Back to home page

OSCL-LXR
 
 

     

0001 /*
0002  * Copyright (C) 2003 Sistina Software Limited.
0003  * Copyright (C) 2005-2008 Red Hat, Inc. All rights reserved.
0004  *
0005  * This file is released under the GPL.
0006  */
0007 
0008 #include "dm-bio-record.h"
0009 
0010 #include <linux/init.h>
0011 #include <linux/mempool.h>
0012 #include <linux/module.h>
0013 #include <linux/pagemap.h>
0014 #include <linux/slab.h>
0015 #include <linux/workqueue.h>
0016 #include <linux/device-mapper.h>
0017 #include <linux/dm-io.h>
0018 #include <linux/dm-dirty-log.h>
0019 #include <linux/dm-kcopyd.h>
0020 #include <linux/dm-region-hash.h>
0021 
0022 #define DM_MSG_PREFIX "raid1"
0023 
0024 #define MAX_RECOVERY 1  /* Maximum number of regions recovered in parallel. */
0025 
0026 #define MAX_NR_MIRRORS  (DM_KCOPYD_MAX_REGIONS + 1)
0027 
0028 #define DM_RAID1_HANDLE_ERRORS  0x01
0029 #define DM_RAID1_KEEP_LOG   0x02
0030 #define errors_handled(p)   ((p)->features & DM_RAID1_HANDLE_ERRORS)
0031 #define keep_log(p)     ((p)->features & DM_RAID1_KEEP_LOG)
0032 
0033 static DECLARE_WAIT_QUEUE_HEAD(_kmirrord_recovery_stopped);
0034 
0035 /*-----------------------------------------------------------------
0036  * Mirror set structures.
0037  *---------------------------------------------------------------*/
0038 enum dm_raid1_error {
0039     DM_RAID1_WRITE_ERROR,
0040     DM_RAID1_FLUSH_ERROR,
0041     DM_RAID1_SYNC_ERROR,
0042     DM_RAID1_READ_ERROR
0043 };
0044 
0045 struct mirror {
0046     struct mirror_set *ms;
0047     atomic_t error_count;
0048     unsigned long error_type;
0049     struct dm_dev *dev;
0050     sector_t offset;
0051 };
0052 
0053 struct mirror_set {
0054     struct dm_target *ti;
0055     struct list_head list;
0056 
0057     uint64_t features;
0058 
0059     spinlock_t lock;    /* protects the lists */
0060     struct bio_list reads;
0061     struct bio_list writes;
0062     struct bio_list failures;
0063     struct bio_list holds;  /* bios are waiting until suspend */
0064 
0065     struct dm_region_hash *rh;
0066     struct dm_kcopyd_client *kcopyd_client;
0067     struct dm_io_client *io_client;
0068 
0069     /* recovery */
0070     region_t nr_regions;
0071     int in_sync;
0072     int log_failure;
0073     int leg_failure;
0074     atomic_t suspend;
0075 
0076     atomic_t default_mirror;    /* Default mirror */
0077 
0078     struct workqueue_struct *kmirrord_wq;
0079     struct work_struct kmirrord_work;
0080     struct timer_list timer;
0081     unsigned long timer_pending;
0082 
0083     struct work_struct trigger_event;
0084 
0085     unsigned nr_mirrors;
0086     struct mirror mirror[];
0087 };
0088 
0089 DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(raid1_resync_throttle,
0090         "A percentage of time allocated for raid resynchronization");
0091 
0092 static void wakeup_mirrord(void *context)
0093 {
0094     struct mirror_set *ms = context;
0095 
0096     queue_work(ms->kmirrord_wq, &ms->kmirrord_work);
0097 }
0098 
0099 static void delayed_wake_fn(struct timer_list *t)
0100 {
0101     struct mirror_set *ms = from_timer(ms, t, timer);
0102 
0103     clear_bit(0, &ms->timer_pending);
0104     wakeup_mirrord(ms);
0105 }
0106 
0107 static void delayed_wake(struct mirror_set *ms)
0108 {
0109     if (test_and_set_bit(0, &ms->timer_pending))
0110         return;
0111 
0112     ms->timer.expires = jiffies + HZ / 5;
0113     add_timer(&ms->timer);
0114 }
0115 
0116 static void wakeup_all_recovery_waiters(void *context)
0117 {
0118     wake_up_all(&_kmirrord_recovery_stopped);
0119 }
0120 
0121 static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw)
0122 {
0123     unsigned long flags;
0124     int should_wake = 0;
0125     struct bio_list *bl;
0126 
0127     bl = (rw == WRITE) ? &ms->writes : &ms->reads;
0128     spin_lock_irqsave(&ms->lock, flags);
0129     should_wake = !(bl->head);
0130     bio_list_add(bl, bio);
0131     spin_unlock_irqrestore(&ms->lock, flags);
0132 
0133     if (should_wake)
0134         wakeup_mirrord(ms);
0135 }
0136 
0137 static void dispatch_bios(void *context, struct bio_list *bio_list)
0138 {
0139     struct mirror_set *ms = context;
0140     struct bio *bio;
0141 
0142     while ((bio = bio_list_pop(bio_list)))
0143         queue_bio(ms, bio, WRITE);
0144 }
0145 
0146 struct dm_raid1_bio_record {
0147     struct mirror *m;
0148     /* if details->bi_bdev == NULL, details were not saved */
0149     struct dm_bio_details details;
0150     region_t write_region;
0151 };
0152 
0153 /*
0154  * Every mirror should look like this one.
0155  */
0156 #define DEFAULT_MIRROR 0
0157 
0158 /*
0159  * This is yucky.  We squirrel the mirror struct away inside
0160  * bi_next for read/write buffers.  This is safe since the bh
0161  * doesn't get submitted to the lower levels of block layer.
0162  */
0163 static struct mirror *bio_get_m(struct bio *bio)
0164 {
0165     return (struct mirror *) bio->bi_next;
0166 }
0167 
0168 static void bio_set_m(struct bio *bio, struct mirror *m)
0169 {
0170     bio->bi_next = (struct bio *) m;
0171 }
0172 
0173 static struct mirror *get_default_mirror(struct mirror_set *ms)
0174 {
0175     return &ms->mirror[atomic_read(&ms->default_mirror)];
0176 }
0177 
0178 static void set_default_mirror(struct mirror *m)
0179 {
0180     struct mirror_set *ms = m->ms;
0181     struct mirror *m0 = &(ms->mirror[0]);
0182 
0183     atomic_set(&ms->default_mirror, m - m0);
0184 }
0185 
0186 static struct mirror *get_valid_mirror(struct mirror_set *ms)
0187 {
0188     struct mirror *m;
0189 
0190     for (m = ms->mirror; m < ms->mirror + ms->nr_mirrors; m++)
0191         if (!atomic_read(&m->error_count))
0192             return m;
0193 
0194     return NULL;
0195 }
0196 
0197 /* fail_mirror
0198  * @m: mirror device to fail
0199  * @error_type: one of the enum's, DM_RAID1_*_ERROR
0200  *
0201  * If errors are being handled, record the type of
0202  * error encountered for this device.  If this type
0203  * of error has already been recorded, we can return;
0204  * otherwise, we must signal userspace by triggering
0205  * an event.  Additionally, if the device is the
0206  * primary device, we must choose a new primary, but
0207  * only if the mirror is in-sync.
0208  *
0209  * This function must not block.
0210  */
0211 static void fail_mirror(struct mirror *m, enum dm_raid1_error error_type)
0212 {
0213     struct mirror_set *ms = m->ms;
0214     struct mirror *new;
0215 
0216     ms->leg_failure = 1;
0217 
0218     /*
0219      * error_count is used for nothing more than a
0220      * simple way to tell if a device has encountered
0221      * errors.
0222      */
0223     atomic_inc(&m->error_count);
0224 
0225     if (test_and_set_bit(error_type, &m->error_type))
0226         return;
0227 
0228     if (!errors_handled(ms))
0229         return;
0230 
0231     if (m != get_default_mirror(ms))
0232         goto out;
0233 
0234     if (!ms->in_sync && !keep_log(ms)) {
0235         /*
0236          * Better to issue requests to same failing device
0237          * than to risk returning corrupt data.
0238          */
0239         DMERR("Primary mirror (%s) failed while out-of-sync: "
0240               "Reads may fail.", m->dev->name);
0241         goto out;
0242     }
0243 
0244     new = get_valid_mirror(ms);
0245     if (new)
0246         set_default_mirror(new);
0247     else
0248         DMWARN("All sides of mirror have failed.");
0249 
0250 out:
0251     schedule_work(&ms->trigger_event);
0252 }
0253 
0254 static int mirror_flush(struct dm_target *ti)
0255 {
0256     struct mirror_set *ms = ti->private;
0257     unsigned long error_bits;
0258 
0259     unsigned int i;
0260     struct dm_io_region io[MAX_NR_MIRRORS];
0261     struct mirror *m;
0262     struct dm_io_request io_req = {
0263         .bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC,
0264         .mem.type = DM_IO_KMEM,
0265         .mem.ptr.addr = NULL,
0266         .client = ms->io_client,
0267     };
0268 
0269     for (i = 0, m = ms->mirror; i < ms->nr_mirrors; i++, m++) {
0270         io[i].bdev = m->dev->bdev;
0271         io[i].sector = 0;
0272         io[i].count = 0;
0273     }
0274 
0275     error_bits = -1;
0276     dm_io(&io_req, ms->nr_mirrors, io, &error_bits);
0277     if (unlikely(error_bits != 0)) {
0278         for (i = 0; i < ms->nr_mirrors; i++)
0279             if (test_bit(i, &error_bits))
0280                 fail_mirror(ms->mirror + i,
0281                         DM_RAID1_FLUSH_ERROR);
0282         return -EIO;
0283     }
0284 
0285     return 0;
0286 }
0287 
0288 /*-----------------------------------------------------------------
0289  * Recovery.
0290  *
0291  * When a mirror is first activated we may find that some regions
0292  * are in the no-sync state.  We have to recover these by
0293  * recopying from the default mirror to all the others.
0294  *---------------------------------------------------------------*/
0295 static void recovery_complete(int read_err, unsigned long write_err,
0296                   void *context)
0297 {
0298     struct dm_region *reg = context;
0299     struct mirror_set *ms = dm_rh_region_context(reg);
0300     int m, bit = 0;
0301 
0302     if (read_err) {
0303         /* Read error means the failure of default mirror. */
0304         DMERR_LIMIT("Unable to read primary mirror during recovery");
0305         fail_mirror(get_default_mirror(ms), DM_RAID1_SYNC_ERROR);
0306     }
0307 
0308     if (write_err) {
0309         DMERR_LIMIT("Write error during recovery (error = 0x%lx)",
0310                 write_err);
0311         /*
0312          * Bits correspond to devices (excluding default mirror).
0313          * The default mirror cannot change during recovery.
0314          */
0315         for (m = 0; m < ms->nr_mirrors; m++) {
0316             if (&ms->mirror[m] == get_default_mirror(ms))
0317                 continue;
0318             if (test_bit(bit, &write_err))
0319                 fail_mirror(ms->mirror + m,
0320                         DM_RAID1_SYNC_ERROR);
0321             bit++;
0322         }
0323     }
0324 
0325     dm_rh_recovery_end(reg, !(read_err || write_err));
0326 }
0327 
0328 static void recover(struct mirror_set *ms, struct dm_region *reg)
0329 {
0330     unsigned i;
0331     struct dm_io_region from, to[DM_KCOPYD_MAX_REGIONS], *dest;
0332     struct mirror *m;
0333     unsigned long flags = 0;
0334     region_t key = dm_rh_get_region_key(reg);
0335     sector_t region_size = dm_rh_get_region_size(ms->rh);
0336 
0337     /* fill in the source */
0338     m = get_default_mirror(ms);
0339     from.bdev = m->dev->bdev;
0340     from.sector = m->offset + dm_rh_region_to_sector(ms->rh, key);
0341     if (key == (ms->nr_regions - 1)) {
0342         /*
0343          * The final region may be smaller than
0344          * region_size.
0345          */
0346         from.count = ms->ti->len & (region_size - 1);
0347         if (!from.count)
0348             from.count = region_size;
0349     } else
0350         from.count = region_size;
0351 
0352     /* fill in the destinations */
0353     for (i = 0, dest = to; i < ms->nr_mirrors; i++) {
0354         if (&ms->mirror[i] == get_default_mirror(ms))
0355             continue;
0356 
0357         m = ms->mirror + i;
0358         dest->bdev = m->dev->bdev;
0359         dest->sector = m->offset + dm_rh_region_to_sector(ms->rh, key);
0360         dest->count = from.count;
0361         dest++;
0362     }
0363 
0364     /* hand to kcopyd */
0365     if (!errors_handled(ms))
0366         flags |= BIT(DM_KCOPYD_IGNORE_ERROR);
0367 
0368     dm_kcopyd_copy(ms->kcopyd_client, &from, ms->nr_mirrors - 1, to,
0369                flags, recovery_complete, reg);
0370 }
0371 
0372 static void reset_ms_flags(struct mirror_set *ms)
0373 {
0374     unsigned int m;
0375 
0376     ms->leg_failure = 0;
0377     for (m = 0; m < ms->nr_mirrors; m++) {
0378         atomic_set(&(ms->mirror[m].error_count), 0);
0379         ms->mirror[m].error_type = 0;
0380     }
0381 }
0382 
0383 static void do_recovery(struct mirror_set *ms)
0384 {
0385     struct dm_region *reg;
0386     struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
0387 
0388     /*
0389      * Start quiescing some regions.
0390      */
0391     dm_rh_recovery_prepare(ms->rh);
0392 
0393     /*
0394      * Copy any already quiesced regions.
0395      */
0396     while ((reg = dm_rh_recovery_start(ms->rh)))
0397         recover(ms, reg);
0398 
0399     /*
0400      * Update the in sync flag.
0401      */
0402     if (!ms->in_sync &&
0403         (log->type->get_sync_count(log) == ms->nr_regions)) {
0404         /* the sync is complete */
0405         dm_table_event(ms->ti->table);
0406         ms->in_sync = 1;
0407         reset_ms_flags(ms);
0408     }
0409 }
0410 
0411 /*-----------------------------------------------------------------
0412  * Reads
0413  *---------------------------------------------------------------*/
0414 static struct mirror *choose_mirror(struct mirror_set *ms, sector_t sector)
0415 {
0416     struct mirror *m = get_default_mirror(ms);
0417 
0418     do {
0419         if (likely(!atomic_read(&m->error_count)))
0420             return m;
0421 
0422         if (m-- == ms->mirror)
0423             m += ms->nr_mirrors;
0424     } while (m != get_default_mirror(ms));
0425 
0426     return NULL;
0427 }
0428 
0429 static int default_ok(struct mirror *m)
0430 {
0431     struct mirror *default_mirror = get_default_mirror(m->ms);
0432 
0433     return !atomic_read(&default_mirror->error_count);
0434 }
0435 
0436 static int mirror_available(struct mirror_set *ms, struct bio *bio)
0437 {
0438     struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
0439     region_t region = dm_rh_bio_to_region(ms->rh, bio);
0440 
0441     if (log->type->in_sync(log, region, 0))
0442         return choose_mirror(ms,  bio->bi_iter.bi_sector) ? 1 : 0;
0443 
0444     return 0;
0445 }
0446 
0447 /*
0448  * remap a buffer to a particular mirror.
0449  */
0450 static sector_t map_sector(struct mirror *m, struct bio *bio)
0451 {
0452     if (unlikely(!bio->bi_iter.bi_size))
0453         return 0;
0454     return m->offset + dm_target_offset(m->ms->ti, bio->bi_iter.bi_sector);
0455 }
0456 
0457 static void map_bio(struct mirror *m, struct bio *bio)
0458 {
0459     bio_set_dev(bio, m->dev->bdev);
0460     bio->bi_iter.bi_sector = map_sector(m, bio);
0461 }
0462 
0463 static void map_region(struct dm_io_region *io, struct mirror *m,
0464                struct bio *bio)
0465 {
0466     io->bdev = m->dev->bdev;
0467     io->sector = map_sector(m, bio);
0468     io->count = bio_sectors(bio);
0469 }
0470 
0471 static void hold_bio(struct mirror_set *ms, struct bio *bio)
0472 {
0473     /*
0474      * Lock is required to avoid race condition during suspend
0475      * process.
0476      */
0477     spin_lock_irq(&ms->lock);
0478 
0479     if (atomic_read(&ms->suspend)) {
0480         spin_unlock_irq(&ms->lock);
0481 
0482         /*
0483          * If device is suspended, complete the bio.
0484          */
0485         if (dm_noflush_suspending(ms->ti))
0486             bio->bi_status = BLK_STS_DM_REQUEUE;
0487         else
0488             bio->bi_status = BLK_STS_IOERR;
0489 
0490         bio_endio(bio);
0491         return;
0492     }
0493 
0494     /*
0495      * Hold bio until the suspend is complete.
0496      */
0497     bio_list_add(&ms->holds, bio);
0498     spin_unlock_irq(&ms->lock);
0499 }
0500 
0501 /*-----------------------------------------------------------------
0502  * Reads
0503  *---------------------------------------------------------------*/
0504 static void read_callback(unsigned long error, void *context)
0505 {
0506     struct bio *bio = context;
0507     struct mirror *m;
0508 
0509     m = bio_get_m(bio);
0510     bio_set_m(bio, NULL);
0511 
0512     if (likely(!error)) {
0513         bio_endio(bio);
0514         return;
0515     }
0516 
0517     fail_mirror(m, DM_RAID1_READ_ERROR);
0518 
0519     if (likely(default_ok(m)) || mirror_available(m->ms, bio)) {
0520         DMWARN_LIMIT("Read failure on mirror device %s.  "
0521                  "Trying alternative device.",
0522                  m->dev->name);
0523         queue_bio(m->ms, bio, bio_data_dir(bio));
0524         return;
0525     }
0526 
0527     DMERR_LIMIT("Read failure on mirror device %s.  Failing I/O.",
0528             m->dev->name);
0529     bio_io_error(bio);
0530 }
0531 
0532 /* Asynchronous read. */
0533 static void read_async_bio(struct mirror *m, struct bio *bio)
0534 {
0535     struct dm_io_region io;
0536     struct dm_io_request io_req = {
0537         .bi_opf = REQ_OP_READ,
0538         .mem.type = DM_IO_BIO,
0539         .mem.ptr.bio = bio,
0540         .notify.fn = read_callback,
0541         .notify.context = bio,
0542         .client = m->ms->io_client,
0543     };
0544 
0545     map_region(&io, m, bio);
0546     bio_set_m(bio, m);
0547     BUG_ON(dm_io(&io_req, 1, &io, NULL));
0548 }
0549 
0550 static inline int region_in_sync(struct mirror_set *ms, region_t region,
0551                  int may_block)
0552 {
0553     int state = dm_rh_get_state(ms->rh, region, may_block);
0554     return state == DM_RH_CLEAN || state == DM_RH_DIRTY;
0555 }
0556 
0557 static void do_reads(struct mirror_set *ms, struct bio_list *reads)
0558 {
0559     region_t region;
0560     struct bio *bio;
0561     struct mirror *m;
0562 
0563     while ((bio = bio_list_pop(reads))) {
0564         region = dm_rh_bio_to_region(ms->rh, bio);
0565         m = get_default_mirror(ms);
0566 
0567         /*
0568          * We can only read balance if the region is in sync.
0569          */
0570         if (likely(region_in_sync(ms, region, 1)))
0571             m = choose_mirror(ms, bio->bi_iter.bi_sector);
0572         else if (m && atomic_read(&m->error_count))
0573             m = NULL;
0574 
0575         if (likely(m))
0576             read_async_bio(m, bio);
0577         else
0578             bio_io_error(bio);
0579     }
0580 }
0581 
0582 /*-----------------------------------------------------------------
0583  * Writes.
0584  *
0585  * We do different things with the write io depending on the
0586  * state of the region that it's in:
0587  *
0588  * SYNC:    increment pending, use kcopyd to write to *all* mirrors
0589  * RECOVERING:  delay the io until recovery completes
0590  * NOSYNC:  increment pending, just write to the default mirror
0591  *---------------------------------------------------------------*/
0592 
0593 
0594 static void write_callback(unsigned long error, void *context)
0595 {
0596     unsigned i;
0597     struct bio *bio = (struct bio *) context;
0598     struct mirror_set *ms;
0599     int should_wake = 0;
0600     unsigned long flags;
0601 
0602     ms = bio_get_m(bio)->ms;
0603     bio_set_m(bio, NULL);
0604 
0605     /*
0606      * NOTE: We don't decrement the pending count here,
0607      * instead it is done by the targets endio function.
0608      * This way we handle both writes to SYNC and NOSYNC
0609      * regions with the same code.
0610      */
0611     if (likely(!error)) {
0612         bio_endio(bio);
0613         return;
0614     }
0615 
0616     /*
0617      * If the bio is discard, return an error, but do not
0618      * degrade the array.
0619      */
0620     if (bio_op(bio) == REQ_OP_DISCARD) {
0621         bio->bi_status = BLK_STS_NOTSUPP;
0622         bio_endio(bio);
0623         return;
0624     }
0625 
0626     for (i = 0; i < ms->nr_mirrors; i++)
0627         if (test_bit(i, &error))
0628             fail_mirror(ms->mirror + i, DM_RAID1_WRITE_ERROR);
0629 
0630     /*
0631      * Need to raise event.  Since raising
0632      * events can block, we need to do it in
0633      * the main thread.
0634      */
0635     spin_lock_irqsave(&ms->lock, flags);
0636     if (!ms->failures.head)
0637         should_wake = 1;
0638     bio_list_add(&ms->failures, bio);
0639     spin_unlock_irqrestore(&ms->lock, flags);
0640     if (should_wake)
0641         wakeup_mirrord(ms);
0642 }
0643 
0644 static void do_write(struct mirror_set *ms, struct bio *bio)
0645 {
0646     unsigned int i;
0647     struct dm_io_region io[MAX_NR_MIRRORS], *dest = io;
0648     struct mirror *m;
0649     blk_opf_t op_flags = bio->bi_opf & (REQ_FUA | REQ_PREFLUSH);
0650     struct dm_io_request io_req = {
0651         .bi_opf = REQ_OP_WRITE | op_flags,
0652         .mem.type = DM_IO_BIO,
0653         .mem.ptr.bio = bio,
0654         .notify.fn = write_callback,
0655         .notify.context = bio,
0656         .client = ms->io_client,
0657     };
0658 
0659     if (bio_op(bio) == REQ_OP_DISCARD) {
0660         io_req.bi_opf = REQ_OP_DISCARD | op_flags;
0661         io_req.mem.type = DM_IO_KMEM;
0662         io_req.mem.ptr.addr = NULL;
0663     }
0664 
0665     for (i = 0, m = ms->mirror; i < ms->nr_mirrors; i++, m++)
0666         map_region(dest++, m, bio);
0667 
0668     /*
0669      * Use default mirror because we only need it to retrieve the reference
0670      * to the mirror set in write_callback().
0671      */
0672     bio_set_m(bio, get_default_mirror(ms));
0673 
0674     BUG_ON(dm_io(&io_req, ms->nr_mirrors, io, NULL));
0675 }
0676 
0677 static void do_writes(struct mirror_set *ms, struct bio_list *writes)
0678 {
0679     int state;
0680     struct bio *bio;
0681     struct bio_list sync, nosync, recover, *this_list = NULL;
0682     struct bio_list requeue;
0683     struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
0684     region_t region;
0685 
0686     if (!writes->head)
0687         return;
0688 
0689     /*
0690      * Classify each write.
0691      */
0692     bio_list_init(&sync);
0693     bio_list_init(&nosync);
0694     bio_list_init(&recover);
0695     bio_list_init(&requeue);
0696 
0697     while ((bio = bio_list_pop(writes))) {
0698         if ((bio->bi_opf & REQ_PREFLUSH) ||
0699             (bio_op(bio) == REQ_OP_DISCARD)) {
0700             bio_list_add(&sync, bio);
0701             continue;
0702         }
0703 
0704         region = dm_rh_bio_to_region(ms->rh, bio);
0705 
0706         if (log->type->is_remote_recovering &&
0707             log->type->is_remote_recovering(log, region)) {
0708             bio_list_add(&requeue, bio);
0709             continue;
0710         }
0711 
0712         state = dm_rh_get_state(ms->rh, region, 1);
0713         switch (state) {
0714         case DM_RH_CLEAN:
0715         case DM_RH_DIRTY:
0716             this_list = &sync;
0717             break;
0718 
0719         case DM_RH_NOSYNC:
0720             this_list = &nosync;
0721             break;
0722 
0723         case DM_RH_RECOVERING:
0724             this_list = &recover;
0725             break;
0726         }
0727 
0728         bio_list_add(this_list, bio);
0729     }
0730 
0731     /*
0732      * Add bios that are delayed due to remote recovery
0733      * back on to the write queue
0734      */
0735     if (unlikely(requeue.head)) {
0736         spin_lock_irq(&ms->lock);
0737         bio_list_merge(&ms->writes, &requeue);
0738         spin_unlock_irq(&ms->lock);
0739         delayed_wake(ms);
0740     }
0741 
0742     /*
0743      * Increment the pending counts for any regions that will
0744      * be written to (writes to recover regions are going to
0745      * be delayed).
0746      */
0747     dm_rh_inc_pending(ms->rh, &sync);
0748     dm_rh_inc_pending(ms->rh, &nosync);
0749 
0750     /*
0751      * If the flush fails on a previous call and succeeds here,
0752      * we must not reset the log_failure variable.  We need
0753      * userspace interaction to do that.
0754      */
0755     ms->log_failure = dm_rh_flush(ms->rh) ? 1 : ms->log_failure;
0756 
0757     /*
0758      * Dispatch io.
0759      */
0760     if (unlikely(ms->log_failure) && errors_handled(ms)) {
0761         spin_lock_irq(&ms->lock);
0762         bio_list_merge(&ms->failures, &sync);
0763         spin_unlock_irq(&ms->lock);
0764         wakeup_mirrord(ms);
0765     } else
0766         while ((bio = bio_list_pop(&sync)))
0767             do_write(ms, bio);
0768 
0769     while ((bio = bio_list_pop(&recover)))
0770         dm_rh_delay(ms->rh, bio);
0771 
0772     while ((bio = bio_list_pop(&nosync))) {
0773         if (unlikely(ms->leg_failure) && errors_handled(ms) && !keep_log(ms)) {
0774             spin_lock_irq(&ms->lock);
0775             bio_list_add(&ms->failures, bio);
0776             spin_unlock_irq(&ms->lock);
0777             wakeup_mirrord(ms);
0778         } else {
0779             map_bio(get_default_mirror(ms), bio);
0780             submit_bio_noacct(bio);
0781         }
0782     }
0783 }
0784 
0785 static void do_failures(struct mirror_set *ms, struct bio_list *failures)
0786 {
0787     struct bio *bio;
0788 
0789     if (likely(!failures->head))
0790         return;
0791 
0792     /*
0793      * If the log has failed, unattempted writes are being
0794      * put on the holds list.  We can't issue those writes
0795      * until a log has been marked, so we must store them.
0796      *
0797      * If a 'noflush' suspend is in progress, we can requeue
0798      * the I/O's to the core.  This give userspace a chance
0799      * to reconfigure the mirror, at which point the core
0800      * will reissue the writes.  If the 'noflush' flag is
0801      * not set, we have no choice but to return errors.
0802      *
0803      * Some writes on the failures list may have been
0804      * submitted before the log failure and represent a
0805      * failure to write to one of the devices.  It is ok
0806      * for us to treat them the same and requeue them
0807      * as well.
0808      */
0809     while ((bio = bio_list_pop(failures))) {
0810         if (!ms->log_failure) {
0811             ms->in_sync = 0;
0812             dm_rh_mark_nosync(ms->rh, bio);
0813         }
0814 
0815         /*
0816          * If all the legs are dead, fail the I/O.
0817          * If the device has failed and keep_log is enabled,
0818          * fail the I/O.
0819          *
0820          * If we have been told to handle errors, and keep_log
0821          * isn't enabled, hold the bio and wait for userspace to
0822          * deal with the problem.
0823          *
0824          * Otherwise pretend that the I/O succeeded. (This would
0825          * be wrong if the failed leg returned after reboot and
0826          * got replicated back to the good legs.)
0827          */
0828         if (unlikely(!get_valid_mirror(ms) || (keep_log(ms) && ms->log_failure)))
0829             bio_io_error(bio);
0830         else if (errors_handled(ms) && !keep_log(ms))
0831             hold_bio(ms, bio);
0832         else
0833             bio_endio(bio);
0834     }
0835 }
0836 
0837 static void trigger_event(struct work_struct *work)
0838 {
0839     struct mirror_set *ms =
0840         container_of(work, struct mirror_set, trigger_event);
0841 
0842     dm_table_event(ms->ti->table);
0843 }
0844 
0845 /*-----------------------------------------------------------------
0846  * kmirrord
0847  *---------------------------------------------------------------*/
0848 static void do_mirror(struct work_struct *work)
0849 {
0850     struct mirror_set *ms = container_of(work, struct mirror_set,
0851                          kmirrord_work);
0852     struct bio_list reads, writes, failures;
0853     unsigned long flags;
0854 
0855     spin_lock_irqsave(&ms->lock, flags);
0856     reads = ms->reads;
0857     writes = ms->writes;
0858     failures = ms->failures;
0859     bio_list_init(&ms->reads);
0860     bio_list_init(&ms->writes);
0861     bio_list_init(&ms->failures);
0862     spin_unlock_irqrestore(&ms->lock, flags);
0863 
0864     dm_rh_update_states(ms->rh, errors_handled(ms));
0865     do_recovery(ms);
0866     do_reads(ms, &reads);
0867     do_writes(ms, &writes);
0868     do_failures(ms, &failures);
0869 }
0870 
0871 /*-----------------------------------------------------------------
0872  * Target functions
0873  *---------------------------------------------------------------*/
0874 static struct mirror_set *alloc_context(unsigned int nr_mirrors,
0875                     uint32_t region_size,
0876                     struct dm_target *ti,
0877                     struct dm_dirty_log *dl)
0878 {
0879     struct mirror_set *ms =
0880         kzalloc(struct_size(ms, mirror, nr_mirrors), GFP_KERNEL);
0881 
0882     if (!ms) {
0883         ti->error = "Cannot allocate mirror context";
0884         return NULL;
0885     }
0886 
0887     spin_lock_init(&ms->lock);
0888     bio_list_init(&ms->reads);
0889     bio_list_init(&ms->writes);
0890     bio_list_init(&ms->failures);
0891     bio_list_init(&ms->holds);
0892 
0893     ms->ti = ti;
0894     ms->nr_mirrors = nr_mirrors;
0895     ms->nr_regions = dm_sector_div_up(ti->len, region_size);
0896     ms->in_sync = 0;
0897     ms->log_failure = 0;
0898     ms->leg_failure = 0;
0899     atomic_set(&ms->suspend, 0);
0900     atomic_set(&ms->default_mirror, DEFAULT_MIRROR);
0901 
0902     ms->io_client = dm_io_client_create();
0903     if (IS_ERR(ms->io_client)) {
0904         ti->error = "Error creating dm_io client";
0905         kfree(ms);
0906         return NULL;
0907     }
0908 
0909     ms->rh = dm_region_hash_create(ms, dispatch_bios, wakeup_mirrord,
0910                        wakeup_all_recovery_waiters,
0911                        ms->ti->begin, MAX_RECOVERY,
0912                        dl, region_size, ms->nr_regions);
0913     if (IS_ERR(ms->rh)) {
0914         ti->error = "Error creating dirty region hash";
0915         dm_io_client_destroy(ms->io_client);
0916         kfree(ms);
0917         return NULL;
0918     }
0919 
0920     return ms;
0921 }
0922 
0923 static void free_context(struct mirror_set *ms, struct dm_target *ti,
0924              unsigned int m)
0925 {
0926     while (m--)
0927         dm_put_device(ti, ms->mirror[m].dev);
0928 
0929     dm_io_client_destroy(ms->io_client);
0930     dm_region_hash_destroy(ms->rh);
0931     kfree(ms);
0932 }
0933 
0934 static int get_mirror(struct mirror_set *ms, struct dm_target *ti,
0935               unsigned int mirror, char **argv)
0936 {
0937     unsigned long long offset;
0938     char dummy;
0939     int ret;
0940 
0941     if (sscanf(argv[1], "%llu%c", &offset, &dummy) != 1 ||
0942         offset != (sector_t)offset) {
0943         ti->error = "Invalid offset";
0944         return -EINVAL;
0945     }
0946 
0947     ret = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table),
0948                 &ms->mirror[mirror].dev);
0949     if (ret) {
0950         ti->error = "Device lookup failure";
0951         return ret;
0952     }
0953 
0954     ms->mirror[mirror].ms = ms;
0955     atomic_set(&(ms->mirror[mirror].error_count), 0);
0956     ms->mirror[mirror].error_type = 0;
0957     ms->mirror[mirror].offset = offset;
0958 
0959     return 0;
0960 }
0961 
0962 /*
0963  * Create dirty log: log_type #log_params <log_params>
0964  */
0965 static struct dm_dirty_log *create_dirty_log(struct dm_target *ti,
0966                          unsigned argc, char **argv,
0967                          unsigned *args_used)
0968 {
0969     unsigned param_count;
0970     struct dm_dirty_log *dl;
0971     char dummy;
0972 
0973     if (argc < 2) {
0974         ti->error = "Insufficient mirror log arguments";
0975         return NULL;
0976     }
0977 
0978     if (sscanf(argv[1], "%u%c", &param_count, &dummy) != 1) {
0979         ti->error = "Invalid mirror log argument count";
0980         return NULL;
0981     }
0982 
0983     *args_used = 2 + param_count;
0984 
0985     if (argc < *args_used) {
0986         ti->error = "Insufficient mirror log arguments";
0987         return NULL;
0988     }
0989 
0990     dl = dm_dirty_log_create(argv[0], ti, mirror_flush, param_count,
0991                  argv + 2);
0992     if (!dl) {
0993         ti->error = "Error creating mirror dirty log";
0994         return NULL;
0995     }
0996 
0997     return dl;
0998 }
0999 
1000 static int parse_features(struct mirror_set *ms, unsigned argc, char **argv,
1001               unsigned *args_used)
1002 {
1003     unsigned num_features;
1004     struct dm_target *ti = ms->ti;
1005     char dummy;
1006     int i;
1007 
1008     *args_used = 0;
1009 
1010     if (!argc)
1011         return 0;
1012 
1013     if (sscanf(argv[0], "%u%c", &num_features, &dummy) != 1) {
1014         ti->error = "Invalid number of features";
1015         return -EINVAL;
1016     }
1017 
1018     argc--;
1019     argv++;
1020     (*args_used)++;
1021 
1022     if (num_features > argc) {
1023         ti->error = "Not enough arguments to support feature count";
1024         return -EINVAL;
1025     }
1026 
1027     for (i = 0; i < num_features; i++) {
1028         if (!strcmp("handle_errors", argv[0]))
1029             ms->features |= DM_RAID1_HANDLE_ERRORS;
1030         else if (!strcmp("keep_log", argv[0]))
1031             ms->features |= DM_RAID1_KEEP_LOG;
1032         else {
1033             ti->error = "Unrecognised feature requested";
1034             return -EINVAL;
1035         }
1036 
1037         argc--;
1038         argv++;
1039         (*args_used)++;
1040     }
1041     if (!errors_handled(ms) && keep_log(ms)) {
1042         ti->error = "keep_log feature requires the handle_errors feature";
1043         return -EINVAL;
1044     }
1045 
1046     return 0;
1047 }
1048 
1049 /*
1050  * Construct a mirror mapping:
1051  *
1052  * log_type #log_params <log_params>
1053  * #mirrors [mirror_path offset]{2,}
1054  * [#features <features>]
1055  *
1056  * log_type is "core" or "disk"
1057  * #log_params is between 1 and 3
1058  *
1059  * If present, supported features are "handle_errors" and "keep_log".
1060  */
1061 static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1062 {
1063     int r;
1064     unsigned int nr_mirrors, m, args_used;
1065     struct mirror_set *ms;
1066     struct dm_dirty_log *dl;
1067     char dummy;
1068 
1069     dl = create_dirty_log(ti, argc, argv, &args_used);
1070     if (!dl)
1071         return -EINVAL;
1072 
1073     argv += args_used;
1074     argc -= args_used;
1075 
1076     if (!argc || sscanf(argv[0], "%u%c", &nr_mirrors, &dummy) != 1 ||
1077         nr_mirrors < 2 || nr_mirrors > MAX_NR_MIRRORS) {
1078         ti->error = "Invalid number of mirrors";
1079         dm_dirty_log_destroy(dl);
1080         return -EINVAL;
1081     }
1082 
1083     argv++, argc--;
1084 
1085     if (argc < nr_mirrors * 2) {
1086         ti->error = "Too few mirror arguments";
1087         dm_dirty_log_destroy(dl);
1088         return -EINVAL;
1089     }
1090 
1091     ms = alloc_context(nr_mirrors, dl->type->get_region_size(dl), ti, dl);
1092     if (!ms) {
1093         dm_dirty_log_destroy(dl);
1094         return -ENOMEM;
1095     }
1096 
1097     /* Get the mirror parameter sets */
1098     for (m = 0; m < nr_mirrors; m++) {
1099         r = get_mirror(ms, ti, m, argv);
1100         if (r) {
1101             free_context(ms, ti, m);
1102             return r;
1103         }
1104         argv += 2;
1105         argc -= 2;
1106     }
1107 
1108     ti->private = ms;
1109 
1110     r = dm_set_target_max_io_len(ti, dm_rh_get_region_size(ms->rh));
1111     if (r)
1112         goto err_free_context;
1113 
1114     ti->num_flush_bios = 1;
1115     ti->num_discard_bios = 1;
1116     ti->per_io_data_size = sizeof(struct dm_raid1_bio_record);
1117 
1118     ms->kmirrord_wq = alloc_workqueue("kmirrord", WQ_MEM_RECLAIM, 0);
1119     if (!ms->kmirrord_wq) {
1120         DMERR("couldn't start kmirrord");
1121         r = -ENOMEM;
1122         goto err_free_context;
1123     }
1124     INIT_WORK(&ms->kmirrord_work, do_mirror);
1125     timer_setup(&ms->timer, delayed_wake_fn, 0);
1126     ms->timer_pending = 0;
1127     INIT_WORK(&ms->trigger_event, trigger_event);
1128 
1129     r = parse_features(ms, argc, argv, &args_used);
1130     if (r)
1131         goto err_destroy_wq;
1132 
1133     argv += args_used;
1134     argc -= args_used;
1135 
1136     /*
1137      * Any read-balancing addition depends on the
1138      * DM_RAID1_HANDLE_ERRORS flag being present.
1139      * This is because the decision to balance depends
1140      * on the sync state of a region.  If the above
1141      * flag is not present, we ignore errors; and
1142      * the sync state may be inaccurate.
1143      */
1144 
1145     if (argc) {
1146         ti->error = "Too many mirror arguments";
1147         r = -EINVAL;
1148         goto err_destroy_wq;
1149     }
1150 
1151     ms->kcopyd_client = dm_kcopyd_client_create(&dm_kcopyd_throttle);
1152     if (IS_ERR(ms->kcopyd_client)) {
1153         r = PTR_ERR(ms->kcopyd_client);
1154         goto err_destroy_wq;
1155     }
1156 
1157     wakeup_mirrord(ms);
1158     return 0;
1159 
1160 err_destroy_wq:
1161     destroy_workqueue(ms->kmirrord_wq);
1162 err_free_context:
1163     free_context(ms, ti, ms->nr_mirrors);
1164     return r;
1165 }
1166 
1167 static void mirror_dtr(struct dm_target *ti)
1168 {
1169     struct mirror_set *ms = (struct mirror_set *) ti->private;
1170 
1171     del_timer_sync(&ms->timer);
1172     flush_workqueue(ms->kmirrord_wq);
1173     flush_work(&ms->trigger_event);
1174     dm_kcopyd_client_destroy(ms->kcopyd_client);
1175     destroy_workqueue(ms->kmirrord_wq);
1176     free_context(ms, ti, ms->nr_mirrors);
1177 }
1178 
1179 /*
1180  * Mirror mapping function
1181  */
1182 static int mirror_map(struct dm_target *ti, struct bio *bio)
1183 {
1184     int r, rw = bio_data_dir(bio);
1185     struct mirror *m;
1186     struct mirror_set *ms = ti->private;
1187     struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1188     struct dm_raid1_bio_record *bio_record =
1189       dm_per_bio_data(bio, sizeof(struct dm_raid1_bio_record));
1190 
1191     bio_record->details.bi_bdev = NULL;
1192 
1193     if (rw == WRITE) {
1194         /* Save region for mirror_end_io() handler */
1195         bio_record->write_region = dm_rh_bio_to_region(ms->rh, bio);
1196         queue_bio(ms, bio, rw);
1197         return DM_MAPIO_SUBMITTED;
1198     }
1199 
1200     r = log->type->in_sync(log, dm_rh_bio_to_region(ms->rh, bio), 0);
1201     if (r < 0 && r != -EWOULDBLOCK)
1202         return DM_MAPIO_KILL;
1203 
1204     /*
1205      * If region is not in-sync queue the bio.
1206      */
1207     if (!r || (r == -EWOULDBLOCK)) {
1208         if (bio->bi_opf & REQ_RAHEAD)
1209             return DM_MAPIO_KILL;
1210 
1211         queue_bio(ms, bio, rw);
1212         return DM_MAPIO_SUBMITTED;
1213     }
1214 
1215     /*
1216      * The region is in-sync and we can perform reads directly.
1217      * Store enough information so we can retry if it fails.
1218      */
1219     m = choose_mirror(ms, bio->bi_iter.bi_sector);
1220     if (unlikely(!m))
1221         return DM_MAPIO_KILL;
1222 
1223     dm_bio_record(&bio_record->details, bio);
1224     bio_record->m = m;
1225 
1226     map_bio(m, bio);
1227 
1228     return DM_MAPIO_REMAPPED;
1229 }
1230 
1231 static int mirror_end_io(struct dm_target *ti, struct bio *bio,
1232         blk_status_t *error)
1233 {
1234     int rw = bio_data_dir(bio);
1235     struct mirror_set *ms = (struct mirror_set *) ti->private;
1236     struct mirror *m = NULL;
1237     struct dm_bio_details *bd = NULL;
1238     struct dm_raid1_bio_record *bio_record =
1239       dm_per_bio_data(bio, sizeof(struct dm_raid1_bio_record));
1240 
1241     /*
1242      * We need to dec pending if this was a write.
1243      */
1244     if (rw == WRITE) {
1245         if (!(bio->bi_opf & REQ_PREFLUSH) &&
1246             bio_op(bio) != REQ_OP_DISCARD)
1247             dm_rh_dec(ms->rh, bio_record->write_region);
1248         return DM_ENDIO_DONE;
1249     }
1250 
1251     if (*error == BLK_STS_NOTSUPP)
1252         goto out;
1253 
1254     if (bio->bi_opf & REQ_RAHEAD)
1255         goto out;
1256 
1257     if (unlikely(*error)) {
1258         if (!bio_record->details.bi_bdev) {
1259             /*
1260              * There wasn't enough memory to record necessary
1261              * information for a retry or there was no other
1262              * mirror in-sync.
1263              */
1264             DMERR_LIMIT("Mirror read failed.");
1265             return DM_ENDIO_DONE;
1266         }
1267 
1268         m = bio_record->m;
1269 
1270         DMERR("Mirror read failed from %s. Trying alternative device.",
1271               m->dev->name);
1272 
1273         fail_mirror(m, DM_RAID1_READ_ERROR);
1274 
1275         /*
1276          * A failed read is requeued for another attempt using an intact
1277          * mirror.
1278          */
1279         if (default_ok(m) || mirror_available(ms, bio)) {
1280             bd = &bio_record->details;
1281 
1282             dm_bio_restore(bd, bio);
1283             bio_record->details.bi_bdev = NULL;
1284             bio->bi_status = 0;
1285 
1286             queue_bio(ms, bio, rw);
1287             return DM_ENDIO_INCOMPLETE;
1288         }
1289         DMERR("All replicated volumes dead, failing I/O");
1290     }
1291 
1292 out:
1293     bio_record->details.bi_bdev = NULL;
1294 
1295     return DM_ENDIO_DONE;
1296 }
1297 
1298 static void mirror_presuspend(struct dm_target *ti)
1299 {
1300     struct mirror_set *ms = (struct mirror_set *) ti->private;
1301     struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1302 
1303     struct bio_list holds;
1304     struct bio *bio;
1305 
1306     atomic_set(&ms->suspend, 1);
1307 
1308     /*
1309      * Process bios in the hold list to start recovery waiting
1310      * for bios in the hold list. After the process, no bio has
1311      * a chance to be added in the hold list because ms->suspend
1312      * is set.
1313      */
1314     spin_lock_irq(&ms->lock);
1315     holds = ms->holds;
1316     bio_list_init(&ms->holds);
1317     spin_unlock_irq(&ms->lock);
1318 
1319     while ((bio = bio_list_pop(&holds)))
1320         hold_bio(ms, bio);
1321 
1322     /*
1323      * We must finish up all the work that we've
1324      * generated (i.e. recovery work).
1325      */
1326     dm_rh_stop_recovery(ms->rh);
1327 
1328     wait_event(_kmirrord_recovery_stopped,
1329            !dm_rh_recovery_in_flight(ms->rh));
1330 
1331     if (log->type->presuspend && log->type->presuspend(log))
1332         /* FIXME: need better error handling */
1333         DMWARN("log presuspend failed");
1334 
1335     /*
1336      * Now that recovery is complete/stopped and the
1337      * delayed bios are queued, we need to wait for
1338      * the worker thread to complete.  This way,
1339      * we know that all of our I/O has been pushed.
1340      */
1341     flush_workqueue(ms->kmirrord_wq);
1342 }
1343 
1344 static void mirror_postsuspend(struct dm_target *ti)
1345 {
1346     struct mirror_set *ms = ti->private;
1347     struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1348 
1349     if (log->type->postsuspend && log->type->postsuspend(log))
1350         /* FIXME: need better error handling */
1351         DMWARN("log postsuspend failed");
1352 }
1353 
1354 static void mirror_resume(struct dm_target *ti)
1355 {
1356     struct mirror_set *ms = ti->private;
1357     struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1358 
1359     atomic_set(&ms->suspend, 0);
1360     if (log->type->resume && log->type->resume(log))
1361         /* FIXME: need better error handling */
1362         DMWARN("log resume failed");
1363     dm_rh_start_recovery(ms->rh);
1364 }
1365 
1366 /*
1367  * device_status_char
1368  * @m: mirror device/leg we want the status of
1369  *
1370  * We return one character representing the most severe error
1371  * we have encountered.
1372  *    A => Alive - No failures
1373  *    D => Dead - A write failure occurred leaving mirror out-of-sync
1374  *    S => Sync - A sychronization failure occurred, mirror out-of-sync
1375  *    R => Read - A read failure occurred, mirror data unaffected
1376  *
1377  * Returns: <char>
1378  */
1379 static char device_status_char(struct mirror *m)
1380 {
1381     if (!atomic_read(&(m->error_count)))
1382         return 'A';
1383 
1384     return (test_bit(DM_RAID1_FLUSH_ERROR, &(m->error_type))) ? 'F' :
1385         (test_bit(DM_RAID1_WRITE_ERROR, &(m->error_type))) ? 'D' :
1386         (test_bit(DM_RAID1_SYNC_ERROR, &(m->error_type))) ? 'S' :
1387         (test_bit(DM_RAID1_READ_ERROR, &(m->error_type))) ? 'R' : 'U';
1388 }
1389 
1390 
1391 static void mirror_status(struct dm_target *ti, status_type_t type,
1392               unsigned status_flags, char *result, unsigned maxlen)
1393 {
1394     unsigned int m, sz = 0;
1395     int num_feature_args = 0;
1396     struct mirror_set *ms = (struct mirror_set *) ti->private;
1397     struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1398     char buffer[MAX_NR_MIRRORS + 1];
1399 
1400     switch (type) {
1401     case STATUSTYPE_INFO:
1402         DMEMIT("%d ", ms->nr_mirrors);
1403         for (m = 0; m < ms->nr_mirrors; m++) {
1404             DMEMIT("%s ", ms->mirror[m].dev->name);
1405             buffer[m] = device_status_char(&(ms->mirror[m]));
1406         }
1407         buffer[m] = '\0';
1408 
1409         DMEMIT("%llu/%llu 1 %s ",
1410               (unsigned long long)log->type->get_sync_count(log),
1411               (unsigned long long)ms->nr_regions, buffer);
1412 
1413         sz += log->type->status(log, type, result+sz, maxlen-sz);
1414 
1415         break;
1416 
1417     case STATUSTYPE_TABLE:
1418         sz = log->type->status(log, type, result, maxlen);
1419 
1420         DMEMIT("%d", ms->nr_mirrors);
1421         for (m = 0; m < ms->nr_mirrors; m++)
1422             DMEMIT(" %s %llu", ms->mirror[m].dev->name,
1423                    (unsigned long long)ms->mirror[m].offset);
1424 
1425         num_feature_args += !!errors_handled(ms);
1426         num_feature_args += !!keep_log(ms);
1427         if (num_feature_args) {
1428             DMEMIT(" %d", num_feature_args);
1429             if (errors_handled(ms))
1430                 DMEMIT(" handle_errors");
1431             if (keep_log(ms))
1432                 DMEMIT(" keep_log");
1433         }
1434 
1435         break;
1436 
1437     case STATUSTYPE_IMA:
1438         DMEMIT_TARGET_NAME_VERSION(ti->type);
1439         DMEMIT(",nr_mirrors=%d", ms->nr_mirrors);
1440         for (m = 0; m < ms->nr_mirrors; m++) {
1441             DMEMIT(",mirror_device_%d=%s", m, ms->mirror[m].dev->name);
1442             DMEMIT(",mirror_device_%d_status=%c",
1443                    m, device_status_char(&(ms->mirror[m])));
1444         }
1445 
1446         DMEMIT(",handle_errors=%c", errors_handled(ms) ? 'y' : 'n');
1447         DMEMIT(",keep_log=%c", keep_log(ms) ? 'y' : 'n');
1448 
1449         DMEMIT(",log_type_status=");
1450         sz += log->type->status(log, type, result+sz, maxlen-sz);
1451         DMEMIT(";");
1452         break;
1453     }
1454 }
1455 
1456 static int mirror_iterate_devices(struct dm_target *ti,
1457                   iterate_devices_callout_fn fn, void *data)
1458 {
1459     struct mirror_set *ms = ti->private;
1460     int ret = 0;
1461     unsigned i;
1462 
1463     for (i = 0; !ret && i < ms->nr_mirrors; i++)
1464         ret = fn(ti, ms->mirror[i].dev,
1465              ms->mirror[i].offset, ti->len, data);
1466 
1467     return ret;
1468 }
1469 
1470 static struct target_type mirror_target = {
1471     .name    = "mirror",
1472     .version = {1, 14, 0},
1473     .module  = THIS_MODULE,
1474     .ctr     = mirror_ctr,
1475     .dtr     = mirror_dtr,
1476     .map     = mirror_map,
1477     .end_io  = mirror_end_io,
1478     .presuspend = mirror_presuspend,
1479     .postsuspend = mirror_postsuspend,
1480     .resume  = mirror_resume,
1481     .status  = mirror_status,
1482     .iterate_devices = mirror_iterate_devices,
1483 };
1484 
1485 static int __init dm_mirror_init(void)
1486 {
1487     int r;
1488 
1489     r = dm_register_target(&mirror_target);
1490     if (r < 0) {
1491         DMERR("Failed to register mirror target");
1492         goto bad_target;
1493     }
1494 
1495     return 0;
1496 
1497 bad_target:
1498     return r;
1499 }
1500 
1501 static void __exit dm_mirror_exit(void)
1502 {
1503     dm_unregister_target(&mirror_target);
1504 }
1505 
1506 /* Module hooks */
1507 module_init(dm_mirror_init);
1508 module_exit(dm_mirror_exit);
1509 
1510 MODULE_DESCRIPTION(DM_NAME " mirror target");
1511 MODULE_AUTHOR("Joe Thornber");
1512 MODULE_LICENSE("GPL");
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!