mtd: Avoid boot crash in RedBoot partition table parser
Subsystems (2)
PATCH 1/1 mtd: Avoid boot crash in RedBoot partition table parser
Given CONFIG_FORTIFY_SOURCE=y, and given a recent compiler, commit 439a1bcac648 ("fortify: Use __builtin_dynamic_object_size() when available") produces the warning below and an oops.
Searching for RedBoot partition table in 50000000.flash at offset 0x7e0000
------------[ cut here ]------------
WARNING: lib/string_helpers.c:1035 at 0xc029e04c, CPU#0: swapper/0/1
memcmp: detected buffer overflow: 15 byte read of buffer size 14
Modules linked in:
CPU: 0 UID: 0 PID: 1 Comm: swapper/0 Not tainted 6.19.0 #1 NONEI couldn't see how memcmp() exceeds the buffer here, so the simplest way to prevent the regression was to perform memcmp() on the original name rather than the copy.
---
I put commit 439a1bcac648 into a Fixes tag because git bisect identified that commit as the source of the regression. But I don't know anything about __builtin_dynamic_object_size() or its limitations. So perhaps the real bug lies elsewhere. The compiler I'm using is this one:
$ armeb-softfloat-linux-musleabi-gcc --version armeb-softfloat-linux-musleabi-gcc (Gentoo Hardened 13.4.1_p20250807 p8) 13.4.1 20250807 --- drivers/mtd/parsers/redboot.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)
drivers/mtd/parsers/redboot.c +5 -5
@@ -269,14 +269,14 @@ static int parse_redboot_partitions(struct mtd_info *master, parts[i].name = names;strcpy (names, fl->img->name); + names +=strlen (names) +1 ; +#ifdef CONFIG_MTD_REDBOOT_PARTS_READONLY -if (!memcmp (names,"RedBoot" ,8 ) || - !memcmp (names,"RedBoot config" ,15 ) || - !memcmp (names,"FIS directory" ,14 )) { +if (!memcmp (fl->img->name,"RedBoot" ,8 ) || + !memcmp (fl->img->name,"RedBoot config" ,15 ) || + !memcmp (fl->img->name,"FIS directory" ,14 )) parts[i].mask_flags =MTD_WRITEABLE ; - }#endif - names +=strlen (names) +1 ;#ifdef CONFIG_MTD_REDBOOT_PARTS_UNALLOCATED if (fl->next && fl->img->flash_base + fl->img->size + master->erasesize <= fl->next->img->flash_base) {
Signature · 1 line
2.49.1
Review
Quoted 3 lines · Finn Thain
memcmp: detected buffer overflow: 15 byte read of buffer size 14
[...](quoted diff) +0
-1- !memcmp(names, "RedBoot config", 15) ||
The warning is saying that "names" is detected to be 14 bytes in size. It is allocated here, and nulllen can be ignored (it is fixed size, either 0 or sizeof(nullstring), and skipped over):
parts = kzalloc(sizeof(*parts) * nrparts + nulllen + namelen, GFP_KERNEL);
...
nullname = (char *)&parts[nrparts];
...
names = nullname + nulllen;so "names" is pointing to the final "namelen" many bytes of the allocation. Calculating "namelen" happens via an earlier for loop:
buf = vmalloc(master->erasesize);
...
ret = mtd_read(master, offset, master->erasesize, &retlen,
(void *)buf);
...
numslots = (master->erasesize / sizeof(struct fis_image_desc));
...
for (i = 0; i < numslots; i++) {
...
namelen += strlen(buf[i].name) + 1;So namelen could be basically any length at all. This fortify warning looks legit to me -- this code used to be reading beyond the end of the allocation.
Your patch looks technically correct, but why not just use strcmp? Both arguments are NUL-terminated. The memcmp() calls were all including the NUL byte, so they're effectively doing strcmp except that they weren't stopping at the first NUL byte. So probably just better to do:
drivers/mtd/parsers/redboot.c +3 -3
@@ -270,9 +270,9 @@ static int parse_redboot_partitions(struct mtd_info *master,strcpy (names, fl->img->name);#ifdef CONFIG_MTD_REDBOOT_PARTS_READONLY -if (!memcmp (names,"RedBoot" ,8 ) || - !memcmp (names,"RedBoot config" ,15 ) || - !memcmp (names,"FIS directory" ,14 )) { +if (!strcmp (names,"RedBoot" ) || + !strcmp (names,"RedBoot config" ) || + !strcmp (names,"FIS directory" )) { parts[i].mask_flags =MTD_WRITEABLE ; }#endif
Signature · 1 line
Kees Cook