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 SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
  53 {
  54         int err = -EBADF;
  55         struct file * file, *tofree;
  56         struct files_struct * files = current->files;
  57         struct fdtable *fdt;
  58
  59         if ((flags & ~O_CLOEXEC) != 0)
  60                 return -EINVAL;
  61
  62         if (unlikely(oldfd == newfd))
  63                 return -EINVAL;
  64
  65         spin_lock(&files->file_lock);
  66         err = expand_files(files, newfd);
  67         file = fcheck(oldfd);
  68         if (unlikely(!file))
  69                 goto Ebadf;
  70         if (unlikely(err < 0)) {
  71                 if (err == -EMFILE)
  72                         goto Ebadf;
  73                 goto out_unlock;
  74         }
  75         /*
  76          * We need to detect attempts to do dup2() over allocated but still
  77          * not finished descriptor.  NB: OpenBSD avoids that at the price of
  78          * extra work in their equivalent of fget() - they insert struct
  79          * file immediately after grabbing descriptor, mark it larval if
  80          * more work (e.g. actual opening) is needed and make sure that
  81          * fget() treats larval files as absent.  Potentially interesting,
  82          * but while extra work in fget() is trivial, locking implications
  83          * and amount of surgery on open()-related paths in VFS are not.
  84          * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
  85          * deadlocks in rather amusing ways, AFAICS.  All of that is out of
  86          * scope of POSIX or SUS, since neither considers shared descriptor
  87          * tables and this condition does not arise without those.
  88          */
  89         err = -EBUSY;
  90         fdt = files_fdtable(files);
  91         tofree = fdt->fd[newfd];
  92         if (!tofree && FD_ISSET(newfd, fdt->open_fds))
  93                 goto out_unlock;
  94         get_file(file);
  95         rcu_assign_pointer(fdt->fd[newfd], file);
  96         FD_SET(newfd, fdt->open_fds);
  97         if (flags & O_CLOEXEC)
  98                 FD_SET(newfd, fdt->close_on_exec);
  99         else
 100                 FD_CLR(newfd, fdt->close_on_exec);
 101         spin_unlock(&files->file_lock);
 102
 103         if (tofree)
 104                 filp_close(tofree, files);
 105
 106         return newfd;
 107
 108 Ebadf:
 109         err = -EBADF;
 110 out_unlock:
 111         spin_unlock(&files->file_lock);
 112         return err;
 113 }
 114
 115 SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
 116 {
 117         if (unlikely(newfd == oldfd)) { /* corner case */
 118                 struct files_struct *files = current->files;
 119                 int retval = oldfd;
 120
 121                 rcu_read_lock();
 122                 if (!fcheck_files(files, oldfd))
 123                         retval = -EBADF;
 124                 rcu_read_unlock();
 125                 return retval;
 126         }
 127         return sys_dup3(oldfd, newfd, 0);
 128 }
 129
 130 SYSCALL_DEFINE1(dup, unsigned int, fildes)
 131 {
 132         int ret = -EBADF;
 133         struct file *file = fget(fildes);
 134
 135         if (file) {
 136                 ret = get_unused_fd();
 137                 if (ret >= 0)
 138                         fd_install(ret, file);
 139                 else
 140                         fput(file);
 141         }
 142         return ret;
 143 }
 144
 145 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
 146
 147 static int setfl(int fd, struct file * filp, unsigned long arg)
 148 {
 149         struct inode * inode = filp->f_path.dentry->d_inode;
 150         int error = 0;
 151
 152         /*
 153          * O_APPEND cannot be cleared if the file is marked as append-only
 154          * and the file is open for write.
 155          */
 156         if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
 157                 return -EPERM;
 158
 159         /* O_NOATIME can only be set by the owner or superuser */
 160         if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
 161                 if (!is_owner_or_cap(inode))
 162                         return -EPERM;
 163
 164         /* required for strict SunOS emulation */
 165         if (O_NONBLOCK != O_NDELAY)
 166                if (arg & O_NDELAY)
 167                    arg |= O_NONBLOCK;
 168
 169         if (arg & O_DIRECT) {
 170                 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
 171                         !filp->f_mapping->a_ops->direct_IO)
 172                                 return -EINVAL;
 173         }
 174
 175         if (filp->f_op && filp->f_op->check_flags)
 176                 error = filp->f_op->check_flags(arg);
 177         if (error)
 178                 return error;
 179
 180         /*
 181          * ->fasync() is responsible for setting the FASYNC bit.
 182          */
 183         if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op &&
 184                         filp->f_op->fasync) {
 185                 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
 186                 if (error < 0)
 187                         goto out;
 188                 if (error > 0)
 189                         error = 0;
 190         }
 191         spin_lock(&filp->f_lock);
 192         filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
 193         spin_unlock(&filp->f_lock);
 194
 195  out:
 196         return error;
 197 }
 198
 199 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
 200                      int force)
 201 {
 202         write_lock_irq(&filp->f_owner.lock);
 203         if (force || !filp->f_owner.pid) {
 204                 put_pid(filp->f_owner.pid);
 205                 filp->f_owner.pid = get_pid(pid);
 206                 filp->f_owner.pid_type = type;
 207
 208                 if (pid) {
 209                         const struct cred *cred = current_cred();
 210                         filp->f_owner.uid = cred->uid;
 211                         filp->f_owner.euid = cred->euid;
 212                 }
 213         }
 214         write_unlock_irq(&filp->f_owner.lock);
 215 }
 216
 217 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
 218                 int force)
 219 {
 220         int err;
 221
 222         err = security_file_set_fowner(filp);
 223         if (err)
 224                 return err;
 225
 226         f_modown(filp, pid, type, force);
 227         return 0;
 228 }
 229 EXPORT_SYMBOL(__f_setown);
 230
 231 int f_setown(struct file *filp, unsigned long arg, int force)
 232 {
 233         enum pid_type type;
 234         struct pid *pid;
 235         int who = arg;
 236         int result;
 237         type = PIDTYPE_PID;
 238         if (who < 0) {
 239                 type = PIDTYPE_PGID;
 240                 who = -who;
 241         }
 242         rcu_read_lock();
 243         pid = find_vpid(who);
 244         result = __f_setown(filp, pid, type, force);
 245         rcu_read_unlock();
 246         return result;
 247 }
 248 EXPORT_SYMBOL(f_setown);
 249
 250 void f_delown(struct file *filp)
 251 {
 252         f_modown(filp, NULL, PIDTYPE_PID, 1);
 253 }
 254
 255 pid_t f_getown(struct file *filp)
 256 {
 257         pid_t pid;
 258         read_lock(&filp->f_owner.lock);
 259         pid = pid_vnr(filp->f_owner.pid);
 260         if (filp->f_owner.pid_type == PIDTYPE_PGID)
 261                 pid = -pid;
 262         read_unlock(&filp->f_owner.lock);
 263         return pid;
 264 }
 265
 266 static int f_setown_ex(struct file *filp, unsigned long arg)
 267 {
 268         struct f_owner_ex * __user owner_p = (void * __user)arg;
 269         struct f_owner_ex owner;
 270         struct pid *pid;
 271         int type;
 272         int ret;
 273
 274         ret = copy_from_user(&owner, owner_p, sizeof(owner));
 275         if (ret)
 276                 return ret;
 277
 278         switch (owner.type) {
 279         case F_OWNER_TID:
 280                 type = PIDTYPE_MAX;
 281                 break;
 282
 283         case F_OWNER_PID:
 284                 type = PIDTYPE_PID;
 285                 break;
 286
 287         case F_OWNER_PGRP:
 288                 type = PIDTYPE_PGID;
 289                 break;
 290
 291         default:
 292                 return -EINVAL;
 293         }
 294
 295         rcu_read_lock();
 296         pid = find_vpid(owner.pid);
 297         if (owner.pid && !pid)
 298                 ret = -ESRCH;
 299         else
 300                 ret = __f_setown(filp, pid, type, 1);
 301         rcu_read_unlock();
 302
 303         return ret;
 304 }
 305
 306 static int f_getown_ex(struct file *filp, unsigned long arg)
 307 {
 308         struct f_owner_ex * __user owner_p = (void * __user)arg;
 309         struct f_owner_ex owner;
 310         int ret = 0;
 311
 312         read_lock(&filp->f_owner.lock);
 313         owner.pid = pid_vnr(filp->f_owner.pid);
 314         switch (filp->f_owner.pid_type) {
 315         case PIDTYPE_MAX:
 316                 owner.type = F_OWNER_TID;
 317                 break;
 318
 319         case PIDTYPE_PID:
 320                 owner.type = F_OWNER_PID;
 321                 break;
 322
 323         case PIDTYPE_PGID:
 324                 owner.type = F_OWNER_PGRP;
 325                 break;
 326
 327         default:
 328                 WARN_ON(1);
 329                 ret = -EINVAL;
 330                 break;
 331         }
 332         read_unlock(&filp->f_owner.lock);
 333
 334         if (!ret)
 335                 ret = copy_to_user(owner_p, &owner, sizeof(owner));
 336         return ret;
 337 }
 338
 339 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
 340                 struct file *filp)
 341 {
 342         long err = -EINVAL;
 343
 344         switch (cmd) {
 345         case F_DUPFD:
 346         case F_DUPFD_CLOEXEC:
 347                 if (arg >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
 348                         break;
 349                 err = alloc_fd(arg, cmd == F_DUPFD_CLOEXEC ? O_CLOEXEC : 0);
 350                 if (err >= 0) {
 351                         get_file(filp);
 352                         fd_install(err, filp);
 353                 }
 354                 break;
 355         case F_GETFD:
 356                 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
 357                 break;
 358         case F_SETFD:
 359                 err = 0;
 360                 set_close_on_exec(fd, arg & FD_CLOEXEC);
 361                 break;
 362         case F_GETFL:
 363                 err = filp->f_flags;
 364                 break;
 365         case F_SETFL:
 366                 err = setfl(fd, filp, arg);
 367                 break;
 368         case F_GETLK:
 369                 err = fcntl_getlk(filp, (struct flock __user *) arg);
 370                 break;
 371         case F_SETLK:
 372         case F_SETLKW:
 373                 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
 374                 break;
 375         case F_GETOWN:
 376                 /*
 377                  * XXX If f_owner is a process group, the
 378                  * negative return value will get converted
 379                  * into an error.  Oops.  If we keep the
 380                  * current syscall conventions, the only way
 381                  * to fix this will be in libc.
 382                  */
 383                 err = f_getown(filp);
 384                 force_successful_syscall_return();
 385                 break;
 386         case F_SETOWN:
 387                 err = f_setown(filp, arg, 1);
 388                 break;
 389         case F_GETOWN_EX:
 390                 err = f_getown_ex(filp, arg);
 391                 break;
 392         case F_SETOWN_EX:
 393                 err = f_setown_ex(filp, arg);
 394                 break;
 395         case F_GETSIG:
 396                 err = filp->f_owner.signum;
 397                 break;
 398         case F_SETSIG:
 399                 /* arg == 0 restores default behaviour. */
 400                 if (!valid_signal(arg)) {
 401                         break;
 402                 }
 403                 err = 0;
 404                 filp->f_owner.signum = arg;
 405                 break;
 406         case F_GETLEASE:
 407                 err = fcntl_getlease(filp);
 408                 break;
 409         case F_SETLEASE:
 410                 err = fcntl_setlease(fd, filp, arg);
 411                 break;
 412         case F_NOTIFY:
 413                 err = fcntl_dirnotify(fd, filp, arg);
 414                 break;
 415         default:
 416                 break;
 417         }
 418         return err;
 419 }
 420
 421 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
 422 {
 423         struct file *filp;
 424         long err = -EBADF;
 425
 426         filp = fget(fd);
 427         if (!filp)
 428                 goto out;
 429
 430         err = security_file_fcntl(filp, cmd, arg);
 431         if (err) {
 432                 fput(filp);
 433                 return err;
 434         }
 435
 436         err = do_fcntl(fd, cmd, arg, filp);
 437
 438         fput(filp);
 439 out:
 440         return err;
 441 }
 442
 443 #if BITS_PER_LONG == 32
 444 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
 445                 unsigned long, arg)
 446 {
 447         struct file * filp;
 448         long err;
 449
 450         err = -EBADF;
 451         filp = fget(fd);
 452         if (!filp)
 453                 goto out;
 454
 455         err = security_file_fcntl(filp, cmd, arg);
 456         if (err) {
 457                 fput(filp);
 458                 return err;
 459         }
 460         err = -EBADF;
 461
 462         switch (cmd) {
 463                 case F_GETLK64:
 464                         err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
 465                         break;
 466                 case F_SETLK64:
 467                 case F_SETLKW64:
 468                         err = fcntl_setlk64(fd, filp, cmd,
 469                                         (struct flock64 __user *) arg);
 470                         break;
 471                 default:
 472                         err = do_fcntl(fd, cmd, arg, filp);
 473                         break;
 474         }
 475         fput(filp);
 476 out:
 477         return err;
 478 }
 479 #endif
 480
 481 /* Table to convert sigio signal codes into poll band bitmaps */
 482
 483 static const long band_table[NSIGPOLL] = {
 484         POLLIN | POLLRDNORM,                    /* POLL_IN */
 485         POLLOUT | POLLWRNORM | POLLWRBAND,      /* POLL_OUT */
 486         POLLIN | POLLRDNORM | POLLMSG,          /* POLL_MSG */
 487         POLLERR,                                /* POLL_ERR */
 488         POLLPRI | POLLRDBAND,                   /* POLL_PRI */
 489         POLLHUP | POLLERR                       /* POLL_HUP */
 490 };
 491
 492 static inline int sigio_perm(struct task_struct *p,
 493                              struct fown_struct *fown, int sig)
 494 {
 495         const struct cred *cred;
 496         int ret;
 497
 498         rcu_read_lock();
 499         cred = __task_cred(p);
 500         ret = ((fown->euid == 0 ||
 501                 fown->euid == cred->suid || fown->euid == cred->uid ||
 502                 fown->uid  == cred->suid || fown->uid  == cred->uid) &&
 503                !security_file_send_sigiotask(p, fown, sig));
 504         rcu_read_unlock();
 505         return ret;
 506 }
 507
 508 static void send_sigio_to_task(struct task_struct *p,
 509                                struct fown_struct *fown,
 510                                int fd, int reason, int group)
 511 {
 512         /*
 513          * F_SETSIG can change ->signum lockless in parallel, make
 514          * sure we read it once and use the same value throughout.
 515          */
 516         int signum = ACCESS_ONCE(fown->signum);
 517
 518         if (!sigio_perm(p, fown, signum))
 519                 return;
 520
 521         switch (signum) {
 522                 siginfo_t si;
 523                 default:
 524                         /* Queue a rt signal with the appropriate fd as its
 525                            value.  We use SI_SIGIO as the source, not
 526                            SI_KERNEL, since kernel signals always get
 527                            delivered even if we can't queue.  Failure to
 528                            queue in this case _should_ be reported; we fall
 529                            back to SIGIO in that case. --sct */
 530                         si.si_signo = signum;
 531                         si.si_errno = 0;
 532                         si.si_code  = reason;
 533                         /* Make sure we are called with one of the POLL_*
 534                            reasons, otherwise we could leak kernel stack into
 535                            userspace.  */
 536                         BUG_ON((reason & __SI_MASK) != __SI_POLL);
 537                         if (reason - POLL_IN >= NSIGPOLL)
 538                                 si.si_band  = ~0L;
 539                         else
 540                                 si.si_band = band_table[reason - POLL_IN];
 541                         si.si_fd    = fd;
 542                         if (!do_send_sig_info(signum, &si, p, group))
 543                                 break;
 544                 /* fall-through: fall back on the old plain SIGIO signal */
 545                 case 0:
 546                         do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
 547         }
 548 }
 549
 550 void send_sigio(struct fown_struct *fown, int fd, int band)
 551 {
 552         struct task_struct *p;
 553         enum pid_type type;
 554         struct pid *pid;
 555         int group = 1;
 556
 557         read_lock(&fown->lock);
 558
 559         type = fown->pid_type;
 560         if (type == PIDTYPE_MAX) {
 561                 group = 0;
 562                 type = PIDTYPE_PID;
 563         }
 564
 565         pid = fown->pid;
 566         if (!pid)
 567                 goto out_unlock_fown;
 568
 569         read_lock(&tasklist_lock);
 570         do_each_pid_task(pid, type, p) {
 571                 send_sigio_to_task(p, fown, fd, band, group);
 572         } while_each_pid_task(pid, type, p);
 573         read_unlock(&tasklist_lock);
 574  out_unlock_fown:
 575         read_unlock(&fown->lock);
 576 }
 577
 578 static void send_sigurg_to_task(struct task_struct *p,
 579                                 struct fown_struct *fown, int group)
 580 {
 581         if (sigio_perm(p, fown, SIGURG))
 582                 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
 583 }
 584
 585 int send_sigurg(struct fown_struct *fown)
 586 {
 587         struct task_struct *p;
 588         enum pid_type type;
 589         struct pid *pid;
 590         int group = 1;
 591         int ret = 0;
 592
 593         read_lock(&fown->lock);
 594
 595         type = fown->pid_type;
 596         if (type == PIDTYPE_MAX) {
 597                 group = 0;
 598                 type = PIDTYPE_PID;
 599         }
 600
 601         pid = fown->pid;
 602         if (!pid)
 603                 goto out_unlock_fown;
 604
 605         ret = 1;
 606
 607         read_lock(&tasklist_lock);
 608         do_each_pid_task(pid, type, p) {
 609                 send_sigurg_to_task(p, fown, group);
 610         } while_each_pid_task(pid, type, p);
 611         read_unlock(&tasklist_lock);
 612  out_unlock_fown:
 613         read_unlock(&fown->lock);
 614         return ret;
 615 }
 616
 617 static DEFINE_RWLOCK(fasync_lock);
 618 static struct kmem_cache *fasync_cache __read_mostly;
 619
 620 /*
 621  * fasync_helper() is used by almost all character device drivers
 622  * to set up the fasync queue. It returns negative on error, 0 if it did
 623  * no changes and positive if it added/deleted the entry.
 624  */
 625 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
 626 {
 627         struct fasync_struct *fa, **fp;
 628         struct fasync_struct *new = NULL;
 629         int result = 0;
 630
 631         if (on) {
 632                 new = kmem_cache_alloc(fasync_cache, GFP_KERNEL);
 633                 if (!new)
 634                         return -ENOMEM;
 635         }
 636
 637         /*
 638          * We need to take f_lock first since it's not an IRQ-safe
 639          * lock.
 640          */
 641         spin_lock(&filp->f_lock);
 642         write_lock_irq(&fasync_lock);
 643         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 644                 if (fa->fa_file == filp) {
 645                         if(on) {
 646                                 fa->fa_fd = fd;
 647                                 kmem_cache_free(fasync_cache, new);
 648                         } else {
 649                                 *fp = fa->fa_next;
 650                                 kmem_cache_free(fasync_cache, fa);
 651                                 result = 1;
 652                         }
 653                         goto out;
 654                 }
 655         }
 656
 657         if (on) {
 658                 new->magic = FASYNC_MAGIC;
 659                 new->fa_file = filp;
 660                 new->fa_fd = fd;
 661                 new->fa_next = *fapp;
 662                 *fapp = new;
 663                 result = 1;
 664         }
 665 out:
 666         if (on)
 667                 filp->f_flags |= FASYNC;
 668         else
 669                 filp->f_flags &= ~FASYNC;
 670         write_unlock_irq(&fasync_lock);
 671         spin_unlock(&filp->f_lock);
 672         return result;
 673 }
 674
 675 EXPORT_SYMBOL(fasync_helper);
 676
 677 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
 678 {
 679         while (fa) {
 680                 struct fown_struct * fown;
 681                 if (fa->magic != FASYNC_MAGIC) {
 682                         printk(KERN_ERR "kill_fasync: bad magic number in "
 683                                "fasync_struct!\n");
 684                         return;
 685                 }
 686                 fown = &fa->fa_file->f_owner;
 687                 /* Don't send SIGURG to processes which have not set a
 688                    queued signum: SIGURG has its own default signalling
 689                    mechanism. */
 690                 if (!(sig == SIGURG && fown->signum == 0))
 691                         send_sigio(fown, fa->fa_fd, band);
 692                 fa = fa->fa_next;
 693         }
 694 }
 695
 696 EXPORT_SYMBOL(__kill_fasync);
 697
 698 void kill_fasync(struct fasync_struct **fp, int sig, int band)
 699 {
 700         /* First a quick test without locking: usually
 701          * the list is empty.
 702          */
 703         if (*fp) {
 704                 read_lock(&fasync_lock);
 705                 /* reread *fp after obtaining the lock */
 706                 __kill_fasync(*fp, sig, band);
 707                 read_unlock(&fasync_lock);
 708         }
 709 }
 710 EXPORT_SYMBOL(kill_fasync);
 711
 712 static int __init fasync_init(void)
 713 {
 714         fasync_cache = kmem_cache_create("fasync_cache",
 715                 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
 716         return 0;
 717 }
 718
 719 module_init(fasync_init)