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