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