Age | Commit message (Collapse) | Author | |
---|---|---|---|
2016-07-26 | bcache: fix setting bool/uint opts from sysfs | Kent Overstreet | |
ugh variable shadowing | |||
2016-07-26 | bcachefs: fix a deadlock | Kent Overstreet | |
pagecache add lock/page lock locking inversion | |||
2016-07-26 | bcache: better journal pipelining | Kent Overstreet | |
previously, when a journal entry filled up or was otherwise ready to be written, we couldn't start on the next journal entry until all outstanding reservations had finished using the current journal entry. this fixes that - we'll start the next journal entry immediately, and the write will happen whenever the last res_put() happens on the old journal entry. | |||
2016-07-26 | bcache: verify btree node lock ordering | Kent Overstreet | |
2016-07-26 | bcache: fix journal_reclaim_work() locking | Kent Overstreet | |
it assumed it was only ever called by one thread, but that changed when freezig was fixed | |||
2016-07-26 | bcache: journal_write -> journal_buf | Kent Overstreet | |
2016-07-26 | bcache: btree_insert_hook is actually only for extents | Kent Overstreet | |
doesn't make sense for non extents | |||
2016-07-26 | bcache: fix bch_btree_node_iter_fix() | Kent Overstreet | |
2016-07-26 | bcache: Simplify reservations for extents | Kent Overstreet | |
2016-07-26 | bcache: reimplement handle_existing_key_newer() | Kent Overstreet | |
2016-07-26 | bcache: more bch_insert_fixup_extent() refactoring | Kent Overstreet | |
more state machine improvements - move responsibility for handling @insert spanning btree nodes into bch_insert_fixup_extent() | |||
2016-07-26 | bcache: tweak extent hook fn | Kent Overstreet | |
2016-07-26 | bcache: btree_trans_iter_cmp() | Kent Overstreet | |
sort btree iters correctly in btree_insert_trans() (by btree_id, not just pos) | |||
2016-07-26 | bcache: fix btree_iter_pos_in_node() | Kent Overstreet | |
2016-07-26 | bcache: improve time stats for journal reservation get blocked | Kent Overstreet | |
2016-07-26 | bcache: no need for journal_res_put() to call journal_entry_close() | Kent Overstreet | |
journal_unlock() -> __journal_write() will close it | |||
2016-07-26 | bcache: fix bch_write() error path | Kent Overstreet | |
on write IO error, bch_write_index() could end up being called with op->insert_keys empty, breaking in bch_btree_insert_list_at() | |||
2016-07-26 | bcache: fix for when we stop before allocator's first start | Kent Overstreet | |
2016-07-26 | bcache: don't pack extents when trimmed version won't fit | Kent Overstreet | |
2016-07-26 | bcache: Clamp journal size so fits in superblock | Kent Overstreet | |
2016-07-26 | bcache: Call bch_async_split_will_free_node() in appropriate places | Kent Overstreet | |
2016-07-26 | bcache: split out bch_btree_insert_list_at() | Kent Overstreet | |
2016-07-26 | bcache: btree insert path refactoring | Kent Overstreet | |
2016-07-26 | bcache: Make bch_insert_fixup_extent() handle failure to pack | Kent Overstreet | |
This means we can be less conservative when calculating btree node formats | |||
2016-07-26 | bcache: refactor bch_insert_fixup_extent() | Kent Overstreet | |
2016-07-26 | bcache: bch_btree_node_insert_fits()/journal_res_insert_fits() | Kent Overstreet | |
2016-07-26 | bcache: optimize inserts of overwriting keys | Kent Overstreet | |
2016-07-26 | bcache: delete dead code | Kent Overstreet | |
2016-07-26 | bcachefs: BCH_READ_FORCE_BOUNCE isn't needed in bchfs_read() | Kent Overstreet | |
2016-07-26 | bcachefs: simplify dio read path a bit | Kent Overstreet | |
2016-07-26 | bcache: fix a null ptr deref in the xattr code | Kent Overstreet | |
2016-07-26 | bcachefs: fix fallocate + compression | Kent Overstreet | |
2016-07-26 | bcache: fall back to vmalloc() in bio_uncompress_lz4() | Kent Overstreet | |
2016-07-26 | bcachefs: fix leak in dio read path | Kent Overstreet | |
2016-07-26 | bcachefs: use FS_REQUIRES_DEV | Kent Overstreet | |
this seems to be needed for root=/dev/blah on kernel command line to work... might break mounting multi device cache sets though.. | |||
2016-07-26 | bcache: don't run copygc unnecessarily | Kent Overstreet | |
2016-07-26 | bcache: don't use TASK_UNINTERRUPTIBLE when sleeping on io | Kent Overstreet | |
TASK_UNINTERRUPTIBLE counts against load average | |||
2016-07-26 | bcachefs: pagecache lock | Kent Overstreet | |
2016-07-26 | mm: pagecache add lock | Kent Overstreet | |
Add a per address space lock around adding pages to the pagecache - making it possible for fallocate INSERT_RANGE/COLLAPSE_RANGE to work correctly, and also hopefully making truncate and dio a bit saner. | |||
2016-07-26 | bcache: Suspend fixes | Kent Overstreet | |
ugh, hate freezing - ought to be able to suspend correctly now though.. | |||
2016-07-26 | bcache: async_split improvements & error handling | Kent Overstreet | |
2016-07-26 | bcache: bch_journal_halt() | Kent Overstreet | |
2016-07-26 | bcache: printk cleanup | Kent Overstreet | |
2016-07-26 | bcache: block bch_write() caller on alloc failure | Kent Overstreet | |
else things such as writeback can queue up unlimited numbers of writes | |||
2016-07-26 | bcache: Make sure to do a journal write after flushing btree | Kent Overstreet | |
2016-07-26 | bcache: fix allocator shutdown | Kent Overstreet | |
the BUG_ON(!atomic_read(&ob->pin)) line was wrong - since there's no lock around putting the ref, have to use the wait_event() method | |||
2016-07-26 | bcache: kill obsolete/incorrect BUG_ON() | Kent Overstreet | |
The journal_seq blacklist mechanism will add journal pins before we do journal replay, causing this BUG_ON() to fire incorrectly | |||
2016-07-26 | bcache: delete dead code | Kent Overstreet | |
2016-07-26 | bcache: do cancel_work_sync() in shutdown path | Kent Overstreet | |
these probably aren't technically needed, but better safe than sorry | |||
2016-07-26 | bcache: bch_cache_set_emergency_read_only() | Kent Overstreet | |
going ro with bch_cache_set_read_only() on error, when the allocator has already stopped, could deadlock - we'd be waiting on foreground writes to complete that are stuck because the allocator already stopped (due to the original error) |