Re: Problem with boot block / softraid on current

[Stefan Sperling]

On Tue, Jan 03, 2017 at 07:37:10AM +0100, Andreas Bartelt wrote:
> One of my amd64 machines does not boot anymore after updating to current
> (attached dmesg was obtained after booting a build of current from today but
> with a boot block from December, 22nd). Interestingly, the same disk (with a
> boot block from today's build) still boots fine with another amd64 machine
> (an old x61s thinkpad).
> 
> The disk image makes use of a softraid(4) encrypted root partition. On the
> affected machine, it triggers a reboot directly after the password has been
> entered.
> 
> I could narrow the problem down to the use of installboot(8), i.e., booting
> current still works on the affected machine with a boot block based on my
> previous build from December, 22nd.

I ran into the same problem on my thinkpad x130e and tracked it down to
this part of the changes committed in softraid_amd64.c r1.3:

@@ -433,7 +435,7 @@
        const char               openbsd_uuid_code[] = GPT_UUID_OPENBSD;
        struct gpt_partition     gp;
        static struct uuid      *openbsd_uuid = NULL, openbsd_uuid_space;
-       static u_char            buf[DEV_BSIZE];
+       static u_char            buf[4096];


So the problem seems to be that boot uses too much stack memory.

The diff below fixes the problem for me. As far as I can tell this
should work on both 512 and 4k disks. But I cannot test with a 4k disk.

Index: softraid_amd64.c
===================================================================
RCS file: /cvs/src/sys/arch/amd64/stand/libsa/softraid_amd64.c,v
retrieving revision 1.3
diff -u -p -r1.3 softraid_amd64.c
--- softraid_amd64.c    24 Dec 2016 22:49:38 -0000      1.3
+++ softraid_amd64.c    4 Jan 2017 18:14:16 -0000
@@ -435,7 +435,7 @@ findopenbsd_gpt(struct sr_boot_volume *b
        const char               openbsd_uuid_code[] = GPT_UUID_OPENBSD;
        struct gpt_partition     gp;
        static struct uuid      *openbsd_uuid = NULL, openbsd_uuid_space;
-       static u_char            buf[4096];
+       u_char                  *buf;
 
        /* Prepare OpenBSD UUID */
        if (openbsd_uuid == NULL) {
@@ -456,6 +456,8 @@ findopenbsd_gpt(struct sr_boot_volume *b
                *err = "disk sector > 4096 bytes\n";
                return (-1);
        }
+       buf = alloc(bv->sbv_secsize);
+       bzero(buf, bv->sbv_secsize);
 
        /* LBA1: GPT Header */
        lba = 1;
@@ -466,17 +468,20 @@ findopenbsd_gpt(struct sr_boot_volume *b
        /* Check signature */
        if (letoh64(gh.gh_sig) != GPTSIGNATURE) {
                *err = "bad GPT signature\n";
+               free(buf, bv->sbv_secsize);
                return (-1);
        }
 
        if (letoh32(gh.gh_rev) != GPTREVISION) {
                *err = "bad GPT revision\n";
+               free(buf, bv->sbv_secsize);
                return (-1);
        }
 
        ghsize = letoh32(gh.gh_size);
        if (ghsize < GPTMINHDRSIZE || ghsize > sizeof(struct gpt_header)) {
                *err = "bad GPT header size\n";
+               free(buf, bv->sbv_secsize);
                return (-1);
        }
 
@@ -487,6 +492,7 @@ findopenbsd_gpt(struct sr_boot_volume *b
        gh.gh_csum = orig_csum;
        if (letoh32(orig_csum) != new_csum) {
                *err = "bad GPT header checksum\n";
+               free(buf, bv->sbv_secsize);
                return (-1);
        }
 
@@ -514,6 +520,9 @@ findopenbsd_gpt(struct sr_boot_volume *b
                                found = 1;
                }
        }
+
+       free(buf, bv->sbv_secsize);
+
        if (new_csum != letoh32(gh.gh_part_csum)) {
                *err = "bad GPT entries checksum\n";
                return (-1);