blob: f4e7267d117fb83b69f6f5be2305fb8b375077c8 [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>
Ingo Molnar29930022017-02-08 18:51:36 +010010#include <linux/sched/task.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/fs.h>
12#include <linux/file.h>
Al Viro9f3acc32008-04-24 07:44:08 -040013#include <linux/fdtable.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080014#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/dnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/slab.h>
17#include <linux/module.h>
Jens Axboe35f3d142010-05-20 10:43:18 +020018#include <linux/pipe_fs_i.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/security.h>
20#include <linux/ptrace.h>
Jesper Juhl7ed20e12005-05-01 08:59:14 -070021#include <linux/signal.h>
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070022#include <linux/rcupdate.h>
Pavel Emelyanovb4888932007-10-18 23:40:14 -070023#include <linux/pid_namespace.h>
Cyrill Gorcunov1d151c32012-07-30 14:43:00 -070024#include <linux/user_namespace.h>
David Herrmann40e041a2014-08-08 14:25:27 -070025#include <linux/shmem_fs.h>
Al Viro80f0cce2017-04-08 18:10:56 -040026#include <linux/compat.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027
28#include <asm/poll.h>
29#include <asm/siginfo.h>
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080030#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
Jonathan Corbet76398422009-02-01 14:26:59 -070032#define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
34static int setfl(int fd, struct file * filp, unsigned long arg)
35{
Al Viro496ad9a2013-01-23 17:07:38 -050036 struct inode * inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -070037 int error = 0;
38
dean gaudet7d95c8f2006-02-03 03:04:30 -080039 /*
40 * O_APPEND cannot be cleared if the file is marked as append-only
41 * and the file is open for write.
42 */
43 if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -070044 return -EPERM;
45
46 /* O_NOATIME can only be set by the owner or superuser */
47 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
Serge E. Hallyn2e149672011-03-23 16:43:26 -070048 if (!inode_owner_or_capable(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -070049 return -EPERM;
50
51 /* required for strict SunOS emulation */
52 if (O_NONBLOCK != O_NDELAY)
53 if (arg & O_NDELAY)
54 arg |= O_NONBLOCK;
55
Stanislav Kinsburskiy0dbf5f22015-12-15 19:41:31 +040056 /* Pipe packetized mode is controlled by O_DIRECT flag */
Al Viro45063092016-12-04 18:24:56 -050057 if (!S_ISFIFO(inode->i_mode) && (arg & O_DIRECT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070058 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
59 !filp->f_mapping->a_ops->direct_IO)
60 return -EINVAL;
61 }
62
Al Viro72c2d532013-09-22 16:27:52 -040063 if (filp->f_op->check_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 error = filp->f_op->check_flags(arg);
65 if (error)
66 return error;
67
Jonathan Corbet218d11a2008-12-05 16:12:48 -070068 /*
Jonathan Corbet76398422009-02-01 14:26:59 -070069 * ->fasync() is responsible for setting the FASYNC bit.
Jonathan Corbet218d11a2008-12-05 16:12:48 -070070 */
Al Viro72c2d532013-09-22 16:27:52 -040071 if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op->fasync) {
Jonathan Corbet76398422009-02-01 14:26:59 -070072 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
73 if (error < 0)
74 goto out;
Jonathan Corbet60aa4922009-02-01 14:52:56 -070075 if (error > 0)
76 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 }
Jonathan Corbetdb1dd4d2009-02-06 15:25:24 -070078 spin_lock(&filp->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
Jonathan Corbetdb1dd4d2009-02-06 15:25:24 -070080 spin_unlock(&filp->f_lock);
Jonathan Corbet76398422009-02-01 14:26:59 -070081
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 return error;
84}
85
Eric W. Biederman609d7fa2006-10-02 02:17:15 -070086static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
Oleg Nesterov2f38d702009-06-16 22:07:46 +020087 int force)
Linus Torvalds1da177e2005-04-16 15:20:36 -070088{
Linus Torvalds80e1e822010-02-07 10:11:23 -080089 write_lock_irq(&filp->f_owner.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070090 if (force || !filp->f_owner.pid) {
Eric W. Biederman609d7fa2006-10-02 02:17:15 -070091 put_pid(filp->f_owner.pid);
92 filp->f_owner.pid = get_pid(pid);
93 filp->f_owner.pid_type = type;
Oleg Nesterov2f38d702009-06-16 22:07:46 +020094
95 if (pid) {
96 const struct cred *cred = current_cred();
97 filp->f_owner.uid = cred->uid;
98 filp->f_owner.euid = cred->euid;
99 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 }
Linus Torvalds80e1e822010-02-07 10:11:23 -0800101 write_unlock_irq(&filp->f_owner.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102}
103
Jeff Laytone0b93ed2014-08-22 11:27:32 -0400104void __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700105 int force)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106{
Jeff Laytone0b93ed2014-08-22 11:27:32 -0400107 security_file_set_fowner(filp);
Oleg Nesterov2f38d702009-06-16 22:07:46 +0200108 f_modown(filp, pid, type, force);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109}
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700110EXPORT_SYMBOL(__f_setown);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111
Jeff Laytone0b93ed2014-08-22 11:27:32 -0400112void f_setown(struct file *filp, unsigned long arg, int force)
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700113{
114 enum pid_type type;
115 struct pid *pid;
116 int who = arg;
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700117 type = PIDTYPE_PID;
118 if (who < 0) {
119 type = PIDTYPE_PGID;
120 who = -who;
121 }
122 rcu_read_lock();
Pavel Emelyanovb4888932007-10-18 23:40:14 -0700123 pid = find_vpid(who);
Jeff Laytone0b93ed2014-08-22 11:27:32 -0400124 __f_setown(filp, pid, type, force);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700125 rcu_read_unlock();
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700126}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127EXPORT_SYMBOL(f_setown);
128
129void f_delown(struct file *filp)
130{
Oleg Nesterov2f38d702009-06-16 22:07:46 +0200131 f_modown(filp, NULL, PIDTYPE_PID, 1);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700132}
133
134pid_t f_getown(struct file *filp)
135{
136 pid_t pid;
Eric W. Biederman43fa1ad2006-10-02 02:17:27 -0700137 read_lock(&filp->f_owner.lock);
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -0800138 pid = pid_vnr(filp->f_owner.pid);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700139 if (filp->f_owner.pid_type == PIDTYPE_PGID)
140 pid = -pid;
Eric W. Biederman43fa1ad2006-10-02 02:17:27 -0700141 read_unlock(&filp->f_owner.lock);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700142 return pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143}
144
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700145static int f_setown_ex(struct file *filp, unsigned long arg)
146{
Al Viro63784dd2012-09-26 21:43:05 -0400147 struct f_owner_ex __user *owner_p = (void __user *)arg;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700148 struct f_owner_ex owner;
149 struct pid *pid;
150 int type;
151 int ret;
152
153 ret = copy_from_user(&owner, owner_p, sizeof(owner));
154 if (ret)
Dan Carpenter5b544702010-06-03 12:35:42 +0200155 return -EFAULT;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700156
157 switch (owner.type) {
158 case F_OWNER_TID:
159 type = PIDTYPE_MAX;
160 break;
161
162 case F_OWNER_PID:
163 type = PIDTYPE_PID;
164 break;
165
Peter Zijlstra978b4052009-11-17 14:06:24 -0800166 case F_OWNER_PGRP:
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700167 type = PIDTYPE_PGID;
168 break;
169
170 default:
171 return -EINVAL;
172 }
173
174 rcu_read_lock();
175 pid = find_vpid(owner.pid);
176 if (owner.pid && !pid)
177 ret = -ESRCH;
178 else
Jeff Laytone0b93ed2014-08-22 11:27:32 -0400179 __f_setown(filp, pid, type, 1);
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700180 rcu_read_unlock();
181
182 return ret;
183}
184
185static int f_getown_ex(struct file *filp, unsigned long arg)
186{
Al Viro63784dd2012-09-26 21:43:05 -0400187 struct f_owner_ex __user *owner_p = (void __user *)arg;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700188 struct f_owner_ex owner;
189 int ret = 0;
190
191 read_lock(&filp->f_owner.lock);
192 owner.pid = pid_vnr(filp->f_owner.pid);
193 switch (filp->f_owner.pid_type) {
194 case PIDTYPE_MAX:
195 owner.type = F_OWNER_TID;
196 break;
197
198 case PIDTYPE_PID:
199 owner.type = F_OWNER_PID;
200 break;
201
202 case PIDTYPE_PGID:
Peter Zijlstra978b4052009-11-17 14:06:24 -0800203 owner.type = F_OWNER_PGRP;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700204 break;
205
206 default:
207 WARN_ON(1);
208 ret = -EINVAL;
209 break;
210 }
211 read_unlock(&filp->f_owner.lock);
212
Dan Carpenter5b544702010-06-03 12:35:42 +0200213 if (!ret) {
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700214 ret = copy_to_user(owner_p, &owner, sizeof(owner));
Dan Carpenter5b544702010-06-03 12:35:42 +0200215 if (ret)
216 ret = -EFAULT;
217 }
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700218 return ret;
219}
220
Cyrill Gorcunov1d151c32012-07-30 14:43:00 -0700221#ifdef CONFIG_CHECKPOINT_RESTORE
222static int f_getowner_uids(struct file *filp, unsigned long arg)
223{
224 struct user_namespace *user_ns = current_user_ns();
Al Viro63784dd2012-09-26 21:43:05 -0400225 uid_t __user *dst = (void __user *)arg;
Cyrill Gorcunov1d151c32012-07-30 14:43:00 -0700226 uid_t src[2];
227 int err;
228
229 read_lock(&filp->f_owner.lock);
230 src[0] = from_kuid(user_ns, filp->f_owner.uid);
231 src[1] = from_kuid(user_ns, filp->f_owner.euid);
232 read_unlock(&filp->f_owner.lock);
233
234 err = put_user(src[0], &dst[0]);
235 err |= put_user(src[1], &dst[1]);
236
237 return err;
238}
239#else
240static int f_getowner_uids(struct file *filp, unsigned long arg)
241{
242 return -EINVAL;
243}
244#endif
245
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
247 struct file *filp)
248{
249 long err = -EINVAL;
250
251 switch (cmd) {
252 case F_DUPFD:
Al Virofe17f222012-08-21 11:48:11 -0400253 err = f_dupfd(arg, filp, 0);
254 break;
Ulrich Drepper22d2b352007-10-16 23:30:26 -0700255 case F_DUPFD_CLOEXEC:
Al Viro12197712012-10-08 23:21:58 +0100256 err = f_dupfd(arg, filp, O_CLOEXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 break;
258 case F_GETFD:
259 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
260 break;
261 case F_SETFD:
262 err = 0;
263 set_close_on_exec(fd, arg & FD_CLOEXEC);
264 break;
265 case F_GETFL:
266 err = filp->f_flags;
267 break;
268 case F_SETFL:
269 err = setfl(fd, filp, arg);
270 break;
Jeff Layton5d50ffd2014-02-03 12:13:10 -0500271#if BITS_PER_LONG != 32
272 /* 32-bit arches must use fcntl64() */
Jeff Layton0d3f7a22014-04-22 08:23:58 -0400273 case F_OFD_GETLK:
Jeff Layton5d50ffd2014-02-03 12:13:10 -0500274#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 case F_GETLK:
Jeff Laytonc1e62b82014-02-03 12:13:09 -0500276 err = fcntl_getlk(filp, cmd, (struct flock __user *) arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 break;
Jeff Layton5d50ffd2014-02-03 12:13:10 -0500278#if BITS_PER_LONG != 32
279 /* 32-bit arches must use fcntl64() */
Jeff Layton0d3f7a22014-04-22 08:23:58 -0400280 case F_OFD_SETLK:
281 case F_OFD_SETLKW:
Jeff Layton5d50ffd2014-02-03 12:13:10 -0500282#endif
283 /* Fallthrough */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 case F_SETLK:
285 case F_SETLKW:
Peter Staubachc2936212005-07-27 11:45:09 -0700286 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 break;
288 case F_GETOWN:
289 /*
290 * XXX If f_owner is a process group, the
291 * negative return value will get converted
292 * into an error. Oops. If we keep the
293 * current syscall conventions, the only way
294 * to fix this will be in libc.
295 */
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700296 err = f_getown(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 force_successful_syscall_return();
298 break;
299 case F_SETOWN:
Jeff Laytone0b93ed2014-08-22 11:27:32 -0400300 f_setown(filp, arg, 1);
301 err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 break;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700303 case F_GETOWN_EX:
304 err = f_getown_ex(filp, arg);
305 break;
306 case F_SETOWN_EX:
307 err = f_setown_ex(filp, arg);
308 break;
Cyrill Gorcunov1d151c32012-07-30 14:43:00 -0700309 case F_GETOWNER_UIDS:
310 err = f_getowner_uids(filp, arg);
311 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 case F_GETSIG:
313 err = filp->f_owner.signum;
314 break;
315 case F_SETSIG:
316 /* arg == 0 restores default behaviour. */
Jesper Juhl7ed20e12005-05-01 08:59:14 -0700317 if (!valid_signal(arg)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 break;
319 }
320 err = 0;
321 filp->f_owner.signum = arg;
322 break;
323 case F_GETLEASE:
324 err = fcntl_getlease(filp);
325 break;
326 case F_SETLEASE:
327 err = fcntl_setlease(fd, filp, arg);
328 break;
329 case F_NOTIFY:
330 err = fcntl_dirnotify(fd, filp, arg);
331 break;
Jens Axboe35f3d142010-05-20 10:43:18 +0200332 case F_SETPIPE_SZ:
333 case F_GETPIPE_SZ:
334 err = pipe_fcntl(filp, cmd, arg);
335 break;
David Herrmann40e041a2014-08-08 14:25:27 -0700336 case F_ADD_SEALS:
337 case F_GET_SEALS:
338 err = shmem_fcntl(filp, cmd, arg);
339 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 default:
341 break;
342 }
343 return err;
344}
345
Al Viro1abf0c72011-03-13 03:51:11 -0400346static int check_fcntl_cmd(unsigned cmd)
347{
348 switch (cmd) {
349 case F_DUPFD:
350 case F_DUPFD_CLOEXEC:
351 case F_GETFD:
352 case F_SETFD:
353 case F_GETFL:
354 return 1;
355 }
356 return 0;
357}
358
Heiko Carstensa26eab22009-01-14 14:14:17 +0100359SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360{
Al Viro2903ff02012-08-28 12:52:22 -0400361 struct fd f = fdget_raw(fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 long err = -EBADF;
363
Al Viro2903ff02012-08-28 12:52:22 -0400364 if (!f.file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 goto out;
366
Al Viro2903ff02012-08-28 12:52:22 -0400367 if (unlikely(f.file->f_mode & FMODE_PATH)) {
Al Viro545ec2c2012-04-21 18:42:19 -0400368 if (!check_fcntl_cmd(cmd))
369 goto out1;
Al Viro1abf0c72011-03-13 03:51:11 -0400370 }
371
Al Viro2903ff02012-08-28 12:52:22 -0400372 err = security_file_fcntl(f.file, cmd, arg);
Al Viro545ec2c2012-04-21 18:42:19 -0400373 if (!err)
Al Viro2903ff02012-08-28 12:52:22 -0400374 err = do_fcntl(fd, cmd, arg, f.file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375
Al Viro545ec2c2012-04-21 18:42:19 -0400376out1:
Al Viro2903ff02012-08-28 12:52:22 -0400377 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378out:
379 return err;
380}
381
382#if BITS_PER_LONG == 32
Heiko Carstensa26eab22009-01-14 14:14:17 +0100383SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
384 unsigned long, arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385{
Al Viro2903ff02012-08-28 12:52:22 -0400386 struct fd f = fdget_raw(fd);
Al Viro545ec2c2012-04-21 18:42:19 -0400387 long err = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
Al Viro2903ff02012-08-28 12:52:22 -0400389 if (!f.file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 goto out;
391
Al Viro2903ff02012-08-28 12:52:22 -0400392 if (unlikely(f.file->f_mode & FMODE_PATH)) {
Al Viro545ec2c2012-04-21 18:42:19 -0400393 if (!check_fcntl_cmd(cmd))
394 goto out1;
Al Viro1abf0c72011-03-13 03:51:11 -0400395 }
396
Al Viro2903ff02012-08-28 12:52:22 -0400397 err = security_file_fcntl(f.file, cmd, arg);
Al Viro545ec2c2012-04-21 18:42:19 -0400398 if (err)
399 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400
401 switch (cmd) {
Jeff Layton5d50ffd2014-02-03 12:13:10 -0500402 case F_GETLK64:
Jeff Layton0d3f7a22014-04-22 08:23:58 -0400403 case F_OFD_GETLK:
Jeff Layton5d50ffd2014-02-03 12:13:10 -0500404 err = fcntl_getlk64(f.file, cmd, (struct flock64 __user *) arg);
405 break;
406 case F_SETLK64:
407 case F_SETLKW64:
Jeff Layton0d3f7a22014-04-22 08:23:58 -0400408 case F_OFD_SETLK:
409 case F_OFD_SETLKW:
Jeff Layton5d50ffd2014-02-03 12:13:10 -0500410 err = fcntl_setlk64(fd, f.file, cmd,
411 (struct flock64 __user *) arg);
412 break;
413 default:
414 err = do_fcntl(fd, cmd, arg, f.file);
415 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 }
Al Viro545ec2c2012-04-21 18:42:19 -0400417out1:
Al Viro2903ff02012-08-28 12:52:22 -0400418 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419out:
420 return err;
421}
422#endif
423
Al Viro80f0cce2017-04-08 18:10:56 -0400424#ifdef CONFIG_COMPAT
425static int get_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
426{
427 if (!access_ok(VERIFY_READ, ufl, sizeof(*ufl)) ||
428 __get_user(kfl->l_type, &ufl->l_type) ||
429 __get_user(kfl->l_whence, &ufl->l_whence) ||
430 __get_user(kfl->l_start, &ufl->l_start) ||
431 __get_user(kfl->l_len, &ufl->l_len) ||
432 __get_user(kfl->l_pid, &ufl->l_pid))
433 return -EFAULT;
434 return 0;
435}
436
437static int put_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
438{
439 if (!access_ok(VERIFY_WRITE, ufl, sizeof(*ufl)) ||
440 __put_user(kfl->l_type, &ufl->l_type) ||
441 __put_user(kfl->l_whence, &ufl->l_whence) ||
442 __put_user(kfl->l_start, &ufl->l_start) ||
443 __put_user(kfl->l_len, &ufl->l_len) ||
444 __put_user(kfl->l_pid, &ufl->l_pid))
445 return -EFAULT;
446 return 0;
447}
448
449#ifndef HAVE_ARCH_GET_COMPAT_FLOCK64
450static int get_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
451{
452 if (!access_ok(VERIFY_READ, ufl, sizeof(*ufl)) ||
453 __get_user(kfl->l_type, &ufl->l_type) ||
454 __get_user(kfl->l_whence, &ufl->l_whence) ||
455 __get_user(kfl->l_start, &ufl->l_start) ||
456 __get_user(kfl->l_len, &ufl->l_len) ||
457 __get_user(kfl->l_pid, &ufl->l_pid))
458 return -EFAULT;
459 return 0;
460}
461#endif
462
463#ifndef HAVE_ARCH_PUT_COMPAT_FLOCK64
464static int put_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
465{
466 if (!access_ok(VERIFY_WRITE, ufl, sizeof(*ufl)) ||
467 __put_user(kfl->l_type, &ufl->l_type) ||
468 __put_user(kfl->l_whence, &ufl->l_whence) ||
469 __put_user(kfl->l_start, &ufl->l_start) ||
470 __put_user(kfl->l_len, &ufl->l_len) ||
471 __put_user(kfl->l_pid, &ufl->l_pid))
472 return -EFAULT;
473 return 0;
474}
475#endif
476
477static unsigned int
478convert_fcntl_cmd(unsigned int cmd)
479{
480 switch (cmd) {
481 case F_GETLK64:
482 return F_GETLK;
483 case F_SETLK64:
484 return F_SETLK;
485 case F_SETLKW64:
486 return F_SETLKW;
487 }
488
489 return cmd;
490}
491
492COMPAT_SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
493 compat_ulong_t, arg)
494{
495 mm_segment_t old_fs;
496 struct flock f;
497 long ret;
498 unsigned int conv_cmd;
499
500 switch (cmd) {
501 case F_GETLK:
502 case F_SETLK:
503 case F_SETLKW:
504 ret = get_compat_flock(&f, compat_ptr(arg));
505 if (ret != 0)
506 break;
507 old_fs = get_fs();
508 set_fs(KERNEL_DS);
509 ret = sys_fcntl(fd, cmd, (unsigned long)&f);
510 set_fs(old_fs);
511 if (cmd == F_GETLK && ret == 0) {
512 /* GETLK was successful and we need to return the data...
513 * but it needs to fit in the compat structure.
514 * l_start shouldn't be too big, unless the original
515 * start + end is greater than COMPAT_OFF_T_MAX, in which
516 * case the app was asking for trouble, so we return
517 * -EOVERFLOW in that case.
518 * l_len could be too big, in which case we just truncate it,
519 * and only allow the app to see that part of the conflicting
520 * lock that might make sense to it anyway
521 */
522
523 if (f.l_start > COMPAT_OFF_T_MAX)
524 ret = -EOVERFLOW;
525 if (f.l_len > COMPAT_OFF_T_MAX)
526 f.l_len = COMPAT_OFF_T_MAX;
527 if (ret == 0)
528 ret = put_compat_flock(&f, compat_ptr(arg));
529 }
530 break;
531
532 case F_GETLK64:
533 case F_SETLK64:
534 case F_SETLKW64:
535 case F_OFD_GETLK:
536 case F_OFD_SETLK:
537 case F_OFD_SETLKW:
538 ret = get_compat_flock64(&f, compat_ptr(arg));
539 if (ret != 0)
540 break;
541 old_fs = get_fs();
542 set_fs(KERNEL_DS);
543 conv_cmd = convert_fcntl_cmd(cmd);
544 ret = sys_fcntl(fd, conv_cmd, (unsigned long)&f);
545 set_fs(old_fs);
546 if ((conv_cmd == F_GETLK || conv_cmd == F_OFD_GETLK) && ret == 0) {
547 /* need to return lock information - see above for commentary */
548 if (f.l_start > COMPAT_LOFF_T_MAX)
549 ret = -EOVERFLOW;
550 if (f.l_len > COMPAT_LOFF_T_MAX)
551 f.l_len = COMPAT_LOFF_T_MAX;
552 if (ret == 0)
553 ret = put_compat_flock64(&f, compat_ptr(arg));
554 }
555 break;
556
557 default:
558 ret = sys_fcntl(fd, cmd, arg);
559 break;
560 }
561 return ret;
562}
563
564COMPAT_SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd,
565 compat_ulong_t, arg)
566{
567 switch (cmd) {
568 case F_GETLK64:
569 case F_SETLK64:
570 case F_SETLKW64:
571 case F_OFD_GETLK:
572 case F_OFD_SETLK:
573 case F_OFD_SETLKW:
574 return -EINVAL;
575 }
576 return compat_sys_fcntl64(fd, cmd, arg);
577}
578#endif
579
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580/* Table to convert sigio signal codes into poll band bitmaps */
581
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800582static const long band_table[NSIGPOLL] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 POLLIN | POLLRDNORM, /* POLL_IN */
584 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
585 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
586 POLLERR, /* POLL_ERR */
587 POLLPRI | POLLRDBAND, /* POLL_PRI */
588 POLLHUP | POLLERR /* POLL_HUP */
589};
590
591static inline int sigio_perm(struct task_struct *p,
592 struct fown_struct *fown, int sig)
593{
David Howellsc69e8d92008-11-14 10:39:19 +1100594 const struct cred *cred;
595 int ret;
596
597 rcu_read_lock();
598 cred = __task_cred(p);
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -0800599 ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
600 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
601 uid_eq(fown->uid, cred->suid) || uid_eq(fown->uid, cred->uid)) &&
David Howellsc69e8d92008-11-14 10:39:19 +1100602 !security_file_send_sigiotask(p, fown, sig));
603 rcu_read_unlock();
604 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605}
606
607static void send_sigio_to_task(struct task_struct *p,
Oleg Nesterov8eeee4e2009-06-17 00:27:10 +0200608 struct fown_struct *fown,
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700609 int fd, int reason, int group)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610{
Oleg Nesterov8eeee4e2009-06-17 00:27:10 +0200611 /*
612 * F_SETSIG can change ->signum lockless in parallel, make
613 * sure we read it once and use the same value throughout.
614 */
615 int signum = ACCESS_ONCE(fown->signum);
616
617 if (!sigio_perm(p, fown, signum))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 return;
619
Oleg Nesterov8eeee4e2009-06-17 00:27:10 +0200620 switch (signum) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 siginfo_t si;
622 default:
623 /* Queue a rt signal with the appropriate fd as its
624 value. We use SI_SIGIO as the source, not
625 SI_KERNEL, since kernel signals always get
626 delivered even if we can't queue. Failure to
627 queue in this case _should_ be reported; we fall
628 back to SIGIO in that case. --sct */
Oleg Nesterov8eeee4e2009-06-17 00:27:10 +0200629 si.si_signo = signum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 si.si_errno = 0;
631 si.si_code = reason;
632 /* Make sure we are called with one of the POLL_*
633 reasons, otherwise we could leak kernel stack into
634 userspace. */
Eric Sesterhennf6298aa2006-04-02 13:37:19 +0200635 BUG_ON((reason & __SI_MASK) != __SI_POLL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 if (reason - POLL_IN >= NSIGPOLL)
637 si.si_band = ~0L;
638 else
639 si.si_band = band_table[reason - POLL_IN];
640 si.si_fd = fd;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700641 if (!do_send_sig_info(signum, &si, p, group))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 break;
643 /* fall-through: fall back on the old plain SIGIO signal */
644 case 0:
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700645 do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 }
647}
648
649void send_sigio(struct fown_struct *fown, int fd, int band)
650{
651 struct task_struct *p;
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700652 enum pid_type type;
653 struct pid *pid;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700654 int group = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655
656 read_lock(&fown->lock);
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700657
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700658 type = fown->pid_type;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700659 if (type == PIDTYPE_MAX) {
660 group = 0;
661 type = PIDTYPE_PID;
662 }
663
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 pid = fown->pid;
665 if (!pid)
666 goto out_unlock_fown;
667
668 read_lock(&tasklist_lock);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700669 do_each_pid_task(pid, type, p) {
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700670 send_sigio_to_task(p, fown, fd, band, group);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700671 } while_each_pid_task(pid, type, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 read_unlock(&tasklist_lock);
673 out_unlock_fown:
674 read_unlock(&fown->lock);
675}
676
677static void send_sigurg_to_task(struct task_struct *p,
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700678 struct fown_struct *fown, int group)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679{
680 if (sigio_perm(p, fown, SIGURG))
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700681 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682}
683
684int send_sigurg(struct fown_struct *fown)
685{
686 struct task_struct *p;
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700687 enum pid_type type;
688 struct pid *pid;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700689 int group = 1;
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700690 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691
692 read_lock(&fown->lock);
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700693
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700694 type = fown->pid_type;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700695 if (type == PIDTYPE_MAX) {
696 group = 0;
697 type = PIDTYPE_PID;
698 }
699
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 pid = fown->pid;
701 if (!pid)
702 goto out_unlock_fown;
703
704 ret = 1;
705
706 read_lock(&tasklist_lock);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700707 do_each_pid_task(pid, type, p) {
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700708 send_sigurg_to_task(p, fown, group);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700709 } while_each_pid_task(pid, type, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 read_unlock(&tasklist_lock);
711 out_unlock_fown:
712 read_unlock(&fown->lock);
713 return ret;
714}
715
Eric Dumazet989a2972010-04-14 09:55:35 +0000716static DEFINE_SPINLOCK(fasync_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800717static struct kmem_cache *fasync_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718
Eric Dumazet989a2972010-04-14 09:55:35 +0000719static void fasync_free_rcu(struct rcu_head *head)
720{
721 kmem_cache_free(fasync_cache,
722 container_of(head, struct fasync_struct, fa_rcu));
723}
724
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725/*
Linus Torvalds53281b62009-12-16 08:23:37 -0800726 * Remove a fasync entry. If successfully removed, return
727 * positive and clear the FASYNC flag. If no entry exists,
728 * do nothing and return 0.
729 *
730 * NOTE! It is very important that the FASYNC flag always
731 * match the state "is the filp on a fasync list".
732 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 */
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400734int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735{
736 struct fasync_struct *fa, **fp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 int result = 0;
738
Jonathan Corbet4a6a4492009-03-27 12:24:31 -0600739 spin_lock(&filp->f_lock);
Eric Dumazet989a2972010-04-14 09:55:35 +0000740 spin_lock(&fasync_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
Linus Torvalds53281b62009-12-16 08:23:37 -0800742 if (fa->fa_file != filp)
743 continue;
Eric Dumazet989a2972010-04-14 09:55:35 +0000744
745 spin_lock_irq(&fa->fa_lock);
746 fa->fa_file = NULL;
747 spin_unlock_irq(&fa->fa_lock);
748
Linus Torvalds53281b62009-12-16 08:23:37 -0800749 *fp = fa->fa_next;
Eric Dumazet989a2972010-04-14 09:55:35 +0000750 call_rcu(&fa->fa_rcu, fasync_free_rcu);
Jonathan Corbet76398422009-02-01 14:26:59 -0700751 filp->f_flags &= ~FASYNC;
Linus Torvalds53281b62009-12-16 08:23:37 -0800752 result = 1;
753 break;
754 }
Eric Dumazet989a2972010-04-14 09:55:35 +0000755 spin_unlock(&fasync_lock);
Jonathan Corbet4a6a4492009-03-27 12:24:31 -0600756 spin_unlock(&filp->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 return result;
758}
759
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400760struct fasync_struct *fasync_alloc(void)
Linus Torvalds53281b62009-12-16 08:23:37 -0800761{
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400762 return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
763}
Linus Torvalds53281b62009-12-16 08:23:37 -0800764
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400765/*
766 * NOTE! This can be used only for unused fasync entries:
767 * entries that actually got inserted on the fasync list
768 * need to be released by rcu - see fasync_remove_entry.
769 */
770void fasync_free(struct fasync_struct *new)
771{
772 kmem_cache_free(fasync_cache, new);
773}
774
775/*
776 * Insert a new entry into the fasync list. Return the pointer to the
777 * old one if we didn't use the new one.
Linus Torvalds55f335a2010-10-27 18:17:02 -0700778 *
779 * NOTE! It is very important that the FASYNC flag always
780 * match the state "is the filp on a fasync list".
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400781 */
782struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
783{
784 struct fasync_struct *fa, **fp;
Linus Torvalds53281b62009-12-16 08:23:37 -0800785
786 spin_lock(&filp->f_lock);
Eric Dumazet989a2972010-04-14 09:55:35 +0000787 spin_lock(&fasync_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800788 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
789 if (fa->fa_file != filp)
790 continue;
Eric Dumazet989a2972010-04-14 09:55:35 +0000791
792 spin_lock_irq(&fa->fa_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800793 fa->fa_fd = fd;
Eric Dumazet989a2972010-04-14 09:55:35 +0000794 spin_unlock_irq(&fa->fa_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800795 goto out;
796 }
797
Eric Dumazet989a2972010-04-14 09:55:35 +0000798 spin_lock_init(&new->fa_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800799 new->magic = FASYNC_MAGIC;
800 new->fa_file = filp;
801 new->fa_fd = fd;
802 new->fa_next = *fapp;
Eric Dumazet989a2972010-04-14 09:55:35 +0000803 rcu_assign_pointer(*fapp, new);
Linus Torvalds53281b62009-12-16 08:23:37 -0800804 filp->f_flags |= FASYNC;
805
806out:
Eric Dumazet989a2972010-04-14 09:55:35 +0000807 spin_unlock(&fasync_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800808 spin_unlock(&filp->f_lock);
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400809 return fa;
810}
811
812/*
813 * Add a fasync entry. Return negative on error, positive if
814 * added, and zero if did nothing but change an existing one.
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400815 */
816static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
817{
818 struct fasync_struct *new;
819
820 new = fasync_alloc();
821 if (!new)
822 return -ENOMEM;
823
824 /*
825 * fasync_insert_entry() returns the old (update) entry if
826 * it existed.
827 *
828 * So free the (unused) new entry and return 0 to let the
829 * caller know that we didn't add any new fasync entries.
830 */
831 if (fasync_insert_entry(fd, filp, fapp, new)) {
832 fasync_free(new);
833 return 0;
834 }
835
836 return 1;
Linus Torvalds53281b62009-12-16 08:23:37 -0800837}
838
839/*
840 * fasync_helper() is used by almost all character device drivers
841 * to set up the fasync queue, and for regular files by the file
842 * lease code. It returns negative on error, 0 if it did no changes
843 * and positive if it added/deleted the entry.
844 */
845int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
846{
847 if (!on)
848 return fasync_remove_entry(filp, fapp);
849 return fasync_add_entry(fd, filp, fapp);
850}
851
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852EXPORT_SYMBOL(fasync_helper);
853
Eric Dumazet989a2972010-04-14 09:55:35 +0000854/*
855 * rcu_read_lock() is held
856 */
857static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858{
859 while (fa) {
Eric Dumazet989a2972010-04-14 09:55:35 +0000860 struct fown_struct *fown;
Andrew Mortonf4985dc2010-06-29 15:05:42 -0700861 unsigned long flags;
862
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 if (fa->magic != FASYNC_MAGIC) {
864 printk(KERN_ERR "kill_fasync: bad magic number in "
865 "fasync_struct!\n");
866 return;
867 }
Andrew Mortonf4985dc2010-06-29 15:05:42 -0700868 spin_lock_irqsave(&fa->fa_lock, flags);
Eric Dumazet989a2972010-04-14 09:55:35 +0000869 if (fa->fa_file) {
870 fown = &fa->fa_file->f_owner;
871 /* Don't send SIGURG to processes which have not set a
872 queued signum: SIGURG has its own default signalling
873 mechanism. */
874 if (!(sig == SIGURG && fown->signum == 0))
875 send_sigio(fown, fa->fa_fd, band);
876 }
Andrew Mortonf4985dc2010-06-29 15:05:42 -0700877 spin_unlock_irqrestore(&fa->fa_lock, flags);
Eric Dumazet989a2972010-04-14 09:55:35 +0000878 fa = rcu_dereference(fa->fa_next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 }
880}
881
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882void kill_fasync(struct fasync_struct **fp, int sig, int band)
883{
884 /* First a quick test without locking: usually
885 * the list is empty.
886 */
887 if (*fp) {
Eric Dumazet989a2972010-04-14 09:55:35 +0000888 rcu_read_lock();
889 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
890 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 }
892}
893EXPORT_SYMBOL(kill_fasync);
894
Wu Fengguang454eedb2010-08-10 18:01:29 -0700895static int __init fcntl_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896{
James Bottomley3ab04d52010-09-09 16:38:12 -0700897 /*
898 * Please add new bits here to ensure allocation uniqueness.
899 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
900 * is defined as O_NONBLOCK on some platforms and not on others.
901 */
Christoph Hellwig80f18372017-04-27 09:42:24 +0200902 BUILD_BUG_ON(21 - 1 /* for O_RDONLY being 0 */ !=
903 HWEIGHT32(
904 (VALID_OPEN_FLAGS & ~(O_NONBLOCK | O_NDELAY)) |
905 __FMODE_EXEC | __FMODE_NONOTIFY));
Wu Fengguang454eedb2010-08-10 18:01:29 -0700906
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 fasync_cache = kmem_cache_create("fasync_cache",
Paul Mundt20c2df82007-07-20 10:11:58 +0900908 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 return 0;
910}
911
Wu Fengguang454eedb2010-08-10 18:01:29 -0700912module_init(fcntl_init)