lustre/liblustre/llite_cl.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *   Copyright (c) 2007 Cluster File Systems, Inc.
   5  *   Author: Nikita Danilov <nikita@clusterfs.com>
   6  *
   7  *   This file is part of Lustre, http://www.lustre.org.
   8  *
   9  *   Lustre is free software; you can redistribute it and/or
  10  *   modify it under the terms of version 2 of the GNU General Public
  11  *   License as published by the Free Software Foundation.
  12  *
  13  *   Lustre is distributed in the hope that it will be useful,
  14  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  16  *   GNU General Public License for more details.
  17  *
  18  *   You should have received a copy of the GNU General Public License
  19  *   along with Lustre; if not, write to the Free Software
  20  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  21  */
  22
  23 #define DEBUG_SUBSYSTEM S_LLITE
  24
  25 #include <stdlib.h>
  26 #include <string.h>
  27 #include <assert.h>
  28 #include <time.h>
  29 #include <sys/types.h>
  30 #include <sys/stat.h>
  31 #include <fcntl.h>
  32 #include <sys/queue.h>
  33 #ifndef __CYGWIN__
  34 # include <sys/statvfs.h>
  35 #else
  36 # include <sys/statfs.h>
  37 #endif
  38
  39 #include <sysio.h>
  40 #ifdef HAVE_XTIO_H
  41 #include <xtio.h>
  42 #endif
  43 #include <fs.h>
  44 #include <mount.h>
  45 #include <inode.h>
  46 #ifdef HAVE_FILE_H
  47 #include <file.h>
  48 #endif
  49 #include <liblustre.h>
  50
  51 #include <obd.h>
  52 #include <obd_support.h>
  53 #include <lustre_fid.h>
  54 #include <lustre_lite.h>
  55 #include <lustre_dlm.h>
  56 #include <lustre_ver.h>
  57 #include <lustre_mdc.h>
  58 #include <cl_object.h>
  59
  60 #include "llite_lib.h"
  61
  62 /*
  63  * slp_ prefix stands for "Sysio Library Posix". It corresponds to historical
  64  * "llu_" prefix.
  65  */
  66
  67 static int   slp_type_init     (struct lu_device_type *t);
  68 static void  slp_type_fini     (struct lu_device_type *t);
  69
  70 static struct cl_page * slp_page_init(const struct lu_env *env,
  71                                      struct cl_object *obj,
  72                                      struct cl_page *page, cfs_page_t *vmpage);
  73 static int   slp_attr_get     (const struct lu_env *env, struct cl_object *obj,
  74                                struct cl_attr *attr);
  75
  76 static struct lu_device  *slp_device_alloc(const struct lu_env *env,
  77                                            struct lu_device_type *t,
  78                                            struct lustre_cfg *cfg);
  79
  80 static int slp_io_init(const struct lu_env *env, struct cl_object *obj,
  81                        struct cl_io *io);
  82 static struct slp_io *cl2slp_io(const struct lu_env *env,
  83                                 const struct cl_io_slice *slice);
  84
  85
  86 static void llu_free_user_page(struct page *page);
  87
  88 static const struct lu_object_operations      slp_lu_obj_ops;
  89 static const struct lu_device_operations      slp_lu_ops;
  90 static const struct cl_device_operations      slp_cl_ops;
  91 static const struct cl_io_operations          ccc_io_ops;
  92 static const struct lu_device_type_operations slp_device_type_ops;
  93              //struct lu_device_type            slp_device_type;
  94 static const struct cl_page_operations        slp_page_ops;
  95 static const struct cl_page_operations        slp_transient_page_ops;
  96 static const struct cl_lock_operations        slp_lock_ops;
  97
  98
  99 /*****************************************************************************
 100  *
 101  * Slp device and device type functions.
 102  *
 103  */
 104
 105 void *slp_session_key_init(const struct lu_context *ctx,
 106                                   struct lu_context_key *key)
 107 {
 108         struct slp_session *session;
 109
 110         OBD_ALLOC_PTR(session);
 111         if (session == NULL)
 112                 session = ERR_PTR(-ENOMEM);
 113         return session;
 114 }
 115
 116 void slp_session_key_fini(const struct lu_context *ctx,
 117                                  struct lu_context_key *key, void *data)
 118 {
 119         struct slp_session *session = data;
 120         OBD_FREE_PTR(session);
 121 }
 122
 123 struct lu_context_key slp_session_key = {
 124         .lct_tags = LCT_SESSION,
 125         .lct_init = slp_session_key_init,
 126         .lct_fini = slp_session_key_fini
 127 };
 128
 129 /* type constructor/destructor: slp_type_{init,fini,start,stop}(). */
 130 LU_TYPE_INIT_FINI(slp, &ccc_key, &ccc_session_key, &slp_session_key);
 131
 132 static struct lu_device *slp_device_alloc(const struct lu_env *env,
 133                                           struct lu_device_type *t,
 134                                           struct lustre_cfg *cfg)
 135 {
 136         return ccc_device_alloc(env, t, cfg, &slp_lu_ops, &slp_cl_ops);
 137 }
 138
 139 static int slp_lock_init(const struct lu_env *env,
 140                          struct cl_object *obj, struct cl_lock *lock,
 141                          const struct cl_io *io)
 142 {
 143         return ccc_lock_init(env, obj, lock, io, &slp_lock_ops);
 144 }
 145
 146 static const struct cl_object_operations slp_ops = {
 147         .coo_page_init = slp_page_init,
 148         .coo_lock_init = slp_lock_init,
 149         .coo_io_init   = slp_io_init,
 150         .coo_attr_get  = slp_attr_get,
 151         .coo_attr_set  = ccc_attr_set,
 152         .coo_conf_set  = ccc_conf_set,
 153         .coo_glimpse   = ccc_object_glimpse
 154 };
 155
 156 static int slp_object_print(const struct lu_env *env, void *cookie,
 157                             lu_printer_t p, const struct lu_object *o)
 158 {
 159         struct ccc_object *obj   = lu2ccc(o);
 160         struct inode      *inode = obj->cob_inode;
 161         struct intnl_stat *st = NULL;
 162
 163         if (inode)
 164                 st = llu_i2stat(inode);
 165
 166         return (*p)(env, cookie, LUSTRE_SLP_NAME"-object@%p(%p:%lu/%u)",
 167                     obj, inode,
 168                     st ? (unsigned long)st->st_ino : 0UL,
 169                     inode ? (unsigned int)llu_i2info(inode)->lli_st_generation
 170                     : 0);
 171 }
 172
 173 static const struct lu_object_operations slp_lu_obj_ops = {
 174         .loo_object_init      = ccc_object_init,
 175         .loo_object_start     = NULL,
 176         .loo_object_delete    = NULL,
 177         .loo_object_release   = NULL,
 178         .loo_object_free      = ccc_object_free,
 179         .loo_object_print     = slp_object_print,
 180         .loo_object_invariant = NULL
 181 };
 182
 183 static struct lu_object *slp_object_alloc(const struct lu_env *env,
 184                                           const struct lu_object_header *hdr,
 185                                           struct lu_device *dev)
 186 {
 187         return ccc_object_alloc(env, hdr, dev, &slp_ops, &slp_lu_obj_ops);
 188 }
 189
 190 static const struct lu_device_operations slp_lu_ops = {
 191         .ldo_object_alloc      = slp_object_alloc
 192 };
 193
 194 static const struct cl_device_operations slp_cl_ops = {
 195         .cdo_req_init = ccc_req_init
 196 };
 197
 198 static const struct lu_device_type_operations slp_device_type_ops = {
 199         .ldto_init = slp_type_init,
 200         .ldto_fini = slp_type_fini,
 201
 202         .ldto_start = slp_type_start,
 203         .ldto_stop  = slp_type_stop,
 204
 205         .ldto_device_alloc = slp_device_alloc,
 206         .ldto_device_free  = ccc_device_free,
 207         .ldto_device_init  = ccc_device_init,
 208         .ldto_device_fini  = ccc_device_fini
 209 };
 210
 211 struct lu_device_type slp_device_type = {
 212         .ldt_tags     = LU_DEVICE_CL,
 213         .ldt_name     = LUSTRE_SLP_NAME,
 214         .ldt_ops      = &slp_device_type_ops,
 215         .ldt_ctx_tags = LCT_CL_THREAD
 216 };
 217
 218 int slp_global_init(void)
 219 {
 220         int result;
 221
 222         result = ccc_global_init(&slp_device_type);
 223         return result;
 224 }
 225
 226 void slp_global_fini(void)
 227 {
 228         ccc_global_fini(&slp_device_type);
 229 }
 230
 231 /*****************************************************************************
 232  *
 233  * Object operations.
 234  *
 235  */
 236
 237 static struct cl_page *slp_page_init(const struct lu_env *env,
 238                                      struct cl_object *obj,
 239                                      struct cl_page *page, cfs_page_t *vmpage)
 240 {
 241         struct ccc_page *cpg;
 242         int result;
 243
 244         CLOBINVRNT(env, obj, ccc_object_invariant(obj));
 245
 246         OBD_ALLOC_PTR(cpg);
 247         if (cpg != NULL) {
 248                 cpg->cpg_page = vmpage;
 249
 250                 if (page->cp_type == CPT_CACHEABLE) {
 251                         LBUG();
 252                 } else {
 253                         struct ccc_object *clobj = cl2ccc(obj);
 254
 255                         cl_page_slice_add(page, &cpg->cpg_cl, obj,
 256                                           &slp_transient_page_ops);
 257                         clobj->cob_transient_pages++;
 258                 }
 259                 result = 0;
 260         } else
 261                 result = -ENOMEM;
 262         return ERR_PTR(result);
 263 }
 264
 265 static int slp_io_init(const struct lu_env *env, struct cl_object *obj,
 266                        struct cl_io *io)
 267 {
 268         struct ccc_io      *vio   = ccc_env_io(env);
 269         int result = 0;
 270
 271         CLOBINVRNT(env, obj, ccc_object_invariant(obj));
 272
 273         cl_io_slice_add(io, &vio->cui_cl, obj, &ccc_io_ops);
 274         if (io->ci_type == CIT_READ || io->ci_type == CIT_WRITE) {
 275                 size_t count;
 276
 277                 count = io->u.ci_rw.crw_count;
 278                 /* "If nbyte is 0, read() will return 0 and have no other
 279                  *  results."  -- Single Unix Spec */
 280                 if (count == 0)
 281                         return 1;
 282                 /* "If nbyte is 0, read() will return 0 and have no other
 283                  *  results."  -- Single Unix Spec */
 284                 if (count == 0)
 285                         result = 1;
 286                 else {
 287                         vio->cui_tot_count = count;
 288                         vio->cui_tot_nrsegs = 0;
 289                 }
 290
 291         }
 292         return 0;
 293 }
 294
 295 static int slp_attr_get(const struct lu_env *env, struct cl_object *obj,
 296                         struct cl_attr *attr)
 297 {
 298         struct inode *inode = ccc_object_inode(obj);
 299         struct intnl_stat *st = llu_i2stat(inode);
 300
 301         attr->cat_size = st->st_size;
 302         attr->cat_blocks = st->st_blocks;
 303         attr->cat_mtime  = st->st_mtime;
 304         attr->cat_atime  = st->st_atime;
 305         attr->cat_ctime  = st->st_ctime;
 306         /* KMS is not known by this layer */
 307         return 0; /* layers below have to fill in the rest */
 308 }
 309
 310 /*****************************************************************************
 311  *
 312  * Page operations.
 313  *
 314  */
 315
 316 static void slp_page_fini_common(struct ccc_page *cp)
 317 {
 318         cfs_page_t *vmpage = cp->cpg_page;
 319
 320         LASSERT(vmpage != NULL);
 321         llu_free_user_page(vmpage);
 322         OBD_FREE_PTR(cp);
 323 }
 324
 325 static void slp_page_completion_common(const struct lu_env *env,
 326                                        struct ccc_page *cp, int ioret)
 327 {
 328         struct cl_sync_io *anchor = cp->cpg_sync_io;
 329
 330         if (anchor) {
 331                 cp->cpg_sync_io  = NULL;
 332                 cl_sync_io_note(anchor, ioret);
 333         } else {
 334                 LBUG();
 335         }
 336 }
 337
 338 static void slp_page_completion_read(const struct lu_env *env,
 339                                      const struct cl_page_slice *slice,
 340                                      int ioret)
 341 {
 342         struct ccc_page *cp      = cl2ccc_page(slice);
 343         ENTRY;
 344
 345         slp_page_completion_common(env, cp, ioret);
 346
 347         EXIT;
 348 }
 349
 350 static void slp_page_completion_write_common(const struct lu_env *env,
 351                                              const struct cl_page_slice *slice,
 352                                              int ioret)
 353 {
 354         struct ccc_page *cp     = cl2ccc_page(slice);
 355
 356         if (ioret == 0) {
 357                 cp->cpg_write_queued = 0;
 358                 /*
 359                  * Only ioret == 0, write succeed, then this page could be
 360                  * deleted from the pending_writing count.
 361                  */
 362         }
 363         slp_page_completion_common(env, cp, ioret);
 364 }
 365
 366 static int slp_page_is_vmlocked(const struct lu_env *env,
 367                                 const struct cl_page_slice *slice)
 368 {
 369         return -EBUSY;
 370 }
 371
 372 static void slp_transient_page_fini(const struct lu_env *env,
 373                                     struct cl_page_slice *slice)
 374 {
 375         struct ccc_page *cp = cl2ccc_page(slice);
 376         struct cl_page *clp = slice->cpl_page;
 377         struct ccc_object *clobj = cl2ccc(clp->cp_obj);
 378
 379         slp_page_fini_common(cp);
 380         clobj->cob_transient_pages--;
 381 }
 382
 383
 384 static const struct cl_page_operations slp_transient_page_ops = {
 385         .cpo_own           = ccc_transient_page_own,
 386         .cpo_assume        = ccc_transient_page_assume,
 387         .cpo_unassume      = ccc_transient_page_unassume,
 388         .cpo_disown        = ccc_transient_page_disown,
 389         .cpo_discard       = ccc_transient_page_discard,
 390         .cpo_vmpage        = ccc_page_vmpage,
 391         .cpo_is_vmlocked   = slp_page_is_vmlocked,
 392         .cpo_fini          = slp_transient_page_fini,
 393         .cpo_is_under_lock = ccc_page_is_under_lock,
 394         .io = {
 395                 [CRT_READ] = {
 396                         .cpo_completion  = slp_page_completion_read,
 397                 },
 398                 [CRT_WRITE] = {
 399                         .cpo_completion  = slp_page_completion_write_common,
 400                 }
 401         }
 402 };
 403
 404 /*****************************************************************************
 405  *
 406  * Lock operations.
 407  *
 408  */
 409
 410 static int slp_lock_enqueue(const struct lu_env *env,
 411                            const struct cl_lock_slice *slice,
 412                            struct cl_io *_, __u32 enqflags)
 413 {
 414         CLOBINVRNT(env, slice->cls_obj, ccc_object_invariant(slice->cls_obj));
 415
 416         liblustre_wait_event(0);
 417         return 0;
 418 }
 419
 420 static const struct cl_lock_operations slp_lock_ops = {
 421         .clo_fini      = ccc_lock_fini,
 422         .clo_enqueue   = slp_lock_enqueue,
 423         .clo_wait      = ccc_lock_wait,
 424         .clo_unuse     = ccc_lock_unuse,
 425         .clo_fits_into = ccc_lock_fits_into,
 426 };
 427
 428 /*****************************************************************************
 429  *
 430  * io operations.
 431  *
 432  */
 433
 434 static int slp_io_rw_lock(const struct lu_env *env, struct cl_io *io,
 435                           enum cl_lock_mode mode, loff_t start, loff_t end)
 436 {
 437         int result;
 438
 439         LASSERT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE);
 440
 441         if (!io->u.ci_wr.wr_append) { // No lock without O_APPEND in liblustre
 442                 return 0;
 443         }
 444
 445         result = ccc_io_one_lock(env, io, 0, mode, start, end);
 446
 447         return result;
 448 }
 449
 450 static int slp_io_write_lock(const struct lu_env *env,
 451                              const struct cl_io_slice *ios)
 452 {
 453         struct cl_io *io = ios->cis_io;
 454         loff_t start;
 455         loff_t end;
 456
 457         if (io->u.ci_wr.wr_append) {
 458                 start = 0;
 459                 end   = OBD_OBJECT_EOF;
 460         } else {
 461                 start = io->u.ci_wr.wr.crw_pos;
 462                 end   = start + io->u.ci_wr.wr.crw_count - 1;
 463         }
 464
 465         return slp_io_rw_lock(env, io, CLM_WRITE, start, end);
 466
 467 }
 468
 469 static int slp_io_trunc_iter_init(const struct lu_env *env,
 470                                   const struct cl_io_slice *ios)
 471 {
 472         return 0;
 473 }
 474
 475 static int slp_io_trunc_start(const struct lu_env *env,
 476                               const struct cl_io_slice *ios)
 477 {
 478         return 0;
 479 }
 480
 481 static struct page *llu_get_user_page(int index, void *addr, int offset,
 482                                       int count)
 483 {
 484         struct page *page;
 485
 486         OBD_ALLOC_PTR(page);
 487         if (!page)
 488                 return NULL;
 489         page->index = index;
 490         page->addr = addr;
 491         page->_offset = offset;
 492         page->_count = count;
 493
 494         CFS_INIT_LIST_HEAD(&page->list);
 495         CFS_INIT_LIST_HEAD(&page->_node);
 496
 497         return page;
 498 }
 499
 500 static void llu_free_user_page(struct page *page)
 501 {
 502         OBD_FREE_PTR(page);
 503 }
 504
 505 static int llu_queue_pio(const struct lu_env *env, struct cl_io *io,
 506                          struct llu_io_group *group,
 507                          char *buf, size_t count, loff_t pos)
 508 {
 509         struct cl_object *obj = io->ci_obj;
 510         struct inode *inode = ccc_object_inode(obj);
 511         struct intnl_stat *st = llu_i2stat(inode);
 512         struct obd_export *exp = llu_i2obdexp(inode);
 513         struct page *page;
 514         int  rc = 0, npages = 0, ret_bytes = 0;
 515         int local_lock;
 516         struct cl_page *clp;
 517         struct ccc_page *clup;
 518         struct cl_2queue *queue;
 519         struct cl_sync_io *anchor = &ccc_env_info(env)->cti_sync_io;
 520         ENTRY;
 521
 522         if (!exp)
 523                 RETURN(-EINVAL);
 524
 525         local_lock = group->lig_params->lrp_lock_mode != LCK_NL;
 526
 527         queue = &io->ci_queue;
 528         cl_2queue_init(queue);
 529
 530
 531         /* prepare the pages array */
 532         do {
 533                 unsigned long index, offset, bytes;
 534
 535                 offset = (pos & ~CFS_PAGE_MASK);
 536                 index = pos >> CFS_PAGE_SHIFT;
 537                 bytes = CFS_PAGE_SIZE - offset;
 538                 if (bytes > count)
 539                         bytes = count;
 540
 541                 /* prevent read beyond file range */
 542                 if (/* local_lock && */
 543                     io->ci_type == CIT_READ && pos + bytes >= st->st_size) {
 544                         if (pos >= st->st_size)
 545                                 break;
 546                         bytes = st->st_size - pos;
 547                 }
 548
 549                 /* prepare page for this index */
 550                 page = llu_get_user_page(index, buf - offset, offset, bytes);
 551                 if (!page) {
 552                         rc = -ENOMEM;
 553                         break;
 554                 }
 555
 556                 clp = cl_page_find(env, obj,
 557                                    cl_index(obj, pos),
 558                                    page, CPT_TRANSIENT);
 559
 560                 if (IS_ERR(clp)) {
 561                         rc = PTR_ERR(clp);
 562                         break;
 563                 }
 564
 565                 rc = cl_page_own(env, io, clp);
 566                 if (rc) {
 567                         LASSERT(clp->cp_state == CPS_FREEING);
 568                         cl_page_put(env, clp);
 569                         break;
 570                 }
 571
 572                 clup = cl2ccc_page(cl_page_at(clp, &slp_device_type));
 573                 clup->cpg_sync_io = anchor;
 574                 cl_2queue_add(queue, clp);
 575
 576                 /* drop the reference count for cl_page_find, so that the page
 577                  * will be freed in cl_2queue_fini. */
 578                 cl_page_put(env, clp);
 579
 580                 cl_page_clip(env, clp, offset, offset+bytes);
 581
 582                 npages++;
 583                 count -= bytes;
 584                 pos += bytes;
 585                 buf += bytes;
 586
 587                 group->lig_rwcount += bytes;
 588                 ret_bytes += bytes;
 589                 page++;
 590         } while (count);
 591
 592         cl_sync_io_init(anchor, npages);
 593         /* printk("Inited anchor with %d pages\n", npages); */
 594
 595         if (rc == 0) {
 596                 rc = cl_io_submit_rw(env, io,
 597                                      io->ci_type == CIT_READ ? CRT_READ :
 598                                                                CRT_WRITE,
 599                                      queue);
 600                 if (rc == 0) {
 601                         /* If some pages weren't sent for any reason, count
 602                          * then as completed, to avoid infinite wait. */
 603                         cl_page_list_for_each(clp, &queue->c2_qin) {
 604                                 CL_PAGE_DEBUG(D_ERROR, env, clp,
 605                                               "not completed\n");
 606                                 cl_sync_io_note(anchor, +1);
 607                         }
 608                         /* wait for the IO to be finished. */
 609                         rc = cl_sync_io_wait(env, io, &queue->c2_qout, anchor);
 610                 }
 611         }
 612
 613         group->lig_rc = rc;
 614
 615         cl_2queue_discard(env, io, queue);
 616         cl_2queue_disown(env, io, queue);
 617         cl_2queue_fini(env, queue);
 618
 619         RETURN(ret_bytes);
 620 }
 621
 622 static
 623 struct llu_io_group * get_io_group(struct inode *inode, int maxpages,
 624                                    struct lustre_rw_params *params)
 625 {
 626         struct llu_io_group *group;
 627
 628         OBD_ALLOC_PTR(group);
 629         if (!group)
 630                 return ERR_PTR(-ENOMEM);
 631
 632         group->lig_params = params;
 633
 634         return group;
 635 }
 636
 637 static int max_io_pages(ssize_t len, int iovlen)
 638 {
 639         return (((len + CFS_PAGE_SIZE -1) / CFS_PAGE_SIZE) + 2 + iovlen - 1);
 640 }
 641
 642 void put_io_group(struct llu_io_group *group)
 643 {
 644         OBD_FREE_PTR(group);
 645 }
 646
 647 static int slp_io_start(const struct lu_env *env, const struct cl_io_slice *ios)
 648 {
 649         struct ccc_io     *cio   = cl2ccc_io(env, ios);
 650         struct cl_io      *io    = ios->cis_io;
 651         struct cl_object  *obj   = io->ci_obj;
 652         struct inode      *inode = ccc_object_inode(obj);
 653         int     err, ret;
 654         loff_t  pos;
 655         size_t  cnt;
 656         struct llu_io_group *iogroup;
 657         struct lustre_rw_params p = {0};
 658         int iovidx;
 659         struct intnl_stat *st = llu_i2stat(inode);
 660         struct llu_inode_info *lli = llu_i2info(inode);
 661         struct llu_io_session *session = cl2slp_io(env, ios)->sio_session;
 662         int write = io->ci_type == CIT_WRITE;
 663
 664         CLOBINVRNT(env, obj, ccc_object_invariant(obj));
 665
 666         if (write) {
 667                 pos = io->u.ci_wr.wr.crw_pos;
 668                 cnt = io->u.ci_wr.wr.crw_count;
 669         } else {
 670                 pos = io->u.ci_rd.rd.crw_pos;
 671                 cnt = io->u.ci_rd.rd.crw_count;
 672         }
 673         if (io->u.ci_wr.wr_append) {
 674                 p.lrp_lock_mode = LCK_PW;
 675         } else {
 676                 p.lrp_brw_flags = OBD_BRW_SRVLOCK;
 677                 p.lrp_lock_mode = LCK_NL;
 678         }
 679
 680         iogroup = get_io_group(inode, max_io_pages(cnt, cio->cui_nrsegs), &p);
 681         if (IS_ERR(iogroup))
 682                 RETURN(PTR_ERR(iogroup));
 683
 684         err = ccc_prep_size(env, obj, io, pos + cnt - 1, 0);
 685         if (err != 0)
 686                 GOTO(out, err);
 687
 688         CDEBUG(D_INODE,
 689                "%s ino %lu, "LPSZ" bytes, offset %lld, i_size %llu\n",
 690                write?"Write":"Read", (unsigned long)st->st_ino,
 691                cnt, (__u64)pos, (__u64)st->st_size);
 692
 693         if (write && io->u.ci_wr.wr_append)
 694                 pos = io->u.ci_wr.wr.crw_pos = st->st_size; /* XXX? Do we need to change io content too here? */
 695                 /* XXX What about if one write syscall writes at 2 different offsets? */
 696
 697         for (iovidx = 0; iovidx < cio->cui_nrsegs; iovidx++) {
 698                 char *buf = (char *) cio->cui_iov[iovidx].iov_base;
 699                 size_t count = cio->cui_iov[iovidx].iov_len;
 700
 701                 if (!count)
 702                         continue;
 703                 if (cnt < count)
 704                         count = cnt;
 705                 if (IS_BAD_PTR(buf) || IS_BAD_PTR(buf + count)) {
 706                         GOTO(out, err = -EFAULT);
 707                 }
 708
 709                 if (io->ci_type == CIT_READ) {
 710                         if (/* local_lock && */ pos >= st->st_size)
 711                                 break;
 712                 } else if (io->ci_type == CIT_WRITE) {
 713                         if (pos >= lli->lli_maxbytes) {
 714                                 GOTO(out, err = -EFBIG);
 715                         }
 716                         if (pos + count >= lli->lli_maxbytes)
 717                                 count = lli->lli_maxbytes - pos;
 718                 } else {
 719                         LBUG();
 720                 }
 721
 722                 ret = llu_queue_pio(env, io, iogroup, buf, count, pos);
 723                 if (ret < 0) {
 724                         GOTO(out, err = ret);
 725                 } else {
 726                         io->ci_nob += ret;
 727                         pos += ret;
 728                         cnt -= ret;
 729                         if (io->ci_type == CIT_WRITE) {
 730 //                                obd_adjust_kms(exp, lsm, pos, 0); // XXX
 731                                 if (pos > st->st_size)
 732                                         st->st_size = pos;
 733                         }
 734                         if (!cnt)
 735                                 break;
 736                 }
 737         }
 738         LASSERT(cnt == 0 || io->ci_type == CIT_READ); /* libsysio should guarantee this */
 739
 740         session->lis_groups[session->lis_ngroups++] = iogroup;
 741
 742         return 0;
 743 out:
 744         put_io_group(iogroup);
 745         return err;
 746 }
 747
 748 static const struct cl_io_operations ccc_io_ops = {
 749         .op = {
 750                 [CIT_READ] = {
 751                         .cio_fini      = ccc_io_fini,
 752                         .cio_start     = slp_io_start,
 753                         .cio_end       = ccc_io_end
 754                 },
 755                 [CIT_WRITE] = {
 756                         .cio_fini      = ccc_io_fini,
 757                         .cio_lock      = slp_io_write_lock,
 758                         .cio_start     = slp_io_start,
 759                         .cio_end       = ccc_io_end
 760                 },
 761                 [CIT_TRUNC] = {
 762                         .cio_fini       = ccc_io_fini,
 763                         .cio_iter_init  = slp_io_trunc_iter_init,
 764                         .cio_start      = slp_io_trunc_start
 765                 },
 766                 [CIT_MISC] = {
 767                         .cio_fini   = ccc_io_fini
 768                 }
 769         }
 770 };
 771
 772 static struct slp_io *cl2slp_io(const struct lu_env *env,
 773                                 const struct cl_io_slice *slice)
 774 {
 775         /* We call it just for assertion here */
 776         cl2ccc_io(env, slice);
 777
 778         return slp_env_io(env);
 779 }
 780
 781 /*****************************************************************************
 782  *
 783  * Temporary prototype thing: mirror obd-devices into cl devices.
 784  *
 785  */
 786
 787 int cl_sb_init(struct llu_sb_info *sbi)
 788 {
 789         struct cl_device  *cl;
 790         struct lu_env     *env;
 791         int rc = 0;
 792         int refcheck;
 793
 794         env = cl_env_get(&refcheck);
 795         if (IS_ERR(env))
 796                 RETURN(PTR_ERR(env));
 797
 798         cl = cl_type_setup(env, NULL, &slp_device_type,
 799                            sbi->ll_dt_exp->exp_obd->obd_lu_dev);
 800         if (IS_ERR(cl))
 801                 GOTO(out, rc = PTR_ERR(cl));
 802
 803         sbi->ll_cl = cl;
 804         sbi->ll_site = cl2lu_dev(cl)->ld_site;
 805 out:
 806         cl_env_put(env, &refcheck);
 807         RETURN(rc);
 808 }
 809
 810 int cl_sb_fini(struct llu_sb_info *sbi)
 811 {
 812         struct lu_env *env;
 813         int refcheck;
 814
 815         ENTRY;
 816
 817         env = cl_env_get(&refcheck);
 818         if (IS_ERR(env))
 819                 RETURN(PTR_ERR(env));
 820
 821         if (sbi->ll_cl != NULL) {
 822                 cl_stack_fini(env, sbi->ll_cl);
 823                 sbi->ll_cl = NULL;
 824                 sbi->ll_site = NULL;
 825         }
 826         cl_env_put(env, &refcheck);
 827         /*
 828          * If mount failed (sbi->ll_cl == NULL), and this there are no other
 829          * mounts, stop device types manually (this usually happens
 830          * automatically when last device is destroyed).
 831          */
 832         lu_types_stop();
 833         cl_env_cache_purge(~0);
 834         RETURN(0);
 835 }