xref: /linux/kernel/bpf/task_iter.c (revision e6a901a00822659181c93c86d8bbc2a17779fddc)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (c) 2020 Facebook */
3 
4 #include <linux/init.h>
5 #include <linux/namei.h>
6 #include <linux/pid_namespace.h>
7 #include <linux/fs.h>
8 #include <linux/fdtable.h>
9 #include <linux/filter.h>
10 #include <linux/bpf_mem_alloc.h>
11 #include <linux/btf_ids.h>
12 #include <linux/mm_types.h>
13 #include "mmap_unlock_work.h"
14 
15 static const char * const iter_task_type_names[] = {
16 	"ALL",
17 	"TID",
18 	"PID",
19 };
20 
21 struct bpf_iter_seq_task_common {
22 	struct pid_namespace *ns;
23 	enum bpf_iter_task_type	type;
24 	u32 pid;
25 	u32 pid_visiting;
26 };
27 
28 struct bpf_iter_seq_task_info {
29 	/* The first field must be struct bpf_iter_seq_task_common.
30 	 * this is assumed by {init, fini}_seq_pidns() callback functions.
31 	 */
32 	struct bpf_iter_seq_task_common common;
33 	u32 tid;
34 };
35 
36 static struct task_struct *task_group_seq_get_next(struct bpf_iter_seq_task_common *common,
37 						   u32 *tid,
38 						   bool skip_if_dup_files)
39 {
40 	struct task_struct *task;
41 	struct pid *pid;
42 	u32 next_tid;
43 
44 	if (!*tid) {
45 		/* The first time, the iterator calls this function. */
46 		pid = find_pid_ns(common->pid, common->ns);
47 		task = get_pid_task(pid, PIDTYPE_TGID);
48 		if (!task)
49 			return NULL;
50 
51 		*tid = common->pid;
52 		common->pid_visiting = common->pid;
53 
54 		return task;
55 	}
56 
57 	/* If the control returns to user space and comes back to the
58 	 * kernel again, *tid and common->pid_visiting should be the
59 	 * same for task_seq_start() to pick up the correct task.
60 	 */
61 	if (*tid == common->pid_visiting) {
62 		pid = find_pid_ns(common->pid_visiting, common->ns);
63 		task = get_pid_task(pid, PIDTYPE_PID);
64 
65 		return task;
66 	}
67 
68 	task = find_task_by_pid_ns(common->pid_visiting, common->ns);
69 	if (!task)
70 		return NULL;
71 
72 retry:
73 	task = __next_thread(task);
74 	if (!task)
75 		return NULL;
76 
77 	next_tid = __task_pid_nr_ns(task, PIDTYPE_PID, common->ns);
78 	if (!next_tid)
79 		goto retry;
80 
81 	if (skip_if_dup_files && task->files == task->group_leader->files)
82 		goto retry;
83 
84 	*tid = common->pid_visiting = next_tid;
85 	get_task_struct(task);
86 	return task;
87 }
88 
89 static struct task_struct *task_seq_get_next(struct bpf_iter_seq_task_common *common,
90 					     u32 *tid,
91 					     bool skip_if_dup_files)
92 {
93 	struct task_struct *task = NULL;
94 	struct pid *pid;
95 
96 	if (common->type == BPF_TASK_ITER_TID) {
97 		if (*tid && *tid != common->pid)
98 			return NULL;
99 		rcu_read_lock();
100 		pid = find_pid_ns(common->pid, common->ns);
101 		if (pid) {
102 			task = get_pid_task(pid, PIDTYPE_TGID);
103 			*tid = common->pid;
104 		}
105 		rcu_read_unlock();
106 
107 		return task;
108 	}
109 
110 	if (common->type == BPF_TASK_ITER_TGID) {
111 		rcu_read_lock();
112 		task = task_group_seq_get_next(common, tid, skip_if_dup_files);
113 		rcu_read_unlock();
114 
115 		return task;
116 	}
117 
118 	rcu_read_lock();
119 retry:
120 	pid = find_ge_pid(*tid, common->ns);
121 	if (pid) {
122 		*tid = pid_nr_ns(pid, common->ns);
123 		task = get_pid_task(pid, PIDTYPE_PID);
124 		if (!task) {
125 			++*tid;
126 			goto retry;
127 		} else if (skip_if_dup_files && !thread_group_leader(task) &&
128 			   task->files == task->group_leader->files) {
129 			put_task_struct(task);
130 			task = NULL;
131 			++*tid;
132 			goto retry;
133 		}
134 	}
135 	rcu_read_unlock();
136 
137 	return task;
138 }
139 
140 static void *task_seq_start(struct seq_file *seq, loff_t *pos)
141 {
142 	struct bpf_iter_seq_task_info *info = seq->private;
143 	struct task_struct *task;
144 
145 	task = task_seq_get_next(&info->common, &info->tid, false);
146 	if (!task)
147 		return NULL;
148 
149 	if (*pos == 0)
150 		++*pos;
151 	return task;
152 }
153 
154 static void *task_seq_next(struct seq_file *seq, void *v, loff_t *pos)
155 {
156 	struct bpf_iter_seq_task_info *info = seq->private;
157 	struct task_struct *task;
158 
159 	++*pos;
160 	++info->tid;
161 	put_task_struct((struct task_struct *)v);
162 	task = task_seq_get_next(&info->common, &info->tid, false);
163 	if (!task)
164 		return NULL;
165 
166 	return task;
167 }
168 
169 struct bpf_iter__task {
170 	__bpf_md_ptr(struct bpf_iter_meta *, meta);
171 	__bpf_md_ptr(struct task_struct *, task);
172 };
173 
174 DEFINE_BPF_ITER_FUNC(task, struct bpf_iter_meta *meta, struct task_struct *task)
175 
176 static int __task_seq_show(struct seq_file *seq, struct task_struct *task,
177 			   bool in_stop)
178 {
179 	struct bpf_iter_meta meta;
180 	struct bpf_iter__task ctx;
181 	struct bpf_prog *prog;
182 
183 	meta.seq = seq;
184 	prog = bpf_iter_get_info(&meta, in_stop);
185 	if (!prog)
186 		return 0;
187 
188 	ctx.meta = &meta;
189 	ctx.task = task;
190 	return bpf_iter_run_prog(prog, &ctx);
191 }
192 
193 static int task_seq_show(struct seq_file *seq, void *v)
194 {
195 	return __task_seq_show(seq, v, false);
196 }
197 
198 static void task_seq_stop(struct seq_file *seq, void *v)
199 {
200 	if (!v)
201 		(void)__task_seq_show(seq, v, true);
202 	else
203 		put_task_struct((struct task_struct *)v);
204 }
205 
206 static int bpf_iter_attach_task(struct bpf_prog *prog,
207 				union bpf_iter_link_info *linfo,
208 				struct bpf_iter_aux_info *aux)
209 {
210 	unsigned int flags;
211 	struct pid *pid;
212 	pid_t tgid;
213 
214 	if ((!!linfo->task.tid + !!linfo->task.pid + !!linfo->task.pid_fd) > 1)
215 		return -EINVAL;
216 
217 	aux->task.type = BPF_TASK_ITER_ALL;
218 	if (linfo->task.tid != 0) {
219 		aux->task.type = BPF_TASK_ITER_TID;
220 		aux->task.pid = linfo->task.tid;
221 	}
222 	if (linfo->task.pid != 0) {
223 		aux->task.type = BPF_TASK_ITER_TGID;
224 		aux->task.pid = linfo->task.pid;
225 	}
226 	if (linfo->task.pid_fd != 0) {
227 		aux->task.type = BPF_TASK_ITER_TGID;
228 
229 		pid = pidfd_get_pid(linfo->task.pid_fd, &flags);
230 		if (IS_ERR(pid))
231 			return PTR_ERR(pid);
232 
233 		tgid = pid_nr_ns(pid, task_active_pid_ns(current));
234 		aux->task.pid = tgid;
235 		put_pid(pid);
236 	}
237 
238 	return 0;
239 }
240 
241 static const struct seq_operations task_seq_ops = {
242 	.start	= task_seq_start,
243 	.next	= task_seq_next,
244 	.stop	= task_seq_stop,
245 	.show	= task_seq_show,
246 };
247 
248 struct bpf_iter_seq_task_file_info {
249 	/* The first field must be struct bpf_iter_seq_task_common.
250 	 * this is assumed by {init, fini}_seq_pidns() callback functions.
251 	 */
252 	struct bpf_iter_seq_task_common common;
253 	struct task_struct *task;
254 	u32 tid;
255 	u32 fd;
256 };
257 
258 static struct file *
259 task_file_seq_get_next(struct bpf_iter_seq_task_file_info *info)
260 {
261 	u32 saved_tid = info->tid;
262 	struct task_struct *curr_task;
263 	unsigned int curr_fd = info->fd;
264 
265 	/* If this function returns a non-NULL file object,
266 	 * it held a reference to the task/file.
267 	 * Otherwise, it does not hold any reference.
268 	 */
269 again:
270 	if (info->task) {
271 		curr_task = info->task;
272 		curr_fd = info->fd;
273 	} else {
274 		curr_task = task_seq_get_next(&info->common, &info->tid, true);
275                 if (!curr_task) {
276                         info->task = NULL;
277                         return NULL;
278                 }
279 
280 		/* set info->task */
281 		info->task = curr_task;
282 		if (saved_tid == info->tid)
283 			curr_fd = info->fd;
284 		else
285 			curr_fd = 0;
286 	}
287 
288 	rcu_read_lock();
289 	for (;; curr_fd++) {
290 		struct file *f;
291 		f = task_lookup_next_fdget_rcu(curr_task, &curr_fd);
292 		if (!f)
293 			break;
294 
295 		/* set info->fd */
296 		info->fd = curr_fd;
297 		rcu_read_unlock();
298 		return f;
299 	}
300 
301 	/* the current task is done, go to the next task */
302 	rcu_read_unlock();
303 	put_task_struct(curr_task);
304 
305 	if (info->common.type == BPF_TASK_ITER_TID) {
306 		info->task = NULL;
307 		return NULL;
308 	}
309 
310 	info->task = NULL;
311 	info->fd = 0;
312 	saved_tid = ++(info->tid);
313 	goto again;
314 }
315 
316 static void *task_file_seq_start(struct seq_file *seq, loff_t *pos)
317 {
318 	struct bpf_iter_seq_task_file_info *info = seq->private;
319 	struct file *file;
320 
321 	info->task = NULL;
322 	file = task_file_seq_get_next(info);
323 	if (file && *pos == 0)
324 		++*pos;
325 
326 	return file;
327 }
328 
329 static void *task_file_seq_next(struct seq_file *seq, void *v, loff_t *pos)
330 {
331 	struct bpf_iter_seq_task_file_info *info = seq->private;
332 
333 	++*pos;
334 	++info->fd;
335 	fput((struct file *)v);
336 	return task_file_seq_get_next(info);
337 }
338 
339 struct bpf_iter__task_file {
340 	__bpf_md_ptr(struct bpf_iter_meta *, meta);
341 	__bpf_md_ptr(struct task_struct *, task);
342 	u32 fd __aligned(8);
343 	__bpf_md_ptr(struct file *, file);
344 };
345 
346 DEFINE_BPF_ITER_FUNC(task_file, struct bpf_iter_meta *meta,
347 		     struct task_struct *task, u32 fd,
348 		     struct file *file)
349 
350 static int __task_file_seq_show(struct seq_file *seq, struct file *file,
351 				bool in_stop)
352 {
353 	struct bpf_iter_seq_task_file_info *info = seq->private;
354 	struct bpf_iter__task_file ctx;
355 	struct bpf_iter_meta meta;
356 	struct bpf_prog *prog;
357 
358 	meta.seq = seq;
359 	prog = bpf_iter_get_info(&meta, in_stop);
360 	if (!prog)
361 		return 0;
362 
363 	ctx.meta = &meta;
364 	ctx.task = info->task;
365 	ctx.fd = info->fd;
366 	ctx.file = file;
367 	return bpf_iter_run_prog(prog, &ctx);
368 }
369 
370 static int task_file_seq_show(struct seq_file *seq, void *v)
371 {
372 	return __task_file_seq_show(seq, v, false);
373 }
374 
375 static void task_file_seq_stop(struct seq_file *seq, void *v)
376 {
377 	struct bpf_iter_seq_task_file_info *info = seq->private;
378 
379 	if (!v) {
380 		(void)__task_file_seq_show(seq, v, true);
381 	} else {
382 		fput((struct file *)v);
383 		put_task_struct(info->task);
384 		info->task = NULL;
385 	}
386 }
387 
388 static int init_seq_pidns(void *priv_data, struct bpf_iter_aux_info *aux)
389 {
390 	struct bpf_iter_seq_task_common *common = priv_data;
391 
392 	common->ns = get_pid_ns(task_active_pid_ns(current));
393 	common->type = aux->task.type;
394 	common->pid = aux->task.pid;
395 
396 	return 0;
397 }
398 
399 static void fini_seq_pidns(void *priv_data)
400 {
401 	struct bpf_iter_seq_task_common *common = priv_data;
402 
403 	put_pid_ns(common->ns);
404 }
405 
406 static const struct seq_operations task_file_seq_ops = {
407 	.start	= task_file_seq_start,
408 	.next	= task_file_seq_next,
409 	.stop	= task_file_seq_stop,
410 	.show	= task_file_seq_show,
411 };
412 
413 struct bpf_iter_seq_task_vma_info {
414 	/* The first field must be struct bpf_iter_seq_task_common.
415 	 * this is assumed by {init, fini}_seq_pidns() callback functions.
416 	 */
417 	struct bpf_iter_seq_task_common common;
418 	struct task_struct *task;
419 	struct mm_struct *mm;
420 	struct vm_area_struct *vma;
421 	u32 tid;
422 	unsigned long prev_vm_start;
423 	unsigned long prev_vm_end;
424 };
425 
426 enum bpf_task_vma_iter_find_op {
427 	task_vma_iter_first_vma,   /* use find_vma() with addr 0 */
428 	task_vma_iter_next_vma,    /* use vma_next() with curr_vma */
429 	task_vma_iter_find_vma,    /* use find_vma() to find next vma */
430 };
431 
432 static struct vm_area_struct *
433 task_vma_seq_get_next(struct bpf_iter_seq_task_vma_info *info)
434 {
435 	enum bpf_task_vma_iter_find_op op;
436 	struct vm_area_struct *curr_vma;
437 	struct task_struct *curr_task;
438 	struct mm_struct *curr_mm;
439 	u32 saved_tid = info->tid;
440 
441 	/* If this function returns a non-NULL vma, it holds a reference to
442 	 * the task_struct, holds a refcount on mm->mm_users, and holds
443 	 * read lock on vma->mm->mmap_lock.
444 	 * If this function returns NULL, it does not hold any reference or
445 	 * lock.
446 	 */
447 	if (info->task) {
448 		curr_task = info->task;
449 		curr_vma = info->vma;
450 		curr_mm = info->mm;
451 		/* In case of lock contention, drop mmap_lock to unblock
452 		 * the writer.
453 		 *
454 		 * After relock, call find(mm, prev_vm_end - 1) to find
455 		 * new vma to process.
456 		 *
457 		 *   +------+------+-----------+
458 		 *   | VMA1 | VMA2 | VMA3      |
459 		 *   +------+------+-----------+
460 		 *   |      |      |           |
461 		 *  4k     8k     16k         400k
462 		 *
463 		 * For example, curr_vma == VMA2. Before unlock, we set
464 		 *
465 		 *    prev_vm_start = 8k
466 		 *    prev_vm_end   = 16k
467 		 *
468 		 * There are a few cases:
469 		 *
470 		 * 1) VMA2 is freed, but VMA3 exists.
471 		 *
472 		 *    find_vma() will return VMA3, just process VMA3.
473 		 *
474 		 * 2) VMA2 still exists.
475 		 *
476 		 *    find_vma() will return VMA2, process VMA2->next.
477 		 *
478 		 * 3) no more vma in this mm.
479 		 *
480 		 *    Process the next task.
481 		 *
482 		 * 4) find_vma() returns a different vma, VMA2'.
483 		 *
484 		 *    4.1) If VMA2 covers same range as VMA2', skip VMA2',
485 		 *         because we already covered the range;
486 		 *    4.2) VMA2 and VMA2' covers different ranges, process
487 		 *         VMA2'.
488 		 */
489 		if (mmap_lock_is_contended(curr_mm)) {
490 			info->prev_vm_start = curr_vma->vm_start;
491 			info->prev_vm_end = curr_vma->vm_end;
492 			op = task_vma_iter_find_vma;
493 			mmap_read_unlock(curr_mm);
494 			if (mmap_read_lock_killable(curr_mm)) {
495 				mmput(curr_mm);
496 				goto finish;
497 			}
498 		} else {
499 			op = task_vma_iter_next_vma;
500 		}
501 	} else {
502 again:
503 		curr_task = task_seq_get_next(&info->common, &info->tid, true);
504 		if (!curr_task) {
505 			info->tid++;
506 			goto finish;
507 		}
508 
509 		if (saved_tid != info->tid) {
510 			/* new task, process the first vma */
511 			op = task_vma_iter_first_vma;
512 		} else {
513 			/* Found the same tid, which means the user space
514 			 * finished data in previous buffer and read more.
515 			 * We dropped mmap_lock before returning to user
516 			 * space, so it is necessary to use find_vma() to
517 			 * find the next vma to process.
518 			 */
519 			op = task_vma_iter_find_vma;
520 		}
521 
522 		curr_mm = get_task_mm(curr_task);
523 		if (!curr_mm)
524 			goto next_task;
525 
526 		if (mmap_read_lock_killable(curr_mm)) {
527 			mmput(curr_mm);
528 			goto finish;
529 		}
530 	}
531 
532 	switch (op) {
533 	case task_vma_iter_first_vma:
534 		curr_vma = find_vma(curr_mm, 0);
535 		break;
536 	case task_vma_iter_next_vma:
537 		curr_vma = find_vma(curr_mm, curr_vma->vm_end);
538 		break;
539 	case task_vma_iter_find_vma:
540 		/* We dropped mmap_lock so it is necessary to use find_vma
541 		 * to find the next vma. This is similar to the  mechanism
542 		 * in show_smaps_rollup().
543 		 */
544 		curr_vma = find_vma(curr_mm, info->prev_vm_end - 1);
545 		/* case 1) and 4.2) above just use curr_vma */
546 
547 		/* check for case 2) or case 4.1) above */
548 		if (curr_vma &&
549 		    curr_vma->vm_start == info->prev_vm_start &&
550 		    curr_vma->vm_end == info->prev_vm_end)
551 			curr_vma = find_vma(curr_mm, curr_vma->vm_end);
552 		break;
553 	}
554 	if (!curr_vma) {
555 		/* case 3) above, or case 2) 4.1) with vma->next == NULL */
556 		mmap_read_unlock(curr_mm);
557 		mmput(curr_mm);
558 		goto next_task;
559 	}
560 	info->task = curr_task;
561 	info->vma = curr_vma;
562 	info->mm = curr_mm;
563 	return curr_vma;
564 
565 next_task:
566 	if (info->common.type == BPF_TASK_ITER_TID)
567 		goto finish;
568 
569 	put_task_struct(curr_task);
570 	info->task = NULL;
571 	info->mm = NULL;
572 	info->tid++;
573 	goto again;
574 
575 finish:
576 	if (curr_task)
577 		put_task_struct(curr_task);
578 	info->task = NULL;
579 	info->vma = NULL;
580 	info->mm = NULL;
581 	return NULL;
582 }
583 
584 static void *task_vma_seq_start(struct seq_file *seq, loff_t *pos)
585 {
586 	struct bpf_iter_seq_task_vma_info *info = seq->private;
587 	struct vm_area_struct *vma;
588 
589 	vma = task_vma_seq_get_next(info);
590 	if (vma && *pos == 0)
591 		++*pos;
592 
593 	return vma;
594 }
595 
596 static void *task_vma_seq_next(struct seq_file *seq, void *v, loff_t *pos)
597 {
598 	struct bpf_iter_seq_task_vma_info *info = seq->private;
599 
600 	++*pos;
601 	return task_vma_seq_get_next(info);
602 }
603 
604 struct bpf_iter__task_vma {
605 	__bpf_md_ptr(struct bpf_iter_meta *, meta);
606 	__bpf_md_ptr(struct task_struct *, task);
607 	__bpf_md_ptr(struct vm_area_struct *, vma);
608 };
609 
610 DEFINE_BPF_ITER_FUNC(task_vma, struct bpf_iter_meta *meta,
611 		     struct task_struct *task, struct vm_area_struct *vma)
612 
613 static int __task_vma_seq_show(struct seq_file *seq, bool in_stop)
614 {
615 	struct bpf_iter_seq_task_vma_info *info = seq->private;
616 	struct bpf_iter__task_vma ctx;
617 	struct bpf_iter_meta meta;
618 	struct bpf_prog *prog;
619 
620 	meta.seq = seq;
621 	prog = bpf_iter_get_info(&meta, in_stop);
622 	if (!prog)
623 		return 0;
624 
625 	ctx.meta = &meta;
626 	ctx.task = info->task;
627 	ctx.vma = info->vma;
628 	return bpf_iter_run_prog(prog, &ctx);
629 }
630 
631 static int task_vma_seq_show(struct seq_file *seq, void *v)
632 {
633 	return __task_vma_seq_show(seq, false);
634 }
635 
636 static void task_vma_seq_stop(struct seq_file *seq, void *v)
637 {
638 	struct bpf_iter_seq_task_vma_info *info = seq->private;
639 
640 	if (!v) {
641 		(void)__task_vma_seq_show(seq, true);
642 	} else {
643 		/* info->vma has not been seen by the BPF program. If the
644 		 * user space reads more, task_vma_seq_get_next should
645 		 * return this vma again. Set prev_vm_start to ~0UL,
646 		 * so that we don't skip the vma returned by the next
647 		 * find_vma() (case task_vma_iter_find_vma in
648 		 * task_vma_seq_get_next()).
649 		 */
650 		info->prev_vm_start = ~0UL;
651 		info->prev_vm_end = info->vma->vm_end;
652 		mmap_read_unlock(info->mm);
653 		mmput(info->mm);
654 		info->mm = NULL;
655 		put_task_struct(info->task);
656 		info->task = NULL;
657 	}
658 }
659 
660 static const struct seq_operations task_vma_seq_ops = {
661 	.start	= task_vma_seq_start,
662 	.next	= task_vma_seq_next,
663 	.stop	= task_vma_seq_stop,
664 	.show	= task_vma_seq_show,
665 };
666 
667 static const struct bpf_iter_seq_info task_seq_info = {
668 	.seq_ops		= &task_seq_ops,
669 	.init_seq_private	= init_seq_pidns,
670 	.fini_seq_private	= fini_seq_pidns,
671 	.seq_priv_size		= sizeof(struct bpf_iter_seq_task_info),
672 };
673 
674 static int bpf_iter_fill_link_info(const struct bpf_iter_aux_info *aux, struct bpf_link_info *info)
675 {
676 	switch (aux->task.type) {
677 	case BPF_TASK_ITER_TID:
678 		info->iter.task.tid = aux->task.pid;
679 		break;
680 	case BPF_TASK_ITER_TGID:
681 		info->iter.task.pid = aux->task.pid;
682 		break;
683 	default:
684 		break;
685 	}
686 	return 0;
687 }
688 
689 static void bpf_iter_task_show_fdinfo(const struct bpf_iter_aux_info *aux, struct seq_file *seq)
690 {
691 	seq_printf(seq, "task_type:\t%s\n", iter_task_type_names[aux->task.type]);
692 	if (aux->task.type == BPF_TASK_ITER_TID)
693 		seq_printf(seq, "tid:\t%u\n", aux->task.pid);
694 	else if (aux->task.type == BPF_TASK_ITER_TGID)
695 		seq_printf(seq, "pid:\t%u\n", aux->task.pid);
696 }
697 
698 static struct bpf_iter_reg task_reg_info = {
699 	.target			= "task",
700 	.attach_target		= bpf_iter_attach_task,
701 	.feature		= BPF_ITER_RESCHED,
702 	.ctx_arg_info_size	= 1,
703 	.ctx_arg_info		= {
704 		{ offsetof(struct bpf_iter__task, task),
705 		  PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED },
706 	},
707 	.seq_info		= &task_seq_info,
708 	.fill_link_info		= bpf_iter_fill_link_info,
709 	.show_fdinfo		= bpf_iter_task_show_fdinfo,
710 };
711 
712 static const struct bpf_iter_seq_info task_file_seq_info = {
713 	.seq_ops		= &task_file_seq_ops,
714 	.init_seq_private	= init_seq_pidns,
715 	.fini_seq_private	= fini_seq_pidns,
716 	.seq_priv_size		= sizeof(struct bpf_iter_seq_task_file_info),
717 };
718 
719 static struct bpf_iter_reg task_file_reg_info = {
720 	.target			= "task_file",
721 	.attach_target		= bpf_iter_attach_task,
722 	.feature		= BPF_ITER_RESCHED,
723 	.ctx_arg_info_size	= 2,
724 	.ctx_arg_info		= {
725 		{ offsetof(struct bpf_iter__task_file, task),
726 		  PTR_TO_BTF_ID_OR_NULL },
727 		{ offsetof(struct bpf_iter__task_file, file),
728 		  PTR_TO_BTF_ID_OR_NULL },
729 	},
730 	.seq_info		= &task_file_seq_info,
731 	.fill_link_info		= bpf_iter_fill_link_info,
732 	.show_fdinfo		= bpf_iter_task_show_fdinfo,
733 };
734 
735 static const struct bpf_iter_seq_info task_vma_seq_info = {
736 	.seq_ops		= &task_vma_seq_ops,
737 	.init_seq_private	= init_seq_pidns,
738 	.fini_seq_private	= fini_seq_pidns,
739 	.seq_priv_size		= sizeof(struct bpf_iter_seq_task_vma_info),
740 };
741 
742 static struct bpf_iter_reg task_vma_reg_info = {
743 	.target			= "task_vma",
744 	.attach_target		= bpf_iter_attach_task,
745 	.feature		= BPF_ITER_RESCHED,
746 	.ctx_arg_info_size	= 2,
747 	.ctx_arg_info		= {
748 		{ offsetof(struct bpf_iter__task_vma, task),
749 		  PTR_TO_BTF_ID_OR_NULL },
750 		{ offsetof(struct bpf_iter__task_vma, vma),
751 		  PTR_TO_BTF_ID_OR_NULL },
752 	},
753 	.seq_info		= &task_vma_seq_info,
754 	.fill_link_info		= bpf_iter_fill_link_info,
755 	.show_fdinfo		= bpf_iter_task_show_fdinfo,
756 };
757 
758 BPF_CALL_5(bpf_find_vma, struct task_struct *, task, u64, start,
759 	   bpf_callback_t, callback_fn, void *, callback_ctx, u64, flags)
760 {
761 	struct mmap_unlock_irq_work *work = NULL;
762 	struct vm_area_struct *vma;
763 	bool irq_work_busy = false;
764 	struct mm_struct *mm;
765 	int ret = -ENOENT;
766 
767 	if (flags)
768 		return -EINVAL;
769 
770 	if (!task)
771 		return -ENOENT;
772 
773 	mm = task->mm;
774 	if (!mm)
775 		return -ENOENT;
776 
777 	irq_work_busy = bpf_mmap_unlock_get_irq_work(&work);
778 
779 	if (irq_work_busy || !mmap_read_trylock(mm))
780 		return -EBUSY;
781 
782 	vma = find_vma(mm, start);
783 
784 	if (vma && vma->vm_start <= start && vma->vm_end > start) {
785 		callback_fn((u64)(long)task, (u64)(long)vma,
786 			    (u64)(long)callback_ctx, 0, 0);
787 		ret = 0;
788 	}
789 	bpf_mmap_unlock_mm(work, mm);
790 	return ret;
791 }
792 
793 const struct bpf_func_proto bpf_find_vma_proto = {
794 	.func		= bpf_find_vma,
795 	.ret_type	= RET_INTEGER,
796 	.arg1_type	= ARG_PTR_TO_BTF_ID,
797 	.arg1_btf_id	= &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
798 	.arg2_type	= ARG_ANYTHING,
799 	.arg3_type	= ARG_PTR_TO_FUNC,
800 	.arg4_type	= ARG_PTR_TO_STACK_OR_NULL,
801 	.arg5_type	= ARG_ANYTHING,
802 };
803 
804 struct bpf_iter_task_vma_kern_data {
805 	struct task_struct *task;
806 	struct mm_struct *mm;
807 	struct mmap_unlock_irq_work *work;
808 	struct vma_iterator vmi;
809 };
810 
811 struct bpf_iter_task_vma {
812 	/* opaque iterator state; having __u64 here allows to preserve correct
813 	 * alignment requirements in vmlinux.h, generated from BTF
814 	 */
815 	__u64 __opaque[1];
816 } __attribute__((aligned(8)));
817 
818 /* Non-opaque version of bpf_iter_task_vma */
819 struct bpf_iter_task_vma_kern {
820 	struct bpf_iter_task_vma_kern_data *data;
821 } __attribute__((aligned(8)));
822 
823 __bpf_kfunc_start_defs();
824 
825 __bpf_kfunc int bpf_iter_task_vma_new(struct bpf_iter_task_vma *it,
826 				      struct task_struct *task, u64 addr)
827 {
828 	struct bpf_iter_task_vma_kern *kit = (void *)it;
829 	bool irq_work_busy = false;
830 	int err;
831 
832 	BUILD_BUG_ON(sizeof(struct bpf_iter_task_vma_kern) != sizeof(struct bpf_iter_task_vma));
833 	BUILD_BUG_ON(__alignof__(struct bpf_iter_task_vma_kern) != __alignof__(struct bpf_iter_task_vma));
834 
835 	/* is_iter_reg_valid_uninit guarantees that kit hasn't been initialized
836 	 * before, so non-NULL kit->data doesn't point to previously
837 	 * bpf_mem_alloc'd bpf_iter_task_vma_kern_data
838 	 */
839 	kit->data = bpf_mem_alloc(&bpf_global_ma, sizeof(struct bpf_iter_task_vma_kern_data));
840 	if (!kit->data)
841 		return -ENOMEM;
842 
843 	kit->data->task = get_task_struct(task);
844 	kit->data->mm = task->mm;
845 	if (!kit->data->mm) {
846 		err = -ENOENT;
847 		goto err_cleanup_iter;
848 	}
849 
850 	/* kit->data->work == NULL is valid after bpf_mmap_unlock_get_irq_work */
851 	irq_work_busy = bpf_mmap_unlock_get_irq_work(&kit->data->work);
852 	if (irq_work_busy || !mmap_read_trylock(kit->data->mm)) {
853 		err = -EBUSY;
854 		goto err_cleanup_iter;
855 	}
856 
857 	vma_iter_init(&kit->data->vmi, kit->data->mm, addr);
858 	return 0;
859 
860 err_cleanup_iter:
861 	if (kit->data->task)
862 		put_task_struct(kit->data->task);
863 	bpf_mem_free(&bpf_global_ma, kit->data);
864 	/* NULL kit->data signals failed bpf_iter_task_vma initialization */
865 	kit->data = NULL;
866 	return err;
867 }
868 
869 __bpf_kfunc struct vm_area_struct *bpf_iter_task_vma_next(struct bpf_iter_task_vma *it)
870 {
871 	struct bpf_iter_task_vma_kern *kit = (void *)it;
872 
873 	if (!kit->data) /* bpf_iter_task_vma_new failed */
874 		return NULL;
875 	return vma_next(&kit->data->vmi);
876 }
877 
878 __bpf_kfunc void bpf_iter_task_vma_destroy(struct bpf_iter_task_vma *it)
879 {
880 	struct bpf_iter_task_vma_kern *kit = (void *)it;
881 
882 	if (kit->data) {
883 		bpf_mmap_unlock_mm(kit->data->work, kit->data->mm);
884 		put_task_struct(kit->data->task);
885 		bpf_mem_free(&bpf_global_ma, kit->data);
886 	}
887 }
888 
889 __bpf_kfunc_end_defs();
890 
891 #ifdef CONFIG_CGROUPS
892 
893 struct bpf_iter_css_task {
894 	__u64 __opaque[1];
895 } __attribute__((aligned(8)));
896 
897 struct bpf_iter_css_task_kern {
898 	struct css_task_iter *css_it;
899 } __attribute__((aligned(8)));
900 
901 __bpf_kfunc_start_defs();
902 
903 __bpf_kfunc int bpf_iter_css_task_new(struct bpf_iter_css_task *it,
904 		struct cgroup_subsys_state *css, unsigned int flags)
905 {
906 	struct bpf_iter_css_task_kern *kit = (void *)it;
907 
908 	BUILD_BUG_ON(sizeof(struct bpf_iter_css_task_kern) != sizeof(struct bpf_iter_css_task));
909 	BUILD_BUG_ON(__alignof__(struct bpf_iter_css_task_kern) !=
910 					__alignof__(struct bpf_iter_css_task));
911 	kit->css_it = NULL;
912 	switch (flags) {
913 	case CSS_TASK_ITER_PROCS | CSS_TASK_ITER_THREADED:
914 	case CSS_TASK_ITER_PROCS:
915 	case 0:
916 		break;
917 	default:
918 		return -EINVAL;
919 	}
920 
921 	kit->css_it = bpf_mem_alloc(&bpf_global_ma, sizeof(struct css_task_iter));
922 	if (!kit->css_it)
923 		return -ENOMEM;
924 	css_task_iter_start(css, flags, kit->css_it);
925 	return 0;
926 }
927 
928 __bpf_kfunc struct task_struct *bpf_iter_css_task_next(struct bpf_iter_css_task *it)
929 {
930 	struct bpf_iter_css_task_kern *kit = (void *)it;
931 
932 	if (!kit->css_it)
933 		return NULL;
934 	return css_task_iter_next(kit->css_it);
935 }
936 
937 __bpf_kfunc void bpf_iter_css_task_destroy(struct bpf_iter_css_task *it)
938 {
939 	struct bpf_iter_css_task_kern *kit = (void *)it;
940 
941 	if (!kit->css_it)
942 		return;
943 	css_task_iter_end(kit->css_it);
944 	bpf_mem_free(&bpf_global_ma, kit->css_it);
945 }
946 
947 __bpf_kfunc_end_defs();
948 
949 #endif /* CONFIG_CGROUPS */
950 
951 struct bpf_iter_task {
952 	__u64 __opaque[3];
953 } __attribute__((aligned(8)));
954 
955 struct bpf_iter_task_kern {
956 	struct task_struct *task;
957 	struct task_struct *pos;
958 	unsigned int flags;
959 } __attribute__((aligned(8)));
960 
961 enum {
962 	/* all process in the system */
963 	BPF_TASK_ITER_ALL_PROCS,
964 	/* all threads in the system */
965 	BPF_TASK_ITER_ALL_THREADS,
966 	/* all threads of a specific process */
967 	BPF_TASK_ITER_PROC_THREADS
968 };
969 
970 __bpf_kfunc_start_defs();
971 
972 __bpf_kfunc int bpf_iter_task_new(struct bpf_iter_task *it,
973 		struct task_struct *task__nullable, unsigned int flags)
974 {
975 	struct bpf_iter_task_kern *kit = (void *)it;
976 
977 	BUILD_BUG_ON(sizeof(struct bpf_iter_task_kern) > sizeof(struct bpf_iter_task));
978 	BUILD_BUG_ON(__alignof__(struct bpf_iter_task_kern) !=
979 					__alignof__(struct bpf_iter_task));
980 
981 	kit->pos = NULL;
982 
983 	switch (flags) {
984 	case BPF_TASK_ITER_ALL_THREADS:
985 	case BPF_TASK_ITER_ALL_PROCS:
986 		break;
987 	case BPF_TASK_ITER_PROC_THREADS:
988 		if (!task__nullable)
989 			return -EINVAL;
990 		break;
991 	default:
992 		return -EINVAL;
993 	}
994 
995 	if (flags == BPF_TASK_ITER_PROC_THREADS)
996 		kit->task = task__nullable;
997 	else
998 		kit->task = &init_task;
999 	kit->pos = kit->task;
1000 	kit->flags = flags;
1001 	return 0;
1002 }
1003 
1004 __bpf_kfunc struct task_struct *bpf_iter_task_next(struct bpf_iter_task *it)
1005 {
1006 	struct bpf_iter_task_kern *kit = (void *)it;
1007 	struct task_struct *pos;
1008 	unsigned int flags;
1009 
1010 	flags = kit->flags;
1011 	pos = kit->pos;
1012 
1013 	if (!pos)
1014 		return pos;
1015 
1016 	if (flags == BPF_TASK_ITER_ALL_PROCS)
1017 		goto get_next_task;
1018 
1019 	kit->pos = __next_thread(kit->pos);
1020 	if (kit->pos || flags == BPF_TASK_ITER_PROC_THREADS)
1021 		return pos;
1022 
1023 get_next_task:
1024 	kit->task = next_task(kit->task);
1025 	if (kit->task == &init_task)
1026 		kit->pos = NULL;
1027 	else
1028 		kit->pos = kit->task;
1029 
1030 	return pos;
1031 }
1032 
1033 __bpf_kfunc void bpf_iter_task_destroy(struct bpf_iter_task *it)
1034 {
1035 }
1036 
1037 __bpf_kfunc_end_defs();
1038 
1039 DEFINE_PER_CPU(struct mmap_unlock_irq_work, mmap_unlock_work);
1040 
1041 static void do_mmap_read_unlock(struct irq_work *entry)
1042 {
1043 	struct mmap_unlock_irq_work *work;
1044 
1045 	if (WARN_ON_ONCE(IS_ENABLED(CONFIG_PREEMPT_RT)))
1046 		return;
1047 
1048 	work = container_of(entry, struct mmap_unlock_irq_work, irq_work);
1049 	mmap_read_unlock_non_owner(work->mm);
1050 }
1051 
1052 static int __init task_iter_init(void)
1053 {
1054 	struct mmap_unlock_irq_work *work;
1055 	int ret, cpu;
1056 
1057 	for_each_possible_cpu(cpu) {
1058 		work = per_cpu_ptr(&mmap_unlock_work, cpu);
1059 		init_irq_work(&work->irq_work, do_mmap_read_unlock);
1060 	}
1061 
1062 	task_reg_info.ctx_arg_info[0].btf_id = btf_tracing_ids[BTF_TRACING_TYPE_TASK];
1063 	ret = bpf_iter_reg_target(&task_reg_info);
1064 	if (ret)
1065 		return ret;
1066 
1067 	task_file_reg_info.ctx_arg_info[0].btf_id = btf_tracing_ids[BTF_TRACING_TYPE_TASK];
1068 	task_file_reg_info.ctx_arg_info[1].btf_id = btf_tracing_ids[BTF_TRACING_TYPE_FILE];
1069 	ret =  bpf_iter_reg_target(&task_file_reg_info);
1070 	if (ret)
1071 		return ret;
1072 
1073 	task_vma_reg_info.ctx_arg_info[0].btf_id = btf_tracing_ids[BTF_TRACING_TYPE_TASK];
1074 	task_vma_reg_info.ctx_arg_info[1].btf_id = btf_tracing_ids[BTF_TRACING_TYPE_VMA];
1075 	return bpf_iter_reg_target(&task_vma_reg_info);
1076 }
1077 late_initcall(task_iter_init);
1078