/*
* pmempool_info_part -- (internal) print info about poolset part
*/
static int
pmempool_info_part(struct pmem_info *pip, unsigned repn, unsigned partn, int v)
{
/* get path of the part file */
const char *path = NULL;
if (repn != UNDEF_REPLICA && partn != UNDEF_PART) {
outv(v, "part %u:\n", partn);
struct pool_set_part *part =
&pip->pfile->poolset->replica[repn]->part[partn];
path = part->path;
} else {
outv(v, "Part file:\n");
path = pip->file_name;
}
outv_field(v, "path", "%s", path);
/* get type of the part file */
int is_dev_dax = util_file_is_device_dax(path);
const char *type_str = is_dev_dax ? "device dax" : "regular file";
outv_field(v, "type", "%s", type_str);
/* get size of the part file */
ssize_t size = util_file_get_size(path);
if (size < 0) {
outv_err("couldn't get size of %s", path);
return -1;
}
outv_field(v, "size", "%s", out_get_size_str((size_t)size,
pip->args.human));
return 0;
}
/*
* util_file_pread -- reads from a file with an offset
*/
ssize_t
util_file_pread(const char *path, void *buffer, size_t size,
off_t offset)
{
if (!util_file_is_device_dax(path)) {
int fd = util_file_open(path, NULL, 0, O_RDONLY);
if (fd < 0)
return -1;
ssize_t read_len = pread(fd, buffer, size, offset);
int olderrno = errno;
(void) close(fd);
errno = olderrno;
return read_len;
}
ssize_t file_size = util_file_get_size(path);
if (file_size < 0)
return -1;
size_t max_size = (size_t)(file_size - offset);
if (size > max_size) {
LOG(1, "Requested size of read goes beyond the mapped memory");
size = max_size;
}
void *addr = util_file_map_whole(path);
if (addr == NULL)
return -1;
memcpy(buffer, ADDR_SUM(addr, offset), size);
util_unmap(addr, (size_t)file_size);
return (ssize_t)size;
}
/*
* util_unlink -- unlinks a file or zeroes a device dax
*/
int
util_unlink(const char *path)
{
if (util_file_is_device_dax(path)) {
return util_file_zero_whole(path);
} else {
return unlink(path);
}
}
/*
* util_unlink -- unlinks a file or zeroes a device dax
*/
int
util_unlink(const char *path)
{
if (util_file_is_device_dax(path)) {
return util_file_zero_whole(path);
} else {
/*
* On Windows we can not unlink Read-Only files
*/
#ifdef _WIN32
_chmod(path, _S_IREAD | _S_IWRITE);
#endif
return unlink(path);
}
}
/*
* is_dev_dax -- checks if given path points to device dax
*/
static int
is_dev_dax(const char *path)
{
if (!util_file_is_device_dax(path)) {
printf("%s -- not device dax\n", path);
return 0;
}
if (access(path, W_OK|R_OK)) {
printf("%s -- permission denied\n", path);
return -1;
}
return 1;
}
/*
* util_file_get_size -- returns size of a file
*/
ssize_t
util_file_get_size(const char *path)
{
#ifndef _WIN32
if (util_file_is_device_dax(path)) {
return device_dax_size(path);
}
#endif
util_stat_t stbuf;
if (util_stat(path, &stbuf) < 0) {
ERR("!fstat %s", path);
return -1;
}
return stbuf.st_size;
}
/*
* 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, 0))
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, 0))
return -1;
}
params->size = set->poolsize;
addr = set->replica[0]->part[0].addr;
/*
* XXX mprotect for device dax with length not aligned to its
* page granularity causes SIGBUS on the next page fault.
* The length argument of this call should be changed to
* set->poolsize once the kernel issue is solved.
*/
if (mprotect(addr, set->replica[0]->repsize,
PROT_READ) < 0) {
ERR("!mprotect");
goto out_unmap;
}
params->is_device_dax = set->replica[0]->part[0].is_dax;
} 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 {
ssize_t s = util_file_get_size(ppc->path);
if (s < 0) {
ret = -1;
goto out_close;
}
params->size = (size_t)s;
addr = mmap(NULL, (uint64_t)params->size, PROT_READ,
MAP_PRIVATE, fd, 0);
if (addr == MAP_FAILED) {
ret = -1;
goto out_close;
}
params->is_device_dax = util_file_is_device_dax(ppc->path);
}
/* 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;
}
/*
* pobj_init - common part of the benchmark initialization functions.
* Parses command line arguments, set variables and creates persistent pools.
*/
static int
pobj_init(struct benchmark *bench, struct benchmark_args *args)
{
unsigned i = 0;
size_t psize;
size_t n_objs;
assert(bench != nullptr);
assert(args != nullptr);
auto *bench_priv =
(struct pobj_bench *)malloc(sizeof(struct pobj_bench));
if (bench_priv == nullptr) {
perror("malloc");
return -1;
}
assert(args->opts != nullptr);
bench_priv->args_priv = (struct pobj_args *)args->opts;
bench_priv->args_priv->obj_size = args->dsize;
bench_priv->args_priv->range =
bench_priv->args_priv->min_size > 0 ? true : false;
bench_priv->n_pools =
!bench_priv->args_priv->one_pool ? args->n_threads : 1;
bench_priv->pool = bench_priv->n_pools > 1 ? diff_num : one_num;
bench_priv->obj = !bench_priv->args_priv->one_obj ? diff_num : one_num;
if ((args->is_poolset || util_file_is_device_dax(args->fname)) &&
bench_priv->n_pools > 1) {
fprintf(stderr,
"cannot use poolset nor device dax for multiple pools,"
" please use -P|--one-pool option instead");
goto free_bench_priv;
}
/*
* Multiplication by FACTOR prevents from out of memory error
* as the actual size of the allocated persistent objects
* is always larger than requested.
*/
n_objs = bench_priv->args_priv->n_objs;
if (bench_priv->n_pools == 1)
n_objs *= args->n_threads;
psize = n_objs * args->dsize * args->n_threads * FACTOR;
if (psize < PMEMOBJ_MIN_POOL)
psize = PMEMOBJ_MIN_POOL;
/* assign type_number determining function */
bench_priv->type_mode =
parse_type_mode(bench_priv->args_priv->type_num);
switch (bench_priv->type_mode) {
case MAX_TYPE_MODE:
fprintf(stderr, "unknown type mode");
goto free_bench_priv;
case TYPE_MODE_RAND:
if (random_types(bench_priv, args))
goto free_bench_priv;
break;
default:
bench_priv->random_types = nullptr;
}
bench_priv->fn_type_num = type_mode_func[bench_priv->type_mode];
/* assign size determining function */
bench_priv->fn_size =
bench_priv->args_priv->range ? range_size : static_size;
bench_priv->rand_sizes = nullptr;
if (bench_priv->args_priv->range) {
if (bench_priv->args_priv->min_size > args->dsize) {
fprintf(stderr, "Invalid allocation size");
goto free_random_types;
}
bench_priv->rand_sizes =
rand_sizes(bench_priv->args_priv->min_size,
bench_priv->args_priv->obj_size,
bench_priv->args_priv->n_objs);
if (bench_priv->rand_sizes == nullptr)
goto free_random_types;
}
assert(bench_priv->n_pools > 0);
bench_priv->pop = (PMEMobjpool **)calloc(bench_priv->n_pools,
sizeof(PMEMobjpool *));
if (bench_priv->pop == nullptr) {
perror("calloc");
goto free_random_sizes;
}
bench_priv->sets = (const char **)calloc(bench_priv->n_pools,
sizeof(const char *));
if (bench_priv->sets == nullptr) {
perror("calloc");
goto free_pop;
}
if (bench_priv->n_pools > 1) {
assert(!args->is_poolset);
if (util_file_mkdir(args->fname, DIR_MODE) != 0) {
fprintf(stderr, "cannot create directory\n");
goto free_sets;
}
size_t path_len = (strlen(PART_NAME) + strlen(args->fname)) +
MAX_DIGITS + 1;
for (i = 0; i < bench_priv->n_pools; i++) {
bench_priv->sets[i] =
(char *)malloc(path_len * sizeof(char));
if (bench_priv->sets[i] == nullptr) {
perror("malloc");
goto free_sets;
}
int ret =
snprintf((char *)bench_priv->sets[i], path_len,
"%s%s%02x", args->fname, PART_NAME, i);
if (ret < 0 || ret >= (int)path_len) {
perror("snprintf");
goto free_sets;
}
bench_priv->pop[i] =
pmemobj_create(bench_priv->sets[i], LAYOUT_NAME,
psize, FILE_MODE);
if (bench_priv->pop[i] == nullptr) {
perror(pmemobj_errormsg());
goto free_sets;
}
}
} else {
if (args->is_poolset || util_file_is_device_dax(args->fname)) {
if (args->fsize < psize) {
fprintf(stderr, "file size too large\n");
goto free_pools;
}
psize = 0;
}
bench_priv->sets[0] = args->fname;
bench_priv->pop[0] = pmemobj_create(
bench_priv->sets[0], LAYOUT_NAME, psize, FILE_MODE);
if (bench_priv->pop[0] == nullptr) {
perror(pmemobj_errormsg());
goto free_pools;
}
}
pmembench_set_priv(bench, bench_priv);
return 0;
free_sets:
for (; i > 0; i--) {
pmemobj_close(bench_priv->pop[i - 1]);
free((char *)bench_priv->sets[i - 1]);
}
free_pools:
free(bench_priv->sets);
free_pop:
free(bench_priv->pop);
free_random_sizes:
free(bench_priv->rand_sizes);
free_random_types:
free(bench_priv->random_types);
free_bench_priv:
free(bench_priv);
return -1;
}
