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 #include "err.h" 10 #include "bpf_experimental.h" 11 12 #ifndef NULL 13 #define NULL 0 14 #endif 15 16 #define O_WRONLY 00000001 17 #define O_RDWR 00000002 18 #define O_DIRECTORY 00200000 19 #define __O_TMPFILE 020000000 20 #define O_TMPFILE (__O_TMPFILE | O_DIRECTORY) 21 #define S_IFMT 00170000 22 #define S_IFSOCK 0140000 23 #define S_IFLNK 0120000 24 #define S_IFREG 0100000 25 #define S_IFBLK 0060000 26 #define S_IFDIR 0040000 27 #define S_IFCHR 0020000 28 #define S_IFIFO 0010000 29 #define S_ISUID 0004000 30 #define S_ISGID 0002000 31 #define S_ISVTX 0001000 32 #define S_ISLNK(m) (((m)&S_IFMT) == S_IFLNK) 33 #define S_ISDIR(m) (((m)&S_IFMT) == S_IFDIR) 34 #define S_ISCHR(m) (((m)&S_IFMT) == S_IFCHR) 35 #define S_ISBLK(m) (((m)&S_IFMT) == S_IFBLK) 36 #define S_ISFIFO(m) (((m)&S_IFMT) == S_IFIFO) 37 #define S_ISSOCK(m) (((m)&S_IFMT) == S_IFSOCK) 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) (int)(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_kernel(dst, len, src)) 160 return 0; 161 } else if (len == 1) { 162 if (bpf_probe_read_kernel(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_kernel_str(payload, MAX_PATH, 220 BPF_CORE_READ(cgroup_node, name)); 221 if (!cgroup_node) 222 return payload; 223 if (cgroup_node == cgroup_root_node) 224 *root_pos = payload - payload_start; 225 if (bpf_cmp_likely(filepart_length, <=, MAX_PATH)) { 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 extern bool CONFIG_CGROUP_PIDS __kconfig __weak; 247 enum cgroup_subsys_id___local { 248 pids_cgrp_id___local = 123, /* value doesn't matter */ 249 }; 250 251 static INLINE void* populate_cgroup_info(struct cgroup_data_t* cgroup_data, 252 struct task_struct* task, 253 void* payload) 254 { 255 struct kernfs_node* root_kernfs = 256 BPF_CORE_READ(task, nsproxy, cgroup_ns, root_cset, dfl_cgrp, kn); 257 struct kernfs_node* proc_kernfs = BPF_CORE_READ(task, cgroups, dfl_cgrp, kn); 258 259 #if __has_builtin(__builtin_preserve_enum_value) 260 if (ENABLE_CGROUP_V1_RESOLVER && CONFIG_CGROUP_PIDS) { 261 int cgrp_id = bpf_core_enum_value(enum cgroup_subsys_id___local, 262 pids_cgrp_id___local); 263 #ifdef UNROLL 264 #pragma unroll 265 #endif 266 for (int i = 0; i < CGROUP_SUBSYS_COUNT; i++) { 267 struct cgroup_subsys_state* subsys = 268 BPF_CORE_READ(task, cgroups, subsys[i]); 269 if (subsys != NULL) { 270 int subsys_id = BPF_CORE_READ(subsys, ss, id); 271 if (subsys_id == cgrp_id) { 272 proc_kernfs = BPF_CORE_READ(subsys, cgroup, kn); 273 root_kernfs = BPF_CORE_READ(subsys, ss, root, kf_root, kn); 274 break; 275 } 276 } 277 } 278 } 279 #endif 280 281 cgroup_data->cgroup_root_inode = get_inode_from_kernfs(root_kernfs); 282 cgroup_data->cgroup_proc_inode = get_inode_from_kernfs(proc_kernfs); 283 284 if (bpf_core_field_exists(root_kernfs->iattr->ia_mtime)) { 285 cgroup_data->cgroup_root_mtime = 286 BPF_CORE_READ(root_kernfs, iattr, ia_mtime.tv_nsec); 287 cgroup_data->cgroup_proc_mtime = 288 BPF_CORE_READ(proc_kernfs, iattr, ia_mtime.tv_nsec); 289 } else { 290 struct kernfs_iattrs___52* root_iattr = 291 (struct kernfs_iattrs___52*)BPF_CORE_READ(root_kernfs, iattr); 292 cgroup_data->cgroup_root_mtime = 293 BPF_CORE_READ(root_iattr, ia_iattr.ia_mtime.tv_nsec); 294 295 struct kernfs_iattrs___52* proc_iattr = 296 (struct kernfs_iattrs___52*)BPF_CORE_READ(proc_kernfs, iattr); 297 cgroup_data->cgroup_proc_mtime = 298 BPF_CORE_READ(proc_iattr, ia_iattr.ia_mtime.tv_nsec); 299 } 300 301 cgroup_data->cgroup_root_length = 0; 302 cgroup_data->cgroup_proc_length = 0; 303 cgroup_data->cgroup_full_length = 0; 304 305 size_t cgroup_root_length = 306 bpf_probe_read_kernel_str(payload, MAX_PATH, 307 BPF_CORE_READ(root_kernfs, name)); 308 if (bpf_cmp_likely(cgroup_root_length, <=, MAX_PATH)) { 309 cgroup_data->cgroup_root_length = cgroup_root_length; 310 payload += cgroup_root_length; 311 } 312 313 size_t cgroup_proc_length = 314 bpf_probe_read_kernel_str(payload, MAX_PATH, 315 BPF_CORE_READ(proc_kernfs, name)); 316 if (bpf_cmp_likely(cgroup_proc_length, <=, MAX_PATH)) { 317 cgroup_data->cgroup_proc_length = cgroup_proc_length; 318 payload += cgroup_proc_length; 319 } 320 321 if (FETCH_CGROUPS_FROM_BPF) { 322 cgroup_data->cgroup_full_path_root_pos = -1; 323 void* payload_end_pos = read_full_cgroup_path(proc_kernfs, root_kernfs, payload, 324 &cgroup_data->cgroup_full_path_root_pos); 325 cgroup_data->cgroup_full_length = payload_end_pos - payload; 326 payload = payload_end_pos; 327 } 328 329 return (void*)payload; 330 } 331 332 static INLINE void* populate_var_metadata(struct var_metadata_t* metadata, 333 struct task_struct* task, 334 u32 pid, void* payload) 335 { 336 u64 uid_gid = bpf_get_current_uid_gid(); 337 338 metadata->uid = (u32)uid_gid; 339 metadata->gid = uid_gid >> 32; 340 metadata->pid = pid; 341 metadata->exec_id = BPF_CORE_READ(task, self_exec_id); 342 metadata->start_time = BPF_CORE_READ(task, start_time); 343 metadata->comm_length = 0; 344 345 size_t comm_length = bpf_core_read_str(payload, TASK_COMM_LEN, &task->comm); 346 if (bpf_cmp_likely(comm_length, <=, TASK_COMM_LEN)) { 347 metadata->comm_length = comm_length; 348 payload += comm_length; 349 } 350 351 return (void*)payload; 352 } 353 354 static INLINE struct var_kill_data_t* 355 get_var_kill_data(struct pt_regs* ctx, int spid, int tpid, int sig) 356 { 357 int zero = 0; 358 struct var_kill_data_t* kill_data = bpf_map_lookup_elem(&data_heap, &zero); 359 360 if (kill_data == NULL) 361 return NULL; 362 struct task_struct* task = (struct task_struct*)bpf_get_current_task(); 363 364 void* payload = populate_var_metadata(&kill_data->meta, task, spid, kill_data->payload); 365 payload = populate_cgroup_info(&kill_data->cgroup_data, task, payload); 366 size_t payload_length = payload - (void*)kill_data->payload; 367 kill_data->payload_length = payload_length; 368 populate_ancestors(task, &kill_data->ancestors_info); 369 kill_data->meta.type = KILL_EVENT; 370 kill_data->kill_target_pid = tpid; 371 kill_data->kill_sig = sig; 372 kill_data->kill_count = 1; 373 kill_data->last_kill_time = bpf_ktime_get_ns(); 374 return kill_data; 375 } 376 377 static INLINE int trace_var_sys_kill(void* ctx, int tpid, int sig) 378 { 379 if ((KILL_SIGNALS & (1ULL << sig)) == 0) 380 return 0; 381 382 u32 spid = get_userspace_pid(); 383 struct var_kill_data_arr_t* arr_struct = bpf_map_lookup_elem(&var_tpid_to_data, &tpid); 384 385 if (arr_struct == NULL) { 386 struct var_kill_data_t* kill_data = get_var_kill_data(ctx, spid, tpid, sig); 387 int zero = 0; 388 389 if (kill_data == NULL) 390 return 0; 391 arr_struct = bpf_map_lookup_elem(&data_heap, &zero); 392 if (arr_struct == NULL) 393 return 0; 394 bpf_probe_read_kernel(&arr_struct->array[0], 395 sizeof(arr_struct->array[0]), kill_data); 396 } else { 397 int index = get_var_spid_index(arr_struct, spid); 398 399 if (index == -1) { 400 struct var_kill_data_t* kill_data = 401 get_var_kill_data(ctx, spid, tpid, sig); 402 if (kill_data == NULL) 403 return 0; 404 #ifdef UNROLL 405 #pragma unroll 406 #endif 407 for (int i = 0; i < ARRAY_SIZE(arr_struct->array); i++) 408 if (arr_struct->array[i].meta.pid == 0) { 409 bpf_probe_read_kernel(&arr_struct->array[i], 410 sizeof(arr_struct->array[i]), 411 kill_data); 412 bpf_map_update_elem(&var_tpid_to_data, &tpid, 413 arr_struct, 0); 414 415 return 0; 416 } 417 return 0; 418 } 419 420 struct var_kill_data_t* kill_data = &arr_struct->array[index]; 421 422 u64 delta_sec = 423 (bpf_ktime_get_ns() - kill_data->last_kill_time) / 1000000000; 424 425 if (delta_sec < STALE_INFO) { 426 kill_data->kill_count++; 427 kill_data->last_kill_time = bpf_ktime_get_ns(); 428 bpf_probe_read_kernel(&arr_struct->array[index], 429 sizeof(arr_struct->array[index]), 430 kill_data); 431 } else { 432 struct var_kill_data_t* kill_data = 433 get_var_kill_data(ctx, spid, tpid, sig); 434 if (kill_data == NULL) 435 return 0; 436 bpf_probe_read_kernel(&arr_struct->array[index], 437 sizeof(arr_struct->array[index]), 438 kill_data); 439 } 440 } 441 bpf_map_update_elem(&var_tpid_to_data, &tpid, arr_struct, 0); 442 return 0; 443 } 444 445 static INLINE void bpf_stats_enter(struct bpf_func_stats_ctx* bpf_stat_ctx, 446 enum bpf_function_id func_id) 447 { 448 int func_id_key = func_id; 449 450 bpf_stat_ctx->start_time_ns = bpf_ktime_get_ns(); 451 bpf_stat_ctx->bpf_func_stats_data_val = 452 bpf_map_lookup_elem(&bpf_func_stats, &func_id_key); 453 if (bpf_stat_ctx->bpf_func_stats_data_val) 454 bpf_stat_ctx->bpf_func_stats_data_val->num_executions++; 455 } 456 457 static INLINE void bpf_stats_exit(struct bpf_func_stats_ctx* bpf_stat_ctx) 458 { 459 if (bpf_stat_ctx->bpf_func_stats_data_val) 460 bpf_stat_ctx->bpf_func_stats_data_val->time_elapsed_ns += 461 bpf_ktime_get_ns() - bpf_stat_ctx->start_time_ns; 462 } 463 464 static INLINE void 465 bpf_stats_pre_submit_var_perf_event(struct bpf_func_stats_ctx* bpf_stat_ctx, 466 struct var_metadata_t* meta) 467 { 468 if (bpf_stat_ctx->bpf_func_stats_data_val) { 469 bpf_stat_ctx->bpf_func_stats_data_val->num_perf_events++; 470 meta->bpf_stats_num_perf_events = 471 bpf_stat_ctx->bpf_func_stats_data_val->num_perf_events; 472 } 473 meta->bpf_stats_start_ktime_ns = bpf_stat_ctx->start_time_ns; 474 meta->cpu_id = bpf_get_smp_processor_id(); 475 } 476 477 static INLINE size_t 478 read_absolute_file_path_from_dentry(struct dentry* filp_dentry, void* payload) 479 { 480 size_t length = 0; 481 size_t filepart_length; 482 struct dentry* parent_dentry; 483 484 #ifdef UNROLL 485 #pragma unroll 486 #endif 487 for (int i = 0; i < MAX_PATH_DEPTH; i++) { 488 filepart_length = 489 bpf_probe_read_kernel_str(payload, MAX_PATH, 490 BPF_CORE_READ(filp_dentry, d_name.name)); 491 bpf_nop_mov(filepart_length); 492 if (bpf_cmp_unlikely(filepart_length, >, MAX_PATH)) 493 break; 494 payload += filepart_length; 495 length += filepart_length; 496 497 parent_dentry = BPF_CORE_READ(filp_dentry, d_parent); 498 if (filp_dentry == parent_dentry) 499 break; 500 filp_dentry = parent_dentry; 501 } 502 503 return length; 504 } 505 506 static INLINE bool 507 is_ancestor_in_allowed_inodes(struct dentry* filp_dentry) 508 { 509 struct dentry* parent_dentry; 510 #ifdef UNROLL 511 #pragma unroll 512 #endif 513 for (int i = 0; i < MAX_PATH_DEPTH; i++) { 514 u64 dir_ino = BPF_CORE_READ(filp_dentry, d_inode, i_ino); 515 bool* allowed_dir = bpf_map_lookup_elem(&allowed_directory_inodes, &dir_ino); 516 517 if (allowed_dir != NULL) 518 return true; 519 parent_dentry = BPF_CORE_READ(filp_dentry, d_parent); 520 if (filp_dentry == parent_dentry) 521 break; 522 filp_dentry = parent_dentry; 523 } 524 return false; 525 } 526 527 static INLINE bool is_dentry_allowed_for_filemod(struct dentry* file_dentry, 528 u32* device_id, 529 u64* file_ino) 530 { 531 u32 dev_id = BPF_CORE_READ(file_dentry, d_sb, s_dev); 532 *device_id = dev_id; 533 bool* allowed_device = bpf_map_lookup_elem(&allowed_devices, &dev_id); 534 535 if (allowed_device == NULL) 536 return false; 537 538 u64 ino = BPF_CORE_READ(file_dentry, d_inode, i_ino); 539 *file_ino = ino; 540 bool* allowed_file = bpf_map_lookup_elem(&allowed_file_inodes, &ino); 541 542 if (allowed_file == NULL) 543 if (!is_ancestor_in_allowed_inodes(BPF_CORE_READ(file_dentry, d_parent))) 544 return false; 545 return true; 546 } 547 548 SEC("kprobe/proc_sys_write") 549 ssize_t BPF_KPROBE(kprobe__proc_sys_write, 550 struct file* filp, const char* buf, 551 size_t count, loff_t* ppos) 552 { 553 struct bpf_func_stats_ctx stats_ctx; 554 bpf_stats_enter(&stats_ctx, profiler_bpf_proc_sys_write); 555 556 u32 pid = get_userspace_pid(); 557 int zero = 0; 558 struct var_sysctl_data_t* sysctl_data = 559 bpf_map_lookup_elem(&data_heap, &zero); 560 if (!sysctl_data) 561 goto out; 562 563 struct task_struct* task = (struct task_struct*)bpf_get_current_task(); 564 sysctl_data->meta.type = SYSCTL_EVENT; 565 void* payload = populate_var_metadata(&sysctl_data->meta, task, pid, sysctl_data->payload); 566 payload = populate_cgroup_info(&sysctl_data->cgroup_data, task, payload); 567 568 populate_ancestors(task, &sysctl_data->ancestors_info); 569 570 sysctl_data->sysctl_val_length = 0; 571 sysctl_data->sysctl_path_length = 0; 572 573 size_t sysctl_val_length = bpf_probe_read_kernel_str(payload, 574 CTL_MAXNAME, buf); 575 if (bpf_cmp_likely(sysctl_val_length, <=, CTL_MAXNAME)) { 576 sysctl_data->sysctl_val_length = sysctl_val_length; 577 payload += sysctl_val_length; 578 } 579 580 size_t sysctl_path_length = 581 bpf_probe_read_kernel_str(payload, MAX_PATH, 582 BPF_CORE_READ(filp, f_path.dentry, 583 d_name.name)); 584 if (bpf_cmp_likely(sysctl_path_length, <=, MAX_PATH)) { 585 sysctl_data->sysctl_path_length = sysctl_path_length; 586 payload += sysctl_path_length; 587 } 588 589 bpf_stats_pre_submit_var_perf_event(&stats_ctx, &sysctl_data->meta); 590 unsigned long data_len = payload - (void*)sysctl_data; 591 data_len = data_len > sizeof(struct var_sysctl_data_t) 592 ? sizeof(struct var_sysctl_data_t) 593 : data_len; 594 bpf_perf_event_output(ctx, &events, BPF_F_CURRENT_CPU, sysctl_data, data_len); 595 out: 596 bpf_stats_exit(&stats_ctx); 597 return 0; 598 } 599 600 SEC("tracepoint/syscalls/sys_enter_kill") 601 int tracepoint__syscalls__sys_enter_kill(struct syscall_trace_enter* ctx) 602 { 603 struct bpf_func_stats_ctx stats_ctx; 604 605 bpf_stats_enter(&stats_ctx, profiler_bpf_sys_enter_kill); 606 int pid = ctx->args[0]; 607 int sig = ctx->args[1]; 608 int ret = trace_var_sys_kill(ctx, pid, sig); 609 bpf_stats_exit(&stats_ctx); 610 return ret; 611 }; 612 613 SEC("raw_tracepoint/sched_process_exit") 614 int raw_tracepoint__sched_process_exit(void* ctx) 615 { 616 int zero = 0; 617 struct bpf_func_stats_ctx stats_ctx; 618 bpf_stats_enter(&stats_ctx, profiler_bpf_sched_process_exit); 619 620 u32 tpid = get_userspace_pid(); 621 622 struct var_kill_data_arr_t* arr_struct = bpf_map_lookup_elem(&var_tpid_to_data, &tpid); 623 struct var_kill_data_t* kill_data = bpf_map_lookup_elem(&data_heap, &zero); 624 625 if (arr_struct == NULL || kill_data == NULL) 626 goto out; 627 628 struct task_struct* task = (struct task_struct*)bpf_get_current_task(); 629 struct kernfs_node* proc_kernfs = BPF_CORE_READ(task, cgroups, dfl_cgrp, kn); 630 631 #ifdef UNROLL 632 #pragma unroll 633 #endif 634 for (int i = 0; i < ARRAY_SIZE(arr_struct->array); i++) { 635 struct var_kill_data_t* past_kill_data = &arr_struct->array[i]; 636 637 if (past_kill_data != NULL && past_kill_data->kill_target_pid == (pid_t)tpid) { 638 bpf_probe_read_kernel(kill_data, sizeof(*past_kill_data), 639 past_kill_data); 640 void* payload = kill_data->payload; 641 size_t offset = kill_data->payload_length; 642 if (offset >= MAX_METADATA_PAYLOAD_LEN + MAX_CGROUP_PAYLOAD_LEN) 643 return 0; 644 payload += offset; 645 646 kill_data->kill_target_name_length = 0; 647 kill_data->kill_target_cgroup_proc_length = 0; 648 649 size_t comm_length = bpf_core_read_str(payload, TASK_COMM_LEN, &task->comm); 650 if (bpf_cmp_likely(comm_length, <=, TASK_COMM_LEN)) { 651 kill_data->kill_target_name_length = comm_length; 652 payload += comm_length; 653 } 654 655 size_t cgroup_proc_length = 656 bpf_probe_read_kernel_str(payload, 657 KILL_TARGET_LEN, 658 BPF_CORE_READ(proc_kernfs, name)); 659 if (bpf_cmp_likely(cgroup_proc_length, <=, KILL_TARGET_LEN)) { 660 kill_data->kill_target_cgroup_proc_length = cgroup_proc_length; 661 payload += cgroup_proc_length; 662 } 663 664 bpf_stats_pre_submit_var_perf_event(&stats_ctx, &kill_data->meta); 665 unsigned long data_len = (void*)payload - (void*)kill_data; 666 data_len = data_len > sizeof(struct var_kill_data_t) 667 ? sizeof(struct var_kill_data_t) 668 : data_len; 669 bpf_perf_event_output(ctx, &events, BPF_F_CURRENT_CPU, kill_data, data_len); 670 } 671 } 672 bpf_map_delete_elem(&var_tpid_to_data, &tpid); 673 out: 674 bpf_stats_exit(&stats_ctx); 675 return 0; 676 } 677 678 SEC("raw_tracepoint/sched_process_exec") 679 int raw_tracepoint__sched_process_exec(struct bpf_raw_tracepoint_args* ctx) 680 { 681 struct bpf_func_stats_ctx stats_ctx; 682 bpf_stats_enter(&stats_ctx, profiler_bpf_sched_process_exec); 683 684 struct linux_binprm* bprm = (struct linux_binprm*)ctx->args[2]; 685 u64 inode = BPF_CORE_READ(bprm, file, f_inode, i_ino); 686 687 bool* should_filter_binprm = bpf_map_lookup_elem(&disallowed_exec_inodes, &inode); 688 if (should_filter_binprm != NULL) 689 goto out; 690 691 int zero = 0; 692 struct var_exec_data_t* proc_exec_data = bpf_map_lookup_elem(&data_heap, &zero); 693 if (!proc_exec_data) 694 goto out; 695 696 if (INODE_FILTER && inode != INODE_FILTER) 697 return 0; 698 699 u32 pid = get_userspace_pid(); 700 struct task_struct* task = (struct task_struct*)bpf_get_current_task(); 701 702 proc_exec_data->meta.type = EXEC_EVENT; 703 proc_exec_data->bin_path_length = 0; 704 proc_exec_data->cmdline_length = 0; 705 proc_exec_data->environment_length = 0; 706 void* payload = populate_var_metadata(&proc_exec_data->meta, task, pid, 707 proc_exec_data->payload); 708 payload = populate_cgroup_info(&proc_exec_data->cgroup_data, task, payload); 709 710 struct task_struct* parent_task = BPF_CORE_READ(task, real_parent); 711 proc_exec_data->parent_pid = BPF_CORE_READ(parent_task, tgid); 712 proc_exec_data->parent_uid = BPF_CORE_READ(parent_task, real_cred, uid.val); 713 proc_exec_data->parent_exec_id = BPF_CORE_READ(parent_task, self_exec_id); 714 proc_exec_data->parent_start_time = BPF_CORE_READ(parent_task, start_time); 715 716 const char* filename = BPF_CORE_READ(bprm, filename); 717 size_t bin_path_length = 718 bpf_probe_read_kernel_str(payload, MAX_FILENAME_LEN, filename); 719 if (bpf_cmp_likely(bin_path_length, <=, MAX_FILENAME_LEN)) { 720 proc_exec_data->bin_path_length = bin_path_length; 721 payload += bin_path_length; 722 } 723 724 void* arg_start = (void*)BPF_CORE_READ(task, mm, arg_start); 725 void* arg_end = (void*)BPF_CORE_READ(task, mm, arg_end); 726 unsigned int cmdline_length = probe_read_lim(payload, arg_start, 727 arg_end - arg_start, MAX_ARGS_LEN); 728 729 if (bpf_cmp_likely(cmdline_length, <=, MAX_ARGS_LEN)) { 730 proc_exec_data->cmdline_length = cmdline_length; 731 payload += cmdline_length; 732 } 733 734 if (READ_ENVIRON_FROM_EXEC) { 735 void* env_start = (void*)BPF_CORE_READ(task, mm, env_start); 736 void* env_end = (void*)BPF_CORE_READ(task, mm, env_end); 737 unsigned long env_len = probe_read_lim(payload, env_start, 738 env_end - env_start, MAX_ENVIRON_LEN); 739 if (cmdline_length <= MAX_ENVIRON_LEN) { 740 proc_exec_data->environment_length = env_len; 741 payload += env_len; 742 } 743 } 744 745 bpf_stats_pre_submit_var_perf_event(&stats_ctx, &proc_exec_data->meta); 746 unsigned long data_len = payload - (void*)proc_exec_data; 747 data_len = data_len > sizeof(struct var_exec_data_t) 748 ? sizeof(struct var_exec_data_t) 749 : data_len; 750 bpf_perf_event_output(ctx, &events, BPF_F_CURRENT_CPU, proc_exec_data, data_len); 751 out: 752 bpf_stats_exit(&stats_ctx); 753 return 0; 754 } 755 756 SEC("kretprobe/do_filp_open") 757 int kprobe_ret__do_filp_open(struct pt_regs* ctx) 758 { 759 struct bpf_func_stats_ctx stats_ctx; 760 bpf_stats_enter(&stats_ctx, profiler_bpf_do_filp_open_ret); 761 762 struct file* filp = (struct file*)PT_REGS_RC_CORE(ctx); 763 764 if (filp == NULL || IS_ERR(filp)) 765 goto out; 766 unsigned int flags = BPF_CORE_READ(filp, f_flags); 767 if ((flags & (O_RDWR | O_WRONLY)) == 0) 768 goto out; 769 if ((flags & O_TMPFILE) > 0) 770 goto out; 771 struct inode* file_inode = BPF_CORE_READ(filp, f_inode); 772 umode_t mode = BPF_CORE_READ(file_inode, i_mode); 773 if (S_ISDIR(mode) || S_ISCHR(mode) || S_ISBLK(mode) || S_ISFIFO(mode) || 774 S_ISSOCK(mode)) 775 goto out; 776 777 struct dentry* filp_dentry = BPF_CORE_READ(filp, f_path.dentry); 778 u32 device_id = 0; 779 u64 file_ino = 0; 780 if (!is_dentry_allowed_for_filemod(filp_dentry, &device_id, &file_ino)) 781 goto out; 782 783 int zero = 0; 784 struct var_filemod_data_t* filemod_data = bpf_map_lookup_elem(&data_heap, &zero); 785 if (!filemod_data) 786 goto out; 787 788 u32 pid = get_userspace_pid(); 789 struct task_struct* task = (struct task_struct*)bpf_get_current_task(); 790 791 filemod_data->meta.type = FILEMOD_EVENT; 792 filemod_data->fmod_type = FMOD_OPEN; 793 filemod_data->dst_flags = flags; 794 filemod_data->src_inode = 0; 795 filemod_data->dst_inode = file_ino; 796 filemod_data->src_device_id = 0; 797 filemod_data->dst_device_id = device_id; 798 filemod_data->src_filepath_length = 0; 799 filemod_data->dst_filepath_length = 0; 800 801 void* payload = populate_var_metadata(&filemod_data->meta, task, pid, 802 filemod_data->payload); 803 payload = populate_cgroup_info(&filemod_data->cgroup_data, task, payload); 804 805 size_t len = read_absolute_file_path_from_dentry(filp_dentry, payload); 806 if (bpf_cmp_likely(len, <=, MAX_FILEPATH_LENGTH)) { 807 payload += len; 808 filemod_data->dst_filepath_length = len; 809 } 810 bpf_stats_pre_submit_var_perf_event(&stats_ctx, &filemod_data->meta); 811 unsigned long data_len = payload - (void*)filemod_data; 812 data_len = data_len > sizeof(*filemod_data) ? sizeof(*filemod_data) : data_len; 813 bpf_perf_event_output(ctx, &events, BPF_F_CURRENT_CPU, filemod_data, data_len); 814 out: 815 bpf_stats_exit(&stats_ctx); 816 return 0; 817 } 818 819 SEC("kprobe/vfs_link") 820 int BPF_KPROBE(kprobe__vfs_link, 821 struct dentry* old_dentry, struct mnt_idmap *idmap, 822 struct inode* dir, struct dentry* new_dentry, 823 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 if (bpf_cmp_likely(len, <=, MAX_FILEPATH_LENGTH)) { 860 payload += len; 861 filemod_data->src_filepath_length = len; 862 } 863 864 len = read_absolute_file_path_from_dentry(new_dentry, payload); 865 if (bpf_cmp_likely(len, <=, MAX_FILEPATH_LENGTH)) { 866 payload += len; 867 filemod_data->dst_filepath_length = len; 868 } 869 870 bpf_stats_pre_submit_var_perf_event(&stats_ctx, &filemod_data->meta); 871 unsigned long data_len = payload - (void*)filemod_data; 872 data_len = data_len > sizeof(*filemod_data) ? sizeof(*filemod_data) : data_len; 873 bpf_perf_event_output(ctx, &events, BPF_F_CURRENT_CPU, filemod_data, data_len); 874 out: 875 bpf_stats_exit(&stats_ctx); 876 return 0; 877 } 878 879 SEC("kprobe/vfs_symlink") 880 int BPF_KPROBE(kprobe__vfs_symlink, struct inode* dir, struct dentry* dentry, 881 const char* oldname) 882 { 883 struct bpf_func_stats_ctx stats_ctx; 884 bpf_stats_enter(&stats_ctx, profiler_bpf_vfs_symlink); 885 886 u32 dst_device_id = 0; 887 u64 dst_file_ino = 0; 888 if (!is_dentry_allowed_for_filemod(dentry, &dst_device_id, &dst_file_ino)) 889 goto out; 890 891 int zero = 0; 892 struct var_filemod_data_t* filemod_data = bpf_map_lookup_elem(&data_heap, &zero); 893 if (!filemod_data) 894 goto out; 895 896 u32 pid = get_userspace_pid(); 897 struct task_struct* task = (struct task_struct*)bpf_get_current_task(); 898 899 filemod_data->meta.type = FILEMOD_EVENT; 900 filemod_data->fmod_type = FMOD_SYMLINK; 901 filemod_data->dst_flags = 0; 902 filemod_data->src_inode = 0; 903 filemod_data->dst_inode = dst_file_ino; 904 filemod_data->src_device_id = 0; 905 filemod_data->dst_device_id = dst_device_id; 906 filemod_data->src_filepath_length = 0; 907 filemod_data->dst_filepath_length = 0; 908 909 void* payload = populate_var_metadata(&filemod_data->meta, task, pid, 910 filemod_data->payload); 911 payload = populate_cgroup_info(&filemod_data->cgroup_data, task, payload); 912 913 size_t len = bpf_probe_read_kernel_str(payload, MAX_FILEPATH_LENGTH, 914 oldname); 915 if (bpf_cmp_likely(len, <=, MAX_FILEPATH_LENGTH)) { 916 payload += len; 917 filemod_data->src_filepath_length = len; 918 } 919 len = read_absolute_file_path_from_dentry(dentry, payload); 920 if (bpf_cmp_likely(len, <=, MAX_FILEPATH_LENGTH)) { 921 payload += len; 922 filemod_data->dst_filepath_length = len; 923 } 924 bpf_stats_pre_submit_var_perf_event(&stats_ctx, &filemod_data->meta); 925 unsigned long data_len = payload - (void*)filemod_data; 926 data_len = data_len > sizeof(*filemod_data) ? sizeof(*filemod_data) : data_len; 927 bpf_perf_event_output(ctx, &events, BPF_F_CURRENT_CPU, filemod_data, data_len); 928 out: 929 bpf_stats_exit(&stats_ctx); 930 return 0; 931 } 932 933 SEC("raw_tracepoint/sched_process_fork") 934 int raw_tracepoint__sched_process_fork(struct bpf_raw_tracepoint_args* ctx) 935 { 936 struct bpf_func_stats_ctx stats_ctx; 937 bpf_stats_enter(&stats_ctx, profiler_bpf_sched_process_fork); 938 939 int zero = 0; 940 struct var_fork_data_t* fork_data = bpf_map_lookup_elem(&data_heap, &zero); 941 if (!fork_data) 942 goto out; 943 944 struct task_struct* parent = (struct task_struct*)ctx->args[0]; 945 struct task_struct* child = (struct task_struct*)ctx->args[1]; 946 fork_data->meta.type = FORK_EVENT; 947 948 void* payload = populate_var_metadata(&fork_data->meta, child, 949 BPF_CORE_READ(child, pid), fork_data->payload); 950 fork_data->parent_pid = BPF_CORE_READ(parent, pid); 951 fork_data->parent_exec_id = BPF_CORE_READ(parent, self_exec_id); 952 fork_data->parent_start_time = BPF_CORE_READ(parent, start_time); 953 bpf_stats_pre_submit_var_perf_event(&stats_ctx, &fork_data->meta); 954 955 unsigned long data_len = payload - (void*)fork_data; 956 data_len = data_len > sizeof(*fork_data) ? sizeof(*fork_data) : data_len; 957 bpf_perf_event_output(ctx, &events, BPF_F_CURRENT_CPU, fork_data, data_len); 958 out: 959 bpf_stats_exit(&stats_ctx); 960 return 0; 961 } 962 char _license[] SEC("license") = "GPL"; 963