from the block notifier, the cache could be deleted before we have the chance
to lock it. We now lock the sCachesLock, and see if this cache is still valid.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@28512 a95241bf-73f2-0310-859d-f6bbb57e9c96
cache_end_transaction(), and cache_start_sub_transaction()).
* Further work on the test application, it's now actually usable, first test
passes.
* dump_block() did erroneously print 'B' for the dirty flag; now both dirty and
discard have the 'D' (3rd and 5th column).
* block_cache::LowMemoryHandler() is now private (and got an underscore prefix).
* Minor cleanup, shuffled some methods around.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@28508 a95241bf-73f2-0310-859d-f6bbb57e9c96
discarding their changes. This functionality currently only works correctly
when no transactions are used.
* Started test application for the block cache, doesn't do anything yet.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@28496 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Turned the checks for all those macros to "#if"s instead of "#ifdef"s.
* Introduced macro KDEBUG_LEVEL which serves as a master setting.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@28248 a95241bf-73f2-0310-859d-f6bbb57e9c96
ignore every other cache (starting from the first).
* The consequence of this was that no blocks were written back automatically
for those caches, and their transactions were never idle, causing bug #2781.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@27831 a95241bf-73f2-0310-859d-f6bbb57e9c96
Currently it only contains KDEBUG and the block cache debugging macros.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@27816 a95241bf-73f2-0310-859d-f6bbb57e9c96
global cache list before locking it. This allows to get rid of the
block_cache::deleting field, as well as simplifies some code.
* This also fixes a possible deadlock I recently introduced (on destruction,
the locking order was wrong).
* Now uses an anonymous condition variable instead.
* Moved the block_cache initialization code into a dedicated method that will
now also fail in case the low resource handler couldn't be registered (as
pointed out by Salvatore).
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@27074 a95241bf-73f2-0310-859d-f6bbb57e9c96
introduces the following relevant changes:
* VMCache:
- Renamed vm_cache to VMCache, merged it with vm_store and made it a
C++ class with virtual methods (replacing the store operations).
Turned the different store implementations into subclasses.
- Introduced MergeStore() callback, changed semantics of Commit().
- Changed locking and referencing semantics. A reference can only be
acquired/released with the cache locked. An unreferenced cache is
deleted and a mergeable cache merged when it is unlocked. This
removes the "busy" state of a cache and simplifies the page fault
code.
* Added VMAnonymousCache, which will implement swap support (work by
Zhao Shuai). It is not integrated and used yet, though.
* Enabled the mutex/recursive lock holder asserts.
* Fixed DoublyLinkedList::Swap().
* Generalized the low memory handler to a low resource handler. And made
semaphores and reserved memory handled resources. Made
vm_try_resource_memory() optionally wait (with timeout), and used that
feature to reserve memory for areas.
...
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@26572 a95241bf-73f2-0310-859d-f6bbb57e9c96
* remove_transaction_listeners() was checking the events_pending field without
holding the sNotificatonLock - that should have been harmless (as we're in the
middle of deleting the transaction), but it now looks better.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@25130 a95241bf-73f2-0310-859d-f6bbb57e9c96
respective Private* base class.
* Changed sigwait() and sigsuspend() to use thread_block() instead of a
condition variable.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@25100 a95241bf-73f2-0310-859d-f6bbb57e9c96
This fixes a potential problem where a volume would be left marked dirty on
a clean reboot as reported by Vasilis; no harm would have been done to the
data on the volume, though (it would just replay a log entry that it didn't
need to).
* Unified the naming of the notifier/writer thread: renamed the thread to "block
notifier/writer", and the variable holding the thread ID to
sNotifierWriterThread.
* Added a few more doxygen descriptions to the private functions.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@24870 a95241bf-73f2-0310-859d-f6bbb57e9c96
was asked to. Unfortunately, it also removed the TRANSACTION_WRITTEN
listeners (that were added before notifying other listeners), so that BFS
could never free up space in the log. Now, it will only remove the listeners
it may remove. This fixes bug #2024.
* Rearranged functions a bit, moved notification/listener related private
functions together.
* Added tracing entry for written transactions as well.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@24853 a95241bf-73f2-0310-859d-f6bbb57e9c96
within the notifier/writer thread, and will then flush the notifications
directly. This should fix#2008 again.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@24806 a95241bf-73f2-0310-859d-f6bbb57e9c96
space left for the new log entry, it did call cache_sync_transaction(), and
then just assumed the space would be ready. But since the transaction could
have been written before that call by the block writer, and since the
_TransactionWritten() hook is now called asynchronously,
cache_sync_transaction() actually has to flush all pending TRANSACTION_WRITTEN
notifications before returning to the caller.
* To implement this, block_cache now publishs a condition variable, and
wait_for_notifications() adds a fake notification that signals that one.
Since the notifications are handled in FIFO order, this guarantees that
the previous TRANSACTION_WRITTEN hook is done.
* notify_transaction_listeners() could accidently delete notifications that
still had pending signals. Now, it will defer the deletion to the notification
thread instead in that case. This should fix bug #2008.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@24792 a95241bf-73f2-0310-859d-f6bbb57e9c96
called without having the cache's lock held.
* The only downside to this approach is that ending a transaction now needs
additional memory, and might therefore fail - that could be fixed by
allocating that memory upfront with the start of the new transaction,
though.
* Therefore, I reverted the cache lock to a benaphore, as the recursive
lock is no longer needed.
* The block writer thread is now a block writer and notifier thread: it will
wait for up to 2 seconds on the sEventSemaphore, and if nothing comes in,
it will continue with its previously sole job.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@24768 a95241bf-73f2-0310-859d-f6bbb57e9c96
check in notify_transaction_listeners() didn't hold, although the
cache_hooks were deleted anyway. I've introduced a listener_change
counter to detect changes to the listener list while iterating, since
any change makes the iteration unsafe. Should fix bug #2008 for real.
Nevertheless I thing the transaction listener concept needs revision.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@24761 a95241bf-73f2-0310-859d-f6bbb57e9c96
block_writer() as well as notify_transaction_listeners() must update their
data or quit. This fixes newly introduced bug #2008.
* Minor cleanup.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@24749 a95241bf-73f2-0310-859d-f6bbb57e9c96
hasn't been used for more than 2 seconds.
* Replaced the block_cache::lock benaphore with a recursive lock, so that
you can call cache functions from within the notification listeners.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@24737 a95241bf-73f2-0310-859d-f6bbb57e9c96
not referenced and clean (is no longer part of a transaction).
This finally gets memory consumption to a moderate level; even
unpacking a large archive will no longer eat all memory.
* The "block_cache" KDL command will now print the number of unused
blocks as well.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@24606 a95241bf-73f2-0310-859d-f6bbb57e9c96
could be found or not, eventually resulting in a read fault.
* It now also uses parse_expression() for its second argument, since it
now returns an uint64 in Haiku.
* Tracing output cut off the label too early.
* Minor cleanup.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@24579 a95241bf-73f2-0310-859d-f6bbb57e9c96
number. Was a problem only for partitions > 2^32 * block size (4TB
for 1KB blocks).
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@24393 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Added block_writer thread that continuously writes back blocks for all current
block caches.
* A block cache now maintains the number of dirty blocks if it doesn't use
transactions. That knowledge could also be used in block_cache_sync()...
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@23698 a95241bf-73f2-0310-859d-f6bbb57e9c96
* This should speed up writing a transaction considerably.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@23694 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Removed the public hash_grow() function again (at least for now, it's only
private).
* Removed the newSize argument from hash_grow(); it will compute the new size
automatically.
* The block cache is now using hash_insert_grow() instead of hash_insert()
which should make hash lookups much faster with some 10 thousand blocks,
also increased the initial table size from 32 to 1024...
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@23692 a95241bf-73f2-0310-859d-f6bbb57e9c96
So now there's a central place to enable tracing in general and for
individual components.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@23645 a95241bf-73f2-0310-859d-f6bbb57e9c96
"block_cache" to dump info about transactions.
* Improved tracing.
* Now uses the new add_debugger_command_etc().
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@23644 a95241bf-73f2-0310-859d-f6bbb57e9c96
blocks into the new transaction, but it would set that transaction on all
blocks of the old transaction, too. Also, it did not correctly update the
num_blocks/sub_num_blocks fields of the old transaction. Even worse, it did
return B_OK instead of the ID of the new transaction...
* get_writable_cached_block() did not correctly maintain the number of blocks
in the sub transaction.
* write_cached_block() did not free the original_data of a block when it wrote
it back as part of a previous transaction.
* Changed "cookie" for cache_next_block_in_transaction() to "long", so it will
be 64 bits when needed.
* Improved the API for detaching sub transactions: you can now get the blocks
of only the main (parent) transaction as well, added new
cache_block_in_main_transaction() function.
* BFS now flushes the log when there is no space left for the current
transaction.
* _WriteTransactionToLog() allocated a "vecs" array, but never freed it.
* _WriteTransactionToLog() now also supports detaching the current sub
transaction if the whole thing is getting too large (it will now also panic
if that doesn't work out).
* Removed a useless optimization: making the blocks available in the cache
isn't really needed, as all blocks in a transaction are locked into the
cache, anyway.
* Implemented Transaction::WriteBlocks().
* Minor cleanup, removed some dead code, fixed warnings in the fs_shell's
block_cache when compiled with debug output on.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@23610 a95241bf-73f2-0310-859d-f6bbb57e9c96
(so we can't allocate more physical pages than virtual address space
anymore, but so what?).
* Used the new CACHE_LARGE_SLAB flag as a temporary work-around; else
the slab would easily create several thousands of areas, which our
area code (and kernel heap) can't really handle that well (gets
awfully slow).
* Block caches with the same size could share the same slab, but we
don't do that yet.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@23567 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Made the block_cache KDL command dump a bit more useful info (number of
referenced and dirty blocks).
* Minor cleanup.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@23540 a95241bf-73f2-0310-859d-f6bbb57e9c96
* improved the "block_cache" KDL command: it can now also dump blocks,
added support for the "--help" argument.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@23532 a95241bf-73f2-0310-859d-f6bbb57e9c96
transaction ends or has been aborted.
* BFS now listens for transactions when it created an inode to see if
the transaction will be aborted without freeing the inode (in which
case it will panic for now).
* Started implementing tracing support, but it's not working yet.
* Minor cleanup.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@23492 a95241bf-73f2-0310-859d-f6bbb57e9c96
the inode's internal last modified and size copies, causing in inconsistent
data. This fixes#1643.
* The block being replayed are now dumped to the debug output.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@23050 a95241bf-73f2-0310-859d-f6bbb57e9c96
Axel Dörfler