int
main(int argc, char *argv[])
{
START(argc, argv, "util_poolset");
common_init(LOG_PREFIX, LOG_LEVEL_VAR, LOG_FILE_VAR,
MAJOR_VERSION, MINOR_VERSION);
if (argc < 3)
UT_FATAL("usage: %s cmd minsize [mockopts] "
"setfile ...", argv[0]);
char *fname;
struct pool_set *set;
int ret;
size_t minsize = strtoul(argv[2], &fname, 0);
for (int arg = 3; arg < argc; arg++) {
arg += mock_options(argv[arg]);
fname = argv[arg];
struct pool_attr attr;
memset(&attr, 0, sizeof(attr));
memcpy(attr.signature, SIG, sizeof(SIG));
attr.major = 1;
switch (argv[1][0]) {
case 'c':
ret = util_pool_create(&set, fname, 0, minsize,
MIN_PART, &attr, NULL, REPLICAS_ENABLED);
if (ret == -1)
UT_OUT("!%s: util_pool_create", fname);
else {
/*
* XXX: On Windows pool files are created with
* R/W permissions, so no need for chmod().
*/
#ifndef _WIN32
util_poolset_chmod(set, S_IWUSR | S_IRUSR);
#endif
poolset_info(fname, set, 0);
util_poolset_close(set, DO_NOT_DELETE_PARTS);
}
break;
case 'o':
attr.incompat_features = TEST_FORMAT_INCOMPAT_CHECK;
ret = util_pool_open(&set, fname, 0 /* rdonly */,
MIN_PART, &attr, NULL, false, NULL);
if (ret == -1)
UT_OUT("!%s: util_pool_open", fname);
else {
poolset_info(fname, set, 1);
util_poolset_close(set, DO_NOT_DELETE_PARTS);
}
break;
case 'e':
attr.incompat_features = TEST_FORMAT_INCOMPAT_CHECK;
ret = util_pool_open(&set, fname, 0 /* rdonly */,
MIN_PART, &attr, NULL, false, NULL);
UT_ASSERTeq(ret, 0);
void *nptr = util_pool_extend(set, Extend_size);
if (nptr == NULL)
UT_OUT("!%s: util_pool_extend", fname);
else {
poolset_info(fname, set, 1);
}
util_poolset_close(set, DO_NOT_DELETE_PARTS);
break;
}
}
common_fini();
DONE(NULL);
}
/*
* 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;
}
/*
* pmempool_setup_poolset -- parse poolset file and setup reading from it
*/
static int
pmempool_setup_poolset(struct pmem_info *pip)
{
struct pool_set *set = NULL;
int fd = -1;
struct pool_hdr hdr;
/* parse poolset file */
if (util_poolset_parse(pip->file_name, pip->fd, &set)) {
outv_err("parsing poolset file failed\n");
return -1;
}
close(pip->fd);
pip->fd = -1;
/* open the first part set file to read the pool header values */
int ret = 0;
fd = util_file_open(set->replica[0]->part[0].path, NULL, 0, O_RDONLY);
if (fd < 0) {
outv_err("cannot open poolset part file\n");
ret = -1;
goto err_pool_set;
}
/* read the pool header from first pool set file */
if (pread(fd, &hdr, sizeof (hdr), 0)
!= sizeof (hdr)) {
outv_err("cannot read pool header from poolset\n");
ret = -1;
goto err_close;
}
close(fd);
fd = -1;
util_convert2h_pool_hdr(&hdr);
/* parse pool type from first pool set file */
pmem_pool_type_t type = pmem_pool_type_parse_hdr(&hdr);
if (type == PMEM_POOL_TYPE_UNKNOWN) {
outv_err("cannot determine pool type from poolset\n");
ret = -1;
goto err_close;
}
/* get minimum size based on pool type for util_pool_open */
size_t minsize = pmem_pool_get_min_size(type);
/*
* Open the poolset, the values passed to util_pool_open are read
* from the first poolset file, these values are then compared with
* the values from all headers of poolset files.
*/
if (util_pool_open(&pip->poolset, pip->file_name, 1, minsize,
sizeof (struct pool_hdr),
hdr.signature, hdr.major,
hdr.compat_features,
hdr.incompat_features,
hdr.ro_compat_features)) {
outv_err("openning poolset failed\n");
ret = -1;
}
err_close:
if (fd != -1)
close(fd);
err_pool_set:
util_poolset_free(set);
return ret;
}
/*
* blk_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 *
blk_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,
BLK_HDR_SIG, BLK_FORMAT_MAJOR,
BLK_FORMAT_COMPAT, BLK_FORMAT_INCOMPAT,
BLK_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];
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);
if (set->nreplicas > 1) {
errno = ENOTSUP;
ERR("!replicas not supported");
goto err;
}
/* validate pool descriptor */
if (blk_descr_check(pbp, &bsize) != 0) {
LOG(2, "descriptor check failed");
goto err;
}
/* initialize runtime parts */
if (blk_runtime_init(pbp, bsize, set->rdonly) != 0) {
ERR("pool initialization failed");
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, DO_NOT_DELETE_PARTS);
errno = oerrno;
return NULL;
}
