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 /*
  53  * locate_fd finds a free file descriptor in the open_fds fdset,
  54  * expanding the fd arrays if necessary.  Must be called with the
  55  * file_lock held for write.
  56  */
  57
  58 static int locate_fd(unsigned int orig_start, int cloexec)
  59 {
  60         struct files_struct *files = current->files;
  61         unsigned int newfd;
  62         unsigned int start;
  63         int error;
  64         struct fdtable *fdt;
  65
  66         spin_lock(&files->file_lock);
  67
  68         error = -EINVAL;
  69         if (orig_start >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
  70                 goto out;
  71
  72 repeat:
  73         fdt = files_fdtable(files);
  74         /*
  75          * Someone might have closed fd's in the range
  76          * orig_start..fdt->next_fd
  77          */
  78         start = orig_start;
  79         if (start < files->next_fd)
  80                 start = files->next_fd;
  81
  82         newfd = start;
  83         if (start < fdt->max_fds)
  84                 newfd = find_next_zero_bit(fdt->open_fds->fds_bits,
  85                                            fdt->max_fds, start);
  86
  87         error = -EMFILE;
  88         if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
  89                 goto out;
  90
  91         error = expand_files(files, newfd);
  92         if (error < 0)
  93                 goto out;
  94
  95         /*
  96          * If we needed to expand the fs array we
  97          * might have blocked - try again.
  98          */
  99         if (error)
 100                 goto repeat;
 101
 102         if (start <= files->next_fd)
 103                 files->next_fd = newfd + 1;
 104
 105         FD_SET(newfd, fdt->open_fds);
 106         if (cloexec)
 107                 FD_SET(newfd, fdt->close_on_exec);
 108         else
 109                 FD_CLR(newfd, fdt->close_on_exec);
 110         error = newfd;
 111
 112 out:
 113         spin_unlock(&files->file_lock);
 114         return error;
 115 }
 116
 117 static int dupfd(struct file *file, unsigned int start, int cloexec)
 118 {
 119         int fd = locate_fd(start, cloexec);
 120         if (fd >= 0)
 121                 fd_install(fd, file);
 122         else
 123                 fput(file);
 124
 125         return fd;
 126 }
 127
 128 asmlinkage long sys_dup3(unsigned int oldfd, unsigned int newfd, int flags)
 129 {
 130         int err = -EBADF;
 131         struct file * file, *tofree;
 132         struct files_struct * files = current->files;
 133         struct fdtable *fdt;
 134
 135         if ((flags & ~O_CLOEXEC) != 0)
 136                 return -EINVAL;
 137
 138         if (unlikely(oldfd == newfd))
 139                 return -EINVAL;
 140
 141         spin_lock(&files->file_lock);
 142         if (!(file = fcheck(oldfd)))
 143                 goto out_unlock;
 144         if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
 145                 goto out_unlock;
 146         get_file(file);                 /* We are now finished with oldfd */
 147
 148         err = expand_files(files, newfd);
 149         if (err < 0)
 150                 goto out_fput;
 151
 152         /* To avoid races with open() and dup(), we will mark the fd as
 153          * in-use in the open-file bitmap throughout the entire dup2()
 154          * process.  This is quite safe: do_close() uses the fd array
 155          * entry, not the bitmap, to decide what work needs to be
 156          * done.  --sct */
 157         /* Doesn't work. open() might be there first. --AV */
 158
 159         /* Yes. It's a race. In user space. Nothing sane to do */
 160         err = -EBUSY;
 161         fdt = files_fdtable(files);
 162         tofree = fdt->fd[newfd];
 163         if (!tofree && FD_ISSET(newfd, fdt->open_fds))
 164                 goto out_fput;
 165
 166         rcu_assign_pointer(fdt->fd[newfd], file);
 167         FD_SET(newfd, fdt->open_fds);
 168         if (flags & O_CLOEXEC)
 169                 FD_SET(newfd, fdt->close_on_exec);
 170         else
 171                 FD_CLR(newfd, fdt->close_on_exec);
 172         spin_unlock(&files->file_lock);
 173
 174         if (tofree)
 175                 filp_close(tofree, files);
 176         err = newfd;
 177 out:
 178         return err;
 179 out_unlock:
 180         spin_unlock(&files->file_lock);
 181         goto out;
 182
 183 out_fput:
 184         spin_unlock(&files->file_lock);
 185         fput(file);
 186         goto out;
 187 }
 188
 189 asmlinkage long sys_dup2(unsigned int oldfd, unsigned int newfd)
 190 {
 191         if (unlikely(newfd == oldfd)) { /* corner case */
 192                 struct files_struct *files = current->files;
 193                 rcu_read_lock();
 194                 if (!fcheck_files(files, oldfd))
 195                         oldfd = -EBADF;
 196                 rcu_read_unlock();
 197                 return oldfd;
 198         }
 199         return sys_dup3(oldfd, newfd, 0);
 200 }
 201
 202 asmlinkage long sys_dup(unsigned int fildes)
 203 {
 204         int ret = -EBADF;
 205         struct file * file = fget(fildes);
 206
 207         if (file)
 208                 ret = dupfd(file, 0, 0);
 209         return ret;
 210 }
 211
 212 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
 213
 214 static int setfl(int fd, struct file * filp, unsigned long arg)
 215 {
 216         struct inode * inode = filp->f_path.dentry->d_inode;
 217         int error = 0;
 218
 219         /*
 220          * O_APPEND cannot be cleared if the file is marked as append-only
 221          * and the file is open for write.
 222          */
 223         if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
 224                 return -EPERM;
 225
 226         /* O_NOATIME can only be set by the owner or superuser */
 227         if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
 228                 if (!is_owner_or_cap(inode))
 229                         return -EPERM;
 230
 231         /* required for strict SunOS emulation */
 232         if (O_NONBLOCK != O_NDELAY)
 233                if (arg & O_NDELAY)
 234                    arg |= O_NONBLOCK;
 235
 236         if (arg & O_DIRECT) {
 237                 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
 238                         !filp->f_mapping->a_ops->direct_IO)
 239                                 return -EINVAL;
 240         }
 241
 242         if (filp->f_op && filp->f_op->check_flags)
 243                 error = filp->f_op->check_flags(arg);
 244         if (error)
 245                 return error;
 246
 247         if ((arg ^ filp->f_flags) & FASYNC) {
 248                 if (filp->f_op && filp->f_op->fasync) {
 249                         error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
 250                         if (error < 0)
 251                                 goto out;
 252                 }
 253         }
 254
 255         filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
 256  out:
 257         return error;
 258 }
 259
 260 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
 261                      uid_t uid, uid_t euid, int force)
 262 {
 263         write_lock_irq(&filp->f_owner.lock);
 264         if (force || !filp->f_owner.pid) {
 265                 put_pid(filp->f_owner.pid);
 266                 filp->f_owner.pid = get_pid(pid);
 267                 filp->f_owner.pid_type = type;
 268                 filp->f_owner.uid = uid;
 269                 filp->f_owner.euid = euid;
 270         }
 271         write_unlock_irq(&filp->f_owner.lock);
 272 }
 273
 274 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
 275                 int force)
 276 {
 277         int err;
 278
 279         err = security_file_set_fowner(filp);
 280         if (err)
 281                 return err;
 282
 283         f_modown(filp, pid, type, current->uid, current->euid, force);
 284         return 0;
 285 }
 286 EXPORT_SYMBOL(__f_setown);
 287
 288 int f_setown(struct file *filp, unsigned long arg, int force)
 289 {
 290         enum pid_type type;
 291         struct pid *pid;
 292         int who = arg;
 293         int result;
 294         type = PIDTYPE_PID;
 295         if (who < 0) {
 296                 type = PIDTYPE_PGID;
 297                 who = -who;
 298         }
 299         rcu_read_lock();
 300         pid = find_vpid(who);
 301         result = __f_setown(filp, pid, type, force);
 302         rcu_read_unlock();
 303         return result;
 304 }
 305 EXPORT_SYMBOL(f_setown);
 306
 307 void f_delown(struct file *filp)
 308 {
 309         f_modown(filp, NULL, PIDTYPE_PID, 0, 0, 1);
 310 }
 311
 312 pid_t f_getown(struct file *filp)
 313 {
 314         pid_t pid;
 315         read_lock(&filp->f_owner.lock);
 316         pid = pid_vnr(filp->f_owner.pid);
 317         if (filp->f_owner.pid_type == PIDTYPE_PGID)
 318                 pid = -pid;
 319         read_unlock(&filp->f_owner.lock);
 320         return pid;
 321 }
 322
 323 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
 324                 struct file *filp)
 325 {
 326         long err = -EINVAL;
 327
 328         switch (cmd) {
 329         case F_DUPFD:
 330         case F_DUPFD_CLOEXEC:
 331                 get_file(filp);
 332                 err = dupfd(filp, arg, cmd == F_DUPFD_CLOEXEC);
 333                 break;
 334         case F_GETFD:
 335                 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
 336                 break;
 337         case F_SETFD:
 338                 err = 0;
 339                 set_close_on_exec(fd, arg & FD_CLOEXEC);
 340                 break;
 341         case F_GETFL:
 342                 err = filp->f_flags;
 343                 break;
 344         case F_SETFL:
 345                 err = setfl(fd, filp, arg);
 346                 break;
 347         case F_GETLK:
 348                 err = fcntl_getlk(filp, (struct flock __user *) arg);
 349                 break;
 350         case F_SETLK:
 351         case F_SETLKW:
 352                 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
 353                 break;
 354         case F_GETOWN:
 355                 /*
 356                  * XXX If f_owner is a process group, the
 357                  * negative return value will get converted
 358                  * into an error.  Oops.  If we keep the
 359                  * current syscall conventions, the only way
 360                  * to fix this will be in libc.
 361                  */
 362                 err = f_getown(filp);
 363                 force_successful_syscall_return();
 364                 break;
 365         case F_SETOWN:
 366                 err = f_setown(filp, arg, 1);
 367                 break;
 368         case F_GETSIG:
 369                 err = filp->f_owner.signum;
 370                 break;
 371         case F_SETSIG:
 372                 /* arg == 0 restores default behaviour. */
 373                 if (!valid_signal(arg)) {
 374                         break;
 375                 }
 376                 err = 0;
 377                 filp->f_owner.signum = arg;
 378                 break;
 379         case F_GETLEASE:
 380                 err = fcntl_getlease(filp);
 381                 break;
 382         case F_SETLEASE:
 383                 err = fcntl_setlease(fd, filp, arg);
 384                 break;
 385         case F_NOTIFY:
 386                 err = fcntl_dirnotify(fd, filp, arg);
 387                 break;
 388         default:
 389                 break;
 390         }
 391         return err;
 392 }
 393
 394 asmlinkage long sys_fcntl(unsigned int fd, unsigned int cmd, unsigned long arg)
 395 {
 396         struct file *filp;
 397         long err = -EBADF;
 398
 399         filp = fget(fd);
 400         if (!filp)
 401                 goto out;
 402
 403         err = security_file_fcntl(filp, cmd, arg);
 404         if (err) {
 405                 fput(filp);
 406                 return err;
 407         }
 408
 409         err = do_fcntl(fd, cmd, arg, filp);
 410
 411         fput(filp);
 412 out:
 413         return err;
 414 }
 415
 416 #if BITS_PER_LONG == 32
 417 asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
 418 {
 419         struct file * filp;
 420         long err;
 421
 422         err = -EBADF;
 423         filp = fget(fd);
 424         if (!filp)
 425                 goto out;
 426
 427         err = security_file_fcntl(filp, cmd, arg);
 428         if (err) {
 429                 fput(filp);
 430                 return err;
 431         }
 432         err = -EBADF;
 433
 434         switch (cmd) {
 435                 case F_GETLK64:
 436                         err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
 437                         break;
 438                 case F_SETLK64:
 439                 case F_SETLKW64:
 440                         err = fcntl_setlk64(fd, filp, cmd,
 441                                         (struct flock64 __user *) arg);
 442                         break;
 443                 default:
 444                         err = do_fcntl(fd, cmd, arg, filp);
 445                         break;
 446         }
 447         fput(filp);
 448 out:
 449         return err;
 450 }
 451 #endif
 452
 453 /* Table to convert sigio signal codes into poll band bitmaps */
 454
 455 static const long band_table[NSIGPOLL] = {
 456         POLLIN | POLLRDNORM,                    /* POLL_IN */
 457         POLLOUT | POLLWRNORM | POLLWRBAND,      /* POLL_OUT */
 458         POLLIN | POLLRDNORM | POLLMSG,          /* POLL_MSG */
 459         POLLERR,                                /* POLL_ERR */
 460         POLLPRI | POLLRDBAND,                   /* POLL_PRI */
 461         POLLHUP | POLLERR                       /* POLL_HUP */
 462 };
 463
 464 static inline int sigio_perm(struct task_struct *p,
 465                              struct fown_struct *fown, int sig)
 466 {
 467         return (((fown->euid == 0) ||
 468                  (fown->euid == p->suid) || (fown->euid == p->uid) ||
 469                  (fown->uid == p->suid) || (fown->uid == p->uid)) &&
 470                 !security_file_send_sigiotask(p, fown, sig));
 471 }
 472
 473 static void send_sigio_to_task(struct task_struct *p,
 474                                struct fown_struct *fown,
 475                                int fd,
 476                                int reason)
 477 {
 478         if (!sigio_perm(p, fown, fown->signum))
 479                 return;
 480
 481         switch (fown->signum) {
 482                 siginfo_t si;
 483                 default:
 484                         /* Queue a rt signal with the appropriate fd as its
 485                            value.  We use SI_SIGIO as the source, not
 486                            SI_KERNEL, since kernel signals always get
 487                            delivered even if we can't queue.  Failure to
 488                            queue in this case _should_ be reported; we fall
 489                            back to SIGIO in that case. --sct */
 490                         si.si_signo = fown->signum;
 491                         si.si_errno = 0;
 492                         si.si_code  = reason;
 493                         /* Make sure we are called with one of the POLL_*
 494                            reasons, otherwise we could leak kernel stack into
 495                            userspace.  */
 496                         BUG_ON((reason & __SI_MASK) != __SI_POLL);
 497                         if (reason - POLL_IN >= NSIGPOLL)
 498                                 si.si_band  = ~0L;
 499                         else
 500                                 si.si_band = band_table[reason - POLL_IN];
 501                         si.si_fd    = fd;
 502                         if (!group_send_sig_info(fown->signum, &si, p))
 503                                 break;
 504                 /* fall-through: fall back on the old plain SIGIO signal */
 505                 case 0:
 506                         group_send_sig_info(SIGIO, SEND_SIG_PRIV, p);
 507         }
 508 }
 509
 510 void send_sigio(struct fown_struct *fown, int fd, int band)
 511 {
 512         struct task_struct *p;
 513         enum pid_type type;
 514         struct pid *pid;
 515
 516         read_lock(&fown->lock);
 517         type = fown->pid_type;
 518         pid = fown->pid;
 519         if (!pid)
 520                 goto out_unlock_fown;
 521
 522         read_lock(&tasklist_lock);
 523         do_each_pid_task(pid, type, p) {
 524                 send_sigio_to_task(p, fown, fd, band);
 525         } while_each_pid_task(pid, type, p);
 526         read_unlock(&tasklist_lock);
 527  out_unlock_fown:
 528         read_unlock(&fown->lock);
 529 }
 530
 531 static void send_sigurg_to_task(struct task_struct *p,
 532                                 struct fown_struct *fown)
 533 {
 534         if (sigio_perm(p, fown, SIGURG))
 535                 group_send_sig_info(SIGURG, SEND_SIG_PRIV, p);
 536 }
 537
 538 int send_sigurg(struct fown_struct *fown)
 539 {
 540         struct task_struct *p;
 541         enum pid_type type;
 542         struct pid *pid;
 543         int ret = 0;
 544
 545         read_lock(&fown->lock);
 546         type = fown->pid_type;
 547         pid = fown->pid;
 548         if (!pid)
 549                 goto out_unlock_fown;
 550
 551         ret = 1;
 552
 553         read_lock(&tasklist_lock);
 554         do_each_pid_task(pid, type, p) {
 555                 send_sigurg_to_task(p, fown);
 556         } while_each_pid_task(pid, type, p);
 557         read_unlock(&tasklist_lock);
 558  out_unlock_fown:
 559         read_unlock(&fown->lock);
 560         return ret;
 561 }
 562
 563 static DEFINE_RWLOCK(fasync_lock);
 564 static struct kmem_cache *fasync_cache __read_mostly;
 565
 566 /*
 567  * fasync_helper() is used by some character device drivers (mainly mice)
 568  * to set up the fasync queue. It returns negative on error, 0 if it did
 569  * no changes and positive if it added/deleted the entry.
 570  */
 571 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
 572 {
 573         struct fasync_struct *fa, **fp;
 574         struct fasync_struct *new = NULL;
 575         int result = 0;
 576
 577         if (on) {
 578                 new = kmem_cache_alloc(fasync_cache, GFP_KERNEL);
 579                 if (!new)
 580                         return -ENOMEM;
 581         }
 582         write_lock_irq(&fasync_lock);
 583         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 584                 if (fa->fa_file == filp) {
 585                         if(on) {
 586                                 fa->fa_fd = fd;
 587                                 kmem_cache_free(fasync_cache, new);
 588                         } else {
 589                                 *fp = fa->fa_next;
 590                                 kmem_cache_free(fasync_cache, fa);
 591                                 result = 1;
 592                         }
 593                         goto out;
 594                 }
 595         }
 596
 597         if (on) {
 598                 new->magic = FASYNC_MAGIC;
 599                 new->fa_file = filp;
 600                 new->fa_fd = fd;
 601                 new->fa_next = *fapp;
 602                 *fapp = new;
 603                 result = 1;
 604         }
 605 out:
 606         write_unlock_irq(&fasync_lock);
 607         return result;
 608 }
 609
 610 EXPORT_SYMBOL(fasync_helper);
 611
 612 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
 613 {
 614         while (fa) {
 615                 struct fown_struct * fown;
 616                 if (fa->magic != FASYNC_MAGIC) {
 617                         printk(KERN_ERR "kill_fasync: bad magic number in "
 618                                "fasync_struct!\n");
 619                         return;
 620                 }
 621                 fown = &fa->fa_file->f_owner;
 622                 /* Don't send SIGURG to processes which have not set a
 623                    queued signum: SIGURG has its own default signalling
 624                    mechanism. */
 625                 if (!(sig == SIGURG && fown->signum == 0))
 626                         send_sigio(fown, fa->fa_fd, band);
 627                 fa = fa->fa_next;
 628         }
 629 }
 630
 631 EXPORT_SYMBOL(__kill_fasync);
 632
 633 void kill_fasync(struct fasync_struct **fp, int sig, int band)
 634 {
 635         /* First a quick test without locking: usually
 636          * the list is empty.
 637          */
 638         if (*fp) {
 639                 read_lock(&fasync_lock);
 640                 /* reread *fp after obtaining the lock */
 641                 __kill_fasync(*fp, sig, band);
 642                 read_unlock(&fasync_lock);
 643         }
 644 }
 645 EXPORT_SYMBOL(kill_fasync);
 646
 647 static int __init fasync_init(void)
 648 {
 649         fasync_cache = kmem_cache_create("fasync_cache",
 650                 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
 651         return 0;
 652 }
 653
 654 module_init(fasync_init)