While we have mechanisms in place to protect ourselves against the read
behind the disk end, there is still one corner case. As the GPT
partition table has backup table at the end of the disk, and we yet
do not know the size of the disk (if the wrong size is provided by the
firmware/bios), we need to limit the reads to avoid read ahead in such case.
Note: this update does add constant into stand.h, so the incremental build
will need to get local stand.h updated first.
Reviewed by: allanjude
Differential Revision: https://reviews.freebsd.org/D10187
The loader bcache is implementing simple read-ahead to boost the cache.
The bcache is built based on 512B block sizes, and the read ahead is attempting
to read number of cache blocks, based on amount of the free bcache space.
However, there are devices using larger sector sizes than 512B, most obviously
the CD media is based on 2k sectors. This means the read-ahead can not be just
random number of blocks, but we should use value suitable also for use with
larger sectors, as for example, with CD devices, we should read multiple of 2KB.
Since the sector size from disk interface is not too reliable, i guess we can
just use "good enough" value, so the implementation is rounding down the read
ahead block count to be multiple of 16.
This means we have covered sector sizes to 8k.
In addition, the update does implement the end of cache marker, to help to
detect the possible memory corruption - I have not seen it happening so far,
but it does not hurt to have the detection mechanism in place.
Reviewed by: allanjude
Approved by: allanjude (mentor)
Differential Revision: https://reviews.freebsd.org/D9179
Apparently the libstand dosfs optimization is a bit too optimistic
and did introduce possible memory corruption.
This patch is backing out the bad part and since this results in
dosfs reading full blocks now, we can also remove extra offset argument
from dv_strategy callback.
The analysis of the issue and the backout patch is provided by Mikhail Kupchik.
PR: 214423
Submitted by: Mikhail Kupchik
Reported by: Mikhail Kupchik
Reviewed by: bapt, allanjude
Approved by: allanjude (mentor)
MFC after: 1 month
Differential Revision: https://reviews.freebsd.org/D8644
dosfs (fat file systems) can perform reads of partial sectors
bcache should support such reads.
Submitted by: Toomas Soome <tsoome@me.com>
Reviewed by: cem
Differential Revision: https://reviews.freebsd.org/D6475
The new bcache code does not know the size of the disk, and therefore may attempt to read past the end of the disk while trying to fill its read-ahead cache.
This is usually not an issue, it fails gracefully on all of my machines, but some BIOSes seem to retry the reads for up to 30 seconds each, resulting in a long stall during boot
Submitted by: Toomas Soome <tsoome@me.com>
Reviewed by: jhb, np
Differential Revision: https://reviews.freebsd.org/D6109
The block cache implementation in loader has proven to be almost useless, and in worst case even slowing down the disk reads due to insufficient cache size and extra memory copy.
Also the current cache implementation does not cache reads from CDs, or work with zfs built on top of multiple disks.
Instead of an LRU, this code uses a simple hash (O(1) read from cache), and instead of a single global cache, a separate cache per block device.
The cache also implements limited read-ahead to increase performance.
To simplify read ahead management, the read ahead will not wrap over bcache end, so in worst case, single block physical read will be performed to fill the last block in bcache.
Booting from a virtual CD over IPMI:
0ms latency, before: 27 second, after: 7 seconds
60ms latency, before: over 12 minutes, after: under 5 minutes.
Submitted by: Toomas Soome <tsoome@me.com>
Reviewed by: delphij (previous version), emaste (previous version)
Relnotes: yes
Differential Revision: https://reviews.freebsd.org/D4713
was the last unit number received. If it changes, it flushes the cache.
Add bcache_flash().
The actual fix is sligthly different from the one in the PR.
PR: 17098
Submitted by: John Hood <jhood@sitaranetworks.com>
the first time block 0 is read. This fix initialises the block
numbers to -1 which isn't the most correct thing for a daddr_t but
it isn't likely to cause a problem in the boot blocks and it could
do with a more thought out fix later.
The bug is probably benign on the i386 but on the alpha it can
cause initial file opens to fail. This is the cause of the "can't
open /boot/boot.conf" errors.
It appears on the alpha because of a number of combining factors.
On the alpha the LABELSECTOR is 0 so block 0 needs to be read in
from the media. The first time this happens you get a false hit
because the bc_blkno field is zero initially. Also, the timestamp
check against this cache hit succeeds because on the alpha a hacked
getsecs() function can return 0 when it starts counting so that
the zero initial timestamp + BCACHE_TIMEOUT is greater than the
current time until getsecs() has counted passed BCACHE_TIMEOUT.
The overall effect is that the first open() that occurs gets a
false cache hit and returns garbage to the bd_strategy() function
which then fails the open() call. This false hit then stays in the
cache until BCACHE_TIMEOUT getsecs() ticks have passed; all open()
calls during this time fail.
This explains why you can generally access the media by the time
you get to interp() and start issuing commands but the earlier
attempts to run the boot scripts are failing.
It's possible that this is causing the problem switching to the
mfsroot floppy as well but I haven't confirmed that.
to the Forth interpreter. Instantiate all of our inbuilt commands
as Forth words, and handle them being called from there.
Add my copyright to the bcache module (oops).
and will bypass transfers for more than 8k. Blocks are invalidated after
2 seconds, so removable media should not confuse the cache.
The 8k threshold is a compromise; all UFS transfers performed by
libstand are 8k or less, so large file reads thrash the cache.
However many filesystem metadata operations are also performed using
8k blocks, so using a lower threshold gives poor performance.
Those of you with an eye for cache algorithms are welcome to tell me
how badly this one sucks; you can start with the 'bcachestats' command
which will print the contents of the cache and access statistics.
