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