1 /* 2 * Functions to manage eBPF programs attached to cgroups 3 * 4 * Copyright (c) 2016 Daniel Mack 5 * 6 * This file is subject to the terms and conditions of version 2 of the GNU 7 * General Public License. See the file COPYING in the main directory of the 8 * Linux distribution for more details. 9 */ 10 11 #include <linux/kernel.h> 12 #include <linux/atomic.h> 13 #include <linux/cgroup.h> 14 #include <linux/filter.h> 15 #include <linux/slab.h> 16 #include <linux/sysctl.h> 17 #include <linux/string.h> 18 #include <linux/bpf.h> 19 #include <linux/bpf-cgroup.h> 20 #include <net/sock.h> 21 22 DEFINE_STATIC_KEY_FALSE(cgroup_bpf_enabled_key); 23 EXPORT_SYMBOL(cgroup_bpf_enabled_key); 24 25 /** 26 * cgroup_bpf_put() - put references of all bpf programs 27 * @cgrp: the cgroup to modify 28 */ 29 void cgroup_bpf_put(struct cgroup *cgrp) 30 { 31 enum bpf_cgroup_storage_type stype; 32 unsigned int type; 33 34 for (type = 0; type < ARRAY_SIZE(cgrp->bpf.progs); type++) { 35 struct list_head *progs = &cgrp->bpf.progs[type]; 36 struct bpf_prog_list *pl, *tmp; 37 38 list_for_each_entry_safe(pl, tmp, progs, node) { 39 list_del(&pl->node); 40 bpf_prog_put(pl->prog); 41 for_each_cgroup_storage_type(stype) { 42 bpf_cgroup_storage_unlink(pl->storage[stype]); 43 bpf_cgroup_storage_free(pl->storage[stype]); 44 } 45 kfree(pl); 46 static_branch_dec(&cgroup_bpf_enabled_key); 47 } 48 bpf_prog_array_free(cgrp->bpf.effective[type]); 49 } 50 } 51 52 /* count number of elements in the list. 53 * it's slow but the list cannot be long 54 */ 55 static u32 prog_list_length(struct list_head *head) 56 { 57 struct bpf_prog_list *pl; 58 u32 cnt = 0; 59 60 list_for_each_entry(pl, head, node) { 61 if (!pl->prog) 62 continue; 63 cnt++; 64 } 65 return cnt; 66 } 67 68 /* if parent has non-overridable prog attached, 69 * disallow attaching new programs to the descendent cgroup. 70 * if parent has overridable or multi-prog, allow attaching 71 */ 72 static bool hierarchy_allows_attach(struct cgroup *cgrp, 73 enum bpf_attach_type type, 74 u32 new_flags) 75 { 76 struct cgroup *p; 77 78 p = cgroup_parent(cgrp); 79 if (!p) 80 return true; 81 do { 82 u32 flags = p->bpf.flags[type]; 83 u32 cnt; 84 85 if (flags & BPF_F_ALLOW_MULTI) 86 return true; 87 cnt = prog_list_length(&p->bpf.progs[type]); 88 WARN_ON_ONCE(cnt > 1); 89 if (cnt == 1) 90 return !!(flags & BPF_F_ALLOW_OVERRIDE); 91 p = cgroup_parent(p); 92 } while (p); 93 return true; 94 } 95 96 /* compute a chain of effective programs for a given cgroup: 97 * start from the list of programs in this cgroup and add 98 * all parent programs. 99 * Note that parent's F_ALLOW_OVERRIDE-type program is yielding 100 * to programs in this cgroup 101 */ 102 static int compute_effective_progs(struct cgroup *cgrp, 103 enum bpf_attach_type type, 104 struct bpf_prog_array __rcu **array) 105 { 106 enum bpf_cgroup_storage_type stype; 107 struct bpf_prog_array *progs; 108 struct bpf_prog_list *pl; 109 struct cgroup *p = cgrp; 110 int cnt = 0; 111 112 /* count number of effective programs by walking parents */ 113 do { 114 if (cnt == 0 || (p->bpf.flags[type] & BPF_F_ALLOW_MULTI)) 115 cnt += prog_list_length(&p->bpf.progs[type]); 116 p = cgroup_parent(p); 117 } while (p); 118 119 progs = bpf_prog_array_alloc(cnt, GFP_KERNEL); 120 if (!progs) 121 return -ENOMEM; 122 123 /* populate the array with effective progs */ 124 cnt = 0; 125 p = cgrp; 126 do { 127 if (cnt > 0 && !(p->bpf.flags[type] & BPF_F_ALLOW_MULTI)) 128 continue; 129 130 list_for_each_entry(pl, &p->bpf.progs[type], node) { 131 if (!pl->prog) 132 continue; 133 134 progs->items[cnt].prog = pl->prog; 135 for_each_cgroup_storage_type(stype) 136 progs->items[cnt].cgroup_storage[stype] = 137 pl->storage[stype]; 138 cnt++; 139 } 140 } while ((p = cgroup_parent(p))); 141 142 rcu_assign_pointer(*array, progs); 143 return 0; 144 } 145 146 static void activate_effective_progs(struct cgroup *cgrp, 147 enum bpf_attach_type type, 148 struct bpf_prog_array __rcu *array) 149 { 150 struct bpf_prog_array __rcu *old_array; 151 152 old_array = xchg(&cgrp->bpf.effective[type], array); 153 /* free prog array after grace period, since __cgroup_bpf_run_*() 154 * might be still walking the array 155 */ 156 bpf_prog_array_free(old_array); 157 } 158 159 /** 160 * cgroup_bpf_inherit() - inherit effective programs from parent 161 * @cgrp: the cgroup to modify 162 */ 163 int cgroup_bpf_inherit(struct cgroup *cgrp) 164 { 165 /* has to use marco instead of const int, since compiler thinks 166 * that array below is variable length 167 */ 168 #define NR ARRAY_SIZE(cgrp->bpf.effective) 169 struct bpf_prog_array __rcu *arrays[NR] = {}; 170 int i; 171 172 for (i = 0; i < NR; i++) 173 INIT_LIST_HEAD(&cgrp->bpf.progs[i]); 174 175 for (i = 0; i < NR; i++) 176 if (compute_effective_progs(cgrp, i, &arrays[i])) 177 goto cleanup; 178 179 for (i = 0; i < NR; i++) 180 activate_effective_progs(cgrp, i, arrays[i]); 181 182 return 0; 183 cleanup: 184 for (i = 0; i < NR; i++) 185 bpf_prog_array_free(arrays[i]); 186 return -ENOMEM; 187 } 188 189 static int update_effective_progs(struct cgroup *cgrp, 190 enum bpf_attach_type type) 191 { 192 struct cgroup_subsys_state *css; 193 int err; 194 195 /* allocate and recompute effective prog arrays */ 196 css_for_each_descendant_pre(css, &cgrp->self) { 197 struct cgroup *desc = container_of(css, struct cgroup, self); 198 199 err = compute_effective_progs(desc, type, &desc->bpf.inactive); 200 if (err) 201 goto cleanup; 202 } 203 204 /* all allocations were successful. Activate all prog arrays */ 205 css_for_each_descendant_pre(css, &cgrp->self) { 206 struct cgroup *desc = container_of(css, struct cgroup, self); 207 208 activate_effective_progs(desc, type, desc->bpf.inactive); 209 desc->bpf.inactive = NULL; 210 } 211 212 return 0; 213 214 cleanup: 215 /* oom while computing effective. Free all computed effective arrays 216 * since they were not activated 217 */ 218 css_for_each_descendant_pre(css, &cgrp->self) { 219 struct cgroup *desc = container_of(css, struct cgroup, self); 220 221 bpf_prog_array_free(desc->bpf.inactive); 222 desc->bpf.inactive = NULL; 223 } 224 225 return err; 226 } 227 228 #define BPF_CGROUP_MAX_PROGS 64 229 230 /** 231 * __cgroup_bpf_attach() - Attach the program to a cgroup, and 232 * propagate the change to descendants 233 * @cgrp: The cgroup which descendants to traverse 234 * @prog: A program to attach 235 * @type: Type of attach operation 236 * @flags: Option flags 237 * 238 * Must be called with cgroup_mutex held. 239 */ 240 int __cgroup_bpf_attach(struct cgroup *cgrp, struct bpf_prog *prog, 241 enum bpf_attach_type type, u32 flags) 242 { 243 struct list_head *progs = &cgrp->bpf.progs[type]; 244 struct bpf_prog *old_prog = NULL; 245 struct bpf_cgroup_storage *storage[MAX_BPF_CGROUP_STORAGE_TYPE], 246 *old_storage[MAX_BPF_CGROUP_STORAGE_TYPE] = {NULL}; 247 enum bpf_cgroup_storage_type stype; 248 struct bpf_prog_list *pl; 249 bool pl_was_allocated; 250 int err; 251 252 if ((flags & BPF_F_ALLOW_OVERRIDE) && (flags & BPF_F_ALLOW_MULTI)) 253 /* invalid combination */ 254 return -EINVAL; 255 256 if (!hierarchy_allows_attach(cgrp, type, flags)) 257 return -EPERM; 258 259 if (!list_empty(progs) && cgrp->bpf.flags[type] != flags) 260 /* Disallow attaching non-overridable on top 261 * of existing overridable in this cgroup. 262 * Disallow attaching multi-prog if overridable or none 263 */ 264 return -EPERM; 265 266 if (prog_list_length(progs) >= BPF_CGROUP_MAX_PROGS) 267 return -E2BIG; 268 269 for_each_cgroup_storage_type(stype) { 270 storage[stype] = bpf_cgroup_storage_alloc(prog, stype); 271 if (IS_ERR(storage[stype])) { 272 storage[stype] = NULL; 273 for_each_cgroup_storage_type(stype) 274 bpf_cgroup_storage_free(storage[stype]); 275 return -ENOMEM; 276 } 277 } 278 279 if (flags & BPF_F_ALLOW_MULTI) { 280 list_for_each_entry(pl, progs, node) { 281 if (pl->prog == prog) { 282 /* disallow attaching the same prog twice */ 283 for_each_cgroup_storage_type(stype) 284 bpf_cgroup_storage_free(storage[stype]); 285 return -EINVAL; 286 } 287 } 288 289 pl = kmalloc(sizeof(*pl), GFP_KERNEL); 290 if (!pl) { 291 for_each_cgroup_storage_type(stype) 292 bpf_cgroup_storage_free(storage[stype]); 293 return -ENOMEM; 294 } 295 296 pl_was_allocated = true; 297 pl->prog = prog; 298 for_each_cgroup_storage_type(stype) 299 pl->storage[stype] = storage[stype]; 300 list_add_tail(&pl->node, progs); 301 } else { 302 if (list_empty(progs)) { 303 pl = kmalloc(sizeof(*pl), GFP_KERNEL); 304 if (!pl) { 305 for_each_cgroup_storage_type(stype) 306 bpf_cgroup_storage_free(storage[stype]); 307 return -ENOMEM; 308 } 309 pl_was_allocated = true; 310 list_add_tail(&pl->node, progs); 311 } else { 312 pl = list_first_entry(progs, typeof(*pl), node); 313 old_prog = pl->prog; 314 for_each_cgroup_storage_type(stype) { 315 old_storage[stype] = pl->storage[stype]; 316 bpf_cgroup_storage_unlink(old_storage[stype]); 317 } 318 pl_was_allocated = false; 319 } 320 pl->prog = prog; 321 for_each_cgroup_storage_type(stype) 322 pl->storage[stype] = storage[stype]; 323 } 324 325 cgrp->bpf.flags[type] = flags; 326 327 err = update_effective_progs(cgrp, type); 328 if (err) 329 goto cleanup; 330 331 static_branch_inc(&cgroup_bpf_enabled_key); 332 for_each_cgroup_storage_type(stype) { 333 if (!old_storage[stype]) 334 continue; 335 bpf_cgroup_storage_free(old_storage[stype]); 336 } 337 if (old_prog) { 338 bpf_prog_put(old_prog); 339 static_branch_dec(&cgroup_bpf_enabled_key); 340 } 341 for_each_cgroup_storage_type(stype) 342 bpf_cgroup_storage_link(storage[stype], cgrp, type); 343 return 0; 344 345 cleanup: 346 /* and cleanup the prog list */ 347 pl->prog = old_prog; 348 for_each_cgroup_storage_type(stype) { 349 bpf_cgroup_storage_free(pl->storage[stype]); 350 pl->storage[stype] = old_storage[stype]; 351 bpf_cgroup_storage_link(old_storage[stype], cgrp, type); 352 } 353 if (pl_was_allocated) { 354 list_del(&pl->node); 355 kfree(pl); 356 } 357 return err; 358 } 359 360 /** 361 * __cgroup_bpf_detach() - Detach the program from a cgroup, and 362 * propagate the change to descendants 363 * @cgrp: The cgroup which descendants to traverse 364 * @prog: A program to detach or NULL 365 * @type: Type of detach operation 366 * 367 * Must be called with cgroup_mutex held. 368 */ 369 int __cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog, 370 enum bpf_attach_type type) 371 { 372 struct list_head *progs = &cgrp->bpf.progs[type]; 373 enum bpf_cgroup_storage_type stype; 374 u32 flags = cgrp->bpf.flags[type]; 375 struct bpf_prog *old_prog = NULL; 376 struct bpf_prog_list *pl; 377 int err; 378 379 if (flags & BPF_F_ALLOW_MULTI) { 380 if (!prog) 381 /* to detach MULTI prog the user has to specify valid FD 382 * of the program to be detached 383 */ 384 return -EINVAL; 385 } else { 386 if (list_empty(progs)) 387 /* report error when trying to detach and nothing is attached */ 388 return -ENOENT; 389 } 390 391 if (flags & BPF_F_ALLOW_MULTI) { 392 /* find the prog and detach it */ 393 list_for_each_entry(pl, progs, node) { 394 if (pl->prog != prog) 395 continue; 396 old_prog = prog; 397 /* mark it deleted, so it's ignored while 398 * recomputing effective 399 */ 400 pl->prog = NULL; 401 break; 402 } 403 if (!old_prog) 404 return -ENOENT; 405 } else { 406 /* to maintain backward compatibility NONE and OVERRIDE cgroups 407 * allow detaching with invalid FD (prog==NULL) 408 */ 409 pl = list_first_entry(progs, typeof(*pl), node); 410 old_prog = pl->prog; 411 pl->prog = NULL; 412 } 413 414 err = update_effective_progs(cgrp, type); 415 if (err) 416 goto cleanup; 417 418 /* now can actually delete it from this cgroup list */ 419 list_del(&pl->node); 420 for_each_cgroup_storage_type(stype) { 421 bpf_cgroup_storage_unlink(pl->storage[stype]); 422 bpf_cgroup_storage_free(pl->storage[stype]); 423 } 424 kfree(pl); 425 if (list_empty(progs)) 426 /* last program was detached, reset flags to zero */ 427 cgrp->bpf.flags[type] = 0; 428 429 bpf_prog_put(old_prog); 430 static_branch_dec(&cgroup_bpf_enabled_key); 431 return 0; 432 433 cleanup: 434 /* and restore back old_prog */ 435 pl->prog = old_prog; 436 return err; 437 } 438 439 /* Must be called with cgroup_mutex held to avoid races. */ 440 int __cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr, 441 union bpf_attr __user *uattr) 442 { 443 __u32 __user *prog_ids = u64_to_user_ptr(attr->query.prog_ids); 444 enum bpf_attach_type type = attr->query.attach_type; 445 struct list_head *progs = &cgrp->bpf.progs[type]; 446 u32 flags = cgrp->bpf.flags[type]; 447 int cnt, ret = 0, i; 448 449 if (attr->query.query_flags & BPF_F_QUERY_EFFECTIVE) 450 cnt = bpf_prog_array_length(cgrp->bpf.effective[type]); 451 else 452 cnt = prog_list_length(progs); 453 454 if (copy_to_user(&uattr->query.attach_flags, &flags, sizeof(flags))) 455 return -EFAULT; 456 if (copy_to_user(&uattr->query.prog_cnt, &cnt, sizeof(cnt))) 457 return -EFAULT; 458 if (attr->query.prog_cnt == 0 || !prog_ids || !cnt) 459 /* return early if user requested only program count + flags */ 460 return 0; 461 if (attr->query.prog_cnt < cnt) { 462 cnt = attr->query.prog_cnt; 463 ret = -ENOSPC; 464 } 465 466 if (attr->query.query_flags & BPF_F_QUERY_EFFECTIVE) { 467 return bpf_prog_array_copy_to_user(cgrp->bpf.effective[type], 468 prog_ids, cnt); 469 } else { 470 struct bpf_prog_list *pl; 471 u32 id; 472 473 i = 0; 474 list_for_each_entry(pl, progs, node) { 475 id = pl->prog->aux->id; 476 if (copy_to_user(prog_ids + i, &id, sizeof(id))) 477 return -EFAULT; 478 if (++i == cnt) 479 break; 480 } 481 } 482 return ret; 483 } 484 485 int cgroup_bpf_prog_attach(const union bpf_attr *attr, 486 enum bpf_prog_type ptype, struct bpf_prog *prog) 487 { 488 struct cgroup *cgrp; 489 int ret; 490 491 cgrp = cgroup_get_from_fd(attr->target_fd); 492 if (IS_ERR(cgrp)) 493 return PTR_ERR(cgrp); 494 495 ret = cgroup_bpf_attach(cgrp, prog, attr->attach_type, 496 attr->attach_flags); 497 cgroup_put(cgrp); 498 return ret; 499 } 500 501 int cgroup_bpf_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype) 502 { 503 struct bpf_prog *prog; 504 struct cgroup *cgrp; 505 int ret; 506 507 cgrp = cgroup_get_from_fd(attr->target_fd); 508 if (IS_ERR(cgrp)) 509 return PTR_ERR(cgrp); 510 511 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype); 512 if (IS_ERR(prog)) 513 prog = NULL; 514 515 ret = cgroup_bpf_detach(cgrp, prog, attr->attach_type, 0); 516 if (prog) 517 bpf_prog_put(prog); 518 519 cgroup_put(cgrp); 520 return ret; 521 } 522 523 int cgroup_bpf_prog_query(const union bpf_attr *attr, 524 union bpf_attr __user *uattr) 525 { 526 struct cgroup *cgrp; 527 int ret; 528 529 cgrp = cgroup_get_from_fd(attr->query.target_fd); 530 if (IS_ERR(cgrp)) 531 return PTR_ERR(cgrp); 532 533 ret = cgroup_bpf_query(cgrp, attr, uattr); 534 535 cgroup_put(cgrp); 536 return ret; 537 } 538 539 /** 540 * __cgroup_bpf_run_filter_skb() - Run a program for packet filtering 541 * @sk: The socket sending or receiving traffic 542 * @skb: The skb that is being sent or received 543 * @type: The type of program to be exectuted 544 * 545 * If no socket is passed, or the socket is not of type INET or INET6, 546 * this function does nothing and returns 0. 547 * 548 * The program type passed in via @type must be suitable for network 549 * filtering. No further check is performed to assert that. 550 * 551 * This function will return %-EPERM if any if an attached program was found 552 * and if it returned != 1 during execution. In all other cases, 0 is returned. 553 */ 554 int __cgroup_bpf_run_filter_skb(struct sock *sk, 555 struct sk_buff *skb, 556 enum bpf_attach_type type) 557 { 558 unsigned int offset = skb->data - skb_network_header(skb); 559 struct sock *save_sk; 560 void *saved_data_end; 561 struct cgroup *cgrp; 562 int ret; 563 564 if (!sk || !sk_fullsock(sk)) 565 return 0; 566 567 if (sk->sk_family != AF_INET && sk->sk_family != AF_INET6) 568 return 0; 569 570 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 571 save_sk = skb->sk; 572 skb->sk = sk; 573 __skb_push(skb, offset); 574 575 /* compute pointers for the bpf prog */ 576 bpf_compute_and_save_data_end(skb, &saved_data_end); 577 578 ret = BPF_PROG_RUN_ARRAY(cgrp->bpf.effective[type], skb, 579 __bpf_prog_run_save_cb); 580 bpf_restore_data_end(skb, saved_data_end); 581 __skb_pull(skb, offset); 582 skb->sk = save_sk; 583 return ret == 1 ? 0 : -EPERM; 584 } 585 EXPORT_SYMBOL(__cgroup_bpf_run_filter_skb); 586 587 /** 588 * __cgroup_bpf_run_filter_sk() - Run a program on a sock 589 * @sk: sock structure to manipulate 590 * @type: The type of program to be exectuted 591 * 592 * socket is passed is expected to be of type INET or INET6. 593 * 594 * The program type passed in via @type must be suitable for sock 595 * filtering. No further check is performed to assert that. 596 * 597 * This function will return %-EPERM if any if an attached program was found 598 * and if it returned != 1 during execution. In all other cases, 0 is returned. 599 */ 600 int __cgroup_bpf_run_filter_sk(struct sock *sk, 601 enum bpf_attach_type type) 602 { 603 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 604 int ret; 605 606 ret = BPF_PROG_RUN_ARRAY(cgrp->bpf.effective[type], sk, BPF_PROG_RUN); 607 return ret == 1 ? 0 : -EPERM; 608 } 609 EXPORT_SYMBOL(__cgroup_bpf_run_filter_sk); 610 611 /** 612 * __cgroup_bpf_run_filter_sock_addr() - Run a program on a sock and 613 * provided by user sockaddr 614 * @sk: sock struct that will use sockaddr 615 * @uaddr: sockaddr struct provided by user 616 * @type: The type of program to be exectuted 617 * @t_ctx: Pointer to attach type specific context 618 * 619 * socket is expected to be of type INET or INET6. 620 * 621 * This function will return %-EPERM if an attached program is found and 622 * returned value != 1 during execution. In all other cases, 0 is returned. 623 */ 624 int __cgroup_bpf_run_filter_sock_addr(struct sock *sk, 625 struct sockaddr *uaddr, 626 enum bpf_attach_type type, 627 void *t_ctx) 628 { 629 struct bpf_sock_addr_kern ctx = { 630 .sk = sk, 631 .uaddr = uaddr, 632 .t_ctx = t_ctx, 633 }; 634 struct sockaddr_storage unspec; 635 struct cgroup *cgrp; 636 int ret; 637 638 /* Check socket family since not all sockets represent network 639 * endpoint (e.g. AF_UNIX). 640 */ 641 if (sk->sk_family != AF_INET && sk->sk_family != AF_INET6) 642 return 0; 643 644 if (!ctx.uaddr) { 645 memset(&unspec, 0, sizeof(unspec)); 646 ctx.uaddr = (struct sockaddr *)&unspec; 647 } 648 649 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 650 ret = BPF_PROG_RUN_ARRAY(cgrp->bpf.effective[type], &ctx, BPF_PROG_RUN); 651 652 return ret == 1 ? 0 : -EPERM; 653 } 654 EXPORT_SYMBOL(__cgroup_bpf_run_filter_sock_addr); 655 656 /** 657 * __cgroup_bpf_run_filter_sock_ops() - Run a program on a sock 658 * @sk: socket to get cgroup from 659 * @sock_ops: bpf_sock_ops_kern struct to pass to program. Contains 660 * sk with connection information (IP addresses, etc.) May not contain 661 * cgroup info if it is a req sock. 662 * @type: The type of program to be exectuted 663 * 664 * socket passed is expected to be of type INET or INET6. 665 * 666 * The program type passed in via @type must be suitable for sock_ops 667 * filtering. No further check is performed to assert that. 668 * 669 * This function will return %-EPERM if any if an attached program was found 670 * and if it returned != 1 during execution. In all other cases, 0 is returned. 671 */ 672 int __cgroup_bpf_run_filter_sock_ops(struct sock *sk, 673 struct bpf_sock_ops_kern *sock_ops, 674 enum bpf_attach_type type) 675 { 676 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 677 int ret; 678 679 ret = BPF_PROG_RUN_ARRAY(cgrp->bpf.effective[type], sock_ops, 680 BPF_PROG_RUN); 681 return ret == 1 ? 0 : -EPERM; 682 } 683 EXPORT_SYMBOL(__cgroup_bpf_run_filter_sock_ops); 684 685 int __cgroup_bpf_check_dev_permission(short dev_type, u32 major, u32 minor, 686 short access, enum bpf_attach_type type) 687 { 688 struct cgroup *cgrp; 689 struct bpf_cgroup_dev_ctx ctx = { 690 .access_type = (access << 16) | dev_type, 691 .major = major, 692 .minor = minor, 693 }; 694 int allow = 1; 695 696 rcu_read_lock(); 697 cgrp = task_dfl_cgroup(current); 698 allow = BPF_PROG_RUN_ARRAY(cgrp->bpf.effective[type], &ctx, 699 BPF_PROG_RUN); 700 rcu_read_unlock(); 701 702 return !allow; 703 } 704 EXPORT_SYMBOL(__cgroup_bpf_check_dev_permission); 705 706 static const struct bpf_func_proto * 707 cgroup_base_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 708 { 709 switch (func_id) { 710 case BPF_FUNC_map_lookup_elem: 711 return &bpf_map_lookup_elem_proto; 712 case BPF_FUNC_map_update_elem: 713 return &bpf_map_update_elem_proto; 714 case BPF_FUNC_map_delete_elem: 715 return &bpf_map_delete_elem_proto; 716 case BPF_FUNC_get_current_uid_gid: 717 return &bpf_get_current_uid_gid_proto; 718 case BPF_FUNC_get_local_storage: 719 return &bpf_get_local_storage_proto; 720 case BPF_FUNC_get_current_cgroup_id: 721 return &bpf_get_current_cgroup_id_proto; 722 case BPF_FUNC_trace_printk: 723 if (capable(CAP_SYS_ADMIN)) 724 return bpf_get_trace_printk_proto(); 725 /* fall through */ 726 default: 727 return NULL; 728 } 729 } 730 731 static const struct bpf_func_proto * 732 cgroup_dev_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 733 { 734 return cgroup_base_func_proto(func_id, prog); 735 } 736 737 static bool cgroup_dev_is_valid_access(int off, int size, 738 enum bpf_access_type type, 739 const struct bpf_prog *prog, 740 struct bpf_insn_access_aux *info) 741 { 742 const int size_default = sizeof(__u32); 743 744 if (type == BPF_WRITE) 745 return false; 746 747 if (off < 0 || off + size > sizeof(struct bpf_cgroup_dev_ctx)) 748 return false; 749 /* The verifier guarantees that size > 0. */ 750 if (off % size != 0) 751 return false; 752 753 switch (off) { 754 case bpf_ctx_range(struct bpf_cgroup_dev_ctx, access_type): 755 bpf_ctx_record_field_size(info, size_default); 756 if (!bpf_ctx_narrow_access_ok(off, size, size_default)) 757 return false; 758 break; 759 default: 760 if (size != size_default) 761 return false; 762 } 763 764 return true; 765 } 766 767 const struct bpf_prog_ops cg_dev_prog_ops = { 768 }; 769 770 const struct bpf_verifier_ops cg_dev_verifier_ops = { 771 .get_func_proto = cgroup_dev_func_proto, 772 .is_valid_access = cgroup_dev_is_valid_access, 773 }; 774 775 /** 776 * __cgroup_bpf_run_filter_sysctl - Run a program on sysctl 777 * 778 * @head: sysctl table header 779 * @table: sysctl table 780 * @write: sysctl is being read (= 0) or written (= 1) 781 * @buf: pointer to buffer passed by user space 782 * @pcount: value-result argument: value is size of buffer pointed to by @buf, 783 * result is size of @new_buf if program set new value, initial value 784 * otherwise 785 * @ppos: value-result argument: value is position at which read from or write 786 * to sysctl is happening, result is new position if program overrode it, 787 * initial value otherwise 788 * @new_buf: pointer to pointer to new buffer that will be allocated if program 789 * overrides new value provided by user space on sysctl write 790 * NOTE: it's caller responsibility to free *new_buf if it was set 791 * @type: type of program to be executed 792 * 793 * Program is run when sysctl is being accessed, either read or written, and 794 * can allow or deny such access. 795 * 796 * This function will return %-EPERM if an attached program is found and 797 * returned value != 1 during execution. In all other cases 0 is returned. 798 */ 799 int __cgroup_bpf_run_filter_sysctl(struct ctl_table_header *head, 800 struct ctl_table *table, int write, 801 void __user *buf, size_t *pcount, 802 loff_t *ppos, void **new_buf, 803 enum bpf_attach_type type) 804 { 805 struct bpf_sysctl_kern ctx = { 806 .head = head, 807 .table = table, 808 .write = write, 809 .ppos = ppos, 810 .cur_val = NULL, 811 .cur_len = PAGE_SIZE, 812 .new_val = NULL, 813 .new_len = 0, 814 .new_updated = 0, 815 }; 816 struct cgroup *cgrp; 817 int ret; 818 819 ctx.cur_val = kmalloc_track_caller(ctx.cur_len, GFP_KERNEL); 820 if (ctx.cur_val) { 821 mm_segment_t old_fs; 822 loff_t pos = 0; 823 824 old_fs = get_fs(); 825 set_fs(KERNEL_DS); 826 if (table->proc_handler(table, 0, (void __user *)ctx.cur_val, 827 &ctx.cur_len, &pos)) { 828 /* Let BPF program decide how to proceed. */ 829 ctx.cur_len = 0; 830 } 831 set_fs(old_fs); 832 } else { 833 /* Let BPF program decide how to proceed. */ 834 ctx.cur_len = 0; 835 } 836 837 if (write && buf && *pcount) { 838 /* BPF program should be able to override new value with a 839 * buffer bigger than provided by user. 840 */ 841 ctx.new_val = kmalloc_track_caller(PAGE_SIZE, GFP_KERNEL); 842 ctx.new_len = min_t(size_t, PAGE_SIZE, *pcount); 843 if (!ctx.new_val || 844 copy_from_user(ctx.new_val, buf, ctx.new_len)) 845 /* Let BPF program decide how to proceed. */ 846 ctx.new_len = 0; 847 } 848 849 rcu_read_lock(); 850 cgrp = task_dfl_cgroup(current); 851 ret = BPF_PROG_RUN_ARRAY(cgrp->bpf.effective[type], &ctx, BPF_PROG_RUN); 852 rcu_read_unlock(); 853 854 kfree(ctx.cur_val); 855 856 if (ret == 1 && ctx.new_updated) { 857 *new_buf = ctx.new_val; 858 *pcount = ctx.new_len; 859 } else { 860 kfree(ctx.new_val); 861 } 862 863 return ret == 1 ? 0 : -EPERM; 864 } 865 EXPORT_SYMBOL(__cgroup_bpf_run_filter_sysctl); 866 867 static ssize_t sysctl_cpy_dir(const struct ctl_dir *dir, char **bufp, 868 size_t *lenp) 869 { 870 ssize_t tmp_ret = 0, ret; 871 872 if (dir->header.parent) { 873 tmp_ret = sysctl_cpy_dir(dir->header.parent, bufp, lenp); 874 if (tmp_ret < 0) 875 return tmp_ret; 876 } 877 878 ret = strscpy(*bufp, dir->header.ctl_table[0].procname, *lenp); 879 if (ret < 0) 880 return ret; 881 *bufp += ret; 882 *lenp -= ret; 883 ret += tmp_ret; 884 885 /* Avoid leading slash. */ 886 if (!ret) 887 return ret; 888 889 tmp_ret = strscpy(*bufp, "/", *lenp); 890 if (tmp_ret < 0) 891 return tmp_ret; 892 *bufp += tmp_ret; 893 *lenp -= tmp_ret; 894 895 return ret + tmp_ret; 896 } 897 898 BPF_CALL_4(bpf_sysctl_get_name, struct bpf_sysctl_kern *, ctx, char *, buf, 899 size_t, buf_len, u64, flags) 900 { 901 ssize_t tmp_ret = 0, ret; 902 903 if (!buf) 904 return -EINVAL; 905 906 if (!(flags & BPF_F_SYSCTL_BASE_NAME)) { 907 if (!ctx->head) 908 return -EINVAL; 909 tmp_ret = sysctl_cpy_dir(ctx->head->parent, &buf, &buf_len); 910 if (tmp_ret < 0) 911 return tmp_ret; 912 } 913 914 ret = strscpy(buf, ctx->table->procname, buf_len); 915 916 return ret < 0 ? ret : tmp_ret + ret; 917 } 918 919 static const struct bpf_func_proto bpf_sysctl_get_name_proto = { 920 .func = bpf_sysctl_get_name, 921 .gpl_only = false, 922 .ret_type = RET_INTEGER, 923 .arg1_type = ARG_PTR_TO_CTX, 924 .arg2_type = ARG_PTR_TO_MEM, 925 .arg3_type = ARG_CONST_SIZE, 926 .arg4_type = ARG_ANYTHING, 927 }; 928 929 static int copy_sysctl_value(char *dst, size_t dst_len, char *src, 930 size_t src_len) 931 { 932 if (!dst) 933 return -EINVAL; 934 935 if (!dst_len) 936 return -E2BIG; 937 938 if (!src || !src_len) { 939 memset(dst, 0, dst_len); 940 return -EINVAL; 941 } 942 943 memcpy(dst, src, min(dst_len, src_len)); 944 945 if (dst_len > src_len) { 946 memset(dst + src_len, '\0', dst_len - src_len); 947 return src_len; 948 } 949 950 dst[dst_len - 1] = '\0'; 951 952 return -E2BIG; 953 } 954 955 BPF_CALL_3(bpf_sysctl_get_current_value, struct bpf_sysctl_kern *, ctx, 956 char *, buf, size_t, buf_len) 957 { 958 return copy_sysctl_value(buf, buf_len, ctx->cur_val, ctx->cur_len); 959 } 960 961 static const struct bpf_func_proto bpf_sysctl_get_current_value_proto = { 962 .func = bpf_sysctl_get_current_value, 963 .gpl_only = false, 964 .ret_type = RET_INTEGER, 965 .arg1_type = ARG_PTR_TO_CTX, 966 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 967 .arg3_type = ARG_CONST_SIZE, 968 }; 969 970 BPF_CALL_3(bpf_sysctl_get_new_value, struct bpf_sysctl_kern *, ctx, char *, buf, 971 size_t, buf_len) 972 { 973 if (!ctx->write) { 974 if (buf && buf_len) 975 memset(buf, '\0', buf_len); 976 return -EINVAL; 977 } 978 return copy_sysctl_value(buf, buf_len, ctx->new_val, ctx->new_len); 979 } 980 981 static const struct bpf_func_proto bpf_sysctl_get_new_value_proto = { 982 .func = bpf_sysctl_get_new_value, 983 .gpl_only = false, 984 .ret_type = RET_INTEGER, 985 .arg1_type = ARG_PTR_TO_CTX, 986 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 987 .arg3_type = ARG_CONST_SIZE, 988 }; 989 990 BPF_CALL_3(bpf_sysctl_set_new_value, struct bpf_sysctl_kern *, ctx, 991 const char *, buf, size_t, buf_len) 992 { 993 if (!ctx->write || !ctx->new_val || !ctx->new_len || !buf || !buf_len) 994 return -EINVAL; 995 996 if (buf_len > PAGE_SIZE - 1) 997 return -E2BIG; 998 999 memcpy(ctx->new_val, buf, buf_len); 1000 ctx->new_len = buf_len; 1001 ctx->new_updated = 1; 1002 1003 return 0; 1004 } 1005 1006 static const struct bpf_func_proto bpf_sysctl_set_new_value_proto = { 1007 .func = bpf_sysctl_set_new_value, 1008 .gpl_only = false, 1009 .ret_type = RET_INTEGER, 1010 .arg1_type = ARG_PTR_TO_CTX, 1011 .arg2_type = ARG_PTR_TO_MEM, 1012 .arg3_type = ARG_CONST_SIZE, 1013 }; 1014 1015 static const struct bpf_func_proto * 1016 sysctl_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 1017 { 1018 switch (func_id) { 1019 case BPF_FUNC_strtol: 1020 return &bpf_strtol_proto; 1021 case BPF_FUNC_strtoul: 1022 return &bpf_strtoul_proto; 1023 case BPF_FUNC_sysctl_get_name: 1024 return &bpf_sysctl_get_name_proto; 1025 case BPF_FUNC_sysctl_get_current_value: 1026 return &bpf_sysctl_get_current_value_proto; 1027 case BPF_FUNC_sysctl_get_new_value: 1028 return &bpf_sysctl_get_new_value_proto; 1029 case BPF_FUNC_sysctl_set_new_value: 1030 return &bpf_sysctl_set_new_value_proto; 1031 default: 1032 return cgroup_base_func_proto(func_id, prog); 1033 } 1034 } 1035 1036 static bool sysctl_is_valid_access(int off, int size, enum bpf_access_type type, 1037 const struct bpf_prog *prog, 1038 struct bpf_insn_access_aux *info) 1039 { 1040 const int size_default = sizeof(__u32); 1041 1042 if (off < 0 || off + size > sizeof(struct bpf_sysctl) || off % size) 1043 return false; 1044 1045 switch (off) { 1046 case offsetof(struct bpf_sysctl, write): 1047 if (type != BPF_READ) 1048 return false; 1049 bpf_ctx_record_field_size(info, size_default); 1050 return bpf_ctx_narrow_access_ok(off, size, size_default); 1051 case offsetof(struct bpf_sysctl, file_pos): 1052 if (type == BPF_READ) { 1053 bpf_ctx_record_field_size(info, size_default); 1054 return bpf_ctx_narrow_access_ok(off, size, size_default); 1055 } else { 1056 return size == size_default; 1057 } 1058 default: 1059 return false; 1060 } 1061 } 1062 1063 static u32 sysctl_convert_ctx_access(enum bpf_access_type type, 1064 const struct bpf_insn *si, 1065 struct bpf_insn *insn_buf, 1066 struct bpf_prog *prog, u32 *target_size) 1067 { 1068 struct bpf_insn *insn = insn_buf; 1069 1070 switch (si->off) { 1071 case offsetof(struct bpf_sysctl, write): 1072 *insn++ = BPF_LDX_MEM( 1073 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 1074 bpf_target_off(struct bpf_sysctl_kern, write, 1075 FIELD_SIZEOF(struct bpf_sysctl_kern, 1076 write), 1077 target_size)); 1078 break; 1079 case offsetof(struct bpf_sysctl, file_pos): 1080 /* ppos is a pointer so it should be accessed via indirect 1081 * loads and stores. Also for stores additional temporary 1082 * register is used since neither src_reg nor dst_reg can be 1083 * overridden. 1084 */ 1085 if (type == BPF_WRITE) { 1086 int treg = BPF_REG_9; 1087 1088 if (si->src_reg == treg || si->dst_reg == treg) 1089 --treg; 1090 if (si->src_reg == treg || si->dst_reg == treg) 1091 --treg; 1092 *insn++ = BPF_STX_MEM( 1093 BPF_DW, si->dst_reg, treg, 1094 offsetof(struct bpf_sysctl_kern, tmp_reg)); 1095 *insn++ = BPF_LDX_MEM( 1096 BPF_FIELD_SIZEOF(struct bpf_sysctl_kern, ppos), 1097 treg, si->dst_reg, 1098 offsetof(struct bpf_sysctl_kern, ppos)); 1099 *insn++ = BPF_STX_MEM( 1100 BPF_SIZEOF(u32), treg, si->src_reg, 0); 1101 *insn++ = BPF_LDX_MEM( 1102 BPF_DW, treg, si->dst_reg, 1103 offsetof(struct bpf_sysctl_kern, tmp_reg)); 1104 } else { 1105 *insn++ = BPF_LDX_MEM( 1106 BPF_FIELD_SIZEOF(struct bpf_sysctl_kern, ppos), 1107 si->dst_reg, si->src_reg, 1108 offsetof(struct bpf_sysctl_kern, ppos)); 1109 *insn++ = BPF_LDX_MEM( 1110 BPF_SIZE(si->code), si->dst_reg, si->dst_reg, 0); 1111 } 1112 *target_size = sizeof(u32); 1113 break; 1114 } 1115 1116 return insn - insn_buf; 1117 } 1118 1119 const struct bpf_verifier_ops cg_sysctl_verifier_ops = { 1120 .get_func_proto = sysctl_func_proto, 1121 .is_valid_access = sysctl_is_valid_access, 1122 .convert_ctx_access = sysctl_convert_ctx_access, 1123 }; 1124 1125 const struct bpf_prog_ops cg_sysctl_prog_ops = { 1126 }; 1127