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 #include <linux/smp_lock.h>
  23
  24 #include <asm/poll.h>
  25 #include <asm/siginfo.h>
  26 #include <asm/uaccess.h>
  27
  28 void set_close_on_exec(unsigned int fd, int flag)
  29 {
  30         struct files_struct *files = current->files;
  31         struct fdtable *fdt;
  32         spin_lock(&files->file_lock);
  33         fdt = files_fdtable(files);
  34         if (flag)
  35                 FD_SET(fd, fdt->close_on_exec);
  36         else
  37                 FD_CLR(fd, fdt->close_on_exec);
  38         spin_unlock(&files->file_lock);
  39 }
  40
  41 static int get_close_on_exec(unsigned int fd)
  42 {
  43         struct files_struct *files = current->files;
  44         struct fdtable *fdt;
  45         int res;
  46         rcu_read_lock();
  47         fdt = files_fdtable(files);
  48         res = FD_ISSET(fd, fdt->close_on_exec);
  49         rcu_read_unlock();
  50         return res;
  51 }
  52
  53 SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
  54 {
  55         int err = -EBADF;
  56         struct file * file, *tofree;
  57         struct files_struct * files = current->files;
  58         struct fdtable *fdt;
  59
  60         if ((flags & ~O_CLOEXEC) != 0)
  61                 return -EINVAL;
  62
  63         if (unlikely(oldfd == newfd))
  64                 return -EINVAL;
  65
  66         spin_lock(&files->file_lock);
  67         err = expand_files(files, newfd);
  68         file = fcheck(oldfd);
  69         if (unlikely(!file))
  70                 goto Ebadf;
  71         if (unlikely(err < 0)) {
  72                 if (err == -EMFILE)
  73                         goto Ebadf;
  74                 goto out_unlock;
  75         }
  76         /*
  77          * We need to detect attempts to do dup2() over allocated but still
  78          * not finished descriptor.  NB: OpenBSD avoids that at the price of
  79          * extra work in their equivalent of fget() - they insert struct
  80          * file immediately after grabbing descriptor, mark it larval if
  81          * more work (e.g. actual opening) is needed and make sure that
  82          * fget() treats larval files as absent.  Potentially interesting,
  83          * but while extra work in fget() is trivial, locking implications
  84          * and amount of surgery on open()-related paths in VFS are not.
  85          * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
  86          * deadlocks in rather amusing ways, AFAICS.  All of that is out of
  87          * scope of POSIX or SUS, since neither considers shared descriptor
  88          * tables and this condition does not arise without those.
  89          */
  90         err = -EBUSY;
  91         fdt = files_fdtable(files);
  92         tofree = fdt->fd[newfd];
  93         if (!tofree && FD_ISSET(newfd, fdt->open_fds))
  94                 goto out_unlock;
  95         get_file(file);
  96         rcu_assign_pointer(fdt->fd[newfd], file);
  97         FD_SET(newfd, fdt->open_fds);
  98         if (flags & O_CLOEXEC)
  99                 FD_SET(newfd, fdt->close_on_exec);
 100         else
 101                 FD_CLR(newfd, fdt->close_on_exec);
 102         spin_unlock(&files->file_lock);
 103
 104         if (tofree)
 105                 filp_close(tofree, files);
 106
 107         return newfd;
 108
 109 Ebadf:
 110         err = -EBADF;
 111 out_unlock:
 112         spin_unlock(&files->file_lock);
 113         return err;
 114 }
 115
 116 SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
 117 {
 118         if (unlikely(newfd == oldfd)) { /* corner case */
 119                 struct files_struct *files = current->files;
 120                 int retval = oldfd;
 121
 122                 rcu_read_lock();
 123                 if (!fcheck_files(files, oldfd))
 124                         retval = -EBADF;
 125                 rcu_read_unlock();
 126                 return retval;
 127         }
 128         return sys_dup3(oldfd, newfd, 0);
 129 }
 130
 131 SYSCALL_DEFINE1(dup, unsigned int, fildes)
 132 {
 133         int ret = -EBADF;
 134         struct file *file = fget(fildes);
 135
 136         if (file) {
 137                 ret = get_unused_fd();
 138                 if (ret >= 0)
 139                         fd_install(ret, file);
 140                 else
 141                         fput(file);
 142         }
 143         return ret;
 144 }
 145
 146 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
 147
 148 static int setfl(int fd, struct file * filp, unsigned long arg)
 149 {
 150         struct inode * inode = filp->f_path.dentry->d_inode;
 151         int error = 0;
 152
 153         /*
 154          * O_APPEND cannot be cleared if the file is marked as append-only
 155          * and the file is open for write.
 156          */
 157         if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
 158                 return -EPERM;
 159
 160         /* O_NOATIME can only be set by the owner or superuser */
 161         if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
 162                 if (!is_owner_or_cap(inode))
 163                         return -EPERM;
 164
 165         /* required for strict SunOS emulation */
 166         if (O_NONBLOCK != O_NDELAY)
 167                if (arg & O_NDELAY)
 168                    arg |= O_NONBLOCK;
 169
 170         if (arg & O_DIRECT) {
 171                 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
 172                         !filp->f_mapping->a_ops->direct_IO)
 173                                 return -EINVAL;
 174         }
 175
 176         if (filp->f_op && filp->f_op->check_flags)
 177                 error = filp->f_op->check_flags(arg);
 178         if (error)
 179                 return error;
 180
 181         /*
 182          * ->fasync() is responsible for setting the FASYNC bit.
 183          */
 184         if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op &&
 185                         filp->f_op->fasync) {
 186                 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
 187                 if (error < 0)
 188                         goto out;
 189                 if (error > 0)
 190                         error = 0;
 191         }
 192         spin_lock(&filp->f_lock);
 193         filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
 194         spin_unlock(&filp->f_lock);
 195
 196  out:
 197         return error;
 198 }
 199
 200 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
 201                      int force)
 202 {
 203         write_lock_irq(&filp->f_owner.lock);
 204         if (force || !filp->f_owner.pid) {
 205                 put_pid(filp->f_owner.pid);
 206                 filp->f_owner.pid = get_pid(pid);
 207                 filp->f_owner.pid_type = type;
 208
 209                 if (pid) {
 210                         const struct cred *cred = current_cred();
 211                         filp->f_owner.uid = cred->uid;
 212                         filp->f_owner.euid = cred->euid;
 213                 }
 214         }
 215         write_unlock_irq(&filp->f_owner.lock);
 216 }
 217
 218 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
 219                 int force)
 220 {
 221         int err;
 222
 223         err = security_file_set_fowner(filp);
 224         if (err)
 225                 return err;
 226
 227         f_modown(filp, pid, type, force);
 228         return 0;
 229 }
 230 EXPORT_SYMBOL(__f_setown);
 231
 232 int f_setown(struct file *filp, unsigned long arg, int force)
 233 {
 234         enum pid_type type;
 235         struct pid *pid;
 236         int who = arg;
 237         int result;
 238         type = PIDTYPE_PID;
 239         if (who < 0) {
 240                 type = PIDTYPE_PGID;
 241                 who = -who;
 242         }
 243         rcu_read_lock();
 244         pid = find_vpid(who);
 245         result = __f_setown(filp, pid, type, force);
 246         rcu_read_unlock();
 247         return result;
 248 }
 249 EXPORT_SYMBOL(f_setown);
 250
 251 void f_delown(struct file *filp)
 252 {
 253         f_modown(filp, NULL, PIDTYPE_PID, 1);
 254 }
 255
 256 pid_t f_getown(struct file *filp)
 257 {
 258         pid_t pid;
 259         read_lock(&filp->f_owner.lock);
 260         pid = pid_vnr(filp->f_owner.pid);
 261         if (filp->f_owner.pid_type == PIDTYPE_PGID)
 262                 pid = -pid;
 263         read_unlock(&filp->f_owner.lock);
 264         return pid;
 265 }
 266
 267 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
 268                 struct file *filp)
 269 {
 270         long err = -EINVAL;
 271
 272         switch (cmd) {
 273         case F_DUPFD:
 274         case F_DUPFD_CLOEXEC:
 275                 if (arg >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
 276                         break;
 277                 err = alloc_fd(arg, cmd == F_DUPFD_CLOEXEC ? O_CLOEXEC : 0);
 278                 if (err >= 0) {
 279                         get_file(filp);
 280                         fd_install(err, filp);
 281                 }
 282                 break;
 283         case F_GETFD:
 284                 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
 285                 break;
 286         case F_SETFD:
 287                 err = 0;
 288                 set_close_on_exec(fd, arg & FD_CLOEXEC);
 289                 break;
 290         case F_GETFL:
 291                 err = filp->f_flags;
 292                 break;
 293         case F_SETFL:
 294                 err = setfl(fd, filp, arg);
 295                 break;
 296         case F_GETLK:
 297                 err = fcntl_getlk(filp, (struct flock __user *) arg);
 298                 break;
 299         case F_SETLK:
 300         case F_SETLKW:
 301                 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
 302                 break;
 303         case F_GETOWN:
 304                 /*
 305                  * XXX If f_owner is a process group, the
 306                  * negative return value will get converted
 307                  * into an error.  Oops.  If we keep the
 308                  * current syscall conventions, the only way
 309                  * to fix this will be in libc.
 310                  */
 311                 err = f_getown(filp);
 312                 force_successful_syscall_return();
 313                 break;
 314         case F_SETOWN:
 315                 err = f_setown(filp, arg, 1);
 316                 break;
 317         case F_GETSIG:
 318                 err = filp->f_owner.signum;
 319                 break;
 320         case F_SETSIG:
 321                 /* arg == 0 restores default behaviour. */
 322                 if (!valid_signal(arg)) {
 323                         break;
 324                 }
 325                 err = 0;
 326                 filp->f_owner.signum = arg;
 327                 break;
 328         case F_GETLEASE:
 329                 err = fcntl_getlease(filp);
 330                 break;
 331         case F_SETLEASE:
 332                 err = fcntl_setlease(fd, filp, arg);
 333                 break;
 334         case F_NOTIFY:
 335                 err = fcntl_dirnotify(fd, filp, arg);
 336                 break;
 337         default:
 338                 break;
 339         }
 340         return err;
 341 }
 342
 343 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
 344 {
 345         struct file *filp;
 346         long err = -EBADF;
 347
 348         filp = fget(fd);
 349         if (!filp)
 350                 goto out;
 351
 352         err = security_file_fcntl(filp, cmd, arg);
 353         if (err) {
 354                 fput(filp);
 355                 return err;
 356         }
 357
 358         err = do_fcntl(fd, cmd, arg, filp);
 359
 360         fput(filp);
 361 out:
 362         return err;
 363 }
 364
 365 #if BITS_PER_LONG == 32
 366 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
 367                 unsigned long, arg)
 368 {
 369         struct file * filp;
 370         long err;
 371
 372         err = -EBADF;
 373         filp = fget(fd);
 374         if (!filp)
 375                 goto out;
 376
 377         err = security_file_fcntl(filp, cmd, arg);
 378         if (err) {
 379                 fput(filp);
 380                 return err;
 381         }
 382         err = -EBADF;
 383
 384         switch (cmd) {
 385                 case F_GETLK64:
 386                         err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
 387                         break;
 388                 case F_SETLK64:
 389                 case F_SETLKW64:
 390                         err = fcntl_setlk64(fd, filp, cmd,
 391                                         (struct flock64 __user *) arg);
 392                         break;
 393                 default:
 394                         err = do_fcntl(fd, cmd, arg, filp);
 395                         break;
 396         }
 397         fput(filp);
 398 out:
 399         return err;
 400 }
 401 #endif
 402
 403 /* Table to convert sigio signal codes into poll band bitmaps */
 404
 405 static const long band_table[NSIGPOLL] = {
 406         POLLIN | POLLRDNORM,                    /* POLL_IN */
 407         POLLOUT | POLLWRNORM | POLLWRBAND,      /* POLL_OUT */
 408         POLLIN | POLLRDNORM | POLLMSG,          /* POLL_MSG */
 409         POLLERR,                                /* POLL_ERR */
 410         POLLPRI | POLLRDBAND,                   /* POLL_PRI */
 411         POLLHUP | POLLERR                       /* POLL_HUP */
 412 };
 413
 414 static inline int sigio_perm(struct task_struct *p,
 415                              struct fown_struct *fown, int sig)
 416 {
 417         const struct cred *cred;
 418         int ret;
 419
 420         rcu_read_lock();
 421         cred = __task_cred(p);
 422         ret = ((fown->euid == 0 ||
 423                 fown->euid == cred->suid || fown->euid == cred->uid ||
 424                 fown->uid  == cred->suid || fown->uid  == cred->uid) &&
 425                !security_file_send_sigiotask(p, fown, sig));
 426         rcu_read_unlock();
 427         return ret;
 428 }
 429
 430 static void send_sigio_to_task(struct task_struct *p,
 431                                struct fown_struct *fown,
 432                                int fd,
 433                                int reason)
 434 {
 435         /*
 436          * F_SETSIG can change ->signum lockless in parallel, make
 437          * sure we read it once and use the same value throughout.
 438          */
 439         int signum = ACCESS_ONCE(fown->signum);
 440
 441         if (!sigio_perm(p, fown, signum))
 442                 return;
 443
 444         switch (signum) {
 445                 siginfo_t si;
 446                 default:
 447                         /* Queue a rt signal with the appropriate fd as its
 448                            value.  We use SI_SIGIO as the source, not
 449                            SI_KERNEL, since kernel signals always get
 450                            delivered even if we can't queue.  Failure to
 451                            queue in this case _should_ be reported; we fall
 452                            back to SIGIO in that case. --sct */
 453                         si.si_signo = signum;
 454                         si.si_errno = 0;
 455                         si.si_code  = reason;
 456                         /* Make sure we are called with one of the POLL_*
 457                            reasons, otherwise we could leak kernel stack into
 458                            userspace.  */
 459                         BUG_ON((reason & __SI_MASK) != __SI_POLL);
 460                         if (reason - POLL_IN >= NSIGPOLL)
 461                                 si.si_band  = ~0L;
 462                         else
 463                                 si.si_band = band_table[reason - POLL_IN];
 464                         si.si_fd    = fd;
 465                         if (!group_send_sig_info(signum, &si, p))
 466                                 break;
 467                 /* fall-through: fall back on the old plain SIGIO signal */
 468                 case 0:
 469                         group_send_sig_info(SIGIO, SEND_SIG_PRIV, p);
 470         }
 471 }
 472
 473 void send_sigio(struct fown_struct *fown, int fd, int band)
 474 {
 475         struct task_struct *p;
 476         enum pid_type type;
 477         struct pid *pid;
 478
 479         read_lock(&fown->lock);
 480         type = fown->pid_type;
 481         pid = fown->pid;
 482         if (!pid)
 483                 goto out_unlock_fown;
 484
 485         read_lock(&tasklist_lock);
 486         do_each_pid_task(pid, type, p) {
 487                 send_sigio_to_task(p, fown, fd, band);
 488         } while_each_pid_task(pid, type, p);
 489         read_unlock(&tasklist_lock);
 490  out_unlock_fown:
 491         read_unlock(&fown->lock);
 492 }
 493
 494 static void send_sigurg_to_task(struct task_struct *p,
 495                                 struct fown_struct *fown)
 496 {
 497         if (sigio_perm(p, fown, SIGURG))
 498                 group_send_sig_info(SIGURG, SEND_SIG_PRIV, p);
 499 }
 500
 501 int send_sigurg(struct fown_struct *fown)
 502 {
 503         struct task_struct *p;
 504         enum pid_type type;
 505         struct pid *pid;
 506         int ret = 0;
 507
 508         read_lock(&fown->lock);
 509         type = fown->pid_type;
 510         pid = fown->pid;
 511         if (!pid)
 512                 goto out_unlock_fown;
 513
 514         ret = 1;
 515
 516         read_lock(&tasklist_lock);
 517         do_each_pid_task(pid, type, p) {
 518                 send_sigurg_to_task(p, fown);
 519         } while_each_pid_task(pid, type, p);
 520         read_unlock(&tasklist_lock);
 521  out_unlock_fown:
 522         read_unlock(&fown->lock);
 523         return ret;
 524 }
 525
 526 static DEFINE_RWLOCK(fasync_lock);
 527 static struct kmem_cache *fasync_cache __read_mostly;
 528
 529 /*
 530  * fasync_helper() is used by almost all character device drivers
 531  * to set up the fasync queue. It returns negative on error, 0 if it did
 532  * no changes and positive if it added/deleted the entry.
 533  */
 534 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
 535 {
 536         struct fasync_struct *fa, **fp;
 537         struct fasync_struct *new = NULL;
 538         int result = 0;
 539
 540         if (on) {
 541                 new = kmem_cache_alloc(fasync_cache, GFP_KERNEL);
 542                 if (!new)
 543                         return -ENOMEM;
 544         }
 545
 546         /*
 547          * We need to take f_lock first since it's not an IRQ-safe
 548          * lock.
 549          */
 550         spin_lock(&filp->f_lock);
 551         write_lock_irq(&fasync_lock);
 552         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 553                 if (fa->fa_file == filp) {
 554                         if(on) {
 555                                 fa->fa_fd = fd;
 556                                 kmem_cache_free(fasync_cache, new);
 557                         } else {
 558                                 *fp = fa->fa_next;
 559                                 kmem_cache_free(fasync_cache, fa);
 560                                 result = 1;
 561                         }
 562                         goto out;
 563                 }
 564         }
 565
 566         if (on) {
 567                 new->magic = FASYNC_MAGIC;
 568                 new->fa_file = filp;
 569                 new->fa_fd = fd;
 570                 new->fa_next = *fapp;
 571                 *fapp = new;
 572                 result = 1;
 573         }
 574 out:
 575         if (on)
 576                 filp->f_flags |= FASYNC;
 577         else
 578                 filp->f_flags &= ~FASYNC;
 579         write_unlock_irq(&fasync_lock);
 580         spin_unlock(&filp->f_lock);
 581         return result;
 582 }
 583
 584 EXPORT_SYMBOL(fasync_helper);
 585
 586 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
 587 {
 588         while (fa) {
 589                 struct fown_struct * fown;
 590                 if (fa->magic != FASYNC_MAGIC) {
 591                         printk(KERN_ERR "kill_fasync: bad magic number in "
 592                                "fasync_struct!\n");
 593                         return;
 594                 }
 595                 fown = &fa->fa_file->f_owner;
 596                 /* Don't send SIGURG to processes which have not set a
 597                    queued signum: SIGURG has its own default signalling
 598                    mechanism. */
 599                 if (!(sig == SIGURG && fown->signum == 0))
 600                         send_sigio(fown, fa->fa_fd, band);
 601                 fa = fa->fa_next;
 602         }
 603 }
 604
 605 EXPORT_SYMBOL(__kill_fasync);
 606
 607 void kill_fasync(struct fasync_struct **fp, int sig, int band)
 608 {
 609         /* First a quick test without locking: usually
 610          * the list is empty.
 611          */
 612         if (*fp) {
 613                 read_lock(&fasync_lock);
 614                 /* reread *fp after obtaining the lock */
 615                 __kill_fasync(*fp, sig, band);
 616                 read_unlock(&fasync_lock);
 617         }
 618 }
 619 EXPORT_SYMBOL(kill_fasync);
 620
 621 static int __init fasync_init(void)
 622 {
 623         fasync_cache = kmem_cache_create("fasync_cache",
 624                 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
 625         return 0;
 626 }
 627
 628 module_init(fasync_init)