projects/linux·blob master · refs

375 lines · 10.4 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2/* Copyright (c) 2022 Google */
3#include <linux/bpf.h>
4#include <linux/btf_ids.h>
5#include <linux/cgroup.h>
6#include <linux/kernel.h>
7#include <linux/seq_file.h>
8
9#include "../cgroup/cgroup-internal.h" /* cgroup_mutex and cgroup_is_dead */
10
11/* cgroup_iter provides five modes of traversal to the cgroup hierarchy.
12 *
13 * 1. Walk the descendants of a cgroup in pre-order.
14 * 2. Walk the descendants of a cgroup in post-order.
15 * 3. Walk the ancestors of a cgroup.
16 * 4. Show the given cgroup only.
17 * 5. Walk the children of a given parent cgroup.
18 *
19 * For walking descendants, cgroup_iter can walk in either pre-order or
20 * post-order. For walking ancestors, the iter walks up from a cgroup to
21 * the root.
22 *
23 * The iter program can terminate the walk early by returning 1. Walk
24 * continues if prog returns 0.
25 *
26 * The prog can check (seq->num == 0) to determine whether this is
27 * the first element. The prog may also be passed a NULL cgroup,
28 * which means the walk has completed and the prog has a chance to
29 * do post-processing, such as outputting an epilogue.
30 *
31 * Note: the iter_prog is called with cgroup_mutex held.
32 *
33 * Currently only one session is supported, which means, depending on the
34 * volume of data bpf program intends to send to user space, the number
35 * of cgroups that can be walked is limited. For example, given the current
36 * buffer size is 8 * PAGE_SIZE, if the program sends 64B data for each
37 * cgroup, assuming PAGE_SIZE is 4kb, the total number of cgroups that can
38 * be walked is 512. This is a limitation of cgroup_iter. If the output data
39 * is larger than the kernel buffer size, after all data in the kernel buffer
40 * is consumed by user space, the subsequent read() syscall will signal
41 * EOPNOTSUPP. In order to work around, the user may have to update their
42 * program to reduce the volume of data sent to output. For example, skip
43 * some uninteresting cgroups.
44 */
45
46struct bpf_iter__cgroup {
47 __bpf_md_ptr(struct bpf_iter_meta *, meta);
48 __bpf_md_ptr(struct cgroup *, cgroup);
49};
50
51struct cgroup_iter_priv {
52 struct cgroup_subsys_state *start_css;
53 bool visited_all;
54 bool terminate;
55 int order;
56};
57
58static void *cgroup_iter_seq_start(struct seq_file *seq, loff_t *pos)
59{
60 struct cgroup_iter_priv *p = seq->private;
61
62 cgroup_lock();
63
64 /* cgroup_iter doesn't support read across multiple sessions. */
65 if (*pos > 0) {
66 if (p->visited_all)
67 return NULL;
68
69 /* Haven't visited all, but because cgroup_mutex has dropped,
70 * return -EOPNOTSUPP to indicate incomplete iteration.
71 */
72 return ERR_PTR(-EOPNOTSUPP);
73 }
74
75 ++*pos;
76 p->terminate = false;
77 p->visited_all = false;
78 if (p->order == BPF_CGROUP_ITER_DESCENDANTS_PRE)
79 return css_next_descendant_pre(NULL, p->start_css);
80 else if (p->order == BPF_CGROUP_ITER_DESCENDANTS_POST)
81 return css_next_descendant_post(NULL, p->start_css);
82 else if (p->order == BPF_CGROUP_ITER_CHILDREN)
83 return css_next_child(NULL, p->start_css);
84 else /* BPF_CGROUP_ITER_SELF_ONLY and BPF_CGROUP_ITER_ANCESTORS_UP */
85 return p->start_css;
86}
87
88static int __cgroup_iter_seq_show(struct seq_file *seq,
89 struct cgroup_subsys_state *css, int in_stop);
90
91static void cgroup_iter_seq_stop(struct seq_file *seq, void *v)
92{
93 struct cgroup_iter_priv *p = seq->private;
94
95 cgroup_unlock();
96
97 /* pass NULL to the prog for post-processing */
98 if (!v) {
99 __cgroup_iter_seq_show(seq, NULL, true);
100 p->visited_all = true;
101 }
102}
103
104static void *cgroup_iter_seq_next(struct seq_file *seq, void *v, loff_t *pos)
105{
106 struct cgroup_subsys_state *curr = (struct cgroup_subsys_state *)v;
107 struct cgroup_iter_priv *p = seq->private;
108
109 ++*pos;
110 if (p->terminate)
111 return NULL;
112
113 if (p->order == BPF_CGROUP_ITER_DESCENDANTS_PRE)
114 return css_next_descendant_pre(curr, p->start_css);
115 else if (p->order == BPF_CGROUP_ITER_DESCENDANTS_POST)
116 return css_next_descendant_post(curr, p->start_css);
117 else if (p->order == BPF_CGROUP_ITER_ANCESTORS_UP)
118 return curr->parent;
119 else if (p->order == BPF_CGROUP_ITER_CHILDREN)
120 return css_next_child(curr, p->start_css);
121 else /* BPF_CGROUP_ITER_SELF_ONLY */
122 return NULL;
123}
124
125static int __cgroup_iter_seq_show(struct seq_file *seq,
126 struct cgroup_subsys_state *css, int in_stop)
127{
128 struct cgroup_iter_priv *p = seq->private;
129 struct bpf_iter__cgroup ctx;
130 struct bpf_iter_meta meta;
131 struct bpf_prog *prog;
132 int ret = 0;
133
134 /* cgroup is dead, skip this element */
135 if (css && cgroup_is_dead(css->cgroup))
136 return 0;
137
138 ctx.meta = &meta;
139 ctx.cgroup = css ? css->cgroup : NULL;
140 meta.seq = seq;
141 prog = bpf_iter_get_info(&meta, in_stop);
142 if (prog)
143 ret = bpf_iter_run_prog(prog, &ctx);
144
145 /* if prog returns > 0, terminate after this element. */
146 if (ret != 0)
147 p->terminate = true;
148
149 return 0;
150}
151
152static int cgroup_iter_seq_show(struct seq_file *seq, void *v)
153{
154 return __cgroup_iter_seq_show(seq, (struct cgroup_subsys_state *)v,
155 false);
156}
157
158static const struct seq_operations cgroup_iter_seq_ops = {
159 .start = cgroup_iter_seq_start,
160 .next = cgroup_iter_seq_next,
161 .stop = cgroup_iter_seq_stop,
162 .show = cgroup_iter_seq_show,
163};
164
165BTF_ID_LIST_GLOBAL_SINGLE(bpf_cgroup_btf_id, struct, cgroup)
166
167static int cgroup_iter_seq_init(void *priv, struct bpf_iter_aux_info *aux)
168{
169 struct cgroup_iter_priv *p = (struct cgroup_iter_priv *)priv;
170 struct cgroup *cgrp = aux->cgroup.start;
171
172 /* bpf_iter_attach_cgroup() has already acquired an extra reference
173 * for the start cgroup, but the reference may be released after
174 * cgroup_iter_seq_init(), so acquire another reference for the
175 * start cgroup.
176 */
177 p->start_css = &cgrp->self;
178 css_get(p->start_css);
179 p->terminate = false;
180 p->visited_all = false;
181 p->order = aux->cgroup.order;
182 return 0;
183}
184
185static void cgroup_iter_seq_fini(void *priv)
186{
187 struct cgroup_iter_priv *p = (struct cgroup_iter_priv *)priv;
188
189 css_put(p->start_css);
190}
191
192static const struct bpf_iter_seq_info cgroup_iter_seq_info = {
193 .seq_ops = &cgroup_iter_seq_ops,
194 .init_seq_private = cgroup_iter_seq_init,
195 .fini_seq_private = cgroup_iter_seq_fini,
196 .seq_priv_size = sizeof(struct cgroup_iter_priv),
197};
198
199static int bpf_iter_attach_cgroup(struct bpf_prog *prog,
200 union bpf_iter_link_info *linfo,
201 struct bpf_iter_aux_info *aux)
202{
203 int fd = linfo->cgroup.cgroup_fd;
204 u64 id = linfo->cgroup.cgroup_id;
205 int order = linfo->cgroup.order;
206 struct cgroup *cgrp;
207
208 switch (order) {
209 case BPF_CGROUP_ITER_DESCENDANTS_PRE:
210 case BPF_CGROUP_ITER_DESCENDANTS_POST:
211 case BPF_CGROUP_ITER_ANCESTORS_UP:
212 case BPF_CGROUP_ITER_SELF_ONLY:
213 case BPF_CGROUP_ITER_CHILDREN:
214 break;
215 default:
216 return -EINVAL;
217 }
218
219 if (fd && id)
220 return -EINVAL;
221
222 if (fd)
223 cgrp = cgroup_v1v2_get_from_fd(fd);
224 else if (id)
225 cgrp = cgroup_get_from_id(id);
226 else /* walk the entire hierarchy by default. */
227 cgrp = cgroup_get_from_path("/");
228
229 if (IS_ERR(cgrp))
230 return PTR_ERR(cgrp);
231
232 aux->cgroup.start = cgrp;
233 aux->cgroup.order = order;
234 return 0;
235}
236
237static void bpf_iter_detach_cgroup(struct bpf_iter_aux_info *aux)
238{
239 cgroup_put(aux->cgroup.start);
240}
241
242static void bpf_iter_cgroup_show_fdinfo(const struct bpf_iter_aux_info *aux,
243 struct seq_file *seq)
244{
245 char *buf;
246
247 buf = kzalloc(PATH_MAX, GFP_KERNEL);
248 if (!buf) {
249 seq_puts(seq, "cgroup_path:\t<unknown>\n");
250 goto show_order;
251 }
252
253 /* If cgroup_path_ns() fails, buf will be an empty string, cgroup_path
254 * will print nothing.
255 *
256 * Path is in the calling process's cgroup namespace.
257 */
258 cgroup_path_ns(aux->cgroup.start, buf, PATH_MAX,
259 current->nsproxy->cgroup_ns);
260 seq_printf(seq, "cgroup_path:\t%s\n", buf);
261 kfree(buf);
262
263show_order:
264 if (aux->cgroup.order == BPF_CGROUP_ITER_DESCENDANTS_PRE)
265 seq_puts(seq, "order: descendants_pre\n");
266 else if (aux->cgroup.order == BPF_CGROUP_ITER_DESCENDANTS_POST)
267 seq_puts(seq, "order: descendants_post\n");
268 else if (aux->cgroup.order == BPF_CGROUP_ITER_ANCESTORS_UP)
269 seq_puts(seq, "order: ancestors_up\n");
270 else if (aux->cgroup.order == BPF_CGROUP_ITER_CHILDREN)
271 seq_puts(seq, "order: children\n");
272 else /* BPF_CGROUP_ITER_SELF_ONLY */
273 seq_puts(seq, "order: self_only\n");
274}
275
276static int bpf_iter_cgroup_fill_link_info(const struct bpf_iter_aux_info *aux,
277 struct bpf_link_info *info)
278{
279 info->iter.cgroup.order = aux->cgroup.order;
280 info->iter.cgroup.cgroup_id = cgroup_id(aux->cgroup.start);
281 return 0;
282}
283
284DEFINE_BPF_ITER_FUNC(cgroup, struct bpf_iter_meta *meta,
285 struct cgroup *cgroup)
286
287static struct bpf_iter_reg bpf_cgroup_reg_info = {
288 .target = "cgroup",
289 .feature = BPF_ITER_RESCHED,
290 .attach_target = bpf_iter_attach_cgroup,
291 .detach_target = bpf_iter_detach_cgroup,
292 .show_fdinfo = bpf_iter_cgroup_show_fdinfo,
293 .fill_link_info = bpf_iter_cgroup_fill_link_info,
294 .ctx_arg_info_size = 1,
295 .ctx_arg_info = {
296 { offsetof(struct bpf_iter__cgroup, cgroup),
297 PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED },
298 },
299 .seq_info = &cgroup_iter_seq_info,
300};
301
302static int __init bpf_cgroup_iter_init(void)
303{
304 bpf_cgroup_reg_info.ctx_arg_info[0].btf_id = bpf_cgroup_btf_id[0];
305 return bpf_iter_reg_target(&bpf_cgroup_reg_info);
306}
307
308late_initcall(bpf_cgroup_iter_init);
309
310struct bpf_iter_css {
311 __u64 __opaque[3];
312} __attribute__((aligned(8)));
313
314struct bpf_iter_css_kern {
315 struct cgroup_subsys_state *start;
316 struct cgroup_subsys_state *pos;
317 unsigned int flags;
318} __attribute__((aligned(8)));
319
320__bpf_kfunc_start_defs();
321
322__bpf_kfunc int bpf_iter_css_new(struct bpf_iter_css *it,
323 struct cgroup_subsys_state *start, unsigned int flags)
324{
325 struct bpf_iter_css_kern *kit = (void *)it;
326
327 BUILD_BUG_ON(sizeof(struct bpf_iter_css_kern) > sizeof(struct bpf_iter_css));
328 BUILD_BUG_ON(__alignof__(struct bpf_iter_css_kern) != __alignof__(struct bpf_iter_css));
329
330 kit->start = NULL;
331 switch (flags) {
332 case BPF_CGROUP_ITER_DESCENDANTS_PRE:
333 case BPF_CGROUP_ITER_DESCENDANTS_POST:
334 case BPF_CGROUP_ITER_ANCESTORS_UP:
335 case BPF_CGROUP_ITER_CHILDREN:
336 break;
337 default:
338 return -EINVAL;
339 }
340
341 kit->start = start;
342 kit->pos = NULL;
343 kit->flags = flags;
344 return 0;
345}
346
347__bpf_kfunc struct cgroup_subsys_state *bpf_iter_css_next(struct bpf_iter_css *it)
348{
349 struct bpf_iter_css_kern *kit = (void *)it;
350
351 if (!kit->start)
352 return NULL;
353
354 switch (kit->flags) {
355 case BPF_CGROUP_ITER_DESCENDANTS_PRE:
356 kit->pos = css_next_descendant_pre(kit->pos, kit->start);
357 break;
358 case BPF_CGROUP_ITER_DESCENDANTS_POST:
359 kit->pos = css_next_descendant_post(kit->pos, kit->start);
360 break;
361 case BPF_CGROUP_ITER_CHILDREN:
362 kit->pos = css_next_child(kit->pos, kit->start);
363 break;
364 case BPF_CGROUP_ITER_ANCESTORS_UP:
365 kit->pos = kit->pos ? kit->pos->parent : kit->start;
366 }
367
368 return kit->pos;
369}
370
371__bpf_kfunc void bpf_iter_css_destroy(struct bpf_iter_css *it)
372{
373}
374
375__bpf_kfunc_end_defs();