fs/fcntl.c

   1 /*
   2  *  linux/fs/fcntl.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/init.h>
   8 #include <linux/mm.h>
   9 #include <linux/fs.h>
  10 #include <linux/file.h>
  11 #include <linux/dnotify.h>
  12 #include <linux/smp_lock.h>
  13 #include <linux/slab.h>
  14 #include <linux/module.h>
  15 #include <linux/security.h>
  16 #include <linux/ptrace.h>
  17
  18 #include <asm/poll.h>
  19 #include <asm/siginfo.h>
  20 #include <asm/uaccess.h>
  21
  22 void fastcall set_close_on_exec(unsigned int fd, int flag)
  23 {
  24         struct files_struct *files = current->files;
  25         spin_lock(&files->file_lock);
  26         if (flag)
  27                 FD_SET(fd, files->close_on_exec);
  28         else
  29                 FD_CLR(fd, files->close_on_exec);
  30         spin_unlock(&files->file_lock);
  31 }
  32
  33 static inline int get_close_on_exec(unsigned int fd)
  34 {
  35         struct files_struct *files = current->files;
  36         int res;
  37         spin_lock(&files->file_lock);
  38         res = FD_ISSET(fd, files->close_on_exec);
  39         spin_unlock(&files->file_lock);
  40         return res;
  41 }
  42
  43
  44 /* Expand files.  Return <0 on error; 0 nothing done; 1 files expanded,
  45  * we may have blocked.
  46  *
  47  * Should be called with the files->file_lock spinlock held for write.
  48  */
  49 static int expand_files(struct files_struct *files, int nr)
  50 {
  51         int err, expand = 0;
  52 #ifdef FDSET_DEBUG
  53         printk (KERN_ERR "%s %d: nr = %d\n", __FUNCTION__, current->pid, nr);
  54 #endif
  55
  56         if (nr >= files->max_fdset) {
  57                 expand = 1;
  58                 if ((err = expand_fdset(files, nr)))
  59                         goto out;
  60         }
  61         if (nr >= files->max_fds) {
  62                 expand = 1;
  63                 if ((err = expand_fd_array(files, nr)))
  64                         goto out;
  65         }
  66         err = expand;
  67  out:
  68 #ifdef FDSET_DEBUG
  69         if (err)
  70                 printk (KERN_ERR "%s %d: return %d\n", __FUNCTION__, current->pid, err);
  71 #endif
  72         return err;
  73 }
  74
  75 /*
  76  * locate_fd finds a free file descriptor in the open_fds fdset,
  77  * expanding the fd arrays if necessary.  Must be called with the
  78  * file_lock held for write.
  79  */
  80
  81 static int locate_fd(struct files_struct *files,
  82                             struct file *file, unsigned int orig_start)
  83 {
  84         unsigned int newfd;
  85         unsigned int start;
  86         int error;
  87
  88         error = -EINVAL;
  89         if (orig_start >= current->rlim[RLIMIT_NOFILE].rlim_cur)
  90                 goto out;
  91
  92 repeat:
  93         /*
  94          * Someone might have closed fd's in the range
  95          * orig_start..files->next_fd
  96          */
  97         start = orig_start;
  98         if (start < files->next_fd)
  99                 start = files->next_fd;
 100
 101         newfd = start;
 102         if (start < files->max_fdset) {
 103                 newfd = find_next_zero_bit(files->open_fds->fds_bits,
 104                         files->max_fdset, start);
 105         }
 106
 107         error = -EMFILE;
 108         if (newfd >= current->rlim[RLIMIT_NOFILE].rlim_cur)
 109                 goto out;
 110
 111         error = expand_files(files, newfd);
 112         if (error < 0)
 113                 goto out;
 114
 115         /*
 116          * If we needed to expand the fs array we
 117          * might have blocked - try again.
 118          */
 119         if (error)
 120                 goto repeat;
 121
 122         if (start <= files->next_fd)
 123                 files->next_fd = newfd + 1;
 124
 125         error = newfd;
 126
 127 out:
 128         return error;
 129 }
 130
 131 static int dupfd(struct file *file, unsigned int start)
 132 {
 133         struct files_struct * files = current->files;
 134         int fd;
 135
 136         spin_lock(&files->file_lock);
 137         fd = locate_fd(files, file, start);
 138         if (fd >= 0) {
 139                 FD_SET(fd, files->open_fds);
 140                 FD_CLR(fd, files->close_on_exec);
 141                 spin_unlock(&files->file_lock);
 142                 fd_install(fd, file);
 143         } else {
 144                 spin_unlock(&files->file_lock);
 145                 fput(file);
 146         }
 147
 148         return fd;
 149 }
 150
 151 asmlinkage long sys_dup2(unsigned int oldfd, unsigned int newfd)
 152 {
 153         int err = -EBADF;
 154         struct file * file, *tofree;
 155         struct files_struct * files = current->files;
 156
 157         spin_lock(&files->file_lock);
 158         if (!(file = fcheck(oldfd)))
 159                 goto out_unlock;
 160         err = newfd;
 161         if (newfd == oldfd)
 162                 goto out_unlock;
 163         err = -EBADF;
 164         if (newfd >= current->rlim[RLIMIT_NOFILE].rlim_cur)
 165                 goto out_unlock;
 166         get_file(file);                 /* We are now finished with oldfd */
 167
 168         err = expand_files(files, newfd);
 169         if (err < 0)
 170                 goto out_fput;
 171
 172         /* To avoid races with open() and dup(), we will mark the fd as
 173          * in-use in the open-file bitmap throughout the entire dup2()
 174          * process.  This is quite safe: do_close() uses the fd array
 175          * entry, not the bitmap, to decide what work needs to be
 176          * done.  --sct */
 177         /* Doesn't work. open() might be there first. --AV */
 178
 179         /* Yes. It's a race. In user space. Nothing sane to do */
 180         err = -EBUSY;
 181         tofree = files->fd[newfd];
 182         if (!tofree && FD_ISSET(newfd, files->open_fds))
 183                 goto out_fput;
 184
 185         files->fd[newfd] = file;
 186         FD_SET(newfd, files->open_fds);
 187         FD_CLR(newfd, files->close_on_exec);
 188         spin_unlock(&files->file_lock);
 189
 190         if (tofree)
 191                 filp_close(tofree, files);
 192         err = newfd;
 193 out:
 194         return err;
 195 out_unlock:
 196         spin_unlock(&files->file_lock);
 197         goto out;
 198
 199 out_fput:
 200         spin_unlock(&files->file_lock);
 201         fput(file);
 202         goto out;
 203 }
 204
 205 asmlinkage long sys_dup(unsigned int fildes)
 206 {
 207         int ret = -EBADF;
 208         struct file * file = fget(fildes);
 209
 210         if (file)
 211                 ret = dupfd(file, 0);
 212         return ret;
 213 }
 214
 215 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT)
 216
 217 static int setfl(int fd, struct file * filp, unsigned long arg)
 218 {
 219         struct inode * inode = filp->f_dentry->d_inode;
 220         int error = 0;
 221
 222         /* O_APPEND cannot be cleared if the file is marked as append-only */
 223         if (!(arg & O_APPEND) && IS_APPEND(inode))
 224                 return -EPERM;
 225
 226         /* required for strict SunOS emulation */
 227         if (O_NONBLOCK != O_NDELAY)
 228                if (arg & O_NDELAY)
 229                    arg |= O_NONBLOCK;
 230
 231         if (arg & O_DIRECT) {
 232                 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
 233                         !filp->f_mapping->a_ops->direct_IO)
 234                                 return -EINVAL;
 235         }
 236
 237         lock_kernel();
 238         if ((arg ^ filp->f_flags) & FASYNC) {
 239                 if (filp->f_op && filp->f_op->fasync) {
 240                         error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
 241                         if (error < 0)
 242                                 goto out;
 243                 }
 244         }
 245
 246         filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
 247  out:
 248         unlock_kernel();
 249         return error;
 250 }
 251
 252 static void f_modown(struct file *filp, unsigned long pid,
 253                      uid_t uid, uid_t euid, int force)
 254 {
 255         write_lock_irq(&filp->f_owner.lock);
 256         if (force || !filp->f_owner.pid) {
 257                 filp->f_owner.pid = pid;
 258                 filp->f_owner.uid = uid;
 259                 filp->f_owner.euid = euid;
 260         }
 261         write_unlock_irq(&filp->f_owner.lock);
 262 }
 263
 264 int f_setown(struct file *filp, unsigned long arg, int force)
 265 {
 266         int err;
 267
 268         err = security_file_set_fowner(filp);
 269         if (err)
 270                 return err;
 271
 272         f_modown(filp, arg, current->uid, current->euid, force);
 273         return 0;
 274 }
 275
 276 EXPORT_SYMBOL(f_setown);
 277
 278 void f_delown(struct file *filp)
 279 {
 280         f_modown(filp, 0, 0, 0, 1);
 281 }
 282
 283 EXPORT_SYMBOL(f_delown);
 284
 285 static long do_fcntl(unsigned int fd, unsigned int cmd,
 286                      unsigned long arg, struct file * filp)
 287 {
 288         long err = -EINVAL;
 289
 290         switch (cmd) {
 291                 case F_DUPFD:
 292                         get_file(filp);
 293                         err = dupfd(filp, arg);
 294                         break;
 295                 case F_GETFD:
 296                         err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
 297                         break;
 298                 case F_SETFD:
 299                         err = 0;
 300                         set_close_on_exec(fd, arg & FD_CLOEXEC);
 301                         break;
 302                 case F_GETFL:
 303                         err = filp->f_flags;
 304                         break;
 305                 case F_SETFL:
 306                         err = setfl(fd, filp, arg);
 307                         break;
 308                 case F_GETLK:
 309                         err = fcntl_getlk(filp, (struct flock __user *) arg);
 310                         break;
 311                 case F_SETLK:
 312                 case F_SETLKW:
 313                         err = fcntl_setlk(filp, cmd, (struct flock __user *) arg);
 314                         break;
 315                 case F_GETOWN:
 316                         /*
 317                          * XXX If f_owner is a process group, the
 318                          * negative return value will get converted
 319                          * into an error.  Oops.  If we keep the
 320                          * current syscall conventions, the only way
 321                          * to fix this will be in libc.
 322                          */
 323                         err = filp->f_owner.pid;
 324                         force_successful_syscall_return();
 325                         break;
 326                 case F_SETOWN:
 327                         err = f_setown(filp, arg, 1);
 328                         break;
 329                 case F_GETSIG:
 330                         err = filp->f_owner.signum;
 331                         break;
 332                 case F_SETSIG:
 333                         /* arg == 0 restores default behaviour. */
 334                         if (arg < 0 || arg > _NSIG) {
 335                                 break;
 336                         }
 337                         err = 0;
 338                         filp->f_owner.signum = arg;
 339                         break;
 340                 case F_GETLEASE:
 341                         err = fcntl_getlease(filp);
 342                         break;
 343                 case F_SETLEASE:
 344                         err = fcntl_setlease(fd, filp, arg);
 345                         break;
 346                 case F_NOTIFY:
 347                         err = fcntl_dirnotify(fd, filp, arg);
 348                         break;
 349                 default:
 350                         break;
 351         }
 352
 353         return err;
 354 }
 355
 356 asmlinkage long sys_fcntl(unsigned int fd, unsigned int cmd, unsigned long arg)
 357 {
 358         struct file * filp;
 359         long err = -EBADF;
 360
 361         filp = fget(fd);
 362         if (!filp)
 363                 goto out;
 364
 365         err = security_file_fcntl(filp, cmd, arg);
 366         if (err) {
 367                 fput(filp);
 368                 return err;
 369         }
 370
 371         err = do_fcntl(fd, cmd, arg, filp);
 372
 373         fput(filp);
 374 out:
 375         return err;
 376 }
 377
 378 #if BITS_PER_LONG == 32
 379 asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
 380 {
 381         struct file * filp;
 382         long err;
 383
 384         err = -EBADF;
 385         filp = fget(fd);
 386         if (!filp)
 387                 goto out;
 388
 389         err = security_file_fcntl(filp, cmd, arg);
 390         if (err) {
 391                 fput(filp);
 392                 return err;
 393         }
 394         err = -EBADF;
 395
 396         switch (cmd) {
 397                 case F_GETLK64:
 398                         err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
 399                         break;
 400                 case F_SETLK64:
 401                 case F_SETLKW64:
 402                         err = fcntl_setlk64(filp, cmd, (struct flock64 __user *) arg);
 403                         break;
 404                 default:
 405                         err = do_fcntl(fd, cmd, arg, filp);
 406                         break;
 407         }
 408         fput(filp);
 409 out:
 410         return err;
 411 }
 412 #endif
 413
 414 /* Table to convert sigio signal codes into poll band bitmaps */
 415
 416 static long band_table[NSIGPOLL] = {
 417         POLLIN | POLLRDNORM,                    /* POLL_IN */
 418         POLLOUT | POLLWRNORM | POLLWRBAND,      /* POLL_OUT */
 419         POLLIN | POLLRDNORM | POLLMSG,          /* POLL_MSG */
 420         POLLERR,                                /* POLL_ERR */
 421         POLLPRI | POLLRDBAND,                   /* POLL_PRI */
 422         POLLHUP | POLLERR                       /* POLL_HUP */
 423 };
 424
 425 static inline int sigio_perm(struct task_struct *p,
 426                              struct fown_struct *fown)
 427 {
 428         return ((fown->euid == 0) ||
 429                 (fown->euid == p->suid) || (fown->euid == p->uid) ||
 430                 (fown->uid == p->suid) || (fown->uid == p->uid));
 431 }
 432
 433 static void send_sigio_to_task(struct task_struct *p,
 434                                struct fown_struct *fown,
 435                                int fd,
 436                                int reason)
 437 {
 438         if (!sigio_perm(p, fown))
 439                 return;
 440
 441         if (security_file_send_sigiotask(p, fown, fd, reason))
 442                 return;
 443
 444         switch (fown->signum) {
 445                 siginfo_t si;
 446                 default:
 447                         /* Queue a rt signal with the appropriate fd as its
 448                            value.  We use SI_SIGIO as the source, not
 449                            SI_KERNEL, since kernel signals always get
 450                            delivered even if we can't queue.  Failure to
 451                            queue in this case _should_ be reported; we fall
 452                            back to SIGIO in that case. --sct */
 453                         si.si_signo = fown->signum;
 454                         si.si_errno = 0;
 455                         si.si_code  = reason;
 456                         /* Make sure we are called with one of the POLL_*
 457                            reasons, otherwise we could leak kernel stack into
 458                            userspace.  */
 459                         if ((reason & __SI_MASK) != __SI_POLL)
 460                                 BUG();
 461                         if (reason - POLL_IN >= NSIGPOLL)
 462                                 si.si_band  = ~0L;
 463                         else
 464                                 si.si_band = band_table[reason - POLL_IN];
 465                         si.si_fd    = fd;
 466                         if (!send_sig_info(fown->signum, &si, p))
 467                                 break;
 468                 /* fall-through: fall back on the old plain SIGIO signal */
 469                 case 0:
 470                         send_group_sig_info(SIGIO, SEND_SIG_PRIV, p);
 471         }
 472 }
 473
 474 void send_sigio(struct fown_struct *fown, int fd, int band)
 475 {
 476         struct task_struct *p;
 477         int pid;
 478
 479         read_lock(&fown->lock);
 480         pid = fown->pid;
 481         if (!pid)
 482                 goto out_unlock_fown;
 483
 484         read_lock(&tasklist_lock);
 485         if (pid > 0) {
 486                 p = find_task_by_pid(pid);
 487                 if (p) {
 488                         send_sigio_to_task(p, fown, fd, band);
 489                 }
 490         } else {
 491                 struct list_head *l;
 492                 struct pid *pidptr;
 493                 for_each_task_pid(-pid, PIDTYPE_PGID, p, l, pidptr) {
 494                         send_sigio_to_task(p, fown, fd, band);
 495                 }
 496         }
 497         read_unlock(&tasklist_lock);
 498  out_unlock_fown:
 499         read_unlock(&fown->lock);
 500 }
 501
 502 static void send_sigurg_to_task(struct task_struct *p,
 503                                 struct fown_struct *fown)
 504 {
 505         if (sigio_perm(p, fown))
 506                 send_group_sig_info(SIGURG, SEND_SIG_PRIV, p);
 507 }
 508
 509 int send_sigurg(struct fown_struct *fown)
 510 {
 511         struct task_struct *p;
 512         int pid, ret = 0;
 513
 514         read_lock(&fown->lock);
 515         pid = fown->pid;
 516         if (!pid)
 517                 goto out_unlock_fown;
 518
 519         ret = 1;
 520
 521         read_lock(&tasklist_lock);
 522         if (pid > 0) {
 523                 p = find_task_by_pid(pid);
 524                 if (p) {
 525                         send_sigurg_to_task(p, fown);
 526                 }
 527         } else {
 528                 struct list_head *l;
 529                 struct pid *pidptr;
 530                 for_each_task_pid(-pid, PIDTYPE_PGID, p, l, pidptr) {
 531                         send_sigurg_to_task(p, fown);
 532                 }
 533         }
 534         read_unlock(&tasklist_lock);
 535  out_unlock_fown:
 536         read_unlock(&fown->lock);
 537         return ret;
 538 }
 539
 540 static rwlock_t fasync_lock = RW_LOCK_UNLOCKED;
 541 static kmem_cache_t *fasync_cache;
 542
 543 /*
 544  * fasync_helper() is used by some character device drivers (mainly mice)
 545  * to set up the fasync queue. It returns negative on error, 0 if it did
 546  * no changes and positive if it added/deleted the entry.
 547  */
 548 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
 549 {
 550         struct fasync_struct *fa, **fp;
 551         struct fasync_struct *new = NULL;
 552         int result = 0;
 553
 554         if (on) {
 555                 new = kmem_cache_alloc(fasync_cache, SLAB_KERNEL);
 556                 if (!new)
 557                         return -ENOMEM;
 558         }
 559         write_lock_irq(&fasync_lock);
 560         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 561                 if (fa->fa_file == filp) {
 562                         if(on) {
 563                                 fa->fa_fd = fd;
 564                                 kmem_cache_free(fasync_cache, new);
 565                         } else {
 566                                 *fp = fa->fa_next;
 567                                 kmem_cache_free(fasync_cache, fa);
 568                                 result = 1;
 569                         }
 570                         goto out;
 571                 }
 572         }
 573
 574         if (on) {
 575                 new->magic = FASYNC_MAGIC;
 576                 new->fa_file = filp;
 577                 new->fa_fd = fd;
 578                 new->fa_next = *fapp;
 579                 *fapp = new;
 580                 result = 1;
 581         }
 582 out:
 583         write_unlock_irq(&fasync_lock);
 584         return result;
 585 }
 586
 587 EXPORT_SYMBOL(fasync_helper);
 588
 589 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
 590 {
 591         while (fa) {
 592                 struct fown_struct * fown;
 593                 if (fa->magic != FASYNC_MAGIC) {
 594                         printk(KERN_ERR "kill_fasync: bad magic number in "
 595                                "fasync_struct!\n");
 596                         return;
 597                 }
 598                 fown = &fa->fa_file->f_owner;
 599                 /* Don't send SIGURG to processes which have not set a
 600                    queued signum: SIGURG has its own default signalling
 601                    mechanism. */
 602                 if (!(sig == SIGURG && fown->signum == 0))
 603                         send_sigio(fown, fa->fa_fd, band);
 604                 fa = fa->fa_next;
 605         }
 606 }
 607
 608 EXPORT_SYMBOL(__kill_fasync);
 609
 610 void kill_fasync(struct fasync_struct **fp, int sig, int band)
 611 {
 612         /* First a quick test without locking: usually
 613          * the list is empty.
 614          */
 615         if (*fp) {
 616                 read_lock(&fasync_lock);
 617                 /* reread *fp after obtaining the lock */
 618                 __kill_fasync(*fp, sig, band);
 619                 read_unlock(&fasync_lock);
 620         }
 621 }
 622
 623 EXPORT_SYMBOL(kill_fasync);
 624
 625 static int __init fasync_init(void)
 626 {
 627         fasync_cache = kmem_cache_create("fasync_cache",
 628                 sizeof(struct fasync_struct), 0, 0, NULL, NULL);
 629         if (!fasync_cache)
 630                 panic("cannot create fasync slab cache");
 631         return 0;
 632 }
 633
 634 module_init(fasync_init)