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
  16 #include <asm/poll.h>
  17 #include <asm/siginfo.h>
  18 #include <asm/uaccess.h>
  19
  20 extern int fcntl_setlease(unsigned int fd, struct file *filp, long arg);
  21 extern int fcntl_getlease(struct file *filp);
  22
  23 void set_close_on_exec(unsigned int fd, int flag)
  24 {
  25         struct files_struct *files = current->files;
  26         spin_lock(&files->file_lock);
  27         if (flag)
  28                 FD_SET(fd, files->close_on_exec);
  29         else
  30                 FD_CLR(fd, files->close_on_exec);
  31         spin_unlock(&files->file_lock);
  32 }
  33
  34 static inline int get_close_on_exec(unsigned int fd)
  35 {
  36         struct files_struct *files = current->files;
  37         int res;
  38         spin_lock(&files->file_lock);
  39         res = FD_ISSET(fd, files->close_on_exec);
  40         spin_unlock(&files->file_lock);
  41         return res;
  42 }
  43
  44
  45 /* Expand files.  Return <0 on error; 0 nothing done; 1 files expanded,
  46  * we may have blocked.
  47  *
  48  * Should be called with the files->file_lock spinlock held for write.
  49  */
  50 static int expand_files(struct files_struct *files, int nr)
  51 {
  52         int err, expand = 0;
  53 #ifdef FDSET_DEBUG
  54         printk (KERN_ERR "%s %d: nr = %d\n", __FUNCTION__, current->pid, nr);
  55 #endif
  56
  57         if (nr >= files->max_fdset) {
  58                 expand = 1;
  59                 if ((err = expand_fdset(files, nr)))
  60                         goto out;
  61         }
  62         if (nr >= files->max_fds) {
  63                 expand = 1;
  64                 if ((err = expand_fd_array(files, nr)))
  65                         goto out;
  66         }
  67         err = expand;
  68  out:
  69 #ifdef FDSET_DEBUG
  70         if (err)
  71                 printk (KERN_ERR "%s %d: return %d\n", __FUNCTION__, current->pid, err);
  72 #endif
  73         return err;
  74 }
  75
  76 /*
  77  * locate_fd finds a free file descriptor in the open_fds fdset,
  78  * expanding the fd arrays if necessary.  Must be called with the
  79  * file_lock held for write.
  80  */
  81
  82 static int locate_fd(struct files_struct *files,
  83                             struct file *file, int orig_start)
  84 {
  85         unsigned int newfd;
  86         int error;
  87         int start;
  88
  89         error = -EINVAL;
  90         if (orig_start >= current->rlim[RLIMIT_NOFILE].rlim_cur)
  91                 goto out;
  92
  93 repeat:
  94         /*
  95          * Someone might have closed fd's in the range
  96          * orig_start..files->next_fd
  97          */
  98         start = orig_start;
  99         if (start < files->next_fd)
 100                 start = files->next_fd;
 101
 102         newfd = start;
 103         if (start < files->max_fdset) {
 104                 newfd = find_next_zero_bit(files->open_fds->fds_bits,
 105                         files->max_fdset, start);
 106         }
 107
 108         error = -EMFILE;
 109         if (newfd >= current->rlim[RLIMIT_NOFILE].rlim_cur)
 110                 goto out;
 111
 112         error = expand_files(files, newfd);
 113         if (error < 0)
 114                 goto out;
 115
 116         /*
 117          * If we needed to expand the fs array we
 118          * might have blocked - try again.
 119          */
 120         if (error)
 121                 goto repeat;
 122
 123         if (start <= files->next_fd)
 124                 files->next_fd = newfd + 1;
 125
 126         error = newfd;
 127
 128 out:
 129         return error;
 130 }
 131
 132 static int dupfd(struct file *file, int start)
 133 {
 134         struct files_struct * files = current->files;
 135         int fd;
 136
 137         spin_lock(&files->file_lock);
 138         fd = locate_fd(files, file, start);
 139         if (fd >= 0) {
 140                 FD_SET(fd, files->open_fds);
 141                 FD_CLR(fd, files->close_on_exec);
 142                 spin_unlock(&files->file_lock);
 143                 fd_install(fd, file);
 144         } else {
 145                 spin_unlock(&files->file_lock);
 146                 fput(file);
 147         }
 148
 149         return fd;
 150 }
 151
 152 asmlinkage long sys_dup2(unsigned int oldfd, unsigned int newfd)
 153 {
 154         int err = -EBADF;
 155         struct file * file, *tofree;
 156         struct files_struct * files = current->files;
 157
 158         spin_lock(&files->file_lock);
 159         if (!(file = fcheck(oldfd)))
 160                 goto out_unlock;
 161         err = newfd;
 162         if (newfd == oldfd)
 163                 goto out_unlock;
 164         err = -EBADF;
 165         if (newfd >= current->rlim[RLIMIT_NOFILE].rlim_cur)
 166                 goto out_unlock;
 167         get_file(file);                 /* We are now finished with oldfd */
 168
 169         err = expand_files(files, newfd);
 170         if (err < 0)
 171                 goto out_fput;
 172
 173         /* To avoid races with open() and dup(), we will mark the fd as
 174          * in-use in the open-file bitmap throughout the entire dup2()
 175          * process.  This is quite safe: do_close() uses the fd array
 176          * entry, not the bitmap, to decide what work needs to be
 177          * done.  --sct */
 178         /* Doesn't work. open() might be there first. --AV */
 179
 180         /* Yes. It's a race. In user space. Nothing sane to do */
 181         err = -EBUSY;
 182         tofree = files->fd[newfd];
 183         if (!tofree && FD_ISSET(newfd, files->open_fds))
 184                 goto out_fput;
 185
 186         files->fd[newfd] = file;
 187         FD_SET(newfd, files->open_fds);
 188         FD_CLR(newfd, files->close_on_exec);
 189         spin_unlock(&files->file_lock);
 190
 191         if (tofree)
 192                 filp_close(tofree, files);
 193         err = newfd;
 194 out:
 195         return err;
 196 out_unlock:
 197         spin_unlock(&files->file_lock);
 198         goto out;
 199
 200 out_fput:
 201         spin_unlock(&files->file_lock);
 202         fput(file);
 203         goto out;
 204 }
 205
 206 asmlinkage long sys_dup(unsigned int fildes)
 207 {
 208         int ret = -EBADF;
 209         struct file * file = fget(fildes);
 210
 211         if (file)
 212                 ret = dupfd(file, 0);
 213         return ret;
 214 }
 215
 216 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT)
 217
 218 static int setfl(int fd, struct file * filp, unsigned long arg)
 219 {
 220         struct inode * inode = filp->f_dentry->d_inode;
 221         int error = 0;
 222
 223         /* O_APPEND cannot be cleared if the file is marked as append-only */
 224         if (!(arg & O_APPEND) && IS_APPEND(inode))
 225                 return -EPERM;
 226
 227         /* required for strict SunOS emulation */
 228         if (O_NONBLOCK != O_NDELAY)
 229                if (arg & O_NDELAY)
 230                    arg |= O_NONBLOCK;
 231
 232         if (arg & O_DIRECT) {
 233                 if (!inode->i_mapping || !inode->i_mapping->a_ops ||
 234                         !inode->i_mapping->a_ops->direct_IO)
 235                                 return -EINVAL;
 236         }
 237
 238         lock_kernel();
 239         if ((arg ^ filp->f_flags) & FASYNC) {
 240                 if (filp->f_op && filp->f_op->fasync) {
 241                         error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
 242                         if (error < 0)
 243                                 goto out;
 244                 }
 245         }
 246
 247         filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
 248  out:
 249         unlock_kernel();
 250         return error;
 251 }
 252
 253 static void f_modown(struct file *filp, unsigned long pid,
 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                 filp->f_owner.pid = pid;
 259                 filp->f_owner.uid = uid;
 260                 filp->f_owner.euid = euid;
 261         }
 262         write_unlock_irq(&filp->f_owner.lock);
 263 }
 264
 265 int f_setown(struct file *filp, unsigned long arg, int force)
 266 {
 267         int err;
 268
 269         err = security_file_set_fowner(filp);
 270         if (err)
 271                 return err;
 272
 273         f_modown(filp, arg, current->uid, current->euid, force);
 274         return 0;
 275 }
 276
 277 void f_delown(struct file *filp)
 278 {
 279         f_modown(filp, 0, 0, 0, 1);
 280 }
 281
 282 static long do_fcntl(unsigned int fd, unsigned int cmd,
 283                      unsigned long arg, struct file * filp)
 284 {
 285         long err = -EINVAL;
 286
 287         switch (cmd) {
 288                 case F_DUPFD:
 289                         if (arg < NR_OPEN) {
 290                                 get_file(filp);
 291                                 err = dupfd(filp, arg);
 292                         }
 293                         break;
 294                 case F_GETFD:
 295                         err = get_close_on_exec(fd);
 296                         break;
 297                 case F_SETFD:
 298                         err = 0;
 299                         set_close_on_exec(fd, arg&1);
 300                         break;
 301                 case F_GETFL:
 302                         err = filp->f_flags;
 303                         break;
 304                 case F_SETFL:
 305                         err = setfl(fd, filp, arg);
 306                         break;
 307                 case F_GETLK:
 308                         err = fcntl_getlk(filp, (struct flock __user *) arg);
 309                         break;
 310                 case F_SETLK:
 311                 case F_SETLKW:
 312                         err = fcntl_setlk(filp, cmd, (struct flock __user *) arg);
 313                         break;
 314                 case F_GETOWN:
 315                         /*
 316                          * XXX If f_owner is a process group, the
 317                          * negative return value will get converted
 318                          * into an error.  Oops.  If we keep the
 319                          * current syscall conventions, the only way
 320                          * to fix this will be in libc.
 321                          */
 322                         err = filp->f_owner.pid;
 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);