1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (c) 2020 Facebook */
5 #include <linux/anon_inodes.h>
6 #include <linux/filter.h>
9 struct bpf_iter_target_info {
10 struct list_head list;
11 const struct bpf_iter_reg *reg_info;
12 u32 btf_id; /* cached value */
15 struct bpf_iter_link {
17 struct bpf_iter_aux_info aux;
18 struct bpf_iter_target_info *tinfo;
21 struct bpf_iter_priv_data {
22 struct bpf_iter_target_info *tinfo;
23 const struct bpf_iter_seq_info *seq_info;
24 struct bpf_prog *prog;
28 u8 target_private[] __aligned(8);
31 static struct list_head targets = LIST_HEAD_INIT(targets);
32 static DEFINE_MUTEX(targets_mutex);
34 /* protect bpf_iter_link changes */
35 static DEFINE_MUTEX(link_mutex);
37 /* incremented on every opened seq_file */
38 static atomic64_t session_id;
40 static int prepare_seq_file(struct file *file, struct bpf_iter_link *link,
41 const struct bpf_iter_seq_info *seq_info);
43 static void bpf_iter_inc_seq_num(struct seq_file *seq)
45 struct bpf_iter_priv_data *iter_priv;
47 iter_priv = container_of(seq->private, struct bpf_iter_priv_data,
52 static void bpf_iter_dec_seq_num(struct seq_file *seq)
54 struct bpf_iter_priv_data *iter_priv;
56 iter_priv = container_of(seq->private, struct bpf_iter_priv_data,
61 static void bpf_iter_done_stop(struct seq_file *seq)
63 struct bpf_iter_priv_data *iter_priv;
65 iter_priv = container_of(seq->private, struct bpf_iter_priv_data,
67 iter_priv->done_stop = true;
70 static bool bpf_iter_support_resched(struct seq_file *seq)
72 struct bpf_iter_priv_data *iter_priv;
74 iter_priv = container_of(seq->private, struct bpf_iter_priv_data,
76 return iter_priv->tinfo->reg_info->feature & BPF_ITER_RESCHED;
79 /* maximum visited objects before bailing out */
80 #define MAX_ITER_OBJECTS 1000000
82 /* bpf_seq_read, a customized and simpler version for bpf iterator.
83 * no_llseek is assumed for this file.
84 * The following are differences from seq_read():
85 * . fixed buffer size (PAGE_SIZE)
86 * . assuming no_llseek
87 * . stop() may call bpf program, handling potential overflow there
89 static ssize_t bpf_seq_read(struct file *file, char __user *buf, size_t size,
92 struct seq_file *seq = file->private_data;
93 size_t n, offs, copied = 0;
94 int err = 0, num_objs = 0;
98 mutex_lock(&seq->lock);
101 seq->size = PAGE_SIZE << 3;
102 seq->buf = kvmalloc(seq->size, GFP_KERNEL);
110 n = min(seq->count, size);
111 err = copy_to_user(buf, seq->buf + seq->from, n);
123 p = seq->op->start(seq, &seq->index);
128 seq->op->stop(seq, p);
133 err = seq->op->show(seq, p);
135 /* object is skipped, decrease seq_num, so next
136 * valid object can reuse the same seq_num.
138 bpf_iter_dec_seq_num(seq);
140 } else if (err < 0 || seq_has_overflowed(seq)) {
143 seq->op->stop(seq, p);
148 can_resched = bpf_iter_support_resched(seq);
150 loff_t pos = seq->index;
154 p = seq->op->next(seq, p, &seq->index);
155 if (pos == seq->index) {
156 pr_info_ratelimited("buggy seq_file .next function %ps "
157 "did not updated position index\n",
162 if (IS_ERR_OR_NULL(p))
165 /* got a valid next object, increase seq_num */
166 bpf_iter_inc_seq_num(seq);
168 if (seq->count >= size)
171 if (num_objs >= MAX_ITER_OBJECTS) {
174 seq->op->stop(seq, p);
180 err = seq->op->show(seq, p);
182 bpf_iter_dec_seq_num(seq);
184 } else if (err < 0 || seq_has_overflowed(seq)) {
189 seq->op->stop(seq, p);
200 /* bpf program called if !p */
201 seq->op->stop(seq, p);
203 if (!seq_has_overflowed(seq)) {
204 bpf_iter_done_stop(seq);
214 n = min(seq->count, size);
215 err = copy_to_user(buf, seq->buf, n);
228 mutex_unlock(&seq->lock);
232 static const struct bpf_iter_seq_info *
233 __get_seq_info(struct bpf_iter_link *link)
235 const struct bpf_iter_seq_info *seq_info;
238 seq_info = link->aux.map->ops->iter_seq_info;
243 return link->tinfo->reg_info->seq_info;
246 static int iter_open(struct inode *inode, struct file *file)
248 struct bpf_iter_link *link = inode->i_private;
250 return prepare_seq_file(file, link, __get_seq_info(link));
253 static int iter_release(struct inode *inode, struct file *file)
255 struct bpf_iter_priv_data *iter_priv;
256 struct seq_file *seq;
258 seq = file->private_data;
262 iter_priv = container_of(seq->private, struct bpf_iter_priv_data,
265 if (iter_priv->seq_info->fini_seq_private)
266 iter_priv->seq_info->fini_seq_private(seq->private);
268 bpf_prog_put(iter_priv->prog);
269 seq->private = iter_priv;
271 return seq_release_private(inode, file);
274 const struct file_operations bpf_iter_fops = {
277 .read = bpf_seq_read,
278 .release = iter_release,
281 /* The argument reg_info will be cached in bpf_iter_target_info.
282 * The common practice is to declare target reg_info as
283 * a const static variable and passed as an argument to
284 * bpf_iter_reg_target().
286 int bpf_iter_reg_target(const struct bpf_iter_reg *reg_info)
288 struct bpf_iter_target_info *tinfo;
290 tinfo = kzalloc(sizeof(*tinfo), GFP_KERNEL);
294 tinfo->reg_info = reg_info;
295 INIT_LIST_HEAD(&tinfo->list);
297 mutex_lock(&targets_mutex);
298 list_add(&tinfo->list, &targets);
299 mutex_unlock(&targets_mutex);
304 void bpf_iter_unreg_target(const struct bpf_iter_reg *reg_info)
306 struct bpf_iter_target_info *tinfo;
309 mutex_lock(&targets_mutex);
310 list_for_each_entry(tinfo, &targets, list) {
311 if (reg_info == tinfo->reg_info) {
312 list_del(&tinfo->list);
318 mutex_unlock(&targets_mutex);
320 WARN_ON(found == false);
323 static void cache_btf_id(struct bpf_iter_target_info *tinfo,
324 struct bpf_prog *prog)
326 tinfo->btf_id = prog->aux->attach_btf_id;
329 bool bpf_iter_prog_supported(struct bpf_prog *prog)
331 const char *attach_fname = prog->aux->attach_func_name;
332 u32 prog_btf_id = prog->aux->attach_btf_id;
333 const char *prefix = BPF_ITER_FUNC_PREFIX;
334 struct bpf_iter_target_info *tinfo;
335 int prefix_len = strlen(prefix);
336 bool supported = false;
338 if (strncmp(attach_fname, prefix, prefix_len))
341 mutex_lock(&targets_mutex);
342 list_for_each_entry(tinfo, &targets, list) {
343 if (tinfo->btf_id && tinfo->btf_id == prog_btf_id) {
347 if (!strcmp(attach_fname + prefix_len, tinfo->reg_info->target)) {
348 cache_btf_id(tinfo, prog);
353 mutex_unlock(&targets_mutex);
356 prog->aux->ctx_arg_info_size = tinfo->reg_info->ctx_arg_info_size;
357 prog->aux->ctx_arg_info = tinfo->reg_info->ctx_arg_info;
363 static void bpf_iter_link_release(struct bpf_link *link)
365 struct bpf_iter_link *iter_link =
366 container_of(link, struct bpf_iter_link, link);
368 if (iter_link->tinfo->reg_info->detach_target)
369 iter_link->tinfo->reg_info->detach_target(&iter_link->aux);
372 static void bpf_iter_link_dealloc(struct bpf_link *link)
374 struct bpf_iter_link *iter_link =
375 container_of(link, struct bpf_iter_link, link);
380 static int bpf_iter_link_replace(struct bpf_link *link,
381 struct bpf_prog *new_prog,
382 struct bpf_prog *old_prog)
386 mutex_lock(&link_mutex);
387 if (old_prog && link->prog != old_prog) {
392 if (link->prog->type != new_prog->type ||
393 link->prog->expected_attach_type != new_prog->expected_attach_type ||
394 link->prog->aux->attach_btf_id != new_prog->aux->attach_btf_id) {
399 old_prog = xchg(&link->prog, new_prog);
400 bpf_prog_put(old_prog);
403 mutex_unlock(&link_mutex);
407 static void bpf_iter_link_show_fdinfo(const struct bpf_link *link,
408 struct seq_file *seq)
410 struct bpf_iter_link *iter_link =
411 container_of(link, struct bpf_iter_link, link);
412 bpf_iter_show_fdinfo_t show_fdinfo;
415 "target_name:\t%s\n",
416 iter_link->tinfo->reg_info->target);
418 show_fdinfo = iter_link->tinfo->reg_info->show_fdinfo;
420 show_fdinfo(&iter_link->aux, seq);
423 static int bpf_iter_link_fill_link_info(const struct bpf_link *link,
424 struct bpf_link_info *info)
426 struct bpf_iter_link *iter_link =
427 container_of(link, struct bpf_iter_link, link);
428 char __user *ubuf = u64_to_user_ptr(info->iter.target_name);
429 bpf_iter_fill_link_info_t fill_link_info;
430 u32 ulen = info->iter.target_name_len;
431 const char *target_name;
437 target_name = iter_link->tinfo->reg_info->target;
438 target_len = strlen(target_name);
439 info->iter.target_name_len = target_len + 1;
442 if (ulen >= target_len + 1) {
443 if (copy_to_user(ubuf, target_name, target_len + 1))
448 if (copy_to_user(ubuf, target_name, ulen - 1))
450 if (put_user(zero, ubuf + ulen - 1))
456 fill_link_info = iter_link->tinfo->reg_info->fill_link_info;
458 return fill_link_info(&iter_link->aux, info);
463 static const struct bpf_link_ops bpf_iter_link_lops = {
464 .release = bpf_iter_link_release,
465 .dealloc = bpf_iter_link_dealloc,
466 .update_prog = bpf_iter_link_replace,
467 .show_fdinfo = bpf_iter_link_show_fdinfo,
468 .fill_link_info = bpf_iter_link_fill_link_info,
471 bool bpf_link_is_iter(struct bpf_link *link)
473 return link->ops == &bpf_iter_link_lops;
476 int bpf_iter_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
478 union bpf_iter_link_info __user *ulinfo;
479 struct bpf_link_primer link_primer;
480 struct bpf_iter_target_info *tinfo;
481 union bpf_iter_link_info linfo;
482 struct bpf_iter_link *link;
483 u32 prog_btf_id, linfo_len;
484 bool existed = false;
487 if (attr->link_create.target_fd || attr->link_create.flags)
490 memset(&linfo, 0, sizeof(union bpf_iter_link_info));
492 ulinfo = u64_to_user_ptr(attr->link_create.iter_info);
493 linfo_len = attr->link_create.iter_info_len;
494 if (!ulinfo ^ !linfo_len)
498 err = bpf_check_uarg_tail_zero(ulinfo, sizeof(linfo),
502 linfo_len = min_t(u32, linfo_len, sizeof(linfo));
503 if (copy_from_user(&linfo, ulinfo, linfo_len))
507 prog_btf_id = prog->aux->attach_btf_id;
508 mutex_lock(&targets_mutex);
509 list_for_each_entry(tinfo, &targets, list) {
510 if (tinfo->btf_id == prog_btf_id) {
515 mutex_unlock(&targets_mutex);
519 link = kzalloc(sizeof(*link), GFP_USER | __GFP_NOWARN);
523 bpf_link_init(&link->link, BPF_LINK_TYPE_ITER, &bpf_iter_link_lops, prog);
526 err = bpf_link_prime(&link->link, &link_primer);
532 if (tinfo->reg_info->attach_target) {
533 err = tinfo->reg_info->attach_target(prog, &linfo, &link->aux);
535 bpf_link_cleanup(&link_primer);
540 return bpf_link_settle(&link_primer);
543 static void init_seq_meta(struct bpf_iter_priv_data *priv_data,
544 struct bpf_iter_target_info *tinfo,
545 const struct bpf_iter_seq_info *seq_info,
546 struct bpf_prog *prog)
548 priv_data->tinfo = tinfo;
549 priv_data->seq_info = seq_info;
550 priv_data->prog = prog;
551 priv_data->session_id = atomic64_inc_return(&session_id);
552 priv_data->seq_num = 0;
553 priv_data->done_stop = false;
556 static int prepare_seq_file(struct file *file, struct bpf_iter_link *link,
557 const struct bpf_iter_seq_info *seq_info)
559 struct bpf_iter_priv_data *priv_data;
560 struct bpf_iter_target_info *tinfo;
561 struct bpf_prog *prog;
562 u32 total_priv_dsize;
563 struct seq_file *seq;
566 mutex_lock(&link_mutex);
567 prog = link->link.prog;
569 mutex_unlock(&link_mutex);
572 total_priv_dsize = offsetof(struct bpf_iter_priv_data, target_private) +
573 seq_info->seq_priv_size;
574 priv_data = __seq_open_private(file, seq_info->seq_ops,
581 if (seq_info->init_seq_private) {
582 err = seq_info->init_seq_private(priv_data->target_private, &link->aux);
584 goto release_seq_file;
587 init_seq_meta(priv_data, tinfo, seq_info, prog);
588 seq = file->private_data;
589 seq->private = priv_data->target_private;
594 seq_release_private(file->f_inode, file);
595 file->private_data = NULL;
601 int bpf_iter_new_fd(struct bpf_link *link)
603 struct bpf_iter_link *iter_link;
608 if (link->ops != &bpf_iter_link_lops)
611 flags = O_RDONLY | O_CLOEXEC;
612 fd = get_unused_fd_flags(flags);
616 file = anon_inode_getfile("bpf_iter", &bpf_iter_fops, NULL, flags);
622 iter_link = container_of(link, struct bpf_iter_link, link);
623 err = prepare_seq_file(file, iter_link, __get_seq_info(iter_link));
627 fd_install(fd, file);
637 struct bpf_prog *bpf_iter_get_info(struct bpf_iter_meta *meta, bool in_stop)
639 struct bpf_iter_priv_data *iter_priv;
640 struct seq_file *seq;
644 if (seq->file->f_op != &bpf_iter_fops)
647 seq_priv = seq->private;
648 iter_priv = container_of(seq_priv, struct bpf_iter_priv_data,
651 if (in_stop && iter_priv->done_stop)
654 meta->session_id = iter_priv->session_id;
655 meta->seq_num = iter_priv->seq_num;
657 return iter_priv->prog;
660 int bpf_iter_run_prog(struct bpf_prog *prog, void *ctx)
666 ret = BPF_PROG_RUN(prog, ctx);
670 /* bpf program can only return 0 or 1:
672 * 1 : retry the same object
673 * The bpf_iter_run_prog() return value
674 * will be seq_ops->show() return value.
676 return ret == 0 ? 0 : -EAGAIN;
679 BPF_CALL_4(bpf_for_each_map_elem, struct bpf_map *, map, void *, callback_fn,
680 void *, callback_ctx, u64, flags)
682 return map->ops->map_for_each_callback(map, callback_fn, callback_ctx, flags);
685 const struct bpf_func_proto bpf_for_each_map_elem_proto = {
686 .func = bpf_for_each_map_elem,
688 .ret_type = RET_INTEGER,
689 .arg1_type = ARG_CONST_MAP_PTR,
690 .arg2_type = ARG_PTR_TO_FUNC,
691 .arg3_type = ARG_PTR_TO_STACK_OR_NULL,
692 .arg4_type = ARG_ANYTHING,