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         spin_lock(&files->file_lock);
 139         if (!(file = fcheck(oldfd)))
 140                 goto out_unlock;
 141         err = newfd;
 142         if (unlikely(newfd == oldfd)) {
 143                 if (flags & O_CLOEXEC) {
 144                         fdt = files_fdtable(files);
 145                         FD_SET(newfd, fdt->close_on_exec);
 146                 }
 147                 goto out_unlock;
 148         }
 149         err = -EBADF;
 150         if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
 151                 goto out_unlock;
 152         get_file(file);                 /* We are now finished with oldfd */
 153
 154         err = expand_files(files, newfd);
 155         if (err < 0)
 156                 goto out_fput;
 157
 158         /* To avoid races with open() and dup(), we will mark the fd as
 159          * in-use in the open-file bitmap throughout the entire dup2()
 160          * process.  This is quite safe: do_close() uses the fd array
 161          * entry, not the bitmap, to decide what work needs to be
 162          * done.  --sct */
 163         /* Doesn't work. open() might be there first. --AV */
 164
 165         /* Yes. It's a race. In user space. Nothing sane to do */
 166         err = -EBUSY;
 167         fdt = files_fdtable(files);
 168         tofree = fdt->fd[newfd];
 169         if (!tofree && FD_ISSET(newfd, fdt->open_fds))
 170                 goto out_fput;
 171
 172         rcu_assign_pointer(fdt->fd[newfd], file);
 173         FD_SET(newfd, fdt->open_fds);
 174         if (flags & O_CLOEXEC)
 175                 FD_SET(newfd, fdt->close_on_exec);
 176         else
 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_dup2(unsigned int oldfd, unsigned int newfd)
 196 {
 197         return sys_dup3(oldfd, newfd, 0);
 198 }
 199
 200 asmlinkage long sys_dup(unsigned int fildes)
 201 {
 202         int ret = -EBADF;
 203         struct file * file = fget(fildes);
 204
 205         if (file)
 206                 ret = dupfd(file, 0, 0);
 207         return ret;
 208 }
 209
 210 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
 211
 212 static int setfl(int fd, struct file * filp, unsigned long arg)
 213 {
 214         struct inode * inode = filp->f_path.dentry->d_inode;
 215         int error = 0;
 216
 217         /*
 218          * O_APPEND cannot be cleared if the file is marked as append-only
 219          * and the file is open for write.
 220          */
 221         if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
 222                 return -EPERM;
 223
 224         /* O_NOATIME can only be set by the owner or superuser */
 225         if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
 226                 if (!is_owner_or_cap(inode))
 227                         return -EPERM;
 228
 229         /* required for strict SunOS emulation */
 230         if (O_NONBLOCK != O_NDELAY)
 231                if (arg & O_NDELAY)
 232                    arg |= O_NONBLOCK;
 233
 234         if (arg & O_DIRECT) {
 235                 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
 236                         !filp->f_mapping->a_ops->direct_IO)
 237                                 return -EINVAL;
 238         }
 239
 240         if (filp->f_op && filp->f_op->check_flags)
 241                 error = filp->f_op->check_flags(arg);
 242         if (error)
 243                 return error;
 244
 245         if ((arg ^ filp->f_flags) & FASYNC) {
 246                 if (filp->f_op && filp->f_op->fasync) {
 247                         error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
 248                         if (error < 0)
 249                                 goto out;
 250                 }
 251         }
 252
 253         filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
 254  out:
 255         return error;
 256 }
 257
 258 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
 259                      uid_t uid, uid_t euid, int force)
 260 {
 261         write_lock_irq(&filp->f_owner.lock);
 262         if (force || !filp->f_owner.pid) {
 263                 put_pid(filp->f_owner.pid);
 264                 filp->f_owner.pid = get_pid(pid);
 265                 filp->f_owner.pid_type = type;
 266                 filp->f_owner.uid = uid;
 267                 filp->f_owner.euid = euid;
 268         }
 269         write_unlock_irq(&filp->f_owner.lock);
 270 }
 271
 272 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
 273                 int force)
 274 {
 275         int err;
 276
 277         err = security_file_set_fowner(filp);
 278         if (err)
 279                 return err;
 280
 281         f_modown(filp, pid, type, current->uid, current->euid, force);
 282         return 0;
 283 }
 284 EXPORT_SYMBOL(__f_setown);
 285
 286 int f_setown(struct file *filp, unsigned long arg, int force)
 287 {
 288         enum pid_type type;
 289         struct pid *pid;
 290         int who = arg;
 291         int result;
 292         type = PIDTYPE_PID;
 293         if (who < 0) {
 294                 type = PIDTYPE_PGID;
 295                 who = -who;
 296         }
 297         rcu_read_lock();
 298         pid = find_vpid(who);
 299         result = __f_setown(filp, pid, type, force);
 300         rcu_read_unlock();
 301         return result;
 302 }
 303 EXPORT_SYMBOL(f_setown);
 304
 305 void f_delown(struct file *filp)
 306 {
 307         f_modown(filp, NULL, PIDTYPE_PID, 0, 0, 1);
 308 }
 309
 310 pid_t f_getown(struct file *filp)
 311 {
 312         pid_t pid;
 313         read_lock(&filp->f_owner.lock);
 314         pid = pid_vnr(filp->f_owner.pid);
 315         if (filp->f_owner.pid_type == PIDTYPE_PGID)
 316                 pid = -pid;
 317         read_unlock(&filp->f_owner.lock);
 318         return pid;
 319 }
 320
 321 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
 322                 struct file *filp)
 323 {
 324         long err = -EINVAL;
 325
 326         switch (cmd) {
 327         case F_DUPFD:
 328         case F_DUPFD_CLOEXEC:
 329                 get_file(filp);
 330                 err = dupfd(filp, arg, cmd == F_DUPFD_CLOEXEC);
 331                 break;
 332         case F_GETFD:
 333                 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
 334                 break;
 335         case F_SETFD:
 336                 err = 0;
 337                 set_close_on_exec(fd, arg & FD_CLOEXEC);
 338                 break;
 339         case F_GETFL:
 340                 err = filp->f_flags;
 341                 break;
 342         case F_SETFL:
 343                 err = setfl(fd, filp, arg);
 344                 break;
 345         case F_GETLK:
 346                 err = fcntl_getlk(filp, (struct flock __user *) arg);
 347                 break;
 348         case F_SETLK:
 349         case F_SETLKW:
 350                 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
 351                 break;
 352         case F_GETOWN:
 353                 /*
 354                  * XXX If f_owner is a process group, the
 355                  * negative return value will get converted
 356                  * into an error.  Oops.  If we keep the
 357                  * current syscall conventions, the only way
 358                  * to fix this will be in libc.
 359                  */
 360                 err = f_getown(filp);
 361                 force_successful_syscall_return();
 362                 break;
 363         case F_SETOWN:
 364                 err = f_setown(filp, arg, 1);
 365                 break;
 366         case F_GETSIG:
 367                 err = filp->f_owner.signum;
 368                 break;
 369         case F_SETSIG:
 370                 /* arg == 0 restores default behaviour. */
 371                 if (!valid_signal(arg)) {
 372                         break;
 373                 }
 374                 err = 0;
 375                 filp->f_owner.signum = arg;
 376                 break;
 377         case F_GETLEASE:
 378                 err = fcntl_getlease(filp);
 379                 break;
 380         case F_SETLEASE:
 381                 err = fcntl_setlease(fd, filp, arg);
 382                 break;
 383         case F_NOTIFY:
 384                 err = fcntl_dirnotify(fd, filp, arg);
 385                 break;
 386         default:
 387                 break;
 388         }
 389         return err;
 390 }
 391
 392 asmlinkage long sys_fcntl(unsigned int fd, unsigned int cmd, unsigned long arg)
 393 {
 394         struct file *filp;
 395         long err = -EBADF;
 396
 397         filp = fget(fd);
 398         if (!filp)
 399                 goto out;
 400
 401         err = security_file_fcntl(filp, cmd, arg);
 402         if (err) {
 403                 fput(filp);
 404                 return err;
 405         }
 406
 407         err = do_fcntl(fd, cmd, arg, filp);
 408
 409         fput(filp);
 410 out:
 411         return err;
 412 }
 413
 414 #if BITS_PER_LONG == 32
 415 asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
 416 {
 417         struct file * filp;
 418         long err;
 419
 420         err = -EBADF;
 421         filp = fget(fd);
 422         if (!filp)
 423                 goto out;
 424
 425         err = security_file_fcntl(filp, cmd, arg);
 426         if (err) {
 427                 fput(filp);
 428                 return err;
 429         }
 430         err = -EBADF;
 431
 432         switch (cmd) {
 433                 case F_GETLK64:
 434                         err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
 435                         break;
 436                 case F_SETLK64:
 437                 case F_SETLKW64:
 438                         err = fcntl_setlk64(fd, filp, cmd,
 439                                         (struct flock64 __user *) arg);
 440                         break;
 441                 default:
 442                         err = do_fcntl(fd, cmd, arg, filp);
 443                         break;
 444         }
 445         fput(filp);
 446 out:
 447         return err;
 448 }
 449 #endif
 450
 451 /* Table to convert sigio signal codes into poll band bitmaps */
 452
 453 static const long band_table[NSIGPOLL] = {
 454         POLLIN | POLLRDNORM,                    /* POLL_IN */
 455         POLLOUT | POLLWRNORM | POLLWRBAND,      /* POLL_OUT */
 456         POLLIN | POLLRDNORM | POLLMSG,          /* POLL_MSG */
 457         POLLERR,                                /* POLL_ERR */
 458         POLLPRI | POLLRDBAND,                   /* POLL_PRI */
 459         POLLHUP | POLLERR                       /* POLL_HUP */
 460 };
 461
 462 static inline int sigio_perm(struct task_struct *p,
 463                              struct fown_struct *fown, int sig)
 464 {
 465         return (((fown->euid == 0) ||
 466                  (fown->euid == p->suid) || (fown->euid == p->uid) ||
 467                  (fown->uid == p->suid) || (fown->uid == p->uid)) &&
 468                 !security_file_send_sigiotask(p, fown, sig));
 469 }
 470
 471 static void send_sigio_to_task(struct task_struct *p,
 472                                struct fown_struct *fown,
 473                                int fd,
 474                                int reason)
 475 {
 476         if (!sigio_perm(p, fown, fown->signum))
 477                 return;
 478
 479         switch (fown->signum) {
 480                 siginfo_t si;
 481                 default:
 482                         /* Queue a rt signal with the appropriate fd as its
 483                            value.  We use SI_SIGIO as the source, not
 484                            SI_KERNEL, since kernel signals always get
 485                            delivered even if we can't queue.  Failure to
 486                            queue in this case _should_ be reported; we fall
 487                            back to SIGIO in that case. --sct */
 488                         si.si_signo = fown->signum;
 489                         si.si_errno = 0;
 490                         si.si_code  = reason;
 491                         /* Make sure we are called with one of the POLL_*
 492                            reasons, otherwise we could leak kernel stack into
 493                            userspace.  */
 494                         BUG_ON((reason & __SI_MASK) != __SI_POLL);
 495                         if (reason - POLL_IN >= NSIGPOLL)
 496                                 si.si_band  = ~0L;
 497                         else
 498                                 si.si_band = band_table[reason - POLL_IN];
 499                         si.si_fd    = fd;
 500                         if (!group_send_sig_info(fown->signum, &si, p))
 501                                 break;
 502                 /* fall-through: fall back on the old plain SIGIO signal */
 503                 case 0:
 504                         group_send_sig_info(SIGIO, SEND_SIG_PRIV, p);
 505         }
 506 }
 507
 508 void send_sigio(struct fown_struct *fown, int fd, int band)
 509 {
 510         struct task_struct *p;
 511         enum pid_type type;
 512         struct pid *pid;
 513
 514         read_lock(&fown->lock);
 515         type = fown->pid_type;
 516         pid = fown->pid;
 517         if (!pid)
 518                 goto out_unlock_fown;
 519
 520         read_lock(&tasklist_lock);
 521         do_each_pid_task(pid, type, p) {
 522                 send_sigio_to_task(p, fown, fd, band);
 523         } while_each_pid_task(pid, type, p);
 524         read_unlock(&tasklist_lock);
 525  out_unlock_fown:
 526         read_unlock(&fown->lock);
 527 }
 528
 529 static void send_sigurg_to_task(struct task_struct *p,
 530                                 struct fown_struct *fown)
 531 {
 532         if (sigio_perm(p, fown, SIGURG))
 533                 group_send_sig_info(SIGURG, SEND_SIG_PRIV, p);
 534 }
 535
 536 int send_sigurg(struct fown_struct *fown)
 537 {
 538         struct task_struct *p;
 539         enum pid_type type;
 540         struct pid *pid;
 541         int ret = 0;
 542
 543         read_lock(&fown->lock);
 544         type = fown->pid_type;
 545         pid = fown->pid;
 546         if (!pid)
 547                 goto out_unlock_fown;
 548
 549         ret = 1;
 550
 551         read_lock(&tasklist_lock);
 552         do_each_pid_task(pid, type, p) {
 553                 send_sigurg_to_task(p, fown);
 554         } while_each_pid_task(pid, type, p);
 555         read_unlock(&tasklist_lock);
 556  out_unlock_fown:
 557         read_unlock(&fown->lock);
 558         return ret;
 559 }
 560
 561 static DEFINE_RWLOCK(fasync_lock);
 562 static struct kmem_cache *fasync_cache __read_mostly;
 563
 564 /*
 565  * fasync_helper() is used by some character device drivers (mainly mice)
 566  * to set up the fasync queue. It returns negative on error, 0 if it did
 567  * no changes and positive if it added/deleted the entry.
 568  */
 569 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
 570 {
 571         struct fasync_struct *fa, **fp;
 572         struct fasync_struct *new = NULL;
 573         int result = 0;
 574
 575         if (on) {
 576                 new = kmem_cache_alloc(fasync_cache, GFP_KERNEL);
 577                 if (!new)
 578                         return -ENOMEM;
 579         }
 580         write_lock_irq(&fasync_lock);
 581         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 582                 if (fa->fa_file == filp) {
 583                         if(on) {
 584                                 fa->fa_fd = fd;
 585                                 kmem_cache_free(fasync_cache, new);
 586                         } else {
 587                                 *fp = fa->fa_next;
 588                                 kmem_cache_free(fasync_cache, fa);
 589                                 result = 1;
 590                         }
 591                         goto out;
 592                 }
 593         }
 594
 595         if (on) {
 596                 new->magic = FASYNC_MAGIC;
 597                 new->fa_file = filp;
 598                 new->fa_fd = fd;
 599                 new->fa_next = *fapp;
 600                 *fapp = new;
 601                 result = 1;
 602         }
 603 out:
 604         write_unlock_irq(&fasync_lock);
 605         return result;
 606 }
 607
 608 EXPORT_SYMBOL(fasync_helper);
 609
 610 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
 611 {
 612         while (fa) {
 613                 struct fown_struct * fown;
 614                 if (fa->magic != FASYNC_MAGIC) {
 615                         printk(KERN_ERR "kill_fasync: bad magic number in "
 616                                "fasync_struct!\n");
 617                         return;
 618                 }
 619                 fown = &fa->fa_file->f_owner;
 620                 /* Don't send SIGURG to processes which have not set a
 621                    queued signum: SIGURG has its own default signalling
 622                    mechanism. */
 623                 if (!(sig == SIGURG && fown->signum == 0))
 624                         send_sigio(fown, fa->fa_fd, band);
 625                 fa = fa->fa_next;
 626         }
 627 }
 628
 629 EXPORT_SYMBOL(__kill_fasync);
 630
 631 void kill_fasync(struct fasync_struct **fp, int sig, int band)
 632 {
 633         /* First a quick test without locking: usually
 634          * the list is empty.
 635          */
 636         if (*fp) {
 637                 read_lock(&fasync_lock);
 638                 /* reread *fp after obtaining the lock */
 639                 __kill_fasync(*fp, sig, band);
 640                 read_unlock(&fasync_lock);
 641         }
 642 }
 643 EXPORT_SYMBOL(kill_fasync);
 644
 645 static int __init fasync_init(void)
 646 {
 647         fasync_cache = kmem_cache_create("fasync_cache",
 648                 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
 649         return 0;
 650 }
 651
 652 module_init(fasync_init)