static int lfsck_namespace_checkpoint(const struct lu_env *env,
struct lfsck_component *com, bool init)
{
struct lfsck_instance *lfsck = com->lc_lfsck;
struct lfsck_namespace *ns = com->lc_file_ram;
int rc;
if (com->lc_new_checked == 0 && !init)
return 0;
down_write(&com->lc_sem);
if (init) {
ns->ln_pos_latest_start = lfsck->li_pos_current;
} else {
ns->ln_pos_last_checkpoint = lfsck->li_pos_current;
ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
HALF_SEC - lfsck->li_time_last_checkpoint);
ns->ln_time_last_checkpoint = cfs_time_current_sec();
ns->ln_items_checked += com->lc_new_checked;
com->lc_new_checked = 0;
}
rc = lfsck_namespace_store(env, com, false);
up_write(&com->lc_sem);
return rc;
}
/**
* Callback handler for receiving incoming glimpse ASTs.
*
* This only can happen on client side. After handling the glimpse AST
* we also consider dropping the lock here if it is unused locally for a
* long time.
*/
static void ldlm_handle_gl_callback(struct ptlrpc_request *req,
struct ldlm_namespace *ns,
struct ldlm_request *dlm_req,
struct ldlm_lock *lock)
{
int rc = -ENOSYS;
LDLM_DEBUG(lock, "client glimpse AST callback handler");
if (lock->l_glimpse_ast != NULL)
rc = lock->l_glimpse_ast(lock, req);
if (req->rq_repmsg != NULL) {
ptlrpc_reply(req);
} else {
req->rq_status = rc;
ptlrpc_error(req);
}
lock_res_and_lock(lock);
if (lock->l_granted_mode == LCK_PW &&
!lock->l_readers && !lock->l_writers &&
cfs_time_after(cfs_time_current(),
cfs_time_add(lock->l_last_used,
cfs_time_seconds(10)))) {
unlock_res_and_lock(lock);
if (ldlm_bl_to_thread_lock(ns, NULL, lock))
ldlm_handle_bl_callback(ns, NULL, lock);
return;
}
unlock_res_and_lock(lock);
LDLM_LOCK_RELEASE(lock);
}
int osc_object_is_contended(struct osc_object *obj)
{
struct osc_device *dev = lu2osc_dev(obj->oo_cl.co_lu.lo_dev);
int osc_contention_time = dev->od_contention_time;
unsigned long cur_time = cfs_time_current();
unsigned long retry_time;
if (OBD_FAIL_CHECK(OBD_FAIL_OSC_OBJECT_CONTENTION))
return 1;
if (!obj->oo_contended)
return 0;
/*
* I like copy-paste. the code is copied from
* ll_file_is_contended.
*/
retry_time = cfs_time_add(obj->oo_contention_time,
cfs_time_seconds(osc_contention_time));
if (cfs_time_after(cur_time, retry_time)) {
osc_object_clear_contended(obj);
return 0;
}
return 1;
}
int cfs_timer_is_armed(cfs_timer_t *l)
{
if (cfs_time_before(cfs_time_current(), l->expires))
return 1;
else
return 0;
}
int
libcfs_sock_write (cfs_socket_t *sock, void *buffer, int nob, int timeout)
{
int rc;
struct pollfd pfd;
cfs_time_t start_time = cfs_time_current();
pfd.fd = sock->s_fd;
pfd.events = POLLOUT;
pfd.revents = 0;
/* poll(2) measures timeout in msec */
timeout *= 1000;
while (nob != 0 && timeout > 0) {
cfs_time_t current_time;
rc = poll(&pfd, 1, timeout);
if (rc < 0)
return -errno;
if (rc == 0)
return -ETIMEDOUT;
if ((pfd.revents & POLLOUT) == 0)
return -EIO;
rc = write(sock->s_fd, buffer, nob);
if (rc < 0)
return -errno;
if (rc == 0)
return -EIO;
buffer = ((char *)buffer) + rc;
nob -= rc;
current_time = cfs_time_current();
timeout -= 1000 *
cfs_duration_sec(cfs_time_sub(current_time,
start_time));
start_time = current_time;
}
if (nob == 0)
return 0;
else
return -ETIMEDOUT;
}
int osp_init_precreate(struct osp_device *d)
{
struct l_wait_info lwi = { 0 };
struct task_struct *task;
ENTRY;
OBD_ALLOC_PTR(d->opd_pre);
if (d->opd_pre == NULL)
RETURN(-ENOMEM);
/* initially precreation isn't ready */
d->opd_pre_status = -EAGAIN;
fid_zero(&d->opd_pre_used_fid);
d->opd_pre_used_fid.f_oid = 1;
fid_zero(&d->opd_pre_last_created_fid);
d->opd_pre_last_created_fid.f_oid = 1;
d->opd_pre_reserved = 0;
d->opd_got_disconnected = 1;
d->opd_pre_grow_slow = 0;
d->opd_pre_grow_count = OST_MIN_PRECREATE;
d->opd_pre_min_grow_count = OST_MIN_PRECREATE;
d->opd_pre_max_grow_count = OST_MAX_PRECREATE;
spin_lock_init(&d->opd_pre_lock);
init_waitqueue_head(&d->opd_pre_waitq);
init_waitqueue_head(&d->opd_pre_user_waitq);
init_waitqueue_head(&d->opd_pre_thread.t_ctl_waitq);
/*
* Initialize statfs-related things
*/
d->opd_statfs_maxage = 5; /* default update interval */
d->opd_statfs_fresh_till = cfs_time_shift(-1000);
CDEBUG(D_OTHER, "current %llu, fresh till %llu\n",
(unsigned long long)cfs_time_current(),
(unsigned long long)d->opd_statfs_fresh_till);
cfs_timer_init(&d->opd_statfs_timer, osp_statfs_timer_cb, d);
/*
* start thread handling precreation and statfs updates
*/
task = kthread_run(osp_precreate_thread, d,
"osp-pre-%u-%u", d->opd_index, d->opd_group);
if (IS_ERR(task)) {
CERROR("can't start precreate thread %ld\n", PTR_ERR(task));
RETURN(PTR_ERR(task));
}
l_wait_event(d->opd_pre_thread.t_ctl_waitq,
osp_precreate_running(d) || osp_precreate_stopped(d),
&lwi);
RETURN(0);
}
ssize_t
libcfs_sock_read(struct lnet_xport *lx, void *buffer, size_t nob,
int timeout)
{
ssize_t rc;
struct pollfd pfd;
cfs_time_t start_time = cfs_time_current();
pfd.fd = lx->lx_fd;
pfd.events = POLLIN;
pfd.revents = 0;
/* poll(2) measures timeout in msec */
timeout *= 1000;
while (nob != 0 && timeout > 0) {
rc = poll(&pfd, 1, timeout);
if (rc < 0)
return -errno;
if (rc == 0)
return -ETIMEDOUT;
if ((pfd.revents & POLLIN) == 0)
return -EIO;
rc = read(lx->lx_fd, buffer, nob);
if (rc < 0)
return -errno;
if (rc == 0)
return -EIO;
buffer = ((char *)buffer) + rc;
nob -= rc;
timeout -= cfs_duration_sec(cfs_time_sub(cfs_time_current(),
start_time));
}
if (nob == 0)
return 0;
else
return -ETIMEDOUT;
}
ssize_t
libcfs_ssl_sock_read(struct lnet_xport *lx, void *buf, size_t n,
int timeout)
{
cfs_time_t start_time = cfs_time_current();
struct pollfd pfd;
ssize_t rc;
pfd.fd = lx->lx_fd;
pfd.events = POLLIN;
pfd.revents = 0;
/* poll(2) measures timeout in msec */
timeout *= 1000;
while (n != 0 && timeout > 0) {
rc = poll(&pfd, 1, timeout);
if (rc < 0)
return (-errno);
if (rc == 0)
return (-ETIMEDOUT);
if ((pfd.revents & POLLIN) == 0)
return (-EIO);
rc = SSL_read(lx->lx_ssl, buf, n);
if (rc < 0)
return (-errno);
if (rc == 0)
return (-EIO);
buf = (char *)buf + rc;
n -= rc;
timeout -= cfs_duration_sec(cfs_time_sub(
cfs_time_current(), start_time));
}
if (n == 0)
return (0);
return (-ETIMEDOUT);
}
static ssize_t osd_read(const struct lu_env *env, struct dt_object *dt,
struct lu_buf *buf, loff_t *pos)
{
struct osd_object *obj = osd_dt_obj(dt);
struct osd_device *osd = osd_obj2dev(obj);
uint64_t old_size;
int size = buf->lb_len;
int rc;
unsigned long start;
LASSERT(dt_object_exists(dt));
LASSERT(obj->oo_db);
start = cfs_time_current();
read_lock(&obj->oo_attr_lock);
old_size = obj->oo_attr.la_size;
read_unlock(&obj->oo_attr_lock);
if (*pos + size > old_size) {
if (old_size < *pos)
return 0;
else
size = old_size - *pos;
}
record_start_io(osd, READ, 0);
rc = -dmu_read(osd->od_os, obj->oo_db->db_object, *pos, size,
buf->lb_buf, DMU_READ_PREFETCH);
record_end_io(osd, READ, cfs_time_current() - start, size,
size >> PAGE_CACHE_SHIFT);
if (rc == 0) {
rc = size;
*pos += size;
}
return rc;
}
/* Read from wire as much data as possible.
* Returns 0 or 1 on succsess, <0 if error or EOF.
* 0 means partial read, 1 - complete */
int
usocklnd_read_data(usock_conn_t *conn)
{
struct iovec *iov;
int nob;
cfs_time_t t;
LASSERT (conn->uc_rx_nob_wanted != 0);
do {
usock_peer_t *peer = conn->uc_peer;
LASSERT (conn->uc_rx_niov > 0);
nob = libcfs_sock_readv(conn->uc_sock,
conn->uc_rx_iov, conn->uc_rx_niov);
if (nob <= 0) {/* read nothing or error */
if (nob < 0)
conn->uc_errored = 1;
return nob;
}
LASSERT (nob <= conn->uc_rx_nob_wanted);
conn->uc_rx_nob_wanted -= nob;
conn->uc_rx_nob_left -= nob;
t = cfs_time_current();
conn->uc_rx_deadline = cfs_time_add(t, cfs_time_seconds(usock_tuns.ut_timeout));
if(peer != NULL)
peer->up_last_alive = t;
/* "consume" iov */
iov = conn->uc_rx_iov;
do {
LASSERT (conn->uc_rx_niov > 0);
if (nob < iov->iov_len) {
iov->iov_base = (void *)(((unsigned long)(iov->iov_base)) + nob);
iov->iov_len -= nob;
break;
}
nob -= iov->iov_len;
conn->uc_rx_iov = ++iov;
conn->uc_rx_niov--;
} while (nob != 0);
} while (conn->uc_rx_nob_wanted != 0);
return 1; /* read complete */
}
struct lc_watchdog *lc_watchdog_add(int timeout,
void (*callback)(pid_t, void *),
void *data)
{
struct lc_watchdog *lcw = NULL;
ENTRY;
LIBCFS_ALLOC(lcw, sizeof(*lcw));
if (lcw == NULL) {
CDEBUG(D_INFO, "Could not allocate new lc_watchdog\n");
RETURN(ERR_PTR(-ENOMEM));
}
cfs_spin_lock_init(&lcw->lcw_lock);
lcw->lcw_refcount = 1; /* refcount for owner */
lcw->lcw_task = cfs_current();
lcw->lcw_pid = cfs_curproc_pid();
lcw->lcw_callback = (callback != NULL) ? callback : lc_watchdog_dumplog;
lcw->lcw_data = data;
lcw->lcw_state = LC_WATCHDOG_DISABLED;
CFS_INIT_LIST_HEAD(&lcw->lcw_list);
cfs_timer_init(&lcw->lcw_timer, lcw_cb, lcw);
cfs_down(&lcw_refcount_sem);
if (++lcw_refcount == 1)
lcw_dispatch_start();
cfs_up(&lcw_refcount_sem);
/* Keep this working in case we enable them by default */
if (lcw->lcw_state == LC_WATCHDOG_ENABLED) {
lcw->lcw_last_touched = cfs_time_current();
cfs_timer_arm(&lcw->lcw_timer, cfs_time_seconds(timeout) +
cfs_time_current());
}
RETURN(lcw);
}
int osp_init_precreate(struct osp_device *d)
{
struct l_wait_info lwi = { 0 };
int rc;
ENTRY;
/* initially precreation isn't ready */
d->opd_pre_status = -EAGAIN;
fid_zero(&d->opd_pre_used_fid);
d->opd_pre_used_fid.f_oid = 1;
fid_zero(&d->opd_pre_last_created_fid);
d->opd_pre_last_created_fid.f_oid = 1;
d->opd_pre_reserved = 0;
d->opd_got_disconnected = 1;
d->opd_pre_grow_slow = 0;
d->opd_pre_grow_count = OST_MIN_PRECREATE;
d->opd_pre_min_grow_count = OST_MIN_PRECREATE;
d->opd_pre_max_grow_count = OST_MAX_PRECREATE;
spin_lock_init(&d->opd_pre_lock);
cfs_waitq_init(&d->opd_pre_waitq);
cfs_waitq_init(&d->opd_pre_user_waitq);
cfs_waitq_init(&d->opd_pre_thread.t_ctl_waitq);
/*
* Initialize statfs-related things
*/
d->opd_statfs_maxage = 5; /* default update interval */
d->opd_statfs_fresh_till = cfs_time_shift(-1000);
CDEBUG(D_OTHER, "current %llu, fresh till %llu\n",
(unsigned long long)cfs_time_current(),
(unsigned long long)d->opd_statfs_fresh_till);
cfs_timer_init(&d->opd_statfs_timer, osp_statfs_timer_cb, d);
/*
* start thread handling precreation and statfs updates
*/
rc = cfs_create_thread(osp_precreate_thread, d, 0);
if (rc < 0) {
CERROR("can't start precreate thread %d\n", rc);
RETURN(rc);
}
l_wait_event(d->opd_pre_thread.t_ctl_waitq,
osp_precreate_running(d) || osp_precreate_stopped(d),
&lwi);
RETURN(0);
}
static int lfsck_namespace_post(const struct lu_env *env,
struct lfsck_component *com,
int result, bool init)
{
struct lfsck_instance *lfsck = com->lc_lfsck;
struct lfsck_namespace *ns = com->lc_file_ram;
int rc;
down_write(&com->lc_sem);
spin_lock(&lfsck->li_lock);
if (!init)
ns->ln_pos_last_checkpoint = lfsck->li_pos_current;
if (result > 0) {
ns->ln_status = LS_SCANNING_PHASE2;
ns->ln_flags |= LF_SCANNED_ONCE;
ns->ln_flags &= ~LF_UPGRADE;
cfs_list_del_init(&com->lc_link);
cfs_list_del_init(&com->lc_link_dir);
cfs_list_add_tail(&com->lc_link, &lfsck->li_list_double_scan);
} else if (result == 0) {
ns->ln_status = lfsck->li_status;
if (ns->ln_status == 0)
ns->ln_status = LS_STOPPED;
if (ns->ln_status != LS_PAUSED) {
cfs_list_del_init(&com->lc_link);
cfs_list_del_init(&com->lc_link_dir);
cfs_list_add_tail(&com->lc_link, &lfsck->li_list_idle);
}
} else {
ns->ln_status = LS_FAILED;
cfs_list_del_init(&com->lc_link);
cfs_list_del_init(&com->lc_link_dir);
cfs_list_add_tail(&com->lc_link, &lfsck->li_list_idle);
}
spin_unlock(&lfsck->li_lock);
if (!init) {
ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
HALF_SEC - lfsck->li_time_last_checkpoint);
ns->ln_time_last_checkpoint = cfs_time_current_sec();
ns->ln_items_checked += com->lc_new_checked;
com->lc_new_checked = 0;
}
rc = lfsck_namespace_store(env, com, false);
up_write(&com->lc_sem);
return rc;
}
/* Send as much tx data as possible.
* Returns 0 or 1 on succsess, <0 if fatal error.
* 0 means partial send or non-fatal error, 1 - complete.
* Rely on libcfs_sock_writev() for differentiating fatal and
* non-fatal errors. An error should be considered as non-fatal if:
* 1) it still makes sense to continue reading &&
* 2) anyway, poll() will set up POLLHUP|POLLERR flags */
int
usocklnd_send_tx(usock_conn_t *conn, usock_tx_t *tx)
{
struct iovec *iov;
int nob;
cfs_time_t t;
LASSERT (tx->tx_resid != 0);
do {
usock_peer_t *peer = conn->uc_peer;
LASSERT (tx->tx_niov > 0);
nob = libcfs_sock_writev(conn->uc_sock,
tx->tx_iov, tx->tx_niov);
if (nob < 0)
conn->uc_errored = 1;
if (nob <= 0) /* write queue is flow-controlled or error */
return nob;
LASSERT (nob <= tx->tx_resid);
tx->tx_resid -= nob;
t = cfs_time_current();
conn->uc_tx_deadline = cfs_time_add(t, cfs_time_seconds(usock_tuns.ut_timeout));
if(peer != NULL)
peer->up_last_alive = t;
/* "consume" iov */
iov = tx->tx_iov;
do {
LASSERT (tx->tx_niov > 0);
if (nob < iov->iov_len) {
iov->iov_base = (void *)(((unsigned long)(iov->iov_base)) + nob);
iov->iov_len -= nob;
break;
}
nob -= iov->iov_len;
tx->tx_iov = ++iov;
tx->tx_niov--;
} while (nob != 0);
} while (tx->tx_resid != 0);
return 1; /* send complete */
}
void lc_watchdog_touch(struct lc_watchdog *lcw, int timeout)
{
ENTRY;
LASSERT(lcw != NULL);
lc_watchdog_del_pending(lcw);
lcw_update_time(lcw, "resumed");
lcw->lcw_state = LC_WATCHDOG_ENABLED;
cfs_timer_arm(&lcw->lcw_timer, cfs_time_current() +
cfs_time_seconds(timeout));
EXIT;
}
static void lcw_dump_stack(struct lc_watchdog *lcw)
{
cfs_time_t current_time;
cfs_duration_t delta_time;
struct timeval timediff;
current_time = cfs_time_current();
delta_time = cfs_time_sub(current_time, lcw->lcw_last_touched);
cfs_duration_usec(delta_time, &timediff);
/*
* Check to see if we should throttle the watchdog timer to avoid
* too many dumps going to the console thus triggering an NMI.
*/
delta_time = cfs_duration_sec(cfs_time_sub(current_time,
lcw_last_watchdog_time));
if (delta_time < libcfs_watchdog_ratelimit &&
lcw_recent_watchdog_count > 3) {
LCONSOLE_WARN("Service thread pid %u was inactive for "
"%lu.%.02lus. Watchdog stack traces are limited "
"to 3 per %d seconds, skipping this one.\n",
(int)lcw->lcw_pid,
timediff.tv_sec,
timediff.tv_usec / 10000,
libcfs_watchdog_ratelimit);
} else {
if (delta_time < libcfs_watchdog_ratelimit) {
lcw_recent_watchdog_count++;
} else {
memcpy(&lcw_last_watchdog_time, ¤t_time,
sizeof(current_time));
lcw_recent_watchdog_count = 0;
}
LCONSOLE_WARN("Service thread pid %u was inactive for "
"%lu.%.02lus. The thread might be hung, or it "
"might only be slow and will resume later. "
"Dumping the stack trace for debugging purposes:"
"\n",
(int)lcw->lcw_pid,
timediff.tv_sec,
timediff.tv_usec / 10000);
lcw_dump(lcw);
}
}
static void ctx_start_timer_kr(struct ptlrpc_cli_ctx *ctx, long timeout)
{
struct gss_cli_ctx_keyring *gctx_kr = ctx2gctx_keyring(ctx);
struct timer_list *timer = gctx_kr->gck_timer;
LASSERT(timer);
CDEBUG(D_SEC, "ctx %p: start timer %lds\n", ctx, timeout);
timeout = msecs_to_jiffies(timeout * MSEC_PER_SEC) +
cfs_time_current();
init_timer(timer);
timer->expires = timeout;
timer->data = (unsigned long ) ctx;
timer->function = ctx_upcall_timeout_kr;
add_timer(timer);
}
static void lcw_update_time(struct lc_watchdog *lcw, const char *message)
{
cfs_time_t newtime = cfs_time_current();;
if (lcw->lcw_state == LC_WATCHDOG_EXPIRED) {
struct timeval timediff;
cfs_time_t delta_time = cfs_time_sub(newtime,
lcw->lcw_last_touched);
cfs_duration_usec(delta_time, &timediff);
LCONSOLE_WARN("Service thread pid %u %s after %lu.%.02lus. "
"This indicates the system was overloaded (too "
"many service threads, or there were not enough "
"hardware resources).\n",
lcw->lcw_pid,
message,
timediff.tv_sec,
timediff.tv_usec / 10000);
}
lcw->lcw_last_touched = newtime;
}
/**
* Prepare buffers for read.
*
* The function maps the range described by \a off and \a len to \a lnb array.
* dmu_buf_hold_array_by_bonus() finds/creates appropriate ARC buffers, then
* we fill \a lnb array with the pages storing ARC buffers. Notice the current
* implementationt passes TRUE to dmu_buf_hold_array_by_bonus() to fill ARC
* buffers with actual data, I/O is done in the conext of osd_bufs_get_read().
* A better implementation would just return the buffers (potentially unfilled)
* and subsequent osd_read_prep() would do I/O for many ranges concurrently.
*
* \param[in] env environment
* \param[in] obj object
* \param[in] off offset in bytes
* \param[in] len the number of bytes to access
* \param[out] lnb array of local niobufs pointing to the buffers with data
*
* \retval 0 for success
* \retval negative error number of failure
*/
static int osd_bufs_get_read(const struct lu_env *env, struct osd_object *obj,
loff_t off, ssize_t len, struct niobuf_local *lnb)
{
struct osd_device *osd = osd_obj2dev(obj);
unsigned long start = cfs_time_current();
int rc, i, numbufs, npages = 0;
dmu_buf_t **dbp;
ENTRY;
record_start_io(osd, READ, 0);
/* grab buffers for read:
* OSD API let us to grab buffers first, then initiate IO(s)
* so that all required IOs will be done in parallel, but at the
* moment DMU doesn't provide us with a method to grab buffers.
* If we discover this is a vital for good performance we
* can get own replacement for dmu_buf_hold_array_by_bonus().
*/
while (len > 0) {
rc = -dmu_buf_hold_array_by_bonus(obj->oo_db, off, len, TRUE,
osd_zerocopy_tag, &numbufs,
&dbp);
if (unlikely(rc))
GOTO(err, rc);
for (i = 0; i < numbufs; i++) {
int bufoff, tocpy, thispage;
void *dbf = dbp[i];
LASSERT(len > 0);
atomic_inc(&osd->od_zerocopy_pin);
bufoff = off - dbp[i]->db_offset;
tocpy = min_t(int, dbp[i]->db_size - bufoff, len);
/* kind of trick to differentiate dbuf vs. arcbuf */
LASSERT(((unsigned long)dbp[i] & 1) == 0);
dbf = (void *) ((unsigned long)dbp[i] | 1);
while (tocpy > 0) {
thispage = PAGE_CACHE_SIZE;
thispage -= bufoff & (PAGE_CACHE_SIZE - 1);
thispage = min(tocpy, thispage);
lnb->lnb_rc = 0;
lnb->lnb_file_offset = off;
lnb->lnb_page_offset = bufoff & ~PAGE_MASK;
lnb->lnb_len = thispage;
lnb->lnb_page = kmem_to_page(dbp[i]->db_data +
bufoff);
/* mark just a single slot: we need this
* reference to dbuf to be released once */
lnb->lnb_data = dbf;
dbf = NULL;
tocpy -= thispage;
len -= thispage;
bufoff += thispage;
off += thispage;
npages++;
lnb++;
}
/* steal dbuf so dmu_buf_rele_array() can't release
* it */
dbp[i] = NULL;
}
dmu_buf_rele_array(dbp, numbufs, osd_zerocopy_tag);
}
record_end_io(osd, READ, cfs_time_current() - start,
npages * PAGE_SIZE, npages);
RETURN(npages);
err:
LASSERT(rc < 0);
osd_bufs_put(env, &obj->oo_dt, lnb - npages, npages);
RETURN(rc);
}
/* All actions that we need after receiving hello on passive conn:
* 1) Stash peer's nid, pid, incarnation and conn type
* 2) Cope with easy case: conn[idx] is empty - just save conn there
* 3) Resolve race:
* a) if our nid is higher - reply with CONN_NONE and make us zombie
* b) if peer's nid is higher - postpone race resolution till
* READY state
* 4) Anyhow, send reply hello
*/
int
usocklnd_passiveconn_hellorecv(usock_conn_t *conn)
{
ksock_hello_msg_t *hello = conn->uc_rx_hello;
int type;
int idx;
int rc;
usock_peer_t *peer;
lnet_ni_t *ni = conn->uc_ni;
__u32 peer_ip = conn->uc_peer_ip;
__u16 peer_port = conn->uc_peer_port;
/* don't know parent peer yet and not zombie */
LASSERT (conn->uc_peer == NULL &&
ni != NULL);
/* don't know peer's nid and incarnation yet */
if (peer_port > LNET_ACCEPTOR_MAX_RESERVED_PORT) {
/* do not trust liblustre clients */
conn->uc_peerid.pid = peer_port | LNET_PID_USERFLAG;
conn->uc_peerid.nid = LNET_MKNID(LNET_NIDNET(ni->ni_nid),
peer_ip);
if (hello->kshm_ctype != SOCKLND_CONN_ANY) {
lnet_ni_decref(ni);
conn->uc_ni = NULL;
CERROR("Refusing to accept connection of type=%d from "
"userspace process %u.%u.%u.%u:%d\n", hello->kshm_ctype,
HIPQUAD(peer_ip), peer_port);
return -EINVAL;
}
} else {
conn->uc_peerid.pid = hello->kshm_src_pid;
conn->uc_peerid.nid = hello->kshm_src_nid;
}
conn->uc_type = type = usocklnd_invert_type(hello->kshm_ctype);
rc = usocklnd_find_or_create_peer(ni, conn->uc_peerid, &peer);
if (rc) {
lnet_ni_decref(ni);
conn->uc_ni = NULL;
return rc;
}
peer->up_last_alive = cfs_time_current();
idx = usocklnd_type2idx(conn->uc_type);
/* safely check whether we're first */
pthread_mutex_lock(&peer->up_lock);
usocklnd_cleanup_stale_conns(peer, hello->kshm_src_incarnation, NULL);
if (peer->up_conns[idx] == NULL) {
peer->up_last_alive = cfs_time_current();
conn->uc_peer = peer;
conn->uc_ni = NULL;
usocklnd_link_conn_to_peer(conn, peer, idx);
usocklnd_conn_addref(conn);
} else {
usocklnd_peer_decref(peer);
/* Resolve race in favour of higher NID */
if (conn->uc_peerid.nid < conn->uc_ni->ni_nid) {
/* make us zombie */
conn->uc_ni = NULL;
type = SOCKLND_CONN_NONE;
}
/* if conn->uc_peerid.nid > conn->uc_ni->ni_nid,
* postpone race resolution till READY state
* (hopefully that conn[idx] will die because of
* incoming hello of CONN_NONE type) */
}
pthread_mutex_unlock(&peer->up_lock);
/* allocate and initialize fake tx with hello */
conn->uc_tx_hello = usocklnd_create_hello_tx(ni, type,
conn->uc_peerid.nid);
if (conn->uc_ni == NULL)
lnet_ni_decref(ni);
if (conn->uc_tx_hello == NULL)
return -ENOMEM;
/* rc == 0 */
pthread_mutex_lock(&conn->uc_lock);
if (conn->uc_state == UC_DEAD)
goto passive_hellorecv_done;
conn->uc_state = UC_SENDING_HELLO;
conn->uc_tx_deadline = cfs_time_shift(usock_tuns.ut_timeout);
conn->uc_tx_flag = 1;
rc = usocklnd_add_pollrequest(conn, POLL_SET_REQUEST, POLLOUT);
passive_hellorecv_done:
pthread_mutex_unlock(&conn->uc_lock);
return rc;
}
/*
* we allocate the requested pages atomically.
*/
int sptlrpc_enc_pool_get_pages(struct ptlrpc_bulk_desc *desc)
{
cfs_waitlink_t waitlink;
unsigned long this_idle = -1;
cfs_time_t tick = 0;
long now;
int p_idx, g_idx;
int i;
LASSERT(desc->bd_iov_count > 0);
LASSERT(desc->bd_iov_count <= page_pools.epp_max_pages);
/* resent bulk, enc iov might have been allocated previously */
if (desc->bd_enc_iov != NULL)
return 0;
OBD_ALLOC(desc->bd_enc_iov,
desc->bd_iov_count * sizeof(*desc->bd_enc_iov));
if (desc->bd_enc_iov == NULL)
return -ENOMEM;
cfs_spin_lock(&page_pools.epp_lock);
page_pools.epp_st_access++;
again:
if (unlikely(page_pools.epp_free_pages < desc->bd_iov_count)) {
if (tick == 0)
tick = cfs_time_current();
now = cfs_time_current_sec();
page_pools.epp_st_missings++;
page_pools.epp_pages_short += desc->bd_iov_count;
if (enc_pools_should_grow(desc->bd_iov_count, now)) {
page_pools.epp_growing = 1;
cfs_spin_unlock(&page_pools.epp_lock);
enc_pools_add_pages(page_pools.epp_pages_short / 2);
cfs_spin_lock(&page_pools.epp_lock);
page_pools.epp_growing = 0;
enc_pools_wakeup();
} else {
if (++page_pools.epp_waitqlen >
page_pools.epp_st_max_wqlen)
page_pools.epp_st_max_wqlen =
page_pools.epp_waitqlen;
cfs_set_current_state(CFS_TASK_UNINT);
cfs_waitlink_init(&waitlink);
cfs_waitq_add(&page_pools.epp_waitq, &waitlink);
cfs_spin_unlock(&page_pools.epp_lock);
cfs_waitq_wait(&waitlink, CFS_TASK_UNINT);
cfs_waitq_del(&page_pools.epp_waitq, &waitlink);
LASSERT(page_pools.epp_waitqlen > 0);
cfs_spin_lock(&page_pools.epp_lock);
page_pools.epp_waitqlen--;
}
LASSERT(page_pools.epp_pages_short >= desc->bd_iov_count);
page_pools.epp_pages_short -= desc->bd_iov_count;
this_idle = 0;
goto again;
}
/* record max wait time */
if (unlikely(tick != 0)) {
tick = cfs_time_current() - tick;
if (tick > page_pools.epp_st_max_wait)
page_pools.epp_st_max_wait = tick;
}
/* proceed with rest of allocation */
page_pools.epp_free_pages -= desc->bd_iov_count;
p_idx = page_pools.epp_free_pages / PAGES_PER_POOL;
g_idx = page_pools.epp_free_pages % PAGES_PER_POOL;
for (i = 0; i < desc->bd_iov_count; i++) {
LASSERT(page_pools.epp_pools[p_idx][g_idx] != NULL);
desc->bd_enc_iov[i].kiov_page =
page_pools.epp_pools[p_idx][g_idx];
page_pools.epp_pools[p_idx][g_idx] = NULL;
if (++g_idx == PAGES_PER_POOL) {
p_idx++;
g_idx = 0;
}
}
if (page_pools.epp_free_pages < page_pools.epp_st_lowfree)
page_pools.epp_st_lowfree = page_pools.epp_free_pages;
/*
* new idle index = (old * weight + new) / (weight + 1)
*/
if (this_idle == -1) {
this_idle = page_pools.epp_free_pages * IDLE_IDX_MAX /
page_pools.epp_total_pages;
}
page_pools.epp_idle_idx = (page_pools.epp_idle_idx * IDLE_IDX_WEIGHT +
this_idle) /
(IDLE_IDX_WEIGHT + 1);
page_pools.epp_last_access = cfs_time_current_sec();
cfs_spin_unlock(&page_pools.epp_lock);
return 0;
}
int
libcfs_sock_read (cfs_socket_t *sock, void *buffer, int nob, int timeout)
{
size_t rcvlen;
int rc;
cfs_duration_t to = cfs_time_seconds(timeout);
cfs_time_t then;
struct timeval tv;
LASSERT(nob > 0);
for (;;) {
struct iovec iov = {
.iov_base = buffer,
.iov_len = nob
};
struct msghdr msg = {
.msg_name = NULL,
.msg_namelen = 0,
.msg_iov = &iov,
.msg_iovlen = 1,
.msg_control = NULL,
.msg_controllen = 0,
.msg_flags = 0,
};
cfs_duration_usec(to, &tv);
rc = -sock_setsockopt(C2B_SOCK(sock), SOL_SOCKET, SO_RCVTIMEO,
&tv, sizeof(tv));
if (rc != 0) {
CERROR("Can't set socket recv timeout "
"%ld.%06d: %d\n",
(long)tv.tv_sec, (int)tv.tv_usec, rc);
return rc;
}
then = cfs_time_current();
rc = -sock_receive(C2B_SOCK(sock), &msg, 0, &rcvlen);
to -= cfs_time_current() - then;
if (rc != 0 && rc != -EWOULDBLOCK)
return rc;
if (rcvlen == nob)
return 0;
if (to <= 0)
return -EAGAIN;
buffer = ((char *)buffer) + rcvlen;
nob -= rcvlen;
}
return 0;
}
int
libcfs_sock_write (cfs_socket_t *sock, void *buffer, int nob, int timeout)
{
size_t sndlen;
int rc;
cfs_duration_t to = cfs_time_seconds(timeout);
cfs_time_t then;
struct timeval tv;
LASSERT(nob > 0);
for (;;) {
struct iovec iov = {
.iov_base = buffer,
.iov_len = nob
};
struct msghdr msg = {
.msg_name = NULL,
.msg_namelen = 0,
.msg_iov = &iov,
.msg_iovlen = 1,
.msg_control = NULL,
.msg_controllen = 0,
.msg_flags = (timeout == 0) ? MSG_DONTWAIT : 0,
};
if (timeout != 0) {
cfs_duration_usec(to, &tv);
rc = -sock_setsockopt(C2B_SOCK(sock), SOL_SOCKET, SO_SNDTIMEO,
&tv, sizeof(tv));
if (rc != 0) {
CERROR("Can't set socket send timeout "
"%ld.%06d: %d\n",
(long)tv.tv_sec, (int)tv.tv_usec, rc);
return rc;
}
}
then = cfs_time_current();
rc = -sock_send(C2B_SOCK(sock), &msg,
((timeout == 0) ? MSG_DONTWAIT : 0), &sndlen);
to -= cfs_time_current() - then;
if (rc != 0 && rc != -EWOULDBLOCK)
return rc;
if (sndlen == nob)
return 0;
if (to <= 0)
return -EAGAIN;
buffer = ((char *)buffer) + sndlen;
nob -= sndlen;
}
return 0;
}
int
libcfs_sock_getaddr (cfs_socket_t *sock, int remote, __u32 *ip, int *port)
{
struct sockaddr_in sin;
int rc;
if (remote != 0)
/* Get remote address */
rc = -sock_getpeername(C2B_SOCK(sock), (struct sockaddr *)&sin, sizeof(sin));
else
/* Get local address */
rc = -sock_getsockname(C2B_SOCK(sock), (struct sockaddr *)&sin, sizeof(sin));
if (rc != 0) {
CERROR ("Error %d getting sock %s IP/port\n",
rc, remote ? "peer" : "local");
return rc;
}
if (ip != NULL)
*ip = ntohl (sin.sin_addr.s_addr);
if (port != NULL)
*port = ntohs (sin.sin_port);
return 0;
}
int
libcfs_sock_setbuf (cfs_socket_t *sock, int txbufsize, int rxbufsize)
{
int option;
int rc;
if (txbufsize != 0) {
option = txbufsize;
rc = -sock_setsockopt(C2B_SOCK(sock), SOL_SOCKET, SO_SNDBUF,
(char *)&option, sizeof (option));
if (rc != 0) {
CERROR ("Can't set send buffer %d: %d\n",
option, rc);
return (rc);
}
}
if (rxbufsize != 0) {
option = rxbufsize;
rc = -sock_setsockopt (C2B_SOCK(sock), SOL_SOCKET, SO_RCVBUF,
(char *)&option, sizeof (option));
if (rc != 0) {
CERROR ("Can't set receive buffer %d: %d\n",
option, rc);
return (rc);
}
}
return 0;
}
int ll_update_remote_perm(struct inode *inode, struct mdt_remote_perm *perm)
{
struct ll_inode_info *lli = ll_i2info(inode);
struct ll_remote_perm *lrp = NULL, *tmp = NULL;
struct hlist_head *head, *perm_hash = NULL;
LASSERT(ll_i2sbi(inode)->ll_flags & LL_SBI_RMT_CLIENT);
#if 0
if (perm->rp_uid != current->uid ||
perm->rp_gid != current->gid ||
perm->rp_fsuid != current->fsuid ||
perm->rp_fsgid != current->fsgid) {
/* user might setxid in this small period */
CDEBUG(D_SEC,
"remote perm user %u/%u/%u/%u != current %u/%u/%u/%u\n",
perm->rp_uid, perm->rp_gid, perm->rp_fsuid,
perm->rp_fsgid, current->uid, current->gid,
current->fsuid, current->fsgid);
return -EAGAIN;
}
#endif
if (!lli->lli_remote_perms) {
perm_hash = alloc_rmtperm_hash();
if (perm_hash == NULL) {
CERROR("alloc lli_remote_perms failed!\n");
return -ENOMEM;
}
}
spin_lock(&lli->lli_lock);
if (!lli->lli_remote_perms)
lli->lli_remote_perms = perm_hash;
else if (perm_hash)
free_rmtperm_hash(perm_hash);
head = lli->lli_remote_perms + remote_perm_hashfunc(perm->rp_uid);
again:
hlist_for_each_entry(tmp, head, lrp_list) {
if (tmp->lrp_uid != perm->rp_uid)
continue;
if (tmp->lrp_gid != perm->rp_gid)
continue;
if (tmp->lrp_fsuid != perm->rp_fsuid)
continue;
if (tmp->lrp_fsgid != perm->rp_fsgid)
continue;
if (lrp)
free_ll_remote_perm(lrp);
lrp = tmp;
break;
}
if (!lrp) {
spin_unlock(&lli->lli_lock);
lrp = alloc_ll_remote_perm();
if (!lrp) {
CERROR("alloc memory for ll_remote_perm failed!\n");
return -ENOMEM;
}
spin_lock(&lli->lli_lock);
goto again;
}
lrp->lrp_access_perm = perm->rp_access_perm;
if (lrp != tmp) {
lrp->lrp_uid = perm->rp_uid;
lrp->lrp_gid = perm->rp_gid;
lrp->lrp_fsuid = perm->rp_fsuid;
lrp->lrp_fsgid = perm->rp_fsgid;
hlist_add_head(&lrp->lrp_list, head);
}
lli->lli_rmtperm_time = cfs_time_current();
spin_unlock(&lli->lli_lock);
CDEBUG(D_SEC, "new remote perm@%p: %u/%u/%u/%u - %#x\n",
lrp, lrp->lrp_uid, lrp->lrp_gid, lrp->lrp_fsuid, lrp->lrp_fsgid,
lrp->lrp_access_perm);
return 0;
}
/*
* statfs
*/
static inline int osp_statfs_need_update(struct osp_device *d)
{
return !cfs_time_before(cfs_time_current(),
d->opd_statfs_fresh_till);
}
static inline int capa_is_to_expire(struct obd_capa *ocapa)
{
return cfs_time_beforeq(capa_renewal_time(ocapa), cfs_time_current());
}
void osc_object_set_contended(struct osc_object *obj)
{
obj->oo_contention_time = cfs_time_current();
/* mb(); */
obj->oo_contended = 1;
}
int
LNetEQPoll (lnet_handle_eq_t *eventqs, int neq, int timeout_ms,
lnet_event_t *event, int *which)
{
int i;
int rc;
#ifdef __KERNEL__
cfs_waitlink_t wl;
cfs_time_t now;
#else
struct timeval then;
struct timeval now;
# ifdef HAVE_LIBPTHREAD
struct timespec ts;
# endif
lnet_ni_t *eqwaitni = the_lnet.ln_eqwaitni;
#endif
ENTRY;
LASSERT (the_lnet.ln_init);
LASSERT (the_lnet.ln_refcount > 0);
if (neq < 1)
RETURN(-ENOENT);
LNET_LOCK();
for (;;) {
#ifndef __KERNEL__
LNET_UNLOCK();
/* Recursion breaker */
if (the_lnet.ln_rc_state == LNET_RC_STATE_RUNNING &&
!LNetHandleIsEqual(eventqs[0], the_lnet.ln_rc_eqh))
lnet_router_checker();
LNET_LOCK();
#endif
for (i = 0; i < neq; i++) {
lnet_eq_t *eq = lnet_handle2eq(&eventqs[i]);
if (eq == NULL) {
LNET_UNLOCK();
RETURN(-ENOENT);
}
rc = lib_get_event (eq, event);
if (rc != 0) {
LNET_UNLOCK();
*which = i;
RETURN(rc);
}
}
#ifdef __KERNEL__
if (timeout_ms == 0) {
LNET_UNLOCK();
RETURN (0);
}
cfs_waitlink_init(&wl);
set_current_state(TASK_INTERRUPTIBLE);
cfs_waitq_add(&the_lnet.ln_waitq, &wl);
LNET_UNLOCK();
if (timeout_ms < 0) {
cfs_waitq_wait (&wl, CFS_TASK_INTERRUPTIBLE);
} else {
struct timeval tv;
now = cfs_time_current();
cfs_waitq_timedwait(&wl, CFS_TASK_INTERRUPTIBLE,
cfs_time_seconds(timeout_ms)/1000);
cfs_duration_usec(cfs_time_sub(cfs_time_current(), now),
&tv);
timeout_ms -= tv.tv_sec * 1000 + tv.tv_usec / 1000;
if (timeout_ms < 0)
timeout_ms = 0;
}
LNET_LOCK();
cfs_waitq_del(&the_lnet.ln_waitq, &wl);
#else
if (eqwaitni != NULL) {
/* I have a single NI that I have to call into, to get
* events queued, or to block. */
lnet_ni_addref_locked(eqwaitni);
LNET_UNLOCK();
if (timeout_ms <= 0) {
(eqwaitni->ni_lnd->lnd_wait)(eqwaitni, timeout_ms);
} else {
gettimeofday(&then, NULL);
(eqwaitni->ni_lnd->lnd_wait)(eqwaitni, timeout_ms);
gettimeofday(&now, NULL);
timeout_ms -= (now.tv_sec - then.tv_sec) * 1000 +
(now.tv_usec - then.tv_usec) / 1000;
if (timeout_ms < 0)
timeout_ms = 0;
}
LNET_LOCK();
lnet_ni_decref_locked(eqwaitni);
/* don't call into eqwaitni again if timeout has
* expired */
if (timeout_ms == 0)
eqwaitni = NULL;
continue; /* go back and check for events */
}
if (timeout_ms == 0) {
LNET_UNLOCK();
RETURN (0);
}
# ifndef HAVE_LIBPTHREAD
/* If I'm single-threaded, LNET fails at startup if it can't
* set the_lnet.ln_eqwaitni correctly. */
LBUG();
# else
if (timeout_ms < 0) {
pthread_cond_wait(&the_lnet.ln_cond,
&the_lnet.ln_lock);
} else {
gettimeofday(&then, NULL);
ts.tv_sec = then.tv_sec + timeout_ms/1000;
ts.tv_nsec = then.tv_usec * 1000 +
(timeout_ms%1000) * 1000000;
if (ts.tv_nsec >= 1000000000) {
ts.tv_sec++;
ts.tv_nsec -= 1000000000;
}
pthread_cond_timedwait(&the_lnet.ln_cond,
&the_lnet.ln_lock, &ts);
gettimeofday(&now, NULL);
timeout_ms -= (now.tv_sec - then.tv_sec) * 1000 +
(now.tv_usec - then.tv_usec) / 1000;
if (timeout_ms < 0)
timeout_ms = 0;
}
# endif
#endif
}
}
int quota_adjust_slave_lqs(struct quota_adjust_qunit *oqaq,
struct lustre_quota_ctxt *qctxt)
{
struct lustre_qunit_size *lqs = NULL;
unsigned long *unit, *tune;
signed long tmp = 0;
cfs_time_t time_limit = 0, *shrink;
int i, rc = 0;
ENTRY;
LASSERT(qctxt);
lqs = quota_search_lqs(LQS_KEY(QAQ_IS_GRP(oqaq), oqaq->qaq_id),
qctxt, QAQ_IS_CREATE_LQS(oqaq) ? 1 : 0);
if (lqs == NULL || IS_ERR(lqs)){
CERROR("fail to find a lqs for %sid %u!\n",
QAQ_IS_GRP(oqaq) ? "g" : "u", oqaq->qaq_id);
RETURN(PTR_ERR(lqs));
}
CDEBUG(D_QUOTA, "before: bunit: %lu, iunit: %lu.\n",
lqs->lqs_bunit_sz, lqs->lqs_iunit_sz);
cfs_spin_lock(&lqs->lqs_lock);
for (i = 0; i < 2; i++) {
if (i == 0 && !QAQ_IS_ADJBLK(oqaq))
continue;
if (i == 1 && !QAQ_IS_ADJINO(oqaq))
continue;
tmp = i ? (lqs->lqs_iunit_sz - oqaq->qaq_iunit_sz) :
(lqs->lqs_bunit_sz - oqaq->qaq_bunit_sz);
shrink = i ? &lqs->lqs_last_ishrink :
&lqs->lqs_last_bshrink;
time_limit = cfs_time_add(i ? lqs->lqs_last_ishrink :
lqs->lqs_last_bshrink,
cfs_time_seconds(qctxt->lqc_switch_seconds));
unit = i ? &lqs->lqs_iunit_sz : &lqs->lqs_bunit_sz;
tune = i ? &lqs->lqs_itune_sz : &lqs->lqs_btune_sz;
/* quota master shrinks */
if (qctxt->lqc_handler && tmp > 0)
*shrink = cfs_time_current();
/* quota master enlarges */
if (qctxt->lqc_handler && tmp < 0) {
/* in case of ping-pong effect, don't enlarge lqs
* in a short time */
if (*shrink &&
cfs_time_before(cfs_time_current(), time_limit))
tmp = 0;
}
/* when setquota, don't enlarge lqs b=18616 */
if (QAQ_IS_CREATE_LQS(oqaq) && tmp < 0)
tmp = 0;
if (tmp != 0) {
*unit = i ? oqaq->qaq_iunit_sz : oqaq->qaq_bunit_sz;
*tune = (*unit) / 2;
}
if (tmp > 0)
rc |= i ? LQS_INO_DECREASE : LQS_BLK_DECREASE;
if (tmp < 0)
rc |= i ? LQS_INO_INCREASE : LQS_BLK_INCREASE;
}
cfs_spin_unlock(&lqs->lqs_lock);
CDEBUG(D_QUOTA, "after: bunit: %lu, iunit: %lu.\n",
lqs->lqs_bunit_sz, lqs->lqs_iunit_sz);
lqs_putref(lqs);
RETURN(rc);
}
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);
}
/*
* called to reserve object in the pool
* return codes:
* ENOSPC - no space on corresponded OST
* EAGAIN - precreation is in progress, try later
* EIO - no access to OST
*/
int osp_precreate_reserve(const struct lu_env *env, struct osp_device *d)
{
struct l_wait_info lwi;
cfs_time_t expire = cfs_time_shift(obd_timeout);
int precreated, rc;
ENTRY;
LASSERTF(osp_objs_precreated(env, d) >= 0, "Last created FID "DFID
"Next FID "DFID"\n", PFID(&d->opd_pre_last_created_fid),
PFID(&d->opd_pre_used_fid));
/*
* wait till:
* - preallocation is done
* - no free space expected soon
* - can't connect to OST for too long (obd_timeout)
* - OST can allocate fid sequence.
*/
while ((rc = d->opd_pre_status) == 0 || rc == -ENOSPC ||
rc == -ENODEV || rc == -EAGAIN || rc == -ENOTCONN) {
/*
* increase number of precreations
*/
precreated = osp_objs_precreated(env, d);
if (d->opd_pre_grow_count < d->opd_pre_max_grow_count &&
d->opd_pre_grow_slow == 0 &&
precreated <= (d->opd_pre_grow_count / 4 + 1)) {
spin_lock(&d->opd_pre_lock);
d->opd_pre_grow_slow = 1;
d->opd_pre_grow_count *= 2;
spin_unlock(&d->opd_pre_lock);
}
spin_lock(&d->opd_pre_lock);
precreated = osp_objs_precreated(env, d);
if (precreated > d->opd_pre_reserved &&
!d->opd_pre_recovering) {
d->opd_pre_reserved++;
spin_unlock(&d->opd_pre_lock);
rc = 0;
/* XXX: don't wake up if precreation is in progress */
if (osp_precreate_near_empty_nolock(env, d) &&
!osp_precreate_end_seq_nolock(env, d))
wake_up(&d->opd_pre_waitq);
break;
}
spin_unlock(&d->opd_pre_lock);
/*
* all precreated objects have been used and no-space
* status leave us no chance to succeed very soon
* but if there is destroy in progress, then we should
* wait till that is done - some space might be released
*/
if (unlikely(rc == -ENOSPC)) {
if (d->opd_syn_changes) {
/* force local commit to release space */
dt_commit_async(env, d->opd_storage);
}
if (d->opd_syn_rpc_in_progress) {
/* just wait till destroys are done */
/* see l_wait_even() few lines below */
}
if (d->opd_syn_changes +
d->opd_syn_rpc_in_progress == 0) {
/* no hope for free space */
break;
}
}
/* XXX: don't wake up if precreation is in progress */
wake_up(&d->opd_pre_waitq);
lwi = LWI_TIMEOUT(expire - cfs_time_current(),
osp_precreate_timeout_condition, d);
if (cfs_time_aftereq(cfs_time_current(), expire)) {
rc = -ETIMEDOUT;
break;
}
l_wait_event(d->opd_pre_user_waitq,
osp_precreate_ready_condition(env, d), &lwi);
}
RETURN(rc);
}
