memstat_uma.c revision 330897 
1/*-
2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3 *
4 * Copyright (c) 2005-2006 Robert N. M. Watson
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 *    notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 *    notice, this list of conditions and the following disclaimer in the
14 *    documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 *
28 * $FreeBSD: stable/11/lib/libmemstat/memstat_uma.c 330897 2018-03-14 03:19:51Z eadler $
29 */
30
31#include <sys/param.h>
32#include <sys/cpuset.h>
33#include <sys/sysctl.h>
34
35#include <vm/vm.h>
36#include <vm/vm_page.h>
37
38#include <vm/uma.h>
39#include <vm/uma_int.h>
40
41#include <err.h>
42#include <errno.h>
43#include <kvm.h>
44#include <nlist.h>
45#include <stddef.h>
46#include <stdio.h>
47#include <stdlib.h>
48#include <string.h>
49#include <unistd.h>
50
51#include "memstat.h"
52#include "memstat_internal.h"
53
54static struct nlist namelist[] = {
55#define	X_UMA_KEGS	0
56	{ .n_name = "_uma_kegs" },
57#define	X_MP_MAXID	1
58	{ .n_name = "_mp_maxid" },
59#define	X_ALL_CPUS	2
60	{ .n_name = "_all_cpus" },
61	{ .n_name = "" },
62};
63
64/*
65 * Extract uma(9) statistics from the running kernel, and store all memory
66 * type information in the passed list.  For each type, check the list for an
67 * existing entry with the right name/allocator -- if present, update that
68 * entry.  Otherwise, add a new entry.  On error, the entire list will be
69 * cleared, as entries will be in an inconsistent state.
70 *
71 * To reduce the level of work for a list that starts empty, we keep around a
72 * hint as to whether it was empty when we began, so we can avoid searching
73 * the list for entries to update.  Updates are O(n^2) due to searching for
74 * each entry before adding it.
75 */
76int
77memstat_sysctl_uma(struct memory_type_list *list, int flags)
78{
79	struct uma_stream_header *ushp;
80	struct uma_type_header *uthp;
81	struct uma_percpu_stat *upsp;
82	struct memory_type *mtp;
83	int count, hint_dontsearch, i, j, maxcpus, maxid;
84	char *buffer, *p;
85	size_t size;
86
87	hint_dontsearch = LIST_EMPTY(&list->mtl_list);
88
89	/*
90	 * Query the number of CPUs, number of malloc types so that we can
91	 * guess an initial buffer size.  We loop until we succeed or really
92	 * fail.  Note that the value of maxcpus we query using sysctl is not
93	 * the version we use when processing the real data -- that is read
94	 * from the header.
95	 */
96retry:
97	size = sizeof(maxid);
98	if (sysctlbyname("kern.smp.maxid", &maxid, &size, NULL, 0) < 0) {
99		if (errno == EACCES || errno == EPERM)
100			list->mtl_error = MEMSTAT_ERROR_PERMISSION;
101		else
102			list->mtl_error = MEMSTAT_ERROR_DATAERROR;
103		return (-1);
104	}
105	if (size != sizeof(maxid)) {
106		list->mtl_error = MEMSTAT_ERROR_DATAERROR;
107		return (-1);
108	}
109
110	size = sizeof(count);
111	if (sysctlbyname("vm.zone_count", &count, &size, NULL, 0) < 0) {
112		if (errno == EACCES || errno == EPERM)
113			list->mtl_error = MEMSTAT_ERROR_PERMISSION;
114		else
115			list->mtl_error = MEMSTAT_ERROR_VERSION;
116		return (-1);
117	}
118	if (size != sizeof(count)) {
119		list->mtl_error = MEMSTAT_ERROR_DATAERROR;
120		return (-1);
121	}
122
123	size = sizeof(*uthp) + count * (sizeof(*uthp) + sizeof(*upsp) *
124	    (maxid + 1));
125
126	buffer = malloc(size);
127	if (buffer == NULL) {
128		list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
129		return (-1);
130	}
131
132	if (sysctlbyname("vm.zone_stats", buffer, &size, NULL, 0) < 0) {
133		/*
134		 * XXXRW: ENOMEM is an ambiguous return, we should bound the
135		 * number of loops, perhaps.
136		 */
137		if (errno == ENOMEM) {
138			free(buffer);
139			goto retry;
140		}
141		if (errno == EACCES || errno == EPERM)
142			list->mtl_error = MEMSTAT_ERROR_PERMISSION;
143		else
144			list->mtl_error = MEMSTAT_ERROR_VERSION;
145		free(buffer);
146		return (-1);
147	}
148
149	if (size == 0) {
150		free(buffer);
151		return (0);
152	}
153
154	if (size < sizeof(*ushp)) {
155		list->mtl_error = MEMSTAT_ERROR_VERSION;
156		free(buffer);
157		return (-1);
158	}
159	p = buffer;
160	ushp = (struct uma_stream_header *)p;
161	p += sizeof(*ushp);
162
163	if (ushp->ush_version != UMA_STREAM_VERSION) {
164		list->mtl_error = MEMSTAT_ERROR_VERSION;
165		free(buffer);
166		return (-1);
167	}
168
169	/*
170	 * For the remainder of this function, we are quite trusting about
171	 * the layout of structures and sizes, since we've determined we have
172	 * a matching version and acceptable CPU count.
173	 */
174	maxcpus = ushp->ush_maxcpus;
175	count = ushp->ush_count;
176	for (i = 0; i < count; i++) {
177		uthp = (struct uma_type_header *)p;
178		p += sizeof(*uthp);
179
180		if (hint_dontsearch == 0) {
181			mtp = memstat_mtl_find(list, ALLOCATOR_UMA,
182			    uthp->uth_name);
183		} else
184			mtp = NULL;
185		if (mtp == NULL)
186			mtp = _memstat_mt_allocate(list, ALLOCATOR_UMA,
187			    uthp->uth_name, maxid + 1);
188		if (mtp == NULL) {
189			_memstat_mtl_empty(list);
190			free(buffer);
191			list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
192			return (-1);
193		}
194
195		/*
196		 * Reset the statistics on a current node.
197		 */
198		_memstat_mt_reset_stats(mtp, maxid + 1);
199
200		mtp->mt_numallocs = uthp->uth_allocs;
201		mtp->mt_numfrees = uthp->uth_frees;
202		mtp->mt_failures = uthp->uth_fails;
203		mtp->mt_sleeps = uthp->uth_sleeps;
204
205		for (j = 0; j < maxcpus; j++) {
206			upsp = (struct uma_percpu_stat *)p;
207			p += sizeof(*upsp);
208
209			mtp->mt_percpu_cache[j].mtp_free =
210			    upsp->ups_cache_free;
211			mtp->mt_free += upsp->ups_cache_free;
212			mtp->mt_numallocs += upsp->ups_allocs;
213			mtp->mt_numfrees += upsp->ups_frees;
214		}
215
216		mtp->mt_size = uthp->uth_size;
217		mtp->mt_rsize = uthp->uth_rsize;
218		mtp->mt_memalloced = mtp->mt_numallocs * uthp->uth_size;
219		mtp->mt_memfreed = mtp->mt_numfrees * uthp->uth_size;
220		mtp->mt_bytes = mtp->mt_memalloced - mtp->mt_memfreed;
221		mtp->mt_countlimit = uthp->uth_limit;
222		mtp->mt_byteslimit = uthp->uth_limit * uthp->uth_size;
223
224		mtp->mt_count = mtp->mt_numallocs - mtp->mt_numfrees;
225		mtp->mt_zonefree = uthp->uth_zone_free;
226
227		/*
228		 * UMA secondary zones share a keg with the primary zone.  To
229		 * avoid double-reporting of free items, report keg free
230		 * items only in the primary zone.
231		 */
232		if (!(uthp->uth_zone_flags & UTH_ZONE_SECONDARY)) {
233			mtp->mt_kegfree = uthp->uth_keg_free;
234			mtp->mt_free += mtp->mt_kegfree;
235		}
236		mtp->mt_free += mtp->mt_zonefree;
237	}
238
239	free(buffer);
240
241	return (0);
242}
243
244static int
245kread(kvm_t *kvm, void *kvm_pointer, void *address, size_t size,
246    size_t offset)
247{
248	ssize_t ret;
249
250	ret = kvm_read(kvm, (unsigned long)kvm_pointer + offset, address,
251	    size);
252	if (ret < 0)
253		return (MEMSTAT_ERROR_KVM);
254	if ((size_t)ret != size)
255		return (MEMSTAT_ERROR_KVM_SHORTREAD);
256	return (0);
257}
258
259static int
260kread_string(kvm_t *kvm, const void *kvm_pointer, char *buffer, int buflen)
261{
262	ssize_t ret;
263	int i;
264
265	for (i = 0; i < buflen; i++) {
266		ret = kvm_read(kvm, (unsigned long)kvm_pointer + i,
267		    &(buffer[i]), sizeof(char));
268		if (ret < 0)
269			return (MEMSTAT_ERROR_KVM);
270		if ((size_t)ret != sizeof(char))
271			return (MEMSTAT_ERROR_KVM_SHORTREAD);
272		if (buffer[i] == '\0')
273			return (0);
274	}
275	/* Truncate. */
276	buffer[i-1] = '\0';
277	return (0);
278}
279
280static int
281kread_symbol(kvm_t *kvm, int index, void *address, size_t size,
282    size_t offset)
283{
284	ssize_t ret;
285
286	ret = kvm_read(kvm, namelist[index].n_value + offset, address, size);
287	if (ret < 0)
288		return (MEMSTAT_ERROR_KVM);
289	if ((size_t)ret != size)
290		return (MEMSTAT_ERROR_KVM_SHORTREAD);
291	return (0);
292}
293
294/*
295 * memstat_kvm_uma() is similar to memstat_sysctl_uma(), only it extracts
296 * UMA(9) statistics from a kernel core/memory file.
297 */
298int
299memstat_kvm_uma(struct memory_type_list *list, void *kvm_handle)
300{
301	LIST_HEAD(, uma_keg) uma_kegs;
302	struct memory_type *mtp;
303	struct uma_bucket *ubp, ub;
304	struct uma_cache *ucp, *ucp_array;
305	struct uma_zone *uzp, uz;
306	struct uma_keg *kzp, kz;
307	int hint_dontsearch, i, mp_maxid, ret;
308	char name[MEMTYPE_MAXNAME];
309	cpuset_t all_cpus;
310	long cpusetsize;
311	kvm_t *kvm;
312
313	kvm = (kvm_t *)kvm_handle;
314	hint_dontsearch = LIST_EMPTY(&list->mtl_list);
315	if (kvm_nlist(kvm, namelist) != 0) {
316		list->mtl_error = MEMSTAT_ERROR_KVM;
317		return (-1);
318	}
319	if (namelist[X_UMA_KEGS].n_type == 0 ||
320	    namelist[X_UMA_KEGS].n_value == 0) {
321		list->mtl_error = MEMSTAT_ERROR_KVM_NOSYMBOL;
322		return (-1);
323	}
324	ret = kread_symbol(kvm, X_MP_MAXID, &mp_maxid, sizeof(mp_maxid), 0);
325	if (ret != 0) {
326		list->mtl_error = ret;
327		return (-1);
328	}
329	ret = kread_symbol(kvm, X_UMA_KEGS, &uma_kegs, sizeof(uma_kegs), 0);
330	if (ret != 0) {
331		list->mtl_error = ret;
332		return (-1);
333	}
334	cpusetsize = sysconf(_SC_CPUSET_SIZE);
335	if (cpusetsize == -1 || (u_long)cpusetsize > sizeof(cpuset_t)) {
336		list->mtl_error = MEMSTAT_ERROR_KVM_NOSYMBOL;
337		return (-1);
338	}
339	CPU_ZERO(&all_cpus);
340	ret = kread_symbol(kvm, X_ALL_CPUS, &all_cpus, cpusetsize, 0);
341	if (ret != 0) {
342		list->mtl_error = ret;
343		return (-1);
344	}
345	ucp_array = malloc(sizeof(struct uma_cache) * (mp_maxid + 1));
346	if (ucp_array == NULL) {
347		list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
348		return (-1);
349	}
350	for (kzp = LIST_FIRST(&uma_kegs); kzp != NULL; kzp =
351	    LIST_NEXT(&kz, uk_link)) {
352		ret = kread(kvm, kzp, &kz, sizeof(kz), 0);
353		if (ret != 0) {
354			free(ucp_array);
355			_memstat_mtl_empty(list);
356			list->mtl_error = ret;
357			return (-1);
358		}
359		for (uzp = LIST_FIRST(&kz.uk_zones); uzp != NULL; uzp =
360		    LIST_NEXT(&uz, uz_link)) {
361			ret = kread(kvm, uzp, &uz, sizeof(uz), 0);
362			if (ret != 0) {
363				free(ucp_array);
364				_memstat_mtl_empty(list);
365				list->mtl_error = ret;
366				return (-1);
367			}
368			ret = kread(kvm, uzp, ucp_array,
369			    sizeof(struct uma_cache) * (mp_maxid + 1),
370			    offsetof(struct uma_zone, uz_cpu[0]));
371			if (ret != 0) {
372				free(ucp_array);
373				_memstat_mtl_empty(list);
374				list->mtl_error = ret;
375				return (-1);
376			}
377			ret = kread_string(kvm, uz.uz_name, name,
378			    MEMTYPE_MAXNAME);
379			if (ret != 0) {
380				free(ucp_array);
381				_memstat_mtl_empty(list);
382				list->mtl_error = ret;
383				return (-1);
384			}
385			if (hint_dontsearch == 0) {
386				mtp = memstat_mtl_find(list, ALLOCATOR_UMA,
387				    name);
388			} else
389				mtp = NULL;
390			if (mtp == NULL)
391				mtp = _memstat_mt_allocate(list, ALLOCATOR_UMA,
392				    name, mp_maxid + 1);
393			if (mtp == NULL) {
394				free(ucp_array);
395				_memstat_mtl_empty(list);
396				list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
397				return (-1);
398			}
399			/*
400			 * Reset the statistics on a current node.
401			 */
402			_memstat_mt_reset_stats(mtp, mp_maxid + 1);
403			mtp->mt_numallocs = uz.uz_allocs;
404			mtp->mt_numfrees = uz.uz_frees;
405			mtp->mt_failures = uz.uz_fails;
406			mtp->mt_sleeps = uz.uz_sleeps;
407			if (kz.uk_flags & UMA_ZFLAG_INTERNAL)
408				goto skip_percpu;
409			for (i = 0; i < mp_maxid + 1; i++) {
410				if (!CPU_ISSET(i, &all_cpus))
411					continue;
412				ucp = &ucp_array[i];
413				mtp->mt_numallocs += ucp->uc_allocs;
414				mtp->mt_numfrees += ucp->uc_frees;
415
416				if (ucp->uc_allocbucket != NULL) {
417					ret = kread(kvm, ucp->uc_allocbucket,
418					    &ub, sizeof(ub), 0);
419					if (ret != 0) {
420						free(ucp_array);
421						_memstat_mtl_empty(list);
422						list->mtl_error = ret;
423						return (-1);
424					}
425					mtp->mt_free += ub.ub_cnt;
426				}
427				if (ucp->uc_freebucket != NULL) {
428					ret = kread(kvm, ucp->uc_freebucket,
429					    &ub, sizeof(ub), 0);
430					if (ret != 0) {
431						free(ucp_array);
432						_memstat_mtl_empty(list);
433						list->mtl_error = ret;
434						return (-1);
435					}
436					mtp->mt_free += ub.ub_cnt;
437				}
438			}
439skip_percpu:
440			mtp->mt_size = kz.uk_size;
441			mtp->mt_rsize = kz.uk_rsize;
442			mtp->mt_memalloced = mtp->mt_numallocs * mtp->mt_size;
443			mtp->mt_memfreed = mtp->mt_numfrees * mtp->mt_size;
444			mtp->mt_bytes = mtp->mt_memalloced - mtp->mt_memfreed;
445			if (kz.uk_ppera > 1)
446				mtp->mt_countlimit = kz.uk_maxpages /
447				    kz.uk_ipers;
448			else
449				mtp->mt_countlimit = kz.uk_maxpages *
450				    kz.uk_ipers;
451			mtp->mt_byteslimit = mtp->mt_countlimit * mtp->mt_size;
452			mtp->mt_count = mtp->mt_numallocs - mtp->mt_numfrees;
453			for (ubp = LIST_FIRST(&uz.uz_buckets); ubp !=
454			    NULL; ubp = LIST_NEXT(&ub, ub_link)) {
455				ret = kread(kvm, ubp, &ub, sizeof(ub), 0);
456				mtp->mt_zonefree += ub.ub_cnt;
457			}
458			if (!((kz.uk_flags & UMA_ZONE_SECONDARY) &&
459			    LIST_FIRST(&kz.uk_zones) != uzp)) {
460				mtp->mt_kegfree = kz.uk_free;
461				mtp->mt_free += mtp->mt_kegfree;
462			}
463			mtp->mt_free += mtp->mt_zonefree;
464		}
465	}
466	free(ucp_array);
467	return (0);
468}
469