2004-10-13 14:59:43

by Petr Vandrovec

Subject: [PATCH] Weak symbols in modules and versioned symbols

Hello,
is there some reason why to not apply patch below? Without patch below
I'm getting

vmmon: no version for "sys_ioctl" found: kernel tainted.

when loading vmmon into kernel which exports sys_ioctl (and vmmon uses
weak import for sys_ioctl because I'm not able to determine sys_ioctl
availablilty for modules at build time) and no error when loading it
into the kernel which does not export sys_ioctl to modules.

With added CRCs it also looks safer to me: I do not want just
random sys_ioctl; I need sys_ioctl with signature 0xABCDEFGH.
Thanks,
Petr Vandrovec


diff -urdN linux/scripts/mod/modpost.c linux/scripts/mod/modpost.c
--- linux/scripts/mod/modpost.c 2004-10-11 13:29:03.000000000 +0000
+++ linux/scripts/mod/modpost.c 2004-10-13 14:50:16.000000000 +0000
@@ -102,6 +102,7 @@
struct module *module;
unsigned int crc;
int crc_valid;
+ unsigned int weak:1;
char name[0];
};

@@ -124,12 +125,13 @@
* the list of unresolved symbols per module */

struct symbol *
-alloc_symbol(const char *name, struct symbol *next)
+alloc_symbol(const char *name, unsigned int weak, struct symbol *next)
{
struct symbol *s = NOFAIL(malloc(sizeof(*s) + strlen(name) + 1));

memset(s, 0, sizeof(*s));
strcpy(s->name, name);
+ s->weak = weak;
s->next = next;
return s;
}
@@ -143,7 +145,7 @@
struct symbol *new;

hash = tdb_hash(name) % SYMBOL_HASH_SIZE;
- new = symbolhash[hash] = alloc_symbol(name, symbolhash[hash]);
+ new = symbolhash[hash] = alloc_symbol(name, 0, symbolhash[hash]);
new->module = module;
if (crc) {
new->crc = *crc;
@@ -347,7 +349,8 @@
break;
case SHN_UNDEF:
/* undefined symbol */
- if (ELF_ST_BIND(sym->st_info) != STB_GLOBAL)
+ if (ELF_ST_BIND(sym->st_info) != STB_GLOBAL &&
+ ELF_ST_BIND(sym->st_info) != STB_WEAK)
break;
/* ignore global offset table */
if (strcmp(symname, "_GLOBAL_OFFSET_TABLE_") == 0)
@@ -368,6 +371,7 @@
strlen(MODULE_SYMBOL_PREFIX)) == 0)
mod->unres = alloc_symbol(symname +
strlen(MODULE_SYMBOL_PREFIX),
+ ELF_ST_BIND(sym->st_info) == STB_WEAK,
mod->unres);
break;
default:
@@ -433,7 +437,7 @@
* the automatic versioning doesn't pick it up, but it's really
* important anyhow */
if (modversions)
- mod->unres = alloc_symbol("struct_module", mod->unres);
+ mod->unres = alloc_symbol("struct_module", 0, mod->unres);
}

#define SZ 500
@@ -505,7 +509,7 @@
for (s = mod->unres; s; s = s->next) {
exp = find_symbol(s->name);
if (!exp || exp->module == mod) {
- if (have_vmlinux)
+ if (have_vmlinux && !s->weak)
fprintf(stderr, "*** Warning: \"%s\" [%s.ko] "
"undefined!\n", s->name, mod->name);
continue;

2004-10-13 23:01:31

by Rusty Russell

Subject: Re: [PATCH] Weak symbols in modules and versioned symbols

On Thu, 2004-10-14 at 00:58, Petr Vandrovec wrote:
> Hello,
> is there some reason why to not apply patch below? Without patch below
> I'm getting
>
> vmmon: no version for "sys_ioctl" found: kernel tainted.

Sure, patch seems fine to me. I guess you're the first one to use weak
symbols in a module. In fact, I think PPC64 might barf on them anyway.

Please ensure you've read Documentation/SubmittingPatches line 262
onwards "11) Sign your work" and resend to me off-line.

Thanks!
Rusty.
--
Anyone who quotes me in their signature is an idiot -- Rusty Russell