fs/fcntl.c

   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/security.h>
  18 #include <linux/ptrace.h>
  19 #include <linux/signal.h>
  20 #include <linux/rcupdate.h>
  21 #include <linux/pid_namespace.h>
  22
  23 #include <asm/poll.h>
  24 #include <asm/siginfo.h>
  25 #include <asm/uaccess.h>
  26
  27 void set_close_on_exec(unsigned int fd, int flag)
  28 {
  29         struct files_struct *files = current->files;
  30         struct fdtable *fdt;
  31         spin_lock(&files->file_lock);
  32         fdt = files_fdtable(files);
  33         if (flag)
  34                 FD_SET(fd, fdt->close_on_exec);
  35         else
  36                 FD_CLR(fd, fdt->close_on_exec);
  37         spin_unlock(&files->file_lock);
  38 }
  39
  40 static int get_close_on_exec(unsigned int fd)
  41 {
  42         struct files_struct *files = current->files;
  43         struct fdtable *fdt;
  44         int res;
  45         rcu_read_lock();
  46         fdt = files_fdtable(files);
  47         res = FD_ISSET(fd, fdt->close_on_exec);
  48         rcu_read_unlock();
  49         return res;
  50 }
  51
  52 asmlinkage long sys_dup3(unsigned int oldfd, unsigned int newfd, int flags)
  53 {
  54         int err = -EBADF;
  55         struct file * file, *tofree;
  56         struct files_struct * files = current->files;
  57         struct fdtable *fdt;
  58
  59         if ((flags & ~O_CLOEXEC) != 0)
  60                 return -EINVAL;
  61
  62         if (unlikely(oldfd == newfd))
  63                 return -EINVAL;
  64
  65         spin_lock(&files->file_lock);
  66         if (!(file = fcheck(oldfd)))
  67                 goto out_unlock;
  68         get_file(file);                 /* We are now finished with oldfd */
  69
  70         err = expand_files(files, newfd);
  71         if (unlikely(err < 0)) {
  72                 if (err == -EMFILE)
  73                         err = -EBADF;
  74                 goto out_fput;
  75         }
  76
  77         /* To avoid races with open() and dup(), we will mark the fd as
  78          * in-use in the open-file bitmap throughout the entire dup2()
  79          * process.  This is quite safe: do_close() uses the fd array
  80          * entry, not the bitmap, to decide what work needs to be
  81          * done.  --sct */
  82         /* Doesn't work. open() might be there first. --AV */
  83
  84         /* Yes. It's a race. In user space. Nothing sane to do */
  85         err = -EBUSY;
  86         fdt = files_fdtable(files);
  87         tofree = fdt->fd[newfd];
  88         if (!tofree && FD_ISSET(newfd, fdt->open_fds))
  89                 goto out_fput;
  90
  91         rcu_assign_pointer(fdt->fd[newfd], file);
  92         FD_SET(newfd, fdt->open_fds);
  93         if (flags & O_CLOEXEC)
  94                 FD_SET(newfd, fdt->close_on_exec);
  95         else
  96                 FD_CLR(newfd, fdt->close_on_exec);
  97         spin_unlock(&files->file_lock);
  98
  99         if (tofree)
 100                 filp_close(tofree, files);
 101         err = newfd;
 102 out:
 103         return err;
 104 out_unlock:
 105         spin_unlock(&files->file_lock);
 106         goto out;
 107
 108 out_fput:
 109         spin_unlock(&files->file_lock);
 110         fput(file);
 111         goto out;
 112 }
 113
 114 asmlinkage long sys_dup2(unsigned int oldfd, unsigned int newfd)
 115 {
 116         if (unlikely(newfd == oldfd)) { /* corner case */
 117                 struct files_struct *files = current->files;
 118                 rcu_read_lock();
 119                 if (!fcheck_files(files, oldfd))
 120                         oldfd = -EBADF;
 121                 rcu_read_unlock();
 122                 return oldfd;
 123         }
 124         return sys_dup3(oldfd, newfd, 0);
 125 }
 126
 127 asmlinkage long sys_dup(unsigned int fildes)
 128 {
 129         int ret = -EBADF;
 130         struct file *file = fget(fildes);
 131
 132         if (file) {
 133                 ret = get_unused_fd();
 134                 if (ret >= 0)
 135                         fd_install(ret, file);
 136                 else
 137                         fput(file);
 138         }
 139         return ret;
 140 }
 141
 142 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
 143
 144 static int setfl(int fd, struct file * filp, unsigned long arg)
 145 {
 146         struct inode * inode = filp->f_path.dentry->d_inode;
 147         int error = 0;
 148
 149         /*
 150          * O_APPEND cannot be cleared if the file is marked as append-only
 151          * and the file is open for write.
 152          */
 153         if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
 154                 return -EPERM;
 155
 156         /* O_NOATIME can only be set by the owner or superuser */
 157         if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
 158                 if (!is_owner_or_cap(inode))
 159                         return -EPERM;
 160
 161         /* required for strict SunOS emulation */
 162         if (O_NONBLOCK != O_NDELAY)
 163                if (arg & O_NDELAY)
 164                    arg |= O_NONBLOCK;
 165
 166         if (arg & O_DIRECT) {
 167                 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
 168                         !filp->f_mapping->a_ops->direct_IO)
 169                                 return -EINVAL;
 170         }
 171
 172         if (filp->f_op && filp->f_op->check_flags)
 173                 error = filp->f_op->check_flags(arg);
 174         if (error)
 175                 return error;
 176
 177         if ((arg ^ filp->f_flags) & FASYNC) {
 178                 if (filp->f_op && filp->f_op->fasync) {
 179                         error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
 180                         if (error < 0)
 181                                 goto out;
 182                 }
 183         }
 184
 185         filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
 186  out:
 187         return error;
 188 }
 189
 190 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
 191                      uid_t uid, uid_t euid, int force)
 192 {
 193         write_lock_irq(&filp->f_owner.lock);
 194         if (force || !filp->f_owner.pid) {
 195                 put_pid(filp->f_owner.pid);
 196                 filp->f_owner.pid = get_pid(pid);
 197                 filp->f_owner.pid_type = type;
 198                 filp->f_owner.uid = uid;
 199                 filp->f_owner.euid = euid;
 200         }
 201         write_unlock_irq(&filp->f_owner.lock);
 202 }
 203
 204 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
 205                 int force)
 206 {
 207         int err;
 208
 209         err = security_file_set_fowner(filp);
 210         if (err)
 211                 return err;
 212
 213         f_modown(filp, pid, type, current->uid, current->euid, force);
 214         return 0;
 215 }
 216 EXPORT_SYMBOL(__f_setown);
 217
 218 int f_setown(struct file *filp, unsigned long arg, int force)
 219 {
 220         enum pid_type type;
 221         struct pid *pid;
 222         int who = arg;
 223         int result;
 224         type = PIDTYPE_PID;
 225         if (who < 0) {
 226                 type = PIDTYPE_PGID;
 227                 who = -who;
 228         }
 229         rcu_read_lock();
 230         pid = find_vpid(who);
 231         result = __f_setown(filp, pid, type, force);
 232         rcu_read_unlock();
 233         return result;
 234 }
 235 EXPORT_SYMBOL(f_setown);
 236
 237 void f_delown(struct file *filp)
 238 {
 239         f_modown(filp, NULL, PIDTYPE_PID, 0, 0, 1);
 240 }
 241
 242 pid_t f_getown(struct file *filp)
 243 {
 244         pid_t pid;
 245         read_lock(&filp->f_owner.lock);
 246         pid = pid_vnr(filp->f_owner.pid);
 247         if (filp->f_owner.pid_type == PIDTYPE_PGID)
 248                 pid = -pid;
 249         read_unlock(&filp->f_owner.lock);
 250         return pid;
 251 }
 252
 253 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
 254                 struct file *filp)
 255 {
 256         long err = -EINVAL;
 257
 258         switch (cmd) {
 259         case F_DUPFD:
 260         case F_DUPFD_CLOEXEC:
 261                 if (arg >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
 262                         break;
 263                 err = alloc_fd(arg, cmd == F_DUPFD_CLOEXEC ? O_CLOEXEC : 0);
 264                 if (err >= 0) {
 265                         get_file(filp);
 266                         fd_install(err, filp);
 267                 }
 268                 break;
 269         case F_GETFD:
 270                 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
 271                 break;
 272         case F_SETFD:
 273                 err = 0;
 274                 set_close_on_exec(fd, arg & FD_CLOEXEC);
 275                 break;
 276         case F_GETFL:
 277                 err = filp->f_flags;
 278                 break;
 279         case F_SETFL:
 280                 err = setfl(fd, filp, arg);
 281                 break;
 282         case F_GETLK:
 283                 err = fcntl_getlk(filp, (struct flock __user *) arg);
 284                 break;
 285         case F_SETLK:
 286         case F_SETLKW:
 287                 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
 288                 break;
 289         case F_GETOWN:
 290                 /*
 291                  * XXX If f_owner is a process group, the
 292                  * negative return value will get converted
 293                  * into an error.  Oops.  If we keep the
 294                  * current syscall conventions, the only way
 295                  * to fix this will be in libc.
 296                  */
 297                 err = f_getown(filp);
 298                 force_successful_syscall_return();
 299                 break;
 300         case F_SETOWN:
 301                 err = f_setown(filp, arg, 1);
 302                 break;
 303         case F_GETSIG:
 304                 err = filp->f_owner.signum;
 305                 break;
 306         case F_SETSIG:
 307                 /* arg == 0 restores default behaviour. */
 308                 if (!valid_signal(arg)) {
 309                         break;
 310                 }
 311                 err = 0;
 312                 filp->f_owner.signum = arg;
 313                 break;
 314         case F_GETLEASE:
 315                 err = fcntl_getlease(filp);
 316                 break;
 317         case F_SETLEASE:
 318                 err = fcntl_setlease(fd, filp, arg);
 319                 break;
 320         case F_NOTIFY:
 321                 err = fcntl_dirnotify(fd, filp, arg);
 322                 break;
 323         default:
 324                 break;
 325         }
 326         return err;
 327 }
 328
 329 asmlinkage long sys_fcntl(unsigned int fd, unsigned int cmd, unsigned long arg)
 330 {
 331         struct file *filp;
 332         long err = -EBADF;
 333
 334         filp = fget(fd);
 335         if (!filp)
 336                 goto out;
 337
 338         err = security_file_fcntl(filp, cmd, arg);
 339         if (err) {
 340                 fput(filp);
 341                 return err;
 342         }
 343
 344         err = do_fcntl(fd, cmd, arg, filp);
 345
 346         fput(filp);
 347 out:
 348         return err;
 349 }
 350
 351 #if BITS_PER_LONG == 32
 352 asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
 353 {
 354         struct file * filp;
 355         long err;
 356
 357         err = -EBADF;
 358         filp = fget(fd);
 359         if (!filp)
 360                 goto out;
 361
 362         err = security_file_fcntl(filp, cmd, arg);
 363         if (err) {
 364                 fput(filp);
 365                 return err;
 366         }
 367         err = -EBADF;
 368
 369         switch (cmd) {
 370                 case F_GETLK64:
 371                         err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
 372                         break;
 373                 case F_SETLK64:
 374                 case F_SETLKW64:
 375                         err = fcntl_setlk64(fd, filp, cmd,
 376                                         (struct flock64 __user *) arg);
 377                         break;
 378                 default:
 379                         err = do_fcntl(fd, cmd, arg, filp);
 380                         break;
 381         }
 382         fput(filp);
 383 out:
 384         return err;
 385 }
 386 #endif
 387
 388 /* Table to convert sigio signal codes into poll band bitmaps */
 389
 390 static const long band_table[NSIGPOLL] = {
 391         POLLIN | POLLRDNORM,                    /* POLL_IN */
 392         POLLOUT | POLLWRNORM | POLLWRBAND,      /* POLL_OUT */
 393         POLLIN | POLLRDNORM | POLLMSG,          /* POLL_MSG */
 394         POLLERR,                                /* POLL_ERR */
 395         POLLPRI | POLLRDBAND,                   /* POLL_PRI */
 396         POLLHUP | POLLERR                       /* POLL_HUP */
 397 };
 398
 399 static inline int sigio_perm(struct task_struct *p,
 400                              struct fown_struct *fown, int sig)
 401 {
 402         return (((fown->euid == 0) ||
 403                  (fown->euid == p->suid) || (fown->euid == p->uid) ||
 404                  (fown->uid == p->suid) || (fown->uid == p->uid)) &&
 405                 !security_file_send_sigiotask(p, fown, sig));
 406 }
 407
 408 static void send_sigio_to_task(struct task_struct *p,
 409                                struct fown_struct *fown,
 410                                int fd,
 411                                int reason)
 412 {
 413         if (!sigio_perm(p, fown, fown->signum))
 414                 return;
 415
 416         switch (fown->signum) {
 417                 siginfo_t si;
 418                 default:
 419                         /* Queue a rt signal with the appropriate fd as its
 420                            value.  We use SI_SIGIO as the source, not
 421                            SI_KERNEL, since kernel signals always get
 422                            delivered even if we can't queue.  Failure to
 423                            queue in this case _should_ be reported; we fall
 424                            back to SIGIO in that case. --sct */
 425                         si.si_signo = fown->signum;
 426                         si.si_errno = 0;
 427                         si.si_code  = reason;
 428                         /* Make sure we are called with one of the POLL_*
 429                            reasons, otherwise we could leak kernel stack into
 430                            userspace.  */
 431                         BUG_ON((reason & __SI_MASK) != __SI_POLL);
 432                         if (reason - POLL_IN >= NSIGPOLL)
 433                                 si.si_band  = ~0L;
 434                         else
 435                                 si.si_band = band_table[reason - POLL_IN];
 436                         si.si_fd    = fd;
 437                         if (!group_send_sig_info(fown->signum, &si, p))
 438                                 break;
 439                 /* fall-through: fall back on the old plain SIGIO signal */
 440                 case 0:
 441                         group_send_sig_info(SIGIO, SEND_SIG_PRIV, p);
 442         }
 443 }
 444
 445 void send_sigio(struct fown_struct *fown, int fd, int band)
 446 {
 447         struct task_struct *p;
 448         enum pid_type type;
 449         struct pid *pid;
 450
 451         read_lock(&fown->lock);
 452         type = fown->pid_type;
 453         pid = fown->pid;
 454         if (!pid)
 455                 goto out_unlock_fown;
 456
 457         read_lock(&tasklist_lock);
 458         do_each_pid_task(pid, type, p) {
 459                 send_sigio_to_task(p, fown, fd, band);
 460         } while_each_pid_task(pid, type, p);
 461         read_unlock(&tasklist_lock);
 462  out_unlock_fown:
 463         read_unlock(&fown->lock);
 464 }
 465
 466 static void send_sigurg_to_task(struct task_struct *p,
 467                                 struct fown_struct *fown)
 468 {
 469         if (sigio_perm(p, fown, SIGURG))
 470                 group_send_sig_info(SIGURG, SEND_SIG_PRIV, p);
 471 }
 472
 473 int send_sigurg(struct fown_struct *fown)
 474 {
 475         struct task_struct *p;
 476         enum pid_type type;
 477         struct pid *pid;
 478         int ret = 0;
 479
 480         read_lock(&fown->lock);
 481         type = fown->pid_type;
 482         pid = fown->pid;
 483         if (!pid)
 484                 goto out_unlock_fown;
 485
 486         ret = 1;
 487
 488         read_lock(&tasklist_lock);
 489         do_each_pid_task(pid, type, p) {
 490                 send_sigurg_to_task(p, fown);
 491         } while_each_pid_task(pid, type, p);
 492         read_unlock(&tasklist_lock);
 493  out_unlock_fown:
 494         read_unlock(&fown->lock);
 495         return ret;
 496 }
 497
 498 static DEFINE_RWLOCK(fasync_lock);
 499 static struct kmem_cache *fasync_cache __read_mostly;
 500
 501 /*
 502  * fasync_helper() is used by some character device drivers (mainly mice)
 503  * to set up the fasync queue. It returns negative on error, 0 if it did
 504  * no changes and positive if it added/deleted the entry.
 505  */
 506 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
 507 {
 508         struct fasync_struct *fa, **fp;
 509         struct fasync_struct *new = NULL;
 510         int result = 0;
 511
 512         if (on) {
 513                 new = kmem_cache_alloc(fasync_cache, GFP_KERNEL);
 514                 if (!new)
 515                         return -ENOMEM;
 516         }
 517         write_lock_irq(&fasync_lock);
 518         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 519                 if (fa->fa_file == filp) {
 520                         if(on) {
 521                                 fa->fa_fd = fd;
 522                                 kmem_cache_free(fasync_cache, new);
 523                         } else {
 524                                 *fp = fa->fa_next;
 525                                 kmem_cache_free(fasync_cache, fa);
 526                                 result = 1;
 527                         }
 528                         goto out;
 529                 }
 530         }
 531
 532         if (on) {
 533                 new->magic = FASYNC_MAGIC;
 534                 new->fa_file = filp;
 535                 new->fa_fd = fd;
 536                 new->fa_next = *fapp;
 537                 *fapp = new;
 538                 result = 1;
 539         }
 540 out:
 541         write_unlock_irq(&fasync_lock);
 542         return result;
 543 }
 544
 545 EXPORT_SYMBOL(fasync_helper);
 546
 547 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
 548 {
 549         while (fa) {
 550                 struct fown_struct * fown;
 551                 if (fa->magic != FASYNC_MAGIC) {
 552                         printk(KERN_ERR "kill_fasync: bad magic number in "
 553                                "fasync_struct!\n");
 554                         return;
 555                 }
 556                 fown = &fa->fa_file->f_owner;
 557                 /* Don't send SIGURG to processes which have not set a
 558                    queued signum: SIGURG has its own default signalling
 559                    mechanism. */
 560                 if (!(sig == SIGURG && fown->signum == 0))
 561                         send_sigio(fown, fa->fa_fd, band);
 562                 fa = fa->fa_next;
 563         }
 564 }
 565
 566 EXPORT_SYMBOL(__kill_fasync);
 567
 568 void kill_fasync(struct fasync_struct **fp, int sig, int band)
 569 {
 570         /* First a quick test without locking: usually
 571          * the list is empty.
 572          */
 573         if (*fp) {
 574                 read_lock(&fasync_lock);
 575                 /* reread *fp after obtaining the lock */
 576                 __kill_fasync(*fp, sig, band);
 577                 read_unlock(&fasync_lock);
 578         }
 579 }
 580 EXPORT_SYMBOL(kill_fasync);
 581
 582 static int __init fasync_init(void)
 583 {
 584         fasync_cache = kmem_cache_create("fasync_cache",
 585                 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
 586         return 0;
 587 }
 588
 589 module_init(fasync_init)