Fix md RAID enumeration
PATCH 1/1 Fix md RAID enumeration
In addition to the prior patch to avoid leaving holes in the returned devicelist, the enumeration logic also needed fixing. GET_ARRAY_INFO's info.nr_disks does not map to GET_DISK_INFO's disk.number, which is an internal kernel index. If an array has had drives added, removed, etc, there may be gaps in GET_DISK_INFO's results. Instead, do what mdadm does and keep scanning until we find enough valid (major/minor != 0) disks.
(This is intended to be applied on top of the earlier patch sent in
https://lists.gnu.org/archive/html/grub-devel/2021-01/msg00027.html)
grub-core/osdep/linux/getroot.c +14 -1
@@ -135,10 +135,18 @@ struct mountinfo_entrychar fstype[ESCAPED_PATH_MAX +1 ], device[ESCAPED_PATH_MAX +1 ]; }; +/* GET_ARRAY_INFO's info.nr_disks does not map to GET_DISK_INFO's +disk.number, which is an internal kernel index. Instead, do what mdadm +does and keep scanning until we find enough valid disks. The limit is +copied from there, which notes that it is sufficiently high given +that the on-disk metadata for v1.x can only support 1920. */ +#define MD_MAX_DISKS 4096 +static char **grub_util_raid_getmembers (const char *name,int bootable) {int fd, ret, i, j; +int remaining;char **devicelist;mdu_version_t version;mdu_array_info_t info;
@@ -170,12 +178,17 @@ grub_util_raid_getmembers (const char *n devicelist =xcalloc (info.nr_disks +1 ,sizeof (char *)); -for (i =0 , j =0 ; i < info.nr_disks; i++) + remaining = info.nr_disks; +for (i =0 , j =0 ; i <MD_MAX_DISKS && remaining >0 ; i++) { disk.number = i; ret =ioctl (fd,GET_DISK_INFO , &disk);if (ret !=0 )grub_util_error (_ ("ioctl GET_DISK_INFO error: %s" ),strerror (errno)); +/* Skip empty disk slot. */ +if (disk.major ==0 && disk.minor ==0 ) +continue ; + remaining--;if (disk.state & (1 <<MD_DISK_REMOVED ))continue ;
Signature · 1 line
Kees Cook @outflux.net
PATCH 1/1 osdep/linux: Fix md array device enumeration
GET_ARRAY_INFO's info.nr_disks does not map to GET_DISK_INFO's disk.number, which is an internal kernel index. If an array has had drives added, removed, etc, there may be gaps in GET_DISK_INFO's results. But since the consumer of devicelist cannot tolerate gaps (it expects to walk a NULL-terminated list of device name strings), the devicelist index (j) must be tracked separately from the disk.number index (i). But grub wants to only examine active (i.e. present and non-failed) disks, so how many disks are left to be queried must be also separately tracked (remaining).
---
Actually, here's a better patch which combines the two I sent...
--- grub-core/osdep/linux/getroot.c | 27 +++++++++++++++++++-------- 1 file changed, 19 insertions(+), 8 deletions(-)
grub-core/osdep/linux/getroot.c +19 -8
@@ -135,10 +135,18 @@ struct mountinfo_entrychar fstype[ESCAPED_PATH_MAX +1 ], device[ESCAPED_PATH_MAX +1 ]; }; +/* GET_DISK_INFO nr_disks (total count) does not map to disk.number, +which is an internal kernel index. Instead, do what mdadm does +and keep scanning until we find enough valid disks. The limit is +copied from there, which notes that it is sufficiently high given +that the on-disk metadata for v1.x can only support 1920. */ +#define MD_MAX_DISKS 4096 +static char **grub_util_raid_getmembers (const char *name,int bootable) {int fd, ret, i, j; +int remaining;char **devicelist;mdu_version_t version;mdu_array_info_t info;
@@ -170,22 +178,25 @@ grub_util_raid_getmembers (const char *name, int bootable) devicelist =xcalloc (info.nr_disks +1 ,sizeof (char *)); -for (i =0 , j =0 ; j < info.nr_disks; i++) + remaining = info.nr_disks; +for (i =0 , j =0 ; i <MD_MAX_DISKS && remaining >0 ; i++) { disk.number = i; ret =ioctl (fd,GET_DISK_INFO , &disk);if (ret !=0 )grub_util_error (_ ("ioctl GET_DISK_INFO error: %s" ),strerror (errno)); - + +/* Skip empty slot: MD_DISK_REMOVED slots don't count toward nr_disks. */ if (disk.state & (1 <<MD_DISK_REMOVED ))continue ; + remaining--; -if (disk.state & (1 <<MD_DISK_ACTIVE )) - devicelist[j] =grub_find_device (NULL , -makedev (disk.major, disk.minor)); -else - devicelist[j] =NULL ; - j++; +/* Only examine disks that are actively participating in the array. */ +if (!(disk.state & (1 <<MD_DISK_ACTIVE ))) +continue ; + + devicelist[j++] =grub_find_device (NULL ,makedev (disk.major, + disk.minor)); } devicelist[j] =NULL ;
Signature · 4 lines
2.25.1 -- Kees Cook