Commit 2ef176f1 authored by Linus Torvalds's avatar Linus Torvalds

Merge tag 'dm-3.14-fixes-3' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm

Pull device mapper fixes from Mike Snitzer:

 - dm-cache memory allocation failure fix
 - fix DM's Kconfig identation
 - dm-snapshot metadata corruption fix for bug introduced in 3.14-rc1
 - important refcount < 0 fix for the DM persistent data library's space
   map metadata interface which fixes corruption reported by a few
   dm-thinp users

and last but not least:

 - more extensive fixes than ideal for dm-thinp's data resize capability
   (which has had growing pain much like we've seen from -ENOSPC
   handling of filesystems that mature).

   The end result is dm-thinp now handles metadata operation failure and
   no data space error conditions much better than before.

* tag 'dm-3.14-fixes-3' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm:
  dm space map metadata: fix refcount decrement below 0 which caused corruption
  dm thin: fix Documentation for held metadata root feature
  dm thin: fix noflush suspend IO queueing
  dm thin: fix deadlock in __requeue_bio_list
  dm thin: fix out of data space handling
  dm thin: ensure user takes action to validate data and metadata consistency
  dm thin: synchronize the pool mode during suspend
  dm snapshot: fix metadata corruption
  dm: fix Kconfig indentation
  dm cache mq: fix memory allocation failure for large cache devices
parents b053940d cebc2de4
......@@ -124,12 +124,11 @@ the default being 204800 sectors (or 100MB).
Updating on-disk metadata
-------------------------
On-disk metadata is committed every time a REQ_SYNC or REQ_FUA bio is
written. If no such requests are made then commits will occur every
second. This means the cache behaves like a physical disk that has a
write cache (the same is true of the thin-provisioning target). If
power is lost you may lose some recent writes. The metadata should
always be consistent in spite of any crash.
On-disk metadata is committed every time a FLUSH or FUA bio is written.
If no such requests are made then commits will occur every second. This
means the cache behaves like a physical disk that has a volatile write
cache. If power is lost you may lose some recent writes. The metadata
should always be consistent in spite of any crash.
The 'dirty' state for a cache block changes far too frequently for us
to keep updating it on the fly. So we treat it as a hint. In normal
......
......@@ -116,6 +116,35 @@ Resuming a device with a new table itself triggers an event so the
userspace daemon can use this to detect a situation where a new table
already exceeds the threshold.
A low water mark for the metadata device is maintained in the kernel and
will trigger a dm event if free space on the metadata device drops below
it.
Updating on-disk metadata
-------------------------
On-disk metadata is committed every time a FLUSH or FUA bio is written.
If no such requests are made then commits will occur every second. This
means the thin-provisioning target behaves like a physical disk that has
a volatile write cache. If power is lost you may lose some recent
writes. The metadata should always be consistent in spite of any crash.
If data space is exhausted the pool will either error or queue IO
according to the configuration (see: error_if_no_space). If metadata
space is exhausted or a metadata operation fails: the pool will error IO
until the pool is taken offline and repair is performed to 1) fix any
potential inconsistencies and 2) clear the flag that imposes repair.
Once the pool's metadata device is repaired it may be resized, which
will allow the pool to return to normal operation. Note that if a pool
is flagged as needing repair, the pool's data and metadata devices
cannot be resized until repair is performed. It should also be noted
that when the pool's metadata space is exhausted the current metadata
transaction is aborted. Given that the pool will cache IO whose
completion may have already been acknowledged to upper IO layers
(e.g. filesystem) it is strongly suggested that consistency checks
(e.g. fsck) be performed on those layers when repair of the pool is
required.
Thin provisioning
-----------------
......@@ -258,10 +287,9 @@ ii) Status
should register for the event and then check the target's status.
held metadata root:
The location, in sectors, of the metadata root that has been
The location, in blocks, of the metadata root that has been
'held' for userspace read access. '-' indicates there is no
held root. This feature is not yet implemented so '-' is
always returned.
held root.
discard_passdown|no_discard_passdown
Whether or not discards are actually being passed down to the
......
......@@ -254,16 +254,6 @@ config DM_THIN_PROVISIONING
---help---
Provides thin provisioning and snapshots that share a data store.
config DM_DEBUG_BLOCK_STACK_TRACING
boolean "Keep stack trace of persistent data block lock holders"
depends on STACKTRACE_SUPPORT && DM_PERSISTENT_DATA
select STACKTRACE
---help---
Enable this for messages that may help debug problems with the
block manager locking used by thin provisioning and caching.
If unsure, say N.
config DM_CACHE
tristate "Cache target (EXPERIMENTAL)"
depends on BLK_DEV_DM
......
......@@ -872,7 +872,7 @@ static void mq_destroy(struct dm_cache_policy *p)
{
struct mq_policy *mq = to_mq_policy(p);
kfree(mq->table);
vfree(mq->table);
epool_exit(&mq->cache_pool);
epool_exit(&mq->pre_cache_pool);
kfree(mq);
......@@ -1245,7 +1245,7 @@ static struct dm_cache_policy *mq_create(dm_cblock_t cache_size,
mq->nr_buckets = next_power(from_cblock(cache_size) / 2, 16);
mq->hash_bits = ffs(mq->nr_buckets) - 1;
mq->table = kzalloc(sizeof(*mq->table) * mq->nr_buckets, GFP_KERNEL);
mq->table = vzalloc(sizeof(*mq->table) * mq->nr_buckets);
if (!mq->table)
goto bad_alloc_table;
......
......@@ -546,6 +546,9 @@ static int read_exceptions(struct pstore *ps,
r = insert_exceptions(ps, area, callback, callback_context,
&full);
if (!full)
memcpy(ps->area, area, ps->store->chunk_size << SECTOR_SHIFT);
dm_bufio_release(bp);
dm_bufio_forget(client, chunk);
......
......@@ -76,7 +76,7 @@
#define THIN_SUPERBLOCK_MAGIC 27022010
#define THIN_SUPERBLOCK_LOCATION 0
#define THIN_VERSION 1
#define THIN_VERSION 2
#define THIN_METADATA_CACHE_SIZE 64
#define SECTOR_TO_BLOCK_SHIFT 3
......@@ -1755,3 +1755,38 @@ int dm_pool_register_metadata_threshold(struct dm_pool_metadata *pmd,
return r;
}
int dm_pool_metadata_set_needs_check(struct dm_pool_metadata *pmd)
{
int r;
struct dm_block *sblock;
struct thin_disk_superblock *disk_super;
down_write(&pmd->root_lock);
pmd->flags |= THIN_METADATA_NEEDS_CHECK_FLAG;
r = superblock_lock(pmd, &sblock);
if (r) {
DMERR("couldn't read superblock");
goto out;
}
disk_super = dm_block_data(sblock);
disk_super->flags = cpu_to_le32(pmd->flags);
dm_bm_unlock(sblock);
out:
up_write(&pmd->root_lock);
return r;
}
bool dm_pool_metadata_needs_check(struct dm_pool_metadata *pmd)
{
bool needs_check;
down_read(&pmd->root_lock);
needs_check = pmd->flags & THIN_METADATA_NEEDS_CHECK_FLAG;
up_read(&pmd->root_lock);
return needs_check;
}
......@@ -25,6 +25,11 @@
/*----------------------------------------------------------------*/
/*
* Thin metadata superblock flags.
*/
#define THIN_METADATA_NEEDS_CHECK_FLAG (1 << 0)
struct dm_pool_metadata;
struct dm_thin_device;
......@@ -202,6 +207,12 @@ int dm_pool_register_metadata_threshold(struct dm_pool_metadata *pmd,
dm_sm_threshold_fn fn,
void *context);
/*
* Updates the superblock immediately.
*/
int dm_pool_metadata_set_needs_check(struct dm_pool_metadata *pmd);
bool dm_pool_metadata_needs_check(struct dm_pool_metadata *pmd);
/*----------------------------------------------------------------*/
#endif
......@@ -130,10 +130,11 @@ static void build_virtual_key(struct dm_thin_device *td, dm_block_t b,
struct dm_thin_new_mapping;
/*
* The pool runs in 3 modes. Ordered in degraded order for comparisons.
* The pool runs in 4 modes. Ordered in degraded order for comparisons.
*/
enum pool_mode {
PM_WRITE, /* metadata may be changed */
PM_OUT_OF_DATA_SPACE, /* metadata may be changed, though data may not be allocated */
PM_READ_ONLY, /* metadata may not be changed */
PM_FAIL, /* all I/O fails */
};
......@@ -198,7 +199,6 @@ struct pool {
};
static enum pool_mode get_pool_mode(struct pool *pool);
static void out_of_data_space(struct pool *pool);
static void metadata_operation_failed(struct pool *pool, const char *op, int r);
/*
......@@ -226,6 +226,7 @@ struct thin_c {
struct pool *pool;
struct dm_thin_device *td;
bool requeue_mode:1;
};
/*----------------------------------------------------------------*/
......@@ -369,14 +370,18 @@ struct dm_thin_endio_hook {
struct dm_thin_new_mapping *overwrite_mapping;
};
static void __requeue_bio_list(struct thin_c *tc, struct bio_list *master)
static void requeue_bio_list(struct thin_c *tc, struct bio_list *master)
{
struct bio *bio;
struct bio_list bios;
unsigned long flags;
bio_list_init(&bios);
spin_lock_irqsave(&tc->pool->lock, flags);
bio_list_merge(&bios, master);
bio_list_init(master);
spin_unlock_irqrestore(&tc->pool->lock, flags);
while ((bio = bio_list_pop(&bios))) {
struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
......@@ -391,12 +396,26 @@ static void __requeue_bio_list(struct thin_c *tc, struct bio_list *master)
static void requeue_io(struct thin_c *tc)
{
struct pool *pool = tc->pool;
requeue_bio_list(tc, &pool->deferred_bios);
requeue_bio_list(tc, &pool->retry_on_resume_list);
}
static void error_retry_list(struct pool *pool)
{
struct bio *bio;
unsigned long flags;
struct bio_list bios;
bio_list_init(&bios);
spin_lock_irqsave(&pool->lock, flags);
__requeue_bio_list(tc, &pool->deferred_bios);
__requeue_bio_list(tc, &pool->retry_on_resume_list);
bio_list_merge(&bios, &pool->retry_on_resume_list);
bio_list_init(&pool->retry_on_resume_list);
spin_unlock_irqrestore(&pool->lock, flags);
while ((bio = bio_list_pop(&bios)))
bio_io_error(bio);
}
/*
......@@ -925,13 +944,15 @@ static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks)
}
}
static void set_pool_mode(struct pool *pool, enum pool_mode new_mode);
static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
{
int r;
dm_block_t free_blocks;
struct pool *pool = tc->pool;
if (get_pool_mode(pool) != PM_WRITE)
if (WARN_ON(get_pool_mode(pool) != PM_WRITE))
return -EINVAL;
r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
......@@ -958,7 +979,7 @@ static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
}
if (!free_blocks) {
out_of_data_space(pool);
set_pool_mode(pool, PM_OUT_OF_DATA_SPACE);
return -ENOSPC;
}
}
......@@ -988,15 +1009,32 @@ static void retry_on_resume(struct bio *bio)
spin_unlock_irqrestore(&pool->lock, flags);
}
static void handle_unserviceable_bio(struct pool *pool, struct bio *bio)
static bool should_error_unserviceable_bio(struct pool *pool)
{
/*
* When pool is read-only, no cell locking is needed because
* nothing is changing.
*/
WARN_ON_ONCE(get_pool_mode(pool) != PM_READ_ONLY);
enum pool_mode m = get_pool_mode(pool);
if (pool->pf.error_if_no_space)
switch (m) {
case PM_WRITE:
/* Shouldn't get here */
DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode");
return true;
case PM_OUT_OF_DATA_SPACE:
return pool->pf.error_if_no_space;
case PM_READ_ONLY:
case PM_FAIL:
return true;
default:
/* Shouldn't get here */
DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode");
return true;
}
}
static void handle_unserviceable_bio(struct pool *pool, struct bio *bio)
{
if (should_error_unserviceable_bio(pool))
bio_io_error(bio);
else
retry_on_resume(bio);
......@@ -1007,11 +1045,20 @@ static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *c
struct bio *bio;
struct bio_list bios;
if (should_error_unserviceable_bio(pool)) {
cell_error(pool, cell);
return;
}
bio_list_init(&bios);
cell_release(pool, cell, &bios);
if (should_error_unserviceable_bio(pool))
while ((bio = bio_list_pop(&bios)))
bio_io_error(bio);
else
while ((bio = bio_list_pop(&bios)))
handle_unserviceable_bio(pool, bio);
retry_on_resume(bio);
}
static void process_discard(struct thin_c *tc, struct bio *bio)
......@@ -1296,6 +1343,11 @@ static void process_bio_read_only(struct thin_c *tc, struct bio *bio)
}
}
static void process_bio_success(struct thin_c *tc, struct bio *bio)
{
bio_endio(bio, 0);
}
static void process_bio_fail(struct thin_c *tc, struct bio *bio)
{
bio_io_error(bio);
......@@ -1328,6 +1380,11 @@ static void process_deferred_bios(struct pool *pool)
struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
struct thin_c *tc = h->tc;
if (tc->requeue_mode) {
bio_endio(bio, DM_ENDIO_REQUEUE);
continue;
}
/*
* If we've got no free new_mapping structs, and processing
* this bio might require one, we pause until there are some
......@@ -1394,51 +1451,134 @@ static void do_waker(struct work_struct *ws)
/*----------------------------------------------------------------*/
struct noflush_work {
struct work_struct worker;
struct thin_c *tc;
atomic_t complete;
wait_queue_head_t wait;
};
static void complete_noflush_work(struct noflush_work *w)
{
atomic_set(&w->complete, 1);
wake_up(&w->wait);
}
static void do_noflush_start(struct work_struct *ws)
{
struct noflush_work *w = container_of(ws, struct noflush_work, worker);
w->tc->requeue_mode = true;
requeue_io(w->tc);
complete_noflush_work(w);
}
static void do_noflush_stop(struct work_struct *ws)
{
struct noflush_work *w = container_of(ws, struct noflush_work, worker);
w->tc->requeue_mode = false;
complete_noflush_work(w);
}
static void noflush_work(struct thin_c *tc, void (*fn)(struct work_struct *))
{
struct noflush_work w;
INIT_WORK(&w.worker, fn);
w.tc = tc;
atomic_set(&w.complete, 0);
init_waitqueue_head(&w.wait);
queue_work(tc->pool->wq, &w.worker);
wait_event(w.wait, atomic_read(&w.complete));
}
/*----------------------------------------------------------------*/
static enum pool_mode get_pool_mode(struct pool *pool)
{
return pool->pf.mode;
}
static void notify_of_pool_mode_change(struct pool *pool, const char *new_mode)
{
dm_table_event(pool->ti->table);
DMINFO("%s: switching pool to %s mode",
dm_device_name(pool->pool_md), new_mode);
}
static void set_pool_mode(struct pool *pool, enum pool_mode new_mode)
{
int r;
enum pool_mode old_mode = pool->pf.mode;
struct pool_c *pt = pool->ti->private;
bool needs_check = dm_pool_metadata_needs_check(pool->pmd);
enum pool_mode old_mode = get_pool_mode(pool);
/*
* Never allow the pool to transition to PM_WRITE mode if user
* intervention is required to verify metadata and data consistency.
*/
if (new_mode == PM_WRITE && needs_check) {
DMERR("%s: unable to switch pool to write mode until repaired.",
dm_device_name(pool->pool_md));
if (old_mode != new_mode)
new_mode = old_mode;
else
new_mode = PM_READ_ONLY;
}
/*
* If we were in PM_FAIL mode, rollback of metadata failed. We're
* not going to recover without a thin_repair. So we never let the
* pool move out of the old mode.
*/
if (old_mode == PM_FAIL)
new_mode = old_mode;
switch (new_mode) {
case PM_FAIL:
if (old_mode != new_mode)
DMERR("%s: switching pool to failure mode",
dm_device_name(pool->pool_md));
notify_of_pool_mode_change(pool, "failure");
dm_pool_metadata_read_only(pool->pmd);
pool->process_bio = process_bio_fail;
pool->process_discard = process_bio_fail;
pool->process_prepared_mapping = process_prepared_mapping_fail;
pool->process_prepared_discard = process_prepared_discard_fail;
error_retry_list(pool);
break;
case PM_READ_ONLY:
if (old_mode != new_mode)
DMERR("%s: switching pool to read-only mode",
dm_device_name(pool->pool_md));
r = dm_pool_abort_metadata(pool->pmd);
if (r) {
DMERR("%s: aborting transaction failed",
dm_device_name(pool->pool_md));
new_mode = PM_FAIL;
set_pool_mode(pool, new_mode);
} else {
notify_of_pool_mode_change(pool, "read-only");
dm_pool_metadata_read_only(pool->pmd);
pool->process_bio = process_bio_read_only;
pool->process_discard = process_discard;
pool->process_discard = process_bio_success;
pool->process_prepared_mapping = process_prepared_mapping_fail;
pool->process_prepared_discard = process_prepared_discard_passdown;
}
error_retry_list(pool);
break;
case PM_OUT_OF_DATA_SPACE:
/*
* Ideally we'd never hit this state; the low water mark
* would trigger userland to extend the pool before we
* completely run out of data space. However, many small
* IOs to unprovisioned space can consume data space at an
* alarming rate. Adjust your low water mark if you're
* frequently seeing this mode.
*/
if (old_mode != new_mode)
notify_of_pool_mode_change(pool, "out-of-data-space");
pool->process_bio = process_bio_read_only;
pool->process_discard = process_discard;
pool->process_prepared_mapping = process_prepared_mapping;
pool->process_prepared_discard = process_prepared_discard_passdown;
break;
case PM_WRITE:
if (old_mode != new_mode)
DMINFO("%s: switching pool to write mode",
dm_device_name(pool->pool_md));
notify_of_pool_mode_change(pool, "write");
dm_pool_metadata_read_write(pool->pmd);
pool->process_bio = process_bio;
pool->process_discard = process_discard;
......@@ -1448,32 +1588,35 @@ static void set_pool_mode(struct pool *pool, enum pool_mode new_mode)
}
pool->pf.mode = new_mode;
/*
* The pool mode may have changed, sync it so bind_control_target()
* doesn't cause an unexpected mode transition on resume.
*/
pt->adjusted_pf.mode = new_mode;
}
/*
* Rather than calling set_pool_mode directly, use these which describe the
* reason for mode degradation.
*/
static void out_of_data_space(struct pool *pool)
static void abort_transaction(struct pool *pool)
{
DMERR_LIMIT("%s: no free data space available.",
dm_device_name(pool->pool_md));
set_pool_mode(pool, PM_READ_ONLY);
const char *dev_name = dm_device_name(pool->pool_md);
DMERR_LIMIT("%s: aborting current metadata transaction", dev_name);
if (dm_pool_abort_metadata(pool->pmd)) {
DMERR("%s: failed to abort metadata transaction", dev_name);
set_pool_mode(pool, PM_FAIL);
}
if (dm_pool_metadata_set_needs_check(pool->pmd)) {
DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name);
set_pool_mode(pool, PM_FAIL);
}
}
static void metadata_operation_failed(struct pool *pool, const char *op, int r)
{
dm_block_t free_blocks;
DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d",
dm_device_name(pool->pool_md), op, r);
if (r == -ENOSPC &&
!dm_pool_get_free_metadata_block_count(pool->pmd, &free_blocks) &&
!free_blocks)
DMERR_LIMIT("%s: no free metadata space available.",
dm_device_name(pool->pool_md));
abort_transaction(pool);
set_pool_mode(pool, PM_READ_ONLY);
}
......@@ -1524,6 +1667,11 @@ static int thin_bio_map(struct dm_target *ti, struct bio *bio)
thin_hook_bio(tc, bio);
if (tc->requeue_mode) {
bio_endio(bio, DM_ENDIO_REQUEUE);
return DM_MAPIO_SUBMITTED;
}
if (get_pool_mode(tc->pool) == PM_FAIL) {
bio_io_error(bio);
return DM_MAPIO_SUBMITTED;
......@@ -1687,7 +1835,7 @@ static int bind_control_target(struct pool *pool, struct dm_target *ti)
/*
* We want to make sure that a pool in PM_FAIL mode is never upgraded.
*/
enum pool_mode old_mode = pool->pf.mode;
enum pool_mode old_mode = get_pool_mode(pool);
enum pool_mode new_mode = pt->adjusted_pf.mode;
/*
......@@ -1701,16 +1849,6 @@ static int bind_control_target(struct pool *pool, struct dm_target *ti)
pool->pf = pt->adjusted_pf;
pool->low_water_blocks = pt->low_water_blocks;
/*
* If we were in PM_FAIL mode, rollback of metadata failed. We're
* not going to recover without a thin_repair. So we never let the
* pool move out of the old mode. On the other hand a PM_READ_ONLY
* may have been due to a lack of metadata or data space, and may
* now work (ie. if the underlying devices have been resized).
*/
if (old_mode == PM_FAIL)
new_mode = old_mode;
set_pool_mode(pool, new_mode);
return 0;
......@@ -2253,6 +2391,12 @@ static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit)
return -EINVAL;
} else if (data_size > sb_data_size) {
if (dm_pool_metadata_needs_check(pool->pmd)) {
DMERR("%s: unable to grow the data device until repaired.",
dm_device_name(pool->pool_md));
return 0;
}
if (sb_data_size)
DMINFO("%s: growing the data device from %llu to %llu blocks",
dm_device_name(pool->pool_md),
......@@ -2294,6 +2438,12 @@ static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit)
return -EINVAL;
} else if (metadata_dev_size > sb_metadata_dev_size) {
if (dm_pool_metadata_needs_check(pool->pmd)) {
DMERR("%s: unable to grow the metadata device until repaired.",
dm_device_name(pool->pool_md));
return 0;
}
warn_if_metadata_device_too_big(pool->md_dev);
DMINFO("%s: growing the metadata device from %llu to %llu blocks",
dm_device_name(pool->pool_md),
......@@ -2681,7 +2831,9 @@ static void pool_status(struct dm_target *ti, status_type_t type,
else
DMEMIT("- ");
if (pool->pf.mode == PM_READ_ONLY)
if (pool->pf.mode == PM_OUT_OF_DATA_SPACE)
DMEMIT("out_of_data_space ");
else if (pool->pf.mode == PM_READ_ONLY)
DMEMIT("ro ");
else
DMEMIT("rw ");
......@@ -2795,7 +2947,7 @@ static struct target_type pool_target = {
.name = "thin-pool",
.features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE |
DM_TARGET_IMMUTABLE,
.version = {1, 10, 0},
.version = {1, 11, 0},
.module = THIS_MODULE,
.ctr = pool_ctr,
.dtr = pool_dtr,
......@@ -2997,10 +3149,23 @@ static int thin_endio(struct dm_target *ti, struct bio *bio, int err)
return 0;
}
static void thin_postsuspend(struct dm_target *ti)
static void thin_presuspend(struct dm_target *ti)
{
struct thin_c *tc = ti->private;
if (dm_noflush_suspending(ti))
requeue_io((struct thin_c *)ti->private);
noflush_work(tc, do_noflush_start);
}
static void thin_postsuspend(struct dm_target *ti)
{
struct thin_c *tc = ti->private;
/*
* The dm_noflush_suspending flag has been cleared by now, so
* unfortunately we must always run this.
*/
noflush_work(tc, do_noflush_stop);
}
/*
......@@ -3085,12 +3250,13 @@ static int thin_iterate_devices(struct dm_target *ti,
static struct target_type thin_target = {
.name = "thin",
.version = {1, 10, 0},
.version = {1, 11, 0},
.module = THIS_MODULE,
.ctr = thin_ctr,
.dtr = thin_dtr,
.map = thin_map,
.end_io = thin_endio,
.presuspend = thin_presuspend,
.postsuspend = thin_postsuspend,
.status = thin_status,
.iterate_devices = thin_iterate_devices,
......
......@@ -6,3 +6,13 @@ config DM_PERSISTENT_DATA
---help---
Library providing immutable on-disk data structure support for
device-mapper targets such as the thin provisioning target.
config DM_DEBUG_BLOCK_STACK_TRACING
boolean "Keep stack trace of persistent data block lock holders"
depends on STACKTRACE_SUPPORT && DM_PERSISTENT_DATA
select STACKTRACE
---help---
Enable this for messages that may help debug problems with the
block manager locking used by thin provisioning and caching.
If unsure, say N.
......@@ -91,6 +91,69 @@ struct block_op {
dm_block_t block;
};
struct bop_ring_buffer {
unsigned begin;
unsigned end;
struct block_op bops[MAX_RECURSIVE_ALLOCATIONS + 1];
};
static void brb_init(struct bop_ring_buffer *brb)
{
brb->begin = 0;
brb->end = 0;
}
static bool brb_empty(struct bop_ring_buffer *brb)
{
return brb->begin == brb->end;
}
static unsigned brb_next(struct bop_ring_buffer *brb, unsigned old)
{
unsigned r = old + 1;
return (r >= (sizeof(brb->bops) / sizeof(*brb->bops))) ? 0 : r;
}
static int brb_push(struct bop_ring_buffer *brb,
enum block_op_type type, dm_block_t b)
{
struct block_op *bop;
unsigned next = brb_next(brb, brb->end);
/*
* We don't allow the last bop to be filled, this way we can
* differentiate between full and empty.
*/
if (next == brb->begin)
return -ENOMEM;
bop = brb->bops + brb->end;
bop->type = type;
bop->block = b;
brb->end = next;
return 0;
}
static int brb_pop(struct bop_ring_buffer *brb, struct block_op *result)
{
struct block_op *bop;
if (brb_empty(brb))
return -ENODATA;
bop = brb->bops + brb->begin;
result->type = bop->type;
result->block = bop->block;
brb->begin = brb_next(brb, brb->begin);
return 0;
}
/*----------------------------------------------------------------*/
struct sm_metadata {
struct dm_space_map sm;
......@@ -101,25 +164,20 @@ struct sm_metadata {
unsigned recursion_count;
unsigned allocated_this_transaction;
unsigned nr_uncommitted;
struct block_op uncommitted[MAX_RECURSIVE_ALLOCATIONS];
struct bop_ring_buffer uncommitted;
struct threshold threshold;
};
static int add_bop(struct sm_metadata *smm, enum block_op_type type, dm_block_t b)
{
struct block_op *op;
int r = brb_push(&smm->uncommitted, type, b);
if (smm->nr_uncommitted == MAX_RECURSIVE_ALLOCATIONS) {
if (r) {
DMERR("too many recursive allocations");
return -ENOMEM;
}
op = smm->uncommitted + smm->nr_uncommitted++;
op->type = type;
op->block = b;
return 0;
}
......@@ -158,11 +216,17 @@ static int out(struct sm_metadata *smm)
return -ENOMEM;
}
if (smm->recursion_count == 1 && smm->nr_uncommitted) {
while (smm->nr_uncommitted && !r) {
smm->nr_uncommitted--;
r = commit_bop(smm, smm->uncommitted +
smm->nr_uncommitted);
if (smm->recursion_count == 1) {
while (!brb_empty(&smm->uncommitted)) {
struct block_op bop;
r = brb_pop(&smm->uncommitted, &bop);
if (r) {
DMERR("bug in bop ring buffer");
break;
}
r = commit_bop(smm, &bop);
if (r)
break;
}
......@@ -217,7 +281,8 @@ static int sm_metadata_get_nr_free(struct dm_space_map *sm, dm_block_t *count)
static int sm_metadata_get_count(struct dm_space_map *sm, dm_block_t b,
uint32_t *result)
{
int r, i;
int r;
unsigned i;
struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
unsigned adjustment = 0;
......@@ -225,8 +290,10 @@ static int sm_metadata_get_count(struct dm_space_map *sm, dm_block_t b,
* We may have some uncommitted adjustments to add. This list
* should always be really short.
*/
for (i = 0; i < smm->nr_uncommitted; i++) {
struct block_op *op = smm->uncommitted + i;
for (i = smm->uncommitted.begin;
i != smm->uncommitted.end;
i = brb_next(&smm->uncommitted, i)) {
struct block_op *op = smm->uncommitted.bops + i;
if (op->block != b)
continue;
......@@ -254,7 +321,8 @@ static int sm_metadata_get_count(struct dm_space_map *sm, dm_block_t b,
static int sm_metadata_count_is_more_than_one(struct dm_space_map *sm,
dm_block_t b, int *result)
{
int r, i, adjustment = 0;
int r, adjustment = 0;
unsigned i;
struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
uint32_t rc;
......@@ -262,8 +330,11 @@ static int sm_metadata_count_is_more_than_one(struct dm_space_map *sm,
* We may have some uncommitted adjustments to add. This list
* should always be really short.
*/
for (i = 0; i < smm->nr_uncommitted; i++) {
struct block_op *op = smm->uncommitted + i;
for (i = smm->uncommitted.begin;
i != smm->uncommitted.end;
i = brb_next(&smm->uncommitted, i)) {
struct block_op *op = smm->uncommitted.bops + i;
if (op->block != b)
continue;
......@@ -671,7 +742,7 @@ int dm_sm_metadata_create(struct dm_space_map *sm,
smm->begin = superblock + 1;
smm->recursion_count = 0;
smm->allocated_this_transaction = 0;
smm->nr_uncommitted = 0;
brb_init(&smm->uncommitted);
threshold_init(&smm->threshold);
memcpy(&smm->sm, &bootstrap_ops, sizeof(smm->sm));
......@@ -715,7 +786,7 @@ int dm_sm_metadata_open(struct dm_space_map *sm,
smm->begin = 0;
smm->recursion_count = 0;
smm->allocated_this_transaction = 0;
smm->nr_uncommitted = 0;
brb_init(&smm->uncommitted);
threshold_init(&smm->threshold);
memcpy(&smm->old_ll, &smm->ll, sizeof(smm->old_ll));
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment