fs/exofs/ios.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 <scsi/scsi_device.h>
  27 #include <asm/div64.h>
  28
  29 #include "exofs.h"
  30
  31 #define EXOFS_DBGMSG2(M...) do {} while (0)
  32 /* #define EXOFS_DBGMSG2 EXOFS_DBGMSG */
  33
  34 void exofs_make_credential(u8 cred_a[OSD_CAP_LEN], const struct osd_obj_id *obj)
  35 {
  36         osd_sec_init_nosec_doall_caps(cred_a, obj, false, true);
  37 }
  38
  39 int exofs_read_kern(struct osd_dev *od, u8 *cred, struct osd_obj_id *obj,
  40                     u64 offset, void *p, unsigned length)
  41 {
  42         struct osd_request *or = osd_start_request(od, GFP_KERNEL);
  43 /*      struct osd_sense_info osi = {.key = 0};*/
  44         int ret;
  45
  46         if (unlikely(!or)) {
  47                 EXOFS_DBGMSG("%s: osd_start_request failed.\n", __func__);
  48                 return -ENOMEM;
  49         }
  50         ret = osd_req_read_kern(or, obj, offset, p, length);
  51         if (unlikely(ret)) {
  52                 EXOFS_DBGMSG("%s: osd_req_read_kern failed.\n", __func__);
  53                 goto out;
  54         }
  55
  56         ret = osd_finalize_request(or, 0, cred, NULL);
  57         if (unlikely(ret)) {
  58                 EXOFS_DBGMSG("Failed to osd_finalize_request() => %d\n", ret);
  59                 goto out;
  60         }
  61
  62         ret = osd_execute_request(or);
  63         if (unlikely(ret))
  64                 EXOFS_DBGMSG("osd_execute_request() => %d\n", ret);
  65         /* osd_req_decode_sense(or, ret); */
  66
  67 out:
  68         osd_end_request(or);
  69         return ret;
  70 }
  71
  72 int exofs_get_io_state(struct exofs_layout *layout,
  73                        struct exofs_io_state **pios)
  74 {
  75         struct exofs_io_state *ios;
  76
  77         /*TODO: Maybe use kmem_cach per sbi of size
  78          * exofs_io_state_size(layout->s_numdevs)
  79          */
  80         ios = kzalloc(exofs_io_state_size(layout->s_numdevs), GFP_KERNEL);
  81         if (unlikely(!ios)) {
  82                 EXOFS_DBGMSG("Failed kzalloc bytes=%d\n",
  83                              exofs_io_state_size(layout->s_numdevs));
  84                 *pios = NULL;
  85                 return -ENOMEM;
  86         }
  87
  88         ios->layout = layout;
  89         ios->obj.partition = layout->s_pid;
  90         *pios = ios;
  91         return 0;
  92 }
  93
  94 void exofs_put_io_state(struct exofs_io_state *ios)
  95 {
  96         if (ios) {
  97                 unsigned i;
  98
  99                 for (i = 0; i < ios->numdevs; i++) {
 100                         struct exofs_per_dev_state *per_dev = &ios->per_dev[i];
 101
 102                         if (per_dev->or)
 103                                 osd_end_request(per_dev->or);
 104                         if (per_dev->bio)
 105                                 bio_put(per_dev->bio);
 106                 }
 107
 108                 kfree(ios);
 109         }
 110 }
 111
 112 unsigned exofs_layout_od_id(struct exofs_layout *layout,
 113                             osd_id obj_no, unsigned layout_index)
 114 {
 115 /*      switch (layout->lay_func) {
 116         case LAYOUT_MOVING_WINDOW:
 117         {*/
 118                 unsigned dev_mod = obj_no;
 119
 120                 return (layout_index + dev_mod * layout->mirrors_p1) %
 121                                                               layout->s_numdevs;
 122 /*      }
 123         case LAYOUT_FUNC_IMPLICT:
 124                 return layout->devs[layout_index];
 125         }*/
 126 }
 127
 128 static inline struct osd_dev *exofs_ios_od(struct exofs_io_state *ios,
 129                                            unsigned layout_index)
 130 {
 131         return ios->layout->s_ods[
 132                 exofs_layout_od_id(ios->layout, ios->obj.id, layout_index)];
 133 }
 134
 135 static void _sync_done(struct exofs_io_state *ios, void *p)
 136 {
 137         struct completion *waiting = p;
 138
 139         complete(waiting);
 140 }
 141
 142 static void _last_io(struct kref *kref)
 143 {
 144         struct exofs_io_state *ios = container_of(
 145                                         kref, struct exofs_io_state, kref);
 146
 147         ios->done(ios, ios->private);
 148 }
 149
 150 static void _done_io(struct osd_request *or, void *p)
 151 {
 152         struct exofs_io_state *ios = p;
 153
 154         kref_put(&ios->kref, _last_io);
 155 }
 156
 157 static int exofs_io_execute(struct exofs_io_state *ios)
 158 {
 159         DECLARE_COMPLETION_ONSTACK(wait);
 160         bool sync = (ios->done == NULL);
 161         int i, ret;
 162
 163         if (sync) {
 164                 ios->done = _sync_done;
 165                 ios->private = &wait;
 166         }
 167
 168         for (i = 0; i < ios->numdevs; i++) {
 169                 struct osd_request *or = ios->per_dev[i].or;
 170                 if (unlikely(!or))
 171                         continue;
 172
 173                 ret = osd_finalize_request(or, 0, ios->cred, NULL);
 174                 if (unlikely(ret)) {
 175                         EXOFS_DBGMSG("Failed to osd_finalize_request() => %d\n",
 176                                      ret);
 177                         return ret;
 178                 }
 179         }
 180
 181         kref_init(&ios->kref);
 182
 183         for (i = 0; i < ios->numdevs; i++) {
 184                 struct osd_request *or = ios->per_dev[i].or;
 185                 if (unlikely(!or))
 186                         continue;
 187
 188                 kref_get(&ios->kref);
 189                 osd_execute_request_async(or, _done_io, ios);
 190         }
 191
 192         kref_put(&ios->kref, _last_io);
 193         ret = 0;
 194
 195         if (sync) {
 196                 wait_for_completion(&wait);
 197                 ret = exofs_check_io(ios, NULL);
 198         }
 199         return ret;
 200 }
 201
 202 static void _clear_bio(struct bio *bio)
 203 {
 204         struct bio_vec *bv;
 205         unsigned i;
 206
 207         __bio_for_each_segment(bv, bio, i, 0) {
 208                 unsigned this_count = bv->bv_len;
 209
 210                 if (likely(PAGE_SIZE == this_count))
 211                         clear_highpage(bv->bv_page);
 212                 else
 213                         zero_user(bv->bv_page, bv->bv_offset, this_count);
 214         }
 215 }
 216
 217 int exofs_check_io(struct exofs_io_state *ios, u64 *resid)
 218 {
 219         enum osd_err_priority acumulated_osd_err = 0;
 220         int acumulated_lin_err = 0;
 221         int i;
 222
 223         for (i = 0; i < ios->numdevs; i++) {
 224                 struct osd_sense_info osi;
 225                 struct osd_request *or = ios->per_dev[i].or;
 226                 int ret;
 227
 228                 if (unlikely(!or))
 229                         continue;
 230
 231                 ret = osd_req_decode_sense(or, &osi);
 232                 if (likely(!ret))
 233                         continue;
 234
 235                 if (OSD_ERR_PRI_CLEAR_PAGES == osi.osd_err_pri) {
 236                         /* start read offset passed endof file */
 237                         _clear_bio(ios->per_dev[i].bio);
 238                         EXOFS_DBGMSG("start read offset passed end of file "
 239                                 "offset=0x%llx, length=0x%llx\n",
 240                                 _LLU(ios->per_dev[i].offset),
 241                                 _LLU(ios->per_dev[i].length));
 242
 243                         continue; /* we recovered */
 244                 }
 245
 246                 if (osi.osd_err_pri >= acumulated_osd_err) {
 247                         acumulated_osd_err = osi.osd_err_pri;
 248                         acumulated_lin_err = ret;
 249                 }
 250         }
 251
 252         /* TODO: raid specific residual calculations */
 253         if (resid) {
 254                 if (likely(!acumulated_lin_err))
 255                         *resid = 0;
 256                 else
 257                         *resid = ios->length;
 258         }
 259
 260         return acumulated_lin_err;
 261 }
 262
 263 /*
 264  * L - logical offset into the file
 265  *
 266  * U - The number of bytes in a stripe within a group
 267  *
 268  *      U = stripe_unit * group_width
 269  *
 270  * T - The number of bytes striped within a group of component objects
 271  *     (before advancing to the next group)
 272  *
 273  *      T = stripe_unit * group_width * group_depth
 274  *
 275  * S - The number of bytes striped across all component objects
 276  *     before the pattern repeats
 277  *
 278  *      S = stripe_unit * group_width * group_depth * group_count
 279  *
 280  * M - The "major" (i.e., across all components) stripe number
 281  *
 282  *      M = L / S
 283  *
 284  * G - Counts the groups from the beginning of the major stripe
 285  *
 286  *      G = (L - (M * S)) / T   [or (L % S) / T]
 287  *
 288  * H - The byte offset within the group
 289  *
 290  *      H = (L - (M * S)) % T   [or (L % S) % T]
 291  *
 292  * N - The "minor" (i.e., across the group) stripe number
 293  *
 294  *      N = H / U
 295  *
 296  * C - The component index coresponding to L
 297  *
 298  *      C = (H - (N * U)) / stripe_unit + G * group_width
 299  *      [or (L % U) / stripe_unit + G * group_width]
 300  *
 301  * O - The component offset coresponding to L
 302  *
 303  *      O = L % stripe_unit + N * stripe_unit + M * group_depth * stripe_unit
 304  */
 305 struct _striping_info {
 306         u64 obj_offset;
 307         u64 group_length;
 308         u64 M; /* for truncate */
 309         unsigned dev;
 310         unsigned unit_off;
 311 };
 312
 313 static void _calc_stripe_info(struct exofs_layout *layout, u64 file_offset,
 314                               struct _striping_info *si)
 315 {
 316         u32     stripe_unit = layout->stripe_unit;
 317         u32     group_width = layout->group_width;
 318         u64     group_depth = layout->group_depth;
 319
 320         u32     U = stripe_unit * group_width;
 321         u64     T = U * group_depth;
 322         u64     S = T * layout->group_count;
 323         u64     M = div64_u64(file_offset, S);
 324
 325         /*
 326         G = (L - (M * S)) / T
 327         H = (L - (M * S)) % T
 328         */
 329         u64     LmodS = file_offset - M * S;
 330         u32     G = div64_u64(LmodS, T);
 331         u64     H = LmodS - G * T;
 332
 333         u32     N = div_u64(H, U);
 334
 335         /* "H - (N * U)" is just "H % U" so it's bound to u32 */
 336         si->dev = (u32)(H - (N * U)) / stripe_unit + G * group_width;
 337         si->dev *= layout->mirrors_p1;
 338
 339         div_u64_rem(file_offset, stripe_unit, &si->unit_off);
 340
 341         si->obj_offset = si->unit_off + (N * stripe_unit) +
 342                                   (M * group_depth * stripe_unit);
 343
 344         si->group_length = T - H;
 345         si->M = M;
 346 }
 347
 348 static int _add_stripe_unit(struct exofs_io_state *ios,  unsigned *cur_pg,
 349                 unsigned pgbase, struct exofs_per_dev_state *per_dev,
 350                 int cur_len)
 351 {
 352         unsigned pg = *cur_pg;
 353         struct request_queue *q =
 354                         osd_request_queue(exofs_ios_od(ios, per_dev->dev));
 355
 356         per_dev->length += cur_len;
 357
 358         if (per_dev->bio == NULL) {
 359                 unsigned pages_in_stripe = ios->layout->group_width *
 360                                         (ios->layout->stripe_unit / PAGE_SIZE);
 361                 unsigned bio_size = (ios->nr_pages + pages_in_stripe) /
 362                                                 ios->layout->group_width;
 363
 364                 per_dev->bio = bio_kmalloc(GFP_KERNEL, bio_size);
 365                 if (unlikely(!per_dev->bio)) {
 366                         EXOFS_DBGMSG("Failed to allocate BIO size=%u\n",
 367                                      bio_size);
 368                         return -ENOMEM;
 369                 }
 370         }
 371
 372         while (cur_len > 0) {
 373                 unsigned pglen = min_t(unsigned, PAGE_SIZE - pgbase, cur_len);
 374                 unsigned added_len;
 375
 376                 BUG_ON(ios->nr_pages <= pg);
 377                 cur_len -= pglen;
 378
 379                 added_len = bio_add_pc_page(q, per_dev->bio, ios->pages[pg],
 380                                             pglen, pgbase);
 381                 if (unlikely(pglen != added_len))
 382                         return -ENOMEM;
 383                 pgbase = 0;
 384                 ++pg;
 385         }
 386         BUG_ON(cur_len);
 387
 388         *cur_pg = pg;
 389         return 0;
 390 }
 391
 392 static int _prepare_one_group(struct exofs_io_state *ios, u64 length,
 393                               struct _striping_info *si)
 394 {
 395         unsigned stripe_unit = ios->layout->stripe_unit;
 396         unsigned mirrors_p1 = ios->layout->mirrors_p1;
 397         unsigned devs_in_group = ios->layout->group_width * mirrors_p1;
 398         unsigned dev = si->dev;
 399         unsigned first_dev = dev - (dev % devs_in_group);
 400         unsigned max_comp = ios->numdevs ? ios->numdevs - mirrors_p1 : 0;
 401         unsigned cur_pg = ios->pages_consumed;
 402         int ret = 0;
 403
 404         while (length) {
 405                 struct exofs_per_dev_state *per_dev = &ios->per_dev[dev];
 406                 unsigned cur_len, page_off = 0;
 407
 408                 if (!per_dev->length) {
 409                         per_dev->dev = dev;
 410                         if (dev < si->dev) {
 411                                 per_dev->offset = si->obj_offset + stripe_unit -
 412                                                                    si->unit_off;
 413                                 cur_len = stripe_unit;
 414                         } else if (dev == si->dev) {
 415                                 per_dev->offset = si->obj_offset;
 416                                 cur_len = stripe_unit - si->unit_off;
 417                                 page_off = si->unit_off & ~PAGE_MASK;
 418                                 BUG_ON(page_off && (page_off != ios->pgbase));
 419                         } else { /* dev > si->dev */
 420                                 per_dev->offset = si->obj_offset - si->unit_off;
 421                                 cur_len = stripe_unit;
 422                         }
 423
 424                         if (max_comp < dev)
 425                                 max_comp = dev;
 426                 } else {
 427                         cur_len = stripe_unit;
 428                 }
 429                 if (cur_len >= length)
 430                         cur_len = length;
 431
 432                 ret = _add_stripe_unit(ios, &cur_pg, page_off , per_dev,
 433                                        cur_len);
 434                 if (unlikely(ret))
 435                         goto out;
 436
 437                 dev += mirrors_p1;
 438                 dev = (dev % devs_in_group) + first_dev;
 439
 440                 length -= cur_len;
 441         }
 442 out:
 443         ios->numdevs = max_comp + mirrors_p1;
 444         ios->pages_consumed = cur_pg;
 445         return ret;
 446 }
 447
 448 static int _prepare_for_striping(struct exofs_io_state *ios)
 449 {
 450         u64 length = ios->length;
 451         u64 offset = ios->offset;
 452         struct _striping_info si;
 453         int ret = 0;
 454
 455         if (!ios->pages) {
 456                 if (ios->kern_buff) {
 457                         struct exofs_per_dev_state *per_dev = &ios->per_dev[0];
 458
 459                         _calc_stripe_info(ios->layout, ios->offset, &si);
 460                         per_dev->offset = si.obj_offset;
 461                         per_dev->dev = si.dev;
 462
 463                         /* no cross device without page array */
 464                         BUG_ON((ios->layout->group_width > 1) &&
 465                                (si.unit_off + ios->length >
 466                                 ios->layout->stripe_unit));
 467                 }
 468                 ios->numdevs = ios->layout->mirrors_p1;
 469                 return 0;
 470         }
 471
 472         while (length) {
 473                 _calc_stripe_info(ios->layout, offset, &si);
 474
 475                 if (length < si.group_length)
 476                         si.group_length = length;
 477
 478                 ret = _prepare_one_group(ios, si.group_length, &si);
 479                 if (unlikely(ret))
 480                         goto out;
 481
 482                 offset += si.group_length;
 483                 length -= si.group_length;
 484         }
 485
 486 out:
 487         return ret;
 488 }
 489
 490 int exofs_sbi_create(struct exofs_io_state *ios)
 491 {
 492         int i, ret;
 493
 494         for (i = 0; i < ios->layout->s_numdevs; i++) {
 495                 struct osd_request *or;
 496
 497                 or = osd_start_request(exofs_ios_od(ios, i), GFP_KERNEL);
 498                 if (unlikely(!or)) {
 499                         EXOFS_ERR("%s: osd_start_request failed\n", __func__);
 500                         ret = -ENOMEM;
 501                         goto out;
 502                 }
 503                 ios->per_dev[i].or = or;
 504                 ios->numdevs++;
 505
 506                 osd_req_create_object(or, &ios->obj);
 507         }
 508         ret = exofs_io_execute(ios);
 509
 510 out:
 511         return ret;
 512 }
 513
 514 int exofs_sbi_remove(struct exofs_io_state *ios)
 515 {
 516         int i, ret;
 517
 518         for (i = 0; i < ios->layout->s_numdevs; i++) {
 519                 struct osd_request *or;
 520
 521                 or = osd_start_request(exofs_ios_od(ios, i), GFP_KERNEL);
 522                 if (unlikely(!or)) {
 523                         EXOFS_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);
 531         }
 532         ret = exofs_io_execute(ios);
 533
 534 out:
 535         return ret;
 536 }
 537
 538 static int _sbi_write_mirror(struct exofs_io_state *ios, int cur_comp)
 539 {
 540         struct exofs_per_dev_state *master_dev = &ios->per_dev[cur_comp];
 541         unsigned dev = ios->per_dev[cur_comp].dev;
 542         unsigned last_comp = cur_comp + ios->layout->mirrors_p1;
 543         int ret = 0;
 544
 545         if (ios->pages && !master_dev->length)
 546                 return 0; /* Just an empty slot */
 547
 548         for (; cur_comp < last_comp; ++cur_comp, ++dev) {
 549                 struct exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp];
 550                 struct osd_request *or;
 551
 552                 or = osd_start_request(exofs_ios_od(ios, dev), GFP_KERNEL);
 553                 if (unlikely(!or)) {
 554                         EXOFS_ERR("%s: osd_start_request failed\n", __func__);
 555                         ret = -ENOMEM;
 556                         goto out;
 557                 }
 558                 per_dev->or = or;
 559                 per_dev->offset = master_dev->offset;
 560
 561                 if (ios->pages) {
 562                         struct bio *bio;
 563
 564                         if (per_dev != master_dev) {
 565                                 bio = bio_kmalloc(GFP_KERNEL,
 566                                                   master_dev->bio->bi_max_vecs);
 567                                 if (unlikely(!bio)) {
 568                                         EXOFS_DBGMSG(
 569                                               "Failed to allocate BIO size=%u\n",
 570                                               master_dev->bio->bi_max_vecs);
 571                                         ret = -ENOMEM;
 572                                         goto out;
 573                                 }
 574
 575                                 __bio_clone(bio, master_dev->bio);
 576                                 bio->bi_bdev = NULL;
 577                                 bio->bi_next = NULL;
 578                                 per_dev->length = master_dev->length;
 579                                 per_dev->bio =  bio;
 580                                 per_dev->dev = dev;
 581                         } else {
 582                                 bio = master_dev->bio;
 583                                 /* FIXME: bio_set_dir() */
 584                                 bio->bi_rw |= REQ_WRITE;
 585                         }
 586
 587                         osd_req_write(or, &ios->obj, per_dev->offset, bio,
 588                                       per_dev->length);
 589                         EXOFS_DBGMSG("write(0x%llx) offset=0x%llx "
 590                                       "length=0x%llx dev=%d\n",
 591                                      _LLU(ios->obj.id), _LLU(per_dev->offset),
 592                                      _LLU(per_dev->length), dev);
 593                 } else if (ios->kern_buff) {
 594                         ret = osd_req_write_kern(or, &ios->obj, per_dev->offset,
 595                                            ios->kern_buff, ios->length);
 596                         if (unlikely(ret))
 597                                 goto out;
 598                         EXOFS_DBGMSG2("write_kern(0x%llx) offset=0x%llx "
 599                                       "length=0x%llx dev=%d\n",
 600                                      _LLU(ios->obj.id), _LLU(per_dev->offset),
 601                                      _LLU(ios->length), dev);
 602                 } else {
 603                         osd_req_set_attributes(or, &ios->obj);
 604                         EXOFS_DBGMSG2("obj(0x%llx) set_attributes=%d dev=%d\n",
 605                                      _LLU(ios->obj.id), ios->out_attr_len, dev);
 606                 }
 607
 608                 if (ios->out_attr)
 609                         osd_req_add_set_attr_list(or, ios->out_attr,
 610                                                   ios->out_attr_len);
 611
 612                 if (ios->in_attr)
 613                         osd_req_add_get_attr_list(or, ios->in_attr,
 614                                                   ios->in_attr_len);
 615         }
 616
 617 out:
 618         return ret;
 619 }
 620
 621 int exofs_sbi_write(struct exofs_io_state *ios)
 622 {
 623         int i;
 624         int ret;
 625
 626         ret = _prepare_for_striping(ios);
 627         if (unlikely(ret))
 628                 return ret;
 629
 630         for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
 631                 ret = _sbi_write_mirror(ios, i);
 632                 if (unlikely(ret))
 633                         return ret;
 634         }
 635
 636         ret = exofs_io_execute(ios);
 637         return ret;
 638 }
 639
 640 static int _sbi_read_mirror(struct exofs_io_state *ios, unsigned cur_comp)
 641 {
 642         struct osd_request *or;
 643         struct exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp];
 644         unsigned first_dev = (unsigned)ios->obj.id;
 645
 646         if (ios->pages && !per_dev->length)
 647                 return 0; /* Just an empty slot */
 648
 649         first_dev = per_dev->dev + first_dev % ios->layout->mirrors_p1;
 650         or = osd_start_request(exofs_ios_od(ios, first_dev), GFP_KERNEL);
 651         if (unlikely(!or)) {
 652                 EXOFS_ERR("%s: osd_start_request failed\n", __func__);
 653                 return -ENOMEM;
 654         }
 655         per_dev->or = or;
 656
 657         if (ios->pages) {
 658                 osd_req_read(or, &ios->obj, per_dev->offset,
 659                                 per_dev->bio, per_dev->length);
 660                 EXOFS_DBGMSG("read(0x%llx) offset=0x%llx length=0x%llx"
 661                              " dev=%d\n", _LLU(ios->obj.id),
 662                              _LLU(per_dev->offset), _LLU(per_dev->length),
 663                              first_dev);
 664         } else if (ios->kern_buff) {
 665                 int ret = osd_req_read_kern(or, &ios->obj, per_dev->offset,
 666                                             ios->kern_buff, ios->length);
 667                 EXOFS_DBGMSG2("read_kern(0x%llx) offset=0x%llx "
 668                               "length=0x%llx dev=%d ret=>%d\n",
 669                               _LLU(ios->obj.id), _LLU(per_dev->offset),
 670                               _LLU(ios->length), first_dev, ret);
 671                 if (unlikely(ret))
 672                         return ret;
 673         } else {
 674                 osd_req_get_attributes(or, &ios->obj);
 675                 EXOFS_DBGMSG2("obj(0x%llx) get_attributes=%d dev=%d\n",
 676                               _LLU(ios->obj.id), ios->in_attr_len, first_dev);
 677         }
 678         if (ios->out_attr)
 679                 osd_req_add_set_attr_list(or, ios->out_attr, ios->out_attr_len);
 680
 681         if (ios->in_attr)
 682                 osd_req_add_get_attr_list(or, ios->in_attr, ios->in_attr_len);
 683
 684         return 0;
 685 }
 686
 687 int exofs_sbi_read(struct exofs_io_state *ios)
 688 {
 689         int i;
 690         int ret;
 691
 692         ret = _prepare_for_striping(ios);
 693         if (unlikely(ret))
 694                 return ret;
 695
 696         for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
 697                 ret = _sbi_read_mirror(ios, i);
 698                 if (unlikely(ret))
 699                         return ret;
 700         }
 701
 702         ret = exofs_io_execute(ios);
 703         return ret;
 704 }
 705
 706 int extract_attr_from_ios(struct exofs_io_state *ios, struct osd_attr *attr)
 707 {
 708         struct osd_attr cur_attr = {.attr_page = 0}; /* start with zeros */
 709         void *iter = NULL;
 710         int nelem;
 711
 712         do {
 713                 nelem = 1;
 714                 osd_req_decode_get_attr_list(ios->per_dev[0].or,
 715                                              &cur_attr, &nelem, &iter);
 716                 if ((cur_attr.attr_page == attr->attr_page) &&
 717                     (cur_attr.attr_id == attr->attr_id)) {
 718                         attr->len = cur_attr.len;
 719                         attr->val_ptr = cur_attr.val_ptr;
 720                         return 0;
 721                 }
 722         } while (iter);
 723
 724         return -EIO;
 725 }
 726
 727 static int _truncate_mirrors(struct exofs_io_state *ios, unsigned cur_comp,
 728                              struct osd_attr *attr)
 729 {
 730         int last_comp = cur_comp + ios->layout->mirrors_p1;
 731
 732         for (; cur_comp < last_comp; ++cur_comp) {
 733                 struct exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp];
 734                 struct osd_request *or;
 735
 736                 or = osd_start_request(exofs_ios_od(ios, cur_comp), GFP_KERNEL);
 737                 if (unlikely(!or)) {
 738                         EXOFS_ERR("%s: osd_start_request failed\n", __func__);
 739                         return -ENOMEM;
 740                 }
 741                 per_dev->or = or;
 742
 743                 osd_req_set_attributes(or, &ios->obj);
 744                 osd_req_add_set_attr_list(or, attr, 1);
 745         }
 746
 747         return 0;
 748 }
 749
 750 struct _trunc_info {
 751         struct _striping_info si;
 752         u64 prev_group_obj_off;
 753         u64 next_group_obj_off;
 754
 755         unsigned first_group_dev;
 756         unsigned nex_group_dev;
 757         unsigned max_devs;
 758 };
 759
 760 void _calc_trunk_info(struct exofs_layout *layout, u64 file_offset,
 761                        struct _trunc_info *ti)
 762 {
 763         unsigned stripe_unit = layout->stripe_unit;
 764
 765         _calc_stripe_info(layout, file_offset, &ti->si);
 766
 767         ti->prev_group_obj_off = ti->si.M * stripe_unit;
 768         ti->next_group_obj_off = ti->si.M ? (ti->si.M - 1) * stripe_unit : 0;
 769
 770         ti->first_group_dev = ti->si.dev - (ti->si.dev % layout->group_width);
 771         ti->nex_group_dev = ti->first_group_dev + layout->group_width;
 772         ti->max_devs = layout->group_width * layout->group_count;
 773 }
 774
 775 int exofs_oi_truncate(struct exofs_i_info *oi, u64 size)
 776 {
 777         struct exofs_sb_info *sbi = oi->vfs_inode.i_sb->s_fs_info;
 778         struct exofs_io_state *ios;
 779         struct exofs_trunc_attr {
 780                 struct osd_attr attr;
 781                 __be64 newsize;
 782         } *size_attrs;
 783         struct _trunc_info ti;
 784         int i, ret;
 785
 786         ret = exofs_get_io_state(&sbi->layout, &ios);
 787         if (unlikely(ret))
 788                 return ret;
 789
 790         _calc_trunk_info(ios->layout, size, &ti);
 791
 792         size_attrs = kcalloc(ti.max_devs, sizeof(*size_attrs),
 793                              GFP_KERNEL);
 794         if (unlikely(!size_attrs)) {
 795                 ret = -ENOMEM;
 796                 goto out;
 797         }
 798
 799         ios->obj.id = exofs_oi_objno(oi);
 800         ios->cred = oi->i_cred;
 801         ios->numdevs = ios->layout->s_numdevs;
 802
 803         for (i = 0; i < ti.max_devs; ++i) {
 804                 struct exofs_trunc_attr *size_attr = &size_attrs[i];
 805                 u64 obj_size;
 806
 807                 if (i < ti.first_group_dev)
 808                         obj_size = ti.prev_group_obj_off;
 809                 else if (i >= ti.nex_group_dev)
 810                         obj_size = ti.next_group_obj_off;
 811                 else if (i < ti.si.dev) /* dev within this group */
 812                         obj_size = ti.si.obj_offset +
 813                                       ios->layout->stripe_unit - ti.si.unit_off;
 814                 else if (i == ti.si.dev)
 815                         obj_size = ti.si.obj_offset;
 816                 else /* i > ti.dev */
 817                         obj_size = ti.si.obj_offset - ti.si.unit_off;
 818
 819                 size_attr->newsize = cpu_to_be64(obj_size);
 820                 size_attr->attr = g_attr_logical_length;
 821                 size_attr->attr.val_ptr = &size_attr->newsize;
 822
 823                 EXOFS_DBGMSG("trunc(0x%llx) obj_offset=0x%llx dev=%d\n",
 824                              _LLU(ios->obj.id), _LLU(obj_size), i);
 825                 ret = _truncate_mirrors(ios, i * ios->layout->mirrors_p1,
 826                                         &size_attr->attr);
 827                 if (unlikely(ret))
 828                         goto out;
 829         }
 830         ret = exofs_io_execute(ios);
 831
 832 out:
 833         kfree(size_attrs);
 834         exofs_put_io_state(ios);
 835         return ret;
 836 }