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/capability.h>
  13 #include <linux/dnotify.h>
  14 #include <linux/smp_lock.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_dup2(unsigned int oldfd, unsigned int newfd)
 129 {
 130         int err = -EBADF;
 131         struct file * file, *tofree;
 132         struct files_struct * files = current->files;
 133         struct fdtable *fdt;
 134
 135         spin_lock(&files->file_lock);
 136         if (!(file = fcheck(oldfd)))
 137                 goto out_unlock;
 138         err = newfd;
 139         if (newfd == oldfd)
 140                 goto out_unlock;
 141         err = -EBADF;
 142         if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
 143                 goto out_unlock;
 144         get_file(file);                 /* We are now finished with oldfd */
 145
 146         err = expand_files(files, newfd);
 147         if (err < 0)
 148                 goto out_fput;
 149
 150         /* To avoid races with open() and dup(), we will mark the fd as
 151          * in-use in the open-file bitmap throughout the entire dup2()
 152          * process.  This is quite safe: do_close() uses the fd array
 153          * entry, not the bitmap, to decide what work needs to be
 154          * done.  --sct */
 155         /* Doesn't work. open() might be there first. --AV */
 156
 157         /* Yes. It's a race. In user space. Nothing sane to do */
 158         err = -EBUSY;
 159         fdt = files_fdtable(files);
 160         tofree = fdt->fd[newfd];
 161         if (!tofree && FD_ISSET(newfd, fdt->open_fds))
 162                 goto out_fput;
 163
 164         rcu_assign_pointer(fdt->fd[newfd], file);
 165         FD_SET(newfd, fdt->open_fds);
 166         FD_CLR(newfd, fdt->close_on_exec);
 167         spin_unlock(&files->file_lock);
 168
 169         if (tofree)
 170                 filp_close(tofree, files);
 171         err = newfd;
 172 out:
 173         return err;
 174 out_unlock:
 175         spin_unlock(&files->file_lock);
 176         goto out;
 177
 178 out_fput:
 179         spin_unlock(&files->file_lock);
 180         fput(file);
 181         goto out;
 182 }
 183
 184 asmlinkage long sys_dup(unsigned int fildes)
 185 {
 186         int ret = -EBADF;
 187         struct file * file = fget(fildes);
 188
 189         if (file)
 190                 ret = dupfd(file, 0, 0);
 191         return ret;
 192 }
 193
 194 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
 195
 196 static int setfl(int fd, struct file * filp, unsigned long arg)
 197 {
 198         struct inode * inode = filp->f_path.dentry->d_inode;
 199         int error = 0;
 200
 201         /*
 202          * O_APPEND cannot be cleared if the file is marked as append-only
 203          * and the file is open for write.
 204          */
 205         if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
 206                 return -EPERM;
 207
 208         /* O_NOATIME can only be set by the owner or superuser */
 209         if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
 210                 if (!is_owner_or_cap(inode))
 211                         return -EPERM;
 212
 213         /* required for strict SunOS emulation */
 214         if (O_NONBLOCK != O_NDELAY)
 215                if (arg & O_NDELAY)
 216                    arg |= O_NONBLOCK;
 217
 218         if (arg & O_DIRECT) {
 219                 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
 220                         !filp->f_mapping->a_ops->direct_IO)
 221                                 return -EINVAL;
 222         }
 223
 224         if (filp->f_op && filp->f_op->check_flags)
 225                 error = filp->f_op->check_flags(arg);
 226         if (error)
 227                 return error;
 228
 229         lock_kernel();
 230         if ((arg ^ filp->f_flags) & FASYNC) {
 231                 if (filp->f_op && filp->f_op->fasync) {
 232                         error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
 233                         if (error < 0)
 234                                 goto out;
 235                 }
 236         }
 237
 238         filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
 239  out:
 240         unlock_kernel();
 241         return error;
 242 }
 243
 244 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
 245                      uid_t uid, uid_t euid, int force)
 246 {
 247         write_lock_irq(&filp->f_owner.lock);
 248         if (force || !filp->f_owner.pid) {
 249                 put_pid(filp->f_owner.pid);
 250                 filp->f_owner.pid = get_pid(pid);
 251                 filp->f_owner.pid_type = type;
 252                 filp->f_owner.uid = uid;
 253                 filp->f_owner.euid = euid;
 254         }
 255         write_unlock_irq(&filp->f_owner.lock);
 256 }
 257
 258 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
 259                 int force)
 260 {
 261         int err;
 262
 263         err = security_file_set_fowner(filp);
 264         if (err)
 265                 return err;
 266
 267         f_modown(filp, pid, type, current->uid, current->euid, force);
 268         return 0;
 269 }
 270 EXPORT_SYMBOL(__f_setown);
 271
 272 int f_setown(struct file *filp, unsigned long arg, int force)
 273 {
 274         enum pid_type type;
 275         struct pid *pid;
 276         int who = arg;
 277         int result;
 278         type = PIDTYPE_PID;
 279         if (who < 0) {
 280                 type = PIDTYPE_PGID;
 281                 who = -who;
 282         }
 283         rcu_read_lock();
 284         pid = find_vpid(who);
 285         result = __f_setown(filp, pid, type, force);
 286         rcu_read_unlock();
 287         return result;
 288 }
 289 EXPORT_SYMBOL(f_setown);
 290
 291 void f_delown(struct file *filp)
 292 {
 293         f_modown(filp, NULL, PIDTYPE_PID, 0, 0, 1);
 294 }
 295
 296 pid_t f_getown(struct file *filp)
 297 {
 298         pid_t pid;
 299         read_lock(&filp->f_owner.lock);
 300         pid = pid_vnr(filp->f_owner.pid);
 301         if (filp->f_owner.pid_type == PIDTYPE_PGID)
 302                 pid = -pid;
 303         read_unlock(&filp->f_owner.lock);
 304         return pid;
 305 }
 306
 307 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
 308                 struct file *filp)
 309 {
 310         long err = -EINVAL;
 311
 312         switch (cmd) {
 313         case F_DUPFD:
 314         case F_DUPFD_CLOEXEC:
 315                 get_file(filp);
 316                 err = dupfd(filp, arg, cmd == F_DUPFD_CLOEXEC);
 317                 break;
 318         case F_GETFD:
 319                 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
 320                 break;
 321         case F_SETFD:
 322                 err = 0;
 323                 set_close_on_exec(fd, arg & FD_CLOEXEC);
 324                 break;
 325         case F_GETFL:
 326                 err = filp->f_flags;
 327                 break;
 328         case F_SETFL:
 329                 err = setfl(fd, filp, arg);
 330                 break;
 331         case F_GETLK:
 332                 err = fcntl_getlk(filp, (struct flock __user *) arg);
 333                 break;
 334         case F_SETLK:
 335         case F_SETLKW:
 336                 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
 337                 break;
 338         case F_GETOWN:
 339                 /*
 340                  * XXX If f_owner is a process group, the
 341                  * negative return value will get converted
 342                  * into an error.  Oops.  If we keep the
 343                  * current syscall conventions, the only way
 344                  * to fix this will be in libc.
 345                  */
 346                 err = f_getown(filp);
 347                 force_successful_syscall_return();
 348                 break;
 349         case F_SETOWN:
 350                 err = f_setown(filp, arg, 1);
 351                 break;
 352         case F_GETSIG:
 353                 err = filp->f_owner.signum;
 354                 break;
 355         case F_SETSIG:
 356                 /* arg == 0 restores default behaviour. */
 357                 if (!valid_signal(arg)) {
 358                         break;
 359                 }
 360                 err = 0;
 361                 filp->f_owner.signum = arg;
 362                 break;
 363         case F_GETLEASE:
 364                 err = fcntl_getlease(filp);
 365                 break;
 366         case F_SETLEASE:
 367                 err = fcntl_setlease(fd, filp, arg);
 368                 break;
 369         case F_NOTIFY:
 370                 err = fcntl_dirnotify(fd, filp, arg);
 371                 break;
 372         default:
 373                 break;
 374         }
 375         return err;
 376 }
 377
 378 asmlinkage long sys_fcntl(unsigned int fd, unsigned int cmd, unsigned long arg)
 379 {
 380         struct file *filp;
 381         long err = -EBADF;
 382
 383         filp = fget(fd);
 384         if (!filp)
 385                 goto out;
 386
 387         err = security_file_fcntl(filp, cmd, arg);
 388         if (err) {
 389                 fput(filp);
 390                 return err;
 391         }
 392
 393         err = do_fcntl(fd, cmd, arg, filp);
 394
 395         fput(filp);
 396 out:
 397         return err;
 398 }
 399
 400 #if BITS_PER_LONG == 32
 401 asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
 402 {
 403         struct file * filp;
 404         long err;
 405
 406         err = -EBADF;
 407         filp = fget(fd);
 408         if (!filp)
 409                 goto out;
 410
 411         err = security_file_fcntl(filp, cmd, arg);
 412         if (err) {
 413                 fput(filp);
 414                 return err;
 415         }
 416         err = -EBADF;
 417
 418         switch (cmd) {
 419                 case F_GETLK64:
 420                         err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
 421                         break;
 422                 case F_SETLK64:
 423                 case F_SETLKW64:
 424                         err = fcntl_setlk64(fd, filp, cmd,
 425                                         (struct flock64 __user *) arg);
 426                         break;
 427                 default:
 428                         err = do_fcntl(fd, cmd, arg, filp);
 429                         break;
 430         }
 431         fput(filp);
 432 out:
 433         return err;
 434 }
 435 #endif
 436
 437 /* Table to convert sigio signal codes into poll band bitmaps */
 438
 439 static const long band_table[NSIGPOLL] = {
 440         POLLIN | POLLRDNORM,                    /* POLL_IN */
 441         POLLOUT | POLLWRNORM | POLLWRBAND,      /* POLL_OUT */
 442         POLLIN | POLLRDNORM | POLLMSG,          /* POLL_MSG */
 443         POLLERR,                                /* POLL_ERR */
 444         POLLPRI | POLLRDBAND,                   /* POLL_PRI */
 445         POLLHUP | POLLERR                       /* POLL_HUP */
 446 };
 447
 448 static inline int sigio_perm(struct task_struct *p,
 449                              struct fown_struct *fown, int sig)
 450 {
 451         return (((fown->euid == 0) ||
 452                  (fown->euid == p->suid) || (fown->euid == p->uid) ||
 453                  (fown->uid == p->suid) || (fown->uid == p->uid)) &&
 454                 !security_file_send_sigiotask(p, fown, sig));
 455 }
 456
 457 static void send_sigio_to_task(struct task_struct *p,
 458                                struct fown_struct *fown,
 459                                int fd,
 460                                int reason)
 461 {
 462         if (!sigio_perm(p, fown, fown->signum))
 463                 return;
 464
 465         switch (fown->signum) {
 466                 siginfo_t si;
 467                 default:
 468                         /* Queue a rt signal with the appropriate fd as its
 469                            value.  We use SI_SIGIO as the source, not
 470                            SI_KERNEL, since kernel signals always get
 471                            delivered even if we can't queue.  Failure to
 472                            queue in this case _should_ be reported; we fall
 473                            back to SIGIO in that case. --sct */
 474                         si.si_signo = fown->signum;
 475                         si.si_errno = 0;
 476                         si.si_code  = reason;
 477                         /* Make sure we are called with one of the POLL_*
 478                            reasons, otherwise we could leak kernel stack into
 479                            userspace.  */
 480                         BUG_ON((reason & __SI_MASK) != __SI_POLL);
 481                         if (reason - POLL_IN >= NSIGPOLL)
 482                                 si.si_band  = ~0L;
 483                         else
 484                                 si.si_band = band_table[reason - POLL_IN];
 485                         si.si_fd    = fd;
 486                         if (!group_send_sig_info(fown->signum, &si, p))
 487                                 break;
 488                 /* fall-through: fall back on the old plain SIGIO signal */
 489                 case 0:
 490                         group_send_sig_info(SIGIO, SEND_SIG_PRIV, p);
 491         }
 492 }
 493
 494 void send_sigio(struct fown_struct *fown, int fd, int band)
 495 {
 496         struct task_struct *p;
 497         enum pid_type type;
 498         struct pid *pid;
 499
 500         read_lock(&fown->lock);
 501         type = fown->pid_type;
 502         pid = fown->pid;
 503         if (!pid)
 504                 goto out_unlock_fown;
 505
 506         read_lock(&tasklist_lock);
 507         do_each_pid_task(pid, type, p) {
 508                 send_sigio_to_task(p, fown, fd, band);
 509         } while_each_pid_task(pid, type, p);
 510         read_unlock(&tasklist_lock);
 511  out_unlock_fown:
 512         read_unlock(&fown->lock);
 513 }
 514
 515 static void send_sigurg_to_task(struct task_struct *p,
 516                                 struct fown_struct *fown)
 517 {
 518         if (sigio_perm(p, fown, SIGURG))
 519                 group_send_sig_info(SIGURG, SEND_SIG_PRIV, p);
 520 }
 521
 522 int send_sigurg(struct fown_struct *fown)
 523 {
 524         struct task_struct *p;
 525         enum pid_type type;
 526         struct pid *pid;
 527         int ret = 0;
 528
 529         read_lock(&fown->lock);
 530         type = fown->pid_type;
 531         pid = fown->pid;
 532         if (!pid)
 533                 goto out_unlock_fown;
 534
 535         ret = 1;
 536
 537         read_lock(&tasklist_lock);
 538         do_each_pid_task(pid, type, p) {
 539                 send_sigurg_to_task(p, fown);
 540         } while_each_pid_task(pid, type, p);
 541         read_unlock(&tasklist_lock);
 542  out_unlock_fown:
 543         read_unlock(&fown->lock);
 544         return ret;
 545 }
 546
 547 static DEFINE_RWLOCK(fasync_lock);
 548 static struct kmem_cache *fasync_cache __read_mostly;
 549
 550 /*
 551  * fasync_helper() is used by some character device drivers (mainly mice)
 552  * to set up the fasync queue. It returns negative on error, 0 if it did
 553  * no changes and positive if it added/deleted the entry.
 554  */
 555 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
 556 {
 557         struct fasync_struct *fa, **fp;
 558         struct fasync_struct *new = NULL;
 559         int result = 0;
 560
 561         if (on) {
 562                 new = kmem_cache_alloc(fasync_cache, GFP_KERNEL);
 563                 if (!new)
 564                         return -ENOMEM;
 565         }
 566         write_lock_irq(&fasync_lock);
 567         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 568                 if (fa->fa_file == filp) {
 569                         if(on) {
 570                                 fa->fa_fd = fd;
 571                                 kmem_cache_free(fasync_cache, new);
 572                         } else {
 573                                 *fp = fa->fa_next;
 574                                 kmem_cache_free(fasync_cache, fa);
 575                                 result = 1;
 576                         }
 577                         goto out;
 578                 }
 579         }
 580
 581         if (on) {
 582                 new->magic = FASYNC_MAGIC;
 583                 new->fa_file = filp;
 584                 new->fa_fd = fd;
 585                 new->fa_next = *fapp;
 586                 *fapp = new;
 587                 result = 1;
 588         }
 589 out:
 590         write_unlock_irq(&fasync_lock);
 591         return result;
 592 }
 593
 594 EXPORT_SYMBOL(fasync_helper);
 595
 596 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
 597 {
 598         while (fa) {
 599                 struct fown_struct * fown;
 600                 if (fa->magic != FASYNC_MAGIC) {
 601                         printk(KERN_ERR "kill_fasync: bad magic number in "
 602                                "fasync_struct!\n");
 603                         return;
 604                 }
 605                 fown = &fa->fa_file->f_owner;
 606                 /* Don't send SIGURG to processes which have not set a
 607                    queued signum: SIGURG has its own default signalling
 608                    mechanism. */
 609                 if (!(sig == SIGURG && fown->signum == 0))
 610                         send_sigio(fown, fa->fa_fd, band);
 611                 fa = fa->fa_next;
 612         }
 613 }
 614
 615 EXPORT_SYMBOL(__kill_fasync);
 616
 617 void kill_fasync(struct fasync_struct **fp, int sig, int band)
 618 {
 619         /* First a quick test without locking: usually
 620          * the list is empty.
 621          */
 622         if (*fp) {
 623                 read_lock(&fasync_lock);
 624                 /* reread *fp after obtaining the lock */
 625                 __kill_fasync(*fp, sig, band);
 626                 read_unlock(&fasync_lock);
 627         }
 628 }
 629 EXPORT_SYMBOL(kill_fasync);
 630
 631 static int __init fasync_init(void)
 632 {
 633         fasync_cache = kmem_cache_create("fasync_cache",
 634                 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
 635         return 0;
 636 }
 637
 638 module_init(fasync_init)