As Miklos reported and suggested:
This pattern repeats two times in trace_uprobe.c and in
kernel/events/core.c as well:
ret = kern_path(filename, LOOKUP_FOLLOW, &path);
if (ret)
goto fail_address_parse;
inode = igrab(d_inode(path.dentry));
path_put(&path);
And it's wrong. You can only hold a reference to the inode if you
have an active ref to the superblock as well (which is normally
through path.mnt) or holding s_umount.
This way unmounting the containing filesystem while the tracepoint is
active will give you the "VFS: Busy inodes after unmount..." message
and a crash when the inode is finally put.
Solution: store path instead of inode.
This patch fixes two instances in trace_uprobe.c.
Fixes: f3f096cfedf8 ("tracing: Provide trace events interface for uprobes")
Fixes: 33ea4b24277b ("perf/core: Implement the 'perf_uprobe' PMU")
Cc: Steven Rostedt <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Howard McLauchlan <[email protected]>
Cc: Josef Bacik <[email protected]>
Cc: Srikar Dronamraju <[email protected]>
Reported-by: Miklos Szeredi <[email protected]>
Signed-off-by: Song Liu <[email protected]>
---
kernel/trace/trace_uprobe.c | 49 +++++++++++++++------------------------------
1 file changed, 16 insertions(+), 33 deletions(-)
diff --git a/kernel/trace/trace_uprobe.c b/kernel/trace/trace_uprobe.c
index 34fd0e0..d9ee522c 100644
--- a/kernel/trace/trace_uprobe.c
+++ b/kernel/trace/trace_uprobe.c
@@ -55,7 +55,7 @@ struct trace_uprobe {
struct list_head list;
struct trace_uprobe_filter filter;
struct uprobe_consumer consumer;
- struct inode *inode;
+ struct path path;
char *filename;
unsigned long offset;
unsigned long nhit;
@@ -289,7 +289,7 @@ static void free_trace_uprobe(struct trace_uprobe *tu)
for (i = 0; i < tu->tp.nr_args; i++)
traceprobe_free_probe_arg(&tu->tp.args[i]);
- iput(tu->inode);
+ path_put(&tu->path);
kfree(tu->tp.call.class->system);
kfree(tu->tp.call.name);
kfree(tu->filename);
@@ -363,7 +363,6 @@ static int register_trace_uprobe(struct trace_uprobe *tu)
static int create_trace_uprobe(int argc, char **argv)
{
struct trace_uprobe *tu;
- struct inode *inode;
char *arg, *event, *group, *filename;
char buf[MAX_EVENT_NAME_LEN];
struct path path;
@@ -371,7 +370,6 @@ static int create_trace_uprobe(int argc, char **argv)
bool is_delete, is_return;
int i, ret;
- inode = NULL;
ret = 0;
is_delete = false;
is_return = false;
@@ -437,24 +435,14 @@ static int create_trace_uprobe(int argc, char **argv)
}
/* Find the last occurrence, in case the path contains ':' too. */
arg = strrchr(argv[1], ':');
- if (!arg) {
- ret = -EINVAL;
- goto fail_address_parse;
- }
+ if (!arg)
+ return -EINVAL;
*arg++ = '\0';
filename = argv[1];
ret = kern_path(filename, LOOKUP_FOLLOW, &path);
if (ret)
- goto fail_address_parse;
-
- inode = igrab(d_real_inode(path.dentry));
- path_put(&path);
-
- if (!inode || !S_ISREG(inode->i_mode)) {
- ret = -EINVAL;
- goto fail_address_parse;
- }
+ return ret;
ret = kstrtoul(arg, 0, &offset);
if (ret)
@@ -490,7 +478,7 @@ static int create_trace_uprobe(int argc, char **argv)
goto fail_address_parse;
}
tu->offset = offset;
- tu->inode = inode;
+ tu->path = path;
tu->filename = kstrdup(filename, GFP_KERNEL);
if (!tu->filename) {
@@ -558,7 +546,7 @@ static int create_trace_uprobe(int argc, char **argv)
return ret;
fail_address_parse:
- iput(inode);
+ path_put(&path);
pr_info("Failed to parse address or file.\n");
@@ -922,7 +910,8 @@ probe_event_enable(struct trace_uprobe *tu, struct trace_event_file *file,
goto err_flags;
tu->consumer.filter = filter;
- ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
+ ret = uprobe_register(d_inode(tu->path.dentry), tu->offset,
+ &tu->consumer);
if (ret)
goto err_buffer;
@@ -966,7 +955,7 @@ probe_event_disable(struct trace_uprobe *tu, struct trace_event_file *file)
WARN_ON(!uprobe_filter_is_empty(&tu->filter));
- uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
+ uprobe_unregister(d_inode(tu->path.dentry), tu->offset, &tu->consumer);
tu->tp.flags &= file ? ~TP_FLAG_TRACE : ~TP_FLAG_PROFILE;
uprobe_buffer_disable();
@@ -1041,7 +1030,8 @@ static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
write_unlock(&tu->filter.rwlock);
if (!done)
- return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
+ return uprobe_apply(d_inode(tu->path.dentry), tu->offset,
+ &tu->consumer, false);
return 0;
}
@@ -1073,7 +1063,8 @@ static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
err = 0;
if (!done) {
- err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
+ err = uprobe_apply(d_inode(tu->path.dentry),
+ tu->offset, &tu->consumer, true);
if (err)
uprobe_perf_close(tu, event);
}
@@ -1337,7 +1328,6 @@ struct trace_event_call *
create_local_trace_uprobe(char *name, unsigned long offs, bool is_return)
{
struct trace_uprobe *tu;
- struct inode *inode;
struct path path;
int ret;
@@ -1345,14 +1335,6 @@ create_local_trace_uprobe(char *name, unsigned long offs, bool is_return)
if (ret)
return ERR_PTR(ret);
- inode = igrab(d_inode(path.dentry));
- path_put(&path);
-
- if (!inode || !S_ISREG(inode->i_mode)) {
- iput(inode);
- return ERR_PTR(-EINVAL);
- }
-
/*
* local trace_kprobes are not added to probe_list, so they are never
* searched in find_trace_kprobe(). Therefore, there is no concern of
@@ -1364,11 +1346,12 @@ create_local_trace_uprobe(char *name, unsigned long offs, bool is_return)
if (IS_ERR(tu)) {
pr_info("Failed to allocate trace_uprobe.(%d)\n",
(int)PTR_ERR(tu));
+ path_put(&path);
return ERR_CAST(tu);
}
tu->offset = offs;
- tu->inode = inode;
+ tu->path = path;
tu->filename = kstrdup(name, GFP_KERNEL);
init_trace_event_call(tu, &tu->tp.call);
--
2.9.5
You forgot to add Miklos again.
Miklos, have any comments on this patch? Or can you give a
"Reviewed-by"?
-- Steve
On Wed, 18 Apr 2018 10:40:14 -0700
Song Liu <[email protected]> wrote:
> As Miklos reported and suggested:
>
> This pattern repeats two times in trace_uprobe.c and in
> kernel/events/core.c as well:
>
> ret = kern_path(filename, LOOKUP_FOLLOW, &path);
> if (ret)
> goto fail_address_parse;
>
> inode = igrab(d_inode(path.dentry));
> path_put(&path);
>
> And it's wrong. You can only hold a reference to the inode if you
> have an active ref to the superblock as well (which is normally
> through path.mnt) or holding s_umount.
>
> This way unmounting the containing filesystem while the tracepoint is
> active will give you the "VFS: Busy inodes after unmount..." message
> and a crash when the inode is finally put.
>
> Solution: store path instead of inode.
>
> This patch fixes two instances in trace_uprobe.c.
>
> Fixes: f3f096cfedf8 ("tracing: Provide trace events interface for uprobes")
> Fixes: 33ea4b24277b ("perf/core: Implement the 'perf_uprobe' PMU")
> Cc: Steven Rostedt <[email protected]>
> Cc: Ingo Molnar <[email protected]>
> Cc: Howard McLauchlan <[email protected]>
> Cc: Josef Bacik <[email protected]>
> Cc: Srikar Dronamraju <[email protected]>
> Reported-by: Miklos Szeredi <[email protected]>
> Signed-off-by: Song Liu <[email protected]>
> ---
> kernel/trace/trace_uprobe.c | 49 +++++++++++++++------------------------------
> 1 file changed, 16 insertions(+), 33 deletions(-)
>
> diff --git a/kernel/trace/trace_uprobe.c b/kernel/trace/trace_uprobe.c
> index 34fd0e0..d9ee522c 100644
> --- a/kernel/trace/trace_uprobe.c
> +++ b/kernel/trace/trace_uprobe.c
> @@ -55,7 +55,7 @@ struct trace_uprobe {
> struct list_head list;
> struct trace_uprobe_filter filter;
> struct uprobe_consumer consumer;
> - struct inode *inode;
> + struct path path;
> char *filename;
> unsigned long offset;
> unsigned long nhit;
> @@ -289,7 +289,7 @@ static void free_trace_uprobe(struct trace_uprobe *tu)
> for (i = 0; i < tu->tp.nr_args; i++)
> traceprobe_free_probe_arg(&tu->tp.args[i]);
>
> - iput(tu->inode);
> + path_put(&tu->path);
> kfree(tu->tp.call.class->system);
> kfree(tu->tp.call.name);
> kfree(tu->filename);
> @@ -363,7 +363,6 @@ static int register_trace_uprobe(struct trace_uprobe *tu)
> static int create_trace_uprobe(int argc, char **argv)
> {
> struct trace_uprobe *tu;
> - struct inode *inode;
> char *arg, *event, *group, *filename;
> char buf[MAX_EVENT_NAME_LEN];
> struct path path;
> @@ -371,7 +370,6 @@ static int create_trace_uprobe(int argc, char **argv)
> bool is_delete, is_return;
> int i, ret;
>
> - inode = NULL;
> ret = 0;
> is_delete = false;
> is_return = false;
> @@ -437,24 +435,14 @@ static int create_trace_uprobe(int argc, char **argv)
> }
> /* Find the last occurrence, in case the path contains ':' too. */
> arg = strrchr(argv[1], ':');
> - if (!arg) {
> - ret = -EINVAL;
> - goto fail_address_parse;
> - }
> + if (!arg)
> + return -EINVAL;
>
> *arg++ = '\0';
> filename = argv[1];
> ret = kern_path(filename, LOOKUP_FOLLOW, &path);
> if (ret)
> - goto fail_address_parse;
> -
> - inode = igrab(d_real_inode(path.dentry));
> - path_put(&path);
> -
> - if (!inode || !S_ISREG(inode->i_mode)) {
> - ret = -EINVAL;
> - goto fail_address_parse;
> - }
> + return ret;
>
> ret = kstrtoul(arg, 0, &offset);
> if (ret)
> @@ -490,7 +478,7 @@ static int create_trace_uprobe(int argc, char **argv)
> goto fail_address_parse;
> }
> tu->offset = offset;
> - tu->inode = inode;
> + tu->path = path;
> tu->filename = kstrdup(filename, GFP_KERNEL);
>
> if (!tu->filename) {
> @@ -558,7 +546,7 @@ static int create_trace_uprobe(int argc, char **argv)
> return ret;
>
> fail_address_parse:
> - iput(inode);
> + path_put(&path);
>
> pr_info("Failed to parse address or file.\n");
>
> @@ -922,7 +910,8 @@ probe_event_enable(struct trace_uprobe *tu, struct trace_event_file *file,
> goto err_flags;
>
> tu->consumer.filter = filter;
> - ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
> + ret = uprobe_register(d_inode(tu->path.dentry), tu->offset,
> + &tu->consumer);
> if (ret)
> goto err_buffer;
>
> @@ -966,7 +955,7 @@ probe_event_disable(struct trace_uprobe *tu, struct trace_event_file *file)
>
> WARN_ON(!uprobe_filter_is_empty(&tu->filter));
>
> - uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
> + uprobe_unregister(d_inode(tu->path.dentry), tu->offset, &tu->consumer);
> tu->tp.flags &= file ? ~TP_FLAG_TRACE : ~TP_FLAG_PROFILE;
>
> uprobe_buffer_disable();
> @@ -1041,7 +1030,8 @@ static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
> write_unlock(&tu->filter.rwlock);
>
> if (!done)
> - return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
> + return uprobe_apply(d_inode(tu->path.dentry), tu->offset,
> + &tu->consumer, false);
>
> return 0;
> }
> @@ -1073,7 +1063,8 @@ static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
>
> err = 0;
> if (!done) {
> - err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
> + err = uprobe_apply(d_inode(tu->path.dentry),
> + tu->offset, &tu->consumer, true);
> if (err)
> uprobe_perf_close(tu, event);
> }
> @@ -1337,7 +1328,6 @@ struct trace_event_call *
> create_local_trace_uprobe(char *name, unsigned long offs, bool is_return)
> {
> struct trace_uprobe *tu;
> - struct inode *inode;
> struct path path;
> int ret;
>
> @@ -1345,14 +1335,6 @@ create_local_trace_uprobe(char *name, unsigned long offs, bool is_return)
> if (ret)
> return ERR_PTR(ret);
>
> - inode = igrab(d_inode(path.dentry));
> - path_put(&path);
> -
> - if (!inode || !S_ISREG(inode->i_mode)) {
> - iput(inode);
> - return ERR_PTR(-EINVAL);
> - }
> -
> /*
> * local trace_kprobes are not added to probe_list, so they are never
> * searched in find_trace_kprobe(). Therefore, there is no concern of
> @@ -1364,11 +1346,12 @@ create_local_trace_uprobe(char *name, unsigned long offs, bool is_return)
> if (IS_ERR(tu)) {
> pr_info("Failed to allocate trace_uprobe.(%d)\n",
> (int)PTR_ERR(tu));
> + path_put(&path);
> return ERR_CAST(tu);
> }
>
> tu->offset = offs;
> - tu->inode = inode;
> + tu->path = path;
> tu->filename = kstrdup(name, GFP_KERNEL);
> init_trace_event_call(tu, &tu->tp.call);
>
> On Apr 18, 2018, at 10:47 AM, Steven Rostedt <[email protected]> wrote:
>
> You forgot to add Miklos again.
>
I am really sorry! I was expecting "Reported-by" to add Miklos.
Song
> Miklos, have any comments on this patch? Or can you give a
> "Reviewed-by"?
>
> -- Steve
>
>
> On Wed, 18 Apr 2018 10:40:14 -0700
> Song Liu <[email protected]> wrote:
>
>> As Miklos reported and suggested:
>>
>> This pattern repeats two times in trace_uprobe.c and in
>> kernel/events/core.c as well:
>>
>> ret = kern_path(filename, LOOKUP_FOLLOW, &path);
>> if (ret)
>> goto fail_address_parse;
>>
>> inode = igrab(d_inode(path.dentry));
>> path_put(&path);
>>
>> And it's wrong. You can only hold a reference to the inode if you
>> have an active ref to the superblock as well (which is normally
>> through path.mnt) or holding s_umount.
>>
>> This way unmounting the containing filesystem while the tracepoint is
>> active will give you the "VFS: Busy inodes after unmount..." message
>> and a crash when the inode is finally put.
>>
>> Solution: store path instead of inode.
>>
>> This patch fixes two instances in trace_uprobe.c.
>>
>> Fixes: f3f096cfedf8 ("tracing: Provide trace events interface for uprobes")
>> Fixes: 33ea4b24277b ("perf/core: Implement the 'perf_uprobe' PMU")
>> Cc: Steven Rostedt <[email protected]>
>> Cc: Ingo Molnar <[email protected]>
>> Cc: Howard McLauchlan <[email protected]>
>> Cc: Josef Bacik <[email protected]>
>> Cc: Srikar Dronamraju <[email protected]>
>> Reported-by: Miklos Szeredi <[email protected]>
>> Signed-off-by: Song Liu <[email protected]>
>> ---
>> kernel/trace/trace_uprobe.c | 49 +++++++++++++++------------------------------
>> 1 file changed, 16 insertions(+), 33 deletions(-)
>>
>> diff --git a/kernel/trace/trace_uprobe.c b/kernel/trace/trace_uprobe.c
>> index 34fd0e0..d9ee522c 100644
>> --- a/kernel/trace/trace_uprobe.c
>> +++ b/kernel/trace/trace_uprobe.c
>> @@ -55,7 +55,7 @@ struct trace_uprobe {
>> struct list_head list;
>> struct trace_uprobe_filter filter;
>> struct uprobe_consumer consumer;
>> - struct inode *inode;
>> + struct path path;
>> char *filename;
>> unsigned long offset;
>> unsigned long nhit;
>> @@ -289,7 +289,7 @@ static void free_trace_uprobe(struct trace_uprobe *tu)
>> for (i = 0; i < tu->tp.nr_args; i++)
>> traceprobe_free_probe_arg(&tu->tp.args[i]);
>>
>> - iput(tu->inode);
>> + path_put(&tu->path);
>> kfree(tu->tp.call.class->system);
>> kfree(tu->tp.call.name);
>> kfree(tu->filename);
>> @@ -363,7 +363,6 @@ static int register_trace_uprobe(struct trace_uprobe *tu)
>> static int create_trace_uprobe(int argc, char **argv)
>> {
>> struct trace_uprobe *tu;
>> - struct inode *inode;
>> char *arg, *event, *group, *filename;
>> char buf[MAX_EVENT_NAME_LEN];
>> struct path path;
>> @@ -371,7 +370,6 @@ static int create_trace_uprobe(int argc, char **argv)
>> bool is_delete, is_return;
>> int i, ret;
>>
>> - inode = NULL;
>> ret = 0;
>> is_delete = false;
>> is_return = false;
>> @@ -437,24 +435,14 @@ static int create_trace_uprobe(int argc, char **argv)
>> }
>> /* Find the last occurrence, in case the path contains ':' too. */
>> arg = strrchr(argv[1], ':');
>> - if (!arg) {
>> - ret = -EINVAL;
>> - goto fail_address_parse;
>> - }
>> + if (!arg)
>> + return -EINVAL;
>>
>> *arg++ = '\0';
>> filename = argv[1];
>> ret = kern_path(filename, LOOKUP_FOLLOW, &path);
>> if (ret)
>> - goto fail_address_parse;
>> -
>> - inode = igrab(d_real_inode(path.dentry));
>> - path_put(&path);
>> -
>> - if (!inode || !S_ISREG(inode->i_mode)) {
>> - ret = -EINVAL;
>> - goto fail_address_parse;
>> - }
>> + return ret;
>>
>> ret = kstrtoul(arg, 0, &offset);
>> if (ret)
>> @@ -490,7 +478,7 @@ static int create_trace_uprobe(int argc, char **argv)
>> goto fail_address_parse;
>> }
>> tu->offset = offset;
>> - tu->inode = inode;
>> + tu->path = path;
>> tu->filename = kstrdup(filename, GFP_KERNEL);
>>
>> if (!tu->filename) {
>> @@ -558,7 +546,7 @@ static int create_trace_uprobe(int argc, char **argv)
>> return ret;
>>
>> fail_address_parse:
>> - iput(inode);
>> + path_put(&path);
>>
>> pr_info("Failed to parse address or file.\n");
>>
>> @@ -922,7 +910,8 @@ probe_event_enable(struct trace_uprobe *tu, struct trace_event_file *file,
>> goto err_flags;
>>
>> tu->consumer.filter = filter;
>> - ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
>> + ret = uprobe_register(d_inode(tu->path.dentry), tu->offset,
>> + &tu->consumer);
>> if (ret)
>> goto err_buffer;
>>
>> @@ -966,7 +955,7 @@ probe_event_disable(struct trace_uprobe *tu, struct trace_event_file *file)
>>
>> WARN_ON(!uprobe_filter_is_empty(&tu->filter));
>>
>> - uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
>> + uprobe_unregister(d_inode(tu->path.dentry), tu->offset, &tu->consumer);
>> tu->tp.flags &= file ? ~TP_FLAG_TRACE : ~TP_FLAG_PROFILE;
>>
>> uprobe_buffer_disable();
>> @@ -1041,7 +1030,8 @@ static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
>> write_unlock(&tu->filter.rwlock);
>>
>> if (!done)
>> - return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
>> + return uprobe_apply(d_inode(tu->path.dentry), tu->offset,
>> + &tu->consumer, false);
>>
>> return 0;
>> }
>> @@ -1073,7 +1063,8 @@ static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
>>
>> err = 0;
>> if (!done) {
>> - err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
>> + err = uprobe_apply(d_inode(tu->path.dentry),
>> + tu->offset, &tu->consumer, true);
>> if (err)
>> uprobe_perf_close(tu, event);
>> }
>> @@ -1337,7 +1328,6 @@ struct trace_event_call *
>> create_local_trace_uprobe(char *name, unsigned long offs, bool is_return)
>> {
>> struct trace_uprobe *tu;
>> - struct inode *inode;
>> struct path path;
>> int ret;
>>
>> @@ -1345,14 +1335,6 @@ create_local_trace_uprobe(char *name, unsigned long offs, bool is_return)
>> if (ret)
>> return ERR_PTR(ret);
>>
>> - inode = igrab(d_inode(path.dentry));
>> - path_put(&path);
>> -
>> - if (!inode || !S_ISREG(inode->i_mode)) {
>> - iput(inode);
>> - return ERR_PTR(-EINVAL);
>> - }
>> -
>> /*
>> * local trace_kprobes are not added to probe_list, so they are never
>> * searched in find_trace_kprobe(). Therefore, there is no concern of
>> @@ -1364,11 +1346,12 @@ create_local_trace_uprobe(char *name, unsigned long offs, bool is_return)
>> if (IS_ERR(tu)) {
>> pr_info("Failed to allocate trace_uprobe.(%d)\n",
>> (int)PTR_ERR(tu));
>> + path_put(&path);
>> return ERR_CAST(tu);
>> }
>>
>> tu->offset = offs;
>> - tu->inode = inode;
>> + tu->path = path;
>> tu->filename = kstrdup(name, GFP_KERNEL);
>> init_trace_event_call(tu, &tu->tp.call);
>>
>
On Wed, Apr 18, 2018 at 7:40 PM, Song Liu <[email protected]> wrote:
> As Miklos reported and suggested:
>
> This pattern repeats two times in trace_uprobe.c and in
> kernel/events/core.c as well:
>
> ret = kern_path(filename, LOOKUP_FOLLOW, &path);
> if (ret)
> goto fail_address_parse;
>
> inode = igrab(d_inode(path.dentry));
> path_put(&path);
>
> And it's wrong. You can only hold a reference to the inode if you
> have an active ref to the superblock as well (which is normally
> through path.mnt) or holding s_umount.
>
> This way unmounting the containing filesystem while the tracepoint is
> active will give you the "VFS: Busy inodes after unmount..." message
> and a crash when the inode is finally put.
>
> Solution: store path instead of inode.
>
> This patch fixes two instances in trace_uprobe.c.
>
> Fixes: f3f096cfedf8 ("tracing: Provide trace events interface for uprobes")
> Fixes: 33ea4b24277b ("perf/core: Implement the 'perf_uprobe' PMU")
> Cc: Steven Rostedt <[email protected]>
> Cc: Ingo Molnar <[email protected]>
> Cc: Howard McLauchlan <[email protected]>
> Cc: Josef Bacik <[email protected]>
> Cc: Srikar Dronamraju <[email protected]>
> Reported-by: Miklos Szeredi <[email protected]>
> Signed-off-by: Song Liu <[email protected]>
> ---
> kernel/trace/trace_uprobe.c | 49 +++++++++++++++------------------------------
> 1 file changed, 16 insertions(+), 33 deletions(-)
>
> diff --git a/kernel/trace/trace_uprobe.c b/kernel/trace/trace_uprobe.c
> index 34fd0e0..d9ee522c 100644
> --- a/kernel/trace/trace_uprobe.c
> +++ b/kernel/trace/trace_uprobe.c
> @@ -55,7 +55,7 @@ struct trace_uprobe {
> struct list_head list;
> struct trace_uprobe_filter filter;
> struct uprobe_consumer consumer;
> - struct inode *inode;
> + struct path path;
> char *filename;
> unsigned long offset;
> unsigned long nhit;
> @@ -289,7 +289,7 @@ static void free_trace_uprobe(struct trace_uprobe *tu)
> for (i = 0; i < tu->tp.nr_args; i++)
> traceprobe_free_probe_arg(&tu->tp.args[i]);
>
> - iput(tu->inode);
> + path_put(&tu->path);
> kfree(tu->tp.call.class->system);
> kfree(tu->tp.call.name);
> kfree(tu->filename);
> @@ -363,7 +363,6 @@ static int register_trace_uprobe(struct trace_uprobe *tu)
> static int create_trace_uprobe(int argc, char **argv)
> {
> struct trace_uprobe *tu;
> - struct inode *inode;
> char *arg, *event, *group, *filename;
> char buf[MAX_EVENT_NAME_LEN];
> struct path path;
> @@ -371,7 +370,6 @@ static int create_trace_uprobe(int argc, char **argv)
> bool is_delete, is_return;
> int i, ret;
>
> - inode = NULL;
> ret = 0;
> is_delete = false;
> is_return = false;
> @@ -437,24 +435,14 @@ static int create_trace_uprobe(int argc, char **argv)
> }
> /* Find the last occurrence, in case the path contains ':' too. */
> arg = strrchr(argv[1], ':');
> - if (!arg) {
> - ret = -EINVAL;
> - goto fail_address_parse;
> - }
> + if (!arg)
> + return -EINVAL;
>
> *arg++ = '\0';
> filename = argv[1];
> ret = kern_path(filename, LOOKUP_FOLLOW, &path);
> if (ret)
> - goto fail_address_parse;
> -
> - inode = igrab(d_real_inode(path.dentry));
> - path_put(&path);
> -
> - if (!inode || !S_ISREG(inode->i_mode)) {
Where has the S_ISREG check gone?
> - ret = -EINVAL;
> - goto fail_address_parse;
> - }
> + return ret;
>
> ret = kstrtoul(arg, 0, &offset);
> if (ret)
> @@ -490,7 +478,7 @@ static int create_trace_uprobe(int argc, char **argv)
> goto fail_address_parse;
> }
> tu->offset = offset;
> - tu->inode = inode;
> + tu->path = path;
> tu->filename = kstrdup(filename, GFP_KERNEL);
>
> if (!tu->filename) {
> @@ -558,7 +546,7 @@ static int create_trace_uprobe(int argc, char **argv)
> return ret;
>
> fail_address_parse:
> - iput(inode);
> + path_put(&path);
>
> pr_info("Failed to parse address or file.\n");
>
> @@ -922,7 +910,8 @@ probe_event_enable(struct trace_uprobe *tu, struct trace_event_file *file,
> goto err_flags;
>
> tu->consumer.filter = filter;
> - ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
> + ret = uprobe_register(d_inode(tu->path.dentry), tu->offset,
> + &tu->consumer);
> if (ret)
> goto err_buffer;
>
> @@ -966,7 +955,7 @@ probe_event_disable(struct trace_uprobe *tu, struct trace_event_file *file)
>
> WARN_ON(!uprobe_filter_is_empty(&tu->filter));
>
> - uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
> + uprobe_unregister(d_inode(tu->path.dentry), tu->offset, &tu->consumer);
> tu->tp.flags &= file ? ~TP_FLAG_TRACE : ~TP_FLAG_PROFILE;
>
> uprobe_buffer_disable();
> @@ -1041,7 +1030,8 @@ static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
> write_unlock(&tu->filter.rwlock);
>
> if (!done)
> - return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
> + return uprobe_apply(d_inode(tu->path.dentry), tu->offset,
> + &tu->consumer, false);
>
> return 0;
> }
> @@ -1073,7 +1063,8 @@ static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
>
> err = 0;
> if (!done) {
> - err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
> + err = uprobe_apply(d_inode(tu->path.dentry),
> + tu->offset, &tu->consumer, true);
> if (err)
> uprobe_perf_close(tu, event);
> }
> @@ -1337,7 +1328,6 @@ struct trace_event_call *
> create_local_trace_uprobe(char *name, unsigned long offs, bool is_return)
> {
> struct trace_uprobe *tu;
> - struct inode *inode;
> struct path path;
> int ret;
>
> @@ -1345,14 +1335,6 @@ create_local_trace_uprobe(char *name, unsigned long offs, bool is_return)
> if (ret)
> return ERR_PTR(ret);
>
> - inode = igrab(d_inode(path.dentry));
> - path_put(&path);
> -
> - if (!inode || !S_ISREG(inode->i_mode)) {
And here, again.
Otherwise ACK.
Also please create a separate patch that removes igrab/iput calls from
kernel/events/uprobes.c and adds a comment to the effect that the
caller is required to keep the inode (and the containing mount)
referenced.
Thanks,
Miklos
On Thu, Apr 19, 2018 at 10:58 AM, Miklos Szeredi <[email protected]> wrote:
> On Wed, Apr 18, 2018 at 7:40 PM, Song Liu <[email protected]> wrote:
>> As Miklos reported and suggested:
>>
>> This pattern repeats two times in trace_uprobe.c and in
>> kernel/events/core.c as well:
>>
>> ret = kern_path(filename, LOOKUP_FOLLOW, &path);
>> if (ret)
>> goto fail_address_parse;
>>
>> inode = igrab(d_inode(path.dentry));
>> path_put(&path);
>>
>> And it's wrong. You can only hold a reference to the inode if you
>> have an active ref to the superblock as well (which is normally
>> through path.mnt) or holding s_umount.
>>
>> This way unmounting the containing filesystem while the tracepoint is
>> active will give you the "VFS: Busy inodes after unmount..." message
>> and a crash when the inode is finally put.
>>
>> Solution: store path instead of inode.
>>
>> This patch fixes two instances in trace_uprobe.c.
>>
>> Fixes: f3f096cfedf8 ("tracing: Provide trace events interface for uprobes")
>> Fixes: 33ea4b24277b ("perf/core: Implement the 'perf_uprobe' PMU")
>> Cc: Steven Rostedt <[email protected]>
>> Cc: Ingo Molnar <[email protected]>
>> Cc: Howard McLauchlan <[email protected]>
>> Cc: Josef Bacik <[email protected]>
>> Cc: Srikar Dronamraju <[email protected]>
>> Reported-by: Miklos Szeredi <[email protected]>
>> Signed-off-by: Song Liu <[email protected]>
>> ---
>> kernel/trace/trace_uprobe.c | 49 +++++++++++++++------------------------------
>> 1 file changed, 16 insertions(+), 33 deletions(-)
>>
>> diff --git a/kernel/trace/trace_uprobe.c b/kernel/trace/trace_uprobe.c
>> index 34fd0e0..d9ee522c 100644
>> --- a/kernel/trace/trace_uprobe.c
>> +++ b/kernel/trace/trace_uprobe.c
>> @@ -55,7 +55,7 @@ struct trace_uprobe {
>> struct list_head list;
>> struct trace_uprobe_filter filter;
>> struct uprobe_consumer consumer;
>> - struct inode *inode;
>> + struct path path;
>> char *filename;
>> unsigned long offset;
>> unsigned long nhit;
>> @@ -289,7 +289,7 @@ static void free_trace_uprobe(struct trace_uprobe *tu)
>> for (i = 0; i < tu->tp.nr_args; i++)
>> traceprobe_free_probe_arg(&tu->tp.args[i]);
>>
>> - iput(tu->inode);
>> + path_put(&tu->path);
>> kfree(tu->tp.call.class->system);
>> kfree(tu->tp.call.name);
>> kfree(tu->filename);
>> @@ -363,7 +363,6 @@ static int register_trace_uprobe(struct trace_uprobe *tu)
>> static int create_trace_uprobe(int argc, char **argv)
>> {
>> struct trace_uprobe *tu;
>> - struct inode *inode;
>> char *arg, *event, *group, *filename;
>> char buf[MAX_EVENT_NAME_LEN];
>> struct path path;
>> @@ -371,7 +370,6 @@ static int create_trace_uprobe(int argc, char **argv)
>> bool is_delete, is_return;
>> int i, ret;
>>
>> - inode = NULL;
>> ret = 0;
>> is_delete = false;
>> is_return = false;
>> @@ -437,24 +435,14 @@ static int create_trace_uprobe(int argc, char **argv)
>> }
>> /* Find the last occurrence, in case the path contains ':' too. */
>> arg = strrchr(argv[1], ':');
>> - if (!arg) {
>> - ret = -EINVAL;
>> - goto fail_address_parse;
>> - }
>> + if (!arg)
>> + return -EINVAL;
>>
>> *arg++ = '\0';
>> filename = argv[1];
>> ret = kern_path(filename, LOOKUP_FOLLOW, &path);
>> if (ret)
>> - goto fail_address_parse;
>> -
>> - inode = igrab(d_real_inode(path.dentry));
Also, where has the d_real_inode() gone?
Looks like we need tu->inode back, since the return value of
d_real_inode() may change over time. I'd do the "tu->inode =
d_real_inode(tu->path.dentry)" just before first use (i.e. when
enabling the tracepoint).
Thanks,
Miklos
>> - path_put(&path);
>> -
>> - if (!inode || !S_ISREG(inode->i_mode)) {
>
> Where has the S_ISREG check gone?
>
>> - ret = -EINVAL;
>> - goto fail_address_parse;
>> - }
>> + return ret;
>>
>> ret = kstrtoul(arg, 0, &offset);
>> if (ret)
>> @@ -490,7 +478,7 @@ static int create_trace_uprobe(int argc, char **argv)
>> goto fail_address_parse;
>> }
>> tu->offset = offset;
>> - tu->inode = inode;
>> + tu->path = path;
>> tu->filename = kstrdup(filename, GFP_KERNEL);
>>
>> if (!tu->filename) {
>> @@ -558,7 +546,7 @@ static int create_trace_uprobe(int argc, char **argv)
>> return ret;
>>
>> fail_address_parse:
>> - iput(inode);
>> + path_put(&path);
>>
>> pr_info("Failed to parse address or file.\n");
>>
>> @@ -922,7 +910,8 @@ probe_event_enable(struct trace_uprobe *tu, struct trace_event_file *file,
>> goto err_flags;
>>
>> tu->consumer.filter = filter;
>> - ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
>> + ret = uprobe_register(d_inode(tu->path.dentry), tu->offset,
>> + &tu->consumer);
>> if (ret)
>> goto err_buffer;
>>
>> @@ -966,7 +955,7 @@ probe_event_disable(struct trace_uprobe *tu, struct trace_event_file *file)
>>
>> WARN_ON(!uprobe_filter_is_empty(&tu->filter));
>>
>> - uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
>> + uprobe_unregister(d_inode(tu->path.dentry), tu->offset, &tu->consumer);
>> tu->tp.flags &= file ? ~TP_FLAG_TRACE : ~TP_FLAG_PROFILE;
>>
>> uprobe_buffer_disable();
>> @@ -1041,7 +1030,8 @@ static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
>> write_unlock(&tu->filter.rwlock);
>>
>> if (!done)
>> - return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
>> + return uprobe_apply(d_inode(tu->path.dentry), tu->offset,
>> + &tu->consumer, false);
>>
>> return 0;
>> }
>> @@ -1073,7 +1063,8 @@ static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
>>
>> err = 0;
>> if (!done) {
>> - err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
>> + err = uprobe_apply(d_inode(tu->path.dentry),
>> + tu->offset, &tu->consumer, true);
>> if (err)
>> uprobe_perf_close(tu, event);
>> }
>> @@ -1337,7 +1328,6 @@ struct trace_event_call *
>> create_local_trace_uprobe(char *name, unsigned long offs, bool is_return)
>> {
>> struct trace_uprobe *tu;
>> - struct inode *inode;
>> struct path path;
>> int ret;
>>
>> @@ -1345,14 +1335,6 @@ create_local_trace_uprobe(char *name, unsigned long offs, bool is_return)
>> if (ret)
>> return ERR_PTR(ret);
>>
>> - inode = igrab(d_inode(path.dentry));
>> - path_put(&path);
>> -
>> - if (!inode || !S_ISREG(inode->i_mode)) {
>
> And here, again.
>
> Otherwise ACK.
>
> Also please create a separate patch that removes igrab/iput calls from
> kernel/events/uprobes.c and adds a comment to the effect that the
> caller is required to keep the inode (and the containing mount)
> referenced.
>
> Thanks,
> Miklos
> On Apr 19, 2018, at 7:44 AM, Miklos Szeredi <[email protected]> wrote:
>
> On Thu, Apr 19, 2018 at 10:58 AM, Miklos Szeredi <[email protected]> wrote:
>> On Wed, Apr 18, 2018 at 7:40 PM, Song Liu <[email protected]> wrote:
>>> As Miklos reported and suggested:
>>>
>>> This pattern repeats two times in trace_uprobe.c and in
>>> kernel/events/core.c as well:
>>>
>>> ret = kern_path(filename, LOOKUP_FOLLOW, &path);
>>> if (ret)
>>> goto fail_address_parse;
>>>
>>> inode = igrab(d_inode(path.dentry));
>>> path_put(&path);
>>>
>>> And it's wrong. You can only hold a reference to the inode if you
>>> have an active ref to the superblock as well (which is normally
>>> through path.mnt) or holding s_umount.
>>>
>>> This way unmounting the containing filesystem while the tracepoint is
>>> active will give you the "VFS: Busy inodes after unmount..." message
>>> and a crash when the inode is finally put.
>>>
>>> Solution: store path instead of inode.
>>>
>>> This patch fixes two instances in trace_uprobe.c.
>>>
>>> Fixes: f3f096cfedf8 ("tracing: Provide trace events interface for uprobes")
>>> Fixes: 33ea4b24277b ("perf/core: Implement the 'perf_uprobe' PMU")
>>> Cc: Steven Rostedt <[email protected]>
>>> Cc: Ingo Molnar <[email protected]>
>>> Cc: Howard McLauchlan <[email protected]>
>>> Cc: Josef Bacik <[email protected]>
>>> Cc: Srikar Dronamraju <[email protected]>
>>> Reported-by: Miklos Szeredi <[email protected]>
>>> Signed-off-by: Song Liu <[email protected]>
>>> ---
>>> kernel/trace/trace_uprobe.c | 49 +++++++++++++++------------------------------
>>> 1 file changed, 16 insertions(+), 33 deletions(-)
>>>
>>> diff --git a/kernel/trace/trace_uprobe.c b/kernel/trace/trace_uprobe.c
>>> index 34fd0e0..d9ee522c 100644
>>> --- a/kernel/trace/trace_uprobe.c
>>> +++ b/kernel/trace/trace_uprobe.c
>>> @@ -55,7 +55,7 @@ struct trace_uprobe {
>>> struct list_head list;
>>> struct trace_uprobe_filter filter;
>>> struct uprobe_consumer consumer;
>>> - struct inode *inode;
>>> + struct path path;
>>> char *filename;
>>> unsigned long offset;
>>> unsigned long nhit;
>>> @@ -289,7 +289,7 @@ static void free_trace_uprobe(struct trace_uprobe *tu)
>>> for (i = 0; i < tu->tp.nr_args; i++)
>>> traceprobe_free_probe_arg(&tu->tp.args[i]);
>>>
>>> - iput(tu->inode);
>>> + path_put(&tu->path);
>>> kfree(tu->tp.call.class->system);
>>> kfree(tu->tp.call.name);
>>> kfree(tu->filename);
>>> @@ -363,7 +363,6 @@ static int register_trace_uprobe(struct trace_uprobe *tu)
>>> static int create_trace_uprobe(int argc, char **argv)
>>> {
>>> struct trace_uprobe *tu;
>>> - struct inode *inode;
>>> char *arg, *event, *group, *filename;
>>> char buf[MAX_EVENT_NAME_LEN];
>>> struct path path;
>>> @@ -371,7 +370,6 @@ static int create_trace_uprobe(int argc, char **argv)
>>> bool is_delete, is_return;
>>> int i, ret;
>>>
>>> - inode = NULL;
>>> ret = 0;
>>> is_delete = false;
>>> is_return = false;
>>> @@ -437,24 +435,14 @@ static int create_trace_uprobe(int argc, char **argv)
>>> }
>>> /* Find the last occurrence, in case the path contains ':' too. */
>>> arg = strrchr(argv[1], ':');
>>> - if (!arg) {
>>> - ret = -EINVAL;
>>> - goto fail_address_parse;
>>> - }
>>> + if (!arg)
>>> + return -EINVAL;
>>>
>>> *arg++ = '\0';
>>> filename = argv[1];
>>> ret = kern_path(filename, LOOKUP_FOLLOW, &path);
>>> if (ret)
>>> - goto fail_address_parse;
>>> -
>>> - inode = igrab(d_real_inode(path.dentry));
>
> Also, where has the d_real_inode() gone?
>
> Looks like we need tu->inode back, since the return value of
> d_real_inode() may change over time. I'd do the "tu->inode =
> d_real_inode(tu->path.dentry)" just before first use (i.e. when
> enabling the tracepoint).
>
> Thanks,
> Miklos
>
Do we need mechanism to prevent the return value of d_real_inode()
to change? Would the following sequence happen?
create trace_uprobe
enable trace_uprobe (uprobe_register)
d_real changes
disable trace_uprobe (uprobe_unregister get wrong inode?)
Another case might be:
create trace_uprobe
enable trace_uprobe (uprobe_register)
disable trace_uprobe (uprobe_unregister)
d_real changes
enable trace_uprobe (do we need new inode for uprobe_register)
Thanks,
Song
>
>>> - path_put(&path);
>>> -
>>> - if (!inode || !S_ISREG(inode->i_mode)) {
>>
>> Where has the S_ISREG check gone?
>>
>>> - ret = -EINVAL;
>>> - goto fail_address_parse;
>>> - }
>>> + return ret;
>>>
>>> ret = kstrtoul(arg, 0, &offset);
>>> if (ret)
>>> @@ -490,7 +478,7 @@ static int create_trace_uprobe(int argc, char **argv)
>>> goto fail_address_parse;
>>> }
>>> tu->offset = offset;
>>> - tu->inode = inode;
>>> + tu->path = path;
>>> tu->filename = kstrdup(filename, GFP_KERNEL);
>>>
>>> if (!tu->filename) {
>>> @@ -558,7 +546,7 @@ static int create_trace_uprobe(int argc, char **argv)
>>> return ret;
>>>
>>> fail_address_parse:
>>> - iput(inode);
>>> + path_put(&path);
>>>
>>> pr_info("Failed to parse address or file.\n");
>>>
>>> @@ -922,7 +910,8 @@ probe_event_enable(struct trace_uprobe *tu, struct trace_event_file *file,
>>> goto err_flags;
>>>
>>> tu->consumer.filter = filter;
>>> - ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
>>> + ret = uprobe_register(d_inode(tu->path.dentry), tu->offset,
>>> + &tu->consumer);
>>> if (ret)
>>> goto err_buffer;
>>>
>>> @@ -966,7 +955,7 @@ probe_event_disable(struct trace_uprobe *tu, struct trace_event_file *file)
>>>
>>> WARN_ON(!uprobe_filter_is_empty(&tu->filter));
>>>
>>> - uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
>>> + uprobe_unregister(d_inode(tu->path.dentry), tu->offset, &tu->consumer);
>>> tu->tp.flags &= file ? ~TP_FLAG_TRACE : ~TP_FLAG_PROFILE;
>>>
>>> uprobe_buffer_disable();
>>> @@ -1041,7 +1030,8 @@ static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
>>> write_unlock(&tu->filter.rwlock);
>>>
>>> if (!done)
>>> - return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
>>> + return uprobe_apply(d_inode(tu->path.dentry), tu->offset,
>>> + &tu->consumer, false);
>>>
>>> return 0;
>>> }
>>> @@ -1073,7 +1063,8 @@ static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
>>>
>>> err = 0;
>>> if (!done) {
>>> - err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
>>> + err = uprobe_apply(d_inode(tu->path.dentry),
>>> + tu->offset, &tu->consumer, true);
>>> if (err)
>>> uprobe_perf_close(tu, event);
>>> }
>>> @@ -1337,7 +1328,6 @@ struct trace_event_call *
>>> create_local_trace_uprobe(char *name, unsigned long offs, bool is_return)
>>> {
>>> struct trace_uprobe *tu;
>>> - struct inode *inode;
>>> struct path path;
>>> int ret;
>>>
>>> @@ -1345,14 +1335,6 @@ create_local_trace_uprobe(char *name, unsigned long offs, bool is_return)
>>> if (ret)
>>> return ERR_PTR(ret);
>>>
>>> - inode = igrab(d_inode(path.dentry));
>>> - path_put(&path);
>>> -
>>> - if (!inode || !S_ISREG(inode->i_mode)) {
>>
>> And here, again.
>>
>> Otherwise ACK.
>>
>> Also please create a separate patch that removes igrab/iput calls from
>> kernel/events/uprobes.c and adds a comment to the effect that the
>> caller is required to keep the inode (and the containing mount)
>> referenced.
>>
>> Thanks,
>> Miklos
On Thu, Apr 19, 2018 at 6:37 PM, Song Liu <[email protected]> wrote:
>
>
>> On Apr 19, 2018, at 7:44 AM, Miklos Szeredi <[email protected]> wrote:
>>
>> On Thu, Apr 19, 2018 at 10:58 AM, Miklos Szeredi <[email protected]> wrote:
>>> On Wed, Apr 18, 2018 at 7:40 PM, Song Liu <[email protected]> wrote:
>>>> *arg++ = '\0';
>>>> filename = argv[1];
>>>> ret = kern_path(filename, LOOKUP_FOLLOW, &path);
>>>> if (ret)
>>>> - goto fail_address_parse;
>>>> -
>>>> - inode = igrab(d_real_inode(path.dentry));
>>
>> Also, where has the d_real_inode() gone?
>>
>> Looks like we need tu->inode back, since the return value of
>> d_real_inode() may change over time. I'd do the "tu->inode =
>> d_real_inode(tu->path.dentry)" just before first use (i.e. when
>> enabling the tracepoint).
>>
>
> Do we need mechanism to prevent the return value of d_real_inode()
> to change? Would the following sequence happen?
>
> create trace_uprobe
> enable trace_uprobe (uprobe_register)
> d_real changes
> disable trace_uprobe (uprobe_unregister get wrong inode?)
Yes.
>
> Another case might be:
>
> create trace_uprobe
> enable trace_uprobe (uprobe_register)
> disable trace_uprobe (uprobe_unregister)
> d_real changes
> enable trace_uprobe (do we need new inode for uprobe_register)
Probably a good idea to use the new one, but doesn't really matter.
Do the one that's simpler.
This corner case is simply not interesting (modifying a binary while
it is being debugged with uprobe). Let's just concentrate on making
this crash and leak free.
Thanks,
Miklos
On Fri, 20 Apr 2018 16:34:25 +0200
Miklos Szeredi <[email protected]> wrote:
> This corner case is simply not interesting (modifying a binary while
> it is being debugged with uprobe). Let's just concentrate on making
> this crash and leak free.
Yes please. I'm waiting for a final patch to run my tests. I have other
patches that need to go mainline as well, and I'll start testing all of
them when I get the final version of this one.
-- Steve