X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosc%2Fosc_cache.c;h=2481e51d460d976ce7638e5b6b2ccab72a477287;hp=f533e48053ed3658905f20a3079bc69cdc1203b4;hb=6bce536725efd166d2772f13fe954f271f9c53b8;hpb=7b63a5dab65cde131627bf22d16d6e13cf259686;ds=sidebyside diff --git a/lustre/osc/osc_cache.c b/lustre/osc/osc_cache.c index f533e48..2481e51 100644 --- a/lustre/osc/osc_cache.c +++ b/lustre/osc/osc_cache.c @@ -15,11 +15,7 @@ * * You should have received a copy of the GNU General Public License * version 2 along with this program; If not, see - * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf - * - * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, - * CA 95054 USA or visit www.sun.com if you need additional information or - * have any questions. + * http://www.gnu.org/licenses/gpl-2.0.html * * GPL HEADER END */ @@ -27,7 +23,7 @@ * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2012, 2014, Intel Corporation. + * Copyright (c) 2012, 2017, Intel Corporation. * */ /* @@ -41,7 +37,8 @@ #define DEBUG_SUBSYSTEM S_OSC -#include "osc_cl_internal.h" +#include + #include "osc_internal.h" static int extent_debug; /* set it to be true for more debug */ @@ -58,12 +55,15 @@ static int osc_refresh_count(const struct lu_env *env, static int osc_io_unplug_async(const struct lu_env *env, struct client_obd *cli, struct osc_object *osc); static void osc_free_grant(struct client_obd *cli, unsigned int nr_pages, - unsigned int lost_grant); + unsigned int lost_grant, unsigned int dirty_grant); -static void osc_extent_tree_dump0(int level, struct osc_object *obj, +static void osc_extent_tree_dump0(int mask, struct osc_object *obj, const char *func, int line); -#define osc_extent_tree_dump(lvl, obj) \ - osc_extent_tree_dump0(lvl, obj, __func__, __LINE__) +#define osc_extent_tree_dump(mask, obj) \ + osc_extent_tree_dump0(mask, obj, __func__, __LINE__) + +static void osc_unreserve_grant(struct client_obd *cli, unsigned int reserved, + unsigned int unused); /** \addtogroup osc * @{ @@ -104,18 +104,19 @@ static inline char list_empty_marker(struct list_head *list) static const char *oes_strings[] = { "inv", "active", "cache", "locking", "lockdone", "rpc", "trunc", NULL }; -#define OSC_EXTENT_DUMP(lvl, extent, fmt, ...) do { \ +#define OSC_EXTENT_DUMP_WITH_LOC(file, func, line, mask, extent, fmt, ...) do {\ + static struct cfs_debug_limit_state cdls; \ struct osc_extent *__ext = (extent); \ char __buf[16]; \ \ - CDEBUG(lvl, \ + __CDEBUG_WITH_LOC(file, func, line, mask, &cdls, \ "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), \ + 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, \ @@ -126,12 +127,16 @@ static const char *oes_strings[] = { __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); \ + if (mask == D_ERROR && __ext->oe_dlmlock != NULL) \ + LDLM_ERROR(__ext->oe_dlmlock, "extent: %p", __ext); \ else \ - LDLM_DEBUG(__ext->oe_dlmlock, "extent: %p\n", __ext); \ + LDLM_DEBUG(__ext->oe_dlmlock, "extent: %p", __ext); \ } while (0) +#define OSC_EXTENT_DUMP(mask, ext, fmt, ...) \ + OSC_EXTENT_DUMP_WITH_LOC(__FILE__, __func__, __LINE__, \ + mask, ext, fmt, ## __VA_ARGS__) + #undef EASSERTF #define EASSERTF(expr, ext, fmt, args...) do { \ if (!(expr)) { \ @@ -215,6 +220,7 @@ static int osc_extent_sanity_check0(struct osc_extent *ext, GOTO(out, rc = 60); if (ext->oe_fsync_wait && !ext->oe_urgent && !ext->oe_hp) GOTO(out, rc = 65); + /* fallthrough */ default: if (atomic_read(&ext->oe_users) > 0) GOTO(out, rc = 70); @@ -226,7 +232,9 @@ static int osc_extent_sanity_check0(struct osc_extent *ext, if (ext->oe_sync && ext->oe_grants > 0) GOTO(out, rc = 90); - if (ext->oe_dlmlock != NULL) { + if (ext->oe_dlmlock != NULL && + ext->oe_dlmlock->l_resource->lr_type == LDLM_EXTENT && + !ldlm_is_failed(ext->oe_dlmlock)) { struct ldlm_extent *extent; extent = &ext->oe_dlmlock->l_policy_data.l_extent; @@ -261,9 +269,9 @@ static int osc_extent_sanity_check0(struct osc_extent *ext, out: if (rc != 0) - OSC_EXTENT_DUMP(D_ERROR, ext, - "%s:%d sanity check %p failed with rc = %d\n", - func, line, ext, rc); + OSC_EXTENT_DUMP_WITH_LOC(__FILE__, func, line, D_ERROR, ext, + "sanity check %p failed: rc = %d\n", + ext, rc); return rc; } @@ -319,12 +327,13 @@ static struct osc_extent *osc_extent_alloc(struct osc_object *obj) { struct osc_extent *ext; - OBD_SLAB_ALLOC_PTR_GFP(ext, osc_extent_kmem, GFP_IOFS); + OBD_SLAB_ALLOC_PTR_GFP(ext, osc_extent_kmem, GFP_NOFS); if (ext == NULL) return NULL; RB_CLEAR_NODE(&ext->oe_node); ext->oe_obj = obj; + cl_object_get(osc2cl(obj)); atomic_set(&ext->oe_refc, 1); atomic_set(&ext->oe_users, 0); INIT_LIST_HEAD(&ext->oe_link); @@ -363,6 +372,7 @@ static void osc_extent_put(const struct lu_env *env, struct osc_extent *ext) LDLM_LOCK_PUT(ext->oe_dlmlock); ext->oe_dlmlock = NULL; } + cl_object_put(env, osc2cl(ext->oe_obj)); osc_extent_free(ext); } } @@ -495,15 +505,16 @@ static void osc_extent_remove(struct osc_extent *ext) /** * This function is used to merge extents to get better performance. It checks - * if @cur and @victim are contiguous at chunk level. + * if @cur and @victim are contiguous at block level. */ static int osc_extent_merge(const struct lu_env *env, struct osc_extent *cur, struct osc_extent *victim) { - struct osc_object *obj = cur->oe_obj; - pgoff_t chunk_start; - pgoff_t chunk_end; - int ppc_bits; + struct osc_object *obj = cur->oe_obj; + struct client_obd *cli = osc_cli(obj); + pgoff_t chunk_start; + pgoff_t chunk_end; + int ppc_bits; LASSERT(cur->oe_state == OES_CACHE); LASSERT(osc_object_is_locked(obj)); @@ -517,18 +528,25 @@ static int osc_extent_merge(const struct lu_env *env, struct osc_extent *cur, return -ERANGE; LASSERT(cur->oe_dlmlock == victim->oe_dlmlock); - ppc_bits = osc_cli(obj)->cl_chunkbits - PAGE_CACHE_SHIFT; + ppc_bits = osc_cli(obj)->cl_chunkbits - PAGE_SHIFT; chunk_start = cur->oe_start >> ppc_bits; chunk_end = cur->oe_end >> ppc_bits; if (chunk_start != (victim->oe_end >> ppc_bits) + 1 && chunk_end + 1 != victim->oe_start >> ppc_bits) return -ERANGE; + /* overall extent size should not exceed the max supported limit + * reported by the server */ + if (cur->oe_end - cur->oe_start + 1 + + victim->oe_end - victim->oe_start + 1 > cli->cl_max_extent_pages) + return -ERANGE; + OSC_EXTENT_DUMP(D_CACHE, victim, "will be merged by %p.\n", cur); cur->oe_start = min(cur->oe_start, victim->oe_start); cur->oe_end = max(cur->oe_end, victim->oe_end); - cur->oe_grants += victim->oe_grants; + /* per-extent tax should be accounted only once for the whole extent */ + cur->oe_grants += victim->oe_grants - cli->cl_grant_extent_tax; cur->oe_nr_pages += victim->oe_nr_pages; /* only the following bits are needed to merge */ cur->oe_urgent |= victim->oe_urgent; @@ -551,6 +569,7 @@ static int osc_extent_merge(const struct lu_env *env, struct osc_extent *cur, int osc_extent_release(const struct lu_env *env, struct osc_extent *ext) { struct osc_object *obj = ext->oe_obj; + struct client_obd *cli = osc_cli(obj); int rc = 0; ENTRY; @@ -566,28 +585,40 @@ int osc_extent_release(const struct lu_env *env, struct osc_extent *ext) * osc_cache_truncate_start(). */ osc_extent_state_set(ext, OES_TRUNC); ext->oe_trunc_pending = 0; + osc_object_unlock(obj); } else { + int grant = 0; + osc_extent_state_set(ext, OES_CACHE); osc_update_pending(obj, OBD_BRW_WRITE, ext->oe_nr_pages); /* try to merge the previous and next extent. */ - osc_extent_merge(env, ext, prev_extent(ext)); - osc_extent_merge(env, ext, next_extent(ext)); + if (osc_extent_merge(env, ext, prev_extent(ext)) == 0) + grant += cli->cl_grant_extent_tax; + if (osc_extent_merge(env, ext, next_extent(ext)) == 0) + grant += cli->cl_grant_extent_tax; if (ext->oe_urgent) list_move_tail(&ext->oe_link, &obj->oo_urgent_exts); + else if (ext->oe_nr_pages == ext->oe_mppr) { + list_move_tail(&ext->oe_link, + &obj->oo_full_exts); + } + osc_object_unlock(obj); + if (grant > 0) + osc_unreserve_grant(cli, 0, grant); } - osc_object_unlock(obj); - osc_io_unplug_async(env, osc_cli(obj), obj); + osc_io_unplug_async(env, cli, obj); } osc_extent_put(env, ext); RETURN(rc); } -static inline int overlapped(struct osc_extent *ex1, struct osc_extent *ex2) +static inline bool +overlapped(const struct osc_extent *ex1, const struct osc_extent *ex2) { return !(ex1->oe_end < ex2->oe_start || ex2->oe_end < ex1->oe_start); } @@ -627,15 +658,20 @@ static struct osc_extent *osc_extent_find(const struct lu_env *env, 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; + LASSERTF(cli->cl_chunkbits >= PAGE_SHIFT, + "chunkbits: %u\n", cli->cl_chunkbits); + ppc_bits = cli->cl_chunkbits - PAGE_SHIFT; chunk_mask = ~((1 << ppc_bits) - 1); chunksize = 1 << cli->cl_chunkbits; chunk = index >> ppc_bits; - /* align end to rpc edge, rpc size may not be a power 2 integer. */ + /* align end to RPC edge. */ max_pages = cli->cl_max_pages_per_rpc; - LASSERT((max_pages & ~chunk_mask) == 0); + if ((max_pages & ~chunk_mask) != 0) { + CERROR("max_pages: %#x chunkbits: %u chunk_mask: %#lx\n", + max_pages, cli->cl_chunkbits, chunk_mask); + RETURN(ERR_PTR(-EINVAL)); + } max_end = index - (index % max_pages) + max_pages - 1; max_end = min_t(pgoff_t, max_end, descr->cld_end); @@ -656,8 +692,8 @@ static struct osc_extent *osc_extent_find(const struct lu_env *env, } /* grants has been allocated by caller */ - LASSERTF(*grants >= chunksize + cli->cl_extent_tax, - "%u/%u/%u.\n", *grants, chunksize, cli->cl_extent_tax); + LASSERTF(*grants >= chunksize + cli->cl_grant_extent_tax, + "%u/%u/%u.\n", *grants, chunksize, cli->cl_grant_extent_tax); LASSERTF((max_end - cur->oe_start) < max_pages, EXTSTR"\n", EXTPARA(cur)); @@ -671,7 +707,7 @@ restart: pgoff_t ext_chk_end = ext->oe_end >> ppc_bits; LASSERT(sanity_check_nolock(ext) == 0); - if (chunk > ext_chk_end + 1) + if (chunk > ext_chk_end + 1 || chunk < ext_chk_start) break; /* if covering by different locks, no chance to match */ @@ -730,6 +766,13 @@ restart: continue; } + /* check whether maximum extent size will be hit */ + if ((ext_chk_end - ext_chk_start + 1 + 1) << ppc_bits > + cli->cl_max_extent_pages) { + ext = next_extent(ext); + continue; + } + /* it's required that an extent must be contiguous at chunk * level so that we know the whole extent is covered by grant * (the pages in the extent are NOT required to be contiguous). @@ -744,6 +787,7 @@ restart: /* pull ext's start back to cover cur */ ext->oe_start = cur->oe_start; ext->oe_grants += chunksize; + LASSERT(*grants >= chunksize); *grants -= chunksize; found = osc_extent_hold(ext); @@ -751,13 +795,14 @@ restart: /* rear merge */ ext->oe_end = cur->oe_end; ext->oe_grants += chunksize; + LASSERT(*grants >= chunksize); *grants -= chunksize; /* try to merge with the next one because we just fill * in a gap */ if (osc_extent_merge(env, ext, next_extent(ext)) == 0) /* we can save extent tax from next extent */ - *grants += cli->cl_extent_tax; + *grants += cli->cl_grant_extent_tax; found = osc_extent_hold(ext); } @@ -778,9 +823,9 @@ restart: } else if (conflict == NULL) { /* create a new extent */ EASSERT(osc_extent_is_overlapped(obj, cur) == 0, cur); - cur->oe_grants = chunksize + cli->cl_extent_tax; + cur->oe_grants = chunksize + cli->cl_grant_extent_tax; + LASSERT(*grants >= cur->oe_grants); *grants -= cur->oe_grants; - LASSERT(*grants >= 0); cur->oe_state = OES_CACHE; found = osc_extent_hold(cur); @@ -807,7 +852,6 @@ restart: out: osc_extent_put(env, cur); - LASSERT(*grants >= 0); return found; } @@ -849,21 +893,21 @@ int osc_extent_finish(const struct lu_env *env, struct osc_extent *ext, if (!sent) { lost_grant = ext->oe_grants; - } else if (blocksize < PAGE_CACHE_SIZE && - last_count != PAGE_CACHE_SIZE) { + } else if (blocksize < PAGE_SIZE && + last_count != PAGE_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. */ - int offset = last_off & ~CFS_PAGE_MASK; + int offset = last_off & ~PAGE_MASK; int count = last_count + (offset & (blocksize - 1)); int end = (offset + last_count) & (blocksize - 1); if (end) count += blocksize - end; - lost_grant = PAGE_CACHE_SIZE - count; + lost_grant = PAGE_SIZE - count; } if (ext->oe_grants > 0) - osc_free_grant(cli, nr_pages, lost_grant); + osc_free_grant(cli, nr_pages, lost_grant, ext->oe_grants); osc_extent_remove(ext); /* put the refcount for RPC */ @@ -889,8 +933,6 @@ static int osc_extent_wait(const struct lu_env *env, struct osc_extent *ext, 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, - LWI_ON_SIGNAL_NOOP, NULL); int rc = 0; ENTRY; @@ -912,18 +954,19 @@ static int osc_extent_wait(const struct lu_env *env, struct osc_extent *ext, osc_extent_release(env, ext); /* wait for the extent until its state becomes @state */ - rc = l_wait_event(ext->oe_waitq, extent_wait_cb(ext, state), &lwi); - if (rc == -ETIMEDOUT) { + rc = wait_event_idle_timeout(ext->oe_waitq, extent_wait_cb(ext, state), + cfs_time_seconds(600)); + if (rc == 0) { OSC_EXTENT_DUMP(D_ERROR, ext, "%s: wait ext to %u timedout, recovery in progress?\n", - osc_export(obj)->exp_obd->obd_name, state); + cli_name(osc_cli(obj)), state); - lwi = LWI_INTR(NULL, NULL); - rc = l_wait_event(ext->oe_waitq, extent_wait_cb(ext, state), - &lwi); + wait_event_idle(ext->oe_waitq, extent_wait_cb(ext, state)); } - if (rc == 0 && ext->oe_rc < 0) + if (ext->oe_rc < 0) rc = ext->oe_rc; + else + rc = 0; RETURN(rc); } @@ -934,20 +977,21 @@ static int osc_extent_wait(const struct lu_env *env, struct osc_extent *ext, static int osc_extent_truncate(struct osc_extent *ext, pgoff_t trunc_index, bool partial) { - struct cl_env_nest nest; struct lu_env *env; struct cl_io *io; struct osc_object *obj = ext->oe_obj; struct client_obd *cli = osc_cli(obj); struct osc_async_page *oap; struct osc_async_page *tmp; + struct pagevec *pvec; int pages_in_chunk = 0; int ppc_bits = cli->cl_chunkbits - - PAGE_CACHE_SHIFT; + PAGE_SHIFT; __u64 trunc_chunk = trunc_index >> ppc_bits; int grants = 0; int nr_pages = 0; int rc = 0; + __u16 refcheck; ENTRY; LASSERT(sanity_check(ext) == 0); @@ -957,14 +1001,20 @@ static int osc_extent_truncate(struct osc_extent *ext, pgoff_t trunc_index, /* Request new lu_env. * We can't use that env from osc_cache_truncate_start() because * it's from lov_io_sub and not fully initialized. */ - env = cl_env_nested_get(&nest); - io = &osc_env_info(env)->oti_io; + env = cl_env_get(&refcheck); + if (IS_ERR(env)) + RETURN(PTR_ERR(env)); + + io = osc_env_thread_io(env); io->ci_obj = cl_object_top(osc2cl(obj)); + io->ci_ignore_layout = 1; + pvec = &osc_env_info(env)->oti_pagevec; + ll_pagevec_init(pvec, 0); rc = cl_io_init(env, io, CIT_MISC, io->ci_obj); if (rc < 0) GOTO(out, rc); - /* discard all pages with index greater then trunc_index */ + /* discard all pages with index greater than trunc_index */ list_for_each_entry_safe(oap, tmp, &ext->oe_pages, oap_pending_item) { pgoff_t index = osc_index(oap2osc(oap)); @@ -997,11 +1047,13 @@ static int osc_extent_truncate(struct osc_extent *ext, pgoff_t trunc_index, } lu_ref_del(&page->cp_reference, "truncate", current); - cl_page_put(env, page); + cl_pagevec_put(env, page, pvec); --ext->oe_nr_pages; ++nr_pages; } + pagevec_release(pvec); + EASSERTF(ergo(ext->oe_start >= trunc_index + !!partial, ext->oe_nr_pages == 0), ext, "trunc_index %lu, partial %d\n", trunc_index, partial); @@ -1038,11 +1090,11 @@ static int osc_extent_truncate(struct osc_extent *ext, pgoff_t trunc_index, osc_object_unlock(obj); if (grants > 0 || nr_pages > 0) - osc_free_grant(cli, nr_pages, grants); + osc_free_grant(cli, nr_pages, grants, grants); out: cl_io_fini(env, io); - cl_env_nested_put(&nest, env); + cl_env_put(env, &refcheck); RETURN(rc); } @@ -1099,7 +1151,7 @@ static int osc_extent_make_ready(const struct lu_env *env, if (!(last->oap_async_flags & ASYNC_COUNT_STABLE)) { 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); + LASSERT(last->oap_page_off + last_oap_count <= PAGE_SIZE); last->oap_count = last_oap_count; spin_lock(&last->oap_lock); last->oap_async_flags |= ASYNC_COUNT_STABLE; @@ -1110,7 +1162,7 @@ static int osc_extent_make_ready(const struct lu_env *env, * because it's known they are not the last page */ list_for_each_entry(oap, &ext->oe_pages, oap_pending_item) { if (!(oap->oap_async_flags & ASYNC_COUNT_STABLE)) { - oap->oap_count = PAGE_CACHE_SIZE - oap->oap_page_off; + oap->oap_count = PAGE_SIZE - oap->oap_page_off; spin_lock(&oap->oap_lock); oap->oap_async_flags |= ASYNC_COUNT_STABLE; spin_unlock(&oap->oap_lock); @@ -1137,7 +1189,7 @@ static int osc_extent_expand(struct osc_extent *ext, pgoff_t index, struct osc_object *obj = ext->oe_obj; struct client_obd *cli = osc_cli(obj); struct osc_extent *next; - int ppc_bits = cli->cl_chunkbits - PAGE_CACHE_SHIFT; + int ppc_bits = cli->cl_chunkbits - PAGE_SHIFT; pgoff_t chunk = index >> ppc_bits; pgoff_t end_chunk; pgoff_t end_index; @@ -1156,9 +1208,14 @@ static int osc_extent_expand(struct osc_extent *ext, pgoff_t index, GOTO(out, rc = 0); LASSERT(end_chunk + 1 == chunk); + /* try to expand this extent to cover @index */ end_index = min(ext->oe_max_end, ((chunk + 1) << ppc_bits) - 1); + /* don't go over the maximum extent size reported by server */ + if (end_index - ext->oe_start + 1 > cli->cl_max_extent_pages) + GOTO(out, rc = -ERANGE); + next = next_extent(ext); if (next != NULL && next->oe_start <= end_index) /* complex mode - overlapped with the next extent, @@ -1167,8 +1224,8 @@ static int osc_extent_expand(struct osc_extent *ext, pgoff_t index, ext->oe_end = end_index; ext->oe_grants += chunksize; + LASSERT(*grants >= chunksize); *grants -= chunksize; - LASSERT(*grants >= 0); EASSERTF(osc_extent_is_overlapped(obj, ext) == 0, ext, "overlapped after expanding for %lu.\n", index); EXIT; @@ -1178,31 +1235,34 @@ out: RETURN(rc); } -static void osc_extent_tree_dump0(int level, struct osc_object *obj, +static void osc_extent_tree_dump0(int mask, struct osc_object *obj, const char *func, int line) { struct osc_extent *ext; int cnt; - CDEBUG(level, "Dump object %p extents at %s:%d, mppr: %u.\n", + if (!cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) + return; + + CDEBUG(mask, "Dump object %p extents at %s:%d, mppr: %u.\n", obj, func, line, osc_cli(obj)->cl_max_pages_per_rpc); /* osc_object_lock(obj); */ cnt = 1; for (ext = first_extent(obj); ext != NULL; ext = next_extent(ext)) - OSC_EXTENT_DUMP(level, ext, "in tree %d.\n", cnt++); + OSC_EXTENT_DUMP(mask, ext, "in tree %d.\n", cnt++); cnt = 1; list_for_each_entry(ext, &obj->oo_hp_exts, oe_link) - OSC_EXTENT_DUMP(level, ext, "hp %d.\n", cnt++); + OSC_EXTENT_DUMP(mask, ext, "hp %d.\n", cnt++); cnt = 1; list_for_each_entry(ext, &obj->oo_urgent_exts, oe_link) - OSC_EXTENT_DUMP(level, ext, "urgent %d.\n", cnt++); + OSC_EXTENT_DUMP(mask, ext, "urgent %d.\n", cnt++); cnt = 1; list_for_each_entry(ext, &obj->oo_reading_exts, oe_link) - OSC_EXTENT_DUMP(level, ext, "reading %d.\n", cnt++); + OSC_EXTENT_DUMP(mask, ext, "reading %d.\n", cnt++); /* osc_object_unlock(obj); */ } @@ -1238,7 +1298,7 @@ static int osc_make_ready(const struct lu_env *env, struct osc_async_page *oap, ENTRY; result = cl_page_make_ready(env, page, CRT_WRITE); if (result == 0) - opg->ops_submit_time = cfs_time_current(); + opg->ops_submit_time = ktime_get(); RETURN(result); } @@ -1249,7 +1309,6 @@ static int osc_refresh_count(const struct lu_env *env, pgoff_t index = osc_index(oap2osc(oap)); struct cl_object *obj; struct cl_attr *attr = &osc_env_info(env)->oti_attr; - int result; loff_t kms; @@ -1269,9 +1328,9 @@ static int osc_refresh_count(const struct lu_env *env, return 0; else if (cl_offset(obj, index + 1) > kms) /* catch sub-page write at end of file */ - return kms % PAGE_CACHE_SIZE; + return kms & ~PAGE_MASK; else - return PAGE_CACHE_SIZE; + return PAGE_SIZE; } static int osc_completion(const struct lu_env *env, struct osc_async_page *oap, @@ -1279,7 +1338,6 @@ static int osc_completion(const struct lu_env *env, struct osc_async_page *oap, { struct osc_page *opg = oap2osc_page(oap); struct cl_page *page = oap2cl_page(oap); - struct osc_object *obj = cl2osc(opg->ops_cl.cpl_obj); enum cl_req_type crt; int srvlock; @@ -1292,26 +1350,11 @@ static int osc_completion(const struct lu_env *env, struct osc_async_page *oap, "cp_state:%u, cmd:%d\n", page->cp_state, cmd); LASSERT(opg->ops_transfer_pinned); - /* - * page->cp_req can be NULL if io submission failed before - * cl_req was allocated. - */ - if (page->cp_req != NULL) - cl_req_page_done(env, page); - LASSERT(page->cp_req == NULL); - crt = cmd == OBD_BRW_READ ? CRT_READ : CRT_WRITE; /* Clear opg->ops_transfer_pinned before VM lock is released. */ opg->ops_transfer_pinned = 0; - spin_lock(&obj->oo_seatbelt); - LASSERT(opg->ops_submitter != NULL); - LASSERT(!list_empty(&opg->ops_inflight)); - list_del_init(&opg->ops_inflight); - opg->ops_submitter = NULL; - spin_unlock(&obj->oo_seatbelt); - - opg->ops_submit_time = 0; + opg->ops_submit_time = ktime_set(0, 0); srvlock = oap->oap_brw_flags & OBD_BRW_SRVLOCK; /* statistic */ @@ -1334,19 +1377,22 @@ static int osc_completion(const struct lu_env *env, struct osc_async_page *oap, lu_ref_del(&page->cp_reference, "transfer", page); cl_page_completion(env, page, crt, rc); + cl_page_put(env, page); RETURN(0); } -#define OSC_DUMP_GRANT(lvl, cli, fmt, args...) do { \ +#define OSC_DUMP_GRANT(mask, cli, fmt, args...) do { \ struct client_obd *__tmp = (cli); \ - 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, \ + CDEBUG(mask, "%s: grant { dirty: %ld/%ld dirty_pages: %ld/%lu " \ + "dropped: %ld avail: %ld, dirty_grant: %ld, " \ + "reserved: %ld, flight: %d } lru {in list: %ld, " \ + "left: %ld, waiters: %d }" fmt "\n", \ + cli_name(__tmp), \ __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_dirty_grant, \ __tmp->cl_reserved_grant, __tmp->cl_w_in_flight, \ atomic_long_read(&__tmp->cl_lru_in_list), \ atomic_long_read(&__tmp->cl_lru_busy), \ @@ -1359,11 +1405,10 @@ static void osc_consume_write_grant(struct client_obd *cli, { assert_spin_locked(&cli->cl_loi_list_lock); LASSERT(!(pga->flag & OBD_BRW_FROM_GRANT)); - 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); + PAGE_SIZE, pga, pga->pg); osc_update_next_shrink(cli); } @@ -1383,11 +1428,6 @@ static void osc_release_write_grant(struct client_obd *cli, pga->flag &= ~OBD_BRW_FROM_GRANT; atomic_long_dec(&obd_dirty_pages); cli->cl_dirty_pages--; - if (pga->flag & OBD_BRW_NOCACHE) { - pga->flag &= ~OBD_BRW_NOCACHE; - atomic_long_dec(&obd_dirty_transit_pages); - cli->cl_dirty_transit--; - } EXIT; } @@ -1420,18 +1460,27 @@ static void __osc_unreserve_grant(struct client_obd *cli, if (unused > reserved) { cli->cl_avail_grant += reserved; cli->cl_lost_grant += unused - reserved; + cli->cl_dirty_grant -= unused - reserved; } else { cli->cl_avail_grant += unused; + cli->cl_dirty_grant += reserved - unused; } } -static void osc_unreserve_grant(struct client_obd *cli, - unsigned int reserved, unsigned int unused) +static void osc_unreserve_grant_nolock(struct client_obd *cli, + unsigned int reserved, + unsigned int unused) { - spin_lock(&cli->cl_loi_list_lock); __osc_unreserve_grant(cli, reserved, unused); if (unused > 0) osc_wake_cache_waiters(cli); +} + +static void osc_unreserve_grant(struct client_obd *cli, + unsigned int reserved, unsigned int unused) +{ + spin_lock(&cli->cl_loi_list_lock); + osc_unreserve_grant_nolock(cli, reserved, unused); spin_unlock(&cli->cl_loi_list_lock); } @@ -1442,21 +1491,24 @@ static void osc_unreserve_grant(struct client_obd *cli, * 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 PAGE_CACHE_SIZE and a partial page was + * 2. blocksize at OST is less than PAGE_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. * See filter_grant_check() for details. */ static void osc_free_grant(struct client_obd *cli, unsigned int nr_pages, - unsigned int lost_grant) + unsigned int lost_grant, unsigned int dirty_grant) { - unsigned long grant = (1 << cli->cl_chunkbits) + cli->cl_extent_tax; + unsigned long grant; + + grant = (1 << cli->cl_chunkbits) + cli->cl_grant_extent_tax; 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; + cli->cl_dirty_grant -= dirty_grant; if (cli->cl_avail_grant < grant && cli->cl_lost_grant >= grant) { /* borrow some grant from truncate to avoid the case that * truncate uses up all avail grant */ @@ -1465,9 +1517,10 @@ static void osc_free_grant(struct client_obd *cli, unsigned int nr_pages, } osc_wake_cache_waiters(cli); spin_unlock(&cli->cl_loi_list_lock); - CDEBUG(D_CACHE, "lost %u grant: %lu avail: %lu dirty: %lu\n", + CDEBUG(D_CACHE, "lost %u grant: %lu avail: %lu dirty: %lu/%lu\n", lost_grant, cli->cl_lost_grant, - cli->cl_avail_grant, cli->cl_dirty_pages << PAGE_CACHE_SHIFT); + cli->cl_avail_grant, cli->cl_dirty_pages << PAGE_SHIFT, + cli->cl_dirty_grant); } /** @@ -1487,7 +1540,7 @@ static void osc_exit_cache(struct client_obd *cli, struct osc_async_page *oap) */ static int osc_enter_cache_try(struct client_obd *cli, struct osc_async_page *oap, - int bytes, int transient) + int bytes) { int rc; @@ -1497,19 +1550,18 @@ static int osc_enter_cache_try(struct client_obd *cli, if (rc < 0) return 0; - 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++; - atomic_long_inc(&obd_dirty_transit_pages); - oap->oap_brw_flags |= OBD_BRW_NOCACHE; - } - rc = 1; - } else { - __osc_unreserve_grant(cli, bytes, bytes); - rc = 0; + if (cli->cl_dirty_pages < cli->cl_dirty_max_pages) { + if (atomic_long_add_return(1, &obd_dirty_pages) <= + obd_max_dirty_pages) { + osc_consume_write_grant(cli, &oap->oap_brw_page); + rc = 1; + goto out; + } else + atomic_long_dec(&obd_dirty_pages); } + __osc_unreserve_grant(cli, bytes, bytes); + +out: return rc; } @@ -1535,13 +1587,9 @@ static int osc_enter_cache(const struct lu_env *env, struct client_obd *cli, 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; - lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(AT_OFF ? obd_timeout : at_max), - NULL, LWI_ON_SIGNAL_NOOP, NULL); - OSC_DUMP_GRANT(D_CACHE, cli, "need:%d\n", bytes); spin_lock(&cli->cl_loi_list_lock); @@ -1556,7 +1604,8 @@ static int osc_enter_cache(const struct lu_env *env, struct client_obd *cli, } /* Hopefully normal case - cache space and write credits available */ - if (osc_enter_cache_try(cli, oap, bytes, 0)) { + if (list_empty(&cli->cl_cache_waiters) && + osc_enter_cache_try(cli, oap, bytes)) { OSC_DUMP_GRANT(D_CACHE, cli, "granted from cache\n"); GOTO(out, rc = 0); } @@ -1578,15 +1627,21 @@ static int osc_enter_cache(const struct lu_env *env, struct client_obd *cli, osc_io_unplug_async(env, cli, NULL); CDEBUG(D_CACHE, "%s: sleeping for cache space @ %p for %p\n", - cli->cl_import->imp_obd->obd_name, &ocw, oap); + cli_name(cli), &ocw, oap); - rc = l_wait_event(ocw.ocw_waitq, ocw_granted(cli, &ocw), &lwi); + rc = wait_event_idle_timeout(ocw.ocw_waitq, + ocw_granted(cli, &ocw), + cfs_time_seconds(AT_OFF ? + obd_timeout : + at_max)); spin_lock(&cli->cl_loi_list_lock); - if (rc < 0) { + if (rc <= 0) { /* l_wait_event is interrupted by signal or timed out */ list_del_init(&ocw.ocw_entry); + if (rc == 0) + rc = -ETIMEDOUT; break; } LASSERT(list_empty(&ocw.ocw_entry)); @@ -1594,7 +1649,7 @@ static int osc_enter_cache(const struct lu_env *env, struct client_obd *cli, if (rc != -EDQUOT) break; - if (osc_enter_cache_try(cli, oap, bytes, 0)) { + if (osc_enter_cache_try(cli, oap, bytes)) { rc = 0; break; } @@ -1623,7 +1678,7 @@ static int osc_enter_cache(const struct lu_env *env, struct client_obd *cli, 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); + cli_name(cli), &ocw, rc); break; } EXIT; @@ -1641,31 +1696,26 @@ void osc_wake_cache_waiters(struct client_obd *cli) ENTRY; list_for_each_safe(l, tmp, &cli->cl_cache_waiters) { ocw = list_entry(l, struct osc_cache_waiter, ocw_entry); - list_del_init(&ocw->ocw_entry); ocw->ocw_rc = -EDQUOT; - /* we can't dirty more */ - 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 %ld\n", - cli->cl_dirty_pages, cli->cl_dirty_max_pages, - obd_max_dirty_pages); - goto wakeup; - } - if (osc_enter_cache_try(cli, ocw->ocw_oap, ocw->ocw_grant, 0)) + if (osc_enter_cache_try(cli, ocw->ocw_oap, ocw->ocw_grant)) 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); - wake_up(&ocw->ocw_waitq); + if (ocw->ocw_rc == 0 || + !(cli->cl_dirty_pages > 0 || cli->cl_w_in_flight > 0)) { + list_del_init(&ocw->ocw_entry); + 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); + + wake_up(&ocw->ocw_waitq); + } } EXIT; } +EXPORT_SYMBOL(osc_wake_cache_waiters); static int osc_max_rpc_in_flight(struct client_obd *cli, struct osc_object *osc) { @@ -1711,9 +1761,10 @@ static int osc_makes_rpc(struct client_obd *cli, struct osc_object *osc, CDEBUG(D_CACHE, "cache waiters forcing RPC\n"); RETURN(1); } - if (atomic_read(&osc->oo_nr_writes) >= - cli->cl_max_pages_per_rpc) + if (!list_empty(&osc->oo_full_exts)) { + CDEBUG(D_CACHE, "full extent ready, make an RPC\n"); RETURN(1); + } } else { if (atomic_read(&osc->oo_nr_reads) == 0) RETURN(0); @@ -1852,6 +1903,47 @@ static void osc_ap_completion(const struct lu_env *env, struct client_obd *cli, EXIT; } +struct extent_rpc_data { + struct list_head *erd_rpc_list; + unsigned int erd_page_count; + unsigned int erd_max_pages; + unsigned int erd_max_chunks; + unsigned int erd_max_extents; +}; + +static inline unsigned osc_extent_chunks(const struct osc_extent *ext) +{ + struct client_obd *cli = osc_cli(ext->oe_obj); + unsigned ppc_bits = cli->cl_chunkbits - PAGE_SHIFT; + + return (ext->oe_end >> ppc_bits) - (ext->oe_start >> ppc_bits) + 1; +} + +static inline bool +can_merge(const struct osc_extent *ext, const struct osc_extent *in_rpc) +{ + if (ext->oe_no_merge || in_rpc->oe_no_merge) + return false; + + if (ext->oe_srvlock != in_rpc->oe_srvlock) + return false; + + if (ext->oe_ndelay != in_rpc->oe_ndelay) + return false; + + if (!ext->oe_grants != !in_rpc->oe_grants) + return false; + + if (ext->oe_dio != in_rpc->oe_dio) + return false; + + /* It's possible to have overlap on DIO */ + if (in_rpc->oe_dio && overlapped(ext, in_rpc)) + return false; + + return true; +} + /** * Try to add extent to one RPC. We need to think about the following things: * - # of pages must not be over max_pages_per_rpc @@ -1859,49 +1951,51 @@ static void osc_ap_completion(const struct lu_env *env, struct client_obd *cli, */ static int try_to_add_extent_for_io(struct client_obd *cli, struct osc_extent *ext, - struct list_head *rpclist, - unsigned int *pc, unsigned int *max_pages) + struct extent_rpc_data *data) { struct osc_extent *tmp; - struct osc_async_page *oap = list_first_entry(&ext->oe_pages, - struct osc_async_page, - oap_pending_item); + unsigned int chunk_count; ENTRY; EASSERT((ext->oe_state == OES_CACHE || ext->oe_state == OES_LOCK_DONE), ext); + OSC_EXTENT_DUMP(D_CACHE, ext, "trying to add this extent\n"); + + if (data->erd_max_extents == 0) + RETURN(0); + + chunk_count = osc_extent_chunks(ext); + EASSERTF(data->erd_page_count != 0 || + chunk_count <= data->erd_max_chunks, ext, + "The first extent to be fit in a RPC contains %u chunks, " + "which is over the limit %u.\n", chunk_count, + data->erd_max_chunks); + if (chunk_count > data->erd_max_chunks) + RETURN(0); - *max_pages = max(ext->oe_mppr, *max_pages); - if (*pc + ext->oe_nr_pages > *max_pages) + data->erd_max_pages = max(ext->oe_mppr, data->erd_max_pages); + EASSERTF(data->erd_page_count != 0 || + ext->oe_nr_pages <= data->erd_max_pages, ext, + "The first extent to be fit in a RPC contains %u pages, " + "which is over the limit %u.\n", ext->oe_nr_pages, + data->erd_max_pages); + if (data->erd_page_count + ext->oe_nr_pages > data->erd_max_pages) 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); + list_for_each_entry(tmp, data->erd_rpc_list, oe_link) { EASSERT(tmp->oe_owner == current, tmp); -#if 0 - if (overlapped(tmp, ext)) { - OSC_EXTENT_DUMP(D_ERROR, tmp, "overlapped %p.\n", 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) + if (!can_merge(ext, tmp)) RETURN(0); /* remove break for strict check */ break; } - *pc += ext->oe_nr_pages; - list_move_tail(&ext->oe_link, rpclist); + data->erd_max_extents--; + data->erd_max_chunks -= chunk_count; + data->erd_page_count += ext->oe_nr_pages; + list_move_tail(&ext->oe_link, data->erd_rpc_list); ext->oe_owner = current; RETURN(1); } @@ -1924,45 +2018,48 @@ static unsigned int get_write_extents(struct osc_object *obj, { struct client_obd *cli = osc_cli(obj); struct osc_extent *ext; - unsigned int page_count = 0; - unsigned int max_pages = cli->cl_max_pages_per_rpc; + struct extent_rpc_data data = { + .erd_rpc_list = rpclist, + .erd_page_count = 0, + .erd_max_pages = cli->cl_max_pages_per_rpc, + .erd_max_chunks = osc_max_write_chunks(cli), + .erd_max_extents = 256, + }; LASSERT(osc_object_is_locked(obj)); while (!list_empty(&obj->oo_hp_exts)) { ext = list_entry(obj->oo_hp_exts.next, struct osc_extent, oe_link); LASSERT(ext->oe_state == OES_CACHE); - if (!try_to_add_extent_for_io(cli, ext, rpclist, &page_count, - &max_pages)) - return page_count; - EASSERT(ext->oe_nr_pages <= max_pages, ext); + if (!try_to_add_extent_for_io(cli, ext, &data)) + return data.erd_page_count; + EASSERT(ext->oe_nr_pages <= data.erd_max_pages, ext); } - if (page_count == max_pages) - return page_count; + if (data.erd_page_count == data.erd_max_pages) + return data.erd_page_count; while (!list_empty(&obj->oo_urgent_exts)) { ext = list_entry(obj->oo_urgent_exts.next, struct osc_extent, oe_link); - if (!try_to_add_extent_for_io(cli, ext, rpclist, &page_count, - &max_pages)) - return page_count; - - if (!ext->oe_intree) - continue; - - while ((ext = next_extent(ext)) != NULL) { - if ((ext->oe_state != OES_CACHE) || - (!list_empty(&ext->oe_link) && - ext->oe_owner != NULL)) - continue; - - if (!try_to_add_extent_for_io(cli, ext, rpclist, - &page_count, &max_pages)) - return page_count; - } + if (!try_to_add_extent_for_io(cli, ext, &data)) + return data.erd_page_count; + } + if (data.erd_page_count == data.erd_max_pages) + return data.erd_page_count; + + /* One key difference between full extents and other extents: full + * extents can usually only be added if the rpclist was empty, so if we + * can't add one, we continue on to trying to add normal extents. This + * is so we don't miss adding extra extents to an RPC containing high + * priority or urgent extents. */ + while (!list_empty(&obj->oo_full_exts)) { + ext = list_entry(obj->oo_full_exts.next, + struct osc_extent, oe_link); + if (!try_to_add_extent_for_io(cli, ext, &data)) + break; } - if (page_count == max_pages) - return page_count; + if (data.erd_page_count == data.erd_max_pages) + return data.erd_page_count; ext = first_extent(obj); while (ext != NULL) { @@ -1973,21 +2070,20 @@ static unsigned int get_write_extents(struct osc_object *obj, continue; } - if (!try_to_add_extent_for_io(cli, ext, rpclist, &page_count, - &max_pages)) - return page_count; + if (!try_to_add_extent_for_io(cli, ext, &data)) + return data.erd_page_count; ext = next_extent(ext); } - return page_count; + return data.erd_page_count; } static int osc_send_write_rpc(const struct lu_env *env, struct client_obd *cli, - struct osc_object *osc, pdl_policy_t pol) + struct osc_object *osc) __must_hold(osc) { - struct list_head rpclist = LIST_HEAD_INIT(rpclist); + LIST_HEAD(rpclist); struct osc_extent *ext; struct osc_extent *tmp; struct osc_extent *first = NULL; @@ -2038,7 +2134,7 @@ __must_hold(osc) if (!list_empty(&rpclist)) { LASSERT(page_count > 0); - rc = osc_build_rpc(env, cli, &rpclist, OBD_BRW_WRITE, pol); + rc = osc_build_rpc(env, cli, &rpclist, OBD_BRW_WRITE); LASSERT(list_empty(&rpclist)); } @@ -2058,36 +2154,38 @@ __must_hold(osc) */ static int osc_send_read_rpc(const struct lu_env *env, struct client_obd *cli, - struct osc_object *osc, pdl_policy_t pol) + struct osc_object *osc) __must_hold(osc) { struct osc_extent *ext; struct osc_extent *next; - struct list_head rpclist = LIST_HEAD_INIT(rpclist); - unsigned int page_count = 0; - unsigned int max_pages = cli->cl_max_pages_per_rpc; + LIST_HEAD(rpclist); + struct extent_rpc_data data = { + .erd_rpc_list = &rpclist, + .erd_page_count = 0, + .erd_max_pages = cli->cl_max_pages_per_rpc, + .erd_max_chunks = UINT_MAX, + .erd_max_extents = UINT_MAX, + }; int rc = 0; ENTRY; LASSERT(osc_object_is_locked(osc)); - list_for_each_entry_safe(ext, next, - &osc->oo_reading_exts, oe_link) { + list_for_each_entry_safe(ext, next, &osc->oo_reading_exts, oe_link) { EASSERT(ext->oe_state == OES_LOCK_DONE, ext); - if (!try_to_add_extent_for_io(cli, ext, &rpclist, &page_count, - &max_pages)) + if (!try_to_add_extent_for_io(cli, ext, &data)) break; osc_extent_state_set(ext, OES_RPC); - EASSERT(ext->oe_nr_pages <= max_pages, ext); + EASSERT(ext->oe_nr_pages <= data.erd_max_pages, ext); } - LASSERT(page_count <= max_pages); + LASSERT(data.erd_page_count <= data.erd_max_pages); - osc_update_pending(osc, OBD_BRW_READ, -page_count); + osc_update_pending(osc, OBD_BRW_READ, -data.erd_page_count); if (!list_empty(&rpclist)) { osc_object_unlock(osc); - LASSERT(page_count > 0); - rc = osc_build_rpc(env, cli, &rpclist, OBD_BRW_READ, pol); + rc = osc_build_rpc(env, cli, &rpclist, OBD_BRW_READ); LASSERT(list_empty(&rpclist)); osc_object_lock(osc); @@ -2137,8 +2235,7 @@ static struct osc_object *osc_next_obj(struct client_obd *cli) } /* 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) +static void osc_check_rpcs(const struct lu_env *env, struct client_obd *cli) __must_hold(&cli->cl_loi_list_lock) { struct osc_object *osc; @@ -2168,7 +2265,7 @@ __must_hold(&cli->cl_loi_list_lock) * do io on writes while there are cache waiters */ osc_object_lock(osc); if (osc_makes_rpc(cli, osc, OBD_BRW_WRITE)) { - rc = osc_send_write_rpc(env, cli, osc, pol); + rc = osc_send_write_rpc(env, cli, osc); if (rc < 0) { CERROR("Write request failed with %d\n", rc); @@ -2192,7 +2289,7 @@ __must_hold(&cli->cl_loi_list_lock) } } if (osc_makes_rpc(cli, osc, OBD_BRW_READ)) { - rc = osc_send_read_rpc(env, cli, osc, pol); + rc = osc_send_read_rpc(env, cli, osc); if (rc < 0) CERROR("Read request failed with %d\n", rc); } @@ -2206,8 +2303,8 @@ __must_hold(&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 osc_io_unplug0(const struct lu_env *env, struct client_obd *cli, + struct osc_object *osc, int async) { int rc = 0; @@ -2215,13 +2312,9 @@ static int osc_io_unplug0(const struct lu_env *env, struct client_obd *cli, return 0; if (!async) { - /* disable osc_lru_shrink() temporarily to avoid - * potential stack overrun problem. LU-2859 */ - atomic_inc(&cli->cl_lru_shrinkers); spin_lock(&cli->cl_loi_list_lock); - osc_check_rpcs(env, cli, pol); + osc_check_rpcs(env, cli); 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); LASSERT(cli->cl_writeback_work != NULL); @@ -2229,19 +2322,7 @@ static int osc_io_unplug0(const struct lu_env *env, struct client_obd *cli, } return rc; } - -static int osc_io_unplug_async(const struct lu_env *env, - struct client_obd *cli, struct osc_object *osc) -{ - /* XXX: policy is no use actually. */ - return osc_io_unplug0(env, cli, osc, PDL_POLICY_ROUND, 1); -} - -void osc_io_unplug(const struct lu_env *env, struct client_obd *cli, - struct osc_object *osc, pdl_policy_t pol) -{ - (void)osc_io_unplug0(env, cli, osc, pol, 0); -} +EXPORT_SYMBOL(osc_io_unplug0); int osc_prep_async_page(struct osc_object *osc, struct osc_page *ops, struct page *page, loff_t offset) @@ -2259,28 +2340,27 @@ int osc_prep_async_page(struct osc_object *osc, struct osc_page *ops, oap->oap_page = page; oap->oap_obj_off = offset; - LASSERT(!(offset & ~CFS_PAGE_MASK)); - - if (!client_is_remote(exp) && cfs_capable(CFS_CAP_SYS_RESOURCE)) - oap->oap_brw_flags = OBD_BRW_NOQUOTA; + LASSERT(!(offset & ~PAGE_MASK)); INIT_LIST_HEAD(&oap->oap_pending_item); INIT_LIST_HEAD(&oap->oap_rpc_item); spin_lock_init(&oap->oap_lock); - CDEBUG(D_INFO, "oap %p page %p obj off "LPU64"\n", + CDEBUG(D_INFO, "oap %p page %p obj off %llu\n", oap, page, oap->oap_obj_off); RETURN(0); } +EXPORT_SYMBOL(osc_prep_async_page); int osc_queue_async_io(const struct lu_env *env, struct cl_io *io, - struct osc_page *ops) + struct osc_page *ops, cl_commit_cbt cb) { struct osc_io *oio = osc_env_io(env); struct osc_extent *ext = NULL; struct osc_async_page *oap = &ops->ops_oap; struct client_obd *cli = oap->oap_cli; struct osc_object *osc = oap->oap_obj; + struct pagevec *pvec = &osc_env_info(env)->oti_pagevec; pgoff_t index; unsigned int tmp; unsigned int grants = 0; @@ -2302,8 +2382,7 @@ int osc_queue_async_io(const struct lu_env *env, struct cl_io *io, /* Set the OBD_BRW_SRVLOCK before the page is queued. */ brw_flags |= ops->ops_srvlock ? OBD_BRW_SRVLOCK : 0; - if (!client_is_remote(osc_export(osc)) && - cfs_capable(CFS_CAP_SYS_RESOURCE)) { + if (oio->oi_cap_sys_resource) { brw_flags |= OBD_BRW_NOQUOTA; cmd |= OBD_BRW_NOQUOTA; } @@ -2312,7 +2391,7 @@ int osc_queue_async_io(const struct lu_env *env, struct cl_io *io, if (!(cmd & OBD_BRW_NOQUOTA)) { struct cl_object *obj; struct cl_attr *attr; - unsigned int qid[MAXQUOTAS]; + unsigned int qid[LL_MAXQUOTAS]; obj = cl_object_top(&osc->oo_cl); attr = &osc_env_info(env)->oti_attr; @@ -2323,6 +2402,7 @@ int osc_queue_async_io(const struct lu_env *env, struct cl_io *io, qid[USRQUOTA] = attr->cat_uid; qid[GRPQUOTA] = attr->cat_gid; + qid[PRJQUOTA] = attr->cat_projid; if (rc == 0 && osc_quota_chkdq(cli, qid) == NO_QUOTA) rc = -EDQUOT; if (rc) @@ -2352,14 +2432,13 @@ int osc_queue_async_io(const struct lu_env *env, struct cl_io *io, if (ext != NULL && ext->oe_start <= index && ext->oe_max_end >= index) { /* one chunk plus extent overhead must be enough to write this * page */ - grants = (1 << cli->cl_chunkbits) + cli->cl_extent_tax; + grants = (1 << cli->cl_chunkbits) + cli->cl_grant_extent_tax; if (ext->oe_end >= index) grants = 0; /* it doesn't need any grant to dirty this page */ spin_lock(&cli->cl_loi_list_lock); - rc = osc_enter_cache_try(cli, oap, grants, 0); - spin_unlock(&cli->cl_loi_list_lock); + rc = osc_enter_cache_try(cli, oap, grants); if (rc == 0) { /* try failed */ grants = 0; need_release = 1; @@ -2373,10 +2452,11 @@ int osc_queue_async_io(const struct lu_env *env, struct cl_io *io, } else { OSC_EXTENT_DUMP(D_CACHE, ext, "expanded for %lu.\n", index); - osc_unreserve_grant(cli, grants, tmp); + osc_unreserve_grant_nolock(cli, grants, tmp); grants = 0; } } + spin_unlock(&cli->cl_loi_list_lock); rc = 0; } else if (ext != NULL) { /* index is located outside of active extent */ @@ -2389,7 +2469,7 @@ int osc_queue_async_io(const struct lu_env *env, struct cl_io *io, } if (ext == NULL) { - tmp = (1 << cli->cl_chunkbits) + cli->cl_extent_tax; + tmp = (1 << cli->cl_chunkbits) + cli->cl_grant_extent_tax; /* try to find new extent to cover this page */ LASSERT(oio->oi_active == NULL); @@ -2399,7 +2479,14 @@ int osc_queue_async_io(const struct lu_env *env, struct cl_io *io, rc = 0; if (grants == 0) { - /* we haven't allocated grant for this page. */ + /* We haven't allocated grant for this page, and we + * must not hold a page lock while we do enter_cache, + * so we must mark dirty & unlock any pages in the + * write commit pagevec. */ + if (pagevec_count(pvec)) { + cb(env, io, pvec); + pagevec_reinit(pvec); + } rc = osc_enter_cache(env, cli, oap, tmp); if (rc == 0) grants = tmp; @@ -2435,7 +2522,11 @@ int osc_queue_async_io(const struct lu_env *env, struct cl_io *io, ++ext->oe_nr_pages; list_add_tail(&oap->oap_pending_item, &ext->oe_pages); osc_object_unlock(osc); + + if (!ext->oe_layout_version) + ext->oe_layout_version = io->ci_layout_version; } + RETURN(rc); } @@ -2443,7 +2534,6 @@ int osc_teardown_async_page(const struct lu_env *env, struct osc_object *obj, struct osc_page *ops) { struct osc_async_page *oap = &ops->ops_oap; - struct osc_extent *ext = NULL; int rc = 0; ENTRY; @@ -2452,12 +2542,15 @@ int osc_teardown_async_page(const struct lu_env *env, CDEBUG(D_INFO, "teardown oap %p page %p at index %lu.\n", oap, ops, osc_index(oap2osc(oap))); - osc_object_lock(obj); if (!list_empty(&oap->oap_rpc_item)) { CDEBUG(D_CACHE, "oap %p is not in cache.\n", oap); rc = -EBUSY; } else if (!list_empty(&oap->oap_pending_item)) { + struct osc_extent *ext = NULL; + + osc_object_lock(obj); ext = osc_extent_lookup(obj, osc_index(oap2osc(oap))); + osc_object_unlock(obj); /* only truncated pages are allowed to be taken out. * See osc_extent_truncate() and osc_cache_truncate_start() * for details. */ @@ -2466,10 +2559,9 @@ int osc_teardown_async_page(const struct lu_env *env, osc_index(oap2osc(oap))); rc = -EBUSY; } + if (ext != NULL) + osc_extent_put(env, ext); } - osc_object_unlock(obj); - if (ext != NULL) - osc_extent_put(env, ext); RETURN(rc); } @@ -2620,26 +2712,33 @@ int osc_cancel_async_page(const struct lu_env *env, struct osc_page *ops) RETURN(rc); } -int osc_queue_sync_pages(const struct lu_env *env, struct osc_object *obj, - struct list_head *list, int cmd, int brw_flags) +int osc_queue_sync_pages(const struct lu_env *env, const struct cl_io *io, + struct osc_object *obj, struct list_head *list, + int brw_flags) { struct client_obd *cli = osc_cli(obj); struct osc_extent *ext; struct osc_async_page *oap; int page_count = 0; int mppr = cli->cl_max_pages_per_rpc; + bool can_merge = true; pgoff_t start = CL_PAGE_EOF; pgoff_t end = 0; ENTRY; list_for_each_entry(oap, list, oap_pending_item) { - pgoff_t index = osc_index(oap2osc(oap)); + struct osc_page *opg = oap2osc_page(oap); + pgoff_t index = osc_index(opg); + if (index > end) end = index; if (index < start) start = index; ++page_count; mppr <<= (page_count > mppr); + + if (unlikely(opg->ops_from > 0 || opg->ops_to < PAGE_SIZE)) + can_merge = false; } ext = osc_extent_alloc(obj); @@ -2653,21 +2752,25 @@ int osc_queue_sync_pages(const struct lu_env *env, struct osc_object *obj, RETURN(-ENOMEM); } - ext->oe_rw = !!(cmd & OBD_BRW_READ); + ext->oe_rw = !!(brw_flags & OBD_BRW_READ); ext->oe_sync = 1; + ext->oe_no_merge = !can_merge; ext->oe_urgent = 1; ext->oe_start = start; ext->oe_end = ext->oe_max_end = end; ext->oe_obj = obj; ext->oe_srvlock = !!(brw_flags & OBD_BRW_SRVLOCK); + ext->oe_ndelay = !!(brw_flags & OBD_BRW_NDELAY); + ext->oe_dio = !!(brw_flags & OBD_BRW_NOCACHE); ext->oe_nr_pages = page_count; ext->oe_mppr = mppr; list_splice_init(list, &ext->oe_pages); + ext->oe_layout_version = io->ci_layout_version; osc_object_lock(obj); /* Reuse the initial refcount for RPC, don't drop it */ osc_extent_state_set(ext, OES_LOCK_DONE); - if (cmd & OBD_BRW_WRITE) { + if (!ext->oe_rw) { /* write */ list_add_tail(&ext->oe_link, &obj->oo_urgent_exts); osc_update_pending(obj, OBD_BRW_WRITE, page_count); } else { @@ -2683,14 +2786,14 @@ int osc_queue_sync_pages(const struct lu_env *env, struct osc_object *obj, /** * Called by osc_io_setattr_start() to freeze and destroy covering extents. */ -int osc_cache_truncate_start(const struct lu_env *env, struct osc_io *oio, - struct osc_object *obj, __u64 size) +int osc_cache_truncate_start(const struct lu_env *env, struct osc_object *obj, + __u64 size, struct osc_extent **extp) { struct client_obd *cli = osc_cli(obj); struct osc_extent *ext; struct osc_extent *waiting = NULL; pgoff_t index; - struct list_head list = LIST_HEAD_INIT(list); + LIST_HEAD(list); int result = 0; bool partial; ENTRY; @@ -2714,14 +2817,13 @@ again: * a page already having been flushed by write_page(). * We have to wait for this extent because we can't * truncate that page. */ - LASSERT(!ext->oe_hp); OSC_EXTENT_DUMP(D_CACHE, ext, "waiting for busy extent\n"); waiting = osc_extent_get(ext); break; } - OSC_EXTENT_DUMP(D_CACHE, ext, "try to trunc:"LPU64".\n", size); + OSC_EXTENT_DUMP(D_CACHE, ext, "try to trunc:%llu.\n", size); osc_extent_get(ext); if (ext->oe_state == OES_ACTIVE) { @@ -2737,8 +2839,12 @@ again: osc_update_pending(obj, OBD_BRW_WRITE, -ext->oe_nr_pages); } - EASSERT(list_empty(&ext->oe_link), ext); - list_add_tail(&ext->oe_link, &list); + /* This extent could be on the full extents list, that's OK */ + EASSERT(!ext->oe_hp && !ext->oe_urgent, ext); + if (!list_empty(&ext->oe_link)) + list_move_tail(&ext->oe_link, &list); + else + list_add_tail(&ext->oe_link, &list); ext = next_extent(ext); } @@ -2778,11 +2884,13 @@ again: /* we need to hold this extent in OES_TRUNC state so * that no writeback will happen. This is to avoid - * BUG 17397. */ - LASSERT(oio->oi_trunc == NULL); - oio->oi_trunc = osc_extent_get(ext); + * BUG 17397. + * Only partial truncate can reach here, if @size is + * not zero, the caller should provide a valid @extp. */ + LASSERT(*extp == NULL); + *extp = osc_extent_get(ext); OSC_EXTENT_DUMP(D_CACHE, ext, - "trunc at "LPU64"\n", size); + "trunc at %llu\n", size); } osc_extent_put(env, ext); } @@ -2801,17 +2909,15 @@ again: } RETURN(result); } +EXPORT_SYMBOL(osc_cache_truncate_start); /** * Called after osc_io_setattr_end to add oio->oi_trunc back to cache. */ -void osc_cache_truncate_end(const struct lu_env *env, struct osc_io *oio, - struct osc_object *obj) +void osc_cache_truncate_end(const struct lu_env *env, struct osc_extent *ext) { - struct osc_extent *ext = oio->oi_trunc; - - oio->oi_trunc = NULL; if (ext != NULL) { + struct osc_object *obj = ext->oe_obj; bool unplug = false; EASSERT(ext->oe_nr_pages > 0, ext); @@ -2890,6 +2996,7 @@ again: OSC_IO_DEBUG(obj, "sync file range.\n"); RETURN(result); } +EXPORT_SYMBOL(osc_cache_wait_range); /** * Called to write out a range of osc object. @@ -2904,7 +3011,7 @@ int osc_cache_writeback_range(const struct lu_env *env, struct osc_object *obj, pgoff_t start, pgoff_t end, int hp, int discard) { struct osc_extent *ext; - struct list_head discard_list = LIST_HEAD_INIT(discard_list); + LIST_HEAD(discard_list); bool unplug = false; int result = 0; ENTRY; @@ -2937,10 +3044,25 @@ int osc_cache_writeback_range(const struct lu_env *env, struct osc_object *obj, list_move_tail(&ext->oe_link, list); unplug = true; } else { + struct client_obd *cli = osc_cli(obj); + int pcc_bits = cli->cl_chunkbits - PAGE_SHIFT; + pgoff_t align_by = (1 << pcc_bits); + pgoff_t a_start = round_down(start, align_by); + pgoff_t a_end = round_up(end, align_by); + + /* overflow case */ + if (end && !a_end) + a_end = CL_PAGE_EOF; /* the only discarder is lock cancelling, so - * [start, end] must contain this extent */ - EASSERT(ext->oe_start >= start && - ext->oe_max_end <= end, ext); + * [start, end], aligned by chunk size, must + * contain this extent */ + LASSERTF(ext->oe_start >= a_start && + ext->oe_end <= a_end, + "ext [%lu, %lu] reg [%lu, %lu] " + "orig [%lu %lu] align %lu bits " + "%d\n", ext->oe_start, ext->oe_end, + a_start, a_end, start, end, + align_by, pcc_bits); osc_extent_state_set(ext, OES_LOCKING); ext->oe_owner = current; list_move_tail(&ext->oe_link, @@ -2994,7 +3116,7 @@ int osc_cache_writeback_range(const struct lu_env *env, struct osc_object *obj, } if (unplug) - osc_io_unplug(env, osc_cli(obj), obj, PDL_POLICY_ROUND); + osc_io_unplug(env, osc_cli(obj), obj); if (hp || discard) { int rc; @@ -3006,6 +3128,7 @@ int osc_cache_writeback_range(const struct lu_env *env, struct osc_object *obj, OSC_IO_DEBUG(obj, "pageout [%lu, %lu], %d.\n", start, end, result); RETURN(result); } +EXPORT_SYMBOL(osc_cache_writeback_range); /** * Returns a list of pages by a given [start, end] of \a obj. @@ -3024,6 +3147,7 @@ int osc_page_gang_lookup(const struct lu_env *env, struct cl_io *io, osc_page_gang_cbt cb, void *cbdata) { struct osc_page *ops; + struct pagevec *pagevec; void **pvec; pgoff_t idx; unsigned int nr; @@ -3035,6 +3159,8 @@ int osc_page_gang_lookup(const struct lu_env *env, struct cl_io *io, idx = start; pvec = osc_env_info(env)->oti_pvec; + pagevec = &osc_env_info(env)->oti_pagevec; + ll_pagevec_init(pagevec, 0); spin_lock(&osc->oo_tree_lock); while ((nr = radix_tree_gang_lookup(&osc->oo_tree, pvec, idx, OTI_PVEC_SIZE)) > 0) { @@ -3081,8 +3207,10 @@ int osc_page_gang_lookup(const struct lu_env *env, struct cl_io *io, page = ops->ops_cl.cpl_page; lu_ref_del(&page->cp_reference, "gang_lookup", current); - cl_page_put(env, page); + cl_pagevec_put(env, page, pagevec); } + pagevec_release(pagevec); + if (nr < OTI_PVEC_SIZE || end_of_region) break; @@ -3098,6 +3226,7 @@ int osc_page_gang_lookup(const struct lu_env *env, struct cl_io *io, spin_unlock(&osc->oo_tree_lock); RETURN(res); } +EXPORT_SYMBOL(osc_page_gang_lookup); /** * Check if page @page is covered by an extra lock or discard it. @@ -3140,8 +3269,8 @@ static int check_and_discard_cb(const struct lu_env *env, struct cl_io *io, return CLP_GANG_OKAY; } -static int discard_cb(const struct lu_env *env, struct cl_io *io, - struct osc_page *ops, void *cbdata) +int osc_discard_cb(const struct lu_env *env, struct cl_io *io, + struct osc_page *ops, void *cbdata) { struct osc_thread_info *info = osc_env_info(env); struct cl_page *page = ops->ops_cl.cpl_page; @@ -3149,8 +3278,10 @@ static int discard_cb(const struct lu_env *env, struct cl_io *io, /* page is top page. */ info->oti_next_index = osc_index(ops) + 1; if (cl_page_own(env, io, page) == 0) { - KLASSERT(ergo(page->cp_type == CPT_CACHEABLE, - !PageDirty(cl_page_vmpage(page)))); + if (!ergo(page->cp_type == CPT_CACHEABLE, + !PageDirty(cl_page_vmpage(page)))) + CL_PAGE_DEBUG(D_ERROR, env, page, + "discard dirty page?\n"); /* discard the page */ cl_page_discard(env, io, page); @@ -3161,6 +3292,7 @@ static int discard_cb(const struct lu_env *env, struct cl_io *io, return CLP_GANG_OKAY; } +EXPORT_SYMBOL(osc_discard_cb); /** * Discard pages protected by the given lock. This function traverses radix @@ -3171,10 +3303,10 @@ static int discard_cb(const struct lu_env *env, struct cl_io *io, * behind this being that lock cancellation cannot be delayed indefinitely). */ 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) + pgoff_t start, pgoff_t end, bool discard) { struct osc_thread_info *info = osc_env_info(env); - struct cl_io *io = &info->oti_io; + struct cl_io *io = osc_env_thread_io(env); osc_page_gang_cbt cb; int res; int result; @@ -3187,7 +3319,7 @@ int osc_lock_discard_pages(const struct lu_env *env, struct osc_object *osc, if (result != 0) GOTO(out, result); - cb = mode == CLM_READ ? check_and_discard_cb : discard_cb; + cb = discard ? osc_discard_cb : check_and_discard_cb; info->oti_fn_index = info->oti_next_index = start; do { res = osc_page_gang_lookup(env, io, osc,