1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1987, 1993
5 * The Regents of the University of California.
6 * Copyright (c) 2005, 2009 Robert N. M. Watson
7 * All rights reserved.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 */
33
34 #ifndef _SYS_MALLOC_H_
35 #define _SYS_MALLOC_H_
36
37 #ifndef _STANDALONE
38 #include <sys/param.h>
39 #ifdef _KERNEL
40 #include <sys/systm.h>
41 #endif
42 #include <sys/queue.h>
43 #include <sys/_lock.h>
44 #include <sys/_mutex.h>
45 #include <machine/_limits.h>
46
47 #define MINALLOCSIZE UMA_SMALLEST_UNIT
48
49 /*
50 * Flags to memory allocation functions.
51 */
52 #define M_NOWAIT 0x0001 /* do not block */
53 #define M_WAITOK 0x0002 /* ok to block */
54 #define M_NORECLAIM 0x0080 /* do not reclaim after failure */
55 #define M_ZERO 0x0100 /* bzero the allocation */
56 #define M_NOVM 0x0200 /* don't ask VM for pages */
57 #define M_USE_RESERVE 0x0400 /* can alloc out of reserve memory */
58 #define M_NODUMP 0x0800 /* don't dump pages in this allocation */
59 #define M_FIRSTFIT 0x1000 /* only for vmem, fast fit */
60 #define M_BESTFIT 0x2000 /* only for vmem, low fragmentation */
61 #define M_EXEC 0x4000 /* allocate executable space */
62 #define M_NEXTFIT 0x8000 /* only for vmem, follow cursor */
63 #define M_NEVERFREED 0x10000 /* chunk will never get freed */
64
65 #define M_VERSION 2024073001
66
67 /*
68 * Two malloc type structures are present: malloc_type, which is used by a
69 * type owner to declare the type, and malloc_type_internal, which holds
70 * malloc-owned statistics and other ABI-sensitive fields, such as the set of
71 * malloc statistics indexed by the compile-time MAXCPU constant.
72 * Applications should avoid introducing dependence on the allocator private
73 * data layout and size.
74 *
75 * The malloc_type ks_next field is protected by malloc_mtx. Other fields in
76 * malloc_type are static after initialization so unsynchronized.
77 *
78 * Statistics in malloc_type_stats are written only when holding a critical
79 * section and running on the CPU associated with the index into the stat
80 * array, but read lock-free resulting in possible (minor) races, which the
81 * monitoring app should take into account.
82 */
83 struct malloc_type_stats {
84 uint64_t mts_memalloced; /* Bytes allocated on CPU. */
85 uint64_t mts_memfreed; /* Bytes freed on CPU. */
86 uint64_t mts_numallocs; /* Number of allocates on CPU. */
87 uint64_t mts_numfrees; /* number of frees on CPU. */
88 uint64_t mts_size; /* Bitmask of sizes allocated on CPU. */
89 uint64_t _mts_reserved1; /* Reserved field. */
90 uint64_t _mts_reserved2; /* Reserved field. */
91 uint64_t _mts_reserved3; /* Reserved field. */
92 };
93
94 _Static_assert(sizeof(struct malloc_type_stats) == 64,
95 "allocations come from pcpu_zone_64");
96
97 /*
98 * Index definitions for the mti_probes[] array.
99 */
100 #define DTMALLOC_PROBE_MALLOC 0
101 #define DTMALLOC_PROBE_FREE 1
102 #define DTMALLOC_PROBE_MAX 2
103
104 struct malloc_type_internal {
105 uint32_t mti_probes[DTMALLOC_PROBE_MAX];
106 /* DTrace probe ID array. */
107 u_char mti_zone;
108 struct malloc_type_stats *mti_stats;
109 u_long mti_spare[8];
110 };
111
112 /*
113 * Public data structure describing a malloc type.
114 */
115 struct malloc_type {
116 struct malloc_type *ks_next; /* Next in global chain. */
117 u_long ks_version; /* Detect programmer error. */
118 const char *ks_shortdesc; /* Printable type name. */
119 struct malloc_type_internal ks_mti;
120 };
121
122 /*
123 * Statistics structure headers for user space. The kern.malloc sysctl
124 * exposes a structure stream consisting of a stream header, then a series of
125 * malloc type headers and statistics structures (quantity maxcpus). For
126 * convenience, the kernel will provide the current value of maxcpus at the
127 * head of the stream.
128 */
129 #define MALLOC_TYPE_STREAM_VERSION 0x00000001
130 struct malloc_type_stream_header {
131 uint32_t mtsh_version; /* Stream format version. */
132 uint32_t mtsh_maxcpus; /* Value of MAXCPU for stream. */
133 uint32_t mtsh_count; /* Number of records. */
134 uint32_t _mtsh_pad; /* Pad/reserved field. */
135 };
136
137 #define MALLOC_MAX_NAME 32
138 struct malloc_type_header {
139 char mth_name[MALLOC_MAX_NAME];
140 };
141
142 #ifdef _KERNEL
143 #define MALLOC_DEFINE(type, shortdesc, longdesc) \
144 struct malloc_type type[1] = { \
145 { \
146 .ks_next = NULL, \
147 .ks_version = M_VERSION, \
148 .ks_shortdesc = shortdesc, \
149 } \
150 }; \
151 SYSINIT(type##_init, SI_SUB_KMEM, SI_ORDER_THIRD, malloc_init, \
152 type); \
153 SYSUNINIT(type##_uninit, SI_SUB_KMEM, SI_ORDER_ANY, \
154 malloc_uninit, type)
155
156 #define MALLOC_DECLARE(type) \
157 extern struct malloc_type type[1]
158
159 MALLOC_DECLARE(M_CACHE);
160 MALLOC_DECLARE(M_DEVBUF);
161 MALLOC_DECLARE(M_PARGS);
162 MALLOC_DECLARE(M_SESSION);
163 MALLOC_DECLARE(M_SUBPROC);
164 MALLOC_DECLARE(M_TEMP);
165
166 /*
167 * XXX this should be declared in <sys/uio.h>, but that tends to fail
168 * because <sys/uio.h> is included in a header before the source file
169 * has a chance to include <sys/malloc.h> to get MALLOC_DECLARE() defined.
170 */
171 MALLOC_DECLARE(M_IOV);
172
173 struct domainset;
174 extern struct mtx malloc_mtx;
175
176 /*
177 * Function type used when iterating over the list of malloc types.
178 */
179 typedef void malloc_type_list_func_t(struct malloc_type *, void *);
180
181 /* contigfree(9) is deprecated. */
182 void contigfree(void *addr, unsigned long, struct malloc_type *type);
183 void *contigmalloc(unsigned long size, struct malloc_type *type, int flags,
184 vm_paddr_t low, vm_paddr_t high, unsigned long alignment,
185 vm_paddr_t boundary) __malloc_like __result_use_check
186 __alloc_size(1) __alloc_align(6);
187 void *contigmalloc_domainset(unsigned long size, struct malloc_type *type,
188 struct domainset *ds, int flags, vm_paddr_t low, vm_paddr_t high,
189 unsigned long alignment, vm_paddr_t boundary)
190 __malloc_like __result_use_check __alloc_size(1) __alloc_align(7);
191 void free(void *addr, struct malloc_type *type);
192 void zfree(void *addr, struct malloc_type *type);
193 void *malloc(size_t size, struct malloc_type *type, int flags) __malloc_like
194 __result_use_check __alloc_size(1);
195 /*
196 * Try to optimize malloc(..., ..., M_ZERO) allocations by doing zeroing in
197 * place if the size is known at compilation time.
198 *
199 * Passing the flag down requires malloc to blindly zero the entire object.
200 * In practice a lot of the zeroing can be avoided if most of the object
201 * gets explicitly initialized after the allocation. Letting the compiler
202 * zero in place gives it the opportunity to take advantage of this state.
203 *
204 * Note that the operation is only applicable if both flags and size are
205 * known at compilation time. If M_ZERO is passed but M_WAITOK is not, the
206 * allocation can fail and a NULL check is needed. However, if M_WAITOK is
207 * passed we know the allocation must succeed and the check can be elided.
208 *
209 * _malloc_item = malloc(_size, type, (flags) &~ M_ZERO);
210 * if (((flags) & M_WAITOK) != 0 || _malloc_item != NULL)
211 * bzero(_malloc_item, _size);
212 *
213 * If the flag is set, the compiler knows the left side is always true,
214 * therefore the entire statement is true and the callsite is:
215 *
216 * _malloc_item = malloc(_size, type, (flags) &~ M_ZERO);
217 * bzero(_malloc_item, _size);
218 *
219 * If the flag is not set, the compiler knows the left size is always false
220 * and the NULL check is needed, therefore the callsite is:
221 *
222 * _malloc_item = malloc(_size, type, (flags) &~ M_ZERO);
223 * if (_malloc_item != NULL)
224 * bzero(_malloc_item, _size);
225 *
226 * The implementation is a macro because of what appears to be a clang 6 bug:
227 * an inline function variant ended up being compiled to a mere malloc call
228 * regardless of argument. gcc generates expected code (like the above).
229 */
230 #define malloc(size, type, flags) ({ \
231 void *_malloc_item; \
232 size_t _size = (size); \
233 if (__builtin_constant_p(size) && __builtin_constant_p(flags) &&\
234 ((flags) & M_ZERO) != 0) { \
235 _malloc_item = malloc(_size, type, (flags) &~ M_ZERO); \
236 if (((flags) & M_WAITOK) != 0 || \
237 __predict_true(_malloc_item != NULL)) \
238 memset(_malloc_item, 0, _size); \
239 } else { \
240 _malloc_item = malloc(_size, type, flags); \
241 } \
242 _malloc_item; \
243 })
244
245 void *malloc_domainset(size_t size, struct malloc_type *type,
246 struct domainset *ds, int flags) __malloc_like __result_use_check
247 __alloc_size(1);
248 void *mallocarray(size_t nmemb, size_t size, struct malloc_type *type,
249 int flags) __malloc_like __result_use_check
250 __alloc_size2(1, 2);
251 void *mallocarray_domainset(size_t nmemb, size_t size, struct malloc_type *type,
252 struct domainset *ds, int flags) __malloc_like __result_use_check
253 __alloc_size2(1, 2);
254 void *malloc_exec(size_t size, struct malloc_type *type, int flags) __malloc_like
255 __result_use_check __alloc_size(1);
256 void *malloc_domainset_exec(size_t size, struct malloc_type *type,
257 struct domainset *ds, int flags) __malloc_like __result_use_check
258 __alloc_size(1);
259 void malloc_init(void *);
260 void malloc_type_allocated(struct malloc_type *type, unsigned long size);
261 void malloc_type_freed(struct malloc_type *type, unsigned long size);
262 void malloc_type_list(malloc_type_list_func_t *, void *);
263 void malloc_uninit(void *);
264 size_t malloc_size(size_t);
265 size_t malloc_usable_size(const void *);
266 void *realloc(void *addr, size_t size, struct malloc_type *type, int flags)
267 __result_use_check __alloc_size(2);
268 void *reallocf(void *addr, size_t size, struct malloc_type *type, int flags)
269 __result_use_check __alloc_size(2);
270 void *malloc_aligned(size_t size, size_t align, struct malloc_type *type,
271 int flags) __malloc_like __result_use_check __alloc_size(1);
272 void *malloc_domainset_aligned(size_t size, size_t align,
273 struct malloc_type *mtp, struct domainset *ds, int flags)
274 __malloc_like __result_use_check __alloc_size(1);
275
276 struct malloc_type *malloc_desc2type(const char *desc);
277
278 /*
279 * This is sqrt(SIZE_MAX+1), as s1*s2 <= SIZE_MAX
280 * if both s1 < MUL_NO_OVERFLOW and s2 < MUL_NO_OVERFLOW
281 */
282 #define MUL_NO_OVERFLOW (1UL << (sizeof(size_t) * 8 / 2))
283 static inline bool
WOULD_OVERFLOW(size_t nmemb,size_t size)284 WOULD_OVERFLOW(size_t nmemb, size_t size)
285 {
286
287 return ((nmemb >= MUL_NO_OVERFLOW || size >= MUL_NO_OVERFLOW) &&
288 nmemb > 0 && __SIZE_T_MAX / nmemb < size);
289 }
290 #undef MUL_NO_OVERFLOW
291 #endif /* _KERNEL */
292
293 #else
294 /*
295 * The native stand malloc / free interface we're mapping to
296 */
297 extern void Free(void *p, const char *file, int line);
298 extern void *Malloc(size_t bytes, const char *file, int line);
299
300 /*
301 * Minimal standalone malloc implementation / environment. None of the
302 * flags mean anything and there's no need declare malloc types.
303 * Define the simple alloc / free routines in terms of Malloc and
304 * Free. None of the kernel features that this stuff disables are needed.
305 */
306 #define M_WAITOK 1
307 #define M_ZERO 0
308 #define M_NOWAIT 2
309 #define MALLOC_DECLARE(x)
310
311 #define kmem_zalloc(size, flags) ({ \
312 void *p = Malloc((size), __FILE__, __LINE__); \
313 if (p == NULL && (flags & M_WAITOK) != 0) \
314 panic("Could not malloc %zd bytes with M_WAITOK from %s line %d", \
315 (size_t)size, __FILE__, __LINE__); \
316 p; \
317 })
318
319 #define kmem_free(p, size) Free(p, __FILE__, __LINE__)
320
321 /*
322 * ZFS mem.h define that's the OpenZFS porting layer way of saying
323 * M_WAITOK. Given the above, it will also be a nop.
324 */
325 #define KM_SLEEP M_WAITOK
326 #define KM_NOSLEEP M_NOWAIT
327 #endif /* _STANDALONE */
328 #endif /* !_SYS_MALLOC_H_ */
329