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 /*
  53  * locate_fd finds a free file descriptor in the open_fds fdset,
  54  * expanding the fd arrays if necessary.  Must be called with the
  55  * file_lock held for write.
  56  */
  57
  58 static int locate_fd(unsigned int orig_start, int cloexec)
  59 {
  60         struct files_struct *files = current->files;
  61         unsigned int newfd;
  62         unsigned int start;
  63         int error;
  64         struct fdtable *fdt;
  65
  66         spin_lock(&files->file_lock);
  67
  68         error = -EINVAL;
  69         if (orig_start >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
  70                 goto out;
  71
  72 repeat:
  73         fdt = files_fdtable(files);
  74         /*
  75          * Someone might have closed fd's in the range
  76          * orig_start..fdt->next_fd
  77          */
  78         start = orig_start;
  79         if (start < files->next_fd)
  80                 start = files->next_fd;
  81
  82         newfd = start;
  83         if (start < fdt->max_fds)
  84                 newfd = find_next_zero_bit(fdt->open_fds->fds_bits,
  85                                            fdt->max_fds, start);
  86
  87         error = -EMFILE;
  88         if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
  89                 goto out;
  90
  91         error = expand_files(files, newfd);
  92         if (error < 0)
  93                 goto out;
  94
  95         /*
  96          * If we needed to expand the fs array we
  97          * might have blocked - try again.
  98          */
  99         if (error)
 100                 goto repeat;
 101
 102         if (start <= files->next_fd)
 103                 files->next_fd = newfd + 1;
 104
 105         FD_SET(newfd, fdt->open_fds);
 106         if (cloexec)
 107                 FD_SET(newfd, fdt->close_on_exec);
 108         else
 109                 FD_CLR(newfd, fdt->close_on_exec);
 110         error = newfd;
 111
 112 out:
 113         spin_unlock(&files->file_lock);
 114         return error;
 115 }
 116
 117 static int dupfd(struct file *file, unsigned int start, int cloexec)
 118 {
 119         int fd = locate_fd(start, cloexec);
 120         if (fd >= 0)
 121                 fd_install(fd, file);
 122         else
 123                 fput(file);
 124
 125         return fd;
 126 }
 127
 128 asmlinkage long sys_dup3(unsigned int oldfd, unsigned int newfd, int flags)
 129 {
 130         int err = -EBADF;
 131         struct file * file, *tofree;
 132         struct files_struct * files = current->files;
 133         struct fdtable *fdt;
 134
 135         if ((flags & ~O_CLOEXEC) != 0)
 136                 return -EINVAL;
 137
 138         spin_lock(&files->file_lock);
 139         if (!(file = fcheck(oldfd)))
 140                 goto out_unlock;
 141         err = newfd;
 142         if (newfd == oldfd)
 143                 goto out_unlock;
 144         err = -EBADF;
 145         if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
 146                 goto out_unlock;
 147         get_file(file);                 /* We are now finished with oldfd */
 148
 149         err = expand_files(files, newfd);
 150         if (err < 0)
 151                 goto out_fput;
 152
 153         /* To avoid races with open() and dup(), we will mark the fd as
 154          * in-use in the open-file bitmap throughout the entire dup2()
 155          * process.  This is quite safe: do_close() uses the fd array
 156          * entry, not the bitmap, to decide what work needs to be
 157          * done.  --sct */
 158         /* Doesn't work. open() might be there first. --AV */
 159
 160         /* Yes. It's a race. In user space. Nothing sane to do */
 161         err = -EBUSY;
 162         fdt = files_fdtable(files);
 163         tofree = fdt->fd[newfd];
 164         if (!tofree && FD_ISSET(newfd, fdt->open_fds))
 165                 goto out_fput;
 166
 167         rcu_assign_pointer(fdt->fd[newfd], file);
 168         FD_SET(newfd, fdt->open_fds);
 169         if (flags & O_CLOEXEC)
 170                 FD_SET(newfd, fdt->close_on_exec);
 171         else
 172                 FD_CLR(newfd, fdt->close_on_exec);
 173         spin_unlock(&files->file_lock);
 174
 175         if (tofree)
 176                 filp_close(tofree, files);
 177         err = newfd;
 178 out:
 179         return err;
 180 out_unlock:
 181         spin_unlock(&files->file_lock);
 182         goto out;
 183
 184 out_fput:
 185         spin_unlock(&files->file_lock);
 186         fput(file);
 187         goto out;
 188 }
 189
 190 asmlinkage long sys_dup2(unsigned int oldfd, unsigned int newfd)
 191 {
 192         return sys_dup3(oldfd, newfd, 0);
 193 }
 194
 195 asmlinkage long sys_dup(unsigned int fildes)
 196 {
 197         int ret = -EBADF;
 198         struct file * file = fget(fildes);
 199
 200         if (file)
 201                 ret = dupfd(file, 0, 0);
 202         return ret;
 203 }
 204
 205 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
 206
 207 static int setfl(int fd, struct file * filp, unsigned long arg)
 208 {
 209         struct inode * inode = filp->f_path.dentry->d_inode;
 210         int error = 0;
 211
 212         /*
 213          * O_APPEND cannot be cleared if the file is marked as append-only
 214          * and the file is open for write.
 215          */
 216         if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
 217                 return -EPERM;
 218
 219         /* O_NOATIME can only be set by the owner or superuser */
 220         if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
 221                 if (!is_owner_or_cap(inode))
 222                         return -EPERM;
 223
 224         /* required for strict SunOS emulation */
 225         if (O_NONBLOCK != O_NDELAY)
 226                if (arg & O_NDELAY)
 227                    arg |= O_NONBLOCK;
 228
 229         if (arg & O_DIRECT) {
 230                 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
 231                         !filp->f_mapping->a_ops->direct_IO)
 232                                 return -EINVAL;
 233         }
 234
 235         if (filp->f_op && filp->f_op->check_flags)
 236                 error = filp->f_op->check_flags(arg);
 237         if (error)
 238                 return error;
 239
 240         if ((arg ^ filp->f_flags) & FASYNC) {
 241                 if (filp->f_op && filp->f_op->fasync) {
 242                         error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
 243                         if (error < 0)
 244                                 goto out;
 245                 }
 246         }
 247
 248         filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
 249  out:
 250         return error;
 251 }
 252
 253 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
 254                      uid_t uid, uid_t euid, int force)
 255 {
 256         write_lock_irq(&filp->f_owner.lock);
 257         if (force || !filp->f_owner.pid) {
 258                 put_pid(filp->f_owner.pid);
 259                 filp->f_owner.pid = get_pid(pid);
 260                 filp->f_owner.pid_type = type;
 261                 filp->f_owner.uid = uid;
 262                 filp->f_owner.euid = euid;
 263         }
 264         write_unlock_irq(&filp->f_owner.lock);
 265 }
 266
 267 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
 268                 int force)
 269 {
 270         int err;
 271
 272         err = security_file_set_fowner(filp);
 273         if (err)
 274                 return err;
 275
 276         f_modown(filp, pid, type, current->uid, current->euid, force);
 277         return 0;
 278 }
 279 EXPORT_SYMBOL(__f_setown);
 280
 281 int f_setown(struct file *filp, unsigned long arg, int force)
 282 {
 283         enum pid_type type;
 284         struct pid *pid;
 285         int who = arg;
 286         int result;
 287         type = PIDTYPE_PID;
 288         if (who < 0) {
 289                 type = PIDTYPE_PGID;
 290                 who = -who;
 291         }
 292         rcu_read_lock();
 293         pid = find_vpid(who);
 294         result = __f_setown(filp, pid, type, force);
 295         rcu_read_unlock();
 296         return result;
 297 }
 298 EXPORT_SYMBOL(f_setown);
 299
 300 void f_delown(struct file *filp)
 301 {
 302         f_modown(filp, NULL, PIDTYPE_PID, 0, 0, 1);
 303 }
 304
 305 pid_t f_getown(struct file *filp)
 306 {
 307         pid_t pid;
 308         read_lock(&filp->f_owner.lock);
 309         pid = pid_vnr(filp->f_owner.pid);
 310         if (filp->f_owner.pid_type == PIDTYPE_PGID)
 311                 pid = -pid;
 312         read_unlock(&filp->f_owner.lock);
 313         return pid;
 314 }
 315
 316 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
 317                 struct file *filp)
 318 {
 319         long err = -EINVAL;
 320
 321         switch (cmd) {
 322         case F_DUPFD:
 323         case F_DUPFD_CLOEXEC:
 324                 get_file(filp);
 325                 err = dupfd(filp, arg, cmd == F_DUPFD_CLOEXEC);
 326                 break;
 327         case F_GETFD:
 328                 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
 329                 break;
 330         case F_SETFD:
 331                 err = 0;
 332                 set_close_on_exec(fd, arg & FD_CLOEXEC);
 333                 break;
 334         case F_GETFL:
 335                 err = filp->f_flags;
 336                 break;
 337         case F_SETFL:
 338                 err = setfl(fd, filp, arg);
 339                 break;
 340         case F_GETLK:
 341                 err = fcntl_getlk(filp, (struct flock __user *) arg);
 342                 break;
 343         case F_SETLK:
 344         case F_SETLKW:
 345                 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
 346                 break;
 347         case F_GETOWN:
 348                 /*
 349                  * XXX If f_owner is a process group, the
 350                  * negative return value will get converted
 351                  * into an error.  Oops.  If we keep the
 352                  * current syscall conventions, the only way
 353                  * to fix this will be in libc.
 354                  */
 355                 err = f_getown(filp);
 356                 force_successful_syscall_return();
 357                 break;
 358         case F_SETOWN:
 359                 err = f_setown(filp, arg, 1);
 360                 break;
 361         case F_GETSIG:
 362                 err = filp->f_owner.signum;
 363                 break;
 364         case F_SETSIG:
 365                 /* arg == 0 restores default behaviour. */
 366                 if (!valid_signal(arg)) {
 367                         break;
 368                 }
 369                 err = 0;
 370                 filp->f_owner.signum = arg;
 371                 break;
 372         case F_GETLEASE:
 373                 err = fcntl_getlease(filp);
 374                 break;
 375         case F_SETLEASE:
 376                 err = fcntl_setlease(fd, filp, arg);
 377                 break;
 378         case F_NOTIFY:
 379                 err = fcntl_dirnotify(fd, filp, arg);
 380                 break;
 381         default:
 382                 break;
 383         }
 384         return err;
 385 }
 386
 387 asmlinkage long sys_fcntl(unsigned int fd, unsigned int cmd, unsigned long arg)
 388 {
 389         struct file *filp;
 390         long err = -EBADF;
 391
 392         filp = fget(fd);
 393         if (!filp)
 394                 goto out;
 395
 396         err = security_file_fcntl(filp, cmd, arg);
 397         if (err) {
 398                 fput(filp);
 399                 return err;
 400         }
 401
 402         err = do_fcntl(fd, cmd, arg, filp);
 403
 404         fput(filp);
 405 out:
 406         return err;
 407 }
 408
 409 #if BITS_PER_LONG == 32
 410 asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
 411 {
 412         struct file * filp;
 413         long err;
 414
 415         err = -EBADF;
 416         filp = fget(fd);
 417         if (!filp)
 418                 goto out;
 419
 420         err = security_file_fcntl(filp, cmd, arg);
 421         if (err) {
 422                 fput(filp);
 423                 return err;
 424         }
 425         err = -EBADF;
 426
 427         switch (cmd) {
 428                 case F_GETLK64:
 429                         err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
 430                         break;
 431                 case F_SETLK64:
 432                 case F_SETLKW64:
 433                         err = fcntl_setlk64(fd, filp, cmd,
 434                                         (struct flock64 __user *) arg);
 435                         break;
 436                 default:
 437                         err = do_fcntl(fd, cmd, arg, filp);
 438                         break;
 439         }
 440         fput(filp);
 441 out:
 442         return err;
 443 }
 444 #endif
 445
 446 /* Table to convert sigio signal codes into poll band bitmaps */
 447
 448 static const long band_table[NSIGPOLL] = {
 449         POLLIN | POLLRDNORM,                    /* POLL_IN */
 450         POLLOUT | POLLWRNORM | POLLWRBAND,      /* POLL_OUT */
 451         POLLIN | POLLRDNORM | POLLMSG,          /* POLL_MSG */
 452         POLLERR,                                /* POLL_ERR */
 453         POLLPRI | POLLRDBAND,                   /* POLL_PRI */
 454         POLLHUP | POLLERR                       /* POLL_HUP */
 455 };
 456
 457 static inline int sigio_perm(struct task_struct *p,
 458                              struct fown_struct *fown, int sig)
 459 {
 460         return (((fown->euid == 0) ||
 461                  (fown->euid == p->suid) || (fown->euid == p->uid) ||
 462                  (fown->uid == p->suid) || (fown->uid == p->uid)) &&
 463                 !security_file_send_sigiotask(p, fown, sig));
 464 }
 465
 466 static void send_sigio_to_task(struct task_struct *p,
 467                                struct fown_struct *fown,
 468                                int fd,
 469                                int reason)
 470 {
 471         if (!sigio_perm(p, fown, fown->signum))
 472                 return;
 473
 474         switch (fown->signum) {
 475                 siginfo_t si;
 476                 default:
 477                         /* Queue a rt signal with the appropriate fd as its
 478                            value.  We use SI_SIGIO as the source, not
 479                            SI_KERNEL, since kernel signals always get
 480                            delivered even if we can't queue.  Failure to
 481                            queue in this case _should_ be reported; we fall
 482                            back to SIGIO in that case. --sct */
 483                         si.si_signo = fown->signum;
 484                         si.si_errno = 0;
 485                         si.si_code  = reason;
 486                         /* Make sure we are called with one of the POLL_*
 487                            reasons, otherwise we could leak kernel stack into
 488                            userspace.  */
 489                         BUG_ON((reason & __SI_MASK) != __SI_POLL);
 490                         if (reason - POLL_IN >= NSIGPOLL)
 491                                 si.si_band  = ~0L;
 492                         else
 493                                 si.si_band = band_table[reason - POLL_IN];
 494                         si.si_fd    = fd;
 495                         if (!group_send_sig_info(fown->signum, &si, p))
 496                                 break;
 497                 /* fall-through: fall back on the old plain SIGIO signal */
 498                 case 0:
 499                         group_send_sig_info(SIGIO, SEND_SIG_PRIV, p);
 500         }
 501 }
 502
 503 void send_sigio(struct fown_struct *fown, int fd, int band)
 504 {
 505         struct task_struct *p;
 506         enum pid_type type;
 507         struct pid *pid;
 508
 509         read_lock(&fown->lock);
 510         type = fown->pid_type;
 511         pid = fown->pid;
 512         if (!pid)
 513                 goto out_unlock_fown;
 514
 515         read_lock(&tasklist_lock);
 516         do_each_pid_task(pid, type, p) {
 517                 send_sigio_to_task(p, fown, fd, band);
 518         } while_each_pid_task(pid, type, p);
 519         read_unlock(&tasklist_lock);
 520  out_unlock_fown:
 521         read_unlock(&fown->lock);
 522 }
 523
 524 static void send_sigurg_to_task(struct task_struct *p,
 525                                 struct fown_struct *fown)
 526 {
 527         if (sigio_perm(p, fown, SIGURG))
 528                 group_send_sig_info(SIGURG, SEND_SIG_PRIV, p);
 529 }
 530
 531 int send_sigurg(struct fown_struct *fown)
 532 {
 533         struct task_struct *p;
 534         enum pid_type type;
 535         struct pid *pid;
 536         int ret = 0;
 537
 538         read_lock(&fown->lock);
 539         type = fown->pid_type;
 540         pid = fown->pid;
 541         if (!pid)
 542                 goto out_unlock_fown;
 543
 544         ret = 1;
 545
 546         read_lock(&tasklist_lock);
 547         do_each_pid_task(pid, type, p) {
 548                 send_sigurg_to_task(p, fown);
 549         } while_each_pid_task(pid, type, p);
 550         read_unlock(&tasklist_lock);
 551  out_unlock_fown:
 552         read_unlock(&fown->lock);
 553         return ret;
 554 }
 555
 556 static DEFINE_RWLOCK(fasync_lock);
 557 static struct kmem_cache *fasync_cache __read_mostly;
 558
 559 /*
 560  * fasync_helper() is used by some character device drivers (mainly mice)
 561  * to set up the fasync queue. It returns negative on error, 0 if it did
 562  * no changes and positive if it added/deleted the entry.
 563  */
 564 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
 565 {
 566         struct fasync_struct *fa, **fp;
 567         struct fasync_struct *new = NULL;
 568         int result = 0;
 569
 570         if (on) {
 571                 new = kmem_cache_alloc(fasync_cache, GFP_KERNEL);
 572                 if (!new)
 573                         return -ENOMEM;
 574         }
 575         write_lock_irq(&fasync_lock);
 576         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 577                 if (fa->fa_file == filp) {
 578                         if(on) {
 579                                 fa->fa_fd = fd;
 580                                 kmem_cache_free(fasync_cache, new);
 581                         } else {
 582                                 *fp = fa->fa_next;
 583                                 kmem_cache_free(fasync_cache, fa);
 584                                 result = 1;
 585                         }
 586                         goto out;
 587                 }
 588         }
 589
 590         if (on) {
 591                 new->magic = FASYNC_MAGIC;
 592                 new->fa_file = filp;
 593                 new->fa_fd = fd;
 594                 new->fa_next = *fapp;
 595                 *fapp = new;
 596                 result = 1;
 597         }
 598 out:
 599         write_unlock_irq(&fasync_lock);
 600         return result;
 601 }
 602
 603 EXPORT_SYMBOL(fasync_helper);
 604
 605 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
 606 {
 607         while (fa) {
 608                 struct fown_struct * fown;
 609                 if (fa->magic != FASYNC_MAGIC) {
 610                         printk(KERN_ERR "kill_fasync: bad magic number in "
 611                                "fasync_struct!\n");
 612                         return;
 613                 }
 614                 fown = &fa->fa_file->f_owner;
 615                 /* Don't send SIGURG to processes which have not set a
 616                    queued signum: SIGURG has its own default signalling
 617                    mechanism. */
 618                 if (!(sig == SIGURG && fown->signum == 0))
 619                         send_sigio(fown, fa->fa_fd, band);
 620                 fa = fa->fa_next;
 621         }
 622 }
 623
 624 EXPORT_SYMBOL(__kill_fasync);
 625
 626 void kill_fasync(struct fasync_struct **fp, int sig, int band)
 627 {
 628         /* First a quick test without locking: usually
 629          * the list is empty.
 630          */
 631         if (*fp) {
 632                 read_lock(&fasync_lock);
 633                 /* reread *fp after obtaining the lock */
 634                 __kill_fasync(*fp, sig, band);
 635                 read_unlock(&fasync_lock);
 636         }
 637 }
 638 EXPORT_SYMBOL(kill_fasync);
 639
 640 static int __init fasync_init(void)
 641 {
 642         fasync_cache = kmem_cache_create("fasync_cache",
 643                 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
 644         return 0;
 645 }
 646
 647 module_init(fasync_init)