Обсуждение: Don't pass NULL pointer to strcmp().

Hi hackers,

I found that there's a nullable pointer being passed to strcmp() and
can make the server crash. It can be reproduced on the latest master
branch by crafting an extension[1]. Patch for fixing it is attatched.

[1] https://github.com/higuoxing/guc_crash/tree/pg

-- 
Best Regards,
Xing

On Wed, Nov 1, 2023 at 5:25 PM Xing Guo <higuoxing@gmail.com> wrote:
>
> Hi hackers,
>
> I found that there's a nullable pointer being passed to strcmp() and
> can make the server crash. It can be reproduced on the latest master
> branch by crafting an extension[1]. Patch for fixing it is attatched.
>
> [1] https://github.com/higuoxing/guc_crash/tree/pg
>

Can we set a string guc to NULL? If not, `*lconf->variable == NULL` would
be unnecessary.

> --
> Best Regards,
> Xing



--
Regards
Junwang Zhao

Hi,

> > I found that there's a nullable pointer being passed to strcmp() and
> > can make the server crash. It can be reproduced on the latest master
> > branch by crafting an extension[1]. Patch for fixing it is attatched.
> >
> > [1] https://github.com/higuoxing/guc_crash/tree/pg

Thanks for reporting. I can confirm that the issue reproduces on the
`master` branch and the proposed patch fixes it.

> Can we set a string guc to NULL? If not, `*lconf->variable == NULL` would
> be unnecessary.

Judging by the rest of the code we better keep it, at least for consistenc.

I see one more place with a similar code in guc.c around line 1472.
Although I don't have exact steps to trigger a crash I suggest adding
a similar check there.

-- 
Best regards,
Aleksander Alekseev

Xing Guo <higuoxing@gmail.com> writes:
> Thanks for your comments. I have updated the patch accordingly.

I'm leery of accepting this patch, as I see no reason that we
should consider it valid for an extension to have a string GUC
with a boot_val of NULL.

I realize that we have a few core GUCs that are like that, but
I'm pretty sure that every one of them has special-case code
that initializes the GUC to something non-null a bit later on
in startup.  I don't think there are any cases where a string
GUC's persistent value will be null, and I don't like the
idea of considering that to be an allowed case.  It would
open the door to more crash situations, and it brings up the
old question of how could a user tell NULL from empty string
(via SHOW or current_setting() or whatever).  Besides, what's
the benefit really?

            regards, tom lane

Xing Guo <higuoxing@gmail.com> writes:
> There're extensions that set their boot_val to NULL. E.g., postgres_fdw

Hmm ... if we're doing it ourselves, I suppose we've got to consider
it supported :-(.  But I'm still wondering how many seldom-used
code paths didn't get the message.  An example here is that this
could lead to GetConfigOptionResetString returning NULL, which
I think is outside its admittedly-vague API spec.

            regards, tom lane

I wrote:
> Hmm ... if we're doing it ourselves, I suppose we've got to consider
> it supported :-(.  But I'm still wondering how many seldom-used
> code paths didn't get the message.  An example here is that this
> could lead to GetConfigOptionResetString returning NULL, which
> I think is outside its admittedly-vague API spec.

After digging around for a bit, I think part of the problem is a lack
of a clearly defined spec for what should happen with NULL string GUCs.
In the attached v3, I attempted to remedy that by adding a comment in
guc_tables.h (which is maybe not the best place but I didn't see a
better one).  That led me to a couple more changes beyond what you had.

It's possible that some of these are unreachable --- for example,
given that a NULL could only be the default value, I'm not sure that
the fix in write_one_nondefault_variable is a live bug.  But we ought
to code all this stuff defensively, and most of it already was
NULL-safe.

            regards, tom lane

diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c
index 39d3775e80..ccffaaad02 100644
--- a/src/backend/utils/misc/guc.c
+++ b/src/backend/utils/misc/guc.c
@@ -1473,7 +1473,9 @@ check_GUC_init(struct config_generic *gconf)
             {
                 struct config_string *conf = (struct config_string *) gconf;

-                if (*conf->variable != NULL && strcmp(*conf->variable, conf->boot_val) != 0)
+                if (*conf->variable != NULL &&
+                    (conf->boot_val == NULL ||
+                     strcmp(*conf->variable, conf->boot_val) != 0))
                 {
                     elog(LOG, "GUC (PGC_STRING) %s, boot_val=%s, C-var=%s",
                          conf->gen.name, conf->boot_val ? conf->boot_val : "<null>", *conf->variable);
@@ -4213,8 +4215,7 @@ SetConfigOption(const char *name, const char *value,
 /*
  * Fetch the current value of the option `name', as a string.
  *
- * If the option doesn't exist, return NULL if missing_ok is true (NOTE that
- * this cannot be distinguished from a string variable with a NULL value!),
+ * If the option doesn't exist, return NULL if missing_ok is true,
  * otherwise throw an ereport and don't return.
  *
  * If restrict_privileged is true, we also enforce that only superusers and
@@ -4257,7 +4258,8 @@ GetConfigOption(const char *name, bool missing_ok, bool restrict_privileged)
             return buffer;

         case PGC_STRING:
-            return *((struct config_string *) record)->variable;
+            return *((struct config_string *) record)->variable ?
+                *((struct config_string *) record)->variable : "";

         case PGC_ENUM:
             return config_enum_lookup_by_value((struct config_enum *) record,
@@ -4304,7 +4306,8 @@ GetConfigOptionResetString(const char *name)
             return buffer;

         case PGC_STRING:
-            return ((struct config_string *) record)->reset_val;
+            return ((struct config_string *) record)->reset_val ?
+                ((struct config_string *) record)->reset_val : "";

         case PGC_ENUM:
             return config_enum_lookup_by_value((struct config_enum *) record,
@@ -5255,7 +5258,14 @@ get_explain_guc_options(int *num)
                 {
                     struct config_string *lconf = (struct config_string *) conf;

-                    modified = (strcmp(lconf->boot_val, *(lconf->variable)) != 0);
+                    if (lconf->boot_val == NULL &&
+                        *lconf->variable == NULL)
+                        modified = false;
+                    else if (lconf->boot_val == NULL ||
+                             *lconf->variable == NULL)
+                        modified = true;
+                    else
+                        modified = (strcmp(lconf->boot_val, *(lconf->variable)) != 0);
                 }
                 break;

@@ -5482,7 +5492,8 @@ write_one_nondefault_variable(FILE *fp, struct config_generic *gconf)
             {
                 struct config_string *conf = (struct config_string *) gconf;

-                fprintf(fp, "%s", *conf->variable);
+                if (*conf->variable)
+                    fprintf(fp, "%s", *conf->variable);
             }
             break;

@@ -6142,7 +6153,8 @@ RestoreGUCState(void *gucstate)
                 {
                     struct config_string *conf = (struct config_string *) gconf;

-                    guc_free(*conf->variable);
+                    if (*conf->variable)
+                        guc_free(*conf->variable);
                     if (conf->reset_val && conf->reset_val != *conf->variable)
                         guc_free(conf->reset_val);
                     if (conf->reset_extra && conf->reset_extra != gconf->extra)
diff --git a/src/include/utils/guc_tables.h b/src/include/utils/guc_tables.h
index 1ec9575570..0c38255961 100644
--- a/src/include/utils/guc_tables.h
+++ b/src/include/utils/guc_tables.h
@@ -240,6 +240,16 @@ struct config_real
     void       *reset_extra;
 };

+/*
+ * A note about string GUCs: the boot_val is allowed to be NULL, which leads
+ * to the reset_val and the actual variable value (*variable) also being NULL.
+ * However, there is no way to set a NULL value subsequently using
+ * set_config_option or any other GUC API.  Also, GUC APIs such as SHOW will
+ * display a NULL value as an empty string.  Callers that choose to use a NULL
+ * boot_val should overwrite the setting later in startup, or else be careful
+ * that NULL doesn't have semantics that are visibly different from an empty
+ * string.
+ */
 struct config_string
 {
     struct config_generic gen;

Thank you Tom!

Your comment
"NULL doesn't have semantics that are visibly different from an empty
string" is exactly what I want to confirm :-)

On 11/2/23, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> I wrote:
>> Hmm ... if we're doing it ourselves, I suppose we've got to consider
>> it supported :-(.  But I'm still wondering how many seldom-used
>> code paths didn't get the message.  An example here is that this
>> could lead to GetConfigOptionResetString returning NULL, which
>> I think is outside its admittedly-vague API spec.
>
> After digging around for a bit, I think part of the problem is a lack
> of a clearly defined spec for what should happen with NULL string GUCs.
> In the attached v3, I attempted to remedy that by adding a comment in
> guc_tables.h (which is maybe not the best place but I didn't see a
> better one).  That led me to a couple more changes beyond what you had.
>
> It's possible that some of these are unreachable --- for example,
> given that a NULL could only be the default value, I'm not sure that
> the fix in write_one_nondefault_variable is a live bug.  But we ought
> to code all this stuff defensively, and most of it already was
> NULL-safe.
>
>             regards, tom lane
>
>


-- 
Best Regards,
Xing

On Wed, Nov 01, 2023 at 09:57:18PM -0400, Tom Lane wrote:
> I wrote:
>> Hmm ... if we're doing it ourselves, I suppose we've got to consider
>> it supported :-(.  But I'm still wondering how many seldom-used
>> code paths didn't get the message.  An example here is that this
>> could lead to GetConfigOptionResetString returning NULL, which
>> I think is outside its admittedly-vague API spec.
> 
> After digging around for a bit, I think part of the problem is a lack
> of a clearly defined spec for what should happen with NULL string GUCs.
> In the attached v3, I attempted to remedy that by adding a comment in
> guc_tables.h (which is maybe not the best place but I didn't see a
> better one).  That led me to a couple more changes beyond what you had.

What if we disallowed NULL string GUCs in v17?  That'd simplify the spec
and future-proof against similar bugs, but it might also break a fair
number of extensions.  If there aren't any other reasons to continue
supporting it, maybe it's the right long-term approach, though.

-- 
Nathan Bossart
Amazon Web Services: https://aws.amazon.com

Nathan Bossart <nathandbossart@gmail.com> writes:
> On Wed, Nov 01, 2023 at 09:57:18PM -0400, Tom Lane wrote:
>> After digging around for a bit, I think part of the problem is a lack
>> of a clearly defined spec for what should happen with NULL string GUCs.

> What if we disallowed NULL string GUCs in v17?

Well, we'd need to devise some other solution for hacks like the
one used by timezone_abbreviations (see comment in
check_timezone_abbreviations).  I think it's not worth the trouble,
especially seeing that 95% of guc.c is already set up for this.
The bugs are mostly in newer code like get_explain_guc_options,
and I think that's directly traceable to the lack of any comments
or docs about this behavior.

            regards, tom lane

On Wed, Nov 01, 2023 at 10:39:04PM -0400, Tom Lane wrote:
> Nathan Bossart <nathandbossart@gmail.com> writes:
>> What if we disallowed NULL string GUCs in v17?
> 
> Well, we'd need to devise some other solution for hacks like the
> one used by timezone_abbreviations (see comment in
> check_timezone_abbreviations).  I think it's not worth the trouble,
> especially seeing that 95% of guc.c is already set up for this.
> The bugs are mostly in newer code like get_explain_guc_options,
> and I think that's directly traceable to the lack of any comments
> or docs about this behavior.

Eh, yeah, it's probably not worth it if we find ourselves trading one set
of hacks for another.

-- 
Nathan Bossart
Amazon Web Services: https://aws.amazon.com

Hi,

Seems that Tom's patch cannot be applied to the current master branch.
I just re-generate the patch for others to play with.

On 11/2/23, Nathan Bossart <nathandbossart@gmail.com> wrote:
> On Wed, Nov 01, 2023 at 10:39:04PM -0400, Tom Lane wrote:
>> Nathan Bossart <nathandbossart@gmail.com> writes:
>>> What if we disallowed NULL string GUCs in v17?
>>
>> Well, we'd need to devise some other solution for hacks like the
>> one used by timezone_abbreviations (see comment in
>> check_timezone_abbreviations).  I think it's not worth the trouble,
>> especially seeing that 95% of guc.c is already set up for this.
>> The bugs are mostly in newer code like get_explain_guc_options,
>> and I think that's directly traceable to the lack of any comments
>> or docs about this behavior.
>
> Eh, yeah, it's probably not worth it if we find ourselves trading one set
> of hacks for another.
>
> --
> Nathan Bossart
> Amazon Web Services: https://aws.amazon.com
>


-- 
Best Regards,
Xing

Looking closer, I realized that my proposed change in RestoreGUCState
is unnecessary, because guc_free() is already permissive about being
passed a NULL.  That leaves us with one live bug in
get_explain_guc_options, two probably-unreachable hazards in
check_GUC_init and write_one_nondefault_variable, and two API changes
in GetConfigOption and GetConfigOptionResetString.  I'm dubious that
back-patching the API changes would be a good idea, so I applied
that to HEAD only.  The rest I backpatched as far as relevant.

Thanks for the report!

            regards, tom lane