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                         result = 1;
 282                 else {
 283                         vio->cui_tot_count = count;
 284                         vio->cui_tot_nrsegs = 0;
 285                 }
 286
 287         }
 288         return result;
 289 }
 290
 291 static int slp_attr_get(const struct lu_env *env, struct cl_object *obj,
 292                         struct cl_attr *attr)
 293 {
 294         struct inode *inode = ccc_object_inode(obj);
 295         struct intnl_stat *st = llu_i2stat(inode);
 296
 297         attr->cat_size = st->st_size;
 298         attr->cat_blocks = st->st_blocks;
 299         attr->cat_mtime  = st->st_mtime;
 300         attr->cat_atime  = st->st_atime;
 301         attr->cat_ctime  = st->st_ctime;
 302         /* KMS is not known by this layer */
 303         return 0; /* layers below have to fill in the rest */
 304 }
 305
 306 /*****************************************************************************
 307  *
 308  * Page operations.
 309  *
 310  */
 311
 312 static void slp_page_fini_common(struct ccc_page *cp)
 313 {
 314         cfs_page_t *vmpage = cp->cpg_page;
 315
 316         LASSERT(vmpage != NULL);
 317         llu_free_user_page(vmpage);
 318         OBD_FREE_PTR(cp);
 319 }
 320
 321 static void slp_page_completion_common(const struct lu_env *env,
 322                                        struct ccc_page *cp, int ioret)
 323 {
 324         struct cl_sync_io *anchor = cp->cpg_sync_io;
 325
 326         if (anchor) {
 327                 cp->cpg_sync_io  = NULL;
 328                 cl_sync_io_note(anchor, ioret);
 329         } else {
 330                 LBUG();
 331         }
 332 }
 333
 334 static void slp_page_completion_read(const struct lu_env *env,
 335                                      const struct cl_page_slice *slice,
 336                                      int ioret)
 337 {
 338         struct ccc_page *cp      = cl2ccc_page(slice);
 339         ENTRY;
 340
 341         slp_page_completion_common(env, cp, ioret);
 342
 343         EXIT;
 344 }
 345
 346 static void slp_page_completion_write_common(const struct lu_env *env,
 347                                              const struct cl_page_slice *slice,
 348                                              int ioret)
 349 {
 350         struct ccc_page *cp     = cl2ccc_page(slice);
 351
 352         if (ioret == 0) {
 353                 cp->cpg_write_queued = 0;
 354                 /*
 355                  * Only ioret == 0, write succeed, then this page could be
 356                  * deleted from the pending_writing count.
 357                  */
 358         }
 359         slp_page_completion_common(env, cp, ioret);
 360 }
 361
 362 static int slp_page_is_vmlocked(const struct lu_env *env,
 363                                 const struct cl_page_slice *slice)
 364 {
 365         return -EBUSY;
 366 }
 367
 368 static void slp_transient_page_fini(const struct lu_env *env,
 369                                     struct cl_page_slice *slice)
 370 {
 371         struct ccc_page *cp = cl2ccc_page(slice);
 372         struct cl_page *clp = slice->cpl_page;
 373         struct ccc_object *clobj = cl2ccc(clp->cp_obj);
 374
 375         slp_page_fini_common(cp);
 376         clobj->cob_transient_pages--;
 377 }
 378
 379
 380 static const struct cl_page_operations slp_transient_page_ops = {
 381         .cpo_own           = ccc_transient_page_own,
 382         .cpo_assume        = ccc_transient_page_assume,
 383         .cpo_unassume      = ccc_transient_page_unassume,
 384         .cpo_disown        = ccc_transient_page_disown,
 385         .cpo_discard       = ccc_transient_page_discard,
 386         .cpo_vmpage        = ccc_page_vmpage,
 387         .cpo_is_vmlocked   = slp_page_is_vmlocked,
 388         .cpo_fini          = slp_transient_page_fini,
 389         .cpo_is_under_lock = ccc_page_is_under_lock,
 390         .io = {
 391                 [CRT_READ] = {
 392                         .cpo_completion  = slp_page_completion_read,
 393                 },
 394                 [CRT_WRITE] = {
 395                         .cpo_completion  = slp_page_completion_write_common,
 396                 }
 397         }
 398 };
 399
 400 /*****************************************************************************
 401  *
 402  * Lock operations.
 403  *
 404  */
 405
 406 static int slp_lock_enqueue(const struct lu_env *env,
 407                            const struct cl_lock_slice *slice,
 408                            struct cl_io *_, __u32 enqflags)
 409 {
 410         CLOBINVRNT(env, slice->cls_obj, ccc_object_invariant(slice->cls_obj));
 411
 412         liblustre_wait_event(0);
 413         return 0;
 414 }
 415
 416 static const struct cl_lock_operations slp_lock_ops = {
 417         .clo_delete    = ccc_lock_delete,
 418         .clo_fini      = ccc_lock_fini,
 419         .clo_enqueue   = slp_lock_enqueue,
 420         .clo_wait      = ccc_lock_wait,
 421         .clo_unuse     = ccc_lock_unuse,
 422         .clo_fits_into = ccc_lock_fits_into,
 423 };
 424
 425 /*****************************************************************************
 426  *
 427  * io operations.
 428  *
 429  */
 430
 431 static int slp_io_rw_lock(const struct lu_env *env,
 432                              const struct cl_io_slice *ios)
 433 {
 434         struct cl_io *io = ios->cis_io;
 435         loff_t start;
 436         loff_t end;
 437
 438         if (cl_io_is_append(io)) {
 439                 start = 0;
 440                 end   = OBD_OBJECT_EOF;
 441         } else {
 442                 start = io->u.ci_wr.wr.crw_pos;
 443                 end   = start + io->u.ci_wr.wr.crw_count - 1;
 444         }
 445         /*
 446          * This acquires real DLM lock only in O_APPEND case, because of
 447          * the io->ci_lockreq setting in llu_io_init().
 448          */
 449         LASSERT(ergo(cl_io_is_append(io), io->ci_lockreq == CILR_MANDATORY));
 450         LASSERT(ergo(!cl_io_is_append(io), io->ci_lockreq == CILR_NEVER));
 451         return ccc_io_one_lock(env, io, 0,
 452                                io->ci_type == CIT_READ ? CLM_READ : CLM_WRITE,
 453                                start, end);
 454
 455 }
 456
 457 static int slp_io_trunc_iter_init(const struct lu_env *env,
 458                                   const struct cl_io_slice *ios)
 459 {
 460         return 0;
 461 }
 462
 463 static int slp_io_trunc_start(const struct lu_env *env,
 464                               const struct cl_io_slice *ios)
 465 {
 466         return 0;
 467 }
 468
 469 static struct page *llu_get_user_page(int index, void *addr, int offset,
 470                                       int count)
 471 {
 472         struct page *page;
 473
 474         OBD_ALLOC_PTR(page);
 475         if (!page)
 476                 return NULL;
 477         page->index = index;
 478         page->addr = addr;
 479         page->_offset = offset;
 480         page->_count = count;
 481
 482         CFS_INIT_LIST_HEAD(&page->list);
 483         CFS_INIT_LIST_HEAD(&page->_node);
 484
 485         return page;
 486 }
 487
 488 static void llu_free_user_page(struct page *page)
 489 {
 490         OBD_FREE_PTR(page);
 491 }
 492
 493 static int llu_queue_pio(const struct lu_env *env, struct cl_io *io,
 494                          struct llu_io_group *group,
 495                          char *buf, size_t count, loff_t pos)
 496 {
 497         struct cl_object *obj = io->ci_obj;
 498         struct inode *inode = ccc_object_inode(obj);
 499         struct intnl_stat *st = llu_i2stat(inode);
 500         struct obd_export *exp = llu_i2obdexp(inode);
 501         struct page *page;
 502         int  rc = 0, npages = 0, ret_bytes = 0;
 503         int local_lock;
 504         struct cl_page *clp;
 505         struct ccc_page *clup;
 506         struct cl_2queue *queue;
 507         struct cl_sync_io *anchor = &ccc_env_info(env)->cti_sync_io;
 508         ENTRY;
 509
 510         if (!exp)
 511                 RETURN(-EINVAL);
 512
 513         local_lock = group->lig_params->lrp_lock_mode != LCK_NL;
 514
 515         queue = &io->ci_queue;
 516         cl_2queue_init(queue);
 517
 518
 519         /* prepare the pages array */
 520         do {
 521                 unsigned long index, offset, bytes;
 522
 523                 offset = (pos & ~CFS_PAGE_MASK);
 524                 index = pos >> CFS_PAGE_SHIFT;
 525                 bytes = CFS_PAGE_SIZE - offset;
 526                 if (bytes > count)
 527                         bytes = count;
 528
 529                 /* prevent read beyond file range */
 530                 if (/* local_lock && */
 531                     io->ci_type == CIT_READ && pos + bytes >= st->st_size) {
 532                         if (pos >= st->st_size)
 533                                 break;
 534                         bytes = st->st_size - pos;
 535                 }
 536
 537                 /* prepare page for this index */
 538                 page = llu_get_user_page(index, buf - offset, offset, bytes);
 539                 if (!page) {
 540                         rc = -ENOMEM;
 541                         break;
 542                 }
 543
 544                 clp = cl_page_find(env, obj,
 545                                    cl_index(obj, pos),
 546                                    page, CPT_TRANSIENT);
 547
 548                 if (IS_ERR(clp)) {
 549                         rc = PTR_ERR(clp);
 550                         break;
 551                 }
 552
 553                 rc = cl_page_own(env, io, clp);
 554                 if (rc) {
 555                         LASSERT(clp->cp_state == CPS_FREEING);
 556                         cl_page_put(env, clp);
 557                         break;
 558                 }
 559
 560                 clup = cl2ccc_page(cl_page_at(clp, &slp_device_type));
 561                 clup->cpg_sync_io = anchor;
 562                 cl_2queue_add(queue, clp);
 563
 564                 /* drop the reference count for cl_page_find, so that the page
 565                  * will be freed in cl_2queue_fini. */
 566                 cl_page_put(env, clp);
 567
 568                 cl_page_clip(env, clp, offset, offset+bytes);
 569
 570                 npages++;
 571                 count -= bytes;
 572                 pos += bytes;
 573                 buf += bytes;
 574
 575                 group->lig_rwcount += bytes;
 576                 ret_bytes += bytes;
 577                 page++;
 578         } while (count);
 579
 580         cl_sync_io_init(anchor, npages);
 581         /* printk("Inited anchor with %d pages\n", npages); */
 582
 583         if (rc == 0) {
 584                 enum cl_req_type crt;
 585
 586                 crt = io->ci_type == CIT_READ ? CRT_READ : CRT_WRITE;
 587                 rc = cl_io_submit_rw(env, io, crt, queue, CRP_NORMAL);
 588                 if (rc == 0) {
 589                         /* If some pages weren't sent for any reason, count
 590                          * then as completed, to avoid infinite wait. */
 591                         cl_page_list_for_each(clp, &queue->c2_qin) {
 592                                 CL_PAGE_DEBUG(D_ERROR, env, clp,
 593                                               "not completed\n");
 594                                 cl_sync_io_note(anchor, +1);
 595                         }
 596                         /* wait for the IO to be finished. */
 597                         rc = cl_sync_io_wait(env, io, &queue->c2_qout, anchor);
 598                 }
 599         }
 600
 601         group->lig_rc = rc;
 602
 603         cl_2queue_discard(env, io, queue);
 604         cl_2queue_disown(env, io, queue);
 605         cl_2queue_fini(env, queue);
 606
 607         RETURN(ret_bytes);
 608 }
 609
 610 static
 611 struct llu_io_group * get_io_group(struct inode *inode, int maxpages,
 612                                    struct lustre_rw_params *params)
 613 {
 614         struct llu_io_group *group;
 615
 616         OBD_ALLOC_PTR(group);
 617         if (!group)
 618                 return ERR_PTR(-ENOMEM);
 619
 620         group->lig_params = params;
 621
 622         return group;
 623 }
 624
 625 static int max_io_pages(ssize_t len, int iovlen)
 626 {
 627         return (((len + CFS_PAGE_SIZE -1) / CFS_PAGE_SIZE) + 2 + iovlen - 1);
 628 }
 629
 630 void put_io_group(struct llu_io_group *group)
 631 {
 632         OBD_FREE_PTR(group);
 633 }
 634
 635 static int slp_io_start(const struct lu_env *env, const struct cl_io_slice *ios)
 636 {
 637         struct ccc_io     *cio   = cl2ccc_io(env, ios);
 638         struct cl_io      *io    = ios->cis_io;
 639         struct cl_object  *obj   = io->ci_obj;
 640         struct inode      *inode = ccc_object_inode(obj);
 641         int     err, ret;
 642         loff_t  pos;
 643         size_t  cnt;
 644         struct llu_io_group *iogroup;
 645         struct lustre_rw_params p = {0};
 646         int iovidx;
 647         struct intnl_stat *st = llu_i2stat(inode);
 648         struct llu_inode_info *lli = llu_i2info(inode);
 649         struct llu_io_session *session = cl2slp_io(env, ios)->sio_session;
 650         int write = io->ci_type == CIT_WRITE;
 651         int exceed = 0;
 652
 653         CLOBINVRNT(env, obj, ccc_object_invariant(obj));
 654
 655         if (write) {
 656                 pos = io->u.ci_wr.wr.crw_pos;
 657                 cnt = io->u.ci_wr.wr.crw_count;
 658         } else {
 659                 pos = io->u.ci_rd.rd.crw_pos;
 660                 cnt = io->u.ci_rd.rd.crw_count;
 661         }
 662         if (io->u.ci_wr.wr_append) {
 663                 p.lrp_lock_mode = LCK_PW;
 664         } else {
 665                 p.lrp_brw_flags = OBD_BRW_SRVLOCK;
 666                 p.lrp_lock_mode = LCK_NL;
 667         }
 668
 669         iogroup = get_io_group(inode, max_io_pages(cnt, cio->cui_nrsegs), &p);
 670         if (IS_ERR(iogroup))
 671                 RETURN(PTR_ERR(iogroup));
 672
 673         err = ccc_prep_size(env, obj, io, pos, cnt, 0, &exceed);
 674         if (err != 0 || (write == 0 && exceed != 0))
 675                 GOTO(out, err);
 676
 677         CDEBUG(D_INODE,
 678                "%s ino %lu, "LPSZ" bytes, offset %lld, i_size %llu\n",
 679                write?"Write":"Read", (unsigned long)st->st_ino,
 680                cnt, (__u64)pos, (__u64)st->st_size);
 681
 682         if (write && io->u.ci_wr.wr_append)
 683                 pos = io->u.ci_wr.wr.crw_pos = st->st_size; /* XXX? Do we need to change io content too here? */
 684                 /* XXX What about if one write syscall writes at 2 different offsets? */
 685
 686         for (iovidx = 0; iovidx < cio->cui_nrsegs; iovidx++) {
 687                 char *buf = (char *) cio->cui_iov[iovidx].iov_base;
 688                 size_t count = cio->cui_iov[iovidx].iov_len;
 689
 690                 if (!count)
 691                         continue;
 692                 if (cnt < count)
 693                         count = cnt;
 694                 if (IS_BAD_PTR(buf) || IS_BAD_PTR(buf + count)) {
 695                         GOTO(out, err = -EFAULT);
 696                 }
 697
 698                 if (io->ci_type == CIT_READ) {
 699                         if (/* local_lock && */ pos >= st->st_size)
 700                                 break;
 701                 } else if (io->ci_type == CIT_WRITE) {
 702                         if (pos >= lli->lli_maxbytes) {
 703                                 GOTO(out, err = -EFBIG);
 704                         }
 705                         if (pos + count >= lli->lli_maxbytes)
 706                                 count = lli->lli_maxbytes - pos;
 707                 } else {
 708                         LBUG();
 709                 }
 710
 711                 ret = llu_queue_pio(env, io, iogroup, buf, count, pos);
 712                 if (ret < 0) {
 713                         GOTO(out, err = ret);
 714                 } else {
 715                         io->ci_nob += ret;
 716                         pos += ret;
 717                         cnt -= ret;
 718                         if (io->ci_type == CIT_WRITE) {
 719 //                                obd_adjust_kms(exp, lsm, pos, 0); // XXX
 720                                 if (pos > st->st_size)
 721                                         st->st_size = pos;
 722                         }
 723                         if (!cnt)
 724                                 break;
 725                 }
 726         }
 727         LASSERT(cnt == 0 || io->ci_type == CIT_READ); /* libsysio should guarantee this */
 728
 729         session->lis_groups[session->lis_ngroups++] = iogroup;
 730
 731         return 0;
 732 out:
 733         put_io_group(iogroup);
 734         return err;
 735 }
 736
 737 static const struct cl_io_operations ccc_io_ops = {
 738         .op = {
 739                 [CIT_READ] = {
 740                         .cio_fini      = ccc_io_fini,
 741                         .cio_lock      = slp_io_rw_lock,
 742                         .cio_start     = slp_io_start,
 743                         .cio_end       = ccc_io_end
 744                 },
 745                 [CIT_WRITE] = {
 746                         .cio_fini      = ccc_io_fini,
 747                         .cio_lock      = slp_io_rw_lock,
 748                         .cio_start     = slp_io_start,
 749                         .cio_end       = ccc_io_end
 750                 },
 751                 [CIT_TRUNC] = {
 752                         .cio_fini       = ccc_io_fini,
 753                         .cio_iter_init  = slp_io_trunc_iter_init,
 754                         .cio_start      = slp_io_trunc_start
 755                 },
 756                 [CIT_MISC] = {
 757                         .cio_fini   = ccc_io_fini
 758                 }
 759         }
 760 };
 761
 762 static struct slp_io *cl2slp_io(const struct lu_env *env,
 763                                 const struct cl_io_slice *slice)
 764 {
 765         /* We call it just for assertion here */
 766         cl2ccc_io(env, slice);
 767
 768         return slp_env_io(env);
 769 }
 770
 771 /*****************************************************************************
 772  *
 773  * Temporary prototype thing: mirror obd-devices into cl devices.
 774  *
 775  */
 776
 777 int cl_sb_init(struct llu_sb_info *sbi)
 778 {
 779         struct cl_device  *cl;
 780         struct lu_env     *env;
 781         int rc = 0;
 782         int refcheck;
 783
 784         env = cl_env_get(&refcheck);
 785         if (IS_ERR(env))
 786                 RETURN(PTR_ERR(env));
 787
 788         cl = cl_type_setup(env, NULL, &slp_device_type,
 789                            sbi->ll_dt_exp->exp_obd->obd_lu_dev);
 790         if (IS_ERR(cl))
 791                 GOTO(out, rc = PTR_ERR(cl));
 792
 793         sbi->ll_cl = cl;
 794         sbi->ll_site = cl2lu_dev(cl)->ld_site;
 795 out:
 796         cl_env_put(env, &refcheck);
 797         RETURN(rc);
 798 }
 799
 800 int cl_sb_fini(struct llu_sb_info *sbi)
 801 {
 802         struct lu_env *env;
 803         int refcheck;
 804
 805         ENTRY;
 806
 807         env = cl_env_get(&refcheck);
 808         if (IS_ERR(env))
 809                 RETURN(PTR_ERR(env));
 810
 811         if (sbi->ll_cl != NULL) {
 812                 cl_stack_fini(env, sbi->ll_cl);
 813                 sbi->ll_cl = NULL;
 814                 sbi->ll_site = NULL;
 815         }
 816         cl_env_put(env, &refcheck);
 817         /*
 818          * If mount failed (sbi->ll_cl == NULL), and this there are no other
 819          * mounts, stop device types manually (this usually happens
 820          * automatically when last device is destroyed).
 821          */
 822         lu_types_stop();
 823         cl_env_cache_purge(~0);
 824         RETURN(0);
 825 }