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 found: 925 BUG_ON(!cg); 926 progs = rcu_dereference_protected( 927 desc->bpf.effective[atype], 928 lockdep_is_held(&cgroup_mutex)); 929 930 /* Remove the program from the array */ 931 WARN_ONCE(bpf_prog_array_delete_safe_at(progs, pos), 932 "Failed to purge a prog from array at index %d", pos); 933 } 934 } 935 936 /** 937 * __cgroup_bpf_detach() - Detach the program or link from a cgroup, and 938 * propagate the change to descendants 939 * @cgrp: The cgroup which descendants to traverse 940 * @prog: A program to detach or NULL 941 * @link: A link to detach or NULL 942 * @type: Type of detach operation 943 * 944 * At most one of @prog or @link can be non-NULL. 945 * Must be called with cgroup_mutex held. 946 */ 947 static int __cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog, 948 struct bpf_cgroup_link *link, enum bpf_attach_type type) 949 { 950 enum cgroup_bpf_attach_type atype; 951 struct bpf_prog *old_prog; 952 struct bpf_prog_list *pl; 953 struct hlist_head *progs; 954 u32 attach_btf_id = 0; 955 u32 flags; 956 957 if (prog) 958 attach_btf_id = prog->aux->attach_btf_id; 959 if (link) 960 attach_btf_id = link->link.prog->aux->attach_btf_id; 961 962 atype = bpf_cgroup_atype_find(type, attach_btf_id); 963 if (atype < 0) 964 return -EINVAL; 965 966 progs = &cgrp->bpf.progs[atype]; 967 flags = cgrp->bpf.flags[atype]; 968 969 if (prog && link) 970 /* only one of prog or link can be specified */ 971 return -EINVAL; 972 973 pl = find_detach_entry(progs, prog, link, flags & BPF_F_ALLOW_MULTI); 974 if (IS_ERR(pl)) 975 return PTR_ERR(pl); 976 977 /* mark it deleted, so it's ignored while recomputing effective */ 978 old_prog = pl->prog; 979 pl->prog = NULL; 980 pl->link = NULL; 981 982 if (update_effective_progs(cgrp, atype)) { 983 /* if update effective array failed replace the prog with a dummy prog*/ 984 pl->prog = old_prog; 985 pl->link = link; 986 purge_effective_progs(cgrp, old_prog, link, atype); 987 } 988 989 /* now can actually delete it from this cgroup list */ 990 hlist_del(&pl->node); 991 992 kfree(pl); 993 if (hlist_empty(progs)) 994 /* last program was detached, reset flags to zero */ 995 cgrp->bpf.flags[atype] = 0; 996 if (old_prog) { 997 if (type == BPF_LSM_CGROUP) 998 bpf_trampoline_unlink_cgroup_shim(old_prog); 999 bpf_prog_put(old_prog); 1000 } 1001 static_branch_dec(&cgroup_bpf_enabled_key[atype]); 1002 return 0; 1003 } 1004 1005 static int cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog, 1006 enum bpf_attach_type type) 1007 { 1008 int ret; 1009 1010 mutex_lock(&cgroup_mutex); 1011 ret = __cgroup_bpf_detach(cgrp, prog, NULL, type); 1012 mutex_unlock(&cgroup_mutex); 1013 return ret; 1014 } 1015 1016 /* Must be called with cgroup_mutex held to avoid races. */ 1017 static int __cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr, 1018 union bpf_attr __user *uattr) 1019 { 1020 __u32 __user *prog_ids = u64_to_user_ptr(attr->query.prog_ids); 1021 enum bpf_attach_type type = attr->query.attach_type; 1022 enum cgroup_bpf_attach_type atype; 1023 struct bpf_prog_array *effective; 1024 struct hlist_head *progs; 1025 struct bpf_prog *prog; 1026 int cnt, ret = 0, i; 1027 u32 flags; 1028 1029 atype = to_cgroup_bpf_attach_type(type); 1030 if (atype < 0) 1031 return -EINVAL; 1032 1033 progs = &cgrp->bpf.progs[atype]; 1034 flags = cgrp->bpf.flags[atype]; 1035 1036 effective = rcu_dereference_protected(cgrp->bpf.effective[atype], 1037 lockdep_is_held(&cgroup_mutex)); 1038 1039 if (attr->query.query_flags & BPF_F_QUERY_EFFECTIVE) 1040 cnt = bpf_prog_array_length(effective); 1041 else 1042 cnt = prog_list_length(progs); 1043 1044 if (copy_to_user(&uattr->query.attach_flags, &flags, sizeof(flags))) 1045 return -EFAULT; 1046 if (copy_to_user(&uattr->query.prog_cnt, &cnt, sizeof(cnt))) 1047 return -EFAULT; 1048 if (attr->query.prog_cnt == 0 || !prog_ids || !cnt) 1049 /* return early if user requested only program count + flags */ 1050 return 0; 1051 if (attr->query.prog_cnt < cnt) { 1052 cnt = attr->query.prog_cnt; 1053 ret = -ENOSPC; 1054 } 1055 1056 if (attr->query.query_flags & BPF_F_QUERY_EFFECTIVE) { 1057 return bpf_prog_array_copy_to_user(effective, prog_ids, cnt); 1058 } else { 1059 struct bpf_prog_list *pl; 1060 u32 id; 1061 1062 i = 0; 1063 hlist_for_each_entry(pl, progs, node) { 1064 prog = prog_list_prog(pl); 1065 id = prog->aux->id; 1066 if (copy_to_user(prog_ids + i, &id, sizeof(id))) 1067 return -EFAULT; 1068 if (++i == cnt) 1069 break; 1070 } 1071 } 1072 return ret; 1073 } 1074 1075 static int cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr, 1076 union bpf_attr __user *uattr) 1077 { 1078 int ret; 1079 1080 mutex_lock(&cgroup_mutex); 1081 ret = __cgroup_bpf_query(cgrp, attr, uattr); 1082 mutex_unlock(&cgroup_mutex); 1083 return ret; 1084 } 1085 1086 int cgroup_bpf_prog_attach(const union bpf_attr *attr, 1087 enum bpf_prog_type ptype, struct bpf_prog *prog) 1088 { 1089 struct bpf_prog *replace_prog = NULL; 1090 struct cgroup *cgrp; 1091 int ret; 1092 1093 cgrp = cgroup_get_from_fd(attr->target_fd); 1094 if (IS_ERR(cgrp)) 1095 return PTR_ERR(cgrp); 1096 1097 if ((attr->attach_flags & BPF_F_ALLOW_MULTI) && 1098 (attr->attach_flags & BPF_F_REPLACE)) { 1099 replace_prog = bpf_prog_get_type(attr->replace_bpf_fd, ptype); 1100 if (IS_ERR(replace_prog)) { 1101 cgroup_put(cgrp); 1102 return PTR_ERR(replace_prog); 1103 } 1104 } 1105 1106 ret = cgroup_bpf_attach(cgrp, prog, replace_prog, NULL, 1107 attr->attach_type, attr->attach_flags); 1108 1109 if (replace_prog) 1110 bpf_prog_put(replace_prog); 1111 cgroup_put(cgrp); 1112 return ret; 1113 } 1114 1115 int cgroup_bpf_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype) 1116 { 1117 struct bpf_prog *prog; 1118 struct cgroup *cgrp; 1119 int ret; 1120 1121 cgrp = cgroup_get_from_fd(attr->target_fd); 1122 if (IS_ERR(cgrp)) 1123 return PTR_ERR(cgrp); 1124 1125 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype); 1126 if (IS_ERR(prog)) 1127 prog = NULL; 1128 1129 ret = cgroup_bpf_detach(cgrp, prog, attr->attach_type); 1130 if (prog) 1131 bpf_prog_put(prog); 1132 1133 cgroup_put(cgrp); 1134 return ret; 1135 } 1136 1137 static void bpf_cgroup_link_release(struct bpf_link *link) 1138 { 1139 struct bpf_cgroup_link *cg_link = 1140 container_of(link, struct bpf_cgroup_link, link); 1141 struct cgroup *cg; 1142 1143 /* link might have been auto-detached by dying cgroup already, 1144 * in that case our work is done here 1145 */ 1146 if (!cg_link->cgroup) 1147 return; 1148 1149 mutex_lock(&cgroup_mutex); 1150 1151 /* re-check cgroup under lock again */ 1152 if (!cg_link->cgroup) { 1153 mutex_unlock(&cgroup_mutex); 1154 return; 1155 } 1156 1157 WARN_ON(__cgroup_bpf_detach(cg_link->cgroup, NULL, cg_link, 1158 cg_link->type)); 1159 if (cg_link->type == BPF_LSM_CGROUP) 1160 bpf_trampoline_unlink_cgroup_shim(cg_link->link.prog); 1161 1162 cg = cg_link->cgroup; 1163 cg_link->cgroup = NULL; 1164 1165 mutex_unlock(&cgroup_mutex); 1166 1167 cgroup_put(cg); 1168 } 1169 1170 static void bpf_cgroup_link_dealloc(struct bpf_link *link) 1171 { 1172 struct bpf_cgroup_link *cg_link = 1173 container_of(link, struct bpf_cgroup_link, link); 1174 1175 kfree(cg_link); 1176 } 1177 1178 static int bpf_cgroup_link_detach(struct bpf_link *link) 1179 { 1180 bpf_cgroup_link_release(link); 1181 1182 return 0; 1183 } 1184 1185 static void bpf_cgroup_link_show_fdinfo(const struct bpf_link *link, 1186 struct seq_file *seq) 1187 { 1188 struct bpf_cgroup_link *cg_link = 1189 container_of(link, struct bpf_cgroup_link, link); 1190 u64 cg_id = 0; 1191 1192 mutex_lock(&cgroup_mutex); 1193 if (cg_link->cgroup) 1194 cg_id = cgroup_id(cg_link->cgroup); 1195 mutex_unlock(&cgroup_mutex); 1196 1197 seq_printf(seq, 1198 "cgroup_id:\t%llu\n" 1199 "attach_type:\t%d\n", 1200 cg_id, 1201 cg_link->type); 1202 } 1203 1204 static int bpf_cgroup_link_fill_link_info(const struct bpf_link *link, 1205 struct bpf_link_info *info) 1206 { 1207 struct bpf_cgroup_link *cg_link = 1208 container_of(link, struct bpf_cgroup_link, link); 1209 u64 cg_id = 0; 1210 1211 mutex_lock(&cgroup_mutex); 1212 if (cg_link->cgroup) 1213 cg_id = cgroup_id(cg_link->cgroup); 1214 mutex_unlock(&cgroup_mutex); 1215 1216 info->cgroup.cgroup_id = cg_id; 1217 info->cgroup.attach_type = cg_link->type; 1218 return 0; 1219 } 1220 1221 static const struct bpf_link_ops bpf_cgroup_link_lops = { 1222 .release = bpf_cgroup_link_release, 1223 .dealloc = bpf_cgroup_link_dealloc, 1224 .detach = bpf_cgroup_link_detach, 1225 .update_prog = cgroup_bpf_replace, 1226 .show_fdinfo = bpf_cgroup_link_show_fdinfo, 1227 .fill_link_info = bpf_cgroup_link_fill_link_info, 1228 }; 1229 1230 int cgroup_bpf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 1231 { 1232 struct bpf_link_primer link_primer; 1233 struct bpf_cgroup_link *link; 1234 struct cgroup *cgrp; 1235 int err; 1236 1237 if (attr->link_create.flags) 1238 return -EINVAL; 1239 1240 cgrp = cgroup_get_from_fd(attr->link_create.target_fd); 1241 if (IS_ERR(cgrp)) 1242 return PTR_ERR(cgrp); 1243 1244 link = kzalloc(sizeof(*link), GFP_USER); 1245 if (!link) { 1246 err = -ENOMEM; 1247 goto out_put_cgroup; 1248 } 1249 bpf_link_init(&link->link, BPF_LINK_TYPE_CGROUP, &bpf_cgroup_link_lops, 1250 prog); 1251 link->cgroup = cgrp; 1252 link->type = attr->link_create.attach_type; 1253 1254 err = bpf_link_prime(&link->link, &link_primer); 1255 if (err) { 1256 kfree(link); 1257 goto out_put_cgroup; 1258 } 1259 1260 err = cgroup_bpf_attach(cgrp, NULL, NULL, link, 1261 link->type, BPF_F_ALLOW_MULTI); 1262 if (err) { 1263 bpf_link_cleanup(&link_primer); 1264 goto out_put_cgroup; 1265 } 1266 1267 return bpf_link_settle(&link_primer); 1268 1269 out_put_cgroup: 1270 cgroup_put(cgrp); 1271 return err; 1272 } 1273 1274 int cgroup_bpf_prog_query(const union bpf_attr *attr, 1275 union bpf_attr __user *uattr) 1276 { 1277 struct cgroup *cgrp; 1278 int ret; 1279 1280 cgrp = cgroup_get_from_fd(attr->query.target_fd); 1281 if (IS_ERR(cgrp)) 1282 return PTR_ERR(cgrp); 1283 1284 ret = cgroup_bpf_query(cgrp, attr, uattr); 1285 1286 cgroup_put(cgrp); 1287 return ret; 1288 } 1289 1290 /** 1291 * __cgroup_bpf_run_filter_skb() - Run a program for packet filtering 1292 * @sk: The socket sending or receiving traffic 1293 * @skb: The skb that is being sent or received 1294 * @type: The type of program to be executed 1295 * 1296 * If no socket is passed, or the socket is not of type INET or INET6, 1297 * this function does nothing and returns 0. 1298 * 1299 * The program type passed in via @type must be suitable for network 1300 * filtering. No further check is performed to assert that. 1301 * 1302 * For egress packets, this function can return: 1303 * NET_XMIT_SUCCESS (0) - continue with packet output 1304 * NET_XMIT_DROP (1) - drop packet and notify TCP to call cwr 1305 * NET_XMIT_CN (2) - continue with packet output and notify TCP 1306 * to call cwr 1307 * -err - drop packet 1308 * 1309 * For ingress packets, this function will return -EPERM if any 1310 * attached program was found and if it returned != 1 during execution. 1311 * Otherwise 0 is returned. 1312 */ 1313 int __cgroup_bpf_run_filter_skb(struct sock *sk, 1314 struct sk_buff *skb, 1315 enum cgroup_bpf_attach_type atype) 1316 { 1317 unsigned int offset = skb->data - skb_network_header(skb); 1318 struct sock *save_sk; 1319 void *saved_data_end; 1320 struct cgroup *cgrp; 1321 int ret; 1322 1323 if (!sk || !sk_fullsock(sk)) 1324 return 0; 1325 1326 if (sk->sk_family != AF_INET && sk->sk_family != AF_INET6) 1327 return 0; 1328 1329 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 1330 save_sk = skb->sk; 1331 skb->sk = sk; 1332 __skb_push(skb, offset); 1333 1334 /* compute pointers for the bpf prog */ 1335 bpf_compute_and_save_data_end(skb, &saved_data_end); 1336 1337 if (atype == CGROUP_INET_EGRESS) { 1338 u32 flags = 0; 1339 bool cn; 1340 1341 ret = bpf_prog_run_array_cg(&cgrp->bpf, atype, skb, 1342 __bpf_prog_run_save_cb, 0, &flags); 1343 1344 /* Return values of CGROUP EGRESS BPF programs are: 1345 * 0: drop packet 1346 * 1: keep packet 1347 * 2: drop packet and cn 1348 * 3: keep packet and cn 1349 * 1350 * The returned value is then converted to one of the NET_XMIT 1351 * or an error code that is then interpreted as drop packet 1352 * (and no cn): 1353 * 0: NET_XMIT_SUCCESS skb should be transmitted 1354 * 1: NET_XMIT_DROP skb should be dropped and cn 1355 * 2: NET_XMIT_CN skb should be transmitted and cn 1356 * 3: -err skb should be dropped 1357 */ 1358 1359 cn = flags & BPF_RET_SET_CN; 1360 if (ret && !IS_ERR_VALUE((long)ret)) 1361 ret = -EFAULT; 1362 if (!ret) 1363 ret = (cn ? NET_XMIT_CN : NET_XMIT_SUCCESS); 1364 else 1365 ret = (cn ? NET_XMIT_DROP : ret); 1366 } else { 1367 ret = bpf_prog_run_array_cg(&cgrp->bpf, atype, 1368 skb, __bpf_prog_run_save_cb, 0, 1369 NULL); 1370 if (ret && !IS_ERR_VALUE((long)ret)) 1371 ret = -EFAULT; 1372 } 1373 bpf_restore_data_end(skb, saved_data_end); 1374 __skb_pull(skb, offset); 1375 skb->sk = save_sk; 1376 1377 return ret; 1378 } 1379 EXPORT_SYMBOL(__cgroup_bpf_run_filter_skb); 1380 1381 /** 1382 * __cgroup_bpf_run_filter_sk() - Run a program on a sock 1383 * @sk: sock structure to manipulate 1384 * @type: The type of program to be executed 1385 * 1386 * socket is passed is expected to be of type INET or INET6. 1387 * 1388 * The program type passed in via @type must be suitable for sock 1389 * filtering. No further check is performed to assert that. 1390 * 1391 * This function will return %-EPERM if any if an attached program was found 1392 * and if it returned != 1 during execution. In all other cases, 0 is returned. 1393 */ 1394 int __cgroup_bpf_run_filter_sk(struct sock *sk, 1395 enum cgroup_bpf_attach_type atype) 1396 { 1397 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 1398 1399 return bpf_prog_run_array_cg(&cgrp->bpf, atype, sk, bpf_prog_run, 0, 1400 NULL); 1401 } 1402 EXPORT_SYMBOL(__cgroup_bpf_run_filter_sk); 1403 1404 /** 1405 * __cgroup_bpf_run_filter_sock_addr() - Run a program on a sock and 1406 * provided by user sockaddr 1407 * @sk: sock struct that will use sockaddr 1408 * @uaddr: sockaddr struct provided by user 1409 * @type: The type of program to be executed 1410 * @t_ctx: Pointer to attach type specific context 1411 * @flags: Pointer to u32 which contains higher bits of BPF program 1412 * return value (OR'ed together). 1413 * 1414 * socket is expected to be of type INET or INET6. 1415 * 1416 * This function will return %-EPERM if an attached program is found and 1417 * returned value != 1 during execution. In all other cases, 0 is returned. 1418 */ 1419 int __cgroup_bpf_run_filter_sock_addr(struct sock *sk, 1420 struct sockaddr *uaddr, 1421 enum cgroup_bpf_attach_type atype, 1422 void *t_ctx, 1423 u32 *flags) 1424 { 1425 struct bpf_sock_addr_kern ctx = { 1426 .sk = sk, 1427 .uaddr = uaddr, 1428 .t_ctx = t_ctx, 1429 }; 1430 struct sockaddr_storage unspec; 1431 struct cgroup *cgrp; 1432 1433 /* Check socket family since not all sockets represent network 1434 * endpoint (e.g. AF_UNIX). 1435 */ 1436 if (sk->sk_family != AF_INET && sk->sk_family != AF_INET6) 1437 return 0; 1438 1439 if (!ctx.uaddr) { 1440 memset(&unspec, 0, sizeof(unspec)); 1441 ctx.uaddr = (struct sockaddr *)&unspec; 1442 } 1443 1444 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 1445 return bpf_prog_run_array_cg(&cgrp->bpf, atype, &ctx, bpf_prog_run, 1446 0, flags); 1447 } 1448 EXPORT_SYMBOL(__cgroup_bpf_run_filter_sock_addr); 1449 1450 /** 1451 * __cgroup_bpf_run_filter_sock_ops() - Run a program on a sock 1452 * @sk: socket to get cgroup from 1453 * @sock_ops: bpf_sock_ops_kern struct to pass to program. Contains 1454 * sk with connection information (IP addresses, etc.) May not contain 1455 * cgroup info if it is a req sock. 1456 * @type: The type of program to be executed 1457 * 1458 * socket passed is expected to be of type INET or INET6. 1459 * 1460 * The program type passed in via @type must be suitable for sock_ops 1461 * filtering. No further check is performed to assert that. 1462 * 1463 * This function will return %-EPERM if any if an attached program was found 1464 * and if it returned != 1 during execution. In all other cases, 0 is returned. 1465 */ 1466 int __cgroup_bpf_run_filter_sock_ops(struct sock *sk, 1467 struct bpf_sock_ops_kern *sock_ops, 1468 enum cgroup_bpf_attach_type atype) 1469 { 1470 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 1471 1472 return bpf_prog_run_array_cg(&cgrp->bpf, atype, sock_ops, bpf_prog_run, 1473 0, NULL); 1474 } 1475 EXPORT_SYMBOL(__cgroup_bpf_run_filter_sock_ops); 1476 1477 int __cgroup_bpf_check_dev_permission(short dev_type, u32 major, u32 minor, 1478 short access, enum cgroup_bpf_attach_type atype) 1479 { 1480 struct cgroup *cgrp; 1481 struct bpf_cgroup_dev_ctx ctx = { 1482 .access_type = (access << 16) | dev_type, 1483 .major = major, 1484 .minor = minor, 1485 }; 1486 int ret; 1487 1488 rcu_read_lock(); 1489 cgrp = task_dfl_cgroup(current); 1490 ret = bpf_prog_run_array_cg(&cgrp->bpf, atype, &ctx, bpf_prog_run, 0, 1491 NULL); 1492 rcu_read_unlock(); 1493 1494 return ret; 1495 } 1496 1497 BPF_CALL_0(bpf_get_retval) 1498 { 1499 struct bpf_cg_run_ctx *ctx = 1500 container_of(current->bpf_ctx, struct bpf_cg_run_ctx, run_ctx); 1501 1502 return ctx->retval; 1503 } 1504 1505 const struct bpf_func_proto bpf_get_retval_proto = { 1506 .func = bpf_get_retval, 1507 .gpl_only = false, 1508 .ret_type = RET_INTEGER, 1509 }; 1510 1511 BPF_CALL_1(bpf_set_retval, int, retval) 1512 { 1513 struct bpf_cg_run_ctx *ctx = 1514 container_of(current->bpf_ctx, struct bpf_cg_run_ctx, run_ctx); 1515 1516 ctx->retval = retval; 1517 return 0; 1518 } 1519 1520 const struct bpf_func_proto bpf_set_retval_proto = { 1521 .func = bpf_set_retval, 1522 .gpl_only = false, 1523 .ret_type = RET_INTEGER, 1524 .arg1_type = ARG_ANYTHING, 1525 }; 1526 1527 static const struct bpf_func_proto * 1528 cgroup_base_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 1529 { 1530 switch (func_id) { 1531 case BPF_FUNC_get_current_uid_gid: 1532 return &bpf_get_current_uid_gid_proto; 1533 case BPF_FUNC_get_local_storage: 1534 return &bpf_get_local_storage_proto; 1535 case BPF_FUNC_get_current_cgroup_id: 1536 return &bpf_get_current_cgroup_id_proto; 1537 case BPF_FUNC_perf_event_output: 1538 return &bpf_event_output_data_proto; 1539 case BPF_FUNC_get_retval: 1540 return &bpf_get_retval_proto; 1541 case BPF_FUNC_set_retval: 1542 return &bpf_set_retval_proto; 1543 default: 1544 return bpf_base_func_proto(func_id); 1545 } 1546 } 1547 1548 static const struct bpf_func_proto * 1549 cgroup_dev_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 1550 { 1551 return cgroup_base_func_proto(func_id, prog); 1552 } 1553 1554 static bool cgroup_dev_is_valid_access(int off, int size, 1555 enum bpf_access_type type, 1556 const struct bpf_prog *prog, 1557 struct bpf_insn_access_aux *info) 1558 { 1559 const int size_default = sizeof(__u32); 1560 1561 if (type == BPF_WRITE) 1562 return false; 1563 1564 if (off < 0 || off + size > sizeof(struct bpf_cgroup_dev_ctx)) 1565 return false; 1566 /* The verifier guarantees that size > 0. */ 1567 if (off % size != 0) 1568 return false; 1569 1570 switch (off) { 1571 case bpf_ctx_range(struct bpf_cgroup_dev_ctx, access_type): 1572 bpf_ctx_record_field_size(info, size_default); 1573 if (!bpf_ctx_narrow_access_ok(off, size, size_default)) 1574 return false; 1575 break; 1576 default: 1577 if (size != size_default) 1578 return false; 1579 } 1580 1581 return true; 1582 } 1583 1584 const struct bpf_prog_ops cg_dev_prog_ops = { 1585 }; 1586 1587 const struct bpf_verifier_ops cg_dev_verifier_ops = { 1588 .get_func_proto = cgroup_dev_func_proto, 1589 .is_valid_access = cgroup_dev_is_valid_access, 1590 }; 1591 1592 /** 1593 * __cgroup_bpf_run_filter_sysctl - Run a program on sysctl 1594 * 1595 * @head: sysctl table header 1596 * @table: sysctl table 1597 * @write: sysctl is being read (= 0) or written (= 1) 1598 * @buf: pointer to buffer (in and out) 1599 * @pcount: value-result argument: value is size of buffer pointed to by @buf, 1600 * result is size of @new_buf if program set new value, initial value 1601 * otherwise 1602 * @ppos: value-result argument: value is position at which read from or write 1603 * to sysctl is happening, result is new position if program overrode it, 1604 * initial value otherwise 1605 * @type: type of program to be executed 1606 * 1607 * Program is run when sysctl is being accessed, either read or written, and 1608 * can allow or deny such access. 1609 * 1610 * This function will return %-EPERM if an attached program is found and 1611 * returned value != 1 during execution. In all other cases 0 is returned. 1612 */ 1613 int __cgroup_bpf_run_filter_sysctl(struct ctl_table_header *head, 1614 struct ctl_table *table, int write, 1615 char **buf, size_t *pcount, loff_t *ppos, 1616 enum cgroup_bpf_attach_type atype) 1617 { 1618 struct bpf_sysctl_kern ctx = { 1619 .head = head, 1620 .table = table, 1621 .write = write, 1622 .ppos = ppos, 1623 .cur_val = NULL, 1624 .cur_len = PAGE_SIZE, 1625 .new_val = NULL, 1626 .new_len = 0, 1627 .new_updated = 0, 1628 }; 1629 struct cgroup *cgrp; 1630 loff_t pos = 0; 1631 int ret; 1632 1633 ctx.cur_val = kmalloc_track_caller(ctx.cur_len, GFP_KERNEL); 1634 if (!ctx.cur_val || 1635 table->proc_handler(table, 0, ctx.cur_val, &ctx.cur_len, &pos)) { 1636 /* Let BPF program decide how to proceed. */ 1637 ctx.cur_len = 0; 1638 } 1639 1640 if (write && *buf && *pcount) { 1641 /* BPF program should be able to override new value with a 1642 * buffer bigger than provided by user. 1643 */ 1644 ctx.new_val = kmalloc_track_caller(PAGE_SIZE, GFP_KERNEL); 1645 ctx.new_len = min_t(size_t, PAGE_SIZE, *pcount); 1646 if (ctx.new_val) { 1647 memcpy(ctx.new_val, *buf, ctx.new_len); 1648 } else { 1649 /* Let BPF program decide how to proceed. */ 1650 ctx.new_len = 0; 1651 } 1652 } 1653 1654 rcu_read_lock(); 1655 cgrp = task_dfl_cgroup(current); 1656 ret = bpf_prog_run_array_cg(&cgrp->bpf, atype, &ctx, bpf_prog_run, 0, 1657 NULL); 1658 rcu_read_unlock(); 1659 1660 kfree(ctx.cur_val); 1661 1662 if (ret == 1 && ctx.new_updated) { 1663 kfree(*buf); 1664 *buf = ctx.new_val; 1665 *pcount = ctx.new_len; 1666 } else { 1667 kfree(ctx.new_val); 1668 } 1669 1670 return ret; 1671 } 1672 1673 #ifdef CONFIG_NET 1674 static int sockopt_alloc_buf(struct bpf_sockopt_kern *ctx, int max_optlen, 1675 struct bpf_sockopt_buf *buf) 1676 { 1677 if (unlikely(max_optlen < 0)) 1678 return -EINVAL; 1679 1680 if (unlikely(max_optlen > PAGE_SIZE)) { 1681 /* We don't expose optvals that are greater than PAGE_SIZE 1682 * to the BPF program. 1683 */ 1684 max_optlen = PAGE_SIZE; 1685 } 1686 1687 if (max_optlen <= sizeof(buf->data)) { 1688 /* When the optval fits into BPF_SOCKOPT_KERN_BUF_SIZE 1689 * bytes avoid the cost of kzalloc. 1690 */ 1691 ctx->optval = buf->data; 1692 ctx->optval_end = ctx->optval + max_optlen; 1693 return max_optlen; 1694 } 1695 1696 ctx->optval = kzalloc(max_optlen, GFP_USER); 1697 if (!ctx->optval) 1698 return -ENOMEM; 1699 1700 ctx->optval_end = ctx->optval + max_optlen; 1701 1702 return max_optlen; 1703 } 1704 1705 static void sockopt_free_buf(struct bpf_sockopt_kern *ctx, 1706 struct bpf_sockopt_buf *buf) 1707 { 1708 if (ctx->optval == buf->data) 1709 return; 1710 kfree(ctx->optval); 1711 } 1712 1713 static bool sockopt_buf_allocated(struct bpf_sockopt_kern *ctx, 1714 struct bpf_sockopt_buf *buf) 1715 { 1716 return ctx->optval != buf->data; 1717 } 1718 1719 int __cgroup_bpf_run_filter_setsockopt(struct sock *sk, int *level, 1720 int *optname, char __user *optval, 1721 int *optlen, char **kernel_optval) 1722 { 1723 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 1724 struct bpf_sockopt_buf buf = {}; 1725 struct bpf_sockopt_kern ctx = { 1726 .sk = sk, 1727 .level = *level, 1728 .optname = *optname, 1729 }; 1730 int ret, max_optlen; 1731 1732 /* Allocate a bit more than the initial user buffer for 1733 * BPF program. The canonical use case is overriding 1734 * TCP_CONGESTION(nv) to TCP_CONGESTION(cubic). 1735 */ 1736 max_optlen = max_t(int, 16, *optlen); 1737 max_optlen = sockopt_alloc_buf(&ctx, max_optlen, &buf); 1738 if (max_optlen < 0) 1739 return max_optlen; 1740 1741 ctx.optlen = *optlen; 1742 1743 if (copy_from_user(ctx.optval, optval, min(*optlen, max_optlen)) != 0) { 1744 ret = -EFAULT; 1745 goto out; 1746 } 1747 1748 lock_sock(sk); 1749 ret = bpf_prog_run_array_cg(&cgrp->bpf, CGROUP_SETSOCKOPT, 1750 &ctx, bpf_prog_run, 0, NULL); 1751 release_sock(sk); 1752 1753 if (ret) 1754 goto out; 1755 1756 if (ctx.optlen == -1) { 1757 /* optlen set to -1, bypass kernel */ 1758 ret = 1; 1759 } else if (ctx.optlen > max_optlen || ctx.optlen < -1) { 1760 /* optlen is out of bounds */ 1761 ret = -EFAULT; 1762 } else { 1763 /* optlen within bounds, run kernel handler */ 1764 ret = 0; 1765 1766 /* export any potential modifications */ 1767 *level = ctx.level; 1768 *optname = ctx.optname; 1769 1770 /* optlen == 0 from BPF indicates that we should 1771 * use original userspace data. 1772 */ 1773 if (ctx.optlen != 0) { 1774 *optlen = ctx.optlen; 1775 /* We've used bpf_sockopt_kern->buf as an intermediary 1776 * storage, but the BPF program indicates that we need 1777 * to pass this data to the kernel setsockopt handler. 1778 * No way to export on-stack buf, have to allocate a 1779 * new buffer. 1780 */ 1781 if (!sockopt_buf_allocated(&ctx, &buf)) { 1782 void *p = kmalloc(ctx.optlen, GFP_USER); 1783 1784 if (!p) { 1785 ret = -ENOMEM; 1786 goto out; 1787 } 1788 memcpy(p, ctx.optval, ctx.optlen); 1789 *kernel_optval = p; 1790 } else { 1791 *kernel_optval = ctx.optval; 1792 } 1793 /* export and don't free sockopt buf */ 1794 return 0; 1795 } 1796 } 1797 1798 out: 1799 sockopt_free_buf(&ctx, &buf); 1800 return ret; 1801 } 1802 1803 int __cgroup_bpf_run_filter_getsockopt(struct sock *sk, int level, 1804 int optname, char __user *optval, 1805 int __user *optlen, int max_optlen, 1806 int retval) 1807 { 1808 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 1809 struct bpf_sockopt_buf buf = {}; 1810 struct bpf_sockopt_kern ctx = { 1811 .sk = sk, 1812 .level = level, 1813 .optname = optname, 1814 .current_task = current, 1815 }; 1816 int ret; 1817 1818 ctx.optlen = max_optlen; 1819 max_optlen = sockopt_alloc_buf(&ctx, max_optlen, &buf); 1820 if (max_optlen < 0) 1821 return max_optlen; 1822 1823 if (!retval) { 1824 /* If kernel getsockopt finished successfully, 1825 * copy whatever was returned to the user back 1826 * into our temporary buffer. Set optlen to the 1827 * one that kernel returned as well to let 1828 * BPF programs inspect the value. 1829 */ 1830 1831 if (get_user(ctx.optlen, optlen)) { 1832 ret = -EFAULT; 1833 goto out; 1834 } 1835 1836 if (ctx.optlen < 0) { 1837 ret = -EFAULT; 1838 goto out; 1839 } 1840 1841 if (copy_from_user(ctx.optval, optval, 1842 min(ctx.optlen, max_optlen)) != 0) { 1843 ret = -EFAULT; 1844 goto out; 1845 } 1846 } 1847 1848 lock_sock(sk); 1849 ret = bpf_prog_run_array_cg(&cgrp->bpf, CGROUP_GETSOCKOPT, 1850 &ctx, bpf_prog_run, retval, NULL); 1851 release_sock(sk); 1852 1853 if (ret < 0) 1854 goto out; 1855 1856 if (ctx.optlen > max_optlen || ctx.optlen < 0) { 1857 ret = -EFAULT; 1858 goto out; 1859 } 1860 1861 if (ctx.optlen != 0) { 1862 if (copy_to_user(optval, ctx.optval, ctx.optlen) || 1863 put_user(ctx.optlen, optlen)) { 1864 ret = -EFAULT; 1865 goto out; 1866 } 1867 } 1868 1869 out: 1870 sockopt_free_buf(&ctx, &buf); 1871 return ret; 1872 } 1873 1874 int __cgroup_bpf_run_filter_getsockopt_kern(struct sock *sk, int level, 1875 int optname, void *optval, 1876 int *optlen, int retval) 1877 { 1878 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 1879 struct bpf_sockopt_kern ctx = { 1880 .sk = sk, 1881 .level = level, 1882 .optname = optname, 1883 .optlen = *optlen, 1884 .optval = optval, 1885 .optval_end = optval + *optlen, 1886 .current_task = current, 1887 }; 1888 int ret; 1889 1890 /* Note that __cgroup_bpf_run_filter_getsockopt doesn't copy 1891 * user data back into BPF buffer when reval != 0. This is 1892 * done as an optimization to avoid extra copy, assuming 1893 * kernel won't populate the data in case of an error. 1894 * Here we always pass the data and memset() should 1895 * be called if that data shouldn't be "exported". 1896 */ 1897 1898 ret = bpf_prog_run_array_cg(&cgrp->bpf, CGROUP_GETSOCKOPT, 1899 &ctx, bpf_prog_run, retval, NULL); 1900 if (ret < 0) 1901 return ret; 1902 1903 if (ctx.optlen > *optlen) 1904 return -EFAULT; 1905 1906 /* BPF programs can shrink the buffer, export the modifications. 1907 */ 1908 if (ctx.optlen != 0) 1909 *optlen = ctx.optlen; 1910 1911 return ret; 1912 } 1913 #endif 1914 1915 static ssize_t sysctl_cpy_dir(const struct ctl_dir *dir, char **bufp, 1916 size_t *lenp) 1917 { 1918 ssize_t tmp_ret = 0, ret; 1919 1920 if (dir->header.parent) { 1921 tmp_ret = sysctl_cpy_dir(dir->header.parent, bufp, lenp); 1922 if (tmp_ret < 0) 1923 return tmp_ret; 1924 } 1925 1926 ret = strscpy(*bufp, dir->header.ctl_table[0].procname, *lenp); 1927 if (ret < 0) 1928 return ret; 1929 *bufp += ret; 1930 *lenp -= ret; 1931 ret += tmp_ret; 1932 1933 /* Avoid leading slash. */ 1934 if (!ret) 1935 return ret; 1936 1937 tmp_ret = strscpy(*bufp, "/", *lenp); 1938 if (tmp_ret < 0) 1939 return tmp_ret; 1940 *bufp += tmp_ret; 1941 *lenp -= tmp_ret; 1942 1943 return ret + tmp_ret; 1944 } 1945 1946 BPF_CALL_4(bpf_sysctl_get_name, struct bpf_sysctl_kern *, ctx, char *, buf, 1947 size_t, buf_len, u64, flags) 1948 { 1949 ssize_t tmp_ret = 0, ret; 1950 1951 if (!buf) 1952 return -EINVAL; 1953 1954 if (!(flags & BPF_F_SYSCTL_BASE_NAME)) { 1955 if (!ctx->head) 1956 return -EINVAL; 1957 tmp_ret = sysctl_cpy_dir(ctx->head->parent, &buf, &buf_len); 1958 if (tmp_ret < 0) 1959 return tmp_ret; 1960 } 1961 1962 ret = strscpy(buf, ctx->table->procname, buf_len); 1963 1964 return ret < 0 ? ret : tmp_ret + ret; 1965 } 1966 1967 static const struct bpf_func_proto bpf_sysctl_get_name_proto = { 1968 .func = bpf_sysctl_get_name, 1969 .gpl_only = false, 1970 .ret_type = RET_INTEGER, 1971 .arg1_type = ARG_PTR_TO_CTX, 1972 .arg2_type = ARG_PTR_TO_MEM, 1973 .arg3_type = ARG_CONST_SIZE, 1974 .arg4_type = ARG_ANYTHING, 1975 }; 1976 1977 static int copy_sysctl_value(char *dst, size_t dst_len, char *src, 1978 size_t src_len) 1979 { 1980 if (!dst) 1981 return -EINVAL; 1982 1983 if (!dst_len) 1984 return -E2BIG; 1985 1986 if (!src || !src_len) { 1987 memset(dst, 0, dst_len); 1988 return -EINVAL; 1989 } 1990 1991 memcpy(dst, src, min(dst_len, src_len)); 1992 1993 if (dst_len > src_len) { 1994 memset(dst + src_len, '\0', dst_len - src_len); 1995 return src_len; 1996 } 1997 1998 dst[dst_len - 1] = '\0'; 1999 2000 return -E2BIG; 2001 } 2002 2003 BPF_CALL_3(bpf_sysctl_get_current_value, struct bpf_sysctl_kern *, ctx, 2004 char *, buf, size_t, buf_len) 2005 { 2006 return copy_sysctl_value(buf, buf_len, ctx->cur_val, ctx->cur_len); 2007 } 2008 2009 static const struct bpf_func_proto bpf_sysctl_get_current_value_proto = { 2010 .func = bpf_sysctl_get_current_value, 2011 .gpl_only = false, 2012 .ret_type = RET_INTEGER, 2013 .arg1_type = ARG_PTR_TO_CTX, 2014 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 2015 .arg3_type = ARG_CONST_SIZE, 2016 }; 2017 2018 BPF_CALL_3(bpf_sysctl_get_new_value, struct bpf_sysctl_kern *, ctx, char *, buf, 2019 size_t, buf_len) 2020 { 2021 if (!ctx->write) { 2022 if (buf && buf_len) 2023 memset(buf, '\0', buf_len); 2024 return -EINVAL; 2025 } 2026 return copy_sysctl_value(buf, buf_len, ctx->new_val, ctx->new_len); 2027 } 2028 2029 static const struct bpf_func_proto bpf_sysctl_get_new_value_proto = { 2030 .func = bpf_sysctl_get_new_value, 2031 .gpl_only = false, 2032 .ret_type = RET_INTEGER, 2033 .arg1_type = ARG_PTR_TO_CTX, 2034 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 2035 .arg3_type = ARG_CONST_SIZE, 2036 }; 2037 2038 BPF_CALL_3(bpf_sysctl_set_new_value, struct bpf_sysctl_kern *, ctx, 2039 const char *, buf, size_t, buf_len) 2040 { 2041 if (!ctx->write || !ctx->new_val || !ctx->new_len || !buf || !buf_len) 2042 return -EINVAL; 2043 2044 if (buf_len > PAGE_SIZE - 1) 2045 return -E2BIG; 2046 2047 memcpy(ctx->new_val, buf, buf_len); 2048 ctx->new_len = buf_len; 2049 ctx->new_updated = 1; 2050 2051 return 0; 2052 } 2053 2054 static const struct bpf_func_proto bpf_sysctl_set_new_value_proto = { 2055 .func = bpf_sysctl_set_new_value, 2056 .gpl_only = false, 2057 .ret_type = RET_INTEGER, 2058 .arg1_type = ARG_PTR_TO_CTX, 2059 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 2060 .arg3_type = ARG_CONST_SIZE, 2061 }; 2062 2063 static const struct bpf_func_proto * 2064 sysctl_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 2065 { 2066 switch (func_id) { 2067 case BPF_FUNC_strtol: 2068 return &bpf_strtol_proto; 2069 case BPF_FUNC_strtoul: 2070 return &bpf_strtoul_proto; 2071 case BPF_FUNC_sysctl_get_name: 2072 return &bpf_sysctl_get_name_proto; 2073 case BPF_FUNC_sysctl_get_current_value: 2074 return &bpf_sysctl_get_current_value_proto; 2075 case BPF_FUNC_sysctl_get_new_value: 2076 return &bpf_sysctl_get_new_value_proto; 2077 case BPF_FUNC_sysctl_set_new_value: 2078 return &bpf_sysctl_set_new_value_proto; 2079 case BPF_FUNC_ktime_get_coarse_ns: 2080 return &bpf_ktime_get_coarse_ns_proto; 2081 default: 2082 return cgroup_base_func_proto(func_id, prog); 2083 } 2084 } 2085 2086 static bool sysctl_is_valid_access(int off, int size, enum bpf_access_type type, 2087 const struct bpf_prog *prog, 2088 struct bpf_insn_access_aux *info) 2089 { 2090 const int size_default = sizeof(__u32); 2091 2092 if (off < 0 || off + size > sizeof(struct bpf_sysctl) || off % size) 2093 return false; 2094 2095 switch (off) { 2096 case bpf_ctx_range(struct bpf_sysctl, write): 2097 if (type != BPF_READ) 2098 return false; 2099 bpf_ctx_record_field_size(info, size_default); 2100 return bpf_ctx_narrow_access_ok(off, size, size_default); 2101 case bpf_ctx_range(struct bpf_sysctl, file_pos): 2102 if (type == BPF_READ) { 2103 bpf_ctx_record_field_size(info, size_default); 2104 return bpf_ctx_narrow_access_ok(off, size, size_default); 2105 } else { 2106 return size == size_default; 2107 } 2108 default: 2109 return false; 2110 } 2111 } 2112 2113 static u32 sysctl_convert_ctx_access(enum bpf_access_type type, 2114 const struct bpf_insn *si, 2115 struct bpf_insn *insn_buf, 2116 struct bpf_prog *prog, u32 *target_size) 2117 { 2118 struct bpf_insn *insn = insn_buf; 2119 u32 read_size; 2120 2121 switch (si->off) { 2122 case offsetof(struct bpf_sysctl, write): 2123 *insn++ = BPF_LDX_MEM( 2124 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 2125 bpf_target_off(struct bpf_sysctl_kern, write, 2126 sizeof_field(struct bpf_sysctl_kern, 2127 write), 2128 target_size)); 2129 break; 2130 case offsetof(struct bpf_sysctl, file_pos): 2131 /* ppos is a pointer so it should be accessed via indirect 2132 * loads and stores. Also for stores additional temporary 2133 * register is used since neither src_reg nor dst_reg can be 2134 * overridden. 2135 */ 2136 if (type == BPF_WRITE) { 2137 int treg = BPF_REG_9; 2138 2139 if (si->src_reg == treg || si->dst_reg == treg) 2140 --treg; 2141 if (si->src_reg == treg || si->dst_reg == treg) 2142 --treg; 2143 *insn++ = BPF_STX_MEM( 2144 BPF_DW, si->dst_reg, treg, 2145 offsetof(struct bpf_sysctl_kern, tmp_reg)); 2146 *insn++ = BPF_LDX_MEM( 2147 BPF_FIELD_SIZEOF(struct bpf_sysctl_kern, ppos), 2148 treg, si->dst_reg, 2149 offsetof(struct bpf_sysctl_kern, ppos)); 2150 *insn++ = BPF_STX_MEM( 2151 BPF_SIZEOF(u32), treg, si->src_reg, 2152 bpf_ctx_narrow_access_offset( 2153 0, sizeof(u32), sizeof(loff_t))); 2154 *insn++ = BPF_LDX_MEM( 2155 BPF_DW, treg, si->dst_reg, 2156 offsetof(struct bpf_sysctl_kern, tmp_reg)); 2157 } else { 2158 *insn++ = BPF_LDX_MEM( 2159 BPF_FIELD_SIZEOF(struct bpf_sysctl_kern, ppos), 2160 si->dst_reg, si->src_reg, 2161 offsetof(struct bpf_sysctl_kern, ppos)); 2162 read_size = bpf_size_to_bytes(BPF_SIZE(si->code)); 2163 *insn++ = BPF_LDX_MEM( 2164 BPF_SIZE(si->code), si->dst_reg, si->dst_reg, 2165 bpf_ctx_narrow_access_offset( 2166 0, read_size, sizeof(loff_t))); 2167 } 2168 *target_size = sizeof(u32); 2169 break; 2170 } 2171 2172 return insn - insn_buf; 2173 } 2174 2175 const struct bpf_verifier_ops cg_sysctl_verifier_ops = { 2176 .get_func_proto = sysctl_func_proto, 2177 .is_valid_access = sysctl_is_valid_access, 2178 .convert_ctx_access = sysctl_convert_ctx_access, 2179 }; 2180 2181 const struct bpf_prog_ops cg_sysctl_prog_ops = { 2182 }; 2183 2184 #ifdef CONFIG_NET 2185 BPF_CALL_1(bpf_get_netns_cookie_sockopt, struct bpf_sockopt_kern *, ctx) 2186 { 2187 const struct net *net = ctx ? sock_net(ctx->sk) : &init_net; 2188 2189 return net->net_cookie; 2190 } 2191 2192 static const struct bpf_func_proto bpf_get_netns_cookie_sockopt_proto = { 2193 .func = bpf_get_netns_cookie_sockopt, 2194 .gpl_only = false, 2195 .ret_type = RET_INTEGER, 2196 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 2197 }; 2198 #endif 2199 2200 static const struct bpf_func_proto * 2201 cg_sockopt_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 2202 { 2203 switch (func_id) { 2204 #ifdef CONFIG_NET 2205 case BPF_FUNC_get_netns_cookie: 2206 return &bpf_get_netns_cookie_sockopt_proto; 2207 case BPF_FUNC_sk_storage_get: 2208 return &bpf_sk_storage_get_proto; 2209 case BPF_FUNC_sk_storage_delete: 2210 return &bpf_sk_storage_delete_proto; 2211 case BPF_FUNC_setsockopt: 2212 if (prog->expected_attach_type == BPF_CGROUP_SETSOCKOPT) 2213 return &bpf_sk_setsockopt_proto; 2214 return NULL; 2215 case BPF_FUNC_getsockopt: 2216 if (prog->expected_attach_type == BPF_CGROUP_SETSOCKOPT) 2217 return &bpf_sk_getsockopt_proto; 2218 return NULL; 2219 #endif 2220 #ifdef CONFIG_INET 2221 case BPF_FUNC_tcp_sock: 2222 return &bpf_tcp_sock_proto; 2223 #endif 2224 default: 2225 return cgroup_base_func_proto(func_id, prog); 2226 } 2227 } 2228 2229 static bool cg_sockopt_is_valid_access(int off, int size, 2230 enum bpf_access_type type, 2231 const struct bpf_prog *prog, 2232 struct bpf_insn_access_aux *info) 2233 { 2234 const int size_default = sizeof(__u32); 2235 2236 if (off < 0 || off >= sizeof(struct bpf_sockopt)) 2237 return false; 2238 2239 if (off % size != 0) 2240 return false; 2241 2242 if (type == BPF_WRITE) { 2243 switch (off) { 2244 case offsetof(struct bpf_sockopt, retval): 2245 if (size != size_default) 2246 return false; 2247 return prog->expected_attach_type == 2248 BPF_CGROUP_GETSOCKOPT; 2249 case offsetof(struct bpf_sockopt, optname): 2250 fallthrough; 2251 case offsetof(struct bpf_sockopt, level): 2252 if (size != size_default) 2253 return false; 2254 return prog->expected_attach_type == 2255 BPF_CGROUP_SETSOCKOPT; 2256 case offsetof(struct bpf_sockopt, optlen): 2257 return size == size_default; 2258 default: 2259 return false; 2260 } 2261 } 2262 2263 switch (off) { 2264 case offsetof(struct bpf_sockopt, sk): 2265 if (size != sizeof(__u64)) 2266 return false; 2267 info->reg_type = PTR_TO_SOCKET; 2268 break; 2269 case offsetof(struct bpf_sockopt, optval): 2270 if (size != sizeof(__u64)) 2271 return false; 2272 info->reg_type = PTR_TO_PACKET; 2273 break; 2274 case offsetof(struct bpf_sockopt, optval_end): 2275 if (size != sizeof(__u64)) 2276 return false; 2277 info->reg_type = PTR_TO_PACKET_END; 2278 break; 2279 case offsetof(struct bpf_sockopt, retval): 2280 if (size != size_default) 2281 return false; 2282 return prog->expected_attach_type == BPF_CGROUP_GETSOCKOPT; 2283 default: 2284 if (size != size_default) 2285 return false; 2286 break; 2287 } 2288 return true; 2289 } 2290 2291 #define CG_SOCKOPT_ACCESS_FIELD(T, F) \ 2292 T(BPF_FIELD_SIZEOF(struct bpf_sockopt_kern, F), \ 2293 si->dst_reg, si->src_reg, \ 2294 offsetof(struct bpf_sockopt_kern, F)) 2295 2296 static u32 cg_sockopt_convert_ctx_access(enum bpf_access_type type, 2297 const struct bpf_insn *si, 2298 struct bpf_insn *insn_buf, 2299 struct bpf_prog *prog, 2300 u32 *target_size) 2301 { 2302 struct bpf_insn *insn = insn_buf; 2303 2304 switch (si->off) { 2305 case offsetof(struct bpf_sockopt, sk): 2306 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, sk); 2307 break; 2308 case offsetof(struct bpf_sockopt, level): 2309 if (type == BPF_WRITE) 2310 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_STX_MEM, level); 2311 else 2312 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, level); 2313 break; 2314 case offsetof(struct bpf_sockopt, optname): 2315 if (type == BPF_WRITE) 2316 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_STX_MEM, optname); 2317 else 2318 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, optname); 2319 break; 2320 case offsetof(struct bpf_sockopt, optlen): 2321 if (type == BPF_WRITE) 2322 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_STX_MEM, optlen); 2323 else 2324 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, optlen); 2325 break; 2326 case offsetof(struct bpf_sockopt, retval): 2327 BUILD_BUG_ON(offsetof(struct bpf_cg_run_ctx, run_ctx) != 0); 2328 2329 if (type == BPF_WRITE) { 2330 int treg = BPF_REG_9; 2331 2332 if (si->src_reg == treg || si->dst_reg == treg) 2333 --treg; 2334 if (si->src_reg == treg || si->dst_reg == treg) 2335 --treg; 2336 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, treg, 2337 offsetof(struct bpf_sockopt_kern, tmp_reg)); 2338 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sockopt_kern, current_task), 2339 treg, si->dst_reg, 2340 offsetof(struct bpf_sockopt_kern, current_task)); 2341 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct task_struct, bpf_ctx), 2342 treg, treg, 2343 offsetof(struct task_struct, bpf_ctx)); 2344 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(struct bpf_cg_run_ctx, retval), 2345 treg, si->src_reg, 2346 offsetof(struct bpf_cg_run_ctx, retval)); 2347 *insn++ = BPF_LDX_MEM(BPF_DW, treg, si->dst_reg, 2348 offsetof(struct bpf_sockopt_kern, tmp_reg)); 2349 } else { 2350 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sockopt_kern, current_task), 2351 si->dst_reg, si->src_reg, 2352 offsetof(struct bpf_sockopt_kern, current_task)); 2353 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct task_struct, bpf_ctx), 2354 si->dst_reg, si->dst_reg, 2355 offsetof(struct task_struct, bpf_ctx)); 2356 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_cg_run_ctx, retval), 2357 si->dst_reg, si->dst_reg, 2358 offsetof(struct bpf_cg_run_ctx, retval)); 2359 } 2360 break; 2361 case offsetof(struct bpf_sockopt, optval): 2362 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, optval); 2363 break; 2364 case offsetof(struct bpf_sockopt, optval_end): 2365 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, optval_end); 2366 break; 2367 } 2368 2369 return insn - insn_buf; 2370 } 2371 2372 static int cg_sockopt_get_prologue(struct bpf_insn *insn_buf, 2373 bool direct_write, 2374 const struct bpf_prog *prog) 2375 { 2376 /* Nothing to do for sockopt argument. The data is kzalloc'ated. 2377 */ 2378 return 0; 2379 } 2380 2381 const struct bpf_verifier_ops cg_sockopt_verifier_ops = { 2382 .get_func_proto = cg_sockopt_func_proto, 2383 .is_valid_access = cg_sockopt_is_valid_access, 2384 .convert_ctx_access = cg_sockopt_convert_ctx_access, 2385 .gen_prologue = cg_sockopt_get_prologue, 2386 }; 2387 2388 const struct bpf_prog_ops cg_sockopt_prog_ops = { 2389 }; 2390