xref: /linux/kernel/bpf/task_iter.c (revision 04317b129e4eb5c6f4a58bb899b2019c1545320b)
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/btf_ids.h>
11 #include "mmap_unlock_work.h"
12 
13 static const char * const iter_task_type_names[] = {
14 	"ALL",
15 	"TID",
16 	"PID",
17 };
18 
19 struct bpf_iter_seq_task_common {
20 	struct pid_namespace *ns;
21 	enum bpf_iter_task_type	type;
22 	u32 pid;
23 	u32 pid_visiting;
24 };
25 
26 struct bpf_iter_seq_task_info {
27 	/* The first field must be struct bpf_iter_seq_task_common.
28 	 * this is assumed by {init, fini}_seq_pidns() callback functions.
29 	 */
30 	struct bpf_iter_seq_task_common common;
31 	u32 tid;
32 };
33 
34 static struct task_struct *task_group_seq_get_next(struct bpf_iter_seq_task_common *common,
35 						   u32 *tid,
36 						   bool skip_if_dup_files)
37 {
38 	struct task_struct *task;
39 	struct pid *pid;
40 	u32 next_tid;
41 
42 	if (!*tid) {
43 		/* The first time, the iterator calls this function. */
44 		pid = find_pid_ns(common->pid, common->ns);
45 		task = get_pid_task(pid, PIDTYPE_TGID);
46 		if (!task)
47 			return NULL;
48 
49 		*tid = common->pid;
50 		common->pid_visiting = common->pid;
51 
52 		return task;
53 	}
54 
55 	/* If the control returns to user space and comes back to the
56 	 * kernel again, *tid and common->pid_visiting should be the
57 	 * same for task_seq_start() to pick up the correct task.
58 	 */
59 	if (*tid == common->pid_visiting) {
60 		pid = find_pid_ns(common->pid_visiting, common->ns);
61 		task = get_pid_task(pid, PIDTYPE_PID);
62 
63 		return task;
64 	}
65 
66 	task = find_task_by_pid_ns(common->pid_visiting, common->ns);
67 	if (!task)
68 		return NULL;
69 
70 retry:
71 	task = next_thread(task);
72 
73 	next_tid = __task_pid_nr_ns(task, PIDTYPE_PID, common->ns);
74 	if (!next_tid || next_tid == common->pid) {
75 		/* Run out of tasks of a process.  The tasks of a
76 		 * thread_group are linked as circular linked list.
77 		 */
78 		return NULL;
79 	}
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_fd_rcu(curr_task, &curr_fd);
292 		if (!f)
293 			break;
294 		if (!get_file_rcu(f))
295 			continue;
296 
297 		/* set info->fd */
298 		info->fd = curr_fd;
299 		rcu_read_unlock();
300 		return f;
301 	}
302 
303 	/* the current task is done, go to the next task */
304 	rcu_read_unlock();
305 	put_task_struct(curr_task);
306 
307 	if (info->common.type == BPF_TASK_ITER_TID) {
308 		info->task = NULL;
309 		return NULL;
310 	}
311 
312 	info->task = NULL;
313 	info->fd = 0;
314 	saved_tid = ++(info->tid);
315 	goto again;
316 }
317 
318 static void *task_file_seq_start(struct seq_file *seq, loff_t *pos)
319 {
320 	struct bpf_iter_seq_task_file_info *info = seq->private;
321 	struct file *file;
322 
323 	info->task = NULL;
324 	file = task_file_seq_get_next(info);
325 	if (file && *pos == 0)
326 		++*pos;
327 
328 	return file;
329 }
330 
331 static void *task_file_seq_next(struct seq_file *seq, void *v, loff_t *pos)
332 {
333 	struct bpf_iter_seq_task_file_info *info = seq->private;
334 
335 	++*pos;
336 	++info->fd;
337 	fput((struct file *)v);
338 	return task_file_seq_get_next(info);
339 }
340 
341 struct bpf_iter__task_file {
342 	__bpf_md_ptr(struct bpf_iter_meta *, meta);
343 	__bpf_md_ptr(struct task_struct *, task);
344 	u32 fd __aligned(8);
345 	__bpf_md_ptr(struct file *, file);
346 };
347 
348 DEFINE_BPF_ITER_FUNC(task_file, struct bpf_iter_meta *meta,
349 		     struct task_struct *task, u32 fd,
350 		     struct file *file)
351 
352 static int __task_file_seq_show(struct seq_file *seq, struct file *file,
353 				bool in_stop)
354 {
355 	struct bpf_iter_seq_task_file_info *info = seq->private;
356 	struct bpf_iter__task_file ctx;
357 	struct bpf_iter_meta meta;
358 	struct bpf_prog *prog;
359 
360 	meta.seq = seq;
361 	prog = bpf_iter_get_info(&meta, in_stop);
362 	if (!prog)
363 		return 0;
364 
365 	ctx.meta = &meta;
366 	ctx.task = info->task;
367 	ctx.fd = info->fd;
368 	ctx.file = file;
369 	return bpf_iter_run_prog(prog, &ctx);
370 }
371 
372 static int task_file_seq_show(struct seq_file *seq, void *v)
373 {
374 	return __task_file_seq_show(seq, v, false);
375 }
376 
377 static void task_file_seq_stop(struct seq_file *seq, void *v)
378 {
379 	struct bpf_iter_seq_task_file_info *info = seq->private;
380 
381 	if (!v) {
382 		(void)__task_file_seq_show(seq, v, true);
383 	} else {
384 		fput((struct file *)v);
385 		put_task_struct(info->task);
386 		info->task = NULL;
387 	}
388 }
389 
390 static int init_seq_pidns(void *priv_data, struct bpf_iter_aux_info *aux)
391 {
392 	struct bpf_iter_seq_task_common *common = priv_data;
393 
394 	common->ns = get_pid_ns(task_active_pid_ns(current));
395 	common->type = aux->task.type;
396 	common->pid = aux->task.pid;
397 
398 	return 0;
399 }
400 
401 static void fini_seq_pidns(void *priv_data)
402 {
403 	struct bpf_iter_seq_task_common *common = priv_data;
404 
405 	put_pid_ns(common->ns);
406 }
407 
408 static const struct seq_operations task_file_seq_ops = {
409 	.start	= task_file_seq_start,
410 	.next	= task_file_seq_next,
411 	.stop	= task_file_seq_stop,
412 	.show	= task_file_seq_show,
413 };
414 
415 struct bpf_iter_seq_task_vma_info {
416 	/* The first field must be struct bpf_iter_seq_task_common.
417 	 * this is assumed by {init, fini}_seq_pidns() callback functions.
418 	 */
419 	struct bpf_iter_seq_task_common common;
420 	struct task_struct *task;
421 	struct mm_struct *mm;
422 	struct vm_area_struct *vma;
423 	u32 tid;
424 	unsigned long prev_vm_start;
425 	unsigned long prev_vm_end;
426 };
427 
428 enum bpf_task_vma_iter_find_op {
429 	task_vma_iter_first_vma,   /* use find_vma() with addr 0 */
430 	task_vma_iter_next_vma,    /* use vma_next() with curr_vma */
431 	task_vma_iter_find_vma,    /* use find_vma() to find next vma */
432 };
433 
434 static struct vm_area_struct *
435 task_vma_seq_get_next(struct bpf_iter_seq_task_vma_info *info)
436 {
437 	enum bpf_task_vma_iter_find_op op;
438 	struct vm_area_struct *curr_vma;
439 	struct task_struct *curr_task;
440 	struct mm_struct *curr_mm;
441 	u32 saved_tid = info->tid;
442 
443 	/* If this function returns a non-NULL vma, it holds a reference to
444 	 * the task_struct, holds a refcount on mm->mm_users, and holds
445 	 * read lock on vma->mm->mmap_lock.
446 	 * If this function returns NULL, it does not hold any reference or
447 	 * lock.
448 	 */
449 	if (info->task) {
450 		curr_task = info->task;
451 		curr_vma = info->vma;
452 		curr_mm = info->mm;
453 		/* In case of lock contention, drop mmap_lock to unblock
454 		 * the writer.
455 		 *
456 		 * After relock, call find(mm, prev_vm_end - 1) to find
457 		 * new vma to process.
458 		 *
459 		 *   +------+------+-----------+
460 		 *   | VMA1 | VMA2 | VMA3      |
461 		 *   +------+------+-----------+
462 		 *   |      |      |           |
463 		 *  4k     8k     16k         400k
464 		 *
465 		 * For example, curr_vma == VMA2. Before unlock, we set
466 		 *
467 		 *    prev_vm_start = 8k
468 		 *    prev_vm_end   = 16k
469 		 *
470 		 * There are a few cases:
471 		 *
472 		 * 1) VMA2 is freed, but VMA3 exists.
473 		 *
474 		 *    find_vma() will return VMA3, just process VMA3.
475 		 *
476 		 * 2) VMA2 still exists.
477 		 *
478 		 *    find_vma() will return VMA2, process VMA2->next.
479 		 *
480 		 * 3) no more vma in this mm.
481 		 *
482 		 *    Process the next task.
483 		 *
484 		 * 4) find_vma() returns a different vma, VMA2'.
485 		 *
486 		 *    4.1) If VMA2 covers same range as VMA2', skip VMA2',
487 		 *         because we already covered the range;
488 		 *    4.2) VMA2 and VMA2' covers different ranges, process
489 		 *         VMA2'.
490 		 */
491 		if (mmap_lock_is_contended(curr_mm)) {
492 			info->prev_vm_start = curr_vma->vm_start;
493 			info->prev_vm_end = curr_vma->vm_end;
494 			op = task_vma_iter_find_vma;
495 			mmap_read_unlock(curr_mm);
496 			if (mmap_read_lock_killable(curr_mm)) {
497 				mmput(curr_mm);
498 				goto finish;
499 			}
500 		} else {
501 			op = task_vma_iter_next_vma;
502 		}
503 	} else {
504 again:
505 		curr_task = task_seq_get_next(&info->common, &info->tid, true);
506 		if (!curr_task) {
507 			info->tid++;
508 			goto finish;
509 		}
510 
511 		if (saved_tid != info->tid) {
512 			/* new task, process the first vma */
513 			op = task_vma_iter_first_vma;
514 		} else {
515 			/* Found the same tid, which means the user space
516 			 * finished data in previous buffer and read more.
517 			 * We dropped mmap_lock before returning to user
518 			 * space, so it is necessary to use find_vma() to
519 			 * find the next vma to process.
520 			 */
521 			op = task_vma_iter_find_vma;
522 		}
523 
524 		curr_mm = get_task_mm(curr_task);
525 		if (!curr_mm)
526 			goto next_task;
527 
528 		if (mmap_read_lock_killable(curr_mm)) {
529 			mmput(curr_mm);
530 			goto finish;
531 		}
532 	}
533 
534 	switch (op) {
535 	case task_vma_iter_first_vma:
536 		curr_vma = find_vma(curr_mm, 0);
537 		break;
538 	case task_vma_iter_next_vma:
539 		curr_vma = find_vma(curr_mm, curr_vma->vm_end);
540 		break;
541 	case task_vma_iter_find_vma:
542 		/* We dropped mmap_lock so it is necessary to use find_vma
543 		 * to find the next vma. This is similar to the  mechanism
544 		 * in show_smaps_rollup().
545 		 */
546 		curr_vma = find_vma(curr_mm, info->prev_vm_end - 1);
547 		/* case 1) and 4.2) above just use curr_vma */
548 
549 		/* check for case 2) or case 4.1) above */
550 		if (curr_vma &&
551 		    curr_vma->vm_start == info->prev_vm_start &&
552 		    curr_vma->vm_end == info->prev_vm_end)
553 			curr_vma = find_vma(curr_mm, curr_vma->vm_end);
554 		break;
555 	}
556 	if (!curr_vma) {
557 		/* case 3) above, or case 2) 4.1) with vma->next == NULL */
558 		mmap_read_unlock(curr_mm);
559 		mmput(curr_mm);
560 		goto next_task;
561 	}
562 	info->task = curr_task;
563 	info->vma = curr_vma;
564 	info->mm = curr_mm;
565 	return curr_vma;
566 
567 next_task:
568 	if (info->common.type == BPF_TASK_ITER_TID)
569 		goto finish;
570 
571 	put_task_struct(curr_task);
572 	info->task = NULL;
573 	info->mm = NULL;
574 	info->tid++;
575 	goto again;
576 
577 finish:
578 	if (curr_task)
579 		put_task_struct(curr_task);
580 	info->task = NULL;
581 	info->vma = NULL;
582 	info->mm = NULL;
583 	return NULL;
584 }
585 
586 static void *task_vma_seq_start(struct seq_file *seq, loff_t *pos)
587 {
588 	struct bpf_iter_seq_task_vma_info *info = seq->private;
589 	struct vm_area_struct *vma;
590 
591 	vma = task_vma_seq_get_next(info);
592 	if (vma && *pos == 0)
593 		++*pos;
594 
595 	return vma;
596 }
597 
598 static void *task_vma_seq_next(struct seq_file *seq, void *v, loff_t *pos)
599 {
600 	struct bpf_iter_seq_task_vma_info *info = seq->private;
601 
602 	++*pos;
603 	return task_vma_seq_get_next(info);
604 }
605 
606 struct bpf_iter__task_vma {
607 	__bpf_md_ptr(struct bpf_iter_meta *, meta);
608 	__bpf_md_ptr(struct task_struct *, task);
609 	__bpf_md_ptr(struct vm_area_struct *, vma);
610 };
611 
612 DEFINE_BPF_ITER_FUNC(task_vma, struct bpf_iter_meta *meta,
613 		     struct task_struct *task, struct vm_area_struct *vma)
614 
615 static int __task_vma_seq_show(struct seq_file *seq, bool in_stop)
616 {
617 	struct bpf_iter_seq_task_vma_info *info = seq->private;
618 	struct bpf_iter__task_vma ctx;
619 	struct bpf_iter_meta meta;
620 	struct bpf_prog *prog;
621 
622 	meta.seq = seq;
623 	prog = bpf_iter_get_info(&meta, in_stop);
624 	if (!prog)
625 		return 0;
626 
627 	ctx.meta = &meta;
628 	ctx.task = info->task;
629 	ctx.vma = info->vma;
630 	return bpf_iter_run_prog(prog, &ctx);
631 }
632 
633 static int task_vma_seq_show(struct seq_file *seq, void *v)
634 {
635 	return __task_vma_seq_show(seq, false);
636 }
637 
638 static void task_vma_seq_stop(struct seq_file *seq, void *v)
639 {
640 	struct bpf_iter_seq_task_vma_info *info = seq->private;
641 
642 	if (!v) {
643 		(void)__task_vma_seq_show(seq, true);
644 	} else {
645 		/* info->vma has not been seen by the BPF program. If the
646 		 * user space reads more, task_vma_seq_get_next should
647 		 * return this vma again. Set prev_vm_start to ~0UL,
648 		 * so that we don't skip the vma returned by the next
649 		 * find_vma() (case task_vma_iter_find_vma in
650 		 * task_vma_seq_get_next()).
651 		 */
652 		info->prev_vm_start = ~0UL;
653 		info->prev_vm_end = info->vma->vm_end;
654 		mmap_read_unlock(info->mm);
655 		mmput(info->mm);
656 		info->mm = NULL;
657 		put_task_struct(info->task);
658 		info->task = NULL;
659 	}
660 }
661 
662 static const struct seq_operations task_vma_seq_ops = {
663 	.start	= task_vma_seq_start,
664 	.next	= task_vma_seq_next,
665 	.stop	= task_vma_seq_stop,
666 	.show	= task_vma_seq_show,
667 };
668 
669 static const struct bpf_iter_seq_info task_seq_info = {
670 	.seq_ops		= &task_seq_ops,
671 	.init_seq_private	= init_seq_pidns,
672 	.fini_seq_private	= fini_seq_pidns,
673 	.seq_priv_size		= sizeof(struct bpf_iter_seq_task_info),
674 };
675 
676 static int bpf_iter_fill_link_info(const struct bpf_iter_aux_info *aux, struct bpf_link_info *info)
677 {
678 	switch (aux->task.type) {
679 	case BPF_TASK_ITER_TID:
680 		info->iter.task.tid = aux->task.pid;
681 		break;
682 	case BPF_TASK_ITER_TGID:
683 		info->iter.task.pid = aux->task.pid;
684 		break;
685 	default:
686 		break;
687 	}
688 	return 0;
689 }
690 
691 static void bpf_iter_task_show_fdinfo(const struct bpf_iter_aux_info *aux, struct seq_file *seq)
692 {
693 	seq_printf(seq, "task_type:\t%s\n", iter_task_type_names[aux->task.type]);
694 	if (aux->task.type == BPF_TASK_ITER_TID)
695 		seq_printf(seq, "tid:\t%u\n", aux->task.pid);
696 	else if (aux->task.type == BPF_TASK_ITER_TGID)
697 		seq_printf(seq, "pid:\t%u\n", aux->task.pid);
698 }
699 
700 static struct bpf_iter_reg task_reg_info = {
701 	.target			= "task",
702 	.attach_target		= bpf_iter_attach_task,
703 	.feature		= BPF_ITER_RESCHED,
704 	.ctx_arg_info_size	= 1,
705 	.ctx_arg_info		= {
706 		{ offsetof(struct bpf_iter__task, task),
707 		  PTR_TO_BTF_ID_OR_NULL },
708 	},
709 	.seq_info		= &task_seq_info,
710 	.fill_link_info		= bpf_iter_fill_link_info,
711 	.show_fdinfo		= bpf_iter_task_show_fdinfo,
712 };
713 
714 static const struct bpf_iter_seq_info task_file_seq_info = {
715 	.seq_ops		= &task_file_seq_ops,
716 	.init_seq_private	= init_seq_pidns,
717 	.fini_seq_private	= fini_seq_pidns,
718 	.seq_priv_size		= sizeof(struct bpf_iter_seq_task_file_info),
719 };
720 
721 static struct bpf_iter_reg task_file_reg_info = {
722 	.target			= "task_file",
723 	.attach_target		= bpf_iter_attach_task,
724 	.feature		= BPF_ITER_RESCHED,
725 	.ctx_arg_info_size	= 2,
726 	.ctx_arg_info		= {
727 		{ offsetof(struct bpf_iter__task_file, task),
728 		  PTR_TO_BTF_ID_OR_NULL },
729 		{ offsetof(struct bpf_iter__task_file, file),
730 		  PTR_TO_BTF_ID_OR_NULL },
731 	},
732 	.seq_info		= &task_file_seq_info,
733 	.fill_link_info		= bpf_iter_fill_link_info,
734 	.show_fdinfo		= bpf_iter_task_show_fdinfo,
735 };
736 
737 static const struct bpf_iter_seq_info task_vma_seq_info = {
738 	.seq_ops		= &task_vma_seq_ops,
739 	.init_seq_private	= init_seq_pidns,
740 	.fini_seq_private	= fini_seq_pidns,
741 	.seq_priv_size		= sizeof(struct bpf_iter_seq_task_vma_info),
742 };
743 
744 static struct bpf_iter_reg task_vma_reg_info = {
745 	.target			= "task_vma",
746 	.attach_target		= bpf_iter_attach_task,
747 	.feature		= BPF_ITER_RESCHED,
748 	.ctx_arg_info_size	= 2,
749 	.ctx_arg_info		= {
750 		{ offsetof(struct bpf_iter__task_vma, task),
751 		  PTR_TO_BTF_ID_OR_NULL },
752 		{ offsetof(struct bpf_iter__task_vma, vma),
753 		  PTR_TO_BTF_ID_OR_NULL },
754 	},
755 	.seq_info		= &task_vma_seq_info,
756 	.fill_link_info		= bpf_iter_fill_link_info,
757 	.show_fdinfo		= bpf_iter_task_show_fdinfo,
758 };
759 
760 BPF_CALL_5(bpf_find_vma, struct task_struct *, task, u64, start,
761 	   bpf_callback_t, callback_fn, void *, callback_ctx, u64, flags)
762 {
763 	struct mmap_unlock_irq_work *work = NULL;
764 	struct vm_area_struct *vma;
765 	bool irq_work_busy = false;
766 	struct mm_struct *mm;
767 	int ret = -ENOENT;
768 
769 	if (flags)
770 		return -EINVAL;
771 
772 	if (!task)
773 		return -ENOENT;
774 
775 	mm = task->mm;
776 	if (!mm)
777 		return -ENOENT;
778 
779 	irq_work_busy = bpf_mmap_unlock_get_irq_work(&work);
780 
781 	if (irq_work_busy || !mmap_read_trylock(mm))
782 		return -EBUSY;
783 
784 	vma = find_vma(mm, start);
785 
786 	if (vma && vma->vm_start <= start && vma->vm_end > start) {
787 		callback_fn((u64)(long)task, (u64)(long)vma,
788 			    (u64)(long)callback_ctx, 0, 0);
789 		ret = 0;
790 	}
791 	bpf_mmap_unlock_mm(work, mm);
792 	return ret;
793 }
794 
795 const struct bpf_func_proto bpf_find_vma_proto = {
796 	.func		= bpf_find_vma,
797 	.ret_type	= RET_INTEGER,
798 	.arg1_type	= ARG_PTR_TO_BTF_ID,
799 	.arg1_btf_id	= &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
800 	.arg2_type	= ARG_ANYTHING,
801 	.arg3_type	= ARG_PTR_TO_FUNC,
802 	.arg4_type	= ARG_PTR_TO_STACK_OR_NULL,
803 	.arg5_type	= ARG_ANYTHING,
804 };
805 
806 DEFINE_PER_CPU(struct mmap_unlock_irq_work, mmap_unlock_work);
807 
808 static void do_mmap_read_unlock(struct irq_work *entry)
809 {
810 	struct mmap_unlock_irq_work *work;
811 
812 	if (WARN_ON_ONCE(IS_ENABLED(CONFIG_PREEMPT_RT)))
813 		return;
814 
815 	work = container_of(entry, struct mmap_unlock_irq_work, irq_work);
816 	mmap_read_unlock_non_owner(work->mm);
817 }
818 
819 static int __init task_iter_init(void)
820 {
821 	struct mmap_unlock_irq_work *work;
822 	int ret, cpu;
823 
824 	for_each_possible_cpu(cpu) {
825 		work = per_cpu_ptr(&mmap_unlock_work, cpu);
826 		init_irq_work(&work->irq_work, do_mmap_read_unlock);
827 	}
828 
829 	task_reg_info.ctx_arg_info[0].btf_id = btf_tracing_ids[BTF_TRACING_TYPE_TASK];
830 	ret = bpf_iter_reg_target(&task_reg_info);
831 	if (ret)
832 		return ret;
833 
834 	task_file_reg_info.ctx_arg_info[0].btf_id = btf_tracing_ids[BTF_TRACING_TYPE_TASK];
835 	task_file_reg_info.ctx_arg_info[1].btf_id = btf_tracing_ids[BTF_TRACING_TYPE_FILE];
836 	ret =  bpf_iter_reg_target(&task_file_reg_info);
837 	if (ret)
838 		return ret;
839 
840 	task_vma_reg_info.ctx_arg_info[0].btf_id = btf_tracing_ids[BTF_TRACING_TYPE_TASK];
841 	task_vma_reg_info.ctx_arg_info[1].btf_id = btf_tracing_ids[BTF_TRACING_TYPE_VMA];
842 	return bpf_iter_reg_target(&task_vma_reg_info);
843 }
844 late_initcall(task_iter_init);
845