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 <linux/bpf_lsm.h> 18 #include <linux/bpf_verifier.h> 19 #include <net/sock.h> 20 #include <net/bpf_sk_storage.h> 21 22 #include "../cgroup/cgroup-internal.h" 23 24 DEFINE_STATIC_KEY_ARRAY_FALSE(cgroup_bpf_enabled_key, MAX_CGROUP_BPF_ATTACH_TYPE); 25 EXPORT_SYMBOL(cgroup_bpf_enabled_key); 26 27 /* __always_inline is necessary to prevent indirect call through run_prog 28 * function pointer. 29 */ 30 static __always_inline int 31 bpf_prog_run_array_cg(const struct cgroup_bpf *cgrp, 32 enum cgroup_bpf_attach_type atype, 33 const void *ctx, bpf_prog_run_fn run_prog, 34 int retval, u32 *ret_flags) 35 { 36 const struct bpf_prog_array_item *item; 37 const struct bpf_prog *prog; 38 const struct bpf_prog_array *array; 39 struct bpf_run_ctx *old_run_ctx; 40 struct bpf_cg_run_ctx run_ctx; 41 u32 func_ret; 42 43 run_ctx.retval = retval; 44 migrate_disable(); 45 rcu_read_lock(); 46 array = rcu_dereference(cgrp->effective[atype]); 47 item = &array->items[0]; 48 old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx); 49 while ((prog = READ_ONCE(item->prog))) { 50 run_ctx.prog_item = item; 51 func_ret = run_prog(prog, ctx); 52 if (ret_flags) { 53 *(ret_flags) |= (func_ret >> 1); 54 func_ret &= 1; 55 } 56 if (!func_ret && !IS_ERR_VALUE((long)run_ctx.retval)) 57 run_ctx.retval = -EPERM; 58 item++; 59 } 60 bpf_reset_run_ctx(old_run_ctx); 61 rcu_read_unlock(); 62 migrate_enable(); 63 return run_ctx.retval; 64 } 65 66 unsigned int __cgroup_bpf_run_lsm_sock(const void *ctx, 67 const struct bpf_insn *insn) 68 { 69 const struct bpf_prog *shim_prog; 70 struct sock *sk; 71 struct cgroup *cgrp; 72 int ret = 0; 73 u64 *args; 74 75 args = (u64 *)ctx; 76 sk = (void *)(unsigned long)args[0]; 77 /*shim_prog = container_of(insn, struct bpf_prog, insnsi);*/ 78 shim_prog = (const struct bpf_prog *)((void *)insn - offsetof(struct bpf_prog, insnsi)); 79 80 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 81 if (likely(cgrp)) 82 ret = bpf_prog_run_array_cg(&cgrp->bpf, 83 shim_prog->aux->cgroup_atype, 84 ctx, bpf_prog_run, 0, NULL); 85 return ret; 86 } 87 88 unsigned int __cgroup_bpf_run_lsm_socket(const void *ctx, 89 const struct bpf_insn *insn) 90 { 91 const struct bpf_prog *shim_prog; 92 struct socket *sock; 93 struct cgroup *cgrp; 94 int ret = 0; 95 u64 *args; 96 97 args = (u64 *)ctx; 98 sock = (void *)(unsigned long)args[0]; 99 /*shim_prog = container_of(insn, struct bpf_prog, insnsi);*/ 100 shim_prog = (const struct bpf_prog *)((void *)insn - offsetof(struct bpf_prog, insnsi)); 101 102 cgrp = sock_cgroup_ptr(&sock->sk->sk_cgrp_data); 103 if (likely(cgrp)) 104 ret = bpf_prog_run_array_cg(&cgrp->bpf, 105 shim_prog->aux->cgroup_atype, 106 ctx, bpf_prog_run, 0, NULL); 107 return ret; 108 } 109 110 unsigned int __cgroup_bpf_run_lsm_current(const void *ctx, 111 const struct bpf_insn *insn) 112 { 113 const struct bpf_prog *shim_prog; 114 struct cgroup *cgrp; 115 int ret = 0; 116 117 /*shim_prog = container_of(insn, struct bpf_prog, insnsi);*/ 118 shim_prog = (const struct bpf_prog *)((void *)insn - offsetof(struct bpf_prog, insnsi)); 119 120 /* We rely on trampoline's __bpf_prog_enter_lsm_cgroup to grab RCU read lock. */ 121 cgrp = task_dfl_cgroup(current); 122 if (likely(cgrp)) 123 ret = bpf_prog_run_array_cg(&cgrp->bpf, 124 shim_prog->aux->cgroup_atype, 125 ctx, bpf_prog_run, 0, NULL); 126 return ret; 127 } 128 129 #ifdef CONFIG_BPF_LSM 130 struct cgroup_lsm_atype { 131 u32 attach_btf_id; 132 int refcnt; 133 }; 134 135 static struct cgroup_lsm_atype cgroup_lsm_atype[CGROUP_LSM_NUM]; 136 137 static enum cgroup_bpf_attach_type 138 bpf_cgroup_atype_find(enum bpf_attach_type attach_type, u32 attach_btf_id) 139 { 140 int i; 141 142 lockdep_assert_held(&cgroup_mutex); 143 144 if (attach_type != BPF_LSM_CGROUP) 145 return to_cgroup_bpf_attach_type(attach_type); 146 147 for (i = 0; i < ARRAY_SIZE(cgroup_lsm_atype); i++) 148 if (cgroup_lsm_atype[i].attach_btf_id == attach_btf_id) 149 return CGROUP_LSM_START + i; 150 151 for (i = 0; i < ARRAY_SIZE(cgroup_lsm_atype); i++) 152 if (cgroup_lsm_atype[i].attach_btf_id == 0) 153 return CGROUP_LSM_START + i; 154 155 return -E2BIG; 156 157 } 158 159 void bpf_cgroup_atype_get(u32 attach_btf_id, int cgroup_atype) 160 { 161 int i = cgroup_atype - CGROUP_LSM_START; 162 163 lockdep_assert_held(&cgroup_mutex); 164 165 WARN_ON_ONCE(cgroup_lsm_atype[i].attach_btf_id && 166 cgroup_lsm_atype[i].attach_btf_id != attach_btf_id); 167 168 cgroup_lsm_atype[i].attach_btf_id = attach_btf_id; 169 cgroup_lsm_atype[i].refcnt++; 170 } 171 172 void bpf_cgroup_atype_put(int cgroup_atype) 173 { 174 int i = cgroup_atype - CGROUP_LSM_START; 175 176 mutex_lock(&cgroup_mutex); 177 if (--cgroup_lsm_atype[i].refcnt <= 0) 178 cgroup_lsm_atype[i].attach_btf_id = 0; 179 WARN_ON_ONCE(cgroup_lsm_atype[i].refcnt < 0); 180 mutex_unlock(&cgroup_mutex); 181 } 182 #else 183 static enum cgroup_bpf_attach_type 184 bpf_cgroup_atype_find(enum bpf_attach_type attach_type, u32 attach_btf_id) 185 { 186 if (attach_type != BPF_LSM_CGROUP) 187 return to_cgroup_bpf_attach_type(attach_type); 188 return -EOPNOTSUPP; 189 } 190 #endif /* CONFIG_BPF_LSM */ 191 192 void cgroup_bpf_offline(struct cgroup *cgrp) 193 { 194 cgroup_get(cgrp); 195 percpu_ref_kill(&cgrp->bpf.refcnt); 196 } 197 198 static void bpf_cgroup_storages_free(struct bpf_cgroup_storage *storages[]) 199 { 200 enum bpf_cgroup_storage_type stype; 201 202 for_each_cgroup_storage_type(stype) 203 bpf_cgroup_storage_free(storages[stype]); 204 } 205 206 static int bpf_cgroup_storages_alloc(struct bpf_cgroup_storage *storages[], 207 struct bpf_cgroup_storage *new_storages[], 208 enum bpf_attach_type type, 209 struct bpf_prog *prog, 210 struct cgroup *cgrp) 211 { 212 enum bpf_cgroup_storage_type stype; 213 struct bpf_cgroup_storage_key key; 214 struct bpf_map *map; 215 216 key.cgroup_inode_id = cgroup_id(cgrp); 217 key.attach_type = type; 218 219 for_each_cgroup_storage_type(stype) { 220 map = prog->aux->cgroup_storage[stype]; 221 if (!map) 222 continue; 223 224 storages[stype] = cgroup_storage_lookup((void *)map, &key, false); 225 if (storages[stype]) 226 continue; 227 228 storages[stype] = bpf_cgroup_storage_alloc(prog, stype); 229 if (IS_ERR(storages[stype])) { 230 bpf_cgroup_storages_free(new_storages); 231 return -ENOMEM; 232 } 233 234 new_storages[stype] = storages[stype]; 235 } 236 237 return 0; 238 } 239 240 static void bpf_cgroup_storages_assign(struct bpf_cgroup_storage *dst[], 241 struct bpf_cgroup_storage *src[]) 242 { 243 enum bpf_cgroup_storage_type stype; 244 245 for_each_cgroup_storage_type(stype) 246 dst[stype] = src[stype]; 247 } 248 249 static void bpf_cgroup_storages_link(struct bpf_cgroup_storage *storages[], 250 struct cgroup *cgrp, 251 enum bpf_attach_type attach_type) 252 { 253 enum bpf_cgroup_storage_type stype; 254 255 for_each_cgroup_storage_type(stype) 256 bpf_cgroup_storage_link(storages[stype], cgrp, attach_type); 257 } 258 259 /* Called when bpf_cgroup_link is auto-detached from dying cgroup. 260 * It drops cgroup and bpf_prog refcounts, and marks bpf_link as defunct. It 261 * doesn't free link memory, which will eventually be done by bpf_link's 262 * release() callback, when its last FD is closed. 263 */ 264 static void bpf_cgroup_link_auto_detach(struct bpf_cgroup_link *link) 265 { 266 cgroup_put(link->cgroup); 267 link->cgroup = NULL; 268 } 269 270 /** 271 * cgroup_bpf_release() - put references of all bpf programs and 272 * release all cgroup bpf data 273 * @work: work structure embedded into the cgroup to modify 274 */ 275 static void cgroup_bpf_release(struct work_struct *work) 276 { 277 struct cgroup *p, *cgrp = container_of(work, struct cgroup, 278 bpf.release_work); 279 struct bpf_prog_array *old_array; 280 struct list_head *storages = &cgrp->bpf.storages; 281 struct bpf_cgroup_storage *storage, *stmp; 282 283 unsigned int atype; 284 285 mutex_lock(&cgroup_mutex); 286 287 for (atype = 0; atype < ARRAY_SIZE(cgrp->bpf.progs); atype++) { 288 struct hlist_head *progs = &cgrp->bpf.progs[atype]; 289 struct bpf_prog_list *pl; 290 struct hlist_node *pltmp; 291 292 hlist_for_each_entry_safe(pl, pltmp, progs, node) { 293 hlist_del(&pl->node); 294 if (pl->prog) { 295 if (pl->prog->expected_attach_type == BPF_LSM_CGROUP) 296 bpf_trampoline_unlink_cgroup_shim(pl->prog); 297 bpf_prog_put(pl->prog); 298 } 299 if (pl->link) { 300 if (pl->link->link.prog->expected_attach_type == BPF_LSM_CGROUP) 301 bpf_trampoline_unlink_cgroup_shim(pl->link->link.prog); 302 bpf_cgroup_link_auto_detach(pl->link); 303 } 304 kfree(pl); 305 static_branch_dec(&cgroup_bpf_enabled_key[atype]); 306 } 307 old_array = rcu_dereference_protected( 308 cgrp->bpf.effective[atype], 309 lockdep_is_held(&cgroup_mutex)); 310 bpf_prog_array_free(old_array); 311 } 312 313 list_for_each_entry_safe(storage, stmp, storages, list_cg) { 314 bpf_cgroup_storage_unlink(storage); 315 bpf_cgroup_storage_free(storage); 316 } 317 318 mutex_unlock(&cgroup_mutex); 319 320 for (p = cgroup_parent(cgrp); p; p = cgroup_parent(p)) 321 cgroup_bpf_put(p); 322 323 percpu_ref_exit(&cgrp->bpf.refcnt); 324 cgroup_put(cgrp); 325 } 326 327 /** 328 * cgroup_bpf_release_fn() - callback used to schedule releasing 329 * of bpf cgroup data 330 * @ref: percpu ref counter structure 331 */ 332 static void cgroup_bpf_release_fn(struct percpu_ref *ref) 333 { 334 struct cgroup *cgrp = container_of(ref, struct cgroup, bpf.refcnt); 335 336 INIT_WORK(&cgrp->bpf.release_work, cgroup_bpf_release); 337 queue_work(system_wq, &cgrp->bpf.release_work); 338 } 339 340 /* Get underlying bpf_prog of bpf_prog_list entry, regardless if it's through 341 * link or direct prog. 342 */ 343 static struct bpf_prog *prog_list_prog(struct bpf_prog_list *pl) 344 { 345 if (pl->prog) 346 return pl->prog; 347 if (pl->link) 348 return pl->link->link.prog; 349 return NULL; 350 } 351 352 /* count number of elements in the list. 353 * it's slow but the list cannot be long 354 */ 355 static u32 prog_list_length(struct hlist_head *head) 356 { 357 struct bpf_prog_list *pl; 358 u32 cnt = 0; 359 360 hlist_for_each_entry(pl, head, node) { 361 if (!prog_list_prog(pl)) 362 continue; 363 cnt++; 364 } 365 return cnt; 366 } 367 368 /* if parent has non-overridable prog attached, 369 * disallow attaching new programs to the descendent cgroup. 370 * if parent has overridable or multi-prog, allow attaching 371 */ 372 static bool hierarchy_allows_attach(struct cgroup *cgrp, 373 enum cgroup_bpf_attach_type atype) 374 { 375 struct cgroup *p; 376 377 p = cgroup_parent(cgrp); 378 if (!p) 379 return true; 380 do { 381 u32 flags = p->bpf.flags[atype]; 382 u32 cnt; 383 384 if (flags & BPF_F_ALLOW_MULTI) 385 return true; 386 cnt = prog_list_length(&p->bpf.progs[atype]); 387 WARN_ON_ONCE(cnt > 1); 388 if (cnt == 1) 389 return !!(flags & BPF_F_ALLOW_OVERRIDE); 390 p = cgroup_parent(p); 391 } while (p); 392 return true; 393 } 394 395 /* compute a chain of effective programs for a given cgroup: 396 * start from the list of programs in this cgroup and add 397 * all parent programs. 398 * Note that parent's F_ALLOW_OVERRIDE-type program is yielding 399 * to programs in this cgroup 400 */ 401 static int compute_effective_progs(struct cgroup *cgrp, 402 enum cgroup_bpf_attach_type atype, 403 struct bpf_prog_array **array) 404 { 405 struct bpf_prog_array_item *item; 406 struct bpf_prog_array *progs; 407 struct bpf_prog_list *pl; 408 struct cgroup *p = cgrp; 409 int cnt = 0; 410 411 /* count number of effective programs by walking parents */ 412 do { 413 if (cnt == 0 || (p->bpf.flags[atype] & BPF_F_ALLOW_MULTI)) 414 cnt += prog_list_length(&p->bpf.progs[atype]); 415 p = cgroup_parent(p); 416 } while (p); 417 418 progs = bpf_prog_array_alloc(cnt, GFP_KERNEL); 419 if (!progs) 420 return -ENOMEM; 421 422 /* populate the array with effective progs */ 423 cnt = 0; 424 p = cgrp; 425 do { 426 if (cnt > 0 && !(p->bpf.flags[atype] & BPF_F_ALLOW_MULTI)) 427 continue; 428 429 hlist_for_each_entry(pl, &p->bpf.progs[atype], node) { 430 if (!prog_list_prog(pl)) 431 continue; 432 433 item = &progs->items[cnt]; 434 item->prog = prog_list_prog(pl); 435 bpf_cgroup_storages_assign(item->cgroup_storage, 436 pl->storage); 437 cnt++; 438 } 439 } while ((p = cgroup_parent(p))); 440 441 *array = progs; 442 return 0; 443 } 444 445 static void activate_effective_progs(struct cgroup *cgrp, 446 enum cgroup_bpf_attach_type atype, 447 struct bpf_prog_array *old_array) 448 { 449 old_array = rcu_replace_pointer(cgrp->bpf.effective[atype], old_array, 450 lockdep_is_held(&cgroup_mutex)); 451 /* free prog array after grace period, since __cgroup_bpf_run_*() 452 * might be still walking the array 453 */ 454 bpf_prog_array_free(old_array); 455 } 456 457 /** 458 * cgroup_bpf_inherit() - inherit effective programs from parent 459 * @cgrp: the cgroup to modify 460 */ 461 int cgroup_bpf_inherit(struct cgroup *cgrp) 462 { 463 /* has to use marco instead of const int, since compiler thinks 464 * that array below is variable length 465 */ 466 #define NR ARRAY_SIZE(cgrp->bpf.effective) 467 struct bpf_prog_array *arrays[NR] = {}; 468 struct cgroup *p; 469 int ret, i; 470 471 ret = percpu_ref_init(&cgrp->bpf.refcnt, cgroup_bpf_release_fn, 0, 472 GFP_KERNEL); 473 if (ret) 474 return ret; 475 476 for (p = cgroup_parent(cgrp); p; p = cgroup_parent(p)) 477 cgroup_bpf_get(p); 478 479 for (i = 0; i < NR; i++) 480 INIT_HLIST_HEAD(&cgrp->bpf.progs[i]); 481 482 INIT_LIST_HEAD(&cgrp->bpf.storages); 483 484 for (i = 0; i < NR; i++) 485 if (compute_effective_progs(cgrp, i, &arrays[i])) 486 goto cleanup; 487 488 for (i = 0; i < NR; i++) 489 activate_effective_progs(cgrp, i, arrays[i]); 490 491 return 0; 492 cleanup: 493 for (i = 0; i < NR; i++) 494 bpf_prog_array_free(arrays[i]); 495 496 for (p = cgroup_parent(cgrp); p; p = cgroup_parent(p)) 497 cgroup_bpf_put(p); 498 499 percpu_ref_exit(&cgrp->bpf.refcnt); 500 501 return -ENOMEM; 502 } 503 504 static int update_effective_progs(struct cgroup *cgrp, 505 enum cgroup_bpf_attach_type atype) 506 { 507 struct cgroup_subsys_state *css; 508 int err; 509 510 /* allocate and recompute effective prog arrays */ 511 css_for_each_descendant_pre(css, &cgrp->self) { 512 struct cgroup *desc = container_of(css, struct cgroup, self); 513 514 if (percpu_ref_is_zero(&desc->bpf.refcnt)) 515 continue; 516 517 err = compute_effective_progs(desc, atype, &desc->bpf.inactive); 518 if (err) 519 goto cleanup; 520 } 521 522 /* all allocations were successful. Activate all prog arrays */ 523 css_for_each_descendant_pre(css, &cgrp->self) { 524 struct cgroup *desc = container_of(css, struct cgroup, self); 525 526 if (percpu_ref_is_zero(&desc->bpf.refcnt)) { 527 if (unlikely(desc->bpf.inactive)) { 528 bpf_prog_array_free(desc->bpf.inactive); 529 desc->bpf.inactive = NULL; 530 } 531 continue; 532 } 533 534 activate_effective_progs(desc, atype, desc->bpf.inactive); 535 desc->bpf.inactive = NULL; 536 } 537 538 return 0; 539 540 cleanup: 541 /* oom while computing effective. Free all computed effective arrays 542 * since they were not activated 543 */ 544 css_for_each_descendant_pre(css, &cgrp->self) { 545 struct cgroup *desc = container_of(css, struct cgroup, self); 546 547 bpf_prog_array_free(desc->bpf.inactive); 548 desc->bpf.inactive = NULL; 549 } 550 551 return err; 552 } 553 554 #define BPF_CGROUP_MAX_PROGS 64 555 556 static struct bpf_prog_list *find_attach_entry(struct hlist_head *progs, 557 struct bpf_prog *prog, 558 struct bpf_cgroup_link *link, 559 struct bpf_prog *replace_prog, 560 bool allow_multi) 561 { 562 struct bpf_prog_list *pl; 563 564 /* single-attach case */ 565 if (!allow_multi) { 566 if (hlist_empty(progs)) 567 return NULL; 568 return hlist_entry(progs->first, typeof(*pl), node); 569 } 570 571 hlist_for_each_entry(pl, progs, node) { 572 if (prog && pl->prog == prog && prog != replace_prog) 573 /* disallow attaching the same prog twice */ 574 return ERR_PTR(-EINVAL); 575 if (link && pl->link == link) 576 /* disallow attaching the same link twice */ 577 return ERR_PTR(-EINVAL); 578 } 579 580 /* direct prog multi-attach w/ replacement case */ 581 if (replace_prog) { 582 hlist_for_each_entry(pl, progs, node) { 583 if (pl->prog == replace_prog) 584 /* a match found */ 585 return pl; 586 } 587 /* prog to replace not found for cgroup */ 588 return ERR_PTR(-ENOENT); 589 } 590 591 return NULL; 592 } 593 594 /** 595 * __cgroup_bpf_attach() - Attach the program or the link to a cgroup, and 596 * propagate the change to descendants 597 * @cgrp: The cgroup which descendants to traverse 598 * @prog: A program to attach 599 * @link: A link to attach 600 * @replace_prog: Previously attached program to replace if BPF_F_REPLACE is set 601 * @type: Type of attach operation 602 * @flags: Option flags 603 * 604 * Exactly one of @prog or @link can be non-null. 605 * Must be called with cgroup_mutex held. 606 */ 607 static int __cgroup_bpf_attach(struct cgroup *cgrp, 608 struct bpf_prog *prog, struct bpf_prog *replace_prog, 609 struct bpf_cgroup_link *link, 610 enum bpf_attach_type type, u32 flags) 611 { 612 u32 saved_flags = (flags & (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)); 613 struct bpf_prog *old_prog = NULL; 614 struct bpf_cgroup_storage *storage[MAX_BPF_CGROUP_STORAGE_TYPE] = {}; 615 struct bpf_cgroup_storage *new_storage[MAX_BPF_CGROUP_STORAGE_TYPE] = {}; 616 struct bpf_prog *new_prog = prog ? : link->link.prog; 617 enum cgroup_bpf_attach_type atype; 618 struct bpf_prog_list *pl; 619 struct hlist_head *progs; 620 int err; 621 622 if (((flags & BPF_F_ALLOW_OVERRIDE) && (flags & BPF_F_ALLOW_MULTI)) || 623 ((flags & BPF_F_REPLACE) && !(flags & BPF_F_ALLOW_MULTI))) 624 /* invalid combination */ 625 return -EINVAL; 626 if (link && (prog || replace_prog)) 627 /* only either link or prog/replace_prog can be specified */ 628 return -EINVAL; 629 if (!!replace_prog != !!(flags & BPF_F_REPLACE)) 630 /* replace_prog implies BPF_F_REPLACE, and vice versa */ 631 return -EINVAL; 632 633 atype = bpf_cgroup_atype_find(type, new_prog->aux->attach_btf_id); 634 if (atype < 0) 635 return -EINVAL; 636 637 progs = &cgrp->bpf.progs[atype]; 638 639 if (!hierarchy_allows_attach(cgrp, atype)) 640 return -EPERM; 641 642 if (!hlist_empty(progs) && cgrp->bpf.flags[atype] != saved_flags) 643 /* Disallow attaching non-overridable on top 644 * of existing overridable in this cgroup. 645 * Disallow attaching multi-prog if overridable or none 646 */ 647 return -EPERM; 648 649 if (prog_list_length(progs) >= BPF_CGROUP_MAX_PROGS) 650 return -E2BIG; 651 652 pl = find_attach_entry(progs, prog, link, replace_prog, 653 flags & BPF_F_ALLOW_MULTI); 654 if (IS_ERR(pl)) 655 return PTR_ERR(pl); 656 657 if (bpf_cgroup_storages_alloc(storage, new_storage, type, 658 prog ? : link->link.prog, cgrp)) 659 return -ENOMEM; 660 661 if (pl) { 662 old_prog = pl->prog; 663 } else { 664 struct hlist_node *last = NULL; 665 666 pl = kmalloc(sizeof(*pl), GFP_KERNEL); 667 if (!pl) { 668 bpf_cgroup_storages_free(new_storage); 669 return -ENOMEM; 670 } 671 if (hlist_empty(progs)) 672 hlist_add_head(&pl->node, progs); 673 else 674 hlist_for_each(last, progs) { 675 if (last->next) 676 continue; 677 hlist_add_behind(&pl->node, last); 678 break; 679 } 680 } 681 682 pl->prog = prog; 683 pl->link = link; 684 bpf_cgroup_storages_assign(pl->storage, storage); 685 cgrp->bpf.flags[atype] = saved_flags; 686 687 if (type == BPF_LSM_CGROUP) { 688 err = bpf_trampoline_link_cgroup_shim(new_prog, atype); 689 if (err) 690 goto cleanup; 691 } 692 693 err = update_effective_progs(cgrp, atype); 694 if (err) 695 goto cleanup_trampoline; 696 697 if (old_prog) { 698 if (type == BPF_LSM_CGROUP) 699 bpf_trampoline_unlink_cgroup_shim(old_prog); 700 bpf_prog_put(old_prog); 701 } else { 702 static_branch_inc(&cgroup_bpf_enabled_key[atype]); 703 } 704 bpf_cgroup_storages_link(new_storage, cgrp, type); 705 return 0; 706 707 cleanup_trampoline: 708 if (type == BPF_LSM_CGROUP) 709 bpf_trampoline_unlink_cgroup_shim(new_prog); 710 711 cleanup: 712 if (old_prog) { 713 pl->prog = old_prog; 714 pl->link = NULL; 715 } 716 bpf_cgroup_storages_free(new_storage); 717 if (!old_prog) { 718 hlist_del(&pl->node); 719 kfree(pl); 720 } 721 return err; 722 } 723 724 static int cgroup_bpf_attach(struct cgroup *cgrp, 725 struct bpf_prog *prog, struct bpf_prog *replace_prog, 726 struct bpf_cgroup_link *link, 727 enum bpf_attach_type type, 728 u32 flags) 729 { 730 int ret; 731 732 mutex_lock(&cgroup_mutex); 733 ret = __cgroup_bpf_attach(cgrp, prog, replace_prog, link, type, flags); 734 mutex_unlock(&cgroup_mutex); 735 return ret; 736 } 737 738 /* Swap updated BPF program for given link in effective program arrays across 739 * all descendant cgroups. This function is guaranteed to succeed. 740 */ 741 static void replace_effective_prog(struct cgroup *cgrp, 742 enum cgroup_bpf_attach_type atype, 743 struct bpf_cgroup_link *link) 744 { 745 struct bpf_prog_array_item *item; 746 struct cgroup_subsys_state *css; 747 struct bpf_prog_array *progs; 748 struct bpf_prog_list *pl; 749 struct hlist_head *head; 750 struct cgroup *cg; 751 int pos; 752 753 css_for_each_descendant_pre(css, &cgrp->self) { 754 struct cgroup *desc = container_of(css, struct cgroup, self); 755 756 if (percpu_ref_is_zero(&desc->bpf.refcnt)) 757 continue; 758 759 /* find position of link in effective progs array */ 760 for (pos = 0, cg = desc; cg; cg = cgroup_parent(cg)) { 761 if (pos && !(cg->bpf.flags[atype] & BPF_F_ALLOW_MULTI)) 762 continue; 763 764 head = &cg->bpf.progs[atype]; 765 hlist_for_each_entry(pl, head, node) { 766 if (!prog_list_prog(pl)) 767 continue; 768 if (pl->link == link) 769 goto found; 770 pos++; 771 } 772 } 773 found: 774 BUG_ON(!cg); 775 progs = rcu_dereference_protected( 776 desc->bpf.effective[atype], 777 lockdep_is_held(&cgroup_mutex)); 778 item = &progs->items[pos]; 779 WRITE_ONCE(item->prog, link->link.prog); 780 } 781 } 782 783 /** 784 * __cgroup_bpf_replace() - Replace link's program and propagate the change 785 * to descendants 786 * @cgrp: The cgroup which descendants to traverse 787 * @link: A link for which to replace BPF program 788 * @type: Type of attach operation 789 * 790 * Must be called with cgroup_mutex held. 791 */ 792 static int __cgroup_bpf_replace(struct cgroup *cgrp, 793 struct bpf_cgroup_link *link, 794 struct bpf_prog *new_prog) 795 { 796 enum cgroup_bpf_attach_type atype; 797 struct bpf_prog *old_prog; 798 struct bpf_prog_list *pl; 799 struct hlist_head *progs; 800 bool found = false; 801 802 atype = bpf_cgroup_atype_find(link->type, new_prog->aux->attach_btf_id); 803 if (atype < 0) 804 return -EINVAL; 805 806 progs = &cgrp->bpf.progs[atype]; 807 808 if (link->link.prog->type != new_prog->type) 809 return -EINVAL; 810 811 hlist_for_each_entry(pl, progs, node) { 812 if (pl->link == link) { 813 found = true; 814 break; 815 } 816 } 817 if (!found) 818 return -ENOENT; 819 820 old_prog = xchg(&link->link.prog, new_prog); 821 replace_effective_prog(cgrp, atype, link); 822 bpf_prog_put(old_prog); 823 return 0; 824 } 825 826 static int cgroup_bpf_replace(struct bpf_link *link, struct bpf_prog *new_prog, 827 struct bpf_prog *old_prog) 828 { 829 struct bpf_cgroup_link *cg_link; 830 int ret; 831 832 cg_link = container_of(link, struct bpf_cgroup_link, link); 833 834 mutex_lock(&cgroup_mutex); 835 /* link might have been auto-released by dying cgroup, so fail */ 836 if (!cg_link->cgroup) { 837 ret = -ENOLINK; 838 goto out_unlock; 839 } 840 if (old_prog && link->prog != old_prog) { 841 ret = -EPERM; 842 goto out_unlock; 843 } 844 ret = __cgroup_bpf_replace(cg_link->cgroup, cg_link, new_prog); 845 out_unlock: 846 mutex_unlock(&cgroup_mutex); 847 return ret; 848 } 849 850 static struct bpf_prog_list *find_detach_entry(struct hlist_head *progs, 851 struct bpf_prog *prog, 852 struct bpf_cgroup_link *link, 853 bool allow_multi) 854 { 855 struct bpf_prog_list *pl; 856 857 if (!allow_multi) { 858 if (hlist_empty(progs)) 859 /* report error when trying to detach and nothing is attached */ 860 return ERR_PTR(-ENOENT); 861 862 /* to maintain backward compatibility NONE and OVERRIDE cgroups 863 * allow detaching with invalid FD (prog==NULL) in legacy mode 864 */ 865 return hlist_entry(progs->first, typeof(*pl), node); 866 } 867 868 if (!prog && !link) 869 /* to detach MULTI prog the user has to specify valid FD 870 * of the program or link to be detached 871 */ 872 return ERR_PTR(-EINVAL); 873 874 /* find the prog or link and detach it */ 875 hlist_for_each_entry(pl, progs, node) { 876 if (pl->prog == prog && pl->link == link) 877 return pl; 878 } 879 return ERR_PTR(-ENOENT); 880 } 881 882 /** 883 * purge_effective_progs() - After compute_effective_progs fails to alloc new 884 * cgrp->bpf.inactive table we can recover by 885 * recomputing the array in place. 886 * 887 * @cgrp: The cgroup which descendants to travers 888 * @prog: A program to detach or NULL 889 * @link: A link to detach or NULL 890 * @atype: Type of detach operation 891 */ 892 static void purge_effective_progs(struct cgroup *cgrp, struct bpf_prog *prog, 893 struct bpf_cgroup_link *link, 894 enum cgroup_bpf_attach_type atype) 895 { 896 struct cgroup_subsys_state *css; 897 struct bpf_prog_array *progs; 898 struct bpf_prog_list *pl; 899 struct hlist_head *head; 900 struct cgroup *cg; 901 int pos; 902 903 /* recompute effective prog array in place */ 904 css_for_each_descendant_pre(css, &cgrp->self) { 905 struct cgroup *desc = container_of(css, struct cgroup, self); 906 907 if (percpu_ref_is_zero(&desc->bpf.refcnt)) 908 continue; 909 910 /* find position of link or prog in effective progs array */ 911 for (pos = 0, cg = desc; cg; cg = cgroup_parent(cg)) { 912 if (pos && !(cg->bpf.flags[atype] & BPF_F_ALLOW_MULTI)) 913 continue; 914 915 head = &cg->bpf.progs[atype]; 916 hlist_for_each_entry(pl, head, node) { 917 if (!prog_list_prog(pl)) 918 continue; 919 if (pl->prog == prog && pl->link == link) 920 goto found; 921 pos++; 922 } 923 } 924 925 /* no link or prog match, skip the cgroup of this layer */ 926 continue; 927 found: 928 progs = rcu_dereference_protected( 929 desc->bpf.effective[atype], 930 lockdep_is_held(&cgroup_mutex)); 931 932 /* Remove the program from the array */ 933 WARN_ONCE(bpf_prog_array_delete_safe_at(progs, pos), 934 "Failed to purge a prog from array at index %d", pos); 935 } 936 } 937 938 /** 939 * __cgroup_bpf_detach() - Detach the program or link from a cgroup, and 940 * propagate the change to descendants 941 * @cgrp: The cgroup which descendants to traverse 942 * @prog: A program to detach or NULL 943 * @link: A link to detach or NULL 944 * @type: Type of detach operation 945 * 946 * At most one of @prog or @link can be non-NULL. 947 * Must be called with cgroup_mutex held. 948 */ 949 static int __cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog, 950 struct bpf_cgroup_link *link, enum bpf_attach_type type) 951 { 952 enum cgroup_bpf_attach_type atype; 953 struct bpf_prog *old_prog; 954 struct bpf_prog_list *pl; 955 struct hlist_head *progs; 956 u32 attach_btf_id = 0; 957 u32 flags; 958 959 if (prog) 960 attach_btf_id = prog->aux->attach_btf_id; 961 if (link) 962 attach_btf_id = link->link.prog->aux->attach_btf_id; 963 964 atype = bpf_cgroup_atype_find(type, attach_btf_id); 965 if (atype < 0) 966 return -EINVAL; 967 968 progs = &cgrp->bpf.progs[atype]; 969 flags = cgrp->bpf.flags[atype]; 970 971 if (prog && link) 972 /* only one of prog or link can be specified */ 973 return -EINVAL; 974 975 pl = find_detach_entry(progs, prog, link, flags & BPF_F_ALLOW_MULTI); 976 if (IS_ERR(pl)) 977 return PTR_ERR(pl); 978 979 /* mark it deleted, so it's ignored while recomputing effective */ 980 old_prog = pl->prog; 981 pl->prog = NULL; 982 pl->link = NULL; 983 984 if (update_effective_progs(cgrp, atype)) { 985 /* if update effective array failed replace the prog with a dummy prog*/ 986 pl->prog = old_prog; 987 pl->link = link; 988 purge_effective_progs(cgrp, old_prog, link, atype); 989 } 990 991 /* now can actually delete it from this cgroup list */ 992 hlist_del(&pl->node); 993 994 kfree(pl); 995 if (hlist_empty(progs)) 996 /* last program was detached, reset flags to zero */ 997 cgrp->bpf.flags[atype] = 0; 998 if (old_prog) { 999 if (type == BPF_LSM_CGROUP) 1000 bpf_trampoline_unlink_cgroup_shim(old_prog); 1001 bpf_prog_put(old_prog); 1002 } 1003 static_branch_dec(&cgroup_bpf_enabled_key[atype]); 1004 return 0; 1005 } 1006 1007 static int cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog, 1008 enum bpf_attach_type type) 1009 { 1010 int ret; 1011 1012 mutex_lock(&cgroup_mutex); 1013 ret = __cgroup_bpf_detach(cgrp, prog, NULL, type); 1014 mutex_unlock(&cgroup_mutex); 1015 return ret; 1016 } 1017 1018 /* Must be called with cgroup_mutex held to avoid races. */ 1019 static int __cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr, 1020 union bpf_attr __user *uattr) 1021 { 1022 __u32 __user *prog_attach_flags = u64_to_user_ptr(attr->query.prog_attach_flags); 1023 __u32 __user *prog_ids = u64_to_user_ptr(attr->query.prog_ids); 1024 enum bpf_attach_type type = attr->query.attach_type; 1025 enum cgroup_bpf_attach_type from_atype, to_atype; 1026 enum cgroup_bpf_attach_type atype; 1027 struct bpf_prog_array *effective; 1028 int cnt, ret = 0, i; 1029 int total_cnt = 0; 1030 u32 flags; 1031 1032 if (type == BPF_LSM_CGROUP) { 1033 if (attr->query.prog_cnt && prog_ids && !prog_attach_flags) 1034 return -EINVAL; 1035 1036 from_atype = CGROUP_LSM_START; 1037 to_atype = CGROUP_LSM_END; 1038 flags = 0; 1039 } else { 1040 from_atype = to_cgroup_bpf_attach_type(type); 1041 if (from_atype < 0) 1042 return -EINVAL; 1043 to_atype = from_atype; 1044 flags = cgrp->bpf.flags[from_atype]; 1045 } 1046 1047 for (atype = from_atype; atype <= to_atype; atype++) { 1048 if (attr->query.query_flags & BPF_F_QUERY_EFFECTIVE) { 1049 effective = rcu_dereference_protected(cgrp->bpf.effective[atype], 1050 lockdep_is_held(&cgroup_mutex)); 1051 total_cnt += bpf_prog_array_length(effective); 1052 } else { 1053 total_cnt += prog_list_length(&cgrp->bpf.progs[atype]); 1054 } 1055 } 1056 1057 if (copy_to_user(&uattr->query.attach_flags, &flags, sizeof(flags))) 1058 return -EFAULT; 1059 if (copy_to_user(&uattr->query.prog_cnt, &total_cnt, sizeof(total_cnt))) 1060 return -EFAULT; 1061 if (attr->query.prog_cnt == 0 || !prog_ids || !total_cnt) 1062 /* return early if user requested only program count + flags */ 1063 return 0; 1064 1065 if (attr->query.prog_cnt < total_cnt) { 1066 total_cnt = attr->query.prog_cnt; 1067 ret = -ENOSPC; 1068 } 1069 1070 for (atype = from_atype; atype <= to_atype && total_cnt; atype++) { 1071 if (attr->query.query_flags & BPF_F_QUERY_EFFECTIVE) { 1072 effective = rcu_dereference_protected(cgrp->bpf.effective[atype], 1073 lockdep_is_held(&cgroup_mutex)); 1074 cnt = min_t(int, bpf_prog_array_length(effective), total_cnt); 1075 ret = bpf_prog_array_copy_to_user(effective, prog_ids, cnt); 1076 } else { 1077 struct hlist_head *progs; 1078 struct bpf_prog_list *pl; 1079 struct bpf_prog *prog; 1080 u32 id; 1081 1082 progs = &cgrp->bpf.progs[atype]; 1083 cnt = min_t(int, prog_list_length(progs), total_cnt); 1084 i = 0; 1085 hlist_for_each_entry(pl, progs, node) { 1086 prog = prog_list_prog(pl); 1087 id = prog->aux->id; 1088 if (copy_to_user(prog_ids + i, &id, sizeof(id))) 1089 return -EFAULT; 1090 if (++i == cnt) 1091 break; 1092 } 1093 } 1094 1095 if (prog_attach_flags) { 1096 flags = cgrp->bpf.flags[atype]; 1097 1098 for (i = 0; i < cnt; i++) 1099 if (copy_to_user(prog_attach_flags + i, &flags, sizeof(flags))) 1100 return -EFAULT; 1101 prog_attach_flags += cnt; 1102 } 1103 1104 prog_ids += cnt; 1105 total_cnt -= cnt; 1106 } 1107 return ret; 1108 } 1109 1110 static int cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr, 1111 union bpf_attr __user *uattr) 1112 { 1113 int ret; 1114 1115 mutex_lock(&cgroup_mutex); 1116 ret = __cgroup_bpf_query(cgrp, attr, uattr); 1117 mutex_unlock(&cgroup_mutex); 1118 return ret; 1119 } 1120 1121 int cgroup_bpf_prog_attach(const union bpf_attr *attr, 1122 enum bpf_prog_type ptype, struct bpf_prog *prog) 1123 { 1124 struct bpf_prog *replace_prog = NULL; 1125 struct cgroup *cgrp; 1126 int ret; 1127 1128 cgrp = cgroup_get_from_fd(attr->target_fd); 1129 if (IS_ERR(cgrp)) 1130 return PTR_ERR(cgrp); 1131 1132 if ((attr->attach_flags & BPF_F_ALLOW_MULTI) && 1133 (attr->attach_flags & BPF_F_REPLACE)) { 1134 replace_prog = bpf_prog_get_type(attr->replace_bpf_fd, ptype); 1135 if (IS_ERR(replace_prog)) { 1136 cgroup_put(cgrp); 1137 return PTR_ERR(replace_prog); 1138 } 1139 } 1140 1141 ret = cgroup_bpf_attach(cgrp, prog, replace_prog, NULL, 1142 attr->attach_type, attr->attach_flags); 1143 1144 if (replace_prog) 1145 bpf_prog_put(replace_prog); 1146 cgroup_put(cgrp); 1147 return ret; 1148 } 1149 1150 int cgroup_bpf_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype) 1151 { 1152 struct bpf_prog *prog; 1153 struct cgroup *cgrp; 1154 int ret; 1155 1156 cgrp = cgroup_get_from_fd(attr->target_fd); 1157 if (IS_ERR(cgrp)) 1158 return PTR_ERR(cgrp); 1159 1160 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype); 1161 if (IS_ERR(prog)) 1162 prog = NULL; 1163 1164 ret = cgroup_bpf_detach(cgrp, prog, attr->attach_type); 1165 if (prog) 1166 bpf_prog_put(prog); 1167 1168 cgroup_put(cgrp); 1169 return ret; 1170 } 1171 1172 static void bpf_cgroup_link_release(struct bpf_link *link) 1173 { 1174 struct bpf_cgroup_link *cg_link = 1175 container_of(link, struct bpf_cgroup_link, link); 1176 struct cgroup *cg; 1177 1178 /* link might have been auto-detached by dying cgroup already, 1179 * in that case our work is done here 1180 */ 1181 if (!cg_link->cgroup) 1182 return; 1183 1184 mutex_lock(&cgroup_mutex); 1185 1186 /* re-check cgroup under lock again */ 1187 if (!cg_link->cgroup) { 1188 mutex_unlock(&cgroup_mutex); 1189 return; 1190 } 1191 1192 WARN_ON(__cgroup_bpf_detach(cg_link->cgroup, NULL, cg_link, 1193 cg_link->type)); 1194 if (cg_link->type == BPF_LSM_CGROUP) 1195 bpf_trampoline_unlink_cgroup_shim(cg_link->link.prog); 1196 1197 cg = cg_link->cgroup; 1198 cg_link->cgroup = NULL; 1199 1200 mutex_unlock(&cgroup_mutex); 1201 1202 cgroup_put(cg); 1203 } 1204 1205 static void bpf_cgroup_link_dealloc(struct bpf_link *link) 1206 { 1207 struct bpf_cgroup_link *cg_link = 1208 container_of(link, struct bpf_cgroup_link, link); 1209 1210 kfree(cg_link); 1211 } 1212 1213 static int bpf_cgroup_link_detach(struct bpf_link *link) 1214 { 1215 bpf_cgroup_link_release(link); 1216 1217 return 0; 1218 } 1219 1220 static void bpf_cgroup_link_show_fdinfo(const struct bpf_link *link, 1221 struct seq_file *seq) 1222 { 1223 struct bpf_cgroup_link *cg_link = 1224 container_of(link, struct bpf_cgroup_link, link); 1225 u64 cg_id = 0; 1226 1227 mutex_lock(&cgroup_mutex); 1228 if (cg_link->cgroup) 1229 cg_id = cgroup_id(cg_link->cgroup); 1230 mutex_unlock(&cgroup_mutex); 1231 1232 seq_printf(seq, 1233 "cgroup_id:\t%llu\n" 1234 "attach_type:\t%d\n", 1235 cg_id, 1236 cg_link->type); 1237 } 1238 1239 static int bpf_cgroup_link_fill_link_info(const struct bpf_link *link, 1240 struct bpf_link_info *info) 1241 { 1242 struct bpf_cgroup_link *cg_link = 1243 container_of(link, struct bpf_cgroup_link, link); 1244 u64 cg_id = 0; 1245 1246 mutex_lock(&cgroup_mutex); 1247 if (cg_link->cgroup) 1248 cg_id = cgroup_id(cg_link->cgroup); 1249 mutex_unlock(&cgroup_mutex); 1250 1251 info->cgroup.cgroup_id = cg_id; 1252 info->cgroup.attach_type = cg_link->type; 1253 return 0; 1254 } 1255 1256 static const struct bpf_link_ops bpf_cgroup_link_lops = { 1257 .release = bpf_cgroup_link_release, 1258 .dealloc = bpf_cgroup_link_dealloc, 1259 .detach = bpf_cgroup_link_detach, 1260 .update_prog = cgroup_bpf_replace, 1261 .show_fdinfo = bpf_cgroup_link_show_fdinfo, 1262 .fill_link_info = bpf_cgroup_link_fill_link_info, 1263 }; 1264 1265 int cgroup_bpf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 1266 { 1267 struct bpf_link_primer link_primer; 1268 struct bpf_cgroup_link *link; 1269 struct cgroup *cgrp; 1270 int err; 1271 1272 if (attr->link_create.flags) 1273 return -EINVAL; 1274 1275 cgrp = cgroup_get_from_fd(attr->link_create.target_fd); 1276 if (IS_ERR(cgrp)) 1277 return PTR_ERR(cgrp); 1278 1279 link = kzalloc(sizeof(*link), GFP_USER); 1280 if (!link) { 1281 err = -ENOMEM; 1282 goto out_put_cgroup; 1283 } 1284 bpf_link_init(&link->link, BPF_LINK_TYPE_CGROUP, &bpf_cgroup_link_lops, 1285 prog); 1286 link->cgroup = cgrp; 1287 link->type = attr->link_create.attach_type; 1288 1289 err = bpf_link_prime(&link->link, &link_primer); 1290 if (err) { 1291 kfree(link); 1292 goto out_put_cgroup; 1293 } 1294 1295 err = cgroup_bpf_attach(cgrp, NULL, NULL, link, 1296 link->type, BPF_F_ALLOW_MULTI); 1297 if (err) { 1298 bpf_link_cleanup(&link_primer); 1299 goto out_put_cgroup; 1300 } 1301 1302 return bpf_link_settle(&link_primer); 1303 1304 out_put_cgroup: 1305 cgroup_put(cgrp); 1306 return err; 1307 } 1308 1309 int cgroup_bpf_prog_query(const union bpf_attr *attr, 1310 union bpf_attr __user *uattr) 1311 { 1312 struct cgroup *cgrp; 1313 int ret; 1314 1315 cgrp = cgroup_get_from_fd(attr->query.target_fd); 1316 if (IS_ERR(cgrp)) 1317 return PTR_ERR(cgrp); 1318 1319 ret = cgroup_bpf_query(cgrp, attr, uattr); 1320 1321 cgroup_put(cgrp); 1322 return ret; 1323 } 1324 1325 /** 1326 * __cgroup_bpf_run_filter_skb() - Run a program for packet filtering 1327 * @sk: The socket sending or receiving traffic 1328 * @skb: The skb that is being sent or received 1329 * @type: The type of program to be executed 1330 * 1331 * If no socket is passed, or the socket is not of type INET or INET6, 1332 * this function does nothing and returns 0. 1333 * 1334 * The program type passed in via @type must be suitable for network 1335 * filtering. No further check is performed to assert that. 1336 * 1337 * For egress packets, this function can return: 1338 * NET_XMIT_SUCCESS (0) - continue with packet output 1339 * NET_XMIT_DROP (1) - drop packet and notify TCP to call cwr 1340 * NET_XMIT_CN (2) - continue with packet output and notify TCP 1341 * to call cwr 1342 * -err - drop packet 1343 * 1344 * For ingress packets, this function will return -EPERM if any 1345 * attached program was found and if it returned != 1 during execution. 1346 * Otherwise 0 is returned. 1347 */ 1348 int __cgroup_bpf_run_filter_skb(struct sock *sk, 1349 struct sk_buff *skb, 1350 enum cgroup_bpf_attach_type atype) 1351 { 1352 unsigned int offset = skb->data - skb_network_header(skb); 1353 struct sock *save_sk; 1354 void *saved_data_end; 1355 struct cgroup *cgrp; 1356 int ret; 1357 1358 if (!sk || !sk_fullsock(sk)) 1359 return 0; 1360 1361 if (sk->sk_family != AF_INET && sk->sk_family != AF_INET6) 1362 return 0; 1363 1364 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 1365 save_sk = skb->sk; 1366 skb->sk = sk; 1367 __skb_push(skb, offset); 1368 1369 /* compute pointers for the bpf prog */ 1370 bpf_compute_and_save_data_end(skb, &saved_data_end); 1371 1372 if (atype == CGROUP_INET_EGRESS) { 1373 u32 flags = 0; 1374 bool cn; 1375 1376 ret = bpf_prog_run_array_cg(&cgrp->bpf, atype, skb, 1377 __bpf_prog_run_save_cb, 0, &flags); 1378 1379 /* Return values of CGROUP EGRESS BPF programs are: 1380 * 0: drop packet 1381 * 1: keep packet 1382 * 2: drop packet and cn 1383 * 3: keep packet and cn 1384 * 1385 * The returned value is then converted to one of the NET_XMIT 1386 * or an error code that is then interpreted as drop packet 1387 * (and no cn): 1388 * 0: NET_XMIT_SUCCESS skb should be transmitted 1389 * 1: NET_XMIT_DROP skb should be dropped and cn 1390 * 2: NET_XMIT_CN skb should be transmitted and cn 1391 * 3: -err skb should be dropped 1392 */ 1393 1394 cn = flags & BPF_RET_SET_CN; 1395 if (ret && !IS_ERR_VALUE((long)ret)) 1396 ret = -EFAULT; 1397 if (!ret) 1398 ret = (cn ? NET_XMIT_CN : NET_XMIT_SUCCESS); 1399 else 1400 ret = (cn ? NET_XMIT_DROP : ret); 1401 } else { 1402 ret = bpf_prog_run_array_cg(&cgrp->bpf, atype, 1403 skb, __bpf_prog_run_save_cb, 0, 1404 NULL); 1405 if (ret && !IS_ERR_VALUE((long)ret)) 1406 ret = -EFAULT; 1407 } 1408 bpf_restore_data_end(skb, saved_data_end); 1409 __skb_pull(skb, offset); 1410 skb->sk = save_sk; 1411 1412 return ret; 1413 } 1414 EXPORT_SYMBOL(__cgroup_bpf_run_filter_skb); 1415 1416 /** 1417 * __cgroup_bpf_run_filter_sk() - Run a program on a sock 1418 * @sk: sock structure to manipulate 1419 * @type: The type of program to be executed 1420 * 1421 * socket is passed is expected to be of type INET or INET6. 1422 * 1423 * The program type passed in via @type must be suitable for sock 1424 * filtering. No further check is performed to assert that. 1425 * 1426 * This function will return %-EPERM if any if an attached program was found 1427 * and if it returned != 1 during execution. In all other cases, 0 is returned. 1428 */ 1429 int __cgroup_bpf_run_filter_sk(struct sock *sk, 1430 enum cgroup_bpf_attach_type atype) 1431 { 1432 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 1433 1434 return bpf_prog_run_array_cg(&cgrp->bpf, atype, sk, bpf_prog_run, 0, 1435 NULL); 1436 } 1437 EXPORT_SYMBOL(__cgroup_bpf_run_filter_sk); 1438 1439 /** 1440 * __cgroup_bpf_run_filter_sock_addr() - Run a program on a sock and 1441 * provided by user sockaddr 1442 * @sk: sock struct that will use sockaddr 1443 * @uaddr: sockaddr struct provided by user 1444 * @type: The type of program to be executed 1445 * @t_ctx: Pointer to attach type specific context 1446 * @flags: Pointer to u32 which contains higher bits of BPF program 1447 * return value (OR'ed together). 1448 * 1449 * socket is expected to be of type INET or INET6. 1450 * 1451 * This function will return %-EPERM if an attached program is found and 1452 * returned value != 1 during execution. In all other cases, 0 is returned. 1453 */ 1454 int __cgroup_bpf_run_filter_sock_addr(struct sock *sk, 1455 struct sockaddr *uaddr, 1456 enum cgroup_bpf_attach_type atype, 1457 void *t_ctx, 1458 u32 *flags) 1459 { 1460 struct bpf_sock_addr_kern ctx = { 1461 .sk = sk, 1462 .uaddr = uaddr, 1463 .t_ctx = t_ctx, 1464 }; 1465 struct sockaddr_storage unspec; 1466 struct cgroup *cgrp; 1467 1468 /* Check socket family since not all sockets represent network 1469 * endpoint (e.g. AF_UNIX). 1470 */ 1471 if (sk->sk_family != AF_INET && sk->sk_family != AF_INET6) 1472 return 0; 1473 1474 if (!ctx.uaddr) { 1475 memset(&unspec, 0, sizeof(unspec)); 1476 ctx.uaddr = (struct sockaddr *)&unspec; 1477 } 1478 1479 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 1480 return bpf_prog_run_array_cg(&cgrp->bpf, atype, &ctx, bpf_prog_run, 1481 0, flags); 1482 } 1483 EXPORT_SYMBOL(__cgroup_bpf_run_filter_sock_addr); 1484 1485 /** 1486 * __cgroup_bpf_run_filter_sock_ops() - Run a program on a sock 1487 * @sk: socket to get cgroup from 1488 * @sock_ops: bpf_sock_ops_kern struct to pass to program. Contains 1489 * sk with connection information (IP addresses, etc.) May not contain 1490 * cgroup info if it is a req sock. 1491 * @type: The type of program to be executed 1492 * 1493 * socket passed is expected to be of type INET or INET6. 1494 * 1495 * The program type passed in via @type must be suitable for sock_ops 1496 * filtering. No further check is performed to assert that. 1497 * 1498 * This function will return %-EPERM if any if an attached program was found 1499 * and if it returned != 1 during execution. In all other cases, 0 is returned. 1500 */ 1501 int __cgroup_bpf_run_filter_sock_ops(struct sock *sk, 1502 struct bpf_sock_ops_kern *sock_ops, 1503 enum cgroup_bpf_attach_type atype) 1504 { 1505 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 1506 1507 return bpf_prog_run_array_cg(&cgrp->bpf, atype, sock_ops, bpf_prog_run, 1508 0, NULL); 1509 } 1510 EXPORT_SYMBOL(__cgroup_bpf_run_filter_sock_ops); 1511 1512 int __cgroup_bpf_check_dev_permission(short dev_type, u32 major, u32 minor, 1513 short access, enum cgroup_bpf_attach_type atype) 1514 { 1515 struct cgroup *cgrp; 1516 struct bpf_cgroup_dev_ctx ctx = { 1517 .access_type = (access << 16) | dev_type, 1518 .major = major, 1519 .minor = minor, 1520 }; 1521 int ret; 1522 1523 rcu_read_lock(); 1524 cgrp = task_dfl_cgroup(current); 1525 ret = bpf_prog_run_array_cg(&cgrp->bpf, atype, &ctx, bpf_prog_run, 0, 1526 NULL); 1527 rcu_read_unlock(); 1528 1529 return ret; 1530 } 1531 1532 BPF_CALL_2(bpf_get_local_storage, struct bpf_map *, map, u64, flags) 1533 { 1534 /* flags argument is not used now, 1535 * but provides an ability to extend the API. 1536 * verifier checks that its value is correct. 1537 */ 1538 enum bpf_cgroup_storage_type stype = cgroup_storage_type(map); 1539 struct bpf_cgroup_storage *storage; 1540 struct bpf_cg_run_ctx *ctx; 1541 void *ptr; 1542 1543 /* get current cgroup storage from BPF run context */ 1544 ctx = container_of(current->bpf_ctx, struct bpf_cg_run_ctx, run_ctx); 1545 storage = ctx->prog_item->cgroup_storage[stype]; 1546 1547 if (stype == BPF_CGROUP_STORAGE_SHARED) 1548 ptr = &READ_ONCE(storage->buf)->data[0]; 1549 else 1550 ptr = this_cpu_ptr(storage->percpu_buf); 1551 1552 return (unsigned long)ptr; 1553 } 1554 1555 const struct bpf_func_proto bpf_get_local_storage_proto = { 1556 .func = bpf_get_local_storage, 1557 .gpl_only = false, 1558 .ret_type = RET_PTR_TO_MAP_VALUE, 1559 .arg1_type = ARG_CONST_MAP_PTR, 1560 .arg2_type = ARG_ANYTHING, 1561 }; 1562 1563 BPF_CALL_0(bpf_get_retval) 1564 { 1565 struct bpf_cg_run_ctx *ctx = 1566 container_of(current->bpf_ctx, struct bpf_cg_run_ctx, run_ctx); 1567 1568 return ctx->retval; 1569 } 1570 1571 const struct bpf_func_proto bpf_get_retval_proto = { 1572 .func = bpf_get_retval, 1573 .gpl_only = false, 1574 .ret_type = RET_INTEGER, 1575 }; 1576 1577 BPF_CALL_1(bpf_set_retval, int, retval) 1578 { 1579 struct bpf_cg_run_ctx *ctx = 1580 container_of(current->bpf_ctx, struct bpf_cg_run_ctx, run_ctx); 1581 1582 ctx->retval = retval; 1583 return 0; 1584 } 1585 1586 const struct bpf_func_proto bpf_set_retval_proto = { 1587 .func = bpf_set_retval, 1588 .gpl_only = false, 1589 .ret_type = RET_INTEGER, 1590 .arg1_type = ARG_ANYTHING, 1591 }; 1592 1593 static const struct bpf_func_proto * 1594 cgroup_dev_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 1595 { 1596 const struct bpf_func_proto *func_proto; 1597 1598 func_proto = cgroup_common_func_proto(func_id, prog); 1599 if (func_proto) 1600 return func_proto; 1601 1602 func_proto = cgroup_current_func_proto(func_id, prog); 1603 if (func_proto) 1604 return func_proto; 1605 1606 switch (func_id) { 1607 case BPF_FUNC_perf_event_output: 1608 return &bpf_event_output_data_proto; 1609 default: 1610 return bpf_base_func_proto(func_id); 1611 } 1612 } 1613 1614 static bool cgroup_dev_is_valid_access(int off, int size, 1615 enum bpf_access_type type, 1616 const struct bpf_prog *prog, 1617 struct bpf_insn_access_aux *info) 1618 { 1619 const int size_default = sizeof(__u32); 1620 1621 if (type == BPF_WRITE) 1622 return false; 1623 1624 if (off < 0 || off + size > sizeof(struct bpf_cgroup_dev_ctx)) 1625 return false; 1626 /* The verifier guarantees that size > 0. */ 1627 if (off % size != 0) 1628 return false; 1629 1630 switch (off) { 1631 case bpf_ctx_range(struct bpf_cgroup_dev_ctx, access_type): 1632 bpf_ctx_record_field_size(info, size_default); 1633 if (!bpf_ctx_narrow_access_ok(off, size, size_default)) 1634 return false; 1635 break; 1636 default: 1637 if (size != size_default) 1638 return false; 1639 } 1640 1641 return true; 1642 } 1643 1644 const struct bpf_prog_ops cg_dev_prog_ops = { 1645 }; 1646 1647 const struct bpf_verifier_ops cg_dev_verifier_ops = { 1648 .get_func_proto = cgroup_dev_func_proto, 1649 .is_valid_access = cgroup_dev_is_valid_access, 1650 }; 1651 1652 /** 1653 * __cgroup_bpf_run_filter_sysctl - Run a program on sysctl 1654 * 1655 * @head: sysctl table header 1656 * @table: sysctl table 1657 * @write: sysctl is being read (= 0) or written (= 1) 1658 * @buf: pointer to buffer (in and out) 1659 * @pcount: value-result argument: value is size of buffer pointed to by @buf, 1660 * result is size of @new_buf if program set new value, initial value 1661 * otherwise 1662 * @ppos: value-result argument: value is position at which read from or write 1663 * to sysctl is happening, result is new position if program overrode it, 1664 * initial value otherwise 1665 * @type: type of program to be executed 1666 * 1667 * Program is run when sysctl is being accessed, either read or written, and 1668 * can allow or deny such access. 1669 * 1670 * This function will return %-EPERM if an attached program is found and 1671 * returned value != 1 during execution. In all other cases 0 is returned. 1672 */ 1673 int __cgroup_bpf_run_filter_sysctl(struct ctl_table_header *head, 1674 struct ctl_table *table, int write, 1675 char **buf, size_t *pcount, loff_t *ppos, 1676 enum cgroup_bpf_attach_type atype) 1677 { 1678 struct bpf_sysctl_kern ctx = { 1679 .head = head, 1680 .table = table, 1681 .write = write, 1682 .ppos = ppos, 1683 .cur_val = NULL, 1684 .cur_len = PAGE_SIZE, 1685 .new_val = NULL, 1686 .new_len = 0, 1687 .new_updated = 0, 1688 }; 1689 struct cgroup *cgrp; 1690 loff_t pos = 0; 1691 int ret; 1692 1693 ctx.cur_val = kmalloc_track_caller(ctx.cur_len, GFP_KERNEL); 1694 if (!ctx.cur_val || 1695 table->proc_handler(table, 0, ctx.cur_val, &ctx.cur_len, &pos)) { 1696 /* Let BPF program decide how to proceed. */ 1697 ctx.cur_len = 0; 1698 } 1699 1700 if (write && *buf && *pcount) { 1701 /* BPF program should be able to override new value with a 1702 * buffer bigger than provided by user. 1703 */ 1704 ctx.new_val = kmalloc_track_caller(PAGE_SIZE, GFP_KERNEL); 1705 ctx.new_len = min_t(size_t, PAGE_SIZE, *pcount); 1706 if (ctx.new_val) { 1707 memcpy(ctx.new_val, *buf, ctx.new_len); 1708 } else { 1709 /* Let BPF program decide how to proceed. */ 1710 ctx.new_len = 0; 1711 } 1712 } 1713 1714 rcu_read_lock(); 1715 cgrp = task_dfl_cgroup(current); 1716 ret = bpf_prog_run_array_cg(&cgrp->bpf, atype, &ctx, bpf_prog_run, 0, 1717 NULL); 1718 rcu_read_unlock(); 1719 1720 kfree(ctx.cur_val); 1721 1722 if (ret == 1 && ctx.new_updated) { 1723 kfree(*buf); 1724 *buf = ctx.new_val; 1725 *pcount = ctx.new_len; 1726 } else { 1727 kfree(ctx.new_val); 1728 } 1729 1730 return ret; 1731 } 1732 1733 #ifdef CONFIG_NET 1734 static int sockopt_alloc_buf(struct bpf_sockopt_kern *ctx, int max_optlen, 1735 struct bpf_sockopt_buf *buf) 1736 { 1737 if (unlikely(max_optlen < 0)) 1738 return -EINVAL; 1739 1740 if (unlikely(max_optlen > PAGE_SIZE)) { 1741 /* We don't expose optvals that are greater than PAGE_SIZE 1742 * to the BPF program. 1743 */ 1744 max_optlen = PAGE_SIZE; 1745 } 1746 1747 if (max_optlen <= sizeof(buf->data)) { 1748 /* When the optval fits into BPF_SOCKOPT_KERN_BUF_SIZE 1749 * bytes avoid the cost of kzalloc. 1750 */ 1751 ctx->optval = buf->data; 1752 ctx->optval_end = ctx->optval + max_optlen; 1753 return max_optlen; 1754 } 1755 1756 ctx->optval = kzalloc(max_optlen, GFP_USER); 1757 if (!ctx->optval) 1758 return -ENOMEM; 1759 1760 ctx->optval_end = ctx->optval + max_optlen; 1761 1762 return max_optlen; 1763 } 1764 1765 static void sockopt_free_buf(struct bpf_sockopt_kern *ctx, 1766 struct bpf_sockopt_buf *buf) 1767 { 1768 if (ctx->optval == buf->data) 1769 return; 1770 kfree(ctx->optval); 1771 } 1772 1773 static bool sockopt_buf_allocated(struct bpf_sockopt_kern *ctx, 1774 struct bpf_sockopt_buf *buf) 1775 { 1776 return ctx->optval != buf->data; 1777 } 1778 1779 int __cgroup_bpf_run_filter_setsockopt(struct sock *sk, int *level, 1780 int *optname, char __user *optval, 1781 int *optlen, char **kernel_optval) 1782 { 1783 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 1784 struct bpf_sockopt_buf buf = {}; 1785 struct bpf_sockopt_kern ctx = { 1786 .sk = sk, 1787 .level = *level, 1788 .optname = *optname, 1789 }; 1790 int ret, max_optlen; 1791 1792 /* Allocate a bit more than the initial user buffer for 1793 * BPF program. The canonical use case is overriding 1794 * TCP_CONGESTION(nv) to TCP_CONGESTION(cubic). 1795 */ 1796 max_optlen = max_t(int, 16, *optlen); 1797 max_optlen = sockopt_alloc_buf(&ctx, max_optlen, &buf); 1798 if (max_optlen < 0) 1799 return max_optlen; 1800 1801 ctx.optlen = *optlen; 1802 1803 if (copy_from_user(ctx.optval, optval, min(*optlen, max_optlen)) != 0) { 1804 ret = -EFAULT; 1805 goto out; 1806 } 1807 1808 lock_sock(sk); 1809 ret = bpf_prog_run_array_cg(&cgrp->bpf, CGROUP_SETSOCKOPT, 1810 &ctx, bpf_prog_run, 0, NULL); 1811 release_sock(sk); 1812 1813 if (ret) 1814 goto out; 1815 1816 if (ctx.optlen == -1) { 1817 /* optlen set to -1, bypass kernel */ 1818 ret = 1; 1819 } else if (ctx.optlen > max_optlen || ctx.optlen < -1) { 1820 /* optlen is out of bounds */ 1821 ret = -EFAULT; 1822 } else { 1823 /* optlen within bounds, run kernel handler */ 1824 ret = 0; 1825 1826 /* export any potential modifications */ 1827 *level = ctx.level; 1828 *optname = ctx.optname; 1829 1830 /* optlen == 0 from BPF indicates that we should 1831 * use original userspace data. 1832 */ 1833 if (ctx.optlen != 0) { 1834 *optlen = ctx.optlen; 1835 /* We've used bpf_sockopt_kern->buf as an intermediary 1836 * storage, but the BPF program indicates that we need 1837 * to pass this data to the kernel setsockopt handler. 1838 * No way to export on-stack buf, have to allocate a 1839 * new buffer. 1840 */ 1841 if (!sockopt_buf_allocated(&ctx, &buf)) { 1842 void *p = kmalloc(ctx.optlen, GFP_USER); 1843 1844 if (!p) { 1845 ret = -ENOMEM; 1846 goto out; 1847 } 1848 memcpy(p, ctx.optval, ctx.optlen); 1849 *kernel_optval = p; 1850 } else { 1851 *kernel_optval = ctx.optval; 1852 } 1853 /* export and don't free sockopt buf */ 1854 return 0; 1855 } 1856 } 1857 1858 out: 1859 sockopt_free_buf(&ctx, &buf); 1860 return ret; 1861 } 1862 1863 int __cgroup_bpf_run_filter_getsockopt(struct sock *sk, int level, 1864 int optname, char __user *optval, 1865 int __user *optlen, int max_optlen, 1866 int retval) 1867 { 1868 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 1869 struct bpf_sockopt_buf buf = {}; 1870 struct bpf_sockopt_kern ctx = { 1871 .sk = sk, 1872 .level = level, 1873 .optname = optname, 1874 .current_task = current, 1875 }; 1876 int ret; 1877 1878 ctx.optlen = max_optlen; 1879 max_optlen = sockopt_alloc_buf(&ctx, max_optlen, &buf); 1880 if (max_optlen < 0) 1881 return max_optlen; 1882 1883 if (!retval) { 1884 /* If kernel getsockopt finished successfully, 1885 * copy whatever was returned to the user back 1886 * into our temporary buffer. Set optlen to the 1887 * one that kernel returned as well to let 1888 * BPF programs inspect the value. 1889 */ 1890 1891 if (get_user(ctx.optlen, optlen)) { 1892 ret = -EFAULT; 1893 goto out; 1894 } 1895 1896 if (ctx.optlen < 0) { 1897 ret = -EFAULT; 1898 goto out; 1899 } 1900 1901 if (copy_from_user(ctx.optval, optval, 1902 min(ctx.optlen, max_optlen)) != 0) { 1903 ret = -EFAULT; 1904 goto out; 1905 } 1906 } 1907 1908 lock_sock(sk); 1909 ret = bpf_prog_run_array_cg(&cgrp->bpf, CGROUP_GETSOCKOPT, 1910 &ctx, bpf_prog_run, retval, NULL); 1911 release_sock(sk); 1912 1913 if (ret < 0) 1914 goto out; 1915 1916 if (ctx.optlen > max_optlen || ctx.optlen < 0) { 1917 ret = -EFAULT; 1918 goto out; 1919 } 1920 1921 if (ctx.optlen != 0) { 1922 if (copy_to_user(optval, ctx.optval, ctx.optlen) || 1923 put_user(ctx.optlen, optlen)) { 1924 ret = -EFAULT; 1925 goto out; 1926 } 1927 } 1928 1929 out: 1930 sockopt_free_buf(&ctx, &buf); 1931 return ret; 1932 } 1933 1934 int __cgroup_bpf_run_filter_getsockopt_kern(struct sock *sk, int level, 1935 int optname, void *optval, 1936 int *optlen, int retval) 1937 { 1938 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 1939 struct bpf_sockopt_kern ctx = { 1940 .sk = sk, 1941 .level = level, 1942 .optname = optname, 1943 .optlen = *optlen, 1944 .optval = optval, 1945 .optval_end = optval + *optlen, 1946 .current_task = current, 1947 }; 1948 int ret; 1949 1950 /* Note that __cgroup_bpf_run_filter_getsockopt doesn't copy 1951 * user data back into BPF buffer when reval != 0. This is 1952 * done as an optimization to avoid extra copy, assuming 1953 * kernel won't populate the data in case of an error. 1954 * Here we always pass the data and memset() should 1955 * be called if that data shouldn't be "exported". 1956 */ 1957 1958 ret = bpf_prog_run_array_cg(&cgrp->bpf, CGROUP_GETSOCKOPT, 1959 &ctx, bpf_prog_run, retval, NULL); 1960 if (ret < 0) 1961 return ret; 1962 1963 if (ctx.optlen > *optlen) 1964 return -EFAULT; 1965 1966 /* BPF programs can shrink the buffer, export the modifications. 1967 */ 1968 if (ctx.optlen != 0) 1969 *optlen = ctx.optlen; 1970 1971 return ret; 1972 } 1973 #endif 1974 1975 static ssize_t sysctl_cpy_dir(const struct ctl_dir *dir, char **bufp, 1976 size_t *lenp) 1977 { 1978 ssize_t tmp_ret = 0, ret; 1979 1980 if (dir->header.parent) { 1981 tmp_ret = sysctl_cpy_dir(dir->header.parent, bufp, lenp); 1982 if (tmp_ret < 0) 1983 return tmp_ret; 1984 } 1985 1986 ret = strscpy(*bufp, dir->header.ctl_table[0].procname, *lenp); 1987 if (ret < 0) 1988 return ret; 1989 *bufp += ret; 1990 *lenp -= ret; 1991 ret += tmp_ret; 1992 1993 /* Avoid leading slash. */ 1994 if (!ret) 1995 return ret; 1996 1997 tmp_ret = strscpy(*bufp, "/", *lenp); 1998 if (tmp_ret < 0) 1999 return tmp_ret; 2000 *bufp += tmp_ret; 2001 *lenp -= tmp_ret; 2002 2003 return ret + tmp_ret; 2004 } 2005 2006 BPF_CALL_4(bpf_sysctl_get_name, struct bpf_sysctl_kern *, ctx, char *, buf, 2007 size_t, buf_len, u64, flags) 2008 { 2009 ssize_t tmp_ret = 0, ret; 2010 2011 if (!buf) 2012 return -EINVAL; 2013 2014 if (!(flags & BPF_F_SYSCTL_BASE_NAME)) { 2015 if (!ctx->head) 2016 return -EINVAL; 2017 tmp_ret = sysctl_cpy_dir(ctx->head->parent, &buf, &buf_len); 2018 if (tmp_ret < 0) 2019 return tmp_ret; 2020 } 2021 2022 ret = strscpy(buf, ctx->table->procname, buf_len); 2023 2024 return ret < 0 ? ret : tmp_ret + ret; 2025 } 2026 2027 static const struct bpf_func_proto bpf_sysctl_get_name_proto = { 2028 .func = bpf_sysctl_get_name, 2029 .gpl_only = false, 2030 .ret_type = RET_INTEGER, 2031 .arg1_type = ARG_PTR_TO_CTX, 2032 .arg2_type = ARG_PTR_TO_MEM, 2033 .arg3_type = ARG_CONST_SIZE, 2034 .arg4_type = ARG_ANYTHING, 2035 }; 2036 2037 static int copy_sysctl_value(char *dst, size_t dst_len, char *src, 2038 size_t src_len) 2039 { 2040 if (!dst) 2041 return -EINVAL; 2042 2043 if (!dst_len) 2044 return -E2BIG; 2045 2046 if (!src || !src_len) { 2047 memset(dst, 0, dst_len); 2048 return -EINVAL; 2049 } 2050 2051 memcpy(dst, src, min(dst_len, src_len)); 2052 2053 if (dst_len > src_len) { 2054 memset(dst + src_len, '\0', dst_len - src_len); 2055 return src_len; 2056 } 2057 2058 dst[dst_len - 1] = '\0'; 2059 2060 return -E2BIG; 2061 } 2062 2063 BPF_CALL_3(bpf_sysctl_get_current_value, struct bpf_sysctl_kern *, ctx, 2064 char *, buf, size_t, buf_len) 2065 { 2066 return copy_sysctl_value(buf, buf_len, ctx->cur_val, ctx->cur_len); 2067 } 2068 2069 static const struct bpf_func_proto bpf_sysctl_get_current_value_proto = { 2070 .func = bpf_sysctl_get_current_value, 2071 .gpl_only = false, 2072 .ret_type = RET_INTEGER, 2073 .arg1_type = ARG_PTR_TO_CTX, 2074 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 2075 .arg3_type = ARG_CONST_SIZE, 2076 }; 2077 2078 BPF_CALL_3(bpf_sysctl_get_new_value, struct bpf_sysctl_kern *, ctx, char *, buf, 2079 size_t, buf_len) 2080 { 2081 if (!ctx->write) { 2082 if (buf && buf_len) 2083 memset(buf, '\0', buf_len); 2084 return -EINVAL; 2085 } 2086 return copy_sysctl_value(buf, buf_len, ctx->new_val, ctx->new_len); 2087 } 2088 2089 static const struct bpf_func_proto bpf_sysctl_get_new_value_proto = { 2090 .func = bpf_sysctl_get_new_value, 2091 .gpl_only = false, 2092 .ret_type = RET_INTEGER, 2093 .arg1_type = ARG_PTR_TO_CTX, 2094 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 2095 .arg3_type = ARG_CONST_SIZE, 2096 }; 2097 2098 BPF_CALL_3(bpf_sysctl_set_new_value, struct bpf_sysctl_kern *, ctx, 2099 const char *, buf, size_t, buf_len) 2100 { 2101 if (!ctx->write || !ctx->new_val || !ctx->new_len || !buf || !buf_len) 2102 return -EINVAL; 2103 2104 if (buf_len > PAGE_SIZE - 1) 2105 return -E2BIG; 2106 2107 memcpy(ctx->new_val, buf, buf_len); 2108 ctx->new_len = buf_len; 2109 ctx->new_updated = 1; 2110 2111 return 0; 2112 } 2113 2114 static const struct bpf_func_proto bpf_sysctl_set_new_value_proto = { 2115 .func = bpf_sysctl_set_new_value, 2116 .gpl_only = false, 2117 .ret_type = RET_INTEGER, 2118 .arg1_type = ARG_PTR_TO_CTX, 2119 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 2120 .arg3_type = ARG_CONST_SIZE, 2121 }; 2122 2123 static const struct bpf_func_proto * 2124 sysctl_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 2125 { 2126 const struct bpf_func_proto *func_proto; 2127 2128 func_proto = cgroup_common_func_proto(func_id, prog); 2129 if (func_proto) 2130 return func_proto; 2131 2132 func_proto = cgroup_current_func_proto(func_id, prog); 2133 if (func_proto) 2134 return func_proto; 2135 2136 switch (func_id) { 2137 case BPF_FUNC_sysctl_get_name: 2138 return &bpf_sysctl_get_name_proto; 2139 case BPF_FUNC_sysctl_get_current_value: 2140 return &bpf_sysctl_get_current_value_proto; 2141 case BPF_FUNC_sysctl_get_new_value: 2142 return &bpf_sysctl_get_new_value_proto; 2143 case BPF_FUNC_sysctl_set_new_value: 2144 return &bpf_sysctl_set_new_value_proto; 2145 case BPF_FUNC_ktime_get_coarse_ns: 2146 return &bpf_ktime_get_coarse_ns_proto; 2147 case BPF_FUNC_perf_event_output: 2148 return &bpf_event_output_data_proto; 2149 default: 2150 return bpf_base_func_proto(func_id); 2151 } 2152 } 2153 2154 static bool sysctl_is_valid_access(int off, int size, enum bpf_access_type type, 2155 const struct bpf_prog *prog, 2156 struct bpf_insn_access_aux *info) 2157 { 2158 const int size_default = sizeof(__u32); 2159 2160 if (off < 0 || off + size > sizeof(struct bpf_sysctl) || off % size) 2161 return false; 2162 2163 switch (off) { 2164 case bpf_ctx_range(struct bpf_sysctl, write): 2165 if (type != BPF_READ) 2166 return false; 2167 bpf_ctx_record_field_size(info, size_default); 2168 return bpf_ctx_narrow_access_ok(off, size, size_default); 2169 case bpf_ctx_range(struct bpf_sysctl, file_pos): 2170 if (type == BPF_READ) { 2171 bpf_ctx_record_field_size(info, size_default); 2172 return bpf_ctx_narrow_access_ok(off, size, size_default); 2173 } else { 2174 return size == size_default; 2175 } 2176 default: 2177 return false; 2178 } 2179 } 2180 2181 static u32 sysctl_convert_ctx_access(enum bpf_access_type type, 2182 const struct bpf_insn *si, 2183 struct bpf_insn *insn_buf, 2184 struct bpf_prog *prog, u32 *target_size) 2185 { 2186 struct bpf_insn *insn = insn_buf; 2187 u32 read_size; 2188 2189 switch (si->off) { 2190 case offsetof(struct bpf_sysctl, write): 2191 *insn++ = BPF_LDX_MEM( 2192 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 2193 bpf_target_off(struct bpf_sysctl_kern, write, 2194 sizeof_field(struct bpf_sysctl_kern, 2195 write), 2196 target_size)); 2197 break; 2198 case offsetof(struct bpf_sysctl, file_pos): 2199 /* ppos is a pointer so it should be accessed via indirect 2200 * loads and stores. Also for stores additional temporary 2201 * register is used since neither src_reg nor dst_reg can be 2202 * overridden. 2203 */ 2204 if (type == BPF_WRITE) { 2205 int treg = BPF_REG_9; 2206 2207 if (si->src_reg == treg || si->dst_reg == treg) 2208 --treg; 2209 if (si->src_reg == treg || si->dst_reg == treg) 2210 --treg; 2211 *insn++ = BPF_STX_MEM( 2212 BPF_DW, si->dst_reg, treg, 2213 offsetof(struct bpf_sysctl_kern, tmp_reg)); 2214 *insn++ = BPF_LDX_MEM( 2215 BPF_FIELD_SIZEOF(struct bpf_sysctl_kern, ppos), 2216 treg, si->dst_reg, 2217 offsetof(struct bpf_sysctl_kern, ppos)); 2218 *insn++ = BPF_STX_MEM( 2219 BPF_SIZEOF(u32), treg, si->src_reg, 2220 bpf_ctx_narrow_access_offset( 2221 0, sizeof(u32), sizeof(loff_t))); 2222 *insn++ = BPF_LDX_MEM( 2223 BPF_DW, treg, si->dst_reg, 2224 offsetof(struct bpf_sysctl_kern, tmp_reg)); 2225 } else { 2226 *insn++ = BPF_LDX_MEM( 2227 BPF_FIELD_SIZEOF(struct bpf_sysctl_kern, ppos), 2228 si->dst_reg, si->src_reg, 2229 offsetof(struct bpf_sysctl_kern, ppos)); 2230 read_size = bpf_size_to_bytes(BPF_SIZE(si->code)); 2231 *insn++ = BPF_LDX_MEM( 2232 BPF_SIZE(si->code), si->dst_reg, si->dst_reg, 2233 bpf_ctx_narrow_access_offset( 2234 0, read_size, sizeof(loff_t))); 2235 } 2236 *target_size = sizeof(u32); 2237 break; 2238 } 2239 2240 return insn - insn_buf; 2241 } 2242 2243 const struct bpf_verifier_ops cg_sysctl_verifier_ops = { 2244 .get_func_proto = sysctl_func_proto, 2245 .is_valid_access = sysctl_is_valid_access, 2246 .convert_ctx_access = sysctl_convert_ctx_access, 2247 }; 2248 2249 const struct bpf_prog_ops cg_sysctl_prog_ops = { 2250 }; 2251 2252 #ifdef CONFIG_NET 2253 BPF_CALL_1(bpf_get_netns_cookie_sockopt, struct bpf_sockopt_kern *, ctx) 2254 { 2255 const struct net *net = ctx ? sock_net(ctx->sk) : &init_net; 2256 2257 return net->net_cookie; 2258 } 2259 2260 static const struct bpf_func_proto bpf_get_netns_cookie_sockopt_proto = { 2261 .func = bpf_get_netns_cookie_sockopt, 2262 .gpl_only = false, 2263 .ret_type = RET_INTEGER, 2264 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 2265 }; 2266 #endif 2267 2268 static const struct bpf_func_proto * 2269 cg_sockopt_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 2270 { 2271 const struct bpf_func_proto *func_proto; 2272 2273 func_proto = cgroup_common_func_proto(func_id, prog); 2274 if (func_proto) 2275 return func_proto; 2276 2277 func_proto = cgroup_current_func_proto(func_id, prog); 2278 if (func_proto) 2279 return func_proto; 2280 2281 switch (func_id) { 2282 #ifdef CONFIG_NET 2283 case BPF_FUNC_get_netns_cookie: 2284 return &bpf_get_netns_cookie_sockopt_proto; 2285 case BPF_FUNC_sk_storage_get: 2286 return &bpf_sk_storage_get_proto; 2287 case BPF_FUNC_sk_storage_delete: 2288 return &bpf_sk_storage_delete_proto; 2289 case BPF_FUNC_setsockopt: 2290 if (prog->expected_attach_type == BPF_CGROUP_SETSOCKOPT) 2291 return &bpf_sk_setsockopt_proto; 2292 return NULL; 2293 case BPF_FUNC_getsockopt: 2294 if (prog->expected_attach_type == BPF_CGROUP_SETSOCKOPT) 2295 return &bpf_sk_getsockopt_proto; 2296 return NULL; 2297 #endif 2298 #ifdef CONFIG_INET 2299 case BPF_FUNC_tcp_sock: 2300 return &bpf_tcp_sock_proto; 2301 #endif 2302 case BPF_FUNC_perf_event_output: 2303 return &bpf_event_output_data_proto; 2304 default: 2305 return bpf_base_func_proto(func_id); 2306 } 2307 } 2308 2309 static bool cg_sockopt_is_valid_access(int off, int size, 2310 enum bpf_access_type type, 2311 const struct bpf_prog *prog, 2312 struct bpf_insn_access_aux *info) 2313 { 2314 const int size_default = sizeof(__u32); 2315 2316 if (off < 0 || off >= sizeof(struct bpf_sockopt)) 2317 return false; 2318 2319 if (off % size != 0) 2320 return false; 2321 2322 if (type == BPF_WRITE) { 2323 switch (off) { 2324 case offsetof(struct bpf_sockopt, retval): 2325 if (size != size_default) 2326 return false; 2327 return prog->expected_attach_type == 2328 BPF_CGROUP_GETSOCKOPT; 2329 case offsetof(struct bpf_sockopt, optname): 2330 fallthrough; 2331 case offsetof(struct bpf_sockopt, level): 2332 if (size != size_default) 2333 return false; 2334 return prog->expected_attach_type == 2335 BPF_CGROUP_SETSOCKOPT; 2336 case offsetof(struct bpf_sockopt, optlen): 2337 return size == size_default; 2338 default: 2339 return false; 2340 } 2341 } 2342 2343 switch (off) { 2344 case offsetof(struct bpf_sockopt, sk): 2345 if (size != sizeof(__u64)) 2346 return false; 2347 info->reg_type = PTR_TO_SOCKET; 2348 break; 2349 case offsetof(struct bpf_sockopt, optval): 2350 if (size != sizeof(__u64)) 2351 return false; 2352 info->reg_type = PTR_TO_PACKET; 2353 break; 2354 case offsetof(struct bpf_sockopt, optval_end): 2355 if (size != sizeof(__u64)) 2356 return false; 2357 info->reg_type = PTR_TO_PACKET_END; 2358 break; 2359 case offsetof(struct bpf_sockopt, retval): 2360 if (size != size_default) 2361 return false; 2362 return prog->expected_attach_type == BPF_CGROUP_GETSOCKOPT; 2363 default: 2364 if (size != size_default) 2365 return false; 2366 break; 2367 } 2368 return true; 2369 } 2370 2371 #define CG_SOCKOPT_ACCESS_FIELD(T, F) \ 2372 T(BPF_FIELD_SIZEOF(struct bpf_sockopt_kern, F), \ 2373 si->dst_reg, si->src_reg, \ 2374 offsetof(struct bpf_sockopt_kern, F)) 2375 2376 static u32 cg_sockopt_convert_ctx_access(enum bpf_access_type type, 2377 const struct bpf_insn *si, 2378 struct bpf_insn *insn_buf, 2379 struct bpf_prog *prog, 2380 u32 *target_size) 2381 { 2382 struct bpf_insn *insn = insn_buf; 2383 2384 switch (si->off) { 2385 case offsetof(struct bpf_sockopt, sk): 2386 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, sk); 2387 break; 2388 case offsetof(struct bpf_sockopt, level): 2389 if (type == BPF_WRITE) 2390 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_STX_MEM, level); 2391 else 2392 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, level); 2393 break; 2394 case offsetof(struct bpf_sockopt, optname): 2395 if (type == BPF_WRITE) 2396 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_STX_MEM, optname); 2397 else 2398 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, optname); 2399 break; 2400 case offsetof(struct bpf_sockopt, optlen): 2401 if (type == BPF_WRITE) 2402 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_STX_MEM, optlen); 2403 else 2404 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, optlen); 2405 break; 2406 case offsetof(struct bpf_sockopt, retval): 2407 BUILD_BUG_ON(offsetof(struct bpf_cg_run_ctx, run_ctx) != 0); 2408 2409 if (type == BPF_WRITE) { 2410 int treg = BPF_REG_9; 2411 2412 if (si->src_reg == treg || si->dst_reg == treg) 2413 --treg; 2414 if (si->src_reg == treg || si->dst_reg == treg) 2415 --treg; 2416 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, treg, 2417 offsetof(struct bpf_sockopt_kern, tmp_reg)); 2418 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sockopt_kern, current_task), 2419 treg, si->dst_reg, 2420 offsetof(struct bpf_sockopt_kern, current_task)); 2421 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct task_struct, bpf_ctx), 2422 treg, treg, 2423 offsetof(struct task_struct, bpf_ctx)); 2424 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(struct bpf_cg_run_ctx, retval), 2425 treg, si->src_reg, 2426 offsetof(struct bpf_cg_run_ctx, retval)); 2427 *insn++ = BPF_LDX_MEM(BPF_DW, treg, si->dst_reg, 2428 offsetof(struct bpf_sockopt_kern, tmp_reg)); 2429 } else { 2430 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sockopt_kern, current_task), 2431 si->dst_reg, si->src_reg, 2432 offsetof(struct bpf_sockopt_kern, current_task)); 2433 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct task_struct, bpf_ctx), 2434 si->dst_reg, si->dst_reg, 2435 offsetof(struct task_struct, bpf_ctx)); 2436 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_cg_run_ctx, retval), 2437 si->dst_reg, si->dst_reg, 2438 offsetof(struct bpf_cg_run_ctx, retval)); 2439 } 2440 break; 2441 case offsetof(struct bpf_sockopt, optval): 2442 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, optval); 2443 break; 2444 case offsetof(struct bpf_sockopt, optval_end): 2445 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, optval_end); 2446 break; 2447 } 2448 2449 return insn - insn_buf; 2450 } 2451 2452 static int cg_sockopt_get_prologue(struct bpf_insn *insn_buf, 2453 bool direct_write, 2454 const struct bpf_prog *prog) 2455 { 2456 /* Nothing to do for sockopt argument. The data is kzalloc'ated. 2457 */ 2458 return 0; 2459 } 2460 2461 const struct bpf_verifier_ops cg_sockopt_verifier_ops = { 2462 .get_func_proto = cg_sockopt_func_proto, 2463 .is_valid_access = cg_sockopt_is_valid_access, 2464 .convert_ctx_access = cg_sockopt_convert_ctx_access, 2465 .gen_prologue = cg_sockopt_get_prologue, 2466 }; 2467 2468 const struct bpf_prog_ops cg_sockopt_prog_ops = { 2469 }; 2470 2471 /* Common helpers for cgroup hooks. */ 2472 const struct bpf_func_proto * 2473 cgroup_common_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 2474 { 2475 switch (func_id) { 2476 case BPF_FUNC_get_local_storage: 2477 return &bpf_get_local_storage_proto; 2478 case BPF_FUNC_get_retval: 2479 switch (prog->expected_attach_type) { 2480 case BPF_CGROUP_INET_INGRESS: 2481 case BPF_CGROUP_INET_EGRESS: 2482 case BPF_CGROUP_SOCK_OPS: 2483 case BPF_CGROUP_UDP4_RECVMSG: 2484 case BPF_CGROUP_UDP6_RECVMSG: 2485 case BPF_CGROUP_INET4_GETPEERNAME: 2486 case BPF_CGROUP_INET6_GETPEERNAME: 2487 case BPF_CGROUP_INET4_GETSOCKNAME: 2488 case BPF_CGROUP_INET6_GETSOCKNAME: 2489 return NULL; 2490 default: 2491 return &bpf_get_retval_proto; 2492 } 2493 case BPF_FUNC_set_retval: 2494 switch (prog->expected_attach_type) { 2495 case BPF_CGROUP_INET_INGRESS: 2496 case BPF_CGROUP_INET_EGRESS: 2497 case BPF_CGROUP_SOCK_OPS: 2498 case BPF_CGROUP_UDP4_RECVMSG: 2499 case BPF_CGROUP_UDP6_RECVMSG: 2500 case BPF_CGROUP_INET4_GETPEERNAME: 2501 case BPF_CGROUP_INET6_GETPEERNAME: 2502 case BPF_CGROUP_INET4_GETSOCKNAME: 2503 case BPF_CGROUP_INET6_GETSOCKNAME: 2504 return NULL; 2505 default: 2506 return &bpf_set_retval_proto; 2507 } 2508 default: 2509 return NULL; 2510 } 2511 } 2512 2513 /* Common helpers for cgroup hooks with valid process context. */ 2514 const struct bpf_func_proto * 2515 cgroup_current_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 2516 { 2517 switch (func_id) { 2518 case BPF_FUNC_get_current_uid_gid: 2519 return &bpf_get_current_uid_gid_proto; 2520 case BPF_FUNC_get_current_pid_tgid: 2521 return &bpf_get_current_pid_tgid_proto; 2522 case BPF_FUNC_get_current_comm: 2523 return &bpf_get_current_comm_proto; 2524 case BPF_FUNC_get_current_cgroup_id: 2525 return &bpf_get_current_cgroup_id_proto; 2526 case BPF_FUNC_get_current_ancestor_cgroup_id: 2527 return &bpf_get_current_ancestor_cgroup_id_proto; 2528 #ifdef CONFIG_CGROUP_NET_CLASSID 2529 case BPF_FUNC_get_cgroup_classid: 2530 return &bpf_get_cgroup_classid_curr_proto; 2531 #endif 2532 default: 2533 return NULL; 2534 } 2535 } 2536