fs/exofs/ore.c

   1 /*
   2  * Copyright (C) 2005, 2006
   3  * Avishay Traeger (avishay@gmail.com)
   4  * Copyright (C) 2008, 2009
   5  * Boaz Harrosh <ooo@electrozaur.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 <linux/module.h>
  27 #include <asm/div64.h>
  28 #include <linux/lcm.h>
  29
  30 #include "ore_raid.h"
  31
  32 MODULE_AUTHOR("Boaz Harrosh <ooo@electrozaur.com>");
  33 MODULE_DESCRIPTION("Objects Raid Engine ore.ko");
  34 MODULE_LICENSE("GPL");
  35
  36 /* ore_verify_layout does a couple of things:
  37  * 1. Given a minimum number of needed parameters fixes up the rest of the
  38  *    members to be operatonals for the ore. The needed parameters are those
  39  *    that are defined by the pnfs-objects layout STD.
  40  * 2. Check to see if the current ore code actually supports these parameters
  41  *    for example stripe_unit must be a multple of the system PAGE_SIZE,
  42  *    and etc...
  43  * 3. Cache some havily used calculations that will be needed by users.
  44  */
  45
  46 enum { BIO_MAX_PAGES_KMALLOC =
  47                 (PAGE_SIZE - sizeof(struct bio)) / sizeof(struct bio_vec),};
  48
  49 int ore_verify_layout(unsigned total_comps, struct ore_layout *layout)
  50 {
  51         u64 stripe_length;
  52
  53         switch (layout->raid_algorithm) {
  54         case PNFS_OSD_RAID_0:
  55                 layout->parity = 0;
  56                 break;
  57         case PNFS_OSD_RAID_5:
  58                 layout->parity = 1;
  59                 break;
  60         case PNFS_OSD_RAID_PQ:
  61                 layout->parity = 2;
  62                 break;
  63         case PNFS_OSD_RAID_4:
  64         default:
  65                 ORE_ERR("Only RAID_0/5/6 for now received-enum=%d\n",
  66                         layout->raid_algorithm);
  67                 return -EINVAL;
  68         }
  69         if (0 != (layout->stripe_unit & ~PAGE_MASK)) {
  70                 ORE_ERR("Stripe Unit(0x%llx)"
  71                           " must be Multples of PAGE_SIZE(0x%lx)\n",
  72                           _LLU(layout->stripe_unit), PAGE_SIZE);
  73                 return -EINVAL;
  74         }
  75         if (layout->group_width) {
  76                 if (!layout->group_depth) {
  77                         ORE_ERR("group_depth == 0 && group_width != 0\n");
  78                         return -EINVAL;
  79                 }
  80                 if (total_comps < (layout->group_width * layout->mirrors_p1)) {
  81                         ORE_ERR("Data Map wrong, "
  82                                 "numdevs=%d < group_width=%d * mirrors=%d\n",
  83                                 total_comps, layout->group_width,
  84                                 layout->mirrors_p1);
  85                         return -EINVAL;
  86                 }
  87                 layout->group_count = total_comps / layout->mirrors_p1 /
  88                                                 layout->group_width;
  89         } else {
  90                 if (layout->group_depth) {
  91                         printk(KERN_NOTICE "Warning: group_depth ignored "
  92                                 "group_width == 0 && group_depth == %lld\n",
  93                                 _LLU(layout->group_depth));
  94                 }
  95                 layout->group_width = total_comps / layout->mirrors_p1;
  96                 layout->group_depth = -1;
  97                 layout->group_count = 1;
  98         }
  99
 100         stripe_length = (u64)layout->group_width * layout->stripe_unit;
 101         if (stripe_length >= (1ULL << 32)) {
 102                 ORE_ERR("Stripe_length(0x%llx) >= 32bit is not supported\n",
 103                         _LLU(stripe_length));
 104                 return -EINVAL;
 105         }
 106
 107         layout->max_io_length =
 108                 (BIO_MAX_PAGES_KMALLOC * PAGE_SIZE - layout->stripe_unit) *
 109                                         (layout->group_width - layout->parity);
 110         if (layout->parity) {
 111                 unsigned stripe_length =
 112                                 (layout->group_width - layout->parity) *
 113                                 layout->stripe_unit;
 114
 115                 layout->max_io_length /= stripe_length;
 116                 layout->max_io_length *= stripe_length;
 117         }
 118         ORE_DBGMSG("max_io_length=0x%lx\n", layout->max_io_length);
 119
 120         return 0;
 121 }
 122 EXPORT_SYMBOL(ore_verify_layout);
 123
 124 static u8 *_ios_cred(struct ore_io_state *ios, unsigned index)
 125 {
 126         return ios->oc->comps[index & ios->oc->single_comp].cred;
 127 }
 128
 129 static struct osd_obj_id *_ios_obj(struct ore_io_state *ios, unsigned index)
 130 {
 131         return &ios->oc->comps[index & ios->oc->single_comp].obj;
 132 }
 133
 134 static struct osd_dev *_ios_od(struct ore_io_state *ios, unsigned index)
 135 {
 136         ORE_DBGMSG2("oc->first_dev=%d oc->numdevs=%d i=%d oc->ods=%p\n",
 137                     ios->oc->first_dev, ios->oc->numdevs, index,
 138                     ios->oc->ods);
 139
 140         return ore_comp_dev(ios->oc, index);
 141 }
 142
 143 int  _ore_get_io_state(struct ore_layout *layout,
 144                         struct ore_components *oc, unsigned numdevs,
 145                         unsigned sgs_per_dev, unsigned num_par_pages,
 146                         struct ore_io_state **pios)
 147 {
 148         struct ore_io_state *ios;
 149         size_t size_ios, size_extra, size_total;
 150         void *ios_extra;
 151
 152         /*
 153          * The desired layout looks like this, with the extra_allocation
 154          * items pointed at from fields within ios or per_dev:
 155
 156         struct __alloc_all_io_state {
 157                 struct ore_io_state ios;
 158                 struct ore_per_dev_state per_dev[numdevs];
 159                 union {
 160                         struct osd_sg_entry sglist[sgs_per_dev * numdevs];
 161                         struct page *pages[num_par_pages];
 162                 } extra_allocation;
 163         } whole_allocation;
 164
 165         */
 166
 167         /* This should never happen, so abort early if it ever does. */
 168         if (sgs_per_dev && num_par_pages) {
 169                 ORE_DBGMSG("Tried to use both pages and sglist\n");
 170                 *pios = NULL;
 171                 return -EINVAL;
 172         }
 173
 174         if (numdevs > (INT_MAX - sizeof(*ios)) /
 175                        sizeof(struct ore_per_dev_state))
 176                 return -ENOMEM;
 177         size_ios = sizeof(*ios) + sizeof(struct ore_per_dev_state) * numdevs;
 178
 179         if (sgs_per_dev * numdevs > INT_MAX / sizeof(struct osd_sg_entry))
 180                 return -ENOMEM;
 181         if (num_par_pages > INT_MAX / sizeof(struct page *))
 182                 return -ENOMEM;
 183         size_extra = max(sizeof(struct osd_sg_entry) * (sgs_per_dev * numdevs),
 184                          sizeof(struct page *) * num_par_pages);
 185
 186         size_total = size_ios + size_extra;
 187
 188         if (likely(size_total <= PAGE_SIZE)) {
 189                 ios = kzalloc(size_total, GFP_KERNEL);
 190                 if (unlikely(!ios)) {
 191                         ORE_DBGMSG("Failed kzalloc bytes=%zd\n", size_total);
 192                         *pios = NULL;
 193                         return -ENOMEM;
 194                 }
 195                 ios_extra = (char *)ios + size_ios;
 196         } else {
 197                 ios = kzalloc(size_ios, GFP_KERNEL);
 198                 if (unlikely(!ios)) {
 199                         ORE_DBGMSG("Failed alloc first part bytes=%zd\n",
 200                                    size_ios);
 201                         *pios = NULL;
 202                         return -ENOMEM;
 203                 }
 204                 ios_extra = kzalloc(size_extra, GFP_KERNEL);
 205                 if (unlikely(!ios_extra)) {
 206                         ORE_DBGMSG("Failed alloc second part bytes=%zd\n",
 207                                    size_extra);
 208                         kfree(ios);
 209                         *pios = NULL;
 210                         return -ENOMEM;
 211                 }
 212
 213                 /* In this case the per_dev[0].sgilist holds the pointer to
 214                  * be freed
 215                  */
 216                 ios->extra_part_alloc = true;
 217         }
 218
 219         if (num_par_pages) {
 220                 ios->parity_pages = ios_extra;
 221                 ios->max_par_pages = num_par_pages;
 222         }
 223         if (sgs_per_dev) {
 224                 struct osd_sg_entry *sgilist = ios_extra;
 225                 unsigned d;
 226
 227                 for (d = 0; d < numdevs; ++d) {
 228                         ios->per_dev[d].sglist = sgilist;
 229                         sgilist += sgs_per_dev;
 230                 }
 231                 ios->sgs_per_dev = sgs_per_dev;
 232         }
 233
 234         ios->layout = layout;
 235         ios->oc = oc;
 236         *pios = ios;
 237         return 0;
 238 }
 239
 240 /* Allocate an io_state for only a single group of devices
 241  *
 242  * If a user needs to call ore_read/write() this version must be used becase it
 243  * allocates extra stuff for striping and raid.
 244  * The ore might decide to only IO less then @length bytes do to alignmets
 245  * and constrains as follows:
 246  * - The IO cannot cross group boundary.
 247  * - In raid5/6 The end of the IO must align at end of a stripe eg.
 248  *   (@offset + @length) % strip_size == 0. Or the complete range is within a
 249  *   single stripe.
 250  * - Memory condition only permitted a shorter IO. (A user can use @length=~0
 251  *   And check the returned ios->length for max_io_size.)
 252  *
 253  * The caller must check returned ios->length (and/or ios->nr_pages) and
 254  * re-issue these pages that fall outside of ios->length
 255  */
 256 int  ore_get_rw_state(struct ore_layout *layout, struct ore_components *oc,
 257                       bool is_reading, u64 offset, u64 length,
 258                       struct ore_io_state **pios)
 259 {
 260         struct ore_io_state *ios;
 261         unsigned numdevs = layout->group_width * layout->mirrors_p1;
 262         unsigned sgs_per_dev = 0, max_par_pages = 0;
 263         int ret;
 264
 265         if (layout->parity && length) {
 266                 unsigned data_devs = layout->group_width - layout->parity;
 267                 unsigned stripe_size = layout->stripe_unit * data_devs;
 268                 unsigned pages_in_unit = layout->stripe_unit / PAGE_SIZE;
 269                 u32 remainder;
 270                 u64 num_stripes;
 271                 u64 num_raid_units;
 272
 273                 num_stripes = div_u64_rem(length, stripe_size, &remainder);
 274                 if (remainder)
 275                         ++num_stripes;
 276
 277                 num_raid_units =  num_stripes * layout->parity;
 278
 279                 if (is_reading) {
 280                         /* For reads add per_dev sglist array */
 281                         /* TODO: Raid 6 we need twice more. Actually:
 282                         *         num_stripes / LCMdP(W,P);
 283                         *         if (W%P != 0) num_stripes *= parity;
 284                         */
 285
 286                         /* first/last seg is split */
 287                         num_raid_units += layout->group_width;
 288                         sgs_per_dev = div_u64(num_raid_units, data_devs) + 2;
 289                 } else {
 290                         /* For Writes add parity pages array. */
 291                         max_par_pages = num_raid_units * pages_in_unit *
 292                                                 sizeof(struct page *);
 293                 }
 294         }
 295
 296         ret = _ore_get_io_state(layout, oc, numdevs, sgs_per_dev, max_par_pages,
 297                                 pios);
 298         if (unlikely(ret))
 299                 return ret;
 300
 301         ios = *pios;
 302         ios->reading = is_reading;
 303         ios->offset = offset;
 304
 305         if (length) {
 306                 ore_calc_stripe_info(layout, offset, length, &ios->si);
 307                 ios->length = ios->si.length;
 308                 ios->nr_pages = ((ios->offset & (PAGE_SIZE - 1)) +
 309                                  ios->length + PAGE_SIZE - 1) / PAGE_SIZE;
 310                 if (layout->parity)
 311                         _ore_post_alloc_raid_stuff(ios);
 312         }
 313
 314         return 0;
 315 }
 316 EXPORT_SYMBOL(ore_get_rw_state);
 317
 318 /* Allocate an io_state for all the devices in the comps array
 319  *
 320  * This version of io_state allocation is used mostly by create/remove
 321  * and trunc where we currently need all the devices. The only wastful
 322  * bit is the read/write_attributes with no IO. Those sites should
 323  * be converted to use ore_get_rw_state() with length=0
 324  */
 325 int  ore_get_io_state(struct ore_layout *layout, struct ore_components *oc,
 326                       struct ore_io_state **pios)
 327 {
 328         return _ore_get_io_state(layout, oc, oc->numdevs, 0, 0, pios);
 329 }
 330 EXPORT_SYMBOL(ore_get_io_state);
 331
 332 void ore_put_io_state(struct ore_io_state *ios)
 333 {
 334         if (ios) {
 335                 unsigned i;
 336
 337                 for (i = 0; i < ios->numdevs; i++) {
 338                         struct ore_per_dev_state *per_dev = &ios->per_dev[i];
 339
 340                         if (per_dev->or)
 341                                 osd_end_request(per_dev->or);
 342                         if (per_dev->bio)
 343                                 bio_put(per_dev->bio);
 344                 }
 345
 346                 _ore_free_raid_stuff(ios);
 347                 kfree(ios);
 348         }
 349 }
 350 EXPORT_SYMBOL(ore_put_io_state);
 351
 352 static void _sync_done(struct ore_io_state *ios, void *p)
 353 {
 354         struct completion *waiting = p;
 355
 356         complete(waiting);
 357 }
 358
 359 static void _last_io(struct kref *kref)
 360 {
 361         struct ore_io_state *ios = container_of(
 362                                         kref, struct ore_io_state, kref);
 363
 364         ios->done(ios, ios->private);
 365 }
 366
 367 static void _done_io(struct osd_request *or, void *p)
 368 {
 369         struct ore_io_state *ios = p;
 370
 371         kref_put(&ios->kref, _last_io);
 372 }
 373
 374 int ore_io_execute(struct ore_io_state *ios)
 375 {
 376         DECLARE_COMPLETION_ONSTACK(wait);
 377         bool sync = (ios->done == NULL);
 378         int i, ret;
 379
 380         if (sync) {
 381                 ios->done = _sync_done;
 382                 ios->private = &wait;
 383         }
 384
 385         for (i = 0; i < ios->numdevs; i++) {
 386                 struct osd_request *or = ios->per_dev[i].or;
 387                 if (unlikely(!or))
 388                         continue;
 389
 390                 ret = osd_finalize_request(or, 0, _ios_cred(ios, i), NULL);
 391                 if (unlikely(ret)) {
 392                         ORE_DBGMSG("Failed to osd_finalize_request() => %d\n",
 393                                      ret);
 394                         return ret;
 395                 }
 396         }
 397
 398         kref_init(&ios->kref);
 399
 400         for (i = 0; i < ios->numdevs; i++) {
 401                 struct osd_request *or = ios->per_dev[i].or;
 402                 if (unlikely(!or))
 403                         continue;
 404
 405                 kref_get(&ios->kref);
 406                 osd_execute_request_async(or, _done_io, ios);
 407         }
 408
 409         kref_put(&ios->kref, _last_io);
 410         ret = 0;
 411
 412         if (sync) {
 413                 wait_for_completion(&wait);
 414                 ret = ore_check_io(ios, NULL);
 415         }
 416         return ret;
 417 }
 418
 419 static void _clear_bio(struct bio *bio)
 420 {
 421         struct bio_vec *bv;
 422         unsigned i;
 423
 424         bio_for_each_segment_all(bv, bio, i) {
 425                 unsigned this_count = bv->bv_len;
 426
 427                 if (likely(PAGE_SIZE == this_count))
 428                         clear_highpage(bv->bv_page);
 429                 else
 430                         zero_user(bv->bv_page, bv->bv_offset, this_count);
 431         }
 432 }
 433
 434 int ore_check_io(struct ore_io_state *ios, ore_on_dev_error on_dev_error)
 435 {
 436         enum osd_err_priority acumulated_osd_err = 0;
 437         int acumulated_lin_err = 0;
 438         int i;
 439
 440         for (i = 0; i < ios->numdevs; i++) {
 441                 struct osd_sense_info osi;
 442                 struct ore_per_dev_state *per_dev = &ios->per_dev[i];
 443                 struct osd_request *or = per_dev->or;
 444                 int ret;
 445
 446                 if (unlikely(!or))
 447                         continue;
 448
 449                 ret = osd_req_decode_sense(or, &osi);
 450                 if (likely(!ret))
 451                         continue;
 452
 453                 if ((OSD_ERR_PRI_CLEAR_PAGES == osi.osd_err_pri) &&
 454                     per_dev->bio) {
 455                         /* start read offset passed endof file.
 456                          * Note: if we do not have bio it means read-attributes
 457                          * In this case we should return error to caller.
 458                          */
 459                         _clear_bio(per_dev->bio);
 460                         ORE_DBGMSG("start read offset passed end of file "
 461                                 "offset=0x%llx, length=0x%llx\n",
 462                                 _LLU(per_dev->offset),
 463                                 _LLU(per_dev->length));
 464
 465                         continue; /* we recovered */
 466                 }
 467
 468                 if (on_dev_error) {
 469                         u64 residual = ios->reading ?
 470                                         or->in.residual : or->out.residual;
 471                         u64 offset = (ios->offset + ios->length) - residual;
 472                         unsigned dev = per_dev->dev - ios->oc->first_dev;
 473                         struct ore_dev *od = ios->oc->ods[dev];
 474
 475                         on_dev_error(ios, od, dev, osi.osd_err_pri,
 476                                      offset, residual);
 477                 }
 478                 if (osi.osd_err_pri >= acumulated_osd_err) {
 479                         acumulated_osd_err = osi.osd_err_pri;
 480                         acumulated_lin_err = ret;
 481                 }
 482         }
 483
 484         return acumulated_lin_err;
 485 }
 486 EXPORT_SYMBOL(ore_check_io);
 487
 488 /*
 489  * L - logical offset into the file
 490  *
 491  * D - number of Data devices
 492  *      D = group_width - parity
 493  *
 494  * U - The number of bytes in a stripe within a group
 495  *      U =  stripe_unit * D
 496  *
 497  * T - The number of bytes striped within a group of component objects
 498  *     (before advancing to the next group)
 499  *      T = U * group_depth
 500  *
 501  * S - The number of bytes striped across all component objects
 502  *     before the pattern repeats
 503  *      S = T * group_count
 504  *
 505  * M - The "major" (i.e., across all components) cycle number
 506  *      M = L / S
 507  *
 508  * G - Counts the groups from the beginning of the major cycle
 509  *      G = (L - (M * S)) / T   [or (L % S) / T]
 510  *
 511  * H - The byte offset within the group
 512  *      H = (L - (M * S)) % T   [or (L % S) % T]
 513  *
 514  * N - The "minor" (i.e., across the group) stripe number
 515  *      N = H / U
 516  *
 517  * C - The component index coresponding to L
 518  *
 519  *      C = (H - (N * U)) / stripe_unit + G * D
 520  *      [or (L % U) / stripe_unit + G * D]
 521  *
 522  * O - The component offset coresponding to L
 523  *      O = L % stripe_unit + N * stripe_unit + M * group_depth * stripe_unit
 524  *
 525  * LCMdP – Parity cycle: Lowest Common Multiple of group_width, parity
 526  *          divide by parity
 527  *      LCMdP = lcm(group_width, parity) / parity
 528  *
 529  * R - The parity Rotation stripe
 530  *     (Note parity cycle always starts at a group's boundary)
 531  *      R = N % LCMdP
 532  *
 533  * I = the first parity device index
 534  *      I = (group_width + group_width - R*parity - parity) % group_width
 535  *
 536  * Craid - The component index Rotated
 537  *      Craid = (group_width + C - R*parity) % group_width
 538  *      (We add the group_width to avoid negative numbers modulo math)
 539  */
 540 void ore_calc_stripe_info(struct ore_layout *layout, u64 file_offset,
 541                           u64 length, struct ore_striping_info *si)
 542 {
 543         u32     stripe_unit = layout->stripe_unit;
 544         u32     group_width = layout->group_width;
 545         u64     group_depth = layout->group_depth;
 546         u32     parity      = layout->parity;
 547
 548         u32     D = group_width - parity;
 549         u32     U = D * stripe_unit;
 550         u64     T = U * group_depth;
 551         u64     S = T * layout->group_count;
 552         u64     M = div64_u64(file_offset, S);
 553
 554         /*
 555         G = (L - (M * S)) / T
 556         H = (L - (M * S)) % T
 557         */
 558         u64     LmodS = file_offset - M * S;
 559         u32     G = div64_u64(LmodS, T);
 560         u64     H = LmodS - G * T;
 561
 562         u32     N = div_u64(H, U);
 563         u32     Nlast;
 564
 565         /* "H - (N * U)" is just "H % U" so it's bound to u32 */
 566         u32     C = (u32)(H - (N * U)) / stripe_unit + G * group_width;
 567         u32 first_dev = C - C % group_width;
 568
 569         div_u64_rem(file_offset, stripe_unit, &si->unit_off);
 570
 571         si->obj_offset = si->unit_off + (N * stripe_unit) +
 572                                   (M * group_depth * stripe_unit);
 573         si->cur_comp = C - first_dev;
 574         si->cur_pg = si->unit_off / PAGE_SIZE;
 575
 576         if (parity) {
 577                 u32 LCMdP = lcm(group_width, parity) / parity;
 578                 /* R     = N % LCMdP; */
 579                 u32 RxP   = (N % LCMdP) * parity;
 580
 581                 si->par_dev = (group_width + group_width - parity - RxP) %
 582                               group_width + first_dev;
 583                 si->dev = (group_width + group_width + C - RxP) %
 584                           group_width + first_dev;
 585                 si->bytes_in_stripe = U;
 586                 si->first_stripe_start = M * S + G * T + N * U;
 587         } else {
 588                 /* Make the math correct see _prepare_one_group */
 589                 si->par_dev = group_width;
 590                 si->dev = C;
 591         }
 592
 593         si->dev *= layout->mirrors_p1;
 594         si->par_dev *= layout->mirrors_p1;
 595         si->offset = file_offset;
 596         si->length = T - H;
 597         if (si->length > length)
 598                 si->length = length;
 599
 600         Nlast = div_u64(H + si->length + U - 1, U);
 601         si->maxdevUnits = Nlast - N;
 602
 603         si->M = M;
 604 }
 605 EXPORT_SYMBOL(ore_calc_stripe_info);
 606
 607 int _ore_add_stripe_unit(struct ore_io_state *ios,  unsigned *cur_pg,
 608                          unsigned pgbase, struct page **pages,
 609                          struct ore_per_dev_state *per_dev, int cur_len)
 610 {
 611         unsigned pg = *cur_pg;
 612         struct request_queue *q =
 613                         osd_request_queue(_ios_od(ios, per_dev->dev));
 614         unsigned len = cur_len;
 615         int ret;
 616
 617         if (per_dev->bio == NULL) {
 618                 unsigned bio_size;
 619
 620                 if (!ios->reading) {
 621                         bio_size = ios->si.maxdevUnits;
 622                 } else {
 623                         bio_size = (ios->si.maxdevUnits + 1) *
 624                              (ios->layout->group_width - ios->layout->parity) /
 625                              ios->layout->group_width;
 626                 }
 627                 bio_size *= (ios->layout->stripe_unit / PAGE_SIZE);
 628
 629                 per_dev->bio = bio_kmalloc(GFP_KERNEL, bio_size);
 630                 if (unlikely(!per_dev->bio)) {
 631                         ORE_DBGMSG("Failed to allocate BIO size=%u\n",
 632                                      bio_size);
 633                         ret = -ENOMEM;
 634                         goto out;
 635                 }
 636         }
 637
 638         while (cur_len > 0) {
 639                 unsigned pglen = min_t(unsigned, PAGE_SIZE - pgbase, cur_len);
 640                 unsigned added_len;
 641
 642                 cur_len -= pglen;
 643
 644                 added_len = bio_add_pc_page(q, per_dev->bio, pages[pg],
 645                                             pglen, pgbase);
 646                 if (unlikely(pglen != added_len)) {
 647                         /* If bi_vcnt == bi_max then this is a SW BUG */
 648                         ORE_DBGMSG("Failed bio_add_pc_page bi_vcnt=0x%x "
 649                                    "bi_max=0x%x BIO_MAX=0x%x cur_len=0x%x\n",
 650                                    per_dev->bio->bi_vcnt,
 651                                    per_dev->bio->bi_max_vecs,
 652                                    BIO_MAX_PAGES_KMALLOC, cur_len);
 653                         ret = -ENOMEM;
 654                         goto out;
 655                 }
 656                 _add_stripe_page(ios->sp2d, &ios->si, pages[pg]);
 657
 658                 pgbase = 0;
 659                 ++pg;
 660         }
 661         BUG_ON(cur_len);
 662
 663         per_dev->length += len;
 664         *cur_pg = pg;
 665         ret = 0;
 666 out:    /* we fail the complete unit on an error eg don't advance
 667          * per_dev->length and cur_pg. This means that we might have a bigger
 668          * bio than the CDB requested length (per_dev->length). That's fine
 669          * only the oposite is fatal.
 670          */
 671         return ret;
 672 }
 673
 674 static int _add_parity_units(struct ore_io_state *ios,
 675                              struct ore_striping_info *si,
 676                              unsigned dev, unsigned first_dev,
 677                              unsigned mirrors_p1, unsigned devs_in_group,
 678                              unsigned cur_len)
 679 {
 680         unsigned do_parity;
 681         int ret = 0;
 682
 683         for (do_parity = ios->layout->parity; do_parity; --do_parity) {
 684                 struct ore_per_dev_state *per_dev;
 685
 686                 per_dev = &ios->per_dev[dev - first_dev];
 687                 if (!per_dev->length && !per_dev->offset) {
 688                         /* Only/always the parity unit of the first
 689                          * stripe will be empty. So this is a chance to
 690                          * initialize the per_dev info.
 691                          */
 692                         per_dev->dev = dev;
 693                         per_dev->offset = si->obj_offset - si->unit_off;
 694                 }
 695
 696                 ret = _ore_add_parity_unit(ios, si, per_dev, cur_len,
 697                                            do_parity == 1);
 698                 if (unlikely(ret))
 699                                 break;
 700
 701                 if (do_parity != 1) {
 702                         dev = ((dev + mirrors_p1) % devs_in_group) + first_dev;
 703                         si->cur_comp = (si->cur_comp + 1) %
 704                                                        ios->layout->group_width;
 705                 }
 706         }
 707
 708         return ret;
 709 }
 710
 711 static int _prepare_for_striping(struct ore_io_state *ios)
 712 {
 713         struct ore_striping_info *si = &ios->si;
 714         unsigned stripe_unit = ios->layout->stripe_unit;
 715         unsigned mirrors_p1 = ios->layout->mirrors_p1;
 716         unsigned group_width = ios->layout->group_width;
 717         unsigned devs_in_group = group_width * mirrors_p1;
 718         unsigned dev = si->dev;
 719         unsigned first_dev = dev - (dev % devs_in_group);
 720         unsigned cur_pg = ios->pages_consumed;
 721         u64 length = ios->length;
 722         int ret = 0;
 723
 724         if (!ios->pages) {
 725                 ios->numdevs = ios->layout->mirrors_p1;
 726                 return 0;
 727         }
 728
 729         BUG_ON(length > si->length);
 730
 731         while (length) {
 732                 struct ore_per_dev_state *per_dev =
 733                                                 &ios->per_dev[dev - first_dev];
 734                 unsigned cur_len, page_off = 0;
 735
 736                 if (!per_dev->length && !per_dev->offset) {
 737                         /* First time initialize the per_dev info. */
 738                         per_dev->dev = dev;
 739                         if (dev == si->dev) {
 740                                 WARN_ON(dev == si->par_dev);
 741                                 per_dev->offset = si->obj_offset;
 742                                 cur_len = stripe_unit - si->unit_off;
 743                                 page_off = si->unit_off & ~PAGE_MASK;
 744                                 BUG_ON(page_off && (page_off != ios->pgbase));
 745                         } else {
 746                                 per_dev->offset = si->obj_offset - si->unit_off;
 747                                 cur_len = stripe_unit;
 748                         }
 749                 } else {
 750                         cur_len = stripe_unit;
 751                 }
 752                 if (cur_len >= length)
 753                         cur_len = length;
 754
 755                 ret = _ore_add_stripe_unit(ios, &cur_pg, page_off, ios->pages,
 756                                            per_dev, cur_len);
 757                 if (unlikely(ret))
 758                         goto out;
 759
 760                 length -= cur_len;
 761
 762                 dev = ((dev + mirrors_p1) % devs_in_group) + first_dev;
 763                 si->cur_comp = (si->cur_comp + 1) % group_width;
 764                 if (unlikely((dev == si->par_dev) || (!length && ios->sp2d))) {
 765                         if (!length && ios->sp2d) {
 766                                 /* If we are writing and this is the very last
 767                                  * stripe. then operate on parity dev.
 768                                  */
 769                                 dev = si->par_dev;
 770                                 /* If last stripe operate on parity comp */
 771                                 si->cur_comp = group_width - ios->layout->parity;
 772                         }
 773
 774                         /* In writes cur_len just means if it's the
 775                          * last one. See _ore_add_parity_unit.
 776                          */
 777                         ret = _add_parity_units(ios, si, dev, first_dev,
 778                                                 mirrors_p1, devs_in_group,
 779                                                 ios->sp2d ? length : cur_len);
 780                         if (unlikely(ret))
 781                                         goto out;
 782
 783                         /* Rotate next par_dev backwards with wraping */
 784                         si->par_dev = (devs_in_group + si->par_dev -
 785                                        ios->layout->parity * mirrors_p1) %
 786                                       devs_in_group + first_dev;
 787                         /* Next stripe, start fresh */
 788                         si->cur_comp = 0;
 789                         si->cur_pg = 0;
 790                         si->obj_offset += cur_len;
 791                         si->unit_off = 0;
 792                 }
 793         }
 794 out:
 795         ios->numdevs = devs_in_group;
 796         ios->pages_consumed = cur_pg;
 797         return ret;
 798 }
 799
 800 int ore_create(struct ore_io_state *ios)
 801 {
 802         int i, ret;
 803
 804         for (i = 0; i < ios->oc->numdevs; i++) {
 805                 struct osd_request *or;
 806
 807                 or = osd_start_request(_ios_od(ios, i));
 808                 if (unlikely(!or)) {
 809                         ORE_ERR("%s: osd_start_request failed\n", __func__);
 810                         ret = -ENOMEM;
 811                         goto out;
 812                 }
 813                 ios->per_dev[i].or = or;
 814                 ios->numdevs++;
 815
 816                 osd_req_create_object(or, _ios_obj(ios, i));
 817         }
 818         ret = ore_io_execute(ios);
 819
 820 out:
 821         return ret;
 822 }
 823 EXPORT_SYMBOL(ore_create);
 824
 825 int ore_remove(struct ore_io_state *ios)
 826 {
 827         int i, ret;
 828
 829         for (i = 0; i < ios->oc->numdevs; i++) {
 830                 struct osd_request *or;
 831
 832                 or = osd_start_request(_ios_od(ios, i));
 833                 if (unlikely(!or)) {
 834                         ORE_ERR("%s: osd_start_request failed\n", __func__);
 835                         ret = -ENOMEM;
 836                         goto out;
 837                 }
 838                 ios->per_dev[i].or = or;
 839                 ios->numdevs++;
 840
 841                 osd_req_remove_object(or, _ios_obj(ios, i));
 842         }
 843         ret = ore_io_execute(ios);
 844
 845 out:
 846         return ret;
 847 }
 848 EXPORT_SYMBOL(ore_remove);
 849
 850 static int _write_mirror(struct ore_io_state *ios, int cur_comp)
 851 {
 852         struct ore_per_dev_state *master_dev = &ios->per_dev[cur_comp];
 853         unsigned dev = ios->per_dev[cur_comp].dev;
 854         unsigned last_comp = cur_comp + ios->layout->mirrors_p1;
 855         int ret = 0;
 856
 857         if (ios->pages && !master_dev->length)
 858                 return 0; /* Just an empty slot */
 859
 860         for (; cur_comp < last_comp; ++cur_comp, ++dev) {
 861                 struct ore_per_dev_state *per_dev = &ios->per_dev[cur_comp];
 862                 struct osd_request *or;
 863
 864                 or = osd_start_request(_ios_od(ios, dev));
 865                 if (unlikely(!or)) {
 866                         ORE_ERR("%s: osd_start_request failed\n", __func__);
 867                         ret = -ENOMEM;
 868                         goto out;
 869                 }
 870                 per_dev->or = or;
 871
 872                 if (ios->pages) {
 873                         struct bio *bio;
 874
 875                         if (per_dev != master_dev) {
 876                                 bio = bio_clone_fast(master_dev->bio,
 877                                                      GFP_KERNEL, NULL);
 878                                 if (unlikely(!bio)) {
 879                                         ORE_DBGMSG(
 880                                               "Failed to allocate BIO size=%u\n",
 881                                               master_dev->bio->bi_max_vecs);
 882                                         ret = -ENOMEM;
 883                                         goto out;
 884                                 }
 885
 886                                 bio->bi_disk = NULL;
 887                                 bio->bi_next = NULL;
 888                                 per_dev->offset = master_dev->offset;
 889                                 per_dev->length = master_dev->length;
 890                                 per_dev->bio =  bio;
 891                                 per_dev->dev = dev;
 892                         } else {
 893                                 bio = master_dev->bio;
 894                                 /* FIXME: bio_set_dir() */
 895                                 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
 896                         }
 897
 898                         osd_req_write(or, _ios_obj(ios, cur_comp),
 899                                       per_dev->offset, bio, per_dev->length);
 900                         ORE_DBGMSG("write(0x%llx) offset=0x%llx "
 901                                       "length=0x%llx dev=%d\n",
 902                                      _LLU(_ios_obj(ios, cur_comp)->id),
 903                                      _LLU(per_dev->offset),
 904                                      _LLU(per_dev->length), dev);
 905                 } else if (ios->kern_buff) {
 906                         per_dev->offset = ios->si.obj_offset;
 907                         per_dev->dev = ios->si.dev + dev;
 908
 909                         /* no cross device without page array */
 910                         BUG_ON((ios->layout->group_width > 1) &&
 911                                (ios->si.unit_off + ios->length >
 912                                 ios->layout->stripe_unit));
 913
 914                         ret = osd_req_write_kern(or, _ios_obj(ios, cur_comp),
 915                                                  per_dev->offset,
 916                                                  ios->kern_buff, ios->length);
 917                         if (unlikely(ret))
 918                                 goto out;
 919                         ORE_DBGMSG2("write_kern(0x%llx) offset=0x%llx "
 920                                       "length=0x%llx dev=%d\n",
 921                                      _LLU(_ios_obj(ios, cur_comp)->id),
 922                                      _LLU(per_dev->offset),
 923                                      _LLU(ios->length), per_dev->dev);
 924                 } else {
 925                         osd_req_set_attributes(or, _ios_obj(ios, cur_comp));
 926                         ORE_DBGMSG2("obj(0x%llx) set_attributes=%d dev=%d\n",
 927                                      _LLU(_ios_obj(ios, cur_comp)->id),
 928                                      ios->out_attr_len, dev);
 929                 }
 930
 931                 if (ios->out_attr)
 932                         osd_req_add_set_attr_list(or, ios->out_attr,
 933                                                   ios->out_attr_len);
 934
 935                 if (ios->in_attr)
 936                         osd_req_add_get_attr_list(or, ios->in_attr,
 937                                                   ios->in_attr_len);
 938         }
 939
 940 out:
 941         return ret;
 942 }
 943
 944 int ore_write(struct ore_io_state *ios)
 945 {
 946         int i;
 947         int ret;
 948
 949         if (unlikely(ios->sp2d && !ios->r4w)) {
 950                 /* A library is attempting a RAID-write without providing
 951                  * a pages lock interface.
 952                  */
 953                 WARN_ON_ONCE(1);
 954                 return -ENOTSUPP;
 955         }
 956
 957         ret = _prepare_for_striping(ios);
 958         if (unlikely(ret))
 959                 return ret;
 960
 961         for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
 962                 ret = _write_mirror(ios, i);
 963                 if (unlikely(ret))
 964                         return ret;
 965         }
 966
 967         ret = ore_io_execute(ios);
 968         return ret;
 969 }
 970 EXPORT_SYMBOL(ore_write);
 971
 972 int _ore_read_mirror(struct ore_io_state *ios, unsigned cur_comp)
 973 {
 974         struct osd_request *or;
 975         struct ore_per_dev_state *per_dev = &ios->per_dev[cur_comp];
 976         struct osd_obj_id *obj = _ios_obj(ios, cur_comp);
 977         unsigned first_dev = (unsigned)obj->id;
 978
 979         if (ios->pages && !per_dev->length)
 980                 return 0; /* Just an empty slot */
 981
 982         first_dev = per_dev->dev + first_dev % ios->layout->mirrors_p1;
 983         or = osd_start_request(_ios_od(ios, first_dev));
 984         if (unlikely(!or)) {
 985                 ORE_ERR("%s: osd_start_request failed\n", __func__);
 986                 return -ENOMEM;
 987         }
 988         per_dev->or = or;
 989
 990         if (ios->pages) {
 991                 if (per_dev->cur_sg) {
 992                         /* finalize the last sg_entry */
 993                         _ore_add_sg_seg(per_dev, 0, false);
 994                         if (unlikely(!per_dev->cur_sg))
 995                                 return 0; /* Skip parity only device */
 996
 997                         osd_req_read_sg(or, obj, per_dev->bio,
 998                                         per_dev->sglist, per_dev->cur_sg);
 999                 } else {
1000                         /* The no raid case */
1001                         osd_req_read(or, obj, per_dev->offset,
1002                                      per_dev->bio, per_dev->length);
1003                 }
1004
1005                 ORE_DBGMSG("read(0x%llx) offset=0x%llx length=0x%llx"
1006                              " dev=%d sg_len=%d\n", _LLU(obj->id),
1007                              _LLU(per_dev->offset), _LLU(per_dev->length),
1008                              first_dev, per_dev->cur_sg);
1009         } else {
1010                 BUG_ON(ios->kern_buff);
1011
1012                 osd_req_get_attributes(or, obj);
1013                 ORE_DBGMSG2("obj(0x%llx) get_attributes=%d dev=%d\n",
1014                               _LLU(obj->id),
1015                               ios->in_attr_len, first_dev);
1016         }
1017         if (ios->out_attr)
1018                 osd_req_add_set_attr_list(or, ios->out_attr, ios->out_attr_len);
1019
1020         if (ios->in_attr)
1021                 osd_req_add_get_attr_list(or, ios->in_attr, ios->in_attr_len);
1022
1023         return 0;
1024 }
1025
1026 int ore_read(struct ore_io_state *ios)
1027 {
1028         int i;
1029         int ret;
1030
1031         ret = _prepare_for_striping(ios);
1032         if (unlikely(ret))
1033                 return ret;
1034
1035         for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
1036                 ret = _ore_read_mirror(ios, i);
1037                 if (unlikely(ret))
1038                         return ret;
1039         }
1040
1041         ret = ore_io_execute(ios);
1042         return ret;
1043 }
1044 EXPORT_SYMBOL(ore_read);
1045
1046 int extract_attr_from_ios(struct ore_io_state *ios, struct osd_attr *attr)
1047 {
1048         struct osd_attr cur_attr = {.attr_page = 0}; /* start with zeros */
1049         void *iter = NULL;
1050         int nelem;
1051
1052         do {
1053                 nelem = 1;
1054                 osd_req_decode_get_attr_list(ios->per_dev[0].or,
1055                                              &cur_attr, &nelem, &iter);
1056                 if ((cur_attr.attr_page == attr->attr_page) &&
1057                     (cur_attr.attr_id == attr->attr_id)) {
1058                         attr->len = cur_attr.len;
1059                         attr->val_ptr = cur_attr.val_ptr;
1060                         return 0;
1061                 }
1062         } while (iter);
1063
1064         return -EIO;
1065 }
1066 EXPORT_SYMBOL(extract_attr_from_ios);
1067
1068 static int _truncate_mirrors(struct ore_io_state *ios, unsigned cur_comp,
1069                              struct osd_attr *attr)
1070 {
1071         int last_comp = cur_comp + ios->layout->mirrors_p1;
1072
1073         for (; cur_comp < last_comp; ++cur_comp) {
1074                 struct ore_per_dev_state *per_dev = &ios->per_dev[cur_comp];
1075                 struct osd_request *or;
1076
1077                 or = osd_start_request(_ios_od(ios, cur_comp));
1078                 if (unlikely(!or)) {
1079                         ORE_ERR("%s: osd_start_request failed\n", __func__);
1080                         return -ENOMEM;
1081                 }
1082                 per_dev->or = or;
1083
1084                 osd_req_set_attributes(or, _ios_obj(ios, cur_comp));
1085                 osd_req_add_set_attr_list(or, attr, 1);
1086         }
1087
1088         return 0;
1089 }
1090
1091 struct _trunc_info {
1092         struct ore_striping_info si;
1093         u64 prev_group_obj_off;
1094         u64 next_group_obj_off;
1095
1096         unsigned first_group_dev;
1097         unsigned nex_group_dev;
1098 };
1099
1100 static void _calc_trunk_info(struct ore_layout *layout, u64 file_offset,
1101                              struct _trunc_info *ti)
1102 {
1103         unsigned stripe_unit = layout->stripe_unit;
1104
1105         ore_calc_stripe_info(layout, file_offset, 0, &ti->si);
1106
1107         ti->prev_group_obj_off = ti->si.M * stripe_unit;
1108         ti->next_group_obj_off = ti->si.M ? (ti->si.M - 1) * stripe_unit : 0;
1109
1110         ti->first_group_dev = ti->si.dev - (ti->si.dev % layout->group_width);
1111         ti->nex_group_dev = ti->first_group_dev + layout->group_width;
1112 }
1113
1114 int ore_truncate(struct ore_layout *layout, struct ore_components *oc,
1115                    u64 size)
1116 {
1117         struct ore_io_state *ios;
1118         struct exofs_trunc_attr {
1119                 struct osd_attr attr;
1120                 __be64 newsize;
1121         } *size_attrs;
1122         struct _trunc_info ti;
1123         int i, ret;
1124
1125         ret = ore_get_io_state(layout, oc, &ios);
1126         if (unlikely(ret))
1127                 return ret;
1128
1129         _calc_trunk_info(ios->layout, size, &ti);
1130
1131         size_attrs = kcalloc(ios->oc->numdevs, sizeof(*size_attrs),
1132                              GFP_KERNEL);
1133         if (unlikely(!size_attrs)) {
1134                 ret = -ENOMEM;
1135                 goto out;
1136         }
1137
1138         ios->numdevs = ios->oc->numdevs;
1139
1140         for (i = 0; i < ios->numdevs; ++i) {
1141                 struct exofs_trunc_attr *size_attr = &size_attrs[i];
1142                 u64 obj_size;
1143
1144                 if (i < ti.first_group_dev)
1145                         obj_size = ti.prev_group_obj_off;
1146                 else if (i >= ti.nex_group_dev)
1147                         obj_size = ti.next_group_obj_off;
1148                 else if (i < ti.si.dev) /* dev within this group */
1149                         obj_size = ti.si.obj_offset +
1150                                       ios->layout->stripe_unit - ti.si.unit_off;
1151                 else if (i == ti.si.dev)
1152                         obj_size = ti.si.obj_offset;
1153                 else /* i > ti.dev */
1154                         obj_size = ti.si.obj_offset - ti.si.unit_off;
1155
1156                 size_attr->newsize = cpu_to_be64(obj_size);
1157                 size_attr->attr = g_attr_logical_length;
1158                 size_attr->attr.val_ptr = &size_attr->newsize;
1159
1160                 ORE_DBGMSG2("trunc(0x%llx) obj_offset=0x%llx dev=%d\n",
1161                              _LLU(oc->comps->obj.id), _LLU(obj_size), i);
1162                 ret = _truncate_mirrors(ios, i * ios->layout->mirrors_p1,
1163                                         &size_attr->attr);
1164                 if (unlikely(ret))
1165                         goto out;
1166         }
1167         ret = ore_io_execute(ios);
1168
1169 out:
1170         kfree(size_attrs);
1171         ore_put_io_state(ios);
1172         return ret;
1173 }
1174 EXPORT_SYMBOL(ore_truncate);
1175
1176 const struct osd_attr g_attr_logical_length = ATTR_DEF(
1177         OSD_APAGE_OBJECT_INFORMATION, OSD_ATTR_OI_LOGICAL_LENGTH, 8);
1178 EXPORT_SYMBOL(g_attr_logical_length);