[PATCH v5 0/2] fs/binfmt_elf: Fix AT_PHDR for unusual ELF files

[PATCH v5 0/2] fs/binfmt_elf: Fix AT_PHDR for unusual ELF files

[Akira Kawata]

These patches fix a bug in AT_PHDR calculation. 

We cannot calculate AT_PHDR as the sum of load_addr and exec->e_phoff.
This is because exec->e_phoff is the offset of PHDRs in the file and the
address of PHDRs in the memory may differ from it. These patches fix the
bug by calculating the address of program headers from PT_LOADs
directly.

Changes in v5
- Reflecting on comments from Kees, add a comment to the first commit.

Changes in v4
- Reflecting on comments from Lukas, add a refactoring commit.

Changes in v3:
- Fix a reported bug from kernel test robot.

Changes in v2:
- Remove unused load_addr from create_elf_tables.
- Improve the commit message.

Akira Kawata (2):
  fs/binfmt_elf: Fix AT_PHDR for unusual ELF files
  fs/binfmt_elf: Refactor load_elf_binary function

 fs/binfmt_elf.c | 40 +++++++++++++++++++++++++---------------
 1 file changed, 25 insertions(+), 15 deletions(-)


base-commit: 0eb96e2c58c03e79fc2ee833ba88bf9226986564
-- 
2.25.1

[PATCH v5 1/2] fs/binfmt_elf: Fix AT_PHDR for unusual ELF files

[Akira Kawata]

BugLink: https://bugzilla.kernel.org/show_bug.cgi?id=197921

As pointed out in the discussion of buglink, we cannot calculate AT_PHDR
as the sum of load_addr and exec->e_phoff.

: The AT_PHDR of ELF auxiliary vectors should point to the memory address
: of program header. But binfmt_elf.c calculates this address as follows:
:
: NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
:
: which is wrong since e_phoff is the file offset of program header and
: load_addr is the memory base address from PT_LOAD entry.
:
: The ld.so uses AT_PHDR as the memory address of program header. In normal
: case, since the e_phoff is usually 64 and in the first PT_LOAD region, it
: is the correct program header address.
:
: But if the address of program header isn't equal to the first PT_LOAD
: address + e_phoff (e.g.  Put the program header in other non-consecutive
: PT_LOAD region), ld.so will try to read program header from wrong address
: then crash or use incorrect program header.

This is because exec->e_phoff
is the offset of PHDRs in the file and the address of PHDRs in the
memory may differ from it. This patch fixes the bug by calculating the
address of program headers from PT_LOADs directly.

Signed-off-by: Akira Kawata <akirakawata1@gmail.com>
Reported-by: kernel test robot <lkp@intel.com>
Acked-by: Kees Cook <keescook@chromium.org>
---
 fs/binfmt_elf.c | 24 ++++++++++++++++++------
 1 file changed, 18 insertions(+), 6 deletions(-)

diff --git a/fs/binfmt_elf.c b/fs/binfmt_elf.c
index 0a25b8049b74..d120ab03795f 100644
--- a/fs/binfmt_elf.c
+++ b/fs/binfmt_elf.c
@@ -170,8 +170,8 @@ static int padzero(unsigned long elf_bss)
 
 static int
 create_elf_tables(struct linux_binprm *bprm, const struct elfhdr *exec,
-		unsigned long load_addr, unsigned long interp_load_addr,
-		unsigned long e_entry)
+		unsigned long interp_load_addr,
+		unsigned long e_entry, unsigned long phdr_addr)
 {
 	struct mm_struct *mm = current->mm;
 	unsigned long p = bprm->p;
@@ -257,7 +257,7 @@ create_elf_tables(struct linux_binprm *bprm, const struct elfhdr *exec,
 	NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
 	NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
 	NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
-	NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
+	NEW_AUX_ENT(AT_PHDR, phdr_addr);
 	NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
 	NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
 	NEW_AUX_ENT(AT_BASE, interp_load_addr);
@@ -822,7 +822,7 @@ static int parse_elf_properties(struct file *f, const struct elf_phdr *phdr,
 static int load_elf_binary(struct linux_binprm *bprm)
 {
 	struct file *interpreter = NULL; /* to shut gcc up */
- 	unsigned long load_addr = 0, load_bias = 0;
+	unsigned long load_addr, load_bias = 0, phdr_addr = 0;
 	int load_addr_set = 0;
 	unsigned long error;
 	struct elf_phdr *elf_ppnt, *elf_phdata, *interp_elf_phdata = NULL;
@@ -1168,6 +1168,17 @@ static int load_elf_binary(struct linux_binprm *bprm)
 				reloc_func_desc = load_bias;
 			}
 		}
+
+		/*
+		 * Figure out which segment in the file contains the Program
+		 * Header table, and map to the associated memory address.
+		 */
+		if (elf_ppnt->p_offset <= elf_ex->e_phoff &&
+		    elf_ex->e_phoff < elf_ppnt->p_offset + elf_ppnt->p_filesz) {
+			phdr_addr = elf_ex->e_phoff - elf_ppnt->p_offset +
+				    elf_ppnt->p_vaddr;
+		}
+
 		k = elf_ppnt->p_vaddr;
 		if ((elf_ppnt->p_flags & PF_X) && k < start_code)
 			start_code = k;
@@ -1203,6 +1214,7 @@ static int load_elf_binary(struct linux_binprm *bprm)
 	}
 
 	e_entry = elf_ex->e_entry + load_bias;
+	phdr_addr += load_bias;
 	elf_bss += load_bias;
 	elf_brk += load_bias;
 	start_code += load_bias;
@@ -1266,8 +1278,8 @@ static int load_elf_binary(struct linux_binprm *bprm)
 		goto out;
 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
 
-	retval = create_elf_tables(bprm, elf_ex,
-			  load_addr, interp_load_addr, e_entry);
+	retval = create_elf_tables(bprm, elf_ex, interp_load_addr,
+				   e_entry, phdr_addr);
 	if (retval < 0)
 		goto out;
 
-- 
2.25.1

[PATCH v5 2/2] fs/binfmt_elf: Refactor load_elf_binary function

[Akira Kawata]

I delete load_addr because it is not used anymore. And I rename
load_addr_set to first_pt_load because it is used only to capture the
first iteration of the loop.

Signed-off-by: Akira Kawata <akirakawata1@gmail.com>
Acked-by: Kees Cook <keescook@chromium.org>
---
 fs/binfmt_elf.c | 18 ++++++++----------
 1 file changed, 8 insertions(+), 10 deletions(-)

diff --git a/fs/binfmt_elf.c b/fs/binfmt_elf.c
index d120ab03795f..3218ebfde409 100644
--- a/fs/binfmt_elf.c
+++ b/fs/binfmt_elf.c
@@ -822,8 +822,8 @@ static int parse_elf_properties(struct file *f, const struct elf_phdr *phdr,
 static int load_elf_binary(struct linux_binprm *bprm)
 {
 	struct file *interpreter = NULL; /* to shut gcc up */
-	unsigned long load_addr, load_bias = 0, phdr_addr = 0;
-	int load_addr_set = 0;
+	unsigned long load_bias = 0, phdr_addr = 0;
+	int first_pt_load = 1;
 	unsigned long error;
 	struct elf_phdr *elf_ppnt, *elf_phdata, *interp_elf_phdata = NULL;
 	struct elf_phdr *elf_property_phdata = NULL;
@@ -1073,12 +1073,12 @@ static int load_elf_binary(struct linux_binprm *bprm)
 
 		vaddr = elf_ppnt->p_vaddr;
 		/*
-		 * The first time through the loop, load_addr_set is false:
+		 * The first time through the loop, first_pt_load is true:
 		 * layout will be calculated. Once set, use MAP_FIXED since
 		 * we know we've already safely mapped the entire region with
 		 * MAP_FIXED_NOREPLACE in the once-per-binary logic following.
 		 */
-		if (load_addr_set) {
+		if (!first_pt_load) {
 			elf_flags |= MAP_FIXED;
 		} else if (elf_ex->e_type == ET_EXEC) {
 			/*
@@ -1138,10 +1138,10 @@ static int load_elf_binary(struct linux_binprm *bprm)
 
 		/*
 		 * Calculate the entire size of the ELF mapping (total_size).
-		 * (Note that load_addr_set is set to true later once the
+		 * (Note that first_pt_load is set to false later once the
 		 * initial mapping is performed.)
 		 */
-		if (!load_addr_set) {
+		if (first_pt_load) {
 			total_size = total_mapping_size(elf_phdata,
 							elf_ex->e_phnum);
 			if (!total_size) {
@@ -1158,13 +1158,11 @@ static int load_elf_binary(struct linux_binprm *bprm)
 			goto out_free_dentry;
 		}
 
-		if (!load_addr_set) {
-			load_addr_set = 1;
-			load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
+		if (first_pt_load) {
+			first_pt_load = 0;
 			if (elf_ex->e_type == ET_DYN) {
 				load_bias += error -
 				             ELF_PAGESTART(load_bias + vaddr);
-				load_addr += load_bias;
 				reloc_func_desc = load_bias;
 			}
 		}
-- 
2.25.1