fs/fcntl.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/fcntl.c
   4  *
   5  *  Copyright (C) 1991, 1992  Linus Torvalds
   6  */
   7
   8 #include <linux/syscalls.h>
   9 #include <linux/init.h>
  10 #include <linux/mm.h>
  11 #include <linux/sched/task.h>
  12 #include <linux/fs.h>
  13 #include <linux/file.h>
  14 #include <linux/fdtable.h>
  15 #include <linux/capability.h>
  16 #include <linux/dnotify.h>
  17 #include <linux/slab.h>
  18 #include <linux/module.h>
  19 #include <linux/pipe_fs_i.h>
  20 #include <linux/security.h>
  21 #include <linux/ptrace.h>
  22 #include <linux/signal.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/pid_namespace.h>
  25 #include <linux/user_namespace.h>
  26 #include <linux/memfd.h>
  27 #include <linux/compat.h>
  28 #include <linux/mount.h>
  29
  30 #include <linux/poll.h>
  31 #include <asm/siginfo.h>
  32 #include <linux/uaccess.h>
  33
  34 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
  35
  36 static int setfl(int fd, struct file * filp, unsigned long arg)
  37 {
  38         struct inode * inode = file_inode(filp);
  39         int error = 0;
  40
  41         /*
  42          * O_APPEND cannot be cleared if the file is marked as append-only
  43          * and the file is open for write.
  44          */
  45         if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
  46                 return -EPERM;
  47
  48         /* O_NOATIME can only be set by the owner or superuser */
  49         if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
  50                 if (!inode_owner_or_capable(file_mnt_user_ns(filp), inode))
  51                         return -EPERM;
  52
  53         /* required for strict SunOS emulation */
  54         if (O_NONBLOCK != O_NDELAY)
  55                if (arg & O_NDELAY)
  56                    arg |= O_NONBLOCK;
  57
  58         /* Pipe packetized mode is controlled by O_DIRECT flag */
  59         if (!S_ISFIFO(inode->i_mode) && (arg & O_DIRECT)) {
  60                 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
  61                         !filp->f_mapping->a_ops->direct_IO)
  62                                 return -EINVAL;
  63         }
  64
  65         if (filp->f_op->check_flags)
  66                 error = filp->f_op->check_flags(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
  88 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
  89                      int force)
  90 {
  91         write_lock_irq(&filp->f_owner.lock);
  92         if (force || !filp->f_owner.pid) {
  93                 put_pid(filp->f_owner.pid);
  94                 filp->f_owner.pid = get_pid(pid);
  95                 filp->f_owner.pid_type = type;
  96
  97                 if (pid) {
  98                         const struct cred *cred = current_cred();
  99                         filp->f_owner.uid = cred->uid;
 100                         filp->f_owner.euid = cred->euid;
 101                 }
 102         }
 103         write_unlock_irq(&filp->f_owner.lock);
 104 }
 105
 106 void __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
 107                 int force)
 108 {
 109         security_file_set_fowner(filp);
 110         f_modown(filp, pid, type, force);
 111 }
 112 EXPORT_SYMBOL(__f_setown);
 113
 114 int f_setown(struct file *filp, unsigned long arg, int force)
 115 {
 116         enum pid_type type;
 117         struct pid *pid = NULL;
 118         int who = arg, ret = 0;
 119
 120         type = PIDTYPE_TGID;
 121         if (who < 0) {
 122                 /* avoid overflow below */
 123                 if (who == INT_MIN)
 124                         return -EINVAL;
 125
 126                 type = PIDTYPE_PGID;
 127                 who = -who;
 128         }
 129
 130         rcu_read_lock();
 131         if (who) {
 132                 pid = find_vpid(who);
 133                 if (!pid)
 134                         ret = -ESRCH;
 135         }
 136
 137         if (!ret)
 138                 __f_setown(filp, pid, type, force);
 139         rcu_read_unlock();
 140
 141         return ret;
 142 }
 143 EXPORT_SYMBOL(f_setown);
 144
 145 void f_delown(struct file *filp)
 146 {
 147         f_modown(filp, NULL, PIDTYPE_TGID, 1);
 148 }
 149
 150 pid_t f_getown(struct file *filp)
 151 {
 152         pid_t pid = 0;
 153         read_lock(&filp->f_owner.lock);
 154         rcu_read_lock();
 155         if (pid_task(filp->f_owner.pid, filp->f_owner.pid_type)) {
 156                 pid = pid_vnr(filp->f_owner.pid);
 157                 if (filp->f_owner.pid_type == PIDTYPE_PGID)
 158                         pid = -pid;
 159         }
 160         rcu_read_unlock();
 161         read_unlock(&filp->f_owner.lock);
 162         return pid;
 163 }
 164
 165 static int f_setown_ex(struct file *filp, unsigned long arg)
 166 {
 167         struct f_owner_ex __user *owner_p = (void __user *)arg;
 168         struct f_owner_ex owner;
 169         struct pid *pid;
 170         int type;
 171         int ret;
 172
 173         ret = copy_from_user(&owner, owner_p, sizeof(owner));
 174         if (ret)
 175                 return -EFAULT;
 176
 177         switch (owner.type) {
 178         case F_OWNER_TID:
 179                 type = PIDTYPE_PID;
 180                 break;
 181
 182         case F_OWNER_PID:
 183                 type = PIDTYPE_TGID;
 184                 break;
 185
 186         case F_OWNER_PGRP:
 187                 type = PIDTYPE_PGID;
 188                 break;
 189
 190         default:
 191                 return -EINVAL;
 192         }
 193
 194         rcu_read_lock();
 195         pid = find_vpid(owner.pid);
 196         if (owner.pid && !pid)
 197                 ret = -ESRCH;
 198         else
 199                  __f_setown(filp, pid, type, 1);
 200         rcu_read_unlock();
 201
 202         return ret;
 203 }
 204
 205 static int f_getown_ex(struct file *filp, unsigned long arg)
 206 {
 207         struct f_owner_ex __user *owner_p = (void __user *)arg;
 208         struct f_owner_ex owner = {};
 209         int ret = 0;
 210
 211         read_lock(&filp->f_owner.lock);
 212         rcu_read_lock();
 213         if (pid_task(filp->f_owner.pid, filp->f_owner.pid_type))
 214                 owner.pid = pid_vnr(filp->f_owner.pid);
 215         rcu_read_unlock();
 216         switch (filp->f_owner.pid_type) {
 217         case PIDTYPE_PID:
 218                 owner.type = F_OWNER_TID;
 219                 break;
 220
 221         case PIDTYPE_TGID:
 222                 owner.type = F_OWNER_PID;
 223                 break;
 224
 225         case PIDTYPE_PGID:
 226                 owner.type = F_OWNER_PGRP;
 227                 break;
 228
 229         default:
 230                 WARN_ON(1);
 231                 ret = -EINVAL;
 232                 break;
 233         }
 234         read_unlock(&filp->f_owner.lock);
 235
 236         if (!ret) {
 237                 ret = copy_to_user(owner_p, &owner, sizeof(owner));
 238                 if (ret)
 239                         ret = -EFAULT;
 240         }
 241         return ret;
 242 }
 243
 244 #ifdef CONFIG_CHECKPOINT_RESTORE
 245 static int f_getowner_uids(struct file *filp, unsigned long arg)
 246 {
 247         struct user_namespace *user_ns = current_user_ns();
 248         uid_t __user *dst = (void __user *)arg;
 249         uid_t src[2];
 250         int err;
 251
 252         read_lock(&filp->f_owner.lock);
 253         src[0] = from_kuid(user_ns, filp->f_owner.uid);
 254         src[1] = from_kuid(user_ns, filp->f_owner.euid);
 255         read_unlock(&filp->f_owner.lock);
 256
 257         err  = put_user(src[0], &dst[0]);
 258         err |= put_user(src[1], &dst[1]);
 259
 260         return err;
 261 }
 262 #else
 263 static int f_getowner_uids(struct file *filp, unsigned long arg)
 264 {
 265         return -EINVAL;
 266 }
 267 #endif
 268
 269 static bool rw_hint_valid(enum rw_hint hint)
 270 {
 271         switch (hint) {
 272         case RWH_WRITE_LIFE_NOT_SET:
 273         case RWH_WRITE_LIFE_NONE:
 274         case RWH_WRITE_LIFE_SHORT:
 275         case RWH_WRITE_LIFE_MEDIUM:
 276         case RWH_WRITE_LIFE_LONG:
 277         case RWH_WRITE_LIFE_EXTREME:
 278                 return true;
 279         default:
 280                 return false;
 281         }
 282 }
 283
 284 static long fcntl_rw_hint(struct file *file, unsigned int cmd,
 285                           unsigned long arg)
 286 {
 287         struct inode *inode = file_inode(file);
 288         u64 __user *argp = (u64 __user *)arg;
 289         enum rw_hint hint;
 290         u64 h;
 291
 292         switch (cmd) {
 293         case F_GET_FILE_RW_HINT:
 294                 h = file_write_hint(file);
 295                 if (copy_to_user(argp, &h, sizeof(*argp)))
 296                         return -EFAULT;
 297                 return 0;
 298         case F_SET_FILE_RW_HINT:
 299                 if (copy_from_user(&h, argp, sizeof(h)))
 300                         return -EFAULT;
 301                 hint = (enum rw_hint) h;
 302                 if (!rw_hint_valid(hint))
 303                         return -EINVAL;
 304
 305                 spin_lock(&file->f_lock);
 306                 file->f_write_hint = hint;
 307                 spin_unlock(&file->f_lock);
 308                 return 0;
 309         case F_GET_RW_HINT:
 310                 h = inode->i_write_hint;
 311                 if (copy_to_user(argp, &h, sizeof(*argp)))
 312                         return -EFAULT;
 313                 return 0;
 314         case F_SET_RW_HINT:
 315                 if (copy_from_user(&h, argp, sizeof(h)))
 316                         return -EFAULT;
 317                 hint = (enum rw_hint) h;
 318                 if (!rw_hint_valid(hint))
 319                         return -EINVAL;
 320
 321                 inode_lock(inode);
 322                 inode->i_write_hint = hint;
 323                 inode_unlock(inode);
 324                 return 0;
 325         default:
 326                 return -EINVAL;
 327         }
 328 }
 329
 330 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
 331                 struct file *filp)
 332 {
 333         void __user *argp = (void __user *)arg;
 334         struct flock flock;
 335         long err = -EINVAL;
 336
 337         switch (cmd) {
 338         case F_DUPFD:
 339                 err = f_dupfd(arg, filp, 0);
 340                 break;
 341         case F_DUPFD_CLOEXEC:
 342                 err = f_dupfd(arg, filp, O_CLOEXEC);
 343                 break;
 344         case F_GETFD:
 345                 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
 346                 break;
 347         case F_SETFD:
 348                 err = 0;
 349                 set_close_on_exec(fd, arg & FD_CLOEXEC);
 350                 break;
 351         case F_GETFL:
 352                 err = filp->f_flags;
 353                 break;
 354         case F_SETFL:
 355                 err = setfl(fd, filp, arg);
 356                 break;
 357 #if BITS_PER_LONG != 32
 358         /* 32-bit arches must use fcntl64() */
 359         case F_OFD_GETLK:
 360 #endif
 361         case F_GETLK:
 362                 if (copy_from_user(&flock, argp, sizeof(flock)))
 363                         return -EFAULT;
 364                 err = fcntl_getlk(filp, cmd, &flock);
 365                 if (!err && copy_to_user(argp, &flock, sizeof(flock)))
 366                         return -EFAULT;
 367                 break;
 368 #if BITS_PER_LONG != 32
 369         /* 32-bit arches must use fcntl64() */
 370         case F_OFD_SETLK:
 371         case F_OFD_SETLKW:
 372 #endif
 373                 fallthrough;
 374         case F_SETLK:
 375         case F_SETLKW:
 376                 if (copy_from_user(&flock, argp, sizeof(flock)))
 377                         return -EFAULT;
 378                 err = fcntl_setlk(fd, filp, cmd, &flock);
 379                 break;
 380         case F_GETOWN:
 381                 /*
 382                  * XXX If f_owner is a process group, the
 383                  * negative return value will get converted
 384                  * into an error.  Oops.  If we keep the
 385                  * current syscall conventions, the only way
 386                  * to fix this will be in libc.
 387                  */
 388                 err = f_getown(filp);
 389                 force_successful_syscall_return();
 390                 break;
 391         case F_SETOWN:
 392                 err = f_setown(filp, arg, 1);
 393                 break;
 394         case F_GETOWN_EX:
 395                 err = f_getown_ex(filp, arg);
 396                 break;
 397         case F_SETOWN_EX:
 398                 err = f_setown_ex(filp, arg);
 399                 break;
 400         case F_GETOWNER_UIDS:
 401                 err = f_getowner_uids(filp, arg);
 402                 break;
 403         case F_GETSIG:
 404                 err = filp->f_owner.signum;
 405                 break;
 406         case F_SETSIG:
 407                 /* arg == 0 restores default behaviour. */
 408                 if (!valid_signal(arg)) {
 409                         break;
 410                 }
 411                 err = 0;
 412                 filp->f_owner.signum = arg;
 413                 break;
 414         case F_GETLEASE:
 415                 err = fcntl_getlease(filp);
 416                 break;
 417         case F_SETLEASE:
 418                 err = fcntl_setlease(fd, filp, arg);
 419                 break;
 420         case F_NOTIFY:
 421                 err = fcntl_dirnotify(fd, filp, arg);
 422                 break;
 423         case F_SETPIPE_SZ:
 424         case F_GETPIPE_SZ:
 425                 err = pipe_fcntl(filp, cmd, arg);
 426                 break;
 427         case F_ADD_SEALS:
 428         case F_GET_SEALS:
 429                 err = memfd_fcntl(filp, cmd, arg);
 430                 break;
 431         case F_GET_RW_HINT:
 432         case F_SET_RW_HINT:
 433         case F_GET_FILE_RW_HINT:
 434         case F_SET_FILE_RW_HINT:
 435                 err = fcntl_rw_hint(filp, cmd, arg);
 436                 break;
 437         default:
 438                 break;
 439         }
 440         return err;
 441 }
 442
 443 static int check_fcntl_cmd(unsigned cmd)
 444 {
 445         switch (cmd) {
 446         case F_DUPFD:
 447         case F_DUPFD_CLOEXEC:
 448         case F_GETFD:
 449         case F_SETFD:
 450         case F_GETFL:
 451                 return 1;
 452         }
 453         return 0;
 454 }
 455
 456 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
 457 {
 458         struct fd f = fdget_raw(fd);
 459         long err = -EBADF;
 460
 461         if (!f.file)
 462                 goto out;
 463
 464         if (unlikely(f.file->f_mode & FMODE_PATH)) {
 465                 if (!check_fcntl_cmd(cmd))
 466                         goto out1;
 467         }
 468
 469         err = security_file_fcntl(f.file, cmd, arg);
 470         if (!err)
 471                 err = do_fcntl(fd, cmd, arg, f.file);
 472
 473 out1:
 474         fdput(f);
 475 out:
 476         return err;
 477 }
 478
 479 #if BITS_PER_LONG == 32
 480 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
 481                 unsigned long, arg)
 482 {
 483         void __user *argp = (void __user *)arg;
 484         struct fd f = fdget_raw(fd);
 485         struct flock64 flock;
 486         long err = -EBADF;
 487
 488         if (!f.file)
 489                 goto out;
 490
 491         if (unlikely(f.file->f_mode & FMODE_PATH)) {
 492                 if (!check_fcntl_cmd(cmd))
 493                         goto out1;
 494         }
 495
 496         err = security_file_fcntl(f.file, cmd, arg);
 497         if (err)
 498                 goto out1;
 499
 500         switch (cmd) {
 501         case F_GETLK64:
 502         case F_OFD_GETLK:
 503                 err = -EFAULT;
 504                 if (copy_from_user(&flock, argp, sizeof(flock)))
 505                         break;
 506                 err = fcntl_getlk64(f.file, cmd, &flock);
 507                 if (!err && copy_to_user(argp, &flock, sizeof(flock)))
 508                         err = -EFAULT;
 509                 break;
 510         case F_SETLK64:
 511         case F_SETLKW64:
 512         case F_OFD_SETLK:
 513         case F_OFD_SETLKW:
 514                 err = -EFAULT;
 515                 if (copy_from_user(&flock, argp, sizeof(flock)))
 516                         break;
 517                 err = fcntl_setlk64(fd, f.file, cmd, &flock);
 518                 break;
 519         default:
 520                 err = do_fcntl(fd, cmd, arg, f.file);
 521                 break;
 522         }
 523 out1:
 524         fdput(f);
 525 out:
 526         return err;
 527 }
 528 #endif
 529
 530 #ifdef CONFIG_COMPAT
 531 /* careful - don't use anywhere else */
 532 #define copy_flock_fields(dst, src)             \
 533         (dst)->l_type = (src)->l_type;          \
 534         (dst)->l_whence = (src)->l_whence;      \
 535         (dst)->l_start = (src)->l_start;        \
 536         (dst)->l_len = (src)->l_len;            \
 537         (dst)->l_pid = (src)->l_pid;
 538
 539 static int get_compat_flock(struct flock *kfl, const struct compat_flock __user *ufl)
 540 {
 541         struct compat_flock fl;
 542
 543         if (copy_from_user(&fl, ufl, sizeof(struct compat_flock)))
 544                 return -EFAULT;
 545         copy_flock_fields(kfl, &fl);
 546         return 0;
 547 }
 548
 549 static int get_compat_flock64(struct flock *kfl, const struct compat_flock64 __user *ufl)
 550 {
 551         struct compat_flock64 fl;
 552
 553         if (copy_from_user(&fl, ufl, sizeof(struct compat_flock64)))
 554                 return -EFAULT;
 555         copy_flock_fields(kfl, &fl);
 556         return 0;
 557 }
 558
 559 static int put_compat_flock(const struct flock *kfl, struct compat_flock __user *ufl)
 560 {
 561         struct compat_flock fl;
 562
 563         memset(&fl, 0, sizeof(struct compat_flock));
 564         copy_flock_fields(&fl, kfl);
 565         if (copy_to_user(ufl, &fl, sizeof(struct compat_flock)))
 566                 return -EFAULT;
 567         return 0;
 568 }
 569
 570 static int put_compat_flock64(const struct flock *kfl, struct compat_flock64 __user *ufl)
 571 {
 572         struct compat_flock64 fl;
 573
 574         BUILD_BUG_ON(sizeof(kfl->l_start) > sizeof(ufl->l_start));
 575         BUILD_BUG_ON(sizeof(kfl->l_len) > sizeof(ufl->l_len));
 576
 577         memset(&fl, 0, sizeof(struct compat_flock64));
 578         copy_flock_fields(&fl, kfl);
 579         if (copy_to_user(ufl, &fl, sizeof(struct compat_flock64)))
 580                 return -EFAULT;
 581         return 0;
 582 }
 583 #undef copy_flock_fields
 584
 585 static unsigned int
 586 convert_fcntl_cmd(unsigned int cmd)
 587 {
 588         switch (cmd) {
 589         case F_GETLK64:
 590                 return F_GETLK;
 591         case F_SETLK64:
 592                 return F_SETLK;
 593         case F_SETLKW64:
 594                 return F_SETLKW;
 595         }
 596
 597         return cmd;
 598 }
 599
 600 /*
 601  * GETLK was successful and we need to return the data, but it needs to fit in
 602  * the compat structure.
 603  * l_start shouldn't be too big, unless the original start + end is greater than
 604  * COMPAT_OFF_T_MAX, in which case the app was asking for trouble, so we return
 605  * -EOVERFLOW in that case.  l_len could be too big, in which case we just
 606  * truncate it, and only allow the app to see that part of the conflicting lock
 607  * that might make sense to it anyway
 608  */
 609 static int fixup_compat_flock(struct flock *flock)
 610 {
 611         if (flock->l_start > COMPAT_OFF_T_MAX)
 612                 return -EOVERFLOW;
 613         if (flock->l_len > COMPAT_OFF_T_MAX)
 614                 flock->l_len = COMPAT_OFF_T_MAX;
 615         return 0;
 616 }
 617
 618 static long do_compat_fcntl64(unsigned int fd, unsigned int cmd,
 619                              compat_ulong_t arg)
 620 {
 621         struct fd f = fdget_raw(fd);
 622         struct flock flock;
 623         long err = -EBADF;
 624
 625         if (!f.file)
 626                 return err;
 627
 628         if (unlikely(f.file->f_mode & FMODE_PATH)) {
 629                 if (!check_fcntl_cmd(cmd))
 630                         goto out_put;
 631         }
 632
 633         err = security_file_fcntl(f.file, cmd, arg);
 634         if (err)
 635                 goto out_put;
 636
 637         switch (cmd) {
 638         case F_GETLK:
 639                 err = get_compat_flock(&flock, compat_ptr(arg));
 640                 if (err)
 641                         break;
 642                 err = fcntl_getlk(f.file, convert_fcntl_cmd(cmd), &flock);
 643                 if (err)
 644                         break;
 645                 err = fixup_compat_flock(&flock);
 646                 if (!err)
 647                         err = put_compat_flock(&flock, compat_ptr(arg));
 648                 break;
 649         case F_GETLK64:
 650         case F_OFD_GETLK:
 651                 err = get_compat_flock64(&flock, compat_ptr(arg));
 652                 if (err)
 653                         break;
 654                 err = fcntl_getlk(f.file, convert_fcntl_cmd(cmd), &flock);
 655                 if (!err)
 656                         err = put_compat_flock64(&flock, compat_ptr(arg));
 657                 break;
 658         case F_SETLK:
 659         case F_SETLKW:
 660                 err = get_compat_flock(&flock, compat_ptr(arg));
 661                 if (err)
 662                         break;
 663                 err = fcntl_setlk(fd, f.file, convert_fcntl_cmd(cmd), &flock);
 664                 break;
 665         case F_SETLK64:
 666         case F_SETLKW64:
 667         case F_OFD_SETLK:
 668         case F_OFD_SETLKW:
 669                 err = get_compat_flock64(&flock, compat_ptr(arg));
 670                 if (err)
 671                         break;
 672                 err = fcntl_setlk(fd, f.file, convert_fcntl_cmd(cmd), &flock);
 673                 break;
 674         default:
 675                 err = do_fcntl(fd, cmd, arg, f.file);
 676                 break;
 677         }
 678 out_put:
 679         fdput(f);
 680         return err;
 681 }
 682
 683 COMPAT_SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
 684                        compat_ulong_t, arg)
 685 {
 686         return do_compat_fcntl64(fd, cmd, arg);
 687 }
 688
 689 COMPAT_SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd,
 690                        compat_ulong_t, arg)
 691 {
 692         switch (cmd) {
 693         case F_GETLK64:
 694         case F_SETLK64:
 695         case F_SETLKW64:
 696         case F_OFD_GETLK:
 697         case F_OFD_SETLK:
 698         case F_OFD_SETLKW:
 699                 return -EINVAL;
 700         }
 701         return do_compat_fcntl64(fd, cmd, arg);
 702 }
 703 #endif
 704
 705 /* Table to convert sigio signal codes into poll band bitmaps */
 706
 707 static const __poll_t band_table[NSIGPOLL] = {
 708         EPOLLIN | EPOLLRDNORM,                  /* POLL_IN */
 709         EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND,   /* POLL_OUT */
 710         EPOLLIN | EPOLLRDNORM | EPOLLMSG,               /* POLL_MSG */
 711         EPOLLERR,                               /* POLL_ERR */
 712         EPOLLPRI | EPOLLRDBAND,                 /* POLL_PRI */
 713         EPOLLHUP | EPOLLERR                     /* POLL_HUP */
 714 };
 715
 716 static inline int sigio_perm(struct task_struct *p,
 717                              struct fown_struct *fown, int sig)
 718 {
 719         const struct cred *cred;
 720         int ret;
 721
 722         rcu_read_lock();
 723         cred = __task_cred(p);
 724         ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
 725                 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
 726                 uid_eq(fown->uid,  cred->suid) || uid_eq(fown->uid,  cred->uid)) &&
 727                !security_file_send_sigiotask(p, fown, sig));
 728         rcu_read_unlock();
 729         return ret;
 730 }
 731
 732 static void send_sigio_to_task(struct task_struct *p,
 733                                struct fown_struct *fown,
 734                                int fd, int reason, enum pid_type type)
 735 {
 736         /*
 737          * F_SETSIG can change ->signum lockless in parallel, make
 738          * sure we read it once and use the same value throughout.
 739          */
 740         int signum = READ_ONCE(fown->signum);
 741
 742         if (!sigio_perm(p, fown, signum))
 743                 return;
 744
 745         switch (signum) {
 746                 default: {
 747                         kernel_siginfo_t si;
 748
 749                         /* Queue a rt signal with the appropriate fd as its
 750                            value.  We use SI_SIGIO as the source, not
 751                            SI_KERNEL, since kernel signals always get
 752                            delivered even if we can't queue.  Failure to
 753                            queue in this case _should_ be reported; we fall
 754                            back to SIGIO in that case. --sct */
 755                         clear_siginfo(&si);
 756                         si.si_signo = signum;
 757                         si.si_errno = 0;
 758                         si.si_code  = reason;
 759                         /*
 760                          * Posix definies POLL_IN and friends to be signal
 761                          * specific si_codes for SIG_POLL.  Linux extended
 762                          * these si_codes to other signals in a way that is
 763                          * ambiguous if other signals also have signal
 764                          * specific si_codes.  In that case use SI_SIGIO instead
 765                          * to remove the ambiguity.
 766                          */
 767                         if ((signum != SIGPOLL) && sig_specific_sicodes(signum))
 768                                 si.si_code = SI_SIGIO;
 769
 770                         /* Make sure we are called with one of the POLL_*
 771                            reasons, otherwise we could leak kernel stack into
 772                            userspace.  */
 773                         BUG_ON((reason < POLL_IN) || ((reason - POLL_IN) >= NSIGPOLL));
 774                         if (reason - POLL_IN >= NSIGPOLL)
 775                                 si.si_band  = ~0L;
 776                         else
 777                                 si.si_band = mangle_poll(band_table[reason - POLL_IN]);
 778                         si.si_fd    = fd;
 779                         if (!do_send_sig_info(signum, &si, p, type))
 780                                 break;
 781                 }
 782                         fallthrough;    /* fall back on the old plain SIGIO signal */
 783                 case 0:
 784                         do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, type);
 785         }
 786 }
 787
 788 void send_sigio(struct fown_struct *fown, int fd, int band)
 789 {
 790         struct task_struct *p;
 791         enum pid_type type;
 792         unsigned long flags;
 793         struct pid *pid;
 794
 795         read_lock_irqsave(&fown->lock, flags);
 796
 797         type = fown->pid_type;
 798         pid = fown->pid;
 799         if (!pid)
 800                 goto out_unlock_fown;
 801
 802         if (type <= PIDTYPE_TGID) {
 803                 rcu_read_lock();
 804                 p = pid_task(pid, PIDTYPE_PID);
 805                 if (p)
 806                         send_sigio_to_task(p, fown, fd, band, type);
 807                 rcu_read_unlock();
 808         } else {
 809                 read_lock(&tasklist_lock);
 810                 do_each_pid_task(pid, type, p) {
 811                         send_sigio_to_task(p, fown, fd, band, type);
 812                 } while_each_pid_task(pid, type, p);
 813                 read_unlock(&tasklist_lock);
 814         }
 815  out_unlock_fown:
 816         read_unlock_irqrestore(&fown->lock, flags);
 817 }
 818
 819 static void send_sigurg_to_task(struct task_struct *p,
 820                                 struct fown_struct *fown, enum pid_type type)
 821 {
 822         if (sigio_perm(p, fown, SIGURG))
 823                 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, type);
 824 }
 825
 826 int send_sigurg(struct fown_struct *fown)
 827 {
 828         struct task_struct *p;
 829         enum pid_type type;
 830         struct pid *pid;
 831         unsigned long flags;
 832         int ret = 0;
 833
 834         read_lock_irqsave(&fown->lock, flags);
 835
 836         type = fown->pid_type;
 837         pid = fown->pid;
 838         if (!pid)
 839                 goto out_unlock_fown;
 840
 841         ret = 1;
 842
 843         if (type <= PIDTYPE_TGID) {
 844                 rcu_read_lock();
 845                 p = pid_task(pid, PIDTYPE_PID);
 846                 if (p)
 847                         send_sigurg_to_task(p, fown, type);
 848                 rcu_read_unlock();
 849         } else {
 850                 read_lock(&tasklist_lock);
 851                 do_each_pid_task(pid, type, p) {
 852                         send_sigurg_to_task(p, fown, type);
 853                 } while_each_pid_task(pid, type, p);
 854                 read_unlock(&tasklist_lock);
 855         }
 856  out_unlock_fown:
 857         read_unlock_irqrestore(&fown->lock, flags);
 858         return ret;
 859 }
 860
 861 static DEFINE_SPINLOCK(fasync_lock);
 862 static struct kmem_cache *fasync_cache __read_mostly;
 863
 864 static void fasync_free_rcu(struct rcu_head *head)
 865 {
 866         kmem_cache_free(fasync_cache,
 867                         container_of(head, struct fasync_struct, fa_rcu));
 868 }
 869
 870 /*
 871  * Remove a fasync entry. If successfully removed, return
 872  * positive and clear the FASYNC flag. If no entry exists,
 873  * do nothing and return 0.
 874  *
 875  * NOTE! It is very important that the FASYNC flag always
 876  * match the state "is the filp on a fasync list".
 877  *
 878  */
 879 int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
 880 {
 881         struct fasync_struct *fa, **fp;
 882         int result = 0;
 883
 884         spin_lock(&filp->f_lock);
 885         spin_lock(&fasync_lock);
 886         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 887                 if (fa->fa_file != filp)
 888                         continue;
 889
 890                 write_lock_irq(&fa->fa_lock);
 891                 fa->fa_file = NULL;
 892                 write_unlock_irq(&fa->fa_lock);
 893
 894                 *fp = fa->fa_next;
 895                 call_rcu(&fa->fa_rcu, fasync_free_rcu);
 896                 filp->f_flags &= ~FASYNC;
 897                 result = 1;
 898                 break;
 899         }
 900         spin_unlock(&fasync_lock);
 901         spin_unlock(&filp->f_lock);
 902         return result;
 903 }
 904
 905 struct fasync_struct *fasync_alloc(void)
 906 {
 907         return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
 908 }
 909
 910 /*
 911  * NOTE! This can be used only for unused fasync entries:
 912  * entries that actually got inserted on the fasync list
 913  * need to be released by rcu - see fasync_remove_entry.
 914  */
 915 void fasync_free(struct fasync_struct *new)
 916 {
 917         kmem_cache_free(fasync_cache, new);
 918 }
 919
 920 /*
 921  * Insert a new entry into the fasync list.  Return the pointer to the
 922  * old one if we didn't use the new one.
 923  *
 924  * NOTE! It is very important that the FASYNC flag always
 925  * match the state "is the filp on a fasync list".
 926  */
 927 struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
 928 {
 929         struct fasync_struct *fa, **fp;
 930
 931         spin_lock(&filp->f_lock);
 932         spin_lock(&fasync_lock);
 933         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 934                 if (fa->fa_file != filp)
 935                         continue;
 936
 937                 write_lock_irq(&fa->fa_lock);
 938                 fa->fa_fd = fd;
 939                 write_unlock_irq(&fa->fa_lock);
 940                 goto out;
 941         }
 942
 943         rwlock_init(&new->fa_lock);
 944         new->magic = FASYNC_MAGIC;
 945         new->fa_file = filp;
 946         new->fa_fd = fd;
 947         new->fa_next = *fapp;
 948         rcu_assign_pointer(*fapp, new);
 949         filp->f_flags |= FASYNC;
 950
 951 out:
 952         spin_unlock(&fasync_lock);
 953         spin_unlock(&filp->f_lock);
 954         return fa;
 955 }
 956
 957 /*
 958  * Add a fasync entry. Return negative on error, positive if
 959  * added, and zero if did nothing but change an existing one.
 960  */
 961 static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
 962 {
 963         struct fasync_struct *new;
 964
 965         new = fasync_alloc();
 966         if (!new)
 967                 return -ENOMEM;
 968
 969         /*
 970          * fasync_insert_entry() returns the old (update) entry if
 971          * it existed.
 972          *
 973          * So free the (unused) new entry and return 0 to let the
 974          * caller know that we didn't add any new fasync entries.
 975          */
 976         if (fasync_insert_entry(fd, filp, fapp, new)) {
 977                 fasync_free(new);
 978                 return 0;
 979         }
 980
 981         return 1;
 982 }
 983
 984 /*
 985  * fasync_helper() is used by almost all character device drivers
 986  * to set up the fasync queue, and for regular files by the file
 987  * lease code. It returns negative on error, 0 if it did no changes
 988  * and positive if it added/deleted the entry.
 989  */
 990 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
 991 {
 992         if (!on)
 993                 return fasync_remove_entry(filp, fapp);
 994         return fasync_add_entry(fd, filp, fapp);
 995 }
 996
 997 EXPORT_SYMBOL(fasync_helper);
 998
 999 /*
1000  * rcu_read_lock() is held
1001  */
1002 static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
1003 {
1004         while (fa) {
1005                 struct fown_struct *fown;
1006
1007                 if (fa->magic != FASYNC_MAGIC) {
1008                         printk(KERN_ERR "kill_fasync: bad magic number in "
1009                                "fasync_struct!\n");
1010                         return;
1011                 }
1012                 read_lock(&fa->fa_lock);
1013                 if (fa->fa_file) {
1014                         fown = &fa->fa_file->f_owner;
1015                         /* Don't send SIGURG to processes which have not set a
1016                            queued signum: SIGURG has its own default signalling
1017                            mechanism. */
1018                         if (!(sig == SIGURG && fown->signum == 0))
1019                                 send_sigio(fown, fa->fa_fd, band);
1020                 }
1021                 read_unlock(&fa->fa_lock);
1022                 fa = rcu_dereference(fa->fa_next);
1023         }
1024 }
1025
1026 void kill_fasync(struct fasync_struct **fp, int sig, int band)
1027 {
1028         /* First a quick test without locking: usually
1029          * the list is empty.
1030          */
1031         if (*fp) {
1032                 rcu_read_lock();
1033                 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
1034                 rcu_read_unlock();
1035         }
1036 }
1037 EXPORT_SYMBOL(kill_fasync);
1038
1039 static int __init fcntl_init(void)
1040 {
1041         /*
1042          * Please add new bits here to ensure allocation uniqueness.
1043          * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
1044          * is defined as O_NONBLOCK on some platforms and not on others.
1045          */
1046         BUILD_BUG_ON(21 - 1 /* for O_RDONLY being 0 */ !=
1047                 HWEIGHT32(
1048                         (VALID_OPEN_FLAGS & ~(O_NONBLOCK | O_NDELAY)) |
1049                         __FMODE_EXEC | __FMODE_NONOTIFY));
1050
1051         fasync_cache = kmem_cache_create("fasync_cache",
1052                 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
1053         return 0;
1054 }
1055
1056 module_init(fcntl_init)