ffs: do not read full direct blocks if they are going to be overwritten.

[kib]

Summary

BA_CLRBUF specifies that existing context of the block will be completely overwritten by caller, so there is no reason to spend io fetching existing data. We do the same for indirect blocks.

Reported by: tmunro

Test Plan

Not tested.

[tmunro]

Hmm. My filesystem has 32kb blocks, and this generates reads, according to dwatch -X io:

openat(AT_FDCWD,"file",O_RDWR|O_FSYNC|O_DIRECT,00) = 3 (0x3)
pwrite(3,"................................"...,32768,0x0) = 32768 (0x8000)
pwrite(3,"................................"...,32768,0x0) = 32768 (0x8000)
pwrite(3,"................................"...,32768,0x0) = 32768 (0x8000)

... while smaller write sizes don't. Isn't that backwards?

[mckusick]

The sentiment is correct, but logic fixes noted are needed.

sys/ufs/ffs/ffs_balloc.c

175–185

This is backwards. It should be "if (flags & BA_CLRBUF) != 0)".

BA_CLRBUF is set when not all of the buffer will be written so we need to clear it so that the unwritten part will be zero. Here where the block is already allocated, we need to read it when not all of it will be written and can getblk it when BA_CLRBUF is cleared because we will write all of it.

779–789

Same comment here: This is backwards. It should be "if (flags & BA_CLRBUF) != 0)".

[kib]

Correct the condition.

[tmunro]

Tested, LGTM.

[mckusick]

Good to go.