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/pipe_fs_i.h>
  18 #include <linux/security.h>
  19 #include <linux/ptrace.h>
  20 #include <linux/signal.h>
  21 #include <linux/rcupdate.h>
  22 #include <linux/pid_namespace.h>
  23
  24 #include <asm/poll.h>
  25 #include <asm/siginfo.h>
  26 #include <asm/uaccess.h>
  27
  28 void set_close_on_exec(unsigned int fd, int flag)
  29 {
  30         struct files_struct *files = current->files;
  31         struct fdtable *fdt;
  32         spin_lock(&files->file_lock);
  33         fdt = files_fdtable(files);
  34         if (flag)
  35                 __set_close_on_exec(fd, fdt);
  36         else
  37                 __clear_close_on_exec(fd, fdt);
  38         spin_unlock(&files->file_lock);
  39 }
  40
  41 static bool get_close_on_exec(unsigned int fd)
  42 {
  43         struct files_struct *files = current->files;
  44         struct fdtable *fdt;
  45         bool res;
  46         rcu_read_lock();
  47         fdt = files_fdtable(files);
  48         res = close_on_exec(fd, fdt);
  49         rcu_read_unlock();
  50         return res;
  51 }
  52
  53 SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
  54 {
  55         int err = -EBADF;
  56         struct file * file, *tofree;
  57         struct files_struct * files = current->files;
  58         struct fdtable *fdt;
  59
  60         if ((flags & ~O_CLOEXEC) != 0)
  61                 return -EINVAL;
  62
  63         if (unlikely(oldfd == newfd))
  64                 return -EINVAL;
  65
  66         spin_lock(&files->file_lock);
  67         err = expand_files(files, newfd);
  68         file = fcheck(oldfd);
  69         if (unlikely(!file))
  70                 goto Ebadf;
  71         if (unlikely(err < 0)) {
  72                 if (err == -EMFILE)
  73                         goto Ebadf;
  74                 goto out_unlock;
  75         }
  76         /*
  77          * We need to detect attempts to do dup2() over allocated but still
  78          * not finished descriptor.  NB: OpenBSD avoids that at the price of
  79          * extra work in their equivalent of fget() - they insert struct
  80          * file immediately after grabbing descriptor, mark it larval if
  81          * more work (e.g. actual opening) is needed and make sure that
  82          * fget() treats larval files as absent.  Potentially interesting,
  83          * but while extra work in fget() is trivial, locking implications
  84          * and amount of surgery on open()-related paths in VFS are not.
  85          * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
  86          * deadlocks in rather amusing ways, AFAICS.  All of that is out of
  87          * scope of POSIX or SUS, since neither considers shared descriptor
  88          * tables and this condition does not arise without those.
  89          */
  90         err = -EBUSY;
  91         fdt = files_fdtable(files);
  92         tofree = fdt->fd[newfd];
  93         if (!tofree && fd_is_open(newfd, fdt))
  94                 goto out_unlock;
  95         get_file(file);
  96         rcu_assign_pointer(fdt->fd[newfd], file);
  97         __set_open_fd(newfd, fdt);
  98         if (flags & O_CLOEXEC)
  99                 __set_close_on_exec(newfd, fdt);
 100         else
 101                 __clear_close_on_exec(newfd, fdt);
 102         spin_unlock(&files->file_lock);
 103
 104         if (tofree)
 105                 filp_close(tofree, files);
 106
 107         return newfd;
 108
 109 Ebadf:
 110         err = -EBADF;
 111 out_unlock:
 112         spin_unlock(&files->file_lock);
 113         return err;
 114 }
 115
 116 SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
 117 {
 118         if (unlikely(newfd == oldfd)) { /* corner case */
 119                 struct files_struct *files = current->files;
 120                 int retval = oldfd;
 121
 122                 rcu_read_lock();
 123                 if (!fcheck_files(files, oldfd))
 124                         retval = -EBADF;
 125                 rcu_read_unlock();
 126                 return retval;
 127         }
 128         return sys_dup3(oldfd, newfd, 0);
 129 }
 130
 131 SYSCALL_DEFINE1(dup, unsigned int, fildes)
 132 {
 133         int ret = -EBADF;
 134         struct file *file = fget_raw(fildes);
 135
 136         if (file) {
 137                 ret = get_unused_fd();
 138                 if (ret >= 0)
 139                         fd_install(ret, file);
 140                 else
 141                         fput(file);
 142         }
 143         return ret;
 144 }
 145
 146 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
 147
 148 static int setfl(int fd, struct file * filp, unsigned long arg)
 149 {
 150         struct inode * inode = filp->f_path.dentry->d_inode;
 151         int error = 0;
 152
 153         /*
 154          * O_APPEND cannot be cleared if the file is marked as append-only
 155          * and the file is open for write.
 156          */
 157         if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
 158                 return -EPERM;
 159
 160         /* O_NOATIME can only be set by the owner or superuser */
 161         if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
 162                 if (!inode_owner_or_capable(inode))
 163                         return -EPERM;
 164
 165         /* required for strict SunOS emulation */
 166         if (O_NONBLOCK != O_NDELAY)
 167                if (arg & O_NDELAY)
 168                    arg |= O_NONBLOCK;
 169
 170         if (arg & O_DIRECT) {
 171                 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
 172                         !filp->f_mapping->a_ops->direct_IO)
 173                                 return -EINVAL;
 174         }
 175
 176         if (filp->f_op && filp->f_op->check_flags)
 177                 error = filp->f_op->check_flags(arg);
 178         if (error)
 179                 return error;
 180
 181         /*
 182          * ->fasync() is responsible for setting the FASYNC bit.
 183          */
 184         if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op &&
 185                         filp->f_op->fasync) {
 186                 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
 187                 if (error < 0)
 188                         goto out;
 189                 if (error > 0)
 190                         error = 0;
 191         }
 192         spin_lock(&filp->f_lock);
 193         filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
 194         spin_unlock(&filp->f_lock);
 195
 196  out:
 197         return error;
 198 }
 199
 200 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
 201                      int force)
 202 {
 203         write_lock_irq(&filp->f_owner.lock);
 204         if (force || !filp->f_owner.pid) {
 205                 put_pid(filp->f_owner.pid);
 206                 filp->f_owner.pid = get_pid(pid);
 207                 filp->f_owner.pid_type = type;
 208
 209                 if (pid) {
 210                         const struct cred *cred = current_cred();
 211                         filp->f_owner.uid = cred->uid;
 212                         filp->f_owner.euid = cred->euid;
 213                 }
 214         }
 215         write_unlock_irq(&filp->f_owner.lock);
 216 }
 217
 218 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
 219                 int force)
 220 {
 221         int err;
 222
 223         err = security_file_set_fowner(filp);
 224         if (err)
 225                 return err;
 226
 227         f_modown(filp, pid, type, force);
 228         return 0;
 229 }
 230 EXPORT_SYMBOL(__f_setown);
 231
 232 int f_setown(struct file *filp, unsigned long arg, int force)
 233 {
 234         enum pid_type type;
 235         struct pid *pid;
 236         int who = arg;
 237         int result;
 238         type = PIDTYPE_PID;
 239         if (who < 0) {
 240                 type = PIDTYPE_PGID;
 241                 who = -who;
 242         }
 243         rcu_read_lock();
 244         pid = find_vpid(who);
 245         result = __f_setown(filp, pid, type, force);
 246         rcu_read_unlock();
 247         return result;
 248 }
 249 EXPORT_SYMBOL(f_setown);
 250
 251 void f_delown(struct file *filp)
 252 {
 253         f_modown(filp, NULL, PIDTYPE_PID, 1);
 254 }
 255
 256 pid_t f_getown(struct file *filp)
 257 {
 258         pid_t pid;
 259         read_lock(&filp->f_owner.lock);
 260         pid = pid_vnr(filp->f_owner.pid);
 261         if (filp->f_owner.pid_type == PIDTYPE_PGID)
 262                 pid = -pid;
 263         read_unlock(&filp->f_owner.lock);
 264         return pid;
 265 }
 266
 267 static int f_setown_ex(struct file *filp, unsigned long arg)
 268 {
 269         struct f_owner_ex * __user owner_p = (void * __user)arg;
 270         struct f_owner_ex owner;
 271         struct pid *pid;
 272         int type;
 273         int ret;
 274
 275         ret = copy_from_user(&owner, owner_p, sizeof(owner));
 276         if (ret)
 277                 return -EFAULT;
 278
 279         switch (owner.type) {
 280         case F_OWNER_TID:
 281                 type = PIDTYPE_MAX;
 282                 break;
 283
 284         case F_OWNER_PID:
 285                 type = PIDTYPE_PID;
 286                 break;
 287
 288         case F_OWNER_PGRP:
 289                 type = PIDTYPE_PGID;
 290                 break;
 291
 292         default:
 293                 return -EINVAL;
 294         }
 295
 296         rcu_read_lock();
 297         pid = find_vpid(owner.pid);
 298         if (owner.pid && !pid)
 299                 ret = -ESRCH;
 300         else
 301                 ret = __f_setown(filp, pid, type, 1);
 302         rcu_read_unlock();
 303
 304         return ret;
 305 }
 306
 307 static int f_getown_ex(struct file *filp, unsigned long arg)
 308 {
 309         struct f_owner_ex * __user owner_p = (void * __user)arg;
 310         struct f_owner_ex owner;
 311         int ret = 0;
 312
 313         read_lock(&filp->f_owner.lock);
 314         owner.pid = pid_vnr(filp->f_owner.pid);
 315         switch (filp->f_owner.pid_type) {
 316         case PIDTYPE_MAX:
 317                 owner.type = F_OWNER_TID;
 318                 break;
 319
 320         case PIDTYPE_PID:
 321                 owner.type = F_OWNER_PID;
 322                 break;
 323
 324         case PIDTYPE_PGID:
 325                 owner.type = F_OWNER_PGRP;
 326                 break;
 327
 328         default:
 329                 WARN_ON(1);
 330                 ret = -EINVAL;
 331                 break;
 332         }
 333         read_unlock(&filp->f_owner.lock);
 334
 335         if (!ret) {
 336                 ret = copy_to_user(owner_p, &owner, sizeof(owner));
 337                 if (ret)
 338                         ret = -EFAULT;
 339         }
 340         return ret;
 341 }
 342
 343 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
 344                 struct file *filp)
 345 {
 346         long err = -EINVAL;
 347
 348         switch (cmd) {
 349         case F_DUPFD:
 350         case F_DUPFD_CLOEXEC:
 351                 if (arg >= rlimit(RLIMIT_NOFILE))
 352                         break;
 353                 err = alloc_fd(arg, cmd == F_DUPFD_CLOEXEC ? O_CLOEXEC : 0);
 354                 if (err >= 0) {
 355                         get_file(filp);
 356                         fd_install(err, filp);
 357                 }
 358                 break;
 359         case F_GETFD:
 360                 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
 361                 break;
 362         case F_SETFD:
 363                 err = 0;
 364                 set_close_on_exec(fd, arg & FD_CLOEXEC);
 365                 break;
 366         case F_GETFL:
 367                 err = filp->f_flags;
 368                 break;
 369         case F_SETFL:
 370                 err = setfl(fd, filp, arg);
 371                 break;
 372         case F_GETLK:
 373                 err = fcntl_getlk(filp, (struct flock __user *) arg);
 374                 break;
 375         case F_SETLK:
 376         case F_SETLKW:
 377                 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
 378                 break;
 379         case F_GETOWN:
 380                 /*
 381                  * XXX If f_owner is a process group, the
 382                  * negative return value will get converted
 383                  * into an error.  Oops.  If we keep the
 384                  * current syscall conventions, the only way
 385                  * to fix this will be in libc.
 386                  */
 387                 err = f_getown(filp);
 388                 force_successful_syscall_return();
 389                 break;
 390         case F_SETOWN:
 391                 err = f_setown(filp, arg, 1);
 392                 break;
 393         case F_GETOWN_EX:
 394                 err = f_getown_ex(filp, arg);
 395                 break;
 396         case F_SETOWN_EX:
 397                 err = f_setown_ex(filp, arg);
 398                 break;
 399         case F_GETSIG:
 400                 err = filp->f_owner.signum;
 401                 break;
 402         case F_SETSIG:
 403                 /* arg == 0 restores default behaviour. */
 404                 if (!valid_signal(arg)) {
 405                         break;
 406                 }
 407                 err = 0;
 408                 filp->f_owner.signum = arg;
 409                 break;
 410         case F_GETLEASE:
 411                 err = fcntl_getlease(filp);
 412                 break;
 413         case F_SETLEASE:
 414                 err = fcntl_setlease(fd, filp, arg);
 415                 break;
 416         case F_NOTIFY:
 417                 err = fcntl_dirnotify(fd, filp, arg);
 418                 break;
 419         case F_SETPIPE_SZ:
 420         case F_GETPIPE_SZ:
 421                 err = pipe_fcntl(filp, cmd, arg);
 422                 break;
 423         default:
 424                 break;
 425         }
 426         return err;
 427 }
 428
 429 static int check_fcntl_cmd(unsigned cmd)
 430 {
 431         switch (cmd) {
 432         case F_DUPFD:
 433         case F_DUPFD_CLOEXEC:
 434         case F_GETFD:
 435         case F_SETFD:
 436         case F_GETFL:
 437                 return 1;
 438         }
 439         return 0;
 440 }
 441
 442 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
 443 {
 444         struct file *filp;
 445         int fput_needed;
 446         long err = -EBADF;
 447
 448         filp = fget_raw_light(fd, &fput_needed);
 449         if (!filp)
 450                 goto out;
 451
 452         if (unlikely(filp->f_mode & FMODE_PATH)) {
 453                 if (!check_fcntl_cmd(cmd))
 454                         goto out1;
 455         }
 456
 457         err = security_file_fcntl(filp, cmd, arg);
 458         if (!err)
 459                 err = do_fcntl(fd, cmd, arg, filp);
 460
 461 out1:
 462         fput_light(filp, fput_needed);
 463 out:
 464         return err;
 465 }
 466
 467 #if BITS_PER_LONG == 32
 468 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
 469                 unsigned long, arg)
 470 {
 471         struct file * filp;
 472         long err = -EBADF;
 473         int fput_needed;
 474
 475         filp = fget_raw_light(fd, &fput_needed);
 476         if (!filp)
 477                 goto out;
 478
 479         if (unlikely(filp->f_mode & FMODE_PATH)) {
 480                 if (!check_fcntl_cmd(cmd))
 481                         goto out1;
 482         }
 483
 484         err = security_file_fcntl(filp, cmd, arg);
 485         if (err)
 486                 goto out1;
 487
 488         switch (cmd) {
 489                 case F_GETLK64:
 490                         err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
 491                         break;
 492                 case F_SETLK64:
 493                 case F_SETLKW64:
 494                         err = fcntl_setlk64(fd, filp, cmd,
 495                                         (struct flock64 __user *) arg);
 496                         break;
 497                 default:
 498                         err = do_fcntl(fd, cmd, arg, filp);
 499                         break;
 500         }
 501 out1:
 502         fput_light(filp, fput_needed);
 503 out:
 504         return err;
 505 }
 506 #endif
 507
 508 /* Table to convert sigio signal codes into poll band bitmaps */
 509
 510 static const long band_table[NSIGPOLL] = {
 511         POLLIN | POLLRDNORM,                    /* POLL_IN */
 512         POLLOUT | POLLWRNORM | POLLWRBAND,      /* POLL_OUT */
 513         POLLIN | POLLRDNORM | POLLMSG,          /* POLL_MSG */
 514         POLLERR,                                /* POLL_ERR */
 515         POLLPRI | POLLRDBAND,                   /* POLL_PRI */
 516         POLLHUP | POLLERR                       /* POLL_HUP */
 517 };
 518
 519 static inline int sigio_perm(struct task_struct *p,
 520                              struct fown_struct *fown, int sig)
 521 {
 522         const struct cred *cred;
 523         int ret;
 524
 525         rcu_read_lock();
 526         cred = __task_cred(p);
 527         ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
 528                 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
 529                 uid_eq(fown->uid,  cred->suid) || uid_eq(fown->uid,  cred->uid)) &&
 530                !security_file_send_sigiotask(p, fown, sig));
 531         rcu_read_unlock();
 532         return ret;
 533 }
 534
 535 static void send_sigio_to_task(struct task_struct *p,
 536                                struct fown_struct *fown,
 537                                int fd, int reason, int group)
 538 {
 539         /*
 540          * F_SETSIG can change ->signum lockless in parallel, make
 541          * sure we read it once and use the same value throughout.
 542          */
 543         int signum = ACCESS_ONCE(fown->signum);
 544
 545         if (!sigio_perm(p, fown, signum))
 546                 return;
 547
 548         switch (signum) {
 549                 siginfo_t si;
 550                 default:
 551                         /* Queue a rt signal with the appropriate fd as its
 552                            value.  We use SI_SIGIO as the source, not
 553                            SI_KERNEL, since kernel signals always get
 554                            delivered even if we can't queue.  Failure to
 555                            queue in this case _should_ be reported; we fall
 556                            back to SIGIO in that case. --sct */
 557                         si.si_signo = signum;
 558                         si.si_errno = 0;
 559                         si.si_code  = reason;
 560                         /* Make sure we are called with one of the POLL_*
 561                            reasons, otherwise we could leak kernel stack into
 562                            userspace.  */
 563                         BUG_ON((reason & __SI_MASK) != __SI_POLL);
 564                         if (reason - POLL_IN >= NSIGPOLL)
 565                                 si.si_band  = ~0L;
 566                         else
 567                                 si.si_band = band_table[reason - POLL_IN];
 568                         si.si_fd    = fd;
 569                         if (!do_send_sig_info(signum, &si, p, group))
 570                                 break;
 571                 /* fall-through: fall back on the old plain SIGIO signal */
 572                 case 0:
 573                         do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
 574         }
 575 }
 576
 577 void send_sigio(struct fown_struct *fown, int fd, int band)
 578 {
 579         struct task_struct *p;
 580         enum pid_type type;
 581         struct pid *pid;
 582         int group = 1;
 583
 584         read_lock(&fown->lock);
 585
 586         type = fown->pid_type;
 587         if (type == PIDTYPE_MAX) {
 588                 group = 0;
 589                 type = PIDTYPE_PID;
 590         }
 591
 592         pid = fown->pid;
 593         if (!pid)
 594                 goto out_unlock_fown;
 595
 596         read_lock(&tasklist_lock);
 597         do_each_pid_task(pid, type, p) {
 598                 send_sigio_to_task(p, fown, fd, band, group);
 599         } while_each_pid_task(pid, type, p);
 600         read_unlock(&tasklist_lock);
 601  out_unlock_fown:
 602         read_unlock(&fown->lock);
 603 }
 604
 605 static void send_sigurg_to_task(struct task_struct *p,
 606                                 struct fown_struct *fown, int group)
 607 {
 608         if (sigio_perm(p, fown, SIGURG))
 609                 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
 610 }
 611
 612 int send_sigurg(struct fown_struct *fown)
 613 {
 614         struct task_struct *p;
 615         enum pid_type type;
 616         struct pid *pid;
 617         int group = 1;
 618         int ret = 0;
 619
 620         read_lock(&fown->lock);
 621
 622         type = fown->pid_type;
 623         if (type == PIDTYPE_MAX) {
 624                 group = 0;
 625                 type = PIDTYPE_PID;
 626         }
 627
 628         pid = fown->pid;
 629         if (!pid)
 630                 goto out_unlock_fown;
 631
 632         ret = 1;
 633
 634         read_lock(&tasklist_lock);
 635         do_each_pid_task(pid, type, p) {
 636                 send_sigurg_to_task(p, fown, group);
 637         } while_each_pid_task(pid, type, p);
 638         read_unlock(&tasklist_lock);
 639  out_unlock_fown:
 640         read_unlock(&fown->lock);
 641         return ret;
 642 }
 643
 644 static DEFINE_SPINLOCK(fasync_lock);
 645 static struct kmem_cache *fasync_cache __read_mostly;
 646
 647 static void fasync_free_rcu(struct rcu_head *head)
 648 {
 649         kmem_cache_free(fasync_cache,
 650                         container_of(head, struct fasync_struct, fa_rcu));
 651 }
 652
 653 /*
 654  * Remove a fasync entry. If successfully removed, return
 655  * positive and clear the FASYNC flag. If no entry exists,
 656  * do nothing and return 0.
 657  *
 658  * NOTE! It is very important that the FASYNC flag always
 659  * match the state "is the filp on a fasync list".
 660  *
 661  */
 662 int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
 663 {
 664         struct fasync_struct *fa, **fp;
 665         int result = 0;
 666
 667         spin_lock(&filp->f_lock);
 668         spin_lock(&fasync_lock);
 669         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 670                 if (fa->fa_file != filp)
 671                         continue;
 672
 673                 spin_lock_irq(&fa->fa_lock);
 674                 fa->fa_file = NULL;
 675                 spin_unlock_irq(&fa->fa_lock);
 676
 677                 *fp = fa->fa_next;
 678                 call_rcu(&fa->fa_rcu, fasync_free_rcu);
 679                 filp->f_flags &= ~FASYNC;
 680                 result = 1;
 681                 break;
 682         }
 683         spin_unlock(&fasync_lock);
 684         spin_unlock(&filp->f_lock);
 685         return result;
 686 }
 687
 688 struct fasync_struct *fasync_alloc(void)
 689 {
 690         return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
 691 }
 692
 693 /*
 694  * NOTE! This can be used only for unused fasync entries:
 695  * entries that actually got inserted on the fasync list
 696  * need to be released by rcu - see fasync_remove_entry.
 697  */
 698 void fasync_free(struct fasync_struct *new)
 699 {
 700         kmem_cache_free(fasync_cache, new);
 701 }
 702
 703 /*
 704  * Insert a new entry into the fasync list.  Return the pointer to the
 705  * old one if we didn't use the new one.
 706  *
 707  * NOTE! It is very important that the FASYNC flag always
 708  * match the state "is the filp on a fasync list".
 709  */
 710 struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
 711 {
 712         struct fasync_struct *fa, **fp;
 713
 714         spin_lock(&filp->f_lock);
 715         spin_lock(&fasync_lock);
 716         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 717                 if (fa->fa_file != filp)
 718                         continue;
 719
 720                 spin_lock_irq(&fa->fa_lock);
 721                 fa->fa_fd = fd;
 722                 spin_unlock_irq(&fa->fa_lock);
 723                 goto out;
 724         }
 725
 726         spin_lock_init(&new->fa_lock);
 727         new->magic = FASYNC_MAGIC;
 728         new->fa_file = filp;
 729         new->fa_fd = fd;
 730         new->fa_next = *fapp;
 731         rcu_assign_pointer(*fapp, new);
 732         filp->f_flags |= FASYNC;
 733
 734 out:
 735         spin_unlock(&fasync_lock);
 736         spin_unlock(&filp->f_lock);
 737         return fa;
 738 }
 739
 740 /*
 741  * Add a fasync entry. Return negative on error, positive if
 742  * added, and zero if did nothing but change an existing one.
 743  */
 744 static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
 745 {
 746         struct fasync_struct *new;
 747
 748         new = fasync_alloc();
 749         if (!new)
 750                 return -ENOMEM;
 751
 752         /*
 753          * fasync_insert_entry() returns the old (update) entry if
 754          * it existed.
 755          *
 756          * So free the (unused) new entry and return 0 to let the
 757          * caller know that we didn't add any new fasync entries.
 758          */
 759         if (fasync_insert_entry(fd, filp, fapp, new)) {
 760                 fasync_free(new);
 761                 return 0;
 762         }
 763
 764         return 1;
 765 }
 766
 767 /*
 768  * fasync_helper() is used by almost all character device drivers
 769  * to set up the fasync queue, and for regular files by the file
 770  * lease code. It returns negative on error, 0 if it did no changes
 771  * and positive if it added/deleted the entry.
 772  */
 773 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
 774 {
 775         if (!on)
 776                 return fasync_remove_entry(filp, fapp);
 777         return fasync_add_entry(fd, filp, fapp);
 778 }
 779
 780 EXPORT_SYMBOL(fasync_helper);
 781
 782 /*
 783  * rcu_read_lock() is held
 784  */
 785 static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
 786 {
 787         while (fa) {
 788                 struct fown_struct *fown;
 789                 unsigned long flags;
 790
 791                 if (fa->magic != FASYNC_MAGIC) {
 792                         printk(KERN_ERR "kill_fasync: bad magic number in "
 793                                "fasync_struct!\n");
 794                         return;
 795                 }
 796                 spin_lock_irqsave(&fa->fa_lock, flags);
 797                 if (fa->fa_file) {
 798                         fown = &fa->fa_file->f_owner;
 799                         /* Don't send SIGURG to processes which have not set a
 800                            queued signum: SIGURG has its own default signalling
 801                            mechanism. */
 802                         if (!(sig == SIGURG && fown->signum == 0))
 803                                 send_sigio(fown, fa->fa_fd, band);
 804                 }
 805                 spin_unlock_irqrestore(&fa->fa_lock, flags);
 806                 fa = rcu_dereference(fa->fa_next);
 807         }
 808 }
 809
 810 void kill_fasync(struct fasync_struct **fp, int sig, int band)
 811 {
 812         /* First a quick test without locking: usually
 813          * the list is empty.
 814          */
 815         if (*fp) {
 816                 rcu_read_lock();
 817                 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
 818                 rcu_read_unlock();
 819         }
 820 }
 821 EXPORT_SYMBOL(kill_fasync);
 822
 823 static int __init fcntl_init(void)
 824 {
 825         /*
 826          * Please add new bits here to ensure allocation uniqueness.
 827          * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
 828          * is defined as O_NONBLOCK on some platforms and not on others.
 829          */
 830         BUILD_BUG_ON(19 - 1 /* for O_RDONLY being 0 */ != HWEIGHT32(
 831                 O_RDONLY        | O_WRONLY      | O_RDWR        |
 832                 O_CREAT         | O_EXCL        | O_NOCTTY      |
 833                 O_TRUNC         | O_APPEND      | /* O_NONBLOCK | */
 834                 __O_SYNC        | O_DSYNC       | FASYNC        |
 835                 O_DIRECT        | O_LARGEFILE   | O_DIRECTORY   |
 836                 O_NOFOLLOW      | O_NOATIME     | O_CLOEXEC     |
 837                 __FMODE_EXEC    | O_PATH
 838                 ));
 839
 840         fasync_cache = kmem_cache_create("fasync_cache",
 841                 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
 842         return 0;
 843 }
 844
 845 module_init(fcntl_init)