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 #include <linux/bpf_mem_alloc.h>
  18
  19 #define HTAB_CREATE_FLAG_MASK                                           \
  20         (BPF_F_NO_PREALLOC | BPF_F_NO_COMMON_LRU | BPF_F_NUMA_NODE |    \
  21          BPF_F_ACCESS_MASK | BPF_F_ZERO_SEED)
  22
  23 #define BATCH_OPS(_name)                        \
  24         .map_lookup_batch =                     \
  25         _name##_map_lookup_batch,               \
  26         .map_lookup_and_delete_batch =          \
  27         _name##_map_lookup_and_delete_batch,    \
  28         .map_update_batch =                     \
  29         generic_map_update_batch,               \
  30         .map_delete_batch =                     \
  31         generic_map_delete_batch
  32
  33 /*
  34  * The bucket lock has two protection scopes:
  35  *
  36  * 1) Serializing concurrent operations from BPF programs on different
  37  *    CPUs
  38  *
  39  * 2) Serializing concurrent operations from BPF programs and sys_bpf()
  40  *
  41  * BPF programs can execute in any context including perf, kprobes and
  42  * tracing. As there are almost no limits where perf, kprobes and tracing
  43  * can be invoked from the lock operations need to be protected against
  44  * deadlocks. Deadlocks can be caused by recursion and by an invocation in
  45  * the lock held section when functions which acquire this lock are invoked
  46  * from sys_bpf(). BPF recursion is prevented by incrementing the per CPU
  47  * variable bpf_prog_active, which prevents BPF programs attached to perf
  48  * events, kprobes and tracing to be invoked before the prior invocation
  49  * from one of these contexts completed. sys_bpf() uses the same mechanism
  50  * by pinning the task to the current CPU and incrementing the recursion
  51  * protection across the map operation.
  52  *
  53  * This has subtle implications on PREEMPT_RT. PREEMPT_RT forbids certain
  54  * operations like memory allocations (even with GFP_ATOMIC) from atomic
  55  * contexts. This is required because even with GFP_ATOMIC the memory
  56  * allocator calls into code paths which acquire locks with long held lock
  57  * sections. To ensure the deterministic behaviour these locks are regular
  58  * spinlocks, which are converted to 'sleepable' spinlocks on RT. The only
  59  * true atomic contexts on an RT kernel are the low level hardware
  60  * handling, scheduling, low level interrupt handling, NMIs etc. None of
  61  * these contexts should ever do memory allocations.
  62  *
  63  * As regular device interrupt handlers and soft interrupts are forced into
  64  * thread context, the existing code which does
  65  *   spin_lock*(); alloc(GFP_ATOMIC); spin_unlock*();
  66  * just works.
  67  *
  68  * In theory the BPF locks could be converted to regular spinlocks as well,
  69  * but the bucket locks and percpu_freelist locks can be taken from
  70  * arbitrary contexts (perf, kprobes, tracepoints) which are required to be
  71  * atomic contexts even on RT. Before the introduction of bpf_mem_alloc,
  72  * it is only safe to use raw spinlock for preallocated hash map on a RT kernel,
  73  * because there is no memory allocation within the lock held sections. However
  74  * after hash map was fully converted to use bpf_mem_alloc, there will be
  75  * non-synchronous memory allocation for non-preallocated hash map, so it is
  76  * safe to always use raw spinlock for bucket lock.
  77  */
  78 struct bucket {
  79         struct hlist_nulls_head head;
  80         raw_spinlock_t raw_lock;
  81 };
  82
  83 #define HASHTAB_MAP_LOCK_COUNT 8
  84 #define HASHTAB_MAP_LOCK_MASK (HASHTAB_MAP_LOCK_COUNT - 1)
  85
  86 struct bpf_htab {
  87         struct bpf_map map;
  88         struct bpf_mem_alloc ma;
  89         struct bpf_mem_alloc pcpu_ma;
  90         struct bucket *buckets;
  91         void *elems;
  92         union {
  93                 struct pcpu_freelist freelist;
  94                 struct bpf_lru lru;
  95         };
  96         struct htab_elem *__percpu *extra_elems;
  97         /* number of elements in non-preallocated hashtable are kept
  98          * in either pcount or count
  99          */
 100         struct percpu_counter pcount;
 101         atomic_t count;
 102         bool use_percpu_counter;
 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 pcpu_freelist_node fnode;
 118                                 struct htab_elem *batch_flink;
 119                         };
 120                 };
 121         };
 122         union {
 123                 /* pointer to per-cpu pointer */
 124                 void *ptr_to_pptr;
 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 void htab_init_buckets(struct bpf_htab *htab)
 137 {
 138         unsigned int i;
 139
 140         for (i = 0; i < htab->n_buckets; i++) {
 141                 INIT_HLIST_NULLS_HEAD(&htab->buckets[i].head, i);
 142                 raw_spin_lock_init(&htab->buckets[i].raw_lock);
 143                 lockdep_set_class(&htab->buckets[i].raw_lock,
 144                                           &htab->lockdep_key);
 145                 cond_resched();
 146         }
 147 }
 148
 149 static inline int htab_lock_bucket(const struct bpf_htab *htab,
 150                                    struct bucket *b, u32 hash,
 151                                    unsigned long *pflags)
 152 {
 153         unsigned long flags;
 154
 155         hash = hash & HASHTAB_MAP_LOCK_MASK;
 156
 157         preempt_disable();
 158         if (unlikely(__this_cpu_inc_return(*(htab->map_locked[hash])) != 1)) {
 159                 __this_cpu_dec(*(htab->map_locked[hash]));
 160                 preempt_enable();
 161                 return -EBUSY;
 162         }
 163
 164         raw_spin_lock_irqsave(&b->raw_lock, flags);
 165         *pflags = flags;
 166
 167         return 0;
 168 }
 169
 170 static inline void htab_unlock_bucket(const struct bpf_htab *htab,
 171                                       struct bucket *b, u32 hash,
 172                                       unsigned long flags)
 173 {
 174         hash = hash & HASHTAB_MAP_LOCK_MASK;
 175         raw_spin_unlock_irqrestore(&b->raw_lock, flags);
 176         __this_cpu_dec(*(htab->map_locked[hash]));
 177         preempt_enable();
 178 }
 179
 180 static bool htab_lru_map_delete_node(void *arg, struct bpf_lru_node *node);
 181
 182 static bool htab_is_lru(const struct bpf_htab *htab)
 183 {
 184         return htab->map.map_type == BPF_MAP_TYPE_LRU_HASH ||
 185                 htab->map.map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH;
 186 }
 187
 188 static bool htab_is_percpu(const struct bpf_htab *htab)
 189 {
 190         return htab->map.map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 191                 htab->map.map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH;
 192 }
 193
 194 static inline void htab_elem_set_ptr(struct htab_elem *l, u32 key_size,
 195                                      void __percpu *pptr)
 196 {
 197         *(void __percpu **)(l->key + key_size) = pptr;
 198 }
 199
 200 static inline void __percpu *htab_elem_get_ptr(struct htab_elem *l, u32 key_size)
 201 {
 202         return *(void __percpu **)(l->key + key_size);
 203 }
 204
 205 static void *fd_htab_map_get_ptr(const struct bpf_map *map, struct htab_elem *l)
 206 {
 207         return *(void **)(l->key + roundup(map->key_size, 8));
 208 }
 209
 210 static struct htab_elem *get_htab_elem(struct bpf_htab *htab, int i)
 211 {
 212         return (struct htab_elem *) (htab->elems + i * (u64)htab->elem_size);
 213 }
 214
 215 static bool htab_has_extra_elems(struct bpf_htab *htab)
 216 {
 217         return !htab_is_percpu(htab) && !htab_is_lru(htab);
 218 }
 219
 220 static void htab_free_prealloced_timers(struct bpf_htab *htab)
 221 {
 222         u32 num_entries = htab->map.max_entries;
 223         int i;
 224
 225         if (!map_value_has_timer(&htab->map))
 226                 return;
 227         if (htab_has_extra_elems(htab))
 228                 num_entries += num_possible_cpus();
 229
 230         for (i = 0; i < num_entries; i++) {
 231                 struct htab_elem *elem;
 232
 233                 elem = get_htab_elem(htab, i);
 234                 bpf_timer_cancel_and_free(elem->key +
 235                                           round_up(htab->map.key_size, 8) +
 236                                           htab->map.timer_off);
 237                 cond_resched();
 238         }
 239 }
 240
 241 static void htab_free_prealloced_kptrs(struct bpf_htab *htab)
 242 {
 243         u32 num_entries = htab->map.max_entries;
 244         int i;
 245
 246         if (!map_value_has_kptrs(&htab->map))
 247                 return;
 248         if (htab_has_extra_elems(htab))
 249                 num_entries += num_possible_cpus();
 250
 251         for (i = 0; i < num_entries; i++) {
 252                 struct htab_elem *elem;
 253
 254                 elem = get_htab_elem(htab, i);
 255                 bpf_map_free_kptrs(&htab->map, elem->key + round_up(htab->map.key_size, 8));
 256                 cond_resched();
 257         }
 258 }
 259
 260 static void htab_free_elems(struct bpf_htab *htab)
 261 {
 262         int i;
 263
 264         if (!htab_is_percpu(htab))
 265                 goto free_elems;
 266
 267         for (i = 0; i < htab->map.max_entries; i++) {
 268                 void __percpu *pptr;
 269
 270                 pptr = htab_elem_get_ptr(get_htab_elem(htab, i),
 271                                          htab->map.key_size);
 272                 free_percpu(pptr);
 273                 cond_resched();
 274         }
 275 free_elems:
 276         bpf_map_area_free(htab->elems);
 277 }
 278
 279 /* The LRU list has a lock (lru_lock). Each htab bucket has a lock
 280  * (bucket_lock). If both locks need to be acquired together, the lock
 281  * order is always lru_lock -> bucket_lock and this only happens in
 282  * bpf_lru_list.c logic. For example, certain code path of
 283  * bpf_lru_pop_free(), which is called by function prealloc_lru_pop(),
 284  * will acquire lru_lock first followed by acquiring bucket_lock.
 285  *
 286  * In hashtab.c, to avoid deadlock, lock acquisition of
 287  * bucket_lock followed by lru_lock is not allowed. In such cases,
 288  * bucket_lock needs to be released first before acquiring lru_lock.
 289  */
 290 static struct htab_elem *prealloc_lru_pop(struct bpf_htab *htab, void *key,
 291                                           u32 hash)
 292 {
 293         struct bpf_lru_node *node = bpf_lru_pop_free(&htab->lru, hash);
 294         struct htab_elem *l;
 295
 296         if (node) {
 297                 l = container_of(node, struct htab_elem, lru_node);
 298                 memcpy(l->key, key, htab->map.key_size);
 299                 return l;
 300         }
 301
 302         return NULL;
 303 }
 304
 305 static int prealloc_init(struct bpf_htab *htab)
 306 {
 307         u32 num_entries = htab->map.max_entries;
 308         int err = -ENOMEM, i;
 309
 310         if (htab_has_extra_elems(htab))
 311                 num_entries += num_possible_cpus();
 312
 313         htab->elems = bpf_map_area_alloc((u64)htab->elem_size * num_entries,
 314                                          htab->map.numa_node);
 315         if (!htab->elems)
 316                 return -ENOMEM;
 317
 318         if (!htab_is_percpu(htab))
 319                 goto skip_percpu_elems;
 320
 321         for (i = 0; i < num_entries; i++) {
 322                 u32 size = round_up(htab->map.value_size, 8);
 323                 void __percpu *pptr;
 324
 325                 pptr = bpf_map_alloc_percpu(&htab->map, size, 8,
 326                                             GFP_USER | __GFP_NOWARN);
 327                 if (!pptr)
 328                         goto free_elems;
 329                 htab_elem_set_ptr(get_htab_elem(htab, i), htab->map.key_size,
 330                                   pptr);
 331                 cond_resched();
 332         }
 333
 334 skip_percpu_elems:
 335         if (htab_is_lru(htab))
 336                 err = bpf_lru_init(&htab->lru,
 337                                    htab->map.map_flags & BPF_F_NO_COMMON_LRU,
 338                                    offsetof(struct htab_elem, hash) -
 339                                    offsetof(struct htab_elem, lru_node),
 340                                    htab_lru_map_delete_node,
 341                                    htab);
 342         else
 343                 err = pcpu_freelist_init(&htab->freelist);
 344
 345         if (err)
 346                 goto free_elems;
 347
 348         if (htab_is_lru(htab))
 349                 bpf_lru_populate(&htab->lru, htab->elems,
 350                                  offsetof(struct htab_elem, lru_node),
 351                                  htab->elem_size, num_entries);
 352         else
 353                 pcpu_freelist_populate(&htab->freelist,
 354                                        htab->elems + offsetof(struct htab_elem, fnode),
 355                                        htab->elem_size, num_entries);
 356
 357         return 0;
 358
 359 free_elems:
 360         htab_free_elems(htab);
 361         return err;
 362 }
 363
 364 static void prealloc_destroy(struct bpf_htab *htab)
 365 {
 366         htab_free_elems(htab);
 367
 368         if (htab_is_lru(htab))
 369                 bpf_lru_destroy(&htab->lru);
 370         else
 371                 pcpu_freelist_destroy(&htab->freelist);
 372 }
 373
 374 static int alloc_extra_elems(struct bpf_htab *htab)
 375 {
 376         struct htab_elem *__percpu *pptr, *l_new;
 377         struct pcpu_freelist_node *l;
 378         int cpu;
 379
 380         pptr = bpf_map_alloc_percpu(&htab->map, sizeof(struct htab_elem *), 8,
 381                                     GFP_USER | __GFP_NOWARN);
 382         if (!pptr)
 383                 return -ENOMEM;
 384
 385         for_each_possible_cpu(cpu) {
 386                 l = pcpu_freelist_pop(&htab->freelist);
 387                 /* pop will succeed, since prealloc_init()
 388                  * preallocated extra num_possible_cpus elements
 389                  */
 390                 l_new = container_of(l, struct htab_elem, fnode);
 391                 *per_cpu_ptr(pptr, cpu) = l_new;
 392         }
 393         htab->extra_elems = pptr;
 394         return 0;
 395 }
 396
 397 /* Called from syscall */
 398 static int htab_map_alloc_check(union bpf_attr *attr)
 399 {
 400         bool percpu = (attr->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 401                        attr->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH);
 402         bool lru = (attr->map_type == BPF_MAP_TYPE_LRU_HASH ||
 403                     attr->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH);
 404         /* percpu_lru means each cpu has its own LRU list.
 405          * it is different from BPF_MAP_TYPE_PERCPU_HASH where
 406          * the map's value itself is percpu.  percpu_lru has
 407          * nothing to do with the map's value.
 408          */
 409         bool percpu_lru = (attr->map_flags & BPF_F_NO_COMMON_LRU);
 410         bool prealloc = !(attr->map_flags & BPF_F_NO_PREALLOC);
 411         bool zero_seed = (attr->map_flags & BPF_F_ZERO_SEED);
 412         int numa_node = bpf_map_attr_numa_node(attr);
 413
 414         BUILD_BUG_ON(offsetof(struct htab_elem, fnode.next) !=
 415                      offsetof(struct htab_elem, hash_node.pprev));
 416
 417         if (lru && !bpf_capable())
 418                 /* LRU implementation is much complicated than other
 419                  * maps.  Hence, limit to CAP_BPF.
 420                  */
 421                 return -EPERM;
 422
 423         if (zero_seed && !capable(CAP_SYS_ADMIN))
 424                 /* Guard against local DoS, and discourage production use. */
 425                 return -EPERM;
 426
 427         if (attr->map_flags & ~HTAB_CREATE_FLAG_MASK ||
 428             !bpf_map_flags_access_ok(attr->map_flags))
 429                 return -EINVAL;
 430
 431         if (!lru && percpu_lru)
 432                 return -EINVAL;
 433
 434         if (lru && !prealloc)
 435                 return -ENOTSUPP;
 436
 437         if (numa_node != NUMA_NO_NODE && (percpu || percpu_lru))
 438                 return -EINVAL;
 439
 440         /* check sanity of attributes.
 441          * value_size == 0 may be allowed in the future to use map as a set
 442          */
 443         if (attr->max_entries == 0 || attr->key_size == 0 ||
 444             attr->value_size == 0)
 445                 return -EINVAL;
 446
 447         if ((u64)attr->key_size + attr->value_size >= KMALLOC_MAX_SIZE -
 448            sizeof(struct htab_elem))
 449                 /* if key_size + value_size is bigger, the user space won't be
 450                  * able to access the elements via bpf syscall. This check
 451                  * also makes sure that the elem_size doesn't overflow and it's
 452                  * kmalloc-able later in htab_map_update_elem()
 453                  */
 454                 return -E2BIG;
 455
 456         return 0;
 457 }
 458
 459 static struct bpf_map *htab_map_alloc(union bpf_attr *attr)
 460 {
 461         bool percpu = (attr->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 462                        attr->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH);
 463         bool lru = (attr->map_type == BPF_MAP_TYPE_LRU_HASH ||
 464                     attr->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH);
 465         /* percpu_lru means each cpu has its own LRU list.
 466          * it is different from BPF_MAP_TYPE_PERCPU_HASH where
 467          * the map's value itself is percpu.  percpu_lru has
 468          * nothing to do with the map's value.
 469          */
 470         bool percpu_lru = (attr->map_flags & BPF_F_NO_COMMON_LRU);
 471         bool prealloc = !(attr->map_flags & BPF_F_NO_PREALLOC);
 472         struct bpf_htab *htab;
 473         int err, i;
 474
 475         htab = bpf_map_area_alloc(sizeof(*htab), NUMA_NO_NODE);
 476         if (!htab)
 477                 return ERR_PTR(-ENOMEM);
 478
 479         lockdep_register_key(&htab->lockdep_key);
 480
 481         bpf_map_init_from_attr(&htab->map, attr);
 482
 483         if (percpu_lru) {
 484                 /* ensure each CPU's lru list has >=1 elements.
 485                  * since we are at it, make each lru list has the same
 486                  * number of elements.
 487                  */
 488                 htab->map.max_entries = roundup(attr->max_entries,
 489                                                 num_possible_cpus());
 490                 if (htab->map.max_entries < attr->max_entries)
 491                         htab->map.max_entries = rounddown(attr->max_entries,
 492                                                           num_possible_cpus());
 493         }
 494
 495         /* hash table size must be power of 2 */
 496         htab->n_buckets = roundup_pow_of_two(htab->map.max_entries);
 497
 498         htab->elem_size = sizeof(struct htab_elem) +
 499                           round_up(htab->map.key_size, 8);
 500         if (percpu)
 501                 htab->elem_size += sizeof(void *);
 502         else
 503                 htab->elem_size += round_up(htab->map.value_size, 8);
 504
 505         err = -E2BIG;
 506         /* prevent zero size kmalloc and check for u32 overflow */
 507         if (htab->n_buckets == 0 ||
 508             htab->n_buckets > U32_MAX / sizeof(struct bucket))
 509                 goto free_htab;
 510
 511         err = -ENOMEM;
 512         htab->buckets = bpf_map_area_alloc(htab->n_buckets *
 513                                            sizeof(struct bucket),
 514                                            htab->map.numa_node);
 515         if (!htab->buckets)
 516                 goto free_htab;
 517
 518         for (i = 0; i < HASHTAB_MAP_LOCK_COUNT; i++) {
 519                 htab->map_locked[i] = bpf_map_alloc_percpu(&htab->map,
 520                                                            sizeof(int),
 521                                                            sizeof(int),
 522                                                            GFP_USER);
 523                 if (!htab->map_locked[i])
 524                         goto free_map_locked;
 525         }
 526
 527         if (htab->map.map_flags & BPF_F_ZERO_SEED)
 528                 htab->hashrnd = 0;
 529         else
 530                 htab->hashrnd = get_random_u32();
 531
 532         htab_init_buckets(htab);
 533
 534 /* compute_batch_value() computes batch value as num_online_cpus() * 2
 535  * and __percpu_counter_compare() needs
 536  * htab->max_entries - cur_number_of_elems to be more than batch * num_online_cpus()
 537  * for percpu_counter to be faster than atomic_t. In practice the average bpf
 538  * hash map size is 10k, which means that a system with 64 cpus will fill
 539  * hashmap to 20% of 10k before percpu_counter becomes ineffective. Therefore
 540  * define our own batch count as 32 then 10k hash map can be filled up to 80%:
 541  * 10k - 8k > 32 _batch_ * 64 _cpus_
 542  * and __percpu_counter_compare() will still be fast. At that point hash map
 543  * collisions will dominate its performance anyway. Assume that hash map filled
 544  * to 50+% isn't going to be O(1) and use the following formula to choose
 545  * between percpu_counter and atomic_t.
 546  */
 547 #define PERCPU_COUNTER_BATCH 32
 548         if (attr->max_entries / 2 > num_online_cpus() * PERCPU_COUNTER_BATCH)
 549                 htab->use_percpu_counter = true;
 550
 551         if (htab->use_percpu_counter) {
 552                 err = percpu_counter_init(&htab->pcount, 0, GFP_KERNEL);
 553                 if (err)
 554                         goto free_map_locked;
 555         }
 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         } else {
 571                 err = bpf_mem_alloc_init(&htab->ma, htab->elem_size, false);
 572                 if (err)
 573                         goto free_map_locked;
 574                 if (percpu) {
 575                         err = bpf_mem_alloc_init(&htab->pcpu_ma,
 576                                                  round_up(htab->map.value_size, 8), true);
 577                         if (err)
 578                                 goto free_map_locked;
 579                 }
 580         }
 581
 582         return &htab->map;
 583
 584 free_prealloc:
 585         prealloc_destroy(htab);
 586 free_map_locked:
 587         if (htab->use_percpu_counter)
 588                 percpu_counter_destroy(&htab->pcount);
 589         for (i = 0; i < HASHTAB_MAP_LOCK_COUNT; i++)
 590                 free_percpu(htab->map_locked[i]);
 591         bpf_map_area_free(htab->buckets);
 592         bpf_mem_alloc_destroy(&htab->pcpu_ma);
 593         bpf_mem_alloc_destroy(&htab->ma);
 594 free_htab:
 595         lockdep_unregister_key(&htab->lockdep_key);
 596         bpf_map_area_free(htab);
 597         return ERR_PTR(err);
 598 }
 599
 600 static inline u32 htab_map_hash(const void *key, u32 key_len, u32 hashrnd)
 601 {
 602         return jhash(key, key_len, hashrnd);
 603 }
 604
 605 static inline struct bucket *__select_bucket(struct bpf_htab *htab, u32 hash)
 606 {
 607         return &htab->buckets[hash & (htab->n_buckets - 1)];
 608 }
 609
 610 static inline struct hlist_nulls_head *select_bucket(struct bpf_htab *htab, u32 hash)
 611 {
 612         return &__select_bucket(htab, hash)->head;
 613 }
 614
 615 /* this lookup function can only be called with bucket lock taken */
 616 static struct htab_elem *lookup_elem_raw(struct hlist_nulls_head *head, u32 hash,
 617                                          void *key, u32 key_size)
 618 {
 619         struct hlist_nulls_node *n;
 620         struct htab_elem *l;
 621
 622         hlist_nulls_for_each_entry_rcu(l, n, head, hash_node)
 623                 if (l->hash == hash && !memcmp(&l->key, key, key_size))
 624                         return l;
 625
 626         return NULL;
 627 }
 628
 629 /* can be called without bucket lock. it will repeat the loop in
 630  * the unlikely event when elements moved from one bucket into another
 631  * while link list is being walked
 632  */
 633 static struct htab_elem *lookup_nulls_elem_raw(struct hlist_nulls_head *head,
 634                                                u32 hash, void *key,
 635                                                u32 key_size, u32 n_buckets)
 636 {
 637         struct hlist_nulls_node *n;
 638         struct htab_elem *l;
 639
 640 again:
 641         hlist_nulls_for_each_entry_rcu(l, n, head, hash_node)
 642                 if (l->hash == hash && !memcmp(&l->key, key, key_size))
 643                         return l;
 644
 645         if (unlikely(get_nulls_value(n) != (hash & (n_buckets - 1))))
 646                 goto again;
 647
 648         return NULL;
 649 }
 650
 651 /* Called from syscall or from eBPF program directly, so
 652  * arguments have to match bpf_map_lookup_elem() exactly.
 653  * The return value is adjusted by BPF instructions
 654  * in htab_map_gen_lookup().
 655  */
 656 static void *__htab_map_lookup_elem(struct bpf_map *map, void *key)
 657 {
 658         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
 659         struct hlist_nulls_head *head;
 660         struct htab_elem *l;
 661         u32 hash, key_size;
 662
 663         WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
 664                      !rcu_read_lock_bh_held());
 665
 666         key_size = map->key_size;
 667
 668         hash = htab_map_hash(key, key_size, htab->hashrnd);
 669
 670         head = select_bucket(htab, hash);
 671
 672         l = lookup_nulls_elem_raw(head, hash, key, key_size, htab->n_buckets);
 673
 674         return l;
 675 }
 676
 677 static void *htab_map_lookup_elem(struct bpf_map *map, void *key)
 678 {
 679         struct htab_elem *l = __htab_map_lookup_elem(map, key);
 680
 681         if (l)
 682                 return l->key + round_up(map->key_size, 8);
 683
 684         return NULL;
 685 }
 686
 687 /* inline bpf_map_lookup_elem() call.
 688  * Instead of:
 689  * bpf_prog
 690  *   bpf_map_lookup_elem
 691  *     map->ops->map_lookup_elem
 692  *       htab_map_lookup_elem
 693  *         __htab_map_lookup_elem
 694  * do:
 695  * bpf_prog
 696  *   __htab_map_lookup_elem
 697  */
 698 static int htab_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
 699 {
 700         struct bpf_insn *insn = insn_buf;
 701         const int ret = BPF_REG_0;
 702
 703         BUILD_BUG_ON(!__same_type(&__htab_map_lookup_elem,
 704                      (void *(*)(struct bpf_map *map, void *key))NULL));
 705         *insn++ = BPF_EMIT_CALL(__htab_map_lookup_elem);
 706         *insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 1);
 707         *insn++ = BPF_ALU64_IMM(BPF_ADD, ret,
 708                                 offsetof(struct htab_elem, key) +
 709                                 round_up(map->key_size, 8));
 710         return insn - insn_buf;
 711 }
 712
 713 static __always_inline void *__htab_lru_map_lookup_elem(struct bpf_map *map,
 714                                                         void *key, const bool mark)
 715 {
 716         struct htab_elem *l = __htab_map_lookup_elem(map, key);
 717
 718         if (l) {
 719                 if (mark)
 720                         bpf_lru_node_set_ref(&l->lru_node);
 721                 return l->key + round_up(map->key_size, 8);
 722         }
 723
 724         return NULL;
 725 }
 726
 727 static void *htab_lru_map_lookup_elem(struct bpf_map *map, void *key)
 728 {
 729         return __htab_lru_map_lookup_elem(map, key, true);
 730 }
 731
 732 static void *htab_lru_map_lookup_elem_sys(struct bpf_map *map, void *key)
 733 {
 734         return __htab_lru_map_lookup_elem(map, key, false);
 735 }
 736
 737 static int htab_lru_map_gen_lookup(struct bpf_map *map,
 738                                    struct bpf_insn *insn_buf)
 739 {
 740         struct bpf_insn *insn = insn_buf;
 741         const int ret = BPF_REG_0;
 742         const int ref_reg = BPF_REG_1;
 743
 744         BUILD_BUG_ON(!__same_type(&__htab_map_lookup_elem,
 745                      (void *(*)(struct bpf_map *map, void *key))NULL));
 746         *insn++ = BPF_EMIT_CALL(__htab_map_lookup_elem);
 747         *insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 4);
 748         *insn++ = BPF_LDX_MEM(BPF_B, ref_reg, ret,
 749                               offsetof(struct htab_elem, lru_node) +
 750                               offsetof(struct bpf_lru_node, ref));
 751         *insn++ = BPF_JMP_IMM(BPF_JNE, ref_reg, 0, 1);
 752         *insn++ = BPF_ST_MEM(BPF_B, ret,
 753                              offsetof(struct htab_elem, lru_node) +
 754                              offsetof(struct bpf_lru_node, ref),
 755                              1);
 756         *insn++ = BPF_ALU64_IMM(BPF_ADD, ret,
 757                                 offsetof(struct htab_elem, key) +
 758                                 round_up(map->key_size, 8));
 759         return insn - insn_buf;
 760 }
 761
 762 static void check_and_free_fields(struct bpf_htab *htab,
 763                                   struct htab_elem *elem)
 764 {
 765         void *map_value = elem->key + round_up(htab->map.key_size, 8);
 766
 767         if (map_value_has_timer(&htab->map))
 768                 bpf_timer_cancel_and_free(map_value + htab->map.timer_off);
 769         if (map_value_has_kptrs(&htab->map))
 770                 bpf_map_free_kptrs(&htab->map, map_value);
 771 }
 772
 773 /* It is called from the bpf_lru_list when the LRU needs to delete
 774  * older elements from the htab.
 775  */
 776 static bool htab_lru_map_delete_node(void *arg, struct bpf_lru_node *node)
 777 {
 778         struct bpf_htab *htab = arg;
 779         struct htab_elem *l = NULL, *tgt_l;
 780         struct hlist_nulls_head *head;
 781         struct hlist_nulls_node *n;
 782         unsigned long flags;
 783         struct bucket *b;
 784         int ret;
 785
 786         tgt_l = container_of(node, struct htab_elem, lru_node);
 787         b = __select_bucket(htab, tgt_l->hash);
 788         head = &b->head;
 789
 790         ret = htab_lock_bucket(htab, b, tgt_l->hash, &flags);
 791         if (ret)
 792                 return false;
 793
 794         hlist_nulls_for_each_entry_rcu(l, n, head, hash_node)
 795                 if (l == tgt_l) {
 796                         hlist_nulls_del_rcu(&l->hash_node);
 797                         check_and_free_fields(htab, l);
 798                         break;
 799                 }
 800
 801         htab_unlock_bucket(htab, b, tgt_l->hash, flags);
 802
 803         return l == tgt_l;
 804 }
 805
 806 /* Called from syscall */
 807 static int htab_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
 808 {
 809         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
 810         struct hlist_nulls_head *head;
 811         struct htab_elem *l, *next_l;
 812         u32 hash, key_size;
 813         int i = 0;
 814
 815         WARN_ON_ONCE(!rcu_read_lock_held());
 816
 817         key_size = map->key_size;
 818
 819         if (!key)
 820                 goto find_first_elem;
 821
 822         hash = htab_map_hash(key, key_size, htab->hashrnd);
 823
 824         head = select_bucket(htab, hash);
 825
 826         /* lookup the key */
 827         l = lookup_nulls_elem_raw(head, hash, key, key_size, htab->n_buckets);
 828
 829         if (!l)
 830                 goto find_first_elem;
 831
 832         /* key was found, get next key in the same bucket */
 833         next_l = hlist_nulls_entry_safe(rcu_dereference_raw(hlist_nulls_next_rcu(&l->hash_node)),
 834                                   struct htab_elem, hash_node);
 835
 836         if (next_l) {
 837                 /* if next elem in this hash list is non-zero, just return it */
 838                 memcpy(next_key, next_l->key, key_size);
 839                 return 0;
 840         }
 841
 842         /* no more elements in this hash list, go to the next bucket */
 843         i = hash & (htab->n_buckets - 1);
 844         i++;
 845
 846 find_first_elem:
 847         /* iterate over buckets */
 848         for (; i < htab->n_buckets; i++) {
 849                 head = select_bucket(htab, i);
 850
 851                 /* pick first element in the bucket */
 852                 next_l = hlist_nulls_entry_safe(rcu_dereference_raw(hlist_nulls_first_rcu(head)),
 853                                           struct htab_elem, hash_node);
 854                 if (next_l) {
 855                         /* if it's not empty, just return it */
 856                         memcpy(next_key, next_l->key, key_size);
 857                         return 0;
 858                 }
 859         }
 860
 861         /* iterated over all buckets and all elements */
 862         return -ENOENT;
 863 }
 864
 865 static void htab_elem_free(struct bpf_htab *htab, struct htab_elem *l)
 866 {
 867         if (htab->map.map_type == BPF_MAP_TYPE_PERCPU_HASH)
 868                 bpf_mem_cache_free(&htab->pcpu_ma, l->ptr_to_pptr);
 869         check_and_free_fields(htab, l);
 870         bpf_mem_cache_free(&htab->ma, l);
 871 }
 872
 873 static void htab_put_fd_value(struct bpf_htab *htab, struct htab_elem *l)
 874 {
 875         struct bpf_map *map = &htab->map;
 876         void *ptr;
 877
 878         if (map->ops->map_fd_put_ptr) {
 879                 ptr = fd_htab_map_get_ptr(map, l);
 880                 map->ops->map_fd_put_ptr(ptr);
 881         }
 882 }
 883
 884 static bool is_map_full(struct bpf_htab *htab)
 885 {
 886         if (htab->use_percpu_counter)
 887                 return __percpu_counter_compare(&htab->pcount, htab->map.max_entries,
 888                                                 PERCPU_COUNTER_BATCH) >= 0;
 889         return atomic_read(&htab->count) >= htab->map.max_entries;
 890 }
 891
 892 static void inc_elem_count(struct bpf_htab *htab)
 893 {
 894         if (htab->use_percpu_counter)
 895                 percpu_counter_add_batch(&htab->pcount, 1, PERCPU_COUNTER_BATCH);
 896         else
 897                 atomic_inc(&htab->count);
 898 }
 899
 900 static void dec_elem_count(struct bpf_htab *htab)
 901 {
 902         if (htab->use_percpu_counter)
 903                 percpu_counter_add_batch(&htab->pcount, -1, PERCPU_COUNTER_BATCH);
 904         else
 905                 atomic_dec(&htab->count);
 906 }
 907
 908
 909 static void free_htab_elem(struct bpf_htab *htab, struct htab_elem *l)
 910 {
 911         htab_put_fd_value(htab, l);
 912
 913         if (htab_is_prealloc(htab)) {
 914                 check_and_free_fields(htab, l);
 915                 __pcpu_freelist_push(&htab->freelist, &l->fnode);
 916         } else {
 917                 dec_elem_count(htab);
 918                 htab_elem_free(htab, l);
 919         }
 920 }
 921
 922 static void pcpu_copy_value(struct bpf_htab *htab, void __percpu *pptr,
 923                             void *value, bool onallcpus)
 924 {
 925         if (!onallcpus) {
 926                 /* copy true value_size bytes */
 927                 memcpy(this_cpu_ptr(pptr), value, htab->map.value_size);
 928         } else {
 929                 u32 size = round_up(htab->map.value_size, 8);
 930                 int off = 0, cpu;
 931
 932                 for_each_possible_cpu(cpu) {
 933                         bpf_long_memcpy(per_cpu_ptr(pptr, cpu),
 934                                         value + off, size);
 935                         off += size;
 936                 }
 937         }
 938 }
 939
 940 static void pcpu_init_value(struct bpf_htab *htab, void __percpu *pptr,
 941                             void *value, bool onallcpus)
 942 {
 943         /* When not setting the initial value on all cpus, zero-fill element
 944          * values for other cpus. Otherwise, bpf program has no way to ensure
 945          * known initial values for cpus other than current one
 946          * (onallcpus=false always when coming from bpf prog).
 947          */
 948         if (!onallcpus) {
 949                 u32 size = round_up(htab->map.value_size, 8);
 950                 int current_cpu = raw_smp_processor_id();
 951                 int cpu;
 952
 953                 for_each_possible_cpu(cpu) {
 954                         if (cpu == current_cpu)
 955                                 bpf_long_memcpy(per_cpu_ptr(pptr, cpu), value,
 956                                                 size);
 957                         else
 958                                 memset(per_cpu_ptr(pptr, cpu), 0, size);
 959                 }
 960         } else {
 961                 pcpu_copy_value(htab, pptr, value, onallcpus);
 962         }
 963 }
 964
 965 static bool fd_htab_map_needs_adjust(const struct bpf_htab *htab)
 966 {
 967         return htab->map.map_type == BPF_MAP_TYPE_HASH_OF_MAPS &&
 968                BITS_PER_LONG == 64;
 969 }
 970
 971 static struct htab_elem *alloc_htab_elem(struct bpf_htab *htab, void *key,
 972                                          void *value, u32 key_size, u32 hash,
 973                                          bool percpu, bool onallcpus,
 974                                          struct htab_elem *old_elem)
 975 {
 976         u32 size = htab->map.value_size;
 977         bool prealloc = htab_is_prealloc(htab);
 978         struct htab_elem *l_new, **pl_new;
 979         void __percpu *pptr;
 980
 981         if (prealloc) {
 982                 if (old_elem) {
 983                         /* if we're updating the existing element,
 984                          * use per-cpu extra elems to avoid freelist_pop/push
 985                          */
 986                         pl_new = this_cpu_ptr(htab->extra_elems);
 987                         l_new = *pl_new;
 988                         htab_put_fd_value(htab, old_elem);
 989                         *pl_new = old_elem;
 990                 } else {
 991                         struct pcpu_freelist_node *l;
 992
 993                         l = __pcpu_freelist_pop(&htab->freelist);
 994                         if (!l)
 995                                 return ERR_PTR(-E2BIG);
 996                         l_new = container_of(l, struct htab_elem, fnode);
 997                 }
 998         } else {
 999                 if (is_map_full(htab))
1000                         if (!old_elem)
1001                                 /* when map is full and update() is replacing
1002                                  * old element, it's ok to allocate, since
1003                                  * old element will be freed immediately.
1004                                  * Otherwise return an error
1005                                  */
1006                                 return ERR_PTR(-E2BIG);
1007                 inc_elem_count(htab);
1008                 l_new = bpf_mem_cache_alloc(&htab->ma);
1009                 if (!l_new) {
1010                         l_new = ERR_PTR(-ENOMEM);
1011                         goto dec_count;
1012                 }
1013                 check_and_init_map_value(&htab->map,
1014                                          l_new->key + round_up(key_size, 8));
1015         }
1016
1017         memcpy(l_new->key, key, key_size);
1018         if (percpu) {
1019                 if (prealloc) {
1020                         pptr = htab_elem_get_ptr(l_new, key_size);
1021                 } else {
1022                         /* alloc_percpu zero-fills */
1023                         pptr = bpf_mem_cache_alloc(&htab->pcpu_ma);
1024                         if (!pptr) {
1025                                 bpf_mem_cache_free(&htab->ma, l_new);
1026                                 l_new = ERR_PTR(-ENOMEM);
1027                                 goto dec_count;
1028                         }
1029                         l_new->ptr_to_pptr = pptr;
1030                         pptr = *(void **)pptr;
1031                 }
1032
1033                 pcpu_init_value(htab, pptr, value, onallcpus);
1034
1035                 if (!prealloc)
1036                         htab_elem_set_ptr(l_new, key_size, pptr);
1037         } else if (fd_htab_map_needs_adjust(htab)) {
1038                 size = round_up(size, 8);
1039                 memcpy(l_new->key + round_up(key_size, 8), value, size);
1040         } else {
1041                 copy_map_value(&htab->map,
1042                                l_new->key + round_up(key_size, 8),
1043                                value);
1044         }
1045
1046         l_new->hash = hash;
1047         return l_new;
1048 dec_count:
1049         dec_elem_count(htab);
1050         return l_new;
1051 }
1052
1053 static int check_flags(struct bpf_htab *htab, struct htab_elem *l_old,
1054                        u64 map_flags)
1055 {
1056         if (l_old && (map_flags & ~BPF_F_LOCK) == BPF_NOEXIST)
1057                 /* elem already exists */
1058                 return -EEXIST;
1059
1060         if (!l_old && (map_flags & ~BPF_F_LOCK) == BPF_EXIST)
1061                 /* elem doesn't exist, cannot update it */
1062                 return -ENOENT;
1063
1064         return 0;
1065 }
1066
1067 /* Called from syscall or from eBPF program */
1068 static int htab_map_update_elem(struct bpf_map *map, void *key, void *value,
1069                                 u64 map_flags)
1070 {
1071         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1072         struct htab_elem *l_new = NULL, *l_old;
1073         struct hlist_nulls_head *head;
1074         unsigned long flags;
1075         struct bucket *b;
1076         u32 key_size, hash;
1077         int ret;
1078
1079         if (unlikely((map_flags & ~BPF_F_LOCK) > BPF_EXIST))
1080                 /* unknown flags */
1081                 return -EINVAL;
1082
1083         WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1084                      !rcu_read_lock_bh_held());
1085
1086         key_size = map->key_size;
1087
1088         hash = htab_map_hash(key, key_size, htab->hashrnd);
1089
1090         b = __select_bucket(htab, hash);
1091         head = &b->head;
1092
1093         if (unlikely(map_flags & BPF_F_LOCK)) {
1094                 if (unlikely(!map_value_has_spin_lock(map)))
1095                         return -EINVAL;
1096                 /* find an element without taking the bucket lock */
1097                 l_old = lookup_nulls_elem_raw(head, hash, key, key_size,
1098                                               htab->n_buckets);
1099                 ret = check_flags(htab, l_old, map_flags);
1100                 if (ret)
1101                         return ret;
1102                 if (l_old) {
1103                         /* grab the element lock and update value in place */
1104                         copy_map_value_locked(map,
1105                                               l_old->key + round_up(key_size, 8),
1106                                               value, false);
1107                         return 0;
1108                 }
1109                 /* fall through, grab the bucket lock and lookup again.
1110                  * 99.9% chance that the element won't be found,
1111                  * but second lookup under lock has to be done.
1112                  */
1113         }
1114
1115         ret = htab_lock_bucket(htab, b, hash, &flags);
1116         if (ret)
1117                 return ret;
1118
1119         l_old = lookup_elem_raw(head, hash, key, key_size);
1120
1121         ret = check_flags(htab, l_old, map_flags);
1122         if (ret)
1123                 goto err;
1124
1125         if (unlikely(l_old && (map_flags & BPF_F_LOCK))) {
1126                 /* first lookup without the bucket lock didn't find the element,
1127                  * but second lookup with the bucket lock found it.
1128                  * This case is highly unlikely, but has to be dealt with:
1129                  * grab the element lock in addition to the bucket lock
1130                  * and update element in place
1131                  */
1132                 copy_map_value_locked(map,
1133                                       l_old->key + round_up(key_size, 8),
1134                                       value, false);
1135                 ret = 0;
1136                 goto err;
1137         }
1138
1139         l_new = alloc_htab_elem(htab, key, value, key_size, hash, false, false,
1140                                 l_old);
1141         if (IS_ERR(l_new)) {
1142                 /* all pre-allocated elements are in use or memory exhausted */
1143                 ret = PTR_ERR(l_new);
1144                 goto err;
1145         }
1146
1147         /* add new element to the head of the list, so that
1148          * concurrent search will find it before old elem
1149          */
1150         hlist_nulls_add_head_rcu(&l_new->hash_node, head);
1151         if (l_old) {
1152                 hlist_nulls_del_rcu(&l_old->hash_node);
1153                 if (!htab_is_prealloc(htab))
1154                         free_htab_elem(htab, l_old);
1155                 else
1156                         check_and_free_fields(htab, l_old);
1157         }
1158         ret = 0;
1159 err:
1160         htab_unlock_bucket(htab, b, hash, flags);
1161         return ret;
1162 }
1163
1164 static void htab_lru_push_free(struct bpf_htab *htab, struct htab_elem *elem)
1165 {
1166         check_and_free_fields(htab, elem);
1167         bpf_lru_push_free(&htab->lru, &elem->lru_node);
1168 }
1169
1170 static int htab_lru_map_update_elem(struct bpf_map *map, void *key, void *value,
1171                                     u64 map_flags)
1172 {
1173         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1174         struct htab_elem *l_new, *l_old = NULL;
1175         struct hlist_nulls_head *head;
1176         unsigned long flags;
1177         struct bucket *b;
1178         u32 key_size, hash;
1179         int ret;
1180
1181         if (unlikely(map_flags > BPF_EXIST))
1182                 /* unknown flags */
1183                 return -EINVAL;
1184
1185         WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1186                      !rcu_read_lock_bh_held());
1187
1188         key_size = map->key_size;
1189
1190         hash = htab_map_hash(key, key_size, htab->hashrnd);
1191
1192         b = __select_bucket(htab, hash);
1193         head = &b->head;
1194
1195         /* For LRU, we need to alloc before taking bucket's
1196          * spinlock because getting free nodes from LRU may need
1197          * to remove older elements from htab and this removal
1198          * operation will need a bucket lock.
1199          */
1200         l_new = prealloc_lru_pop(htab, key, hash);
1201         if (!l_new)
1202                 return -ENOMEM;
1203         copy_map_value(&htab->map,
1204                        l_new->key + round_up(map->key_size, 8), value);
1205
1206         ret = htab_lock_bucket(htab, b, hash, &flags);
1207         if (ret)
1208                 return ret;
1209
1210         l_old = lookup_elem_raw(head, hash, key, key_size);
1211
1212         ret = check_flags(htab, l_old, map_flags);
1213         if (ret)
1214                 goto err;
1215
1216         /* add new element to the head of the list, so that
1217          * concurrent search will find it before old elem
1218          */
1219         hlist_nulls_add_head_rcu(&l_new->hash_node, head);
1220         if (l_old) {
1221                 bpf_lru_node_set_ref(&l_new->lru_node);
1222                 hlist_nulls_del_rcu(&l_old->hash_node);
1223         }
1224         ret = 0;
1225
1226 err:
1227         htab_unlock_bucket(htab, b, hash, flags);
1228
1229         if (ret)
1230                 htab_lru_push_free(htab, l_new);
1231         else if (l_old)
1232                 htab_lru_push_free(htab, l_old);
1233
1234         return ret;
1235 }
1236
1237 static int __htab_percpu_map_update_elem(struct bpf_map *map, void *key,
1238                                          void *value, u64 map_flags,
1239                                          bool onallcpus)
1240 {
1241         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1242         struct htab_elem *l_new = NULL, *l_old;
1243         struct hlist_nulls_head *head;
1244         unsigned long flags;
1245         struct bucket *b;
1246         u32 key_size, hash;
1247         int ret;
1248
1249         if (unlikely(map_flags > BPF_EXIST))
1250                 /* unknown flags */
1251                 return -EINVAL;
1252
1253         WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1254                      !rcu_read_lock_bh_held());
1255
1256         key_size = map->key_size;
1257
1258         hash = htab_map_hash(key, key_size, htab->hashrnd);
1259
1260         b = __select_bucket(htab, hash);
1261         head = &b->head;
1262
1263         ret = htab_lock_bucket(htab, b, hash, &flags);
1264         if (ret)
1265                 return ret;
1266
1267         l_old = lookup_elem_raw(head, hash, key, key_size);
1268
1269         ret = check_flags(htab, l_old, map_flags);
1270         if (ret)
1271                 goto err;
1272
1273         if (l_old) {
1274                 /* per-cpu hash map can update value in-place */
1275                 pcpu_copy_value(htab, htab_elem_get_ptr(l_old, key_size),
1276                                 value, onallcpus);
1277         } else {
1278                 l_new = alloc_htab_elem(htab, key, value, key_size,
1279                                         hash, true, onallcpus, NULL);
1280                 if (IS_ERR(l_new)) {
1281                         ret = PTR_ERR(l_new);
1282                         goto err;
1283                 }
1284                 hlist_nulls_add_head_rcu(&l_new->hash_node, head);
1285         }
1286         ret = 0;
1287 err:
1288         htab_unlock_bucket(htab, b, hash, flags);
1289         return ret;
1290 }
1291
1292 static int __htab_lru_percpu_map_update_elem(struct bpf_map *map, void *key,
1293                                              void *value, u64 map_flags,
1294                                              bool onallcpus)
1295 {
1296         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1297         struct htab_elem *l_new = NULL, *l_old;
1298         struct hlist_nulls_head *head;
1299         unsigned long flags;
1300         struct bucket *b;
1301         u32 key_size, hash;
1302         int ret;
1303
1304         if (unlikely(map_flags > BPF_EXIST))
1305                 /* unknown flags */
1306                 return -EINVAL;
1307
1308         WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1309                      !rcu_read_lock_bh_held());
1310
1311         key_size = map->key_size;
1312
1313         hash = htab_map_hash(key, key_size, htab->hashrnd);
1314
1315         b = __select_bucket(htab, hash);
1316         head = &b->head;
1317
1318         /* For LRU, we need to alloc before taking bucket's
1319          * spinlock because LRU's elem alloc may need
1320          * to remove older elem from htab and this removal
1321          * operation will need a bucket lock.
1322          */
1323         if (map_flags != BPF_EXIST) {
1324                 l_new = prealloc_lru_pop(htab, key, hash);
1325                 if (!l_new)
1326                         return -ENOMEM;
1327         }
1328
1329         ret = htab_lock_bucket(htab, b, hash, &flags);
1330         if (ret)
1331                 return ret;
1332
1333         l_old = lookup_elem_raw(head, hash, key, key_size);
1334
1335         ret = check_flags(htab, l_old, map_flags);
1336         if (ret)
1337                 goto err;
1338
1339         if (l_old) {
1340                 bpf_lru_node_set_ref(&l_old->lru_node);
1341
1342                 /* per-cpu hash map can update value in-place */
1343                 pcpu_copy_value(htab, htab_elem_get_ptr(l_old, key_size),
1344                                 value, onallcpus);
1345         } else {
1346                 pcpu_init_value(htab, htab_elem_get_ptr(l_new, key_size),
1347                                 value, onallcpus);
1348                 hlist_nulls_add_head_rcu(&l_new->hash_node, head);
1349                 l_new = NULL;
1350         }
1351         ret = 0;
1352 err:
1353         htab_unlock_bucket(htab, b, hash, flags);
1354         if (l_new)
1355                 bpf_lru_push_free(&htab->lru, &l_new->lru_node);
1356         return ret;
1357 }
1358
1359 static int htab_percpu_map_update_elem(struct bpf_map *map, void *key,
1360                                        void *value, u64 map_flags)
1361 {
1362         return __htab_percpu_map_update_elem(map, key, value, map_flags, false);
1363 }
1364
1365 static int htab_lru_percpu_map_update_elem(struct bpf_map *map, void *key,
1366                                            void *value, u64 map_flags)
1367 {
1368         return __htab_lru_percpu_map_update_elem(map, key, value, map_flags,
1369                                                  false);
1370 }
1371
1372 /* Called from syscall or from eBPF program */
1373 static int htab_map_delete_elem(struct bpf_map *map, void *key)
1374 {
1375         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1376         struct hlist_nulls_head *head;
1377         struct bucket *b;
1378         struct htab_elem *l;
1379         unsigned long flags;
1380         u32 hash, key_size;
1381         int ret;
1382
1383         WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1384                      !rcu_read_lock_bh_held());
1385
1386         key_size = map->key_size;
1387
1388         hash = htab_map_hash(key, key_size, htab->hashrnd);
1389         b = __select_bucket(htab, hash);
1390         head = &b->head;
1391
1392         ret = htab_lock_bucket(htab, b, hash, &flags);
1393         if (ret)
1394                 return ret;
1395
1396         l = lookup_elem_raw(head, hash, key, key_size);
1397
1398         if (l) {
1399                 hlist_nulls_del_rcu(&l->hash_node);
1400                 free_htab_elem(htab, l);
1401         } else {
1402                 ret = -ENOENT;
1403         }
1404
1405         htab_unlock_bucket(htab, b, hash, flags);
1406         return ret;
1407 }
1408
1409 static int htab_lru_map_delete_elem(struct bpf_map *map, void *key)
1410 {
1411         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1412         struct hlist_nulls_head *head;
1413         struct bucket *b;
1414         struct htab_elem *l;
1415         unsigned long flags;
1416         u32 hash, key_size;
1417         int ret;
1418
1419         WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1420                      !rcu_read_lock_bh_held());
1421
1422         key_size = map->key_size;
1423
1424         hash = htab_map_hash(key, key_size, htab->hashrnd);
1425         b = __select_bucket(htab, hash);
1426         head = &b->head;
1427
1428         ret = htab_lock_bucket(htab, b, hash, &flags);
1429         if (ret)
1430                 return ret;
1431
1432         l = lookup_elem_raw(head, hash, key, key_size);
1433
1434         if (l)
1435                 hlist_nulls_del_rcu(&l->hash_node);
1436         else
1437                 ret = -ENOENT;
1438
1439         htab_unlock_bucket(htab, b, hash, flags);
1440         if (l)
1441                 htab_lru_push_free(htab, l);
1442         return ret;
1443 }
1444
1445 static void delete_all_elements(struct bpf_htab *htab)
1446 {
1447         int i;
1448
1449         /* It's called from a worker thread, so disable migration here,
1450          * since bpf_mem_cache_free() relies on that.
1451          */
1452         migrate_disable();
1453         for (i = 0; i < htab->n_buckets; i++) {
1454                 struct hlist_nulls_head *head = select_bucket(htab, i);
1455                 struct hlist_nulls_node *n;
1456                 struct htab_elem *l;
1457
1458                 hlist_nulls_for_each_entry_safe(l, n, head, hash_node) {
1459                         hlist_nulls_del_rcu(&l->hash_node);
1460                         htab_elem_free(htab, l);
1461                 }
1462         }
1463         migrate_enable();
1464 }
1465
1466 static void htab_free_malloced_timers(struct bpf_htab *htab)
1467 {
1468         int i;
1469
1470         rcu_read_lock();
1471         for (i = 0; i < htab->n_buckets; i++) {
1472                 struct hlist_nulls_head *head = select_bucket(htab, i);
1473                 struct hlist_nulls_node *n;
1474                 struct htab_elem *l;
1475
1476                 hlist_nulls_for_each_entry(l, n, head, hash_node) {
1477                         /* We don't reset or free kptr on uref dropping to zero,
1478                          * hence just free timer.
1479                          */
1480                         bpf_timer_cancel_and_free(l->key +
1481                                                   round_up(htab->map.key_size, 8) +
1482                                                   htab->map.timer_off);
1483                 }
1484                 cond_resched_rcu();
1485         }
1486         rcu_read_unlock();
1487 }
1488
1489 static void htab_map_free_timers(struct bpf_map *map)
1490 {
1491         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1492
1493         /* We don't reset or free kptr on uref dropping to zero. */
1494         if (!map_value_has_timer(&htab->map))
1495                 return;
1496         if (!htab_is_prealloc(htab))
1497                 htab_free_malloced_timers(htab);
1498         else
1499                 htab_free_prealloced_timers(htab);
1500 }
1501
1502 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
1503 static void htab_map_free(struct bpf_map *map)
1504 {
1505         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1506         int i;
1507
1508         /* bpf_free_used_maps() or close(map_fd) will trigger this map_free callback.
1509          * bpf_free_used_maps() is called after bpf prog is no longer executing.
1510          * There is no need to synchronize_rcu() here to protect map elements.
1511          */
1512
1513         /* htab no longer uses call_rcu() directly. bpf_mem_alloc does it
1514          * underneath and is reponsible for waiting for callbacks to finish
1515          * during bpf_mem_alloc_destroy().
1516          */
1517         if (!htab_is_prealloc(htab)) {
1518                 delete_all_elements(htab);
1519         } else {
1520                 htab_free_prealloced_kptrs(htab);
1521                 prealloc_destroy(htab);
1522         }
1523
1524         bpf_map_free_kptr_off_tab(map);
1525         free_percpu(htab->extra_elems);
1526         bpf_map_area_free(htab->buckets);
1527         bpf_mem_alloc_destroy(&htab->pcpu_ma);
1528         bpf_mem_alloc_destroy(&htab->ma);
1529         if (htab->use_percpu_counter)
1530                 percpu_counter_destroy(&htab->pcount);
1531         for (i = 0; i < HASHTAB_MAP_LOCK_COUNT; i++)
1532                 free_percpu(htab->map_locked[i]);
1533         lockdep_unregister_key(&htab->lockdep_key);
1534         bpf_map_area_free(htab);
1535 }
1536
1537 static void htab_map_seq_show_elem(struct bpf_map *map, void *key,
1538                                    struct seq_file *m)
1539 {
1540         void *value;
1541
1542         rcu_read_lock();
1543
1544         value = htab_map_lookup_elem(map, key);
1545         if (!value) {
1546                 rcu_read_unlock();
1547                 return;
1548         }
1549
1550         btf_type_seq_show(map->btf, map->btf_key_type_id, key, m);
1551         seq_puts(m, ": ");
1552         btf_type_seq_show(map->btf, map->btf_value_type_id, value, m);
1553         seq_puts(m, "\n");
1554
1555         rcu_read_unlock();
1556 }
1557
1558 static int __htab_map_lookup_and_delete_elem(struct bpf_map *map, void *key,
1559                                              void *value, bool is_lru_map,
1560                                              bool is_percpu, u64 flags)
1561 {
1562         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1563         struct hlist_nulls_head *head;
1564         unsigned long bflags;
1565         struct htab_elem *l;
1566         u32 hash, key_size;
1567         struct bucket *b;
1568         int ret;
1569
1570         key_size = map->key_size;
1571
1572         hash = htab_map_hash(key, key_size, htab->hashrnd);
1573         b = __select_bucket(htab, hash);
1574         head = &b->head;
1575
1576         ret = htab_lock_bucket(htab, b, hash, &bflags);
1577         if (ret)
1578                 return ret;
1579
1580         l = lookup_elem_raw(head, hash, key, key_size);
1581         if (!l) {
1582                 ret = -ENOENT;
1583         } else {
1584                 if (is_percpu) {
1585                         u32 roundup_value_size = round_up(map->value_size, 8);
1586                         void __percpu *pptr;
1587                         int off = 0, cpu;
1588
1589                         pptr = htab_elem_get_ptr(l, key_size);
1590                         for_each_possible_cpu(cpu) {
1591                                 bpf_long_memcpy(value + off,
1592                                                 per_cpu_ptr(pptr, cpu),
1593                                                 roundup_value_size);
1594                                 off += roundup_value_size;
1595                         }
1596                 } else {
1597                         u32 roundup_key_size = round_up(map->key_size, 8);
1598
1599                         if (flags & BPF_F_LOCK)
1600                                 copy_map_value_locked(map, value, l->key +
1601                                                       roundup_key_size,
1602                                                       true);
1603                         else
1604                                 copy_map_value(map, value, l->key +
1605                                                roundup_key_size);
1606                         check_and_init_map_value(map, value);
1607                 }
1608
1609                 hlist_nulls_del_rcu(&l->hash_node);
1610                 if (!is_lru_map)
1611                         free_htab_elem(htab, l);
1612         }
1613
1614         htab_unlock_bucket(htab, b, hash, bflags);
1615
1616         if (is_lru_map && l)
1617                 htab_lru_push_free(htab, l);
1618
1619         return ret;
1620 }
1621
1622 static int htab_map_lookup_and_delete_elem(struct bpf_map *map, void *key,
1623                                            void *value, u64 flags)
1624 {
1625         return __htab_map_lookup_and_delete_elem(map, key, value, false, false,
1626                                                  flags);
1627 }
1628
1629 static int htab_percpu_map_lookup_and_delete_elem(struct bpf_map *map,
1630                                                   void *key, void *value,
1631                                                   u64 flags)
1632 {
1633         return __htab_map_lookup_and_delete_elem(map, key, value, false, true,
1634                                                  flags);
1635 }
1636
1637 static int htab_lru_map_lookup_and_delete_elem(struct bpf_map *map, void *key,
1638                                                void *value, u64 flags)
1639 {
1640         return __htab_map_lookup_and_delete_elem(map, key, value, true, false,
1641                                                  flags);
1642 }
1643
1644 static int htab_lru_percpu_map_lookup_and_delete_elem(struct bpf_map *map,
1645                                                       void *key, void *value,
1646                                                       u64 flags)
1647 {
1648         return __htab_map_lookup_and_delete_elem(map, key, value, true, true,
1649                                                  flags);
1650 }
1651
1652 static int
1653 __htab_map_lookup_and_delete_batch(struct bpf_map *map,
1654                                    const union bpf_attr *attr,
1655                                    union bpf_attr __user *uattr,
1656                                    bool do_delete, bool is_lru_map,
1657                                    bool is_percpu)
1658 {
1659         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1660         u32 bucket_cnt, total, key_size, value_size, roundup_key_size;
1661         void *keys = NULL, *values = NULL, *value, *dst_key, *dst_val;
1662         void __user *uvalues = u64_to_user_ptr(attr->batch.values);
1663         void __user *ukeys = u64_to_user_ptr(attr->batch.keys);
1664         void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1665         u32 batch, max_count, size, bucket_size, map_id;
1666         struct htab_elem *node_to_free = NULL;
1667         u64 elem_map_flags, map_flags;
1668         struct hlist_nulls_head *head;
1669         struct hlist_nulls_node *n;
1670         unsigned long flags = 0;
1671         bool locked = false;
1672         struct htab_elem *l;
1673         struct bucket *b;
1674         int ret = 0;
1675
1676         elem_map_flags = attr->batch.elem_flags;
1677         if ((elem_map_flags & ~BPF_F_LOCK) ||
1678             ((elem_map_flags & BPF_F_LOCK) && !map_value_has_spin_lock(map)))
1679                 return -EINVAL;
1680
1681         map_flags = attr->batch.flags;
1682         if (map_flags)
1683                 return -EINVAL;
1684
1685         max_count = attr->batch.count;
1686         if (!max_count)
1687                 return 0;
1688
1689         if (put_user(0, &uattr->batch.count))
1690                 return -EFAULT;
1691
1692         batch = 0;
1693         if (ubatch && copy_from_user(&batch, ubatch, sizeof(batch)))
1694                 return -EFAULT;
1695
1696         if (batch >= htab->n_buckets)
1697                 return -ENOENT;
1698
1699         key_size = htab->map.key_size;
1700         roundup_key_size = round_up(htab->map.key_size, 8);
1701         value_size = htab->map.value_size;
1702         size = round_up(value_size, 8);
1703         if (is_percpu)
1704                 value_size = size * num_possible_cpus();
1705         total = 0;
1706         /* while experimenting with hash tables with sizes ranging from 10 to
1707          * 1000, it was observed that a bucket can have up to 5 entries.
1708          */
1709         bucket_size = 5;
1710
1711 alloc:
1712         /* We cannot do copy_from_user or copy_to_user inside
1713          * the rcu_read_lock. Allocate enough space here.
1714          */
1715         keys = kvmalloc_array(key_size, bucket_size, GFP_USER | __GFP_NOWARN);
1716         values = kvmalloc_array(value_size, bucket_size, GFP_USER | __GFP_NOWARN);
1717         if (!keys || !values) {
1718                 ret = -ENOMEM;
1719                 goto after_loop;
1720         }
1721
1722 again:
1723         bpf_disable_instrumentation();
1724         rcu_read_lock();
1725 again_nocopy:
1726         dst_key = keys;
1727         dst_val = values;
1728         b = &htab->buckets[batch];
1729         head = &b->head;
1730         /* do not grab the lock unless need it (bucket_cnt > 0). */
1731         if (locked) {
1732                 ret = htab_lock_bucket(htab, b, batch, &flags);
1733                 if (ret) {
1734                         rcu_read_unlock();
1735                         bpf_enable_instrumentation();
1736                         goto after_loop;
1737                 }
1738         }
1739
1740         bucket_cnt = 0;
1741         hlist_nulls_for_each_entry_rcu(l, n, head, hash_node)
1742                 bucket_cnt++;
1743
1744         if (bucket_cnt && !locked) {
1745                 locked = true;
1746                 goto again_nocopy;
1747         }
1748
1749         if (bucket_cnt > (max_count - total)) {
1750                 if (total == 0)
1751                         ret = -ENOSPC;
1752                 /* Note that since bucket_cnt > 0 here, it is implicit
1753                  * that the locked was grabbed, so release it.
1754                  */
1755                 htab_unlock_bucket(htab, b, batch, flags);
1756                 rcu_read_unlock();
1757                 bpf_enable_instrumentation();
1758                 goto after_loop;
1759         }
1760
1761         if (bucket_cnt > bucket_size) {
1762                 bucket_size = bucket_cnt;
1763                 /* Note that since bucket_cnt > 0 here, it is implicit
1764                  * that the locked was grabbed, so release it.
1765                  */
1766                 htab_unlock_bucket(htab, b, batch, flags);
1767                 rcu_read_unlock();
1768                 bpf_enable_instrumentation();
1769                 kvfree(keys);
1770                 kvfree(values);
1771                 goto alloc;
1772         }
1773
1774         /* Next block is only safe to run if you have grabbed the lock */
1775         if (!locked)
1776                 goto next_batch;
1777
1778         hlist_nulls_for_each_entry_safe(l, n, head, hash_node) {
1779                 memcpy(dst_key, l->key, key_size);
1780
1781                 if (is_percpu) {
1782                         int off = 0, cpu;
1783                         void __percpu *pptr;
1784
1785                         pptr = htab_elem_get_ptr(l, map->key_size);
1786                         for_each_possible_cpu(cpu) {
1787                                 bpf_long_memcpy(dst_val + off,
1788                                                 per_cpu_ptr(pptr, cpu), size);
1789                                 off += size;
1790                         }
1791                 } else {
1792                         value = l->key + roundup_key_size;
1793                         if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
1794                                 struct bpf_map **inner_map = value;
1795
1796                                  /* Actual value is the id of the inner map */
1797                                 map_id = map->ops->map_fd_sys_lookup_elem(*inner_map);
1798                                 value = &map_id;
1799                         }
1800
1801                         if (elem_map_flags & BPF_F_LOCK)
1802                                 copy_map_value_locked(map, dst_val, value,
1803                                                       true);
1804                         else
1805                                 copy_map_value(map, dst_val, value);
1806                         check_and_init_map_value(map, dst_val);
1807                 }
1808                 if (do_delete) {
1809                         hlist_nulls_del_rcu(&l->hash_node);
1810
1811                         /* bpf_lru_push_free() will acquire lru_lock, which
1812                          * may cause deadlock. See comments in function
1813                          * prealloc_lru_pop(). Let us do bpf_lru_push_free()
1814                          * after releasing the bucket lock.
1815                          */
1816                         if (is_lru_map) {
1817                                 l->batch_flink = node_to_free;
1818                                 node_to_free = l;
1819                         } else {
1820                                 free_htab_elem(htab, l);
1821                         }
1822                 }
1823                 dst_key += key_size;
1824                 dst_val += value_size;
1825         }
1826
1827         htab_unlock_bucket(htab, b, batch, flags);
1828         locked = false;
1829
1830         while (node_to_free) {
1831                 l = node_to_free;
1832                 node_to_free = node_to_free->batch_flink;
1833                 htab_lru_push_free(htab, l);
1834         }
1835
1836 next_batch:
1837         /* If we are not copying data, we can go to next bucket and avoid
1838          * unlocking the rcu.
1839          */
1840         if (!bucket_cnt && (batch + 1 < htab->n_buckets)) {
1841                 batch++;
1842                 goto again_nocopy;
1843         }
1844
1845         rcu_read_unlock();
1846         bpf_enable_instrumentation();
1847         if (bucket_cnt && (copy_to_user(ukeys + total * key_size, keys,
1848             key_size * bucket_cnt) ||
1849             copy_to_user(uvalues + total * value_size, values,
1850             value_size * bucket_cnt))) {
1851                 ret = -EFAULT;
1852                 goto after_loop;
1853         }
1854
1855         total += bucket_cnt;
1856         batch++;
1857         if (batch >= htab->n_buckets) {
1858                 ret = -ENOENT;
1859                 goto after_loop;
1860         }
1861         goto again;
1862
1863 after_loop:
1864         if (ret == -EFAULT)
1865                 goto out;
1866
1867         /* copy # of entries and next batch */
1868         ubatch = u64_to_user_ptr(attr->batch.out_batch);
1869         if (copy_to_user(ubatch, &batch, sizeof(batch)) ||
1870             put_user(total, &uattr->batch.count))
1871                 ret = -EFAULT;
1872
1873 out:
1874         kvfree(keys);
1875         kvfree(values);
1876         return ret;
1877 }
1878
1879 static int
1880 htab_percpu_map_lookup_batch(struct bpf_map *map, const union bpf_attr *attr,
1881                              union bpf_attr __user *uattr)
1882 {
1883         return __htab_map_lookup_and_delete_batch(map, attr, uattr, false,
1884                                                   false, true);
1885 }
1886
1887 static int
1888 htab_percpu_map_lookup_and_delete_batch(struct bpf_map *map,
1889                                         const union bpf_attr *attr,
1890                                         union bpf_attr __user *uattr)
1891 {
1892         return __htab_map_lookup_and_delete_batch(map, attr, uattr, true,
1893                                                   false, true);
1894 }
1895
1896 static int
1897 htab_map_lookup_batch(struct bpf_map *map, const union bpf_attr *attr,
1898                       union bpf_attr __user *uattr)
1899 {
1900         return __htab_map_lookup_and_delete_batch(map, attr, uattr, false,
1901                                                   false, false);
1902 }
1903
1904 static int
1905 htab_map_lookup_and_delete_batch(struct bpf_map *map,
1906                                  const union bpf_attr *attr,
1907                                  union bpf_attr __user *uattr)
1908 {
1909         return __htab_map_lookup_and_delete_batch(map, attr, uattr, true,
1910                                                   false, false);
1911 }
1912
1913 static int
1914 htab_lru_percpu_map_lookup_batch(struct bpf_map *map,
1915                                  const union bpf_attr *attr,
1916                                  union bpf_attr __user *uattr)
1917 {
1918         return __htab_map_lookup_and_delete_batch(map, attr, uattr, false,
1919                                                   true, true);
1920 }
1921
1922 static int
1923 htab_lru_percpu_map_lookup_and_delete_batch(struct bpf_map *map,
1924                                             const union bpf_attr *attr,
1925                                             union bpf_attr __user *uattr)
1926 {
1927         return __htab_map_lookup_and_delete_batch(map, attr, uattr, true,
1928                                                   true, true);
1929 }
1930
1931 static int
1932 htab_lru_map_lookup_batch(struct bpf_map *map, const union bpf_attr *attr,
1933                           union bpf_attr __user *uattr)
1934 {
1935         return __htab_map_lookup_and_delete_batch(map, attr, uattr, false,
1936                                                   true, false);
1937 }
1938
1939 static int
1940 htab_lru_map_lookup_and_delete_batch(struct bpf_map *map,
1941                                      const union bpf_attr *attr,
1942                                      union bpf_attr __user *uattr)
1943 {
1944         return __htab_map_lookup_and_delete_batch(map, attr, uattr, true,
1945                                                   true, false);
1946 }
1947
1948 struct bpf_iter_seq_hash_map_info {
1949         struct bpf_map *map;
1950         struct bpf_htab *htab;
1951         void *percpu_value_buf; // non-zero means percpu hash
1952         u32 bucket_id;
1953         u32 skip_elems;
1954 };
1955
1956 static struct htab_elem *
1957 bpf_hash_map_seq_find_next(struct bpf_iter_seq_hash_map_info *info,
1958                            struct htab_elem *prev_elem)
1959 {
1960         const struct bpf_htab *htab = info->htab;
1961         u32 skip_elems = info->skip_elems;
1962         u32 bucket_id = info->bucket_id;
1963         struct hlist_nulls_head *head;
1964         struct hlist_nulls_node *n;
1965         struct htab_elem *elem;
1966         struct bucket *b;
1967         u32 i, count;
1968
1969         if (bucket_id >= htab->n_buckets)
1970                 return NULL;
1971
1972         /* try to find next elem in the same bucket */
1973         if (prev_elem) {
1974                 /* no update/deletion on this bucket, prev_elem should be still valid
1975                  * and we won't skip elements.
1976                  */
1977                 n = rcu_dereference_raw(hlist_nulls_next_rcu(&prev_elem->hash_node));
1978                 elem = hlist_nulls_entry_safe(n, struct htab_elem, hash_node);
1979                 if (elem)
1980                         return elem;
1981
1982                 /* not found, unlock and go to the next bucket */
1983                 b = &htab->buckets[bucket_id++];
1984                 rcu_read_unlock();
1985                 skip_elems = 0;
1986         }
1987
1988         for (i = bucket_id; i < htab->n_buckets; i++) {
1989                 b = &htab->buckets[i];
1990                 rcu_read_lock();
1991
1992                 count = 0;
1993                 head = &b->head;
1994                 hlist_nulls_for_each_entry_rcu(elem, n, head, hash_node) {
1995                         if (count >= skip_elems) {
1996                                 info->bucket_id = i;
1997                                 info->skip_elems = count;
1998                                 return elem;
1999                         }
2000                         count++;
2001                 }
2002
2003                 rcu_read_unlock();
2004                 skip_elems = 0;
2005         }
2006
2007         info->bucket_id = i;
2008         info->skip_elems = 0;
2009         return NULL;
2010 }
2011
2012 static void *bpf_hash_map_seq_start(struct seq_file *seq, loff_t *pos)
2013 {
2014         struct bpf_iter_seq_hash_map_info *info = seq->private;
2015         struct htab_elem *elem;
2016
2017         elem = bpf_hash_map_seq_find_next(info, NULL);
2018         if (!elem)
2019                 return NULL;
2020
2021         if (*pos == 0)
2022                 ++*pos;
2023         return elem;
2024 }
2025
2026 static void *bpf_hash_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2027 {
2028         struct bpf_iter_seq_hash_map_info *info = seq->private;
2029
2030         ++*pos;
2031         ++info->skip_elems;
2032         return bpf_hash_map_seq_find_next(info, v);
2033 }
2034
2035 static int __bpf_hash_map_seq_show(struct seq_file *seq, struct htab_elem *elem)
2036 {
2037         struct bpf_iter_seq_hash_map_info *info = seq->private;
2038         u32 roundup_key_size, roundup_value_size;
2039         struct bpf_iter__bpf_map_elem ctx = {};
2040         struct bpf_map *map = info->map;
2041         struct bpf_iter_meta meta;
2042         int ret = 0, off = 0, cpu;
2043         struct bpf_prog *prog;
2044         void __percpu *pptr;
2045
2046         meta.seq = seq;
2047         prog = bpf_iter_get_info(&meta, elem == NULL);
2048         if (prog) {
2049                 ctx.meta = &meta;
2050                 ctx.map = info->map;
2051                 if (elem) {
2052                         roundup_key_size = round_up(map->key_size, 8);
2053                         ctx.key = elem->key;
2054                         if (!info->percpu_value_buf) {
2055                                 ctx.value = elem->key + roundup_key_size;
2056                         } else {
2057                                 roundup_value_size = round_up(map->value_size, 8);
2058                                 pptr = htab_elem_get_ptr(elem, map->key_size);
2059                                 for_each_possible_cpu(cpu) {
2060                                         bpf_long_memcpy(info->percpu_value_buf + off,
2061                                                         per_cpu_ptr(pptr, cpu),
2062                                                         roundup_value_size);
2063                                         off += roundup_value_size;
2064                                 }
2065                                 ctx.value = info->percpu_value_buf;
2066                         }
2067                 }
2068                 ret = bpf_iter_run_prog(prog, &ctx);
2069         }
2070
2071         return ret;
2072 }
2073
2074 static int bpf_hash_map_seq_show(struct seq_file *seq, void *v)
2075 {
2076         return __bpf_hash_map_seq_show(seq, v);
2077 }
2078
2079 static void bpf_hash_map_seq_stop(struct seq_file *seq, void *v)
2080 {
2081         if (!v)
2082                 (void)__bpf_hash_map_seq_show(seq, NULL);
2083         else
2084                 rcu_read_unlock();
2085 }
2086
2087 static int bpf_iter_init_hash_map(void *priv_data,
2088                                   struct bpf_iter_aux_info *aux)
2089 {
2090         struct bpf_iter_seq_hash_map_info *seq_info = priv_data;
2091         struct bpf_map *map = aux->map;
2092         void *value_buf;
2093         u32 buf_size;
2094
2095         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
2096             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
2097                 buf_size = round_up(map->value_size, 8) * num_possible_cpus();
2098                 value_buf = kmalloc(buf_size, GFP_USER | __GFP_NOWARN);
2099                 if (!value_buf)
2100                         return -ENOMEM;
2101
2102                 seq_info->percpu_value_buf = value_buf;
2103         }
2104
2105         bpf_map_inc_with_uref(map);
2106         seq_info->map = map;
2107         seq_info->htab = container_of(map, struct bpf_htab, map);
2108         return 0;
2109 }
2110
2111 static void bpf_iter_fini_hash_map(void *priv_data)
2112 {
2113         struct bpf_iter_seq_hash_map_info *seq_info = priv_data;
2114
2115         bpf_map_put_with_uref(seq_info->map);
2116         kfree(seq_info->percpu_value_buf);
2117 }
2118
2119 static const struct seq_operations bpf_hash_map_seq_ops = {
2120         .start  = bpf_hash_map_seq_start,
2121         .next   = bpf_hash_map_seq_next,
2122         .stop   = bpf_hash_map_seq_stop,
2123         .show   = bpf_hash_map_seq_show,
2124 };
2125
2126 static const struct bpf_iter_seq_info iter_seq_info = {
2127         .seq_ops                = &bpf_hash_map_seq_ops,
2128         .init_seq_private       = bpf_iter_init_hash_map,
2129         .fini_seq_private       = bpf_iter_fini_hash_map,
2130         .seq_priv_size          = sizeof(struct bpf_iter_seq_hash_map_info),
2131 };
2132
2133 static int bpf_for_each_hash_elem(struct bpf_map *map, bpf_callback_t callback_fn,
2134                                   void *callback_ctx, u64 flags)
2135 {
2136         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
2137         struct hlist_nulls_head *head;
2138         struct hlist_nulls_node *n;
2139         struct htab_elem *elem;
2140         u32 roundup_key_size;
2141         int i, num_elems = 0;
2142         void __percpu *pptr;
2143         struct bucket *b;
2144         void *key, *val;
2145         bool is_percpu;
2146         u64 ret = 0;
2147
2148         if (flags != 0)
2149                 return -EINVAL;
2150
2151         is_percpu = htab_is_percpu(htab);
2152
2153         roundup_key_size = round_up(map->key_size, 8);
2154         /* disable migration so percpu value prepared here will be the
2155          * same as the one seen by the bpf program with bpf_map_lookup_elem().
2156          */
2157         if (is_percpu)
2158                 migrate_disable();
2159         for (i = 0; i < htab->n_buckets; i++) {
2160                 b = &htab->buckets[i];
2161                 rcu_read_lock();
2162                 head = &b->head;
2163                 hlist_nulls_for_each_entry_rcu(elem, n, head, hash_node) {
2164                         key = elem->key;
2165                         if (is_percpu) {
2166                                 /* current cpu value for percpu map */
2167                                 pptr = htab_elem_get_ptr(elem, map->key_size);
2168                                 val = this_cpu_ptr(pptr);
2169                         } else {
2170                                 val = elem->key + roundup_key_size;
2171                         }
2172                         num_elems++;
2173                         ret = callback_fn((u64)(long)map, (u64)(long)key,
2174                                           (u64)(long)val, (u64)(long)callback_ctx, 0);
2175                         /* return value: 0 - continue, 1 - stop and return */
2176                         if (ret) {
2177                                 rcu_read_unlock();
2178                                 goto out;
2179                         }
2180                 }
2181                 rcu_read_unlock();
2182         }
2183 out:
2184         if (is_percpu)
2185                 migrate_enable();
2186         return num_elems;
2187 }
2188
2189 BTF_ID_LIST_SINGLE(htab_map_btf_ids, struct, bpf_htab)
2190 const struct bpf_map_ops htab_map_ops = {
2191         .map_meta_equal = bpf_map_meta_equal,
2192         .map_alloc_check = htab_map_alloc_check,
2193         .map_alloc = htab_map_alloc,
2194         .map_free = htab_map_free,
2195         .map_get_next_key = htab_map_get_next_key,
2196         .map_release_uref = htab_map_free_timers,
2197         .map_lookup_elem = htab_map_lookup_elem,
2198         .map_lookup_and_delete_elem = htab_map_lookup_and_delete_elem,
2199         .map_update_elem = htab_map_update_elem,
2200         .map_delete_elem = htab_map_delete_elem,
2201         .map_gen_lookup = htab_map_gen_lookup,
2202         .map_seq_show_elem = htab_map_seq_show_elem,
2203         .map_set_for_each_callback_args = map_set_for_each_callback_args,
2204         .map_for_each_callback = bpf_for_each_hash_elem,
2205         BATCH_OPS(htab),
2206         .map_btf_id = &htab_map_btf_ids[0],
2207         .iter_seq_info = &iter_seq_info,
2208 };
2209
2210 const struct bpf_map_ops htab_lru_map_ops = {
2211         .map_meta_equal = bpf_map_meta_equal,
2212         .map_alloc_check = htab_map_alloc_check,
2213         .map_alloc = htab_map_alloc,
2214         .map_free = htab_map_free,
2215         .map_get_next_key = htab_map_get_next_key,
2216         .map_release_uref = htab_map_free_timers,
2217         .map_lookup_elem = htab_lru_map_lookup_elem,
2218         .map_lookup_and_delete_elem = htab_lru_map_lookup_and_delete_elem,
2219         .map_lookup_elem_sys_only = htab_lru_map_lookup_elem_sys,
2220         .map_update_elem = htab_lru_map_update_elem,
2221         .map_delete_elem = htab_lru_map_delete_elem,
2222         .map_gen_lookup = htab_lru_map_gen_lookup,
2223         .map_seq_show_elem = htab_map_seq_show_elem,
2224         .map_set_for_each_callback_args = map_set_for_each_callback_args,
2225         .map_for_each_callback = bpf_for_each_hash_elem,
2226         BATCH_OPS(htab_lru),
2227         .map_btf_id = &htab_map_btf_ids[0],
2228         .iter_seq_info = &iter_seq_info,
2229 };
2230
2231 /* Called from eBPF program */
2232 static void *htab_percpu_map_lookup_elem(struct bpf_map *map, void *key)
2233 {
2234         struct htab_elem *l = __htab_map_lookup_elem(map, key);
2235
2236         if (l)
2237                 return this_cpu_ptr(htab_elem_get_ptr(l, map->key_size));
2238         else
2239                 return NULL;
2240 }
2241
2242 static void *htab_percpu_map_lookup_percpu_elem(struct bpf_map *map, void *key, u32 cpu)
2243 {
2244         struct htab_elem *l;
2245
2246         if (cpu >= nr_cpu_ids)
2247                 return NULL;
2248
2249         l = __htab_map_lookup_elem(map, key);
2250         if (l)
2251                 return per_cpu_ptr(htab_elem_get_ptr(l, map->key_size), cpu);
2252         else
2253                 return NULL;
2254 }
2255
2256 static void *htab_lru_percpu_map_lookup_elem(struct bpf_map *map, void *key)
2257 {
2258         struct htab_elem *l = __htab_map_lookup_elem(map, key);
2259
2260         if (l) {
2261                 bpf_lru_node_set_ref(&l->lru_node);
2262                 return this_cpu_ptr(htab_elem_get_ptr(l, map->key_size));
2263         }
2264
2265         return NULL;
2266 }
2267
2268 static void *htab_lru_percpu_map_lookup_percpu_elem(struct bpf_map *map, void *key, u32 cpu)
2269 {
2270         struct htab_elem *l;
2271
2272         if (cpu >= nr_cpu_ids)
2273                 return NULL;
2274
2275         l = __htab_map_lookup_elem(map, key);
2276         if (l) {
2277                 bpf_lru_node_set_ref(&l->lru_node);
2278                 return per_cpu_ptr(htab_elem_get_ptr(l, map->key_size), cpu);
2279         }
2280
2281         return NULL;
2282 }
2283
2284 int bpf_percpu_hash_copy(struct bpf_map *map, void *key, void *value)
2285 {
2286         struct htab_elem *l;
2287         void __percpu *pptr;
2288         int ret = -ENOENT;
2289         int cpu, off = 0;
2290         u32 size;
2291
2292         /* per_cpu areas are zero-filled and bpf programs can only
2293          * access 'value_size' of them, so copying rounded areas
2294          * will not leak any kernel data
2295          */
2296         size = round_up(map->value_size, 8);
2297         rcu_read_lock();
2298         l = __htab_map_lookup_elem(map, key);
2299         if (!l)
2300                 goto out;
2301         /* We do not mark LRU map element here in order to not mess up
2302          * eviction heuristics when user space does a map walk.
2303          */
2304         pptr = htab_elem_get_ptr(l, map->key_size);
2305         for_each_possible_cpu(cpu) {
2306                 bpf_long_memcpy(value + off,
2307                                 per_cpu_ptr(pptr, cpu), size);
2308                 off += size;
2309         }
2310         ret = 0;
2311 out:
2312         rcu_read_unlock();
2313         return ret;
2314 }
2315
2316 int bpf_percpu_hash_update(struct bpf_map *map, void *key, void *value,
2317                            u64 map_flags)
2318 {
2319         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
2320         int ret;
2321
2322         rcu_read_lock();
2323         if (htab_is_lru(htab))
2324                 ret = __htab_lru_percpu_map_update_elem(map, key, value,
2325                                                         map_flags, true);
2326         else
2327                 ret = __htab_percpu_map_update_elem(map, key, value, map_flags,
2328                                                     true);
2329         rcu_read_unlock();
2330
2331         return ret;
2332 }
2333
2334 static void htab_percpu_map_seq_show_elem(struct bpf_map *map, void *key,
2335                                           struct seq_file *m)
2336 {
2337         struct htab_elem *l;
2338         void __percpu *pptr;
2339         int cpu;
2340
2341         rcu_read_lock();
2342
2343         l = __htab_map_lookup_elem(map, key);
2344         if (!l) {
2345                 rcu_read_unlock();
2346                 return;
2347         }
2348
2349         btf_type_seq_show(map->btf, map->btf_key_type_id, key, m);
2350         seq_puts(m, ": {\n");
2351         pptr = htab_elem_get_ptr(l, map->key_size);
2352         for_each_possible_cpu(cpu) {
2353                 seq_printf(m, "\tcpu%d: ", cpu);
2354                 btf_type_seq_show(map->btf, map->btf_value_type_id,
2355                                   per_cpu_ptr(pptr, cpu), m);
2356                 seq_puts(m, "\n");
2357         }
2358         seq_puts(m, "}\n");
2359
2360         rcu_read_unlock();
2361 }
2362
2363 const struct bpf_map_ops htab_percpu_map_ops = {
2364         .map_meta_equal = bpf_map_meta_equal,
2365         .map_alloc_check = htab_map_alloc_check,
2366         .map_alloc = htab_map_alloc,
2367         .map_free = htab_map_free,
2368         .map_get_next_key = htab_map_get_next_key,
2369         .map_lookup_elem = htab_percpu_map_lookup_elem,
2370         .map_lookup_and_delete_elem = htab_percpu_map_lookup_and_delete_elem,
2371         .map_update_elem = htab_percpu_map_update_elem,
2372         .map_delete_elem = htab_map_delete_elem,
2373         .map_lookup_percpu_elem = htab_percpu_map_lookup_percpu_elem,
2374         .map_seq_show_elem = htab_percpu_map_seq_show_elem,
2375         .map_set_for_each_callback_args = map_set_for_each_callback_args,
2376         .map_for_each_callback = bpf_for_each_hash_elem,
2377         BATCH_OPS(htab_percpu),
2378         .map_btf_id = &htab_map_btf_ids[0],
2379         .iter_seq_info = &iter_seq_info,
2380 };
2381
2382 const struct bpf_map_ops htab_lru_percpu_map_ops = {
2383         .map_meta_equal = bpf_map_meta_equal,
2384         .map_alloc_check = htab_map_alloc_check,
2385         .map_alloc = htab_map_alloc,
2386         .map_free = htab_map_free,
2387         .map_get_next_key = htab_map_get_next_key,
2388         .map_lookup_elem = htab_lru_percpu_map_lookup_elem,
2389         .map_lookup_and_delete_elem = htab_lru_percpu_map_lookup_and_delete_elem,
2390         .map_update_elem = htab_lru_percpu_map_update_elem,
2391         .map_delete_elem = htab_lru_map_delete_elem,
2392         .map_lookup_percpu_elem = htab_lru_percpu_map_lookup_percpu_elem,
2393         .map_seq_show_elem = htab_percpu_map_seq_show_elem,
2394         .map_set_for_each_callback_args = map_set_for_each_callback_args,
2395         .map_for_each_callback = bpf_for_each_hash_elem,
2396         BATCH_OPS(htab_lru_percpu),
2397         .map_btf_id = &htab_map_btf_ids[0],
2398         .iter_seq_info = &iter_seq_info,
2399 };
2400
2401 static int fd_htab_map_alloc_check(union bpf_attr *attr)
2402 {
2403         if (attr->value_size != sizeof(u32))
2404                 return -EINVAL;
2405         return htab_map_alloc_check(attr);
2406 }
2407
2408 static void fd_htab_map_free(struct bpf_map *map)
2409 {
2410         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
2411         struct hlist_nulls_node *n;
2412         struct hlist_nulls_head *head;
2413         struct htab_elem *l;
2414         int i;
2415
2416         for (i = 0; i < htab->n_buckets; i++) {
2417                 head = select_bucket(htab, i);
2418
2419                 hlist_nulls_for_each_entry_safe(l, n, head, hash_node) {
2420                         void *ptr = fd_htab_map_get_ptr(map, l);
2421
2422                         map->ops->map_fd_put_ptr(ptr);
2423                 }
2424         }
2425
2426         htab_map_free(map);
2427 }
2428
2429 /* only called from syscall */
2430 int bpf_fd_htab_map_lookup_elem(struct bpf_map *map, void *key, u32 *value)
2431 {
2432         void **ptr;
2433         int ret = 0;
2434
2435         if (!map->ops->map_fd_sys_lookup_elem)
2436                 return -ENOTSUPP;
2437
2438         rcu_read_lock();
2439         ptr = htab_map_lookup_elem(map, key);
2440         if (ptr)
2441                 *value = map->ops->map_fd_sys_lookup_elem(READ_ONCE(*ptr));
2442         else
2443                 ret = -ENOENT;
2444         rcu_read_unlock();
2445
2446         return ret;
2447 }
2448
2449 /* only called from syscall */
2450 int bpf_fd_htab_map_update_elem(struct bpf_map *map, struct file *map_file,
2451                                 void *key, void *value, u64 map_flags)
2452 {
2453         void *ptr;
2454         int ret;
2455         u32 ufd = *(u32 *)value;
2456
2457         ptr = map->ops->map_fd_get_ptr(map, map_file, ufd);
2458         if (IS_ERR(ptr))
2459                 return PTR_ERR(ptr);
2460
2461         ret = htab_map_update_elem(map, key, &ptr, map_flags);
2462         if (ret)
2463                 map->ops->map_fd_put_ptr(ptr);
2464
2465         return ret;
2466 }
2467
2468 static struct bpf_map *htab_of_map_alloc(union bpf_attr *attr)
2469 {
2470         struct bpf_map *map, *inner_map_meta;
2471
2472         inner_map_meta = bpf_map_meta_alloc(attr->inner_map_fd);
2473         if (IS_ERR(inner_map_meta))
2474                 return inner_map_meta;
2475
2476         map = htab_map_alloc(attr);
2477         if (IS_ERR(map)) {
2478                 bpf_map_meta_free(inner_map_meta);
2479                 return map;
2480         }
2481
2482         map->inner_map_meta = inner_map_meta;
2483
2484         return map;
2485 }
2486
2487 static void *htab_of_map_lookup_elem(struct bpf_map *map, void *key)
2488 {
2489         struct bpf_map **inner_map  = htab_map_lookup_elem(map, key);
2490
2491         if (!inner_map)
2492                 return NULL;
2493
2494         return READ_ONCE(*inner_map);
2495 }
2496
2497 static int htab_of_map_gen_lookup(struct bpf_map *map,
2498                                   struct bpf_insn *insn_buf)
2499 {
2500         struct bpf_insn *insn = insn_buf;
2501         const int ret = BPF_REG_0;
2502
2503         BUILD_BUG_ON(!__same_type(&__htab_map_lookup_elem,
2504                      (void *(*)(struct bpf_map *map, void *key))NULL));
2505         *insn++ = BPF_EMIT_CALL(__htab_map_lookup_elem);
2506         *insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 2);
2507         *insn++ = BPF_ALU64_IMM(BPF_ADD, ret,
2508                                 offsetof(struct htab_elem, key) +
2509                                 round_up(map->key_size, 8));
2510         *insn++ = BPF_LDX_MEM(BPF_DW, ret, ret, 0);
2511
2512         return insn - insn_buf;
2513 }
2514
2515 static void htab_of_map_free(struct bpf_map *map)
2516 {
2517         bpf_map_meta_free(map->inner_map_meta);
2518         fd_htab_map_free(map);
2519 }
2520
2521 const struct bpf_map_ops htab_of_maps_map_ops = {
2522         .map_alloc_check = fd_htab_map_alloc_check,
2523         .map_alloc = htab_of_map_alloc,
2524         .map_free = htab_of_map_free,
2525         .map_get_next_key = htab_map_get_next_key,
2526         .map_lookup_elem = htab_of_map_lookup_elem,
2527         .map_delete_elem = htab_map_delete_elem,
2528         .map_fd_get_ptr = bpf_map_fd_get_ptr,
2529         .map_fd_put_ptr = bpf_map_fd_put_ptr,
2530         .map_fd_sys_lookup_elem = bpf_map_fd_sys_lookup_elem,
2531         .map_gen_lookup = htab_of_map_gen_lookup,
2532         .map_check_btf = map_check_no_btf,
2533         BATCH_OPS(htab),
2534         .map_btf_id = &htab_map_btf_ids[0],
2535 };