blob: 1ad703150dee08d9bec6c481f480b99ae7b149d5 [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>
17#include <linux/security.h>
18#include <linux/ptrace.h>
Jesper Juhl7ed20e12005-05-01 08:59:14 -070019#include <linux/signal.h>
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070020#include <linux/rcupdate.h>
Pavel Emelyanovb4888932007-10-18 23:40:14 -070021#include <linux/pid_namespace.h>
Jonathan Corbet218d11a2008-12-05 16:12:48 -070022#include <linux/smp_lock.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 Viro1027abe2008-07-30 04:13:04 -0400134 struct file *file = fget(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))
Satyam Sharma3bd858a2007-07-17 15:00:08 +0530162 if (!is_owner_or_cap(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,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 uid_t uid, uid_t euid, int force)
202{
203 write_lock_irq(&filp->f_owner.lock);
204 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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 filp->f_owner.uid = uid;
209 filp->f_owner.euid = euid;
210 }
211 write_unlock_irq(&filp->f_owner.lock);
212}
213
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700214int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
215 int force)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216{
David Howells86a264a2008-11-14 10:39:18 +1100217 const struct cred *cred = current_cred();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 int err;
219
220 err = security_file_set_fowner(filp);
221 if (err)
222 return err;
223
David Howells86a264a2008-11-14 10:39:18 +1100224 f_modown(filp, pid, type, cred->uid, cred->euid, force);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 return 0;
226}
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700227EXPORT_SYMBOL(__f_setown);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700229int f_setown(struct file *filp, unsigned long arg, int force)
230{
231 enum pid_type type;
232 struct pid *pid;
233 int who = arg;
234 int result;
235 type = PIDTYPE_PID;
236 if (who < 0) {
237 type = PIDTYPE_PGID;
238 who = -who;
239 }
240 rcu_read_lock();
Pavel Emelyanovb4888932007-10-18 23:40:14 -0700241 pid = find_vpid(who);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700242 result = __f_setown(filp, pid, type, force);
243 rcu_read_unlock();
244 return result;
245}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246EXPORT_SYMBOL(f_setown);
247
248void f_delown(struct file *filp)
249{
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700250 f_modown(filp, NULL, PIDTYPE_PID, 0, 0, 1);
251}
252
253pid_t f_getown(struct file *filp)
254{
255 pid_t pid;
Eric W. Biederman43fa1ad2006-10-02 02:17:27 -0700256 read_lock(&filp->f_owner.lock);
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -0800257 pid = pid_vnr(filp->f_owner.pid);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700258 if (filp->f_owner.pid_type == PIDTYPE_PGID)
259 pid = -pid;
Eric W. Biederman43fa1ad2006-10-02 02:17:27 -0700260 read_unlock(&filp->f_owner.lock);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700261 return pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262}
263
264static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
265 struct file *filp)
266{
267 long err = -EINVAL;
268
269 switch (cmd) {
270 case F_DUPFD:
Ulrich Drepper22d2b352007-10-16 23:30:26 -0700271 case F_DUPFD_CLOEXEC:
Al Viro4e1e0182008-07-26 16:01:20 -0400272 if (arg >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
273 break;
Al Viro1027abe2008-07-30 04:13:04 -0400274 err = alloc_fd(arg, cmd == F_DUPFD_CLOEXEC ? O_CLOEXEC : 0);
275 if (err >= 0) {
276 get_file(filp);
277 fd_install(err, filp);
278 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 break;
280 case F_GETFD:
281 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
282 break;
283 case F_SETFD:
284 err = 0;
285 set_close_on_exec(fd, arg & FD_CLOEXEC);
286 break;
287 case F_GETFL:
288 err = filp->f_flags;
289 break;
290 case F_SETFL:
291 err = setfl(fd, filp, arg);
292 break;
293 case F_GETLK:
294 err = fcntl_getlk(filp, (struct flock __user *) arg);
295 break;
296 case F_SETLK:
297 case F_SETLKW:
Peter Staubachc2936212005-07-27 11:45:09 -0700298 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 break;
300 case F_GETOWN:
301 /*
302 * XXX If f_owner is a process group, the
303 * negative return value will get converted
304 * into an error. Oops. If we keep the
305 * current syscall conventions, the only way
306 * to fix this will be in libc.
307 */
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700308 err = f_getown(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 force_successful_syscall_return();
310 break;
311 case F_SETOWN:
312 err = f_setown(filp, arg, 1);
313 break;
314 case F_GETSIG:
315 err = filp->f_owner.signum;
316 break;
317 case F_SETSIG:
318 /* arg == 0 restores default behaviour. */
Jesper Juhl7ed20e12005-05-01 08:59:14 -0700319 if (!valid_signal(arg)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 break;
321 }
322 err = 0;
323 filp->f_owner.signum = arg;
324 break;
325 case F_GETLEASE:
326 err = fcntl_getlease(filp);
327 break;
328 case F_SETLEASE:
329 err = fcntl_setlease(fd, filp, arg);
330 break;
331 case F_NOTIFY:
332 err = fcntl_dirnotify(fd, filp, arg);
333 break;
334 default:
335 break;
336 }
337 return err;
338}
339
Heiko Carstensa26eab22009-01-14 14:14:17 +0100340SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341{
342 struct file *filp;
343 long err = -EBADF;
344
345 filp = fget(fd);
346 if (!filp)
347 goto out;
348
349 err = security_file_fcntl(filp, cmd, arg);
350 if (err) {
351 fput(filp);
352 return err;
353 }
354
355 err = do_fcntl(fd, cmd, arg, filp);
356
357 fput(filp);
358out:
359 return err;
360}
361
362#if BITS_PER_LONG == 32
Heiko Carstensa26eab22009-01-14 14:14:17 +0100363SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
364 unsigned long, arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365{
366 struct file * filp;
367 long err;
368
369 err = -EBADF;
370 filp = fget(fd);
371 if (!filp)
372 goto out;
373
374 err = security_file_fcntl(filp, cmd, arg);
375 if (err) {
376 fput(filp);
377 return err;
378 }
379 err = -EBADF;
380
381 switch (cmd) {
382 case F_GETLK64:
383 err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
384 break;
385 case F_SETLK64:
386 case F_SETLKW64:
Peter Staubachc2936212005-07-27 11:45:09 -0700387 err = fcntl_setlk64(fd, filp, cmd,
388 (struct flock64 __user *) arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 break;
390 default:
391 err = do_fcntl(fd, cmd, arg, filp);
392 break;
393 }
394 fput(filp);
395out:
396 return err;
397}
398#endif
399
400/* Table to convert sigio signal codes into poll band bitmaps */
401
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800402static const long band_table[NSIGPOLL] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 POLLIN | POLLRDNORM, /* POLL_IN */
404 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
405 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
406 POLLERR, /* POLL_ERR */
407 POLLPRI | POLLRDBAND, /* POLL_PRI */
408 POLLHUP | POLLERR /* POLL_HUP */
409};
410
411static inline int sigio_perm(struct task_struct *p,
412 struct fown_struct *fown, int sig)
413{
David Howellsc69e8d92008-11-14 10:39:19 +1100414 const struct cred *cred;
415 int ret;
416
417 rcu_read_lock();
418 cred = __task_cred(p);
419 ret = ((fown->euid == 0 ||
420 fown->euid == cred->suid || fown->euid == cred->uid ||
421 fown->uid == cred->suid || fown->uid == cred->uid) &&
422 !security_file_send_sigiotask(p, fown, sig));
423 rcu_read_unlock();
424 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425}
426
427static void send_sigio_to_task(struct task_struct *p,
428 struct fown_struct *fown,
429 int fd,
430 int reason)
431{
432 if (!sigio_perm(p, fown, fown->signum))
433 return;
434
435 switch (fown->signum) {
436 siginfo_t si;
437 default:
438 /* Queue a rt signal with the appropriate fd as its
439 value. We use SI_SIGIO as the source, not
440 SI_KERNEL, since kernel signals always get
441 delivered even if we can't queue. Failure to
442 queue in this case _should_ be reported; we fall
443 back to SIGIO in that case. --sct */
444 si.si_signo = fown->signum;
445 si.si_errno = 0;
446 si.si_code = reason;
447 /* Make sure we are called with one of the POLL_*
448 reasons, otherwise we could leak kernel stack into
449 userspace. */
Eric Sesterhennf6298aa2006-04-02 13:37:19 +0200450 BUG_ON((reason & __SI_MASK) != __SI_POLL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 if (reason - POLL_IN >= NSIGPOLL)
452 si.si_band = ~0L;
453 else
454 si.si_band = band_table[reason - POLL_IN];
455 si.si_fd = fd;
Oleg Nesterov850d6fb2006-01-08 01:03:29 -0800456 if (!group_send_sig_info(fown->signum, &si, p))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 break;
458 /* fall-through: fall back on the old plain SIGIO signal */
459 case 0:
Oleg Nesterov850d6fb2006-01-08 01:03:29 -0800460 group_send_sig_info(SIGIO, SEND_SIG_PRIV, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 }
462}
463
464void send_sigio(struct fown_struct *fown, int fd, int band)
465{
466 struct task_struct *p;
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700467 enum pid_type type;
468 struct pid *pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469
470 read_lock(&fown->lock);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700471 type = fown->pid_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 pid = fown->pid;
473 if (!pid)
474 goto out_unlock_fown;
475
476 read_lock(&tasklist_lock);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700477 do_each_pid_task(pid, type, p) {
478 send_sigio_to_task(p, fown, fd, band);
479 } while_each_pid_task(pid, type, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 read_unlock(&tasklist_lock);
481 out_unlock_fown:
482 read_unlock(&fown->lock);
483}
484
485static void send_sigurg_to_task(struct task_struct *p,
486 struct fown_struct *fown)
487{
488 if (sigio_perm(p, fown, SIGURG))
Oleg Nesterov850d6fb2006-01-08 01:03:29 -0800489 group_send_sig_info(SIGURG, SEND_SIG_PRIV, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490}
491
492int send_sigurg(struct fown_struct *fown)
493{
494 struct task_struct *p;
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700495 enum pid_type type;
496 struct pid *pid;
497 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498
499 read_lock(&fown->lock);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700500 type = fown->pid_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 pid = fown->pid;
502 if (!pid)
503 goto out_unlock_fown;
504
505 ret = 1;
506
507 read_lock(&tasklist_lock);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700508 do_each_pid_task(pid, type, p) {
509 send_sigurg_to_task(p, fown);
510 } while_each_pid_task(pid, type, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 read_unlock(&tasklist_lock);
512 out_unlock_fown:
513 read_unlock(&fown->lock);
514 return ret;
515}
516
517static DEFINE_RWLOCK(fasync_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800518static struct kmem_cache *fasync_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519
520/*
Jonathan Corbet76398422009-02-01 14:26:59 -0700521 * fasync_helper() is used by almost all character device drivers
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 * to set up the fasync queue. It returns negative on error, 0 if it did
523 * no changes and positive if it added/deleted the entry.
524 */
525int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
526{
527 struct fasync_struct *fa, **fp;
528 struct fasync_struct *new = NULL;
529 int result = 0;
530
531 if (on) {
Christoph Lametere94b1762006-12-06 20:33:17 -0800532 new = kmem_cache_alloc(fasync_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 if (!new)
534 return -ENOMEM;
535 }
Jonathan Corbet4a6a4492009-03-27 12:24:31 -0600536
537 /*
538 * We need to take f_lock first since it's not an IRQ-safe
539 * lock.
540 */
541 spin_lock(&filp->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 write_lock_irq(&fasync_lock);
543 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
544 if (fa->fa_file == filp) {
545 if(on) {
546 fa->fa_fd = fd;
547 kmem_cache_free(fasync_cache, new);
548 } else {
549 *fp = fa->fa_next;
550 kmem_cache_free(fasync_cache, fa);
551 result = 1;
552 }
553 goto out;
554 }
555 }
556
557 if (on) {
558 new->magic = FASYNC_MAGIC;
559 new->fa_file = filp;
560 new->fa_fd = fd;
561 new->fa_next = *fapp;
562 *fapp = new;
563 result = 1;
564 }
565out:
Jonathan Corbet76398422009-02-01 14:26:59 -0700566 if (on)
567 filp->f_flags |= FASYNC;
568 else
569 filp->f_flags &= ~FASYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 write_unlock_irq(&fasync_lock);
Jonathan Corbet4a6a4492009-03-27 12:24:31 -0600571 spin_unlock(&filp->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 return result;
573}
574
575EXPORT_SYMBOL(fasync_helper);
576
577void __kill_fasync(struct fasync_struct *fa, int sig, int band)
578{
579 while (fa) {
580 struct fown_struct * fown;
581 if (fa->magic != FASYNC_MAGIC) {
582 printk(KERN_ERR "kill_fasync: bad magic number in "
583 "fasync_struct!\n");
584 return;
585 }
586 fown = &fa->fa_file->f_owner;
587 /* Don't send SIGURG to processes which have not set a
588 queued signum: SIGURG has its own default signalling
589 mechanism. */
590 if (!(sig == SIGURG && fown->signum == 0))
591 send_sigio(fown, fa->fa_fd, band);
592 fa = fa->fa_next;
593 }
594}
595
596EXPORT_SYMBOL(__kill_fasync);
597
598void kill_fasync(struct fasync_struct **fp, int sig, int band)
599{
600 /* First a quick test without locking: usually
601 * the list is empty.
602 */
603 if (*fp) {
604 read_lock(&fasync_lock);
605 /* reread *fp after obtaining the lock */
606 __kill_fasync(*fp, sig, band);
607 read_unlock(&fasync_lock);
608 }
609}
610EXPORT_SYMBOL(kill_fasync);
611
612static int __init fasync_init(void)
613{
614 fasync_cache = kmem_cache_create("fasync_cache",
Paul Mundt20c2df82007-07-20 10:11:58 +0900615 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 return 0;
617}
618
619module_init(fasync_init)