blob: 03c789560fb87b6213abc2b33cf5dea494fb583d [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>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080012#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/dnotify.h>
14#include <linux/smp_lock.h>
15#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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021
22#include <asm/poll.h>
23#include <asm/siginfo.h>
24#include <asm/uaccess.h>
25
26void fastcall set_close_on_exec(unsigned int fd, int flag)
27{
28 struct files_struct *files = current->files;
Dipankar Sarmabadf1662005-09-09 13:04:10 -070029 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -070030 spin_lock(&files->file_lock);
Dipankar Sarmabadf1662005-09-09 13:04:10 -070031 fdt = files_fdtable(files);
Linus Torvalds1da177e2005-04-16 15:20:36 -070032 if (flag)
Dipankar Sarmabadf1662005-09-09 13:04:10 -070033 FD_SET(fd, fdt->close_on_exec);
Linus Torvalds1da177e2005-04-16 15:20:36 -070034 else
Dipankar Sarmabadf1662005-09-09 13:04:10 -070035 FD_CLR(fd, fdt->close_on_exec);
Linus Torvalds1da177e2005-04-16 15:20:36 -070036 spin_unlock(&files->file_lock);
37}
38
Arjan van de Ven858119e2006-01-14 13:20:43 -080039static int get_close_on_exec(unsigned int fd)
Linus Torvalds1da177e2005-04-16 15:20:36 -070040{
41 struct files_struct *files = current->files;
Dipankar Sarmabadf1662005-09-09 13:04:10 -070042 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -070043 int res;
Dipankar Sarmab8359962005-09-09 13:04:14 -070044 rcu_read_lock();
Dipankar Sarmabadf1662005-09-09 13:04:10 -070045 fdt = files_fdtable(files);
46 res = FD_ISSET(fd, fdt->close_on_exec);
Dipankar Sarmab8359962005-09-09 13:04:14 -070047 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -070048 return res;
49}
50
51/*
52 * locate_fd finds a free file descriptor in the open_fds fdset,
53 * expanding the fd arrays if necessary. Must be called with the
54 * file_lock held for write.
55 */
56
57static int locate_fd(struct files_struct *files,
58 struct file *file, unsigned int orig_start)
59{
60 unsigned int newfd;
61 unsigned int start;
62 int error;
Dipankar Sarmabadf1662005-09-09 13:04:10 -070063 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
65 error = -EINVAL;
66 if (orig_start >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
67 goto out;
68
69repeat:
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070070 fdt = files_fdtable(files);
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 /*
72 * Someone might have closed fd's in the range
Dipankar Sarmabadf1662005-09-09 13:04:10 -070073 * orig_start..fdt->next_fd
Linus Torvalds1da177e2005-04-16 15:20:36 -070074 */
75 start = orig_start;
Eric Dumazet0c9e63f2006-03-23 03:00:12 -080076 if (start < files->next_fd)
77 start = files->next_fd;
Linus Torvalds1da177e2005-04-16 15:20:36 -070078
79 newfd = start;
Dipankar Sarmabadf1662005-09-09 13:04:10 -070080 if (start < fdt->max_fdset) {
81 newfd = find_next_zero_bit(fdt->open_fds->fds_bits,
82 fdt->max_fdset, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 }
84
85 error = -EMFILE;
86 if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
87 goto out;
88
89 error = expand_files(files, newfd);
90 if (error < 0)
91 goto out;
92
93 /*
94 * If we needed to expand the fs array we
95 * might have blocked - try again.
96 */
97 if (error)
98 goto repeat;
99
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700100 /*
101 * We reacquired files_lock, so we are safe as long as
102 * we reacquire the fdtable pointer and use it while holding
103 * the lock, no one can free it during that time.
104 */
Eric Dumazet0c9e63f2006-03-23 03:00:12 -0800105 if (start <= files->next_fd)
106 files->next_fd = newfd + 1;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700107
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 error = newfd;
109
110out:
111 return error;
112}
113
114static int dupfd(struct file *file, unsigned int start)
115{
116 struct files_struct * files = current->files;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700117 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 int fd;
119
120 spin_lock(&files->file_lock);
121 fd = locate_fd(files, file, start);
122 if (fd >= 0) {
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700123 /* locate_fd() may have expanded fdtable, load the ptr */
124 fdt = files_fdtable(files);
125 FD_SET(fd, fdt->open_fds);
126 FD_CLR(fd, fdt->close_on_exec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 spin_unlock(&files->file_lock);
128 fd_install(fd, file);
129 } else {
130 spin_unlock(&files->file_lock);
131 fput(file);
132 }
133
134 return fd;
135}
136
137asmlinkage long sys_dup2(unsigned int oldfd, unsigned int newfd)
138{
139 int err = -EBADF;
140 struct file * file, *tofree;
141 struct files_struct * files = current->files;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700142 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143
144 spin_lock(&files->file_lock);
145 if (!(file = fcheck(oldfd)))
146 goto out_unlock;
147 err = newfd;
148 if (newfd == oldfd)
149 goto out_unlock;
150 err = -EBADF;
151 if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
152 goto out_unlock;
153 get_file(file); /* We are now finished with oldfd */
154
155 err = expand_files(files, newfd);
156 if (err < 0)
157 goto out_fput;
158
159 /* To avoid races with open() and dup(), we will mark the fd as
160 * in-use in the open-file bitmap throughout the entire dup2()
161 * process. This is quite safe: do_close() uses the fd array
162 * entry, not the bitmap, to decide what work needs to be
163 * done. --sct */
164 /* Doesn't work. open() might be there first. --AV */
165
166 /* Yes. It's a race. In user space. Nothing sane to do */
167 err = -EBUSY;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700168 fdt = files_fdtable(files);
169 tofree = fdt->fd[newfd];
170 if (!tofree && FD_ISSET(newfd, fdt->open_fds))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 goto out_fput;
172
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700173 rcu_assign_pointer(fdt->fd[newfd], file);
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700174 FD_SET(newfd, fdt->open_fds);
175 FD_CLR(newfd, fdt->close_on_exec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 spin_unlock(&files->file_lock);
177
178 if (tofree)
179 filp_close(tofree, files);
180 err = newfd;
181out:
182 return err;
183out_unlock:
184 spin_unlock(&files->file_lock);
185 goto out;
186
187out_fput:
188 spin_unlock(&files->file_lock);
189 fput(file);
190 goto out;
191}
192
193asmlinkage long sys_dup(unsigned int fildes)
194{
195 int ret = -EBADF;
196 struct file * file = fget(fildes);
197
198 if (file)
199 ret = dupfd(file, 0);
200 return ret;
201}
202
203#define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
204
205static int setfl(int fd, struct file * filp, unsigned long arg)
206{
207 struct inode * inode = filp->f_dentry->d_inode;
208 int error = 0;
209
dean gaudet7d95c8f2006-02-03 03:04:30 -0800210 /*
211 * O_APPEND cannot be cleared if the file is marked as append-only
212 * and the file is open for write.
213 */
214 if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 return -EPERM;
216
217 /* O_NOATIME can only be set by the owner or superuser */
218 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
219 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
220 return -EPERM;
221
222 /* required for strict SunOS emulation */
223 if (O_NONBLOCK != O_NDELAY)
224 if (arg & O_NDELAY)
225 arg |= O_NONBLOCK;
226
227 if (arg & O_DIRECT) {
228 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
229 !filp->f_mapping->a_ops->direct_IO)
230 return -EINVAL;
231 }
232
233 if (filp->f_op && filp->f_op->check_flags)
234 error = filp->f_op->check_flags(arg);
235 if (error)
236 return error;
237
238 lock_kernel();
239 if ((arg ^ filp->f_flags) & FASYNC) {
240 if (filp->f_op && filp->f_op->fasync) {
241 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
242 if (error < 0)
243 goto out;
244 }
245 }
246
247 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
248 out:
249 unlock_kernel();
250 return error;
251}
252
253static void f_modown(struct file *filp, unsigned long pid,
254 uid_t uid, uid_t euid, int force)
255{
256 write_lock_irq(&filp->f_owner.lock);
257 if (force || !filp->f_owner.pid) {
258 filp->f_owner.pid = pid;
259 filp->f_owner.uid = uid;
260 filp->f_owner.euid = euid;
261 }
262 write_unlock_irq(&filp->f_owner.lock);
263}
264
265int f_setown(struct file *filp, unsigned long arg, int force)
266{
267 int err;
268
269 err = security_file_set_fowner(filp);
270 if (err)
271 return err;
272
273 f_modown(filp, arg, current->uid, current->euid, force);
274 return 0;
275}
276
277EXPORT_SYMBOL(f_setown);
278
279void f_delown(struct file *filp)
280{
281 f_modown(filp, 0, 0, 0, 1);
282}
283
284static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
285 struct file *filp)
286{
287 long err = -EINVAL;
288
289 switch (cmd) {
290 case F_DUPFD:
291 get_file(filp);
292 err = dupfd(filp, arg);
293 break;
294 case F_GETFD:
295 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
296 break;
297 case F_SETFD:
298 err = 0;
299 set_close_on_exec(fd, arg & FD_CLOEXEC);
300 break;
301 case F_GETFL:
302 err = filp->f_flags;
303 break;
304 case F_SETFL:
305 err = setfl(fd, filp, arg);
306 break;
307 case F_GETLK:
308 err = fcntl_getlk(filp, (struct flock __user *) arg);
309 break;
310 case F_SETLK:
311 case F_SETLKW:
Peter Staubachc2936212005-07-27 11:45:09 -0700312 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 break;
314 case F_GETOWN:
315 /*
316 * XXX If f_owner is a process group, the
317 * negative return value will get converted
318 * into an error. Oops. If we keep the
319 * current syscall conventions, the only way
320 * to fix this will be in libc.
321 */
322 err = filp->f_owner.pid;
323 force_successful_syscall_return();
324 break;
325 case F_SETOWN:
326 err = f_setown(filp, arg, 1);
327 break;
328 case F_GETSIG:
329 err = filp->f_owner.signum;
330 break;
331 case F_SETSIG:
332 /* arg == 0 restores default behaviour. */
Jesper Juhl7ed20e12005-05-01 08:59:14 -0700333 if (!valid_signal(arg)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 break;
335 }
336 err = 0;
337 filp->f_owner.signum = arg;
338 break;
339 case F_GETLEASE:
340 err = fcntl_getlease(filp);
341 break;
342 case F_SETLEASE:
343 err = fcntl_setlease(fd, filp, arg);
344 break;
345 case F_NOTIFY:
346 err = fcntl_dirnotify(fd, filp, arg);
347 break;
348 default:
349 break;
350 }
351 return err;
352}
353
354asmlinkage long sys_fcntl(unsigned int fd, unsigned int cmd, unsigned long arg)
355{
356 struct file *filp;
357 long err = -EBADF;
358
359 filp = fget(fd);
360 if (!filp)
361 goto out;
362
363 err = security_file_fcntl(filp, cmd, arg);
364 if (err) {
365 fput(filp);
366 return err;
367 }
368
369 err = do_fcntl(fd, cmd, arg, filp);
370
371 fput(filp);
372out:
373 return err;
374}
375
376#if BITS_PER_LONG == 32
377asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
378{
379 struct file * filp;
380 long err;
381
382 err = -EBADF;
383 filp = fget(fd);
384 if (!filp)
385 goto out;
386
387 err = security_file_fcntl(filp, cmd, arg);
388 if (err) {
389 fput(filp);
390 return err;
391 }
392 err = -EBADF;
393
394 switch (cmd) {
395 case F_GETLK64:
396 err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
397 break;
398 case F_SETLK64:
399 case F_SETLKW64:
Peter Staubachc2936212005-07-27 11:45:09 -0700400 err = fcntl_setlk64(fd, filp, cmd,
401 (struct flock64 __user *) arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 break;
403 default:
404 err = do_fcntl(fd, cmd, arg, filp);
405 break;
406 }
407 fput(filp);
408out:
409 return err;
410}
411#endif
412
413/* Table to convert sigio signal codes into poll band bitmaps */
414
415static long band_table[NSIGPOLL] = {
416 POLLIN | POLLRDNORM, /* POLL_IN */
417 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
418 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
419 POLLERR, /* POLL_ERR */
420 POLLPRI | POLLRDBAND, /* POLL_PRI */
421 POLLHUP | POLLERR /* POLL_HUP */
422};
423
424static inline int sigio_perm(struct task_struct *p,
425 struct fown_struct *fown, int sig)
426{
427 return (((fown->euid == 0) ||
428 (fown->euid == p->suid) || (fown->euid == p->uid) ||
429 (fown->uid == p->suid) || (fown->uid == p->uid)) &&
430 !security_file_send_sigiotask(p, fown, sig));
431}
432
433static void send_sigio_to_task(struct task_struct *p,
434 struct fown_struct *fown,
435 int fd,
436 int reason)
437{
438 if (!sigio_perm(p, fown, fown->signum))
439 return;
440
441 switch (fown->signum) {
442 siginfo_t si;
443 default:
444 /* Queue a rt signal with the appropriate fd as its
445 value. We use SI_SIGIO as the source, not
446 SI_KERNEL, since kernel signals always get
447 delivered even if we can't queue. Failure to
448 queue in this case _should_ be reported; we fall
449 back to SIGIO in that case. --sct */
450 si.si_signo = fown->signum;
451 si.si_errno = 0;
452 si.si_code = reason;
453 /* Make sure we are called with one of the POLL_*
454 reasons, otherwise we could leak kernel stack into
455 userspace. */
456 if ((reason & __SI_MASK) != __SI_POLL)
457 BUG();
458 if (reason - POLL_IN >= NSIGPOLL)
459 si.si_band = ~0L;
460 else
461 si.si_band = band_table[reason - POLL_IN];
462 si.si_fd = fd;
Oleg Nesterov850d6fb2006-01-08 01:03:29 -0800463 if (!group_send_sig_info(fown->signum, &si, p))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 break;
465 /* fall-through: fall back on the old plain SIGIO signal */
466 case 0:
Oleg Nesterov850d6fb2006-01-08 01:03:29 -0800467 group_send_sig_info(SIGIO, SEND_SIG_PRIV, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 }
469}
470
471void send_sigio(struct fown_struct *fown, int fd, int band)
472{
473 struct task_struct *p;
474 int pid;
475
476 read_lock(&fown->lock);
477 pid = fown->pid;
478 if (!pid)
479 goto out_unlock_fown;
480
481 read_lock(&tasklist_lock);
482 if (pid > 0) {
483 p = find_task_by_pid(pid);
484 if (p) {
485 send_sigio_to_task(p, fown, fd, band);
486 }
487 } else {
488 do_each_task_pid(-pid, PIDTYPE_PGID, p) {
489 send_sigio_to_task(p, fown, fd, band);
490 } while_each_task_pid(-pid, PIDTYPE_PGID, p);
491 }
492 read_unlock(&tasklist_lock);
493 out_unlock_fown:
494 read_unlock(&fown->lock);
495}
496
497static void send_sigurg_to_task(struct task_struct *p,
498 struct fown_struct *fown)
499{
500 if (sigio_perm(p, fown, SIGURG))
Oleg Nesterov850d6fb2006-01-08 01:03:29 -0800501 group_send_sig_info(SIGURG, SEND_SIG_PRIV, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502}
503
504int send_sigurg(struct fown_struct *fown)
505{
506 struct task_struct *p;
507 int pid, ret = 0;
508
509 read_lock(&fown->lock);
510 pid = fown->pid;
511 if (!pid)
512 goto out_unlock_fown;
513
514 ret = 1;
515
516 read_lock(&tasklist_lock);
517 if (pid > 0) {
518 p = find_task_by_pid(pid);
519 if (p) {
520 send_sigurg_to_task(p, fown);
521 }
522 } else {
523 do_each_task_pid(-pid, PIDTYPE_PGID, p) {
524 send_sigurg_to_task(p, fown);
525 } while_each_task_pid(-pid, PIDTYPE_PGID, p);
526 }
527 read_unlock(&tasklist_lock);
528 out_unlock_fown:
529 read_unlock(&fown->lock);
530 return ret;
531}
532
533static DEFINE_RWLOCK(fasync_lock);
534static kmem_cache_t *fasync_cache;
535
536/*
537 * fasync_helper() is used by some character device drivers (mainly mice)
538 * to set up the fasync queue. It returns negative on error, 0 if it did
539 * no changes and positive if it added/deleted the entry.
540 */
541int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
542{
543 struct fasync_struct *fa, **fp;
544 struct fasync_struct *new = NULL;
545 int result = 0;
546
547 if (on) {
548 new = kmem_cache_alloc(fasync_cache, SLAB_KERNEL);
549 if (!new)
550 return -ENOMEM;
551 }
552 write_lock_irq(&fasync_lock);
553 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
554 if (fa->fa_file == filp) {
555 if(on) {
556 fa->fa_fd = fd;
557 kmem_cache_free(fasync_cache, new);
558 } else {
559 *fp = fa->fa_next;
560 kmem_cache_free(fasync_cache, fa);
561 result = 1;
562 }
563 goto out;
564 }
565 }
566
567 if (on) {
568 new->magic = FASYNC_MAGIC;
569 new->fa_file = filp;
570 new->fa_fd = fd;
571 new->fa_next = *fapp;
572 *fapp = new;
573 result = 1;
574 }
575out:
576 write_unlock_irq(&fasync_lock);
577 return result;
578}
579
580EXPORT_SYMBOL(fasync_helper);
581
582void __kill_fasync(struct fasync_struct *fa, int sig, int band)
583{
584 while (fa) {
585 struct fown_struct * fown;
586 if (fa->magic != FASYNC_MAGIC) {
587 printk(KERN_ERR "kill_fasync: bad magic number in "
588 "fasync_struct!\n");
589 return;
590 }
591 fown = &fa->fa_file->f_owner;
592 /* Don't send SIGURG to processes which have not set a
593 queued signum: SIGURG has its own default signalling
594 mechanism. */
595 if (!(sig == SIGURG && fown->signum == 0))
596 send_sigio(fown, fa->fa_fd, band);
597 fa = fa->fa_next;
598 }
599}
600
601EXPORT_SYMBOL(__kill_fasync);
602
603void kill_fasync(struct fasync_struct **fp, int sig, int band)
604{
605 /* First a quick test without locking: usually
606 * the list is empty.
607 */
608 if (*fp) {
609 read_lock(&fasync_lock);
610 /* reread *fp after obtaining the lock */
611 __kill_fasync(*fp, sig, band);
612 read_unlock(&fasync_lock);
613 }
614}
615EXPORT_SYMBOL(kill_fasync);
616
617static int __init fasync_init(void)
618{
619 fasync_cache = kmem_cache_create("fasync_cache",
620 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL, NULL);
621 return 0;
622}
623
624module_init(fasync_init)