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 int 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         return 0;
 128 }
 129 EXPORT_SYMBOL(f_setown);
 130
 131 void f_delown(struct file *filp)
 132 {
 133         f_modown(filp, NULL, PIDTYPE_PID, 1);
 134 }
 135
 136 pid_t f_getown(struct file *filp)
 137 {
 138         pid_t pid;
 139         read_lock(&filp->f_owner.lock);
 140         pid = pid_vnr(filp->f_owner.pid);
 141         if (filp->f_owner.pid_type == PIDTYPE_PGID)
 142                 pid = -pid;
 143         read_unlock(&filp->f_owner.lock);
 144         return pid;
 145 }
 146
 147 static int f_setown_ex(struct file *filp, unsigned long arg)
 148 {
 149         struct f_owner_ex __user *owner_p = (void __user *)arg;
 150         struct f_owner_ex owner;
 151         struct pid *pid;
 152         int type;
 153         int ret;
 154
 155         ret = copy_from_user(&owner, owner_p, sizeof(owner));
 156         if (ret)
 157                 return -EFAULT;
 158
 159         switch (owner.type) {
 160         case F_OWNER_TID:
 161                 type = PIDTYPE_MAX;
 162                 break;
 163
 164         case F_OWNER_PID:
 165                 type = PIDTYPE_PID;
 166                 break;
 167
 168         case F_OWNER_PGRP:
 169                 type = PIDTYPE_PGID;
 170                 break;
 171
 172         default:
 173                 return -EINVAL;
 174         }
 175
 176         rcu_read_lock();
 177         pid = find_vpid(owner.pid);
 178         if (owner.pid && !pid)
 179                 ret = -ESRCH;
 180         else
 181                  __f_setown(filp, pid, type, 1);
 182         rcu_read_unlock();
 183
 184         return ret;
 185 }
 186
 187 static int f_getown_ex(struct file *filp, unsigned long arg)
 188 {
 189         struct f_owner_ex __user *owner_p = (void __user *)arg;
 190         struct f_owner_ex owner;
 191         int ret = 0;
 192
 193         read_lock(&filp->f_owner.lock);
 194         owner.pid = pid_vnr(filp->f_owner.pid);
 195         switch (filp->f_owner.pid_type) {
 196         case PIDTYPE_MAX:
 197                 owner.type = F_OWNER_TID;
 198                 break;
 199
 200         case PIDTYPE_PID:
 201                 owner.type = F_OWNER_PID;
 202                 break;
 203
 204         case PIDTYPE_PGID:
 205                 owner.type = F_OWNER_PGRP;
 206                 break;
 207
 208         default:
 209                 WARN_ON(1);
 210                 ret = -EINVAL;
 211                 break;
 212         }
 213         read_unlock(&filp->f_owner.lock);
 214
 215         if (!ret) {
 216                 ret = copy_to_user(owner_p, &owner, sizeof(owner));
 217                 if (ret)
 218                         ret = -EFAULT;
 219         }
 220         return ret;
 221 }
 222
 223 #ifdef CONFIG_CHECKPOINT_RESTORE
 224 static int f_getowner_uids(struct file *filp, unsigned long arg)
 225 {
 226         struct user_namespace *user_ns = current_user_ns();
 227         uid_t __user *dst = (void __user *)arg;
 228         uid_t src[2];
 229         int err;
 230
 231         read_lock(&filp->f_owner.lock);
 232         src[0] = from_kuid(user_ns, filp->f_owner.uid);
 233         src[1] = from_kuid(user_ns, filp->f_owner.euid);
 234         read_unlock(&filp->f_owner.lock);
 235
 236         err  = put_user(src[0], &dst[0]);
 237         err |= put_user(src[1], &dst[1]);
 238
 239         return err;
 240 }
 241 #else
 242 static int f_getowner_uids(struct file *filp, unsigned long arg)
 243 {
 244         return -EINVAL;
 245 }
 246 #endif
 247
 248 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
 249                 struct file *filp)
 250 {
 251         void __user *argp = (void __user *)arg;
 252         struct flock flock;
 253         long err = -EINVAL;
 254
 255         switch (cmd) {
 256         case F_DUPFD:
 257                 err = f_dupfd(arg, filp, 0);
 258                 break;
 259         case F_DUPFD_CLOEXEC:
 260                 err = f_dupfd(arg, filp, O_CLOEXEC);
 261                 break;
 262         case F_GETFD:
 263                 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
 264                 break;
 265         case F_SETFD:
 266                 err = 0;
 267                 set_close_on_exec(fd, arg & FD_CLOEXEC);
 268                 break;
 269         case F_GETFL:
 270                 err = filp->f_flags;
 271                 break;
 272         case F_SETFL:
 273                 err = setfl(fd, filp, arg);
 274                 break;
 275 #if BITS_PER_LONG != 32
 276         /* 32-bit arches must use fcntl64() */
 277         case F_OFD_GETLK:
 278 #endif
 279         case F_GETLK:
 280                 if (copy_from_user(&flock, argp, sizeof(flock)))
 281                         return -EFAULT;
 282                 err = fcntl_getlk(filp, cmd, &flock);
 283                 if (!err && copy_to_user(argp, &flock, sizeof(flock)))
 284                         return -EFAULT;
 285                 break;
 286 #if BITS_PER_LONG != 32
 287         /* 32-bit arches must use fcntl64() */
 288         case F_OFD_SETLK:
 289         case F_OFD_SETLKW:
 290 #endif
 291                 /* Fallthrough */
 292         case F_SETLK:
 293         case F_SETLKW:
 294                 if (copy_from_user(&flock, argp, sizeof(flock)))
 295                         return -EFAULT;
 296                 err = fcntl_setlk(fd, filp, cmd, &flock);
 297                 break;
 298         case F_GETOWN:
 299                 /*
 300                  * XXX If f_owner is a process group, the
 301                  * negative return value will get converted
 302                  * into an error.  Oops.  If we keep the
 303                  * current syscall conventions, the only way
 304                  * to fix this will be in libc.
 305                  */
 306                 err = f_getown(filp);
 307                 force_successful_syscall_return();
 308                 break;
 309         case F_SETOWN:
 310                 err = f_setown(filp, arg, 1);
 311                 break;
 312         case F_GETOWN_EX:
 313                 err = f_getown_ex(filp, arg);
 314                 break;
 315         case F_SETOWN_EX:
 316                 err = f_setown_ex(filp, arg);
 317                 break;
 318         case F_GETOWNER_UIDS:
 319                 err = f_getowner_uids(filp, arg);
 320                 break;
 321         case F_GETSIG:
 322                 err = filp->f_owner.signum;
 323                 break;
 324         case F_SETSIG:
 325                 /* arg == 0 restores default behaviour. */
 326                 if (!valid_signal(arg)) {
 327                         break;
 328                 }
 329                 err = 0;
 330                 filp->f_owner.signum = arg;
 331                 break;
 332         case F_GETLEASE:
 333                 err = fcntl_getlease(filp);
 334                 break;
 335         case F_SETLEASE:
 336                 err = fcntl_setlease(fd, filp, arg);
 337                 break;
 338         case F_NOTIFY:
 339                 err = fcntl_dirnotify(fd, filp, arg);
 340                 break;
 341         case F_SETPIPE_SZ:
 342         case F_GETPIPE_SZ:
 343                 err = pipe_fcntl(filp, cmd, arg);
 344                 break;
 345         case F_ADD_SEALS:
 346         case F_GET_SEALS:
 347                 err = shmem_fcntl(filp, cmd, arg);
 348                 break;
 349         default:
 350                 break;
 351         }
 352         return err;
 353 }
 354
 355 static int check_fcntl_cmd(unsigned cmd)
 356 {
 357         switch (cmd) {
 358         case F_DUPFD:
 359         case F_DUPFD_CLOEXEC:
 360         case F_GETFD:
 361         case F_SETFD:
 362         case F_GETFL:
 363                 return 1;
 364         }
 365         return 0;
 366 }
 367
 368 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
 369 {
 370         struct fd f = fdget_raw(fd);
 371         long err = -EBADF;
 372
 373         if (!f.file)
 374                 goto out;
 375
 376         if (unlikely(f.file->f_mode & FMODE_PATH)) {
 377                 if (!check_fcntl_cmd(cmd))
 378                         goto out1;
 379         }
 380
 381         err = security_file_fcntl(f.file, cmd, arg);
 382         if (!err)
 383                 err = do_fcntl(fd, cmd, arg, f.file);
 384
 385 out1:
 386         fdput(f);
 387 out:
 388         return err;
 389 }
 390
 391 #if BITS_PER_LONG == 32
 392 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
 393                 unsigned long, arg)
 394 {
 395         void __user *argp = (void __user *)arg;
 396         struct fd f = fdget_raw(fd);
 397         struct flock64 flock;
 398         long err = -EBADF;
 399
 400         if (!f.file)
 401                 goto out;
 402
 403         if (unlikely(f.file->f_mode & FMODE_PATH)) {
 404                 if (!check_fcntl_cmd(cmd))
 405                         goto out1;
 406         }
 407
 408         err = security_file_fcntl(f.file, cmd, arg);
 409         if (err)
 410                 goto out1;
 411
 412         switch (cmd) {
 413         case F_GETLK64:
 414         case F_OFD_GETLK:
 415                 err = -EFAULT;
 416                 if (copy_from_user(&flock, argp, sizeof(flock)))
 417                         break;
 418                 err = fcntl_getlk64(f.file, cmd, &flock);
 419                 if (!err && copy_to_user(argp, &flock, sizeof(flock)))
 420                         err = -EFAULT;
 421                 break;
 422         case F_SETLK64:
 423         case F_SETLKW64:
 424         case F_OFD_SETLK:
 425         case F_OFD_SETLKW:
 426                 err = -EFAULT;
 427                 if (copy_from_user(&flock, argp, sizeof(flock)))
 428                         break;
 429                 err = fcntl_setlk64(fd, f.file, cmd, &flock);
 430                 break;
 431         default:
 432                 err = do_fcntl(fd, cmd, arg, f.file);
 433                 break;
 434         }
 435 out1:
 436         fdput(f);
 437 out:
 438         return err;
 439 }
 440 #endif
 441
 442 #ifdef CONFIG_COMPAT
 443 static int get_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
 444 {
 445         if (!access_ok(VERIFY_READ, ufl, sizeof(*ufl)) ||
 446             __get_user(kfl->l_type, &ufl->l_type) ||
 447             __get_user(kfl->l_whence, &ufl->l_whence) ||
 448             __get_user(kfl->l_start, &ufl->l_start) ||
 449             __get_user(kfl->l_len, &ufl->l_len) ||
 450             __get_user(kfl->l_pid, &ufl->l_pid))
 451                 return -EFAULT;
 452         return 0;
 453 }
 454
 455 static int put_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
 456 {
 457         if (!access_ok(VERIFY_WRITE, ufl, sizeof(*ufl)) ||
 458             __put_user(kfl->l_type, &ufl->l_type) ||
 459             __put_user(kfl->l_whence, &ufl->l_whence) ||
 460             __put_user(kfl->l_start, &ufl->l_start) ||
 461             __put_user(kfl->l_len, &ufl->l_len) ||
 462             __put_user(kfl->l_pid, &ufl->l_pid))
 463                 return -EFAULT;
 464         return 0;
 465 }
 466
 467 #ifndef HAVE_ARCH_GET_COMPAT_FLOCK64
 468 static int get_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
 469 {
 470         if (!access_ok(VERIFY_READ, ufl, sizeof(*ufl)) ||
 471             __get_user(kfl->l_type, &ufl->l_type) ||
 472             __get_user(kfl->l_whence, &ufl->l_whence) ||
 473             __get_user(kfl->l_start, &ufl->l_start) ||
 474             __get_user(kfl->l_len, &ufl->l_len) ||
 475             __get_user(kfl->l_pid, &ufl->l_pid))
 476                 return -EFAULT;
 477         return 0;
 478 }
 479 #endif
 480
 481 #ifndef HAVE_ARCH_PUT_COMPAT_FLOCK64
 482 static int put_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
 483 {
 484         if (!access_ok(VERIFY_WRITE, ufl, sizeof(*ufl)) ||
 485             __put_user(kfl->l_type, &ufl->l_type) ||
 486             __put_user(kfl->l_whence, &ufl->l_whence) ||
 487             __put_user(kfl->l_start, &ufl->l_start) ||
 488             __put_user(kfl->l_len, &ufl->l_len) ||
 489             __put_user(kfl->l_pid, &ufl->l_pid))
 490                 return -EFAULT;
 491         return 0;
 492 }
 493 #endif
 494
 495 static unsigned int
 496 convert_fcntl_cmd(unsigned int cmd)
 497 {
 498         switch (cmd) {
 499         case F_GETLK64:
 500                 return F_GETLK;
 501         case F_SETLK64:
 502                 return F_SETLK;
 503         case F_SETLKW64:
 504                 return F_SETLKW;
 505         }
 506
 507         return cmd;
 508 }
 509
 510 /*
 511  * GETLK was successful and we need to return the data, but it needs to fit in
 512  * the compat structure.
 513  * l_start shouldn't be too big, unless the original start + end is greater than
 514  * COMPAT_OFF_T_MAX, in which case the app was asking for trouble, so we return
 515  * -EOVERFLOW in that case.  l_len could be too big, in which case we just
 516  * truncate it, and only allow the app to see that part of the conflicting lock
 517  * that might make sense to it anyway
 518  */
 519 static int fixup_compat_flock(struct flock *flock)
 520 {
 521         if (flock->l_start > COMPAT_OFF_T_MAX)
 522                 return -EOVERFLOW;
 523         if (flock->l_len > COMPAT_OFF_T_MAX)
 524                 flock->l_len = COMPAT_OFF_T_MAX;
 525         return 0;
 526 }
 527
 528 COMPAT_SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
 529                        compat_ulong_t, arg)
 530 {
 531         struct fd f = fdget_raw(fd);
 532         struct flock flock;
 533         long err = -EBADF;
 534
 535         if (!f.file)
 536                 return err;
 537
 538         if (unlikely(f.file->f_mode & FMODE_PATH)) {
 539                 if (!check_fcntl_cmd(cmd))
 540                         goto out_put;
 541         }
 542
 543         err = security_file_fcntl(f.file, cmd, arg);
 544         if (err)
 545                 goto out_put;
 546
 547         switch (cmd) {
 548         case F_GETLK:
 549                 err = get_compat_flock(&flock, compat_ptr(arg));
 550                 if (err)
 551                         break;
 552                 err = fcntl_getlk(f.file, convert_fcntl_cmd(cmd), &flock);
 553                 if (err)
 554                         break;
 555                 err = fixup_compat_flock(&flock);
 556                 if (err)
 557                         return err;
 558                 err = put_compat_flock(&flock, compat_ptr(arg));
 559                 break;
 560         case F_GETLK64:
 561         case F_OFD_GETLK:
 562                 err = get_compat_flock64(&flock, compat_ptr(arg));
 563                 if (err)
 564                         break;
 565                 err = fcntl_getlk(f.file, convert_fcntl_cmd(cmd), &flock);
 566                 if (err)
 567                         break;
 568                 err = fixup_compat_flock(&flock);
 569                 if (err)
 570                         return err;
 571                 err = put_compat_flock64(&flock, compat_ptr(arg));
 572                 break;
 573         case F_SETLK:
 574         case F_SETLKW:
 575                 err = get_compat_flock(&flock, compat_ptr(arg));
 576                 if (err)
 577                         break;
 578                 err = fcntl_setlk(fd, f.file, convert_fcntl_cmd(cmd), &flock);
 579                 break;
 580         case F_SETLK64:
 581         case F_SETLKW64:
 582         case F_OFD_SETLK:
 583         case F_OFD_SETLKW:
 584                 err = get_compat_flock64(&flock, compat_ptr(arg));
 585                 if (err)
 586                         break;
 587                 err = fcntl_setlk(fd, f.file, convert_fcntl_cmd(cmd), &flock);
 588                 break;
 589         default:
 590                 err = do_fcntl(fd, cmd, arg, f.file);
 591                 break;
 592         }
 593 out_put:
 594         fdput(f);
 595         return err;
 596 }
 597
 598 COMPAT_SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd,
 599                        compat_ulong_t, arg)
 600 {
 601         switch (cmd) {
 602         case F_GETLK64:
 603         case F_SETLK64:
 604         case F_SETLKW64:
 605         case F_OFD_GETLK:
 606         case F_OFD_SETLK:
 607         case F_OFD_SETLKW:
 608                 return -EINVAL;
 609         }
 610         return compat_sys_fcntl64(fd, cmd, arg);
 611 }
 612 #endif
 613
 614 /* Table to convert sigio signal codes into poll band bitmaps */
 615
 616 static const long band_table[NSIGPOLL] = {
 617         POLLIN | POLLRDNORM,                    /* POLL_IN */
 618         POLLOUT | POLLWRNORM | POLLWRBAND,      /* POLL_OUT */
 619         POLLIN | POLLRDNORM | POLLMSG,          /* POLL_MSG */
 620         POLLERR,                                /* POLL_ERR */
 621         POLLPRI | POLLRDBAND,                   /* POLL_PRI */
 622         POLLHUP | POLLERR                       /* POLL_HUP */
 623 };
 624
 625 static inline int sigio_perm(struct task_struct *p,
 626                              struct fown_struct *fown, int sig)
 627 {
 628         const struct cred *cred;
 629         int ret;
 630
 631         rcu_read_lock();
 632         cred = __task_cred(p);
 633         ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
 634                 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
 635                 uid_eq(fown->uid,  cred->suid) || uid_eq(fown->uid,  cred->uid)) &&
 636                !security_file_send_sigiotask(p, fown, sig));
 637         rcu_read_unlock();
 638         return ret;
 639 }
 640
 641 static void send_sigio_to_task(struct task_struct *p,
 642                                struct fown_struct *fown,
 643                                int fd, int reason, int group)
 644 {
 645         /*
 646          * F_SETSIG can change ->signum lockless in parallel, make
 647          * sure we read it once and use the same value throughout.
 648          */
 649         int signum = ACCESS_ONCE(fown->signum);
 650
 651         if (!sigio_perm(p, fown, signum))
 652                 return;
 653
 654         switch (signum) {
 655                 siginfo_t si;
 656                 default:
 657                         /* Queue a rt signal with the appropriate fd as its
 658                            value.  We use SI_SIGIO as the source, not
 659                            SI_KERNEL, since kernel signals always get
 660                            delivered even if we can't queue.  Failure to
 661                            queue in this case _should_ be reported; we fall
 662                            back to SIGIO in that case. --sct */
 663                         si.si_signo = signum;
 664                         si.si_errno = 0;
 665                         si.si_code  = reason;
 666                         /* Make sure we are called with one of the POLL_*
 667                            reasons, otherwise we could leak kernel stack into
 668                            userspace.  */
 669                         BUG_ON((reason & __SI_MASK) != __SI_POLL);
 670                         if (reason - POLL_IN >= NSIGPOLL)
 671                                 si.si_band  = ~0L;
 672                         else
 673                                 si.si_band = band_table[reason - POLL_IN];
 674                         si.si_fd    = fd;
 675                         if (!do_send_sig_info(signum, &si, p, group))
 676                                 break;
 677                 /* fall-through: fall back on the old plain SIGIO signal */
 678                 case 0:
 679                         do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
 680         }
 681 }
 682
 683 void send_sigio(struct fown_struct *fown, int fd, int band)
 684 {
 685         struct task_struct *p;
 686         enum pid_type type;
 687         struct pid *pid;
 688         int group = 1;
 689
 690         read_lock(&fown->lock);
 691
 692         type = fown->pid_type;
 693         if (type == PIDTYPE_MAX) {
 694                 group = 0;
 695                 type = PIDTYPE_PID;
 696         }
 697
 698         pid = fown->pid;
 699         if (!pid)
 700                 goto out_unlock_fown;
 701
 702         read_lock(&tasklist_lock);
 703         do_each_pid_task(pid, type, p) {
 704                 send_sigio_to_task(p, fown, fd, band, group);
 705         } while_each_pid_task(pid, type, p);
 706         read_unlock(&tasklist_lock);
 707  out_unlock_fown:
 708         read_unlock(&fown->lock);
 709 }
 710
 711 static void send_sigurg_to_task(struct task_struct *p,
 712                                 struct fown_struct *fown, int group)
 713 {
 714         if (sigio_perm(p, fown, SIGURG))
 715                 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
 716 }
 717
 718 int send_sigurg(struct fown_struct *fown)
 719 {
 720         struct task_struct *p;
 721         enum pid_type type;
 722         struct pid *pid;
 723         int group = 1;
 724         int ret = 0;
 725
 726         read_lock(&fown->lock);
 727
 728         type = fown->pid_type;
 729         if (type == PIDTYPE_MAX) {
 730                 group = 0;
 731                 type = PIDTYPE_PID;
 732         }
 733
 734         pid = fown->pid;
 735         if (!pid)
 736                 goto out_unlock_fown;
 737
 738         ret = 1;
 739
 740         read_lock(&tasklist_lock);
 741         do_each_pid_task(pid, type, p) {
 742                 send_sigurg_to_task(p, fown, group);
 743         } while_each_pid_task(pid, type, p);
 744         read_unlock(&tasklist_lock);
 745  out_unlock_fown:
 746         read_unlock(&fown->lock);
 747         return ret;
 748 }
 749
 750 static DEFINE_SPINLOCK(fasync_lock);
 751 static struct kmem_cache *fasync_cache __read_mostly;
 752
 753 static void fasync_free_rcu(struct rcu_head *head)
 754 {
 755         kmem_cache_free(fasync_cache,
 756                         container_of(head, struct fasync_struct, fa_rcu));
 757 }
 758
 759 /*
 760  * Remove a fasync entry. If successfully removed, return
 761  * positive and clear the FASYNC flag. If no entry exists,
 762  * do nothing and return 0.
 763  *
 764  * NOTE! It is very important that the FASYNC flag always
 765  * match the state "is the filp on a fasync list".
 766  *
 767  */
 768 int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
 769 {
 770         struct fasync_struct *fa, **fp;
 771         int result = 0;
 772
 773         spin_lock(&filp->f_lock);
 774         spin_lock(&fasync_lock);
 775         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 776                 if (fa->fa_file != filp)
 777                         continue;
 778
 779                 spin_lock_irq(&fa->fa_lock);
 780                 fa->fa_file = NULL;
 781                 spin_unlock_irq(&fa->fa_lock);
 782
 783                 *fp = fa->fa_next;
 784                 call_rcu(&fa->fa_rcu, fasync_free_rcu);
 785                 filp->f_flags &= ~FASYNC;
 786                 result = 1;
 787                 break;
 788         }
 789         spin_unlock(&fasync_lock);
 790         spin_unlock(&filp->f_lock);
 791         return result;
 792 }
 793
 794 struct fasync_struct *fasync_alloc(void)
 795 {
 796         return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
 797 }
 798
 799 /*
 800  * NOTE! This can be used only for unused fasync entries:
 801  * entries that actually got inserted on the fasync list
 802  * need to be released by rcu - see fasync_remove_entry.
 803  */
 804 void fasync_free(struct fasync_struct *new)
 805 {
 806         kmem_cache_free(fasync_cache, new);
 807 }
 808
 809 /*
 810  * Insert a new entry into the fasync list.  Return the pointer to the
 811  * old one if we didn't use the new one.
 812  *
 813  * NOTE! It is very important that the FASYNC flag always
 814  * match the state "is the filp on a fasync list".
 815  */
 816 struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
 817 {
 818         struct fasync_struct *fa, **fp;
 819
 820         spin_lock(&filp->f_lock);
 821         spin_lock(&fasync_lock);
 822         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 823                 if (fa->fa_file != filp)
 824                         continue;
 825
 826                 spin_lock_irq(&fa->fa_lock);
 827                 fa->fa_fd = fd;
 828                 spin_unlock_irq(&fa->fa_lock);
 829                 goto out;
 830         }
 831
 832         spin_lock_init(&new->fa_lock);
 833         new->magic = FASYNC_MAGIC;
 834         new->fa_file = filp;
 835         new->fa_fd = fd;
 836         new->fa_next = *fapp;
 837         rcu_assign_pointer(*fapp, new);
 838         filp->f_flags |= FASYNC;
 839
 840 out:
 841         spin_unlock(&fasync_lock);
 842         spin_unlock(&filp->f_lock);
 843         return fa;
 844 }
 845
 846 /*
 847  * Add a fasync entry. Return negative on error, positive if
 848  * added, and zero if did nothing but change an existing one.
 849  */
 850 static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
 851 {
 852         struct fasync_struct *new;
 853
 854         new = fasync_alloc();
 855         if (!new)
 856                 return -ENOMEM;
 857
 858         /*
 859          * fasync_insert_entry() returns the old (update) entry if
 860          * it existed.
 861          *
 862          * So free the (unused) new entry and return 0 to let the
 863          * caller know that we didn't add any new fasync entries.
 864          */
 865         if (fasync_insert_entry(fd, filp, fapp, new)) {
 866                 fasync_free(new);
 867                 return 0;
 868         }
 869
 870         return 1;
 871 }
 872
 873 /*
 874  * fasync_helper() is used by almost all character device drivers
 875  * to set up the fasync queue, and for regular files by the file
 876  * lease code. It returns negative on error, 0 if it did no changes
 877  * and positive if it added/deleted the entry.
 878  */
 879 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
 880 {
 881         if (!on)
 882                 return fasync_remove_entry(filp, fapp);
 883         return fasync_add_entry(fd, filp, fapp);
 884 }
 885
 886 EXPORT_SYMBOL(fasync_helper);
 887
 888 /*
 889  * rcu_read_lock() is held
 890  */
 891 static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
 892 {
 893         while (fa) {
 894                 struct fown_struct *fown;
 895                 unsigned long flags;
 896
 897                 if (fa->magic != FASYNC_MAGIC) {
 898                         printk(KERN_ERR "kill_fasync: bad magic number in "
 899                                "fasync_struct!\n");
 900                         return;
 901                 }
 902                 spin_lock_irqsave(&fa->fa_lock, flags);
 903                 if (fa->fa_file) {
 904                         fown = &fa->fa_file->f_owner;
 905                         /* Don't send SIGURG to processes which have not set a
 906                            queued signum: SIGURG has its own default signalling
 907                            mechanism. */
 908                         if (!(sig == SIGURG && fown->signum == 0))
 909                                 send_sigio(fown, fa->fa_fd, band);
 910                 }
 911                 spin_unlock_irqrestore(&fa->fa_lock, flags);
 912                 fa = rcu_dereference(fa->fa_next);
 913         }
 914 }
 915
 916 void kill_fasync(struct fasync_struct **fp, int sig, int band)
 917 {
 918         /* First a quick test without locking: usually
 919          * the list is empty.
 920          */
 921         if (*fp) {
 922                 rcu_read_lock();
 923                 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
 924                 rcu_read_unlock();
 925         }
 926 }
 927 EXPORT_SYMBOL(kill_fasync);
 928
 929 static int __init fcntl_init(void)
 930 {
 931         /*
 932          * Please add new bits here to ensure allocation uniqueness.
 933          * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
 934          * is defined as O_NONBLOCK on some platforms and not on others.
 935          */
 936         BUILD_BUG_ON(21 - 1 /* for O_RDONLY being 0 */ !=
 937                 HWEIGHT32(
 938                         (VALID_OPEN_FLAGS & ~(O_NONBLOCK | O_NDELAY)) |
 939                         __FMODE_EXEC | __FMODE_NONOTIFY));
 940
 941         fasync_cache = kmem_cache_create("fasync_cache",
 942                 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
 943         return 0;
 944 }
 945
 946 module_init(fcntl_init)