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