Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 1 | /* |
Willy Tarreau | 62405a2 | 2014-12-23 13:51:28 +0100 | [diff] [blame] | 2 | * include/common/memory.h |
| 3 | * Memory management definitions.. |
| 4 | * |
| 5 | * Copyright (C) 2000-2014 Willy Tarreau - w@1wt.eu |
| 6 | * |
| 7 | * This library is free software; you can redistribute it and/or |
| 8 | * modify it under the terms of the GNU Lesser General Public |
| 9 | * License as published by the Free Software Foundation, version 2.1 |
| 10 | * exclusively. |
| 11 | * |
| 12 | * This library is distributed in the hope that it will be useful, |
| 13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 15 | * Lesser General Public License for more details. |
| 16 | * |
| 17 | * You should have received a copy of the GNU Lesser General Public |
| 18 | * License along with this library; if not, write to the Free Software |
| 19 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 20 | */ |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 21 | |
Willy Tarreau | 2dd0d47 | 2006-06-29 17:53:05 +0200 | [diff] [blame] | 22 | #ifndef _COMMON_MEMORY_H |
| 23 | #define _COMMON_MEMORY_H |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 24 | |
Willy Tarreau | 158fa75 | 2017-11-22 15:47:29 +0100 | [diff] [blame] | 25 | #include <sys/mman.h> |
| 26 | |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 27 | #include <stdlib.h> |
Willy Tarreau | e430e77 | 2014-12-23 14:13:16 +0100 | [diff] [blame] | 28 | #include <string.h> |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 29 | |
Willy Tarreau | 2dd0d47 | 2006-06-29 17:53:05 +0200 | [diff] [blame] | 30 | #include <common/config.h> |
Willy Tarreau | 50e608d | 2007-05-13 18:26:08 +0200 | [diff] [blame] | 31 | #include <common/mini-clist.h> |
Christopher Faulet | b349e48 | 2017-08-29 09:52:38 +0200 | [diff] [blame] | 32 | #include <common/hathreads.h> |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 33 | |
Willy Tarreau | a84dcb8 | 2015-10-28 12:04:02 +0100 | [diff] [blame] | 34 | #ifndef DEBUG_DONT_SHARE_POOLS |
Willy Tarreau | 50e608d | 2007-05-13 18:26:08 +0200 | [diff] [blame] | 35 | #define MEM_F_SHARED 0x1 |
Willy Tarreau | a84dcb8 | 2015-10-28 12:04:02 +0100 | [diff] [blame] | 36 | #else |
| 37 | #define MEM_F_SHARED 0 |
| 38 | #endif |
Willy Tarreau | 581bf81 | 2016-01-25 02:19:13 +0100 | [diff] [blame] | 39 | #define MEM_F_EXACT 0x2 |
Willy Tarreau | 50e608d | 2007-05-13 18:26:08 +0200 | [diff] [blame] | 40 | |
Willy Tarreau | ac42111 | 2015-10-28 15:09:29 +0100 | [diff] [blame] | 41 | /* reserve an extra void* at the end of a pool for linking */ |
| 42 | #ifdef DEBUG_MEMORY_POOLS |
| 43 | #define POOL_EXTRA (sizeof(void *)) |
| 44 | #define POOL_LINK(pool, item) (void **)(((char *)item) + (pool->size)) |
| 45 | #else |
| 46 | #define POOL_EXTRA (0) |
| 47 | #define POOL_LINK(pool, item) ((void **)(item)) |
| 48 | #endif |
| 49 | |
Willy Tarreau | 50e608d | 2007-05-13 18:26:08 +0200 | [diff] [blame] | 50 | struct pool_head { |
| 51 | void **free_list; |
Christopher Faulet | 9dcf9b6 | 2017-11-13 10:34:01 +0100 | [diff] [blame] | 52 | __decl_hathreads(HA_SPINLOCK_T lock); /* the spin lock */ |
Willy Tarreau | 50e608d | 2007-05-13 18:26:08 +0200 | [diff] [blame] | 53 | struct list list; /* list of all known pools */ |
| 54 | unsigned int used; /* how many chunks are currently in use */ |
| 55 | unsigned int allocated; /* how many chunks have been allocated */ |
| 56 | unsigned int limit; /* hard limit on the number of chunks */ |
| 57 | unsigned int minavail; /* how many chunks are expected to be used */ |
| 58 | unsigned int size; /* chunk size */ |
| 59 | unsigned int flags; /* MEM_F_* */ |
Willy Tarreau | 7dcd46d | 2007-05-14 00:16:13 +0200 | [diff] [blame] | 60 | unsigned int users; /* number of pools sharing this zone */ |
Willy Tarreau | 58102cf | 2015-10-28 16:24:21 +0100 | [diff] [blame] | 61 | unsigned int failed; /* failed allocations */ |
Willy Tarreau | 7dcd46d | 2007-05-14 00:16:13 +0200 | [diff] [blame] | 62 | char name[12]; /* name of the pool */ |
Willy Tarreau | 50e608d | 2007-05-13 18:26:08 +0200 | [diff] [blame] | 63 | }; |
| 64 | |
Willy Tarreau | 067ac9f | 2015-10-08 14:12:13 +0200 | [diff] [blame] | 65 | /* poison each newly allocated area with this byte if >= 0 */ |
| 66 | extern int mem_poison_byte; |
Willy Tarreau | 50e608d | 2007-05-13 18:26:08 +0200 | [diff] [blame] | 67 | |
Willy Tarreau | a885f6d | 2014-12-03 15:25:28 +0100 | [diff] [blame] | 68 | /* Allocates new entries for pool <pool> until there are at least <avail> + 1 |
| 69 | * available, then returns the last one for immediate use, so that at least |
| 70 | * <avail> are left available in the pool upon return. NULL is returned if the |
| 71 | * last entry could not be allocated. It's important to note that at least one |
| 72 | * allocation is always performed even if there are enough entries in the pool. |
| 73 | * A call to the garbage collector is performed at most once in case malloc() |
| 74 | * returns an error, before returning NULL. |
Willy Tarreau | 50e608d | 2007-05-13 18:26:08 +0200 | [diff] [blame] | 75 | */ |
Christopher Faulet | b349e48 | 2017-08-29 09:52:38 +0200 | [diff] [blame] | 76 | void *__pool_refill_alloc(struct pool_head *pool, unsigned int avail); |
Willy Tarreau | a885f6d | 2014-12-03 15:25:28 +0100 | [diff] [blame] | 77 | void *pool_refill_alloc(struct pool_head *pool, unsigned int avail); |
Willy Tarreau | 50e608d | 2007-05-13 18:26:08 +0200 | [diff] [blame] | 78 | |
| 79 | /* Try to find an existing shared pool with the same characteristics and |
| 80 | * returns it, otherwise creates this one. NULL is returned if no memory |
| 81 | * is available for a new creation. |
| 82 | */ |
| 83 | struct pool_head *create_pool(char *name, unsigned int size, unsigned int flags); |
| 84 | |
| 85 | /* Dump statistics on pools usage. |
| 86 | */ |
Willy Tarreau | 12833bb | 2014-01-28 16:49:56 +0100 | [diff] [blame] | 87 | void dump_pools_to_trash(); |
Willy Tarreau | 50e608d | 2007-05-13 18:26:08 +0200 | [diff] [blame] | 88 | void dump_pools(void); |
Willy Tarreau | 58102cf | 2015-10-28 16:24:21 +0100 | [diff] [blame] | 89 | int pool_total_failures(); |
| 90 | unsigned long pool_total_allocated(); |
| 91 | unsigned long pool_total_used(); |
Willy Tarreau | 50e608d | 2007-05-13 18:26:08 +0200 | [diff] [blame] | 92 | |
| 93 | /* |
Willy Tarreau | e6ce59d | 2007-05-13 19:38:49 +0200 | [diff] [blame] | 94 | * This function frees whatever can be freed in pool <pool>. |
| 95 | */ |
| 96 | void pool_flush2(struct pool_head *pool); |
| 97 | |
| 98 | /* |
| 99 | * This function frees whatever can be freed in all pools, but respecting |
| 100 | * the minimum thresholds imposed by owners. |
Christopher Faulet | b349e48 | 2017-08-29 09:52:38 +0200 | [diff] [blame] | 101 | * |
| 102 | * <pool_ctx> is used when pool_gc2 is called to release resources to allocate |
| 103 | * an element in __pool_refill_alloc. It is important because <pool_ctx> is |
| 104 | * already locked, so we need to skip the lock here. |
Willy Tarreau | e6ce59d | 2007-05-13 19:38:49 +0200 | [diff] [blame] | 105 | */ |
Christopher Faulet | b349e48 | 2017-08-29 09:52:38 +0200 | [diff] [blame] | 106 | void pool_gc2(struct pool_head *pool_ctx); |
Willy Tarreau | e6ce59d | 2007-05-13 19:38:49 +0200 | [diff] [blame] | 107 | |
| 108 | /* |
| 109 | * This function destroys a pull by freeing it completely. |
| 110 | * This should be called only under extreme circumstances. |
| 111 | */ |
Willy Tarreau | 4d2d098 | 2007-05-14 00:39:29 +0200 | [diff] [blame] | 112 | void *pool_destroy2(struct pool_head *pool); |
Willy Tarreau | e6ce59d | 2007-05-13 19:38:49 +0200 | [diff] [blame] | 113 | |
| 114 | /* |
Willy Tarreau | 0262241 | 2014-12-08 16:35:23 +0100 | [diff] [blame] | 115 | * Returns a pointer to type <type> taken from the pool <pool_type> if |
| 116 | * available, otherwise returns NULL. No malloc() is attempted, and poisonning |
| 117 | * is never performed. The purpose is to get the fastest possible allocation. |
Willy Tarreau | 50e608d | 2007-05-13 18:26:08 +0200 | [diff] [blame] | 118 | */ |
Christopher Faulet | b349e48 | 2017-08-29 09:52:38 +0200 | [diff] [blame] | 119 | static inline void *__pool_get_first(struct pool_head *pool) |
Willy Tarreau | e430e77 | 2014-12-23 14:13:16 +0100 | [diff] [blame] | 120 | { |
| 121 | void *p; |
| 122 | |
Willy Tarreau | 0262241 | 2014-12-08 16:35:23 +0100 | [diff] [blame] | 123 | if ((p = pool->free_list) != NULL) { |
Willy Tarreau | ac42111 | 2015-10-28 15:09:29 +0100 | [diff] [blame] | 124 | pool->free_list = *POOL_LINK(pool, p); |
Willy Tarreau | e430e77 | 2014-12-23 14:13:16 +0100 | [diff] [blame] | 125 | pool->used++; |
Willy Tarreau | de30a68 | 2015-10-28 15:23:51 +0100 | [diff] [blame] | 126 | #ifdef DEBUG_MEMORY_POOLS |
| 127 | /* keep track of where the element was allocated from */ |
| 128 | *POOL_LINK(pool, p) = (void *)pool; |
| 129 | #endif |
Willy Tarreau | e430e77 | 2014-12-23 14:13:16 +0100 | [diff] [blame] | 130 | } |
| 131 | return p; |
| 132 | } |
Willy Tarreau | 50e608d | 2007-05-13 18:26:08 +0200 | [diff] [blame] | 133 | |
Christopher Faulet | b349e48 | 2017-08-29 09:52:38 +0200 | [diff] [blame] | 134 | static inline void *pool_get_first(struct pool_head *pool) |
| 135 | { |
| 136 | void *ret; |
| 137 | |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 138 | HA_SPIN_LOCK(POOL_LOCK, &pool->lock); |
Christopher Faulet | b349e48 | 2017-08-29 09:52:38 +0200 | [diff] [blame] | 139 | ret = __pool_get_first(pool); |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 140 | HA_SPIN_UNLOCK(POOL_LOCK, &pool->lock); |
Christopher Faulet | b349e48 | 2017-08-29 09:52:38 +0200 | [diff] [blame] | 141 | return ret; |
| 142 | } |
Willy Tarreau | 50e608d | 2007-05-13 18:26:08 +0200 | [diff] [blame] | 143 | /* |
Willy Tarreau | 0262241 | 2014-12-08 16:35:23 +0100 | [diff] [blame] | 144 | * Returns a pointer to type <type> taken from the pool <pool_type> or |
| 145 | * dynamically allocated. In the first case, <pool_type> is updated to point to |
| 146 | * the next element in the list. No memory poisonning is ever performed on the |
| 147 | * returned area. |
| 148 | */ |
| 149 | static inline void *pool_alloc_dirty(struct pool_head *pool) |
| 150 | { |
| 151 | void *p; |
| 152 | |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 153 | HA_SPIN_LOCK(POOL_LOCK, &pool->lock); |
Christopher Faulet | b349e48 | 2017-08-29 09:52:38 +0200 | [diff] [blame] | 154 | if ((p = __pool_get_first(pool)) == NULL) |
| 155 | p = __pool_refill_alloc(pool, 0); |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 156 | HA_SPIN_UNLOCK(POOL_LOCK, &pool->lock); |
Willy Tarreau | 0262241 | 2014-12-08 16:35:23 +0100 | [diff] [blame] | 157 | return p; |
| 158 | } |
| 159 | |
Willy Tarreau | 158fa75 | 2017-11-22 15:47:29 +0100 | [diff] [blame] | 160 | #ifndef DEBUG_UAF /* normal allocator */ |
| 161 | |
Willy Tarreau | f13322e | 2017-11-22 10:50:54 +0100 | [diff] [blame] | 162 | /* allocates an area of size <size> and returns it. The semantics are similar |
| 163 | * to those of malloc(). |
| 164 | */ |
| 165 | static inline void *pool_alloc_area(size_t size) |
| 166 | { |
| 167 | return malloc(size); |
| 168 | } |
| 169 | |
| 170 | /* frees an area <area> of size <size> allocated by pool_alloc_area(). The |
| 171 | * semantics are identical to free() except that the size is specified and |
| 172 | * may be ignored. |
| 173 | */ |
| 174 | static inline void pool_free_area(void *area, size_t __maybe_unused size) |
| 175 | { |
| 176 | free(area); |
| 177 | } |
| 178 | |
Willy Tarreau | 158fa75 | 2017-11-22 15:47:29 +0100 | [diff] [blame] | 179 | #else /* use-after-free detector */ |
| 180 | |
| 181 | /* allocates an area of size <size> and returns it. The semantics are similar |
| 182 | * to those of malloc(). However the allocation is rounded up to 4kB so that a |
| 183 | * full page is allocated. This ensures the object can be freed alone so that |
| 184 | * future dereferences are easily detected. The returned object is always |
| 185 | * 16-bytes aligned to avoid issues with unaligned structure objects. |
| 186 | */ |
| 187 | static inline void *pool_alloc_area(size_t size) |
| 188 | { |
| 189 | size_t pad = (4096 - size) & 0xFF0; |
| 190 | |
| 191 | return mmap(NULL, (size + 4095) & -4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, 0, 0) + pad; |
| 192 | } |
| 193 | |
| 194 | /* frees an area <area> of size <size> allocated by pool_alloc_area(). The |
| 195 | * semantics are identical to free() except that the size must absolutely match |
| 196 | * the one passed to pool_alloc_area(). |
| 197 | */ |
| 198 | static inline void pool_free_area(void *area, size_t size) |
| 199 | { |
| 200 | size_t pad = (4096 - size) & 0xFF0; |
| 201 | |
| 202 | munmap(area - pad, (size + 4095) & -4096); |
| 203 | } |
| 204 | |
| 205 | #endif /* DEBUG_UAF */ |
| 206 | |
Willy Tarreau | 0262241 | 2014-12-08 16:35:23 +0100 | [diff] [blame] | 207 | /* |
| 208 | * Returns a pointer to type <type> taken from the pool <pool_type> or |
| 209 | * dynamically allocated. In the first case, <pool_type> is updated to point to |
| 210 | * the next element in the list. Memory poisonning is performed if enabled. |
| 211 | */ |
| 212 | static inline void *pool_alloc2(struct pool_head *pool) |
| 213 | { |
| 214 | void *p; |
| 215 | |
| 216 | p = pool_alloc_dirty(pool); |
Willy Tarreau | de30a68 | 2015-10-28 15:23:51 +0100 | [diff] [blame] | 217 | #ifdef DEBUG_MEMORY_POOLS |
| 218 | if (p) { |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 219 | HA_SPIN_LOCK(POOL_LOCK, &pool->lock); |
Willy Tarreau | de30a68 | 2015-10-28 15:23:51 +0100 | [diff] [blame] | 220 | /* keep track of where the element was allocated from */ |
| 221 | *POOL_LINK(pool, p) = (void *)pool; |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 222 | HA_SPIN_UNLOCK(POOL_LOCK, &pool->lock); |
Willy Tarreau | de30a68 | 2015-10-28 15:23:51 +0100 | [diff] [blame] | 223 | } |
| 224 | #endif |
| 225 | if (p && mem_poison_byte >= 0) { |
Willy Tarreau | 0262241 | 2014-12-08 16:35:23 +0100 | [diff] [blame] | 226 | memset(p, mem_poison_byte, pool->size); |
Willy Tarreau | de30a68 | 2015-10-28 15:23:51 +0100 | [diff] [blame] | 227 | } |
| 228 | |
Willy Tarreau | 0262241 | 2014-12-08 16:35:23 +0100 | [diff] [blame] | 229 | return p; |
| 230 | } |
| 231 | |
| 232 | /* |
Willy Tarreau | 50e608d | 2007-05-13 18:26:08 +0200 | [diff] [blame] | 233 | * Puts a memory area back to the corresponding pool. |
| 234 | * Items are chained directly through a pointer that |
| 235 | * is written in the beginning of the memory area, so |
| 236 | * there's no need for any carrier cell. This implies |
| 237 | * that each memory area is at least as big as one |
Willy Tarreau | 48d63db | 2008-08-03 17:41:33 +0200 | [diff] [blame] | 238 | * pointer. Just like with the libc's free(), nothing |
| 239 | * is done if <ptr> is NULL. |
Willy Tarreau | 50e608d | 2007-05-13 18:26:08 +0200 | [diff] [blame] | 240 | */ |
Willy Tarreau | e430e77 | 2014-12-23 14:13:16 +0100 | [diff] [blame] | 241 | static inline void pool_free2(struct pool_head *pool, void *ptr) |
| 242 | { |
| 243 | if (likely(ptr != NULL)) { |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 244 | HA_SPIN_LOCK(POOL_LOCK, &pool->lock); |
Willy Tarreau | de30a68 | 2015-10-28 15:23:51 +0100 | [diff] [blame] | 245 | #ifdef DEBUG_MEMORY_POOLS |
| 246 | /* we'll get late corruption if we refill to the wrong pool or double-free */ |
| 247 | if (*POOL_LINK(pool, ptr) != (void *)pool) |
| 248 | *(int *)0 = 0; |
| 249 | #endif |
Willy Tarreau | 158fa75 | 2017-11-22 15:47:29 +0100 | [diff] [blame] | 250 | |
| 251 | #ifndef DEBUG_UAF /* normal pool behaviour */ |
Willy Tarreau | ac42111 | 2015-10-28 15:09:29 +0100 | [diff] [blame] | 252 | *POOL_LINK(pool, ptr) = (void *)pool->free_list; |
Willy Tarreau | e430e77 | 2014-12-23 14:13:16 +0100 | [diff] [blame] | 253 | pool->free_list = (void *)ptr; |
Willy Tarreau | 158fa75 | 2017-11-22 15:47:29 +0100 | [diff] [blame] | 254 | #else /* release the entry for real to detect use after free */ |
| 255 | /* ensure we crash on double free or free of a const area*/ |
| 256 | *(uint32_t *)ptr = 0xDEADADD4; |
| 257 | pool_free_area(ptr, pool->size + POOL_EXTRA); |
| 258 | pool->allocated--; |
| 259 | #endif /* DEBUG_UAF */ |
Willy Tarreau | e430e77 | 2014-12-23 14:13:16 +0100 | [diff] [blame] | 260 | pool->used--; |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 261 | HA_SPIN_UNLOCK(POOL_LOCK, &pool->lock); |
Willy Tarreau | e430e77 | 2014-12-23 14:13:16 +0100 | [diff] [blame] | 262 | } |
| 263 | } |
Willy Tarreau | 2dd0d47 | 2006-06-29 17:53:05 +0200 | [diff] [blame] | 264 | #endif /* _COMMON_MEMORY_H */ |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 265 | |
| 266 | /* |
| 267 | * Local variables: |
| 268 | * c-indent-level: 8 |
| 269 | * c-basic-offset: 8 |
| 270 | * End: |
| 271 | */ |