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