xref: /linux/kernel/bpf/arraymap.c (revision 0a91336e287ca2557fead5221d2c79e0effd034e)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
3  * Copyright (c) 2016,2017 Facebook
4  */
5 #include <linux/bpf.h>
6 #include <linux/btf.h>
7 #include <linux/err.h>
8 #include <linux/slab.h>
9 #include <linux/mm.h>
10 #include <linux/filter.h>
11 #include <linux/perf_event.h>
12 #include <uapi/linux/btf.h>
13 #include <linux/rcupdate_trace.h>
14 #include <linux/btf_ids.h>
15 
16 #include "map_in_map.h"
17 
18 #define ARRAY_CREATE_FLAG_MASK \
19 	(BPF_F_NUMA_NODE | BPF_F_MMAPABLE | BPF_F_ACCESS_MASK | \
20 	 BPF_F_PRESERVE_ELEMS | BPF_F_INNER_MAP)
21 
bpf_array_free_percpu(struct bpf_array * array)22 static void bpf_array_free_percpu(struct bpf_array *array)
23 {
24 	int i;
25 
26 	for (i = 0; i < array->map.max_entries; i++) {
27 		free_percpu(array->pptrs[i]);
28 		cond_resched();
29 	}
30 }
31 
bpf_array_alloc_percpu(struct bpf_array * array)32 static int bpf_array_alloc_percpu(struct bpf_array *array)
33 {
34 	void __percpu *ptr;
35 	int i;
36 
37 	for (i = 0; i < array->map.max_entries; i++) {
38 		ptr = bpf_map_alloc_percpu(&array->map, array->elem_size, 8,
39 					   GFP_USER | __GFP_NOWARN);
40 		if (!ptr) {
41 			bpf_array_free_percpu(array);
42 			return -ENOMEM;
43 		}
44 		array->pptrs[i] = ptr;
45 		cond_resched();
46 	}
47 
48 	return 0;
49 }
50 
51 /* Called from syscall */
array_map_alloc_check(union bpf_attr * attr)52 int array_map_alloc_check(union bpf_attr *attr)
53 {
54 	bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
55 	int numa_node = bpf_map_attr_numa_node(attr);
56 
57 	/* check sanity of attributes */
58 	if (attr->max_entries == 0 || attr->key_size != 4 ||
59 	    attr->value_size == 0 ||
60 	    attr->map_flags & ~ARRAY_CREATE_FLAG_MASK ||
61 	    !bpf_map_flags_access_ok(attr->map_flags) ||
62 	    (percpu && numa_node != NUMA_NO_NODE))
63 		return -EINVAL;
64 
65 	if (attr->map_type != BPF_MAP_TYPE_ARRAY &&
66 	    attr->map_flags & (BPF_F_MMAPABLE | BPF_F_INNER_MAP))
67 		return -EINVAL;
68 
69 	if (attr->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY &&
70 	    attr->map_flags & BPF_F_PRESERVE_ELEMS)
71 		return -EINVAL;
72 
73 	/* avoid overflow on round_up(map->value_size) */
74 	if (attr->value_size > INT_MAX)
75 		return -E2BIG;
76 	/* percpu map value size is bound by PCPU_MIN_UNIT_SIZE */
77 	if (percpu && round_up(attr->value_size, 8) > PCPU_MIN_UNIT_SIZE)
78 		return -E2BIG;
79 
80 	return 0;
81 }
82 
array_map_alloc(union bpf_attr * attr)83 static struct bpf_map *array_map_alloc(union bpf_attr *attr)
84 {
85 	bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
86 	int numa_node = bpf_map_attr_numa_node(attr);
87 	u32 elem_size, index_mask, max_entries;
88 	bool bypass_spec_v1 = bpf_bypass_spec_v1(NULL);
89 	u64 array_size, mask64;
90 	struct bpf_array *array;
91 
92 	elem_size = round_up(attr->value_size, 8);
93 
94 	max_entries = attr->max_entries;
95 
96 	/* On 32 bit archs roundup_pow_of_two() with max_entries that has
97 	 * upper most bit set in u32 space is undefined behavior due to
98 	 * resulting 1U << 32, so do it manually here in u64 space.
99 	 */
100 	mask64 = fls_long(max_entries - 1);
101 	mask64 = 1ULL << mask64;
102 	mask64 -= 1;
103 
104 	index_mask = mask64;
105 	if (!bypass_spec_v1) {
106 		/* round up array size to nearest power of 2,
107 		 * since cpu will speculate within index_mask limits
108 		 */
109 		max_entries = index_mask + 1;
110 		/* Check for overflows. */
111 		if (max_entries < attr->max_entries)
112 			return ERR_PTR(-E2BIG);
113 	}
114 
115 	array_size = sizeof(*array);
116 	if (percpu) {
117 		array_size += (u64) max_entries * sizeof(void *);
118 	} else {
119 		/* rely on vmalloc() to return page-aligned memory and
120 		 * ensure array->value is exactly page-aligned
121 		 */
122 		if (attr->map_flags & BPF_F_MMAPABLE) {
123 			array_size = PAGE_ALIGN(array_size);
124 			array_size += PAGE_ALIGN((u64) max_entries * elem_size);
125 		} else {
126 			array_size += (u64) max_entries * elem_size;
127 		}
128 	}
129 
130 	/* allocate all map elements and zero-initialize them */
131 	if (attr->map_flags & BPF_F_MMAPABLE) {
132 		void *data;
133 
134 		/* kmalloc'ed memory can't be mmap'ed, use explicit vmalloc */
135 		data = bpf_map_area_mmapable_alloc(array_size, numa_node);
136 		if (!data)
137 			return ERR_PTR(-ENOMEM);
138 		array = data + PAGE_ALIGN(sizeof(struct bpf_array))
139 			- offsetof(struct bpf_array, value);
140 	} else {
141 		array = bpf_map_area_alloc(array_size, numa_node);
142 	}
143 	if (!array)
144 		return ERR_PTR(-ENOMEM);
145 	array->index_mask = index_mask;
146 	array->map.bypass_spec_v1 = bypass_spec_v1;
147 
148 	/* copy mandatory map attributes */
149 	bpf_map_init_from_attr(&array->map, attr);
150 	array->elem_size = elem_size;
151 
152 	if (percpu && bpf_array_alloc_percpu(array)) {
153 		bpf_map_area_free(array);
154 		return ERR_PTR(-ENOMEM);
155 	}
156 
157 	return &array->map;
158 }
159 
array_map_elem_ptr(struct bpf_array * array,u32 index)160 static void *array_map_elem_ptr(struct bpf_array* array, u32 index)
161 {
162 	return array->value + (u64)array->elem_size * index;
163 }
164 
165 /* Called from syscall or from eBPF program */
array_map_lookup_elem(struct bpf_map * map,void * key)166 static void *array_map_lookup_elem(struct bpf_map *map, void *key)
167 {
168 	struct bpf_array *array = container_of(map, struct bpf_array, map);
169 	u32 index = *(u32 *)key;
170 
171 	if (unlikely(index >= array->map.max_entries))
172 		return NULL;
173 
174 	return array->value + (u64)array->elem_size * (index & array->index_mask);
175 }
176 
array_map_direct_value_addr(const struct bpf_map * map,u64 * imm,u32 off)177 static int array_map_direct_value_addr(const struct bpf_map *map, u64 *imm,
178 				       u32 off)
179 {
180 	struct bpf_array *array = container_of(map, struct bpf_array, map);
181 
182 	if (map->max_entries != 1)
183 		return -ENOTSUPP;
184 	if (off >= map->value_size)
185 		return -EINVAL;
186 
187 	*imm = (unsigned long)array->value;
188 	return 0;
189 }
190 
array_map_direct_value_meta(const struct bpf_map * map,u64 imm,u32 * off)191 static int array_map_direct_value_meta(const struct bpf_map *map, u64 imm,
192 				       u32 *off)
193 {
194 	struct bpf_array *array = container_of(map, struct bpf_array, map);
195 	u64 base = (unsigned long)array->value;
196 	u64 range = array->elem_size;
197 
198 	if (map->max_entries != 1)
199 		return -ENOTSUPP;
200 	if (imm < base || imm >= base + range)
201 		return -ENOENT;
202 
203 	*off = imm - base;
204 	return 0;
205 }
206 
207 /* emit BPF instructions equivalent to C code of array_map_lookup_elem() */
array_map_gen_lookup(struct bpf_map * map,struct bpf_insn * insn_buf)208 static int array_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
209 {
210 	struct bpf_array *array = container_of(map, struct bpf_array, map);
211 	struct bpf_insn *insn = insn_buf;
212 	u32 elem_size = array->elem_size;
213 	const int ret = BPF_REG_0;
214 	const int map_ptr = BPF_REG_1;
215 	const int index = BPF_REG_2;
216 
217 	if (map->map_flags & BPF_F_INNER_MAP)
218 		return -EOPNOTSUPP;
219 
220 	*insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
221 	*insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
222 	if (!map->bypass_spec_v1) {
223 		*insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 4);
224 		*insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask);
225 	} else {
226 		*insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 3);
227 	}
228 
229 	if (is_power_of_2(elem_size)) {
230 		*insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
231 	} else {
232 		*insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
233 	}
234 	*insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
235 	*insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
236 	*insn++ = BPF_MOV64_IMM(ret, 0);
237 	return insn - insn_buf;
238 }
239 
240 /* Called from eBPF program */
percpu_array_map_lookup_elem(struct bpf_map * map,void * key)241 static void *percpu_array_map_lookup_elem(struct bpf_map *map, void *key)
242 {
243 	struct bpf_array *array = container_of(map, struct bpf_array, map);
244 	u32 index = *(u32 *)key;
245 
246 	if (unlikely(index >= array->map.max_entries))
247 		return NULL;
248 
249 	return this_cpu_ptr(array->pptrs[index & array->index_mask]);
250 }
251 
252 /* emit BPF instructions equivalent to C code of percpu_array_map_lookup_elem() */
percpu_array_map_gen_lookup(struct bpf_map * map,struct bpf_insn * insn_buf)253 static int percpu_array_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
254 {
255 	struct bpf_array *array = container_of(map, struct bpf_array, map);
256 	struct bpf_insn *insn = insn_buf;
257 
258 	if (!bpf_jit_supports_percpu_insn())
259 		return -EOPNOTSUPP;
260 
261 	if (map->map_flags & BPF_F_INNER_MAP)
262 		return -EOPNOTSUPP;
263 
264 	BUILD_BUG_ON(offsetof(struct bpf_array, map) != 0);
265 	*insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, offsetof(struct bpf_array, pptrs));
266 
267 	*insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_0, BPF_REG_2, 0);
268 	if (!map->bypass_spec_v1) {
269 		*insn++ = BPF_JMP_IMM(BPF_JGE, BPF_REG_0, map->max_entries, 6);
270 		*insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_0, array->index_mask);
271 	} else {
272 		*insn++ = BPF_JMP_IMM(BPF_JGE, BPF_REG_0, map->max_entries, 5);
273 	}
274 
275 	*insn++ = BPF_ALU64_IMM(BPF_LSH, BPF_REG_0, 3);
276 	*insn++ = BPF_ALU64_REG(BPF_ADD, BPF_REG_0, BPF_REG_1);
277 	*insn++ = BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_0, 0);
278 	*insn++ = BPF_MOV64_PERCPU_REG(BPF_REG_0, BPF_REG_0);
279 	*insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
280 	*insn++ = BPF_MOV64_IMM(BPF_REG_0, 0);
281 	return insn - insn_buf;
282 }
283 
percpu_array_map_lookup_percpu_elem(struct bpf_map * map,void * key,u32 cpu)284 static void *percpu_array_map_lookup_percpu_elem(struct bpf_map *map, void *key, u32 cpu)
285 {
286 	struct bpf_array *array = container_of(map, struct bpf_array, map);
287 	u32 index = *(u32 *)key;
288 
289 	if (cpu >= nr_cpu_ids)
290 		return NULL;
291 
292 	if (unlikely(index >= array->map.max_entries))
293 		return NULL;
294 
295 	return per_cpu_ptr(array->pptrs[index & array->index_mask], cpu);
296 }
297 
bpf_percpu_array_copy(struct bpf_map * map,void * key,void * value)298 int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value)
299 {
300 	struct bpf_array *array = container_of(map, struct bpf_array, map);
301 	u32 index = *(u32 *)key;
302 	void __percpu *pptr;
303 	int cpu, off = 0;
304 	u32 size;
305 
306 	if (unlikely(index >= array->map.max_entries))
307 		return -ENOENT;
308 
309 	/* per_cpu areas are zero-filled and bpf programs can only
310 	 * access 'value_size' of them, so copying rounded areas
311 	 * will not leak any kernel data
312 	 */
313 	size = array->elem_size;
314 	rcu_read_lock();
315 	pptr = array->pptrs[index & array->index_mask];
316 	for_each_possible_cpu(cpu) {
317 		copy_map_value_long(map, value + off, per_cpu_ptr(pptr, cpu));
318 		check_and_init_map_value(map, value + off);
319 		off += size;
320 	}
321 	rcu_read_unlock();
322 	return 0;
323 }
324 
325 /* Called from syscall */
array_map_get_next_key(struct bpf_map * map,void * key,void * next_key)326 static int array_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
327 {
328 	struct bpf_array *array = container_of(map, struct bpf_array, map);
329 	u32 index = key ? *(u32 *)key : U32_MAX;
330 	u32 *next = (u32 *)next_key;
331 
332 	if (index >= array->map.max_entries) {
333 		*next = 0;
334 		return 0;
335 	}
336 
337 	if (index == array->map.max_entries - 1)
338 		return -ENOENT;
339 
340 	*next = index + 1;
341 	return 0;
342 }
343 
344 /* Called from syscall or from eBPF program */
array_map_update_elem(struct bpf_map * map,void * key,void * value,u64 map_flags)345 static long array_map_update_elem(struct bpf_map *map, void *key, void *value,
346 				  u64 map_flags)
347 {
348 	struct bpf_array *array = container_of(map, struct bpf_array, map);
349 	u32 index = *(u32 *)key;
350 	char *val;
351 
352 	if (unlikely((map_flags & ~BPF_F_LOCK) > BPF_EXIST))
353 		/* unknown flags */
354 		return -EINVAL;
355 
356 	if (unlikely(index >= array->map.max_entries))
357 		/* all elements were pre-allocated, cannot insert a new one */
358 		return -E2BIG;
359 
360 	if (unlikely(map_flags & BPF_NOEXIST))
361 		/* all elements already exist */
362 		return -EEXIST;
363 
364 	if (unlikely((map_flags & BPF_F_LOCK) &&
365 		     !btf_record_has_field(map->record, BPF_SPIN_LOCK)))
366 		return -EINVAL;
367 
368 	if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
369 		val = this_cpu_ptr(array->pptrs[index & array->index_mask]);
370 		copy_map_value(map, val, value);
371 		bpf_obj_free_fields(array->map.record, val);
372 	} else {
373 		val = array->value +
374 			(u64)array->elem_size * (index & array->index_mask);
375 		if (map_flags & BPF_F_LOCK)
376 			copy_map_value_locked(map, val, value, false);
377 		else
378 			copy_map_value(map, val, value);
379 		bpf_obj_free_fields(array->map.record, val);
380 	}
381 	return 0;
382 }
383 
bpf_percpu_array_update(struct bpf_map * map,void * key,void * value,u64 map_flags)384 int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value,
385 			    u64 map_flags)
386 {
387 	struct bpf_array *array = container_of(map, struct bpf_array, map);
388 	u32 index = *(u32 *)key;
389 	void __percpu *pptr;
390 	int cpu, off = 0;
391 	u32 size;
392 
393 	if (unlikely(map_flags > BPF_EXIST))
394 		/* unknown flags */
395 		return -EINVAL;
396 
397 	if (unlikely(index >= array->map.max_entries))
398 		/* all elements were pre-allocated, cannot insert a new one */
399 		return -E2BIG;
400 
401 	if (unlikely(map_flags == BPF_NOEXIST))
402 		/* all elements already exist */
403 		return -EEXIST;
404 
405 	/* the user space will provide round_up(value_size, 8) bytes that
406 	 * will be copied into per-cpu area. bpf programs can only access
407 	 * value_size of it. During lookup the same extra bytes will be
408 	 * returned or zeros which were zero-filled by percpu_alloc,
409 	 * so no kernel data leaks possible
410 	 */
411 	size = array->elem_size;
412 	rcu_read_lock();
413 	pptr = array->pptrs[index & array->index_mask];
414 	for_each_possible_cpu(cpu) {
415 		copy_map_value_long(map, per_cpu_ptr(pptr, cpu), value + off);
416 		bpf_obj_free_fields(array->map.record, per_cpu_ptr(pptr, cpu));
417 		off += size;
418 	}
419 	rcu_read_unlock();
420 	return 0;
421 }
422 
423 /* Called from syscall or from eBPF program */
array_map_delete_elem(struct bpf_map * map,void * key)424 static long array_map_delete_elem(struct bpf_map *map, void *key)
425 {
426 	return -EINVAL;
427 }
428 
array_map_vmalloc_addr(struct bpf_array * array)429 static void *array_map_vmalloc_addr(struct bpf_array *array)
430 {
431 	return (void *)round_down((unsigned long)array, PAGE_SIZE);
432 }
433 
array_map_free_timers_wq(struct bpf_map * map)434 static void array_map_free_timers_wq(struct bpf_map *map)
435 {
436 	struct bpf_array *array = container_of(map, struct bpf_array, map);
437 	int i;
438 
439 	/* We don't reset or free fields other than timer and workqueue
440 	 * on uref dropping to zero.
441 	 */
442 	if (btf_record_has_field(map->record, BPF_TIMER | BPF_WORKQUEUE)) {
443 		for (i = 0; i < array->map.max_entries; i++) {
444 			if (btf_record_has_field(map->record, BPF_TIMER))
445 				bpf_obj_free_timer(map->record, array_map_elem_ptr(array, i));
446 			if (btf_record_has_field(map->record, BPF_WORKQUEUE))
447 				bpf_obj_free_workqueue(map->record, array_map_elem_ptr(array, i));
448 		}
449 	}
450 }
451 
452 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
array_map_free(struct bpf_map * map)453 static void array_map_free(struct bpf_map *map)
454 {
455 	struct bpf_array *array = container_of(map, struct bpf_array, map);
456 	int i;
457 
458 	if (!IS_ERR_OR_NULL(map->record)) {
459 		if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
460 			for (i = 0; i < array->map.max_entries; i++) {
461 				void __percpu *pptr = array->pptrs[i & array->index_mask];
462 				int cpu;
463 
464 				for_each_possible_cpu(cpu) {
465 					bpf_obj_free_fields(map->record, per_cpu_ptr(pptr, cpu));
466 					cond_resched();
467 				}
468 			}
469 		} else {
470 			for (i = 0; i < array->map.max_entries; i++)
471 				bpf_obj_free_fields(map->record, array_map_elem_ptr(array, i));
472 		}
473 	}
474 
475 	if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
476 		bpf_array_free_percpu(array);
477 
478 	if (array->map.map_flags & BPF_F_MMAPABLE)
479 		bpf_map_area_free(array_map_vmalloc_addr(array));
480 	else
481 		bpf_map_area_free(array);
482 }
483 
array_map_seq_show_elem(struct bpf_map * map,void * key,struct seq_file * m)484 static void array_map_seq_show_elem(struct bpf_map *map, void *key,
485 				    struct seq_file *m)
486 {
487 	void *value;
488 
489 	rcu_read_lock();
490 
491 	value = array_map_lookup_elem(map, key);
492 	if (!value) {
493 		rcu_read_unlock();
494 		return;
495 	}
496 
497 	if (map->btf_key_type_id)
498 		seq_printf(m, "%u: ", *(u32 *)key);
499 	btf_type_seq_show(map->btf, map->btf_value_type_id, value, m);
500 	seq_putc(m, '\n');
501 
502 	rcu_read_unlock();
503 }
504 
percpu_array_map_seq_show_elem(struct bpf_map * map,void * key,struct seq_file * m)505 static void percpu_array_map_seq_show_elem(struct bpf_map *map, void *key,
506 					   struct seq_file *m)
507 {
508 	struct bpf_array *array = container_of(map, struct bpf_array, map);
509 	u32 index = *(u32 *)key;
510 	void __percpu *pptr;
511 	int cpu;
512 
513 	rcu_read_lock();
514 
515 	seq_printf(m, "%u: {\n", *(u32 *)key);
516 	pptr = array->pptrs[index & array->index_mask];
517 	for_each_possible_cpu(cpu) {
518 		seq_printf(m, "\tcpu%d: ", cpu);
519 		btf_type_seq_show(map->btf, map->btf_value_type_id,
520 				  per_cpu_ptr(pptr, cpu), m);
521 		seq_putc(m, '\n');
522 	}
523 	seq_puts(m, "}\n");
524 
525 	rcu_read_unlock();
526 }
527 
array_map_check_btf(const struct bpf_map * map,const struct btf * btf,const struct btf_type * key_type,const struct btf_type * value_type)528 static int array_map_check_btf(const struct bpf_map *map,
529 			       const struct btf *btf,
530 			       const struct btf_type *key_type,
531 			       const struct btf_type *value_type)
532 {
533 	/* One exception for keyless BTF: .bss/.data/.rodata map */
534 	if (btf_type_is_void(key_type)) {
535 		if (map->map_type != BPF_MAP_TYPE_ARRAY ||
536 		    map->max_entries != 1)
537 			return -EINVAL;
538 
539 		if (BTF_INFO_KIND(value_type->info) != BTF_KIND_DATASEC)
540 			return -EINVAL;
541 
542 		return 0;
543 	}
544 
545 	/*
546 	 * Bpf array can only take a u32 key. This check makes sure
547 	 * that the btf matches the attr used during map_create.
548 	 */
549 	if (!btf_type_is_i32(key_type))
550 		return -EINVAL;
551 
552 	return 0;
553 }
554 
array_map_mmap(struct bpf_map * map,struct vm_area_struct * vma)555 static int array_map_mmap(struct bpf_map *map, struct vm_area_struct *vma)
556 {
557 	struct bpf_array *array = container_of(map, struct bpf_array, map);
558 	pgoff_t pgoff = PAGE_ALIGN(sizeof(*array)) >> PAGE_SHIFT;
559 
560 	if (!(map->map_flags & BPF_F_MMAPABLE))
561 		return -EINVAL;
562 
563 	if (vma->vm_pgoff * PAGE_SIZE + (vma->vm_end - vma->vm_start) >
564 	    PAGE_ALIGN((u64)array->map.max_entries * array->elem_size))
565 		return -EINVAL;
566 
567 	return remap_vmalloc_range(vma, array_map_vmalloc_addr(array),
568 				   vma->vm_pgoff + pgoff);
569 }
570 
array_map_meta_equal(const struct bpf_map * meta0,const struct bpf_map * meta1)571 static bool array_map_meta_equal(const struct bpf_map *meta0,
572 				 const struct bpf_map *meta1)
573 {
574 	if (!bpf_map_meta_equal(meta0, meta1))
575 		return false;
576 	return meta0->map_flags & BPF_F_INNER_MAP ? true :
577 	       meta0->max_entries == meta1->max_entries;
578 }
579 
580 struct bpf_iter_seq_array_map_info {
581 	struct bpf_map *map;
582 	void *percpu_value_buf;
583 	u32 index;
584 };
585 
bpf_array_map_seq_start(struct seq_file * seq,loff_t * pos)586 static void *bpf_array_map_seq_start(struct seq_file *seq, loff_t *pos)
587 {
588 	struct bpf_iter_seq_array_map_info *info = seq->private;
589 	struct bpf_map *map = info->map;
590 	struct bpf_array *array;
591 	u32 index;
592 
593 	if (info->index >= map->max_entries)
594 		return NULL;
595 
596 	if (*pos == 0)
597 		++*pos;
598 	array = container_of(map, struct bpf_array, map);
599 	index = info->index & array->index_mask;
600 	if (info->percpu_value_buf)
601 		return (void *)(uintptr_t)array->pptrs[index];
602 	return array_map_elem_ptr(array, index);
603 }
604 
bpf_array_map_seq_next(struct seq_file * seq,void * v,loff_t * pos)605 static void *bpf_array_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
606 {
607 	struct bpf_iter_seq_array_map_info *info = seq->private;
608 	struct bpf_map *map = info->map;
609 	struct bpf_array *array;
610 	u32 index;
611 
612 	++*pos;
613 	++info->index;
614 	if (info->index >= map->max_entries)
615 		return NULL;
616 
617 	array = container_of(map, struct bpf_array, map);
618 	index = info->index & array->index_mask;
619 	if (info->percpu_value_buf)
620 		return (void *)(uintptr_t)array->pptrs[index];
621 	return array_map_elem_ptr(array, index);
622 }
623 
__bpf_array_map_seq_show(struct seq_file * seq,void * v)624 static int __bpf_array_map_seq_show(struct seq_file *seq, void *v)
625 {
626 	struct bpf_iter_seq_array_map_info *info = seq->private;
627 	struct bpf_iter__bpf_map_elem ctx = {};
628 	struct bpf_map *map = info->map;
629 	struct bpf_array *array = container_of(map, struct bpf_array, map);
630 	struct bpf_iter_meta meta;
631 	struct bpf_prog *prog;
632 	int off = 0, cpu = 0;
633 	void __percpu *pptr;
634 	u32 size;
635 
636 	meta.seq = seq;
637 	prog = bpf_iter_get_info(&meta, v == NULL);
638 	if (!prog)
639 		return 0;
640 
641 	ctx.meta = &meta;
642 	ctx.map = info->map;
643 	if (v) {
644 		ctx.key = &info->index;
645 
646 		if (!info->percpu_value_buf) {
647 			ctx.value = v;
648 		} else {
649 			pptr = (void __percpu *)(uintptr_t)v;
650 			size = array->elem_size;
651 			for_each_possible_cpu(cpu) {
652 				copy_map_value_long(map, info->percpu_value_buf + off,
653 						    per_cpu_ptr(pptr, cpu));
654 				check_and_init_map_value(map, info->percpu_value_buf + off);
655 				off += size;
656 			}
657 			ctx.value = info->percpu_value_buf;
658 		}
659 	}
660 
661 	return bpf_iter_run_prog(prog, &ctx);
662 }
663 
bpf_array_map_seq_show(struct seq_file * seq,void * v)664 static int bpf_array_map_seq_show(struct seq_file *seq, void *v)
665 {
666 	return __bpf_array_map_seq_show(seq, v);
667 }
668 
bpf_array_map_seq_stop(struct seq_file * seq,void * v)669 static void bpf_array_map_seq_stop(struct seq_file *seq, void *v)
670 {
671 	if (!v)
672 		(void)__bpf_array_map_seq_show(seq, NULL);
673 }
674 
bpf_iter_init_array_map(void * priv_data,struct bpf_iter_aux_info * aux)675 static int bpf_iter_init_array_map(void *priv_data,
676 				   struct bpf_iter_aux_info *aux)
677 {
678 	struct bpf_iter_seq_array_map_info *seq_info = priv_data;
679 	struct bpf_map *map = aux->map;
680 	struct bpf_array *array = container_of(map, struct bpf_array, map);
681 	void *value_buf;
682 	u32 buf_size;
683 
684 	if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
685 		buf_size = array->elem_size * num_possible_cpus();
686 		value_buf = kmalloc(buf_size, GFP_USER | __GFP_NOWARN);
687 		if (!value_buf)
688 			return -ENOMEM;
689 
690 		seq_info->percpu_value_buf = value_buf;
691 	}
692 
693 	/* bpf_iter_attach_map() acquires a map uref, and the uref may be
694 	 * released before or in the middle of iterating map elements, so
695 	 * acquire an extra map uref for iterator.
696 	 */
697 	bpf_map_inc_with_uref(map);
698 	seq_info->map = map;
699 	return 0;
700 }
701 
bpf_iter_fini_array_map(void * priv_data)702 static void bpf_iter_fini_array_map(void *priv_data)
703 {
704 	struct bpf_iter_seq_array_map_info *seq_info = priv_data;
705 
706 	bpf_map_put_with_uref(seq_info->map);
707 	kfree(seq_info->percpu_value_buf);
708 }
709 
710 static const struct seq_operations bpf_array_map_seq_ops = {
711 	.start	= bpf_array_map_seq_start,
712 	.next	= bpf_array_map_seq_next,
713 	.stop	= bpf_array_map_seq_stop,
714 	.show	= bpf_array_map_seq_show,
715 };
716 
717 static const struct bpf_iter_seq_info iter_seq_info = {
718 	.seq_ops		= &bpf_array_map_seq_ops,
719 	.init_seq_private	= bpf_iter_init_array_map,
720 	.fini_seq_private	= bpf_iter_fini_array_map,
721 	.seq_priv_size		= sizeof(struct bpf_iter_seq_array_map_info),
722 };
723 
bpf_for_each_array_elem(struct bpf_map * map,bpf_callback_t callback_fn,void * callback_ctx,u64 flags)724 static long bpf_for_each_array_elem(struct bpf_map *map, bpf_callback_t callback_fn,
725 				    void *callback_ctx, u64 flags)
726 {
727 	u32 i, key, num_elems = 0;
728 	struct bpf_array *array;
729 	bool is_percpu;
730 	u64 ret = 0;
731 	void *val;
732 
733 	cant_migrate();
734 
735 	if (flags != 0)
736 		return -EINVAL;
737 
738 	is_percpu = map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
739 	array = container_of(map, struct bpf_array, map);
740 	for (i = 0; i < map->max_entries; i++) {
741 		if (is_percpu)
742 			val = this_cpu_ptr(array->pptrs[i]);
743 		else
744 			val = array_map_elem_ptr(array, i);
745 		num_elems++;
746 		key = i;
747 		ret = callback_fn((u64)(long)map, (u64)(long)&key,
748 				  (u64)(long)val, (u64)(long)callback_ctx, 0);
749 		/* return value: 0 - continue, 1 - stop and return */
750 		if (ret)
751 			break;
752 	}
753 
754 	return num_elems;
755 }
756 
array_map_mem_usage(const struct bpf_map * map)757 static u64 array_map_mem_usage(const struct bpf_map *map)
758 {
759 	struct bpf_array *array = container_of(map, struct bpf_array, map);
760 	bool percpu = map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
761 	u32 elem_size = array->elem_size;
762 	u64 entries = map->max_entries;
763 	u64 usage = sizeof(*array);
764 
765 	if (percpu) {
766 		usage += entries * sizeof(void *);
767 		usage += entries * elem_size * num_possible_cpus();
768 	} else {
769 		if (map->map_flags & BPF_F_MMAPABLE) {
770 			usage = PAGE_ALIGN(usage);
771 			usage += PAGE_ALIGN(entries * elem_size);
772 		} else {
773 			usage += entries * elem_size;
774 		}
775 	}
776 	return usage;
777 }
778 
779 BTF_ID_LIST_SINGLE(array_map_btf_ids, struct, bpf_array)
780 const struct bpf_map_ops array_map_ops = {
781 	.map_meta_equal = array_map_meta_equal,
782 	.map_alloc_check = array_map_alloc_check,
783 	.map_alloc = array_map_alloc,
784 	.map_free = array_map_free,
785 	.map_get_next_key = array_map_get_next_key,
786 	.map_release_uref = array_map_free_timers_wq,
787 	.map_lookup_elem = array_map_lookup_elem,
788 	.map_update_elem = array_map_update_elem,
789 	.map_delete_elem = array_map_delete_elem,
790 	.map_gen_lookup = array_map_gen_lookup,
791 	.map_direct_value_addr = array_map_direct_value_addr,
792 	.map_direct_value_meta = array_map_direct_value_meta,
793 	.map_mmap = array_map_mmap,
794 	.map_seq_show_elem = array_map_seq_show_elem,
795 	.map_check_btf = array_map_check_btf,
796 	.map_lookup_batch = generic_map_lookup_batch,
797 	.map_update_batch = generic_map_update_batch,
798 	.map_set_for_each_callback_args = map_set_for_each_callback_args,
799 	.map_for_each_callback = bpf_for_each_array_elem,
800 	.map_mem_usage = array_map_mem_usage,
801 	.map_btf_id = &array_map_btf_ids[0],
802 	.iter_seq_info = &iter_seq_info,
803 };
804 
805 const struct bpf_map_ops percpu_array_map_ops = {
806 	.map_meta_equal = bpf_map_meta_equal,
807 	.map_alloc_check = array_map_alloc_check,
808 	.map_alloc = array_map_alloc,
809 	.map_free = array_map_free,
810 	.map_get_next_key = array_map_get_next_key,
811 	.map_lookup_elem = percpu_array_map_lookup_elem,
812 	.map_gen_lookup = percpu_array_map_gen_lookup,
813 	.map_update_elem = array_map_update_elem,
814 	.map_delete_elem = array_map_delete_elem,
815 	.map_lookup_percpu_elem = percpu_array_map_lookup_percpu_elem,
816 	.map_seq_show_elem = percpu_array_map_seq_show_elem,
817 	.map_check_btf = array_map_check_btf,
818 	.map_lookup_batch = generic_map_lookup_batch,
819 	.map_update_batch = generic_map_update_batch,
820 	.map_set_for_each_callback_args = map_set_for_each_callback_args,
821 	.map_for_each_callback = bpf_for_each_array_elem,
822 	.map_mem_usage = array_map_mem_usage,
823 	.map_btf_id = &array_map_btf_ids[0],
824 	.iter_seq_info = &iter_seq_info,
825 };
826 
fd_array_map_alloc_check(union bpf_attr * attr)827 static int fd_array_map_alloc_check(union bpf_attr *attr)
828 {
829 	/* only file descriptors can be stored in this type of map */
830 	if (attr->value_size != sizeof(u32))
831 		return -EINVAL;
832 	/* Program read-only/write-only not supported for special maps yet. */
833 	if (attr->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG))
834 		return -EINVAL;
835 	return array_map_alloc_check(attr);
836 }
837 
fd_array_map_free(struct bpf_map * map)838 static void fd_array_map_free(struct bpf_map *map)
839 {
840 	struct bpf_array *array = container_of(map, struct bpf_array, map);
841 	int i;
842 
843 	/* make sure it's empty */
844 	for (i = 0; i < array->map.max_entries; i++)
845 		BUG_ON(array->ptrs[i] != NULL);
846 
847 	bpf_map_area_free(array);
848 }
849 
fd_array_map_lookup_elem(struct bpf_map * map,void * key)850 static void *fd_array_map_lookup_elem(struct bpf_map *map, void *key)
851 {
852 	return ERR_PTR(-EOPNOTSUPP);
853 }
854 
855 /* only called from syscall */
bpf_fd_array_map_lookup_elem(struct bpf_map * map,void * key,u32 * value)856 int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value)
857 {
858 	void **elem, *ptr;
859 	int ret =  0;
860 
861 	if (!map->ops->map_fd_sys_lookup_elem)
862 		return -ENOTSUPP;
863 
864 	rcu_read_lock();
865 	elem = array_map_lookup_elem(map, key);
866 	if (elem && (ptr = READ_ONCE(*elem)))
867 		*value = map->ops->map_fd_sys_lookup_elem(ptr);
868 	else
869 		ret = -ENOENT;
870 	rcu_read_unlock();
871 
872 	return ret;
873 }
874 
875 /* only called from syscall */
bpf_fd_array_map_update_elem(struct bpf_map * map,struct file * map_file,void * key,void * value,u64 map_flags)876 int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file,
877 				 void *key, void *value, u64 map_flags)
878 {
879 	struct bpf_array *array = container_of(map, struct bpf_array, map);
880 	void *new_ptr, *old_ptr;
881 	u32 index = *(u32 *)key, ufd;
882 
883 	if (map_flags != BPF_ANY)
884 		return -EINVAL;
885 
886 	if (index >= array->map.max_entries)
887 		return -E2BIG;
888 
889 	ufd = *(u32 *)value;
890 	new_ptr = map->ops->map_fd_get_ptr(map, map_file, ufd);
891 	if (IS_ERR(new_ptr))
892 		return PTR_ERR(new_ptr);
893 
894 	if (map->ops->map_poke_run) {
895 		mutex_lock(&array->aux->poke_mutex);
896 		old_ptr = xchg(array->ptrs + index, new_ptr);
897 		map->ops->map_poke_run(map, index, old_ptr, new_ptr);
898 		mutex_unlock(&array->aux->poke_mutex);
899 	} else {
900 		old_ptr = xchg(array->ptrs + index, new_ptr);
901 	}
902 
903 	if (old_ptr)
904 		map->ops->map_fd_put_ptr(map, old_ptr, true);
905 	return 0;
906 }
907 
__fd_array_map_delete_elem(struct bpf_map * map,void * key,bool need_defer)908 static long __fd_array_map_delete_elem(struct bpf_map *map, void *key, bool need_defer)
909 {
910 	struct bpf_array *array = container_of(map, struct bpf_array, map);
911 	void *old_ptr;
912 	u32 index = *(u32 *)key;
913 
914 	if (index >= array->map.max_entries)
915 		return -E2BIG;
916 
917 	if (map->ops->map_poke_run) {
918 		mutex_lock(&array->aux->poke_mutex);
919 		old_ptr = xchg(array->ptrs + index, NULL);
920 		map->ops->map_poke_run(map, index, old_ptr, NULL);
921 		mutex_unlock(&array->aux->poke_mutex);
922 	} else {
923 		old_ptr = xchg(array->ptrs + index, NULL);
924 	}
925 
926 	if (old_ptr) {
927 		map->ops->map_fd_put_ptr(map, old_ptr, need_defer);
928 		return 0;
929 	} else {
930 		return -ENOENT;
931 	}
932 }
933 
fd_array_map_delete_elem(struct bpf_map * map,void * key)934 static long fd_array_map_delete_elem(struct bpf_map *map, void *key)
935 {
936 	return __fd_array_map_delete_elem(map, key, true);
937 }
938 
prog_fd_array_get_ptr(struct bpf_map * map,struct file * map_file,int fd)939 static void *prog_fd_array_get_ptr(struct bpf_map *map,
940 				   struct file *map_file, int fd)
941 {
942 	struct bpf_prog *prog = bpf_prog_get(fd);
943 	bool is_extended;
944 
945 	if (IS_ERR(prog))
946 		return prog;
947 
948 	if (prog->type == BPF_PROG_TYPE_EXT ||
949 	    !bpf_prog_map_compatible(map, prog)) {
950 		bpf_prog_put(prog);
951 		return ERR_PTR(-EINVAL);
952 	}
953 
954 	mutex_lock(&prog->aux->ext_mutex);
955 	is_extended = prog->aux->is_extended;
956 	if (!is_extended)
957 		prog->aux->prog_array_member_cnt++;
958 	mutex_unlock(&prog->aux->ext_mutex);
959 	if (is_extended) {
960 		/* Extended prog can not be tail callee. It's to prevent a
961 		 * potential infinite loop like:
962 		 * tail callee prog entry -> tail callee prog subprog ->
963 		 * freplace prog entry --tailcall-> tail callee prog entry.
964 		 */
965 		bpf_prog_put(prog);
966 		return ERR_PTR(-EBUSY);
967 	}
968 
969 	return prog;
970 }
971 
prog_fd_array_put_ptr(struct bpf_map * map,void * ptr,bool need_defer)972 static void prog_fd_array_put_ptr(struct bpf_map *map, void *ptr, bool need_defer)
973 {
974 	struct bpf_prog *prog = ptr;
975 
976 	mutex_lock(&prog->aux->ext_mutex);
977 	prog->aux->prog_array_member_cnt--;
978 	mutex_unlock(&prog->aux->ext_mutex);
979 	/* bpf_prog is freed after one RCU or tasks trace grace period */
980 	bpf_prog_put(prog);
981 }
982 
prog_fd_array_sys_lookup_elem(void * ptr)983 static u32 prog_fd_array_sys_lookup_elem(void *ptr)
984 {
985 	return ((struct bpf_prog *)ptr)->aux->id;
986 }
987 
988 /* decrement refcnt of all bpf_progs that are stored in this map */
bpf_fd_array_map_clear(struct bpf_map * map,bool need_defer)989 static void bpf_fd_array_map_clear(struct bpf_map *map, bool need_defer)
990 {
991 	struct bpf_array *array = container_of(map, struct bpf_array, map);
992 	int i;
993 
994 	for (i = 0; i < array->map.max_entries; i++)
995 		__fd_array_map_delete_elem(map, &i, need_defer);
996 }
997 
prog_array_map_seq_show_elem(struct bpf_map * map,void * key,struct seq_file * m)998 static void prog_array_map_seq_show_elem(struct bpf_map *map, void *key,
999 					 struct seq_file *m)
1000 {
1001 	void **elem, *ptr;
1002 	u32 prog_id;
1003 
1004 	rcu_read_lock();
1005 
1006 	elem = array_map_lookup_elem(map, key);
1007 	if (elem) {
1008 		ptr = READ_ONCE(*elem);
1009 		if (ptr) {
1010 			seq_printf(m, "%u: ", *(u32 *)key);
1011 			prog_id = prog_fd_array_sys_lookup_elem(ptr);
1012 			btf_type_seq_show(map->btf, map->btf_value_type_id,
1013 					  &prog_id, m);
1014 			seq_putc(m, '\n');
1015 		}
1016 	}
1017 
1018 	rcu_read_unlock();
1019 }
1020 
1021 struct prog_poke_elem {
1022 	struct list_head list;
1023 	struct bpf_prog_aux *aux;
1024 };
1025 
prog_array_map_poke_track(struct bpf_map * map,struct bpf_prog_aux * prog_aux)1026 static int prog_array_map_poke_track(struct bpf_map *map,
1027 				     struct bpf_prog_aux *prog_aux)
1028 {
1029 	struct prog_poke_elem *elem;
1030 	struct bpf_array_aux *aux;
1031 	int ret = 0;
1032 
1033 	aux = container_of(map, struct bpf_array, map)->aux;
1034 	mutex_lock(&aux->poke_mutex);
1035 	list_for_each_entry(elem, &aux->poke_progs, list) {
1036 		if (elem->aux == prog_aux)
1037 			goto out;
1038 	}
1039 
1040 	elem = kmalloc(sizeof(*elem), GFP_KERNEL);
1041 	if (!elem) {
1042 		ret = -ENOMEM;
1043 		goto out;
1044 	}
1045 
1046 	INIT_LIST_HEAD(&elem->list);
1047 	/* We must track the program's aux info at this point in time
1048 	 * since the program pointer itself may not be stable yet, see
1049 	 * also comment in prog_array_map_poke_run().
1050 	 */
1051 	elem->aux = prog_aux;
1052 
1053 	list_add_tail(&elem->list, &aux->poke_progs);
1054 out:
1055 	mutex_unlock(&aux->poke_mutex);
1056 	return ret;
1057 }
1058 
prog_array_map_poke_untrack(struct bpf_map * map,struct bpf_prog_aux * prog_aux)1059 static void prog_array_map_poke_untrack(struct bpf_map *map,
1060 					struct bpf_prog_aux *prog_aux)
1061 {
1062 	struct prog_poke_elem *elem, *tmp;
1063 	struct bpf_array_aux *aux;
1064 
1065 	aux = container_of(map, struct bpf_array, map)->aux;
1066 	mutex_lock(&aux->poke_mutex);
1067 	list_for_each_entry_safe(elem, tmp, &aux->poke_progs, list) {
1068 		if (elem->aux == prog_aux) {
1069 			list_del_init(&elem->list);
1070 			kfree(elem);
1071 			break;
1072 		}
1073 	}
1074 	mutex_unlock(&aux->poke_mutex);
1075 }
1076 
bpf_arch_poke_desc_update(struct bpf_jit_poke_descriptor * poke,struct bpf_prog * new,struct bpf_prog * old)1077 void __weak bpf_arch_poke_desc_update(struct bpf_jit_poke_descriptor *poke,
1078 				      struct bpf_prog *new, struct bpf_prog *old)
1079 {
1080 	WARN_ON_ONCE(1);
1081 }
1082 
prog_array_map_poke_run(struct bpf_map * map,u32 key,struct bpf_prog * old,struct bpf_prog * new)1083 static void prog_array_map_poke_run(struct bpf_map *map, u32 key,
1084 				    struct bpf_prog *old,
1085 				    struct bpf_prog *new)
1086 {
1087 	struct prog_poke_elem *elem;
1088 	struct bpf_array_aux *aux;
1089 
1090 	aux = container_of(map, struct bpf_array, map)->aux;
1091 	WARN_ON_ONCE(!mutex_is_locked(&aux->poke_mutex));
1092 
1093 	list_for_each_entry(elem, &aux->poke_progs, list) {
1094 		struct bpf_jit_poke_descriptor *poke;
1095 		int i;
1096 
1097 		for (i = 0; i < elem->aux->size_poke_tab; i++) {
1098 			poke = &elem->aux->poke_tab[i];
1099 
1100 			/* Few things to be aware of:
1101 			 *
1102 			 * 1) We can only ever access aux in this context, but
1103 			 *    not aux->prog since it might not be stable yet and
1104 			 *    there could be danger of use after free otherwise.
1105 			 * 2) Initially when we start tracking aux, the program
1106 			 *    is not JITed yet and also does not have a kallsyms
1107 			 *    entry. We skip these as poke->tailcall_target_stable
1108 			 *    is not active yet. The JIT will do the final fixup
1109 			 *    before setting it stable. The various
1110 			 *    poke->tailcall_target_stable are successively
1111 			 *    activated, so tail call updates can arrive from here
1112 			 *    while JIT is still finishing its final fixup for
1113 			 *    non-activated poke entries.
1114 			 * 3) Also programs reaching refcount of zero while patching
1115 			 *    is in progress is okay since we're protected under
1116 			 *    poke_mutex and untrack the programs before the JIT
1117 			 *    buffer is freed.
1118 			 */
1119 			if (!READ_ONCE(poke->tailcall_target_stable))
1120 				continue;
1121 			if (poke->reason != BPF_POKE_REASON_TAIL_CALL)
1122 				continue;
1123 			if (poke->tail_call.map != map ||
1124 			    poke->tail_call.key != key)
1125 				continue;
1126 
1127 			bpf_arch_poke_desc_update(poke, new, old);
1128 		}
1129 	}
1130 }
1131 
prog_array_map_clear_deferred(struct work_struct * work)1132 static void prog_array_map_clear_deferred(struct work_struct *work)
1133 {
1134 	struct bpf_map *map = container_of(work, struct bpf_array_aux,
1135 					   work)->map;
1136 	bpf_fd_array_map_clear(map, true);
1137 	bpf_map_put(map);
1138 }
1139 
prog_array_map_clear(struct bpf_map * map)1140 static void prog_array_map_clear(struct bpf_map *map)
1141 {
1142 	struct bpf_array_aux *aux = container_of(map, struct bpf_array,
1143 						 map)->aux;
1144 	bpf_map_inc(map);
1145 	schedule_work(&aux->work);
1146 }
1147 
prog_array_map_alloc(union bpf_attr * attr)1148 static struct bpf_map *prog_array_map_alloc(union bpf_attr *attr)
1149 {
1150 	struct bpf_array_aux *aux;
1151 	struct bpf_map *map;
1152 
1153 	aux = kzalloc(sizeof(*aux), GFP_KERNEL_ACCOUNT);
1154 	if (!aux)
1155 		return ERR_PTR(-ENOMEM);
1156 
1157 	INIT_WORK(&aux->work, prog_array_map_clear_deferred);
1158 	INIT_LIST_HEAD(&aux->poke_progs);
1159 	mutex_init(&aux->poke_mutex);
1160 
1161 	map = array_map_alloc(attr);
1162 	if (IS_ERR(map)) {
1163 		kfree(aux);
1164 		return map;
1165 	}
1166 
1167 	container_of(map, struct bpf_array, map)->aux = aux;
1168 	aux->map = map;
1169 
1170 	return map;
1171 }
1172 
prog_array_map_free(struct bpf_map * map)1173 static void prog_array_map_free(struct bpf_map *map)
1174 {
1175 	struct prog_poke_elem *elem, *tmp;
1176 	struct bpf_array_aux *aux;
1177 
1178 	aux = container_of(map, struct bpf_array, map)->aux;
1179 	list_for_each_entry_safe(elem, tmp, &aux->poke_progs, list) {
1180 		list_del_init(&elem->list);
1181 		kfree(elem);
1182 	}
1183 	kfree(aux);
1184 	fd_array_map_free(map);
1185 }
1186 
1187 /* prog_array->aux->{type,jited} is a runtime binding.
1188  * Doing static check alone in the verifier is not enough.
1189  * Thus, prog_array_map cannot be used as an inner_map
1190  * and map_meta_equal is not implemented.
1191  */
1192 const struct bpf_map_ops prog_array_map_ops = {
1193 	.map_alloc_check = fd_array_map_alloc_check,
1194 	.map_alloc = prog_array_map_alloc,
1195 	.map_free = prog_array_map_free,
1196 	.map_poke_track = prog_array_map_poke_track,
1197 	.map_poke_untrack = prog_array_map_poke_untrack,
1198 	.map_poke_run = prog_array_map_poke_run,
1199 	.map_get_next_key = array_map_get_next_key,
1200 	.map_lookup_elem = fd_array_map_lookup_elem,
1201 	.map_delete_elem = fd_array_map_delete_elem,
1202 	.map_fd_get_ptr = prog_fd_array_get_ptr,
1203 	.map_fd_put_ptr = prog_fd_array_put_ptr,
1204 	.map_fd_sys_lookup_elem = prog_fd_array_sys_lookup_elem,
1205 	.map_release_uref = prog_array_map_clear,
1206 	.map_seq_show_elem = prog_array_map_seq_show_elem,
1207 	.map_mem_usage = array_map_mem_usage,
1208 	.map_btf_id = &array_map_btf_ids[0],
1209 };
1210 
bpf_event_entry_gen(struct file * perf_file,struct file * map_file)1211 static struct bpf_event_entry *bpf_event_entry_gen(struct file *perf_file,
1212 						   struct file *map_file)
1213 {
1214 	struct bpf_event_entry *ee;
1215 
1216 	ee = kzalloc(sizeof(*ee), GFP_KERNEL);
1217 	if (ee) {
1218 		ee->event = perf_file->private_data;
1219 		ee->perf_file = perf_file;
1220 		ee->map_file = map_file;
1221 	}
1222 
1223 	return ee;
1224 }
1225 
__bpf_event_entry_free(struct rcu_head * rcu)1226 static void __bpf_event_entry_free(struct rcu_head *rcu)
1227 {
1228 	struct bpf_event_entry *ee;
1229 
1230 	ee = container_of(rcu, struct bpf_event_entry, rcu);
1231 	fput(ee->perf_file);
1232 	kfree(ee);
1233 }
1234 
bpf_event_entry_free_rcu(struct bpf_event_entry * ee)1235 static void bpf_event_entry_free_rcu(struct bpf_event_entry *ee)
1236 {
1237 	call_rcu(&ee->rcu, __bpf_event_entry_free);
1238 }
1239 
perf_event_fd_array_get_ptr(struct bpf_map * map,struct file * map_file,int fd)1240 static void *perf_event_fd_array_get_ptr(struct bpf_map *map,
1241 					 struct file *map_file, int fd)
1242 {
1243 	struct bpf_event_entry *ee;
1244 	struct perf_event *event;
1245 	struct file *perf_file;
1246 	u64 value;
1247 
1248 	perf_file = perf_event_get(fd);
1249 	if (IS_ERR(perf_file))
1250 		return perf_file;
1251 
1252 	ee = ERR_PTR(-EOPNOTSUPP);
1253 	event = perf_file->private_data;
1254 	if (perf_event_read_local(event, &value, NULL, NULL) == -EOPNOTSUPP)
1255 		goto err_out;
1256 
1257 	ee = bpf_event_entry_gen(perf_file, map_file);
1258 	if (ee)
1259 		return ee;
1260 	ee = ERR_PTR(-ENOMEM);
1261 err_out:
1262 	fput(perf_file);
1263 	return ee;
1264 }
1265 
perf_event_fd_array_put_ptr(struct bpf_map * map,void * ptr,bool need_defer)1266 static void perf_event_fd_array_put_ptr(struct bpf_map *map, void *ptr, bool need_defer)
1267 {
1268 	/* bpf_perf_event is freed after one RCU grace period */
1269 	bpf_event_entry_free_rcu(ptr);
1270 }
1271 
perf_event_fd_array_release(struct bpf_map * map,struct file * map_file)1272 static void perf_event_fd_array_release(struct bpf_map *map,
1273 					struct file *map_file)
1274 {
1275 	struct bpf_array *array = container_of(map, struct bpf_array, map);
1276 	struct bpf_event_entry *ee;
1277 	int i;
1278 
1279 	if (map->map_flags & BPF_F_PRESERVE_ELEMS)
1280 		return;
1281 
1282 	rcu_read_lock();
1283 	for (i = 0; i < array->map.max_entries; i++) {
1284 		ee = READ_ONCE(array->ptrs[i]);
1285 		if (ee && ee->map_file == map_file)
1286 			__fd_array_map_delete_elem(map, &i, true);
1287 	}
1288 	rcu_read_unlock();
1289 }
1290 
perf_event_fd_array_map_free(struct bpf_map * map)1291 static void perf_event_fd_array_map_free(struct bpf_map *map)
1292 {
1293 	if (map->map_flags & BPF_F_PRESERVE_ELEMS)
1294 		bpf_fd_array_map_clear(map, false);
1295 	fd_array_map_free(map);
1296 }
1297 
1298 const struct bpf_map_ops perf_event_array_map_ops = {
1299 	.map_meta_equal = bpf_map_meta_equal,
1300 	.map_alloc_check = fd_array_map_alloc_check,
1301 	.map_alloc = array_map_alloc,
1302 	.map_free = perf_event_fd_array_map_free,
1303 	.map_get_next_key = array_map_get_next_key,
1304 	.map_lookup_elem = fd_array_map_lookup_elem,
1305 	.map_delete_elem = fd_array_map_delete_elem,
1306 	.map_fd_get_ptr = perf_event_fd_array_get_ptr,
1307 	.map_fd_put_ptr = perf_event_fd_array_put_ptr,
1308 	.map_release = perf_event_fd_array_release,
1309 	.map_check_btf = map_check_no_btf,
1310 	.map_mem_usage = array_map_mem_usage,
1311 	.map_btf_id = &array_map_btf_ids[0],
1312 };
1313 
1314 #ifdef CONFIG_CGROUPS
cgroup_fd_array_get_ptr(struct bpf_map * map,struct file * map_file,int fd)1315 static void *cgroup_fd_array_get_ptr(struct bpf_map *map,
1316 				     struct file *map_file /* not used */,
1317 				     int fd)
1318 {
1319 	return cgroup_get_from_fd(fd);
1320 }
1321 
cgroup_fd_array_put_ptr(struct bpf_map * map,void * ptr,bool need_defer)1322 static void cgroup_fd_array_put_ptr(struct bpf_map *map, void *ptr, bool need_defer)
1323 {
1324 	/* cgroup_put free cgrp after a rcu grace period */
1325 	cgroup_put(ptr);
1326 }
1327 
cgroup_fd_array_free(struct bpf_map * map)1328 static void cgroup_fd_array_free(struct bpf_map *map)
1329 {
1330 	bpf_fd_array_map_clear(map, false);
1331 	fd_array_map_free(map);
1332 }
1333 
1334 const struct bpf_map_ops cgroup_array_map_ops = {
1335 	.map_meta_equal = bpf_map_meta_equal,
1336 	.map_alloc_check = fd_array_map_alloc_check,
1337 	.map_alloc = array_map_alloc,
1338 	.map_free = cgroup_fd_array_free,
1339 	.map_get_next_key = array_map_get_next_key,
1340 	.map_lookup_elem = fd_array_map_lookup_elem,
1341 	.map_delete_elem = fd_array_map_delete_elem,
1342 	.map_fd_get_ptr = cgroup_fd_array_get_ptr,
1343 	.map_fd_put_ptr = cgroup_fd_array_put_ptr,
1344 	.map_check_btf = map_check_no_btf,
1345 	.map_mem_usage = array_map_mem_usage,
1346 	.map_btf_id = &array_map_btf_ids[0],
1347 };
1348 #endif
1349 
array_of_map_alloc(union bpf_attr * attr)1350 static struct bpf_map *array_of_map_alloc(union bpf_attr *attr)
1351 {
1352 	struct bpf_map *map, *inner_map_meta;
1353 
1354 	inner_map_meta = bpf_map_meta_alloc(attr->inner_map_fd);
1355 	if (IS_ERR(inner_map_meta))
1356 		return inner_map_meta;
1357 
1358 	map = array_map_alloc(attr);
1359 	if (IS_ERR(map)) {
1360 		bpf_map_meta_free(inner_map_meta);
1361 		return map;
1362 	}
1363 
1364 	map->inner_map_meta = inner_map_meta;
1365 
1366 	return map;
1367 }
1368 
array_of_map_free(struct bpf_map * map)1369 static void array_of_map_free(struct bpf_map *map)
1370 {
1371 	/* map->inner_map_meta is only accessed by syscall which
1372 	 * is protected by fdget/fdput.
1373 	 */
1374 	bpf_map_meta_free(map->inner_map_meta);
1375 	bpf_fd_array_map_clear(map, false);
1376 	fd_array_map_free(map);
1377 }
1378 
array_of_map_lookup_elem(struct bpf_map * map,void * key)1379 static void *array_of_map_lookup_elem(struct bpf_map *map, void *key)
1380 {
1381 	struct bpf_map **inner_map = array_map_lookup_elem(map, key);
1382 
1383 	if (!inner_map)
1384 		return NULL;
1385 
1386 	return READ_ONCE(*inner_map);
1387 }
1388 
array_of_map_gen_lookup(struct bpf_map * map,struct bpf_insn * insn_buf)1389 static int array_of_map_gen_lookup(struct bpf_map *map,
1390 				   struct bpf_insn *insn_buf)
1391 {
1392 	struct bpf_array *array = container_of(map, struct bpf_array, map);
1393 	u32 elem_size = array->elem_size;
1394 	struct bpf_insn *insn = insn_buf;
1395 	const int ret = BPF_REG_0;
1396 	const int map_ptr = BPF_REG_1;
1397 	const int index = BPF_REG_2;
1398 
1399 	*insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
1400 	*insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
1401 	if (!map->bypass_spec_v1) {
1402 		*insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 6);
1403 		*insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask);
1404 	} else {
1405 		*insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 5);
1406 	}
1407 	if (is_power_of_2(elem_size))
1408 		*insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
1409 	else
1410 		*insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
1411 	*insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
1412 	*insn++ = BPF_LDX_MEM(BPF_DW, ret, ret, 0);
1413 	*insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 1);
1414 	*insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
1415 	*insn++ = BPF_MOV64_IMM(ret, 0);
1416 
1417 	return insn - insn_buf;
1418 }
1419 
1420 const struct bpf_map_ops array_of_maps_map_ops = {
1421 	.map_alloc_check = fd_array_map_alloc_check,
1422 	.map_alloc = array_of_map_alloc,
1423 	.map_free = array_of_map_free,
1424 	.map_get_next_key = array_map_get_next_key,
1425 	.map_lookup_elem = array_of_map_lookup_elem,
1426 	.map_delete_elem = fd_array_map_delete_elem,
1427 	.map_fd_get_ptr = bpf_map_fd_get_ptr,
1428 	.map_fd_put_ptr = bpf_map_fd_put_ptr,
1429 	.map_fd_sys_lookup_elem = bpf_map_fd_sys_lookup_elem,
1430 	.map_gen_lookup = array_of_map_gen_lookup,
1431 	.map_lookup_batch = generic_map_lookup_batch,
1432 	.map_update_batch = generic_map_update_batch,
1433 	.map_check_btf = map_check_no_btf,
1434 	.map_mem_usage = array_map_mem_usage,
1435 	.map_btf_id = &array_map_btf_ids[0],
1436 };
1437