Root/
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> |
12 | #include <linux/fdtable.h> |
13 | #include <linux/capability.h> |
14 | #include <linux/dnotify.h> |
15 | #include <linux/slab.h> |
16 | #include <linux/module.h> |
17 | #include <linux/pipe_fs_i.h> |
18 | #include <linux/security.h> |
19 | #include <linux/ptrace.h> |
20 | #include <linux/signal.h> |
21 | #include <linux/rcupdate.h> |
22 | #include <linux/pid_namespace.h> |
23 | |
24 | #include <asm/poll.h> |
25 | #include <asm/siginfo.h> |
26 | #include <asm/uaccess.h> |
27 | |
28 | void set_close_on_exec(unsigned int fd, int flag) |
29 | { |
30 | struct files_struct *files = current->files; |
31 | struct fdtable *fdt; |
32 | spin_lock(&files->file_lock); |
33 | fdt = files_fdtable(files); |
34 | if (flag) |
35 | FD_SET(fd, fdt->close_on_exec); |
36 | else |
37 | FD_CLR(fd, fdt->close_on_exec); |
38 | spin_unlock(&files->file_lock); |
39 | } |
40 | |
41 | static int get_close_on_exec(unsigned int fd) |
42 | { |
43 | struct files_struct *files = current->files; |
44 | struct fdtable *fdt; |
45 | int res; |
46 | rcu_read_lock(); |
47 | fdt = files_fdtable(files); |
48 | res = FD_ISSET(fd, fdt->close_on_exec); |
49 | rcu_read_unlock(); |
50 | return res; |
51 | } |
52 | |
53 | SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags) |
54 | { |
55 | int err = -EBADF; |
56 | struct file * file, *tofree; |
57 | struct files_struct * files = current->files; |
58 | struct fdtable *fdt; |
59 | |
60 | if ((flags & ~O_CLOEXEC) != 0) |
61 | return -EINVAL; |
62 | |
63 | if (unlikely(oldfd == newfd)) |
64 | return -EINVAL; |
65 | |
66 | spin_lock(&files->file_lock); |
67 | err = expand_files(files, newfd); |
68 | file = fcheck(oldfd); |
69 | if (unlikely(!file)) |
70 | goto Ebadf; |
71 | if (unlikely(err < 0)) { |
72 | if (err == -EMFILE) |
73 | goto Ebadf; |
74 | goto out_unlock; |
75 | } |
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 | */ |
90 | err = -EBUSY; |
91 | fdt = files_fdtable(files); |
92 | tofree = fdt->fd[newfd]; |
93 | if (!tofree && FD_ISSET(newfd, fdt->open_fds)) |
94 | goto out_unlock; |
95 | get_file(file); |
96 | rcu_assign_pointer(fdt->fd[newfd], file); |
97 | FD_SET(newfd, fdt->open_fds); |
98 | if (flags & O_CLOEXEC) |
99 | FD_SET(newfd, fdt->close_on_exec); |
100 | else |
101 | FD_CLR(newfd, fdt->close_on_exec); |
102 | spin_unlock(&files->file_lock); |
103 | |
104 | if (tofree) |
105 | filp_close(tofree, files); |
106 | |
107 | return newfd; |
108 | |
109 | Ebadf: |
110 | err = -EBADF; |
111 | out_unlock: |
112 | spin_unlock(&files->file_lock); |
113 | return err; |
114 | } |
115 | |
116 | SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd) |
117 | { |
118 | if (unlikely(newfd == oldfd)) { /* corner case */ |
119 | struct files_struct *files = current->files; |
120 | int retval = oldfd; |
121 | |
122 | rcu_read_lock(); |
123 | if (!fcheck_files(files, oldfd)) |
124 | retval = -EBADF; |
125 | rcu_read_unlock(); |
126 | return retval; |
127 | } |
128 | return sys_dup3(oldfd, newfd, 0); |
129 | } |
130 | |
131 | SYSCALL_DEFINE1(dup, unsigned int, fildes) |
132 | { |
133 | int ret = -EBADF; |
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 | } |
143 | return ret; |
144 | } |
145 | |
146 | #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME) |
147 | |
148 | static int setfl(int fd, struct file * filp, unsigned long arg) |
149 | { |
150 | struct inode * inode = filp->f_path.dentry->d_inode; |
151 | int error = 0; |
152 | |
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)) |
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)) |
162 | if (!is_owner_or_cap(inode)) |
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 | |
181 | /* |
182 | * ->fasync() is responsible for setting the FASYNC bit. |
183 | */ |
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; |
189 | if (error > 0) |
190 | error = 0; |
191 | } |
192 | spin_lock(&filp->f_lock); |
193 | filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK); |
194 | spin_unlock(&filp->f_lock); |
195 | |
196 | out: |
197 | return error; |
198 | } |
199 | |
200 | static void f_modown(struct file *filp, struct pid *pid, enum pid_type type, |
201 | int force) |
202 | { |
203 | write_lock_irq(&filp->f_owner.lock); |
204 | if (force || !filp->f_owner.pid) { |
205 | put_pid(filp->f_owner.pid); |
206 | filp->f_owner.pid = get_pid(pid); |
207 | filp->f_owner.pid_type = type; |
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 | } |
214 | } |
215 | write_unlock_irq(&filp->f_owner.lock); |
216 | } |
217 | |
218 | int __f_setown(struct file *filp, struct pid *pid, enum pid_type type, |
219 | int force) |
220 | { |
221 | int err; |
222 | |
223 | err = security_file_set_fowner(filp); |
224 | if (err) |
225 | return err; |
226 | |
227 | f_modown(filp, pid, type, force); |
228 | return 0; |
229 | } |
230 | EXPORT_SYMBOL(__f_setown); |
231 | |
232 | int 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(); |
244 | pid = find_vpid(who); |
245 | result = __f_setown(filp, pid, type, force); |
246 | rcu_read_unlock(); |
247 | return result; |
248 | } |
249 | EXPORT_SYMBOL(f_setown); |
250 | |
251 | void f_delown(struct file *filp) |
252 | { |
253 | f_modown(filp, NULL, PIDTYPE_PID, 1); |
254 | } |
255 | |
256 | pid_t f_getown(struct file *filp) |
257 | { |
258 | pid_t pid; |
259 | read_lock(&filp->f_owner.lock); |
260 | pid = pid_vnr(filp->f_owner.pid); |
261 | if (filp->f_owner.pid_type == PIDTYPE_PGID) |
262 | pid = -pid; |
263 | read_unlock(&filp->f_owner.lock); |
264 | return pid; |
265 | } |
266 | |
267 | static 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) |
277 | return -EFAULT; |
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 | |
288 | case F_OWNER_PGRP: |
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 | |
307 | static 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: |
325 | owner.type = F_OWNER_PGRP; |
326 | break; |
327 | |
328 | default: |
329 | WARN_ON(1); |
330 | ret = -EINVAL; |
331 | break; |
332 | } |
333 | read_unlock(&filp->f_owner.lock); |
334 | |
335 | if (!ret) { |
336 | ret = copy_to_user(owner_p, &owner, sizeof(owner)); |
337 | if (ret) |
338 | ret = -EFAULT; |
339 | } |
340 | return ret; |
341 | } |
342 | |
343 | static 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: |
350 | case F_DUPFD_CLOEXEC: |
351 | if (arg >= rlimit(RLIMIT_NOFILE)) |
352 | break; |
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 | } |
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: |
377 | err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg); |
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 | */ |
387 | err = f_getown(filp); |
388 | force_successful_syscall_return(); |
389 | break; |
390 | case F_SETOWN: |
391 | err = f_setown(filp, arg, 1); |
392 | break; |
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; |
399 | case F_GETSIG: |
400 | err = filp->f_owner.signum; |
401 | break; |
402 | case F_SETSIG: |
403 | /* arg == 0 restores default behaviour. */ |
404 | if (!valid_signal(arg)) { |
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; |
419 | case F_SETPIPE_SZ: |
420 | case F_GETPIPE_SZ: |
421 | err = pipe_fcntl(filp, cmd, arg); |
422 | break; |
423 | default: |
424 | break; |
425 | } |
426 | return err; |
427 | } |
428 | |
429 | SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg) |
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); |
447 | out: |
448 | return err; |
449 | } |
450 | |
451 | #if BITS_PER_LONG == 32 |
452 | SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd, |
453 | unsigned long, arg) |
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: |
476 | err = fcntl_setlk64(fd, filp, cmd, |
477 | (struct flock64 __user *) arg); |
478 | break; |
479 | default: |
480 | err = do_fcntl(fd, cmd, arg, filp); |
481 | break; |
482 | } |
483 | fput(filp); |
484 | out: |
485 | return err; |
486 | } |
487 | #endif |
488 | |
489 | /* Table to convert sigio signal codes into poll band bitmaps */ |
490 | |
491 | static const long band_table[NSIGPOLL] = { |
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 | |
500 | static inline int sigio_perm(struct task_struct *p, |
501 | struct fown_struct *fown, int sig) |
502 | { |
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; |
514 | } |
515 | |
516 | static void send_sigio_to_task(struct task_struct *p, |
517 | struct fown_struct *fown, |
518 | int fd, int reason, int group) |
519 | { |
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)) |
527 | return; |
528 | |
529 | switch (signum) { |
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 */ |
538 | si.si_signo = signum; |
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. */ |
544 | BUG_ON((reason & __SI_MASK) != __SI_POLL); |
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; |
550 | if (!do_send_sig_info(signum, &si, p, group)) |
551 | break; |
552 | /* fall-through: fall back on the old plain SIGIO signal */ |
553 | case 0: |
554 | do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group); |
555 | } |
556 | } |
557 | |
558 | void send_sigio(struct fown_struct *fown, int fd, int band) |
559 | { |
560 | struct task_struct *p; |
561 | enum pid_type type; |
562 | struct pid *pid; |
563 | int group = 1; |
564 | |
565 | read_lock(&fown->lock); |
566 | |
567 | type = fown->pid_type; |
568 | if (type == PIDTYPE_MAX) { |
569 | group = 0; |
570 | type = PIDTYPE_PID; |
571 | } |
572 | |
573 | pid = fown->pid; |
574 | if (!pid) |
575 | goto out_unlock_fown; |
576 | |
577 | read_lock(&tasklist_lock); |
578 | do_each_pid_task(pid, type, p) { |
579 | send_sigio_to_task(p, fown, fd, band, group); |
580 | } while_each_pid_task(pid, type, p); |
581 | read_unlock(&tasklist_lock); |
582 | out_unlock_fown: |
583 | read_unlock(&fown->lock); |
584 | } |
585 | |
586 | static void send_sigurg_to_task(struct task_struct *p, |
587 | struct fown_struct *fown, int group) |
588 | { |
589 | if (sigio_perm(p, fown, SIGURG)) |
590 | do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group); |
591 | } |
592 | |
593 | int send_sigurg(struct fown_struct *fown) |
594 | { |
595 | struct task_struct *p; |
596 | enum pid_type type; |
597 | struct pid *pid; |
598 | int group = 1; |
599 | int ret = 0; |
600 | |
601 | read_lock(&fown->lock); |
602 | |
603 | type = fown->pid_type; |
604 | if (type == PIDTYPE_MAX) { |
605 | group = 0; |
606 | type = PIDTYPE_PID; |
607 | } |
608 | |
609 | pid = fown->pid; |
610 | if (!pid) |
611 | goto out_unlock_fown; |
612 | |
613 | ret = 1; |
614 | |
615 | read_lock(&tasklist_lock); |
616 | do_each_pid_task(pid, type, p) { |
617 | send_sigurg_to_task(p, fown, group); |
618 | } while_each_pid_task(pid, type, p); |
619 | read_unlock(&tasklist_lock); |
620 | out_unlock_fown: |
621 | read_unlock(&fown->lock); |
622 | return ret; |
623 | } |
624 | |
625 | static DEFINE_SPINLOCK(fasync_lock); |
626 | static struct kmem_cache *fasync_cache __read_mostly; |
627 | |
628 | static 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 | |
634 | /* |
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 | * |
642 | */ |
643 | int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp) |
644 | { |
645 | struct fasync_struct *fa, **fp; |
646 | int result = 0; |
647 | |
648 | spin_lock(&filp->f_lock); |
649 | spin_lock(&fasync_lock); |
650 | for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) { |
651 | if (fa->fa_file != filp) |
652 | continue; |
653 | |
654 | spin_lock_irq(&fa->fa_lock); |
655 | fa->fa_file = NULL; |
656 | spin_unlock_irq(&fa->fa_lock); |
657 | |
658 | *fp = fa->fa_next; |
659 | call_rcu(&fa->fa_rcu, fasync_free_rcu); |
660 | filp->f_flags &= ~FASYNC; |
661 | result = 1; |
662 | break; |
663 | } |
664 | spin_unlock(&fasync_lock); |
665 | spin_unlock(&filp->f_lock); |
666 | return result; |
667 | } |
668 | |
669 | struct fasync_struct *fasync_alloc(void) |
670 | { |
671 | return kmem_cache_alloc(fasync_cache, GFP_KERNEL); |
672 | } |
673 | |
674 | /* |
675 | * NOTE! This can be used only for unused fasync entries: |
676 | * entries that actually got inserted on the fasync list |
677 | * need to be released by rcu - see fasync_remove_entry. |
678 | */ |
679 | void fasync_free(struct fasync_struct *new) |
680 | { |
681 | kmem_cache_free(fasync_cache, new); |
682 | } |
683 | |
684 | /* |
685 | * Insert a new entry into the fasync list. Return the pointer to the |
686 | * old one if we didn't use the new one. |
687 | * |
688 | * NOTE! It is very important that the FASYNC flag always |
689 | * match the state "is the filp on a fasync list". |
690 | */ |
691 | struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new) |
692 | { |
693 | struct fasync_struct *fa, **fp; |
694 | |
695 | spin_lock(&filp->f_lock); |
696 | spin_lock(&fasync_lock); |
697 | for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) { |
698 | if (fa->fa_file != filp) |
699 | continue; |
700 | |
701 | spin_lock_irq(&fa->fa_lock); |
702 | fa->fa_fd = fd; |
703 | spin_unlock_irq(&fa->fa_lock); |
704 | goto out; |
705 | } |
706 | |
707 | spin_lock_init(&new->fa_lock); |
708 | new->magic = FASYNC_MAGIC; |
709 | new->fa_file = filp; |
710 | new->fa_fd = fd; |
711 | new->fa_next = *fapp; |
712 | rcu_assign_pointer(*fapp, new); |
713 | filp->f_flags |= FASYNC; |
714 | |
715 | out: |
716 | spin_unlock(&fasync_lock); |
717 | spin_unlock(&filp->f_lock); |
718 | return fa; |
719 | } |
720 | |
721 | /* |
722 | * Add a fasync entry. Return negative on error, positive if |
723 | * added, and zero if did nothing but change an existing one. |
724 | */ |
725 | static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp) |
726 | { |
727 | struct fasync_struct *new; |
728 | |
729 | new = fasync_alloc(); |
730 | if (!new) |
731 | return -ENOMEM; |
732 | |
733 | /* |
734 | * fasync_insert_entry() returns the old (update) entry if |
735 | * it existed. |
736 | * |
737 | * So free the (unused) new entry and return 0 to let the |
738 | * caller know that we didn't add any new fasync entries. |
739 | */ |
740 | if (fasync_insert_entry(fd, filp, fapp, new)) { |
741 | fasync_free(new); |
742 | return 0; |
743 | } |
744 | |
745 | return 1; |
746 | } |
747 | |
748 | /* |
749 | * fasync_helper() is used by almost all character device drivers |
750 | * to set up the fasync queue, and for regular files by the file |
751 | * lease code. It returns negative on error, 0 if it did no changes |
752 | * and positive if it added/deleted the entry. |
753 | */ |
754 | int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp) |
755 | { |
756 | if (!on) |
757 | return fasync_remove_entry(filp, fapp); |
758 | return fasync_add_entry(fd, filp, fapp); |
759 | } |
760 | |
761 | EXPORT_SYMBOL(fasync_helper); |
762 | |
763 | /* |
764 | * rcu_read_lock() is held |
765 | */ |
766 | static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band) |
767 | { |
768 | while (fa) { |
769 | struct fown_struct *fown; |
770 | unsigned long flags; |
771 | |
772 | if (fa->magic != FASYNC_MAGIC) { |
773 | printk(KERN_ERR "kill_fasync: bad magic number in " |
774 | "fasync_struct!\n"); |
775 | return; |
776 | } |
777 | spin_lock_irqsave(&fa->fa_lock, flags); |
778 | if (fa->fa_file) { |
779 | fown = &fa->fa_file->f_owner; |
780 | /* Don't send SIGURG to processes which have not set a |
781 | queued signum: SIGURG has its own default signalling |
782 | mechanism. */ |
783 | if (!(sig == SIGURG && fown->signum == 0)) |
784 | send_sigio(fown, fa->fa_fd, band); |
785 | } |
786 | spin_unlock_irqrestore(&fa->fa_lock, flags); |
787 | fa = rcu_dereference(fa->fa_next); |
788 | } |
789 | } |
790 | |
791 | void kill_fasync(struct fasync_struct **fp, int sig, int band) |
792 | { |
793 | /* First a quick test without locking: usually |
794 | * the list is empty. |
795 | */ |
796 | if (*fp) { |
797 | rcu_read_lock(); |
798 | kill_fasync_rcu(rcu_dereference(*fp), sig, band); |
799 | rcu_read_unlock(); |
800 | } |
801 | } |
802 | EXPORT_SYMBOL(kill_fasync); |
803 | |
804 | static int __init fcntl_init(void) |
805 | { |
806 | /* |
807 | * Please add new bits here to ensure allocation uniqueness. |
808 | * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY |
809 | * is defined as O_NONBLOCK on some platforms and not on others. |
810 | */ |
811 | BUILD_BUG_ON(18 - 1 /* for O_RDONLY being 0 */ != HWEIGHT32( |
812 | O_RDONLY | O_WRONLY | O_RDWR | |
813 | O_CREAT | O_EXCL | O_NOCTTY | |
814 | O_TRUNC | O_APPEND | /* O_NONBLOCK | */ |
815 | __O_SYNC | O_DSYNC | FASYNC | |
816 | O_DIRECT | O_LARGEFILE | O_DIRECTORY | |
817 | O_NOFOLLOW | O_NOATIME | O_CLOEXEC | |
818 | __FMODE_EXEC |
819 | )); |
820 | |
821 | fasync_cache = kmem_cache_create("fasync_cache", |
822 | sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL); |
823 | return 0; |
824 | } |
825 | |
826 | module_init(fcntl_init) |
827 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9