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/fdtable.h> 16#include <linux/capability.h> 17#include <linux/dnotify.h> 18#include <linux/slab.h> 19#include <linux/module.h> 20#include <linux/pipe_fs_i.h> 21#include <linux/security.h> 22#include <linux/ptrace.h> 23#include <linux/signal.h> 24#include <linux/rcupdate.h> 25#include <linux/pid_namespace.h> 26#include <linux/user_namespace.h> 27#include <linux/memfd.h> 28#include <linux/compat.h> 29#include <linux/mount.h> 30#include <linux/rw_hint.h> 31 32#include <linux/poll.h> 33#include <asm/siginfo.h> 34#include <linux/uaccess.h> 35 36#define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME) 37 38static int setfl(int fd, struct file * filp, unsigned int arg) 39{ 40 struct inode * inode = file_inode(filp); 41 int error = 0; 42 43 /* 44 * O_APPEND cannot be cleared if the file is marked as append-only 45 * and the file is open for write. 46 */ 47 if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode)) 48 return -EPERM; 49 50 /* O_NOATIME can only be set by the owner or superuser */ 51 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME)) 52 if (!inode_owner_or_capable(file_mnt_idmap(filp), inode)) 53 return -EPERM; 54 55 /* required for strict SunOS emulation */ 56 if (O_NONBLOCK != O_NDELAY) 57 if (arg & O_NDELAY) 58 arg |= O_NONBLOCK; 59 60 /* Pipe packetized mode is controlled by O_DIRECT flag */ 61 if (!S_ISFIFO(inode->i_mode) && 62 (arg & O_DIRECT) && 63 !(filp->f_mode & FMODE_CAN_ODIRECT)) 64 return -EINVAL; 65 66 if (filp->f_op->check_flags) 67 error = filp->f_op->check_flags(arg); 68 if (error) 69 return error; 70 71 /* 72 * ->fasync() is responsible for setting the FASYNC bit. 73 */ 74 if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op->fasync) { 75 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0); 76 if (error < 0) 77 goto out; 78 if (error > 0) 79 error = 0; 80 } 81 spin_lock(&filp->f_lock); 82 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK); 83 filp->f_iocb_flags = iocb_flags(filp); 84 spin_unlock(&filp->f_lock); 85 86 out: 87 return error; 88} 89 90static void f_modown(struct file *filp, struct pid *pid, enum pid_type type, 91 int force) 92{ 93 write_lock_irq(&filp->f_owner.lock); 94 if (force || !filp->f_owner.pid) { 95 put_pid(filp->f_owner.pid); 96 filp->f_owner.pid = get_pid(pid); 97 filp->f_owner.pid_type = type; 98 99 if (pid) { 100 const struct cred *cred = current_cred(); 101 filp->f_owner.uid = cred->uid; 102 filp->f_owner.euid = cred->euid; 103 } 104 } 105 write_unlock_irq(&filp->f_owner.lock); 106} 107 108void __f_setown(struct file *filp, struct pid *pid, enum pid_type type, 109 int force) 110{ 111 security_file_set_fowner(filp); 112 f_modown(filp, pid, type, force); 113} 114EXPORT_SYMBOL(__f_setown); 115 116int f_setown(struct file *filp, int who, int force) 117{ 118 enum pid_type type; 119 struct pid *pid = NULL; 120 int ret = 0; 121 122 type = PIDTYPE_TGID; 123 if (who < 0) { 124 /* avoid overflow below */ 125 if (who == INT_MIN) 126 return -EINVAL; 127 128 type = PIDTYPE_PGID; 129 who = -who; 130 } 131 132 rcu_read_lock(); 133 if (who) { 134 pid = find_vpid(who); 135 if (!pid) 136 ret = -ESRCH; 137 } 138 139 if (!ret) 140 __f_setown(filp, pid, type, force); 141 rcu_read_unlock(); 142 143 return ret; 144} 145EXPORT_SYMBOL(f_setown); 146 147void f_delown(struct file *filp) 148{ 149 f_modown(filp, NULL, PIDTYPE_TGID, 1); 150} 151 152pid_t f_getown(struct file *filp) 153{ 154 pid_t pid = 0; 155 156 read_lock_irq(&filp->f_owner.lock); 157 rcu_read_lock(); 158 if (pid_task(filp->f_owner.pid, filp->f_owner.pid_type)) { 159 pid = pid_vnr(filp->f_owner.pid); 160 if (filp->f_owner.pid_type == PIDTYPE_PGID) 161 pid = -pid; 162 } 163 rcu_read_unlock(); 164 read_unlock_irq(&filp->f_owner.lock); 165 return pid; 166} 167 168static int f_setown_ex(struct file *filp, unsigned long arg) 169{ 170 struct f_owner_ex __user *owner_p = (void __user *)arg; 171 struct f_owner_ex owner; 172 struct pid *pid; 173 int type; 174 int ret; 175 176 ret = copy_from_user(&owner, owner_p, sizeof(owner)); 177 if (ret) 178 return -EFAULT; 179 180 switch (owner.type) { 181 case F_OWNER_TID: 182 type = PIDTYPE_PID; 183 break; 184 185 case F_OWNER_PID: 186 type = PIDTYPE_TGID; 187 break; 188 189 case F_OWNER_PGRP: 190 type = PIDTYPE_PGID; 191 break; 192 193 default: 194 return -EINVAL; 195 } 196 197 rcu_read_lock(); 198 pid = find_vpid(owner.pid); 199 if (owner.pid && !pid) 200 ret = -ESRCH; 201 else 202 __f_setown(filp, pid, type, 1); 203 rcu_read_unlock(); 204 205 return ret; 206} 207 208static int f_getown_ex(struct file *filp, unsigned long arg) 209{ 210 struct f_owner_ex __user *owner_p = (void __user *)arg; 211 struct f_owner_ex owner = {}; 212 int ret = 0; 213 214 read_lock_irq(&filp->f_owner.lock); 215 rcu_read_lock(); 216 if (pid_task(filp->f_owner.pid, filp->f_owner.pid_type)) 217 owner.pid = pid_vnr(filp->f_owner.pid); 218 rcu_read_unlock(); 219 switch (filp->f_owner.pid_type) { 220 case PIDTYPE_PID: 221 owner.type = F_OWNER_TID; 222 break; 223 224 case PIDTYPE_TGID: 225 owner.type = F_OWNER_PID; 226 break; 227 228 case PIDTYPE_PGID: 229 owner.type = F_OWNER_PGRP; 230 break; 231 232 default: 233 WARN_ON(1); 234 ret = -EINVAL; 235 break; 236 } 237 read_unlock_irq(&filp->f_owner.lock); 238 239 if (!ret) { 240 ret = copy_to_user(owner_p, &owner, sizeof(owner)); 241 if (ret) 242 ret = -EFAULT; 243 } 244 return ret; 245} 246 247#ifdef CONFIG_CHECKPOINT_RESTORE 248static int f_getowner_uids(struct file *filp, unsigned long arg) 249{ 250 struct user_namespace *user_ns = current_user_ns(); 251 uid_t __user *dst = (void __user *)arg; 252 uid_t src[2]; 253 int err; 254 255 read_lock_irq(&filp->f_owner.lock); 256 src[0] = from_kuid(user_ns, filp->f_owner.uid); 257 src[1] = from_kuid(user_ns, filp->f_owner.euid); 258 read_unlock_irq(&filp->f_owner.lock); 259 260 err = put_user(src[0], &dst[0]); 261 err |= put_user(src[1], &dst[1]); 262 263 return err; 264} 265#else 266static int f_getowner_uids(struct file *filp, unsigned long arg) 267{ 268 return -EINVAL; 269} 270#endif 271 272static bool rw_hint_valid(u64 hint) 273{ 274 BUILD_BUG_ON(WRITE_LIFE_NOT_SET != RWH_WRITE_LIFE_NOT_SET); 275 BUILD_BUG_ON(WRITE_LIFE_NONE != RWH_WRITE_LIFE_NONE); 276 BUILD_BUG_ON(WRITE_LIFE_SHORT != RWH_WRITE_LIFE_SHORT); 277 BUILD_BUG_ON(WRITE_LIFE_MEDIUM != RWH_WRITE_LIFE_MEDIUM); 278 BUILD_BUG_ON(WRITE_LIFE_LONG != RWH_WRITE_LIFE_LONG); 279 BUILD_BUG_ON(WRITE_LIFE_EXTREME != RWH_WRITE_LIFE_EXTREME); 280 281 switch (hint) { 282 case RWH_WRITE_LIFE_NOT_SET: 283 case RWH_WRITE_LIFE_NONE: 284 case RWH_WRITE_LIFE_SHORT: 285 case RWH_WRITE_LIFE_MEDIUM: 286 case RWH_WRITE_LIFE_LONG: 287 case RWH_WRITE_LIFE_EXTREME: 288 return true; 289 default: 290 return false; 291 } 292} 293 294static long fcntl_get_rw_hint(struct file *file, unsigned int cmd, 295 unsigned long arg) 296{ 297 struct inode *inode = file_inode(file); 298 u64 __user *argp = (u64 __user *)arg; 299 u64 hint = READ_ONCE(inode->i_write_hint); 300 301 if (copy_to_user(argp, &hint, sizeof(*argp))) 302 return -EFAULT; 303 return 0; 304} 305 306static long fcntl_set_rw_hint(struct file *file, unsigned int cmd, 307 unsigned long arg) 308{ 309 struct inode *inode = file_inode(file); 310 u64 __user *argp = (u64 __user *)arg; 311 u64 hint; 312 313 if (copy_from_user(&hint, argp, sizeof(hint))) 314 return -EFAULT; 315 if (!rw_hint_valid(hint)) 316 return -EINVAL; 317 318 WRITE_ONCE(inode->i_write_hint, hint); 319 320 /* 321 * file->f_mapping->host may differ from inode. As an example, 322 * blkdev_open() modifies file->f_mapping. 323 */ 324 if (file->f_mapping->host != inode) 325 WRITE_ONCE(file->f_mapping->host->i_write_hint, hint); 326 327 return 0; 328} 329 330/* Is the file descriptor a dup of the file? */ 331static long f_dupfd_query(int fd, struct file *filp) 332{ 333 CLASS(fd_raw, f)(fd); 334 335 /* 336 * We can do the 'fdput()' immediately, as the only thing that 337 * matters is the pointer value which isn't changed by the fdput. 338 * 339 * Technically we didn't need a ref at all, and 'fdget()' was 340 * overkill, but given our lockless file pointer lookup, the 341 * alternatives are complicated. 342 */ 343 return f.file == filp; 344} 345 346static long do_fcntl(int fd, unsigned int cmd, unsigned long arg, 347 struct file *filp) 348{ 349 void __user *argp = (void __user *)arg; 350 int argi = (int)arg; 351 struct flock flock; 352 long err = -EINVAL; 353 354 switch (cmd) { 355 case F_DUPFD: 356 err = f_dupfd(argi, filp, 0); 357 break; 358 case F_DUPFD_CLOEXEC: 359 err = f_dupfd(argi, filp, O_CLOEXEC); 360 break; 361 case F_DUPFD_QUERY: 362 err = f_dupfd_query(argi, filp); 363 break; 364 case F_GETFD: 365 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0; 366 break; 367 case F_SETFD: 368 err = 0; 369 set_close_on_exec(fd, argi & FD_CLOEXEC); 370 break; 371 case F_GETFL: 372 err = filp->f_flags; 373 break; 374 case F_SETFL: 375 err = setfl(fd, filp, argi); 376 break; 377#if BITS_PER_LONG != 32 378 /* 32-bit arches must use fcntl64() */ 379 case F_OFD_GETLK: 380#endif 381 case F_GETLK: 382 if (copy_from_user(&flock, argp, sizeof(flock))) 383 return -EFAULT; 384 err = fcntl_getlk(filp, cmd, &flock); 385 if (!err && copy_to_user(argp, &flock, sizeof(flock))) 386 return -EFAULT; 387 break; 388#if BITS_PER_LONG != 32 389 /* 32-bit arches must use fcntl64() */ 390 case F_OFD_SETLK: 391 case F_OFD_SETLKW: 392 fallthrough; 393#endif 394 case F_SETLK: 395 case F_SETLKW: 396 if (copy_from_user(&flock, argp, sizeof(flock))) 397 return -EFAULT; 398 err = fcntl_setlk(fd, filp, cmd, &flock); 399 break; 400 case F_GETOWN: 401 /* 402 * XXX If f_owner is a process group, the 403 * negative return value will get converted 404 * into an error. Oops. If we keep the 405 * current syscall conventions, the only way 406 * to fix this will be in libc. 407 */ 408 err = f_getown(filp); 409 force_successful_syscall_return(); 410 break; 411 case F_SETOWN: 412 err = f_setown(filp, argi, 1); 413 break; 414 case F_GETOWN_EX: 415 err = f_getown_ex(filp, arg); 416 break; 417 case F_SETOWN_EX: 418 err = f_setown_ex(filp, arg); 419 break; 420 case F_GETOWNER_UIDS: 421 err = f_getowner_uids(filp, arg); 422 break; 423 case F_GETSIG: 424 err = filp->f_owner.signum; 425 break; 426 case F_SETSIG: 427 /* arg == 0 restores default behaviour. */ 428 if (!valid_signal(argi)) { 429 break; 430 } 431 err = 0; 432 filp->f_owner.signum = argi; 433 break; 434 case F_GETLEASE: 435 err = fcntl_getlease(filp); 436 break; 437 case F_SETLEASE: 438 err = fcntl_setlease(fd, filp, argi); 439 break; 440 case F_NOTIFY: 441 err = fcntl_dirnotify(fd, filp, argi); 442 break; 443 case F_SETPIPE_SZ: 444 case F_GETPIPE_SZ: 445 err = pipe_fcntl(filp, cmd, argi); 446 break; 447 case F_ADD_SEALS: 448 case F_GET_SEALS: 449 err = memfd_fcntl(filp, cmd, argi); 450 break; 451 case F_GET_RW_HINT: 452 err = fcntl_get_rw_hint(filp, cmd, arg); 453 break; 454 case F_SET_RW_HINT: 455 err = fcntl_set_rw_hint(filp, cmd, arg); 456 break; 457 default: 458 break; 459 } 460 return err; 461} 462 463static int check_fcntl_cmd(unsigned cmd) 464{ 465 switch (cmd) { 466 case F_DUPFD: 467 case F_DUPFD_CLOEXEC: 468 case F_DUPFD_QUERY: 469 case F_GETFD: 470 case F_SETFD: 471 case F_GETFL: 472 return 1; 473 } 474 return 0; 475} 476 477SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg) 478{ 479 struct fd f = fdget_raw(fd); 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 err = do_fcntl(fd, cmd, arg, f.file); 493 494out1: 495 fdput(f); 496out: 497 return err; 498} 499 500#if BITS_PER_LONG == 32 501SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd, 502 unsigned long, arg) 503{ 504 void __user *argp = (void __user *)arg; 505 struct fd f = fdget_raw(fd); 506 struct flock64 flock; 507 long err = -EBADF; 508 509 if (!f.file) 510 goto out; 511 512 if (unlikely(f.file->f_mode & FMODE_PATH)) { 513 if (!check_fcntl_cmd(cmd)) 514 goto out1; 515 } 516 517 err = security_file_fcntl(f.file, cmd, arg); 518 if (err) 519 goto out1; 520 521 switch (cmd) { 522 case F_GETLK64: 523 case F_OFD_GETLK: 524 err = -EFAULT; 525 if (copy_from_user(&flock, argp, sizeof(flock))) 526 break; 527 err = fcntl_getlk64(f.file, cmd, &flock); 528 if (!err && copy_to_user(argp, &flock, sizeof(flock))) 529 err = -EFAULT; 530 break; 531 case F_SETLK64: 532 case F_SETLKW64: 533 case F_OFD_SETLK: 534 case F_OFD_SETLKW: 535 err = -EFAULT; 536 if (copy_from_user(&flock, argp, sizeof(flock))) 537 break; 538 err = fcntl_setlk64(fd, f.file, cmd, &flock); 539 break; 540 default: 541 err = do_fcntl(fd, cmd, arg, f.file); 542 break; 543 } 544out1: 545 fdput(f); 546out: 547 return err; 548} 549#endif 550 551#ifdef CONFIG_COMPAT 552/* careful - don't use anywhere else */ 553#define copy_flock_fields(dst, src) \ 554 (dst)->l_type = (src)->l_type; \ 555 (dst)->l_whence = (src)->l_whence; \ 556 (dst)->l_start = (src)->l_start; \ 557 (dst)->l_len = (src)->l_len; \ 558 (dst)->l_pid = (src)->l_pid; 559 560static int get_compat_flock(struct flock *kfl, const struct compat_flock __user *ufl) 561{ 562 struct compat_flock fl; 563 564 if (copy_from_user(&fl, ufl, sizeof(struct compat_flock))) 565 return -EFAULT; 566 copy_flock_fields(kfl, &fl); 567 return 0; 568} 569 570static int get_compat_flock64(struct flock *kfl, const struct compat_flock64 __user *ufl) 571{ 572 struct compat_flock64 fl; 573 574 if (copy_from_user(&fl, ufl, sizeof(struct compat_flock64))) 575 return -EFAULT; 576 copy_flock_fields(kfl, &fl); 577 return 0; 578} 579 580static int put_compat_flock(const struct flock *kfl, struct compat_flock __user *ufl) 581{ 582 struct compat_flock fl; 583 584 memset(&fl, 0, sizeof(struct compat_flock)); 585 copy_flock_fields(&fl, kfl); 586 if (copy_to_user(ufl, &fl, sizeof(struct compat_flock))) 587 return -EFAULT; 588 return 0; 589} 590 591static int put_compat_flock64(const struct flock *kfl, struct compat_flock64 __user *ufl) 592{ 593 struct compat_flock64 fl; 594 595 BUILD_BUG_ON(sizeof(kfl->l_start) > sizeof(ufl->l_start)); 596 BUILD_BUG_ON(sizeof(kfl->l_len) > sizeof(ufl->l_len)); 597 598 memset(&fl, 0, sizeof(struct compat_flock64)); 599 copy_flock_fields(&fl, kfl); 600 if (copy_to_user(ufl, &fl, sizeof(struct compat_flock64))) 601 return -EFAULT; 602 return 0; 603} 604#undef copy_flock_fields 605 606static unsigned int 607convert_fcntl_cmd(unsigned int cmd) 608{ 609 switch (cmd) { 610 case F_GETLK64: 611 return F_GETLK; 612 case F_SETLK64: 613 return F_SETLK; 614 case F_SETLKW64: 615 return F_SETLKW; 616 } 617 618 return cmd; 619} 620 621/* 622 * GETLK was successful and we need to return the data, but it needs to fit in 623 * the compat structure. 624 * l_start shouldn't be too big, unless the original start + end is greater than 625 * COMPAT_OFF_T_MAX, in which case the app was asking for trouble, so we return 626 * -EOVERFLOW in that case. l_len could be too big, in which case we just 627 * truncate it, and only allow the app to see that part of the conflicting lock 628 * that might make sense to it anyway 629 */ 630static int fixup_compat_flock(struct flock *flock) 631{ 632 if (flock->l_start > COMPAT_OFF_T_MAX) 633 return -EOVERFLOW; 634 if (flock->l_len > COMPAT_OFF_T_MAX) 635 flock->l_len = COMPAT_OFF_T_MAX; 636 return 0; 637} 638 639static long do_compat_fcntl64(unsigned int fd, unsigned int cmd, 640 compat_ulong_t arg) 641{ 642 struct fd f = fdget_raw(fd); 643 struct flock flock; 644 long err = -EBADF; 645 646 if (!f.file) 647 return err; 648 649 if (unlikely(f.file->f_mode & FMODE_PATH)) { 650 if (!check_fcntl_cmd(cmd)) 651 goto out_put; 652 } 653 654 err = security_file_fcntl(f.file, cmd, arg); 655 if (err) 656 goto out_put; 657 658 switch (cmd) { 659 case F_GETLK: 660 err = get_compat_flock(&flock, compat_ptr(arg)); 661 if (err) 662 break; 663 err = fcntl_getlk(f.file, convert_fcntl_cmd(cmd), &flock); 664 if (err) 665 break; 666 err = fixup_compat_flock(&flock); 667 if (!err) 668 err = put_compat_flock(&flock, compat_ptr(arg)); 669 break; 670 case F_GETLK64: 671 case F_OFD_GETLK: 672 err = get_compat_flock64(&flock, compat_ptr(arg)); 673 if (err) 674 break; 675 err = fcntl_getlk(f.file, convert_fcntl_cmd(cmd), &flock); 676 if (!err) 677 err = put_compat_flock64(&flock, compat_ptr(arg)); 678 break; 679 case F_SETLK: 680 case F_SETLKW: 681 err = get_compat_flock(&flock, compat_ptr(arg)); 682 if (err) 683 break; 684 err = fcntl_setlk(fd, f.file, convert_fcntl_cmd(cmd), &flock); 685 break; 686 case F_SETLK64: 687 case F_SETLKW64: 688 case F_OFD_SETLK: 689 case F_OFD_SETLKW: 690 err = get_compat_flock64(&flock, compat_ptr(arg)); 691 if (err) 692 break; 693 err = fcntl_setlk(fd, f.file, convert_fcntl_cmd(cmd), &flock); 694 break; 695 default: 696 err = do_fcntl(fd, cmd, arg, f.file); 697 break; 698 } 699out_put: 700 fdput(f); 701 return err; 702} 703 704COMPAT_SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd, 705 compat_ulong_t, arg) 706{ 707 return do_compat_fcntl64(fd, cmd, arg); 708} 709 710COMPAT_SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, 711 compat_ulong_t, arg) 712{ 713 switch (cmd) { 714 case F_GETLK64: 715 case F_SETLK64: 716 case F_SETLKW64: 717 case F_OFD_GETLK: 718 case F_OFD_SETLK: 719 case F_OFD_SETLKW: 720 return -EINVAL; 721 } 722 return do_compat_fcntl64(fd, cmd, arg); 723} 724#endif 725 726/* Table to convert sigio signal codes into poll band bitmaps */ 727 728static const __poll_t band_table[NSIGPOLL] = { 729 EPOLLIN | EPOLLRDNORM, /* POLL_IN */ 730 EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND, /* POLL_OUT */ 731 EPOLLIN | EPOLLRDNORM | EPOLLMSG, /* POLL_MSG */ 732 EPOLLERR, /* POLL_ERR */ 733 EPOLLPRI | EPOLLRDBAND, /* POLL_PRI */ 734 EPOLLHUP | EPOLLERR /* POLL_HUP */ 735}; 736 737static inline int sigio_perm(struct task_struct *p, 738 struct fown_struct *fown, int sig) 739{ 740 const struct cred *cred; 741 int ret; 742 743 rcu_read_lock(); 744 cred = __task_cred(p); 745 ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) || 746 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) || 747 uid_eq(fown->uid, cred->suid) || uid_eq(fown->uid, cred->uid)) && 748 !security_file_send_sigiotask(p, fown, sig)); 749 rcu_read_unlock(); 750 return ret; 751} 752 753static void send_sigio_to_task(struct task_struct *p, 754 struct fown_struct *fown, 755 int fd, int reason, enum pid_type type) 756{ 757 /* 758 * F_SETSIG can change ->signum lockless in parallel, make 759 * sure we read it once and use the same value throughout. 760 */ 761 int signum = READ_ONCE(fown->signum); 762 763 if (!sigio_perm(p, fown, signum)) 764 return; 765 766 switch (signum) { 767 default: { 768 kernel_siginfo_t si; 769 770 /* Queue a rt signal with the appropriate fd as its 771 value. We use SI_SIGIO as the source, not 772 SI_KERNEL, since kernel signals always get 773 delivered even if we can't queue. Failure to 774 queue in this case _should_ be reported; we fall 775 back to SIGIO in that case. --sct */ 776 clear_siginfo(&si); 777 si.si_signo = signum; 778 si.si_errno = 0; 779 si.si_code = reason; 780 /* 781 * Posix definies POLL_IN and friends to be signal 782 * specific si_codes for SIG_POLL. Linux extended 783 * these si_codes to other signals in a way that is 784 * ambiguous if other signals also have signal 785 * specific si_codes. In that case use SI_SIGIO instead 786 * to remove the ambiguity. 787 */ 788 if ((signum != SIGPOLL) && sig_specific_sicodes(signum)) 789 si.si_code = SI_SIGIO; 790 791 /* Make sure we are called with one of the POLL_* 792 reasons, otherwise we could leak kernel stack into 793 userspace. */ 794 BUG_ON((reason < POLL_IN) || ((reason - POLL_IN) >= NSIGPOLL)); 795 if (reason - POLL_IN >= NSIGPOLL) 796 si.si_band = ~0L; 797 else 798 si.si_band = mangle_poll(band_table[reason - POLL_IN]); 799 si.si_fd = fd; 800 if (!do_send_sig_info(signum, &si, p, type)) 801 break; 802 } 803 fallthrough; /* fall back on the old plain SIGIO signal */ 804 case 0: 805 do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, type); 806 } 807} 808 809void send_sigio(struct fown_struct *fown, int fd, int band) 810{ 811 struct task_struct *p; 812 enum pid_type type; 813 unsigned long flags; 814 struct pid *pid; 815 816 read_lock_irqsave(&fown->lock, flags); 817 818 type = fown->pid_type; 819 pid = fown->pid; 820 if (!pid) 821 goto out_unlock_fown; 822 823 if (type <= PIDTYPE_TGID) { 824 rcu_read_lock(); 825 p = pid_task(pid, PIDTYPE_PID); 826 if (p) 827 send_sigio_to_task(p, fown, fd, band, type); 828 rcu_read_unlock(); 829 } else { 830 read_lock(&tasklist_lock); 831 do_each_pid_task(pid, type, p) { 832 send_sigio_to_task(p, fown, fd, band, type); 833 } while_each_pid_task(pid, type, p); 834 read_unlock(&tasklist_lock); 835 } 836 out_unlock_fown: 837 read_unlock_irqrestore(&fown->lock, flags); 838} 839 840static void send_sigurg_to_task(struct task_struct *p, 841 struct fown_struct *fown, enum pid_type type) 842{ 843 if (sigio_perm(p, fown, SIGURG)) 844 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, type); 845} 846 847int send_sigurg(struct fown_struct *fown) 848{ 849 struct task_struct *p; 850 enum pid_type type; 851 struct pid *pid; 852 unsigned long flags; 853 int ret = 0; 854 855 read_lock_irqsave(&fown->lock, flags); 856 857 type = fown->pid_type; 858 pid = fown->pid; 859 if (!pid) 860 goto out_unlock_fown; 861 862 ret = 1; 863 864 if (type <= PIDTYPE_TGID) { 865 rcu_read_lock(); 866 p = pid_task(pid, PIDTYPE_PID); 867 if (p) 868 send_sigurg_to_task(p, fown, type); 869 rcu_read_unlock(); 870 } else { 871 read_lock(&tasklist_lock); 872 do_each_pid_task(pid, type, p) { 873 send_sigurg_to_task(p, fown, type); 874 } while_each_pid_task(pid, type, p); 875 read_unlock(&tasklist_lock); 876 } 877 out_unlock_fown: 878 read_unlock_irqrestore(&fown->lock, flags); 879 return ret; 880} 881 882static DEFINE_SPINLOCK(fasync_lock); 883static struct kmem_cache *fasync_cache __ro_after_init; 884 885/* 886 * Remove a fasync entry. If successfully removed, return 887 * positive and clear the FASYNC flag. If no entry exists, 888 * do nothing and return 0. 889 * 890 * NOTE! It is very important that the FASYNC flag always 891 * match the state "is the filp on a fasync list". 892 * 893 */ 894int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp) 895{ 896 struct fasync_struct *fa, **fp; 897 int result = 0; 898 899 spin_lock(&filp->f_lock); 900 spin_lock(&fasync_lock); 901 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) { 902 if (fa->fa_file != filp) 903 continue; 904 905 write_lock_irq(&fa->fa_lock); 906 fa->fa_file = NULL; 907 write_unlock_irq(&fa->fa_lock); 908 909 *fp = fa->fa_next; 910 kfree_rcu(fa, fa_rcu); 911 filp->f_flags &= ~FASYNC; 912 result = 1; 913 break; 914 } 915 spin_unlock(&fasync_lock); 916 spin_unlock(&filp->f_lock); 917 return result; 918} 919 920struct fasync_struct *fasync_alloc(void) 921{ 922 return kmem_cache_alloc(fasync_cache, GFP_KERNEL); 923} 924 925/* 926 * NOTE! This can be used only for unused fasync entries: 927 * entries that actually got inserted on the fasync list 928 * need to be released by rcu - see fasync_remove_entry. 929 */ 930void fasync_free(struct fasync_struct *new) 931{ 932 kmem_cache_free(fasync_cache, new); 933} 934 935/* 936 * Insert a new entry into the fasync list. Return the pointer to the 937 * old one if we didn't use the new one. 938 * 939 * NOTE! It is very important that the FASYNC flag always 940 * match the state "is the filp on a fasync list". 941 */ 942struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new) 943{ 944 struct fasync_struct *fa, **fp; 945 946 spin_lock(&filp->f_lock); 947 spin_lock(&fasync_lock); 948 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) { 949 if (fa->fa_file != filp) 950 continue; 951 952 write_lock_irq(&fa->fa_lock); 953 fa->fa_fd = fd; 954 write_unlock_irq(&fa->fa_lock); 955 goto out; 956 } 957 958 rwlock_init(&new->fa_lock); 959 new->magic = FASYNC_MAGIC; 960 new->fa_file = filp; 961 new->fa_fd = fd; 962 new->fa_next = *fapp; 963 rcu_assign_pointer(*fapp, new); 964 filp->f_flags |= FASYNC; 965 966out: 967 spin_unlock(&fasync_lock); 968 spin_unlock(&filp->f_lock); 969 return fa; 970} 971 972/* 973 * Add a fasync entry. Return negative on error, positive if 974 * added, and zero if did nothing but change an existing one. 975 */ 976static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp) 977{ 978 struct fasync_struct *new; 979 980 new = fasync_alloc(); 981 if (!new) 982 return -ENOMEM; 983 984 /* 985 * fasync_insert_entry() returns the old (update) entry if 986 * it existed. 987 * 988 * So free the (unused) new entry and return 0 to let the 989 * caller know that we didn't add any new fasync entries. 990 */ 991 if (fasync_insert_entry(fd, filp, fapp, new)) { 992 fasync_free(new); 993 return 0; 994 } 995 996 return 1; 997} 998 999/* 1000 * fasync_helper() is used by almost all character device drivers 1001 * to set up the fasync queue, and for regular files by the file 1002 * lease code. It returns negative on error, 0 if it did no changes 1003 * and positive if it added/deleted the entry. 1004 */ 1005int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp) 1006{ 1007 if (!on) 1008 return fasync_remove_entry(filp, fapp); 1009 return fasync_add_entry(fd, filp, fapp); 1010} 1011 1012EXPORT_SYMBOL(fasync_helper); 1013 1014/* 1015 * rcu_read_lock() is held 1016 */ 1017static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band) 1018{ 1019 while (fa) { 1020 struct fown_struct *fown; 1021 unsigned long flags; 1022 1023 if (fa->magic != FASYNC_MAGIC) { 1024 printk(KERN_ERR "kill_fasync: bad magic number in " 1025 "fasync_struct!\n"); 1026 return; 1027 } 1028 read_lock_irqsave(&fa->fa_lock, flags); 1029 if (fa->fa_file) { 1030 fown = &fa->fa_file->f_owner; 1031 /* Don't send SIGURG to processes which have not set a 1032 queued signum: SIGURG has its own default signalling 1033 mechanism. */ 1034 if (!(sig == SIGURG && fown->signum == 0)) 1035 send_sigio(fown, fa->fa_fd, band); 1036 } 1037 read_unlock_irqrestore(&fa->fa_lock, flags); 1038 fa = rcu_dereference(fa->fa_next); 1039 } 1040} 1041 1042void kill_fasync(struct fasync_struct **fp, int sig, int band) 1043{ 1044 /* First a quick test without locking: usually 1045 * the list is empty. 1046 */ 1047 if (*fp) { 1048 rcu_read_lock(); 1049 kill_fasync_rcu(rcu_dereference(*fp), sig, band); 1050 rcu_read_unlock(); 1051 } 1052} 1053EXPORT_SYMBOL(kill_fasync); 1054 1055static int __init fcntl_init(void) 1056{ 1057 /* 1058 * Please add new bits here to ensure allocation uniqueness. 1059 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY 1060 * is defined as O_NONBLOCK on some platforms and not on others. 1061 */ 1062 BUILD_BUG_ON(21 - 1 /* for O_RDONLY being 0 */ != 1063 HWEIGHT32( 1064 (VALID_OPEN_FLAGS & ~(O_NONBLOCK | O_NDELAY)) | 1065 __FMODE_EXEC | __FMODE_NONOTIFY)); 1066 1067 fasync_cache = kmem_cache_create("fasync_cache", 1068 sizeof(struct fasync_struct), 0, 1069 SLAB_PANIC | SLAB_ACCOUNT, NULL); 1070 return 0; 1071} 1072 1073module_init(fcntl_init) 1074