fs/exofs/ore.c

   1 /*
   2  * Copyright (C) 2005, 2006
   3  * Avishay Traeger (avishay@gmail.com)
   4  * Copyright (C) 2008, 2009
   5  * Boaz Harrosh <bharrosh@panasas.com>
   6  *
   7  * This file is part of exofs.
   8  *
   9  * exofs is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License as published by
  11  * the Free Software Foundation.  Since it is based on ext2, and the only
  12  * valid version of GPL for the Linux kernel is version 2, the only valid
  13  * version of GPL for exofs is version 2.
  14  *
  15  * exofs is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18  * GNU General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public License
  21  * along with exofs; if not, write to the Free Software
  22  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  23  */
  24
  25 #include <linux/slab.h>
  26 #include <asm/div64.h>
  27
  28 #include <scsi/osd_ore.h>
  29
  30 #define ORE_ERR(fmt, a...) printk(KERN_ERR "ore: " fmt, ##a)
  31
  32 #ifdef CONFIG_EXOFS_DEBUG
  33 #define ORE_DBGMSG(fmt, a...) \
  34         printk(KERN_NOTICE "ore @%s:%d: " fmt, __func__, __LINE__, ##a)
  35 #else
  36 #define ORE_DBGMSG(fmt, a...) \
  37         do { if (0) printk(fmt, ##a); } while (0)
  38 #endif
  39
  40 /* u64 has problems with printk this will cast it to unsigned long long */
  41 #define _LLU(x) (unsigned long long)(x)
  42
  43 #define ORE_DBGMSG2(M...) do {} while (0)
  44 /* #define ORE_DBGMSG2 ORE_DBGMSG */
  45
  46 MODULE_AUTHOR("Boaz Harrosh <bharrosh@panasas.com>");
  47 MODULE_DESCRIPTION("Objects Raid Engine ore.ko");
  48 MODULE_LICENSE("GPL");
  49
  50 static void ore_calc_stripe_info(struct ore_layout *layout, u64 file_offset,
  51                                  struct ore_striping_info *si);
  52
  53 static u8 *_ios_cred(struct ore_io_state *ios, unsigned index)
  54 {
  55         return ios->oc->comps[index & ios->oc->single_comp].cred;
  56 }
  57
  58 static struct osd_obj_id *_ios_obj(struct ore_io_state *ios, unsigned index)
  59 {
  60         return &ios->oc->comps[index & ios->oc->single_comp].obj;
  61 }
  62
  63 static struct osd_dev *_ios_od(struct ore_io_state *ios, unsigned index)
  64 {
  65         return ore_comp_dev(ios->oc, index);
  66 }
  67
  68 static int  _get_io_state(struct ore_layout *layout,
  69                           struct ore_components *oc, unsigned numdevs,
  70                           struct ore_io_state **pios)
  71 {
  72         struct ore_io_state *ios;
  73
  74         /*TODO: Maybe use kmem_cach per sbi of size
  75          * exofs_io_state_size(layout->s_numdevs)
  76          */
  77         ios = kzalloc(ore_io_state_size(numdevs), GFP_KERNEL);
  78         if (unlikely(!ios)) {
  79                 ORE_DBGMSG("Failed kzalloc bytes=%d\n",
  80                            ore_io_state_size(numdevs));
  81                 *pios = NULL;
  82                 return -ENOMEM;
  83         }
  84
  85         ios->layout = layout;
  86         ios->oc = oc;
  87         *pios = ios;
  88         return 0;
  89 }
  90
  91 /* Allocate an io_state for only a single group of devices
  92  *
  93  * If a user needs to call ore_read/write() this version must be used becase it
  94  * allocates extra stuff for striping and raid.
  95  * The ore might decide to only IO less then @length bytes do to alignmets
  96  * and constrains as follows:
  97  * - The IO cannot cross group boundary.
  98  * - In raid5/6 The end of the IO must align at end of a stripe eg.
  99  *   (@offset + @length) % strip_size == 0. Or the complete range is within a
 100  *   single stripe.
 101  * - Memory condition only permitted a shorter IO. (A user can use @length=~0
 102  *   And check the returned ios->length for max_io_size.)
 103  *
 104  * The caller must check returned ios->length (and/or ios->nr_pages) and
 105  * re-issue these pages that fall outside of ios->length
 106  */
 107 int  ore_get_rw_state(struct ore_layout *layout, struct ore_components *oc,
 108                       bool is_reading, u64 offset, u64 length,
 109                       struct ore_io_state **pios)
 110 {
 111         struct ore_io_state *ios;
 112         unsigned numdevs = layout->group_width * layout->mirrors_p1;
 113         int ret;
 114
 115         ret = _get_io_state(layout, oc, numdevs, pios);
 116         if (unlikely(ret))
 117                 return ret;
 118
 119         ios = *pios;
 120         ios->reading = is_reading;
 121         ios->offset = offset;
 122
 123         if (length) {
 124                 ore_calc_stripe_info(layout, offset, &ios->si);
 125                 ios->length = (length <= ios->si.group_length) ? length :
 126                                                         ios->si.group_length;
 127                 ios->nr_pages = (ios->length + PAGE_SIZE - 1) / PAGE_SIZE;
 128         }
 129
 130         return 0;
 131 }
 132 EXPORT_SYMBOL(ore_get_rw_state);
 133
 134 /* Allocate an io_state for all the devices in the comps array
 135  *
 136  * This version of io_state allocation is used mostly by create/remove
 137  * and trunc where we currently need all the devices. The only wastful
 138  * bit is the read/write_attributes with no IO. Those sites should
 139  * be converted to use ore_get_rw_state() with length=0
 140  */
 141 int  ore_get_io_state(struct ore_layout *layout, struct ore_components *oc,
 142                       struct ore_io_state **pios)
 143 {
 144         return _get_io_state(layout, oc, oc->numdevs, pios);
 145 }
 146 EXPORT_SYMBOL(ore_get_io_state);
 147
 148 void ore_put_io_state(struct ore_io_state *ios)
 149 {
 150         if (ios) {
 151                 unsigned i;
 152
 153                 for (i = 0; i < ios->numdevs; i++) {
 154                         struct ore_per_dev_state *per_dev = &ios->per_dev[i];
 155
 156                         if (per_dev->or)
 157                                 osd_end_request(per_dev->or);
 158                         if (per_dev->bio)
 159                                 bio_put(per_dev->bio);
 160                 }
 161
 162                 kfree(ios);
 163         }
 164 }
 165 EXPORT_SYMBOL(ore_put_io_state);
 166
 167 static void _sync_done(struct ore_io_state *ios, void *p)
 168 {
 169         struct completion *waiting = p;
 170
 171         complete(waiting);
 172 }
 173
 174 static void _last_io(struct kref *kref)
 175 {
 176         struct ore_io_state *ios = container_of(
 177                                         kref, struct ore_io_state, kref);
 178
 179         ios->done(ios, ios->private);
 180 }
 181
 182 static void _done_io(struct osd_request *or, void *p)
 183 {
 184         struct ore_io_state *ios = p;
 185
 186         kref_put(&ios->kref, _last_io);
 187 }
 188
 189 static int ore_io_execute(struct ore_io_state *ios)
 190 {
 191         DECLARE_COMPLETION_ONSTACK(wait);
 192         bool sync = (ios->done == NULL);
 193         int i, ret;
 194
 195         if (sync) {
 196                 ios->done = _sync_done;
 197                 ios->private = &wait;
 198         }
 199
 200         for (i = 0; i < ios->numdevs; i++) {
 201                 struct osd_request *or = ios->per_dev[i].or;
 202                 if (unlikely(!or))
 203                         continue;
 204
 205                 ret = osd_finalize_request(or, 0, _ios_cred(ios, i), NULL);
 206                 if (unlikely(ret)) {
 207                         ORE_DBGMSG("Failed to osd_finalize_request() => %d\n",
 208                                      ret);
 209                         return ret;
 210                 }
 211         }
 212
 213         kref_init(&ios->kref);
 214
 215         for (i = 0; i < ios->numdevs; i++) {
 216                 struct osd_request *or = ios->per_dev[i].or;
 217                 if (unlikely(!or))
 218                         continue;
 219
 220                 kref_get(&ios->kref);
 221                 osd_execute_request_async(or, _done_io, ios);
 222         }
 223
 224         kref_put(&ios->kref, _last_io);
 225         ret = 0;
 226
 227         if (sync) {
 228                 wait_for_completion(&wait);
 229                 ret = ore_check_io(ios, NULL);
 230         }
 231         return ret;
 232 }
 233
 234 static void _clear_bio(struct bio *bio)
 235 {
 236         struct bio_vec *bv;
 237         unsigned i;
 238
 239         __bio_for_each_segment(bv, bio, i, 0) {
 240                 unsigned this_count = bv->bv_len;
 241
 242                 if (likely(PAGE_SIZE == this_count))
 243                         clear_highpage(bv->bv_page);
 244                 else
 245                         zero_user(bv->bv_page, bv->bv_offset, this_count);
 246         }
 247 }
 248
 249 int ore_check_io(struct ore_io_state *ios, u64 *resid)
 250 {
 251         enum osd_err_priority acumulated_osd_err = 0;
 252         int acumulated_lin_err = 0;
 253         int i;
 254
 255         for (i = 0; i < ios->numdevs; i++) {
 256                 struct osd_sense_info osi;
 257                 struct osd_request *or = ios->per_dev[i].or;
 258                 int ret;
 259
 260                 if (unlikely(!or))
 261                         continue;
 262
 263                 ret = osd_req_decode_sense(or, &osi);
 264                 if (likely(!ret))
 265                         continue;
 266
 267                 if (OSD_ERR_PRI_CLEAR_PAGES == osi.osd_err_pri) {
 268                         /* start read offset passed endof file */
 269                         _clear_bio(ios->per_dev[i].bio);
 270                         ORE_DBGMSG("start read offset passed end of file "
 271                                 "offset=0x%llx, length=0x%llx\n",
 272                                 _LLU(ios->per_dev[i].offset),
 273                                 _LLU(ios->per_dev[i].length));
 274
 275                         continue; /* we recovered */
 276                 }
 277
 278                 if (osi.osd_err_pri >= acumulated_osd_err) {
 279                         acumulated_osd_err = osi.osd_err_pri;
 280                         acumulated_lin_err = ret;
 281                 }
 282         }
 283
 284         /* TODO: raid specific residual calculations */
 285         if (resid) {
 286                 if (likely(!acumulated_lin_err))
 287                         *resid = 0;
 288                 else
 289                         *resid = ios->length;
 290         }
 291
 292         return acumulated_lin_err;
 293 }
 294 EXPORT_SYMBOL(ore_check_io);
 295
 296 /*
 297  * L - logical offset into the file
 298  *
 299  * U - The number of bytes in a stripe within a group
 300  *
 301  *      U = stripe_unit * group_width
 302  *
 303  * T - The number of bytes striped within a group of component objects
 304  *     (before advancing to the next group)
 305  *
 306  *      T = stripe_unit * group_width * group_depth
 307  *
 308  * S - The number of bytes striped across all component objects
 309  *     before the pattern repeats
 310  *
 311  *      S = stripe_unit * group_width * group_depth * group_count
 312  *
 313  * M - The "major" (i.e., across all components) stripe number
 314  *
 315  *      M = L / S
 316  *
 317  * G - Counts the groups from the beginning of the major stripe
 318  *
 319  *      G = (L - (M * S)) / T   [or (L % S) / T]
 320  *
 321  * H - The byte offset within the group
 322  *
 323  *      H = (L - (M * S)) % T   [or (L % S) % T]
 324  *
 325  * N - The "minor" (i.e., across the group) stripe number
 326  *
 327  *      N = H / U
 328  *
 329  * C - The component index coresponding to L
 330  *
 331  *      C = (H - (N * U)) / stripe_unit + G * group_width
 332  *      [or (L % U) / stripe_unit + G * group_width]
 333  *
 334  * O - The component offset coresponding to L
 335  *
 336  *      O = L % stripe_unit + N * stripe_unit + M * group_depth * stripe_unit
 337  */
 338 static void ore_calc_stripe_info(struct ore_layout *layout, u64 file_offset,
 339                                  struct ore_striping_info *si)
 340 {
 341         u32     stripe_unit = layout->stripe_unit;
 342         u32     group_width = layout->group_width;
 343         u64     group_depth = layout->group_depth;
 344
 345         u32     U = stripe_unit * group_width;
 346         u64     T = U * group_depth;
 347         u64     S = T * layout->group_count;
 348         u64     M = div64_u64(file_offset, S);
 349
 350         /*
 351         G = (L - (M * S)) / T
 352         H = (L - (M * S)) % T
 353         */
 354         u64     LmodS = file_offset - M * S;
 355         u32     G = div64_u64(LmodS, T);
 356         u64     H = LmodS - G * T;
 357
 358         u32     N = div_u64(H, U);
 359
 360         /* "H - (N * U)" is just "H % U" so it's bound to u32 */
 361         si->dev = (u32)(H - (N * U)) / stripe_unit + G * group_width;
 362         si->dev *= layout->mirrors_p1;
 363
 364         div_u64_rem(file_offset, stripe_unit, &si->unit_off);
 365
 366         si->obj_offset = si->unit_off + (N * stripe_unit) +
 367                                   (M * group_depth * stripe_unit);
 368
 369         si->group_length = T - H;
 370         si->M = M;
 371 }
 372
 373 static int _add_stripe_unit(struct ore_io_state *ios,  unsigned *cur_pg,
 374                 unsigned pgbase, struct ore_per_dev_state *per_dev,
 375                 int cur_len)
 376 {
 377         unsigned pg = *cur_pg;
 378         struct request_queue *q =
 379                         osd_request_queue(_ios_od(ios, per_dev->dev));
 380
 381         per_dev->length += cur_len;
 382
 383         if (per_dev->bio == NULL) {
 384                 unsigned pages_in_stripe = ios->layout->group_width *
 385                                         (ios->layout->stripe_unit / PAGE_SIZE);
 386                 unsigned bio_size = (ios->nr_pages + pages_in_stripe) /
 387                                                 ios->layout->group_width;
 388
 389                 per_dev->bio = bio_kmalloc(GFP_KERNEL, bio_size);
 390                 if (unlikely(!per_dev->bio)) {
 391                         ORE_DBGMSG("Failed to allocate BIO size=%u\n",
 392                                      bio_size);
 393                         return -ENOMEM;
 394                 }
 395         }
 396
 397         while (cur_len > 0) {
 398                 unsigned pglen = min_t(unsigned, PAGE_SIZE - pgbase, cur_len);
 399                 unsigned added_len;
 400
 401                 BUG_ON(ios->nr_pages <= pg);
 402                 cur_len -= pglen;
 403
 404                 added_len = bio_add_pc_page(q, per_dev->bio, ios->pages[pg],
 405                                             pglen, pgbase);
 406                 if (unlikely(pglen != added_len))
 407                         return -ENOMEM;
 408                 pgbase = 0;
 409                 ++pg;
 410         }
 411         BUG_ON(cur_len);
 412
 413         *cur_pg = pg;
 414         return 0;
 415 }
 416
 417 static int _prepare_for_striping(struct ore_io_state *ios)
 418 {
 419         struct ore_striping_info *si = &ios->si;
 420         unsigned stripe_unit = ios->layout->stripe_unit;
 421         unsigned mirrors_p1 = ios->layout->mirrors_p1;
 422         unsigned devs_in_group = ios->layout->group_width * mirrors_p1;
 423         unsigned dev = si->dev;
 424         unsigned first_dev = dev - (dev % devs_in_group);
 425         unsigned cur_pg = ios->pages_consumed;
 426         u64 length = ios->length;
 427         int ret = 0;
 428
 429         if (!ios->pages) {
 430                 if (ios->kern_buff) {
 431                         struct ore_per_dev_state *per_dev = &ios->per_dev[0];
 432
 433                         per_dev->offset = si->obj_offset;
 434                         per_dev->dev = si->dev;
 435
 436                         /* no cross device without page array */
 437                         BUG_ON((ios->layout->group_width > 1) &&
 438                                (si->unit_off + ios->length >
 439                                 ios->layout->stripe_unit));
 440                 }
 441                 ios->numdevs = ios->layout->mirrors_p1;
 442                 return 0;
 443         }
 444
 445         BUG_ON(length > si->group_length);
 446
 447         while (length) {
 448                 unsigned comp = dev - first_dev;
 449                 struct ore_per_dev_state *per_dev = &ios->per_dev[comp];
 450                 unsigned cur_len, page_off = 0;
 451
 452                 if (!per_dev->length) {
 453                         per_dev->dev = dev;
 454                         if (dev < si->dev) {
 455                                 per_dev->offset = si->obj_offset + stripe_unit -
 456                                                                    si->unit_off;
 457                                 cur_len = stripe_unit;
 458                         } else if (dev == si->dev) {
 459                                 per_dev->offset = si->obj_offset;
 460                                 cur_len = stripe_unit - si->unit_off;
 461                                 page_off = si->unit_off & ~PAGE_MASK;
 462                                 BUG_ON(page_off && (page_off != ios->pgbase));
 463                         } else { /* dev > si->dev */
 464                                 per_dev->offset = si->obj_offset - si->unit_off;
 465                                 cur_len = stripe_unit;
 466                         }
 467                 } else {
 468                         cur_len = stripe_unit;
 469                 }
 470                 if (cur_len >= length)
 471                         cur_len = length;
 472
 473                 ret = _add_stripe_unit(ios, &cur_pg, page_off , per_dev,
 474                                        cur_len);
 475                 if (unlikely(ret))
 476                         goto out;
 477
 478                 dev += mirrors_p1;
 479                 dev = (dev % devs_in_group) + first_dev;
 480
 481                 length -= cur_len;
 482         }
 483 out:
 484         ios->numdevs = devs_in_group;
 485         ios->pages_consumed = cur_pg;
 486         return ret;
 487 }
 488
 489 int ore_create(struct ore_io_state *ios)
 490 {
 491         int i, ret;
 492
 493         for (i = 0; i < ios->oc->numdevs; i++) {
 494                 struct osd_request *or;
 495
 496                 or = osd_start_request(_ios_od(ios, i), GFP_KERNEL);
 497                 if (unlikely(!or)) {
 498                         ORE_ERR("%s: osd_start_request failed\n", __func__);
 499                         ret = -ENOMEM;
 500                         goto out;
 501                 }
 502                 ios->per_dev[i].or = or;
 503                 ios->numdevs++;
 504
 505                 osd_req_create_object(or, _ios_obj(ios, i));
 506         }
 507         ret = ore_io_execute(ios);
 508
 509 out:
 510         return ret;
 511 }
 512 EXPORT_SYMBOL(ore_create);
 513
 514 int ore_remove(struct ore_io_state *ios)
 515 {
 516         int i, ret;
 517
 518         for (i = 0; i < ios->oc->numdevs; i++) {
 519                 struct osd_request *or;
 520
 521                 or = osd_start_request(_ios_od(ios, i), GFP_KERNEL);
 522                 if (unlikely(!or)) {
 523                         ORE_ERR("%s: osd_start_request failed\n", __func__);
 524                         ret = -ENOMEM;
 525                         goto out;
 526                 }
 527                 ios->per_dev[i].or = or;
 528                 ios->numdevs++;
 529
 530                 osd_req_remove_object(or, _ios_obj(ios, i));
 531         }
 532         ret = ore_io_execute(ios);
 533
 534 out:
 535         return ret;
 536 }
 537 EXPORT_SYMBOL(ore_remove);
 538
 539 static int _write_mirror(struct ore_io_state *ios, int cur_comp)
 540 {
 541         struct ore_per_dev_state *master_dev = &ios->per_dev[cur_comp];
 542         unsigned dev = ios->per_dev[cur_comp].dev;
 543         unsigned last_comp = cur_comp + ios->layout->mirrors_p1;
 544         int ret = 0;
 545
 546         if (ios->pages && !master_dev->length)
 547                 return 0; /* Just an empty slot */
 548
 549         for (; cur_comp < last_comp; ++cur_comp, ++dev) {
 550                 struct ore_per_dev_state *per_dev = &ios->per_dev[cur_comp];
 551                 struct osd_request *or;
 552
 553                 or = osd_start_request(_ios_od(ios, dev), GFP_KERNEL);
 554                 if (unlikely(!or)) {
 555                         ORE_ERR("%s: osd_start_request failed\n", __func__);
 556                         ret = -ENOMEM;
 557                         goto out;
 558                 }
 559                 per_dev->or = or;
 560                 per_dev->offset = master_dev->offset;
 561
 562                 if (ios->pages) {
 563                         struct bio *bio;
 564
 565                         if (per_dev != master_dev) {
 566                                 bio = bio_kmalloc(GFP_KERNEL,
 567                                                   master_dev->bio->bi_max_vecs);
 568                                 if (unlikely(!bio)) {
 569                                         ORE_DBGMSG(
 570                                               "Failed to allocate BIO size=%u\n",
 571                                               master_dev->bio->bi_max_vecs);
 572                                         ret = -ENOMEM;
 573                                         goto out;
 574                                 }
 575
 576                                 __bio_clone(bio, master_dev->bio);
 577                                 bio->bi_bdev = NULL;
 578                                 bio->bi_next = NULL;
 579                                 per_dev->length = master_dev->length;
 580                                 per_dev->bio =  bio;
 581                                 per_dev->dev = dev;
 582                         } else {
 583                                 bio = master_dev->bio;
 584                                 /* FIXME: bio_set_dir() */
 585                                 bio->bi_rw |= REQ_WRITE;
 586                         }
 587
 588                         osd_req_write(or, _ios_obj(ios, dev), per_dev->offset,
 589                                       bio, per_dev->length);
 590                         ORE_DBGMSG("write(0x%llx) offset=0x%llx "
 591                                       "length=0x%llx dev=%d\n",
 592                                      _LLU(_ios_obj(ios, dev)->id),
 593                                      _LLU(per_dev->offset),
 594                                      _LLU(per_dev->length), dev);
 595                 } else if (ios->kern_buff) {
 596                         ret = osd_req_write_kern(or, _ios_obj(ios, dev),
 597                                                  per_dev->offset,
 598                                                  ios->kern_buff, ios->length);
 599                         if (unlikely(ret))
 600                                 goto out;
 601                         ORE_DBGMSG2("write_kern(0x%llx) offset=0x%llx "
 602                                       "length=0x%llx dev=%d\n",
 603                                      _LLU(_ios_obj(ios, dev)->id),
 604                                      _LLU(per_dev->offset),
 605                                      _LLU(ios->length), dev);
 606                 } else {
 607                         osd_req_set_attributes(or, _ios_obj(ios, dev));
 608                         ORE_DBGMSG2("obj(0x%llx) set_attributes=%d dev=%d\n",
 609                                      _LLU(_ios_obj(ios, dev)->id),
 610                                      ios->out_attr_len, dev);
 611                 }
 612
 613                 if (ios->out_attr)
 614                         osd_req_add_set_attr_list(or, ios->out_attr,
 615                                                   ios->out_attr_len);
 616
 617                 if (ios->in_attr)
 618                         osd_req_add_get_attr_list(or, ios->in_attr,
 619                                                   ios->in_attr_len);
 620         }
 621
 622 out:
 623         return ret;
 624 }
 625
 626 int ore_write(struct ore_io_state *ios)
 627 {
 628         int i;
 629         int ret;
 630
 631         ret = _prepare_for_striping(ios);
 632         if (unlikely(ret))
 633                 return ret;
 634
 635         for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
 636                 ret = _write_mirror(ios, i);
 637                 if (unlikely(ret))
 638                         return ret;
 639         }
 640
 641         ret = ore_io_execute(ios);
 642         return ret;
 643 }
 644 EXPORT_SYMBOL(ore_write);
 645
 646 static int _read_mirror(struct ore_io_state *ios, unsigned cur_comp)
 647 {
 648         struct osd_request *or;
 649         struct ore_per_dev_state *per_dev = &ios->per_dev[cur_comp];
 650         struct osd_obj_id *obj = _ios_obj(ios, cur_comp);
 651         unsigned first_dev = (unsigned)obj->id;
 652
 653         if (ios->pages && !per_dev->length)
 654                 return 0; /* Just an empty slot */
 655
 656         first_dev = per_dev->dev + first_dev % ios->layout->mirrors_p1;
 657         or = osd_start_request(_ios_od(ios, first_dev), GFP_KERNEL);
 658         if (unlikely(!or)) {
 659                 ORE_ERR("%s: osd_start_request failed\n", __func__);
 660                 return -ENOMEM;
 661         }
 662         per_dev->or = or;
 663
 664         if (ios->pages) {
 665                 osd_req_read(or, obj, per_dev->offset,
 666                                 per_dev->bio, per_dev->length);
 667                 ORE_DBGMSG("read(0x%llx) offset=0x%llx length=0x%llx"
 668                              " dev=%d\n", _LLU(obj->id),
 669                              _LLU(per_dev->offset), _LLU(per_dev->length),
 670                              first_dev);
 671         } else if (ios->kern_buff) {
 672                 int ret = osd_req_read_kern(or, obj, per_dev->offset,
 673                                             ios->kern_buff, ios->length);
 674                 ORE_DBGMSG2("read_kern(0x%llx) offset=0x%llx "
 675                               "length=0x%llx dev=%d ret=>%d\n",
 676                               _LLU(obj->id), _LLU(per_dev->offset),
 677                               _LLU(ios->length), first_dev, ret);
 678                 if (unlikely(ret))
 679                         return ret;
 680         } else {
 681                 osd_req_get_attributes(or, obj);
 682                 ORE_DBGMSG2("obj(0x%llx) get_attributes=%d dev=%d\n",
 683                               _LLU(obj->id),
 684                               ios->in_attr_len, first_dev);
 685         }
 686         if (ios->out_attr)
 687                 osd_req_add_set_attr_list(or, ios->out_attr, ios->out_attr_len);
 688
 689         if (ios->in_attr)
 690                 osd_req_add_get_attr_list(or, ios->in_attr, ios->in_attr_len);
 691
 692         return 0;
 693 }
 694
 695 int ore_read(struct ore_io_state *ios)
 696 {
 697         int i;
 698         int ret;
 699
 700         ret = _prepare_for_striping(ios);
 701         if (unlikely(ret))
 702                 return ret;
 703
 704         for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
 705                 ret = _read_mirror(ios, i);
 706                 if (unlikely(ret))
 707                         return ret;
 708         }
 709
 710         ret = ore_io_execute(ios);
 711         return ret;
 712 }
 713 EXPORT_SYMBOL(ore_read);
 714
 715 int extract_attr_from_ios(struct ore_io_state *ios, struct osd_attr *attr)
 716 {
 717         struct osd_attr cur_attr = {.attr_page = 0}; /* start with zeros */
 718         void *iter = NULL;
 719         int nelem;
 720
 721         do {
 722                 nelem = 1;
 723                 osd_req_decode_get_attr_list(ios->per_dev[0].or,
 724                                              &cur_attr, &nelem, &iter);
 725                 if ((cur_attr.attr_page == attr->attr_page) &&
 726                     (cur_attr.attr_id == attr->attr_id)) {
 727                         attr->len = cur_attr.len;
 728                         attr->val_ptr = cur_attr.val_ptr;
 729                         return 0;
 730                 }
 731         } while (iter);
 732
 733         return -EIO;
 734 }
 735 EXPORT_SYMBOL(extract_attr_from_ios);
 736
 737 static int _truncate_mirrors(struct ore_io_state *ios, unsigned cur_comp,
 738                              struct osd_attr *attr)
 739 {
 740         int last_comp = cur_comp + ios->layout->mirrors_p1;
 741
 742         for (; cur_comp < last_comp; ++cur_comp) {
 743                 struct ore_per_dev_state *per_dev = &ios->per_dev[cur_comp];
 744                 struct osd_request *or;
 745
 746                 or = osd_start_request(_ios_od(ios, cur_comp), GFP_KERNEL);
 747                 if (unlikely(!or)) {
 748                         ORE_ERR("%s: osd_start_request failed\n", __func__);
 749                         return -ENOMEM;
 750                 }
 751                 per_dev->or = or;
 752
 753                 osd_req_set_attributes(or, _ios_obj(ios, cur_comp));
 754                 osd_req_add_set_attr_list(or, attr, 1);
 755         }
 756
 757         return 0;
 758 }
 759
 760 struct _trunc_info {
 761         struct ore_striping_info si;
 762         u64 prev_group_obj_off;
 763         u64 next_group_obj_off;
 764
 765         unsigned first_group_dev;
 766         unsigned nex_group_dev;
 767 };
 768
 769 static void _calc_trunk_info(struct ore_layout *layout, u64 file_offset,
 770                              struct _trunc_info *ti)
 771 {
 772         unsigned stripe_unit = layout->stripe_unit;
 773
 774         ore_calc_stripe_info(layout, file_offset, &ti->si);
 775
 776         ti->prev_group_obj_off = ti->si.M * stripe_unit;
 777         ti->next_group_obj_off = ti->si.M ? (ti->si.M - 1) * stripe_unit : 0;
 778
 779         ti->first_group_dev = ti->si.dev - (ti->si.dev % layout->group_width);
 780         ti->nex_group_dev = ti->first_group_dev + layout->group_width;
 781 }
 782
 783 int ore_truncate(struct ore_layout *layout, struct ore_components *oc,
 784                    u64 size)
 785 {
 786         struct ore_io_state *ios;
 787         struct exofs_trunc_attr {
 788                 struct osd_attr attr;
 789                 __be64 newsize;
 790         } *size_attrs;
 791         struct _trunc_info ti;
 792         int i, ret;
 793
 794         ret = ore_get_io_state(layout, oc, &ios);
 795         if (unlikely(ret))
 796                 return ret;
 797
 798         _calc_trunk_info(ios->layout, size, &ti);
 799
 800         size_attrs = kcalloc(ios->oc->numdevs, sizeof(*size_attrs),
 801                              GFP_KERNEL);
 802         if (unlikely(!size_attrs)) {
 803                 ret = -ENOMEM;
 804                 goto out;
 805         }
 806
 807         ios->numdevs = ios->oc->numdevs;
 808
 809         for (i = 0; i < ios->numdevs; ++i) {
 810                 struct exofs_trunc_attr *size_attr = &size_attrs[i];
 811                 u64 obj_size;
 812
 813                 if (i < ti.first_group_dev)
 814                         obj_size = ti.prev_group_obj_off;
 815                 else if (i >= ti.nex_group_dev)
 816                         obj_size = ti.next_group_obj_off;
 817                 else if (i < ti.si.dev) /* dev within this group */
 818                         obj_size = ti.si.obj_offset +
 819                                       ios->layout->stripe_unit - ti.si.unit_off;
 820                 else if (i == ti.si.dev)
 821                         obj_size = ti.si.obj_offset;
 822                 else /* i > ti.dev */
 823                         obj_size = ti.si.obj_offset - ti.si.unit_off;
 824
 825                 size_attr->newsize = cpu_to_be64(obj_size);
 826                 size_attr->attr = g_attr_logical_length;
 827                 size_attr->attr.val_ptr = &size_attr->newsize;
 828
 829                 ORE_DBGMSG("trunc(0x%llx) obj_offset=0x%llx dev=%d\n",
 830                              _LLU(oc->comps->obj.id), _LLU(obj_size), i);
 831                 ret = _truncate_mirrors(ios, i * ios->layout->mirrors_p1,
 832                                         &size_attr->attr);
 833                 if (unlikely(ret))
 834                         goto out;
 835         }
 836         ret = ore_io_execute(ios);
 837
 838 out:
 839         kfree(size_attrs);
 840         ore_put_io_state(ios);
 841         return ret;
 842 }
 843 EXPORT_SYMBOL(ore_truncate);
 844
 845 const struct osd_attr g_attr_logical_length = ATTR_DEF(
 846         OSD_APAGE_OBJECT_INFORMATION, OSD_ATTR_OI_LOGICAL_LENGTH, 8);
 847 EXPORT_SYMBOL(g_attr_logical_length);