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 #include <linux/user_namespace.h>
  24
  25 #include <asm/poll.h>
  26 #include <asm/siginfo.h>
  27 #include <asm/uaccess.h>
  28
  29 void set_close_on_exec(unsigned int fd, int flag)
  30 {
  31         struct files_struct *files = current->files;
  32         struct fdtable *fdt;
  33         spin_lock(&files->file_lock);
  34         fdt = files_fdtable(files);
  35         if (flag)
  36                 __set_close_on_exec(fd, fdt);
  37         else
  38                 __clear_close_on_exec(fd, fdt);
  39         spin_unlock(&files->file_lock);
  40 }
  41
  42 static bool get_close_on_exec(unsigned int fd)
  43 {
  44         struct files_struct *files = current->files;
  45         struct fdtable *fdt;
  46         bool res;
  47         rcu_read_lock();
  48         fdt = files_fdtable(files);
  49         res = close_on_exec(fd, fdt);
  50         rcu_read_unlock();
  51         return res;
  52 }
  53
  54 SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
  55 {
  56         int err = -EBADF;
  57         struct file * file, *tofree;
  58         struct files_struct * files = current->files;
  59         struct fdtable *fdt;
  60
  61         if ((flags & ~O_CLOEXEC) != 0)
  62                 return -EINVAL;
  63
  64         if (unlikely(oldfd == newfd))
  65                 return -EINVAL;
  66
  67         spin_lock(&files->file_lock);
  68         err = expand_files(files, newfd);
  69         file = fcheck(oldfd);
  70         if (unlikely(!file))
  71                 goto Ebadf;
  72         if (unlikely(err < 0)) {
  73                 if (err == -EMFILE)
  74                         goto Ebadf;
  75                 goto out_unlock;
  76         }
  77         /*
  78          * We need to detect attempts to do dup2() over allocated but still
  79          * not finished descriptor.  NB: OpenBSD avoids that at the price of
  80          * extra work in their equivalent of fget() - they insert struct
  81          * file immediately after grabbing descriptor, mark it larval if
  82          * more work (e.g. actual opening) is needed and make sure that
  83          * fget() treats larval files as absent.  Potentially interesting,
  84          * but while extra work in fget() is trivial, locking implications
  85          * and amount of surgery on open()-related paths in VFS are not.
  86          * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
  87          * deadlocks in rather amusing ways, AFAICS.  All of that is out of
  88          * scope of POSIX or SUS, since neither considers shared descriptor
  89          * tables and this condition does not arise without those.
  90          */
  91         err = -EBUSY;
  92         fdt = files_fdtable(files);
  93         tofree = fdt->fd[newfd];
  94         if (!tofree && fd_is_open(newfd, fdt))
  95                 goto out_unlock;
  96         get_file(file);
  97         rcu_assign_pointer(fdt->fd[newfd], file);
  98         __set_open_fd(newfd, fdt);
  99         if (flags & O_CLOEXEC)
 100                 __set_close_on_exec(newfd, fdt);
 101         else
 102                 __clear_close_on_exec(newfd, fdt);
 103         spin_unlock(&files->file_lock);
 104
 105         if (tofree)
 106                 filp_close(tofree, files);
 107
 108         return newfd;
 109
 110 Ebadf:
 111         err = -EBADF;
 112 out_unlock:
 113         spin_unlock(&files->file_lock);
 114         return err;
 115 }
 116
 117 SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
 118 {
 119         if (unlikely(newfd == oldfd)) { /* corner case */
 120                 struct files_struct *files = current->files;
 121                 int retval = oldfd;
 122
 123                 rcu_read_lock();
 124                 if (!fcheck_files(files, oldfd))
 125                         retval = -EBADF;
 126                 rcu_read_unlock();
 127                 return retval;
 128         }
 129         return sys_dup3(oldfd, newfd, 0);
 130 }
 131
 132 SYSCALL_DEFINE1(dup, unsigned int, fildes)
 133 {
 134         int ret = -EBADF;
 135         struct file *file = fget_raw(fildes);
 136
 137         if (file) {
 138                 ret = get_unused_fd();
 139                 if (ret >= 0)
 140                         fd_install(ret, file);
 141                 else
 142                         fput(file);
 143         }
 144         return ret;
 145 }
 146
 147 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
 148
 149 static int setfl(int fd, struct file * filp, unsigned long arg)
 150 {
 151         struct inode * inode = filp->f_path.dentry->d_inode;
 152         int error = 0;
 153
 154         /*
 155          * O_APPEND cannot be cleared if the file is marked as append-only
 156          * and the file is open for write.
 157          */
 158         if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
 159                 return -EPERM;
 160
 161         /* O_NOATIME can only be set by the owner or superuser */
 162         if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
 163                 if (!inode_owner_or_capable(inode))
 164                         return -EPERM;
 165
 166         /* required for strict SunOS emulation */
 167         if (O_NONBLOCK != O_NDELAY)
 168                if (arg & O_NDELAY)
 169                    arg |= O_NONBLOCK;
 170
 171         if (arg & O_DIRECT) {
 172                 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
 173                         !filp->f_mapping->a_ops->direct_IO)
 174                                 return -EINVAL;
 175         }
 176
 177         if (filp->f_op && filp->f_op->check_flags)
 178                 error = filp->f_op->check_flags(arg);
 179         if (error)
 180                 return error;
 181
 182         /*
 183          * ->fasync() is responsible for setting the FASYNC bit.
 184          */
 185         if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op &&
 186                         filp->f_op->fasync) {
 187                 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
 188                 if (error < 0)
 189                         goto out;
 190                 if (error > 0)
 191                         error = 0;
 192         }
 193         spin_lock(&filp->f_lock);
 194         filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
 195         spin_unlock(&filp->f_lock);
 196
 197  out:
 198         return error;
 199 }
 200
 201 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
 202                      int force)
 203 {
 204         write_lock_irq(&filp->f_owner.lock);
 205         if (force || !filp->f_owner.pid) {
 206                 put_pid(filp->f_owner.pid);
 207                 filp->f_owner.pid = get_pid(pid);
 208                 filp->f_owner.pid_type = type;
 209
 210                 if (pid) {
 211                         const struct cred *cred = current_cred();
 212                         filp->f_owner.uid = cred->uid;
 213                         filp->f_owner.euid = cred->euid;
 214                 }
 215         }
 216         write_unlock_irq(&filp->f_owner.lock);
 217 }
 218
 219 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
 220                 int force)
 221 {
 222         int err;
 223
 224         err = security_file_set_fowner(filp);
 225         if (err)
 226                 return err;
 227
 228         f_modown(filp, pid, type, force);
 229         return 0;
 230 }
 231 EXPORT_SYMBOL(__f_setown);
 232
 233 int f_setown(struct file *filp, unsigned long arg, int force)
 234 {
 235         enum pid_type type;
 236         struct pid *pid;
 237         int who = arg;
 238         int result;
 239         type = PIDTYPE_PID;
 240         if (who < 0) {
 241                 type = PIDTYPE_PGID;
 242                 who = -who;
 243         }
 244         rcu_read_lock();
 245         pid = find_vpid(who);
 246         result = __f_setown(filp, pid, type, force);
 247         rcu_read_unlock();
 248         return result;
 249 }
 250 EXPORT_SYMBOL(f_setown);
 251
 252 void f_delown(struct file *filp)
 253 {
 254         f_modown(filp, NULL, PIDTYPE_PID, 1);
 255 }
 256
 257 pid_t f_getown(struct file *filp)
 258 {
 259         pid_t pid;
 260         read_lock(&filp->f_owner.lock);
 261         pid = pid_vnr(filp->f_owner.pid);
 262         if (filp->f_owner.pid_type == PIDTYPE_PGID)
 263                 pid = -pid;
 264         read_unlock(&filp->f_owner.lock);
 265         return pid;
 266 }
 267
 268 static int f_setown_ex(struct file *filp, unsigned long arg)
 269 {
 270         struct f_owner_ex * __user owner_p = (void * __user)arg;
 271         struct f_owner_ex owner;
 272         struct pid *pid;
 273         int type;
 274         int ret;
 275
 276         ret = copy_from_user(&owner, owner_p, sizeof(owner));
 277         if (ret)
 278                 return -EFAULT;
 279
 280         switch (owner.type) {
 281         case F_OWNER_TID:
 282                 type = PIDTYPE_MAX;
 283                 break;
 284
 285         case F_OWNER_PID:
 286                 type = PIDTYPE_PID;
 287                 break;
 288
 289         case F_OWNER_PGRP:
 290                 type = PIDTYPE_PGID;
 291                 break;
 292
 293         default:
 294                 return -EINVAL;
 295         }
 296
 297         rcu_read_lock();
 298         pid = find_vpid(owner.pid);
 299         if (owner.pid && !pid)
 300                 ret = -ESRCH;
 301         else
 302                 ret = __f_setown(filp, pid, type, 1);
 303         rcu_read_unlock();
 304
 305         return ret;
 306 }
 307
 308 static int f_getown_ex(struct file *filp, unsigned long arg)
 309 {
 310         struct f_owner_ex * __user owner_p = (void * __user)arg;
 311         struct f_owner_ex owner;
 312         int ret = 0;
 313
 314         read_lock(&filp->f_owner.lock);
 315         owner.pid = pid_vnr(filp->f_owner.pid);
 316         switch (filp->f_owner.pid_type) {
 317         case PIDTYPE_MAX:
 318                 owner.type = F_OWNER_TID;
 319                 break;
 320
 321         case PIDTYPE_PID:
 322                 owner.type = F_OWNER_PID;
 323                 break;
 324
 325         case PIDTYPE_PGID:
 326                 owner.type = F_OWNER_PGRP;
 327                 break;
 328
 329         default:
 330                 WARN_ON(1);
 331                 ret = -EINVAL;
 332                 break;
 333         }
 334         read_unlock(&filp->f_owner.lock);
 335
 336         if (!ret) {
 337                 ret = copy_to_user(owner_p, &owner, sizeof(owner));
 338                 if (ret)
 339                         ret = -EFAULT;
 340         }
 341         return ret;
 342 }
 343
 344 #ifdef CONFIG_CHECKPOINT_RESTORE
 345 static int f_getowner_uids(struct file *filp, unsigned long arg)
 346 {
 347         struct user_namespace *user_ns = current_user_ns();
 348         uid_t * __user dst = (void * __user)arg;
 349         uid_t src[2];
 350         int err;
 351
 352         read_lock(&filp->f_owner.lock);
 353         src[0] = from_kuid(user_ns, filp->f_owner.uid);
 354         src[1] = from_kuid(user_ns, filp->f_owner.euid);
 355         read_unlock(&filp->f_owner.lock);
 356
 357         err  = put_user(src[0], &dst[0]);
 358         err |= put_user(src[1], &dst[1]);
 359
 360         return err;
 361 }
 362 #else
 363 static int f_getowner_uids(struct file *filp, unsigned long arg)
 364 {
 365         return -EINVAL;
 366 }
 367 #endif
 368
 369 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
 370                 struct file *filp)
 371 {
 372         long err = -EINVAL;
 373
 374         switch (cmd) {
 375         case F_DUPFD:
 376         case F_DUPFD_CLOEXEC:
 377                 if (arg >= rlimit(RLIMIT_NOFILE))
 378                         break;
 379                 err = alloc_fd(arg, cmd == F_DUPFD_CLOEXEC ? O_CLOEXEC : 0);
 380                 if (err >= 0) {
 381                         get_file(filp);
 382                         fd_install(err, filp);
 383                 }
 384                 break;
 385         case F_GETFD:
 386                 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
 387                 break;
 388         case F_SETFD:
 389                 err = 0;
 390                 set_close_on_exec(fd, arg & FD_CLOEXEC);
 391                 break;
 392         case F_GETFL:
 393                 err = filp->f_flags;
 394                 break;
 395         case F_SETFL:
 396                 err = setfl(fd, filp, arg);
 397                 break;
 398         case F_GETLK:
 399                 err = fcntl_getlk(filp, (struct flock __user *) arg);
 400                 break;
 401         case F_SETLK:
 402         case F_SETLKW:
 403                 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
 404                 break;
 405         case F_GETOWN:
 406                 /*
 407                  * XXX If f_owner is a process group, the
 408                  * negative return value will get converted
 409                  * into an error.  Oops.  If we keep the
 410                  * current syscall conventions, the only way
 411                  * to fix this will be in libc.
 412                  */
 413                 err = f_getown(filp);
 414                 force_successful_syscall_return();
 415                 break;
 416         case F_SETOWN:
 417                 err = f_setown(filp, arg, 1);
 418                 break;
 419         case F_GETOWN_EX:
 420                 err = f_getown_ex(filp, arg);
 421                 break;
 422         case F_SETOWN_EX:
 423                 err = f_setown_ex(filp, arg);
 424                 break;
 425         case F_GETOWNER_UIDS:
 426                 err = f_getowner_uids(filp, arg);
 427                 break;
 428         case F_GETSIG:
 429                 err = filp->f_owner.signum;
 430                 break;
 431         case F_SETSIG:
 432                 /* arg == 0 restores default behaviour. */
 433                 if (!valid_signal(arg)) {
 434                         break;
 435                 }
 436                 err = 0;
 437                 filp->f_owner.signum = arg;
 438                 break;
 439         case F_GETLEASE:
 440                 err = fcntl_getlease(filp);
 441                 break;
 442         case F_SETLEASE:
 443                 err = fcntl_setlease(fd, filp, arg);
 444                 break;
 445         case F_NOTIFY:
 446                 err = fcntl_dirnotify(fd, filp, arg);
 447                 break;
 448         case F_SETPIPE_SZ:
 449         case F_GETPIPE_SZ:
 450                 err = pipe_fcntl(filp, cmd, arg);
 451                 break;
 452         default:
 453                 break;
 454         }
 455         return err;
 456 }
 457
 458 static int check_fcntl_cmd(unsigned cmd)
 459 {
 460         switch (cmd) {
 461         case F_DUPFD:
 462         case F_DUPFD_CLOEXEC:
 463         case F_GETFD:
 464         case F_SETFD:
 465         case F_GETFL:
 466                 return 1;
 467         }
 468         return 0;
 469 }
 470
 471 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
 472 {
 473         struct file *filp;
 474         int fput_needed;
 475         long err = -EBADF;
 476
 477         filp = fget_raw_light(fd, &fput_needed);
 478         if (!filp)
 479                 goto out;
 480
 481         if (unlikely(filp->f_mode & FMODE_PATH)) {
 482                 if (!check_fcntl_cmd(cmd))
 483                         goto out1;
 484         }
 485
 486         err = security_file_fcntl(filp, cmd, arg);
 487         if (!err)
 488                 err = do_fcntl(fd, cmd, arg, filp);
 489
 490 out1:
 491         fput_light(filp, fput_needed);
 492 out:
 493         return err;
 494 }
 495
 496 #if BITS_PER_LONG == 32
 497 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
 498                 unsigned long, arg)
 499 {
 500         struct file * filp;
 501         long err = -EBADF;
 502         int fput_needed;
 503
 504         filp = fget_raw_light(fd, &fput_needed);
 505         if (!filp)
 506                 goto out;
 507
 508         if (unlikely(filp->f_mode & FMODE_PATH)) {
 509                 if (!check_fcntl_cmd(cmd))
 510                         goto out1;
 511         }
 512
 513         err = security_file_fcntl(filp, cmd, arg);
 514         if (err)
 515                 goto out1;
 516
 517         switch (cmd) {
 518                 case F_GETLK64:
 519                         err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
 520                         break;
 521                 case F_SETLK64:
 522                 case F_SETLKW64:
 523                         err = fcntl_setlk64(fd, filp, cmd,
 524                                         (struct flock64 __user *) arg);
 525                         break;
 526                 default:
 527                         err = do_fcntl(fd, cmd, arg, filp);
 528                         break;
 529         }
 530 out1:
 531         fput_light(filp, fput_needed);
 532 out:
 533         return err;
 534 }
 535 #endif
 536
 537 /* Table to convert sigio signal codes into poll band bitmaps */
 538
 539 static const long band_table[NSIGPOLL] = {
 540         POLLIN | POLLRDNORM,                    /* POLL_IN */
 541         POLLOUT | POLLWRNORM | POLLWRBAND,      /* POLL_OUT */
 542         POLLIN | POLLRDNORM | POLLMSG,          /* POLL_MSG */
 543         POLLERR,                                /* POLL_ERR */
 544         POLLPRI | POLLRDBAND,                   /* POLL_PRI */
 545         POLLHUP | POLLERR                       /* POLL_HUP */
 546 };
 547
 548 static inline int sigio_perm(struct task_struct *p,
 549                              struct fown_struct *fown, int sig)
 550 {
 551         const struct cred *cred;
 552         int ret;
 553
 554         rcu_read_lock();
 555         cred = __task_cred(p);
 556         ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
 557                 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
 558                 uid_eq(fown->uid,  cred->suid) || uid_eq(fown->uid,  cred->uid)) &&
 559                !security_file_send_sigiotask(p, fown, sig));
 560         rcu_read_unlock();
 561         return ret;
 562 }
 563
 564 static void send_sigio_to_task(struct task_struct *p,
 565                                struct fown_struct *fown,
 566                                int fd, int reason, int group)
 567 {
 568         /*
 569          * F_SETSIG can change ->signum lockless in parallel, make
 570          * sure we read it once and use the same value throughout.
 571          */
 572         int signum = ACCESS_ONCE(fown->signum);
 573
 574         if (!sigio_perm(p, fown, signum))
 575                 return;
 576
 577         switch (signum) {
 578                 siginfo_t si;
 579                 default:
 580                         /* Queue a rt signal with the appropriate fd as its
 581                            value.  We use SI_SIGIO as the source, not
 582                            SI_KERNEL, since kernel signals always get
 583                            delivered even if we can't queue.  Failure to
 584                            queue in this case _should_ be reported; we fall
 585                            back to SIGIO in that case. --sct */
 586                         si.si_signo = signum;
 587                         si.si_errno = 0;
 588                         si.si_code  = reason;
 589                         /* Make sure we are called with one of the POLL_*
 590                            reasons, otherwise we could leak kernel stack into
 591                            userspace.  */
 592                         BUG_ON((reason & __SI_MASK) != __SI_POLL);
 593                         if (reason - POLL_IN >= NSIGPOLL)
 594                                 si.si_band  = ~0L;
 595                         else
 596                                 si.si_band = band_table[reason - POLL_IN];
 597                         si.si_fd    = fd;
 598                         if (!do_send_sig_info(signum, &si, p, group))
 599                                 break;
 600                 /* fall-through: fall back on the old plain SIGIO signal */
 601                 case 0:
 602                         do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
 603         }
 604 }
 605
 606 void send_sigio(struct fown_struct *fown, int fd, int band)
 607 {
 608         struct task_struct *p;
 609         enum pid_type type;
 610         struct pid *pid;
 611         int group = 1;
 612
 613         read_lock(&fown->lock);
 614
 615         type = fown->pid_type;
 616         if (type == PIDTYPE_MAX) {
 617                 group = 0;
 618                 type = PIDTYPE_PID;
 619         }
 620
 621         pid = fown->pid;
 622         if (!pid)
 623                 goto out_unlock_fown;
 624
 625         read_lock(&tasklist_lock);
 626         do_each_pid_task(pid, type, p) {
 627                 send_sigio_to_task(p, fown, fd, band, group);
 628         } while_each_pid_task(pid, type, p);
 629         read_unlock(&tasklist_lock);
 630  out_unlock_fown:
 631         read_unlock(&fown->lock);
 632 }
 633
 634 static void send_sigurg_to_task(struct task_struct *p,
 635                                 struct fown_struct *fown, int group)
 636 {
 637         if (sigio_perm(p, fown, SIGURG))
 638                 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
 639 }
 640
 641 int send_sigurg(struct fown_struct *fown)
 642 {
 643         struct task_struct *p;
 644         enum pid_type type;
 645         struct pid *pid;
 646         int group = 1;
 647         int ret = 0;
 648
 649         read_lock(&fown->lock);
 650
 651         type = fown->pid_type;
 652         if (type == PIDTYPE_MAX) {
 653                 group = 0;
 654                 type = PIDTYPE_PID;
 655         }
 656
 657         pid = fown->pid;
 658         if (!pid)
 659                 goto out_unlock_fown;
 660
 661         ret = 1;
 662
 663         read_lock(&tasklist_lock);
 664         do_each_pid_task(pid, type, p) {
 665                 send_sigurg_to_task(p, fown, group);
 666         } while_each_pid_task(pid, type, p);
 667         read_unlock(&tasklist_lock);
 668  out_unlock_fown:
 669         read_unlock(&fown->lock);
 670         return ret;
 671 }
 672
 673 static DEFINE_SPINLOCK(fasync_lock);
 674 static struct kmem_cache *fasync_cache __read_mostly;
 675
 676 static void fasync_free_rcu(struct rcu_head *head)
 677 {
 678         kmem_cache_free(fasync_cache,
 679                         container_of(head, struct fasync_struct, fa_rcu));
 680 }
 681
 682 /*
 683  * Remove a fasync entry. If successfully removed, return
 684  * positive and clear the FASYNC flag. If no entry exists,
 685  * do nothing and return 0.
 686  *
 687  * NOTE! It is very important that the FASYNC flag always
 688  * match the state "is the filp on a fasync list".
 689  *
 690  */
 691 int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
 692 {
 693         struct fasync_struct *fa, **fp;
 694         int result = 0;
 695
 696         spin_lock(&filp->f_lock);
 697         spin_lock(&fasync_lock);
 698         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 699                 if (fa->fa_file != filp)
 700                         continue;
 701
 702                 spin_lock_irq(&fa->fa_lock);
 703                 fa->fa_file = NULL;
 704                 spin_unlock_irq(&fa->fa_lock);
 705
 706                 *fp = fa->fa_next;
 707                 call_rcu(&fa->fa_rcu, fasync_free_rcu);
 708                 filp->f_flags &= ~FASYNC;
 709                 result = 1;
 710                 break;
 711         }
 712         spin_unlock(&fasync_lock);
 713         spin_unlock(&filp->f_lock);
 714         return result;
 715 }
 716
 717 struct fasync_struct *fasync_alloc(void)
 718 {
 719         return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
 720 }
 721
 722 /*
 723  * NOTE! This can be used only for unused fasync entries:
 724  * entries that actually got inserted on the fasync list
 725  * need to be released by rcu - see fasync_remove_entry.
 726  */
 727 void fasync_free(struct fasync_struct *new)
 728 {
 729         kmem_cache_free(fasync_cache, new);
 730 }
 731
 732 /*
 733  * Insert a new entry into the fasync list.  Return the pointer to the
 734  * old one if we didn't use the new one.
 735  *
 736  * NOTE! It is very important that the FASYNC flag always
 737  * match the state "is the filp on a fasync list".
 738  */
 739 struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
 740 {
 741         struct fasync_struct *fa, **fp;
 742
 743         spin_lock(&filp->f_lock);
 744         spin_lock(&fasync_lock);
 745         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 746                 if (fa->fa_file != filp)
 747                         continue;
 748
 749                 spin_lock_irq(&fa->fa_lock);
 750                 fa->fa_fd = fd;
 751                 spin_unlock_irq(&fa->fa_lock);
 752                 goto out;
 753         }
 754
 755         spin_lock_init(&new->fa_lock);
 756         new->magic = FASYNC_MAGIC;
 757         new->fa_file = filp;
 758         new->fa_fd = fd;
 759         new->fa_next = *fapp;
 760         rcu_assign_pointer(*fapp, new);
 761         filp->f_flags |= FASYNC;
 762
 763 out:
 764         spin_unlock(&fasync_lock);
 765         spin_unlock(&filp->f_lock);
 766         return fa;
 767 }
 768
 769 /*
 770  * Add a fasync entry. Return negative on error, positive if
 771  * added, and zero if did nothing but change an existing one.
 772  */
 773 static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
 774 {
 775         struct fasync_struct *new;
 776
 777         new = fasync_alloc();
 778         if (!new)
 779                 return -ENOMEM;
 780
 781         /*
 782          * fasync_insert_entry() returns the old (update) entry if
 783          * it existed.
 784          *
 785          * So free the (unused) new entry and return 0 to let the
 786          * caller know that we didn't add any new fasync entries.
 787          */
 788         if (fasync_insert_entry(fd, filp, fapp, new)) {
 789                 fasync_free(new);
 790                 return 0;
 791         }
 792
 793         return 1;
 794 }
 795
 796 /*
 797  * fasync_helper() is used by almost all character device drivers
 798  * to set up the fasync queue, and for regular files by the file
 799  * lease code. It returns negative on error, 0 if it did no changes
 800  * and positive if it added/deleted the entry.
 801  */
 802 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
 803 {
 804         if (!on)
 805                 return fasync_remove_entry(filp, fapp);
 806         return fasync_add_entry(fd, filp, fapp);
 807 }
 808
 809 EXPORT_SYMBOL(fasync_helper);
 810
 811 /*
 812  * rcu_read_lock() is held
 813  */
 814 static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
 815 {
 816         while (fa) {
 817                 struct fown_struct *fown;
 818                 unsigned long flags;
 819
 820                 if (fa->magic != FASYNC_MAGIC) {
 821                         printk(KERN_ERR "kill_fasync: bad magic number in "
 822                                "fasync_struct!\n");
 823                         return;
 824                 }
 825                 spin_lock_irqsave(&fa->fa_lock, flags);
 826                 if (fa->fa_file) {
 827                         fown = &fa->fa_file->f_owner;
 828                         /* Don't send SIGURG to processes which have not set a
 829                            queued signum: SIGURG has its own default signalling
 830                            mechanism. */
 831                         if (!(sig == SIGURG && fown->signum == 0))
 832                                 send_sigio(fown, fa->fa_fd, band);
 833                 }
 834                 spin_unlock_irqrestore(&fa->fa_lock, flags);
 835                 fa = rcu_dereference(fa->fa_next);
 836         }
 837 }
 838
 839 void kill_fasync(struct fasync_struct **fp, int sig, int band)
 840 {
 841         /* First a quick test without locking: usually
 842          * the list is empty.
 843          */
 844         if (*fp) {
 845                 rcu_read_lock();
 846                 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
 847                 rcu_read_unlock();
 848         }
 849 }
 850 EXPORT_SYMBOL(kill_fasync);
 851
 852 static int __init fcntl_init(void)
 853 {
 854         /*
 855          * Please add new bits here to ensure allocation uniqueness.
 856          * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
 857          * is defined as O_NONBLOCK on some platforms and not on others.
 858          */
 859         BUILD_BUG_ON(19 - 1 /* for O_RDONLY being 0 */ != HWEIGHT32(
 860                 O_RDONLY        | O_WRONLY      | O_RDWR        |
 861                 O_CREAT         | O_EXCL        | O_NOCTTY      |
 862                 O_TRUNC         | O_APPEND      | /* O_NONBLOCK | */
 863                 __O_SYNC        | O_DSYNC       | FASYNC        |
 864                 O_DIRECT        | O_LARGEFILE   | O_DIRECTORY   |
 865                 O_NOFOLLOW      | O_NOATIME     | O_CLOEXEC     |
 866                 __FMODE_EXEC    | O_PATH
 867                 ));
 868
 869         fasync_cache = kmem_cache_create("fasync_cache",
 870                 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
 871         return 0;
 872 }
 873
 874 module_init(fcntl_init)