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 
IS_ERR(const void * ptr)139 static INLINE bool IS_ERR(const void* ptr)
140 {
141 	return IS_ERR_VALUE((unsigned long)ptr);
142 }
143 
get_userspace_pid()144 static INLINE u32 get_userspace_pid()
145 {
146 	return bpf_get_current_pid_tgid() >> 32;
147 }
148 
is_init_process(u32 tgid)149 static INLINE bool is_init_process(u32 tgid)
150 {
151 	return tgid == 1 || tgid == 0;
152 }
153 
154 static INLINE unsigned long
probe_read_lim(void * dst,void * src,unsigned long len,unsigned long max)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 
get_var_spid_index(struct var_kill_data_arr_t * arr_struct,int spid)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 
populate_ancestors(struct task_struct * task,struct ancestors_data_t * ancestors_data)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 
read_full_cgroup_path(struct kernfs_node * cgroup_node,struct kernfs_node * cgroup_root_node,void * payload,int * root_pos)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 
get_inode_from_kernfs(struct kernfs_node * node)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 
populate_cgroup_info(struct cgroup_data_t * cgroup_data,struct task_struct * task,void * payload)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_str(payload, MAX_PATH, BPF_CORE_READ(root_kernfs, name));
307 	barrier_var(cgroup_root_length);
308 	if (cgroup_root_length <= MAX_PATH) {
309 		barrier_var(cgroup_root_length);
310 		cgroup_data->cgroup_root_length = cgroup_root_length;
311 		payload += cgroup_root_length;
312 	}
313 
314 	size_t cgroup_proc_length =
315 		bpf_probe_read_str(payload, MAX_PATH, BPF_CORE_READ(proc_kernfs, name));
316 	barrier_var(cgroup_proc_length);
317 	if (cgroup_proc_length <= MAX_PATH) {
318 		barrier_var(cgroup_proc_length);
319 		cgroup_data->cgroup_proc_length = cgroup_proc_length;
320 		payload += cgroup_proc_length;
321 	}
322 
323 	if (FETCH_CGROUPS_FROM_BPF) {
324 		cgroup_data->cgroup_full_path_root_pos = -1;
325 		void* payload_end_pos = read_full_cgroup_path(proc_kernfs, root_kernfs, payload,
326 							      &cgroup_data->cgroup_full_path_root_pos);
327 		cgroup_data->cgroup_full_length = payload_end_pos - payload;
328 		payload = payload_end_pos;
329 	}
330 
331 	return (void*)payload;
332 }
333 
populate_var_metadata(struct var_metadata_t * metadata,struct task_struct * task,u32 pid,void * payload)334 static INLINE void* populate_var_metadata(struct var_metadata_t* metadata,
335 					  struct task_struct* task,
336 					  u32 pid, void* payload)
337 {
338 	u64 uid_gid = bpf_get_current_uid_gid();
339 
340 	metadata->uid = (u32)uid_gid;
341 	metadata->gid = uid_gid >> 32;
342 	metadata->pid = pid;
343 	metadata->exec_id = BPF_CORE_READ(task, self_exec_id);
344 	metadata->start_time = BPF_CORE_READ(task, start_time);
345 	metadata->comm_length = 0;
346 
347 	size_t comm_length = bpf_core_read_str(payload, TASK_COMM_LEN, &task->comm);
348 	barrier_var(comm_length);
349 	if (comm_length <= TASK_COMM_LEN) {
350 		barrier_var(comm_length);
351 		metadata->comm_length = comm_length;
352 		payload += comm_length;
353 	}
354 
355 	return (void*)payload;
356 }
357 
358 static INLINE struct var_kill_data_t*
get_var_kill_data(struct pt_regs * ctx,int spid,int tpid,int sig)359 get_var_kill_data(struct pt_regs* ctx, int spid, int tpid, int sig)
360 {
361 	int zero = 0;
362 	struct var_kill_data_t* kill_data = bpf_map_lookup_elem(&data_heap, &zero);
363 
364 	if (kill_data == NULL)
365 		return NULL;
366 	struct task_struct* task = (struct task_struct*)bpf_get_current_task();
367 
368 	void* payload = populate_var_metadata(&kill_data->meta, task, spid, kill_data->payload);
369 	payload = populate_cgroup_info(&kill_data->cgroup_data, task, payload);
370 	size_t payload_length = payload - (void*)kill_data->payload;
371 	kill_data->payload_length = payload_length;
372 	populate_ancestors(task, &kill_data->ancestors_info);
373 	kill_data->meta.type = KILL_EVENT;
374 	kill_data->kill_target_pid = tpid;
375 	kill_data->kill_sig = sig;
376 	kill_data->kill_count = 1;
377 	kill_data->last_kill_time = bpf_ktime_get_ns();
378 	return kill_data;
379 }
380 
trace_var_sys_kill(void * ctx,int tpid,int sig)381 static INLINE int trace_var_sys_kill(void* ctx, int tpid, int sig)
382 {
383 	if ((KILL_SIGNALS & (1ULL << sig)) == 0)
384 		return 0;
385 
386 	u32 spid = get_userspace_pid();
387 	struct var_kill_data_arr_t* arr_struct = bpf_map_lookup_elem(&var_tpid_to_data, &tpid);
388 
389 	if (arr_struct == NULL) {
390 		struct var_kill_data_t* kill_data = get_var_kill_data(ctx, spid, tpid, sig);
391 		int zero = 0;
392 
393 		if (kill_data == NULL)
394 			return 0;
395 		arr_struct = bpf_map_lookup_elem(&data_heap, &zero);
396 		if (arr_struct == NULL)
397 			return 0;
398 		bpf_probe_read(&arr_struct->array[0], sizeof(arr_struct->array[0]), kill_data);
399 	} else {
400 		int index = get_var_spid_index(arr_struct, spid);
401 
402 		if (index == -1) {
403 			struct var_kill_data_t* kill_data =
404 				get_var_kill_data(ctx, spid, tpid, sig);
405 			if (kill_data == NULL)
406 				return 0;
407 #ifdef UNROLL
408 #pragma unroll
409 #endif
410 			for (int i = 0; i < ARRAY_SIZE(arr_struct->array); i++)
411 				if (arr_struct->array[i].meta.pid == 0) {
412 					bpf_probe_read(&arr_struct->array[i],
413 						       sizeof(arr_struct->array[i]), kill_data);
414 					bpf_map_update_elem(&var_tpid_to_data, &tpid,
415 							    arr_struct, 0);
416 
417 					return 0;
418 				}
419 			return 0;
420 		}
421 
422 		struct var_kill_data_t* kill_data = &arr_struct->array[index];
423 
424 		u64 delta_sec =
425 			(bpf_ktime_get_ns() - kill_data->last_kill_time) / 1000000000;
426 
427 		if (delta_sec < STALE_INFO) {
428 			kill_data->kill_count++;
429 			kill_data->last_kill_time = bpf_ktime_get_ns();
430 			bpf_probe_read(&arr_struct->array[index],
431 				       sizeof(arr_struct->array[index]),
432 				       kill_data);
433 		} else {
434 			struct var_kill_data_t* kill_data =
435 				get_var_kill_data(ctx, spid, tpid, sig);
436 			if (kill_data == NULL)
437 				return 0;
438 			bpf_probe_read(&arr_struct->array[index],
439 				       sizeof(arr_struct->array[index]),
440 				       kill_data);
441 		}
442 	}
443 	bpf_map_update_elem(&var_tpid_to_data, &tpid, arr_struct, 0);
444 	return 0;
445 }
446 
bpf_stats_enter(struct bpf_func_stats_ctx * bpf_stat_ctx,enum bpf_function_id func_id)447 static INLINE void bpf_stats_enter(struct bpf_func_stats_ctx* bpf_stat_ctx,
448 				   enum bpf_function_id func_id)
449 {
450 	int func_id_key = func_id;
451 
452 	bpf_stat_ctx->start_time_ns = bpf_ktime_get_ns();
453 	bpf_stat_ctx->bpf_func_stats_data_val =
454 		bpf_map_lookup_elem(&bpf_func_stats, &func_id_key);
455 	if (bpf_stat_ctx->bpf_func_stats_data_val)
456 		bpf_stat_ctx->bpf_func_stats_data_val->num_executions++;
457 }
458 
bpf_stats_exit(struct bpf_func_stats_ctx * bpf_stat_ctx)459 static INLINE void bpf_stats_exit(struct bpf_func_stats_ctx* bpf_stat_ctx)
460 {
461 	if (bpf_stat_ctx->bpf_func_stats_data_val)
462 		bpf_stat_ctx->bpf_func_stats_data_val->time_elapsed_ns +=
463 			bpf_ktime_get_ns() - bpf_stat_ctx->start_time_ns;
464 }
465 
466 static INLINE void
bpf_stats_pre_submit_var_perf_event(struct bpf_func_stats_ctx * bpf_stat_ctx,struct var_metadata_t * meta)467 bpf_stats_pre_submit_var_perf_event(struct bpf_func_stats_ctx* bpf_stat_ctx,
468 				    struct var_metadata_t* meta)
469 {
470 	if (bpf_stat_ctx->bpf_func_stats_data_val) {
471 		bpf_stat_ctx->bpf_func_stats_data_val->num_perf_events++;
472 		meta->bpf_stats_num_perf_events =
473 			bpf_stat_ctx->bpf_func_stats_data_val->num_perf_events;
474 	}
475 	meta->bpf_stats_start_ktime_ns = bpf_stat_ctx->start_time_ns;
476 	meta->cpu_id = bpf_get_smp_processor_id();
477 }
478 
479 static INLINE size_t
read_absolute_file_path_from_dentry(struct dentry * filp_dentry,void * payload)480 read_absolute_file_path_from_dentry(struct dentry* filp_dentry, void* payload)
481 {
482 	size_t length = 0;
483 	size_t filepart_length;
484 	struct dentry* parent_dentry;
485 
486 #ifdef UNROLL
487 #pragma unroll
488 #endif
489 	for (int i = 0; i < MAX_PATH_DEPTH; i++) {
490 		filepart_length = bpf_probe_read_str(payload, MAX_PATH,
491 						     BPF_CORE_READ(filp_dentry, d_name.name));
492 		barrier_var(filepart_length);
493 		if (filepart_length > MAX_PATH)
494 			break;
495 		barrier_var(filepart_length);
496 		payload += filepart_length;
497 		length += filepart_length;
498 
499 		parent_dentry = BPF_CORE_READ(filp_dentry, d_parent);
500 		if (filp_dentry == parent_dentry)
501 			break;
502 		filp_dentry = parent_dentry;
503 	}
504 
505 	return length;
506 }
507 
508 static INLINE bool
is_ancestor_in_allowed_inodes(struct dentry * filp_dentry)509 is_ancestor_in_allowed_inodes(struct dentry* filp_dentry)
510 {
511 	struct dentry* parent_dentry;
512 #ifdef UNROLL
513 #pragma unroll
514 #endif
515 	for (int i = 0; i < MAX_PATH_DEPTH; i++) {
516 		u64 dir_ino = BPF_CORE_READ(filp_dentry, d_inode, i_ino);
517 		bool* allowed_dir = bpf_map_lookup_elem(&allowed_directory_inodes, &dir_ino);
518 
519 		if (allowed_dir != NULL)
520 			return true;
521 		parent_dentry = BPF_CORE_READ(filp_dentry, d_parent);
522 		if (filp_dentry == parent_dentry)
523 			break;
524 		filp_dentry = parent_dentry;
525 	}
526 	return false;
527 }
528 
is_dentry_allowed_for_filemod(struct dentry * file_dentry,u32 * device_id,u64 * file_ino)529 static INLINE bool is_dentry_allowed_for_filemod(struct dentry* file_dentry,
530 						 u32* device_id,
531 						 u64* file_ino)
532 {
533 	u32 dev_id = BPF_CORE_READ(file_dentry, d_sb, s_dev);
534 	*device_id = dev_id;
535 	bool* allowed_device = bpf_map_lookup_elem(&allowed_devices, &dev_id);
536 
537 	if (allowed_device == NULL)
538 		return false;
539 
540 	u64 ino = BPF_CORE_READ(file_dentry, d_inode, i_ino);
541 	*file_ino = ino;
542 	bool* allowed_file = bpf_map_lookup_elem(&allowed_file_inodes, &ino);
543 
544 	if (allowed_file == NULL)
545 		if (!is_ancestor_in_allowed_inodes(BPF_CORE_READ(file_dentry, d_parent)))
546 			return false;
547 	return true;
548 }
549 
550 SEC("kprobe/proc_sys_write")
BPF_KPROBE(kprobe__proc_sys_write,struct file * filp,const char * buf,size_t count,loff_t * ppos)551 ssize_t BPF_KPROBE(kprobe__proc_sys_write,
552 		   struct file* filp, const char* buf,
553 		   size_t count, loff_t* ppos)
554 {
555 	struct bpf_func_stats_ctx stats_ctx;
556 	bpf_stats_enter(&stats_ctx, profiler_bpf_proc_sys_write);
557 
558 	u32 pid = get_userspace_pid();
559 	int zero = 0;
560 	struct var_sysctl_data_t* sysctl_data =
561 		bpf_map_lookup_elem(&data_heap, &zero);
562 	if (!sysctl_data)
563 		goto out;
564 
565 	struct task_struct* task = (struct task_struct*)bpf_get_current_task();
566 	sysctl_data->meta.type = SYSCTL_EVENT;
567 	void* payload = populate_var_metadata(&sysctl_data->meta, task, pid, sysctl_data->payload);
568 	payload = populate_cgroup_info(&sysctl_data->cgroup_data, task, payload);
569 
570 	populate_ancestors(task, &sysctl_data->ancestors_info);
571 
572 	sysctl_data->sysctl_val_length = 0;
573 	sysctl_data->sysctl_path_length = 0;
574 
575 	size_t sysctl_val_length = bpf_probe_read_str(payload, CTL_MAXNAME, buf);
576 	barrier_var(sysctl_val_length);
577 	if (sysctl_val_length <= CTL_MAXNAME) {
578 		barrier_var(sysctl_val_length);
579 		sysctl_data->sysctl_val_length = sysctl_val_length;
580 		payload += sysctl_val_length;
581 	}
582 
583 	size_t sysctl_path_length = bpf_probe_read_str(payload, MAX_PATH,
584 						       BPF_CORE_READ(filp, f_path.dentry, d_name.name));
585 	barrier_var(sysctl_path_length);
586 	if (sysctl_path_length <= MAX_PATH) {
587 		barrier_var(sysctl_path_length);
588 		sysctl_data->sysctl_path_length = sysctl_path_length;
589 		payload += sysctl_path_length;
590 	}
591 
592 	bpf_stats_pre_submit_var_perf_event(&stats_ctx, &sysctl_data->meta);
593 	unsigned long data_len = payload - (void*)sysctl_data;
594 	data_len = data_len > sizeof(struct var_sysctl_data_t)
595 		? sizeof(struct var_sysctl_data_t)
596 		: data_len;
597 	bpf_perf_event_output(ctx, &events, BPF_F_CURRENT_CPU, sysctl_data, data_len);
598 out:
599 	bpf_stats_exit(&stats_ctx);
600 	return 0;
601 }
602 
603 SEC("tracepoint/syscalls/sys_enter_kill")
tracepoint__syscalls__sys_enter_kill(struct trace_event_raw_sys_enter * ctx)604 int tracepoint__syscalls__sys_enter_kill(struct trace_event_raw_sys_enter* ctx)
605 {
606 	struct bpf_func_stats_ctx stats_ctx;
607 
608 	bpf_stats_enter(&stats_ctx, profiler_bpf_sys_enter_kill);
609 	int pid = ctx->args[0];
610 	int sig = ctx->args[1];
611 	int ret = trace_var_sys_kill(ctx, pid, sig);
612 	bpf_stats_exit(&stats_ctx);
613 	return ret;
614 };
615 
616 SEC("raw_tracepoint/sched_process_exit")
raw_tracepoint__sched_process_exit(void * ctx)617 int raw_tracepoint__sched_process_exit(void* ctx)
618 {
619 	int zero = 0;
620 	struct bpf_func_stats_ctx stats_ctx;
621 	bpf_stats_enter(&stats_ctx, profiler_bpf_sched_process_exit);
622 
623 	u32 tpid = get_userspace_pid();
624 
625 	struct var_kill_data_arr_t* arr_struct = bpf_map_lookup_elem(&var_tpid_to_data, &tpid);
626 	struct var_kill_data_t* kill_data = bpf_map_lookup_elem(&data_heap, &zero);
627 
628 	if (arr_struct == NULL || kill_data == NULL)
629 		goto out;
630 
631 	struct task_struct* task = (struct task_struct*)bpf_get_current_task();
632 	struct kernfs_node* proc_kernfs = BPF_CORE_READ(task, cgroups, dfl_cgrp, kn);
633 
634 #ifdef UNROLL
635 #pragma unroll
636 #endif
637 	for (int i = 0; i < ARRAY_SIZE(arr_struct->array); i++) {
638 		struct var_kill_data_t* past_kill_data = &arr_struct->array[i];
639 
640 		if (past_kill_data != NULL && past_kill_data->kill_target_pid == tpid) {
641 			bpf_probe_read(kill_data, sizeof(*past_kill_data), past_kill_data);
642 			void* payload = kill_data->payload;
643 			size_t offset = kill_data->payload_length;
644 			if (offset >= MAX_METADATA_PAYLOAD_LEN + MAX_CGROUP_PAYLOAD_LEN)
645 				return 0;
646 			payload += offset;
647 
648 			kill_data->kill_target_name_length = 0;
649 			kill_data->kill_target_cgroup_proc_length = 0;
650 
651 			size_t comm_length = bpf_core_read_str(payload, TASK_COMM_LEN, &task->comm);
652 			barrier_var(comm_length);
653 			if (comm_length <= TASK_COMM_LEN) {
654 				barrier_var(comm_length);
655 				kill_data->kill_target_name_length = comm_length;
656 				payload += comm_length;
657 			}
658 
659 			size_t cgroup_proc_length = bpf_probe_read_str(payload, KILL_TARGET_LEN,
660 								       BPF_CORE_READ(proc_kernfs, name));
661 			barrier_var(cgroup_proc_length);
662 			if (cgroup_proc_length <= KILL_TARGET_LEN) {
663 				barrier_var(cgroup_proc_length);
664 				kill_data->kill_target_cgroup_proc_length = cgroup_proc_length;
665 				payload += cgroup_proc_length;
666 			}
667 
668 			bpf_stats_pre_submit_var_perf_event(&stats_ctx, &kill_data->meta);
669 			unsigned long data_len = (void*)payload - (void*)kill_data;
670 			data_len = data_len > sizeof(struct var_kill_data_t)
671 				? sizeof(struct var_kill_data_t)
672 				: data_len;
673 			bpf_perf_event_output(ctx, &events, BPF_F_CURRENT_CPU, kill_data, data_len);
674 		}
675 	}
676 	bpf_map_delete_elem(&var_tpid_to_data, &tpid);
677 out:
678 	bpf_stats_exit(&stats_ctx);
679 	return 0;
680 }
681 
682 SEC("raw_tracepoint/sched_process_exec")
raw_tracepoint__sched_process_exec(struct bpf_raw_tracepoint_args * ctx)683 int raw_tracepoint__sched_process_exec(struct bpf_raw_tracepoint_args* ctx)
684 {
685 	struct bpf_func_stats_ctx stats_ctx;
686 	bpf_stats_enter(&stats_ctx, profiler_bpf_sched_process_exec);
687 
688 	struct linux_binprm* bprm = (struct linux_binprm*)ctx->args[2];
689 	u64 inode = BPF_CORE_READ(bprm, file, f_inode, i_ino);
690 
691 	bool* should_filter_binprm = bpf_map_lookup_elem(&disallowed_exec_inodes, &inode);
692 	if (should_filter_binprm != NULL)
693 		goto out;
694 
695 	int zero = 0;
696 	struct var_exec_data_t* proc_exec_data = bpf_map_lookup_elem(&data_heap, &zero);
697 	if (!proc_exec_data)
698 		goto out;
699 
700 	if (INODE_FILTER && inode != INODE_FILTER)
701 		return 0;
702 
703 	u32 pid = get_userspace_pid();
704 	struct task_struct* task = (struct task_struct*)bpf_get_current_task();
705 
706 	proc_exec_data->meta.type = EXEC_EVENT;
707 	proc_exec_data->bin_path_length = 0;
708 	proc_exec_data->cmdline_length = 0;
709 	proc_exec_data->environment_length = 0;
710 	void* payload = populate_var_metadata(&proc_exec_data->meta, task, pid,
711 					      proc_exec_data->payload);
712 	payload = populate_cgroup_info(&proc_exec_data->cgroup_data, task, payload);
713 
714 	struct task_struct* parent_task = BPF_CORE_READ(task, real_parent);
715 	proc_exec_data->parent_pid = BPF_CORE_READ(parent_task, tgid);
716 	proc_exec_data->parent_uid = BPF_CORE_READ(parent_task, real_cred, uid.val);
717 	proc_exec_data->parent_exec_id = BPF_CORE_READ(parent_task, self_exec_id);
718 	proc_exec_data->parent_start_time = BPF_CORE_READ(parent_task, start_time);
719 
720 	const char* filename = BPF_CORE_READ(bprm, filename);
721 	size_t bin_path_length = bpf_probe_read_str(payload, MAX_FILENAME_LEN, filename);
722 	barrier_var(bin_path_length);
723 	if (bin_path_length <= MAX_FILENAME_LEN) {
724 		barrier_var(bin_path_length);
725 		proc_exec_data->bin_path_length = bin_path_length;
726 		payload += bin_path_length;
727 	}
728 
729 	void* arg_start = (void*)BPF_CORE_READ(task, mm, arg_start);
730 	void* arg_end = (void*)BPF_CORE_READ(task, mm, arg_end);
731 	unsigned int cmdline_length = probe_read_lim(payload, arg_start,
732 						     arg_end - arg_start, MAX_ARGS_LEN);
733 
734 	if (cmdline_length <= MAX_ARGS_LEN) {
735 		barrier_var(cmdline_length);
736 		proc_exec_data->cmdline_length = cmdline_length;
737 		payload += cmdline_length;
738 	}
739 
740 	if (READ_ENVIRON_FROM_EXEC) {
741 		void* env_start = (void*)BPF_CORE_READ(task, mm, env_start);
742 		void* env_end = (void*)BPF_CORE_READ(task, mm, env_end);
743 		unsigned long env_len = probe_read_lim(payload, env_start,
744 						       env_end - env_start, MAX_ENVIRON_LEN);
745 		if (cmdline_length <= MAX_ENVIRON_LEN) {
746 			proc_exec_data->environment_length = env_len;
747 			payload += env_len;
748 		}
749 	}
750 
751 	bpf_stats_pre_submit_var_perf_event(&stats_ctx, &proc_exec_data->meta);
752 	unsigned long data_len = payload - (void*)proc_exec_data;
753 	data_len = data_len > sizeof(struct var_exec_data_t)
754 		? sizeof(struct var_exec_data_t)
755 		: data_len;
756 	bpf_perf_event_output(ctx, &events, BPF_F_CURRENT_CPU, proc_exec_data, data_len);
757 out:
758 	bpf_stats_exit(&stats_ctx);
759 	return 0;
760 }
761 
762 SEC("kretprobe/do_filp_open")
kprobe_ret__do_filp_open(struct pt_regs * ctx)763 int kprobe_ret__do_filp_open(struct pt_regs* ctx)
764 {
765 	struct bpf_func_stats_ctx stats_ctx;
766 	bpf_stats_enter(&stats_ctx, profiler_bpf_do_filp_open_ret);
767 
768 	struct file* filp = (struct file*)PT_REGS_RC_CORE(ctx);
769 
770 	if (filp == NULL || IS_ERR(filp))
771 		goto out;
772 	unsigned int flags = BPF_CORE_READ(filp, f_flags);
773 	if ((flags & (O_RDWR | O_WRONLY)) == 0)
774 		goto out;
775 	if ((flags & O_TMPFILE) > 0)
776 		goto out;
777 	struct inode* file_inode = BPF_CORE_READ(filp, f_inode);
778 	umode_t mode = BPF_CORE_READ(file_inode, i_mode);
779 	if (S_ISDIR(mode) || S_ISCHR(mode) || S_ISBLK(mode) || S_ISFIFO(mode) ||
780 	    S_ISSOCK(mode))
781 		goto out;
782 
783 	struct dentry* filp_dentry = BPF_CORE_READ(filp, f_path.dentry);
784 	u32 device_id = 0;
785 	u64 file_ino = 0;
786 	if (!is_dentry_allowed_for_filemod(filp_dentry, &device_id, &file_ino))
787 		goto out;
788 
789 	int zero = 0;
790 	struct var_filemod_data_t* filemod_data = bpf_map_lookup_elem(&data_heap, &zero);
791 	if (!filemod_data)
792 		goto out;
793 
794 	u32 pid = get_userspace_pid();
795 	struct task_struct* task = (struct task_struct*)bpf_get_current_task();
796 
797 	filemod_data->meta.type = FILEMOD_EVENT;
798 	filemod_data->fmod_type = FMOD_OPEN;
799 	filemod_data->dst_flags = flags;
800 	filemod_data->src_inode = 0;
801 	filemod_data->dst_inode = file_ino;
802 	filemod_data->src_device_id = 0;
803 	filemod_data->dst_device_id = device_id;
804 	filemod_data->src_filepath_length = 0;
805 	filemod_data->dst_filepath_length = 0;
806 
807 	void* payload = populate_var_metadata(&filemod_data->meta, task, pid,
808 					      filemod_data->payload);
809 	payload = populate_cgroup_info(&filemod_data->cgroup_data, task, payload);
810 
811 	size_t len = read_absolute_file_path_from_dentry(filp_dentry, payload);
812 	barrier_var(len);
813 	if (len <= MAX_FILEPATH_LENGTH) {
814 		barrier_var(len);
815 		payload += len;
816 		filemod_data->dst_filepath_length = len;
817 	}
818 	bpf_stats_pre_submit_var_perf_event(&stats_ctx, &filemod_data->meta);
819 	unsigned long data_len = payload - (void*)filemod_data;
820 	data_len = data_len > sizeof(*filemod_data) ? sizeof(*filemod_data) : data_len;
821 	bpf_perf_event_output(ctx, &events, BPF_F_CURRENT_CPU, filemod_data, data_len);
822 out:
823 	bpf_stats_exit(&stats_ctx);
824 	return 0;
825 }
826 
827 SEC("kprobe/vfs_link")
BPF_KPROBE(kprobe__vfs_link,struct dentry * old_dentry,struct user_namespace * mnt_userns,struct inode * dir,struct dentry * new_dentry,struct inode ** delegated_inode)828 int BPF_KPROBE(kprobe__vfs_link,
829 	       struct dentry* old_dentry, struct user_namespace *mnt_userns,
830 	       struct inode* dir, struct dentry* new_dentry,
831 	       struct inode** delegated_inode)
832 {
833 	struct bpf_func_stats_ctx stats_ctx;
834 	bpf_stats_enter(&stats_ctx, profiler_bpf_vfs_link);
835 
836 	u32 src_device_id = 0;
837 	u64 src_file_ino = 0;
838 	u32 dst_device_id = 0;
839 	u64 dst_file_ino = 0;
840 	if (!is_dentry_allowed_for_filemod(old_dentry, &src_device_id, &src_file_ino) &&
841 	    !is_dentry_allowed_for_filemod(new_dentry, &dst_device_id, &dst_file_ino))
842 		goto out;
843 
844 	int zero = 0;
845 	struct var_filemod_data_t* filemod_data = bpf_map_lookup_elem(&data_heap, &zero);
846 	if (!filemod_data)
847 		goto out;
848 
849 	u32 pid = get_userspace_pid();
850 	struct task_struct* task = (struct task_struct*)bpf_get_current_task();
851 
852 	filemod_data->meta.type = FILEMOD_EVENT;
853 	filemod_data->fmod_type = FMOD_LINK;
854 	filemod_data->dst_flags = 0;
855 	filemod_data->src_inode = src_file_ino;
856 	filemod_data->dst_inode = dst_file_ino;
857 	filemod_data->src_device_id = src_device_id;
858 	filemod_data->dst_device_id = dst_device_id;
859 	filemod_data->src_filepath_length = 0;
860 	filemod_data->dst_filepath_length = 0;
861 
862 	void* payload = populate_var_metadata(&filemod_data->meta, task, pid,
863 					      filemod_data->payload);
864 	payload = populate_cgroup_info(&filemod_data->cgroup_data, task, payload);
865 
866 	size_t len = read_absolute_file_path_from_dentry(old_dentry, payload);
867 	barrier_var(len);
868 	if (len <= MAX_FILEPATH_LENGTH) {
869 		barrier_var(len);
870 		payload += len;
871 		filemod_data->src_filepath_length = len;
872 	}
873 
874 	len = read_absolute_file_path_from_dentry(new_dentry, payload);
875 	barrier_var(len);
876 	if (len <= MAX_FILEPATH_LENGTH) {
877 		barrier_var(len);
878 		payload += len;
879 		filemod_data->dst_filepath_length = len;
880 	}
881 
882 	bpf_stats_pre_submit_var_perf_event(&stats_ctx, &filemod_data->meta);
883 	unsigned long data_len = payload - (void*)filemod_data;
884 	data_len = data_len > sizeof(*filemod_data) ? sizeof(*filemod_data) : data_len;
885 	bpf_perf_event_output(ctx, &events, BPF_F_CURRENT_CPU, filemod_data, data_len);
886 out:
887 	bpf_stats_exit(&stats_ctx);
888 	return 0;
889 }
890 
891 SEC("kprobe/vfs_symlink")
BPF_KPROBE(kprobe__vfs_symlink,struct inode * dir,struct dentry * dentry,const char * oldname)892 int BPF_KPROBE(kprobe__vfs_symlink, struct inode* dir, struct dentry* dentry,
893 	       const char* oldname)
894 {
895 	struct bpf_func_stats_ctx stats_ctx;
896 	bpf_stats_enter(&stats_ctx, profiler_bpf_vfs_symlink);
897 
898 	u32 dst_device_id = 0;
899 	u64 dst_file_ino = 0;
900 	if (!is_dentry_allowed_for_filemod(dentry, &dst_device_id, &dst_file_ino))
901 		goto out;
902 
903 	int zero = 0;
904 	struct var_filemod_data_t* filemod_data = bpf_map_lookup_elem(&data_heap, &zero);
905 	if (!filemod_data)
906 		goto out;
907 
908 	u32 pid = get_userspace_pid();
909 	struct task_struct* task = (struct task_struct*)bpf_get_current_task();
910 
911 	filemod_data->meta.type = FILEMOD_EVENT;
912 	filemod_data->fmod_type = FMOD_SYMLINK;
913 	filemod_data->dst_flags = 0;
914 	filemod_data->src_inode = 0;
915 	filemod_data->dst_inode = dst_file_ino;
916 	filemod_data->src_device_id = 0;
917 	filemod_data->dst_device_id = dst_device_id;
918 	filemod_data->src_filepath_length = 0;
919 	filemod_data->dst_filepath_length = 0;
920 
921 	void* payload = populate_var_metadata(&filemod_data->meta, task, pid,
922 					      filemod_data->payload);
923 	payload = populate_cgroup_info(&filemod_data->cgroup_data, task, payload);
924 
925 	size_t len = bpf_probe_read_str(payload, MAX_FILEPATH_LENGTH, oldname);
926 	barrier_var(len);
927 	if (len <= MAX_FILEPATH_LENGTH) {
928 		barrier_var(len);
929 		payload += len;
930 		filemod_data->src_filepath_length = len;
931 	}
932 	len = read_absolute_file_path_from_dentry(dentry, payload);
933 	barrier_var(len);
934 	if (len <= MAX_FILEPATH_LENGTH) {
935 		barrier_var(len);
936 		payload += len;
937 		filemod_data->dst_filepath_length = len;
938 	}
939 	bpf_stats_pre_submit_var_perf_event(&stats_ctx, &filemod_data->meta);
940 	unsigned long data_len = payload - (void*)filemod_data;
941 	data_len = data_len > sizeof(*filemod_data) ? sizeof(*filemod_data) : data_len;
942 	bpf_perf_event_output(ctx, &events, BPF_F_CURRENT_CPU, filemod_data, data_len);
943 out:
944 	bpf_stats_exit(&stats_ctx);
945 	return 0;
946 }
947 
948 SEC("raw_tracepoint/sched_process_fork")
raw_tracepoint__sched_process_fork(struct bpf_raw_tracepoint_args * ctx)949 int raw_tracepoint__sched_process_fork(struct bpf_raw_tracepoint_args* ctx)
950 {
951 	struct bpf_func_stats_ctx stats_ctx;
952 	bpf_stats_enter(&stats_ctx, profiler_bpf_sched_process_fork);
953 
954 	int zero = 0;
955 	struct var_fork_data_t* fork_data = bpf_map_lookup_elem(&data_heap, &zero);
956 	if (!fork_data)
957 		goto out;
958 
959 	struct task_struct* parent = (struct task_struct*)ctx->args[0];
960 	struct task_struct* child = (struct task_struct*)ctx->args[1];
961 	fork_data->meta.type = FORK_EVENT;
962 
963 	void* payload = populate_var_metadata(&fork_data->meta, child,
964 					      BPF_CORE_READ(child, pid), fork_data->payload);
965 	fork_data->parent_pid = BPF_CORE_READ(parent, pid);
966 	fork_data->parent_exec_id = BPF_CORE_READ(parent, self_exec_id);
967 	fork_data->parent_start_time = BPF_CORE_READ(parent, start_time);
968 	bpf_stats_pre_submit_var_perf_event(&stats_ctx, &fork_data->meta);
969 
970 	unsigned long data_len = payload - (void*)fork_data;
971 	data_len = data_len > sizeof(*fork_data) ? sizeof(*fork_data) : data_len;
972 	bpf_perf_event_output(ctx, &events, BPF_F_CURRENT_CPU, fork_data, data_len);
973 out:
974 	bpf_stats_exit(&stats_ctx);
975 	return 0;
976 }
977 char _license[] SEC("license") = "GPL";
978