#define EXTSTR "[%lu -> %lu/%lu]"
#define EXTPARA(ext) (ext)->oe_start, (ext)->oe_end, (ext)->oe_max_end
+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); \
- const char *__str[] = OES_STRINGS; \
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 ----- */ \
+ /* ----- extent part 0 ----- */ \
__ext, EXTPARA(__ext), \
/* ----- part 1 ----- */ \
cfs_atomic_read(&__ext->oe_refc), \
cfs_atomic_read(&__ext->oe_users), \
list_empty_marker(&__ext->oe_link), \
- __str[__ext->oe_state], ext_flags(__ext, __buf), \
+ oes_strings[__ext->oe_state], ext_flags(__ext, __buf), \
__ext->oe_obj, \
/* ----- part 2 ----- */ \
__ext->oe_grants, __ext->oe_nr_pages, \
#undef EASSERTF
#define EASSERTF(expr, ext, fmt, args...) do { \
if (!(expr)) { \
- OSC_EXTENT_DUMP(D_ERROR, (ext), fmt, ##args); \
- osc_extent_tree_dump(D_ERROR, (ext)->oe_obj); \
+ OSC_EXTENT_DUMP(D_ERROR, (ext), fmt, ##args); \
+ osc_extent_tree_dump(D_ERROR, (ext)->oe_obj); \
LASSERT(expr); \
- } \
+ } \
} while (0)
#undef EASSERT
{
struct osc_extent *ext;
- OBD_SLAB_ALLOC_PTR_GFP(ext, osc_extent_kmem, CFS_ALLOC_STD);
+ OBD_SLAB_ALLOC_PTR_GFP(ext, osc_extent_kmem, GFP_IOFS);
if (ext == NULL)
return NULL;
return -ERANGE;
LASSERT(cur->oe_osclock == victim->oe_osclock);
- ppc_bits = osc_cli(obj)->cl_chunkbits - CFS_PAGE_SHIFT;
+ ppc_bits = osc_cli(obj)->cl_chunkbits - PAGE_CACHE_SHIFT;
chunk_start = cur->oe_start >> ppc_bits;
chunk_end = cur->oe_end >> ppc_bits;
if (chunk_start != (victim->oe_end >> ppc_bits) + 1 &&
LASSERT(lock != NULL);
LASSERT(lock->cll_descr.cld_mode >= CLM_WRITE);
- LASSERT(cli->cl_chunkbits >= CFS_PAGE_SHIFT);
- ppc_bits = cli->cl_chunkbits - CFS_PAGE_SHIFT;
+ LASSERT(cli->cl_chunkbits >= PAGE_CACHE_SHIFT);
+ ppc_bits = cli->cl_chunkbits - PAGE_CACHE_SHIFT;
chunk_mask = ~((1 << ppc_bits) - 1);
chunksize = 1 << cli->cl_chunkbits;
chunk = index >> ppc_bits;
if (!sent) {
lost_grant = ext->oe_grants;
- } else if (blocksize < CFS_PAGE_SIZE &&
- last_count != CFS_PAGE_SIZE) {
+ } else if (blocksize < PAGE_CACHE_SIZE &&
+ last_count != PAGE_CACHE_SIZE) {
/* For short writes we shouldn't count parts of pages that
* span a whole chunk on the OST side, or our accounting goes
* wrong. Should match the code in filter_grant_check. */
if (end)
count += blocksize - end;
- lost_grant = CFS_PAGE_SIZE - count;
+ lost_grant = PAGE_CACHE_SIZE - count;
}
if (ext->oe_grants > 0)
osc_free_grant(cli, nr_pages, lost_grant);
struct osc_async_page *oap;
struct osc_async_page *tmp;
int pages_in_chunk = 0;
- int ppc_bits = cli->cl_chunkbits - CFS_PAGE_SHIFT;
+ int ppc_bits = cli->cl_chunkbits -
+ PAGE_CACHE_SHIFT;
__u64 trunc_chunk = trunc_index >> ppc_bits;
int grants = 0;
int nr_pages = 0;
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 <= CFS_PAGE_SIZE);
+ LASSERT(last->oap_page_off + last->oap_count <= PAGE_CACHE_SIZE);
last->oap_async_flags |= ASYNC_COUNT_STABLE;
}
* because it's known they are not the last page */
cfs_list_for_each_entry(oap, &ext->oe_pages, oap_pending_item) {
if (!(oap->oap_async_flags & ASYNC_COUNT_STABLE)) {
- oap->oap_count = CFS_PAGE_SIZE - oap->oap_page_off;
+ oap->oap_count = PAGE_CACHE_SIZE - oap->oap_page_off;
oap->oap_async_flags |= ASYNC_COUNT_STABLE;
}
}
struct osc_object *obj = ext->oe_obj;
struct client_obd *cli = osc_cli(obj);
struct osc_extent *next;
- int ppc_bits = cli->cl_chunkbits - CFS_PAGE_SHIFT;
+ int ppc_bits = cli->cl_chunkbits - PAGE_CACHE_SHIFT;
pgoff_t chunk = index >> ppc_bits;
pgoff_t end_chunk;
pgoff_t end_index;
return 0;
else if (cl_offset(obj, page->cp_index + 1) > kms)
/* catch sub-page write at end of file */
- return kms % CFS_PAGE_SIZE;
+ return kms % PAGE_CACHE_SIZE;
else
- return CFS_PAGE_SIZE;
+ return PAGE_CACHE_SIZE;
}
static int osc_completion(const struct lu_env *env, struct osc_async_page *oap,
#define OSC_DUMP_GRANT(cli, fmt, args...) do { \
struct client_obd *__tmp = (cli); \
CDEBUG(D_CACHE, "%s: { dirty: %ld/%ld dirty_pages: %d/%d " \
- "unstable_pages: %d/%d dropped: %ld avail: %ld, " \
- "reserved: %ld, flight: %d } " fmt, \
+ "dropped: %ld avail: %ld, reserved: %ld, flight: %d } " fmt, \
__tmp->cl_import->imp_obd->obd_name, \
__tmp->cl_dirty, __tmp->cl_dirty_max, \
cfs_atomic_read(&obd_dirty_pages), obd_max_dirty_pages, \
- cfs_atomic_read(&obd_unstable_pages), obd_max_dirty_pages, \
__tmp->cl_lost_grant, __tmp->cl_avail_grant, \
__tmp->cl_reserved_grant, __tmp->cl_w_in_flight, ##args); \
} while (0)
static void osc_consume_write_grant(struct client_obd *cli,
struct brw_page *pga)
{
- LASSERT_SPIN_LOCKED(&cli->cl_loi_list_lock.lock);
+ LASSERT(spin_is_locked(&cli->cl_loi_list_lock.lock));
LASSERT(!(pga->flag & OBD_BRW_FROM_GRANT));
cfs_atomic_inc(&obd_dirty_pages);
- cli->cl_dirty += CFS_PAGE_SIZE;
+ cli->cl_dirty += PAGE_CACHE_SIZE;
pga->flag |= OBD_BRW_FROM_GRANT;
CDEBUG(D_CACHE, "using %lu grant credits for brw %p page %p\n",
- CFS_PAGE_SIZE, pga, pga->pg);
+ PAGE_CACHE_SIZE, pga, pga->pg);
osc_update_next_shrink(cli);
}
{
ENTRY;
- LASSERT_SPIN_LOCKED(&cli->cl_loi_list_lock.lock);
+ LASSERT(spin_is_locked(&cli->cl_loi_list_lock.lock));
if (!(pga->flag & OBD_BRW_FROM_GRANT)) {
EXIT;
return;
pga->flag &= ~OBD_BRW_FROM_GRANT;
cfs_atomic_dec(&obd_dirty_pages);
- cli->cl_dirty -= CFS_PAGE_SIZE;
+ cli->cl_dirty -= PAGE_CACHE_SIZE;
if (pga->flag & OBD_BRW_NOCACHE) {
pga->flag &= ~OBD_BRW_NOCACHE;
cfs_atomic_dec(&obd_dirty_transit_pages);
- cli->cl_dirty_transit -= CFS_PAGE_SIZE;
+ cli->cl_dirty_transit -= PAGE_CACHE_SIZE;
}
EXIT;
}
* used, we should return these grants to OST. There're two cases where grants
* can be lost:
* 1. truncate;
- * 2. blocksize at OST is less than CFS_PAGE_SIZE and a partial page was
+ * 2. blocksize at OST is less than PAGE_CACHE_SIZE and a partial page was
* written. In this case OST may use less chunks to serve this partial
* write. OSTs don't actually know the page size on the client side. so
* clients have to calculate lost grant by the blocksize on the OST.
client_obd_list_lock(&cli->cl_loi_list_lock);
cfs_atomic_sub(nr_pages, &obd_dirty_pages);
- cli->cl_dirty -= nr_pages << CFS_PAGE_SHIFT;
+ cli->cl_dirty -= nr_pages << PAGE_CACHE_SHIFT;
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
if (rc < 0)
return 0;
- if (cli->cl_dirty + CFS_PAGE_SIZE <= cli->cl_dirty_max &&
- cfs_atomic_read(&obd_unstable_pages) + 1 +
- cfs_atomic_read(&obd_dirty_pages) <= obd_max_dirty_pages) {
+ if (cli->cl_dirty + PAGE_CACHE_SIZE <= cli->cl_dirty_max &&
+ cfs_atomic_read(&obd_dirty_pages) + 1 <= obd_max_dirty_pages) {
osc_consume_write_grant(cli, &oap->oap_brw_page);
if (transient) {
- cli->cl_dirty_transit += CFS_PAGE_SIZE;
+ cli->cl_dirty_transit += PAGE_CACHE_SIZE;
cfs_atomic_inc(&obd_dirty_transit_pages);
oap->oap_brw_flags |= OBD_BRW_NOCACHE;
}
/* 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 < CFS_PAGE_SIZE ||
+ cli->cl_dirty_max < PAGE_CACHE_SIZE ||
cli->cl_ar.ar_force_sync || loi->loi_ar.ar_force_sync)
GOTO(out, rc = -EDQUOT);
ocw->ocw_rc = -EDQUOT;
/* we can't dirty more */
- if (cli->cl_dirty + CFS_PAGE_SIZE > cli->cl_dirty_max ||
- cfs_atomic_read(&obd_unstable_pages) + 1 +
- cfs_atomic_read(&obd_dirty_pages) > obd_max_dirty_pages) {
+ if ((cli->cl_dirty + PAGE_CACHE_SIZE > cli->cl_dirty_max) ||
+ (cfs_atomic_read(&obd_dirty_pages) + 1 >
+ 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);
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;
- obd_count 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);
-
- cfs_atomic_sub(page_count, &cli->cl_cache->ccc_unstable_nr);
- LASSERT(cfs_atomic_read(&cli->cl_cache->ccc_unstable_nr) >= 0);
-
- cfs_atomic_sub(page_count, &obd_unstable_pages);
- LASSERT(cfs_atomic_read(&obd_unstable_pages) >= 0);
-
- spin_lock(&req->rq_lock);
- req->rq_committed = 1;
- req->rq_unstable = 0;
- spin_unlock(&req->rq_lock);
-
- cfs_waitq_broadcast(&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;
- obd_count 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(cfs_atomic_read(&cli->cl_cache->ccc_unstable_nr) >= 0);
- cfs_atomic_add(page_count, &cli->cl_cache->ccc_unstable_nr);
-
- LASSERT(cfs_atomic_read(&obd_unstable_pages) >= 0);
- cfs_atomic_add(page_count, &obd_unstable_pages);
-
- spin_lock(&req->rq_lock);
-
- /* 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. Otherwise, just set the commit callback so the
- * unstable page accounting is properly updated when the request
- * is committed */
- if (req->rq_committed) {
- /* Drop lock before calling osc_dec_unstable_pages */
- spin_unlock(&req->rq_lock);
- osc_dec_unstable_pages(req);
- spin_lock(&req->rq_lock);
- } else {
- req->rq_unstable = 1;
- req->rq_commit_cb = osc_dec_unstable_pages;
- }
-
- 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;
while ((osc = osc_next_obj(cli)) != NULL) {
struct cl_object *obj = osc2cl(osc);
- struct lu_ref_link *link;
+ struct lu_ref_link link;
OSC_IO_DEBUG(osc, "%lu in flight\n", rpcs_in_flight(cli));
cl_object_get(obj);
client_obd_list_unlock(&cli->cl_loi_list_lock);
- link = lu_object_ref_add(&obj->co_lu, "check", cfs_current());
+ lu_object_ref_add_at(&obj->co_lu, &link, "check",
+ cfs_current());
/* attempt some read/write balancing by alternating between
* reads and writes in an object. The makes_rpc checks here
osc_object_unlock(osc);
osc_list_maint(cli, osc);
- lu_object_ref_del_at(&obj->co_lu, link, "check", cfs_current());
+ lu_object_ref_del_at(&obj->co_lu, &link, "check",
+ cfs_current());
cl_object_put(env, obj);
client_obd_list_lock(&cli->cl_loi_list_lock);
static int osc_io_unplug0(const struct lu_env *env, struct client_obd *cli,
struct osc_object *osc, pdl_policy_t pol, int async)
{
- int has_rpcs = 1;
int rc = 0;
- client_obd_list_lock(&cli->cl_loi_list_lock);
- if (osc != NULL)
- has_rpcs = __osc_list_maint(cli, osc);
- if (has_rpcs) {
- if (!async) {
- /* disable osc_lru_shrink() temporarily to avoid
- * potential stack overrun problem. LU-2859 */
- cfs_atomic_inc(&cli->cl_lru_shrinkers);
- osc_check_rpcs(env, cli, pol);
- cfs_atomic_dec(&cli->cl_lru_shrinkers);
- } else {
- CDEBUG(D_CACHE, "Queue writeback work for client %p.\n",
- cli);
- LASSERT(cli->cl_writeback_work != NULL);
- rc = ptlrpcd_queue_work(cli->cl_writeback_work);
- }
+ if (osc != NULL && osc_list_maint(cli, osc) == 0)
+ return 0;
+
+ if (!async) {
+ /* disable osc_lru_shrink() temporarily to avoid
+ * potential stack overrun problem. LU-2859 */
+ cfs_atomic_inc(&cli->cl_lru_shrinkers);
+ client_obd_list_lock(&cli->cl_loi_list_lock);
+ osc_check_rpcs(env, cli, pol);
+ client_obd_list_unlock(&cli->cl_loi_list_lock);
+ cfs_atomic_dec(&cli->cl_lru_shrinkers);
+ } else {
+ CDEBUG(D_CACHE, "Queue writeback work for client %p.\n", cli);
+ LASSERT(cli->cl_writeback_work != NULL);
+ rc = ptlrpcd_queue_work(cli->cl_writeback_work);
}
- client_obd_list_unlock(&cli->cl_loi_list_lock);
return rc;
}
}
int osc_prep_async_page(struct osc_object *osc, struct osc_page *ops,
- cfs_page_t *page, loff_t offset)
+ struct page *page, loff_t offset)
{
struct obd_export *exp = osc_export(osc);
struct osc_async_page *oap = &ops->ops_oap;
oap->oap_async_flags |= ASYNC_READY|ASYNC_URGENT;
spin_unlock(&oap->oap_lock);
- if (cfs_memory_pressure_get())
+ if (memory_pressure_get())
ext->oe_memalloc = 1;
ext->oe_urgent = 1;
}
osc_object_unlock(obj);
- osc_io_unplug_async(env, cli, obj);
+ osc_io_unplug(env, cli, obj, PDL_POLICY_ROUND);
RETURN(0);
}
* should take care of it. */
rc = osc_extent_wait(env, waiting, OES_INV);
if (rc < 0)
- OSC_EXTENT_DUMP(D_CACHE, ext, "wait error: %d.\n", rc);
+ OSC_EXTENT_DUMP(D_CACHE, waiting, "error: %d.\n", rc);
osc_extent_put(env, waiting);
waiting = NULL;
git://git.whamcloud.com / fs / lustre-release.git / blobdiff