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
64 #define M_VERSION 2020110501
65
66 /*
67 * Two malloc type structures are present: malloc_type, which is used by a
68 * type owner to declare the type, and malloc_type_internal, which holds
69 * malloc-owned statistics and other ABI-sensitive fields, such as the set of
70 * malloc statistics indexed by the compile-time MAXCPU constant.
71 * Applications should avoid introducing dependence on the allocator private
72 * data layout and size.
73 *
74 * The malloc_type ks_next field is protected by malloc_mtx. Other fields in
75 * malloc_type are static after initialization so unsynchronized.
76 *
77 * Statistics in malloc_type_stats are written only when holding a critical
78 * section and running on the CPU associated with the index into the stat
79 * array, but read lock-free resulting in possible (minor) races, which the
80 * monitoring app should take into account.
81 */
82 struct malloc_type_stats {
83 uint64_t mts_memalloced; /* Bytes allocated on CPU. */
84 uint64_t mts_memfreed; /* Bytes freed on CPU. */
85 uint64_t mts_numallocs; /* Number of allocates on CPU. */
86 uint64_t mts_numfrees; /* number of frees on CPU. */
87 uint64_t mts_size; /* Bitmask of sizes allocated on CPU. */
88 uint64_t _mts_reserved1; /* Reserved field. */
89 uint64_t _mts_reserved2; /* Reserved field. */
90 uint64_t _mts_reserved3; /* Reserved field. */
91 };
92
93 _Static_assert(sizeof(struct malloc_type_stats) == 64,
94 "allocations come from pcpu_zone_64");
95
96 /*
97 * Index definitions for the mti_probes[] array.
98 */
99 #define DTMALLOC_PROBE_MALLOC 0
100 #define DTMALLOC_PROBE_FREE 1
101 #define DTMALLOC_PROBE_MAX 2
102
103 struct malloc_type_internal {
104 uint32_t mti_probes[DTMALLOC_PROBE_MAX];
105 /* DTrace probe ID array. */
106 u_char mti_zone;
107 struct malloc_type_stats *mti_stats;
108 u_long mti_spare[8];
109 };
110
111 /*
112 * Public data structure describing a malloc type.
113 */
114 struct malloc_type {
115 struct malloc_type *ks_next; /* Next in global chain. */
116 u_long ks_version; /* Detect programmer error. */
117 const char *ks_shortdesc; /* Printable type name. */
118 struct malloc_type_internal ks_mti;
119 };
120
121 /*
122 * Statistics structure headers for user space. The kern.malloc sysctl
123 * exposes a structure stream consisting of a stream header, then a series of
124 * malloc type headers and statistics structures (quantity maxcpus). For
125 * convenience, the kernel will provide the current value of maxcpus at the
126 * head of the stream.
127 */
128 #define MALLOC_TYPE_STREAM_VERSION 0x00000001
129 struct malloc_type_stream_header {
130 uint32_t mtsh_version; /* Stream format version. */
131 uint32_t mtsh_maxcpus; /* Value of MAXCPU for stream. */
132 uint32_t mtsh_count; /* Number of records. */
133 uint32_t _mtsh_pad; /* Pad/reserved field. */
134 };
135
136 #define MALLOC_MAX_NAME 32
137 struct malloc_type_header {
138 char mth_name[MALLOC_MAX_NAME];
139 };
140
141 #ifdef _KERNEL
142 #define MALLOC_DEFINE(type, shortdesc, longdesc) \
143 struct malloc_type type[1] = { \
144 { \
145 .ks_next = NULL, \
146 .ks_version = M_VERSION, \
147 .ks_shortdesc = shortdesc, \
148 } \
149 }; \
150 SYSINIT(type##_init, SI_SUB_KMEM, SI_ORDER_THIRD, malloc_init, \
151 type); \
152 SYSUNINIT(type##_uninit, SI_SUB_KMEM, SI_ORDER_ANY, \
153 malloc_uninit, type)
154
155 #define MALLOC_DECLARE(type) \
156 extern struct malloc_type type[1]
157
158 MALLOC_DECLARE(M_CACHE);
159 MALLOC_DECLARE(M_DEVBUF);
160 MALLOC_DECLARE(M_PARGS);
161 MALLOC_DECLARE(M_SESSION);
162 MALLOC_DECLARE(M_SUBPROC);
163 MALLOC_DECLARE(M_TEMP);
164
165 /*
166 * XXX this should be declared in <sys/uio.h>, but that tends to fail
167 * because <sys/uio.h> is included in a header before the source file
168 * has a chance to include <sys/malloc.h> to get MALLOC_DECLARE() defined.
169 */
170 MALLOC_DECLARE(M_IOV);
171
172 struct domainset;
173 extern struct mtx malloc_mtx;
174
175 /*
176 * Function type used when iterating over the list of malloc types.
177 */
178 typedef void malloc_type_list_func_t(struct malloc_type *, void *);
179
180 void contigfree(void *addr, unsigned long size, struct malloc_type *type);
181 void *contigmalloc(unsigned long size, struct malloc_type *type, int flags,
182 vm_paddr_t low, vm_paddr_t high, unsigned long alignment,
183 vm_paddr_t boundary) __malloc_like __result_use_check
184 __alloc_size(1) __alloc_align(6);
185 void *contigmalloc_domainset(unsigned long size, struct malloc_type *type,
186 struct domainset *ds, int flags, vm_paddr_t low, vm_paddr_t high,
187 unsigned long alignment, vm_paddr_t boundary)
188 __malloc_like __result_use_check __alloc_size(1) __alloc_align(7);
189 void free(void *addr, struct malloc_type *type);
190 void zfree(void *addr, struct malloc_type *type);
191 void *malloc(size_t size, struct malloc_type *type, int flags) __malloc_like
192 __result_use_check __alloc_size(1);
193 /*
194 * Try to optimize malloc(..., ..., M_ZERO) allocations by doing zeroing in
195 * place if the size is known at compilation time.
196 *
197 * Passing the flag down requires malloc to blindly zero the entire object.
198 * In practice a lot of the zeroing can be avoided if most of the object
199 * gets explicitly initialized after the allocation. Letting the compiler
200 * zero in place gives it the opportunity to take advantage of this state.
201 *
202 * Note that the operation is only applicable if both flags and size are
203 * known at compilation time. If M_ZERO is passed but M_WAITOK is not, the
204 * allocation can fail and a NULL check is needed. However, if M_WAITOK is
205 * passed we know the allocation must succeed and the check can be elided.
206 *
207 * _malloc_item = malloc(_size, type, (flags) &~ M_ZERO);
208 * if (((flags) & M_WAITOK) != 0 || _malloc_item != NULL)
209 * bzero(_malloc_item, _size);
210 *
211 * If the flag is set, the compiler knows the left side is always true,
212 * therefore the entire statement is true and the callsite is:
213 *
214 * _malloc_item = malloc(_size, type, (flags) &~ M_ZERO);
215 * bzero(_malloc_item, _size);
216 *
217 * If the flag is not set, the compiler knows the left size is always false
218 * and the NULL check is needed, therefore the callsite is:
219 *
220 * _malloc_item = malloc(_size, type, (flags) &~ M_ZERO);
221 * if (_malloc_item != NULL)
222 * bzero(_malloc_item, _size);
223 *
224 * The implementation is a macro because of what appears to be a clang 6 bug:
225 * an inline function variant ended up being compiled to a mere malloc call
226 * regardless of argument. gcc generates expected code (like the above).
227 */
228 #define malloc(size, type, flags) ({ \
229 void *_malloc_item; \
230 size_t _size = (size); \
231 if (__builtin_constant_p(size) && __builtin_constant_p(flags) &&\
232 ((flags) & M_ZERO) != 0) { \
233 _malloc_item = malloc(_size, type, (flags) &~ M_ZERO); \
234 if (((flags) & M_WAITOK) != 0 || \
235 __predict_true(_malloc_item != NULL)) \
236 memset(_malloc_item, 0, _size); \
237 } else { \
238 _malloc_item = malloc(_size, type, flags); \
239 } \
240 _malloc_item; \
241 })
242
243 void *malloc_domainset(size_t size, struct malloc_type *type,
244 struct domainset *ds, int flags) __malloc_like __result_use_check
245 __alloc_size(1);
246 void *mallocarray(size_t nmemb, size_t size, struct malloc_type *type,
247 int flags) __malloc_like __result_use_check
248 __alloc_size2(1, 2);
249 void *mallocarray_domainset(size_t nmemb, size_t size, struct malloc_type *type,
250 struct domainset *ds, int flags) __malloc_like __result_use_check
251 __alloc_size2(1, 2);
252 void *malloc_exec(size_t size, struct malloc_type *type, int flags) __malloc_like
253 __result_use_check __alloc_size(1);
254 void *malloc_domainset_exec(size_t size, struct malloc_type *type,
255 struct domainset *ds, int flags) __malloc_like __result_use_check
256 __alloc_size(1);
257 void malloc_init(void *);
258 void malloc_type_allocated(struct malloc_type *type, unsigned long size);
259 void malloc_type_freed(struct malloc_type *type, unsigned long size);
260 void malloc_type_list(malloc_type_list_func_t *, void *);
261 void malloc_uninit(void *);
262 size_t malloc_size(size_t);
263 size_t malloc_usable_size(const void *);
264 void *realloc(void *addr, size_t size, struct malloc_type *type, int flags)
265 __result_use_check __alloc_size(2);
266 void *reallocf(void *addr, size_t size, struct malloc_type *type, int flags)
267 __result_use_check __alloc_size(2);
268 void *malloc_aligned(size_t size, size_t align, struct malloc_type *type,
269 int flags) __malloc_like __result_use_check __alloc_size(1);
270 void *malloc_domainset_aligned(size_t size, size_t align,
271 struct malloc_type *mtp, struct domainset *ds, int flags)
272 __malloc_like __result_use_check __alloc_size(1);
273
274 struct malloc_type *malloc_desc2type(const char *desc);
275
276 /*
277 * This is sqrt(SIZE_MAX+1), as s1*s2 <= SIZE_MAX
278 * if both s1 < MUL_NO_OVERFLOW and s2 < MUL_NO_OVERFLOW
279 */
280 #define MUL_NO_OVERFLOW (1UL << (sizeof(size_t) * 8 / 2))
281 static inline bool
WOULD_OVERFLOW(size_t nmemb,size_t size)282 WOULD_OVERFLOW(size_t nmemb, size_t size)
283 {
284
285 return ((nmemb >= MUL_NO_OVERFLOW || size >= MUL_NO_OVERFLOW) &&
286 nmemb > 0 && __SIZE_T_MAX / nmemb < size);
287 }
288 #undef MUL_NO_OVERFLOW
289 #endif /* _KERNEL */
290
291 #else
292 /*
293 * The native stand malloc / free interface we're mapping to
294 */
295 extern void Free(void *p, const char *file, int line);
296 extern void *Malloc(size_t bytes, const char *file, int line);
297
298 /*
299 * Minimal standalone malloc implementation / environment. None of the
300 * flags mean anything and there's no need declare malloc types.
301 * Define the simple alloc / free routines in terms of Malloc and
302 * Free. None of the kernel features that this stuff disables are needed.
303 */
304 #define M_WAITOK 1
305 #define M_ZERO 0
306 #define M_NOWAIT 2
307 #define MALLOC_DECLARE(x)
308
309 #define kmem_zalloc(size, flags) ({ \
310 void *p = Malloc((size), __FILE__, __LINE__); \
311 if (p == NULL && (flags & M_WAITOK) != 0) \
312 panic("Could not malloc %zd bytes with M_WAITOK from %s line %d", \
313 (size_t)size, __FILE__, __LINE__); \
314 p; \
315 })
316
317 #define kmem_free(p, size) Free(p, __FILE__, __LINE__)
318
319 /*
320 * ZFS mem.h define that's the OpenZFS porting layer way of saying
321 * M_WAITOK. Given the above, it will also be a nop.
322 */
323 #define KM_SLEEP M_WAITOK
324 #define KM_NOSLEEP M_NOWAIT
325 #endif /* _STANDALONE */
326 #endif /* !_SYS_MALLOC_H_ */
327