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
  22 #include <asm/poll.h>
  23 #include <asm/siginfo.h>
  24 #include <asm/uaccess.h>
  25
  26 void fastcall set_close_on_exec(unsigned int fd, int flag)
  27 {
  28         struct files_struct *files = current->files;
  29         struct fdtable *fdt;
  30         spin_lock(&files->file_lock);
  31         fdt = files_fdtable(files);
  32         if (flag)
  33                 FD_SET(fd, fdt->close_on_exec);
  34         else
  35                 FD_CLR(fd, fdt->close_on_exec);
  36         spin_unlock(&files->file_lock);
  37 }
  38
  39 static int get_close_on_exec(unsigned int fd)
  40 {
  41         struct files_struct *files = current->files;
  42         struct fdtable *fdt;
  43         int res;
  44         rcu_read_lock();
  45         fdt = files_fdtable(files);
  46         res = FD_ISSET(fd, fdt->close_on_exec);
  47         rcu_read_unlock();
  48         return res;
  49 }
  50
  51 /*
  52  * locate_fd finds a free file descriptor in the open_fds fdset,
  53  * expanding the fd arrays if necessary.  Must be called with the
  54  * file_lock held for write.
  55  */
  56
  57 static int locate_fd(struct files_struct *files,
  58                             struct file *file, unsigned int orig_start)
  59 {
  60         unsigned int newfd;
  61         unsigned int start;
  62         int error;
  63         struct fdtable *fdt;
  64
  65         error = -EINVAL;
  66         if (orig_start >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
  67                 goto out;
  68
  69 repeat:
  70         fdt = files_fdtable(files);
  71         /*
  72          * Someone might have closed fd's in the range
  73          * orig_start..fdt->next_fd
  74          */
  75         start = orig_start;
  76         if (start < fdt->next_fd)
  77                 start = fdt->next_fd;
  78
  79         newfd = start;
  80         if (start < fdt->max_fdset) {
  81                 newfd = find_next_zero_bit(fdt->open_fds->fds_bits,
  82                         fdt->max_fdset, start);
  83         }
  84
  85         error = -EMFILE;
  86         if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
  87                 goto out;
  88
  89         error = expand_files(files, newfd);
  90         if (error < 0)
  91                 goto out;
  92
  93         /*
  94          * If we needed to expand the fs array we
  95          * might have blocked - try again.
  96          */
  97         if (error)
  98                 goto repeat;
  99
 100         /*
 101          * We reacquired files_lock, so we are safe as long as
 102          * we reacquire the fdtable pointer and use it while holding
 103          * the lock, no one can free it during that time.
 104          */
 105         fdt = files_fdtable(files);
 106         if (start <= fdt->next_fd)
 107                 fdt->next_fd = newfd + 1;
 108
 109         error = newfd;
 110
 111 out:
 112         return error;
 113 }
 114
 115 static int dupfd(struct file *file, unsigned int start)
 116 {
 117         struct files_struct * files = current->files;
 118         struct fdtable *fdt;
 119         int fd;
 120
 121         spin_lock(&files->file_lock);
 122         fd = locate_fd(files, file, start);
 123         if (fd >= 0) {
 124                 /* locate_fd() may have expanded fdtable, load the ptr */
 125                 fdt = files_fdtable(files);
 126                 FD_SET(fd, fdt->open_fds);
 127                 FD_CLR(fd, fdt->close_on_exec);
 128                 spin_unlock(&files->file_lock);
 129                 fd_install(fd, file);
 130         } else {
 131                 spin_unlock(&files->file_lock);
 132                 fput(file);
 133         }
 134
 135         return fd;
 136 }
 137
 138 asmlinkage long sys_dup2(unsigned int oldfd, unsigned int newfd)
 139 {
 140         int err = -EBADF;
 141         struct file * file, *tofree;
 142         struct files_struct * files = current->files;
 143         struct fdtable *fdt;
 144
 145         spin_lock(&files->file_lock);
 146         if (!(file = fcheck(oldfd)))
 147                 goto out_unlock;
 148         err = newfd;
 149         if (newfd == oldfd)
 150                 goto out_unlock;
 151         err = -EBADF;
 152         if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
 153                 goto out_unlock;
 154         get_file(file);                 /* We are now finished with oldfd */
 155
 156         err = expand_files(files, newfd);
 157         if (err < 0)
 158                 goto out_fput;
 159
 160         /* To avoid races with open() and dup(), we will mark the fd as
 161          * in-use in the open-file bitmap throughout the entire dup2()
 162          * process.  This is quite safe: do_close() uses the fd array
 163          * entry, not the bitmap, to decide what work needs to be
 164          * done.  --sct */
 165         /* Doesn't work. open() might be there first. --AV */
 166
 167         /* Yes. It's a race. In user space. Nothing sane to do */
 168         err = -EBUSY;
 169         fdt = files_fdtable(files);
 170         tofree = fdt->fd[newfd];
 171         if (!tofree && FD_ISSET(newfd, fdt->open_fds))
 172                 goto out_fput;
 173
 174         rcu_assign_pointer(fdt->fd[newfd], file);
 175         FD_SET(newfd, fdt->open_fds);
 176         FD_CLR(newfd, fdt->close_on_exec);
 177         spin_unlock(&files->file_lock);
 178
 179         if (tofree)
 180                 filp_close(tofree, files);
 181         err = newfd;
 182 out:
 183         return err;
 184 out_unlock:
 185         spin_unlock(&files->file_lock);
 186         goto out;
 187
 188 out_fput:
 189         spin_unlock(&files->file_lock);
 190         fput(file);
 191         goto out;
 192 }
 193
 194 asmlinkage long sys_dup(unsigned int fildes)
 195 {
 196         int ret = -EBADF;
 197         struct file * file = fget(fildes);
 198
 199         if (file)
 200                 ret = dupfd(file, 0);
 201         return ret;
 202 }
 203
 204 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
 205
 206 static int setfl(int fd, struct file * filp, unsigned long arg)
 207 {
 208         struct inode * inode = filp->f_dentry->d_inode;
 209         int error = 0;
 210
 211         /*
 212          * O_APPEND cannot be cleared if the file is marked as append-only
 213          * and the file is open for write.
 214          */
 215         if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
 216                 return -EPERM;
 217
 218         /* O_NOATIME can only be set by the owner or superuser */
 219         if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
 220                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
 221                         return -EPERM;
 222
 223         /* required for strict SunOS emulation */
 224         if (O_NONBLOCK != O_NDELAY)
 225                if (arg & O_NDELAY)
 226                    arg |= O_NONBLOCK;
 227
 228         if (arg & O_DIRECT) {
 229                 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
 230                         !filp->f_mapping->a_ops->direct_IO)
 231                                 return -EINVAL;
 232         }
 233
 234         if (filp->f_op && filp->f_op->check_flags)
 235                 error = filp->f_op->check_flags(arg);
 236         if (error)
 237                 return error;
 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 EXPORT_SYMBOL(f_setown);
 279
 280 void f_delown(struct file *filp)
 281 {
 282         f_modown(filp, 0, 0, 0, 1);
 283 }
 284
 285 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
 286                 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(fd, 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 (!valid_signal(arg)) {
 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         return err;
 353 }
 354
 355 asmlinkage long sys_fcntl(unsigned int fd, unsigned int cmd, unsigned long arg)
 356 {
 357         struct file *filp;
 358         long err = -EBADF;
 359
 360         filp = fget(fd);
 361         if (!filp)
 362                 goto out;
 363
 364         err = security_file_fcntl(filp, cmd, arg);
 365         if (err) {
 366                 fput(filp);
 367                 return err;
 368         }
 369
 370         err = do_fcntl(fd, cmd, arg, filp);
 371
 372         fput(filp);
 373 out:
 374         return err;
 375 }
 376
 377 #if BITS_PER_LONG == 32
 378 asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
 379 {
 380         struct file * filp;
 381         long err;
 382
 383         err = -EBADF;
 384         filp = fget(fd);
 385         if (!filp)
 386                 goto out;
 387
 388         err = security_file_fcntl(filp, cmd, arg);
 389         if (err) {
 390                 fput(filp);
 391                 return err;
 392         }
 393         err = -EBADF;
 394
 395         switch (cmd) {
 396                 case F_GETLK64:
 397                         err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
 398                         break;
 399                 case F_SETLK64:
 400                 case F_SETLKW64:
 401                         err = fcntl_setlk64(fd, filp, cmd,
 402                                         (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, int sig)
 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                 !security_file_send_sigiotask(p, fown, sig));
 432 }
 433
 434 static void send_sigio_to_task(struct task_struct *p,
 435                                struct fown_struct *fown,
 436                                int fd,
 437                                int reason)
 438 {
 439         if (!sigio_perm(p, fown, fown->signum))
 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 (!group_send_sig_info(fown->signum, &si, p))
 465                                 break;
 466                 /* fall-through: fall back on the old plain SIGIO signal */
 467                 case 0:
 468                         group_send_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                 do_each_task_pid(-pid, PIDTYPE_PGID, p) {
 490                         send_sigio_to_task(p, fown, fd, band);
 491                 } while_each_task_pid(-pid, PIDTYPE_PGID, p);
 492         }
 493         read_unlock(&tasklist_lock);
 494  out_unlock_fown:
 495         read_unlock(&fown->lock);
 496 }
 497
 498 static void send_sigurg_to_task(struct task_struct *p,
 499                                 struct fown_struct *fown)
 500 {
 501         if (sigio_perm(p, fown, SIGURG))
 502                 group_send_sig_info(SIGURG, SEND_SIG_PRIV, p);
 503 }
 504
 505 int send_sigurg(struct fown_struct *fown)
 506 {
 507         struct task_struct *p;
 508         int pid, ret = 0;
 509
 510         read_lock(&fown->lock);
 511         pid = fown->pid;
 512         if (!pid)
 513                 goto out_unlock_fown;
 514
 515         ret = 1;
 516
 517         read_lock(&tasklist_lock);
 518         if (pid > 0) {
 519                 p = find_task_by_pid(pid);
 520                 if (p) {
 521                         send_sigurg_to_task(p, fown);
 522                 }
 523         } else {
 524                 do_each_task_pid(-pid, PIDTYPE_PGID, p) {
 525                         send_sigurg_to_task(p, fown);
 526                 } while_each_task_pid(-pid, PIDTYPE_PGID, p);
 527         }
 528         read_unlock(&tasklist_lock);
 529  out_unlock_fown:
 530         read_unlock(&fown->lock);
 531         return ret;
 532 }
 533
 534 static DEFINE_RWLOCK(fasync_lock);
 535 static kmem_cache_t *fasync_cache;
 536
 537 /*
 538  * fasync_helper() is used by some character device drivers (mainly mice)
 539  * to set up the fasync queue. It returns negative on error, 0 if it did
 540  * no changes and positive if it added/deleted the entry.
 541  */
 542 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
 543 {
 544         struct fasync_struct *fa, **fp;
 545         struct fasync_struct *new = NULL;
 546         int result = 0;
 547
 548         if (on) {
 549                 new = kmem_cache_alloc(fasync_cache, SLAB_KERNEL);
 550                 if (!new)
 551                         return -ENOMEM;
 552         }
 553         write_lock_irq(&fasync_lock);
 554         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 555                 if (fa->fa_file == filp) {
 556                         if(on) {
 557                                 fa->fa_fd = fd;
 558                                 kmem_cache_free(fasync_cache, new);
 559                         } else {
 560                                 *fp = fa->fa_next;
 561                                 kmem_cache_free(fasync_cache, fa);
 562                                 result = 1;
 563                         }
 564                         goto out;
 565                 }
 566         }
 567
 568         if (on) {
 569                 new->magic = FASYNC_MAGIC;
 570                 new->fa_file = filp;
 571                 new->fa_fd = fd;
 572                 new->fa_next = *fapp;
 573                 *fapp = new;
 574                 result = 1;
 575         }
 576 out:
 577         write_unlock_irq(&fasync_lock);
 578         return result;
 579 }
 580
 581 EXPORT_SYMBOL(fasync_helper);
 582
 583 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
 584 {
 585         while (fa) {
 586                 struct fown_struct * fown;
 587                 if (fa->magic != FASYNC_MAGIC) {
 588                         printk(KERN_ERR "kill_fasync: bad magic number in "
 589                                "fasync_struct!\n");
 590                         return;
 591                 }
 592                 fown = &fa->fa_file->f_owner;
 593                 /* Don't send SIGURG to processes which have not set a
 594                    queued signum: SIGURG has its own default signalling
 595                    mechanism. */
 596                 if (!(sig == SIGURG && fown->signum == 0))
 597                         send_sigio(fown, fa->fa_fd, band);
 598                 fa = fa->fa_next;
 599         }
 600 }
 601
 602 EXPORT_SYMBOL(__kill_fasync);
 603
 604 void kill_fasync(struct fasync_struct **fp, int sig, int band)
 605 {
 606         /* First a quick test without locking: usually
 607          * the list is empty.
 608          */
 609         if (*fp) {
 610                 read_lock(&fasync_lock);
 611                 /* reread *fp after obtaining the lock */
 612                 __kill_fasync(*fp, sig, band);
 613                 read_unlock(&fasync_lock);
 614         }
 615 }
 616 EXPORT_SYMBOL(kill_fasync);
 617
 618 static int __init fasync_init(void)
 619 {
 620         fasync_cache = kmem_cache_create("fasync_cache",
 621                 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL, NULL);
 622         return 0;
 623 }
 624
 625 module_init(fasync_init)