[PATCH 1/2] module: Overwrite st_size instead of st_info

st_info is currently overwritten after relocation and used to store the
elf_type().  However, we're going to need it fix kallsyms on ARM's
Thumb-2 kernels, so preserve st_info and overwrite the st_size field
instead.  st_size is neither used by the module core nor by any
architecture.

Signed-off-by: Vincent Whitchurch <redacted>
---
v4: Split out to separate patch.  Use st_size instead of st_other.
v1-v3: See PATCH 2/2

 kernel/module.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/kernel/module.c b/kernel/module.c
index 49a405891587..3d86a38b580c 100644
--- a/kernel/module.c
+++ b/kernel/module.c

@@ -2682,7 +2682,7 @@ static void add_kallsyms(struct module *mod, const struct load_info *info)
 
 	/* Set types up while we still have access to sections. */
 	for (i = 0; i < mod->kallsyms->num_symtab; i++)
-		mod->kallsyms->symtab[i].st_info
+		mod->kallsyms->symtab[i].st_size
 			= elf_type(&mod->kallsyms->symtab[i], info);
 
 	/* Now populate the cut down core kallsyms for after init. */

@@ -4061,7 +4061,7 @@ int module_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
 		kallsyms = rcu_dereference_sched(mod->kallsyms);
 		if (symnum < kallsyms->num_symtab) {
 			*value = kallsyms->symtab[symnum].st_value;
-			*type = kallsyms->symtab[symnum].st_info;
+			*type = kallsyms->symtab[symnum].st_size;
 			strlcpy(name, symname(kallsyms, symnum), KSYM_NAME_LEN);
 			strlcpy(module_name, mod->name, MODULE_NAME_LEN);
 			*exported = is_exported(name, *value, mod);

-- 
2.11.0

Thumb-2 functions have the lowest bit set in the symbol value in the
symtab.  When kallsyms are generated for the vmlinux, the kallsyms are
generated from the output of nm, and nm clears the lowest bit.

 $ arm-linux-gnueabihf-readelf -a vmlinux | grep show_interrupts
  95947: 8015dc89   686 FUNC    GLOBAL DEFAULT    2 show_interrupts
 $ arm-linux-gnueabihf-nm vmlinux | grep show_interrupts
 8015dc88 T show_interrupts
 $ cat /proc/kallsyms | grep show_interrupts
 8015dc88 T show_interrupts

However, for modules, the kallsyms uses the values in the symbol table
without modification, so for functions in modules, the lowest bit is set
in kallsyms.

 $ arm-linux-gnueabihf-readelf -a drivers/net/tun.ko | grep tun_get_socket
    333: 00002d4d    36 FUNC    GLOBAL DEFAULT    1 tun_get_socket
 $ arm-linux-gnueabihf-nm drivers/net/tun.ko | grep tun_get_socket
 00002d4c T tun_get_socket
 $ cat /proc/kallsyms | grep tun_get_socket
 7f802d4d t tun_get_socket      [tun]

Because of this, the symbol+offset of the crashing instruction shown in
oopses is incorrect when the crash is in a module.  For example, given a
tun_get_socket which starts like this,

 00002d4c <tun_get_socket>:
     2d4c:       6943            ldr     r3, [r0, #20]
     2d4e:       4a07            ldr     r2, [pc, #28]
     2d50:       4293            cmp     r3, r2

a crash when tun_get_socket is called with NULL results in:

 PC is at tun_xdp+0xa3/0xa4 [tun]
 pc : [<7f802d4c>]

As can be seen, the "PC is at" line reports the wrong symbol name, and
the symbol+offset will point to the wrong source line if it is passed to
gdb.

To solve this, add a way for archs to fixup the reading of these module
kallsyms values, and use that to clear the lowest bit for function
symbols on Thumb-2.

After the fix:

 # cat /proc/kallsyms | grep tun_get_socket
 7f802d4c t tun_get_socket       [tun]

 PC is at tun_get_socket+0x0/0x24 [tun]
 pc : [<7f802d4c>]

Signed-off-by: Vincent Whitchurch <redacted>
---
v4: Split out st_value overwrite change.  Add HAVE* macro to avoid function call.
v3: Do not overwrite st_value
v2: Fix build warning with !MODULES

 arch/arm/include/asm/module.h | 11 +++++++++++
 include/linux/module.h        |  7 +++++++
 kernel/module.c               | 25 ++++++++++++++-----------
 3 files changed, 32 insertions(+), 11 deletions(-)

diff --git a/arch/arm/include/asm/module.h b/arch/arm/include/asm/module.h
index 9e81b7c498d8..e2ccec651e71 100644
--- a/arch/arm/include/asm/module.h
+++ b/arch/arm/include/asm/module.h

@@ -61,4 +61,15 @@ u32 get_module_plt(struct module *mod, unsigned long loc, Elf32_Addr val);
 	MODULE_ARCH_VERMAGIC_ARMTHUMB \
 	MODULE_ARCH_VERMAGIC_P2V
 
+#ifdef CONFIG_THUMB2_KERNEL
+#define HAVE_ARCH_MODULE_KALLSYMS_SYMBOL_VALUE
+static inline unsigned long module_kallsyms_symbol_value(Elf_Sym *sym)
+{
+	if (ELF_ST_TYPE(sym->st_info) == STT_FUNC)
+		return sym->st_value & ~1;
+
+	return sym->st_value;
+}
+#endif
+
 #endif /* _ASM_ARM_MODULE_H */

diff --git a/include/linux/module.h b/include/linux/module.h
index fce6b4335e36..cfc55f358800 100644
--- a/include/linux/module.h
+++ b/include/linux/module.h

@@ -486,6 +486,13 @@ struct module {
 #define MODULE_ARCH_INIT {}
 #endif
 
+#ifndef HAVE_ARCH_MODULE_KALLSYMS_SYMBOL_VALUE
+static inline unsigned long module_kallsyms_symbol_value(Elf_Sym *sym)
+{
+	return sym->st_value;
+}
+#endif
+
 extern struct mutex module_mutex;
 
 /* FIXME: It'd be nice to isolate modules during init, too, so they

diff --git a/kernel/module.c b/kernel/module.c
index 3d86a38b580c..a36d7915ed2b 100644
--- a/kernel/module.c
+++ b/kernel/module.c

@@ -3934,6 +3934,9 @@ static const char *get_ksymbol(struct module *mod,
 	/* Scan for closest preceding symbol, and next symbol. (ELF
 	   starts real symbols at 1). */
 	for (i = 1; i < kallsyms->num_symtab; i++) {
+		unsigned long thisval = module_kallsyms_symbol_value(&kallsyms->symtab[i]);
+		unsigned long bestval = module_kallsyms_symbol_value(&kallsyms->symtab[best]);
+
 		if (kallsyms->symtab[i].st_shndx == SHN_UNDEF)
 			continue;
 

@@ -3943,21 +3946,21 @@ static const char *get_ksymbol(struct module *mod,
 		    || is_arm_mapping_symbol(symname(kallsyms, i)))
 			continue;
 
-		if (kallsyms->symtab[i].st_value <= addr
-		    && kallsyms->symtab[i].st_value > kallsyms->symtab[best].st_value)
+		if (thisval <= addr
+		    && thisval > bestval)
 			best = i;
-		if (kallsyms->symtab[i].st_value > addr
-		    && kallsyms->symtab[i].st_value < nextval)
-			nextval = kallsyms->symtab[i].st_value;
+		if (thisval > addr
+		    && thisval < nextval)
+			nextval = thisval;
 	}
 
 	if (!best)
 		return NULL;
 
 	if (size)
-		*size = nextval - kallsyms->symtab[best].st_value;
+		*size = nextval - module_kallsyms_symbol_value(&kallsyms->symtab[best]);
 	if (offset)
-		*offset = addr - kallsyms->symtab[best].st_value;
+		*offset = addr - module_kallsyms_symbol_value(&kallsyms->symtab[best]);
 	return symname(kallsyms, best);
 }
 

@@ -4060,7 +4063,7 @@ int module_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
 			continue;
 		kallsyms = rcu_dereference_sched(mod->kallsyms);
 		if (symnum < kallsyms->num_symtab) {
-			*value = kallsyms->symtab[symnum].st_value;
+			*value = module_kallsyms_symbol_value(&kallsyms->symtab[symnum]);
 			*type = kallsyms->symtab[symnum].st_size;
 			strlcpy(name, symname(kallsyms, symnum), KSYM_NAME_LEN);
 			strlcpy(module_name, mod->name, MODULE_NAME_LEN);

@@ -4082,7 +4085,7 @@ static unsigned long mod_find_symname(struct module *mod, const char *name)
 	for (i = 0; i < kallsyms->num_symtab; i++)
 		if (strcmp(name, symname(kallsyms, i)) == 0 &&
 		    kallsyms->symtab[i].st_shndx != SHN_UNDEF)
-			return kallsyms->symtab[i].st_value;
+			return module_kallsyms_symbol_value(&kallsyms->symtab[i]);
 	return 0;
 }
 

@@ -4131,8 +4134,8 @@ int module_kallsyms_on_each_symbol(int (*fn)(void *, const char *,
 			if (kallsyms->symtab[i].st_shndx == SHN_UNDEF)
 				continue;
 
-			ret = fn(data, symname(kallsyms, i),
-				 mod, kallsyms->symtab[i].st_value);
+			ret = fn(data, symname(kallsyms, i), mod,
+				 module_kallsyms_symbol_value(&kallsyms->symtab[i]));
 			if (ret != 0)
 				return ret;
 		}

-- 
2.11.0

On Mon, Nov 19, 2018 at 05:25:12PM +0100, Vincent Whitchurch wrote:

This is fine if st_size is really unused, but how sure are you of that?

grepping for st_size throws up some hits that appear ELF-related, but
I've not investigated them in detail.

(The fact that struct stat has an identically named field throws up
a load of false positives too.)

Cheers
---Dave

On Thu, Nov 22, 2018 at 12:01:54PM +0000, Dave Martin wrote:

On Mon, Nov 19, 2018 at 05:25:12PM +0100, Vincent Whitchurch wrote:

quoted

st_info is currently overwritten after relocation and used to store the
elf_type().  However, we're going to need it fix kallsyms on ARM's
Thumb-2 kernels, so preserve st_info and overwrite the st_size field
instead.  st_size is neither used by the module core nor by any
architecture.

Signed-off-by: Vincent Whitchurch <redacted>
---
v4: Split out to separate patch.  Use st_size instead of st_other.
v1-v3: See PATCH 2/2

 kernel/module.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/kernel/module.c b/kernel/module.c
index 49a405891587..3d86a38b580c 100644
--- a/kernel/module.c
+++ b/kernel/module.c

@@ -2682,7 +2682,7 @@ static void add_kallsyms(struct module *mod, const struct load_info *info)
 
 	/* Set types up while we still have access to sections. */
 	for (i = 0; i < mod->kallsyms->num_symtab; i++)
-		mod->kallsyms->symtab[i].st_info
+		mod->kallsyms->symtab[i].st_size
 			= elf_type(&mod->kallsyms->symtab[i], info);
 
 	/* Now populate the cut down core kallsyms for after init. */

@@ -4061,7 +4061,7 @@ int module_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
 		kallsyms = rcu_dereference_sched(mod->kallsyms);
 		if (symnum < kallsyms->num_symtab) {
 			*value = kallsyms->symtab[symnum].st_value;
-			*type = kallsyms->symtab[symnum].st_info;
+			*type = kallsyms->symtab[symnum].st_size;
 			strlcpy(name, symname(kallsyms, symnum), KSYM_NAME_LEN);
 			strlcpy(module_name, mod->name, MODULE_NAME_LEN);
 			*exported = is_exported(name, *value, mod);

This is fine if st_size is really unused, but how sure are you of that?

grepping for st_size throws up some hits that appear ELF-related, but
I've not investigated them in detail.

(The fact that struct stat has an identically named field throws up
a load of false positives too.)

$ git describe --tags
v4.20-rc3-93-g92b419289cee
 
$ rg -m1 '[\.>]st_size'  --iglob '!**/tools/**' --iglob '!**/vdso*' --iglob '!**/scripts/**' --iglob '!**/usr/**' --iglob '!**/samples/**' | cat
| kernel/kexec_file.c:	if (sym->st_size != size) {

Symbols in kexec kernel.

| fs/stat.c:	tmp.st_size = stat->size;
| Documentation/networking/tls.txt:  sendfile(sock, file, &offset, stat.st_size);
| net/9p/client.c:		ret->st_rdev, ret->st_size, ret->st_blksize,
| net/9p/protocol.c:					&stbuf->st_rdev, &stbuf->st_size,
| fs/9p/vfs_inode_dotl.c:		i_size_write(inode, stat->st_size);
| fs/hostfs/hostfs_user.c:	p->size = buf->st_size;
| arch/powerpc/boot/mktree.c:	nblks = (st.st_size + IMGBLK) / IMGBLK;
| arch/alpha/kernel/osf_sys.c:	tmp.st_size	= lstat->size;
| arch/x86/ia32/sys_ia32.c:	    __put_user(stat->size, &ubuf->st_size) ||

Not Elf_Sym.

| arch/x86/kernel/machine_kexec_64.c:			 sym->st_size);

Symbols in kexec kernel.

| arch/sparc/boot/piggyback.c:	st4(buffer + 12, s.st_size);
| arch/sparc/kernel/sys_sparc32.c:	err |= put_user(stat->size, &statbuf->st_size);
| arch/um/os-Linux/file.c:		.ust_size    = src->st_size,    /* total size, in bytes */
| arch/um/os-Linux/start_up.c:	size = (buf.st_size + UM_KERN_PAGE_SIZE) & ~(UM_KERN_PAGE_SIZE - 1);
| arch/s390/kernel/compat_linux.c:	tmp.st_size = stat->size;
| arch/arm/kernel/sys_oabi-compat.c:	tmp.st_size = stat->size;
| arch/mips/boot/compressed/calc_vmlinuz_load_addr.c:	vmlinux_size = (uint64_t)sb.st_size;
| drivers/net/ethernet/marvell/sky2.c:		hw->st_idx = RING_NEXT(hw->st_idx, hw->st_size);

Not Elf_Sym.

+++ Vincent Whitchurch [22/11/18 13:24 +0100]:

On Thu, Nov 22, 2018 at 12:01:54PM +0000, Dave Martin wrote:

quoted

On Mon, Nov 19, 2018 at 05:25:12PM +0100, Vincent Whitchurch wrote:

quoted

st_info is currently overwritten after relocation and used to store the
elf_type().  However, we're going to need it fix kallsyms on ARM's
Thumb-2 kernels, so preserve st_info and overwrite the st_size field
instead.  st_size is neither used by the module core nor by any
architecture.

Signed-off-by: Vincent Whitchurch <redacted>
---
v4: Split out to separate patch.  Use st_size instead of st_other.
v1-v3: See PATCH 2/2

 kernel/module.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/kernel/module.c b/kernel/module.c
index 49a405891587..3d86a38b580c 100644
--- a/kernel/module.c
+++ b/kernel/module.c

@@ -2682,7 +2682,7 @@ static void add_kallsyms(struct module *mod, const struct load_info *info)

 	/* Set types up while we still have access to sections. */
 	for (i = 0; i < mod->kallsyms->num_symtab; i++)
-		mod->kallsyms->symtab[i].st_info
+		mod->kallsyms->symtab[i].st_size
 			= elf_type(&mod->kallsyms->symtab[i], info);

 	/* Now populate the cut down core kallsyms for after init. */

@@ -4061,7 +4061,7 @@ int module_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
 		kallsyms = rcu_dereference_sched(mod->kallsyms);
 		if (symnum < kallsyms->num_symtab) {
 			*value = kallsyms->symtab[symnum].st_value;
-			*type = kallsyms->symtab[symnum].st_info;
+			*type = kallsyms->symtab[symnum].st_size;
 			strlcpy(name, symname(kallsyms, symnum), KSYM_NAME_LEN);
 			strlcpy(module_name, mod->name, MODULE_NAME_LEN);
 			*exported = is_exported(name, *value, mod);

This is fine if st_size is really unused, but how sure are you of that?

grepping for st_size throws up some hits that appear ELF-related, but
I've not investigated them in detail.

(The fact that struct stat has an identically named field throws up
a load of false positives too.)

$ git describe --tags
v4.20-rc3-93-g92b419289cee

$ rg -m1 '[\.>]st_size'  --iglob '!**/tools/**' --iglob '!**/vdso*' --iglob '!**/scripts/**' --iglob '!**/usr/**' --iglob '!**/samples/**' | cat
| kernel/kexec_file.c:	if (sym->st_size != size) {

Symbols in kexec kernel.

| fs/stat.c:	tmp.st_size = stat->size;
| Documentation/networking/tls.txt:  sendfile(sock, file, &offset, stat.st_size);
| net/9p/client.c:		ret->st_rdev, ret->st_size, ret->st_blksize,
| net/9p/protocol.c:					&stbuf->st_rdev, &stbuf->st_size,
| fs/9p/vfs_inode_dotl.c:		i_size_write(inode, stat->st_size);
| fs/hostfs/hostfs_user.c:	p->size = buf->st_size;
| arch/powerpc/boot/mktree.c:	nblks = (st.st_size + IMGBLK) / IMGBLK;
| arch/alpha/kernel/osf_sys.c:	tmp.st_size	= lstat->size;
| arch/x86/ia32/sys_ia32.c:	    __put_user(stat->size, &ubuf->st_size) ||

Not Elf_Sym.

| arch/x86/kernel/machine_kexec_64.c:			 sym->st_size);

Symbols in kexec kernel.

| arch/sparc/boot/piggyback.c:	st4(buffer + 12, s.st_size);
| arch/sparc/kernel/sys_sparc32.c:	err |= put_user(stat->size, &statbuf->st_size);
| arch/um/os-Linux/file.c:		.ust_size    = src->st_size,    /* total size, in bytes */
| arch/um/os-Linux/start_up.c:	size = (buf.st_size + UM_KERN_PAGE_SIZE) & ~(UM_KERN_PAGE_SIZE - 1);
| arch/s390/kernel/compat_linux.c:	tmp.st_size = stat->size;
| arch/arm/kernel/sys_oabi-compat.c:	tmp.st_size = stat->size;
| arch/mips/boot/compressed/calc_vmlinuz_load_addr.c:	vmlinux_size = (uint64_t)sb.st_size;
| drivers/net/ethernet/marvell/sky2.c:		hw->st_idx = RING_NEXT(hw->st_idx, hw->st_size);

Not Elf_Sym.

[ added Miroslav to CC, just in case he would like to check :) ]

I have just double checked as well, and am fairly certain that the
Elf_Sym st_size field is not used to apply module relocations in any
arches, and it is not used in the core module loader nor in the module
kallsyms code. We'd like to avoid overwriting st_info in any case, to
fix kallsyms on Thumb-2 and also so that livepatch won't run into any
issues with delayed relocations, should livepatch support ever expand
to arches (e.g., arm) that rely on st_info for module relocations.

Thanks,

Jessica

On Thu, 22 Nov 2018 at 17:29, Jessica Yu [off-list ref] wrote:

Also note that st_size cannot be relied upon in general, since we
overwrite the addresses of undefined symbols in a module's symbol
table when resolve them against ksyms, while the st_size field is kept
at 0. At relocation time, we don't really distinguish anymore between
local and external module symbols, and so relying on st_size to be
accurate would definitely break things.

On Thu, Nov 22, 2018 at 06:40:45PM +0100, Ard Biesheuvel wrote:

On Thu, 22 Nov 2018 at 17:29, Jessica Yu [off-list ref] wrote:

quoted

+++ Vincent Whitchurch [22/11/18 13:24 +0100]:

quoted

On Thu, Nov 22, 2018 at 12:01:54PM +0000, Dave Martin wrote:

quoted

On Mon, Nov 19, 2018 at 05:25:12PM +0100, Vincent Whitchurch wrote:

quoted

st_info is currently overwritten after relocation and used to store the
elf_type().  However, we're going to need it fix kallsyms on ARM's
Thumb-2 kernels, so preserve st_info and overwrite the st_size field
instead.  st_size is neither used by the module core nor by any
architecture.

Signed-off-by: Vincent Whitchurch <redacted>
---
v4: Split out to separate patch.  Use st_size instead of st_other.
v1-v3: See PATCH 2/2

 kernel/module.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/kernel/module.c b/kernel/module.c
index 49a405891587..3d86a38b580c 100644
--- a/kernel/module.c
+++ b/kernel/module.c

@@ -2682,7 +2682,7 @@ static void add_kallsyms(struct module *mod, const struct load_info *info)

   /* Set types up while we still have access to sections. */
   for (i = 0; i < mod->kallsyms->num_symtab; i++)
-          mod->kallsyms->symtab[i].st_info
+          mod->kallsyms->symtab[i].st_size
                   = elf_type(&mod->kallsyms->symtab[i], info);

   /* Now populate the cut down core kallsyms for after init. */

@@ -4061,7 +4061,7 @@ int module_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
           kallsyms = rcu_dereference_sched(mod->kallsyms);
           if (symnum < kallsyms->num_symtab) {
                   *value = kallsyms->symtab[symnum].st_value;
-                  *type = kallsyms->symtab[symnum].st_info;
+                  *type = kallsyms->symtab[symnum].st_size;
                   strlcpy(name, symname(kallsyms, symnum), KSYM_NAME_LEN);
                   strlcpy(module_name, mod->name, MODULE_NAME_LEN);
                   *exported = is_exported(name, *value, mod);

This is fine if st_size is really unused, but how sure are you of that?

grepping for st_size throws up some hits that appear ELF-related, but
I've not investigated them in detail.

(The fact that struct stat has an identically named field throws up
a load of false positives too.)

$ git describe --tags
v4.20-rc3-93-g92b419289cee

$ rg -m1 '[\.>]st_size'  --iglob '!**/tools/**' --iglob '!**/vdso*' --iglob '!**/scripts/**' --iglob '!**/usr/**' --iglob '!**/samples/**' | cat
| kernel/kexec_file.c: if (sym->st_size != size) {

Symbols in kexec kernel.

| fs/stat.c:   tmp.st_size = stat->size;
| Documentation/networking/tls.txt:  sendfile(sock, file, &offset, stat.st_size);
| net/9p/client.c:             ret->st_rdev, ret->st_size, ret->st_blksize,
| net/9p/protocol.c:                                   &stbuf->st_rdev, &stbuf->st_size,
| fs/9p/vfs_inode_dotl.c:              i_size_write(inode, stat->st_size);
| fs/hostfs/hostfs_user.c:     p->size = buf->st_size;
| arch/powerpc/boot/mktree.c:  nblks = (st.st_size + IMGBLK) / IMGBLK;
| arch/alpha/kernel/osf_sys.c: tmp.st_size     = lstat->size;
| arch/x86/ia32/sys_ia32.c:        __put_user(stat->size, &ubuf->st_size) ||

Not Elf_Sym.

| arch/x86/kernel/machine_kexec_64.c:                   sym->st_size);

Symbols in kexec kernel.

| arch/sparc/boot/piggyback.c: st4(buffer + 12, s.st_size);
| arch/sparc/kernel/sys_sparc32.c:     err |= put_user(stat->size, &statbuf->st_size);
| arch/um/os-Linux/file.c:             .ust_size    = src->st_size,    /* total size, in bytes */
| arch/um/os-Linux/start_up.c: size = (buf.st_size + UM_KERN_PAGE_SIZE) & ~(UM_KERN_PAGE_SIZE - 1);
| arch/s390/kernel/compat_linux.c:     tmp.st_size = stat->size;
| arch/arm/kernel/sys_oabi-compat.c:   tmp.st_size = stat->size;
| arch/mips/boot/compressed/calc_vmlinuz_load_addr.c:  vmlinux_size = (uint64_t)sb.st_size;
| drivers/net/ethernet/marvell/sky2.c:         hw->st_idx = RING_NEXT(hw->st_idx, hw->st_size);

Not Elf_Sym.

[ added Miroslav to CC, just in case he would like to check :) ]

I have just double checked as well, and am fairly certain that the
Elf_Sym st_size field is not used to apply module relocations in any
arches, and it is not used in the core module loader nor in the module
kallsyms code. We'd like to avoid overwriting st_info in any case, to
fix kallsyms on Thumb-2 and also so that livepatch won't run into any
issues with delayed relocations, should livepatch support ever expand
to arches (e.g., arm) that rely on st_info for module relocations.

Also note that st_size cannot be relied upon in general, since we
overwrite the addresses of undefined symbols in a module's symbol
table when resolve them against ksyms, while the st_size field is kept
at 0. At relocation time, we don't really distinguish anymore between
local and external module symbols, and so relying on st_size to be
accurate would definitely break things.

Umm.

Undefined symbols of course have a zero size, because it's not known
how large the definition that the symbol refers to is.  However,
the non-undefined symbols should have a size with them - anything
emitted by the compiler should, but only if .size has been used in
the assembler will assembly have correct sizes.

Aren't SHN_UNDEF symbols ignored, except when applying the
relocations, where we rely on st_value to be set on SHN_UNDEF
symbols?

-- 
RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 12.1Mbps down 622kbps up
According to speedtest.net: 11.9Mbps down 500kbps up

On Thu, 22 Nov 2018, Jessica Yu wrote:

I checked and came to the same conclusion.

Regards,
Miroslav

On Thu, Nov 22, 2018 at 05:49:23PM +0000, Russell King - ARM Linux wrote:

On Thu, Nov 22, 2018 at 06:40:45PM +0100, Ard Biesheuvel wrote:

quoted

On Thu, 22 Nov 2018 at 17:29, Jessica Yu [off-list ref] wrote:

quoted

+++ Vincent Whitchurch [22/11/18 13:24 +0100]:

quoted

On Thu, Nov 22, 2018 at 12:01:54PM +0000, Dave Martin wrote:

quoted

On Mon, Nov 19, 2018 at 05:25:12PM +0100, Vincent Whitchurch wrote:

quoted

st_info is currently overwritten after relocation and used to store the
elf_type().  However, we're going to need it fix kallsyms on ARM's
Thumb-2 kernels, so preserve st_info and overwrite the st_size field
instead.  st_size is neither used by the module core nor by any
architecture.

Signed-off-by: Vincent Whitchurch <redacted>
---
v4: Split out to separate patch.  Use st_size instead of st_other.
v1-v3: See PATCH 2/2

 kernel/module.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/kernel/module.c b/kernel/module.c
index 49a405891587..3d86a38b580c 100644
--- a/kernel/module.c
+++ b/kernel/module.c

@@ -2682,7 +2682,7 @@ static void add_kallsyms(struct module *mod, const struct load_info *info)

   /* Set types up while we still have access to sections. */
   for (i = 0; i < mod->kallsyms->num_symtab; i++)
-          mod->kallsyms->symtab[i].st_info
+          mod->kallsyms->symtab[i].st_size
                   = elf_type(&mod->kallsyms->symtab[i], info);

   /* Now populate the cut down core kallsyms for after init. */

@@ -4061,7 +4061,7 @@ int module_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
           kallsyms = rcu_dereference_sched(mod->kallsyms);
           if (symnum < kallsyms->num_symtab) {
                   *value = kallsyms->symtab[symnum].st_value;
-                  *type = kallsyms->symtab[symnum].st_info;
+                  *type = kallsyms->symtab[symnum].st_size;
                   strlcpy(name, symname(kallsyms, symnum), KSYM_NAME_LEN);
                   strlcpy(module_name, mod->name, MODULE_NAME_LEN);
                   *exported = is_exported(name, *value, mod);

This is fine if st_size is really unused, but how sure are you of that?

grepping for st_size throws up some hits that appear ELF-related, but
I've not investigated them in detail.

(The fact that struct stat has an identically named field throws up
a load of false positives too.)

$ git describe --tags
v4.20-rc3-93-g92b419289cee

$ rg -m1 '[\.>]st_size'  --iglob '!**/tools/**' --iglob '!**/vdso*' --iglob '!**/scripts/**' --iglob '!**/usr/**' --iglob '!**/samples/**' | cat
| kernel/kexec_file.c: if (sym->st_size != size) {

Symbols in kexec kernel.

| fs/stat.c:   tmp.st_size = stat->size;
| Documentation/networking/tls.txt:  sendfile(sock, file, &offset, stat.st_size);
| net/9p/client.c:             ret->st_rdev, ret->st_size, ret->st_blksize,
| net/9p/protocol.c:                                   &stbuf->st_rdev, &stbuf->st_size,
| fs/9p/vfs_inode_dotl.c:              i_size_write(inode, stat->st_size);
| fs/hostfs/hostfs_user.c:     p->size = buf->st_size;
| arch/powerpc/boot/mktree.c:  nblks = (st.st_size + IMGBLK) / IMGBLK;
| arch/alpha/kernel/osf_sys.c: tmp.st_size     = lstat->size;
| arch/x86/ia32/sys_ia32.c:        __put_user(stat->size, &ubuf->st_size) ||

Not Elf_Sym.

| arch/x86/kernel/machine_kexec_64.c:                   sym->st_size);

Symbols in kexec kernel.

| arch/sparc/boot/piggyback.c: st4(buffer + 12, s.st_size);
| arch/sparc/kernel/sys_sparc32.c:     err |= put_user(stat->size, &statbuf->st_size);
| arch/um/os-Linux/file.c:             .ust_size    = src->st_size,    /* total size, in bytes */
| arch/um/os-Linux/start_up.c: size = (buf.st_size + UM_KERN_PAGE_SIZE) & ~(UM_KERN_PAGE_SIZE - 1);
| arch/s390/kernel/compat_linux.c:     tmp.st_size = stat->size;
| arch/arm/kernel/sys_oabi-compat.c:   tmp.st_size = stat->size;
| arch/mips/boot/compressed/calc_vmlinuz_load_addr.c:  vmlinux_size = (uint64_t)sb.st_size;
| drivers/net/ethernet/marvell/sky2.c:         hw->st_idx = RING_NEXT(hw->st_idx, hw->st_size);

Not Elf_Sym.

[ added Miroslav to CC, just in case he would like to check :) ]

I have just double checked as well, and am fairly certain that the
Elf_Sym st_size field is not used to apply module relocations in any
arches, and it is not used in the core module loader nor in the module
kallsyms code. We'd like to avoid overwriting st_info in any case, to
fix kallsyms on Thumb-2 and also so that livepatch won't run into any
issues with delayed relocations, should livepatch support ever expand
to arches (e.g., arm) that rely on st_info for module relocations.

Also note that st_size cannot be relied upon in general, since we
overwrite the addresses of undefined symbols in a module's symbol
table when resolve them against ksyms, while the st_size field is kept
at 0. At relocation time, we don't really distinguish anymore between
local and external module symbols, and so relying on st_size to be
accurate would definitely break things.

Umm.

Undefined symbols of course have a zero size, because it's not known
how large the definition that the symbol refers to is.  However,
the non-undefined symbols should have a size with them - anything
emitted by the compiler should, but only if .size has been used in
the assembler will assembly have correct sizes.

which I guess is one reason st_size is a bit unreliable.

For kallsyms purposes we assume that the size of each symbol is just the
offset from it to the next symbol, which seems good enough in practice.

Aren't SHN_UNDEF symbols ignored, except when applying the
relocations, where we rely on st_value to be set on SHN_UNDEF
symbols?

Presumably, but it looks like st_size is pretty much don't care.

I renamed st_size in the struct in elf.h to a garbage name and the
kernel and modules still build fine, though I only tried this for arm64
so far.

This suggests that at runtime at least, there is no runtime usage of
this field at all unless it's in arch-specific code for other arches.
Others seem confident that it's not used anywhere, and although I've not
gone through all the grep hits, I've only encountered false positives so
far.

Cheers
---Dave

On Mon, Nov 19, 2018 at 05:25:12PM +0100, Vincent Whitchurch wrote:

Based on the discussion here:

Reviewed-by: Dave Martin <Dave.Martin@arm.com>

(I still think this is a bit gross, but without resorting to overkill
there's not an obvious nicer option.  Even before this patch, we were
abusing st_info.  Of course, someday somebody may find a use for
st_size and then we'd need to revisit this code...)

Cheers
---Dave

On Mon, Nov 19, 2018 at 05:25:13PM +0100, Vincent Whitchurch wrote:

Nit: this fits easily on one line now.

 			best = i;

Can we declare bestval outside the loop and update it here, so that
we always have
bestval == module_kallsyms_symbol_value(&kallsyms->symtab[best]) ?

Then we wouldn't need to call module_kallsyms_symbol_value() again
afterwards at [1], [2] below.

-		if (kallsyms->symtab[i].st_value > addr
-		    && kallsyms->symtab[i].st_value < nextval)
-			nextval = kallsyms->symtab[i].st_value;
+		if (thisval > addr
+		    && thisval < nextval)
+			nextval = thisval;

Nit: same again.

 	}
 
 	if (!best)
 		return NULL;
 
 	if (size)
-		*size = nextval - kallsyms->symtab[best].st_value;
+		*size = nextval - module_kallsyms_symbol_value(&kallsyms->symtab[best]);

[1]

 	if (offset)
-		*offset = addr - kallsyms->symtab[best].st_value;
+		*offset = addr - module_kallsyms_symbol_value(&kallsyms->symtab[best]);

[2]

Nit: We have some more overlong lines throughout this file now, which is
mildly annoying (though hardly a big deal).

In may places the expression kallsyms->symtab[foo] appears multiple
times and adds to verbosity.

Is it worth stashing the pointer first?  For example, in this case:

			const Elf_Sym *sym = &kallsyms->symtab[i];

			if (sym->st_shndx == SHN_UNDEF)
				continue;

			ret = fn(data, symname(kallsyms, i), mod,
				module_kallsyms_symbol_value(sym));

This adds two lines in the diffstat of course.  Take your pick!

Cheers
---Dave