* Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2012, 2013, Intel Corporation.
+ * Copyright (c) 2012, 2014, Intel Corporation.
*
*/
/*
static void osc_update_pending(struct osc_object *obj, int cmd, int delta);
static int osc_extent_wait(const struct lu_env *env, struct osc_extent *ext,
- int state);
+ enum osc_extent_state state);
static void osc_ap_completion(const struct lu_env *env, struct client_obd *cli,
struct osc_async_page *oap, int sent, int rc);
static int osc_make_ready(const struct lu_env *env, struct osc_async_page *oap,
*buf++ = ext->oe_rw ? 'r' : 'w';
if (ext->oe_intree)
*buf++ = 'i';
+ if (ext->oe_sync)
+ *buf++ = 'S';
if (ext->oe_srvlock)
*buf++ = 's';
if (ext->oe_hp)
static const char *oes_strings[] = {
"inv", "active", "cache", "locking", "lockdone", "rpc", "trunc", NULL };
-#define OSC_EXTENT_DUMP(lvl, extent, fmt, ...) do { \
- struct osc_extent *__ext = (extent); \
- char __buf[16]; \
- \
- CDEBUG(lvl, \
- "extent %p@{" EXTSTR ", " \
- "[%d|%d|%c|%s|%s|%p], [%d|%d|%c|%c|%p|%u|%p]} " fmt, \
- /* ----- extent part 0 ----- */ \
- __ext, EXTPARA(__ext), \
- /* ----- part 1 ----- */ \
- atomic_read(&__ext->oe_refc), \
- atomic_read(&__ext->oe_users), \
- list_empty_marker(&__ext->oe_link), \
- oes_strings[__ext->oe_state], ext_flags(__ext, __buf), \
- __ext->oe_obj, \
- /* ----- part 2 ----- */ \
- __ext->oe_grants, __ext->oe_nr_pages, \
- list_empty_marker(&__ext->oe_pages), \
- waitqueue_active(&__ext->oe_waitq) ? '+' : '-', \
- __ext->oe_osclock, __ext->oe_mppr, __ext->oe_owner, \
- /* ----- part 4 ----- */ \
- ## __VA_ARGS__); \
+#define OSC_EXTENT_DUMP(lvl, extent, fmt, ...) do { \
+ struct osc_extent *__ext = (extent); \
+ char __buf[16]; \
+ \
+ CDEBUG(lvl, \
+ "extent %p@{" EXTSTR ", " \
+ "[%d|%d|%c|%s|%s|%p], [%d|%d|%c|%c|%p|%u|%p]} " fmt, \
+ /* ----- extent part 0 ----- */ \
+ __ext, EXTPARA(__ext), \
+ /* ----- part 1 ----- */ \
+ atomic_read(&__ext->oe_refc), \
+ atomic_read(&__ext->oe_users), \
+ list_empty_marker(&__ext->oe_link), \
+ oes_strings[__ext->oe_state], ext_flags(__ext, __buf), \
+ __ext->oe_obj, \
+ /* ----- part 2 ----- */ \
+ __ext->oe_grants, __ext->oe_nr_pages, \
+ list_empty_marker(&__ext->oe_pages), \
+ waitqueue_active(&__ext->oe_waitq) ? '+' : '-', \
+ __ext->oe_dlmlock, __ext->oe_mppr, __ext->oe_owner, \
+ /* ----- part 4 ----- */ \
+ ## __VA_ARGS__); \
+ if (lvl == D_ERROR && __ext->oe_dlmlock != NULL) \
+ LDLM_ERROR(__ext->oe_dlmlock, "extent: %p\n", __ext); \
+ else \
+ LDLM_DEBUG(__ext->oe_dlmlock, "extent: %p\n", __ext); \
} while (0)
#undef EASSERTF
{
struct osc_object *obj = ext->oe_obj;
struct osc_async_page *oap;
- int page_count;
+ size_t page_count;
int rc = 0;
if (!osc_object_is_locked(obj))
if (ext->oe_max_end < ext->oe_end || ext->oe_end < ext->oe_start)
GOTO(out, rc = 80);
- if (ext->oe_osclock == NULL && ext->oe_grants > 0)
+ if (ext->oe_sync && ext->oe_grants > 0)
GOTO(out, rc = 90);
- if (ext->oe_osclock) {
- struct cl_lock_descr *descr;
- descr = &ext->oe_osclock->cll_descr;
- if (!(descr->cld_start <= ext->oe_start &&
- descr->cld_end >= ext->oe_max_end))
+ if (ext->oe_dlmlock != NULL) {
+ struct ldlm_extent *extent;
+
+ extent = &ext->oe_dlmlock->l_policy_data.l_extent;
+ if (!(extent->start <= cl_offset(osc2cl(obj), ext->oe_start) &&
+ extent->end >= cl_offset(osc2cl(obj), ext->oe_max_end)))
GOTO(out, rc = 100);
+
+ if (!(ext->oe_dlmlock->l_granted_mode & (LCK_PW | LCK_GROUP)))
+ GOTO(out, rc = 102);
}
if (ext->oe_nr_pages > ext->oe_mppr)
ext->oe_state = OES_INV;
INIT_LIST_HEAD(&ext->oe_pages);
init_waitqueue_head(&ext->oe_waitq);
- ext->oe_osclock = NULL;
+ ext->oe_dlmlock = NULL;
return ext;
}
LASSERT(ext->oe_state == OES_INV);
LASSERT(!ext->oe_intree);
- if (ext->oe_osclock) {
- cl_lock_put(env, ext->oe_osclock);
- ext->oe_osclock = NULL;
+ if (ext->oe_dlmlock != NULL) {
+ lu_ref_add(&ext->oe_dlmlock->l_reference,
+ "osc_extent", ext);
+ LDLM_LOCK_PUT(ext->oe_dlmlock);
+ ext->oe_dlmlock = NULL;
}
osc_extent_free(ext);
}
else if (ext->oe_start > tmp->oe_end)
n = &(*n)->rb_right;
else
- EASSERTF(0, tmp, EXTSTR, EXTPARA(ext));
+ EASSERTF(0, tmp, EXTSTR"\n", EXTPARA(ext));
}
rb_link_node(&ext->oe_node, parent, n);
rb_insert_color(&ext->oe_node, &obj->oo_root);
if (cur->oe_max_end != victim->oe_max_end)
return -ERANGE;
- LASSERT(cur->oe_osclock == victim->oe_osclock);
+ LASSERT(cur->oe_dlmlock == victim->oe_dlmlock);
ppc_bits = osc_cli(obj)->cl_chunkbits - PAGE_CACHE_SHIFT;
chunk_start = cur->oe_start >> ppc_bits;
chunk_end = cur->oe_end >> ppc_bits;
* Find or create an extent which includes @index, core function to manage
* extent tree.
*/
-struct osc_extent *osc_extent_find(const struct lu_env *env,
- struct osc_object *obj, pgoff_t index,
- int *grants)
-
+static struct osc_extent *osc_extent_find(const struct lu_env *env,
+ struct osc_object *obj, pgoff_t index,
+ unsigned int *grants)
{
struct client_obd *cli = osc_cli(obj);
- struct cl_lock *lock;
+ struct osc_lock *olck;
+ struct cl_lock_descr *descr;
struct osc_extent *cur;
struct osc_extent *ext;
struct osc_extent *conflict = NULL;
struct osc_extent *found = NULL;
pgoff_t chunk;
pgoff_t max_end;
- int max_pages; /* max_pages_per_rpc */
- int chunksize;
+ unsigned int max_pages; /* max_pages_per_rpc */
+ unsigned int chunksize;
int ppc_bits; /* pages per chunk bits */
- int chunk_mask;
+ pgoff_t chunk_mask;
int rc;
ENTRY;
if (cur == NULL)
RETURN(ERR_PTR(-ENOMEM));
- lock = cl_lock_at_pgoff(env, osc2cl(obj), index, NULL, 1, 0);
- LASSERT(lock != NULL);
- LASSERT(lock->cll_descr.cld_mode >= CLM_WRITE);
+ olck = osc_env_io(env)->oi_write_osclock;
+ LASSERTF(olck != NULL, "page %lu is not covered by lock\n", index);
+ LASSERT(olck->ols_state == OLS_GRANTED);
+
+ descr = &olck->ols_cl.cls_lock->cll_descr;
+ LASSERT(descr->cld_mode >= CLM_WRITE);
LASSERT(cli->cl_chunkbits >= PAGE_CACHE_SHIFT);
ppc_bits = cli->cl_chunkbits - PAGE_CACHE_SHIFT;
max_pages = cli->cl_max_pages_per_rpc;
LASSERT((max_pages & ~chunk_mask) == 0);
max_end = index - (index % max_pages) + max_pages - 1;
- max_end = min_t(pgoff_t, max_end, lock->cll_descr.cld_end);
+ max_end = min_t(pgoff_t, max_end, descr->cld_end);
/* initialize new extent by parameters so far */
cur->oe_max_end = max_end;
cur->oe_start = index & chunk_mask;
cur->oe_end = ((index + ~chunk_mask + 1) & chunk_mask) - 1;
- if (cur->oe_start < lock->cll_descr.cld_start)
- cur->oe_start = lock->cll_descr.cld_start;
+ if (cur->oe_start < descr->cld_start)
+ cur->oe_start = descr->cld_start;
if (cur->oe_end > max_end)
cur->oe_end = max_end;
- cur->oe_osclock = lock;
cur->oe_grants = 0;
cur->oe_mppr = max_pages;
+ if (olck->ols_dlmlock != NULL) {
+ LASSERT(olck->ols_hold);
+ cur->oe_dlmlock = LDLM_LOCK_GET(olck->ols_dlmlock);
+ lu_ref_add(&olck->ols_dlmlock->l_reference, "osc_extent", cur);
+ }
/* grants has been allocated by caller */
LASSERTF(*grants >= chunksize + cli->cl_extent_tax,
"%u/%u/%u.\n", *grants, chunksize, cli->cl_extent_tax);
- LASSERTF((max_end - cur->oe_start) < max_pages, EXTSTR, EXTPARA(cur));
+ LASSERTF((max_end - cur->oe_start) < max_pages, EXTSTR"\n",
+ EXTPARA(cur));
restart:
osc_object_lock(obj);
if (ext == NULL)
ext = first_extent(obj);
while (ext != NULL) {
- loff_t ext_chk_start = ext->oe_start >> ppc_bits;
- loff_t ext_chk_end = ext->oe_end >> ppc_bits;
+ pgoff_t ext_chk_start = ext->oe_start >> ppc_bits;
+ pgoff_t ext_chk_end = ext->oe_end >> ppc_bits;
LASSERT(sanity_check_nolock(ext) == 0);
if (chunk > ext_chk_end + 1)
break;
/* if covering by different locks, no chance to match */
- if (lock != ext->oe_osclock) {
+ if (olck->ols_dlmlock != ext->oe_dlmlock) {
EASSERTF(!overlapped(ext, cur), ext,
- EXTSTR, EXTPARA(cur));
+ EXTSTR"\n", EXTPARA(cur));
ext = next_extent(ext);
continue;
* full contain. */
EASSERTF((ext->oe_start <= cur->oe_start &&
ext->oe_end >= cur->oe_end),
- ext, EXTSTR, EXTPARA(cur));
+ ext, EXTSTR"\n", EXTPARA(cur));
if (ext->oe_state > OES_CACHE || ext->oe_fsync_wait) {
/* for simplicity, we wait for this extent to
if (found != NULL) {
LASSERT(conflict == NULL);
if (!IS_ERR(found)) {
- LASSERT(found->oe_osclock == cur->oe_osclock);
+ LASSERT(found->oe_dlmlock == cur->oe_dlmlock);
OSC_EXTENT_DUMP(D_CACHE, found,
"found caching ext for %lu.\n", index);
}
found = osc_extent_hold(cur);
osc_extent_insert(obj, cur);
OSC_EXTENT_DUMP(D_CACHE, cur, "add into tree %lu/%lu.\n",
- index, lock->cll_descr.cld_end);
+ index, descr->cld_end);
}
osc_object_unlock(obj);
RETURN(0);
}
-static int extent_wait_cb(struct osc_extent *ext, int state)
+static int extent_wait_cb(struct osc_extent *ext, enum osc_extent_state state)
{
int ret;
* Wait for the extent's state to become @state.
*/
static int osc_extent_wait(const struct lu_env *env, struct osc_extent *ext,
- int state)
+ enum osc_extent_state state)
{
struct osc_object *obj = ext->oe_obj;
struct l_wait_info lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(600), NULL,
rc = l_wait_event(ext->oe_waitq, extent_wait_cb(ext, state), &lwi);
if (rc == -ETIMEDOUT) {
OSC_EXTENT_DUMP(D_ERROR, ext,
- "%s: wait ext to %d timedout, recovery in progress?\n",
+ "%s: wait ext to %u timedout, recovery in progress?\n",
osc_export(obj)->exp_obd->obd_name, state);
lwi = LWI_INTR(NULL, NULL);
struct osc_async_page *oap;
struct osc_async_page *last = NULL;
struct osc_object *obj = ext->oe_obj;
- int page_count = 0;
+ unsigned int page_count = 0;
int rc;
ENTRY;
/* the last page is the only one we need to refresh its count by
* the size of file. */
if (!(last->oap_async_flags & ASYNC_COUNT_STABLE)) {
- last->oap_count = osc_refresh_count(env, last, OBD_BRW_WRITE);
- LASSERT(last->oap_count > 0);
- LASSERT(last->oap_page_off + last->oap_count <= PAGE_CACHE_SIZE);
+ int last_oap_count = osc_refresh_count(env, last, OBD_BRW_WRITE);
+ LASSERT(last_oap_count > 0);
+ LASSERT(last->oap_page_off + last_oap_count <= PAGE_CACHE_SIZE);
+ last->oap_count = last_oap_count;
spin_lock(&last->oap_lock);
last->oap_async_flags |= ASYNC_COUNT_STABLE;
spin_unlock(&last->oap_lock);
* called to expand the extent for the same IO. To expand the extent, the
* page index must be in the same or next chunk of ext->oe_end.
*/
-static int osc_extent_expand(struct osc_extent *ext, pgoff_t index, int *grants)
+static int osc_extent_expand(struct osc_extent *ext, pgoff_t index,
+ unsigned int *grants)
{
struct osc_object *obj = ext->oe_obj;
struct client_obd *cli = osc_cli(obj);
pgoff_t chunk = index >> ppc_bits;
pgoff_t end_chunk;
pgoff_t end_index;
- int chunksize = 1 << cli->cl_chunkbits;
+ unsigned int chunksize = 1 << cli->cl_chunkbits;
int rc = 0;
ENTRY;
if (rc == 0 && srvlock) {
struct lu_device *ld = opg->ops_cl.cpl_obj->co_lu.lo_dev;
struct osc_stats *stats = &lu2osc_dev(ld)->od_stats;
- int bytes = oap->oap_count;
+ size_t bytes = oap->oap_count;
if (crt == CRT_READ)
stats->os_lockless_reads += bytes;
#define OSC_DUMP_GRANT(lvl, cli, fmt, args...) do { \
struct client_obd *__tmp = (cli); \
- CDEBUG(lvl, "%s: grant { dirty: %ld/%ld dirty_pages: %d/%d " \
- "unstable_pages: %d/%d dropped: %ld avail: %ld, " \
- "reserved: %ld, flight: %d } lru {in list: %d, " \
- "left: %d, waiters: %d }" fmt, \
+ CDEBUG(lvl, "%s: grant { dirty: %ld/%ld dirty_pages: %ld/%lu " \
+ "dropped: %ld avail: %ld, reserved: %ld, flight: %d }" \
+ "lru {in list: %ld, left: %ld, waiters: %d }"fmt"\n", \
__tmp->cl_import->imp_obd->obd_name, \
- __tmp->cl_dirty, __tmp->cl_dirty_max, \
- atomic_read(&obd_dirty_pages), obd_max_dirty_pages, \
- atomic_read(&obd_unstable_pages), obd_max_dirty_pages, \
+ __tmp->cl_dirty_pages, __tmp->cl_dirty_max_pages, \
+ atomic_long_read(&obd_dirty_pages), obd_max_dirty_pages, \
__tmp->cl_lost_grant, __tmp->cl_avail_grant, \
__tmp->cl_reserved_grant, __tmp->cl_w_in_flight, \
- atomic_read(&__tmp->cl_lru_in_list), \
- atomic_read(&__tmp->cl_lru_busy), \
+ atomic_long_read(&__tmp->cl_lru_in_list), \
+ atomic_long_read(&__tmp->cl_lru_busy), \
atomic_read(&__tmp->cl_lru_shrinkers), ##args); \
} while (0)
static void osc_consume_write_grant(struct client_obd *cli,
struct brw_page *pga)
{
- assert_spin_locked(&cli->cl_loi_list_lock.lock);
+ assert_spin_locked(&cli->cl_loi_list_lock);
LASSERT(!(pga->flag & OBD_BRW_FROM_GRANT));
- atomic_inc(&obd_dirty_pages);
- cli->cl_dirty += PAGE_CACHE_SIZE;
+ atomic_long_inc(&obd_dirty_pages);
+ cli->cl_dirty_pages++;
pga->flag |= OBD_BRW_FROM_GRANT;
CDEBUG(D_CACHE, "using %lu grant credits for brw %p page %p\n",
PAGE_CACHE_SIZE, pga, pga->pg);
{
ENTRY;
- assert_spin_locked(&cli->cl_loi_list_lock.lock);
+ assert_spin_locked(&cli->cl_loi_list_lock);
if (!(pga->flag & OBD_BRW_FROM_GRANT)) {
EXIT;
return;
}
pga->flag &= ~OBD_BRW_FROM_GRANT;
- atomic_dec(&obd_dirty_pages);
- cli->cl_dirty -= PAGE_CACHE_SIZE;
+ atomic_long_dec(&obd_dirty_pages);
+ cli->cl_dirty_pages--;
if (pga->flag & OBD_BRW_NOCACHE) {
pga->flag &= ~OBD_BRW_NOCACHE;
- atomic_dec(&obd_dirty_transit_pages);
- cli->cl_dirty_transit -= PAGE_CACHE_SIZE;
+ atomic_long_dec(&obd_dirty_transit_pages);
+ cli->cl_dirty_transit--;
}
EXIT;
}
}
}
-void osc_unreserve_grant(struct client_obd *cli,
- unsigned int reserved, unsigned int unused)
+static void osc_unreserve_grant(struct client_obd *cli,
+ unsigned int reserved, unsigned int unused)
{
- client_obd_list_lock(&cli->cl_loi_list_lock);
+ spin_lock(&cli->cl_loi_list_lock);
__osc_unreserve_grant(cli, reserved, unused);
if (unused > 0)
osc_wake_cache_waiters(cli);
- client_obd_list_unlock(&cli->cl_loi_list_lock);
+ spin_unlock(&cli->cl_loi_list_lock);
}
/**
static void osc_free_grant(struct client_obd *cli, unsigned int nr_pages,
unsigned int lost_grant)
{
- int grant = (1 << cli->cl_chunkbits) + cli->cl_extent_tax;
+ unsigned long grant = (1 << cli->cl_chunkbits) + cli->cl_extent_tax;
- client_obd_list_lock(&cli->cl_loi_list_lock);
- atomic_sub(nr_pages, &obd_dirty_pages);
- cli->cl_dirty -= nr_pages << PAGE_CACHE_SHIFT;
+ spin_lock(&cli->cl_loi_list_lock);
+ atomic_long_sub(nr_pages, &obd_dirty_pages);
+ cli->cl_dirty_pages -= nr_pages;
cli->cl_lost_grant += lost_grant;
if (cli->cl_avail_grant < grant && cli->cl_lost_grant >= grant) {
/* borrow some grant from truncate to avoid the case that
cli->cl_avail_grant += grant;
}
osc_wake_cache_waiters(cli);
- client_obd_list_unlock(&cli->cl_loi_list_lock);
+ spin_unlock(&cli->cl_loi_list_lock);
CDEBUG(D_CACHE, "lost %u grant: %lu avail: %lu dirty: %lu\n",
lost_grant, cli->cl_lost_grant,
- cli->cl_avail_grant, cli->cl_dirty);
+ cli->cl_avail_grant, cli->cl_dirty_pages << PAGE_CACHE_SHIFT);
}
/**
*/
static void osc_exit_cache(struct client_obd *cli, struct osc_async_page *oap)
{
- client_obd_list_lock(&cli->cl_loi_list_lock);
+ spin_lock(&cli->cl_loi_list_lock);
osc_release_write_grant(cli, &oap->oap_brw_page);
- client_obd_list_unlock(&cli->cl_loi_list_lock);
+ spin_unlock(&cli->cl_loi_list_lock);
}
/**
{
int rc;
- OSC_DUMP_GRANT(D_CACHE, cli, "need:%d.\n", bytes);
+ OSC_DUMP_GRANT(D_CACHE, cli, "need:%d\n", bytes);
rc = osc_reserve_grant(cli, bytes);
if (rc < 0)
return 0;
- if (cli->cl_dirty + PAGE_CACHE_SIZE <= cli->cl_dirty_max &&
- atomic_read(&obd_unstable_pages) + 1 +
- atomic_read(&obd_dirty_pages) <= obd_max_dirty_pages) {
+ if (cli->cl_dirty_pages < cli->cl_dirty_max_pages &&
+ 1 + atomic_long_read(&obd_dirty_pages) <= obd_max_dirty_pages) {
osc_consume_write_grant(cli, &oap->oap_brw_page);
if (transient) {
- cli->cl_dirty_transit += PAGE_CACHE_SIZE;
- atomic_inc(&obd_dirty_transit_pages);
+ cli->cl_dirty_transit++;
+ atomic_long_inc(&obd_dirty_transit_pages);
oap->oap_brw_flags |= OBD_BRW_NOCACHE;
}
rc = 1;
static int ocw_granted(struct client_obd *cli, struct osc_cache_waiter *ocw)
{
int rc;
- client_obd_list_lock(&cli->cl_loi_list_lock);
+ spin_lock(&cli->cl_loi_list_lock);
rc = list_empty(&ocw->ocw_entry);
- client_obd_list_unlock(&cli->cl_loi_list_lock);
+ spin_unlock(&cli->cl_loi_list_lock);
return rc;
}
static int osc_enter_cache(const struct lu_env *env, struct client_obd *cli,
struct osc_async_page *oap, int bytes)
{
- struct osc_object *osc = oap->oap_obj;
- struct lov_oinfo *loi = osc->oo_oinfo;
- struct osc_cache_waiter ocw;
- struct l_wait_info lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(600), NULL,
- LWI_ON_SIGNAL_NOOP, NULL);
- int rc = -EDQUOT;
+ struct osc_object *osc = oap->oap_obj;
+ struct lov_oinfo *loi = osc->oo_oinfo;
+ struct osc_cache_waiter ocw;
+ struct l_wait_info lwi;
+ int rc = -EDQUOT;
ENTRY;
- OSC_DUMP_GRANT(D_CACHE, cli, "need:%d.\n", bytes);
+ lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(AT_OFF ? obd_timeout : at_max),
+ NULL, LWI_ON_SIGNAL_NOOP, NULL);
- client_obd_list_lock(&cli->cl_loi_list_lock);
+ OSC_DUMP_GRANT(D_CACHE, cli, "need:%d\n", bytes);
+
+ spin_lock(&cli->cl_loi_list_lock);
/* force the caller to try sync io. this can jump the list
* of queued writes and create a discontiguous rpc stream */
if (OBD_FAIL_CHECK(OBD_FAIL_OSC_NO_GRANT) ||
- cli->cl_dirty_max < PAGE_CACHE_SIZE ||
- cli->cl_ar.ar_force_sync || loi->loi_ar.ar_force_sync)
+ cli->cl_dirty_max_pages == 0 ||
+ cli->cl_ar.ar_force_sync || loi->loi_ar.ar_force_sync) {
+ OSC_DUMP_GRANT(D_CACHE, cli, "forced sync i/o\n");
GOTO(out, rc = -EDQUOT);
+ }
/* Hopefully normal case - cache space and write credits available */
- if (osc_enter_cache_try(cli, oap, bytes, 0))
+ if (osc_enter_cache_try(cli, oap, bytes, 0)) {
+ OSC_DUMP_GRANT(D_CACHE, cli, "granted from cache\n");
GOTO(out, rc = 0);
+ }
/* We can get here for two reasons: too many dirty pages in cache, or
* run out of grants. In both cases we should write dirty pages out.
init_waitqueue_head(&ocw.ocw_waitq);
ocw.ocw_oap = oap;
ocw.ocw_grant = bytes;
- while (cli->cl_dirty > 0 || cli->cl_w_in_flight > 0) {
+ while (cli->cl_dirty_pages > 0 || cli->cl_w_in_flight > 0) {
list_add_tail(&ocw.ocw_entry, &cli->cl_cache_waiters);
ocw.ocw_rc = 0;
- client_obd_list_unlock(&cli->cl_loi_list_lock);
+ spin_unlock(&cli->cl_loi_list_lock);
osc_io_unplug_async(env, cli, NULL);
rc = l_wait_event(ocw.ocw_waitq, ocw_granted(cli, &ocw), &lwi);
- client_obd_list_lock(&cli->cl_loi_list_lock);
+ spin_lock(&cli->cl_loi_list_lock);
- /* l_wait_event is interrupted by signal, or timed out */
if (rc < 0) {
- switch (rc) {
- case -ETIMEDOUT:
- OSC_DUMP_GRANT(D_ERROR, cli,
- "try to reserve %d.\n", bytes);
- osc_extent_tree_dump(D_ERROR, osc);
- rc = -EDQUOT;
- break;
- case -EINTR:
- /* Ensures restartability - LU-3581 */
- rc = -ERESTARTSYS;
- break;
- default:
- CDEBUG(D_CACHE, "%s: event for cache space @"
- " %p never arrived due to %d\n",
- cli->cl_import->imp_obd->obd_name,
- &ocw, rc);
- break;
- }
+ /* l_wait_event is interrupted by signal or timed out */
list_del_init(&ocw.ocw_entry);
- GOTO(out, rc);
+ break;
}
-
LASSERT(list_empty(&ocw.ocw_entry));
rc = ocw.ocw_rc;
if (rc != -EDQUOT)
- GOTO(out, rc);
- if (osc_enter_cache_try(cli, oap, bytes, 0))
- GOTO(out, rc = 0);
+ break;
+ if (osc_enter_cache_try(cli, oap, bytes, 0)) {
+ rc = 0;
+ break;
+ }
+ }
+
+ switch (rc) {
+ case 0:
+ OSC_DUMP_GRANT(D_CACHE, cli, "finally got grant space\n");
+ break;
+ case -ETIMEDOUT:
+ OSC_DUMP_GRANT(D_CACHE, cli,
+ "timeout, fall back to sync i/o\n");
+ osc_extent_tree_dump(D_CACHE, osc);
+ /* fall back to synchronous I/O */
+ rc = -EDQUOT;
+ break;
+ case -EINTR:
+ /* Ensures restartability - LU-3581 */
+ OSC_DUMP_GRANT(D_CACHE, cli, "interrupted\n");
+ rc = -ERESTARTSYS;
+ break;
+ case -EDQUOT:
+ OSC_DUMP_GRANT(D_CACHE, cli,
+ "no grant space, fall back to sync i/o\n");
+ break;
+ default:
+ CDEBUG(D_CACHE, "%s: event for cache space @ %p never arrived "
+ "due to %d, fall back to sync i/o\n",
+ cli->cl_import->imp_obd->obd_name, &ocw, rc);
+ break;
}
EXIT;
out:
- client_obd_list_unlock(&cli->cl_loi_list_lock);
- OSC_DUMP_GRANT(D_CACHE, cli, "returned %d.\n", rc);
+ spin_unlock(&cli->cl_loi_list_lock);
RETURN(rc);
}
ocw->ocw_rc = -EDQUOT;
/* we can't dirty more */
- if ((cli->cl_dirty + PAGE_CACHE_SIZE > cli->cl_dirty_max) ||
- (atomic_read(&obd_unstable_pages) + 1 +
- atomic_read(&obd_dirty_pages) > obd_max_dirty_pages)) {
+ if ((cli->cl_dirty_pages >= cli->cl_dirty_max_pages) ||
+ (1 + atomic_long_read(&obd_dirty_pages) >
+ obd_max_dirty_pages)) {
CDEBUG(D_CACHE, "no dirty room: dirty: %ld "
- "osc max %ld, sys max %d\n", cli->cl_dirty,
- cli->cl_dirty_max, obd_max_dirty_pages);
+ "osc max %ld, sys max %ld\n",
+ cli->cl_dirty_pages, cli->cl_dirty_max_pages,
+ obd_max_dirty_pages);
goto wakeup;
}
- ocw->ocw_rc = 0;
- if (!osc_enter_cache_try(cli, ocw->ocw_oap, ocw->ocw_grant, 0))
- ocw->ocw_rc = -EDQUOT;
-
+ if (osc_enter_cache_try(cli, ocw->ocw_oap, ocw->ocw_grant, 0))
+ ocw->ocw_rc = 0;
wakeup:
CDEBUG(D_CACHE, "wake up %p for oap %p, avail grant %ld, %d\n",
ocw, ocw->ocw_oap, cli->cl_avail_grant, ocw->ocw_rc);
{
int is_ready;
- client_obd_list_lock(&cli->cl_loi_list_lock);
+ spin_lock(&cli->cl_loi_list_lock);
is_ready = __osc_list_maint(cli, osc);
- client_obd_list_unlock(&cli->cl_loi_list_lock);
+ spin_unlock(&cli->cl_loi_list_lock);
return is_ready;
}
ar->ar_force_sync = 0;
}
-/* Performs "unstable" page accounting. This function balances the
- * increment operations performed in osc_inc_unstable_pages. It is
- * registered as the RPC request callback, and is executed when the
- * bulk RPC is committed on the server. Thus at this point, the pages
- * involved in the bulk transfer are no longer considered unstable. */
-void osc_dec_unstable_pages(struct ptlrpc_request *req)
-{
- struct ptlrpc_bulk_desc *desc = req->rq_bulk;
- struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
- int page_count = desc->bd_iov_count;
- int i;
-
- /* No unstable page tracking */
- if (cli->cl_cache == NULL)
- return;
-
- LASSERT(page_count >= 0);
-
- for (i = 0; i < page_count; i++)
- dec_zone_page_state(desc->bd_iov[i].kiov_page, NR_UNSTABLE_NFS);
-
- atomic_sub(page_count, &cli->cl_cache->ccc_unstable_nr);
- LASSERT(atomic_read(&cli->cl_cache->ccc_unstable_nr) >= 0);
-
- atomic_sub(page_count, &cli->cl_unstable_count);
- LASSERT(atomic_read(&cli->cl_unstable_count) >= 0);
-
- atomic_sub(page_count, &obd_unstable_pages);
- LASSERT(atomic_read(&obd_unstable_pages) >= 0);
-
- wake_up_all(&cli->cl_cache->ccc_unstable_waitq);
-}
-
-/* "unstable" page accounting. See: osc_dec_unstable_pages. */
-void osc_inc_unstable_pages(struct ptlrpc_request *req)
-{
- struct ptlrpc_bulk_desc *desc = req->rq_bulk;
- struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
- int page_count = desc->bd_iov_count;
- int i;
-
- /* No unstable page tracking */
- if (cli->cl_cache == NULL)
- return;
-
- LASSERT(page_count >= 0);
-
- for (i = 0; i < page_count; i++)
- inc_zone_page_state(desc->bd_iov[i].kiov_page, NR_UNSTABLE_NFS);
-
- LASSERT(atomic_read(&cli->cl_cache->ccc_unstable_nr) >= 0);
- atomic_add(page_count, &cli->cl_cache->ccc_unstable_nr);
-
- LASSERT(atomic_read(&cli->cl_unstable_count) >= 0);
- atomic_add(page_count, &cli->cl_unstable_count);
-
- LASSERT(atomic_read(&obd_unstable_pages) >= 0);
- atomic_add(page_count, &obd_unstable_pages);
-
- /* If the request has already been committed (i.e. brw_commit
- * called via rq_commit_cb), we need to undo the unstable page
- * increments we just performed because rq_commit_cb wont be
- * called again. */
- spin_lock(&req->rq_lock);
- if (unlikely(req->rq_committed)) {
- spin_unlock(&req->rq_lock);
-
- osc_dec_unstable_pages(req);
- } else {
- req->rq_unstable = 1;
- spin_unlock(&req->rq_lock);
- }
-}
-
/* this must be called holding the loi list lock to give coverage to exit_cache,
* async_flag maintenance, and oap_request */
static void osc_ap_completion(const struct lu_env *env, struct client_obd *cli,
ENTRY;
if (oap->oap_request != NULL) {
- if (rc == 0)
- osc_inc_unstable_pages(oap->oap_request);
-
xid = ptlrpc_req_xid(oap->oap_request);
ptlrpc_req_finished(oap->oap_request);
oap->oap_request = NULL;
oap->oap_interrupted = 0;
if (oap->oap_cmd & OBD_BRW_WRITE && xid > 0) {
- client_obd_list_lock(&cli->cl_loi_list_lock);
+ spin_lock(&cli->cl_loi_list_lock);
osc_process_ar(&cli->cl_ar, xid, rc);
osc_process_ar(&loi->loi_ar, xid, rc);
- client_obd_list_unlock(&cli->cl_loi_list_lock);
+ spin_unlock(&cli->cl_loi_list_lock);
}
rc = osc_completion(env, oap, oap->oap_cmd, rc);
static int try_to_add_extent_for_io(struct client_obd *cli,
struct osc_extent *ext,
struct list_head *rpclist,
- int *pc, unsigned int *max_pages)
+ unsigned int *pc, unsigned int *max_pages)
{
struct osc_extent *tmp;
+ struct osc_async_page *oap = list_first_entry(&ext->oe_pages,
+ struct osc_async_page,
+ oap_pending_item);
ENTRY;
EASSERT((ext->oe_state == OES_CACHE || ext->oe_state == OES_LOCK_DONE),
RETURN(0);
list_for_each_entry(tmp, rpclist, oe_link) {
+ struct osc_async_page *oap2;
+ oap2 = list_first_entry(&tmp->oe_pages, struct osc_async_page,
+ oap_pending_item);
EASSERT(tmp->oe_owner == current, tmp);
#if 0
if (overlapped(tmp, ext)) {
EASSERT(0, ext);
}
#endif
+ if (oap2cl_page(oap)->cp_type != oap2cl_page(oap2)->cp_type) {
+ CDEBUG(D_CACHE, "Do not permit different type of IO"
+ " for a same RPC\n");
+ RETURN(0);
+ }
if (tmp->oe_srvlock != ext->oe_srvlock ||
!tmp->oe_grants != !ext->oe_grants)
* 5. Traverse the extent tree from the 1st extent;
* 6. Above steps exit if there is no space in this RPC.
*/
-static int get_write_extents(struct osc_object *obj, struct list_head *rpclist)
+static unsigned int get_write_extents(struct osc_object *obj,
+ struct list_head *rpclist)
{
struct client_obd *cli = osc_cli(obj);
struct osc_extent *ext;
- int page_count = 0;
+ unsigned int page_count = 0;
unsigned int max_pages = cli->cl_max_pages_per_rpc;
LASSERT(osc_object_is_locked(obj));
static int
osc_send_write_rpc(const struct lu_env *env, struct client_obd *cli,
struct osc_object *osc, pdl_policy_t pol)
+__must_hold(osc)
{
struct list_head rpclist = LIST_HEAD_INIT(rpclist);
struct osc_extent *ext;
struct osc_extent *tmp;
struct osc_extent *first = NULL;
- obd_count page_count = 0;
+ unsigned int page_count = 0;
int srvlock = 0;
int rc = 0;
ENTRY;
static int
osc_send_read_rpc(const struct lu_env *env, struct client_obd *cli,
struct osc_object *osc, pdl_policy_t pol)
+__must_hold(osc)
{
struct osc_extent *ext;
struct osc_extent *next;
struct list_head rpclist = LIST_HEAD_INIT(rpclist);
- int page_count = 0;
+ unsigned int page_count = 0;
unsigned int max_pages = cli->cl_max_pages_per_rpc;
int rc = 0;
ENTRY;
/* called with the loi list lock held */
static void osc_check_rpcs(const struct lu_env *env, struct client_obd *cli,
pdl_policy_t pol)
+__must_hold(&cli->cl_loi_list_lock)
{
struct osc_object *osc;
int rc = 0;
}
cl_object_get(obj);
- client_obd_list_unlock(&cli->cl_loi_list_lock);
+ spin_unlock(&cli->cl_loi_list_lock);
lu_object_ref_add_at(&obj->co_lu, &link, "check", current);
/* attempt some read/write balancing by alternating between
lu_object_ref_del_at(&obj->co_lu, &link, "check", current);
cl_object_put(env, obj);
- client_obd_list_lock(&cli->cl_loi_list_lock);
+ spin_lock(&cli->cl_loi_list_lock);
}
}
/* disable osc_lru_shrink() temporarily to avoid
* potential stack overrun problem. LU-2859 */
atomic_inc(&cli->cl_lru_shrinkers);
- client_obd_list_lock(&cli->cl_loi_list_lock);
+ spin_lock(&cli->cl_loi_list_lock);
osc_check_rpcs(env, cli, pol);
- client_obd_list_unlock(&cli->cl_loi_list_lock);
+ spin_unlock(&cli->cl_loi_list_lock);
atomic_dec(&cli->cl_lru_shrinkers);
} else {
CDEBUG(D_CACHE, "Queue writeback work for client %p.\n", cli);
struct client_obd *cli = oap->oap_cli;
struct osc_object *osc = oap->oap_obj;
pgoff_t index;
- int grants = 0;
+ unsigned int tmp;
+ unsigned int grants = 0;
int brw_flags = OBD_BRW_ASYNC;
int cmd = OBD_BRW_WRITE;
int need_release = 0;
RETURN(rc);
}
- if (osc_over_unstable_soft_limit(cli))
- brw_flags |= OBD_BRW_SOFT_SYNC;
-
oap->oap_cmd = cmd;
oap->oap_page_off = ops->ops_from;
oap->oap_count = ops->ops_to - ops->ops_from;
grants = 0;
/* it doesn't need any grant to dirty this page */
- client_obd_list_lock(&cli->cl_loi_list_lock);
+ spin_lock(&cli->cl_loi_list_lock);
rc = osc_enter_cache_try(cli, oap, grants, 0);
- client_obd_list_unlock(&cli->cl_loi_list_lock);
+ spin_unlock(&cli->cl_loi_list_lock);
if (rc == 0) { /* try failed */
grants = 0;
need_release = 1;
} else if (ext->oe_end < index) {
- int tmp = grants;
+ tmp = grants;
/* try to expand this extent */
rc = osc_extent_expand(ext, index, &tmp);
if (rc < 0) {
}
if (ext == NULL) {
- int tmp = (1 << cli->cl_chunkbits) + cli->cl_extent_tax;
+ tmp = (1 << cli->cl_chunkbits) + cli->cl_extent_tax;
/* try to find new extent to cover this page */
LASSERT(oio->oi_active == NULL);
}
ext->oe_rw = !!(cmd & OBD_BRW_READ);
+ ext->oe_sync = 1;
ext->oe_urgent = 1;
ext->oe_start = start;
ext->oe_end = ext->oe_max_end = end;
}
osc_object_unlock(obj);
- osc_io_unplug(env, cli, obj, PDL_POLICY_ROUND);
+ osc_io_unplug_async(env, cli, obj);
RETURN(0);
}
struct osc_page *ops, void *cbdata)
{
struct osc_thread_info *info = osc_env_info(env);
- struct cl_lock *lock = cbdata;
+ struct osc_object *osc = cbdata;
pgoff_t index;
index = osc_index(ops);
if (index >= info->oti_fn_index) {
- struct cl_lock *tmp;
+ struct ldlm_lock *tmp;
struct cl_page *page = ops->ops_cl.cpl_page;
/* refresh non-overlapped index */
- tmp = cl_lock_at_pgoff(env, lock->cll_descr.cld_obj, index,
- lock, 1, 0);
+ tmp = osc_dlmlock_at_pgoff(env, osc, index, 0, 0);
if (tmp != NULL) {
+ __u64 end = tmp->l_policy_data.l_extent.end;
/* Cache the first-non-overlapped index so as to skip
- * all pages within [index, oti_fn_index). This
- * is safe because if tmp lock is canceled, it will
- * discard these pages. */
- info->oti_fn_index = tmp->cll_descr.cld_end + 1;
- if (tmp->cll_descr.cld_end == CL_PAGE_EOF)
+ * all pages within [index, oti_fn_index). This is safe
+ * because if tmp lock is canceled, it will discard
+ * these pages. */
+ info->oti_fn_index = cl_index(osc2cl(osc), end + 1);
+ if (end == OBD_OBJECT_EOF)
info->oti_fn_index = CL_PAGE_EOF;
- cl_lock_put(env, tmp);
+ LDLM_LOCK_PUT(tmp);
} else if (cl_page_own(env, io, page) == 0) {
/* discard the page */
cl_page_discard(env, io, page);
struct osc_page *ops, void *cbdata)
{
struct osc_thread_info *info = osc_env_info(env);
- struct cl_lock *lock = cbdata;
struct cl_page *page = ops->ops_cl.cpl_page;
- LASSERT(lock->cll_descr.cld_mode >= CLM_WRITE);
-
/* page is top page. */
info->oti_next_index = osc_index(ops) + 1;
if (cl_page_own(env, io, page) == 0) {
* If error happens on any step, the process continues anyway (the reasoning
* behind this being that lock cancellation cannot be delayed indefinitely).
*/
-int osc_lock_discard_pages(const struct lu_env *env, struct osc_lock *ols)
+int osc_lock_discard_pages(const struct lu_env *env, struct osc_object *osc,
+ pgoff_t start, pgoff_t end, enum cl_lock_mode mode)
{
struct osc_thread_info *info = osc_env_info(env);
struct cl_io *io = &info->oti_io;
- struct cl_object *osc = ols->ols_cl.cls_obj;
- struct cl_lock *lock = ols->ols_cl.cls_lock;
- struct cl_lock_descr *descr = &lock->cll_descr;
osc_page_gang_cbt cb;
int res;
int result;
ENTRY;
- io->ci_obj = cl_object_top(osc);
+ io->ci_obj = cl_object_top(osc2cl(osc));
io->ci_ignore_layout = 1;
result = cl_io_init(env, io, CIT_MISC, io->ci_obj);
if (result != 0)
GOTO(out, result);
- cb = descr->cld_mode == CLM_READ ? check_and_discard_cb : discard_cb;
- info->oti_fn_index = info->oti_next_index = descr->cld_start;
+ cb = mode == CLM_READ ? check_and_discard_cb : discard_cb;
+ info->oti_fn_index = info->oti_next_index = start;
do {
- res = osc_page_gang_lookup(env, io, cl2osc(osc),
- info->oti_next_index, descr->cld_end,
- cb, (void *)lock);
- if (info->oti_next_index > descr->cld_end)
+ res = osc_page_gang_lookup(env, io, osc,
+ info->oti_next_index, end, cb, osc);
+ if (info->oti_next_index > end)
break;
if (res == CLP_GANG_RESCHED)
git://git.whamcloud.com / fs / lustre-release.git / blobdiff