1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 #include <linux/sched.h>
40 #include <linux/smp_lock.h>
41 #include <linux/highmem.h>
42 #include <linux/pagemap.h>
44 #define DEBUG_SUBSYSTEM S_LLITE
46 #include <obd_support.h>
47 #include <lustre_lite.h>
48 #include <lustre_dlm.h>
49 #include <linux/lustre_version.h>
50 #include "llite_internal.h"
53 struct list_head se_list;
54 unsigned int se_index;
56 struct ptlrpc_request *se_req;
57 struct md_enqueue_info *se_minfo;
61 SA_ENTRY_UNSTATED = 0,
65 static unsigned int sai_generation = 0;
66 static spinlock_t sai_generation_lock = SPIN_LOCK_UNLOCKED;
69 * Check whether first entry was stated already or not.
70 * No need to hold lli_lock, for:
71 * (1) it is me that remove entry from the list
72 * (2) the statahead thread only add new entry to the list
74 static int ll_sai_entry_stated(struct ll_statahead_info *sai)
76 struct ll_sai_entry *entry;
79 if (!list_empty(&sai->sai_entries_stated)) {
80 entry = list_entry(sai->sai_entries_stated.next,
81 struct ll_sai_entry, se_list);
82 if (entry->se_index == sai->sai_index_next)
88 static inline int sa_received_empty(struct ll_statahead_info *sai)
90 return list_empty(&sai->sai_entries_received);
93 static inline int sa_not_full(struct ll_statahead_info *sai)
95 return (sai->sai_index < sai->sai_hit + sai->sai_miss + sai->sai_max);
98 static inline int sa_is_running(struct ll_statahead_info *sai)
100 return !!(sai->sai_thread.t_flags & SVC_RUNNING);
103 static inline int sa_is_stopping(struct ll_statahead_info *sai)
105 return !!(sai->sai_thread.t_flags & SVC_STOPPING);
108 static inline int sa_is_stopped(struct ll_statahead_info *sai)
110 return !!(sai->sai_thread.t_flags & SVC_STOPPED);
114 * (1) hit ratio less than 80%
116 * (2) consecutive miss more than 8
118 static inline int sa_low_hit(struct ll_statahead_info *sai)
120 return ((sai->sai_hit < 4 * sai->sai_miss && sai->sai_hit > 7) ||
121 (sai->sai_consecutive_miss > 8));
125 * process the deleted entry's member and free the entry.
127 * (2) free md_enqueue_info
128 * (3) drop dentry's ref count
129 * (4) release request's ref count
131 static void ll_sai_entry_cleanup(struct ll_sai_entry *entry, int free)
133 struct ptlrpc_request *req = entry->se_req;
134 struct md_enqueue_info *minfo = entry->se_minfo;
138 struct dentry *dentry = minfo->mi_dentry;
139 struct lookup_intent *it = &minfo->mi_it;
141 entry->se_minfo = NULL;
142 ll_intent_release(it);
147 entry->se_req = NULL;
148 ptlrpc_req_finished(req);
151 LASSERT(list_empty(&entry->se_list));
158 static struct ll_statahead_info *ll_sai_alloc(void)
160 struct ll_statahead_info *sai;
166 spin_lock(&sai_generation_lock);
167 sai->sai_generation = ++sai_generation;
168 if (unlikely(sai_generation == 0))
169 sai->sai_generation = ++sai_generation;
170 spin_unlock(&sai_generation_lock);
171 atomic_set(&sai->sai_refcount, 1);
172 sai->sai_max = LL_SA_RPC_MIN;
173 cfs_waitq_init(&sai->sai_waitq);
174 cfs_waitq_init(&sai->sai_thread.t_ctl_waitq);
175 CFS_INIT_LIST_HEAD(&sai->sai_entries_sent);
176 CFS_INIT_LIST_HEAD(&sai->sai_entries_received);
177 CFS_INIT_LIST_HEAD(&sai->sai_entries_stated);
182 struct ll_statahead_info *ll_sai_get(struct ll_statahead_info *sai)
185 atomic_inc(&sai->sai_refcount);
189 static void ll_sai_put(struct ll_statahead_info *sai)
191 struct inode *inode = sai->sai_inode;
192 struct ll_inode_info *lli;
195 LASSERT(inode != NULL);
196 lli = ll_i2info(inode);
197 if (atomic_dec_and_lock(&sai->sai_refcount, &lli->lli_lock)) {
198 struct ll_sai_entry *entry, *next;
200 LASSERT(lli->lli_opendir_key == NULL);
202 lli->lli_opendir_pid = 0;
203 spin_unlock(&lli->lli_lock);
205 LASSERT(sa_is_stopped(sai));
207 if (sai->sai_sent > sai->sai_replied)
208 CDEBUG(D_READA,"statahead for dir "DFID" does not "
209 "finish: [sent:%u] [replied:%u]\n",
211 sai->sai_sent, sai->sai_replied);
213 list_for_each_entry_safe(entry, next, &sai->sai_entries_sent,
215 list_del_init(&entry->se_list);
216 ll_sai_entry_cleanup(entry, 1);
218 list_for_each_entry_safe(entry, next, &sai->sai_entries_received,
220 list_del_init(&entry->se_list);
221 ll_sai_entry_cleanup(entry, 1);
223 list_for_each_entry_safe(entry, next, &sai->sai_entries_stated,
225 list_del_init(&entry->se_list);
226 ll_sai_entry_cleanup(entry, 1);
235 * insert it into sai_entries_sent tail when init.
237 static struct ll_sai_entry *
238 ll_sai_entry_init(struct ll_statahead_info *sai, unsigned int index)
240 struct ll_inode_info *lli = ll_i2info(sai->sai_inode);
241 struct ll_sai_entry *entry;
244 OBD_ALLOC_PTR(entry);
246 RETURN(ERR_PTR(-ENOMEM));
248 CDEBUG(D_READA, "alloc sai entry %p index %u\n",
250 entry->se_index = index;
251 entry->se_stat = SA_ENTRY_UNSTATED;
253 spin_lock(&lli->lli_lock);
254 list_add_tail(&entry->se_list, &sai->sai_entries_sent);
255 spin_unlock(&lli->lli_lock);
261 * delete it from sai_entries_stated head when fini, it need not
262 * to process entry's member.
264 static void ll_sai_entry_fini(struct ll_statahead_info *sai)
266 struct ll_inode_info *lli = ll_i2info(sai->sai_inode);
267 struct ll_sai_entry *entry;
270 spin_lock(&lli->lli_lock);
271 sai->sai_index_next++;
272 if (likely(!list_empty(&sai->sai_entries_stated))) {
273 entry = list_entry(sai->sai_entries_stated.next,
274 struct ll_sai_entry, se_list);
275 if (entry->se_index < sai->sai_index_next) {
276 list_del(&entry->se_list);
280 LASSERT(sa_is_stopped(sai));
281 spin_unlock(&lli->lli_lock);
288 * \retval NULL : can not find the entry in sai_entries_sent with the index
289 * \retval entry: find the entry in sai_entries_sent with the index
291 static struct ll_sai_entry *
292 ll_sai_entry_set(struct ll_statahead_info *sai, unsigned int index, int stat,
293 struct ptlrpc_request *req, struct md_enqueue_info *minfo)
295 struct ll_sai_entry *entry;
298 if (!list_empty(&sai->sai_entries_sent)) {
299 list_for_each_entry(entry, &sai->sai_entries_sent, se_list) {
300 if (entry->se_index == index) {
301 entry->se_stat = stat;
302 entry->se_req = ptlrpc_request_addref(req);
303 entry->se_minfo = minfo;
305 } else if (entry->se_index > index)
314 * Move entry to sai_entries_received and
315 * insert it into sai_entries_received tail.
318 ll_sai_entry_to_received(struct ll_statahead_info *sai, struct ll_sai_entry *entry)
320 if (!list_empty(&entry->se_list))
321 list_del_init(&entry->se_list);
322 list_add_tail(&entry->se_list, &sai->sai_entries_received);
326 * Move entry to sai_entries_stated and
327 * sort with the index.
330 ll_sai_entry_to_stated(struct ll_statahead_info *sai, struct ll_sai_entry *entry)
332 struct ll_inode_info *lli = ll_i2info(sai->sai_inode);
333 struct ll_sai_entry *se;
336 ll_sai_entry_cleanup(entry, 0);
338 spin_lock(&lli->lli_lock);
339 if (!list_empty(&entry->se_list))
340 list_del_init(&entry->se_list);
342 if (unlikely(entry->se_index < sai->sai_index_next)) {
343 spin_unlock(&lli->lli_lock);
348 list_for_each_entry_reverse(se, &sai->sai_entries_stated, se_list) {
349 if (se->se_index < entry->se_index) {
350 list_add(&entry->se_list, &se->se_list);
351 spin_unlock(&lli->lli_lock);
357 * I am the first entry.
359 list_add(&entry->se_list, &sai->sai_entries_stated);
360 spin_unlock(&lli->lli_lock);
365 * finish lookup/revalidate.
367 static int do_statahead_interpret(struct ll_statahead_info *sai)
369 struct ll_inode_info *lli = ll_i2info(sai->sai_inode);
370 struct ll_sai_entry *entry;
371 struct ptlrpc_request *req;
372 struct md_enqueue_info *minfo;
373 struct dentry *dentry;
374 struct lookup_intent *it;
376 struct mdt_body *body;
379 spin_lock(&lli->lli_lock);
380 LASSERT(!sa_received_empty(sai));
381 entry = list_entry(sai->sai_entries_received.next, struct ll_sai_entry,
383 list_del_init(&entry->se_list);
384 spin_unlock(&lli->lli_lock);
386 if (unlikely(entry->se_index < sai->sai_index_next)) {
387 ll_sai_entry_cleanup(entry, 1);
391 if (entry->se_stat != SA_ENTRY_STATED)
392 GOTO(out, rc = entry->se_stat);
395 minfo = entry->se_minfo;
396 dentry = minfo->mi_dentry;
399 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
401 GOTO(out, rc = -EFAULT);
403 if (dentry->d_inode == NULL) {
407 struct dentry *save = dentry;
408 struct it_cb_data icbd = {
409 .icbd_parent = dentry->d_parent->d_inode,
410 .icbd_childp = &dentry
413 LASSERT(fid_is_zero(&minfo->mi_data.op_fid2));
416 * XXX: No fid in reply, this is probaly cross-ref case.
417 * SA can't handle it yet.
419 if (body->valid & OBD_MD_MDS)
420 GOTO(out, rc = -EAGAIN);
422 rc = ll_lookup_it_finish(req, it, &icbd);
425 * Here dentry->d_inode might be NULL,
426 * because the entry may have been removed before
427 * we start doing stat ahead.
429 ll_lookup_finish_locks(it, dentry);
431 if (dentry != save) {
432 minfo->mi_dentry = dentry;
439 if (!lu_fid_eq(&minfo->mi_data.op_fid2, &body->fid1)) {
440 ll_unhash_aliases(dentry->d_inode);
441 GOTO(out, rc = -EAGAIN);
444 rc = ll_revalidate_it_finish(req, it, dentry);
446 ll_unhash_aliases(dentry->d_inode);
450 spin_lock(&ll_lookup_lock);
451 spin_lock(&dcache_lock);
454 #ifdef DCACHE_LUSTRE_INVALID
455 dentry->d_flags &= ~DCACHE_LUSTRE_INVALID;
457 unlock_dentry(dentry);
458 d_rehash_cond(dentry, 0);
459 spin_unlock(&dcache_lock);
460 spin_unlock(&ll_lookup_lock);
462 ll_lookup_finish_locks(it, dentry);
467 if (likely(ll_sai_entry_to_stated(sai, entry)))
468 cfs_waitq_signal(&sai->sai_waitq);
472 static int ll_statahead_interpret(struct ptlrpc_request *req,
473 struct md_enqueue_info *minfo,
476 struct dentry *dentry = minfo->mi_dentry;
477 struct lookup_intent *it = &minfo->mi_it;
478 struct inode *dir = dentry->d_parent->d_inode;
479 struct ll_inode_info *lli = ll_i2info(dir);
480 struct ll_statahead_info *sai;
481 struct ll_sai_entry *entry;
484 CDEBUG(D_READA, "interpret statahead %.*s rc %d\n",
485 dentry->d_name.len, dentry->d_name.name, rc);
487 spin_lock(&lli->lli_lock);
488 if (unlikely(lli->lli_sai == NULL ||
489 lli->lli_sai->sai_generation != minfo->mi_generation)) {
490 spin_unlock(&lli->lli_lock);
491 ll_intent_release(it);
496 sai = ll_sai_get(lli->lli_sai);
497 if (rc || dir == NULL)
500 entry = ll_sai_entry_set(sai,
501 (unsigned int)(long)minfo->mi_cbdata,
502 rc ? SA_ENTRY_UNSTATED :
503 SA_ENTRY_STATED, req, minfo);
504 LASSERT(entry != NULL);
505 if (likely(sa_is_running(sai))) {
506 ll_sai_entry_to_received(sai, entry);
508 spin_unlock(&lli->lli_lock);
509 cfs_waitq_signal(&sai->sai_thread.t_ctl_waitq);
511 if (!list_empty(&entry->se_list))
512 list_del_init(&entry->se_list);
514 spin_unlock(&lli->lli_lock);
515 ll_sai_entry_cleanup(entry, 1);
522 static void sa_args_fini(struct md_enqueue_info *minfo,
523 struct ldlm_enqueue_info *einfo)
525 LASSERT(minfo && einfo);
526 capa_put(minfo->mi_data.op_capa1);
527 capa_put(minfo->mi_data.op_capa2);
533 * There is race condition between "capa_put" and "ll_statahead_interpret" for
534 * accessing "op_data.op_capa[1,2]" as following:
535 * "capa_put" releases "op_data.op_capa[1,2]"'s reference count after calling
536 * "md_intent_getattr_async". But "ll_statahead_interpret" maybe run first, and
537 * fill "op_data.op_capa[1,2]" as POISON, then cause "capa_put" access invalid
538 * "ocapa". So here reserve "op_data.op_capa[1,2]" in "pcapa" before calling
539 * "md_intent_getattr_async".
541 static int sa_args_init(struct inode *dir, struct dentry *dentry,
542 struct md_enqueue_info **pmi,
543 struct ldlm_enqueue_info **pei,
544 struct obd_capa **pcapa)
546 struct ll_inode_info *lli = ll_i2info(dir);
547 struct md_enqueue_info *minfo;
548 struct ldlm_enqueue_info *einfo;
549 struct md_op_data *op_data;
551 OBD_ALLOC_PTR(einfo);
555 OBD_ALLOC_PTR(minfo);
561 op_data = ll_prep_md_op_data(&minfo->mi_data, dir, dentry->d_inode,
562 dentry->d_name.name, dentry->d_name.len,
563 0, LUSTRE_OPC_ANY, NULL);
564 if (IS_ERR(op_data)) {
567 return PTR_ERR(op_data);
570 minfo->mi_it.it_op = IT_GETATTR;
571 minfo->mi_dentry = dentry;
572 minfo->mi_cb = ll_statahead_interpret;
573 minfo->mi_generation = lli->lli_sai->sai_generation;
574 minfo->mi_cbdata = (void *)(long)lli->lli_sai->sai_index;
576 einfo->ei_type = LDLM_IBITS;
577 einfo->ei_mode = it_to_lock_mode(&minfo->mi_it);
578 einfo->ei_cb_bl = ll_md_blocking_ast;
579 einfo->ei_cb_cp = ldlm_completion_ast;
580 einfo->ei_cb_gl = NULL;
581 einfo->ei_cbdata = NULL;
585 pcapa[0] = op_data->op_capa1;
586 pcapa[1] = op_data->op_capa2;
592 * similar to ll_lookup_it().
594 static int do_sa_lookup(struct inode *dir, struct dentry *dentry)
596 struct md_enqueue_info *minfo;
597 struct ldlm_enqueue_info *einfo;
598 struct obd_capa *capas[2];
602 rc = sa_args_init(dir, dentry, &minfo, &einfo, capas);
606 rc = md_intent_getattr_async(ll_i2mdexp(dir), minfo, einfo);
611 sa_args_fini(minfo, einfo);
618 * similar to ll_revalidate_it().
619 * \retval 1 -- dentry valid
620 * \retval 0 -- will send stat-ahead request
621 * \retval others -- prepare stat-ahead request failed
623 static int do_sa_revalidate(struct dentry *dentry)
625 struct inode *inode = dentry->d_inode;
626 struct inode *dir = dentry->d_parent->d_inode;
627 struct lookup_intent it = { .it_op = IT_GETATTR };
628 struct md_enqueue_info *minfo;
629 struct ldlm_enqueue_info *einfo;
630 struct obd_capa *capas[2];
637 if (d_mountpoint(dentry))
640 if (dentry == dentry->d_sb->s_root)
643 rc = md_revalidate_lock(ll_i2mdexp(dir), &it, ll_inode2fid(inode));
645 ll_intent_release(&it);
649 rc = sa_args_init(dir, dentry, &minfo, &einfo, capas);
653 rc = md_intent_getattr_async(ll_i2mdexp(dir), minfo, einfo);
658 sa_args_fini(minfo, einfo);
664 static inline void ll_name2qstr(struct qstr *q, const char *name, int namelen)
668 q->hash = full_name_hash(name, namelen);
671 static int ll_statahead_one(struct dentry *parent, const char* entry_name,
674 struct inode *dir = parent->d_inode;
675 struct ll_inode_info *lli = ll_i2info(dir);
676 struct ll_statahead_info *sai = lli->lli_sai;
678 struct dentry *dentry;
679 struct ll_sai_entry *se;
683 #ifdef DCACHE_LUSTRE_INVALID
684 if (parent->d_flags & DCACHE_LUSTRE_INVALID) {
686 if (d_unhashed(parent)) {
688 CDEBUG(D_READA, "parent dentry@%p %.*s is "
689 "invalid, skip statahead\n",
690 parent, parent->d_name.len, parent->d_name.name);
694 se = ll_sai_entry_init(sai, sai->sai_index);
698 ll_name2qstr(&name, entry_name, entry_name_len);
699 dentry = d_lookup(parent, &name);
701 dentry = d_alloc(parent, &name);
703 rc = do_sa_lookup(dir, dentry);
707 GOTO(out, rc = -ENOMEM);
710 rc = do_sa_revalidate(dentry);
719 CDEBUG(D_READA, "set sai entry %p index %u stat %d rc %d\n",
720 se, se->se_index, se->se_stat, rc);
722 if (ll_sai_entry_to_stated(sai, se))
723 cfs_waitq_signal(&sai->sai_waitq);
732 struct ll_sa_thread_args {
733 struct dentry *sta_parent;
737 static int ll_statahead_thread(void *arg)
739 struct ll_sa_thread_args *sta = arg;
740 struct dentry *parent = dget(sta->sta_parent);
741 struct inode *dir = parent->d_inode;
742 struct ll_inode_info *lli = ll_i2info(dir);
743 struct ll_sb_info *sbi = ll_i2sbi(dir);
744 struct ll_statahead_info *sai = ll_sai_get(lli->lli_sai);
745 struct ptlrpc_thread *thread = &sai->sai_thread;
750 struct ll_dir_chain chain;
755 snprintf(pname, 15, "ll_sa_%u", sta->sta_pid);
756 cfs_daemonize(pname);
760 spin_lock(&lli->lli_lock);
761 thread->t_flags = SVC_RUNNING;
762 spin_unlock(&lli->lli_lock);
763 cfs_waitq_signal(&thread->t_ctl_waitq);
764 CDEBUG(D_READA, "start doing statahead for %s\n", parent->d_name.name);
766 ll_dir_chain_init(&chain);
767 page = ll_get_dir_page(dir, pos, 0, &chain);
770 struct l_wait_info lwi = { 0 };
771 struct lu_dirpage *dp;
772 struct lu_dirent *ent;
776 CERROR("error reading dir "DFID" at "LPU64"/%u: rc %d\n",
777 PFID(ll_inode2fid(dir)), pos,
782 dp = page_address(page);
783 for (ent = lu_dirent_start(dp); ent != NULL;
784 ent = lu_dirent_next(ent)) {
785 char *name = ent->lde_name;
786 int namelen = le16_to_cpu(ent->lde_namelen);
794 if (name[0] == '.') {
800 } else if (name[1] == '.' && namelen == 2) {
805 } else if (!sai->sai_ls_all) {
809 sai->sai_skip_hidden++;
815 * don't stat-ahead first entry.
817 if (unlikely(!first)) {
823 l_wait_event(thread->t_ctl_waitq,
824 !sa_is_running(sai) || sa_not_full(sai) ||
825 !sa_received_empty(sai),
828 while (!sa_received_empty(sai) && sa_is_running(sai))
829 do_statahead_interpret(sai);
831 if (unlikely(!sa_is_running(sai))) {
836 if (!sa_not_full(sai))
838 * do not skip the current de.
842 rc = ll_statahead_one(parent, name, namelen);
848 pos = le64_to_cpu(dp->ldp_hash_end);
850 if (pos == DIR_END_OFF) {
852 * End of directory reached.
855 l_wait_event(thread->t_ctl_waitq,
856 !sa_is_running(sai) ||
857 !sa_received_empty(sai) ||
858 sai->sai_sent == sai->sai_replied,
860 if (!sa_received_empty(sai) &&
862 do_statahead_interpret(sai);
868 * chain is exhausted.
869 * Normal case: continue to the next page.
871 page = ll_get_dir_page(dir, pos, 1, &chain);
874 * go into overflow page.
881 ll_dir_chain_fini(&chain);
882 spin_lock(&lli->lli_lock);
883 thread->t_flags = SVC_STOPPED;
884 spin_unlock(&lli->lli_lock);
885 cfs_waitq_signal(&sai->sai_waitq);
886 cfs_waitq_signal(&thread->t_ctl_waitq);
889 CDEBUG(D_READA, "statahead thread stopped, pid %d\n",
895 * called in ll_file_release().
897 void ll_stop_statahead(struct inode *inode, void *key)
899 struct ll_inode_info *lli = ll_i2info(inode);
901 if (unlikely(key == NULL))
904 spin_lock(&lli->lli_lock);
905 if (lli->lli_opendir_key != key || lli->lli_opendir_pid == 0) {
906 spin_unlock(&lli->lli_lock);
910 lli->lli_opendir_key = NULL;
913 struct l_wait_info lwi = { 0 };
914 struct ptlrpc_thread *thread = &lli->lli_sai->sai_thread;
916 if (!sa_is_stopped(lli->lli_sai)) {
917 thread->t_flags = SVC_STOPPING;
918 spin_unlock(&lli->lli_lock);
919 cfs_waitq_signal(&thread->t_ctl_waitq);
921 CDEBUG(D_READA, "stopping statahead thread, pid %d\n",
923 l_wait_event(thread->t_ctl_waitq,
924 sa_is_stopped(lli->lli_sai),
927 spin_unlock(&lli->lli_lock);
931 * Put the ref which was held when first statahead_enter.
932 * It maybe not the last ref for some statahead requests
935 ll_sai_put(lli->lli_sai);
937 lli->lli_opendir_pid = 0;
938 spin_unlock(&lli->lli_lock);
944 * not first dirent, or is "."
946 LS_NONE_FIRST_DE = 0,
948 * the first non-hidden dirent
952 * the first hidden dirent, that is ".xxx
957 static int is_first_dirent(struct inode *dir, struct dentry *dentry)
959 struct ll_dir_chain chain;
960 struct qstr *target = &dentry->d_name;
964 int rc = LS_NONE_FIRST_DE;
967 ll_dir_chain_init(&chain);
968 page = ll_get_dir_page(dir, pos, 0, &chain);
971 struct lu_dirpage *dp;
972 struct lu_dirent *ent;
976 CERROR("error reading dir "DFID" at "LPU64": rc %d\n",
977 PFID(ll_inode2fid(dir)), pos, rc);
981 dp = page_address(page);
982 for (ent = lu_dirent_start(dp); ent != NULL;
983 ent = lu_dirent_next(ent)) {
984 char *name = ent->lde_name;
985 int namelen = le16_to_cpu(ent->lde_namelen);
993 if (name[0] == '.') {
999 else if (name[1] == '.' && namelen == 2)
1010 if (dot_de && target->name[0] != '.') {
1011 CDEBUG(D_READA, "%.*s skip hidden file %.*s\n",
1012 target->len, target->name,
1017 if (target->len == namelen &&
1018 memcmp(target->name, name, namelen) == 0)
1019 rc = LS_FIRST_DE + dot_de;
1021 rc = LS_NONE_FIRST_DE;
1025 pos = le64_to_cpu(dp->ldp_hash_end);
1027 if (pos == DIR_END_OFF) {
1029 * End of directory reached.
1034 * chain is exhausted
1035 * Normal case: continue to the next page.
1037 page = ll_get_dir_page(dir, pos, 1, &chain);
1040 * go into overflow page.
1047 ll_dir_chain_fini(&chain);
1052 * Start statahead thread if this is the first dir entry.
1053 * Otherwise if a thread is started already, wait it until it is ahead of me.
1054 * \retval 0 -- stat ahead thread process such dentry, for lookup, it miss
1055 * \retval 1 -- stat ahead thread process such dentry, for lookup, it hit
1056 * \retval -EEXIST -- stat ahead thread started, and this is the first dentry
1057 * \retval -EBADFD -- statahead thread exit and not dentry available
1058 * \retval others -- error
1060 int do_statahead_enter(struct inode *dir, struct dentry **dentryp, int lookup)
1062 struct ll_sb_info *sbi = ll_i2sbi(dir);
1063 struct ll_inode_info *lli = ll_i2info(dir);
1064 struct ll_statahead_info *sai = lli->lli_sai;
1065 struct ll_sa_thread_args sta;
1066 struct l_wait_info lwi = { 0 };
1070 LASSERT(lli->lli_opendir_pid == cfs_curproc_pid());
1073 if (unlikely(sa_is_stopped(sai) &&
1074 list_empty(&sai->sai_entries_stated)))
1077 if ((*dentryp)->d_name.name[0] == '.') {
1078 if (likely(sai->sai_ls_all ||
1079 sai->sai_miss_hidden >= sai->sai_skip_hidden)) {
1081 * Hidden dentry is the first one, or statahead
1082 * thread does not skip so many hidden dentries
1083 * before "sai_ls_all" enabled as below.
1086 if (!sai->sai_ls_all)
1088 * It maybe because hidden dentry is not
1089 * the first one, "sai_ls_all" was not
1090 * set, then "ls -al" missed. Enable
1091 * "sai_ls_all" for such case.
1093 sai->sai_ls_all = 1;
1096 * Such "getattr" has been skipped before
1097 * "sai_ls_all" enabled as above.
1099 sai->sai_miss_hidden++;
1104 if (ll_sai_entry_stated(sai)) {
1105 sbi->ll_sa_cached++;
1107 sbi->ll_sa_blocked++;
1109 * thread started already, avoid double-stat.
1111 l_wait_event(sai->sai_waitq,
1112 ll_sai_entry_stated(sai) || sa_is_stopped(sai),
1117 struct dentry *result;
1119 result = d_lookup((*dentryp)->d_parent,
1120 &(*dentryp)->d_name);
1122 LASSERT(result != *dentryp);
1123 /* BUG 16303: do not drop reference count for
1124 * "*dentryp", VFS will do that by itself. */
1130 * do nothing for revalidate.
1136 * I am the "lli_opendir_pid" owner, only me can set "lli_sai".
1138 LASSERT(lli->lli_sai == NULL);
1140 rc = is_first_dirent(dir, *dentryp);
1141 if (rc == LS_NONE_FIRST_DE) {
1143 * It is not "ls -{a}l" operation, no need statahead for it.
1145 spin_lock(&lli->lli_lock);
1146 lli->lli_opendir_key = NULL;
1147 lli->lli_opendir_pid = 0;
1148 spin_unlock(&lli->lli_lock);
1152 sai = ll_sai_alloc();
1156 sai->sai_ls_all = (rc == LS_FIRST_DOT_DE);
1157 sai->sai_inode = igrab(dir);
1158 if (unlikely(sai->sai_inode == NULL)) {
1159 CWARN("Do not start stat ahead on dying inode "DFID" .\n",
1160 PFID(&lli->lli_fid));
1165 LASSERT(sai->sai_inode == (*dentryp)->d_parent->d_inode);
1167 sta.sta_parent = (*dentryp)->d_parent;
1168 sta.sta_pid = cfs_curproc_pid();
1171 rc = cfs_kernel_thread(ll_statahead_thread, &sta, 0);
1173 CERROR("can't start ll_sa thread, rc: %d\n", rc);
1174 lli->lli_opendir_key = NULL;
1175 sai->sai_thread.t_flags = SVC_STOPPED;
1177 LASSERT(lli->lli_sai == NULL);
1181 l_wait_event(sai->sai_thread.t_ctl_waitq,
1182 sa_is_running(sai) || sa_is_stopped(sai),
1186 * We don't stat-ahead for the first dirent since we are already in
1187 * lookup, and -EEXIST also indicates that this is the first dirent.
1193 * update hit/miss count.
1195 int ll_statahead_exit(struct dentry *dentry, int result)
1197 struct dentry *parent = dentry->d_parent;
1198 struct ll_inode_info *lli = ll_i2info(parent->d_inode);
1199 struct ll_sb_info *sbi = ll_i2sbi(parent->d_inode);
1203 if (lli->lli_opendir_pid != cfs_curproc_pid())
1207 struct ll_statahead_info *sai = lli->lli_sai;
1212 sai->sai_consecutive_miss = 0;
1213 sai->sai_max = min(2 * sai->sai_max, sbi->ll_sa_max);
1217 sai->sai_consecutive_miss++;
1218 if (sa_low_hit(sai) && sa_is_running(sai)) {
1220 CDEBUG(D_READA, "statahead for dir %.*s hit "
1221 "ratio too low: hit/miss %u/%u, "
1222 "sent/replied %u/%u. stopping statahead "
1224 parent->d_name.len, parent->d_name.name,
1225 sai->sai_hit, sai->sai_miss,
1226 sai->sai_sent, sai->sai_replied,
1228 spin_lock(&lli->lli_lock);
1229 if (!sa_is_stopped(sai))
1230 sai->sai_thread.t_flags = SVC_STOPPING;
1231 spin_unlock(&lli->lli_lock);
1235 if (!sa_is_stopped(sai))
1236 cfs_waitq_signal(&sai->sai_thread.t_ctl_waitq);
1237 ll_sai_entry_fini(sai);
1238 rc = ll_statahead_mark(dentry);