4 #include <linux/raid/md.h>
5 #include <linux/raid/xor.h>
9 * Each stripe contains one buffer per disc. Each buffer can be in
10 * one of a number of states determined by bh_state. Changes between
11 * these states happen *almost* exclusively under a per-stripe
12 * spinlock. Some very specific changes can happen in b_end_io, and
13 * these are not protected by the spin lock.
15 * The bh_state bits that are used to represent these states are:
16 * BH_Uptodate, BH_Lock
18 * State Empty == !Uptodate, !Lock
19 * We have no data, and there is no active request
20 * State Want == !Uptodate, Lock
21 * A read request is being submitted for this block
22 * State Dirty == Uptodate, Lock
23 * Some new data is in this buffer, and it is being written out
24 * State Clean == Uptodate, !Lock
25 * We have valid data which is the same as on disc
27 * The possible state transitions are:
29 * Empty -> Want - on read or write to get old data for parity calc
30 * Empty -> Dirty - on compute_parity to satisfy write/sync request.(RECONSTRUCT_WRITE)
31 * Empty -> Clean - on compute_block when computing a block for failed drive
32 * Want -> Empty - on failed read
33 * Want -> Clean - on successful completion of read request
34 * Dirty -> Clean - on successful completion of write request
35 * Dirty -> Clean - on failed write
36 * Clean -> Dirty - on compute_parity to satisfy write/sync (RECONSTRUCT or RMW)
38 * The Want->Empty, Want->Clean, Dirty->Clean, transitions
39 * all happen in b_end_io at interrupt time.
40 * Each sets the Uptodate bit before releasing the Lock bit.
41 * This leaves one multi-stage transition:
43 * This is safe because thinking that a Clean buffer is actually dirty
44 * will at worst delay some action, and the stripe will be scheduled
45 * for attention after the transition is complete.
47 * There is one possibility that is not covered by these states. That
48 * is if one drive has failed and there is a spare being rebuilt. We
49 * can't distinguish between a clean block that has been generated
50 * from parity calculations, and a clean block that has been
51 * successfully written to the spare ( or to parity when resyncing).
52 * To distingush these states we have a stripe bit STRIPE_INSYNC that
53 * is set whenever a write is scheduled to the spare, or to the parity
54 * disc if there is no spare. A sync request clears this bit, and
55 * when we find it set with no buffers locked, we know the sync is
58 * Buffers for the md device that arrive via make_request are attached
59 * to the appropriate stripe in one of two lists linked on b_reqnext.
60 * One list (bh_read) for read requests, one (bh_write) for write.
61 * There should never be more than one buffer on the two lists
62 * together, but we are not guaranteed of that so we allow for more.
64 * If a buffer is on the read list when the associated cache buffer is
65 * Uptodate, the data is copied into the read buffer and it's b_end_io
66 * routine is called. This may happen in the end_request routine only
67 * if the buffer has just successfully been read. end_request should
68 * remove the buffers from the list and then set the Uptodate bit on
69 * the buffer. Other threads may do this only if they first check
70 * that the Uptodate bit is set. Once they have checked that they may
71 * take buffers off the read queue.
73 * When a buffer on the write list is committed for write is it copied
74 * into the cache buffer, which is then marked dirty, and moved onto a
75 * third list, the written list (bh_written). Once both the parity
76 * block and the cached buffer are successfully written, any buffer on
77 * a written list can be returned with b_end_io.
79 * The write list and read list both act as fifos. The read list is
80 * protected by the device_lock. The write and written lists are
81 * protected by the stripe lock. The device_lock, which can be
82 * claimed while the stipe lock is held, is only for list
83 * manipulations and will only be held for a very short time. It can
84 * be claimed from interrupts.
87 * Stripes in the stripe cache can be on one of two lists (or on
88 * neither). The "inactive_list" contains stripes which are not
89 * currently being used for any request. They can freely be reused
90 * for another stripe. The "handle_list" contains stripes that need
91 * to be handled in some way. Both of these are fifo queues. Each
92 * stripe is also (potentially) linked to a hash bucket in the hash
93 * table so that it can be found by sector number. Stripes that are
94 * not hashed must be on the inactive_list, and will normally be at
95 * the front. All stripes start life this way.
97 * The inactive_list, handle_list and hash bucket lists are all protected by the
99 * - stripes on the inactive_list never have their stripe_lock held.
100 * - stripes have a reference counter. If count==0, they are on a list.
101 * - If a stripe might need handling, STRIPE_HANDLE is set.
102 * - When refcount reaches zero, then if STRIPE_HANDLE it is put on
103 * handle_list else inactive_list
105 * This, combined with the fact that STRIPE_HANDLE is only ever
106 * cleared while a stripe has a non-zero count means that if the
107 * refcount is 0 and STRIPE_HANDLE is set, then it is on the
108 * handle_list and if recount is 0 and STRIPE_HANDLE is not set, then
109 * the stripe is on inactive_list.
111 * The possible transitions are:
112 * activate an unhashed/inactive stripe (get_active_stripe())
113 * lockdev check-hash unlink-stripe cnt++ clean-stripe hash-stripe unlockdev
114 * activate a hashed, possibly active stripe (get_active_stripe())
115 * lockdev check-hash if(!cnt++)unlink-stripe unlockdev
116 * attach a request to an active stripe (add_stripe_bh())
117 * lockdev attach-buffer unlockdev
118 * handle a stripe (handle_stripe())
119 * lockstripe clrSTRIPE_HANDLE ... (lockdev check-buffers unlockdev) .. change-state .. record io needed unlockstripe schedule io
120 * release an active stripe (release_stripe())
121 * lockdev if (!--cnt) { if STRIPE_HANDLE, add to handle_list else add to inactive-list } unlockdev
123 * The refcount counts each thread that have activated the stripe,
124 * plus raid5d if it is handling it, plus one for each active request
125 * on a cached buffer.
128 struct stripe_head *hash_next, **hash_pprev; /* hash pointers */
129 struct list_head lru; /* inactive_list or handle_list */
130 struct raid5_private_data *raid_conf;
131 struct buffer_head *bh_cache[MD_SB_DISKS]; /* buffered copy */
132 struct buffer_head *bh_read[MD_SB_DISKS]; /* read request buffers of the MD device */
133 struct buffer_head *bh_write[MD_SB_DISKS]; /* write request buffers of the MD device */
134 struct buffer_head *bh_written[MD_SB_DISKS]; /* write request buffers of the MD device that have been scheduled for write */
135 struct page *bh_page[MD_SB_DISKS]; /* saved bh_cache[n]->b_page when reading around the cache */
136 unsigned long sector; /* sector of this row */
137 int size; /* buffers size */
138 int pd_idx; /* parity disk index */
139 unsigned long state; /* state flags */
140 atomic_t count; /* nr of active thread/requests */
149 #define RECONSTRUCT_WRITE 1
150 #define READ_MODIFY_WRITE 2
151 /* not a write method, but a compute_parity mode */
152 #define CHECK_PARITY 3
157 #define STRIPE_ERROR 1
158 #define STRIPE_HANDLE 2
159 #define STRIPE_SYNCING 3
160 #define STRIPE_INSYNC 4
161 #define STRIPE_PREREAD_ACTIVE 5
162 #define STRIPE_DELAYED 6
167 * To improve write throughput, we need to delay the handling of some
168 * stripes until there has been a chance that several write requests
169 * for the one stripe have all been collected.
170 * In particular, any write request that would require pre-reading
171 * is put on a "delayed" queue until there are no stripes currently
172 * in a pre-read phase. Further, if the "delayed" queue is empty when
173 * a stripe is put on it then we "plug" the queue and do not process it
174 * until an unplg call is made. (the tq_disk list is run).
176 * When preread is initiated on a stripe, we set PREREAD_ACTIVE and add
177 * it to the count of prereading stripes.
178 * When write is initiated, or the stripe refcnt == 0 (just in case) we
179 * clear the PREREAD_ACTIVE flag and decrement the count
180 * Whenever the delayed queue is empty and the device is not plugged, we
181 * move any strips from delayed to handle and clear the DELAYED flag and set PREREAD_ACTIVE.
182 * In stripe_handle, if we find pre-reading is necessary, we do it if
183 * PREREAD_ACTIVE is set, else we set DELAYED which will send it to the delayed queue.
184 * HANDLE gets cleared if stripe_handle leave nothing locked.
198 struct raid5_private_data {
199 struct stripe_head **stripe_hashtbl;
201 mdk_thread_t *thread, *resync_thread;
202 struct disk_info disks[MD_SB_DISKS];
203 struct disk_info *spare;
205 int chunk_size, level, algorithm;
206 int raid_disks, working_disks, failed_disks;
210 struct list_head handle_list; /* stripes needing handling */
211 struct list_head delayed_list; /* stripes that have plugged requests */
212 atomic_t preread_active_stripes; /* stripes with scheduled io */
216 atomic_t active_stripes;
217 struct list_head inactive_list;
218 md_wait_queue_head_t wait_for_stripe;
219 int inactive_blocked; /* release of inactive stripes blocked,
220 * waiting for 25% to be free
222 md_spinlock_t device_lock;
225 struct tq_struct plug_tq;
228 typedef struct raid5_private_data raid5_conf_t;
230 #define mddev_to_conf(mddev) ((raid5_conf_t *) mddev->private)
233 * Our supported algorithms
235 #define ALGORITHM_LEFT_ASYMMETRIC 0
236 #define ALGORITHM_RIGHT_ASYMMETRIC 1
237 #define ALGORITHM_LEFT_SYMMETRIC 2
238 #define ALGORITHM_RIGHT_SYMMETRIC 3