brassow This patch contains a cluster-aware log module. When used by dm-raid1, device-mapper mirroring can be cluster-aware. There is a kernel component (provided in this patch) and a user space component. The kernel component implements the logging interface and passes all requests to userspace via 'connector' (a netlink wrapper). The userspace daemon is built upon OpenAIS for cluster communication and is fault tolerant. The logging interface code (found in dm-log-cluster.c) is fairly simple and should be close to completion. I would really enjoy feedback on: 1) Connector code (found in dm-log-cluster-transfer.c) 2) Suggestions on possible break-ups of dm-log-cluster-transfer.h, since some of that code is meant to be used by both userspace and kernel space, and some is meant only for kernel space... (I think the code meant for both should go in include/linux, while the code just meant for kernel space should go in drivers/md... yes?) Looking-for-review: Jonathan Brassow --- drivers/md/Kconfig | 9 drivers/md/Makefile | 2 drivers/md/dm-log-cluster-transfer.c | 270 ++++++++++++ drivers/md/dm-log-cluster-transfer.h | 75 +++ drivers/md/dm-log-cluster.c | 779 +++++++++++++++++++++++++++++++++++ 5 files changed, 1135 insertions(+) Index: linux/drivers/md/Kconfig =================================================================== --- linux.orig/drivers/md/Kconfig 2008-08-04 14:38:30.000000000 +0100 +++ linux/drivers/md/Kconfig 2008-08-04 14:40:22.000000000 +0100 @@ -242,6 +242,15 @@ config DM_MIRROR Allow volume managers to mirror logical volumes, also needed for live data migration tools such as 'pvmove'. +config DM_LOG_CLUSTERED + tristate "Mirror cluster logging (EXPERIMENTAL)" + depends on DM_MIRROR && EXPERIMENTAL + ---help--- + Cluster logging allows device-mapper mirroring to be + cluster-aware. Mirror devices can be used by multiple + machines at the same time. Note: this will not make + your applications cluster-aware. + config DM_ZERO tristate "Zero target" depends on BLK_DEV_DM Index: linux/drivers/md/Makefile =================================================================== --- linux.orig/drivers/md/Makefile 2008-08-04 14:38:30.000000000 +0100 +++ linux/drivers/md/Makefile 2008-08-04 14:40:22.000000000 +0100 @@ -7,6 +7,7 @@ dm-mod-objs := dm.o dm-table.o dm-target dm-multipath-objs := dm-path-selector.o dm-mpath.o dm-snapshot-objs := dm-snap.o dm-exception-store.o dm-mirror-objs := dm-raid1.o +dm-log-clustered-objs := dm-log-cluster.o dm-log-cluster-transfer.o md-mod-objs := md.o bitmap.o raid456-objs := raid5.o raid6algos.o raid6recov.o raid6tables.o \ raid6int1.o raid6int2.o raid6int4.o \ @@ -35,6 +36,7 @@ obj-$(CONFIG_DM_DELAY) += dm-delay.o obj-$(CONFIG_DM_MULTIPATH) += dm-multipath.o dm-round-robin.o obj-$(CONFIG_DM_SNAPSHOT) += dm-snapshot.o obj-$(CONFIG_DM_MIRROR) += dm-mirror.o dm-log.o +obj-$(CONFIG_DM_LOG_CLUSTERED) += dm-log-clustered.o obj-$(CONFIG_DM_ZERO) += dm-zero.o quiet_cmd_unroll = UNROLL $@ Index: linux/drivers/md/dm-log-cluster-transfer.c =================================================================== --- /dev/null 1970-01-01 00:00:00.000000000 +0000 +++ linux/drivers/md/dm-log-cluster-transfer.c 2008-08-04 14:40:22.000000000 +0100 @@ -0,0 +1,270 @@ +/* + * Copyright (C) 2006 Red Hat, Inc. + * + * This file is released under the LGPL. + */ + +#include +#include +#include +#include +#include +#include "dm.h" +#include "dm-log-cluster-transfer.h" + +#include /* Unnecessary */ + +#define SHORT_UUID(x) (strlen(x) > 8) ? ((x) + (strlen(x) - 8)) : (x) + +static uint32_t seq; + +/* + * Pre-allocated space for speed + */ +#define DM_CLOG_PREALLOCED_SIZE 512 +static struct cn_msg *prealloced_cn_msg; +static struct clog_tfr *prealloced_clog_tfr; + +static struct cb_id cn_clog_id = { 0x4, 0x1 }; +static DEFINE_MUTEX(_lock); + +struct receiving_pkg { + struct list_head list; + struct completion complete; + + unsigned long long start_time; + uint32_t seq; + + int error; + int *data_size; + char *data; +}; + +static DEFINE_SPINLOCK(receiving_list_lock); +static struct list_head receiving_list; + +static int dm_clog_sendto_server(struct clog_tfr *tfr) +{ + int r; + int size; + struct cn_msg *msg = prealloced_cn_msg; + + if (tfr != prealloced_clog_tfr) { + size = sizeof(struct cn_msg) + + sizeof(struct clog_tfr) + tfr->data_size; + msg = kmalloc(size, GFP_NOIO); + if (!msg) + return -ENOMEM; + memcpy((msg + 1), tfr, + sizeof(struct clog_tfr) + tfr->data_size); + } + + memset(msg, 0, sizeof(struct cn_msg)); + + msg->id.idx = cn_clog_id.idx; + msg->id.val = cn_clog_id.val; + msg->ack = 0; + msg->seq = tfr->seq; + msg->len = sizeof(struct clog_tfr) + tfr->data_size; + + r = cn_netlink_send(msg, 0, gfp_any()); + + if (msg != prealloced_cn_msg) + kfree(msg); + + return r; +} + +/* + * fill_pkg + * @msg + * @tfr + * + * Parameters can be either msg or tfr, but not both. This + * function fills in the reply for a waiting request. If just + * msg is given, then the reply is simply an ACK from userspace + * that the request was received. + * + * Returns: 0 on success, -ENOENT on failure + */ +static int fill_pkg(struct cn_msg *msg, struct clog_tfr *tfr) +{ + uint32_t rtn_seq = (msg) ? msg->seq : (tfr) ? tfr->seq : 0; + struct receiving_pkg *pkg; + + list_for_each_entry(pkg, &receiving_list, list) { + if (rtn_seq != pkg->seq) + continue; + + if (msg) { + pkg->error = -msg->ack; + } else if (tfr->data_size > *(pkg->data_size)) { + DMERR("Insufficient space to receive package [%s]", + RQ_TYPE(tfr->request_type)); + *(pkg->data_size) = 0; + pkg->error = -ENOSPC; + } else { + pkg->error = tfr->error; + memcpy(pkg->data, tfr->data, tfr->data_size); + *(pkg->data_size) = tfr->data_size; + } + complete(&pkg->complete); + return 0; + } + + return -ENOENT; +} + +/* + * cn_clog_callback + * @data + * + * This is the connector callback that delivers data + * that was sent from userspace. + */ +static void cn_clog_callback(void *data) +{ + struct cn_msg *msg = (struct cn_msg *)data; + struct clog_tfr *tfr = (struct clog_tfr *)(msg + 1); + + spin_lock(&receiving_list_lock); + if (msg->len == 0) + fill_pkg(msg, NULL); + else if (msg->len < sizeof(*tfr)) + DMERR("Incomplete message received: [%u]", msg->seq); + else + fill_pkg(NULL, tfr); + spin_unlock(&receiving_list_lock); +} + +/* + * dm_clog_consult_server + * @uuid: log's uuid (must be DM_UUID_LEN in size) + * @request_type: + * @data: data to tx to the server + * @data_size: size of data in bytes + * @rdata: place to put return data from server + * @rdata_size: value-result (amount of space given/amount of space used) + * + * Only one process at a time can communicate with the server. + * rdata_size is undefined on failure. + * + * Returns: 0 on success, -EXXX on failure + */ +int dm_clog_consult_server(const char *uuid, int request_type, + char *data, int data_size, + char *rdata, int *rdata_size) +{ + int r = 0; + int dummy = 0; + int overhead_size = sizeof(struct clog_tfr *) + sizeof(struct cn_msg); + struct clog_tfr *tfr = prealloced_clog_tfr; + struct receiving_pkg pkg; + + if (data_size > (DM_CLOG_PREALLOCED_SIZE - overhead_size)) { + DMINFO("Size of tfr exceeds preallocated size"); + /* FIXME: is kmalloc sufficient if we need this much space? */ + tfr = kzalloc(data_size + sizeof(*tfr), GFP_NOIO); + } + + if (!tfr) + return -ENOMEM; + + if (!rdata_size) + rdata_size = &dummy; +resend: + /* + * We serialize the sending of requests so we can + * use the preallocated space. + */ + mutex_lock(&_lock); + + memset(tfr, 0, DM_CLOG_PREALLOCED_SIZE - overhead_size); + memcpy(tfr->uuid, uuid, DM_UUID_LEN); + tfr->seq = seq++; + tfr->request_type = request_type; + tfr->data_size = data_size; + if (data && data_size) + memcpy(tfr->data, data, data_size); + + memset(&pkg, 0, sizeof(pkg)); + init_completion(&pkg.complete); + pkg.seq = tfr->seq; + pkg.data_size = rdata_size; + pkg.data = rdata; + spin_lock(&receiving_list_lock); + list_add(&(pkg.list), &receiving_list); + spin_unlock(&receiving_list_lock); + pkg.start_time = jiffies; + + r = dm_clog_sendto_server(tfr); + + mutex_unlock(&_lock); + + if (r) { + DMERR("Unable to send cluster log request [%s] to server: %d", + RQ_TYPE(request_type), r); + spin_lock(&receiving_list_lock); + list_del_init(&(pkg.list)); + spin_unlock(&receiving_list_lock); + + goto out; + } + + r = wait_for_completion_timeout(&(pkg.complete), 15 * HZ); + spin_lock(&receiving_list_lock); + list_del_init(&(pkg.list)); + spin_unlock(&receiving_list_lock); + if (!r) { + DMWARN("[%s] Request timed out: [%s/%u] - retrying", + SHORT_UUID(uuid), RQ_TYPE(request_type), pkg.seq); + goto resend; + } + + /* FIXME: Pull time checking code - it's just for profiling */ + pkg.start_time = (jiffies - pkg.start_time); + do_div(pkg.start_time, HZ); + if (pkg.start_time > 0) + DMWARN("Excessive delay in request processing, %llu sec: " + "[%s/%u]", pkg.start_time, RQ_TYPE(request_type), + pkg.seq); + + r = pkg.error; + if (r == -EAGAIN) + goto resend; + +out: + if (tfr != (struct clog_tfr *)prealloced_clog_tfr) + kfree(tfr); + + return r; +} + +int dm_clog_tfr_init(void) +{ + int r; + void *prealloced; + + INIT_LIST_HEAD(&receiving_list); + + prealloced = kmalloc(DM_CLOG_PREALLOCED_SIZE, GFP_KERNEL); + if (!prealloced) + return -ENOMEM; + + prealloced_cn_msg = prealloced; + prealloced_clog_tfr = prealloced + sizeof(struct cn_msg); + + r = cn_add_callback(&cn_clog_id, "clulog", cn_clog_callback); + if (r) { + cn_del_callback(&cn_clog_id); + return r; + } + + return 0; +} + +void dm_clog_tfr_exit(void) +{ + cn_del_callback(&cn_clog_id); + kfree(prealloced_cn_msg); +} Index: linux/drivers/md/dm-log-cluster-transfer.h =================================================================== --- /dev/null 1970-01-01 00:00:00.000000000 +0000 +++ linux/drivers/md/dm-log-cluster-transfer.h 2008-08-04 14:40:22.000000000 +0100 @@ -0,0 +1,75 @@ +/* + * Copyright (C) 2006 Red Hat, Inc. + * + * This file is released under the LGPL. + */ + +#ifndef __DM_CLOG_TFR_H__ +#define __DM_CLOG_TFR_H__ + +#include /* For DM_UUID_LEN */ + +#define DM_CLOG_TFR_SIZE 1024 + +#define DM_CLOG_CTR 1 +#define DM_CLOG_DTR 2 +#define DM_CLOG_PRESUSPEND 3 +#define DM_CLOG_POSTSUSPEND 4 +#define DM_CLOG_RESUME 5 +#define DM_CLOG_GET_REGION_SIZE 6 +#define DM_CLOG_IS_CLEAN 7 +#define DM_CLOG_IN_SYNC 8 +#define DM_CLOG_FLUSH 9 +#define DM_CLOG_MARK_REGION 10 +#define DM_CLOG_CLEAR_REGION 11 +#define DM_CLOG_GET_RESYNC_WORK 12 +#define DM_CLOG_SET_REGION_SYNC 13 +#define DM_CLOG_GET_SYNC_COUNT 14 +#define DM_CLOG_STATUS_INFO 15 +#define DM_CLOG_STATUS_TABLE 16 +#define DM_CLOG_IS_REMOTE_RECOVERING 17 + +#define RQ_TYPE(x) \ + ((x) == DM_CLOG_CTR) ? "DM_CLOG_CTR" : \ + ((x) == DM_CLOG_DTR) ? "DM_CLOG_DTR" : \ + ((x) == DM_CLOG_PRESUSPEND) ? "DM_CLOG_PRESUSPEND" : \ + ((x) == DM_CLOG_POSTSUSPEND) ? "DM_CLOG_POSTSUSPEND" : \ + ((x) == DM_CLOG_RESUME) ? "DM_CLOG_RESUME" : \ + ((x) == DM_CLOG_GET_REGION_SIZE) ? "DM_CLOG_GET_REGION_SIZE" : \ + ((x) == DM_CLOG_IS_CLEAN) ? "DM_CLOG_IS_CLEAN" : \ + ((x) == DM_CLOG_IN_SYNC) ? "DM_CLOG_IN_SYNC" : \ + ((x) == DM_CLOG_FLUSH) ? "DM_CLOG_FLUSH" : \ + ((x) == DM_CLOG_MARK_REGION) ? "DM_CLOG_MARK_REGION" : \ + ((x) == DM_CLOG_CLEAR_REGION) ? "DM_CLOG_CLEAR_REGION" : \ + ((x) == DM_CLOG_GET_RESYNC_WORK) ? "DM_CLOG_GET_RESYNC_WORK" : \ + ((x) == DM_CLOG_SET_REGION_SYNC) ? "DM_CLOG_SET_REGION_SYNC" : \ + ((x) == DM_CLOG_GET_SYNC_COUNT) ? "DM_CLOG_GET_SYNC_COUNT" : \ + ((x) == DM_CLOG_STATUS_INFO) ? "DM_CLOG_STATUS_INFO" : \ + ((x) == DM_CLOG_STATUS_TABLE) ? "DM_CLOG_STATUS_TABLE" : \ + ((x) == DM_CLOG_IS_REMOTE_RECOVERING) ? \ + "DM_CLOG_IS_REMOTE_RECOVERING" : NULL + +struct clog_tfr { + uint64_t private[2]; + char uuid[DM_UUID_LEN]; /* Ties a request to a specific mirror log */ + + int error; /* Used by server to inform of errors */ + uint32_t originator; /* Cluster ID of this machine */ + + uint32_t seq; /* Sequence number for request */ + uint32_t request_type; /* DM_CLOG_* */ + uint32_t data_size; /* How much data (not including this struct) */ + char data[0]; +}; + +#ifdef __KERNEL__ +#define DM_MSG_PREFIX "dm-log-clustered" + +int dm_clog_tfr_init(void); +void dm_clog_tfr_exit(void); +int dm_clog_consult_server(const char *uuid, int request_type, + char *data, int data_size, + char *rdata, int *rdata_size); +#endif + +#endif /* __DM_CLOG_TFR_H__ */ Index: linux/drivers/md/dm-log-cluster.c =================================================================== --- /dev/null 1970-01-01 00:00:00.000000000 +0000 +++ linux/drivers/md/dm-log-cluster.c 2008-08-04 14:40:22.000000000 +0100 @@ -0,0 +1,779 @@ +/* + * Copyright (C) 2006 Red Hat, Inc. + * + * This file is released under the LGPL. + */ + +#include "dm.h" +#include "dm-log-cluster-transfer.h" +#include + +struct flush_entry { + int type; + region_t region; + struct list_head list; +}; + +struct log_c { + struct dm_target *ti; + uint32_t region_size; + region_t region_count; + char uuid[DM_UUID_LEN]; + + char *ctr_str; /* Gives ability to restart if userspace dies */ + uint32_t ctr_size; + + /* + * in_sync_hint gets set when doing is_remote_recovering. It + * represents the first region that needs recovery. IOW, the + * first zero bit of sync_bits. This can be useful for to limit + * traffic for calls like is_remote_recovering and get_resync_work, + * but be take care in its use for anything else. + */ + uint64_t in_sync_hint; + + spinlock_t flush_lock; + struct list_head flush_list; /* only for clear and mark requests */ + + struct dm_dev *disk_log; +}; + +static mempool_t *flush_entry_pool; + +static void *flush_entry_alloc(gfp_t gfp_mask, void *pool_data) +{ + return kmalloc(sizeof(struct flush_entry), gfp_mask); +} + +static void flush_entry_free(void *element, void *pool_data) +{ + kfree(element); +} + +int cluster_do_request(struct log_c *lc, const char *uuid, int request_type, + char *data, int data_size, char *rdata, int *rdata_size) +{ + int r; + + /* + * If the server isn't there, -ESRCH is returned, + * and we must keep trying until the server is + * restored. + */ +retry: + r = dm_clog_consult_server(uuid, request_type, data, + data_size, rdata, rdata_size); + + if (r != -ESRCH) + return r; + + DMERR(" Userspace cluster log server not found."); + while (1) { + set_current_state(TASK_INTERRUPTIBLE); + schedule_timeout(2*HZ); + DMWARN("Attempting to contact cluster log server..."); + r = dm_clog_consult_server(uuid, DM_CLOG_CTR, lc->ctr_str, + lc->ctr_size, NULL, NULL); + if (!r) + break; + } + DMINFO("Reconnected to cluster log server... CTR complete"); + r = dm_clog_consult_server(uuid, DM_CLOG_RESUME, NULL, + 0, NULL, NULL); + if (!r) + goto retry; + + DMERR("Error trying to resume cluster log: %d", r); + + return -ESRCH; +} + +static int cluster_ctr(struct dm_dirty_log *log, struct dm_target *ti, + unsigned int argc, char **argv, + struct dm_dev *disk_log) +{ + int i; + int r = 0; + int str_size; + int offset = (disk_log) ? 1 : 0; + char *ctr_str = NULL; + struct log_c *lc = NULL; + uint32_t region_size; + region_t region_count; + + /* Already checked argument count */ + + if (sscanf(argv[offset], "%u", ®ion_size) != 1) { + DMWARN("Invalid region size string"); + return -EINVAL; + } + + region_count = dm_sector_div_up(ti->len, region_size); + + lc = kmalloc(sizeof(*lc), GFP_KERNEL); + if (!lc) { + DMWARN("Unable to allocate cluster log context."); + return -ENOMEM; + } + + lc->ti = ti; + lc->region_size = region_size; + lc->region_count = region_count; + lc->disk_log = disk_log; + + /* FIXME: Need to check size of uuid arg */ + memcpy(lc->uuid, argv[1 + offset], DM_UUID_LEN); + spin_lock_init(&lc->flush_lock); + INIT_LIST_HEAD(&lc->flush_list); + + for (i = 0, str_size = 0; i < argc; i++) + str_size += strlen(argv[i]) + 1; /* +1 for space between args */ + + str_size += 20; /* Max number of chars in a printed u64 number */ + + ctr_str = kzalloc(str_size, GFP_KERNEL); + if (!ctr_str) { + DMWARN("Unable to allocate memory for constructor string"); + kfree(lc); + return -ENOMEM; + } + + for (i = 0, str_size = 0; i < argc; i++) + str_size += sprintf(ctr_str + str_size, "%s ", argv[i]); + str_size += sprintf(ctr_str + str_size, "%llu", + (long long unsigned int)ti->len); + + /* Send table string */ + r = dm_clog_consult_server(lc->uuid, DM_CLOG_CTR, + ctr_str, str_size, NULL, NULL); + + if (r == -ESRCH) + DMERR(" Userspace cluster log server not found"); + + if (r) { + kfree(lc); + kfree(ctr_str); + } else { + lc->ctr_str = ctr_str; + lc->ctr_size = str_size; + log->context = lc; + } + + return r; +} + +/* + * cluster_core_ctr + * @log + * @ti + * @argc + * @argv + * + * argv contains: + * [[no]sync] + * + * Returns: 0 on success, -XXX on failure + */ +static int cluster_core_ctr(struct dm_dirty_log *log, struct dm_target *ti, + unsigned int argc, char **argv) +{ + int i, r; + if ((argc < 2) || (argc > 3)) { + DMERR("Too %s arguments to clustered-core mirror log type.", + (argc < 2) ? "few" : "many"); + DMERR(" %d arguments supplied:", argc); + for (i = 0; i < argc; i++) + DMERR(" %s", argv[i]); + return -EINVAL; + } + + r = cluster_ctr(log, ti, argc, argv, NULL); + + return r; +} + + +/* + * cluster_core_ctr + * @log + * @ti + * @argc + * @argv + * + * argv contains: + * [[no]sync] + * + * Returns: 0 on success, -XXX on failure + */ +static int cluster_disk_ctr(struct dm_dirty_log *log, struct dm_target *ti, + unsigned int argc, char **argv) +{ + int r, i; + struct dm_dev *dev; + + if ((argc < 3) || (argc > 4)) { + DMERR("Too %s arguments to clustered-disk mirror log type.", + (argc < 3) ? "few" : "many"); + DMERR(" %d arguments supplied:", argc); + for (i = 0; i < argc; i++) + DMERR(" %s", argv[i]); + return -EINVAL; + } + + r = dm_get_device(ti, argv[0], 0, 0, FMODE_READ | FMODE_WRITE, &dev); + if (r) + return r; + + r = cluster_ctr(log, ti, argc, argv, dev); + if (r) + dm_put_device(ti, dev); + + return r; +} + +/* + * cluster_dtr + * @log + */ +static void cluster_dtr(struct dm_dirty_log *log) +{ + int r; + struct log_c *lc = (struct log_c *)log->context; + + r = dm_clog_consult_server(lc->uuid, DM_CLOG_DTR, + NULL, 0, + NULL, NULL); + + if (lc->disk_log) + dm_put_device(lc->ti, lc->disk_log); + kfree(lc->ctr_str); + kfree(lc); + + return; +} + +/* + * cluster_presuspend + * @log + */ +static int cluster_presuspend(struct dm_dirty_log *log) +{ + int r; + struct log_c *lc = (struct log_c *)log->context; + + r = dm_clog_consult_server(lc->uuid, DM_CLOG_PRESUSPEND, + NULL, 0, + NULL, NULL); + + return r; +} + +/* + * cluster_postsuspend + * @log + */ +static int cluster_postsuspend(struct dm_dirty_log *log) +{ + int r; + struct log_c *lc = (struct log_c *)log->context; + + r = dm_clog_consult_server(lc->uuid, DM_CLOG_POSTSUSPEND, + NULL, 0, + NULL, NULL); + + return r; +} + +/* + * cluster_resume + * @log + */ +static int cluster_resume(struct dm_dirty_log *log) +{ + int r; + struct log_c *lc = (struct log_c *)log->context; + + lc->in_sync_hint = 0; + r = dm_clog_consult_server(lc->uuid, DM_CLOG_RESUME, + NULL, 0, + NULL, NULL); + + return r; +} + +/* + * cluster_get_region_size + * @log + * + * Only called during mirror construction, ok to block. + * + * Returns: region size (doesn't fail) + */ +static uint32_t cluster_get_region_size(struct dm_dirty_log *log) +{ + struct log_c *lc = (struct log_c *)log->context; + + return lc->region_size; +} + +/* + * cluster_is_clean + * @log + * @region + * + * Check whether a region is clean. If there is any sort of + * failure when consulting the server, we return not clean. + * + * Returns: 1 if clean, 0 otherwise + */ +static int cluster_is_clean(struct dm_dirty_log *log, region_t region) +{ + int r; + int is_clean; + int rdata_size; + struct log_c *lc = (struct log_c *)log->context; + + rdata_size = sizeof(is_clean); + r = cluster_do_request(lc, lc->uuid, DM_CLOG_IS_CLEAN, + (char *)®ion, sizeof(region), + (char *)&is_clean, &rdata_size); + + return (r) ? 0 : is_clean; +} + +/* + * cluster_in_sync + * @log + * @region + * @can_block: if set, return immediately + * + * Check if the region is in-sync. If there is any sort + * of failure when consulting the server, we assume that + * the region is not in sync. + * + * Returns: 1 if in-sync, 0 if not-in-sync, -EWOULDBLOCK + */ +static int cluster_in_sync(struct dm_dirty_log *log, region_t region, + int can_block) +{ + int r; + int in_sync; + int rdata_size; + struct log_c *lc = (struct log_c *)log->context; + + /* + * We can never respond directly - even if in_sync_hint is + * set. This is because another machine could see a device + * failure and mark the region out-of-sync. If we don't go + * to userspace to ask, we might think the region is in-sync + * and allow a read to pick up data that is stale. (This is + * very unlikely if a device actually fails; but it is very + * likely if a connection to one device from one machine fails.) + * + * There still might be a problem if the mirror caches the region + * state as in-sync... but then this call would not be made. So, + * that is a mirror problem. + */ + if (!can_block) + return -EWOULDBLOCK; + + rdata_size = sizeof(in_sync); + r = cluster_do_request(lc, lc->uuid, DM_CLOG_IN_SYNC, + (char *)®ion, sizeof(region), + (char *)&in_sync, &rdata_size); + return (r) ? 0 : in_sync; +} + +/* + * cluster_flush + * @log + * + * This function is ok to block. + * The flush happens in two stages. First, it sends all + * clear/mark requests that are on the list. Then it + * tells the server to commit them. This gives the + * server a chance to optimise the commit to the cluster + * and/or disk, instead of doing it for every request. + * + * Additionally, we could implement another thread that + * sends the requests up to the server - reducing the + * load on flush. Then the flush would have less in + * the list and be responsible for the finishing commit. + * + * Returns: 0 on success, < 0 on failure + */ +static int cluster_flush(struct dm_dirty_log *log) +{ + int r = 0; + int flags; + struct log_c *lc = (struct log_c *)log->context; + LIST_HEAD(flush_list); + struct flush_entry *fe, *tmp_fe; + + spin_lock_irqsave(&lc->flush_lock, flags); + list_splice_init(&lc->flush_list, &flush_list); + spin_unlock_irqrestore(&lc->flush_lock, flags); + + if (list_empty(&flush_list)) + return 0; + + /* + * FIXME: Count up requests, group request types, + * allocate memory to stick all requests in and + * send to server in one go. Failing the allocation, + * do it one by one. + */ + + list_for_each_entry(fe, &flush_list, list) { + r = cluster_do_request(lc, lc->uuid, fe->type, + (char *)&fe->region, + sizeof(fe->region), + NULL, NULL); + if (r) + goto fail; + } + + r = cluster_do_request(lc, lc->uuid, DM_CLOG_FLUSH, + NULL, 0, NULL, NULL); + +fail: + /* + * We can safely remove these entries, even if failure. + * Calling code will receive an error and will know that + * the log facility has failed. + */ + list_for_each_entry_safe(fe, tmp_fe, &flush_list, list) { + list_del(&fe->list); + mempool_free(fe, flush_entry_pool); + } + + if (r) + dm_table_event(lc->ti->table); + + return r; +} + +/* + * cluster_mark_region + * @log + * @region + * + * This function should avoid blocking unless absolutely required. + * (Memory allocation is valid for blocking.) + */ +static void cluster_mark_region(struct dm_dirty_log *log, region_t region) +{ + int flags; + struct log_c *lc = (struct log_c *)log->context; + struct flush_entry *fe; + + /* Wait for an allocation, but _never_ fail */ + fe = mempool_alloc(flush_entry_pool, GFP_NOIO); + BUG_ON(!fe); + + spin_lock_irqsave(&lc->flush_lock, flags); + fe->type = DM_CLOG_MARK_REGION; + fe->region = region; + list_add(&fe->list, &lc->flush_list); + spin_unlock_irqrestore(&lc->flush_lock, flags); + + return; +} + +/* + * cluster_clear_region + * @log + * @region + * + * This function must not block. + * So, the alloc can't block. In the worst case, it is ok to + * fail. It would simply mean we can't clear the region. + * Does nothing to current sync context, but does mean + * the region will be re-sync'ed on a reload of the mirror + * even though it is in-sync. + */ +static void cluster_clear_region(struct dm_dirty_log *log, region_t region) +{ + int flags; + struct log_c *lc = (struct log_c *)log->context; + struct flush_entry *fe; + + /* + * If we fail to allocate, we skip the clearing of + * the region. This doesn't hurt us in any way, except + * to cause the region to be resync'ed when the + * device is activated next time. + */ + fe = mempool_alloc(flush_entry_pool, GFP_ATOMIC); + if (!fe) { + DMERR("Failed to allocate memory to clear region."); + return; + } + + spin_lock_irqsave(&lc->flush_lock, flags); + fe->type = DM_CLOG_CLEAR_REGION; + fe->region = region; + list_add(&fe->list, &lc->flush_list); + spin_unlock_irqrestore(&lc->flush_lock, flags); + + return; +} + +/* + * cluster_get_resync_work + * @log + * @region + * + * Get a region that needs recovery. It is valid to return + * an error for this function. + * + * Returns: 1 if region filled, 0 if no work, <0 on error + */ +static int cluster_get_resync_work(struct dm_dirty_log *log, region_t *region) +{ + int r; + int rdata_size; + struct log_c *lc = (struct log_c *)log->context; + struct { int i; region_t r; } pkg; + + if (lc->in_sync_hint >= lc->region_count) + return 0; + + rdata_size = sizeof(pkg); + r = cluster_do_request(lc, lc->uuid, DM_CLOG_GET_RESYNC_WORK, + NULL, 0, + (char *)&pkg, &rdata_size); + + *region = pkg.r; + return (r) ? r : pkg.i; +} + +/* + * cluster_set_region_sync + * @log + * @region + * @in_sync + * + * Set the sync status of a given region. This function + * must not fail. + */ +static void cluster_set_region_sync(struct dm_dirty_log *log, + region_t region, int in_sync) +{ + int r; + struct log_c *lc = (struct log_c *)log->context; + struct { region_t r; int i; } pkg; + + pkg.r = region; + pkg.i = in_sync; + + r = cluster_do_request(lc, lc->uuid, DM_CLOG_SET_REGION_SYNC, + (char *)&pkg, sizeof(pkg), + NULL, NULL); + + return; +} + +/* + * cluster_get_sync_count + * @log + * + * If there is any sort of failure when consulting the server, + * we assume that the sync count is zero. + * + * Returns: sync count on success, 0 on failure + */ +static region_t cluster_get_sync_count(struct dm_dirty_log *log) +{ + int r; + int rdata_size; + region_t sync_count; + struct log_c *lc = (struct log_c *)log->context; + + rdata_size = sizeof(sync_count); + r = cluster_do_request(lc, lc->uuid, DM_CLOG_GET_SYNC_COUNT, + NULL, 0, + (char *)&sync_count, &rdata_size); + + if (r) + return 0; + + if (sync_count >= lc->region_count) + lc->in_sync_hint = lc->region_count; + + return sync_count; +} + +/* + * cluster_status + * @log + * @status_type + * @result + * @maxlen + * + * Returns: amount of space consumed + */ +static int cluster_status(struct dm_dirty_log *log, status_type_t status_type, + char *result, unsigned int maxlen) +{ + int r = 0; + unsigned int sz = maxlen; + struct log_c *lc = (struct log_c *)log->context; + + switch (status_type) { + case STATUSTYPE_INFO: + r = cluster_do_request(lc, lc->uuid, DM_CLOG_STATUS_INFO, + NULL, 0, + result, &sz); + /* + * FIXME: If we fail to contact server, we should still + * populate this with parsible results + */ + break; + case STATUSTYPE_TABLE: + r = cluster_do_request(lc, lc->uuid, DM_CLOG_STATUS_TABLE, + NULL, 0, + result, &sz); + break; + } + return (r) ? 0: sz; +} + +/* + * cluster_is_remote_recovering + * @log + * @region + * + * Returns: 1 if region recovering, 0 otherwise + */ +static int cluster_is_remote_recovering(struct dm_dirty_log *log, + region_t region) +{ + int r; + struct log_c *lc = (struct log_c *)log->context; + static unsigned long long limit; + struct { int is_recovering; uint64_t sync_search; } pkg; + int rdata_size = sizeof(pkg); + + /* + * Once the mirror has been reported to be in-sync, + * it will never again ask for recovery work. So, + * we can safely say there is not a remote machine + * recovering if the device is in-sync. (in_sync_hint + * must be reset at resume time.) + */ + if (region < lc->in_sync_hint) + return 0; + else if (jiffies < limit) + return 1; + + limit = jiffies + (HZ / 4); + r = cluster_do_request(lc, lc->uuid, DM_CLOG_IS_REMOTE_RECOVERING, + (char *)®ion, sizeof(region), + (char *)&pkg, &rdata_size); + if (r) + return 1; + + lc->in_sync_hint = pkg.sync_search; + + return pkg.is_recovering; +} + +static struct dm_dirty_log_type _clustered_core_type = { + .name = "clustered-core", + .module = THIS_MODULE, + .ctr = cluster_core_ctr, + .dtr = cluster_dtr, + .presuspend = cluster_presuspend, + .postsuspend = cluster_postsuspend, + .resume = cluster_resume, + .get_region_size = cluster_get_region_size, + .is_clean = cluster_is_clean, + .in_sync = cluster_in_sync, + .flush = cluster_flush, + .mark_region = cluster_mark_region, + .clear_region = cluster_clear_region, + .get_resync_work = cluster_get_resync_work, + .set_region_sync = cluster_set_region_sync, + .get_sync_count = cluster_get_sync_count, + .status = cluster_status, + .is_remote_recovering = cluster_is_remote_recovering, +}; + +static struct dm_dirty_log_type _clustered_disk_type = { + .name = "clustered-disk", + .module = THIS_MODULE, + .ctr = cluster_disk_ctr, + .dtr = cluster_dtr, + .presuspend = cluster_presuspend, + .postsuspend = cluster_postsuspend, + .resume = cluster_resume, + .get_region_size = cluster_get_region_size, + .is_clean = cluster_is_clean, + .in_sync = cluster_in_sync, + .flush = cluster_flush, + .mark_region = cluster_mark_region, + .clear_region = cluster_clear_region, + .get_resync_work = cluster_get_resync_work, + .set_region_sync = cluster_set_region_sync, + .get_sync_count = cluster_get_sync_count, + .status = cluster_status, + .is_remote_recovering = cluster_is_remote_recovering, +}; + +static int __init cluster_dirty_log_init(void) +{ + int r = 0; + + flush_entry_pool = mempool_create(100, flush_entry_alloc, + flush_entry_free, NULL); + + if (!flush_entry_pool) { + DMWARN("Unable to create flush_entry_pool: No memory."); + return -ENOMEM; + } + + r = dm_clog_tfr_init(); + if (r) { + DMWARN("Unable to initialize cluster log communications"); + mempool_destroy(flush_entry_pool); + return r; + } + + r = dm_dirty_log_type_register(&_clustered_core_type); + if (r) { + DMWARN("Couldn't register clustered-core dirty log type"); + dm_clog_tfr_exit(); + mempool_destroy(flush_entry_pool); + return r; + } + + r = dm_dirty_log_type_register(&_clustered_disk_type); + if (r) { + DMWARN("Couldn't register clustered-disk dirty log type"); + dm_dirty_log_type_unregister(&_clustered_core_type); + dm_clog_tfr_exit(); + mempool_destroy(flush_entry_pool); + return r; + } + + DMINFO("(built %s %s) installed", __DATE__, __TIME__); + return 0; +} + +static void __exit cluster_dirty_log_exit(void) +{ + dm_dirty_log_type_unregister(&_clustered_disk_type); + dm_dirty_log_type_unregister(&_clustered_core_type); + dm_clog_tfr_exit(); + mempool_destroy(flush_entry_pool); + DMINFO("(built %s %s) removed", __DATE__, __TIME__); + return; +} + +module_init(cluster_dirty_log_init); +module_exit(cluster_dirty_log_exit); + +MODULE_DESCRIPTION(DM_NAME " mirror cluster-aware log"); +MODULE_AUTHOR("Jonathan Brassow"); +MODULE_LICENSE("GPL");