| 1 | |
| 2 | |
| 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 <linux/mmap_lock.h>
|
| 13 | #include "percpu_freelist.h"
|
| 14 | #include "mmap_unlock_work.h"
|
| 15 |
|
| 16 | #define STACK_CREATE_FLAG_MASK \
|
| 17 | (BPF_F_NUMA_NODE | BPF_F_RDONLY | BPF_F_WRONLY | \
|
| 18 | BPF_F_STACK_BUILD_ID)
|
| 19 |
|
| 20 | struct stack_map_bucket {
|
| 21 | struct pcpu_freelist_node fnode;
|
| 22 | u32 hash;
|
| 23 | u32 nr;
|
| 24 | u64 data[];
|
| 25 | };
|
| 26 |
|
| 27 | struct bpf_stack_map {
|
| 28 | struct bpf_map map;
|
| 29 | void *elems;
|
| 30 | struct pcpu_freelist freelist;
|
| 31 | u32 n_buckets;
|
| 32 | struct stack_map_bucket *buckets[] __counted_by(n_buckets);
|
| 33 | };
|
| 34 |
|
| 35 | static inline bool stack_map_use_build_id(struct bpf_map *map)
|
| 36 | {
|
| 37 | return (map->map_flags & BPF_F_STACK_BUILD_ID);
|
| 38 | }
|
| 39 |
|
| 40 | static inline int stack_map_data_size(struct bpf_map *map)
|
| 41 | {
|
| 42 | return stack_map_use_build_id(map) ?
|
| 43 | sizeof(struct bpf_stack_build_id) : sizeof(u64);
|
| 44 | }
|
| 45 |
|
| 46 | |
| 47 | * stack_map_calculate_max_depth - Calculate maximum allowed stack trace depth
|
| 48 | * @size: Size of the buffer/map value in bytes
|
| 49 | * @elem_size: Size of each stack trace element
|
| 50 | * @flags: BPF stack trace flags (BPF_F_USER_STACK, BPF_F_USER_BUILD_ID, ...)
|
| 51 | *
|
| 52 | * Return: Maximum number of stack trace entries that can be safely stored
|
| 53 |
|
| 54 | static u32 stack_map_calculate_max_depth(u32 size, u32 elem_size, u64 flags)
|
| 55 | {
|
| 56 | u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
|
| 57 | u32 max_depth;
|
| 58 | u32 curr_sysctl_max_stack = READ_ONCE(sysctl_perf_event_max_stack);
|
| 59 |
|
| 60 | max_depth = size / elem_size;
|
| 61 | max_depth += skip;
|
| 62 | if (max_depth > curr_sysctl_max_stack)
|
| 63 | return curr_sysctl_max_stack;
|
| 64 |
|
| 65 | return max_depth;
|
| 66 | }
|
| 67 |
|
| 68 | static int prealloc_elems_and_freelist(struct bpf_stack_map *smap)
|
| 69 | {
|
| 70 | u64 elem_size = sizeof(struct stack_map_bucket) +
|
| 71 | (u64)smap->map.value_size;
|
| 72 | int err;
|
| 73 |
|
| 74 | smap->elems = bpf_map_area_alloc(elem_size * smap->map.max_entries,
|
| 75 | smap->map.numa_node);
|
| 76 | if (!smap->elems)
|
| 77 | return -ENOMEM;
|
| 78 |
|
| 79 | err = pcpu_freelist_init(&smap->freelist);
|
| 80 | if (err)
|
| 81 | goto free_elems;
|
| 82 |
|
| 83 | pcpu_freelist_populate(&smap->freelist, smap->elems, elem_size,
|
| 84 | smap->map.max_entries);
|
| 85 | return 0;
|
| 86 |
|
| 87 | free_elems:
|
| 88 | bpf_map_area_free(smap->elems);
|
| 89 | return err;
|
| 90 | }
|
| 91 |
|
| 92 |
|
| 93 | static struct bpf_map *stack_map_alloc(union bpf_attr *attr)
|
| 94 | {
|
| 95 | u32 value_size = attr->value_size;
|
| 96 | struct bpf_stack_map *smap;
|
| 97 | u64 cost, n_buckets;
|
| 98 | int err;
|
| 99 |
|
| 100 | if (attr->map_flags & ~STACK_CREATE_FLAG_MASK)
|
| 101 | return ERR_PTR(-EINVAL);
|
| 102 |
|
| 103 |
|
| 104 | if (attr->max_entries == 0 || attr->key_size != 4 ||
|
| 105 | value_size < 8 || value_size % 8)
|
| 106 | return ERR_PTR(-EINVAL);
|
| 107 |
|
| 108 | BUILD_BUG_ON(sizeof(struct bpf_stack_build_id) % sizeof(u64));
|
| 109 | if (attr->map_flags & BPF_F_STACK_BUILD_ID) {
|
| 110 | if (value_size % sizeof(struct bpf_stack_build_id) ||
|
| 111 | value_size / sizeof(struct bpf_stack_build_id)
|
| 112 | > sysctl_perf_event_max_stack)
|
| 113 | return ERR_PTR(-EINVAL);
|
| 114 | } else if (value_size / 8 > sysctl_perf_event_max_stack)
|
| 115 | return ERR_PTR(-EINVAL);
|
| 116 |
|
| 117 | |
| 118 | * into UB on 32-bit arches, so check that first
|
| 119 |
|
| 120 | if (attr->max_entries > 1UL << 31)
|
| 121 | return ERR_PTR(-E2BIG);
|
| 122 |
|
| 123 | n_buckets = roundup_pow_of_two(attr->max_entries);
|
| 124 |
|
| 125 | cost = n_buckets * sizeof(struct stack_map_bucket *) + sizeof(*smap);
|
| 126 | smap = bpf_map_area_alloc(cost, bpf_map_attr_numa_node(attr));
|
| 127 | if (!smap)
|
| 128 | return ERR_PTR(-ENOMEM);
|
| 129 |
|
| 130 | bpf_map_init_from_attr(&smap->map, attr);
|
| 131 | smap->n_buckets = n_buckets;
|
| 132 |
|
| 133 | err = get_callchain_buffers(sysctl_perf_event_max_stack);
|
| 134 | if (err)
|
| 135 | goto free_smap;
|
| 136 |
|
| 137 | err = prealloc_elems_and_freelist(smap);
|
| 138 | if (err)
|
| 139 | goto put_buffers;
|
| 140 |
|
| 141 | return &smap->map;
|
| 142 |
|
| 143 | put_buffers:
|
| 144 | put_callchain_buffers();
|
| 145 | free_smap:
|
| 146 | bpf_map_area_free(smap);
|
| 147 | return ERR_PTR(err);
|
| 148 | }
|
| 149 |
|
| 150 | static int fetch_build_id(struct vm_area_struct *vma, unsigned char *build_id, bool may_fault)
|
| 151 | {
|
| 152 | return may_fault ? build_id_parse(vma, build_id, NULL)
|
| 153 | : build_id_parse_nofault(vma, build_id, NULL);
|
| 154 | }
|
| 155 |
|
| 156 | static inline void stack_map_build_id_set_ip(struct bpf_stack_build_id *id)
|
| 157 | {
|
| 158 | id->status = BPF_STACK_BUILD_ID_IP;
|
| 159 | memset(id->build_id, 0, BUILD_ID_SIZE_MAX);
|
| 160 | }
|
| 161 |
|
| 162 | static inline u64 stack_map_build_id_offset(unsigned long vm_pgoff,
|
| 163 | unsigned long vm_start, u64 ip)
|
| 164 | {
|
| 165 | return (vm_pgoff << PAGE_SHIFT) + ip - vm_start;
|
| 166 | }
|
| 167 |
|
| 168 | static inline void stack_map_build_id_set_valid(struct bpf_stack_build_id *id,
|
| 169 | u64 offset,
|
| 170 | const unsigned char *build_id)
|
| 171 | {
|
| 172 | id->status = BPF_STACK_BUILD_ID_VALID;
|
| 173 | id->offset = offset;
|
| 174 | if (id->build_id != build_id)
|
| 175 | memcpy(id->build_id, build_id, BUILD_ID_SIZE_MAX);
|
| 176 | }
|
| 177 |
|
| 178 | |
| 179 | * A cached VMA lookup result. The range [vm_start, vm_end) is always set.
|
| 180 | * vm_pgoff, file, build_id are set only when the build ID was resolved.
|
| 181 | * Zero vm_end marks the slot empty. build_id aliases the id_offs[] entry.
|
| 182 |
|
| 183 | struct stack_map_cached_vma {
|
| 184 | unsigned long vm_start;
|
| 185 | unsigned long vm_end;
|
| 186 | unsigned long vm_pgoff;
|
| 187 | struct file *file;
|
| 188 | const unsigned char *build_id;
|
| 189 | };
|
| 190 |
|
| 191 | |
| 192 | * Per stack_map_get_build_id_offset() call cache of the last VMA with a build ID
|
| 193 | * resolved and the last VMA with no usable build ID. Adjacent stack frames tend
|
| 194 | * to land in the same VMA or the same backing file, so caching the last result
|
| 195 | * of each kind lets us skip unnecessary VMA lookups and build ID parse calls.
|
| 196 | * Keeping the two slots independent means a build-ID-less VMA doesn't evict the
|
| 197 | * last resolved build ID.
|
| 198 |
|
| 199 | struct stack_map_build_id_cache {
|
| 200 | struct stack_map_cached_vma resolved;
|
| 201 | struct stack_map_cached_vma unresolved;
|
| 202 | };
|
| 203 |
|
| 204 | |
| 205 | * Fill @id from a cached range covering @ip. On a hit this writes @id (resolved
|
| 206 | * range -> build ID + offset, unresolved range -> raw ip) and returns 0; on a
|
| 207 | * miss it leaves @id untouched and returns -ENOENT.
|
| 208 |
|
| 209 | static int stack_map_build_id_set_from_cache(struct stack_map_build_id_cache *cache,
|
| 210 | struct bpf_stack_build_id *id, u64 ip)
|
| 211 | {
|
| 212 | unsigned long vm_start, vm_end, vm_pgoff;
|
| 213 | u64 offset;
|
| 214 |
|
| 215 | vm_start = cache->resolved.vm_start;
|
| 216 | vm_end = cache->resolved.vm_end;
|
| 217 | if (vm_end && ip >= vm_start && ip < vm_end) {
|
| 218 | vm_pgoff = cache->resolved.vm_pgoff;
|
| 219 | offset = stack_map_build_id_offset(vm_pgoff, vm_start, ip);
|
| 220 | stack_map_build_id_set_valid(id, offset, cache->resolved.build_id);
|
| 221 | return 0;
|
| 222 | }
|
| 223 |
|
| 224 | vm_start = cache->unresolved.vm_start;
|
| 225 | vm_end = cache->unresolved.vm_end;
|
| 226 | if (vm_end && ip >= vm_start && ip < vm_end) {
|
| 227 | stack_map_build_id_set_ip(id);
|
| 228 | return 0;
|
| 229 | }
|
| 230 |
|
| 231 | return -ENOENT;
|
| 232 | }
|
| 233 |
|
| 234 | |
| 235 | * Record @vma's build ID as the last resolved one. @file is the pinned backing
|
| 236 | * file in the sleepable path (released when evicted), or NULL otherwise.
|
| 237 |
|
| 238 | static void stack_map_build_id_cache_set_resolved(struct stack_map_build_id_cache *cache,
|
| 239 | struct file *file,
|
| 240 | const unsigned char *build_id,
|
| 241 | unsigned long vm_start,
|
| 242 | unsigned long vm_end,
|
| 243 | unsigned long vm_pgoff)
|
| 244 | {
|
| 245 | if (cache->resolved.file)
|
| 246 | fput(cache->resolved.file);
|
| 247 | cache->resolved = (struct stack_map_cached_vma){
|
| 248 | .vm_start = vm_start,
|
| 249 | .vm_end = vm_end,
|
| 250 | .vm_pgoff = vm_pgoff,
|
| 251 | .file = file,
|
| 252 | .build_id = build_id,
|
| 253 | };
|
| 254 | }
|
| 255 |
|
| 256 |
|
| 257 | static void stack_map_build_id_cache_set_unresolved(struct stack_map_build_id_cache *cache,
|
| 258 | unsigned long vm_start,
|
| 259 | unsigned long vm_end)
|
| 260 | {
|
| 261 | cache->unresolved = (struct stack_map_cached_vma){
|
| 262 | .vm_start = vm_start,
|
| 263 | .vm_end = vm_end,
|
| 264 | };
|
| 265 | }
|
| 266 |
|
| 267 | struct stack_map_vma_lock {
|
| 268 | struct vm_area_struct *vma;
|
| 269 | struct mm_struct *mm;
|
| 270 | };
|
| 271 |
|
| 272 | |
| 273 | * Acquire a stable read-side reference on the VMA covering @ip.
|
| 274 | *
|
| 275 | * With CONFIG_PER_VMA_LOCK=y this returns a VMA with its per-VMA read
|
| 276 | * lock held and mmap_lock dropped, so the caller may sleep.
|
| 277 | *
|
| 278 | * With CONFIG_PER_VMA_LOCK=n it returns a VMA with mmap_lock still
|
| 279 | * held; the caller must snapshot any fields it needs and pin vm_file
|
| 280 | * with get_file() before stack_map_unlock_vma() drops mmap_lock, as
|
| 281 | * the VMA may be split, merged, or freed after that.
|
| 282 | *
|
| 283 | * Returns NULL on failure, in which case no lock is held.
|
| 284 |
|
| 285 | static struct vm_area_struct *
|
| 286 | stack_map_lock_vma(struct stack_map_vma_lock *lock, unsigned long ip)
|
| 287 | {
|
| 288 | struct mm_struct *mm = lock->mm;
|
| 289 | struct vm_area_struct *vma;
|
| 290 |
|
| 291 |
|
| 292 | vma = lock_vma_under_rcu(mm, ip);
|
| 293 | if (vma) {
|
| 294 | lock->vma = vma;
|
| 295 | return vma;
|
| 296 | }
|
| 297 |
|
| 298 | |
| 299 | * Taking mmap_read_lock() is unsafe here, because the caller BPF
|
| 300 | * program might already hold it, causing a deadlock.
|
| 301 |
|
| 302 | if (!mmap_read_trylock(mm))
|
| 303 | return NULL;
|
| 304 |
|
| 305 | vma = vma_lookup(mm, ip);
|
| 306 | if (!vma) {
|
| 307 | mmap_read_unlock(mm);
|
| 308 | return NULL;
|
| 309 | }
|
| 310 |
|
| 311 | #ifdef CONFIG_PER_VMA_LOCK
|
| 312 | if (!vma_start_read_locked(vma)) {
|
| 313 | mmap_read_unlock(mm);
|
| 314 | return NULL;
|
| 315 | }
|
| 316 | mmap_read_unlock(mm);
|
| 317 | #endif
|
| 318 |
|
| 319 | lock->vma = vma;
|
| 320 | return vma;
|
| 321 | }
|
| 322 |
|
| 323 | static void stack_map_unlock_vma(struct stack_map_vma_lock *lock)
|
| 324 | {
|
| 325 | #ifdef CONFIG_PER_VMA_LOCK
|
| 326 | vma_end_read(lock->vma);
|
| 327 | #else
|
| 328 | mmap_read_unlock(lock->mm);
|
| 329 | #endif
|
| 330 | lock->vma = NULL;
|
| 331 | }
|
| 332 |
|
| 333 | static void stack_map_get_build_id_offset_sleepable(struct bpf_stack_build_id *id_offs,
|
| 334 | u32 trace_nr)
|
| 335 | {
|
| 336 | struct stack_map_vma_lock lock = { .mm = current->mm };
|
| 337 | struct stack_map_build_id_cache cache = {};
|
| 338 | struct stack_map_cached_vma *res = &cache.resolved;
|
| 339 | unsigned long vm_pgoff, vm_start, vm_end;
|
| 340 | struct vm_area_struct *vma;
|
| 341 | struct file *file;
|
| 342 | u64 offset;
|
| 343 | u64 ip;
|
| 344 |
|
| 345 | for (u32 i = 0; i < trace_nr; i++) {
|
| 346 | ip = READ_ONCE(id_offs[i].ip);
|
| 347 |
|
| 348 | if (!stack_map_build_id_set_from_cache(&cache, &id_offs[i], ip))
|
| 349 | continue;
|
| 350 |
|
| 351 | vma = stack_map_lock_vma(&lock, ip);
|
| 352 | if (!vma) {
|
| 353 | stack_map_build_id_set_ip(&id_offs[i]);
|
| 354 | continue;
|
| 355 | }
|
| 356 |
|
| 357 | vm_pgoff = vma->vm_pgoff;
|
| 358 | vm_start = vma->vm_start;
|
| 359 | vm_end = vma->vm_end;
|
| 360 |
|
| 361 | if (vma_is_anonymous(vma) || !vma->vm_file) {
|
| 362 | stack_map_unlock_vma(&lock);
|
| 363 | stack_map_build_id_set_ip(&id_offs[i]);
|
| 364 | stack_map_build_id_cache_set_unresolved(&cache, vm_start, vm_end);
|
| 365 | continue;
|
| 366 | }
|
| 367 |
|
| 368 | file = vma->vm_file;
|
| 369 | offset = stack_map_build_id_offset(vm_pgoff, vm_start, ip);
|
| 370 |
|
| 371 | |
| 372 | * Same backing file as the last resolved VMA (another mapping
|
| 373 | * of the same ELF binary): reuse its build_id without re-parsing.
|
| 374 |
|
| 375 | if (file == res->file) {
|
| 376 | stack_map_unlock_vma(&lock);
|
| 377 | stack_map_build_id_set_valid(&id_offs[i], offset, res->build_id);
|
| 378 | res->vm_start = vm_start;
|
| 379 | res->vm_end = vm_end;
|
| 380 | res->vm_pgoff = vm_pgoff;
|
| 381 | continue;
|
| 382 | }
|
| 383 |
|
| 384 | file = get_file(file);
|
| 385 | stack_map_unlock_vma(&lock);
|
| 386 |
|
| 387 |
|
| 388 | if (build_id_parse_file(file, id_offs[i].build_id, NULL)) {
|
| 389 | stack_map_build_id_set_ip(&id_offs[i]);
|
| 390 | fput(file);
|
| 391 | stack_map_build_id_cache_set_unresolved(&cache, vm_start, vm_end);
|
| 392 | continue;
|
| 393 | }
|
| 394 |
|
| 395 | stack_map_build_id_set_valid(&id_offs[i], offset, id_offs[i].build_id);
|
| 396 | stack_map_build_id_cache_set_resolved(&cache, file, id_offs[i].build_id,
|
| 397 | vm_start, vm_end, vm_pgoff);
|
| 398 | }
|
| 399 |
|
| 400 | if (res->file)
|
| 401 | fput(res->file);
|
| 402 | }
|
| 403 |
|
| 404 | |
| 405 | * Expects all id_offs[i].ip values to be set to correct initial IPs.
|
| 406 | * They will be subsequently:
|
| 407 | * - either adjusted in place to a file offset, if build ID fetching
|
| 408 | * succeeds; in this case id_offs[i].build_id is set to correct build ID,
|
| 409 | * and id_offs[i].status is set to BPF_STACK_BUILD_ID_VALID;
|
| 410 | * - or IP will be kept intact, if build ID fetching failed; in this case
|
| 411 | * id_offs[i].build_id is zeroed out and id_offs[i].status is set to
|
| 412 | * BPF_STACK_BUILD_ID_IP.
|
| 413 |
|
| 414 | static void stack_map_get_build_id_offset(struct bpf_stack_build_id *id_offs,
|
| 415 | u32 trace_nr, bool user, bool may_fault)
|
| 416 | {
|
| 417 | struct mmap_unlock_irq_work *work;
|
| 418 | bool has_user_ctx = user && current && current->mm;
|
| 419 | struct stack_map_build_id_cache cache = {};
|
| 420 | struct vm_area_struct *vma;
|
| 421 | int i;
|
| 422 |
|
| 423 | if (may_fault && has_user_ctx) {
|
| 424 | stack_map_get_build_id_offset_sleepable(id_offs, trace_nr);
|
| 425 | return;
|
| 426 | }
|
| 427 |
|
| 428 | if (!has_user_ctx)
|
| 429 | goto fallback;
|
| 430 |
|
| 431 | work = bpf_mmap_unlock_guard_get();
|
| 432 | if (IS_ERR(work))
|
| 433 | goto fallback;
|
| 434 |
|
| 435 | if (!mmap_read_trylock(current->mm)) {
|
| 436 | bpf_mmap_unlock_guard_put(work);
|
| 437 | goto fallback;
|
| 438 | }
|
| 439 |
|
| 440 | for (i = 0; i < trace_nr; i++) {
|
| 441 | u64 ip = READ_ONCE(id_offs[i].ip);
|
| 442 |
|
| 443 | if (!stack_map_build_id_set_from_cache(&cache, &id_offs[i], ip))
|
| 444 | continue;
|
| 445 |
|
| 446 | vma = find_vma(current->mm, ip);
|
| 447 | if (!vma || vma_is_anonymous(vma) ||
|
| 448 | fetch_build_id(vma, id_offs[i].build_id, may_fault)) {
|
| 449 |
|
| 450 | stack_map_build_id_set_ip(&id_offs[i]);
|
| 451 | if (vma)
|
| 452 | stack_map_build_id_cache_set_unresolved(&cache,
|
| 453 | vma->vm_start, vma->vm_end);
|
| 454 | continue;
|
| 455 | }
|
| 456 | |
| 457 | * mmap_lock is held for the whole loop, so the cached VMA
|
| 458 | * fields stay valid; no file pinning is needed here.
|
| 459 |
|
| 460 | stack_map_build_id_set_valid(&id_offs[i],
|
| 461 | stack_map_build_id_offset(vma->vm_pgoff, vma->vm_start, ip),
|
| 462 | id_offs[i].build_id);
|
| 463 | stack_map_build_id_cache_set_resolved(&cache, NULL, id_offs[i].build_id,
|
| 464 | vma->vm_start, vma->vm_end,
|
| 465 | vma->vm_pgoff);
|
| 466 | }
|
| 467 | bpf_mmap_unlock_mm(work, current->mm);
|
| 468 | return;
|
| 469 |
|
| 470 | fallback:
|
| 471 |
|
| 472 | for (i = 0; i < trace_nr; i++)
|
| 473 | stack_map_build_id_set_ip(&id_offs[i]);
|
| 474 | }
|
| 475 |
|
| 476 | static struct perf_callchain_entry *
|
| 477 | get_callchain_entry_for_task(struct task_struct *task, u32 max_depth)
|
| 478 | {
|
| 479 | #ifdef CONFIG_STACKTRACE
|
| 480 | struct perf_callchain_entry *entry;
|
| 481 | int rctx;
|
| 482 |
|
| 483 | entry = get_callchain_entry(&rctx);
|
| 484 |
|
| 485 | if (!entry)
|
| 486 | return NULL;
|
| 487 |
|
| 488 | entry->nr = stack_trace_save_tsk(task, (unsigned long *)entry->ip,
|
| 489 | max_depth, 0);
|
| 490 |
|
| 491 | |
| 492 | * perf_callchain_entry uses u64 array. For 32-bit systems, it is
|
| 493 | * necessary to fix this mismatch.
|
| 494 |
|
| 495 | if (__BITS_PER_LONG != 64) {
|
| 496 | unsigned long *from = (unsigned long *) entry->ip;
|
| 497 | u64 *to = entry->ip;
|
| 498 | int i;
|
| 499 |
|
| 500 |
|
| 501 | for (i = entry->nr - 1; i >= 0; i--)
|
| 502 | to[i] = (u64)(from[i]);
|
| 503 | }
|
| 504 |
|
| 505 | put_callchain_entry(rctx);
|
| 506 |
|
| 507 | return entry;
|
| 508 | #else
|
| 509 | return NULL;
|
| 510 | #endif
|
| 511 | }
|
| 512 |
|
| 513 | struct stackid {
|
| 514 | struct stack_map_bucket *bucket;
|
| 515 | const u64 *ips;
|
| 516 | u32 nr;
|
| 517 | u32 len;
|
| 518 | u32 hash;
|
| 519 | u32 id;
|
| 520 | bool hash_matches;
|
| 521 | };
|
| 522 |
|
| 523 | static int stackid_init(struct stackid *stackid, struct bpf_map *map,
|
| 524 | const struct perf_callchain_entry *trace, u32 trace_nr, u64 flags)
|
| 525 | {
|
| 526 | struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
|
| 527 | u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
|
| 528 | u32 max_depth;
|
| 529 |
|
| 530 | if (trace_nr <= skip)
|
| 531 |
|
| 532 | return -EFAULT;
|
| 533 |
|
| 534 | max_depth = stack_map_calculate_max_depth(map->value_size, stack_map_data_size(map), flags);
|
| 535 | stackid->nr = min_t(u32, trace_nr - skip, max_depth - skip);
|
| 536 | stackid->len = stackid->nr * sizeof(u64);
|
| 537 | stackid->ips = trace->ip + skip;
|
| 538 | stackid->hash = jhash2((const u32 *)stackid->ips, stackid->len / sizeof(u32), 0);
|
| 539 | stackid->id = stackid->hash & (smap->n_buckets - 1);
|
| 540 | stackid->bucket = READ_ONCE(smap->buckets[stackid->id]);
|
| 541 | stackid->hash_matches = stackid->bucket && stackid->bucket->hash == stackid->hash;
|
| 542 | return 0;
|
| 543 | }
|
| 544 |
|
| 545 | static int stackid_fastpath(struct stackid *stackid, struct bpf_map *map,
|
| 546 | const struct perf_callchain_entry *trace, u32 trace_nr,
|
| 547 | u64 flags)
|
| 548 | {
|
| 549 | int err;
|
| 550 |
|
| 551 | err = stackid_init(stackid, map, trace, trace_nr, flags);
|
| 552 | if (err)
|
| 553 | return err;
|
| 554 |
|
| 555 |
|
| 556 | if (stackid->hash_matches && flags & BPF_F_FAST_STACK_CMP)
|
| 557 | return stackid->id;
|
| 558 |
|
| 559 | if (stack_map_use_build_id(map))
|
| 560 | return -ENOENT;
|
| 561 | if (stackid->hash_matches && stackid->bucket->nr == stackid->nr &&
|
| 562 | memcmp(stackid->bucket->data, stackid->ips, stackid->len) == 0)
|
| 563 | return stackid->id;
|
| 564 | if (stackid->bucket && !(flags & BPF_F_REUSE_STACKID))
|
| 565 | return -EEXIST;
|
| 566 | return -ENOENT;
|
| 567 | }
|
| 568 |
|
| 569 | static struct stack_map_bucket *
|
| 570 | stackid_new_bucket(struct stackid *stackid, struct bpf_map *map)
|
| 571 | {
|
| 572 | struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
|
| 573 | struct bpf_stack_build_id *id_offs;
|
| 574 | struct stack_map_bucket *bucket;
|
| 575 | u32 i;
|
| 576 |
|
| 577 | bucket = (struct stack_map_bucket *) pcpu_freelist_pop(&smap->freelist);
|
| 578 | if (unlikely(!bucket))
|
| 579 | return NULL;
|
| 580 |
|
| 581 | if (stack_map_use_build_id(map)) {
|
| 582 | id_offs = (struct bpf_stack_build_id *)bucket->data;
|
| 583 | for (i = 0; i < stackid->nr; i++)
|
| 584 | id_offs[i].ip = stackid->ips[i];
|
| 585 | } else {
|
| 586 | memcpy(bucket->data, stackid->ips, stackid->len);
|
| 587 | }
|
| 588 |
|
| 589 | bucket->hash = stackid->hash;
|
| 590 | bucket->nr = stackid->nr;
|
| 591 | return bucket;
|
| 592 | }
|
| 593 |
|
| 594 | static long stackid_install(struct stackid *stackid, struct bpf_map *map,
|
| 595 | struct stack_map_bucket *new_bucket, u64 flags)
|
| 596 | {
|
| 597 | struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
|
| 598 | bool user = flags & BPF_F_USER_STACK;
|
| 599 | struct stack_map_bucket *old_bucket;
|
| 600 | u32 trace_len;
|
| 601 |
|
| 602 | if (stack_map_use_build_id(map)) {
|
| 603 | struct bpf_stack_build_id *id_offs;
|
| 604 |
|
| 605 | id_offs = (struct bpf_stack_build_id *)new_bucket->data;
|
| 606 | stack_map_get_build_id_offset(id_offs, stackid->nr, user, false );
|
| 607 | trace_len = stackid->nr * sizeof(struct bpf_stack_build_id);
|
| 608 | if (stackid->hash_matches && stackid->bucket->nr == stackid->nr &&
|
| 609 | memcmp(stackid->bucket->data, new_bucket->data, trace_len) == 0) {
|
| 610 | pcpu_freelist_push(&smap->freelist, &new_bucket->fnode);
|
| 611 | return stackid->id;
|
| 612 | }
|
| 613 | if (stackid->bucket && !(flags & BPF_F_REUSE_STACKID)) {
|
| 614 | pcpu_freelist_push(&smap->freelist, &new_bucket->fnode);
|
| 615 | return -EEXIST;
|
| 616 | }
|
| 617 | }
|
| 618 |
|
| 619 | old_bucket = xchg(&smap->buckets[stackid->id], new_bucket);
|
| 620 | if (old_bucket)
|
| 621 | pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
|
| 622 | return stackid->id;
|
| 623 | }
|
| 624 |
|
| 625 | BPF_CALL_3(bpf_get_stackid, struct pt_regs *, regs, struct bpf_map *, map,
|
| 626 | u64, flags)
|
| 627 | {
|
| 628 | u32 elem_size = stack_map_data_size(map);
|
| 629 | bool user = flags & BPF_F_USER_STACK;
|
| 630 | struct stack_map_bucket *new_bucket;
|
| 631 | struct perf_callchain_entry *trace;
|
| 632 | struct stackid stackid;
|
| 633 | bool kernel = !user;
|
| 634 | u32 max_depth;
|
| 635 | int err;
|
| 636 |
|
| 637 | if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
|
| 638 | BPF_F_FAST_STACK_CMP | BPF_F_REUSE_STACKID)))
|
| 639 | return -EINVAL;
|
| 640 |
|
| 641 | max_depth = stack_map_calculate_max_depth(map->value_size, elem_size, flags);
|
| 642 |
|
| 643 | scoped_guard(preempt) {
|
| 644 | trace = get_perf_callchain(regs, kernel, user, max_depth,
|
| 645 | false, false, 0);
|
| 646 | if (unlikely(!trace))
|
| 647 |
|
| 648 | return -EFAULT;
|
| 649 |
|
| 650 | err = stackid_fastpath(&stackid, map, trace, trace->nr, flags);
|
| 651 | if (err != -ENOENT)
|
| 652 | return err;
|
| 653 |
|
| 654 | new_bucket = stackid_new_bucket(&stackid, map);
|
| 655 | if (!new_bucket)
|
| 656 | return -ENOMEM;
|
| 657 | }
|
| 658 |
|
| 659 | return stackid_install(&stackid, map, new_bucket, flags);
|
| 660 | }
|
| 661 |
|
| 662 | const struct bpf_func_proto bpf_get_stackid_proto = {
|
| 663 | .func = bpf_get_stackid,
|
| 664 | .gpl_only = true,
|
| 665 | .ret_type = RET_INTEGER,
|
| 666 | .arg1_type = ARG_PTR_TO_CTX,
|
| 667 | .arg2_type = ARG_CONST_MAP_PTR,
|
| 668 | .arg3_type = ARG_ANYTHING,
|
| 669 | };
|
| 670 |
|
| 671 | static __u64 count_kernel_ip(const struct perf_callchain_entry *trace)
|
| 672 | {
|
| 673 | __u64 nr_kernel = 0;
|
| 674 |
|
| 675 | while (nr_kernel < trace->nr) {
|
| 676 | if (trace->ip[nr_kernel] == PERF_CONTEXT_USER)
|
| 677 | break;
|
| 678 | nr_kernel++;
|
| 679 | }
|
| 680 | return nr_kernel;
|
| 681 | }
|
| 682 |
|
| 683 | BPF_CALL_3(bpf_get_stackid_pe, struct bpf_perf_event_data_kern *, ctx,
|
| 684 | struct bpf_map *, map, u64, flags)
|
| 685 | {
|
| 686 | const struct perf_callchain_entry *trace;
|
| 687 | struct perf_event *event = ctx->event;
|
| 688 | struct stack_map_bucket *new_bucket;
|
| 689 | struct stackid stackid;
|
| 690 | bool kernel, user;
|
| 691 | __u64 nr_kernel;
|
| 692 | u32 trace_nr;
|
| 693 | int ret;
|
| 694 |
|
| 695 |
|
| 696 | if (!(event->attr.sample_type & PERF_SAMPLE_CALLCHAIN))
|
| 697 | return bpf_get_stackid((unsigned long)(ctx->regs),
|
| 698 | (unsigned long) map, flags, 0, 0);
|
| 699 |
|
| 700 | if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
|
| 701 | BPF_F_FAST_STACK_CMP | BPF_F_REUSE_STACKID)))
|
| 702 | return -EINVAL;
|
| 703 |
|
| 704 | user = flags & BPF_F_USER_STACK;
|
| 705 | kernel = !user;
|
| 706 |
|
| 707 | trace = ctx->data->callchain;
|
| 708 | if (unlikely(!trace))
|
| 709 | return -EFAULT;
|
| 710 |
|
| 711 | nr_kernel = count_kernel_ip(trace);
|
| 712 |
|
| 713 | if (kernel) {
|
| 714 | trace_nr = nr_kernel;
|
| 715 | } else {
|
| 716 | u64 skip = flags & BPF_F_SKIP_FIELD_MASK;
|
| 717 |
|
| 718 | trace_nr = trace->nr;
|
| 719 | skip += nr_kernel;
|
| 720 | if (skip > BPF_F_SKIP_FIELD_MASK)
|
| 721 | return -EFAULT;
|
| 722 |
|
| 723 | flags = (flags & ~BPF_F_SKIP_FIELD_MASK) | skip;
|
| 724 | }
|
| 725 |
|
| 726 | ret = stackid_fastpath(&stackid, map, trace, trace_nr, flags);
|
| 727 | if (ret != -ENOENT)
|
| 728 | return ret;
|
| 729 |
|
| 730 | new_bucket = stackid_new_bucket(&stackid, map);
|
| 731 | if (new_bucket)
|
| 732 | return stackid_install(&stackid, map, new_bucket, flags);
|
| 733 | return -ENOMEM;
|
| 734 | }
|
| 735 |
|
| 736 | const struct bpf_func_proto bpf_get_stackid_proto_pe = {
|
| 737 | .func = bpf_get_stackid_pe,
|
| 738 | .gpl_only = false,
|
| 739 | .ret_type = RET_INTEGER,
|
| 740 | .arg1_type = ARG_PTR_TO_CTX,
|
| 741 | .arg2_type = ARG_CONST_MAP_PTR,
|
| 742 | .arg3_type = ARG_ANYTHING,
|
| 743 | };
|
| 744 |
|
| 745 | static u32 callchain_store(const struct perf_callchain_entry *trace, u32 trace_nr,
|
| 746 | void *buf, u32 elem_size, u64 flags)
|
| 747 | {
|
| 748 | bool user_build_id = flags & BPF_F_USER_BUILD_ID;
|
| 749 | u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
|
| 750 | const u64 *ips;
|
| 751 | u32 copy_len;
|
| 752 |
|
| 753 | trace_nr = trace_nr - skip;
|
| 754 | copy_len = trace_nr * elem_size;
|
| 755 |
|
| 756 | ips = trace->ip + skip;
|
| 757 | if (user_build_id) {
|
| 758 | struct bpf_stack_build_id *id_offs = buf;
|
| 759 |
|
| 760 | for (u32 i = 0; i < trace_nr; i++)
|
| 761 | id_offs[i].ip = ips[i];
|
| 762 | } else {
|
| 763 | memcpy(buf, ips, copy_len);
|
| 764 | }
|
| 765 | return trace_nr;
|
| 766 | }
|
| 767 |
|
| 768 | static long callchain_finalize(void *buf, u32 size, u32 trace_nr, u32 elem_size,
|
| 769 | u64 flags, bool may_fault)
|
| 770 | {
|
| 771 | bool user_build_id = flags & BPF_F_USER_BUILD_ID;
|
| 772 | bool user = flags & BPF_F_USER_STACK;
|
| 773 | u32 copy_len = trace_nr * elem_size;
|
| 774 |
|
| 775 | if (user_build_id)
|
| 776 | stack_map_get_build_id_offset(buf, trace_nr, user, may_fault);
|
| 777 |
|
| 778 | if (size > copy_len)
|
| 779 | memset(buf + copy_len, 0, size - copy_len);
|
| 780 | return copy_len;
|
| 781 | }
|
| 782 |
|
| 783 | static long __bpf_get_stack(struct pt_regs *regs, struct task_struct *task,
|
| 784 | void *buf, u32 size, u64 flags, bool may_fault)
|
| 785 | {
|
| 786 | bool user_build_id = flags & BPF_F_USER_BUILD_ID;
|
| 787 | bool crosstask = task && task != current;
|
| 788 | u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
|
| 789 | bool user = flags & BPF_F_USER_STACK;
|
| 790 | struct perf_callchain_entry *trace;
|
| 791 | u32 trace_nr, elem_size, max_depth;
|
| 792 | bool kernel = !user;
|
| 793 | int err = -EINVAL;
|
| 794 |
|
| 795 | if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
|
| 796 | BPF_F_USER_BUILD_ID)))
|
| 797 | goto clear;
|
| 798 | if (kernel && user_build_id)
|
| 799 | goto clear;
|
| 800 |
|
| 801 | elem_size = user_build_id ? sizeof(struct bpf_stack_build_id) : sizeof(u64);
|
| 802 | if (unlikely(size % elem_size))
|
| 803 | goto clear;
|
| 804 |
|
| 805 |
|
| 806 | if (task && user && !user_mode(regs))
|
| 807 | goto err_fault;
|
| 808 |
|
| 809 | |
| 810 | * but returns an empty stack instead of NULL.
|
| 811 |
|
| 812 | if (crosstask && user) {
|
| 813 | err = -EOPNOTSUPP;
|
| 814 | goto clear;
|
| 815 | }
|
| 816 |
|
| 817 | max_depth = stack_map_calculate_max_depth(size, elem_size, flags);
|
| 818 |
|
| 819 | preempt_disable();
|
| 820 | if (may_fault)
|
| 821 | rcu_read_lock();
|
| 822 |
|
| 823 | if (kernel && task) {
|
| 824 | trace = get_callchain_entry_for_task(task, max_depth);
|
| 825 | } else {
|
| 826 | trace = get_perf_callchain(regs, kernel, user, max_depth,
|
| 827 | crosstask, false, 0);
|
| 828 | }
|
| 829 |
|
| 830 | if (unlikely(!trace) || trace->nr < skip) {
|
| 831 | if (may_fault)
|
| 832 | rcu_read_unlock();
|
| 833 | preempt_enable();
|
| 834 | goto err_fault;
|
| 835 | }
|
| 836 |
|
| 837 | trace_nr = callchain_store(trace, trace->nr, buf, elem_size, flags);
|
| 838 |
|
| 839 |
|
| 840 | if (may_fault)
|
| 841 | rcu_read_unlock();
|
| 842 | preempt_enable();
|
| 843 |
|
| 844 | return callchain_finalize(buf, size, trace_nr, elem_size, flags, may_fault);
|
| 845 |
|
| 846 | err_fault:
|
| 847 | err = -EFAULT;
|
| 848 | clear:
|
| 849 | memset(buf, 0, size);
|
| 850 | return err;
|
| 851 | }
|
| 852 |
|
| 853 | BPF_CALL_4(bpf_get_stack, struct pt_regs *, regs, void *, buf, u32, size,
|
| 854 | u64, flags)
|
| 855 | {
|
| 856 | return __bpf_get_stack(regs, NULL, buf, size, flags, false );
|
| 857 | }
|
| 858 |
|
| 859 | const struct bpf_func_proto bpf_get_stack_proto = {
|
| 860 | .func = bpf_get_stack,
|
| 861 | .gpl_only = true,
|
| 862 | .ret_type = RET_INTEGER,
|
| 863 | .arg1_type = ARG_PTR_TO_CTX,
|
| 864 | .arg2_type = ARG_PTR_TO_UNINIT_MEM,
|
| 865 | .arg3_type = ARG_MEM_SIZE_OR_ZERO,
|
| 866 | .arg4_type = ARG_ANYTHING,
|
| 867 | };
|
| 868 |
|
| 869 | BPF_CALL_4(bpf_get_stack_sleepable, struct pt_regs *, regs, void *, buf, u32, size,
|
| 870 | u64, flags)
|
| 871 | {
|
| 872 | return __bpf_get_stack(regs, NULL, buf, size, flags, true );
|
| 873 | }
|
| 874 |
|
| 875 | const struct bpf_func_proto bpf_get_stack_sleepable_proto = {
|
| 876 | .func = bpf_get_stack_sleepable,
|
| 877 | .gpl_only = true,
|
| 878 | .might_sleep = true,
|
| 879 | .ret_type = RET_INTEGER,
|
| 880 | .arg1_type = ARG_PTR_TO_CTX,
|
| 881 | .arg2_type = ARG_PTR_TO_UNINIT_MEM,
|
| 882 | .arg3_type = ARG_MEM_SIZE_OR_ZERO,
|
| 883 | .arg4_type = ARG_ANYTHING,
|
| 884 | };
|
| 885 |
|
| 886 | static long __bpf_get_task_stack(struct task_struct *task, void *buf, u32 size,
|
| 887 | u64 flags, bool may_fault)
|
| 888 | {
|
| 889 | struct pt_regs *regs;
|
| 890 | long res = -EINVAL;
|
| 891 |
|
| 892 | if (!try_get_task_stack(task)) {
|
| 893 | memset(buf, 0, size);
|
| 894 | return -EFAULT;
|
| 895 | }
|
| 896 |
|
| 897 | regs = task_pt_regs(task);
|
| 898 | if (regs)
|
| 899 | res = __bpf_get_stack(regs, task, buf, size, flags, may_fault);
|
| 900 | else
|
| 901 | memset(buf, 0, size);
|
| 902 | put_task_stack(task);
|
| 903 | return res;
|
| 904 | }
|
| 905 |
|
| 906 | BPF_CALL_4(bpf_get_task_stack, struct task_struct *, task, void *, buf,
|
| 907 | u32, size, u64, flags)
|
| 908 | {
|
| 909 | return __bpf_get_task_stack(task, buf, size, flags, false );
|
| 910 | }
|
| 911 |
|
| 912 | const struct bpf_func_proto bpf_get_task_stack_proto = {
|
| 913 | .func = bpf_get_task_stack,
|
| 914 | .gpl_only = false,
|
| 915 | .ret_type = RET_INTEGER,
|
| 916 | .arg1_type = ARG_PTR_TO_BTF_ID,
|
| 917 | .arg1_btf_id = &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
|
| 918 | .arg2_type = ARG_PTR_TO_UNINIT_MEM,
|
| 919 | .arg3_type = ARG_MEM_SIZE_OR_ZERO,
|
| 920 | .arg4_type = ARG_ANYTHING,
|
| 921 | };
|
| 922 |
|
| 923 | BPF_CALL_4(bpf_get_task_stack_sleepable, struct task_struct *, task, void *, buf,
|
| 924 | u32, size, u64, flags)
|
| 925 | {
|
| 926 | return __bpf_get_task_stack(task, buf, size, flags, true );
|
| 927 | }
|
| 928 |
|
| 929 | const struct bpf_func_proto bpf_get_task_stack_sleepable_proto = {
|
| 930 | .func = bpf_get_task_stack_sleepable,
|
| 931 | .gpl_only = false,
|
| 932 | .might_sleep = true,
|
| 933 | .ret_type = RET_INTEGER,
|
| 934 | .arg1_type = ARG_PTR_TO_BTF_ID,
|
| 935 | .arg1_btf_id = &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
|
| 936 | .arg2_type = ARG_PTR_TO_UNINIT_MEM,
|
| 937 | .arg3_type = ARG_MEM_SIZE_OR_ZERO,
|
| 938 | .arg4_type = ARG_ANYTHING,
|
| 939 | };
|
| 940 |
|
| 941 | static int __bpf_get_stack_pe(const struct perf_callchain_entry *trace, u32 trace_nr,
|
| 942 | void *buf, u32 size, u64 flags)
|
| 943 | {
|
| 944 | bool user_build_id = flags & BPF_F_USER_BUILD_ID;
|
| 945 | u64 skip = flags & BPF_F_SKIP_FIELD_MASK;
|
| 946 | bool user = flags & BPF_F_USER_STACK;
|
| 947 | u32 elem_size, max_depth, nr_trace;
|
| 948 | bool kernel = !user;
|
| 949 |
|
| 950 | if (kernel && user_build_id)
|
| 951 | return -EINVAL;
|
| 952 |
|
| 953 | elem_size = user_build_id ? sizeof(struct bpf_stack_build_id) : sizeof(u64);
|
| 954 | if (unlikely(size % elem_size))
|
| 955 | return -EINVAL;
|
| 956 |
|
| 957 | max_depth = stack_map_calculate_max_depth(size, elem_size, flags);
|
| 958 | trace_nr = min_t(u32, trace_nr, max_depth);
|
| 959 |
|
| 960 | if (trace_nr < skip)
|
| 961 | return -EFAULT;
|
| 962 |
|
| 963 | nr_trace = callchain_store(trace, trace_nr, buf, elem_size, flags);
|
| 964 | return callchain_finalize(buf, size, nr_trace, elem_size, flags, false );
|
| 965 | }
|
| 966 |
|
| 967 | BPF_CALL_4(bpf_get_stack_pe, struct bpf_perf_event_data_kern *, ctx,
|
| 968 | void *, buf, u32, size, u64, flags)
|
| 969 | {
|
| 970 | struct pt_regs *regs = (struct pt_regs *)(ctx->regs);
|
| 971 | const struct perf_callchain_entry *trace;
|
| 972 | struct perf_event *event = ctx->event;
|
| 973 | bool kernel, user;
|
| 974 | int err = -EINVAL;
|
| 975 | __u64 nr_kernel;
|
| 976 |
|
| 977 | if (!(event->attr.sample_type & PERF_SAMPLE_CALLCHAIN))
|
| 978 | return __bpf_get_stack(regs, NULL, buf, size, flags, false );
|
| 979 |
|
| 980 | if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
|
| 981 | BPF_F_USER_BUILD_ID)))
|
| 982 | goto clear;
|
| 983 |
|
| 984 | user = flags & BPF_F_USER_STACK;
|
| 985 | kernel = !user;
|
| 986 |
|
| 987 | err = -EFAULT;
|
| 988 | trace = ctx->data->callchain;
|
| 989 | if (unlikely(!trace))
|
| 990 | goto clear;
|
| 991 |
|
| 992 | nr_kernel = count_kernel_ip(trace);
|
| 993 |
|
| 994 | if (kernel) {
|
| 995 | err = __bpf_get_stack_pe(trace, nr_kernel, buf, size, flags);
|
| 996 | } else {
|
| 997 | u64 skip = flags & BPF_F_SKIP_FIELD_MASK;
|
| 998 |
|
| 999 | skip += nr_kernel;
|
| 1000 | if (skip > BPF_F_SKIP_FIELD_MASK)
|
| 1001 | goto clear;
|
| 1002 | flags = (flags & ~BPF_F_SKIP_FIELD_MASK) | skip;
|
| 1003 | err = __bpf_get_stack_pe(trace, trace->nr, buf, size, flags);
|
| 1004 | }
|
| 1005 |
|
| 1006 | clear:
|
| 1007 | if (err < 0)
|
| 1008 | memset(buf, 0, size);
|
| 1009 | return err;
|
| 1010 |
|
| 1011 | }
|
| 1012 |
|
| 1013 | const struct bpf_func_proto bpf_get_stack_proto_pe = {
|
| 1014 | .func = bpf_get_stack_pe,
|
| 1015 | .gpl_only = true,
|
| 1016 | .ret_type = RET_INTEGER,
|
| 1017 | .arg1_type = ARG_PTR_TO_CTX,
|
| 1018 | .arg2_type = ARG_PTR_TO_UNINIT_MEM,
|
| 1019 | .arg3_type = ARG_MEM_SIZE_OR_ZERO,
|
| 1020 | .arg4_type = ARG_ANYTHING,
|
| 1021 | };
|
| 1022 |
|
| 1023 |
|
| 1024 | static void *stack_map_lookup_elem(struct bpf_map *map, void *key)
|
| 1025 | {
|
| 1026 | return ERR_PTR(-EOPNOTSUPP);
|
| 1027 | }
|
| 1028 |
|
| 1029 |
|
| 1030 | static int stack_map_lookup_and_delete_elem(struct bpf_map *map, void *key,
|
| 1031 | void *value, u64 flags)
|
| 1032 | {
|
| 1033 | return bpf_stackmap_extract(map, key, value, true);
|
| 1034 | }
|
| 1035 |
|
| 1036 |
|
| 1037 | int bpf_stackmap_extract(struct bpf_map *map, void *key, void *value,
|
| 1038 | bool delete)
|
| 1039 | {
|
| 1040 | struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
|
| 1041 | struct stack_map_bucket *bucket, *old_bucket;
|
| 1042 | u32 id = *(u32 *)key, trace_len;
|
| 1043 |
|
| 1044 | if (unlikely(id >= smap->n_buckets))
|
| 1045 | return -ENOENT;
|
| 1046 |
|
| 1047 | bucket = xchg(&smap->buckets[id], NULL);
|
| 1048 | if (!bucket)
|
| 1049 | return -ENOENT;
|
| 1050 |
|
| 1051 | trace_len = bucket->nr * stack_map_data_size(map);
|
| 1052 | memcpy(value, bucket->data, trace_len);
|
| 1053 | memset(value + trace_len, 0, map->value_size - trace_len);
|
| 1054 |
|
| 1055 | if (delete)
|
| 1056 | old_bucket = bucket;
|
| 1057 | else
|
| 1058 | old_bucket = xchg(&smap->buckets[id], bucket);
|
| 1059 | if (old_bucket)
|
| 1060 | pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
|
| 1061 | return 0;
|
| 1062 | }
|
| 1063 |
|
| 1064 | static int stack_map_get_next_key(struct bpf_map *map, void *key,
|
| 1065 | void *next_key)
|
| 1066 | {
|
| 1067 | struct bpf_stack_map *smap = container_of(map,
|
| 1068 | struct bpf_stack_map, map);
|
| 1069 | u32 id;
|
| 1070 |
|
| 1071 | WARN_ON_ONCE(!rcu_read_lock_held());
|
| 1072 |
|
| 1073 | if (!key) {
|
| 1074 | id = 0;
|
| 1075 | } else {
|
| 1076 | id = *(u32 *)key;
|
| 1077 | if (id >= smap->n_buckets || !smap->buckets[id])
|
| 1078 | id = 0;
|
| 1079 | else
|
| 1080 | id++;
|
| 1081 | }
|
| 1082 |
|
| 1083 | while (id < smap->n_buckets && !smap->buckets[id])
|
| 1084 | id++;
|
| 1085 |
|
| 1086 | if (id >= smap->n_buckets)
|
| 1087 | return -ENOENT;
|
| 1088 |
|
| 1089 | *(u32 *)next_key = id;
|
| 1090 | return 0;
|
| 1091 | }
|
| 1092 |
|
| 1093 | static long stack_map_update_elem(struct bpf_map *map, void *key, void *value,
|
| 1094 | u64 map_flags)
|
| 1095 | {
|
| 1096 | return -EINVAL;
|
| 1097 | }
|
| 1098 |
|
| 1099 |
|
| 1100 | static long stack_map_delete_elem(struct bpf_map *map, void *key)
|
| 1101 | {
|
| 1102 | struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
|
| 1103 | struct stack_map_bucket *old_bucket;
|
| 1104 | u32 id = *(u32 *)key;
|
| 1105 |
|
| 1106 | if (unlikely(id >= smap->n_buckets))
|
| 1107 | return -E2BIG;
|
| 1108 |
|
| 1109 | old_bucket = xchg(&smap->buckets[id], NULL);
|
| 1110 | if (old_bucket) {
|
| 1111 | pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
|
| 1112 | return 0;
|
| 1113 | } else {
|
| 1114 | return -ENOENT;
|
| 1115 | }
|
| 1116 | }
|
| 1117 |
|
| 1118 |
|
| 1119 | static void stack_map_free(struct bpf_map *map)
|
| 1120 | {
|
| 1121 | struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
|
| 1122 |
|
| 1123 | bpf_map_area_free(smap->elems);
|
| 1124 | pcpu_freelist_destroy(&smap->freelist);
|
| 1125 | bpf_map_area_free(smap);
|
| 1126 | put_callchain_buffers();
|
| 1127 | }
|
| 1128 |
|
| 1129 | static u64 stack_map_mem_usage(const struct bpf_map *map)
|
| 1130 | {
|
| 1131 | struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
|
| 1132 | u64 value_size = map->value_size;
|
| 1133 | u64 n_buckets = smap->n_buckets;
|
| 1134 | u64 enties = map->max_entries;
|
| 1135 | u64 usage = sizeof(*smap);
|
| 1136 |
|
| 1137 | usage += n_buckets * sizeof(struct stack_map_bucket *);
|
| 1138 | usage += enties * (sizeof(struct stack_map_bucket) + value_size);
|
| 1139 | return usage;
|
| 1140 | }
|
| 1141 |
|
| 1142 | BTF_ID_LIST_SINGLE(stack_trace_map_btf_ids, struct, bpf_stack_map)
|
| 1143 | const struct bpf_map_ops stack_trace_map_ops = {
|
| 1144 | .map_meta_equal = bpf_map_meta_equal,
|
| 1145 | .map_alloc = stack_map_alloc,
|
| 1146 | .map_free = stack_map_free,
|
| 1147 | .map_get_next_key = stack_map_get_next_key,
|
| 1148 | .map_lookup_elem = stack_map_lookup_elem,
|
| 1149 | .map_lookup_and_delete_elem = stack_map_lookup_and_delete_elem,
|
| 1150 | .map_update_elem = stack_map_update_elem,
|
| 1151 | .map_delete_elem = stack_map_delete_elem,
|
| 1152 | .map_check_btf = map_check_no_btf,
|
| 1153 | .map_mem_usage = stack_map_mem_usage,
|
| 1154 | .map_btf_id = &stack_trace_map_btf_ids[0],
|
| 1155 | };
|