]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - fs/fcntl.c
BKL: Remove BKL from HFS
[mirror_ubuntu-artful-kernel.git] / fs / fcntl.c
CommitLineData
1da177e4
LT
1/*
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>
9f3acc31 12#include <linux/fdtable.h>
16f7e0fe 13#include <linux/capability.h>
1da177e4 14#include <linux/dnotify.h>
1da177e4
LT
15#include <linux/slab.h>
16#include <linux/module.h>
35f3d14d 17#include <linux/pipe_fs_i.h>
1da177e4
LT
18#include <linux/security.h>
19#include <linux/ptrace.h>
7ed20e1a 20#include <linux/signal.h>
ab2af1f5 21#include <linux/rcupdate.h>
b488893a 22#include <linux/pid_namespace.h>
1da177e4
LT
23
24#include <asm/poll.h>
25#include <asm/siginfo.h>
26#include <asm/uaccess.h>
27
fc9b52cd 28void set_close_on_exec(unsigned int fd, int flag)
1da177e4
LT
29{
30 struct files_struct *files = current->files;
badf1662 31 struct fdtable *fdt;
1da177e4 32 spin_lock(&files->file_lock);
badf1662 33 fdt = files_fdtable(files);
1da177e4 34 if (flag)
badf1662 35 FD_SET(fd, fdt->close_on_exec);
1da177e4 36 else
badf1662 37 FD_CLR(fd, fdt->close_on_exec);
1da177e4
LT
38 spin_unlock(&files->file_lock);
39}
40
858119e1 41static int get_close_on_exec(unsigned int fd)
1da177e4
LT
42{
43 struct files_struct *files = current->files;
badf1662 44 struct fdtable *fdt;
1da177e4 45 int res;
b835996f 46 rcu_read_lock();
badf1662
DS
47 fdt = files_fdtable(files);
48 res = FD_ISSET(fd, fdt->close_on_exec);
b835996f 49 rcu_read_unlock();
1da177e4
LT
50 return res;
51}
52
a26eab24 53SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
1da177e4
LT
54{
55 int err = -EBADF;
56 struct file * file, *tofree;
57 struct files_struct * files = current->files;
badf1662 58 struct fdtable *fdt;
1da177e4 59
336dd1f7
UD
60 if ((flags & ~O_CLOEXEC) != 0)
61 return -EINVAL;
62
6c5d0512
AV
63 if (unlikely(oldfd == newfd))
64 return -EINVAL;
65
1da177e4 66 spin_lock(&files->file_lock);
1da177e4 67 err = expand_files(files, newfd);
1b7e190b
AV
68 file = fcheck(oldfd);
69 if (unlikely(!file))
70 goto Ebadf;
4e1e018e
AV
71 if (unlikely(err < 0)) {
72 if (err == -EMFILE)
1b7e190b
AV
73 goto Ebadf;
74 goto out_unlock;
4e1e018e 75 }
1b7e190b
AV
76 /*
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 */
1da177e4 90 err = -EBUSY;
badf1662
DS
91 fdt = files_fdtable(files);
92 tofree = fdt->fd[newfd];
93 if (!tofree && FD_ISSET(newfd, fdt->open_fds))
1b7e190b
AV
94 goto out_unlock;
95 get_file(file);
ab2af1f5 96 rcu_assign_pointer(fdt->fd[newfd], file);
badf1662 97 FD_SET(newfd, fdt->open_fds);
336dd1f7
UD
98 if (flags & O_CLOEXEC)
99 FD_SET(newfd, fdt->close_on_exec);
100 else
101 FD_CLR(newfd, fdt->close_on_exec);
1da177e4
LT
102 spin_unlock(&files->file_lock);
103
104 if (tofree)
105 filp_close(tofree, files);
1da177e4 106
1b7e190b
AV
107 return newfd;
108
109Ebadf:
110 err = -EBADF;
111out_unlock:
1da177e4 112 spin_unlock(&files->file_lock);
1b7e190b 113 return err;
1da177e4 114}
336dd1f7 115
a26eab24 116SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
336dd1f7 117{
6c5d0512
AV
118 if (unlikely(newfd == oldfd)) { /* corner case */
119 struct files_struct *files = current->files;
2b79bc4f
JM
120 int retval = oldfd;
121
6c5d0512
AV
122 rcu_read_lock();
123 if (!fcheck_files(files, oldfd))
2b79bc4f 124 retval = -EBADF;
6c5d0512 125 rcu_read_unlock();
2b79bc4f 126 return retval;
6c5d0512 127 }
336dd1f7
UD
128 return sys_dup3(oldfd, newfd, 0);
129}
1da177e4 130
a26eab24 131SYSCALL_DEFINE1(dup, unsigned int, fildes)
1da177e4
LT
132{
133 int ret = -EBADF;
1027abe8
AV
134 struct file *file = fget(fildes);
135
136 if (file) {
137 ret = get_unused_fd();
138 if (ret >= 0)
139 fd_install(ret, file);
140 else
141 fput(file);
142 }
1da177e4
LT
143 return ret;
144}
145
76398425 146#define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
1da177e4
LT
147
148static int setfl(int fd, struct file * filp, unsigned long arg)
149{
0f7fc9e4 150 struct inode * inode = filp->f_path.dentry->d_inode;
1da177e4
LT
151 int error = 0;
152
7d95c8f2 153 /*
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))
1da177e4
LT
158 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))
3bd858ab 162 if (!is_owner_or_cap(inode))
1da177e4
LT
163 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
218d11a8 181 /*
76398425 182 * ->fasync() is responsible for setting the FASYNC bit.
218d11a8 183 */
76398425
JC
184 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;
60aa4924
JC
189 if (error > 0)
190 error = 0;
1da177e4 191 }
db1dd4d3 192 spin_lock(&filp->f_lock);
1da177e4 193 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
db1dd4d3 194 spin_unlock(&filp->f_lock);
76398425 195
1da177e4 196 out:
1da177e4
LT
197 return error;
198}
199
609d7fa9 200static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
2f38d70f 201 int force)
1da177e4 202{
80e1e823 203 write_lock_irq(&filp->f_owner.lock);
1da177e4 204 if (force || !filp->f_owner.pid) {
609d7fa9
EB
205 put_pid(filp->f_owner.pid);
206 filp->f_owner.pid = get_pid(pid);
207 filp->f_owner.pid_type = type;
2f38d70f
ON
208
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 }
1da177e4 214 }
80e1e823 215 write_unlock_irq(&filp->f_owner.lock);
1da177e4
LT
216}
217
609d7fa9
EB
218int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
219 int force)
1da177e4
LT
220{
221 int err;
2f38d70f 222
1da177e4
LT
223 err = security_file_set_fowner(filp);
224 if (err)
225 return err;
226
2f38d70f 227 f_modown(filp, pid, type, force);
1da177e4
LT
228 return 0;
229}
609d7fa9 230EXPORT_SYMBOL(__f_setown);
1da177e4 231
609d7fa9
EB
232int 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();
b488893a 244 pid = find_vpid(who);
609d7fa9
EB
245 result = __f_setown(filp, pid, type, force);
246 rcu_read_unlock();
247 return result;
248}
1da177e4
LT
249EXPORT_SYMBOL(f_setown);
250
251void f_delown(struct file *filp)
252{
2f38d70f 253 f_modown(filp, NULL, PIDTYPE_PID, 1);
609d7fa9
EB
254}
255
256pid_t f_getown(struct file *filp)
257{
258 pid_t pid;
43fa1adb 259 read_lock(&filp->f_owner.lock);
6c5f3e7b 260 pid = pid_vnr(filp->f_owner.pid);
609d7fa9
EB
261 if (filp->f_owner.pid_type == PIDTYPE_PGID)
262 pid = -pid;
43fa1adb 263 read_unlock(&filp->f_owner.lock);
609d7fa9 264 return pid;
1da177e4
LT
265}
266
ba0a6c9f
PZ
267static 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)
5b54470d 277 return -EFAULT;
ba0a6c9f
PZ
278
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
978b4053 288 case F_OWNER_PGRP:
ba0a6c9f
PZ
289 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:
978b4053 325 owner.type = F_OWNER_PGRP;
ba0a6c9f
PZ
326 break;
327
328 default:
329 WARN_ON(1);
330 ret = -EINVAL;
331 break;
332 }
333 read_unlock(&filp->f_owner.lock);
334
5b54470d 335 if (!ret) {
ba0a6c9f 336 ret = copy_to_user(owner_p, &owner, sizeof(owner));
5b54470d
DC
337 if (ret)
338 ret = -EFAULT;
339 }
ba0a6c9f
PZ
340 return ret;
341}
342
1da177e4
LT
343static 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:
22d2b35b 350 case F_DUPFD_CLOEXEC:
d554ed89 351 if (arg >= rlimit(RLIMIT_NOFILE))
4e1e018e 352 break;
1027abe8
AV
353 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 }
1da177e4
LT
358 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:
c293621b 377 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
1da177e4
LT
378 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 */
609d7fa9 387 err = f_getown(filp);
1da177e4
LT
388 force_successful_syscall_return();
389 break;
390 case F_SETOWN:
391 err = f_setown(filp, arg, 1);
392 break;
ba0a6c9f
PZ
393 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;
1da177e4
LT
399 case F_GETSIG:
400 err = filp->f_owner.signum;
401 break;
402 case F_SETSIG:
403 /* arg == 0 restores default behaviour. */
7ed20e1a 404 if (!valid_signal(arg)) {
1da177e4
LT
405 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;
35f3d14d
JA
419 case F_SETPIPE_SZ:
420 case F_GETPIPE_SZ:
421 err = pipe_fcntl(filp, cmd, arg);
422 break;
1da177e4
LT
423 default:
424 break;
425 }
426 return err;
427}
428
a26eab24 429SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
1da177e4
LT
430{
431 struct file *filp;
432 long err = -EBADF;
433
434 filp = fget(fd);
435 if (!filp)
436 goto out;
437
438 err = security_file_fcntl(filp, cmd, arg);
439 if (err) {
440 fput(filp);
441 return err;
442 }
443
444 err = do_fcntl(fd, cmd, arg, filp);
445
446 fput(filp);
447out:
448 return err;
449}
450
451#if BITS_PER_LONG == 32
a26eab24
HC
452SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
453 unsigned long, arg)
1da177e4
LT
454{
455 struct file * filp;
456 long err;
457
458 err = -EBADF;
459 filp = fget(fd);
460 if (!filp)
461 goto out;
462
463 err = security_file_fcntl(filp, cmd, arg);
464 if (err) {
465 fput(filp);
466 return err;
467 }
468 err = -EBADF;
469
470 switch (cmd) {
471 case F_GETLK64:
472 err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
473 break;
474 case F_SETLK64:
475 case F_SETLKW64:
c293621b
PS
476 err = fcntl_setlk64(fd, filp, cmd,
477 (struct flock64 __user *) arg);
1da177e4
LT
478 break;
479 default:
480 err = do_fcntl(fd, cmd, arg, filp);
481 break;
482 }
483 fput(filp);
484out:
485 return err;
486}
487#endif
488
489/* Table to convert sigio signal codes into poll band bitmaps */
490
fa3536cc 491static const long band_table[NSIGPOLL] = {
1da177e4
LT
492 POLLIN | POLLRDNORM, /* POLL_IN */
493 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
494 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
495 POLLERR, /* POLL_ERR */
496 POLLPRI | POLLRDBAND, /* POLL_PRI */
497 POLLHUP | POLLERR /* POLL_HUP */
498};
499
500static inline int sigio_perm(struct task_struct *p,
501 struct fown_struct *fown, int sig)
502{
c69e8d9c
DH
503 const struct cred *cred;
504 int ret;
505
506 rcu_read_lock();
507 cred = __task_cred(p);
508 ret = ((fown->euid == 0 ||
509 fown->euid == cred->suid || fown->euid == cred->uid ||
510 fown->uid == cred->suid || fown->uid == cred->uid) &&
511 !security_file_send_sigiotask(p, fown, sig));
512 rcu_read_unlock();
513 return ret;
1da177e4
LT
514}
515
516static void send_sigio_to_task(struct task_struct *p,
8eeee4e2 517 struct fown_struct *fown,
ba0a6c9f 518 int fd, int reason, int group)
1da177e4 519{
8eeee4e2
ON
520 /*
521 * F_SETSIG can change ->signum lockless in parallel, make
522 * sure we read it once and use the same value throughout.
523 */
524 int signum = ACCESS_ONCE(fown->signum);
525
526 if (!sigio_perm(p, fown, signum))
1da177e4
LT
527 return;
528
8eeee4e2 529 switch (signum) {
1da177e4
LT
530 siginfo_t si;
531 default:
532 /* Queue a rt signal with the appropriate fd as its
533 value. We use SI_SIGIO as the source, not
534 SI_KERNEL, since kernel signals always get
535 delivered even if we can't queue. Failure to
536 queue in this case _should_ be reported; we fall
537 back to SIGIO in that case. --sct */
8eeee4e2 538 si.si_signo = signum;
1da177e4
LT
539 si.si_errno = 0;
540 si.si_code = reason;
541 /* Make sure we are called with one of the POLL_*
542 reasons, otherwise we could leak kernel stack into
543 userspace. */
f6298aab 544 BUG_ON((reason & __SI_MASK) != __SI_POLL);
1da177e4
LT
545 if (reason - POLL_IN >= NSIGPOLL)
546 si.si_band = ~0L;
547 else
548 si.si_band = band_table[reason - POLL_IN];
549 si.si_fd = fd;
ba0a6c9f 550 if (!do_send_sig_info(signum, &si, p, group))
1da177e4
LT
551 break;
552 /* fall-through: fall back on the old plain SIGIO signal */
553 case 0:
ba0a6c9f 554 do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
1da177e4
LT
555 }
556}
557
558void send_sigio(struct fown_struct *fown, int fd, int band)
559{
560 struct task_struct *p;
609d7fa9
EB
561 enum pid_type type;
562 struct pid *pid;
ba0a6c9f 563 int group = 1;
1da177e4
LT
564
565 read_lock(&fown->lock);
ba0a6c9f 566
609d7fa9 567 type = fown->pid_type;
ba0a6c9f
PZ
568 if (type == PIDTYPE_MAX) {
569 group = 0;
570 type = PIDTYPE_PID;
571 }
572
1da177e4
LT
573 pid = fown->pid;
574 if (!pid)
575 goto out_unlock_fown;
576
577 read_lock(&tasklist_lock);
609d7fa9 578 do_each_pid_task(pid, type, p) {
ba0a6c9f 579 send_sigio_to_task(p, fown, fd, band, group);
609d7fa9 580 } while_each_pid_task(pid, type, p);
1da177e4
LT
581 read_unlock(&tasklist_lock);
582 out_unlock_fown:
583 read_unlock(&fown->lock);
584}
585
586static void send_sigurg_to_task(struct task_struct *p,
ba0a6c9f 587 struct fown_struct *fown, int group)
1da177e4
LT
588{
589 if (sigio_perm(p, fown, SIGURG))
ba0a6c9f 590 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
1da177e4
LT
591}
592
593int send_sigurg(struct fown_struct *fown)
594{
595 struct task_struct *p;
609d7fa9
EB
596 enum pid_type type;
597 struct pid *pid;
ba0a6c9f 598 int group = 1;
609d7fa9 599 int ret = 0;
1da177e4
LT
600
601 read_lock(&fown->lock);
ba0a6c9f 602
609d7fa9 603 type = fown->pid_type;
ba0a6c9f
PZ
604 if (type == PIDTYPE_MAX) {
605 group = 0;
606 type = PIDTYPE_PID;
607 }
608
1da177e4
LT
609 pid = fown->pid;
610 if (!pid)
611 goto out_unlock_fown;
612
613 ret = 1;
614
615 read_lock(&tasklist_lock);
609d7fa9 616 do_each_pid_task(pid, type, p) {
ba0a6c9f 617 send_sigurg_to_task(p, fown, group);
609d7fa9 618 } while_each_pid_task(pid, type, p);
1da177e4
LT
619 read_unlock(&tasklist_lock);
620 out_unlock_fown:
621 read_unlock(&fown->lock);
622 return ret;
623}
624
989a2979 625static DEFINE_SPINLOCK(fasync_lock);
e18b890b 626static struct kmem_cache *fasync_cache __read_mostly;
1da177e4 627
989a2979
ED
628static void fasync_free_rcu(struct rcu_head *head)
629{
630 kmem_cache_free(fasync_cache,
631 container_of(head, struct fasync_struct, fa_rcu));
632}
633
1da177e4 634/*
53281b6d
LT
635 * Remove a fasync entry. If successfully removed, return
636 * positive and clear the FASYNC flag. If no entry exists,
637 * do nothing and return 0.
638 *
639 * NOTE! It is very important that the FASYNC flag always
640 * match the state "is the filp on a fasync list".
641 *
1da177e4 642 */
53281b6d 643static int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
1da177e4
LT
644{
645 struct fasync_struct *fa, **fp;
1da177e4
LT
646 int result = 0;
647
53281b6d 648 spin_lock(&filp->f_lock);
989a2979 649 spin_lock(&fasync_lock);
53281b6d
LT
650 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
651 if (fa->fa_file != filp)
652 continue;
989a2979
ED
653
654 spin_lock_irq(&fa->fa_lock);
655 fa->fa_file = NULL;
656 spin_unlock_irq(&fa->fa_lock);
657
53281b6d 658 *fp = fa->fa_next;
989a2979 659 call_rcu(&fa->fa_rcu, fasync_free_rcu);
53281b6d
LT
660 filp->f_flags &= ~FASYNC;
661 result = 1;
662 break;
1da177e4 663 }
989a2979 664 spin_unlock(&fasync_lock);
53281b6d
LT
665 spin_unlock(&filp->f_lock);
666 return result;
667}
668
669/*
670 * Add a fasync entry. Return negative on error, positive if
671 * added, and zero if did nothing but change an existing one.
672 *
673 * NOTE! It is very important that the FASYNC flag always
674 * match the state "is the filp on a fasync list".
675 */
676static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
677{
678 struct fasync_struct *new, *fa, **fp;
679 int result = 0;
680
681 new = kmem_cache_alloc(fasync_cache, GFP_KERNEL);
682 if (!new)
683 return -ENOMEM;
4a6a4499 684
4a6a4499 685 spin_lock(&filp->f_lock);
989a2979 686 spin_lock(&fasync_lock);
1da177e4 687 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
53281b6d
LT
688 if (fa->fa_file != filp)
689 continue;
989a2979
ED
690
691 spin_lock_irq(&fa->fa_lock);
53281b6d 692 fa->fa_fd = fd;
989a2979
ED
693 spin_unlock_irq(&fa->fa_lock);
694
53281b6d
LT
695 kmem_cache_free(fasync_cache, new);
696 goto out;
1da177e4
LT
697 }
698
989a2979 699 spin_lock_init(&new->fa_lock);
53281b6d
LT
700 new->magic = FASYNC_MAGIC;
701 new->fa_file = filp;
702 new->fa_fd = fd;
703 new->fa_next = *fapp;
989a2979 704 rcu_assign_pointer(*fapp, new);
53281b6d
LT
705 result = 1;
706 filp->f_flags |= FASYNC;
707
1da177e4 708out:
989a2979 709 spin_unlock(&fasync_lock);
4a6a4499 710 spin_unlock(&filp->f_lock);
1da177e4
LT
711 return result;
712}
713
53281b6d
LT
714/*
715 * fasync_helper() is used by almost all character device drivers
716 * to set up the fasync queue, and for regular files by the file
717 * lease code. It returns negative on error, 0 if it did no changes
718 * and positive if it added/deleted the entry.
719 */
720int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
721{
722 if (!on)
723 return fasync_remove_entry(filp, fapp);
724 return fasync_add_entry(fd, filp, fapp);
725}
726
1da177e4
LT
727EXPORT_SYMBOL(fasync_helper);
728
989a2979
ED
729/*
730 * rcu_read_lock() is held
731 */
732static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
1da177e4
LT
733{
734 while (fa) {
989a2979 735 struct fown_struct *fown;
f4985dc7
AM
736 unsigned long flags;
737
1da177e4
LT
738 if (fa->magic != FASYNC_MAGIC) {
739 printk(KERN_ERR "kill_fasync: bad magic number in "
740 "fasync_struct!\n");
741 return;
742 }
f4985dc7 743 spin_lock_irqsave(&fa->fa_lock, flags);
989a2979
ED
744 if (fa->fa_file) {
745 fown = &fa->fa_file->f_owner;
746 /* Don't send SIGURG to processes which have not set a
747 queued signum: SIGURG has its own default signalling
748 mechanism. */
749 if (!(sig == SIGURG && fown->signum == 0))
750 send_sigio(fown, fa->fa_fd, band);
751 }
f4985dc7 752 spin_unlock_irqrestore(&fa->fa_lock, flags);
989a2979 753 fa = rcu_dereference(fa->fa_next);
1da177e4
LT
754 }
755}
756
1da177e4
LT
757void kill_fasync(struct fasync_struct **fp, int sig, int band)
758{
759 /* First a quick test without locking: usually
760 * the list is empty.
761 */
762 if (*fp) {
989a2979
ED
763 rcu_read_lock();
764 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
765 rcu_read_unlock();
1da177e4
LT
766 }
767}
768EXPORT_SYMBOL(kill_fasync);
769
454eedb8 770static int __init fcntl_init(void)
1da177e4 771{
3ab04d5c
JB
772 /*
773 * Please add new bits here to ensure allocation uniqueness.
774 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
775 * is defined as O_NONBLOCK on some platforms and not on others.
776 */
777 BUILD_BUG_ON(18 - 1 /* for O_RDONLY being 0 */ != HWEIGHT32(
454eedb8
WF
778 O_RDONLY | O_WRONLY | O_RDWR |
779 O_CREAT | O_EXCL | O_NOCTTY |
3ab04d5c 780 O_TRUNC | O_APPEND | /* O_NONBLOCK | */
454eedb8
WF
781 __O_SYNC | O_DSYNC | FASYNC |
782 O_DIRECT | O_LARGEFILE | O_DIRECTORY |
783 O_NOFOLLOW | O_NOATIME | O_CLOEXEC |
784 FMODE_EXEC
785 ));
786
1da177e4 787 fasync_cache = kmem_cache_create("fasync_cache",
20c2df83 788 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
1da177e4
LT
789 return 0;
790}
791
454eedb8 792module_init(fcntl_init)