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