Skip to contentSkip to conversation

positional variable problem

Cocci Nicholas Mc Guire 2 participants 4 messages Nov 7, 2016

lore
Share & tools
lore
#0 lore raw

Hi !

On LKML Philippe Reynes posted an API update for ethtools API for the pcnet32 care, basically to replace deprecated {get|set}_settings by the new API calls {get|set}_link_ksettings.
see: http://lkml.org/lkml/2016/11/6/2

So the scanner for this is simple:

<snip>
@find exists@
identifier oldapi,cmd,dev;
position p;
expression E;
@@

*static int oldapi at p(struct net_device *dev, struct ethtool_cmd *cmd) {
<+...
*mii_ethtool_gset(&E, cmd);
...+>
}

@script:python@
p << find.p;
@@

msg = "old ethtool API in use."
coccilib.report.print_report(p[0], msg)
<snip>

The problem was in the API update - as the get and set basically have the
same prototype so the find function seems to trigger in both - the "solution" is a bit brute force regex in the "mod" rule - which I basicall think should not be needed.

<snip>
@initialize:python@
@@
import re
g = re.compile('get_')
s = re.compile('set_')

@find1 exists@
identifier get,cmd,dev;
expression E;
@@
static int get(struct net_device *dev, struct ethtool_cmd *cmd) {
...
mii_ethtool_gset(&E, cmd);
...
}

@find2 exists@
identifier set,cmd,dev,ret;
expression E;
@@

static int set(struct net_device *dev, struct ethtool_cmd *cmd) {
<...
ret = mii_ethtool_sset(&E, cmd);
...>
}

@script:python mod@
oldg << find1.get;
olds << find2.set;
newg;
news;
@@

// the problem here is that the prototypes of get and set
// are identical so we only can keep them appart with regex
if g.search(oldg) != None:
coccinelle.newg = oldg.replace("get_settings","") + "get_link_ksettings"
else:
cocci.include_match(False)
if s.search(olds) != None:
coccinelle.news = olds.replace("set_settings","") + "set_link_ksettings"
else:
cocci.include_match(False)

@updateget@
identifier find1.get,find1.cmd,find1.dev,mod.newg; expression E;
@@

-static int get(struct net_device *dev, struct ethtool_cmd *cmd) +static int newg(struct net_device *dev, struct ethtool_link_ksettings *cmd) {
<...
- mii_ethtool_gset(&E, cmd);
+ mii_ethtool_get_link_ksettings(&E, cmd);
...>
}

@updateset@
identifier find2.set,find2.cmd,find2.dev,mod.news; identifier ret;
expression E;
@@

-static int set(struct net_device *dev, struct ethtool_cmd *cmd) +static int news(struct net_device *dev, const struct ethtool_link_ksettings *cmd) {
<...
(
ret =
- mii_ethtool_sset(&E, cmd);
+ mii_ethtool_set_link_ksettings(&E, cmd);
|
- return mii_ethtool_sset(&E, cmd);
+ return mii_ethtool_set_link_ksettings(&E, cmd);
)
...>
}

@ksettings@
identifier ops;
identifier mod.newg;
identifier mod.news;
expression E1,E2,E3;
@@

struct ethtool_ops ops = {
...,
-.get_settings = E1,
-.set_settings = E2,
...,
.get_link = E3,
+.get_link_ksettings = newg,
+.set_link_ksettings = news,
...,
};
<snip>

Is there a more resonable solution to this ?
it seems to be working correctly but I did not see the need to actually use regex if the get and set fuction should be differenciateable by the calls in the body (mii_ethtool_gset vs. mii_ethtool_sset) but I could not get it to work on that basis.

The original attempt that fails due to the position collision (atleat that is what I assume that is happeining) is below - the two lines in the @updateset@ rule will trigger the problem if uncommented. giving:

Fatal error: exception Failure("meta: semantic error: position cannot be inherited over modifications: p
=File "eth.cocci", line 63, column 16, charpos = 1123

    around = ';', whole content = position find2.p;

")

find1 and find2 with the respective update functions seem to be ok if they are put in seperate files. So the spatch that causes this is:

<snip>
@find1 exists@
identifier get,cmd,dev;
position p;
expression E;
@@
static int get at p(struct net_device *dev, struct ethtool_cmd *cmd) {
<+...
mii_ethtool_gset(&E, cmd);
...+>
}

@find2 exists@
identifier set,cmd,dev,ret;
position p;
expression E;
@@

static int set at p(struct net_device *dev, struct ethtool_cmd *cmd) {
<+...
(
ret = mii_ethtool_sset(&E, cmd);
|
return mii_ethtool_sset(&E, cmd);
)
...+>
}

@script:python mod@
oldg << find1.get;
olds << find2.set;
newg;
news;
@@

coccinelle.newg = oldg.replace("get_settings","") + "get_link_ksettings" coccinelle.news = olds.replace("set_settings","") + "set_link_ksettings"

@updateget@
identifier find1.get,find1.cmd,find1.dev,mod.newg; position find1.p;
expression E;
@@

-static int get@p(struct net_device *dev, struct ethtool_cmd *cmd) +static int newg(struct net_device *dev, struct ethtool_link_ksettings *cmd) {
<...
- mii_ethtool_gset(&E, cmd);
+ mii_ethtool_get_link_ksettings(&E, cmd);
...>
}

@updateset@
identifier find2.set,find2.cmd,find2.dev,mod.news; identifier ret;
// the position causing the problem //position find2.p;
expression E;
@@

//-static int set at p(struct net_device *dev, struct ethtool_cmd *cmd) -static int set(struct net_device *dev, struct ethtool_cmd *cmd) +static int news(struct net_device *dev, const struct ethtool_link_ksettings *cmd) {
<...
(
ret =
- mii_ethtool_sset(&E, cmd);
+ mii_ethtool_set_link_ksettings(&E, cmd);
|
- return mii_ethtool_sset(&E, cmd);
+ return mii_ethtool_set_link_ksettings(&E, cmd);
)
...>
}

@ksettings@
identifier ops;
identifier mod.newg;
identifier mod.news;
expression E1,E2,E3;
@@

struct ethtool_ops ops = {
...,
-.get_settings = E1,
-.set_settings = E2,
...,
.get_link = E3,
+.get_link_ksettings = newg,
+.set_link_ksettings = news,
...,
};
<snip>

thx!
hofrat

#1 lore raw
Quoted 28 lines · Nicholas Mc Guire

Hi !

On LKML Philippe Reynes posted an API update for ethtools API for the pcnet32 care, basically to replace deprecated {get|set}_settings by the new API calls {get|set}_link_ksettings.
see: http://lkml.org/lkml/2016/11/6/2

So the scanner for this is simple:

<snip>
@find exists@
identifier oldapi,cmd,dev;
position p;
expression E;
@@

*static int oldapi at p(struct net_device *dev, struct ethtool_cmd *cmd) {
<+...
*mii_ethtool_gset(&E, cmd);
...+>
}

@script:python@
p << find.p;
@@

Quoted 8 lines

msg = "old ethtool API in use."
coccilib.report.print_report(p[0], msg)
<snip>

The problem was in the API update - as the get and set basically have the
same prototype so the find function seems to trigger in both - the "solution" is a bit brute force regex in the "mod" rule - which I basicall think should not be needed.

I'm not completely following this... Would it work to start with the ethtool_ops structure? Then you could see what entry point functions need to be updated? It would be good to also have a rule afterwards that looks for other calls to mii_ethtool_gset and mii_ethtool_sset to be sure that you haven't missed anything. Is the issue is that you don't want to modify some call to eg mii_ethtool_gset that may be in the definition of mii_ethtool_get_link_ksettings itself? In that case, you can do soemthing like the following:

@exists@
identifier f != {function_to_ignore};
@@

f(...) {
<+...
*function_to_warn_about(...)
...+> }

Another approach that would be more efficient is to use the new ability to attach scripts to metavariables. This is illustrated in tests/python_poscon.cocci.

@@
position p : script:python() { p[0].current_element != "function_to_ignore" }; @@

function_to_warn_about at p(...)

Let me know if I've missed understanding the issue completely.

julia

Quoted 227 lines

<snip>
@initialize:python@
@@
import re
g = re.compile('get_')
s = re.compile('set_')

@find1 exists@
identifier get,cmd,dev;
expression E;
@@
static int get(struct net_device *dev, struct ethtool_cmd *cmd) {
...
mii_ethtool_gset(&E, cmd);
...
}

@find2 exists@
identifier set,cmd,dev,ret;
expression E;
@@

static int set(struct net_device *dev, struct ethtool_cmd *cmd) {
<...
ret = mii_ethtool_sset(&E, cmd);
...>
}

@script:python mod@
oldg << find1.get;
olds << find2.set;
newg;
news;
@@

// the problem here is that the prototypes of get and set
// are identical so we only can keep them appart with regex
if g.search(oldg) != None:
coccinelle.newg = oldg.replace("get_settings","") + "get_link_ksettings"
else:
cocci.include_match(False)
if s.search(olds) != None:
coccinelle.news = olds.replace("set_settings","") + "set_link_ksettings"
else:
cocci.include_match(False)

@updateget@
identifier find1.get,find1.cmd,find1.dev,mod.newg; expression E;
@@

(quoted diff) +1 -1
-static int get(struct net_device *dev, struct ethtool_cmd *cmd)
+static int newg(struct net_device *dev, struct ethtool_link_ksettings *cmd)

{
<...
- mii_ethtool_gset(&E, cmd);
+ mii_ethtool_get_link_ksettings(&E, cmd);
...>
}

@updateset@
identifier find2.set,find2.cmd,find2.dev,mod.news; identifier ret;
expression E;
@@

(quoted diff) +1 -1
-static int set(struct net_device *dev, struct ethtool_cmd *cmd)
+static int news(struct net_device *dev, const struct ethtool_link_ksettings *cmd)

{
<...
(
ret =
- mii_ethtool_sset(&E, cmd);
+ mii_ethtool_set_link_ksettings(&E, cmd);
|
- return mii_ethtool_sset(&E, cmd);
+ return mii_ethtool_set_link_ksettings(&E, cmd);
)
...>
}

@ksettings@
identifier ops;
identifier mod.newg;
identifier mod.news;
expression E1,E2,E3;
@@

struct ethtool_ops ops = {
...,
-.get_settings = E1,
-.set_settings = E2,
...,
.get_link = E3,
+.get_link_ksettings = newg,
+.set_link_ksettings = news,
...,
};
<snip>

Is there a more resonable solution to this ?
it seems to be working correctly but I did not see the need to actually use regex if the get and set fuction should be differenciateable by the calls in the body (mii_ethtool_gset vs. mii_ethtool_sset) but I could not get it to work on that basis.

The original attempt that fails due to the position collision (atleat that is what I assume that is happeining) is below - the two lines in the @updateset@ rule will trigger the problem if uncommented. giving:

Fatal error: exception Failure("meta: semantic error: position cannot be inherited over modifications: p
=File "eth.cocci", line 63, column 16, charpos = 1123

    around = ';', whole content = position find2.p;

")

find1 and find2 with the respective update functions seem to be ok if they are put in seperate files. So the spatch that causes this is:

<snip>
@find1 exists@
identifier get,cmd,dev;
position p;
expression E;
@@
static int get at p(struct net_device *dev, struct ethtool_cmd *cmd) {
<+...
mii_ethtool_gset(&E, cmd);
...+>
}

@find2 exists@
identifier set,cmd,dev,ret;
position p;
expression E;
@@

static int set at p(struct net_device *dev, struct ethtool_cmd *cmd) {
<+...
(
ret = mii_ethtool_sset(&E, cmd);
|
return mii_ethtool_sset(&E, cmd);
)
...+>
}

@script:python mod@
oldg << find1.get;
olds << find2.set;
newg;
news;
@@

coccinelle.newg = oldg.replace("get_settings","") + "get_link_ksettings" coccinelle.news = olds.replace("set_settings","") + "set_link_ksettings"

@updateget@
identifier find1.get,find1.cmd,find1.dev,mod.newg; position find1.p;
expression E;
@@

(quoted diff) +1 -1
-static int get at p(struct net_device *dev, struct ethtool_cmd *cmd)
+static int newg(struct net_device *dev, struct ethtool_link_ksettings *cmd)

{
<...
- mii_ethtool_gset(&E, cmd);
+ mii_ethtool_get_link_ksettings(&E, cmd);
...>
}

@updateset@
identifier find2.set,find2.cmd,find2.dev,mod.news; identifier ret;
// the position causing the problem //position find2.p;
expression E;
@@

//-static int set at p(struct net_device *dev, struct ethtool_cmd *cmd)

(quoted diff) +1 -1
-static int set(struct net_device *dev, struct ethtool_cmd *cmd)
+static int news(struct net_device *dev, const struct ethtool_link_ksettings *cmd)

{
<...
(
ret =
- mii_ethtool_sset(&E, cmd);
+ mii_ethtool_set_link_ksettings(&E, cmd);
|
- return mii_ethtool_sset(&E, cmd);
+ return mii_ethtool_set_link_ksettings(&E, cmd);
)
...>
}

@ksettings@
identifier ops;
identifier mod.newg;
identifier mod.news;
expression E1,E2,E3;
@@

struct ethtool_ops ops = {
...,
-.get_settings = E1,
-.set_settings = E2,
...,
.get_link = E3,
+.get_link_ksettings = newg,
+.set_link_ksettings = news,
...,
};
<snip>

thx!
hofrat
_______________________________________________
Cocci mailing list
Cocci at systeme.lip6.fr
https://systeme.lip6.fr/mailman/listinfo/cocci

#2 lore raw
Quoted 36 lines · Julia Lawall
Quoted 32 lines · Nicholas Mc Guire

Hi !

On LKML Philippe Reynes posted an API update for ethtools API for the pcnet32 care, basically to replace deprecated {get|set}_settings by the new API calls {get|set}_link_ksettings.
see: http://lkml.org/lkml/2016/11/6/2

So the scanner for this is simple:

<snip>
@find exists@
identifier oldapi,cmd,dev;
position p;
expression E;
@@

*static int oldapi at p(struct net_device *dev, struct ethtool_cmd *cmd) {
<+...
*mii_ethtool_gset(&E, cmd);
...+>
}

@script:python@
p << find.p;
@@

msg = "old ethtool API in use."
coccilib.report.print_report(p[0], msg)
<snip>

Quoted 7 lines
Quoted 4 lines

The problem was in the API update - as the get and set basically have the
same prototype so the find function seems to trigger in both - the "solution" is a bit brute force regex in the "mod" rule - which I basicall think should not be needed.

I'm not completely following this... Would it work to start with the ethtool_ops structure? Then you could see what entry point functions need

The problem with that was that one can not tell from the ops if its the old API or the new API - so I thought it is simpler to scan for the old api prototypes and use the *gset/*sset functions as check.

Quoted 20 lines

to be updated? It would be good to also have a rule afterwards that looks for other calls to mii_ethtool_gset and mii_ethtool_sset to be sure that you haven't missed anything. Is the issue is that you don't want to modify some call to eg mii_ethtool_gset that may be in the definition of mii_ethtool_get_link_ksettings itself? In that case, you can do soemthing like the following:

@exists@
identifier f != {function_to_ignore};
@@

f(...) {
<+...
*function_to_warn_about(...)
...+> }

Another approach that would be more efficient is to use the new ability to attach scripts to metavariables. This is illustrated in tests/python_poscon.cocci.

Quoted 8 lines

@@
position p : script:python() { p[0].current_element != "function_to_ignore" }; @@

function_to_warn_about at p(...)

Let me know if I've missed understanding the issue completely.

nop - I?ve just managed to ignore the available capabiliteis and come up with a complicated (and as you point out possibly incomplete) solution. Notably the positional scripts sounds like a simple way to resolve this in a cleaner way.

will try to unrole it from the ops side.

thx!
hofrat

#3 lore raw
Quoted 42 lines · Nicholas Mc Guire
Quoted 41 lines · Julia Lawall
Quoted 36 lines · Nicholas Mc Guire

Hi !

On LKML Philippe Reynes posted an API update for ethtools API for the pcnet32 care, basically to replace deprecated {get|set}_settings by the new API calls {get|set}_link_ksettings.
see: http://lkml.org/lkml/2016/11/6/2

So the scanner for this is simple:

<snip>
@find exists@
identifier oldapi,cmd,dev;
position p;
expression E;
@@

*static int oldapi at p(struct net_device *dev, struct ethtool_cmd *cmd) {
<+...
*mii_ethtool_gset(&E, cmd);
...+>
}

@script:python@
p << find.p;
@@

msg = "old ethtool API in use."
coccilib.report.print_report(p[0], msg)
<snip>

The problem was in the API update - as the get and set basically have the
same prototype so the find function seems to trigger in both - the "solution" is a bit brute force regex in the "mod" rule - which I basicall think should not be needed.

Quoted 7 lines
Quoted 2 lines

I'm not completely following this... Would it work to start with the ethtool_ops structure? Then you could see what entry point functions need

The problem with that was that one can not tell from the ops if its the old API or the new API - so I thought it is simpler to scan for the old api prototypes and use the *gset/*sset functions as check.

I had the impression that it was clear from the structure, because the field names are different.

julia

Quoted 39 lines
Quoted 21 lines

to be updated? It would be good to also have a rule afterwards that looks for other calls to mii_ethtool_gset and mii_ethtool_sset to be sure that you haven't missed anything. Is the issue is that you don't want to modify some call to eg mii_ethtool_gset that may be in the definition of mii_ethtool_get_link_ksettings itself? In that case, you can do soemthing like the following:

@exists@
identifier f != {function_to_ignore};
@@

f(...) {
<+...
*function_to_warn_about(...)
...+> }

Another approach that would be more efficient is to use the new ability to attach scripts to metavariables. This is illustrated in tests/python_poscon.cocci.

Quoted 8 lines

@@
position p : script:python() { p[0].current_element != "function_to_ignore" }; @@

function_to_warn_about at p(...)

Let me know if I've missed understanding the issue completely.

nop - I?ve just managed to ignore the available capabiliteis and come up with a complicated (and as you point out possibly incomplete) solution. Notably the positional scripts sounds like a simple way to resolve this in a cleaner way.

will try to unrole it from the ops side.

thx!
hofrat