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/filelock.h>
  14 #include <linux/file.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 #include <linux/rw_hint.h>
  30
  31 #include <linux/poll.h>
  32 #include <asm/siginfo.h>
  33 #include <linux/uaccess.h>
  34
  35 #include "internal.h"
  36
  37 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
  38
  39 static int setfl(int fd, struct file * filp, unsigned int arg)
  40 {
  41         struct inode * inode = file_inode(filp);
  42         int error = 0;
  43
  44         /*
  45          * O_APPEND cannot be cleared if the file is marked as append-only
  46          * and the file is open for write.
  47          */
  48         if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
  49                 return -EPERM;
  50
  51         /* O_NOATIME can only be set by the owner or superuser */
  52         if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
  53                 if (!inode_owner_or_capable(file_mnt_idmap(filp), inode))
  54                         return -EPERM;
  55
  56         /* required for strict SunOS emulation */
  57         if (O_NONBLOCK != O_NDELAY)
  58                if (arg & O_NDELAY)
  59                    arg |= O_NONBLOCK;
  60
  61         /* Pipe packetized mode is controlled by O_DIRECT flag */
  62         if (!S_ISFIFO(inode->i_mode) &&
  63             (arg & O_DIRECT) &&
  64             !(filp->f_mode & FMODE_CAN_ODIRECT))
  65                 return -EINVAL;
  66
  67         if (filp->f_op->check_flags)
  68                 error = filp->f_op->check_flags(arg);
  69         if (error)
  70                 return error;
  71
  72         /*
  73          * ->fasync() is responsible for setting the FASYNC bit.
  74          */
  75         if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op->fasync) {
  76                 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
  77                 if (error < 0)
  78                         goto out;
  79                 if (error > 0)
  80                         error = 0;
  81         }
  82         spin_lock(&filp->f_lock);
  83         filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
  84         filp->f_iocb_flags = iocb_flags(filp);
  85         spin_unlock(&filp->f_lock);
  86
  87  out:
  88         return error;
  89 }
  90
  91 /*
  92  * Allocate an file->f_owner struct if it doesn't exist, handling racing
  93  * allocations correctly.
  94  */
  95 int file_f_owner_allocate(struct file *file)
  96 {
  97         struct fown_struct *f_owner;
  98
  99         f_owner = file_f_owner(file);
 100         if (f_owner)
 101                 return 0;
 102
 103         f_owner = kzalloc(sizeof(struct fown_struct), GFP_KERNEL);
 104         if (!f_owner)
 105                 return -ENOMEM;
 106
 107         rwlock_init(&f_owner->lock);
 108         f_owner->file = file;
 109         /* If someone else raced us, drop our allocation. */
 110         if (unlikely(cmpxchg(&file->f_owner, NULL, f_owner)))
 111                 kfree(f_owner);
 112         return 0;
 113 }
 114 EXPORT_SYMBOL(file_f_owner_allocate);
 115
 116 void file_f_owner_release(struct file *file)
 117 {
 118         struct fown_struct *f_owner;
 119
 120         f_owner = file_f_owner(file);
 121         if (f_owner) {
 122                 put_pid(f_owner->pid);
 123                 kfree(f_owner);
 124         }
 125 }
 126
 127 void __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
 128                 int force)
 129 {
 130         struct fown_struct *f_owner;
 131
 132         f_owner = file_f_owner(filp);
 133         if (WARN_ON_ONCE(!f_owner))
 134                 return;
 135
 136         write_lock_irq(&f_owner->lock);
 137         if (force || !f_owner->pid) {
 138                 put_pid(f_owner->pid);
 139                 f_owner->pid = get_pid(pid);
 140                 f_owner->pid_type = type;
 141
 142                 if (pid) {
 143                         const struct cred *cred = current_cred();
 144                         security_file_set_fowner(filp);
 145                         f_owner->uid = cred->uid;
 146                         f_owner->euid = cred->euid;
 147                 }
 148         }
 149         write_unlock_irq(&f_owner->lock);
 150 }
 151 EXPORT_SYMBOL(__f_setown);
 152
 153 int f_setown(struct file *filp, int who, int force)
 154 {
 155         enum pid_type type;
 156         struct pid *pid = NULL;
 157         int ret = 0;
 158
 159         might_sleep();
 160
 161         type = PIDTYPE_TGID;
 162         if (who < 0) {
 163                 /* avoid overflow below */
 164                 if (who == INT_MIN)
 165                         return -EINVAL;
 166
 167                 type = PIDTYPE_PGID;
 168                 who = -who;
 169         }
 170
 171         ret = file_f_owner_allocate(filp);
 172         if (ret)
 173                 return ret;
 174
 175         rcu_read_lock();
 176         if (who) {
 177                 pid = find_vpid(who);
 178                 if (!pid)
 179                         ret = -ESRCH;
 180         }
 181
 182         if (!ret)
 183                 __f_setown(filp, pid, type, force);
 184         rcu_read_unlock();
 185
 186         return ret;
 187 }
 188 EXPORT_SYMBOL(f_setown);
 189
 190 void f_delown(struct file *filp)
 191 {
 192         __f_setown(filp, NULL, PIDTYPE_TGID, 1);
 193 }
 194
 195 pid_t f_getown(struct file *filp)
 196 {
 197         pid_t pid = 0;
 198         struct fown_struct *f_owner;
 199
 200         f_owner = file_f_owner(filp);
 201         if (!f_owner)
 202                 return pid;
 203
 204         read_lock_irq(&f_owner->lock);
 205         rcu_read_lock();
 206         if (pid_task(f_owner->pid, f_owner->pid_type)) {
 207                 pid = pid_vnr(f_owner->pid);
 208                 if (f_owner->pid_type == PIDTYPE_PGID)
 209                         pid = -pid;
 210         }
 211         rcu_read_unlock();
 212         read_unlock_irq(&f_owner->lock);
 213         return pid;
 214 }
 215
 216 static int f_setown_ex(struct file *filp, unsigned long arg)
 217 {
 218         struct f_owner_ex __user *owner_p = (void __user *)arg;
 219         struct f_owner_ex owner;
 220         struct pid *pid;
 221         int type;
 222         int ret;
 223
 224         ret = copy_from_user(&owner, owner_p, sizeof(owner));
 225         if (ret)
 226                 return -EFAULT;
 227
 228         switch (owner.type) {
 229         case F_OWNER_TID:
 230                 type = PIDTYPE_PID;
 231                 break;
 232
 233         case F_OWNER_PID:
 234                 type = PIDTYPE_TGID;
 235                 break;
 236
 237         case F_OWNER_PGRP:
 238                 type = PIDTYPE_PGID;
 239                 break;
 240
 241         default:
 242                 return -EINVAL;
 243         }
 244
 245         ret = file_f_owner_allocate(filp);
 246         if (ret)
 247                 return ret;
 248
 249         rcu_read_lock();
 250         pid = find_vpid(owner.pid);
 251         if (owner.pid && !pid)
 252                 ret = -ESRCH;
 253         else
 254                  __f_setown(filp, pid, type, 1);
 255         rcu_read_unlock();
 256
 257         return ret;
 258 }
 259
 260 static int f_getown_ex(struct file *filp, unsigned long arg)
 261 {
 262         struct f_owner_ex __user *owner_p = (void __user *)arg;
 263         struct f_owner_ex owner = {};
 264         int ret = 0;
 265         struct fown_struct *f_owner;
 266         enum pid_type pid_type = PIDTYPE_PID;
 267
 268         f_owner = file_f_owner(filp);
 269         if (f_owner) {
 270                 read_lock_irq(&f_owner->lock);
 271                 rcu_read_lock();
 272                 if (pid_task(f_owner->pid, f_owner->pid_type))
 273                         owner.pid = pid_vnr(f_owner->pid);
 274                 rcu_read_unlock();
 275                 pid_type = f_owner->pid_type;
 276         }
 277
 278         switch (pid_type) {
 279         case PIDTYPE_PID:
 280                 owner.type = F_OWNER_TID;
 281                 break;
 282
 283         case PIDTYPE_TGID:
 284                 owner.type = F_OWNER_PID;
 285                 break;
 286
 287         case PIDTYPE_PGID:
 288                 owner.type = F_OWNER_PGRP;
 289                 break;
 290
 291         default:
 292                 WARN_ON(1);
 293                 ret = -EINVAL;
 294                 break;
 295         }
 296         if (f_owner)
 297                 read_unlock_irq(&f_owner->lock);
 298
 299         if (!ret) {
 300                 ret = copy_to_user(owner_p, &owner, sizeof(owner));
 301                 if (ret)
 302                         ret = -EFAULT;
 303         }
 304         return ret;
 305 }
 306
 307 #ifdef CONFIG_CHECKPOINT_RESTORE
 308 static int f_getowner_uids(struct file *filp, unsigned long arg)
 309 {
 310         struct user_namespace *user_ns = current_user_ns();
 311         struct fown_struct *f_owner;
 312         uid_t __user *dst = (void __user *)arg;
 313         uid_t src[2] = {0, 0};
 314         int err;
 315
 316         f_owner = file_f_owner(filp);
 317         if (f_owner) {
 318                 read_lock_irq(&f_owner->lock);
 319                 src[0] = from_kuid(user_ns, f_owner->uid);
 320                 src[1] = from_kuid(user_ns, f_owner->euid);
 321                 read_unlock_irq(&f_owner->lock);
 322         }
 323
 324         err  = put_user(src[0], &dst[0]);
 325         err |= put_user(src[1], &dst[1]);
 326
 327         return err;
 328 }
 329 #else
 330 static int f_getowner_uids(struct file *filp, unsigned long arg)
 331 {
 332         return -EINVAL;
 333 }
 334 #endif
 335
 336 static bool rw_hint_valid(u64 hint)
 337 {
 338         BUILD_BUG_ON(WRITE_LIFE_NOT_SET != RWH_WRITE_LIFE_NOT_SET);
 339         BUILD_BUG_ON(WRITE_LIFE_NONE != RWH_WRITE_LIFE_NONE);
 340         BUILD_BUG_ON(WRITE_LIFE_SHORT != RWH_WRITE_LIFE_SHORT);
 341         BUILD_BUG_ON(WRITE_LIFE_MEDIUM != RWH_WRITE_LIFE_MEDIUM);
 342         BUILD_BUG_ON(WRITE_LIFE_LONG != RWH_WRITE_LIFE_LONG);
 343         BUILD_BUG_ON(WRITE_LIFE_EXTREME != RWH_WRITE_LIFE_EXTREME);
 344
 345         switch (hint) {
 346         case RWH_WRITE_LIFE_NOT_SET:
 347         case RWH_WRITE_LIFE_NONE:
 348         case RWH_WRITE_LIFE_SHORT:
 349         case RWH_WRITE_LIFE_MEDIUM:
 350         case RWH_WRITE_LIFE_LONG:
 351         case RWH_WRITE_LIFE_EXTREME:
 352                 return true;
 353         default:
 354                 return false;
 355         }
 356 }
 357
 358 static long fcntl_get_rw_hint(struct file *file, unsigned int cmd,
 359                               unsigned long arg)
 360 {
 361         struct inode *inode = file_inode(file);
 362         u64 __user *argp = (u64 __user *)arg;
 363         u64 hint = READ_ONCE(inode->i_write_hint);
 364
 365         if (copy_to_user(argp, &hint, sizeof(*argp)))
 366                 return -EFAULT;
 367         return 0;
 368 }
 369
 370 static long fcntl_set_rw_hint(struct file *file, unsigned int cmd,
 371                               unsigned long arg)
 372 {
 373         struct inode *inode = file_inode(file);
 374         u64 __user *argp = (u64 __user *)arg;
 375         u64 hint;
 376
 377         if (copy_from_user(&hint, argp, sizeof(hint)))
 378                 return -EFAULT;
 379         if (!rw_hint_valid(hint))
 380                 return -EINVAL;
 381
 382         WRITE_ONCE(inode->i_write_hint, hint);
 383
 384         /*
 385          * file->f_mapping->host may differ from inode. As an example,
 386          * blkdev_open() modifies file->f_mapping.
 387          */
 388         if (file->f_mapping->host != inode)
 389                 WRITE_ONCE(file->f_mapping->host->i_write_hint, hint);
 390
 391         return 0;
 392 }
 393
 394 /* Is the file descriptor a dup of the file? */
 395 static long f_dupfd_query(int fd, struct file *filp)
 396 {
 397         CLASS(fd_raw, f)(fd);
 398
 399         if (fd_empty(f))
 400                 return -EBADF;
 401
 402         /*
 403          * We can do the 'fdput()' immediately, as the only thing that
 404          * matters is the pointer value which isn't changed by the fdput.
 405          *
 406          * Technically we didn't need a ref at all, and 'fdget()' was
 407          * overkill, but given our lockless file pointer lookup, the
 408          * alternatives are complicated.
 409          */
 410         return fd_file(f) == filp;
 411 }
 412
 413 /* Let the caller figure out whether a given file was just created. */
 414 static long f_created_query(const struct file *filp)
 415 {
 416         return !!(filp->f_mode & FMODE_CREATED);
 417 }
 418
 419 static int f_owner_sig(struct file *filp, int signum, bool setsig)
 420 {
 421         int ret = 0;
 422         struct fown_struct *f_owner;
 423
 424         might_sleep();
 425
 426         if (setsig) {
 427                 if (!valid_signal(signum))
 428                         return -EINVAL;
 429
 430                 ret = file_f_owner_allocate(filp);
 431                 if (ret)
 432                         return ret;
 433         }
 434
 435         f_owner = file_f_owner(filp);
 436         if (setsig)
 437                 f_owner->signum = signum;
 438         else if (f_owner)
 439                 ret = f_owner->signum;
 440         return ret;
 441 }
 442
 443 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
 444                 struct file *filp)
 445 {
 446         void __user *argp = (void __user *)arg;
 447         int argi = (int)arg;
 448         struct flock flock;
 449         long err = -EINVAL;
 450
 451         switch (cmd) {
 452         case F_CREATED_QUERY:
 453                 err = f_created_query(filp);
 454                 break;
 455         case F_DUPFD:
 456                 err = f_dupfd(argi, filp, 0);
 457                 break;
 458         case F_DUPFD_CLOEXEC:
 459                 err = f_dupfd(argi, filp, O_CLOEXEC);
 460                 break;
 461         case F_DUPFD_QUERY:
 462                 err = f_dupfd_query(argi, filp);
 463                 break;
 464         case F_GETFD:
 465                 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
 466                 break;
 467         case F_SETFD:
 468                 err = 0;
 469                 set_close_on_exec(fd, argi & FD_CLOEXEC);
 470                 break;
 471         case F_GETFL:
 472                 err = filp->f_flags;
 473                 break;
 474         case F_SETFL:
 475                 err = setfl(fd, filp, argi);
 476                 break;
 477 #if BITS_PER_LONG != 32
 478         /* 32-bit arches must use fcntl64() */
 479         case F_OFD_GETLK:
 480 #endif
 481         case F_GETLK:
 482                 if (copy_from_user(&flock, argp, sizeof(flock)))
 483                         return -EFAULT;
 484                 err = fcntl_getlk(filp, cmd, &flock);
 485                 if (!err && copy_to_user(argp, &flock, sizeof(flock)))
 486                         return -EFAULT;
 487                 break;
 488 #if BITS_PER_LONG != 32
 489         /* 32-bit arches must use fcntl64() */
 490         case F_OFD_SETLK:
 491         case F_OFD_SETLKW:
 492                 fallthrough;
 493 #endif
 494         case F_SETLK:
 495         case F_SETLKW:
 496                 if (copy_from_user(&flock, argp, sizeof(flock)))
 497                         return -EFAULT;
 498                 err = fcntl_setlk(fd, filp, cmd, &flock);
 499                 break;
 500         case F_GETOWN:
 501                 /*
 502                  * XXX If f_owner is a process group, the
 503                  * negative return value will get converted
 504                  * into an error.  Oops.  If we keep the
 505                  * current syscall conventions, the only way
 506                  * to fix this will be in libc.
 507                  */
 508                 err = f_getown(filp);
 509                 force_successful_syscall_return();
 510                 break;
 511         case F_SETOWN:
 512                 err = f_setown(filp, argi, 1);
 513                 break;
 514         case F_GETOWN_EX:
 515                 err = f_getown_ex(filp, arg);
 516                 break;
 517         case F_SETOWN_EX:
 518                 err = f_setown_ex(filp, arg);
 519                 break;
 520         case F_GETOWNER_UIDS:
 521                 err = f_getowner_uids(filp, arg);
 522                 break;
 523         case F_GETSIG:
 524                 err = f_owner_sig(filp, 0, false);
 525                 break;
 526         case F_SETSIG:
 527                 err = f_owner_sig(filp, argi, true);
 528                 break;
 529         case F_GETLEASE:
 530                 err = fcntl_getlease(filp);
 531                 break;
 532         case F_SETLEASE:
 533                 err = fcntl_setlease(fd, filp, argi);
 534                 break;
 535         case F_NOTIFY:
 536                 err = fcntl_dirnotify(fd, filp, argi);
 537                 break;
 538         case F_SETPIPE_SZ:
 539         case F_GETPIPE_SZ:
 540                 err = pipe_fcntl(filp, cmd, argi);
 541                 break;
 542         case F_ADD_SEALS:
 543         case F_GET_SEALS:
 544                 err = memfd_fcntl(filp, cmd, argi);
 545                 break;
 546         case F_GET_RW_HINT:
 547                 err = fcntl_get_rw_hint(filp, cmd, arg);
 548                 break;
 549         case F_SET_RW_HINT:
 550                 err = fcntl_set_rw_hint(filp, cmd, arg);
 551                 break;
 552         default:
 553                 break;
 554         }
 555         return err;
 556 }
 557
 558 static int check_fcntl_cmd(unsigned cmd)
 559 {
 560         switch (cmd) {
 561         case F_CREATED_QUERY:
 562         case F_DUPFD:
 563         case F_DUPFD_CLOEXEC:
 564         case F_DUPFD_QUERY:
 565         case F_GETFD:
 566         case F_SETFD:
 567         case F_GETFL:
 568                 return 1;
 569         }
 570         return 0;
 571 }
 572
 573 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
 574 {
 575         CLASS(fd_raw, f)(fd);
 576         long err;
 577
 578         if (fd_empty(f))
 579                 return -EBADF;
 580
 581         if (unlikely(fd_file(f)->f_mode & FMODE_PATH)) {
 582                 if (!check_fcntl_cmd(cmd))
 583                         return -EBADF;
 584         }
 585
 586         err = security_file_fcntl(fd_file(f), cmd, arg);
 587         if (!err)
 588                 err = do_fcntl(fd, cmd, arg, fd_file(f));
 589
 590         return err;
 591 }
 592
 593 #if BITS_PER_LONG == 32
 594 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
 595                 unsigned long, arg)
 596 {
 597         void __user *argp = (void __user *)arg;
 598         CLASS(fd_raw, f)(fd);
 599         struct flock64 flock;
 600         long err;
 601
 602         if (fd_empty(f))
 603                 return -EBADF;
 604
 605         if (unlikely(fd_file(f)->f_mode & FMODE_PATH)) {
 606                 if (!check_fcntl_cmd(cmd))
 607                         return -EBADF;
 608         }
 609
 610         err = security_file_fcntl(fd_file(f), cmd, arg);
 611         if (err)
 612                 return err;
 613
 614         switch (cmd) {
 615         case F_GETLK64:
 616         case F_OFD_GETLK:
 617                 err = -EFAULT;
 618                 if (copy_from_user(&flock, argp, sizeof(flock)))
 619                         break;
 620                 err = fcntl_getlk64(fd_file(f), cmd, &flock);
 621                 if (!err && copy_to_user(argp, &flock, sizeof(flock)))
 622                         err = -EFAULT;
 623                 break;
 624         case F_SETLK64:
 625         case F_SETLKW64:
 626         case F_OFD_SETLK:
 627         case F_OFD_SETLKW:
 628                 err = -EFAULT;
 629                 if (copy_from_user(&flock, argp, sizeof(flock)))
 630                         break;
 631                 err = fcntl_setlk64(fd, fd_file(f), cmd, &flock);
 632                 break;
 633         default:
 634                 err = do_fcntl(fd, cmd, arg, fd_file(f));
 635                 break;
 636         }
 637         return err;
 638 }
 639 #endif
 640
 641 #ifdef CONFIG_COMPAT
 642 /* careful - don't use anywhere else */
 643 #define copy_flock_fields(dst, src)             \
 644         (dst)->l_type = (src)->l_type;          \
 645         (dst)->l_whence = (src)->l_whence;      \
 646         (dst)->l_start = (src)->l_start;        \
 647         (dst)->l_len = (src)->l_len;            \
 648         (dst)->l_pid = (src)->l_pid;
 649
 650 static int get_compat_flock(struct flock *kfl, const struct compat_flock __user *ufl)
 651 {
 652         struct compat_flock fl;
 653
 654         if (copy_from_user(&fl, ufl, sizeof(struct compat_flock)))
 655                 return -EFAULT;
 656         copy_flock_fields(kfl, &fl);
 657         return 0;
 658 }
 659
 660 static int get_compat_flock64(struct flock *kfl, const struct compat_flock64 __user *ufl)
 661 {
 662         struct compat_flock64 fl;
 663
 664         if (copy_from_user(&fl, ufl, sizeof(struct compat_flock64)))
 665                 return -EFAULT;
 666         copy_flock_fields(kfl, &fl);
 667         return 0;
 668 }
 669
 670 static int put_compat_flock(const struct flock *kfl, struct compat_flock __user *ufl)
 671 {
 672         struct compat_flock fl;
 673
 674         memset(&fl, 0, sizeof(struct compat_flock));
 675         copy_flock_fields(&fl, kfl);
 676         if (copy_to_user(ufl, &fl, sizeof(struct compat_flock)))
 677                 return -EFAULT;
 678         return 0;
 679 }
 680
 681 static int put_compat_flock64(const struct flock *kfl, struct compat_flock64 __user *ufl)
 682 {
 683         struct compat_flock64 fl;
 684
 685         BUILD_BUG_ON(sizeof(kfl->l_start) > sizeof(ufl->l_start));
 686         BUILD_BUG_ON(sizeof(kfl->l_len) > sizeof(ufl->l_len));
 687
 688         memset(&fl, 0, sizeof(struct compat_flock64));
 689         copy_flock_fields(&fl, kfl);
 690         if (copy_to_user(ufl, &fl, sizeof(struct compat_flock64)))
 691                 return -EFAULT;
 692         return 0;
 693 }
 694 #undef copy_flock_fields
 695
 696 static unsigned int
 697 convert_fcntl_cmd(unsigned int cmd)
 698 {
 699         switch (cmd) {
 700         case F_GETLK64:
 701                 return F_GETLK;
 702         case F_SETLK64:
 703                 return F_SETLK;
 704         case F_SETLKW64:
 705                 return F_SETLKW;
 706         }
 707
 708         return cmd;
 709 }
 710
 711 /*
 712  * GETLK was successful and we need to return the data, but it needs to fit in
 713  * the compat structure.
 714  * l_start shouldn't be too big, unless the original start + end is greater than
 715  * COMPAT_OFF_T_MAX, in which case the app was asking for trouble, so we return
 716  * -EOVERFLOW in that case.  l_len could be too big, in which case we just
 717  * truncate it, and only allow the app to see that part of the conflicting lock
 718  * that might make sense to it anyway
 719  */
 720 static int fixup_compat_flock(struct flock *flock)
 721 {
 722         if (flock->l_start > COMPAT_OFF_T_MAX)
 723                 return -EOVERFLOW;
 724         if (flock->l_len > COMPAT_OFF_T_MAX)
 725                 flock->l_len = COMPAT_OFF_T_MAX;
 726         return 0;
 727 }
 728
 729 static long do_compat_fcntl64(unsigned int fd, unsigned int cmd,
 730                              compat_ulong_t arg)
 731 {
 732         CLASS(fd_raw, f)(fd);
 733         struct flock flock;
 734         long err;
 735
 736         if (fd_empty(f))
 737                 return -EBADF;
 738
 739         if (unlikely(fd_file(f)->f_mode & FMODE_PATH)) {
 740                 if (!check_fcntl_cmd(cmd))
 741                         return -EBADF;
 742         }
 743
 744         err = security_file_fcntl(fd_file(f), cmd, arg);
 745         if (err)
 746                 return err;
 747
 748         switch (cmd) {
 749         case F_GETLK:
 750                 err = get_compat_flock(&flock, compat_ptr(arg));
 751                 if (err)
 752                         break;
 753                 err = fcntl_getlk(fd_file(f), convert_fcntl_cmd(cmd), &flock);
 754                 if (err)
 755                         break;
 756                 err = fixup_compat_flock(&flock);
 757                 if (!err)
 758                         err = put_compat_flock(&flock, compat_ptr(arg));
 759                 break;
 760         case F_GETLK64:
 761         case F_OFD_GETLK:
 762                 err = get_compat_flock64(&flock, compat_ptr(arg));
 763                 if (err)
 764                         break;
 765                 err = fcntl_getlk(fd_file(f), convert_fcntl_cmd(cmd), &flock);
 766                 if (!err)
 767                         err = put_compat_flock64(&flock, compat_ptr(arg));
 768                 break;
 769         case F_SETLK:
 770         case F_SETLKW:
 771                 err = get_compat_flock(&flock, compat_ptr(arg));
 772                 if (err)
 773                         break;
 774                 err = fcntl_setlk(fd, fd_file(f), convert_fcntl_cmd(cmd), &flock);
 775                 break;
 776         case F_SETLK64:
 777         case F_SETLKW64:
 778         case F_OFD_SETLK:
 779         case F_OFD_SETLKW:
 780                 err = get_compat_flock64(&flock, compat_ptr(arg));
 781                 if (err)
 782                         break;
 783                 err = fcntl_setlk(fd, fd_file(f), convert_fcntl_cmd(cmd), &flock);
 784                 break;
 785         default:
 786                 err = do_fcntl(fd, cmd, arg, fd_file(f));
 787                 break;
 788         }
 789         return err;
 790 }
 791
 792 COMPAT_SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
 793                        compat_ulong_t, arg)
 794 {
 795         return do_compat_fcntl64(fd, cmd, arg);
 796 }
 797
 798 COMPAT_SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd,
 799                        compat_ulong_t, arg)
 800 {
 801         switch (cmd) {
 802         case F_GETLK64:
 803         case F_SETLK64:
 804         case F_SETLKW64:
 805         case F_OFD_GETLK:
 806         case F_OFD_SETLK:
 807         case F_OFD_SETLKW:
 808                 return -EINVAL;
 809         }
 810         return do_compat_fcntl64(fd, cmd, arg);
 811 }
 812 #endif
 813
 814 /* Table to convert sigio signal codes into poll band bitmaps */
 815
 816 static const __poll_t band_table[NSIGPOLL] = {
 817         EPOLLIN | EPOLLRDNORM,                  /* POLL_IN */
 818         EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND,   /* POLL_OUT */
 819         EPOLLIN | EPOLLRDNORM | EPOLLMSG,               /* POLL_MSG */
 820         EPOLLERR,                               /* POLL_ERR */
 821         EPOLLPRI | EPOLLRDBAND,                 /* POLL_PRI */
 822         EPOLLHUP | EPOLLERR                     /* POLL_HUP */
 823 };
 824
 825 static inline int sigio_perm(struct task_struct *p,
 826                              struct fown_struct *fown, int sig)
 827 {
 828         const struct cred *cred;
 829         int ret;
 830
 831         rcu_read_lock();
 832         cred = __task_cred(p);
 833         ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
 834                 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
 835                 uid_eq(fown->uid,  cred->suid) || uid_eq(fown->uid,  cred->uid)) &&
 836                !security_file_send_sigiotask(p, fown, sig));
 837         rcu_read_unlock();
 838         return ret;
 839 }
 840
 841 static void send_sigio_to_task(struct task_struct *p,
 842                                struct fown_struct *fown,
 843                                int fd, int reason, enum pid_type type)
 844 {
 845         /*
 846          * F_SETSIG can change ->signum lockless in parallel, make
 847          * sure we read it once and use the same value throughout.
 848          */
 849         int signum = READ_ONCE(fown->signum);
 850
 851         if (!sigio_perm(p, fown, signum))
 852                 return;
 853
 854         switch (signum) {
 855                 default: {
 856                         kernel_siginfo_t si;
 857
 858                         /* Queue a rt signal with the appropriate fd as its
 859                            value.  We use SI_SIGIO as the source, not
 860                            SI_KERNEL, since kernel signals always get
 861                            delivered even if we can't queue.  Failure to
 862                            queue in this case _should_ be reported; we fall
 863                            back to SIGIO in that case. --sct */
 864                         clear_siginfo(&si);
 865                         si.si_signo = signum;
 866                         si.si_errno = 0;
 867                         si.si_code  = reason;
 868                         /*
 869                          * Posix definies POLL_IN and friends to be signal
 870                          * specific si_codes for SIG_POLL.  Linux extended
 871                          * these si_codes to other signals in a way that is
 872                          * ambiguous if other signals also have signal
 873                          * specific si_codes.  In that case use SI_SIGIO instead
 874                          * to remove the ambiguity.
 875                          */
 876                         if ((signum != SIGPOLL) && sig_specific_sicodes(signum))
 877                                 si.si_code = SI_SIGIO;
 878
 879                         /* Make sure we are called with one of the POLL_*
 880                            reasons, otherwise we could leak kernel stack into
 881                            userspace.  */
 882                         BUG_ON((reason < POLL_IN) || ((reason - POLL_IN) >= NSIGPOLL));
 883                         if (reason - POLL_IN >= NSIGPOLL)
 884                                 si.si_band  = ~0L;
 885                         else
 886                                 si.si_band = mangle_poll(band_table[reason - POLL_IN]);
 887                         si.si_fd    = fd;
 888                         if (!do_send_sig_info(signum, &si, p, type))
 889                                 break;
 890                 }
 891                         fallthrough;    /* fall back on the old plain SIGIO signal */
 892                 case 0:
 893                         do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, type);
 894         }
 895 }
 896
 897 void send_sigio(struct fown_struct *fown, int fd, int band)
 898 {
 899         struct task_struct *p;
 900         enum pid_type type;
 901         unsigned long flags;
 902         struct pid *pid;
 903
 904         read_lock_irqsave(&fown->lock, flags);
 905
 906         type = fown->pid_type;
 907         pid = fown->pid;
 908         if (!pid)
 909                 goto out_unlock_fown;
 910
 911         if (type <= PIDTYPE_TGID) {
 912                 rcu_read_lock();
 913                 p = pid_task(pid, PIDTYPE_PID);
 914                 if (p)
 915                         send_sigio_to_task(p, fown, fd, band, type);
 916                 rcu_read_unlock();
 917         } else {
 918                 read_lock(&tasklist_lock);
 919                 do_each_pid_task(pid, type, p) {
 920                         send_sigio_to_task(p, fown, fd, band, type);
 921                 } while_each_pid_task(pid, type, p);
 922                 read_unlock(&tasklist_lock);
 923         }
 924  out_unlock_fown:
 925         read_unlock_irqrestore(&fown->lock, flags);
 926 }
 927
 928 static void send_sigurg_to_task(struct task_struct *p,
 929                                 struct fown_struct *fown, enum pid_type type)
 930 {
 931         if (sigio_perm(p, fown, SIGURG))
 932                 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, type);
 933 }
 934
 935 int send_sigurg(struct file *file)
 936 {
 937         struct fown_struct *fown;
 938         struct task_struct *p;
 939         enum pid_type type;
 940         struct pid *pid;
 941         unsigned long flags;
 942         int ret = 0;
 943
 944         fown = file_f_owner(file);
 945         if (!fown)
 946                 return 0;
 947
 948         read_lock_irqsave(&fown->lock, flags);
 949
 950         type = fown->pid_type;
 951         pid = fown->pid;
 952         if (!pid)
 953                 goto out_unlock_fown;
 954
 955         ret = 1;
 956
 957         if (type <= PIDTYPE_TGID) {
 958                 rcu_read_lock();
 959                 p = pid_task(pid, PIDTYPE_PID);
 960                 if (p)
 961                         send_sigurg_to_task(p, fown, type);
 962                 rcu_read_unlock();
 963         } else {
 964                 read_lock(&tasklist_lock);
 965                 do_each_pid_task(pid, type, p) {
 966                         send_sigurg_to_task(p, fown, type);
 967                 } while_each_pid_task(pid, type, p);
 968                 read_unlock(&tasklist_lock);
 969         }
 970  out_unlock_fown:
 971         read_unlock_irqrestore(&fown->lock, flags);
 972         return ret;
 973 }
 974
 975 static DEFINE_SPINLOCK(fasync_lock);
 976 static struct kmem_cache *fasync_cache __ro_after_init;
 977
 978 /*
 979  * Remove a fasync entry. If successfully removed, return
 980  * positive and clear the FASYNC flag. If no entry exists,
 981  * do nothing and return 0.
 982  *
 983  * NOTE! It is very important that the FASYNC flag always
 984  * match the state "is the filp on a fasync list".
 985  *
 986  */
 987 int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
 988 {
 989         struct fasync_struct *fa, **fp;
 990         int result = 0;
 991
 992         spin_lock(&filp->f_lock);
 993         spin_lock(&fasync_lock);
 994         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 995                 if (fa->fa_file != filp)
 996                         continue;
 997
 998                 write_lock_irq(&fa->fa_lock);
 999                 fa->fa_file = NULL;
1000                 write_unlock_irq(&fa->fa_lock);
1001
1002                 *fp = fa->fa_next;
1003                 kfree_rcu(fa, fa_rcu);
1004                 filp->f_flags &= ~FASYNC;
1005                 result = 1;
1006                 break;
1007         }
1008         spin_unlock(&fasync_lock);
1009         spin_unlock(&filp->f_lock);
1010         return result;
1011 }
1012
1013 struct fasync_struct *fasync_alloc(void)
1014 {
1015         return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
1016 }
1017
1018 /*
1019  * NOTE! This can be used only for unused fasync entries:
1020  * entries that actually got inserted on the fasync list
1021  * need to be released by rcu - see fasync_remove_entry.
1022  */
1023 void fasync_free(struct fasync_struct *new)
1024 {
1025         kmem_cache_free(fasync_cache, new);
1026 }
1027
1028 /*
1029  * Insert a new entry into the fasync list.  Return the pointer to the
1030  * old one if we didn't use the new one.
1031  *
1032  * NOTE! It is very important that the FASYNC flag always
1033  * match the state "is the filp on a fasync list".
1034  */
1035 struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
1036 {
1037         struct fasync_struct *fa, **fp;
1038
1039         spin_lock(&filp->f_lock);
1040         spin_lock(&fasync_lock);
1041         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
1042                 if (fa->fa_file != filp)
1043                         continue;
1044
1045                 write_lock_irq(&fa->fa_lock);
1046                 fa->fa_fd = fd;
1047                 write_unlock_irq(&fa->fa_lock);
1048                 goto out;
1049         }
1050
1051         rwlock_init(&new->fa_lock);
1052         new->magic = FASYNC_MAGIC;
1053         new->fa_file = filp;
1054         new->fa_fd = fd;
1055         new->fa_next = *fapp;
1056         rcu_assign_pointer(*fapp, new);
1057         filp->f_flags |= FASYNC;
1058
1059 out:
1060         spin_unlock(&fasync_lock);
1061         spin_unlock(&filp->f_lock);
1062         return fa;
1063 }
1064
1065 /*
1066  * Add a fasync entry. Return negative on error, positive if
1067  * added, and zero if did nothing but change an existing one.
1068  */
1069 static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
1070 {
1071         struct fasync_struct *new;
1072
1073         new = fasync_alloc();
1074         if (!new)
1075                 return -ENOMEM;
1076
1077         /*
1078          * fasync_insert_entry() returns the old (update) entry if
1079          * it existed.
1080          *
1081          * So free the (unused) new entry and return 0 to let the
1082          * caller know that we didn't add any new fasync entries.
1083          */
1084         if (fasync_insert_entry(fd, filp, fapp, new)) {
1085                 fasync_free(new);
1086                 return 0;
1087         }
1088
1089         return 1;
1090 }
1091
1092 /*
1093  * fasync_helper() is used by almost all character device drivers
1094  * to set up the fasync queue, and for regular files by the file
1095  * lease code. It returns negative on error, 0 if it did no changes
1096  * and positive if it added/deleted the entry.
1097  */
1098 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
1099 {
1100         if (!on)
1101                 return fasync_remove_entry(filp, fapp);
1102         return fasync_add_entry(fd, filp, fapp);
1103 }
1104
1105 EXPORT_SYMBOL(fasync_helper);
1106
1107 /*
1108  * rcu_read_lock() is held
1109  */
1110 static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
1111 {
1112         while (fa) {
1113                 struct fown_struct *fown;
1114                 unsigned long flags;
1115
1116                 if (fa->magic != FASYNC_MAGIC) {
1117                         printk(KERN_ERR "kill_fasync: bad magic number in "
1118                                "fasync_struct!\n");
1119                         return;
1120                 }
1121                 read_lock_irqsave(&fa->fa_lock, flags);
1122                 if (fa->fa_file) {
1123                         fown = file_f_owner(fa->fa_file);
1124                         if (!fown)
1125                                 goto next;
1126                         /* Don't send SIGURG to processes which have not set a
1127                            queued signum: SIGURG has its own default signalling
1128                            mechanism. */
1129                         if (!(sig == SIGURG && fown->signum == 0))
1130                                 send_sigio(fown, fa->fa_fd, band);
1131                 }
1132 next:
1133                 read_unlock_irqrestore(&fa->fa_lock, flags);
1134                 fa = rcu_dereference(fa->fa_next);
1135         }
1136 }
1137
1138 void kill_fasync(struct fasync_struct **fp, int sig, int band)
1139 {
1140         /* First a quick test without locking: usually
1141          * the list is empty.
1142          */
1143         if (*fp) {
1144                 rcu_read_lock();
1145                 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
1146                 rcu_read_unlock();
1147         }
1148 }
1149 EXPORT_SYMBOL(kill_fasync);
1150
1151 static int __init fcntl_init(void)
1152 {
1153         /*
1154          * Please add new bits here to ensure allocation uniqueness.
1155          * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
1156          * is defined as O_NONBLOCK on some platforms and not on others.
1157          */
1158         BUILD_BUG_ON(21 - 1 /* for O_RDONLY being 0 */ !=
1159                 HWEIGHT32(
1160                         (VALID_OPEN_FLAGS & ~(O_NONBLOCK | O_NDELAY)) |
1161                         __FMODE_EXEC | __FMODE_NONOTIFY));
1162
1163         fasync_cache = kmem_cache_create("fasync_cache",
1164                                          sizeof(struct fasync_struct), 0,
1165                                          SLAB_PANIC | SLAB_ACCOUNT, NULL);
1166         return 0;
1167 }
1168
1169 module_init(fcntl_init)