1 // SPDX-License-Identifier: GPL-2.0 2 /* Copyright (c) 2020 Facebook */ 3 #include <vmlinux.h> 4 #include <bpf/bpf_core_read.h> 5 #include <bpf/bpf_helpers.h> 6 #include <bpf/bpf_tracing.h> 7 8 #include "profiler.h" 9 10 #ifndef NULL 11 #define NULL 0 12 #endif 13 14 #define O_WRONLY 00000001 15 #define O_RDWR 00000002 16 #define O_DIRECTORY 00200000 17 #define __O_TMPFILE 020000000 18 #define O_TMPFILE (__O_TMPFILE | O_DIRECTORY) 19 #define MAX_ERRNO 4095 20 #define S_IFMT 00170000 21 #define S_IFSOCK 0140000 22 #define S_IFLNK 0120000 23 #define S_IFREG 0100000 24 #define S_IFBLK 0060000 25 #define S_IFDIR 0040000 26 #define S_IFCHR 0020000 27 #define S_IFIFO 0010000 28 #define S_ISUID 0004000 29 #define S_ISGID 0002000 30 #define S_ISVTX 0001000 31 #define S_ISLNK(m) (((m)&S_IFMT) == S_IFLNK) 32 #define S_ISDIR(m) (((m)&S_IFMT) == S_IFDIR) 33 #define S_ISCHR(m) (((m)&S_IFMT) == S_IFCHR) 34 #define S_ISBLK(m) (((m)&S_IFMT) == S_IFBLK) 35 #define S_ISFIFO(m) (((m)&S_IFMT) == S_IFIFO) 36 #define S_ISSOCK(m) (((m)&S_IFMT) == S_IFSOCK) 37 #define IS_ERR_VALUE(x) (unsigned long)(void*)(x) >= (unsigned long)-MAX_ERRNO 38 39 #define KILL_DATA_ARRAY_SIZE 8 40 41 struct var_kill_data_arr_t { 42 struct var_kill_data_t array[KILL_DATA_ARRAY_SIZE]; 43 }; 44 45 union any_profiler_data_t { 46 struct var_exec_data_t var_exec; 47 struct var_kill_data_t var_kill; 48 struct var_sysctl_data_t var_sysctl; 49 struct var_filemod_data_t var_filemod; 50 struct var_fork_data_t var_fork; 51 struct var_kill_data_arr_t var_kill_data_arr; 52 }; 53 54 volatile struct profiler_config_struct bpf_config = {}; 55 56 #define FETCH_CGROUPS_FROM_BPF (bpf_config.fetch_cgroups_from_bpf) 57 #define CGROUP_FS_INODE (bpf_config.cgroup_fs_inode) 58 #define CGROUP_LOGIN_SESSION_INODE \ 59 (bpf_config.cgroup_login_session_inode) 60 #define KILL_SIGNALS (bpf_config.kill_signals_mask) 61 #define STALE_INFO (bpf_config.stale_info_secs) 62 #define INODE_FILTER (bpf_config.inode_filter) 63 #define READ_ENVIRON_FROM_EXEC (bpf_config.read_environ_from_exec) 64 #define ENABLE_CGROUP_V1_RESOLVER (bpf_config.enable_cgroup_v1_resolver) 65 66 struct kernfs_iattrs___52 { 67 struct iattr ia_iattr; 68 }; 69 70 struct kernfs_node___52 { 71 union /* kernfs_node_id */ { 72 struct { 73 u32 ino; 74 u32 generation; 75 }; 76 u64 id; 77 } id; 78 }; 79 80 struct { 81 __uint(type, BPF_MAP_TYPE_PERCPU_ARRAY); 82 __uint(max_entries, 1); 83 __type(key, u32); 84 __type(value, union any_profiler_data_t); 85 } data_heap SEC(".maps"); 86 87 struct { 88 __uint(type, BPF_MAP_TYPE_PERF_EVENT_ARRAY); 89 __uint(key_size, sizeof(int)); 90 __uint(value_size, sizeof(int)); 91 } events SEC(".maps"); 92 93 struct { 94 __uint(type, BPF_MAP_TYPE_HASH); 95 __uint(max_entries, KILL_DATA_ARRAY_SIZE); 96 __type(key, u32); 97 __type(value, struct var_kill_data_arr_t); 98 } var_tpid_to_data SEC(".maps"); 99 100 struct { 101 __uint(type, BPF_MAP_TYPE_PERCPU_ARRAY); 102 __uint(max_entries, profiler_bpf_max_function_id); 103 __type(key, u32); 104 __type(value, struct bpf_func_stats_data); 105 } bpf_func_stats SEC(".maps"); 106 107 struct { 108 __uint(type, BPF_MAP_TYPE_HASH); 109 __type(key, u32); 110 __type(value, bool); 111 __uint(max_entries, 16); 112 } allowed_devices SEC(".maps"); 113 114 struct { 115 __uint(type, BPF_MAP_TYPE_HASH); 116 __type(key, u64); 117 __type(value, bool); 118 __uint(max_entries, 1024); 119 } allowed_file_inodes SEC(".maps"); 120 121 struct { 122 __uint(type, BPF_MAP_TYPE_HASH); 123 __type(key, u64); 124 __type(value, bool); 125 __uint(max_entries, 1024); 126 } allowed_directory_inodes SEC(".maps"); 127 128 struct { 129 __uint(type, BPF_MAP_TYPE_HASH); 130 __type(key, u32); 131 __type(value, bool); 132 __uint(max_entries, 16); 133 } disallowed_exec_inodes SEC(".maps"); 134 135 #ifndef ARRAY_SIZE 136 #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0])) 137 #endif 138 139 static INLINE bool IS_ERR(const void* ptr) 140 { 141 return IS_ERR_VALUE((unsigned long)ptr); 142 } 143 144 static INLINE u32 get_userspace_pid() 145 { 146 return bpf_get_current_pid_tgid() >> 32; 147 } 148 149 static INLINE bool is_init_process(u32 tgid) 150 { 151 return tgid == 1 || tgid == 0; 152 } 153 154 static INLINE unsigned long 155 probe_read_lim(void* dst, void* src, unsigned long len, unsigned long max) 156 { 157 len = len < max ? len : max; 158 if (len > 1) { 159 if (bpf_probe_read(dst, len, src)) 160 return 0; 161 } else if (len == 1) { 162 if (bpf_probe_read(dst, 1, src)) 163 return 0; 164 } 165 return len; 166 } 167 168 static INLINE int get_var_spid_index(struct var_kill_data_arr_t* arr_struct, 169 int spid) 170 { 171 #ifdef UNROLL 172 #pragma unroll 173 #endif 174 for (int i = 0; i < ARRAY_SIZE(arr_struct->array); i++) 175 if (arr_struct->array[i].meta.pid == spid) 176 return i; 177 return -1; 178 } 179 180 static INLINE void populate_ancestors(struct task_struct* task, 181 struct ancestors_data_t* ancestors_data) 182 { 183 struct task_struct* parent = task; 184 u32 num_ancestors, ppid; 185 186 ancestors_data->num_ancestors = 0; 187 #ifdef UNROLL 188 #pragma unroll 189 #endif 190 for (num_ancestors = 0; num_ancestors < MAX_ANCESTORS; num_ancestors++) { 191 parent = BPF_CORE_READ(parent, real_parent); 192 if (parent == NULL) 193 break; 194 ppid = BPF_CORE_READ(parent, tgid); 195 if (is_init_process(ppid)) 196 break; 197 ancestors_data->ancestor_pids[num_ancestors] = ppid; 198 ancestors_data->ancestor_exec_ids[num_ancestors] = 199 BPF_CORE_READ(parent, self_exec_id); 200 ancestors_data->ancestor_start_times[num_ancestors] = 201 BPF_CORE_READ(parent, start_time); 202 ancestors_data->num_ancestors = num_ancestors; 203 } 204 } 205 206 static INLINE void* read_full_cgroup_path(struct kernfs_node* cgroup_node, 207 struct kernfs_node* cgroup_root_node, 208 void* payload, 209 int* root_pos) 210 { 211 void* payload_start = payload; 212 size_t filepart_length; 213 214 #ifdef UNROLL 215 #pragma unroll 216 #endif 217 for (int i = 0; i < MAX_CGROUPS_PATH_DEPTH; i++) { 218 filepart_length = 219 bpf_probe_read_str(payload, MAX_PATH, BPF_CORE_READ(cgroup_node, name)); 220 if (!cgroup_node) 221 return payload; 222 if (cgroup_node == cgroup_root_node) 223 *root_pos = payload - payload_start; 224 if (filepart_length <= MAX_PATH) { 225 barrier_var(filepart_length); 226 payload += filepart_length; 227 } 228 cgroup_node = BPF_CORE_READ(cgroup_node, parent); 229 } 230 return payload; 231 } 232 233 static ino_t get_inode_from_kernfs(struct kernfs_node* node) 234 { 235 struct kernfs_node___52* node52 = (void*)node; 236 237 if (bpf_core_field_exists(node52->id.ino)) { 238 barrier_var(node52); 239 return BPF_CORE_READ(node52, id.ino); 240 } else { 241 barrier_var(node); 242 return (u64)BPF_CORE_READ(node, id); 243 } 244 } 245 246 int pids_cgrp_id = 1; 247 248 static INLINE void* populate_cgroup_info(struct cgroup_data_t* cgroup_data, 249 struct task_struct* task, 250 void* payload) 251 { 252 struct kernfs_node* root_kernfs = 253 BPF_CORE_READ(task, nsproxy, cgroup_ns, root_cset, dfl_cgrp, kn); 254 struct kernfs_node* proc_kernfs = BPF_CORE_READ(task, cgroups, dfl_cgrp, kn); 255 256 if (ENABLE_CGROUP_V1_RESOLVER) { 257 #ifdef UNROLL 258 #pragma unroll 259 #endif 260 for (int i = 0; i < CGROUP_SUBSYS_COUNT; i++) { 261 struct cgroup_subsys_state* subsys = 262 BPF_CORE_READ(task, cgroups, subsys[i]); 263 if (subsys != NULL) { 264 int subsys_id = BPF_CORE_READ(subsys, ss, id); 265 if (subsys_id == pids_cgrp_id) { 266 proc_kernfs = BPF_CORE_READ(subsys, cgroup, kn); 267 root_kernfs = BPF_CORE_READ(subsys, ss, root, kf_root, kn); 268 break; 269 } 270 } 271 } 272 } 273 274 cgroup_data->cgroup_root_inode = get_inode_from_kernfs(root_kernfs); 275 cgroup_data->cgroup_proc_inode = get_inode_from_kernfs(proc_kernfs); 276 277 if (bpf_core_field_exists(root_kernfs->iattr->ia_mtime)) { 278 cgroup_data->cgroup_root_mtime = 279 BPF_CORE_READ(root_kernfs, iattr, ia_mtime.tv_nsec); 280 cgroup_data->cgroup_proc_mtime = 281 BPF_CORE_READ(proc_kernfs, iattr, ia_mtime.tv_nsec); 282 } else { 283 struct kernfs_iattrs___52* root_iattr = 284 (struct kernfs_iattrs___52*)BPF_CORE_READ(root_kernfs, iattr); 285 cgroup_data->cgroup_root_mtime = 286 BPF_CORE_READ(root_iattr, ia_iattr.ia_mtime.tv_nsec); 287 288 struct kernfs_iattrs___52* proc_iattr = 289 (struct kernfs_iattrs___52*)BPF_CORE_READ(proc_kernfs, iattr); 290 cgroup_data->cgroup_proc_mtime = 291 BPF_CORE_READ(proc_iattr, ia_iattr.ia_mtime.tv_nsec); 292 } 293 294 cgroup_data->cgroup_root_length = 0; 295 cgroup_data->cgroup_proc_length = 0; 296 cgroup_data->cgroup_full_length = 0; 297 298 size_t cgroup_root_length = 299 bpf_probe_read_str(payload, MAX_PATH, BPF_CORE_READ(root_kernfs, name)); 300 barrier_var(cgroup_root_length); 301 if (cgroup_root_length <= MAX_PATH) { 302 barrier_var(cgroup_root_length); 303 cgroup_data->cgroup_root_length = cgroup_root_length; 304 payload += cgroup_root_length; 305 } 306 307 size_t cgroup_proc_length = 308 bpf_probe_read_str(payload, MAX_PATH, BPF_CORE_READ(proc_kernfs, name)); 309 barrier_var(cgroup_proc_length); 310 if (cgroup_proc_length <= MAX_PATH) { 311 barrier_var(cgroup_proc_length); 312 cgroup_data->cgroup_proc_length = cgroup_proc_length; 313 payload += cgroup_proc_length; 314 } 315 316 if (FETCH_CGROUPS_FROM_BPF) { 317 cgroup_data->cgroup_full_path_root_pos = -1; 318 void* payload_end_pos = read_full_cgroup_path(proc_kernfs, root_kernfs, payload, 319 &cgroup_data->cgroup_full_path_root_pos); 320 cgroup_data->cgroup_full_length = payload_end_pos - payload; 321 payload = payload_end_pos; 322 } 323 324 return (void*)payload; 325 } 326 327 static INLINE void* populate_var_metadata(struct var_metadata_t* metadata, 328 struct task_struct* task, 329 u32 pid, void* payload) 330 { 331 u64 uid_gid = bpf_get_current_uid_gid(); 332 333 metadata->uid = (u32)uid_gid; 334 metadata->gid = uid_gid >> 32; 335 metadata->pid = pid; 336 metadata->exec_id = BPF_CORE_READ(task, self_exec_id); 337 metadata->start_time = BPF_CORE_READ(task, start_time); 338 metadata->comm_length = 0; 339 340 size_t comm_length = bpf_core_read_str(payload, TASK_COMM_LEN, &task->comm); 341 barrier_var(comm_length); 342 if (comm_length <= TASK_COMM_LEN) { 343 barrier_var(comm_length); 344 metadata->comm_length = comm_length; 345 payload += comm_length; 346 } 347 348 return (void*)payload; 349 } 350 351 static INLINE struct var_kill_data_t* 352 get_var_kill_data(struct pt_regs* ctx, int spid, int tpid, int sig) 353 { 354 int zero = 0; 355 struct var_kill_data_t* kill_data = bpf_map_lookup_elem(&data_heap, &zero); 356 357 if (kill_data == NULL) 358 return NULL; 359 struct task_struct* task = (struct task_struct*)bpf_get_current_task(); 360 361 void* payload = populate_var_metadata(&kill_data->meta, task, spid, kill_data->payload); 362 payload = populate_cgroup_info(&kill_data->cgroup_data, task, payload); 363 size_t payload_length = payload - (void*)kill_data->payload; 364 kill_data->payload_length = payload_length; 365 populate_ancestors(task, &kill_data->ancestors_info); 366 kill_data->meta.type = KILL_EVENT; 367 kill_data->kill_target_pid = tpid; 368 kill_data->kill_sig = sig; 369 kill_data->kill_count = 1; 370 kill_data->last_kill_time = bpf_ktime_get_ns(); 371 return kill_data; 372 } 373 374 static INLINE int trace_var_sys_kill(void* ctx, int tpid, int sig) 375 { 376 if ((KILL_SIGNALS & (1ULL << sig)) == 0) 377 return 0; 378 379 u32 spid = get_userspace_pid(); 380 struct var_kill_data_arr_t* arr_struct = bpf_map_lookup_elem(&var_tpid_to_data, &tpid); 381 382 if (arr_struct == NULL) { 383 struct var_kill_data_t* kill_data = get_var_kill_data(ctx, spid, tpid, sig); 384 int zero = 0; 385 386 if (kill_data == NULL) 387 return 0; 388 arr_struct = bpf_map_lookup_elem(&data_heap, &zero); 389 if (arr_struct == NULL) 390 return 0; 391 bpf_probe_read(&arr_struct->array[0], sizeof(arr_struct->array[0]), kill_data); 392 } else { 393 int index = get_var_spid_index(arr_struct, spid); 394 395 if (index == -1) { 396 struct var_kill_data_t* kill_data = 397 get_var_kill_data(ctx, spid, tpid, sig); 398 if (kill_data == NULL) 399 return 0; 400 #ifdef UNROLL 401 #pragma unroll 402 #endif 403 for (int i = 0; i < ARRAY_SIZE(arr_struct->array); i++) 404 if (arr_struct->array[i].meta.pid == 0) { 405 bpf_probe_read(&arr_struct->array[i], 406 sizeof(arr_struct->array[i]), kill_data); 407 bpf_map_update_elem(&var_tpid_to_data, &tpid, 408 arr_struct, 0); 409 410 return 0; 411 } 412 return 0; 413 } 414 415 struct var_kill_data_t* kill_data = &arr_struct->array[index]; 416 417 u64 delta_sec = 418 (bpf_ktime_get_ns() - kill_data->last_kill_time) / 1000000000; 419 420 if (delta_sec < STALE_INFO) { 421 kill_data->kill_count++; 422 kill_data->last_kill_time = bpf_ktime_get_ns(); 423 bpf_probe_read(&arr_struct->array[index], 424 sizeof(arr_struct->array[index]), 425 kill_data); 426 } else { 427 struct var_kill_data_t* kill_data = 428 get_var_kill_data(ctx, spid, tpid, sig); 429 if (kill_data == NULL) 430 return 0; 431 bpf_probe_read(&arr_struct->array[index], 432 sizeof(arr_struct->array[index]), 433 kill_data); 434 } 435 } 436 bpf_map_update_elem(&var_tpid_to_data, &tpid, arr_struct, 0); 437 return 0; 438 } 439 440 static INLINE void bpf_stats_enter(struct bpf_func_stats_ctx* bpf_stat_ctx, 441 enum bpf_function_id func_id) 442 { 443 int func_id_key = func_id; 444 445 bpf_stat_ctx->start_time_ns = bpf_ktime_get_ns(); 446 bpf_stat_ctx->bpf_func_stats_data_val = 447 bpf_map_lookup_elem(&bpf_func_stats, &func_id_key); 448 if (bpf_stat_ctx->bpf_func_stats_data_val) 449 bpf_stat_ctx->bpf_func_stats_data_val->num_executions++; 450 } 451 452 static INLINE void bpf_stats_exit(struct bpf_func_stats_ctx* bpf_stat_ctx) 453 { 454 if (bpf_stat_ctx->bpf_func_stats_data_val) 455 bpf_stat_ctx->bpf_func_stats_data_val->time_elapsed_ns += 456 bpf_ktime_get_ns() - bpf_stat_ctx->start_time_ns; 457 } 458 459 static INLINE void 460 bpf_stats_pre_submit_var_perf_event(struct bpf_func_stats_ctx* bpf_stat_ctx, 461 struct var_metadata_t* meta) 462 { 463 if (bpf_stat_ctx->bpf_func_stats_data_val) { 464 bpf_stat_ctx->bpf_func_stats_data_val->num_perf_events++; 465 meta->bpf_stats_num_perf_events = 466 bpf_stat_ctx->bpf_func_stats_data_val->num_perf_events; 467 } 468 meta->bpf_stats_start_ktime_ns = bpf_stat_ctx->start_time_ns; 469 meta->cpu_id = bpf_get_smp_processor_id(); 470 } 471 472 static INLINE size_t 473 read_absolute_file_path_from_dentry(struct dentry* filp_dentry, void* payload) 474 { 475 size_t length = 0; 476 size_t filepart_length; 477 struct dentry* parent_dentry; 478 479 #ifdef UNROLL 480 #pragma unroll 481 #endif 482 for (int i = 0; i < MAX_PATH_DEPTH; i++) { 483 filepart_length = bpf_probe_read_str(payload, MAX_PATH, 484 BPF_CORE_READ(filp_dentry, d_name.name)); 485 barrier_var(filepart_length); 486 if (filepart_length > MAX_PATH) 487 break; 488 barrier_var(filepart_length); 489 payload += filepart_length; 490 length += filepart_length; 491 492 parent_dentry = BPF_CORE_READ(filp_dentry, d_parent); 493 if (filp_dentry == parent_dentry) 494 break; 495 filp_dentry = parent_dentry; 496 } 497 498 return length; 499 } 500 501 static INLINE bool 502 is_ancestor_in_allowed_inodes(struct dentry* filp_dentry) 503 { 504 struct dentry* parent_dentry; 505 #ifdef UNROLL 506 #pragma unroll 507 #endif 508 for (int i = 0; i < MAX_PATH_DEPTH; i++) { 509 u64 dir_ino = BPF_CORE_READ(filp_dentry, d_inode, i_ino); 510 bool* allowed_dir = bpf_map_lookup_elem(&allowed_directory_inodes, &dir_ino); 511 512 if (allowed_dir != NULL) 513 return true; 514 parent_dentry = BPF_CORE_READ(filp_dentry, d_parent); 515 if (filp_dentry == parent_dentry) 516 break; 517 filp_dentry = parent_dentry; 518 } 519 return false; 520 } 521 522 static INLINE bool is_dentry_allowed_for_filemod(struct dentry* file_dentry, 523 u32* device_id, 524 u64* file_ino) 525 { 526 u32 dev_id = BPF_CORE_READ(file_dentry, d_sb, s_dev); 527 *device_id = dev_id; 528 bool* allowed_device = bpf_map_lookup_elem(&allowed_devices, &dev_id); 529 530 if (allowed_device == NULL) 531 return false; 532 533 u64 ino = BPF_CORE_READ(file_dentry, d_inode, i_ino); 534 *file_ino = ino; 535 bool* allowed_file = bpf_map_lookup_elem(&allowed_file_inodes, &ino); 536 537 if (allowed_file == NULL) 538 if (!is_ancestor_in_allowed_inodes(BPF_CORE_READ(file_dentry, d_parent))) 539 return false; 540 return true; 541 } 542 543 SEC("kprobe/proc_sys_write") 544 ssize_t BPF_KPROBE(kprobe__proc_sys_write, 545 struct file* filp, const char* buf, 546 size_t count, loff_t* ppos) 547 { 548 struct bpf_func_stats_ctx stats_ctx; 549 bpf_stats_enter(&stats_ctx, profiler_bpf_proc_sys_write); 550 551 u32 pid = get_userspace_pid(); 552 int zero = 0; 553 struct var_sysctl_data_t* sysctl_data = 554 bpf_map_lookup_elem(&data_heap, &zero); 555 if (!sysctl_data) 556 goto out; 557 558 struct task_struct* task = (struct task_struct*)bpf_get_current_task(); 559 sysctl_data->meta.type = SYSCTL_EVENT; 560 void* payload = populate_var_metadata(&sysctl_data->meta, task, pid, sysctl_data->payload); 561 payload = populate_cgroup_info(&sysctl_data->cgroup_data, task, payload); 562 563 populate_ancestors(task, &sysctl_data->ancestors_info); 564 565 sysctl_data->sysctl_val_length = 0; 566 sysctl_data->sysctl_path_length = 0; 567 568 size_t sysctl_val_length = bpf_probe_read_str(payload, CTL_MAXNAME, buf); 569 barrier_var(sysctl_val_length); 570 if (sysctl_val_length <= CTL_MAXNAME) { 571 barrier_var(sysctl_val_length); 572 sysctl_data->sysctl_val_length = sysctl_val_length; 573 payload += sysctl_val_length; 574 } 575 576 size_t sysctl_path_length = bpf_probe_read_str(payload, MAX_PATH, 577 BPF_CORE_READ(filp, f_path.dentry, d_name.name)); 578 barrier_var(sysctl_path_length); 579 if (sysctl_path_length <= MAX_PATH) { 580 barrier_var(sysctl_path_length); 581 sysctl_data->sysctl_path_length = sysctl_path_length; 582 payload += sysctl_path_length; 583 } 584 585 bpf_stats_pre_submit_var_perf_event(&stats_ctx, &sysctl_data->meta); 586 unsigned long data_len = payload - (void*)sysctl_data; 587 data_len = data_len > sizeof(struct var_sysctl_data_t) 588 ? sizeof(struct var_sysctl_data_t) 589 : data_len; 590 bpf_perf_event_output(ctx, &events, BPF_F_CURRENT_CPU, sysctl_data, data_len); 591 out: 592 bpf_stats_exit(&stats_ctx); 593 return 0; 594 } 595 596 SEC("tracepoint/syscalls/sys_enter_kill") 597 int tracepoint__syscalls__sys_enter_kill(struct trace_event_raw_sys_enter* ctx) 598 { 599 struct bpf_func_stats_ctx stats_ctx; 600 601 bpf_stats_enter(&stats_ctx, profiler_bpf_sys_enter_kill); 602 int pid = ctx->args[0]; 603 int sig = ctx->args[1]; 604 int ret = trace_var_sys_kill(ctx, pid, sig); 605 bpf_stats_exit(&stats_ctx); 606 return ret; 607 }; 608 609 SEC("raw_tracepoint/sched_process_exit") 610 int raw_tracepoint__sched_process_exit(void* ctx) 611 { 612 int zero = 0; 613 struct bpf_func_stats_ctx stats_ctx; 614 bpf_stats_enter(&stats_ctx, profiler_bpf_sched_process_exit); 615 616 u32 tpid = get_userspace_pid(); 617 618 struct var_kill_data_arr_t* arr_struct = bpf_map_lookup_elem(&var_tpid_to_data, &tpid); 619 struct var_kill_data_t* kill_data = bpf_map_lookup_elem(&data_heap, &zero); 620 621 if (arr_struct == NULL || kill_data == NULL) 622 goto out; 623 624 struct task_struct* task = (struct task_struct*)bpf_get_current_task(); 625 struct kernfs_node* proc_kernfs = BPF_CORE_READ(task, cgroups, dfl_cgrp, kn); 626 627 #ifdef UNROLL 628 #pragma unroll 629 #endif 630 for (int i = 0; i < ARRAY_SIZE(arr_struct->array); i++) { 631 struct var_kill_data_t* past_kill_data = &arr_struct->array[i]; 632 633 if (past_kill_data != NULL && past_kill_data->kill_target_pid == tpid) { 634 bpf_probe_read(kill_data, sizeof(*past_kill_data), past_kill_data); 635 void* payload = kill_data->payload; 636 size_t offset = kill_data->payload_length; 637 if (offset >= MAX_METADATA_PAYLOAD_LEN + MAX_CGROUP_PAYLOAD_LEN) 638 return 0; 639 payload += offset; 640 641 kill_data->kill_target_name_length = 0; 642 kill_data->kill_target_cgroup_proc_length = 0; 643 644 size_t comm_length = bpf_core_read_str(payload, TASK_COMM_LEN, &task->comm); 645 barrier_var(comm_length); 646 if (comm_length <= TASK_COMM_LEN) { 647 barrier_var(comm_length); 648 kill_data->kill_target_name_length = comm_length; 649 payload += comm_length; 650 } 651 652 size_t cgroup_proc_length = bpf_probe_read_str(payload, KILL_TARGET_LEN, 653 BPF_CORE_READ(proc_kernfs, name)); 654 barrier_var(cgroup_proc_length); 655 if (cgroup_proc_length <= KILL_TARGET_LEN) { 656 barrier_var(cgroup_proc_length); 657 kill_data->kill_target_cgroup_proc_length = cgroup_proc_length; 658 payload += cgroup_proc_length; 659 } 660 661 bpf_stats_pre_submit_var_perf_event(&stats_ctx, &kill_data->meta); 662 unsigned long data_len = (void*)payload - (void*)kill_data; 663 data_len = data_len > sizeof(struct var_kill_data_t) 664 ? sizeof(struct var_kill_data_t) 665 : data_len; 666 bpf_perf_event_output(ctx, &events, BPF_F_CURRENT_CPU, kill_data, data_len); 667 } 668 } 669 bpf_map_delete_elem(&var_tpid_to_data, &tpid); 670 out: 671 bpf_stats_exit(&stats_ctx); 672 return 0; 673 } 674 675 SEC("raw_tracepoint/sched_process_exec") 676 int raw_tracepoint__sched_process_exec(struct bpf_raw_tracepoint_args* ctx) 677 { 678 struct bpf_func_stats_ctx stats_ctx; 679 bpf_stats_enter(&stats_ctx, profiler_bpf_sched_process_exec); 680 681 struct linux_binprm* bprm = (struct linux_binprm*)ctx->args[2]; 682 u64 inode = BPF_CORE_READ(bprm, file, f_inode, i_ino); 683 684 bool* should_filter_binprm = bpf_map_lookup_elem(&disallowed_exec_inodes, &inode); 685 if (should_filter_binprm != NULL) 686 goto out; 687 688 int zero = 0; 689 struct var_exec_data_t* proc_exec_data = bpf_map_lookup_elem(&data_heap, &zero); 690 if (!proc_exec_data) 691 goto out; 692 693 if (INODE_FILTER && inode != INODE_FILTER) 694 return 0; 695 696 u32 pid = get_userspace_pid(); 697 struct task_struct* task = (struct task_struct*)bpf_get_current_task(); 698 699 proc_exec_data->meta.type = EXEC_EVENT; 700 proc_exec_data->bin_path_length = 0; 701 proc_exec_data->cmdline_length = 0; 702 proc_exec_data->environment_length = 0; 703 void* payload = populate_var_metadata(&proc_exec_data->meta, task, pid, 704 proc_exec_data->payload); 705 payload = populate_cgroup_info(&proc_exec_data->cgroup_data, task, payload); 706 707 struct task_struct* parent_task = BPF_CORE_READ(task, real_parent); 708 proc_exec_data->parent_pid = BPF_CORE_READ(parent_task, tgid); 709 proc_exec_data->parent_uid = BPF_CORE_READ(parent_task, real_cred, uid.val); 710 proc_exec_data->parent_exec_id = BPF_CORE_READ(parent_task, self_exec_id); 711 proc_exec_data->parent_start_time = BPF_CORE_READ(parent_task, start_time); 712 713 const char* filename = BPF_CORE_READ(bprm, filename); 714 size_t bin_path_length = bpf_probe_read_str(payload, MAX_FILENAME_LEN, filename); 715 barrier_var(bin_path_length); 716 if (bin_path_length <= MAX_FILENAME_LEN) { 717 barrier_var(bin_path_length); 718 proc_exec_data->bin_path_length = bin_path_length; 719 payload += bin_path_length; 720 } 721 722 void* arg_start = (void*)BPF_CORE_READ(task, mm, arg_start); 723 void* arg_end = (void*)BPF_CORE_READ(task, mm, arg_end); 724 unsigned int cmdline_length = probe_read_lim(payload, arg_start, 725 arg_end - arg_start, MAX_ARGS_LEN); 726 727 if (cmdline_length <= MAX_ARGS_LEN) { 728 barrier_var(cmdline_length); 729 proc_exec_data->cmdline_length = cmdline_length; 730 payload += cmdline_length; 731 } 732 733 if (READ_ENVIRON_FROM_EXEC) { 734 void* env_start = (void*)BPF_CORE_READ(task, mm, env_start); 735 void* env_end = (void*)BPF_CORE_READ(task, mm, env_end); 736 unsigned long env_len = probe_read_lim(payload, env_start, 737 env_end - env_start, MAX_ENVIRON_LEN); 738 if (cmdline_length <= MAX_ENVIRON_LEN) { 739 proc_exec_data->environment_length = env_len; 740 payload += env_len; 741 } 742 } 743 744 bpf_stats_pre_submit_var_perf_event(&stats_ctx, &proc_exec_data->meta); 745 unsigned long data_len = payload - (void*)proc_exec_data; 746 data_len = data_len > sizeof(struct var_exec_data_t) 747 ? sizeof(struct var_exec_data_t) 748 : data_len; 749 bpf_perf_event_output(ctx, &events, BPF_F_CURRENT_CPU, proc_exec_data, data_len); 750 out: 751 bpf_stats_exit(&stats_ctx); 752 return 0; 753 } 754 755 SEC("kretprobe/do_filp_open") 756 int kprobe_ret__do_filp_open(struct pt_regs* ctx) 757 { 758 struct bpf_func_stats_ctx stats_ctx; 759 bpf_stats_enter(&stats_ctx, profiler_bpf_do_filp_open_ret); 760 761 struct file* filp = (struct file*)PT_REGS_RC_CORE(ctx); 762 763 if (filp == NULL || IS_ERR(filp)) 764 goto out; 765 unsigned int flags = BPF_CORE_READ(filp, f_flags); 766 if ((flags & (O_RDWR | O_WRONLY)) == 0) 767 goto out; 768 if ((flags & O_TMPFILE) > 0) 769 goto out; 770 struct inode* file_inode = BPF_CORE_READ(filp, f_inode); 771 umode_t mode = BPF_CORE_READ(file_inode, i_mode); 772 if (S_ISDIR(mode) || S_ISCHR(mode) || S_ISBLK(mode) || S_ISFIFO(mode) || 773 S_ISSOCK(mode)) 774 goto out; 775 776 struct dentry* filp_dentry = BPF_CORE_READ(filp, f_path.dentry); 777 u32 device_id = 0; 778 u64 file_ino = 0; 779 if (!is_dentry_allowed_for_filemod(filp_dentry, &device_id, &file_ino)) 780 goto out; 781 782 int zero = 0; 783 struct var_filemod_data_t* filemod_data = bpf_map_lookup_elem(&data_heap, &zero); 784 if (!filemod_data) 785 goto out; 786 787 u32 pid = get_userspace_pid(); 788 struct task_struct* task = (struct task_struct*)bpf_get_current_task(); 789 790 filemod_data->meta.type = FILEMOD_EVENT; 791 filemod_data->fmod_type = FMOD_OPEN; 792 filemod_data->dst_flags = flags; 793 filemod_data->src_inode = 0; 794 filemod_data->dst_inode = file_ino; 795 filemod_data->src_device_id = 0; 796 filemod_data->dst_device_id = device_id; 797 filemod_data->src_filepath_length = 0; 798 filemod_data->dst_filepath_length = 0; 799 800 void* payload = populate_var_metadata(&filemod_data->meta, task, pid, 801 filemod_data->payload); 802 payload = populate_cgroup_info(&filemod_data->cgroup_data, task, payload); 803 804 size_t len = read_absolute_file_path_from_dentry(filp_dentry, payload); 805 barrier_var(len); 806 if (len <= MAX_FILEPATH_LENGTH) { 807 barrier_var(len); 808 payload += len; 809 filemod_data->dst_filepath_length = len; 810 } 811 bpf_stats_pre_submit_var_perf_event(&stats_ctx, &filemod_data->meta); 812 unsigned long data_len = payload - (void*)filemod_data; 813 data_len = data_len > sizeof(*filemod_data) ? sizeof(*filemod_data) : data_len; 814 bpf_perf_event_output(ctx, &events, BPF_F_CURRENT_CPU, filemod_data, data_len); 815 out: 816 bpf_stats_exit(&stats_ctx); 817 return 0; 818 } 819 820 SEC("kprobe/vfs_link") 821 int BPF_KPROBE(kprobe__vfs_link, 822 struct dentry* old_dentry, struct inode* dir, 823 struct dentry* new_dentry, struct inode** delegated_inode) 824 { 825 struct bpf_func_stats_ctx stats_ctx; 826 bpf_stats_enter(&stats_ctx, profiler_bpf_vfs_link); 827 828 u32 src_device_id = 0; 829 u64 src_file_ino = 0; 830 u32 dst_device_id = 0; 831 u64 dst_file_ino = 0; 832 if (!is_dentry_allowed_for_filemod(old_dentry, &src_device_id, &src_file_ino) && 833 !is_dentry_allowed_for_filemod(new_dentry, &dst_device_id, &dst_file_ino)) 834 goto out; 835 836 int zero = 0; 837 struct var_filemod_data_t* filemod_data = bpf_map_lookup_elem(&data_heap, &zero); 838 if (!filemod_data) 839 goto out; 840 841 u32 pid = get_userspace_pid(); 842 struct task_struct* task = (struct task_struct*)bpf_get_current_task(); 843 844 filemod_data->meta.type = FILEMOD_EVENT; 845 filemod_data->fmod_type = FMOD_LINK; 846 filemod_data->dst_flags = 0; 847 filemod_data->src_inode = src_file_ino; 848 filemod_data->dst_inode = dst_file_ino; 849 filemod_data->src_device_id = src_device_id; 850 filemod_data->dst_device_id = dst_device_id; 851 filemod_data->src_filepath_length = 0; 852 filemod_data->dst_filepath_length = 0; 853 854 void* payload = populate_var_metadata(&filemod_data->meta, task, pid, 855 filemod_data->payload); 856 payload = populate_cgroup_info(&filemod_data->cgroup_data, task, payload); 857 858 size_t len = read_absolute_file_path_from_dentry(old_dentry, payload); 859 barrier_var(len); 860 if (len <= MAX_FILEPATH_LENGTH) { 861 barrier_var(len); 862 payload += len; 863 filemod_data->src_filepath_length = len; 864 } 865 866 len = read_absolute_file_path_from_dentry(new_dentry, payload); 867 barrier_var(len); 868 if (len <= MAX_FILEPATH_LENGTH) { 869 barrier_var(len); 870 payload += len; 871 filemod_data->dst_filepath_length = len; 872 } 873 874 bpf_stats_pre_submit_var_perf_event(&stats_ctx, &filemod_data->meta); 875 unsigned long data_len = payload - (void*)filemod_data; 876 data_len = data_len > sizeof(*filemod_data) ? sizeof(*filemod_data) : data_len; 877 bpf_perf_event_output(ctx, &events, BPF_F_CURRENT_CPU, filemod_data, data_len); 878 out: 879 bpf_stats_exit(&stats_ctx); 880 return 0; 881 } 882 883 SEC("kprobe/vfs_symlink") 884 int BPF_KPROBE(kprobe__vfs_symlink, struct inode* dir, struct dentry* dentry, 885 const char* oldname) 886 { 887 struct bpf_func_stats_ctx stats_ctx; 888 bpf_stats_enter(&stats_ctx, profiler_bpf_vfs_symlink); 889 890 u32 dst_device_id = 0; 891 u64 dst_file_ino = 0; 892 if (!is_dentry_allowed_for_filemod(dentry, &dst_device_id, &dst_file_ino)) 893 goto out; 894 895 int zero = 0; 896 struct var_filemod_data_t* filemod_data = bpf_map_lookup_elem(&data_heap, &zero); 897 if (!filemod_data) 898 goto out; 899 900 u32 pid = get_userspace_pid(); 901 struct task_struct* task = (struct task_struct*)bpf_get_current_task(); 902 903 filemod_data->meta.type = FILEMOD_EVENT; 904 filemod_data->fmod_type = FMOD_SYMLINK; 905 filemod_data->dst_flags = 0; 906 filemod_data->src_inode = 0; 907 filemod_data->dst_inode = dst_file_ino; 908 filemod_data->src_device_id = 0; 909 filemod_data->dst_device_id = dst_device_id; 910 filemod_data->src_filepath_length = 0; 911 filemod_data->dst_filepath_length = 0; 912 913 void* payload = populate_var_metadata(&filemod_data->meta, task, pid, 914 filemod_data->payload); 915 payload = populate_cgroup_info(&filemod_data->cgroup_data, task, payload); 916 917 size_t len = bpf_probe_read_str(payload, MAX_FILEPATH_LENGTH, oldname); 918 barrier_var(len); 919 if (len <= MAX_FILEPATH_LENGTH) { 920 barrier_var(len); 921 payload += len; 922 filemod_data->src_filepath_length = len; 923 } 924 len = read_absolute_file_path_from_dentry(dentry, payload); 925 barrier_var(len); 926 if (len <= MAX_FILEPATH_LENGTH) { 927 barrier_var(len); 928 payload += len; 929 filemod_data->dst_filepath_length = len; 930 } 931 bpf_stats_pre_submit_var_perf_event(&stats_ctx, &filemod_data->meta); 932 unsigned long data_len = payload - (void*)filemod_data; 933 data_len = data_len > sizeof(*filemod_data) ? sizeof(*filemod_data) : data_len; 934 bpf_perf_event_output(ctx, &events, BPF_F_CURRENT_CPU, filemod_data, data_len); 935 out: 936 bpf_stats_exit(&stats_ctx); 937 return 0; 938 } 939 940 SEC("raw_tracepoint/sched_process_fork") 941 int raw_tracepoint__sched_process_fork(struct bpf_raw_tracepoint_args* ctx) 942 { 943 struct bpf_func_stats_ctx stats_ctx; 944 bpf_stats_enter(&stats_ctx, profiler_bpf_sched_process_fork); 945 946 int zero = 0; 947 struct var_fork_data_t* fork_data = bpf_map_lookup_elem(&data_heap, &zero); 948 if (!fork_data) 949 goto out; 950 951 struct task_struct* parent = (struct task_struct*)ctx->args[0]; 952 struct task_struct* child = (struct task_struct*)ctx->args[1]; 953 fork_data->meta.type = FORK_EVENT; 954 955 void* payload = populate_var_metadata(&fork_data->meta, child, 956 BPF_CORE_READ(child, pid), fork_data->payload); 957 fork_data->parent_pid = BPF_CORE_READ(parent, pid); 958 fork_data->parent_exec_id = BPF_CORE_READ(parent, self_exec_id); 959 fork_data->parent_start_time = BPF_CORE_READ(parent, start_time); 960 bpf_stats_pre_submit_var_perf_event(&stats_ctx, &fork_data->meta); 961 962 unsigned long data_len = payload - (void*)fork_data; 963 data_len = data_len > sizeof(*fork_data) ? sizeof(*fork_data) : data_len; 964 bpf_perf_event_output(ctx, &events, BPF_F_CURRENT_CPU, fork_data, data_len); 965 out: 966 bpf_stats_exit(&stats_ctx); 967 return 0; 968 } 969 char _license[] SEC("license") = "GPL"; 970