xref: /freebsd/lib/libmemstat/memstat_uma.c (revision 262e143bd46171a6415a5b28af260a5efa2a3db8)
1 /*-
2  * Copyright (c) 2005 Robert N. M. Watson
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  * $FreeBSD$
27  */
28 
29 #include <sys/param.h>
30 #include <sys/sysctl.h>
31 
32 #define	LIBMEMSTAT	/* Cause vm_page.h not to include opt_vmpage.h */
33 #include <vm/vm.h>
34 #include <vm/vm_page.h>
35 
36 #include <vm/uma.h>
37 #include <vm/uma_int.h>
38 
39 #include <err.h>
40 #include <errno.h>
41 #include <kvm.h>
42 #include <nlist.h>
43 #include <stdio.h>
44 #include <stdlib.h>
45 #include <string.h>
46 
47 #include "memstat.h"
48 #include "memstat_internal.h"
49 
50 static struct nlist namelist[] = {
51 #define	X_UMA_KEGS	0
52 	{ .n_name = "_uma_kegs" },
53 #define	X_MP_MAXID	1
54 	{ .n_name = "_mp_maxid" },
55 	{ .n_name = "" },
56 };
57 
58 /*
59  * Extract uma(9) statistics from the running kernel, and store all memory
60  * type information in the passed list.  For each type, check the list for an
61  * existing entry with the right name/allocator -- if present, update that
62  * entry.  Otherwise, add a new entry.  On error, the entire list will be
63  * cleared, as entries will be in an inconsistent state.
64  *
65  * To reduce the level of work for a list that starts empty, we keep around a
66  * hint as to whether it was empty when we began, so we can avoid searching
67  * the list for entries to update.  Updates are O(n^2) due to searching for
68  * each entry before adding it.
69  */
70 int
71 memstat_sysctl_uma(struct memory_type_list *list, int flags)
72 {
73 	struct uma_stream_header *ushp;
74 	struct uma_type_header *uthp;
75 	struct uma_percpu_stat *upsp;
76 	struct memory_type *mtp;
77 	int count, hint_dontsearch, i, j, maxcpus;
78 	char *buffer, *p;
79 	size_t size;
80 
81 	hint_dontsearch = LIST_EMPTY(&list->mtl_list);
82 
83 	/*
84 	 * Query the number of CPUs, number of malloc types so that we can
85 	 * guess an initial buffer size.  We loop until we succeed or really
86 	 * fail.  Note that the value of maxcpus we query using sysctl is not
87 	 * the version we use when processing the real data -- that is read
88 	 * from the header.
89 	 */
90 retry:
91 	size = sizeof(maxcpus);
92 	if (sysctlbyname("kern.smp.maxcpus", &maxcpus, &size, NULL, 0) < 0) {
93 		if (errno == EACCES || errno == EPERM)
94 			list->mtl_error = MEMSTAT_ERROR_PERMISSION;
95 		else
96 			list->mtl_error = MEMSTAT_ERROR_DATAERROR;
97 		return (-1);
98 	}
99 	if (size != sizeof(maxcpus)) {
100 		list->mtl_error = MEMSTAT_ERROR_DATAERROR;
101 		return (-1);
102 	}
103 
104 	if (maxcpus > MEMSTAT_MAXCPU) {
105 		list->mtl_error = MEMSTAT_ERROR_TOOMANYCPUS;
106 		return (-1);
107 	}
108 
109 	size = sizeof(count);
110 	if (sysctlbyname("vm.zone_count", &count, &size, NULL, 0) < 0) {
111 		if (errno == EACCES || errno == EPERM)
112 			list->mtl_error = MEMSTAT_ERROR_PERMISSION;
113 		else
114 			list->mtl_error = MEMSTAT_ERROR_VERSION;
115 		return (-1);
116 	}
117 	if (size != sizeof(count)) {
118 		list->mtl_error = MEMSTAT_ERROR_DATAERROR;
119 		return (-1);
120 	}
121 
122 	size = sizeof(*uthp) + count * (sizeof(*uthp) + sizeof(*upsp) *
123 	    maxcpus);
124 
125 	buffer = malloc(size);
126 	if (buffer == NULL) {
127 		list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
128 		return (-1);
129 	}
130 
131 	if (sysctlbyname("vm.zone_stats", buffer, &size, NULL, 0) < 0) {
132 		/*
133 		 * XXXRW: ENOMEM is an ambiguous return, we should bound the
134 		 * number of loops, perhaps.
135 		 */
136 		if (errno == ENOMEM) {
137 			free(buffer);
138 			goto retry;
139 		}
140 		if (errno == EACCES || errno == EPERM)
141 			list->mtl_error = MEMSTAT_ERROR_PERMISSION;
142 		else
143 			list->mtl_error = MEMSTAT_ERROR_VERSION;
144 		free(buffer);
145 		return (-1);
146 	}
147 
148 	if (size == 0) {
149 		free(buffer);
150 		return (0);
151 	}
152 
153 	if (size < sizeof(*ushp)) {
154 		list->mtl_error = MEMSTAT_ERROR_VERSION;
155 		free(buffer);
156 		return (-1);
157 	}
158 	p = buffer;
159 	ushp = (struct uma_stream_header *)p;
160 	p += sizeof(*ushp);
161 
162 	if (ushp->ush_version != UMA_STREAM_VERSION) {
163 		list->mtl_error = MEMSTAT_ERROR_VERSION;
164 		free(buffer);
165 		return (-1);
166 	}
167 
168 	if (ushp->ush_maxcpus > MEMSTAT_MAXCPU) {
169 		list->mtl_error = MEMSTAT_ERROR_TOOMANYCPUS;
170 		free(buffer);
171 		return (-1);
172 	}
173 
174 	/*
175 	 * For the remainder of this function, we are quite trusting about
176 	 * the layout of structures and sizes, since we've determined we have
177 	 * a matching version and acceptable CPU count.
178 	 */
179 	maxcpus = ushp->ush_maxcpus;
180 	count = ushp->ush_count;
181 	for (i = 0; i < count; i++) {
182 		uthp = (struct uma_type_header *)p;
183 		p += sizeof(*uthp);
184 
185 		if (hint_dontsearch == 0) {
186 			mtp = memstat_mtl_find(list, ALLOCATOR_UMA,
187 			    uthp->uth_name);
188 		} else
189 			mtp = NULL;
190 		if (mtp == NULL)
191 			mtp = _memstat_mt_allocate(list, ALLOCATOR_UMA,
192 			    uthp->uth_name);
193 		if (mtp == NULL) {
194 			_memstat_mtl_empty(list);
195 			free(buffer);
196 			list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
197 			return (-1);
198 		}
199 
200 		/*
201 		 * Reset the statistics on a current node.
202 		 */
203 		_memstat_mt_reset_stats(mtp);
204 
205 		mtp->mt_numallocs = uthp->uth_allocs;
206 		mtp->mt_numfrees = uthp->uth_frees;
207 		mtp->mt_failures = uthp->uth_fails;
208 
209 		for (j = 0; j < maxcpus; j++) {
210 			upsp = (struct uma_percpu_stat *)p;
211 			p += sizeof(*upsp);
212 
213 			mtp->mt_percpu_cache[j].mtp_free =
214 			    upsp->ups_cache_free;
215 			mtp->mt_free += upsp->ups_cache_free;
216 			mtp->mt_numallocs += upsp->ups_allocs;
217 			mtp->mt_numfrees += upsp->ups_frees;
218 		}
219 
220 		mtp->mt_size = uthp->uth_size;
221 		mtp->mt_memalloced = mtp->mt_numallocs * uthp->uth_size;
222 		mtp->mt_memfreed = mtp->mt_numfrees * uthp->uth_size;
223 		mtp->mt_bytes = mtp->mt_memalloced - mtp->mt_memfreed;
224 		mtp->mt_countlimit = uthp->uth_limit;
225 		mtp->mt_byteslimit = uthp->uth_limit * uthp->uth_size;
226 
227 		mtp->mt_count = mtp->mt_numallocs - mtp->mt_numfrees;
228 		mtp->mt_zonefree = uthp->uth_zone_free;
229 
230 		/*
231 		 * UMA secondary zones share a keg with the primary zone.  To
232 		 * avoid double-reporting of free items, report keg free
233 		 * items only in the primary zone.
234 		 */
235 		if (!(uthp->uth_zone_flags & UTH_ZONE_SECONDARY)) {
236 			mtp->mt_kegfree = uthp->uth_keg_free;
237 			mtp->mt_free += mtp->mt_kegfree;
238 		}
239 		mtp->mt_free += mtp->mt_zonefree;
240 	}
241 
242 	free(buffer);
243 
244 	return (0);
245 }
246 
247 static int
248 kread(kvm_t *kvm, void *kvm_pointer, void *address, size_t size,
249     size_t offset)
250 {
251 	ssize_t ret;
252 
253 	ret = kvm_read(kvm, (unsigned long)kvm_pointer + offset, address,
254 	    size);
255 	if (ret < 0)
256 		return (MEMSTAT_ERROR_KVM);
257 	if ((size_t)ret != size)
258 		return (MEMSTAT_ERROR_KVM_SHORTREAD);
259 	return (0);
260 }
261 
262 static int
263 kread_string(kvm_t *kvm, void *kvm_pointer, char *buffer, int buflen)
264 {
265 	ssize_t ret;
266 	int i;
267 
268 	for (i = 0; i < buflen; i++) {
269 		ret = kvm_read(kvm, (unsigned long)kvm_pointer + i,
270 		    &(buffer[i]), sizeof(char));
271 		if (ret < 0)
272 			return (MEMSTAT_ERROR_KVM);
273 		if ((size_t)ret != sizeof(char))
274 			return (MEMSTAT_ERROR_KVM_SHORTREAD);
275 		if (buffer[i] == '\0')
276 			return (0);
277 	}
278 	/* Truncate. */
279 	buffer[i-1] = '\0';
280 	return (0);
281 }
282 
283 static int
284 kread_symbol(kvm_t *kvm, int index, void *address, size_t size,
285     size_t offset)
286 {
287 	ssize_t ret;
288 
289 	ret = kvm_read(kvm, namelist[index].n_value + offset, address, size);
290 	if (ret < 0)
291 		return (MEMSTAT_ERROR_KVM);
292 	if ((size_t)ret != size)
293 		return (MEMSTAT_ERROR_KVM_SHORTREAD);
294 	return (0);
295 }
296 
297 /*
298  * memstat_kvm_uma() is similar to memstat_sysctl_uma(), only it extracts
299  * UMA(9) statistics from a kernel core/memory file.
300  */
301 int
302 memstat_kvm_uma(struct memory_type_list *list, void *kvm_handle)
303 {
304 	static LIST_HEAD(, uma_keg) uma_kegs;
305 	struct memory_type *mtp;
306 	struct uma_bucket *ubp, ub;
307 	struct uma_cache *ucp;
308 	struct uma_zone *uzp, uz;
309 	struct uma_keg *kzp, kz;
310 	int hint_dontsearch, i, mp_maxid, ret;
311 	char name[MEMTYPE_MAXNAME];
312 	kvm_t *kvm;
313 
314 	kvm = (kvm_t *)kvm_handle;
315 	hint_dontsearch = LIST_EMPTY(&list->mtl_list);
316 	if (kvm_nlist(kvm, namelist) != 0) {
317 		list->mtl_error = MEMSTAT_ERROR_KVM;
318 		return (-1);
319 	}
320 	if (namelist[X_UMA_KEGS].n_type == 0 ||
321 	    namelist[X_UMA_KEGS].n_value == 0) {
322 		list->mtl_error = MEMSTAT_ERROR_KVM_NOSYMBOL;
323 		return (-1);
324 	}
325 	ret = kread_symbol(kvm, X_MP_MAXID, &mp_maxid, sizeof(mp_maxid), 0);
326 	if (ret != 0) {
327 		list->mtl_error = ret;
328 		return (-1);
329 	}
330 	ret = kread_symbol(kvm, X_UMA_KEGS, &uma_kegs, sizeof(uma_kegs), 0);
331 	if (ret != 0) {
332 		list->mtl_error = ret;
333 		return (-1);
334 	}
335 	for (kzp = LIST_FIRST(&uma_kegs); kzp != NULL; kzp =
336 	    LIST_NEXT(&kz, uk_link)) {
337 		ret = kread(kvm, kzp, &kz, sizeof(kz), 0);
338 		if (ret != 0) {
339 			_memstat_mtl_empty(list);
340 			list->mtl_error = ret;
341 			return (-1);
342 		}
343 		for (uzp = LIST_FIRST(&kz.uk_zones); uzp != NULL; uzp =
344 		    LIST_NEXT(&uz, uz_link)) {
345 			ret = kread(kvm, uzp, &uz, sizeof(uz), 0);
346 			if (ret != 0) {
347 				_memstat_mtl_empty(list);
348 				list->mtl_error = ret;
349 				return (-1);
350 			}
351 			ret = kread_string(kvm, uz.uz_name, name,
352 			    MEMTYPE_MAXNAME);
353 			if (ret != 0) {
354 				_memstat_mtl_empty(list);
355 				list->mtl_error = ret;
356 				return (-1);
357 			}
358 			if (hint_dontsearch == 0) {
359 				mtp = memstat_mtl_find(list, ALLOCATOR_UMA,
360 				    name);
361 			} else
362 				mtp = NULL;
363 			if (mtp == NULL)
364 				mtp = _memstat_mt_allocate(list, ALLOCATOR_UMA,
365 				    name);
366 			if (mtp == NULL) {
367 				_memstat_mtl_empty(list);
368 				list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
369 				return (-1);
370 			}
371 			/*
372 			 * Reset the statistics on a current node.
373 			 */
374 			_memstat_mt_reset_stats(mtp);
375 			mtp->mt_numallocs = uz.uz_allocs;
376 			mtp->mt_numfrees = uz.uz_frees;
377 			mtp->mt_failures = uz.uz_fails;
378 			if (kz.uk_flags & UMA_ZFLAG_INTERNAL)
379 				goto skip_percpu;
380 			for (i = 0; i < mp_maxid + 1; i++) {
381 				ucp = &uz.uz_cpu[i];
382 				mtp->mt_numallocs += ucp->uc_allocs;
383 				mtp->mt_numfrees += ucp->uc_frees;
384 
385 				if (ucp->uc_allocbucket != NULL) {
386 					ret = kread(kvm, ucp->uc_allocbucket,
387 					    &ub, sizeof(ub), 0);
388 					if (ret != 0) {
389 						_memstat_mtl_empty(list);
390 						list->mtl_error =
391 						    MEMSTAT_ERROR_NOMEMORY;
392 						return (-1);
393 					}
394 					mtp->mt_free += ub.ub_cnt;
395 				}
396 				if (ucp->uc_freebucket != NULL) {
397 					ret = kread(kvm, ucp->uc_freebucket,
398 					    &ub, sizeof(ub), 0);
399 					if (ret != 0) {
400 						_memstat_mtl_empty(list);
401 						list->mtl_error =
402 						    MEMSTAT_ERROR_NOMEMORY;
403 						return (-1);
404 					}
405 					mtp->mt_free += ub.ub_cnt;
406 				}
407 			}
408 skip_percpu:
409 			mtp->mt_size = kz.uk_size;
410 			mtp->mt_memalloced = mtp->mt_numallocs * mtp->mt_size;
411 			mtp->mt_memfreed = mtp->mt_numfrees * mtp->mt_size;
412 			mtp->mt_bytes = mtp->mt_memalloced = mtp->mt_memfreed;
413 			if (kz.uk_ppera > 1)
414 				mtp->mt_countlimit = kz.uk_maxpages /
415 				    kz.uk_ipers;
416 			else
417 				mtp->mt_countlimit = kz.uk_maxpages *
418 				    kz.uk_ipers;
419 			mtp->mt_byteslimit = mtp->mt_countlimit * mtp->mt_size;
420 			mtp->mt_count = mtp->mt_numallocs - mtp->mt_numfrees;
421 			for (ubp = LIST_FIRST(&uz.uz_full_bucket); ubp !=
422 			    NULL; ubp = LIST_NEXT(&ub, ub_link)) {
423 				ret = kread(kvm, ubp, &ub, sizeof(ub), 0);
424 				mtp->mt_zonefree += ub.ub_cnt;
425 			}
426 			if (!((kz.uk_flags & UMA_ZONE_SECONDARY) &&
427 			    LIST_FIRST(&kz.uk_zones) != uzp)) {
428 				mtp->mt_kegfree = kz.uk_free;
429 				mtp->mt_free += mtp->mt_kegfree;
430 			}
431 			mtp->mt_free += mtp->mt_zonefree;
432 		}
433 	}
434 	return (0);
435 }
436