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/shmem_fs.h> 27 #include <linux/compat.h> 28 29 #include <asm/poll.h> 30 #include <asm/siginfo.h> 31 #include <linux/uaccess.h> 32 33 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME) 34 35 static int setfl(int fd, struct file * filp, unsigned long arg) 36 { 37 struct inode * inode = file_inode(filp); 38 int error = 0; 39 40 /* 41 * O_APPEND cannot be cleared if the file is marked as append-only 42 * and the file is open for write. 43 */ 44 if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode)) 45 return -EPERM; 46 47 /* O_NOATIME can only be set by the owner or superuser */ 48 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME)) 49 if (!inode_owner_or_capable(inode)) 50 return -EPERM; 51 52 /* required for strict SunOS emulation */ 53 if (O_NONBLOCK != O_NDELAY) 54 if (arg & O_NDELAY) 55 arg |= O_NONBLOCK; 56 57 /* Pipe packetized mode is controlled by O_DIRECT flag */ 58 if (!S_ISFIFO(inode->i_mode) && (arg & O_DIRECT)) { 59 if (!filp->f_mapping || !filp->f_mapping->a_ops || 60 !filp->f_mapping->a_ops->direct_IO) 61 return -EINVAL; 62 } 63 64 if (filp->f_op->check_flags) 65 error = filp->f_op->check_flags(arg); 66 if (error) 67 return error; 68 69 /* 70 * ->fasync() is responsible for setting the FASYNC bit. 71 */ 72 if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op->fasync) { 73 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0); 74 if (error < 0) 75 goto out; 76 if (error > 0) 77 error = 0; 78 } 79 spin_lock(&filp->f_lock); 80 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK); 81 spin_unlock(&filp->f_lock); 82 83 out: 84 return error; 85 } 86 87 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type, 88 int force) 89 { 90 write_lock_irq(&filp->f_owner.lock); 91 if (force || !filp->f_owner.pid) { 92 put_pid(filp->f_owner.pid); 93 filp->f_owner.pid = get_pid(pid); 94 filp->f_owner.pid_type = type; 95 96 if (pid) { 97 const struct cred *cred = current_cred(); 98 filp->f_owner.uid = cred->uid; 99 filp->f_owner.euid = cred->euid; 100 } 101 } 102 write_unlock_irq(&filp->f_owner.lock); 103 } 104 105 void __f_setown(struct file *filp, struct pid *pid, enum pid_type type, 106 int force) 107 { 108 security_file_set_fowner(filp); 109 f_modown(filp, pid, type, force); 110 } 111 EXPORT_SYMBOL(__f_setown); 112 113 int f_setown(struct file *filp, unsigned long arg, int force) 114 { 115 enum pid_type type; 116 struct pid *pid = NULL; 117 int who = arg, ret = 0; 118 119 type = PIDTYPE_PID; 120 if (who < 0) { 121 /* avoid overflow below */ 122 if (who == INT_MIN) 123 return -EINVAL; 124 125 type = PIDTYPE_PGID; 126 who = -who; 127 } 128 129 rcu_read_lock(); 130 if (who) { 131 pid = find_vpid(who); 132 if (!pid) 133 ret = -ESRCH; 134 } 135 136 if (!ret) 137 __f_setown(filp, pid, type, force); 138 rcu_read_unlock(); 139 140 return ret; 141 } 142 EXPORT_SYMBOL(f_setown); 143 144 void f_delown(struct file *filp) 145 { 146 f_modown(filp, NULL, PIDTYPE_PID, 1); 147 } 148 149 pid_t f_getown(struct file *filp) 150 { 151 pid_t pid; 152 read_lock(&filp->f_owner.lock); 153 pid = pid_vnr(filp->f_owner.pid); 154 if (filp->f_owner.pid_type == PIDTYPE_PGID) 155 pid = -pid; 156 read_unlock(&filp->f_owner.lock); 157 return pid; 158 } 159 160 static int f_setown_ex(struct file *filp, unsigned long arg) 161 { 162 struct f_owner_ex __user *owner_p = (void __user *)arg; 163 struct f_owner_ex owner; 164 struct pid *pid; 165 int type; 166 int ret; 167 168 ret = copy_from_user(&owner, owner_p, sizeof(owner)); 169 if (ret) 170 return -EFAULT; 171 172 switch (owner.type) { 173 case F_OWNER_TID: 174 type = PIDTYPE_MAX; 175 break; 176 177 case F_OWNER_PID: 178 type = PIDTYPE_PID; 179 break; 180 181 case F_OWNER_PGRP: 182 type = PIDTYPE_PGID; 183 break; 184 185 default: 186 return -EINVAL; 187 } 188 189 rcu_read_lock(); 190 pid = find_vpid(owner.pid); 191 if (owner.pid && !pid) 192 ret = -ESRCH; 193 else 194 __f_setown(filp, pid, type, 1); 195 rcu_read_unlock(); 196 197 return ret; 198 } 199 200 static int f_getown_ex(struct file *filp, unsigned long arg) 201 { 202 struct f_owner_ex __user *owner_p = (void __user *)arg; 203 struct f_owner_ex owner; 204 int ret = 0; 205 206 read_lock(&filp->f_owner.lock); 207 owner.pid = pid_vnr(filp->f_owner.pid); 208 switch (filp->f_owner.pid_type) { 209 case PIDTYPE_MAX: 210 owner.type = F_OWNER_TID; 211 break; 212 213 case PIDTYPE_PID: 214 owner.type = F_OWNER_PID; 215 break; 216 217 case PIDTYPE_PGID: 218 owner.type = F_OWNER_PGRP; 219 break; 220 221 default: 222 WARN_ON(1); 223 ret = -EINVAL; 224 break; 225 } 226 read_unlock(&filp->f_owner.lock); 227 228 if (!ret) { 229 ret = copy_to_user(owner_p, &owner, sizeof(owner)); 230 if (ret) 231 ret = -EFAULT; 232 } 233 return ret; 234 } 235 236 #ifdef CONFIG_CHECKPOINT_RESTORE 237 static int f_getowner_uids(struct file *filp, unsigned long arg) 238 { 239 struct user_namespace *user_ns = current_user_ns(); 240 uid_t __user *dst = (void __user *)arg; 241 uid_t src[2]; 242 int err; 243 244 read_lock(&filp->f_owner.lock); 245 src[0] = from_kuid(user_ns, filp->f_owner.uid); 246 src[1] = from_kuid(user_ns, filp->f_owner.euid); 247 read_unlock(&filp->f_owner.lock); 248 249 err = put_user(src[0], &dst[0]); 250 err |= put_user(src[1], &dst[1]); 251 252 return err; 253 } 254 #else 255 static int f_getowner_uids(struct file *filp, unsigned long arg) 256 { 257 return -EINVAL; 258 } 259 #endif 260 261 static bool rw_hint_valid(enum rw_hint hint) 262 { 263 switch (hint) { 264 case RWF_WRITE_LIFE_NOT_SET: 265 case RWH_WRITE_LIFE_NONE: 266 case RWH_WRITE_LIFE_SHORT: 267 case RWH_WRITE_LIFE_MEDIUM: 268 case RWH_WRITE_LIFE_LONG: 269 case RWH_WRITE_LIFE_EXTREME: 270 return true; 271 default: 272 return false; 273 } 274 } 275 276 static long fcntl_rw_hint(struct file *file, unsigned int cmd, 277 unsigned long arg) 278 { 279 struct inode *inode = file_inode(file); 280 u64 *argp = (u64 __user *)arg; 281 enum rw_hint hint; 282 u64 h; 283 284 switch (cmd) { 285 case F_GET_FILE_RW_HINT: 286 h = file_write_hint(file); 287 if (copy_to_user(argp, &h, sizeof(*argp))) 288 return -EFAULT; 289 return 0; 290 case F_SET_FILE_RW_HINT: 291 if (copy_from_user(&h, argp, sizeof(h))) 292 return -EFAULT; 293 hint = (enum rw_hint) h; 294 if (!rw_hint_valid(hint)) 295 return -EINVAL; 296 297 spin_lock(&file->f_lock); 298 file->f_write_hint = hint; 299 spin_unlock(&file->f_lock); 300 return 0; 301 case F_GET_RW_HINT: 302 h = inode->i_write_hint; 303 if (copy_to_user(argp, &h, sizeof(*argp))) 304 return -EFAULT; 305 return 0; 306 case F_SET_RW_HINT: 307 if (copy_from_user(&h, argp, sizeof(h))) 308 return -EFAULT; 309 hint = (enum rw_hint) h; 310 if (!rw_hint_valid(hint)) 311 return -EINVAL; 312 313 inode_lock(inode); 314 inode->i_write_hint = hint; 315 inode_unlock(inode); 316 return 0; 317 default: 318 return -EINVAL; 319 } 320 } 321 322 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg, 323 struct file *filp) 324 { 325 void __user *argp = (void __user *)arg; 326 struct flock flock; 327 long err = -EINVAL; 328 329 switch (cmd) { 330 case F_DUPFD: 331 err = f_dupfd(arg, filp, 0); 332 break; 333 case F_DUPFD_CLOEXEC: 334 err = f_dupfd(arg, filp, O_CLOEXEC); 335 break; 336 case F_GETFD: 337 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0; 338 break; 339 case F_SETFD: 340 err = 0; 341 set_close_on_exec(fd, arg & FD_CLOEXEC); 342 break; 343 case F_GETFL: 344 err = filp->f_flags; 345 break; 346 case F_SETFL: 347 err = setfl(fd, filp, arg); 348 break; 349 #if BITS_PER_LONG != 32 350 /* 32-bit arches must use fcntl64() */ 351 case F_OFD_GETLK: 352 #endif 353 case F_GETLK: 354 if (copy_from_user(&flock, argp, sizeof(flock))) 355 return -EFAULT; 356 err = fcntl_getlk(filp, cmd, &flock); 357 if (!err && copy_to_user(argp, &flock, sizeof(flock))) 358 return -EFAULT; 359 break; 360 #if BITS_PER_LONG != 32 361 /* 32-bit arches must use fcntl64() */ 362 case F_OFD_SETLK: 363 case F_OFD_SETLKW: 364 #endif 365 /* Fallthrough */ 366 case F_SETLK: 367 case F_SETLKW: 368 if (copy_from_user(&flock, argp, sizeof(flock))) 369 return -EFAULT; 370 err = fcntl_setlk(fd, filp, cmd, &flock); 371 break; 372 case F_GETOWN: 373 /* 374 * XXX If f_owner is a process group, the 375 * negative return value will get converted 376 * into an error. Oops. If we keep the 377 * current syscall conventions, the only way 378 * to fix this will be in libc. 379 */ 380 err = f_getown(filp); 381 force_successful_syscall_return(); 382 break; 383 case F_SETOWN: 384 err = f_setown(filp, arg, 1); 385 break; 386 case F_GETOWN_EX: 387 err = f_getown_ex(filp, arg); 388 break; 389 case F_SETOWN_EX: 390 err = f_setown_ex(filp, arg); 391 break; 392 case F_GETOWNER_UIDS: 393 err = f_getowner_uids(filp, arg); 394 break; 395 case F_GETSIG: 396 err = filp->f_owner.signum; 397 break; 398 case F_SETSIG: 399 /* arg == 0 restores default behaviour. */ 400 if (!valid_signal(arg)) { 401 break; 402 } 403 err = 0; 404 filp->f_owner.signum = arg; 405 break; 406 case F_GETLEASE: 407 err = fcntl_getlease(filp); 408 break; 409 case F_SETLEASE: 410 err = fcntl_setlease(fd, filp, arg); 411 break; 412 case F_NOTIFY: 413 err = fcntl_dirnotify(fd, filp, arg); 414 break; 415 case F_SETPIPE_SZ: 416 case F_GETPIPE_SZ: 417 err = pipe_fcntl(filp, cmd, arg); 418 break; 419 case F_ADD_SEALS: 420 case F_GET_SEALS: 421 err = shmem_fcntl(filp, cmd, arg); 422 break; 423 case F_GET_RW_HINT: 424 case F_SET_RW_HINT: 425 case F_GET_FILE_RW_HINT: 426 case F_SET_FILE_RW_HINT: 427 err = fcntl_rw_hint(filp, cmd, arg); 428 break; 429 default: 430 break; 431 } 432 return err; 433 } 434 435 static int check_fcntl_cmd(unsigned cmd) 436 { 437 switch (cmd) { 438 case F_DUPFD: 439 case F_DUPFD_CLOEXEC: 440 case F_GETFD: 441 case F_SETFD: 442 case F_GETFL: 443 return 1; 444 } 445 return 0; 446 } 447 448 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg) 449 { 450 struct fd f = fdget_raw(fd); 451 long err = -EBADF; 452 453 if (!f.file) 454 goto out; 455 456 if (unlikely(f.file->f_mode & FMODE_PATH)) { 457 if (!check_fcntl_cmd(cmd)) 458 goto out1; 459 } 460 461 err = security_file_fcntl(f.file, cmd, arg); 462 if (!err) 463 err = do_fcntl(fd, cmd, arg, f.file); 464 465 out1: 466 fdput(f); 467 out: 468 return err; 469 } 470 471 #if BITS_PER_LONG == 32 472 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd, 473 unsigned long, arg) 474 { 475 void __user *argp = (void __user *)arg; 476 struct fd f = fdget_raw(fd); 477 struct flock64 flock; 478 long err = -EBADF; 479 480 if (!f.file) 481 goto out; 482 483 if (unlikely(f.file->f_mode & FMODE_PATH)) { 484 if (!check_fcntl_cmd(cmd)) 485 goto out1; 486 } 487 488 err = security_file_fcntl(f.file, cmd, arg); 489 if (err) 490 goto out1; 491 492 switch (cmd) { 493 case F_GETLK64: 494 case F_OFD_GETLK: 495 err = -EFAULT; 496 if (copy_from_user(&flock, argp, sizeof(flock))) 497 break; 498 err = fcntl_getlk64(f.file, cmd, &flock); 499 if (!err && copy_to_user(argp, &flock, sizeof(flock))) 500 err = -EFAULT; 501 break; 502 case F_SETLK64: 503 case F_SETLKW64: 504 case F_OFD_SETLK: 505 case F_OFD_SETLKW: 506 err = -EFAULT; 507 if (copy_from_user(&flock, argp, sizeof(flock))) 508 break; 509 err = fcntl_setlk64(fd, f.file, cmd, &flock); 510 break; 511 default: 512 err = do_fcntl(fd, cmd, arg, f.file); 513 break; 514 } 515 out1: 516 fdput(f); 517 out: 518 return err; 519 } 520 #endif 521 522 #ifdef CONFIG_COMPAT 523 /* careful - don't use anywhere else */ 524 #define copy_flock_fields(dst, src) \ 525 (dst)->l_type = (src)->l_type; \ 526 (dst)->l_whence = (src)->l_whence; \ 527 (dst)->l_start = (src)->l_start; \ 528 (dst)->l_len = (src)->l_len; \ 529 (dst)->l_pid = (src)->l_pid; 530 531 static int get_compat_flock(struct flock *kfl, const struct compat_flock __user *ufl) 532 { 533 struct compat_flock fl; 534 535 if (copy_from_user(&fl, ufl, sizeof(struct compat_flock))) 536 return -EFAULT; 537 copy_flock_fields(kfl, &fl); 538 return 0; 539 } 540 541 static int get_compat_flock64(struct flock *kfl, const struct compat_flock64 __user *ufl) 542 { 543 struct compat_flock64 fl; 544 545 if (copy_from_user(&fl, ufl, sizeof(struct compat_flock64))) 546 return -EFAULT; 547 copy_flock_fields(kfl, &fl); 548 return 0; 549 } 550 551 static int put_compat_flock(const struct flock *kfl, struct compat_flock __user *ufl) 552 { 553 struct compat_flock fl; 554 555 memset(&fl, 0, sizeof(struct compat_flock)); 556 copy_flock_fields(&fl, kfl); 557 if (copy_to_user(ufl, &fl, sizeof(struct compat_flock))) 558 return -EFAULT; 559 return 0; 560 } 561 562 static int put_compat_flock64(const struct flock *kfl, struct compat_flock64 __user *ufl) 563 { 564 struct compat_flock64 fl; 565 566 memset(&fl, 0, sizeof(struct compat_flock64)); 567 copy_flock_fields(&fl, kfl); 568 if (copy_to_user(ufl, &fl, sizeof(struct compat_flock64))) 569 return -EFAULT; 570 return 0; 571 } 572 #undef copy_flock_fields 573 574 static unsigned int 575 convert_fcntl_cmd(unsigned int cmd) 576 { 577 switch (cmd) { 578 case F_GETLK64: 579 return F_GETLK; 580 case F_SETLK64: 581 return F_SETLK; 582 case F_SETLKW64: 583 return F_SETLKW; 584 } 585 586 return cmd; 587 } 588 589 /* 590 * GETLK was successful and we need to return the data, but it needs to fit in 591 * the compat structure. 592 * l_start shouldn't be too big, unless the original start + end is greater than 593 * COMPAT_OFF_T_MAX, in which case the app was asking for trouble, so we return 594 * -EOVERFLOW in that case. l_len could be too big, in which case we just 595 * truncate it, and only allow the app to see that part of the conflicting lock 596 * that might make sense to it anyway 597 */ 598 static int fixup_compat_flock(struct flock *flock) 599 { 600 if (flock->l_start > COMPAT_OFF_T_MAX) 601 return -EOVERFLOW; 602 if (flock->l_len > COMPAT_OFF_T_MAX) 603 flock->l_len = COMPAT_OFF_T_MAX; 604 return 0; 605 } 606 607 COMPAT_SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd, 608 compat_ulong_t, arg) 609 { 610 struct fd f = fdget_raw(fd); 611 struct flock flock; 612 long err = -EBADF; 613 614 if (!f.file) 615 return err; 616 617 if (unlikely(f.file->f_mode & FMODE_PATH)) { 618 if (!check_fcntl_cmd(cmd)) 619 goto out_put; 620 } 621 622 err = security_file_fcntl(f.file, cmd, arg); 623 if (err) 624 goto out_put; 625 626 switch (cmd) { 627 case F_GETLK: 628 err = get_compat_flock(&flock, compat_ptr(arg)); 629 if (err) 630 break; 631 err = fcntl_getlk(f.file, convert_fcntl_cmd(cmd), &flock); 632 if (err) 633 break; 634 err = fixup_compat_flock(&flock); 635 if (err) 636 return err; 637 err = put_compat_flock(&flock, compat_ptr(arg)); 638 break; 639 case F_GETLK64: 640 case F_OFD_GETLK: 641 err = get_compat_flock64(&flock, compat_ptr(arg)); 642 if (err) 643 break; 644 err = fcntl_getlk(f.file, convert_fcntl_cmd(cmd), &flock); 645 if (err) 646 break; 647 err = fixup_compat_flock(&flock); 648 if (err) 649 return err; 650 err = put_compat_flock64(&flock, compat_ptr(arg)); 651 break; 652 case F_SETLK: 653 case F_SETLKW: 654 err = get_compat_flock(&flock, compat_ptr(arg)); 655 if (err) 656 break; 657 err = fcntl_setlk(fd, f.file, convert_fcntl_cmd(cmd), &flock); 658 break; 659 case F_SETLK64: 660 case F_SETLKW64: 661 case F_OFD_SETLK: 662 case F_OFD_SETLKW: 663 err = get_compat_flock64(&flock, compat_ptr(arg)); 664 if (err) 665 break; 666 err = fcntl_setlk(fd, f.file, convert_fcntl_cmd(cmd), &flock); 667 break; 668 default: 669 err = do_fcntl(fd, cmd, arg, f.file); 670 break; 671 } 672 out_put: 673 fdput(f); 674 return err; 675 } 676 677 COMPAT_SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, 678 compat_ulong_t, arg) 679 { 680 switch (cmd) { 681 case F_GETLK64: 682 case F_SETLK64: 683 case F_SETLKW64: 684 case F_OFD_GETLK: 685 case F_OFD_SETLK: 686 case F_OFD_SETLKW: 687 return -EINVAL; 688 } 689 return compat_sys_fcntl64(fd, cmd, arg); 690 } 691 #endif 692 693 /* Table to convert sigio signal codes into poll band bitmaps */ 694 695 static const long band_table[NSIGPOLL] = { 696 POLLIN | POLLRDNORM, /* POLL_IN */ 697 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */ 698 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */ 699 POLLERR, /* POLL_ERR */ 700 POLLPRI | POLLRDBAND, /* POLL_PRI */ 701 POLLHUP | POLLERR /* POLL_HUP */ 702 }; 703 704 static inline int sigio_perm(struct task_struct *p, 705 struct fown_struct *fown, int sig) 706 { 707 const struct cred *cred; 708 int ret; 709 710 rcu_read_lock(); 711 cred = __task_cred(p); 712 ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) || 713 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) || 714 uid_eq(fown->uid, cred->suid) || uid_eq(fown->uid, cred->uid)) && 715 !security_file_send_sigiotask(p, fown, sig)); 716 rcu_read_unlock(); 717 return ret; 718 } 719 720 static void send_sigio_to_task(struct task_struct *p, 721 struct fown_struct *fown, 722 int fd, int reason, int group) 723 { 724 /* 725 * F_SETSIG can change ->signum lockless in parallel, make 726 * sure we read it once and use the same value throughout. 727 */ 728 int signum = ACCESS_ONCE(fown->signum); 729 730 if (!sigio_perm(p, fown, signum)) 731 return; 732 733 switch (signum) { 734 siginfo_t si; 735 default: 736 /* Queue a rt signal with the appropriate fd as its 737 value. We use SI_SIGIO as the source, not 738 SI_KERNEL, since kernel signals always get 739 delivered even if we can't queue. Failure to 740 queue in this case _should_ be reported; we fall 741 back to SIGIO in that case. --sct */ 742 si.si_signo = signum; 743 si.si_errno = 0; 744 si.si_code = reason; 745 /* 746 * Posix definies POLL_IN and friends to be signal 747 * specific si_codes for SIG_POLL. Linux extended 748 * these si_codes to other signals in a way that is 749 * ambiguous if other signals also have signal 750 * specific si_codes. In that case use SI_SIGIO instead 751 * to remove the ambiguity. 752 */ 753 if ((signum != SIGPOLL) && sig_specific_sicodes(signum)) 754 si.si_code = SI_SIGIO; 755 756 /* Make sure we are called with one of the POLL_* 757 reasons, otherwise we could leak kernel stack into 758 userspace. */ 759 BUG_ON((reason < POLL_IN) || ((reason - POLL_IN) >= NSIGPOLL)); 760 if (reason - POLL_IN >= NSIGPOLL) 761 si.si_band = ~0L; 762 else 763 si.si_band = band_table[reason - POLL_IN]; 764 si.si_fd = fd; 765 if (!do_send_sig_info(signum, &si, p, group)) 766 break; 767 /* fall-through: fall back on the old plain SIGIO signal */ 768 case 0: 769 do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group); 770 } 771 } 772 773 void send_sigio(struct fown_struct *fown, int fd, int band) 774 { 775 struct task_struct *p; 776 enum pid_type type; 777 struct pid *pid; 778 int group = 1; 779 780 read_lock(&fown->lock); 781 782 type = fown->pid_type; 783 if (type == PIDTYPE_MAX) { 784 group = 0; 785 type = PIDTYPE_PID; 786 } 787 788 pid = fown->pid; 789 if (!pid) 790 goto out_unlock_fown; 791 792 read_lock(&tasklist_lock); 793 do_each_pid_task(pid, type, p) { 794 send_sigio_to_task(p, fown, fd, band, group); 795 } while_each_pid_task(pid, type, p); 796 read_unlock(&tasklist_lock); 797 out_unlock_fown: 798 read_unlock(&fown->lock); 799 } 800 801 static void send_sigurg_to_task(struct task_struct *p, 802 struct fown_struct *fown, int group) 803 { 804 if (sigio_perm(p, fown, SIGURG)) 805 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group); 806 } 807 808 int send_sigurg(struct fown_struct *fown) 809 { 810 struct task_struct *p; 811 enum pid_type type; 812 struct pid *pid; 813 int group = 1; 814 int ret = 0; 815 816 read_lock(&fown->lock); 817 818 type = fown->pid_type; 819 if (type == PIDTYPE_MAX) { 820 group = 0; 821 type = PIDTYPE_PID; 822 } 823 824 pid = fown->pid; 825 if (!pid) 826 goto out_unlock_fown; 827 828 ret = 1; 829 830 read_lock(&tasklist_lock); 831 do_each_pid_task(pid, type, p) { 832 send_sigurg_to_task(p, fown, group); 833 } while_each_pid_task(pid, type, p); 834 read_unlock(&tasklist_lock); 835 out_unlock_fown: 836 read_unlock(&fown->lock); 837 return ret; 838 } 839 840 static DEFINE_SPINLOCK(fasync_lock); 841 static struct kmem_cache *fasync_cache __read_mostly; 842 843 static void fasync_free_rcu(struct rcu_head *head) 844 { 845 kmem_cache_free(fasync_cache, 846 container_of(head, struct fasync_struct, fa_rcu)); 847 } 848 849 /* 850 * Remove a fasync entry. If successfully removed, return 851 * positive and clear the FASYNC flag. If no entry exists, 852 * do nothing and return 0. 853 * 854 * NOTE! It is very important that the FASYNC flag always 855 * match the state "is the filp on a fasync list". 856 * 857 */ 858 int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp) 859 { 860 struct fasync_struct *fa, **fp; 861 int result = 0; 862 863 spin_lock(&filp->f_lock); 864 spin_lock(&fasync_lock); 865 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) { 866 if (fa->fa_file != filp) 867 continue; 868 869 spin_lock_irq(&fa->fa_lock); 870 fa->fa_file = NULL; 871 spin_unlock_irq(&fa->fa_lock); 872 873 *fp = fa->fa_next; 874 call_rcu(&fa->fa_rcu, fasync_free_rcu); 875 filp->f_flags &= ~FASYNC; 876 result = 1; 877 break; 878 } 879 spin_unlock(&fasync_lock); 880 spin_unlock(&filp->f_lock); 881 return result; 882 } 883 884 struct fasync_struct *fasync_alloc(void) 885 { 886 return kmem_cache_alloc(fasync_cache, GFP_KERNEL); 887 } 888 889 /* 890 * NOTE! This can be used only for unused fasync entries: 891 * entries that actually got inserted on the fasync list 892 * need to be released by rcu - see fasync_remove_entry. 893 */ 894 void fasync_free(struct fasync_struct *new) 895 { 896 kmem_cache_free(fasync_cache, new); 897 } 898 899 /* 900 * Insert a new entry into the fasync list. Return the pointer to the 901 * old one if we didn't use the new one. 902 * 903 * NOTE! It is very important that the FASYNC flag always 904 * match the state "is the filp on a fasync list". 905 */ 906 struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new) 907 { 908 struct fasync_struct *fa, **fp; 909 910 spin_lock(&filp->f_lock); 911 spin_lock(&fasync_lock); 912 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) { 913 if (fa->fa_file != filp) 914 continue; 915 916 spin_lock_irq(&fa->fa_lock); 917 fa->fa_fd = fd; 918 spin_unlock_irq(&fa->fa_lock); 919 goto out; 920 } 921 922 spin_lock_init(&new->fa_lock); 923 new->magic = FASYNC_MAGIC; 924 new->fa_file = filp; 925 new->fa_fd = fd; 926 new->fa_next = *fapp; 927 rcu_assign_pointer(*fapp, new); 928 filp->f_flags |= FASYNC; 929 930 out: 931 spin_unlock(&fasync_lock); 932 spin_unlock(&filp->f_lock); 933 return fa; 934 } 935 936 /* 937 * Add a fasync entry. Return negative on error, positive if 938 * added, and zero if did nothing but change an existing one. 939 */ 940 static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp) 941 { 942 struct fasync_struct *new; 943 944 new = fasync_alloc(); 945 if (!new) 946 return -ENOMEM; 947 948 /* 949 * fasync_insert_entry() returns the old (update) entry if 950 * it existed. 951 * 952 * So free the (unused) new entry and return 0 to let the 953 * caller know that we didn't add any new fasync entries. 954 */ 955 if (fasync_insert_entry(fd, filp, fapp, new)) { 956 fasync_free(new); 957 return 0; 958 } 959 960 return 1; 961 } 962 963 /* 964 * fasync_helper() is used by almost all character device drivers 965 * to set up the fasync queue, and for regular files by the file 966 * lease code. It returns negative on error, 0 if it did no changes 967 * and positive if it added/deleted the entry. 968 */ 969 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp) 970 { 971 if (!on) 972 return fasync_remove_entry(filp, fapp); 973 return fasync_add_entry(fd, filp, fapp); 974 } 975 976 EXPORT_SYMBOL(fasync_helper); 977 978 /* 979 * rcu_read_lock() is held 980 */ 981 static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band) 982 { 983 while (fa) { 984 struct fown_struct *fown; 985 unsigned long flags; 986 987 if (fa->magic != FASYNC_MAGIC) { 988 printk(KERN_ERR "kill_fasync: bad magic number in " 989 "fasync_struct!\n"); 990 return; 991 } 992 spin_lock_irqsave(&fa->fa_lock, flags); 993 if (fa->fa_file) { 994 fown = &fa->fa_file->f_owner; 995 /* Don't send SIGURG to processes which have not set a 996 queued signum: SIGURG has its own default signalling 997 mechanism. */ 998 if (!(sig == SIGURG && fown->signum == 0)) 999 send_sigio(fown, fa->fa_fd, band); 1000 } 1001 spin_unlock_irqrestore(&fa->fa_lock, flags); 1002 fa = rcu_dereference(fa->fa_next); 1003 } 1004 } 1005 1006 void kill_fasync(struct fasync_struct **fp, int sig, int band) 1007 { 1008 /* First a quick test without locking: usually 1009 * the list is empty. 1010 */ 1011 if (*fp) { 1012 rcu_read_lock(); 1013 kill_fasync_rcu(rcu_dereference(*fp), sig, band); 1014 rcu_read_unlock(); 1015 } 1016 } 1017 EXPORT_SYMBOL(kill_fasync); 1018 1019 static int __init fcntl_init(void) 1020 { 1021 /* 1022 * Please add new bits here to ensure allocation uniqueness. 1023 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY 1024 * is defined as O_NONBLOCK on some platforms and not on others. 1025 */ 1026 BUILD_BUG_ON(21 - 1 /* for O_RDONLY being 0 */ != 1027 HWEIGHT32( 1028 (VALID_OPEN_FLAGS & ~(O_NONBLOCK | O_NDELAY)) | 1029 __FMODE_EXEC | __FMODE_NONOTIFY)); 1030 1031 fasync_cache = kmem_cache_create("fasync_cache", 1032 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL); 1033 return 0; 1034 } 1035 1036 module_init(fcntl_init) 1037