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 long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
 267                 struct file *filp)
 268 {
 269         long err = -EINVAL;
 270
 271         switch (cmd) {
 272         case F_DUPFD:
 273         case F_DUPFD_CLOEXEC:
 274                 if (arg >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
 275                         break;
 276                 err = alloc_fd(arg, cmd == F_DUPFD_CLOEXEC ? O_CLOEXEC : 0);
 277                 if (err >= 0) {
 278                         get_file(filp);
 279                         fd_install(err, filp);
 280                 }
 281                 break;
 282         case F_GETFD:
 283                 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
 284                 break;
 285         case F_SETFD:
 286                 err = 0;
 287                 set_close_on_exec(fd, arg & FD_CLOEXEC);
 288                 break;
 289         case F_GETFL:
 290                 err = filp->f_flags;
 291                 break;
 292         case F_SETFL:
 293                 err = setfl(fd, filp, arg);
 294                 break;
 295         case F_GETLK:
 296                 err = fcntl_getlk(filp, (struct flock __user *) arg);
 297                 break;
 298         case F_SETLK:
 299         case F_SETLKW:
 300                 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
 301                 break;
 302         case F_GETOWN:
 303                 /*
 304                  * XXX If f_owner is a process group, the
 305                  * negative return value will get converted
 306                  * into an error.  Oops.  If we keep the
 307                  * current syscall conventions, the only way
 308                  * to fix this will be in libc.
 309                  */
 310                 err = f_getown(filp);
 311                 force_successful_syscall_return();
 312                 break;
 313         case F_SETOWN:
 314                 err = f_setown(filp, arg, 1);
 315                 break;
 316         case F_GETSIG:
 317                 err = filp->f_owner.signum;
 318                 break;
 319         case F_SETSIG:
 320                 /* arg == 0 restores default behaviour. */
 321                 if (!valid_signal(arg)) {
 322                         break;
 323                 }
 324                 err = 0;
 325                 filp->f_owner.signum = arg;
 326                 break;
 327         case F_GETLEASE:
 328                 err = fcntl_getlease(filp);
 329                 break;
 330         case F_SETLEASE:
 331                 err = fcntl_setlease(fd, filp, arg);
 332                 break;
 333         case F_NOTIFY:
 334                 err = fcntl_dirnotify(fd, filp, arg);
 335                 break;
 336         default:
 337                 break;
 338         }
 339         return err;
 340 }
 341
 342 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
 343 {
 344         struct file *filp;
 345         long err = -EBADF;
 346
 347         filp = fget(fd);
 348         if (!filp)
 349                 goto out;
 350
 351         err = security_file_fcntl(filp, cmd, arg);
 352         if (err) {
 353                 fput(filp);
 354                 return err;
 355         }
 356
 357         err = do_fcntl(fd, cmd, arg, filp);
 358
 359         fput(filp);
 360 out:
 361         return err;
 362 }
 363
 364 #if BITS_PER_LONG == 32
 365 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
 366                 unsigned long, arg)
 367 {
 368         struct file * filp;
 369         long err;
 370
 371         err = -EBADF;
 372         filp = fget(fd);
 373         if (!filp)
 374                 goto out;
 375
 376         err = security_file_fcntl(filp, cmd, arg);
 377         if (err) {
 378                 fput(filp);
 379                 return err;
 380         }
 381         err = -EBADF;
 382
 383         switch (cmd) {
 384                 case F_GETLK64:
 385                         err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
 386                         break;
 387                 case F_SETLK64:
 388                 case F_SETLKW64:
 389                         err = fcntl_setlk64(fd, filp, cmd,
 390                                         (struct flock64 __user *) arg);
 391                         break;
 392                 default:
 393                         err = do_fcntl(fd, cmd, arg, filp);
 394                         break;
 395         }
 396         fput(filp);
 397 out:
 398         return err;
 399 }
 400 #endif
 401
 402 /* Table to convert sigio signal codes into poll band bitmaps */
 403
 404 static const long band_table[NSIGPOLL] = {
 405         POLLIN | POLLRDNORM,                    /* POLL_IN */
 406         POLLOUT | POLLWRNORM | POLLWRBAND,      /* POLL_OUT */
 407         POLLIN | POLLRDNORM | POLLMSG,          /* POLL_MSG */
 408         POLLERR,                                /* POLL_ERR */
 409         POLLPRI | POLLRDBAND,                   /* POLL_PRI */
 410         POLLHUP | POLLERR                       /* POLL_HUP */
 411 };
 412
 413 static inline int sigio_perm(struct task_struct *p,
 414                              struct fown_struct *fown, int sig)
 415 {
 416         const struct cred *cred;
 417         int ret;
 418
 419         rcu_read_lock();
 420         cred = __task_cred(p);
 421         ret = ((fown->euid == 0 ||
 422                 fown->euid == cred->suid || fown->euid == cred->uid ||
 423                 fown->uid  == cred->suid || fown->uid  == cred->uid) &&
 424                !security_file_send_sigiotask(p, fown, sig));
 425         rcu_read_unlock();
 426         return ret;
 427 }
 428
 429 static void send_sigio_to_task(struct task_struct *p,
 430                                struct fown_struct *fown,
 431                                int fd,
 432                                int reason)
 433 {
 434         /*
 435          * F_SETSIG can change ->signum lockless in parallel, make
 436          * sure we read it once and use the same value throughout.
 437          */
 438         int signum = ACCESS_ONCE(fown->signum);
 439
 440         if (!sigio_perm(p, fown, signum))
 441                 return;
 442
 443         switch (signum) {
 444                 siginfo_t si;
 445                 default:
 446                         /* Queue a rt signal with the appropriate fd as its
 447                            value.  We use SI_SIGIO as the source, not
 448                            SI_KERNEL, since kernel signals always get
 449                            delivered even if we can't queue.  Failure to
 450                            queue in this case _should_ be reported; we fall
 451                            back to SIGIO in that case. --sct */
 452                         si.si_signo = signum;
 453                         si.si_errno = 0;
 454                         si.si_code  = reason;
 455                         /* Make sure we are called with one of the POLL_*
 456                            reasons, otherwise we could leak kernel stack into
 457                            userspace.  */
 458                         BUG_ON((reason & __SI_MASK) != __SI_POLL);
 459                         if (reason - POLL_IN >= NSIGPOLL)
 460                                 si.si_band  = ~0L;
 461                         else
 462                                 si.si_band = band_table[reason - POLL_IN];
 463                         si.si_fd    = fd;
 464                         if (!group_send_sig_info(signum, &si, p))
 465                                 break;
 466                 /* fall-through: fall back on the old plain SIGIO signal */
 467                 case 0:
 468                         group_send_sig_info(SIGIO, SEND_SIG_PRIV, p);
 469         }
 470 }
 471
 472 void send_sigio(struct fown_struct *fown, int fd, int band)
 473 {
 474         struct task_struct *p;
 475         enum pid_type type;
 476         struct pid *pid;
 477
 478         read_lock(&fown->lock);
 479         type = fown->pid_type;
 480         pid = fown->pid;
 481         if (!pid)
 482                 goto out_unlock_fown;
 483
 484         read_lock(&tasklist_lock);
 485         do_each_pid_task(pid, type, p) {
 486                 send_sigio_to_task(p, fown, fd, band);
 487         } while_each_pid_task(pid, type, p);
 488         read_unlock(&tasklist_lock);
 489  out_unlock_fown:
 490         read_unlock(&fown->lock);
 491 }
 492
 493 static void send_sigurg_to_task(struct task_struct *p,
 494                                 struct fown_struct *fown)
 495 {
 496         if (sigio_perm(p, fown, SIGURG))
 497                 group_send_sig_info(SIGURG, SEND_SIG_PRIV, p);
 498 }
 499
 500 int send_sigurg(struct fown_struct *fown)
 501 {
 502         struct task_struct *p;
 503         enum pid_type type;
 504         struct pid *pid;
 505         int ret = 0;
 506
 507         read_lock(&fown->lock);
 508         type = fown->pid_type;
 509         pid = fown->pid;
 510         if (!pid)
 511                 goto out_unlock_fown;
 512
 513         ret = 1;
 514
 515         read_lock(&tasklist_lock);
 516         do_each_pid_task(pid, type, p) {
 517                 send_sigurg_to_task(p, fown);
 518         } while_each_pid_task(pid, type, p);
 519         read_unlock(&tasklist_lock);
 520  out_unlock_fown:
 521         read_unlock(&fown->lock);
 522         return ret;
 523 }
 524
 525 static DEFINE_RWLOCK(fasync_lock);
 526 static struct kmem_cache *fasync_cache __read_mostly;
 527
 528 /*
 529  * fasync_helper() is used by almost all character device drivers
 530  * to set up the fasync queue. It returns negative on error, 0 if it did
 531  * no changes and positive if it added/deleted the entry.
 532  */
 533 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
 534 {
 535         struct fasync_struct *fa, **fp;
 536         struct fasync_struct *new = NULL;
 537         int result = 0;
 538
 539         if (on) {
 540                 new = kmem_cache_alloc(fasync_cache, GFP_KERNEL);
 541                 if (!new)
 542                         return -ENOMEM;
 543         }
 544
 545         /*
 546          * We need to take f_lock first since it's not an IRQ-safe
 547          * lock.
 548          */
 549         spin_lock(&filp->f_lock);
 550         write_lock_irq(&fasync_lock);
 551         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 552                 if (fa->fa_file == filp) {
 553                         if(on) {
 554                                 fa->fa_fd = fd;
 555                                 kmem_cache_free(fasync_cache, new);
 556                         } else {
 557                                 *fp = fa->fa_next;
 558                                 kmem_cache_free(fasync_cache, fa);
 559                                 result = 1;
 560                         }
 561                         goto out;
 562                 }
 563         }
 564
 565         if (on) {
 566                 new->magic = FASYNC_MAGIC;
 567                 new->fa_file = filp;
 568                 new->fa_fd = fd;
 569                 new->fa_next = *fapp;
 570                 *fapp = new;
 571                 result = 1;
 572         }
 573 out:
 574         if (on)
 575                 filp->f_flags |= FASYNC;
 576         else
 577                 filp->f_flags &= ~FASYNC;
 578         write_unlock_irq(&fasync_lock);
 579         spin_unlock(&filp->f_lock);
 580         return result;
 581 }
 582
 583 EXPORT_SYMBOL(fasync_helper);
 584
 585 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
 586 {
 587         while (fa) {
 588                 struct fown_struct * fown;
 589                 if (fa->magic != FASYNC_MAGIC) {
 590                         printk(KERN_ERR "kill_fasync: bad magic number in "
 591                                "fasync_struct!\n");
 592                         return;
 593                 }
 594                 fown = &fa->fa_file->f_owner;
 595                 /* Don't send SIGURG to processes which have not set a
 596                    queued signum: SIGURG has its own default signalling
 597                    mechanism. */
 598                 if (!(sig == SIGURG && fown->signum == 0))
 599                         send_sigio(fown, fa->fa_fd, band);
 600                 fa = fa->fa_next;
 601         }
 602 }
 603
 604 EXPORT_SYMBOL(__kill_fasync);
 605
 606 void kill_fasync(struct fasync_struct **fp, int sig, int band)
 607 {
 608         /* First a quick test without locking: usually
 609          * the list is empty.
 610          */
 611         if (*fp) {
 612                 read_lock(&fasync_lock);
 613                 /* reread *fp after obtaining the lock */
 614                 __kill_fasync(*fp, sig, band);
 615                 read_unlock(&fasync_lock);
 616         }
 617 }
 618 EXPORT_SYMBOL(kill_fasync);
 619
 620 static int __init fasync_init(void)
 621 {
 622         fasync_cache = kmem_cache_create("fasync_cache",
 623                 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
 624         return 0;
 625 }
 626
 627 module_init(fasync_init)