projects / deliverable / linux.git / blob
? search:


d06129e8d96596bd1891e3d8079450ac5cae7b28
[deliverable/linux.git] / drivers / staging / lustre / lustre / llite / statahead.c
1 /*
2 * GPL HEADER START
3 *
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19 *
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
22 * have any questions.
23 *
24 * GPL HEADER END
25 */
26 /*
27 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
29 *
30 * Copyright (c) 2011, 2015, Intel Corporation.
31 */
32 /*
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
35 */
36
37 #include <linux/fs.h>
38 #include <linux/sched.h>
39 #include <linux/mm.h>
40 #include <linux/highmem.h>
41 #include <linux/pagemap.h>
42
43 #define DEBUG_SUBSYSTEM S_LLITE
44
45 #include "../include/obd_support.h"
46 #include "../include/lustre_lite.h"
47 #include "../include/lustre_dlm.h"
48 #include "llite_internal.h"
49
50 #define SA_OMITTED_ENTRY_MAX 8ULL
51
52 enum se_stat {
53 /** negative values are for error cases */
54 SA_ENTRY_INIT = 0, /** init entry */
55 SA_ENTRY_SUCC = 1, /** stat succeed */
56 SA_ENTRY_INVA = 2, /** invalid entry */
57 SA_ENTRY_DEST = 3, /** entry to be destroyed */
58 };
59
60 struct ll_sa_entry {
61 /* link into sai->sai_entries */
62 struct list_head se_link;
63 /* link into sai->sai_entries_{received,stated} */
64 struct list_head se_list;
65 /* link into sai hash table locally */
66 struct list_head se_hash;
67 /* entry reference count */
68 atomic_t se_refcount;
69 /* entry index in the sai */
70 __u64 se_index;
71 /* low layer ldlm lock handle */
72 __u64 se_handle;
73 /* entry status */
74 enum se_stat se_stat;
75 /* entry size, contains name */
76 int se_size;
77 /* pointer to async getattr enqueue info */
78 struct md_enqueue_info *se_minfo;
79 /* pointer to the async getattr request */
80 struct ptlrpc_request *se_req;
81 /* pointer to the target inode */
82 struct inode *se_inode;
83 /* entry name */
84 struct qstr se_qstr;
85 };
86
87 static unsigned int sai_generation;
88 static DEFINE_SPINLOCK(sai_generation_lock);
89
90 /*
91 * The entry only can be released by the caller, it is necessary to hold lock.
92 */
93 static inline int ll_sa_entry_stated(struct ll_sa_entry *entry)
94 {
95 smp_rmb();
96 return (entry->se_stat != SA_ENTRY_INIT);
97 }
98
99 static inline int ll_sa_entry_hash(int val)
100 {
101 return val & LL_SA_CACHE_MASK;
102 }
103
104 /*
105 * Insert entry to hash SA table.
106 */
107 static inline void
108 ll_sa_entry_enhash(struct ll_statahead_info *sai, struct ll_sa_entry *entry)
109 {
110 int i = ll_sa_entry_hash(entry->se_qstr.hash);
111
112 spin_lock(&sai->sai_cache_lock[i]);
113 list_add_tail(&entry->se_hash, &sai->sai_cache[i]);
114 spin_unlock(&sai->sai_cache_lock[i]);
115 }
116
117 /*
118 * Remove entry from SA table.
119 */
120 static inline void
121 ll_sa_entry_unhash(struct ll_statahead_info *sai, struct ll_sa_entry *entry)
122 {
123 int i = ll_sa_entry_hash(entry->se_qstr.hash);
124
125 spin_lock(&sai->sai_cache_lock[i]);
126 list_del_init(&entry->se_hash);
127 spin_unlock(&sai->sai_cache_lock[i]);
128 }
129
130 static inline int agl_should_run(struct ll_statahead_info *sai,
131 struct inode *inode)
132 {
133 return (inode && S_ISREG(inode->i_mode) && sai->sai_agl_valid);
134 }
135
136 static inline int sa_sent_full(struct ll_statahead_info *sai)
137 {
138 return atomic_read(&sai->sai_cache_count) >= sai->sai_max;
139 }
140
141 static inline int sa_received_empty(struct ll_statahead_info *sai)
142 {
143 return list_empty(&sai->sai_entries_received);
144 }
145
146 static inline int agl_list_empty(struct ll_statahead_info *sai)
147 {
148 return list_empty(&sai->sai_entries_agl);
149 }
150
151 /**
152 * (1) hit ratio less than 80%
153 * or
154 * (2) consecutive miss more than 8
155 * then means low hit.
156 */
157 static inline int sa_low_hit(struct ll_statahead_info *sai)
158 {
159 return ((sai->sai_hit > 7 && sai->sai_hit < 4 * sai->sai_miss) ||
160 (sai->sai_consecutive_miss > 8));
161 }
162
163 /*
164 * If the given index is behind of statahead window more than
165 * SA_OMITTED_ENTRY_MAX, then it is old.
166 */
167 static inline int is_omitted_entry(struct ll_statahead_info *sai, __u64 index)
168 {
169 return ((__u64)sai->sai_max + index + SA_OMITTED_ENTRY_MAX <
170 sai->sai_index);
171 }
172
173 /*
174 * Insert it into sai_entries tail when init.
175 */
176 static struct ll_sa_entry *
177 ll_sa_entry_alloc(struct ll_statahead_info *sai, __u64 index,
178 const char *name, int len)
179 {
180 struct ll_inode_info *lli;
181 struct ll_sa_entry *entry;
182 int entry_size;
183 char *dname;
184
185 entry_size = sizeof(struct ll_sa_entry) + (len & ~3) + 4;
186 entry = kzalloc(entry_size, GFP_NOFS);
187 if (unlikely(!entry))
188 return ERR_PTR(-ENOMEM);
189
190 CDEBUG(D_READA, "alloc sa entry %.*s(%p) index %llu\n",
191 len, name, entry, index);
192
193 entry->se_index = index;
194
195 /*
196 * Statahead entry reference rules:
197 *
198 * 1) When statahead entry is initialized, its reference is set as 2.
199 * One reference is used by the directory scanner. When the scanner
200 * searches the statahead cache for the given name, it can perform
201 * lockless hash lookup (only the scanner can remove entry from hash
202 * list), and once found, it needn't to call "atomic_inc()" for the
203 * entry reference. So the performance is improved. After using the
204 * statahead entry, the scanner will call "atomic_dec()" to drop the
205 * reference held when initialization. If it is the last reference,
206 * the statahead entry will be freed.
207 *
208 * 2) All other threads, including statahead thread and ptlrpcd thread,
209 * when they process the statahead entry, the reference for target
210 * should be held to guarantee the entry will not be released by the
211 * directory scanner. After processing the entry, these threads will
212 * drop the entry reference. If it is the last reference, the entry
213 * will be freed.
214 *
215 * The second reference when initializes the statahead entry is used
216 * by the statahead thread, following the rule 2).
217 */
218 atomic_set(&entry->se_refcount, 2);
219 entry->se_stat = SA_ENTRY_INIT;
220 entry->se_size = entry_size;
221 dname = (char *)entry + sizeof(struct ll_sa_entry);
222 memcpy(dname, name, len);
223 dname[len] = 0;
224 entry->se_qstr.hash = full_name_hash(name, len);
225 entry->se_qstr.len = len;
226 entry->se_qstr.name = dname;
227
228 lli = ll_i2info(sai->sai_inode);
229 spin_lock(&lli->lli_sa_lock);
230 list_add_tail(&entry->se_link, &sai->sai_entries);
231 INIT_LIST_HEAD(&entry->se_list);
232 ll_sa_entry_enhash(sai, entry);
233 spin_unlock(&lli->lli_sa_lock);
234
235 atomic_inc(&sai->sai_cache_count);
236
237 return entry;
238 }
239
240 /*
241 * Used by the directory scanner to search entry with name.
242 *
243 * Only the caller can remove the entry from hash, so it is unnecessary to hold
244 * hash lock. It is caller's duty to release the init refcount on the entry, so
245 * it is also unnecessary to increase refcount on the entry.
246 */
247 static struct ll_sa_entry *
248 ll_sa_entry_get_byname(struct ll_statahead_info *sai, const struct qstr *qstr)
249 {
250 struct ll_sa_entry *entry;
251 int i = ll_sa_entry_hash(qstr->hash);
252
253 list_for_each_entry(entry, &sai->sai_cache[i], se_hash) {
254 if (entry->se_qstr.hash == qstr->hash &&
255 entry->se_qstr.len == qstr->len &&
256 memcmp(entry->se_qstr.name, qstr->name, qstr->len) == 0)
257 return entry;
258 }
259 return NULL;
260 }
261
262 /*
263 * Used by the async getattr request callback to find entry with index.
264 *
265 * Inside lli_sa_lock to prevent others to change the list during the search.
266 * It needs to increase entry refcount before returning to guarantee that the
267 * entry cannot be freed by others.
268 */
269 static struct ll_sa_entry *
270 ll_sa_entry_get_byindex(struct ll_statahead_info *sai, __u64 index)
271 {
272 struct ll_sa_entry *entry;
273
274 list_for_each_entry(entry, &sai->sai_entries, se_link) {
275 if (entry->se_index == index) {
276 LASSERT(atomic_read(&entry->se_refcount) > 0);
277 atomic_inc(&entry->se_refcount);
278 return entry;
279 }
280 if (entry->se_index > index)
281 break;
282 }
283 return NULL;
284 }
285
286 static void ll_sa_entry_cleanup(struct ll_statahead_info *sai,
287 struct ll_sa_entry *entry)
288 {
289 struct md_enqueue_info *minfo = entry->se_minfo;
290 struct ptlrpc_request *req = entry->se_req;
291
292 if (minfo) {
293 entry->se_minfo = NULL;
294 ll_intent_release(&minfo->mi_it);
295 iput(minfo->mi_dir);
296 kfree(minfo);
297 }
298
299 if (req) {
300 entry->se_req = NULL;
301 ptlrpc_req_finished(req);
302 }
303 }
304
305 static void ll_sa_entry_put(struct ll_statahead_info *sai,
306 struct ll_sa_entry *entry)
307 {
308 if (atomic_dec_and_test(&entry->se_refcount)) {
309 CDEBUG(D_READA, "free sa entry %.*s(%p) index %llu\n",
310 entry->se_qstr.len, entry->se_qstr.name, entry,
311 entry->se_index);
312
313 LASSERT(list_empty(&entry->se_link));
314 LASSERT(list_empty(&entry->se_list));
315 LASSERT(list_empty(&entry->se_hash));
316
317 ll_sa_entry_cleanup(sai, entry);
318 iput(entry->se_inode);
319
320 kfree(entry);
321 atomic_dec(&sai->sai_cache_count);
322 }
323 }
324
325 static inline void
326 do_sa_entry_fini(struct ll_statahead_info *sai, struct ll_sa_entry *entry)
327 {
328 struct ll_inode_info *lli = ll_i2info(sai->sai_inode);
329
330 LASSERT(!list_empty(&entry->se_hash));
331 LASSERT(!list_empty(&entry->se_link));
332
333 ll_sa_entry_unhash(sai, entry);
334
335 spin_lock(&lli->lli_sa_lock);
336 entry->se_stat = SA_ENTRY_DEST;
337 list_del_init(&entry->se_link);
338 if (likely(!list_empty(&entry->se_list)))
339 list_del_init(&entry->se_list);
340 spin_unlock(&lli->lli_sa_lock);
341
342 ll_sa_entry_put(sai, entry);
343 }
344
345 /*
346 * Delete it from sai_entries_stated list when fini.
347 */
348 static void
349 ll_sa_entry_fini(struct ll_statahead_info *sai, struct ll_sa_entry *entry)
350 {
351 struct ll_sa_entry *pos, *next;
352
353 if (entry)
354 do_sa_entry_fini(sai, entry);
355
356 /* drop old entry, only 'scanner' process does this, no need to lock */
357 list_for_each_entry_safe(pos, next, &sai->sai_entries, se_link) {
358 if (!is_omitted_entry(sai, pos->se_index))
359 break;
360 do_sa_entry_fini(sai, pos);
361 }
362 }
363
364 /*
365 * Inside lli_sa_lock.
366 */
367 static void
368 do_sa_entry_to_stated(struct ll_statahead_info *sai,
369 struct ll_sa_entry *entry, enum se_stat stat)
370 {
371 struct ll_sa_entry *se;
372 struct list_head *pos = &sai->sai_entries_stated;
373
374 if (!list_empty(&entry->se_list))
375 list_del_init(&entry->se_list);
376
377 list_for_each_entry_reverse(se, &sai->sai_entries_stated, se_list) {
378 if (se->se_index < entry->se_index) {
379 pos = &se->se_list;
380 break;
381 }
382 }
383
384 list_add(&entry->se_list, pos);
385 entry->se_stat = stat;
386 }
387
388 /*
389 * Move entry to sai_entries_stated and sort with the index.
390 * \retval 1 -- entry to be destroyed.
391 * \retval 0 -- entry is inserted into stated list.
392 */
393 static int
394 ll_sa_entry_to_stated(struct ll_statahead_info *sai,
395 struct ll_sa_entry *entry, enum se_stat stat)
396 {
397 struct ll_inode_info *lli = ll_i2info(sai->sai_inode);
398 int ret = 1;
399
400 ll_sa_entry_cleanup(sai, entry);
401
402 spin_lock(&lli->lli_sa_lock);
403 if (likely(entry->se_stat != SA_ENTRY_DEST)) {
404 do_sa_entry_to_stated(sai, entry, stat);
405 ret = 0;
406 }
407 spin_unlock(&lli->lli_sa_lock);
408
409 return ret;
410 }
411
412 /*
413 * Insert inode into the list of sai_entries_agl.
414 */
415 static void ll_agl_add(struct ll_statahead_info *sai,
416 struct inode *inode, int index)
417 {
418 struct ll_inode_info *child = ll_i2info(inode);
419 struct ll_inode_info *parent = ll_i2info(sai->sai_inode);
420 int added = 0;
421
422 spin_lock(&child->lli_agl_lock);
423 if (child->lli_agl_index == 0) {
424 child->lli_agl_index = index;
425 spin_unlock(&child->lli_agl_lock);
426
427 LASSERT(list_empty(&child->lli_agl_list));
428
429 igrab(inode);
430 spin_lock(&parent->lli_agl_lock);
431 if (list_empty(&sai->sai_entries_agl))
432 added = 1;
433 list_add_tail(&child->lli_agl_list, &sai->sai_entries_agl);
434 spin_unlock(&parent->lli_agl_lock);
435 } else {
436 spin_unlock(&child->lli_agl_lock);
437 }
438
439 if (added > 0)
440 wake_up(&sai->sai_agl_thread.t_ctl_waitq);
441 }
442
443 static struct ll_statahead_info *ll_sai_alloc(void)
444 {
445 struct ll_statahead_info *sai;
446 int i;
447
448 sai = kzalloc(sizeof(*sai), GFP_NOFS);
449 if (!sai)
450 return NULL;
451
452 atomic_set(&sai->sai_refcount, 1);
453
454 spin_lock(&sai_generation_lock);
455 sai->sai_generation = ++sai_generation;
456 if (unlikely(sai_generation == 0))
457 sai->sai_generation = ++sai_generation;
458 spin_unlock(&sai_generation_lock);
459
460 sai->sai_max = LL_SA_RPC_MIN;
461 sai->sai_index = 1;
462 init_waitqueue_head(&sai->sai_waitq);
463 init_waitqueue_head(&sai->sai_thread.t_ctl_waitq);
464 init_waitqueue_head(&sai->sai_agl_thread.t_ctl_waitq);
465
466 INIT_LIST_HEAD(&sai->sai_entries);
467 INIT_LIST_HEAD(&sai->sai_entries_received);
468 INIT_LIST_HEAD(&sai->sai_entries_stated);
469 INIT_LIST_HEAD(&sai->sai_entries_agl);
470
471 for (i = 0; i < LL_SA_CACHE_SIZE; i++) {
472 INIT_LIST_HEAD(&sai->sai_cache[i]);
473 spin_lock_init(&sai->sai_cache_lock[i]);
474 }
475 atomic_set(&sai->sai_cache_count, 0);
476
477 return sai;
478 }
479
480 static inline struct ll_statahead_info *
481 ll_sai_get(struct ll_statahead_info *sai)
482 {
483 atomic_inc(&sai->sai_refcount);
484 return sai;
485 }
486
487 static void ll_sai_put(struct ll_statahead_info *sai)
488 {
489 struct inode *inode = sai->sai_inode;
490 struct ll_inode_info *lli = ll_i2info(inode);
491
492 if (atomic_dec_and_lock(&sai->sai_refcount, &lli->lli_sa_lock)) {
493 struct ll_sa_entry *entry, *next;
494
495 if (unlikely(atomic_read(&sai->sai_refcount) > 0)) {
496 /* It is race case, the interpret callback just hold
497 * a reference count
498 */
499 spin_unlock(&lli->lli_sa_lock);
500 return;
501 }
502
503 LASSERT(!lli->lli_opendir_key);
504 LASSERT(thread_is_stopped(&sai->sai_thread));
505 LASSERT(thread_is_stopped(&sai->sai_agl_thread));
506
507 lli->lli_sai = NULL;
508 lli->lli_opendir_pid = 0;
509 spin_unlock(&lli->lli_sa_lock);
510
511 if (sai->sai_sent > sai->sai_replied)
512 CDEBUG(D_READA, "statahead for dir "DFID
513 " does not finish: [sent:%llu] [replied:%llu]\n",
514 PFID(&lli->lli_fid),
515 sai->sai_sent, sai->sai_replied);
516
517 list_for_each_entry_safe(entry, next, &sai->sai_entries,
518 se_link)
519 do_sa_entry_fini(sai, entry);
520
521 LASSERT(list_empty(&sai->sai_entries));
522 LASSERT(list_empty(&sai->sai_entries_received));
523 LASSERT(list_empty(&sai->sai_entries_stated));
524
525 LASSERT(atomic_read(&sai->sai_cache_count) == 0);
526 LASSERT(list_empty(&sai->sai_entries_agl));
527
528 iput(inode);
529 kfree(sai);
530 }
531 }
532
533 /* Do NOT forget to drop inode refcount when into sai_entries_agl. */
534 static void ll_agl_trigger(struct inode *inode, struct ll_statahead_info *sai)
535 {
536 struct ll_inode_info *lli = ll_i2info(inode);
537 __u64 index = lli->lli_agl_index;
538 int rc;
539
540 LASSERT(list_empty(&lli->lli_agl_list));
541
542 /* AGL maybe fall behind statahead with one entry */
543 if (is_omitted_entry(sai, index + 1)) {
544 lli->lli_agl_index = 0;
545 iput(inode);
546 return;
547 }
548
549 /* Someone is in glimpse (sync or async), do nothing. */
550 rc = down_write_trylock(&lli->lli_glimpse_sem);
551 if (rc == 0) {
552 lli->lli_agl_index = 0;
553 iput(inode);
554 return;
555 }
556
557 /*
558 * Someone triggered glimpse within 1 sec before.
559 * 1) The former glimpse succeeded with glimpse lock granted by OST, and
560 * if the lock is still cached on client, AGL needs to do nothing. If
561 * it is cancelled by other client, AGL maybe cannot obtain new lock
562 * for no glimpse callback triggered by AGL.
563 * 2) The former glimpse succeeded, but OST did not grant glimpse lock.
564 * Under such case, it is quite possible that the OST will not grant
565 * glimpse lock for AGL also.
566 * 3) The former glimpse failed, compared with other two cases, it is
567 * relative rare. AGL can ignore such case, and it will not muchly
568 * affect the performance.
569 */
570 if (lli->lli_glimpse_time != 0 &&
571 time_before(cfs_time_shift(-1), lli->lli_glimpse_time)) {
572 up_write(&lli->lli_glimpse_sem);
573 lli->lli_agl_index = 0;
574 iput(inode);
575 return;
576 }
577
578 CDEBUG(D_READA, "Handling (init) async glimpse: inode = "
579 DFID", idx = %llu\n", PFID(&lli->lli_fid), index);
580
581 cl_agl(inode);
582 lli->lli_agl_index = 0;
583 lli->lli_glimpse_time = cfs_time_current();
584 up_write(&lli->lli_glimpse_sem);
585
586 CDEBUG(D_READA, "Handled (init) async glimpse: inode= "
587 DFID", idx = %llu, rc = %d\n",
588 PFID(&lli->lli_fid), index, rc);
589
590 iput(inode);
591 }
592
593 static void ll_post_statahead(struct ll_statahead_info *sai)
594 {
595 struct inode *dir = sai->sai_inode;
596 struct inode *child;
597 struct ll_inode_info *lli = ll_i2info(dir);
598 struct ll_sa_entry *entry;
599 struct md_enqueue_info *minfo;
600 struct lookup_intent *it;
601 struct ptlrpc_request *req;
602 struct mdt_body *body;
603 int rc = 0;
604
605 spin_lock(&lli->lli_sa_lock);
606 if (unlikely(list_empty(&sai->sai_entries_received))) {
607 spin_unlock(&lli->lli_sa_lock);
608 return;
609 }
610 entry = list_entry(sai->sai_entries_received.next,
611 struct ll_sa_entry, se_list);
612 atomic_inc(&entry->se_refcount);
613 list_del_init(&entry->se_list);
614 spin_unlock(&lli->lli_sa_lock);
615
616 LASSERT(entry->se_handle != 0);
617
618 minfo = entry->se_minfo;
619 it = &minfo->mi_it;
620 req = entry->se_req;
621 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
622 if (!body) {
623 rc = -EFAULT;
624 goto out;
625 }
626
627 child = entry->se_inode;
628 if (!child) {
629 /*
630 * lookup.
631 */
632 LASSERT(fid_is_zero(&minfo->mi_data.op_fid2));
633
634 /* XXX: No fid in reply, this is probably cross-ref case.
635 * SA can't handle it yet.
636 */
637 if (body->valid & OBD_MD_MDS) {
638 rc = -EAGAIN;
639 goto out;
640 }
641 } else {
642 /*
643 * revalidate.
644 */
645 /* unlinked and re-created with the same name */
646 if (unlikely(!lu_fid_eq(&minfo->mi_data.op_fid2, &body->fid1))) {
647 entry->se_inode = NULL;
648 iput(child);
649 child = NULL;
650 }
651 }
652
653 it->d.lustre.it_lock_handle = entry->se_handle;
654 rc = md_revalidate_lock(ll_i2mdexp(dir), it, ll_inode2fid(dir), NULL);
655 if (rc != 1) {
656 rc = -EAGAIN;
657 goto out;
658 }
659
660 rc = ll_prep_inode(&child, req, dir->i_sb, it);
661 if (rc)
662 goto out;
663
664 CDEBUG(D_DLMTRACE, "%s: setting l_data to inode "DFID"%p\n",
665 ll_get_fsname(child->i_sb, NULL, 0),
666 PFID(ll_inode2fid(child)), child);
667 ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, child, it, NULL);
668
669 entry->se_inode = child;
670
671 if (agl_should_run(sai, child))
672 ll_agl_add(sai, child, entry->se_index);
673
674 out:
675 /* The "ll_sa_entry_to_stated()" will drop related ldlm ibits lock
676 * reference count by calling "ll_intent_drop_lock()" in spite of the
677 * above operations failed or not. Do not worry about calling
678 * "ll_intent_drop_lock()" more than once.
679 */
680 rc = ll_sa_entry_to_stated(sai, entry,
681 rc < 0 ? SA_ENTRY_INVA : SA_ENTRY_SUCC);
682 if (rc == 0 && entry->se_index == sai->sai_index_wait)
683 wake_up(&sai->sai_waitq);
684 ll_sa_entry_put(sai, entry);
685 }
686
687 static int ll_statahead_interpret(struct ptlrpc_request *req,
688 struct md_enqueue_info *minfo, int rc)
689 {
690 struct lookup_intent *it = &minfo->mi_it;
691 struct inode *dir = minfo->mi_dir;
692 struct ll_inode_info *lli = ll_i2info(dir);
693 struct ll_statahead_info *sai = NULL;
694 struct ll_sa_entry *entry;
695 __u64 handle = 0;
696 int wakeup;
697
698 if (it_disposition(it, DISP_LOOKUP_NEG))
699 rc = -ENOENT;
700
701 if (rc == 0) {
702 /* release ibits lock ASAP to avoid deadlock when statahead
703 * thread enqueues lock on parent in readdir and another
704 * process enqueues lock on child with parent lock held, eg.
705 * unlink.
706 */
707 handle = it->d.lustre.it_lock_handle;
708 ll_intent_drop_lock(it);
709 }
710
711 spin_lock(&lli->lli_sa_lock);
712 /* stale entry */
713 if (unlikely(!lli->lli_sai ||
714 lli->lli_sai->sai_generation != minfo->mi_generation)) {
715 spin_unlock(&lli->lli_sa_lock);
716 rc = -ESTALE;
717 goto out;
718 } else {
719 sai = ll_sai_get(lli->lli_sai);
720 if (unlikely(!thread_is_running(&sai->sai_thread))) {
721 sai->sai_replied++;
722 spin_unlock(&lli->lli_sa_lock);
723 rc = -EBADFD;
724 goto out;
725 }
726
727 entry = ll_sa_entry_get_byindex(sai, minfo->mi_cbdata);
728 if (!entry) {
729 sai->sai_replied++;
730 spin_unlock(&lli->lli_sa_lock);
731 rc = -EIDRM;
732 goto out;
733 }
734
735 if (rc != 0) {
736 do_sa_entry_to_stated(sai, entry, SA_ENTRY_INVA);
737 wakeup = (entry->se_index == sai->sai_index_wait);
738 } else {
739 entry->se_minfo = minfo;
740 entry->se_req = ptlrpc_request_addref(req);
741 /* Release the async ibits lock ASAP to avoid deadlock
742 * when statahead thread tries to enqueue lock on parent
743 * for readpage and other tries to enqueue lock on child
744 * with parent's lock held, for example: unlink.
745 */
746 entry->se_handle = handle;
747 wakeup = list_empty(&sai->sai_entries_received);
748 list_add_tail(&entry->se_list,
749 &sai->sai_entries_received);
750 }
751 sai->sai_replied++;
752 spin_unlock(&lli->lli_sa_lock);
753
754 ll_sa_entry_put(sai, entry);
755 if (wakeup)
756 wake_up(&sai->sai_thread.t_ctl_waitq);
757 }
758
759 out:
760 if (rc != 0) {
761 ll_intent_release(it);
762 iput(dir);
763 kfree(minfo);
764 }
765 if (sai)
766 ll_sai_put(sai);
767 return rc;
768 }
769
770 static void sa_args_fini(struct md_enqueue_info *minfo,
771 struct ldlm_enqueue_info *einfo)
772 {
773 LASSERT(minfo && einfo);
774 iput(minfo->mi_dir);
775 kfree(minfo);
776 kfree(einfo);
777 }
778
779 /**
780 * prepare arguments for async stat RPC.
781 */
782 static int sa_args_init(struct inode *dir, struct inode *child,
783 struct ll_sa_entry *entry, struct md_enqueue_info **pmi,
784 struct ldlm_enqueue_info **pei)
785 {
786 const struct qstr *qstr = &entry->se_qstr;
787 struct ll_inode_info *lli = ll_i2info(dir);
788 struct md_enqueue_info *minfo;
789 struct ldlm_enqueue_info *einfo;
790 struct md_op_data *op_data;
791
792 einfo = kzalloc(sizeof(*einfo), GFP_NOFS);
793 if (!einfo)
794 return -ENOMEM;
795
796 minfo = kzalloc(sizeof(*minfo), GFP_NOFS);
797 if (!minfo) {
798 kfree(einfo);
799 return -ENOMEM;
800 }
801
802 op_data = ll_prep_md_op_data(&minfo->mi_data, dir, child, qstr->name,
803 qstr->len, 0, LUSTRE_OPC_ANY, NULL);
804 if (IS_ERR(op_data)) {
805 kfree(einfo);
806 kfree(minfo);
807 return PTR_ERR(op_data);
808 }
809
810 minfo->mi_it.it_op = IT_GETATTR;
811 minfo->mi_dir = igrab(dir);
812 minfo->mi_cb = ll_statahead_interpret;
813 minfo->mi_generation = lli->lli_sai->sai_generation;
814 minfo->mi_cbdata = entry->se_index;
815
816 einfo->ei_type = LDLM_IBITS;
817 einfo->ei_mode = it_to_lock_mode(&minfo->mi_it);
818 einfo->ei_cb_bl = ll_md_blocking_ast;
819 einfo->ei_cb_cp = ldlm_completion_ast;
820 einfo->ei_cb_gl = NULL;
821 einfo->ei_cbdata = NULL;
822
823 *pmi = minfo;
824 *pei = einfo;
825
826 return 0;
827 }
828
829 static int do_sa_lookup(struct inode *dir, struct ll_sa_entry *entry)
830 {
831 struct md_enqueue_info *minfo;
832 struct ldlm_enqueue_info *einfo;
833 int rc;
834
835 rc = sa_args_init(dir, NULL, entry, &minfo, &einfo);
836 if (rc)
837 return rc;
838
839 rc = md_intent_getattr_async(ll_i2mdexp(dir), minfo, einfo);
840 if (rc < 0)
841 sa_args_fini(minfo, einfo);
842
843 return rc;
844 }
845
846 /**
847 * similar to ll_revalidate_it().
848 * \retval 1 -- dentry valid
849 * \retval 0 -- will send stat-ahead request
850 * \retval others -- prepare stat-ahead request failed
851 */
852 static int do_sa_revalidate(struct inode *dir, struct ll_sa_entry *entry,
853 struct dentry *dentry)
854 {
855 struct inode *inode = d_inode(dentry);
856 struct lookup_intent it = { .it_op = IT_GETATTR,
857 .d.lustre.it_lock_handle = 0 };
858 struct md_enqueue_info *minfo;
859 struct ldlm_enqueue_info *einfo;
860 int rc;
861
862 if (unlikely(!inode))
863 return 1;
864
865 if (d_mountpoint(dentry))
866 return 1;
867
868 entry->se_inode = igrab(inode);
869 rc = md_revalidate_lock(ll_i2mdexp(dir), &it, ll_inode2fid(inode),
870 NULL);
871 if (rc == 1) {
872 entry->se_handle = it.d.lustre.it_lock_handle;
873 ll_intent_release(&it);
874 return 1;
875 }
876
877 rc = sa_args_init(dir, inode, entry, &minfo, &einfo);
878 if (rc) {
879 entry->se_inode = NULL;
880 iput(inode);
881 return rc;
882 }
883
884 rc = md_intent_getattr_async(ll_i2mdexp(dir), minfo, einfo);
885 if (rc < 0) {
886 entry->se_inode = NULL;
887 iput(inode);
888 sa_args_fini(minfo, einfo);
889 }
890
891 return rc;
892 }
893
894 static void ll_statahead_one(struct dentry *parent, const char *entry_name,
895 int entry_name_len)
896 {
897 struct inode *dir = d_inode(parent);
898 struct ll_inode_info *lli = ll_i2info(dir);
899 struct ll_statahead_info *sai = lli->lli_sai;
900 struct dentry *dentry = NULL;
901 struct ll_sa_entry *entry;
902 int rc;
903 int rc1;
904
905 entry = ll_sa_entry_alloc(sai, sai->sai_index, entry_name,
906 entry_name_len);
907 if (IS_ERR(entry))
908 return;
909
910 dentry = d_lookup(parent, &entry->se_qstr);
911 if (!dentry) {
912 rc = do_sa_lookup(dir, entry);
913 } else {
914 rc = do_sa_revalidate(dir, entry, dentry);
915 if (rc == 1 && agl_should_run(sai, d_inode(dentry)))
916 ll_agl_add(sai, d_inode(dentry), entry->se_index);
917
918 dput(dentry);
919 }
920
921 if (rc) {
922 rc1 = ll_sa_entry_to_stated(sai, entry,
923 rc < 0 ? SA_ENTRY_INVA : SA_ENTRY_SUCC);
924 if (rc1 == 0 && entry->se_index == sai->sai_index_wait)
925 wake_up(&sai->sai_waitq);
926 } else {
927 sai->sai_sent++;
928 }
929
930 sai->sai_index++;
931 /* drop one refcount on entry by ll_sa_entry_alloc */
932 ll_sa_entry_put(sai, entry);
933 }
934
935 static int ll_agl_thread(void *arg)
936 {
937 struct dentry *parent = arg;
938 struct inode *dir = d_inode(parent);
939 struct ll_inode_info *plli = ll_i2info(dir);
940 struct ll_inode_info *clli;
941 struct ll_sb_info *sbi = ll_i2sbi(dir);
942 struct ll_statahead_info *sai = ll_sai_get(plli->lli_sai);
943 struct ptlrpc_thread *thread = &sai->sai_agl_thread;
944 struct l_wait_info lwi = { 0 };
945
946 thread->t_pid = current_pid();
947 CDEBUG(D_READA, "agl thread started: sai %p, parent %pd\n",
948 sai, parent);
949
950 atomic_inc(&sbi->ll_agl_total);
951 spin_lock(&plli->lli_agl_lock);
952 sai->sai_agl_valid = 1;
953 if (thread_is_init(thread))
954 /* If someone else has changed the thread state
955 * (e.g. already changed to SVC_STOPPING), we can't just
956 * blindly overwrite that setting.
957 */
958 thread_set_flags(thread, SVC_RUNNING);
959 spin_unlock(&plli->lli_agl_lock);
960 wake_up(&thread->t_ctl_waitq);
961
962 while (1) {
963 l_wait_event(thread->t_ctl_waitq,
964 !list_empty(&sai->sai_entries_agl) ||
965 !thread_is_running(thread),
966 &lwi);
967
968 if (!thread_is_running(thread))
969 break;
970
971 spin_lock(&plli->lli_agl_lock);
972 /* The statahead thread maybe help to process AGL entries,
973 * so check whether list empty again.
974 */
975 if (!list_empty(&sai->sai_entries_agl)) {
976 clli = list_entry(sai->sai_entries_agl.next,
977 struct ll_inode_info, lli_agl_list);
978 list_del_init(&clli->lli_agl_list);
979 spin_unlock(&plli->lli_agl_lock);
980 ll_agl_trigger(&clli->lli_vfs_inode, sai);
981 } else {
982 spin_unlock(&plli->lli_agl_lock);
983 }
984 }
985
986 spin_lock(&plli->lli_agl_lock);
987 sai->sai_agl_valid = 0;
988 while (!list_empty(&sai->sai_entries_agl)) {
989 clli = list_entry(sai->sai_entries_agl.next,
990 struct ll_inode_info, lli_agl_list);
991 list_del_init(&clli->lli_agl_list);
992 spin_unlock(&plli->lli_agl_lock);
993 clli->lli_agl_index = 0;
994 iput(&clli->lli_vfs_inode);
995 spin_lock(&plli->lli_agl_lock);
996 }
997 thread_set_flags(thread, SVC_STOPPED);
998 spin_unlock(&plli->lli_agl_lock);
999 wake_up(&thread->t_ctl_waitq);
1000 ll_sai_put(sai);
1001 CDEBUG(D_READA, "agl thread stopped: sai %p, parent %pd\n",
1002 sai, parent);
1003 return 0;
1004 }
1005
1006 static void ll_start_agl(struct dentry *parent, struct ll_statahead_info *sai)
1007 {
1008 struct ptlrpc_thread *thread = &sai->sai_agl_thread;
1009 struct l_wait_info lwi = { 0 };
1010 struct ll_inode_info *plli;
1011 struct task_struct *task;
1012
1013 CDEBUG(D_READA, "start agl thread: sai %p, parent %pd\n",
1014 sai, parent);
1015
1016 plli = ll_i2info(d_inode(parent));
1017 task = kthread_run(ll_agl_thread, parent, "ll_agl_%u",
1018 plli->lli_opendir_pid);
1019 if (IS_ERR(task)) {
1020 CERROR("can't start ll_agl thread, rc: %ld\n", PTR_ERR(task));
1021 thread_set_flags(thread, SVC_STOPPED);
1022 return;
1023 }
1024
1025 l_wait_event(thread->t_ctl_waitq,
1026 thread_is_running(thread) || thread_is_stopped(thread),
1027 &lwi);
1028 }
1029
1030 static int ll_statahead_thread(void *arg)
1031 {
1032 struct dentry *parent = arg;
1033 struct inode *dir = d_inode(parent);
1034 struct ll_inode_info *plli = ll_i2info(dir);
1035 struct ll_inode_info *clli;
1036 struct ll_sb_info *sbi = ll_i2sbi(dir);
1037 struct ll_statahead_info *sai = ll_sai_get(plli->lli_sai);
1038 struct ptlrpc_thread *thread = &sai->sai_thread;
1039 struct ptlrpc_thread *agl_thread = &sai->sai_agl_thread;
1040 struct page *page;
1041 __u64 pos = 0;
1042 int first = 0;
1043 int rc = 0;
1044 struct ll_dir_chain chain;
1045 struct l_wait_info lwi = { 0 };
1046
1047 thread->t_pid = current_pid();
1048 CDEBUG(D_READA, "statahead thread starting: sai %p, parent %pd\n",
1049 sai, parent);
1050
1051 if (sbi->ll_flags & LL_SBI_AGL_ENABLED)
1052 ll_start_agl(parent, sai);
1053
1054 atomic_inc(&sbi->ll_sa_total);
1055 spin_lock(&plli->lli_sa_lock);
1056 if (thread_is_init(thread))
1057 /* If someone else has changed the thread state
1058 * (e.g. already changed to SVC_STOPPING), we can't just
1059 * blindly overwrite that setting.
1060 */
1061 thread_set_flags(thread, SVC_RUNNING);
1062 spin_unlock(&plli->lli_sa_lock);
1063 wake_up(&thread->t_ctl_waitq);
1064
1065 ll_dir_chain_init(&chain);
1066 page = ll_get_dir_page(dir, pos, &chain);
1067
1068 while (1) {
1069 struct lu_dirpage *dp;
1070 struct lu_dirent *ent;
1071
1072 if (IS_ERR(page)) {
1073 rc = PTR_ERR(page);
1074 CDEBUG(D_READA, "error reading dir "DFID" at %llu/%llu: [rc %d] [parent %u]\n",
1075 PFID(ll_inode2fid(dir)), pos, sai->sai_index,
1076 rc, plli->lli_opendir_pid);
1077 goto out;
1078 }
1079
1080 dp = page_address(page);
1081 for (ent = lu_dirent_start(dp); ent;
1082 ent = lu_dirent_next(ent)) {
1083 __u64 hash;
1084 int namelen;
1085 char *name;
1086
1087 hash = le64_to_cpu(ent->lde_hash);
1088 if (unlikely(hash < pos))
1089 /*
1090 * Skip until we find target hash value.
1091 */
1092 continue;
1093
1094 namelen = le16_to_cpu(ent->lde_namelen);
1095 if (unlikely(namelen == 0))
1096 /*
1097 * Skip dummy record.
1098 */
1099 continue;
1100
1101 name = ent->lde_name;
1102 if (name[0] == '.') {
1103 if (namelen == 1) {
1104 /*
1105 * skip "."
1106 */
1107 continue;
1108 } else if (name[1] == '.' && namelen == 2) {
1109 /*
1110 * skip ".."
1111 */
1112 continue;
1113 } else if (!sai->sai_ls_all) {
1114 /*
1115 * skip hidden files.
1116 */
1117 sai->sai_skip_hidden++;
1118 continue;
1119 }
1120 }
1121
1122 /*
1123 * don't stat-ahead first entry.
1124 */
1125 if (unlikely(++first == 1))
1126 continue;
1127
1128 keep_it:
1129 l_wait_event(thread->t_ctl_waitq,
1130 !sa_sent_full(sai) ||
1131 !list_empty(&sai->sai_entries_received) ||
1132 !list_empty(&sai->sai_entries_agl) ||
1133 !thread_is_running(thread),
1134 &lwi);
1135
1136 interpret_it:
1137 while (!list_empty(&sai->sai_entries_received))
1138 ll_post_statahead(sai);
1139
1140 if (unlikely(!thread_is_running(thread))) {
1141 ll_release_page(page, 0);
1142 rc = 0;
1143 goto out;
1144 }
1145
1146 /* If no window for metadata statahead, but there are
1147 * some AGL entries to be triggered, then try to help
1148 * to process the AGL entries.
1149 */
1150 if (sa_sent_full(sai)) {
1151 spin_lock(&plli->lli_agl_lock);
1152 while (!list_empty(&sai->sai_entries_agl)) {
1153 clli = list_entry(sai->sai_entries_agl.next,
1154 struct ll_inode_info, lli_agl_list);
1155 list_del_init(&clli->lli_agl_list);
1156 spin_unlock(&plli->lli_agl_lock);
1157 ll_agl_trigger(&clli->lli_vfs_inode,
1158 sai);
1159
1160 if (!list_empty(&sai->sai_entries_received))
1161 goto interpret_it;
1162
1163 if (unlikely(
1164 !thread_is_running(thread))) {
1165 ll_release_page(page, 0);
1166 rc = 0;
1167 goto out;
1168 }
1169
1170 if (!sa_sent_full(sai))
1171 goto do_it;
1172
1173 spin_lock(&plli->lli_agl_lock);
1174 }
1175 spin_unlock(&plli->lli_agl_lock);
1176
1177 goto keep_it;
1178 }
1179
1180 do_it:
1181 ll_statahead_one(parent, name, namelen);
1182 }
1183 pos = le64_to_cpu(dp->ldp_hash_end);
1184 if (pos == MDS_DIR_END_OFF) {
1185 /*
1186 * End of directory reached.
1187 */
1188 ll_release_page(page, 0);
1189 while (1) {
1190 l_wait_event(thread->t_ctl_waitq,
1191 !list_empty(&sai->sai_entries_received) ||
1192 sai->sai_sent == sai->sai_replied ||
1193 !thread_is_running(thread),
1194 &lwi);
1195
1196 while (!list_empty(&sai->sai_entries_received))
1197 ll_post_statahead(sai);
1198
1199 if (unlikely(!thread_is_running(thread))) {
1200 rc = 0;
1201 goto out;
1202 }
1203
1204 if (sai->sai_sent == sai->sai_replied &&
1205 list_empty(&sai->sai_entries_received))
1206 break;
1207 }
1208
1209 spin_lock(&plli->lli_agl_lock);
1210 while (!list_empty(&sai->sai_entries_agl) &&
1211 thread_is_running(thread)) {
1212 clli = list_entry(sai->sai_entries_agl.next,
1213 struct ll_inode_info, lli_agl_list);
1214 list_del_init(&clli->lli_agl_list);
1215 spin_unlock(&plli->lli_agl_lock);
1216 ll_agl_trigger(&clli->lli_vfs_inode, sai);
1217 spin_lock(&plli->lli_agl_lock);
1218 }
1219 spin_unlock(&plli->lli_agl_lock);
1220
1221 rc = 0;
1222 goto out;
1223 } else if (1) {
1224 /*
1225 * chain is exhausted.
1226 * Normal case: continue to the next page.
1227 */
1228 ll_release_page(page, le32_to_cpu(dp->ldp_flags) &
1229 LDF_COLLIDE);
1230 page = ll_get_dir_page(dir, pos, &chain);
1231 } else {
1232 LASSERT(le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1233 ll_release_page(page, 1);
1234 /*
1235 * go into overflow page.
1236 */
1237 }
1238 }
1239
1240 out:
1241 if (sai->sai_agl_valid) {
1242 spin_lock(&plli->lli_agl_lock);
1243 thread_set_flags(agl_thread, SVC_STOPPING);
1244 spin_unlock(&plli->lli_agl_lock);
1245 wake_up(&agl_thread->t_ctl_waitq);
1246
1247 CDEBUG(D_READA, "stop agl thread: sai %p pid %u\n",
1248 sai, (unsigned int)agl_thread->t_pid);
1249 l_wait_event(agl_thread->t_ctl_waitq,
1250 thread_is_stopped(agl_thread),
1251 &lwi);
1252 } else {
1253 /* Set agl_thread flags anyway. */
1254 thread_set_flags(&sai->sai_agl_thread, SVC_STOPPED);
1255 }
1256 ll_dir_chain_fini(&chain);
1257 spin_lock(&plli->lli_sa_lock);
1258 if (!list_empty(&sai->sai_entries_received)) {
1259 thread_set_flags(thread, SVC_STOPPING);
1260 spin_unlock(&plli->lli_sa_lock);
1261
1262 /* To release the resources held by received entries. */
1263 while (!list_empty(&sai->sai_entries_received))
1264 ll_post_statahead(sai);
1265
1266 spin_lock(&plli->lli_sa_lock);
1267 }
1268 thread_set_flags(thread, SVC_STOPPED);
1269 spin_unlock(&plli->lli_sa_lock);
1270 wake_up(&sai->sai_waitq);
1271 wake_up(&thread->t_ctl_waitq);
1272 ll_sai_put(sai);
1273 dput(parent);
1274 CDEBUG(D_READA, "statahead thread stopped: sai %p, parent %pd\n",
1275 sai, parent);
1276 return rc;
1277 }
1278
1279 /**
1280 * called in ll_file_release().
1281 */
1282 void ll_stop_statahead(struct inode *dir, void *key)
1283 {
1284 struct ll_inode_info *lli = ll_i2info(dir);
1285
1286 if (unlikely(!key))
1287 return;
1288
1289 spin_lock(&lli->lli_sa_lock);
1290 if (lli->lli_opendir_key != key || lli->lli_opendir_pid == 0) {
1291 spin_unlock(&lli->lli_sa_lock);
1292 return;
1293 }
1294
1295 lli->lli_opendir_key = NULL;
1296
1297 if (lli->lli_sai) {
1298 struct l_wait_info lwi = { 0 };
1299 struct ptlrpc_thread *thread = &lli->lli_sai->sai_thread;
1300
1301 if (!thread_is_stopped(thread)) {
1302 thread_set_flags(thread, SVC_STOPPING);
1303 spin_unlock(&lli->lli_sa_lock);
1304 wake_up(&thread->t_ctl_waitq);
1305
1306 CDEBUG(D_READA, "stop statahead thread: sai %p pid %u\n",
1307 lli->lli_sai, (unsigned int)thread->t_pid);
1308 l_wait_event(thread->t_ctl_waitq,
1309 thread_is_stopped(thread),
1310 &lwi);
1311 } else {
1312 spin_unlock(&lli->lli_sa_lock);
1313 }
1314
1315 /*
1316 * Put the ref which was held when first statahead_enter.
1317 * It maybe not the last ref for some statahead requests
1318 * maybe inflight.
1319 */
1320 ll_sai_put(lli->lli_sai);
1321 } else {
1322 lli->lli_opendir_pid = 0;
1323 spin_unlock(&lli->lli_sa_lock);
1324 }
1325 }
1326
1327 enum {
1328 /**
1329 * not first dirent, or is "."
1330 */
1331 LS_NONE_FIRST_DE = 0,
1332 /**
1333 * the first non-hidden dirent
1334 */
1335 LS_FIRST_DE,
1336 /**
1337 * the first hidden dirent, that is "."
1338 */
1339 LS_FIRST_DOT_DE
1340 };
1341
1342 static int is_first_dirent(struct inode *dir, struct dentry *dentry)
1343 {
1344 struct ll_dir_chain chain;
1345 const struct qstr *target = &dentry->d_name;
1346 struct page *page;
1347 __u64 pos = 0;
1348 int dot_de;
1349 int rc = LS_NONE_FIRST_DE;
1350
1351 ll_dir_chain_init(&chain);
1352 page = ll_get_dir_page(dir, pos, &chain);
1353
1354 while (1) {
1355 struct lu_dirpage *dp;
1356 struct lu_dirent *ent;
1357
1358 if (IS_ERR(page)) {
1359 struct ll_inode_info *lli = ll_i2info(dir);
1360
1361 rc = PTR_ERR(page);
1362 CERROR("error reading dir "DFID" at %llu: [rc %d] [parent %u]\n",
1363 PFID(ll_inode2fid(dir)), pos,
1364 rc, lli->lli_opendir_pid);
1365 break;
1366 }
1367
1368 dp = page_address(page);
1369 for (ent = lu_dirent_start(dp); ent;
1370 ent = lu_dirent_next(ent)) {
1371 __u64 hash;
1372 int namelen;
1373 char *name;
1374
1375 hash = le64_to_cpu(ent->lde_hash);
1376 /* The ll_get_dir_page() can return any page containing
1377 * the given hash which may be not the start hash.
1378 */
1379 if (unlikely(hash < pos))
1380 continue;
1381
1382 namelen = le16_to_cpu(ent->lde_namelen);
1383 if (unlikely(namelen == 0))
1384 /*
1385 * skip dummy record.
1386 */
1387 continue;
1388
1389 name = ent->lde_name;
1390 if (name[0] == '.') {
1391 if (namelen == 1)
1392 /*
1393 * skip "."
1394 */
1395 continue;
1396 else if (name[1] == '.' && namelen == 2)
1397 /*
1398 * skip ".."
1399 */
1400 continue;
1401 else
1402 dot_de = 1;
1403 } else {
1404 dot_de = 0;
1405 }
1406
1407 if (dot_de && target->name[0] != '.') {
1408 CDEBUG(D_READA, "%.*s skip hidden file %.*s\n",
1409 target->len, target->name,
1410 namelen, name);
1411 continue;
1412 }
1413
1414 if (target->len != namelen ||
1415 memcmp(target->name, name, namelen) != 0)
1416 rc = LS_NONE_FIRST_DE;
1417 else if (!dot_de)
1418 rc = LS_FIRST_DE;
1419 else
1420 rc = LS_FIRST_DOT_DE;
1421
1422 ll_release_page(page, 0);
1423 goto out;
1424 }
1425 pos = le64_to_cpu(dp->ldp_hash_end);
1426 if (pos == MDS_DIR_END_OFF) {
1427 /*
1428 * End of directory reached.
1429 */
1430 ll_release_page(page, 0);
1431 break;
1432 } else if (1) {
1433 /*
1434 * chain is exhausted
1435 * Normal case: continue to the next page.
1436 */
1437 ll_release_page(page, le32_to_cpu(dp->ldp_flags) &
1438 LDF_COLLIDE);
1439 page = ll_get_dir_page(dir, pos, &chain);
1440 } else {
1441 /*
1442 * go into overflow page.
1443 */
1444 LASSERT(le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1445 ll_release_page(page, 1);
1446 }
1447 }
1448
1449 out:
1450 ll_dir_chain_fini(&chain);
1451 return rc;
1452 }
1453
1454 static void
1455 ll_sai_unplug(struct ll_statahead_info *sai, struct ll_sa_entry *entry)
1456 {
1457 struct ptlrpc_thread *thread = &sai->sai_thread;
1458 struct ll_sb_info *sbi = ll_i2sbi(sai->sai_inode);
1459 int hit;
1460
1461 if (entry && entry->se_stat == SA_ENTRY_SUCC)
1462 hit = 1;
1463 else
1464 hit = 0;
1465
1466 ll_sa_entry_fini(sai, entry);
1467 if (hit) {
1468 sai->sai_hit++;
1469 sai->sai_consecutive_miss = 0;
1470 sai->sai_max = min(2 * sai->sai_max, sbi->ll_sa_max);
1471 } else {
1472 struct ll_inode_info *lli = ll_i2info(sai->sai_inode);
1473
1474 sai->sai_miss++;
1475 sai->sai_consecutive_miss++;
1476 if (sa_low_hit(sai) && thread_is_running(thread)) {
1477 atomic_inc(&sbi->ll_sa_wrong);
1478 CDEBUG(D_READA, "Statahead for dir " DFID " hit ratio too low: hit/miss %llu/%llu, sent/replied %llu/%llu, stopping statahead thread\n",
1479 PFID(&lli->lli_fid), sai->sai_hit,
1480 sai->sai_miss, sai->sai_sent,
1481 sai->sai_replied);
1482 spin_lock(&lli->lli_sa_lock);
1483 if (!thread_is_stopped(thread))
1484 thread_set_flags(thread, SVC_STOPPING);
1485 spin_unlock(&lli->lli_sa_lock);
1486 }
1487 }
1488
1489 if (!thread_is_stopped(thread))
1490 wake_up(&thread->t_ctl_waitq);
1491 }
1492
1493 /**
1494 * Start statahead thread if this is the first dir entry.
1495 * Otherwise if a thread is started already, wait it until it is ahead of me.
1496 * \retval 1 -- find entry with lock in cache, the caller needs to do
1497 * nothing.
1498 * \retval 0 -- find entry in cache, but without lock, the caller needs
1499 * refresh from MDS.
1500 * \retval others -- the caller need to process as non-statahead.
1501 */
1502 int do_statahead_enter(struct inode *dir, struct dentry **dentryp,
1503 int only_unplug)
1504 {
1505 struct ll_inode_info *lli = ll_i2info(dir);
1506 struct ll_statahead_info *sai = lli->lli_sai;
1507 struct dentry *parent;
1508 struct ll_sa_entry *entry;
1509 struct ptlrpc_thread *thread;
1510 struct l_wait_info lwi = { 0 };
1511 struct task_struct *task;
1512 int rc = 0;
1513 struct ll_inode_info *plli;
1514
1515 LASSERT(lli->lli_opendir_pid == current_pid());
1516
1517 if (sai) {
1518 thread = &sai->sai_thread;
1519 if (unlikely(thread_is_stopped(thread) &&
1520 list_empty(&sai->sai_entries_stated))) {
1521 /* to release resource */
1522 ll_stop_statahead(dir, lli->lli_opendir_key);
1523 return -EAGAIN;
1524 }
1525
1526 if ((*dentryp)->d_name.name[0] == '.') {
1527 if (sai->sai_ls_all ||
1528 sai->sai_miss_hidden >= sai->sai_skip_hidden) {
1529 /*
1530 * Hidden dentry is the first one, or statahead
1531 * thread does not skip so many hidden dentries
1532 * before "sai_ls_all" enabled as below.
1533 */
1534 } else {
1535 if (!sai->sai_ls_all)
1536 /*
1537 * It maybe because hidden dentry is not
1538 * the first one, "sai_ls_all" was not
1539 * set, then "ls -al" missed. Enable
1540 * "sai_ls_all" for such case.
1541 */
1542 sai->sai_ls_all = 1;
1543
1544 /*
1545 * Such "getattr" has been skipped before
1546 * "sai_ls_all" enabled as above.
1547 */
1548 sai->sai_miss_hidden++;
1549 return -EAGAIN;
1550 }
1551 }
1552
1553 entry = ll_sa_entry_get_byname(sai, &(*dentryp)->d_name);
1554 if (!entry || only_unplug) {
1555 ll_sai_unplug(sai, entry);
1556 return entry ? 1 : -EAGAIN;
1557 }
1558
1559 if (!ll_sa_entry_stated(entry)) {
1560 sai->sai_index_wait = entry->se_index;
1561 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(30), NULL,
1562 LWI_ON_SIGNAL_NOOP, NULL);
1563 rc = l_wait_event(sai->sai_waitq,
1564 ll_sa_entry_stated(entry) ||
1565 thread_is_stopped(thread),
1566 &lwi);
1567 if (rc < 0) {
1568 ll_sai_unplug(sai, entry);
1569 return -EAGAIN;
1570 }
1571 }
1572
1573 if (entry->se_stat == SA_ENTRY_SUCC && entry->se_inode) {
1574 struct inode *inode = entry->se_inode;
1575 struct lookup_intent it = { .it_op = IT_GETATTR,
1576 .d.lustre.it_lock_handle =
1577 entry->se_handle };
1578 __u64 bits;
1579
1580 rc = md_revalidate_lock(ll_i2mdexp(dir), &it,
1581 ll_inode2fid(inode), &bits);
1582 if (rc == 1) {
1583 if (!d_inode(*dentryp)) {
1584 struct dentry *alias;
1585
1586 alias = ll_splice_alias(inode,
1587 *dentryp);
1588 if (IS_ERR(alias)) {
1589 ll_sai_unplug(sai, entry);
1590 return PTR_ERR(alias);
1591 }
1592 *dentryp = alias;
1593 } else if (d_inode(*dentryp) != inode) {
1594 /* revalidate, but inode is recreated */
1595 CDEBUG(D_READA, "%s: stale dentry %pd inode "DFID", statahead inode "DFID"\n",
1596 ll_get_fsname(d_inode(*dentryp)->i_sb, NULL, 0),
1597 *dentryp,
1598 PFID(ll_inode2fid(d_inode(*dentryp))),
1599 PFID(ll_inode2fid(inode)));
1600 ll_sai_unplug(sai, entry);
1601 return -ESTALE;
1602 } else {
1603 iput(inode);
1604 }
1605 entry->se_inode = NULL;
1606
1607 if ((bits & MDS_INODELOCK_LOOKUP) &&
1608 d_lustre_invalid(*dentryp))
1609 d_lustre_revalidate(*dentryp);
1610 ll_intent_release(&it);
1611 }
1612 }
1613
1614 ll_sai_unplug(sai, entry);
1615 return rc;
1616 }
1617
1618 /* I am the "lli_opendir_pid" owner, only me can set "lli_sai". */
1619 rc = is_first_dirent(dir, *dentryp);
1620 if (rc == LS_NONE_FIRST_DE) {
1621 /* It is not "ls -{a}l" operation, no need statahead for it. */
1622 rc = -EAGAIN;
1623 goto out;
1624 }
1625
1626 sai = ll_sai_alloc();
1627 if (!sai) {
1628 rc = -ENOMEM;
1629 goto out;
1630 }
1631
1632 sai->sai_ls_all = (rc == LS_FIRST_DOT_DE);
1633 sai->sai_inode = igrab(dir);
1634 if (unlikely(!sai->sai_inode)) {
1635 CWARN("Do not start stat ahead on dying inode "DFID"\n",
1636 PFID(&lli->lli_fid));
1637 rc = -ESTALE;
1638 goto out;
1639 }
1640
1641 /* get parent reference count here, and put it in ll_statahead_thread */
1642 parent = dget((*dentryp)->d_parent);
1643 if (unlikely(sai->sai_inode != d_inode(parent))) {
1644 struct ll_inode_info *nlli = ll_i2info(d_inode(parent));
1645
1646 CWARN("Race condition, someone changed %pd just now: old parent "DFID", new parent "DFID"\n",
1647 *dentryp,
1648 PFID(&lli->lli_fid), PFID(&nlli->lli_fid));
1649 dput(parent);
1650 iput(sai->sai_inode);
1651 rc = -EAGAIN;
1652 goto out;
1653 }
1654
1655 CDEBUG(D_READA, "start statahead thread: sai %p, parent %pd\n",
1656 sai, parent);
1657
1658 /* The sai buffer already has one reference taken at allocation time,
1659 * but as soon as we expose the sai by attaching it to the lli that
1660 * default reference can be dropped by another thread calling
1661 * ll_stop_statahead. We need to take a local reference to protect
1662 * the sai buffer while we intend to access it.
1663 */
1664 ll_sai_get(sai);
1665 lli->lli_sai = sai;
1666
1667 plli = ll_i2info(d_inode(parent));
1668 task = kthread_run(ll_statahead_thread, parent, "ll_sa_%u",
1669 plli->lli_opendir_pid);
1670 thread = &sai->sai_thread;
1671 if (IS_ERR(task)) {
1672 rc = PTR_ERR(task);
1673 CERROR("can't start ll_sa thread, rc: %d\n", rc);
1674 dput(parent);
1675 lli->lli_opendir_key = NULL;
1676 thread_set_flags(thread, SVC_STOPPED);
1677 thread_set_flags(&sai->sai_agl_thread, SVC_STOPPED);
1678 /* Drop both our own local reference and the default
1679 * reference from allocation time.
1680 */
1681 ll_sai_put(sai);
1682 ll_sai_put(sai);
1683 LASSERT(!lli->lli_sai);
1684 return -EAGAIN;
1685 }
1686
1687 l_wait_event(thread->t_ctl_waitq,
1688 thread_is_running(thread) || thread_is_stopped(thread),
1689 &lwi);
1690 ll_sai_put(sai);
1691
1692 /*
1693 * We don't stat-ahead for the first dirent since we are already in
1694 * lookup.
1695 */
1696 return -EAGAIN;
1697
1698 out:
1699 kfree(sai);
1700 spin_lock(&lli->lli_sa_lock);
1701 lli->lli_opendir_key = NULL;
1702 lli->lli_opendir_pid = 0;
1703 spin_unlock(&lli->lli_sa_lock);
1704 return rc;
1705 }
Linux kernel - Deliverables
This page took 0.067581 seconds and 5 git commands to generate.