1 /*
2 * linux/fs/fcntl.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7 #include <linux/init.h>
8 #include <linux/mm.h>
9 #include <linux/file.h>
10 #include <linux/dnotify.h>
11 #include <linux/smp_lock.h>
12 #include <linux/slab.h>
13
14 #include <asm/poll.h>
15 #include <asm/siginfo.h>
16 #include <asm/uaccess.h>
17
18 extern int sock_fcntl (struct file *, unsigned int cmd, unsigned long arg);
19 extern int fcntl_setlease(unsigned int fd, struct file *filp, long arg);
20 extern int fcntl_getlease(struct file *filp);
21
22 /* Expand files. Return <0 on error; 0 nothing done; 1 files expanded,
23 * we may have blocked.
24 *
25 * Should be called with the files->file_lock spinlock held for write.
26 */
27 static int expand_files(struct files_struct *files, int nr)
28 {
29 int err, expand = 0;
30 #ifdef FDSET_DEBUG
31 printk (KERN_ERR __FUNCTION__ " %d: nr = %d\n", current->pid, nr);
32 #endif
33
34 if (nr >= files->max_fdset) {
35 expand = 1;
36 if ((err = expand_fdset(files, nr)))
37 goto out;
38 }
39 if (nr >= files->max_fds) {
40 expand = 1;
41 if ((err = expand_fd_array(files, nr)))
42 goto out;
43 }
44 err = expand;
45 out:
46 #ifdef FDSET_DEBUG
47 if (err)
48 printk (KERN_ERR __FUNCTION__ " %d: return %d\n", current->pid, err);
49 #endif
50 return err;
51 }
52
53 /*
54 * locate_fd finds a free file descriptor in the open_fds fdset,
55 * expanding the fd arrays if necessary. The files write lock will be
56 * held on exit to ensure that the fd can be entered atomically.
57 */
58
59 static int locate_fd(struct files_struct *files,
60 struct file *file, int orig_start)
61 {
62 unsigned int newfd;
63 int error;
64 int start;
65
66 write_lock(&files->file_lock);
67
68 repeat:
69 /*
70 * Someone might have closed fd's in the range
71 * orig_start..files->next_fd
72 */
73 start = orig_start;
74 if (start < files->next_fd)
75 start = files->next_fd;
76
77 newfd = start;
78 if (start < files->max_fdset) {
79 newfd = find_next_zero_bit(files->open_fds->fds_bits,
80 files->max_fdset, start);
81 }
82
83 error = -EMFILE;
84 if (newfd >= current->rlim[RLIMIT_NOFILE].rlim_cur)
85 goto out;
86
87 error = expand_files(files, newfd);
88 if (error < 0)
89 goto out;
90
91 /*
92 * If we needed to expand the fs array we
93 * might have blocked - try again.
94 */
95 if (error)
96 goto repeat;
97
98 if (start <= files->next_fd)
99 files->next_fd = newfd + 1;
100
101 error = newfd;
102
103 out:
104 return error;
105 }
106
107 static inline void allocate_fd(struct files_struct *files,
108 struct file *file, int fd)
109 {
110 FD_SET(fd, files->open_fds);
111 FD_CLR(fd, files->close_on_exec);
112 write_unlock(&files->file_lock);
113 fd_install(fd, file);
114 }
115
116 static int dupfd(struct file *file, int start)
117 {
118 struct files_struct * files = current->files;
119 int ret;
120
121 ret = locate_fd(files, file, start);
122 if (ret < 0)
123 goto out_putf;
124 allocate_fd(files, file, ret);
125 return ret;
126
127 out_putf:
128 write_unlock(&files->file_lock);
129 fput(file);
130 return ret;
131 }
132
133 asmlinkage long sys_dup2(unsigned int oldfd, unsigned int newfd)
134 {
135 int err = -EBADF;
136 struct file * file, *tofree;
137 struct files_struct * files = current->files;
138
139 write_lock(&files->file_lock);
140 if (!(file = fcheck(oldfd)))
141 goto out_unlock;
142 err = newfd;
143 if (newfd == oldfd)
144 goto out_unlock;
145 err = -EBADF;
146 if (newfd >= current->rlim[RLIMIT_NOFILE].rlim_cur)
147 goto out_unlock;
148 get_file(file); /* We are now finished with oldfd */
149
150 err = expand_files(files, newfd);
151 if (err < 0)
152 goto out_fput;
153
154 /* To avoid races with open() and dup(), we will mark the fd as
155 * in-use in the open-file bitmap throughout the entire dup2()
156 * process. This is quite safe: do_close() uses the fd array
157 * entry, not the bitmap, to decide what work needs to be
158 * done. --sct */
159 /* Doesn't work. open() might be there first. --AV */
160
161 /* Yes. It's a race. In user space. Nothing sane to do */
162 err = -EBUSY;
163 tofree = files->fd[newfd];
164 if (!tofree && FD_ISSET(newfd, files->open_fds))
165 goto out_fput;
166
167 files->fd[newfd] = file;
168 FD_SET(newfd, files->open_fds);
169 FD_CLR(newfd, files->close_on_exec);
170 write_unlock(&files->file_lock);
171
172 if (tofree)
173 filp_close(tofree, files);
174 err = newfd;
175 out:
176 return err;
177 out_unlock:
178 write_unlock(&files->file_lock);
179 goto out;
180
181 out_fput:
182 write_unlock(&files->file_lock);
183 fput(file);
184 goto out;
185 }
186
187 asmlinkage long sys_dup(unsigned int fildes)
188 {
189 int ret = -EBADF;
190 struct file * file = fget(fildes);
191
192 if (file)
193 ret = dupfd(file, 0);
194 return ret;
195 }
196
197 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC)
198
199 static int setfl(int fd, struct file * filp, unsigned long arg)
200 {
201 struct inode * inode = filp->f_dentry->d_inode;
202 int error;
203
204 /*
205 * In the case of an append-only file, O_APPEND
206 * cannot be cleared
207 */
208 if (!(arg & O_APPEND) && IS_APPEND(inode))
209 return -EPERM;
210
211 /* Did FASYNC state change? */
212 if ((arg ^ filp->f_flags) & FASYNC) {
213 if (filp->f_op && filp->f_op->fasync) {
214 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
215 if (error < 0)
216 return error;
217 }
218 }
219
220 /* required for strict SunOS emulation */
221 if (O_NONBLOCK != O_NDELAY)
222 if (arg & O_NDELAY)
223 arg |= O_NONBLOCK;
224
225 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
226 return 0;
227 }
228
229 static long do_fcntl(unsigned int fd, unsigned int cmd,
230 unsigned long arg, struct file * filp)
231 {
232 long err = -EINVAL;
233
234 switch (cmd) {
235 case F_DUPFD:
236 if (arg < NR_OPEN) {
237 get_file(filp);
238 err = dupfd(filp, arg);
239 }
240 break;
241 case F_GETFD:
242 err = get_close_on_exec(fd);
243 break;
244 case F_SETFD:
245 err = 0;
246 set_close_on_exec(fd, arg&1);
247 break;
248 case F_GETFL:
249 err = filp->f_flags;
250 break;
251 case F_SETFL:
252 lock_kernel();
253 err = setfl(fd, filp, arg);
254 unlock_kernel();
255 break;
256 case F_GETLK:
257 err = fcntl_getlk(fd, (struct flock *) arg);
258 break;
259 case F_SETLK:
260 case F_SETLKW:
261 err = fcntl_setlk(fd, cmd, (struct flock *) arg);
262 break;
263 case F_GETOWN:
264 /*
265 * XXX If f_owner is a process group, the
266 * negative return value will get converted
267 * into an error. Oops. If we keep the
268 * current syscall conventions, the only way
269 * to fix this will be in libc.
270 */
271 err = filp->f_owner.pid;
272 break;
273 case F_SETOWN:
274 lock_kernel();
275 filp->f_owner.pid = arg;
276 filp->f_owner.uid = current->uid;
277 filp->f_owner.euid = current->euid;
278 err = 0;
279 if (S_ISSOCK (filp->f_dentry->d_inode->i_mode))
280 err = sock_fcntl (filp, F_SETOWN, arg);
281 unlock_kernel();
282 break;
283 case F_GETSIG:
284 err = filp->f_owner.signum;
285 break;
286 case F_SETSIG:
287 /* arg == 0 restores default behaviour. */
288 if (arg < 0 || arg > _NSIG) {
289 break;
290 }
291 err = 0;
292 filp->f_owner.signum = arg;
293 break;
294 case F_GETLEASE:
295 err = fcntl_getlease(filp);
296 break;
297 case F_SETLEASE:
298 err = fcntl_setlease(fd, filp, arg);
299 break;
300 case F_NOTIFY:
301 err = fcntl_dirnotify(fd, filp, arg);
302 break;
303 default:
304 /* sockets need a few special fcntls. */
305 err = -EINVAL;
306 if (S_ISSOCK (filp->f_dentry->d_inode->i_mode))
307 err = sock_fcntl (filp, cmd, arg);
308 break;
309 }
310
311 return err;
312 }
313
314 asmlinkage long sys_fcntl(unsigned int fd, unsigned int cmd, unsigned long arg)
315 {
316 struct file * filp;
317 long err = -EBADF;
318
319 filp = fget(fd);
320 if (!filp)
321 goto out;
322
323 err = do_fcntl(fd, cmd, arg, filp);
324
325 fput(filp);
326 out:
327 return err;
328 }
329
330 #if BITS_PER_LONG == 32
331 asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
332 {
333 struct file * filp;
334 long err;
335
336 err = -EBADF;
337 filp = fget(fd);
338 if (!filp)
339 goto out;
340
341 lock_kernel();
342 switch (cmd) {
343 case F_GETLK64:
344 err = fcntl_getlk64(fd, (struct flock64 *) arg);
345 break;
346 case F_SETLK64:
347 err = fcntl_setlk64(fd, cmd, (struct flock64 *) arg);
348 break;
349 case F_SETLKW64:
350 err = fcntl_setlk64(fd, cmd, (struct flock64 *) arg);
351 break;
352 default:
353 err = do_fcntl(fd, cmd, arg, filp);
354 break;
355 }
356 unlock_kernel();
357 fput(filp);
358 out:
359 return err;
360 }
361 #endif
362
363 /* Table to convert sigio signal codes into poll band bitmaps */
364
365 static long band_table[NSIGPOLL] = {
366 POLLIN | POLLRDNORM, /* POLL_IN */
367 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
368 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
369 POLLERR, /* POLL_ERR */
370 POLLPRI | POLLRDBAND, /* POLL_PRI */
371 POLLHUP | POLLERR /* POLL_HUP */
372 };
373
374 static void send_sigio_to_task(struct task_struct *p,
375 struct fown_struct *fown,
376 int fd,
377 int reason)
378 {
379 if ((fown->euid != 0) &&
380 (fown->euid ^ p->suid) && (fown->euid ^ p->uid) &&
381 (fown->uid ^ p->suid) && (fown->uid ^ p->uid))
382 return;
383 switch (fown->signum) {
384 siginfo_t si;
385 default:
386 /* Queue a rt signal with the appropriate fd as its
387 value. We use SI_SIGIO as the source, not
388 SI_KERNEL, since kernel signals always get
389 delivered even if we can't queue. Failure to
390 queue in this case _should_ be reported; we fall
391 back to SIGIO in that case. --sct */
392 si.si_signo = fown->signum;
393 si.si_errno = 0;
394 si.si_code = reason & ~__SI_MASK;
395 /* Make sure we are called with one of the POLL_*
396 reasons, otherwise we could leak kernel stack into
397 userspace. */
398 if ((reason & __SI_MASK) != __SI_POLL)
399 BUG();
400 if (reason - POLL_IN >= NSIGPOLL)
401 si.si_band = ~0L;
402 else
403 si.si_band = band_table[reason - POLL_IN];
404 si.si_fd = fd;
405 if (!send_sig_info(fown->signum, &si, p))
406 break;
407 /* fall-through: fall back on the old plain SIGIO signal */
408 case 0:
409 send_sig(SIGIO, p, 1);
410 }
411 }
412
413 void send_sigio(struct fown_struct *fown, int fd, int band)
414 {
415 struct task_struct * p;
416 int pid = fown->pid;
417
418 read_lock(&tasklist_lock);
419 if ( (pid > 0) && (p = find_task_by_pid(pid)) ) {
420 send_sigio_to_task(p, fown, fd, band);
421 goto out;
422 }
423 for_each_task(p) {
424 int match = p->pid;
425 if (pid < 0)
426 match = -p->pgrp;
427 if (pid != match)
428 continue;
429 send_sigio_to_task(p, fown, fd, band);
430 }
431 out:
432 read_unlock(&tasklist_lock);
433 }
434
435 static rwlock_t fasync_lock = RW_LOCK_UNLOCKED;
436 static kmem_cache_t *fasync_cache;
437
438 /*
439 * fasync_helper() is used by some character device drivers (mainly mice)
440 * to set up the fasync queue. It returns negative on error, 0 if it did
441 * no changes and positive if it added/deleted the entry.
442 */
443 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
444 {
445 struct fasync_struct *fa, **fp;
446 struct fasync_struct *new = NULL;
447 int result = 0;
448
449 if (on) {
450 new = kmem_cache_alloc(fasync_cache, SLAB_KERNEL);
451 if (!new)
452 return -ENOMEM;
453 }
454 write_lock_irq(&fasync_lock);
455 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
456 if (fa->fa_file == filp) {
457 if(on) {
458 fa->fa_fd = fd;
459 kmem_cache_free(fasync_cache, new);
460 } else {
461 *fp = fa->fa_next;
462 kmem_cache_free(fasync_cache, fa);
463 result = 1;
464 }
465 goto out;
466 }
467 }
468
469 if (on) {
470 new->magic = FASYNC_MAGIC;
471 new->fa_file = filp;
472 new->fa_fd = fd;
473 new->fa_next = *fapp;
474 *fapp = new;
475 result = 1;
476 }
477 out:
478 write_unlock_irq(&fasync_lock);
479 return result;
480 }
481
482 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
483 {
484 while (fa) {
485 struct fown_struct * fown;
486 if (fa->magic != FASYNC_MAGIC) {
487 printk(KERN_ERR "kill_fasync: bad magic number in "
488 "fasync_struct!\n");
489 return;
490 }
491 fown = &fa->fa_file->f_owner;
492 /* Don't send SIGURG to processes which have not set a
493 queued signum: SIGURG has its own default signalling
494 mechanism. */
495 if (fown->pid && !(sig == SIGURG && fown->signum == 0))
496 send_sigio(fown, fa->fa_fd, band);
497 fa = fa->fa_next;
498 }
499 }
500
501 void kill_fasync(struct fasync_struct **fp, int sig, int band)
502 {
503 read_lock(&fasync_lock);
504 __kill_fasync(*fp, sig, band);
505 read_unlock(&fasync_lock);
506 }
507
508 static int __init fasync_init(void)
509 {
510 fasync_cache = kmem_cache_create("fasync cache",
511 sizeof(struct fasync_struct), 0, 0, NULL, NULL);
512 if (!fasync_cache)
513 panic("cannot create fasync slab cache");
514 return 0;
515 }
516
517 module_init(fasync_init)
518
This page was automatically generated by the
LXR engine.
Visit the LXR main site for more
information.