xref: /linux/kernel/bpf/cgroup_iter.c (revision 90e0d94d369d342e735a75174439482119b6c393)
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 four 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  *
18  * For walking descendants, cgroup_iter can walk in either pre-order or
19  * post-order. For walking ancestors, the iter walks up from a cgroup to
20  * the root.
21  *
22  * The iter program can terminate the walk early by returning 1. Walk
23  * continues if prog returns 0.
24  *
25  * The prog can check (seq->num == 0) to determine whether this is
26  * the first element. The prog may also be passed a NULL cgroup,
27  * which means the walk has completed and the prog has a chance to
28  * do post-processing, such as outputting an epilogue.
29  *
30  * Note: the iter_prog is called with cgroup_mutex held.
31  *
32  * Currently only one session is supported, which means, depending on the
33  * volume of data bpf program intends to send to user space, the number
34  * of cgroups that can be walked is limited. For example, given the current
35  * buffer size is 8 * PAGE_SIZE, if the program sends 64B data for each
36  * cgroup, assuming PAGE_SIZE is 4kb, the total number of cgroups that can
37  * be walked is 512. This is a limitation of cgroup_iter. If the output data
38  * is larger than the kernel buffer size, after all data in the kernel buffer
39  * is consumed by user space, the subsequent read() syscall will signal
40  * EOPNOTSUPP. In order to work around, the user may have to update their
41  * program to reduce the volume of data sent to output. For example, skip
42  * some uninteresting cgroups.
43  */
44 
45 struct bpf_iter__cgroup {
46 	__bpf_md_ptr(struct bpf_iter_meta *, meta);
47 	__bpf_md_ptr(struct cgroup *, cgroup);
48 };
49 
50 struct cgroup_iter_priv {
51 	struct cgroup_subsys_state *start_css;
52 	bool visited_all;
53 	bool terminate;
54 	int order;
55 };
56 
57 static void *cgroup_iter_seq_start(struct seq_file *seq, loff_t *pos)
58 {
59 	struct cgroup_iter_priv *p = seq->private;
60 
61 	mutex_lock(&cgroup_mutex);
62 
63 	/* cgroup_iter doesn't support read across multiple sessions. */
64 	if (*pos > 0) {
65 		if (p->visited_all)
66 			return NULL;
67 
68 		/* Haven't visited all, but because cgroup_mutex has dropped,
69 		 * return -EOPNOTSUPP to indicate incomplete iteration.
70 		 */
71 		return ERR_PTR(-EOPNOTSUPP);
72 	}
73 
74 	++*pos;
75 	p->terminate = false;
76 	p->visited_all = false;
77 	if (p->order == BPF_CGROUP_ITER_DESCENDANTS_PRE)
78 		return css_next_descendant_pre(NULL, p->start_css);
79 	else if (p->order == BPF_CGROUP_ITER_DESCENDANTS_POST)
80 		return css_next_descendant_post(NULL, p->start_css);
81 	else /* BPF_CGROUP_ITER_SELF_ONLY and BPF_CGROUP_ITER_ANCESTORS_UP */
82 		return p->start_css;
83 }
84 
85 static int __cgroup_iter_seq_show(struct seq_file *seq,
86 				  struct cgroup_subsys_state *css, int in_stop);
87 
88 static void cgroup_iter_seq_stop(struct seq_file *seq, void *v)
89 {
90 	struct cgroup_iter_priv *p = seq->private;
91 
92 	mutex_unlock(&cgroup_mutex);
93 
94 	/* pass NULL to the prog for post-processing */
95 	if (!v) {
96 		__cgroup_iter_seq_show(seq, NULL, true);
97 		p->visited_all = true;
98 	}
99 }
100 
101 static void *cgroup_iter_seq_next(struct seq_file *seq, void *v, loff_t *pos)
102 {
103 	struct cgroup_subsys_state *curr = (struct cgroup_subsys_state *)v;
104 	struct cgroup_iter_priv *p = seq->private;
105 
106 	++*pos;
107 	if (p->terminate)
108 		return NULL;
109 
110 	if (p->order == BPF_CGROUP_ITER_DESCENDANTS_PRE)
111 		return css_next_descendant_pre(curr, p->start_css);
112 	else if (p->order == BPF_CGROUP_ITER_DESCENDANTS_POST)
113 		return css_next_descendant_post(curr, p->start_css);
114 	else if (p->order == BPF_CGROUP_ITER_ANCESTORS_UP)
115 		return curr->parent;
116 	else  /* BPF_CGROUP_ITER_SELF_ONLY */
117 		return NULL;
118 }
119 
120 static int __cgroup_iter_seq_show(struct seq_file *seq,
121 				  struct cgroup_subsys_state *css, int in_stop)
122 {
123 	struct cgroup_iter_priv *p = seq->private;
124 	struct bpf_iter__cgroup ctx;
125 	struct bpf_iter_meta meta;
126 	struct bpf_prog *prog;
127 	int ret = 0;
128 
129 	/* cgroup is dead, skip this element */
130 	if (css && cgroup_is_dead(css->cgroup))
131 		return 0;
132 
133 	ctx.meta = &meta;
134 	ctx.cgroup = css ? css->cgroup : NULL;
135 	meta.seq = seq;
136 	prog = bpf_iter_get_info(&meta, in_stop);
137 	if (prog)
138 		ret = bpf_iter_run_prog(prog, &ctx);
139 
140 	/* if prog returns > 0, terminate after this element. */
141 	if (ret != 0)
142 		p->terminate = true;
143 
144 	return 0;
145 }
146 
147 static int cgroup_iter_seq_show(struct seq_file *seq, void *v)
148 {
149 	return __cgroup_iter_seq_show(seq, (struct cgroup_subsys_state *)v,
150 				      false);
151 }
152 
153 static const struct seq_operations cgroup_iter_seq_ops = {
154 	.start  = cgroup_iter_seq_start,
155 	.next   = cgroup_iter_seq_next,
156 	.stop   = cgroup_iter_seq_stop,
157 	.show   = cgroup_iter_seq_show,
158 };
159 
160 BTF_ID_LIST_SINGLE(bpf_cgroup_btf_id, struct, cgroup)
161 
162 static int cgroup_iter_seq_init(void *priv, struct bpf_iter_aux_info *aux)
163 {
164 	struct cgroup_iter_priv *p = (struct cgroup_iter_priv *)priv;
165 	struct cgroup *cgrp = aux->cgroup.start;
166 
167 	p->start_css = &cgrp->self;
168 	p->terminate = false;
169 	p->visited_all = false;
170 	p->order = aux->cgroup.order;
171 	return 0;
172 }
173 
174 static const struct bpf_iter_seq_info cgroup_iter_seq_info = {
175 	.seq_ops		= &cgroup_iter_seq_ops,
176 	.init_seq_private	= cgroup_iter_seq_init,
177 	.seq_priv_size		= sizeof(struct cgroup_iter_priv),
178 };
179 
180 static int bpf_iter_attach_cgroup(struct bpf_prog *prog,
181 				  union bpf_iter_link_info *linfo,
182 				  struct bpf_iter_aux_info *aux)
183 {
184 	int fd = linfo->cgroup.cgroup_fd;
185 	u64 id = linfo->cgroup.cgroup_id;
186 	int order = linfo->cgroup.order;
187 	struct cgroup *cgrp;
188 
189 	if (order != BPF_CGROUP_ITER_DESCENDANTS_PRE &&
190 	    order != BPF_CGROUP_ITER_DESCENDANTS_POST &&
191 	    order != BPF_CGROUP_ITER_ANCESTORS_UP &&
192 	    order != BPF_CGROUP_ITER_SELF_ONLY)
193 		return -EINVAL;
194 
195 	if (fd && id)
196 		return -EINVAL;
197 
198 	if (fd)
199 		cgrp = cgroup_get_from_fd(fd);
200 	else if (id)
201 		cgrp = cgroup_get_from_id(id);
202 	else /* walk the entire hierarchy by default. */
203 		cgrp = cgroup_get_from_path("/");
204 
205 	if (IS_ERR(cgrp))
206 		return PTR_ERR(cgrp);
207 
208 	aux->cgroup.start = cgrp;
209 	aux->cgroup.order = order;
210 	return 0;
211 }
212 
213 static void bpf_iter_detach_cgroup(struct bpf_iter_aux_info *aux)
214 {
215 	cgroup_put(aux->cgroup.start);
216 }
217 
218 static void bpf_iter_cgroup_show_fdinfo(const struct bpf_iter_aux_info *aux,
219 					struct seq_file *seq)
220 {
221 	char *buf;
222 
223 	buf = kzalloc(PATH_MAX, GFP_KERNEL);
224 	if (!buf) {
225 		seq_puts(seq, "cgroup_path:\t<unknown>\n");
226 		goto show_order;
227 	}
228 
229 	/* If cgroup_path_ns() fails, buf will be an empty string, cgroup_path
230 	 * will print nothing.
231 	 *
232 	 * Path is in the calling process's cgroup namespace.
233 	 */
234 	cgroup_path_ns(aux->cgroup.start, buf, PATH_MAX,
235 		       current->nsproxy->cgroup_ns);
236 	seq_printf(seq, "cgroup_path:\t%s\n", buf);
237 	kfree(buf);
238 
239 show_order:
240 	if (aux->cgroup.order == BPF_CGROUP_ITER_DESCENDANTS_PRE)
241 		seq_puts(seq, "order: descendants_pre\n");
242 	else if (aux->cgroup.order == BPF_CGROUP_ITER_DESCENDANTS_POST)
243 		seq_puts(seq, "order: descendants_post\n");
244 	else if (aux->cgroup.order == BPF_CGROUP_ITER_ANCESTORS_UP)
245 		seq_puts(seq, "order: ancestors_up\n");
246 	else /* BPF_CGROUP_ITER_SELF_ONLY */
247 		seq_puts(seq, "order: self_only\n");
248 }
249 
250 static int bpf_iter_cgroup_fill_link_info(const struct bpf_iter_aux_info *aux,
251 					  struct bpf_link_info *info)
252 {
253 	info->iter.cgroup.order = aux->cgroup.order;
254 	info->iter.cgroup.cgroup_id = cgroup_id(aux->cgroup.start);
255 	return 0;
256 }
257 
258 DEFINE_BPF_ITER_FUNC(cgroup, struct bpf_iter_meta *meta,
259 		     struct cgroup *cgroup)
260 
261 static struct bpf_iter_reg bpf_cgroup_reg_info = {
262 	.target			= "cgroup",
263 	.feature		= BPF_ITER_RESCHED,
264 	.attach_target		= bpf_iter_attach_cgroup,
265 	.detach_target		= bpf_iter_detach_cgroup,
266 	.show_fdinfo		= bpf_iter_cgroup_show_fdinfo,
267 	.fill_link_info		= bpf_iter_cgroup_fill_link_info,
268 	.ctx_arg_info_size	= 1,
269 	.ctx_arg_info		= {
270 		{ offsetof(struct bpf_iter__cgroup, cgroup),
271 		  PTR_TO_BTF_ID_OR_NULL },
272 	},
273 	.seq_info		= &cgroup_iter_seq_info,
274 };
275 
276 static int __init bpf_cgroup_iter_init(void)
277 {
278 	bpf_cgroup_reg_info.ctx_arg_info[0].btf_id = bpf_cgroup_btf_id[0];
279 	return bpf_iter_reg_target(&bpf_cgroup_reg_info);
280 }
281 
282 late_initcall(bpf_cgroup_iter_init);
283