xref: /linux/kernel/bpf/mprog.c (revision 7a92fc8b4d20680e4c20289a670d8fca2d1f2c1b)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2023 Isovalent */
3 
4 #include <linux/bpf.h>
5 #include <linux/bpf_mprog.h>
6 
7 static int bpf_mprog_link(struct bpf_tuple *tuple,
8 			  u32 id_or_fd, u32 flags,
9 			  enum bpf_prog_type type)
10 {
11 	struct bpf_link *link = ERR_PTR(-EINVAL);
12 	bool id = flags & BPF_F_ID;
13 
14 	if (id)
15 		link = bpf_link_by_id(id_or_fd);
16 	else if (id_or_fd)
17 		link = bpf_link_get_from_fd(id_or_fd);
18 	if (IS_ERR(link))
19 		return PTR_ERR(link);
20 	if (type && link->prog->type != type) {
21 		bpf_link_put(link);
22 		return -EINVAL;
23 	}
24 
25 	tuple->link = link;
26 	tuple->prog = link->prog;
27 	return 0;
28 }
29 
30 static int bpf_mprog_prog(struct bpf_tuple *tuple,
31 			  u32 id_or_fd, u32 flags,
32 			  enum bpf_prog_type type)
33 {
34 	struct bpf_prog *prog = ERR_PTR(-EINVAL);
35 	bool id = flags & BPF_F_ID;
36 
37 	if (id)
38 		prog = bpf_prog_by_id(id_or_fd);
39 	else if (id_or_fd)
40 		prog = bpf_prog_get(id_or_fd);
41 	if (IS_ERR(prog))
42 		return PTR_ERR(prog);
43 	if (type && prog->type != type) {
44 		bpf_prog_put(prog);
45 		return -EINVAL;
46 	}
47 
48 	tuple->link = NULL;
49 	tuple->prog = prog;
50 	return 0;
51 }
52 
53 static int bpf_mprog_tuple_relative(struct bpf_tuple *tuple,
54 				    u32 id_or_fd, u32 flags,
55 				    enum bpf_prog_type type)
56 {
57 	bool link = flags & BPF_F_LINK;
58 	bool id = flags & BPF_F_ID;
59 
60 	memset(tuple, 0, sizeof(*tuple));
61 	if (link)
62 		return bpf_mprog_link(tuple, id_or_fd, flags, type);
63 	/* If no relevant flag is set and no id_or_fd was passed, then
64 	 * tuple link/prog is just NULLed. This is the case when before/
65 	 * after selects first/last position without passing fd.
66 	 */
67 	if (!id && !id_or_fd)
68 		return 0;
69 	return bpf_mprog_prog(tuple, id_or_fd, flags, type);
70 }
71 
72 static void bpf_mprog_tuple_put(struct bpf_tuple *tuple)
73 {
74 	if (tuple->link)
75 		bpf_link_put(tuple->link);
76 	else if (tuple->prog)
77 		bpf_prog_put(tuple->prog);
78 }
79 
80 /* The bpf_mprog_{replace,delete}() operate on exact idx position with the
81  * one exception that for deletion we support delete from front/back. In
82  * case of front idx is -1, in case of back idx is bpf_mprog_total(entry).
83  * Adjustment to first and last entry is trivial. The bpf_mprog_insert()
84  * we have to deal with the following cases:
85  *
86  * idx + before:
87  *
88  * Insert P4 before P3: idx for old array is 1, idx for new array is 2,
89  * hence we adjust target idx for the new array, so that memmove copies
90  * P1 and P2 to the new entry, and we insert P4 into idx 2. Inserting
91  * before P1 would have old idx -1 and new idx 0.
92  *
93  * +--+--+--+     +--+--+--+--+     +--+--+--+--+
94  * |P1|P2|P3| ==> |P1|P2|  |P3| ==> |P1|P2|P4|P3|
95  * +--+--+--+     +--+--+--+--+     +--+--+--+--+
96  *
97  * idx + after:
98  *
99  * Insert P4 after P2: idx for old array is 2, idx for new array is 2.
100  * Again, memmove copies P1 and P2 to the new entry, and we insert P4
101  * into idx 2. Inserting after P3 would have both old/new idx at 4 aka
102  * bpf_mprog_total(entry).
103  *
104  * +--+--+--+     +--+--+--+--+     +--+--+--+--+
105  * |P1|P2|P3| ==> |P1|P2|  |P3| ==> |P1|P2|P4|P3|
106  * +--+--+--+     +--+--+--+--+     +--+--+--+--+
107  */
108 static int bpf_mprog_replace(struct bpf_mprog_entry *entry,
109 			     struct bpf_mprog_entry **entry_new,
110 			     struct bpf_tuple *ntuple, int idx)
111 {
112 	struct bpf_mprog_fp *fp;
113 	struct bpf_mprog_cp *cp;
114 	struct bpf_prog *oprog;
115 
116 	bpf_mprog_read(entry, idx, &fp, &cp);
117 	oprog = READ_ONCE(fp->prog);
118 	bpf_mprog_write(fp, cp, ntuple);
119 	if (!ntuple->link) {
120 		WARN_ON_ONCE(cp->link);
121 		bpf_prog_put(oprog);
122 	}
123 	*entry_new = entry;
124 	return 0;
125 }
126 
127 static int bpf_mprog_insert(struct bpf_mprog_entry *entry,
128 			    struct bpf_mprog_entry **entry_new,
129 			    struct bpf_tuple *ntuple, int idx, u32 flags)
130 {
131 	int total = bpf_mprog_total(entry);
132 	struct bpf_mprog_entry *peer;
133 	struct bpf_mprog_fp *fp;
134 	struct bpf_mprog_cp *cp;
135 
136 	peer = bpf_mprog_peer(entry);
137 	bpf_mprog_entry_copy(peer, entry);
138 	if (idx == total)
139 		goto insert;
140 	else if (flags & BPF_F_BEFORE)
141 		idx += 1;
142 	bpf_mprog_entry_grow(peer, idx);
143 insert:
144 	bpf_mprog_read(peer, idx, &fp, &cp);
145 	bpf_mprog_write(fp, cp, ntuple);
146 	bpf_mprog_inc(peer);
147 	*entry_new = peer;
148 	return 0;
149 }
150 
151 static int bpf_mprog_delete(struct bpf_mprog_entry *entry,
152 			    struct bpf_mprog_entry **entry_new,
153 			    struct bpf_tuple *dtuple, int idx)
154 {
155 	int total = bpf_mprog_total(entry);
156 	struct bpf_mprog_entry *peer;
157 
158 	peer = bpf_mprog_peer(entry);
159 	bpf_mprog_entry_copy(peer, entry);
160 	if (idx == -1)
161 		idx = 0;
162 	else if (idx == total)
163 		idx = total - 1;
164 	bpf_mprog_entry_shrink(peer, idx);
165 	bpf_mprog_dec(peer);
166 	bpf_mprog_mark_for_release(peer, dtuple);
167 	*entry_new = peer;
168 	return 0;
169 }
170 
171 /* In bpf_mprog_pos_*() we evaluate the target position for the BPF
172  * program/link that needs to be replaced, inserted or deleted for
173  * each "rule" independently. If all rules agree on that position
174  * or existing element, then enact replacement, addition or deletion.
175  * If this is not the case, then the request cannot be satisfied and
176  * we bail out with an error.
177  */
178 static int bpf_mprog_pos_exact(struct bpf_mprog_entry *entry,
179 			       struct bpf_tuple *tuple)
180 {
181 	struct bpf_mprog_fp *fp;
182 	struct bpf_mprog_cp *cp;
183 	int i;
184 
185 	for (i = 0; i < bpf_mprog_total(entry); i++) {
186 		bpf_mprog_read(entry, i, &fp, &cp);
187 		if (tuple->prog == READ_ONCE(fp->prog))
188 			return tuple->link == cp->link ? i : -EBUSY;
189 	}
190 	return -ENOENT;
191 }
192 
193 static int bpf_mprog_pos_before(struct bpf_mprog_entry *entry,
194 				struct bpf_tuple *tuple)
195 {
196 	struct bpf_mprog_fp *fp;
197 	struct bpf_mprog_cp *cp;
198 	int i;
199 
200 	for (i = 0; i < bpf_mprog_total(entry); i++) {
201 		bpf_mprog_read(entry, i, &fp, &cp);
202 		if (tuple->prog == READ_ONCE(fp->prog) &&
203 		    (!tuple->link || tuple->link == cp->link))
204 			return i - 1;
205 	}
206 	return tuple->prog ? -ENOENT : -1;
207 }
208 
209 static int bpf_mprog_pos_after(struct bpf_mprog_entry *entry,
210 			       struct bpf_tuple *tuple)
211 {
212 	struct bpf_mprog_fp *fp;
213 	struct bpf_mprog_cp *cp;
214 	int i;
215 
216 	for (i = 0; i < bpf_mprog_total(entry); i++) {
217 		bpf_mprog_read(entry, i, &fp, &cp);
218 		if (tuple->prog == READ_ONCE(fp->prog) &&
219 		    (!tuple->link || tuple->link == cp->link))
220 			return i + 1;
221 	}
222 	return tuple->prog ? -ENOENT : bpf_mprog_total(entry);
223 }
224 
225 int bpf_mprog_attach(struct bpf_mprog_entry *entry,
226 		     struct bpf_mprog_entry **entry_new,
227 		     struct bpf_prog *prog_new, struct bpf_link *link,
228 		     struct bpf_prog *prog_old,
229 		     u32 flags, u32 id_or_fd, u64 revision)
230 {
231 	struct bpf_tuple rtuple, ntuple = {
232 		.prog = prog_new,
233 		.link = link,
234 	}, otuple = {
235 		.prog = prog_old,
236 		.link = link,
237 	};
238 	int ret, idx = -ERANGE, tidx;
239 
240 	if (revision && revision != bpf_mprog_revision(entry))
241 		return -ESTALE;
242 	if (bpf_mprog_exists(entry, prog_new))
243 		return -EEXIST;
244 	ret = bpf_mprog_tuple_relative(&rtuple, id_or_fd,
245 				       flags & ~BPF_F_REPLACE,
246 				       prog_new->type);
247 	if (ret)
248 		return ret;
249 	if (flags & BPF_F_REPLACE) {
250 		tidx = bpf_mprog_pos_exact(entry, &otuple);
251 		if (tidx < 0) {
252 			ret = tidx;
253 			goto out;
254 		}
255 		idx = tidx;
256 	} else if (bpf_mprog_total(entry) == bpf_mprog_max()) {
257 		ret = -ERANGE;
258 		goto out;
259 	}
260 	if (flags & BPF_F_BEFORE) {
261 		tidx = bpf_mprog_pos_before(entry, &rtuple);
262 		if (tidx < -1 || (idx >= -1 && tidx != idx)) {
263 			ret = tidx < -1 ? tidx : -ERANGE;
264 			goto out;
265 		}
266 		idx = tidx;
267 	}
268 	if (flags & BPF_F_AFTER) {
269 		tidx = bpf_mprog_pos_after(entry, &rtuple);
270 		if (tidx < -1 || (idx >= -1 && tidx != idx)) {
271 			ret = tidx < 0 ? tidx : -ERANGE;
272 			goto out;
273 		}
274 		idx = tidx;
275 	}
276 	if (idx < -1) {
277 		if (rtuple.prog || flags) {
278 			ret = -EINVAL;
279 			goto out;
280 		}
281 		idx = bpf_mprog_total(entry);
282 		flags = BPF_F_AFTER;
283 	}
284 	if (idx >= bpf_mprog_max()) {
285 		ret = -ERANGE;
286 		goto out;
287 	}
288 	if (flags & BPF_F_REPLACE)
289 		ret = bpf_mprog_replace(entry, entry_new, &ntuple, idx);
290 	else
291 		ret = bpf_mprog_insert(entry, entry_new, &ntuple, idx, flags);
292 out:
293 	bpf_mprog_tuple_put(&rtuple);
294 	return ret;
295 }
296 
297 static int bpf_mprog_fetch(struct bpf_mprog_entry *entry,
298 			   struct bpf_tuple *tuple, int idx)
299 {
300 	int total = bpf_mprog_total(entry);
301 	struct bpf_mprog_cp *cp;
302 	struct bpf_mprog_fp *fp;
303 	struct bpf_prog *prog;
304 	struct bpf_link *link;
305 
306 	if (idx == -1)
307 		idx = 0;
308 	else if (idx == total)
309 		idx = total - 1;
310 	bpf_mprog_read(entry, idx, &fp, &cp);
311 	prog = READ_ONCE(fp->prog);
312 	link = cp->link;
313 	/* The deletion request can either be without filled tuple in which
314 	 * case it gets populated here based on idx, or with filled tuple
315 	 * where the only thing we end up doing is the WARN_ON_ONCE() assert.
316 	 * If we hit a BPF link at the given index, it must not be removed
317 	 * from opts path.
318 	 */
319 	if (link && !tuple->link)
320 		return -EBUSY;
321 	WARN_ON_ONCE(tuple->prog && tuple->prog != prog);
322 	WARN_ON_ONCE(tuple->link && tuple->link != link);
323 	tuple->prog = prog;
324 	tuple->link = link;
325 	return 0;
326 }
327 
328 int bpf_mprog_detach(struct bpf_mprog_entry *entry,
329 		     struct bpf_mprog_entry **entry_new,
330 		     struct bpf_prog *prog, struct bpf_link *link,
331 		     u32 flags, u32 id_or_fd, u64 revision)
332 {
333 	struct bpf_tuple rtuple, dtuple = {
334 		.prog = prog,
335 		.link = link,
336 	};
337 	int ret, idx = -ERANGE, tidx;
338 
339 	if (flags & BPF_F_REPLACE)
340 		return -EINVAL;
341 	if (revision && revision != bpf_mprog_revision(entry))
342 		return -ESTALE;
343 	if (!bpf_mprog_total(entry))
344 		return -ENOENT;
345 	ret = bpf_mprog_tuple_relative(&rtuple, id_or_fd, flags,
346 				       prog ? prog->type :
347 				       BPF_PROG_TYPE_UNSPEC);
348 	if (ret)
349 		return ret;
350 	if (dtuple.prog) {
351 		tidx = bpf_mprog_pos_exact(entry, &dtuple);
352 		if (tidx < 0) {
353 			ret = tidx;
354 			goto out;
355 		}
356 		idx = tidx;
357 	}
358 	if (flags & BPF_F_BEFORE) {
359 		tidx = bpf_mprog_pos_before(entry, &rtuple);
360 		if (tidx < -1 || (idx >= -1 && tidx != idx)) {
361 			ret = tidx < -1 ? tidx : -ERANGE;
362 			goto out;
363 		}
364 		idx = tidx;
365 	}
366 	if (flags & BPF_F_AFTER) {
367 		tidx = bpf_mprog_pos_after(entry, &rtuple);
368 		if (tidx < -1 || (idx >= -1 && tidx != idx)) {
369 			ret = tidx < 0 ? tidx : -ERANGE;
370 			goto out;
371 		}
372 		idx = tidx;
373 	}
374 	if (idx < -1) {
375 		if (rtuple.prog || flags) {
376 			ret = -EINVAL;
377 			goto out;
378 		}
379 		idx = bpf_mprog_total(entry);
380 		flags = BPF_F_AFTER;
381 	}
382 	if (idx >= bpf_mprog_max()) {
383 		ret = -ERANGE;
384 		goto out;
385 	}
386 	ret = bpf_mprog_fetch(entry, &dtuple, idx);
387 	if (ret)
388 		goto out;
389 	ret = bpf_mprog_delete(entry, entry_new, &dtuple, idx);
390 out:
391 	bpf_mprog_tuple_put(&rtuple);
392 	return ret;
393 }
394 
395 int bpf_mprog_query(const union bpf_attr *attr, union bpf_attr __user *uattr,
396 		    struct bpf_mprog_entry *entry)
397 {
398 	u32 __user *uprog_flags, *ulink_flags;
399 	u32 __user *uprog_id, *ulink_id;
400 	struct bpf_mprog_fp *fp;
401 	struct bpf_mprog_cp *cp;
402 	struct bpf_prog *prog;
403 	const u32 flags = 0;
404 	u32 id, count = 0;
405 	u64 revision = 1;
406 	int i, ret = 0;
407 
408 	if (attr->query.query_flags || attr->query.attach_flags)
409 		return -EINVAL;
410 	if (entry) {
411 		revision = bpf_mprog_revision(entry);
412 		count = bpf_mprog_total(entry);
413 	}
414 	if (copy_to_user(&uattr->query.attach_flags, &flags, sizeof(flags)))
415 		return -EFAULT;
416 	if (copy_to_user(&uattr->query.revision, &revision, sizeof(revision)))
417 		return -EFAULT;
418 	if (copy_to_user(&uattr->query.count, &count, sizeof(count)))
419 		return -EFAULT;
420 	uprog_id = u64_to_user_ptr(attr->query.prog_ids);
421 	uprog_flags = u64_to_user_ptr(attr->query.prog_attach_flags);
422 	ulink_id = u64_to_user_ptr(attr->query.link_ids);
423 	ulink_flags = u64_to_user_ptr(attr->query.link_attach_flags);
424 	if (attr->query.count == 0 || !uprog_id || !count)
425 		return 0;
426 	if (attr->query.count < count) {
427 		count = attr->query.count;
428 		ret = -ENOSPC;
429 	}
430 	for (i = 0; i < bpf_mprog_max(); i++) {
431 		bpf_mprog_read(entry, i, &fp, &cp);
432 		prog = READ_ONCE(fp->prog);
433 		if (!prog)
434 			break;
435 		id = prog->aux->id;
436 		if (copy_to_user(uprog_id + i, &id, sizeof(id)))
437 			return -EFAULT;
438 		if (uprog_flags &&
439 		    copy_to_user(uprog_flags + i, &flags, sizeof(flags)))
440 			return -EFAULT;
441 		id = cp->link ? cp->link->id : 0;
442 		if (ulink_id &&
443 		    copy_to_user(ulink_id + i, &id, sizeof(id)))
444 			return -EFAULT;
445 		if (ulink_flags &&
446 		    copy_to_user(ulink_flags + i, &flags, sizeof(flags)))
447 			return -EFAULT;
448 		if (i + 1 == count)
449 			break;
450 	}
451 	return ret;
452 }
453