blob: ee85cd4e136abbff33409fb018343028d21578e2 [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
Jeff Laytone0b93ed2014-08-22 11:27:32 -0400101void __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700102 int force)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103{
Jeff Laytone0b93ed2014-08-22 11:27:32 -0400104 security_file_set_fowner(filp);
Oleg Nesterov2f38d702009-06-16 22:07:46 +0200105 f_modown(filp, pid, type, force);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106}
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700107EXPORT_SYMBOL(__f_setown);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108
Jeff Laytone0b93ed2014-08-22 11:27:32 -0400109void f_setown(struct file *filp, unsigned long arg, int force)
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700110{
111 enum pid_type type;
112 struct pid *pid;
113 int who = arg;
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700114 type = PIDTYPE_PID;
115 if (who < 0) {
116 type = PIDTYPE_PGID;
117 who = -who;
118 }
119 rcu_read_lock();
Pavel Emelyanovb4888932007-10-18 23:40:14 -0700120 pid = find_vpid(who);
Jeff Laytone0b93ed2014-08-22 11:27:32 -0400121 __f_setown(filp, pid, type, force);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700122 rcu_read_unlock();
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700123}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124EXPORT_SYMBOL(f_setown);
125
126void f_delown(struct file *filp)
127{
Oleg Nesterov2f38d702009-06-16 22:07:46 +0200128 f_modown(filp, NULL, PIDTYPE_PID, 1);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700129}
130
131pid_t f_getown(struct file *filp)
132{
133 pid_t pid;
Eric W. Biederman43fa1ad2006-10-02 02:17:27 -0700134 read_lock(&filp->f_owner.lock);
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -0800135 pid = pid_vnr(filp->f_owner.pid);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700136 if (filp->f_owner.pid_type == PIDTYPE_PGID)
137 pid = -pid;
Eric W. Biederman43fa1ad2006-10-02 02:17:27 -0700138 read_unlock(&filp->f_owner.lock);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700139 return pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140}
141
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700142static int f_setown_ex(struct file *filp, unsigned long arg)
143{
Al Viro63784dd2012-09-26 21:43:05 -0400144 struct f_owner_ex __user *owner_p = (void __user *)arg;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700145 struct f_owner_ex owner;
146 struct pid *pid;
147 int type;
148 int ret;
149
150 ret = copy_from_user(&owner, owner_p, sizeof(owner));
151 if (ret)
Dan Carpenter5b544702010-06-03 12:35:42 +0200152 return -EFAULT;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700153
154 switch (owner.type) {
155 case F_OWNER_TID:
156 type = PIDTYPE_MAX;
157 break;
158
159 case F_OWNER_PID:
160 type = PIDTYPE_PID;
161 break;
162
Peter Zijlstra978b4052009-11-17 14:06:24 -0800163 case F_OWNER_PGRP:
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700164 type = PIDTYPE_PGID;
165 break;
166
167 default:
168 return -EINVAL;
169 }
170
171 rcu_read_lock();
172 pid = find_vpid(owner.pid);
173 if (owner.pid && !pid)
174 ret = -ESRCH;
175 else
Jeff Laytone0b93ed2014-08-22 11:27:32 -0400176 __f_setown(filp, pid, type, 1);
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700177 rcu_read_unlock();
178
179 return ret;
180}
181
182static int f_getown_ex(struct file *filp, unsigned long arg)
183{
Al Viro63784dd2012-09-26 21:43:05 -0400184 struct f_owner_ex __user *owner_p = (void __user *)arg;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700185 struct f_owner_ex owner;
186 int ret = 0;
187
188 read_lock(&filp->f_owner.lock);
189 owner.pid = pid_vnr(filp->f_owner.pid);
190 switch (filp->f_owner.pid_type) {
191 case PIDTYPE_MAX:
192 owner.type = F_OWNER_TID;
193 break;
194
195 case PIDTYPE_PID:
196 owner.type = F_OWNER_PID;
197 break;
198
199 case PIDTYPE_PGID:
Peter Zijlstra978b4052009-11-17 14:06:24 -0800200 owner.type = F_OWNER_PGRP;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700201 break;
202
203 default:
204 WARN_ON(1);
205 ret = -EINVAL;
206 break;
207 }
208 read_unlock(&filp->f_owner.lock);
209
Dan Carpenter5b544702010-06-03 12:35:42 +0200210 if (!ret) {
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700211 ret = copy_to_user(owner_p, &owner, sizeof(owner));
Dan Carpenter5b544702010-06-03 12:35:42 +0200212 if (ret)
213 ret = -EFAULT;
214 }
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700215 return ret;
216}
217
Cyrill Gorcunov1d151c32012-07-30 14:43:00 -0700218#ifdef CONFIG_CHECKPOINT_RESTORE
219static int f_getowner_uids(struct file *filp, unsigned long arg)
220{
221 struct user_namespace *user_ns = current_user_ns();
Al Viro63784dd2012-09-26 21:43:05 -0400222 uid_t __user *dst = (void __user *)arg;
Cyrill Gorcunov1d151c32012-07-30 14:43:00 -0700223 uid_t src[2];
224 int err;
225
226 read_lock(&filp->f_owner.lock);
227 src[0] = from_kuid(user_ns, filp->f_owner.uid);
228 src[1] = from_kuid(user_ns, filp->f_owner.euid);
229 read_unlock(&filp->f_owner.lock);
230
231 err = put_user(src[0], &dst[0]);
232 err |= put_user(src[1], &dst[1]);
233
234 return err;
235}
236#else
237static int f_getowner_uids(struct file *filp, unsigned long arg)
238{
239 return -EINVAL;
240}
241#endif
242
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
244 struct file *filp)
245{
246 long err = -EINVAL;
247
248 switch (cmd) {
249 case F_DUPFD:
Al Virofe17f222012-08-21 11:48:11 -0400250 err = f_dupfd(arg, filp, 0);
251 break;
Ulrich Drepper22d2b352007-10-16 23:30:26 -0700252 case F_DUPFD_CLOEXEC:
Al Viro12197712012-10-08 23:21:58 +0100253 err = f_dupfd(arg, filp, O_CLOEXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 break;
255 case F_GETFD:
256 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
257 break;
258 case F_SETFD:
259 err = 0;
260 set_close_on_exec(fd, arg & FD_CLOEXEC);
261 break;
262 case F_GETFL:
263 err = filp->f_flags;
264 break;
265 case F_SETFL:
266 err = setfl(fd, filp, arg);
267 break;
Jeff Layton5d50ffd2014-02-03 12:13:10 -0500268#if BITS_PER_LONG != 32
269 /* 32-bit arches must use fcntl64() */
Jeff Layton0d3f7a22014-04-22 08:23:58 -0400270 case F_OFD_GETLK:
Jeff Layton5d50ffd2014-02-03 12:13:10 -0500271#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 case F_GETLK:
Jeff Laytonc1e62b82014-02-03 12:13:09 -0500273 err = fcntl_getlk(filp, cmd, (struct flock __user *) arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 break;
Jeff Layton5d50ffd2014-02-03 12:13:10 -0500275#if BITS_PER_LONG != 32
276 /* 32-bit arches must use fcntl64() */
Jeff Layton0d3f7a22014-04-22 08:23:58 -0400277 case F_OFD_SETLK:
278 case F_OFD_SETLKW:
Jeff Layton5d50ffd2014-02-03 12:13:10 -0500279#endif
280 /* Fallthrough */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 case F_SETLK:
282 case F_SETLKW:
Peter Staubachc2936212005-07-27 11:45:09 -0700283 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 break;
285 case F_GETOWN:
286 /*
287 * XXX If f_owner is a process group, the
288 * negative return value will get converted
289 * into an error. Oops. If we keep the
290 * current syscall conventions, the only way
291 * to fix this will be in libc.
292 */
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700293 err = f_getown(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 force_successful_syscall_return();
295 break;
296 case F_SETOWN:
Jeff Laytone0b93ed2014-08-22 11:27:32 -0400297 f_setown(filp, arg, 1);
298 err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 break;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700300 case F_GETOWN_EX:
301 err = f_getown_ex(filp, arg);
302 break;
303 case F_SETOWN_EX:
304 err = f_setown_ex(filp, arg);
305 break;
Cyrill Gorcunov1d151c32012-07-30 14:43:00 -0700306 case F_GETOWNER_UIDS:
307 err = f_getowner_uids(filp, arg);
308 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 case F_GETSIG:
310 err = filp->f_owner.signum;
311 break;
312 case F_SETSIG:
313 /* arg == 0 restores default behaviour. */
Jesper Juhl7ed20e12005-05-01 08:59:14 -0700314 if (!valid_signal(arg)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 break;
316 }
317 err = 0;
318 filp->f_owner.signum = arg;
319 break;
320 case F_GETLEASE:
321 err = fcntl_getlease(filp);
322 break;
323 case F_SETLEASE:
324 err = fcntl_setlease(fd, filp, arg);
325 break;
326 case F_NOTIFY:
327 err = fcntl_dirnotify(fd, filp, arg);
328 break;
Jens Axboe35f3d142010-05-20 10:43:18 +0200329 case F_SETPIPE_SZ:
330 case F_GETPIPE_SZ:
331 err = pipe_fcntl(filp, cmd, arg);
332 break;
David Herrmann40e041a2014-08-08 14:25:27 -0700333 case F_ADD_SEALS:
334 case F_GET_SEALS:
335 err = shmem_fcntl(filp, cmd, arg);
336 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 default:
338 break;
339 }
340 return err;
341}
342
Al Viro1abf0c72011-03-13 03:51:11 -0400343static int check_fcntl_cmd(unsigned cmd)
344{
345 switch (cmd) {
346 case F_DUPFD:
347 case F_DUPFD_CLOEXEC:
348 case F_GETFD:
349 case F_SETFD:
350 case F_GETFL:
351 return 1;
352 }
353 return 0;
354}
355
Heiko Carstensa26eab22009-01-14 14:14:17 +0100356SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357{
Al Viro2903ff02012-08-28 12:52:22 -0400358 struct fd f = fdget_raw(fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 long err = -EBADF;
360
Al Viro2903ff02012-08-28 12:52:22 -0400361 if (!f.file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 goto out;
363
Al Viro2903ff02012-08-28 12:52:22 -0400364 if (unlikely(f.file->f_mode & FMODE_PATH)) {
Al Viro545ec2c2012-04-21 18:42:19 -0400365 if (!check_fcntl_cmd(cmd))
366 goto out1;
Al Viro1abf0c72011-03-13 03:51:11 -0400367 }
368
Al Viro2903ff02012-08-28 12:52:22 -0400369 err = security_file_fcntl(f.file, cmd, arg);
Al Viro545ec2c2012-04-21 18:42:19 -0400370 if (!err)
Al Viro2903ff02012-08-28 12:52:22 -0400371 err = do_fcntl(fd, cmd, arg, f.file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372
Al Viro545ec2c2012-04-21 18:42:19 -0400373out1:
Al Viro2903ff02012-08-28 12:52:22 -0400374 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375out:
376 return err;
377}
378
379#if BITS_PER_LONG == 32
Heiko Carstensa26eab22009-01-14 14:14:17 +0100380SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
381 unsigned long, arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382{
Al Viro2903ff02012-08-28 12:52:22 -0400383 struct fd f = fdget_raw(fd);
Al Viro545ec2c2012-04-21 18:42:19 -0400384 long err = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385
Al Viro2903ff02012-08-28 12:52:22 -0400386 if (!f.file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 goto out;
388
Al Viro2903ff02012-08-28 12:52:22 -0400389 if (unlikely(f.file->f_mode & FMODE_PATH)) {
Al Viro545ec2c2012-04-21 18:42:19 -0400390 if (!check_fcntl_cmd(cmd))
391 goto out1;
Al Viro1abf0c72011-03-13 03:51:11 -0400392 }
393
Al Viro2903ff02012-08-28 12:52:22 -0400394 err = security_file_fcntl(f.file, cmd, arg);
Al Viro545ec2c2012-04-21 18:42:19 -0400395 if (err)
396 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397
398 switch (cmd) {
Jeff Layton5d50ffd2014-02-03 12:13:10 -0500399 case F_GETLK64:
Jeff Layton0d3f7a22014-04-22 08:23:58 -0400400 case F_OFD_GETLK:
Jeff Layton5d50ffd2014-02-03 12:13:10 -0500401 err = fcntl_getlk64(f.file, cmd, (struct flock64 __user *) arg);
402 break;
403 case F_SETLK64:
404 case F_SETLKW64:
Jeff Layton0d3f7a22014-04-22 08:23:58 -0400405 case F_OFD_SETLK:
406 case F_OFD_SETLKW:
Jeff Layton5d50ffd2014-02-03 12:13:10 -0500407 err = fcntl_setlk64(fd, f.file, cmd,
408 (struct flock64 __user *) arg);
409 break;
410 default:
411 err = do_fcntl(fd, cmd, arg, f.file);
412 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 }
Al Viro545ec2c2012-04-21 18:42:19 -0400414out1:
Al Viro2903ff02012-08-28 12:52:22 -0400415 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416out:
417 return err;
418}
419#endif
420
421/* Table to convert sigio signal codes into poll band bitmaps */
422
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800423static const long band_table[NSIGPOLL] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 POLLIN | POLLRDNORM, /* POLL_IN */
425 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
426 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
427 POLLERR, /* POLL_ERR */
428 POLLPRI | POLLRDBAND, /* POLL_PRI */
429 POLLHUP | POLLERR /* POLL_HUP */
430};
431
432static inline int sigio_perm(struct task_struct *p,
433 struct fown_struct *fown, int sig)
434{
David Howellsc69e8d92008-11-14 10:39:19 +1100435 const struct cred *cred;
436 int ret;
437
438 rcu_read_lock();
439 cred = __task_cred(p);
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -0800440 ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
441 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
442 uid_eq(fown->uid, cred->suid) || uid_eq(fown->uid, cred->uid)) &&
David Howellsc69e8d92008-11-14 10:39:19 +1100443 !security_file_send_sigiotask(p, fown, sig));
444 rcu_read_unlock();
445 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446}
447
448static void send_sigio_to_task(struct task_struct *p,
Oleg Nesterov8eeee4e2009-06-17 00:27:10 +0200449 struct fown_struct *fown,
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700450 int fd, int reason, int group)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451{
Oleg Nesterov8eeee4e2009-06-17 00:27:10 +0200452 /*
453 * F_SETSIG can change ->signum lockless in parallel, make
454 * sure we read it once and use the same value throughout.
455 */
456 int signum = ACCESS_ONCE(fown->signum);
457
458 if (!sigio_perm(p, fown, signum))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 return;
460
Oleg Nesterov8eeee4e2009-06-17 00:27:10 +0200461 switch (signum) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 siginfo_t si;
463 default:
464 /* Queue a rt signal with the appropriate fd as its
465 value. We use SI_SIGIO as the source, not
466 SI_KERNEL, since kernel signals always get
467 delivered even if we can't queue. Failure to
468 queue in this case _should_ be reported; we fall
469 back to SIGIO in that case. --sct */
Oleg Nesterov8eeee4e2009-06-17 00:27:10 +0200470 si.si_signo = signum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 si.si_errno = 0;
472 si.si_code = reason;
473 /* Make sure we are called with one of the POLL_*
474 reasons, otherwise we could leak kernel stack into
475 userspace. */
Eric Sesterhennf6298aa2006-04-02 13:37:19 +0200476 BUG_ON((reason & __SI_MASK) != __SI_POLL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 if (reason - POLL_IN >= NSIGPOLL)
478 si.si_band = ~0L;
479 else
480 si.si_band = band_table[reason - POLL_IN];
481 si.si_fd = fd;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700482 if (!do_send_sig_info(signum, &si, p, group))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 break;
484 /* fall-through: fall back on the old plain SIGIO signal */
485 case 0:
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700486 do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 }
488}
489
490void send_sigio(struct fown_struct *fown, int fd, int band)
491{
492 struct task_struct *p;
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700493 enum pid_type type;
494 struct pid *pid;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700495 int group = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496
497 read_lock(&fown->lock);
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700498
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700499 type = fown->pid_type;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700500 if (type == PIDTYPE_MAX) {
501 group = 0;
502 type = PIDTYPE_PID;
503 }
504
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 pid = fown->pid;
506 if (!pid)
507 goto out_unlock_fown;
508
509 read_lock(&tasklist_lock);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700510 do_each_pid_task(pid, type, p) {
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700511 send_sigio_to_task(p, fown, fd, band, group);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700512 } while_each_pid_task(pid, type, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 read_unlock(&tasklist_lock);
514 out_unlock_fown:
515 read_unlock(&fown->lock);
516}
517
518static void send_sigurg_to_task(struct task_struct *p,
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700519 struct fown_struct *fown, int group)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520{
521 if (sigio_perm(p, fown, SIGURG))
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700522 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523}
524
525int send_sigurg(struct fown_struct *fown)
526{
527 struct task_struct *p;
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700528 enum pid_type type;
529 struct pid *pid;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700530 int group = 1;
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700531 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532
533 read_lock(&fown->lock);
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700534
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700535 type = fown->pid_type;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700536 if (type == PIDTYPE_MAX) {
537 group = 0;
538 type = PIDTYPE_PID;
539 }
540
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 pid = fown->pid;
542 if (!pid)
543 goto out_unlock_fown;
544
545 ret = 1;
546
547 read_lock(&tasklist_lock);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700548 do_each_pid_task(pid, type, p) {
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700549 send_sigurg_to_task(p, fown, group);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700550 } while_each_pid_task(pid, type, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 read_unlock(&tasklist_lock);
552 out_unlock_fown:
553 read_unlock(&fown->lock);
554 return ret;
555}
556
Eric Dumazet989a2972010-04-14 09:55:35 +0000557static DEFINE_SPINLOCK(fasync_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800558static struct kmem_cache *fasync_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559
Eric Dumazet989a2972010-04-14 09:55:35 +0000560static void fasync_free_rcu(struct rcu_head *head)
561{
562 kmem_cache_free(fasync_cache,
563 container_of(head, struct fasync_struct, fa_rcu));
564}
565
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566/*
Linus Torvalds53281b62009-12-16 08:23:37 -0800567 * Remove a fasync entry. If successfully removed, return
568 * positive and clear the FASYNC flag. If no entry exists,
569 * do nothing and return 0.
570 *
571 * NOTE! It is very important that the FASYNC flag always
572 * match the state "is the filp on a fasync list".
573 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 */
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400575int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576{
577 struct fasync_struct *fa, **fp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 int result = 0;
579
Jonathan Corbet4a6a4492009-03-27 12:24:31 -0600580 spin_lock(&filp->f_lock);
Eric Dumazet989a2972010-04-14 09:55:35 +0000581 spin_lock(&fasync_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
Linus Torvalds53281b62009-12-16 08:23:37 -0800583 if (fa->fa_file != filp)
584 continue;
Eric Dumazet989a2972010-04-14 09:55:35 +0000585
586 spin_lock_irq(&fa->fa_lock);
587 fa->fa_file = NULL;
588 spin_unlock_irq(&fa->fa_lock);
589
Linus Torvalds53281b62009-12-16 08:23:37 -0800590 *fp = fa->fa_next;
Eric Dumazet989a2972010-04-14 09:55:35 +0000591 call_rcu(&fa->fa_rcu, fasync_free_rcu);
Jonathan Corbet76398422009-02-01 14:26:59 -0700592 filp->f_flags &= ~FASYNC;
Linus Torvalds53281b62009-12-16 08:23:37 -0800593 result = 1;
594 break;
595 }
Eric Dumazet989a2972010-04-14 09:55:35 +0000596 spin_unlock(&fasync_lock);
Jonathan Corbet4a6a4492009-03-27 12:24:31 -0600597 spin_unlock(&filp->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 return result;
599}
600
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400601struct fasync_struct *fasync_alloc(void)
Linus Torvalds53281b62009-12-16 08:23:37 -0800602{
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400603 return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
604}
Linus Torvalds53281b62009-12-16 08:23:37 -0800605
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400606/*
607 * NOTE! This can be used only for unused fasync entries:
608 * entries that actually got inserted on the fasync list
609 * need to be released by rcu - see fasync_remove_entry.
610 */
611void fasync_free(struct fasync_struct *new)
612{
613 kmem_cache_free(fasync_cache, new);
614}
615
616/*
617 * Insert a new entry into the fasync list. Return the pointer to the
618 * old one if we didn't use the new one.
Linus Torvalds55f335a2010-10-27 18:17:02 -0700619 *
620 * NOTE! It is very important that the FASYNC flag always
621 * match the state "is the filp on a fasync list".
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400622 */
623struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
624{
625 struct fasync_struct *fa, **fp;
Linus Torvalds53281b62009-12-16 08:23:37 -0800626
627 spin_lock(&filp->f_lock);
Eric Dumazet989a2972010-04-14 09:55:35 +0000628 spin_lock(&fasync_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800629 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
630 if (fa->fa_file != filp)
631 continue;
Eric Dumazet989a2972010-04-14 09:55:35 +0000632
633 spin_lock_irq(&fa->fa_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800634 fa->fa_fd = fd;
Eric Dumazet989a2972010-04-14 09:55:35 +0000635 spin_unlock_irq(&fa->fa_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800636 goto out;
637 }
638
Eric Dumazet989a2972010-04-14 09:55:35 +0000639 spin_lock_init(&new->fa_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800640 new->magic = FASYNC_MAGIC;
641 new->fa_file = filp;
642 new->fa_fd = fd;
643 new->fa_next = *fapp;
Eric Dumazet989a2972010-04-14 09:55:35 +0000644 rcu_assign_pointer(*fapp, new);
Linus Torvalds53281b62009-12-16 08:23:37 -0800645 filp->f_flags |= FASYNC;
646
647out:
Eric Dumazet989a2972010-04-14 09:55:35 +0000648 spin_unlock(&fasync_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800649 spin_unlock(&filp->f_lock);
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400650 return fa;
651}
652
653/*
654 * Add a fasync entry. Return negative on error, positive if
655 * added, and zero if did nothing but change an existing one.
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400656 */
657static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
658{
659 struct fasync_struct *new;
660
661 new = fasync_alloc();
662 if (!new)
663 return -ENOMEM;
664
665 /*
666 * fasync_insert_entry() returns the old (update) entry if
667 * it existed.
668 *
669 * So free the (unused) new entry and return 0 to let the
670 * caller know that we didn't add any new fasync entries.
671 */
672 if (fasync_insert_entry(fd, filp, fapp, new)) {
673 fasync_free(new);
674 return 0;
675 }
676
677 return 1;
Linus Torvalds53281b62009-12-16 08:23:37 -0800678}
679
680/*
681 * fasync_helper() is used by almost all character device drivers
682 * to set up the fasync queue, and for regular files by the file
683 * lease code. It returns negative on error, 0 if it did no changes
684 * and positive if it added/deleted the entry.
685 */
686int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
687{
688 if (!on)
689 return fasync_remove_entry(filp, fapp);
690 return fasync_add_entry(fd, filp, fapp);
691}
692
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693EXPORT_SYMBOL(fasync_helper);
694
Eric Dumazet989a2972010-04-14 09:55:35 +0000695/*
696 * rcu_read_lock() is held
697 */
698static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699{
700 while (fa) {
Eric Dumazet989a2972010-04-14 09:55:35 +0000701 struct fown_struct *fown;
Andrew Mortonf4985dc2010-06-29 15:05:42 -0700702 unsigned long flags;
703
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 if (fa->magic != FASYNC_MAGIC) {
705 printk(KERN_ERR "kill_fasync: bad magic number in "
706 "fasync_struct!\n");
707 return;
708 }
Andrew Mortonf4985dc2010-06-29 15:05:42 -0700709 spin_lock_irqsave(&fa->fa_lock, flags);
Eric Dumazet989a2972010-04-14 09:55:35 +0000710 if (fa->fa_file) {
711 fown = &fa->fa_file->f_owner;
712 /* Don't send SIGURG to processes which have not set a
713 queued signum: SIGURG has its own default signalling
714 mechanism. */
715 if (!(sig == SIGURG && fown->signum == 0))
716 send_sigio(fown, fa->fa_fd, band);
717 }
Andrew Mortonf4985dc2010-06-29 15:05:42 -0700718 spin_unlock_irqrestore(&fa->fa_lock, flags);
Eric Dumazet989a2972010-04-14 09:55:35 +0000719 fa = rcu_dereference(fa->fa_next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 }
721}
722
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723void kill_fasync(struct fasync_struct **fp, int sig, int band)
724{
725 /* First a quick test without locking: usually
726 * the list is empty.
727 */
728 if (*fp) {
Eric Dumazet989a2972010-04-14 09:55:35 +0000729 rcu_read_lock();
730 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
731 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 }
733}
734EXPORT_SYMBOL(kill_fasync);
735
Wu Fengguang454eedb2010-08-10 18:01:29 -0700736static int __init fcntl_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737{
James Bottomley3ab04d52010-09-09 16:38:12 -0700738 /*
739 * Please add new bits here to ensure allocation uniqueness.
740 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
741 * is defined as O_NONBLOCK on some platforms and not on others.
742 */
David Drysdale75069f22015-01-08 14:32:29 -0800743 BUILD_BUG_ON(21 - 1 /* for O_RDONLY being 0 */ != HWEIGHT32(
Wu Fengguang454eedb2010-08-10 18:01:29 -0700744 O_RDONLY | O_WRONLY | O_RDWR |
745 O_CREAT | O_EXCL | O_NOCTTY |
James Bottomley3ab04d52010-09-09 16:38:12 -0700746 O_TRUNC | O_APPEND | /* O_NONBLOCK | */
Wu Fengguang454eedb2010-08-10 18:01:29 -0700747 __O_SYNC | O_DSYNC | FASYNC |
748 O_DIRECT | O_LARGEFILE | O_DIRECTORY |
749 O_NOFOLLOW | O_NOATIME | O_CLOEXEC |
David Drysdale75069f22015-01-08 14:32:29 -0800750 __FMODE_EXEC | O_PATH | __O_TMPFILE |
751 __FMODE_NONOTIFY
Wu Fengguang454eedb2010-08-10 18:01:29 -0700752 ));
753
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 fasync_cache = kmem_cache_create("fasync_cache",
Paul Mundt20c2df82007-07-20 10:11:58 +0900755 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 return 0;
757}
758
Wu Fengguang454eedb2010-08-10 18:01:29 -0700759module_init(fcntl_init)