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, 2013, 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 <obd_support.h>
  46 #include <lustre_dlm.h>
  47 #include "llite_internal.h"
  48
  49 #define SA_OMITTED_ENTRY_MAX 8ULL
  50
  51 typedef enum {
  52         /** negative values are for error cases */
  53         SA_ENTRY_INIT = 0,      /** init entry */
  54         SA_ENTRY_SUCC = 1,      /** stat succeed */
  55         SA_ENTRY_INVA = 2,      /** invalid entry */
  56         SA_ENTRY_DEST = 3,      /** entry to be destroyed */
  57 } se_stat_t;
  58
  59 struct ll_sa_entry {
  60         /* link into sai->sai_entries */
  61         struct list_head        se_link;
  62         /* link into sai->sai_entries_{received,stated} */
  63         struct list_head        se_list;
  64         /* link into sai hash table locally */
  65         struct list_head        se_hash;
  66         /* entry reference count */
  67         atomic_t                se_refcount;
  68         /* entry index in the sai */
  69         __u64                   se_index;
  70         /* low layer ldlm lock handle */
  71         __u64                   se_handle;
  72         /* entry status */
  73         se_stat_t               se_stat;
  74         /* entry size, contains name */
  75         int                     se_size;
  76         /* pointer to async getattr enqueue info */
  77         struct md_enqueue_info *se_minfo;
  78         /* pointer to the async getattr request */
  79         struct ptlrpc_request  *se_req;
  80         /* pointer to the target inode */
  81         struct inode           *se_inode;
  82         /* entry name */
  83         struct qstr             se_qstr;
  84 };
  85
  86 static unsigned int sai_generation = 0;
  87 static DEFINE_SPINLOCK(sai_generation_lock);
  88
  89 static inline int ll_sa_entry_unhashed(struct ll_sa_entry *entry)
  90 {
  91         return list_empty(&entry->se_hash);
  92 }
  93
  94 /*
  95  * The entry only can be released by the caller, it is necessary to hold lock.
  96  */
  97 static inline int ll_sa_entry_stated(struct ll_sa_entry *entry)
  98 {
  99         smp_rmb();
 100         return (entry->se_stat != SA_ENTRY_INIT);
 101 }
 102
 103 static inline int ll_sa_entry_hash(int val)
 104 {
 105         return val & LL_SA_CACHE_MASK;
 106 }
 107
 108 /*
 109  * Insert entry to hash SA table.
 110  */
 111 static inline void
 112 ll_sa_entry_enhash(struct ll_statahead_info *sai, struct ll_sa_entry *entry)
 113 {
 114         int i = ll_sa_entry_hash(entry->se_qstr.hash);
 115
 116         spin_lock(&sai->sai_cache_lock[i]);
 117         list_add_tail(&entry->se_hash, &sai->sai_cache[i]);
 118         spin_unlock(&sai->sai_cache_lock[i]);
 119 }
 120
 121 /*
 122  * Remove entry from SA table.
 123  */
 124 static inline void
 125 ll_sa_entry_unhash(struct ll_statahead_info *sai, struct ll_sa_entry *entry)
 126 {
 127         int i = ll_sa_entry_hash(entry->se_qstr.hash);
 128
 129         spin_lock(&sai->sai_cache_lock[i]);
 130         list_del_init(&entry->se_hash);
 131         spin_unlock(&sai->sai_cache_lock[i]);
 132 }
 133
 134 static inline int agl_should_run(struct ll_statahead_info *sai,
 135                                  struct inode *inode)
 136 {
 137         return (inode != NULL && S_ISREG(inode->i_mode) && sai->sai_agl_valid);
 138 }
 139
 140 static inline struct ll_sa_entry *
 141 sa_first_received_entry(struct ll_statahead_info *sai)
 142 {
 143         return list_entry(sai->sai_entries_received.next,
 144                           struct ll_sa_entry, se_list);
 145 }
 146
 147 static inline struct ll_inode_info *
 148 agl_first_entry(struct ll_statahead_info *sai)
 149 {
 150         return list_entry(sai->sai_entries_agl.next,
 151                           struct ll_inode_info, lli_agl_list);
 152 }
 153
 154 static inline int sa_sent_full(struct ll_statahead_info *sai)
 155 {
 156         return atomic_read(&sai->sai_cache_count) >= sai->sai_max;
 157 }
 158
 159 static inline int sa_received_empty(struct ll_statahead_info *sai)
 160 {
 161         return list_empty(&sai->sai_entries_received);
 162 }
 163
 164 static inline int agl_list_empty(struct ll_statahead_info *sai)
 165 {
 166         return list_empty(&sai->sai_entries_agl);
 167 }
 168
 169 /**
 170  * (1) hit ratio less than 80%
 171  * or
 172  * (2) consecutive miss more than 8
 173  * then means low hit.
 174  */
 175 static inline int sa_low_hit(struct ll_statahead_info *sai)
 176 {
 177         return ((sai->sai_hit > 7 && sai->sai_hit < 4 * sai->sai_miss) ||
 178                 (sai->sai_consecutive_miss > 8));
 179 }
 180
 181 /*
 182  * If the given index is behind of statahead window more than
 183  * SA_OMITTED_ENTRY_MAX, then it is old.
 184  */
 185 static inline int is_omitted_entry(struct ll_statahead_info *sai, __u64 index)
 186 {
 187         return ((__u64)sai->sai_max + index + SA_OMITTED_ENTRY_MAX <
 188                  sai->sai_index);
 189 }
 190
 191 /*
 192  * Insert it into sai_entries tail when init.
 193  */
 194 static struct ll_sa_entry *
 195 ll_sa_entry_alloc(struct ll_statahead_info *sai, __u64 index, const char *name,
 196                  int len)
 197 {
 198         struct ll_inode_info *lli;
 199         struct ll_sa_entry   *entry;
 200         int                   entry_size;
 201         char                 *dname;
 202         ENTRY;
 203
 204         entry_size = sizeof(struct ll_sa_entry) + (len & ~3) + 4;
 205         OBD_ALLOC(entry, entry_size);
 206         if (unlikely(entry == NULL))
 207                 RETURN(ERR_PTR(-ENOMEM));
 208
 209         CDEBUG(D_READA, "alloc sa entry %.*s(%p) index "LPU64"\n",
 210                len, name, entry, index);
 211
 212         entry->se_index = index;
 213
 214         /*
 215          * Statahead entry reference rules:
 216          *
 217          * 1) When statahead entry is initialized, its reference is set as 2.
 218          *    One reference is used by the directory scanner. When the scanner
 219          *    searches the statahead cache for the given name, it can perform
 220          *    lockless hash lookup (only the scanner can remove entry from hash
 221          *    list), and once found, it needn't to call "atomic_inc()" for the
 222          *    entry reference. So the performance is improved. After using the
 223          *    statahead entry, the scanner will call "atomic_dec()" to drop the
 224          *    reference held when initialization. If it is the last reference,
 225          *    the statahead entry will be freed.
 226          *
 227          * 2) All other threads, including statahead thread and ptlrpcd thread,
 228          *    when they process the statahead entry, the reference for target
 229          *    should be held to guarantee the entry will not be released by the
 230          *    directory scanner. After processing the entry, these threads will
 231          *    drop the entry reference. If it is the last reference, the entry
 232          *    will be freed.
 233          *
 234          *    The second reference when initializes the statahead entry is used
 235          *    by the statahead thread, following the rule 2).
 236          */
 237         atomic_set(&entry->se_refcount, 2);
 238         entry->se_stat = SA_ENTRY_INIT;
 239         entry->se_size = entry_size;
 240         dname = (char *)entry + sizeof(struct ll_sa_entry);
 241         memcpy(dname, name, len);
 242         dname[len] = 0;
 243         entry->se_qstr.hash = full_name_hash(name, len);
 244         entry->se_qstr.len = len;
 245         entry->se_qstr.name = dname;
 246
 247         lli = ll_i2info(sai->sai_inode);
 248         spin_lock(&lli->lli_sa_lock);
 249         list_add_tail(&entry->se_link, &sai->sai_entries);
 250         INIT_LIST_HEAD(&entry->se_list);
 251         ll_sa_entry_enhash(sai, entry);
 252         spin_unlock(&lli->lli_sa_lock);
 253
 254         atomic_inc(&sai->sai_cache_count);
 255
 256         RETURN(entry);
 257 }
 258
 259 /*
 260  * Used by the directory scanner to search entry with name.
 261  *
 262  * Only the caller can remove the entry from hash, so it is unnecessary to hold
 263  * hash lock. It is caller's duty to release the init refcount on the entry, so
 264  * it is also unnecessary to increase refcount on the entry.
 265  */
 266 static struct ll_sa_entry *
 267 ll_sa_entry_get_byname(struct ll_statahead_info *sai, const struct qstr *qstr)
 268 {
 269         struct ll_sa_entry *entry;
 270         int i = ll_sa_entry_hash(qstr->hash);
 271
 272         list_for_each_entry(entry, &sai->sai_cache[i], se_hash) {
 273                 if (entry->se_qstr.hash == qstr->hash &&
 274                     entry->se_qstr.len == qstr->len &&
 275                     memcmp(entry->se_qstr.name, qstr->name, qstr->len) == 0)
 276                         return entry;
 277         }
 278         return NULL;
 279 }
 280
 281 /*
 282  * Used by the async getattr request callback to find entry with index.
 283  *
 284  * Inside lli_sa_lock to prevent others to change the list during the search.
 285  * It needs to increase entry refcount before returning to guarantee that the
 286  * entry cannot be freed by others.
 287  */
 288 static struct ll_sa_entry *
 289 ll_sa_entry_get_byindex(struct ll_statahead_info *sai, __u64 index)
 290 {
 291         struct ll_sa_entry *entry;
 292
 293         list_for_each_entry(entry, &sai->sai_entries, se_link) {
 294                 if (entry->se_index == index) {
 295                         LASSERT(atomic_read(&entry->se_refcount) > 0);
 296                         atomic_inc(&entry->se_refcount);
 297                         return entry;
 298                 }
 299                 if (entry->se_index > index)
 300                         break;
 301         }
 302         return NULL;
 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 "LPU64"\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(ll_sa_entry_unhashed(entry));
 316
 317                 if (entry->se_inode)
 318                         iput(entry->se_inode);
 319
 320                 OBD_FREE(entry, entry->se_size);
 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(!ll_sa_entry_unhashed(entry));
 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                 /* keep those whose statahead RPC not finished */
 361                 if (pos->se_stat == SA_ENTRY_SUCC ||
 362                     pos->se_stat == SA_ENTRY_INVA)
 363                         do_sa_entry_fini(sai, pos);
 364         }
 365 }
 366
 367 /*
 368  * Inside lli_sa_lock.
 369  */
 370 static void
 371 __sa_entry_post_stat(struct ll_statahead_info *sai, struct ll_sa_entry *entry,
 372                      se_stat_t stat)
 373 {
 374         struct ll_sa_entry *se;
 375         struct list_head *pos = &sai->sai_entries_stated;
 376
 377         LASSERT(entry->se_stat == SA_ENTRY_INIT);
 378
 379         if (!list_empty(&entry->se_list))
 380                 list_del_init(&entry->se_list);
 381
 382         list_for_each_entry_reverse(se, &sai->sai_entries_stated, se_list) {
 383                 if (se->se_index < entry->se_index) {
 384                         pos = &se->se_list;
 385                         break;
 386                 }
 387         }
 388
 389         list_add(&entry->se_list, pos);
 390         entry->se_stat = stat;
 391 }
 392
 393 /*
 394  * Move entry to sai_entries_stated and sort with the index.
 395  * \retval 1    -- entry to be destroyed.
 396  * \retval 0    -- entry is inserted into stated list.
 397  */
 398 static void
 399 sa_entry_post_stat(struct ll_statahead_info *sai, struct ll_sa_entry *entry,
 400                    se_stat_t stat)
 401 {
 402         struct ll_inode_info *lli = ll_i2info(sai->sai_inode);
 403         struct md_enqueue_info *minfo = entry->se_minfo;
 404         struct ptlrpc_request *req = entry->se_req;
 405
 406         /* release resources used in RPC */
 407         if (minfo) {
 408                 entry->se_minfo = NULL;
 409                 ll_intent_release(&minfo->mi_it);
 410                 iput(minfo->mi_dir);
 411                 OBD_FREE_PTR(minfo);
 412         }
 413
 414         if (req) {
 415                 entry->se_req = NULL;
 416                 ptlrpc_req_finished(req);
 417         }
 418
 419         spin_lock(&lli->lli_sa_lock);
 420         __sa_entry_post_stat(sai, entry, stat);
 421         spin_unlock(&lli->lli_sa_lock);
 422 }
 423
 424 /*
 425  * Insert inode into the list of sai_entries_agl.
 426  */
 427 static void ll_agl_add(struct ll_statahead_info *sai,
 428                        struct inode *inode, int index)
 429 {
 430         struct ll_inode_info *child  = ll_i2info(inode);
 431         struct ll_inode_info *parent = ll_i2info(sai->sai_inode);
 432         int                   added  = 0;
 433
 434         spin_lock(&child->lli_agl_lock);
 435         if (child->lli_agl_index == 0) {
 436                 child->lli_agl_index = index;
 437                 spin_unlock(&child->lli_agl_lock);
 438
 439                 LASSERT(list_empty(&child->lli_agl_list));
 440
 441                 igrab(inode);
 442                 spin_lock(&parent->lli_agl_lock);
 443                 if (agl_list_empty(sai))
 444                         added = 1;
 445                 list_add_tail(&child->lli_agl_list, &sai->sai_entries_agl);
 446                 spin_unlock(&parent->lli_agl_lock);
 447         } else {
 448                 spin_unlock(&child->lli_agl_lock);
 449         }
 450
 451         if (added > 0)
 452                 wake_up(&sai->sai_agl_thread.t_ctl_waitq);
 453 }
 454
 455 static struct ll_statahead_info *ll_sai_alloc(void)
 456 {
 457         struct ll_statahead_info *sai;
 458         int                       i;
 459         ENTRY;
 460
 461         OBD_ALLOC_PTR(sai);
 462         if (!sai)
 463                 RETURN(NULL);
 464
 465         atomic_set(&sai->sai_refcount, 1);
 466
 467         spin_lock(&sai_generation_lock);
 468         sai->sai_generation = ++sai_generation;
 469         if (unlikely(sai_generation == 0))
 470                 sai->sai_generation = ++sai_generation;
 471         spin_unlock(&sai_generation_lock);
 472
 473         sai->sai_max = LL_SA_RPC_MIN;
 474         sai->sai_index = 1;
 475         init_waitqueue_head(&sai->sai_waitq);
 476         init_waitqueue_head(&sai->sai_thread.t_ctl_waitq);
 477         init_waitqueue_head(&sai->sai_agl_thread.t_ctl_waitq);
 478
 479         INIT_LIST_HEAD(&sai->sai_entries);
 480         INIT_LIST_HEAD(&sai->sai_entries_received);
 481         INIT_LIST_HEAD(&sai->sai_entries_stated);
 482         INIT_LIST_HEAD(&sai->sai_entries_agl);
 483
 484         for (i = 0; i < LL_SA_CACHE_SIZE; i++) {
 485                 INIT_LIST_HEAD(&sai->sai_cache[i]);
 486                 spin_lock_init(&sai->sai_cache_lock[i]);
 487         }
 488         atomic_set(&sai->sai_cache_count, 0);
 489
 490         RETURN(sai);
 491 }
 492
 493 static inline struct ll_statahead_info *ll_sai_get(struct inode *dir)
 494 {
 495         struct ll_inode_info *lli = ll_i2info(dir);
 496         struct ll_statahead_info *sai = NULL;
 497
 498         spin_lock(&lli->lli_sa_lock);
 499         sai = lli->lli_sai;
 500         if (sai != NULL)
 501                 atomic_inc(&sai->sai_refcount);
 502         spin_unlock(&lli->lli_sa_lock);
 503
 504         return sai;
 505 }
 506
 507 static void ll_sai_put(struct ll_statahead_info *sai)
 508 {
 509         struct ll_inode_info *lli = ll_i2info(sai->sai_inode);
 510
 511         if (atomic_dec_and_lock(&sai->sai_refcount, &lli->lli_sa_lock)) {
 512                 struct ll_sa_entry *entry, *next;
 513                 struct ll_sb_info *sbi = ll_i2sbi(sai->sai_inode);
 514
 515                 lli->lli_sai = NULL;
 516                 spin_unlock(&lli->lli_sa_lock);
 517
 518                 LASSERT(thread_is_stopped(&sai->sai_thread));
 519                 LASSERT(thread_is_stopped(&sai->sai_agl_thread));
 520                 LASSERT(sai->sai_sent == sai->sai_replied);
 521
 522                 list_for_each_entry_safe(entry, next, &sai->sai_entries,
 523                                          se_link)
 524                         do_sa_entry_fini(sai, entry);
 525
 526                 LASSERT(atomic_read(&sai->sai_cache_count) == 0);
 527                 LASSERT(agl_list_empty(sai));
 528                 LASSERT(atomic_read(&sai->sai_refcount) == 0);
 529
 530                 iput(sai->sai_inode);
 531                 OBD_FREE_PTR(sai);
 532                 atomic_dec(&sbi->ll_sa_running);
 533         }
 534 }
 535
 536 /* Do NOT forget to drop inode refcount when into sai_entries_agl. */
 537 static void ll_agl_trigger(struct inode *inode, struct ll_statahead_info *sai)
 538 {
 539         struct ll_inode_info *lli   = ll_i2info(inode);
 540         __u64                 index = lli->lli_agl_index;
 541         int                   rc;
 542         ENTRY;
 543
 544         LASSERT(list_empty(&lli->lli_agl_list));
 545
 546         /* AGL maybe fall behind statahead with one entry */
 547         if (is_omitted_entry(sai, index + 1)) {
 548                 lli->lli_agl_index = 0;
 549                 iput(inode);
 550                 RETURN_EXIT;
 551         }
 552
 553         /* Someone is in glimpse (sync or async), do nothing. */
 554         rc = down_write_trylock(&lli->lli_glimpse_sem);
 555         if (rc == 0) {
 556                 lli->lli_agl_index = 0;
 557                 iput(inode);
 558                 RETURN_EXIT;
 559         }
 560
 561         /*
 562          * Someone triggered glimpse within 1 sec before.
 563          * 1) The former glimpse succeeded with glimpse lock granted by OST, and
 564          *    if the lock is still cached on client, AGL needs to do nothing. If
 565          *    it is cancelled by other client, AGL maybe cannot obtaion new lock
 566          *    for no glimpse callback triggered by AGL.
 567          * 2) The former glimpse succeeded, but OST did not grant glimpse lock.
 568          *    Under such case, it is quite possible that the OST will not grant
 569          *    glimpse lock for AGL also.
 570          * 3) The former glimpse failed, compared with other two cases, it is
 571          *    relative rare. AGL can ignore such case, and it will not muchly
 572          *    affect the performance.
 573          */
 574         if (lli->lli_glimpse_time != 0 &&
 575             cfs_time_before(cfs_time_shift(-1), lli->lli_glimpse_time)) {
 576                 up_write(&lli->lli_glimpse_sem);
 577                 lli->lli_agl_index = 0;
 578                 iput(inode);
 579                 RETURN_EXIT;
 580         }
 581
 582         CDEBUG(D_READA, "Handling (init) async glimpse: inode = "
 583                DFID", idx = "LPU64"\n", PFID(&lli->lli_fid), index);
 584
 585         cl_agl(inode);
 586         lli->lli_agl_index = 0;
 587         lli->lli_glimpse_time = cfs_time_current();
 588         up_write(&lli->lli_glimpse_sem);
 589
 590         CDEBUG(D_READA, "Handled (init) async glimpse: inode= "
 591                DFID", idx = "LPU64", rc = %d\n",
 592                PFID(&lli->lli_fid), index, rc);
 593
 594         iput(inode);
 595
 596         EXIT;
 597 }
 598
 599 /* prepare inode for received statahead entry, and add it into agl list */
 600 static void sa_post_one(struct ll_statahead_info *sai,
 601                         struct ll_sa_entry *entry)
 602 {
 603         struct inode           *dir   = sai->sai_inode;
 604         struct inode           *child;
 605         struct md_enqueue_info *minfo;
 606         struct lookup_intent   *it;
 607         struct ptlrpc_request  *req;
 608         struct mdt_body        *body;
 609         int                     rc    = 0;
 610         ENTRY;
 611
 612         LASSERT(entry->se_handle != 0);
 613
 614         minfo = entry->se_minfo;
 615         it = &minfo->mi_it;
 616         req = entry->se_req;
 617         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 618         if (body == NULL)
 619                 GOTO(out, rc = -EFAULT);
 620
 621         child = entry->se_inode;
 622         if (child == NULL) {
 623                 /*
 624                  * lookup.
 625                  */
 626                 LASSERT(fid_is_zero(&minfo->mi_data.op_fid2));
 627
 628                 /* XXX: No fid in reply, this is probaly cross-ref case.
 629                  * SA can't handle it yet. */
 630                 if (body->mbo_valid & OBD_MD_MDS)
 631                         GOTO(out, rc = -EAGAIN);
 632         } else {
 633                 /*
 634                  * revalidate.
 635                  */
 636                 /* unlinked and re-created with the same name */
 637                 if (unlikely(!lu_fid_eq(&minfo->mi_data.op_fid2,
 638                                         &body->mbo_fid1))) {
 639                         entry->se_inode = NULL;
 640                         iput(child);
 641                         child = NULL;
 642                 }
 643         }
 644
 645         it->d.lustre.it_lock_handle = entry->se_handle;
 646         rc = md_revalidate_lock(ll_i2mdexp(dir), it, ll_inode2fid(dir), NULL);
 647         if (rc != 1)
 648                 GOTO(out, rc = -EAGAIN);
 649
 650         rc = ll_prep_inode(&child, req, dir->i_sb, it);
 651         if (rc)
 652                 GOTO(out, rc);
 653
 654         CDEBUG(D_DLMTRACE, "%s: setting l_data to inode "DFID"(%p)\n",
 655                ll_get_fsname(child->i_sb, NULL, 0),
 656                PFID(ll_inode2fid(child)), child);
 657         ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, child, it, NULL);
 658
 659         entry->se_inode = child;
 660
 661         if (agl_should_run(sai, child))
 662                 ll_agl_add(sai, child, entry->se_index);
 663
 664         EXIT;
 665
 666 out:
 667         /* The "sa_entry_post_stat()" will drop related ldlm ibits lock
 668          * reference count by calling "ll_intent_drop_lock()" in spite of the
 669          * above operations failed or not. Do not worry about calling
 670          * "ll_intent_drop_lock()" more than once. */
 671         sa_entry_post_stat(sai, entry, rc < 0 ? SA_ENTRY_INVA : SA_ENTRY_SUCC);
 672         if (entry->se_index == sai->sai_index_wait)
 673                 wake_up(&sai->sai_waitq);
 674         ll_sa_entry_put(sai, entry);
 675 }
 676
 677 static void ll_post_statahead(struct ll_statahead_info *sai)
 678 {
 679         struct ll_inode_info *lli;
 680
 681         lli = ll_i2info(sai->sai_inode);
 682
 683         while (!sa_received_empty(sai)) {
 684                 struct ll_sa_entry *entry;
 685
 686                 spin_lock(&lli->lli_sa_lock);
 687                 if (unlikely(sa_received_empty(sai))) {
 688                         spin_unlock(&lli->lli_sa_lock);
 689                         break;
 690                 }
 691                 entry = sa_first_received_entry(sai);
 692                 atomic_inc(&entry->se_refcount);
 693                 list_del_init(&entry->se_list);
 694                 spin_unlock(&lli->lli_sa_lock);
 695
 696                 sa_post_one(sai, entry);
 697         }
 698
 699         spin_lock(&lli->lli_agl_lock);
 700         while (!agl_list_empty(sai)) {
 701                 struct ll_inode_info *clli;
 702
 703                 clli = agl_first_entry(sai);
 704                 list_del_init(&clli->lli_agl_list);
 705                 spin_unlock(&lli->lli_agl_lock);
 706
 707                 ll_agl_trigger(&clli->lli_vfs_inode, sai);
 708
 709                 spin_lock(&lli->lli_agl_lock);
 710         }
 711         spin_unlock(&lli->lli_agl_lock);
 712 }
 713
 714 static int ll_statahead_interpret(struct ptlrpc_request *req,
 715                                   struct md_enqueue_info *minfo, int rc)
 716 {
 717         struct lookup_intent *it = &minfo->mi_it;
 718         struct inode *dir = minfo->mi_dir;
 719         struct ll_inode_info *lli = ll_i2info(dir);
 720         struct ll_statahead_info *sai;
 721         struct ll_sa_entry *entry;
 722         int wakeup;
 723         ENTRY;
 724
 725         if (it_disposition(it, DISP_LOOKUP_NEG))
 726                 rc = -ENOENT;
 727
 728         sai = ll_sai_get(dir);
 729         LASSERT(sai != NULL);
 730         LASSERT(!thread_is_stopped(&sai->sai_thread));
 731
 732         spin_lock(&lli->lli_sa_lock);
 733         entry = ll_sa_entry_get_byindex(sai, minfo->mi_cbdata);
 734         LASSERT(entry != NULL);
 735         if (rc != 0) {
 736                 __sa_entry_post_stat(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                 entry->se_handle = it->d.lustre.it_lock_handle;
 746                 ll_intent_drop_lock(it);
 747                 wakeup = sa_received_empty(sai);
 748                 list_add_tail(&entry->se_list, &sai->sai_entries_received);
 749         }
 750         sai->sai_replied++;
 751         spin_unlock(&lli->lli_sa_lock);
 752
 753         ll_sa_entry_put(sai, entry);
 754         if (wakeup)
 755                 wake_up(&sai->sai_thread.t_ctl_waitq);
 756
 757         if (rc != 0) {
 758                 ll_intent_release(it);
 759                 iput(dir);
 760                 OBD_FREE_PTR(minfo);
 761         }
 762         ll_sai_put(sai);
 763         RETURN(rc);
 764 }
 765
 766 static void sa_args_fini(struct md_enqueue_info *minfo,
 767                          struct ldlm_enqueue_info *einfo)
 768 {
 769         LASSERT(minfo && einfo);
 770         iput(minfo->mi_dir);
 771         capa_put(minfo->mi_data.op_capa1);
 772         capa_put(minfo->mi_data.op_capa2);
 773         OBD_FREE_PTR(minfo);
 774         OBD_FREE_PTR(einfo);
 775 }
 776
 777 /**
 778  * There is race condition between "capa_put" and "ll_statahead_interpret" for
 779  * accessing "op_data.op_capa[1,2]" as following:
 780  * "capa_put" releases "op_data.op_capa[1,2]"'s reference count after calling
 781  * "md_intent_getattr_async". But "ll_statahead_interpret" maybe run first, and
 782  * fill "op_data.op_capa[1,2]" as POISON, then cause "capa_put" access invalid
 783  * "ocapa". So here reserve "op_data.op_capa[1,2]" in "pcapa" before calling
 784  * "md_intent_getattr_async".
 785  */
 786 static int sa_args_init(struct inode *dir, struct inode *child,
 787                         struct ll_sa_entry *entry, struct md_enqueue_info **pmi,
 788                         struct ldlm_enqueue_info **pei,
 789                         struct obd_capa **pcapa)
 790 {
 791         struct qstr              *qstr = &entry->se_qstr;
 792         struct md_enqueue_info   *minfo;
 793         struct ldlm_enqueue_info *einfo;
 794         struct md_op_data        *op_data;
 795
 796         OBD_ALLOC_PTR(einfo);
 797         if (einfo == NULL)
 798                 return -ENOMEM;
 799
 800         OBD_ALLOC_PTR(minfo);
 801         if (minfo == NULL) {
 802                 OBD_FREE_PTR(einfo);
 803                 return -ENOMEM;
 804         }
 805
 806         op_data = ll_prep_md_op_data(&minfo->mi_data, dir, child, qstr->name,
 807                                      qstr->len, 0, LUSTRE_OPC_ANY, NULL);
 808         if (IS_ERR(op_data)) {
 809                 OBD_FREE_PTR(einfo);
 810                 OBD_FREE_PTR(minfo);
 811                 return PTR_ERR(op_data);
 812         }
 813
 814         minfo->mi_it.it_op = IT_GETATTR;
 815         minfo->mi_dir = igrab(dir);
 816         minfo->mi_cb = ll_statahead_interpret;
 817         minfo->mi_cbdata = entry->se_index;
 818
 819         einfo->ei_type   = LDLM_IBITS;
 820         einfo->ei_mode   = it_to_lock_mode(&minfo->mi_it);
 821         einfo->ei_cb_bl  = ll_md_blocking_ast;
 822         einfo->ei_cb_cp  = ldlm_completion_ast;
 823         einfo->ei_cb_gl  = NULL;
 824         einfo->ei_cbdata = NULL;
 825
 826         *pmi = minfo;
 827         *pei = einfo;
 828         pcapa[0] = op_data->op_capa1;
 829         pcapa[1] = op_data->op_capa2;
 830
 831         return 0;
 832 }
 833
 834 static int do_sa_lookup(struct inode *dir, struct ll_sa_entry *entry)
 835 {
 836         struct md_enqueue_info   *minfo;
 837         struct ldlm_enqueue_info *einfo;
 838         struct obd_capa          *capas[2];
 839         int                       rc;
 840         ENTRY;
 841
 842         rc = sa_args_init(dir, NULL, entry, &minfo, &einfo, capas);
 843         if (rc)
 844                 RETURN(rc);
 845
 846         rc = md_intent_getattr_async(ll_i2mdexp(dir), minfo, einfo);
 847         if (!rc) {
 848                 capa_put(capas[0]);
 849                 capa_put(capas[1]);
 850         } else {
 851                 sa_args_fini(minfo, einfo);
 852         }
 853
 854         RETURN(rc);
 855 }
 856
 857 /**
 858  * similar to ll_revalidate_it().
 859  * \retval      1 -- dentry valid
 860  * \retval      0 -- will send stat-ahead request
 861  * \retval others -- prepare stat-ahead request failed
 862  */
 863 static int do_sa_revalidate(struct inode *dir, struct ll_sa_entry *entry,
 864                             struct dentry *dentry)
 865 {
 866         struct inode             *inode = dentry->d_inode;
 867         struct lookup_intent      it = { .it_op = IT_GETATTR,
 868                                          .d.lustre.it_lock_handle = 0 };
 869         struct md_enqueue_info   *minfo;
 870         struct ldlm_enqueue_info *einfo;
 871         struct obd_capa          *capas[2];
 872         int rc;
 873         ENTRY;
 874
 875         if (unlikely(inode == NULL))
 876                 RETURN(1);
 877
 878         if (d_mountpoint(dentry))
 879                 RETURN(1);
 880
 881         entry->se_inode = igrab(inode);
 882         rc = md_revalidate_lock(ll_i2mdexp(dir), &it, ll_inode2fid(inode),NULL);
 883         if (rc == 1) {
 884                 entry->se_handle = it.d.lustre.it_lock_handle;
 885                 ll_intent_release(&it);
 886                 RETURN(1);
 887         }
 888
 889         rc = sa_args_init(dir, inode, entry, &minfo, &einfo, capas);
 890         if (rc) {
 891                 entry->se_inode = NULL;
 892                 iput(inode);
 893                 RETURN(rc);
 894         }
 895
 896         rc = md_intent_getattr_async(ll_i2mdexp(dir), minfo, einfo);
 897         if (!rc) {
 898                 capa_put(capas[0]);
 899                 capa_put(capas[1]);
 900         } else {
 901                 entry->se_inode = NULL;
 902                 iput(inode);
 903                 sa_args_fini(minfo, einfo);
 904         }
 905
 906         RETURN(rc);
 907 }
 908
 909 static void ll_statahead_one(struct dentry *parent, const char *name,
 910                              const int namelen)
 911 {
 912         struct inode             *dir    = parent->d_inode;
 913         struct ll_inode_info     *lli    = ll_i2info(dir);
 914         struct ll_statahead_info *sai    = lli->lli_sai;
 915         struct dentry            *dentry = NULL;
 916         struct ll_sa_entry       *entry;
 917         int                       rc;
 918         ENTRY;
 919
 920         entry = ll_sa_entry_alloc(sai, sai->sai_index, name,namelen);
 921         if (IS_ERR(entry))
 922                 RETURN_EXIT;
 923
 924         dentry = d_lookup(parent, &entry->se_qstr);
 925         if (!dentry) {
 926                 rc = do_sa_lookup(dir, entry);
 927         } else {
 928                 rc = do_sa_revalidate(dir, entry, dentry);
 929                 if (rc == 1 && agl_should_run(sai, dentry->d_inode))
 930                         ll_agl_add(sai, dentry->d_inode, entry->se_index);
 931         }
 932
 933         if (dentry != NULL)
 934                 dput(dentry);
 935
 936         if (rc) {
 937                 sa_entry_post_stat(sai, entry,
 938                                    rc < 0 ? SA_ENTRY_INVA : SA_ENTRY_SUCC);
 939                 if (entry->se_index == sai->sai_index_wait)
 940                         wake_up(&sai->sai_waitq);
 941         } else {
 942                 sai->sai_sent++;
 943         }
 944
 945         sai->sai_index++;
 946         /* drop one refcount on entry by ll_sa_entry_alloc */
 947         ll_sa_entry_put(sai, entry);
 948
 949         EXIT;
 950 }
 951
 952 static int ll_agl_thread(void *arg)
 953 {
 954         struct dentry *parent = (struct dentry *)arg;
 955         struct inode *dir = parent->d_inode;
 956         struct ll_inode_info *plli = ll_i2info(dir);
 957         struct ll_inode_info *clli;
 958         struct ll_sb_info *sbi = ll_i2sbi(dir);
 959         struct ll_statahead_info *sai;
 960         struct ptlrpc_thread *thread;
 961         struct l_wait_info lwi = { 0 };
 962         ENTRY;
 963
 964
 965         sai = ll_sai_get(dir);
 966         thread = &sai->sai_agl_thread;
 967         thread->t_pid = current_pid();
 968         CDEBUG(D_READA, "agl thread started: sai %p, parent %.*s\n",
 969                sai, parent->d_name.len, parent->d_name.name);
 970
 971         atomic_inc(&sbi->ll_agl_total);
 972         spin_lock(&plli->lli_agl_lock);
 973         sai->sai_agl_valid = 1;
 974         if (thread_is_init(thread))
 975                 /* If someone else has changed the thread state
 976                  * (e.g. already changed to SVC_STOPPING), we can't just
 977                  * blindly overwrite that setting. */
 978                 thread_set_flags(thread, SVC_RUNNING);
 979         spin_unlock(&plli->lli_agl_lock);
 980         wake_up(&thread->t_ctl_waitq);
 981
 982         while (1) {
 983                 l_wait_event(thread->t_ctl_waitq,
 984                              !agl_list_empty(sai) ||
 985                              !thread_is_running(thread),
 986                              &lwi);
 987
 988                 if (!thread_is_running(thread))
 989                         break;
 990
 991                 spin_lock(&plli->lli_agl_lock);
 992                 /* The statahead thread maybe help to process AGL entries,
 993                  * so check whether list empty again. */
 994                 if (!agl_list_empty(sai)) {
 995                         clli = agl_first_entry(sai);
 996                         list_del_init(&clli->lli_agl_list);
 997                         spin_unlock(&plli->lli_agl_lock);
 998                         ll_agl_trigger(&clli->lli_vfs_inode, sai);
 999                 } else {
1000                         spin_unlock(&plli->lli_agl_lock);
1001                 }
1002         }
1003
1004         spin_lock(&plli->lli_agl_lock);
1005         sai->sai_agl_valid = 0;
1006         while (!agl_list_empty(sai)) {
1007                 clli = agl_first_entry(sai);
1008                 list_del_init(&clli->lli_agl_list);
1009                 spin_unlock(&plli->lli_agl_lock);
1010                 clli->lli_agl_index = 0;
1011                 iput(&clli->lli_vfs_inode);
1012                 spin_lock(&plli->lli_agl_lock);
1013         }
1014         thread_set_flags(thread, SVC_STOPPED);
1015         spin_unlock(&plli->lli_agl_lock);
1016         wake_up(&thread->t_ctl_waitq);
1017         ll_sai_put(sai);
1018         CDEBUG(D_READA, "agl thread stopped: sai %p, parent %.*s\n",
1019                sai, parent->d_name.len, parent->d_name.name);
1020         RETURN(0);
1021 }
1022
1023 static void ll_start_agl(struct dentry *parent, struct ll_statahead_info *sai)
1024 {
1025         struct ptlrpc_thread *thread = &sai->sai_agl_thread;
1026         struct l_wait_info    lwi    = { 0 };
1027         struct ll_inode_info  *plli;
1028         struct task_struct            *task;
1029         ENTRY;
1030
1031         CDEBUG(D_READA, "start agl thread: sai %p, parent %.*s\n",
1032                sai, parent->d_name.len, parent->d_name.name);
1033
1034         plli = ll_i2info(parent->d_inode);
1035         task = kthread_run(ll_agl_thread, parent,
1036                                "ll_agl_%u", plli->lli_opendir_pid);
1037         if (IS_ERR(task)) {
1038                 CERROR("can't start ll_agl thread, rc: %ld\n", PTR_ERR(task));
1039                 thread_set_flags(thread, SVC_STOPPED);
1040                 RETURN_EXIT;
1041         }
1042
1043         l_wait_event(thread->t_ctl_waitq,
1044                      thread_is_running(thread) || thread_is_stopped(thread),
1045                      &lwi);
1046         EXIT;
1047 }
1048
1049 static int ll_statahead_thread(void *arg)
1050 {
1051         struct dentry *parent = (struct dentry *)arg;
1052         struct inode *dir = parent->d_inode;
1053         struct ll_inode_info *lli = ll_i2info(dir);
1054         struct ll_sb_info *sbi = ll_i2sbi(dir);
1055         struct ll_statahead_info *sai;
1056         struct ptlrpc_thread *thread;
1057         struct ptlrpc_thread *agl_thread;
1058         int first = 0;
1059         struct md_op_data *op_data;
1060         struct ll_dir_chain chain;
1061         struct l_wait_info lwi = { 0 };
1062         struct page *page = NULL;
1063         __u64 pos = 0;
1064         int rc = 0;
1065         ENTRY;
1066
1067         sai = ll_sai_get(dir);
1068         thread = &sai->sai_thread;
1069         agl_thread = &sai->sai_agl_thread;
1070         thread->t_pid = current_pid();
1071         CDEBUG(D_READA, "statahead thread starting: sai %p, parent %.*s\n",
1072                sai, parent->d_name.len, parent->d_name.name);
1073
1074         op_data = ll_prep_md_op_data(NULL, dir, dir, NULL, 0, 0,
1075                                      LUSTRE_OPC_ANY, dir);
1076         if (IS_ERR(op_data))
1077                 GOTO(out, rc = PTR_ERR(op_data));
1078
1079         op_data->op_max_pages = ll_i2sbi(dir)->ll_md_brw_pages;
1080
1081         if (sbi->ll_flags & LL_SBI_AGL_ENABLED)
1082                 ll_start_agl(parent, sai);
1083
1084         atomic_inc(&sbi->ll_sa_total);
1085         spin_lock(&lli->lli_sa_lock);
1086         if (thread_is_init(thread))
1087                 /* If someone else has changed the thread state
1088                  * (e.g. already changed to SVC_STOPPING), we can't just
1089                  * blindly overwrite that setting. */
1090                 thread_set_flags(thread, SVC_RUNNING);
1091         spin_unlock(&lli->lli_sa_lock);
1092         wake_up(&thread->t_ctl_waitq);
1093
1094         ll_dir_chain_init(&chain);
1095         while (pos != MDS_DIR_END_OFF && thread_is_running(thread)) {
1096                 struct lu_dirpage *dp;
1097                 struct lu_dirent  *ent;
1098
1099                 sai->sai_in_readpage = 1;
1100                 page = ll_get_dir_page(dir, op_data, pos, &chain);
1101                 sai->sai_in_readpage = 0;
1102                 if (IS_ERR(page)) {
1103                         rc = PTR_ERR(page);
1104                         CDEBUG(D_READA, "error reading dir "DFID" at "LPU64
1105                                "/"LPU64" opendir_pid = %u: rc = %d\n",
1106                                PFID(ll_inode2fid(dir)), pos, sai->sai_index,
1107                                lli->lli_opendir_pid, rc);
1108                         break;
1109                 }
1110
1111                 dp = page_address(page);
1112                 for (ent = lu_dirent_start(dp);
1113                      ent != NULL && thread_is_running(thread) &&
1114                      !sa_low_hit(sai);
1115                      ent = lu_dirent_next(ent)) {
1116                         __u64 hash;
1117                         int namelen;
1118                         char *name;
1119
1120                         hash = le64_to_cpu(ent->lde_hash);
1121                         if (unlikely(hash < pos))
1122                                 /*
1123                                  * Skip until we find target hash value.
1124                                  */
1125                                 continue;
1126
1127                         namelen = le16_to_cpu(ent->lde_namelen);
1128                         if (unlikely(namelen == 0))
1129                                 /*
1130                                  * Skip dummy record.
1131                                  */
1132                                 continue;
1133
1134                         name = ent->lde_name;
1135                         if (name[0] == '.') {
1136                                 if (namelen == 1) {
1137                                         /*
1138                                          * skip "."
1139                                          */
1140                                         continue;
1141                                 } else if (name[1] == '.' && namelen == 2) {
1142                                         /*
1143                                          * skip ".."
1144                                          */
1145                                         continue;
1146                                 } else if (!sai->sai_ls_all) {
1147                                         /*
1148                                          * skip hidden files.
1149                                          */
1150                                         sai->sai_skip_hidden++;
1151                                         continue;
1152                                 }
1153                         }
1154
1155                         /*
1156                          * don't stat-ahead first entry.
1157                          */
1158                         if (unlikely(++first == 1))
1159                                 continue;
1160
1161                         /* wait for spare statahead window */
1162                         do {
1163                                 l_wait_event(thread->t_ctl_waitq,
1164                                              !sa_sent_full(sai) ||
1165                                              !sa_received_empty(sai) ||
1166                                              !agl_list_empty(sai) ||
1167                                              !thread_is_running(thread),
1168                                              &lwi);
1169
1170                                 ll_post_statahead(sai);
1171                         } while (sa_sent_full(sai) &&
1172                                  thread_is_running(thread));
1173
1174                         ll_statahead_one(parent, name, namelen);
1175                 }
1176
1177                 pos = le64_to_cpu(dp->ldp_hash_end);
1178                 ll_release_page(dir, page,
1179                                 le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1180
1181                 if (sa_low_hit(sai)) {
1182                         rc = -EFAULT;
1183                         atomic_inc(&sbi->ll_sa_wrong);
1184                         CDEBUG(D_READA, "Statahead for dir "DFID" hit "
1185                                "ratio too low: hit/miss "LPU64"/"LPU64
1186                                ", sent/replied "LPU64"/"LPU64", stopping "
1187                                "statahead thread: pid %d\n",
1188                                PFID(&lli->lli_fid), sai->sai_hit,
1189                                sai->sai_miss, sai->sai_sent,
1190                                sai->sai_replied, current_pid());
1191                         break;
1192                 }
1193         }
1194         ll_dir_chain_fini(&chain);
1195         ll_finish_md_op_data(op_data);
1196
1197         if (rc < 0) {
1198                 spin_lock(&lli->lli_sa_lock);
1199                 thread_set_flags(thread, SVC_STOPPING);
1200                 lli->lli_sa_enabled = 0;
1201                 spin_unlock(&lli->lli_sa_lock);
1202         }
1203
1204         /* statahead is finished, but statahead entries need to be cached, wait
1205          * for file release to stop me. */
1206         while (thread_is_running(thread)) {
1207                 l_wait_event(thread->t_ctl_waitq,
1208                              !sa_received_empty(sai) ||
1209                              !agl_list_empty(sai) ||
1210                              !thread_is_running(thread),
1211                              &lwi);
1212
1213                 ll_post_statahead(sai);
1214         }
1215
1216         EXIT;
1217 out:
1218         if (sai->sai_agl_valid) {
1219                 spin_lock(&lli->lli_agl_lock);
1220                 thread_set_flags(agl_thread, SVC_STOPPING);
1221                 spin_unlock(&lli->lli_agl_lock);
1222                 wake_up(&agl_thread->t_ctl_waitq);
1223
1224                 CDEBUG(D_READA, "stop agl thread: sai %p pid %u\n",
1225                        sai, (unsigned int)agl_thread->t_pid);
1226                 l_wait_event(agl_thread->t_ctl_waitq,
1227                              thread_is_stopped(agl_thread),
1228                              &lwi);
1229         } else {
1230                 /* Set agl_thread flags anyway. */
1231                 thread_set_flags(agl_thread, SVC_STOPPED);
1232         }
1233
1234         /* wait for inflight statahead RPCs to finish, and then we can free sai
1235          * safely because statahead RPC will access sai data */
1236         while (sai->sai_sent != sai->sai_replied) {
1237                 /* in case we're not woken up, timeout wait */
1238                 lwi = LWI_TIMEOUT(HZ >> 3, NULL, NULL);
1239                 l_wait_event(thread->t_ctl_waitq,
1240                         sai->sai_sent == sai->sai_replied, &lwi);
1241         }
1242
1243         /* release resources held by received entries. */
1244         ll_post_statahead(sai);
1245
1246         spin_lock(&lli->lli_sa_lock);
1247         thread_set_flags(thread, SVC_STOPPED);
1248         spin_unlock(&lli->lli_sa_lock);
1249
1250         wake_up(&sai->sai_waitq);
1251         wake_up(&thread->t_ctl_waitq);
1252         ll_sai_put(sai);
1253         CDEBUG(D_READA, "statahead thread stopped: sai %p, parent %.*s\n",
1254                sai, parent->d_name.len, parent->d_name.name);
1255         dput(parent);
1256         return rc;
1257 }
1258
1259 /* authorize opened dir handle @key to statahead later */
1260 void ll_authorize_statahead(struct inode *dir, void *key)
1261 {
1262         struct ll_inode_info *lli = ll_i2info(dir);
1263
1264         spin_lock(&lli->lli_sa_lock);
1265         if (lli->lli_opendir_key == NULL && lli->lli_sai == NULL) {
1266                 /*
1267                  * if lli_sai is not NULL, it means previous statahead is not
1268                  * finished yet, we'd better not start a new statahead for now.
1269                  */
1270                 LASSERT(lli->lli_opendir_pid == 0);
1271                 lli->lli_opendir_key = key;
1272                 lli->lli_opendir_pid = current_pid();
1273                 lli->lli_sa_enabled = 1;
1274         }
1275         spin_unlock(&lli->lli_sa_lock);
1276 }
1277
1278 /* deauthorize opened dir handle @key to statahead, but statahead thread may
1279  * still be running, notify it to quit. */
1280 void ll_deauthorize_statahead(struct inode *dir, void *key)
1281 {
1282         struct ll_inode_info *lli = ll_i2info(dir);
1283         struct ll_statahead_info *sai;
1284
1285         LASSERT(lli->lli_opendir_key == key);
1286         LASSERT(lli->lli_opendir_pid != 0);
1287
1288         CDEBUG(D_READA, "deauthorize statahead for "DFID"\n",
1289                 PFID(&lli->lli_fid));
1290
1291         spin_lock(&lli->lli_sa_lock);
1292         lli->lli_opendir_key = NULL;
1293         lli->lli_opendir_pid = 0;
1294         lli->lli_sa_enabled = 0;
1295         sai = lli->lli_sai;
1296         if (sai != NULL && thread_is_running(&sai->sai_thread)) {
1297                 /*
1298                  * statahead thread may not quit yet because it needs to cache
1299                  * stated entries, now it's time to tell it to quit.
1300                  */
1301                 thread_set_flags(&sai->sai_thread, SVC_STOPPING);
1302                 wake_up(&sai->sai_thread.t_ctl_waitq);
1303         }
1304         spin_unlock(&lli->lli_sa_lock);
1305 }
1306
1307 enum {
1308         /**
1309          * not first dirent, or is "."
1310          */
1311         LS_NONE_FIRST_DE = 0,
1312         /**
1313          * the first non-hidden dirent
1314          */
1315         LS_FIRST_DE,
1316         /**
1317          * the first hidden dirent, that is "."
1318          */
1319         LS_FIRST_DOT_DE
1320 };
1321
1322 static int is_first_dirent(struct inode *dir, struct dentry *dentry)
1323 {
1324         struct ll_dir_chain   chain;
1325         struct qstr          *target = &dentry->d_name;
1326         struct md_op_data    *op_data;
1327         int                   dot_de;
1328         struct page          *page = NULL;
1329         int                   rc     = LS_NONE_FIRST_DE;
1330         __u64                 pos = 0;
1331         ENTRY;
1332
1333         op_data = ll_prep_md_op_data(NULL, dir, dir, NULL, 0, 0,
1334                                      LUSTRE_OPC_ANY, dir);
1335         if (IS_ERR(op_data))
1336                 RETURN(PTR_ERR(op_data));
1337         /**
1338          *FIXME choose the start offset of the readdir
1339          */
1340         op_data->op_stripe_offset = 0;
1341         op_data->op_max_pages = ll_i2sbi(dir)->ll_md_brw_pages;
1342
1343         ll_dir_chain_init(&chain);
1344         page = ll_get_dir_page(dir, op_data, 0, &chain);
1345
1346         while (1) {
1347                 struct lu_dirpage *dp;
1348                 struct lu_dirent  *ent;
1349
1350                 if (IS_ERR(page)) {
1351                         struct ll_inode_info *lli = ll_i2info(dir);
1352
1353                         rc = PTR_ERR(page);
1354                         CERROR("%s: reading dir "DFID" at "LPU64
1355                                "opendir_pid = %u : rc = %d\n",
1356                                ll_get_fsname(dir->i_sb, NULL, 0),
1357                                PFID(ll_inode2fid(dir)), pos,
1358                                lli->lli_opendir_pid, rc);
1359                         break;
1360                 }
1361
1362                 dp = page_address(page);
1363                 for (ent = lu_dirent_start(dp); ent != NULL;
1364                      ent = lu_dirent_next(ent)) {
1365                         __u64 hash;
1366                         int namelen;
1367                         char *name;
1368
1369                         hash = le64_to_cpu(ent->lde_hash);
1370                         /* The ll_get_dir_page() can return any page containing
1371                          * the given hash which may be not the start hash. */
1372                         if (unlikely(hash < pos))
1373                                 continue;
1374
1375                         namelen = le16_to_cpu(ent->lde_namelen);
1376                         if (unlikely(namelen == 0))
1377                                 /*
1378                                  * skip dummy record.
1379                                  */
1380                                 continue;
1381
1382                         name = ent->lde_name;
1383                         if (name[0] == '.') {
1384                                 if (namelen == 1)
1385                                         /*
1386                                          * skip "."
1387                                          */
1388                                         continue;
1389                                 else if (name[1] == '.' && namelen == 2)
1390                                         /*
1391                                          * skip ".."
1392                                          */
1393                                         continue;
1394                                 else
1395                                         dot_de = 1;
1396                         } else {
1397                                 dot_de = 0;
1398                         }
1399
1400                         if (dot_de && target->name[0] != '.') {
1401                                 CDEBUG(D_READA, "%.*s skip hidden file %.*s\n",
1402                                        target->len, target->name,
1403                                        namelen, name);
1404                                 continue;
1405                         }
1406
1407                         if (target->len != namelen ||
1408                             memcmp(target->name, name, namelen) != 0)
1409                                 rc = LS_NONE_FIRST_DE;
1410                         else if (!dot_de)
1411                                 rc = LS_FIRST_DE;
1412                         else
1413                                 rc = LS_FIRST_DOT_DE;
1414
1415                         ll_release_page(dir, page, false);
1416                         GOTO(out, rc);
1417                 }
1418                 pos = le64_to_cpu(dp->ldp_hash_end);
1419                 if (pos == MDS_DIR_END_OFF) {
1420                         /*
1421                          * End of directory reached.
1422                          */
1423                         ll_release_page(dir, page, false);
1424                         GOTO(out, rc);
1425                 } else {
1426                         /*
1427                          * chain is exhausted
1428                          * Normal case: continue to the next page.
1429                          */
1430                         ll_release_page(dir, page, le32_to_cpu(dp->ldp_flags) &
1431                                               LDF_COLLIDE);
1432                         page = ll_get_dir_page(dir, op_data, pos, &chain);
1433                 }
1434         }
1435         EXIT;
1436 out:
1437         ll_dir_chain_fini(&chain);
1438         ll_finish_md_op_data(op_data);
1439         return rc;
1440 }
1441
1442 static void
1443 ll_sai_unplug(struct ll_statahead_info *sai, struct ll_sa_entry *entry)
1444 {
1445         if (entry != NULL && entry->se_stat == SA_ENTRY_SUCC) {
1446                 struct ll_sb_info *sbi = ll_i2sbi(sai->sai_inode);
1447
1448                 sai->sai_hit++;
1449                 sai->sai_consecutive_miss = 0;
1450                 sai->sai_max = min(2 * sai->sai_max, sbi->ll_sa_max);
1451         } else {
1452                 sai->sai_miss++;
1453                 sai->sai_consecutive_miss++;
1454         }
1455         ll_sa_entry_fini(sai, entry);
1456         wake_up(&sai->sai_thread.t_ctl_waitq);
1457 }
1458
1459 static int revalidate_statahead_dentry(struct inode *dir,
1460                                         struct ll_statahead_info *sai,
1461                                         struct dentry **dentryp,
1462                                         int only_unplug)
1463 {
1464         struct ll_sa_entry *entry = NULL;
1465         struct l_wait_info lwi = { 0 };
1466         int rc = 0;
1467         ENTRY;
1468
1469         if ((*dentryp)->d_name.name[0] == '.') {
1470                 if (sai->sai_ls_all ||
1471                     sai->sai_miss_hidden >= sai->sai_skip_hidden) {
1472                         /*
1473                          * Hidden dentry is the first one, or statahead
1474                          * thread does not skip so many hidden dentries
1475                          * before "sai_ls_all" enabled as below.
1476                          */
1477                 } else {
1478                         if (!sai->sai_ls_all)
1479                                 /*
1480                                  * It maybe because hidden dentry is not
1481                                  * the first one, "sai_ls_all" was not
1482                                  * set, then "ls -al" missed. Enable
1483                                  * "sai_ls_all" for such case.
1484                                  */
1485                                 sai->sai_ls_all = 1;
1486
1487                         /*
1488                          * Such "getattr" has been skipped before
1489                          * "sai_ls_all" enabled as above.
1490                          */
1491                         sai->sai_miss_hidden++;
1492                         RETURN(-EAGAIN);
1493                 }
1494         }
1495
1496         entry = ll_sa_entry_get_byname(sai, &(*dentryp)->d_name);
1497         if (entry == NULL || only_unplug) {
1498                 ll_sai_unplug(sai, entry);
1499                 RETURN(entry ? 1 : -EAGAIN);
1500         }
1501
1502         /* if statahead is busy in readdir, help it do post-work */
1503         if (!ll_sa_entry_stated(entry) && sai->sai_in_readpage)
1504                 ll_post_statahead(sai);
1505
1506         if (!ll_sa_entry_stated(entry)) {
1507                 sai->sai_index_wait = entry->se_index;
1508                 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(30), NULL,
1509                                         LWI_ON_SIGNAL_NOOP, NULL);
1510                 rc = l_wait_event(sai->sai_waitq,
1511                                 ll_sa_entry_stated(entry) ||
1512                                 thread_is_stopped(&sai->sai_thread),
1513                                 &lwi);
1514                 if (rc < 0) {
1515                         ll_sai_unplug(sai, entry);
1516                         RETURN(-EAGAIN);
1517                 }
1518         }
1519
1520         if (entry->se_stat == SA_ENTRY_SUCC && entry->se_inode != NULL) {
1521                 struct inode *inode = entry->se_inode;
1522                 struct lookup_intent it = { .it_op = IT_GETATTR,
1523                                             .d.lustre.it_lock_handle =
1524                                                 entry->se_handle };
1525                 __u64 bits;
1526
1527                 rc = md_revalidate_lock(ll_i2mdexp(dir), &it,
1528                                         ll_inode2fid(inode), &bits);
1529                 if (rc == 1) {
1530                         if ((*dentryp)->d_inode == NULL) {
1531                                 struct dentry *alias;
1532
1533                                 alias = ll_splice_alias(inode, *dentryp);
1534                                 if (IS_ERR(alias)) {
1535                                         ll_sai_unplug(sai, entry);
1536                                         RETURN(PTR_ERR(alias));
1537                                 }
1538                                 *dentryp = alias;
1539                         } else if ((*dentryp)->d_inode != inode) {
1540                                 /* revalidate, but inode is recreated */
1541                                 CDEBUG(D_READA,
1542                                         "%s: stale dentry %.*s inode "
1543                                         DFID", statahead inode "DFID
1544                                         "\n",
1545                                         ll_get_fsname((*dentryp)->d_inode->i_sb,
1546                                                       NULL, 0),
1547                                         (*dentryp)->d_name.len,
1548                                         (*dentryp)->d_name.name,
1549                                         PFID(ll_inode2fid((*dentryp)->d_inode)),
1550                                         PFID(ll_inode2fid(inode)));
1551                                 ll_sai_unplug(sai, entry);
1552                                 RETURN(-ESTALE);
1553                         } else {
1554                                 iput(inode);
1555                         }
1556                         entry->se_inode = NULL;
1557
1558                         if ((bits & MDS_INODELOCK_LOOKUP) &&
1559                             d_lustre_invalid(*dentryp))
1560                                 d_lustre_revalidate(*dentryp);
1561                         ll_intent_release(&it);
1562                 }
1563         }
1564
1565         ll_sai_unplug(sai, entry);
1566         RETURN(rc);
1567 }
1568
1569 static int start_statahead_thread(struct inode *dir, struct dentry *dentry)
1570 {
1571         struct ll_inode_info *lli = ll_i2info(dir);
1572         struct ll_statahead_info *sai = NULL;
1573         struct dentry *parent;
1574         struct ptlrpc_thread *thread;
1575         struct l_wait_info lwi = { 0 };
1576         struct task_struct *task;
1577         int rc;
1578         ENTRY;
1579
1580         /* I am the "lli_opendir_pid" owner, only me can set "lli_sai". */
1581         rc = is_first_dirent(dir, dentry);
1582         if (rc == LS_NONE_FIRST_DE)
1583                 /* It is not "ls -{a}l" operation, no need statahead for it. */
1584                 GOTO(out, rc = -EAGAIN);
1585
1586         sai = ll_sai_alloc();
1587         if (sai == NULL)
1588                 GOTO(out, rc = -ENOMEM);
1589
1590         sai->sai_ls_all = (rc == LS_FIRST_DOT_DE);
1591         sai->sai_inode = igrab(dir);
1592         if (unlikely(sai->sai_inode == NULL)) {
1593                 CWARN("Do not start stat ahead on dying inode "DFID"\n",
1594                         PFID(&lli->lli_fid));
1595                 GOTO(out, rc = -ESTALE);
1596         }
1597
1598         /* get parent reference count here, and put it in ll_statahead_thread */
1599         parent = dget(dentry->d_parent);
1600         if (unlikely(sai->sai_inode != parent->d_inode)) {
1601                 struct ll_inode_info *nlli = ll_i2info(parent->d_inode);
1602
1603                 CWARN("Race condition, someone changed %.*s just now: "
1604                         "old parent "DFID", new parent "DFID"\n",
1605                         dentry->d_name.len, dentry->d_name.name,
1606                         PFID(&lli->lli_fid), PFID(&nlli->lli_fid));
1607                 dput(parent);
1608                 iput(sai->sai_inode);
1609                 GOTO(out, rc = -EAGAIN);
1610         }
1611
1612         CDEBUG(D_READA, "start statahead thread: sai %p, parent %.*s\n",
1613                sai, parent->d_name.len, parent->d_name.name);
1614
1615         lli->lli_sai = sai;
1616
1617         task = kthread_run(ll_statahead_thread, parent, "ll_sa_%u",
1618                            lli->lli_opendir_pid);
1619         thread = &sai->sai_thread;
1620         if (IS_ERR(task)) {
1621                 rc = PTR_ERR(task);
1622                 CERROR("cannot start ll_sa thread: rc = %d\n", rc);
1623                 dput(parent);
1624                 lli->lli_opendir_key = NULL;
1625                 thread_set_flags(thread, SVC_STOPPED);
1626                 thread_set_flags(&sai->sai_agl_thread, SVC_STOPPED);
1627                 ll_sai_put(sai);
1628                 LASSERT(lli->lli_sai == NULL);
1629                 RETURN(-EAGAIN);
1630         }
1631
1632         l_wait_event(thread->t_ctl_waitq,
1633                      thread_is_running(thread) || thread_is_stopped(thread),
1634                      &lwi);
1635         atomic_inc(&ll_i2sbi(parent->d_inode)->ll_sa_running);
1636         ll_sai_put(sai);
1637
1638         /*
1639          * We don't stat-ahead for the first dirent since we are already in
1640          * lookup.
1641          */
1642         RETURN(-EAGAIN);
1643
1644 out:
1645         if (sai != NULL)
1646                 OBD_FREE_PTR(sai);
1647         spin_lock(&lli->lli_sa_lock);
1648         lli->lli_opendir_key = NULL;
1649         lli->lli_opendir_pid = 0;
1650         lli->lli_sa_enabled = 0;
1651         spin_unlock(&lli->lli_sa_lock);
1652
1653         RETURN(rc);
1654 }
1655
1656 /**
1657  * Start statahead thread if this is the first dir entry.
1658  * Otherwise if a thread is started already, wait it until it is ahead of me.
1659  * \retval 1       -- find entry with lock in cache, the caller needs to do
1660  *                    nothing.
1661  * \retval 0       -- find entry in cache, but without lock, the caller needs
1662  *                    refresh from MDS.
1663  * \retval others  -- the caller need to process as non-statahead.
1664  */
1665 int do_statahead_enter(struct inode *dir, struct dentry **dentryp,
1666                        int only_unplug)
1667 {
1668         struct ll_statahead_info *sai;
1669
1670         sai = ll_sai_get(dir);
1671         if (sai != NULL) {
1672                 int rc;
1673
1674                 rc = revalidate_statahead_dentry(dir, sai, dentryp,
1675                                                  only_unplug);
1676                 CDEBUG(D_READA, "revalidate statahead %.*s: %d.\n",
1677                         (*dentryp)->d_name.len, (*dentryp)->d_name.name, rc);
1678                 ll_sai_put(sai);
1679                 return rc;
1680         }
1681
1682         return start_statahead_thread(dir, *dentryp);
1683 }