mkimg: Ensure GPT Entry Array is at least 16k

[imp]

Summary

The GPT standard requires that we reserve either 34 512 sectors, or 6 4k
sectors before starting partitions. However, we don't currently don't
enforce these minimums. This is 2 sectors + 16k of space.

Sponsored by: Netflix

Test Plan

The old code produces

00000200  45 46 49 20 50 41 52 54  00 00 01 00 5c 00 00 00  |EFI PART....\...|
00000210  ea c7 2f 95 00 00 00 00  01 00 00 00 00 00 00 00  |../.............|
00000220  04 60 09 00 00 00 00 00  03 00 00 00 00 00 00 00  |.`..............|
00000230  02 60 09 00 00 00 00 00  12 7a a9 63 4a f6 ed 11  |.`.......z.cJ...|
00000240  a7 d3 e0 d5 5e 1e 73 bd  02 00 00 00 00 00 00 00  |....^.s.........|
00000250  04 00 00 00 80 00 00 00  58 9c b0 81 00 00 00 00  |........X.......|

Which has 3 is the FirstLBA and 4 as the number of entries.

The correct output I claim is:

00000200  45 46 49 20 50 41 52 54  00 00 01 00 5c 00 00 00  |EFI PART....\...|
00000210  d8 2e 74 5a 00 00 00 00  01 00 00 00 00 00 00 00  |..tZ............|
00000220  42 60 09 00 00 00 00 00  22 00 00 00 00 00 00 00  |B`......".......|
00000230  21 60 09 00 00 00 00 00  84 7b f1 cc 98 6c ee 11  |!`.......{...l..|
00000240  a8 54 a0 36 9f 09 4f 1e  02 00 00 00 00 00 00 00  |.T.6..O.........|
00000250  80 00 00 00 80 00 00 00  71 5c 49 7b 00 00 00 00  |........q\I{....|

which has first LBA is 0x22 / 34 and num entires of 0x80 / 128. These are the desired values.

[imp]

Include sys/param.h for MIN macro

[imp]

braino: min will return the smaller number. We want it to return the
larger of the computed size and the minimal size.

[emaste]

sys/sys/disk/gpt.h

87

[emaste]

usr.bin/mkimg/gpt.c

131

I find this function a little confusing as written, but that predates and is independent of your change.

[imp]

usr.bin/mkimg/gpt.c

131

Yea... we need a roundup macro...

[jrtc27]

usr.bin/mkimg/gpt.c

29

param comes before all headers other than cdefs, though not really sure why cdefs is included in the first place (left over from $FreeBSD$?), and there should be a blank line between the sys/ and non-sys/ includes anyway, so this isn't exactly the most style-conforming already

131

howmany from sys/param.h? #define howmany(x, y) (((x)+((y)-1))/(y))

[emaste]

usr.bin/mkimg/gpt.c

29

Indeed cdefs is not needed. Changing the first include to #include <sys/param.h> results in an identical binary (other than debug info).

[emaste]

It would be really good to get (some version of) this into the tree and on the build machine before the next RC build.
CC releng

[imp]

Update headers included, use howmany and improve (I claim) the comments
and commit message (which isn't updated automatically, alas)

[emaste]

which isn't updated automatically, alas

You can edit it through the Phab web UI though.

[emaste]

my comment in gpt.h still not addressed but gpt_tblsize is definitely an improvement

usr.bin/mkimg/gpt.c

133

Thanks, variable name ents was part of the confusion before

141

maybe also homany(GPT_MIN_RESERVED, secsz)?

[imp]

usr.bin/mkimg/gpt.c

141

No. Sector sizes are required to be a power of 2. There's no rounding needed. I think elsewhere we don't bother... And I'd like to keep the number of 'one more things' to a minimum.

Not sure what a 16k sector size would do, or how larger sector sizes would work here. But both a 16k and 64k sector size would result in this returning 1. The standard is silent on sector sizes that aren't 512 or 4k.

[imp]

address comments about gpt.h

[emaste]

usr.bin/mkimg/gpt.c

141

It's not secsz I was thinking about, but GPT_MIN_RESERVED. 16384 is (obviously) a power of two, but the constant is defined elsewhere and so it's not immediately obvious when looking at this expression.