I don't quite understand this code, but... what's 'p', and aren't you using its uninitialized value?

You're right. I copied the _ALIGN(p) - (size_t)p from a few lines below, but p isn't initialized yet. Surprising that the compiler didn't warn about it.

Just for my understanding: orig_buf_size is the size of the buffer to which we're copying results, and buffer_size is the size of the cached data returned by nscd?

In nscd's on_read_request_process() I see that if the agent's lookup function returns NS_NOTFOUND, nscd seemingly writes negative_data, which is a single nul byte. Is that the source of these short reads? Doesn't getpwent() have the same problem?

Sorry, I missed that you also modified getpwent().

This brace should be on the preceding line. There are several instances of this in the diff.

Should this be NS_NOTFOUND? Hard to say unless we know exactly why nscd sends small buffers.

Yes.

Fixed. Note that the brace on line 338 needs its own line; otherwise it overflows 80 columns.

Probably so. At the very least, it's no more wrong than NS_UNAVAIL. I'll change it.