blob: 22764c7c8382ce607ca5f9ec9b1a2f48e3d15c4e [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023
24#include <asm/poll.h>
25#include <asm/siginfo.h>
26#include <asm/uaccess.h>
27
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -080028void set_close_on_exec(unsigned int fd, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -070029{
30 struct files_struct *files = current->files;
Dipankar Sarmabadf1662005-09-09 13:04:10 -070031 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -070032 spin_lock(&files->file_lock);
Dipankar Sarmabadf1662005-09-09 13:04:10 -070033 fdt = files_fdtable(files);
Linus Torvalds1da177e2005-04-16 15:20:36 -070034 if (flag)
Dipankar Sarmabadf1662005-09-09 13:04:10 -070035 FD_SET(fd, fdt->close_on_exec);
Linus Torvalds1da177e2005-04-16 15:20:36 -070036 else
Dipankar Sarmabadf1662005-09-09 13:04:10 -070037 FD_CLR(fd, fdt->close_on_exec);
Linus Torvalds1da177e2005-04-16 15:20:36 -070038 spin_unlock(&files->file_lock);
39}
40
Arjan van de Ven858119e2006-01-14 13:20:43 -080041static int get_close_on_exec(unsigned int fd)
Linus Torvalds1da177e2005-04-16 15:20:36 -070042{
43 struct files_struct *files = current->files;
Dipankar Sarmabadf1662005-09-09 13:04:10 -070044 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -070045 int res;
Dipankar Sarmab8359962005-09-09 13:04:14 -070046 rcu_read_lock();
Dipankar Sarmabadf1662005-09-09 13:04:10 -070047 fdt = files_fdtable(files);
48 res = FD_ISSET(fd, fdt->close_on_exec);
Dipankar Sarmab8359962005-09-09 13:04:14 -070049 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -070050 return res;
51}
52
Heiko Carstensa26eab22009-01-14 14:14:17 +010053SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -070054{
55 int err = -EBADF;
56 struct file * file, *tofree;
57 struct files_struct * files = current->files;
Dipankar Sarmabadf1662005-09-09 13:04:10 -070058 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
Ulrich Drepper336dd1f2008-07-23 21:29:29 -070060 if ((flags & ~O_CLOEXEC) != 0)
61 return -EINVAL;
62
Al Viro6c5d0512008-07-26 13:38:19 -040063 if (unlikely(oldfd == newfd))
64 return -EINVAL;
65
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 spin_lock(&files->file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 err = expand_files(files, newfd);
Al Viro1b7e1902008-07-30 06:18:03 -040068 file = fcheck(oldfd);
69 if (unlikely(!file))
70 goto Ebadf;
Al Viro4e1e0182008-07-26 16:01:20 -040071 if (unlikely(err < 0)) {
72 if (err == -EMFILE)
Al Viro1b7e1902008-07-30 06:18:03 -040073 goto Ebadf;
74 goto out_unlock;
Al Viro4e1e0182008-07-26 16:01:20 -040075 }
Al Viro1b7e1902008-07-30 06:18:03 -040076 /*
77 * We need to detect attempts to do dup2() over allocated but still
78 * not finished descriptor. NB: OpenBSD avoids that at the price of
79 * extra work in their equivalent of fget() - they insert struct
80 * file immediately after grabbing descriptor, mark it larval if
81 * more work (e.g. actual opening) is needed and make sure that
82 * fget() treats larval files as absent. Potentially interesting,
83 * but while extra work in fget() is trivial, locking implications
84 * and amount of surgery on open()-related paths in VFS are not.
85 * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
86 * deadlocks in rather amusing ways, AFAICS. All of that is out of
87 * scope of POSIX or SUS, since neither considers shared descriptor
88 * tables and this condition does not arise without those.
89 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070090 err = -EBUSY;
Dipankar Sarmabadf1662005-09-09 13:04:10 -070091 fdt = files_fdtable(files);
92 tofree = fdt->fd[newfd];
93 if (!tofree && FD_ISSET(newfd, fdt->open_fds))
Al Viro1b7e1902008-07-30 06:18:03 -040094 goto out_unlock;
95 get_file(file);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070096 rcu_assign_pointer(fdt->fd[newfd], file);
Dipankar Sarmabadf1662005-09-09 13:04:10 -070097 FD_SET(newfd, fdt->open_fds);
Ulrich Drepper336dd1f2008-07-23 21:29:29 -070098 if (flags & O_CLOEXEC)
99 FD_SET(newfd, fdt->close_on_exec);
100 else
101 FD_CLR(newfd, fdt->close_on_exec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 spin_unlock(&files->file_lock);
103
104 if (tofree)
105 filp_close(tofree, files);
Al Viro1b7e1902008-07-30 06:18:03 -0400106
107 return newfd;
108
109Ebadf:
110 err = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111out_unlock:
112 spin_unlock(&files->file_lock);
Al Viro1b7e1902008-07-30 06:18:03 -0400113 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114}
115
Heiko Carstensa26eab22009-01-14 14:14:17 +0100116SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
Ulrich Drepper336dd1f2008-07-23 21:29:29 -0700117{
Al Viro6c5d0512008-07-26 13:38:19 -0400118 if (unlikely(newfd == oldfd)) { /* corner case */
119 struct files_struct *files = current->files;
Jeff Mahoney2b79bc42009-05-11 14:25:34 -0400120 int retval = oldfd;
121
Al Viro6c5d0512008-07-26 13:38:19 -0400122 rcu_read_lock();
123 if (!fcheck_files(files, oldfd))
Jeff Mahoney2b79bc42009-05-11 14:25:34 -0400124 retval = -EBADF;
Al Viro6c5d0512008-07-26 13:38:19 -0400125 rcu_read_unlock();
Jeff Mahoney2b79bc42009-05-11 14:25:34 -0400126 return retval;
Al Viro6c5d0512008-07-26 13:38:19 -0400127 }
Ulrich Drepper336dd1f2008-07-23 21:29:29 -0700128 return sys_dup3(oldfd, newfd, 0);
129}
130
Heiko Carstensa26eab22009-01-14 14:14:17 +0100131SYSCALL_DEFINE1(dup, unsigned int, fildes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132{
133 int ret = -EBADF;
Al Viro1abf0c72011-03-13 03:51:11 -0400134 struct file *file = fget_raw(fildes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135
Al Viro1027abe2008-07-30 04:13:04 -0400136 if (file) {
137 ret = get_unused_fd();
138 if (ret >= 0)
139 fd_install(ret, file);
140 else
141 fput(file);
142 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 return ret;
144}
145
Jonathan Corbet76398422009-02-01 14:26:59 -0700146#define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147
148static int setfl(int fd, struct file * filp, unsigned long arg)
149{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800150 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 int error = 0;
152
dean gaudet7d95c8f2006-02-03 03:04:30 -0800153 /*
154 * O_APPEND cannot be cleared if the file is marked as append-only
155 * and the file is open for write.
156 */
157 if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 return -EPERM;
159
160 /* O_NOATIME can only be set by the owner or superuser */
161 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
Serge E. Hallyn2e149672011-03-23 16:43:26 -0700162 if (!inode_owner_or_capable(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 return -EPERM;
164
165 /* required for strict SunOS emulation */
166 if (O_NONBLOCK != O_NDELAY)
167 if (arg & O_NDELAY)
168 arg |= O_NONBLOCK;
169
170 if (arg & O_DIRECT) {
171 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
172 !filp->f_mapping->a_ops->direct_IO)
173 return -EINVAL;
174 }
175
176 if (filp->f_op && filp->f_op->check_flags)
177 error = filp->f_op->check_flags(arg);
178 if (error)
179 return error;
180
Jonathan Corbet218d11a2008-12-05 16:12:48 -0700181 /*
Jonathan Corbet76398422009-02-01 14:26:59 -0700182 * ->fasync() is responsible for setting the FASYNC bit.
Jonathan Corbet218d11a2008-12-05 16:12:48 -0700183 */
Jonathan Corbet76398422009-02-01 14:26:59 -0700184 if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op &&
185 filp->f_op->fasync) {
186 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
187 if (error < 0)
188 goto out;
Jonathan Corbet60aa4922009-02-01 14:52:56 -0700189 if (error > 0)
190 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 }
Jonathan Corbetdb1dd4d2009-02-06 15:25:24 -0700192 spin_lock(&filp->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
Jonathan Corbetdb1dd4d2009-02-06 15:25:24 -0700194 spin_unlock(&filp->f_lock);
Jonathan Corbet76398422009-02-01 14:26:59 -0700195
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 return error;
198}
199
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700200static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
Oleg Nesterov2f38d702009-06-16 22:07:46 +0200201 int force)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202{
Linus Torvalds80e1e822010-02-07 10:11:23 -0800203 write_lock_irq(&filp->f_owner.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 if (force || !filp->f_owner.pid) {
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700205 put_pid(filp->f_owner.pid);
206 filp->f_owner.pid = get_pid(pid);
207 filp->f_owner.pid_type = type;
Oleg Nesterov2f38d702009-06-16 22:07:46 +0200208
209 if (pid) {
210 const struct cred *cred = current_cred();
211 filp->f_owner.uid = cred->uid;
212 filp->f_owner.euid = cred->euid;
213 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 }
Linus Torvalds80e1e822010-02-07 10:11:23 -0800215 write_unlock_irq(&filp->f_owner.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216}
217
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700218int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
219 int force)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220{
221 int err;
Oleg Nesterov2f38d702009-06-16 22:07:46 +0200222
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 err = security_file_set_fowner(filp);
224 if (err)
225 return err;
226
Oleg Nesterov2f38d702009-06-16 22:07:46 +0200227 f_modown(filp, pid, type, force);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 return 0;
229}
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700230EXPORT_SYMBOL(__f_setown);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700232int f_setown(struct file *filp, unsigned long arg, int force)
233{
234 enum pid_type type;
235 struct pid *pid;
236 int who = arg;
237 int result;
238 type = PIDTYPE_PID;
239 if (who < 0) {
240 type = PIDTYPE_PGID;
241 who = -who;
242 }
243 rcu_read_lock();
Pavel Emelyanovb4888932007-10-18 23:40:14 -0700244 pid = find_vpid(who);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700245 result = __f_setown(filp, pid, type, force);
246 rcu_read_unlock();
247 return result;
248}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249EXPORT_SYMBOL(f_setown);
250
251void f_delown(struct file *filp)
252{
Oleg Nesterov2f38d702009-06-16 22:07:46 +0200253 f_modown(filp, NULL, PIDTYPE_PID, 1);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700254}
255
256pid_t f_getown(struct file *filp)
257{
258 pid_t pid;
Eric W. Biederman43fa1ad2006-10-02 02:17:27 -0700259 read_lock(&filp->f_owner.lock);
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -0800260 pid = pid_vnr(filp->f_owner.pid);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700261 if (filp->f_owner.pid_type == PIDTYPE_PGID)
262 pid = -pid;
Eric W. Biederman43fa1ad2006-10-02 02:17:27 -0700263 read_unlock(&filp->f_owner.lock);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700264 return pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265}
266
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700267static int f_setown_ex(struct file *filp, unsigned long arg)
268{
269 struct f_owner_ex * __user owner_p = (void * __user)arg;
270 struct f_owner_ex owner;
271 struct pid *pid;
272 int type;
273 int ret;
274
275 ret = copy_from_user(&owner, owner_p, sizeof(owner));
276 if (ret)
Dan Carpenter5b544702010-06-03 12:35:42 +0200277 return -EFAULT;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700278
279 switch (owner.type) {
280 case F_OWNER_TID:
281 type = PIDTYPE_MAX;
282 break;
283
284 case F_OWNER_PID:
285 type = PIDTYPE_PID;
286 break;
287
Peter Zijlstra978b4052009-11-17 14:06:24 -0800288 case F_OWNER_PGRP:
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700289 type = PIDTYPE_PGID;
290 break;
291
292 default:
293 return -EINVAL;
294 }
295
296 rcu_read_lock();
297 pid = find_vpid(owner.pid);
298 if (owner.pid && !pid)
299 ret = -ESRCH;
300 else
301 ret = __f_setown(filp, pid, type, 1);
302 rcu_read_unlock();
303
304 return ret;
305}
306
307static int f_getown_ex(struct file *filp, unsigned long arg)
308{
309 struct f_owner_ex * __user owner_p = (void * __user)arg;
310 struct f_owner_ex owner;
311 int ret = 0;
312
313 read_lock(&filp->f_owner.lock);
314 owner.pid = pid_vnr(filp->f_owner.pid);
315 switch (filp->f_owner.pid_type) {
316 case PIDTYPE_MAX:
317 owner.type = F_OWNER_TID;
318 break;
319
320 case PIDTYPE_PID:
321 owner.type = F_OWNER_PID;
322 break;
323
324 case PIDTYPE_PGID:
Peter Zijlstra978b4052009-11-17 14:06:24 -0800325 owner.type = F_OWNER_PGRP;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700326 break;
327
328 default:
329 WARN_ON(1);
330 ret = -EINVAL;
331 break;
332 }
333 read_unlock(&filp->f_owner.lock);
334
Dan Carpenter5b544702010-06-03 12:35:42 +0200335 if (!ret) {
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700336 ret = copy_to_user(owner_p, &owner, sizeof(owner));
Dan Carpenter5b544702010-06-03 12:35:42 +0200337 if (ret)
338 ret = -EFAULT;
339 }
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700340 return ret;
341}
342
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
344 struct file *filp)
345{
346 long err = -EINVAL;
347
348 switch (cmd) {
349 case F_DUPFD:
Ulrich Drepper22d2b352007-10-16 23:30:26 -0700350 case F_DUPFD_CLOEXEC:
Jiri Slabyd554ed892010-03-05 13:42:42 -0800351 if (arg >= rlimit(RLIMIT_NOFILE))
Al Viro4e1e0182008-07-26 16:01:20 -0400352 break;
Al Viro1027abe2008-07-30 04:13:04 -0400353 err = alloc_fd(arg, cmd == F_DUPFD_CLOEXEC ? O_CLOEXEC : 0);
354 if (err >= 0) {
355 get_file(filp);
356 fd_install(err, filp);
357 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 break;
359 case F_GETFD:
360 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
361 break;
362 case F_SETFD:
363 err = 0;
364 set_close_on_exec(fd, arg & FD_CLOEXEC);
365 break;
366 case F_GETFL:
367 err = filp->f_flags;
368 break;
369 case F_SETFL:
370 err = setfl(fd, filp, arg);
371 break;
372 case F_GETLK:
373 err = fcntl_getlk(filp, (struct flock __user *) arg);
374 break;
375 case F_SETLK:
376 case F_SETLKW:
Peter Staubachc2936212005-07-27 11:45:09 -0700377 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 break;
379 case F_GETOWN:
380 /*
381 * XXX If f_owner is a process group, the
382 * negative return value will get converted
383 * into an error. Oops. If we keep the
384 * current syscall conventions, the only way
385 * to fix this will be in libc.
386 */
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700387 err = f_getown(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 force_successful_syscall_return();
389 break;
390 case F_SETOWN:
391 err = f_setown(filp, arg, 1);
392 break;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700393 case F_GETOWN_EX:
394 err = f_getown_ex(filp, arg);
395 break;
396 case F_SETOWN_EX:
397 err = f_setown_ex(filp, arg);
398 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 case F_GETSIG:
400 err = filp->f_owner.signum;
401 break;
402 case F_SETSIG:
403 /* arg == 0 restores default behaviour. */
Jesper Juhl7ed20e12005-05-01 08:59:14 -0700404 if (!valid_signal(arg)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 break;
406 }
407 err = 0;
408 filp->f_owner.signum = arg;
409 break;
410 case F_GETLEASE:
411 err = fcntl_getlease(filp);
412 break;
413 case F_SETLEASE:
414 err = fcntl_setlease(fd, filp, arg);
415 break;
416 case F_NOTIFY:
417 err = fcntl_dirnotify(fd, filp, arg);
418 break;
Jens Axboe35f3d142010-05-20 10:43:18 +0200419 case F_SETPIPE_SZ:
420 case F_GETPIPE_SZ:
421 err = pipe_fcntl(filp, cmd, arg);
422 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 default:
424 break;
425 }
426 return err;
427}
428
Al Viro1abf0c72011-03-13 03:51:11 -0400429static int check_fcntl_cmd(unsigned cmd)
430{
431 switch (cmd) {
432 case F_DUPFD:
433 case F_DUPFD_CLOEXEC:
434 case F_GETFD:
435 case F_SETFD:
436 case F_GETFL:
437 return 1;
438 }
439 return 0;
440}
441
Heiko Carstensa26eab22009-01-14 14:14:17 +0100442SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443{
444 struct file *filp;
445 long err = -EBADF;
446
Al Viro1abf0c72011-03-13 03:51:11 -0400447 filp = fget_raw(fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 if (!filp)
449 goto out;
450
Al Viro1abf0c72011-03-13 03:51:11 -0400451 if (unlikely(filp->f_mode & FMODE_PATH)) {
452 if (!check_fcntl_cmd(cmd)) {
453 fput(filp);
454 goto out;
455 }
456 }
457
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 err = security_file_fcntl(filp, cmd, arg);
459 if (err) {
460 fput(filp);
461 return err;
462 }
463
464 err = do_fcntl(fd, cmd, arg, filp);
465
466 fput(filp);
467out:
468 return err;
469}
470
471#if BITS_PER_LONG == 32
Heiko Carstensa26eab22009-01-14 14:14:17 +0100472SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
473 unsigned long, arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474{
475 struct file * filp;
476 long err;
477
478 err = -EBADF;
Al Viro1abf0c72011-03-13 03:51:11 -0400479 filp = fget_raw(fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 if (!filp)
481 goto out;
482
Al Viro1abf0c72011-03-13 03:51:11 -0400483 if (unlikely(filp->f_mode & FMODE_PATH)) {
484 if (!check_fcntl_cmd(cmd)) {
485 fput(filp);
486 goto out;
487 }
488 }
489
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 err = security_file_fcntl(filp, cmd, arg);
491 if (err) {
492 fput(filp);
493 return err;
494 }
495 err = -EBADF;
496
497 switch (cmd) {
498 case F_GETLK64:
499 err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
500 break;
501 case F_SETLK64:
502 case F_SETLKW64:
Peter Staubachc2936212005-07-27 11:45:09 -0700503 err = fcntl_setlk64(fd, filp, cmd,
504 (struct flock64 __user *) arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 break;
506 default:
507 err = do_fcntl(fd, cmd, arg, filp);
508 break;
509 }
510 fput(filp);
511out:
512 return err;
513}
514#endif
515
516/* Table to convert sigio signal codes into poll band bitmaps */
517
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800518static const long band_table[NSIGPOLL] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 POLLIN | POLLRDNORM, /* POLL_IN */
520 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
521 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
522 POLLERR, /* POLL_ERR */
523 POLLPRI | POLLRDBAND, /* POLL_PRI */
524 POLLHUP | POLLERR /* POLL_HUP */
525};
526
527static inline int sigio_perm(struct task_struct *p,
528 struct fown_struct *fown, int sig)
529{
David Howellsc69e8d92008-11-14 10:39:19 +1100530 const struct cred *cred;
531 int ret;
532
533 rcu_read_lock();
534 cred = __task_cred(p);
535 ret = ((fown->euid == 0 ||
536 fown->euid == cred->suid || fown->euid == cred->uid ||
537 fown->uid == cred->suid || fown->uid == cred->uid) &&
538 !security_file_send_sigiotask(p, fown, sig));
539 rcu_read_unlock();
540 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541}
542
543static void send_sigio_to_task(struct task_struct *p,
Oleg Nesterov8eeee4e2009-06-17 00:27:10 +0200544 struct fown_struct *fown,
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700545 int fd, int reason, int group)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546{
Oleg Nesterov8eeee4e2009-06-17 00:27:10 +0200547 /*
548 * F_SETSIG can change ->signum lockless in parallel, make
549 * sure we read it once and use the same value throughout.
550 */
551 int signum = ACCESS_ONCE(fown->signum);
552
553 if (!sigio_perm(p, fown, signum))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 return;
555
Oleg Nesterov8eeee4e2009-06-17 00:27:10 +0200556 switch (signum) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 siginfo_t si;
558 default:
559 /* Queue a rt signal with the appropriate fd as its
560 value. We use SI_SIGIO as the source, not
561 SI_KERNEL, since kernel signals always get
562 delivered even if we can't queue. Failure to
563 queue in this case _should_ be reported; we fall
564 back to SIGIO in that case. --sct */
Oleg Nesterov8eeee4e2009-06-17 00:27:10 +0200565 si.si_signo = signum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 si.si_errno = 0;
567 si.si_code = reason;
568 /* Make sure we are called with one of the POLL_*
569 reasons, otherwise we could leak kernel stack into
570 userspace. */
Eric Sesterhennf6298aa2006-04-02 13:37:19 +0200571 BUG_ON((reason & __SI_MASK) != __SI_POLL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 if (reason - POLL_IN >= NSIGPOLL)
573 si.si_band = ~0L;
574 else
575 si.si_band = band_table[reason - POLL_IN];
576 si.si_fd = fd;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700577 if (!do_send_sig_info(signum, &si, p, group))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 break;
579 /* fall-through: fall back on the old plain SIGIO signal */
580 case 0:
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700581 do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 }
583}
584
585void send_sigio(struct fown_struct *fown, int fd, int band)
586{
587 struct task_struct *p;
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700588 enum pid_type type;
589 struct pid *pid;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700590 int group = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591
592 read_lock(&fown->lock);
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700593
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700594 type = fown->pid_type;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700595 if (type == PIDTYPE_MAX) {
596 group = 0;
597 type = PIDTYPE_PID;
598 }
599
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 pid = fown->pid;
601 if (!pid)
602 goto out_unlock_fown;
603
604 read_lock(&tasklist_lock);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700605 do_each_pid_task(pid, type, p) {
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700606 send_sigio_to_task(p, fown, fd, band, group);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700607 } while_each_pid_task(pid, type, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 read_unlock(&tasklist_lock);
609 out_unlock_fown:
610 read_unlock(&fown->lock);
611}
612
613static void send_sigurg_to_task(struct task_struct *p,
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700614 struct fown_struct *fown, int group)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615{
616 if (sigio_perm(p, fown, SIGURG))
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700617 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618}
619
620int send_sigurg(struct fown_struct *fown)
621{
622 struct task_struct *p;
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700623 enum pid_type type;
624 struct pid *pid;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700625 int group = 1;
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700626 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627
628 read_lock(&fown->lock);
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700629
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700630 type = fown->pid_type;
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700631 if (type == PIDTYPE_MAX) {
632 group = 0;
633 type = PIDTYPE_PID;
634 }
635
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 pid = fown->pid;
637 if (!pid)
638 goto out_unlock_fown;
639
640 ret = 1;
641
642 read_lock(&tasklist_lock);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700643 do_each_pid_task(pid, type, p) {
Peter Zijlstraba0a6c92009-09-23 15:57:03 -0700644 send_sigurg_to_task(p, fown, group);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700645 } while_each_pid_task(pid, type, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 read_unlock(&tasklist_lock);
647 out_unlock_fown:
648 read_unlock(&fown->lock);
649 return ret;
650}
651
Eric Dumazet989a2972010-04-14 09:55:35 +0000652static DEFINE_SPINLOCK(fasync_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800653static struct kmem_cache *fasync_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654
Eric Dumazet989a2972010-04-14 09:55:35 +0000655static void fasync_free_rcu(struct rcu_head *head)
656{
657 kmem_cache_free(fasync_cache,
658 container_of(head, struct fasync_struct, fa_rcu));
659}
660
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661/*
Linus Torvalds53281b62009-12-16 08:23:37 -0800662 * Remove a fasync entry. If successfully removed, return
663 * positive and clear the FASYNC flag. If no entry exists,
664 * do nothing and return 0.
665 *
666 * NOTE! It is very important that the FASYNC flag always
667 * match the state "is the filp on a fasync list".
668 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 */
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400670int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671{
672 struct fasync_struct *fa, **fp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 int result = 0;
674
Jonathan Corbet4a6a4492009-03-27 12:24:31 -0600675 spin_lock(&filp->f_lock);
Eric Dumazet989a2972010-04-14 09:55:35 +0000676 spin_lock(&fasync_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
Linus Torvalds53281b62009-12-16 08:23:37 -0800678 if (fa->fa_file != filp)
679 continue;
Eric Dumazet989a2972010-04-14 09:55:35 +0000680
681 spin_lock_irq(&fa->fa_lock);
682 fa->fa_file = NULL;
683 spin_unlock_irq(&fa->fa_lock);
684
Linus Torvalds53281b62009-12-16 08:23:37 -0800685 *fp = fa->fa_next;
Eric Dumazet989a2972010-04-14 09:55:35 +0000686 call_rcu(&fa->fa_rcu, fasync_free_rcu);
Jonathan Corbet76398422009-02-01 14:26:59 -0700687 filp->f_flags &= ~FASYNC;
Linus Torvalds53281b62009-12-16 08:23:37 -0800688 result = 1;
689 break;
690 }
Eric Dumazet989a2972010-04-14 09:55:35 +0000691 spin_unlock(&fasync_lock);
Jonathan Corbet4a6a4492009-03-27 12:24:31 -0600692 spin_unlock(&filp->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 return result;
694}
695
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400696struct fasync_struct *fasync_alloc(void)
Linus Torvalds53281b62009-12-16 08:23:37 -0800697{
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400698 return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
699}
Linus Torvalds53281b62009-12-16 08:23:37 -0800700
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400701/*
702 * NOTE! This can be used only for unused fasync entries:
703 * entries that actually got inserted on the fasync list
704 * need to be released by rcu - see fasync_remove_entry.
705 */
706void fasync_free(struct fasync_struct *new)
707{
708 kmem_cache_free(fasync_cache, new);
709}
710
711/*
712 * Insert a new entry into the fasync list. Return the pointer to the
713 * old one if we didn't use the new one.
Linus Torvalds55f335a2010-10-27 18:17:02 -0700714 *
715 * NOTE! It is very important that the FASYNC flag always
716 * match the state "is the filp on a fasync list".
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400717 */
718struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
719{
720 struct fasync_struct *fa, **fp;
Linus Torvalds53281b62009-12-16 08:23:37 -0800721
722 spin_lock(&filp->f_lock);
Eric Dumazet989a2972010-04-14 09:55:35 +0000723 spin_lock(&fasync_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800724 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
725 if (fa->fa_file != filp)
726 continue;
Eric Dumazet989a2972010-04-14 09:55:35 +0000727
728 spin_lock_irq(&fa->fa_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800729 fa->fa_fd = fd;
Eric Dumazet989a2972010-04-14 09:55:35 +0000730 spin_unlock_irq(&fa->fa_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800731 goto out;
732 }
733
Eric Dumazet989a2972010-04-14 09:55:35 +0000734 spin_lock_init(&new->fa_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800735 new->magic = FASYNC_MAGIC;
736 new->fa_file = filp;
737 new->fa_fd = fd;
738 new->fa_next = *fapp;
Eric Dumazet989a2972010-04-14 09:55:35 +0000739 rcu_assign_pointer(*fapp, new);
Linus Torvalds53281b62009-12-16 08:23:37 -0800740 filp->f_flags |= FASYNC;
741
742out:
Eric Dumazet989a2972010-04-14 09:55:35 +0000743 spin_unlock(&fasync_lock);
Linus Torvalds53281b62009-12-16 08:23:37 -0800744 spin_unlock(&filp->f_lock);
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400745 return fa;
746}
747
748/*
749 * Add a fasync entry. Return negative on error, positive if
750 * added, and zero if did nothing but change an existing one.
Linus Torvaldsf7347ce2010-10-27 12:38:12 -0400751 */
752static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
753{
754 struct fasync_struct *new;
755
756 new = fasync_alloc();
757 if (!new)
758 return -ENOMEM;
759
760 /*
761 * fasync_insert_entry() returns the old (update) entry if
762 * it existed.
763 *
764 * So free the (unused) new entry and return 0 to let the
765 * caller know that we didn't add any new fasync entries.
766 */
767 if (fasync_insert_entry(fd, filp, fapp, new)) {
768 fasync_free(new);
769 return 0;
770 }
771
772 return 1;
Linus Torvalds53281b62009-12-16 08:23:37 -0800773}
774
775/*
776 * fasync_helper() is used by almost all character device drivers
777 * to set up the fasync queue, and for regular files by the file
778 * lease code. It returns negative on error, 0 if it did no changes
779 * and positive if it added/deleted the entry.
780 */
781int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
782{
783 if (!on)
784 return fasync_remove_entry(filp, fapp);
785 return fasync_add_entry(fd, filp, fapp);
786}
787
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788EXPORT_SYMBOL(fasync_helper);
789
Eric Dumazet989a2972010-04-14 09:55:35 +0000790/*
791 * rcu_read_lock() is held
792 */
793static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794{
795 while (fa) {
Eric Dumazet989a2972010-04-14 09:55:35 +0000796 struct fown_struct *fown;
Andrew Mortonf4985dc2010-06-29 15:05:42 -0700797 unsigned long flags;
798
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 if (fa->magic != FASYNC_MAGIC) {
800 printk(KERN_ERR "kill_fasync: bad magic number in "
801 "fasync_struct!\n");
802 return;
803 }
Andrew Mortonf4985dc2010-06-29 15:05:42 -0700804 spin_lock_irqsave(&fa->fa_lock, flags);
Eric Dumazet989a2972010-04-14 09:55:35 +0000805 if (fa->fa_file) {
806 fown = &fa->fa_file->f_owner;
807 /* Don't send SIGURG to processes which have not set a
808 queued signum: SIGURG has its own default signalling
809 mechanism. */
810 if (!(sig == SIGURG && fown->signum == 0))
811 send_sigio(fown, fa->fa_fd, band);
812 }
Andrew Mortonf4985dc2010-06-29 15:05:42 -0700813 spin_unlock_irqrestore(&fa->fa_lock, flags);
Eric Dumazet989a2972010-04-14 09:55:35 +0000814 fa = rcu_dereference(fa->fa_next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 }
816}
817
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818void kill_fasync(struct fasync_struct **fp, int sig, int band)
819{
820 /* First a quick test without locking: usually
821 * the list is empty.
822 */
823 if (*fp) {
Eric Dumazet989a2972010-04-14 09:55:35 +0000824 rcu_read_lock();
825 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
826 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 }
828}
829EXPORT_SYMBOL(kill_fasync);
830
Wu Fengguang454eedb2010-08-10 18:01:29 -0700831static int __init fcntl_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832{
James Bottomley3ab04d52010-09-09 16:38:12 -0700833 /*
834 * Please add new bits here to ensure allocation uniqueness.
835 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
836 * is defined as O_NONBLOCK on some platforms and not on others.
837 */
Al Viro1abf0c72011-03-13 03:51:11 -0400838 BUILD_BUG_ON(19 - 1 /* for O_RDONLY being 0 */ != HWEIGHT32(
Wu Fengguang454eedb2010-08-10 18:01:29 -0700839 O_RDONLY | O_WRONLY | O_RDWR |
840 O_CREAT | O_EXCL | O_NOCTTY |
James Bottomley3ab04d52010-09-09 16:38:12 -0700841 O_TRUNC | O_APPEND | /* O_NONBLOCK | */
Wu Fengguang454eedb2010-08-10 18:01:29 -0700842 __O_SYNC | O_DSYNC | FASYNC |
843 O_DIRECT | O_LARGEFILE | O_DIRECTORY |
844 O_NOFOLLOW | O_NOATIME | O_CLOEXEC |
Al Viro1abf0c72011-03-13 03:51:11 -0400845 __FMODE_EXEC | O_PATH
Wu Fengguang454eedb2010-08-10 18:01:29 -0700846 ));
847
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 fasync_cache = kmem_cache_create("fasync_cache",
Paul Mundt20c2df82007-07-20 10:11:58 +0900849 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 return 0;
851}
852
Wu Fengguang454eedb2010-08-10 18:01:29 -0700853module_init(fcntl_init)