/*
* 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;
}
/*
* util_pool_open -- open a memory pool (set or a single file)
*
* This routine does all the work, but takes a rdonly flag so internal
* calls can map a read-only pool if required.
*/
int
util_pool_open(struct pool_set **setp, const char *path, int rdonly,
size_t minsize, size_t hdrsize, const char *sig,
uint32_t major, uint32_t compat, uint32_t incompat, uint32_t ro_compat)
{
LOG(3, "setp %p path %s rdonly %d minsize %zu "
"hdrsize %zu sig %s major %u "
"compat %#x incompat %#x ro_comapt %#x",
setp, path, rdonly, minsize, hdrsize,
sig, major, compat, incompat, ro_compat);
int flags = rdonly ? MAP_PRIVATE|MAP_NORESERVE : MAP_SHARED;
int ret = util_poolset_open(setp, path, minsize);
if (ret < 0) {
LOG(2, "cannot open pool set");
return -1;
}
struct pool_set *set = *setp;
ASSERT(set->nreplicas > 0);
set->rdonly = 0;
set->poolsize = SIZE_MAX;
for (unsigned r = 0; r < set->nreplicas; r++) {
if (util_replica_open(set, r, flags, hdrsize, sig,
major, compat, incompat, ro_compat) != 0) {
LOG(2, "replica open failed");
goto err;
}
}
/* check replicas linkage */
for (unsigned r = 0; r < set->nreplicas; r++) {
if (memcmp(HDR(REP(set, r - 1), 0)->uuid,
HDR(REP(set, r), 0)->prev_repl_uuid,
POOL_HDR_UUID_LEN) ||
memcmp(HDR(REP(set, r + 1), 0)->uuid,
HDR(REP(set, r), 0)->next_repl_uuid,
POOL_HDR_UUID_LEN)) {
ERR("wrong replica UUID");
errno = EINVAL;
goto err;
}
}
return 0;
err:
LOG(4, "error clean up");
int oerrno = errno;
for (unsigned r = 0; r < set->nreplicas; r++) {
struct pool_replica *rep = set->replica[r];
VALGRIND_REMOVE_PMEM_MAPPING(rep->part[0].addr,
rep->part[0].size);
util_unmap(rep->part[0].addr, rep->part[0].size);
}
util_poolset_close(set, 0);
errno = oerrno;
return -1;
}
/*
* pmempool_transform -- alter poolset structure
*/
int
pmempool_transform(const char *poolset_file_src,
const char *poolset_file_dst, unsigned flags)
{
ASSERTne(poolset_file_src, NULL);
ASSERTne(poolset_file_dst, NULL);
/* check if the source poolset has correct signature */
if (util_is_poolset_file(poolset_file_src) != 1) {
ERR("source file is not a poolset file");
goto err;
}
/* check if the destination poolset has correct signature */
if (util_is_poolset_file(poolset_file_dst) != 1) {
ERR("destination file is not a poolset file");
goto err;
}
/* open the source poolset file */
int fd_in = util_file_open(poolset_file_src, NULL, 0, O_RDONLY);
if (fd_in < 0) {
ERR("cannot open source poolset file");
goto err;
}
/* parse the source poolset file */
struct pool_set *set_in = NULL;
if (util_poolset_parse(&set_in, poolset_file_src, fd_in)) {
ERR("parsing source poolset failed");
close(fd_in);
goto err;
}
close(fd_in);
/* open the destination poolset file */
int fd_out = util_file_open(poolset_file_dst, NULL, 0, O_RDONLY);
if (fd_out < 0) {
ERR("cannot open destination poolset file");
goto err;
}
/* parse the destination poolset file */
struct pool_set *set_out = NULL;
if (util_poolset_parse(&set_out, poolset_file_dst, fd_out)) {
ERR("parsing destination poolset failed");
close(fd_out);
goto err_free_poolin;
}
close(fd_out);
/* check if the source poolset is of a correct type */
if (pool_set_type(set_in) != POOL_TYPE_OBJ) {
ERR("source poolset is of a wrong type");
goto err_free_poolout;
}
/* check if the source poolset is healthy */
struct poolset_health_status *set_in_hs = NULL;
if (replica_check_poolset_health(set_in, &set_in_hs, flags)) {
ERR("source poolset health check failed");
goto err_free_poolout;
}
if (!replica_is_poolset_healthy(set_in_hs)) {
ERR("source poolset is broken");
replica_free_poolset_health_status(set_in_hs);
goto err_free_poolout;
}
replica_free_poolset_health_status(set_in_hs);
/* transform poolset */
if (transform_replica(set_in, set_out, flags)) {
ERR("transformation failed");
goto err_free_poolin;
}
util_poolset_close(set_in, 0);
util_poolset_close(set_out, 0);
return 0;
err_free_poolout:
util_poolset_close(set_out, 0);
err_free_poolin:
util_poolset_close(set_in, 0);
err:
if (errno == 0)
errno = EINVAL;
return -1;
}
/*
* 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;
}
/*
* pool_parse_params -- parse pool type, file size and block size
*/
static int
pool_params_parse(const PMEMpoolcheck *ppc, struct pool_params *params,
int check)
{
LOG(3, NULL);
int is_btt = ppc->args.pool_type == PMEMPOOL_POOL_TYPE_BTT;
params->type = POOL_TYPE_UNKNOWN;
params->is_poolset = util_is_poolset_file(ppc->path) == 1;
int fd = util_file_open(ppc->path, NULL, 0, O_RDONLY);
if (fd < 0)
return -1;
int ret = 0;
util_stat_t stat_buf;
ret = util_fstat(fd, &stat_buf);
if (ret)
goto out_close;
ASSERT(stat_buf.st_size >= 0);
params->mode = stat_buf.st_mode;
struct pool_set *set;
void *addr;
if (params->is_poolset) {
/*
* Need to close the poolset because it will be opened with
* flock in the following instructions.
*/
close(fd);
fd = -1;
if (check) {
if (pool_set_map(ppc->path, &set, 1))
return -1;
} else {
ret = util_poolset_create_set(&set, ppc->path, 0, 0);
if (ret < 0) {
LOG(2, "cannot open pool set -- '%s'",
ppc->path);
return -1;
}
if (set->remote) {
ERR("poolsets with remote replicas are not "
"supported");
return -1;
}
if (util_pool_open_nocheck(set, 1))
return -1;
}
params->size = set->poolsize;
addr = set->replica[0]->part[0].addr;
} else if (is_btt) {
params->size = (size_t)stat_buf.st_size;
#ifndef _WIN32
if (params->mode & S_IFBLK)
if (ioctl(fd, BLKGETSIZE64, ¶ms->size)) {
ERR("!ioctl");
goto out_close;
}
#endif
addr = NULL;
} else {
params->size = (size_t)stat_buf.st_size;
addr = mmap(NULL, (uint64_t)stat_buf.st_size, PROT_READ,
MAP_PRIVATE, fd, 0);
if (addr == MAP_FAILED) {
ret = -1;
goto out_close;
}
}
/* stop processing for BTT device */
if (is_btt) {
params->type = POOL_TYPE_BTT;
params->is_part = false;
goto out_close;
}
struct pool_hdr hdr;
memcpy(&hdr, addr, sizeof(hdr));
util_convert2h_hdr_nocheck(&hdr);
pool_params_from_header(params, &hdr);
if (ppc->args.pool_type != PMEMPOOL_POOL_TYPE_DETECT) {
enum pool_type declared_type =
pmempool_check_type_to_pool_type(ppc->args.pool_type);
if ((params->type & ~declared_type) != 0) {
ERR("declared pool type does not match");
ret = 1;
goto out_unmap;
}
}
if (params->type == POOL_TYPE_BLK) {
struct pmemblk pbp;
memcpy(&pbp, addr, sizeof(pbp));
params->blk.bsize = le32toh(pbp.bsize);
} else if (params->type == POOL_TYPE_OBJ) {
struct pmemobjpool pop;
memcpy(&pop, addr, sizeof(pop));
memcpy(params->obj.layout, pop.layout,
PMEMOBJ_MAX_LAYOUT);
}
out_unmap:
if (params->is_poolset) {
ASSERTeq(fd, -1);
ASSERTne(addr, NULL);
util_poolset_close(set, 0);
} else if (!is_btt) {
ASSERTne(fd, -1);
ASSERTne(addr, NULL);
munmap(addr, params->size);
}
out_close:
if (fd != -1)
close(fd);
return ret;
}
/*
* util_poolset_open -- (internal) open memory pool set
*
* On success returns 0 and a pointer to a newly allocated structure
* containing the info of all the parts of the pool set and replicas.
*/
static int
util_poolset_open(struct pool_set **setp, const char *path, size_t minsize)
{
LOG(3, "setp %p path %s minsize %zu", setp, path, minsize);
int oerrno;
int ret = 0;
int fd;
size_t size = 0;
/* do not check minsize */
if ((fd = util_file_open(path, &size, 0, O_RDONLY)) == -1)
return -1;
char signature[POOLSET_HDR_SIG_LEN];
/*
* read returns ssize_t, but we know it will return value between -1
* and POOLSET_HDR_SIG_LEN (11), so we can safely cast it to int
*/
ret = (int)read(fd, signature, POOLSET_HDR_SIG_LEN);
if (ret < 0) {
ERR("!read %d", fd);
goto err;
}
if (ret < POOLSET_HDR_SIG_LEN ||
strncmp(signature, POOLSET_HDR_SIG, POOLSET_HDR_SIG_LEN)) {
LOG(4, "not a pool set header");
if (size < minsize) {
ERR("size %zu smaller than %zu", size, minsize);
errno = EINVAL;
ret = -1;
goto err;
}
/* close the file and open with O_RDWR */
(void) close(fd);
size = 0;
if ((fd = util_file_open(path, &size, 0, O_RDWR)) == -1)
return -1;
*setp = util_poolset_single(path, size, fd, 0);
if (*setp == NULL) {
ret = -1;
goto err;
}
/* do not close the file */
return 0;
}
ret = util_poolset_parse(path, fd, setp);
if (ret != 0)
goto err;
ret = util_poolset_files(*setp, minsize, 0);
if (ret != 0)
util_poolset_close(*setp, 0);
err:
oerrno = errno;
(void) close(fd);
errno = oerrno;
return ret;
}
/*
* util_pool_open -- open a memory pool (set or a single file)
*
* This routine does all the work, but takes a rdonly flag so internal
* calls can map a read-only pool if required.
*/
int
util_pool_open(struct pool_set **setp, const char *path, int rdonly,
size_t minsize, size_t hdrsize, const char *sig,
uint32_t major, uint32_t compat, uint32_t incompat, uint32_t ro_compat)
{
LOG(3, "setp %p path %s rdonly %d minsize %zu "
"hdrsize %zu sig %.8s major %u "
"compat %#x incompat %#x ro_comapt %#x",
setp, path, rdonly, minsize, hdrsize,
sig, major, compat, incompat, ro_compat);
int flags = rdonly ? MAP_PRIVATE|MAP_NORESERVE : MAP_SHARED;
int ret = util_poolset_open(setp, path, minsize);
if (ret < 0) {
LOG(2, "cannot open pool set");
return -1;
}
struct pool_set *set = *setp;
ASSERT(set->nreplicas > 0);
set->rdonly = 0;
set->poolsize = SIZE_MAX;
for (unsigned r = 0; r < set->nreplicas; r++) {
if (util_replica_open(set, r, flags, hdrsize) != 0) {
LOG(2, "replica open failed");
goto err;
}
}
/* check headers, check UUID's, check replicas linkage */
for (unsigned r = 0; r < set->nreplicas; r++) {
struct pool_replica *rep = set->replica[r];
for (unsigned p = 0; p < rep->nparts; p++) {
if (util_header_check(set, r, p, sig, major,
compat, incompat, ro_compat) != 0) {
LOG(2, "header check failed - part #%d", p);
goto err;
}
set->rdonly |= rep->part[p].rdonly;
}
if (memcmp(HDR(REP(set, r - 1), 0)->uuid,
HDR(REP(set, r), 0)->prev_repl_uuid,
POOL_HDR_UUID_LEN) ||
memcmp(HDR(REP(set, r + 1), 0)->uuid,
HDR(REP(set, r), 0)->next_repl_uuid,
POOL_HDR_UUID_LEN)) {
ERR("wrong replica UUID");
errno = EINVAL;
goto err;
}
}
/* unmap all headers */
for (unsigned r = 0; r < set->nreplicas; r++) {
struct pool_replica *rep = set->replica[r];
for (unsigned p = 0; p < rep->nparts; p++)
util_unmap_hdr(&rep->part[p]);
}
return 0;
err:
LOG(4, "error clean up");
int oerrno = errno;
for (unsigned r = 0; r < set->nreplicas; r++)
util_replica_close(set, r);
util_poolset_close(set, 0);
errno = oerrno;
return -1;
}
/*
* 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;
}
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;
}
