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/sched/task.h>
  11 #include <linux/fs.h>
  12 #include <linux/file.h>
  13 #include <linux/fdtable.h>
  14 #include <linux/capability.h>
  15 #include <linux/dnotify.h>
  16 #include <linux/slab.h>
  17 #include <linux/module.h>
  18 #include <linux/pipe_fs_i.h>
  19 #include <linux/security.h>
  20 #include <linux/ptrace.h>
  21 #include <linux/signal.h>
  22 #include <linux/rcupdate.h>
  23 #include <linux/pid_namespace.h>
  24 #include <linux/user_namespace.h>
  25 #include <linux/shmem_fs.h>
  26 #include <linux/compat.h>
  27
  28 #include <asm/poll.h>
  29 #include <asm/siginfo.h>
  30 #include <linux/uaccess.h>
  31
  32 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
  33
  34 static int setfl(int fd, struct file * filp, unsigned long arg)
  35 {
  36         struct inode * inode = file_inode(filp);
  37         int error = 0;
  38
  39         /*
  40          * O_APPEND cannot be cleared if the file is marked as append-only
  41          * and the file is open for write.
  42          */
  43         if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
  44                 return -EPERM;
  45
  46         /* O_NOATIME can only be set by the owner or superuser */
  47         if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
  48                 if (!inode_owner_or_capable(inode))
  49                         return -EPERM;
  50
  51         /* required for strict SunOS emulation */
  52         if (O_NONBLOCK != O_NDELAY)
  53                if (arg & O_NDELAY)
  54                    arg |= O_NONBLOCK;
  55
  56         /* Pipe packetized mode is controlled by O_DIRECT flag */
  57         if (!S_ISFIFO(inode->i_mode) && (arg & O_DIRECT)) {
  58                 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
  59                         !filp->f_mapping->a_ops->direct_IO)
  60                                 return -EINVAL;
  61         }
  62
  63         if (filp->f_op->check_flags)
  64                 error = filp->f_op->check_flags(arg);
  65         if (error)
  66                 return error;
  67
  68         /*
  69          * ->fasync() is responsible for setting the FASYNC bit.
  70          */
  71         if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op->fasync) {
  72                 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
  73                 if (error < 0)
  74                         goto out;
  75                 if (error > 0)
  76                         error = 0;
  77         }
  78         spin_lock(&filp->f_lock);
  79         filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
  80         spin_unlock(&filp->f_lock);
  81
  82  out:
  83         return error;
  84 }
  85
  86 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
  87                      int force)
  88 {
  89         write_lock_irq(&filp->f_owner.lock);
  90         if (force || !filp->f_owner.pid) {
  91                 put_pid(filp->f_owner.pid);
  92                 filp->f_owner.pid = get_pid(pid);
  93                 filp->f_owner.pid_type = type;
  94
  95                 if (pid) {
  96                         const struct cred *cred = current_cred();
  97                         filp->f_owner.uid = cred->uid;
  98                         filp->f_owner.euid = cred->euid;
  99                 }
 100         }
 101         write_unlock_irq(&filp->f_owner.lock);
 102 }
 103
 104 void __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
 105                 int force)
 106 {
 107         security_file_set_fowner(filp);
 108         f_modown(filp, pid, type, force);
 109 }
 110 EXPORT_SYMBOL(__f_setown);
 111
 112 void f_setown(struct file *filp, unsigned long arg, int force)
 113 {
 114         enum pid_type type;
 115         struct pid *pid;
 116         int who = arg;
 117         type = PIDTYPE_PID;
 118         if (who < 0) {
 119                 type = PIDTYPE_PGID;
 120                 who = -who;
 121         }
 122         rcu_read_lock();
 123         pid = find_vpid(who);
 124         __f_setown(filp, pid, type, force);
 125         rcu_read_unlock();
 126 }
 127 EXPORT_SYMBOL(f_setown);
 128
 129 void f_delown(struct file *filp)
 130 {
 131         f_modown(filp, NULL, PIDTYPE_PID, 1);
 132 }
 133
 134 pid_t f_getown(struct file *filp)
 135 {
 136         pid_t pid;
 137         read_lock(&filp->f_owner.lock);
 138         pid = pid_vnr(filp->f_owner.pid);
 139         if (filp->f_owner.pid_type == PIDTYPE_PGID)
 140                 pid = -pid;
 141         read_unlock(&filp->f_owner.lock);
 142         return pid;
 143 }
 144
 145 static int f_setown_ex(struct file *filp, unsigned long arg)
 146 {
 147         struct f_owner_ex __user *owner_p = (void __user *)arg;
 148         struct f_owner_ex owner;
 149         struct pid *pid;
 150         int type;
 151         int ret;
 152
 153         ret = copy_from_user(&owner, owner_p, sizeof(owner));
 154         if (ret)
 155                 return -EFAULT;
 156
 157         switch (owner.type) {
 158         case F_OWNER_TID:
 159                 type = PIDTYPE_MAX;
 160                 break;
 161
 162         case F_OWNER_PID:
 163                 type = PIDTYPE_PID;
 164                 break;
 165
 166         case F_OWNER_PGRP:
 167                 type = PIDTYPE_PGID;
 168                 break;
 169
 170         default:
 171                 return -EINVAL;
 172         }
 173
 174         rcu_read_lock();
 175         pid = find_vpid(owner.pid);
 176         if (owner.pid && !pid)
 177                 ret = -ESRCH;
 178         else
 179                  __f_setown(filp, pid, type, 1);
 180         rcu_read_unlock();
 181
 182         return ret;
 183 }
 184
 185 static int f_getown_ex(struct file *filp, unsigned long arg)
 186 {
 187         struct f_owner_ex __user *owner_p = (void __user *)arg;
 188         struct f_owner_ex owner;
 189         int ret = 0;
 190
 191         read_lock(&filp->f_owner.lock);
 192         owner.pid = pid_vnr(filp->f_owner.pid);
 193         switch (filp->f_owner.pid_type) {
 194         case PIDTYPE_MAX:
 195                 owner.type = F_OWNER_TID;
 196                 break;
 197
 198         case PIDTYPE_PID:
 199                 owner.type = F_OWNER_PID;
 200                 break;
 201
 202         case PIDTYPE_PGID:
 203                 owner.type = F_OWNER_PGRP;
 204                 break;
 205
 206         default:
 207                 WARN_ON(1);
 208                 ret = -EINVAL;
 209                 break;
 210         }
 211         read_unlock(&filp->f_owner.lock);
 212
 213         if (!ret) {
 214                 ret = copy_to_user(owner_p, &owner, sizeof(owner));
 215                 if (ret)
 216                         ret = -EFAULT;
 217         }
 218         return ret;
 219 }
 220
 221 #ifdef CONFIG_CHECKPOINT_RESTORE
 222 static int f_getowner_uids(struct file *filp, unsigned long arg)
 223 {
 224         struct user_namespace *user_ns = current_user_ns();
 225         uid_t __user *dst = (void __user *)arg;
 226         uid_t src[2];
 227         int err;
 228
 229         read_lock(&filp->f_owner.lock);
 230         src[0] = from_kuid(user_ns, filp->f_owner.uid);
 231         src[1] = from_kuid(user_ns, filp->f_owner.euid);
 232         read_unlock(&filp->f_owner.lock);
 233
 234         err  = put_user(src[0], &dst[0]);
 235         err |= put_user(src[1], &dst[1]);
 236
 237         return err;
 238 }
 239 #else
 240 static int f_getowner_uids(struct file *filp, unsigned long arg)
 241 {
 242         return -EINVAL;
 243 }
 244 #endif
 245
 246 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
 247                 struct file *filp)
 248 {
 249         long err = -EINVAL;
 250
 251         switch (cmd) {
 252         case F_DUPFD:
 253                 err = f_dupfd(arg, filp, 0);
 254                 break;
 255         case F_DUPFD_CLOEXEC:
 256                 err = f_dupfd(arg, filp, O_CLOEXEC);
 257                 break;
 258         case F_GETFD:
 259                 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
 260                 break;
 261         case F_SETFD:
 262                 err = 0;
 263                 set_close_on_exec(fd, arg & FD_CLOEXEC);
 264                 break;
 265         case F_GETFL:
 266                 err = filp->f_flags;
 267                 break;
 268         case F_SETFL:
 269                 err = setfl(fd, filp, arg);
 270                 break;
 271 #if BITS_PER_LONG != 32
 272         /* 32-bit arches must use fcntl64() */
 273         case F_OFD_GETLK:
 274 #endif
 275         case F_GETLK:
 276                 err = fcntl_getlk(filp, cmd, (struct flock __user *) arg);
 277                 break;
 278 #if BITS_PER_LONG != 32
 279         /* 32-bit arches must use fcntl64() */
 280         case F_OFD_SETLK:
 281         case F_OFD_SETLKW:
 282 #endif
 283                 /* Fallthrough */
 284         case F_SETLK:
 285         case F_SETLKW:
 286                 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
 287                 break;
 288         case F_GETOWN:
 289                 /*
 290                  * XXX If f_owner is a process group, the
 291                  * negative return value will get converted
 292                  * into an error.  Oops.  If we keep the
 293                  * current syscall conventions, the only way
 294                  * to fix this will be in libc.
 295                  */
 296                 err = f_getown(filp);
 297                 force_successful_syscall_return();
 298                 break;
 299         case F_SETOWN:
 300                 f_setown(filp, arg, 1);
 301                 err = 0;
 302                 break;
 303         case F_GETOWN_EX:
 304                 err = f_getown_ex(filp, arg);
 305                 break;
 306         case F_SETOWN_EX:
 307                 err = f_setown_ex(filp, arg);
 308                 break;
 309         case F_GETOWNER_UIDS:
 310                 err = f_getowner_uids(filp, arg);
 311                 break;
 312         case F_GETSIG:
 313                 err = filp->f_owner.signum;
 314                 break;
 315         case F_SETSIG:
 316                 /* arg == 0 restores default behaviour. */
 317                 if (!valid_signal(arg)) {
 318                         break;
 319                 }
 320                 err = 0;
 321                 filp->f_owner.signum = arg;
 322                 break;
 323         case F_GETLEASE:
 324                 err = fcntl_getlease(filp);
 325                 break;
 326         case F_SETLEASE:
 327                 err = fcntl_setlease(fd, filp, arg);
 328                 break;
 329         case F_NOTIFY:
 330                 err = fcntl_dirnotify(fd, filp, arg);
 331                 break;
 332         case F_SETPIPE_SZ:
 333         case F_GETPIPE_SZ:
 334                 err = pipe_fcntl(filp, cmd, arg);
 335                 break;
 336         case F_ADD_SEALS:
 337         case F_GET_SEALS:
 338                 err = shmem_fcntl(filp, cmd, arg);
 339                 break;
 340         default:
 341                 break;
 342         }
 343         return err;
 344 }
 345
 346 static int check_fcntl_cmd(unsigned cmd)
 347 {
 348         switch (cmd) {
 349         case F_DUPFD:
 350         case F_DUPFD_CLOEXEC:
 351         case F_GETFD:
 352         case F_SETFD:
 353         case F_GETFL:
 354                 return 1;
 355         }
 356         return 0;
 357 }
 358
 359 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
 360 {
 361         struct fd f = fdget_raw(fd);
 362         long err = -EBADF;
 363
 364         if (!f.file)
 365                 goto out;
 366
 367         if (unlikely(f.file->f_mode & FMODE_PATH)) {
 368                 if (!check_fcntl_cmd(cmd))
 369                         goto out1;
 370         }
 371
 372         err = security_file_fcntl(f.file, cmd, arg);
 373         if (!err)
 374                 err = do_fcntl(fd, cmd, arg, f.file);
 375
 376 out1:
 377         fdput(f);
 378 out:
 379         return err;
 380 }
 381
 382 #if BITS_PER_LONG == 32
 383 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
 384                 unsigned long, arg)
 385 {
 386         struct fd f = fdget_raw(fd);
 387         long err = -EBADF;
 388
 389         if (!f.file)
 390                 goto out;
 391
 392         if (unlikely(f.file->f_mode & FMODE_PATH)) {
 393                 if (!check_fcntl_cmd(cmd))
 394                         goto out1;
 395         }
 396
 397         err = security_file_fcntl(f.file, cmd, arg);
 398         if (err)
 399                 goto out1;
 400
 401         switch (cmd) {
 402         case F_GETLK64:
 403         case F_OFD_GETLK:
 404                 err = fcntl_getlk64(f.file, cmd, (struct flock64 __user *) arg);
 405                 break;
 406         case F_SETLK64:
 407         case F_SETLKW64:
 408         case F_OFD_SETLK:
 409         case F_OFD_SETLKW:
 410                 err = fcntl_setlk64(fd, f.file, cmd,
 411                                 (struct flock64 __user *) arg);
 412                 break;
 413         default:
 414                 err = do_fcntl(fd, cmd, arg, f.file);
 415                 break;
 416         }
 417 out1:
 418         fdput(f);
 419 out:
 420         return err;
 421 }
 422 #endif
 423
 424 #ifdef CONFIG_COMPAT
 425 static int get_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
 426 {
 427         if (!access_ok(VERIFY_READ, ufl, sizeof(*ufl)) ||
 428             __get_user(kfl->l_type, &ufl->l_type) ||
 429             __get_user(kfl->l_whence, &ufl->l_whence) ||
 430             __get_user(kfl->l_start, &ufl->l_start) ||
 431             __get_user(kfl->l_len, &ufl->l_len) ||
 432             __get_user(kfl->l_pid, &ufl->l_pid))
 433                 return -EFAULT;
 434         return 0;
 435 }
 436
 437 static int put_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
 438 {
 439         if (!access_ok(VERIFY_WRITE, ufl, sizeof(*ufl)) ||
 440             __put_user(kfl->l_type, &ufl->l_type) ||
 441             __put_user(kfl->l_whence, &ufl->l_whence) ||
 442             __put_user(kfl->l_start, &ufl->l_start) ||
 443             __put_user(kfl->l_len, &ufl->l_len) ||
 444             __put_user(kfl->l_pid, &ufl->l_pid))
 445                 return -EFAULT;
 446         return 0;
 447 }
 448
 449 #ifndef HAVE_ARCH_GET_COMPAT_FLOCK64
 450 static int get_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
 451 {
 452         if (!access_ok(VERIFY_READ, ufl, sizeof(*ufl)) ||
 453             __get_user(kfl->l_type, &ufl->l_type) ||
 454             __get_user(kfl->l_whence, &ufl->l_whence) ||
 455             __get_user(kfl->l_start, &ufl->l_start) ||
 456             __get_user(kfl->l_len, &ufl->l_len) ||
 457             __get_user(kfl->l_pid, &ufl->l_pid))
 458                 return -EFAULT;
 459         return 0;
 460 }
 461 #endif
 462
 463 #ifndef HAVE_ARCH_PUT_COMPAT_FLOCK64
 464 static int put_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
 465 {
 466         if (!access_ok(VERIFY_WRITE, ufl, sizeof(*ufl)) ||
 467             __put_user(kfl->l_type, &ufl->l_type) ||
 468             __put_user(kfl->l_whence, &ufl->l_whence) ||
 469             __put_user(kfl->l_start, &ufl->l_start) ||
 470             __put_user(kfl->l_len, &ufl->l_len) ||
 471             __put_user(kfl->l_pid, &ufl->l_pid))
 472                 return -EFAULT;
 473         return 0;
 474 }
 475 #endif
 476
 477 static unsigned int
 478 convert_fcntl_cmd(unsigned int cmd)
 479 {
 480         switch (cmd) {
 481         case F_GETLK64:
 482                 return F_GETLK;
 483         case F_SETLK64:
 484                 return F_SETLK;
 485         case F_SETLKW64:
 486                 return F_SETLKW;
 487         }
 488
 489         return cmd;
 490 }
 491
 492 COMPAT_SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
 493                        compat_ulong_t, arg)
 494 {
 495         mm_segment_t old_fs;
 496         struct flock f;
 497         long ret;
 498         unsigned int conv_cmd;
 499
 500         switch (cmd) {
 501         case F_GETLK:
 502         case F_SETLK:
 503         case F_SETLKW:
 504                 ret = get_compat_flock(&f, compat_ptr(arg));
 505                 if (ret != 0)
 506                         break;
 507                 old_fs = get_fs();
 508                 set_fs(KERNEL_DS);
 509                 ret = sys_fcntl(fd, cmd, (unsigned long)&f);
 510                 set_fs(old_fs);
 511                 if (cmd == F_GETLK && ret == 0) {
 512                         /* GETLK was successful and we need to return the data...
 513                          * but it needs to fit in the compat structure.
 514                          * l_start shouldn't be too big, unless the original
 515                          * start + end is greater than COMPAT_OFF_T_MAX, in which
 516                          * case the app was asking for trouble, so we return
 517                          * -EOVERFLOW in that case.
 518                          * l_len could be too big, in which case we just truncate it,
 519                          * and only allow the app to see that part of the conflicting
 520                          * lock that might make sense to it anyway
 521                          */
 522
 523                         if (f.l_start > COMPAT_OFF_T_MAX)
 524                                 ret = -EOVERFLOW;
 525                         if (f.l_len > COMPAT_OFF_T_MAX)
 526                                 f.l_len = COMPAT_OFF_T_MAX;
 527                         if (ret == 0)
 528                                 ret = put_compat_flock(&f, compat_ptr(arg));
 529                 }
 530                 break;
 531
 532         case F_GETLK64:
 533         case F_SETLK64:
 534         case F_SETLKW64:
 535         case F_OFD_GETLK:
 536         case F_OFD_SETLK:
 537         case F_OFD_SETLKW:
 538                 ret = get_compat_flock64(&f, compat_ptr(arg));
 539                 if (ret != 0)
 540                         break;
 541                 old_fs = get_fs();
 542                 set_fs(KERNEL_DS);
 543                 conv_cmd = convert_fcntl_cmd(cmd);
 544                 ret = sys_fcntl(fd, conv_cmd, (unsigned long)&f);
 545                 set_fs(old_fs);
 546                 if ((conv_cmd == F_GETLK || conv_cmd == F_OFD_GETLK) && ret == 0) {
 547                         /* need to return lock information - see above for commentary */
 548                         if (f.l_start > COMPAT_LOFF_T_MAX)
 549                                 ret = -EOVERFLOW;
 550                         if (f.l_len > COMPAT_LOFF_T_MAX)
 551                                 f.l_len = COMPAT_LOFF_T_MAX;
 552                         if (ret == 0)
 553                                 ret = put_compat_flock64(&f, compat_ptr(arg));
 554                 }
 555                 break;
 556
 557         default:
 558                 ret = sys_fcntl(fd, cmd, arg);
 559                 break;
 560         }
 561         return ret;
 562 }
 563
 564 COMPAT_SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd,
 565                        compat_ulong_t, arg)
 566 {
 567         switch (cmd) {
 568         case F_GETLK64:
 569         case F_SETLK64:
 570         case F_SETLKW64:
 571         case F_OFD_GETLK:
 572         case F_OFD_SETLK:
 573         case F_OFD_SETLKW:
 574                 return -EINVAL;
 575         }
 576         return compat_sys_fcntl64(fd, cmd, arg);
 577 }
 578 #endif
 579
 580 /* Table to convert sigio signal codes into poll band bitmaps */
 581
 582 static const long band_table[NSIGPOLL] = {
 583         POLLIN | POLLRDNORM,                    /* POLL_IN */
 584         POLLOUT | POLLWRNORM | POLLWRBAND,      /* POLL_OUT */
 585         POLLIN | POLLRDNORM | POLLMSG,          /* POLL_MSG */
 586         POLLERR,                                /* POLL_ERR */
 587         POLLPRI | POLLRDBAND,                   /* POLL_PRI */
 588         POLLHUP | POLLERR                       /* POLL_HUP */
 589 };
 590
 591 static inline int sigio_perm(struct task_struct *p,
 592                              struct fown_struct *fown, int sig)
 593 {
 594         const struct cred *cred;
 595         int ret;
 596
 597         rcu_read_lock();
 598         cred = __task_cred(p);
 599         ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
 600                 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
 601                 uid_eq(fown->uid,  cred->suid) || uid_eq(fown->uid,  cred->uid)) &&
 602                !security_file_send_sigiotask(p, fown, sig));
 603         rcu_read_unlock();
 604         return ret;
 605 }
 606
 607 static void send_sigio_to_task(struct task_struct *p,
 608                                struct fown_struct *fown,
 609                                int fd, int reason, int group)
 610 {
 611         /*
 612          * F_SETSIG can change ->signum lockless in parallel, make
 613          * sure we read it once and use the same value throughout.
 614          */
 615         int signum = ACCESS_ONCE(fown->signum);
 616
 617         if (!sigio_perm(p, fown, signum))
 618                 return;
 619
 620         switch (signum) {
 621                 siginfo_t si;
 622                 default:
 623                         /* Queue a rt signal with the appropriate fd as its
 624                            value.  We use SI_SIGIO as the source, not
 625                            SI_KERNEL, since kernel signals always get
 626                            delivered even if we can't queue.  Failure to
 627                            queue in this case _should_ be reported; we fall
 628                            back to SIGIO in that case. --sct */
 629                         si.si_signo = signum;
 630                         si.si_errno = 0;
 631                         si.si_code  = reason;
 632                         /* Make sure we are called with one of the POLL_*
 633                            reasons, otherwise we could leak kernel stack into
 634                            userspace.  */
 635                         BUG_ON((reason & __SI_MASK) != __SI_POLL);
 636                         if (reason - POLL_IN >= NSIGPOLL)
 637                                 si.si_band  = ~0L;
 638                         else
 639                                 si.si_band = band_table[reason - POLL_IN];
 640                         si.si_fd    = fd;
 641                         if (!do_send_sig_info(signum, &si, p, group))
 642                                 break;
 643                 /* fall-through: fall back on the old plain SIGIO signal */
 644                 case 0:
 645                         do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
 646         }
 647 }
 648
 649 void send_sigio(struct fown_struct *fown, int fd, int band)
 650 {
 651         struct task_struct *p;
 652         enum pid_type type;
 653         struct pid *pid;
 654         int group = 1;
 655
 656         read_lock(&fown->lock);
 657
 658         type = fown->pid_type;
 659         if (type == PIDTYPE_MAX) {
 660                 group = 0;
 661                 type = PIDTYPE_PID;
 662         }
 663
 664         pid = fown->pid;
 665         if (!pid)
 666                 goto out_unlock_fown;
 667
 668         read_lock(&tasklist_lock);
 669         do_each_pid_task(pid, type, p) {
 670                 send_sigio_to_task(p, fown, fd, band, group);
 671         } while_each_pid_task(pid, type, p);
 672         read_unlock(&tasklist_lock);
 673  out_unlock_fown:
 674         read_unlock(&fown->lock);
 675 }
 676
 677 static void send_sigurg_to_task(struct task_struct *p,
 678                                 struct fown_struct *fown, int group)
 679 {
 680         if (sigio_perm(p, fown, SIGURG))
 681                 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
 682 }
 683
 684 int send_sigurg(struct fown_struct *fown)
 685 {
 686         struct task_struct *p;
 687         enum pid_type type;
 688         struct pid *pid;
 689         int group = 1;
 690         int ret = 0;
 691
 692         read_lock(&fown->lock);
 693
 694         type = fown->pid_type;
 695         if (type == PIDTYPE_MAX) {
 696                 group = 0;
 697                 type = PIDTYPE_PID;
 698         }
 699
 700         pid = fown->pid;
 701         if (!pid)
 702                 goto out_unlock_fown;
 703
 704         ret = 1;
 705
 706         read_lock(&tasklist_lock);
 707         do_each_pid_task(pid, type, p) {
 708                 send_sigurg_to_task(p, fown, group);
 709         } while_each_pid_task(pid, type, p);
 710         read_unlock(&tasklist_lock);
 711  out_unlock_fown:
 712         read_unlock(&fown->lock);
 713         return ret;
 714 }
 715
 716 static DEFINE_SPINLOCK(fasync_lock);
 717 static struct kmem_cache *fasync_cache __read_mostly;
 718
 719 static void fasync_free_rcu(struct rcu_head *head)
 720 {
 721         kmem_cache_free(fasync_cache,
 722                         container_of(head, struct fasync_struct, fa_rcu));
 723 }
 724
 725 /*
 726  * Remove a fasync entry. If successfully removed, return
 727  * positive and clear the FASYNC flag. If no entry exists,
 728  * do nothing and return 0.
 729  *
 730  * NOTE! It is very important that the FASYNC flag always
 731  * match the state "is the filp on a fasync list".
 732  *
 733  */
 734 int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
 735 {
 736         struct fasync_struct *fa, **fp;
 737         int result = 0;
 738
 739         spin_lock(&filp->f_lock);
 740         spin_lock(&fasync_lock);
 741         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 742                 if (fa->fa_file != filp)
 743                         continue;
 744
 745                 spin_lock_irq(&fa->fa_lock);
 746                 fa->fa_file = NULL;
 747                 spin_unlock_irq(&fa->fa_lock);
 748
 749                 *fp = fa->fa_next;
 750                 call_rcu(&fa->fa_rcu, fasync_free_rcu);
 751                 filp->f_flags &= ~FASYNC;
 752                 result = 1;
 753                 break;
 754         }
 755         spin_unlock(&fasync_lock);
 756         spin_unlock(&filp->f_lock);
 757         return result;
 758 }
 759
 760 struct fasync_struct *fasync_alloc(void)
 761 {
 762         return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
 763 }
 764
 765 /*
 766  * NOTE! This can be used only for unused fasync entries:
 767  * entries that actually got inserted on the fasync list
 768  * need to be released by rcu - see fasync_remove_entry.
 769  */
 770 void fasync_free(struct fasync_struct *new)
 771 {
 772         kmem_cache_free(fasync_cache, new);
 773 }
 774
 775 /*
 776  * Insert a new entry into the fasync list.  Return the pointer to the
 777  * old one if we didn't use the new one.
 778  *
 779  * NOTE! It is very important that the FASYNC flag always
 780  * match the state "is the filp on a fasync list".
 781  */
 782 struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
 783 {
 784         struct fasync_struct *fa, **fp;
 785
 786         spin_lock(&filp->f_lock);
 787         spin_lock(&fasync_lock);
 788         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 789                 if (fa->fa_file != filp)
 790                         continue;
 791
 792                 spin_lock_irq(&fa->fa_lock);
 793                 fa->fa_fd = fd;
 794                 spin_unlock_irq(&fa->fa_lock);
 795                 goto out;
 796         }
 797
 798         spin_lock_init(&new->fa_lock);
 799         new->magic = FASYNC_MAGIC;
 800         new->fa_file = filp;
 801         new->fa_fd = fd;
 802         new->fa_next = *fapp;
 803         rcu_assign_pointer(*fapp, new);
 804         filp->f_flags |= FASYNC;
 805
 806 out:
 807         spin_unlock(&fasync_lock);
 808         spin_unlock(&filp->f_lock);
 809         return fa;
 810 }
 811
 812 /*
 813  * Add a fasync entry. Return negative on error, positive if
 814  * added, and zero if did nothing but change an existing one.
 815  */
 816 static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
 817 {
 818         struct fasync_struct *new;
 819
 820         new = fasync_alloc();
 821         if (!new)
 822                 return -ENOMEM;
 823
 824         /*
 825          * fasync_insert_entry() returns the old (update) entry if
 826          * it existed.
 827          *
 828          * So free the (unused) new entry and return 0 to let the
 829          * caller know that we didn't add any new fasync entries.
 830          */
 831         if (fasync_insert_entry(fd, filp, fapp, new)) {
 832                 fasync_free(new);
 833                 return 0;
 834         }
 835
 836         return 1;
 837 }
 838
 839 /*
 840  * fasync_helper() is used by almost all character device drivers
 841  * to set up the fasync queue, and for regular files by the file
 842  * lease code. It returns negative on error, 0 if it did no changes
 843  * and positive if it added/deleted the entry.
 844  */
 845 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
 846 {
 847         if (!on)
 848                 return fasync_remove_entry(filp, fapp);
 849         return fasync_add_entry(fd, filp, fapp);
 850 }
 851
 852 EXPORT_SYMBOL(fasync_helper);
 853
 854 /*
 855  * rcu_read_lock() is held
 856  */
 857 static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
 858 {
 859         while (fa) {
 860                 struct fown_struct *fown;
 861                 unsigned long flags;
 862
 863                 if (fa->magic != FASYNC_MAGIC) {
 864                         printk(KERN_ERR "kill_fasync: bad magic number in "
 865                                "fasync_struct!\n");
 866                         return;
 867                 }
 868                 spin_lock_irqsave(&fa->fa_lock, flags);
 869                 if (fa->fa_file) {
 870                         fown = &fa->fa_file->f_owner;
 871                         /* Don't send SIGURG to processes which have not set a
 872                            queued signum: SIGURG has its own default signalling
 873                            mechanism. */
 874                         if (!(sig == SIGURG && fown->signum == 0))
 875                                 send_sigio(fown, fa->fa_fd, band);
 876                 }
 877                 spin_unlock_irqrestore(&fa->fa_lock, flags);
 878                 fa = rcu_dereference(fa->fa_next);
 879         }
 880 }
 881
 882 void kill_fasync(struct fasync_struct **fp, int sig, int band)
 883 {
 884         /* First a quick test without locking: usually
 885          * the list is empty.
 886          */
 887         if (*fp) {
 888                 rcu_read_lock();
 889                 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
 890                 rcu_read_unlock();
 891         }
 892 }
 893 EXPORT_SYMBOL(kill_fasync);
 894
 895 static int __init fcntl_init(void)
 896 {
 897         /*
 898          * Please add new bits here to ensure allocation uniqueness.
 899          * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
 900          * is defined as O_NONBLOCK on some platforms and not on others.
 901          */
 902         BUILD_BUG_ON(21 - 1 /* for O_RDONLY being 0 */ !=
 903                 HWEIGHT32(
 904                         (VALID_OPEN_FLAGS & ~(O_NONBLOCK | O_NDELAY)) |
 905                         __FMODE_EXEC | __FMODE_NONOTIFY));
 906
 907         fasync_cache = kmem_cache_create("fasync_cache",
 908                 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
 909         return 0;
 910 }
 911
 912 module_init(fcntl_init)