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 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 && filp->f_op->setfl)
  66                 error = filp->f_op->setfl(filp, arg);
  67         if (error)
  68                 return error;
  69
  70         /*
  71          * ->fasync() is responsible for setting the FASYNC bit.
  72          */
  73         if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op->fasync) {
  74                 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
  75                 if (error < 0)
  76                         goto out;
  77                 if (error > 0)
  78                         error = 0;
  79         }
  80         spin_lock(&filp->f_lock);
  81         filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
  82         spin_unlock(&filp->f_lock);
  83
  84  out:
  85         return error;
  86 }
  87 EXPORT_SYMBOL_GPL(setfl);
  88
  89 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
  90                      int force)
  91 {
  92         write_lock_irq(&filp->f_owner.lock);
  93         if (force || !filp->f_owner.pid) {
  94                 put_pid(filp->f_owner.pid);
  95                 filp->f_owner.pid = get_pid(pid);
  96                 filp->f_owner.pid_type = type;
  97
  98                 if (pid) {
  99                         const struct cred *cred = current_cred();
 100                         filp->f_owner.uid = cred->uid;
 101                         filp->f_owner.euid = cred->euid;
 102                 }
 103         }
 104         write_unlock_irq(&filp->f_owner.lock);
 105 }
 106
 107 void __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
 108                 int force)
 109 {
 110         security_file_set_fowner(filp);
 111         f_modown(filp, pid, type, force);
 112 }
 113 EXPORT_SYMBOL(__f_setown);
 114
 115 int f_setown(struct file *filp, unsigned long arg, int force)
 116 {
 117         enum pid_type type;
 118         struct pid *pid = NULL;
 119         int who = arg, ret = 0;
 120
 121         type = PIDTYPE_PID;
 122         if (who < 0) {
 123                 /* avoid overflow below */
 124                 if (who == INT_MIN)
 125                         return -EINVAL;
 126
 127                 type = PIDTYPE_PGID;
 128                 who = -who;
 129         }
 130
 131         rcu_read_lock();
 132         if (who) {
 133                 pid = find_vpid(who);
 134                 if (!pid)
 135                         ret = -ESRCH;
 136         }
 137
 138         if (!ret)
 139                 __f_setown(filp, pid, type, force);
 140         rcu_read_unlock();
 141
 142         return ret;
 143 }
 144 EXPORT_SYMBOL(f_setown);
 145
 146 void f_delown(struct file *filp)
 147 {
 148         f_modown(filp, NULL, PIDTYPE_PID, 1);
 149 }
 150
 151 pid_t f_getown(struct file *filp)
 152 {
 153         pid_t pid;
 154         read_lock(&filp->f_owner.lock);
 155         pid = pid_vnr(filp->f_owner.pid);
 156         if (filp->f_owner.pid_type == PIDTYPE_PGID)
 157                 pid = -pid;
 158         read_unlock(&filp->f_owner.lock);
 159         return pid;
 160 }
 161
 162 static int f_setown_ex(struct file *filp, unsigned long arg)
 163 {
 164         struct f_owner_ex __user *owner_p = (void __user *)arg;
 165         struct f_owner_ex owner;
 166         struct pid *pid;
 167         int type;
 168         int ret;
 169
 170         ret = copy_from_user(&owner, owner_p, sizeof(owner));
 171         if (ret)
 172                 return -EFAULT;
 173
 174         switch (owner.type) {
 175         case F_OWNER_TID:
 176                 type = PIDTYPE_MAX;
 177                 break;
 178
 179         case F_OWNER_PID:
 180                 type = PIDTYPE_PID;
 181                 break;
 182
 183         case F_OWNER_PGRP:
 184                 type = PIDTYPE_PGID;
 185                 break;
 186
 187         default:
 188                 return -EINVAL;
 189         }
 190
 191         rcu_read_lock();
 192         pid = find_vpid(owner.pid);
 193         if (owner.pid && !pid)
 194                 ret = -ESRCH;
 195         else
 196                  __f_setown(filp, pid, type, 1);
 197         rcu_read_unlock();
 198
 199         return ret;
 200 }
 201
 202 static int f_getown_ex(struct file *filp, unsigned long arg)
 203 {
 204         struct f_owner_ex __user *owner_p = (void __user *)arg;
 205         struct f_owner_ex owner;
 206         int ret = 0;
 207
 208         read_lock(&filp->f_owner.lock);
 209         owner.pid = pid_vnr(filp->f_owner.pid);
 210         switch (filp->f_owner.pid_type) {
 211         case PIDTYPE_MAX:
 212                 owner.type = F_OWNER_TID;
 213                 break;
 214
 215         case PIDTYPE_PID:
 216                 owner.type = F_OWNER_PID;
 217                 break;
 218
 219         case PIDTYPE_PGID:
 220                 owner.type = F_OWNER_PGRP;
 221                 break;
 222
 223         default:
 224                 WARN_ON(1);
 225                 ret = -EINVAL;
 226                 break;
 227         }
 228         read_unlock(&filp->f_owner.lock);
 229
 230         if (!ret) {
 231                 ret = copy_to_user(owner_p, &owner, sizeof(owner));
 232                 if (ret)
 233                         ret = -EFAULT;
 234         }
 235         return ret;
 236 }
 237
 238 #ifdef CONFIG_CHECKPOINT_RESTORE
 239 static int f_getowner_uids(struct file *filp, unsigned long arg)
 240 {
 241         struct user_namespace *user_ns = current_user_ns();
 242         uid_t __user *dst = (void __user *)arg;
 243         uid_t src[2];
 244         int err;
 245
 246         read_lock(&filp->f_owner.lock);
 247         src[0] = from_kuid(user_ns, filp->f_owner.uid);
 248         src[1] = from_kuid(user_ns, filp->f_owner.euid);
 249         read_unlock(&filp->f_owner.lock);
 250
 251         err  = put_user(src[0], &dst[0]);
 252         err |= put_user(src[1], &dst[1]);
 253
 254         return err;
 255 }
 256 #else
 257 static int f_getowner_uids(struct file *filp, unsigned long arg)
 258 {
 259         return -EINVAL;
 260 }
 261 #endif
 262
 263 static bool rw_hint_valid(enum rw_hint hint)
 264 {
 265         switch (hint) {
 266         case RWF_WRITE_LIFE_NOT_SET:
 267         case RWH_WRITE_LIFE_NONE:
 268         case RWH_WRITE_LIFE_SHORT:
 269         case RWH_WRITE_LIFE_MEDIUM:
 270         case RWH_WRITE_LIFE_LONG:
 271         case RWH_WRITE_LIFE_EXTREME:
 272                 return true;
 273         default:
 274                 return false;
 275         }
 276 }
 277
 278 static long fcntl_rw_hint(struct file *file, unsigned int cmd,
 279                           unsigned long arg)
 280 {
 281         struct inode *inode = file_inode(file);
 282         u64 *argp = (u64 __user *)arg;
 283         enum rw_hint hint;
 284         u64 h;
 285
 286         switch (cmd) {
 287         case F_GET_FILE_RW_HINT:
 288                 h = file_write_hint(file);
 289                 if (copy_to_user(argp, &h, sizeof(*argp)))
 290                         return -EFAULT;
 291                 return 0;
 292         case F_SET_FILE_RW_HINT:
 293                 if (copy_from_user(&h, argp, sizeof(h)))
 294                         return -EFAULT;
 295                 hint = (enum rw_hint) h;
 296                 if (!rw_hint_valid(hint))
 297                         return -EINVAL;
 298
 299                 spin_lock(&file->f_lock);
 300                 file->f_write_hint = hint;
 301                 spin_unlock(&file->f_lock);
 302                 return 0;
 303         case F_GET_RW_HINT:
 304                 h = inode->i_write_hint;
 305                 if (copy_to_user(argp, &h, sizeof(*argp)))
 306                         return -EFAULT;
 307                 return 0;
 308         case F_SET_RW_HINT:
 309                 if (copy_from_user(&h, argp, sizeof(h)))
 310                         return -EFAULT;
 311                 hint = (enum rw_hint) h;
 312                 if (!rw_hint_valid(hint))
 313                         return -EINVAL;
 314
 315                 inode_lock(inode);
 316                 inode->i_write_hint = hint;
 317                 inode_unlock(inode);
 318                 return 0;
 319         default:
 320                 return -EINVAL;
 321         }
 322 }
 323
 324 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
 325                 struct file *filp)
 326 {
 327         void __user *argp = (void __user *)arg;
 328         struct flock flock;
 329         long err = -EINVAL;
 330
 331         switch (cmd) {
 332         case F_DUPFD:
 333                 err = f_dupfd(arg, filp, 0);
 334                 break;
 335         case F_DUPFD_CLOEXEC:
 336                 err = f_dupfd(arg, filp, O_CLOEXEC);
 337                 break;
 338         case F_GETFD:
 339                 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
 340                 break;
 341         case F_SETFD:
 342                 err = 0;
 343                 set_close_on_exec(fd, arg & FD_CLOEXEC);
 344                 break;
 345         case F_GETFL:
 346                 err = filp->f_flags;
 347                 break;
 348         case F_SETFL:
 349                 err = setfl(fd, filp, arg);
 350                 break;
 351 #if BITS_PER_LONG != 32
 352         /* 32-bit arches must use fcntl64() */
 353         case F_OFD_GETLK:
 354 #endif
 355         case F_GETLK:
 356                 if (copy_from_user(&flock, argp, sizeof(flock)))
 357                         return -EFAULT;
 358                 err = fcntl_getlk(filp, cmd, &flock);
 359                 if (!err && copy_to_user(argp, &flock, sizeof(flock)))
 360                         return -EFAULT;
 361                 break;
 362 #if BITS_PER_LONG != 32
 363         /* 32-bit arches must use fcntl64() */
 364         case F_OFD_SETLK:
 365         case F_OFD_SETLKW:
 366 #endif
 367                 /* Fallthrough */
 368         case F_SETLK:
 369         case F_SETLKW:
 370                 if (copy_from_user(&flock, argp, sizeof(flock)))
 371                         return -EFAULT;
 372                 err = fcntl_setlk(fd, filp, cmd, &flock);
 373                 break;
 374         case F_GETOWN:
 375                 /*
 376                  * XXX If f_owner is a process group, the
 377                  * negative return value will get converted
 378                  * into an error.  Oops.  If we keep the
 379                  * current syscall conventions, the only way
 380                  * to fix this will be in libc.
 381                  */
 382                 err = f_getown(filp);
 383                 force_successful_syscall_return();
 384                 break;
 385         case F_SETOWN:
 386                 err = f_setown(filp, arg, 1);
 387                 break;
 388         case F_GETOWN_EX:
 389                 err = f_getown_ex(filp, arg);
 390                 break;
 391         case F_SETOWN_EX:
 392                 err = f_setown_ex(filp, arg);
 393                 break;
 394         case F_GETOWNER_UIDS:
 395                 err = f_getowner_uids(filp, arg);
 396                 break;
 397         case F_GETSIG:
 398                 err = filp->f_owner.signum;
 399                 break;
 400         case F_SETSIG:
 401                 /* arg == 0 restores default behaviour. */
 402                 if (!valid_signal(arg)) {
 403                         break;
 404                 }
 405                 err = 0;
 406                 filp->f_owner.signum = arg;
 407                 break;
 408         case F_GETLEASE:
 409                 err = fcntl_getlease(filp);
 410                 break;
 411         case F_SETLEASE:
 412                 err = fcntl_setlease(fd, filp, arg);
 413                 break;
 414         case F_NOTIFY:
 415                 err = fcntl_dirnotify(fd, filp, arg);
 416                 break;
 417         case F_SETPIPE_SZ:
 418         case F_GETPIPE_SZ:
 419                 err = pipe_fcntl(filp, cmd, arg);
 420                 break;
 421         case F_ADD_SEALS:
 422         case F_GET_SEALS:
 423                 err = shmem_fcntl(filp, cmd, arg);
 424                 break;
 425         case F_GET_RW_HINT:
 426         case F_SET_RW_HINT:
 427         case F_GET_FILE_RW_HINT:
 428         case F_SET_FILE_RW_HINT:
 429                 err = fcntl_rw_hint(filp, cmd, arg);
 430                 break;
 431         default:
 432                 break;
 433         }
 434         return err;
 435 }
 436
 437 static int check_fcntl_cmd(unsigned cmd)
 438 {
 439         switch (cmd) {
 440         case F_DUPFD:
 441         case F_DUPFD_CLOEXEC:
 442         case F_GETFD:
 443         case F_SETFD:
 444         case F_GETFL:
 445                 return 1;
 446         }
 447         return 0;
 448 }
 449
 450 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
 451 {
 452         struct fd f = fdget_raw(fd);
 453         long err = -EBADF;
 454
 455         if (!f.file)
 456                 goto out;
 457
 458         if (unlikely(f.file->f_mode & FMODE_PATH)) {
 459                 if (!check_fcntl_cmd(cmd))
 460                         goto out1;
 461         }
 462
 463         err = security_file_fcntl(f.file, cmd, arg);
 464         if (!err)
 465                 err = do_fcntl(fd, cmd, arg, f.file);
 466
 467 out1:
 468         fdput(f);
 469 out:
 470         return err;
 471 }
 472
 473 #if BITS_PER_LONG == 32
 474 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
 475                 unsigned long, arg)
 476 {
 477         void __user *argp = (void __user *)arg;
 478         struct fd f = fdget_raw(fd);
 479         struct flock64 flock;
 480         long err = -EBADF;
 481
 482         if (!f.file)
 483                 goto out;
 484
 485         if (unlikely(f.file->f_mode & FMODE_PATH)) {
 486                 if (!check_fcntl_cmd(cmd))
 487                         goto out1;
 488         }
 489
 490         err = security_file_fcntl(f.file, cmd, arg);
 491         if (err)
 492                 goto out1;
 493
 494         switch (cmd) {
 495         case F_GETLK64:
 496         case F_OFD_GETLK:
 497                 err = -EFAULT;
 498                 if (copy_from_user(&flock, argp, sizeof(flock)))
 499                         break;
 500                 err = fcntl_getlk64(f.file, cmd, &flock);
 501                 if (!err && copy_to_user(argp, &flock, sizeof(flock)))
 502                         err = -EFAULT;
 503                 break;
 504         case F_SETLK64:
 505         case F_SETLKW64:
 506         case F_OFD_SETLK:
 507         case F_OFD_SETLKW:
 508                 err = -EFAULT;
 509                 if (copy_from_user(&flock, argp, sizeof(flock)))
 510                         break;
 511                 err = fcntl_setlk64(fd, f.file, cmd, &flock);
 512                 break;
 513         default:
 514                 err = do_fcntl(fd, cmd, arg, f.file);
 515                 break;
 516         }
 517 out1:
 518         fdput(f);
 519 out:
 520         return err;
 521 }
 522 #endif
 523
 524 #ifdef CONFIG_COMPAT
 525 /* careful - don't use anywhere else */
 526 #define copy_flock_fields(dst, src)             \
 527         (dst)->l_type = (src)->l_type;          \
 528         (dst)->l_whence = (src)->l_whence;      \
 529         (dst)->l_start = (src)->l_start;        \
 530         (dst)->l_len = (src)->l_len;            \
 531         (dst)->l_pid = (src)->l_pid;
 532
 533 static int get_compat_flock(struct flock *kfl, const struct compat_flock __user *ufl)
 534 {
 535         struct compat_flock fl;
 536
 537         if (copy_from_user(&fl, ufl, sizeof(struct compat_flock)))
 538                 return -EFAULT;
 539         copy_flock_fields(kfl, &fl);
 540         return 0;
 541 }
 542
 543 static int get_compat_flock64(struct flock *kfl, const struct compat_flock64 __user *ufl)
 544 {
 545         struct compat_flock64 fl;
 546
 547         if (copy_from_user(&fl, ufl, sizeof(struct compat_flock64)))
 548                 return -EFAULT;
 549         copy_flock_fields(kfl, &fl);
 550         return 0;
 551 }
 552
 553 static int put_compat_flock(const struct flock *kfl, struct compat_flock __user *ufl)
 554 {
 555         struct compat_flock fl;
 556
 557         memset(&fl, 0, sizeof(struct compat_flock));
 558         copy_flock_fields(&fl, kfl);
 559         if (copy_to_user(ufl, &fl, sizeof(struct compat_flock)))
 560                 return -EFAULT;
 561         return 0;
 562 }
 563
 564 static int put_compat_flock64(const struct flock *kfl, struct compat_flock64 __user *ufl)
 565 {
 566         struct compat_flock64 fl;
 567
 568         memset(&fl, 0, sizeof(struct compat_flock64));
 569         copy_flock_fields(&fl, kfl);
 570         if (copy_to_user(ufl, &fl, sizeof(struct compat_flock64)))
 571                 return -EFAULT;
 572         return 0;
 573 }
 574 #undef copy_flock_fields
 575
 576 static unsigned int
 577 convert_fcntl_cmd(unsigned int cmd)
 578 {
 579         switch (cmd) {
 580         case F_GETLK64:
 581                 return F_GETLK;
 582         case F_SETLK64:
 583                 return F_SETLK;
 584         case F_SETLKW64:
 585                 return F_SETLKW;
 586         }
 587
 588         return cmd;
 589 }
 590
 591 /*
 592  * GETLK was successful and we need to return the data, but it needs to fit in
 593  * the compat structure.
 594  * l_start shouldn't be too big, unless the original start + end is greater than
 595  * COMPAT_OFF_T_MAX, in which case the app was asking for trouble, so we return
 596  * -EOVERFLOW in that case.  l_len could be too big, in which case we just
 597  * truncate it, and only allow the app to see that part of the conflicting lock
 598  * that might make sense to it anyway
 599  */
 600 static int fixup_compat_flock(struct flock *flock)
 601 {
 602         if (flock->l_start > COMPAT_OFF_T_MAX)
 603                 return -EOVERFLOW;
 604         if (flock->l_len > COMPAT_OFF_T_MAX)
 605                 flock->l_len = COMPAT_OFF_T_MAX;
 606         return 0;
 607 }
 608
 609 COMPAT_SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
 610                        compat_ulong_t, arg)
 611 {
 612         struct fd f = fdget_raw(fd);
 613         struct flock flock;
 614         long err = -EBADF;
 615
 616         if (!f.file)
 617                 return err;
 618
 619         if (unlikely(f.file->f_mode & FMODE_PATH)) {
 620                 if (!check_fcntl_cmd(cmd))
 621                         goto out_put;
 622         }
 623
 624         err = security_file_fcntl(f.file, cmd, arg);
 625         if (err)
 626                 goto out_put;
 627
 628         switch (cmd) {
 629         case F_GETLK:
 630                 err = get_compat_flock(&flock, compat_ptr(arg));
 631                 if (err)
 632                         break;
 633                 err = fcntl_getlk(f.file, convert_fcntl_cmd(cmd), &flock);
 634                 if (err)
 635                         break;
 636                 err = fixup_compat_flock(&flock);
 637                 if (err)
 638                         return err;
 639                 err = put_compat_flock(&flock, compat_ptr(arg));
 640                 break;
 641         case F_GETLK64:
 642         case F_OFD_GETLK:
 643                 err = get_compat_flock64(&flock, compat_ptr(arg));
 644                 if (err)
 645                         break;
 646                 err = fcntl_getlk(f.file, convert_fcntl_cmd(cmd), &flock);
 647                 if (err)
 648                         break;
 649                 err = fixup_compat_flock(&flock);
 650                 if (err)
 651                         return err;
 652                 err = put_compat_flock64(&flock, compat_ptr(arg));
 653                 break;
 654         case F_SETLK:
 655         case F_SETLKW:
 656                 err = get_compat_flock(&flock, compat_ptr(arg));
 657                 if (err)
 658                         break;
 659                 err = fcntl_setlk(fd, f.file, convert_fcntl_cmd(cmd), &flock);
 660                 break;
 661         case F_SETLK64:
 662         case F_SETLKW64:
 663         case F_OFD_SETLK:
 664         case F_OFD_SETLKW:
 665                 err = get_compat_flock64(&flock, compat_ptr(arg));
 666                 if (err)
 667                         break;
 668                 err = fcntl_setlk(fd, f.file, convert_fcntl_cmd(cmd), &flock);
 669                 break;
 670         default:
 671                 err = do_fcntl(fd, cmd, arg, f.file);
 672                 break;
 673         }
 674 out_put:
 675         fdput(f);
 676         return err;
 677 }
 678
 679 COMPAT_SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd,
 680                        compat_ulong_t, arg)
 681 {
 682         switch (cmd) {
 683         case F_GETLK64:
 684         case F_SETLK64:
 685         case F_SETLKW64:
 686         case F_OFD_GETLK:
 687         case F_OFD_SETLK:
 688         case F_OFD_SETLKW:
 689                 return -EINVAL;
 690         }
 691         return compat_sys_fcntl64(fd, cmd, arg);
 692 }
 693 #endif
 694
 695 /* Table to convert sigio signal codes into poll band bitmaps */
 696
 697 static const long band_table[NSIGPOLL] = {
 698         POLLIN | POLLRDNORM,                    /* POLL_IN */
 699         POLLOUT | POLLWRNORM | POLLWRBAND,      /* POLL_OUT */
 700         POLLIN | POLLRDNORM | POLLMSG,          /* POLL_MSG */
 701         POLLERR,                                /* POLL_ERR */
 702         POLLPRI | POLLRDBAND,                   /* POLL_PRI */
 703         POLLHUP | POLLERR                       /* POLL_HUP */
 704 };
 705
 706 static inline int sigio_perm(struct task_struct *p,
 707                              struct fown_struct *fown, int sig)
 708 {
 709         const struct cred *cred;
 710         int ret;
 711
 712         rcu_read_lock();
 713         cred = __task_cred(p);
 714         ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
 715                 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
 716                 uid_eq(fown->uid,  cred->suid) || uid_eq(fown->uid,  cred->uid)) &&
 717                !security_file_send_sigiotask(p, fown, sig));
 718         rcu_read_unlock();
 719         return ret;
 720 }
 721
 722 static void send_sigio_to_task(struct task_struct *p,
 723                                struct fown_struct *fown,
 724                                int fd, int reason, int group)
 725 {
 726         /*
 727          * F_SETSIG can change ->signum lockless in parallel, make
 728          * sure we read it once and use the same value throughout.
 729          */
 730         int signum = ACCESS_ONCE(fown->signum);
 731
 732         if (!sigio_perm(p, fown, signum))
 733                 return;
 734
 735         switch (signum) {
 736                 siginfo_t si;
 737                 default:
 738                         /* Queue a rt signal with the appropriate fd as its
 739                            value.  We use SI_SIGIO as the source, not
 740                            SI_KERNEL, since kernel signals always get
 741                            delivered even if we can't queue.  Failure to
 742                            queue in this case _should_ be reported; we fall
 743                            back to SIGIO in that case. --sct */
 744                         si.si_signo = signum;
 745                         si.si_errno = 0;
 746                         si.si_code  = reason;
 747                         /* Make sure we are called with one of the POLL_*
 748                            reasons, otherwise we could leak kernel stack into
 749                            userspace.  */
 750                         BUG_ON((reason & __SI_MASK) != __SI_POLL);
 751                         if (reason - POLL_IN >= NSIGPOLL)
 752                                 si.si_band  = ~0L;
 753                         else
 754                                 si.si_band = band_table[reason - POLL_IN];
 755                         si.si_fd    = fd;
 756                         if (!do_send_sig_info(signum, &si, p, group))
 757                                 break;
 758                 /* fall-through: fall back on the old plain SIGIO signal */
 759                 case 0:
 760                         do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
 761         }
 762 }
 763
 764 void send_sigio(struct fown_struct *fown, int fd, int band)
 765 {
 766         struct task_struct *p;
 767         enum pid_type type;
 768         struct pid *pid;
 769         int group = 1;
 770
 771         read_lock(&fown->lock);
 772
 773         type = fown->pid_type;
 774         if (type == PIDTYPE_MAX) {
 775                 group = 0;
 776                 type = PIDTYPE_PID;
 777         }
 778
 779         pid = fown->pid;
 780         if (!pid)
 781                 goto out_unlock_fown;
 782
 783         read_lock(&tasklist_lock);
 784         do_each_pid_task(pid, type, p) {
 785                 send_sigio_to_task(p, fown, fd, band, group);
 786         } while_each_pid_task(pid, type, p);
 787         read_unlock(&tasklist_lock);
 788  out_unlock_fown:
 789         read_unlock(&fown->lock);
 790 }
 791
 792 static void send_sigurg_to_task(struct task_struct *p,
 793                                 struct fown_struct *fown, int group)
 794 {
 795         if (sigio_perm(p, fown, SIGURG))
 796                 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
 797 }
 798
 799 int send_sigurg(struct fown_struct *fown)
 800 {
 801         struct task_struct *p;
 802         enum pid_type type;
 803         struct pid *pid;
 804         int group = 1;
 805         int ret = 0;
 806
 807         read_lock(&fown->lock);
 808
 809         type = fown->pid_type;
 810         if (type == PIDTYPE_MAX) {
 811                 group = 0;
 812                 type = PIDTYPE_PID;
 813         }
 814
 815         pid = fown->pid;
 816         if (!pid)
 817                 goto out_unlock_fown;
 818
 819         ret = 1;
 820
 821         read_lock(&tasklist_lock);
 822         do_each_pid_task(pid, type, p) {
 823                 send_sigurg_to_task(p, fown, group);
 824         } while_each_pid_task(pid, type, p);
 825         read_unlock(&tasklist_lock);
 826  out_unlock_fown:
 827         read_unlock(&fown->lock);
 828         return ret;
 829 }
 830
 831 static DEFINE_SPINLOCK(fasync_lock);
 832 static struct kmem_cache *fasync_cache __read_mostly;
 833
 834 static void fasync_free_rcu(struct rcu_head *head)
 835 {
 836         kmem_cache_free(fasync_cache,
 837                         container_of(head, struct fasync_struct, fa_rcu));
 838 }
 839
 840 /*
 841  * Remove a fasync entry. If successfully removed, return
 842  * positive and clear the FASYNC flag. If no entry exists,
 843  * do nothing and return 0.
 844  *
 845  * NOTE! It is very important that the FASYNC flag always
 846  * match the state "is the filp on a fasync list".
 847  *
 848  */
 849 int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
 850 {
 851         struct fasync_struct *fa, **fp;
 852         int result = 0;
 853
 854         spin_lock(&filp->f_lock);
 855         spin_lock(&fasync_lock);
 856         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 857                 if (fa->fa_file != filp)
 858                         continue;
 859
 860                 spin_lock_irq(&fa->fa_lock);
 861                 fa->fa_file = NULL;
 862                 spin_unlock_irq(&fa->fa_lock);
 863
 864                 *fp = fa->fa_next;
 865                 call_rcu(&fa->fa_rcu, fasync_free_rcu);
 866                 filp->f_flags &= ~FASYNC;
 867                 result = 1;
 868                 break;
 869         }
 870         spin_unlock(&fasync_lock);
 871         spin_unlock(&filp->f_lock);
 872         return result;
 873 }
 874
 875 struct fasync_struct *fasync_alloc(void)
 876 {
 877         return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
 878 }
 879
 880 /*
 881  * NOTE! This can be used only for unused fasync entries:
 882  * entries that actually got inserted on the fasync list
 883  * need to be released by rcu - see fasync_remove_entry.
 884  */
 885 void fasync_free(struct fasync_struct *new)
 886 {
 887         kmem_cache_free(fasync_cache, new);
 888 }
 889
 890 /*
 891  * Insert a new entry into the fasync list.  Return the pointer to the
 892  * old one if we didn't use the new one.
 893  *
 894  * NOTE! It is very important that the FASYNC flag always
 895  * match the state "is the filp on a fasync list".
 896  */
 897 struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
 898 {
 899         struct fasync_struct *fa, **fp;
 900
 901         spin_lock(&filp->f_lock);
 902         spin_lock(&fasync_lock);
 903         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 904                 if (fa->fa_file != filp)
 905                         continue;
 906
 907                 spin_lock_irq(&fa->fa_lock);
 908                 fa->fa_fd = fd;
 909                 spin_unlock_irq(&fa->fa_lock);
 910                 goto out;
 911         }
 912
 913         spin_lock_init(&new->fa_lock);
 914         new->magic = FASYNC_MAGIC;
 915         new->fa_file = filp;
 916         new->fa_fd = fd;
 917         new->fa_next = *fapp;
 918         rcu_assign_pointer(*fapp, new);
 919         filp->f_flags |= FASYNC;
 920
 921 out:
 922         spin_unlock(&fasync_lock);
 923         spin_unlock(&filp->f_lock);
 924         return fa;
 925 }
 926
 927 /*
 928  * Add a fasync entry. Return negative on error, positive if
 929  * added, and zero if did nothing but change an existing one.
 930  */
 931 static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
 932 {
 933         struct fasync_struct *new;
 934
 935         new = fasync_alloc();
 936         if (!new)
 937                 return -ENOMEM;
 938
 939         /*
 940          * fasync_insert_entry() returns the old (update) entry if
 941          * it existed.
 942          *
 943          * So free the (unused) new entry and return 0 to let the
 944          * caller know that we didn't add any new fasync entries.
 945          */
 946         if (fasync_insert_entry(fd, filp, fapp, new)) {
 947                 fasync_free(new);
 948                 return 0;
 949         }
 950
 951         return 1;
 952 }
 953
 954 /*
 955  * fasync_helper() is used by almost all character device drivers
 956  * to set up the fasync queue, and for regular files by the file
 957  * lease code. It returns negative on error, 0 if it did no changes
 958  * and positive if it added/deleted the entry.
 959  */
 960 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
 961 {
 962         if (!on)
 963                 return fasync_remove_entry(filp, fapp);
 964         return fasync_add_entry(fd, filp, fapp);
 965 }
 966
 967 EXPORT_SYMBOL(fasync_helper);
 968
 969 /*
 970  * rcu_read_lock() is held
 971  */
 972 static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
 973 {
 974         while (fa) {
 975                 struct fown_struct *fown;
 976                 unsigned long flags;
 977
 978                 if (fa->magic != FASYNC_MAGIC) {
 979                         printk(KERN_ERR "kill_fasync: bad magic number in "
 980                                "fasync_struct!\n");
 981                         return;
 982                 }
 983                 spin_lock_irqsave(&fa->fa_lock, flags);
 984                 if (fa->fa_file) {
 985                         fown = &fa->fa_file->f_owner;
 986                         /* Don't send SIGURG to processes which have not set a
 987                            queued signum: SIGURG has its own default signalling
 988                            mechanism. */
 989                         if (!(sig == SIGURG && fown->signum == 0))
 990                                 send_sigio(fown, fa->fa_fd, band);
 991                 }
 992                 spin_unlock_irqrestore(&fa->fa_lock, flags);
 993                 fa = rcu_dereference(fa->fa_next);
 994         }
 995 }
 996
 997 void kill_fasync(struct fasync_struct **fp, int sig, int band)
 998 {
 999         /* First a quick test without locking: usually
1000          * the list is empty.
1001          */
1002         if (*fp) {
1003                 rcu_read_lock();
1004                 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
1005                 rcu_read_unlock();
1006         }
1007 }
1008 EXPORT_SYMBOL(kill_fasync);
1009
1010 static int __init fcntl_init(void)
1011 {
1012         /*
1013          * Please add new bits here to ensure allocation uniqueness.
1014          * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
1015          * is defined as O_NONBLOCK on some platforms and not on others.
1016          */
1017         BUILD_BUG_ON(21 - 1 /* for O_RDONLY being 0 */ !=
1018                 HWEIGHT32(
1019                         (VALID_OPEN_FLAGS & ~(O_NONBLOCK | O_NDELAY)) |
1020                         __FMODE_EXEC | __FMODE_NONOTIFY));
1021
1022         fasync_cache = kmem_cache_create("fasync_cache",
1023                 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
1024         return 0;
1025 }
1026
1027 module_init(fcntl_init)