RepoMirrors
/
mars
mirror of https://github.com/schoebel/mars
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity
1,005 Commits 20 Branches 262 Tags 174 MiB
Commit Graph

15 Commits

Author SHA1 Message Date
Thomas Schoebel-Theuer f45c3fbdd3 aio: use aio_sync_mode=2 by default 2014-03-19 11:30:22 +01:00
Thomas Schoebel-Theuer d70a415b9a aio: fix dirty_head completion 2013-07-10 09:53:59 +02:00
Thomas Schoebel-Theuer 5103e7b46c aio: increase robustness of file descriptors 2013-07-10 07:16:19 +02:00
Thomas Schoebel-Theuer c9bb358239 aio: improve debugging 2013-07-10 07:16:19 +02:00
Thomas Schoebel-Theuer 62e2f5944b aio: finally fix race on dirty area at the end 2013-07-04 07:21:01 +02:00
Thomas Schoebel-Theuer 83b75845c3 aio: remove obsolete clever_sync 2013-07-04 07:21:01 +02:00
Thomas Schoebel-Theuer f3613177a2 aio: prefer fdatasync() over filemap_write_and_wait_range() 2013-07-04 07:21:00 +02:00
Thomas Schoebel-Theuer 81ca278291 aio: fix race on file size 
The old code used mf->mf_max for correcting the file size, but that
was wrong for multiple writes in flight.

A really correct solution would have to remember all in-flight writes
and compute their minimum IO position. Since that would be too
costly, we just use the old size before any writes have started.

This might be too conservative for extremely high load patterns
(possible starvation problem). For now, take this and check whether
we really need higher effort.
 2013-06-20 15:08:28 +02:00
Thomas Schoebel-Theuer a1caef264c aio: avoid cluttering the log messages 2013-05-13 12:50:29 +02:00
Thomas Schoebel-Theuer 5a5ea9bc52 infra: initialize IO transfer metadata correctly 2013-04-16 10:42:16 +02:00
Thomas Schoebel-Theuer 27fd43cb31 infra: remove obsolete black interface 
The backing file should not be exposed from the brick blackbox.
 2013-04-16 10:42:16 +02:00
Thomas Schoebel-Theuer 814b92ba19 aio: fix race on file length (workaround) 
It appears that concurrent reads and writes on AIO seem to
result in inconsistent reads in some very rare cases, due to
races. Sometimes, the inode claims that the file has been already
appended by a write operation, but the data has not actually hit
the page cache, such that a concurrent read gets NULL blocks.

This is a workaround by "misusing" the mapfree infrastructure.
It depends on uniqueness of file names, which is not the right thing
to do.

Best fix would be in the kernel.
 2013-04-16 10:42:16 +02:00
Thomas Schoebel-Theuer fa48bb94d0 infra: add tracing of max region to mapfree 2013-04-16 10:04:43 +02:00
Thomas Schoebel-Theuer 5fc2fa0adb aio: fix comm info of event threads 2013-04-15 18:37:12 +02:00
Thomas Schoebel-Theuer c58417d271 all: move kernel source into separate directory 2013-04-08 17:01:37 +02:00