Add memstat_kvm_uma(), an implementation of a libmemstat(3) query routine

that knows how to extract UMA(9) allocator statistics from a core dump or
live memory image using kvm(3).  The caller is expected to provide the
necessary kvm_t handle, which is then used by libmemstat(3).

With these changes, it is trivially straight forward to re-introduce
vmstat -z support on core dumps, which was lost when UMA was introduced.

In the short term, this requires including vm/ include files that are not
intended for extra-kernel use, requiring in turn some ugliness.

5 files changed, 248 insertions, 2 deletions

diff --git a/lib/libmemstat/libmemstat.3 b/lib/libmemstat/libmemstat.3
index 0bc3410..a851e5e 100644
--- a/lib/libmemstat/libmemstat.3
+++ b/lib/libmemstat/libmemstat.3
@@ -53,6 +53,8 @@
 .Fn memstat_mtl_geterror "struct memory_type_list *list"
 .Ss Allocator Query Functions
 .Ft int
+.Fn memstat_kvm_uma "struct memory_type_list *list" "void *kvm_handle"
+.Ft int
 .Fn memstat_sysctl_all "struct memory_type_list *list" "int flags"
 .Ft int
 .Fn memstat_sysctl_malloc "struct memory_type_list *list" "int flags"
@@ -145,6 +147,7 @@ and freed on completion using
 .Fn memstat_mtl_free .
 Lists of memory types are populated via calls that query the kernel for
 statistics information; currently:
+.Fn memstat_kvm_uma ,
 .Fn memstat_sysctl_all ,
 .Fn memstat_sysctl_uma ,
 and
@@ -443,8 +446,10 @@ monitoring library, along with the library, were written by
 .An Robert Watson Aq rwatson@FreeBSD.org
 .Sh BUGS
 .Nm
-cannot yet extract statistics from kernel core dumps, although this should be
-straight forward to implement.
+cannot yet extract
+.Xr malloc 9
+statistics from kernel core dumps, although this should be straight forward
+to implement.
 .Pp
 Once a memory type is present on a memory type list, it will not be removed
 even if the kernel no longer presents information on the type via its
diff --git a/lib/libmemstat/memstat.c b/lib/libmemstat/memstat.c
index de8ae65..28e4813 100644
--- a/lib/libmemstat/memstat.c
+++ b/lib/libmemstat/memstat.c
@@ -53,6 +53,12 @@ memstat_strerror(int error)
 		return ("Too many CPUs");
 	case MEMSTAT_ERROR_DATAERROR:
 		return ("Data format error");
+	case MEMSTAT_ERROR_KVM:
+		return ("KVM error");
+	case MEMSTAT_ERROR_KVM_NOSYMBOL:
+		return ("KVM unable to find symbol");
+	case MEMSTAT_ERROR_KVM_SHORTREAD:
+		return ("KVM short read");
 	case MEMSTAT_ERROR_UNDEFINED:
 	default:
 		return ("Unknown error");
diff --git a/lib/libmemstat/memstat.h b/lib/libmemstat/memstat.h
index f466408..561f6ee 100644
--- a/lib/libmemstat/memstat.h
+++ b/lib/libmemstat/memstat.h
@@ -72,6 +72,9 @@
 #define	MEMSTAT_ERROR_PERMISSION	3	/* Permission denied. */
 #define	MEMSTAT_ERROR_TOOMANYCPUS	4	/* Too many CPUs. */
 #define	MEMSTAT_ERROR_DATAERROR		5	/* Error in stat data. */
+#define	MEMSTAT_ERROR_KVM		6	/* See kvm_geterr() for err. */
+#define	MEMSTAT_ERROR_KVM_NOSYMBOL	7	/* Symbol not available. */
+#define	MEMSTAT_ERROR_KVM_SHORTREAD	8	/* Short kvm_read return. */
 
 /*
  * Forward declare struct memory_type, which holds per-type properties and
@@ -113,6 +116,12 @@ int	memstat_sysctl_malloc(struct memory_type_list *list, int flags);
 int	memstat_sysctl_uma(struct memory_type_list *list, int flags);
 
 /*
+ * Functions to retrieve data from a kernel core (or /dev/kmem).
+ */
+int	memstat_kvm_all(struct memory_type_list *list, void *kvm_handle);
+int	memstat_kvm_uma(struct memory_type_list *list, void *kvm_handle);
+
+/*
  * Accessor methods for struct memory_type.
  */
 const char	*memstat_get_name(const struct memory_type *mtp);
diff --git a/lib/libmemstat/memstat_all.c b/lib/libmemstat/memstat_all.c
index e32047c..b363ab9 100644
--- a/lib/libmemstat/memstat_all.c
+++ b/lib/libmemstat/memstat_all.c
@@ -45,3 +45,16 @@ memstat_sysctl_all(struct memory_type_list *mtlp, int flags)
 		return (-1);
 	return (0);
 }
+
+int
+memstat_kvm_all(struct memory_type_list *mtlp, void *kvm_handle)
+{
+
+#if NOTYET
+	if (memstat_kvm_malloc(mtlp, kvm_handle) < 0)
+		return (-1);
+#endif
+	if (memstat_kvm_uma(mtlp, kvm_handle) < 0)
+		return (-1);
+	return (0);
+}
diff --git a/lib/libmemstat/memstat_uma.c b/lib/libmemstat/memstat_uma.c
index f1fb710..9572384 100644
--- a/lib/libmemstat/memstat_uma.c
+++ b/lib/libmemstat/memstat_uma.c
@@ -29,10 +29,25 @@
 #include <sys/param.h>
 #include <sys/sysctl.h>
 
+/*
+ * XXX: Grubbing around in UMA(9) using libkvm requires internal knowledge of
+ * a number of VM-related bits.  The ifdefs around those bits are not
+ * designed with a nosy user-space consumer in mind.
+ */
+#include <vm/vm.h>
+#define	_KERNEL		/* XXX: vm_page.h confusion. */
+#define	KLD_MODULE	/* XXX: vm_page.h shouldn't include opt_vmpage.h. */
+#include <vm/vm_page.h>
+#undef KLD_MODULE
+#undef _KERNEL
+
 #include <vm/uma.h>
+#include <vm/uma_int.h>
 
 #include <err.h>
 #include <errno.h>
+#include <kvm.h>
+#include <nlist.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
@@ -40,6 +55,14 @@
 #include "memstat.h"
 #include "memstat_internal.h"
 
+static struct nlist namelist[] = {
+#define	X_UMA_KEGS	0
+	{ .n_name = "_uma_kegs" },
+#define	X_MP_MAXID	1
+	{ .n_name = "_mp_maxid" },
+	{ .n_name = "" },
+};
+
 /*
  * Extract uma(9) statistics from the running kernel, and store all memory
  * type information in the passed list.  For each type, check the list for an
@@ -228,3 +251,193 @@ retry:
 
 	return (0);
 }
+
+static int
+kread(kvm_t *kvm, void *kvm_pointer, void *address, size_t size,
+    size_t offset)
+{
+	ssize_t ret;
+
+	ret = kvm_read(kvm, (unsigned long)kvm_pointer + offset, address,
+	    size);
+	if (ret < 0)
+		return (MEMSTAT_ERROR_KVM);
+	if ((size_t)ret != size)
+		return (MEMSTAT_ERROR_KVM_SHORTREAD);
+	return (0);
+}
+
+static int
+kread_string(kvm_t *kvm, void *kvm_pointer, char *buffer, int buflen)
+{
+	ssize_t ret;
+	int i;
+
+	for (i = 0; i < buflen; i++) {
+		ret = kvm_read(kvm, (unsigned long)kvm_pointer + i,
+		    &(buffer[i]), sizeof(char));
+		if (ret < 0)
+			return (MEMSTAT_ERROR_KVM);
+		if ((size_t)ret != sizeof(char))
+			return (MEMSTAT_ERROR_KVM_SHORTREAD);
+		if (buffer[i] == '\0')
+			return (0);
+	}
+	/* Truncate. */
+	buffer[i-1] = '\0';
+	return (0);
+}
+
+static int
+kread_symbol(kvm_t *kvm, int index, void *address, size_t size,
+    size_t offset)
+{
+	ssize_t ret;
+
+	ret = kvm_read(kvm, namelist[index].n_value + offset, address, size);
+	if (ret < 0)
+		return (MEMSTAT_ERROR_KVM);
+	if ((size_t)ret != size)
+		return (MEMSTAT_ERROR_KVM_SHORTREAD);
+	return (0);
+}
+
+/*
+ * memstat_kvm_uma() is similar to memstat_sysctl_uma(), only it extracts
+ * UMA(9) statistics from a kernel core/memory file.
+ */
+int
+memstat_kvm_uma(struct memory_type_list *list, void *kvm_handle)
+{
+	static LIST_HEAD(, uma_keg) uma_kegs;
+	struct memory_type *mtp;
+	struct uma_bucket *ubp, ub;
+	struct uma_cache *ucp;
+	struct uma_zone *uzp, uz;
+	struct uma_keg *kzp, kz;
+	int hint_dontsearch, i, mp_maxid, ret;
+	char name[MEMTYPE_MAXNAME];
+	kvm_t *kvm;
+
+	kvm = (kvm_t *)kvm_handle;
+	hint_dontsearch = LIST_EMPTY(&list->mtl_list);
+	if (kvm_nlist(kvm, namelist) != 0) {
+		list->mtl_error = MEMSTAT_ERROR_KVM;
+		return (-1);
+	}
+	if (namelist[X_UMA_KEGS].n_type == 0 ||
+	    namelist[X_UMA_KEGS].n_value == 0) {
+		list->mtl_error = MEMSTAT_ERROR_KVM_NOSYMBOL;
+		return (-1);
+	}
+	ret = kread_symbol(kvm, X_MP_MAXID, &mp_maxid, sizeof(mp_maxid), 0);
+	if (ret != 0) {
+		list->mtl_error = ret;
+		return (-1);
+	}
+	ret = kread_symbol(kvm, X_UMA_KEGS, &uma_kegs, sizeof(uma_kegs), 0);
+	if (ret != 0) {
+		list->mtl_error = ret;
+		return (-1);
+	}
+	for (kzp = LIST_FIRST(&uma_kegs); kzp != NULL; kzp =
+	    LIST_NEXT(&kz, uk_link)) {
+		ret = kread(kvm, kzp, &kz, sizeof(kz), 0);
+		if (ret != 0) {
+			_memstat_mtl_empty(list);
+			list->mtl_error = ret;
+			return (-1);
+		}
+		for (uzp = LIST_FIRST(&kz.uk_zones); uzp != NULL; uzp =
+		    LIST_NEXT(&uz, uz_link)) {
+			ret = kread(kvm, uzp, &uz, sizeof(uz), 0);
+			if (ret != 0) {
+				_memstat_mtl_empty(list);
+				list->mtl_error = ret;
+				return (-1);
+			}
+			ret = kread_string(kvm, uz.uz_name, name,
+			    MEMTYPE_MAXNAME);
+			if (ret != 0) {
+				_memstat_mtl_empty(list);
+				list->mtl_error = ret;
+				return (-1);
+			}
+			if (hint_dontsearch == 0) {
+				mtp = memstat_mtl_find(list, ALLOCATOR_UMA,
+				    name);
+			} else
+				mtp = NULL;
+			if (mtp == NULL)
+				mtp = _memstat_mt_allocate(list, ALLOCATOR_UMA,
+				    name);
+			if (mtp == NULL) {
+				_memstat_mtl_empty(list);
+				list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
+				return (-1);
+			}
+			/*
+			 * Reset the statistics on a current node.
+			 */
+			_memstat_mt_reset_stats(mtp);
+			mtp->mt_numallocs = uz.uz_allocs;
+			mtp->mt_numfrees = uz.uz_frees;
+			mtp->mt_failures = uz.uz_fails;
+			if (kz.uk_flags & UMA_ZFLAG_INTERNAL)
+				goto skip_percpu;
+			for (i = 0; i < mp_maxid + 1; i++) {
+				ucp = &uz.uz_cpu[i];
+				mtp->mt_numallocs += ucp->uc_allocs;
+				mtp->mt_numfrees += ucp->uc_frees;
+
+				if (ucp->uc_allocbucket != NULL) {
+					ret = kread(kvm, ucp->uc_allocbucket,
+					    &ub, sizeof(ub), 0);
+					if (ret != 0) {
+						_memstat_mtl_empty(list);
+						list->mtl_error =
+						    MEMSTAT_ERROR_NOMEMORY;
+						return (-1);
+					}
+					mtp->mt_free += ub.ub_cnt;
+				}
+				if (ucp->uc_freebucket != NULL) {
+					ret = kread(kvm, ucp->uc_freebucket,
+					    &ub, sizeof(ub), 0);
+					if (ret != 0) {
+						_memstat_mtl_empty(list);
+						list->mtl_error =
+						    MEMSTAT_ERROR_NOMEMORY;
+						return (-1);
+					}
+					mtp->mt_free += ub.ub_cnt;
+				}
+			}
+skip_percpu:
+			mtp->mt_size = kz.uk_size;
+			mtp->mt_memalloced = mtp->mt_numallocs * mtp->mt_size;
+			mtp->mt_memfreed = mtp->mt_numfrees * mtp->mt_size;
+			mtp->mt_bytes = mtp->mt_memalloced = mtp->mt_memfreed;
+			if (kz.uk_ppera > 1)
+				mtp->mt_countlimit = kz.uk_maxpages /
+				    kz.uk_ipers;
+			else
+				mtp->mt_countlimit = kz.uk_maxpages *
+				    kz.uk_ipers;
+			mtp->mt_byteslimit = mtp->mt_countlimit * mtp->mt_size;
+			mtp->mt_count = mtp->mt_numallocs - mtp->mt_numfrees;
+			for (ubp = LIST_FIRST(&uz.uz_full_bucket); ubp !=
+			    NULL; ubp = LIST_NEXT(&ub, ub_link)) {
+				ret = kread(kvm, ubp, &ub, sizeof(ub), 0);
+				mtp->mt_zonefree += ub.ub_cnt;
+			}
+			if (!((kz.uk_flags & UMA_ZONE_SECONDARY) &&
+			    LIST_FIRST(&kz.uk_zones) != uzp)) {
+				mtp->mt_kegfree = kz.uk_free;
+				mtp->mt_free += mtp->mt_kegfree;
+			}
+			mtp->mt_free += mtp->mt_zonefree;
+		}
+	}
+	return (0);
+}