static void
list_update(list_t *list, lwp_info_t *lwp)
{
id_info_t *id;
if (list->l_head == NULL) { /* first element */
list->l_head = list->l_tail = id = Zalloc(sizeof (id_info_t));
id_update(id, lwp, list->l_type);
list->l_count++;
return;
}
for (id = list->l_head; id; id = id->id_next) {
if ((list->l_type == LT_PROCESS) &&
(id->id_pid != lwp->li_psinfo->pr_pid))
continue;
if ((list->l_type == LT_USERS) &&
(id->id_uid != lwp->li_psinfo->pr_uid))
continue;
if ((list->l_type == LT_PROJECTS) &&
(id->id_projid != lwp->li_psinfo->pr_projid))
continue;
id_update(id, lwp, list->l_type);
return;
}
/* a new element */
id = list->l_tail;
id->id_next = Zalloc(sizeof (id_info_t));
id->id_next->id_prev = list->l_tail;
id->id_next->id_next = NULL;
list->l_tail = id->id_next;
id = list->l_tail;
id_update(id, lwp, list->l_type);
list->l_count++;
}
/*
* get swap info
*/
static int
get_swapinfo(
struct swaptable **swtpp,
int *nswap,
md_error_t *ep
)
{
int i;
size_t swtsize;
*swtpp = NULL;
/* get number of entries */
if ((*nswap = swapctl(SC_GETNSWP, NULL)) < 0) {
return (mdsyserror(ep, errno, "swapctl(SC_GETNSWP)"));
}
/* allocate structure */
swtsize = sizeof ((*swtpp)->swt_n) +
((*nswap) * sizeof ((*swtpp)->swt_ent[0]));
*swtpp = (struct swaptable *)Zalloc(swtsize);
(*swtpp)->swt_n = *nswap;
for (i = 0; (i < (*nswap)); ++i)
(*swtpp)->swt_ent[i].ste_path = Zalloc(MAXPATHLEN);
/* get info */
if (((*nswap) = swapctl(SC_LIST, (*swtpp))) < 0) {
(void) mdsyserror(ep, errno, "swapctl(SC_LIST)");
free_swapinfo(*swtpp);
return (-1);
}
/* return success */
return (0);
}
/*
* util_parse_add_replica -- (internal) add a new replica to the pool set info
*/
static int
util_parse_add_replica(struct pool_set **setp)
{
LOG(3, "setp %p", setp);
ASSERTne(setp, NULL);
struct pool_set *set = *setp;
ASSERTne(set, NULL);
set = Realloc(set, sizeof(struct pool_set) +
(set->nreplicas + 1) * sizeof(struct pool_replica *));
if (set == NULL) {
ERR("!Realloc");
return -1;
}
*setp = set;
struct pool_replica *rep;
rep = Zalloc(sizeof(struct pool_replica));
if (rep == NULL) {
ERR("!Malloc");
return -1;
}
unsigned r = set->nreplicas++;
set->replica[r] = rep;
return 0;
}
/*
* util_parse_add_remote_replica -- (internal) add a new remote replica
* to the pool set info
*/
static int
util_parse_add_remote_replica(struct pool_set **setp, char *node_addr,
char *pool_desc)
{
LOG(3, "setp %p node_addr %s pool_desc %s", setp, node_addr, pool_desc);
ASSERTne(setp, NULL);
ASSERTne(node_addr, NULL);
ASSERTne(pool_desc, NULL);
int ret = util_parse_add_replica(setp);
if (ret != 0)
return ret;
/* a remote replica has one 'fake' part */
ret = util_parse_add_part(*setp, NULL, 0);
if (ret != 0)
return ret;
struct pool_set *set = *setp;
struct pool_replica *rep = set->replica[set->nreplicas - 1];
ASSERTne(rep, NULL);
rep->remote = Zalloc(sizeof(struct remote_replica));
if (rep->remote == NULL) {
ERR("!Malloc");
return -1;
}
rep->remote->node_addr = node_addr;
rep->remote->pool_desc = pool_desc;
set->remote = 1;
return 0;
}
/*
* replica_create_poolset_health_status -- create helping structure for storing
* poolset's health status
*/
int
replica_create_poolset_health_status(struct pool_set *set,
struct poolset_health_status **set_hsp)
{
LOG(3, "set %p, set_hsp %p", set, set_hsp);
unsigned nreplicas = set->nreplicas;
struct poolset_health_status *set_hs;
set_hs = Zalloc(sizeof(struct poolset_health_status) +
nreplicas * sizeof(struct replica_health_status *));
if (set_hs == NULL) {
ERR("!Zalloc for poolset health state");
return -1;
}
set_hs->nreplicas = nreplicas;
for (unsigned i = 0; i < nreplicas; ++i) {
struct replica_health_status *replica_hs =
create_replica_health_status(set, i);
if (replica_hs == NULL) {
replica_free_poolset_health_status(set_hs);
return -1;
}
set_hs->replica[i] = replica_hs;
}
*set_hsp = set_hs;
return 0;
}
/*
* open RPC connection to rpc.metamhd
*/
static mhd_handle_t *
open_metamhd(
char *hostname,
md_error_t *ep
)
{
CLIENT *clientp;
mhd_handle_t *hp;
/* default to local host */
if ((hostname == NULL) || (*hostname == '\0'))
hostname = mynode();
/* open RPC connection */
assert(hostname != NULL);
if ((clientp = meta_client_create(hostname, METAMHD, METAMHD_VERSION,
"tcp")) == NULL) {
clnt_pcreateerror(hostname);
(void) mdrpccreateerror(ep, hostname, "metamhd clnt_create");
return (NULL);
} else {
auth_destroy(clientp->cl_auth);
clientp->cl_auth = authsys_create_default();
assert(clientp->cl_auth != NULL);
}
/* return connection */
hp = Zalloc(sizeof (*hp));
hp->hostname = Strdup(hostname);
hp->clientp = clientp;
return (hp);
}
/*
* util_poolset_single -- (internal) create a one-part pool set
*
* On success returns a pointer to a newly allocated and initialized
* pool set structure. Otherwise, NULL is returned.
*/
static struct pool_set *
util_poolset_single(const char *path, size_t filesize, int fd, int create)
{
LOG(3, "path %s filesize %zu fd %d create %d",
path, filesize, fd, create);
struct pool_set *set;
set = Zalloc(sizeof(struct pool_set) +
sizeof(struct pool_replica *));
if (set == NULL) {
ERR("!Malloc for pool set");
return NULL;
}
struct pool_replica *rep;
rep = Zalloc(sizeof(struct pool_replica) +
sizeof(struct pool_set_part));
if (rep == NULL) {
ERR("!Malloc for pool set replica");
Free(set);
return NULL;
}
set->replica[0] = rep;
rep->part[0].filesize = filesize;
rep->part[0].path = Strdup(path);
rep->part[0].fd = fd;
rep->part[0].created = create;
rep->part[0].hdr = NULL;
rep->part[0].addr = NULL;
rep->nparts = 1;
/* it does not have a remote replica */
rep->remote = NULL;
set->remote = 0;
/* round down to the nearest page boundary */
rep->repsize = rep->part[0].filesize & ~(Pagesize - 1);
set->poolsize = rep->repsize;
set->nreplicas = 1;
return set;
}
void
list_alloc(list_t *list, int size)
{
if (size > 0) {
list->l_size = size;
list->l_ptrs = Zalloc(sizeof (void *) * (size + 1));
}
}
void
list_alloc(list_t *list, int size)
{
list->l_size = size;
if (size > 0)
list->l_ptrs = Zalloc(sizeof (void *) * (size + 1));
else
list->l_ptrs = NULL;
}
static lwp_info_t *
list_add_lwp(list_t *list, pid_t pid, id_t lwpid)
{
lwp_info_t *lwp;
if (list->l_head == NULL) {
list->l_head = list->l_tail = lwp = Zalloc(sizeof (lwp_info_t));
} else {
lwp = Zalloc(sizeof (lwp_info_t));
lwp->li_prev = list->l_tail;
((lwp_info_t *)list->l_tail)->li_next = lwp;
list->l_tail = lwp;
}
lwp->li_info.pr_pid = pid;
lwp->li_info.pr_lwp.pr_lwpid = lwpid;
lwpid_add(lwp, pid, lwpid);
list->l_count++;
return (lwp);
}
static int
list_read(int listt, int elemn)
{
char idstr[P_MAXVAL];
list_t *list;
id_info_t *id;
if (listt == L_LWP)
return (lwp_read(elemn));
while (elemn-- > 0) {
switch (listt) {
case L_PRC_SI : list = &processes;
break;
case L_USR_SI : list = &users;
break;
case L_PRJ_SI : list = &projects;
break;
}
if (list->l_head == NULL) { /* first element */
list->l_head = list->l_tail = id =
Zalloc(sizeof (id_info_t));
list->l_count++;
} else {
/* a new element */
id = list->l_tail;
id->id_next = Zalloc(sizeof (id_info_t));
id->id_next->id_prev = list->l_tail;
id->id_next->id_next = NULL;
list->l_tail = id->id_next;
id = list->l_tail;
list->l_count++;
}
if (r_element((char *)id, idstr) == -1) {
list_clear(list);
return (-1);
}
}
return (0);
}
void
lwpid_add(lwp_info_t *lwp, pid_t pid, id_t lwpid)
{
plwp_t *elm = Zalloc(sizeof (plwp_t));
int hash = pid % PLWP_TBL_SZ;
elm->l_pid = pid;
elm->l_lwpid = lwpid;
elm->l_lwp = lwp;
elm->l_next = plwp_tbl[hash]; /* add in front of chain */
plwp_tbl[hash] = elm;
}
/*
* status drives in set
*/
int
mhd_status_drives(
mhd_set_t *mhsp,
mhd_opts_t options,
mhd_drive_status_t **status,
mhd_error_t *mhep
)
{
mhd_drive_list_t dl = mhd_null_list;
mhd_drive_list_t *dlp = &dl;
mhd_drive_set_t *sp;
uint_t i;
int rval = 0;
/* grab global lock */
mhd_mx_lock(&mhd_set_mx);
/* create or update set */
if ((sp = mhd_create_set(mhsp, options, &dl, mhep)) == NULL) {
mhd_mx_unlock(&mhd_set_mx);
mhd_free_list(&dl);
return (-1);
}
/* lock set */
mhd_mx_lock(&sp->sr_mx);
/* status drives */
if (mhd_status_set(sp, &dl, mhep) != 0) {
rval = -1;
goto out;
}
/* build list */
*status = Zalloc(dlp->dl_ndrive * sizeof (**status));
for (i = 0; (i < dlp->dl_ndrive); ++i) {
mhd_drive_t *dp = dlp->dl_drives[i];
mhd_drive_status_t *statusp = &(*status)[i];
statusp->drive = Strdup(dp->dr_rname);
statusp->errnum = dp->dr_errnum;
}
assert(i == dlp->dl_ndrive);
rval = dlp->dl_ndrive;
/* unlock, return count */
out:
mhd_mx_unlock(&sp->sr_mx);
mhd_mx_unlock(&mhd_set_mx);
mhd_free_list(&dl);
return (rval);
}
/*
* create_replica_health_status -- (internal) create helping structure for
* storing replica's health status
*/
static struct replica_health_status *
create_replica_health_status(struct pool_set *set, unsigned repn)
{
unsigned nparts = set->replica[repn]->nparts;
struct replica_health_status *replica_hs;
replica_hs = Zalloc(sizeof(struct replica_health_status)
+ nparts * sizeof(unsigned));
if (replica_hs == NULL) {
ERR("!Zalloc for replica health status");
return NULL;
}
replica_hs->nparts = nparts;
return replica_hs;
}
/*
* lane_boot -- initializes all lanes
*/
int
lane_boot(PMEMobjpool *pop)
{
int err = 0;
pop->lanes_desc.lane = Malloc(sizeof(struct lane) * pop->nlanes);
if (pop->lanes_desc.lane == NULL) {
err = ENOMEM;
ERR("!Malloc of volatile lanes");
goto error_lanes_malloc;
}
pop->lanes_desc.next_lane_idx = 0;
pop->lanes_desc.lane_locks =
Zalloc(sizeof(*pop->lanes_desc.lane_locks) * pop->nlanes);
if (pop->lanes_desc.lane_locks == NULL) {
ERR("!Malloc for lane locks");
goto error_locks_malloc;
}
/* add lanes to pmemcheck ignored list */
VALGRIND_ADD_TO_GLOBAL_TX_IGNORE((char *)pop + pop->lanes_offset,
(sizeof(struct lane_layout) * pop->nlanes));
uint64_t i;
for (i = 0; i < pop->nlanes; ++i) {
struct lane_layout *layout = lane_get_layout(pop, i);
if ((err = lane_init(pop, &pop->lanes_desc.lane[i], layout))) {
ERR("!lane_init");
goto error_lane_init;
}
}
return 0;
error_lane_init:
for (; i >= 1; --i)
lane_destroy(pop, &pop->lanes_desc.lane[i - 1]);
Free(pop->lanes_desc.lane_locks);
pop->lanes_desc.lane_locks = NULL;
error_locks_malloc:
Free(pop->lanes_desc.lane);
pop->lanes_desc.lane = NULL;
error_lanes_malloc:
return err;
}
/*
* Add a device/device id tuple to the devname cache
*/
static void
add_to_devname_list(
char *device_name, /* fully qualified dev name */
ddi_devid_t devid /* device id */
)
{
dev_list_t *dnlp;
dnlp = Zalloc(sizeof (*dnlp));
dnlp->dev_name = Strdup(device_name);
dnlp->devid = devid;
/* link the node into the devname list */
dnlp->dev_nxt = devnamelist;
devnamelist = dnlp;
}
/*
* shutdown_state_add_part -- adds file uuid and usc to shutdown_state struct
*
* if path does not exist it will fail which does NOT mean shutdown failure
*/
int
shutdown_state_add_part(struct shutdown_state *sds, const char *path,
struct pool_replica *rep)
{
LOG(3, "sds %p, path %s", sds, path);
size_t len = 0;
char *uid;
uint64_t usc;
if (os_dimm_usc(path, &usc)) {
ERR("cannot read unsafe shutdown count of %s", path);
return 1;
}
if (os_dimm_uid(path, NULL, &len)) {
ERR("cannot read uuid of %s", path);
return 1;
}
len += 4 - len % 4;
uid = Zalloc(len);
if (uid == NULL) {
ERR("!Zalloc");
return 1;
}
if (os_dimm_uid(path, uid, &len)) {
ERR("cannot read uuid of %s", path);
Free(uid);
return 1;
}
sds->usc = htole64(le64toh(sds->usc) + usc);
uint64_t tmp;
util_checksum(uid, len, &tmp, 1, 0);
sds->uuid = htole64(le64toh(sds->uuid) + tmp);
FLUSH_SDS(sds, rep);
Free(uid);
shutdown_state_checksum(sds, rep);
return 0;
}
/*
* calloc
*/
void *
Calloc(
size_t n,
size_t s
)
{
unsigned long total;
if (n == 0 || s == 0) {
total = 0;
} else {
total = (unsigned long)n * s;
/* check for overflow */
if (total / n != s)
return (NULL);
}
return (Zalloc(total));
}
/*
* create poolset_compare_status -- (internal) create structure for gathering
* status of poolset comparison
*/
static int
create_poolset_compare_status(struct pool_set *set,
struct poolset_compare_status **set_sp)
{
LOG(3, "set %p, set_sp %p", set, set_sp);
struct poolset_compare_status *set_s;
set_s = Zalloc(sizeof(struct poolset_compare_status)
+ set->nreplicas * sizeof(unsigned));
if (set_s == NULL) {
ERR("!Zalloc for poolset status");
return -1;
}
for (unsigned r = 0; r < set->nreplicas; ++r)
set_s->replica[r] = UNDEF_REPLICA;
set_s->nreplicas = set->nreplicas;
*set_sp = set_s;
return 0;
}
int
os_execv(const char *path, char *const argv[])
{
wchar_t *wpath = util_toUTF16(path);
if (wpath == NULL)
return -1;
int argc = 0;
while (argv[argc])
argc++;
int ret;
wchar_t **wargv = Zalloc((argc + 1) * sizeof(wargv[0]));
if (!wargv) {
ret = -1;
goto wargv_alloc_failed;
}
for (int i = 0; i < argc; ++i) {
wargv[i] = util_toUTF16(argv[i]);
if (!wargv[i]) {
ret = -1;
goto end;
}
}
intptr_t iret = _wexecv(wpath, wargv);
if (iret == 0)
ret = 0;
else
ret = -1;
end:
for (int i = 0; i < argc; ++i)
util_free_UTF16(wargv[i]);
Free(wargv);
wargv_alloc_failed:
util_free_UTF16(wpath);
return ret;
}
/*
* build disk status list from drivename list
*/
md_disk_status_list_t *
meta_drive_to_disk_status_list(
mddrivenamelist_t *dnlp
)
{
md_disk_status_list_t *head = NULL;
md_disk_status_list_t **tailp = &head;
mddrivenamelist_t *p;
/* copy list */
for (p = dnlp; (p != NULL); p = p->next) {
md_disk_status_list_t *dsp;
dsp = *tailp = Zalloc(sizeof (*dsp));
tailp = &dsp->next;
dsp->drivenamep = p->drivenamep;
}
/* return list */
return (head);
}
static void
setmovecur()
{
int i, n;
if ((opts.o_outpmode & OPT_FULLSCREEN) &&
(opts.o_outpmode & OPT_USEHOME)) {
movecur = t_home;
return;
}
if (opts.o_outpmode & OPT_SPLIT) {
n = opts.o_ntop + opts.o_nbottom + 2;
} else {
if (opts.o_outpmode & OPT_USERS)
n = opts.o_nbottom + 1;
else
n = opts.o_ntop + 1;
}
if (movecur != NULL && movecur != empty_string && movecur != t_home)
free(movecur);
movecur = Zalloc(strlen(t_up) * (n + 5));
for (i = 0; i <= n; i++)
(void) strcat(movecur, t_up);
}
/*
* util_poolset_parse -- (internal) parse pool set config file
*
* Returns 1 if the file is a valid pool set config file, 0 if the file
* is not a pool set header, and -1 in case of any error.
*
* XXX: use memory mapped file
*/
int
util_poolset_parse(const char *path, int fd, struct pool_set **setp)
{
LOG(3, "path %s fd %d setp %p", path, fd, setp);
struct pool_set *set;
enum parser_codes result;
char line[PARSER_MAX_LINE];
char *s;
char *ppath;
char *pool_desc;
char *node_addr;
char *cp;
size_t psize;
FILE *fs;
if (util_lseek(fd, 0, SEEK_SET) != 0) {
ERR("!lseek %d", fd);
return -1;
}
fd = dup(fd);
if (fd < 0) {
ERR("!dup");
return -1;
}
/* associate a stream with the file descriptor */
if ((fs = fdopen(fd, "r")) == NULL) {
ERR("!fdopen %d", fd);
close(fd);
return -1;
}
unsigned nlines = 0;
unsigned nparts = 0; /* number of parts in current replica */
/* read the first line */
s = fgets(line, PARSER_MAX_LINE, fs);
nlines++;
set = Zalloc(sizeof(struct pool_set));
if (set == NULL) {
ERR("!Malloc for pool set");
goto err;
}
/* check also if the last character is '\n' */
if (s && strncmp(line, POOLSET_HDR_SIG, POOLSET_HDR_SIG_LEN) == 0 &&
line[POOLSET_HDR_SIG_LEN] == '\n') {
/* 'PMEMPOOLSET' signature detected */
LOG(10, "PMEMPOOLSET");
int ret = util_parse_add_replica(&set);
if (ret != 0)
goto err;
nparts = 0;
result = PARSER_CONTINUE;
} else {
result = PARSER_PMEMPOOLSET;
}
while (result == PARSER_CONTINUE) {
/* read next line */
s = fgets(line, PARSER_MAX_LINE, fs);
nlines++;
if (s) {
/* chop off newline and comments */
if ((cp = strchr(line, '\n')) != NULL)
*cp = '\0';
if (cp != s && (cp = strchr(line, '#')) != NULL)
*cp = '\0';
/* skip comments and blank lines */
if (cp == s)
continue;
}
if (!s) {
if (nparts >= 1) {
result = PARSER_FORMAT_OK;
} else {
if (set->nreplicas == 1)
result = PARSER_SET_NO_PARTS;
else
result = PARSER_REP_NO_PARTS;
}
} else if (strncmp(line, POOLSET_REPLICA_SIG,
POOLSET_REPLICA_SIG_LEN) == 0) {
if (line[POOLSET_REPLICA_SIG_LEN] != '\0') {
/* something more than 'REPLICA' */
if (!isblank(line[POOLSET_REPLICA_SIG_LEN])) {
result = PARSER_REPLICA;
continue;
}
/* check if it is a remote replica */
result = parser_read_replica(
line + POOLSET_REPLICA_SIG_LEN,
&node_addr, &pool_desc);
if (result == PARSER_CONTINUE) {
/* remote REPLICA */
LOG(10, "REMOTE REPLICA "
"node address '%s' "
"pool set descriptor '%s'",
node_addr, pool_desc);
if (util_parse_add_remote_replica(&set,
node_addr, pool_desc))
goto err;
}
} else if (nparts >= 1) {
/* 'REPLICA' signature detected */
LOG(10, "REPLICA");
int ret = util_parse_add_replica(&set);
if (ret != 0)
goto err;
nparts = 0;
result = PARSER_CONTINUE;
} else {
if (set->nreplicas == 1)
result = PARSER_SET_NO_PARTS;
else
result = PARSER_REP_NO_PARTS;
}
} else {
/* read size and path */
result = parser_read_line(line, &psize, &ppath);
if (result == PARSER_CONTINUE) {
/* add a new pool's part to the list */
int ret = util_parse_add_part(set,
ppath, psize);
if (ret != 0)
goto err;
nparts++;
}
}
}
if (result == PARSER_FORMAT_OK) {
LOG(4, "set file format correct (%s)", path);
(void) fclose(fs);
util_poolset_set_size(set);
*setp = set;
return 0;
} else {
ERR("%s [%s:%d]", path, parser_errstr[result], nlines);
errno = EINVAL;
}
err:
(void) fclose(fs);
if (set)
util_poolset_free(set);
return -1;
}
/*
* alloc_class_collection_new -- creates a new collection of allocation classes
*/
struct alloc_class_collection *
alloc_class_collection_new()
{
LOG(10, NULL);
struct alloc_class_collection *ac = Zalloc(sizeof(*ac));
if (ac == NULL)
return NULL;
memset(ac->aclasses, 0, sizeof(ac->aclasses));
ac->granularity = ALLOC_BLOCK_SIZE;
ac->last_run_max_size = MAX_RUN_SIZE;
ac->fail_on_missing_class = 0;
ac->autogenerate_on_missing_class = 1;
size_t maps_size = (MAX_RUN_SIZE / ac->granularity) + 1;
if ((ac->class_map_by_alloc_size = Malloc(maps_size)) == NULL)
goto error;
if ((ac->class_map_by_unit_size = cuckoo_new()) == NULL)
goto error;
memset(ac->class_map_by_alloc_size, 0xFF, maps_size);
if (alloc_class_new(-1, ac, CLASS_HUGE, HEADER_COMPACT,
CHUNKSIZE, 0, 1) == NULL)
goto error;
struct alloc_class *predefined_class =
alloc_class_new(-1, ac, CLASS_RUN, HEADER_COMPACT,
MIN_RUN_SIZE, 0, 1);
if (predefined_class == NULL)
goto error;
for (size_t i = 0; i < FIRST_GENERATED_CLASS_SIZE / ac->granularity;
++i) {
ac->class_map_by_alloc_size[i] = predefined_class->id;
}
/*
* Based on the defined categories, a set of allocation classes is
* created. The unit size of those classes is depended on the category
* initial size and step.
*/
size_t granularity_mask = ALLOC_BLOCK_SIZE_GEN - 1;
for (int c = 1; c < MAX_ALLOC_CATEGORIES; ++c) {
size_t n = categories[c - 1].size + ALLOC_BLOCK_SIZE_GEN;
do {
if (alloc_class_find_or_create(ac, n) == NULL)
goto error;
float stepf = (float)n * categories[c].step;
size_t stepi = (size_t)stepf;
stepi = (stepf - (float)stepi < FLT_EPSILON) ?
stepi : stepi + 1;
n += (stepi + (granularity_mask)) & ~granularity_mask;
} while (n <= categories[c].size);
}
/*
* Find the largest alloc class and use it's unit size as run allocation
* threshold.
*/
uint8_t largest_aclass_slot;
for (largest_aclass_slot = MAX_ALLOCATION_CLASSES - 1;
largest_aclass_slot > 0 &&
ac->aclasses[largest_aclass_slot] == NULL;
--largest_aclass_slot) {
/* intentional NOP */
}
struct alloc_class *c = ac->aclasses[largest_aclass_slot];
/*
* The actual run might contain less unit blocks than the theoretical
* unit max variable. This may be the case for very large unit sizes.
*/
size_t real_unit_max = c->run.bitmap_nallocs < RUN_UNIT_MAX_ALLOC ?
c->run.bitmap_nallocs : RUN_UNIT_MAX_ALLOC;
size_t theoretical_run_max_size = c->unit_size * real_unit_max;
ac->last_run_max_size = MAX_RUN_SIZE > theoretical_run_max_size ?
theoretical_run_max_size : MAX_RUN_SIZE;
#ifdef DEBUG
/*
* Verify that each bucket's unit size points back to the bucket by the
* bucket map. This must be true for the default allocation classes,
* otherwise duplicate buckets will be created.
*/
for (size_t i = 0; i < MAX_ALLOCATION_CLASSES; ++i) {
struct alloc_class *c = ac->aclasses[i];
if (c != NULL && c->type == CLASS_RUN) {
ASSERTeq(i, c->id);
ASSERTeq(alloc_class_by_run(ac, c->unit_size,
c->flags, c->run.size_idx), c);
}
}
#endif
return ac;
error:
alloc_class_collection_delete(ac);
return NULL;
}
/*
* FUNCTION: meta_replicaslice()
* INPUT: dnp - the name of the drive to check
* OUTPUT: slicep - pointer to slice number
* ep - pointer to an md_error_t structure in which
* to return errors to the caller
* RETURNS: int - 0 - value pointed to by slicep is valid
* -1 - otherwise
*
* PURPOSE: Determine which slice of the specified drive to
* reserve, presumably for metadb replica usage.
*
* NOTE: If slicep is NULL, the return code will indicate
* whether or not the slice number could be determined
*/
int
meta_replicaslice(
mddrivename_t *dnp,
uint_t *slicep,
md_error_t *ep
)
{
int err = 0;
int ioctl_return;
int fd;
char *rname;
struct dk_geom geom;
rname = dnp->rname;
if ((fd = open(rname, (O_RDONLY|O_NDELAY), 0)) < 0) {
char *n;
int open_errno;
size_t len;
if (errno != ENOENT)
return (mdsyserror(ep, errno, rname));
len = strlen(rname) + 3;
n = Zalloc(len);
(void) snprintf(n, len, "%ss0", rname);
fd = open(n, (O_RDONLY|O_NDELAY), 0);
open_errno = errno;
Free(n);
if (fd < 0) {
return (mdsyserror(ep, open_errno, rname));
}
}
/*
* if our drivenamep points to a device not supporting
* DKIOCGGEOM, we have an EFI label.
*/
errno = 0;
ioctl_return = ioctl(fd, DKIOCGGEOM, &geom);
err = errno;
(void) close(fd);
/*
* If the DKIOCGGEOM ioctl succeeded, then the device has a
* VTOC style label. In this case, we use slice 7.
*/
if (ioctl_return == 0) {
if (slicep != NULL) {
*slicep = MD_SLICE7;
}
return (0);
}
/*
* ENOTSUP indicates an EFI style label, in which case slice 7
* cannot be used because its minor number is reserved. In
* this case, use slice 6.
*/
if (err == ENOTSUP) {
if (slicep != NULL) {
*slicep = MD_SLICE6;
}
return (0);
}
/*
* Those are the only two cases we know how to deal with;
* either the drivenamep didn't point to a disk, or the ioctl
* failed for some other reason.
*/
if (err == ENOTTY) {
return (mddeverror(ep, MDE_NOT_DISK, NODEV, rname));
}
return (mdsyserror(ep, err, rname));
}
/*
* os_badblocks_get -- returns 0 and bad blocks in the 'bbs' array
* (that has to be pre-allocated)
* or -1 in case of an error
*/
int
os_badblocks_get(const char *file, struct badblocks *bbs)
{
LOG(3, "file %s badblocks %p", file, bbs);
ASSERTne(bbs, NULL);
VEC(bbsvec, struct bad_block) bbv = VEC_INITIALIZER;
struct extents *exts = NULL;
long extents = 0;
unsigned long long bb_beg;
unsigned long long bb_end;
unsigned long long bb_len;
unsigned long long bb_off;
unsigned long long ext_beg;
unsigned long long ext_end;
unsigned long long not_block_aligned;
int bb_found = -1; /* -1 means an error */
memset(bbs, 0, sizeof(*bbs));
if (os_dimm_files_namespace_badblocks(file, bbs)) {
LOG(1, "checking the file for bad blocks failed -- '%s'", file);
goto error_free_all;
}
if (bbs->bb_cnt == 0) {
bb_found = 0;
goto exit_free_all;
}
exts = Zalloc(sizeof(struct extents));
if (exts == NULL) {
ERR("!Zalloc");
goto error_free_all;
}
extents = os_extents_count(file, exts);
if (extents < 0) {
LOG(1, "counting file's extents failed -- '%s'", file);
goto error_free_all;
}
if (extents == 0) {
/* dax device has no extents */
bb_found = (int)bbs->bb_cnt;
for (unsigned b = 0; b < bbs->bb_cnt; b++) {
LOG(4, "bad block found: offset: %llu, length: %u",
bbs->bbv[b].offset,
bbs->bbv[b].length);
}
goto exit_free_all;
}
exts->extents = Zalloc(exts->extents_count * sizeof(struct extent));
if (exts->extents == NULL) {
ERR("!Zalloc");
goto error_free_all;
}
if (os_extents_get(file, exts)) {
LOG(1, "getting file's extents failed -- '%s'", file);
goto error_free_all;
}
bb_found = 0;
for (unsigned b = 0; b < bbs->bb_cnt; b++) {
bb_beg = bbs->bbv[b].offset;
bb_end = bb_beg + bbs->bbv[b].length - 1;
for (unsigned e = 0; e < exts->extents_count; e++) {
ext_beg = exts->extents[e].offset_physical;
ext_end = ext_beg + exts->extents[e].length - 1;
/* check if the bad block overlaps with file's extent */
if (bb_beg > ext_end || ext_beg > bb_end)
continue;
bb_found++;
bb_beg = (bb_beg > ext_beg) ? bb_beg : ext_beg;
bb_end = (bb_end < ext_end) ? bb_end : ext_end;
bb_len = bb_end - bb_beg + 1;
bb_off = bb_beg + exts->extents[e].offset_logical
- exts->extents[e].offset_physical;
LOG(10,
"bad block found: physical offset: %llu, length: %llu",
bb_beg, bb_len);
/* check if offset is block-aligned */
not_block_aligned = bb_off & (exts->blksize - 1);
if (not_block_aligned) {
bb_off -= not_block_aligned;
bb_len += not_block_aligned;
}
/* check if length is block-aligned */
bb_len = ALIGN_UP(bb_len, exts->blksize);
LOG(4,
"bad block found: logical offset: %llu, length: %llu",
bb_off, bb_len);
/*
* Form a new bad block structure with offset and length
* expressed in bytes and offset relative
* to the beginning of the file.
*/
struct bad_block bb;
bb.offset = bb_off;
bb.length = (unsigned)(bb_len);
/* unknown healthy replica */
bb.nhealthy = NO_HEALTHY_REPLICA;
/* add the new bad block to the vector */
if (VEC_PUSH_BACK(&bbv, bb)) {
VEC_DELETE(&bbv);
bb_found = -1;
goto error_free_all;
}
}
}
error_free_all:
Free(bbs->bbv);
bbs->bbv = NULL;
bbs->bb_cnt = 0;
exit_free_all:
if (exts) {
Free(exts->extents);
Free(exts);
}
if (extents > 0 && bb_found > 0) {
bbs->bbv = VEC_ARR(&bbv);
bbs->bb_cnt = (unsigned)VEC_SIZE(&bbv);
LOG(10, "number of bad blocks detected: %u", bbs->bb_cnt);
/* sanity check */
ASSERTeq((unsigned)bb_found, bbs->bb_cnt);
}
return (bb_found >= 0) ? 0 : -1;
}
/*
* list all the drives
*/
int
mhd_list_drives(
char *path,
mhd_did_flags_t flags,
mhd_list_res_t *resultsp,
mhd_error_t *mhep
)
{
mhd_state_t state;
uint_t ndrive, i, j, c;
/* grab lock */
mhd_mx_lock(&mhd_set_mx);
/* add path to list */
if (mhd_create_drives(path, mhep) != 0) {
mhd_mx_unlock(&mhd_set_mx);
return (-1);
}
/* get what we want */
state = 0;
if (flags & MHD_DID_SERIAL)
state |= DRIVE_SERIALING;
if (flags & MHD_DID_TIME)
state |= DRIVE_VTOCING;
if (flags & MHD_DID_CINFO)
state |= DRIVE_CINFOING;
/* ident and count drives */
for (ndrive = 0, i = 0; (i < mhd_nset); ++i) {
mhd_drive_set_t *sp = mhd_sets[i];
mhd_drive_list_t *dlp = &sp->sr_drives;
/* count drives */
ndrive += dlp->dl_ndrive;
/* ident drives */
if (state != 0) {
mhd_mx_lock(&sp->sr_mx);
for (j = 0; (j < dlp->dl_ndrive); ++j) {
mhd_drive_t *dp = dlp->dl_drives[j];
if (mhd_state_set(dp, state, mhep) != 0) {
mhd_mx_unlock(&sp->sr_mx);
mhd_mx_unlock(&mhd_set_mx);
return (-1);
}
}
mhd_wait_set(sp, dlp, state);
mhd_mx_unlock(&sp->sr_mx);
}
}
/* build list */
assert(resultsp->results.mhd_drive_info_list_t_len == 0);
assert(resultsp->results.mhd_drive_info_list_t_val == NULL);
resultsp->results.mhd_drive_info_list_t_len = ndrive;
resultsp->results.mhd_drive_info_list_t_val = Zalloc(
ndrive * sizeof (*resultsp->results.mhd_drive_info_list_t_val));
for (c = 0, i = 0; (i < mhd_nset); ++i) {
mhd_drive_set_t *sp = mhd_sets[i];
mhd_drive_list_t *dlp = &sp->sr_drives;
mhd_mx_lock(&sp->sr_mx);
for (j = 0; (j < dlp->dl_ndrive); ++j) {
mhd_drive_t *dp = dlp->dl_drives[j];
mhd_drive_info_t *ip =
&resultsp->results.mhd_drive_info_list_t_val[c++];
ip->dif_name = Strdup(dp->dr_rname);
ip->dif_id = dp->dr_drive_id;
}
mhd_mx_unlock(&sp->sr_mx);
}
assert(c == ndrive);
/* unlock, return count */
mhd_mx_unlock(&mhd_set_mx);
return (ndrive);
}
/*
* create or update new set
*/
mhd_drive_set_t *
mhd_create_set(
mhd_set_t *mhsp,
mhd_opts_t options,
mhd_drive_list_t *dlp,
mhd_error_t *mhep
)
{
char *setname;
mhd_drive_set_t *sp;
mhd_drive_list_t *sp_dlp;
mhd_drive_set_t *null_sp;
uint_t i;
/* check locks */
assert(MUTEX_HELD(&mhd_set_mx));
/* get setname */
if (mhsp == NULL)
setname = "";
else
setname = mhsp->setname;
/* find or create set */
if ((sp = mhd_find_set(setname)) == NULL) {
/* allocate and initialize set */
sp = Zalloc(sizeof (*sp));
sp->sr_name = Strdup(setname);
mhd_mx_init(&sp->sr_mx);
mhd_cv_init(&sp->sr_cv);
sp->sr_ff = -1;
/* append to set list */
++mhd_nset;
mhd_sets = Realloc(mhd_sets, (mhd_nset * sizeof (*mhd_sets)));
mhd_sets[mhd_nset - 1] = sp;
}
sp_dlp = &sp->sr_drives;
/* if just grabbing null set, return */
if (mhsp == NULL)
return (sp);
assert(strcmp(setname, "") != 0);
assert(mhep != NULL);
/* get null set */
null_sp = mhd_create_set(NULL, 0, NULL, NULL);
assert(null_sp != NULL);
assert(sp != null_sp);
/* grab set lock */
mhd_mx_lock(&sp->sr_mx);
/* save options */
if (options & MHD_SERIAL)
sp->sr_options |= MHD_SERIAL;
else
sp->sr_options &= ~MHD_SERIAL;
/* move drives no longer in set to null set */
if (! (options & MHD_PARTIAL_SET)) {
for (i = 0; (i < sp_dlp->dl_ndrive); /* void */) {
mhd_drive_t *dp = sp_dlp->dl_drives[i];
uint_t j;
/* check still there */
for (j = 0; (j < mhsp->drives.drives_len); ++j) {
mhd_drivename_t mhdp;
mhdp = mhsp->drives.drives_val[j];
if (strcmp(dp->dr_rname, mhdp) == 0)
break;
}
if (j < mhsp->drives.drives_len) {
++i;
continue;
}
/* idle the drive */
if (mhd_idle(dp, mhep) != 0)
mhd_clrerror(mhep);
/* move to null set */
mhd_del_drive_from_set(dp);
mhd_mx_unlock(&sp->sr_mx);
mhd_mx_lock(&null_sp->sr_mx);
mhd_add_drive_to_set(null_sp, dp);
mhd_mx_unlock(&null_sp->sr_mx);
mhd_mx_lock(&sp->sr_mx);
}
}
/* add new drives to lists */
for (i = 0; (i < mhsp->drives.drives_len); ++i) {
mhd_drivename_t mhdp = mhsp->drives.drives_val[i];
uint_t j;
mhd_drive_t *dp;
/* check already there */
for (j = 0; (j < dlp->dl_ndrive); ++j) {
dp = dlp->dl_drives[j];
if (strcmp(mhdp, dp->dr_rname) == 0)
break;
}
if (j < dlp->dl_ndrive) {
mhd_add_drive(dlp, dp);
continue;
}
/* add drive to set */
if ((dp = mhd_create_drive(sp, mhdp, NULL, mhep)) == NULL) {
mhde_perror(mhep, "mhd_create_drive: %s", mhdp);
continue;
}
mhd_add_drive(dlp, dp);
}
/* debug */
#ifdef MHD_DEBUG
if (mhd_debug > 0) {
for (i = 0; (i < mhd_nset); ++i) {
mhd_drive_set_t *sp = mhd_sets[i];
mhd_drive_list_t *dlp = &sp->sr_drives;
char buf[10240];
uint_t j;
(void) snprintf(buf, sizeof (buf), "set '%s':",
sp->sr_name);
for (j = 0; (j < dlp->dl_ndrive); ++j) {
mhd_drive_t *dp = dlp->dl_drives[j];
char *p;
if ((p = strrchr(dp->dr_rname, '/')) != NULL)
++p;
else
p = dp->dr_rname;
(void) strncat(buf, " ", sizeof (buf));
(void) strncat(buf, p, sizeof (buf));
}
buf[sizeof (buf) - 1] = '\0';
mhd_eprintf("%s\n", buf);
}
}
#endif /* MHD_DEBUG */
/* unlock, return set */
mhd_mx_unlock(&sp->sr_mx);
return (sp);
}
static void
list_update(list_t *list, lwp_info_t *lwp)
{
id_info_t *id;
if (list->l_head == NULL) { /* first element */
list->l_head = list->l_tail = id = Zalloc(sizeof (id_info_t));
goto update;
}
for (id = list->l_head; id; id = id->id_next) {
if ((list->l_type == LT_USERS) &&
(id->id_uid != lwp->li_info.pr_uid))
continue;
if ((list->l_type == LT_TASKS) &&
(id->id_taskid != lwp->li_info.pr_taskid))
continue;
if ((list->l_type == LT_PROJECTS) &&
(id->id_projid != lwp->li_info.pr_projid))
continue;
if ((list->l_type == LT_ZONES) &&
(id->id_zoneid != lwp->li_info.pr_zoneid))
continue;
id->id_nproc++;
id->id_taskid = lwp->li_info.pr_taskid;
id->id_projid = lwp->li_info.pr_projid;
id->id_zoneid = lwp->li_info.pr_zoneid;
if (lwp->li_flags & LWP_REPRESENT) {
id->id_size += lwp->li_info.pr_size;
id->id_rssize += lwp->li_info.pr_rssize;
}
id->id_pctcpu += FRC2PCT(lwp->li_info.pr_lwp.pr_pctcpu);
if (opts.o_outpmode & OPT_LWPS)
id->id_time += TIME2SEC(lwp->li_info.pr_lwp.pr_time);
else
id->id_time += TIME2SEC(lwp->li_info.pr_time);
id->id_pctmem += FRC2PCT(lwp->li_info.pr_pctmem);
id->id_key += lwp->li_key;
total_cpu += FRC2PCT(lwp->li_info.pr_lwp.pr_pctcpu);
total_mem += FRC2PCT(lwp->li_info.pr_pctmem);
return;
}
id = list->l_tail;
id->id_next = Zalloc(sizeof (id_info_t));
id->id_next->id_prev = list->l_tail;
id->id_next->id_next = NULL;
list->l_tail = id->id_next;
id = list->l_tail;
update:
id->id_uid = lwp->li_info.pr_uid;
id->id_projid = lwp->li_info.pr_projid;
id->id_taskid = lwp->li_info.pr_taskid;
id->id_zoneid = lwp->li_info.pr_zoneid;
id->id_nproc++;
if (lwp->li_flags & LWP_REPRESENT) {
id->id_size = lwp->li_info.pr_size;
id->id_rssize = lwp->li_info.pr_rssize;
}
id->id_pctcpu = FRC2PCT(lwp->li_info.pr_lwp.pr_pctcpu);
if (opts.o_outpmode & OPT_LWPS)
id->id_time = TIME2SEC(lwp->li_info.pr_lwp.pr_time);
else
id->id_time = TIME2SEC(lwp->li_info.pr_time);
id->id_pctmem = FRC2PCT(lwp->li_info.pr_pctmem);
id->id_key = lwp->li_key;
total_cpu += id->id_pctcpu;
total_mem += id->id_pctmem;
list->l_count++;
}