1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Functions to manage eBPF programs attached to cgroups 4 * 5 * Copyright (c) 2016 Daniel Mack 6 */ 7 8 #include <linux/kernel.h> 9 #include <linux/atomic.h> 10 #include <linux/cgroup.h> 11 #include <linux/filter.h> 12 #include <linux/slab.h> 13 #include <linux/sysctl.h> 14 #include <linux/string.h> 15 #include <linux/bpf.h> 16 #include <linux/bpf-cgroup.h> 17 #include <net/sock.h> 18 #include <net/bpf_sk_storage.h> 19 20 #include "../cgroup/cgroup-internal.h" 21 22 DEFINE_STATIC_KEY_FALSE(cgroup_bpf_enabled_key); 23 EXPORT_SYMBOL(cgroup_bpf_enabled_key); 24 25 void cgroup_bpf_offline(struct cgroup *cgrp) 26 { 27 cgroup_get(cgrp); 28 percpu_ref_kill(&cgrp->bpf.refcnt); 29 } 30 31 static void bpf_cgroup_storages_free(struct bpf_cgroup_storage *storages[]) 32 { 33 enum bpf_cgroup_storage_type stype; 34 35 for_each_cgroup_storage_type(stype) 36 bpf_cgroup_storage_free(storages[stype]); 37 } 38 39 static int bpf_cgroup_storages_alloc(struct bpf_cgroup_storage *storages[], 40 struct bpf_prog *prog) 41 { 42 enum bpf_cgroup_storage_type stype; 43 44 for_each_cgroup_storage_type(stype) { 45 storages[stype] = bpf_cgroup_storage_alloc(prog, stype); 46 if (IS_ERR(storages[stype])) { 47 storages[stype] = NULL; 48 bpf_cgroup_storages_free(storages); 49 return -ENOMEM; 50 } 51 } 52 53 return 0; 54 } 55 56 static void bpf_cgroup_storages_assign(struct bpf_cgroup_storage *dst[], 57 struct bpf_cgroup_storage *src[]) 58 { 59 enum bpf_cgroup_storage_type stype; 60 61 for_each_cgroup_storage_type(stype) 62 dst[stype] = src[stype]; 63 } 64 65 static void bpf_cgroup_storages_link(struct bpf_cgroup_storage *storages[], 66 struct cgroup* cgrp, 67 enum bpf_attach_type attach_type) 68 { 69 enum bpf_cgroup_storage_type stype; 70 71 for_each_cgroup_storage_type(stype) 72 bpf_cgroup_storage_link(storages[stype], cgrp, attach_type); 73 } 74 75 static void bpf_cgroup_storages_unlink(struct bpf_cgroup_storage *storages[]) 76 { 77 enum bpf_cgroup_storage_type stype; 78 79 for_each_cgroup_storage_type(stype) 80 bpf_cgroup_storage_unlink(storages[stype]); 81 } 82 83 /* Called when bpf_cgroup_link is auto-detached from dying cgroup. 84 * It drops cgroup and bpf_prog refcounts, and marks bpf_link as defunct. It 85 * doesn't free link memory, which will eventually be done by bpf_link's 86 * release() callback, when its last FD is closed. 87 */ 88 static void bpf_cgroup_link_auto_detach(struct bpf_cgroup_link *link) 89 { 90 cgroup_put(link->cgroup); 91 link->cgroup = NULL; 92 } 93 94 /** 95 * cgroup_bpf_release() - put references of all bpf programs and 96 * release all cgroup bpf data 97 * @work: work structure embedded into the cgroup to modify 98 */ 99 static void cgroup_bpf_release(struct work_struct *work) 100 { 101 struct cgroup *p, *cgrp = container_of(work, struct cgroup, 102 bpf.release_work); 103 struct bpf_prog_array *old_array; 104 unsigned int type; 105 106 mutex_lock(&cgroup_mutex); 107 108 for (type = 0; type < ARRAY_SIZE(cgrp->bpf.progs); type++) { 109 struct list_head *progs = &cgrp->bpf.progs[type]; 110 struct bpf_prog_list *pl, *tmp; 111 112 list_for_each_entry_safe(pl, tmp, progs, node) { 113 list_del(&pl->node); 114 if (pl->prog) 115 bpf_prog_put(pl->prog); 116 if (pl->link) 117 bpf_cgroup_link_auto_detach(pl->link); 118 bpf_cgroup_storages_unlink(pl->storage); 119 bpf_cgroup_storages_free(pl->storage); 120 kfree(pl); 121 static_branch_dec(&cgroup_bpf_enabled_key); 122 } 123 old_array = rcu_dereference_protected( 124 cgrp->bpf.effective[type], 125 lockdep_is_held(&cgroup_mutex)); 126 bpf_prog_array_free(old_array); 127 } 128 129 mutex_unlock(&cgroup_mutex); 130 131 for (p = cgroup_parent(cgrp); p; p = cgroup_parent(p)) 132 cgroup_bpf_put(p); 133 134 percpu_ref_exit(&cgrp->bpf.refcnt); 135 cgroup_put(cgrp); 136 } 137 138 /** 139 * cgroup_bpf_release_fn() - callback used to schedule releasing 140 * of bpf cgroup data 141 * @ref: percpu ref counter structure 142 */ 143 static void cgroup_bpf_release_fn(struct percpu_ref *ref) 144 { 145 struct cgroup *cgrp = container_of(ref, struct cgroup, bpf.refcnt); 146 147 INIT_WORK(&cgrp->bpf.release_work, cgroup_bpf_release); 148 queue_work(system_wq, &cgrp->bpf.release_work); 149 } 150 151 /* Get underlying bpf_prog of bpf_prog_list entry, regardless if it's through 152 * link or direct prog. 153 */ 154 static struct bpf_prog *prog_list_prog(struct bpf_prog_list *pl) 155 { 156 if (pl->prog) 157 return pl->prog; 158 if (pl->link) 159 return pl->link->link.prog; 160 return NULL; 161 } 162 163 /* count number of elements in the list. 164 * it's slow but the list cannot be long 165 */ 166 static u32 prog_list_length(struct list_head *head) 167 { 168 struct bpf_prog_list *pl; 169 u32 cnt = 0; 170 171 list_for_each_entry(pl, head, node) { 172 if (!prog_list_prog(pl)) 173 continue; 174 cnt++; 175 } 176 return cnt; 177 } 178 179 /* if parent has non-overridable prog attached, 180 * disallow attaching new programs to the descendent cgroup. 181 * if parent has overridable or multi-prog, allow attaching 182 */ 183 static bool hierarchy_allows_attach(struct cgroup *cgrp, 184 enum bpf_attach_type type) 185 { 186 struct cgroup *p; 187 188 p = cgroup_parent(cgrp); 189 if (!p) 190 return true; 191 do { 192 u32 flags = p->bpf.flags[type]; 193 u32 cnt; 194 195 if (flags & BPF_F_ALLOW_MULTI) 196 return true; 197 cnt = prog_list_length(&p->bpf.progs[type]); 198 WARN_ON_ONCE(cnt > 1); 199 if (cnt == 1) 200 return !!(flags & BPF_F_ALLOW_OVERRIDE); 201 p = cgroup_parent(p); 202 } while (p); 203 return true; 204 } 205 206 /* compute a chain of effective programs for a given cgroup: 207 * start from the list of programs in this cgroup and add 208 * all parent programs. 209 * Note that parent's F_ALLOW_OVERRIDE-type program is yielding 210 * to programs in this cgroup 211 */ 212 static int compute_effective_progs(struct cgroup *cgrp, 213 enum bpf_attach_type type, 214 struct bpf_prog_array **array) 215 { 216 struct bpf_prog_array_item *item; 217 struct bpf_prog_array *progs; 218 struct bpf_prog_list *pl; 219 struct cgroup *p = cgrp; 220 int cnt = 0; 221 222 /* count number of effective programs by walking parents */ 223 do { 224 if (cnt == 0 || (p->bpf.flags[type] & BPF_F_ALLOW_MULTI)) 225 cnt += prog_list_length(&p->bpf.progs[type]); 226 p = cgroup_parent(p); 227 } while (p); 228 229 progs = bpf_prog_array_alloc(cnt, GFP_KERNEL); 230 if (!progs) 231 return -ENOMEM; 232 233 /* populate the array with effective progs */ 234 cnt = 0; 235 p = cgrp; 236 do { 237 if (cnt > 0 && !(p->bpf.flags[type] & BPF_F_ALLOW_MULTI)) 238 continue; 239 240 list_for_each_entry(pl, &p->bpf.progs[type], node) { 241 if (!prog_list_prog(pl)) 242 continue; 243 244 item = &progs->items[cnt]; 245 item->prog = prog_list_prog(pl); 246 bpf_cgroup_storages_assign(item->cgroup_storage, 247 pl->storage); 248 cnt++; 249 } 250 } while ((p = cgroup_parent(p))); 251 252 *array = progs; 253 return 0; 254 } 255 256 static void activate_effective_progs(struct cgroup *cgrp, 257 enum bpf_attach_type type, 258 struct bpf_prog_array *old_array) 259 { 260 old_array = rcu_replace_pointer(cgrp->bpf.effective[type], old_array, 261 lockdep_is_held(&cgroup_mutex)); 262 /* free prog array after grace period, since __cgroup_bpf_run_*() 263 * might be still walking the array 264 */ 265 bpf_prog_array_free(old_array); 266 } 267 268 /** 269 * cgroup_bpf_inherit() - inherit effective programs from parent 270 * @cgrp: the cgroup to modify 271 */ 272 int cgroup_bpf_inherit(struct cgroup *cgrp) 273 { 274 /* has to use marco instead of const int, since compiler thinks 275 * that array below is variable length 276 */ 277 #define NR ARRAY_SIZE(cgrp->bpf.effective) 278 struct bpf_prog_array *arrays[NR] = {}; 279 struct cgroup *p; 280 int ret, i; 281 282 ret = percpu_ref_init(&cgrp->bpf.refcnt, cgroup_bpf_release_fn, 0, 283 GFP_KERNEL); 284 if (ret) 285 return ret; 286 287 for (p = cgroup_parent(cgrp); p; p = cgroup_parent(p)) 288 cgroup_bpf_get(p); 289 290 for (i = 0; i < NR; i++) 291 INIT_LIST_HEAD(&cgrp->bpf.progs[i]); 292 293 for (i = 0; i < NR; i++) 294 if (compute_effective_progs(cgrp, i, &arrays[i])) 295 goto cleanup; 296 297 for (i = 0; i < NR; i++) 298 activate_effective_progs(cgrp, i, arrays[i]); 299 300 return 0; 301 cleanup: 302 for (i = 0; i < NR; i++) 303 bpf_prog_array_free(arrays[i]); 304 305 percpu_ref_exit(&cgrp->bpf.refcnt); 306 307 return -ENOMEM; 308 } 309 310 static int update_effective_progs(struct cgroup *cgrp, 311 enum bpf_attach_type type) 312 { 313 struct cgroup_subsys_state *css; 314 int err; 315 316 /* allocate and recompute effective prog arrays */ 317 css_for_each_descendant_pre(css, &cgrp->self) { 318 struct cgroup *desc = container_of(css, struct cgroup, self); 319 320 if (percpu_ref_is_zero(&desc->bpf.refcnt)) 321 continue; 322 323 err = compute_effective_progs(desc, type, &desc->bpf.inactive); 324 if (err) 325 goto cleanup; 326 } 327 328 /* all allocations were successful. Activate all prog arrays */ 329 css_for_each_descendant_pre(css, &cgrp->self) { 330 struct cgroup *desc = container_of(css, struct cgroup, self); 331 332 if (percpu_ref_is_zero(&desc->bpf.refcnt)) { 333 if (unlikely(desc->bpf.inactive)) { 334 bpf_prog_array_free(desc->bpf.inactive); 335 desc->bpf.inactive = NULL; 336 } 337 continue; 338 } 339 340 activate_effective_progs(desc, type, desc->bpf.inactive); 341 desc->bpf.inactive = NULL; 342 } 343 344 return 0; 345 346 cleanup: 347 /* oom while computing effective. Free all computed effective arrays 348 * since they were not activated 349 */ 350 css_for_each_descendant_pre(css, &cgrp->self) { 351 struct cgroup *desc = container_of(css, struct cgroup, self); 352 353 bpf_prog_array_free(desc->bpf.inactive); 354 desc->bpf.inactive = NULL; 355 } 356 357 return err; 358 } 359 360 #define BPF_CGROUP_MAX_PROGS 64 361 362 static struct bpf_prog_list *find_attach_entry(struct list_head *progs, 363 struct bpf_prog *prog, 364 struct bpf_cgroup_link *link, 365 struct bpf_prog *replace_prog, 366 bool allow_multi) 367 { 368 struct bpf_prog_list *pl; 369 370 /* single-attach case */ 371 if (!allow_multi) { 372 if (list_empty(progs)) 373 return NULL; 374 return list_first_entry(progs, typeof(*pl), node); 375 } 376 377 list_for_each_entry(pl, progs, node) { 378 if (prog && pl->prog == prog) 379 /* disallow attaching the same prog twice */ 380 return ERR_PTR(-EINVAL); 381 if (link && pl->link == link) 382 /* disallow attaching the same link twice */ 383 return ERR_PTR(-EINVAL); 384 } 385 386 /* direct prog multi-attach w/ replacement case */ 387 if (replace_prog) { 388 list_for_each_entry(pl, progs, node) { 389 if (pl->prog == replace_prog) 390 /* a match found */ 391 return pl; 392 } 393 /* prog to replace not found for cgroup */ 394 return ERR_PTR(-ENOENT); 395 } 396 397 return NULL; 398 } 399 400 /** 401 * __cgroup_bpf_attach() - Attach the program or the link to a cgroup, and 402 * propagate the change to descendants 403 * @cgrp: The cgroup which descendants to traverse 404 * @prog: A program to attach 405 * @link: A link to attach 406 * @replace_prog: Previously attached program to replace if BPF_F_REPLACE is set 407 * @type: Type of attach operation 408 * @flags: Option flags 409 * 410 * Exactly one of @prog or @link can be non-null. 411 * Must be called with cgroup_mutex held. 412 */ 413 int __cgroup_bpf_attach(struct cgroup *cgrp, 414 struct bpf_prog *prog, struct bpf_prog *replace_prog, 415 struct bpf_cgroup_link *link, 416 enum bpf_attach_type type, u32 flags) 417 { 418 u32 saved_flags = (flags & (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)); 419 struct list_head *progs = &cgrp->bpf.progs[type]; 420 struct bpf_prog *old_prog = NULL; 421 struct bpf_cgroup_storage *storage[MAX_BPF_CGROUP_STORAGE_TYPE], 422 *old_storage[MAX_BPF_CGROUP_STORAGE_TYPE] = {NULL}; 423 struct bpf_prog_list *pl; 424 int err; 425 426 if (((flags & BPF_F_ALLOW_OVERRIDE) && (flags & BPF_F_ALLOW_MULTI)) || 427 ((flags & BPF_F_REPLACE) && !(flags & BPF_F_ALLOW_MULTI))) 428 /* invalid combination */ 429 return -EINVAL; 430 if (link && (prog || replace_prog)) 431 /* only either link or prog/replace_prog can be specified */ 432 return -EINVAL; 433 if (!!replace_prog != !!(flags & BPF_F_REPLACE)) 434 /* replace_prog implies BPF_F_REPLACE, and vice versa */ 435 return -EINVAL; 436 437 if (!hierarchy_allows_attach(cgrp, type)) 438 return -EPERM; 439 440 if (!list_empty(progs) && cgrp->bpf.flags[type] != saved_flags) 441 /* Disallow attaching non-overridable on top 442 * of existing overridable in this cgroup. 443 * Disallow attaching multi-prog if overridable or none 444 */ 445 return -EPERM; 446 447 if (prog_list_length(progs) >= BPF_CGROUP_MAX_PROGS) 448 return -E2BIG; 449 450 pl = find_attach_entry(progs, prog, link, replace_prog, 451 flags & BPF_F_ALLOW_MULTI); 452 if (IS_ERR(pl)) 453 return PTR_ERR(pl); 454 455 if (bpf_cgroup_storages_alloc(storage, prog ? : link->link.prog)) 456 return -ENOMEM; 457 458 if (pl) { 459 old_prog = pl->prog; 460 bpf_cgroup_storages_unlink(pl->storage); 461 bpf_cgroup_storages_assign(old_storage, pl->storage); 462 } else { 463 pl = kmalloc(sizeof(*pl), GFP_KERNEL); 464 if (!pl) { 465 bpf_cgroup_storages_free(storage); 466 return -ENOMEM; 467 } 468 list_add_tail(&pl->node, progs); 469 } 470 471 pl->prog = prog; 472 pl->link = link; 473 bpf_cgroup_storages_assign(pl->storage, storage); 474 cgrp->bpf.flags[type] = saved_flags; 475 476 err = update_effective_progs(cgrp, type); 477 if (err) 478 goto cleanup; 479 480 bpf_cgroup_storages_free(old_storage); 481 if (old_prog) 482 bpf_prog_put(old_prog); 483 else 484 static_branch_inc(&cgroup_bpf_enabled_key); 485 bpf_cgroup_storages_link(pl->storage, cgrp, type); 486 return 0; 487 488 cleanup: 489 if (old_prog) { 490 pl->prog = old_prog; 491 pl->link = NULL; 492 } 493 bpf_cgroup_storages_free(pl->storage); 494 bpf_cgroup_storages_assign(pl->storage, old_storage); 495 bpf_cgroup_storages_link(pl->storage, cgrp, type); 496 if (!old_prog) { 497 list_del(&pl->node); 498 kfree(pl); 499 } 500 return err; 501 } 502 503 static struct bpf_prog_list *find_detach_entry(struct list_head *progs, 504 struct bpf_prog *prog, 505 struct bpf_cgroup_link *link, 506 bool allow_multi) 507 { 508 struct bpf_prog_list *pl; 509 510 if (!allow_multi) { 511 if (list_empty(progs)) 512 /* report error when trying to detach and nothing is attached */ 513 return ERR_PTR(-ENOENT); 514 515 /* to maintain backward compatibility NONE and OVERRIDE cgroups 516 * allow detaching with invalid FD (prog==NULL) in legacy mode 517 */ 518 return list_first_entry(progs, typeof(*pl), node); 519 } 520 521 if (!prog && !link) 522 /* to detach MULTI prog the user has to specify valid FD 523 * of the program or link to be detached 524 */ 525 return ERR_PTR(-EINVAL); 526 527 /* find the prog or link and detach it */ 528 list_for_each_entry(pl, progs, node) { 529 if (pl->prog == prog && pl->link == link) 530 return pl; 531 } 532 return ERR_PTR(-ENOENT); 533 } 534 535 /** 536 * __cgroup_bpf_detach() - Detach the program or link from a cgroup, and 537 * propagate the change to descendants 538 * @cgrp: The cgroup which descendants to traverse 539 * @prog: A program to detach or NULL 540 * @prog: A link to detach or NULL 541 * @type: Type of detach operation 542 * 543 * At most one of @prog or @link can be non-NULL. 544 * Must be called with cgroup_mutex held. 545 */ 546 int __cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog, 547 struct bpf_cgroup_link *link, enum bpf_attach_type type) 548 { 549 struct list_head *progs = &cgrp->bpf.progs[type]; 550 u32 flags = cgrp->bpf.flags[type]; 551 struct bpf_prog_list *pl; 552 struct bpf_prog *old_prog; 553 int err; 554 555 if (prog && link) 556 /* only one of prog or link can be specified */ 557 return -EINVAL; 558 559 pl = find_detach_entry(progs, prog, link, flags & BPF_F_ALLOW_MULTI); 560 if (IS_ERR(pl)) 561 return PTR_ERR(pl); 562 563 /* mark it deleted, so it's ignored while recomputing effective */ 564 old_prog = pl->prog; 565 pl->prog = NULL; 566 pl->link = NULL; 567 568 err = update_effective_progs(cgrp, type); 569 if (err) 570 goto cleanup; 571 572 /* now can actually delete it from this cgroup list */ 573 list_del(&pl->node); 574 bpf_cgroup_storages_unlink(pl->storage); 575 bpf_cgroup_storages_free(pl->storage); 576 kfree(pl); 577 if (list_empty(progs)) 578 /* last program was detached, reset flags to zero */ 579 cgrp->bpf.flags[type] = 0; 580 if (old_prog) 581 bpf_prog_put(old_prog); 582 static_branch_dec(&cgroup_bpf_enabled_key); 583 return 0; 584 585 cleanup: 586 /* restore back prog or link */ 587 pl->prog = old_prog; 588 pl->link = link; 589 return err; 590 } 591 592 /* Must be called with cgroup_mutex held to avoid races. */ 593 int __cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr, 594 union bpf_attr __user *uattr) 595 { 596 __u32 __user *prog_ids = u64_to_user_ptr(attr->query.prog_ids); 597 enum bpf_attach_type type = attr->query.attach_type; 598 struct list_head *progs = &cgrp->bpf.progs[type]; 599 u32 flags = cgrp->bpf.flags[type]; 600 struct bpf_prog_array *effective; 601 struct bpf_prog *prog; 602 int cnt, ret = 0, i; 603 604 effective = rcu_dereference_protected(cgrp->bpf.effective[type], 605 lockdep_is_held(&cgroup_mutex)); 606 607 if (attr->query.query_flags & BPF_F_QUERY_EFFECTIVE) 608 cnt = bpf_prog_array_length(effective); 609 else 610 cnt = prog_list_length(progs); 611 612 if (copy_to_user(&uattr->query.attach_flags, &flags, sizeof(flags))) 613 return -EFAULT; 614 if (copy_to_user(&uattr->query.prog_cnt, &cnt, sizeof(cnt))) 615 return -EFAULT; 616 if (attr->query.prog_cnt == 0 || !prog_ids || !cnt) 617 /* return early if user requested only program count + flags */ 618 return 0; 619 if (attr->query.prog_cnt < cnt) { 620 cnt = attr->query.prog_cnt; 621 ret = -ENOSPC; 622 } 623 624 if (attr->query.query_flags & BPF_F_QUERY_EFFECTIVE) { 625 return bpf_prog_array_copy_to_user(effective, prog_ids, cnt); 626 } else { 627 struct bpf_prog_list *pl; 628 u32 id; 629 630 i = 0; 631 list_for_each_entry(pl, progs, node) { 632 prog = prog_list_prog(pl); 633 id = prog->aux->id; 634 if (copy_to_user(prog_ids + i, &id, sizeof(id))) 635 return -EFAULT; 636 if (++i == cnt) 637 break; 638 } 639 } 640 return ret; 641 } 642 643 int cgroup_bpf_prog_attach(const union bpf_attr *attr, 644 enum bpf_prog_type ptype, struct bpf_prog *prog) 645 { 646 struct bpf_prog *replace_prog = NULL; 647 struct cgroup *cgrp; 648 int ret; 649 650 cgrp = cgroup_get_from_fd(attr->target_fd); 651 if (IS_ERR(cgrp)) 652 return PTR_ERR(cgrp); 653 654 if ((attr->attach_flags & BPF_F_ALLOW_MULTI) && 655 (attr->attach_flags & BPF_F_REPLACE)) { 656 replace_prog = bpf_prog_get_type(attr->replace_bpf_fd, ptype); 657 if (IS_ERR(replace_prog)) { 658 cgroup_put(cgrp); 659 return PTR_ERR(replace_prog); 660 } 661 } 662 663 ret = cgroup_bpf_attach(cgrp, prog, replace_prog, NULL, 664 attr->attach_type, attr->attach_flags); 665 666 if (replace_prog) 667 bpf_prog_put(replace_prog); 668 cgroup_put(cgrp); 669 return ret; 670 } 671 672 int cgroup_bpf_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype) 673 { 674 struct bpf_prog *prog; 675 struct cgroup *cgrp; 676 int ret; 677 678 cgrp = cgroup_get_from_fd(attr->target_fd); 679 if (IS_ERR(cgrp)) 680 return PTR_ERR(cgrp); 681 682 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype); 683 if (IS_ERR(prog)) 684 prog = NULL; 685 686 ret = cgroup_bpf_detach(cgrp, prog, attr->attach_type); 687 if (prog) 688 bpf_prog_put(prog); 689 690 cgroup_put(cgrp); 691 return ret; 692 } 693 694 static void bpf_cgroup_link_release(struct bpf_link *link) 695 { 696 struct bpf_cgroup_link *cg_link = 697 container_of(link, struct bpf_cgroup_link, link); 698 699 /* link might have been auto-detached by dying cgroup already, 700 * in that case our work is done here 701 */ 702 if (!cg_link->cgroup) 703 return; 704 705 mutex_lock(&cgroup_mutex); 706 707 /* re-check cgroup under lock again */ 708 if (!cg_link->cgroup) { 709 mutex_unlock(&cgroup_mutex); 710 return; 711 } 712 713 WARN_ON(__cgroup_bpf_detach(cg_link->cgroup, NULL, cg_link, 714 cg_link->type)); 715 716 mutex_unlock(&cgroup_mutex); 717 cgroup_put(cg_link->cgroup); 718 } 719 720 static void bpf_cgroup_link_dealloc(struct bpf_link *link) 721 { 722 struct bpf_cgroup_link *cg_link = 723 container_of(link, struct bpf_cgroup_link, link); 724 725 kfree(cg_link); 726 } 727 728 const struct bpf_link_ops bpf_cgroup_link_lops = { 729 .release = bpf_cgroup_link_release, 730 .dealloc = bpf_cgroup_link_dealloc, 731 }; 732 733 int cgroup_bpf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 734 { 735 struct bpf_cgroup_link *link; 736 struct file *link_file; 737 struct cgroup *cgrp; 738 int err, link_fd; 739 740 if (attr->link_create.flags) 741 return -EINVAL; 742 743 cgrp = cgroup_get_from_fd(attr->link_create.target_fd); 744 if (IS_ERR(cgrp)) 745 return PTR_ERR(cgrp); 746 747 link = kzalloc(sizeof(*link), GFP_USER); 748 if (!link) { 749 err = -ENOMEM; 750 goto out_put_cgroup; 751 } 752 bpf_link_init(&link->link, &bpf_cgroup_link_lops, prog); 753 link->cgroup = cgrp; 754 link->type = attr->link_create.attach_type; 755 756 link_file = bpf_link_new_file(&link->link, &link_fd); 757 if (IS_ERR(link_file)) { 758 kfree(link); 759 err = PTR_ERR(link_file); 760 goto out_put_cgroup; 761 } 762 763 err = cgroup_bpf_attach(cgrp, NULL, NULL, link, link->type, 764 BPF_F_ALLOW_MULTI); 765 if (err) { 766 bpf_link_cleanup(&link->link, link_file, link_fd); 767 goto out_put_cgroup; 768 } 769 770 fd_install(link_fd, link_file); 771 return link_fd; 772 773 out_put_cgroup: 774 cgroup_put(cgrp); 775 return err; 776 } 777 778 int cgroup_bpf_prog_query(const union bpf_attr *attr, 779 union bpf_attr __user *uattr) 780 { 781 struct cgroup *cgrp; 782 int ret; 783 784 cgrp = cgroup_get_from_fd(attr->query.target_fd); 785 if (IS_ERR(cgrp)) 786 return PTR_ERR(cgrp); 787 788 ret = cgroup_bpf_query(cgrp, attr, uattr); 789 790 cgroup_put(cgrp); 791 return ret; 792 } 793 794 /** 795 * __cgroup_bpf_run_filter_skb() - Run a program for packet filtering 796 * @sk: The socket sending or receiving traffic 797 * @skb: The skb that is being sent or received 798 * @type: The type of program to be exectuted 799 * 800 * If no socket is passed, or the socket is not of type INET or INET6, 801 * this function does nothing and returns 0. 802 * 803 * The program type passed in via @type must be suitable for network 804 * filtering. No further check is performed to assert that. 805 * 806 * For egress packets, this function can return: 807 * NET_XMIT_SUCCESS (0) - continue with packet output 808 * NET_XMIT_DROP (1) - drop packet and notify TCP to call cwr 809 * NET_XMIT_CN (2) - continue with packet output and notify TCP 810 * to call cwr 811 * -EPERM - drop packet 812 * 813 * For ingress packets, this function will return -EPERM if any 814 * attached program was found and if it returned != 1 during execution. 815 * Otherwise 0 is returned. 816 */ 817 int __cgroup_bpf_run_filter_skb(struct sock *sk, 818 struct sk_buff *skb, 819 enum bpf_attach_type type) 820 { 821 unsigned int offset = skb->data - skb_network_header(skb); 822 struct sock *save_sk; 823 void *saved_data_end; 824 struct cgroup *cgrp; 825 int ret; 826 827 if (!sk || !sk_fullsock(sk)) 828 return 0; 829 830 if (sk->sk_family != AF_INET && sk->sk_family != AF_INET6) 831 return 0; 832 833 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 834 save_sk = skb->sk; 835 skb->sk = sk; 836 __skb_push(skb, offset); 837 838 /* compute pointers for the bpf prog */ 839 bpf_compute_and_save_data_end(skb, &saved_data_end); 840 841 if (type == BPF_CGROUP_INET_EGRESS) { 842 ret = BPF_PROG_CGROUP_INET_EGRESS_RUN_ARRAY( 843 cgrp->bpf.effective[type], skb, __bpf_prog_run_save_cb); 844 } else { 845 ret = BPF_PROG_RUN_ARRAY(cgrp->bpf.effective[type], skb, 846 __bpf_prog_run_save_cb); 847 ret = (ret == 1 ? 0 : -EPERM); 848 } 849 bpf_restore_data_end(skb, saved_data_end); 850 __skb_pull(skb, offset); 851 skb->sk = save_sk; 852 853 return ret; 854 } 855 EXPORT_SYMBOL(__cgroup_bpf_run_filter_skb); 856 857 /** 858 * __cgroup_bpf_run_filter_sk() - Run a program on a sock 859 * @sk: sock structure to manipulate 860 * @type: The type of program to be exectuted 861 * 862 * socket is passed is expected to be of type INET or INET6. 863 * 864 * The program type passed in via @type must be suitable for sock 865 * filtering. No further check is performed to assert that. 866 * 867 * This function will return %-EPERM if any if an attached program was found 868 * and if it returned != 1 during execution. In all other cases, 0 is returned. 869 */ 870 int __cgroup_bpf_run_filter_sk(struct sock *sk, 871 enum bpf_attach_type type) 872 { 873 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 874 int ret; 875 876 ret = BPF_PROG_RUN_ARRAY(cgrp->bpf.effective[type], sk, BPF_PROG_RUN); 877 return ret == 1 ? 0 : -EPERM; 878 } 879 EXPORT_SYMBOL(__cgroup_bpf_run_filter_sk); 880 881 /** 882 * __cgroup_bpf_run_filter_sock_addr() - Run a program on a sock and 883 * provided by user sockaddr 884 * @sk: sock struct that will use sockaddr 885 * @uaddr: sockaddr struct provided by user 886 * @type: The type of program to be exectuted 887 * @t_ctx: Pointer to attach type specific context 888 * 889 * socket is expected to be of type INET or INET6. 890 * 891 * This function will return %-EPERM if an attached program is found and 892 * returned value != 1 during execution. In all other cases, 0 is returned. 893 */ 894 int __cgroup_bpf_run_filter_sock_addr(struct sock *sk, 895 struct sockaddr *uaddr, 896 enum bpf_attach_type type, 897 void *t_ctx) 898 { 899 struct bpf_sock_addr_kern ctx = { 900 .sk = sk, 901 .uaddr = uaddr, 902 .t_ctx = t_ctx, 903 }; 904 struct sockaddr_storage unspec; 905 struct cgroup *cgrp; 906 int ret; 907 908 /* Check socket family since not all sockets represent network 909 * endpoint (e.g. AF_UNIX). 910 */ 911 if (sk->sk_family != AF_INET && sk->sk_family != AF_INET6) 912 return 0; 913 914 if (!ctx.uaddr) { 915 memset(&unspec, 0, sizeof(unspec)); 916 ctx.uaddr = (struct sockaddr *)&unspec; 917 } 918 919 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 920 ret = BPF_PROG_RUN_ARRAY(cgrp->bpf.effective[type], &ctx, BPF_PROG_RUN); 921 922 return ret == 1 ? 0 : -EPERM; 923 } 924 EXPORT_SYMBOL(__cgroup_bpf_run_filter_sock_addr); 925 926 /** 927 * __cgroup_bpf_run_filter_sock_ops() - Run a program on a sock 928 * @sk: socket to get cgroup from 929 * @sock_ops: bpf_sock_ops_kern struct to pass to program. Contains 930 * sk with connection information (IP addresses, etc.) May not contain 931 * cgroup info if it is a req sock. 932 * @type: The type of program to be exectuted 933 * 934 * socket passed is expected to be of type INET or INET6. 935 * 936 * The program type passed in via @type must be suitable for sock_ops 937 * filtering. No further check is performed to assert that. 938 * 939 * This function will return %-EPERM if any if an attached program was found 940 * and if it returned != 1 during execution. In all other cases, 0 is returned. 941 */ 942 int __cgroup_bpf_run_filter_sock_ops(struct sock *sk, 943 struct bpf_sock_ops_kern *sock_ops, 944 enum bpf_attach_type type) 945 { 946 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 947 int ret; 948 949 ret = BPF_PROG_RUN_ARRAY(cgrp->bpf.effective[type], sock_ops, 950 BPF_PROG_RUN); 951 return ret == 1 ? 0 : -EPERM; 952 } 953 EXPORT_SYMBOL(__cgroup_bpf_run_filter_sock_ops); 954 955 int __cgroup_bpf_check_dev_permission(short dev_type, u32 major, u32 minor, 956 short access, enum bpf_attach_type type) 957 { 958 struct cgroup *cgrp; 959 struct bpf_cgroup_dev_ctx ctx = { 960 .access_type = (access << 16) | dev_type, 961 .major = major, 962 .minor = minor, 963 }; 964 int allow = 1; 965 966 rcu_read_lock(); 967 cgrp = task_dfl_cgroup(current); 968 allow = BPF_PROG_RUN_ARRAY(cgrp->bpf.effective[type], &ctx, 969 BPF_PROG_RUN); 970 rcu_read_unlock(); 971 972 return !allow; 973 } 974 EXPORT_SYMBOL(__cgroup_bpf_check_dev_permission); 975 976 static const struct bpf_func_proto * 977 cgroup_base_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 978 { 979 switch (func_id) { 980 case BPF_FUNC_map_lookup_elem: 981 return &bpf_map_lookup_elem_proto; 982 case BPF_FUNC_map_update_elem: 983 return &bpf_map_update_elem_proto; 984 case BPF_FUNC_map_delete_elem: 985 return &bpf_map_delete_elem_proto; 986 case BPF_FUNC_map_push_elem: 987 return &bpf_map_push_elem_proto; 988 case BPF_FUNC_map_pop_elem: 989 return &bpf_map_pop_elem_proto; 990 case BPF_FUNC_map_peek_elem: 991 return &bpf_map_peek_elem_proto; 992 case BPF_FUNC_get_current_uid_gid: 993 return &bpf_get_current_uid_gid_proto; 994 case BPF_FUNC_get_local_storage: 995 return &bpf_get_local_storage_proto; 996 case BPF_FUNC_get_current_cgroup_id: 997 return &bpf_get_current_cgroup_id_proto; 998 case BPF_FUNC_trace_printk: 999 if (capable(CAP_SYS_ADMIN)) 1000 return bpf_get_trace_printk_proto(); 1001 /* fall through */ 1002 default: 1003 return NULL; 1004 } 1005 } 1006 1007 static const struct bpf_func_proto * 1008 cgroup_dev_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 1009 { 1010 return cgroup_base_func_proto(func_id, prog); 1011 } 1012 1013 static bool cgroup_dev_is_valid_access(int off, int size, 1014 enum bpf_access_type type, 1015 const struct bpf_prog *prog, 1016 struct bpf_insn_access_aux *info) 1017 { 1018 const int size_default = sizeof(__u32); 1019 1020 if (type == BPF_WRITE) 1021 return false; 1022 1023 if (off < 0 || off + size > sizeof(struct bpf_cgroup_dev_ctx)) 1024 return false; 1025 /* The verifier guarantees that size > 0. */ 1026 if (off % size != 0) 1027 return false; 1028 1029 switch (off) { 1030 case bpf_ctx_range(struct bpf_cgroup_dev_ctx, access_type): 1031 bpf_ctx_record_field_size(info, size_default); 1032 if (!bpf_ctx_narrow_access_ok(off, size, size_default)) 1033 return false; 1034 break; 1035 default: 1036 if (size != size_default) 1037 return false; 1038 } 1039 1040 return true; 1041 } 1042 1043 const struct bpf_prog_ops cg_dev_prog_ops = { 1044 }; 1045 1046 const struct bpf_verifier_ops cg_dev_verifier_ops = { 1047 .get_func_proto = cgroup_dev_func_proto, 1048 .is_valid_access = cgroup_dev_is_valid_access, 1049 }; 1050 1051 /** 1052 * __cgroup_bpf_run_filter_sysctl - Run a program on sysctl 1053 * 1054 * @head: sysctl table header 1055 * @table: sysctl table 1056 * @write: sysctl is being read (= 0) or written (= 1) 1057 * @buf: pointer to buffer passed by user space 1058 * @pcount: value-result argument: value is size of buffer pointed to by @buf, 1059 * result is size of @new_buf if program set new value, initial value 1060 * otherwise 1061 * @ppos: value-result argument: value is position at which read from or write 1062 * to sysctl is happening, result is new position if program overrode it, 1063 * initial value otherwise 1064 * @new_buf: pointer to pointer to new buffer that will be allocated if program 1065 * overrides new value provided by user space on sysctl write 1066 * NOTE: it's caller responsibility to free *new_buf if it was set 1067 * @type: type of program to be executed 1068 * 1069 * Program is run when sysctl is being accessed, either read or written, and 1070 * can allow or deny such access. 1071 * 1072 * This function will return %-EPERM if an attached program is found and 1073 * returned value != 1 during execution. In all other cases 0 is returned. 1074 */ 1075 int __cgroup_bpf_run_filter_sysctl(struct ctl_table_header *head, 1076 struct ctl_table *table, int write, 1077 void __user *buf, size_t *pcount, 1078 loff_t *ppos, void **new_buf, 1079 enum bpf_attach_type type) 1080 { 1081 struct bpf_sysctl_kern ctx = { 1082 .head = head, 1083 .table = table, 1084 .write = write, 1085 .ppos = ppos, 1086 .cur_val = NULL, 1087 .cur_len = PAGE_SIZE, 1088 .new_val = NULL, 1089 .new_len = 0, 1090 .new_updated = 0, 1091 }; 1092 struct cgroup *cgrp; 1093 int ret; 1094 1095 ctx.cur_val = kmalloc_track_caller(ctx.cur_len, GFP_KERNEL); 1096 if (ctx.cur_val) { 1097 mm_segment_t old_fs; 1098 loff_t pos = 0; 1099 1100 old_fs = get_fs(); 1101 set_fs(KERNEL_DS); 1102 if (table->proc_handler(table, 0, (void __user *)ctx.cur_val, 1103 &ctx.cur_len, &pos)) { 1104 /* Let BPF program decide how to proceed. */ 1105 ctx.cur_len = 0; 1106 } 1107 set_fs(old_fs); 1108 } else { 1109 /* Let BPF program decide how to proceed. */ 1110 ctx.cur_len = 0; 1111 } 1112 1113 if (write && buf && *pcount) { 1114 /* BPF program should be able to override new value with a 1115 * buffer bigger than provided by user. 1116 */ 1117 ctx.new_val = kmalloc_track_caller(PAGE_SIZE, GFP_KERNEL); 1118 ctx.new_len = min_t(size_t, PAGE_SIZE, *pcount); 1119 if (!ctx.new_val || 1120 copy_from_user(ctx.new_val, buf, ctx.new_len)) 1121 /* Let BPF program decide how to proceed. */ 1122 ctx.new_len = 0; 1123 } 1124 1125 rcu_read_lock(); 1126 cgrp = task_dfl_cgroup(current); 1127 ret = BPF_PROG_RUN_ARRAY(cgrp->bpf.effective[type], &ctx, BPF_PROG_RUN); 1128 rcu_read_unlock(); 1129 1130 kfree(ctx.cur_val); 1131 1132 if (ret == 1 && ctx.new_updated) { 1133 *new_buf = ctx.new_val; 1134 *pcount = ctx.new_len; 1135 } else { 1136 kfree(ctx.new_val); 1137 } 1138 1139 return ret == 1 ? 0 : -EPERM; 1140 } 1141 EXPORT_SYMBOL(__cgroup_bpf_run_filter_sysctl); 1142 1143 #ifdef CONFIG_NET 1144 static bool __cgroup_bpf_prog_array_is_empty(struct cgroup *cgrp, 1145 enum bpf_attach_type attach_type) 1146 { 1147 struct bpf_prog_array *prog_array; 1148 bool empty; 1149 1150 rcu_read_lock(); 1151 prog_array = rcu_dereference(cgrp->bpf.effective[attach_type]); 1152 empty = bpf_prog_array_is_empty(prog_array); 1153 rcu_read_unlock(); 1154 1155 return empty; 1156 } 1157 1158 static int sockopt_alloc_buf(struct bpf_sockopt_kern *ctx, int max_optlen) 1159 { 1160 if (unlikely(max_optlen > PAGE_SIZE) || max_optlen < 0) 1161 return -EINVAL; 1162 1163 ctx->optval = kzalloc(max_optlen, GFP_USER); 1164 if (!ctx->optval) 1165 return -ENOMEM; 1166 1167 ctx->optval_end = ctx->optval + max_optlen; 1168 1169 return 0; 1170 } 1171 1172 static void sockopt_free_buf(struct bpf_sockopt_kern *ctx) 1173 { 1174 kfree(ctx->optval); 1175 } 1176 1177 int __cgroup_bpf_run_filter_setsockopt(struct sock *sk, int *level, 1178 int *optname, char __user *optval, 1179 int *optlen, char **kernel_optval) 1180 { 1181 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 1182 struct bpf_sockopt_kern ctx = { 1183 .sk = sk, 1184 .level = *level, 1185 .optname = *optname, 1186 }; 1187 int ret, max_optlen; 1188 1189 /* Opportunistic check to see whether we have any BPF program 1190 * attached to the hook so we don't waste time allocating 1191 * memory and locking the socket. 1192 */ 1193 if (!cgroup_bpf_enabled || 1194 __cgroup_bpf_prog_array_is_empty(cgrp, BPF_CGROUP_SETSOCKOPT)) 1195 return 0; 1196 1197 /* Allocate a bit more than the initial user buffer for 1198 * BPF program. The canonical use case is overriding 1199 * TCP_CONGESTION(nv) to TCP_CONGESTION(cubic). 1200 */ 1201 max_optlen = max_t(int, 16, *optlen); 1202 1203 ret = sockopt_alloc_buf(&ctx, max_optlen); 1204 if (ret) 1205 return ret; 1206 1207 ctx.optlen = *optlen; 1208 1209 if (copy_from_user(ctx.optval, optval, *optlen) != 0) { 1210 ret = -EFAULT; 1211 goto out; 1212 } 1213 1214 lock_sock(sk); 1215 ret = BPF_PROG_RUN_ARRAY(cgrp->bpf.effective[BPF_CGROUP_SETSOCKOPT], 1216 &ctx, BPF_PROG_RUN); 1217 release_sock(sk); 1218 1219 if (!ret) { 1220 ret = -EPERM; 1221 goto out; 1222 } 1223 1224 if (ctx.optlen == -1) { 1225 /* optlen set to -1, bypass kernel */ 1226 ret = 1; 1227 } else if (ctx.optlen > max_optlen || ctx.optlen < -1) { 1228 /* optlen is out of bounds */ 1229 ret = -EFAULT; 1230 } else { 1231 /* optlen within bounds, run kernel handler */ 1232 ret = 0; 1233 1234 /* export any potential modifications */ 1235 *level = ctx.level; 1236 *optname = ctx.optname; 1237 *optlen = ctx.optlen; 1238 *kernel_optval = ctx.optval; 1239 } 1240 1241 out: 1242 if (ret) 1243 sockopt_free_buf(&ctx); 1244 return ret; 1245 } 1246 EXPORT_SYMBOL(__cgroup_bpf_run_filter_setsockopt); 1247 1248 int __cgroup_bpf_run_filter_getsockopt(struct sock *sk, int level, 1249 int optname, char __user *optval, 1250 int __user *optlen, int max_optlen, 1251 int retval) 1252 { 1253 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 1254 struct bpf_sockopt_kern ctx = { 1255 .sk = sk, 1256 .level = level, 1257 .optname = optname, 1258 .retval = retval, 1259 }; 1260 int ret; 1261 1262 /* Opportunistic check to see whether we have any BPF program 1263 * attached to the hook so we don't waste time allocating 1264 * memory and locking the socket. 1265 */ 1266 if (!cgroup_bpf_enabled || 1267 __cgroup_bpf_prog_array_is_empty(cgrp, BPF_CGROUP_GETSOCKOPT)) 1268 return retval; 1269 1270 ret = sockopt_alloc_buf(&ctx, max_optlen); 1271 if (ret) 1272 return ret; 1273 1274 ctx.optlen = max_optlen; 1275 1276 if (!retval) { 1277 /* If kernel getsockopt finished successfully, 1278 * copy whatever was returned to the user back 1279 * into our temporary buffer. Set optlen to the 1280 * one that kernel returned as well to let 1281 * BPF programs inspect the value. 1282 */ 1283 1284 if (get_user(ctx.optlen, optlen)) { 1285 ret = -EFAULT; 1286 goto out; 1287 } 1288 1289 if (ctx.optlen > max_optlen) 1290 ctx.optlen = max_optlen; 1291 1292 if (copy_from_user(ctx.optval, optval, ctx.optlen) != 0) { 1293 ret = -EFAULT; 1294 goto out; 1295 } 1296 } 1297 1298 lock_sock(sk); 1299 ret = BPF_PROG_RUN_ARRAY(cgrp->bpf.effective[BPF_CGROUP_GETSOCKOPT], 1300 &ctx, BPF_PROG_RUN); 1301 release_sock(sk); 1302 1303 if (!ret) { 1304 ret = -EPERM; 1305 goto out; 1306 } 1307 1308 if (ctx.optlen > max_optlen) { 1309 ret = -EFAULT; 1310 goto out; 1311 } 1312 1313 /* BPF programs only allowed to set retval to 0, not some 1314 * arbitrary value. 1315 */ 1316 if (ctx.retval != 0 && ctx.retval != retval) { 1317 ret = -EFAULT; 1318 goto out; 1319 } 1320 1321 if (copy_to_user(optval, ctx.optval, ctx.optlen) || 1322 put_user(ctx.optlen, optlen)) { 1323 ret = -EFAULT; 1324 goto out; 1325 } 1326 1327 ret = ctx.retval; 1328 1329 out: 1330 sockopt_free_buf(&ctx); 1331 return ret; 1332 } 1333 EXPORT_SYMBOL(__cgroup_bpf_run_filter_getsockopt); 1334 #endif 1335 1336 static ssize_t sysctl_cpy_dir(const struct ctl_dir *dir, char **bufp, 1337 size_t *lenp) 1338 { 1339 ssize_t tmp_ret = 0, ret; 1340 1341 if (dir->header.parent) { 1342 tmp_ret = sysctl_cpy_dir(dir->header.parent, bufp, lenp); 1343 if (tmp_ret < 0) 1344 return tmp_ret; 1345 } 1346 1347 ret = strscpy(*bufp, dir->header.ctl_table[0].procname, *lenp); 1348 if (ret < 0) 1349 return ret; 1350 *bufp += ret; 1351 *lenp -= ret; 1352 ret += tmp_ret; 1353 1354 /* Avoid leading slash. */ 1355 if (!ret) 1356 return ret; 1357 1358 tmp_ret = strscpy(*bufp, "/", *lenp); 1359 if (tmp_ret < 0) 1360 return tmp_ret; 1361 *bufp += tmp_ret; 1362 *lenp -= tmp_ret; 1363 1364 return ret + tmp_ret; 1365 } 1366 1367 BPF_CALL_4(bpf_sysctl_get_name, struct bpf_sysctl_kern *, ctx, char *, buf, 1368 size_t, buf_len, u64, flags) 1369 { 1370 ssize_t tmp_ret = 0, ret; 1371 1372 if (!buf) 1373 return -EINVAL; 1374 1375 if (!(flags & BPF_F_SYSCTL_BASE_NAME)) { 1376 if (!ctx->head) 1377 return -EINVAL; 1378 tmp_ret = sysctl_cpy_dir(ctx->head->parent, &buf, &buf_len); 1379 if (tmp_ret < 0) 1380 return tmp_ret; 1381 } 1382 1383 ret = strscpy(buf, ctx->table->procname, buf_len); 1384 1385 return ret < 0 ? ret : tmp_ret + ret; 1386 } 1387 1388 static const struct bpf_func_proto bpf_sysctl_get_name_proto = { 1389 .func = bpf_sysctl_get_name, 1390 .gpl_only = false, 1391 .ret_type = RET_INTEGER, 1392 .arg1_type = ARG_PTR_TO_CTX, 1393 .arg2_type = ARG_PTR_TO_MEM, 1394 .arg3_type = ARG_CONST_SIZE, 1395 .arg4_type = ARG_ANYTHING, 1396 }; 1397 1398 static int copy_sysctl_value(char *dst, size_t dst_len, char *src, 1399 size_t src_len) 1400 { 1401 if (!dst) 1402 return -EINVAL; 1403 1404 if (!dst_len) 1405 return -E2BIG; 1406 1407 if (!src || !src_len) { 1408 memset(dst, 0, dst_len); 1409 return -EINVAL; 1410 } 1411 1412 memcpy(dst, src, min(dst_len, src_len)); 1413 1414 if (dst_len > src_len) { 1415 memset(dst + src_len, '\0', dst_len - src_len); 1416 return src_len; 1417 } 1418 1419 dst[dst_len - 1] = '\0'; 1420 1421 return -E2BIG; 1422 } 1423 1424 BPF_CALL_3(bpf_sysctl_get_current_value, struct bpf_sysctl_kern *, ctx, 1425 char *, buf, size_t, buf_len) 1426 { 1427 return copy_sysctl_value(buf, buf_len, ctx->cur_val, ctx->cur_len); 1428 } 1429 1430 static const struct bpf_func_proto bpf_sysctl_get_current_value_proto = { 1431 .func = bpf_sysctl_get_current_value, 1432 .gpl_only = false, 1433 .ret_type = RET_INTEGER, 1434 .arg1_type = ARG_PTR_TO_CTX, 1435 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 1436 .arg3_type = ARG_CONST_SIZE, 1437 }; 1438 1439 BPF_CALL_3(bpf_sysctl_get_new_value, struct bpf_sysctl_kern *, ctx, char *, buf, 1440 size_t, buf_len) 1441 { 1442 if (!ctx->write) { 1443 if (buf && buf_len) 1444 memset(buf, '\0', buf_len); 1445 return -EINVAL; 1446 } 1447 return copy_sysctl_value(buf, buf_len, ctx->new_val, ctx->new_len); 1448 } 1449 1450 static const struct bpf_func_proto bpf_sysctl_get_new_value_proto = { 1451 .func = bpf_sysctl_get_new_value, 1452 .gpl_only = false, 1453 .ret_type = RET_INTEGER, 1454 .arg1_type = ARG_PTR_TO_CTX, 1455 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 1456 .arg3_type = ARG_CONST_SIZE, 1457 }; 1458 1459 BPF_CALL_3(bpf_sysctl_set_new_value, struct bpf_sysctl_kern *, ctx, 1460 const char *, buf, size_t, buf_len) 1461 { 1462 if (!ctx->write || !ctx->new_val || !ctx->new_len || !buf || !buf_len) 1463 return -EINVAL; 1464 1465 if (buf_len > PAGE_SIZE - 1) 1466 return -E2BIG; 1467 1468 memcpy(ctx->new_val, buf, buf_len); 1469 ctx->new_len = buf_len; 1470 ctx->new_updated = 1; 1471 1472 return 0; 1473 } 1474 1475 static const struct bpf_func_proto bpf_sysctl_set_new_value_proto = { 1476 .func = bpf_sysctl_set_new_value, 1477 .gpl_only = false, 1478 .ret_type = RET_INTEGER, 1479 .arg1_type = ARG_PTR_TO_CTX, 1480 .arg2_type = ARG_PTR_TO_MEM, 1481 .arg3_type = ARG_CONST_SIZE, 1482 }; 1483 1484 static const struct bpf_func_proto * 1485 sysctl_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 1486 { 1487 switch (func_id) { 1488 case BPF_FUNC_strtol: 1489 return &bpf_strtol_proto; 1490 case BPF_FUNC_strtoul: 1491 return &bpf_strtoul_proto; 1492 case BPF_FUNC_sysctl_get_name: 1493 return &bpf_sysctl_get_name_proto; 1494 case BPF_FUNC_sysctl_get_current_value: 1495 return &bpf_sysctl_get_current_value_proto; 1496 case BPF_FUNC_sysctl_get_new_value: 1497 return &bpf_sysctl_get_new_value_proto; 1498 case BPF_FUNC_sysctl_set_new_value: 1499 return &bpf_sysctl_set_new_value_proto; 1500 default: 1501 return cgroup_base_func_proto(func_id, prog); 1502 } 1503 } 1504 1505 static bool sysctl_is_valid_access(int off, int size, enum bpf_access_type type, 1506 const struct bpf_prog *prog, 1507 struct bpf_insn_access_aux *info) 1508 { 1509 const int size_default = sizeof(__u32); 1510 1511 if (off < 0 || off + size > sizeof(struct bpf_sysctl) || off % size) 1512 return false; 1513 1514 switch (off) { 1515 case bpf_ctx_range(struct bpf_sysctl, write): 1516 if (type != BPF_READ) 1517 return false; 1518 bpf_ctx_record_field_size(info, size_default); 1519 return bpf_ctx_narrow_access_ok(off, size, size_default); 1520 case bpf_ctx_range(struct bpf_sysctl, file_pos): 1521 if (type == BPF_READ) { 1522 bpf_ctx_record_field_size(info, size_default); 1523 return bpf_ctx_narrow_access_ok(off, size, size_default); 1524 } else { 1525 return size == size_default; 1526 } 1527 default: 1528 return false; 1529 } 1530 } 1531 1532 static u32 sysctl_convert_ctx_access(enum bpf_access_type type, 1533 const struct bpf_insn *si, 1534 struct bpf_insn *insn_buf, 1535 struct bpf_prog *prog, u32 *target_size) 1536 { 1537 struct bpf_insn *insn = insn_buf; 1538 u32 read_size; 1539 1540 switch (si->off) { 1541 case offsetof(struct bpf_sysctl, write): 1542 *insn++ = BPF_LDX_MEM( 1543 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 1544 bpf_target_off(struct bpf_sysctl_kern, write, 1545 sizeof_field(struct bpf_sysctl_kern, 1546 write), 1547 target_size)); 1548 break; 1549 case offsetof(struct bpf_sysctl, file_pos): 1550 /* ppos is a pointer so it should be accessed via indirect 1551 * loads and stores. Also for stores additional temporary 1552 * register is used since neither src_reg nor dst_reg can be 1553 * overridden. 1554 */ 1555 if (type == BPF_WRITE) { 1556 int treg = BPF_REG_9; 1557 1558 if (si->src_reg == treg || si->dst_reg == treg) 1559 --treg; 1560 if (si->src_reg == treg || si->dst_reg == treg) 1561 --treg; 1562 *insn++ = BPF_STX_MEM( 1563 BPF_DW, si->dst_reg, treg, 1564 offsetof(struct bpf_sysctl_kern, tmp_reg)); 1565 *insn++ = BPF_LDX_MEM( 1566 BPF_FIELD_SIZEOF(struct bpf_sysctl_kern, ppos), 1567 treg, si->dst_reg, 1568 offsetof(struct bpf_sysctl_kern, ppos)); 1569 *insn++ = BPF_STX_MEM( 1570 BPF_SIZEOF(u32), treg, si->src_reg, 1571 bpf_ctx_narrow_access_offset( 1572 0, sizeof(u32), sizeof(loff_t))); 1573 *insn++ = BPF_LDX_MEM( 1574 BPF_DW, treg, si->dst_reg, 1575 offsetof(struct bpf_sysctl_kern, tmp_reg)); 1576 } else { 1577 *insn++ = BPF_LDX_MEM( 1578 BPF_FIELD_SIZEOF(struct bpf_sysctl_kern, ppos), 1579 si->dst_reg, si->src_reg, 1580 offsetof(struct bpf_sysctl_kern, ppos)); 1581 read_size = bpf_size_to_bytes(BPF_SIZE(si->code)); 1582 *insn++ = BPF_LDX_MEM( 1583 BPF_SIZE(si->code), si->dst_reg, si->dst_reg, 1584 bpf_ctx_narrow_access_offset( 1585 0, read_size, sizeof(loff_t))); 1586 } 1587 *target_size = sizeof(u32); 1588 break; 1589 } 1590 1591 return insn - insn_buf; 1592 } 1593 1594 const struct bpf_verifier_ops cg_sysctl_verifier_ops = { 1595 .get_func_proto = sysctl_func_proto, 1596 .is_valid_access = sysctl_is_valid_access, 1597 .convert_ctx_access = sysctl_convert_ctx_access, 1598 }; 1599 1600 const struct bpf_prog_ops cg_sysctl_prog_ops = { 1601 }; 1602 1603 static const struct bpf_func_proto * 1604 cg_sockopt_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 1605 { 1606 switch (func_id) { 1607 #ifdef CONFIG_NET 1608 case BPF_FUNC_sk_storage_get: 1609 return &bpf_sk_storage_get_proto; 1610 case BPF_FUNC_sk_storage_delete: 1611 return &bpf_sk_storage_delete_proto; 1612 #endif 1613 #ifdef CONFIG_INET 1614 case BPF_FUNC_tcp_sock: 1615 return &bpf_tcp_sock_proto; 1616 #endif 1617 default: 1618 return cgroup_base_func_proto(func_id, prog); 1619 } 1620 } 1621 1622 static bool cg_sockopt_is_valid_access(int off, int size, 1623 enum bpf_access_type type, 1624 const struct bpf_prog *prog, 1625 struct bpf_insn_access_aux *info) 1626 { 1627 const int size_default = sizeof(__u32); 1628 1629 if (off < 0 || off >= sizeof(struct bpf_sockopt)) 1630 return false; 1631 1632 if (off % size != 0) 1633 return false; 1634 1635 if (type == BPF_WRITE) { 1636 switch (off) { 1637 case offsetof(struct bpf_sockopt, retval): 1638 if (size != size_default) 1639 return false; 1640 return prog->expected_attach_type == 1641 BPF_CGROUP_GETSOCKOPT; 1642 case offsetof(struct bpf_sockopt, optname): 1643 /* fallthrough */ 1644 case offsetof(struct bpf_sockopt, level): 1645 if (size != size_default) 1646 return false; 1647 return prog->expected_attach_type == 1648 BPF_CGROUP_SETSOCKOPT; 1649 case offsetof(struct bpf_sockopt, optlen): 1650 return size == size_default; 1651 default: 1652 return false; 1653 } 1654 } 1655 1656 switch (off) { 1657 case offsetof(struct bpf_sockopt, sk): 1658 if (size != sizeof(__u64)) 1659 return false; 1660 info->reg_type = PTR_TO_SOCKET; 1661 break; 1662 case offsetof(struct bpf_sockopt, optval): 1663 if (size != sizeof(__u64)) 1664 return false; 1665 info->reg_type = PTR_TO_PACKET; 1666 break; 1667 case offsetof(struct bpf_sockopt, optval_end): 1668 if (size != sizeof(__u64)) 1669 return false; 1670 info->reg_type = PTR_TO_PACKET_END; 1671 break; 1672 case offsetof(struct bpf_sockopt, retval): 1673 if (size != size_default) 1674 return false; 1675 return prog->expected_attach_type == BPF_CGROUP_GETSOCKOPT; 1676 default: 1677 if (size != size_default) 1678 return false; 1679 break; 1680 } 1681 return true; 1682 } 1683 1684 #define CG_SOCKOPT_ACCESS_FIELD(T, F) \ 1685 T(BPF_FIELD_SIZEOF(struct bpf_sockopt_kern, F), \ 1686 si->dst_reg, si->src_reg, \ 1687 offsetof(struct bpf_sockopt_kern, F)) 1688 1689 static u32 cg_sockopt_convert_ctx_access(enum bpf_access_type type, 1690 const struct bpf_insn *si, 1691 struct bpf_insn *insn_buf, 1692 struct bpf_prog *prog, 1693 u32 *target_size) 1694 { 1695 struct bpf_insn *insn = insn_buf; 1696 1697 switch (si->off) { 1698 case offsetof(struct bpf_sockopt, sk): 1699 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, sk); 1700 break; 1701 case offsetof(struct bpf_sockopt, level): 1702 if (type == BPF_WRITE) 1703 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_STX_MEM, level); 1704 else 1705 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, level); 1706 break; 1707 case offsetof(struct bpf_sockopt, optname): 1708 if (type == BPF_WRITE) 1709 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_STX_MEM, optname); 1710 else 1711 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, optname); 1712 break; 1713 case offsetof(struct bpf_sockopt, optlen): 1714 if (type == BPF_WRITE) 1715 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_STX_MEM, optlen); 1716 else 1717 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, optlen); 1718 break; 1719 case offsetof(struct bpf_sockopt, retval): 1720 if (type == BPF_WRITE) 1721 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_STX_MEM, retval); 1722 else 1723 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, retval); 1724 break; 1725 case offsetof(struct bpf_sockopt, optval): 1726 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, optval); 1727 break; 1728 case offsetof(struct bpf_sockopt, optval_end): 1729 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, optval_end); 1730 break; 1731 } 1732 1733 return insn - insn_buf; 1734 } 1735 1736 static int cg_sockopt_get_prologue(struct bpf_insn *insn_buf, 1737 bool direct_write, 1738 const struct bpf_prog *prog) 1739 { 1740 /* Nothing to do for sockopt argument. The data is kzalloc'ated. 1741 */ 1742 return 0; 1743 } 1744 1745 const struct bpf_verifier_ops cg_sockopt_verifier_ops = { 1746 .get_func_proto = cg_sockopt_func_proto, 1747 .is_valid_access = cg_sockopt_is_valid_access, 1748 .convert_ctx_access = cg_sockopt_convert_ctx_access, 1749 .gen_prologue = cg_sockopt_get_prologue, 1750 }; 1751 1752 const struct bpf_prog_ops cg_sockopt_prog_ops = { 1753 }; 1754