fail(9) support:

Add support for kernel fault injection using KFAIL_POINT_* macros and
fail_point_* infrastructure. Add example fail point in vfs_bio.c to
simulate VM buf pressure.

Approved by:        dfr (mentor)

7 files changed, 1088 insertions, 3 deletions

diff --git a/share/man/man9/Makefile b/share/man/man9/Makefile
index 4901fb4..45b5438 100644
--- a/share/man/man9/Makefile
+++ b/share/man/man9/Makefile
@@ -99,6 +99,7 @@ MAN=	accept_filter.9 \
 	DRIVER_MODULE.9 \
 	EVENTHANDLER.9 \
 	extattr.9 \
+	fail.9 \
 	fetch.9 \
 	firmware.9 \
 	g_access.9 \
diff --git a/share/man/man9/fail.9 b/share/man/man9/fail.9
new file mode 100644
index 0000000..648f9d9
--- /dev/null
+++ b/share/man/man9/fail.9
@@ -0,0 +1,197 @@
+.\"
+.\" Copyright (c) 2009 Isilon Inc http://www.isilon.com/
+.\"
+.\" Redistribution and use in source and binary forms, with or without
+.\" modification, are permitted provided that the following conditions
+.\" are met:
+.\" 1. Redistributions of source code must retain the above copyright
+.\"    notice(s), this list of conditions and the following disclaimer as
+.\"    the first lines of this file unmodified other than the possible
+.\"    addition of one or more copyright notices.
+.\" 2. Redistributions in binary form must reproduce the above copyright
+.\"    notice(s), this list of conditions and the following disclaimer in the
+.\"    documentation and/or other materials provided with the distribution.
+.\"
+.\" THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``AS IS'' AND ANY
+.\" EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+.\" WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+.\" DISCLAIMED.  IN NO EVENT SHALL THE COPYRIGHT HOLDER(S) BE LIABLE FOR ANY
+.\" DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+.\" (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+.\" SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+.\" CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
+.\" DAMAGE.
+.\"
+.\" $FreeBSD$
+.\"
+.Dd May 10, 2009
+.Dt FAIL 9
+.Os
+.Sh NAME
+.Nm KFAIL_POINT_CODE ,
+.Nm KFAIL_POINT_RETURN ,
+.Nm KFAIL_POINT_RETURN_VOID ,
+.Nm KFAIL_POINT_ERROR ,
+.Nm KFAIL_POINT_GOTO ,
+.Nm fail_point ,
+.Nm DEBUG_FP
+.
+.Nd fail points
+.Sh SYNOPSIS
+.In sys/fail.h
+.Fn KFAIL_POINT_CODE "parent" "name" "code"
+.Fn KFAIL_POINT_RETURN "parent" "name"
+.Fn KFAIL_POINT_RETURN_VOID "parent" "name"
+.Fn KFAIL_POINT_ERROR "parent" "name" "error_var"
+.Fn KFAIL_POINT_GOTO "parent" "name" "error_var" "label"
+.Sh DESCRIPTION
+Fail points are used to add code points where errors may be injected
+in a user controlled fashion. Fail points provide a convenient wrapper
+around user provided error injection code, providing a
+.Xr sysctl 9 MIB , and a parser for that MIB that describes how the error
+injection code should fire.
+.Pp
+The base fail point macro is
+.Fn KFAIL_POINT_CODE
+where
+.Fa parent
+is a sysctl tree (frequently
+.Sy DEBUG_FP
+for kernel fail points, but various subsystems may wish to provide
+their own fail point trees), and
+.Fa name
+is the name of the MIB in that tree, and
+.Fa code
+is the error injection code. The
+.Fa code
+argument does not require braces, but it is considered good style to
+use braces for any multi-line code arguments. Inside the
+.Fa code
+argument, the evaluation of
+.Sy RETURN_VALUE
+is derived from the
+.Fn return
+value set in the sysctl MIB. See
+.Sx SYSCTL SETTINGS
+below.
+.Pp
+The remaining
+.Fn KFAIL_POINT_*
+macros are wrappers around common error injection paths:
+.Bl -tag -width 8
+.It Fn KFAIL_POINT_RETURN parent name
+is the equivalent of
+.Sy KFAIL_POINT_CODE(..., return RETURN_VALUE)
+.It Fn KFAIL_POINT_RETURN_VOID parent name
+is the equivalent of
+.Sy KFAIL_POINT_CODE(..., return)
+.It Fn KFAIL_POINT_ERROR parent name error_var
+is the equivalent of
+.Sy KFAIL_POINT_CODE(..., error_var = RETURN_VALUE)
+.It Fn KFAIL_POINT_GOTO parent name error_var label
+is the equivalent of
+.Sy KFAIL_POINT_CODE(...,
+  { error_var = RETURN_VALUE; goto label;})
+.El
+.Pp
+.Sh SYSCTL VARIABLES
+The
+.Fn KFAIL_POINT_*
+macros add sysctl MIBs where specified. Many base kernel MIBs can be
+found in the
+.Sy debug.fail_point
+tree (referenced in code by
+.Sy DEBUG_FP
+).
+.Pp
+The sysctl setting recognizes the following grammar:
+.Pp
+  <fail_point> ::
+      <term> ( "->" <term> )*
+.Pp
+  <term> ::
+      ( (<float> "%") | (<integer> "*" ) )*
+      <type>
+      [ "(" <integer> ")" ]
+.Pp
+  <float> ::
+      <integer> [ "." <integer> ] |
+      "." <integer>
+.Pp
+  <type> ::
+      "off" | "return" | "sleep" | "panic" | "break" | "print"
+.Pp
+The <type>
+argument specifies which action to take:
+.Bl -tag -width ".Dv return"
+.It Sy off
+Take no action (does not trigger fail point code)
+.It Sy return
+Trigger fail point code with specified argument
+.It Sy sleep
+Sleep the specified number of milliseconds
+.It Sy panic
+Panic
+.It Sy break
+Break into the debugger.
+.It Sy print
+Print that the fail point executed
+.El
+.Pp
+The <float>% and <integer>* modifiers prior to <type> control when
+<type> is executed. The <float>% form (e.g. "1.2%") can be used to
+specify a probability that <type> will execute. The <integer>* form
+(e.g. "5*") can be used to specify the number of times <type> should
+be executed before this <term> is disabled. Only the last probability
+and the last count are used if multiple are specified, i.e. "1.2%2%"
+is the same as "2%". When both a probability and a count are
+specified, the probability is evaluated before the count, i.e. "2%5*"
+means "2% of the time, but only execute it 5 times total".
+.Pp
+The operator -> can be used to express cascading terms. If you specify
+<term1>-><term2>, it means that if <term1> doesn't 'execute', <term2>
+is evaluated. For the purpose of this operator, the return() and
+print() operators are the only types that cascade. A return() term
+only cascades if the code executes, and a print() term only cascades
+when passed a non-zero argument.
+.Pp
+.Sh EXAMPLES
+.Bl -tag
+.It Sy sysctl debug.fail_point.foobar="2.1%return(5)"
+21/1000ths of the time, execute
+.Fa code
+with RETURN_VALUE set to 5.
+.It Sy sysctl debug.fail_point.foobar="2%return(5)->5%return(22)"
+2/100th of the time, execute
+.Fa code
+with RETURN_VALUE set to 5. If that doesn't happen, 5% of the time
+execute
+.Fa code
+with RETURN_VALUE set to 22.
+.It Sy sysctl debug.fail_point.foobar="5*return(5)->0.1%return(22)"
+For 5 times, return 5. After that, 1/1000ths of the time, return 22.
+.It Sy sysctl debug.fail_point.foobar="0.1%5*return(5)"
+Return 5 for 1 in 1000 executions, but only execute 5 times total.
+.It Sy sysctl debug.fail_point.foobar="1%*sleep(50)"
+1/100ths of the time, sleep 50ms.
+.El
+.Pp
+.Sh CAVEATS
+It's easy to shoot yourself in the foot by setting fail points too
+aggressively or setting too many in combination. For example, forcing
+.Fn malloc
+to fail consistently is potentially harmful to uptime.
+.Pp
+The
+.Fn sleep
+sysctl setting may not be appropriate in all situations. Currently,
+.Fn fail_point_eval
+does not verify whether the context is appropriate for calling
+.Fn msleep .
+.Pp
+.Sh AUTHORS
+.An -nosplit
+This manual page was written by
+.An Zach Loafman Aq zml@FreeBSD.org .
diff --git a/sys/conf/files b/sys/conf/files
index 166bed8..1e958b8 100644
--- a/sys/conf/files
+++ b/sys/conf/files
@@ -1860,6 +1860,7 @@ kern/kern_environment.c		standard
 kern/kern_event.c		standard
 kern/kern_exec.c		standard
 kern/kern_exit.c		standard
+kern/kern_fail.c		standard
 kern/kern_fork.c		standard
 kern/kern_idle.c		standard
 kern/kern_intr.c		standard
diff --git a/sys/kern/kern_fail.c b/sys/kern/kern_fail.c
new file mode 100644
index 0000000..06dfc38
--- /dev/null
+++ b/sys/kern/kern_fail.c
@@ -0,0 +1,575 @@
+/*-
+ * Copyright (c) 2009 Isilon Inc http://www.isilon.com/
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+/**
+ * @file
+ *
+ * fail(9) Facility.
+ *
+ * @ingroup failpoint_private
+ */
+/**
+ * @defgroup failpoint fail(9) Facility
+ *
+ * Failpoints allow for injecting fake errors into running code on the fly,
+ * without modifying code or recompiling with flags.  Failpoints are always
+ * present, and are very efficient when disabled.  Failpoints are described
+ * in man fail(9).
+ */
+/**
+ * @defgroup failpoint_private Private fail(9) Implementation functions
+ *
+ * Private implementations for the actual failpoint code.
+ *
+ * @ingroup failpoint
+ */
+/**
+ * @addtogroup failpoint_private
+ * @{
+ */
+
+#include <sys/cdefs.h>
+__FBSDID("$FreeBSD$");
+
+#include <sys/errno.h>
+#include <sys/fail.h>
+#include <sys/kernel.h>
+#include <sys/libkern.h>
+#include <sys/lock.h>
+#include <sys/malloc.h>
+#include <sys/mutex.h>
+#include <sys/sbuf.h>
+
+#include <machine/stdarg.h>
+
+#ifdef ILOG_DEFINE_FOR_FILE
+ILOG_DEFINE_FOR_FILE(L_ISI_FAIL_POINT, L_ILOG, fail_point);
+#endif
+
+MALLOC_DEFINE(M_FAIL_POINT, "Fail Points", "fail points system");
+#define fp_free(ptr) free(ptr, M_FAIL_POINT)
+#define fp_malloc(size, flags) malloc((size), M_FAIL_POINT, (flags))
+
+static struct mtx g_fp_mtx;
+MTX_SYSINIT(g_fp_mtx, &g_fp_mtx, "fail point mtx", MTX_DEF);
+#define FP_LOCK()	mtx_lock(&g_fp_mtx)
+#define FP_UNLOCK()	mtx_unlock(&g_fp_mtx)
+
+static inline void
+fail_point_sleep(struct fail_point *fp, struct fail_point_entry *ent,
+    int msecs, enum fail_point_return_code *pret)
+{
+	/* convert from millisecs to ticks, rounding up */
+	int timo = ((msecs * hz) + 999) / 1000;
+
+	if (timo) {
+		if (fp->fp_sleep_fn == NULL) {
+			msleep(fp, &g_fp_mtx, PWAIT, "failpt", timo);
+		} else {
+			timeout(fp->fp_sleep_fn, fp->fp_sleep_arg, timo);
+			*pret = FAIL_POINT_RC_QUEUED;
+		}
+	}
+}
+
+
+/**
+ * Defines stating the equivalent of probablilty one (100%)
+ */
+enum {
+	PROB_MAX = 1000000,	/* probability between zero and this number */
+	PROB_DIGITS = 6,        /* number of zero's in above number */
+};
+
+static const char *fail_type_strings[] = {
+	"off",
+	"panic",
+	"return",
+	"break",
+	"print",
+	"sleep",
+};
+
+static char *parse_fail_point(struct fail_point_entries *, char *);
+static char *parse_term(struct fail_point_entries *, char *);
+static char *parse_number(int *out_units, int *out_decimal, char *);
+static char *parse_type(struct fail_point_entry *, char *);
+static void free_entry(struct fail_point_entries *, struct fail_point_entry *);
+static void clear_entries(struct fail_point_entries *);
+
+/**
+ * Initialize a fail_point.  The name is formed in a printf-like fashion
+ * from "fmt" and subsequent arguments.  This function is generally used
+ * for custom failpoints located at odd places in the sysctl tree, and is
+ * not explicitly needed for standard in-line-declared failpoints.
+ *
+ * @ingroup failpoint
+ */
+void
+fail_point_init(struct fail_point *fp, const char *fmt, ...)
+{
+	va_list ap;
+	char *name;
+	int n;
+
+	TAILQ_INIT(&fp->fp_entries);
+	fp->fp_flags = 0;
+
+	/* Figure out the size of the name. */
+	va_start(ap, fmt);
+	n = vsnprintf(NULL, 0, fmt, ap);
+	va_end(ap);
+
+	/* Allocate the name and fill it in. */
+	name = fp_malloc(n + 1, M_WAITOK);
+	if (name != NULL) {
+		va_start(ap, fmt);
+		vsnprintf(name, n + 1, fmt, ap);
+		va_end(ap);
+	}
+	fp->fp_name = name;
+	fp->fp_flags |= FAIL_POINT_DYNAMIC_NAME;
+	fp->fp_sleep_fn = NULL;
+	fp->fp_sleep_arg = NULL;
+}
+
+/**
+ * Free the resources held by a fail_point.
+ *
+ * @ingroup failpoint
+ */
+void
+fail_point_destroy(struct fail_point *fp)
+{
+	struct fail_point_entry *ent;
+
+	if (fp->fp_flags & FAIL_POINT_DYNAMIC_NAME && fp->fp_name != NULL) {
+		fp_free((void *)(intptr_t)fp->fp_name);
+		fp->fp_name = NULL;
+	}
+	fp->fp_flags = 0;
+
+	while (!TAILQ_EMPTY(&fp->fp_entries)) {
+		ent = TAILQ_FIRST(&fp->fp_entries);
+		TAILQ_REMOVE(&fp->fp_entries, ent, fe_entries);
+		fp_free(ent);
+	}
+}
+
+/**
+ * This does the real work of evaluating a fail point. If the fail point tells
+ * us to return a value, this function returns 1 and fills in 'return_value'
+ * (return_value is allowed to be null). If the fail point tells us to panic,
+ * we never return. Otherwise we just return 0 after doing some work, which
+ * means "keep going".
+ */
+enum fail_point_return_code
+fail_point_eval_nontrivial(struct fail_point *fp, int *return_value)
+{
+	enum fail_point_return_code ret = FAIL_POINT_RC_CONTINUE;
+	struct fail_point_entry *ent, *next;
+	int msecs;
+
+	FP_LOCK();
+
+	ent = TAILQ_FIRST(&fp->fp_entries);
+	while (ent) {
+		int cont = 0; /* don't continue by default */
+		next = TAILQ_NEXT(ent, fe_entries);
+
+		if (ent->fe_prob < PROB_MAX &&
+		    ent->fe_prob < random() % PROB_MAX) {
+			cont = 1;
+			goto loop_end;
+		}
+
+		switch (ent->fe_type) {
+		case FAIL_POINT_PANIC:
+			panic("fail point %s panicking", fp->fp_name);
+			/* NOTREACHED */
+
+		case FAIL_POINT_RETURN:
+			if (return_value)
+				*return_value = ent->fe_arg;
+			ret = FAIL_POINT_RC_RETURN;
+			break;
+
+		case FAIL_POINT_BREAK:
+			printf("fail point %s breaking to debugger\n", fp->fp_name);
+			breakpoint();
+			break;
+
+		case FAIL_POINT_PRINT:
+			printf("fail point %s executing\n", fp->fp_name);
+			cont = ent->fe_arg;
+			break;
+
+		case FAIL_POINT_SLEEP:
+			/*
+			 * Free the entry now if necessary, since
+			 * we're about to drop the mutex and sleep.
+			 */
+			msecs = ent->fe_arg;
+			if (ent->fe_count > 0 && --ent->fe_count == 0) {
+				free_entry(&fp->fp_entries, ent);
+				ent = NULL;
+			}
+
+			if (msecs)
+				fail_point_sleep(fp, ent, msecs, &ret);
+			break;
+
+		default:
+			break;
+		}
+
+		if (ent && ent->fe_count > 0 && --ent->fe_count == 0)
+			free_entry(&fp->fp_entries, ent);
+
+loop_end:
+		if (cont)
+			ent = next;
+		else
+			break;
+	}
+
+	/* Get rid of "off"s at the end. */
+	while ((ent = TAILQ_LAST(&fp->fp_entries, fail_point_entries)) &&
+	       ent->fe_type == FAIL_POINT_OFF)
+		free_entry(&fp->fp_entries, ent);
+
+	FP_UNLOCK();
+
+	return ret;
+}
+
+/**
+ * Translate internal fail_point structure into human-readable text.
+ */
+static void
+fail_point_get(struct fail_point *fp, struct sbuf *sb)
+{
+	struct fail_point_entry *ent;
+
+	FP_LOCK();
+
+	TAILQ_FOREACH(ent, &fp->fp_entries, fe_entries) {
+		if (ent->fe_prob < PROB_MAX) {
+			int decimal = ent->fe_prob % (PROB_MAX / 100);
+			int units = ent->fe_prob / (PROB_MAX / 100);
+			sbuf_printf(sb, "%d", units);
+			if (decimal) {
+				int digits = PROB_DIGITS - 2;
+				while (!(decimal % 10)) {
+					digits--;
+					decimal /= 10;
+				}
+				sbuf_printf(sb, ".%0*d", digits, decimal);
+			}
+			sbuf_printf(sb, "%%");
+		}
+		if (ent->fe_count > 0)
+			sbuf_printf(sb, "%d*", ent->fe_count);
+		sbuf_printf(sb, "%s", fail_type_strings[ent->fe_type]);
+		if (ent->fe_arg)
+			sbuf_printf(sb, "(%d)", ent->fe_arg);
+		if (TAILQ_NEXT(ent, fe_entries))
+			sbuf_printf(sb, "->");
+	}
+	if (TAILQ_EMPTY(&fp->fp_entries))
+		sbuf_printf(sb, "off");
+
+	FP_UNLOCK();
+}
+
+/**
+ * Set an internal fail_point structure from a human-readable failpoint string
+ * in a lock-safe manner.
+ */
+static int
+fail_point_set(struct fail_point *fp, char *buf)
+{
+	int error = 0;
+	struct fail_point_entry *ent, *ent_next;
+	struct fail_point_entries new_entries;
+
+	/* Parse new entries. */
+	TAILQ_INIT(&new_entries);
+	if (!parse_fail_point(&new_entries, buf)) {
+	        clear_entries(&new_entries);
+		error = EINVAL;
+		goto end;
+	}
+
+	FP_LOCK();
+
+	/* Move new entries in. */
+	TAILQ_SWAP(&fp->fp_entries, &new_entries, fail_point_entry, fe_entries);
+	clear_entries(&new_entries);
+
+	/* Get rid of useless zero probability entries. */
+	TAILQ_FOREACH_SAFE(ent, &fp->fp_entries, fe_entries, ent_next) {
+		if (ent->fe_prob == 0)
+			free_entry(&fp->fp_entries, ent);
+	}
+
+	/* Get rid of "off"s at the end. */
+	while ((ent = TAILQ_LAST(&fp->fp_entries, fail_point_entries)) &&
+		ent->fe_type == FAIL_POINT_OFF)
+		free_entry(&fp->fp_entries, ent);
+
+	FP_UNLOCK();
+
+ end:
+#ifdef IWARNING
+	if (error)
+		IWARNING("Failed to set %s (%s) to %s",
+		    fp->fp_name, fp->fp_location, buf);
+	else
+		INOTICE("Set %s (%s) to %s",
+		    fp->fp_name, fp->fp_location, buf);
+#endif /* IWARNING */
+
+	return error;
+}
+
+#define MAX_FAIL_POINT_BUF	1023
+
+/**
+ * Handle kernel failpoint set/get.
+ */
+int
+fail_point_sysctl(SYSCTL_HANDLER_ARGS)
+{
+	struct fail_point *fp = arg1;
+	char *buf = NULL;
+	struct sbuf sb;
+	int error;
+
+	/* Retrieving */
+	sbuf_new(&sb, NULL, 128, SBUF_AUTOEXTEND);
+	fail_point_get(fp, &sb);
+	sbuf_trim(&sb);
+	sbuf_finish(&sb);
+	error = SYSCTL_OUT(req, sbuf_data(&sb), sbuf_len(&sb));
+	sbuf_delete(&sb);
+
+	/* Setting */
+	if (!error && req->newptr) {
+		if (req->newlen > MAX_FAIL_POINT_BUF) {
+			error = EINVAL;
+			goto out;
+		}
+
+		buf = fp_malloc(req->newlen + 1, M_WAITOK);
+
+		error = SYSCTL_IN(req, buf, req->newlen);
+		if (error)
+			goto out;
+		buf[req->newlen] = '\0';
+
+		error = fail_point_set(fp, buf);
+        }
+
+out:
+	if (buf)
+		fp_free(buf);
+	return error;
+}
+
+/**
+ * Internal helper function to translate a human-readable failpoint string
+ * into a internally-parsable fail_point structure.
+ */
+static char *
+parse_fail_point(struct fail_point_entries *ents, char *p)
+{
+	/*  <fail_point> ::
+	 *      <term> ( "->" <term> )*
+	 */
+	if (!(p = parse_term(ents, p)))
+		return 0;
+	while (*p)
+		if (p[0] != '-' || p[1] != '>' || !(p = parse_term(ents, p+2)))
+			return 0;
+	return p;
+}
+
+/**
+ * Internal helper function to parse an individual term from a failpoint.
+ */
+static char *
+parse_term(struct fail_point_entries *ents, char *p)
+{
+	struct fail_point_entry *ent;
+
+	ent = fp_malloc(sizeof *ent, M_WAITOK | M_ZERO);
+	ent->fe_prob = PROB_MAX;
+	TAILQ_INSERT_TAIL(ents, ent, fe_entries);
+
+	/*
+	 * <term> ::
+	 *     ( (<float> "%") | (<integer> "*" ) )*
+	 *     <type>
+	 *     [ "(" <integer> ")" ]
+	 */
+
+	/* ( (<float> "%") | (<integer> "*" ) )* */
+	while (('0' <= *p && *p <= '9') || *p == '.') {
+		int units, decimal;
+
+		if (!(p = parse_number(&units, &decimal, p)))
+			return 0;
+
+		if (*p == '%') {
+			if (units > 100) /* prevent overflow early */
+				units = 100;
+			ent->fe_prob = units * (PROB_MAX / 100) + decimal;
+			if (ent->fe_prob > PROB_MAX)
+				ent->fe_prob = PROB_MAX;
+
+		} else if (*p == '*') {
+			if (!units || decimal)
+				return 0;
+			ent->fe_count = units;;
+
+		} else {
+			return 0;
+		}
+
+		p++;
+	}
+
+	/* <type> */
+	if (!(p = parse_type(ent, p)))
+		return 0;
+	if (*p == '\0')
+		return p;
+
+	/* [ "(" <integer> ")" ] */
+	if (*p != '(')
+		return p;
+	p++;
+	if (('0' <= *p && *p <= '9') || *p == '-')
+		ent->fe_arg = strtol(p, &p, 0);
+	else
+		return 0;
+	if (*p++ != ')')
+		return 0;
+
+	return p;
+}
+
+/**
+ * Internal helper function to parse a numeric for a failpoint term.
+ */
+static char *
+parse_number(int *out_units, int *out_decimal, char *p)
+{
+	char *old_p;
+
+	/*
+	 *  <number> ::
+	 *      <integer> [ "." <integer> ] |
+	 *      "." <integer>
+	 */
+
+	/* whole part */
+	old_p = p;
+	*out_units = strtol(p, &p, 10);;
+	if (p == old_p && *p != '.')
+		return 0;
+
+	/* fractional part */
+	*out_decimal = 0;
+	if (*p == '.') {
+		int digits = 0;
+		p++;
+		while ('0' <= *p && *p <= '9') {
+			int digit = *p - '0';
+			if (digits < PROB_DIGITS - 2)
+				*out_decimal = *out_decimal * 10 + digit;
+			else if (digits == PROB_DIGITS - 2 && digit >= 5)
+				(*out_decimal)++;
+			digits++;
+			p++;
+		}
+		if (!digits) /* need at least one digit after '.' */
+			return 0;
+		while (digits++ < PROB_DIGITS - 2) /* add implicit zeros */
+			*out_decimal *= 10;
+	}
+
+	return p; /* success */
+}
+
+/**
+ * Internal helper function to parse an individual type for a failpoint term.
+ */
+static char *
+parse_type(struct fail_point_entry *ent, char *beg)
+{
+	enum fail_point_t type;
+	char *end = beg;
+	while ('a' <= *end && *end <= 'z')
+		end++;
+	if (beg == end)
+		return 0;
+	for (type = FAIL_POINT_OFF; type != FAIL_POINT_INVALID; type++) {
+		const char *p = fail_type_strings[type];
+		const char *q = beg;
+		while (q < end && *p++ == *q++);
+		if (q == end && *p == '\0') {
+			ent->fe_type = type;
+			return end;
+		}
+	}
+	return 0;
+}
+
+/**
+ * Internal helper function to free an individual failpoint term.
+ */
+static void
+free_entry(struct fail_point_entries *ents, struct fail_point_entry *ent)
+{
+	TAILQ_REMOVE(ents, ent, fe_entries);
+	fp_free(ent);
+}
+
+/**
+ * Internal helper function to clear out all failpoint terms for a single
+ * failpoint.
+ */
+static void
+clear_entries(struct fail_point_entries *ents)
+{
+	struct fail_point_entry *ent, *ent_next;
+	TAILQ_FOREACH_SAFE(ent, ents, fe_entries, ent_next)
+		fp_free(ent);
+	TAILQ_INIT(ents);
+}
+
+/* The fail point sysctl tree. */
+SYSCTL_NODE(_debug, OID_AUTO, fail_point, CTLFLAG_RW, 0, "fail points");
diff --git a/sys/kern/vfs_bio.c b/sys/kern/vfs_bio.c
index a3c3e97..0817c7f 100644
--- a/sys/kern/vfs_bio.c
+++ b/sys/kern/vfs_bio.c
@@ -48,6 +48,7 @@ __FBSDID("$FreeBSD$");
 #include <sys/buf.h>
 #include <sys/devicestat.h>
 #include <sys/eventhandler.h>
+#include <sys/fail.h>
 #include <sys/limits.h>
 #include <sys/lock.h>
 #include <sys/malloc.h>
@@ -1167,6 +1168,15 @@ buf_dirty_count_severe(void)
 	return(numdirtybuffers >= hidirtybuffers);
 }
 
+static __noinline int
+buf_vm_page_count_severe(void)
+{
+
+	KFAIL_POINT_CODE(DEBUG_FP, buf_pressure, return 1);
+
+	return vm_page_count_severe();
+}
+
 /*
  *	brelse:
  *
@@ -1233,7 +1243,7 @@ brelse(struct buf *bp)
 	 */
 	if (bp->b_flags & B_DELWRI)
 		bp->b_flags &= ~B_RELBUF;
-	else if (vm_page_count_severe()) {
+	else if (buf_vm_page_count_severe()) {
 		/*
 		 * The locking of the BO_LOCK is not necessary since
 		 * BKGRDINPROG cannot be set while we hold the buf
@@ -1502,7 +1512,7 @@ bqrelse(struct buf *bp)
 		 * lock, it can only be cleared if it is already
 		 * pending.
 		 */
-		if (!vm_page_count_severe() || (bp->b_vflags & BV_BKGRDINPROG)) {
+		if (!buf_vm_page_count_severe() || (bp->b_vflags & BV_BKGRDINPROG)) {
 			bp->b_qindex = QUEUE_CLEAN;
 			TAILQ_INSERT_TAIL(&bufqueues[QUEUE_CLEAN], bp,
 			    b_freelist);
@@ -1571,7 +1581,7 @@ vfs_vmio_release(struct buf *bp)
 				vm_page_free(m);
 			} else if (bp->b_flags & B_DIRECT) {
 				vm_page_try_to_free(m);
-			} else if (vm_page_count_severe()) {
+			} else if (buf_vm_page_count_severe()) {
 				vm_page_try_to_cache(m);
 			}
 		}
diff --git a/sys/sys/fail.h b/sys/sys/fail.h
new file mode 100644
index 0000000..3bc4b06
--- /dev/null
+++ b/sys/sys/fail.h
@@ -0,0 +1,260 @@
+/*-
+ * Copyright (c) 2009 Isilon Inc http://www.isilon.com/
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ *
+ * $FreeBSD$
+ */
+/**
+ * @file
+ *
+ * Main header for failpoint facility.
+ */
+#ifndef _SYS_FAIL_H_
+#define _SYS_FAIL_H_
+
+#include <sys/types.h>
+
+#include <sys/linker_set.h>
+#include <sys/param.h>
+#include <sys/queue.h>
+#include <sys/sysctl.h>
+
+
+/**
+ * Failpoint types.
+ * Don't change these without changing fail_type_strings in fail.c.
+ * @ingroup failpoint_private
+ */
+enum fail_point_t {
+	FAIL_POINT_OFF,		/**< don't fail */
+	FAIL_POINT_PANIC,	/**< panic */
+	FAIL_POINT_RETURN,	/**< return an errorcode */
+	FAIL_POINT_BREAK,	/**< break into the debugger */
+	FAIL_POINT_PRINT,	/**< print a message */
+	FAIL_POINT_SLEEP,	/**< sleep for some msecs */
+	FAIL_POINT_INVALID,	/**< placeholder */
+};
+
+/**
+ * Failpoint return codes, used internally.
+ * @ingroup failpoint_private
+ */
+enum fail_point_return_code {
+	FAIL_POINT_RC_CONTINUE = 0,	/**< Continue with normal execution */
+	FAIL_POINT_RC_RETURN,		/**< FP evaluated to 'return' */
+	FAIL_POINT_RC_QUEUED,		/**< sleep_fn will be called */
+};
+
+TAILQ_HEAD(fail_point_entries, fail_point_entry);
+/**
+ * Internal failpoint structure, tracking all the current details of the
+ * failpoint.  This structure is the core component shared between the
+ * failure-injection code and the user-interface.
+ * @ingroup failpoint_private
+ */
+struct fail_point {
+	const char *fp_name;		/**< name of fail point */
+	const char *fp_location;	/**< file:line of fail point */
+	struct fail_point_entries fp_entries;	/**< list of entries */
+	int fp_flags;
+	void (*fp_sleep_fn)(void *);	/**< Function to call at end of
+					 * sleep for sleep failpoints */
+	void *fp_sleep_arg;		/**< Arg for sleep_fn */
+};
+
+#define	FAIL_POINT_DYNAMIC_NAME	0x01	/**< Must free name on destroy */
+
+/**
+ * Internal structure tracking a single term of a complete failpoint.
+ * @ingroup failpoint_private
+ */
+struct fail_point_entry {
+	enum fail_point_t fe_type;	/**< type of entry */
+	int		fe_arg;		/**< argument to type (e.g. return value) */
+	int		fe_prob;	/**< likelihood of firing in millionths */
+	int		fe_count;	/**< number of times to fire, 0 means always */
+
+	TAILQ_ENTRY(fail_point_entry) fe_entries; /**< next entry in fail point */
+};
+
+/* Private failpoint eval function -- use fail_point_eval() instead. */
+enum fail_point_return_code fail_point_eval_nontrivial(struct fail_point *,
+	int *ret);
+
+/**
+ * @addtogroup failpoint
+ * @{
+ */
+/*
+ * Initialize a fail-point.  The name is formed in printf-like fashion
+ * from "fmt" and the subsequent arguments.
+ * Pair with fail_point_destroy().
+ */
+void fail_point_init(struct fail_point *, const char *fmt, ...)
+    __printflike(2, 3);
+
+/**
+ * Set the sleep function for a fail point
+ * If sleep_fn is specified, then FAIL_POINT_SLEEP will result in a
+ * (*fp->sleep_fn)(fp->sleep_arg) call by the timer thread.  Otherwise,
+ * if sleep_fn is NULL (default), then FAIL_POINT_SLEEP will result in the
+ * fail_point_eval() call sleeping.
+ */
+static __inline void
+fail_point_sleep_set_func(struct fail_point *fp, void (*sleep_fn)(void *))
+{
+	fp->fp_sleep_fn = sleep_fn;
+}
+
+/**
+ * Set the argument for the sleep function for a fail point
+ */
+static __inline void
+fail_point_sleep_set_arg(struct fail_point *fp, void *sleep_arg)
+{
+	fp->fp_sleep_arg = sleep_arg;
+}
+
+/**
+ * Free the resources used by a fail-point.  Pair with fail_point_init().
+ */
+void fail_point_destroy(struct fail_point *);
+
+/**
+ * Evaluate a failpoint.
+ */
+static __inline enum fail_point_return_code
+fail_point_eval(struct fail_point *fp, int *ret)
+{
+	if (TAILQ_EMPTY(&fp->fp_entries)) {
+		return (FAIL_POINT_RC_CONTINUE);
+	}
+	return (fail_point_eval_nontrivial(fp, ret));
+}
+
+/* Declare a fail_point and its sysctl in a function. */
+#define _FAIL_POINT_NAME(name) _fail_point_##name
+#define _STRINGIFY_HELPER(x) #x
+#define _STRINGIFY(x) _STRINGIFY_HELPER(x)
+#define _FAIL_POINT_LOCATION() __FILE__ ":" _STRINGIFY(__LINE__)
+
+/**
+ * Instantiate a failpoint which returns "value" from the function when triggered.
+ * @param parent  The parent sysctl under which to locate the sysctl
+ * @param name    The name of the failpoint in the sysctl tree (and printouts)
+ * @return        Instantly returns the return("value") specified in the
+ *                failpoint, if triggered.
+ */
+#define KFAIL_POINT_RETURN(parent, name) \
+	KFAIL_POINT_CODE(parent, name, return RETURN_VALUE)
+
+/**
+ * Instantiate a failpoint which returns (void) from the function when triggered.
+ * @param parent  The parent sysctl under which to locate the sysctl
+ * @param name    The name of the failpoint in the sysctl tree (and printouts)
+ * @return        Instantly returns void, if triggered in the failpoint.
+ */
+#define KFAIL_POINT_RETURN_VOID(parent, name) \
+	KFAIL_POINT_CODE(parent, name, return)
+
+/**
+ * Instantiate a failpoint which sets an error when triggered.
+ * @param parent     The parent sysctl under which to locate the sysctl
+ * @param name       The name of the failpoint in the sysctl tree (and printouts)
+ * @param error_var  A variable to set to the failpoint's specified
+ *                   return-value when triggered
+ */
+#define KFAIL_POINT_ERROR(parent, name, error_var) \
+	KFAIL_POINT_CODE(parent, name, (error_var) = RETURN_VALUE)
+
+/**
+ * Instantiate a failpoint which sets an error and then goes to a
+ * specified label in the function when triggered.
+ * @param parent     The parent sysctl under which to locate the sysctl
+ * @param name       The name of the failpoint in the sysctl tree (and printouts)
+ * @param error_var  A variable to set to the failpoint's specified
+ *                   return-value when triggered
+ * @param label      The location to goto when triggered.
+ */
+#define KFAIL_POINT_GOTO(parent, name, error_var, label) \
+	KFAIL_POINT_CODE(parent, name, (error_var) = RETURN_VALUE; goto label)
+
+/**
+ * Instantiate a failpoint which runs arbitrary code when triggered.
+ * @param parent     The parent sysctl under which to locate the sysctl
+ * @param name       The name of the failpoint in the sysctl tree (and printouts)
+ * @param code       The arbitrary code to run when triggered.  Can reference
+ *                   "RETURN_VALUE" if desired to extract the specified user
+ *                   return-value when triggered
+ */
+#define KFAIL_POINT_CODE(parent, name, code)				\
+	KFAIL_POINT_START(parent, name) {				\
+		code;							\
+	} FAIL_POINT_END
+
+/**
+ * @}
+ * (end group failpoint)
+ */
+
+/**
+ * Internal macro to implement above #defines -- should not be used directly.
+ * @ingroup failpoint_private
+ */
+#define KFAIL_POINT_START(parent, name)					\
+	do {								\
+		int RETURN_VALUE;					\
+		static struct fail_point _FAIL_POINT_NAME(name) = {	\
+			#name,						\
+			_FAIL_POINT_LOCATION(),				\
+			TAILQ_HEAD_INITIALIZER(				\
+				_FAIL_POINT_NAME(name).fp_entries),	\
+			0,						\
+			NULL, NULL,					\
+		};							\
+		SYSCTL_OID(parent, OID_AUTO, name,			\
+			CTLTYPE_STRING | CTLFLAG_RW,			\
+			&_FAIL_POINT_NAME(name), 0, fail_point_sysctl,	\
+			"A", "");					\
+									\
+		if (__predict_false(					\
+		    fail_point_eval(&_FAIL_POINT_NAME(name),		\
+		    &RETURN_VALUE))) {
+
+/**
+ * Internal macro to implement above #defines -- should not be used directly.
+ * @ingroup failpoint_private
+ */
+#define FAIL_POINT_END							\
+		}							\
+	} while (0)
+
+#ifdef _KERNEL
+int fail_point_sysctl(SYSCTL_HANDLER_ARGS);
+
+/* The fail point sysctl tree. */
+SYSCTL_DECL(_debug_fail_point);
+#endif
+#define DEBUG_FP _debug_fail_point
+
+#endif /* _SYS_FAIL_H_ */
diff --git a/sys/sys/queue.h b/sys/sys/queue.h
index 3ac69e4..d9f339a 100644
--- a/sys/sys/queue.h
+++ b/sys/sys/queue.h
@@ -309,6 +309,20 @@ struct {								\
 		(head)->stqh_last = &STAILQ_NEXT((elm), field);		\
 } while (0)
 
+#define STAILQ_SWAP(head1, head2, type) do {				\
+	struct type *swap_first = STAILQ_FIRST(head1);			\
+	struct type **swap_last = (head1)->stqh_last;			\
+	STAILQ_FIRST(head1) = STAILQ_FIRST(head2);			\
+	(head1)->stqh_last = (head2)->stqh_last;			\
+	STAILQ_FIRST(head2) = swap_first;				\
+	(head2)->stqh_last = swap_last;					\
+	if (STAILQ_EMPTY(head1))					\
+		(head1)->stqh_last = &STAILQ_FIRST(head1);		\
+	if (STAILQ_EMPTY(head2))					\
+		(head2)->stqh_last = &STAILQ_FIRST(head2);		\
+} while (0)
+
+
 /*
  * List declarations.
  */
@@ -411,6 +425,16 @@ struct {								\
 	TRASHIT((elm)->field.le_prev);					\
 } while (0)
 
+#define LIST_SWAP(head1, head2, type, field) do {			\
+	struct type *swap_tmp = LIST_FIRST((head1));			\
+	LIST_FIRST((head1)) = LIST_FIRST((head2));			\
+	LIST_FIRST((head2)) = swap_tmp;					\
+	if ((swap_tmp = LIST_FIRST((head1))) != NULL)			\
+		swap_tmp->field.le_prev = &LIST_FIRST((head1));		\
+	if ((swap_tmp = LIST_FIRST((head2))) != NULL)			\
+		swap_tmp->field.le_prev = &LIST_FIRST((head2));		\
+} while (0)
+
 /*
  * Tail queue declarations.
  */
@@ -578,4 +602,21 @@ struct {								\
 	QMD_TRACE_ELEM(&(elm)->field);					\
 } while (0)
 
+#define TAILQ_SWAP(head1, head2, type, field) do {			\
+	struct type *swap_first = (head1)->tqh_first;			\
+	struct type **swap_last = (head1)->tqh_last;			\
+	(head1)->tqh_first = (head2)->tqh_first;			\
+	(head1)->tqh_last = (head2)->tqh_last;				\
+	(head2)->tqh_first = swap_first;				\
+	(head2)->tqh_last = swap_last;					\
+	if ((swap_first = (head1)->tqh_first) != NULL)			\
+		swap_first->field.tqe_prev = &(head1)->tqh_first;	\
+	else								\
+		(head1)->tqh_last = &(head1)->tqh_first;		\
+	if ((swap_first = (head2)->tqh_first) != NULL)			\
+		swap_first->field.tqe_prev = &(head2)->tqh_first;	\
+	else								\
+		(head2)->tqh_last = &(head2)->tqh_first;		\
+} while (0)
+
 #endif /* !_SYS_QUEUE_H_ */