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