WSL2-Linux-Kernel/drivers/md/dm-thin.c

3016 строки
73 KiB
C
Исходник Обычный вид История

/*
* Copyright (C) 2011-2012 Red Hat UK.
*
* This file is released under the GPL.
*/
#include "dm-thin-metadata.h"
#include "dm-bio-prison.h"
#include "dm.h"
#include <linux/device-mapper.h>
#include <linux/dm-io.h>
#include <linux/dm-kcopyd.h>
#include <linux/list.h>
#include <linux/init.h>
#include <linux/module.h>
#include <linux/slab.h>
#define DM_MSG_PREFIX "thin"
/*
* Tunable constants
*/
#define ENDIO_HOOK_POOL_SIZE 1024
#define MAPPING_POOL_SIZE 1024
#define PRISON_CELLS 1024
#define COMMIT_PERIOD HZ
DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle,
"A percentage of time allocated for copy on write");
/*
* The block size of the device holding pool data must be
* between 64KB and 1GB.
*/
#define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT)
#define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
/*
* Device id is restricted to 24 bits.
*/
#define MAX_DEV_ID ((1 << 24) - 1)
/*
* How do we handle breaking sharing of data blocks?
* =================================================
*
* We use a standard copy-on-write btree to store the mappings for the
* devices (note I'm talking about copy-on-write of the metadata here, not
* the data). When you take an internal snapshot you clone the root node
* of the origin btree. After this there is no concept of an origin or a
* snapshot. They are just two device trees that happen to point to the
* same data blocks.
*
* When we get a write in we decide if it's to a shared data block using
* some timestamp magic. If it is, we have to break sharing.
*
* Let's say we write to a shared block in what was the origin. The
* steps are:
*
* i) plug io further to this physical block. (see bio_prison code).
*
* ii) quiesce any read io to that shared data block. Obviously
* including all devices that share this block. (see dm_deferred_set code)
*
* iii) copy the data block to a newly allocate block. This step can be
* missed out if the io covers the block. (schedule_copy).
*
* iv) insert the new mapping into the origin's btree
* (process_prepared_mapping). This act of inserting breaks some
* sharing of btree nodes between the two devices. Breaking sharing only
* effects the btree of that specific device. Btrees for the other
* devices that share the block never change. The btree for the origin
* device as it was after the last commit is untouched, ie. we're using
* persistent data structures in the functional programming sense.
*
* v) unplug io to this physical block, including the io that triggered
* the breaking of sharing.
*
* Steps (ii) and (iii) occur in parallel.
*
* The metadata _doesn't_ need to be committed before the io continues. We
* get away with this because the io is always written to a _new_ block.
* If there's a crash, then:
*
* - The origin mapping will point to the old origin block (the shared
* one). This will contain the data as it was before the io that triggered
* the breaking of sharing came in.
*
* - The snap mapping still points to the old block. As it would after
* the commit.
*
* The downside of this scheme is the timestamp magic isn't perfect, and
* will continue to think that data block in the snapshot device is shared
* even after the write to the origin has broken sharing. I suspect data
* blocks will typically be shared by many different devices, so we're
* breaking sharing n + 1 times, rather than n, where n is the number of
* devices that reference this data block. At the moment I think the
* benefits far, far outweigh the disadvantages.
*/
/*----------------------------------------------------------------*/
/*
* Key building.
*/
static void build_data_key(struct dm_thin_device *td,
dm_block_t b, struct dm_cell_key *key)
{
key->virtual = 0;
key->dev = dm_thin_dev_id(td);
key->block = b;
}
static void build_virtual_key(struct dm_thin_device *td, dm_block_t b,
struct dm_cell_key *key)
{
key->virtual = 1;
key->dev = dm_thin_dev_id(td);
key->block = b;
}
/*----------------------------------------------------------------*/
/*
* A pool device ties together a metadata device and a data device. It
* also provides the interface for creating and destroying internal
* devices.
*/
struct dm_thin_new_mapping;
/*
* The pool runs in 3 modes. Ordered in degraded order for comparisons.
*/
enum pool_mode {
PM_WRITE, /* metadata may be changed */
PM_READ_ONLY, /* metadata may not be changed */
PM_FAIL, /* all I/O fails */
};
struct pool_features {
enum pool_mode mode;
bool zero_new_blocks:1;
bool discard_enabled:1;
bool discard_passdown:1;
};
struct thin_c;
typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio);
typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m);
struct pool {
struct list_head list;
struct dm_target *ti; /* Only set if a pool target is bound */
struct mapped_device *pool_md;
struct block_device *md_dev;
struct dm_pool_metadata *pmd;
dm_block_t low_water_blocks;
uint32_t sectors_per_block;
int sectors_per_block_shift;
struct pool_features pf;
unsigned low_water_triggered:1; /* A dm event has been sent */
unsigned no_free_space:1; /* A -ENOSPC warning has been issued */
struct dm_bio_prison *prison;
struct dm_kcopyd_client *copier;
struct workqueue_struct *wq;
struct work_struct worker;
struct delayed_work waker;
unsigned long last_commit_jiffies;
unsigned ref_count;
spinlock_t lock;
struct bio_list deferred_bios;
struct bio_list deferred_flush_bios;
struct list_head prepared_mappings;
struct list_head prepared_discards;
struct bio_list retry_on_resume_list;
struct dm_deferred_set *shared_read_ds;
struct dm_deferred_set *all_io_ds;
struct dm_thin_new_mapping *next_mapping;
mempool_t *mapping_pool;
process_bio_fn process_bio;
process_bio_fn process_discard;
process_mapping_fn process_prepared_mapping;
process_mapping_fn process_prepared_discard;
};
static enum pool_mode get_pool_mode(struct pool *pool);
static void set_pool_mode(struct pool *pool, enum pool_mode mode);
/*
* Target context for a pool.
*/
struct pool_c {
struct dm_target *ti;
struct pool *pool;
struct dm_dev *data_dev;
struct dm_dev *metadata_dev;
struct dm_target_callbacks callbacks;
dm_block_t low_water_blocks;
struct pool_features requested_pf; /* Features requested during table load */
struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */
};
/*
* Target context for a thin.
*/
struct thin_c {
struct dm_dev *pool_dev;
struct dm_dev *origin_dev;
dm_thin_id dev_id;
struct pool *pool;
struct dm_thin_device *td;
};
/*----------------------------------------------------------------*/
/*
* wake_worker() is used when new work is queued and when pool_resume is
* ready to continue deferred IO processing.
*/
static void wake_worker(struct pool *pool)
{
queue_work(pool->wq, &pool->worker);
}
/*----------------------------------------------------------------*/
static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio,
struct dm_bio_prison_cell **cell_result)
{
int r;
struct dm_bio_prison_cell *cell_prealloc;
/*
* Allocate a cell from the prison's mempool.
* This might block but it can't fail.
*/
cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO);
r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result);
if (r)
/*
* We reused an old cell; we can get rid of
* the new one.
*/
dm_bio_prison_free_cell(pool->prison, cell_prealloc);
return r;
}
static void cell_release(struct pool *pool,
struct dm_bio_prison_cell *cell,
struct bio_list *bios)
{
dm_cell_release(pool->prison, cell, bios);
dm_bio_prison_free_cell(pool->prison, cell);
}
static void cell_release_no_holder(struct pool *pool,
struct dm_bio_prison_cell *cell,
struct bio_list *bios)
{
dm_cell_release_no_holder(pool->prison, cell, bios);
dm_bio_prison_free_cell(pool->prison, cell);
}
static void cell_defer_no_holder_no_free(struct thin_c *tc,
struct dm_bio_prison_cell *cell)
{
struct pool *pool = tc->pool;
unsigned long flags;
spin_lock_irqsave(&pool->lock, flags);
dm_cell_release_no_holder(pool->prison, cell, &pool->deferred_bios);
spin_unlock_irqrestore(&pool->lock, flags);
wake_worker(pool);
}
static void cell_error(struct pool *pool,
struct dm_bio_prison_cell *cell)
{
dm_cell_error(pool->prison, cell);
dm_bio_prison_free_cell(pool->prison, cell);
}
/*----------------------------------------------------------------*/
/*
* A global list of pools that uses a struct mapped_device as a key.
*/
static struct dm_thin_pool_table {
struct mutex mutex;
struct list_head pools;
} dm_thin_pool_table;
static void pool_table_init(void)
{
mutex_init(&dm_thin_pool_table.mutex);
INIT_LIST_HEAD(&dm_thin_pool_table.pools);
}
static void __pool_table_insert(struct pool *pool)
{
BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
list_add(&pool->list, &dm_thin_pool_table.pools);
}
static void __pool_table_remove(struct pool *pool)
{
BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
list_del(&pool->list);
}
static struct pool *__pool_table_lookup(struct mapped_device *md)
{
struct pool *pool = NULL, *tmp;
BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
if (tmp->pool_md == md) {
pool = tmp;
break;
}
}
return pool;
}
static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev)
{
struct pool *pool = NULL, *tmp;
BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
if (tmp->md_dev == md_dev) {
pool = tmp;
break;
}
}
return pool;
}
/*----------------------------------------------------------------*/
struct dm_thin_endio_hook {
struct thin_c *tc;
struct dm_deferred_entry *shared_read_entry;
struct dm_deferred_entry *all_io_entry;
struct dm_thin_new_mapping *overwrite_mapping;
};
static void __requeue_bio_list(struct thin_c *tc, struct bio_list *master)
{
struct bio *bio;
struct bio_list bios;
bio_list_init(&bios);
bio_list_merge(&bios, master);
bio_list_init(master);
while ((bio = bio_list_pop(&bios))) {
struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
if (h->tc == tc)
bio_endio(bio, DM_ENDIO_REQUEUE);
else
bio_list_add(master, bio);
}
}
static void requeue_io(struct thin_c *tc)
{
struct pool *pool = tc->pool;
unsigned long flags;
spin_lock_irqsave(&pool->lock, flags);
__requeue_bio_list(tc, &pool->deferred_bios);
__requeue_bio_list(tc, &pool->retry_on_resume_list);
spin_unlock_irqrestore(&pool->lock, flags);
}
/*
* This section of code contains the logic for processing a thin device's IO.
* Much of the code depends on pool object resources (lists, workqueues, etc)
* but most is exclusively called from the thin target rather than the thin-pool
* target.
*/
static bool block_size_is_power_of_two(struct pool *pool)
{
return pool->sectors_per_block_shift >= 0;
}
static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio)
{
struct pool *pool = tc->pool;
sector_t block_nr = bio->bi_sector;
if (block_size_is_power_of_two(pool))
block_nr >>= pool->sectors_per_block_shift;
else
(void) sector_div(block_nr, pool->sectors_per_block);
return block_nr;
}
static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block)
{
struct pool *pool = tc->pool;
sector_t bi_sector = bio->bi_sector;
bio->bi_bdev = tc->pool_dev->bdev;
if (block_size_is_power_of_two(pool))
bio->bi_sector = (block << pool->sectors_per_block_shift) |
(bi_sector & (pool->sectors_per_block - 1));
else
bio->bi_sector = (block * pool->sectors_per_block) +
sector_div(bi_sector, pool->sectors_per_block);
}
static void remap_to_origin(struct thin_c *tc, struct bio *bio)
{
bio->bi_bdev = tc->origin_dev->bdev;
}
static int bio_triggers_commit(struct thin_c *tc, struct bio *bio)
{
return (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) &&
dm_thin_changed_this_transaction(tc->td);
}
dm thin: fix race between simultaneous io and discards to same block There is a race when discard bios and non-discard bios are issued simultaneously to the same block. Discard support is expensive for all thin devices precisely because you have to be careful to quiesce the area you're discarding. DM thin must handle this conflicting IO pattern (simultaneous non-discard vs discard) even though a sane application shouldn't be issuing such IO. The race manifests as follows: 1. A non-discard bio is mapped in thin_bio_map. This doesn't lock out parallel activity to the same block. 2. A discard bio is issued to the same block as the non-discard bio. 3. The discard bio is locked in a dm_bio_prison_cell in process_discard to lock out parallel activity against the same block. 4. The non-discard bio's mapping continues and its all_io_entry is incremented so the bio is accounted for in the thin pool's all_io_ds which is a dm_deferred_set used to track time locality of non-discard IO. 5. The non-discard bio is finally locked in a dm_bio_prison_cell in process_bio. The race can result in deadlock, leaving the block layer hanging waiting for completion of a discard bio that never completes, e.g.: INFO: task ruby:15354 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. ruby D ffffffff8160f0e0 0 15354 15314 0x00000000 ffff8802fb08bc58 0000000000000082 ffff8802fb08bfd8 0000000000012900 ffff8802fb08a010 0000000000012900 0000000000012900 0000000000012900 ffff8802fb08bfd8 0000000000012900 ffff8803324b9480 ffff88032c6f14c0 Call Trace: [<ffffffff814e5a19>] schedule+0x29/0x70 [<ffffffff814e3d85>] schedule_timeout+0x195/0x220 [<ffffffffa06b9bc1>] ? _dm_request+0x111/0x160 [dm_mod] [<ffffffff814e589e>] wait_for_common+0x11e/0x190 [<ffffffff8107a170>] ? try_to_wake_up+0x2b0/0x2b0 [<ffffffff814e59ed>] wait_for_completion+0x1d/0x20 [<ffffffff81233289>] blkdev_issue_discard+0x219/0x260 [<ffffffff81233e79>] blkdev_ioctl+0x6e9/0x7b0 [<ffffffff8119a65c>] block_ioctl+0x3c/0x40 [<ffffffff8117539c>] do_vfs_ioctl+0x8c/0x340 [<ffffffff8119a547>] ? block_llseek+0x67/0xb0 [<ffffffff811756f1>] sys_ioctl+0xa1/0xb0 [<ffffffff810561f6>] ? sys_rt_sigprocmask+0x86/0xd0 [<ffffffff814ef099>] system_call_fastpath+0x16/0x1b The thinp-test-suite's test_discard_random_sectors reliably hits this deadlock on fast SSD storage. The fix for this race is that the all_io_entry for a bio must be incremented whilst the dm_bio_prison_cell is held for the bio's associated virtual and physical blocks. That cell locking wasn't occurring early enough in thin_bio_map. This patch fixes this. Care is taken to always call the new function inc_all_io_entry() with the relevant cells locked, but they are generally unlocked before calling issue() to try to avoid holding the cells locked across generic_submit_request. Also, now that thin_bio_map may lock bios in a cell, process_bio() is no longer the only thread that will do so. Because of this we must be sure to use cell_defer_except() to release all non-holder entries, that were added by the other thread, because they must be deferred. This patch depends on "dm thin: replace dm_cell_release_singleton with cell_defer_except". Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Cc: stable@vger.kernel.org
2012-12-22 00:23:31 +04:00
static void inc_all_io_entry(struct pool *pool, struct bio *bio)
{
struct dm_thin_endio_hook *h;
if (bio->bi_rw & REQ_DISCARD)
return;
h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
dm thin: fix race between simultaneous io and discards to same block There is a race when discard bios and non-discard bios are issued simultaneously to the same block. Discard support is expensive for all thin devices precisely because you have to be careful to quiesce the area you're discarding. DM thin must handle this conflicting IO pattern (simultaneous non-discard vs discard) even though a sane application shouldn't be issuing such IO. The race manifests as follows: 1. A non-discard bio is mapped in thin_bio_map. This doesn't lock out parallel activity to the same block. 2. A discard bio is issued to the same block as the non-discard bio. 3. The discard bio is locked in a dm_bio_prison_cell in process_discard to lock out parallel activity against the same block. 4. The non-discard bio's mapping continues and its all_io_entry is incremented so the bio is accounted for in the thin pool's all_io_ds which is a dm_deferred_set used to track time locality of non-discard IO. 5. The non-discard bio is finally locked in a dm_bio_prison_cell in process_bio. The race can result in deadlock, leaving the block layer hanging waiting for completion of a discard bio that never completes, e.g.: INFO: task ruby:15354 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. ruby D ffffffff8160f0e0 0 15354 15314 0x00000000 ffff8802fb08bc58 0000000000000082 ffff8802fb08bfd8 0000000000012900 ffff8802fb08a010 0000000000012900 0000000000012900 0000000000012900 ffff8802fb08bfd8 0000000000012900 ffff8803324b9480 ffff88032c6f14c0 Call Trace: [<ffffffff814e5a19>] schedule+0x29/0x70 [<ffffffff814e3d85>] schedule_timeout+0x195/0x220 [<ffffffffa06b9bc1>] ? _dm_request+0x111/0x160 [dm_mod] [<ffffffff814e589e>] wait_for_common+0x11e/0x190 [<ffffffff8107a170>] ? try_to_wake_up+0x2b0/0x2b0 [<ffffffff814e59ed>] wait_for_completion+0x1d/0x20 [<ffffffff81233289>] blkdev_issue_discard+0x219/0x260 [<ffffffff81233e79>] blkdev_ioctl+0x6e9/0x7b0 [<ffffffff8119a65c>] block_ioctl+0x3c/0x40 [<ffffffff8117539c>] do_vfs_ioctl+0x8c/0x340 [<ffffffff8119a547>] ? block_llseek+0x67/0xb0 [<ffffffff811756f1>] sys_ioctl+0xa1/0xb0 [<ffffffff810561f6>] ? sys_rt_sigprocmask+0x86/0xd0 [<ffffffff814ef099>] system_call_fastpath+0x16/0x1b The thinp-test-suite's test_discard_random_sectors reliably hits this deadlock on fast SSD storage. The fix for this race is that the all_io_entry for a bio must be incremented whilst the dm_bio_prison_cell is held for the bio's associated virtual and physical blocks. That cell locking wasn't occurring early enough in thin_bio_map. This patch fixes this. Care is taken to always call the new function inc_all_io_entry() with the relevant cells locked, but they are generally unlocked before calling issue() to try to avoid holding the cells locked across generic_submit_request. Also, now that thin_bio_map may lock bios in a cell, process_bio() is no longer the only thread that will do so. Because of this we must be sure to use cell_defer_except() to release all non-holder entries, that were added by the other thread, because they must be deferred. This patch depends on "dm thin: replace dm_cell_release_singleton with cell_defer_except". Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Cc: stable@vger.kernel.org
2012-12-22 00:23:31 +04:00
h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds);
}
static void issue(struct thin_c *tc, struct bio *bio)
{
struct pool *pool = tc->pool;
unsigned long flags;
if (!bio_triggers_commit(tc, bio)) {
generic_make_request(bio);
return;
}
/*
* Complete bio with an error if earlier I/O caused changes to
* the metadata that can't be committed e.g, due to I/O errors
* on the metadata device.
*/
if (dm_thin_aborted_changes(tc->td)) {
bio_io_error(bio);
return;
}
/*
* Batch together any bios that trigger commits and then issue a
* single commit for them in process_deferred_bios().
*/
spin_lock_irqsave(&pool->lock, flags);
bio_list_add(&pool->deferred_flush_bios, bio);
spin_unlock_irqrestore(&pool->lock, flags);
}
static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio)
{
remap_to_origin(tc, bio);
issue(tc, bio);
}
static void remap_and_issue(struct thin_c *tc, struct bio *bio,
dm_block_t block)
{
remap(tc, bio, block);
issue(tc, bio);
}
/*----------------------------------------------------------------*/
/*
* Bio endio functions.
*/
struct dm_thin_new_mapping {
struct list_head list;
unsigned quiesced:1;
unsigned prepared:1;
unsigned pass_discard:1;
struct thin_c *tc;
dm_block_t virt_block;
dm_block_t data_block;
struct dm_bio_prison_cell *cell, *cell2;
int err;
/*
* If the bio covers the whole area of a block then we can avoid
* zeroing or copying. Instead this bio is hooked. The bio will
* still be in the cell, so care has to be taken to avoid issuing
* the bio twice.
*/
struct bio *bio;
bio_end_io_t *saved_bi_end_io;
};
static void __maybe_add_mapping(struct dm_thin_new_mapping *m)
{
struct pool *pool = m->tc->pool;
if (m->quiesced && m->prepared) {
list_add(&m->list, &pool->prepared_mappings);
wake_worker(pool);
}
}
static void copy_complete(int read_err, unsigned long write_err, void *context)
{
unsigned long flags;
struct dm_thin_new_mapping *m = context;
struct pool *pool = m->tc->pool;
m->err = read_err || write_err ? -EIO : 0;
spin_lock_irqsave(&pool->lock, flags);
m->prepared = 1;
__maybe_add_mapping(m);
spin_unlock_irqrestore(&pool->lock, flags);
}
static void overwrite_endio(struct bio *bio, int err)
{
unsigned long flags;
struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
struct dm_thin_new_mapping *m = h->overwrite_mapping;
struct pool *pool = m->tc->pool;
m->err = err;
spin_lock_irqsave(&pool->lock, flags);
m->prepared = 1;
__maybe_add_mapping(m);
spin_unlock_irqrestore(&pool->lock, flags);
}
/*----------------------------------------------------------------*/
/*
* Workqueue.
*/
/*
* Prepared mapping jobs.
*/
/*
* This sends the bios in the cell back to the deferred_bios list.
*/
static void cell_defer(struct thin_c *tc, struct dm_bio_prison_cell *cell)
{
struct pool *pool = tc->pool;
unsigned long flags;
spin_lock_irqsave(&pool->lock, flags);
cell_release(pool, cell, &pool->deferred_bios);
spin_unlock_irqrestore(&tc->pool->lock, flags);
wake_worker(pool);
}
/*
* Same as cell_defer above, except it omits the original holder of the cell.
*/
static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell)
{
struct pool *pool = tc->pool;
unsigned long flags;
spin_lock_irqsave(&pool->lock, flags);
cell_release_no_holder(pool, cell, &pool->deferred_bios);
spin_unlock_irqrestore(&pool->lock, flags);
wake_worker(pool);
}
static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m)
{
if (m->bio)
m->bio->bi_end_io = m->saved_bi_end_io;
cell_error(m->tc->pool, m->cell);
list_del(&m->list);
mempool_free(m, m->tc->pool->mapping_pool);
}
static void process_prepared_mapping(struct dm_thin_new_mapping *m)
{
struct thin_c *tc = m->tc;
struct pool *pool = tc->pool;
struct bio *bio;
int r;
bio = m->bio;
if (bio)
bio->bi_end_io = m->saved_bi_end_io;
if (m->err) {
cell_error(pool, m->cell);
goto out;
}
/*
* Commit the prepared block into the mapping btree.
* Any I/O for this block arriving after this point will get
* remapped to it directly.
*/
r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block);
if (r) {
DMERR_LIMIT("dm_thin_insert_block() failed");
cell_error(pool, m->cell);
goto out;
}
/*
* Release any bios held while the block was being provisioned.
* If we are processing a write bio that completely covers the block,
* we already processed it so can ignore it now when processing
* the bios in the cell.
*/
if (bio) {
cell_defer_no_holder(tc, m->cell);
bio_endio(bio, 0);
} else
cell_defer(tc, m->cell);
out:
list_del(&m->list);
mempool_free(m, pool->mapping_pool);
}
static void process_prepared_discard_fail(struct dm_thin_new_mapping *m)
{
struct thin_c *tc = m->tc;
bio_io_error(m->bio);
cell_defer_no_holder(tc, m->cell);
cell_defer_no_holder(tc, m->cell2);
mempool_free(m, tc->pool->mapping_pool);
}
static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m)
{
struct thin_c *tc = m->tc;
dm thin: fix race between simultaneous io and discards to same block There is a race when discard bios and non-discard bios are issued simultaneously to the same block. Discard support is expensive for all thin devices precisely because you have to be careful to quiesce the area you're discarding. DM thin must handle this conflicting IO pattern (simultaneous non-discard vs discard) even though a sane application shouldn't be issuing such IO. The race manifests as follows: 1. A non-discard bio is mapped in thin_bio_map. This doesn't lock out parallel activity to the same block. 2. A discard bio is issued to the same block as the non-discard bio. 3. The discard bio is locked in a dm_bio_prison_cell in process_discard to lock out parallel activity against the same block. 4. The non-discard bio's mapping continues and its all_io_entry is incremented so the bio is accounted for in the thin pool's all_io_ds which is a dm_deferred_set used to track time locality of non-discard IO. 5. The non-discard bio is finally locked in a dm_bio_prison_cell in process_bio. The race can result in deadlock, leaving the block layer hanging waiting for completion of a discard bio that never completes, e.g.: INFO: task ruby:15354 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. ruby D ffffffff8160f0e0 0 15354 15314 0x00000000 ffff8802fb08bc58 0000000000000082 ffff8802fb08bfd8 0000000000012900 ffff8802fb08a010 0000000000012900 0000000000012900 0000000000012900 ffff8802fb08bfd8 0000000000012900 ffff8803324b9480 ffff88032c6f14c0 Call Trace: [<ffffffff814e5a19>] schedule+0x29/0x70 [<ffffffff814e3d85>] schedule_timeout+0x195/0x220 [<ffffffffa06b9bc1>] ? _dm_request+0x111/0x160 [dm_mod] [<ffffffff814e589e>] wait_for_common+0x11e/0x190 [<ffffffff8107a170>] ? try_to_wake_up+0x2b0/0x2b0 [<ffffffff814e59ed>] wait_for_completion+0x1d/0x20 [<ffffffff81233289>] blkdev_issue_discard+0x219/0x260 [<ffffffff81233e79>] blkdev_ioctl+0x6e9/0x7b0 [<ffffffff8119a65c>] block_ioctl+0x3c/0x40 [<ffffffff8117539c>] do_vfs_ioctl+0x8c/0x340 [<ffffffff8119a547>] ? block_llseek+0x67/0xb0 [<ffffffff811756f1>] sys_ioctl+0xa1/0xb0 [<ffffffff810561f6>] ? sys_rt_sigprocmask+0x86/0xd0 [<ffffffff814ef099>] system_call_fastpath+0x16/0x1b The thinp-test-suite's test_discard_random_sectors reliably hits this deadlock on fast SSD storage. The fix for this race is that the all_io_entry for a bio must be incremented whilst the dm_bio_prison_cell is held for the bio's associated virtual and physical blocks. That cell locking wasn't occurring early enough in thin_bio_map. This patch fixes this. Care is taken to always call the new function inc_all_io_entry() with the relevant cells locked, but they are generally unlocked before calling issue() to try to avoid holding the cells locked across generic_submit_request. Also, now that thin_bio_map may lock bios in a cell, process_bio() is no longer the only thread that will do so. Because of this we must be sure to use cell_defer_except() to release all non-holder entries, that were added by the other thread, because they must be deferred. This patch depends on "dm thin: replace dm_cell_release_singleton with cell_defer_except". Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Cc: stable@vger.kernel.org
2012-12-22 00:23:31 +04:00
inc_all_io_entry(tc->pool, m->bio);
cell_defer_no_holder(tc, m->cell);
cell_defer_no_holder(tc, m->cell2);
dm thin: fix race between simultaneous io and discards to same block There is a race when discard bios and non-discard bios are issued simultaneously to the same block. Discard support is expensive for all thin devices precisely because you have to be careful to quiesce the area you're discarding. DM thin must handle this conflicting IO pattern (simultaneous non-discard vs discard) even though a sane application shouldn't be issuing such IO. The race manifests as follows: 1. A non-discard bio is mapped in thin_bio_map. This doesn't lock out parallel activity to the same block. 2. A discard bio is issued to the same block as the non-discard bio. 3. The discard bio is locked in a dm_bio_prison_cell in process_discard to lock out parallel activity against the same block. 4. The non-discard bio's mapping continues and its all_io_entry is incremented so the bio is accounted for in the thin pool's all_io_ds which is a dm_deferred_set used to track time locality of non-discard IO. 5. The non-discard bio is finally locked in a dm_bio_prison_cell in process_bio. The race can result in deadlock, leaving the block layer hanging waiting for completion of a discard bio that never completes, e.g.: INFO: task ruby:15354 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. ruby D ffffffff8160f0e0 0 15354 15314 0x00000000 ffff8802fb08bc58 0000000000000082 ffff8802fb08bfd8 0000000000012900 ffff8802fb08a010 0000000000012900 0000000000012900 0000000000012900 ffff8802fb08bfd8 0000000000012900 ffff8803324b9480 ffff88032c6f14c0 Call Trace: [<ffffffff814e5a19>] schedule+0x29/0x70 [<ffffffff814e3d85>] schedule_timeout+0x195/0x220 [<ffffffffa06b9bc1>] ? _dm_request+0x111/0x160 [dm_mod] [<ffffffff814e589e>] wait_for_common+0x11e/0x190 [<ffffffff8107a170>] ? try_to_wake_up+0x2b0/0x2b0 [<ffffffff814e59ed>] wait_for_completion+0x1d/0x20 [<ffffffff81233289>] blkdev_issue_discard+0x219/0x260 [<ffffffff81233e79>] blkdev_ioctl+0x6e9/0x7b0 [<ffffffff8119a65c>] block_ioctl+0x3c/0x40 [<ffffffff8117539c>] do_vfs_ioctl+0x8c/0x340 [<ffffffff8119a547>] ? block_llseek+0x67/0xb0 [<ffffffff811756f1>] sys_ioctl+0xa1/0xb0 [<ffffffff810561f6>] ? sys_rt_sigprocmask+0x86/0xd0 [<ffffffff814ef099>] system_call_fastpath+0x16/0x1b The thinp-test-suite's test_discard_random_sectors reliably hits this deadlock on fast SSD storage. The fix for this race is that the all_io_entry for a bio must be incremented whilst the dm_bio_prison_cell is held for the bio's associated virtual and physical blocks. That cell locking wasn't occurring early enough in thin_bio_map. This patch fixes this. Care is taken to always call the new function inc_all_io_entry() with the relevant cells locked, but they are generally unlocked before calling issue() to try to avoid holding the cells locked across generic_submit_request. Also, now that thin_bio_map may lock bios in a cell, process_bio() is no longer the only thread that will do so. Because of this we must be sure to use cell_defer_except() to release all non-holder entries, that were added by the other thread, because they must be deferred. This patch depends on "dm thin: replace dm_cell_release_singleton with cell_defer_except". Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Cc: stable@vger.kernel.org
2012-12-22 00:23:31 +04:00
if (m->pass_discard)
remap_and_issue(tc, m->bio, m->data_block);
else
bio_endio(m->bio, 0);
mempool_free(m, tc->pool->mapping_pool);
}
static void process_prepared_discard(struct dm_thin_new_mapping *m)
{
int r;
struct thin_c *tc = m->tc;
r = dm_thin_remove_block(tc->td, m->virt_block);
if (r)
DMERR_LIMIT("dm_thin_remove_block() failed");
process_prepared_discard_passdown(m);
}
static void process_prepared(struct pool *pool, struct list_head *head,
process_mapping_fn *fn)
{
unsigned long flags;
struct list_head maps;
struct dm_thin_new_mapping *m, *tmp;
INIT_LIST_HEAD(&maps);
spin_lock_irqsave(&pool->lock, flags);
list_splice_init(head, &maps);
spin_unlock_irqrestore(&pool->lock, flags);
list_for_each_entry_safe(m, tmp, &maps, list)
(*fn)(m);
}
/*
* Deferred bio jobs.
*/
static int io_overlaps_block(struct pool *pool, struct bio *bio)
{
return bio->bi_size == (pool->sectors_per_block << SECTOR_SHIFT);
}
static int io_overwrites_block(struct pool *pool, struct bio *bio)
{
return (bio_data_dir(bio) == WRITE) &&
io_overlaps_block(pool, bio);
}
static void save_and_set_endio(struct bio *bio, bio_end_io_t **save,
bio_end_io_t *fn)
{
*save = bio->bi_end_io;
bio->bi_end_io = fn;
}
static int ensure_next_mapping(struct pool *pool)
{
if (pool->next_mapping)
return 0;
pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC);
return pool->next_mapping ? 0 : -ENOMEM;
}
static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool)
{
struct dm_thin_new_mapping *r = pool->next_mapping;
BUG_ON(!pool->next_mapping);
pool->next_mapping = NULL;
return r;
}
static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
struct dm_dev *origin, dm_block_t data_origin,
dm_block_t data_dest,
struct dm_bio_prison_cell *cell, struct bio *bio)
{
int r;
struct pool *pool = tc->pool;
struct dm_thin_new_mapping *m = get_next_mapping(pool);
INIT_LIST_HEAD(&m->list);
m->quiesced = 0;
m->prepared = 0;
m->tc = tc;
m->virt_block = virt_block;
m->data_block = data_dest;
m->cell = cell;
m->err = 0;
m->bio = NULL;
if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list))
m->quiesced = 1;
/*
* IO to pool_dev remaps to the pool target's data_dev.
*
* If the whole block of data is being overwritten, we can issue the
* bio immediately. Otherwise we use kcopyd to clone the data first.
*/
if (io_overwrites_block(pool, bio)) {
struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
h->overwrite_mapping = m;
m->bio = bio;
save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
dm thin: fix race between simultaneous io and discards to same block There is a race when discard bios and non-discard bios are issued simultaneously to the same block. Discard support is expensive for all thin devices precisely because you have to be careful to quiesce the area you're discarding. DM thin must handle this conflicting IO pattern (simultaneous non-discard vs discard) even though a sane application shouldn't be issuing such IO. The race manifests as follows: 1. A non-discard bio is mapped in thin_bio_map. This doesn't lock out parallel activity to the same block. 2. A discard bio is issued to the same block as the non-discard bio. 3. The discard bio is locked in a dm_bio_prison_cell in process_discard to lock out parallel activity against the same block. 4. The non-discard bio's mapping continues and its all_io_entry is incremented so the bio is accounted for in the thin pool's all_io_ds which is a dm_deferred_set used to track time locality of non-discard IO. 5. The non-discard bio is finally locked in a dm_bio_prison_cell in process_bio. The race can result in deadlock, leaving the block layer hanging waiting for completion of a discard bio that never completes, e.g.: INFO: task ruby:15354 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. ruby D ffffffff8160f0e0 0 15354 15314 0x00000000 ffff8802fb08bc58 0000000000000082 ffff8802fb08bfd8 0000000000012900 ffff8802fb08a010 0000000000012900 0000000000012900 0000000000012900 ffff8802fb08bfd8 0000000000012900 ffff8803324b9480 ffff88032c6f14c0 Call Trace: [<ffffffff814e5a19>] schedule+0x29/0x70 [<ffffffff814e3d85>] schedule_timeout+0x195/0x220 [<ffffffffa06b9bc1>] ? _dm_request+0x111/0x160 [dm_mod] [<ffffffff814e589e>] wait_for_common+0x11e/0x190 [<ffffffff8107a170>] ? try_to_wake_up+0x2b0/0x2b0 [<ffffffff814e59ed>] wait_for_completion+0x1d/0x20 [<ffffffff81233289>] blkdev_issue_discard+0x219/0x260 [<ffffffff81233e79>] blkdev_ioctl+0x6e9/0x7b0 [<ffffffff8119a65c>] block_ioctl+0x3c/0x40 [<ffffffff8117539c>] do_vfs_ioctl+0x8c/0x340 [<ffffffff8119a547>] ? block_llseek+0x67/0xb0 [<ffffffff811756f1>] sys_ioctl+0xa1/0xb0 [<ffffffff810561f6>] ? sys_rt_sigprocmask+0x86/0xd0 [<ffffffff814ef099>] system_call_fastpath+0x16/0x1b The thinp-test-suite's test_discard_random_sectors reliably hits this deadlock on fast SSD storage. The fix for this race is that the all_io_entry for a bio must be incremented whilst the dm_bio_prison_cell is held for the bio's associated virtual and physical blocks. That cell locking wasn't occurring early enough in thin_bio_map. This patch fixes this. Care is taken to always call the new function inc_all_io_entry() with the relevant cells locked, but they are generally unlocked before calling issue() to try to avoid holding the cells locked across generic_submit_request. Also, now that thin_bio_map may lock bios in a cell, process_bio() is no longer the only thread that will do so. Because of this we must be sure to use cell_defer_except() to release all non-holder entries, that were added by the other thread, because they must be deferred. This patch depends on "dm thin: replace dm_cell_release_singleton with cell_defer_except". Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Cc: stable@vger.kernel.org
2012-12-22 00:23:31 +04:00
inc_all_io_entry(pool, bio);
remap_and_issue(tc, bio, data_dest);
} else {
struct dm_io_region from, to;
from.bdev = origin->bdev;
from.sector = data_origin * pool->sectors_per_block;
from.count = pool->sectors_per_block;
to.bdev = tc->pool_dev->bdev;
to.sector = data_dest * pool->sectors_per_block;
to.count = pool->sectors_per_block;
r = dm_kcopyd_copy(pool->copier, &from, 1, &to,
0, copy_complete, m);
if (r < 0) {
mempool_free(m, pool->mapping_pool);
DMERR_LIMIT("dm_kcopyd_copy() failed");
cell_error(pool, cell);
}
}
}
static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block,
dm_block_t data_origin, dm_block_t data_dest,
struct dm_bio_prison_cell *cell, struct bio *bio)
{
schedule_copy(tc, virt_block, tc->pool_dev,
data_origin, data_dest, cell, bio);
}
static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block,
dm_block_t data_dest,
struct dm_bio_prison_cell *cell, struct bio *bio)
{
schedule_copy(tc, virt_block, tc->origin_dev,
virt_block, data_dest, cell, bio);
}
static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
dm_block_t data_block, struct dm_bio_prison_cell *cell,
struct bio *bio)
{
struct pool *pool = tc->pool;
struct dm_thin_new_mapping *m = get_next_mapping(pool);
INIT_LIST_HEAD(&m->list);
m->quiesced = 1;
m->prepared = 0;
m->tc = tc;
m->virt_block = virt_block;
m->data_block = data_block;
m->cell = cell;
m->err = 0;
m->bio = NULL;
/*
* If the whole block of data is being overwritten or we are not
* zeroing pre-existing data, we can issue the bio immediately.
* Otherwise we use kcopyd to zero the data first.
*/
if (!pool->pf.zero_new_blocks)
process_prepared_mapping(m);
else if (io_overwrites_block(pool, bio)) {
struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
h->overwrite_mapping = m;
m->bio = bio;
save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
dm thin: fix race between simultaneous io and discards to same block There is a race when discard bios and non-discard bios are issued simultaneously to the same block. Discard support is expensive for all thin devices precisely because you have to be careful to quiesce the area you're discarding. DM thin must handle this conflicting IO pattern (simultaneous non-discard vs discard) even though a sane application shouldn't be issuing such IO. The race manifests as follows: 1. A non-discard bio is mapped in thin_bio_map. This doesn't lock out parallel activity to the same block. 2. A discard bio is issued to the same block as the non-discard bio. 3. The discard bio is locked in a dm_bio_prison_cell in process_discard to lock out parallel activity against the same block. 4. The non-discard bio's mapping continues and its all_io_entry is incremented so the bio is accounted for in the thin pool's all_io_ds which is a dm_deferred_set used to track time locality of non-discard IO. 5. The non-discard bio is finally locked in a dm_bio_prison_cell in process_bio. The race can result in deadlock, leaving the block layer hanging waiting for completion of a discard bio that never completes, e.g.: INFO: task ruby:15354 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. ruby D ffffffff8160f0e0 0 15354 15314 0x00000000 ffff8802fb08bc58 0000000000000082 ffff8802fb08bfd8 0000000000012900 ffff8802fb08a010 0000000000012900 0000000000012900 0000000000012900 ffff8802fb08bfd8 0000000000012900 ffff8803324b9480 ffff88032c6f14c0 Call Trace: [<ffffffff814e5a19>] schedule+0x29/0x70 [<ffffffff814e3d85>] schedule_timeout+0x195/0x220 [<ffffffffa06b9bc1>] ? _dm_request+0x111/0x160 [dm_mod] [<ffffffff814e589e>] wait_for_common+0x11e/0x190 [<ffffffff8107a170>] ? try_to_wake_up+0x2b0/0x2b0 [<ffffffff814e59ed>] wait_for_completion+0x1d/0x20 [<ffffffff81233289>] blkdev_issue_discard+0x219/0x260 [<ffffffff81233e79>] blkdev_ioctl+0x6e9/0x7b0 [<ffffffff8119a65c>] block_ioctl+0x3c/0x40 [<ffffffff8117539c>] do_vfs_ioctl+0x8c/0x340 [<ffffffff8119a547>] ? block_llseek+0x67/0xb0 [<ffffffff811756f1>] sys_ioctl+0xa1/0xb0 [<ffffffff810561f6>] ? sys_rt_sigprocmask+0x86/0xd0 [<ffffffff814ef099>] system_call_fastpath+0x16/0x1b The thinp-test-suite's test_discard_random_sectors reliably hits this deadlock on fast SSD storage. The fix for this race is that the all_io_entry for a bio must be incremented whilst the dm_bio_prison_cell is held for the bio's associated virtual and physical blocks. That cell locking wasn't occurring early enough in thin_bio_map. This patch fixes this. Care is taken to always call the new function inc_all_io_entry() with the relevant cells locked, but they are generally unlocked before calling issue() to try to avoid holding the cells locked across generic_submit_request. Also, now that thin_bio_map may lock bios in a cell, process_bio() is no longer the only thread that will do so. Because of this we must be sure to use cell_defer_except() to release all non-holder entries, that were added by the other thread, because they must be deferred. This patch depends on "dm thin: replace dm_cell_release_singleton with cell_defer_except". Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Cc: stable@vger.kernel.org
2012-12-22 00:23:31 +04:00
inc_all_io_entry(pool, bio);
remap_and_issue(tc, bio, data_block);
} else {
int r;
struct dm_io_region to;
to.bdev = tc->pool_dev->bdev;
to.sector = data_block * pool->sectors_per_block;
to.count = pool->sectors_per_block;
r = dm_kcopyd_zero(pool->copier, 1, &to, 0, copy_complete, m);
if (r < 0) {
mempool_free(m, pool->mapping_pool);
DMERR_LIMIT("dm_kcopyd_zero() failed");
cell_error(pool, cell);
}
}
}
static int commit(struct pool *pool)
{
int r;
r = dm_pool_commit_metadata(pool->pmd);
if (r)
DMERR_LIMIT("commit failed: error = %d", r);
return r;
}
/*
* A non-zero return indicates read_only or fail_io mode.
* Many callers don't care about the return value.
*/
static int commit_or_fallback(struct pool *pool)
{
int r;
if (get_pool_mode(pool) != PM_WRITE)
return -EINVAL;
r = commit(pool);
if (r)
set_pool_mode(pool, PM_READ_ONLY);
return r;
}
static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
{
int r;
dm_block_t free_blocks;
unsigned long flags;
struct pool *pool = tc->pool;
r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
if (r)
return r;
if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) {
DMWARN("%s: reached low water mark for data device: sending event.",
dm_device_name(pool->pool_md));
spin_lock_irqsave(&pool->lock, flags);
pool->low_water_triggered = 1;
spin_unlock_irqrestore(&pool->lock, flags);
dm_table_event(pool->ti->table);
}
if (!free_blocks) {
if (pool->no_free_space)
return -ENOSPC;
else {
/*
* Try to commit to see if that will free up some
* more space.
*/
(void) commit_or_fallback(pool);
r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
if (r)
return r;
/*
* If we still have no space we set a flag to avoid
* doing all this checking and return -ENOSPC.
*/
if (!free_blocks) {
DMWARN("%s: no free space available.",
dm_device_name(pool->pool_md));
spin_lock_irqsave(&pool->lock, flags);
pool->no_free_space = 1;
spin_unlock_irqrestore(&pool->lock, flags);
return -ENOSPC;
}
}
}
r = dm_pool_alloc_data_block(pool->pmd, result);
if (r)
return r;
return 0;
}
/*
* If we have run out of space, queue bios until the device is
* resumed, presumably after having been reloaded with more space.
*/
static void retry_on_resume(struct bio *bio)
{
struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
struct thin_c *tc = h->tc;
struct pool *pool = tc->pool;
unsigned long flags;
spin_lock_irqsave(&pool->lock, flags);
bio_list_add(&pool->retry_on_resume_list, bio);
spin_unlock_irqrestore(&pool->lock, flags);
}
static void no_space(struct pool *pool, struct dm_bio_prison_cell *cell)
{
struct bio *bio;
struct bio_list bios;
bio_list_init(&bios);
cell_release(pool, cell, &bios);
while ((bio = bio_list_pop(&bios)))
retry_on_resume(bio);
}
static void process_discard(struct thin_c *tc, struct bio *bio)
{
int r;
dm thin: fix unprotected use of prepared_discards list Fix two places in commit 104655fd4dce ("dm thin: support discards") that didn't use pool->lock to protect against concurrent changes to the prepared_discards list. Without this fix, thin_endio() can race with process_discard(), leading to concurrent list_add()s that result in the processes locking up with an error like the following: WARNING: at lib/list_debug.c:32 __list_add+0x8f/0xa0() ... list_add corruption. next->prev should be prev (ffff880323b96140), but was ffff8801d2c48440. (next=ffff8801d2c485c0). ... Pid: 17205, comm: kworker/u:1 Tainted: G W O 3.4.0-rc3.snitm+ #1 Call Trace: [<ffffffff8103ca1f>] warn_slowpath_common+0x7f/0xc0 [<ffffffff8103cb16>] warn_slowpath_fmt+0x46/0x50 [<ffffffffa04f6ce6>] ? bio_detain+0xc6/0x210 [dm_thin_pool] [<ffffffff8124ff3f>] __list_add+0x8f/0xa0 [<ffffffffa04f70d2>] process_discard+0x2a2/0x2d0 [dm_thin_pool] [<ffffffffa04f6a78>] ? remap_and_issue+0x38/0x50 [dm_thin_pool] [<ffffffffa04f7c3b>] process_deferred_bios+0x7b/0x230 [dm_thin_pool] [<ffffffffa04f7df0>] ? process_deferred_bios+0x230/0x230 [dm_thin_pool] [<ffffffffa04f7e42>] do_worker+0x52/0x60 [dm_thin_pool] [<ffffffff81056fa9>] process_one_work+0x129/0x450 [<ffffffff81059b9c>] worker_thread+0x17c/0x3c0 [<ffffffff81059a20>] ? manage_workers+0x120/0x120 [<ffffffff8105eabe>] kthread+0x9e/0xb0 [<ffffffff814ceda4>] kernel_thread_helper+0x4/0x10 [<ffffffff8105ea20>] ? kthread_freezable_should_stop+0x70/0x70 [<ffffffff814ceda0>] ? gs_change+0x13/0x13 ---[ end trace 7e0a523bc5e52692 ]--- Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com>
2012-05-12 04:43:16 +04:00
unsigned long flags;
struct pool *pool = tc->pool;
struct dm_bio_prison_cell *cell, *cell2;
struct dm_cell_key key, key2;
dm_block_t block = get_bio_block(tc, bio);
struct dm_thin_lookup_result lookup_result;
struct dm_thin_new_mapping *m;
build_virtual_key(tc->td, block, &key);
if (bio_detain(tc->pool, &key, bio, &cell))
return;
r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
switch (r) {
case 0:
/*
* Check nobody is fiddling with this pool block. This can
* happen if someone's in the process of breaking sharing
* on this block.
*/
build_data_key(tc->td, lookup_result.block, &key2);
if (bio_detain(tc->pool, &key2, bio, &cell2)) {
cell_defer_no_holder(tc, cell);
break;
}
if (io_overlaps_block(pool, bio)) {
/*
* IO may still be going to the destination block. We must
* quiesce before we can do the removal.
*/
m = get_next_mapping(pool);
m->tc = tc;
m->pass_discard = (!lookup_result.shared) && pool->pf.discard_passdown;
m->virt_block = block;
m->data_block = lookup_result.block;
m->cell = cell;
m->cell2 = cell2;
m->err = 0;
m->bio = bio;
if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) {
dm thin: fix unprotected use of prepared_discards list Fix two places in commit 104655fd4dce ("dm thin: support discards") that didn't use pool->lock to protect against concurrent changes to the prepared_discards list. Without this fix, thin_endio() can race with process_discard(), leading to concurrent list_add()s that result in the processes locking up with an error like the following: WARNING: at lib/list_debug.c:32 __list_add+0x8f/0xa0() ... list_add corruption. next->prev should be prev (ffff880323b96140), but was ffff8801d2c48440. (next=ffff8801d2c485c0). ... Pid: 17205, comm: kworker/u:1 Tainted: G W O 3.4.0-rc3.snitm+ #1 Call Trace: [<ffffffff8103ca1f>] warn_slowpath_common+0x7f/0xc0 [<ffffffff8103cb16>] warn_slowpath_fmt+0x46/0x50 [<ffffffffa04f6ce6>] ? bio_detain+0xc6/0x210 [dm_thin_pool] [<ffffffff8124ff3f>] __list_add+0x8f/0xa0 [<ffffffffa04f70d2>] process_discard+0x2a2/0x2d0 [dm_thin_pool] [<ffffffffa04f6a78>] ? remap_and_issue+0x38/0x50 [dm_thin_pool] [<ffffffffa04f7c3b>] process_deferred_bios+0x7b/0x230 [dm_thin_pool] [<ffffffffa04f7df0>] ? process_deferred_bios+0x230/0x230 [dm_thin_pool] [<ffffffffa04f7e42>] do_worker+0x52/0x60 [dm_thin_pool] [<ffffffff81056fa9>] process_one_work+0x129/0x450 [<ffffffff81059b9c>] worker_thread+0x17c/0x3c0 [<ffffffff81059a20>] ? manage_workers+0x120/0x120 [<ffffffff8105eabe>] kthread+0x9e/0xb0 [<ffffffff814ceda4>] kernel_thread_helper+0x4/0x10 [<ffffffff8105ea20>] ? kthread_freezable_should_stop+0x70/0x70 [<ffffffff814ceda0>] ? gs_change+0x13/0x13 ---[ end trace 7e0a523bc5e52692 ]--- Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com>
2012-05-12 04:43:16 +04:00
spin_lock_irqsave(&pool->lock, flags);
list_add(&m->list, &pool->prepared_discards);
dm thin: fix unprotected use of prepared_discards list Fix two places in commit 104655fd4dce ("dm thin: support discards") that didn't use pool->lock to protect against concurrent changes to the prepared_discards list. Without this fix, thin_endio() can race with process_discard(), leading to concurrent list_add()s that result in the processes locking up with an error like the following: WARNING: at lib/list_debug.c:32 __list_add+0x8f/0xa0() ... list_add corruption. next->prev should be prev (ffff880323b96140), but was ffff8801d2c48440. (next=ffff8801d2c485c0). ... Pid: 17205, comm: kworker/u:1 Tainted: G W O 3.4.0-rc3.snitm+ #1 Call Trace: [<ffffffff8103ca1f>] warn_slowpath_common+0x7f/0xc0 [<ffffffff8103cb16>] warn_slowpath_fmt+0x46/0x50 [<ffffffffa04f6ce6>] ? bio_detain+0xc6/0x210 [dm_thin_pool] [<ffffffff8124ff3f>] __list_add+0x8f/0xa0 [<ffffffffa04f70d2>] process_discard+0x2a2/0x2d0 [dm_thin_pool] [<ffffffffa04f6a78>] ? remap_and_issue+0x38/0x50 [dm_thin_pool] [<ffffffffa04f7c3b>] process_deferred_bios+0x7b/0x230 [dm_thin_pool] [<ffffffffa04f7df0>] ? process_deferred_bios+0x230/0x230 [dm_thin_pool] [<ffffffffa04f7e42>] do_worker+0x52/0x60 [dm_thin_pool] [<ffffffff81056fa9>] process_one_work+0x129/0x450 [<ffffffff81059b9c>] worker_thread+0x17c/0x3c0 [<ffffffff81059a20>] ? manage_workers+0x120/0x120 [<ffffffff8105eabe>] kthread+0x9e/0xb0 [<ffffffff814ceda4>] kernel_thread_helper+0x4/0x10 [<ffffffff8105ea20>] ? kthread_freezable_should_stop+0x70/0x70 [<ffffffff814ceda0>] ? gs_change+0x13/0x13 ---[ end trace 7e0a523bc5e52692 ]--- Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com>
2012-05-12 04:43:16 +04:00
spin_unlock_irqrestore(&pool->lock, flags);
wake_worker(pool);
}
} else {
dm thin: fix race between simultaneous io and discards to same block There is a race when discard bios and non-discard bios are issued simultaneously to the same block. Discard support is expensive for all thin devices precisely because you have to be careful to quiesce the area you're discarding. DM thin must handle this conflicting IO pattern (simultaneous non-discard vs discard) even though a sane application shouldn't be issuing such IO. The race manifests as follows: 1. A non-discard bio is mapped in thin_bio_map. This doesn't lock out parallel activity to the same block. 2. A discard bio is issued to the same block as the non-discard bio. 3. The discard bio is locked in a dm_bio_prison_cell in process_discard to lock out parallel activity against the same block. 4. The non-discard bio's mapping continues and its all_io_entry is incremented so the bio is accounted for in the thin pool's all_io_ds which is a dm_deferred_set used to track time locality of non-discard IO. 5. The non-discard bio is finally locked in a dm_bio_prison_cell in process_bio. The race can result in deadlock, leaving the block layer hanging waiting for completion of a discard bio that never completes, e.g.: INFO: task ruby:15354 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. ruby D ffffffff8160f0e0 0 15354 15314 0x00000000 ffff8802fb08bc58 0000000000000082 ffff8802fb08bfd8 0000000000012900 ffff8802fb08a010 0000000000012900 0000000000012900 0000000000012900 ffff8802fb08bfd8 0000000000012900 ffff8803324b9480 ffff88032c6f14c0 Call Trace: [<ffffffff814e5a19>] schedule+0x29/0x70 [<ffffffff814e3d85>] schedule_timeout+0x195/0x220 [<ffffffffa06b9bc1>] ? _dm_request+0x111/0x160 [dm_mod] [<ffffffff814e589e>] wait_for_common+0x11e/0x190 [<ffffffff8107a170>] ? try_to_wake_up+0x2b0/0x2b0 [<ffffffff814e59ed>] wait_for_completion+0x1d/0x20 [<ffffffff81233289>] blkdev_issue_discard+0x219/0x260 [<ffffffff81233e79>] blkdev_ioctl+0x6e9/0x7b0 [<ffffffff8119a65c>] block_ioctl+0x3c/0x40 [<ffffffff8117539c>] do_vfs_ioctl+0x8c/0x340 [<ffffffff8119a547>] ? block_llseek+0x67/0xb0 [<ffffffff811756f1>] sys_ioctl+0xa1/0xb0 [<ffffffff810561f6>] ? sys_rt_sigprocmask+0x86/0xd0 [<ffffffff814ef099>] system_call_fastpath+0x16/0x1b The thinp-test-suite's test_discard_random_sectors reliably hits this deadlock on fast SSD storage. The fix for this race is that the all_io_entry for a bio must be incremented whilst the dm_bio_prison_cell is held for the bio's associated virtual and physical blocks. That cell locking wasn't occurring early enough in thin_bio_map. This patch fixes this. Care is taken to always call the new function inc_all_io_entry() with the relevant cells locked, but they are generally unlocked before calling issue() to try to avoid holding the cells locked across generic_submit_request. Also, now that thin_bio_map may lock bios in a cell, process_bio() is no longer the only thread that will do so. Because of this we must be sure to use cell_defer_except() to release all non-holder entries, that were added by the other thread, because they must be deferred. This patch depends on "dm thin: replace dm_cell_release_singleton with cell_defer_except". Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Cc: stable@vger.kernel.org
2012-12-22 00:23:31 +04:00
inc_all_io_entry(pool, bio);
cell_defer_no_holder(tc, cell);
cell_defer_no_holder(tc, cell2);
dm thin: fix race between simultaneous io and discards to same block There is a race when discard bios and non-discard bios are issued simultaneously to the same block. Discard support is expensive for all thin devices precisely because you have to be careful to quiesce the area you're discarding. DM thin must handle this conflicting IO pattern (simultaneous non-discard vs discard) even though a sane application shouldn't be issuing such IO. The race manifests as follows: 1. A non-discard bio is mapped in thin_bio_map. This doesn't lock out parallel activity to the same block. 2. A discard bio is issued to the same block as the non-discard bio. 3. The discard bio is locked in a dm_bio_prison_cell in process_discard to lock out parallel activity against the same block. 4. The non-discard bio's mapping continues and its all_io_entry is incremented so the bio is accounted for in the thin pool's all_io_ds which is a dm_deferred_set used to track time locality of non-discard IO. 5. The non-discard bio is finally locked in a dm_bio_prison_cell in process_bio. The race can result in deadlock, leaving the block layer hanging waiting for completion of a discard bio that never completes, e.g.: INFO: task ruby:15354 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. ruby D ffffffff8160f0e0 0 15354 15314 0x00000000 ffff8802fb08bc58 0000000000000082 ffff8802fb08bfd8 0000000000012900 ffff8802fb08a010 0000000000012900 0000000000012900 0000000000012900 ffff8802fb08bfd8 0000000000012900 ffff8803324b9480 ffff88032c6f14c0 Call Trace: [<ffffffff814e5a19>] schedule+0x29/0x70 [<ffffffff814e3d85>] schedule_timeout+0x195/0x220 [<ffffffffa06b9bc1>] ? _dm_request+0x111/0x160 [dm_mod] [<ffffffff814e589e>] wait_for_common+0x11e/0x190 [<ffffffff8107a170>] ? try_to_wake_up+0x2b0/0x2b0 [<ffffffff814e59ed>] wait_for_completion+0x1d/0x20 [<ffffffff81233289>] blkdev_issue_discard+0x219/0x260 [<ffffffff81233e79>] blkdev_ioctl+0x6e9/0x7b0 [<ffffffff8119a65c>] block_ioctl+0x3c/0x40 [<ffffffff8117539c>] do_vfs_ioctl+0x8c/0x340 [<ffffffff8119a547>] ? block_llseek+0x67/0xb0 [<ffffffff811756f1>] sys_ioctl+0xa1/0xb0 [<ffffffff810561f6>] ? sys_rt_sigprocmask+0x86/0xd0 [<ffffffff814ef099>] system_call_fastpath+0x16/0x1b The thinp-test-suite's test_discard_random_sectors reliably hits this deadlock on fast SSD storage. The fix for this race is that the all_io_entry for a bio must be incremented whilst the dm_bio_prison_cell is held for the bio's associated virtual and physical blocks. That cell locking wasn't occurring early enough in thin_bio_map. This patch fixes this. Care is taken to always call the new function inc_all_io_entry() with the relevant cells locked, but they are generally unlocked before calling issue() to try to avoid holding the cells locked across generic_submit_request. Also, now that thin_bio_map may lock bios in a cell, process_bio() is no longer the only thread that will do so. Because of this we must be sure to use cell_defer_except() to release all non-holder entries, that were added by the other thread, because they must be deferred. This patch depends on "dm thin: replace dm_cell_release_singleton with cell_defer_except". Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Cc: stable@vger.kernel.org
2012-12-22 00:23:31 +04:00
/*
* The DM core makes sure that the discard doesn't span
* a block boundary. So we submit the discard of a
* partial block appropriately.
*/
if ((!lookup_result.shared) && pool->pf.discard_passdown)
remap_and_issue(tc, bio, lookup_result.block);
else
bio_endio(bio, 0);
}
break;
case -ENODATA:
/*
* It isn't provisioned, just forget it.
*/
cell_defer_no_holder(tc, cell);
bio_endio(bio, 0);
break;
default:
DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
__func__, r);
cell_defer_no_holder(tc, cell);
bio_io_error(bio);
break;
}
}
static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block,
struct dm_cell_key *key,
struct dm_thin_lookup_result *lookup_result,
struct dm_bio_prison_cell *cell)
{
int r;
dm_block_t data_block;
r = alloc_data_block(tc, &data_block);
switch (r) {
case 0:
schedule_internal_copy(tc, block, lookup_result->block,
data_block, cell, bio);
break;
case -ENOSPC:
no_space(tc->pool, cell);
break;
default:
DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
__func__, r);
cell_error(tc->pool, cell);
break;
}
}
static void process_shared_bio(struct thin_c *tc, struct bio *bio,
dm_block_t block,
struct dm_thin_lookup_result *lookup_result)
{
struct dm_bio_prison_cell *cell;
struct pool *pool = tc->pool;
struct dm_cell_key key;
/*
* If cell is already occupied, then sharing is already in the process
* of being broken so we have nothing further to do here.
*/
build_data_key(tc->td, lookup_result->block, &key);
if (bio_detain(pool, &key, bio, &cell))
return;
if (bio_data_dir(bio) == WRITE && bio->bi_size)
break_sharing(tc, bio, block, &key, lookup_result, cell);
else {
struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds);
dm thin: fix race between simultaneous io and discards to same block There is a race when discard bios and non-discard bios are issued simultaneously to the same block. Discard support is expensive for all thin devices precisely because you have to be careful to quiesce the area you're discarding. DM thin must handle this conflicting IO pattern (simultaneous non-discard vs discard) even though a sane application shouldn't be issuing such IO. The race manifests as follows: 1. A non-discard bio is mapped in thin_bio_map. This doesn't lock out parallel activity to the same block. 2. A discard bio is issued to the same block as the non-discard bio. 3. The discard bio is locked in a dm_bio_prison_cell in process_discard to lock out parallel activity against the same block. 4. The non-discard bio's mapping continues and its all_io_entry is incremented so the bio is accounted for in the thin pool's all_io_ds which is a dm_deferred_set used to track time locality of non-discard IO. 5. The non-discard bio is finally locked in a dm_bio_prison_cell in process_bio. The race can result in deadlock, leaving the block layer hanging waiting for completion of a discard bio that never completes, e.g.: INFO: task ruby:15354 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. ruby D ffffffff8160f0e0 0 15354 15314 0x00000000 ffff8802fb08bc58 0000000000000082 ffff8802fb08bfd8 0000000000012900 ffff8802fb08a010 0000000000012900 0000000000012900 0000000000012900 ffff8802fb08bfd8 0000000000012900 ffff8803324b9480 ffff88032c6f14c0 Call Trace: [<ffffffff814e5a19>] schedule+0x29/0x70 [<ffffffff814e3d85>] schedule_timeout+0x195/0x220 [<ffffffffa06b9bc1>] ? _dm_request+0x111/0x160 [dm_mod] [<ffffffff814e589e>] wait_for_common+0x11e/0x190 [<ffffffff8107a170>] ? try_to_wake_up+0x2b0/0x2b0 [<ffffffff814e59ed>] wait_for_completion+0x1d/0x20 [<ffffffff81233289>] blkdev_issue_discard+0x219/0x260 [<ffffffff81233e79>] blkdev_ioctl+0x6e9/0x7b0 [<ffffffff8119a65c>] block_ioctl+0x3c/0x40 [<ffffffff8117539c>] do_vfs_ioctl+0x8c/0x340 [<ffffffff8119a547>] ? block_llseek+0x67/0xb0 [<ffffffff811756f1>] sys_ioctl+0xa1/0xb0 [<ffffffff810561f6>] ? sys_rt_sigprocmask+0x86/0xd0 [<ffffffff814ef099>] system_call_fastpath+0x16/0x1b The thinp-test-suite's test_discard_random_sectors reliably hits this deadlock on fast SSD storage. The fix for this race is that the all_io_entry for a bio must be incremented whilst the dm_bio_prison_cell is held for the bio's associated virtual and physical blocks. That cell locking wasn't occurring early enough in thin_bio_map. This patch fixes this. Care is taken to always call the new function inc_all_io_entry() with the relevant cells locked, but they are generally unlocked before calling issue() to try to avoid holding the cells locked across generic_submit_request. Also, now that thin_bio_map may lock bios in a cell, process_bio() is no longer the only thread that will do so. Because of this we must be sure to use cell_defer_except() to release all non-holder entries, that were added by the other thread, because they must be deferred. This patch depends on "dm thin: replace dm_cell_release_singleton with cell_defer_except". Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Cc: stable@vger.kernel.org
2012-12-22 00:23:31 +04:00
inc_all_io_entry(pool, bio);
cell_defer_no_holder(tc, cell);
dm thin: fix race between simultaneous io and discards to same block There is a race when discard bios and non-discard bios are issued simultaneously to the same block. Discard support is expensive for all thin devices precisely because you have to be careful to quiesce the area you're discarding. DM thin must handle this conflicting IO pattern (simultaneous non-discard vs discard) even though a sane application shouldn't be issuing such IO. The race manifests as follows: 1. A non-discard bio is mapped in thin_bio_map. This doesn't lock out parallel activity to the same block. 2. A discard bio is issued to the same block as the non-discard bio. 3. The discard bio is locked in a dm_bio_prison_cell in process_discard to lock out parallel activity against the same block. 4. The non-discard bio's mapping continues and its all_io_entry is incremented so the bio is accounted for in the thin pool's all_io_ds which is a dm_deferred_set used to track time locality of non-discard IO. 5. The non-discard bio is finally locked in a dm_bio_prison_cell in process_bio. The race can result in deadlock, leaving the block layer hanging waiting for completion of a discard bio that never completes, e.g.: INFO: task ruby:15354 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. ruby D ffffffff8160f0e0 0 15354 15314 0x00000000 ffff8802fb08bc58 0000000000000082 ffff8802fb08bfd8 0000000000012900 ffff8802fb08a010 0000000000012900 0000000000012900 0000000000012900 ffff8802fb08bfd8 0000000000012900 ffff8803324b9480 ffff88032c6f14c0 Call Trace: [<ffffffff814e5a19>] schedule+0x29/0x70 [<ffffffff814e3d85>] schedule_timeout+0x195/0x220 [<ffffffffa06b9bc1>] ? _dm_request+0x111/0x160 [dm_mod] [<ffffffff814e589e>] wait_for_common+0x11e/0x190 [<ffffffff8107a170>] ? try_to_wake_up+0x2b0/0x2b0 [<ffffffff814e59ed>] wait_for_completion+0x1d/0x20 [<ffffffff81233289>] blkdev_issue_discard+0x219/0x260 [<ffffffff81233e79>] blkdev_ioctl+0x6e9/0x7b0 [<ffffffff8119a65c>] block_ioctl+0x3c/0x40 [<ffffffff8117539c>] do_vfs_ioctl+0x8c/0x340 [<ffffffff8119a547>] ? block_llseek+0x67/0xb0 [<ffffffff811756f1>] sys_ioctl+0xa1/0xb0 [<ffffffff810561f6>] ? sys_rt_sigprocmask+0x86/0xd0 [<ffffffff814ef099>] system_call_fastpath+0x16/0x1b The thinp-test-suite's test_discard_random_sectors reliably hits this deadlock on fast SSD storage. The fix for this race is that the all_io_entry for a bio must be incremented whilst the dm_bio_prison_cell is held for the bio's associated virtual and physical blocks. That cell locking wasn't occurring early enough in thin_bio_map. This patch fixes this. Care is taken to always call the new function inc_all_io_entry() with the relevant cells locked, but they are generally unlocked before calling issue() to try to avoid holding the cells locked across generic_submit_request. Also, now that thin_bio_map may lock bios in a cell, process_bio() is no longer the only thread that will do so. Because of this we must be sure to use cell_defer_except() to release all non-holder entries, that were added by the other thread, because they must be deferred. This patch depends on "dm thin: replace dm_cell_release_singleton with cell_defer_except". Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Cc: stable@vger.kernel.org
2012-12-22 00:23:31 +04:00
remap_and_issue(tc, bio, lookup_result->block);
}
}
static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block,
struct dm_bio_prison_cell *cell)
{
int r;
dm_block_t data_block;
struct pool *pool = tc->pool;
/*
* Remap empty bios (flushes) immediately, without provisioning.
*/
if (!bio->bi_size) {
inc_all_io_entry(pool, bio);
cell_defer_no_holder(tc, cell);
dm thin: fix race between simultaneous io and discards to same block There is a race when discard bios and non-discard bios are issued simultaneously to the same block. Discard support is expensive for all thin devices precisely because you have to be careful to quiesce the area you're discarding. DM thin must handle this conflicting IO pattern (simultaneous non-discard vs discard) even though a sane application shouldn't be issuing such IO. The race manifests as follows: 1. A non-discard bio is mapped in thin_bio_map. This doesn't lock out parallel activity to the same block. 2. A discard bio is issued to the same block as the non-discard bio. 3. The discard bio is locked in a dm_bio_prison_cell in process_discard to lock out parallel activity against the same block. 4. The non-discard bio's mapping continues and its all_io_entry is incremented so the bio is accounted for in the thin pool's all_io_ds which is a dm_deferred_set used to track time locality of non-discard IO. 5. The non-discard bio is finally locked in a dm_bio_prison_cell in process_bio. The race can result in deadlock, leaving the block layer hanging waiting for completion of a discard bio that never completes, e.g.: INFO: task ruby:15354 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. ruby D ffffffff8160f0e0 0 15354 15314 0x00000000 ffff8802fb08bc58 0000000000000082 ffff8802fb08bfd8 0000000000012900 ffff8802fb08a010 0000000000012900 0000000000012900 0000000000012900 ffff8802fb08bfd8 0000000000012900 ffff8803324b9480 ffff88032c6f14c0 Call Trace: [<ffffffff814e5a19>] schedule+0x29/0x70 [<ffffffff814e3d85>] schedule_timeout+0x195/0x220 [<ffffffffa06b9bc1>] ? _dm_request+0x111/0x160 [dm_mod] [<ffffffff814e589e>] wait_for_common+0x11e/0x190 [<ffffffff8107a170>] ? try_to_wake_up+0x2b0/0x2b0 [<ffffffff814e59ed>] wait_for_completion+0x1d/0x20 [<ffffffff81233289>] blkdev_issue_discard+0x219/0x260 [<ffffffff81233e79>] blkdev_ioctl+0x6e9/0x7b0 [<ffffffff8119a65c>] block_ioctl+0x3c/0x40 [<ffffffff8117539c>] do_vfs_ioctl+0x8c/0x340 [<ffffffff8119a547>] ? block_llseek+0x67/0xb0 [<ffffffff811756f1>] sys_ioctl+0xa1/0xb0 [<ffffffff810561f6>] ? sys_rt_sigprocmask+0x86/0xd0 [<ffffffff814ef099>] system_call_fastpath+0x16/0x1b The thinp-test-suite's test_discard_random_sectors reliably hits this deadlock on fast SSD storage. The fix for this race is that the all_io_entry for a bio must be incremented whilst the dm_bio_prison_cell is held for the bio's associated virtual and physical blocks. That cell locking wasn't occurring early enough in thin_bio_map. This patch fixes this. Care is taken to always call the new function inc_all_io_entry() with the relevant cells locked, but they are generally unlocked before calling issue() to try to avoid holding the cells locked across generic_submit_request. Also, now that thin_bio_map may lock bios in a cell, process_bio() is no longer the only thread that will do so. Because of this we must be sure to use cell_defer_except() to release all non-holder entries, that were added by the other thread, because they must be deferred. This patch depends on "dm thin: replace dm_cell_release_singleton with cell_defer_except". Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Cc: stable@vger.kernel.org
2012-12-22 00:23:31 +04:00
remap_and_issue(tc, bio, 0);
return;
}
/*
* Fill read bios with zeroes and complete them immediately.
*/
if (bio_data_dir(bio) == READ) {
zero_fill_bio(bio);
cell_defer_no_holder(tc, cell);
bio_endio(bio, 0);
return;
}
r = alloc_data_block(tc, &data_block);
switch (r) {
case 0:
if (tc->origin_dev)
schedule_external_copy(tc, block, data_block, cell, bio);
else
schedule_zero(tc, block, data_block, cell, bio);
break;
case -ENOSPC:
no_space(pool, cell);
break;
default:
DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
__func__, r);
set_pool_mode(pool, PM_READ_ONLY);
cell_error(pool, cell);
break;
}
}
static void process_bio(struct thin_c *tc, struct bio *bio)
{
int r;
struct pool *pool = tc->pool;
dm_block_t block = get_bio_block(tc, bio);
struct dm_bio_prison_cell *cell;
struct dm_cell_key key;
struct dm_thin_lookup_result lookup_result;
/*
* If cell is already occupied, then the block is already
* being provisioned so we have nothing further to do here.
*/
build_virtual_key(tc->td, block, &key);
if (bio_detain(pool, &key, bio, &cell))
return;
r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
switch (r) {
case 0:
dm thin: fix race between simultaneous io and discards to same block There is a race when discard bios and non-discard bios are issued simultaneously to the same block. Discard support is expensive for all thin devices precisely because you have to be careful to quiesce the area you're discarding. DM thin must handle this conflicting IO pattern (simultaneous non-discard vs discard) even though a sane application shouldn't be issuing such IO. The race manifests as follows: 1. A non-discard bio is mapped in thin_bio_map. This doesn't lock out parallel activity to the same block. 2. A discard bio is issued to the same block as the non-discard bio. 3. The discard bio is locked in a dm_bio_prison_cell in process_discard to lock out parallel activity against the same block. 4. The non-discard bio's mapping continues and its all_io_entry is incremented so the bio is accounted for in the thin pool's all_io_ds which is a dm_deferred_set used to track time locality of non-discard IO. 5. The non-discard bio is finally locked in a dm_bio_prison_cell in process_bio. The race can result in deadlock, leaving the block layer hanging waiting for completion of a discard bio that never completes, e.g.: INFO: task ruby:15354 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. ruby D ffffffff8160f0e0 0 15354 15314 0x00000000 ffff8802fb08bc58 0000000000000082 ffff8802fb08bfd8 0000000000012900 ffff8802fb08a010 0000000000012900 0000000000012900 0000000000012900 ffff8802fb08bfd8 0000000000012900 ffff8803324b9480 ffff88032c6f14c0 Call Trace: [<ffffffff814e5a19>] schedule+0x29/0x70 [<ffffffff814e3d85>] schedule_timeout+0x195/0x220 [<ffffffffa06b9bc1>] ? _dm_request+0x111/0x160 [dm_mod] [<ffffffff814e589e>] wait_for_common+0x11e/0x190 [<ffffffff8107a170>] ? try_to_wake_up+0x2b0/0x2b0 [<ffffffff814e59ed>] wait_for_completion+0x1d/0x20 [<ffffffff81233289>] blkdev_issue_discard+0x219/0x260 [<ffffffff81233e79>] blkdev_ioctl+0x6e9/0x7b0 [<ffffffff8119a65c>] block_ioctl+0x3c/0x40 [<ffffffff8117539c>] do_vfs_ioctl+0x8c/0x340 [<ffffffff8119a547>] ? block_llseek+0x67/0xb0 [<ffffffff811756f1>] sys_ioctl+0xa1/0xb0 [<ffffffff810561f6>] ? sys_rt_sigprocmask+0x86/0xd0 [<ffffffff814ef099>] system_call_fastpath+0x16/0x1b The thinp-test-suite's test_discard_random_sectors reliably hits this deadlock on fast SSD storage. The fix for this race is that the all_io_entry for a bio must be incremented whilst the dm_bio_prison_cell is held for the bio's associated virtual and physical blocks. That cell locking wasn't occurring early enough in thin_bio_map. This patch fixes this. Care is taken to always call the new function inc_all_io_entry() with the relevant cells locked, but they are generally unlocked before calling issue() to try to avoid holding the cells locked across generic_submit_request. Also, now that thin_bio_map may lock bios in a cell, process_bio() is no longer the only thread that will do so. Because of this we must be sure to use cell_defer_except() to release all non-holder entries, that were added by the other thread, because they must be deferred. This patch depends on "dm thin: replace dm_cell_release_singleton with cell_defer_except". Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Cc: stable@vger.kernel.org
2012-12-22 00:23:31 +04:00
if (lookup_result.shared) {
process_shared_bio(tc, bio, block, &lookup_result);
cell_defer_no_holder(tc, cell); /* FIXME: pass this cell into process_shared? */
dm thin: fix race between simultaneous io and discards to same block There is a race when discard bios and non-discard bios are issued simultaneously to the same block. Discard support is expensive for all thin devices precisely because you have to be careful to quiesce the area you're discarding. DM thin must handle this conflicting IO pattern (simultaneous non-discard vs discard) even though a sane application shouldn't be issuing such IO. The race manifests as follows: 1. A non-discard bio is mapped in thin_bio_map. This doesn't lock out parallel activity to the same block. 2. A discard bio is issued to the same block as the non-discard bio. 3. The discard bio is locked in a dm_bio_prison_cell in process_discard to lock out parallel activity against the same block. 4. The non-discard bio's mapping continues and its all_io_entry is incremented so the bio is accounted for in the thin pool's all_io_ds which is a dm_deferred_set used to track time locality of non-discard IO. 5. The non-discard bio is finally locked in a dm_bio_prison_cell in process_bio. The race can result in deadlock, leaving the block layer hanging waiting for completion of a discard bio that never completes, e.g.: INFO: task ruby:15354 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. ruby D ffffffff8160f0e0 0 15354 15314 0x00000000 ffff8802fb08bc58 0000000000000082 ffff8802fb08bfd8 0000000000012900 ffff8802fb08a010 0000000000012900 0000000000012900 0000000000012900 ffff8802fb08bfd8 0000000000012900 ffff8803324b9480 ffff88032c6f14c0 Call Trace: [<ffffffff814e5a19>] schedule+0x29/0x70 [<ffffffff814e3d85>] schedule_timeout+0x195/0x220 [<ffffffffa06b9bc1>] ? _dm_request+0x111/0x160 [dm_mod] [<ffffffff814e589e>] wait_for_common+0x11e/0x190 [<ffffffff8107a170>] ? try_to_wake_up+0x2b0/0x2b0 [<ffffffff814e59ed>] wait_for_completion+0x1d/0x20 [<ffffffff81233289>] blkdev_issue_discard+0x219/0x260 [<ffffffff81233e79>] blkdev_ioctl+0x6e9/0x7b0 [<ffffffff8119a65c>] block_ioctl+0x3c/0x40 [<ffffffff8117539c>] do_vfs_ioctl+0x8c/0x340 [<ffffffff8119a547>] ? block_llseek+0x67/0xb0 [<ffffffff811756f1>] sys_ioctl+0xa1/0xb0 [<ffffffff810561f6>] ? sys_rt_sigprocmask+0x86/0xd0 [<ffffffff814ef099>] system_call_fastpath+0x16/0x1b The thinp-test-suite's test_discard_random_sectors reliably hits this deadlock on fast SSD storage. The fix for this race is that the all_io_entry for a bio must be incremented whilst the dm_bio_prison_cell is held for the bio's associated virtual and physical blocks. That cell locking wasn't occurring early enough in thin_bio_map. This patch fixes this. Care is taken to always call the new function inc_all_io_entry() with the relevant cells locked, but they are generally unlocked before calling issue() to try to avoid holding the cells locked across generic_submit_request. Also, now that thin_bio_map may lock bios in a cell, process_bio() is no longer the only thread that will do so. Because of this we must be sure to use cell_defer_except() to release all non-holder entries, that were added by the other thread, because they must be deferred. This patch depends on "dm thin: replace dm_cell_release_singleton with cell_defer_except". Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Cc: stable@vger.kernel.org
2012-12-22 00:23:31 +04:00
} else {
inc_all_io_entry(pool, bio);
cell_defer_no_holder(tc, cell);
dm thin: fix race between simultaneous io and discards to same block There is a race when discard bios and non-discard bios are issued simultaneously to the same block. Discard support is expensive for all thin devices precisely because you have to be careful to quiesce the area you're discarding. DM thin must handle this conflicting IO pattern (simultaneous non-discard vs discard) even though a sane application shouldn't be issuing such IO. The race manifests as follows: 1. A non-discard bio is mapped in thin_bio_map. This doesn't lock out parallel activity to the same block. 2. A discard bio is issued to the same block as the non-discard bio. 3. The discard bio is locked in a dm_bio_prison_cell in process_discard to lock out parallel activity against the same block. 4. The non-discard bio's mapping continues and its all_io_entry is incremented so the bio is accounted for in the thin pool's all_io_ds which is a dm_deferred_set used to track time locality of non-discard IO. 5. The non-discard bio is finally locked in a dm_bio_prison_cell in process_bio. The race can result in deadlock, leaving the block layer hanging waiting for completion of a discard bio that never completes, e.g.: INFO: task ruby:15354 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. ruby D ffffffff8160f0e0 0 15354 15314 0x00000000 ffff8802fb08bc58 0000000000000082 ffff8802fb08bfd8 0000000000012900 ffff8802fb08a010 0000000000012900 0000000000012900 0000000000012900 ffff8802fb08bfd8 0000000000012900 ffff8803324b9480 ffff88032c6f14c0 Call Trace: [<ffffffff814e5a19>] schedule+0x29/0x70 [<ffffffff814e3d85>] schedule_timeout+0x195/0x220 [<ffffffffa06b9bc1>] ? _dm_request+0x111/0x160 [dm_mod] [<ffffffff814e589e>] wait_for_common+0x11e/0x190 [<ffffffff8107a170>] ? try_to_wake_up+0x2b0/0x2b0 [<ffffffff814e59ed>] wait_for_completion+0x1d/0x20 [<ffffffff81233289>] blkdev_issue_discard+0x219/0x260 [<ffffffff81233e79>] blkdev_ioctl+0x6e9/0x7b0 [<ffffffff8119a65c>] block_ioctl+0x3c/0x40 [<ffffffff8117539c>] do_vfs_ioctl+0x8c/0x340 [<ffffffff8119a547>] ? block_llseek+0x67/0xb0 [<ffffffff811756f1>] sys_ioctl+0xa1/0xb0 [<ffffffff810561f6>] ? sys_rt_sigprocmask+0x86/0xd0 [<ffffffff814ef099>] system_call_fastpath+0x16/0x1b The thinp-test-suite's test_discard_random_sectors reliably hits this deadlock on fast SSD storage. The fix for this race is that the all_io_entry for a bio must be incremented whilst the dm_bio_prison_cell is held for the bio's associated virtual and physical blocks. That cell locking wasn't occurring early enough in thin_bio_map. This patch fixes this. Care is taken to always call the new function inc_all_io_entry() with the relevant cells locked, but they are generally unlocked before calling issue() to try to avoid holding the cells locked across generic_submit_request. Also, now that thin_bio_map may lock bios in a cell, process_bio() is no longer the only thread that will do so. Because of this we must be sure to use cell_defer_except() to release all non-holder entries, that were added by the other thread, because they must be deferred. This patch depends on "dm thin: replace dm_cell_release_singleton with cell_defer_except". Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Cc: stable@vger.kernel.org
2012-12-22 00:23:31 +04:00
remap_and_issue(tc, bio, lookup_result.block);
dm thin: fix race between simultaneous io and discards to same block There is a race when discard bios and non-discard bios are issued simultaneously to the same block. Discard support is expensive for all thin devices precisely because you have to be careful to quiesce the area you're discarding. DM thin must handle this conflicting IO pattern (simultaneous non-discard vs discard) even though a sane application shouldn't be issuing such IO. The race manifests as follows: 1. A non-discard bio is mapped in thin_bio_map. This doesn't lock out parallel activity to the same block. 2. A discard bio is issued to the same block as the non-discard bio. 3. The discard bio is locked in a dm_bio_prison_cell in process_discard to lock out parallel activity against the same block. 4. The non-discard bio's mapping continues and its all_io_entry is incremented so the bio is accounted for in the thin pool's all_io_ds which is a dm_deferred_set used to track time locality of non-discard IO. 5. The non-discard bio is finally locked in a dm_bio_prison_cell in process_bio. The race can result in deadlock, leaving the block layer hanging waiting for completion of a discard bio that never completes, e.g.: INFO: task ruby:15354 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. ruby D ffffffff8160f0e0 0 15354 15314 0x00000000 ffff8802fb08bc58 0000000000000082 ffff8802fb08bfd8 0000000000012900 ffff8802fb08a010 0000000000012900 0000000000012900 0000000000012900 ffff8802fb08bfd8 0000000000012900 ffff8803324b9480 ffff88032c6f14c0 Call Trace: [<ffffffff814e5a19>] schedule+0x29/0x70 [<ffffffff814e3d85>] schedule_timeout+0x195/0x220 [<ffffffffa06b9bc1>] ? _dm_request+0x111/0x160 [dm_mod] [<ffffffff814e589e>] wait_for_common+0x11e/0x190 [<ffffffff8107a170>] ? try_to_wake_up+0x2b0/0x2b0 [<ffffffff814e59ed>] wait_for_completion+0x1d/0x20 [<ffffffff81233289>] blkdev_issue_discard+0x219/0x260 [<ffffffff81233e79>] blkdev_ioctl+0x6e9/0x7b0 [<ffffffff8119a65c>] block_ioctl+0x3c/0x40 [<ffffffff8117539c>] do_vfs_ioctl+0x8c/0x340 [<ffffffff8119a547>] ? block_llseek+0x67/0xb0 [<ffffffff811756f1>] sys_ioctl+0xa1/0xb0 [<ffffffff810561f6>] ? sys_rt_sigprocmask+0x86/0xd0 [<ffffffff814ef099>] system_call_fastpath+0x16/0x1b The thinp-test-suite's test_discard_random_sectors reliably hits this deadlock on fast SSD storage. The fix for this race is that the all_io_entry for a bio must be incremented whilst the dm_bio_prison_cell is held for the bio's associated virtual and physical blocks. That cell locking wasn't occurring early enough in thin_bio_map. This patch fixes this. Care is taken to always call the new function inc_all_io_entry() with the relevant cells locked, but they are generally unlocked before calling issue() to try to avoid holding the cells locked across generic_submit_request. Also, now that thin_bio_map may lock bios in a cell, process_bio() is no longer the only thread that will do so. Because of this we must be sure to use cell_defer_except() to release all non-holder entries, that were added by the other thread, because they must be deferred. This patch depends on "dm thin: replace dm_cell_release_singleton with cell_defer_except". Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Cc: stable@vger.kernel.org
2012-12-22 00:23:31 +04:00
}
break;
case -ENODATA:
if (bio_data_dir(bio) == READ && tc->origin_dev) {
inc_all_io_entry(pool, bio);
cell_defer_no_holder(tc, cell);
dm thin: fix race between simultaneous io and discards to same block There is a race when discard bios and non-discard bios are issued simultaneously to the same block. Discard support is expensive for all thin devices precisely because you have to be careful to quiesce the area you're discarding. DM thin must handle this conflicting IO pattern (simultaneous non-discard vs discard) even though a sane application shouldn't be issuing such IO. The race manifests as follows: 1. A non-discard bio is mapped in thin_bio_map. This doesn't lock out parallel activity to the same block. 2. A discard bio is issued to the same block as the non-discard bio. 3. The discard bio is locked in a dm_bio_prison_cell in process_discard to lock out parallel activity against the same block. 4. The non-discard bio's mapping continues and its all_io_entry is incremented so the bio is accounted for in the thin pool's all_io_ds which is a dm_deferred_set used to track time locality of non-discard IO. 5. The non-discard bio is finally locked in a dm_bio_prison_cell in process_bio. The race can result in deadlock, leaving the block layer hanging waiting for completion of a discard bio that never completes, e.g.: INFO: task ruby:15354 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. ruby D ffffffff8160f0e0 0 15354 15314 0x00000000 ffff8802fb08bc58 0000000000000082 ffff8802fb08bfd8 0000000000012900 ffff8802fb08a010 0000000000012900 0000000000012900 0000000000012900 ffff8802fb08bfd8 0000000000012900 ffff8803324b9480 ffff88032c6f14c0 Call Trace: [<ffffffff814e5a19>] schedule+0x29/0x70 [<ffffffff814e3d85>] schedule_timeout+0x195/0x220 [<ffffffffa06b9bc1>] ? _dm_request+0x111/0x160 [dm_mod] [<ffffffff814e589e>] wait_for_common+0x11e/0x190 [<ffffffff8107a170>] ? try_to_wake_up+0x2b0/0x2b0 [<ffffffff814e59ed>] wait_for_completion+0x1d/0x20 [<ffffffff81233289>] blkdev_issue_discard+0x219/0x260 [<ffffffff81233e79>] blkdev_ioctl+0x6e9/0x7b0 [<ffffffff8119a65c>] block_ioctl+0x3c/0x40 [<ffffffff8117539c>] do_vfs_ioctl+0x8c/0x340 [<ffffffff8119a547>] ? block_llseek+0x67/0xb0 [<ffffffff811756f1>] sys_ioctl+0xa1/0xb0 [<ffffffff810561f6>] ? sys_rt_sigprocmask+0x86/0xd0 [<ffffffff814ef099>] system_call_fastpath+0x16/0x1b The thinp-test-suite's test_discard_random_sectors reliably hits this deadlock on fast SSD storage. The fix for this race is that the all_io_entry for a bio must be incremented whilst the dm_bio_prison_cell is held for the bio's associated virtual and physical blocks. That cell locking wasn't occurring early enough in thin_bio_map. This patch fixes this. Care is taken to always call the new function inc_all_io_entry() with the relevant cells locked, but they are generally unlocked before calling issue() to try to avoid holding the cells locked across generic_submit_request. Also, now that thin_bio_map may lock bios in a cell, process_bio() is no longer the only thread that will do so. Because of this we must be sure to use cell_defer_except() to release all non-holder entries, that were added by the other thread, because they must be deferred. This patch depends on "dm thin: replace dm_cell_release_singleton with cell_defer_except". Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Cc: stable@vger.kernel.org
2012-12-22 00:23:31 +04:00
remap_to_origin_and_issue(tc, bio);
} else
provision_block(tc, bio, block, cell);
break;
default:
DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
__func__, r);
cell_defer_no_holder(tc, cell);
bio_io_error(bio);
break;
}
}
static void process_bio_read_only(struct thin_c *tc, struct bio *bio)
{
int r;
int rw = bio_data_dir(bio);
dm_block_t block = get_bio_block(tc, bio);
struct dm_thin_lookup_result lookup_result;
r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
switch (r) {
case 0:
if (lookup_result.shared && (rw == WRITE) && bio->bi_size)
bio_io_error(bio);
dm thin: fix race between simultaneous io and discards to same block There is a race when discard bios and non-discard bios are issued simultaneously to the same block. Discard support is expensive for all thin devices precisely because you have to be careful to quiesce the area you're discarding. DM thin must handle this conflicting IO pattern (simultaneous non-discard vs discard) even though a sane application shouldn't be issuing such IO. The race manifests as follows: 1. A non-discard bio is mapped in thin_bio_map. This doesn't lock out parallel activity to the same block. 2. A discard bio is issued to the same block as the non-discard bio. 3. The discard bio is locked in a dm_bio_prison_cell in process_discard to lock out parallel activity against the same block. 4. The non-discard bio's mapping continues and its all_io_entry is incremented so the bio is accounted for in the thin pool's all_io_ds which is a dm_deferred_set used to track time locality of non-discard IO. 5. The non-discard bio is finally locked in a dm_bio_prison_cell in process_bio. The race can result in deadlock, leaving the block layer hanging waiting for completion of a discard bio that never completes, e.g.: INFO: task ruby:15354 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. ruby D ffffffff8160f0e0 0 15354 15314 0x00000000 ffff8802fb08bc58 0000000000000082 ffff8802fb08bfd8 0000000000012900 ffff8802fb08a010 0000000000012900 0000000000012900 0000000000012900 ffff8802fb08bfd8 0000000000012900 ffff8803324b9480 ffff88032c6f14c0 Call Trace: [<ffffffff814e5a19>] schedule+0x29/0x70 [<ffffffff814e3d85>] schedule_timeout+0x195/0x220 [<ffffffffa06b9bc1>] ? _dm_request+0x111/0x160 [dm_mod] [<ffffffff814e589e>] wait_for_common+0x11e/0x190 [<ffffffff8107a170>] ? try_to_wake_up+0x2b0/0x2b0 [<ffffffff814e59ed>] wait_for_completion+0x1d/0x20 [<ffffffff81233289>] blkdev_issue_discard+0x219/0x260 [<ffffffff81233e79>] blkdev_ioctl+0x6e9/0x7b0 [<ffffffff8119a65c>] block_ioctl+0x3c/0x40 [<ffffffff8117539c>] do_vfs_ioctl+0x8c/0x340 [<ffffffff8119a547>] ? block_llseek+0x67/0xb0 [<ffffffff811756f1>] sys_ioctl+0xa1/0xb0 [<ffffffff810561f6>] ? sys_rt_sigprocmask+0x86/0xd0 [<ffffffff814ef099>] system_call_fastpath+0x16/0x1b The thinp-test-suite's test_discard_random_sectors reliably hits this deadlock on fast SSD storage. The fix for this race is that the all_io_entry for a bio must be incremented whilst the dm_bio_prison_cell is held for the bio's associated virtual and physical blocks. That cell locking wasn't occurring early enough in thin_bio_map. This patch fixes this. Care is taken to always call the new function inc_all_io_entry() with the relevant cells locked, but they are generally unlocked before calling issue() to try to avoid holding the cells locked across generic_submit_request. Also, now that thin_bio_map may lock bios in a cell, process_bio() is no longer the only thread that will do so. Because of this we must be sure to use cell_defer_except() to release all non-holder entries, that were added by the other thread, because they must be deferred. This patch depends on "dm thin: replace dm_cell_release_singleton with cell_defer_except". Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Cc: stable@vger.kernel.org
2012-12-22 00:23:31 +04:00
else {
inc_all_io_entry(tc->pool, bio);
remap_and_issue(tc, bio, lookup_result.block);
dm thin: fix race between simultaneous io and discards to same block There is a race when discard bios and non-discard bios are issued simultaneously to the same block. Discard support is expensive for all thin devices precisely because you have to be careful to quiesce the area you're discarding. DM thin must handle this conflicting IO pattern (simultaneous non-discard vs discard) even though a sane application shouldn't be issuing such IO. The race manifests as follows: 1. A non-discard bio is mapped in thin_bio_map. This doesn't lock out parallel activity to the same block. 2. A discard bio is issued to the same block as the non-discard bio. 3. The discard bio is locked in a dm_bio_prison_cell in process_discard to lock out parallel activity against the same block. 4. The non-discard bio's mapping continues and its all_io_entry is incremented so the bio is accounted for in the thin pool's all_io_ds which is a dm_deferred_set used to track time locality of non-discard IO. 5. The non-discard bio is finally locked in a dm_bio_prison_cell in process_bio. The race can result in deadlock, leaving the block layer hanging waiting for completion of a discard bio that never completes, e.g.: INFO: task ruby:15354 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. ruby D ffffffff8160f0e0 0 15354 15314 0x00000000 ffff8802fb08bc58 0000000000000082 ffff8802fb08bfd8 0000000000012900 ffff8802fb08a010 0000000000012900 0000000000012900 0000000000012900 ffff8802fb08bfd8 0000000000012900 ffff8803324b9480 ffff88032c6f14c0 Call Trace: [<ffffffff814e5a19>] schedule+0x29/0x70 [<ffffffff814e3d85>] schedule_timeout+0x195/0x220 [<ffffffffa06b9bc1>] ? _dm_request+0x111/0x160 [dm_mod] [<ffffffff814e589e>] wait_for_common+0x11e/0x190 [<ffffffff8107a170>] ? try_to_wake_up+0x2b0/0x2b0 [<ffffffff814e59ed>] wait_for_completion+0x1d/0x20 [<ffffffff81233289>] blkdev_issue_discard+0x219/0x260 [<ffffffff81233e79>] blkdev_ioctl+0x6e9/0x7b0 [<ffffffff8119a65c>] block_ioctl+0x3c/0x40 [<ffffffff8117539c>] do_vfs_ioctl+0x8c/0x340 [<ffffffff8119a547>] ? block_llseek+0x67/0xb0 [<ffffffff811756f1>] sys_ioctl+0xa1/0xb0 [<ffffffff810561f6>] ? sys_rt_sigprocmask+0x86/0xd0 [<ffffffff814ef099>] system_call_fastpath+0x16/0x1b The thinp-test-suite's test_discard_random_sectors reliably hits this deadlock on fast SSD storage. The fix for this race is that the all_io_entry for a bio must be incremented whilst the dm_bio_prison_cell is held for the bio's associated virtual and physical blocks. That cell locking wasn't occurring early enough in thin_bio_map. This patch fixes this. Care is taken to always call the new function inc_all_io_entry() with the relevant cells locked, but they are generally unlocked before calling issue() to try to avoid holding the cells locked across generic_submit_request. Also, now that thin_bio_map may lock bios in a cell, process_bio() is no longer the only thread that will do so. Because of this we must be sure to use cell_defer_except() to release all non-holder entries, that were added by the other thread, because they must be deferred. This patch depends on "dm thin: replace dm_cell_release_singleton with cell_defer_except". Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Cc: stable@vger.kernel.org
2012-12-22 00:23:31 +04:00
}
break;
case -ENODATA:
if (rw != READ) {
bio_io_error(bio);
break;
}
if (tc->origin_dev) {
dm thin: fix race between simultaneous io and discards to same block There is a race when discard bios and non-discard bios are issued simultaneously to the same block. Discard support is expensive for all thin devices precisely because you have to be careful to quiesce the area you're discarding. DM thin must handle this conflicting IO pattern (simultaneous non-discard vs discard) even though a sane application shouldn't be issuing such IO. The race manifests as follows: 1. A non-discard bio is mapped in thin_bio_map. This doesn't lock out parallel activity to the same block. 2. A discard bio is issued to the same block as the non-discard bio. 3. The discard bio is locked in a dm_bio_prison_cell in process_discard to lock out parallel activity against the same block. 4. The non-discard bio's mapping continues and its all_io_entry is incremented so the bio is accounted for in the thin pool's all_io_ds which is a dm_deferred_set used to track time locality of non-discard IO. 5. The non-discard bio is finally locked in a dm_bio_prison_cell in process_bio. The race can result in deadlock, leaving the block layer hanging waiting for completion of a discard bio that never completes, e.g.: INFO: task ruby:15354 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. ruby D ffffffff8160f0e0 0 15354 15314 0x00000000 ffff8802fb08bc58 0000000000000082 ffff8802fb08bfd8 0000000000012900 ffff8802fb08a010 0000000000012900 0000000000012900 0000000000012900 ffff8802fb08bfd8 0000000000012900 ffff8803324b9480 ffff88032c6f14c0 Call Trace: [<ffffffff814e5a19>] schedule+0x29/0x70 [<ffffffff814e3d85>] schedule_timeout+0x195/0x220 [<ffffffffa06b9bc1>] ? _dm_request+0x111/0x160 [dm_mod] [<ffffffff814e589e>] wait_for_common+0x11e/0x190 [<ffffffff8107a170>] ? try_to_wake_up+0x2b0/0x2b0 [<ffffffff814e59ed>] wait_for_completion+0x1d/0x20 [<ffffffff81233289>] blkdev_issue_discard+0x219/0x260 [<ffffffff81233e79>] blkdev_ioctl+0x6e9/0x7b0 [<ffffffff8119a65c>] block_ioctl+0x3c/0x40 [<ffffffff8117539c>] do_vfs_ioctl+0x8c/0x340 [<ffffffff8119a547>] ? block_llseek+0x67/0xb0 [<ffffffff811756f1>] sys_ioctl+0xa1/0xb0 [<ffffffff810561f6>] ? sys_rt_sigprocmask+0x86/0xd0 [<ffffffff814ef099>] system_call_fastpath+0x16/0x1b The thinp-test-suite's test_discard_random_sectors reliably hits this deadlock on fast SSD storage. The fix for this race is that the all_io_entry for a bio must be incremented whilst the dm_bio_prison_cell is held for the bio's associated virtual and physical blocks. That cell locking wasn't occurring early enough in thin_bio_map. This patch fixes this. Care is taken to always call the new function inc_all_io_entry() with the relevant cells locked, but they are generally unlocked before calling issue() to try to avoid holding the cells locked across generic_submit_request. Also, now that thin_bio_map may lock bios in a cell, process_bio() is no longer the only thread that will do so. Because of this we must be sure to use cell_defer_except() to release all non-holder entries, that were added by the other thread, because they must be deferred. This patch depends on "dm thin: replace dm_cell_release_singleton with cell_defer_except". Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Cc: stable@vger.kernel.org
2012-12-22 00:23:31 +04:00
inc_all_io_entry(tc->pool, bio);
remap_to_origin_and_issue(tc, bio);
break;
}
zero_fill_bio(bio);
bio_endio(bio, 0);
break;
default:
DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
__func__, r);
bio_io_error(bio);
break;
}
}
static void process_bio_fail(struct thin_c *tc, struct bio *bio)
{
bio_io_error(bio);
}
/*
* FIXME: should we also commit due to size of transaction, measured in
* metadata blocks?
*/
static int need_commit_due_to_time(struct pool *pool)
{
return jiffies < pool->last_commit_jiffies ||
jiffies > pool->last_commit_jiffies + COMMIT_PERIOD;
}
static void process_deferred_bios(struct pool *pool)
{
unsigned long flags;
struct bio *bio;
struct bio_list bios;
bio_list_init(&bios);
spin_lock_irqsave(&pool->lock, flags);
bio_list_merge(&bios, &pool->deferred_bios);
bio_list_init(&pool->deferred_bios);
spin_unlock_irqrestore(&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));
struct thin_c *tc = h->tc;
/*
* If we've got no free new_mapping structs, and processing
* this bio might require one, we pause until there are some
* prepared mappings to process.
*/
if (ensure_next_mapping(pool)) {
spin_lock_irqsave(&pool->lock, flags);
bio_list_merge(&pool->deferred_bios, &bios);
spin_unlock_irqrestore(&pool->lock, flags);
break;
}
if (bio->bi_rw & REQ_DISCARD)
pool->process_discard(tc, bio);
else
pool->process_bio(tc, bio);
}
/*
* If there are any deferred flush bios, we must commit
* the metadata before issuing them.
*/
bio_list_init(&bios);
spin_lock_irqsave(&pool->lock, flags);
bio_list_merge(&bios, &pool->deferred_flush_bios);
bio_list_init(&pool->deferred_flush_bios);
spin_unlock_irqrestore(&pool->lock, flags);
if (bio_list_empty(&bios) && !need_commit_due_to_time(pool))
return;
if (commit_or_fallback(pool)) {
while ((bio = bio_list_pop(&bios)))
bio_io_error(bio);
return;
}
pool->last_commit_jiffies = jiffies;
while ((bio = bio_list_pop(&bios)))
generic_make_request(bio);
}
static void do_worker(struct work_struct *ws)
{
struct pool *pool = container_of(ws, struct pool, worker);
process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping);
process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard);
process_deferred_bios(pool);
}
/*
* We want to commit periodically so that not too much
* unwritten data builds up.
*/
static void do_waker(struct work_struct *ws)
{
struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker);
wake_worker(pool);
queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD);
}
/*----------------------------------------------------------------*/
static enum pool_mode get_pool_mode(struct pool *pool)
{
return pool->pf.mode;
}
static void set_pool_mode(struct pool *pool, enum pool_mode mode)
{
int r;
pool->pf.mode = mode;
switch (mode) {
case PM_FAIL:
DMERR("switching pool to failure mode");
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;
break;
case PM_READ_ONLY:
DMERR("switching pool to read-only mode");
r = dm_pool_abort_metadata(pool->pmd);
if (r) {
DMERR("aborting transaction failed");
set_pool_mode(pool, PM_FAIL);
} else {
dm_pool_metadata_read_only(pool->pmd);
pool->process_bio = process_bio_read_only;
pool->process_discard = process_discard;
pool->process_prepared_mapping = process_prepared_mapping_fail;
pool->process_prepared_discard = process_prepared_discard_passdown;
}
break;
case PM_WRITE:
pool->process_bio = process_bio;
pool->process_discard = process_discard;
pool->process_prepared_mapping = process_prepared_mapping;
pool->process_prepared_discard = process_prepared_discard;
break;
}
}
/*----------------------------------------------------------------*/
/*
* Mapping functions.
*/
/*
* Called only while mapping a thin bio to hand it over to the workqueue.
*/
static void thin_defer_bio(struct thin_c *tc, struct bio *bio)
{
unsigned long flags;
struct pool *pool = tc->pool;
spin_lock_irqsave(&pool->lock, flags);
bio_list_add(&pool->deferred_bios, bio);
spin_unlock_irqrestore(&pool->lock, flags);
wake_worker(pool);
}
static void thin_hook_bio(struct thin_c *tc, struct bio *bio)
{
struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
h->tc = tc;
h->shared_read_entry = NULL;
dm thin: fix race between simultaneous io and discards to same block There is a race when discard bios and non-discard bios are issued simultaneously to the same block. Discard support is expensive for all thin devices precisely because you have to be careful to quiesce the area you're discarding. DM thin must handle this conflicting IO pattern (simultaneous non-discard vs discard) even though a sane application shouldn't be issuing such IO. The race manifests as follows: 1. A non-discard bio is mapped in thin_bio_map. This doesn't lock out parallel activity to the same block. 2. A discard bio is issued to the same block as the non-discard bio. 3. The discard bio is locked in a dm_bio_prison_cell in process_discard to lock out parallel activity against the same block. 4. The non-discard bio's mapping continues and its all_io_entry is incremented so the bio is accounted for in the thin pool's all_io_ds which is a dm_deferred_set used to track time locality of non-discard IO. 5. The non-discard bio is finally locked in a dm_bio_prison_cell in process_bio. The race can result in deadlock, leaving the block layer hanging waiting for completion of a discard bio that never completes, e.g.: INFO: task ruby:15354 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. ruby D ffffffff8160f0e0 0 15354 15314 0x00000000 ffff8802fb08bc58 0000000000000082 ffff8802fb08bfd8 0000000000012900 ffff8802fb08a010 0000000000012900 0000000000012900 0000000000012900 ffff8802fb08bfd8 0000000000012900 ffff8803324b9480 ffff88032c6f14c0 Call Trace: [<ffffffff814e5a19>] schedule+0x29/0x70 [<ffffffff814e3d85>] schedule_timeout+0x195/0x220 [<ffffffffa06b9bc1>] ? _dm_request+0x111/0x160 [dm_mod] [<ffffffff814e589e>] wait_for_common+0x11e/0x190 [<ffffffff8107a170>] ? try_to_wake_up+0x2b0/0x2b0 [<ffffffff814e59ed>] wait_for_completion+0x1d/0x20 [<ffffffff81233289>] blkdev_issue_discard+0x219/0x260 [<ffffffff81233e79>] blkdev_ioctl+0x6e9/0x7b0 [<ffffffff8119a65c>] block_ioctl+0x3c/0x40 [<ffffffff8117539c>] do_vfs_ioctl+0x8c/0x340 [<ffffffff8119a547>] ? block_llseek+0x67/0xb0 [<ffffffff811756f1>] sys_ioctl+0xa1/0xb0 [<ffffffff810561f6>] ? sys_rt_sigprocmask+0x86/0xd0 [<ffffffff814ef099>] system_call_fastpath+0x16/0x1b The thinp-test-suite's test_discard_random_sectors reliably hits this deadlock on fast SSD storage. The fix for this race is that the all_io_entry for a bio must be incremented whilst the dm_bio_prison_cell is held for the bio's associated virtual and physical blocks. That cell locking wasn't occurring early enough in thin_bio_map. This patch fixes this. Care is taken to always call the new function inc_all_io_entry() with the relevant cells locked, but they are generally unlocked before calling issue() to try to avoid holding the cells locked across generic_submit_request. Also, now that thin_bio_map may lock bios in a cell, process_bio() is no longer the only thread that will do so. Because of this we must be sure to use cell_defer_except() to release all non-holder entries, that were added by the other thread, because they must be deferred. This patch depends on "dm thin: replace dm_cell_release_singleton with cell_defer_except". Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Cc: stable@vger.kernel.org
2012-12-22 00:23:31 +04:00
h->all_io_entry = NULL;
h->overwrite_mapping = NULL;
}
/*
* Non-blocking function called from the thin target's map function.
*/
static int thin_bio_map(struct dm_target *ti, struct bio *bio)
{
int r;
struct thin_c *tc = ti->private;
dm_block_t block = get_bio_block(tc, bio);
struct dm_thin_device *td = tc->td;
struct dm_thin_lookup_result result;
struct dm_bio_prison_cell cell1, cell2;
struct dm_bio_prison_cell *cell_result;
dm thin: fix race between simultaneous io and discards to same block There is a race when discard bios and non-discard bios are issued simultaneously to the same block. Discard support is expensive for all thin devices precisely because you have to be careful to quiesce the area you're discarding. DM thin must handle this conflicting IO pattern (simultaneous non-discard vs discard) even though a sane application shouldn't be issuing such IO. The race manifests as follows: 1. A non-discard bio is mapped in thin_bio_map. This doesn't lock out parallel activity to the same block. 2. A discard bio is issued to the same block as the non-discard bio. 3. The discard bio is locked in a dm_bio_prison_cell in process_discard to lock out parallel activity against the same block. 4. The non-discard bio's mapping continues and its all_io_entry is incremented so the bio is accounted for in the thin pool's all_io_ds which is a dm_deferred_set used to track time locality of non-discard IO. 5. The non-discard bio is finally locked in a dm_bio_prison_cell in process_bio. The race can result in deadlock, leaving the block layer hanging waiting for completion of a discard bio that never completes, e.g.: INFO: task ruby:15354 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. ruby D ffffffff8160f0e0 0 15354 15314 0x00000000 ffff8802fb08bc58 0000000000000082 ffff8802fb08bfd8 0000000000012900 ffff8802fb08a010 0000000000012900 0000000000012900 0000000000012900 ffff8802fb08bfd8 0000000000012900 ffff8803324b9480 ffff88032c6f14c0 Call Trace: [<ffffffff814e5a19>] schedule+0x29/0x70 [<ffffffff814e3d85>] schedule_timeout+0x195/0x220 [<ffffffffa06b9bc1>] ? _dm_request+0x111/0x160 [dm_mod] [<ffffffff814e589e>] wait_for_common+0x11e/0x190 [<ffffffff8107a170>] ? try_to_wake_up+0x2b0/0x2b0 [<ffffffff814e59ed>] wait_for_completion+0x1d/0x20 [<ffffffff81233289>] blkdev_issue_discard+0x219/0x260 [<ffffffff81233e79>] blkdev_ioctl+0x6e9/0x7b0 [<ffffffff8119a65c>] block_ioctl+0x3c/0x40 [<ffffffff8117539c>] do_vfs_ioctl+0x8c/0x340 [<ffffffff8119a547>] ? block_llseek+0x67/0xb0 [<ffffffff811756f1>] sys_ioctl+0xa1/0xb0 [<ffffffff810561f6>] ? sys_rt_sigprocmask+0x86/0xd0 [<ffffffff814ef099>] system_call_fastpath+0x16/0x1b The thinp-test-suite's test_discard_random_sectors reliably hits this deadlock on fast SSD storage. The fix for this race is that the all_io_entry for a bio must be incremented whilst the dm_bio_prison_cell is held for the bio's associated virtual and physical blocks. That cell locking wasn't occurring early enough in thin_bio_map. This patch fixes this. Care is taken to always call the new function inc_all_io_entry() with the relevant cells locked, but they are generally unlocked before calling issue() to try to avoid holding the cells locked across generic_submit_request. Also, now that thin_bio_map may lock bios in a cell, process_bio() is no longer the only thread that will do so. Because of this we must be sure to use cell_defer_except() to release all non-holder entries, that were added by the other thread, because they must be deferred. This patch depends on "dm thin: replace dm_cell_release_singleton with cell_defer_except". Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Cc: stable@vger.kernel.org
2012-12-22 00:23:31 +04:00
struct dm_cell_key key;
thin_hook_bio(tc, bio);
if (get_pool_mode(tc->pool) == PM_FAIL) {
bio_io_error(bio);
return DM_MAPIO_SUBMITTED;
}
if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) {
thin_defer_bio(tc, bio);
return DM_MAPIO_SUBMITTED;
}
r = dm_thin_find_block(td, block, 0, &result);
/*
* Note that we defer readahead too.
*/
switch (r) {
case 0:
if (unlikely(result.shared)) {
/*
* We have a race condition here between the
* result.shared value returned by the lookup and
* snapshot creation, which may cause new
* sharing.
*
* To avoid this always quiesce the origin before
* taking the snap. You want to do this anyway to
* ensure a consistent application view
* (i.e. lockfs).
*
* More distant ancestors are irrelevant. The
* shared flag will be set in their case.
*/
thin_defer_bio(tc, bio);
dm thin: fix race between simultaneous io and discards to same block There is a race when discard bios and non-discard bios are issued simultaneously to the same block. Discard support is expensive for all thin devices precisely because you have to be careful to quiesce the area you're discarding. DM thin must handle this conflicting IO pattern (simultaneous non-discard vs discard) even though a sane application shouldn't be issuing such IO. The race manifests as follows: 1. A non-discard bio is mapped in thin_bio_map. This doesn't lock out parallel activity to the same block. 2. A discard bio is issued to the same block as the non-discard bio. 3. The discard bio is locked in a dm_bio_prison_cell in process_discard to lock out parallel activity against the same block. 4. The non-discard bio's mapping continues and its all_io_entry is incremented so the bio is accounted for in the thin pool's all_io_ds which is a dm_deferred_set used to track time locality of non-discard IO. 5. The non-discard bio is finally locked in a dm_bio_prison_cell in process_bio. The race can result in deadlock, leaving the block layer hanging waiting for completion of a discard bio that never completes, e.g.: INFO: task ruby:15354 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. ruby D ffffffff8160f0e0 0 15354 15314 0x00000000 ffff8802fb08bc58 0000000000000082 ffff8802fb08bfd8 0000000000012900 ffff8802fb08a010 0000000000012900 0000000000012900 0000000000012900 ffff8802fb08bfd8 0000000000012900 ffff8803324b9480 ffff88032c6f14c0 Call Trace: [<ffffffff814e5a19>] schedule+0x29/0x70 [<ffffffff814e3d85>] schedule_timeout+0x195/0x220 [<ffffffffa06b9bc1>] ? _dm_request+0x111/0x160 [dm_mod] [<ffffffff814e589e>] wait_for_common+0x11e/0x190 [<ffffffff8107a170>] ? try_to_wake_up+0x2b0/0x2b0 [<ffffffff814e59ed>] wait_for_completion+0x1d/0x20 [<ffffffff81233289>] blkdev_issue_discard+0x219/0x260 [<ffffffff81233e79>] blkdev_ioctl+0x6e9/0x7b0 [<ffffffff8119a65c>] block_ioctl+0x3c/0x40 [<ffffffff8117539c>] do_vfs_ioctl+0x8c/0x340 [<ffffffff8119a547>] ? block_llseek+0x67/0xb0 [<ffffffff811756f1>] sys_ioctl+0xa1/0xb0 [<ffffffff810561f6>] ? sys_rt_sigprocmask+0x86/0xd0 [<ffffffff814ef099>] system_call_fastpath+0x16/0x1b The thinp-test-suite's test_discard_random_sectors reliably hits this deadlock on fast SSD storage. The fix for this race is that the all_io_entry for a bio must be incremented whilst the dm_bio_prison_cell is held for the bio's associated virtual and physical blocks. That cell locking wasn't occurring early enough in thin_bio_map. This patch fixes this. Care is taken to always call the new function inc_all_io_entry() with the relevant cells locked, but they are generally unlocked before calling issue() to try to avoid holding the cells locked across generic_submit_request. Also, now that thin_bio_map may lock bios in a cell, process_bio() is no longer the only thread that will do so. Because of this we must be sure to use cell_defer_except() to release all non-holder entries, that were added by the other thread, because they must be deferred. This patch depends on "dm thin: replace dm_cell_release_singleton with cell_defer_except". Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Cc: stable@vger.kernel.org
2012-12-22 00:23:31 +04:00
return DM_MAPIO_SUBMITTED;
}
dm thin: fix race between simultaneous io and discards to same block There is a race when discard bios and non-discard bios are issued simultaneously to the same block. Discard support is expensive for all thin devices precisely because you have to be careful to quiesce the area you're discarding. DM thin must handle this conflicting IO pattern (simultaneous non-discard vs discard) even though a sane application shouldn't be issuing such IO. The race manifests as follows: 1. A non-discard bio is mapped in thin_bio_map. This doesn't lock out parallel activity to the same block. 2. A discard bio is issued to the same block as the non-discard bio. 3. The discard bio is locked in a dm_bio_prison_cell in process_discard to lock out parallel activity against the same block. 4. The non-discard bio's mapping continues and its all_io_entry is incremented so the bio is accounted for in the thin pool's all_io_ds which is a dm_deferred_set used to track time locality of non-discard IO. 5. The non-discard bio is finally locked in a dm_bio_prison_cell in process_bio. The race can result in deadlock, leaving the block layer hanging waiting for completion of a discard bio that never completes, e.g.: INFO: task ruby:15354 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. ruby D ffffffff8160f0e0 0 15354 15314 0x00000000 ffff8802fb08bc58 0000000000000082 ffff8802fb08bfd8 0000000000012900 ffff8802fb08a010 0000000000012900 0000000000012900 0000000000012900 ffff8802fb08bfd8 0000000000012900 ffff8803324b9480 ffff88032c6f14c0 Call Trace: [<ffffffff814e5a19>] schedule+0x29/0x70 [<ffffffff814e3d85>] schedule_timeout+0x195/0x220 [<ffffffffa06b9bc1>] ? _dm_request+0x111/0x160 [dm_mod] [<ffffffff814e589e>] wait_for_common+0x11e/0x190 [<ffffffff8107a170>] ? try_to_wake_up+0x2b0/0x2b0 [<ffffffff814e59ed>] wait_for_completion+0x1d/0x20 [<ffffffff81233289>] blkdev_issue_discard+0x219/0x260 [<ffffffff81233e79>] blkdev_ioctl+0x6e9/0x7b0 [<ffffffff8119a65c>] block_ioctl+0x3c/0x40 [<ffffffff8117539c>] do_vfs_ioctl+0x8c/0x340 [<ffffffff8119a547>] ? block_llseek+0x67/0xb0 [<ffffffff811756f1>] sys_ioctl+0xa1/0xb0 [<ffffffff810561f6>] ? sys_rt_sigprocmask+0x86/0xd0 [<ffffffff814ef099>] system_call_fastpath+0x16/0x1b The thinp-test-suite's test_discard_random_sectors reliably hits this deadlock on fast SSD storage. The fix for this race is that the all_io_entry for a bio must be incremented whilst the dm_bio_prison_cell is held for the bio's associated virtual and physical blocks. That cell locking wasn't occurring early enough in thin_bio_map. This patch fixes this. Care is taken to always call the new function inc_all_io_entry() with the relevant cells locked, but they are generally unlocked before calling issue() to try to avoid holding the cells locked across generic_submit_request. Also, now that thin_bio_map may lock bios in a cell, process_bio() is no longer the only thread that will do so. Because of this we must be sure to use cell_defer_except() to release all non-holder entries, that were added by the other thread, because they must be deferred. This patch depends on "dm thin: replace dm_cell_release_singleton with cell_defer_except". Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Cc: stable@vger.kernel.org
2012-12-22 00:23:31 +04:00
build_virtual_key(tc->td, block, &key);
if (dm_bio_detain(tc->pool->prison, &key, bio, &cell1, &cell_result))
dm thin: fix race between simultaneous io and discards to same block There is a race when discard bios and non-discard bios are issued simultaneously to the same block. Discard support is expensive for all thin devices precisely because you have to be careful to quiesce the area you're discarding. DM thin must handle this conflicting IO pattern (simultaneous non-discard vs discard) even though a sane application shouldn't be issuing such IO. The race manifests as follows: 1. A non-discard bio is mapped in thin_bio_map. This doesn't lock out parallel activity to the same block. 2. A discard bio is issued to the same block as the non-discard bio. 3. The discard bio is locked in a dm_bio_prison_cell in process_discard to lock out parallel activity against the same block. 4. The non-discard bio's mapping continues and its all_io_entry is incremented so the bio is accounted for in the thin pool's all_io_ds which is a dm_deferred_set used to track time locality of non-discard IO. 5. The non-discard bio is finally locked in a dm_bio_prison_cell in process_bio. The race can result in deadlock, leaving the block layer hanging waiting for completion of a discard bio that never completes, e.g.: INFO: task ruby:15354 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. ruby D ffffffff8160f0e0 0 15354 15314 0x00000000 ffff8802fb08bc58 0000000000000082 ffff8802fb08bfd8 0000000000012900 ffff8802fb08a010 0000000000012900 0000000000012900 0000000000012900 ffff8802fb08bfd8 0000000000012900 ffff8803324b9480 ffff88032c6f14c0 Call Trace: [<ffffffff814e5a19>] schedule+0x29/0x70 [<ffffffff814e3d85>] schedule_timeout+0x195/0x220 [<ffffffffa06b9bc1>] ? _dm_request+0x111/0x160 [dm_mod] [<ffffffff814e589e>] wait_for_common+0x11e/0x190 [<ffffffff8107a170>] ? try_to_wake_up+0x2b0/0x2b0 [<ffffffff814e59ed>] wait_for_completion+0x1d/0x20 [<ffffffff81233289>] blkdev_issue_discard+0x219/0x260 [<ffffffff81233e79>] blkdev_ioctl+0x6e9/0x7b0 [<ffffffff8119a65c>] block_ioctl+0x3c/0x40 [<ffffffff8117539c>] do_vfs_ioctl+0x8c/0x340 [<ffffffff8119a547>] ? block_llseek+0x67/0xb0 [<ffffffff811756f1>] sys_ioctl+0xa1/0xb0 [<ffffffff810561f6>] ? sys_rt_sigprocmask+0x86/0xd0 [<ffffffff814ef099>] system_call_fastpath+0x16/0x1b The thinp-test-suite's test_discard_random_sectors reliably hits this deadlock on fast SSD storage. The fix for this race is that the all_io_entry for a bio must be incremented whilst the dm_bio_prison_cell is held for the bio's associated virtual and physical blocks. That cell locking wasn't occurring early enough in thin_bio_map. This patch fixes this. Care is taken to always call the new function inc_all_io_entry() with the relevant cells locked, but they are generally unlocked before calling issue() to try to avoid holding the cells locked across generic_submit_request. Also, now that thin_bio_map may lock bios in a cell, process_bio() is no longer the only thread that will do so. Because of this we must be sure to use cell_defer_except() to release all non-holder entries, that were added by the other thread, because they must be deferred. This patch depends on "dm thin: replace dm_cell_release_singleton with cell_defer_except". Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Cc: stable@vger.kernel.org
2012-12-22 00:23:31 +04:00
return DM_MAPIO_SUBMITTED;
build_data_key(tc->td, result.block, &key);
if (dm_bio_detain(tc->pool->prison, &key, bio, &cell2, &cell_result)) {
cell_defer_no_holder_no_free(tc, &cell1);
dm thin: fix race between simultaneous io and discards to same block There is a race when discard bios and non-discard bios are issued simultaneously to the same block. Discard support is expensive for all thin devices precisely because you have to be careful to quiesce the area you're discarding. DM thin must handle this conflicting IO pattern (simultaneous non-discard vs discard) even though a sane application shouldn't be issuing such IO. The race manifests as follows: 1. A non-discard bio is mapped in thin_bio_map. This doesn't lock out parallel activity to the same block. 2. A discard bio is issued to the same block as the non-discard bio. 3. The discard bio is locked in a dm_bio_prison_cell in process_discard to lock out parallel activity against the same block. 4. The non-discard bio's mapping continues and its all_io_entry is incremented so the bio is accounted for in the thin pool's all_io_ds which is a dm_deferred_set used to track time locality of non-discard IO. 5. The non-discard bio is finally locked in a dm_bio_prison_cell in process_bio. The race can result in deadlock, leaving the block layer hanging waiting for completion of a discard bio that never completes, e.g.: INFO: task ruby:15354 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. ruby D ffffffff8160f0e0 0 15354 15314 0x00000000 ffff8802fb08bc58 0000000000000082 ffff8802fb08bfd8 0000000000012900 ffff8802fb08a010 0000000000012900 0000000000012900 0000000000012900 ffff8802fb08bfd8 0000000000012900 ffff8803324b9480 ffff88032c6f14c0 Call Trace: [<ffffffff814e5a19>] schedule+0x29/0x70 [<ffffffff814e3d85>] schedule_timeout+0x195/0x220 [<ffffffffa06b9bc1>] ? _dm_request+0x111/0x160 [dm_mod] [<ffffffff814e589e>] wait_for_common+0x11e/0x190 [<ffffffff8107a170>] ? try_to_wake_up+0x2b0/0x2b0 [<ffffffff814e59ed>] wait_for_completion+0x1d/0x20 [<ffffffff81233289>] blkdev_issue_discard+0x219/0x260 [<ffffffff81233e79>] blkdev_ioctl+0x6e9/0x7b0 [<ffffffff8119a65c>] block_ioctl+0x3c/0x40 [<ffffffff8117539c>] do_vfs_ioctl+0x8c/0x340 [<ffffffff8119a547>] ? block_llseek+0x67/0xb0 [<ffffffff811756f1>] sys_ioctl+0xa1/0xb0 [<ffffffff810561f6>] ? sys_rt_sigprocmask+0x86/0xd0 [<ffffffff814ef099>] system_call_fastpath+0x16/0x1b The thinp-test-suite's test_discard_random_sectors reliably hits this deadlock on fast SSD storage. The fix for this race is that the all_io_entry for a bio must be incremented whilst the dm_bio_prison_cell is held for the bio's associated virtual and physical blocks. That cell locking wasn't occurring early enough in thin_bio_map. This patch fixes this. Care is taken to always call the new function inc_all_io_entry() with the relevant cells locked, but they are generally unlocked before calling issue() to try to avoid holding the cells locked across generic_submit_request. Also, now that thin_bio_map may lock bios in a cell, process_bio() is no longer the only thread that will do so. Because of this we must be sure to use cell_defer_except() to release all non-holder entries, that were added by the other thread, because they must be deferred. This patch depends on "dm thin: replace dm_cell_release_singleton with cell_defer_except". Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Cc: stable@vger.kernel.org
2012-12-22 00:23:31 +04:00
return DM_MAPIO_SUBMITTED;
}
inc_all_io_entry(tc->pool, bio);
cell_defer_no_holder_no_free(tc, &cell2);
cell_defer_no_holder_no_free(tc, &cell1);
dm thin: fix race between simultaneous io and discards to same block There is a race when discard bios and non-discard bios are issued simultaneously to the same block. Discard support is expensive for all thin devices precisely because you have to be careful to quiesce the area you're discarding. DM thin must handle this conflicting IO pattern (simultaneous non-discard vs discard) even though a sane application shouldn't be issuing such IO. The race manifests as follows: 1. A non-discard bio is mapped in thin_bio_map. This doesn't lock out parallel activity to the same block. 2. A discard bio is issued to the same block as the non-discard bio. 3. The discard bio is locked in a dm_bio_prison_cell in process_discard to lock out parallel activity against the same block. 4. The non-discard bio's mapping continues and its all_io_entry is incremented so the bio is accounted for in the thin pool's all_io_ds which is a dm_deferred_set used to track time locality of non-discard IO. 5. The non-discard bio is finally locked in a dm_bio_prison_cell in process_bio. The race can result in deadlock, leaving the block layer hanging waiting for completion of a discard bio that never completes, e.g.: INFO: task ruby:15354 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. ruby D ffffffff8160f0e0 0 15354 15314 0x00000000 ffff8802fb08bc58 0000000000000082 ffff8802fb08bfd8 0000000000012900 ffff8802fb08a010 0000000000012900 0000000000012900 0000000000012900 ffff8802fb08bfd8 0000000000012900 ffff8803324b9480 ffff88032c6f14c0 Call Trace: [<ffffffff814e5a19>] schedule+0x29/0x70 [<ffffffff814e3d85>] schedule_timeout+0x195/0x220 [<ffffffffa06b9bc1>] ? _dm_request+0x111/0x160 [dm_mod] [<ffffffff814e589e>] wait_for_common+0x11e/0x190 [<ffffffff8107a170>] ? try_to_wake_up+0x2b0/0x2b0 [<ffffffff814e59ed>] wait_for_completion+0x1d/0x20 [<ffffffff81233289>] blkdev_issue_discard+0x219/0x260 [<ffffffff81233e79>] blkdev_ioctl+0x6e9/0x7b0 [<ffffffff8119a65c>] block_ioctl+0x3c/0x40 [<ffffffff8117539c>] do_vfs_ioctl+0x8c/0x340 [<ffffffff8119a547>] ? block_llseek+0x67/0xb0 [<ffffffff811756f1>] sys_ioctl+0xa1/0xb0 [<ffffffff810561f6>] ? sys_rt_sigprocmask+0x86/0xd0 [<ffffffff814ef099>] system_call_fastpath+0x16/0x1b The thinp-test-suite's test_discard_random_sectors reliably hits this deadlock on fast SSD storage. The fix for this race is that the all_io_entry for a bio must be incremented whilst the dm_bio_prison_cell is held for the bio's associated virtual and physical blocks. That cell locking wasn't occurring early enough in thin_bio_map. This patch fixes this. Care is taken to always call the new function inc_all_io_entry() with the relevant cells locked, but they are generally unlocked before calling issue() to try to avoid holding the cells locked across generic_submit_request. Also, now that thin_bio_map may lock bios in a cell, process_bio() is no longer the only thread that will do so. Because of this we must be sure to use cell_defer_except() to release all non-holder entries, that were added by the other thread, because they must be deferred. This patch depends on "dm thin: replace dm_cell_release_singleton with cell_defer_except". Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Cc: stable@vger.kernel.org
2012-12-22 00:23:31 +04:00
remap(tc, bio, result.block);
return DM_MAPIO_REMAPPED;
case -ENODATA:
if (get_pool_mode(tc->pool) == PM_READ_ONLY) {
/*
* This block isn't provisioned, and we have no way
* of doing so. Just error it.
*/
bio_io_error(bio);
return DM_MAPIO_SUBMITTED;
}
/* fall through */
case -EWOULDBLOCK:
/*
* In future, the failed dm_thin_find_block above could
* provide the hint to load the metadata into cache.
*/
thin_defer_bio(tc, bio);
return DM_MAPIO_SUBMITTED;
default:
/*
* Must always call bio_io_error on failure.
* dm_thin_find_block can fail with -EINVAL if the
* pool is switched to fail-io mode.
*/
bio_io_error(bio);
return DM_MAPIO_SUBMITTED;
}
}
static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
{
int r;
unsigned long flags;
struct pool_c *pt = container_of(cb, struct pool_c, callbacks);
spin_lock_irqsave(&pt->pool->lock, flags);
r = !bio_list_empty(&pt->pool->retry_on_resume_list);
spin_unlock_irqrestore(&pt->pool->lock, flags);
if (!r) {
struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
r = bdi_congested(&q->backing_dev_info, bdi_bits);
}
return r;
}
static void __requeue_bios(struct pool *pool)
{
bio_list_merge(&pool->deferred_bios, &pool->retry_on_resume_list);
bio_list_init(&pool->retry_on_resume_list);
}
/*----------------------------------------------------------------
* Binding of control targets to a pool object
*--------------------------------------------------------------*/
static bool data_dev_supports_discard(struct pool_c *pt)
{
struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
return q && blk_queue_discard(q);
}
static bool is_factor(sector_t block_size, uint32_t n)
{
return !sector_div(block_size, n);
}
/*
* If discard_passdown was enabled verify that the data device
* supports discards. Disable discard_passdown if not.
*/
static void disable_passdown_if_not_supported(struct pool_c *pt)
{
struct pool *pool = pt->pool;
struct block_device *data_bdev = pt->data_dev->bdev;
struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits;
sector_t block_size = pool->sectors_per_block << SECTOR_SHIFT;
const char *reason = NULL;
char buf[BDEVNAME_SIZE];
if (!pt->adjusted_pf.discard_passdown)
return;
if (!data_dev_supports_discard(pt))
reason = "discard unsupported";
else if (data_limits->max_discard_sectors < pool->sectors_per_block)
reason = "max discard sectors smaller than a block";
else if (data_limits->discard_granularity > block_size)
reason = "discard granularity larger than a block";
else if (!is_factor(block_size, data_limits->discard_granularity))
reason = "discard granularity not a factor of block size";
if (reason) {
DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason);
pt->adjusted_pf.discard_passdown = false;
}
}
static int bind_control_target(struct pool *pool, struct dm_target *ti)
{
struct pool_c *pt = ti->private;
/*
* We want to make sure that degraded pools are never upgraded.
*/
enum pool_mode old_mode = pool->pf.mode;
enum pool_mode new_mode = pt->adjusted_pf.mode;
if (old_mode > new_mode)
new_mode = old_mode;
pool->ti = ti;
pool->low_water_blocks = pt->low_water_blocks;
pool->pf = pt->adjusted_pf;
set_pool_mode(pool, new_mode);
return 0;
}
static void unbind_control_target(struct pool *pool, struct dm_target *ti)
{
if (pool->ti == ti)
pool->ti = NULL;
}
/*----------------------------------------------------------------
* Pool creation
*--------------------------------------------------------------*/
/* Initialize pool features. */
static void pool_features_init(struct pool_features *pf)
{
pf->mode = PM_WRITE;
pf->zero_new_blocks = true;
pf->discard_enabled = true;
pf->discard_passdown = true;
}
static void __pool_destroy(struct pool *pool)
{
__pool_table_remove(pool);
if (dm_pool_metadata_close(pool->pmd) < 0)
DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
dm_bio_prison_destroy(pool->prison);
dm_kcopyd_client_destroy(pool->copier);
if (pool->wq)
destroy_workqueue(pool->wq);
if (pool->next_mapping)
mempool_free(pool->next_mapping, pool->mapping_pool);
mempool_destroy(pool->mapping_pool);
dm_deferred_set_destroy(pool->shared_read_ds);
dm_deferred_set_destroy(pool->all_io_ds);
kfree(pool);
}
static struct kmem_cache *_new_mapping_cache;
static struct pool *pool_create(struct mapped_device *pool_md,
struct block_device *metadata_dev,
unsigned long block_size,
int read_only, char **error)
{
int r;
void *err_p;
struct pool *pool;
struct dm_pool_metadata *pmd;
bool format_device = read_only ? false : true;
pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device);
if (IS_ERR(pmd)) {
*error = "Error creating metadata object";
return (struct pool *)pmd;
}
pool = kmalloc(sizeof(*pool), GFP_KERNEL);
if (!pool) {
*error = "Error allocating memory for pool";
err_p = ERR_PTR(-ENOMEM);
goto bad_pool;
}
pool->pmd = pmd;
pool->sectors_per_block = block_size;
if (block_size & (block_size - 1))
pool->sectors_per_block_shift = -1;
else
pool->sectors_per_block_shift = __ffs(block_size);
pool->low_water_blocks = 0;
pool_features_init(&pool->pf);
pool->prison = dm_bio_prison_create(PRISON_CELLS);
if (!pool->prison) {
*error = "Error creating pool's bio prison";
err_p = ERR_PTR(-ENOMEM);
goto bad_prison;
}
pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle);
if (IS_ERR(pool->copier)) {
r = PTR_ERR(pool->copier);
*error = "Error creating pool's kcopyd client";
err_p = ERR_PTR(r);
goto bad_kcopyd_client;
}
/*
* Create singlethreaded workqueue that will service all devices
* that use this metadata.
*/
pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
if (!pool->wq) {
*error = "Error creating pool's workqueue";
err_p = ERR_PTR(-ENOMEM);
goto bad_wq;
}
INIT_WORK(&pool->worker, do_worker);
INIT_DELAYED_WORK(&pool->waker, do_waker);
spin_lock_init(&pool->lock);
bio_list_init(&pool->deferred_bios);
bio_list_init(&pool->deferred_flush_bios);
INIT_LIST_HEAD(&pool->prepared_mappings);
INIT_LIST_HEAD(&pool->prepared_discards);
pool->low_water_triggered = 0;
pool->no_free_space = 0;
bio_list_init(&pool->retry_on_resume_list);
pool->shared_read_ds = dm_deferred_set_create();
if (!pool->shared_read_ds) {
*error = "Error creating pool's shared read deferred set";
err_p = ERR_PTR(-ENOMEM);
goto bad_shared_read_ds;
}
pool->all_io_ds = dm_deferred_set_create();
if (!pool->all_io_ds) {
*error = "Error creating pool's all io deferred set";
err_p = ERR_PTR(-ENOMEM);
goto bad_all_io_ds;
}
pool->next_mapping = NULL;
pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE,
_new_mapping_cache);
if (!pool->mapping_pool) {
*error = "Error creating pool's mapping mempool";
err_p = ERR_PTR(-ENOMEM);
goto bad_mapping_pool;
}
pool->ref_count = 1;
pool->last_commit_jiffies = jiffies;
pool->pool_md = pool_md;
pool->md_dev = metadata_dev;
__pool_table_insert(pool);
return pool;
bad_mapping_pool:
dm_deferred_set_destroy(pool->all_io_ds);
bad_all_io_ds:
dm_deferred_set_destroy(pool->shared_read_ds);
bad_shared_read_ds:
destroy_workqueue(pool->wq);
bad_wq:
dm_kcopyd_client_destroy(pool->copier);
bad_kcopyd_client:
dm_bio_prison_destroy(pool->prison);
bad_prison:
kfree(pool);
bad_pool:
if (dm_pool_metadata_close(pmd))
DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
return err_p;
}
static void __pool_inc(struct pool *pool)
{
BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
pool->ref_count++;
}
static void __pool_dec(struct pool *pool)
{
BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
BUG_ON(!pool->ref_count);
if (!--pool->ref_count)
__pool_destroy(pool);
}
static struct pool *__pool_find(struct mapped_device *pool_md,
struct block_device *metadata_dev,
unsigned long block_size, int read_only,
char **error, int *created)
{
struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev);
if (pool) {
if (pool->pool_md != pool_md) {
*error = "metadata device already in use by a pool";
return ERR_PTR(-EBUSY);
}
__pool_inc(pool);
} else {
pool = __pool_table_lookup(pool_md);
if (pool) {
if (pool->md_dev != metadata_dev) {
*error = "different pool cannot replace a pool";
return ERR_PTR(-EINVAL);
}
__pool_inc(pool);
} else {
pool = pool_create(pool_md, metadata_dev, block_size, read_only, error);
*created = 1;
}
}
return pool;
}
/*----------------------------------------------------------------
* Pool target methods
*--------------------------------------------------------------*/
static void pool_dtr(struct dm_target *ti)
{
struct pool_c *pt = ti->private;
mutex_lock(&dm_thin_pool_table.mutex);
unbind_control_target(pt->pool, ti);
__pool_dec(pt->pool);
dm_put_device(ti, pt->metadata_dev);
dm_put_device(ti, pt->data_dev);
kfree(pt);
mutex_unlock(&dm_thin_pool_table.mutex);
}
static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf,
struct dm_target *ti)
{
int r;
unsigned argc;
const char *arg_name;
static struct dm_arg _args[] = {
{0, 3, "Invalid number of pool feature arguments"},
};
/*
* No feature arguments supplied.
*/
if (!as->argc)
return 0;
r = dm_read_arg_group(_args, as, &argc, &ti->error);
if (r)
return -EINVAL;
while (argc && !r) {
arg_name = dm_shift_arg(as);
argc--;
if (!strcasecmp(arg_name, "skip_block_zeroing"))
pf->zero_new_blocks = false;
else if (!strcasecmp(arg_name, "ignore_discard"))
pf->discard_enabled = false;
else if (!strcasecmp(arg_name, "no_discard_passdown"))
pf->discard_passdown = false;
else if (!strcasecmp(arg_name, "read_only"))
pf->mode = PM_READ_ONLY;
else {
ti->error = "Unrecognised pool feature requested";
r = -EINVAL;
break;
}
}
return r;
}
static void metadata_low_callback(void *context)
{
struct pool *pool = context;
DMWARN("%s: reached low water mark for metadata device: sending event.",
dm_device_name(pool->pool_md));
dm_table_event(pool->ti->table);
}
static sector_t get_metadata_dev_size(struct block_device *bdev)
{
sector_t metadata_dev_size = i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
char buffer[BDEVNAME_SIZE];
if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING) {
DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS);
metadata_dev_size = THIN_METADATA_MAX_SECTORS_WARNING;
}
return metadata_dev_size;
}
static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev)
{
sector_t metadata_dev_size = get_metadata_dev_size(bdev);
sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE >> SECTOR_SHIFT);
return metadata_dev_size;
}
/*
* When a metadata threshold is crossed a dm event is triggered, and
* userland should respond by growing the metadata device. We could let
* userland set the threshold, like we do with the data threshold, but I'm
* not sure they know enough to do this well.
*/
static dm_block_t calc_metadata_threshold(struct pool_c *pt)
{
/*
* 4M is ample for all ops with the possible exception of thin
* device deletion which is harmless if it fails (just retry the
* delete after you've grown the device).
*/
dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4;
return min((dm_block_t)1024ULL /* 4M */, quarter);
}
/*
* thin-pool <metadata dev> <data dev>
* <data block size (sectors)>
* <low water mark (blocks)>
* [<#feature args> [<arg>]*]
*
* Optional feature arguments are:
* skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
* ignore_discard: disable discard
* no_discard_passdown: don't pass discards down to the data device
*/
static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv)
{
int r, pool_created = 0;
struct pool_c *pt;
struct pool *pool;
struct pool_features pf;
struct dm_arg_set as;
struct dm_dev *data_dev;
unsigned long block_size;
dm_block_t low_water_blocks;
struct dm_dev *metadata_dev;
fmode_t metadata_mode;
/*
* FIXME Remove validation from scope of lock.
*/
mutex_lock(&dm_thin_pool_table.mutex);
if (argc < 4) {
ti->error = "Invalid argument count";
r = -EINVAL;
goto out_unlock;
}
as.argc = argc;
as.argv = argv;
/*
* Set default pool features.
*/
pool_features_init(&pf);
dm_consume_args(&as, 4);
r = parse_pool_features(&as, &pf, ti);
if (r)
goto out_unlock;
metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE);
r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev);
if (r) {
ti->error = "Error opening metadata block device";
goto out_unlock;
}
/*
* Run for the side-effect of possibly issuing a warning if the
* device is too big.
*/
(void) get_metadata_dev_size(metadata_dev->bdev);
r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev);
if (r) {
ti->error = "Error getting data device";
goto out_metadata;
}
if (kstrtoul(argv[2], 10, &block_size) || !block_size ||
block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) {
ti->error = "Invalid block size";
r = -EINVAL;
goto out;
}
if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) {
ti->error = "Invalid low water mark";
r = -EINVAL;
goto out;
}
pt = kzalloc(sizeof(*pt), GFP_KERNEL);
if (!pt) {
r = -ENOMEM;
goto out;
}
pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev,
block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created);
if (IS_ERR(pool)) {
r = PTR_ERR(pool);
goto out_free_pt;
}
/*
* 'pool_created' reflects whether this is the first table load.
* Top level discard support is not allowed to be changed after
* initial load. This would require a pool reload to trigger thin
* device changes.
*/
if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) {
ti->error = "Discard support cannot be disabled once enabled";
r = -EINVAL;
goto out_flags_changed;
}
pt->pool = pool;
pt->ti = ti;
pt->metadata_dev = metadata_dev;
pt->data_dev = data_dev;
pt->low_water_blocks = low_water_blocks;
pt->adjusted_pf = pt->requested_pf = pf;
ti->num_flush_bios = 1;
/*
* Only need to enable discards if the pool should pass
* them down to the data device. The thin device's discard
* processing will cause mappings to be removed from the btree.
*/
if (pf.discard_enabled && pf.discard_passdown) {
ti->num_discard_bios = 1;
/*
* Setting 'discards_supported' circumvents the normal
* stacking of discard limits (this keeps the pool and
* thin devices' discard limits consistent).
*/
ti->discards_supported = true;
ti->discard_zeroes_data_unsupported = true;
}
ti->private = pt;
r = dm_pool_register_metadata_threshold(pt->pool->pmd,
calc_metadata_threshold(pt),
metadata_low_callback,
pool);
if (r)
goto out_free_pt;
pt->callbacks.congested_fn = pool_is_congested;
dm_table_add_target_callbacks(ti->table, &pt->callbacks);
mutex_unlock(&dm_thin_pool_table.mutex);
return 0;
out_flags_changed:
__pool_dec(pool);
out_free_pt:
kfree(pt);
out:
dm_put_device(ti, data_dev);
out_metadata:
dm_put_device(ti, metadata_dev);
out_unlock:
mutex_unlock(&dm_thin_pool_table.mutex);
return r;
}
static int pool_map(struct dm_target *ti, struct bio *bio)
{
int r;
struct pool_c *pt = ti->private;
struct pool *pool = pt->pool;
unsigned long flags;
/*
* As this is a singleton target, ti->begin is always zero.
*/
spin_lock_irqsave(&pool->lock, flags);
bio->bi_bdev = pt->data_dev->bdev;
r = DM_MAPIO_REMAPPED;
spin_unlock_irqrestore(&pool->lock, flags);
return r;
}
static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit)
{
int r;
struct pool_c *pt = ti->private;
struct pool *pool = pt->pool;
sector_t data_size = ti->len;
dm_block_t sb_data_size;
*need_commit = false;
(void) sector_div(data_size, pool->sectors_per_block);
r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size);
if (r) {
DMERR("failed to retrieve data device size");
return r;
}
if (data_size < sb_data_size) {
DMERR("pool target (%llu blocks) too small: expected %llu",
(unsigned long long)data_size, sb_data_size);
return -EINVAL;
} else if (data_size > sb_data_size) {
r = dm_pool_resize_data_dev(pool->pmd, data_size);
if (r) {
DMERR("failed to resize data device");
set_pool_mode(pool, PM_READ_ONLY);
return r;
}
*need_commit = true;
}
return 0;
}
static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit)
{
int r;
struct pool_c *pt = ti->private;
struct pool *pool = pt->pool;
dm_block_t metadata_dev_size, sb_metadata_dev_size;
*need_commit = false;
metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev);
r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size);
if (r) {
DMERR("failed to retrieve data device size");
return r;
}
if (metadata_dev_size < sb_metadata_dev_size) {
DMERR("metadata device (%llu blocks) too small: expected %llu",
metadata_dev_size, sb_metadata_dev_size);
return -EINVAL;
} else if (metadata_dev_size > sb_metadata_dev_size) {
r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size);
if (r) {
DMERR("failed to resize metadata device");
return r;
}
*need_commit = true;
}
return 0;
}
/*
* Retrieves the number of blocks of the data device from
* the superblock and compares it to the actual device size,
* thus resizing the data device in case it has grown.
*
* This both copes with opening preallocated data devices in the ctr
* being followed by a resume
* -and-
* calling the resume method individually after userspace has
* grown the data device in reaction to a table event.
*/
static int pool_preresume(struct dm_target *ti)
{
int r;
bool need_commit1, need_commit2;
struct pool_c *pt = ti->private;
struct pool *pool = pt->pool;
/*
* Take control of the pool object.
*/
r = bind_control_target(pool, ti);
if (r)
return r;
r = maybe_resize_data_dev(ti, &need_commit1);
if (r)
return r;
r = maybe_resize_metadata_dev(ti, &need_commit2);
if (r)
return r;
if (need_commit1 || need_commit2)
(void) commit_or_fallback(pool);
return 0;
}
static void pool_resume(struct dm_target *ti)
{
struct pool_c *pt = ti->private;
struct pool *pool = pt->pool;
unsigned long flags;
spin_lock_irqsave(&pool->lock, flags);
pool->low_water_triggered = 0;
pool->no_free_space = 0;
__requeue_bios(pool);
spin_unlock_irqrestore(&pool->lock, flags);
do_waker(&pool->waker.work);
}
static void pool_postsuspend(struct dm_target *ti)
{
struct pool_c *pt = ti->private;
struct pool *pool = pt->pool;
cancel_delayed_work(&pool->waker);
flush_workqueue(pool->wq);
(void) commit_or_fallback(pool);
}
static int check_arg_count(unsigned argc, unsigned args_required)
{
if (argc != args_required) {
DMWARN("Message received with %u arguments instead of %u.",
argc, args_required);
return -EINVAL;
}
return 0;
}
static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning)
{
if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) &&
*dev_id <= MAX_DEV_ID)
return 0;
if (warning)
DMWARN("Message received with invalid device id: %s", arg);
return -EINVAL;
}
static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool)
{
dm_thin_id dev_id;
int r;
r = check_arg_count(argc, 2);
if (r)
return r;
r = read_dev_id(argv[1], &dev_id, 1);
if (r)
return r;
r = dm_pool_create_thin(pool->pmd, dev_id);
if (r) {
DMWARN("Creation of new thinly-provisioned device with id %s failed.",
argv[1]);
return r;
}
return 0;
}
static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool)
{
dm_thin_id dev_id;
dm_thin_id origin_dev_id;
int r;
r = check_arg_count(argc, 3);
if (r)
return r;
r = read_dev_id(argv[1], &dev_id, 1);
if (r)
return r;
r = read_dev_id(argv[2], &origin_dev_id, 1);
if (r)
return r;
r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id);
if (r) {
DMWARN("Creation of new snapshot %s of device %s failed.",
argv[1], argv[2]);
return r;
}
return 0;
}
static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool)
{
dm_thin_id dev_id;
int r;
r = check_arg_count(argc, 2);
if (r)
return r;
r = read_dev_id(argv[1], &dev_id, 1);
if (r)
return r;
r = dm_pool_delete_thin_device(pool->pmd, dev_id);
if (r)
DMWARN("Deletion of thin device %s failed.", argv[1]);
return r;
}
static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool)
{
dm_thin_id old_id, new_id;
int r;
r = check_arg_count(argc, 3);
if (r)
return r;
if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) {
DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]);
return -EINVAL;
}
if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) {
DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]);
return -EINVAL;
}
r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id);
if (r) {
DMWARN("Failed to change transaction id from %s to %s.",
argv[1], argv[2]);
return r;
}
return 0;
}
static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
{
int r;
r = check_arg_count(argc, 1);
if (r)
return r;
(void) commit_or_fallback(pool);
r = dm_pool_reserve_metadata_snap(pool->pmd);
if (r)
DMWARN("reserve_metadata_snap message failed.");
return r;
}
static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
{
int r;
r = check_arg_count(argc, 1);
if (r)
return r;
r = dm_pool_release_metadata_snap(pool->pmd);
if (r)
DMWARN("release_metadata_snap message failed.");
return r;
}
/*
* Messages supported:
* create_thin <dev_id>
* create_snap <dev_id> <origin_id>
* delete <dev_id>
* trim <dev_id> <new_size_in_sectors>
* set_transaction_id <current_trans_id> <new_trans_id>
* reserve_metadata_snap
* release_metadata_snap
*/
static int pool_message(struct dm_target *ti, unsigned argc, char **argv)
{
int r = -EINVAL;
struct pool_c *pt = ti->private;
struct pool *pool = pt->pool;
if (!strcasecmp(argv[0], "create_thin"))
r = process_create_thin_mesg(argc, argv, pool);
else if (!strcasecmp(argv[0], "create_snap"))
r = process_create_snap_mesg(argc, argv, pool);
else if (!strcasecmp(argv[0], "delete"))
r = process_delete_mesg(argc, argv, pool);
else if (!strcasecmp(argv[0], "set_transaction_id"))
r = process_set_transaction_id_mesg(argc, argv, pool);
else if (!strcasecmp(argv[0], "reserve_metadata_snap"))
r = process_reserve_metadata_snap_mesg(argc, argv, pool);
else if (!strcasecmp(argv[0], "release_metadata_snap"))
r = process_release_metadata_snap_mesg(argc, argv, pool);
else
DMWARN("Unrecognised thin pool target message received: %s", argv[0]);
if (!r)
(void) commit_or_fallback(pool);
return r;
}
static void emit_flags(struct pool_features *pf, char *result,
unsigned sz, unsigned maxlen)
{
unsigned count = !pf->zero_new_blocks + !pf->discard_enabled +
!pf->discard_passdown + (pf->mode == PM_READ_ONLY);
DMEMIT("%u ", count);
if (!pf->zero_new_blocks)
DMEMIT("skip_block_zeroing ");
if (!pf->discard_enabled)
DMEMIT("ignore_discard ");
if (!pf->discard_passdown)
DMEMIT("no_discard_passdown ");
if (pf->mode == PM_READ_ONLY)
DMEMIT("read_only ");
}
/*
* Status line is:
* <transaction id> <used metadata sectors>/<total metadata sectors>
* <used data sectors>/<total data sectors> <held metadata root>
*/
dm: fix truncated status strings Avoid returning a truncated table or status string instead of setting the DM_BUFFER_FULL_FLAG when the last target of a table fills the buffer. When processing a table or status request, the function retrieve_status calls ti->type->status. If ti->type->status returns non-zero, retrieve_status assumes that the buffer overflowed and sets DM_BUFFER_FULL_FLAG. However, targets don't return non-zero values from their status method on overflow. Most targets returns always zero. If a buffer overflow happens in a target that is not the last in the table, it gets noticed during the next iteration of the loop in retrieve_status; but if a buffer overflow happens in the last target, it goes unnoticed and erroneously truncated data is returned. In the current code, the targets behave in the following way: * dm-crypt returns -ENOMEM if there is not enough space to store the key, but it returns 0 on all other overflows. * dm-thin returns errors from the status method if a disk error happened. This is incorrect because retrieve_status doesn't check the error code, it assumes that all non-zero values mean buffer overflow. * all the other targets always return 0. This patch changes the ti->type->status function to return void (because most targets don't use the return code). Overflow is detected in retrieve_status: if the status method fills up the remaining space completely, it is assumed that buffer overflow happened. Cc: stable@vger.kernel.org Signed-off-by: Mikulas Patocka <mpatocka@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com>
2013-03-02 02:45:44 +04:00
static void pool_status(struct dm_target *ti, status_type_t type,
unsigned status_flags, char *result, unsigned maxlen)
{
int r;
unsigned sz = 0;
uint64_t transaction_id;
dm_block_t nr_free_blocks_data;
dm_block_t nr_free_blocks_metadata;
dm_block_t nr_blocks_data;
dm_block_t nr_blocks_metadata;
dm_block_t held_root;
char buf[BDEVNAME_SIZE];
char buf2[BDEVNAME_SIZE];
struct pool_c *pt = ti->private;
struct pool *pool = pt->pool;
switch (type) {
case STATUSTYPE_INFO:
if (get_pool_mode(pool) == PM_FAIL) {
DMEMIT("Fail");
break;
}
/* Commit to ensure statistics aren't out-of-date */
if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti))
(void) commit_or_fallback(pool);
dm: fix truncated status strings Avoid returning a truncated table or status string instead of setting the DM_BUFFER_FULL_FLAG when the last target of a table fills the buffer. When processing a table or status request, the function retrieve_status calls ti->type->status. If ti->type->status returns non-zero, retrieve_status assumes that the buffer overflowed and sets DM_BUFFER_FULL_FLAG. However, targets don't return non-zero values from their status method on overflow. Most targets returns always zero. If a buffer overflow happens in a target that is not the last in the table, it gets noticed during the next iteration of the loop in retrieve_status; but if a buffer overflow happens in the last target, it goes unnoticed and erroneously truncated data is returned. In the current code, the targets behave in the following way: * dm-crypt returns -ENOMEM if there is not enough space to store the key, but it returns 0 on all other overflows. * dm-thin returns errors from the status method if a disk error happened. This is incorrect because retrieve_status doesn't check the error code, it assumes that all non-zero values mean buffer overflow. * all the other targets always return 0. This patch changes the ti->type->status function to return void (because most targets don't use the return code). Overflow is detected in retrieve_status: if the status method fills up the remaining space completely, it is assumed that buffer overflow happened. Cc: stable@vger.kernel.org Signed-off-by: Mikulas Patocka <mpatocka@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com>
2013-03-02 02:45:44 +04:00
r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id);
if (r) {
DMERR("dm_pool_get_metadata_transaction_id returned %d", r);
goto err;
}
dm: fix truncated status strings Avoid returning a truncated table or status string instead of setting the DM_BUFFER_FULL_FLAG when the last target of a table fills the buffer. When processing a table or status request, the function retrieve_status calls ti->type->status. If ti->type->status returns non-zero, retrieve_status assumes that the buffer overflowed and sets DM_BUFFER_FULL_FLAG. However, targets don't return non-zero values from their status method on overflow. Most targets returns always zero. If a buffer overflow happens in a target that is not the last in the table, it gets noticed during the next iteration of the loop in retrieve_status; but if a buffer overflow happens in the last target, it goes unnoticed and erroneously truncated data is returned. In the current code, the targets behave in the following way: * dm-crypt returns -ENOMEM if there is not enough space to store the key, but it returns 0 on all other overflows. * dm-thin returns errors from the status method if a disk error happened. This is incorrect because retrieve_status doesn't check the error code, it assumes that all non-zero values mean buffer overflow. * all the other targets always return 0. This patch changes the ti->type->status function to return void (because most targets don't use the return code). Overflow is detected in retrieve_status: if the status method fills up the remaining space completely, it is assumed that buffer overflow happened. Cc: stable@vger.kernel.org Signed-off-by: Mikulas Patocka <mpatocka@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com>
2013-03-02 02:45:44 +04:00
r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata);
if (r) {
DMERR("dm_pool_get_free_metadata_block_count returned %d", r);
goto err;
}
r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata);
dm: fix truncated status strings Avoid returning a truncated table or status string instead of setting the DM_BUFFER_FULL_FLAG when the last target of a table fills the buffer. When processing a table or status request, the function retrieve_status calls ti->type->status. If ti->type->status returns non-zero, retrieve_status assumes that the buffer overflowed and sets DM_BUFFER_FULL_FLAG. However, targets don't return non-zero values from their status method on overflow. Most targets returns always zero. If a buffer overflow happens in a target that is not the last in the table, it gets noticed during the next iteration of the loop in retrieve_status; but if a buffer overflow happens in the last target, it goes unnoticed and erroneously truncated data is returned. In the current code, the targets behave in the following way: * dm-crypt returns -ENOMEM if there is not enough space to store the key, but it returns 0 on all other overflows. * dm-thin returns errors from the status method if a disk error happened. This is incorrect because retrieve_status doesn't check the error code, it assumes that all non-zero values mean buffer overflow. * all the other targets always return 0. This patch changes the ti->type->status function to return void (because most targets don't use the return code). Overflow is detected in retrieve_status: if the status method fills up the remaining space completely, it is assumed that buffer overflow happened. Cc: stable@vger.kernel.org Signed-off-by: Mikulas Patocka <mpatocka@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com>
2013-03-02 02:45:44 +04:00
if (r) {
DMERR("dm_pool_get_metadata_dev_size returned %d", r);
goto err;
}
dm: fix truncated status strings Avoid returning a truncated table or status string instead of setting the DM_BUFFER_FULL_FLAG when the last target of a table fills the buffer. When processing a table or status request, the function retrieve_status calls ti->type->status. If ti->type->status returns non-zero, retrieve_status assumes that the buffer overflowed and sets DM_BUFFER_FULL_FLAG. However, targets don't return non-zero values from their status method on overflow. Most targets returns always zero. If a buffer overflow happens in a target that is not the last in the table, it gets noticed during the next iteration of the loop in retrieve_status; but if a buffer overflow happens in the last target, it goes unnoticed and erroneously truncated data is returned. In the current code, the targets behave in the following way: * dm-crypt returns -ENOMEM if there is not enough space to store the key, but it returns 0 on all other overflows. * dm-thin returns errors from the status method if a disk error happened. This is incorrect because retrieve_status doesn't check the error code, it assumes that all non-zero values mean buffer overflow. * all the other targets always return 0. This patch changes the ti->type->status function to return void (because most targets don't use the return code). Overflow is detected in retrieve_status: if the status method fills up the remaining space completely, it is assumed that buffer overflow happened. Cc: stable@vger.kernel.org Signed-off-by: Mikulas Patocka <mpatocka@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com>
2013-03-02 02:45:44 +04:00
r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data);
if (r) {
DMERR("dm_pool_get_free_block_count returned %d", r);
goto err;
}
r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data);
dm: fix truncated status strings Avoid returning a truncated table or status string instead of setting the DM_BUFFER_FULL_FLAG when the last target of a table fills the buffer. When processing a table or status request, the function retrieve_status calls ti->type->status. If ti->type->status returns non-zero, retrieve_status assumes that the buffer overflowed and sets DM_BUFFER_FULL_FLAG. However, targets don't return non-zero values from their status method on overflow. Most targets returns always zero. If a buffer overflow happens in a target that is not the last in the table, it gets noticed during the next iteration of the loop in retrieve_status; but if a buffer overflow happens in the last target, it goes unnoticed and erroneously truncated data is returned. In the current code, the targets behave in the following way: * dm-crypt returns -ENOMEM if there is not enough space to store the key, but it returns 0 on all other overflows. * dm-thin returns errors from the status method if a disk error happened. This is incorrect because retrieve_status doesn't check the error code, it assumes that all non-zero values mean buffer overflow. * all the other targets always return 0. This patch changes the ti->type->status function to return void (because most targets don't use the return code). Overflow is detected in retrieve_status: if the status method fills up the remaining space completely, it is assumed that buffer overflow happened. Cc: stable@vger.kernel.org Signed-off-by: Mikulas Patocka <mpatocka@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com>
2013-03-02 02:45:44 +04:00
if (r) {
DMERR("dm_pool_get_data_dev_size returned %d", r);
goto err;
}
r = dm_pool_get_metadata_snap(pool->pmd, &held_root);
dm: fix truncated status strings Avoid returning a truncated table or status string instead of setting the DM_BUFFER_FULL_FLAG when the last target of a table fills the buffer. When processing a table or status request, the function retrieve_status calls ti->type->status. If ti->type->status returns non-zero, retrieve_status assumes that the buffer overflowed and sets DM_BUFFER_FULL_FLAG. However, targets don't return non-zero values from their status method on overflow. Most targets returns always zero. If a buffer overflow happens in a target that is not the last in the table, it gets noticed during the next iteration of the loop in retrieve_status; but if a buffer overflow happens in the last target, it goes unnoticed and erroneously truncated data is returned. In the current code, the targets behave in the following way: * dm-crypt returns -ENOMEM if there is not enough space to store the key, but it returns 0 on all other overflows. * dm-thin returns errors from the status method if a disk error happened. This is incorrect because retrieve_status doesn't check the error code, it assumes that all non-zero values mean buffer overflow. * all the other targets always return 0. This patch changes the ti->type->status function to return void (because most targets don't use the return code). Overflow is detected in retrieve_status: if the status method fills up the remaining space completely, it is assumed that buffer overflow happened. Cc: stable@vger.kernel.org Signed-off-by: Mikulas Patocka <mpatocka@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com>
2013-03-02 02:45:44 +04:00
if (r) {
DMERR("dm_pool_get_metadata_snap returned %d", r);
goto err;
}
DMEMIT("%llu %llu/%llu %llu/%llu ",
(unsigned long long)transaction_id,
(unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
(unsigned long long)nr_blocks_metadata,
(unsigned long long)(nr_blocks_data - nr_free_blocks_data),
(unsigned long long)nr_blocks_data);
if (held_root)
DMEMIT("%llu ", held_root);
else
DMEMIT("- ");
if (pool->pf.mode == PM_READ_ONLY)
DMEMIT("ro ");
else
DMEMIT("rw ");
if (!pool->pf.discard_enabled)
DMEMIT("ignore_discard");
else if (pool->pf.discard_passdown)
DMEMIT("discard_passdown");
else
DMEMIT("no_discard_passdown");
break;
case STATUSTYPE_TABLE:
DMEMIT("%s %s %lu %llu ",
format_dev_t(buf, pt->metadata_dev->bdev->bd_dev),
format_dev_t(buf2, pt->data_dev->bdev->bd_dev),
(unsigned long)pool->sectors_per_block,
(unsigned long long)pt->low_water_blocks);
emit_flags(&pt->requested_pf, result, sz, maxlen);
break;
}
dm: fix truncated status strings Avoid returning a truncated table or status string instead of setting the DM_BUFFER_FULL_FLAG when the last target of a table fills the buffer. When processing a table or status request, the function retrieve_status calls ti->type->status. If ti->type->status returns non-zero, retrieve_status assumes that the buffer overflowed and sets DM_BUFFER_FULL_FLAG. However, targets don't return non-zero values from their status method on overflow. Most targets returns always zero. If a buffer overflow happens in a target that is not the last in the table, it gets noticed during the next iteration of the loop in retrieve_status; but if a buffer overflow happens in the last target, it goes unnoticed and erroneously truncated data is returned. In the current code, the targets behave in the following way: * dm-crypt returns -ENOMEM if there is not enough space to store the key, but it returns 0 on all other overflows. * dm-thin returns errors from the status method if a disk error happened. This is incorrect because retrieve_status doesn't check the error code, it assumes that all non-zero values mean buffer overflow. * all the other targets always return 0. This patch changes the ti->type->status function to return void (because most targets don't use the return code). Overflow is detected in retrieve_status: if the status method fills up the remaining space completely, it is assumed that buffer overflow happened. Cc: stable@vger.kernel.org Signed-off-by: Mikulas Patocka <mpatocka@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com>
2013-03-02 02:45:44 +04:00
return;
dm: fix truncated status strings Avoid returning a truncated table or status string instead of setting the DM_BUFFER_FULL_FLAG when the last target of a table fills the buffer. When processing a table or status request, the function retrieve_status calls ti->type->status. If ti->type->status returns non-zero, retrieve_status assumes that the buffer overflowed and sets DM_BUFFER_FULL_FLAG. However, targets don't return non-zero values from their status method on overflow. Most targets returns always zero. If a buffer overflow happens in a target that is not the last in the table, it gets noticed during the next iteration of the loop in retrieve_status; but if a buffer overflow happens in the last target, it goes unnoticed and erroneously truncated data is returned. In the current code, the targets behave in the following way: * dm-crypt returns -ENOMEM if there is not enough space to store the key, but it returns 0 on all other overflows. * dm-thin returns errors from the status method if a disk error happened. This is incorrect because retrieve_status doesn't check the error code, it assumes that all non-zero values mean buffer overflow. * all the other targets always return 0. This patch changes the ti->type->status function to return void (because most targets don't use the return code). Overflow is detected in retrieve_status: if the status method fills up the remaining space completely, it is assumed that buffer overflow happened. Cc: stable@vger.kernel.org Signed-off-by: Mikulas Patocka <mpatocka@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com>
2013-03-02 02:45:44 +04:00
err:
DMEMIT("Error");
}
static int pool_iterate_devices(struct dm_target *ti,
iterate_devices_callout_fn fn, void *data)
{
struct pool_c *pt = ti->private;
return fn(ti, pt->data_dev, 0, ti->len, data);
}
static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
struct bio_vec *biovec, int max_size)
{
struct pool_c *pt = ti->private;
struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
if (!q->merge_bvec_fn)
return max_size;
bvm->bi_bdev = pt->data_dev->bdev;
return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
}
static void set_discard_limits(struct pool_c *pt, struct queue_limits *limits)
{
struct pool *pool = pt->pool;
struct queue_limits *data_limits;
limits->max_discard_sectors = pool->sectors_per_block;
/*
* discard_granularity is just a hint, and not enforced.
*/
if (pt->adjusted_pf.discard_passdown) {
data_limits = &bdev_get_queue(pt->data_dev->bdev)->limits;
limits->discard_granularity = data_limits->discard_granularity;
} else
limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT;
}
static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits)
{
struct pool_c *pt = ti->private;
struct pool *pool = pt->pool;
blk_limits_io_min(limits, 0);
blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
/*
* pt->adjusted_pf is a staging area for the actual features to use.
* They get transferred to the live pool in bind_control_target()
* called from pool_preresume().
*/
if (!pt->adjusted_pf.discard_enabled)
return;
disable_passdown_if_not_supported(pt);
set_discard_limits(pt, limits);
}
static struct target_type pool_target = {
.name = "thin-pool",
.features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE |
DM_TARGET_IMMUTABLE,
.version = {1, 8, 0},
.module = THIS_MODULE,
.ctr = pool_ctr,
.dtr = pool_dtr,
.map = pool_map,
.postsuspend = pool_postsuspend,
.preresume = pool_preresume,
.resume = pool_resume,
.message = pool_message,
.status = pool_status,
.merge = pool_merge,
.iterate_devices = pool_iterate_devices,
.io_hints = pool_io_hints,
};
/*----------------------------------------------------------------
* Thin target methods
*--------------------------------------------------------------*/
static void thin_dtr(struct dm_target *ti)
{
struct thin_c *tc = ti->private;
mutex_lock(&dm_thin_pool_table.mutex);
__pool_dec(tc->pool);
dm_pool_close_thin_device(tc->td);
dm_put_device(ti, tc->pool_dev);
if (tc->origin_dev)
dm_put_device(ti, tc->origin_dev);
kfree(tc);
mutex_unlock(&dm_thin_pool_table.mutex);
}
/*
* Thin target parameters:
*
* <pool_dev> <dev_id> [origin_dev]
*
* pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
* dev_id: the internal device identifier
* origin_dev: a device external to the pool that should act as the origin
*
* If the pool device has discards disabled, they get disabled for the thin
* device as well.
*/
static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv)
{
int r;
struct thin_c *tc;
struct dm_dev *pool_dev, *origin_dev;
struct mapped_device *pool_md;
mutex_lock(&dm_thin_pool_table.mutex);
if (argc != 2 && argc != 3) {
ti->error = "Invalid argument count";
r = -EINVAL;
goto out_unlock;
}
tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL);
if (!tc) {
ti->error = "Out of memory";
r = -ENOMEM;
goto out_unlock;
}
if (argc == 3) {
r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev);
if (r) {
ti->error = "Error opening origin device";
goto bad_origin_dev;
}
tc->origin_dev = origin_dev;
}
r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev);
if (r) {
ti->error = "Error opening pool device";
goto bad_pool_dev;
}
tc->pool_dev = pool_dev;
if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) {
ti->error = "Invalid device id";
r = -EINVAL;
goto bad_common;
}
pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev);
if (!pool_md) {
ti->error = "Couldn't get pool mapped device";
r = -EINVAL;
goto bad_common;
}
tc->pool = __pool_table_lookup(pool_md);
if (!tc->pool) {
ti->error = "Couldn't find pool object";
r = -EINVAL;
goto bad_pool_lookup;
}
__pool_inc(tc->pool);
if (get_pool_mode(tc->pool) == PM_FAIL) {
ti->error = "Couldn't open thin device, Pool is in fail mode";
goto bad_thin_open;
}
r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td);
if (r) {
ti->error = "Couldn't open thin internal device";
goto bad_thin_open;
}
r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block);
if (r)
goto bad_thin_open;
ti->num_flush_bios = 1;
ti->flush_supported = true;
ti->per_bio_data_size = sizeof(struct dm_thin_endio_hook);
/* In case the pool supports discards, pass them on. */
if (tc->pool->pf.discard_enabled) {
ti->discards_supported = true;
ti->num_discard_bios = 1;
ti->discard_zeroes_data_unsupported = true;
/* Discard bios must be split on a block boundary */
ti->split_discard_bios = true;
}
dm_put(pool_md);
mutex_unlock(&dm_thin_pool_table.mutex);
return 0;
bad_thin_open:
__pool_dec(tc->pool);
bad_pool_lookup:
dm_put(pool_md);
bad_common:
dm_put_device(ti, tc->pool_dev);
bad_pool_dev:
if (tc->origin_dev)
dm_put_device(ti, tc->origin_dev);
bad_origin_dev:
kfree(tc);
out_unlock:
mutex_unlock(&dm_thin_pool_table.mutex);
return r;
}
static int thin_map(struct dm_target *ti, struct bio *bio)
{
bio->bi_sector = dm_target_offset(ti, bio->bi_sector);
return thin_bio_map(ti, bio);
}
static int thin_endio(struct dm_target *ti, struct bio *bio, int err)
{
unsigned long flags;
struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
struct list_head work;
struct dm_thin_new_mapping *m, *tmp;
struct pool *pool = h->tc->pool;
if (h->shared_read_entry) {
INIT_LIST_HEAD(&work);
dm_deferred_entry_dec(h->shared_read_entry, &work);
spin_lock_irqsave(&pool->lock, flags);
list_for_each_entry_safe(m, tmp, &work, list) {
list_del(&m->list);
m->quiesced = 1;
__maybe_add_mapping(m);
}
spin_unlock_irqrestore(&pool->lock, flags);
}
if (h->all_io_entry) {
INIT_LIST_HEAD(&work);
dm_deferred_entry_dec(h->all_io_entry, &work);
if (!list_empty(&work)) {
spin_lock_irqsave(&pool->lock, flags);
list_for_each_entry_safe(m, tmp, &work, list)
list_add(&m->list, &pool->prepared_discards);
spin_unlock_irqrestore(&pool->lock, flags);
wake_worker(pool);
}
}
return 0;
}
static void thin_postsuspend(struct dm_target *ti)
{
if (dm_noflush_suspending(ti))
requeue_io((struct thin_c *)ti->private);
}
/*
* <nr mapped sectors> <highest mapped sector>
*/
dm: fix truncated status strings Avoid returning a truncated table or status string instead of setting the DM_BUFFER_FULL_FLAG when the last target of a table fills the buffer. When processing a table or status request, the function retrieve_status calls ti->type->status. If ti->type->status returns non-zero, retrieve_status assumes that the buffer overflowed and sets DM_BUFFER_FULL_FLAG. However, targets don't return non-zero values from their status method on overflow. Most targets returns always zero. If a buffer overflow happens in a target that is not the last in the table, it gets noticed during the next iteration of the loop in retrieve_status; but if a buffer overflow happens in the last target, it goes unnoticed and erroneously truncated data is returned. In the current code, the targets behave in the following way: * dm-crypt returns -ENOMEM if there is not enough space to store the key, but it returns 0 on all other overflows. * dm-thin returns errors from the status method if a disk error happened. This is incorrect because retrieve_status doesn't check the error code, it assumes that all non-zero values mean buffer overflow. * all the other targets always return 0. This patch changes the ti->type->status function to return void (because most targets don't use the return code). Overflow is detected in retrieve_status: if the status method fills up the remaining space completely, it is assumed that buffer overflow happened. Cc: stable@vger.kernel.org Signed-off-by: Mikulas Patocka <mpatocka@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com>
2013-03-02 02:45:44 +04:00
static void thin_status(struct dm_target *ti, status_type_t type,
unsigned status_flags, char *result, unsigned maxlen)
{
int r;
ssize_t sz = 0;
dm_block_t mapped, highest;
char buf[BDEVNAME_SIZE];
struct thin_c *tc = ti->private;
if (get_pool_mode(tc->pool) == PM_FAIL) {
DMEMIT("Fail");
dm: fix truncated status strings Avoid returning a truncated table or status string instead of setting the DM_BUFFER_FULL_FLAG when the last target of a table fills the buffer. When processing a table or status request, the function retrieve_status calls ti->type->status. If ti->type->status returns non-zero, retrieve_status assumes that the buffer overflowed and sets DM_BUFFER_FULL_FLAG. However, targets don't return non-zero values from their status method on overflow. Most targets returns always zero. If a buffer overflow happens in a target that is not the last in the table, it gets noticed during the next iteration of the loop in retrieve_status; but if a buffer overflow happens in the last target, it goes unnoticed and erroneously truncated data is returned. In the current code, the targets behave in the following way: * dm-crypt returns -ENOMEM if there is not enough space to store the key, but it returns 0 on all other overflows. * dm-thin returns errors from the status method if a disk error happened. This is incorrect because retrieve_status doesn't check the error code, it assumes that all non-zero values mean buffer overflow. * all the other targets always return 0. This patch changes the ti->type->status function to return void (because most targets don't use the return code). Overflow is detected in retrieve_status: if the status method fills up the remaining space completely, it is assumed that buffer overflow happened. Cc: stable@vger.kernel.org Signed-off-by: Mikulas Patocka <mpatocka@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com>
2013-03-02 02:45:44 +04:00
return;
}
if (!tc->td)
DMEMIT("-");
else {
switch (type) {
case STATUSTYPE_INFO:
r = dm_thin_get_mapped_count(tc->td, &mapped);
dm: fix truncated status strings Avoid returning a truncated table or status string instead of setting the DM_BUFFER_FULL_FLAG when the last target of a table fills the buffer. When processing a table or status request, the function retrieve_status calls ti->type->status. If ti->type->status returns non-zero, retrieve_status assumes that the buffer overflowed and sets DM_BUFFER_FULL_FLAG. However, targets don't return non-zero values from their status method on overflow. Most targets returns always zero. If a buffer overflow happens in a target that is not the last in the table, it gets noticed during the next iteration of the loop in retrieve_status; but if a buffer overflow happens in the last target, it goes unnoticed and erroneously truncated data is returned. In the current code, the targets behave in the following way: * dm-crypt returns -ENOMEM if there is not enough space to store the key, but it returns 0 on all other overflows. * dm-thin returns errors from the status method if a disk error happened. This is incorrect because retrieve_status doesn't check the error code, it assumes that all non-zero values mean buffer overflow. * all the other targets always return 0. This patch changes the ti->type->status function to return void (because most targets don't use the return code). Overflow is detected in retrieve_status: if the status method fills up the remaining space completely, it is assumed that buffer overflow happened. Cc: stable@vger.kernel.org Signed-off-by: Mikulas Patocka <mpatocka@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com>
2013-03-02 02:45:44 +04:00
if (r) {
DMERR("dm_thin_get_mapped_count returned %d", r);
goto err;
}
r = dm_thin_get_highest_mapped_block(tc->td, &highest);
dm: fix truncated status strings Avoid returning a truncated table or status string instead of setting the DM_BUFFER_FULL_FLAG when the last target of a table fills the buffer. When processing a table or status request, the function retrieve_status calls ti->type->status. If ti->type->status returns non-zero, retrieve_status assumes that the buffer overflowed and sets DM_BUFFER_FULL_FLAG. However, targets don't return non-zero values from their status method on overflow. Most targets returns always zero. If a buffer overflow happens in a target that is not the last in the table, it gets noticed during the next iteration of the loop in retrieve_status; but if a buffer overflow happens in the last target, it goes unnoticed and erroneously truncated data is returned. In the current code, the targets behave in the following way: * dm-crypt returns -ENOMEM if there is not enough space to store the key, but it returns 0 on all other overflows. * dm-thin returns errors from the status method if a disk error happened. This is incorrect because retrieve_status doesn't check the error code, it assumes that all non-zero values mean buffer overflow. * all the other targets always return 0. This patch changes the ti->type->status function to return void (because most targets don't use the return code). Overflow is detected in retrieve_status: if the status method fills up the remaining space completely, it is assumed that buffer overflow happened. Cc: stable@vger.kernel.org Signed-off-by: Mikulas Patocka <mpatocka@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com>
2013-03-02 02:45:44 +04:00
if (r < 0) {
DMERR("dm_thin_get_highest_mapped_block returned %d", r);
goto err;
}
DMEMIT("%llu ", mapped * tc->pool->sectors_per_block);
if (r)
DMEMIT("%llu", ((highest + 1) *
tc->pool->sectors_per_block) - 1);
else
DMEMIT("-");
break;
case STATUSTYPE_TABLE:
DMEMIT("%s %lu",
format_dev_t(buf, tc->pool_dev->bdev->bd_dev),
(unsigned long) tc->dev_id);
if (tc->origin_dev)
DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev));
break;
}
}
dm: fix truncated status strings Avoid returning a truncated table or status string instead of setting the DM_BUFFER_FULL_FLAG when the last target of a table fills the buffer. When processing a table or status request, the function retrieve_status calls ti->type->status. If ti->type->status returns non-zero, retrieve_status assumes that the buffer overflowed and sets DM_BUFFER_FULL_FLAG. However, targets don't return non-zero values from their status method on overflow. Most targets returns always zero. If a buffer overflow happens in a target that is not the last in the table, it gets noticed during the next iteration of the loop in retrieve_status; but if a buffer overflow happens in the last target, it goes unnoticed and erroneously truncated data is returned. In the current code, the targets behave in the following way: * dm-crypt returns -ENOMEM if there is not enough space to store the key, but it returns 0 on all other overflows. * dm-thin returns errors from the status method if a disk error happened. This is incorrect because retrieve_status doesn't check the error code, it assumes that all non-zero values mean buffer overflow. * all the other targets always return 0. This patch changes the ti->type->status function to return void (because most targets don't use the return code). Overflow is detected in retrieve_status: if the status method fills up the remaining space completely, it is assumed that buffer overflow happened. Cc: stable@vger.kernel.org Signed-off-by: Mikulas Patocka <mpatocka@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com>
2013-03-02 02:45:44 +04:00
return;
err:
DMEMIT("Error");
}
static int thin_iterate_devices(struct dm_target *ti,
iterate_devices_callout_fn fn, void *data)
{
sector_t blocks;
struct thin_c *tc = ti->private;
struct pool *pool = tc->pool;
/*
* We can't call dm_pool_get_data_dev_size() since that blocks. So
* we follow a more convoluted path through to the pool's target.
*/
if (!pool->ti)
return 0; /* nothing is bound */
blocks = pool->ti->len;
(void) sector_div(blocks, pool->sectors_per_block);
if (blocks)
return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data);
return 0;
}
static struct target_type thin_target = {
.name = "thin",
.version = {1, 8, 0},
.module = THIS_MODULE,
.ctr = thin_ctr,
.dtr = thin_dtr,
.map = thin_map,
.end_io = thin_endio,
.postsuspend = thin_postsuspend,
.status = thin_status,
.iterate_devices = thin_iterate_devices,
};
/*----------------------------------------------------------------*/
static int __init dm_thin_init(void)
{
int r;
pool_table_init();
r = dm_register_target(&thin_target);
if (r)
return r;
r = dm_register_target(&pool_target);
if (r)
goto bad_pool_target;
r = -ENOMEM;
_new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0);
if (!_new_mapping_cache)
goto bad_new_mapping_cache;
return 0;
bad_new_mapping_cache:
dm_unregister_target(&pool_target);
bad_pool_target:
dm_unregister_target(&thin_target);
return r;
}
static void dm_thin_exit(void)
{
dm_unregister_target(&thin_target);
dm_unregister_target(&pool_target);
kmem_cache_destroy(_new_mapping_cache);
}
module_init(dm_thin_init);
module_exit(dm_thin_exit);
MODULE_DESCRIPTION(DM_NAME " thin provisioning target");
MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
MODULE_LICENSE("GPL");