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 (!is_owner_or_cap(inode))
 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         result = __f_setown(filp, pid, type, force);
 297         rcu_read_unlock();
 298         return result;
 299 }
 300 EXPORT_SYMBOL(f_setown);
 301
 302 void f_delown(struct file *filp)
 303 {
 304         f_modown(filp, NULL, PIDTYPE_PID, 0, 0, 1);
 305 }
 306
 307 pid_t f_getown(struct file *filp)
 308 {
 309         pid_t pid;
 310         read_lock(&filp->f_owner.lock);
 311         pid = pid_nr(filp->f_owner.pid);
 312         if (filp->f_owner.pid_type == PIDTYPE_PGID)
 313                 pid = -pid;
 314         read_unlock(&filp->f_owner.lock);
 315         return pid;
 316 }
 317
 318 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
 319                 struct file *filp)
 320 {
 321         long err = -EINVAL;
 322
 323         switch (cmd) {
 324         case F_DUPFD:
 325         case F_DUPFD_CLOEXEC:
 326                 get_file(filp);
 327                 err = dupfd(filp, arg, cmd == F_DUPFD_CLOEXEC);
 328                 break;
 329         case F_GETFD:
 330                 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
 331                 break;
 332         case F_SETFD:
 333                 err = 0;
 334                 set_close_on_exec(fd, arg & FD_CLOEXEC);
 335                 break;
 336         case F_GETFL:
 337                 err = filp->f_flags;
 338                 break;
 339         case F_SETFL:
 340                 err = setfl(fd, filp, arg);
 341                 break;
 342         case F_GETLK:
 343                 err = fcntl_getlk(filp, (struct flock __user *) arg);
 344                 break;
 345         case F_SETLK:
 346         case F_SETLKW:
 347                 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
 348                 break;
 349         case F_GETOWN:
 350                 /*
 351                  * XXX If f_owner is a process group, the
 352                  * negative return value will get converted
 353                  * into an error.  Oops.  If we keep the
 354                  * current syscall conventions, the only way
 355                  * to fix this will be in libc.
 356                  */
 357                 err = f_getown(filp);
 358                 force_successful_syscall_return();
 359                 break;
 360         case F_SETOWN:
 361                 err = f_setown(filp, arg, 1);
 362                 break;
 363         case F_GETSIG:
 364                 err = filp->f_owner.signum;
 365                 break;
 366         case F_SETSIG:
 367                 /* arg == 0 restores default behaviour. */
 368                 if (!valid_signal(arg)) {
 369                         break;
 370                 }
 371                 err = 0;
 372                 filp->f_owner.signum = arg;
 373                 break;
 374         case F_GETLEASE:
 375                 err = fcntl_getlease(filp);
 376                 break;
 377         case F_SETLEASE:
 378                 err = fcntl_setlease(fd, filp, arg);
 379                 break;
 380         case F_NOTIFY:
 381                 err = fcntl_dirnotify(fd, filp, arg);
 382                 break;
 383         default:
 384                 break;
 385         }
 386         return err;
 387 }
 388
 389 asmlinkage long sys_fcntl(unsigned int fd, unsigned int cmd, unsigned long arg)
 390 {
 391         struct file *filp;
 392         long err = -EBADF;
 393
 394         filp = fget(fd);
 395         if (!filp)
 396                 goto out;
 397
 398         err = security_file_fcntl(filp, cmd, arg);
 399         if (err) {
 400                 fput(filp);
 401                 return err;
 402         }
 403
 404         err = do_fcntl(fd, cmd, arg, filp);
 405
 406         fput(filp);
 407 out:
 408         return err;
 409 }
 410
 411 #if BITS_PER_LONG == 32
 412 asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
 413 {
 414         struct file * filp;
 415         long err;
 416
 417         err = -EBADF;
 418         filp = fget(fd);
 419         if (!filp)
 420                 goto out;
 421
 422         err = security_file_fcntl(filp, cmd, arg);
 423         if (err) {
 424                 fput(filp);
 425                 return err;
 426         }
 427         err = -EBADF;
 428
 429         switch (cmd) {
 430                 case F_GETLK64:
 431                         err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
 432                         break;
 433                 case F_SETLK64:
 434                 case F_SETLKW64:
 435                         err = fcntl_setlk64(fd, filp, cmd,
 436                                         (struct flock64 __user *) arg);
 437                         break;
 438                 default:
 439                         err = do_fcntl(fd, cmd, arg, filp);
 440                         break;
 441         }
 442         fput(filp);
 443 out:
 444         return err;
 445 }
 446 #endif
 447
 448 /* Table to convert sigio signal codes into poll band bitmaps */
 449
 450 static const long band_table[NSIGPOLL] = {
 451         POLLIN | POLLRDNORM,                    /* POLL_IN */
 452         POLLOUT | POLLWRNORM | POLLWRBAND,      /* POLL_OUT */
 453         POLLIN | POLLRDNORM | POLLMSG,          /* POLL_MSG */
 454         POLLERR,                                /* POLL_ERR */
 455         POLLPRI | POLLRDBAND,                   /* POLL_PRI */
 456         POLLHUP | POLLERR                       /* POLL_HUP */
 457 };
 458
 459 static inline int sigio_perm(struct task_struct *p,
 460                              struct fown_struct *fown, int sig)
 461 {
 462         return (((fown->euid == 0) ||
 463                  (fown->euid == p->suid) || (fown->euid == p->uid) ||
 464                  (fown->uid == p->suid) || (fown->uid == p->uid)) &&
 465                 !security_file_send_sigiotask(p, fown, sig));
 466 }
 467
 468 static void send_sigio_to_task(struct task_struct *p,
 469                                struct fown_struct *fown,
 470                                int fd,
 471                                int reason)
 472 {
 473         if (!sigio_perm(p, fown, fown->signum))
 474                 return;
 475
 476         switch (fown->signum) {
 477                 siginfo_t si;
 478                 default:
 479                         /* Queue a rt signal with the appropriate fd as its
 480                            value.  We use SI_SIGIO as the source, not
 481                            SI_KERNEL, since kernel signals always get
 482                            delivered even if we can't queue.  Failure to
 483                            queue in this case _should_ be reported; we fall
 484                            back to SIGIO in that case. --sct */
 485                         si.si_signo = fown->signum;
 486                         si.si_errno = 0;
 487                         si.si_code  = reason;
 488                         /* Make sure we are called with one of the POLL_*
 489                            reasons, otherwise we could leak kernel stack into
 490                            userspace.  */
 491                         BUG_ON((reason & __SI_MASK) != __SI_POLL);
 492                         if (reason - POLL_IN >= NSIGPOLL)
 493                                 si.si_band  = ~0L;
 494                         else
 495                                 si.si_band = band_table[reason - POLL_IN];
 496                         si.si_fd    = fd;
 497                         if (!group_send_sig_info(fown->signum, &si, p))
 498                                 break;
 499                 /* fall-through: fall back on the old plain SIGIO signal */
 500                 case 0:
 501                         group_send_sig_info(SIGIO, SEND_SIG_PRIV, p);
 502         }
 503 }
 504
 505 void send_sigio(struct fown_struct *fown, int fd, int band)
 506 {
 507         struct task_struct *p;
 508         enum pid_type type;
 509         struct pid *pid;
 510
 511         read_lock(&fown->lock);
 512         type = fown->pid_type;
 513         pid = fown->pid;
 514         if (!pid)
 515                 goto out_unlock_fown;
 516
 517         read_lock(&tasklist_lock);
 518         do_each_pid_task(pid, type, p) {
 519                 send_sigio_to_task(p, fown, fd, band);
 520         } while_each_pid_task(pid, type, p);
 521         read_unlock(&tasklist_lock);
 522  out_unlock_fown:
 523         read_unlock(&fown->lock);
 524 }
 525
 526 static void send_sigurg_to_task(struct task_struct *p,
 527                                 struct fown_struct *fown)
 528 {
 529         if (sigio_perm(p, fown, SIGURG))
 530                 group_send_sig_info(SIGURG, SEND_SIG_PRIV, p);
 531 }
 532
 533 int send_sigurg(struct fown_struct *fown)
 534 {
 535         struct task_struct *p;
 536         enum pid_type type;
 537         struct pid *pid;
 538         int ret = 0;
 539
 540         read_lock(&fown->lock);
 541         type = fown->pid_type;
 542         pid = fown->pid;
 543         if (!pid)
 544                 goto out_unlock_fown;
 545
 546         ret = 1;
 547
 548         read_lock(&tasklist_lock);
 549         do_each_pid_task(pid, type, p) {
 550                 send_sigurg_to_task(p, fown);
 551         } while_each_pid_task(pid, type, p);
 552         read_unlock(&tasklist_lock);
 553  out_unlock_fown:
 554         read_unlock(&fown->lock);
 555         return ret;
 556 }
 557
 558 static DEFINE_RWLOCK(fasync_lock);
 559 static struct kmem_cache *fasync_cache __read_mostly;
 560
 561 /*
 562  * fasync_helper() is used by some character device drivers (mainly mice)
 563  * to set up the fasync queue. It returns negative on error, 0 if it did
 564  * no changes and positive if it added/deleted the entry.
 565  */
 566 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
 567 {
 568         struct fasync_struct *fa, **fp;
 569         struct fasync_struct *new = NULL;
 570         int result = 0;
 571
 572         if (on) {
 573                 new = kmem_cache_alloc(fasync_cache, GFP_KERNEL);
 574                 if (!new)
 575                         return -ENOMEM;
 576         }
 577         write_lock_irq(&fasync_lock);
 578         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 579                 if (fa->fa_file == filp) {
 580                         if(on) {
 581                                 fa->fa_fd = fd;
 582                                 kmem_cache_free(fasync_cache, new);
 583                         } else {
 584                                 *fp = fa->fa_next;
 585                                 kmem_cache_free(fasync_cache, fa);
 586                                 result = 1;
 587                         }
 588                         goto out;
 589                 }
 590         }
 591
 592         if (on) {
 593                 new->magic = FASYNC_MAGIC;
 594                 new->fa_file = filp;
 595                 new->fa_fd = fd;
 596                 new->fa_next = *fapp;
 597                 *fapp = new;
 598                 result = 1;
 599         }
 600 out:
 601         write_unlock_irq(&fasync_lock);
 602         return result;
 603 }
 604
 605 EXPORT_SYMBOL(fasync_helper);
 606
 607 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
 608 {
 609         while (fa) {
 610                 struct fown_struct * fown;
 611                 if (fa->magic != FASYNC_MAGIC) {
 612                         printk(KERN_ERR "kill_fasync: bad magic number in "
 613                                "fasync_struct!\n");
 614                         return;
 615                 }
 616                 fown = &fa->fa_file->f_owner;
 617                 /* Don't send SIGURG to processes which have not set a
 618                    queued signum: SIGURG has its own default signalling
 619                    mechanism. */
 620                 if (!(sig == SIGURG && fown->signum == 0))
 621                         send_sigio(fown, fa->fa_fd, band);
 622                 fa = fa->fa_next;
 623         }
 624 }
 625
 626 EXPORT_SYMBOL(__kill_fasync);
 627
 628 void kill_fasync(struct fasync_struct **fp, int sig, int band)
 629 {
 630         /* First a quick test without locking: usually
 631          * the list is empty.
 632          */
 633         if (*fp) {
 634                 read_lock(&fasync_lock);
 635                 /* reread *fp after obtaining the lock */
 636                 __kill_fasync(*fp, sig, band);
 637                 read_unlock(&fasync_lock);
 638         }
 639 }
 640 EXPORT_SYMBOL(kill_fasync);
 641
 642 static int __init fasync_init(void)
 643 {
 644         fasync_cache = kmem_cache_create("fasync_cache",
 645                 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
 646         return 0;
 647 }
 648
 649 module_init(fasync_init)