kernel/bpf/hashtab.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
   3  * Copyright (c) 2016 Facebook
   4  */
   5 #include <linux/bpf.h>
   6 #include <linux/btf.h>
   7 #include <linux/jhash.h>
   8 #include <linux/filter.h>
   9 #include <linux/rculist_nulls.h>
  10 #include <linux/random.h>
  11 #include <uapi/linux/btf.h>
  12 #include <linux/rcupdate_trace.h>
  13 #include <linux/btf_ids.h>
  14 #include "percpu_freelist.h"
  15 #include "bpf_lru_list.h"
  16 #include "map_in_map.h"
  17
  18 #define HTAB_CREATE_FLAG_MASK                                           \
  19         (BPF_F_NO_PREALLOC | BPF_F_NO_COMMON_LRU | BPF_F_NUMA_NODE |    \
  20          BPF_F_ACCESS_MASK | BPF_F_ZERO_SEED)
  21
  22 #define BATCH_OPS(_name)                        \
  23         .map_lookup_batch =                     \
  24         _name##_map_lookup_batch,               \
  25         .map_lookup_and_delete_batch =          \
  26         _name##_map_lookup_and_delete_batch,    \
  27         .map_update_batch =                     \
  28         generic_map_update_batch,               \
  29         .map_delete_batch =                     \
  30         generic_map_delete_batch
  31
  32 /*
  33  * The bucket lock has two protection scopes:
  34  *
  35  * 1) Serializing concurrent operations from BPF programs on different
  36  *    CPUs
  37  *
  38  * 2) Serializing concurrent operations from BPF programs and sys_bpf()
  39  *
  40  * BPF programs can execute in any context including perf, kprobes and
  41  * tracing. As there are almost no limits where perf, kprobes and tracing
  42  * can be invoked from the lock operations need to be protected against
  43  * deadlocks. Deadlocks can be caused by recursion and by an invocation in
  44  * the lock held section when functions which acquire this lock are invoked
  45  * from sys_bpf(). BPF recursion is prevented by incrementing the per CPU
  46  * variable bpf_prog_active, which prevents BPF programs attached to perf
  47  * events, kprobes and tracing to be invoked before the prior invocation
  48  * from one of these contexts completed. sys_bpf() uses the same mechanism
  49  * by pinning the task to the current CPU and incrementing the recursion
  50  * protection across the map operation.
  51  *
  52  * This has subtle implications on PREEMPT_RT. PREEMPT_RT forbids certain
  53  * operations like memory allocations (even with GFP_ATOMIC) from atomic
  54  * contexts. This is required because even with GFP_ATOMIC the memory
  55  * allocator calls into code paths which acquire locks with long held lock
  56  * sections. To ensure the deterministic behaviour these locks are regular
  57  * spinlocks, which are converted to 'sleepable' spinlocks on RT. The only
  58  * true atomic contexts on an RT kernel are the low level hardware
  59  * handling, scheduling, low level interrupt handling, NMIs etc. None of
  60  * these contexts should ever do memory allocations.
  61  *
  62  * As regular device interrupt handlers and soft interrupts are forced into
  63  * thread context, the existing code which does
  64  *   spin_lock*(); alloc(GFP_ATOMIC); spin_unlock*();
  65  * just works.
  66  *
  67  * In theory the BPF locks could be converted to regular spinlocks as well,
  68  * but the bucket locks and percpu_freelist locks can be taken from
  69  * arbitrary contexts (perf, kprobes, tracepoints) which are required to be
  70  * atomic contexts even on RT. These mechanisms require preallocated maps,
  71  * so there is no need to invoke memory allocations within the lock held
  72  * sections.
  73  *
  74  * BPF maps which need dynamic allocation are only used from (forced)
  75  * thread context on RT and can therefore use regular spinlocks which in
  76  * turn allows to invoke memory allocations from the lock held section.
  77  *
  78  * On a non RT kernel this distinction is neither possible nor required.
  79  * spinlock maps to raw_spinlock and the extra code is optimized out by the
  80  * compiler.
  81  */
  82 struct bucket {
  83         struct hlist_nulls_head head;
  84         union {
  85                 raw_spinlock_t raw_lock;
  86                 spinlock_t     lock;
  87         };
  88 };
  89
  90 #define HASHTAB_MAP_LOCK_COUNT 8
  91 #define HASHTAB_MAP_LOCK_MASK (HASHTAB_MAP_LOCK_COUNT - 1)
  92
  93 struct bpf_htab {
  94         struct bpf_map map;
  95         struct bucket *buckets;
  96         void *elems;
  97         union {
  98                 struct pcpu_freelist freelist;
  99                 struct bpf_lru lru;
 100         };
 101         struct htab_elem *__percpu *extra_elems;
 102         atomic_t count; /* number of elements in this hashtable */
 103         u32 n_buckets;  /* number of hash buckets */
 104         u32 elem_size;  /* size of each element in bytes */
 105         u32 hashrnd;
 106         struct lock_class_key lockdep_key;
 107         int __percpu *map_locked[HASHTAB_MAP_LOCK_COUNT];
 108 };
 109
 110 /* each htab element is struct htab_elem + key + value */
 111 struct htab_elem {
 112         union {
 113                 struct hlist_nulls_node hash_node;
 114                 struct {
 115                         void *padding;
 116                         union {
 117                                 struct bpf_htab *htab;
 118                                 struct pcpu_freelist_node fnode;
 119                                 struct htab_elem *batch_flink;
 120                         };
 121                 };
 122         };
 123         union {
 124                 struct rcu_head rcu;
 125                 struct bpf_lru_node lru_node;
 126         };
 127         u32 hash;
 128         char key[] __aligned(8);
 129 };
 130
 131 static inline bool htab_is_prealloc(const struct bpf_htab *htab)
 132 {
 133         return !(htab->map.map_flags & BPF_F_NO_PREALLOC);
 134 }
 135
 136 static inline bool htab_use_raw_lock(const struct bpf_htab *htab)
 137 {
 138         return (!IS_ENABLED(CONFIG_PREEMPT_RT) || htab_is_prealloc(htab));
 139 }
 140
 141 static void htab_init_buckets(struct bpf_htab *htab)
 142 {
 143         unsigned int i;
 144
 145         for (i = 0; i < htab->n_buckets; i++) {
 146                 INIT_HLIST_NULLS_HEAD(&htab->buckets[i].head, i);
 147                 if (htab_use_raw_lock(htab)) {
 148                         raw_spin_lock_init(&htab->buckets[i].raw_lock);
 149                         lockdep_set_class(&htab->buckets[i].raw_lock,
 150                                           &htab->lockdep_key);
 151                 } else {
 152                         spin_lock_init(&htab->buckets[i].lock);
 153                         lockdep_set_class(&htab->buckets[i].lock,
 154                                           &htab->lockdep_key);
 155                 }
 156                 cond_resched();
 157         }
 158 }
 159
 160 static inline int htab_lock_bucket(const struct bpf_htab *htab,
 161                                    struct bucket *b, u32 hash,
 162                                    unsigned long *pflags)
 163 {
 164         unsigned long flags;
 165
 166         hash = hash & HASHTAB_MAP_LOCK_MASK;
 167
 168         migrate_disable();
 169         if (unlikely(__this_cpu_inc_return(*(htab->map_locked[hash])) != 1)) {
 170                 __this_cpu_dec(*(htab->map_locked[hash]));
 171                 migrate_enable();
 172                 return -EBUSY;
 173         }
 174
 175         if (htab_use_raw_lock(htab))
 176                 raw_spin_lock_irqsave(&b->raw_lock, flags);
 177         else
 178                 spin_lock_irqsave(&b->lock, flags);
 179         *pflags = flags;
 180
 181         return 0;
 182 }
 183
 184 static inline void htab_unlock_bucket(const struct bpf_htab *htab,
 185                                       struct bucket *b, u32 hash,
 186                                       unsigned long flags)
 187 {
 188         hash = hash & HASHTAB_MAP_LOCK_MASK;
 189         if (htab_use_raw_lock(htab))
 190                 raw_spin_unlock_irqrestore(&b->raw_lock, flags);
 191         else
 192                 spin_unlock_irqrestore(&b->lock, flags);
 193         __this_cpu_dec(*(htab->map_locked[hash]));
 194         migrate_enable();
 195 }
 196
 197 static bool htab_lru_map_delete_node(void *arg, struct bpf_lru_node *node);
 198
 199 static bool htab_is_lru(const struct bpf_htab *htab)
 200 {
 201         return htab->map.map_type == BPF_MAP_TYPE_LRU_HASH ||
 202                 htab->map.map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH;
 203 }
 204
 205 static bool htab_is_percpu(const struct bpf_htab *htab)
 206 {
 207         return htab->map.map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 208                 htab->map.map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH;
 209 }
 210
 211 static inline void htab_elem_set_ptr(struct htab_elem *l, u32 key_size,
 212                                      void __percpu *pptr)
 213 {
 214         *(void __percpu **)(l->key + key_size) = pptr;
 215 }
 216
 217 static inline void __percpu *htab_elem_get_ptr(struct htab_elem *l, u32 key_size)
 218 {
 219         return *(void __percpu **)(l->key + key_size);
 220 }
 221
 222 static void *fd_htab_map_get_ptr(const struct bpf_map *map, struct htab_elem *l)
 223 {
 224         return *(void **)(l->key + roundup(map->key_size, 8));
 225 }
 226
 227 static struct htab_elem *get_htab_elem(struct bpf_htab *htab, int i)
 228 {
 229         return (struct htab_elem *) (htab->elems + i * (u64)htab->elem_size);
 230 }
 231
 232 static bool htab_has_extra_elems(struct bpf_htab *htab)
 233 {
 234         return !htab_is_percpu(htab) && !htab_is_lru(htab);
 235 }
 236
 237 static void htab_free_prealloced_timers(struct bpf_htab *htab)
 238 {
 239         u32 num_entries = htab->map.max_entries;
 240         int i;
 241
 242         if (!map_value_has_timer(&htab->map))
 243                 return;
 244         if (htab_has_extra_elems(htab))
 245                 num_entries += num_possible_cpus();
 246
 247         for (i = 0; i < num_entries; i++) {
 248                 struct htab_elem *elem;
 249
 250                 elem = get_htab_elem(htab, i);
 251                 bpf_timer_cancel_and_free(elem->key +
 252                                           round_up(htab->map.key_size, 8) +
 253                                           htab->map.timer_off);
 254                 cond_resched();
 255         }
 256 }
 257
 258 static void htab_free_prealloced_kptrs(struct bpf_htab *htab)
 259 {
 260         u32 num_entries = htab->map.max_entries;
 261         int i;
 262
 263         if (!map_value_has_kptrs(&htab->map))
 264                 return;
 265         if (htab_has_extra_elems(htab))
 266                 num_entries += num_possible_cpus();
 267
 268         for (i = 0; i < num_entries; i++) {
 269                 struct htab_elem *elem;
 270
 271                 elem = get_htab_elem(htab, i);
 272                 bpf_map_free_kptrs(&htab->map, elem->key + round_up(htab->map.key_size, 8));
 273                 cond_resched();
 274         }
 275 }
 276
 277 static void htab_free_elems(struct bpf_htab *htab)
 278 {
 279         int i;
 280
 281         if (!htab_is_percpu(htab))
 282                 goto free_elems;
 283
 284         for (i = 0; i < htab->map.max_entries; i++) {
 285                 void __percpu *pptr;
 286
 287                 pptr = htab_elem_get_ptr(get_htab_elem(htab, i),
 288                                          htab->map.key_size);
 289                 free_percpu(pptr);
 290                 cond_resched();
 291         }
 292 free_elems:
 293         bpf_map_area_free(htab->elems);
 294 }
 295
 296 /* The LRU list has a lock (lru_lock). Each htab bucket has a lock
 297  * (bucket_lock). If both locks need to be acquired together, the lock
 298  * order is always lru_lock -> bucket_lock and this only happens in
 299  * bpf_lru_list.c logic. For example, certain code path of
 300  * bpf_lru_pop_free(), which is called by function prealloc_lru_pop(),
 301  * will acquire lru_lock first followed by acquiring bucket_lock.
 302  *
 303  * In hashtab.c, to avoid deadlock, lock acquisition of
 304  * bucket_lock followed by lru_lock is not allowed. In such cases,
 305  * bucket_lock needs to be released first before acquiring lru_lock.
 306  */
 307 static struct htab_elem *prealloc_lru_pop(struct bpf_htab *htab, void *key,
 308                                           u32 hash)
 309 {
 310         struct bpf_lru_node *node = bpf_lru_pop_free(&htab->lru, hash);
 311         struct htab_elem *l;
 312
 313         if (node) {
 314                 u32 key_size = htab->map.key_size;
 315
 316                 l = container_of(node, struct htab_elem, lru_node);
 317                 memcpy(l->key, key, key_size);
 318                 check_and_init_map_value(&htab->map,
 319                                          l->key + round_up(key_size, 8));
 320                 return l;
 321         }
 322
 323         return NULL;
 324 }
 325
 326 static int prealloc_init(struct bpf_htab *htab)
 327 {
 328         u32 num_entries = htab->map.max_entries;
 329         int err = -ENOMEM, i;
 330
 331         if (htab_has_extra_elems(htab))
 332                 num_entries += num_possible_cpus();
 333
 334         htab->elems = bpf_map_area_alloc((u64)htab->elem_size * num_entries,
 335                                          htab->map.numa_node);
 336         if (!htab->elems)
 337                 return -ENOMEM;
 338
 339         if (!htab_is_percpu(htab))
 340                 goto skip_percpu_elems;
 341
 342         for (i = 0; i < num_entries; i++) {
 343                 u32 size = round_up(htab->map.value_size, 8);
 344                 void __percpu *pptr;
 345
 346                 pptr = bpf_map_alloc_percpu(&htab->map, size, 8,
 347                                             GFP_USER | __GFP_NOWARN);
 348                 if (!pptr)
 349                         goto free_elems;
 350                 htab_elem_set_ptr(get_htab_elem(htab, i), htab->map.key_size,
 351                                   pptr);
 352                 cond_resched();
 353         }
 354
 355 skip_percpu_elems:
 356         if (htab_is_lru(htab))
 357                 err = bpf_lru_init(&htab->lru,
 358                                    htab->map.map_flags & BPF_F_NO_COMMON_LRU,
 359                                    offsetof(struct htab_elem, hash) -
 360                                    offsetof(struct htab_elem, lru_node),
 361                                    htab_lru_map_delete_node,
 362                                    htab);
 363         else
 364                 err = pcpu_freelist_init(&htab->freelist);
 365
 366         if (err)
 367                 goto free_elems;
 368
 369         if (htab_is_lru(htab))
 370                 bpf_lru_populate(&htab->lru, htab->elems,
 371                                  offsetof(struct htab_elem, lru_node),
 372                                  htab->elem_size, num_entries);
 373         else
 374                 pcpu_freelist_populate(&htab->freelist,
 375                                        htab->elems + offsetof(struct htab_elem, fnode),
 376                                        htab->elem_size, num_entries);
 377
 378         return 0;
 379
 380 free_elems:
 381         htab_free_elems(htab);
 382         return err;
 383 }
 384
 385 static void prealloc_destroy(struct bpf_htab *htab)
 386 {
 387         htab_free_elems(htab);
 388
 389         if (htab_is_lru(htab))
 390                 bpf_lru_destroy(&htab->lru);
 391         else
 392                 pcpu_freelist_destroy(&htab->freelist);
 393 }
 394
 395 static int alloc_extra_elems(struct bpf_htab *htab)
 396 {
 397         struct htab_elem *__percpu *pptr, *l_new;
 398         struct pcpu_freelist_node *l;
 399         int cpu;
 400
 401         pptr = bpf_map_alloc_percpu(&htab->map, sizeof(struct htab_elem *), 8,
 402                                     GFP_USER | __GFP_NOWARN);
 403         if (!pptr)
 404                 return -ENOMEM;
 405
 406         for_each_possible_cpu(cpu) {
 407                 l = pcpu_freelist_pop(&htab->freelist);
 408                 /* pop will succeed, since prealloc_init()
 409                  * preallocated extra num_possible_cpus elements
 410                  */
 411                 l_new = container_of(l, struct htab_elem, fnode);
 412                 *per_cpu_ptr(pptr, cpu) = l_new;
 413         }
 414         htab->extra_elems = pptr;
 415         return 0;
 416 }
 417
 418 /* Called from syscall */
 419 static int htab_map_alloc_check(union bpf_attr *attr)
 420 {
 421         bool percpu = (attr->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 422                        attr->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH);
 423         bool lru = (attr->map_type == BPF_MAP_TYPE_LRU_HASH ||
 424                     attr->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH);
 425         /* percpu_lru means each cpu has its own LRU list.
 426          * it is different from BPF_MAP_TYPE_PERCPU_HASH where
 427          * the map's value itself is percpu.  percpu_lru has
 428          * nothing to do with the map's value.
 429          */
 430         bool percpu_lru = (attr->map_flags & BPF_F_NO_COMMON_LRU);
 431         bool prealloc = !(attr->map_flags & BPF_F_NO_PREALLOC);
 432         bool zero_seed = (attr->map_flags & BPF_F_ZERO_SEED);
 433         int numa_node = bpf_map_attr_numa_node(attr);
 434
 435         BUILD_BUG_ON(offsetof(struct htab_elem, htab) !=
 436                      offsetof(struct htab_elem, hash_node.pprev));
 437         BUILD_BUG_ON(offsetof(struct htab_elem, fnode.next) !=
 438                      offsetof(struct htab_elem, hash_node.pprev));
 439
 440         if (lru && !bpf_capable())
 441                 /* LRU implementation is much complicated than other
 442                  * maps.  Hence, limit to CAP_BPF.
 443                  */
 444                 return -EPERM;
 445
 446         if (zero_seed && !capable(CAP_SYS_ADMIN))
 447                 /* Guard against local DoS, and discourage production use. */
 448                 return -EPERM;
 449
 450         if (attr->map_flags & ~HTAB_CREATE_FLAG_MASK ||
 451             !bpf_map_flags_access_ok(attr->map_flags))
 452                 return -EINVAL;
 453
 454         if (!lru && percpu_lru)
 455                 return -EINVAL;
 456
 457         if (lru && !prealloc)
 458                 return -ENOTSUPP;
 459
 460         if (numa_node != NUMA_NO_NODE && (percpu || percpu_lru))
 461                 return -EINVAL;
 462
 463         /* check sanity of attributes.
 464          * value_size == 0 may be allowed in the future to use map as a set
 465          */
 466         if (attr->max_entries == 0 || attr->key_size == 0 ||
 467             attr->value_size == 0)
 468                 return -EINVAL;
 469
 470         if ((u64)attr->key_size + attr->value_size >= KMALLOC_MAX_SIZE -
 471            sizeof(struct htab_elem))
 472                 /* if key_size + value_size is bigger, the user space won't be
 473                  * able to access the elements via bpf syscall. This check
 474                  * also makes sure that the elem_size doesn't overflow and it's
 475                  * kmalloc-able later in htab_map_update_elem()
 476                  */
 477                 return -E2BIG;
 478
 479         return 0;
 480 }
 481
 482 static struct bpf_map *htab_map_alloc(union bpf_attr *attr)
 483 {
 484         bool percpu = (attr->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 485                        attr->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH);
 486         bool lru = (attr->map_type == BPF_MAP_TYPE_LRU_HASH ||
 487                     attr->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH);
 488         /* percpu_lru means each cpu has its own LRU list.
 489          * it is different from BPF_MAP_TYPE_PERCPU_HASH where
 490          * the map's value itself is percpu.  percpu_lru has
 491          * nothing to do with the map's value.
 492          */
 493         bool percpu_lru = (attr->map_flags & BPF_F_NO_COMMON_LRU);
 494         bool prealloc = !(attr->map_flags & BPF_F_NO_PREALLOC);
 495         struct bpf_htab *htab;
 496         int err, i;
 497
 498         htab = kzalloc(sizeof(*htab), GFP_USER | __GFP_ACCOUNT);
 499         if (!htab)
 500                 return ERR_PTR(-ENOMEM);
 501
 502         lockdep_register_key(&htab->lockdep_key);
 503
 504         bpf_map_init_from_attr(&htab->map, attr);
 505
 506         if (percpu_lru) {
 507                 /* ensure each CPU's lru list has >=1 elements.
 508                  * since we are at it, make each lru list has the same
 509                  * number of elements.
 510                  */
 511                 htab->map.max_entries = roundup(attr->max_entries,
 512                                                 num_possible_cpus());
 513                 if (htab->map.max_entries < attr->max_entries)
 514                         htab->map.max_entries = rounddown(attr->max_entries,
 515                                                           num_possible_cpus());
 516         }
 517
 518         /* hash table size must be power of 2 */
 519         htab->n_buckets = roundup_pow_of_two(htab->map.max_entries);
 520
 521         htab->elem_size = sizeof(struct htab_elem) +
 522                           round_up(htab->map.key_size, 8);
 523         if (percpu)
 524                 htab->elem_size += sizeof(void *);
 525         else
 526                 htab->elem_size += round_up(htab->map.value_size, 8);
 527
 528         err = -E2BIG;
 529         /* prevent zero size kmalloc and check for u32 overflow */
 530         if (htab->n_buckets == 0 ||
 531             htab->n_buckets > U32_MAX / sizeof(struct bucket))
 532                 goto free_htab;
 533
 534         err = -ENOMEM;
 535         htab->buckets = bpf_map_area_alloc(htab->n_buckets *
 536                                            sizeof(struct bucket),
 537                                            htab->map.numa_node);
 538         if (!htab->buckets)
 539                 goto free_htab;
 540
 541         for (i = 0; i < HASHTAB_MAP_LOCK_COUNT; i++) {
 542                 htab->map_locked[i] = bpf_map_alloc_percpu(&htab->map,
 543                                                            sizeof(int),
 544                                                            sizeof(int),
 545                                                            GFP_USER);
 546                 if (!htab->map_locked[i])
 547                         goto free_map_locked;
 548         }
 549
 550         if (htab->map.map_flags & BPF_F_ZERO_SEED)
 551                 htab->hashrnd = 0;
 552         else
 553                 htab->hashrnd = get_random_int();
 554
 555         htab_init_buckets(htab);
 556
 557         if (prealloc) {
 558                 err = prealloc_init(htab);
 559                 if (err)
 560                         goto free_map_locked;
 561
 562                 if (!percpu && !lru) {
 563                         /* lru itself can remove the least used element, so
 564                          * there is no need for an extra elem during map_update.
 565                          */
 566                         err = alloc_extra_elems(htab);
 567                         if (err)
 568                                 goto free_prealloc;
 569                 }
 570         }
 571
 572         return &htab->map;
 573
 574 free_prealloc:
 575         prealloc_destroy(htab);
 576 free_map_locked:
 577         for (i = 0; i < HASHTAB_MAP_LOCK_COUNT; i++)
 578                 free_percpu(htab->map_locked[i]);
 579         bpf_map_area_free(htab->buckets);
 580 free_htab:
 581         lockdep_unregister_key(&htab->lockdep_key);
 582         kfree(htab);
 583         return ERR_PTR(err);
 584 }
 585
 586 static inline u32 htab_map_hash(const void *key, u32 key_len, u32 hashrnd)
 587 {
 588         return jhash(key, key_len, hashrnd);
 589 }
 590
 591 static inline struct bucket *__select_bucket(struct bpf_htab *htab, u32 hash)
 592 {
 593         return &htab->buckets[hash & (htab->n_buckets - 1)];
 594 }
 595
 596 static inline struct hlist_nulls_head *select_bucket(struct bpf_htab *htab, u32 hash)
 597 {
 598         return &__select_bucket(htab, hash)->head;
 599 }
 600
 601 /* this lookup function can only be called with bucket lock taken */
 602 static struct htab_elem *lookup_elem_raw(struct hlist_nulls_head *head, u32 hash,
 603                                          void *key, u32 key_size)
 604 {
 605         struct hlist_nulls_node *n;
 606         struct htab_elem *l;
 607
 608         hlist_nulls_for_each_entry_rcu(l, n, head, hash_node)
 609                 if (l->hash == hash && !memcmp(&l->key, key, key_size))
 610                         return l;
 611
 612         return NULL;
 613 }
 614
 615 /* can be called without bucket lock. it will repeat the loop in
 616  * the unlikely event when elements moved from one bucket into another
 617  * while link list is being walked
 618  */
 619 static struct htab_elem *lookup_nulls_elem_raw(struct hlist_nulls_head *head,
 620                                                u32 hash, void *key,
 621                                                u32 key_size, u32 n_buckets)
 622 {
 623         struct hlist_nulls_node *n;
 624         struct htab_elem *l;
 625
 626 again:
 627         hlist_nulls_for_each_entry_rcu(l, n, head, hash_node)
 628                 if (l->hash == hash && !memcmp(&l->key, key, key_size))
 629                         return l;
 630
 631         if (unlikely(get_nulls_value(n) != (hash & (n_buckets - 1))))
 632                 goto again;
 633
 634         return NULL;
 635 }
 636
 637 /* Called from syscall or from eBPF program directly, so
 638  * arguments have to match bpf_map_lookup_elem() exactly.
 639  * The return value is adjusted by BPF instructions
 640  * in htab_map_gen_lookup().
 641  */
 642 static void *__htab_map_lookup_elem(struct bpf_map *map, void *key)
 643 {
 644         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
 645         struct hlist_nulls_head *head;
 646         struct htab_elem *l;
 647         u32 hash, key_size;
 648
 649         WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
 650                      !rcu_read_lock_bh_held());
 651
 652         key_size = map->key_size;
 653
 654         hash = htab_map_hash(key, key_size, htab->hashrnd);
 655
 656         head = select_bucket(htab, hash);
 657
 658         l = lookup_nulls_elem_raw(head, hash, key, key_size, htab->n_buckets);
 659
 660         return l;
 661 }
 662
 663 static void *htab_map_lookup_elem(struct bpf_map *map, void *key)
 664 {
 665         struct htab_elem *l = __htab_map_lookup_elem(map, key);
 666
 667         if (l)
 668                 return l->key + round_up(map->key_size, 8);
 669
 670         return NULL;
 671 }
 672
 673 /* inline bpf_map_lookup_elem() call.
 674  * Instead of:
 675  * bpf_prog
 676  *   bpf_map_lookup_elem
 677  *     map->ops->map_lookup_elem
 678  *       htab_map_lookup_elem
 679  *         __htab_map_lookup_elem
 680  * do:
 681  * bpf_prog
 682  *   __htab_map_lookup_elem
 683  */
 684 static int htab_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
 685 {
 686         struct bpf_insn *insn = insn_buf;
 687         const int ret = BPF_REG_0;
 688
 689         BUILD_BUG_ON(!__same_type(&__htab_map_lookup_elem,
 690                      (void *(*)(struct bpf_map *map, void *key))NULL));
 691         *insn++ = BPF_EMIT_CALL(__htab_map_lookup_elem);
 692         *insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 1);
 693         *insn++ = BPF_ALU64_IMM(BPF_ADD, ret,
 694                                 offsetof(struct htab_elem, key) +
 695                                 round_up(map->key_size, 8));
 696         return insn - insn_buf;
 697 }
 698
 699 static __always_inline void *__htab_lru_map_lookup_elem(struct bpf_map *map,
 700                                                         void *key, const bool mark)
 701 {
 702         struct htab_elem *l = __htab_map_lookup_elem(map, key);
 703
 704         if (l) {
 705                 if (mark)
 706                         bpf_lru_node_set_ref(&l->lru_node);
 707                 return l->key + round_up(map->key_size, 8);
 708         }
 709
 710         return NULL;
 711 }
 712
 713 static void *htab_lru_map_lookup_elem(struct bpf_map *map, void *key)
 714 {
 715         return __htab_lru_map_lookup_elem(map, key, true);
 716 }
 717
 718 static void *htab_lru_map_lookup_elem_sys(struct bpf_map *map, void *key)
 719 {
 720         return __htab_lru_map_lookup_elem(map, key, false);
 721 }
 722
 723 static int htab_lru_map_gen_lookup(struct bpf_map *map,
 724                                    struct bpf_insn *insn_buf)
 725 {
 726         struct bpf_insn *insn = insn_buf;
 727         const int ret = BPF_REG_0;
 728         const int ref_reg = BPF_REG_1;
 729
 730         BUILD_BUG_ON(!__same_type(&__htab_map_lookup_elem,
 731                      (void *(*)(struct bpf_map *map, void *key))NULL));
 732         *insn++ = BPF_EMIT_CALL(__htab_map_lookup_elem);
 733         *insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 4);
 734         *insn++ = BPF_LDX_MEM(BPF_B, ref_reg, ret,
 735                               offsetof(struct htab_elem, lru_node) +
 736                               offsetof(struct bpf_lru_node, ref));
 737         *insn++ = BPF_JMP_IMM(BPF_JNE, ref_reg, 0, 1);
 738         *insn++ = BPF_ST_MEM(BPF_B, ret,
 739                              offsetof(struct htab_elem, lru_node) +
 740                              offsetof(struct bpf_lru_node, ref),
 741                              1);
 742         *insn++ = BPF_ALU64_IMM(BPF_ADD, ret,
 743                                 offsetof(struct htab_elem, key) +
 744                                 round_up(map->key_size, 8));
 745         return insn - insn_buf;
 746 }
 747
 748 static void check_and_free_fields(struct bpf_htab *htab,
 749                                   struct htab_elem *elem)
 750 {
 751         void *map_value = elem->key + round_up(htab->map.key_size, 8);
 752
 753         if (map_value_has_timer(&htab->map))
 754                 bpf_timer_cancel_and_free(map_value + htab->map.timer_off);
 755         if (map_value_has_kptrs(&htab->map))
 756                 bpf_map_free_kptrs(&htab->map, map_value);
 757 }
 758
 759 /* It is called from the bpf_lru_list when the LRU needs to delete
 760  * older elements from the htab.
 761  */
 762 static bool htab_lru_map_delete_node(void *arg, struct bpf_lru_node *node)
 763 {
 764         struct bpf_htab *htab = arg;
 765         struct htab_elem *l = NULL, *tgt_l;
 766         struct hlist_nulls_head *head;
 767         struct hlist_nulls_node *n;
 768         unsigned long flags;
 769         struct bucket *b;
 770         int ret;
 771
 772         tgt_l = container_of(node, struct htab_elem, lru_node);
 773         b = __select_bucket(htab, tgt_l->hash);
 774         head = &b->head;
 775
 776         ret = htab_lock_bucket(htab, b, tgt_l->hash, &flags);
 777         if (ret)
 778                 return false;
 779
 780         hlist_nulls_for_each_entry_rcu(l, n, head, hash_node)
 781                 if (l == tgt_l) {
 782                         hlist_nulls_del_rcu(&l->hash_node);
 783                         check_and_free_fields(htab, l);
 784                         break;
 785                 }
 786
 787         htab_unlock_bucket(htab, b, tgt_l->hash, flags);
 788
 789         return l == tgt_l;
 790 }
 791
 792 /* Called from syscall */
 793 static int htab_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
 794 {
 795         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
 796         struct hlist_nulls_head *head;
 797         struct htab_elem *l, *next_l;
 798         u32 hash, key_size;
 799         int i = 0;
 800
 801         WARN_ON_ONCE(!rcu_read_lock_held());
 802
 803         key_size = map->key_size;
 804
 805         if (!key)
 806                 goto find_first_elem;
 807
 808         hash = htab_map_hash(key, key_size, htab->hashrnd);
 809
 810         head = select_bucket(htab, hash);
 811
 812         /* lookup the key */
 813         l = lookup_nulls_elem_raw(head, hash, key, key_size, htab->n_buckets);
 814
 815         if (!l)
 816                 goto find_first_elem;
 817
 818         /* key was found, get next key in the same bucket */
 819         next_l = hlist_nulls_entry_safe(rcu_dereference_raw(hlist_nulls_next_rcu(&l->hash_node)),
 820                                   struct htab_elem, hash_node);
 821
 822         if (next_l) {
 823                 /* if next elem in this hash list is non-zero, just return it */
 824                 memcpy(next_key, next_l->key, key_size);
 825                 return 0;
 826         }
 827
 828         /* no more elements in this hash list, go to the next bucket */
 829         i = hash & (htab->n_buckets - 1);
 830         i++;
 831
 832 find_first_elem:
 833         /* iterate over buckets */
 834         for (; i < htab->n_buckets; i++) {
 835                 head = select_bucket(htab, i);
 836
 837                 /* pick first element in the bucket */
 838                 next_l = hlist_nulls_entry_safe(rcu_dereference_raw(hlist_nulls_first_rcu(head)),
 839                                           struct htab_elem, hash_node);
 840                 if (next_l) {
 841                         /* if it's not empty, just return it */
 842                         memcpy(next_key, next_l->key, key_size);
 843                         return 0;
 844                 }
 845         }
 846
 847         /* iterated over all buckets and all elements */
 848         return -ENOENT;
 849 }
 850
 851 static void htab_elem_free(struct bpf_htab *htab, struct htab_elem *l)
 852 {
 853         if (htab->map.map_type == BPF_MAP_TYPE_PERCPU_HASH)
 854                 free_percpu(htab_elem_get_ptr(l, htab->map.key_size));
 855         check_and_free_fields(htab, l);
 856         kfree(l);
 857 }
 858
 859 static void htab_elem_free_rcu(struct rcu_head *head)
 860 {
 861         struct htab_elem *l = container_of(head, struct htab_elem, rcu);
 862         struct bpf_htab *htab = l->htab;
 863
 864         htab_elem_free(htab, l);
 865 }
 866
 867 static void htab_put_fd_value(struct bpf_htab *htab, struct htab_elem *l)
 868 {
 869         struct bpf_map *map = &htab->map;
 870         void *ptr;
 871
 872         if (map->ops->map_fd_put_ptr) {
 873                 ptr = fd_htab_map_get_ptr(map, l);
 874                 map->ops->map_fd_put_ptr(ptr);
 875         }
 876 }
 877
 878 static void free_htab_elem(struct bpf_htab *htab, struct htab_elem *l)
 879 {
 880         htab_put_fd_value(htab, l);
 881
 882         if (htab_is_prealloc(htab)) {
 883                 check_and_free_fields(htab, l);
 884                 __pcpu_freelist_push(&htab->freelist, &l->fnode);
 885         } else {
 886                 atomic_dec(&htab->count);
 887                 l->htab = htab;
 888                 call_rcu(&l->rcu, htab_elem_free_rcu);
 889         }
 890 }
 891
 892 static void pcpu_copy_value(struct bpf_htab *htab, void __percpu *pptr,
 893                             void *value, bool onallcpus)
 894 {
 895         if (!onallcpus) {
 896                 /* copy true value_size bytes */
 897                 memcpy(this_cpu_ptr(pptr), value, htab->map.value_size);
 898         } else {
 899                 u32 size = round_up(htab->map.value_size, 8);
 900                 int off = 0, cpu;
 901
 902                 for_each_possible_cpu(cpu) {
 903                         bpf_long_memcpy(per_cpu_ptr(pptr, cpu),
 904                                         value + off, size);
 905                         off += size;
 906                 }
 907         }
 908 }
 909
 910 static void pcpu_init_value(struct bpf_htab *htab, void __percpu *pptr,
 911                             void *value, bool onallcpus)
 912 {
 913         /* When using prealloc and not setting the initial value on all cpus,
 914          * zero-fill element values for other cpus (just as what happens when
 915          * not using prealloc). Otherwise, bpf program has no way to ensure
 916          * known initial values for cpus other than current one
 917          * (onallcpus=false always when coming from bpf prog).
 918          */
 919         if (htab_is_prealloc(htab) && !onallcpus) {
 920                 u32 size = round_up(htab->map.value_size, 8);
 921                 int current_cpu = raw_smp_processor_id();
 922                 int cpu;
 923
 924                 for_each_possible_cpu(cpu) {
 925                         if (cpu == current_cpu)
 926                                 bpf_long_memcpy(per_cpu_ptr(pptr, cpu), value,
 927                                                 size);
 928                         else
 929                                 memset(per_cpu_ptr(pptr, cpu), 0, size);
 930                 }
 931         } else {
 932                 pcpu_copy_value(htab, pptr, value, onallcpus);
 933         }
 934 }
 935
 936 static bool fd_htab_map_needs_adjust(const struct bpf_htab *htab)
 937 {
 938         return htab->map.map_type == BPF_MAP_TYPE_HASH_OF_MAPS &&
 939                BITS_PER_LONG == 64;
 940 }
 941
 942 static struct htab_elem *alloc_htab_elem(struct bpf_htab *htab, void *key,
 943                                          void *value, u32 key_size, u32 hash,
 944                                          bool percpu, bool onallcpus,
 945                                          struct htab_elem *old_elem)
 946 {
 947         u32 size = htab->map.value_size;
 948         bool prealloc = htab_is_prealloc(htab);
 949         struct htab_elem *l_new, **pl_new;
 950         void __percpu *pptr;
 951
 952         if (prealloc) {
 953                 if (old_elem) {
 954                         /* if we're updating the existing element,
 955                          * use per-cpu extra elems to avoid freelist_pop/push
 956                          */
 957                         pl_new = this_cpu_ptr(htab->extra_elems);
 958                         l_new = *pl_new;
 959                         htab_put_fd_value(htab, old_elem);
 960                         *pl_new = old_elem;
 961                 } else {
 962                         struct pcpu_freelist_node *l;
 963
 964                         l = __pcpu_freelist_pop(&htab->freelist);
 965                         if (!l)
 966                                 return ERR_PTR(-E2BIG);
 967                         l_new = container_of(l, struct htab_elem, fnode);
 968                 }
 969         } else {
 970                 if (atomic_inc_return(&htab->count) > htab->map.max_entries)
 971                         if (!old_elem) {
 972                                 /* when map is full and update() is replacing
 973                                  * old element, it's ok to allocate, since
 974                                  * old element will be freed immediately.
 975                                  * Otherwise return an error
 976                                  */
 977                                 l_new = ERR_PTR(-E2BIG);
 978                                 goto dec_count;
 979                         }
 980                 l_new = bpf_map_kmalloc_node(&htab->map, htab->elem_size,
 981                                              GFP_NOWAIT | __GFP_NOWARN,
 982                                              htab->map.numa_node);
 983                 if (!l_new) {
 984                         l_new = ERR_PTR(-ENOMEM);
 985                         goto dec_count;
 986                 }
 987                 check_and_init_map_value(&htab->map,
 988                                          l_new->key + round_up(key_size, 8));
 989         }
 990
 991         memcpy(l_new->key, key, key_size);
 992         if (percpu) {
 993                 size = round_up(size, 8);
 994                 if (prealloc) {
 995                         pptr = htab_elem_get_ptr(l_new, key_size);
 996                 } else {
 997                         /* alloc_percpu zero-fills */
 998                         pptr = bpf_map_alloc_percpu(&htab->map, size, 8,
 999                                                     GFP_NOWAIT | __GFP_NOWARN);
1000                         if (!pptr) {
1001                                 kfree(l_new);
1002                                 l_new = ERR_PTR(-ENOMEM);
1003                                 goto dec_count;
1004                         }
1005                 }
1006
1007                 pcpu_init_value(htab, pptr, value, onallcpus);
1008
1009                 if (!prealloc)
1010                         htab_elem_set_ptr(l_new, key_size, pptr);
1011         } else if (fd_htab_map_needs_adjust(htab)) {
1012                 size = round_up(size, 8);
1013                 memcpy(l_new->key + round_up(key_size, 8), value, size);
1014         } else {
1015                 copy_map_value(&htab->map,
1016                                l_new->key + round_up(key_size, 8),
1017                                value);
1018         }
1019
1020         l_new->hash = hash;
1021         return l_new;
1022 dec_count:
1023         atomic_dec(&htab->count);
1024         return l_new;
1025 }
1026
1027 static int check_flags(struct bpf_htab *htab, struct htab_elem *l_old,
1028                        u64 map_flags)
1029 {
1030         if (l_old && (map_flags & ~BPF_F_LOCK) == BPF_NOEXIST)
1031                 /* elem already exists */
1032                 return -EEXIST;
1033
1034         if (!l_old && (map_flags & ~BPF_F_LOCK) == BPF_EXIST)
1035                 /* elem doesn't exist, cannot update it */
1036                 return -ENOENT;
1037
1038         return 0;
1039 }
1040
1041 /* Called from syscall or from eBPF program */
1042 static int htab_map_update_elem(struct bpf_map *map, void *key, void *value,
1043                                 u64 map_flags)
1044 {
1045         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1046         struct htab_elem *l_new = NULL, *l_old;
1047         struct hlist_nulls_head *head;
1048         unsigned long flags;
1049         struct bucket *b;
1050         u32 key_size, hash;
1051         int ret;
1052
1053         if (unlikely((map_flags & ~BPF_F_LOCK) > BPF_EXIST))
1054                 /* unknown flags */
1055                 return -EINVAL;
1056
1057         WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1058                      !rcu_read_lock_bh_held());
1059
1060         key_size = map->key_size;
1061
1062         hash = htab_map_hash(key, key_size, htab->hashrnd);
1063
1064         b = __select_bucket(htab, hash);
1065         head = &b->head;
1066
1067         if (unlikely(map_flags & BPF_F_LOCK)) {
1068                 if (unlikely(!map_value_has_spin_lock(map)))
1069                         return -EINVAL;
1070                 /* find an element without taking the bucket lock */
1071                 l_old = lookup_nulls_elem_raw(head, hash, key, key_size,
1072                                               htab->n_buckets);
1073                 ret = check_flags(htab, l_old, map_flags);
1074                 if (ret)
1075                         return ret;
1076                 if (l_old) {
1077                         /* grab the element lock and update value in place */
1078                         copy_map_value_locked(map,
1079                                               l_old->key + round_up(key_size, 8),
1080                                               value, false);
1081                         return 0;
1082                 }
1083                 /* fall through, grab the bucket lock and lookup again.
1084                  * 99.9% chance that the element won't be found,
1085                  * but second lookup under lock has to be done.
1086                  */
1087         }
1088
1089         ret = htab_lock_bucket(htab, b, hash, &flags);
1090         if (ret)
1091                 return ret;
1092
1093         l_old = lookup_elem_raw(head, hash, key, key_size);
1094
1095         ret = check_flags(htab, l_old, map_flags);
1096         if (ret)
1097                 goto err;
1098
1099         if (unlikely(l_old && (map_flags & BPF_F_LOCK))) {
1100                 /* first lookup without the bucket lock didn't find the element,
1101                  * but second lookup with the bucket lock found it.
1102                  * This case is highly unlikely, but has to be dealt with:
1103                  * grab the element lock in addition to the bucket lock
1104                  * and update element in place
1105                  */
1106                 copy_map_value_locked(map,
1107                                       l_old->key + round_up(key_size, 8),
1108                                       value, false);
1109                 ret = 0;
1110                 goto err;
1111         }
1112
1113         l_new = alloc_htab_elem(htab, key, value, key_size, hash, false, false,
1114                                 l_old);
1115         if (IS_ERR(l_new)) {
1116                 /* all pre-allocated elements are in use or memory exhausted */
1117                 ret = PTR_ERR(l_new);
1118                 goto err;
1119         }
1120
1121         /* add new element to the head of the list, so that
1122          * concurrent search will find it before old elem
1123          */
1124         hlist_nulls_add_head_rcu(&l_new->hash_node, head);
1125         if (l_old) {
1126                 hlist_nulls_del_rcu(&l_old->hash_node);
1127                 if (!htab_is_prealloc(htab))
1128                         free_htab_elem(htab, l_old);
1129                 else
1130                         check_and_free_fields(htab, l_old);
1131         }
1132         ret = 0;
1133 err:
1134         htab_unlock_bucket(htab, b, hash, flags);
1135         return ret;
1136 }
1137
1138 static void htab_lru_push_free(struct bpf_htab *htab, struct htab_elem *elem)
1139 {
1140         check_and_free_fields(htab, elem);
1141         bpf_lru_push_free(&htab->lru, &elem->lru_node);
1142 }
1143
1144 static int htab_lru_map_update_elem(struct bpf_map *map, void *key, void *value,
1145                                     u64 map_flags)
1146 {
1147         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1148         struct htab_elem *l_new, *l_old = NULL;
1149         struct hlist_nulls_head *head;
1150         unsigned long flags;
1151         struct bucket *b;
1152         u32 key_size, hash;
1153         int ret;
1154
1155         if (unlikely(map_flags > BPF_EXIST))
1156                 /* unknown flags */
1157                 return -EINVAL;
1158
1159         WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1160                      !rcu_read_lock_bh_held());
1161
1162         key_size = map->key_size;
1163
1164         hash = htab_map_hash(key, key_size, htab->hashrnd);
1165
1166         b = __select_bucket(htab, hash);
1167         head = &b->head;
1168
1169         /* For LRU, we need to alloc before taking bucket's
1170          * spinlock because getting free nodes from LRU may need
1171          * to remove older elements from htab and this removal
1172          * operation will need a bucket lock.
1173          */
1174         l_new = prealloc_lru_pop(htab, key, hash);
1175         if (!l_new)
1176                 return -ENOMEM;
1177         copy_map_value(&htab->map,
1178                        l_new->key + round_up(map->key_size, 8), value);
1179
1180         ret = htab_lock_bucket(htab, b, hash, &flags);
1181         if (ret)
1182                 return ret;
1183
1184         l_old = lookup_elem_raw(head, hash, key, key_size);
1185
1186         ret = check_flags(htab, l_old, map_flags);
1187         if (ret)
1188                 goto err;
1189
1190         /* add new element to the head of the list, so that
1191          * concurrent search will find it before old elem
1192          */
1193         hlist_nulls_add_head_rcu(&l_new->hash_node, head);
1194         if (l_old) {
1195                 bpf_lru_node_set_ref(&l_new->lru_node);
1196                 hlist_nulls_del_rcu(&l_old->hash_node);
1197         }
1198         ret = 0;
1199
1200 err:
1201         htab_unlock_bucket(htab, b, hash, flags);
1202
1203         if (ret)
1204                 htab_lru_push_free(htab, l_new);
1205         else if (l_old)
1206                 htab_lru_push_free(htab, l_old);
1207
1208         return ret;
1209 }
1210
1211 static int __htab_percpu_map_update_elem(struct bpf_map *map, void *key,
1212                                          void *value, u64 map_flags,
1213                                          bool onallcpus)
1214 {
1215         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1216         struct htab_elem *l_new = NULL, *l_old;
1217         struct hlist_nulls_head *head;
1218         unsigned long flags;
1219         struct bucket *b;
1220         u32 key_size, hash;
1221         int ret;
1222
1223         if (unlikely(map_flags > BPF_EXIST))
1224                 /* unknown flags */
1225                 return -EINVAL;
1226
1227         WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1228                      !rcu_read_lock_bh_held());
1229
1230         key_size = map->key_size;
1231
1232         hash = htab_map_hash(key, key_size, htab->hashrnd);
1233
1234         b = __select_bucket(htab, hash);
1235         head = &b->head;
1236
1237         ret = htab_lock_bucket(htab, b, hash, &flags);
1238         if (ret)
1239                 return ret;
1240
1241         l_old = lookup_elem_raw(head, hash, key, key_size);
1242
1243         ret = check_flags(htab, l_old, map_flags);
1244         if (ret)
1245                 goto err;
1246
1247         if (l_old) {
1248                 /* per-cpu hash map can update value in-place */
1249                 pcpu_copy_value(htab, htab_elem_get_ptr(l_old, key_size),
1250                                 value, onallcpus);
1251         } else {
1252                 l_new = alloc_htab_elem(htab, key, value, key_size,
1253                                         hash, true, onallcpus, NULL);
1254                 if (IS_ERR(l_new)) {
1255                         ret = PTR_ERR(l_new);
1256                         goto err;
1257                 }
1258                 hlist_nulls_add_head_rcu(&l_new->hash_node, head);
1259         }
1260         ret = 0;
1261 err:
1262         htab_unlock_bucket(htab, b, hash, flags);
1263         return ret;
1264 }
1265
1266 static int __htab_lru_percpu_map_update_elem(struct bpf_map *map, void *key,
1267                                              void *value, u64 map_flags,
1268                                              bool onallcpus)
1269 {
1270         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1271         struct htab_elem *l_new = NULL, *l_old;
1272         struct hlist_nulls_head *head;
1273         unsigned long flags;
1274         struct bucket *b;
1275         u32 key_size, hash;
1276         int ret;
1277
1278         if (unlikely(map_flags > BPF_EXIST))
1279                 /* unknown flags */
1280                 return -EINVAL;
1281
1282         WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1283                      !rcu_read_lock_bh_held());
1284
1285         key_size = map->key_size;
1286
1287         hash = htab_map_hash(key, key_size, htab->hashrnd);
1288
1289         b = __select_bucket(htab, hash);
1290         head = &b->head;
1291
1292         /* For LRU, we need to alloc before taking bucket's
1293          * spinlock because LRU's elem alloc may need
1294          * to remove older elem from htab and this removal
1295          * operation will need a bucket lock.
1296          */
1297         if (map_flags != BPF_EXIST) {
1298                 l_new = prealloc_lru_pop(htab, key, hash);
1299                 if (!l_new)
1300                         return -ENOMEM;
1301         }
1302
1303         ret = htab_lock_bucket(htab, b, hash, &flags);
1304         if (ret)
1305                 return ret;
1306
1307         l_old = lookup_elem_raw(head, hash, key, key_size);
1308
1309         ret = check_flags(htab, l_old, map_flags);
1310         if (ret)
1311                 goto err;
1312
1313         if (l_old) {
1314                 bpf_lru_node_set_ref(&l_old->lru_node);
1315
1316                 /* per-cpu hash map can update value in-place */
1317                 pcpu_copy_value(htab, htab_elem_get_ptr(l_old, key_size),
1318                                 value, onallcpus);
1319         } else {
1320                 pcpu_init_value(htab, htab_elem_get_ptr(l_new, key_size),
1321                                 value, onallcpus);
1322                 hlist_nulls_add_head_rcu(&l_new->hash_node, head);
1323                 l_new = NULL;
1324         }
1325         ret = 0;
1326 err:
1327         htab_unlock_bucket(htab, b, hash, flags);
1328         if (l_new)
1329                 bpf_lru_push_free(&htab->lru, &l_new->lru_node);
1330         return ret;
1331 }
1332
1333 static int htab_percpu_map_update_elem(struct bpf_map *map, void *key,
1334                                        void *value, u64 map_flags)
1335 {
1336         return __htab_percpu_map_update_elem(map, key, value, map_flags, false);
1337 }
1338
1339 static int htab_lru_percpu_map_update_elem(struct bpf_map *map, void *key,
1340                                            void *value, u64 map_flags)
1341 {
1342         return __htab_lru_percpu_map_update_elem(map, key, value, map_flags,
1343                                                  false);
1344 }
1345
1346 /* Called from syscall or from eBPF program */
1347 static int htab_map_delete_elem(struct bpf_map *map, void *key)
1348 {
1349         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1350         struct hlist_nulls_head *head;
1351         struct bucket *b;
1352         struct htab_elem *l;
1353         unsigned long flags;
1354         u32 hash, key_size;
1355         int ret;
1356
1357         WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1358                      !rcu_read_lock_bh_held());
1359
1360         key_size = map->key_size;
1361
1362         hash = htab_map_hash(key, key_size, htab->hashrnd);
1363         b = __select_bucket(htab, hash);
1364         head = &b->head;
1365
1366         ret = htab_lock_bucket(htab, b, hash, &flags);
1367         if (ret)
1368                 return ret;
1369
1370         l = lookup_elem_raw(head, hash, key, key_size);
1371
1372         if (l) {
1373                 hlist_nulls_del_rcu(&l->hash_node);
1374                 free_htab_elem(htab, l);
1375         } else {
1376                 ret = -ENOENT;
1377         }
1378
1379         htab_unlock_bucket(htab, b, hash, flags);
1380         return ret;
1381 }
1382
1383 static int htab_lru_map_delete_elem(struct bpf_map *map, void *key)
1384 {
1385         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1386         struct hlist_nulls_head *head;
1387         struct bucket *b;
1388         struct htab_elem *l;
1389         unsigned long flags;
1390         u32 hash, key_size;
1391         int ret;
1392
1393         WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1394                      !rcu_read_lock_bh_held());
1395
1396         key_size = map->key_size;
1397
1398         hash = htab_map_hash(key, key_size, htab->hashrnd);
1399         b = __select_bucket(htab, hash);
1400         head = &b->head;
1401
1402         ret = htab_lock_bucket(htab, b, hash, &flags);
1403         if (ret)
1404                 return ret;
1405
1406         l = lookup_elem_raw(head, hash, key, key_size);
1407
1408         if (l)
1409                 hlist_nulls_del_rcu(&l->hash_node);
1410         else
1411                 ret = -ENOENT;
1412
1413         htab_unlock_bucket(htab, b, hash, flags);
1414         if (l)
1415                 htab_lru_push_free(htab, l);
1416         return ret;
1417 }
1418
1419 static void delete_all_elements(struct bpf_htab *htab)
1420 {
1421         int i;
1422
1423         for (i = 0; i < htab->n_buckets; i++) {
1424                 struct hlist_nulls_head *head = select_bucket(htab, i);
1425                 struct hlist_nulls_node *n;
1426                 struct htab_elem *l;
1427
1428                 hlist_nulls_for_each_entry_safe(l, n, head, hash_node) {
1429                         hlist_nulls_del_rcu(&l->hash_node);
1430                         htab_elem_free(htab, l);
1431                 }
1432         }
1433 }
1434
1435 static void htab_free_malloced_timers(struct bpf_htab *htab)
1436 {
1437         int i;
1438
1439         rcu_read_lock();
1440         for (i = 0; i < htab->n_buckets; i++) {
1441                 struct hlist_nulls_head *head = select_bucket(htab, i);
1442                 struct hlist_nulls_node *n;
1443                 struct htab_elem *l;
1444
1445                 hlist_nulls_for_each_entry(l, n, head, hash_node) {
1446                         /* We don't reset or free kptr on uref dropping to zero,
1447                          * hence just free timer.
1448                          */
1449                         bpf_timer_cancel_and_free(l->key +
1450                                                   round_up(htab->map.key_size, 8) +
1451                                                   htab->map.timer_off);
1452                 }
1453                 cond_resched_rcu();
1454         }
1455         rcu_read_unlock();
1456 }
1457
1458 static void htab_map_free_timers(struct bpf_map *map)
1459 {
1460         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1461
1462         /* We don't reset or free kptr on uref dropping to zero. */
1463         if (!map_value_has_timer(&htab->map))
1464                 return;
1465         if (!htab_is_prealloc(htab))
1466                 htab_free_malloced_timers(htab);
1467         else
1468                 htab_free_prealloced_timers(htab);
1469 }
1470
1471 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
1472 static void htab_map_free(struct bpf_map *map)
1473 {
1474         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1475         int i;
1476
1477         /* bpf_free_used_maps() or close(map_fd) will trigger this map_free callback.
1478          * bpf_free_used_maps() is called after bpf prog is no longer executing.
1479          * There is no need to synchronize_rcu() here to protect map elements.
1480          */
1481
1482         /* some of free_htab_elem() callbacks for elements of this map may
1483          * not have executed. Wait for them.
1484          */
1485         rcu_barrier();
1486         if (!htab_is_prealloc(htab)) {
1487                 delete_all_elements(htab);
1488         } else {
1489                 htab_free_prealloced_kptrs(htab);
1490                 prealloc_destroy(htab);
1491         }
1492
1493         bpf_map_free_kptr_off_tab(map);
1494         free_percpu(htab->extra_elems);
1495         bpf_map_area_free(htab->buckets);
1496         for (i = 0; i < HASHTAB_MAP_LOCK_COUNT; i++)
1497                 free_percpu(htab->map_locked[i]);
1498         lockdep_unregister_key(&htab->lockdep_key);
1499         kfree(htab);
1500 }
1501
1502 static void htab_map_seq_show_elem(struct bpf_map *map, void *key,
1503                                    struct seq_file *m)
1504 {
1505         void *value;
1506
1507         rcu_read_lock();
1508
1509         value = htab_map_lookup_elem(map, key);
1510         if (!value) {
1511                 rcu_read_unlock();
1512                 return;
1513         }
1514
1515         btf_type_seq_show(map->btf, map->btf_key_type_id, key, m);
1516         seq_puts(m, ": ");
1517         btf_type_seq_show(map->btf, map->btf_value_type_id, value, m);
1518         seq_puts(m, "\n");
1519
1520         rcu_read_unlock();
1521 }
1522
1523 static int __htab_map_lookup_and_delete_elem(struct bpf_map *map, void *key,
1524                                              void *value, bool is_lru_map,
1525                                              bool is_percpu, u64 flags)
1526 {
1527         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1528         struct hlist_nulls_head *head;
1529         unsigned long bflags;
1530         struct htab_elem *l;
1531         u32 hash, key_size;
1532         struct bucket *b;
1533         int ret;
1534
1535         key_size = map->key_size;
1536
1537         hash = htab_map_hash(key, key_size, htab->hashrnd);
1538         b = __select_bucket(htab, hash);
1539         head = &b->head;
1540
1541         ret = htab_lock_bucket(htab, b, hash, &bflags);
1542         if (ret)
1543                 return ret;
1544
1545         l = lookup_elem_raw(head, hash, key, key_size);
1546         if (!l) {
1547                 ret = -ENOENT;
1548         } else {
1549                 if (is_percpu) {
1550                         u32 roundup_value_size = round_up(map->value_size, 8);
1551                         void __percpu *pptr;
1552                         int off = 0, cpu;
1553
1554                         pptr = htab_elem_get_ptr(l, key_size);
1555                         for_each_possible_cpu(cpu) {
1556                                 bpf_long_memcpy(value + off,
1557                                                 per_cpu_ptr(pptr, cpu),
1558                                                 roundup_value_size);
1559                                 off += roundup_value_size;
1560                         }
1561                 } else {
1562                         u32 roundup_key_size = round_up(map->key_size, 8);
1563
1564                         if (flags & BPF_F_LOCK)
1565                                 copy_map_value_locked(map, value, l->key +
1566                                                       roundup_key_size,
1567                                                       true);
1568                         else
1569                                 copy_map_value(map, value, l->key +
1570                                                roundup_key_size);
1571                         check_and_init_map_value(map, value);
1572                 }
1573
1574                 hlist_nulls_del_rcu(&l->hash_node);
1575                 if (!is_lru_map)
1576                         free_htab_elem(htab, l);
1577         }
1578
1579         htab_unlock_bucket(htab, b, hash, bflags);
1580
1581         if (is_lru_map && l)
1582                 htab_lru_push_free(htab, l);
1583
1584         return ret;
1585 }
1586
1587 static int htab_map_lookup_and_delete_elem(struct bpf_map *map, void *key,
1588                                            void *value, u64 flags)
1589 {
1590         return __htab_map_lookup_and_delete_elem(map, key, value, false, false,
1591                                                  flags);
1592 }
1593
1594 static int htab_percpu_map_lookup_and_delete_elem(struct bpf_map *map,
1595                                                   void *key, void *value,
1596                                                   u64 flags)
1597 {
1598         return __htab_map_lookup_and_delete_elem(map, key, value, false, true,
1599                                                  flags);
1600 }
1601
1602 static int htab_lru_map_lookup_and_delete_elem(struct bpf_map *map, void *key,
1603                                                void *value, u64 flags)
1604 {
1605         return __htab_map_lookup_and_delete_elem(map, key, value, true, false,
1606                                                  flags);
1607 }
1608
1609 static int htab_lru_percpu_map_lookup_and_delete_elem(struct bpf_map *map,
1610                                                       void *key, void *value,
1611                                                       u64 flags)
1612 {
1613         return __htab_map_lookup_and_delete_elem(map, key, value, true, true,
1614                                                  flags);
1615 }
1616
1617 static int
1618 __htab_map_lookup_and_delete_batch(struct bpf_map *map,
1619                                    const union bpf_attr *attr,
1620                                    union bpf_attr __user *uattr,
1621                                    bool do_delete, bool is_lru_map,
1622                                    bool is_percpu)
1623 {
1624         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1625         u32 bucket_cnt, total, key_size, value_size, roundup_key_size;
1626         void *keys = NULL, *values = NULL, *value, *dst_key, *dst_val;
1627         void __user *uvalues = u64_to_user_ptr(attr->batch.values);
1628         void __user *ukeys = u64_to_user_ptr(attr->batch.keys);
1629         void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1630         u32 batch, max_count, size, bucket_size, map_id;
1631         struct htab_elem *node_to_free = NULL;
1632         u64 elem_map_flags, map_flags;
1633         struct hlist_nulls_head *head;
1634         struct hlist_nulls_node *n;
1635         unsigned long flags = 0;
1636         bool locked = false;
1637         struct htab_elem *l;
1638         struct bucket *b;
1639         int ret = 0;
1640
1641         elem_map_flags = attr->batch.elem_flags;
1642         if ((elem_map_flags & ~BPF_F_LOCK) ||
1643             ((elem_map_flags & BPF_F_LOCK) && !map_value_has_spin_lock(map)))
1644                 return -EINVAL;
1645
1646         map_flags = attr->batch.flags;
1647         if (map_flags)
1648                 return -EINVAL;
1649
1650         max_count = attr->batch.count;
1651         if (!max_count)
1652                 return 0;
1653
1654         if (put_user(0, &uattr->batch.count))
1655                 return -EFAULT;
1656
1657         batch = 0;
1658         if (ubatch && copy_from_user(&batch, ubatch, sizeof(batch)))
1659                 return -EFAULT;
1660
1661         if (batch >= htab->n_buckets)
1662                 return -ENOENT;
1663
1664         key_size = htab->map.key_size;
1665         roundup_key_size = round_up(htab->map.key_size, 8);
1666         value_size = htab->map.value_size;
1667         size = round_up(value_size, 8);
1668         if (is_percpu)
1669                 value_size = size * num_possible_cpus();
1670         total = 0;
1671         /* while experimenting with hash tables with sizes ranging from 10 to
1672          * 1000, it was observed that a bucket can have up to 5 entries.
1673          */
1674         bucket_size = 5;
1675
1676 alloc:
1677         /* We cannot do copy_from_user or copy_to_user inside
1678          * the rcu_read_lock. Allocate enough space here.
1679          */
1680         keys = kvmalloc_array(key_size, bucket_size, GFP_USER | __GFP_NOWARN);
1681         values = kvmalloc_array(value_size, bucket_size, GFP_USER | __GFP_NOWARN);
1682         if (!keys || !values) {
1683                 ret = -ENOMEM;
1684                 goto after_loop;
1685         }
1686
1687 again:
1688         bpf_disable_instrumentation();
1689         rcu_read_lock();
1690 again_nocopy:
1691         dst_key = keys;
1692         dst_val = values;
1693         b = &htab->buckets[batch];
1694         head = &b->head;
1695         /* do not grab the lock unless need it (bucket_cnt > 0). */
1696         if (locked) {
1697                 ret = htab_lock_bucket(htab, b, batch, &flags);
1698                 if (ret)
1699                         goto next_batch;
1700         }
1701
1702         bucket_cnt = 0;
1703         hlist_nulls_for_each_entry_rcu(l, n, head, hash_node)
1704                 bucket_cnt++;
1705
1706         if (bucket_cnt && !locked) {
1707                 locked = true;
1708                 goto again_nocopy;
1709         }
1710
1711         if (bucket_cnt > (max_count - total)) {
1712                 if (total == 0)
1713                         ret = -ENOSPC;
1714                 /* Note that since bucket_cnt > 0 here, it is implicit
1715                  * that the locked was grabbed, so release it.
1716                  */
1717                 htab_unlock_bucket(htab, b, batch, flags);
1718                 rcu_read_unlock();
1719                 bpf_enable_instrumentation();
1720                 goto after_loop;
1721         }
1722
1723         if (bucket_cnt > bucket_size) {
1724                 bucket_size = bucket_cnt;
1725                 /* Note that since bucket_cnt > 0 here, it is implicit
1726                  * that the locked was grabbed, so release it.
1727                  */
1728                 htab_unlock_bucket(htab, b, batch, flags);
1729                 rcu_read_unlock();
1730                 bpf_enable_instrumentation();
1731                 kvfree(keys);
1732                 kvfree(values);
1733                 goto alloc;
1734         }
1735
1736         /* Next block is only safe to run if you have grabbed the lock */
1737         if (!locked)
1738                 goto next_batch;
1739
1740         hlist_nulls_for_each_entry_safe(l, n, head, hash_node) {
1741                 memcpy(dst_key, l->key, key_size);
1742
1743                 if (is_percpu) {
1744                         int off = 0, cpu;
1745                         void __percpu *pptr;
1746
1747                         pptr = htab_elem_get_ptr(l, map->key_size);
1748                         for_each_possible_cpu(cpu) {
1749                                 bpf_long_memcpy(dst_val + off,
1750                                                 per_cpu_ptr(pptr, cpu), size);
1751                                 off += size;
1752                         }
1753                 } else {
1754                         value = l->key + roundup_key_size;
1755                         if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
1756                                 struct bpf_map **inner_map = value;
1757
1758                                  /* Actual value is the id of the inner map */
1759                                 map_id = map->ops->map_fd_sys_lookup_elem(*inner_map);
1760                                 value = &map_id;
1761                         }
1762
1763                         if (elem_map_flags & BPF_F_LOCK)
1764                                 copy_map_value_locked(map, dst_val, value,
1765                                                       true);
1766                         else
1767                                 copy_map_value(map, dst_val, value);
1768                         check_and_init_map_value(map, dst_val);
1769                 }
1770                 if (do_delete) {
1771                         hlist_nulls_del_rcu(&l->hash_node);
1772
1773                         /* bpf_lru_push_free() will acquire lru_lock, which
1774                          * may cause deadlock. See comments in function
1775                          * prealloc_lru_pop(). Let us do bpf_lru_push_free()
1776                          * after releasing the bucket lock.
1777                          */
1778                         if (is_lru_map) {
1779                                 l->batch_flink = node_to_free;
1780                                 node_to_free = l;
1781                         } else {
1782                                 free_htab_elem(htab, l);
1783                         }
1784                 }
1785                 dst_key += key_size;
1786                 dst_val += value_size;
1787         }
1788
1789         htab_unlock_bucket(htab, b, batch, flags);
1790         locked = false;
1791
1792         while (node_to_free) {
1793                 l = node_to_free;
1794                 node_to_free = node_to_free->batch_flink;
1795                 htab_lru_push_free(htab, l);
1796         }
1797
1798 next_batch:
1799         /* If we are not copying data, we can go to next bucket and avoid
1800          * unlocking the rcu.
1801          */
1802         if (!bucket_cnt && (batch + 1 < htab->n_buckets)) {
1803                 batch++;
1804                 goto again_nocopy;
1805         }
1806
1807         rcu_read_unlock();
1808         bpf_enable_instrumentation();
1809         if (bucket_cnt && (copy_to_user(ukeys + total * key_size, keys,
1810             key_size * bucket_cnt) ||
1811             copy_to_user(uvalues + total * value_size, values,
1812             value_size * bucket_cnt))) {
1813                 ret = -EFAULT;
1814                 goto after_loop;
1815         }
1816
1817         total += bucket_cnt;
1818         batch++;
1819         if (batch >= htab->n_buckets) {
1820                 ret = -ENOENT;
1821                 goto after_loop;
1822         }
1823         goto again;
1824
1825 after_loop:
1826         if (ret == -EFAULT)
1827                 goto out;
1828
1829         /* copy # of entries and next batch */
1830         ubatch = u64_to_user_ptr(attr->batch.out_batch);
1831         if (copy_to_user(ubatch, &batch, sizeof(batch)) ||
1832             put_user(total, &uattr->batch.count))
1833                 ret = -EFAULT;
1834
1835 out:
1836         kvfree(keys);
1837         kvfree(values);
1838         return ret;
1839 }
1840
1841 static int
1842 htab_percpu_map_lookup_batch(struct bpf_map *map, const union bpf_attr *attr,
1843                              union bpf_attr __user *uattr)
1844 {
1845         return __htab_map_lookup_and_delete_batch(map, attr, uattr, false,
1846                                                   false, true);
1847 }
1848
1849 static int
1850 htab_percpu_map_lookup_and_delete_batch(struct bpf_map *map,
1851                                         const union bpf_attr *attr,
1852                                         union bpf_attr __user *uattr)
1853 {
1854         return __htab_map_lookup_and_delete_batch(map, attr, uattr, true,
1855                                                   false, true);
1856 }
1857
1858 static int
1859 htab_map_lookup_batch(struct bpf_map *map, const union bpf_attr *attr,
1860                       union bpf_attr __user *uattr)
1861 {
1862         return __htab_map_lookup_and_delete_batch(map, attr, uattr, false,
1863                                                   false, false);
1864 }
1865
1866 static int
1867 htab_map_lookup_and_delete_batch(struct bpf_map *map,
1868                                  const union bpf_attr *attr,
1869                                  union bpf_attr __user *uattr)
1870 {
1871         return __htab_map_lookup_and_delete_batch(map, attr, uattr, true,
1872                                                   false, false);
1873 }
1874
1875 static int
1876 htab_lru_percpu_map_lookup_batch(struct bpf_map *map,
1877                                  const union bpf_attr *attr,
1878                                  union bpf_attr __user *uattr)
1879 {
1880         return __htab_map_lookup_and_delete_batch(map, attr, uattr, false,
1881                                                   true, true);
1882 }
1883
1884 static int
1885 htab_lru_percpu_map_lookup_and_delete_batch(struct bpf_map *map,
1886                                             const union bpf_attr *attr,
1887                                             union bpf_attr __user *uattr)
1888 {
1889         return __htab_map_lookup_and_delete_batch(map, attr, uattr, true,
1890                                                   true, true);
1891 }
1892
1893 static int
1894 htab_lru_map_lookup_batch(struct bpf_map *map, const union bpf_attr *attr,
1895                           union bpf_attr __user *uattr)
1896 {
1897         return __htab_map_lookup_and_delete_batch(map, attr, uattr, false,
1898                                                   true, false);
1899 }
1900
1901 static int
1902 htab_lru_map_lookup_and_delete_batch(struct bpf_map *map,
1903                                      const union bpf_attr *attr,
1904                                      union bpf_attr __user *uattr)
1905 {
1906         return __htab_map_lookup_and_delete_batch(map, attr, uattr, true,
1907                                                   true, false);
1908 }
1909
1910 struct bpf_iter_seq_hash_map_info {
1911         struct bpf_map *map;
1912         struct bpf_htab *htab;
1913         void *percpu_value_buf; // non-zero means percpu hash
1914         u32 bucket_id;
1915         u32 skip_elems;
1916 };
1917
1918 static struct htab_elem *
1919 bpf_hash_map_seq_find_next(struct bpf_iter_seq_hash_map_info *info,
1920                            struct htab_elem *prev_elem)
1921 {
1922         const struct bpf_htab *htab = info->htab;
1923         u32 skip_elems = info->skip_elems;
1924         u32 bucket_id = info->bucket_id;
1925         struct hlist_nulls_head *head;
1926         struct hlist_nulls_node *n;
1927         struct htab_elem *elem;
1928         struct bucket *b;
1929         u32 i, count;
1930
1931         if (bucket_id >= htab->n_buckets)
1932                 return NULL;
1933
1934         /* try to find next elem in the same bucket */
1935         if (prev_elem) {
1936                 /* no update/deletion on this bucket, prev_elem should be still valid
1937                  * and we won't skip elements.
1938                  */
1939                 n = rcu_dereference_raw(hlist_nulls_next_rcu(&prev_elem->hash_node));
1940                 elem = hlist_nulls_entry_safe(n, struct htab_elem, hash_node);
1941                 if (elem)
1942                         return elem;
1943
1944                 /* not found, unlock and go to the next bucket */
1945                 b = &htab->buckets[bucket_id++];
1946                 rcu_read_unlock();
1947                 skip_elems = 0;
1948         }
1949
1950         for (i = bucket_id; i < htab->n_buckets; i++) {
1951                 b = &htab->buckets[i];
1952                 rcu_read_lock();
1953
1954                 count = 0;
1955                 head = &b->head;
1956                 hlist_nulls_for_each_entry_rcu(elem, n, head, hash_node) {
1957                         if (count >= skip_elems) {
1958                                 info->bucket_id = i;
1959                                 info->skip_elems = count;
1960                                 return elem;
1961                         }
1962                         count++;
1963                 }
1964
1965                 rcu_read_unlock();
1966                 skip_elems = 0;
1967         }
1968
1969         info->bucket_id = i;
1970         info->skip_elems = 0;
1971         return NULL;
1972 }
1973
1974 static void *bpf_hash_map_seq_start(struct seq_file *seq, loff_t *pos)
1975 {
1976         struct bpf_iter_seq_hash_map_info *info = seq->private;
1977         struct htab_elem *elem;
1978
1979         elem = bpf_hash_map_seq_find_next(info, NULL);
1980         if (!elem)
1981                 return NULL;
1982
1983         if (*pos == 0)
1984                 ++*pos;
1985         return elem;
1986 }
1987
1988 static void *bpf_hash_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1989 {
1990         struct bpf_iter_seq_hash_map_info *info = seq->private;
1991
1992         ++*pos;
1993         ++info->skip_elems;
1994         return bpf_hash_map_seq_find_next(info, v);
1995 }
1996
1997 static int __bpf_hash_map_seq_show(struct seq_file *seq, struct htab_elem *elem)
1998 {
1999         struct bpf_iter_seq_hash_map_info *info = seq->private;
2000         u32 roundup_key_size, roundup_value_size;
2001         struct bpf_iter__bpf_map_elem ctx = {};
2002         struct bpf_map *map = info->map;
2003         struct bpf_iter_meta meta;
2004         int ret = 0, off = 0, cpu;
2005         struct bpf_prog *prog;
2006         void __percpu *pptr;
2007
2008         meta.seq = seq;
2009         prog = bpf_iter_get_info(&meta, elem == NULL);
2010         if (prog) {
2011                 ctx.meta = &meta;
2012                 ctx.map = info->map;
2013                 if (elem) {
2014                         roundup_key_size = round_up(map->key_size, 8);
2015                         ctx.key = elem->key;
2016                         if (!info->percpu_value_buf) {
2017                                 ctx.value = elem->key + roundup_key_size;
2018                         } else {
2019                                 roundup_value_size = round_up(map->value_size, 8);
2020                                 pptr = htab_elem_get_ptr(elem, map->key_size);
2021                                 for_each_possible_cpu(cpu) {
2022                                         bpf_long_memcpy(info->percpu_value_buf + off,
2023                                                         per_cpu_ptr(pptr, cpu),
2024                                                         roundup_value_size);
2025                                         off += roundup_value_size;
2026                                 }
2027                                 ctx.value = info->percpu_value_buf;
2028                         }
2029                 }
2030                 ret = bpf_iter_run_prog(prog, &ctx);
2031         }
2032
2033         return ret;
2034 }
2035
2036 static int bpf_hash_map_seq_show(struct seq_file *seq, void *v)
2037 {
2038         return __bpf_hash_map_seq_show(seq, v);
2039 }
2040
2041 static void bpf_hash_map_seq_stop(struct seq_file *seq, void *v)
2042 {
2043         if (!v)
2044                 (void)__bpf_hash_map_seq_show(seq, NULL);
2045         else
2046                 rcu_read_unlock();
2047 }
2048
2049 static int bpf_iter_init_hash_map(void *priv_data,
2050                                   struct bpf_iter_aux_info *aux)
2051 {
2052         struct bpf_iter_seq_hash_map_info *seq_info = priv_data;
2053         struct bpf_map *map = aux->map;
2054         void *value_buf;
2055         u32 buf_size;
2056
2057         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
2058             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
2059                 buf_size = round_up(map->value_size, 8) * num_possible_cpus();
2060                 value_buf = kmalloc(buf_size, GFP_USER | __GFP_NOWARN);
2061                 if (!value_buf)
2062                         return -ENOMEM;
2063
2064                 seq_info->percpu_value_buf = value_buf;
2065         }
2066
2067         seq_info->map = map;
2068         seq_info->htab = container_of(map, struct bpf_htab, map);
2069         return 0;
2070 }
2071
2072 static void bpf_iter_fini_hash_map(void *priv_data)
2073 {
2074         struct bpf_iter_seq_hash_map_info *seq_info = priv_data;
2075
2076         kfree(seq_info->percpu_value_buf);
2077 }
2078
2079 static const struct seq_operations bpf_hash_map_seq_ops = {
2080         .start  = bpf_hash_map_seq_start,
2081         .next   = bpf_hash_map_seq_next,
2082         .stop   = bpf_hash_map_seq_stop,
2083         .show   = bpf_hash_map_seq_show,
2084 };
2085
2086 static const struct bpf_iter_seq_info iter_seq_info = {
2087         .seq_ops                = &bpf_hash_map_seq_ops,
2088         .init_seq_private       = bpf_iter_init_hash_map,
2089         .fini_seq_private       = bpf_iter_fini_hash_map,
2090         .seq_priv_size          = sizeof(struct bpf_iter_seq_hash_map_info),
2091 };
2092
2093 static int bpf_for_each_hash_elem(struct bpf_map *map, bpf_callback_t callback_fn,
2094                                   void *callback_ctx, u64 flags)
2095 {
2096         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
2097         struct hlist_nulls_head *head;
2098         struct hlist_nulls_node *n;
2099         struct htab_elem *elem;
2100         u32 roundup_key_size;
2101         int i, num_elems = 0;
2102         void __percpu *pptr;
2103         struct bucket *b;
2104         void *key, *val;
2105         bool is_percpu;
2106         u64 ret = 0;
2107
2108         if (flags != 0)
2109                 return -EINVAL;
2110
2111         is_percpu = htab_is_percpu(htab);
2112
2113         roundup_key_size = round_up(map->key_size, 8);
2114         /* disable migration so percpu value prepared here will be the
2115          * same as the one seen by the bpf program with bpf_map_lookup_elem().
2116          */
2117         if (is_percpu)
2118                 migrate_disable();
2119         for (i = 0; i < htab->n_buckets; i++) {
2120                 b = &htab->buckets[i];
2121                 rcu_read_lock();
2122                 head = &b->head;
2123                 hlist_nulls_for_each_entry_rcu(elem, n, head, hash_node) {
2124                         key = elem->key;
2125                         if (is_percpu) {
2126                                 /* current cpu value for percpu map */
2127                                 pptr = htab_elem_get_ptr(elem, map->key_size);
2128                                 val = this_cpu_ptr(pptr);
2129                         } else {
2130                                 val = elem->key + roundup_key_size;
2131                         }
2132                         num_elems++;
2133                         ret = callback_fn((u64)(long)map, (u64)(long)key,
2134                                           (u64)(long)val, (u64)(long)callback_ctx, 0);
2135                         /* return value: 0 - continue, 1 - stop and return */
2136                         if (ret) {
2137                                 rcu_read_unlock();
2138                                 goto out;
2139                         }
2140                 }
2141                 rcu_read_unlock();
2142         }
2143 out:
2144         if (is_percpu)
2145                 migrate_enable();
2146         return num_elems;
2147 }
2148
2149 BTF_ID_LIST_SINGLE(htab_map_btf_ids, struct, bpf_htab)
2150 const struct bpf_map_ops htab_map_ops = {
2151         .map_meta_equal = bpf_map_meta_equal,
2152         .map_alloc_check = htab_map_alloc_check,
2153         .map_alloc = htab_map_alloc,
2154         .map_free = htab_map_free,
2155         .map_get_next_key = htab_map_get_next_key,
2156         .map_release_uref = htab_map_free_timers,
2157         .map_lookup_elem = htab_map_lookup_elem,
2158         .map_lookup_and_delete_elem = htab_map_lookup_and_delete_elem,
2159         .map_update_elem = htab_map_update_elem,
2160         .map_delete_elem = htab_map_delete_elem,
2161         .map_gen_lookup = htab_map_gen_lookup,
2162         .map_seq_show_elem = htab_map_seq_show_elem,
2163         .map_set_for_each_callback_args = map_set_for_each_callback_args,
2164         .map_for_each_callback = bpf_for_each_hash_elem,
2165         BATCH_OPS(htab),
2166         .map_btf_id = &htab_map_btf_ids[0],
2167         .iter_seq_info = &iter_seq_info,
2168 };
2169
2170 const struct bpf_map_ops htab_lru_map_ops = {
2171         .map_meta_equal = bpf_map_meta_equal,
2172         .map_alloc_check = htab_map_alloc_check,
2173         .map_alloc = htab_map_alloc,
2174         .map_free = htab_map_free,
2175         .map_get_next_key = htab_map_get_next_key,
2176         .map_release_uref = htab_map_free_timers,
2177         .map_lookup_elem = htab_lru_map_lookup_elem,
2178         .map_lookup_and_delete_elem = htab_lru_map_lookup_and_delete_elem,
2179         .map_lookup_elem_sys_only = htab_lru_map_lookup_elem_sys,
2180         .map_update_elem = htab_lru_map_update_elem,
2181         .map_delete_elem = htab_lru_map_delete_elem,
2182         .map_gen_lookup = htab_lru_map_gen_lookup,
2183         .map_seq_show_elem = htab_map_seq_show_elem,
2184         .map_set_for_each_callback_args = map_set_for_each_callback_args,
2185         .map_for_each_callback = bpf_for_each_hash_elem,
2186         BATCH_OPS(htab_lru),
2187         .map_btf_id = &htab_map_btf_ids[0],
2188         .iter_seq_info = &iter_seq_info,
2189 };
2190
2191 /* Called from eBPF program */
2192 static void *htab_percpu_map_lookup_elem(struct bpf_map *map, void *key)
2193 {
2194         struct htab_elem *l = __htab_map_lookup_elem(map, key);
2195
2196         if (l)
2197                 return this_cpu_ptr(htab_elem_get_ptr(l, map->key_size));
2198         else
2199                 return NULL;
2200 }
2201
2202 static void *htab_percpu_map_lookup_percpu_elem(struct bpf_map *map, void *key, u32 cpu)
2203 {
2204         struct htab_elem *l;
2205
2206         if (cpu >= nr_cpu_ids)
2207                 return NULL;
2208
2209         l = __htab_map_lookup_elem(map, key);
2210         if (l)
2211                 return per_cpu_ptr(htab_elem_get_ptr(l, map->key_size), cpu);
2212         else
2213                 return NULL;
2214 }
2215
2216 static void *htab_lru_percpu_map_lookup_elem(struct bpf_map *map, void *key)
2217 {
2218         struct htab_elem *l = __htab_map_lookup_elem(map, key);
2219
2220         if (l) {
2221                 bpf_lru_node_set_ref(&l->lru_node);
2222                 return this_cpu_ptr(htab_elem_get_ptr(l, map->key_size));
2223         }
2224
2225         return NULL;
2226 }
2227
2228 static void *htab_lru_percpu_map_lookup_percpu_elem(struct bpf_map *map, void *key, u32 cpu)
2229 {
2230         struct htab_elem *l;
2231
2232         if (cpu >= nr_cpu_ids)
2233                 return NULL;
2234
2235         l = __htab_map_lookup_elem(map, key);
2236         if (l) {
2237                 bpf_lru_node_set_ref(&l->lru_node);
2238                 return per_cpu_ptr(htab_elem_get_ptr(l, map->key_size), cpu);
2239         }
2240
2241         return NULL;
2242 }
2243
2244 int bpf_percpu_hash_copy(struct bpf_map *map, void *key, void *value)
2245 {
2246         struct htab_elem *l;
2247         void __percpu *pptr;
2248         int ret = -ENOENT;
2249         int cpu, off = 0;
2250         u32 size;
2251
2252         /* per_cpu areas are zero-filled and bpf programs can only
2253          * access 'value_size' of them, so copying rounded areas
2254          * will not leak any kernel data
2255          */
2256         size = round_up(map->value_size, 8);
2257         rcu_read_lock();
2258         l = __htab_map_lookup_elem(map, key);
2259         if (!l)
2260                 goto out;
2261         /* We do not mark LRU map element here in order to not mess up
2262          * eviction heuristics when user space does a map walk.
2263          */
2264         pptr = htab_elem_get_ptr(l, map->key_size);
2265         for_each_possible_cpu(cpu) {
2266                 bpf_long_memcpy(value + off,
2267                                 per_cpu_ptr(pptr, cpu), size);
2268                 off += size;
2269         }
2270         ret = 0;
2271 out:
2272         rcu_read_unlock();
2273         return ret;
2274 }
2275
2276 int bpf_percpu_hash_update(struct bpf_map *map, void *key, void *value,
2277                            u64 map_flags)
2278 {
2279         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
2280         int ret;
2281
2282         rcu_read_lock();
2283         if (htab_is_lru(htab))
2284                 ret = __htab_lru_percpu_map_update_elem(map, key, value,
2285                                                         map_flags, true);
2286         else
2287                 ret = __htab_percpu_map_update_elem(map, key, value, map_flags,
2288                                                     true);
2289         rcu_read_unlock();
2290
2291         return ret;
2292 }
2293
2294 static void htab_percpu_map_seq_show_elem(struct bpf_map *map, void *key,
2295                                           struct seq_file *m)
2296 {
2297         struct htab_elem *l;
2298         void __percpu *pptr;
2299         int cpu;
2300
2301         rcu_read_lock();
2302
2303         l = __htab_map_lookup_elem(map, key);
2304         if (!l) {
2305                 rcu_read_unlock();
2306                 return;
2307         }
2308
2309         btf_type_seq_show(map->btf, map->btf_key_type_id, key, m);
2310         seq_puts(m, ": {\n");
2311         pptr = htab_elem_get_ptr(l, map->key_size);
2312         for_each_possible_cpu(cpu) {
2313                 seq_printf(m, "\tcpu%d: ", cpu);
2314                 btf_type_seq_show(map->btf, map->btf_value_type_id,
2315                                   per_cpu_ptr(pptr, cpu), m);
2316                 seq_puts(m, "\n");
2317         }
2318         seq_puts(m, "}\n");
2319
2320         rcu_read_unlock();
2321 }
2322
2323 const struct bpf_map_ops htab_percpu_map_ops = {
2324         .map_meta_equal = bpf_map_meta_equal,
2325         .map_alloc_check = htab_map_alloc_check,
2326         .map_alloc = htab_map_alloc,
2327         .map_free = htab_map_free,
2328         .map_get_next_key = htab_map_get_next_key,
2329         .map_lookup_elem = htab_percpu_map_lookup_elem,
2330         .map_lookup_and_delete_elem = htab_percpu_map_lookup_and_delete_elem,
2331         .map_update_elem = htab_percpu_map_update_elem,
2332         .map_delete_elem = htab_map_delete_elem,
2333         .map_lookup_percpu_elem = htab_percpu_map_lookup_percpu_elem,
2334         .map_seq_show_elem = htab_percpu_map_seq_show_elem,
2335         .map_set_for_each_callback_args = map_set_for_each_callback_args,
2336         .map_for_each_callback = bpf_for_each_hash_elem,
2337         BATCH_OPS(htab_percpu),
2338         .map_btf_id = &htab_map_btf_ids[0],
2339         .iter_seq_info = &iter_seq_info,
2340 };
2341
2342 const struct bpf_map_ops htab_lru_percpu_map_ops = {
2343         .map_meta_equal = bpf_map_meta_equal,
2344         .map_alloc_check = htab_map_alloc_check,
2345         .map_alloc = htab_map_alloc,
2346         .map_free = htab_map_free,
2347         .map_get_next_key = htab_map_get_next_key,
2348         .map_lookup_elem = htab_lru_percpu_map_lookup_elem,
2349         .map_lookup_and_delete_elem = htab_lru_percpu_map_lookup_and_delete_elem,
2350         .map_update_elem = htab_lru_percpu_map_update_elem,
2351         .map_delete_elem = htab_lru_map_delete_elem,
2352         .map_lookup_percpu_elem = htab_lru_percpu_map_lookup_percpu_elem,
2353         .map_seq_show_elem = htab_percpu_map_seq_show_elem,
2354         .map_set_for_each_callback_args = map_set_for_each_callback_args,
2355         .map_for_each_callback = bpf_for_each_hash_elem,
2356         BATCH_OPS(htab_lru_percpu),
2357         .map_btf_id = &htab_map_btf_ids[0],
2358         .iter_seq_info = &iter_seq_info,
2359 };
2360
2361 static int fd_htab_map_alloc_check(union bpf_attr *attr)
2362 {
2363         if (attr->value_size != sizeof(u32))
2364                 return -EINVAL;
2365         return htab_map_alloc_check(attr);
2366 }
2367
2368 static void fd_htab_map_free(struct bpf_map *map)
2369 {
2370         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
2371         struct hlist_nulls_node *n;
2372         struct hlist_nulls_head *head;
2373         struct htab_elem *l;
2374         int i;
2375
2376         for (i = 0; i < htab->n_buckets; i++) {
2377                 head = select_bucket(htab, i);
2378
2379                 hlist_nulls_for_each_entry_safe(l, n, head, hash_node) {
2380                         void *ptr = fd_htab_map_get_ptr(map, l);
2381
2382                         map->ops->map_fd_put_ptr(ptr);
2383                 }
2384         }
2385
2386         htab_map_free(map);
2387 }
2388
2389 /* only called from syscall */
2390 int bpf_fd_htab_map_lookup_elem(struct bpf_map *map, void *key, u32 *value)
2391 {
2392         void **ptr;
2393         int ret = 0;
2394
2395         if (!map->ops->map_fd_sys_lookup_elem)
2396                 return -ENOTSUPP;
2397
2398         rcu_read_lock();
2399         ptr = htab_map_lookup_elem(map, key);
2400         if (ptr)
2401                 *value = map->ops->map_fd_sys_lookup_elem(READ_ONCE(*ptr));
2402         else
2403                 ret = -ENOENT;
2404         rcu_read_unlock();
2405
2406         return ret;
2407 }
2408
2409 /* only called from syscall */
2410 int bpf_fd_htab_map_update_elem(struct bpf_map *map, struct file *map_file,
2411                                 void *key, void *value, u64 map_flags)
2412 {
2413         void *ptr;
2414         int ret;
2415         u32 ufd = *(u32 *)value;
2416
2417         ptr = map->ops->map_fd_get_ptr(map, map_file, ufd);
2418         if (IS_ERR(ptr))
2419                 return PTR_ERR(ptr);
2420
2421         ret = htab_map_update_elem(map, key, &ptr, map_flags);
2422         if (ret)
2423                 map->ops->map_fd_put_ptr(ptr);
2424
2425         return ret;
2426 }
2427
2428 static struct bpf_map *htab_of_map_alloc(union bpf_attr *attr)
2429 {
2430         struct bpf_map *map, *inner_map_meta;
2431
2432         inner_map_meta = bpf_map_meta_alloc(attr->inner_map_fd);
2433         if (IS_ERR(inner_map_meta))
2434                 return inner_map_meta;
2435
2436         map = htab_map_alloc(attr);
2437         if (IS_ERR(map)) {
2438                 bpf_map_meta_free(inner_map_meta);
2439                 return map;
2440         }
2441
2442         map->inner_map_meta = inner_map_meta;
2443
2444         return map;
2445 }
2446
2447 static void *htab_of_map_lookup_elem(struct bpf_map *map, void *key)
2448 {
2449         struct bpf_map **inner_map  = htab_map_lookup_elem(map, key);
2450
2451         if (!inner_map)
2452                 return NULL;
2453
2454         return READ_ONCE(*inner_map);
2455 }
2456
2457 static int htab_of_map_gen_lookup(struct bpf_map *map,
2458                                   struct bpf_insn *insn_buf)
2459 {
2460         struct bpf_insn *insn = insn_buf;
2461         const int ret = BPF_REG_0;
2462
2463         BUILD_BUG_ON(!__same_type(&__htab_map_lookup_elem,
2464                      (void *(*)(struct bpf_map *map, void *key))NULL));
2465         *insn++ = BPF_EMIT_CALL(__htab_map_lookup_elem);
2466         *insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 2);
2467         *insn++ = BPF_ALU64_IMM(BPF_ADD, ret,
2468                                 offsetof(struct htab_elem, key) +
2469                                 round_up(map->key_size, 8));
2470         *insn++ = BPF_LDX_MEM(BPF_DW, ret, ret, 0);
2471
2472         return insn - insn_buf;
2473 }
2474
2475 static void htab_of_map_free(struct bpf_map *map)
2476 {
2477         bpf_map_meta_free(map->inner_map_meta);
2478         fd_htab_map_free(map);
2479 }
2480
2481 const struct bpf_map_ops htab_of_maps_map_ops = {
2482         .map_alloc_check = fd_htab_map_alloc_check,
2483         .map_alloc = htab_of_map_alloc,
2484         .map_free = htab_of_map_free,
2485         .map_get_next_key = htab_map_get_next_key,
2486         .map_lookup_elem = htab_of_map_lookup_elem,
2487         .map_delete_elem = htab_map_delete_elem,
2488         .map_fd_get_ptr = bpf_map_fd_get_ptr,
2489         .map_fd_put_ptr = bpf_map_fd_put_ptr,
2490         .map_fd_sys_lookup_elem = bpf_map_fd_sys_lookup_elem,
2491         .map_gen_lookup = htab_of_map_gen_lookup,
2492         .map_check_btf = map_check_no_btf,
2493         BATCH_OPS(htab),
2494         .map_btf_id = &htab_map_btf_ids[0],
2495 };