Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/* Copyright (c) 2022 Meta Platforms, Inc. and affiliates. */
3#include <linux/mm.h>
4#include <linux/llist.h>
5#include <linux/bpf.h>
6#include <linux/irq_work.h>
7#include <linux/bpf_mem_alloc.h>
8#include <linux/memcontrol.h>
9#include <asm/local.h>
10
11/* Any context (including NMI) BPF specific memory allocator.
12 *
13 * Tracing BPF programs can attach to kprobe and fentry. Hence they
14 * run in unknown context where calling plain kmalloc() might not be safe.
15 *
16 * Front-end kmalloc() with per-cpu per-bucket cache of free elements.
17 * Refill this cache asynchronously from irq_work.
18 *
19 * CPU_0 buckets
20 * 16 32 64 96 128 196 256 512 1024 2048 4096
21 * ...
22 * CPU_N buckets
23 * 16 32 64 96 128 196 256 512 1024 2048 4096
24 *
25 * The buckets are prefilled at the start.
26 * BPF programs always run with migration disabled.
27 * It's safe to allocate from cache of the current cpu with irqs disabled.
28 * Free-ing is always done into bucket of the current cpu as well.
29 * irq_work trims extra free elements from buckets with kfree
30 * and refills them with kmalloc, so global kmalloc logic takes care
31 * of freeing objects allocated by one cpu and freed on another.
32 *
33 * Every allocated objected is padded with extra 8 bytes that contains
34 * struct llist_node.
35 */
36#define LLIST_NODE_SZ sizeof(struct llist_node)
37
38/* similar to kmalloc, but sizeof == 8 bucket is gone */
39static u8 size_index[24] __ro_after_init = {
40 3, /* 8 */
41 3, /* 16 */
42 4, /* 24 */
43 4, /* 32 */
44 5, /* 40 */
45 5, /* 48 */
46 5, /* 56 */
47 5, /* 64 */
48 1, /* 72 */
49 1, /* 80 */
50 1, /* 88 */
51 1, /* 96 */
52 6, /* 104 */
53 6, /* 112 */
54 6, /* 120 */
55 6, /* 128 */
56 2, /* 136 */
57 2, /* 144 */
58 2, /* 152 */
59 2, /* 160 */
60 2, /* 168 */
61 2, /* 176 */
62 2, /* 184 */
63 2 /* 192 */
64};
65
66static int bpf_mem_cache_idx(size_t size)
67{
68 if (!size || size > 4096)
69 return -1;
70
71 if (size <= 192)
72 return size_index[(size - 1) / 8] - 1;
73
74 return fls(size - 1) - 2;
75}
76
77#define NUM_CACHES 11
78
79struct bpf_mem_cache {
80 /* per-cpu list of free objects of size 'unit_size'.
81 * All accesses are done with interrupts disabled and 'active' counter
82 * protection with __llist_add() and __llist_del_first().
83 */
84 struct llist_head free_llist;
85 local_t active;
86
87 /* Operations on the free_list from unit_alloc/unit_free/bpf_mem_refill
88 * are sequenced by per-cpu 'active' counter. But unit_free() cannot
89 * fail. When 'active' is busy the unit_free() will add an object to
90 * free_llist_extra.
91 */
92 struct llist_head free_llist_extra;
93
94 struct irq_work refill_work;
95 struct obj_cgroup *objcg;
96 int unit_size;
97 /* count of objects in free_llist */
98 int free_cnt;
99 int low_watermark, high_watermark, batch;
100 int percpu_size;
101
102 struct rcu_head rcu;
103 struct llist_head free_by_rcu;
104 struct llist_head waiting_for_gp;
105 atomic_t call_rcu_in_progress;
106};
107
108struct bpf_mem_caches {
109 struct bpf_mem_cache cache[NUM_CACHES];
110};
111
112static struct llist_node notrace *__llist_del_first(struct llist_head *head)
113{
114 struct llist_node *entry, *next;
115
116 entry = head->first;
117 if (!entry)
118 return NULL;
119 next = entry->next;
120 head->first = next;
121 return entry;
122}
123
124static void *__alloc(struct bpf_mem_cache *c, int node)
125{
126 /* Allocate, but don't deplete atomic reserves that typical
127 * GFP_ATOMIC would do. irq_work runs on this cpu and kmalloc
128 * will allocate from the current numa node which is what we
129 * want here.
130 */
131 gfp_t flags = GFP_NOWAIT | __GFP_NOWARN | __GFP_ACCOUNT;
132
133 if (c->percpu_size) {
134 void **obj = kmalloc_node(c->percpu_size, flags, node);
135 void *pptr = __alloc_percpu_gfp(c->unit_size, 8, flags);
136
137 if (!obj || !pptr) {
138 free_percpu(pptr);
139 kfree(obj);
140 return NULL;
141 }
142 obj[1] = pptr;
143 return obj;
144 }
145
146 return kmalloc_node(c->unit_size, flags, node);
147}
148
149static struct mem_cgroup *get_memcg(const struct bpf_mem_cache *c)
150{
151#ifdef CONFIG_MEMCG_KMEM
152 if (c->objcg)
153 return get_mem_cgroup_from_objcg(c->objcg);
154#endif
155
156#ifdef CONFIG_MEMCG
157 return root_mem_cgroup;
158#else
159 return NULL;
160#endif
161}
162
163/* Mostly runs from irq_work except __init phase. */
164static void alloc_bulk(struct bpf_mem_cache *c, int cnt, int node)
165{
166 struct mem_cgroup *memcg = NULL, *old_memcg;
167 unsigned long flags;
168 void *obj;
169 int i;
170
171 memcg = get_memcg(c);
172 old_memcg = set_active_memcg(memcg);
173 for (i = 0; i < cnt; i++) {
174 /*
175 * free_by_rcu is only manipulated by irq work refill_work().
176 * IRQ works on the same CPU are called sequentially, so it is
177 * safe to use __llist_del_first() here. If alloc_bulk() is
178 * invoked by the initial prefill, there will be no running
179 * refill_work(), so __llist_del_first() is fine as well.
180 *
181 * In most cases, objects on free_by_rcu are from the same CPU.
182 * If some objects come from other CPUs, it doesn't incur any
183 * harm because NUMA_NO_NODE means the preference for current
184 * numa node and it is not a guarantee.
185 */
186 obj = __llist_del_first(&c->free_by_rcu);
187 if (!obj) {
188 obj = __alloc(c, node);
189 if (!obj)
190 break;
191 }
192 if (IS_ENABLED(CONFIG_PREEMPT_RT))
193 /* In RT irq_work runs in per-cpu kthread, so disable
194 * interrupts to avoid preemption and interrupts and
195 * reduce the chance of bpf prog executing on this cpu
196 * when active counter is busy.
197 */
198 local_irq_save(flags);
199 /* alloc_bulk runs from irq_work which will not preempt a bpf
200 * program that does unit_alloc/unit_free since IRQs are
201 * disabled there. There is no race to increment 'active'
202 * counter. It protects free_llist from corruption in case NMI
203 * bpf prog preempted this loop.
204 */
205 WARN_ON_ONCE(local_inc_return(&c->active) != 1);
206 __llist_add(obj, &c->free_llist);
207 c->free_cnt++;
208 local_dec(&c->active);
209 if (IS_ENABLED(CONFIG_PREEMPT_RT))
210 local_irq_restore(flags);
211 }
212 set_active_memcg(old_memcg);
213 mem_cgroup_put(memcg);
214}
215
216static void free_one(struct bpf_mem_cache *c, void *obj)
217{
218 if (c->percpu_size) {
219 free_percpu(((void **)obj)[1]);
220 kfree(obj);
221 return;
222 }
223
224 kfree(obj);
225}
226
227static void __free_rcu(struct rcu_head *head)
228{
229 struct bpf_mem_cache *c = container_of(head, struct bpf_mem_cache, rcu);
230 struct llist_node *llnode = llist_del_all(&c->waiting_for_gp);
231 struct llist_node *pos, *t;
232
233 llist_for_each_safe(pos, t, llnode)
234 free_one(c, pos);
235 atomic_set(&c->call_rcu_in_progress, 0);
236}
237
238static void __free_rcu_tasks_trace(struct rcu_head *head)
239{
240 /* If RCU Tasks Trace grace period implies RCU grace period,
241 * there is no need to invoke call_rcu().
242 */
243 if (rcu_trace_implies_rcu_gp())
244 __free_rcu(head);
245 else
246 call_rcu(head, __free_rcu);
247}
248
249static void enque_to_free(struct bpf_mem_cache *c, void *obj)
250{
251 struct llist_node *llnode = obj;
252
253 /* bpf_mem_cache is a per-cpu object. Freeing happens in irq_work.
254 * Nothing races to add to free_by_rcu list.
255 */
256 __llist_add(llnode, &c->free_by_rcu);
257}
258
259static void do_call_rcu(struct bpf_mem_cache *c)
260{
261 struct llist_node *llnode, *t;
262
263 if (atomic_xchg(&c->call_rcu_in_progress, 1))
264 return;
265
266 WARN_ON_ONCE(!llist_empty(&c->waiting_for_gp));
267 llist_for_each_safe(llnode, t, __llist_del_all(&c->free_by_rcu))
268 /* There is no concurrent __llist_add(waiting_for_gp) access.
269 * It doesn't race with llist_del_all either.
270 * But there could be two concurrent llist_del_all(waiting_for_gp):
271 * from __free_rcu() and from drain_mem_cache().
272 */
273 __llist_add(llnode, &c->waiting_for_gp);
274 /* Use call_rcu_tasks_trace() to wait for sleepable progs to finish.
275 * If RCU Tasks Trace grace period implies RCU grace period, free
276 * these elements directly, else use call_rcu() to wait for normal
277 * progs to finish and finally do free_one() on each element.
278 */
279 call_rcu_tasks_trace(&c->rcu, __free_rcu_tasks_trace);
280}
281
282static void free_bulk(struct bpf_mem_cache *c)
283{
284 struct llist_node *llnode, *t;
285 unsigned long flags;
286 int cnt;
287
288 do {
289 if (IS_ENABLED(CONFIG_PREEMPT_RT))
290 local_irq_save(flags);
291 WARN_ON_ONCE(local_inc_return(&c->active) != 1);
292 llnode = __llist_del_first(&c->free_llist);
293 if (llnode)
294 cnt = --c->free_cnt;
295 else
296 cnt = 0;
297 local_dec(&c->active);
298 if (IS_ENABLED(CONFIG_PREEMPT_RT))
299 local_irq_restore(flags);
300 if (llnode)
301 enque_to_free(c, llnode);
302 } while (cnt > (c->high_watermark + c->low_watermark) / 2);
303
304 /* and drain free_llist_extra */
305 llist_for_each_safe(llnode, t, llist_del_all(&c->free_llist_extra))
306 enque_to_free(c, llnode);
307 do_call_rcu(c);
308}
309
310static void bpf_mem_refill(struct irq_work *work)
311{
312 struct bpf_mem_cache *c = container_of(work, struct bpf_mem_cache, refill_work);
313 int cnt;
314
315 /* Racy access to free_cnt. It doesn't need to be 100% accurate */
316 cnt = c->free_cnt;
317 if (cnt < c->low_watermark)
318 /* irq_work runs on this cpu and kmalloc will allocate
319 * from the current numa node which is what we want here.
320 */
321 alloc_bulk(c, c->batch, NUMA_NO_NODE);
322 else if (cnt > c->high_watermark)
323 free_bulk(c);
324}
325
326static void notrace irq_work_raise(struct bpf_mem_cache *c)
327{
328 irq_work_queue(&c->refill_work);
329}
330
331/* For typical bpf map case that uses bpf_mem_cache_alloc and single bucket
332 * the freelist cache will be elem_size * 64 (or less) on each cpu.
333 *
334 * For bpf programs that don't have statically known allocation sizes and
335 * assuming (low_mark + high_mark) / 2 as an average number of elements per
336 * bucket and all buckets are used the total amount of memory in freelists
337 * on each cpu will be:
338 * 64*16 + 64*32 + 64*64 + 64*96 + 64*128 + 64*196 + 64*256 + 32*512 + 16*1024 + 8*2048 + 4*4096
339 * == ~ 116 Kbyte using below heuristic.
340 * Initialized, but unused bpf allocator (not bpf map specific one) will
341 * consume ~ 11 Kbyte per cpu.
342 * Typical case will be between 11K and 116K closer to 11K.
343 * bpf progs can and should share bpf_mem_cache when possible.
344 */
345
346static void prefill_mem_cache(struct bpf_mem_cache *c, int cpu)
347{
348 init_irq_work(&c->refill_work, bpf_mem_refill);
349 if (c->unit_size <= 256) {
350 c->low_watermark = 32;
351 c->high_watermark = 96;
352 } else {
353 /* When page_size == 4k, order-0 cache will have low_mark == 2
354 * and high_mark == 6 with batch alloc of 3 individual pages at
355 * a time.
356 * 8k allocs and above low == 1, high == 3, batch == 1.
357 */
358 c->low_watermark = max(32 * 256 / c->unit_size, 1);
359 c->high_watermark = max(96 * 256 / c->unit_size, 3);
360 }
361 c->batch = max((c->high_watermark - c->low_watermark) / 4 * 3, 1);
362
363 /* To avoid consuming memory assume that 1st run of bpf
364 * prog won't be doing more than 4 map_update_elem from
365 * irq disabled region
366 */
367 alloc_bulk(c, c->unit_size <= 256 ? 4 : 1, cpu_to_node(cpu));
368}
369
370/* When size != 0 bpf_mem_cache for each cpu.
371 * This is typical bpf hash map use case when all elements have equal size.
372 *
373 * When size == 0 allocate 11 bpf_mem_cache-s for each cpu, then rely on
374 * kmalloc/kfree. Max allocation size is 4096 in this case.
375 * This is bpf_dynptr and bpf_kptr use case.
376 */
377int bpf_mem_alloc_init(struct bpf_mem_alloc *ma, int size, bool percpu)
378{
379 static u16 sizes[NUM_CACHES] = {96, 192, 16, 32, 64, 128, 256, 512, 1024, 2048, 4096};
380 struct bpf_mem_caches *cc, __percpu *pcc;
381 struct bpf_mem_cache *c, __percpu *pc;
382 struct obj_cgroup *objcg = NULL;
383 int cpu, i, unit_size, percpu_size = 0;
384
385 if (size) {
386 pc = __alloc_percpu_gfp(sizeof(*pc), 8, GFP_KERNEL);
387 if (!pc)
388 return -ENOMEM;
389
390 if (percpu)
391 /* room for llist_node and per-cpu pointer */
392 percpu_size = LLIST_NODE_SZ + sizeof(void *);
393 else
394 size += LLIST_NODE_SZ; /* room for llist_node */
395 unit_size = size;
396
397#ifdef CONFIG_MEMCG_KMEM
398 objcg = get_obj_cgroup_from_current();
399#endif
400 for_each_possible_cpu(cpu) {
401 c = per_cpu_ptr(pc, cpu);
402 c->unit_size = unit_size;
403 c->objcg = objcg;
404 c->percpu_size = percpu_size;
405 prefill_mem_cache(c, cpu);
406 }
407 ma->cache = pc;
408 return 0;
409 }
410
411 /* size == 0 && percpu is an invalid combination */
412 if (WARN_ON_ONCE(percpu))
413 return -EINVAL;
414
415 pcc = __alloc_percpu_gfp(sizeof(*cc), 8, GFP_KERNEL);
416 if (!pcc)
417 return -ENOMEM;
418#ifdef CONFIG_MEMCG_KMEM
419 objcg = get_obj_cgroup_from_current();
420#endif
421 for_each_possible_cpu(cpu) {
422 cc = per_cpu_ptr(pcc, cpu);
423 for (i = 0; i < NUM_CACHES; i++) {
424 c = &cc->cache[i];
425 c->unit_size = sizes[i];
426 c->objcg = objcg;
427 prefill_mem_cache(c, cpu);
428 }
429 }
430 ma->caches = pcc;
431 return 0;
432}
433
434static void drain_mem_cache(struct bpf_mem_cache *c)
435{
436 struct llist_node *llnode, *t;
437
438 /* No progs are using this bpf_mem_cache, but htab_map_free() called
439 * bpf_mem_cache_free() for all remaining elements and they can be in
440 * free_by_rcu or in waiting_for_gp lists, so drain those lists now.
441 *
442 * Except for waiting_for_gp list, there are no concurrent operations
443 * on these lists, so it is safe to use __llist_del_all().
444 */
445 llist_for_each_safe(llnode, t, __llist_del_all(&c->free_by_rcu))
446 free_one(c, llnode);
447 llist_for_each_safe(llnode, t, llist_del_all(&c->waiting_for_gp))
448 free_one(c, llnode);
449 llist_for_each_safe(llnode, t, __llist_del_all(&c->free_llist))
450 free_one(c, llnode);
451 llist_for_each_safe(llnode, t, __llist_del_all(&c->free_llist_extra))
452 free_one(c, llnode);
453}
454
455static void free_mem_alloc_no_barrier(struct bpf_mem_alloc *ma)
456{
457 free_percpu(ma->cache);
458 free_percpu(ma->caches);
459 ma->cache = NULL;
460 ma->caches = NULL;
461}
462
463static void free_mem_alloc(struct bpf_mem_alloc *ma)
464{
465 /* waiting_for_gp lists was drained, but __free_rcu might
466 * still execute. Wait for it now before we freeing percpu caches.
467 *
468 * rcu_barrier_tasks_trace() doesn't imply synchronize_rcu_tasks_trace(),
469 * but rcu_barrier_tasks_trace() and rcu_barrier() below are only used
470 * to wait for the pending __free_rcu_tasks_trace() and __free_rcu(),
471 * so if call_rcu(head, __free_rcu) is skipped due to
472 * rcu_trace_implies_rcu_gp(), it will be OK to skip rcu_barrier() by
473 * using rcu_trace_implies_rcu_gp() as well.
474 */
475 rcu_barrier_tasks_trace();
476 if (!rcu_trace_implies_rcu_gp())
477 rcu_barrier();
478 free_mem_alloc_no_barrier(ma);
479}
480
481static void free_mem_alloc_deferred(struct work_struct *work)
482{
483 struct bpf_mem_alloc *ma = container_of(work, struct bpf_mem_alloc, work);
484
485 free_mem_alloc(ma);
486 kfree(ma);
487}
488
489static void destroy_mem_alloc(struct bpf_mem_alloc *ma, int rcu_in_progress)
490{
491 struct bpf_mem_alloc *copy;
492
493 if (!rcu_in_progress) {
494 /* Fast path. No callbacks are pending, hence no need to do
495 * rcu_barrier-s.
496 */
497 free_mem_alloc_no_barrier(ma);
498 return;
499 }
500
501 copy = kmalloc(sizeof(*ma), GFP_KERNEL);
502 if (!copy) {
503 /* Slow path with inline barrier-s */
504 free_mem_alloc(ma);
505 return;
506 }
507
508 /* Defer barriers into worker to let the rest of map memory to be freed */
509 copy->cache = ma->cache;
510 ma->cache = NULL;
511 copy->caches = ma->caches;
512 ma->caches = NULL;
513 INIT_WORK(©->work, free_mem_alloc_deferred);
514 queue_work(system_unbound_wq, ©->work);
515}
516
517void bpf_mem_alloc_destroy(struct bpf_mem_alloc *ma)
518{
519 struct bpf_mem_caches *cc;
520 struct bpf_mem_cache *c;
521 int cpu, i, rcu_in_progress;
522
523 if (ma->cache) {
524 rcu_in_progress = 0;
525 for_each_possible_cpu(cpu) {
526 c = per_cpu_ptr(ma->cache, cpu);
527 /*
528 * refill_work may be unfinished for PREEMPT_RT kernel
529 * in which irq work is invoked in a per-CPU RT thread.
530 * It is also possible for kernel with
531 * arch_irq_work_has_interrupt() being false and irq
532 * work is invoked in timer interrupt. So waiting for
533 * the completion of irq work to ease the handling of
534 * concurrency.
535 */
536 irq_work_sync(&c->refill_work);
537 drain_mem_cache(c);
538 rcu_in_progress += atomic_read(&c->call_rcu_in_progress);
539 }
540 /* objcg is the same across cpus */
541 if (c->objcg)
542 obj_cgroup_put(c->objcg);
543 destroy_mem_alloc(ma, rcu_in_progress);
544 }
545 if (ma->caches) {
546 rcu_in_progress = 0;
547 for_each_possible_cpu(cpu) {
548 cc = per_cpu_ptr(ma->caches, cpu);
549 for (i = 0; i < NUM_CACHES; i++) {
550 c = &cc->cache[i];
551 irq_work_sync(&c->refill_work);
552 drain_mem_cache(c);
553 rcu_in_progress += atomic_read(&c->call_rcu_in_progress);
554 }
555 }
556 if (c->objcg)
557 obj_cgroup_put(c->objcg);
558 destroy_mem_alloc(ma, rcu_in_progress);
559 }
560}
561
562/* notrace is necessary here and in other functions to make sure
563 * bpf programs cannot attach to them and cause llist corruptions.
564 */
565static void notrace *unit_alloc(struct bpf_mem_cache *c)
566{
567 struct llist_node *llnode = NULL;
568 unsigned long flags;
569 int cnt = 0;
570
571 /* Disable irqs to prevent the following race for majority of prog types:
572 * prog_A
573 * bpf_mem_alloc
574 * preemption or irq -> prog_B
575 * bpf_mem_alloc
576 *
577 * but prog_B could be a perf_event NMI prog.
578 * Use per-cpu 'active' counter to order free_list access between
579 * unit_alloc/unit_free/bpf_mem_refill.
580 */
581 local_irq_save(flags);
582 if (local_inc_return(&c->active) == 1) {
583 llnode = __llist_del_first(&c->free_llist);
584 if (llnode)
585 cnt = --c->free_cnt;
586 }
587 local_dec(&c->active);
588 local_irq_restore(flags);
589
590 WARN_ON(cnt < 0);
591
592 if (cnt < c->low_watermark)
593 irq_work_raise(c);
594 return llnode;
595}
596
597/* Though 'ptr' object could have been allocated on a different cpu
598 * add it to the free_llist of the current cpu.
599 * Let kfree() logic deal with it when it's later called from irq_work.
600 */
601static void notrace unit_free(struct bpf_mem_cache *c, void *ptr)
602{
603 struct llist_node *llnode = ptr - LLIST_NODE_SZ;
604 unsigned long flags;
605 int cnt = 0;
606
607 BUILD_BUG_ON(LLIST_NODE_SZ > 8);
608
609 local_irq_save(flags);
610 if (local_inc_return(&c->active) == 1) {
611 __llist_add(llnode, &c->free_llist);
612 cnt = ++c->free_cnt;
613 } else {
614 /* unit_free() cannot fail. Therefore add an object to atomic
615 * llist. free_bulk() will drain it. Though free_llist_extra is
616 * a per-cpu list we have to use atomic llist_add here, since
617 * it also can be interrupted by bpf nmi prog that does another
618 * unit_free() into the same free_llist_extra.
619 */
620 llist_add(llnode, &c->free_llist_extra);
621 }
622 local_dec(&c->active);
623 local_irq_restore(flags);
624
625 if (cnt > c->high_watermark)
626 /* free few objects from current cpu into global kmalloc pool */
627 irq_work_raise(c);
628}
629
630/* Called from BPF program or from sys_bpf syscall.
631 * In both cases migration is disabled.
632 */
633void notrace *bpf_mem_alloc(struct bpf_mem_alloc *ma, size_t size)
634{
635 int idx;
636 void *ret;
637
638 if (!size)
639 return ZERO_SIZE_PTR;
640
641 idx = bpf_mem_cache_idx(size + LLIST_NODE_SZ);
642 if (idx < 0)
643 return NULL;
644
645 ret = unit_alloc(this_cpu_ptr(ma->caches)->cache + idx);
646 return !ret ? NULL : ret + LLIST_NODE_SZ;
647}
648
649void notrace bpf_mem_free(struct bpf_mem_alloc *ma, void *ptr)
650{
651 int idx;
652
653 if (!ptr)
654 return;
655
656 idx = bpf_mem_cache_idx(ksize(ptr - LLIST_NODE_SZ));
657 if (idx < 0)
658 return;
659
660 unit_free(this_cpu_ptr(ma->caches)->cache + idx, ptr);
661}
662
663void notrace *bpf_mem_cache_alloc(struct bpf_mem_alloc *ma)
664{
665 void *ret;
666
667 ret = unit_alloc(this_cpu_ptr(ma->cache));
668 return !ret ? NULL : ret + LLIST_NODE_SZ;
669}
670
671void notrace bpf_mem_cache_free(struct bpf_mem_alloc *ma, void *ptr)
672{
673 if (!ptr)
674 return;
675
676 unit_free(this_cpu_ptr(ma->cache), ptr);
677}
1// SPDX-License-Identifier: GPL-2.0-only
2/* Copyright (c) 2022 Meta Platforms, Inc. and affiliates. */
3#include <linux/mm.h>
4#include <linux/llist.h>
5#include <linux/bpf.h>
6#include <linux/irq_work.h>
7#include <linux/bpf_mem_alloc.h>
8#include <linux/memcontrol.h>
9#include <asm/local.h>
10
11/* Any context (including NMI) BPF specific memory allocator.
12 *
13 * Tracing BPF programs can attach to kprobe and fentry. Hence they
14 * run in unknown context where calling plain kmalloc() might not be safe.
15 *
16 * Front-end kmalloc() with per-cpu per-bucket cache of free elements.
17 * Refill this cache asynchronously from irq_work.
18 *
19 * CPU_0 buckets
20 * 16 32 64 96 128 196 256 512 1024 2048 4096
21 * ...
22 * CPU_N buckets
23 * 16 32 64 96 128 196 256 512 1024 2048 4096
24 *
25 * The buckets are prefilled at the start.
26 * BPF programs always run with migration disabled.
27 * It's safe to allocate from cache of the current cpu with irqs disabled.
28 * Free-ing is always done into bucket of the current cpu as well.
29 * irq_work trims extra free elements from buckets with kfree
30 * and refills them with kmalloc, so global kmalloc logic takes care
31 * of freeing objects allocated by one cpu and freed on another.
32 *
33 * Every allocated objected is padded with extra 8 bytes that contains
34 * struct llist_node.
35 */
36#define LLIST_NODE_SZ sizeof(struct llist_node)
37
38/* similar to kmalloc, but sizeof == 8 bucket is gone */
39static u8 size_index[24] __ro_after_init = {
40 3, /* 8 */
41 3, /* 16 */
42 4, /* 24 */
43 4, /* 32 */
44 5, /* 40 */
45 5, /* 48 */
46 5, /* 56 */
47 5, /* 64 */
48 1, /* 72 */
49 1, /* 80 */
50 1, /* 88 */
51 1, /* 96 */
52 6, /* 104 */
53 6, /* 112 */
54 6, /* 120 */
55 6, /* 128 */
56 2, /* 136 */
57 2, /* 144 */
58 2, /* 152 */
59 2, /* 160 */
60 2, /* 168 */
61 2, /* 176 */
62 2, /* 184 */
63 2 /* 192 */
64};
65
66static int bpf_mem_cache_idx(size_t size)
67{
68 if (!size || size > 4096)
69 return -1;
70
71 if (size <= 192)
72 return size_index[(size - 1) / 8] - 1;
73
74 return fls(size - 1) - 2;
75}
76
77#define NUM_CACHES 11
78
79struct bpf_mem_cache {
80 /* per-cpu list of free objects of size 'unit_size'.
81 * All accesses are done with interrupts disabled and 'active' counter
82 * protection with __llist_add() and __llist_del_first().
83 */
84 struct llist_head free_llist;
85 local_t active;
86
87 /* Operations on the free_list from unit_alloc/unit_free/bpf_mem_refill
88 * are sequenced by per-cpu 'active' counter. But unit_free() cannot
89 * fail. When 'active' is busy the unit_free() will add an object to
90 * free_llist_extra.
91 */
92 struct llist_head free_llist_extra;
93
94 struct irq_work refill_work;
95 struct obj_cgroup *objcg;
96 int unit_size;
97 /* count of objects in free_llist */
98 int free_cnt;
99 int low_watermark, high_watermark, batch;
100 int percpu_size;
101 bool draining;
102 struct bpf_mem_cache *tgt;
103
104 /* list of objects to be freed after RCU GP */
105 struct llist_head free_by_rcu;
106 struct llist_node *free_by_rcu_tail;
107 struct llist_head waiting_for_gp;
108 struct llist_node *waiting_for_gp_tail;
109 struct rcu_head rcu;
110 atomic_t call_rcu_in_progress;
111 struct llist_head free_llist_extra_rcu;
112
113 /* list of objects to be freed after RCU tasks trace GP */
114 struct llist_head free_by_rcu_ttrace;
115 struct llist_head waiting_for_gp_ttrace;
116 struct rcu_head rcu_ttrace;
117 atomic_t call_rcu_ttrace_in_progress;
118};
119
120struct bpf_mem_caches {
121 struct bpf_mem_cache cache[NUM_CACHES];
122};
123
124static const u16 sizes[NUM_CACHES] = {96, 192, 16, 32, 64, 128, 256, 512, 1024, 2048, 4096};
125
126static struct llist_node notrace *__llist_del_first(struct llist_head *head)
127{
128 struct llist_node *entry, *next;
129
130 entry = head->first;
131 if (!entry)
132 return NULL;
133 next = entry->next;
134 head->first = next;
135 return entry;
136}
137
138static void *__alloc(struct bpf_mem_cache *c, int node, gfp_t flags)
139{
140 if (c->percpu_size) {
141 void **obj = kmalloc_node(c->percpu_size, flags, node);
142 void *pptr = __alloc_percpu_gfp(c->unit_size, 8, flags);
143
144 if (!obj || !pptr) {
145 free_percpu(pptr);
146 kfree(obj);
147 return NULL;
148 }
149 obj[1] = pptr;
150 return obj;
151 }
152
153 return kmalloc_node(c->unit_size, flags | __GFP_ZERO, node);
154}
155
156static struct mem_cgroup *get_memcg(const struct bpf_mem_cache *c)
157{
158#ifdef CONFIG_MEMCG_KMEM
159 if (c->objcg)
160 return get_mem_cgroup_from_objcg(c->objcg);
161#endif
162
163#ifdef CONFIG_MEMCG
164 return root_mem_cgroup;
165#else
166 return NULL;
167#endif
168}
169
170static void inc_active(struct bpf_mem_cache *c, unsigned long *flags)
171{
172 if (IS_ENABLED(CONFIG_PREEMPT_RT))
173 /* In RT irq_work runs in per-cpu kthread, so disable
174 * interrupts to avoid preemption and interrupts and
175 * reduce the chance of bpf prog executing on this cpu
176 * when active counter is busy.
177 */
178 local_irq_save(*flags);
179 /* alloc_bulk runs from irq_work which will not preempt a bpf
180 * program that does unit_alloc/unit_free since IRQs are
181 * disabled there. There is no race to increment 'active'
182 * counter. It protects free_llist from corruption in case NMI
183 * bpf prog preempted this loop.
184 */
185 WARN_ON_ONCE(local_inc_return(&c->active) != 1);
186}
187
188static void dec_active(struct bpf_mem_cache *c, unsigned long *flags)
189{
190 local_dec(&c->active);
191 if (IS_ENABLED(CONFIG_PREEMPT_RT))
192 local_irq_restore(*flags);
193}
194
195static void add_obj_to_free_list(struct bpf_mem_cache *c, void *obj)
196{
197 unsigned long flags;
198
199 inc_active(c, &flags);
200 __llist_add(obj, &c->free_llist);
201 c->free_cnt++;
202 dec_active(c, &flags);
203}
204
205/* Mostly runs from irq_work except __init phase. */
206static void alloc_bulk(struct bpf_mem_cache *c, int cnt, int node, bool atomic)
207{
208 struct mem_cgroup *memcg = NULL, *old_memcg;
209 gfp_t gfp;
210 void *obj;
211 int i;
212
213 gfp = __GFP_NOWARN | __GFP_ACCOUNT;
214 gfp |= atomic ? GFP_NOWAIT : GFP_KERNEL;
215
216 for (i = 0; i < cnt; i++) {
217 /*
218 * For every 'c' llist_del_first(&c->free_by_rcu_ttrace); is
219 * done only by one CPU == current CPU. Other CPUs might
220 * llist_add() and llist_del_all() in parallel.
221 */
222 obj = llist_del_first(&c->free_by_rcu_ttrace);
223 if (!obj)
224 break;
225 add_obj_to_free_list(c, obj);
226 }
227 if (i >= cnt)
228 return;
229
230 for (; i < cnt; i++) {
231 obj = llist_del_first(&c->waiting_for_gp_ttrace);
232 if (!obj)
233 break;
234 add_obj_to_free_list(c, obj);
235 }
236 if (i >= cnt)
237 return;
238
239 memcg = get_memcg(c);
240 old_memcg = set_active_memcg(memcg);
241 for (; i < cnt; i++) {
242 /* Allocate, but don't deplete atomic reserves that typical
243 * GFP_ATOMIC would do. irq_work runs on this cpu and kmalloc
244 * will allocate from the current numa node which is what we
245 * want here.
246 */
247 obj = __alloc(c, node, gfp);
248 if (!obj)
249 break;
250 add_obj_to_free_list(c, obj);
251 }
252 set_active_memcg(old_memcg);
253 mem_cgroup_put(memcg);
254}
255
256static void free_one(void *obj, bool percpu)
257{
258 if (percpu) {
259 free_percpu(((void **)obj)[1]);
260 kfree(obj);
261 return;
262 }
263
264 kfree(obj);
265}
266
267static int free_all(struct llist_node *llnode, bool percpu)
268{
269 struct llist_node *pos, *t;
270 int cnt = 0;
271
272 llist_for_each_safe(pos, t, llnode) {
273 free_one(pos, percpu);
274 cnt++;
275 }
276 return cnt;
277}
278
279static void __free_rcu(struct rcu_head *head)
280{
281 struct bpf_mem_cache *c = container_of(head, struct bpf_mem_cache, rcu_ttrace);
282
283 free_all(llist_del_all(&c->waiting_for_gp_ttrace), !!c->percpu_size);
284 atomic_set(&c->call_rcu_ttrace_in_progress, 0);
285}
286
287static void __free_rcu_tasks_trace(struct rcu_head *head)
288{
289 /* If RCU Tasks Trace grace period implies RCU grace period,
290 * there is no need to invoke call_rcu().
291 */
292 if (rcu_trace_implies_rcu_gp())
293 __free_rcu(head);
294 else
295 call_rcu(head, __free_rcu);
296}
297
298static void enque_to_free(struct bpf_mem_cache *c, void *obj)
299{
300 struct llist_node *llnode = obj;
301
302 /* bpf_mem_cache is a per-cpu object. Freeing happens in irq_work.
303 * Nothing races to add to free_by_rcu_ttrace list.
304 */
305 llist_add(llnode, &c->free_by_rcu_ttrace);
306}
307
308static void do_call_rcu_ttrace(struct bpf_mem_cache *c)
309{
310 struct llist_node *llnode, *t;
311
312 if (atomic_xchg(&c->call_rcu_ttrace_in_progress, 1)) {
313 if (unlikely(READ_ONCE(c->draining))) {
314 llnode = llist_del_all(&c->free_by_rcu_ttrace);
315 free_all(llnode, !!c->percpu_size);
316 }
317 return;
318 }
319
320 WARN_ON_ONCE(!llist_empty(&c->waiting_for_gp_ttrace));
321 llist_for_each_safe(llnode, t, llist_del_all(&c->free_by_rcu_ttrace))
322 llist_add(llnode, &c->waiting_for_gp_ttrace);
323
324 if (unlikely(READ_ONCE(c->draining))) {
325 __free_rcu(&c->rcu_ttrace);
326 return;
327 }
328
329 /* Use call_rcu_tasks_trace() to wait for sleepable progs to finish.
330 * If RCU Tasks Trace grace period implies RCU grace period, free
331 * these elements directly, else use call_rcu() to wait for normal
332 * progs to finish and finally do free_one() on each element.
333 */
334 call_rcu_tasks_trace(&c->rcu_ttrace, __free_rcu_tasks_trace);
335}
336
337static void free_bulk(struct bpf_mem_cache *c)
338{
339 struct bpf_mem_cache *tgt = c->tgt;
340 struct llist_node *llnode, *t;
341 unsigned long flags;
342 int cnt;
343
344 WARN_ON_ONCE(tgt->unit_size != c->unit_size);
345 WARN_ON_ONCE(tgt->percpu_size != c->percpu_size);
346
347 do {
348 inc_active(c, &flags);
349 llnode = __llist_del_first(&c->free_llist);
350 if (llnode)
351 cnt = --c->free_cnt;
352 else
353 cnt = 0;
354 dec_active(c, &flags);
355 if (llnode)
356 enque_to_free(tgt, llnode);
357 } while (cnt > (c->high_watermark + c->low_watermark) / 2);
358
359 /* and drain free_llist_extra */
360 llist_for_each_safe(llnode, t, llist_del_all(&c->free_llist_extra))
361 enque_to_free(tgt, llnode);
362 do_call_rcu_ttrace(tgt);
363}
364
365static void __free_by_rcu(struct rcu_head *head)
366{
367 struct bpf_mem_cache *c = container_of(head, struct bpf_mem_cache, rcu);
368 struct bpf_mem_cache *tgt = c->tgt;
369 struct llist_node *llnode;
370
371 WARN_ON_ONCE(tgt->unit_size != c->unit_size);
372 WARN_ON_ONCE(tgt->percpu_size != c->percpu_size);
373
374 llnode = llist_del_all(&c->waiting_for_gp);
375 if (!llnode)
376 goto out;
377
378 llist_add_batch(llnode, c->waiting_for_gp_tail, &tgt->free_by_rcu_ttrace);
379
380 /* Objects went through regular RCU GP. Send them to RCU tasks trace */
381 do_call_rcu_ttrace(tgt);
382out:
383 atomic_set(&c->call_rcu_in_progress, 0);
384}
385
386static void check_free_by_rcu(struct bpf_mem_cache *c)
387{
388 struct llist_node *llnode, *t;
389 unsigned long flags;
390
391 /* drain free_llist_extra_rcu */
392 if (unlikely(!llist_empty(&c->free_llist_extra_rcu))) {
393 inc_active(c, &flags);
394 llist_for_each_safe(llnode, t, llist_del_all(&c->free_llist_extra_rcu))
395 if (__llist_add(llnode, &c->free_by_rcu))
396 c->free_by_rcu_tail = llnode;
397 dec_active(c, &flags);
398 }
399
400 if (llist_empty(&c->free_by_rcu))
401 return;
402
403 if (atomic_xchg(&c->call_rcu_in_progress, 1)) {
404 /*
405 * Instead of kmalloc-ing new rcu_head and triggering 10k
406 * call_rcu() to hit rcutree.qhimark and force RCU to notice
407 * the overload just ask RCU to hurry up. There could be many
408 * objects in free_by_rcu list.
409 * This hint reduces memory consumption for an artificial
410 * benchmark from 2 Gbyte to 150 Mbyte.
411 */
412 rcu_request_urgent_qs_task(current);
413 return;
414 }
415
416 WARN_ON_ONCE(!llist_empty(&c->waiting_for_gp));
417
418 inc_active(c, &flags);
419 WRITE_ONCE(c->waiting_for_gp.first, __llist_del_all(&c->free_by_rcu));
420 c->waiting_for_gp_tail = c->free_by_rcu_tail;
421 dec_active(c, &flags);
422
423 if (unlikely(READ_ONCE(c->draining))) {
424 free_all(llist_del_all(&c->waiting_for_gp), !!c->percpu_size);
425 atomic_set(&c->call_rcu_in_progress, 0);
426 } else {
427 call_rcu_hurry(&c->rcu, __free_by_rcu);
428 }
429}
430
431static void bpf_mem_refill(struct irq_work *work)
432{
433 struct bpf_mem_cache *c = container_of(work, struct bpf_mem_cache, refill_work);
434 int cnt;
435
436 /* Racy access to free_cnt. It doesn't need to be 100% accurate */
437 cnt = c->free_cnt;
438 if (cnt < c->low_watermark)
439 /* irq_work runs on this cpu and kmalloc will allocate
440 * from the current numa node which is what we want here.
441 */
442 alloc_bulk(c, c->batch, NUMA_NO_NODE, true);
443 else if (cnt > c->high_watermark)
444 free_bulk(c);
445
446 check_free_by_rcu(c);
447}
448
449static void notrace irq_work_raise(struct bpf_mem_cache *c)
450{
451 irq_work_queue(&c->refill_work);
452}
453
454/* For typical bpf map case that uses bpf_mem_cache_alloc and single bucket
455 * the freelist cache will be elem_size * 64 (or less) on each cpu.
456 *
457 * For bpf programs that don't have statically known allocation sizes and
458 * assuming (low_mark + high_mark) / 2 as an average number of elements per
459 * bucket and all buckets are used the total amount of memory in freelists
460 * on each cpu will be:
461 * 64*16 + 64*32 + 64*64 + 64*96 + 64*128 + 64*196 + 64*256 + 32*512 + 16*1024 + 8*2048 + 4*4096
462 * == ~ 116 Kbyte using below heuristic.
463 * Initialized, but unused bpf allocator (not bpf map specific one) will
464 * consume ~ 11 Kbyte per cpu.
465 * Typical case will be between 11K and 116K closer to 11K.
466 * bpf progs can and should share bpf_mem_cache when possible.
467 *
468 * Percpu allocation is typically rare. To avoid potential unnecessary large
469 * memory consumption, set low_mark = 1 and high_mark = 3, resulting in c->batch = 1.
470 */
471static void init_refill_work(struct bpf_mem_cache *c)
472{
473 init_irq_work(&c->refill_work, bpf_mem_refill);
474 if (c->percpu_size) {
475 c->low_watermark = 1;
476 c->high_watermark = 3;
477 } else if (c->unit_size <= 256) {
478 c->low_watermark = 32;
479 c->high_watermark = 96;
480 } else {
481 /* When page_size == 4k, order-0 cache will have low_mark == 2
482 * and high_mark == 6 with batch alloc of 3 individual pages at
483 * a time.
484 * 8k allocs and above low == 1, high == 3, batch == 1.
485 */
486 c->low_watermark = max(32 * 256 / c->unit_size, 1);
487 c->high_watermark = max(96 * 256 / c->unit_size, 3);
488 }
489 c->batch = max((c->high_watermark - c->low_watermark) / 4 * 3, 1);
490}
491
492static void prefill_mem_cache(struct bpf_mem_cache *c, int cpu)
493{
494 int cnt = 1;
495
496 /* To avoid consuming memory, for non-percpu allocation, assume that
497 * 1st run of bpf prog won't be doing more than 4 map_update_elem from
498 * irq disabled region if unit size is less than or equal to 256.
499 * For all other cases, let us just do one allocation.
500 */
501 if (!c->percpu_size && c->unit_size <= 256)
502 cnt = 4;
503 alloc_bulk(c, cnt, cpu_to_node(cpu), false);
504}
505
506/* When size != 0 bpf_mem_cache for each cpu.
507 * This is typical bpf hash map use case when all elements have equal size.
508 *
509 * When size == 0 allocate 11 bpf_mem_cache-s for each cpu, then rely on
510 * kmalloc/kfree. Max allocation size is 4096 in this case.
511 * This is bpf_dynptr and bpf_kptr use case.
512 */
513int bpf_mem_alloc_init(struct bpf_mem_alloc *ma, int size, bool percpu)
514{
515 struct bpf_mem_caches *cc, __percpu *pcc;
516 struct bpf_mem_cache *c, __percpu *pc;
517 struct obj_cgroup *objcg = NULL;
518 int cpu, i, unit_size, percpu_size = 0;
519
520 if (percpu && size == 0)
521 return -EINVAL;
522
523 /* room for llist_node and per-cpu pointer */
524 if (percpu)
525 percpu_size = LLIST_NODE_SZ + sizeof(void *);
526 ma->percpu = percpu;
527
528 if (size) {
529 pc = __alloc_percpu_gfp(sizeof(*pc), 8, GFP_KERNEL);
530 if (!pc)
531 return -ENOMEM;
532
533 if (!percpu)
534 size += LLIST_NODE_SZ; /* room for llist_node */
535 unit_size = size;
536
537#ifdef CONFIG_MEMCG_KMEM
538 if (memcg_bpf_enabled())
539 objcg = get_obj_cgroup_from_current();
540#endif
541 ma->objcg = objcg;
542
543 for_each_possible_cpu(cpu) {
544 c = per_cpu_ptr(pc, cpu);
545 c->unit_size = unit_size;
546 c->objcg = objcg;
547 c->percpu_size = percpu_size;
548 c->tgt = c;
549 init_refill_work(c);
550 prefill_mem_cache(c, cpu);
551 }
552 ma->cache = pc;
553 return 0;
554 }
555
556 pcc = __alloc_percpu_gfp(sizeof(*cc), 8, GFP_KERNEL);
557 if (!pcc)
558 return -ENOMEM;
559#ifdef CONFIG_MEMCG_KMEM
560 objcg = get_obj_cgroup_from_current();
561#endif
562 ma->objcg = objcg;
563 for_each_possible_cpu(cpu) {
564 cc = per_cpu_ptr(pcc, cpu);
565 for (i = 0; i < NUM_CACHES; i++) {
566 c = &cc->cache[i];
567 c->unit_size = sizes[i];
568 c->objcg = objcg;
569 c->percpu_size = percpu_size;
570 c->tgt = c;
571
572 init_refill_work(c);
573 prefill_mem_cache(c, cpu);
574 }
575 }
576
577 ma->caches = pcc;
578 return 0;
579}
580
581int bpf_mem_alloc_percpu_init(struct bpf_mem_alloc *ma, struct obj_cgroup *objcg)
582{
583 struct bpf_mem_caches __percpu *pcc;
584
585 pcc = __alloc_percpu_gfp(sizeof(struct bpf_mem_caches), 8, GFP_KERNEL);
586 if (!pcc)
587 return -ENOMEM;
588
589 ma->caches = pcc;
590 ma->objcg = objcg;
591 ma->percpu = true;
592 return 0;
593}
594
595int bpf_mem_alloc_percpu_unit_init(struct bpf_mem_alloc *ma, int size)
596{
597 struct bpf_mem_caches *cc, __percpu *pcc;
598 int cpu, i, unit_size, percpu_size;
599 struct obj_cgroup *objcg;
600 struct bpf_mem_cache *c;
601
602 i = bpf_mem_cache_idx(size);
603 if (i < 0)
604 return -EINVAL;
605
606 /* room for llist_node and per-cpu pointer */
607 percpu_size = LLIST_NODE_SZ + sizeof(void *);
608
609 unit_size = sizes[i];
610 objcg = ma->objcg;
611 pcc = ma->caches;
612
613 for_each_possible_cpu(cpu) {
614 cc = per_cpu_ptr(pcc, cpu);
615 c = &cc->cache[i];
616 if (c->unit_size)
617 break;
618
619 c->unit_size = unit_size;
620 c->objcg = objcg;
621 c->percpu_size = percpu_size;
622 c->tgt = c;
623
624 init_refill_work(c);
625 prefill_mem_cache(c, cpu);
626 }
627
628 return 0;
629}
630
631static void drain_mem_cache(struct bpf_mem_cache *c)
632{
633 bool percpu = !!c->percpu_size;
634
635 /* No progs are using this bpf_mem_cache, but htab_map_free() called
636 * bpf_mem_cache_free() for all remaining elements and they can be in
637 * free_by_rcu_ttrace or in waiting_for_gp_ttrace lists, so drain those lists now.
638 *
639 * Except for waiting_for_gp_ttrace list, there are no concurrent operations
640 * on these lists, so it is safe to use __llist_del_all().
641 */
642 free_all(llist_del_all(&c->free_by_rcu_ttrace), percpu);
643 free_all(llist_del_all(&c->waiting_for_gp_ttrace), percpu);
644 free_all(__llist_del_all(&c->free_llist), percpu);
645 free_all(__llist_del_all(&c->free_llist_extra), percpu);
646 free_all(__llist_del_all(&c->free_by_rcu), percpu);
647 free_all(__llist_del_all(&c->free_llist_extra_rcu), percpu);
648 free_all(llist_del_all(&c->waiting_for_gp), percpu);
649}
650
651static void check_mem_cache(struct bpf_mem_cache *c)
652{
653 WARN_ON_ONCE(!llist_empty(&c->free_by_rcu_ttrace));
654 WARN_ON_ONCE(!llist_empty(&c->waiting_for_gp_ttrace));
655 WARN_ON_ONCE(!llist_empty(&c->free_llist));
656 WARN_ON_ONCE(!llist_empty(&c->free_llist_extra));
657 WARN_ON_ONCE(!llist_empty(&c->free_by_rcu));
658 WARN_ON_ONCE(!llist_empty(&c->free_llist_extra_rcu));
659 WARN_ON_ONCE(!llist_empty(&c->waiting_for_gp));
660}
661
662static void check_leaked_objs(struct bpf_mem_alloc *ma)
663{
664 struct bpf_mem_caches *cc;
665 struct bpf_mem_cache *c;
666 int cpu, i;
667
668 if (ma->cache) {
669 for_each_possible_cpu(cpu) {
670 c = per_cpu_ptr(ma->cache, cpu);
671 check_mem_cache(c);
672 }
673 }
674 if (ma->caches) {
675 for_each_possible_cpu(cpu) {
676 cc = per_cpu_ptr(ma->caches, cpu);
677 for (i = 0; i < NUM_CACHES; i++) {
678 c = &cc->cache[i];
679 check_mem_cache(c);
680 }
681 }
682 }
683}
684
685static void free_mem_alloc_no_barrier(struct bpf_mem_alloc *ma)
686{
687 check_leaked_objs(ma);
688 free_percpu(ma->cache);
689 free_percpu(ma->caches);
690 ma->cache = NULL;
691 ma->caches = NULL;
692}
693
694static void free_mem_alloc(struct bpf_mem_alloc *ma)
695{
696 /* waiting_for_gp[_ttrace] lists were drained, but RCU callbacks
697 * might still execute. Wait for them.
698 *
699 * rcu_barrier_tasks_trace() doesn't imply synchronize_rcu_tasks_trace(),
700 * but rcu_barrier_tasks_trace() and rcu_barrier() below are only used
701 * to wait for the pending __free_rcu_tasks_trace() and __free_rcu(),
702 * so if call_rcu(head, __free_rcu) is skipped due to
703 * rcu_trace_implies_rcu_gp(), it will be OK to skip rcu_barrier() by
704 * using rcu_trace_implies_rcu_gp() as well.
705 */
706 rcu_barrier(); /* wait for __free_by_rcu */
707 rcu_barrier_tasks_trace(); /* wait for __free_rcu */
708 if (!rcu_trace_implies_rcu_gp())
709 rcu_barrier();
710 free_mem_alloc_no_barrier(ma);
711}
712
713static void free_mem_alloc_deferred(struct work_struct *work)
714{
715 struct bpf_mem_alloc *ma = container_of(work, struct bpf_mem_alloc, work);
716
717 free_mem_alloc(ma);
718 kfree(ma);
719}
720
721static void destroy_mem_alloc(struct bpf_mem_alloc *ma, int rcu_in_progress)
722{
723 struct bpf_mem_alloc *copy;
724
725 if (!rcu_in_progress) {
726 /* Fast path. No callbacks are pending, hence no need to do
727 * rcu_barrier-s.
728 */
729 free_mem_alloc_no_barrier(ma);
730 return;
731 }
732
733 copy = kmemdup(ma, sizeof(*ma), GFP_KERNEL);
734 if (!copy) {
735 /* Slow path with inline barrier-s */
736 free_mem_alloc(ma);
737 return;
738 }
739
740 /* Defer barriers into worker to let the rest of map memory to be freed */
741 memset(ma, 0, sizeof(*ma));
742 INIT_WORK(©->work, free_mem_alloc_deferred);
743 queue_work(system_unbound_wq, ©->work);
744}
745
746void bpf_mem_alloc_destroy(struct bpf_mem_alloc *ma)
747{
748 struct bpf_mem_caches *cc;
749 struct bpf_mem_cache *c;
750 int cpu, i, rcu_in_progress;
751
752 if (ma->cache) {
753 rcu_in_progress = 0;
754 for_each_possible_cpu(cpu) {
755 c = per_cpu_ptr(ma->cache, cpu);
756 WRITE_ONCE(c->draining, true);
757 irq_work_sync(&c->refill_work);
758 drain_mem_cache(c);
759 rcu_in_progress += atomic_read(&c->call_rcu_ttrace_in_progress);
760 rcu_in_progress += atomic_read(&c->call_rcu_in_progress);
761 }
762 if (ma->objcg)
763 obj_cgroup_put(ma->objcg);
764 destroy_mem_alloc(ma, rcu_in_progress);
765 }
766 if (ma->caches) {
767 rcu_in_progress = 0;
768 for_each_possible_cpu(cpu) {
769 cc = per_cpu_ptr(ma->caches, cpu);
770 for (i = 0; i < NUM_CACHES; i++) {
771 c = &cc->cache[i];
772 WRITE_ONCE(c->draining, true);
773 irq_work_sync(&c->refill_work);
774 drain_mem_cache(c);
775 rcu_in_progress += atomic_read(&c->call_rcu_ttrace_in_progress);
776 rcu_in_progress += atomic_read(&c->call_rcu_in_progress);
777 }
778 }
779 if (ma->objcg)
780 obj_cgroup_put(ma->objcg);
781 destroy_mem_alloc(ma, rcu_in_progress);
782 }
783}
784
785/* notrace is necessary here and in other functions to make sure
786 * bpf programs cannot attach to them and cause llist corruptions.
787 */
788static void notrace *unit_alloc(struct bpf_mem_cache *c)
789{
790 struct llist_node *llnode = NULL;
791 unsigned long flags;
792 int cnt = 0;
793
794 /* Disable irqs to prevent the following race for majority of prog types:
795 * prog_A
796 * bpf_mem_alloc
797 * preemption or irq -> prog_B
798 * bpf_mem_alloc
799 *
800 * but prog_B could be a perf_event NMI prog.
801 * Use per-cpu 'active' counter to order free_list access between
802 * unit_alloc/unit_free/bpf_mem_refill.
803 */
804 local_irq_save(flags);
805 if (local_inc_return(&c->active) == 1) {
806 llnode = __llist_del_first(&c->free_llist);
807 if (llnode) {
808 cnt = --c->free_cnt;
809 *(struct bpf_mem_cache **)llnode = c;
810 }
811 }
812 local_dec(&c->active);
813
814 WARN_ON(cnt < 0);
815
816 if (cnt < c->low_watermark)
817 irq_work_raise(c);
818 /* Enable IRQ after the enqueue of irq work completes, so irq work
819 * will run after IRQ is enabled and free_llist may be refilled by
820 * irq work before other task preempts current task.
821 */
822 local_irq_restore(flags);
823
824 return llnode;
825}
826
827/* Though 'ptr' object could have been allocated on a different cpu
828 * add it to the free_llist of the current cpu.
829 * Let kfree() logic deal with it when it's later called from irq_work.
830 */
831static void notrace unit_free(struct bpf_mem_cache *c, void *ptr)
832{
833 struct llist_node *llnode = ptr - LLIST_NODE_SZ;
834 unsigned long flags;
835 int cnt = 0;
836
837 BUILD_BUG_ON(LLIST_NODE_SZ > 8);
838
839 /*
840 * Remember bpf_mem_cache that allocated this object.
841 * The hint is not accurate.
842 */
843 c->tgt = *(struct bpf_mem_cache **)llnode;
844
845 local_irq_save(flags);
846 if (local_inc_return(&c->active) == 1) {
847 __llist_add(llnode, &c->free_llist);
848 cnt = ++c->free_cnt;
849 } else {
850 /* unit_free() cannot fail. Therefore add an object to atomic
851 * llist. free_bulk() will drain it. Though free_llist_extra is
852 * a per-cpu list we have to use atomic llist_add here, since
853 * it also can be interrupted by bpf nmi prog that does another
854 * unit_free() into the same free_llist_extra.
855 */
856 llist_add(llnode, &c->free_llist_extra);
857 }
858 local_dec(&c->active);
859
860 if (cnt > c->high_watermark)
861 /* free few objects from current cpu into global kmalloc pool */
862 irq_work_raise(c);
863 /* Enable IRQ after irq_work_raise() completes, otherwise when current
864 * task is preempted by task which does unit_alloc(), unit_alloc() may
865 * return NULL unexpectedly because irq work is already pending but can
866 * not been triggered and free_llist can not be refilled timely.
867 */
868 local_irq_restore(flags);
869}
870
871static void notrace unit_free_rcu(struct bpf_mem_cache *c, void *ptr)
872{
873 struct llist_node *llnode = ptr - LLIST_NODE_SZ;
874 unsigned long flags;
875
876 c->tgt = *(struct bpf_mem_cache **)llnode;
877
878 local_irq_save(flags);
879 if (local_inc_return(&c->active) == 1) {
880 if (__llist_add(llnode, &c->free_by_rcu))
881 c->free_by_rcu_tail = llnode;
882 } else {
883 llist_add(llnode, &c->free_llist_extra_rcu);
884 }
885 local_dec(&c->active);
886
887 if (!atomic_read(&c->call_rcu_in_progress))
888 irq_work_raise(c);
889 local_irq_restore(flags);
890}
891
892/* Called from BPF program or from sys_bpf syscall.
893 * In both cases migration is disabled.
894 */
895void notrace *bpf_mem_alloc(struct bpf_mem_alloc *ma, size_t size)
896{
897 int idx;
898 void *ret;
899
900 if (!size)
901 return NULL;
902
903 if (!ma->percpu)
904 size += LLIST_NODE_SZ;
905 idx = bpf_mem_cache_idx(size);
906 if (idx < 0)
907 return NULL;
908
909 ret = unit_alloc(this_cpu_ptr(ma->caches)->cache + idx);
910 return !ret ? NULL : ret + LLIST_NODE_SZ;
911}
912
913void notrace bpf_mem_free(struct bpf_mem_alloc *ma, void *ptr)
914{
915 struct bpf_mem_cache *c;
916 int idx;
917
918 if (!ptr)
919 return;
920
921 c = *(void **)(ptr - LLIST_NODE_SZ);
922 idx = bpf_mem_cache_idx(c->unit_size);
923 if (WARN_ON_ONCE(idx < 0))
924 return;
925
926 unit_free(this_cpu_ptr(ma->caches)->cache + idx, ptr);
927}
928
929void notrace bpf_mem_free_rcu(struct bpf_mem_alloc *ma, void *ptr)
930{
931 struct bpf_mem_cache *c;
932 int idx;
933
934 if (!ptr)
935 return;
936
937 c = *(void **)(ptr - LLIST_NODE_SZ);
938 idx = bpf_mem_cache_idx(c->unit_size);
939 if (WARN_ON_ONCE(idx < 0))
940 return;
941
942 unit_free_rcu(this_cpu_ptr(ma->caches)->cache + idx, ptr);
943}
944
945void notrace *bpf_mem_cache_alloc(struct bpf_mem_alloc *ma)
946{
947 void *ret;
948
949 ret = unit_alloc(this_cpu_ptr(ma->cache));
950 return !ret ? NULL : ret + LLIST_NODE_SZ;
951}
952
953void notrace bpf_mem_cache_free(struct bpf_mem_alloc *ma, void *ptr)
954{
955 if (!ptr)
956 return;
957
958 unit_free(this_cpu_ptr(ma->cache), ptr);
959}
960
961void notrace bpf_mem_cache_free_rcu(struct bpf_mem_alloc *ma, void *ptr)
962{
963 if (!ptr)
964 return;
965
966 unit_free_rcu(this_cpu_ptr(ma->cache), ptr);
967}
968
969/* Directly does a kfree() without putting 'ptr' back to the free_llist
970 * for reuse and without waiting for a rcu_tasks_trace gp.
971 * The caller must first go through the rcu_tasks_trace gp for 'ptr'
972 * before calling bpf_mem_cache_raw_free().
973 * It could be used when the rcu_tasks_trace callback does not have
974 * a hold on the original bpf_mem_alloc object that allocated the
975 * 'ptr'. This should only be used in the uncommon code path.
976 * Otherwise, the bpf_mem_alloc's free_llist cannot be refilled
977 * and may affect performance.
978 */
979void bpf_mem_cache_raw_free(void *ptr)
980{
981 if (!ptr)
982 return;
983
984 kfree(ptr - LLIST_NODE_SZ);
985}
986
987/* When flags == GFP_KERNEL, it signals that the caller will not cause
988 * deadlock when using kmalloc. bpf_mem_cache_alloc_flags() will use
989 * kmalloc if the free_llist is empty.
990 */
991void notrace *bpf_mem_cache_alloc_flags(struct bpf_mem_alloc *ma, gfp_t flags)
992{
993 struct bpf_mem_cache *c;
994 void *ret;
995
996 c = this_cpu_ptr(ma->cache);
997
998 ret = unit_alloc(c);
999 if (!ret && flags == GFP_KERNEL) {
1000 struct mem_cgroup *memcg, *old_memcg;
1001
1002 memcg = get_memcg(c);
1003 old_memcg = set_active_memcg(memcg);
1004 ret = __alloc(c, NUMA_NO_NODE, GFP_KERNEL | __GFP_NOWARN | __GFP_ACCOUNT);
1005 if (ret)
1006 *(struct bpf_mem_cache **)ret = c;
1007 set_active_memcg(old_memcg);
1008 mem_cgroup_put(memcg);
1009 }
1010
1011 return !ret ? NULL : ret + LLIST_NODE_SZ;
1012}