/*
* replica_check_poolset_health -- check if a given poolset can be considered as
* healthy, and store the status in a helping structure
*/
int
replica_check_poolset_health(struct pool_set *set,
struct poolset_health_status **set_hsp, unsigned flags)
{
if (replica_create_poolset_health_status(set, set_hsp)) {
LOG(1, "Creating poolset health status failed");
return -1;
}
struct poolset_health_status *set_hs = *set_hsp;
/* check if part files exist, and if not - create them, and open them */
check_and_open_poolset_part_files(set, set_hs, flags);
/* map all headers */
map_all_unbroken_headers(set, set_hs);
/* check if checksums are correct for parts in all replicas */
check_checksums(set, set_hs);
/* check if uuids in parts across each replica are consistent */
if (check_replicas_consistency(set, set_hs)) {
LOG(1, "Replica consistency check failed");
goto err;
}
/* check poolset_uuid values between replicas */
if (check_poolset_uuids(set, set_hs)) {
LOG(1, "Poolset uuids check failed");
goto err;
}
/* check if uuids for adjacent replicas are consistent */
if (check_uuids_between_replicas(set, set_hs)) {
LOG(1, "Replica uuids check failed");
goto err;
}
if (check_store_all_sizes(set, set_hs)) {
LOG(1, "Reading pool sizes failed");
goto err;
}
unmap_all_headers(set);
util_poolset_fdclose(set);
return 0;
err:
unmap_all_headers(set);
util_poolset_fdclose(set);
replica_free_poolset_health_status(set_hs);
return -1;
}
/*
* pmemlog_open_common -- (internal) open a log memory pool
*
* This routine does all the work, but takes a cow flag so internal
* calls can map a read-only pool if required.
*/
static PMEMlogpool *
pmemlog_open_common(const char *path, int cow)
{
LOG(3, "path %s cow %d", path, cow);
struct pool_set *set;
if (util_pool_open(&set, path, cow, PMEMLOG_MIN_POOL,
LOG_HDR_SIG, LOG_FORMAT_MAJOR,
LOG_FORMAT_COMPAT, LOG_FORMAT_INCOMPAT,
LOG_FORMAT_RO_COMPAT, NULL) != 0) {
LOG(2, "cannot open pool or pool set");
return NULL;
}
ASSERT(set->nreplicas > 0);
struct pool_replica *rep = set->replica[0];
PMEMlogpool *plp = rep->part[0].addr;
VALGRIND_REMOVE_PMEM_MAPPING(&plp->addr,
sizeof(struct pmemlog) -
((uintptr_t)&plp->addr - (uintptr_t)&plp->hdr));
plp->addr = plp;
plp->size = rep->repsize;
plp->set = set;
plp->is_pmem = rep->is_pmem;
plp->is_dax = rep->part[0].is_dax;
if (set->nreplicas > 1) {
errno = ENOTSUP;
ERR("!replicas not supported");
goto err;
}
/* validate pool descriptor */
if (pmemlog_descr_check(plp, rep->repsize) != 0) {
LOG(2, "descriptor check failed");
goto err;
}
/* initialize runtime parts */
if (pmemlog_runtime_init(plp, set->rdonly) != 0) {
ERR("pool initialization failed");
goto err;
}
util_poolset_fdclose(set);
LOG(3, "plp %p", plp);
return plp;
err:
LOG(4, "error clean up");
int oerrno = errno;
util_poolset_close(set, 0);
errno = oerrno;
return NULL;
}
/*
* pmemblk_open_common -- (internal) open a block memory pool
*
* This routine does all the work, but takes a cow flag so internal
* calls can map a read-only pool if required.
*
* Passing in bsize == 0 means a valid pool header must exist (which
* will supply the block size).
*/
static PMEMblkpool *
pmemblk_open_common(const char *path, size_t bsize, int cow)
{
LOG(3, "path %s bsize %zu cow %d", path, bsize, cow);
struct pool_set *set;
if (util_pool_open(&set, path, cow, PMEMBLK_MIN_POOL,
roundup(sizeof (struct pmemblk), Pagesize),
BLK_HDR_SIG, BLK_FORMAT_MAJOR,
BLK_FORMAT_COMPAT, BLK_FORMAT_INCOMPAT,
BLK_FORMAT_RO_COMPAT) != 0) {
LOG(2, "cannot open pool or pool set");
return NULL;
}
ASSERT(set->nreplicas > 0);
struct pool_replica *rep = set->replica[0];
PMEMblkpool *pbp = rep->part[0].addr;
pbp->addr = pbp;
pbp->size = rep->repsize;
VALGRIND_REMOVE_PMEM_MAPPING(&pbp->addr,
sizeof (struct pmemblk) -
((uintptr_t)&pbp->addr - (uintptr_t)&pbp->hdr));
if (set->nreplicas > 1) {
ERR("replicas not supported");
goto err;
}
/* validate pool descriptor */
if (pmemblk_descr_check(pbp, &bsize) != 0) {
LOG(2, "descriptor check failed");
goto err;
}
/* initialize runtime parts */
if (pmemblk_runtime_init(pbp, bsize, set->rdonly, rep->is_pmem) != 0) {
ERR("pool initialization failed");
goto err;
}
util_poolset_fdclose(set);
util_poolset_free(set);
LOG(3, "pbp %p", pbp);
return pbp;
err:
LOG(4, "error clean up");
int oerrno = errno;
util_poolset_close(set, 0);
errno = oerrno;
return NULL;
}
/*
* pmemlog_create -- create a log memory pool
*/
PMEMlogpool *
pmemlog_create(const char *path, size_t poolsize, mode_t mode)
{
LOG(3, "path %s poolsize %zu mode %d", path, poolsize, mode);
struct pool_set *set;
if (util_pool_create(&set, path, poolsize, PMEMLOG_MIN_POOL,
LOG_HDR_SIG, LOG_FORMAT_MAJOR,
LOG_FORMAT_COMPAT, LOG_FORMAT_INCOMPAT,
LOG_FORMAT_RO_COMPAT, NULL,
REPLICAS_DISABLED) != 0) {
LOG(2, "cannot create pool or pool set");
return NULL;
}
ASSERT(set->nreplicas > 0);
struct pool_replica *rep = set->replica[0];
PMEMlogpool *plp = rep->part[0].addr;
VALGRIND_REMOVE_PMEM_MAPPING(&plp->addr,
sizeof(struct pmemlog) -
((uintptr_t)&plp->addr - (uintptr_t)&plp->hdr));
plp->addr = plp;
plp->size = rep->repsize;
plp->set = set;
plp->is_pmem = rep->is_pmem;
plp->is_dax = rep->part[0].is_dax;
/* create pool descriptor */
if (pmemlog_descr_create(plp, rep->repsize) != 0) {
LOG(2, "descriptor creation failed");
goto err;
}
/* initialize runtime parts */
if (pmemlog_runtime_init(plp, 0) != 0) {
ERR("pool initialization failed");
goto err;
}
if (util_poolset_chmod(set, mode))
goto err;
util_poolset_fdclose(set);
LOG(3, "plp %p", plp);
return plp;
err:
LOG(4, "error clean up");
int oerrno = errno;
util_poolset_close(set, 1);
errno = oerrno;
return NULL;
}
/*
* replica_open_poolset_part_files -- open all part files for a poolset
*/
int
replica_open_poolset_part_files(struct pool_set *set)
{
for (unsigned r = 0; r < set->nreplicas; ++r) {
if (set->replica[r]->remote)
continue;
if (replica_open_replica_part_files(set, r)) {
LOG(1, "Opening replica %u, part files failed", r);
goto err;
}
}
return 0;
err:
util_poolset_fdclose(set);
return -1;
}
/*
* pmemobj_open_common -- open a transactional memory pool (set)
*
* This routine does all the work, but takes a cow flag so internal
* calls can map a read-only pool if required.
*/
static PMEMobjpool *
pmemobj_open_common(const char *path, const char *layout, int cow, int boot)
{
LOG(3, "path %s layout %s cow %d", path, layout, cow);
struct pool_set *set;
if (util_pool_open(&set, path, cow, PMEMOBJ_MIN_POOL,
roundup(sizeof (struct pmemobjpool), Pagesize),
OBJ_HDR_SIG, OBJ_FORMAT_MAJOR,
OBJ_FORMAT_COMPAT, OBJ_FORMAT_INCOMPAT,
OBJ_FORMAT_RO_COMPAT) != 0) {
LOG(2, "cannot open pool or pool set");
return NULL;
}
ASSERT(set->nreplicas > 0);
/* read-only mode is not supported in libpmemobj */
if (set->rdonly) {
ERR("read-only mode is not supported");
errno = EINVAL;
goto err;
}
PMEMobjpool *pop;
for (unsigned r = 0; r < set->nreplicas; r++) {
struct pool_replica *rep = set->replica[r];
pop = rep->part[0].addr;
VALGRIND_REMOVE_PMEM_MAPPING(&pop->addr,
sizeof (struct pmemobjpool) -
((uintptr_t)&pop->addr - (uintptr_t)&pop->hdr));
pop->addr = pop;
pop->size = rep->repsize;
if (pmemobj_descr_check(pop, layout, set->poolsize) != 0) {
LOG(2, "descriptor check failed");
goto err;
}
/* initialize replica runtime - is_pmem, funcs, ... */
if (pmemobj_replica_init(pop, rep->is_pmem) != 0) {
ERR("pool initialization failed");
goto err;
}
/* link replicas */
if (r < set->nreplicas - 1)
pop->replica = set->replica[r + 1]->part[0].addr;
}
/*
* If there is more than one replica, check if all of them are
* consistent (recoverable).
* On success, choose any replica and copy entire lanes (redo logs)
* to all the other replicas to synchronize them.
*/
if (set->nreplicas > 1) {
for (unsigned r = 0; r < set->nreplicas; r++) {
pop = set->replica[r]->part[0].addr;
if (pmemobj_check_basic(pop) == 0) {
ERR("inconsistent replica #%u", r);
goto err;
}
}
/* copy lanes */
pop = set->replica[0]->part[0].addr;
void *src = (void *)((uintptr_t)pop + pop->lanes_offset);
size_t len = pop->nlanes * sizeof (struct lane_layout);
for (unsigned r = 1; r < set->nreplicas; r++) {
pop = set->replica[r]->part[0].addr;
void *dst = (void *)((uintptr_t)pop +
pop->lanes_offset);
pop->memcpy_persist_local(dst, src, len);
}
}
pop = set->replica[0]->part[0].addr;
pop->is_master_replica = 1;
for (unsigned r = 1; r < set->nreplicas; r++) {
PMEMobjpool *rep = set->replica[r]->part[0].addr;
rep->is_master_replica = 0;
}
#ifdef USE_VG_MEMCHECK
heap_vg_open(pop);
#endif
VALGRIND_DO_CREATE_MEMPOOL(pop, 0, 0);
/* initialize runtime parts - lanes, obj stores, ... */
if (pmemobj_runtime_init(pop, 0, boot) != 0) {
ERR("pool initialization failed");
goto err;
}
util_poolset_fdclose(set);
util_poolset_free(set);
#ifdef USE_VG_MEMCHECK
if (boot)
pmemobj_vg_boot(pop);
#endif
LOG(3, "pop %p", pop);
return pop;
err:
LOG(4, "error clean up");
int oerrno = errno;
util_poolset_close(set, 0);
errno = oerrno;
return NULL;
}
/*
* pmemobj_create -- create a transactional memory pool (set)
*/
PMEMobjpool *
pmemobj_create(const char *path, const char *layout, size_t poolsize,
mode_t mode)
{
LOG(3, "path %s layout %s poolsize %zu mode %o",
path, layout, poolsize, mode);
/* check length of layout */
if (layout && (strlen(layout) >= PMEMOBJ_MAX_LAYOUT)) {
ERR("Layout too long");
errno = EINVAL;
return NULL;
}
struct pool_set *set;
if (util_pool_create(&set, path, poolsize, PMEMOBJ_MIN_POOL,
roundup(sizeof (struct pmemobjpool), Pagesize),
OBJ_HDR_SIG, OBJ_FORMAT_MAJOR,
OBJ_FORMAT_COMPAT, OBJ_FORMAT_INCOMPAT,
OBJ_FORMAT_RO_COMPAT) != 0) {
LOG(2, "cannot create pool or pool set");
return NULL;
}
ASSERT(set->nreplicas > 0);
PMEMobjpool *pop;
for (unsigned r = 0; r < set->nreplicas; r++) {
struct pool_replica *rep = set->replica[r];
pop = rep->part[0].addr;
VALGRIND_REMOVE_PMEM_MAPPING(&pop->addr,
sizeof (struct pmemobjpool) -
((uintptr_t)&pop->addr - (uintptr_t)&pop->hdr));
pop->addr = pop;
pop->size = rep->repsize;
/* create pool descriptor */
if (pmemobj_descr_create(pop, layout, set->poolsize) != 0) {
LOG(2, "descriptor creation failed");
goto err;
}
/* initialize replica runtime - is_pmem, funcs, ... */
if (pmemobj_replica_init(pop, rep->is_pmem) != 0) {
ERR("pool initialization failed");
goto err;
}
/* link replicas */
if (r < set->nreplicas - 1)
pop->replica = set->replica[r + 1]->part[0].addr;
}
pop = set->replica[0]->part[0].addr;
pop->is_master_replica = 1;
for (unsigned r = 1; r < set->nreplicas; r++) {
PMEMobjpool *rep = set->replica[r]->part[0].addr;
rep->is_master_replica = 0;
}
VALGRIND_DO_CREATE_MEMPOOL(pop, 0, 0);
/* initialize runtime parts - lanes, obj stores, ... */
if (pmemobj_runtime_init(pop, 0, 1 /* boot*/) != 0) {
ERR("pool initialization failed");
goto err;
}
if (util_poolset_chmod(set, mode))
goto err;
util_poolset_fdclose(set);
util_poolset_free(set);
LOG(3, "pop %p", pop);
return pop;
err:
LOG(4, "error clean up");
int oerrno = errno;
util_poolset_close(set, 1);
errno = oerrno;
return NULL;
}
/*
* pmemblk_create -- create a block memory pool
*/
PMEMblkpool *
pmemblk_create(const char *path, size_t bsize, size_t poolsize,
mode_t mode)
{
LOG(3, "path %s bsize %zu poolsize %zu mode %o",
path, bsize, poolsize, mode);
/* check if bsize is valid */
if (bsize == 0) {
ERR("Invalid block size %zu", bsize);
errno = EINVAL;
return NULL;
}
if (bsize > UINT32_MAX) {
ERR("Invalid block size %zu", bsize);
errno = EINVAL;
return NULL;
}
struct pool_set *set;
if (util_pool_create(&set, path, poolsize, PMEMBLK_MIN_POOL,
roundup(sizeof (struct pmemblk), Pagesize),
BLK_HDR_SIG, BLK_FORMAT_MAJOR,
BLK_FORMAT_COMPAT, BLK_FORMAT_INCOMPAT,
BLK_FORMAT_RO_COMPAT) != 0) {
LOG(2, "cannot create pool or pool set");
return NULL;
}
ASSERT(set->nreplicas > 0);
struct pool_replica *rep = set->replica[0];
PMEMblkpool *pbp = rep->part[0].addr;
pbp->addr = pbp;
pbp->size = rep->repsize;
VALGRIND_REMOVE_PMEM_MAPPING(&pbp->addr,
sizeof (struct pmemblk) -
((uintptr_t)&pbp->addr - (uintptr_t)&pbp->hdr));
if (set->nreplicas > 1) {
ERR("replicas not supported");
goto err;
}
/* create pool descriptor */
if (pmemblk_descr_create(pbp, (uint32_t)bsize, set->zeroed) != 0) {
LOG(2, "descriptor creation failed");
goto err;
}
/* initialize runtime parts */
if (pmemblk_runtime_init(pbp, bsize, 0, rep->is_pmem) != 0) {
ERR("pool initialization failed");
goto err;
}
if (util_poolset_chmod(set, mode))
goto err;
util_poolset_fdclose(set);
util_poolset_free(set);
LOG(3, "pbp %p", pbp);
return pbp;
err:
LOG(4, "error clean up");
int oerrno = errno;
util_poolset_close(set, 1);
errno = oerrno;
return NULL;
}
static inline
#endif
PMEMblkpool *
pmemblk_createU(const char *path, size_t bsize, size_t poolsize, mode_t mode)
{
LOG(3, "path %s bsize %zu poolsize %zu mode %o",
path, bsize, poolsize, mode);
/* check if bsize is valid */
if (bsize == 0) {
ERR("Invalid block size %zu", bsize);
errno = EINVAL;
return NULL;
}
if (bsize > UINT32_MAX) {
ERR("Invalid block size %zu", bsize);
errno = EINVAL;
return NULL;
}
struct pool_set *set;
if (util_pool_create(&set, path, poolsize, PMEMBLK_MIN_POOL,
BLK_HDR_SIG, BLK_FORMAT_MAJOR,
BLK_FORMAT_COMPAT, BLK_FORMAT_INCOMPAT,
BLK_FORMAT_RO_COMPAT, NULL,
REPLICAS_DISABLED) != 0) {
LOG(2, "cannot create pool or pool set");
return NULL;
}
ASSERT(set->nreplicas > 0);
struct pool_replica *rep = set->replica[0];
PMEMblkpool *pbp = rep->part[0].addr;
VALGRIND_REMOVE_PMEM_MAPPING(&pbp->addr,
sizeof(struct pmemblk) -
((uintptr_t)&pbp->addr - (uintptr_t)&pbp->hdr));
pbp->addr = pbp;
pbp->size = rep->repsize;
pbp->set = set;
pbp->is_pmem = rep->is_pmem;
pbp->is_dev_dax = rep->part[0].is_dev_dax;
/* is_dev_dax implies is_pmem */
ASSERT(!pbp->is_dev_dax || pbp->is_pmem);
/* create pool descriptor */
blk_descr_create(pbp, (uint32_t)bsize, set->zeroed);
/* initialize runtime parts */
if (blk_runtime_init(pbp, bsize, 0) != 0) {
ERR("pool initialization failed");
goto err;
}
if (util_poolset_chmod(set, mode))
goto err;
util_poolset_fdclose(set);
LOG(3, "pbp %p", pbp);
return pbp;
err:
LOG(4, "error clean up");
int oerrno = errno;
util_poolset_close(set, DELETE_CREATED_PARTS);
errno = oerrno;
return NULL;
}