blob: 863b46e0d78a6594371a744358104faae657c307 [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>
12#include <linux/dnotify.h>
13#include <linux/smp_lock.h>
14#include <linux/slab.h>
15#include <linux/module.h>
16#include <linux/security.h>
17#include <linux/ptrace.h>
Jesper Juhl7ed20e12005-05-01 08:59:14 -070018#include <linux/signal.h>
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070019#include <linux/rcupdate.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020
21#include <asm/poll.h>
22#include <asm/siginfo.h>
23#include <asm/uaccess.h>
24
25void fastcall set_close_on_exec(unsigned int fd, int flag)
26{
27 struct files_struct *files = current->files;
Dipankar Sarmabadf1662005-09-09 13:04:10 -070028 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -070029 spin_lock(&files->file_lock);
Dipankar Sarmabadf1662005-09-09 13:04:10 -070030 fdt = files_fdtable(files);
Linus Torvalds1da177e2005-04-16 15:20:36 -070031 if (flag)
Dipankar Sarmabadf1662005-09-09 13:04:10 -070032 FD_SET(fd, fdt->close_on_exec);
Linus Torvalds1da177e2005-04-16 15:20:36 -070033 else
Dipankar Sarmabadf1662005-09-09 13:04:10 -070034 FD_CLR(fd, fdt->close_on_exec);
Linus Torvalds1da177e2005-04-16 15:20:36 -070035 spin_unlock(&files->file_lock);
36}
37
38static inline int get_close_on_exec(unsigned int fd)
39{
40 struct files_struct *files = current->files;
Dipankar Sarmabadf1662005-09-09 13:04:10 -070041 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -070042 int res;
Dipankar Sarmab8359962005-09-09 13:04:14 -070043 rcu_read_lock();
Dipankar Sarmabadf1662005-09-09 13:04:10 -070044 fdt = files_fdtable(files);
45 res = FD_ISSET(fd, fdt->close_on_exec);
Dipankar Sarmab8359962005-09-09 13:04:14 -070046 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -070047 return res;
48}
49
50/*
51 * locate_fd finds a free file descriptor in the open_fds fdset,
52 * expanding the fd arrays if necessary. Must be called with the
53 * file_lock held for write.
54 */
55
56static int locate_fd(struct files_struct *files,
57 struct file *file, unsigned int orig_start)
58{
59 unsigned int newfd;
60 unsigned int start;
61 int error;
Dipankar Sarmabadf1662005-09-09 13:04:10 -070062 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -070063
64 error = -EINVAL;
65 if (orig_start >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
66 goto out;
67
68repeat:
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070069 fdt = files_fdtable(files);
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 /*
71 * Someone might have closed fd's in the range
Dipankar Sarmabadf1662005-09-09 13:04:10 -070072 * orig_start..fdt->next_fd
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 */
74 start = orig_start;
Dipankar Sarmabadf1662005-09-09 13:04:10 -070075 if (start < fdt->next_fd)
76 start = fdt->next_fd;
Linus Torvalds1da177e2005-04-16 15:20:36 -070077
78 newfd = start;
Dipankar Sarmabadf1662005-09-09 13:04:10 -070079 if (start < fdt->max_fdset) {
80 newfd = find_next_zero_bit(fdt->open_fds->fds_bits,
81 fdt->max_fdset, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 }
83
84 error = -EMFILE;
85 if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
86 goto out;
87
88 error = expand_files(files, newfd);
89 if (error < 0)
90 goto out;
91
92 /*
93 * If we needed to expand the fs array we
94 * might have blocked - try again.
95 */
96 if (error)
97 goto repeat;
98
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070099 /*
100 * We reacquired files_lock, so we are safe as long as
101 * we reacquire the fdtable pointer and use it while holding
102 * the lock, no one can free it during that time.
103 */
104 fdt = files_fdtable(files);
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700105 if (start <= fdt->next_fd)
106 fdt->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
210 /* O_APPEND cannot be cleared if the file is marked as append-only */
211 if (!(arg & O_APPEND) && IS_APPEND(inode))
212 return -EPERM;
213
214 /* O_NOATIME can only be set by the owner or superuser */
215 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
216 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
217 return -EPERM;
218
219 /* required for strict SunOS emulation */
220 if (O_NONBLOCK != O_NDELAY)
221 if (arg & O_NDELAY)
222 arg |= O_NONBLOCK;
223
224 if (arg & O_DIRECT) {
225 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
226 !filp->f_mapping->a_ops->direct_IO)
227 return -EINVAL;
228 }
229
230 if (filp->f_op && filp->f_op->check_flags)
231 error = filp->f_op->check_flags(arg);
232 if (error)
233 return error;
234
235 lock_kernel();
236 if ((arg ^ filp->f_flags) & FASYNC) {
237 if (filp->f_op && filp->f_op->fasync) {
238 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
239 if (error < 0)
240 goto out;
241 }
242 }
243
244 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
245 out:
246 unlock_kernel();
247 return error;
248}
249
250static void f_modown(struct file *filp, unsigned long pid,
251 uid_t uid, uid_t euid, int force)
252{
253 write_lock_irq(&filp->f_owner.lock);
254 if (force || !filp->f_owner.pid) {
255 filp->f_owner.pid = pid;
256 filp->f_owner.uid = uid;
257 filp->f_owner.euid = euid;
258 }
259 write_unlock_irq(&filp->f_owner.lock);
260}
261
262int f_setown(struct file *filp, unsigned long arg, int force)
263{
264 int err;
265
266 err = security_file_set_fowner(filp);
267 if (err)
268 return err;
269
270 f_modown(filp, arg, current->uid, current->euid, force);
271 return 0;
272}
273
274EXPORT_SYMBOL(f_setown);
275
276void f_delown(struct file *filp)
277{
278 f_modown(filp, 0, 0, 0, 1);
279}
280
281static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
282 struct file *filp)
283{
284 long err = -EINVAL;
285
286 switch (cmd) {
287 case F_DUPFD:
288 get_file(filp);
289 err = dupfd(filp, arg);
290 break;
291 case F_GETFD:
292 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
293 break;
294 case F_SETFD:
295 err = 0;
296 set_close_on_exec(fd, arg & FD_CLOEXEC);
297 break;
298 case F_GETFL:
299 err = filp->f_flags;
300 break;
301 case F_SETFL:
302 err = setfl(fd, filp, arg);
303 break;
304 case F_GETLK:
305 err = fcntl_getlk(filp, (struct flock __user *) arg);
306 break;
307 case F_SETLK:
308 case F_SETLKW:
Peter Staubachc2936212005-07-27 11:45:09 -0700309 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 break;
311 case F_GETOWN:
312 /*
313 * XXX If f_owner is a process group, the
314 * negative return value will get converted
315 * into an error. Oops. If we keep the
316 * current syscall conventions, the only way
317 * to fix this will be in libc.
318 */
319 err = filp->f_owner.pid;
320 force_successful_syscall_return();
321 break;
322 case F_SETOWN:
323 err = f_setown(filp, arg, 1);
324 break;
325 case F_GETSIG:
326 err = filp->f_owner.signum;
327 break;
328 case F_SETSIG:
329 /* arg == 0 restores default behaviour. */
Jesper Juhl7ed20e12005-05-01 08:59:14 -0700330 if (!valid_signal(arg)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 break;
332 }
333 err = 0;
334 filp->f_owner.signum = arg;
335 break;
336 case F_GETLEASE:
337 err = fcntl_getlease(filp);
338 break;
339 case F_SETLEASE:
340 err = fcntl_setlease(fd, filp, arg);
341 break;
342 case F_NOTIFY:
343 err = fcntl_dirnotify(fd, filp, arg);
344 break;
345 default:
346 break;
347 }
348 return err;
349}
350
351asmlinkage long sys_fcntl(unsigned int fd, unsigned int cmd, unsigned long arg)
352{
353 struct file *filp;
354 long err = -EBADF;
355
356 filp = fget(fd);
357 if (!filp)
358 goto out;
359
360 err = security_file_fcntl(filp, cmd, arg);
361 if (err) {
362 fput(filp);
363 return err;
364 }
365
366 err = do_fcntl(fd, cmd, arg, filp);
367
368 fput(filp);
369out:
370 return err;
371}
372
373#if BITS_PER_LONG == 32
374asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
375{
376 struct file * filp;
377 long err;
378
379 err = -EBADF;
380 filp = fget(fd);
381 if (!filp)
382 goto out;
383
384 err = security_file_fcntl(filp, cmd, arg);
385 if (err) {
386 fput(filp);
387 return err;
388 }
389 err = -EBADF;
390
391 switch (cmd) {
392 case F_GETLK64:
393 err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
394 break;
395 case F_SETLK64:
396 case F_SETLKW64:
Peter Staubachc2936212005-07-27 11:45:09 -0700397 err = fcntl_setlk64(fd, filp, cmd,
398 (struct flock64 __user *) arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 break;
400 default:
401 err = do_fcntl(fd, cmd, arg, filp);
402 break;
403 }
404 fput(filp);
405out:
406 return err;
407}
408#endif
409
410/* Table to convert sigio signal codes into poll band bitmaps */
411
412static long band_table[NSIGPOLL] = {
413 POLLIN | POLLRDNORM, /* POLL_IN */
414 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
415 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
416 POLLERR, /* POLL_ERR */
417 POLLPRI | POLLRDBAND, /* POLL_PRI */
418 POLLHUP | POLLERR /* POLL_HUP */
419};
420
421static inline int sigio_perm(struct task_struct *p,
422 struct fown_struct *fown, int sig)
423{
424 return (((fown->euid == 0) ||
425 (fown->euid == p->suid) || (fown->euid == p->uid) ||
426 (fown->uid == p->suid) || (fown->uid == p->uid)) &&
427 !security_file_send_sigiotask(p, fown, sig));
428}
429
430static void send_sigio_to_task(struct task_struct *p,
431 struct fown_struct *fown,
432 int fd,
433 int reason)
434{
435 if (!sigio_perm(p, fown, fown->signum))
436 return;
437
438 switch (fown->signum) {
439 siginfo_t si;
440 default:
441 /* Queue a rt signal with the appropriate fd as its
442 value. We use SI_SIGIO as the source, not
443 SI_KERNEL, since kernel signals always get
444 delivered even if we can't queue. Failure to
445 queue in this case _should_ be reported; we fall
446 back to SIGIO in that case. --sct */
447 si.si_signo = fown->signum;
448 si.si_errno = 0;
449 si.si_code = reason;
450 /* Make sure we are called with one of the POLL_*
451 reasons, otherwise we could leak kernel stack into
452 userspace. */
453 if ((reason & __SI_MASK) != __SI_POLL)
454 BUG();
455 if (reason - POLL_IN >= NSIGPOLL)
456 si.si_band = ~0L;
457 else
458 si.si_band = band_table[reason - POLL_IN];
459 si.si_fd = fd;
Bharath Rameshfc9c9ab2005-04-16 15:25:41 -0700460 if (!send_group_sig_info(fown->signum, &si, p))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 break;
462 /* fall-through: fall back on the old plain SIGIO signal */
463 case 0:
464 send_group_sig_info(SIGIO, SEND_SIG_PRIV, p);
465 }
466}
467
468void send_sigio(struct fown_struct *fown, int fd, int band)
469{
470 struct task_struct *p;
471 int pid;
472
473 read_lock(&fown->lock);
474 pid = fown->pid;
475 if (!pid)
476 goto out_unlock_fown;
477
478 read_lock(&tasklist_lock);
479 if (pid > 0) {
480 p = find_task_by_pid(pid);
481 if (p) {
482 send_sigio_to_task(p, fown, fd, band);
483 }
484 } else {
485 do_each_task_pid(-pid, PIDTYPE_PGID, p) {
486 send_sigio_to_task(p, fown, fd, band);
487 } while_each_task_pid(-pid, PIDTYPE_PGID, p);
488 }
489 read_unlock(&tasklist_lock);
490 out_unlock_fown:
491 read_unlock(&fown->lock);
492}
493
494static void send_sigurg_to_task(struct task_struct *p,
495 struct fown_struct *fown)
496{
497 if (sigio_perm(p, fown, SIGURG))
498 send_group_sig_info(SIGURG, SEND_SIG_PRIV, p);
499}
500
501int send_sigurg(struct fown_struct *fown)
502{
503 struct task_struct *p;
504 int pid, ret = 0;
505
506 read_lock(&fown->lock);
507 pid = fown->pid;
508 if (!pid)
509 goto out_unlock_fown;
510
511 ret = 1;
512
513 read_lock(&tasklist_lock);
514 if (pid > 0) {
515 p = find_task_by_pid(pid);
516 if (p) {
517 send_sigurg_to_task(p, fown);
518 }
519 } else {
520 do_each_task_pid(-pid, PIDTYPE_PGID, p) {
521 send_sigurg_to_task(p, fown);
522 } while_each_task_pid(-pid, PIDTYPE_PGID, p);
523 }
524 read_unlock(&tasklist_lock);
525 out_unlock_fown:
526 read_unlock(&fown->lock);
527 return ret;
528}
529
530static DEFINE_RWLOCK(fasync_lock);
531static kmem_cache_t *fasync_cache;
532
533/*
534 * fasync_helper() is used by some character device drivers (mainly mice)
535 * to set up the fasync queue. It returns negative on error, 0 if it did
536 * no changes and positive if it added/deleted the entry.
537 */
538int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
539{
540 struct fasync_struct *fa, **fp;
541 struct fasync_struct *new = NULL;
542 int result = 0;
543
544 if (on) {
545 new = kmem_cache_alloc(fasync_cache, SLAB_KERNEL);
546 if (!new)
547 return -ENOMEM;
548 }
549 write_lock_irq(&fasync_lock);
550 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
551 if (fa->fa_file == filp) {
552 if(on) {
553 fa->fa_fd = fd;
554 kmem_cache_free(fasync_cache, new);
555 } else {
556 *fp = fa->fa_next;
557 kmem_cache_free(fasync_cache, fa);
558 result = 1;
559 }
560 goto out;
561 }
562 }
563
564 if (on) {
565 new->magic = FASYNC_MAGIC;
566 new->fa_file = filp;
567 new->fa_fd = fd;
568 new->fa_next = *fapp;
569 *fapp = new;
570 result = 1;
571 }
572out:
573 write_unlock_irq(&fasync_lock);
574 return result;
575}
576
577EXPORT_SYMBOL(fasync_helper);
578
579void __kill_fasync(struct fasync_struct *fa, int sig, int band)
580{
581 while (fa) {
582 struct fown_struct * fown;
583 if (fa->magic != FASYNC_MAGIC) {
584 printk(KERN_ERR "kill_fasync: bad magic number in "
585 "fasync_struct!\n");
586 return;
587 }
588 fown = &fa->fa_file->f_owner;
589 /* Don't send SIGURG to processes which have not set a
590 queued signum: SIGURG has its own default signalling
591 mechanism. */
592 if (!(sig == SIGURG && fown->signum == 0))
593 send_sigio(fown, fa->fa_fd, band);
594 fa = fa->fa_next;
595 }
596}
597
598EXPORT_SYMBOL(__kill_fasync);
599
600void kill_fasync(struct fasync_struct **fp, int sig, int band)
601{
602 /* First a quick test without locking: usually
603 * the list is empty.
604 */
605 if (*fp) {
606 read_lock(&fasync_lock);
607 /* reread *fp after obtaining the lock */
608 __kill_fasync(*fp, sig, band);
609 read_unlock(&fasync_lock);
610 }
611}
612EXPORT_SYMBOL(kill_fasync);
613
614static int __init fasync_init(void)
615{
616 fasync_cache = kmem_cache_create("fasync_cache",
617 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL, NULL);
618 return 0;
619}
620
621module_init(fasync_init)