Re: report expected contrecord size

A pretty minor issue: when reporting that WAL appears invalid because
contrecord length doesn't match, we may as well print to the server log
the value that we're expecting. Patch attached.

--
Álvaro Herrera http://www.flickr.com/photos/alvherre/

Alvaro Herrera <alvherre(at)2ndquadrant(dot)com> writes:
> A pretty minor issue: when reporting that WAL appears invalid because
> contrecord length doesn't match, we may as well print to the server log
> the value that we're expecting. Patch attached.

ITYW

+ (long long) (total_len - gotlen),

just to be sure about what's getting casted to what. Otherwise +1.

regards, tom lane

On 2020-Sep-03, Tom Lane wrote:

> Alvaro Herrera <alvherre(at)2ndquadrant(dot)com> writes:
> > A pretty minor issue: when reporting that WAL appears invalid because
> > contrecord length doesn't match, we may as well print to the server log
> > the value that we're expecting. Patch attached.
>
> ITYW
>
> + (long long) (total_len - gotlen),
>
> just to be sure about what's getting casted to what.

Well, the intention there is to cast the first operand (which is uint32)
so that it turns into signed 64-bits; the subtraction then occurs in 64
bit arithmetic normally. If I let the subtraction occur in 32-bit width
unsigned, the result might overflow 32 bits. I'm thinking in
1 - UINT32_MAX or some such.

Maybe to make that more explicit, it should be

+ ((long long) total_len) - gotlen,

(If I understand the precedence correctly, it's the same thing I wrote).

--
Álvaro Herrera https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

Alvaro Herrera <alvherre(at)2ndquadrant(dot)com> writes:
> Well, the intention there is to cast the first operand (which is uint32)
> so that it turns into signed 64-bits; the subtraction then occurs in 64
> bit arithmetic normally. If I let the subtraction occur in 32-bit width
> unsigned, the result might overflow 32 bits.

Uh ... is it really possible for gotlen to be more than total_len?
(I've not looked at the surrounding code here, but that seems weird.)

regards, tom lane

On 2020-Sep-03, Tom Lane wrote:

> Alvaro Herrera <alvherre(at)2ndquadrant(dot)com> writes:
> > Well, the intention there is to cast the first operand (which is uint32)
> > so that it turns into signed 64-bits; the subtraction then occurs in 64
> > bit arithmetic normally. If I let the subtraction occur in 32-bit width
> > unsigned, the result might overflow 32 bits.
>
> Uh ... is it really possible for gotlen to be more than total_len?
> (I've not looked at the surrounding code here, but that seems weird.)

Well, as I understand, total_len comes from one page, and gotlen comes
from the continuation record(s) in the next page(s) of WAL. So if
things are messed up, it could happen. (This *is* the code that
validates the record, so it can't make too many assumptions.)

--
Álvaro Herrera https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

Alvaro Herrera <alvherre(at)2ndquadrant(dot)com> writes:
> On 2020-Sep-03, Tom Lane wrote:
>> Uh ... is it really possible for gotlen to be more than total_len?
>> (I've not looked at the surrounding code here, but that seems weird.)

> Well, as I understand, total_len comes from one page, and gotlen comes
> from the continuation record(s) in the next page(s) of WAL. So if
> things are messed up, it could happen. (This *is* the code that
> validates the record, so it can't make too many assumptions.)

Got it. No further objections.

regards, tom lane

On 2020-Sep-03, Tom Lane wrote:

> Alvaro Herrera <alvherre(at)2ndquadrant(dot)com> writes:
> > On 2020-Sep-03, Tom Lane wrote:
> >> Uh ... is it really possible for gotlen to be more than total_len?
> >> (I've not looked at the surrounding code here, but that seems weird.)
>
> > Well, as I understand, total_len comes from one page, and gotlen comes
> > from the continuation record(s) in the next page(s) of WAL. So if
> > things are messed up, it could happen. (This *is* the code that
> > validates the record, so it can't make too many assumptions.)
>
> Got it. No further objections.

Many thanks for looking! Pushed now.

--
Álvaro Herrera https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services