lustre/cmobd/cmobd_write.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  Copyright (c) 2002 Cluster File Systems, Inc. <info@clusterfs.com>
   5  *
   6  *   This file is part of Lustre, http://www.lustre.org.
   7  *
   8  *   Lustre is free software; you can redistribute it and/or
   9  *   modify it under the terms of version 2 of the GNU General Public
  10  *   License as published by the Free Software Foundation.
  11  *
  12  *   Lustre is distributed in the hope that it will be useful,
  13  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  *   GNU General Public License for more details.
  16  *
  17  *   You should have received a copy of the GNU General Public License
  18  *   along with Lustre; if not, write to the Free Software
  19  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  20  */
  21
  22 #define DEBUG_SUBSYSTEM S_CMOBD
  23
  24 #include <linux/version.h>
  25 #include <linux/init.h>
  26 #include <linux/obd_support.h>
  27 #include <linux/lustre_lib.h>
  28 #include <linux/lustre_net.h>
  29 #include <linux/lustre_idl.h>
  30 #include <linux/obd_class.h>
  31 #include <linux/lustre_mds.h>
  32 #include <linux/lustre_cmobd.h>
  33
  34 #include <asm/div64.h>
  35 #include <linux/pagemap.h>
  36
  37 #include "cmobd_internal.h"
  38
  39 extern kmem_cache_t *cmobd_extent_slab;
  40
  41 /* helper function to split an extent */
  42 static obd_count split_extent(struct ldlm_extent *ext, unsigned long interval)
  43 {
  44         obd_count buf_count, remainder;
  45         ENTRY;
  46
  47         buf_count = ext->end - ext->start + 1;
  48         LASSERT(buf_count > 0);
  49
  50         remainder = do_div(buf_count, interval);
  51         if (remainder)
  52                 buf_count++;
  53
  54         RETURN(buf_count);
  55 }
  56
  57 static int cmobd_ap_make_ready(void *data, int cmd)
  58 {
  59         struct cmobd_async_page *cmap = (struct cmobd_async_page *)data;
  60         struct page *page = cmap->cmap_page;
  61         ENTRY;
  62
  63         if (cmd == OBD_BRW_READ)
  64                 RETURN(0);
  65
  66         if (TryLockPage(page))
  67                 RETURN(-EAGAIN);
  68
  69         RETURN(0);
  70 }
  71
  72 static int cmobd_ap_refresh_count(void *data, int cmd)
  73 {
  74         struct cmobd_async_page *cmap = (struct cmobd_async_page *)data;
  75         struct page *page = cmap->cmap_page;
  76         struct inode *inode = page->mapping->host;
  77         ENTRY;
  78
  79         LASSERT(cmd != OBD_BRW_READ);
  80
  81         /* catch race with truncate */
  82         if (((loff_t)page->index << PAGE_SHIFT) >= inode->i_size)
  83                 RETURN(0);
  84
  85         /* catch sub-page write at end of file */
  86         if (((loff_t)page->index << PAGE_SHIFT) + PAGE_SIZE > inode->i_size)
  87                 RETURN(inode->i_size % PAGE_SIZE);
  88
  89         RETURN(PAGE_SIZE);
  90 }
  91
  92 static void cmobd_ap_fill_obdo(void *data, int cmd, struct obdo *oa)
  93 {
  94         struct cmobd_async_page *cmap = (struct cmobd_async_page *)data;
  95         obd_flag valid_flags;
  96         struct inode *inode;
  97         ENTRY;
  98
  99         if (IS_ERR(cmap)) {
 100                 EXIT;
 101                 return;
 102         }
 103
 104         inode = cmap->cmap_page->mapping->host;
 105         oa->o_id = cmap->cmap_es->es_oa.o_id;
 106         oa->o_gr = cmap->cmap_es->es_oa.o_gr;
 107         oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
 108         valid_flags = OBD_MD_FLTYPE | OBD_MD_FLATIME;
 109         if (cmd == OBD_BRW_WRITE) {
 110                 oa->o_valid |= OBD_MD_FLIFID;
 111                 mdc_pack_fid(obdo_fid(oa), inode->i_ino, 0, inode->i_mode);
 112
 113                 valid_flags |= OBD_MD_FLMTIME | OBD_MD_FLCTIME;
 114         }
 115
 116         obdo_from_inode(oa, inode, valid_flags);
 117
 118         EXIT;
 119         return;
 120 }
 121
 122 static void cmobd_ap_completion(void *data, int cmd, struct obdo *oa, int rc)
 123 {
 124         struct cmobd_async_page *cmap = (struct cmobd_async_page *)data;
 125         struct cmobd_extent_set *set = cmap->cmap_es;
 126         unsigned long flags;
 127         struct page *page;
 128         int wakeup = 0;
 129         ENTRY;
 130
 131         page = cmap->cmap_page;
 132         LASSERT(PageLocked(page));
 133
 134         /* XXX */
 135         if (rc)
 136                 SetPageError(page);
 137
 138         spin_lock_irqsave(&set->es_lock, flags);
 139         LASSERT(!list_empty(&set->es_pages));
 140         LASSERT(!list_empty(&cmap->cmap_link));
 141
 142         list_del_init(&cmap->cmap_link);
 143         if (list_empty(&set->es_pages) && !set->es_count)
 144                 wakeup = 1;
 145         spin_unlock_irqrestore(&set->es_lock, flags);
 146
 147         obd_teardown_async_page(set->es_exp, set->es_lsm, NULL,
 148                                 cmap->cmap_cookie);
 149         OBD_FREE(cmap, sizeof(*cmap));
 150
 151         unlock_page(page);
 152         page_cache_release(page);
 153
 154         if (wakeup)
 155                 wake_up(&set->es_waitq);
 156         EXIT;
 157         return;
 158 }
 159
 160 static struct obd_async_page_ops cmobd_async_page_ops = {
 161         .ap_make_ready =        cmobd_ap_make_ready,
 162         .ap_refresh_count =     cmobd_ap_refresh_count,
 163         .ap_fill_obdo =         cmobd_ap_fill_obdo,
 164         .ap_completion =        cmobd_ap_completion,
 165 };
 166
 167 static int cmobd_send_pages(struct obd_device *obd,
 168                             struct niobuf_local *lnb,
 169                             obd_count oa_bufs,
 170                             struct cmobd_extent_set *set)
 171 {
 172         struct cache_manager_obd *cmobd = &obd->u.cmobd;
 173         struct obd_export *exp = cmobd->cm_master_exp;
 174         struct cmobd_async_page *cmap = NULL;
 175         obd_count i;
 176         int rc = 0;
 177         unsigned long flags;
 178         ENTRY;
 179
 180         for (i = 0; i < oa_bufs; i++, lnb++) {
 181
 182                 OBD_ALLOC(cmap, sizeof(*cmap));
 183                 if (cmap == NULL) {
 184                         CERROR("Not enought memory\n");
 185                         rc = -ENOMEM;
 186                         break;
 187                 }
 188                 INIT_LIST_HEAD(&cmap->cmap_link);
 189                 cmap->cmap_page = lnb->page;
 190                 cmap->cmap_es = set;
 191
 192                 rc = obd_prep_async_page(exp, set->es_lsm, NULL, lnb->page,
 193                                          lnb->offset, &cmobd_async_page_ops,
 194                                          cmap, &cmap->cmap_cookie);
 195                 if (rc) {
 196                         CERROR("cmobd prep async page failed page(%p) rc(%d)\n",
 197                                lnb->page, rc);
 198                         OBD_FREE(cmap, sizeof(*cmap));
 199                         break;
 200                 }
 201
 202                 LASSERT(cmap->cmap_page);
 203                 LASSERT(!PageLocked(cmap->cmap_page));
 204                 LASSERT(Page_Uptodate(cmap->cmap_page));
 205                 page_cache_get(cmap->cmap_page);
 206
 207                 spin_lock_irqsave(&set->es_lock, flags);
 208                 list_add_tail(&cmap->cmap_link, &set->es_pages);
 209                 spin_unlock_irqrestore(&set->es_lock, flags);
 210
 211                 rc = obd_queue_async_io(exp, set->es_lsm, NULL, cmap->cmap_cookie,
 212                                         OBD_BRW_WRITE, 0, 0, 0, 0);
 213                 if (rc) {  /* try sync io */
 214                         struct obd_io_group *oig;
 215
 216                         spin_lock_irqsave(&set->es_lock, flags);
 217                         list_del_init(&cmap->cmap_link);
 218                         spin_unlock_irqrestore(&set->es_lock, flags);
 219
 220                         lock_page(cmap->cmap_page);
 221
 222                         rc = oig_init(&oig);
 223                         if (rc)
 224                                 GOTO(free_page, rc);
 225
 226                         rc = obd_queue_group_io(exp, set->es_lsm, NULL, oig,
 227                                                 cmap->cmap_cookie,
 228                                                 OBD_BRW_WRITE, 0, lnb->len, 0,
 229                                                 ASYNC_READY | ASYNC_URGENT |
 230                                                 ASYNC_COUNT_STABLE |
 231                                                 ASYNC_GROUP_SYNC);
 232
 233                         if (rc)
 234                                 GOTO(free_oig, rc);
 235
 236                         rc = obd_trigger_group_io(exp, set->es_lsm, NULL, oig);
 237                         if (rc)
 238                                 GOTO(free_oig, rc);
 239
 240                         rc = oig_wait(oig);
 241 free_oig:
 242                         oig_release(oig);
 243 free_page:
 244                         unlock_page(cmap->cmap_page);
 245                         page_cache_release(cmap->cmap_page);
 246                         obd_teardown_async_page(exp, set->es_lsm, NULL,
 247                                                 cmap->cmap_cookie);
 248                         OBD_FREE(cmap, sizeof(*cmap));
 249                         if (rc) {
 250                                 CERROR("cmobd sync io failed\n");
 251                                 break;
 252                         }
 253                 }
 254         }
 255         RETURN(rc);
 256 }
 257
 258 static int cmobd_write_extent(struct obd_device *obd,
 259                               struct cmobd_extent_info *ei)
 260 {
 261         struct cmobd_extent_set *set = ei->ei_set;
 262         struct cache_manager_obd *cmobd = &obd->u.cmobd;
 263         unsigned long flags;
 264         struct obd_ioobj ioo;
 265         struct niobuf_local *lnb;
 266         struct niobuf_remote *rnb;
 267         obd_count i, oa_bufs;
 268         struct obdo *oa;
 269         obd_off offset;
 270         int ret, rc = 0, wakeup = 0;
 271         ENTRY;
 272
 273         oa_bufs = split_extent(&ei->ei_extent, PAGE_SIZE);
 274         LASSERT(oa_bufs > 0);
 275
 276         OBD_ALLOC(lnb, oa_bufs * sizeof(struct niobuf_local));
 277         OBD_ALLOC(rnb, oa_bufs * sizeof(struct niobuf_remote));
 278         oa = obdo_alloc();
 279
 280         if (lnb == NULL || rnb == NULL || oa == NULL)
 281                 GOTO(out, rc = -ENOMEM);
 282
 283         LASSERT(ei->ei_extent.end >= ei->ei_extent.start);
 284         LASSERT((ei->ei_extent.start & (PAGE_SIZE -1)) == 0);
 285
 286         for (i = 0, offset = ei->ei_extent.start; i < oa_bufs;
 287              i++, offset += PAGE_SIZE) {
 288                 rnb[i].offset = offset;
 289                 rnb[i].len = MIN(PAGE_SIZE, ei->ei_extent.end - offset + 1);
 290         }
 291
 292         memcpy(oa, &set->es_oa, sizeof(*oa));
 293         obdo_to_ioobj(oa, &ioo);
 294         ioo.ioo_bufcnt = oa_bufs;
 295
 296         ret = obd_preprw(OBD_BRW_READ, cmobd->cm_cache_exp, oa, 1, &ioo,
 297                          oa_bufs, rnb, lnb, NULL);
 298         if (ret)
 299                 GOTO(out, rc = ret);
 300
 301         rc = cmobd_send_pages(obd, lnb, oa_bufs, set);
 302         if (rc)
 303                 CERROR("cmobd_send_pages failed %d\n", rc);
 304
 305         rc = obd_commitrw(OBD_BRW_READ, cmobd->cm_cache_exp, oa, 1, &ioo,
 306                           oa_bufs, lnb, NULL, ret);
 307
 308         /* countdown and wake up */
 309         spin_lock_irqsave(&set->es_lock, flags);
 310         LASSERT(set->es_count);
 311         set->es_count--;
 312         if (!set->es_count)
 313                 wakeup = 1;
 314         spin_unlock_irqrestore(&set->es_lock, flags);
 315
 316         if (wakeup)
 317                 wake_up(&set->es_waitq);
 318
 319 out:
 320         if (lnb)
 321                 OBD_FREE(lnb, oa_bufs * sizeof(struct niobuf_local));
 322         if (rnb)
 323                 OBD_FREE(rnb, oa_bufs * sizeof(struct niobuf_remote));
 324         if (oa)
 325                 obdo_free(oa);
 326
 327         RETURN(rc);
 328 }
 329
 330 static struct cmobd_extent_info* get_next_ei(struct cmobd_write_service *ws)
 331 {
 332         struct cmobd_extent_info *ei = NULL;
 333         unsigned long flags;
 334         int wakeup = 0;
 335
 336         spin_lock_irqsave(&ws->ws_extent_lock, flags);
 337         if (!list_empty(&ws->ws_extents)) {
 338                 ei = list_entry(ws->ws_extents.next,
 339                                 struct cmobd_extent_info, ei_link);
 340                 list_del_init(&ei->ei_link);
 341                 ws->ws_nextents--;
 342                 if (ws->ws_nextents < CMOBD_MAX_EXTENTS)
 343                         wakeup = 1;
 344         }
 345         spin_unlock_irqrestore(&ws->ws_extent_lock, flags);
 346
 347         if (wakeup)
 348                 wake_up_all(&ws->ws_waitq_provider);
 349
 350         return ei;
 351 }
 352
 353 static int cmobd_write_main(void *arg)
 354 {
 355         struct ptlrpc_svc_data *data = (struct ptlrpc_svc_data *)arg;
 356         struct ptlrpc_thread   *thread = data->thread;
 357         struct obd_device *obd = data->dev;
 358         struct cache_manager_obd *cmobd = &obd->u.cmobd;
 359         struct cmobd_write_service *ws = cmobd->cm_write_srv;
 360         struct cmobd_extent_info *extent = NULL;
 361         unsigned long flags;
 362         int rc;
 363         ENTRY;
 364
 365         lock_kernel();
 366
 367         ptlrpc_daemonize();
 368
 369         SIGNAL_MASK_LOCK(current, flags);
 370         sigfillset(&current->blocked);
 371         RECALC_SIGPENDING;
 372         SIGNAL_MASK_UNLOCK(current, flags);
 373
 374         LASSERTF(strlen(data->name) < sizeof(current->comm),
 375                  "name %d > len %d\n",strlen(data->name),sizeof(current->comm));
 376         THREAD_NAME(current->comm, sizeof(current->comm) - 1, "%s", data->name);
 377
 378         unlock_kernel();
 379
 380         thread->t_flags = SVC_RUNNING;
 381         wake_up(&thread->t_ctl_waitq);
 382
 383         /* Record that the thread is running */
 384         spin_lock_irqsave(&ws->ws_thread_lock, flags);
 385         ws->ws_nthreads++;
 386         spin_unlock_irqrestore(&ws->ws_thread_lock, flags);
 387
 388         while ((thread->t_flags & SVC_STOPPING) == 0) {
 389                 struct l_wait_info lwi = { 0 };
 390
 391                 l_wait_event_exclusive(ws->ws_waitq_consumer,
 392                                        ((thread->t_flags & SVC_STOPPING) ||
 393                                         ((extent = get_next_ei(ws)) !=
 394                                           NULL)),
 395                                        &lwi);
 396                 if (extent == NULL)
 397                         continue;
 398                 rc = cmobd_write_extent(obd, extent);
 399                 if (rc)
 400                         CERROR("write extent failed rc=%d\n", rc);
 401                 OBD_SLAB_FREE(extent, cmobd_extent_slab, sizeof(*extent));
 402                 extent = NULL;
 403         }
 404
 405         thread->t_flags = SVC_STOPPED;
 406         wake_up(&thread->t_ctl_waitq);
 407
 408         spin_lock_irqsave(&ws->ws_thread_lock, flags);
 409         ws->ws_nthreads--;                    /* must know immediately */
 410         spin_unlock_irqrestore(&ws->ws_thread_lock, flags);
 411
 412         RETURN(0);
 413 }
 414
 415 /* functions for manipulating cmobd write replay threads, similar with
 416  * ptlrpc threads functions */
 417 static int cmobd_start_thread(struct obd_device *obd, char *name)
 418 {
 419         struct cache_manager_obd *cmobd = &obd->u.cmobd;
 420         struct cmobd_write_service *ws = cmobd->cm_write_srv;
 421         struct l_wait_info lwi = { 0 };
 422         struct ptlrpc_svc_data d;
 423         struct ptlrpc_thread *thread;
 424         unsigned long flags;
 425         int rc;
 426         ENTRY;
 427
 428         OBD_ALLOC(thread, sizeof(*thread));
 429         if (thread == NULL)
 430                 RETURN(-ENOMEM);
 431         init_waitqueue_head(&thread->t_ctl_waitq);
 432
 433         d.dev = obd;
 434         d.svc = NULL;
 435         d.name = name;
 436         d.thread = thread;
 437
 438         spin_lock_irqsave(&ws->ws_thread_lock, flags);
 439         list_add(&thread->t_link, &ws->ws_threads);
 440         spin_unlock_irqrestore(&ws->ws_thread_lock, flags);
 441
 442         /* CLONE_VM and CLONE_FILES just avoid a needless copy, because we
 443          * just drop the VM and FILES in ptlrpc_daemonize() right away.
 444          */
 445         rc = kernel_thread(cmobd_write_main, &d, CLONE_VM | CLONE_FILES);
 446         if (rc < 0) {
 447                 CERROR("cannot start thread: %d\n", rc);
 448                 spin_lock_irqsave(&ws->ws_thread_lock, flags);
 449                 list_del_init(&thread->t_link);
 450                 spin_unlock_irqrestore(&ws->ws_thread_lock, flags);
 451                 OBD_FREE(thread, sizeof(*thread));
 452                 RETURN(rc);
 453         }
 454         l_wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_RUNNING, &lwi);
 455
 456         RETURN(0);
 457
 458 }
 459
 460 static void cmobd_stop_thread(struct obd_device *obd,
 461                               struct ptlrpc_thread *thread)
 462 {
 463         struct cache_manager_obd *cmobd = &obd->u.cmobd;
 464         struct cmobd_write_service *ws = cmobd->cm_write_srv;
 465         struct l_wait_info lwi = { 0 };
 466         unsigned long flags;
 467         ENTRY;
 468
 469         thread->t_flags = SVC_STOPPING;
 470         wake_up_all(&ws->ws_waitq_consumer);
 471
 472         l_wait_event(thread->t_ctl_waitq, (thread->t_flags & SVC_STOPPED),
 473                      &lwi);
 474
 475         spin_lock_irqsave(&ws->ws_thread_lock, flags);
 476         list_del(&thread->t_link);
 477         spin_unlock_irqrestore(&ws->ws_thread_lock, flags);
 478
 479         OBD_FREE(thread, sizeof(*thread));
 480         EXIT;
 481 }
 482
 483 static void cmobd_stop_all_threads(struct obd_device *obd)
 484 {
 485         struct cache_manager_obd *cmobd = &obd->u.cmobd;
 486         struct cmobd_write_service *ws = cmobd->cm_write_srv;
 487         unsigned long flags;
 488         struct ptlrpc_thread *thread;
 489         ENTRY;
 490
 491         spin_lock_irqsave(&ws->ws_thread_lock, flags);
 492         while (!list_empty(&ws->ws_threads)) {
 493                 thread = list_entry(ws->ws_threads.next,
 494                                     struct ptlrpc_thread, t_link);
 495
 496                 spin_unlock_irqrestore(&ws->ws_thread_lock, flags);
 497                 cmobd_stop_thread(obd, thread);
 498                 spin_lock_irqsave(&ws->ws_thread_lock, flags);
 499         }
 500
 501         spin_unlock_irqrestore(&ws->ws_thread_lock, flags);
 502         EXIT;
 503 }
 504
 505 static int cmobd_start_n_threads(struct obd_device *obd, int num_threads,
 506                                  char *base_name)
 507 {
 508         int i, rc = 0;
 509         ENTRY;
 510
 511         for (i = 0; i < num_threads; i++) {
 512                 char name[32];
 513                 snprintf(name, sizeof(name) - 1, "%s_%02d", base_name, i);
 514                 rc = cmobd_start_thread(obd, name);
 515                 if (rc) {
 516                         CERROR("cannot start %s thread #%d: rc %d\n", base_name,
 517                                i, rc);
 518                         cmobd_stop_all_threads(obd);
 519                 }
 520         }
 521         RETURN(rc);
 522 }
 523
 524 void cmobd_cleanup_write_srv(struct obd_device *obd)
 525 {
 526         struct cache_manager_obd *cmobd = &obd->u.cmobd;
 527         struct list_head *pos, *n;
 528         struct cmobd_extent_info *ei;
 529         ENTRY;
 530
 531         cmobd_stop_all_threads(obd);
 532
 533         list_for_each_safe(pos, n, &cmobd->cm_write_srv->ws_extents) {
 534                 ei = list_entry(pos, struct cmobd_extent_info, ei_link);
 535                 list_del_init(&ei->ei_link);
 536                 OBD_FREE(ei, sizeof(*ei));
 537         }
 538         OBD_FREE(cmobd->cm_write_srv, sizeof(*cmobd->cm_write_srv));
 539         EXIT;
 540 }
 541
 542 int cmobd_init_write_srv(struct obd_device *obd)
 543 {
 544         struct cache_manager_obd *cmobd = &obd->u.cmobd;
 545         struct cmobd_write_service *ws;
 546         int rc;
 547         ENTRY;
 548
 549         OBD_ALLOC(cmobd->cm_write_srv, sizeof(*cmobd->cm_write_srv));
 550         if (cmobd->cm_write_srv == NULL)
 551                 RETURN(-ENOMEM);
 552         ws = cmobd->cm_write_srv;
 553
 554         INIT_LIST_HEAD(&ws->ws_threads);
 555         spin_lock_init(&ws->ws_thread_lock);
 556         ws->ws_nthreads = 0;
 557
 558         INIT_LIST_HEAD(&ws->ws_extents);
 559         spin_lock_init(&ws->ws_extent_lock);
 560         ws->ws_nextents = 0;
 561         init_waitqueue_head(&ws->ws_waitq_provider);
 562         init_waitqueue_head(&ws->ws_waitq_consumer);
 563
 564         rc = cmobd_start_n_threads(obd, CMOBD_NUM_THREADS, "cm_write");
 565         if (rc)
 566                 cmobd_cleanup_write_srv(obd);
 567
 568         RETURN(rc);
 569 }
 570
 571 static int extent_queue_full(struct cmobd_write_service *ws)
 572 {
 573         unsigned long flags;
 574         int full = 0;
 575
 576         spin_lock_irqsave(&ws->ws_extent_lock, flags);
 577         full = (ws->ws_nextents >= CMOBD_MAX_EXTENTS) ? 1 : 0;
 578         spin_unlock_irqrestore(&ws->ws_extent_lock, flags);
 579
 580         return full;
 581 }
 582
 583 static void cmobd_queue_extent(struct obd_device *obd,
 584                                struct cmobd_extent_info *ex)
 585 {
 586         struct cache_manager_obd *cmobd = &obd->u.cmobd;
 587         struct cmobd_write_service *ws = cmobd->cm_write_srv;
 588         struct cmobd_extent_set *set = ex->ei_set;
 589         unsigned long flags;
 590         struct l_wait_info lwi = { 0 };
 591         ENTRY;
 592
 593 wait:
 594         l_wait_event(ws->ws_waitq_provider, !extent_queue_full(ws), &lwi);
 595
 596         spin_lock_irqsave(&ws->ws_extent_lock, flags);
 597         if (ws->ws_nextents >= CMOBD_MAX_EXTENTS) {
 598                 spin_unlock_irqrestore(&ws->ws_extent_lock, flags);
 599                 goto wait;
 600         }
 601         list_add_tail(&ex->ei_link, &ws->ws_extents);
 602         ws->ws_nextents++;
 603         spin_unlock_irqrestore(&ws->ws_extent_lock, flags);
 604
 605         spin_lock_irqsave(&set->es_lock, flags);
 606         set->es_count++;
 607         spin_unlock_irqrestore(&set->es_lock, flags);
 608
 609         wake_up_all(&ws->ws_waitq_consumer);
 610
 611         EXIT;
 612 }
 613
 614 static obd_size cmobd_fid2size(struct obd_export *exp, obd_id id, obd_gr grp)
 615 {
 616         struct lvfs_run_ctxt saved;
 617         struct dentry *de = NULL;
 618         obd_size size;
 619         ENTRY;
 620
 621         push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
 622
 623         de = obd_lvfs_fid2dentry(exp, id, 0, grp);
 624         LASSERT(de);
 625
 626         size = de->d_inode->i_size;
 627
 628         dput(de);
 629         pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
 630
 631         RETURN(size);
 632 }
 633
 634 static int extent_set_done(struct cmobd_extent_set *set, int phase)
 635 {
 636         int done = 0;
 637         unsigned long flags;
 638
 639         spin_lock_irqsave(&set->es_lock, flags);
 640         if (phase == 1)
 641                 done = set->es_count ? 0 : 1;
 642         else if (phase == 2)
 643                 done = (!set->es_count && list_empty(&set->es_pages)) ? 1 : 0;
 644         spin_unlock_irqrestore(&set->es_lock, flags);
 645
 646         return done;
 647 }
 648
 649 int cmobd_replay_write(struct obd_device *obd, struct obdo *oa,
 650                        struct ldlm_extent *ext)
 651 {
 652         struct cache_manager_obd *cmobd = &obd->u.cmobd;
 653         struct lov_obd *lov = &cmobd->cm_master_obd->u.lov;
 654         struct lov_stripe_md *lsm = NULL;
 655         struct cmobd_extent_set set;
 656         struct cmobd_extent_info *ex;
 657         struct l_wait_info lwi = { 0 };
 658         struct list_head *pos, *n;
 659         struct cmobd_async_page *cmap;
 660         unsigned long flags;
 661         obd_count i, buf_count;
 662         obd_off start;
 663         int rc = 0;
 664         ENTRY;
 665
 666         rc = cmobd_dummy_lsm(&lsm, lov->desc.ld_tgt_count, oa,
 667                              (__u32)lov->desc.ld_default_stripe_size);
 668         if (rc)
 669                 RETURN(-ENOMEM);
 670
 671         set.es_extent.start = ext->start;
 672         set.es_extent.end = ext->end;
 673         set.es_lsm = lsm;
 674         set.es_exp = cmobd->cm_master_exp;
 675         set.es_ext_sz = CMOBD_MAX_EXTENT_SZ;
 676         set.es_count = 0;
 677         memcpy(&set.es_oa, oa, sizeof(*oa));
 678
 679         INIT_LIST_HEAD(&set.es_pages);
 680         spin_lock_init(&set.es_lock);
 681         init_waitqueue_head(&set.es_waitq);
 682
 683         if (set.es_extent.end < set.es_extent.start) {
 684                 CDEBUG(D_HA, "illegal extent in write replay\n");
 685                 GOTO(out, rc = -EINVAL);
 686         }
 687         /* start of extent is extended to page boundaries */
 688         set.es_extent.start -= set.es_extent.start & ~PAGE_MASK;
 689         /* if the end of extent is EOF, set it as file size */
 690         if (set.es_extent.end == OBD_OBJECT_EOF) {
 691                 set.es_extent.end = cmobd_fid2size(cmobd->cm_cache_exp,
 692                                                    oa->o_id, oa->o_gr) - 1;
 693                 if (set.es_extent.end <= 0)
 694                         GOTO(out, rc = 0);
 695         }
 696
 697         buf_count = split_extent(&set.es_extent, set.es_ext_sz);
 698         for (i = 0, start = set.es_extent.start; i < buf_count;
 699              i++, start += set.es_ext_sz) {
 700                 OBD_SLAB_ALLOC(ex, cmobd_extent_slab, SLAB_NOFS, sizeof(*ex));
 701                 if (ex == NULL) {
 702                         CERROR("not enough memory\n");
 703                         break;
 704                 }
 705
 706                 INIT_LIST_HEAD(&ex->ei_link);
 707                 ex->ei_set = &set;
 708                 ex->ei_extent.start = start;
 709                 ex->ei_extent.end = start + set.es_ext_sz - 1;
 710                 if (ex->ei_extent.end > set.es_extent.end)
 711                         ex->ei_extent.end = set.es_extent.end;
 712
 713                 cmobd_queue_extent(obd, ex);
 714         }
 715
 716         l_wait_event(set.es_waitq, extent_set_done(&set, 1), &lwi);
 717
 718         /* fire remaining ios */
 719         spin_lock_irqsave(&set.es_lock, flags);
 720         list_for_each_safe (pos, n, &set.es_pages) {
 721                 cmap = list_entry(pos, struct cmobd_async_page, cmap_link);
 722
 723                 /* locked pages are in flight */
 724                 if (PageLocked(cmap->cmap_page))
 725                         continue;
 726
 727                 spin_unlock_irqrestore(&set.es_lock, flags);
 728                 rc = obd_set_async_flags(set.es_exp, set.es_lsm, NULL,
 729                                          cmap->cmap_cookie,
 730                                          ASYNC_URGENT);
 731                 if (rc)
 732                         CERROR("cmobd set async flags failed\n");
 733                 spin_lock_irqsave(&set.es_lock, flags);
 734                 break;
 735         }
 736         spin_unlock_irqrestore(&set.es_lock, flags);
 737
 738         l_wait_event(set.es_waitq, extent_set_done(&set, 2), &lwi);
 739 out:
 740         cmobd_free_lsm(&lsm);
 741         RETURN(rc);
 742 }