int mds_quota_ctl(struct obd_device *obd, struct obd_export *unused,
struct obd_quotactl *oqctl)
{
struct obd_device_target *obt = &obd->u.obt;
struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
struct timeval work_start;
struct timeval work_end;
long timediff;
int rc = 0;
ENTRY;
cfs_gettimeofday(&work_start);
switch (oqctl->qc_cmd) {
case Q_QUOTAON:
oqctl->qc_id = obt->obt_qfmt; /* override qfmt version */
rc = mds_quota_on(obd, oqctl);
break;
case Q_QUOTAOFF:
oqctl->qc_id = obt->obt_qfmt; /* override qfmt version */
rc = mds_quota_off(obd, oqctl);
break;
case Q_SETINFO:
rc = mds_set_dqinfo(obd, oqctl);
break;
case Q_GETINFO:
rc = mds_get_dqinfo(obd, oqctl);
break;
case Q_SETQUOTA:
rc = mds_set_dqblk(obd, oqctl);
break;
case Q_GETQUOTA:
rc = mds_get_dqblk(obd, oqctl);
break;
case Q_GETOINFO:
case Q_GETOQUOTA:
rc = mds_get_obd_quota(obd, oqctl);
break;
case LUSTRE_Q_INVALIDATE:
rc = mds_quota_invalidate(obd, oqctl);
break;
case LUSTRE_Q_FINVALIDATE:
oqctl->qc_id = obt->obt_qfmt; /* override qfmt version */
rc = mds_quota_finvalidate(obd, oqctl);
break;
default:
CERROR("%s: unsupported mds_quotactl command: %d\n",
obd->obd_name, oqctl->qc_cmd);
RETURN(-EFAULT);
}
if (rc)
CDEBUG(D_INFO, "mds_quotactl admin quota command %d, id %u, "
"type %d, failed: rc = %d\n",
oqctl->qc_cmd, oqctl->qc_id, oqctl->qc_type, rc);
cfs_gettimeofday(&work_end);
timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
lprocfs_counter_add(qctxt->lqc_stats, LQUOTA_QUOTA_CTL, timediff);
RETURN(rc);
}
static int osc_stats_seq_show(struct seq_file *seq, void *v)
{
struct timeval now;
struct obd_device *dev = seq->private;
struct osc_stats *stats = &obd2osc_dev(dev)->od_stats;
cfs_gettimeofday(&now);
seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
now.tv_sec, now.tv_usec);
seq_printf(seq, "lockless_write_bytes\t\t"LPU64"\n",
stats->os_lockless_writes);
seq_printf(seq, "lockless_read_bytes\t\t"LPU64"\n",
stats->os_lockless_reads);
seq_printf(seq, "lockless_truncate\t\t"LPU64"\n",
stats->os_lockless_truncates);
return 0;
}
static void brw_stats_show(struct seq_file *seq, struct brw_stats *brw_stats)
{
struct timeval now;
/* this sampling races with updates */
cfs_gettimeofday(&now);
seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
now.tv_sec, now.tv_usec);
display_brw_stats(seq, "pages per bulk r/w", "rpcs",
&brw_stats->hist[BRW_R_PAGES],
&brw_stats->hist[BRW_W_PAGES], 1);
display_brw_stats(seq, "discontiguous pages", "rpcs",
&brw_stats->hist[BRW_R_DISCONT_PAGES],
&brw_stats->hist[BRW_W_DISCONT_PAGES], 0);
display_brw_stats(seq, "discontiguous blocks", "rpcs",
&brw_stats->hist[BRW_R_DISCONT_BLOCKS],
&brw_stats->hist[BRW_W_DISCONT_BLOCKS], 0);
display_brw_stats(seq, "disk fragmented I/Os", "ios",
&brw_stats->hist[BRW_R_DIO_FRAGS],
&brw_stats->hist[BRW_W_DIO_FRAGS], 0);
display_brw_stats(seq, "disk I/Os in flight", "ios",
&brw_stats->hist[BRW_R_RPC_HIST],
&brw_stats->hist[BRW_W_RPC_HIST], 0);
display_brw_stats(seq, "I/O time (1/1000s)", "ios",
&brw_stats->hist[BRW_R_IO_TIME],
&brw_stats->hist[BRW_W_IO_TIME], 1000 / CFS_HZ);
display_brw_stats(seq, "disk I/O size", "ios",
&brw_stats->hist[BRW_R_DISK_IOSIZE],
&brw_stats->hist[BRW_W_DISK_IOSIZE], 1);
}
int filter_quota_ctl(struct obd_device *unused, struct obd_export *exp,
struct obd_quotactl *oqctl)
{
struct obd_device *obd = exp->exp_obd;
struct obd_device_target *obt = &obd->u.obt;
struct lvfs_run_ctxt saved;
struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
struct lustre_qunit_size *lqs;
void *handle = NULL;
struct timeval work_start;
struct timeval work_end;
long timediff;
int rc = 0;
ENTRY;
cfs_gettimeofday(&work_start);
switch (oqctl->qc_cmd) {
case Q_QUOTAON:
oqctl->qc_id = obt->obt_qfmt;
rc = generic_quota_on(obd, oqctl, 0);
break;
case Q_FINVALIDATE:
case Q_QUOTAOFF:
cfs_down(&obt->obt_quotachecking);
if (oqctl->qc_cmd == Q_FINVALIDATE &&
(obt->obt_qctxt.lqc_flags & UGQUOTA2LQC(oqctl->qc_type))) {
CWARN("quota[%u] is on yet\n", oqctl->qc_type);
cfs_up(&obt->obt_quotachecking);
rc = -EBUSY;
break;
}
oqctl->qc_id = obt->obt_qfmt; /* override qfmt version */
case Q_GETOINFO:
case Q_GETOQUOTA:
case Q_GETQUOTA:
/* In recovery scenario, this pending dqacq/dqrel might have
* been processed by master successfully before it's dquot
* on master enter recovery mode. We must wait for this
* dqacq/dqrel done then return the correct limits to master */
if (oqctl->qc_stat == QUOTA_RECOVERING)
handle = quota_barrier(&obd->u.obt.obt_qctxt, oqctl, 1);
push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
rc = fsfilt_quotactl(obd, obt->obt_sb, oqctl);
pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
if (oqctl->qc_stat == QUOTA_RECOVERING)
quota_unbarrier(handle);
if (oqctl->qc_cmd == Q_QUOTAOFF ||
oqctl->qc_cmd == Q_FINVALIDATE) {
if (oqctl->qc_cmd == Q_QUOTAOFF) {
if (!rc)
obt->obt_qctxt.lqc_flags &=
~UGQUOTA2LQC(oqctl->qc_type);
else if (quota_is_off(qctxt, oqctl))
rc = -EALREADY;
CDEBUG(D_QUOTA, "%s: quotaoff type:flags:rc "
"%u:%lu:%d\n", obd->obd_name,
oqctl->qc_type, qctxt->lqc_flags, rc);
}
cfs_up(&obt->obt_quotachecking);
}
break;
case Q_SETQUOTA:
/* currently, it is only used for nullifying the quota */
handle = quota_barrier(&obd->u.obt.obt_qctxt, oqctl, 1);
push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
if (!rc) {
oqctl->qc_cmd = Q_SYNC;
fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
oqctl->qc_cmd = Q_SETQUOTA;
}
pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
quota_unbarrier(handle);
lqs = quota_search_lqs(LQS_KEY(oqctl->qc_type, oqctl->qc_id),
qctxt, 0);
if (lqs == NULL || IS_ERR(lqs)){
CERROR("fail to create lqs during setquota operation "
"for %sid %u\n", oqctl->qc_type ? "g" : "u",
oqctl->qc_id);
} else {
lqs->lqs_flags &= ~QB_SET;
lqs_putref(lqs);
}
break;
case Q_INITQUOTA:
{
unsigned int id[MAXQUOTAS] = { 0, 0 };
/* Initialize quota limit to MIN_QLIMIT */
LASSERT(oqctl->qc_dqblk.dqb_valid == QIF_BLIMITS);
LASSERT(oqctl->qc_dqblk.dqb_bsoftlimit == 0);
if (!oqctl->qc_dqblk.dqb_bhardlimit)
goto adjust;
/* There might be a pending dqacq/dqrel (which is going to
* clear stale limits on slave). we should wait for it's
* completion then initialize limits */
handle = quota_barrier(&obd->u.obt.obt_qctxt, oqctl, 1);
LASSERT(oqctl->qc_dqblk.dqb_bhardlimit == MIN_QLIMIT);
push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
/* Update on-disk quota, in case of lose the changed limits
* (MIN_QLIMIT) on crash, which cannot be recovered.*/
if (!rc) {
oqctl->qc_cmd = Q_SYNC;
fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
oqctl->qc_cmd = Q_INITQUOTA;
}
pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
quota_unbarrier(handle);
if (rc)
RETURN(rc);
adjust:
lqs = quota_search_lqs(LQS_KEY(oqctl->qc_type, oqctl->qc_id),
qctxt, 1);
if (lqs == NULL || IS_ERR(lqs)){
CERROR("fail to create lqs during setquota operation "
"for %sid %u\n", oqctl->qc_type ? "g" : "u",
oqctl->qc_id);
break;
} else {
lqs->lqs_flags |= QB_SET;
lqs_putref(lqs);
}
/* Trigger qunit pre-acquire */
if (oqctl->qc_type == USRQUOTA)
id[USRQUOTA] = oqctl->qc_id;
else
id[GRPQUOTA] = oqctl->qc_id;
rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt,
id, 1, 0, NULL);
if (rc == -EDQUOT || rc == -EBUSY) {
CDEBUG(D_QUOTA, "rc: %d.\n", rc);
rc = 0;
}
break;
}
default:
CERROR("%s: unsupported filter_quotactl command: %d\n",
obd->obd_name, oqctl->qc_cmd);
RETURN(-EFAULT);
}
cfs_gettimeofday(&work_end);
timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
lprocfs_counter_add(qctxt->lqc_stats, LQUOTA_QUOTA_CTL, timediff);
RETURN(rc);
}
static int __init init_lustre_lite(void)
{
int i, rc, seed[2];
struct timeval tv;
lnet_process_id_t lnet_id;
CLASSERT(sizeof(LUSTRE_VOLATILE_HDR) == LUSTRE_VOLATILE_HDR_LEN + 1);
/* print an address of _any_ initialized kernel symbol from this
* module, to allow debugging with gdb that doesn't support data
* symbols from modules.*/
CDEBUG(D_INFO, "Lustre client module (%p).\n",
&lustre_super_operations);
rc = ll_init_inodecache();
if (rc)
return -ENOMEM;
ll_file_data_slab = cfs_mem_cache_create("ll_file_data",
sizeof(struct ll_file_data), 0,
CFS_SLAB_HWCACHE_ALIGN);
if (ll_file_data_slab == NULL) {
ll_destroy_inodecache();
return -ENOMEM;
}
ll_remote_perm_cachep = cfs_mem_cache_create("ll_remote_perm_cache",
sizeof(struct ll_remote_perm),
0, 0);
if (ll_remote_perm_cachep == NULL) {
cfs_mem_cache_destroy(ll_file_data_slab);
ll_file_data_slab = NULL;
ll_destroy_inodecache();
return -ENOMEM;
}
ll_rmtperm_hash_cachep = cfs_mem_cache_create("ll_rmtperm_hash_cache",
REMOTE_PERM_HASHSIZE *
sizeof(cfs_list_t),
0, 0);
if (ll_rmtperm_hash_cachep == NULL) {
cfs_mem_cache_destroy(ll_remote_perm_cachep);
ll_remote_perm_cachep = NULL;
cfs_mem_cache_destroy(ll_file_data_slab);
ll_file_data_slab = NULL;
ll_destroy_inodecache();
return -ENOMEM;
}
proc_lustre_fs_root = proc_lustre_root ?
lprocfs_register("llite", proc_lustre_root, NULL, NULL) : NULL;
lustre_register_client_fill_super(ll_fill_super);
lustre_register_kill_super_cb(ll_kill_super);
lustre_register_client_process_config(ll_process_config);
cfs_get_random_bytes(seed, sizeof(seed));
/* Nodes with small feet have little entropy
* the NID for this node gives the most entropy in the low bits */
for (i=0; ; i++) {
if (LNetGetId(i, &lnet_id) == -ENOENT) {
break;
}
if (LNET_NETTYP(LNET_NIDNET(lnet_id.nid)) != LOLND) {
seed[0] ^= LNET_NIDADDR(lnet_id.nid);
}
}
cfs_gettimeofday(&tv);
cfs_srand(tv.tv_sec ^ seed[0], tv.tv_usec ^ seed[1]);
init_timer(&ll_capa_timer);
ll_capa_timer.function = ll_capa_timer_callback;
rc = ll_capa_thread_start();
/*
* XXX normal cleanup is needed here.
*/
if (rc == 0)
rc = vvp_global_init();
return rc;
}
static int osc_rpc_stats_seq_show(struct seq_file *seq, void *v)
{
struct timeval now;
struct obd_device *dev = seq->private;
struct client_obd *cli = &dev->u.cli;
unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
int i;
cfs_gettimeofday(&now);
client_obd_list_lock(&cli->cl_loi_list_lock);
seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
now.tv_sec, now.tv_usec);
seq_printf(seq, "read RPCs in flight: %d\n",
cli->cl_r_in_flight);
seq_printf(seq, "write RPCs in flight: %d\n",
cli->cl_w_in_flight);
seq_printf(seq, "pending write pages: %d\n",
cli->cl_pending_w_pages);
seq_printf(seq, "pending read pages: %d\n",
cli->cl_pending_r_pages);
seq_printf(seq, "\n\t\t\tread\t\t\twrite\n");
seq_printf(seq, "pages per rpc rpcs %% cum %% |");
seq_printf(seq, " rpcs %% cum %%\n");
read_tot = lprocfs_oh_sum(&cli->cl_read_page_hist);
write_tot = lprocfs_oh_sum(&cli->cl_write_page_hist);
read_cum = 0;
write_cum = 0;
for (i = 0; i < OBD_HIST_MAX; i++) {
unsigned long r = cli->cl_read_page_hist.oh_buckets[i];
unsigned long w = cli->cl_write_page_hist.oh_buckets[i];
read_cum += r;
write_cum += w;
seq_printf(seq, "%d:\t\t%10lu %3lu %3lu | %10lu %3lu %3lu\n",
1 << i, r, pct(r, read_tot),
pct(read_cum, read_tot), w,
pct(w, write_tot),
pct(write_cum, write_tot));
if (read_cum == read_tot && write_cum == write_tot)
break;
}
seq_printf(seq, "\n\t\t\tread\t\t\twrite\n");
seq_printf(seq, "rpcs in flight rpcs %% cum %% |");
seq_printf(seq, " rpcs %% cum %%\n");
read_tot = lprocfs_oh_sum(&cli->cl_read_rpc_hist);
write_tot = lprocfs_oh_sum(&cli->cl_write_rpc_hist);
read_cum = 0;
write_cum = 0;
for (i = 0; i < OBD_HIST_MAX; i++) {
unsigned long r = cli->cl_read_rpc_hist.oh_buckets[i];
unsigned long w = cli->cl_write_rpc_hist.oh_buckets[i];
read_cum += r;
write_cum += w;
seq_printf(seq, "%d:\t\t%10lu %3lu %3lu | %10lu %3lu %3lu\n",
i, r, pct(r, read_tot),
pct(read_cum, read_tot), w,
pct(w, write_tot),
pct(write_cum, write_tot));
if (read_cum == read_tot && write_cum == write_tot)
break;
}
seq_printf(seq, "\n\t\t\tread\t\t\twrite\n");
seq_printf(seq, "offset rpcs %% cum %% |");
seq_printf(seq, " rpcs %% cum %%\n");
read_tot = lprocfs_oh_sum(&cli->cl_read_offset_hist);
write_tot = lprocfs_oh_sum(&cli->cl_write_offset_hist);
read_cum = 0;
write_cum = 0;
for (i = 0; i < OBD_HIST_MAX; i++) {
unsigned long r = cli->cl_read_offset_hist.oh_buckets[i];
unsigned long w = cli->cl_write_offset_hist.oh_buckets[i];
read_cum += r;
write_cum += w;
seq_printf(seq, "%d:\t\t%10lu %3lu %3lu | %10lu %3lu %3lu\n",
(i == 0) ? 0 : 1 << (i - 1),
r, pct(r, read_tot), pct(read_cum, read_tot),
w, pct(w, write_tot), pct(write_cum, write_tot));
if (read_cum == read_tot && write_cum == write_tot)
break;
}
client_obd_list_unlock(&cli->cl_loi_list_lock);
return 0;
}
/**
* when a block_write or inode_create rpc is finished, adjust the record for
* pending blocks and inodes
*/
static int quota_pending_commit(struct obd_device *obd, const unsigned int id[],
int pending[], int isblk)
{
struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
struct timeval work_start;
struct timeval work_end;
long timediff;
int i;
struct qunit_data qdata[MAXQUOTAS];
ENTRY;
CDEBUG(D_QUOTA, "commit pending quota for %s\n", obd->obd_name);
CLASSERT(MAXQUOTAS < 4);
if (!ll_sb_any_quota_active(qctxt->lqc_sb))
RETURN(0);
cfs_gettimeofday(&work_start);
for (i = 0; i < MAXQUOTAS; i++) {
struct lustre_qunit_size *lqs = NULL;
LASSERT(pending[i] >= 0);
if (pending[i] == 0)
continue;
qdata[i].qd_id = id[i];
qdata[i].qd_flags = i;
if (isblk)
QDATA_SET_BLK(&qdata[i]);
qdata[i].qd_count = 0;
if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i]))
continue;
lqs = quota_search_lqs(LQS_KEY(i, qdata[i].qd_id), qctxt, 0);
if (lqs == NULL || IS_ERR(lqs)) {
CERROR("can not find lqs for pending_commit: "
"[id %u] [%c] [pending %u] [isblk %d] (rc %ld), "
"maybe cause unexpected lqs refcount error!\n",
id[i], i ? 'g': 'u', pending[i], isblk,
lqs ? PTR_ERR(lqs) : -1);
continue;
}
cfs_spin_lock(&lqs->lqs_lock);
if (isblk) {
LASSERTF(lqs->lqs_bwrite_pending >= pending[i],
"there are too many blocks! [id %u] [%c] "
"[bwrite_pending %lu] [pending %u]\n",
id[i], i % 2 ? 'g' : 'u',
lqs->lqs_bwrite_pending, pending[i]);
lqs->lqs_bwrite_pending -= pending[i];
} else {
LASSERTF(lqs->lqs_iwrite_pending >= pending[i],
"there are too many files! [id %u] [%c] "
"[iwrite_pending %lu] [pending %u]\n",
id[i], i % 2 ? 'g' : 'u',
lqs->lqs_iwrite_pending, pending[i]);
lqs->lqs_iwrite_pending -= pending[i];
}
CDEBUG(D_QUOTA, "%s: lqs_pending=%lu pending[%d]=%d isblk=%d\n",
obd->obd_name,
isblk ? lqs->lqs_bwrite_pending : lqs->lqs_iwrite_pending,
i, pending[i], isblk);
cfs_spin_unlock(&lqs->lqs_lock);
/* for quota_search_lqs in pending_commit */
lqs_putref(lqs);
/* for quota_search_lqs in quota_check */
lqs_putref(lqs);
}
cfs_gettimeofday(&work_end);
timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
lprocfs_counter_add(qctxt->lqc_stats,
isblk ? LQUOTA_WAIT_FOR_COMMIT_BLK :
LQUOTA_WAIT_FOR_COMMIT_INO,
timediff);
RETURN(0);
}
static int quota_chk_acq_common(struct obd_device *obd, struct obd_export *exp,
const unsigned int id[], int pending[],
int count, quota_acquire acquire,
struct obd_trans_info *oti, int isblk,
struct inode *inode, int frags)
{
struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
struct timeval work_start;
struct timeval work_end;
long timediff;
struct l_wait_info lwi = { 0 };
int rc = 0, cycle = 0, count_err = 1;
ENTRY;
if (!quota_is_set(obd, id, isblk ? QB_SET : QI_SET))
RETURN(0);
if (isblk && (exp->exp_failed || exp->exp_abort_active_req))
/* If the client has been evicted or if it
* timed out and tried to reconnect already,
* abort the request immediately */
RETURN(-ENOTCONN);
CDEBUG(D_QUOTA, "check quota for %s\n", obd->obd_name);
pending[USRQUOTA] = pending[GRPQUOTA] = 0;
/* Unfortunately, if quota master is too busy to handle the
* pre-dqacq in time and quota hash on ost is used up, we
* have to wait for the completion of in flight dqacq/dqrel,
* in order to get enough quota for write b=12588 */
cfs_gettimeofday(&work_start);
while ((rc = quota_check_common(obd, id, pending, count, cycle, isblk,
inode, frags)) &
QUOTA_RET_ACQUOTA) {
cfs_spin_lock(&qctxt->lqc_lock);
if (!qctxt->lqc_import && oti) {
cfs_spin_unlock(&qctxt->lqc_lock);
LASSERT(oti && oti->oti_thread &&
oti->oti_thread->t_watchdog);
lc_watchdog_disable(oti->oti_thread->t_watchdog);
CDEBUG(D_QUOTA, "sleep for quota master\n");
l_wait_event(qctxt->lqc_wait_for_qmaster, check_qm(qctxt),
&lwi);
CDEBUG(D_QUOTA, "wake up when quota master is back\n");
lc_watchdog_touch(oti->oti_thread->t_watchdog,
CFS_GET_TIMEOUT(oti->oti_thread->t_svc));
} else {
cfs_spin_unlock(&qctxt->lqc_lock);
}
cycle++;
if (isblk)
OBD_FAIL_TIMEOUT(OBD_FAIL_OST_HOLD_WRITE_RPC, 90);
/* after acquire(), we should run quota_check_common again
* so that we confirm there are enough quota to finish write */
rc = acquire(obd, id, oti, isblk);
/* please reference to dqacq_completion for the below */
/* a new request is finished, try again */
if (rc == QUOTA_REQ_RETURNED) {
CDEBUG(D_QUOTA, "finish a quota req, try again\n");
continue;
}
/* it is out of quota already */
if (rc == -EDQUOT) {
CDEBUG(D_QUOTA, "out of quota, return -EDQUOT\n");
break;
}
/* Related quota has been disabled by master, but enabled by
* slave, do not try again. */
if (unlikely(rc == -ESRCH)) {
CERROR("mismatched quota configuration, stop try.\n");
break;
}
if (isblk && (exp->exp_failed || exp->exp_abort_active_req))
/* The client has been evicted or tried to
* to reconnect already, abort the request */
RETURN(-ENOTCONN);
/* -EBUSY and others, wait a second and try again */
if (rc < 0) {
cfs_waitq_t waitq;
struct l_wait_info lwi;
if (oti && oti->oti_thread && oti->oti_thread->t_watchdog)
lc_watchdog_touch(oti->oti_thread->t_watchdog,
CFS_GET_TIMEOUT(oti->oti_thread->t_svc));
CDEBUG(D_QUOTA, "rc: %d, count_err: %d\n", rc,
count_err++);
cfs_waitq_init(&waitq);
lwi = LWI_TIMEOUT(cfs_time_seconds(min(cycle, 10)), NULL,
NULL);
l_wait_event(waitq, 0, &lwi);
}
if (rc < 0 || cycle % 10 == 0) {
cfs_spin_lock(&last_print_lock);
if (last_print == 0 ||
cfs_time_before((last_print + cfs_time_seconds(30)),
cfs_time_current())) {
last_print = cfs_time_current();
cfs_spin_unlock(&last_print_lock);
CWARN("still haven't managed to acquire quota "
"space from the quota master after %d "
"retries (err=%d, rc=%d)\n",
cycle, count_err - 1, rc);
} else {
cfs_spin_unlock(&last_print_lock);
}
}
CDEBUG(D_QUOTA, "recheck quota with rc: %d, cycle: %d\n", rc,
cycle);
}
cfs_gettimeofday(&work_end);
timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
lprocfs_counter_add(qctxt->lqc_stats,
isblk ? LQUOTA_WAIT_FOR_CHK_BLK :
LQUOTA_WAIT_FOR_CHK_INO,
timediff);
if (rc > 0)
rc = 0;
RETURN(rc);
}
/**
* Send request \a request.
* if \a noreply is set, don't expect any reply back and don't set up
* reply buffers.
* Returns 0 on success or error code.
*/
int ptl_send_rpc(struct ptlrpc_request *request, int noreply)
{
int rc;
int rc2;
int mpflag = 0;
struct ptlrpc_connection *connection;
lnet_handle_me_t reply_me_h;
lnet_md_t reply_md;
struct obd_device *obd = request->rq_import->imp_obd;
ENTRY;
if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_DROP_RPC))
RETURN(0);
LASSERT(request->rq_type == PTL_RPC_MSG_REQUEST);
LASSERT(request->rq_wait_ctx == 0);
/* If this is a re-transmit, we're required to have disengaged
* cleanly from the previous attempt */
LASSERT(!request->rq_receiving_reply);
if (request->rq_import->imp_obd &&
request->rq_import->imp_obd->obd_fail) {
CDEBUG(D_HA, "muting rpc for failed imp obd %s\n",
request->rq_import->imp_obd->obd_name);
/* this prevents us from waiting in ptlrpc_queue_wait */
request->rq_err = 1;
request->rq_status = -ENODEV;
RETURN(-ENODEV);
}
connection = request->rq_import->imp_connection;
lustre_msg_set_handle(request->rq_reqmsg,
&request->rq_import->imp_remote_handle);
lustre_msg_set_type(request->rq_reqmsg, PTL_RPC_MSG_REQUEST);
lustre_msg_set_conn_cnt(request->rq_reqmsg,
request->rq_import->imp_conn_cnt);
lustre_msghdr_set_flags(request->rq_reqmsg,
request->rq_import->imp_msghdr_flags);
if (request->rq_resend)
lustre_msg_add_flags(request->rq_reqmsg, MSG_RESENT);
if (request->rq_memalloc)
mpflag = cfs_memory_pressure_get_and_set();
rc = sptlrpc_cli_wrap_request(request);
if (rc)
GOTO(out, rc);
/* bulk register should be done after wrap_request() */
if (request->rq_bulk != NULL) {
rc = ptlrpc_register_bulk (request);
if (rc != 0)
GOTO(out, rc);
}
if (!noreply) {
LASSERT (request->rq_replen != 0);
if (request->rq_repbuf == NULL) {
LASSERT(request->rq_repdata == NULL);
LASSERT(request->rq_repmsg == NULL);
rc = sptlrpc_cli_alloc_repbuf(request,
request->rq_replen);
if (rc) {
/* this prevents us from looping in
* ptlrpc_queue_wait */
request->rq_err = 1;
request->rq_status = rc;
GOTO(cleanup_bulk, rc);
}
} else {
request->rq_repdata = NULL;
request->rq_repmsg = NULL;
}
rc = LNetMEAttach(request->rq_reply_portal,/*XXX FIXME bug 249*/
connection->c_peer, request->rq_xid, 0,
LNET_UNLINK, LNET_INS_AFTER, &reply_me_h);
if (rc != 0) {
CERROR("LNetMEAttach failed: %d\n", rc);
LASSERT (rc == -ENOMEM);
GOTO(cleanup_bulk, rc = -ENOMEM);
}
}
spin_lock(&request->rq_lock);
/* If the MD attach succeeds, there _will_ be a reply_in callback */
request->rq_receiving_reply = !noreply;
/* We are responsible for unlinking the reply buffer */
request->rq_must_unlink = !noreply;
/* Clear any flags that may be present from previous sends. */
request->rq_replied = 0;
request->rq_err = 0;
request->rq_timedout = 0;
request->rq_net_err = 0;
request->rq_resend = 0;
request->rq_restart = 0;
request->rq_reply_truncate = 0;
spin_unlock(&request->rq_lock);
if (!noreply) {
reply_md.start = request->rq_repbuf;
reply_md.length = request->rq_repbuf_len;
/* Allow multiple early replies */
reply_md.threshold = LNET_MD_THRESH_INF;
/* Manage remote for early replies */
reply_md.options = PTLRPC_MD_OPTIONS | LNET_MD_OP_PUT |
LNET_MD_MANAGE_REMOTE |
LNET_MD_TRUNCATE; /* allow to make EOVERFLOW error */;
reply_md.user_ptr = &request->rq_reply_cbid;
reply_md.eq_handle = ptlrpc_eq_h;
/* We must see the unlink callback to unset rq_must_unlink,
so we can't auto-unlink */
rc = LNetMDAttach(reply_me_h, reply_md, LNET_RETAIN,
&request->rq_reply_md_h);
if (rc != 0) {
CERROR("LNetMDAttach failed: %d\n", rc);
LASSERT (rc == -ENOMEM);
spin_lock(&request->rq_lock);
/* ...but the MD attach didn't succeed... */
request->rq_receiving_reply = 0;
spin_unlock(&request->rq_lock);
GOTO(cleanup_me, rc = -ENOMEM);
}
CDEBUG(D_NET, "Setup reply buffer: %u bytes, xid "LPU64
", portal %u\n",
request->rq_repbuf_len, request->rq_xid,
request->rq_reply_portal);
}
/* add references on request for request_out_callback */
ptlrpc_request_addref(request);
if (obd->obd_svc_stats != NULL)
lprocfs_counter_add(obd->obd_svc_stats, PTLRPC_REQACTIVE_CNTR,
cfs_atomic_read(&request->rq_import->imp_inflight));
OBD_FAIL_TIMEOUT(OBD_FAIL_PTLRPC_DELAY_SEND, request->rq_timeout + 5);
cfs_gettimeofday(&request->rq_arrival_time);
request->rq_sent = cfs_time_current_sec();
/* We give the server rq_timeout secs to process the req, and
add the network latency for our local timeout. */
request->rq_deadline = request->rq_sent + request->rq_timeout +
ptlrpc_at_get_net_latency(request);
ptlrpc_pinger_sending_on_import(request->rq_import);
DEBUG_REQ(D_INFO, request, "send flg=%x",
lustre_msg_get_flags(request->rq_reqmsg));
rc = ptl_send_buf(&request->rq_req_md_h,
request->rq_reqbuf, request->rq_reqdata_len,
LNET_NOACK_REQ, &request->rq_req_cbid,
connection,
request->rq_request_portal,
request->rq_xid, 0);
if (rc == 0)
GOTO(out, rc);
ptlrpc_req_finished(request);
if (noreply)
GOTO(out, rc);
cleanup_me:
/* MEUnlink is safe; the PUT didn't even get off the ground, and
* nobody apart from the PUT's target has the right nid+XID to
* access the reply buffer. */
rc2 = LNetMEUnlink(reply_me_h);
LASSERT (rc2 == 0);
/* UNLINKED callback called synchronously */
LASSERT(!request->rq_receiving_reply);
cleanup_bulk:
/* We do sync unlink here as there was no real transfer here so
* the chance to have long unlink to sluggish net is smaller here. */
ptlrpc_unregister_bulk(request, 0);
out:
if (request->rq_memalloc)
cfs_memory_pressure_restore(mpflag);
return rc;
}
