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/smp_lock.h>
  16 #include <linux/slab.h>
  17 #include <linux/module.h>
  18 #include <linux/security.h>
  19 #include <linux/ptrace.h>
  20 #include <linux/signal.h>
  21 #include <linux/rcupdate.h>
  22 #include <linux/pid_namespace.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 /*
  54  * locate_fd finds a free file descriptor in the open_fds fdset,
  55  * expanding the fd arrays if necessary.  Must be called with the
  56  * file_lock held for write.
  57  */
  58
  59 static int locate_fd(unsigned int orig_start, int cloexec)
  60 {
  61         struct files_struct *files = current->files;
  62         unsigned int newfd;
  63         unsigned int start;
  64         int error;
  65         struct fdtable *fdt;
  66
  67         spin_lock(&files->file_lock);
  68
  69         error = -EINVAL;
  70         if (orig_start >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
  71                 goto out;
  72
  73 repeat:
  74         fdt = files_fdtable(files);
  75         /*
  76          * Someone might have closed fd's in the range
  77          * orig_start..fdt->next_fd
  78          */
  79         start = orig_start;
  80         if (start < files->next_fd)
  81                 start = files->next_fd;
  82
  83         newfd = start;
  84         if (start < fdt->max_fds)
  85                 newfd = find_next_zero_bit(fdt->open_fds->fds_bits,
  86                                            fdt->max_fds, start);
  87
  88         error = -EMFILE;
  89         if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
  90                 goto out;
  91
  92         error = expand_files(files, newfd);
  93         if (error < 0)
  94                 goto out;
  95
  96         /*
  97          * If we needed to expand the fs array we
  98          * might have blocked - try again.
  99          */
 100         if (error)
 101                 goto repeat;
 102
 103         if (start <= files->next_fd)
 104                 files->next_fd = newfd + 1;
 105
 106         FD_SET(newfd, fdt->open_fds);
 107         if (cloexec)
 108                 FD_SET(newfd, fdt->close_on_exec);
 109         else
 110                 FD_CLR(newfd, fdt->close_on_exec);
 111         error = newfd;
 112
 113 out:
 114         spin_unlock(&files->file_lock);
 115         return error;
 116 }
 117
 118 static int dupfd(struct file *file, unsigned int start, int cloexec)
 119 {
 120         int fd = locate_fd(start, cloexec);
 121         if (fd >= 0)
 122                 fd_install(fd, file);
 123         else
 124                 fput(file);
 125
 126         return fd;
 127 }
 128
 129 asmlinkage long sys_dup2(unsigned int oldfd, unsigned int newfd)
 130 {
 131         int err = -EBADF;
 132         struct file * file, *tofree;
 133         struct files_struct * files = current->files;
 134         struct fdtable *fdt;
 135
 136         spin_lock(&files->file_lock);
 137         if (!(file = fcheck(oldfd)))
 138                 goto out_unlock;
 139         err = newfd;
 140         if (newfd == oldfd)
 141                 goto out_unlock;
 142         err = -EBADF;
 143         if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
 144                 goto out_unlock;
 145         get_file(file);                 /* We are now finished with oldfd */
 146
 147         err = expand_files(files, newfd);
 148         if (err < 0)
 149                 goto out_fput;
 150
 151         /* To avoid races with open() and dup(), we will mark the fd as
 152          * in-use in the open-file bitmap throughout the entire dup2()
 153          * process.  This is quite safe: do_close() uses the fd array
 154          * entry, not the bitmap, to decide what work needs to be
 155          * done.  --sct */
 156         /* Doesn't work. open() might be there first. --AV */
 157
 158         /* Yes. It's a race. In user space. Nothing sane to do */
 159         err = -EBUSY;
 160         fdt = files_fdtable(files);
 161         tofree = fdt->fd[newfd];
 162         if (!tofree && FD_ISSET(newfd, fdt->open_fds))
 163                 goto out_fput;
 164
 165         rcu_assign_pointer(fdt->fd[newfd], file);
 166         FD_SET(newfd, fdt->open_fds);
 167         FD_CLR(newfd, fdt->close_on_exec);
 168         spin_unlock(&files->file_lock);
 169
 170         if (tofree)
 171                 filp_close(tofree, files);
 172         err = newfd;
 173 out:
 174         return err;
 175 out_unlock:
 176         spin_unlock(&files->file_lock);
 177         goto out;
 178
 179 out_fput:
 180         spin_unlock(&files->file_lock);
 181         fput(file);
 182         goto out;
 183 }
 184
 185 asmlinkage long sys_dup(unsigned int fildes)
 186 {
 187         int ret = -EBADF;
 188         struct file * file = fget(fildes);
 189
 190         if (file)
 191                 ret = dupfd(file, 0, 0);
 192         return ret;
 193 }
 194
 195 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
 196
 197 static int setfl(int fd, struct file * filp, unsigned long arg)
 198 {
 199         struct inode * inode = filp->f_path.dentry->d_inode;
 200         int error = 0;
 201
 202         /*
 203          * O_APPEND cannot be cleared if the file is marked as append-only
 204          * and the file is open for write.
 205          */
 206         if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
 207                 return -EPERM;
 208
 209         /* O_NOATIME can only be set by the owner or superuser */
 210         if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
 211                 if (!is_owner_or_cap(inode))
 212                         return -EPERM;
 213
 214         /* required for strict SunOS emulation */
 215         if (O_NONBLOCK != O_NDELAY)
 216                if (arg & O_NDELAY)
 217                    arg |= O_NONBLOCK;
 218
 219         if (arg & O_DIRECT) {
 220                 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
 221                         !filp->f_mapping->a_ops->direct_IO)
 222                                 return -EINVAL;
 223         }
 224
 225         if (filp->f_op && filp->f_op->check_flags)
 226                 error = filp->f_op->check_flags(arg);
 227         if (error)
 228                 return error;
 229
 230         lock_kernel();
 231         if ((arg ^ filp->f_flags) & FASYNC) {
 232                 if (filp->f_op && filp->f_op->fasync) {
 233                         error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
 234                         if (error < 0)
 235                                 goto out;
 236                 }
 237         }
 238
 239         filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
 240  out:
 241         unlock_kernel();
 242         return error;
 243 }
 244
 245 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
 246                      uid_t uid, uid_t euid, int force)
 247 {
 248         write_lock_irq(&filp->f_owner.lock);
 249         if (force || !filp->f_owner.pid) {
 250                 put_pid(filp->f_owner.pid);
 251                 filp->f_owner.pid = get_pid(pid);
 252                 filp->f_owner.pid_type = type;
 253                 filp->f_owner.uid = uid;
 254                 filp->f_owner.euid = euid;
 255         }
 256         write_unlock_irq(&filp->f_owner.lock);
 257 }
 258
 259 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
 260                 int force)
 261 {
 262         int err;
 263
 264         err = security_file_set_fowner(filp);
 265         if (err)
 266                 return err;
 267
 268         f_modown(filp, pid, type, current->uid, current->euid, force);
 269         return 0;
 270 }
 271 EXPORT_SYMBOL(__f_setown);
 272
 273 int f_setown(struct file *filp, unsigned long arg, int force)
 274 {
 275         enum pid_type type;
 276         struct pid *pid;
 277         int who = arg;
 278         int result;
 279         type = PIDTYPE_PID;
 280         if (who < 0) {
 281                 type = PIDTYPE_PGID;
 282                 who = -who;
 283         }
 284         rcu_read_lock();
 285         pid = find_vpid(who);
 286         result = __f_setown(filp, pid, type, force);
 287         rcu_read_unlock();
 288         return result;
 289 }
 290 EXPORT_SYMBOL(f_setown);
 291
 292 void f_delown(struct file *filp)
 293 {
 294         f_modown(filp, NULL, PIDTYPE_PID, 0, 0, 1);
 295 }
 296
 297 pid_t f_getown(struct file *filp)
 298 {
 299         pid_t pid;
 300         read_lock(&filp->f_owner.lock);
 301         pid = pid_vnr(filp->f_owner.pid);
 302         if (filp->f_owner.pid_type == PIDTYPE_PGID)
 303                 pid = -pid;
 304         read_unlock(&filp->f_owner.lock);
 305         return pid;
 306 }
 307
 308 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
 309                 struct file *filp)
 310 {
 311         long err = -EINVAL;
 312
 313         switch (cmd) {
 314         case F_DUPFD:
 315         case F_DUPFD_CLOEXEC:
 316                 get_file(filp);
 317                 err = dupfd(filp, arg, cmd == F_DUPFD_CLOEXEC);
 318                 break;
 319         case F_GETFD:
 320                 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
 321                 break;
 322         case F_SETFD:
 323                 err = 0;
 324                 set_close_on_exec(fd, arg & FD_CLOEXEC);
 325                 break;
 326         case F_GETFL:
 327                 err = filp->f_flags;
 328                 break;
 329         case F_SETFL:
 330                 err = setfl(fd, filp, arg);
 331                 break;
 332         case F_GETLK:
 333                 err = fcntl_getlk(filp, (struct flock __user *) arg);
 334                 break;
 335         case F_SETLK:
 336         case F_SETLKW:
 337                 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
 338                 break;
 339         case F_GETOWN:
 340                 /*
 341                  * XXX If f_owner is a process group, the
 342                  * negative return value will get converted
 343                  * into an error.  Oops.  If we keep the
 344                  * current syscall conventions, the only way
 345                  * to fix this will be in libc.
 346                  */
 347                 err = f_getown(filp);
 348                 force_successful_syscall_return();
 349                 break;
 350         case F_SETOWN:
 351                 err = f_setown(filp, arg, 1);
 352                 break;
 353         case F_GETSIG:
 354                 err = filp->f_owner.signum;
 355                 break;
 356         case F_SETSIG:
 357                 /* arg == 0 restores default behaviour. */
 358                 if (!valid_signal(arg)) {
 359                         break;
 360                 }
 361                 err = 0;
 362                 filp->f_owner.signum = arg;
 363                 break;
 364         case F_GETLEASE:
 365                 err = fcntl_getlease(filp);
 366                 break;
 367         case F_SETLEASE:
 368                 err = fcntl_setlease(fd, filp, arg);
 369                 break;
 370         case F_NOTIFY:
 371                 err = fcntl_dirnotify(fd, filp, arg);
 372                 break;
 373         default:
 374                 break;
 375         }
 376         return err;
 377 }
 378
 379 asmlinkage long sys_fcntl(unsigned int fd, unsigned int cmd, unsigned long arg)
 380 {
 381         struct file *filp;
 382         long err = -EBADF;
 383
 384         filp = fget(fd);
 385         if (!filp)
 386                 goto out;
 387
 388         err = security_file_fcntl(filp, cmd, arg);
 389         if (err) {
 390                 fput(filp);
 391                 return err;
 392         }
 393
 394         err = do_fcntl(fd, cmd, arg, filp);
 395
 396         fput(filp);
 397 out:
 398         return err;
 399 }
 400
 401 #if BITS_PER_LONG == 32
 402 asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
 403 {
 404         struct file * filp;
 405         long err;
 406
 407         err = -EBADF;
 408         filp = fget(fd);
 409         if (!filp)
 410                 goto out;
 411
 412         err = security_file_fcntl(filp, cmd, arg);
 413         if (err) {
 414                 fput(filp);
 415                 return err;
 416         }
 417         err = -EBADF;
 418
 419         switch (cmd) {
 420                 case F_GETLK64:
 421                         err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
 422                         break;
 423                 case F_SETLK64:
 424                 case F_SETLKW64:
 425                         err = fcntl_setlk64(fd, filp, cmd,
 426                                         (struct flock64 __user *) arg);
 427                         break;
 428                 default:
 429                         err = do_fcntl(fd, cmd, arg, filp);
 430                         break;
 431         }
 432         fput(filp);
 433 out:
 434         return err;
 435 }
 436 #endif
 437
 438 /* Table to convert sigio signal codes into poll band bitmaps */
 439
 440 static const long band_table[NSIGPOLL] = {
 441         POLLIN | POLLRDNORM,                    /* POLL_IN */
 442         POLLOUT | POLLWRNORM | POLLWRBAND,      /* POLL_OUT */
 443         POLLIN | POLLRDNORM | POLLMSG,          /* POLL_MSG */
 444         POLLERR,                                /* POLL_ERR */
 445         POLLPRI | POLLRDBAND,                   /* POLL_PRI */
 446         POLLHUP | POLLERR                       /* POLL_HUP */
 447 };
 448
 449 static inline int sigio_perm(struct task_struct *p,
 450                              struct fown_struct *fown, int sig)
 451 {
 452         return (((fown->euid == 0) ||
 453                  (fown->euid == p->suid) || (fown->euid == p->uid) ||
 454                  (fown->uid == p->suid) || (fown->uid == p->uid)) &&
 455                 !security_file_send_sigiotask(p, fown, sig));
 456 }
 457
 458 static void send_sigio_to_task(struct task_struct *p,
 459                                struct fown_struct *fown,
 460                                int fd,
 461                                int reason)
 462 {
 463         if (!sigio_perm(p, fown, fown->signum))
 464                 return;
 465
 466         switch (fown->signum) {
 467                 siginfo_t si;
 468                 default:
 469                         /* Queue a rt signal with the appropriate fd as its
 470                            value.  We use SI_SIGIO as the source, not
 471                            SI_KERNEL, since kernel signals always get
 472                            delivered even if we can't queue.  Failure to
 473                            queue in this case _should_ be reported; we fall
 474                            back to SIGIO in that case. --sct */
 475                         si.si_signo = fown->signum;
 476                         si.si_errno = 0;
 477                         si.si_code  = reason;
 478                         /* Make sure we are called with one of the POLL_*
 479                            reasons, otherwise we could leak kernel stack into
 480                            userspace.  */
 481                         BUG_ON((reason & __SI_MASK) != __SI_POLL);
 482                         if (reason - POLL_IN >= NSIGPOLL)
 483                                 si.si_band  = ~0L;
 484                         else
 485                                 si.si_band = band_table[reason - POLL_IN];
 486                         si.si_fd    = fd;
 487                         if (!group_send_sig_info(fown->signum, &si, p))
 488                                 break;
 489                 /* fall-through: fall back on the old plain SIGIO signal */
 490                 case 0:
 491                         group_send_sig_info(SIGIO, SEND_SIG_PRIV, p);
 492         }
 493 }
 494
 495 void send_sigio(struct fown_struct *fown, int fd, int band)
 496 {
 497         struct task_struct *p;
 498         enum pid_type type;
 499         struct pid *pid;
 500
 501         read_lock(&fown->lock);
 502         type = fown->pid_type;
 503         pid = fown->pid;
 504         if (!pid)
 505                 goto out_unlock_fown;
 506
 507         read_lock(&tasklist_lock);
 508         do_each_pid_task(pid, type, p) {
 509                 send_sigio_to_task(p, fown, fd, band);
 510         } while_each_pid_task(pid, type, p);
 511         read_unlock(&tasklist_lock);
 512  out_unlock_fown:
 513         read_unlock(&fown->lock);
 514 }
 515
 516 static void send_sigurg_to_task(struct task_struct *p,
 517                                 struct fown_struct *fown)
 518 {
 519         if (sigio_perm(p, fown, SIGURG))
 520                 group_send_sig_info(SIGURG, SEND_SIG_PRIV, p);
 521 }
 522
 523 int send_sigurg(struct fown_struct *fown)
 524 {
 525         struct task_struct *p;
 526         enum pid_type type;
 527         struct pid *pid;
 528         int ret = 0;
 529
 530         read_lock(&fown->lock);
 531         type = fown->pid_type;
 532         pid = fown->pid;
 533         if (!pid)
 534                 goto out_unlock_fown;
 535
 536         ret = 1;
 537
 538         read_lock(&tasklist_lock);
 539         do_each_pid_task(pid, type, p) {
 540                 send_sigurg_to_task(p, fown);
 541         } while_each_pid_task(pid, type, p);
 542         read_unlock(&tasklist_lock);
 543  out_unlock_fown:
 544         read_unlock(&fown->lock);
 545         return ret;
 546 }
 547
 548 static DEFINE_RWLOCK(fasync_lock);
 549 static struct kmem_cache *fasync_cache __read_mostly;
 550
 551 /*
 552  * fasync_helper() is used by some character device drivers (mainly mice)
 553  * to set up the fasync queue. It returns negative on error, 0 if it did
 554  * no changes and positive if it added/deleted the entry.
 555  */
 556 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
 557 {
 558         struct fasync_struct *fa, **fp;
 559         struct fasync_struct *new = NULL;
 560         int result = 0;
 561
 562         if (on) {
 563                 new = kmem_cache_alloc(fasync_cache, GFP_KERNEL);
 564                 if (!new)
 565                         return -ENOMEM;
 566         }
 567         write_lock_irq(&fasync_lock);
 568         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 569                 if (fa->fa_file == filp) {
 570                         if(on) {
 571                                 fa->fa_fd = fd;
 572                                 kmem_cache_free(fasync_cache, new);
 573                         } else {
 574                                 *fp = fa->fa_next;
 575                                 kmem_cache_free(fasync_cache, fa);
 576                                 result = 1;
 577                         }
 578                         goto out;
 579                 }
 580         }
 581
 582         if (on) {
 583                 new->magic = FASYNC_MAGIC;
 584                 new->fa_file = filp;
 585                 new->fa_fd = fd;
 586                 new->fa_next = *fapp;
 587                 *fapp = new;
 588                 result = 1;
 589         }
 590 out:
 591         write_unlock_irq(&fasync_lock);
 592         return result;
 593 }
 594
 595 EXPORT_SYMBOL(fasync_helper);
 596
 597 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
 598 {
 599         while (fa) {
 600                 struct fown_struct * fown;
 601                 if (fa->magic != FASYNC_MAGIC) {
 602                         printk(KERN_ERR "kill_fasync: bad magic number in "
 603                                "fasync_struct!\n");
 604                         return;
 605                 }
 606                 fown = &fa->fa_file->f_owner;
 607                 /* Don't send SIGURG to processes which have not set a
 608                    queued signum: SIGURG has its own default signalling
 609                    mechanism. */
 610                 if (!(sig == SIGURG && fown->signum == 0))
 611                         send_sigio(fown, fa->fa_fd, band);
 612                 fa = fa->fa_next;
 613         }
 614 }
 615
 616 EXPORT_SYMBOL(__kill_fasync);
 617
 618 void kill_fasync(struct fasync_struct **fp, int sig, int band)
 619 {
 620         /* First a quick test without locking: usually
 621          * the list is empty.
 622          */
 623         if (*fp) {
 624                 read_lock(&fasync_lock);
 625                 /* reread *fp after obtaining the lock */
 626                 __kill_fasync(*fp, sig, band);
 627                 read_unlock(&fasync_lock);
 628         }
 629 }
 630 EXPORT_SYMBOL(kill_fasync);
 631
 632 static int __init fasync_init(void)
 633 {
 634         fasync_cache = kmem_cache_create("fasync_cache",
 635                 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
 636         return 0;
 637 }
 638
 639 module_init(fasync_init)