Skip to content

Commit b29d498

Browse files
jthornbersnitm
jthornber
authored and
snitm
committed
dm cache: significant rework to leverage dm-bio-prison-v2
The cache policy interfaces have been updated to work well with the new bio-prison v2 interface's ability to queue work immediately (promotion, demotion, etc) -- overriding benefit being reduced latency on processing IO through the cache. Previously such work would be left for the DM cache core to queue on various lists and then process in batches later -- this caused a serious delay in latency for IO driven by the cache. The background tracker code was factored out so that all cache policies can make use of it. Also, the "cleaner" policy has been removed and is now a variant of the smq policy that simply disallows migrations. Signed-off-by: Joe Thornber <[email protected]> Signed-off-by: Mike Snitzer <[email protected]>
1 parent 742c8fd commit b29d498

10 files changed

+1930
-2407
lines changed

‎drivers/md/Kconfig

Lines changed: 0 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -325,14 +325,6 @@ config DM_CACHE_SMQ
325325
of less memory utilization, improved performance and increased
326326
adaptability in the face of changing workloads.
327327

328-
config DM_CACHE_CLEANER
329-
tristate "Cleaner Cache Policy (EXPERIMENTAL)"
330-
depends on DM_CACHE
331-
default y
332-
---help---
333-
A simple cache policy that writes back all data to the
334-
origin. Used when decommissioning a dm-cache.
335-
336328
config DM_ERA
337329
tristate "Era target (EXPERIMENTAL)"
338330
depends on BLK_DEV_DM

‎drivers/md/Makefile

Lines changed: 2 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -13,9 +13,9 @@ dm-log-userspace-y \
1313
+= dm-log-userspace-base.o dm-log-userspace-transfer.o
1414
dm-bio-prison-y += dm-bio-prison-v1.o dm-bio-prison-v2.o
1515
dm-thin-pool-y += dm-thin.o dm-thin-metadata.o
16-
dm-cache-y += dm-cache-target.o dm-cache-metadata.o dm-cache-policy.o
16+
dm-cache-y += dm-cache-target.o dm-cache-metadata.o dm-cache-policy.o \
17+
dm-cache-background-tracker.o
1718
dm-cache-smq-y += dm-cache-policy-smq.o
18-
dm-cache-cleaner-y += dm-cache-policy-cleaner.o
1919
dm-era-y += dm-era-target.o
2020
dm-verity-y += dm-verity-target.o
2121
md-mod-y += md.o bitmap.o
@@ -57,7 +57,6 @@ obj-$(CONFIG_DM_THIN_PROVISIONING) += dm-thin-pool.o
5757
obj-$(CONFIG_DM_VERITY) += dm-verity.o
5858
obj-$(CONFIG_DM_CACHE) += dm-cache.o
5959
obj-$(CONFIG_DM_CACHE_SMQ) += dm-cache-smq.o
60-
obj-$(CONFIG_DM_CACHE_CLEANER) += dm-cache-cleaner.o
6160
obj-$(CONFIG_DM_ERA) += dm-era.o
6261
obj-$(CONFIG_DM_LOG_WRITES) += dm-log-writes.o
6362

‎drivers/md/dm-cache-background-tracker.c

Lines changed: 238 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,238 @@
1+
/*
2+
* Copyright (C) 2017 Red Hat. All rights reserved.
3+
*
4+
* This file is released under the GPL.
5+
*/
6+
7+
#include "dm-cache-background-tracker.h"
8+
9+
/*----------------------------------------------------------------*/
10+
11+
#define DM_MSG_PREFIX "dm-background-tracker"
12+
13+
struct bt_work {
14+
struct list_head list;
15+
struct rb_node node;
16+
struct policy_work work;
17+
};
18+
19+
struct background_tracker {
20+
unsigned max_work;
21+
atomic_t pending_promotes;
22+
atomic_t pending_writebacks;
23+
atomic_t pending_demotes;
24+
25+
struct list_head issued;
26+
struct list_head queued;
27+
struct rb_root pending;
28+
29+
struct kmem_cache *work_cache;
30+
};
31+
32+
struct background_tracker *btracker_create(unsigned max_work)
33+
{
34+
struct background_tracker *b = kmalloc(sizeof(*b), GFP_KERNEL);
35+
36+
b->max_work = max_work;
37+
atomic_set(&b->pending_promotes, 0);
38+
atomic_set(&b->pending_writebacks, 0);
39+
atomic_set(&b->pending_demotes, 0);
40+
41+
INIT_LIST_HEAD(&b->issued);
42+
INIT_LIST_HEAD(&b->queued);
43+
44+
b->pending = RB_ROOT;
45+
b->work_cache = KMEM_CACHE(bt_work, 0);
46+
if (!b->work_cache) {
47+
DMERR("couldn't create mempool for background work items");
48+
kfree(b);
49+
b = NULL;
50+
}
51+
52+
return b;
53+
}
54+
EXPORT_SYMBOL_GPL(btracker_create);
55+
56+
void btracker_destroy(struct background_tracker *b)
57+
{
58+
kmem_cache_destroy(b->work_cache);
59+
kfree(b);
60+
}
61+
EXPORT_SYMBOL_GPL(btracker_destroy);
62+
63+
static int cmp_oblock(dm_oblock_t lhs, dm_oblock_t rhs)
64+
{
65+
if (from_oblock(lhs) < from_oblock(rhs))
66+
return -1;
67+
68+
if (from_oblock(rhs) < from_oblock(lhs))
69+
return 1;
70+
71+
return 0;
72+
}
73+
74+
static bool __insert_pending(struct background_tracker *b,
75+
struct bt_work *nw)
76+
{
77+
int cmp;
78+
struct bt_work *w;
79+
struct rb_node **new = &b->pending.rb_node, *parent = NULL;
80+
81+
while (*new) {
82+
w = container_of(*new, struct bt_work, node);
83+
84+
parent = *new;
85+
cmp = cmp_oblock(w->work.oblock, nw->work.oblock);
86+
if (cmp < 0)
87+
new = &((*new)->rb_left);
88+
89+
else if (cmp > 0)
90+
new = &((*new)->rb_right);
91+
92+
else
93+
/* already present */
94+
return false;
95+
}
96+
97+
rb_link_node(&nw->node, parent, new);
98+
rb_insert_color(&nw->node, &b->pending);
99+
100+
return true;
101+
}
102+
103+
static struct bt_work *__find_pending(struct background_tracker *b,
104+
dm_oblock_t oblock)
105+
{
106+
int cmp;
107+
struct bt_work *w;
108+
struct rb_node **new = &b->pending.rb_node;
109+
110+
while (*new) {
111+
w = container_of(*new, struct bt_work, node);
112+
113+
cmp = cmp_oblock(w->work.oblock, oblock);
114+
if (cmp < 0)
115+
new = &((*new)->rb_left);
116+
117+
else if (cmp > 0)
118+
new = &((*new)->rb_right);
119+
120+
else
121+
break;
122+
}
123+
124+
return *new ? w : NULL;
125+
}
126+
127+
128+
static void update_stats(struct background_tracker *b, struct policy_work *w, int delta)
129+
{
130+
switch (w->op) {
131+
case POLICY_PROMOTE:
132+
atomic_add(delta, &b->pending_promotes);
133+
break;
134+
135+
case POLICY_DEMOTE:
136+
atomic_add(delta, &b->pending_demotes);
137+
break;
138+
139+
case POLICY_WRITEBACK:
140+
atomic_add(delta, &b->pending_writebacks);
141+
break;
142+
}
143+
}
144+
145+
unsigned btracker_nr_writebacks_queued(struct background_tracker *b)
146+
{
147+
return atomic_read(&b->pending_writebacks);
148+
}
149+
EXPORT_SYMBOL_GPL(btracker_nr_writebacks_queued);
150+
151+
unsigned btracker_nr_demotions_queued(struct background_tracker *b)
152+
{
153+
return atomic_read(&b->pending_demotes);
154+
}
155+
EXPORT_SYMBOL_GPL(btracker_nr_demotions_queued);
156+
157+
static bool max_work_reached(struct background_tracker *b)
158+
{
159+
// FIXME: finish
160+
return false;
161+
}
162+
163+
int btracker_queue(struct background_tracker *b,
164+
struct policy_work *work,
165+
struct policy_work **pwork)
166+
{
167+
struct bt_work *w;
168+
169+
if (pwork)
170+
*pwork = NULL;
171+
172+
if (max_work_reached(b))
173+
return -ENOMEM;
174+
175+
w = kmem_cache_alloc(b->work_cache, GFP_NOWAIT);
176+
if (!w)
177+
return -ENOMEM;
178+
179+
memcpy(&w->work, work, sizeof(*work));
180+
181+
if (!__insert_pending(b, w)) {
182+
/*
183+
* There was a race, we'll just ignore this second
184+
* bit of work for the same oblock.
185+
*/
186+
kmem_cache_free(b->work_cache, w);
187+
return -EINVAL;
188+
}
189+
190+
if (pwork) {
191+
*pwork = &w->work;
192+
list_add(&w->list, &b->issued);
193+
} else
194+
list_add(&w->list, &b->queued);
195+
update_stats(b, &w->work, 1);
196+
197+
return 0;
198+
}
199+
EXPORT_SYMBOL_GPL(btracker_queue);
200+
201+
/*
202+
* Returns -ENODATA if there's no work.
203+
*/
204+
int btracker_issue(struct background_tracker *b, struct policy_work **work)
205+
{
206+
struct bt_work *w;
207+
208+
if (list_empty(&b->queued))
209+
return -ENODATA;
210+
211+
w = list_first_entry(&b->queued, struct bt_work, list);
212+
list_move(&w->list, &b->issued);
213+
*work = &w->work;
214+
215+
return 0;
216+
}
217+
EXPORT_SYMBOL_GPL(btracker_issue);
218+
219+
void btracker_complete(struct background_tracker *b,
220+
struct policy_work *op)
221+
{
222+
struct bt_work *w = container_of(op, struct bt_work, work);
223+
224+
update_stats(b, &w->work, -1);
225+
rb_erase(&w->node, &b->pending);
226+
list_del(&w->list);
227+
kmem_cache_free(b->work_cache, w);
228+
}
229+
EXPORT_SYMBOL_GPL(btracker_complete);
230+
231+
bool btracker_promotion_already_present(struct background_tracker *b,
232+
dm_oblock_t oblock)
233+
{
234+
return __find_pending(b, oblock) != NULL;
235+
}
236+
EXPORT_SYMBOL_GPL(btracker_promotion_already_present);
237+
238+
/*----------------------------------------------------------------*/

‎drivers/md/dm-cache-background-tracker.h

Lines changed: 46 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,46 @@
1+
/*
2+
* Copyright (C) 2017 Red Hat. All rights reserved.
3+
*
4+
* This file is released under the GPL.
5+
*/
6+
7+
#ifndef DM_CACHE_BACKGROUND_WORK_H
8+
#define DM_CACHE_BACKGROUND_WORK_H
9+
10+
#include <linux/vmalloc.h>
11+
#include "dm-cache-policy.h"
12+
13+
/*----------------------------------------------------------------*/
14+
15+
struct background_work;
16+
struct background_tracker;
17+
18+
/*
19+
* FIXME: discuss lack of locking in all methods.
20+
*/
21+
struct background_tracker *btracker_create(unsigned max_work);
22+
void btracker_destroy(struct background_tracker *b);
23+
24+
unsigned btracker_nr_writebacks_queued(struct background_tracker *b);
25+
unsigned btracker_nr_demotions_queued(struct background_tracker *b);
26+
27+
/*
28+
* returns -EINVAL iff the work is already queued. -ENOMEM if the work
29+
* couldn't be queued for another reason.
30+
*/
31+
int btracker_queue(struct background_tracker *b,
32+
struct policy_work *work,
33+
struct policy_work **pwork);
34+
35+
/*
36+
* Returns -ENODATA if there's no work.
37+
*/
38+
int btracker_issue(struct background_tracker *b, struct policy_work **work);
39+
void btracker_complete(struct background_tracker *b,
40+
struct policy_work *op);
41+
bool btracker_promotion_already_present(struct background_tracker *b,
42+
dm_oblock_t oblock);
43+
44+
/*----------------------------------------------------------------*/
45+
46+
#endif

‎drivers/md/dm-cache-metadata.h

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -50,6 +50,8 @@
5050
#define DM_CACHE_FEATURE_COMPAT_RO_SUPP 0UL
5151
#define DM_CACHE_FEATURE_INCOMPAT_SUPP 0UL
5252

53+
struct dm_cache_metadata;
54+
5355
/*
5456
* Reopens or creates a new, empty metadata volume. Returns an ERR_PTR on
5557
* failure. If reopening then features must match.

0 commit comments

Comments
 (0)