OS-4869 writes to /dev/kmem should be logged to a ring buffer

8 files changed, 280 insertions, 0 deletions

diff --git a/usr/src/cmd/mdb/Makefile.common b/usr/src/cmd/mdb/Makefile.common
index 6644e8bc25..9b8d438a88 100644
--- a/usr/src/cmd/mdb/Makefile.common
+++ b/usr/src/cmd/mdb/Makefile.common
@@ -20,6 +20,7 @@
 #
 # Copyright (c) 1999, 2010, Oracle and/or its affiliates. All rights reserved.
 # Copyright 2013 Nexenta Systems, Inc.  All rights reserved.
+# Copyright (c) 2015, Joyent, Inc. All rights reserved.
 #
 # MDB modules used for debugging user processes that every ISA's build
 # subdirectory will need to build.
@@ -75,6 +76,7 @@ COMMON_MODULES_KVM = \
 	logindmux \
 	mac \
 	md \
+	mm \
 	mpt_sas \
 	mr_sas \
 	nca \
diff --git a/usr/src/cmd/mdb/common/modules/mm/mm.c b/usr/src/cmd/mdb/common/modules/mm/mm.c
new file mode 100644
index 0000000000..08b6aac770
--- /dev/null
+++ b/usr/src/cmd/mdb/common/modules/mm/mm.c
@@ -0,0 +1,140 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright (c) 2015, Joyent, Inc.  All rights reserved.
+ */
+
+#include <sys/mdb_modapi.h>
+#include <sys/time.h>
+#include <sys/mem.h>
+
+typedef struct kmemlog_walk {
+	uintptr_t kmlw_addr;
+	mm_logentry_t *kmlw_entries;
+	int kmlw_nentries;
+	int kmlw_entry;
+	int kmlw_oldest;
+} kmemlog_walk_t;
+
+static int
+kmemlog_walk_init(mdb_walk_state_t *wsp)
+{
+	kmemlog_walk_t *kw;
+	GElf_Sym sym;
+
+	if (mdb_lookup_by_name("mm_kmemlog", &sym) != 0) {
+		mdb_warn("couldn't find symbol 'mm_kmemlog'");
+		return (WALK_ERR);
+	}
+
+	kw = mdb_zalloc(sizeof (kmemlog_walk_t), UM_SLEEP);
+	kw->kmlw_entries = mdb_zalloc(sym.st_size, UM_SLEEP);
+	kw->kmlw_addr = sym.st_value;
+
+	if (mdb_vread(kw->kmlw_entries, sym.st_size, sym.st_value) == -1) {
+		mdb_warn("couldn't read log at %p", sym.st_value);
+		mdb_free(kw->kmlw_entries, sym.st_size);
+		mdb_free(kw, sizeof (kmemlog_walk_t));
+		return (WALK_ERR);
+	}
+
+	kw->kmlw_nentries = sym.st_size / sizeof (mm_logentry_t);
+
+	mdb_readvar(&kw->kmlw_entry, "mm_kmemlogent");
+	kw->kmlw_oldest = kw->kmlw_entry;
+	wsp->walk_data = kw;
+
+	return (WALK_NEXT);
+}
+
+static int
+kmemlog_walk_step(mdb_walk_state_t *wsp)
+{
+	kmemlog_walk_t *kw = wsp->walk_data;
+	mm_logentry_t *ent;
+	int rval = WALK_NEXT;
+
+	ent = &kw->kmlw_entries[kw->kmlw_entry];
+
+	if (++kw->kmlw_entry == kw->kmlw_nentries)
+		kw->kmlw_entry = 0;
+
+	if (ent->mle_hrtime != 0) {
+		rval = wsp->walk_callback(kw->kmlw_addr + ((uintptr_t)ent -
+		    (uintptr_t)kw->kmlw_entries), ent, wsp->walk_cbdata);
+	}
+
+	if (rval == WALK_NEXT && kw->kmlw_entry == kw->kmlw_oldest)
+		return (WALK_DONE);
+
+	return (rval);
+}
+
+static void
+kmemlog_walk_fini(mdb_walk_state_t *wsp)
+{
+	kmemlog_walk_t *kw = wsp->walk_data;
+
+	mdb_free(kw->kmlw_entries, kw->kmlw_nentries * sizeof (mm_logentry_t));
+	mdb_free(kw, sizeof (kmemlog_walk_t));
+}
+
+static int
+kmemlog(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
+{
+	mm_logentry_t ent;
+
+	if (!(flags & DCMD_ADDRSPEC)) {
+		if (mdb_walk_dcmd("kmemlog", "kmemlog", argc, argv) == -1) {
+			mdb_warn("can't walk 'kmemlog'");
+			return (DCMD_ERR);
+		}
+		return (DCMD_OK);
+	}
+
+	if (DCMD_HDRSPEC(flags)) {
+		mdb_printf("%?s %-20s %?s %5s %s\n",
+		    "ADDR", "TIME", "VADDR", "PID", "PSARGS");
+	}
+
+	if (mdb_vread(&ent, sizeof (ent), addr) == -1) {
+		mdb_warn("can't read mm_logentry_t at %p", addr);
+		return (DCMD_ERR);
+	}
+
+	mdb_printf("%?p %-20Y %?p %5d %s\n",
+	    addr, ent.mle_hrestime.tv_sec, ent.mle_vaddr, ent.mle_pid,
+	    ent.mle_psargs);
+
+	return (DCMD_OK);
+}
+
+static const mdb_dcmd_t dcmds[] = {
+	{ "kmemlog", NULL, "print log of writes via /dev/kmem", kmemlog },
+	{ NULL }
+};
+
+static const mdb_walker_t walkers[] = {
+	{ "kmemlog", "walk entries in /dev/kmem write log",
+		kmemlog_walk_init, kmemlog_walk_step, kmemlog_walk_fini },
+	{ NULL }
+};
+
+static const mdb_modinfo_t modinfo = {
+	MDB_API_VERSION, dcmds, walkers
+};
+
+const mdb_modinfo_t *
+_mdb_init(void)
+{
+	return (&modinfo);
+}
diff --git a/usr/src/cmd/mdb/intel/amd64/mm/Makefile b/usr/src/cmd/mdb/intel/amd64/mm/Makefile
new file mode 100644
index 0000000000..216385a60c
--- /dev/null
+++ b/usr/src/cmd/mdb/intel/amd64/mm/Makefile
@@ -0,0 +1,24 @@
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+
+#
+# Copyright (c) 2015, Joyent, Inc. All rights reserved.
+#
+
+MODULE = mm.so
+MDBTGT = kvm
+
+MODSRCS = mm.c
+
+include ../../../../Makefile.cmd
+include ../../../../Makefile.cmd.64
+include ../../Makefile.amd64
+include ../../../Makefile.module
diff --git a/usr/src/cmd/mdb/intel/ia32/mm/Makefile b/usr/src/cmd/mdb/intel/ia32/mm/Makefile
new file mode 100644
index 0000000000..e37ee86753
--- /dev/null
+++ b/usr/src/cmd/mdb/intel/ia32/mm/Makefile
@@ -0,0 +1,24 @@
+#
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+
+#
+# Copyright (c) 2015, Joyent, Inc. All rights reserved.
+#
+
+MODULE = mm.so
+MDBTGT = kvm
+
+MODSRCS = mm.c
+
+include ../../../../Makefile.cmd
+include ../../Makefile.ia32
+include ../../../Makefile.module
diff --git a/usr/src/cmd/mdb/sparc/v9/mm/Makefile b/usr/src/cmd/mdb/sparc/v9/mm/Makefile
new file mode 100644
index 0000000000..c49f4f4484
--- /dev/null
+++ b/usr/src/cmd/mdb/sparc/v9/mm/Makefile
@@ -0,0 +1,25 @@
+#
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+
+#
+# Copyright (c) 2015, Joyent, Inc. All rights reserved.
+#
+
+MODULE = mm.so
+MDBTGT = kvm
+
+MODSRCS = mm.c
+
+include ../../../../Makefile.cmd
+include ../../../../Makefile.cmd.64
+include ../../Makefile.sparcv9
+include ../../../Makefile.module
diff --git a/usr/src/pkg/manifests/developer-debug-mdb.mf b/usr/src/pkg/manifests/developer-debug-mdb.mf
index 926e63d69f..b7d6bb4630 100644
--- a/usr/src/pkg/manifests/developer-debug-mdb.mf
+++ b/usr/src/pkg/manifests/developer-debug-mdb.mf
@@ -225,6 +225,7 @@ file path=usr/lib/mdb/kvm/$(ARCH64)/mac.so group=sys mode=0555
 file path=usr/lib/mdb/kvm/$(ARCH64)/md.so group=sys mode=0555
 $(i386_ONLY)file path=usr/lib/mdb/kvm/$(ARCH64)/mdb_kb.so group=sys mode=0555
 file path=usr/lib/mdb/kvm/$(ARCH64)/mdb_ks.so group=sys mode=0555
+file path=usr/lib/mdb/kvm/$(ARCH64)/mm.so group=sys mode=0555
 file path=usr/lib/mdb/kvm/$(ARCH64)/mpt.so group=sys mode=0555
 file path=usr/lib/mdb/kvm/$(ARCH64)/mpt_sas.so group=sys mode=0555
 file path=usr/lib/mdb/kvm/$(ARCH64)/mr_sas.so group=sys mode=0555
@@ -263,6 +264,7 @@ $(i386_ONLY)file path=usr/lib/mdb/kvm/mac.so group=sys mode=0555
 $(i386_ONLY)file path=usr/lib/mdb/kvm/md.so group=sys mode=0555
 $(i386_ONLY)file path=usr/lib/mdb/kvm/mdb_kb.so group=sys mode=0555
 $(i386_ONLY)file path=usr/lib/mdb/kvm/mdb_ks.so group=sys mode=0555
+$(i386_ONLY)file path=usr/lib/mdb/kvm/mm.so group=sys mode=0555
 $(i386_ONLY)file path=usr/lib/mdb/kvm/mpt.so group=sys mode=0555
 $(i386_ONLY)file path=usr/lib/mdb/kvm/mpt_sas.so group=sys mode=0555
 $(i386_ONLY)file path=usr/lib/mdb/kvm/mr_sas.so group=sys mode=0555
diff --git a/usr/src/uts/common/io/mem.c b/usr/src/uts/common/io/mem.c
index 7a703c5750..c0173a6f42 100644
--- a/usr/src/uts/common/io/mem.c
+++ b/usr/src/uts/common/io/mem.c
@@ -24,6 +24,10 @@
  */
 
 /*
+ * Copyright (c) 2015, Joyent, Inc.  All rights reserved.
+ */
+
+/*
  * Memory special file
  */
 
@@ -95,6 +99,49 @@ static int mm_kstat_snapshot(kstat_t *ksp, void *buf, int rw);
 
 static int mm_read_mem_name(intptr_t data, mem_name_t *mem_name);
 
+#define	MM_KMEMLOG_NENTRIES	64
+
+static int mm_kmemlogent;
+static mm_logentry_t mm_kmemlog[MM_KMEMLOG_NENTRIES];
+
+/*
+ * On kmem/allmem writes, we log information that might be useful in the event
+ * that a write is errant (that is, due to operator error) and induces a later
+ * problem.  Note that (in particular) in the event of such operator-induced
+ * corruption, a search over the kernel address space for the corrupted
+ * address will yield the ring buffer entry that recorded the write.  And
+ * should it seem baroque or otherwise unnecessary, yes, we need this kind of
+ * auditing facility and yes, we learned that the hard way: disturbingly,
+ * there exist recommendations for "tuning" the system that involve writing to
+ * kernel memory addresses via the kernel debugger, and -- as we discovered --
+ * these can easily be applied incorrectly or unsafely, yielding an entirely
+ * undebuggable "can't happen" kind of panic.
+ */
+static void
+mm_logkmem(struct uio *uio)
+{
+	mm_logentry_t *ent;
+	proc_t *p = curthread->t_procp;
+
+	mutex_enter(&mm_lock);
+
+	ent = &mm_kmemlog[mm_kmemlogent++];
+
+	if (mm_kmemlogent == MM_KMEMLOG_NENTRIES)
+		mm_kmemlogent = 0;
+
+	ent->mle_vaddr = (uintptr_t)uio->uio_loffset;
+	ent->mle_len = uio->uio_resid;
+	gethrestime(&ent->mle_hrestime);
+	ent->mle_hrtime = gethrtime();
+	ent->mle_pid = p->p_pidp->pid_id;
+
+	(void) strncpy(ent->mle_psargs,
+	    p->p_user.u_psargs, sizeof (ent->mle_psargs));
+
+	mutex_exit(&mm_lock);
+}
+
 /*ARGSUSED1*/
 static int
 mm_attach(dev_info_t *devi, ddi_attach_cmd_t cmd)
@@ -354,6 +401,9 @@ mmrw(dev_t dev, struct uio *uio, enum uio_rw rw, cred_t *cred)
 			if ((error = plat_mem_do_mmio(uio, rw)) != ENOTSUP)
 				break;
 
+			if (rw == UIO_WRITE)
+				mm_logkmem(uio);
+
 			/*
 			 * If vaddr does not map a valid page, as_pagelock()
 			 * will return failure. Hence we can't check the
diff --git a/usr/src/uts/common/sys/mem.h b/usr/src/uts/common/sys/mem.h
index d0ac1223fc..5101b3ec17 100644
--- a/usr/src/uts/common/sys/mem.h
+++ b/usr/src/uts/common/sys/mem.h
@@ -23,6 +23,10 @@
  * Use is subject to license terms.
  */
 
+/*
+ * Copyright (c) 2015, Joyent, Inc.  All rights reserved.
+ */
+
 #ifndef _SYS_MEM_H
 #define	_SYS_MEM_H
 
@@ -128,6 +132,15 @@ extern pfn_t impl_obmem_pfnum(pfn_t);
 
 extern int plat_mem_do_mmio(struct uio *, enum uio_rw);
 
+typedef struct mm_logentry {
+	uintptr_t	mle_vaddr;	/* vaddr being written to */
+	size_t		mle_len;	/* length of write */
+	timespec_t	mle_hrestime;	/* hrestime at time of write */
+	hrtime_t	mle_hrtime;	/* hrtime at time of write */
+	pid_t		mle_pid;	/* pid of writing process */
+	char		mle_psargs[80];	/* psargs of writing process */
+} mm_logentry_t;
+
 #endif	/* _KERNEL */
 
 #ifdef	__cplusplus