kernel/bpf/stackmap.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /* Copyright (c) 2016 Facebook
   3  */
   4 #include <linux/bpf.h>
   5 #include <linux/jhash.h>
   6 #include <linux/filter.h>
   7 #include <linux/kernel.h>
   8 #include <linux/stacktrace.h>
   9 #include <linux/perf_event.h>
  10 #include <linux/btf_ids.h>
  11 #include <linux/buildid.h>
  12 #include "percpu_freelist.h"
  13 #include "mmap_unlock_work.h"
  14
  15 #define STACK_CREATE_FLAG_MASK                                  \
  16         (BPF_F_NUMA_NODE | BPF_F_RDONLY | BPF_F_WRONLY |        \
  17          BPF_F_STACK_BUILD_ID)
  18
  19 struct stack_map_bucket {
  20         struct pcpu_freelist_node fnode;
  21         u32 hash;
  22         u32 nr;
  23         u64 data[];
  24 };
  25
  26 struct bpf_stack_map {
  27         struct bpf_map map;
  28         void *elems;
  29         struct pcpu_freelist freelist;
  30         u32 n_buckets;
  31         struct stack_map_bucket *buckets[];
  32 };
  33
  34 static inline bool stack_map_use_build_id(struct bpf_map *map)
  35 {
  36         return (map->map_flags & BPF_F_STACK_BUILD_ID);
  37 }
  38
  39 static inline int stack_map_data_size(struct bpf_map *map)
  40 {
  41         return stack_map_use_build_id(map) ?
  42                 sizeof(struct bpf_stack_build_id) : sizeof(u64);
  43 }
  44
  45 static int prealloc_elems_and_freelist(struct bpf_stack_map *smap)
  46 {
  47         u64 elem_size = sizeof(struct stack_map_bucket) +
  48                         (u64)smap->map.value_size;
  49         int err;
  50
  51         smap->elems = bpf_map_area_alloc(elem_size * smap->map.max_entries,
  52                                          smap->map.numa_node);
  53         if (!smap->elems)
  54                 return -ENOMEM;
  55
  56         err = pcpu_freelist_init(&smap->freelist);
  57         if (err)
  58                 goto free_elems;
  59
  60         pcpu_freelist_populate(&smap->freelist, smap->elems, elem_size,
  61                                smap->map.max_entries);
  62         return 0;
  63
  64 free_elems:
  65         bpf_map_area_free(smap->elems);
  66         return err;
  67 }
  68
  69 /* Called from syscall */
  70 static struct bpf_map *stack_map_alloc(union bpf_attr *attr)
  71 {
  72         u32 value_size = attr->value_size;
  73         struct bpf_stack_map *smap;
  74         u64 cost, n_buckets;
  75         int err;
  76
  77         if (!bpf_capable())
  78                 return ERR_PTR(-EPERM);
  79
  80         if (attr->map_flags & ~STACK_CREATE_FLAG_MASK)
  81                 return ERR_PTR(-EINVAL);
  82
  83         /* check sanity of attributes */
  84         if (attr->max_entries == 0 || attr->key_size != 4 ||
  85             value_size < 8 || value_size % 8)
  86                 return ERR_PTR(-EINVAL);
  87
  88         BUILD_BUG_ON(sizeof(struct bpf_stack_build_id) % sizeof(u64));
  89         if (attr->map_flags & BPF_F_STACK_BUILD_ID) {
  90                 if (value_size % sizeof(struct bpf_stack_build_id) ||
  91                     value_size / sizeof(struct bpf_stack_build_id)
  92                     > sysctl_perf_event_max_stack)
  93                         return ERR_PTR(-EINVAL);
  94         } else if (value_size / 8 > sysctl_perf_event_max_stack)
  95                 return ERR_PTR(-EINVAL);
  96
  97         /* hash table size must be power of 2 */
  98         n_buckets = roundup_pow_of_two(attr->max_entries);
  99         if (!n_buckets)
 100                 return ERR_PTR(-E2BIG);
 101
 102         cost = n_buckets * sizeof(struct stack_map_bucket *) + sizeof(*smap);
 103         cost += n_buckets * (value_size + sizeof(struct stack_map_bucket));
 104         smap = bpf_map_area_alloc(cost, bpf_map_attr_numa_node(attr));
 105         if (!smap)
 106                 return ERR_PTR(-ENOMEM);
 107
 108         bpf_map_init_from_attr(&smap->map, attr);
 109         smap->map.value_size = value_size;
 110         smap->n_buckets = n_buckets;
 111
 112         err = get_callchain_buffers(sysctl_perf_event_max_stack);
 113         if (err)
 114                 goto free_smap;
 115
 116         err = prealloc_elems_and_freelist(smap);
 117         if (err)
 118                 goto put_buffers;
 119
 120         return &smap->map;
 121
 122 put_buffers:
 123         put_callchain_buffers();
 124 free_smap:
 125         bpf_map_area_free(smap);
 126         return ERR_PTR(err);
 127 }
 128
 129 static void stack_map_get_build_id_offset(struct bpf_stack_build_id *id_offs,
 130                                           u64 *ips, u32 trace_nr, bool user)
 131 {
 132         int i;
 133         struct mmap_unlock_irq_work *work = NULL;
 134         bool irq_work_busy = bpf_mmap_unlock_get_irq_work(&work);
 135         struct vm_area_struct *vma;
 136
 137         /* If the irq_work is in use, fall back to report ips. Same
 138          * fallback is used for kernel stack (!user) on a stackmap with
 139          * build_id.
 140          */
 141         if (!user || !current || !current->mm || irq_work_busy ||
 142             !mmap_read_trylock(current->mm)) {
 143                 /* cannot access current->mm, fall back to ips */
 144                 for (i = 0; i < trace_nr; i++) {
 145                         id_offs[i].status = BPF_STACK_BUILD_ID_IP;
 146                         id_offs[i].ip = ips[i];
 147                         memset(id_offs[i].build_id, 0, BUILD_ID_SIZE_MAX);
 148                 }
 149                 return;
 150         }
 151
 152         for (i = 0; i < trace_nr; i++) {
 153                 vma = find_vma(current->mm, ips[i]);
 154                 if (!vma || build_id_parse(vma, id_offs[i].build_id, NULL)) {
 155                         /* per entry fall back to ips */
 156                         id_offs[i].status = BPF_STACK_BUILD_ID_IP;
 157                         id_offs[i].ip = ips[i];
 158                         memset(id_offs[i].build_id, 0, BUILD_ID_SIZE_MAX);
 159                         continue;
 160                 }
 161                 id_offs[i].offset = (vma->vm_pgoff << PAGE_SHIFT) + ips[i]
 162                         - vma->vm_start;
 163                 id_offs[i].status = BPF_STACK_BUILD_ID_VALID;
 164         }
 165         bpf_mmap_unlock_mm(work, current->mm);
 166 }
 167
 168 static struct perf_callchain_entry *
 169 get_callchain_entry_for_task(struct task_struct *task, u32 max_depth)
 170 {
 171 #ifdef CONFIG_STACKTRACE
 172         struct perf_callchain_entry *entry;
 173         int rctx;
 174
 175         entry = get_callchain_entry(&rctx);
 176
 177         if (!entry)
 178                 return NULL;
 179
 180         entry->nr = stack_trace_save_tsk(task, (unsigned long *)entry->ip,
 181                                          max_depth, 0);
 182
 183         /* stack_trace_save_tsk() works on unsigned long array, while
 184          * perf_callchain_entry uses u64 array. For 32-bit systems, it is
 185          * necessary to fix this mismatch.
 186          */
 187         if (__BITS_PER_LONG != 64) {
 188                 unsigned long *from = (unsigned long *) entry->ip;
 189                 u64 *to = entry->ip;
 190                 int i;
 191
 192                 /* copy data from the end to avoid using extra buffer */
 193                 for (i = entry->nr - 1; i >= 0; i--)
 194                         to[i] = (u64)(from[i]);
 195         }
 196
 197         put_callchain_entry(rctx);
 198
 199         return entry;
 200 #else /* CONFIG_STACKTRACE */
 201         return NULL;
 202 #endif
 203 }
 204
 205 static long __bpf_get_stackid(struct bpf_map *map,
 206                               struct perf_callchain_entry *trace, u64 flags)
 207 {
 208         struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
 209         struct stack_map_bucket *bucket, *new_bucket, *old_bucket;
 210         u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
 211         u32 hash, id, trace_nr, trace_len;
 212         bool user = flags & BPF_F_USER_STACK;
 213         u64 *ips;
 214         bool hash_matches;
 215
 216         if (trace->nr <= skip)
 217                 /* skipping more than usable stack trace */
 218                 return -EFAULT;
 219
 220         trace_nr = trace->nr - skip;
 221         trace_len = trace_nr * sizeof(u64);
 222         ips = trace->ip + skip;
 223         hash = jhash2((u32 *)ips, trace_len / sizeof(u32), 0);
 224         id = hash & (smap->n_buckets - 1);
 225         bucket = READ_ONCE(smap->buckets[id]);
 226
 227         hash_matches = bucket && bucket->hash == hash;
 228         /* fast cmp */
 229         if (hash_matches && flags & BPF_F_FAST_STACK_CMP)
 230                 return id;
 231
 232         if (stack_map_use_build_id(map)) {
 233                 /* for build_id+offset, pop a bucket before slow cmp */
 234                 new_bucket = (struct stack_map_bucket *)
 235                         pcpu_freelist_pop(&smap->freelist);
 236                 if (unlikely(!new_bucket))
 237                         return -ENOMEM;
 238                 new_bucket->nr = trace_nr;
 239                 stack_map_get_build_id_offset(
 240                         (struct bpf_stack_build_id *)new_bucket->data,
 241                         ips, trace_nr, user);
 242                 trace_len = trace_nr * sizeof(struct bpf_stack_build_id);
 243                 if (hash_matches && bucket->nr == trace_nr &&
 244                     memcmp(bucket->data, new_bucket->data, trace_len) == 0) {
 245                         pcpu_freelist_push(&smap->freelist, &new_bucket->fnode);
 246                         return id;
 247                 }
 248                 if (bucket && !(flags & BPF_F_REUSE_STACKID)) {
 249                         pcpu_freelist_push(&smap->freelist, &new_bucket->fnode);
 250                         return -EEXIST;
 251                 }
 252         } else {
 253                 if (hash_matches && bucket->nr == trace_nr &&
 254                     memcmp(bucket->data, ips, trace_len) == 0)
 255                         return id;
 256                 if (bucket && !(flags & BPF_F_REUSE_STACKID))
 257                         return -EEXIST;
 258
 259                 new_bucket = (struct stack_map_bucket *)
 260                         pcpu_freelist_pop(&smap->freelist);
 261                 if (unlikely(!new_bucket))
 262                         return -ENOMEM;
 263                 memcpy(new_bucket->data, ips, trace_len);
 264         }
 265
 266         new_bucket->hash = hash;
 267         new_bucket->nr = trace_nr;
 268
 269         old_bucket = xchg(&smap->buckets[id], new_bucket);
 270         if (old_bucket)
 271                 pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
 272         return id;
 273 }
 274
 275 BPF_CALL_3(bpf_get_stackid, struct pt_regs *, regs, struct bpf_map *, map,
 276            u64, flags)
 277 {
 278         u32 max_depth = map->value_size / stack_map_data_size(map);
 279         u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
 280         bool user = flags & BPF_F_USER_STACK;
 281         struct perf_callchain_entry *trace;
 282         bool kernel = !user;
 283
 284         if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
 285                                BPF_F_FAST_STACK_CMP | BPF_F_REUSE_STACKID)))
 286                 return -EINVAL;
 287
 288         max_depth += skip;
 289         if (max_depth > sysctl_perf_event_max_stack)
 290                 max_depth = sysctl_perf_event_max_stack;
 291
 292         trace = get_perf_callchain(regs, 0, kernel, user, max_depth,
 293                                    false, false);
 294
 295         if (unlikely(!trace))
 296                 /* couldn't fetch the stack trace */
 297                 return -EFAULT;
 298
 299         return __bpf_get_stackid(map, trace, flags);
 300 }
 301
 302 const struct bpf_func_proto bpf_get_stackid_proto = {
 303         .func           = bpf_get_stackid,
 304         .gpl_only       = true,
 305         .ret_type       = RET_INTEGER,
 306         .arg1_type      = ARG_PTR_TO_CTX,
 307         .arg2_type      = ARG_CONST_MAP_PTR,
 308         .arg3_type      = ARG_ANYTHING,
 309 };
 310
 311 static __u64 count_kernel_ip(struct perf_callchain_entry *trace)
 312 {
 313         __u64 nr_kernel = 0;
 314
 315         while (nr_kernel < trace->nr) {
 316                 if (trace->ip[nr_kernel] == PERF_CONTEXT_USER)
 317                         break;
 318                 nr_kernel++;
 319         }
 320         return nr_kernel;
 321 }
 322
 323 BPF_CALL_3(bpf_get_stackid_pe, struct bpf_perf_event_data_kern *, ctx,
 324            struct bpf_map *, map, u64, flags)
 325 {
 326         struct perf_event *event = ctx->event;
 327         struct perf_callchain_entry *trace;
 328         bool kernel, user;
 329         __u64 nr_kernel;
 330         int ret;
 331
 332         /* perf_sample_data doesn't have callchain, use bpf_get_stackid */
 333         if (!(event->attr.sample_type & __PERF_SAMPLE_CALLCHAIN_EARLY))
 334                 return bpf_get_stackid((unsigned long)(ctx->regs),
 335                                        (unsigned long) map, flags, 0, 0);
 336
 337         if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
 338                                BPF_F_FAST_STACK_CMP | BPF_F_REUSE_STACKID)))
 339                 return -EINVAL;
 340
 341         user = flags & BPF_F_USER_STACK;
 342         kernel = !user;
 343
 344         trace = ctx->data->callchain;
 345         if (unlikely(!trace))
 346                 return -EFAULT;
 347
 348         nr_kernel = count_kernel_ip(trace);
 349
 350         if (kernel) {
 351                 __u64 nr = trace->nr;
 352
 353                 trace->nr = nr_kernel;
 354                 ret = __bpf_get_stackid(map, trace, flags);
 355
 356                 /* restore nr */
 357                 trace->nr = nr;
 358         } else { /* user */
 359                 u64 skip = flags & BPF_F_SKIP_FIELD_MASK;
 360
 361                 skip += nr_kernel;
 362                 if (skip > BPF_F_SKIP_FIELD_MASK)
 363                         return -EFAULT;
 364
 365                 flags = (flags & ~BPF_F_SKIP_FIELD_MASK) | skip;
 366                 ret = __bpf_get_stackid(map, trace, flags);
 367         }
 368         return ret;
 369 }
 370
 371 const struct bpf_func_proto bpf_get_stackid_proto_pe = {
 372         .func           = bpf_get_stackid_pe,
 373         .gpl_only       = false,
 374         .ret_type       = RET_INTEGER,
 375         .arg1_type      = ARG_PTR_TO_CTX,
 376         .arg2_type      = ARG_CONST_MAP_PTR,
 377         .arg3_type      = ARG_ANYTHING,
 378 };
 379
 380 static long __bpf_get_stack(struct pt_regs *regs, struct task_struct *task,
 381                             struct perf_callchain_entry *trace_in,
 382                             void *buf, u32 size, u64 flags)
 383 {
 384         u32 trace_nr, copy_len, elem_size, num_elem, max_depth;
 385         bool user_build_id = flags & BPF_F_USER_BUILD_ID;
 386         u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
 387         bool user = flags & BPF_F_USER_STACK;
 388         struct perf_callchain_entry *trace;
 389         bool kernel = !user;
 390         int err = -EINVAL;
 391         u64 *ips;
 392
 393         if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
 394                                BPF_F_USER_BUILD_ID)))
 395                 goto clear;
 396         if (kernel && user_build_id)
 397                 goto clear;
 398
 399         elem_size = (user && user_build_id) ? sizeof(struct bpf_stack_build_id)
 400                                             : sizeof(u64);
 401         if (unlikely(size % elem_size))
 402                 goto clear;
 403
 404         /* cannot get valid user stack for task without user_mode regs */
 405         if (task && user && !user_mode(regs))
 406                 goto err_fault;
 407
 408         num_elem = size / elem_size;
 409         max_depth = num_elem + skip;
 410         if (sysctl_perf_event_max_stack < max_depth)
 411                 max_depth = sysctl_perf_event_max_stack;
 412
 413         if (trace_in)
 414                 trace = trace_in;
 415         else if (kernel && task)
 416                 trace = get_callchain_entry_for_task(task, max_depth);
 417         else
 418                 trace = get_perf_callchain(regs, 0, kernel, user, max_depth,
 419                                            false, false);
 420         if (unlikely(!trace))
 421                 goto err_fault;
 422
 423         if (trace->nr < skip)
 424                 goto err_fault;
 425
 426         trace_nr = trace->nr - skip;
 427         trace_nr = (trace_nr <= num_elem) ? trace_nr : num_elem;
 428         copy_len = trace_nr * elem_size;
 429
 430         ips = trace->ip + skip;
 431         if (user && user_build_id)
 432                 stack_map_get_build_id_offset(buf, ips, trace_nr, user);
 433         else
 434                 memcpy(buf, ips, copy_len);
 435
 436         if (size > copy_len)
 437                 memset(buf + copy_len, 0, size - copy_len);
 438         return copy_len;
 439
 440 err_fault:
 441         err = -EFAULT;
 442 clear:
 443         memset(buf, 0, size);
 444         return err;
 445 }
 446
 447 BPF_CALL_4(bpf_get_stack, struct pt_regs *, regs, void *, buf, u32, size,
 448            u64, flags)
 449 {
 450         return __bpf_get_stack(regs, NULL, NULL, buf, size, flags);
 451 }
 452
 453 const struct bpf_func_proto bpf_get_stack_proto = {
 454         .func           = bpf_get_stack,
 455         .gpl_only       = true,
 456         .ret_type       = RET_INTEGER,
 457         .arg1_type      = ARG_PTR_TO_CTX,
 458         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
 459         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
 460         .arg4_type      = ARG_ANYTHING,
 461 };
 462
 463 BPF_CALL_4(bpf_get_task_stack, struct task_struct *, task, void *, buf,
 464            u32, size, u64, flags)
 465 {
 466         struct pt_regs *regs;
 467         long res = -EINVAL;
 468
 469         if (!try_get_task_stack(task))
 470                 return -EFAULT;
 471
 472         regs = task_pt_regs(task);
 473         if (regs)
 474                 res = __bpf_get_stack(regs, task, NULL, buf, size, flags);
 475         put_task_stack(task);
 476
 477         return res;
 478 }
 479
 480 const struct bpf_func_proto bpf_get_task_stack_proto = {
 481         .func           = bpf_get_task_stack,
 482         .gpl_only       = false,
 483         .ret_type       = RET_INTEGER,
 484         .arg1_type      = ARG_PTR_TO_BTF_ID,
 485         .arg1_btf_id    = &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
 486         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
 487         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
 488         .arg4_type      = ARG_ANYTHING,
 489 };
 490
 491 BPF_CALL_4(bpf_get_stack_pe, struct bpf_perf_event_data_kern *, ctx,
 492            void *, buf, u32, size, u64, flags)
 493 {
 494         struct pt_regs *regs = (struct pt_regs *)(ctx->regs);
 495         struct perf_event *event = ctx->event;
 496         struct perf_callchain_entry *trace;
 497         bool kernel, user;
 498         int err = -EINVAL;
 499         __u64 nr_kernel;
 500
 501         if (!(event->attr.sample_type & __PERF_SAMPLE_CALLCHAIN_EARLY))
 502                 return __bpf_get_stack(regs, NULL, NULL, buf, size, flags);
 503
 504         if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
 505                                BPF_F_USER_BUILD_ID)))
 506                 goto clear;
 507
 508         user = flags & BPF_F_USER_STACK;
 509         kernel = !user;
 510
 511         err = -EFAULT;
 512         trace = ctx->data->callchain;
 513         if (unlikely(!trace))
 514                 goto clear;
 515
 516         nr_kernel = count_kernel_ip(trace);
 517
 518         if (kernel) {
 519                 __u64 nr = trace->nr;
 520
 521                 trace->nr = nr_kernel;
 522                 err = __bpf_get_stack(regs, NULL, trace, buf, size, flags);
 523
 524                 /* restore nr */
 525                 trace->nr = nr;
 526         } else { /* user */
 527                 u64 skip = flags & BPF_F_SKIP_FIELD_MASK;
 528
 529                 skip += nr_kernel;
 530                 if (skip > BPF_F_SKIP_FIELD_MASK)
 531                         goto clear;
 532
 533                 flags = (flags & ~BPF_F_SKIP_FIELD_MASK) | skip;
 534                 err = __bpf_get_stack(regs, NULL, trace, buf, size, flags);
 535         }
 536         return err;
 537
 538 clear:
 539         memset(buf, 0, size);
 540         return err;
 541
 542 }
 543
 544 const struct bpf_func_proto bpf_get_stack_proto_pe = {
 545         .func           = bpf_get_stack_pe,
 546         .gpl_only       = true,
 547         .ret_type       = RET_INTEGER,
 548         .arg1_type      = ARG_PTR_TO_CTX,
 549         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
 550         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
 551         .arg4_type      = ARG_ANYTHING,
 552 };
 553
 554 /* Called from eBPF program */
 555 static void *stack_map_lookup_elem(struct bpf_map *map, void *key)
 556 {
 557         return ERR_PTR(-EOPNOTSUPP);
 558 }
 559
 560 /* Called from syscall */
 561 int bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
 562 {
 563         struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
 564         struct stack_map_bucket *bucket, *old_bucket;
 565         u32 id = *(u32 *)key, trace_len;
 566
 567         if (unlikely(id >= smap->n_buckets))
 568                 return -ENOENT;
 569
 570         bucket = xchg(&smap->buckets[id], NULL);
 571         if (!bucket)
 572                 return -ENOENT;
 573
 574         trace_len = bucket->nr * stack_map_data_size(map);
 575         memcpy(value, bucket->data, trace_len);
 576         memset(value + trace_len, 0, map->value_size - trace_len);
 577
 578         old_bucket = xchg(&smap->buckets[id], bucket);
 579         if (old_bucket)
 580                 pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
 581         return 0;
 582 }
 583
 584 static int stack_map_get_next_key(struct bpf_map *map, void *key,
 585                                   void *next_key)
 586 {
 587         struct bpf_stack_map *smap = container_of(map,
 588                                                   struct bpf_stack_map, map);
 589         u32 id;
 590
 591         WARN_ON_ONCE(!rcu_read_lock_held());
 592
 593         if (!key) {
 594                 id = 0;
 595         } else {
 596                 id = *(u32 *)key;
 597                 if (id >= smap->n_buckets || !smap->buckets[id])
 598                         id = 0;
 599                 else
 600                         id++;
 601         }
 602
 603         while (id < smap->n_buckets && !smap->buckets[id])
 604                 id++;
 605
 606         if (id >= smap->n_buckets)
 607                 return -ENOENT;
 608
 609         *(u32 *)next_key = id;
 610         return 0;
 611 }
 612
 613 static int stack_map_update_elem(struct bpf_map *map, void *key, void *value,
 614                                  u64 map_flags)
 615 {
 616         return -EINVAL;
 617 }
 618
 619 /* Called from syscall or from eBPF program */
 620 static int stack_map_delete_elem(struct bpf_map *map, void *key)
 621 {
 622         struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
 623         struct stack_map_bucket *old_bucket;
 624         u32 id = *(u32 *)key;
 625
 626         if (unlikely(id >= smap->n_buckets))
 627                 return -E2BIG;
 628
 629         old_bucket = xchg(&smap->buckets[id], NULL);
 630         if (old_bucket) {
 631                 pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
 632                 return 0;
 633         } else {
 634                 return -ENOENT;
 635         }
 636 }
 637
 638 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
 639 static void stack_map_free(struct bpf_map *map)
 640 {
 641         struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
 642
 643         bpf_map_area_free(smap->elems);
 644         pcpu_freelist_destroy(&smap->freelist);
 645         bpf_map_area_free(smap);
 646         put_callchain_buffers();
 647 }
 648
 649 static int stack_trace_map_btf_id;
 650 const struct bpf_map_ops stack_trace_map_ops = {
 651         .map_meta_equal = bpf_map_meta_equal,
 652         .map_alloc = stack_map_alloc,
 653         .map_free = stack_map_free,
 654         .map_get_next_key = stack_map_get_next_key,
 655         .map_lookup_elem = stack_map_lookup_elem,
 656         .map_update_elem = stack_map_update_elem,
 657         .map_delete_elem = stack_map_delete_elem,
 658         .map_check_btf = map_check_no_btf,
 659         .map_btf_name = "bpf_stack_map",
 660         .map_btf_id = &stack_trace_map_btf_id,
 661 };