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/fdtable.h>
  13 #include <linux/capability.h>
  14 #include <linux/dnotify.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 #include <linux/pid_namespace.h>
  22
  23 #include <asm/poll.h>
  24 #include <asm/siginfo.h>
  25 #include <asm/uaccess.h>
  26
  27 void set_close_on_exec(unsigned int fd, int flag)
  28 {
  29         struct files_struct *files = current->files;
  30         struct fdtable *fdt;
  31         spin_lock(&files->file_lock);
  32         fdt = files_fdtable(files);
  33         if (flag)
  34                 FD_SET(fd, fdt->close_on_exec);
  35         else
  36                 FD_CLR(fd, fdt->close_on_exec);
  37         spin_unlock(&files->file_lock);
  38 }
  39
  40 static int get_close_on_exec(unsigned int fd)
  41 {
  42         struct files_struct *files = current->files;
  43         struct fdtable *fdt;
  44         int res;
  45         rcu_read_lock();
  46         fdt = files_fdtable(files);
  47         res = FD_ISSET(fd, fdt->close_on_exec);
  48         rcu_read_unlock();
  49         return res;
  50 }
  51
  52 /*
  53  * locate_fd finds a free file descriptor in the open_fds fdset,
  54  * expanding the fd arrays if necessary.  Must be called with the
  55  * file_lock held for write.
  56  */
  57
  58 static int locate_fd(unsigned int orig_start, int cloexec)
  59 {
  60         struct files_struct *files = current->files;
  61         unsigned int newfd;
  62         unsigned int start;
  63         int error;
  64         struct fdtable *fdt;
  65
  66         spin_lock(&files->file_lock);
  67
  68         error = -EINVAL;
  69         if (orig_start >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
  70                 goto out;
  71
  72 repeat:
  73         fdt = files_fdtable(files);
  74         /*
  75          * Someone might have closed fd's in the range
  76          * orig_start..fdt->next_fd
  77          */
  78         start = orig_start;
  79         if (start < files->next_fd)
  80                 start = files->next_fd;
  81
  82         newfd = start;
  83         if (start < fdt->max_fds)
  84                 newfd = find_next_zero_bit(fdt->open_fds->fds_bits,
  85                                            fdt->max_fds, start);
  86
  87         error = -EMFILE;
  88         if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
  89                 goto out;
  90
  91         error = expand_files(files, newfd);
  92         if (error < 0)
  93                 goto out;
  94
  95         /*
  96          * If we needed to expand the fs array we
  97          * might have blocked - try again.
  98          */
  99         if (error)
 100                 goto repeat;
 101
 102         if (start <= files->next_fd)
 103                 files->next_fd = newfd + 1;
 104
 105         FD_SET(newfd, fdt->open_fds);
 106         if (cloexec)
 107                 FD_SET(newfd, fdt->close_on_exec);
 108         else
 109                 FD_CLR(newfd, fdt->close_on_exec);
 110         error = newfd;
 111
 112 out:
 113         spin_unlock(&files->file_lock);
 114         return error;
 115 }
 116
 117 static int dupfd(struct file *file, unsigned int start, int cloexec)
 118 {
 119         int fd = locate_fd(start, cloexec);
 120         if (fd >= 0)
 121                 fd_install(fd, file);
 122         else
 123                 fput(file);
 124
 125         return fd;
 126 }
 127
 128 asmlinkage long sys_dup2(unsigned int oldfd, unsigned int newfd)
 129 {
 130         int err = -EBADF;
 131         struct file * file, *tofree;
 132         struct files_struct * files = current->files;
 133         struct fdtable *fdt;
 134
 135         spin_lock(&files->file_lock);
 136         if (!(file = fcheck(oldfd)))
 137                 goto out_unlock;
 138         err = newfd;
 139         if (newfd == oldfd)
 140                 goto out_unlock;
 141         err = -EBADF;
 142         if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
 143                 goto out_unlock;
 144         get_file(file);                 /* We are now finished with oldfd */
 145
 146         err = expand_files(files, newfd);
 147         if (err < 0)
 148                 goto out_fput;
 149
 150         /* To avoid races with open() and dup(), we will mark the fd as
 151          * in-use in the open-file bitmap throughout the entire dup2()
 152          * process.  This is quite safe: do_close() uses the fd array
 153          * entry, not the bitmap, to decide what work needs to be
 154          * done.  --sct */
 155         /* Doesn't work. open() might be there first. --AV */
 156
 157         /* Yes. It's a race. In user space. Nothing sane to do */
 158         err = -EBUSY;
 159         fdt = files_fdtable(files);
 160         tofree = fdt->fd[newfd];
 161         if (!tofree && FD_ISSET(newfd, fdt->open_fds))
 162                 goto out_fput;
 163
 164         rcu_assign_pointer(fdt->fd[newfd], file);
 165         FD_SET(newfd, fdt->open_fds);
 166         FD_CLR(newfd, fdt->close_on_exec);
 167         spin_unlock(&files->file_lock);
 168
 169         if (tofree)
 170                 filp_close(tofree, files);
 171         err = newfd;
 172 out:
 173         return err;
 174 out_unlock:
 175         spin_unlock(&files->file_lock);
 176         goto out;
 177
 178 out_fput:
 179         spin_unlock(&files->file_lock);
 180         fput(file);
 181         goto out;
 182 }
 183
 184 asmlinkage long sys_dup(unsigned int fildes)
 185 {
 186         int ret = -EBADF;
 187         struct file * file = fget(fildes);
 188
 189         if (file)
 190                 ret = dupfd(file, 0, 0);
 191         return ret;
 192 }
 193
 194 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
 195
 196 static int setfl(int fd, struct file * filp, unsigned long arg)
 197 {
 198         struct inode * inode = filp->f_path.dentry->d_inode;
 199         int error = 0;
 200
 201         /*
 202          * O_APPEND cannot be cleared if the file is marked as append-only
 203          * and the file is open for write.
 204          */
 205         if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
 206                 return -EPERM;
 207
 208         /* O_NOATIME can only be set by the owner or superuser */
 209         if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
 210                 if (!is_owner_or_cap(inode))
 211                         return -EPERM;
 212
 213         /* required for strict SunOS emulation */
 214         if (O_NONBLOCK != O_NDELAY)
 215                if (arg & O_NDELAY)
 216                    arg |= O_NONBLOCK;
 217
 218         if (arg & O_DIRECT) {
 219                 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
 220                         !filp->f_mapping->a_ops->direct_IO)
 221                                 return -EINVAL;
 222         }
 223
 224         if (filp->f_op && filp->f_op->check_flags)
 225                 error = filp->f_op->check_flags(arg);
 226         if (error)
 227                 return error;
 228
 229         if ((arg ^ filp->f_flags) & FASYNC) {
 230                 if (filp->f_op && filp->f_op->fasync) {
 231                         error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
 232                         if (error < 0)
 233                                 goto out;
 234                 }
 235         }
 236
 237         filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
 238  out:
 239         return error;
 240 }
 241
 242 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
 243                      uid_t uid, uid_t euid, int force)
 244 {
 245         write_lock_irq(&filp->f_owner.lock);
 246         if (force || !filp->f_owner.pid) {
 247                 put_pid(filp->f_owner.pid);
 248                 filp->f_owner.pid = get_pid(pid);
 249                 filp->f_owner.pid_type = type;
 250                 filp->f_owner.uid = uid;
 251                 filp->f_owner.euid = euid;
 252         }
 253         write_unlock_irq(&filp->f_owner.lock);
 254 }
 255
 256 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
 257                 int force)
 258 {
 259         int err;
 260
 261         err = security_file_set_fowner(filp);
 262         if (err)
 263                 return err;
 264
 265         f_modown(filp, pid, type, current->uid, current->euid, force);
 266         return 0;
 267 }
 268 EXPORT_SYMBOL(__f_setown);
 269
 270 int f_setown(struct file *filp, unsigned long arg, int force)
 271 {
 272         enum pid_type type;
 273         struct pid *pid;
 274         int who = arg;
 275         int result;
 276         type = PIDTYPE_PID;
 277         if (who < 0) {
 278                 type = PIDTYPE_PGID;
 279                 who = -who;
 280         }
 281         rcu_read_lock();
 282         pid = find_vpid(who);
 283         result = __f_setown(filp, pid, type, force);
 284         rcu_read_unlock();
 285         return result;
 286 }
 287 EXPORT_SYMBOL(f_setown);
 288
 289 void f_delown(struct file *filp)
 290 {
 291         f_modown(filp, NULL, PIDTYPE_PID, 0, 0, 1);
 292 }
 293
 294 pid_t f_getown(struct file *filp)
 295 {
 296         pid_t pid;
 297         read_lock(&filp->f_owner.lock);
 298         pid = pid_vnr(filp->f_owner.pid);
 299         if (filp->f_owner.pid_type == PIDTYPE_PGID)
 300                 pid = -pid;
 301         read_unlock(&filp->f_owner.lock);
 302         return pid;
 303 }
 304
 305 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
 306                 struct file *filp)
 307 {
 308         long err = -EINVAL;
 309
 310         switch (cmd) {
 311         case F_DUPFD:
 312         case F_DUPFD_CLOEXEC:
 313                 get_file(filp);
 314                 err = dupfd(filp, arg, cmd == F_DUPFD_CLOEXEC);
 315                 break;
 316         case F_GETFD:
 317                 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
 318                 break;
 319         case F_SETFD:
 320                 err = 0;
 321                 set_close_on_exec(fd, arg & FD_CLOEXEC);
 322                 break;
 323         case F_GETFL:
 324                 err = filp->f_flags;
 325                 break;
 326         case F_SETFL:
 327                 err = setfl(fd, filp, arg);
 328                 break;
 329         case F_GETLK:
 330                 err = fcntl_getlk(filp, (struct flock __user *) arg);
 331                 break;
 332         case F_SETLK:
 333         case F_SETLKW:
 334                 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
 335                 break;
 336         case F_GETOWN:
 337                 /*
 338                  * XXX If f_owner is a process group, the
 339                  * negative return value will get converted
 340                  * into an error.  Oops.  If we keep the
 341                  * current syscall conventions, the only way
 342                  * to fix this will be in libc.
 343                  */
 344                 err = f_getown(filp);
 345                 force_successful_syscall_return();
 346                 break;
 347         case F_SETOWN:
 348                 err = f_setown(filp, arg, 1);
 349                 break;
 350         case F_GETSIG:
 351                 err = filp->f_owner.signum;
 352                 break;
 353         case F_SETSIG:
 354                 /* arg == 0 restores default behaviour. */
 355                 if (!valid_signal(arg)) {
 356                         break;
 357                 }
 358                 err = 0;
 359                 filp->f_owner.signum = arg;
 360                 break;
 361         case F_GETLEASE:
 362                 err = fcntl_getlease(filp);
 363                 break;
 364         case F_SETLEASE:
 365                 err = fcntl_setlease(fd, filp, arg);
 366                 break;
 367         case F_NOTIFY:
 368                 err = fcntl_dirnotify(fd, filp, arg);
 369                 break;
 370         default:
 371                 break;
 372         }
 373         return err;
 374 }
 375
 376 asmlinkage long sys_fcntl(unsigned int fd, unsigned int cmd, unsigned long arg)
 377 {
 378         struct file *filp;
 379         long err = -EBADF;
 380
 381         filp = fget(fd);
 382         if (!filp)
 383                 goto out;
 384
 385         err = security_file_fcntl(filp, cmd, arg);
 386         if (err) {
 387                 fput(filp);
 388                 return err;
 389         }
 390
 391         err = do_fcntl(fd, cmd, arg, filp);
 392
 393         fput(filp);
 394 out:
 395         return err;
 396 }
 397
 398 #if BITS_PER_LONG == 32
 399 asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
 400 {
 401         struct file * filp;
 402         long err;
 403
 404         err = -EBADF;
 405         filp = fget(fd);
 406         if (!filp)
 407                 goto out;
 408
 409         err = security_file_fcntl(filp, cmd, arg);
 410         if (err) {
 411                 fput(filp);
 412                 return err;
 413         }
 414         err = -EBADF;
 415
 416         switch (cmd) {
 417                 case F_GETLK64:
 418                         err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
 419                         break;
 420                 case F_SETLK64:
 421                 case F_SETLKW64:
 422                         err = fcntl_setlk64(fd, filp, cmd,
 423                                         (struct flock64 __user *) arg);
 424                         break;
 425                 default:
 426                         err = do_fcntl(fd, cmd, arg, filp);
 427                         break;
 428         }
 429         fput(filp);
 430 out:
 431         return err;
 432 }
 433 #endif
 434
 435 /* Table to convert sigio signal codes into poll band bitmaps */
 436
 437 static const long band_table[NSIGPOLL] = {
 438         POLLIN | POLLRDNORM,                    /* POLL_IN */
 439         POLLOUT | POLLWRNORM | POLLWRBAND,      /* POLL_OUT */
 440         POLLIN | POLLRDNORM | POLLMSG,          /* POLL_MSG */
 441         POLLERR,                                /* POLL_ERR */
 442         POLLPRI | POLLRDBAND,                   /* POLL_PRI */
 443         POLLHUP | POLLERR                       /* POLL_HUP */
 444 };
 445
 446 static inline int sigio_perm(struct task_struct *p,
 447                              struct fown_struct *fown, int sig)
 448 {
 449         return (((fown->euid == 0) ||
 450                  (fown->euid == p->suid) || (fown->euid == p->uid) ||
 451                  (fown->uid == p->suid) || (fown->uid == p->uid)) &&
 452                 !security_file_send_sigiotask(p, fown, sig));
 453 }
 454
 455 static void send_sigio_to_task(struct task_struct *p,
 456                                struct fown_struct *fown,
 457                                int fd,
 458                                int reason)
 459 {
 460         if (!sigio_perm(p, fown, fown->signum))
 461                 return;
 462
 463         switch (fown->signum) {
 464                 siginfo_t si;
 465                 default:
 466                         /* Queue a rt signal with the appropriate fd as its
 467                            value.  We use SI_SIGIO as the source, not
 468                            SI_KERNEL, since kernel signals always get
 469                            delivered even if we can't queue.  Failure to
 470                            queue in this case _should_ be reported; we fall
 471                            back to SIGIO in that case. --sct */
 472                         si.si_signo = fown->signum;
 473                         si.si_errno = 0;
 474                         si.si_code  = reason;
 475                         /* Make sure we are called with one of the POLL_*
 476                            reasons, otherwise we could leak kernel stack into
 477                            userspace.  */
 478                         BUG_ON((reason & __SI_MASK) != __SI_POLL);
 479                         if (reason - POLL_IN >= NSIGPOLL)
 480                                 si.si_band  = ~0L;
 481                         else
 482                                 si.si_band = band_table[reason - POLL_IN];
 483                         si.si_fd    = fd;
 484                         if (!group_send_sig_info(fown->signum, &si, p))
 485                                 break;
 486                 /* fall-through: fall back on the old plain SIGIO signal */
 487                 case 0:
 488                         group_send_sig_info(SIGIO, SEND_SIG_PRIV, p);
 489         }
 490 }
 491
 492 void send_sigio(struct fown_struct *fown, int fd, int band)
 493 {
 494         struct task_struct *p;
 495         enum pid_type type;
 496         struct pid *pid;
 497
 498         read_lock(&fown->lock);
 499         type = fown->pid_type;
 500         pid = fown->pid;
 501         if (!pid)
 502                 goto out_unlock_fown;
 503
 504         read_lock(&tasklist_lock);
 505         do_each_pid_task(pid, type, p) {
 506                 send_sigio_to_task(p, fown, fd, band);
 507         } while_each_pid_task(pid, type, p);
 508         read_unlock(&tasklist_lock);
 509  out_unlock_fown:
 510         read_unlock(&fown->lock);
 511 }
 512
 513 static void send_sigurg_to_task(struct task_struct *p,
 514                                 struct fown_struct *fown)
 515 {
 516         if (sigio_perm(p, fown, SIGURG))
 517                 group_send_sig_info(SIGURG, SEND_SIG_PRIV, p);
 518 }
 519
 520 int send_sigurg(struct fown_struct *fown)
 521 {
 522         struct task_struct *p;
 523         enum pid_type type;
 524         struct pid *pid;
 525         int ret = 0;
 526
 527         read_lock(&fown->lock);
 528         type = fown->pid_type;
 529         pid = fown->pid;
 530         if (!pid)
 531                 goto out_unlock_fown;
 532
 533         ret = 1;
 534
 535         read_lock(&tasklist_lock);
 536         do_each_pid_task(pid, type, p) {
 537                 send_sigurg_to_task(p, fown);
 538         } while_each_pid_task(pid, type, p);
 539         read_unlock(&tasklist_lock);
 540  out_unlock_fown:
 541         read_unlock(&fown->lock);
 542         return ret;
 543 }
 544
 545 static DEFINE_RWLOCK(fasync_lock);
 546 static struct kmem_cache *fasync_cache __read_mostly;
 547
 548 /*
 549  * fasync_helper() is used by some character device drivers (mainly mice)
 550  * to set up the fasync queue. It returns negative on error, 0 if it did
 551  * no changes and positive if it added/deleted the entry.
 552  */
 553 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
 554 {
 555         struct fasync_struct *fa, **fp;
 556         struct fasync_struct *new = NULL;
 557         int result = 0;
 558
 559         if (on) {
 560                 new = kmem_cache_alloc(fasync_cache, GFP_KERNEL);
 561                 if (!new)
 562                         return -ENOMEM;
 563         }
 564         write_lock_irq(&fasync_lock);
 565         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 566                 if (fa->fa_file == filp) {
 567                         if(on) {
 568                                 fa->fa_fd = fd;
 569                                 kmem_cache_free(fasync_cache, new);
 570                         } else {
 571                                 *fp = fa->fa_next;
 572                                 kmem_cache_free(fasync_cache, fa);
 573                                 result = 1;
 574                         }
 575                         goto out;
 576                 }
 577         }
 578
 579         if (on) {
 580                 new->magic = FASYNC_MAGIC;
 581                 new->fa_file = filp;
 582                 new->fa_fd = fd;
 583                 new->fa_next = *fapp;
 584                 *fapp = new;
 585                 result = 1;
 586         }
 587 out:
 588         write_unlock_irq(&fasync_lock);
 589         return result;
 590 }
 591
 592 EXPORT_SYMBOL(fasync_helper);
 593
 594 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
 595 {
 596         while (fa) {
 597                 struct fown_struct * fown;
 598                 if (fa->magic != FASYNC_MAGIC) {
 599                         printk(KERN_ERR "kill_fasync: bad magic number in "
 600                                "fasync_struct!\n");
 601                         return;
 602                 }
 603                 fown = &fa->fa_file->f_owner;
 604                 /* Don't send SIGURG to processes which have not set a
 605                    queued signum: SIGURG has its own default signalling
 606                    mechanism. */
 607                 if (!(sig == SIGURG && fown->signum == 0))
 608                         send_sigio(fown, fa->fa_fd, band);
 609                 fa = fa->fa_next;
 610         }
 611 }
 612
 613 EXPORT_SYMBOL(__kill_fasync);
 614
 615 void kill_fasync(struct fasync_struct **fp, int sig, int band)
 616 {
 617         /* First a quick test without locking: usually
 618          * the list is empty.
 619          */
 620         if (*fp) {
 621                 read_lock(&fasync_lock);
 622                 /* reread *fp after obtaining the lock */
 623                 __kill_fasync(*fp, sig, band);
 624                 read_unlock(&fasync_lock);
 625         }
 626 }
 627 EXPORT_SYMBOL(kill_fasync);
 628
 629 static int __init fasync_init(void)
 630 {
 631         fasync_cache = kmem_cache_create("fasync_cache",
 632                 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
 633         return 0;
 634 }
 635
 636 module_init(fasync_init)