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 /* maximum visited objects before bailing out */
71 #define MAX_ITER_OBJECTS 1000000
73 /* bpf_seq_read, a customized and simpler version for bpf iterator.
74 * no_llseek is assumed for this file.
75 * The following are differences from seq_read():
76 * . fixed buffer size (PAGE_SIZE)
77 * . assuming no_llseek
78 * . stop() may call bpf program, handling potential overflow there
80 static ssize_t bpf_seq_read(struct file *file, char __user *buf, size_t size,
83 struct seq_file *seq = file->private_data;
84 size_t n, offs, copied = 0;
85 int err = 0, num_objs = 0;
88 mutex_lock(&seq->lock);
91 seq->size = PAGE_SIZE;
92 seq->buf = kmalloc(seq->size, GFP_KERNEL);
100 n = min(seq->count, size);
101 err = copy_to_user(buf, seq->buf + seq->from, n);
113 p = seq->op->start(seq, &seq->index);
118 seq->op->stop(seq, p);
123 err = seq->op->show(seq, p);
125 /* object is skipped, decrease seq_num, so next
126 * valid object can reuse the same seq_num.
128 bpf_iter_dec_seq_num(seq);
130 } else if (err < 0 || seq_has_overflowed(seq)) {
133 seq->op->stop(seq, p);
139 loff_t pos = seq->index;
143 p = seq->op->next(seq, p, &seq->index);
144 if (pos == seq->index) {
145 pr_info_ratelimited("buggy seq_file .next function %ps "
146 "did not updated position index\n",
151 if (IS_ERR_OR_NULL(p))
154 /* got a valid next object, increase seq_num */
155 bpf_iter_inc_seq_num(seq);
157 if (seq->count >= size)
160 if (num_objs >= MAX_ITER_OBJECTS) {
163 seq->op->stop(seq, p);
169 err = seq->op->show(seq, p);
171 bpf_iter_dec_seq_num(seq);
173 } else if (err < 0 || seq_has_overflowed(seq)) {
178 seq->op->stop(seq, p);
186 /* bpf program called if !p */
187 seq->op->stop(seq, p);
189 if (!seq_has_overflowed(seq)) {
190 bpf_iter_done_stop(seq);
200 n = min(seq->count, size);
201 err = copy_to_user(buf, seq->buf, n);
214 mutex_unlock(&seq->lock);
218 static const struct bpf_iter_seq_info *
219 __get_seq_info(struct bpf_iter_link *link)
221 const struct bpf_iter_seq_info *seq_info;
224 seq_info = link->aux.map->ops->iter_seq_info;
229 return link->tinfo->reg_info->seq_info;
232 static int iter_open(struct inode *inode, struct file *file)
234 struct bpf_iter_link *link = inode->i_private;
236 return prepare_seq_file(file, link, __get_seq_info(link));
239 static int iter_release(struct inode *inode, struct file *file)
241 struct bpf_iter_priv_data *iter_priv;
242 struct seq_file *seq;
244 seq = file->private_data;
248 iter_priv = container_of(seq->private, struct bpf_iter_priv_data,
251 if (iter_priv->seq_info->fini_seq_private)
252 iter_priv->seq_info->fini_seq_private(seq->private);
254 bpf_prog_put(iter_priv->prog);
255 seq->private = iter_priv;
257 return seq_release_private(inode, file);
260 const struct file_operations bpf_iter_fops = {
263 .read = bpf_seq_read,
264 .release = iter_release,
267 /* The argument reg_info will be cached in bpf_iter_target_info.
268 * The common practice is to declare target reg_info as
269 * a const static variable and passed as an argument to
270 * bpf_iter_reg_target().
272 int bpf_iter_reg_target(const struct bpf_iter_reg *reg_info)
274 struct bpf_iter_target_info *tinfo;
276 tinfo = kmalloc(sizeof(*tinfo), GFP_KERNEL);
280 tinfo->reg_info = reg_info;
281 INIT_LIST_HEAD(&tinfo->list);
283 mutex_lock(&targets_mutex);
284 list_add(&tinfo->list, &targets);
285 mutex_unlock(&targets_mutex);
290 void bpf_iter_unreg_target(const struct bpf_iter_reg *reg_info)
292 struct bpf_iter_target_info *tinfo;
295 mutex_lock(&targets_mutex);
296 list_for_each_entry(tinfo, &targets, list) {
297 if (reg_info == tinfo->reg_info) {
298 list_del(&tinfo->list);
304 mutex_unlock(&targets_mutex);
306 WARN_ON(found == false);
309 static void cache_btf_id(struct bpf_iter_target_info *tinfo,
310 struct bpf_prog *prog)
312 tinfo->btf_id = prog->aux->attach_btf_id;
315 bool bpf_iter_prog_supported(struct bpf_prog *prog)
317 const char *attach_fname = prog->aux->attach_func_name;
318 u32 prog_btf_id = prog->aux->attach_btf_id;
319 const char *prefix = BPF_ITER_FUNC_PREFIX;
320 struct bpf_iter_target_info *tinfo;
321 int prefix_len = strlen(prefix);
322 bool supported = false;
324 if (strncmp(attach_fname, prefix, prefix_len))
327 mutex_lock(&targets_mutex);
328 list_for_each_entry(tinfo, &targets, list) {
329 if (tinfo->btf_id && tinfo->btf_id == prog_btf_id) {
333 if (!strcmp(attach_fname + prefix_len, tinfo->reg_info->target)) {
334 cache_btf_id(tinfo, prog);
339 mutex_unlock(&targets_mutex);
342 prog->aux->ctx_arg_info_size = tinfo->reg_info->ctx_arg_info_size;
343 prog->aux->ctx_arg_info = tinfo->reg_info->ctx_arg_info;
349 static void bpf_iter_link_release(struct bpf_link *link)
351 struct bpf_iter_link *iter_link =
352 container_of(link, struct bpf_iter_link, link);
354 if (iter_link->tinfo->reg_info->detach_target)
355 iter_link->tinfo->reg_info->detach_target(&iter_link->aux);
358 static void bpf_iter_link_dealloc(struct bpf_link *link)
360 struct bpf_iter_link *iter_link =
361 container_of(link, struct bpf_iter_link, link);
366 static int bpf_iter_link_replace(struct bpf_link *link,
367 struct bpf_prog *new_prog,
368 struct bpf_prog *old_prog)
372 mutex_lock(&link_mutex);
373 if (old_prog && link->prog != old_prog) {
378 if (link->prog->type != new_prog->type ||
379 link->prog->expected_attach_type != new_prog->expected_attach_type ||
380 link->prog->aux->attach_btf_id != new_prog->aux->attach_btf_id) {
385 old_prog = xchg(&link->prog, new_prog);
386 bpf_prog_put(old_prog);
389 mutex_unlock(&link_mutex);
393 static const struct bpf_link_ops bpf_iter_link_lops = {
394 .release = bpf_iter_link_release,
395 .dealloc = bpf_iter_link_dealloc,
396 .update_prog = bpf_iter_link_replace,
399 bool bpf_link_is_iter(struct bpf_link *link)
401 return link->ops == &bpf_iter_link_lops;
404 int bpf_iter_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
406 union bpf_iter_link_info __user *ulinfo;
407 struct bpf_link_primer link_primer;
408 struct bpf_iter_target_info *tinfo;
409 union bpf_iter_link_info linfo;
410 struct bpf_iter_link *link;
411 u32 prog_btf_id, linfo_len;
412 bool existed = false;
415 if (attr->link_create.target_fd || attr->link_create.flags)
418 memset(&linfo, 0, sizeof(union bpf_iter_link_info));
420 ulinfo = u64_to_user_ptr(attr->link_create.iter_info);
421 linfo_len = attr->link_create.iter_info_len;
422 if (!ulinfo ^ !linfo_len)
426 err = bpf_check_uarg_tail_zero(ulinfo, sizeof(linfo),
430 linfo_len = min_t(u32, linfo_len, sizeof(linfo));
431 if (copy_from_user(&linfo, ulinfo, linfo_len))
435 prog_btf_id = prog->aux->attach_btf_id;
436 mutex_lock(&targets_mutex);
437 list_for_each_entry(tinfo, &targets, list) {
438 if (tinfo->btf_id == prog_btf_id) {
443 mutex_unlock(&targets_mutex);
447 link = kzalloc(sizeof(*link), GFP_USER | __GFP_NOWARN);
451 bpf_link_init(&link->link, BPF_LINK_TYPE_ITER, &bpf_iter_link_lops, prog);
454 err = bpf_link_prime(&link->link, &link_primer);
460 if (tinfo->reg_info->attach_target) {
461 err = tinfo->reg_info->attach_target(prog, &linfo, &link->aux);
463 bpf_link_cleanup(&link_primer);
468 return bpf_link_settle(&link_primer);
471 static void init_seq_meta(struct bpf_iter_priv_data *priv_data,
472 struct bpf_iter_target_info *tinfo,
473 const struct bpf_iter_seq_info *seq_info,
474 struct bpf_prog *prog)
476 priv_data->tinfo = tinfo;
477 priv_data->seq_info = seq_info;
478 priv_data->prog = prog;
479 priv_data->session_id = atomic64_inc_return(&session_id);
480 priv_data->seq_num = 0;
481 priv_data->done_stop = false;
484 static int prepare_seq_file(struct file *file, struct bpf_iter_link *link,
485 const struct bpf_iter_seq_info *seq_info)
487 struct bpf_iter_priv_data *priv_data;
488 struct bpf_iter_target_info *tinfo;
489 struct bpf_prog *prog;
490 u32 total_priv_dsize;
491 struct seq_file *seq;
494 mutex_lock(&link_mutex);
495 prog = link->link.prog;
497 mutex_unlock(&link_mutex);
500 total_priv_dsize = offsetof(struct bpf_iter_priv_data, target_private) +
501 seq_info->seq_priv_size;
502 priv_data = __seq_open_private(file, seq_info->seq_ops,
509 if (seq_info->init_seq_private) {
510 err = seq_info->init_seq_private(priv_data->target_private, &link->aux);
512 goto release_seq_file;
515 init_seq_meta(priv_data, tinfo, seq_info, prog);
516 seq = file->private_data;
517 seq->private = priv_data->target_private;
522 seq_release_private(file->f_inode, file);
523 file->private_data = NULL;
529 int bpf_iter_new_fd(struct bpf_link *link)
531 struct bpf_iter_link *iter_link;
536 if (link->ops != &bpf_iter_link_lops)
539 flags = O_RDONLY | O_CLOEXEC;
540 fd = get_unused_fd_flags(flags);
544 file = anon_inode_getfile("bpf_iter", &bpf_iter_fops, NULL, flags);
550 iter_link = container_of(link, struct bpf_iter_link, link);
551 err = prepare_seq_file(file, iter_link, __get_seq_info(iter_link));
555 fd_install(fd, file);
565 struct bpf_prog *bpf_iter_get_info(struct bpf_iter_meta *meta, bool in_stop)
567 struct bpf_iter_priv_data *iter_priv;
568 struct seq_file *seq;
572 if (seq->file->f_op != &bpf_iter_fops)
575 seq_priv = seq->private;
576 iter_priv = container_of(seq_priv, struct bpf_iter_priv_data,
579 if (in_stop && iter_priv->done_stop)
582 meta->session_id = iter_priv->session_id;
583 meta->seq_num = iter_priv->seq_num;
585 return iter_priv->prog;
588 int bpf_iter_run_prog(struct bpf_prog *prog, void *ctx)
594 ret = BPF_PROG_RUN(prog, ctx);
598 /* bpf program can only return 0 or 1:
600 * 1 : retry the same object
601 * The bpf_iter_run_prog() return value
602 * will be seq_ops->show() return value.
604 return ret == 0 ? 0 : -EAGAIN;