linux/drivers/md/raid5.h

637 lines
24 KiB
C
Raw Normal View History

#ifndef _RAID5_H
#define _RAID5_H
#include <linux/raid/xor.h>
#include <linux/dmaengine.h>
/*
*
* Each stripe contains one buffer per device. Each buffer can be in
* one of a number of states stored in "flags". Changes between
* these states happen *almost* exclusively under the protection of the
* STRIPE_ACTIVE flag. Some very specific changes can happen in bi_end_io, and
* these are not protected by STRIPE_ACTIVE.
*
* The flag bits that are used to represent these states are:
* R5_UPTODATE and R5_LOCKED
*
* State Empty == !UPTODATE, !LOCK
* We have no data, and there is no active request
* State Want == !UPTODATE, LOCK
* A read request is being submitted for this block
* State Dirty == UPTODATE, LOCK
* Some new data is in this buffer, and it is being written out
* State Clean == UPTODATE, !LOCK
* We have valid data which is the same as on disc
*
* The possible state transitions are:
*
* Empty -> Want - on read or write to get old data for parity calc
* Empty -> Dirty - on compute_parity to satisfy write/sync request.
* Empty -> Clean - on compute_block when computing a block for failed drive
* Want -> Empty - on failed read
* Want -> Clean - on successful completion of read request
* Dirty -> Clean - on successful completion of write request
* Dirty -> Clean - on failed write
* Clean -> Dirty - on compute_parity to satisfy write/sync (RECONSTRUCT or RMW)
*
* The Want->Empty, Want->Clean, Dirty->Clean, transitions
* all happen in b_end_io at interrupt time.
* Each sets the Uptodate bit before releasing the Lock bit.
* This leaves one multi-stage transition:
* Want->Dirty->Clean
* This is safe because thinking that a Clean buffer is actually dirty
* will at worst delay some action, and the stripe will be scheduled
* for attention after the transition is complete.
*
* There is one possibility that is not covered by these states. That
* is if one drive has failed and there is a spare being rebuilt. We
* can't distinguish between a clean block that has been generated
* from parity calculations, and a clean block that has been
* successfully written to the spare ( or to parity when resyncing).
* To distinguish these states we have a stripe bit STRIPE_INSYNC that
* is set whenever a write is scheduled to the spare, or to the parity
* disc if there is no spare. A sync request clears this bit, and
* when we find it set with no buffers locked, we know the sync is
* complete.
*
* Buffers for the md device that arrive via make_request are attached
* to the appropriate stripe in one of two lists linked on b_reqnext.
* One list (bh_read) for read requests, one (bh_write) for write.
* There should never be more than one buffer on the two lists
* together, but we are not guaranteed of that so we allow for more.
*
* If a buffer is on the read list when the associated cache buffer is
* Uptodate, the data is copied into the read buffer and it's b_end_io
* routine is called. This may happen in the end_request routine only
* if the buffer has just successfully been read. end_request should
* remove the buffers from the list and then set the Uptodate bit on
* the buffer. Other threads may do this only if they first check
* that the Uptodate bit is set. Once they have checked that they may
* take buffers off the read queue.
*
* When a buffer on the write list is committed for write it is copied
* into the cache buffer, which is then marked dirty, and moved onto a
* third list, the written list (bh_written). Once both the parity
* block and the cached buffer are successfully written, any buffer on
* a written list can be returned with b_end_io.
*
* The write list and read list both act as fifos. The read list,
* write list and written list are protected by the device_lock.
* The device_lock is only for list manipulations and will only be
* held for a very short time. It can be claimed from interrupts.
*
*
* Stripes in the stripe cache can be on one of two lists (or on
* neither). The "inactive_list" contains stripes which are not
* currently being used for any request. They can freely be reused
* for another stripe. The "handle_list" contains stripes that need
* to be handled in some way. Both of these are fifo queues. Each
* stripe is also (potentially) linked to a hash bucket in the hash
* table so that it can be found by sector number. Stripes that are
* not hashed must be on the inactive_list, and will normally be at
* the front. All stripes start life this way.
*
* The inactive_list, handle_list and hash bucket lists are all protected by the
* device_lock.
* - stripes have a reference counter. If count==0, they are on a list.
* - If a stripe might need handling, STRIPE_HANDLE is set.
* - When refcount reaches zero, then if STRIPE_HANDLE it is put on
* handle_list else inactive_list
*
* This, combined with the fact that STRIPE_HANDLE is only ever
* cleared while a stripe has a non-zero count means that if the
* refcount is 0 and STRIPE_HANDLE is set, then it is on the
* handle_list and if recount is 0 and STRIPE_HANDLE is not set, then
* the stripe is on inactive_list.
*
* The possible transitions are:
* activate an unhashed/inactive stripe (get_active_stripe())
* lockdev check-hash unlink-stripe cnt++ clean-stripe hash-stripe unlockdev
* activate a hashed, possibly active stripe (get_active_stripe())
* lockdev check-hash if(!cnt++)unlink-stripe unlockdev
* attach a request to an active stripe (add_stripe_bh())
* lockdev attach-buffer unlockdev
* handle a stripe (handle_stripe())
* setSTRIPE_ACTIVE, clrSTRIPE_HANDLE ...
md: raid5_run_ops - run stripe operations outside sh->lock When the raid acceleration work was proposed, Neil laid out the following attack plan: 1/ move the xor and copy operations outside spin_lock(&sh->lock) 2/ find/implement an asynchronous offload api The raid5_run_ops routine uses the asynchronous offload api (async_tx) and the stripe_operations member of a stripe_head to carry out xor+copy operations asynchronously, outside the lock. To perform operations outside the lock a new set of state flags is needed to track new requests, in-flight requests, and completed requests. In this new model handle_stripe is tasked with scanning the stripe_head for work, updating the stripe_operations structure, and finally dropping the lock and calling raid5_run_ops for processing. The following flags outline the requests that handle_stripe can make of raid5_run_ops: STRIPE_OP_BIOFILL - copy data into request buffers to satisfy a read request STRIPE_OP_COMPUTE_BLK - generate a missing block in the cache from the other blocks STRIPE_OP_PREXOR - subtract existing data as part of the read-modify-write process STRIPE_OP_BIODRAIN - copy data out of request buffers to satisfy a write request STRIPE_OP_POSTXOR - recalculate parity for new data that has entered the cache STRIPE_OP_CHECK - verify that the parity is correct STRIPE_OP_IO - submit i/o to the member disks (note this was already performed outside the stripe lock, but it made sense to add it as an operation type The flow is: 1/ handle_stripe sets STRIPE_OP_* in sh->ops.pending 2/ raid5_run_ops reads sh->ops.pending, sets sh->ops.ack, and submits the operation to the async_tx api 3/ async_tx triggers the completion callback routine to set sh->ops.complete and release the stripe 4/ handle_stripe runs again to finish the operation and optionally submit new operations that were previously blocked Note this patch just defines raid5_run_ops, subsequent commits (one per major operation type) modify handle_stripe to take advantage of this routine. Changelog: * removed ops_complete_biodrain in favor of ops_complete_postxor and ops_complete_write. * removed the raid5_run_ops workqueue * call bi_end_io for reads in ops_complete_biofill, saves a call to handle_stripe * explicitly handle the 2-disk raid5 case (xor becomes memcpy), Neil Brown * fix race between async engines and bi_end_io call for reads, Neil Brown * remove unnecessary spin_lock from ops_complete_biofill * remove test_and_set/test_and_clear BUG_ONs, Neil Brown * remove explicit interrupt handling for channel switching, this feature was absorbed (i.e. it is now implicit) by the async_tx api * use return_io in ops_complete_biofill Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-By: NeilBrown <neilb@suse.de>
2007-01-03 04:52:30 +08:00
* (lockdev check-buffers unlockdev) ..
* change-state ..
* record io/ops needed clearSTRIPE_ACTIVE schedule io/ops
* release an active stripe (release_stripe())
* lockdev if (!--cnt) { if STRIPE_HANDLE, add to handle_list else add to inactive-list } unlockdev
*
* The refcount counts each thread that have activated the stripe,
* plus raid5d if it is handling it, plus one for each active request
md: raid5_run_ops - run stripe operations outside sh->lock When the raid acceleration work was proposed, Neil laid out the following attack plan: 1/ move the xor and copy operations outside spin_lock(&sh->lock) 2/ find/implement an asynchronous offload api The raid5_run_ops routine uses the asynchronous offload api (async_tx) and the stripe_operations member of a stripe_head to carry out xor+copy operations asynchronously, outside the lock. To perform operations outside the lock a new set of state flags is needed to track new requests, in-flight requests, and completed requests. In this new model handle_stripe is tasked with scanning the stripe_head for work, updating the stripe_operations structure, and finally dropping the lock and calling raid5_run_ops for processing. The following flags outline the requests that handle_stripe can make of raid5_run_ops: STRIPE_OP_BIOFILL - copy data into request buffers to satisfy a read request STRIPE_OP_COMPUTE_BLK - generate a missing block in the cache from the other blocks STRIPE_OP_PREXOR - subtract existing data as part of the read-modify-write process STRIPE_OP_BIODRAIN - copy data out of request buffers to satisfy a write request STRIPE_OP_POSTXOR - recalculate parity for new data that has entered the cache STRIPE_OP_CHECK - verify that the parity is correct STRIPE_OP_IO - submit i/o to the member disks (note this was already performed outside the stripe lock, but it made sense to add it as an operation type The flow is: 1/ handle_stripe sets STRIPE_OP_* in sh->ops.pending 2/ raid5_run_ops reads sh->ops.pending, sets sh->ops.ack, and submits the operation to the async_tx api 3/ async_tx triggers the completion callback routine to set sh->ops.complete and release the stripe 4/ handle_stripe runs again to finish the operation and optionally submit new operations that were previously blocked Note this patch just defines raid5_run_ops, subsequent commits (one per major operation type) modify handle_stripe to take advantage of this routine. Changelog: * removed ops_complete_biodrain in favor of ops_complete_postxor and ops_complete_write. * removed the raid5_run_ops workqueue * call bi_end_io for reads in ops_complete_biofill, saves a call to handle_stripe * explicitly handle the 2-disk raid5 case (xor becomes memcpy), Neil Brown * fix race between async engines and bi_end_io call for reads, Neil Brown * remove unnecessary spin_lock from ops_complete_biofill * remove test_and_set/test_and_clear BUG_ONs, Neil Brown * remove explicit interrupt handling for channel switching, this feature was absorbed (i.e. it is now implicit) by the async_tx api * use return_io in ops_complete_biofill Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-By: NeilBrown <neilb@suse.de>
2007-01-03 04:52:30 +08:00
* on a cached buffer, and plus one if the stripe is undergoing stripe
* operations.
*
* The stripe operations are:
md: raid5_run_ops - run stripe operations outside sh->lock When the raid acceleration work was proposed, Neil laid out the following attack plan: 1/ move the xor and copy operations outside spin_lock(&sh->lock) 2/ find/implement an asynchronous offload api The raid5_run_ops routine uses the asynchronous offload api (async_tx) and the stripe_operations member of a stripe_head to carry out xor+copy operations asynchronously, outside the lock. To perform operations outside the lock a new set of state flags is needed to track new requests, in-flight requests, and completed requests. In this new model handle_stripe is tasked with scanning the stripe_head for work, updating the stripe_operations structure, and finally dropping the lock and calling raid5_run_ops for processing. The following flags outline the requests that handle_stripe can make of raid5_run_ops: STRIPE_OP_BIOFILL - copy data into request buffers to satisfy a read request STRIPE_OP_COMPUTE_BLK - generate a missing block in the cache from the other blocks STRIPE_OP_PREXOR - subtract existing data as part of the read-modify-write process STRIPE_OP_BIODRAIN - copy data out of request buffers to satisfy a write request STRIPE_OP_POSTXOR - recalculate parity for new data that has entered the cache STRIPE_OP_CHECK - verify that the parity is correct STRIPE_OP_IO - submit i/o to the member disks (note this was already performed outside the stripe lock, but it made sense to add it as an operation type The flow is: 1/ handle_stripe sets STRIPE_OP_* in sh->ops.pending 2/ raid5_run_ops reads sh->ops.pending, sets sh->ops.ack, and submits the operation to the async_tx api 3/ async_tx triggers the completion callback routine to set sh->ops.complete and release the stripe 4/ handle_stripe runs again to finish the operation and optionally submit new operations that were previously blocked Note this patch just defines raid5_run_ops, subsequent commits (one per major operation type) modify handle_stripe to take advantage of this routine. Changelog: * removed ops_complete_biodrain in favor of ops_complete_postxor and ops_complete_write. * removed the raid5_run_ops workqueue * call bi_end_io for reads in ops_complete_biofill, saves a call to handle_stripe * explicitly handle the 2-disk raid5 case (xor becomes memcpy), Neil Brown * fix race between async engines and bi_end_io call for reads, Neil Brown * remove unnecessary spin_lock from ops_complete_biofill * remove test_and_set/test_and_clear BUG_ONs, Neil Brown * remove explicit interrupt handling for channel switching, this feature was absorbed (i.e. it is now implicit) by the async_tx api * use return_io in ops_complete_biofill Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-By: NeilBrown <neilb@suse.de>
2007-01-03 04:52:30 +08:00
* -copying data between the stripe cache and user application buffers
* -computing blocks to save a disk access, or to recover a missing block
* -updating the parity on a write operation (reconstruct write and
* read-modify-write)
* -checking parity correctness
* -running i/o to disk
* These operations are carried out by raid5_run_ops which uses the async_tx
* api to (optionally) offload operations to dedicated hardware engines.
* When requesting an operation handle_stripe sets the pending bit for the
* operation and increments the count. raid5_run_ops is then run whenever
* the count is non-zero.
* There are some critical dependencies between the operations that prevent some
* from being requested while another is in flight.
* 1/ Parity check operations destroy the in cache version of the parity block,
* so we prevent parity dependent operations like writes and compute_blocks
* from starting while a check is in progress. Some dma engines can perform
* the check without damaging the parity block, in these cases the parity
* block is re-marked up to date (assuming the check was successful) and is
* not re-read from disk.
* 2/ When a write operation is requested we immediately lock the affected
* blocks, and mark them as not up to date. This causes new read requests
* to be held off, as well as parity checks and compute block operations.
* 3/ Once a compute block operation has been requested handle_stripe treats
* that block as if it is up to date. raid5_run_ops guaruntees that any
* operation that is dependent on the compute block result is initiated after
* the compute block completes.
*/
/*
* Operations state - intermediate states that are visible outside of
* STRIPE_ACTIVE.
* In general _idle indicates nothing is running, _run indicates a data
* processing operation is active, and _result means the data processing result
* is stable and can be acted upon. For simple operations like biofill and
* compute that only have an _idle and _run state they are indicated with
* sh->state flags (STRIPE_BIOFILL_RUN and STRIPE_COMPUTE_RUN)
*/
/**
* enum check_states - handles syncing / repairing a stripe
* @check_state_idle - check operations are quiesced
* @check_state_run - check operation is running
* @check_state_result - set outside lock when check result is valid
* @check_state_compute_run - check failed and we are repairing
* @check_state_compute_result - set outside lock when compute result is valid
*/
enum check_states {
check_state_idle = 0,
check_state_run, /* xor parity check */
check_state_run_q, /* q-parity check */
check_state_run_pq, /* pq dual parity check */
check_state_check_result,
check_state_compute_run, /* parity repair */
check_state_compute_result,
};
/**
* enum reconstruct_states - handles writing or expanding a stripe
*/
enum reconstruct_states {
reconstruct_state_idle = 0,
reconstruct_state_prexor_drain_run, /* prexor-write */
reconstruct_state_drain_run, /* write */
reconstruct_state_run, /* expand */
reconstruct_state_prexor_drain_result,
reconstruct_state_drain_result,
reconstruct_state_result,
};
struct stripe_head {
struct hlist_node hash;
struct list_head lru; /* inactive_list or handle_list */
struct llist_node release_list;
struct r5conf *raid_conf;
short generation; /* increments with every
* reshape */
sector_t sector; /* sector of this row */
short pd_idx; /* parity disk index */
short qd_idx; /* 'Q' disk index for raid6 */
short ddf_layout;/* use DDF ordering to calculate Q */
raid5: relieve lock contention in get_active_stripe() get_active_stripe() is the last place we have lock contention. It has two paths. One is stripe isn't found and new stripe is allocated, the other is stripe is found. The first path basically calls __find_stripe and init_stripe. It accesses conf->generation, conf->previous_raid_disks, conf->raid_disks, conf->prev_chunk_sectors, conf->chunk_sectors, conf->max_degraded, conf->prev_algo, conf->algorithm, the stripe_hashtbl and inactive_list. Except stripe_hashtbl and inactive_list, other fields are changed very rarely. With this patch, we split inactive_list and add new hash locks. Each free stripe belongs to a specific inactive list. Which inactive list is determined by stripe's lock_hash. Note, even a stripe hasn't a sector assigned, it has a lock_hash assigned. Stripe's inactive list is protected by a hash lock, which is determined by it's lock_hash too. The lock_hash is derivied from current stripe_hashtbl hash, which guarantees any stripe_hashtbl list will be assigned to a specific lock_hash, so we can use new hash lock to protect stripe_hashtbl list too. The goal of the new hash locks introduced is we can only use the new locks in the first path of get_active_stripe(). Since we have several hash locks, lock contention is relieved significantly. The first path of get_active_stripe() accesses other fields, since they are changed rarely, changing them now need take conf->device_lock and all hash locks. For a slow path, this isn't a problem. If we need lock device_lock and hash lock, we always lock hash lock first. The tricky part is release_stripe and friends. We need take device_lock first. Neil's suggestion is we put inactive stripes to a temporary list and readd it to inactive_list after device_lock is released. In this way, we add stripes to temporary list with device_lock hold and remove stripes from the list with hash lock hold. So we don't allow concurrent access to the temporary list, which means we need allocate temporary list for all participants of release_stripe. One downside is free stripes are maintained in their inactive list, they can't across between the lists. By default, we have total 256 stripes and 8 lists, so each list will have 32 stripes. It's possible one list has free stripe but other list hasn't. The chance should be rare because stripes allocation are even distributed. And we can always allocate more stripes for cache, several mega bytes memory isn't a big deal. This completely removes the lock contention of the first path of get_active_stripe(). It slows down the second code path a little bit though because we now need takes two locks, but since the hash lock isn't contended, the overhead should be quite small (several atomic instructions). The second path of get_active_stripe() (basically sequential write or big request size randwrite) still has lock contentions. Signed-off-by: Shaohua Li <shli@fusionio.com> Signed-off-by: NeilBrown <neilb@suse.de>
2013-11-14 12:16:17 +08:00
short hash_lock_index;
unsigned long state; /* state flags */
atomic_t count; /* nr of active thread/requests */
int bm_seq; /* sequence number for bitmap flushes */
int disks; /* disks in stripe */
int overwrite_disks; /* total overwrite disks in stripe,
* this is only checked when stripe
* has STRIPE_BATCH_READY
*/
enum check_states check_state;
enum reconstruct_states reconstruct_state;
spinlock_t stripe_lock;
raid5: offload stripe handle to workqueue This is another attempt to create multiple threads to handle raid5 stripes. This time I use workqueue. raid5 handles request (especially write) in stripe unit. A stripe is page size aligned/long and acrosses all disks. Writing to any disk sector, raid5 runs a state machine for the corresponding stripe, which includes reading some disks of the stripe, calculating parity, and writing some disks of the stripe. The state machine is running in raid5d thread currently. Since there is only one thread, it doesn't scale well for high speed storage. An obvious solution is multi-threading. To get better performance, we have some requirements: a. locality. stripe corresponding to request submitted from one cpu is better handled in thread in local cpu or local node. local cpu is preferred but some times could be a bottleneck, for example, parity calculation is too heavy. local node running has wide adaptability. b. configurablity. Different setup of raid5 array might need diffent configuration. Especially the thread number. More threads don't always mean better performance because of lock contentions. My original implementation is creating some kernel threads. There are interfaces to control which cpu's stripe each thread should handle. And userspace can set affinity of the threads. This provides biggest flexibility and configurability. But it's hard to use and apparently a new thread pool implementation is disfavor. Recent workqueue improvement is quite promising. unbound workqueue will be bound to numa node. If WQ_SYSFS is set in workqueue, there are sysfs option to do affinity setting. For example, we can only include one HT sibling in affinity. Since work is non-reentrant by default, and we can control running thread number by limiting dispatched work_struct number. In this patch, I created several stripe worker group. A group is a numa node. stripes from cpus of one node will be added to a group list. Workqueue thread of one node will only handle stripes of worker group of the node. In this way, stripe handling has numa node locality. And as I said, we can control thread number by limiting dispatched work_struct number. The work_struct callback function handles several stripes in one run. A typical work queue usage is to run one unit in each work_struct. In raid5 case, the unit is a stripe. But we can't do that: a. Though handling a stripe doesn't need lock because of reference accounting and stripe isn't in any list, queuing a work_struct for each stripe will make workqueue lock contended very heavily. b. blk_start_plug()/blk_finish_plug() should surround stripe handle, as we might dispatch request. If each work_struct only handles one stripe, such block plug is meaningless. This implementation can't do very fine grained configuration. But the numa binding is most popular usage model, should be enough for most workloads. Note: since we have only one stripe queue, switching to multi-thread might decrease request size dispatching down to low level layer. The impact depends on thread number, raid configuration and workload. So multi-thread raid5 might not be proper for all setups. Changes V1 -> V2: 1. remove WQ_NON_REENTRANT 2. disabling multi-threading by default 3. Add more descriptions in changelog Signed-off-by: Shaohua Li <shli@fusionio.com> Signed-off-by: NeilBrown <neilb@suse.de>
2013-08-28 14:30:16 +08:00
int cpu;
struct r5worker_group *group;
RAID5: batch adjacent full stripe write stripe cache is 4k size. Even adjacent full stripe writes are handled in 4k unit. Idealy we should use big size for adjacent full stripe writes. Bigger stripe cache size means less stripes runing in the state machine so can reduce cpu overhead. And also bigger size can cause bigger IO size dispatched to under layer disks. With below patch, we will automatically batch adjacent full stripe write together. Such stripes will be added to the batch list. Only the first stripe of the list will be put to handle_list and so run handle_stripe(). Some steps of handle_stripe() are extended to cover all stripes of the list, including ops_run_io, ops_run_biodrain and so on. With this patch, we have less stripes running in handle_stripe() and we send IO of whole stripe list together to increase IO size. Stripes added to a batch list have some limitations. A batch list can only include full stripe write and can't cross chunk boundary to make sure stripes have the same parity disks. Stripes in a batch list must be in the same state (no written, toread and so on). If a stripe is in a batch list, all new read/write to add_stripe_bio will be blocked to overlap conflict till the batch list is handled. The limitations will make sure stripes in a batch list be in exactly the same state in the life circly. I did test running 160k randwrite in a RAID5 array with 32k chunk size and 6 PCIe SSD. This patch improves around 30% performance and IO size to under layer disk is exactly 32k. I also run a 4k randwrite test in the same array to make sure the performance isn't changed with the patch. Signed-off-by: Shaohua Li <shli@fusionio.com> Signed-off-by: NeilBrown <neilb@suse.de>
2014-12-15 09:57:03 +08:00
struct stripe_head *batch_head; /* protected by stripe lock */
spinlock_t batch_lock; /* only header's lock is useful */
struct list_head batch_list; /* protected by head's batch lock*/
raid5: add basic stripe log This introduces a simple log for raid5. Data/parity writing to raid array first writes to the log, then write to raid array disks. If crash happens, we can recovery data from the log. This can speed up raid resync and fix write hole issue. The log structure is pretty simple. Data/meta data is stored in block unit, which is 4k generally. It has only one type of meta data block. The meta data block can track 3 types of data, stripe data, stripe parity and flush block. MD superblock will point to the last valid meta data block. Each meta data block has checksum/seq number, so recovery can scan the log correctly. We store a checksum of stripe data/parity to the metadata block, so meta data and stripe data/parity can be written to log disk together. otherwise, meta data write must wait till stripe data/parity is finished. For stripe data, meta data block will record stripe data sector and size. Currently the size is always 4k. This meta data record can be made simpler if we just fix write hole (eg, we can record data of a stripe's different disks together), but this format can be extended to support caching in the future, which must record data address/size. For stripe parity, meta data block will record stripe sector. It's size should be 4k (for raid5) or 8k (for raid6). We always store p parity first. This format should work for caching too. flush block indicates a stripe is in raid array disks. Fixing write hole doesn't need this type of meta data, it's for caching extension. Signed-off-by: Shaohua Li <shli@fb.com> Signed-off-by: NeilBrown <neilb@suse.com>
2015-08-14 05:31:59 +08:00
struct r5l_io_unit *log_io;
struct list_head log_list;
/**
* struct stripe_operations
md: raid5_run_ops - run stripe operations outside sh->lock When the raid acceleration work was proposed, Neil laid out the following attack plan: 1/ move the xor and copy operations outside spin_lock(&sh->lock) 2/ find/implement an asynchronous offload api The raid5_run_ops routine uses the asynchronous offload api (async_tx) and the stripe_operations member of a stripe_head to carry out xor+copy operations asynchronously, outside the lock. To perform operations outside the lock a new set of state flags is needed to track new requests, in-flight requests, and completed requests. In this new model handle_stripe is tasked with scanning the stripe_head for work, updating the stripe_operations structure, and finally dropping the lock and calling raid5_run_ops for processing. The following flags outline the requests that handle_stripe can make of raid5_run_ops: STRIPE_OP_BIOFILL - copy data into request buffers to satisfy a read request STRIPE_OP_COMPUTE_BLK - generate a missing block in the cache from the other blocks STRIPE_OP_PREXOR - subtract existing data as part of the read-modify-write process STRIPE_OP_BIODRAIN - copy data out of request buffers to satisfy a write request STRIPE_OP_POSTXOR - recalculate parity for new data that has entered the cache STRIPE_OP_CHECK - verify that the parity is correct STRIPE_OP_IO - submit i/o to the member disks (note this was already performed outside the stripe lock, but it made sense to add it as an operation type The flow is: 1/ handle_stripe sets STRIPE_OP_* in sh->ops.pending 2/ raid5_run_ops reads sh->ops.pending, sets sh->ops.ack, and submits the operation to the async_tx api 3/ async_tx triggers the completion callback routine to set sh->ops.complete and release the stripe 4/ handle_stripe runs again to finish the operation and optionally submit new operations that were previously blocked Note this patch just defines raid5_run_ops, subsequent commits (one per major operation type) modify handle_stripe to take advantage of this routine. Changelog: * removed ops_complete_biodrain in favor of ops_complete_postxor and ops_complete_write. * removed the raid5_run_ops workqueue * call bi_end_io for reads in ops_complete_biofill, saves a call to handle_stripe * explicitly handle the 2-disk raid5 case (xor becomes memcpy), Neil Brown * fix race between async engines and bi_end_io call for reads, Neil Brown * remove unnecessary spin_lock from ops_complete_biofill * remove test_and_set/test_and_clear BUG_ONs, Neil Brown * remove explicit interrupt handling for channel switching, this feature was absorbed (i.e. it is now implicit) by the async_tx api * use return_io in ops_complete_biofill Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-By: NeilBrown <neilb@suse.de>
2007-01-03 04:52:30 +08:00
* @target - STRIPE_OP_COMPUTE_BLK target
* @target2 - 2nd compute target in the raid6 case
* @zero_sum_result - P and Q verification flags
* @request - async service request flags for raid_run_ops
md: raid5_run_ops - run stripe operations outside sh->lock When the raid acceleration work was proposed, Neil laid out the following attack plan: 1/ move the xor and copy operations outside spin_lock(&sh->lock) 2/ find/implement an asynchronous offload api The raid5_run_ops routine uses the asynchronous offload api (async_tx) and the stripe_operations member of a stripe_head to carry out xor+copy operations asynchronously, outside the lock. To perform operations outside the lock a new set of state flags is needed to track new requests, in-flight requests, and completed requests. In this new model handle_stripe is tasked with scanning the stripe_head for work, updating the stripe_operations structure, and finally dropping the lock and calling raid5_run_ops for processing. The following flags outline the requests that handle_stripe can make of raid5_run_ops: STRIPE_OP_BIOFILL - copy data into request buffers to satisfy a read request STRIPE_OP_COMPUTE_BLK - generate a missing block in the cache from the other blocks STRIPE_OP_PREXOR - subtract existing data as part of the read-modify-write process STRIPE_OP_BIODRAIN - copy data out of request buffers to satisfy a write request STRIPE_OP_POSTXOR - recalculate parity for new data that has entered the cache STRIPE_OP_CHECK - verify that the parity is correct STRIPE_OP_IO - submit i/o to the member disks (note this was already performed outside the stripe lock, but it made sense to add it as an operation type The flow is: 1/ handle_stripe sets STRIPE_OP_* in sh->ops.pending 2/ raid5_run_ops reads sh->ops.pending, sets sh->ops.ack, and submits the operation to the async_tx api 3/ async_tx triggers the completion callback routine to set sh->ops.complete and release the stripe 4/ handle_stripe runs again to finish the operation and optionally submit new operations that were previously blocked Note this patch just defines raid5_run_ops, subsequent commits (one per major operation type) modify handle_stripe to take advantage of this routine. Changelog: * removed ops_complete_biodrain in favor of ops_complete_postxor and ops_complete_write. * removed the raid5_run_ops workqueue * call bi_end_io for reads in ops_complete_biofill, saves a call to handle_stripe * explicitly handle the 2-disk raid5 case (xor becomes memcpy), Neil Brown * fix race between async engines and bi_end_io call for reads, Neil Brown * remove unnecessary spin_lock from ops_complete_biofill * remove test_and_set/test_and_clear BUG_ONs, Neil Brown * remove explicit interrupt handling for channel switching, this feature was absorbed (i.e. it is now implicit) by the async_tx api * use return_io in ops_complete_biofill Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-By: NeilBrown <neilb@suse.de>
2007-01-03 04:52:30 +08:00
*/
struct stripe_operations {
int target, target2;
enum sum_check_flags zero_sum_result;
md: raid5_run_ops - run stripe operations outside sh->lock When the raid acceleration work was proposed, Neil laid out the following attack plan: 1/ move the xor and copy operations outside spin_lock(&sh->lock) 2/ find/implement an asynchronous offload api The raid5_run_ops routine uses the asynchronous offload api (async_tx) and the stripe_operations member of a stripe_head to carry out xor+copy operations asynchronously, outside the lock. To perform operations outside the lock a new set of state flags is needed to track new requests, in-flight requests, and completed requests. In this new model handle_stripe is tasked with scanning the stripe_head for work, updating the stripe_operations structure, and finally dropping the lock and calling raid5_run_ops for processing. The following flags outline the requests that handle_stripe can make of raid5_run_ops: STRIPE_OP_BIOFILL - copy data into request buffers to satisfy a read request STRIPE_OP_COMPUTE_BLK - generate a missing block in the cache from the other blocks STRIPE_OP_PREXOR - subtract existing data as part of the read-modify-write process STRIPE_OP_BIODRAIN - copy data out of request buffers to satisfy a write request STRIPE_OP_POSTXOR - recalculate parity for new data that has entered the cache STRIPE_OP_CHECK - verify that the parity is correct STRIPE_OP_IO - submit i/o to the member disks (note this was already performed outside the stripe lock, but it made sense to add it as an operation type The flow is: 1/ handle_stripe sets STRIPE_OP_* in sh->ops.pending 2/ raid5_run_ops reads sh->ops.pending, sets sh->ops.ack, and submits the operation to the async_tx api 3/ async_tx triggers the completion callback routine to set sh->ops.complete and release the stripe 4/ handle_stripe runs again to finish the operation and optionally submit new operations that were previously blocked Note this patch just defines raid5_run_ops, subsequent commits (one per major operation type) modify handle_stripe to take advantage of this routine. Changelog: * removed ops_complete_biodrain in favor of ops_complete_postxor and ops_complete_write. * removed the raid5_run_ops workqueue * call bi_end_io for reads in ops_complete_biofill, saves a call to handle_stripe * explicitly handle the 2-disk raid5 case (xor becomes memcpy), Neil Brown * fix race between async engines and bi_end_io call for reads, Neil Brown * remove unnecessary spin_lock from ops_complete_biofill * remove test_and_set/test_and_clear BUG_ONs, Neil Brown * remove explicit interrupt handling for channel switching, this feature was absorbed (i.e. it is now implicit) by the async_tx api * use return_io in ops_complete_biofill Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-By: NeilBrown <neilb@suse.de>
2007-01-03 04:52:30 +08:00
} ops;
struct r5dev {
/* rreq and rvec are used for the replacement device when
* writing data to both devices.
*/
struct bio req, rreq;
struct bio_vec vec, rvec;
struct page *page, *orig_page;
md: raid5_run_ops - run stripe operations outside sh->lock When the raid acceleration work was proposed, Neil laid out the following attack plan: 1/ move the xor and copy operations outside spin_lock(&sh->lock) 2/ find/implement an asynchronous offload api The raid5_run_ops routine uses the asynchronous offload api (async_tx) and the stripe_operations member of a stripe_head to carry out xor+copy operations asynchronously, outside the lock. To perform operations outside the lock a new set of state flags is needed to track new requests, in-flight requests, and completed requests. In this new model handle_stripe is tasked with scanning the stripe_head for work, updating the stripe_operations structure, and finally dropping the lock and calling raid5_run_ops for processing. The following flags outline the requests that handle_stripe can make of raid5_run_ops: STRIPE_OP_BIOFILL - copy data into request buffers to satisfy a read request STRIPE_OP_COMPUTE_BLK - generate a missing block in the cache from the other blocks STRIPE_OP_PREXOR - subtract existing data as part of the read-modify-write process STRIPE_OP_BIODRAIN - copy data out of request buffers to satisfy a write request STRIPE_OP_POSTXOR - recalculate parity for new data that has entered the cache STRIPE_OP_CHECK - verify that the parity is correct STRIPE_OP_IO - submit i/o to the member disks (note this was already performed outside the stripe lock, but it made sense to add it as an operation type The flow is: 1/ handle_stripe sets STRIPE_OP_* in sh->ops.pending 2/ raid5_run_ops reads sh->ops.pending, sets sh->ops.ack, and submits the operation to the async_tx api 3/ async_tx triggers the completion callback routine to set sh->ops.complete and release the stripe 4/ handle_stripe runs again to finish the operation and optionally submit new operations that were previously blocked Note this patch just defines raid5_run_ops, subsequent commits (one per major operation type) modify handle_stripe to take advantage of this routine. Changelog: * removed ops_complete_biodrain in favor of ops_complete_postxor and ops_complete_write. * removed the raid5_run_ops workqueue * call bi_end_io for reads in ops_complete_biofill, saves a call to handle_stripe * explicitly handle the 2-disk raid5 case (xor becomes memcpy), Neil Brown * fix race between async engines and bi_end_io call for reads, Neil Brown * remove unnecessary spin_lock from ops_complete_biofill * remove test_and_set/test_and_clear BUG_ONs, Neil Brown * remove explicit interrupt handling for channel switching, this feature was absorbed (i.e. it is now implicit) by the async_tx api * use return_io in ops_complete_biofill Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-By: NeilBrown <neilb@suse.de>
2007-01-03 04:52:30 +08:00
struct bio *toread, *read, *towrite, *written;
sector_t sector; /* sector of this page */
unsigned long flags;
raid5: add basic stripe log This introduces a simple log for raid5. Data/parity writing to raid array first writes to the log, then write to raid array disks. If crash happens, we can recovery data from the log. This can speed up raid resync and fix write hole issue. The log structure is pretty simple. Data/meta data is stored in block unit, which is 4k generally. It has only one type of meta data block. The meta data block can track 3 types of data, stripe data, stripe parity and flush block. MD superblock will point to the last valid meta data block. Each meta data block has checksum/seq number, so recovery can scan the log correctly. We store a checksum of stripe data/parity to the metadata block, so meta data and stripe data/parity can be written to log disk together. otherwise, meta data write must wait till stripe data/parity is finished. For stripe data, meta data block will record stripe data sector and size. Currently the size is always 4k. This meta data record can be made simpler if we just fix write hole (eg, we can record data of a stripe's different disks together), but this format can be extended to support caching in the future, which must record data address/size. For stripe parity, meta data block will record stripe sector. It's size should be 4k (for raid5) or 8k (for raid6). We always store p parity first. This format should work for caching too. flush block indicates a stripe is in raid array disks. Fixing write hole doesn't need this type of meta data, it's for caching extension. Signed-off-by: Shaohua Li <shli@fb.com> Signed-off-by: NeilBrown <neilb@suse.com>
2015-08-14 05:31:59 +08:00
u32 log_checksum;
} dev[1]; /* allocated with extra space depending of RAID geometry */
};
/* stripe_head_state - collects and tracks the dynamic state of a stripe_head
* for handle_stripe.
*/
struct stripe_head_state {
/* 'syncing' means that we need to read all devices, either
* to check/correct parity, or to reconstruct a missing device.
* 'replacing' means we are replacing one or more drives and
* the source is valid at this point so we don't need to
* read all devices, just the replacement targets.
*/
int syncing, expanding, expanded, replacing;
int locked, uptodate, to_read, to_write, failed, written;
int to_fill, compute, req_compute, non_overwrite;
int failed_num[2];
int p_failed, q_failed;
int dec_preread_active;
unsigned long ops_request;
struct bio_list return_bi;
struct md_rdev *blocked_rdev;
int handle_bad_blocks;
int log_failed;
};
/* Flags for struct r5dev.flags */
enum r5dev_flags {
R5_UPTODATE, /* page contains current data */
R5_LOCKED, /* IO has been submitted on "req" */
R5_DOUBLE_LOCKED,/* Cannot clear R5_LOCKED until 2 writes complete */
R5_OVERWRITE, /* towrite covers whole page */
/* and some that are internal to handle_stripe */
R5_Insync, /* rdev && rdev->in_sync at start */
R5_Wantread, /* want to schedule a read */
R5_Wantwrite,
R5_Overlap, /* There is a pending overlapping request
* on this block */
R5_ReadNoMerge, /* prevent bio from merging in block-layer */
R5_ReadError, /* seen a read error here recently */
R5_ReWrite, /* have tried to over-write the readerror */
R5_Expanded, /* This block now has post-expand data */
R5_Wantcompute, /* compute_block in progress treat as
* uptodate
*/
R5_Wantfill, /* dev->toread contains a bio that needs
* filling
*/
R5_Wantdrain, /* dev->towrite needs to be drained */
R5_WantFUA, /* Write should be FUA */
R5_SyncIO, /* The IO is sync */
R5_WriteError, /* got a write error - need to record it */
R5_MadeGood, /* A bad block has been fixed by writing to it */
R5_ReadRepl, /* Will/did read from replacement rather than orig */
R5_MadeGoodRepl,/* A bad block on the replacement device has been
* fixed by writing to it */
R5_NeedReplace, /* This device has a replacement which is not
* up-to-date at this stripe. */
R5_WantReplace, /* We need to update the replacement, we have read
* data in, and now is a good time to write it out.
*/
MD: raid5 trim support Discard for raid4/5/6 has limitation. If discard request size is small, we do discard for one disk, but we need calculate parity and write parity disk. To correctly calculate parity, zero_after_discard must be guaranteed. Even it's true, we need do discard for one disk but write another disks, which makes the parity disks wear out fast. This doesn't make sense. So an efficient discard for raid4/5/6 should discard all data disks and parity disks, which requires the write pattern to be (A, A+chunk_size, A+chunk_size*2...). If A's size is smaller than chunk_size, such pattern is almost impossible in practice. So in this patch, I only handle the case that A's size equals to chunk_size. That is discard request should be aligned to stripe size and its size is multiple of stripe size. Since we can only handle request with specific alignment and size (or part of the request fitting stripes), we can't guarantee zero_after_discard even zero_after_discard is true in low level drives. The block layer doesn't send down correctly aligned requests even correct discard alignment is set, so I must filter out. For raid4/5/6 parity calculation, if data is 0, parity is 0. So if zero_after_discard is true for all disks, data is consistent after discard. Otherwise, data might be lost. Let's consider a scenario: discard a stripe, write data to one disk and write parity disk. The stripe could be still inconsistent till then depending on using data from other data disks or parity disks to calculate new parity. If the disk is broken, we can't restore it. So in this patch, we only enable discard support if all disks have zero_after_discard. If discard fails in one disk, we face the similar inconsistent issue above. The patch will make discard follow the same path as normal write request. If discard fails, a resync will be scheduled to make the data consistent. This isn't good to have extra writes, but data consistency is important. If a subsequent read/write request hits raid5 cache of a discarded stripe, the discarded dev page should have zero filled, so the data is consistent. This patch will always zero dev page for discarded request stripe. This isn't optimal because discard request doesn't need such payload. Next patch will avoid it. Signed-off-by: Shaohua Li <shli@fusionio.com> Signed-off-by: NeilBrown <neilb@suse.de>
2012-10-11 10:49:05 +08:00
R5_Discard, /* Discard the stripe */
R5_SkipCopy, /* Don't copy data from bio to stripe cache */
};
/*
* Stripe state
*/
enum {
STRIPE_ACTIVE,
STRIPE_HANDLE,
STRIPE_SYNC_REQUESTED,
STRIPE_SYNCING,
STRIPE_INSYNC,
md/raid5: fix interaction of 'replace' and 'recovery'. If a device in a RAID4/5/6 is being replaced while another is being recovered, then the writes to the replacement device currently don't happen, resulting in corruption when the replacement completes and the new drive takes over. This is because the replacement writes are only triggered when 's.replacing' is set and not when the similar 's.sync' is set (which is the case during resync and recovery - it means all devices need to be read). So schedule those writes when s.replacing is set as well. In this case we cannot use "STRIPE_INSYNC" to record that the replacement has happened as that is needed for recording that any parity calculation is complete. So introduce STRIPE_REPLACED to record if the replacement has happened. For safety we should also check that STRIPE_COMPUTE_RUN is not set. This has a similar effect to the "s.locked == 0" test. The latter ensure that now IO has been flagged but not started. The former checks if any parity calculation has been flagged by not started. We must wait for both of these to complete before triggering the 'replace'. Add a similar test to the subsequent check for "are we finished yet". This possibly isn't needed (is subsumed in the STRIPE_INSYNC test), but it makes it more obvious that the REPLACE will happen before we think we are finished. Finally if a NeedReplace device is not UPTODATE then that is an error. We really must trigger a warning. This bug was introduced in commit 9a3e1101b827a59ac9036a672f5fa8d5279d0fe2 (md/raid5: detect and handle replacements during recovery.) which introduced replacement for raid5. That was in 3.3-rc3, so any stable kernel since then would benefit from this fix. Cc: stable@vger.kernel.org (3.3+) Reported-by: qindehua <13691222965@163.com> Tested-by: qindehua <qindehua@163.com> Signed-off-by: NeilBrown <neilb@suse.de>
2013-07-22 10:57:21 +08:00
STRIPE_REPLACED,
STRIPE_PREREAD_ACTIVE,
STRIPE_DELAYED,
STRIPE_DEGRADED,
STRIPE_BIT_DELAY,
STRIPE_EXPANDING,
STRIPE_EXPAND_SOURCE,
STRIPE_EXPAND_READY,
STRIPE_IO_STARTED, /* do not count towards 'bypass_count' */
STRIPE_FULL_WRITE, /* all blocks are set to be overwritten */
STRIPE_BIOFILL_RUN,
STRIPE_COMPUTE_RUN,
STRIPE_OPS_REQ_PENDING,
STRIPE_ON_UNPLUG_LIST,
md/raid5: ensure sync and DISCARD don't happen at the same time. A number of problems can occur due to races between resync/recovery and discard. - if sync_request calls handle_stripe() while a discard is happening on the stripe, it might call handle_stripe_clean_event before all of the individual discard requests have completed (so some devices are still locked, but not all). Since commit ca64cae96037de16e4af92678814f5d4bf0c1c65 md/raid5: Make sure we clear R5_Discard when discard is finished. this will cause R5_Discard to be cleared for the parity device, so handle_stripe_clean_event() will not be called when the other devices do become unlocked, so their ->written will not be cleared. This ultimately leads to a WARN_ON in init_stripe and a lock-up. - If handle_stripe_clean_event() does clear R5_UPTODATE at an awkward time for resync, it can lead to s->uptodate being less than disks in handle_parity_checks5(), which triggers a BUG (because it is one). So: - keep R5_Discard on the parity device until all other devices have completed their discard request - make sure we don't try to have a 'discard' and a 'sync' action at the same time. This involves a new stripe flag to we know when a 'discard' is happening, and the use of R5_Overlap on the parity disk so when a discard is wanted while a sync is active, so we know to wake up the discard at the appropriate time. Discard support for RAID5 was added in 3.7, so this is suitable for any -stable kernel since 3.7. Cc: stable@vger.kernel.org (v3.7+) Reported-by: Jes Sorensen <Jes.Sorensen@redhat.com> Tested-by: Jes Sorensen <Jes.Sorensen@redhat.com> Signed-off-by: NeilBrown <neilb@suse.de>
2013-03-12 09:18:06 +08:00
STRIPE_DISCARD,
STRIPE_ON_RELEASE_LIST,
STRIPE_BATCH_READY,
STRIPE_BATCH_ERR,
STRIPE_BITMAP_PENDING, /* Being added to bitmap, don't add
* to batch yet.
*/
STRIPE_LOG_TRAPPED, /* trapped into log */
};
#define STRIPE_EXPAND_SYNC_FLAGS \
((1 << STRIPE_EXPAND_SOURCE) |\
(1 << STRIPE_EXPAND_READY) |\
(1 << STRIPE_EXPANDING) |\
(1 << STRIPE_SYNC_REQUESTED))
md: raid5_run_ops - run stripe operations outside sh->lock When the raid acceleration work was proposed, Neil laid out the following attack plan: 1/ move the xor and copy operations outside spin_lock(&sh->lock) 2/ find/implement an asynchronous offload api The raid5_run_ops routine uses the asynchronous offload api (async_tx) and the stripe_operations member of a stripe_head to carry out xor+copy operations asynchronously, outside the lock. To perform operations outside the lock a new set of state flags is needed to track new requests, in-flight requests, and completed requests. In this new model handle_stripe is tasked with scanning the stripe_head for work, updating the stripe_operations structure, and finally dropping the lock and calling raid5_run_ops for processing. The following flags outline the requests that handle_stripe can make of raid5_run_ops: STRIPE_OP_BIOFILL - copy data into request buffers to satisfy a read request STRIPE_OP_COMPUTE_BLK - generate a missing block in the cache from the other blocks STRIPE_OP_PREXOR - subtract existing data as part of the read-modify-write process STRIPE_OP_BIODRAIN - copy data out of request buffers to satisfy a write request STRIPE_OP_POSTXOR - recalculate parity for new data that has entered the cache STRIPE_OP_CHECK - verify that the parity is correct STRIPE_OP_IO - submit i/o to the member disks (note this was already performed outside the stripe lock, but it made sense to add it as an operation type The flow is: 1/ handle_stripe sets STRIPE_OP_* in sh->ops.pending 2/ raid5_run_ops reads sh->ops.pending, sets sh->ops.ack, and submits the operation to the async_tx api 3/ async_tx triggers the completion callback routine to set sh->ops.complete and release the stripe 4/ handle_stripe runs again to finish the operation and optionally submit new operations that were previously blocked Note this patch just defines raid5_run_ops, subsequent commits (one per major operation type) modify handle_stripe to take advantage of this routine. Changelog: * removed ops_complete_biodrain in favor of ops_complete_postxor and ops_complete_write. * removed the raid5_run_ops workqueue * call bi_end_io for reads in ops_complete_biofill, saves a call to handle_stripe * explicitly handle the 2-disk raid5 case (xor becomes memcpy), Neil Brown * fix race between async engines and bi_end_io call for reads, Neil Brown * remove unnecessary spin_lock from ops_complete_biofill * remove test_and_set/test_and_clear BUG_ONs, Neil Brown * remove explicit interrupt handling for channel switching, this feature was absorbed (i.e. it is now implicit) by the async_tx api * use return_io in ops_complete_biofill Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-By: NeilBrown <neilb@suse.de>
2007-01-03 04:52:30 +08:00
/*
* Operation request flags
md: raid5_run_ops - run stripe operations outside sh->lock When the raid acceleration work was proposed, Neil laid out the following attack plan: 1/ move the xor and copy operations outside spin_lock(&sh->lock) 2/ find/implement an asynchronous offload api The raid5_run_ops routine uses the asynchronous offload api (async_tx) and the stripe_operations member of a stripe_head to carry out xor+copy operations asynchronously, outside the lock. To perform operations outside the lock a new set of state flags is needed to track new requests, in-flight requests, and completed requests. In this new model handle_stripe is tasked with scanning the stripe_head for work, updating the stripe_operations structure, and finally dropping the lock and calling raid5_run_ops for processing. The following flags outline the requests that handle_stripe can make of raid5_run_ops: STRIPE_OP_BIOFILL - copy data into request buffers to satisfy a read request STRIPE_OP_COMPUTE_BLK - generate a missing block in the cache from the other blocks STRIPE_OP_PREXOR - subtract existing data as part of the read-modify-write process STRIPE_OP_BIODRAIN - copy data out of request buffers to satisfy a write request STRIPE_OP_POSTXOR - recalculate parity for new data that has entered the cache STRIPE_OP_CHECK - verify that the parity is correct STRIPE_OP_IO - submit i/o to the member disks (note this was already performed outside the stripe lock, but it made sense to add it as an operation type The flow is: 1/ handle_stripe sets STRIPE_OP_* in sh->ops.pending 2/ raid5_run_ops reads sh->ops.pending, sets sh->ops.ack, and submits the operation to the async_tx api 3/ async_tx triggers the completion callback routine to set sh->ops.complete and release the stripe 4/ handle_stripe runs again to finish the operation and optionally submit new operations that were previously blocked Note this patch just defines raid5_run_ops, subsequent commits (one per major operation type) modify handle_stripe to take advantage of this routine. Changelog: * removed ops_complete_biodrain in favor of ops_complete_postxor and ops_complete_write. * removed the raid5_run_ops workqueue * call bi_end_io for reads in ops_complete_biofill, saves a call to handle_stripe * explicitly handle the 2-disk raid5 case (xor becomes memcpy), Neil Brown * fix race between async engines and bi_end_io call for reads, Neil Brown * remove unnecessary spin_lock from ops_complete_biofill * remove test_and_set/test_and_clear BUG_ONs, Neil Brown * remove explicit interrupt handling for channel switching, this feature was absorbed (i.e. it is now implicit) by the async_tx api * use return_io in ops_complete_biofill Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-By: NeilBrown <neilb@suse.de>
2007-01-03 04:52:30 +08:00
*/
enum {
STRIPE_OP_BIOFILL,
STRIPE_OP_COMPUTE_BLK,
STRIPE_OP_PREXOR,
STRIPE_OP_BIODRAIN,
STRIPE_OP_RECONSTRUCT,
STRIPE_OP_CHECK,
};
md/raid5: activate raid6 rmw feature Glue it altogehter. The raid6 rmw path should work the same as the already existing raid5 logic. So emulate the prexor handling/flags and split functions as needed. 1) Enable xor_syndrome() in the async layer. 2) Split ops_run_prexor() into RAID4/5 and RAID6 logic. Xor the syndrome at the start of a rmw run as we did it before for the single parity. 3) Take care of rmw run in ops_run_reconstruct6(). Again process only the changed pages to get syndrome back into sync. 4) Enhance set_syndrome_sources() to fill NULL pages if we are in a rmw run. The lower layers will calculate start & end pages from that and call the xor_syndrome() correspondingly. 5) Adapt the several places where we ignored Q handling up to now. Performance numbers for a single E5630 system with a mix of 10 7200k desktop/server disks. 300 seconds random write with 8 threads onto a 3,2TB (10*400GB) RAID6 64K chunk without spare (group_thread_cnt=4) bsize rmw_level=1 rmw_level=0 rmw_level=1 rmw_level=0 skip_copy=1 skip_copy=1 skip_copy=0 skip_copy=0 4K 115 KB/s 141 KB/s 165 KB/s 140 KB/s 8K 225 KB/s 275 KB/s 324 KB/s 274 KB/s 16K 434 KB/s 536 KB/s 640 KB/s 534 KB/s 32K 751 KB/s 1,051 KB/s 1,234 KB/s 1,045 KB/s 64K 1,339 KB/s 1,958 KB/s 2,282 KB/s 1,962 KB/s 128K 2,673 KB/s 3,862 KB/s 4,113 KB/s 3,898 KB/s 256K 7,685 KB/s 7,539 KB/s 7,557 KB/s 7,638 KB/s 512K 19,556 KB/s 19,558 KB/s 19,652 KB/s 19,688 Kb/s Signed-off-by: Markus Stockhausen <stockhausen@collogia.de> Signed-off-by: NeilBrown <neilb@suse.de>
2014-12-15 09:57:05 +08:00
/*
* RAID parity calculation preferences
*/
enum {
PARITY_DISABLE_RMW = 0,
PARITY_ENABLE_RMW,
PARITY_PREFER_RMW,
md/raid5: activate raid6 rmw feature Glue it altogehter. The raid6 rmw path should work the same as the already existing raid5 logic. So emulate the prexor handling/flags and split functions as needed. 1) Enable xor_syndrome() in the async layer. 2) Split ops_run_prexor() into RAID4/5 and RAID6 logic. Xor the syndrome at the start of a rmw run as we did it before for the single parity. 3) Take care of rmw run in ops_run_reconstruct6(). Again process only the changed pages to get syndrome back into sync. 4) Enhance set_syndrome_sources() to fill NULL pages if we are in a rmw run. The lower layers will calculate start & end pages from that and call the xor_syndrome() correspondingly. 5) Adapt the several places where we ignored Q handling up to now. Performance numbers for a single E5630 system with a mix of 10 7200k desktop/server disks. 300 seconds random write with 8 threads onto a 3,2TB (10*400GB) RAID6 64K chunk without spare (group_thread_cnt=4) bsize rmw_level=1 rmw_level=0 rmw_level=1 rmw_level=0 skip_copy=1 skip_copy=1 skip_copy=0 skip_copy=0 4K 115 KB/s 141 KB/s 165 KB/s 140 KB/s 8K 225 KB/s 275 KB/s 324 KB/s 274 KB/s 16K 434 KB/s 536 KB/s 640 KB/s 534 KB/s 32K 751 KB/s 1,051 KB/s 1,234 KB/s 1,045 KB/s 64K 1,339 KB/s 1,958 KB/s 2,282 KB/s 1,962 KB/s 128K 2,673 KB/s 3,862 KB/s 4,113 KB/s 3,898 KB/s 256K 7,685 KB/s 7,539 KB/s 7,557 KB/s 7,638 KB/s 512K 19,556 KB/s 19,558 KB/s 19,652 KB/s 19,688 Kb/s Signed-off-by: Markus Stockhausen <stockhausen@collogia.de> Signed-off-by: NeilBrown <neilb@suse.de>
2014-12-15 09:57:05 +08:00
};
/*
* Pages requested from set_syndrome_sources()
*/
enum {
SYNDROME_SRC_ALL,
SYNDROME_SRC_WANT_DRAIN,
SYNDROME_SRC_WRITTEN,
};
/*
* Plugging:
*
* To improve write throughput, we need to delay the handling of some
* stripes until there has been a chance that several write requests
* for the one stripe have all been collected.
* In particular, any write request that would require pre-reading
* is put on a "delayed" queue until there are no stripes currently
* in a pre-read phase. Further, if the "delayed" queue is empty when
* a stripe is put on it then we "plug" the queue and do not process it
* until an unplug call is made. (the unplug_io_fn() is called).
*
* When preread is initiated on a stripe, we set PREREAD_ACTIVE and add
* it to the count of prereading stripes.
* When write is initiated, or the stripe refcnt == 0 (just in case) we
* clear the PREREAD_ACTIVE flag and decrement the count
* Whenever the 'handle' queue is empty and the device is not plugged, we
* move any strips from delayed to handle and clear the DELAYED flag and set
* PREREAD_ACTIVE.
* In stripe_handle, if we find pre-reading is necessary, we do it if
* PREREAD_ACTIVE is set, else we set DELAYED which will send it to the delayed queue.
* HANDLE gets cleared if stripe_handle leaves nothing locked.
*/
struct disk_info {
struct md_rdev *rdev, *replacement;
};
raid5: relieve lock contention in get_active_stripe() get_active_stripe() is the last place we have lock contention. It has two paths. One is stripe isn't found and new stripe is allocated, the other is stripe is found. The first path basically calls __find_stripe and init_stripe. It accesses conf->generation, conf->previous_raid_disks, conf->raid_disks, conf->prev_chunk_sectors, conf->chunk_sectors, conf->max_degraded, conf->prev_algo, conf->algorithm, the stripe_hashtbl and inactive_list. Except stripe_hashtbl and inactive_list, other fields are changed very rarely. With this patch, we split inactive_list and add new hash locks. Each free stripe belongs to a specific inactive list. Which inactive list is determined by stripe's lock_hash. Note, even a stripe hasn't a sector assigned, it has a lock_hash assigned. Stripe's inactive list is protected by a hash lock, which is determined by it's lock_hash too. The lock_hash is derivied from current stripe_hashtbl hash, which guarantees any stripe_hashtbl list will be assigned to a specific lock_hash, so we can use new hash lock to protect stripe_hashtbl list too. The goal of the new hash locks introduced is we can only use the new locks in the first path of get_active_stripe(). Since we have several hash locks, lock contention is relieved significantly. The first path of get_active_stripe() accesses other fields, since they are changed rarely, changing them now need take conf->device_lock and all hash locks. For a slow path, this isn't a problem. If we need lock device_lock and hash lock, we always lock hash lock first. The tricky part is release_stripe and friends. We need take device_lock first. Neil's suggestion is we put inactive stripes to a temporary list and readd it to inactive_list after device_lock is released. In this way, we add stripes to temporary list with device_lock hold and remove stripes from the list with hash lock hold. So we don't allow concurrent access to the temporary list, which means we need allocate temporary list for all participants of release_stripe. One downside is free stripes are maintained in their inactive list, they can't across between the lists. By default, we have total 256 stripes and 8 lists, so each list will have 32 stripes. It's possible one list has free stripe but other list hasn't. The chance should be rare because stripes allocation are even distributed. And we can always allocate more stripes for cache, several mega bytes memory isn't a big deal. This completely removes the lock contention of the first path of get_active_stripe(). It slows down the second code path a little bit though because we now need takes two locks, but since the hash lock isn't contended, the overhead should be quite small (several atomic instructions). The second path of get_active_stripe() (basically sequential write or big request size randwrite) still has lock contentions. Signed-off-by: Shaohua Li <shli@fusionio.com> Signed-off-by: NeilBrown <neilb@suse.de>
2013-11-14 12:16:17 +08:00
/* NOTE NR_STRIPE_HASH_LOCKS must remain below 64.
* This is because we sometimes take all the spinlocks
* and creating that much locking depth can cause
* problems.
*/
#define NR_STRIPE_HASH_LOCKS 8
#define STRIPE_HASH_LOCKS_MASK (NR_STRIPE_HASH_LOCKS - 1)
raid5: offload stripe handle to workqueue This is another attempt to create multiple threads to handle raid5 stripes. This time I use workqueue. raid5 handles request (especially write) in stripe unit. A stripe is page size aligned/long and acrosses all disks. Writing to any disk sector, raid5 runs a state machine for the corresponding stripe, which includes reading some disks of the stripe, calculating parity, and writing some disks of the stripe. The state machine is running in raid5d thread currently. Since there is only one thread, it doesn't scale well for high speed storage. An obvious solution is multi-threading. To get better performance, we have some requirements: a. locality. stripe corresponding to request submitted from one cpu is better handled in thread in local cpu or local node. local cpu is preferred but some times could be a bottleneck, for example, parity calculation is too heavy. local node running has wide adaptability. b. configurablity. Different setup of raid5 array might need diffent configuration. Especially the thread number. More threads don't always mean better performance because of lock contentions. My original implementation is creating some kernel threads. There are interfaces to control which cpu's stripe each thread should handle. And userspace can set affinity of the threads. This provides biggest flexibility and configurability. But it's hard to use and apparently a new thread pool implementation is disfavor. Recent workqueue improvement is quite promising. unbound workqueue will be bound to numa node. If WQ_SYSFS is set in workqueue, there are sysfs option to do affinity setting. For example, we can only include one HT sibling in affinity. Since work is non-reentrant by default, and we can control running thread number by limiting dispatched work_struct number. In this patch, I created several stripe worker group. A group is a numa node. stripes from cpus of one node will be added to a group list. Workqueue thread of one node will only handle stripes of worker group of the node. In this way, stripe handling has numa node locality. And as I said, we can control thread number by limiting dispatched work_struct number. The work_struct callback function handles several stripes in one run. A typical work queue usage is to run one unit in each work_struct. In raid5 case, the unit is a stripe. But we can't do that: a. Though handling a stripe doesn't need lock because of reference accounting and stripe isn't in any list, queuing a work_struct for each stripe will make workqueue lock contended very heavily. b. blk_start_plug()/blk_finish_plug() should surround stripe handle, as we might dispatch request. If each work_struct only handles one stripe, such block plug is meaningless. This implementation can't do very fine grained configuration. But the numa binding is most popular usage model, should be enough for most workloads. Note: since we have only one stripe queue, switching to multi-thread might decrease request size dispatching down to low level layer. The impact depends on thread number, raid configuration and workload. So multi-thread raid5 might not be proper for all setups. Changes V1 -> V2: 1. remove WQ_NON_REENTRANT 2. disabling multi-threading by default 3. Add more descriptions in changelog Signed-off-by: Shaohua Li <shli@fusionio.com> Signed-off-by: NeilBrown <neilb@suse.de>
2013-08-28 14:30:16 +08:00
struct r5worker {
struct work_struct work;
struct r5worker_group *group;
raid5: relieve lock contention in get_active_stripe() get_active_stripe() is the last place we have lock contention. It has two paths. One is stripe isn't found and new stripe is allocated, the other is stripe is found. The first path basically calls __find_stripe and init_stripe. It accesses conf->generation, conf->previous_raid_disks, conf->raid_disks, conf->prev_chunk_sectors, conf->chunk_sectors, conf->max_degraded, conf->prev_algo, conf->algorithm, the stripe_hashtbl and inactive_list. Except stripe_hashtbl and inactive_list, other fields are changed very rarely. With this patch, we split inactive_list and add new hash locks. Each free stripe belongs to a specific inactive list. Which inactive list is determined by stripe's lock_hash. Note, even a stripe hasn't a sector assigned, it has a lock_hash assigned. Stripe's inactive list is protected by a hash lock, which is determined by it's lock_hash too. The lock_hash is derivied from current stripe_hashtbl hash, which guarantees any stripe_hashtbl list will be assigned to a specific lock_hash, so we can use new hash lock to protect stripe_hashtbl list too. The goal of the new hash locks introduced is we can only use the new locks in the first path of get_active_stripe(). Since we have several hash locks, lock contention is relieved significantly. The first path of get_active_stripe() accesses other fields, since they are changed rarely, changing them now need take conf->device_lock and all hash locks. For a slow path, this isn't a problem. If we need lock device_lock and hash lock, we always lock hash lock first. The tricky part is release_stripe and friends. We need take device_lock first. Neil's suggestion is we put inactive stripes to a temporary list and readd it to inactive_list after device_lock is released. In this way, we add stripes to temporary list with device_lock hold and remove stripes from the list with hash lock hold. So we don't allow concurrent access to the temporary list, which means we need allocate temporary list for all participants of release_stripe. One downside is free stripes are maintained in their inactive list, they can't across between the lists. By default, we have total 256 stripes and 8 lists, so each list will have 32 stripes. It's possible one list has free stripe but other list hasn't. The chance should be rare because stripes allocation are even distributed. And we can always allocate more stripes for cache, several mega bytes memory isn't a big deal. This completely removes the lock contention of the first path of get_active_stripe(). It slows down the second code path a little bit though because we now need takes two locks, but since the hash lock isn't contended, the overhead should be quite small (several atomic instructions). The second path of get_active_stripe() (basically sequential write or big request size randwrite) still has lock contentions. Signed-off-by: Shaohua Li <shli@fusionio.com> Signed-off-by: NeilBrown <neilb@suse.de>
2013-11-14 12:16:17 +08:00
struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS];
bool working;
raid5: offload stripe handle to workqueue This is another attempt to create multiple threads to handle raid5 stripes. This time I use workqueue. raid5 handles request (especially write) in stripe unit. A stripe is page size aligned/long and acrosses all disks. Writing to any disk sector, raid5 runs a state machine for the corresponding stripe, which includes reading some disks of the stripe, calculating parity, and writing some disks of the stripe. The state machine is running in raid5d thread currently. Since there is only one thread, it doesn't scale well for high speed storage. An obvious solution is multi-threading. To get better performance, we have some requirements: a. locality. stripe corresponding to request submitted from one cpu is better handled in thread in local cpu or local node. local cpu is preferred but some times could be a bottleneck, for example, parity calculation is too heavy. local node running has wide adaptability. b. configurablity. Different setup of raid5 array might need diffent configuration. Especially the thread number. More threads don't always mean better performance because of lock contentions. My original implementation is creating some kernel threads. There are interfaces to control which cpu's stripe each thread should handle. And userspace can set affinity of the threads. This provides biggest flexibility and configurability. But it's hard to use and apparently a new thread pool implementation is disfavor. Recent workqueue improvement is quite promising. unbound workqueue will be bound to numa node. If WQ_SYSFS is set in workqueue, there are sysfs option to do affinity setting. For example, we can only include one HT sibling in affinity. Since work is non-reentrant by default, and we can control running thread number by limiting dispatched work_struct number. In this patch, I created several stripe worker group. A group is a numa node. stripes from cpus of one node will be added to a group list. Workqueue thread of one node will only handle stripes of worker group of the node. In this way, stripe handling has numa node locality. And as I said, we can control thread number by limiting dispatched work_struct number. The work_struct callback function handles several stripes in one run. A typical work queue usage is to run one unit in each work_struct. In raid5 case, the unit is a stripe. But we can't do that: a. Though handling a stripe doesn't need lock because of reference accounting and stripe isn't in any list, queuing a work_struct for each stripe will make workqueue lock contended very heavily. b. blk_start_plug()/blk_finish_plug() should surround stripe handle, as we might dispatch request. If each work_struct only handles one stripe, such block plug is meaningless. This implementation can't do very fine grained configuration. But the numa binding is most popular usage model, should be enough for most workloads. Note: since we have only one stripe queue, switching to multi-thread might decrease request size dispatching down to low level layer. The impact depends on thread number, raid configuration and workload. So multi-thread raid5 might not be proper for all setups. Changes V1 -> V2: 1. remove WQ_NON_REENTRANT 2. disabling multi-threading by default 3. Add more descriptions in changelog Signed-off-by: Shaohua Li <shli@fusionio.com> Signed-off-by: NeilBrown <neilb@suse.de>
2013-08-28 14:30:16 +08:00
};
struct r5worker_group {
struct list_head handle_list;
struct r5conf *conf;
struct r5worker *workers;
int stripes_cnt;
raid5: offload stripe handle to workqueue This is another attempt to create multiple threads to handle raid5 stripes. This time I use workqueue. raid5 handles request (especially write) in stripe unit. A stripe is page size aligned/long and acrosses all disks. Writing to any disk sector, raid5 runs a state machine for the corresponding stripe, which includes reading some disks of the stripe, calculating parity, and writing some disks of the stripe. The state machine is running in raid5d thread currently. Since there is only one thread, it doesn't scale well for high speed storage. An obvious solution is multi-threading. To get better performance, we have some requirements: a. locality. stripe corresponding to request submitted from one cpu is better handled in thread in local cpu or local node. local cpu is preferred but some times could be a bottleneck, for example, parity calculation is too heavy. local node running has wide adaptability. b. configurablity. Different setup of raid5 array might need diffent configuration. Especially the thread number. More threads don't always mean better performance because of lock contentions. My original implementation is creating some kernel threads. There are interfaces to control which cpu's stripe each thread should handle. And userspace can set affinity of the threads. This provides biggest flexibility and configurability. But it's hard to use and apparently a new thread pool implementation is disfavor. Recent workqueue improvement is quite promising. unbound workqueue will be bound to numa node. If WQ_SYSFS is set in workqueue, there are sysfs option to do affinity setting. For example, we can only include one HT sibling in affinity. Since work is non-reentrant by default, and we can control running thread number by limiting dispatched work_struct number. In this patch, I created several stripe worker group. A group is a numa node. stripes from cpus of one node will be added to a group list. Workqueue thread of one node will only handle stripes of worker group of the node. In this way, stripe handling has numa node locality. And as I said, we can control thread number by limiting dispatched work_struct number. The work_struct callback function handles several stripes in one run. A typical work queue usage is to run one unit in each work_struct. In raid5 case, the unit is a stripe. But we can't do that: a. Though handling a stripe doesn't need lock because of reference accounting and stripe isn't in any list, queuing a work_struct for each stripe will make workqueue lock contended very heavily. b. blk_start_plug()/blk_finish_plug() should surround stripe handle, as we might dispatch request. If each work_struct only handles one stripe, such block plug is meaningless. This implementation can't do very fine grained configuration. But the numa binding is most popular usage model, should be enough for most workloads. Note: since we have only one stripe queue, switching to multi-thread might decrease request size dispatching down to low level layer. The impact depends on thread number, raid configuration and workload. So multi-thread raid5 might not be proper for all setups. Changes V1 -> V2: 1. remove WQ_NON_REENTRANT 2. disabling multi-threading by default 3. Add more descriptions in changelog Signed-off-by: Shaohua Li <shli@fusionio.com> Signed-off-by: NeilBrown <neilb@suse.de>
2013-08-28 14:30:16 +08:00
};
struct r5conf {
struct hlist_head *stripe_hashtbl;
raid5: relieve lock contention in get_active_stripe() get_active_stripe() is the last place we have lock contention. It has two paths. One is stripe isn't found and new stripe is allocated, the other is stripe is found. The first path basically calls __find_stripe and init_stripe. It accesses conf->generation, conf->previous_raid_disks, conf->raid_disks, conf->prev_chunk_sectors, conf->chunk_sectors, conf->max_degraded, conf->prev_algo, conf->algorithm, the stripe_hashtbl and inactive_list. Except stripe_hashtbl and inactive_list, other fields are changed very rarely. With this patch, we split inactive_list and add new hash locks. Each free stripe belongs to a specific inactive list. Which inactive list is determined by stripe's lock_hash. Note, even a stripe hasn't a sector assigned, it has a lock_hash assigned. Stripe's inactive list is protected by a hash lock, which is determined by it's lock_hash too. The lock_hash is derivied from current stripe_hashtbl hash, which guarantees any stripe_hashtbl list will be assigned to a specific lock_hash, so we can use new hash lock to protect stripe_hashtbl list too. The goal of the new hash locks introduced is we can only use the new locks in the first path of get_active_stripe(). Since we have several hash locks, lock contention is relieved significantly. The first path of get_active_stripe() accesses other fields, since they are changed rarely, changing them now need take conf->device_lock and all hash locks. For a slow path, this isn't a problem. If we need lock device_lock and hash lock, we always lock hash lock first. The tricky part is release_stripe and friends. We need take device_lock first. Neil's suggestion is we put inactive stripes to a temporary list and readd it to inactive_list after device_lock is released. In this way, we add stripes to temporary list with device_lock hold and remove stripes from the list with hash lock hold. So we don't allow concurrent access to the temporary list, which means we need allocate temporary list for all participants of release_stripe. One downside is free stripes are maintained in their inactive list, they can't across between the lists. By default, we have total 256 stripes and 8 lists, so each list will have 32 stripes. It's possible one list has free stripe but other list hasn't. The chance should be rare because stripes allocation are even distributed. And we can always allocate more stripes for cache, several mega bytes memory isn't a big deal. This completely removes the lock contention of the first path of get_active_stripe(). It slows down the second code path a little bit though because we now need takes two locks, but since the hash lock isn't contended, the overhead should be quite small (several atomic instructions). The second path of get_active_stripe() (basically sequential write or big request size randwrite) still has lock contentions. Signed-off-by: Shaohua Li <shli@fusionio.com> Signed-off-by: NeilBrown <neilb@suse.de>
2013-11-14 12:16:17 +08:00
/* only protect corresponding hash list and inactive_list */
spinlock_t hash_locks[NR_STRIPE_HASH_LOCKS];
struct mddev *mddev;
int chunk_sectors;
md/raid5: activate raid6 rmw feature Glue it altogehter. The raid6 rmw path should work the same as the already existing raid5 logic. So emulate the prexor handling/flags and split functions as needed. 1) Enable xor_syndrome() in the async layer. 2) Split ops_run_prexor() into RAID4/5 and RAID6 logic. Xor the syndrome at the start of a rmw run as we did it before for the single parity. 3) Take care of rmw run in ops_run_reconstruct6(). Again process only the changed pages to get syndrome back into sync. 4) Enhance set_syndrome_sources() to fill NULL pages if we are in a rmw run. The lower layers will calculate start & end pages from that and call the xor_syndrome() correspondingly. 5) Adapt the several places where we ignored Q handling up to now. Performance numbers for a single E5630 system with a mix of 10 7200k desktop/server disks. 300 seconds random write with 8 threads onto a 3,2TB (10*400GB) RAID6 64K chunk without spare (group_thread_cnt=4) bsize rmw_level=1 rmw_level=0 rmw_level=1 rmw_level=0 skip_copy=1 skip_copy=1 skip_copy=0 skip_copy=0 4K 115 KB/s 141 KB/s 165 KB/s 140 KB/s 8K 225 KB/s 275 KB/s 324 KB/s 274 KB/s 16K 434 KB/s 536 KB/s 640 KB/s 534 KB/s 32K 751 KB/s 1,051 KB/s 1,234 KB/s 1,045 KB/s 64K 1,339 KB/s 1,958 KB/s 2,282 KB/s 1,962 KB/s 128K 2,673 KB/s 3,862 KB/s 4,113 KB/s 3,898 KB/s 256K 7,685 KB/s 7,539 KB/s 7,557 KB/s 7,638 KB/s 512K 19,556 KB/s 19,558 KB/s 19,652 KB/s 19,688 Kb/s Signed-off-by: Markus Stockhausen <stockhausen@collogia.de> Signed-off-by: NeilBrown <neilb@suse.de>
2014-12-15 09:57:05 +08:00
int level, algorithm, rmw_level;
int max_degraded;
int raid_disks;
int max_nr_stripes;
md/raid5: allow the stripe_cache to grow and shrink. The default setting of 256 stripe_heads is probably much too small for many configurations. So it is best to make it auto-configure. Shrinking the cache under memory pressure is easy. The only interesting part here is that we put a fairly high cost ('seeks') on shrinking the cache as the cost is greater than just having to read more data, it reduces parallelism. Growing the cache on demand needs to be done carefully. If we allow fast growth, that can upset memory balance as lots of dirty memory can quickly turn into lots of memory queued in the stripe_cache. It is important for the raid5 block device to appear congested to allow write-throttling to work. So we only add stripes slowly. We set a flag when an allocation fails because all stripes are in use, allocate at a convenient time when that flag is set, and don't allow it to be set again until at least one stripe_head has been released for re-use. This means that a spurt of requests will only cause one stripe_head to be allocated, but a steady stream of requests will slowly increase the cache size - until memory pressure puts it back again. It could take hours to reach a steady state. The value written to, and displayed in, stripe_cache_size is used as a minimum. The cache can grow above this and shrink back down to it. The actual size is not directly visible, though it can be deduced to some extent by watching stripe_cache_active. Signed-off-by: NeilBrown <neilb@suse.de>
2015-02-26 09:47:56 +08:00
int min_nr_stripes;
/* reshape_progress is the leading edge of a 'reshape'
* It has value MaxSector when no reshape is happening
* If delta_disks < 0, it is the last sector we started work on,
* else is it the next sector to work on.
*/
sector_t reshape_progress;
/* reshape_safe is the trailing edge of a reshape. We know that
* before (or after) this address, all reshape has completed.
*/
sector_t reshape_safe;
int previous_raid_disks;
int prev_chunk_sectors;
int prev_algo;
short generation; /* increments with every reshape */
seqcount_t gen_lock; /* lock against generation changes */
unsigned long reshape_checkpoint; /* Time we last updated
* metadata */
long long min_offset_diff; /* minimum difference between
* data_offset and
* new_data_offset across all
* devices. May be negative,
* but is closest to zero.
*/
struct list_head handle_list; /* stripes needing handling */
md: introduce get_priority_stripe() to improve raid456 write performance Improve write performance by preventing the delayed_list from dumping all its stripes onto the handle_list in one shot. Delayed stripes are now further delayed by being held on the 'hold_list'. The 'hold_list' is bypassed when: * a STRIPE_IO_STARTED stripe is found at the head of 'handle_list' * 'handle_list' is empty and i/o is being done to satisfy full stripe-width write requests * 'bypass_count' is less than 'bypass_threshold'. By default the threshold is 1, i.e. every other stripe handled is a preread stripe provided the top two conditions are false. Benchmark data: System: 2x Xeon 5150, 4x SATA, mem=1GB Baseline: 2.6.24-rc7 Configuration: mdadm --create /dev/md0 /dev/sd[b-e] -n 4 -l 5 --assume-clean Test1: dd if=/dev/zero of=/dev/md0 bs=1024k count=2048 * patched: +33% (stripe_cache_size = 256), +25% (stripe_cache_size = 512) Test2: tiobench --size 2048 --numruns 5 --block 4096 --block 131072 (XFS) * patched: +13% * patched + preread_bypass_threshold = 0: +37% Changes since v1: * reduce bypass_threshold from (chunk_size / sectors_per_chunk) to (1) and make it configurable. This defaults to fairness and modest performance gains out of the box. Changes since v2: * [neilb@suse.de]: kill STRIPE_PRIO_HI and preread_needed as they are not necessary, the important change was clearing STRIPE_DELAYED in add_stripe_bio and this has been moved out to make_request for the hang fix. * [neilb@suse.de]: simplify get_priority_stripe * [dan.j.williams@intel.com]: reset the bypass_count when ->hold_list is sampled empty (+11%) * [dan.j.williams@intel.com]: decrement the bypass_count at the detection of stripes being naturally promoted off of hold_list +2%. Note, resetting bypass_count instead of decrementing on these events yields +4% but that is probably too aggressive. Changes since v3: * cosmetic fixups Tested-by: James W. Laferriere <babydr@baby-dragons.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Signed-off-by: Neil Brown <neilb@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-28 17:15:53 +08:00
struct list_head hold_list; /* preread ready stripes */
struct list_head delayed_list; /* stripes that have plugged requests */
struct list_head bitmap_list; /* stripes delaying awaiting bitmap update */
[PATCH] md: allow reads that have bypassed the cache to be retried on failure If a bypass-the-cache read fails, we simply try again through the cache. If it fails again it will trigger normal recovery precedures. update 1: From: NeilBrown <neilb@suse.de> 1/ chunk_aligned_read and retry_aligned_read assume that data_disks == raid_disks - 1 which is not true for raid6. So when an aligned read request bypasses the cache, we can get the wrong data. 2/ The cloned bio is being used-after-free in raid5_align_endio (to test BIO_UPTODATE). 3/ We forgot to add rdev->data_offset when submitting a bio for aligned-read 4/ clone_bio calls blk_recount_segments and then we change bi_bdev, so we need to invalidate the segment counts. 5/ We don't de-reference the rdev when the read completes. This means we need to record the rdev to so it is still available in the end_io routine. Fortunately bi_next in the original bio is unused at this point so we can stuff it in there. 6/ We leak a cloned bio if the target rdev is not usable. From: NeilBrown <neilb@suse.de> update 2: 1/ When aligned requests fail (read error) they need to be retried via the normal method (stripe cache). As we cannot be sure that we can process a single read in one go (we may not be able to allocate all the stripes needed) we store a bio-being-retried and a list of bioes-that-still-need-to-be-retried. When find a bio that needs to be retried, we should add it to the list, not to single-bio... 2/ We were never incrementing 'scnt' when resubmitting failed aligned requests. [akpm@osdl.org: build fix] Signed-off-by: Neil Brown <neilb@suse.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 18:20:47 +08:00
struct bio *retry_read_aligned; /* currently retrying aligned bios */
struct bio *retry_read_aligned_list; /* aligned bios retry list */
atomic_t preread_active_stripes; /* stripes with scheduled io */
[PATCH] md: allow reads that have bypassed the cache to be retried on failure If a bypass-the-cache read fails, we simply try again through the cache. If it fails again it will trigger normal recovery precedures. update 1: From: NeilBrown <neilb@suse.de> 1/ chunk_aligned_read and retry_aligned_read assume that data_disks == raid_disks - 1 which is not true for raid6. So when an aligned read request bypasses the cache, we can get the wrong data. 2/ The cloned bio is being used-after-free in raid5_align_endio (to test BIO_UPTODATE). 3/ We forgot to add rdev->data_offset when submitting a bio for aligned-read 4/ clone_bio calls blk_recount_segments and then we change bi_bdev, so we need to invalidate the segment counts. 5/ We don't de-reference the rdev when the read completes. This means we need to record the rdev to so it is still available in the end_io routine. Fortunately bi_next in the original bio is unused at this point so we can stuff it in there. 6/ We leak a cloned bio if the target rdev is not usable. From: NeilBrown <neilb@suse.de> update 2: 1/ When aligned requests fail (read error) they need to be retried via the normal method (stripe cache). As we cannot be sure that we can process a single read in one go (we may not be able to allocate all the stripes needed) we store a bio-being-retried and a list of bioes-that-still-need-to-be-retried. When find a bio that needs to be retried, we should add it to the list, not to single-bio... 2/ We were never incrementing 'scnt' when resubmitting failed aligned requests. [akpm@osdl.org: build fix] Signed-off-by: Neil Brown <neilb@suse.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 18:20:47 +08:00
atomic_t active_aligned_reads;
md: introduce get_priority_stripe() to improve raid456 write performance Improve write performance by preventing the delayed_list from dumping all its stripes onto the handle_list in one shot. Delayed stripes are now further delayed by being held on the 'hold_list'. The 'hold_list' is bypassed when: * a STRIPE_IO_STARTED stripe is found at the head of 'handle_list' * 'handle_list' is empty and i/o is being done to satisfy full stripe-width write requests * 'bypass_count' is less than 'bypass_threshold'. By default the threshold is 1, i.e. every other stripe handled is a preread stripe provided the top two conditions are false. Benchmark data: System: 2x Xeon 5150, 4x SATA, mem=1GB Baseline: 2.6.24-rc7 Configuration: mdadm --create /dev/md0 /dev/sd[b-e] -n 4 -l 5 --assume-clean Test1: dd if=/dev/zero of=/dev/md0 bs=1024k count=2048 * patched: +33% (stripe_cache_size = 256), +25% (stripe_cache_size = 512) Test2: tiobench --size 2048 --numruns 5 --block 4096 --block 131072 (XFS) * patched: +13% * patched + preread_bypass_threshold = 0: +37% Changes since v1: * reduce bypass_threshold from (chunk_size / sectors_per_chunk) to (1) and make it configurable. This defaults to fairness and modest performance gains out of the box. Changes since v2: * [neilb@suse.de]: kill STRIPE_PRIO_HI and preread_needed as they are not necessary, the important change was clearing STRIPE_DELAYED in add_stripe_bio and this has been moved out to make_request for the hang fix. * [neilb@suse.de]: simplify get_priority_stripe * [dan.j.williams@intel.com]: reset the bypass_count when ->hold_list is sampled empty (+11%) * [dan.j.williams@intel.com]: decrement the bypass_count at the detection of stripes being naturally promoted off of hold_list +2%. Note, resetting bypass_count instead of decrementing on these events yields +4% but that is probably too aggressive. Changes since v3: * cosmetic fixups Tested-by: James W. Laferriere <babydr@baby-dragons.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Signed-off-by: Neil Brown <neilb@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-28 17:15:53 +08:00
atomic_t pending_full_writes; /* full write backlog */
int bypass_count; /* bypassed prereads */
int bypass_threshold; /* preread nice */
int skip_copy; /* Don't copy data from bio to stripe cache */
md: introduce get_priority_stripe() to improve raid456 write performance Improve write performance by preventing the delayed_list from dumping all its stripes onto the handle_list in one shot. Delayed stripes are now further delayed by being held on the 'hold_list'. The 'hold_list' is bypassed when: * a STRIPE_IO_STARTED stripe is found at the head of 'handle_list' * 'handle_list' is empty and i/o is being done to satisfy full stripe-width write requests * 'bypass_count' is less than 'bypass_threshold'. By default the threshold is 1, i.e. every other stripe handled is a preread stripe provided the top two conditions are false. Benchmark data: System: 2x Xeon 5150, 4x SATA, mem=1GB Baseline: 2.6.24-rc7 Configuration: mdadm --create /dev/md0 /dev/sd[b-e] -n 4 -l 5 --assume-clean Test1: dd if=/dev/zero of=/dev/md0 bs=1024k count=2048 * patched: +33% (stripe_cache_size = 256), +25% (stripe_cache_size = 512) Test2: tiobench --size 2048 --numruns 5 --block 4096 --block 131072 (XFS) * patched: +13% * patched + preread_bypass_threshold = 0: +37% Changes since v1: * reduce bypass_threshold from (chunk_size / sectors_per_chunk) to (1) and make it configurable. This defaults to fairness and modest performance gains out of the box. Changes since v2: * [neilb@suse.de]: kill STRIPE_PRIO_HI and preread_needed as they are not necessary, the important change was clearing STRIPE_DELAYED in add_stripe_bio and this has been moved out to make_request for the hang fix. * [neilb@suse.de]: simplify get_priority_stripe * [dan.j.williams@intel.com]: reset the bypass_count when ->hold_list is sampled empty (+11%) * [dan.j.williams@intel.com]: decrement the bypass_count at the detection of stripes being naturally promoted off of hold_list +2%. Note, resetting bypass_count instead of decrementing on these events yields +4% but that is probably too aggressive. Changes since v3: * cosmetic fixups Tested-by: James W. Laferriere <babydr@baby-dragons.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Signed-off-by: Neil Brown <neilb@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-28 17:15:53 +08:00
struct list_head *last_hold; /* detect hold_list promotions */
/* bios to have bi_end_io called after metadata is synced */
struct bio_list return_bi;
atomic_t reshape_stripes; /* stripes with pending writes for reshape */
/* unfortunately we need two cache names as we temporarily have
* two caches.
*/
int active_name;
char cache_name[2][32];
struct kmem_cache *slab_cache; /* for allocating stripes */
struct mutex cache_size_mutex; /* Protect changes to cache size */
int seq_flush, seq_write;
int quiesce;
int fullsync; /* set to 1 if a full sync is needed,
* (fresh device added).
* Cleared when a sync completes.
*/
int recovery_disabled;
/* per cpu variables */
struct raid5_percpu {
struct page *spare_page; /* Used when checking P/Q in raid6 */
struct flex_array *scribble; /* space for constructing buffer
* lists and performing address
* conversions
*/
} __percpu *percpu;
#ifdef CONFIG_HOTPLUG_CPU
struct notifier_block cpu_notify;
#endif
/*
* Free stripes pool
*/
atomic_t active_stripes;
raid5: relieve lock contention in get_active_stripe() get_active_stripe() is the last place we have lock contention. It has two paths. One is stripe isn't found and new stripe is allocated, the other is stripe is found. The first path basically calls __find_stripe and init_stripe. It accesses conf->generation, conf->previous_raid_disks, conf->raid_disks, conf->prev_chunk_sectors, conf->chunk_sectors, conf->max_degraded, conf->prev_algo, conf->algorithm, the stripe_hashtbl and inactive_list. Except stripe_hashtbl and inactive_list, other fields are changed very rarely. With this patch, we split inactive_list and add new hash locks. Each free stripe belongs to a specific inactive list. Which inactive list is determined by stripe's lock_hash. Note, even a stripe hasn't a sector assigned, it has a lock_hash assigned. Stripe's inactive list is protected by a hash lock, which is determined by it's lock_hash too. The lock_hash is derivied from current stripe_hashtbl hash, which guarantees any stripe_hashtbl list will be assigned to a specific lock_hash, so we can use new hash lock to protect stripe_hashtbl list too. The goal of the new hash locks introduced is we can only use the new locks in the first path of get_active_stripe(). Since we have several hash locks, lock contention is relieved significantly. The first path of get_active_stripe() accesses other fields, since they are changed rarely, changing them now need take conf->device_lock and all hash locks. For a slow path, this isn't a problem. If we need lock device_lock and hash lock, we always lock hash lock first. The tricky part is release_stripe and friends. We need take device_lock first. Neil's suggestion is we put inactive stripes to a temporary list and readd it to inactive_list after device_lock is released. In this way, we add stripes to temporary list with device_lock hold and remove stripes from the list with hash lock hold. So we don't allow concurrent access to the temporary list, which means we need allocate temporary list for all participants of release_stripe. One downside is free stripes are maintained in their inactive list, they can't across between the lists. By default, we have total 256 stripes and 8 lists, so each list will have 32 stripes. It's possible one list has free stripe but other list hasn't. The chance should be rare because stripes allocation are even distributed. And we can always allocate more stripes for cache, several mega bytes memory isn't a big deal. This completely removes the lock contention of the first path of get_active_stripe(). It slows down the second code path a little bit though because we now need takes two locks, but since the hash lock isn't contended, the overhead should be quite small (several atomic instructions). The second path of get_active_stripe() (basically sequential write or big request size randwrite) still has lock contentions. Signed-off-by: Shaohua Li <shli@fusionio.com> Signed-off-by: NeilBrown <neilb@suse.de>
2013-11-14 12:16:17 +08:00
struct list_head inactive_list[NR_STRIPE_HASH_LOCKS];
atomic_t empty_inactive_list_nr;
struct llist_head released_stripes;
wait_queue_head_t wait_for_quiescent;
md/raid5: per hash value and exclusive wait_for_stripe I noticed heavy spin lock contention at get_active_stripe() with fsmark multiple thread write workloads. Here is how this hot contention comes from. We have limited stripes, and it's a multiple thread write workload. Hence, those stripes will be taken soon, which puts later processes to sleep for waiting free stripes. When enough stripes(>= 1/4 total stripes) are released, all process are woken, trying to get the lock. But there is one only being able to get this lock for each hash lock, making other processes spinning out there for acquiring the lock. Thus, it's effectiveless to wakeup all processes and let them battle for a lock that permits one to access only each time. Instead, we could make it be a exclusive wake up: wake up one process only. That avoids the heavy spin lock contention naturally. To do the exclusive wake up, we've to split wait_for_stripe into multiple wait queues, to make it per hash value, just like the hash lock. Here are some test results I have got with this patch applied(all test run 3 times): `fsmark.files_per_sec' ===================== next-20150317 this patch ------------------------- ------------------------- metric_value ±stddev metric_value ±stddev change testbox/benchmark/testcase-params ------------------------- ------------------------- -------- ------------------------------ 25.600 ±0.0 92.700 ±2.5 262.1% ivb44/fsmark/1x-64t-4BRD_12G-RAID5-btrfs-4M-30G-fsyncBeforeClose 25.600 ±0.0 77.800 ±0.6 203.9% ivb44/fsmark/1x-64t-9BRD_6G-RAID5-btrfs-4M-30G-fsyncBeforeClose 32.000 ±0.0 93.800 ±1.7 193.1% ivb44/fsmark/1x-64t-4BRD_12G-RAID5-ext4-4M-30G-fsyncBeforeClose 32.000 ±0.0 81.233 ±1.7 153.9% ivb44/fsmark/1x-64t-9BRD_6G-RAID5-ext4-4M-30G-fsyncBeforeClose 48.800 ±14.5 99.667 ±2.0 104.2% ivb44/fsmark/1x-64t-4BRD_12G-RAID5-xfs-4M-30G-fsyncBeforeClose 6.400 ±0.0 12.800 ±0.0 100.0% ivb44/fsmark/1x-64t-3HDD-RAID5-btrfs-4M-40G-fsyncBeforeClose 63.133 ±8.2 82.800 ±0.7 31.2% ivb44/fsmark/1x-64t-9BRD_6G-RAID5-xfs-4M-30G-fsyncBeforeClose 245.067 ±0.7 306.567 ±7.9 25.1% ivb44/fsmark/1x-64t-4BRD_12G-RAID5-f2fs-4M-30G-fsyncBeforeClose 17.533 ±0.3 21.000 ±0.8 19.8% ivb44/fsmark/1x-1t-3HDD-RAID5-xfs-4M-40G-fsyncBeforeClose 188.167 ±1.9 215.033 ±3.1 14.3% ivb44/fsmark/1x-1t-4BRD_12G-RAID5-btrfs-4M-30G-NoSync 254.500 ±1.8 290.733 ±2.4 14.2% ivb44/fsmark/1x-1t-9BRD_6G-RAID5-btrfs-4M-30G-NoSync `time.system_time' ===================== next-20150317 this patch ------------------------- ------------------------- metric_value ±stddev metric_value ±stddev change testbox/benchmark/testcase-params ------------------------- ------------------------- -------- ------------------------------ 7235.603 ±1.2 185.163 ±1.9 -97.4% ivb44/fsmark/1x-64t-4BRD_12G-RAID5-btrfs-4M-30G-fsyncBeforeClose 7666.883 ±2.9 202.750 ±1.0 -97.4% ivb44/fsmark/1x-64t-9BRD_6G-RAID5-btrfs-4M-30G-fsyncBeforeClose 14567.893 ±0.7 421.230 ±0.4 -97.1% ivb44/fsmark/1x-64t-3HDD-RAID5-btrfs-4M-40G-fsyncBeforeClose 3697.667 ±14.0 148.190 ±1.7 -96.0% ivb44/fsmark/1x-64t-4BRD_12G-RAID5-xfs-4M-30G-fsyncBeforeClose 5572.867 ±3.8 310.717 ±1.4 -94.4% ivb44/fsmark/1x-64t-9BRD_6G-RAID5-ext4-4M-30G-fsyncBeforeClose 5565.050 ±0.5 313.277 ±1.5 -94.4% ivb44/fsmark/1x-64t-4BRD_12G-RAID5-ext4-4M-30G-fsyncBeforeClose 2420.707 ±17.1 171.043 ±2.7 -92.9% ivb44/fsmark/1x-64t-9BRD_6G-RAID5-xfs-4M-30G-fsyncBeforeClose 3743.300 ±4.6 379.827 ±3.5 -89.9% ivb44/fsmark/1x-64t-3HDD-RAID5-ext4-4M-40G-fsyncBeforeClose 3308.687 ±6.3 363.050 ±2.0 -89.0% ivb44/fsmark/1x-64t-3HDD-RAID5-xfs-4M-40G-fsyncBeforeClose Where, 1x: where 'x' means iterations or loop, corresponding to the 'L' option of fsmark 1t, 64t: where 't' means thread 4M: means the single file size, corresponding to the '-s' option of fsmark 40G, 30G, 120G: means the total test size 4BRD_12G: BRD is the ramdisk, where '4' means 4 ramdisk, and where '12G' means the size of one ramdisk. So, it would be 48G in total. And we made a raid on those ramdisk As you can see, though there are no much performance gain for hard disk workload, the system time is dropped heavily, up to 97%. And as expected, the performance increased a lot, up to 260%, for fast device(ram disk). v2: use bits instead of array to note down wait queue need to wake up. Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com> Signed-off-by: NeilBrown <neilb@suse.de>
2015-05-08 16:19:07 +08:00
wait_queue_head_t wait_for_stripe[NR_STRIPE_HASH_LOCKS];
wait_queue_head_t wait_for_overlap;
unsigned long cache_state;
#define R5_INACTIVE_BLOCKED 1 /* release of inactive stripes blocked,
* waiting for 25% to be free
*/
md/raid5: allow the stripe_cache to grow and shrink. The default setting of 256 stripe_heads is probably much too small for many configurations. So it is best to make it auto-configure. Shrinking the cache under memory pressure is easy. The only interesting part here is that we put a fairly high cost ('seeks') on shrinking the cache as the cost is greater than just having to read more data, it reduces parallelism. Growing the cache on demand needs to be done carefully. If we allow fast growth, that can upset memory balance as lots of dirty memory can quickly turn into lots of memory queued in the stripe_cache. It is important for the raid5 block device to appear congested to allow write-throttling to work. So we only add stripes slowly. We set a flag when an allocation fails because all stripes are in use, allocate at a convenient time when that flag is set, and don't allow it to be set again until at least one stripe_head has been released for re-use. This means that a spurt of requests will only cause one stripe_head to be allocated, but a steady stream of requests will slowly increase the cache size - until memory pressure puts it back again. It could take hours to reach a steady state. The value written to, and displayed in, stripe_cache_size is used as a minimum. The cache can grow above this and shrink back down to it. The actual size is not directly visible, though it can be deduced to some extent by watching stripe_cache_active. Signed-off-by: NeilBrown <neilb@suse.de>
2015-02-26 09:47:56 +08:00
#define R5_ALLOC_MORE 2 /* It might help to allocate another
* stripe.
*/
#define R5_DID_ALLOC 4 /* A stripe was allocated, don't allocate
* more until at least one has been
* released. This avoids flooding
* the cache.
*/
struct shrinker shrinker;
int pool_size; /* number of disks in stripeheads in pool */
spinlock_t device_lock;
struct disk_info *disks;
/* When taking over an array from a different personality, we store
* the new thread here until we fully activate the array.
*/
struct md_thread *thread;
raid5: relieve lock contention in get_active_stripe() get_active_stripe() is the last place we have lock contention. It has two paths. One is stripe isn't found and new stripe is allocated, the other is stripe is found. The first path basically calls __find_stripe and init_stripe. It accesses conf->generation, conf->previous_raid_disks, conf->raid_disks, conf->prev_chunk_sectors, conf->chunk_sectors, conf->max_degraded, conf->prev_algo, conf->algorithm, the stripe_hashtbl and inactive_list. Except stripe_hashtbl and inactive_list, other fields are changed very rarely. With this patch, we split inactive_list and add new hash locks. Each free stripe belongs to a specific inactive list. Which inactive list is determined by stripe's lock_hash. Note, even a stripe hasn't a sector assigned, it has a lock_hash assigned. Stripe's inactive list is protected by a hash lock, which is determined by it's lock_hash too. The lock_hash is derivied from current stripe_hashtbl hash, which guarantees any stripe_hashtbl list will be assigned to a specific lock_hash, so we can use new hash lock to protect stripe_hashtbl list too. The goal of the new hash locks introduced is we can only use the new locks in the first path of get_active_stripe(). Since we have several hash locks, lock contention is relieved significantly. The first path of get_active_stripe() accesses other fields, since they are changed rarely, changing them now need take conf->device_lock and all hash locks. For a slow path, this isn't a problem. If we need lock device_lock and hash lock, we always lock hash lock first. The tricky part is release_stripe and friends. We need take device_lock first. Neil's suggestion is we put inactive stripes to a temporary list and readd it to inactive_list after device_lock is released. In this way, we add stripes to temporary list with device_lock hold and remove stripes from the list with hash lock hold. So we don't allow concurrent access to the temporary list, which means we need allocate temporary list for all participants of release_stripe. One downside is free stripes are maintained in their inactive list, they can't across between the lists. By default, we have total 256 stripes and 8 lists, so each list will have 32 stripes. It's possible one list has free stripe but other list hasn't. The chance should be rare because stripes allocation are even distributed. And we can always allocate more stripes for cache, several mega bytes memory isn't a big deal. This completely removes the lock contention of the first path of get_active_stripe(). It slows down the second code path a little bit though because we now need takes two locks, but since the hash lock isn't contended, the overhead should be quite small (several atomic instructions). The second path of get_active_stripe() (basically sequential write or big request size randwrite) still has lock contentions. Signed-off-by: Shaohua Li <shli@fusionio.com> Signed-off-by: NeilBrown <neilb@suse.de>
2013-11-14 12:16:17 +08:00
struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS];
raid5: offload stripe handle to workqueue This is another attempt to create multiple threads to handle raid5 stripes. This time I use workqueue. raid5 handles request (especially write) in stripe unit. A stripe is page size aligned/long and acrosses all disks. Writing to any disk sector, raid5 runs a state machine for the corresponding stripe, which includes reading some disks of the stripe, calculating parity, and writing some disks of the stripe. The state machine is running in raid5d thread currently. Since there is only one thread, it doesn't scale well for high speed storage. An obvious solution is multi-threading. To get better performance, we have some requirements: a. locality. stripe corresponding to request submitted from one cpu is better handled in thread in local cpu or local node. local cpu is preferred but some times could be a bottleneck, for example, parity calculation is too heavy. local node running has wide adaptability. b. configurablity. Different setup of raid5 array might need diffent configuration. Especially the thread number. More threads don't always mean better performance because of lock contentions. My original implementation is creating some kernel threads. There are interfaces to control which cpu's stripe each thread should handle. And userspace can set affinity of the threads. This provides biggest flexibility and configurability. But it's hard to use and apparently a new thread pool implementation is disfavor. Recent workqueue improvement is quite promising. unbound workqueue will be bound to numa node. If WQ_SYSFS is set in workqueue, there are sysfs option to do affinity setting. For example, we can only include one HT sibling in affinity. Since work is non-reentrant by default, and we can control running thread number by limiting dispatched work_struct number. In this patch, I created several stripe worker group. A group is a numa node. stripes from cpus of one node will be added to a group list. Workqueue thread of one node will only handle stripes of worker group of the node. In this way, stripe handling has numa node locality. And as I said, we can control thread number by limiting dispatched work_struct number. The work_struct callback function handles several stripes in one run. A typical work queue usage is to run one unit in each work_struct. In raid5 case, the unit is a stripe. But we can't do that: a. Though handling a stripe doesn't need lock because of reference accounting and stripe isn't in any list, queuing a work_struct for each stripe will make workqueue lock contended very heavily. b. blk_start_plug()/blk_finish_plug() should surround stripe handle, as we might dispatch request. If each work_struct only handles one stripe, such block plug is meaningless. This implementation can't do very fine grained configuration. But the numa binding is most popular usage model, should be enough for most workloads. Note: since we have only one stripe queue, switching to multi-thread might decrease request size dispatching down to low level layer. The impact depends on thread number, raid configuration and workload. So multi-thread raid5 might not be proper for all setups. Changes V1 -> V2: 1. remove WQ_NON_REENTRANT 2. disabling multi-threading by default 3. Add more descriptions in changelog Signed-off-by: Shaohua Li <shli@fusionio.com> Signed-off-by: NeilBrown <neilb@suse.de>
2013-08-28 14:30:16 +08:00
struct r5worker_group *worker_groups;
int group_cnt;
int worker_cnt_per_group;
raid5: add basic stripe log This introduces a simple log for raid5. Data/parity writing to raid array first writes to the log, then write to raid array disks. If crash happens, we can recovery data from the log. This can speed up raid resync and fix write hole issue. The log structure is pretty simple. Data/meta data is stored in block unit, which is 4k generally. It has only one type of meta data block. The meta data block can track 3 types of data, stripe data, stripe parity and flush block. MD superblock will point to the last valid meta data block. Each meta data block has checksum/seq number, so recovery can scan the log correctly. We store a checksum of stripe data/parity to the metadata block, so meta data and stripe data/parity can be written to log disk together. otherwise, meta data write must wait till stripe data/parity is finished. For stripe data, meta data block will record stripe data sector and size. Currently the size is always 4k. This meta data record can be made simpler if we just fix write hole (eg, we can record data of a stripe's different disks together), but this format can be extended to support caching in the future, which must record data address/size. For stripe parity, meta data block will record stripe sector. It's size should be 4k (for raid5) or 8k (for raid6). We always store p parity first. This format should work for caching too. flush block indicates a stripe is in raid array disks. Fixing write hole doesn't need this type of meta data, it's for caching extension. Signed-off-by: Shaohua Li <shli@fb.com> Signed-off-by: NeilBrown <neilb@suse.com>
2015-08-14 05:31:59 +08:00
struct r5l_log *log;
};
/*
* Our supported algorithms
*/
#define ALGORITHM_LEFT_ASYMMETRIC 0 /* Rotating Parity N with Data Restart */
#define ALGORITHM_RIGHT_ASYMMETRIC 1 /* Rotating Parity 0 with Data Restart */
#define ALGORITHM_LEFT_SYMMETRIC 2 /* Rotating Parity N with Data Continuation */
#define ALGORITHM_RIGHT_SYMMETRIC 3 /* Rotating Parity 0 with Data Continuation */
/* Define non-rotating (raid4) algorithms. These allow
* conversion of raid4 to raid5.
*/
#define ALGORITHM_PARITY_0 4 /* P or P,Q are initial devices */
#define ALGORITHM_PARITY_N 5 /* P or P,Q are final devices. */
/* DDF RAID6 layouts differ from md/raid6 layouts in two ways.
* Firstly, the exact positioning of the parity block is slightly
* different between the 'LEFT_*' modes of md and the "_N_*" modes
* of DDF.
* Secondly, or order of datablocks over which the Q syndrome is computed
* is different.
* Consequently we have different layouts for DDF/raid6 than md/raid6.
* These layouts are from the DDFv1.2 spec.
* Interestingly DDFv1.2-Errata-A does not specify N_CONTINUE but
* leaves RLQ=3 as 'Vendor Specific'
*/
#define ALGORITHM_ROTATING_ZERO_RESTART 8 /* DDF PRL=6 RLQ=1 */
#define ALGORITHM_ROTATING_N_RESTART 9 /* DDF PRL=6 RLQ=2 */
#define ALGORITHM_ROTATING_N_CONTINUE 10 /*DDF PRL=6 RLQ=3 */
/* For every RAID5 algorithm we define a RAID6 algorithm
* with exactly the same layout for data and parity, and
* with the Q block always on the last device (N-1).
* This allows trivial conversion from RAID5 to RAID6
*/
#define ALGORITHM_LEFT_ASYMMETRIC_6 16
#define ALGORITHM_RIGHT_ASYMMETRIC_6 17
#define ALGORITHM_LEFT_SYMMETRIC_6 18
#define ALGORITHM_RIGHT_SYMMETRIC_6 19
#define ALGORITHM_PARITY_0_6 20
#define ALGORITHM_PARITY_N_6 ALGORITHM_PARITY_N
static inline int algorithm_valid_raid5(int layout)
{
return (layout >= 0) &&
(layout <= 5);
}
static inline int algorithm_valid_raid6(int layout)
{
return (layout >= 0 && layout <= 5)
||
(layout >= 8 && layout <= 10)
||
(layout >= 16 && layout <= 20);
}
static inline int algorithm_is_DDF(int layout)
{
return layout >= 8 && layout <= 10;
}
extern void md_raid5_kick_device(struct r5conf *conf);
extern int raid5_set_cache_size(struct mddev *mddev, int size);
extern sector_t raid5_compute_blocknr(struct stripe_head *sh, int i, int previous);
extern void raid5_release_stripe(struct stripe_head *sh);
extern sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector,
int previous, int *dd_idx,
struct stripe_head *sh);
extern struct stripe_head *
raid5_get_active_stripe(struct r5conf *conf, sector_t sector,
int previous, int noblock, int noquiesce);
raid5: add basic stripe log This introduces a simple log for raid5. Data/parity writing to raid array first writes to the log, then write to raid array disks. If crash happens, we can recovery data from the log. This can speed up raid resync and fix write hole issue. The log structure is pretty simple. Data/meta data is stored in block unit, which is 4k generally. It has only one type of meta data block. The meta data block can track 3 types of data, stripe data, stripe parity and flush block. MD superblock will point to the last valid meta data block. Each meta data block has checksum/seq number, so recovery can scan the log correctly. We store a checksum of stripe data/parity to the metadata block, so meta data and stripe data/parity can be written to log disk together. otherwise, meta data write must wait till stripe data/parity is finished. For stripe data, meta data block will record stripe data sector and size. Currently the size is always 4k. This meta data record can be made simpler if we just fix write hole (eg, we can record data of a stripe's different disks together), but this format can be extended to support caching in the future, which must record data address/size. For stripe parity, meta data block will record stripe sector. It's size should be 4k (for raid5) or 8k (for raid6). We always store p parity first. This format should work for caching too. flush block indicates a stripe is in raid array disks. Fixing write hole doesn't need this type of meta data, it's for caching extension. Signed-off-by: Shaohua Li <shli@fb.com> Signed-off-by: NeilBrown <neilb@suse.com>
2015-08-14 05:31:59 +08:00
extern int r5l_init_log(struct r5conf *conf, struct md_rdev *rdev);
extern void r5l_exit_log(struct r5l_log *log);
extern int r5l_write_stripe(struct r5l_log *log, struct stripe_head *head_sh);
extern void r5l_write_stripe_run(struct r5l_log *log);
raid5: log reclaim support This is the reclaim support for raid5 log. A stripe write will have following steps: 1. reconstruct the stripe, read data/calculate parity. ops_run_io prepares to write data/parity to raid disks 2. hijack ops_run_io. stripe data/parity is appending to log disk 3. flush log disk cache 4. ops_run_io run again and do normal operation. stripe data/parity is written in raid array disks. raid core can return io to upper layer. 5. flush cache of all raid array disks 6. update super block 7. log disk space used by the stripe can be reused In practice, several stripes consist of an io_unit and we will batch several io_unit in different steps, but the whole process doesn't change. It's possible io return just after data/parity hit log disk, but then read IO will need read from log disk. For simplicity, IO return happens at step 4, where read IO can directly read from raid disks. Currently reclaim run if there is specific reclaimable space (1/4 disk size or 10G) or we are out of space. Reclaim is just to free log disk spaces, it doesn't impact data consistency. The size based force reclaim is to make sure log isn't too big, so recovery doesn't scan log too much. Recovery make sure raid disks and log disk have the same data of a stripe. If crash happens before 4, recovery might/might not recovery stripe's data/parity depending on if data/parity and its checksum matches. In either case, this doesn't change the syntax of an IO write. After step 3, stripe is guaranteed recoverable, because stripe's data/parity is persistent in log disk. In some cases, log disk content and raid disks content of a stripe are the same, but recovery will still copy log disk content to raid disks, this doesn't impact data consistency. space reuse happens after superblock update and cache flush. There is one situation we want to avoid. A broken meta in the middle of a log causes recovery can't find meta at the head of log. If operations require meta at the head persistent in log, we must make sure meta before it persistent in log too. The case is stripe data/parity is in log and we start write stripe to raid disks (before step 4). stripe data/parity must be persistent in log before we do the write to raid disks. The solution is we restrictly maintain io_unit list order. In this case, we only write stripes of an io_unit to raid disks till the io_unit is the first one whose data/parity is in log. The io_unit list order is important for other cases too. For example, some io_unit are reclaimable and others not. They can be mixed in the list, we shouldn't reuse space of an unreclaimable io_unit. Includes fixes to problems which were... Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Shaohua Li <shli@fb.com> Signed-off-by: NeilBrown <neilb@suse.com>
2015-08-14 05:32:00 +08:00
extern void r5l_flush_stripe_to_raid(struct r5l_log *log);
extern void r5l_stripe_write_finished(struct stripe_head *sh);
extern int r5l_handle_flush_request(struct r5l_log *log, struct bio *bio);
extern void r5l_quiesce(struct r5l_log *log, int state);
extern bool r5l_log_disk_error(struct r5conf *conf);
#endif