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 < files->next_fd)
  77                 start = files->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         if (start <= files->next_fd)
 106                 files->next_fd = newfd + 1;
 107
 108         error = newfd;
 109
 110 out:
 111         return error;
 112 }
 113
 114 static int dupfd(struct file *file, unsigned int start)
 115 {
 116         struct files_struct * files = current->files;
 117         struct fdtable *fdt;
 118         int fd;
 119
 120         spin_lock(&files->file_lock);
 121         fd = locate_fd(files, file, start);
 122         if (fd >= 0) {
 123                 /* locate_fd() may have expanded fdtable, load the ptr */
 124                 fdt = files_fdtable(files);
 125                 FD_SET(fd, fdt->open_fds);
 126                 FD_CLR(fd, fdt->close_on_exec);
 127                 spin_unlock(&files->file_lock);
 128                 fd_install(fd, file);
 129         } else {
 130                 spin_unlock(&files->file_lock);
 131                 fput(file);
 132         }
 133
 134         return fd;
 135 }
 136
 137 asmlinkage long sys_dup2(unsigned int oldfd, unsigned int newfd)
 138 {
 139         int err = -EBADF;
 140         struct file * file, *tofree;
 141         struct files_struct * files = current->files;
 142         struct fdtable *fdt;
 143
 144         spin_lock(&files->file_lock);
 145         if (!(file = fcheck(oldfd)))
 146                 goto out_unlock;
 147         err = newfd;
 148         if (newfd == oldfd)
 149                 goto out_unlock;
 150         err = -EBADF;
 151         if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
 152                 goto out_unlock;
 153         get_file(file);                 /* We are now finished with oldfd */
 154
 155         err = expand_files(files, newfd);
 156         if (err < 0)
 157                 goto out_fput;
 158
 159         /* To avoid races with open() and dup(), we will mark the fd as
 160          * in-use in the open-file bitmap throughout the entire dup2()
 161          * process.  This is quite safe: do_close() uses the fd array
 162          * entry, not the bitmap, to decide what work needs to be
 163          * done.  --sct */
 164         /* Doesn't work. open() might be there first. --AV */
 165
 166         /* Yes. It's a race. In user space. Nothing sane to do */
 167         err = -EBUSY;
 168         fdt = files_fdtable(files);
 169         tofree = fdt->fd[newfd];
 170         if (!tofree && FD_ISSET(newfd, fdt->open_fds))
 171                 goto out_fput;
 172
 173         rcu_assign_pointer(fdt->fd[newfd], file);
 174         FD_SET(newfd, fdt->open_fds);
 175         FD_CLR(newfd, fdt->close_on_exec);
 176         spin_unlock(&files->file_lock);
 177
 178         if (tofree)
 179                 filp_close(tofree, files);
 180         err = newfd;
 181 out:
 182         return err;
 183 out_unlock:
 184         spin_unlock(&files->file_lock);
 185         goto out;
 186
 187 out_fput:
 188         spin_unlock(&files->file_lock);
 189         fput(file);
 190         goto out;
 191 }
 192
 193 asmlinkage long sys_dup(unsigned int fildes)
 194 {
 195         int ret = -EBADF;
 196         struct file * file = fget(fildes);
 197
 198         if (file)
 199                 ret = dupfd(file, 0);
 200         return ret;
 201 }
 202
 203 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
 204
 205 static int setfl(int fd, struct file * filp, unsigned long arg)
 206 {
 207         struct inode * inode = filp->f_dentry->d_inode;
 208         int error = 0;
 209
 210         /*
 211          * O_APPEND cannot be cleared if the file is marked as append-only
 212          * and the file is open for write.
 213          */
 214         if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
 215                 return -EPERM;
 216
 217         /* O_NOATIME can only be set by the owner or superuser */
 218         if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
 219                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
 220                         return -EPERM;
 221
 222         /* required for strict SunOS emulation */
 223         if (O_NONBLOCK != O_NDELAY)
 224                if (arg & O_NDELAY)
 225                    arg |= O_NONBLOCK;
 226
 227         if (arg & O_DIRECT) {
 228                 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
 229                         !filp->f_mapping->a_ops->direct_IO)
 230                                 return -EINVAL;
 231         }
 232
 233         if (filp->f_op && filp->f_op->check_flags)
 234                 error = filp->f_op->check_flags(arg);
 235         if (error)
 236                 return error;
 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, struct pid *pid, enum pid_type type,
 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                 put_pid(filp->f_owner.pid);
 259                 filp->f_owner.pid = get_pid(pid);
 260                 filp->f_owner.pid_type = type;
 261                 filp->f_owner.uid = uid;
 262                 filp->f_owner.euid = euid;
 263         }
 264         write_unlock_irq(&filp->f_owner.lock);
 265 }
 266
 267 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
 268                 int force)
 269 {
 270         int err;
 271
 272         err = security_file_set_fowner(filp);
 273         if (err)
 274                 return err;
 275
 276         f_modown(filp, pid, type, current->uid, current->euid, force);
 277         return 0;
 278 }
 279 EXPORT_SYMBOL(__f_setown);
 280
 281 int f_setown(struct file *filp, unsigned long arg, int force)
 282 {
 283         enum pid_type type;
 284         struct pid *pid;
 285         int who = arg;
 286         int result;
 287         type = PIDTYPE_PID;
 288         if (who < 0) {
 289                 type = PIDTYPE_PGID;
 290                 who = -who;
 291         }
 292         rcu_read_lock();
 293         pid = find_pid(who);
 294         result = __f_setown(filp, pid, type, force);
 295         rcu_read_unlock();
 296         return result;
 297 }
 298 EXPORT_SYMBOL(f_setown);
 299
 300 void f_delown(struct file *filp)
 301 {
 302         f_modown(filp, NULL, PIDTYPE_PID, 0, 0, 1);
 303 }
 304
 305 pid_t f_getown(struct file *filp)
 306 {
 307         pid_t pid;
 308         read_lock(&filp->f_owner.lock);
 309         pid = pid_nr(filp->f_owner.pid);
 310         if (filp->f_owner.pid_type == PIDTYPE_PGID)
 311                 pid = -pid;
 312         read_unlock(&filp->f_owner.lock);
 313         return pid;
 314 }
 315
 316 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
 317                 struct file *filp)
 318 {
 319         long err = -EINVAL;
 320
 321         switch (cmd) {
 322         case F_DUPFD:
 323                 get_file(filp);
 324                 err = dupfd(filp, arg);
 325                 break;
 326         case F_GETFD:
 327                 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
 328                 break;
 329         case F_SETFD:
 330                 err = 0;
 331                 set_close_on_exec(fd, arg & FD_CLOEXEC);
 332                 break;
 333         case F_GETFL:
 334                 err = filp->f_flags;
 335                 break;
 336         case F_SETFL:
 337                 err = setfl(fd, filp, arg);
 338                 break;
 339         case F_GETLK:
 340                 err = fcntl_getlk(filp, (struct flock __user *) arg);
 341                 break;
 342         case F_SETLK:
 343         case F_SETLKW:
 344                 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
 345                 break;
 346         case F_GETOWN:
 347                 /*
 348                  * XXX If f_owner is a process group, the
 349                  * negative return value will get converted
 350                  * into an error.  Oops.  If we keep the
 351                  * current syscall conventions, the only way
 352                  * to fix this will be in libc.
 353                  */
 354                 err = f_getown(filp);
 355                 force_successful_syscall_return();
 356                 break;
 357         case F_SETOWN:
 358                 err = f_setown(filp, arg, 1);
 359                 break;
 360         case F_GETSIG:
 361                 err = filp->f_owner.signum;
 362                 break;
 363         case F_SETSIG:
 364                 /* arg == 0 restores default behaviour. */
 365                 if (!valid_signal(arg)) {
 366                         break;
 367                 }
 368                 err = 0;
 369                 filp->f_owner.signum = arg;
 370                 break;
 371         case F_GETLEASE:
 372                 err = fcntl_getlease(filp);
 373                 break;
 374         case F_SETLEASE:
 375                 err = fcntl_setlease(fd, filp, arg);
 376                 break;
 377         case F_NOTIFY:
 378                 err = fcntl_dirnotify(fd, filp, arg);
 379                 break;
 380         default:
 381                 break;
 382         }
 383         return err;
 384 }
 385
 386 asmlinkage long sys_fcntl(unsigned int fd, unsigned int cmd, unsigned long arg)
 387 {
 388         struct file *filp;
 389         long err = -EBADF;
 390
 391         filp = fget(fd);
 392         if (!filp)
 393                 goto out;
 394
 395         err = security_file_fcntl(filp, cmd, arg);
 396         if (err) {
 397                 fput(filp);
 398                 return err;
 399         }
 400
 401         err = do_fcntl(fd, cmd, arg, filp);
 402
 403         fput(filp);
 404 out:
 405         return err;
 406 }
 407
 408 #if BITS_PER_LONG == 32
 409 asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
 410 {
 411         struct file * filp;
 412         long err;
 413
 414         err = -EBADF;
 415         filp = fget(fd);
 416         if (!filp)
 417                 goto out;
 418
 419         err = security_file_fcntl(filp, cmd, arg);
 420         if (err) {
 421                 fput(filp);
 422                 return err;
 423         }
 424         err = -EBADF;
 425
 426         switch (cmd) {
 427                 case F_GETLK64:
 428                         err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
 429                         break;
 430                 case F_SETLK64:
 431                 case F_SETLKW64:
 432                         err = fcntl_setlk64(fd, filp, cmd,
 433                                         (struct flock64 __user *) arg);
 434                         break;
 435                 default:
 436                         err = do_fcntl(fd, cmd, arg, filp);
 437                         break;
 438         }
 439         fput(filp);
 440 out:
 441         return err;
 442 }
 443 #endif
 444
 445 /* Table to convert sigio signal codes into poll band bitmaps */
 446
 447 static const long band_table[NSIGPOLL] = {
 448         POLLIN | POLLRDNORM,                    /* POLL_IN */
 449         POLLOUT | POLLWRNORM | POLLWRBAND,      /* POLL_OUT */
 450         POLLIN | POLLRDNORM | POLLMSG,          /* POLL_MSG */
 451         POLLERR,                                /* POLL_ERR */
 452         POLLPRI | POLLRDBAND,                   /* POLL_PRI */
 453         POLLHUP | POLLERR                       /* POLL_HUP */
 454 };
 455
 456 static inline int sigio_perm(struct task_struct *p,
 457                              struct fown_struct *fown, int sig)
 458 {
 459         return (((fown->euid == 0) ||
 460                  (fown->euid == p->suid) || (fown->euid == p->uid) ||
 461                  (fown->uid == p->suid) || (fown->uid == p->uid)) &&
 462                 !security_file_send_sigiotask(p, fown, sig));
 463 }
 464
 465 static void send_sigio_to_task(struct task_struct *p,
 466                                struct fown_struct *fown,
 467                                int fd,
 468                                int reason)
 469 {
 470         if (!sigio_perm(p, fown, fown->signum))
 471                 return;
 472
 473         switch (fown->signum) {
 474                 siginfo_t si;
 475                 default:
 476                         /* Queue a rt signal with the appropriate fd as its
 477                            value.  We use SI_SIGIO as the source, not
 478                            SI_KERNEL, since kernel signals always get
 479                            delivered even if we can't queue.  Failure to
 480                            queue in this case _should_ be reported; we fall
 481                            back to SIGIO in that case. --sct */
 482                         si.si_signo = fown->signum;
 483                         si.si_errno = 0;
 484                         si.si_code  = reason;
 485                         /* Make sure we are called with one of the POLL_*
 486                            reasons, otherwise we could leak kernel stack into
 487                            userspace.  */
 488                         BUG_ON((reason & __SI_MASK) != __SI_POLL);
 489                         if (reason - POLL_IN >= NSIGPOLL)
 490                                 si.si_band  = ~0L;
 491                         else
 492                                 si.si_band = band_table[reason - POLL_IN];
 493                         si.si_fd    = fd;
 494                         if (!group_send_sig_info(fown->signum, &si, p))
 495                                 break;
 496                 /* fall-through: fall back on the old plain SIGIO signal */
 497                 case 0:
 498                         group_send_sig_info(SIGIO, SEND_SIG_PRIV, p);
 499         }
 500 }
 501
 502 void send_sigio(struct fown_struct *fown, int fd, int band)
 503 {
 504         struct task_struct *p;
 505         enum pid_type type;
 506         struct pid *pid;
 507
 508         read_lock(&fown->lock);
 509         type = fown->pid_type;
 510         pid = fown->pid;
 511         if (!pid)
 512                 goto out_unlock_fown;
 513
 514         read_lock(&tasklist_lock);
 515         do_each_pid_task(pid, type, p) {
 516                 send_sigio_to_task(p, fown, fd, band);
 517         } while_each_pid_task(pid, type, p);
 518         read_unlock(&tasklist_lock);
 519  out_unlock_fown:
 520         read_unlock(&fown->lock);
 521 }
 522
 523 static void send_sigurg_to_task(struct task_struct *p,
 524                                 struct fown_struct *fown)
 525 {
 526         if (sigio_perm(p, fown, SIGURG))
 527                 group_send_sig_info(SIGURG, SEND_SIG_PRIV, p);
 528 }
 529
 530 int send_sigurg(struct fown_struct *fown)
 531 {
 532         struct task_struct *p;
 533         enum pid_type type;
 534         struct pid *pid;
 535         int ret = 0;
 536
 537         read_lock(&fown->lock);
 538         type = fown->pid_type;
 539         pid = fown->pid;
 540         if (!pid)
 541                 goto out_unlock_fown;
 542
 543         ret = 1;
 544
 545         read_lock(&tasklist_lock);
 546         do_each_pid_task(pid, type, p) {
 547                 send_sigurg_to_task(p, fown);
 548         } while_each_pid_task(pid, type, p);
 549         read_unlock(&tasklist_lock);
 550  out_unlock_fown:
 551         read_unlock(&fown->lock);
 552         return ret;
 553 }
 554
 555 static DEFINE_RWLOCK(fasync_lock);
 556 static kmem_cache_t *fasync_cache __read_mostly;
 557
 558 /*
 559  * fasync_helper() is used by some character device drivers (mainly mice)
 560  * to set up the fasync queue. It returns negative on error, 0 if it did
 561  * no changes and positive if it added/deleted the entry.
 562  */
 563 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
 564 {
 565         struct fasync_struct *fa, **fp;
 566         struct fasync_struct *new = NULL;
 567         int result = 0;
 568
 569         if (on) {
 570                 new = kmem_cache_alloc(fasync_cache, SLAB_KERNEL);
 571                 if (!new)
 572                         return -ENOMEM;
 573         }
 574         write_lock_irq(&fasync_lock);
 575         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 576                 if (fa->fa_file == filp) {
 577                         if(on) {
 578                                 fa->fa_fd = fd;
 579                                 kmem_cache_free(fasync_cache, new);
 580                         } else {
 581                                 *fp = fa->fa_next;
 582                                 kmem_cache_free(fasync_cache, fa);
 583                                 result = 1;
 584                         }
 585                         goto out;
 586                 }
 587         }
 588
 589         if (on) {
 590                 new->magic = FASYNC_MAGIC;
 591                 new->fa_file = filp;
 592                 new->fa_fd = fd;
 593                 new->fa_next = *fapp;
 594                 *fapp = new;
 595                 result = 1;
 596         }
 597 out:
 598         write_unlock_irq(&fasync_lock);
 599         return result;
 600 }
 601
 602 EXPORT_SYMBOL(fasync_helper);
 603
 604 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
 605 {
 606         while (fa) {
 607                 struct fown_struct * fown;
 608                 if (fa->magic != FASYNC_MAGIC) {
 609                         printk(KERN_ERR "kill_fasync: bad magic number in "
 610                                "fasync_struct!\n");
 611                         return;
 612                 }
 613                 fown = &fa->fa_file->f_owner;
 614                 /* Don't send SIGURG to processes which have not set a
 615                    queued signum: SIGURG has its own default signalling
 616                    mechanism. */
 617                 if (!(sig == SIGURG && fown->signum == 0))
 618                         send_sigio(fown, fa->fa_fd, band);
 619                 fa = fa->fa_next;
 620         }
 621 }
 622
 623 EXPORT_SYMBOL(__kill_fasync);
 624
 625 void kill_fasync(struct fasync_struct **fp, int sig, int band)
 626 {
 627         /* First a quick test without locking: usually
 628          * the list is empty.
 629          */
 630         if (*fp) {
 631                 read_lock(&fasync_lock);
 632                 /* reread *fp after obtaining the lock */
 633                 __kill_fasync(*fp, sig, band);
 634                 read_unlock(&fasync_lock);
 635         }
 636 }
 637 EXPORT_SYMBOL(kill_fasync);
 638
 639 static int __init fasync_init(void)
 640 {
 641         fasync_cache = kmem_cache_create("fasync_cache",
 642                 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL, NULL);
 643         return 0;
 644 }
 645
 646 module_init(fasync_init)