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 init_nr)
 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 = init_nr +
 181                 stack_trace_save_tsk(task, (unsigned long *)(entry->ip + init_nr),
 182                                      sysctl_perf_event_max_stack - init_nr, 0);
 183
 184         /* stack_trace_save_tsk() works on unsigned long array, while
 185          * perf_callchain_entry uses u64 array. For 32-bit systems, it is
 186          * necessary to fix this mismatch.
 187          */
 188         if (__BITS_PER_LONG != 64) {
 189                 unsigned long *from = (unsigned long *) entry->ip;
 190                 u64 *to = entry->ip;
 191                 int i;
 192
 193                 /* copy data from the end to avoid using extra buffer */
 194                 for (i = entry->nr - 1; i >= (int)init_nr; i--)
 195                         to[i] = (u64)(from[i]);
 196         }
 197
 198         put_callchain_entry(rctx);
 199
 200         return entry;
 201 #else /* CONFIG_STACKTRACE */
 202         return NULL;
 203 #endif
 204 }
 205
 206 static long __bpf_get_stackid(struct bpf_map *map,
 207                               struct perf_callchain_entry *trace, u64 flags)
 208 {
 209         struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
 210         struct stack_map_bucket *bucket, *new_bucket, *old_bucket;
 211         u32 max_depth = map->value_size / stack_map_data_size(map);
 212         /* stack_map_alloc() checks that max_depth <= sysctl_perf_event_max_stack */
 213         u32 init_nr = sysctl_perf_event_max_stack - max_depth;
 214         u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
 215         u32 hash, id, trace_nr, trace_len;
 216         bool user = flags & BPF_F_USER_STACK;
 217         u64 *ips;
 218         bool hash_matches;
 219
 220         /* get_perf_callchain() guarantees that trace->nr >= init_nr
 221          * and trace-nr <= sysctl_perf_event_max_stack, so trace_nr <= max_depth
 222          */
 223         trace_nr = trace->nr - init_nr;
 224
 225         if (trace_nr <= skip)
 226                 /* skipping more than usable stack trace */
 227                 return -EFAULT;
 228
 229         trace_nr -= skip;
 230         trace_len = trace_nr * sizeof(u64);
 231         ips = trace->ip + skip + init_nr;
 232         hash = jhash2((u32 *)ips, trace_len / sizeof(u32), 0);
 233         id = hash & (smap->n_buckets - 1);
 234         bucket = READ_ONCE(smap->buckets[id]);
 235
 236         hash_matches = bucket && bucket->hash == hash;
 237         /* fast cmp */
 238         if (hash_matches && flags & BPF_F_FAST_STACK_CMP)
 239                 return id;
 240
 241         if (stack_map_use_build_id(map)) {
 242                 /* for build_id+offset, pop a bucket before slow cmp */
 243                 new_bucket = (struct stack_map_bucket *)
 244                         pcpu_freelist_pop(&smap->freelist);
 245                 if (unlikely(!new_bucket))
 246                         return -ENOMEM;
 247                 new_bucket->nr = trace_nr;
 248                 stack_map_get_build_id_offset(
 249                         (struct bpf_stack_build_id *)new_bucket->data,
 250                         ips, trace_nr, user);
 251                 trace_len = trace_nr * sizeof(struct bpf_stack_build_id);
 252                 if (hash_matches && bucket->nr == trace_nr &&
 253                     memcmp(bucket->data, new_bucket->data, trace_len) == 0) {
 254                         pcpu_freelist_push(&smap->freelist, &new_bucket->fnode);
 255                         return id;
 256                 }
 257                 if (bucket && !(flags & BPF_F_REUSE_STACKID)) {
 258                         pcpu_freelist_push(&smap->freelist, &new_bucket->fnode);
 259                         return -EEXIST;
 260                 }
 261         } else {
 262                 if (hash_matches && bucket->nr == trace_nr &&
 263                     memcmp(bucket->data, ips, trace_len) == 0)
 264                         return id;
 265                 if (bucket && !(flags & BPF_F_REUSE_STACKID))
 266                         return -EEXIST;
 267
 268                 new_bucket = (struct stack_map_bucket *)
 269                         pcpu_freelist_pop(&smap->freelist);
 270                 if (unlikely(!new_bucket))
 271                         return -ENOMEM;
 272                 memcpy(new_bucket->data, ips, trace_len);
 273         }
 274
 275         new_bucket->hash = hash;
 276         new_bucket->nr = trace_nr;
 277
 278         old_bucket = xchg(&smap->buckets[id], new_bucket);
 279         if (old_bucket)
 280                 pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
 281         return id;
 282 }
 283
 284 BPF_CALL_3(bpf_get_stackid, struct pt_regs *, regs, struct bpf_map *, map,
 285            u64, flags)
 286 {
 287         u32 max_depth = map->value_size / stack_map_data_size(map);
 288         /* stack_map_alloc() checks that max_depth <= sysctl_perf_event_max_stack */
 289         u32 init_nr = sysctl_perf_event_max_stack - max_depth;
 290         bool user = flags & BPF_F_USER_STACK;
 291         struct perf_callchain_entry *trace;
 292         bool kernel = !user;
 293
 294         if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
 295                                BPF_F_FAST_STACK_CMP | BPF_F_REUSE_STACKID)))
 296                 return -EINVAL;
 297
 298         trace = get_perf_callchain(regs, init_nr, kernel, user,
 299                                    sysctl_perf_event_max_stack, false, false);
 300
 301         if (unlikely(!trace))
 302                 /* couldn't fetch the stack trace */
 303                 return -EFAULT;
 304
 305         return __bpf_get_stackid(map, trace, flags);
 306 }
 307
 308 const struct bpf_func_proto bpf_get_stackid_proto = {
 309         .func           = bpf_get_stackid,
 310         .gpl_only       = true,
 311         .ret_type       = RET_INTEGER,
 312         .arg1_type      = ARG_PTR_TO_CTX,
 313         .arg2_type      = ARG_CONST_MAP_PTR,
 314         .arg3_type      = ARG_ANYTHING,
 315 };
 316
 317 static __u64 count_kernel_ip(struct perf_callchain_entry *trace)
 318 {
 319         __u64 nr_kernel = 0;
 320
 321         while (nr_kernel < trace->nr) {
 322                 if (trace->ip[nr_kernel] == PERF_CONTEXT_USER)
 323                         break;
 324                 nr_kernel++;
 325         }
 326         return nr_kernel;
 327 }
 328
 329 BPF_CALL_3(bpf_get_stackid_pe, struct bpf_perf_event_data_kern *, ctx,
 330            struct bpf_map *, map, u64, flags)
 331 {
 332         struct perf_event *event = ctx->event;
 333         struct perf_callchain_entry *trace;
 334         bool kernel, user;
 335         __u64 nr_kernel;
 336         int ret;
 337
 338         /* perf_sample_data doesn't have callchain, use bpf_get_stackid */
 339         if (!(event->attr.sample_type & __PERF_SAMPLE_CALLCHAIN_EARLY))
 340                 return bpf_get_stackid((unsigned long)(ctx->regs),
 341                                        (unsigned long) map, flags, 0, 0);
 342
 343         if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
 344                                BPF_F_FAST_STACK_CMP | BPF_F_REUSE_STACKID)))
 345                 return -EINVAL;
 346
 347         user = flags & BPF_F_USER_STACK;
 348         kernel = !user;
 349
 350         trace = ctx->data->callchain;
 351         if (unlikely(!trace))
 352                 return -EFAULT;
 353
 354         nr_kernel = count_kernel_ip(trace);
 355
 356         if (kernel) {
 357                 __u64 nr = trace->nr;
 358
 359                 trace->nr = nr_kernel;
 360                 ret = __bpf_get_stackid(map, trace, flags);
 361
 362                 /* restore nr */
 363                 trace->nr = nr;
 364         } else { /* user */
 365                 u64 skip = flags & BPF_F_SKIP_FIELD_MASK;
 366
 367                 skip += nr_kernel;
 368                 if (skip > BPF_F_SKIP_FIELD_MASK)
 369                         return -EFAULT;
 370
 371                 flags = (flags & ~BPF_F_SKIP_FIELD_MASK) | skip;
 372                 ret = __bpf_get_stackid(map, trace, flags);
 373         }
 374         return ret;
 375 }
 376
 377 const struct bpf_func_proto bpf_get_stackid_proto_pe = {
 378         .func           = bpf_get_stackid_pe,
 379         .gpl_only       = false,
 380         .ret_type       = RET_INTEGER,
 381         .arg1_type      = ARG_PTR_TO_CTX,
 382         .arg2_type      = ARG_CONST_MAP_PTR,
 383         .arg3_type      = ARG_ANYTHING,
 384 };
 385
 386 static long __bpf_get_stack(struct pt_regs *regs, struct task_struct *task,
 387                             struct perf_callchain_entry *trace_in,
 388                             void *buf, u32 size, u64 flags)
 389 {
 390         u32 init_nr, trace_nr, copy_len, elem_size, num_elem;
 391         bool user_build_id = flags & BPF_F_USER_BUILD_ID;
 392         u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
 393         bool user = flags & BPF_F_USER_STACK;
 394         struct perf_callchain_entry *trace;
 395         bool kernel = !user;
 396         int err = -EINVAL;
 397         u64 *ips;
 398
 399         if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
 400                                BPF_F_USER_BUILD_ID)))
 401                 goto clear;
 402         if (kernel && user_build_id)
 403                 goto clear;
 404
 405         elem_size = (user && user_build_id) ? sizeof(struct bpf_stack_build_id)
 406                                             : sizeof(u64);
 407         if (unlikely(size % elem_size))
 408                 goto clear;
 409
 410         /* cannot get valid user stack for task without user_mode regs */
 411         if (task && user && !user_mode(regs))
 412                 goto err_fault;
 413
 414         num_elem = size / elem_size;
 415         if (sysctl_perf_event_max_stack < num_elem)
 416                 init_nr = 0;
 417         else
 418                 init_nr = sysctl_perf_event_max_stack - num_elem;
 419
 420         if (trace_in)
 421                 trace = trace_in;
 422         else if (kernel && task)
 423                 trace = get_callchain_entry_for_task(task, init_nr);
 424         else
 425                 trace = get_perf_callchain(regs, init_nr, kernel, user,
 426                                            sysctl_perf_event_max_stack,
 427                                            false, false);
 428         if (unlikely(!trace))
 429                 goto err_fault;
 430
 431         trace_nr = trace->nr - init_nr;
 432         if (trace_nr < skip)
 433                 goto err_fault;
 434
 435         trace_nr -= skip;
 436         trace_nr = (trace_nr <= num_elem) ? trace_nr : num_elem;
 437         copy_len = trace_nr * elem_size;
 438         ips = trace->ip + skip + init_nr;
 439         if (user && user_build_id)
 440                 stack_map_get_build_id_offset(buf, ips, trace_nr, user);
 441         else
 442                 memcpy(buf, ips, copy_len);
 443
 444         if (size > copy_len)
 445                 memset(buf + copy_len, 0, size - copy_len);
 446         return copy_len;
 447
 448 err_fault:
 449         err = -EFAULT;
 450 clear:
 451         memset(buf, 0, size);
 452         return err;
 453 }
 454
 455 BPF_CALL_4(bpf_get_stack, struct pt_regs *, regs, void *, buf, u32, size,
 456            u64, flags)
 457 {
 458         return __bpf_get_stack(regs, NULL, NULL, buf, size, flags);
 459 }
 460
 461 const struct bpf_func_proto bpf_get_stack_proto = {
 462         .func           = bpf_get_stack,
 463         .gpl_only       = true,
 464         .ret_type       = RET_INTEGER,
 465         .arg1_type      = ARG_PTR_TO_CTX,
 466         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
 467         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
 468         .arg4_type      = ARG_ANYTHING,
 469 };
 470
 471 BPF_CALL_4(bpf_get_task_stack, struct task_struct *, task, void *, buf,
 472            u32, size, u64, flags)
 473 {
 474         struct pt_regs *regs;
 475         long res = -EINVAL;
 476
 477         if (!try_get_task_stack(task))
 478                 return -EFAULT;
 479
 480         regs = task_pt_regs(task);
 481         if (regs)
 482                 res = __bpf_get_stack(regs, task, NULL, buf, size, flags);
 483         put_task_stack(task);
 484
 485         return res;
 486 }
 487
 488 const struct bpf_func_proto bpf_get_task_stack_proto = {
 489         .func           = bpf_get_task_stack,
 490         .gpl_only       = false,
 491         .ret_type       = RET_INTEGER,
 492         .arg1_type      = ARG_PTR_TO_BTF_ID,
 493         .arg1_btf_id    = &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
 494         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
 495         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
 496         .arg4_type      = ARG_ANYTHING,
 497 };
 498
 499 BPF_CALL_4(bpf_get_stack_pe, struct bpf_perf_event_data_kern *, ctx,
 500            void *, buf, u32, size, u64, flags)
 501 {
 502         struct pt_regs *regs = (struct pt_regs *)(ctx->regs);
 503         struct perf_event *event = ctx->event;
 504         struct perf_callchain_entry *trace;
 505         bool kernel, user;
 506         int err = -EINVAL;
 507         __u64 nr_kernel;
 508
 509         if (!(event->attr.sample_type & __PERF_SAMPLE_CALLCHAIN_EARLY))
 510                 return __bpf_get_stack(regs, NULL, NULL, buf, size, flags);
 511
 512         if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
 513                                BPF_F_USER_BUILD_ID)))
 514                 goto clear;
 515
 516         user = flags & BPF_F_USER_STACK;
 517         kernel = !user;
 518
 519         err = -EFAULT;
 520         trace = ctx->data->callchain;
 521         if (unlikely(!trace))
 522                 goto clear;
 523
 524         nr_kernel = count_kernel_ip(trace);
 525
 526         if (kernel) {
 527                 __u64 nr = trace->nr;
 528
 529                 trace->nr = nr_kernel;
 530                 err = __bpf_get_stack(regs, NULL, trace, buf, size, flags);
 531
 532                 /* restore nr */
 533                 trace->nr = nr;
 534         } else { /* user */
 535                 u64 skip = flags & BPF_F_SKIP_FIELD_MASK;
 536
 537                 skip += nr_kernel;
 538                 if (skip > BPF_F_SKIP_FIELD_MASK)
 539                         goto clear;
 540
 541                 flags = (flags & ~BPF_F_SKIP_FIELD_MASK) | skip;
 542                 err = __bpf_get_stack(regs, NULL, trace, buf, size, flags);
 543         }
 544         return err;
 545
 546 clear:
 547         memset(buf, 0, size);
 548         return err;
 549
 550 }
 551
 552 const struct bpf_func_proto bpf_get_stack_proto_pe = {
 553         .func           = bpf_get_stack_pe,
 554         .gpl_only       = true,
 555         .ret_type       = RET_INTEGER,
 556         .arg1_type      = ARG_PTR_TO_CTX,
 557         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
 558         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
 559         .arg4_type      = ARG_ANYTHING,
 560 };
 561
 562 /* Called from eBPF program */
 563 static void *stack_map_lookup_elem(struct bpf_map *map, void *key)
 564 {
 565         return ERR_PTR(-EOPNOTSUPP);
 566 }
 567
 568 /* Called from syscall */
 569 int bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
 570 {
 571         struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
 572         struct stack_map_bucket *bucket, *old_bucket;
 573         u32 id = *(u32 *)key, trace_len;
 574
 575         if (unlikely(id >= smap->n_buckets))
 576                 return -ENOENT;
 577
 578         bucket = xchg(&smap->buckets[id], NULL);
 579         if (!bucket)
 580                 return -ENOENT;
 581
 582         trace_len = bucket->nr * stack_map_data_size(map);
 583         memcpy(value, bucket->data, trace_len);
 584         memset(value + trace_len, 0, map->value_size - trace_len);
 585
 586         old_bucket = xchg(&smap->buckets[id], bucket);
 587         if (old_bucket)
 588                 pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
 589         return 0;
 590 }
 591
 592 static int stack_map_get_next_key(struct bpf_map *map, void *key,
 593                                   void *next_key)
 594 {
 595         struct bpf_stack_map *smap = container_of(map,
 596                                                   struct bpf_stack_map, map);
 597         u32 id;
 598
 599         WARN_ON_ONCE(!rcu_read_lock_held());
 600
 601         if (!key) {
 602                 id = 0;
 603         } else {
 604                 id = *(u32 *)key;
 605                 if (id >= smap->n_buckets || !smap->buckets[id])
 606                         id = 0;
 607                 else
 608                         id++;
 609         }
 610
 611         while (id < smap->n_buckets && !smap->buckets[id])
 612                 id++;
 613
 614         if (id >= smap->n_buckets)
 615                 return -ENOENT;
 616
 617         *(u32 *)next_key = id;
 618         return 0;
 619 }
 620
 621 static int stack_map_update_elem(struct bpf_map *map, void *key, void *value,
 622                                  u64 map_flags)
 623 {
 624         return -EINVAL;
 625 }
 626
 627 /* Called from syscall or from eBPF program */
 628 static int stack_map_delete_elem(struct bpf_map *map, void *key)
 629 {
 630         struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
 631         struct stack_map_bucket *old_bucket;
 632         u32 id = *(u32 *)key;
 633
 634         if (unlikely(id >= smap->n_buckets))
 635                 return -E2BIG;
 636
 637         old_bucket = xchg(&smap->buckets[id], NULL);
 638         if (old_bucket) {
 639                 pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
 640                 return 0;
 641         } else {
 642                 return -ENOENT;
 643         }
 644 }
 645
 646 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
 647 static void stack_map_free(struct bpf_map *map)
 648 {
 649         struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
 650
 651         bpf_map_area_free(smap->elems);
 652         pcpu_freelist_destroy(&smap->freelist);
 653         bpf_map_area_free(smap);
 654         put_callchain_buffers();
 655 }
 656
 657 static int stack_trace_map_btf_id;
 658 const struct bpf_map_ops stack_trace_map_ops = {
 659         .map_meta_equal = bpf_map_meta_equal,
 660         .map_alloc = stack_map_alloc,
 661         .map_free = stack_map_free,
 662         .map_get_next_key = stack_map_get_next_key,
 663         .map_lookup_elem = stack_map_lookup_elem,
 664         .map_update_elem = stack_map_update_elem,
 665         .map_delete_elem = stack_map_delete_elem,
 666         .map_check_btf = map_check_no_btf,
 667         .map_btf_name = "bpf_stack_map",
 668         .map_btf_id = &stack_trace_map_btf_id,
 669 };