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_rw_state(struct exofs_layout *layout, bool is_reading,
  73                         u64 offset, u64 length, 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         ios->offset = offset;
  91         ios->length = length;
  92         ios->reading = is_reading;
  93
  94         *pios = ios;
  95         return 0;
  96 }
  97
  98 int  exofs_get_io_state(struct exofs_layout *layout,
  99                         struct exofs_io_state **ios)
 100 {
 101         return exofs_get_rw_state(layout, true, 0, 0, ios);
 102 }
 103
 104 void exofs_put_io_state(struct exofs_io_state *ios)
 105 {
 106         if (ios) {
 107                 unsigned i;
 108
 109                 for (i = 0; i < ios->numdevs; i++) {
 110                         struct exofs_per_dev_state *per_dev = &ios->per_dev[i];
 111
 112                         if (per_dev->or)
 113                                 osd_end_request(per_dev->or);
 114                         if (per_dev->bio)
 115                                 bio_put(per_dev->bio);
 116                 }
 117
 118                 kfree(ios);
 119         }
 120 }
 121
 122 unsigned exofs_layout_od_id(struct exofs_layout *layout,
 123                             osd_id obj_no, unsigned layout_index)
 124 {
 125 /*      switch (layout->lay_func) {
 126         case LAYOUT_MOVING_WINDOW:
 127         {*/
 128                 unsigned dev_mod = obj_no;
 129
 130                 return (layout_index + dev_mod * layout->mirrors_p1) %
 131                                                               layout->s_numdevs;
 132 /*      }
 133         case LAYOUT_FUNC_IMPLICT:
 134                 return layout->devs[layout_index];
 135         }*/
 136 }
 137
 138 static inline struct osd_dev *exofs_ios_od(struct exofs_io_state *ios,
 139                                            unsigned layout_index)
 140 {
 141         return ios->layout->s_ods[
 142                 exofs_layout_od_id(ios->layout, ios->obj.id, layout_index)];
 143 }
 144
 145 static void _sync_done(struct exofs_io_state *ios, void *p)
 146 {
 147         struct completion *waiting = p;
 148
 149         complete(waiting);
 150 }
 151
 152 static void _last_io(struct kref *kref)
 153 {
 154         struct exofs_io_state *ios = container_of(
 155                                         kref, struct exofs_io_state, kref);
 156
 157         ios->done(ios, ios->private);
 158 }
 159
 160 static void _done_io(struct osd_request *or, void *p)
 161 {
 162         struct exofs_io_state *ios = p;
 163
 164         kref_put(&ios->kref, _last_io);
 165 }
 166
 167 static int exofs_io_execute(struct exofs_io_state *ios)
 168 {
 169         DECLARE_COMPLETION_ONSTACK(wait);
 170         bool sync = (ios->done == NULL);
 171         int i, ret;
 172
 173         if (sync) {
 174                 ios->done = _sync_done;
 175                 ios->private = &wait;
 176         }
 177
 178         for (i = 0; i < ios->numdevs; i++) {
 179                 struct osd_request *or = ios->per_dev[i].or;
 180                 if (unlikely(!or))
 181                         continue;
 182
 183                 ret = osd_finalize_request(or, 0, ios->cred, NULL);
 184                 if (unlikely(ret)) {
 185                         EXOFS_DBGMSG("Failed to osd_finalize_request() => %d\n",
 186                                      ret);
 187                         return ret;
 188                 }
 189         }
 190
 191         kref_init(&ios->kref);
 192
 193         for (i = 0; i < ios->numdevs; i++) {
 194                 struct osd_request *or = ios->per_dev[i].or;
 195                 if (unlikely(!or))
 196                         continue;
 197
 198                 kref_get(&ios->kref);
 199                 osd_execute_request_async(or, _done_io, ios);
 200         }
 201
 202         kref_put(&ios->kref, _last_io);
 203         ret = 0;
 204
 205         if (sync) {
 206                 wait_for_completion(&wait);
 207                 ret = exofs_check_io(ios, NULL);
 208         }
 209         return ret;
 210 }
 211
 212 static void _clear_bio(struct bio *bio)
 213 {
 214         struct bio_vec *bv;
 215         unsigned i;
 216
 217         __bio_for_each_segment(bv, bio, i, 0) {
 218                 unsigned this_count = bv->bv_len;
 219
 220                 if (likely(PAGE_SIZE == this_count))
 221                         clear_highpage(bv->bv_page);
 222                 else
 223                         zero_user(bv->bv_page, bv->bv_offset, this_count);
 224         }
 225 }
 226
 227 int exofs_check_io(struct exofs_io_state *ios, u64 *resid)
 228 {
 229         enum osd_err_priority acumulated_osd_err = 0;
 230         int acumulated_lin_err = 0;
 231         int i;
 232
 233         for (i = 0; i < ios->numdevs; i++) {
 234                 struct osd_sense_info osi;
 235                 struct osd_request *or = ios->per_dev[i].or;
 236                 int ret;
 237
 238                 if (unlikely(!or))
 239                         continue;
 240
 241                 ret = osd_req_decode_sense(or, &osi);
 242                 if (likely(!ret))
 243                         continue;
 244
 245                 if (OSD_ERR_PRI_CLEAR_PAGES == osi.osd_err_pri) {
 246                         /* start read offset passed endof file */
 247                         _clear_bio(ios->per_dev[i].bio);
 248                         EXOFS_DBGMSG("start read offset passed end of file "
 249                                 "offset=0x%llx, length=0x%llx\n",
 250                                 _LLU(ios->per_dev[i].offset),
 251                                 _LLU(ios->per_dev[i].length));
 252
 253                         continue; /* we recovered */
 254                 }
 255
 256                 if (osi.osd_err_pri >= acumulated_osd_err) {
 257                         acumulated_osd_err = osi.osd_err_pri;
 258                         acumulated_lin_err = ret;
 259                 }
 260         }
 261
 262         /* TODO: raid specific residual calculations */
 263         if (resid) {
 264                 if (likely(!acumulated_lin_err))
 265                         *resid = 0;
 266                 else
 267                         *resid = ios->length;
 268         }
 269
 270         return acumulated_lin_err;
 271 }
 272
 273 /*
 274  * L - logical offset into the file
 275  *
 276  * U - The number of bytes in a stripe within a group
 277  *
 278  *      U = stripe_unit * group_width
 279  *
 280  * T - The number of bytes striped within a group of component objects
 281  *     (before advancing to the next group)
 282  *
 283  *      T = stripe_unit * group_width * group_depth
 284  *
 285  * S - The number of bytes striped across all component objects
 286  *     before the pattern repeats
 287  *
 288  *      S = stripe_unit * group_width * group_depth * group_count
 289  *
 290  * M - The "major" (i.e., across all components) stripe number
 291  *
 292  *      M = L / S
 293  *
 294  * G - Counts the groups from the beginning of the major stripe
 295  *
 296  *      G = (L - (M * S)) / T   [or (L % S) / T]
 297  *
 298  * H - The byte offset within the group
 299  *
 300  *      H = (L - (M * S)) % T   [or (L % S) % T]
 301  *
 302  * N - The "minor" (i.e., across the group) stripe number
 303  *
 304  *      N = H / U
 305  *
 306  * C - The component index coresponding to L
 307  *
 308  *      C = (H - (N * U)) / stripe_unit + G * group_width
 309  *      [or (L % U) / stripe_unit + G * group_width]
 310  *
 311  * O - The component offset coresponding to L
 312  *
 313  *      O = L % stripe_unit + N * stripe_unit + M * group_depth * stripe_unit
 314  */
 315 struct _striping_info {
 316         u64 obj_offset;
 317         u64 group_length;
 318         u64 M; /* for truncate */
 319         unsigned dev;
 320         unsigned unit_off;
 321 };
 322
 323 static void _calc_stripe_info(struct exofs_layout *layout, u64 file_offset,
 324                               struct _striping_info *si)
 325 {
 326         u32     stripe_unit = layout->stripe_unit;
 327         u32     group_width = layout->group_width;
 328         u64     group_depth = layout->group_depth;
 329
 330         u32     U = stripe_unit * group_width;
 331         u64     T = U * group_depth;
 332         u64     S = T * layout->group_count;
 333         u64     M = div64_u64(file_offset, S);
 334
 335         /*
 336         G = (L - (M * S)) / T
 337         H = (L - (M * S)) % T
 338         */
 339         u64     LmodS = file_offset - M * S;
 340         u32     G = div64_u64(LmodS, T);
 341         u64     H = LmodS - G * T;
 342
 343         u32     N = div_u64(H, U);
 344
 345         /* "H - (N * U)" is just "H % U" so it's bound to u32 */
 346         si->dev = (u32)(H - (N * U)) / stripe_unit + G * group_width;
 347         si->dev *= layout->mirrors_p1;
 348
 349         div_u64_rem(file_offset, stripe_unit, &si->unit_off);
 350
 351         si->obj_offset = si->unit_off + (N * stripe_unit) +
 352                                   (M * group_depth * stripe_unit);
 353
 354         si->group_length = T - H;
 355         si->M = M;
 356 }
 357
 358 static int _add_stripe_unit(struct exofs_io_state *ios,  unsigned *cur_pg,
 359                 unsigned pgbase, struct exofs_per_dev_state *per_dev,
 360                 int cur_len)
 361 {
 362         unsigned pg = *cur_pg;
 363         struct request_queue *q =
 364                         osd_request_queue(exofs_ios_od(ios, per_dev->dev));
 365
 366         per_dev->length += cur_len;
 367
 368         if (per_dev->bio == NULL) {
 369                 unsigned pages_in_stripe = ios->layout->group_width *
 370                                         (ios->layout->stripe_unit / PAGE_SIZE);
 371                 unsigned bio_size = (ios->nr_pages + pages_in_stripe) /
 372                                                 ios->layout->group_width;
 373
 374                 per_dev->bio = bio_kmalloc(GFP_KERNEL, bio_size);
 375                 if (unlikely(!per_dev->bio)) {
 376                         EXOFS_DBGMSG("Failed to allocate BIO size=%u\n",
 377                                      bio_size);
 378                         return -ENOMEM;
 379                 }
 380         }
 381
 382         while (cur_len > 0) {
 383                 unsigned pglen = min_t(unsigned, PAGE_SIZE - pgbase, cur_len);
 384                 unsigned added_len;
 385
 386                 BUG_ON(ios->nr_pages <= pg);
 387                 cur_len -= pglen;
 388
 389                 added_len = bio_add_pc_page(q, per_dev->bio, ios->pages[pg],
 390                                             pglen, pgbase);
 391                 if (unlikely(pglen != added_len))
 392                         return -ENOMEM;
 393                 pgbase = 0;
 394                 ++pg;
 395         }
 396         BUG_ON(cur_len);
 397
 398         *cur_pg = pg;
 399         return 0;
 400 }
 401
 402 static int _prepare_one_group(struct exofs_io_state *ios, u64 length,
 403                               struct _striping_info *si)
 404 {
 405         unsigned stripe_unit = ios->layout->stripe_unit;
 406         unsigned mirrors_p1 = ios->layout->mirrors_p1;
 407         unsigned devs_in_group = ios->layout->group_width * mirrors_p1;
 408         unsigned dev = si->dev;
 409         unsigned first_dev = dev - (dev % devs_in_group);
 410         unsigned max_comp = ios->numdevs ? ios->numdevs - mirrors_p1 : 0;
 411         unsigned cur_pg = ios->pages_consumed;
 412         int ret = 0;
 413
 414         while (length) {
 415                 struct exofs_per_dev_state *per_dev = &ios->per_dev[dev];
 416                 unsigned cur_len, page_off = 0;
 417
 418                 if (!per_dev->length) {
 419                         per_dev->dev = dev;
 420                         if (dev < si->dev) {
 421                                 per_dev->offset = si->obj_offset + stripe_unit -
 422                                                                    si->unit_off;
 423                                 cur_len = stripe_unit;
 424                         } else if (dev == si->dev) {
 425                                 per_dev->offset = si->obj_offset;
 426                                 cur_len = stripe_unit - si->unit_off;
 427                                 page_off = si->unit_off & ~PAGE_MASK;
 428                                 BUG_ON(page_off && (page_off != ios->pgbase));
 429                         } else { /* dev > si->dev */
 430                                 per_dev->offset = si->obj_offset - si->unit_off;
 431                                 cur_len = stripe_unit;
 432                         }
 433
 434                         if (max_comp < dev)
 435                                 max_comp = dev;
 436                 } else {
 437                         cur_len = stripe_unit;
 438                 }
 439                 if (cur_len >= length)
 440                         cur_len = length;
 441
 442                 ret = _add_stripe_unit(ios, &cur_pg, page_off , per_dev,
 443                                        cur_len);
 444                 if (unlikely(ret))
 445                         goto out;
 446
 447                 dev += mirrors_p1;
 448                 dev = (dev % devs_in_group) + first_dev;
 449
 450                 length -= cur_len;
 451         }
 452 out:
 453         ios->numdevs = max_comp + mirrors_p1;
 454         ios->pages_consumed = cur_pg;
 455         return ret;
 456 }
 457
 458 static int _prepare_for_striping(struct exofs_io_state *ios)
 459 {
 460         u64 length = ios->length;
 461         u64 offset = ios->offset;
 462         struct _striping_info si;
 463         int ret = 0;
 464
 465         if (!ios->pages) {
 466                 if (ios->kern_buff) {
 467                         struct exofs_per_dev_state *per_dev = &ios->per_dev[0];
 468
 469                         _calc_stripe_info(ios->layout, ios->offset, &si);
 470                         per_dev->offset = si.obj_offset;
 471                         per_dev->dev = si.dev;
 472
 473                         /* no cross device without page array */
 474                         BUG_ON((ios->layout->group_width > 1) &&
 475                                (si.unit_off + ios->length >
 476                                 ios->layout->stripe_unit));
 477                 }
 478                 ios->numdevs = ios->layout->mirrors_p1;
 479                 return 0;
 480         }
 481
 482         while (length) {
 483                 _calc_stripe_info(ios->layout, offset, &si);
 484
 485                 if (length < si.group_length)
 486                         si.group_length = length;
 487
 488                 ret = _prepare_one_group(ios, si.group_length, &si);
 489                 if (unlikely(ret))
 490                         goto out;
 491
 492                 offset += si.group_length;
 493                 length -= si.group_length;
 494         }
 495
 496 out:
 497         return ret;
 498 }
 499
 500 int exofs_sbi_create(struct exofs_io_state *ios)
 501 {
 502         int i, ret;
 503
 504         for (i = 0; i < ios->layout->s_numdevs; i++) {
 505                 struct osd_request *or;
 506
 507                 or = osd_start_request(exofs_ios_od(ios, i), GFP_KERNEL);
 508                 if (unlikely(!or)) {
 509                         EXOFS_ERR("%s: osd_start_request failed\n", __func__);
 510                         ret = -ENOMEM;
 511                         goto out;
 512                 }
 513                 ios->per_dev[i].or = or;
 514                 ios->numdevs++;
 515
 516                 osd_req_create_object(or, &ios->obj);
 517         }
 518         ret = exofs_io_execute(ios);
 519
 520 out:
 521         return ret;
 522 }
 523
 524 int exofs_sbi_remove(struct exofs_io_state *ios)
 525 {
 526         int i, ret;
 527
 528         for (i = 0; i < ios->layout->s_numdevs; i++) {
 529                 struct osd_request *or;
 530
 531                 or = osd_start_request(exofs_ios_od(ios, i), GFP_KERNEL);
 532                 if (unlikely(!or)) {
 533                         EXOFS_ERR("%s: osd_start_request failed\n", __func__);
 534                         ret = -ENOMEM;
 535                         goto out;
 536                 }
 537                 ios->per_dev[i].or = or;
 538                 ios->numdevs++;
 539
 540                 osd_req_remove_object(or, &ios->obj);
 541         }
 542         ret = exofs_io_execute(ios);
 543
 544 out:
 545         return ret;
 546 }
 547
 548 static int _sbi_write_mirror(struct exofs_io_state *ios, int cur_comp)
 549 {
 550         struct exofs_per_dev_state *master_dev = &ios->per_dev[cur_comp];
 551         unsigned dev = ios->per_dev[cur_comp].dev;
 552         unsigned last_comp = cur_comp + ios->layout->mirrors_p1;
 553         int ret = 0;
 554
 555         if (ios->pages && !master_dev->length)
 556                 return 0; /* Just an empty slot */
 557
 558         for (; cur_comp < last_comp; ++cur_comp, ++dev) {
 559                 struct exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp];
 560                 struct osd_request *or;
 561
 562                 or = osd_start_request(exofs_ios_od(ios, dev), GFP_KERNEL);
 563                 if (unlikely(!or)) {
 564                         EXOFS_ERR("%s: osd_start_request failed\n", __func__);
 565                         ret = -ENOMEM;
 566                         goto out;
 567                 }
 568                 per_dev->or = or;
 569                 per_dev->offset = master_dev->offset;
 570
 571                 if (ios->pages) {
 572                         struct bio *bio;
 573
 574                         if (per_dev != master_dev) {
 575                                 bio = bio_kmalloc(GFP_KERNEL,
 576                                                   master_dev->bio->bi_max_vecs);
 577                                 if (unlikely(!bio)) {
 578                                         EXOFS_DBGMSG(
 579                                               "Failed to allocate BIO size=%u\n",
 580                                               master_dev->bio->bi_max_vecs);
 581                                         ret = -ENOMEM;
 582                                         goto out;
 583                                 }
 584
 585                                 __bio_clone(bio, master_dev->bio);
 586                                 bio->bi_bdev = NULL;
 587                                 bio->bi_next = NULL;
 588                                 per_dev->length = master_dev->length;
 589                                 per_dev->bio =  bio;
 590                                 per_dev->dev = dev;
 591                         } else {
 592                                 bio = master_dev->bio;
 593                                 /* FIXME: bio_set_dir() */
 594                                 bio->bi_rw |= REQ_WRITE;
 595                         }
 596
 597                         osd_req_write(or, &ios->obj, per_dev->offset, bio,
 598                                       per_dev->length);
 599                         EXOFS_DBGMSG("write(0x%llx) offset=0x%llx "
 600                                       "length=0x%llx dev=%d\n",
 601                                      _LLU(ios->obj.id), _LLU(per_dev->offset),
 602                                      _LLU(per_dev->length), dev);
 603                 } else if (ios->kern_buff) {
 604                         ret = osd_req_write_kern(or, &ios->obj, per_dev->offset,
 605                                            ios->kern_buff, ios->length);
 606                         if (unlikely(ret))
 607                                 goto out;
 608                         EXOFS_DBGMSG2("write_kern(0x%llx) offset=0x%llx "
 609                                       "length=0x%llx dev=%d\n",
 610                                      _LLU(ios->obj.id), _LLU(per_dev->offset),
 611                                      _LLU(ios->length), dev);
 612                 } else {
 613                         osd_req_set_attributes(or, &ios->obj);
 614                         EXOFS_DBGMSG2("obj(0x%llx) set_attributes=%d dev=%d\n",
 615                                      _LLU(ios->obj.id), ios->out_attr_len, dev);
 616                 }
 617
 618                 if (ios->out_attr)
 619                         osd_req_add_set_attr_list(or, ios->out_attr,
 620                                                   ios->out_attr_len);
 621
 622                 if (ios->in_attr)
 623                         osd_req_add_get_attr_list(or, ios->in_attr,
 624                                                   ios->in_attr_len);
 625         }
 626
 627 out:
 628         return ret;
 629 }
 630
 631 int exofs_sbi_write(struct exofs_io_state *ios)
 632 {
 633         int i;
 634         int ret;
 635
 636         ret = _prepare_for_striping(ios);
 637         if (unlikely(ret))
 638                 return ret;
 639
 640         for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
 641                 ret = _sbi_write_mirror(ios, i);
 642                 if (unlikely(ret))
 643                         return ret;
 644         }
 645
 646         ret = exofs_io_execute(ios);
 647         return ret;
 648 }
 649
 650 static int _sbi_read_mirror(struct exofs_io_state *ios, unsigned cur_comp)
 651 {
 652         struct osd_request *or;
 653         struct exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp];
 654         unsigned first_dev = (unsigned)ios->obj.id;
 655
 656         if (ios->pages && !per_dev->length)
 657                 return 0; /* Just an empty slot */
 658
 659         first_dev = per_dev->dev + first_dev % ios->layout->mirrors_p1;
 660         or = osd_start_request(exofs_ios_od(ios, first_dev), GFP_KERNEL);
 661         if (unlikely(!or)) {
 662                 EXOFS_ERR("%s: osd_start_request failed\n", __func__);
 663                 return -ENOMEM;
 664         }
 665         per_dev->or = or;
 666
 667         if (ios->pages) {
 668                 osd_req_read(or, &ios->obj, per_dev->offset,
 669                                 per_dev->bio, per_dev->length);
 670                 EXOFS_DBGMSG("read(0x%llx) offset=0x%llx length=0x%llx"
 671                              " dev=%d\n", _LLU(ios->obj.id),
 672                              _LLU(per_dev->offset), _LLU(per_dev->length),
 673                              first_dev);
 674         } else if (ios->kern_buff) {
 675                 int ret = osd_req_read_kern(or, &ios->obj, per_dev->offset,
 676                                             ios->kern_buff, ios->length);
 677                 EXOFS_DBGMSG2("read_kern(0x%llx) offset=0x%llx "
 678                               "length=0x%llx dev=%d ret=>%d\n",
 679                               _LLU(ios->obj.id), _LLU(per_dev->offset),
 680                               _LLU(ios->length), first_dev, ret);
 681                 if (unlikely(ret))
 682                         return ret;
 683         } else {
 684                 osd_req_get_attributes(or, &ios->obj);
 685                 EXOFS_DBGMSG2("obj(0x%llx) get_attributes=%d dev=%d\n",
 686                               _LLU(ios->obj.id), ios->in_attr_len, first_dev);
 687         }
 688         if (ios->out_attr)
 689                 osd_req_add_set_attr_list(or, ios->out_attr, ios->out_attr_len);
 690
 691         if (ios->in_attr)
 692                 osd_req_add_get_attr_list(or, ios->in_attr, ios->in_attr_len);
 693
 694         return 0;
 695 }
 696
 697 int exofs_sbi_read(struct exofs_io_state *ios)
 698 {
 699         int i;
 700         int ret;
 701
 702         ret = _prepare_for_striping(ios);
 703         if (unlikely(ret))
 704                 return ret;
 705
 706         for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
 707                 ret = _sbi_read_mirror(ios, i);
 708                 if (unlikely(ret))
 709                         return ret;
 710         }
 711
 712         ret = exofs_io_execute(ios);
 713         return ret;
 714 }
 715
 716 int extract_attr_from_ios(struct exofs_io_state *ios, struct osd_attr *attr)
 717 {
 718         struct osd_attr cur_attr = {.attr_page = 0}; /* start with zeros */
 719         void *iter = NULL;
 720         int nelem;
 721
 722         do {
 723                 nelem = 1;
 724                 osd_req_decode_get_attr_list(ios->per_dev[0].or,
 725                                              &cur_attr, &nelem, &iter);
 726                 if ((cur_attr.attr_page == attr->attr_page) &&
 727                     (cur_attr.attr_id == attr->attr_id)) {
 728                         attr->len = cur_attr.len;
 729                         attr->val_ptr = cur_attr.val_ptr;
 730                         return 0;
 731                 }
 732         } while (iter);
 733
 734         return -EIO;
 735 }
 736
 737 static int _truncate_mirrors(struct exofs_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 exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp];
 744                 struct osd_request *or;
 745
 746                 or = osd_start_request(exofs_ios_od(ios, cur_comp), GFP_KERNEL);
 747                 if (unlikely(!or)) {
 748                         EXOFS_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);
 754                 osd_req_add_set_attr_list(or, attr, 1);
 755         }
 756
 757         return 0;
 758 }
 759
 760 struct _trunc_info {
 761         struct _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         unsigned max_devs;
 768 };
 769
 770 void _calc_trunk_info(struct exofs_layout *layout, u64 file_offset,
 771                        struct _trunc_info *ti)
 772 {
 773         unsigned stripe_unit = layout->stripe_unit;
 774
 775         _calc_stripe_info(layout, file_offset, &ti->si);
 776
 777         ti->prev_group_obj_off = ti->si.M * stripe_unit;
 778         ti->next_group_obj_off = ti->si.M ? (ti->si.M - 1) * stripe_unit : 0;
 779
 780         ti->first_group_dev = ti->si.dev - (ti->si.dev % layout->group_width);
 781         ti->nex_group_dev = ti->first_group_dev + layout->group_width;
 782         ti->max_devs = layout->group_width * layout->group_count;
 783 }
 784
 785 int exofs_oi_truncate(struct exofs_i_info *oi, u64 size)
 786 {
 787         struct exofs_sb_info *sbi = oi->vfs_inode.i_sb->s_fs_info;
 788         struct exofs_io_state *ios;
 789         struct exofs_trunc_attr {
 790                 struct osd_attr attr;
 791                 __be64 newsize;
 792         } *size_attrs;
 793         struct _trunc_info ti;
 794         int i, ret;
 795
 796         ret = exofs_get_io_state(&sbi->layout, &ios);
 797         if (unlikely(ret))
 798                 return ret;
 799
 800         _calc_trunk_info(ios->layout, size, &ti);
 801
 802         size_attrs = kcalloc(ti.max_devs, sizeof(*size_attrs),
 803                              GFP_KERNEL);
 804         if (unlikely(!size_attrs)) {
 805                 ret = -ENOMEM;
 806                 goto out;
 807         }
 808
 809         ios->obj.id = exofs_oi_objno(oi);
 810         ios->cred = oi->i_cred;
 811         ios->numdevs = ios->layout->s_numdevs;
 812
 813         for (i = 0; i < ti.max_devs; ++i) {
 814                 struct exofs_trunc_attr *size_attr = &size_attrs[i];
 815                 u64 obj_size;
 816
 817                 if (i < ti.first_group_dev)
 818                         obj_size = ti.prev_group_obj_off;
 819                 else if (i >= ti.nex_group_dev)
 820                         obj_size = ti.next_group_obj_off;
 821                 else if (i < ti.si.dev) /* dev within this group */
 822                         obj_size = ti.si.obj_offset +
 823                                       ios->layout->stripe_unit - ti.si.unit_off;
 824                 else if (i == ti.si.dev)
 825                         obj_size = ti.si.obj_offset;
 826                 else /* i > ti.dev */
 827                         obj_size = ti.si.obj_offset - ti.si.unit_off;
 828
 829                 size_attr->newsize = cpu_to_be64(obj_size);
 830                 size_attr->attr = g_attr_logical_length;
 831                 size_attr->attr.val_ptr = &size_attr->newsize;
 832
 833                 EXOFS_DBGMSG("trunc(0x%llx) obj_offset=0x%llx dev=%d\n",
 834                              _LLU(ios->obj.id), _LLU(obj_size), i);
 835                 ret = _truncate_mirrors(ios, i * ios->layout->mirrors_p1,
 836                                         &size_attr->attr);
 837                 if (unlikely(ret))
 838                         goto out;
 839         }
 840         ret = exofs_io_execute(ios);
 841
 842 out:
 843         kfree(size_attrs);
 844         exofs_put_io_state(ios);
 845         return ret;
 846 }