mirror of
https://git.proxmox.com/git/mirror_zfs.git
synced 2025-01-14 12:10:26 +03:00
9d5b524597
The timeline of the race condition is the following:
[1] Thread A is about to finish condesing the first vdev in
spa_condense_indirect_thread(), so it calls the
spa_condense_indirect_complete_sync() sync task which sets
the spa_condensing_indirect field to NULL. Waiting for the
sync task to finish, thread A sleeps until the txg is done.
When this happens, thread A will acquire spa_async_lock and
set spa_condense_thread to NULL.
[2] While thread A waits for the txg to finish, thread B which is
running spa_sync() checks whether it should condense the
second vdev in vdev_indirect_should_condense() by checking the
spa_condensing_indirect field which was set to NULL by
spa_condense_indirect_thread() from thread A. So it goes on
and tries to spawn a new condensing thread in
spa_condense_indirect_start_sync() and the aforementioned
assertions fails because thread A has not set spa_condense_thread
to NULL (which is basically the last thing it does before returning).
The main issue here is that we rely on both spa_condensing_indirect
and spa_condense_thread to signify whether a condensing thread is
running. Ideally we would only use one throughout the codebase. In
addition, for managing spa_condense_thread we currently use
spa_async_lock which basically tights condensing to scrubing when
it comes to pausing and resuming those actions during spa export.
This commit introduces the ZTHR infrastructure, which is basically
threads created during spa_load()/spa_create() and exist until we
export or destroy the pool. ZTHRs sleep the majority of the time,
until they are notified to wake up and do some predefined type of work.
In the context of the current bug, a zthr to does the condensing of
indirect mappings replacing the older code that used bare kthreads.
When a pool is created, the condensing zthr is spawned but sleeps
right away, until it is awaken by a signal from spa_sync(). If an
existing pool is loaded, the condensing zthr looks if there is
anything to condense before going to sleep, in case we were condensing
mappings in the pool before it got exported.
The benefits of this solution are the following:
- The current bug is fixed
- spa_condensing_indirect is the sole indicator of whether we are
currently condensing or not
- condensing is more decoupled from the spa_async_thread related
functionality.
As a final note, this commit also sets up the path on upstreaming
other features that use the ZTHR code like zpool checkpoint and
fast clone deletion.
Authored by: Serapheim Dimitropoulos <serapheim@delphix.com>
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com>
Approved by: Hans Rosenfeld <rosenfeld@grumpf.hope-2000.org>
Ported-by: Tim Chase <tim@chase2k.com>
OpenZFS-issue: https://illumos.org/issues/9079
OpenZFS-commit: https://github.com/openzfs/openzfs/commit/3dc606ee
Closes #6900
94 lines
2.6 KiB
C
94 lines
2.6 KiB
C
/*
|
|
* CDDL HEADER START
|
|
*
|
|
* This file and its contents are supplied under the terms of the
|
|
* Common Development and Distribution License ("CDDL"), version 1.0.
|
|
* You may only use this file in accordance with the terms of version
|
|
* 1.0 of the CDDL.
|
|
*
|
|
* A full copy of the text of the CDDL should have accompanied this
|
|
* source. A copy of the CDDL is also available via the Internet at
|
|
* http://www.illumos.org/license/CDDL.
|
|
*
|
|
* CDDL HEADER END
|
|
*/
|
|
|
|
/*
|
|
* Copyright (c) 2014, 2015 by Delphix. All rights reserved.
|
|
*/
|
|
|
|
#ifndef _SYS_VDEV_REMOVAL_H
|
|
#define _SYS_VDEV_REMOVAL_H
|
|
|
|
#include <sys/spa.h>
|
|
#include <sys/bpobj.h>
|
|
#include <sys/vdev_indirect_mapping.h>
|
|
#include <sys/vdev_indirect_births.h>
|
|
|
|
#ifdef __cplusplus
|
|
extern "C" {
|
|
#endif
|
|
|
|
typedef struct spa_vdev_removal {
|
|
uint64_t svr_vdev_id;
|
|
uint64_t svr_max_offset_to_sync[TXG_SIZE];
|
|
/* Thread performing a vdev removal. */
|
|
kthread_t *svr_thread;
|
|
/* Segments left to copy from the current metaslab. */
|
|
range_tree_t *svr_allocd_segs;
|
|
kmutex_t svr_lock;
|
|
kcondvar_t svr_cv;
|
|
boolean_t svr_thread_exit;
|
|
|
|
/*
|
|
* New mappings to write out each txg.
|
|
*/
|
|
list_t svr_new_segments[TXG_SIZE];
|
|
|
|
/*
|
|
* Ranges that were freed while a mapping was in flight. This is
|
|
* a subset of the ranges covered by vdev_im_new_segments.
|
|
*/
|
|
range_tree_t *svr_frees[TXG_SIZE];
|
|
|
|
/*
|
|
* Number of bytes which we have finished our work for
|
|
* in each txg. This could be data copied (which will be part of
|
|
* the mappings in vdev_im_new_segments), or data freed before
|
|
* we got around to copying it.
|
|
*/
|
|
uint64_t svr_bytes_done[TXG_SIZE];
|
|
|
|
/* List of leaf zap objects to be unlinked */
|
|
nvlist_t *svr_zaplist;
|
|
} spa_vdev_removal_t;
|
|
|
|
typedef struct spa_condensing_indirect {
|
|
/*
|
|
* New mappings to write out each txg.
|
|
*/
|
|
list_t sci_new_mapping_entries[TXG_SIZE];
|
|
|
|
vdev_indirect_mapping_t *sci_new_mapping;
|
|
} spa_condensing_indirect_t;
|
|
|
|
extern int spa_remove_init(spa_t *);
|
|
extern void spa_restart_removal(spa_t *);
|
|
extern int spa_condense_init(spa_t *);
|
|
extern void spa_condense_fini(spa_t *);
|
|
extern void spa_start_indirect_condensing_thread(spa_t *);
|
|
extern void spa_vdev_condense_suspend(spa_t *);
|
|
extern int spa_vdev_remove(spa_t *, uint64_t, boolean_t);
|
|
extern void free_from_removing_vdev(vdev_t *, uint64_t, uint64_t, uint64_t);
|
|
extern int spa_removal_get_stats(spa_t *, pool_removal_stat_t *);
|
|
extern void svr_sync(spa_t *spa, dmu_tx_t *tx);
|
|
extern void spa_vdev_remove_suspend(spa_t *);
|
|
extern int spa_vdev_remove_cancel(spa_t *);
|
|
extern void spa_vdev_removal_destroy(spa_vdev_removal_t *svr);
|
|
|
|
#ifdef __cplusplus
|
|
}
|
|
#endif
|
|
|
|
#endif /* _SYS_VDEV_REMOVAL_H */
|