4 * Copyright (C) 1991, 1992 Linus Torvalds
7 #include <linux/syscalls.h>
8 #include <linux/init.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>
18 #include <linux/vs_limit.h>
21 #include <asm/siginfo.h>
22 #include <asm/uaccess.h>
24 void fastcall set_close_on_exec(unsigned int fd, int flag)
26 struct files_struct *files = current->files;
27 spin_lock(&files->file_lock);
29 FD_SET(fd, files->close_on_exec);
31 FD_CLR(fd, files->close_on_exec);
32 spin_unlock(&files->file_lock);
35 static inline int get_close_on_exec(unsigned int fd)
37 struct files_struct *files = current->files;
39 spin_lock(&files->file_lock);
40 res = FD_ISSET(fd, files->close_on_exec);
41 spin_unlock(&files->file_lock);
46 * locate_fd finds a free file descriptor in the open_fds fdset,
47 * expanding the fd arrays if necessary. Must be called with the
48 * file_lock held for write.
51 static int locate_fd(struct files_struct *files,
52 struct file *file, unsigned int orig_start)
59 if (orig_start >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
64 * Someone might have closed fd's in the range
65 * orig_start..files->next_fd
68 if (start < files->next_fd)
69 start = files->next_fd;
72 if (start < files->max_fdset) {
73 newfd = find_next_zero_bit(files->open_fds->fds_bits,
74 files->max_fdset, start);
78 if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
80 if (!vx_files_avail(1))
83 error = expand_files(files, newfd);
88 * If we needed to expand the fs array we
89 * might have blocked - try again.
94 if (start <= files->next_fd)
95 files->next_fd = newfd + 1;
103 static int dupfd(struct file *file, unsigned int start)
105 struct files_struct * files = current->files;
108 spin_lock(&files->file_lock);
109 fd = locate_fd(files, file, start);
111 FD_SET(fd, files->open_fds);
112 FD_CLR(fd, files->close_on_exec);
113 spin_unlock(&files->file_lock);
115 fd_install(fd, file);
117 spin_unlock(&files->file_lock);
124 asmlinkage long sys_dup2(unsigned int oldfd, unsigned int newfd)
127 struct file * file, *tofree;
128 struct files_struct * files = current->files;
130 spin_lock(&files->file_lock);
131 if (!(file = fcheck(oldfd)))
137 if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
139 get_file(file); /* We are now finished with oldfd */
141 err = expand_files(files, newfd);
145 /* To avoid races with open() and dup(), we will mark the fd as
146 * in-use in the open-file bitmap throughout the entire dup2()
147 * process. This is quite safe: do_close() uses the fd array
148 * entry, not the bitmap, to decide what work needs to be
150 /* Doesn't work. open() might be there first. --AV */
152 /* Yes. It's a race. In user space. Nothing sane to do */
154 tofree = files->fd[newfd];
155 if (!tofree && FD_ISSET(newfd, files->open_fds))
158 files->fd[newfd] = file;
159 FD_SET(newfd, files->open_fds);
160 FD_CLR(newfd, files->close_on_exec);
161 spin_unlock(&files->file_lock);
164 filp_close(tofree, files);
166 vx_openfd_inc(newfd); /* fd was unused */
172 spin_unlock(&files->file_lock);
176 spin_unlock(&files->file_lock);
181 asmlinkage long sys_dup(unsigned int fildes)
184 struct file * file = fget(fildes);
187 ret = dupfd(file, 0);
191 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
193 static int setfl(int fd, struct file * filp, unsigned long arg)
195 struct inode * inode = filp->f_dentry->d_inode;
198 /* O_APPEND cannot be cleared if the file is marked as append-only */
199 if (!(arg & O_APPEND) && IS_APPEND(inode))
202 /* O_NOATIME can only be set by the owner or superuser */
203 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
204 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
207 /* required for strict SunOS emulation */
208 if (O_NONBLOCK != O_NDELAY)
212 if (arg & O_DIRECT) {
213 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
214 !filp->f_mapping->a_ops->direct_IO)
218 if (filp->f_op && filp->f_op->check_flags)
219 error = filp->f_op->check_flags(arg);
224 if ((arg ^ filp->f_flags) & FASYNC) {
225 if (filp->f_op && filp->f_op->fasync) {
226 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
232 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
238 static void f_modown(struct file *filp, unsigned long pid,
239 uid_t uid, uid_t euid, int force)
241 write_lock_irq(&filp->f_owner.lock);
242 if (force || !filp->f_owner.pid) {
243 filp->f_owner.pid = pid;
244 filp->f_owner.uid = uid;
245 filp->f_owner.euid = euid;
247 write_unlock_irq(&filp->f_owner.lock);
250 int f_setown(struct file *filp, unsigned long arg, int force)
254 err = security_file_set_fowner(filp);
258 f_modown(filp, arg, current->uid, current->euid, force);
262 EXPORT_SYMBOL(f_setown);
264 void f_delown(struct file *filp)
266 f_modown(filp, 0, 0, 0, 1);
269 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
277 err = dupfd(filp, arg);
280 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
284 set_close_on_exec(fd, arg & FD_CLOEXEC);
290 err = setfl(fd, filp, arg);
293 err = fcntl_getlk(filp, (struct flock __user *) arg);
297 err = fcntl_setlk(filp, cmd, (struct flock __user *) arg);
301 * XXX If f_owner is a process group, the
302 * negative return value will get converted
303 * into an error. Oops. If we keep the
304 * current syscall conventions, the only way
305 * to fix this will be in libc.
307 err = filp->f_owner.pid;
308 force_successful_syscall_return();
311 err = f_setown(filp, arg, 1);
314 err = filp->f_owner.signum;
317 /* arg == 0 restores default behaviour. */
318 if (arg < 0 || arg > _NSIG) {
322 filp->f_owner.signum = arg;
325 err = fcntl_getlease(filp);
328 err = fcntl_setlease(fd, filp, arg);
331 err = fcntl_dirnotify(fd, filp, arg);
339 asmlinkage long sys_fcntl(unsigned int fd, unsigned int cmd, unsigned long arg)
348 err = security_file_fcntl(filp, cmd, arg);
354 err = do_fcntl(fd, cmd, arg, filp);
361 #if BITS_PER_LONG == 32
362 asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
372 err = security_file_fcntl(filp, cmd, arg);
381 err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
385 err = fcntl_setlk64(filp, cmd, (struct flock64 __user *) arg);
388 err = do_fcntl(fd, cmd, arg, filp);
397 /* Table to convert sigio signal codes into poll band bitmaps */
399 static long band_table[NSIGPOLL] = {
400 POLLIN | POLLRDNORM, /* POLL_IN */
401 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
402 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
403 POLLERR, /* POLL_ERR */
404 POLLPRI | POLLRDBAND, /* POLL_PRI */
405 POLLHUP | POLLERR /* POLL_HUP */
408 static inline int sigio_perm(struct task_struct *p,
409 struct fown_struct *fown, int sig)
411 return (((fown->euid == 0) ||
412 (fown->euid == p->suid) || (fown->euid == p->uid) ||
413 (fown->uid == p->suid) || (fown->uid == p->uid)) &&
414 !security_file_send_sigiotask(p, fown, sig));
417 static void send_sigio_to_task(struct task_struct *p,
418 struct fown_struct *fown,
422 if (!sigio_perm(p, fown, fown->signum))
425 switch (fown->signum) {
428 /* Queue a rt signal with the appropriate fd as its
429 value. We use SI_SIGIO as the source, not
430 SI_KERNEL, since kernel signals always get
431 delivered even if we can't queue. Failure to
432 queue in this case _should_ be reported; we fall
433 back to SIGIO in that case. --sct */
434 si.si_signo = fown->signum;
437 /* Make sure we are called with one of the POLL_*
438 reasons, otherwise we could leak kernel stack into
440 if ((reason & __SI_MASK) != __SI_POLL)
442 if (reason - POLL_IN >= NSIGPOLL)
445 si.si_band = band_table[reason - POLL_IN];
447 if (!send_sig_info(fown->signum, &si, p))
449 /* fall-through: fall back on the old plain SIGIO signal */
451 send_group_sig_info(SIGIO, SEND_SIG_PRIV, p);
455 void send_sigio(struct fown_struct *fown, int fd, int band)
457 struct task_struct *p;
460 read_lock(&fown->lock);
463 goto out_unlock_fown;
465 read_lock(&tasklist_lock);
467 p = find_task_by_real_pid(pid);
469 send_sigio_to_task(p, fown, fd, band);
472 do_each_task_pid(-pid, PIDTYPE_PGID, p) {
473 send_sigio_to_task(p, fown, fd, band);
474 } while_each_task_pid(-pid, PIDTYPE_PGID, p);
476 read_unlock(&tasklist_lock);
478 read_unlock(&fown->lock);
481 static void send_sigurg_to_task(struct task_struct *p,
482 struct fown_struct *fown)
484 if (sigio_perm(p, fown, SIGURG))
485 send_group_sig_info(SIGURG, SEND_SIG_PRIV, p);
488 int send_sigurg(struct fown_struct *fown)
490 struct task_struct *p;
493 read_lock(&fown->lock);
496 goto out_unlock_fown;
500 read_lock(&tasklist_lock);
502 p = find_task_by_real_pid(pid);
504 send_sigurg_to_task(p, fown);
507 do_each_task_pid(-pid, PIDTYPE_PGID, p) {
508 send_sigurg_to_task(p, fown);
509 } while_each_task_pid(-pid, PIDTYPE_PGID, p);
511 read_unlock(&tasklist_lock);
513 read_unlock(&fown->lock);
517 static DEFINE_RWLOCK(fasync_lock);
518 static kmem_cache_t *fasync_cache;
521 * fasync_helper() is used by some character device drivers (mainly mice)
522 * to set up the fasync queue. It returns negative on error, 0 if it did
523 * no changes and positive if it added/deleted the entry.
525 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
527 struct fasync_struct *fa, **fp;
528 struct fasync_struct *new = NULL;
532 new = kmem_cache_alloc(fasync_cache, SLAB_KERNEL);
536 write_lock_irq(&fasync_lock);
537 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
538 if (fa->fa_file == filp) {
541 kmem_cache_free(fasync_cache, new);
544 kmem_cache_free(fasync_cache, fa);
552 new->magic = FASYNC_MAGIC;
555 new->fa_next = *fapp;
560 write_unlock_irq(&fasync_lock);
564 EXPORT_SYMBOL(fasync_helper);
566 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
569 struct fown_struct * fown;
570 if (fa->magic != FASYNC_MAGIC) {
571 printk(KERN_ERR "kill_fasync: bad magic number in "
575 fown = &fa->fa_file->f_owner;
576 /* Don't send SIGURG to processes which have not set a
577 queued signum: SIGURG has its own default signalling
579 if (!(sig == SIGURG && fown->signum == 0))
580 send_sigio(fown, fa->fa_fd, band);
585 EXPORT_SYMBOL(__kill_fasync);
587 void kill_fasync(struct fasync_struct **fp, int sig, int band)
589 /* First a quick test without locking: usually
593 read_lock(&fasync_lock);
594 /* reread *fp after obtaining the lock */
595 __kill_fasync(*fp, sig, band);
596 read_unlock(&fasync_lock);
599 EXPORT_SYMBOL(kill_fasync);
601 static int __init fasync_init(void)
603 fasync_cache = kmem_cache_create("fasync_cache",
604 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL, NULL);
608 module_init(fasync_init)