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                 rcu_read_lock();
 121                 if (!fcheck_files(files, oldfd))
 122                         oldfd = -EBADF;
 123                 rcu_read_unlock();
 124                 return oldfd;
 125         }
 126         return sys_dup3(oldfd, newfd, 0);
 127 }
 128
 129 SYSCALL_DEFINE1(dup, unsigned int, fildes)
 130 {
 131         int ret = -EBADF;
 132         struct file *file = fget(fildes);
 133
 134         if (file) {
 135                 ret = get_unused_fd();
 136                 if (ret >= 0)
 137                         fd_install(ret, file);
 138                 else
 139                         fput(file);
 140         }
 141         return ret;
 142 }
 143
 144 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
 145
 146 static int setfl(int fd, struct file * filp, unsigned long arg)
 147 {
 148         struct inode * inode = filp->f_path.dentry->d_inode;
 149         int error = 0;
 150
 151         /*
 152          * O_APPEND cannot be cleared if the file is marked as append-only
 153          * and the file is open for write.
 154          */
 155         if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
 156                 return -EPERM;
 157
 158         /* O_NOATIME can only be set by the owner or superuser */
 159         if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
 160                 if (!is_owner_or_cap(inode))
 161                         return -EPERM;
 162
 163         /* required for strict SunOS emulation */
 164         if (O_NONBLOCK != O_NDELAY)
 165                if (arg & O_NDELAY)
 166                    arg |= O_NONBLOCK;
 167
 168         if (arg & O_DIRECT) {
 169                 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
 170                         !filp->f_mapping->a_ops->direct_IO)
 171                                 return -EINVAL;
 172         }
 173
 174         if (filp->f_op && filp->f_op->check_flags)
 175                 error = filp->f_op->check_flags(arg);
 176         if (error)
 177                 return error;
 178
 179         /*
 180          * ->fasync() is responsible for setting the FASYNC bit.
 181          */
 182         if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op &&
 183                         filp->f_op->fasync) {
 184                 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
 185                 if (error < 0)
 186                         goto out;
 187                 if (error > 0)
 188                         error = 0;
 189         }
 190         spin_lock(&filp->f_lock);
 191         filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
 192         spin_unlock(&filp->f_lock);
 193
 194  out:
 195         return error;
 196 }
 197
 198 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
 199                      uid_t uid, uid_t euid, int force)
 200 {
 201         write_lock_irq(&filp->f_owner.lock);
 202         if (force || !filp->f_owner.pid) {
 203                 put_pid(filp->f_owner.pid);
 204                 filp->f_owner.pid = get_pid(pid);
 205                 filp->f_owner.pid_type = type;
 206                 filp->f_owner.uid = uid;
 207                 filp->f_owner.euid = euid;
 208         }
 209         write_unlock_irq(&filp->f_owner.lock);
 210 }
 211
 212 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
 213                 int force)
 214 {
 215         const struct cred *cred = current_cred();
 216         int err;
 217
 218         err = security_file_set_fowner(filp);
 219         if (err)
 220                 return err;
 221
 222         f_modown(filp, pid, type, cred->uid, cred->euid, force);
 223         return 0;
 224 }
 225 EXPORT_SYMBOL(__f_setown);
 226
 227 int f_setown(struct file *filp, unsigned long arg, int force)
 228 {
 229         enum pid_type type;
 230         struct pid *pid;
 231         int who = arg;
 232         int result;
 233         type = PIDTYPE_PID;
 234         if (who < 0) {
 235                 type = PIDTYPE_PGID;
 236                 who = -who;
 237         }
 238         rcu_read_lock();
 239         pid = find_vpid(who);
 240         result = __f_setown(filp, pid, type, force);
 241         rcu_read_unlock();
 242         return result;
 243 }
 244 EXPORT_SYMBOL(f_setown);
 245
 246 void f_delown(struct file *filp)
 247 {
 248         f_modown(filp, NULL, PIDTYPE_PID, 0, 0, 1);
 249 }
 250
 251 pid_t f_getown(struct file *filp)
 252 {
 253         pid_t pid;
 254         read_lock(&filp->f_owner.lock);
 255         pid = pid_vnr(filp->f_owner.pid);
 256         if (filp->f_owner.pid_type == PIDTYPE_PGID)
 257                 pid = -pid;
 258         read_unlock(&filp->f_owner.lock);
 259         return pid;
 260 }
 261
 262 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
 263                 struct file *filp)
 264 {
 265         long err = -EINVAL;
 266
 267         switch (cmd) {
 268         case F_DUPFD:
 269         case F_DUPFD_CLOEXEC:
 270                 if (arg >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
 271                         break;
 272                 err = alloc_fd(arg, cmd == F_DUPFD_CLOEXEC ? O_CLOEXEC : 0);
 273                 if (err >= 0) {
 274                         get_file(filp);
 275                         fd_install(err, filp);
 276                 }
 277                 break;
 278         case F_GETFD:
 279                 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
 280                 break;
 281         case F_SETFD:
 282                 err = 0;
 283                 set_close_on_exec(fd, arg & FD_CLOEXEC);
 284                 break;
 285         case F_GETFL:
 286                 err = filp->f_flags;
 287                 break;
 288         case F_SETFL:
 289                 err = setfl(fd, filp, arg);
 290                 break;
 291         case F_GETLK:
 292                 err = fcntl_getlk(filp, (struct flock __user *) arg);
 293                 break;
 294         case F_SETLK:
 295         case F_SETLKW:
 296                 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
 297                 break;
 298         case F_GETOWN:
 299                 /*
 300                  * XXX If f_owner is a process group, the
 301                  * negative return value will get converted
 302                  * into an error.  Oops.  If we keep the
 303                  * current syscall conventions, the only way
 304                  * to fix this will be in libc.
 305                  */
 306                 err = f_getown(filp);
 307                 force_successful_syscall_return();
 308                 break;
 309         case F_SETOWN:
 310                 err = f_setown(filp, arg, 1);
 311                 break;
 312         case F_GETSIG:
 313                 err = filp->f_owner.signum;
 314                 break;
 315         case F_SETSIG:
 316                 /* arg == 0 restores default behaviour. */
 317                 if (!valid_signal(arg)) {
 318                         break;
 319                 }
 320                 err = 0;
 321                 filp->f_owner.signum = arg;
 322                 break;
 323         case F_GETLEASE:
 324                 err = fcntl_getlease(filp);
 325                 break;
 326         case F_SETLEASE:
 327                 err = fcntl_setlease(fd, filp, arg);
 328                 break;
 329         case F_NOTIFY:
 330                 err = fcntl_dirnotify(fd, filp, arg);
 331                 break;
 332         default:
 333                 break;
 334         }
 335         return err;
 336 }
 337
 338 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
 339 {
 340         struct file *filp;
 341         long err = -EBADF;
 342
 343         filp = fget(fd);
 344         if (!filp)
 345                 goto out;
 346
 347         err = security_file_fcntl(filp, cmd, arg);
 348         if (err) {
 349                 fput(filp);
 350                 return err;
 351         }
 352
 353         err = do_fcntl(fd, cmd, arg, filp);
 354
 355         fput(filp);
 356 out:
 357         return err;
 358 }
 359
 360 #if BITS_PER_LONG == 32
 361 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
 362                 unsigned long, arg)
 363 {
 364         struct file * filp;
 365         long err;
 366
 367         err = -EBADF;
 368         filp = fget(fd);
 369         if (!filp)
 370                 goto out;
 371
 372         err = security_file_fcntl(filp, cmd, arg);
 373         if (err) {
 374                 fput(filp);
 375                 return err;
 376         }
 377         err = -EBADF;
 378
 379         switch (cmd) {
 380                 case F_GETLK64:
 381                         err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
 382                         break;
 383                 case F_SETLK64:
 384                 case F_SETLKW64:
 385                         err = fcntl_setlk64(fd, filp, cmd,
 386                                         (struct flock64 __user *) arg);
 387                         break;
 388                 default:
 389                         err = do_fcntl(fd, cmd, arg, filp);
 390                         break;
 391         }
 392         fput(filp);
 393 out:
 394         return err;
 395 }
 396 #endif
 397
 398 /* Table to convert sigio signal codes into poll band bitmaps */
 399
 400 static const long band_table[NSIGPOLL] = {
 401         POLLIN | POLLRDNORM,                    /* POLL_IN */
 402         POLLOUT | POLLWRNORM | POLLWRBAND,      /* POLL_OUT */
 403         POLLIN | POLLRDNORM | POLLMSG,          /* POLL_MSG */
 404         POLLERR,                                /* POLL_ERR */
 405         POLLPRI | POLLRDBAND,                   /* POLL_PRI */
 406         POLLHUP | POLLERR                       /* POLL_HUP */
 407 };
 408
 409 static inline int sigio_perm(struct task_struct *p,
 410                              struct fown_struct *fown, int sig)
 411 {
 412         const struct cred *cred;
 413         int ret;
 414
 415         rcu_read_lock();
 416         cred = __task_cred(p);
 417         ret = ((fown->euid == 0 ||
 418                 fown->euid == cred->suid || fown->euid == cred->uid ||
 419                 fown->uid  == cred->suid || fown->uid  == cred->uid) &&
 420                !security_file_send_sigiotask(p, fown, sig));
 421         rcu_read_unlock();
 422         return ret;
 423 }
 424
 425 static void send_sigio_to_task(struct task_struct *p,
 426                                struct fown_struct *fown,
 427                                int fd,
 428                                int reason)
 429 {
 430         if (!sigio_perm(p, fown, fown->signum))
 431                 return;
 432
 433         switch (fown->signum) {
 434                 siginfo_t si;
 435                 default:
 436                         /* Queue a rt signal with the appropriate fd as its
 437                            value.  We use SI_SIGIO as the source, not
 438                            SI_KERNEL, since kernel signals always get
 439                            delivered even if we can't queue.  Failure to
 440                            queue in this case _should_ be reported; we fall
 441                            back to SIGIO in that case. --sct */
 442                         si.si_signo = fown->signum;
 443                         si.si_errno = 0;
 444                         si.si_code  = reason;
 445                         /* Make sure we are called with one of the POLL_*
 446                            reasons, otherwise we could leak kernel stack into
 447                            userspace.  */
 448                         BUG_ON((reason & __SI_MASK) != __SI_POLL);
 449                         if (reason - POLL_IN >= NSIGPOLL)
 450                                 si.si_band  = ~0L;
 451                         else
 452                                 si.si_band = band_table[reason - POLL_IN];
 453                         si.si_fd    = fd;
 454                         if (!group_send_sig_info(fown->signum, &si, p))
 455                                 break;
 456                 /* fall-through: fall back on the old plain SIGIO signal */
 457                 case 0:
 458                         group_send_sig_info(SIGIO, SEND_SIG_PRIV, p);
 459         }
 460 }
 461
 462 void send_sigio(struct fown_struct *fown, int fd, int band)
 463 {
 464         struct task_struct *p;
 465         enum pid_type type;
 466         struct pid *pid;
 467
 468         read_lock(&fown->lock);
 469         type = fown->pid_type;
 470         pid = fown->pid;
 471         if (!pid)
 472                 goto out_unlock_fown;
 473
 474         read_lock(&tasklist_lock);
 475         do_each_pid_task(pid, type, p) {
 476                 send_sigio_to_task(p, fown, fd, band);
 477         } while_each_pid_task(pid, type, p);
 478         read_unlock(&tasklist_lock);
 479  out_unlock_fown:
 480         read_unlock(&fown->lock);
 481 }
 482
 483 static void send_sigurg_to_task(struct task_struct *p,
 484                                 struct fown_struct *fown)
 485 {
 486         if (sigio_perm(p, fown, SIGURG))
 487                 group_send_sig_info(SIGURG, SEND_SIG_PRIV, p);
 488 }
 489
 490 int send_sigurg(struct fown_struct *fown)
 491 {
 492         struct task_struct *p;
 493         enum pid_type type;
 494         struct pid *pid;
 495         int ret = 0;
 496
 497         read_lock(&fown->lock);
 498         type = fown->pid_type;
 499         pid = fown->pid;
 500         if (!pid)
 501                 goto out_unlock_fown;
 502
 503         ret = 1;
 504
 505         read_lock(&tasklist_lock);
 506         do_each_pid_task(pid, type, p) {
 507                 send_sigurg_to_task(p, fown);
 508         } while_each_pid_task(pid, type, p);
 509         read_unlock(&tasklist_lock);
 510  out_unlock_fown:
 511         read_unlock(&fown->lock);
 512         return ret;
 513 }
 514
 515 static DEFINE_RWLOCK(fasync_lock);
 516 static struct kmem_cache *fasync_cache __read_mostly;
 517
 518 /*
 519  * fasync_helper() is used by almost all character device drivers
 520  * to set up the fasync queue. It returns negative on error, 0 if it did
 521  * no changes and positive if it added/deleted the entry.
 522  */
 523 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
 524 {
 525         struct fasync_struct *fa, **fp;
 526         struct fasync_struct *new = NULL;
 527         int result = 0;
 528
 529         if (on) {
 530                 new = kmem_cache_alloc(fasync_cache, GFP_KERNEL);
 531                 if (!new)
 532                         return -ENOMEM;
 533         }
 534         write_lock_irq(&fasync_lock);
 535         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 536                 if (fa->fa_file == filp) {
 537                         if(on) {
 538                                 fa->fa_fd = fd;
 539                                 kmem_cache_free(fasync_cache, new);
 540                         } else {
 541                                 *fp = fa->fa_next;
 542                                 kmem_cache_free(fasync_cache, fa);
 543                                 result = 1;
 544                         }
 545                         goto out;
 546                 }
 547         }
 548
 549         if (on) {
 550                 new->magic = FASYNC_MAGIC;
 551                 new->fa_file = filp;
 552                 new->fa_fd = fd;
 553                 new->fa_next = *fapp;
 554                 *fapp = new;
 555                 result = 1;
 556         }
 557 out:
 558         /* Fix up FASYNC bit while still holding fasync_lock */
 559         spin_lock(&filp->f_lock);
 560         if (on)
 561                 filp->f_flags |= FASYNC;
 562         else
 563                 filp->f_flags &= ~FASYNC;
 564         spin_unlock(&filp->f_lock);
 565         write_unlock_irq(&fasync_lock);
 566         return result;
 567 }
 568
 569 EXPORT_SYMBOL(fasync_helper);
 570
 571 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
 572 {
 573         while (fa) {
 574                 struct fown_struct * fown;
 575                 if (fa->magic != FASYNC_MAGIC) {
 576                         printk(KERN_ERR "kill_fasync: bad magic number in "
 577                                "fasync_struct!\n");
 578                         return;
 579                 }
 580                 fown = &fa->fa_file->f_owner;
 581                 /* Don't send SIGURG to processes which have not set a
 582                    queued signum: SIGURG has its own default signalling
 583                    mechanism. */
 584                 if (!(sig == SIGURG && fown->signum == 0))
 585                         send_sigio(fown, fa->fa_fd, band);
 586                 fa = fa->fa_next;
 587         }
 588 }
 589
 590 EXPORT_SYMBOL(__kill_fasync);
 591
 592 void kill_fasync(struct fasync_struct **fp, int sig, int band)
 593 {
 594         /* First a quick test without locking: usually
 595          * the list is empty.
 596          */
 597         if (*fp) {
 598                 read_lock(&fasync_lock);
 599                 /* reread *fp after obtaining the lock */
 600                 __kill_fasync(*fp, sig, band);
 601                 read_unlock(&fasync_lock);
 602         }
 603 }
 604 EXPORT_SYMBOL(kill_fasync);
 605
 606 static int __init fasync_init(void)
 607 {
 608         fasync_cache = kmem_cache_create("fasync_cache",
 609                 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
 610         return 0;
 611 }
 612
 613 module_init(fasync_init)