release/src-rt-6.x/linux/linux-2.6/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/capability.h>
  13 #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>
  19 #include <linux/signal.h>
  20 #include <linux/rcupdate.h>
  21
  22 #include <asm/poll.h>
  23 #include <asm/siginfo.h>
  24 #include <asm/uaccess.h>
  25
  26 void fastcall set_close_on_exec(unsigned int fd, int flag)
  27 {
  28         struct files_struct *files = current->files;
  29         struct fdtable *fdt;
  30         spin_lock(&files->file_lock);
  31         fdt = files_fdtable(files);
  32         if (flag)
  33                 FD_SET(fd, fdt->close_on_exec);
  34         else
  35                 FD_CLR(fd, fdt->close_on_exec);
  36         spin_unlock(&files->file_lock);
  37 }
  38
  39 static int get_close_on_exec(unsigned int fd)
  40 {
  41         struct files_struct *files = current->files;
  42         struct fdtable *fdt;
  43         int res;
  44         rcu_read_lock();
  45         fdt = files_fdtable(files);
  46         res = FD_ISSET(fd, fdt->close_on_exec);
  47         rcu_read_unlock();
  48         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
  57 static 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;
  63         struct fdtable *fdt;
  64
  65         error = -EINVAL;
  66         if (orig_start >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
  67                 goto out;
  68
  69 repeat:
  70         fdt = files_fdtable(files);
  71         /*
  72          * Someone might have closed fd's in the range
  73          * orig_start..fdt->next_fd
  74          */
  75         start = orig_start;
  76         if (start < files->next_fd)
  77                 start = files->next_fd;
  78
  79         newfd = start;
  80         if (start < fdt->max_fds)
  81                 newfd = find_next_zero_bit(fdt->open_fds->fds_bits,
  82                                            fdt->max_fds, start);
  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
  99         /*
 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         if (start <= files->next_fd)
 105                 files->next_fd = newfd + 1;
 106
 107         error = newfd;
 108
 109 out:
 110         return error;
 111 }
 112
 113 static int dupfd(struct file *file, unsigned int start, int cloexec)
 114 {
 115         struct files_struct * files = current->files;
 116         struct fdtable *fdt;
 117         int fd;
 118
 119         spin_lock(&files->file_lock);
 120         fd = locate_fd(files, file, start);
 121         if (fd >= 0) {
 122                 /* locate_fd() may have expanded fdtable, load the ptr */
 123                 fdt = files_fdtable(files);
 124                 FD_SET(fd, fdt->open_fds);
 125                 if (cloexec)
 126                         FD_SET(fd, fdt->close_on_exec);
 127                 else
 128                         FD_CLR(fd, fdt->close_on_exec);
 129                 spin_unlock(&files->file_lock);
 130                 fd_install(fd, file);
 131         } else {
 132                 spin_unlock(&files->file_lock);
 133                 fput(file);
 134         }
 135
 136         return fd;
 137 }
 138
 139 asmlinkage long sys_dup2(unsigned int oldfd, unsigned int newfd)
 140 {
 141         int err = -EBADF;
 142         struct file * file, *tofree;
 143         struct files_struct * files = current->files;
 144         struct fdtable *fdt;
 145
 146         spin_lock(&files->file_lock);
 147         if (!(file = fcheck(oldfd)))
 148                 goto out_unlock;
 149         err = newfd;
 150         if (newfd == oldfd)
 151                 goto out_unlock;
 152         err = -EBADF;
 153         if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
 154                 goto out_unlock;
 155         get_file(file);                 /* We are now finished with oldfd */
 156
 157         err = expand_files(files, newfd);
 158         if (err < 0)
 159                 goto out_fput;
 160
 161         /* To avoid races with open() and dup(), we will mark the fd as
 162          * in-use in the open-file bitmap throughout the entire dup2()
 163          * process.  This is quite safe: do_close() uses the fd array
 164          * entry, not the bitmap, to decide what work needs to be
 165          * done.  --sct */
 166         /* Doesn't work. open() might be there first. --AV */
 167
 168         /* Yes. It's a race. In user space. Nothing sane to do */
 169         err = -EBUSY;
 170         fdt = files_fdtable(files);
 171         tofree = fdt->fd[newfd];
 172         if (!tofree && FD_ISSET(newfd, fdt->open_fds))
 173                 goto out_fput;
 174
 175         rcu_assign_pointer(fdt->fd[newfd], file);
 176         FD_SET(newfd, fdt->open_fds);
 177         FD_CLR(newfd, fdt->close_on_exec);
 178         spin_unlock(&files->file_lock);
 179
 180         if (tofree)
 181                 filp_close(tofree, files);
 182         err = newfd;
 183 out:
 184         return err;
 185 out_unlock:
 186         spin_unlock(&files->file_lock);
 187         goto out;
 188
 189 out_fput:
 190         spin_unlock(&files->file_lock);
 191         fput(file);
 192         goto out;
 193 }
 194
 195 asmlinkage long sys_dup(unsigned int fildes)
 196 {
 197         int ret = -EBADF;
 198         struct file * file = fget(fildes);
 199
 200         if (file)
 201                 ret = dupfd(file, 0, 0);
 202         return ret;
 203 }
 204
 205 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
 206
 207 static int setfl(int fd, struct file * filp, unsigned long arg)
 208 {
 209         struct inode * inode = filp->f_path.dentry->d_inode;
 210         int error = 0;
 211
 212         /*
 213          * O_APPEND cannot be cleared if the file is marked as append-only
 214          * and the file is open for write.
 215          */
 216         if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
 217                 return -EPERM;
 218
 219         /* O_NOATIME can only be set by the owner or superuser */
 220         if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
 221                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
 222                         return -EPERM;
 223
 224         /* required for strict SunOS emulation */
 225         if (O_NONBLOCK != O_NDELAY)
 226                if (arg & O_NDELAY)
 227                    arg |= O_NONBLOCK;
 228
 229         if (arg & O_DIRECT) {
 230                 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
 231                         !filp->f_mapping->a_ops->direct_IO)
 232                                 return -EINVAL;
 233         }
 234
 235         if (filp->f_op && filp->f_op->check_flags)
 236                 error = filp->f_op->check_flags(arg);
 237         if (error)
 238                 return error;
 239
 240         lock_kernel();
 241         if ((arg ^ filp->f_flags) & FASYNC) {
 242                 if (filp->f_op && filp->f_op->fasync) {
 243                         error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
 244                         if (error < 0)
 245                                 goto out;
 246                 }
 247         }
 248
 249         filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
 250  out:
 251         unlock_kernel();
 252         return error;
 253 }
 254
 255 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
 256                      uid_t uid, uid_t euid, int force)
 257 {
 258         write_lock_irq(&filp->f_owner.lock);
 259         if (force || !filp->f_owner.pid) {
 260                 put_pid(filp->f_owner.pid);
 261                 filp->f_owner.pid = get_pid(pid);
 262                 filp->f_owner.pid_type = type;
 263                 filp->f_owner.uid = uid;
 264                 filp->f_owner.euid = euid;
 265         }
 266         write_unlock_irq(&filp->f_owner.lock);
 267 }
 268
 269 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
 270                 int force)
 271 {
 272         int err;
 273
 274         err = security_file_set_fowner(filp);
 275         if (err)
 276                 return err;
 277
 278         f_modown(filp, pid, type, current->uid, current->euid, force);
 279         return 0;
 280 }
 281 EXPORT_SYMBOL(__f_setown);
 282
 283 int f_setown(struct file *filp, unsigned long arg, int force)
 284 {
 285         enum pid_type type;
 286         struct pid *pid;
 287         int who = arg;
 288         int result;
 289         type = PIDTYPE_PID;
 290         if (who < 0) {
 291                 type = PIDTYPE_PGID;
 292                 who = -who;
 293         }
 294         rcu_read_lock();
 295         pid = find_pid(who);
 296 #if 0
 297         if (pid == NULL)
 298                 printk("dbg: NULL pid\n");
 299         else
 300                 printk("dbg: pid = %d\n", pid);
 301 #endif
 302         result = __f_setown(filp, pid, type, force);
 303         rcu_read_unlock();
 304         return result;
 305 }
 306 EXPORT_SYMBOL(f_setown);
 307
 308 void f_delown(struct file *filp)
 309 {
 310         f_modown(filp, NULL, PIDTYPE_PID, 0, 0, 1);
 311 }
 312
 313 pid_t f_getown(struct file *filp)
 314 {
 315         pid_t pid;
 316         read_lock(&filp->f_owner.lock);
 317         pid = pid_nr(filp->f_owner.pid);
 318         if (filp->f_owner.pid_type == PIDTYPE_PGID)
 319                 pid = -pid;
 320         read_unlock(&filp->f_owner.lock);
 321         return pid;
 322 }
 323
 324 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
 325                 struct file *filp)
 326 {
 327         long err = -EINVAL;
 328
 329         switch (cmd) {
 330         case F_DUPFD:
 331         case F_DUPFD_CLOEXEC:
 332                 get_file(filp);
 333                 err = dupfd(filp, arg, cmd == F_DUPFD_CLOEXEC);
 334                 break;
 335         case F_GETFD:
 336                 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
 337                 break;
 338         case F_SETFD:
 339                 err = 0;
 340                 set_close_on_exec(fd, arg & FD_CLOEXEC);
 341                 break;
 342         case F_GETFL:
 343                 err = filp->f_flags;
 344                 break;
 345         case F_SETFL:
 346                 err = setfl(fd, filp, arg);
 347                 break;
 348         case F_GETLK:
 349                 err = fcntl_getlk(filp, (struct flock __user *) arg);
 350                 break;
 351         case F_SETLK:
 352         case F_SETLKW:
 353                 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
 354                 break;
 355         case F_GETOWN:
 356                 /*
 357                  * XXX If f_owner is a process group, the
 358                  * negative return value will get converted
 359                  * into an error.  Oops.  If we keep the
 360                  * current syscall conventions, the only way
 361                  * to fix this will be in libc.
 362                  */
 363                 err = f_getown(filp);
 364                 force_successful_syscall_return();
 365                 break;
 366         case F_SETOWN:
 367                 err = f_setown(filp, arg, 1);
 368                 break;
 369         case F_GETSIG:
 370                 err = filp->f_owner.signum;
 371                 break;
 372         case F_SETSIG:
 373                 /* arg == 0 restores default behaviour. */
 374                 if (!valid_signal(arg)) {
 375                         break;
 376                 }
 377                 err = 0;
 378                 filp->f_owner.signum = arg;
 379                 break;
 380         case F_GETLEASE:
 381                 err = fcntl_getlease(filp);
 382                 break;
 383         case F_SETLEASE:
 384                 err = fcntl_setlease(fd, filp, arg);
 385                 break;
 386         case F_NOTIFY:
 387                 err = fcntl_dirnotify(fd, filp, arg);
 388                 break;
 389         default:
 390                 break;
 391         }
 392         return err;
 393 }
 394
 395 asmlinkage long sys_fcntl(unsigned int fd, unsigned int cmd, unsigned long arg)
 396 {
 397         struct file *filp;
 398         long err = -EBADF;
 399
 400         filp = fget(fd);
 401         if (!filp)
 402                 goto out;
 403
 404         err = security_file_fcntl(filp, cmd, arg);
 405         if (err) {
 406                 fput(filp);
 407                 return err;
 408         }
 409
 410         err = do_fcntl(fd, cmd, arg, filp);
 411
 412         fput(filp);
 413 out:
 414         return err;
 415 }
 416
 417 #if BITS_PER_LONG == 32
 418 asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
 419 {
 420         struct file * filp;
 421         long err;
 422
 423         err = -EBADF;
 424         filp = fget(fd);
 425         if (!filp)
 426                 goto out;
 427
 428         err = security_file_fcntl(filp, cmd, arg);
 429         if (err) {
 430                 fput(filp);
 431                 return err;
 432         }
 433         err = -EBADF;
 434
 435         switch (cmd) {
 436                 case F_GETLK64:
 437                         err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
 438                         break;
 439                 case F_SETLK64:
 440                 case F_SETLKW64:
 441                         err = fcntl_setlk64(fd, filp, cmd,
 442                                         (struct flock64 __user *) arg);
 443                         break;
 444                 default:
 445                         err = do_fcntl(fd, cmd, arg, filp);
 446                         break;
 447         }
 448         fput(filp);
 449 out:
 450         return err;
 451 }
 452 #endif
 453
 454 /* Table to convert sigio signal codes into poll band bitmaps */
 455
 456 static const long band_table[NSIGPOLL] = {
 457         POLLIN | POLLRDNORM,                    /* POLL_IN */
 458         POLLOUT | POLLWRNORM | POLLWRBAND,      /* POLL_OUT */
 459         POLLIN | POLLRDNORM | POLLMSG,          /* POLL_MSG */
 460         POLLERR,                                /* POLL_ERR */
 461         POLLPRI | POLLRDBAND,                   /* POLL_PRI */
 462         POLLHUP | POLLERR                       /* POLL_HUP */
 463 };
 464
 465 static inline int sigio_perm(struct task_struct *p,
 466                              struct fown_struct *fown, int sig)
 467 {
 468         return (((fown->euid == 0) ||
 469                  (fown->euid == p->suid) || (fown->euid == p->uid) ||
 470                  (fown->uid == p->suid) || (fown->uid == p->uid)) &&
 471                 !security_file_send_sigiotask(p, fown, sig));
 472 }
 473
 474 static void send_sigio_to_task(struct task_struct *p,
 475                                struct fown_struct *fown,
 476                                int fd,
 477                                int reason)
 478 {
 479         if (!sigio_perm(p, fown, fown->signum))
 480                 return;
 481
 482         switch (fown->signum) {
 483                 siginfo_t si;
 484                 default:
 485                         /* Queue a rt signal with the appropriate fd as its
 486                            value.  We use SI_SIGIO as the source, not
 487                            SI_KERNEL, since kernel signals always get
 488                            delivered even if we can't queue.  Failure to
 489                            queue in this case _should_ be reported; we fall
 490                            back to SIGIO in that case. --sct */
 491                         si.si_signo = fown->signum;
 492                         si.si_errno = 0;
 493                         si.si_code  = reason;
 494                         /* Make sure we are called with one of the POLL_*
 495                            reasons, otherwise we could leak kernel stack into
 496                            userspace.  */
 497                         BUG_ON((reason & __SI_MASK) != __SI_POLL);
 498                         if (reason - POLL_IN >= NSIGPOLL)
 499                                 si.si_band  = ~0L;
 500                         else
 501                                 si.si_band = band_table[reason - POLL_IN];
 502                         si.si_fd    = fd;
 503                         if (!group_send_sig_info(fown->signum, &si, p))
 504                                 break;
 505                 /* fall-through: fall back on the old plain SIGIO signal */
 506                 case 0:
 507                         group_send_sig_info(SIGIO, SEND_SIG_PRIV, p);
 508         }
 509 }
 510
 511 void send_sigio(struct fown_struct *fown, int fd, int band)
 512 {
 513         struct task_struct *p;
 514         enum pid_type type;
 515         struct pid *pid;
 516
 517         read_lock(&fown->lock);
 518         type = fown->pid_type;
 519         pid = fown->pid;
 520         if (!pid)
 521                 goto out_unlock_fown;
 522
 523         read_lock(&tasklist_lock);
 524         do_each_pid_task(pid, type, p) {
 525                 send_sigio_to_task(p, fown, fd, band);
 526         } while_each_pid_task(pid, type, p);
 527         read_unlock(&tasklist_lock);
 528  out_unlock_fown:
 529         read_unlock(&fown->lock);
 530 }
 531
 532 static void send_sigurg_to_task(struct task_struct *p,
 533                                 struct fown_struct *fown)
 534 {
 535         if (sigio_perm(p, fown, SIGURG))
 536                 group_send_sig_info(SIGURG, SEND_SIG_PRIV, p);
 537 }
 538
 539 int send_sigurg(struct fown_struct *fown)
 540 {
 541         struct task_struct *p;
 542         enum pid_type type;
 543         struct pid *pid;
 544         int ret = 0;
 545
 546         read_lock(&fown->lock);
 547         type = fown->pid_type;
 548         pid = fown->pid;
 549         if (!pid)
 550                 goto out_unlock_fown;
 551
 552         ret = 1;
 553
 554         read_lock(&tasklist_lock);
 555         do_each_pid_task(pid, type, p) {
 556                 send_sigurg_to_task(p, fown);
 557         } while_each_pid_task(pid, type, p);
 558         read_unlock(&tasklist_lock);
 559  out_unlock_fown:
 560         read_unlock(&fown->lock);
 561         return ret;
 562 }
 563
 564 static DEFINE_RWLOCK(fasync_lock);
 565 static struct kmem_cache *fasync_cache __read_mostly;
 566
 567 /*
 568  * fasync_helper() is used by some character device drivers (mainly mice)
 569  * to set up the fasync queue. It returns negative on error, 0 if it did
 570  * no changes and positive if it added/deleted the entry.
 571  */
 572 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
 573 {
 574         struct fasync_struct *fa, **fp;
 575         struct fasync_struct *new = NULL;
 576         int result = 0;
 577
 578         if (on) {
 579                 new = kmem_cache_alloc(fasync_cache, GFP_KERNEL);
 580                 if (!new)
 581                         return -ENOMEM;
 582         }
 583         write_lock_irq(&fasync_lock);
 584         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 585                 if (fa->fa_file == filp) {
 586                         if(on) {
 587                                 fa->fa_fd = fd;
 588                                 kmem_cache_free(fasync_cache, new);
 589                         } else {
 590                                 *fp = fa->fa_next;
 591                                 kmem_cache_free(fasync_cache, fa);
 592                                 result = 1;
 593                         }
 594                         goto out;
 595                 }
 596         }
 597
 598         if (on) {
 599                 new->magic = FASYNC_MAGIC;
 600                 new->fa_file = filp;
 601                 new->fa_fd = fd;
 602                 new->fa_next = *fapp;
 603                 *fapp = new;
 604                 result = 1;
 605         }
 606 out:
 607         write_unlock_irq(&fasync_lock);
 608         return result;
 609 }
 610
 611 EXPORT_SYMBOL(fasync_helper);
 612
 613 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
 614 {
 615         while (fa) {
 616                 struct fown_struct * fown;
 617                 if (fa->magic != FASYNC_MAGIC) {
 618                         printk(KERN_ERR "kill_fasync: bad magic number in "
 619                                "fasync_struct!\n");
 620                         return;
 621                 }
 622                 fown = &fa->fa_file->f_owner;
 623                 /* Don't send SIGURG to processes which have not set a
 624                    queued signum: SIGURG has its own default signalling
 625                    mechanism. */
 626                 if (!(sig == SIGURG && fown->signum == 0))
 627                         send_sigio(fown, fa->fa_fd, band);
 628                 fa = fa->fa_next;
 629         }
 630 }
 631
 632 EXPORT_SYMBOL(__kill_fasync);
 633
 634 void kill_fasync(struct fasync_struct **fp, int sig, int band)
 635 {
 636         /* First a quick test without locking: usually
 637          * the list is empty.
 638          */
 639         if (*fp) {
 640                 read_lock(&fasync_lock);
 641                 /* reread *fp after obtaining the lock */
 642                 __kill_fasync(*fp, sig, band);
 643                 read_unlock(&fasync_lock);
 644         }
 645 }
 646 EXPORT_SYMBOL(kill_fasync);
 647
 648 static int __init fasync_init(void)
 649 {
 650         fasync_cache = kmem_cache_create("fasync_cache",
 651                 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL, NULL);
 652         return 0;
 653 }
 654
 655 module_init(fasync_init)