntl(fd, cmd, arg, f.file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365
Al Viro545ec2c2012-04-21 18:42:19 -0400366out1:
Al Viro2903ff02012-08-28 12:52:22 -0400367 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368out:
369 return err;
370}
371
372#if BITS_PER_LONG == 32
Heiko Carstensa26eab22009-01-14 14:14:17 +0100373SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
374 unsigned long, arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375{
Al Viro2903ff02012-08-28 12:52:22 -0400376 struct fd f = fdget_raw(fd);
Al Viro545ec2c2012-04-21 18:42:19 -0400377 long err = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378
Al Viro2903ff02012-08-28 12:52:22 -0400379 if (!f.file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 goto out;
381
Al Viro2903ff02012-08-28 12:52:22 -0400382 if (unlikely(f.file->f_mode & FMODE_PATH)) {
Al Viro545ec2c2012-04-21 18:42:19 -0400383 if (!check_fcntl_cmd(cmd))
384 goto out1;
Al Viro1abf0c72011-03-13 03:51:11 -0400385 }
386
Al Viro2903ff02012-08-28 12:52:22 -0400387 err = security_file_fcntl(f.file, cmd, arg);
Al Viro545ec2c2012-04-21 18:42:19 -0400388 if (err)
389 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390
391 switch (cmd) {
392 case F_GETLK64:
Al Viro2903ff02012-08-28 12:52:22 -0400393 err = fcntl_getlk64(f.file, (struct flock64 __user *) arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 break;
395 case F_SETLK64:
396 case F_SETLKW64:
Al Viro2903ff02012-08-28 12:52:22 -0400397 err = fcntl_setlk64(fd, f.file, cmd,
Peter Staubachc2936212005-07-27 11:45:09 -0700398 (struct flock64 __user *) arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 break;
400 default:
Al Viro2903ff02012-08-28 12:52:22 -0400401 err = do_fcntl(fd, cmd, arg, f.file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 break;
403 }
Al Viro545ec2c2012-04-21 18:42:19 -0400404out1:
Al Viro2903ff02012-08-28 12:52:22 -0400405 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406out:
407 return err;
408}
409#endif
410
411/* Table to convert sigio signal codes into poll band bitmaps */
412
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800413static const long band_table[NSIGPOLL] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 POLLIN | POLLRDNORM, /* POLL_IN */
415 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
416 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
417 POLLERR, /* POLL_ERR */
418 POLLPRI | POLLRDBAND, /* POLL_PRI */
419 POLLHUP | POLLERR /* POLL_HUP */
420};
421
422static inline int sigio_perm(struct task_struct *p,
423 struct fown_struct *fown, int sig)
424{
David Howellsc69e8d92008-11-14 10:39:19 +1100425 const struct cred *cred;
426 int ret;
427
428 rcu_read_lock();
429 cred = __task_cred(p);
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -0800430 ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
431 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
432 uid_eq(fown->uid, cred->suid) || uid_eq(fown->uid, cred->uid)) &&
David Howellsc69e8d92008-11-14 10:39:19 +1100433 !security_file_send_sigiotask(p, fown, sig));
434 rcu_read_unlock();
435 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436}
437
438static void send_sigio_to_task(struct task_struct *p,
Oleg Nesterov8eeee4e2009-06-17 00:27:10 +0200439 struct fown_struct *fown,
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700440 int fd, int reason, int group)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441{
Oleg Nesterov8eeee4e2009-06-17 00:27:10 +0200442 /*
443 * F_SETSIG can change ->signum lockless in parallel, make
444 * sure we read it once and use the same value throughout.
445 */
446 int signum = ACCESS_ONCE(fown->signum);
447
448 if (!sigio_perm(p, fown, signum))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 return;
450
Oleg Nesterov8eeee4e2009-06-17 00:27:10 +0200451 switch (signum) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 siginfo_t si;
453 default:
454 /* Queue a rt signal with the appropriate fd as its
455 value. We use SI_SIGIO as the source, not
456 SI_KERNEL, since kernel signals always get
457 delivered even if we can't queue. Failure to
458 queue in this case _should_ be reported; we fall
459 back to SIGIO in that case. --sct */
Oleg Nesterov8eeee4e2009-06-17 00:27:10 +0200460 si.si_signo = signum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 si.si_errno = 0;
462 si.si_code = reason;
463 /* Make sure we are called with one of the POLL_*
464 reasons, otherwise we could leak kernel stack into
465 userspace. */
Eric Sesterhennf6298aa2006-04-02 13:37:19 +0200466 BUG_ON((reason & __SI_MASK) != __SI_POLL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 if (reason - POLL_IN >= NSIGPOLL)
468 si.si_band = ~0L;
469 else
470 si.si_band = band_table[reason - POLL_IN];
471 si.si_fd = fd;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700472 if (!do_send_sig_info(signum, &si, p, group))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 break;
474 /* fall-through: fall back on the old plain SIGIO signal */
475 case 0:
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700476 do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 }
478}
479
480void send_sigio(struct fown_struct *fown, int fd, int band)
481{
482 struct task_struct *p;
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700483 enum pid_type type;
484 struct pid *pid;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700485 int group = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486
487 read_lock(&fown->lock);
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700488
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700489 type = fown->pid_type;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700490 if (type == PIDTYPE_MAX) {
491 group = 0;
492 type = PIDTYPE_PID;
493 }
494
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 pid = fown->pid;
496 if (!pid)
497 goto out_unlock_fown;
498
499 read_lock(&tasklist_lock);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700500 do_each_pid_task(pid, type, p) {
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700501 send_sigio_to_task(p, fown, fd, band, group);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700502 } while_each_pid_task(pid, type, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 read_unlock(&tasklist_lock);
504 out_unlock_fown:
505 read_unlock(&fown->lock);
506}
507
508static void send_sigurg_to_task(struct task_struct *p,
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700509 struct fown_struct *fown, int group)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510{
511 if (sigio_perm(p, fown, SIGURG))
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700512 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513}
514
515int send_sigurg(struct fown_struct *fown)
516{
517 struct task_struct *p;
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700518 enum pid_type type;
519 struct pid *pid;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700520 int group = 1;
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700521 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522
523 read_lock(&fown->lock);
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700524
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700525 type = fown->pid_type;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700526 if (type == PIDTYPE_MAX) {
527 group = 0;
528 type = PIDTYPE_PID;
529 }
530
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 pid = fown->pid;
532 if (!pid)
533 goto out_unlock_fown;
534
535 ret = 1;
536
537 read_lock(&tasklist_lock);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700538 do_each_pid_task(pid, type, p) {
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700539 send_sigurg_to_task(p, fown, group);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700540 } while_each_pid_task(pid, type, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 read_unlock(&tasklist_lock);
542 out_unlock_fown:
543 read_unlock(&fown->lock);
544 return ret;
545}
546
Eric Dumazet989a2972010-04-14 09:55:35 +0000547static DEFINE_SPINLOCK(fasync_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800548static struct kmem_cache *fasync_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549
Eric Dumazet989a2972010-04-14 09:55:35 +0000550static void fasync_free_rcu(struct rcu_head *head)
551{
552 kmem_cache_free(fasync_cache,
553 container_of(head, struct fasync_struct, fa_rcu));
554}
555
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556/*
Linus Torvalds53281b62009-12-16 08:23:37 -0800557 * Remove a fasync entry. If successfully removed, return
558 * positive and clear the FASYNC flag. If no entry exists,
559 * do nothing and return 0.
560 *
561 * NOTE! It is very important that the FASYNC flag always
562 * match the state "is the filp on a fasync list".
563 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 */
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400565int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566{
567 struct fasync_struct *fa, **fp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 int result = 0;
569
Jonathan Corbet4a6a4492009-03-27 12:24:31 -0600570 spin_lock(&filp->f_lock);
Eric Dumazet989a2972010-04-14 09:55:35 +0000571 spin_lock(&fasync_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
Linus Torvalds53281b62009-12-16 08:23:37 -0800573 if (fa->fa_file != filp)
574 continue;
Eric Dumazet989a2972010-04-14 09:55:35 +0000575
576 spin_lock_irq(&fa->fa_lock);
577 fa->fa_file = NULL;
578 spin_unlock_irq(&fa->fa_lock);
579
Linus Torvalds53281b62009-12-16 08:23:37 -0800580 *fp = fa->fa_next;
Eric Dumazet989a2972010-04-14 09:55:35 +0000581 call_rcu(&fa->fa_rcu, fasync_free_rcu);
Jonathan Corbet76398422009-02-01 14:26:59 -0700582 filp->f_flags &= ~FASYNC;
Linus Torvalds53281b62009-12-16 08:23:37 -0800583 result = 1;
584 break;
585 }
Eric Dumazet989a2972010-04-14 09:55:35 +0000586 spin_unlock(&fasync_lock);
Jonathan Corbet4a6a4492009-03-27 12:24:31 -0600587 spin_unlock(&filp->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 return result;
589}
590
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400591struct fasync_struct *fasync_alloc(void)
Linus Torvalds53281b62009-12-16 08:23:37 -0800592{
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400593 return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
594}
Linus Torvalds53281b62009-12-16 08:23:37 -0800595
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400596/*
597 * NOTE! This can be used only for unused fasync entries:
598 * entries that actually got inserted on the fasync list
599 * need to be released by rcu - see fasync_remove_entry.
600 */
601void fasync_free(struct fasync_struct *new)
602{
603 kmem_cache_free(fasync_cache, new);
604}
605
606/*
607 * Insert a new entry into the fasync list. Return the pointer to the
608 * old one if we didn't use the new one.
Linus Torvalds55f335a2010-10-27 18:17:02 -0700609 *
610 * NOTE! It is very important that the FASYNC flag always
611 * match the state "is the filp on a fasync list".
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400612 */
613struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
614{
615 struct fasync_struct *fa, **fp;
Linus Torvalds53281b62009-12-16 08:23:37 -0800616
617 spin_lock(&filp->f_lock);
Eric Dumazet989a2972010-04-14 09:55:35 +0000618 spin_lock(&fasync_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800619 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
620 if (fa->fa_file != filp)
621 continue;
Eric Dumazet989a2972010-04-14 09:55:35 +0000622
623 spin_lock_irq(&fa->fa_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800624 fa->fa_fd = fd;
Eric Dumazet989a2972010-04-14 09:55:35 +0000625 spin_unlock_irq(&fa->fa_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800626 goto out;
627 }
628
Eric Dumazet989a2972010-04-14 09:55:35 +0000629 spin_lock_init(&new->fa_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800630 new->magic = FASYNC_MAGIC;
631 new->fa_file = filp;
632 new->fa_fd = fd;
633 new->fa_next = *fapp;
Eric Dumazet989a2972010-04-14 09:55:35 +0000634 rcu_assign_pointer(*fapp, new);
Linus Torvalds53281b62009-12-16 08:23:37 -0800635 filp->f_flags |= FASYNC;
636
637out:
Eric Dumazet989a2972010-04-14 09:55:35 +0000638 spin_unlock(&fasync_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800639 spin_unlock(&filp->f_lock);
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400640 return fa;
641}
642
643/*
644 * Add a fasync entry. Return negative on error, positive if
645 * added, and zero if did nothing but change an existing one.
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400646 */
647static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
648{
649 struct fasync_struct *new;
650
651 new = fasync_alloc();
652 if (!new)
653 return -ENOMEM;
654
655 /*
656 * fasync_insert_entry() returns the old (update) entry if
657 * it existed.
658 *
659 * So free the (unused) new entry and return 0 to let the
660 * caller know that we didn't add any new fasync entries.
661 */
662 if (fasync_insert_entry(fd, filp, fapp, new)) {
663 fasync_free(new);
664 return 0;
665 }
666
667 return 1;
Linus Torvalds53281b62009-12-16 08:23:37 -0800668}
669
670/*
671 * fasync_helper() is used by almost all character device drivers
672 * to set up the fasync queue, and for regular files by the file
673 * lease code. It returns negative on error, 0 if it did no changes
674 * and positive if it added/deleted the entry.
675 */
676int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
677{
678 if (!on)
679 return fasync_remove_entry(filp, fapp);
680 return fasync_add_entry(fd, filp, fapp);
681}
682
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683EXPORT_SYMBOL(fasync_helper);
684
Eric Dumazet989a2972010-04-14 09:55:35 +0000685/*
686 * rcu_read_lock() is held
687 */
688static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689{
690 while (fa) {
Eric Dumazet989a2972010-04-14 09:55:35 +0000691 struct fown_struct *fown;
Andrew Mortonf4985dc2010-06-29 15:05:42 -0700692 unsigned long flags;
693
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 if (fa->magic != FASYNC_MAGIC) {
695 printk(KERN_ERR "kill_fasync: bad magic number in "
696 "fasync_struct!\n");
697 return;
698 }
Andrew Mortonf4985dc2010-06-29 15:05:42 -0700699 spin_lock_irqsave(&fa->fa_lock, flags);
Eric Dumazet989a2972010-04-14 09:55:35 +0000700 if (fa->fa_file) {
701 fown = &fa->fa_file->f_owner;
702 /* Don't send SIGURG to processes which have not set a
703 queued signum: SIGURG has its own default signalling
704 mechanism. */
705 if (!(sig == SIGURG && fown->signum == 0))
706 send_sigio(fown, fa->fa_fd, band);
707 }
Andrew Mortonf4985dc2010-06-29 15:05:42 -0700708 spin_unlock_irqrestore(&fa->fa_lock, flags);
Eric Dumazet989a2972010-04-14 09:55:35 +0000709 fa = rcu_dereference(fa->fa_next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 }
711}
712
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713void kill_fasync(struct fasync_struct **fp, int sig, int band)
714{
715 /* First a quick test without locking: usually
716 * the list is empty.
717 */
718 if (*fp) {
Eric Dumazet989a2972010-04-14 09:55:35 +0000719 rcu_read_lock();
720 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
721 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 }
723}
724EXPORT_SYMBOL(kill_fasync);
725
Wu Fengguang454eedb2010-08-10 18:01:29 -0700726static int __init fcntl_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727{
James Bottomley3ab04d52010-09-09 16:38:12 -0700728 /*
729 * Please add new bits here to ensure allocation uniqueness.
730 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
731 * is defined as O_NONBLOCK on some platforms and not on others.
732 */
Al Viro1abf0c72011-03-13 03:51:11 -0400733 BUILD_BUG_ON(19 - 1 /* for O_RDONLY being 0 */ != HWEIGHT32(
Wu Fengguang454eedb2010-08-10 18:01:29 -0700734 O_RDONLY | O_WRONLY | O_RDWR |
735 O_CREAT | O_EXCL | O_NOCTTY |
James Bottomley3ab04d52010-09-09 16:38:12 -0700736 O_TRUNC | O_APPEND | /* O_NONBLOCK | */
Wu Fengguang454eedb2010-08-10 18:01:29 -0700737 __O_SYNC | O_DSYNC | FASYNC |
738 O_DIRECT | O_LARGEFILE | O_DIRECTORY |
739 O_NOFOLLOW | O_NOATIME | O_CLOEXEC |
Al Viro1abf0c72011-03-13 03:51:11 -0400740 __FMODE_EXEC | O_PATH
Wu Fengguang454eedb2010-08-10 18:01:29 -0700741 ));
742
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 fasync_cache = kmem_cache_create("fasync_cache",
Paul Mundt20c2df82007-07-20 10:11:58 +0900744 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 return 0;
746}
747
Wu Fengguang454eedb2010-08-10 18:01:29 -0700748module_init(fcntl_init)