drivers/staging/lustre/lustre/osc/osc_io.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, 2012, 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  * Implementation of cl_io for OSC layer.
  37  *
  38  *   Author: Nikita Danilov <nikita.danilov@sun.com>
  39  *   Author: Jinshan Xiong <jinshan.xiong@whamcloud.com>
  40  */
  41
  42 #define DEBUG_SUBSYSTEM S_OSC
  43
  44 #include "osc_cl_internal.h"
  45
  46 /** \addtogroup osc
  47  *  @{
  48  */
  49
  50 /*****************************************************************************
  51  *
  52  * Type conversions.
  53  *
  54  */
  55
  56 static struct osc_req *cl2osc_req(const struct cl_req_slice *slice)
  57 {
  58         LINVRNT(slice->crs_dev->cd_lu_dev.ld_type == &osc_device_type);
  59         return container_of0(slice, struct osc_req, or_cl);
  60 }
  61
  62 static struct osc_io *cl2osc_io(const struct lu_env *env,
  63                                 const struct cl_io_slice *slice)
  64 {
  65         struct osc_io *oio = container_of0(slice, struct osc_io, oi_cl);
  66         LINVRNT(oio == osc_env_io(env));
  67         return oio;
  68 }
  69
  70 static struct osc_page *osc_cl_page_osc(struct cl_page *page)
  71 {
  72         const struct cl_page_slice *slice;
  73
  74         slice = cl_page_at(page, &osc_device_type);
  75         LASSERT(slice != NULL);
  76
  77         return cl2osc_page(slice);
  78 }
  79
  80
  81 /*****************************************************************************
  82  *
  83  * io operations.
  84  *
  85  */
  86
  87 static void osc_io_fini(const struct lu_env *env, const struct cl_io_slice *io)
  88 {
  89 }
  90
  91 /**
  92  * An implementation of cl_io_operations::cio_io_submit() method for osc
  93  * layer. Iterates over pages in the in-queue, prepares each for io by calling
  94  * cl_page_prep() and then either submits them through osc_io_submit_page()
  95  * or, if page is already submitted, changes osc flags through
  96  * osc_set_async_flags().
  97  */
  98 static int osc_io_submit(const struct lu_env *env,
  99                          const struct cl_io_slice *ios,
 100                          enum cl_req_type crt, struct cl_2queue *queue)
 101 {
 102         struct cl_page    *page;
 103         struct cl_page    *tmp;
 104         struct client_obd *cli  = NULL;
 105         struct osc_object *osc  = NULL; /* to keep gcc happy */
 106         struct osc_page   *opg;
 107         struct cl_io      *io;
 108         LIST_HEAD(list);
 109
 110         struct cl_page_list *qin      = &queue->c2_qin;
 111         struct cl_page_list *qout     = &queue->c2_qout;
 112         int queued = 0;
 113         int result = 0;
 114         int cmd;
 115         int brw_flags;
 116         int max_pages;
 117
 118         LASSERT(qin->pl_nr > 0);
 119
 120         CDEBUG(D_CACHE, "%d %d\n", qin->pl_nr, crt);
 121
 122         osc = cl2osc(ios->cis_obj);
 123         cli = osc_cli(osc);
 124         max_pages = cli->cl_max_pages_per_rpc;
 125
 126         cmd = crt == CRT_WRITE ? OBD_BRW_WRITE : OBD_BRW_READ;
 127         brw_flags = osc_io_srvlock(cl2osc_io(env, ios)) ? OBD_BRW_SRVLOCK : 0;
 128
 129         /*
 130          * NOTE: here @page is a top-level page. This is done to avoid
 131          *       creation of sub-page-list.
 132          */
 133         cl_page_list_for_each_safe(page, tmp, qin) {
 134                 struct osc_async_page *oap;
 135
 136                 /* Top level IO. */
 137                 io = page->cp_owner;
 138                 LASSERT(io != NULL);
 139
 140                 opg = osc_cl_page_osc(page);
 141                 oap = &opg->ops_oap;
 142                 LASSERT(osc == oap->oap_obj);
 143
 144                 if (!list_empty(&oap->oap_pending_item) ||
 145                     !list_empty(&oap->oap_rpc_item)) {
 146                         CDEBUG(D_CACHE, "Busy oap %p page %p for submit.\n",
 147                                oap, opg);
 148                         result = -EBUSY;
 149                         break;
 150                 }
 151
 152                 result = cl_page_prep(env, io, page, crt);
 153                 if (result != 0) {
 154                         LASSERT(result < 0);
 155                         if (result != -EALREADY)
 156                                 break;
 157                         /*
 158                          * Handle -EALREADY error: for read case, the page is
 159                          * already in UPTODATE state; for write, the page
 160                          * is not dirty.
 161                          */
 162                         result = 0;
 163                         continue;
 164                 }
 165
 166                 cl_page_list_move(qout, qin, page);
 167                 oap->oap_async_flags = ASYNC_URGENT|ASYNC_READY;
 168                 oap->oap_async_flags |= ASYNC_COUNT_STABLE;
 169
 170                 osc_page_submit(env, opg, crt, brw_flags);
 171                 list_add_tail(&oap->oap_pending_item, &list);
 172                 if (++queued == max_pages) {
 173                         queued = 0;
 174                         result = osc_queue_sync_pages(env, osc, &list, cmd,
 175                                                       brw_flags);
 176                         if (result < 0)
 177                                 break;
 178                 }
 179         }
 180
 181         if (queued > 0)
 182                 result = osc_queue_sync_pages(env, osc, &list, cmd, brw_flags);
 183
 184         CDEBUG(D_INFO, "%d/%d %d\n", qin->pl_nr, qout->pl_nr, result);
 185         return qout->pl_nr > 0 ? 0 : result;
 186 }
 187
 188 static void osc_page_touch_at(const struct lu_env *env,
 189                               struct cl_object *obj, pgoff_t idx, unsigned to)
 190 {
 191         struct lov_oinfo  *loi  = cl2osc(obj)->oo_oinfo;
 192         struct cl_attr    *attr = &osc_env_info(env)->oti_attr;
 193         int valid;
 194         __u64 kms;
 195
 196         /* offset within stripe */
 197         kms = cl_offset(obj, idx) + to;
 198
 199         cl_object_attr_lock(obj);
 200         /*
 201          * XXX old code used
 202          *
 203          *       ll_inode_size_lock(inode, 0); lov_stripe_lock(lsm);
 204          *
 205          * here
 206          */
 207         CDEBUG(D_INODE, "stripe KMS %sincreasing "LPU64"->"LPU64" "LPU64"\n",
 208                kms > loi->loi_kms ? "" : "not ", loi->loi_kms, kms,
 209                loi->loi_lvb.lvb_size);
 210
 211         valid = 0;
 212         if (kms > loi->loi_kms) {
 213                 attr->cat_kms = kms;
 214                 valid |= CAT_KMS;
 215         }
 216         if (kms > loi->loi_lvb.lvb_size) {
 217                 attr->cat_size = kms;
 218                 valid |= CAT_SIZE;
 219         }
 220         cl_object_attr_set(env, obj, attr, valid);
 221         cl_object_attr_unlock(obj);
 222 }
 223
 224 /**
 225  * This is called when a page is accessed within file in a way that creates
 226  * new page, if one were missing (i.e., if there were a hole at that place in
 227  * the file, or accessed page is beyond the current file size). Examples:
 228  * ->commit_write() and ->nopage() methods.
 229  *
 230  * Expand stripe KMS if necessary.
 231  */
 232 static void osc_page_touch(const struct lu_env *env,
 233                            struct osc_page *opage, unsigned to)
 234 {
 235         struct cl_page    *page = opage->ops_cl.cpl_page;
 236         struct cl_object  *obj  = opage->ops_cl.cpl_obj;
 237
 238         osc_page_touch_at(env, obj, page->cp_index, to);
 239 }
 240
 241 /**
 242  * Implements cl_io_operations::cio_prepare_write() method for osc layer.
 243  *
 244  * \retval -EIO transfer initiated against this osc will most likely fail
 245  * \retval 0    transfer initiated against this osc will most likely succeed.
 246  *
 247  * The reason for this check is to immediately return an error to the caller
 248  * in the case of a deactivated import. Note, that import can be deactivated
 249  * later, while pages, dirtied by this IO, are still in the cache, but this is
 250  * irrelevant, because that would still return an error to the application (if
 251  * it does fsync), but many applications don't do fsync because of performance
 252  * issues, and we wanted to return an -EIO at write time to notify the
 253  * application.
 254  */
 255 static int osc_io_prepare_write(const struct lu_env *env,
 256                                 const struct cl_io_slice *ios,
 257                                 const struct cl_page_slice *slice,
 258                                 unsigned from, unsigned to)
 259 {
 260         struct osc_device *dev = lu2osc_dev(slice->cpl_obj->co_lu.lo_dev);
 261         struct obd_import *imp = class_exp2cliimp(dev->od_exp);
 262         struct osc_io     *oio = cl2osc_io(env, ios);
 263         int result = 0;
 264
 265         /*
 266          * This implements OBD_BRW_CHECK logic from old client.
 267          */
 268
 269         if (imp == NULL || imp->imp_invalid)
 270                 result = -EIO;
 271         if (result == 0 && oio->oi_lockless)
 272                 /* this page contains `invalid' data, but who cares?
 273                  * nobody can access the invalid data.
 274                  * in osc_io_commit_write(), we're going to write exact
 275                  * [from, to) bytes of this page to OST. -jay */
 276                 cl_page_export(env, slice->cpl_page, 1);
 277
 278         return result;
 279 }
 280
 281 static int osc_io_commit_write(const struct lu_env *env,
 282                                const struct cl_io_slice *ios,
 283                                const struct cl_page_slice *slice,
 284                                unsigned from, unsigned to)
 285 {
 286         struct osc_io    *oio = cl2osc_io(env, ios);
 287         struct osc_page       *opg = cl2osc_page(slice);
 288         struct osc_object     *obj = cl2osc(opg->ops_cl.cpl_obj);
 289         struct osc_async_page *oap = &opg->ops_oap;
 290
 291         LASSERT(to > 0);
 292         /*
 293          * XXX instead of calling osc_page_touch() here and in
 294          * osc_io_fault_start() it might be more logical to introduce
 295          * cl_page_touch() method, that generic cl_io_commit_write() and page
 296          * fault code calls.
 297          */
 298         osc_page_touch(env, cl2osc_page(slice), to);
 299         if (!client_is_remote(osc_export(obj)) &&
 300             capable(CFS_CAP_SYS_RESOURCE))
 301                 oap->oap_brw_flags |= OBD_BRW_NOQUOTA;
 302
 303         if (oio->oi_lockless)
 304                 /* see osc_io_prepare_write() for lockless io handling. */
 305                 cl_page_clip(env, slice->cpl_page, from, to);
 306
 307         return 0;
 308 }
 309
 310 static int osc_io_fault_start(const struct lu_env *env,
 311                               const struct cl_io_slice *ios)
 312 {
 313         struct cl_io       *io;
 314         struct cl_fault_io *fio;
 315
 316         io  = ios->cis_io;
 317         fio = &io->u.ci_fault;
 318         CDEBUG(D_INFO, "%lu %d %d\n",
 319                fio->ft_index, fio->ft_writable, fio->ft_nob);
 320         /*
 321          * If mapping is writeable, adjust kms to cover this page,
 322          * but do not extend kms beyond actual file size.
 323          * See bug 10919.
 324          */
 325         if (fio->ft_writable)
 326                 osc_page_touch_at(env, ios->cis_obj,
 327                                   fio->ft_index, fio->ft_nob);
 328         return 0;
 329 }
 330
 331 static int osc_async_upcall(void *a, int rc)
 332 {
 333         struct osc_async_cbargs *args = a;
 334
 335         args->opc_rc = rc;
 336         complete(&args->opc_sync);
 337         return 0;
 338 }
 339
 340 /**
 341  * Checks that there are no pages being written in the extent being truncated.
 342  */
 343 static int trunc_check_cb(const struct lu_env *env, struct cl_io *io,
 344                           struct cl_page *page, void *cbdata)
 345 {
 346         const struct cl_page_slice *slice;
 347         struct osc_page *ops;
 348         struct osc_async_page *oap;
 349         __u64 start = *(__u64 *)cbdata;
 350
 351         slice = cl_page_at(page, &osc_device_type);
 352         LASSERT(slice != NULL);
 353         ops = cl2osc_page(slice);
 354         oap = &ops->ops_oap;
 355
 356         if (oap->oap_cmd & OBD_BRW_WRITE &&
 357             !list_empty(&oap->oap_pending_item))
 358                 CL_PAGE_DEBUG(D_ERROR, env, page, "exists " LPU64 "/%s.\n",
 359                                 start, current->comm);
 360
 361         {
 362                 struct page *vmpage = cl_page_vmpage(env, page);
 363                 if (PageLocked(vmpage))
 364                         CDEBUG(D_CACHE, "page %p index %lu locked for %d.\n",
 365                                ops, page->cp_index,
 366                                (oap->oap_cmd & OBD_BRW_RWMASK));
 367         }
 368
 369         return CLP_GANG_OKAY;
 370 }
 371
 372 static void osc_trunc_check(const struct lu_env *env, struct cl_io *io,
 373                             struct osc_io *oio, __u64 size)
 374 {
 375         struct cl_object *clob;
 376         int     partial;
 377         pgoff_t start;
 378
 379         clob    = oio->oi_cl.cis_obj;
 380         start   = cl_index(clob, size);
 381         partial = cl_offset(clob, start) < size;
 382
 383         /*
 384          * Complain if there are pages in the truncated region.
 385          */
 386         cl_page_gang_lookup(env, clob, io, start + partial, CL_PAGE_EOF,
 387                             trunc_check_cb, (void *)&size);
 388 }
 389
 390 static int osc_io_setattr_start(const struct lu_env *env,
 391                                 const struct cl_io_slice *slice)
 392 {
 393         struct cl_io        *io     = slice->cis_io;
 394         struct osc_io      *oio    = cl2osc_io(env, slice);
 395         struct cl_object        *obj    = slice->cis_obj;
 396         struct lov_oinfo        *loi    = cl2osc(obj)->oo_oinfo;
 397         struct cl_attr    *attr   = &osc_env_info(env)->oti_attr;
 398         struct obdo          *oa     = &oio->oi_oa;
 399         struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
 400         __u64               size   = io->u.ci_setattr.sa_attr.lvb_size;
 401         unsigned int         ia_valid = io->u.ci_setattr.sa_valid;
 402         int                   result = 0;
 403         struct obd_info   oinfo = { { { 0 } } };
 404
 405         /* truncate cache dirty pages first */
 406         if (cl_io_is_trunc(io))
 407                 result = osc_cache_truncate_start(env, oio, cl2osc(obj), size);
 408
 409         if (result == 0 && oio->oi_lockless == 0) {
 410                 cl_object_attr_lock(obj);
 411                 result = cl_object_attr_get(env, obj, attr);
 412                 if (result == 0) {
 413                         struct ost_lvb *lvb = &io->u.ci_setattr.sa_attr;
 414                         unsigned int cl_valid = 0;
 415
 416                         if (ia_valid & ATTR_SIZE) {
 417                                 attr->cat_size = attr->cat_kms = size;
 418                                 cl_valid = (CAT_SIZE | CAT_KMS);
 419                         }
 420                         if (ia_valid & ATTR_MTIME_SET) {
 421                                 attr->cat_mtime = lvb->lvb_mtime;
 422                                 cl_valid |= CAT_MTIME;
 423                         }
 424                         if (ia_valid & ATTR_ATIME_SET) {
 425                                 attr->cat_atime = lvb->lvb_atime;
 426                                 cl_valid |= CAT_ATIME;
 427                         }
 428                         if (ia_valid & ATTR_CTIME_SET) {
 429                                 attr->cat_ctime = lvb->lvb_ctime;
 430                                 cl_valid |= CAT_CTIME;
 431                         }
 432                         result = cl_object_attr_set(env, obj, attr, cl_valid);
 433                 }
 434                 cl_object_attr_unlock(obj);
 435         }
 436         memset(oa, 0, sizeof(*oa));
 437         if (result == 0) {
 438                 oa->o_oi = loi->loi_oi;
 439                 oa->o_mtime = attr->cat_mtime;
 440                 oa->o_atime = attr->cat_atime;
 441                 oa->o_ctime = attr->cat_ctime;
 442                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP | OBD_MD_FLATIME |
 443                         OBD_MD_FLCTIME | OBD_MD_FLMTIME;
 444                 if (ia_valid & ATTR_SIZE) {
 445                         oa->o_size = size;
 446                         oa->o_blocks = OBD_OBJECT_EOF;
 447                         oa->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 448
 449                         if (oio->oi_lockless) {
 450                                 oa->o_flags = OBD_FL_SRVLOCK;
 451                                 oa->o_valid |= OBD_MD_FLFLAGS;
 452                         }
 453                 } else {
 454                         LASSERT(oio->oi_lockless == 0);
 455                 }
 456
 457                 oinfo.oi_oa = oa;
 458                 oinfo.oi_capa = io->u.ci_setattr.sa_capa;
 459                 init_completion(&cbargs->opc_sync);
 460
 461                 if (ia_valid & ATTR_SIZE)
 462                         result = osc_punch_base(osc_export(cl2osc(obj)),
 463                                                 &oinfo, osc_async_upcall,
 464                                                 cbargs, PTLRPCD_SET);
 465                 else
 466                         result = osc_setattr_async_base(osc_export(cl2osc(obj)),
 467                                                         &oinfo, NULL,
 468                                                         osc_async_upcall,
 469                                                         cbargs, PTLRPCD_SET);
 470                 cbargs->opc_rpc_sent = result == 0;
 471         }
 472         return result;
 473 }
 474
 475 static void osc_io_setattr_end(const struct lu_env *env,
 476                                const struct cl_io_slice *slice)
 477 {
 478         struct cl_io     *io  = slice->cis_io;
 479         struct osc_io    *oio = cl2osc_io(env, slice);
 480         struct cl_object *obj = slice->cis_obj;
 481         struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
 482         int result = 0;
 483
 484         if (cbargs->opc_rpc_sent) {
 485                 wait_for_completion(&cbargs->opc_sync);
 486                 result = io->ci_result = cbargs->opc_rc;
 487         }
 488         if (result == 0) {
 489                 if (oio->oi_lockless) {
 490                         /* lockless truncate */
 491                         struct osc_device *osd = lu2osc_dev(obj->co_lu.lo_dev);
 492
 493                         LASSERT(cl_io_is_trunc(io));
 494                         /* XXX: Need a lock. */
 495                         osd->od_stats.os_lockless_truncates++;
 496                 }
 497         }
 498
 499         if (cl_io_is_trunc(io)) {
 500                 __u64 size = io->u.ci_setattr.sa_attr.lvb_size;
 501                 osc_trunc_check(env, io, oio, size);
 502                 if (oio->oi_trunc != NULL) {
 503                         osc_cache_truncate_end(env, oio, cl2osc(obj));
 504                         oio->oi_trunc = NULL;
 505                 }
 506         }
 507 }
 508
 509 static int osc_io_read_start(const struct lu_env *env,
 510                              const struct cl_io_slice *slice)
 511 {
 512         struct osc_io    *oio   = cl2osc_io(env, slice);
 513         struct cl_object *obj   = slice->cis_obj;
 514         struct cl_attr   *attr  = &osc_env_info(env)->oti_attr;
 515         int rc = 0;
 516
 517         if (oio->oi_lockless == 0 && !slice->cis_io->ci_noatime) {
 518                 cl_object_attr_lock(obj);
 519                 attr->cat_atime = LTIME_S(CURRENT_TIME);
 520                 rc = cl_object_attr_set(env, obj, attr, CAT_ATIME);
 521                 cl_object_attr_unlock(obj);
 522         }
 523         return rc;
 524 }
 525
 526 static int osc_io_write_start(const struct lu_env *env,
 527                               const struct cl_io_slice *slice)
 528 {
 529         struct osc_io    *oio   = cl2osc_io(env, slice);
 530         struct cl_object *obj   = slice->cis_obj;
 531         struct cl_attr   *attr  = &osc_env_info(env)->oti_attr;
 532         int           result = 0;
 533
 534         if (oio->oi_lockless == 0) {
 535                 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_DELAY_SETTIME, 1);
 536                 cl_object_attr_lock(obj);
 537                 result = cl_object_attr_get(env, obj, attr);
 538                 if (result == 0) {
 539                         attr->cat_mtime = attr->cat_ctime =
 540                                 LTIME_S(CURRENT_TIME);
 541                         result = cl_object_attr_set(env, obj, attr,
 542                                                     CAT_MTIME | CAT_CTIME);
 543                 }
 544                 cl_object_attr_unlock(obj);
 545         }
 546         return result;
 547 }
 548
 549 static int osc_fsync_ost(const struct lu_env *env, struct osc_object *obj,
 550                          struct cl_fsync_io *fio)
 551 {
 552         struct osc_io    *oio   = osc_env_io(env);
 553         struct obdo      *oa    = &oio->oi_oa;
 554         struct obd_info  *oinfo = &oio->oi_info;
 555         struct lov_oinfo *loi   = obj->oo_oinfo;
 556         struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
 557         int rc = 0;
 558
 559         memset(oa, 0, sizeof(*oa));
 560         oa->o_oi = loi->loi_oi;
 561         oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
 562
 563         /* reload size abd blocks for start and end of sync range */
 564         oa->o_size = fio->fi_start;
 565         oa->o_blocks = fio->fi_end;
 566         oa->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 567
 568         obdo_set_parent_fid(oa, fio->fi_fid);
 569
 570         memset(oinfo, 0, sizeof(*oinfo));
 571         oinfo->oi_oa = oa;
 572         oinfo->oi_capa = fio->fi_capa;
 573         init_completion(&cbargs->opc_sync);
 574
 575         rc = osc_sync_base(osc_export(obj), oinfo, osc_async_upcall, cbargs,
 576                            PTLRPCD_SET);
 577         return rc;
 578 }
 579
 580 static int osc_io_fsync_start(const struct lu_env *env,
 581                               const struct cl_io_slice *slice)
 582 {
 583         struct cl_io       *io  = slice->cis_io;
 584         struct cl_fsync_io *fio = &io->u.ci_fsync;
 585         struct cl_object   *obj = slice->cis_obj;
 586         struct osc_object  *osc = cl2osc(obj);
 587         pgoff_t start  = cl_index(obj, fio->fi_start);
 588         pgoff_t end    = cl_index(obj, fio->fi_end);
 589         int     result = 0;
 590
 591         if (fio->fi_end == OBD_OBJECT_EOF)
 592                 end = CL_PAGE_EOF;
 593
 594         result = osc_cache_writeback_range(env, osc, start, end, 0,
 595                                            fio->fi_mode == CL_FSYNC_DISCARD);
 596         if (result > 0) {
 597                 fio->fi_nr_written += result;
 598                 result = 0;
 599         }
 600         if (fio->fi_mode == CL_FSYNC_ALL) {
 601                 int rc;
 602
 603                 /* we have to wait for writeback to finish before we can
 604                  * send OST_SYNC RPC. This is bad because it causes extents
 605                  * to be written osc by osc. However, we usually start
 606                  * writeback before CL_FSYNC_ALL so this won't have any real
 607                  * problem. */
 608                 rc = osc_cache_wait_range(env, osc, start, end);
 609                 if (result == 0)
 610                         result = rc;
 611                 rc = osc_fsync_ost(env, osc, fio);
 612                 if (result == 0)
 613                         result = rc;
 614         }
 615
 616         return result;
 617 }
 618
 619 static void osc_io_fsync_end(const struct lu_env *env,
 620                              const struct cl_io_slice *slice)
 621 {
 622         struct cl_fsync_io *fio = &slice->cis_io->u.ci_fsync;
 623         struct cl_object   *obj = slice->cis_obj;
 624         pgoff_t start = cl_index(obj, fio->fi_start);
 625         pgoff_t end   = cl_index(obj, fio->fi_end);
 626         int result = 0;
 627
 628         if (fio->fi_mode == CL_FSYNC_LOCAL) {
 629                 result = osc_cache_wait_range(env, cl2osc(obj), start, end);
 630         } else if (fio->fi_mode == CL_FSYNC_ALL) {
 631                 struct osc_io      *oio    = cl2osc_io(env, slice);
 632                 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
 633
 634                 wait_for_completion(&cbargs->opc_sync);
 635                 if (result == 0)
 636                         result = cbargs->opc_rc;
 637         }
 638         slice->cis_io->ci_result = result;
 639 }
 640
 641 static void osc_io_end(const struct lu_env *env,
 642                        const struct cl_io_slice *slice)
 643 {
 644         struct osc_io *oio = cl2osc_io(env, slice);
 645
 646         if (oio->oi_active) {
 647                 osc_extent_release(env, oio->oi_active);
 648                 oio->oi_active = NULL;
 649         }
 650 }
 651
 652 static const struct cl_io_operations osc_io_ops = {
 653         .op = {
 654                 [CIT_READ] = {
 655                         .cio_start  = osc_io_read_start,
 656                         .cio_fini   = osc_io_fini
 657                 },
 658                 [CIT_WRITE] = {
 659                         .cio_start  = osc_io_write_start,
 660                         .cio_end    = osc_io_end,
 661                         .cio_fini   = osc_io_fini
 662                 },
 663                 [CIT_SETATTR] = {
 664                         .cio_start  = osc_io_setattr_start,
 665                         .cio_end    = osc_io_setattr_end
 666                 },
 667                 [CIT_FAULT] = {
 668                         .cio_start  = osc_io_fault_start,
 669                         .cio_end    = osc_io_end,
 670                         .cio_fini   = osc_io_fini
 671                 },
 672                 [CIT_FSYNC] = {
 673                         .cio_start  = osc_io_fsync_start,
 674                         .cio_end    = osc_io_fsync_end,
 675                         .cio_fini   = osc_io_fini
 676                 },
 677                 [CIT_MISC] = {
 678                         .cio_fini   = osc_io_fini
 679                 }
 680         },
 681         .req_op = {
 682                  [CRT_READ] = {
 683                          .cio_submit    = osc_io_submit
 684                  },
 685                  [CRT_WRITE] = {
 686                          .cio_submit    = osc_io_submit
 687                  }
 688          },
 689         .cio_prepare_write = osc_io_prepare_write,
 690         .cio_commit_write  = osc_io_commit_write
 691 };
 692
 693 /*****************************************************************************
 694  *
 695  * Transfer operations.
 696  *
 697  */
 698
 699 static int osc_req_prep(const struct lu_env *env,
 700                         const struct cl_req_slice *slice)
 701 {
 702         return 0;
 703 }
 704
 705 static void osc_req_completion(const struct lu_env *env,
 706                                const struct cl_req_slice *slice, int ioret)
 707 {
 708         struct osc_req *or;
 709
 710         or = cl2osc_req(slice);
 711         OBD_SLAB_FREE_PTR(or, osc_req_kmem);
 712 }
 713
 714 /**
 715  * Implementation of struct cl_req_operations::cro_attr_set() for osc
 716  * layer. osc is responsible for struct obdo::o_id and struct obdo::o_seq
 717  * fields.
 718  */
 719 static void osc_req_attr_set(const struct lu_env *env,
 720                              const struct cl_req_slice *slice,
 721                              const struct cl_object *obj,
 722                              struct cl_req_attr *attr, obd_valid flags)
 723 {
 724         struct lov_oinfo *oinfo;
 725         struct cl_req    *clerq;
 726         struct cl_page   *apage; /* _some_ page in @clerq */
 727         struct cl_lock   *lock;  /* _some_ lock protecting @apage */
 728         struct osc_lock  *olck;
 729         struct osc_page  *opg;
 730         struct obdo      *oa;
 731         struct ost_lvb   *lvb;
 732
 733         oinfo   = cl2osc(obj)->oo_oinfo;
 734         lvb     = &oinfo->loi_lvb;
 735         oa      = attr->cra_oa;
 736
 737         if ((flags & OBD_MD_FLMTIME) != 0) {
 738                 oa->o_mtime = lvb->lvb_mtime;
 739                 oa->o_valid |= OBD_MD_FLMTIME;
 740         }
 741         if ((flags & OBD_MD_FLATIME) != 0) {
 742                 oa->o_atime = lvb->lvb_atime;
 743                 oa->o_valid |= OBD_MD_FLATIME;
 744         }
 745         if ((flags & OBD_MD_FLCTIME) != 0) {
 746                 oa->o_ctime = lvb->lvb_ctime;
 747                 oa->o_valid |= OBD_MD_FLCTIME;
 748         }
 749         if (flags & OBD_MD_FLGROUP) {
 750                 ostid_set_seq(&oa->o_oi, ostid_seq(&oinfo->loi_oi));
 751                 oa->o_valid |= OBD_MD_FLGROUP;
 752         }
 753         if (flags & OBD_MD_FLID) {
 754                 ostid_set_id(&oa->o_oi, ostid_id(&oinfo->loi_oi));
 755                 oa->o_valid |= OBD_MD_FLID;
 756         }
 757         if (flags & OBD_MD_FLHANDLE) {
 758                 clerq = slice->crs_req;
 759                 LASSERT(!list_empty(&clerq->crq_pages));
 760                 apage = container_of(clerq->crq_pages.next,
 761                                      struct cl_page, cp_flight);
 762                 opg = osc_cl_page_osc(apage);
 763                 apage = opg->ops_cl.cpl_page; /* now apage is a sub-page */
 764                 lock = cl_lock_at_page(env, apage->cp_obj, apage, NULL, 1, 1);
 765                 if (lock == NULL) {
 766                         struct cl_object_header *head;
 767                         struct cl_lock    *scan;
 768
 769                         head = cl_object_header(apage->cp_obj);
 770                         list_for_each_entry(scan, &head->coh_locks,
 771                                                 cll_linkage)
 772                                 CL_LOCK_DEBUG(D_ERROR, env, scan,
 773                                               "no cover page!\n");
 774                         CL_PAGE_DEBUG(D_ERROR, env, apage,
 775                                       "dump uncover page!\n");
 776                         dump_stack();
 777                         LBUG();
 778                 }
 779
 780                 olck = osc_lock_at(lock);
 781                 LASSERT(olck != NULL);
 782                 LASSERT(ergo(opg->ops_srvlock, olck->ols_lock == NULL));
 783                 /* check for lockless io. */
 784                 if (olck->ols_lock != NULL) {
 785                         oa->o_handle = olck->ols_lock->l_remote_handle;
 786                         oa->o_valid |= OBD_MD_FLHANDLE;
 787                 }
 788                 cl_lock_put(env, lock);
 789         }
 790 }
 791
 792 static const struct cl_req_operations osc_req_ops = {
 793         .cro_prep       = osc_req_prep,
 794         .cro_attr_set   = osc_req_attr_set,
 795         .cro_completion = osc_req_completion
 796 };
 797
 798
 799 int osc_io_init(const struct lu_env *env,
 800                 struct cl_object *obj, struct cl_io *io)
 801 {
 802         struct osc_io *oio = osc_env_io(env);
 803
 804         CL_IO_SLICE_CLEAN(oio, oi_cl);
 805         cl_io_slice_add(io, &oio->oi_cl, obj, &osc_io_ops);
 806         return 0;
 807 }
 808
 809 int osc_req_init(const struct lu_env *env, struct cl_device *dev,
 810                  struct cl_req *req)
 811 {
 812         struct osc_req *or;
 813         int result;
 814
 815         OBD_SLAB_ALLOC_PTR_GFP(or, osc_req_kmem, __GFP_IO);
 816         if (or != NULL) {
 817                 cl_req_slice_add(req, &or->or_cl, dev, &osc_req_ops);
 818                 result = 0;
 819         } else
 820                 result = -ENOMEM;
 821         return result;
 822 }
 823
 824 /** @} osc */