vmmc.rar_Mirror Mirror

/* * Copyright (C) 2003 Sistina Software Limited. * Copyright (C) 2005-2008 Red Hat, Inc. All rights reserved. * * This file is released under the GPL. */ #include "dm-bio-record.h" #include <linux/init.h> #include <linux/mempool.h> #include <linux/module.h> #include <linux/pagemap.h> #include <linux/slab.h> #include <linux/workqueue.h> #include <linux/device-mapper.h> #include <linux/dm-io.h> #include <linux/dm-dirty-log.h> #include <linux/dm-kcopyd.h> #include <linux/dm-region-hash.h> #define DM_MSG_PREFIX "raid1" #define MAX_RECOVERY 1 /* Maximum number of regions recovered in parallel. */ #define DM_RAID1_HANDLE_ERRORS 0x01 #define errors_handled(p) ((p)->features & DM_RAID1_HANDLE_ERRORS) static DECLARE_WAIT_QUEUE_HEAD(_kmirrord_recovery_stopped); /*----------------------------------------------------------------- * Mirror set structures. *---------------------------------------------------------------*/ enum dm_raid1_error { DM_RAID1_WRITE_ERROR, DM_RAID1_FLUSH_ERROR, DM_RAID1_SYNC_ERROR, DM_RAID1_READ_ERROR }; struct mirror { struct mirror_set *ms; atomic_t error_count; unsigned long error_type; struct dm_dev *dev; sector_t offset; }; struct mirror_set { struct dm_target *ti; struct list_head list; uint64_t features; spinlock_t lock; /* protects the lists */ struct bio_list reads; struct bio_list writes; struct bio_list failures; struct bio_list holds; /* bios are waiting until suspend */ struct dm_region_hash *rh; struct dm_kcopyd_client *kcopyd_client; struct dm_io_client *io_client; /* recovery */ region_t nr_regions; int in_sync; int log_failure; int leg_failure; atomic_t suspend; atomic_t default_mirror; /* Default mirror */ struct workqueue_struct *kmirrord_wq; struct work_struct kmirrord_work; struct timer_list timer; unsigned long timer_pending; struct work_struct trigger_event; unsigned nr_mirrors; struct mirror mirror[0]; }; DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(raid1_resync_throttle, "A percentage of time allocated for raid resynchronization"); static void wakeup_mirrord(void *context) { struct mirror_set *ms = context; queue_work(ms->kmirrord_wq, &ms->kmirrord_work); } static void delayed_wake_fn(unsigned long data) { struct mirror_set *ms = (struct mirror_set *) data; clear_bit(0, &ms->timer_pending); wakeup_mirrord(ms); } static void delayed_wake(struct mirror_set *ms) { if (test_and_set_bit(0, &ms->timer_pending)) return; ms->timer.expires = jiffies + HZ / 5; ms->timer.data = (unsigned long) ms; ms->timer.function = delayed_wake_fn; add_timer(&ms->timer); } static void wakeup_all_recovery_waiters(void *context) { wake_up_all(&_kmirrord_recovery_stopped); } static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw) { unsigned long flags; int should_wake = 0; struct bio_list *bl; bl = (rw == WRITE) ? &ms->writes : &ms->reads; spin_lock_irqsave(&ms->lock, flags); should_wake = !(bl->head); bio_list_add(bl, bio); spin_unlock_irqrestore(&ms->lock, flags); if (should_wake) wakeup_mirrord(ms); } static void dispatch_bios(void *context, struct bio_list *bio_list) { struct mirror_set *ms = context; struct bio *bio; while ((bio = bio_list_pop(bio_list))) queue_bio(ms, bio, WRITE); } struct dm_raid1_bio_record { struct mirror *m; /* if details->bi_bdev == NULL, details were not saved */ struct dm_bio_details details; region_t write_region; }; /* * Every mirror should look like this one. */ #define DEFAULT_MIRROR 0 /* * This is yucky. We squirrel the mirror struct away inside * bi_next for read/write buffers. This is safe since the bh * doesn't get submitted to the lower levels of block layer. */ static struct mirror *bio_get_m(struct bio *bio) { return (struct mirror *) bio->bi_next; } static void bio_set_m(struct bio *bio, struct mirror *m) { bio->bi_next = (struct bio *) m; } static struct mirror *get_default_mirror(struct mirror_set *ms) { return &ms->mirror[atomic_read(&ms->default_mirror)]; } static void set_default_mirror(struct mirror *m) { struct mirror_set *ms = m->ms; struct mirror *m0 = &(ms->mirror[0]); atomic_set(&ms->default_mirror, m - m0); } static struct mirror *get_valid_mirror(struct mirror_set *ms) { struct mirror *m; for (m = ms->mirror; m < ms->mirror + ms->nr_mirrors; m++) if (!atomic_read(&m->error_count)) return m; return NULL; } /* fail_mirror * @m: mirror device to fail * @error_type: one of the enum's, DM_RAID1_*_ERROR * * If errors are being handled, record the type of * error encountered for this device. If this type * of error has already been recorded, we can return; * otherwise, we must signal userspace by triggering * an event. Additionally, if the device is the * primary device, we must choose a new primary, but * only if the mirror is in-sync. * * This function must not block. */ static void fail_mirror(struct mirror *m, enum dm_raid1_error error_type) { struct mirror_set *ms = m->ms; struct mirror *new; ms->leg_failure = 1; /* * error_count is used for nothing more than a * simple way to tell if a device has encountered * errors. */ atomic_inc(&m->error_count); if (test_and_set_bit(error_type, &m->error_type)) return; if (!errors_handled(ms)) return; if (m != get_default_mirror(ms)) goto out; if (!ms->in_sync) { /* * Better to issue requests to same failing device * than to risk returning corrupt data. */ DMERR("Primary mirror (%s) failed while out-of-sync: " "Reads may fail.", m->dev->name); goto out; } new = get_valid_mirror(ms); if (new) set_default_mirror(new); else DMWARN("All sides of mirror have failed."); out: schedule_work(&ms->trigger_event); } static int mirror_flush(struct dm_target *ti) { struct mirror_set *ms = ti->private; unsigned long error_bits; unsigned int i; struct dm_io_region io[ms->nr_mirrors]; struct mirror *m; struct dm_io_request io_req = { .bi_rw = WRITE_FLUSH, .mem.type = DM_IO_KMEM, .mem.ptr.addr = NULL, .client = ms->io_client, }; for (i = 0, m = ms->mirror; i < ms->nr_mirrors; i++, m++) { io[i].bdev = m->dev->bdev; io[i].sector = 0; io[i].count = 0; } error_bits = -1; dm_io(&io_req, ms->nr_mirrors, io, &error_bits); if (unlikely(error_bits != 0)) { for (i = 0; i < ms->nr_mirrors; i++) if (test_bit(i, &error_bits)) fail_mirror(ms->mirror + i, DM_RAID1_FLUSH_ERROR); return -EIO; } return 0; } /*----------------------------------------------------------------- * Recovery. * * When a mirror is first activated we may find that some regions * are in the no-sync state. We have to recover these by * recopying from the default mirror to all the others. *---------------------------------------------------------------*/ static void recovery_complete(int read_err, unsigned long write_err, void *context) { struct dm_region *reg = context; struct mirror_set *ms = dm_rh_region_context(reg); int m, bit = 0; if (read_err) { /* Read error means the failure of default mirror. */ DMERR_LIMIT("Unable to read primary mirror during recovery"); fail_mirror(get_default_mirror(ms), DM_RAID1_SYNC_ERROR); } if (write_err) { DMERR_LIMIT("Write error during recovery (error = 0x%lx)", write_err); /* * Bits correspond to devices (excluding default mirror). * The default mirror cannot change during recovery. */ for (m = 0; m < ms->nr_mirrors; m++) { if (&ms->mirror[m] == get_default_mirror(ms)) continue; if (test_bit(bit, &write_err)) fail_mirror(ms->mirror + m, DM_RAID1_SYNC_ERROR); bit++; } } dm_rh_recovery_end(reg, !(read_err || write_err)); } static int recover(struct mirror_set *ms, struct dm_region *reg) { int r; unsigned i; struct dm_io_region from, to[DM_KCOPYD_MAX_REGIONS], *dest; struct mirror *m; unsigned long flags = 0; region_t key = dm_rh_get_region_key(reg); sector_t region_size = dm_rh_get_region_size(ms->rh); /*