X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosc%2Fosc_cache.c;h=c15cac5b8fa097971366a9132bfa6b8f0db03ceb;hp=fc896eb406ee9fc825758e4df50ae25e31981a2b;hb=4311cdaa832fc4a444d48e50174dde09f21146d2;hpb=d1eb3357bab215018bcb29877ed8b2990e65cc73;ds=sidebyside diff --git a/lustre/osc/osc_cache.c b/lustre/osc/osc_cache.c index fc896eb..c15cac5 100644 --- a/lustre/osc/osc_cache.c +++ b/lustre/osc/osc_cache.c @@ -27,7 +27,7 @@ * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2012, Intel Corporation. + * Copyright (c) 2012, 2014, Intel Corporation. * */ /* @@ -76,6 +76,8 @@ static inline char *ext_flags(struct osc_extent *ext, char *flags) *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) @@ -92,46 +94,51 @@ static inline char *ext_flags(struct osc_extent *ext, char *flags) return flags; } -static inline char list_empty_marker(cfs_list_t *list) +static inline char list_empty_marker(struct list_head *list) { - return cfs_list_empty(list) ? '-' : '+'; + return list_empty(list) ? '-' : '+'; } #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), \ + atomic_read(&__ext->oe_refc), \ + 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, \ list_empty_marker(&__ext->oe_pages), \ - cfs_waitq_active(&__ext->oe_waitq) ? '+' : '-', \ - __ext->oe_osclock, __ext->oe_mppr, __ext->oe_owner, \ + 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 #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 @@ -183,20 +190,20 @@ static int osc_extent_sanity_check0(struct osc_extent *ext, if (ext->oe_state >= OES_STATE_MAX) GOTO(out, rc = 10); - if (cfs_atomic_read(&ext->oe_refc) <= 0) + if (atomic_read(&ext->oe_refc) <= 0) GOTO(out, rc = 20); - if (cfs_atomic_read(&ext->oe_refc) < cfs_atomic_read(&ext->oe_users)) + if (atomic_read(&ext->oe_refc) < atomic_read(&ext->oe_users)) GOTO(out, rc = 30); switch (ext->oe_state) { case OES_INV: - if (ext->oe_nr_pages > 0 || !cfs_list_empty(&ext->oe_pages)) + if (ext->oe_nr_pages > 0 || !list_empty(&ext->oe_pages)) GOTO(out, rc = 35); GOTO(out, rc = 0); break; case OES_ACTIVE: - if (cfs_atomic_read(&ext->oe_users) == 0) + if (atomic_read(&ext->oe_users) == 0) GOTO(out, rc = 40); if (ext->oe_hp) GOTO(out, rc = 50); @@ -209,22 +216,26 @@ static int osc_extent_sanity_check0(struct osc_extent *ext, if (ext->oe_fsync_wait && !ext->oe_urgent && !ext->oe_hp) GOTO(out, rc = 65); default: - if (cfs_atomic_read(&ext->oe_users) > 0) + if (atomic_read(&ext->oe_users) > 0) GOTO(out, rc = 70); } 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) @@ -239,8 +250,8 @@ static int osc_extent_sanity_check0(struct osc_extent *ext, GOTO(out, rc = 0); page_count = 0; - cfs_list_for_each_entry(oap, &ext->oe_pages, oap_pending_item) { - pgoff_t index = oap2cl_page(oap)->cp_index; + list_for_each_entry(oap, &ext->oe_pages, oap_pending_item) { + pgoff_t index = osc_index(oap2osc(oap)); ++page_count; if (index > ext->oe_end || index < ext->oe_start) GOTO(out, rc = 110); @@ -301,26 +312,26 @@ static void osc_extent_state_set(struct osc_extent *ext, int state) /* TODO: validate the state machine */ ext->oe_state = state; - cfs_waitq_broadcast(&ext->oe_waitq); + wake_up_all(&ext->oe_waitq); } static struct osc_extent *osc_extent_alloc(struct osc_object *obj) { 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; RB_CLEAR_NODE(&ext->oe_node); ext->oe_obj = obj; - cfs_atomic_set(&ext->oe_refc, 1); - cfs_atomic_set(&ext->oe_users, 0); - CFS_INIT_LIST_HEAD(&ext->oe_link); + atomic_set(&ext->oe_refc, 1); + atomic_set(&ext->oe_users, 0); + INIT_LIST_HEAD(&ext->oe_link); ext->oe_state = OES_INV; - CFS_INIT_LIST_HEAD(&ext->oe_pages); - cfs_waitq_init(&ext->oe_waitq); - ext->oe_osclock = NULL; + INIT_LIST_HEAD(&ext->oe_pages); + init_waitqueue_head(&ext->oe_waitq); + ext->oe_dlmlock = NULL; return ext; } @@ -332,23 +343,25 @@ static void osc_extent_free(struct osc_extent *ext) static struct osc_extent *osc_extent_get(struct osc_extent *ext) { - LASSERT(cfs_atomic_read(&ext->oe_refc) >= 0); - cfs_atomic_inc(&ext->oe_refc); + LASSERT(atomic_read(&ext->oe_refc) >= 0); + atomic_inc(&ext->oe_refc); return ext; } static void osc_extent_put(const struct lu_env *env, struct osc_extent *ext) { - LASSERT(cfs_atomic_read(&ext->oe_refc) > 0); - if (cfs_atomic_dec_and_test(&ext->oe_refc)) { - LASSERT(cfs_list_empty(&ext->oe_link)); - LASSERT(cfs_atomic_read(&ext->oe_users) == 0); + LASSERT(atomic_read(&ext->oe_refc) > 0); + if (atomic_dec_and_test(&ext->oe_refc)) { + LASSERT(list_empty(&ext->oe_link)); + LASSERT(atomic_read(&ext->oe_users) == 0); 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); } @@ -361,9 +374,9 @@ static void osc_extent_put(const struct lu_env *env, struct osc_extent *ext) */ static void osc_extent_put_trust(struct osc_extent *ext) { - LASSERT(cfs_atomic_read(&ext->oe_refc) > 1); + LASSERT(atomic_read(&ext->oe_refc) > 1); LASSERT(osc_object_is_locked(ext->oe_obj)); - cfs_atomic_dec(&ext->oe_refc); + atomic_dec(&ext->oe_refc); } /** @@ -425,7 +438,7 @@ static void osc_extent_insert(struct osc_object *obj, struct osc_extent *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); @@ -456,17 +469,17 @@ static struct osc_extent *osc_extent_hold(struct osc_extent *ext) osc_extent_state_set(ext, OES_ACTIVE); osc_update_pending(obj, OBD_BRW_WRITE, -ext->oe_nr_pages); } - cfs_atomic_inc(&ext->oe_users); - cfs_list_del_init(&ext->oe_link); + atomic_inc(&ext->oe_users); + list_del_init(&ext->oe_link); return osc_extent_get(ext); } static void __osc_extent_remove(struct osc_extent *ext) { LASSERT(osc_object_is_locked(ext->oe_obj)); - LASSERT(cfs_list_empty(&ext->oe_pages)); + LASSERT(list_empty(&ext->oe_pages)); osc_extent_erase(ext); - cfs_list_del_init(&ext->oe_link); + list_del_init(&ext->oe_link); osc_extent_state_set(ext, OES_INV); OSC_EXTENT_DUMP(D_CACHE, ext, "destroyed.\n"); } @@ -503,8 +516,8 @@ static int osc_extent_merge(const struct lu_env *env, struct osc_extent *cur, if (cur->oe_max_end != victim->oe_max_end) return -ERANGE; - LASSERT(cur->oe_osclock == victim->oe_osclock); - ppc_bits = osc_cli(obj)->cl_chunkbits - CFS_PAGE_SHIFT; + 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; if (chunk_start != (victim->oe_end >> ppc_bits) + 1 && @@ -520,8 +533,8 @@ static int osc_extent_merge(const struct lu_env *env, struct osc_extent *cur, /* only the following bits are needed to merge */ cur->oe_urgent |= victim->oe_urgent; cur->oe_memalloc |= victim->oe_memalloc; - cfs_list_splice_init(&victim->oe_pages, &cur->oe_pages); - cfs_list_del_init(&victim->oe_link); + list_splice_init(&victim->oe_pages, &cur->oe_pages); + list_del_init(&victim->oe_link); victim->oe_nr_pages = 0; osc_extent_get(victim); @@ -541,11 +554,11 @@ int osc_extent_release(const struct lu_env *env, struct osc_extent *ext) int rc = 0; ENTRY; - LASSERT(cfs_atomic_read(&ext->oe_users) > 0); + LASSERT(atomic_read(&ext->oe_users) > 0); LASSERT(sanity_check(ext) == 0); LASSERT(ext->oe_grants > 0); - if (cfs_atomic_dec_and_lock(&ext->oe_users, &obj->oo_lock)) { + if (atomic_dec_and_lock(&ext->oe_users, &obj->oo_lock)) { LASSERT(ext->oe_state == OES_ACTIVE); if (ext->oe_trunc_pending) { /* a truncate process is waiting for this extent. @@ -563,8 +576,8 @@ int osc_extent_release(const struct lu_env *env, struct osc_extent *ext) osc_extent_merge(env, ext, next_extent(ext)); if (ext->oe_urgent) - cfs_list_move_tail(&ext->oe_link, - &obj->oo_urgent_exts); + list_move_tail(&ext->oe_link, + &obj->oo_urgent_exts); } osc_object_unlock(obj); @@ -583,13 +596,14 @@ static inline int overlapped(struct osc_extent *ex1, struct osc_extent *ex2) * 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, + 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; @@ -607,12 +621,15 @@ struct osc_extent *osc_extent_find(const struct lu_env *env, 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 >= 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; @@ -621,24 +638,29 @@ struct osc_extent *osc_extent_find(const struct lu_env *env, 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); @@ -654,9 +676,9 @@ restart: 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; @@ -677,7 +699,7 @@ restart: * 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 @@ -750,7 +772,7 @@ restart: 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); } @@ -765,7 +787,7 @@ restart: 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); @@ -810,10 +832,12 @@ int osc_extent_finish(const struct lu_env *env, struct osc_extent *ext, ext->oe_rc = rc ?: ext->oe_nr_pages; EASSERT(ergo(rc == 0, ext->oe_state == OES_RPC), ext); - cfs_list_for_each_entry_safe(oap, tmp, &ext->oe_pages, + + osc_lru_add_batch(cli, &ext->oe_pages); + list_for_each_entry_safe(oap, tmp, &ext->oe_pages, oap_pending_item) { - cfs_list_del_init(&oap->oap_rpc_item); - cfs_list_del_init(&oap->oap_pending_item); + list_del_init(&oap->oap_rpc_item); + list_del_init(&oap->oap_pending_item); if (last_off <= oap->oap_obj_off) { last_off = oap->oap_obj_off; last_count = oap->oap_count; @@ -826,18 +850,18 @@ int osc_extent_finish(const struct lu_env *env, struct osc_extent *ext, 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. */ - int offset = oap->oap_page_off & ~CFS_PAGE_MASK; - int count = oap->oap_count + (offset & (blocksize - 1)); - int end = (offset + oap->oap_count) & (blocksize - 1); + int offset = last_off & ~CFS_PAGE_MASK; + int count = last_count + (offset & (blocksize - 1)); + int end = (offset + last_count) & (blocksize - 1); 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); @@ -895,7 +919,7 @@ static int osc_extent_wait(const struct lu_env *env, struct osc_extent *ext, "%s: wait ext to %d timedout, recovery in progress?\n", osc_export(obj)->exp_obd->obd_name, state); - lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL); + lwi = LWI_INTR(NULL, NULL); rc = l_wait_event(ext->oe_waitq, extent_wait_cb(ext, state), &lwi); } @@ -919,7 +943,8 @@ static int osc_extent_truncate(struct osc_extent *ext, pgoff_t trunc_index, 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; @@ -941,31 +966,30 @@ static int osc_extent_truncate(struct osc_extent *ext, pgoff_t trunc_index, GOTO(out, rc); /* discard all pages with index greater then trunc_index */ - cfs_list_for_each_entry_safe(oap, tmp, &ext->oe_pages, + list_for_each_entry_safe(oap, tmp, &ext->oe_pages, oap_pending_item) { - struct cl_page *sub = oap2cl_page(oap); - struct cl_page *page = cl_page_top(sub); + pgoff_t index = osc_index(oap2osc(oap)); + struct cl_page *page = oap2cl_page(oap); - LASSERT(cfs_list_empty(&oap->oap_rpc_item)); + LASSERT(list_empty(&oap->oap_rpc_item)); /* only discard the pages with their index greater than * trunc_index, and ... */ - if (sub->cp_index < trunc_index || - (sub->cp_index == trunc_index && partial)) { + if (index < trunc_index || + (index == trunc_index && partial)) { /* accounting how many pages remaining in the chunk * so that we can calculate grants correctly. */ - if (sub->cp_index >> ppc_bits == trunc_chunk) + if (index >> ppc_bits == trunc_chunk) ++pages_in_chunk; continue; } - cfs_list_del_init(&oap->oap_pending_item); + list_del_init(&oap->oap_pending_item); cl_page_get(page); - lu_ref_add(&page->cp_reference, "truncate", cfs_current()); + lu_ref_add(&page->cp_reference, "truncate", current); if (cl_page_own(env, io, page) == 0) { - cl_page_unmap(env, io, page); cl_page_discard(env, io, page); cl_page_disown(env, io, page); } else { @@ -973,7 +997,7 @@ static int osc_extent_truncate(struct osc_extent *ext, pgoff_t trunc_index, LASSERT(0); } - lu_ref_del(&page->cp_reference, "truncate", cfs_current()); + lu_ref_del(&page->cp_reference, "truncate", current); cl_page_put(env, page); --ext->oe_nr_pages; @@ -1033,7 +1057,7 @@ static int osc_extent_make_ready(const struct lu_env *env, 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; @@ -1045,7 +1069,7 @@ static int osc_extent_make_ready(const struct lu_env *env, OSC_EXTENT_DUMP(D_CACHE, ext, "make ready\n"); - cfs_list_for_each_entry(oap, &ext->oe_pages, oap_pending_item) { + list_for_each_entry(oap, &ext->oe_pages, oap_pending_item) { ++page_count; if (last == NULL || last->oap_obj_off < oap->oap_obj_off) last = oap; @@ -1074,18 +1098,23 @@ static int osc_extent_make_ready(const struct lu_env *env, /* 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 <= CFS_PAGE_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); } /* for the rest of pages, we don't need to call osf_refresh_count() * because it's known they are not the last page */ - cfs_list_for_each_entry(oap, &ext->oe_pages, oap_pending_item) { + 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; + spin_lock(&oap->oap_lock); oap->oap_async_flags |= ASYNC_COUNT_STABLE; + spin_unlock(&oap->oap_lock); } } @@ -1108,7 +1137,7 @@ static int osc_extent_expand(struct osc_extent *ext, pgoff_t index, int *grants) 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; @@ -1164,15 +1193,15 @@ static void osc_extent_tree_dump0(int level, struct osc_object *obj, OSC_EXTENT_DUMP(level, ext, "in tree %d.\n", cnt++); cnt = 1; - cfs_list_for_each_entry(ext, &obj->oo_hp_exts, oe_link) + list_for_each_entry(ext, &obj->oo_hp_exts, oe_link) OSC_EXTENT_DUMP(level, ext, "hp %d.\n", cnt++); cnt = 1; - cfs_list_for_each_entry(ext, &obj->oo_urgent_exts, oe_link) + list_for_each_entry(ext, &obj->oo_urgent_exts, oe_link) OSC_EXTENT_DUMP(level, ext, "urgent %d.\n", cnt++); cnt = 1; - cfs_list_for_each_entry(ext, &obj->oo_reading_exts, oe_link) + list_for_each_entry(ext, &obj->oo_reading_exts, oe_link) OSC_EXTENT_DUMP(level, ext, "reading %d.\n", cnt++); /* osc_object_unlock(obj); */ } @@ -1181,8 +1210,8 @@ static void osc_extent_tree_dump0(int level, struct osc_object *obj, static inline int osc_is_ready(struct osc_object *osc) { - return !cfs_list_empty(&osc->oo_ready_item) || - !cfs_list_empty(&osc->oo_hp_ready_item); + return !list_empty(&osc->oo_ready_item) || + !list_empty(&osc->oo_hp_ready_item); } #define OSC_IO_DEBUG(OSC, STR, args...) \ @@ -1190,10 +1219,10 @@ static inline int osc_is_ready(struct osc_object *osc) (OSC), osc_is_ready(OSC), \ list_empty_marker(&(OSC)->oo_hp_ready_item), \ list_empty_marker(&(OSC)->oo_ready_item), \ - cfs_atomic_read(&(OSC)->oo_nr_writes), \ + atomic_read(&(OSC)->oo_nr_writes), \ list_empty_marker(&(OSC)->oo_hp_exts), \ list_empty_marker(&(OSC)->oo_urgent_exts), \ - cfs_atomic_read(&(OSC)->oo_nr_reads), \ + atomic_read(&(OSC)->oo_nr_reads), \ list_empty_marker(&(OSC)->oo_reading_exts), \ ##args) @@ -1201,7 +1230,7 @@ static int osc_make_ready(const struct lu_env *env, struct osc_async_page *oap, int cmd) { struct osc_page *opg = oap2osc_page(oap); - struct cl_page *page = cl_page_top(oap2cl_page(oap)); + struct cl_page *page = oap2cl_page(oap); int result; LASSERT(cmd == OBD_BRW_WRITE); /* no cached reads */ @@ -1217,7 +1246,7 @@ static int osc_refresh_count(const struct lu_env *env, struct osc_async_page *oap, int cmd) { struct osc_page *opg = oap2osc_page(oap); - struct cl_page *page = oap2cl_page(oap); + pgoff_t index = osc_index(oap2osc(oap)); struct cl_object *obj; struct cl_attr *attr = &osc_env_info(env)->oti_attr; @@ -1235,21 +1264,21 @@ static int osc_refresh_count(const struct lu_env *env, if (result < 0) return result; kms = attr->cat_kms; - if (cl_offset(obj, page->cp_index) >= kms) + if (cl_offset(obj, index) >= kms) /* catch race with truncate */ return 0; - else if (cl_offset(obj, page->cp_index + 1) > kms) + else if (cl_offset(obj, 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, int cmd, int rc) { struct osc_page *opg = oap2osc_page(oap); - struct cl_page *page = cl_page_top(oap2cl_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; @@ -1257,8 +1286,10 @@ static int osc_completion(const struct lu_env *env, struct osc_async_page *oap, ENTRY; cmd &= ~OBD_BRW_NOQUOTA; - LASSERT(equi(page->cp_state == CPS_PAGEIN, cmd == OBD_BRW_READ)); - LASSERT(equi(page->cp_state == CPS_PAGEOUT, cmd == OBD_BRW_WRITE)); + LASSERTF(equi(page->cp_state == CPS_PAGEIN, cmd == OBD_BRW_READ), + "cp_state:%u, cmd:%d\n", page->cp_state, cmd); + LASSERTF(equi(page->cp_state == CPS_PAGEOUT, cmd == OBD_BRW_WRITE), + "cp_state:%u, cmd:%d\n", page->cp_state, cmd); LASSERT(opg->ops_transfer_pinned); /* @@ -1275,16 +1306,14 @@ static int osc_completion(const struct lu_env *env, struct osc_async_page *oap, spin_lock(&obj->oo_seatbelt); LASSERT(opg->ops_submitter != NULL); - LASSERT(!cfs_list_empty(&opg->ops_inflight)); - cfs_list_del_init(&opg->ops_inflight); + 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; srvlock = oap->oap_brw_flags & OBD_BRW_SRVLOCK; - cl_page_completion(env, page, crt, rc); - /* statistic */ if (rc == 0 && srvlock) { struct lu_device *ld = opg->ops_cl.cpl_obj->co_lu.lo_dev; @@ -1303,37 +1332,38 @@ static int osc_completion(const struct lu_env *env, struct osc_async_page *oap, * reference counter protects page from concurrent reclaim. */ lu_ref_del(&page->cp_reference, "transfer", page); - /* - * As page->cp_obj is pinned by a reference from page->cp_req, it is - * safe to call cl_page_put() without risking object destruction in a - * non-blocking context. - */ - cl_page_put(env, page); + + cl_page_completion(env, page, crt, rc); + RETURN(0); } -#define OSC_DUMP_GRANT(cli, fmt, args...) do { \ - struct client_obd *__tmp = (cli); \ - CDEBUG(D_CACHE, "%s: { dirty: %ld/%ld dirty_pages: %d/%d " \ - "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, \ - __tmp->cl_lost_grant, __tmp->cl_avail_grant, \ - __tmp->cl_reserved_grant, __tmp->cl_w_in_flight, ##args); \ +#define OSC_DUMP_GRANT(lvl, 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, \ + __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_long_read(&__tmp->cl_lru_in_list), \ + atomic_long_read(&__tmp->cl_lru_busy), \ + atomic_read(&__tmp->cl_lru_shrinkers), ##args); \ } while (0) /* caller must hold loi_list_lock */ static void osc_consume_write_grant(struct client_obd *cli, struct brw_page *pga) { - LASSERT_SPIN_LOCKED(&cli->cl_loi_list_lock.lock); + assert_spin_locked(&cli->cl_loi_list_lock); LASSERT(!(pga->flag & OBD_BRW_FROM_GRANT)); - cfs_atomic_inc(&obd_dirty_pages); - cli->cl_dirty += CFS_PAGE_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", - CFS_PAGE_SIZE, pga, pga->pg); + PAGE_CACHE_SIZE, pga, pga->pg); osc_update_next_shrink(cli); } @@ -1344,19 +1374,19 @@ static void osc_release_write_grant(struct client_obd *cli, { ENTRY; - LASSERT_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; - cfs_atomic_dec(&obd_dirty_pages); - cli->cl_dirty -= CFS_PAGE_SIZE; + atomic_long_dec(&obd_dirty_pages); + cli->cl_dirty_pages--; 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; + atomic_long_dec(&obd_dirty_transit_pages); + cli->cl_dirty_transit--; } EXIT; } @@ -1395,14 +1425,14 @@ static void __osc_unreserve_grant(struct client_obd *cli, } } -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); } /** @@ -1412,7 +1442,7 @@ 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 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. @@ -1423,9 +1453,9 @@ static void osc_free_grant(struct client_obd *cli, unsigned int nr_pages, { int grant = (1 << cli->cl_chunkbits) + cli->cl_extent_tax; - 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; + 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 @@ -1434,10 +1464,10 @@ static void osc_free_grant(struct client_obd *cli, unsigned int nr_pages, 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); } /** @@ -1446,9 +1476,9 @@ static void osc_free_grant(struct client_obd *cli, unsigned int nr_pages, */ 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); } /** @@ -1461,18 +1491,18 @@ static int osc_enter_cache_try(struct client_obd *cli, { int rc; - OSC_DUMP_GRANT(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 + CFS_PAGE_SIZE <= cli->cl_dirty_max && - cfs_atomic_read(&obd_dirty_pages) + 1 <= 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 += CFS_PAGE_SIZE; - cfs_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; @@ -1486,9 +1516,9 @@ static int osc_enter_cache_try(struct client_obd *cli, static int ocw_granted(struct client_obd *cli, struct osc_cache_waiter *ocw) { int rc; - client_obd_list_lock(&cli->cl_loi_list_lock); - rc = cfs_list_empty(&ocw->ocw_entry) || cli->cl_w_in_flight == 0; - client_obd_list_unlock(&cli->cl_loi_list_lock); + spin_lock(&cli->cl_loi_list_lock); + rc = list_empty(&ocw->ocw_entry); + spin_unlock(&cli->cl_loi_list_lock); return rc; } @@ -1505,18 +1535,19 @@ 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 = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL); + struct l_wait_info lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(600), NULL, + LWI_ON_SIGNAL_NOOP, NULL); int rc = -EDQUOT; ENTRY; - OSC_DUMP_GRANT(cli, "need:%d.\n", bytes); + OSC_DUMP_GRANT(D_CACHE, cli, "need:%d.\n", bytes); - client_obd_list_lock(&cli->cl_loi_list_lock); + 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 < CFS_PAGE_SIZE || + cli->cl_dirty_max_pages == 0 || cli->cl_ar.ar_force_sync || loi->loi_ar.ar_force_sync) GOTO(out, rc = -EDQUOT); @@ -1530,93 +1561,94 @@ static int osc_enter_cache(const struct lu_env *env, struct client_obd *cli, * RPC size will be. * The exiting condition is no avail grants and no dirty pages caching, * that really means there is no space on the OST. */ - cfs_waitq_init(&ocw.ocw_waitq); + init_waitqueue_head(&ocw.ocw_waitq); ocw.ocw_oap = oap; ocw.ocw_grant = bytes; - if (cli->cl_dirty > 0 || cli->cl_w_in_flight > 0) { - cfs_list_add_tail(&ocw.ocw_entry, &cli->cl_cache_waiters); + 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); - /* First osc_io_unplug() tries to put current object - * on ready list, second osc_io_unplug() makes sure that - * dirty flush can still be triggered even if current - * object hasn't any dirty pages */ - osc_io_unplug(env, cli, osc, PDL_POLICY_ROUND); - osc_io_unplug(env, cli, NULL, PDL_POLICY_ROUND); + 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); 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 */ + /* l_wait_event is interrupted by signal, or timed out */ if (rc < 0) { - cfs_list_del_init(&ocw.ocw_entry); + 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; + } + list_del_init(&ocw.ocw_entry); GOTO(out, rc); } - /* If ocw_entry isn't empty, which means it's not waked up - * by osc_wake_cache_waiters(), then the page must not be - * granted yet. */ - if (!cfs_list_empty(&ocw.ocw_entry)) { - rc = -EDQUOT; - cfs_list_del_init(&ocw.ocw_entry); - } else { - rc = ocw.ocw_rc; - } + 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)) - rc = 0; + GOTO(out, rc = 0); } EXIT; out: - client_obd_list_unlock(&cli->cl_loi_list_lock); - OSC_DUMP_GRANT(cli, "returned %d.\n", rc); + spin_unlock(&cli->cl_loi_list_lock); + OSC_DUMP_GRANT(D_CACHE, cli, "returned %d.\n", rc); RETURN(rc); } /* caller must hold loi_list_lock */ void osc_wake_cache_waiters(struct client_obd *cli) { - cfs_list_t *l, *tmp; + struct list_head *l, *tmp; struct osc_cache_waiter *ocw; ENTRY; - cfs_list_for_each_safe(l, tmp, &cli->cl_cache_waiters) { - /* if we can't dirty more, we must wait until some is written */ - if ((cli->cl_dirty + CFS_PAGE_SIZE > cli->cl_dirty_max) || - (cfs_atomic_read(&obd_dirty_pages) + 1 > + 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 %d\n", cli->cl_dirty, - cli->cl_dirty_max, obd_max_dirty_pages); - return; + "osc max %ld, sys max %ld\n", + cli->cl_dirty_pages, cli->cl_dirty_max_pages, + obd_max_dirty_pages); + goto wakeup; } - /* if still dirty cache but no grant wait for pending RPCs that - * may yet return us some grant before doing sync writes */ - if (cli->cl_w_in_flight && cli->cl_avail_grant < CFS_PAGE_SIZE) { - CDEBUG(D_CACHE, "%u BRW writes in flight, no grant\n", - cli->cl_w_in_flight); - return; - } - - ocw = cfs_list_entry(l, struct osc_cache_waiter, ocw_entry); - cfs_list_del_init(&ocw->ocw_entry); - ocw->ocw_rc = 0; if (!osc_enter_cache_try(cli, ocw->ocw_oap, ocw->ocw_grant, 0)) ocw->ocw_rc = -EDQUOT; +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); - cfs_waitq_signal(&ocw->ocw_waitq); + wake_up(&ocw->ocw_waitq); } EXIT; @@ -1624,7 +1656,7 @@ void osc_wake_cache_waiters(struct client_obd *cli) static int osc_max_rpc_in_flight(struct client_obd *cli, struct osc_object *osc) { - int hprpc = !!cfs_list_empty(&osc->oo_hp_exts); + int hprpc = !!list_empty(&osc->oo_hp_exts); return rpcs_in_flight(cli) >= cli->cl_max_rpcs_in_flight + hprpc; } @@ -1645,39 +1677,39 @@ static int osc_makes_rpc(struct client_obd *cli, struct osc_object *osc, invalid_import = 1; if (cmd & OBD_BRW_WRITE) { - if (cfs_atomic_read(&osc->oo_nr_writes) == 0) + if (atomic_read(&osc->oo_nr_writes) == 0) RETURN(0); if (invalid_import) { CDEBUG(D_CACHE, "invalid import forcing RPC\n"); RETURN(1); } - if (!cfs_list_empty(&osc->oo_hp_exts)) { + if (!list_empty(&osc->oo_hp_exts)) { CDEBUG(D_CACHE, "high prio request forcing RPC\n"); RETURN(1); } - if (!cfs_list_empty(&osc->oo_urgent_exts)) { + if (!list_empty(&osc->oo_urgent_exts)) { CDEBUG(D_CACHE, "urgent request forcing RPC\n"); RETURN(1); } /* trigger a write rpc stream as long as there are dirtiers * waiting for space. as they're waiting, they're not going to * create more pages to coalesce with what's waiting.. */ - if (!cfs_list_empty(&cli->cl_cache_waiters)) { + if (!list_empty(&cli->cl_cache_waiters)) { CDEBUG(D_CACHE, "cache waiters forcing RPC\n"); RETURN(1); } - if (cfs_atomic_read(&osc->oo_nr_writes) >= + if (atomic_read(&osc->oo_nr_writes) >= cli->cl_max_pages_per_rpc) RETURN(1); } else { - if (cfs_atomic_read(&osc->oo_nr_reads) == 0) + if (atomic_read(&osc->oo_nr_reads) == 0) RETURN(0); if (invalid_import) { CDEBUG(D_CACHE, "invalid import forcing RPC\n"); RETURN(1); } /* all read are urgent. */ - if (!cfs_list_empty(&osc->oo_reading_exts)) + if (!list_empty(&osc->oo_reading_exts)) RETURN(1); } @@ -1688,28 +1720,29 @@ static void osc_update_pending(struct osc_object *obj, int cmd, int delta) { struct client_obd *cli = osc_cli(obj); if (cmd & OBD_BRW_WRITE) { - cfs_atomic_add(delta, &obj->oo_nr_writes); - cfs_atomic_add(delta, &cli->cl_pending_w_pages); - LASSERT(cfs_atomic_read(&obj->oo_nr_writes) >= 0); + atomic_add(delta, &obj->oo_nr_writes); + atomic_add(delta, &cli->cl_pending_w_pages); + LASSERT(atomic_read(&obj->oo_nr_writes) >= 0); } else { - cfs_atomic_add(delta, &obj->oo_nr_reads); - cfs_atomic_add(delta, &cli->cl_pending_r_pages); - LASSERT(cfs_atomic_read(&obj->oo_nr_reads) >= 0); + atomic_add(delta, &obj->oo_nr_reads); + atomic_add(delta, &cli->cl_pending_r_pages); + LASSERT(atomic_read(&obj->oo_nr_reads) >= 0); } OSC_IO_DEBUG(obj, "update pending cmd %d delta %d.\n", cmd, delta); } static int osc_makes_hprpc(struct osc_object *obj) { - return !cfs_list_empty(&obj->oo_hp_exts); + return !list_empty(&obj->oo_hp_exts); } -static void on_list(cfs_list_t *item, cfs_list_t *list, int should_be_on) +static void on_list(struct list_head *item, struct list_head *list, + int should_be_on) { - if (cfs_list_empty(item) && should_be_on) - cfs_list_add_tail(item, list); - else if (!cfs_list_empty(item) && !should_be_on) - cfs_list_del_init(item); + if (list_empty(item) && should_be_on) + list_add_tail(item, list); + else if (!list_empty(item) && !should_be_on) + list_del_init(item); } /* maintain the osc's cli list membership invariants so that osc_send_oap_rpc @@ -1728,10 +1761,10 @@ static int __osc_list_maint(struct client_obd *cli, struct osc_object *osc) } on_list(&osc->oo_write_item, &cli->cl_loi_write_list, - cfs_atomic_read(&osc->oo_nr_writes) > 0); + atomic_read(&osc->oo_nr_writes) > 0); on_list(&osc->oo_read_item, &cli->cl_loi_read_list, - cfs_atomic_read(&osc->oo_nr_reads) > 0); + atomic_read(&osc->oo_nr_reads) > 0); return osc_is_ready(osc); } @@ -1740,9 +1773,9 @@ static int osc_list_maint(struct client_obd *cli, struct osc_object *osc) { 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; } @@ -1792,10 +1825,10 @@ static void osc_ap_completion(const struct lu_env *env, struct client_obd *cli, 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); @@ -1812,10 +1845,14 @@ static void osc_ap_completion(const struct lu_env *env, struct client_obd *cli, * - extent must be compatible with previous ones */ static int try_to_add_extent_for_io(struct client_obd *cli, - struct osc_extent *ext, cfs_list_t *rpclist, + struct osc_extent *ext, + struct list_head *rpclist, 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), @@ -1825,14 +1862,22 @@ static int try_to_add_extent_for_io(struct client_obd *cli, if (*pc + ext->oe_nr_pages > *max_pages) RETURN(0); - cfs_list_for_each_entry(tmp, rpclist, oe_link) { - EASSERT(tmp->oe_owner == cfs_current(), tmp); + 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)) { 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) @@ -1843,8 +1888,8 @@ static int try_to_add_extent_for_io(struct client_obd *cli, } *pc += ext->oe_nr_pages; - cfs_list_move_tail(&ext->oe_link, rpclist); - ext->oe_owner = cfs_current(); + list_move_tail(&ext->oe_link, rpclist); + ext->oe_owner = current; RETURN(1); } @@ -1861,7 +1906,7 @@ static int try_to_add_extent_for_io(struct client_obd *cli, * 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, cfs_list_t *rpclist) +static int get_write_extents(struct osc_object *obj, struct list_head *rpclist) { struct client_obd *cli = osc_cli(obj); struct osc_extent *ext; @@ -1869,9 +1914,9 @@ static int get_write_extents(struct osc_object *obj, cfs_list_t *rpclist) unsigned int max_pages = cli->cl_max_pages_per_rpc; LASSERT(osc_object_is_locked(obj)); - while (!cfs_list_empty(&obj->oo_hp_exts)) { - ext = cfs_list_entry(obj->oo_hp_exts.next, struct osc_extent, - oe_link); + 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)) @@ -1881,9 +1926,9 @@ static int get_write_extents(struct osc_object *obj, cfs_list_t *rpclist) if (page_count == max_pages) return page_count; - while (!cfs_list_empty(&obj->oo_urgent_exts)) { - ext = cfs_list_entry(obj->oo_urgent_exts.next, - struct osc_extent, oe_link); + 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; @@ -1893,7 +1938,7 @@ static int get_write_extents(struct osc_object *obj, cfs_list_t *rpclist) while ((ext = next_extent(ext)) != NULL) { if ((ext->oe_state != OES_CACHE) || - (!cfs_list_empty(&ext->oe_link) && + (!list_empty(&ext->oe_link) && ext->oe_owner != NULL)) continue; @@ -1909,7 +1954,7 @@ static int get_write_extents(struct osc_object *obj, cfs_list_t *rpclist) while (ext != NULL) { if ((ext->oe_state != OES_CACHE) || /* this extent may be already in current rpclist */ - (!cfs_list_empty(&ext->oe_link) && ext->oe_owner != NULL)) { + (!list_empty(&ext->oe_link) && ext->oe_owner != NULL)) { ext = next_extent(ext); continue; } @@ -1926,8 +1971,9 @@ static int get_write_extents(struct osc_object *obj, cfs_list_t *rpclist) 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) { - CFS_LIST_HEAD(rpclist); + struct list_head rpclist = LIST_HEAD_INIT(rpclist); struct osc_extent *ext; struct osc_extent *tmp; struct osc_extent *first = NULL; @@ -1939,14 +1985,14 @@ osc_send_write_rpc(const struct lu_env *env, struct client_obd *cli, LASSERT(osc_object_is_locked(osc)); page_count = get_write_extents(osc, &rpclist); - LASSERT(equi(page_count == 0, cfs_list_empty(&rpclist))); + LASSERT(equi(page_count == 0, list_empty(&rpclist))); - if (cfs_list_empty(&rpclist)) + if (list_empty(&rpclist)) RETURN(0); osc_update_pending(osc, OBD_BRW_WRITE, -page_count); - cfs_list_for_each_entry(ext, &rpclist, oe_link) { + list_for_each_entry(ext, &rpclist, oe_link) { LASSERT(ext->oe_state == OES_CACHE || ext->oe_state == OES_LOCK_DONE); if (ext->oe_state == OES_CACHE) @@ -1959,11 +2005,11 @@ osc_send_write_rpc(const struct lu_env *env, struct client_obd *cli, * lock order is page lock -> object lock. */ osc_object_unlock(osc); - cfs_list_for_each_entry_safe(ext, tmp, &rpclist, oe_link) { + list_for_each_entry_safe(ext, tmp, &rpclist, oe_link) { if (ext->oe_state == OES_LOCKING) { rc = osc_extent_make_ready(env, ext); if (unlikely(rc < 0)) { - cfs_list_del_init(&ext->oe_link); + list_del_init(&ext->oe_link); osc_extent_finish(env, ext, 0, rc); continue; } @@ -1976,10 +2022,10 @@ osc_send_write_rpc(const struct lu_env *env, struct client_obd *cli, } } - if (!cfs_list_empty(&rpclist)) { + if (!list_empty(&rpclist)) { LASSERT(page_count > 0); rc = osc_build_rpc(env, cli, &rpclist, OBD_BRW_WRITE, pol); - LASSERT(cfs_list_empty(&rpclist)); + LASSERT(list_empty(&rpclist)); } osc_object_lock(osc); @@ -1999,17 +2045,18 @@ osc_send_write_rpc(const struct lu_env *env, struct client_obd *cli, 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; - CFS_LIST_HEAD(rpclist); + struct list_head rpclist = LIST_HEAD_INIT(rpclist); int page_count = 0; unsigned int max_pages = cli->cl_max_pages_per_rpc; int rc = 0; ENTRY; LASSERT(osc_object_is_locked(osc)); - cfs_list_for_each_entry_safe(ext, next, + 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, @@ -2022,12 +2069,12 @@ osc_send_read_rpc(const struct lu_env *env, struct client_obd *cli, osc_update_pending(osc, OBD_BRW_READ, -page_count); - if (!cfs_list_empty(&rpclist)) { + if (!list_empty(&rpclist)) { osc_object_unlock(osc); LASSERT(page_count > 0); rc = osc_build_rpc(env, cli, &rpclist, OBD_BRW_READ, pol); - LASSERT(cfs_list_empty(&rpclist)); + LASSERT(list_empty(&rpclist)); osc_object_lock(osc); } @@ -2035,9 +2082,9 @@ osc_send_read_rpc(const struct lu_env *env, struct client_obd *cli, } #define list_to_obj(list, item) ({ \ - cfs_list_t *__tmp = (list)->next; \ - cfs_list_del_init(__tmp); \ - cfs_list_entry(__tmp, struct osc_object, oo_##item); \ + struct list_head *__tmp = (list)->next; \ + list_del_init(__tmp); \ + list_entry(__tmp, struct osc_object, oo_##item); \ }) /* This is called by osc_check_rpcs() to find which objects have pages that @@ -2049,26 +2096,26 @@ static struct osc_object *osc_next_obj(struct client_obd *cli) /* First return objects that have blocked locks so that they * will be flushed quickly and other clients can get the lock, * then objects which have pages ready to be stuffed into RPCs */ - if (!cfs_list_empty(&cli->cl_loi_hp_ready_list)) + if (!list_empty(&cli->cl_loi_hp_ready_list)) RETURN(list_to_obj(&cli->cl_loi_hp_ready_list, hp_ready_item)); - if (!cfs_list_empty(&cli->cl_loi_ready_list)) + if (!list_empty(&cli->cl_loi_ready_list)) RETURN(list_to_obj(&cli->cl_loi_ready_list, ready_item)); /* then if we have cache waiters, return all objects with queued * writes. This is especially important when many small files * have filled up the cache and not been fired into rpcs because * they don't pass the nr_pending/object threshhold */ - if (!cfs_list_empty(&cli->cl_cache_waiters) && - !cfs_list_empty(&cli->cl_loi_write_list)) + if (!list_empty(&cli->cl_cache_waiters) && + !list_empty(&cli->cl_loi_write_list)) RETURN(list_to_obj(&cli->cl_loi_write_list, write_item)); /* then return all queued objects when we have an invalid import * so that they get flushed */ if (cli->cl_import == NULL || cli->cl_import->imp_invalid) { - if (!cfs_list_empty(&cli->cl_loi_write_list)) + if (!list_empty(&cli->cl_loi_write_list)) RETURN(list_to_obj(&cli->cl_loi_write_list, write_item)); - if (!cfs_list_empty(&cli->cl_loi_read_list)) + if (!list_empty(&cli->cl_loi_read_list)) RETURN(list_to_obj(&cli->cl_loi_read_list, read_item)); } @@ -2078,6 +2125,7 @@ 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) +__must_hold(&cli->cl_loi_list_lock) { struct osc_object *osc; int rc = 0; @@ -2085,7 +2133,7 @@ static void osc_check_rpcs(const struct lu_env *env, struct client_obd *cli, 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)); @@ -2095,8 +2143,8 @@ static void osc_check_rpcs(const struct lu_env *env, struct client_obd *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()); + 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 * reads and writes in an object. The makes_rpc checks here @@ -2137,33 +2185,34 @@ static void osc_check_rpcs(const struct lu_env *env, struct client_obd *cli, 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", current); cl_object_put(env, obj); - client_obd_list_lock(&cli->cl_loi_list_lock); + spin_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) { - osc_check_rpcs(env, cli, pol); - } 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 */ + atomic_inc(&cli->cl_lru_shrinkers); + spin_lock(&cli->cl_loi_list_lock); + osc_check_rpcs(env, cli, pol); + 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); + rc = ptlrpcd_queue_work(cli->cl_writeback_work); } - client_obd_list_unlock(&cli->cl_loi_list_lock); return rc; } @@ -2181,7 +2230,7 @@ void osc_io_unplug(const struct lu_env *env, struct client_obd *cli, } 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; @@ -2201,8 +2250,8 @@ int osc_prep_async_page(struct osc_object *osc, struct osc_page *ops, if (!client_is_remote(exp) && cfs_capable(CFS_CAP_SYS_RESOURCE)) oap->oap_brw_flags = OBD_BRW_NOQUOTA; - CFS_INIT_LIST_HEAD(&oap->oap_pending_item); - CFS_INIT_LIST_HEAD(&oap->oap_rpc_item); + 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", @@ -2232,8 +2281,8 @@ int osc_queue_async_io(const struct lu_env *env, struct cl_io *io, if (cli->cl_import == NULL || cli->cl_import->imp_invalid) RETURN(-EIO); - if (!cfs_list_empty(&oap->oap_pending_item) || - !cfs_list_empty(&oap->oap_rpc_item)) + if (!list_empty(&oap->oap_pending_item) || + !list_empty(&oap->oap_rpc_item)) RETURN(-EBUSY); /* Set the OBD_BRW_SRVLOCK before the page is queued. */ @@ -2268,13 +2317,15 @@ int osc_queue_async_io(const struct lu_env *env, struct cl_io *io, oap->oap_cmd = cmd; oap->oap_page_off = ops->ops_from; oap->oap_count = ops->ops_to - ops->ops_from; + /* No need to hold a lock here, + * since this page is not in any list yet. */ oap->oap_async_flags = 0; oap->oap_brw_flags = brw_flags; OSC_IO_DEBUG(osc, "oap %p page %p added for cmd %d\n", oap, oap->oap_page, oap->oap_cmd & OBD_BRW_RWMASK); - index = oap2cl_page(oap)->cp_index; + index = osc_index(oap2osc(oap)); /* Add this page into extent by the following steps: * 1. if there exists an active extent for this IO, mostly this page @@ -2291,9 +2342,9 @@ int osc_queue_async_io(const struct lu_env *env, struct cl_io *io, 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; @@ -2367,7 +2418,7 @@ int osc_queue_async_io(const struct lu_env *env, struct cl_io *io, else LASSERT(ext->oe_srvlock == ops->ops_srvlock); ++ext->oe_nr_pages; - cfs_list_add_tail(&oap->oap_pending_item, &ext->oe_pages); + list_add_tail(&oap->oap_pending_item, &ext->oe_pages); osc_object_unlock(osc); } RETURN(rc); @@ -2384,20 +2435,20 @@ int osc_teardown_async_page(const struct lu_env *env, LASSERT(oap->oap_magic == OAP_MAGIC); CDEBUG(D_INFO, "teardown oap %p page %p at index %lu.\n", - oap, ops, oap2cl_page(oap)->cp_index); + oap, ops, osc_index(oap2osc(oap))); osc_object_lock(obj); - if (!cfs_list_empty(&oap->oap_rpc_item)) { + if (!list_empty(&oap->oap_rpc_item)) { CDEBUG(D_CACHE, "oap %p is not in cache.\n", oap); rc = -EBUSY; - } else if (!cfs_list_empty(&oap->oap_pending_item)) { - ext = osc_extent_lookup(obj, oap2cl_page(oap)->cp_index); + } else if (!list_empty(&oap->oap_pending_item)) { + ext = osc_extent_lookup(obj, osc_index(oap2osc(oap))); /* only truncated pages are allowed to be taken out. * See osc_extent_truncate() and osc_cache_truncate_start() * for details. */ if (ext != NULL && ext->oe_state != OES_TRUNC) { OSC_EXTENT_DUMP(D_ERROR, ext, "trunc at %lu.\n", - oap2cl_page(oap)->cp_index); + osc_index(oap2osc(oap))); rc = -EBUSY; } } @@ -2420,7 +2471,7 @@ int osc_flush_async_page(const struct lu_env *env, struct cl_io *io, struct osc_extent *ext = NULL; struct osc_object *obj = cl2osc(ops->ops_cl.cpl_obj); struct cl_page *cp = ops->ops_cl.cpl_page; - pgoff_t index = cp->cp_index; + pgoff_t index = osc_index(ops); struct osc_async_page *oap = &ops->ops_oap; bool unplug = false; int rc = 0; @@ -2436,8 +2487,7 @@ int osc_flush_async_page(const struct lu_env *env, struct cl_io *io, switch (ext->oe_state) { case OES_RPC: case OES_LOCK_DONE: - CL_PAGE_DEBUG(D_ERROR, env, cl_page_top(cp), - "flush an in-rpc page?\n"); + CL_PAGE_DEBUG(D_ERROR, env, cp, "flush an in-rpc page?\n"); LASSERT(0); break; case OES_LOCKING: @@ -2449,12 +2499,18 @@ int osc_flush_async_page(const struct lu_env *env, struct cl_io *io, * really sending the RPC. */ case OES_TRUNC: /* race with truncate, page will be redirtied */ + case OES_ACTIVE: + /* The extent is active so we need to abort and let the caller + * re-dirty the page. If we continued on here, and we were the + * one making the extent active, we could deadlock waiting for + * the page writeback to clear but it won't because the extent + * is active and won't be written out. */ GOTO(out, rc = -EAGAIN); default: break; } - rc = cl_page_prep(env, io, cl_page_top(cp), CRT_WRITE); + rc = cl_page_prep(env, io, cp, CRT_WRITE); if (rc) GOTO(out, rc); @@ -2462,15 +2518,15 @@ int osc_flush_async_page(const struct lu_env *env, struct cl_io *io, 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; if (ext->oe_state == OES_CACHE) { OSC_EXTENT_DUMP(D_CACHE, ext, "flush page %p make it urgent.\n", oap); - if (cfs_list_empty(&ext->oe_link)) - cfs_list_add_tail(&ext->oe_link, &obj->oo_urgent_exts); + if (list_empty(&ext->oe_link)) + list_add_tail(&ext->oe_link, &obj->oo_urgent_exts); unplug = true; } rc = 0; @@ -2498,8 +2554,8 @@ int osc_cancel_async_page(const struct lu_env *env, struct osc_page *ops) struct client_obd *cli = osc_cli(obj); struct osc_extent *ext; struct osc_extent *found = NULL; - cfs_list_t *plist; - pgoff_t index = oap2cl_page(oap)->cp_index; + struct list_head *plist; + pgoff_t index = osc_index(ops); int rc = -EBUSY; int cmd; ENTRY; @@ -2516,7 +2572,7 @@ int osc_cancel_async_page(const struct lu_env *env, struct osc_page *ops) plist = &obj->oo_reading_exts; cmd = OBD_BRW_READ; } - cfs_list_for_each_entry(ext, plist, oe_link) { + list_for_each_entry(ext, plist, oe_link) { if (ext->oe_start <= index && ext->oe_end >= index) { LASSERT(ext->oe_state == OES_LOCK_DONE); /* For OES_LOCK_DONE state extent, it has already held @@ -2526,7 +2582,7 @@ int osc_cancel_async_page(const struct lu_env *env, struct osc_page *ops) } } if (found != NULL) { - cfs_list_del_init(&found->oe_link); + list_del_init(&found->oe_link); osc_update_pending(obj, cmd, -found->oe_nr_pages); osc_object_unlock(obj); @@ -2550,7 +2606,7 @@ int osc_cancel_async_page(const struct lu_env *env, struct osc_page *ops) } int osc_queue_sync_pages(const struct lu_env *env, struct osc_object *obj, - cfs_list_t *list, int cmd, int brw_flags) + struct list_head *list, int cmd, int brw_flags) { struct client_obd *cli = osc_cli(obj); struct osc_extent *ext; @@ -2561,26 +2617,27 @@ int osc_queue_sync_pages(const struct lu_env *env, struct osc_object *obj, pgoff_t end = 0; ENTRY; - cfs_list_for_each_entry(oap, list, oap_pending_item) { - struct cl_page *cp = oap2cl_page(oap); - if (cp->cp_index > end) - end = cp->cp_index; - if (cp->cp_index < start) - start = cp->cp_index; + list_for_each_entry(oap, list, oap_pending_item) { + pgoff_t index = osc_index(oap2osc(oap)); + if (index > end) + end = index; + if (index < start) + start = index; ++page_count; mppr <<= (page_count > mppr); } ext = osc_extent_alloc(obj); if (ext == NULL) { - cfs_list_for_each_entry(oap, list, oap_pending_item) { - cfs_list_del_init(&oap->oap_pending_item); + list_for_each_entry(oap, list, oap_pending_item) { + list_del_init(&oap->oap_pending_item); osc_ap_completion(env, cli, oap, 0, -ENOMEM); } RETURN(-ENOMEM); } 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; @@ -2588,21 +2645,21 @@ int osc_queue_sync_pages(const struct lu_env *env, struct osc_object *obj, ext->oe_srvlock = !!(brw_flags & OBD_BRW_SRVLOCK); ext->oe_nr_pages = page_count; ext->oe_mppr = mppr; - cfs_list_splice_init(list, &ext->oe_pages); + list_splice_init(list, &ext->oe_pages); 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) { - cfs_list_add_tail(&ext->oe_link, &obj->oo_urgent_exts); + list_add_tail(&ext->oe_link, &obj->oo_urgent_exts); osc_update_pending(obj, OBD_BRW_WRITE, page_count); } else { - cfs_list_add_tail(&ext->oe_link, &obj->oo_reading_exts); + list_add_tail(&ext->oe_link, &obj->oo_reading_exts); osc_update_pending(obj, OBD_BRW_READ, page_count); } osc_object_unlock(obj); - osc_io_unplug(env, cli, obj, PDL_POLICY_ROUND); + osc_io_unplug_async(env, cli, obj); RETURN(0); } @@ -2616,7 +2673,7 @@ int osc_cache_truncate_start(const struct lu_env *env, struct osc_io *oio, struct osc_extent *ext; struct osc_extent *waiting = NULL; pgoff_t index; - CFS_LIST_HEAD(list); + struct list_head list = LIST_HEAD_INIT(list); int result = 0; bool partial; ENTRY; @@ -2663,8 +2720,8 @@ again: osc_update_pending(obj, OBD_BRW_WRITE, -ext->oe_nr_pages); } - EASSERT(cfs_list_empty(&ext->oe_link), ext); - cfs_list_add_tail(&ext->oe_link, &list); + EASSERT(list_empty(&ext->oe_link), ext); + list_add_tail(&ext->oe_link, &list); ext = next_extent(ext); } @@ -2672,11 +2729,11 @@ again: osc_list_maint(cli, obj); - while (!cfs_list_empty(&list)) { + while (!list_empty(&list)) { int rc; - ext = cfs_list_entry(list.next, struct osc_extent, oe_link); - cfs_list_del_init(&ext->oe_link); + ext = list_entry(list.next, struct osc_extent, oe_link); + list_del_init(&ext->oe_link); /* extent may be in OES_ACTIVE state because inode mutex * is released before osc_io_end() in file write case */ @@ -2719,7 +2776,7 @@ again: * 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; @@ -2749,7 +2806,7 @@ void osc_cache_truncate_end(const struct lu_env *env, struct osc_io *oio, osc_extent_state_set(ext, OES_CACHE); if (ext->oe_fsync_wait && !ext->oe_urgent) { ext->oe_urgent = 1; - cfs_list_move_tail(&ext->oe_link, &obj->oo_urgent_exts); + list_move_tail(&ext->oe_link, &obj->oo_urgent_exts); unplug = true; } osc_update_pending(obj, OBD_BRW_WRITE, ext->oe_nr_pages); @@ -2830,7 +2887,7 @@ 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; - CFS_LIST_HEAD(discard_list); + struct list_head discard_list = LIST_HEAD_INIT(discard_list); bool unplug = false; int result = 0; ENTRY; @@ -2850,7 +2907,7 @@ int osc_cache_writeback_range(const struct lu_env *env, struct osc_object *obj, case OES_CACHE: result += ext->oe_nr_pages; if (!discard) { - cfs_list_t *list = NULL; + struct list_head *list = NULL; if (hp) { EASSERT(!ext->oe_hp, ext); ext->oe_hp = 1; @@ -2860,7 +2917,7 @@ int osc_cache_writeback_range(const struct lu_env *env, struct osc_object *obj, list = &obj->oo_urgent_exts; } if (list != NULL) - cfs_list_move_tail(&ext->oe_link, list); + list_move_tail(&ext->oe_link, list); unplug = true; } else { /* the only discarder is lock cancelling, so @@ -2868,8 +2925,8 @@ int osc_cache_writeback_range(const struct lu_env *env, struct osc_object *obj, EASSERT(ext->oe_start >= start && ext->oe_max_end <= end, ext); osc_extent_state_set(ext, OES_LOCKING); - ext->oe_owner = cfs_current(); - cfs_list_move_tail(&ext->oe_link, + ext->oe_owner = current; + list_move_tail(&ext->oe_link, &discard_list); osc_update_pending(obj, OBD_BRW_WRITE, -ext->oe_nr_pages); @@ -2894,14 +2951,14 @@ int osc_cache_writeback_range(const struct lu_env *env, struct osc_object *obj, } osc_object_unlock(obj); - LASSERT(ergo(!discard, cfs_list_empty(&discard_list))); - if (!cfs_list_empty(&discard_list)) { + LASSERT(ergo(!discard, list_empty(&discard_list))); + if (!list_empty(&discard_list)) { struct osc_extent *tmp; int rc; osc_list_maint(osc_cli(obj), obj); - cfs_list_for_each_entry_safe(ext, tmp, &discard_list, oe_link) { - cfs_list_del_init(&ext->oe_link); + list_for_each_entry_safe(ext, tmp, &discard_list, oe_link) { + list_del_init(&ext->oe_link); EASSERT(ext->oe_state == OES_LOCKING, ext); /* Discard caching pages. We don't actually write this @@ -2929,8 +2986,204 @@ int osc_cache_writeback_range(const struct lu_env *env, struct osc_object *obj, result = rc; } - OSC_IO_DEBUG(obj, "cache page out.\n"); + OSC_IO_DEBUG(obj, "pageout [%lu, %lu], %d.\n", start, end, result); + RETURN(result); +} + +/** + * Returns a list of pages by a given [start, end] of \a obj. + * + * \param resched If not NULL, then we give up before hogging CPU for too + * long and set *resched = 1, in that case caller should implement a retry + * logic. + * + * Gang tree lookup (radix_tree_gang_lookup()) optimization is absolutely + * crucial in the face of [offset, EOF] locks. + * + * Return at least one page in @queue unless there is no covered page. + */ +int osc_page_gang_lookup(const struct lu_env *env, struct cl_io *io, + struct osc_object *osc, pgoff_t start, pgoff_t end, + osc_page_gang_cbt cb, void *cbdata) +{ + struct osc_page *ops; + void **pvec; + pgoff_t idx; + unsigned int nr; + unsigned int i; + unsigned int j; + int res = CLP_GANG_OKAY; + bool tree_lock = true; + ENTRY; + + idx = start; + pvec = osc_env_info(env)->oti_pvec; + spin_lock(&osc->oo_tree_lock); + while ((nr = radix_tree_gang_lookup(&osc->oo_tree, pvec, + idx, OTI_PVEC_SIZE)) > 0) { + struct cl_page *page; + bool end_of_region = false; + + for (i = 0, j = 0; i < nr; ++i) { + ops = pvec[i]; + pvec[i] = NULL; + + idx = osc_index(ops); + if (idx > end) { + end_of_region = true; + break; + } + + page = ops->ops_cl.cpl_page; + LASSERT(page->cp_type == CPT_CACHEABLE); + if (page->cp_state == CPS_FREEING) + continue; + + cl_page_get(page); + lu_ref_add_atomic(&page->cp_reference, + "gang_lookup", current); + pvec[j++] = ops; + } + ++idx; + + /* + * Here a delicate locking dance is performed. Current thread + * holds a reference to a page, but has to own it before it + * can be placed into queue. Owning implies waiting, so + * radix-tree lock is to be released. After a wait one has to + * check that pages weren't truncated (cl_page_own() returns + * error in the latter case). + */ + spin_unlock(&osc->oo_tree_lock); + tree_lock = false; + + for (i = 0; i < j; ++i) { + ops = pvec[i]; + if (res == CLP_GANG_OKAY) + res = (*cb)(env, io, ops, cbdata); + + page = ops->ops_cl.cpl_page; + lu_ref_del(&page->cp_reference, "gang_lookup", current); + cl_page_put(env, page); + } + if (nr < OTI_PVEC_SIZE || end_of_region) + break; + + if (res == CLP_GANG_OKAY && need_resched()) + res = CLP_GANG_RESCHED; + if (res != CLP_GANG_OKAY) + break; + + spin_lock(&osc->oo_tree_lock); + tree_lock = true; + } + if (tree_lock) + spin_unlock(&osc->oo_tree_lock); + RETURN(res); +} + +/** + * Check if page @page is covered by an extra lock or discard it. + */ +static int check_and_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 osc_object *osc = cbdata; + pgoff_t index; + + index = osc_index(ops); + if (index >= info->oti_fn_index) { + struct ldlm_lock *tmp; + struct cl_page *page = ops->ops_cl.cpl_page; + + /* refresh non-overlapped index */ + 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 = cl_index(osc2cl(osc), end + 1); + if (end == OBD_OBJECT_EOF) + info->oti_fn_index = CL_PAGE_EOF; + LDLM_LOCK_PUT(tmp); + } else if (cl_page_own(env, io, page) == 0) { + /* discard the page */ + cl_page_discard(env, io, page); + cl_page_disown(env, io, page); + } else { + LASSERT(page->cp_state == CPS_FREEING); + } + } + + info->oti_next_index = index + 1; + return CLP_GANG_OKAY; +} + +static int 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; + + /* 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)))); + + /* discard the page */ + cl_page_discard(env, io, page); + cl_page_disown(env, io, page); + } else { + LASSERT(page->cp_state == CPS_FREEING); + } + + return CLP_GANG_OKAY; +} + +/** + * Discard pages protected by the given lock. This function traverses radix + * tree to find all covering pages and discard them. If a page is being covered + * by other locks, it should remain in cache. + * + * 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_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; + osc_page_gang_cbt cb; + int res; + int result; + + ENTRY; + + 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 = 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, osc, + info->oti_next_index, end, cb, osc); + if (info->oti_next_index > end) + break; + + if (res == CLP_GANG_RESCHED) + cond_resched(); + } while (res != CLP_GANG_OKAY); +out: + cl_io_fini(env, io); RETURN(result); } + /** @} osc */