fs/exofs/inode.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  * Copyrights for code taken from ext2:
   8  *     Copyright (C) 1992, 1993, 1994, 1995
   9  *     Remy Card (card@masi.ibp.fr)
  10  *     Laboratoire MASI - Institut Blaise Pascal
  11  *     Universite Pierre et Marie Curie (Paris VI)
  12  *     from
  13  *     linux/fs/minix/inode.c
  14  *     Copyright (C) 1991, 1992  Linus Torvalds
  15  *
  16  * This file is part of exofs.
  17  *
  18  * exofs is free software; you can redistribute it and/or modify
  19  * it under the terms of the GNU General Public License as published by
  20  * the Free Software Foundation.  Since it is based on ext2, and the only
  21  * valid version of GPL for the Linux kernel is version 2, the only valid
  22  * version of GPL for exofs is version 2.
  23  *
  24  * exofs is distributed in the hope that it will be useful,
  25  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  26  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  27  * GNU General Public License for more details.
  28  *
  29  * You should have received a copy of the GNU General Public License
  30  * along with exofs; if not, write to the Free Software
  31  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  32  */
  33
  34 #include <linux/slab.h>
  35
  36 #include "exofs.h"
  37
  38 #define EXOFS_DBGMSG2(M...) do {} while (0)
  39
  40 unsigned exofs_max_io_pages(struct ore_layout *layout,
  41                             unsigned expected_pages)
  42 {
  43         unsigned pages = min_t(unsigned, expected_pages,
  44                                layout->max_io_length / PAGE_SIZE);
  45
  46         return pages;
  47 }
  48
  49 struct page_collect {
  50         struct exofs_sb_info *sbi;
  51         struct inode *inode;
  52         unsigned expected_pages;
  53         struct ore_io_state *ios;
  54
  55         struct page **pages;
  56         unsigned alloc_pages;
  57         unsigned nr_pages;
  58         unsigned long length;
  59         loff_t pg_first; /* keep 64bit also in 32-arches */
  60         bool read_4_write; /* This means two things: that the read is sync
  61                             * And the pages should not be unlocked.
  62                             */
  63         struct page *that_locked_page;
  64 };
  65
  66 static void _pcol_init(struct page_collect *pcol, unsigned expected_pages,
  67                        struct inode *inode)
  68 {
  69         struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
  70
  71         pcol->sbi = sbi;
  72         pcol->inode = inode;
  73         pcol->expected_pages = expected_pages;
  74
  75         pcol->ios = NULL;
  76         pcol->pages = NULL;
  77         pcol->alloc_pages = 0;
  78         pcol->nr_pages = 0;
  79         pcol->length = 0;
  80         pcol->pg_first = -1;
  81         pcol->read_4_write = false;
  82         pcol->that_locked_page = NULL;
  83 }
  84
  85 static void _pcol_reset(struct page_collect *pcol)
  86 {
  87         pcol->expected_pages -= min(pcol->nr_pages, pcol->expected_pages);
  88
  89         pcol->pages = NULL;
  90         pcol->alloc_pages = 0;
  91         pcol->nr_pages = 0;
  92         pcol->length = 0;
  93         pcol->pg_first = -1;
  94         pcol->ios = NULL;
  95         pcol->that_locked_page = NULL;
  96
  97         /* this is probably the end of the loop but in writes
  98          * it might not end here. don't be left with nothing
  99          */
 100         if (!pcol->expected_pages)
 101                 pcol->expected_pages =
 102                                 exofs_max_io_pages(&pcol->sbi->layout, ~0);
 103 }
 104
 105 static int pcol_try_alloc(struct page_collect *pcol)
 106 {
 107         unsigned pages;
 108
 109         /* TODO: easily support bio chaining */
 110         pages =  exofs_max_io_pages(&pcol->sbi->layout, pcol->expected_pages);
 111
 112         for (; pages; pages >>= 1) {
 113                 pcol->pages = kmalloc(pages * sizeof(struct page *),
 114                                       GFP_KERNEL);
 115                 if (likely(pcol->pages)) {
 116                         pcol->alloc_pages = pages;
 117                         return 0;
 118                 }
 119         }
 120
 121         EXOFS_ERR("Failed to kmalloc expected_pages=%u\n",
 122                   pcol->expected_pages);
 123         return -ENOMEM;
 124 }
 125
 126 static void pcol_free(struct page_collect *pcol)
 127 {
 128         kfree(pcol->pages);
 129         pcol->pages = NULL;
 130
 131         if (pcol->ios) {
 132                 ore_put_io_state(pcol->ios);
 133                 pcol->ios = NULL;
 134         }
 135 }
 136
 137 static int pcol_add_page(struct page_collect *pcol, struct page *page,
 138                          unsigned len)
 139 {
 140         if (unlikely(pcol->nr_pages >= pcol->alloc_pages))
 141                 return -ENOMEM;
 142
 143         pcol->pages[pcol->nr_pages++] = page;
 144         pcol->length += len;
 145         return 0;
 146 }
 147
 148 enum {PAGE_WAS_NOT_IN_IO = 17};
 149 static int update_read_page(struct page *page, int ret)
 150 {
 151         switch (ret) {
 152         case 0:
 153                 /* Everything is OK */
 154                 SetPageUptodate(page);
 155                 if (PageError(page))
 156                         ClearPageError(page);
 157                 break;
 158         case -EFAULT:
 159                 /* In this case we were trying to read something that wasn't on
 160                  * disk yet - return a page full of zeroes.  This should be OK,
 161                  * because the object should be empty (if there was a write
 162                  * before this read, the read would be waiting with the page
 163                  * locked */
 164                 clear_highpage(page);
 165
 166                 SetPageUptodate(page);
 167                 if (PageError(page))
 168                         ClearPageError(page);
 169                 EXOFS_DBGMSG("recovered read error\n");
 170                 /* fall through */
 171         case PAGE_WAS_NOT_IN_IO:
 172                 ret = 0; /* recovered error */
 173                 break;
 174         default:
 175                 SetPageError(page);
 176         }
 177         return ret;
 178 }
 179
 180 static void update_write_page(struct page *page, int ret)
 181 {
 182         if (unlikely(ret == PAGE_WAS_NOT_IN_IO))
 183                 return; /* don't pass start don't collect $200 */
 184
 185         if (ret) {
 186                 mapping_set_error(page->mapping, ret);
 187                 SetPageError(page);
 188         }
 189         end_page_writeback(page);
 190 }
 191
 192 /* Called at the end of reads, to optionally unlock pages and update their
 193  * status.
 194  */
 195 static int __readpages_done(struct page_collect *pcol)
 196 {
 197         int i;
 198         u64 good_bytes;
 199         u64 length = 0;
 200         int ret = ore_check_io(pcol->ios, NULL);
 201
 202         if (likely(!ret)) {
 203                 good_bytes = pcol->length;
 204                 ret = PAGE_WAS_NOT_IN_IO;
 205         } else {
 206                 good_bytes = 0;
 207         }
 208
 209         EXOFS_DBGMSG2("readpages_done(0x%lx) good_bytes=0x%llx"
 210                      " length=0x%lx nr_pages=%u\n",
 211                      pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
 212                      pcol->nr_pages);
 213
 214         for (i = 0; i < pcol->nr_pages; i++) {
 215                 struct page *page = pcol->pages[i];
 216                 struct inode *inode = page->mapping->host;
 217                 int page_stat;
 218
 219                 if (inode != pcol->inode)
 220                         continue; /* osd might add more pages at end */
 221
 222                 if (likely(length < good_bytes))
 223                         page_stat = 0;
 224                 else
 225                         page_stat = ret;
 226
 227                 EXOFS_DBGMSG2("    readpages_done(0x%lx, 0x%lx) %s\n",
 228                           inode->i_ino, page->index,
 229                           page_stat ? "bad_bytes" : "good_bytes");
 230
 231                 ret = update_read_page(page, page_stat);
 232                 if (!pcol->read_4_write)
 233                         unlock_page(page);
 234                 length += PAGE_SIZE;
 235         }
 236
 237         pcol_free(pcol);
 238         EXOFS_DBGMSG2("readpages_done END\n");
 239         return ret;
 240 }
 241
 242 /* callback of async reads */
 243 static void readpages_done(struct ore_io_state *ios, void *p)
 244 {
 245         struct page_collect *pcol = p;
 246
 247         __readpages_done(pcol);
 248         atomic_dec(&pcol->sbi->s_curr_pending);
 249         kfree(pcol);
 250 }
 251
 252 static void _unlock_pcol_pages(struct page_collect *pcol, int ret, int rw)
 253 {
 254         int i;
 255
 256         for (i = 0; i < pcol->nr_pages; i++) {
 257                 struct page *page = pcol->pages[i];
 258
 259                 if (rw == READ)
 260                         update_read_page(page, ret);
 261                 else
 262                         update_write_page(page, ret);
 263
 264                 unlock_page(page);
 265         }
 266 }
 267
 268 static int _maybe_not_all_in_one_io(struct ore_io_state *ios,
 269         struct page_collect *pcol_src, struct page_collect *pcol)
 270 {
 271         /* length was wrong or offset was not page aligned */
 272         BUG_ON(pcol_src->nr_pages < ios->nr_pages);
 273
 274         if (pcol_src->nr_pages > ios->nr_pages) {
 275                 struct page **src_page;
 276                 unsigned pages_less = pcol_src->nr_pages - ios->nr_pages;
 277                 unsigned long len_less = pcol_src->length - ios->length;
 278                 unsigned i;
 279                 int ret;
 280
 281                 /* This IO was trimmed */
 282                 pcol_src->nr_pages = ios->nr_pages;
 283                 pcol_src->length = ios->length;
 284
 285                 /* Left over pages are passed to the next io */
 286                 pcol->expected_pages += pages_less;
 287                 pcol->nr_pages = pages_less;
 288                 pcol->length = len_less;
 289                 src_page = pcol_src->pages + pcol_src->nr_pages;
 290                 pcol->pg_first = (*src_page)->index;
 291
 292                 ret = pcol_try_alloc(pcol);
 293                 if (unlikely(ret))
 294                         return ret;
 295
 296                 for (i = 0; i < pages_less; ++i)
 297                         pcol->pages[i] = *src_page++;
 298
 299                 EXOFS_DBGMSG("Length was adjusted nr_pages=0x%x "
 300                         "pages_less=0x%x expected_pages=0x%x "
 301                         "next_offset=0x%llx next_len=0x%lx\n",
 302                         pcol_src->nr_pages, pages_less, pcol->expected_pages,
 303                         pcol->pg_first * PAGE_SIZE, pcol->length);
 304         }
 305         return 0;
 306 }
 307
 308 static int read_exec(struct page_collect *pcol)
 309 {
 310         struct exofs_i_info *oi = exofs_i(pcol->inode);
 311         struct ore_io_state *ios;
 312         struct page_collect *pcol_copy = NULL;
 313         int ret;
 314
 315         if (!pcol->pages)
 316                 return 0;
 317
 318         if (!pcol->ios) {
 319                 int ret = ore_get_rw_state(&pcol->sbi->layout, &oi->oc, true,
 320                                              pcol->pg_first << PAGE_CACHE_SHIFT,
 321                                              pcol->length, &pcol->ios);
 322
 323                 if (ret)
 324                         return ret;
 325         }
 326
 327         ios = pcol->ios;
 328         ios->pages = pcol->pages;
 329
 330         if (pcol->read_4_write) {
 331                 ore_read(pcol->ios);
 332                 return __readpages_done(pcol);
 333         }
 334
 335         pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
 336         if (!pcol_copy) {
 337                 ret = -ENOMEM;
 338                 goto err;
 339         }
 340
 341         *pcol_copy = *pcol;
 342         ios->done = readpages_done;
 343         ios->private = pcol_copy;
 344
 345         /* pages ownership was passed to pcol_copy */
 346         _pcol_reset(pcol);
 347
 348         ret = _maybe_not_all_in_one_io(ios, pcol_copy, pcol);
 349         if (unlikely(ret))
 350                 goto err;
 351
 352         EXOFS_DBGMSG2("read_exec(0x%lx) offset=0x%llx length=0x%llx\n",
 353                 pcol->inode->i_ino, _LLU(ios->offset), _LLU(ios->length));
 354
 355         ret = ore_read(ios);
 356         if (unlikely(ret))
 357                 goto err;
 358
 359         atomic_inc(&pcol->sbi->s_curr_pending);
 360
 361         return 0;
 362
 363 err:
 364         if (!pcol_copy) /* Failed before ownership transfer */
 365                 pcol_copy = pcol;
 366         _unlock_pcol_pages(pcol_copy, ret, READ);
 367         pcol_free(pcol_copy);
 368         kfree(pcol_copy);
 369
 370         return ret;
 371 }
 372
 373 /* readpage_strip is called either directly from readpage() or by the VFS from
 374  * within read_cache_pages(), to add one more page to be read. It will try to
 375  * collect as many contiguous pages as posible. If a discontinuity is
 376  * encountered, or it runs out of resources, it will submit the previous segment
 377  * and will start a new collection. Eventually caller must submit the last
 378  * segment if present.
 379  */
 380 static int readpage_strip(void *data, struct page *page)
 381 {
 382         struct page_collect *pcol = data;
 383         struct inode *inode = pcol->inode;
 384         struct exofs_i_info *oi = exofs_i(inode);
 385         loff_t i_size = i_size_read(inode);
 386         pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
 387         size_t len;
 388         int ret;
 389
 390         BUG_ON(!PageLocked(page));
 391
 392         /* FIXME: Just for debugging, will be removed */
 393         if (PageUptodate(page))
 394                 EXOFS_ERR("PageUptodate(0x%lx, 0x%lx)\n", pcol->inode->i_ino,
 395                           page->index);
 396
 397         pcol->that_locked_page = page;
 398
 399         if (page->index < end_index)
 400                 len = PAGE_CACHE_SIZE;
 401         else if (page->index == end_index)
 402                 len = i_size & ~PAGE_CACHE_MASK;
 403         else
 404                 len = 0;
 405
 406         if (!len || !obj_created(oi)) {
 407                 /* this will be out of bounds, or doesn't exist yet.
 408                  * Current page is cleared and the request is split
 409                  */
 410                 clear_highpage(page);
 411
 412                 SetPageUptodate(page);
 413                 if (PageError(page))
 414                         ClearPageError(page);
 415
 416                 if (!pcol->read_4_write)
 417                         unlock_page(page);
 418                 EXOFS_DBGMSG("readpage_strip(0x%lx) empty page len=%zx "
 419                              "read_4_write=%d index=0x%lx end_index=0x%lx "
 420                              "splitting\n", inode->i_ino, len,
 421                              pcol->read_4_write, page->index, end_index);
 422
 423                 return read_exec(pcol);
 424         }
 425
 426 try_again:
 427
 428         if (unlikely(pcol->pg_first == -1)) {
 429                 pcol->pg_first = page->index;
 430         } else if (unlikely((pcol->pg_first + pcol->nr_pages) !=
 431                    page->index)) {
 432                 /* Discontinuity detected, split the request */
 433                 ret = read_exec(pcol);
 434                 if (unlikely(ret))
 435                         goto fail;
 436                 goto try_again;
 437         }
 438
 439         if (!pcol->pages) {
 440                 ret = pcol_try_alloc(pcol);
 441                 if (unlikely(ret))
 442                         goto fail;
 443         }
 444
 445         if (len != PAGE_CACHE_SIZE)
 446                 zero_user(page, len, PAGE_CACHE_SIZE - len);
 447
 448         EXOFS_DBGMSG2("    readpage_strip(0x%lx, 0x%lx) len=0x%zx\n",
 449                      inode->i_ino, page->index, len);
 450
 451         ret = pcol_add_page(pcol, page, len);
 452         if (ret) {
 453                 EXOFS_DBGMSG2("Failed pcol_add_page pages[i]=%p "
 454                           "this_len=0x%zx nr_pages=%u length=0x%lx\n",
 455                           page, len, pcol->nr_pages, pcol->length);
 456
 457                 /* split the request, and start again with current page */
 458                 ret = read_exec(pcol);
 459                 if (unlikely(ret))
 460                         goto fail;
 461
 462                 goto try_again;
 463         }
 464
 465         return 0;
 466
 467 fail:
 468         /* SetPageError(page); ??? */
 469         unlock_page(page);
 470         return ret;
 471 }
 472
 473 static int exofs_readpages(struct file *file, struct address_space *mapping,
 474                            struct list_head *pages, unsigned nr_pages)
 475 {
 476         struct page_collect pcol;
 477         int ret;
 478
 479         _pcol_init(&pcol, nr_pages, mapping->host);
 480
 481         ret = read_cache_pages(mapping, pages, readpage_strip, &pcol);
 482         if (ret) {
 483                 EXOFS_ERR("read_cache_pages => %d\n", ret);
 484                 return ret;
 485         }
 486
 487         ret = read_exec(&pcol);
 488         if (unlikely(ret))
 489                 return ret;
 490
 491         return read_exec(&pcol);
 492 }
 493
 494 static int _readpage(struct page *page, bool read_4_write)
 495 {
 496         struct page_collect pcol;
 497         int ret;
 498
 499         _pcol_init(&pcol, 1, page->mapping->host);
 500
 501         pcol.read_4_write = read_4_write;
 502         ret = readpage_strip(&pcol, page);
 503         if (ret) {
 504                 EXOFS_ERR("_readpage => %d\n", ret);
 505                 return ret;
 506         }
 507
 508         return read_exec(&pcol);
 509 }
 510
 511 /*
 512  * We don't need the file
 513  */
 514 static int exofs_readpage(struct file *file, struct page *page)
 515 {
 516         return _readpage(page, false);
 517 }
 518
 519 /* Callback for osd_write. All writes are asynchronous */
 520 static void writepages_done(struct ore_io_state *ios, void *p)
 521 {
 522         struct page_collect *pcol = p;
 523         int i;
 524         u64  good_bytes;
 525         u64  length = 0;
 526         int ret = ore_check_io(ios, NULL);
 527
 528         atomic_dec(&pcol->sbi->s_curr_pending);
 529
 530         if (likely(!ret)) {
 531                 good_bytes = pcol->length;
 532                 ret = PAGE_WAS_NOT_IN_IO;
 533         } else {
 534                 good_bytes = 0;
 535         }
 536
 537         EXOFS_DBGMSG2("writepages_done(0x%lx) good_bytes=0x%llx"
 538                      " length=0x%lx nr_pages=%u\n",
 539                      pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
 540                      pcol->nr_pages);
 541
 542         for (i = 0; i < pcol->nr_pages; i++) {
 543                 struct page *page = pcol->pages[i];
 544                 struct inode *inode = page->mapping->host;
 545                 int page_stat;
 546
 547                 if (inode != pcol->inode)
 548                         continue; /* osd might add more pages to a bio */
 549
 550                 if (likely(length < good_bytes))
 551                         page_stat = 0;
 552                 else
 553                         page_stat = ret;
 554
 555                 update_write_page(page, page_stat);
 556                 unlock_page(page);
 557                 EXOFS_DBGMSG2("    writepages_done(0x%lx, 0x%lx) status=%d\n",
 558                              inode->i_ino, page->index, page_stat);
 559
 560                 length += PAGE_SIZE;
 561         }
 562
 563         pcol_free(pcol);
 564         kfree(pcol);
 565         EXOFS_DBGMSG2("writepages_done END\n");
 566 }
 567
 568 static struct page *__r4w_get_page(void *priv, u64 offset, bool *uptodate)
 569 {
 570         struct page_collect *pcol = priv;
 571         pgoff_t index = offset / PAGE_SIZE;
 572
 573         if (!pcol->that_locked_page ||
 574             (pcol->that_locked_page->index != index)) {
 575                 struct page *page;
 576                 loff_t i_size = i_size_read(pcol->inode);
 577
 578                 if (offset >= i_size) {
 579                         *uptodate = true;
 580                         EXOFS_DBGMSG2("offset >= i_size index=0x%lx\n", index);
 581                         return ZERO_PAGE(0);
 582                 }
 583
 584                 page =  find_get_page(pcol->inode->i_mapping, index);
 585                 if (!page) {
 586                         page = find_or_create_page(pcol->inode->i_mapping,
 587                                                    index, GFP_NOFS);
 588                         if (unlikely(!page)) {
 589                                 EXOFS_DBGMSG("grab_cache_page Failed "
 590                                         "index=0x%llx\n", _LLU(index));
 591                                 return NULL;
 592                         }
 593                         unlock_page(page);
 594                 }
 595                 if (PageDirty(page) || PageWriteback(page))
 596                         *uptodate = true;
 597                 else
 598                         *uptodate = PageUptodate(page);
 599                 EXOFS_DBGMSG2("index=0x%lx uptodate=%d\n", index, *uptodate);
 600                 return page;
 601         } else {
 602                 EXOFS_DBGMSG2("YES that_locked_page index=0x%lx\n",
 603                              pcol->that_locked_page->index);
 604                 *uptodate = true;
 605                 return pcol->that_locked_page;
 606         }
 607 }
 608
 609 static void __r4w_put_page(void *priv, struct page *page)
 610 {
 611         struct page_collect *pcol = priv;
 612
 613         if ((pcol->that_locked_page != page) && (ZERO_PAGE(0) != page)) {
 614                 EXOFS_DBGMSG2("index=0x%lx\n", page->index);
 615                 page_cache_release(page);
 616                 return;
 617         }
 618         EXOFS_DBGMSG2("that_locked_page index=0x%lx\n",
 619                      ZERO_PAGE(0) == page ? -1 : page->index);
 620 }
 621
 622 static const struct _ore_r4w_op _r4w_op = {
 623         .get_page = &__r4w_get_page,
 624         .put_page = &__r4w_put_page,
 625 };
 626
 627 static int write_exec(struct page_collect *pcol)
 628 {
 629         struct exofs_i_info *oi = exofs_i(pcol->inode);
 630         struct ore_io_state *ios;
 631         struct page_collect *pcol_copy = NULL;
 632         int ret;
 633
 634         if (!pcol->pages)
 635                 return 0;
 636
 637         BUG_ON(pcol->ios);
 638         ret = ore_get_rw_state(&pcol->sbi->layout, &oi->oc, false,
 639                                  pcol->pg_first << PAGE_CACHE_SHIFT,
 640                                  pcol->length, &pcol->ios);
 641         if (unlikely(ret))
 642                 goto err;
 643
 644         pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
 645         if (!pcol_copy) {
 646                 EXOFS_ERR("write_exec: Failed to kmalloc(pcol)\n");
 647                 ret = -ENOMEM;
 648                 goto err;
 649         }
 650
 651         *pcol_copy = *pcol;
 652
 653         ios = pcol->ios;
 654         ios->pages = pcol_copy->pages;
 655         ios->done = writepages_done;
 656         ios->r4w = &_r4w_op;
 657         ios->private = pcol_copy;
 658
 659         /* pages ownership was passed to pcol_copy */
 660         _pcol_reset(pcol);
 661
 662         ret = _maybe_not_all_in_one_io(ios, pcol_copy, pcol);
 663         if (unlikely(ret))
 664                 goto err;
 665
 666         EXOFS_DBGMSG2("write_exec(0x%lx) offset=0x%llx length=0x%llx\n",
 667                 pcol->inode->i_ino, _LLU(ios->offset), _LLU(ios->length));
 668
 669         ret = ore_write(ios);
 670         if (unlikely(ret)) {
 671                 EXOFS_ERR("write_exec: ore_write() Failed\n");
 672                 goto err;
 673         }
 674
 675         atomic_inc(&pcol->sbi->s_curr_pending);
 676         return 0;
 677
 678 err:
 679         if (!pcol_copy) /* Failed before ownership transfer */
 680                 pcol_copy = pcol;
 681         _unlock_pcol_pages(pcol_copy, ret, WRITE);
 682         pcol_free(pcol_copy);
 683         kfree(pcol_copy);
 684
 685         return ret;
 686 }
 687
 688 /* writepage_strip is called either directly from writepage() or by the VFS from
 689  * within write_cache_pages(), to add one more page to be written to storage.
 690  * It will try to collect as many contiguous pages as possible. If a
 691  * discontinuity is encountered or it runs out of resources it will submit the
 692  * previous segment and will start a new collection.
 693  * Eventually caller must submit the last segment if present.
 694  */
 695 static int writepage_strip(struct page *page,
 696                            struct writeback_control *wbc_unused, void *data)
 697 {
 698         struct page_collect *pcol = data;
 699         struct inode *inode = pcol->inode;
 700         struct exofs_i_info *oi = exofs_i(inode);
 701         loff_t i_size = i_size_read(inode);
 702         pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
 703         size_t len;
 704         int ret;
 705
 706         BUG_ON(!PageLocked(page));
 707
 708         ret = wait_obj_created(oi);
 709         if (unlikely(ret))
 710                 goto fail;
 711
 712         if (page->index < end_index)
 713                 /* in this case, the page is within the limits of the file */
 714                 len = PAGE_CACHE_SIZE;
 715         else {
 716                 len = i_size & ~PAGE_CACHE_MASK;
 717
 718                 if (page->index > end_index || !len) {
 719                         /* in this case, the page is outside the limits
 720                          * (truncate in progress)
 721                          */
 722                         ret = write_exec(pcol);
 723                         if (unlikely(ret))
 724                                 goto fail;
 725                         if (PageError(page))
 726                                 ClearPageError(page);
 727                         unlock_page(page);
 728                         EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) "
 729                                      "outside the limits\n",
 730                                      inode->i_ino, page->index);
 731                         return 0;
 732                 }
 733         }
 734
 735 try_again:
 736
 737         if (unlikely(pcol->pg_first == -1)) {
 738                 pcol->pg_first = page->index;
 739         } else if (unlikely((pcol->pg_first + pcol->nr_pages) !=
 740                    page->index)) {
 741                 /* Discontinuity detected, split the request */
 742                 ret = write_exec(pcol);
 743                 if (unlikely(ret))
 744                         goto fail;
 745
 746                 EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) Discontinuity\n",
 747                              inode->i_ino, page->index);
 748                 goto try_again;
 749         }
 750
 751         if (!pcol->pages) {
 752                 ret = pcol_try_alloc(pcol);
 753                 if (unlikely(ret))
 754                         goto fail;
 755         }
 756
 757         EXOFS_DBGMSG2("    writepage_strip(0x%lx, 0x%lx) len=0x%zx\n",
 758                      inode->i_ino, page->index, len);
 759
 760         ret = pcol_add_page(pcol, page, len);
 761         if (unlikely(ret)) {
 762                 EXOFS_DBGMSG2("Failed pcol_add_page "
 763                              "nr_pages=%u total_length=0x%lx\n",
 764                              pcol->nr_pages, pcol->length);
 765
 766                 /* split the request, next loop will start again */
 767                 ret = write_exec(pcol);
 768                 if (unlikely(ret)) {
 769                         EXOFS_DBGMSG("write_exec failed => %d", ret);
 770                         goto fail;
 771                 }
 772
 773                 goto try_again;
 774         }
 775
 776         BUG_ON(PageWriteback(page));
 777         set_page_writeback(page);
 778
 779         return 0;
 780
 781 fail:
 782         EXOFS_DBGMSG("Error: writepage_strip(0x%lx, 0x%lx)=>%d\n",
 783                      inode->i_ino, page->index, ret);
 784         set_bit(AS_EIO, &page->mapping->flags);
 785         unlock_page(page);
 786         return ret;
 787 }
 788
 789 static int exofs_writepages(struct address_space *mapping,
 790                        struct writeback_control *wbc)
 791 {
 792         struct page_collect pcol;
 793         long start, end, expected_pages;
 794         int ret;
 795
 796         start = wbc->range_start >> PAGE_CACHE_SHIFT;
 797         end = (wbc->range_end == LLONG_MAX) ?
 798                         start + mapping->nrpages :
 799                         wbc->range_end >> PAGE_CACHE_SHIFT;
 800
 801         if (start || end)
 802                 expected_pages = end - start + 1;
 803         else
 804                 expected_pages = mapping->nrpages;
 805
 806         if (expected_pages < 32L)
 807                 expected_pages = 32L;
 808
 809         EXOFS_DBGMSG2("inode(0x%lx) wbc->start=0x%llx wbc->end=0x%llx "
 810                      "nrpages=%lu start=0x%lx end=0x%lx expected_pages=%ld\n",
 811                      mapping->host->i_ino, wbc->range_start, wbc->range_end,
 812                      mapping->nrpages, start, end, expected_pages);
 813
 814         _pcol_init(&pcol, expected_pages, mapping->host);
 815
 816         ret = write_cache_pages(mapping, wbc, writepage_strip, &pcol);
 817         if (unlikely(ret)) {
 818                 EXOFS_ERR("write_cache_pages => %d\n", ret);
 819                 return ret;
 820         }
 821
 822         ret = write_exec(&pcol);
 823         if (unlikely(ret))
 824                 return ret;
 825
 826         if (wbc->sync_mode == WB_SYNC_ALL) {
 827                 return write_exec(&pcol); /* pump the last reminder */
 828         } else if (pcol.nr_pages) {
 829                 /* not SYNC let the reminder join the next writeout */
 830                 unsigned i;
 831
 832                 for (i = 0; i < pcol.nr_pages; i++) {
 833                         struct page *page = pcol.pages[i];
 834
 835                         end_page_writeback(page);
 836                         set_page_dirty(page);
 837                         unlock_page(page);
 838                 }
 839         }
 840         return 0;
 841 }
 842
 843 /*
 844 static int exofs_writepage(struct page *page, struct writeback_control *wbc)
 845 {
 846         struct page_collect pcol;
 847         int ret;
 848
 849         _pcol_init(&pcol, 1, page->mapping->host);
 850
 851         ret = writepage_strip(page, NULL, &pcol);
 852         if (ret) {
 853                 EXOFS_ERR("exofs_writepage => %d\n", ret);
 854                 return ret;
 855         }
 856
 857         return write_exec(&pcol);
 858 }
 859 */
 860 /* i_mutex held using inode->i_size directly */
 861 static void _write_failed(struct inode *inode, loff_t to)
 862 {
 863         if (to > inode->i_size)
 864                 truncate_pagecache(inode, inode->i_size);
 865 }
 866
 867 int exofs_write_begin(struct file *file, struct address_space *mapping,
 868                 loff_t pos, unsigned len, unsigned flags,
 869                 struct page **pagep, void **fsdata)
 870 {
 871         int ret = 0;
 872         struct page *page;
 873
 874         page = *pagep;
 875         if (page == NULL) {
 876                 ret = simple_write_begin(file, mapping, pos, len, flags, pagep,
 877                                          fsdata);
 878                 if (ret) {
 879                         EXOFS_DBGMSG("simple_write_begin failed\n");
 880                         goto out;
 881                 }
 882
 883                 page = *pagep;
 884         }
 885
 886          /* read modify write */
 887         if (!PageUptodate(page) && (len != PAGE_CACHE_SIZE)) {
 888                 loff_t i_size = i_size_read(mapping->host);
 889                 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
 890                 size_t rlen;
 891
 892                 if (page->index < end_index)
 893                         rlen = PAGE_CACHE_SIZE;
 894                 else if (page->index == end_index)
 895                         rlen = i_size & ~PAGE_CACHE_MASK;
 896                 else
 897                         rlen = 0;
 898
 899                 if (!rlen) {
 900                         clear_highpage(page);
 901                         SetPageUptodate(page);
 902                         goto out;
 903                 }
 904
 905                 ret = _readpage(page, true);
 906                 if (ret) {
 907                         /*SetPageError was done by _readpage. Is it ok?*/
 908                         unlock_page(page);
 909                         EXOFS_DBGMSG("__readpage failed\n");
 910                 }
 911         }
 912 out:
 913         if (unlikely(ret))
 914                 _write_failed(mapping->host, pos + len);
 915
 916         return ret;
 917 }
 918
 919 static int exofs_write_begin_export(struct file *file,
 920                 struct address_space *mapping,
 921                 loff_t pos, unsigned len, unsigned flags,
 922                 struct page **pagep, void **fsdata)
 923 {
 924         *pagep = NULL;
 925
 926         return exofs_write_begin(file, mapping, pos, len, flags, pagep,
 927                                         fsdata);
 928 }
 929
 930 static int exofs_write_end(struct file *file, struct address_space *mapping,
 931                         loff_t pos, unsigned len, unsigned copied,
 932                         struct page *page, void *fsdata)
 933 {
 934         struct inode *inode = mapping->host;
 935         /* According to comment in simple_write_end i_mutex is held */
 936         loff_t i_size = inode->i_size;
 937         int ret;
 938
 939         ret = simple_write_end(file, mapping,pos, len, copied, page, fsdata);
 940         if (unlikely(ret))
 941                 _write_failed(inode, pos + len);
 942
 943         /* TODO: once simple_write_end marks inode dirty remove */
 944         if (i_size != inode->i_size)
 945                 mark_inode_dirty(inode);
 946         return ret;
 947 }
 948
 949 static int exofs_releasepage(struct page *page, gfp_t gfp)
 950 {
 951         EXOFS_DBGMSG("page 0x%lx\n", page->index);
 952         WARN_ON(1);
 953         return 0;
 954 }
 955
 956 static void exofs_invalidatepage(struct page *page, unsigned int offset,
 957                                  unsigned int length)
 958 {
 959         EXOFS_DBGMSG("page 0x%lx offset 0x%x length 0x%x\n",
 960                      page->index, offset, length);
 961         WARN_ON(1);
 962 }
 963
 964
 965  /* TODO: Should be easy enough to do proprly */
 966 static ssize_t exofs_direct_IO(int rw, struct kiocb *iocb,
 967                 const struct iovec *iov, loff_t offset, unsigned long nr_segs)
 968 {
 969         return 0;
 970 }
 971
 972 const struct address_space_operations exofs_aops = {
 973         .readpage       = exofs_readpage,
 974         .readpages      = exofs_readpages,
 975         .writepage      = NULL,
 976         .writepages     = exofs_writepages,
 977         .write_begin    = exofs_write_begin_export,
 978         .write_end      = exofs_write_end,
 979         .releasepage    = exofs_releasepage,
 980         .set_page_dirty = __set_page_dirty_nobuffers,
 981         .invalidatepage = exofs_invalidatepage,
 982
 983         /* Not implemented Yet */
 984         .bmap           = NULL, /* TODO: use osd's OSD_ACT_READ_MAP */
 985         .direct_IO      = exofs_direct_IO,
 986
 987         /* With these NULL has special meaning or default is not exported */
 988         .get_xip_mem    = NULL,
 989         .migratepage    = NULL,
 990         .launder_page   = NULL,
 991         .is_partially_uptodate = NULL,
 992         .error_remove_page = NULL,
 993 };
 994
 995 /******************************************************************************
 996  * INODE OPERATIONS
 997  *****************************************************************************/
 998
 999 /*
1000  * Test whether an inode is a fast symlink.
1001  */
1002 static inline int exofs_inode_is_fast_symlink(struct inode *inode)
1003 {
1004         struct exofs_i_info *oi = exofs_i(inode);
1005
1006         return S_ISLNK(inode->i_mode) && (oi->i_data[0] != 0);
1007 }
1008
1009 static int _do_truncate(struct inode *inode, loff_t newsize)
1010 {
1011         struct exofs_i_info *oi = exofs_i(inode);
1012         struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
1013         int ret;
1014
1015         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1016
1017         ret = ore_truncate(&sbi->layout, &oi->oc, (u64)newsize);
1018         if (likely(!ret))
1019                 truncate_setsize(inode, newsize);
1020
1021         EXOFS_DBGMSG2("(0x%lx) size=0x%llx ret=>%d\n",
1022                      inode->i_ino, newsize, ret);
1023         return ret;
1024 }
1025
1026 /*
1027  * Set inode attributes - update size attribute on OSD if needed,
1028  *                        otherwise just call generic functions.
1029  */
1030 int exofs_setattr(struct dentry *dentry, struct iattr *iattr)
1031 {
1032         struct inode *inode = dentry->d_inode;
1033         int error;
1034
1035         /* if we are about to modify an object, and it hasn't been
1036          * created yet, wait
1037          */
1038         error = wait_obj_created(exofs_i(inode));
1039         if (unlikely(error))
1040                 return error;
1041
1042         error = inode_change_ok(inode, iattr);
1043         if (unlikely(error))
1044                 return error;
1045
1046         if ((iattr->ia_valid & ATTR_SIZE) &&
1047             iattr->ia_size != i_size_read(inode)) {
1048                 error = _do_truncate(inode, iattr->ia_size);
1049                 if (unlikely(error))
1050                         return error;
1051         }
1052
1053         setattr_copy(inode, iattr);
1054         mark_inode_dirty(inode);
1055         return 0;
1056 }
1057
1058 static const struct osd_attr g_attr_inode_file_layout = ATTR_DEF(
1059         EXOFS_APAGE_FS_DATA,
1060         EXOFS_ATTR_INODE_FILE_LAYOUT,
1061         0);
1062 static const struct osd_attr g_attr_inode_dir_layout = ATTR_DEF(
1063         EXOFS_APAGE_FS_DATA,
1064         EXOFS_ATTR_INODE_DIR_LAYOUT,
1065         0);
1066
1067 /*
1068  * Read the Linux inode info from the OSD, and return it as is. In exofs the
1069  * inode info is in an application specific page/attribute of the osd-object.
1070  */
1071 static int exofs_get_inode(struct super_block *sb, struct exofs_i_info *oi,
1072                     struct exofs_fcb *inode)
1073 {
1074         struct exofs_sb_info *sbi = sb->s_fs_info;
1075         struct osd_attr attrs[] = {
1076                 [0] = g_attr_inode_data,
1077                 [1] = g_attr_inode_file_layout,
1078                 [2] = g_attr_inode_dir_layout,
1079         };
1080         struct ore_io_state *ios;
1081         struct exofs_on_disk_inode_layout *layout;
1082         int ret;
1083
1084         ret = ore_get_io_state(&sbi->layout, &oi->oc, &ios);
1085         if (unlikely(ret)) {
1086                 EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__);
1087                 return ret;
1088         }
1089
1090         attrs[1].len = exofs_on_disk_inode_layout_size(sbi->oc.numdevs);
1091         attrs[2].len = exofs_on_disk_inode_layout_size(sbi->oc.numdevs);
1092
1093         ios->in_attr = attrs;
1094         ios->in_attr_len = ARRAY_SIZE(attrs);
1095
1096         ret = ore_read(ios);
1097         if (unlikely(ret)) {
1098                 EXOFS_ERR("object(0x%llx) corrupted, return empty file=>%d\n",
1099                           _LLU(oi->one_comp.obj.id), ret);
1100                 memset(inode, 0, sizeof(*inode));
1101                 inode->i_mode = 0040000 | (0777 & ~022);
1102                 /* If object is lost on target we might as well enable it's
1103                  * delete.
1104                  */
1105                 ret = 0;
1106                 goto out;
1107         }
1108
1109         ret = extract_attr_from_ios(ios, &attrs[0]);
1110         if (ret) {
1111                 EXOFS_ERR("%s: extract_attr 0 of inode failed\n", __func__);
1112                 goto out;
1113         }
1114         WARN_ON(attrs[0].len != EXOFS_INO_ATTR_SIZE);
1115         memcpy(inode, attrs[0].val_ptr, EXOFS_INO_ATTR_SIZE);
1116
1117         ret = extract_attr_from_ios(ios, &attrs[1]);
1118         if (ret) {
1119                 EXOFS_ERR("%s: extract_attr 1 of inode failed\n", __func__);
1120                 goto out;
1121         }
1122         if (attrs[1].len) {
1123                 layout = attrs[1].val_ptr;
1124                 if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) {
1125                         EXOFS_ERR("%s: unsupported files layout %d\n",
1126                                 __func__, layout->gen_func);
1127                         ret = -ENOTSUPP;
1128                         goto out;
1129                 }
1130         }
1131
1132         ret = extract_attr_from_ios(ios, &attrs[2]);
1133         if (ret) {
1134                 EXOFS_ERR("%s: extract_attr 2 of inode failed\n", __func__);
1135                 goto out;
1136         }
1137         if (attrs[2].len) {
1138                 layout = attrs[2].val_ptr;
1139                 if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) {
1140                         EXOFS_ERR("%s: unsupported meta-data layout %d\n",
1141                                 __func__, layout->gen_func);
1142                         ret = -ENOTSUPP;
1143                         goto out;
1144                 }
1145         }
1146
1147 out:
1148         ore_put_io_state(ios);
1149         return ret;
1150 }
1151
1152 static void __oi_init(struct exofs_i_info *oi)
1153 {
1154         init_waitqueue_head(&oi->i_wq);
1155         oi->i_flags = 0;
1156 }
1157 /*
1158  * Fill in an inode read from the OSD and set it up for use
1159  */
1160 struct inode *exofs_iget(struct super_block *sb, unsigned long ino)
1161 {
1162         struct exofs_i_info *oi;
1163         struct exofs_fcb fcb;
1164         struct inode *inode;
1165         int ret;
1166
1167         inode = iget_locked(sb, ino);
1168         if (!inode)
1169                 return ERR_PTR(-ENOMEM);
1170         if (!(inode->i_state & I_NEW))
1171                 return inode;
1172         oi = exofs_i(inode);
1173         __oi_init(oi);
1174         exofs_init_comps(&oi->oc, &oi->one_comp, sb->s_fs_info,
1175                          exofs_oi_objno(oi));
1176
1177         /* read the inode from the osd */
1178         ret = exofs_get_inode(sb, oi, &fcb);
1179         if (ret)
1180                 goto bad_inode;
1181
1182         set_obj_created(oi);
1183
1184         /* copy stuff from on-disk struct to in-memory struct */
1185         inode->i_mode = le16_to_cpu(fcb.i_mode);
1186         i_uid_write(inode, le32_to_cpu(fcb.i_uid));
1187         i_gid_write(inode, le32_to_cpu(fcb.i_gid));
1188         set_nlink(inode, le16_to_cpu(fcb.i_links_count));
1189         inode->i_ctime.tv_sec = (signed)le32_to_cpu(fcb.i_ctime);
1190         inode->i_atime.tv_sec = (signed)le32_to_cpu(fcb.i_atime);
1191         inode->i_mtime.tv_sec = (signed)le32_to_cpu(fcb.i_mtime);
1192         inode->i_ctime.tv_nsec =
1193                 inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = 0;
1194         oi->i_commit_size = le64_to_cpu(fcb.i_size);
1195         i_size_write(inode, oi->i_commit_size);
1196         inode->i_blkbits = EXOFS_BLKSHIFT;
1197         inode->i_generation = le32_to_cpu(fcb.i_generation);
1198
1199         oi->i_dir_start_lookup = 0;
1200
1201         if ((inode->i_nlink == 0) && (inode->i_mode == 0)) {
1202                 ret = -ESTALE;
1203                 goto bad_inode;
1204         }
1205
1206         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1207                 if (fcb.i_data[0])
1208                         inode->i_rdev =
1209                                 old_decode_dev(le32_to_cpu(fcb.i_data[0]));
1210                 else
1211                         inode->i_rdev =
1212                                 new_decode_dev(le32_to_cpu(fcb.i_data[1]));
1213         } else {
1214                 memcpy(oi->i_data, fcb.i_data, sizeof(fcb.i_data));
1215         }
1216
1217         inode->i_mapping->backing_dev_info = sb->s_bdi;
1218         if (S_ISREG(inode->i_mode)) {
1219                 inode->i_op = &exofs_file_inode_operations;
1220                 inode->i_fop = &exofs_file_operations;
1221                 inode->i_mapping->a_ops = &exofs_aops;
1222         } else if (S_ISDIR(inode->i_mode)) {
1223                 inode->i_op = &exofs_dir_inode_operations;
1224                 inode->i_fop = &exofs_dir_operations;
1225                 inode->i_mapping->a_ops = &exofs_aops;
1226         } else if (S_ISLNK(inode->i_mode)) {
1227                 if (exofs_inode_is_fast_symlink(inode))
1228                         inode->i_op = &exofs_fast_symlink_inode_operations;
1229                 else {
1230                         inode->i_op = &exofs_symlink_inode_operations;
1231                         inode->i_mapping->a_ops = &exofs_aops;
1232                 }
1233         } else {
1234                 inode->i_op = &exofs_special_inode_operations;
1235                 if (fcb.i_data[0])
1236                         init_special_inode(inode, inode->i_mode,
1237                            old_decode_dev(le32_to_cpu(fcb.i_data[0])));
1238                 else
1239                         init_special_inode(inode, inode->i_mode,
1240                            new_decode_dev(le32_to_cpu(fcb.i_data[1])));
1241         }
1242
1243         unlock_new_inode(inode);
1244         return inode;
1245
1246 bad_inode:
1247         iget_failed(inode);
1248         return ERR_PTR(ret);
1249 }
1250
1251 int __exofs_wait_obj_created(struct exofs_i_info *oi)
1252 {
1253         if (!obj_created(oi)) {
1254                 EXOFS_DBGMSG("!obj_created\n");
1255                 BUG_ON(!obj_2bcreated(oi));
1256                 wait_event(oi->i_wq, obj_created(oi));
1257                 EXOFS_DBGMSG("wait_event done\n");
1258         }
1259         return unlikely(is_bad_inode(&oi->vfs_inode)) ? -EIO : 0;
1260 }
1261
1262 /*
1263  * Callback function from exofs_new_inode().  The important thing is that we
1264  * set the obj_created flag so that other methods know that the object exists on
1265  * the OSD.
1266  */
1267 static void create_done(struct ore_io_state *ios, void *p)
1268 {
1269         struct inode *inode = p;
1270         struct exofs_i_info *oi = exofs_i(inode);
1271         struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
1272         int ret;
1273
1274         ret = ore_check_io(ios, NULL);
1275         ore_put_io_state(ios);
1276
1277         atomic_dec(&sbi->s_curr_pending);
1278
1279         if (unlikely(ret)) {
1280                 EXOFS_ERR("object=0x%llx creation failed in pid=0x%llx",
1281                           _LLU(exofs_oi_objno(oi)),
1282                           _LLU(oi->one_comp.obj.partition));
1283                 /*TODO: When FS is corrupted creation can fail, object already
1284                  * exist. Get rid of this asynchronous creation, if exist
1285                  * increment the obj counter and try the next object. Until we
1286                  * succeed. All these dangling objects will be made into lost
1287                  * files by chkfs.exofs
1288                  */
1289         }
1290
1291         set_obj_created(oi);
1292
1293         wake_up(&oi->i_wq);
1294 }
1295
1296 /*
1297  * Set up a new inode and create an object for it on the OSD
1298  */
1299 struct inode *exofs_new_inode(struct inode *dir, umode_t mode)
1300 {
1301         struct super_block *sb = dir->i_sb;
1302         struct exofs_sb_info *sbi = sb->s_fs_info;
1303         struct inode *inode;
1304         struct exofs_i_info *oi;
1305         struct ore_io_state *ios;
1306         int ret;
1307
1308         inode = new_inode(sb);
1309         if (!inode)
1310                 return ERR_PTR(-ENOMEM);
1311
1312         oi = exofs_i(inode);
1313         __oi_init(oi);
1314
1315         set_obj_2bcreated(oi);
1316
1317         inode->i_mapping->backing_dev_info = sb->s_bdi;
1318         inode_init_owner(inode, dir, mode);
1319         inode->i_ino = sbi->s_nextid++;
1320         inode->i_blkbits = EXOFS_BLKSHIFT;
1321         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1322         oi->i_commit_size = inode->i_size = 0;
1323         spin_lock(&sbi->s_next_gen_lock);
1324         inode->i_generation = sbi->s_next_generation++;
1325         spin_unlock(&sbi->s_next_gen_lock);
1326         insert_inode_hash(inode);
1327
1328         exofs_init_comps(&oi->oc, &oi->one_comp, sb->s_fs_info,
1329                          exofs_oi_objno(oi));
1330         exofs_sbi_write_stats(sbi); /* Make sure new sbi->s_nextid is on disk */
1331
1332         mark_inode_dirty(inode);
1333
1334         ret = ore_get_io_state(&sbi->layout, &oi->oc, &ios);
1335         if (unlikely(ret)) {
1336                 EXOFS_ERR("exofs_new_inode: ore_get_io_state failed\n");
1337                 return ERR_PTR(ret);
1338         }
1339
1340         ios->done = create_done;
1341         ios->private = inode;
1342
1343         ret = ore_create(ios);
1344         if (ret) {
1345                 ore_put_io_state(ios);
1346                 return ERR_PTR(ret);
1347         }
1348         atomic_inc(&sbi->s_curr_pending);
1349
1350         return inode;
1351 }
1352
1353 /*
1354  * struct to pass two arguments to update_inode's callback
1355  */
1356 struct updatei_args {
1357         struct exofs_sb_info    *sbi;
1358         struct exofs_fcb        fcb;
1359 };
1360
1361 /*
1362  * Callback function from exofs_update_inode().
1363  */
1364 static void updatei_done(struct ore_io_state *ios, void *p)
1365 {
1366         struct updatei_args *args = p;
1367
1368         ore_put_io_state(ios);
1369
1370         atomic_dec(&args->sbi->s_curr_pending);
1371
1372         kfree(args);
1373 }
1374
1375 /*
1376  * Write the inode to the OSD.  Just fill up the struct, and set the attribute
1377  * synchronously or asynchronously depending on the do_sync flag.
1378  */
1379 static int exofs_update_inode(struct inode *inode, int do_sync)
1380 {
1381         struct exofs_i_info *oi = exofs_i(inode);
1382         struct super_block *sb = inode->i_sb;
1383         struct exofs_sb_info *sbi = sb->s_fs_info;
1384         struct ore_io_state *ios;
1385         struct osd_attr attr;
1386         struct exofs_fcb *fcb;
1387         struct updatei_args *args;
1388         int ret;
1389
1390         args = kzalloc(sizeof(*args), GFP_KERNEL);
1391         if (!args) {
1392                 EXOFS_DBGMSG("Failed kzalloc of args\n");
1393                 return -ENOMEM;
1394         }
1395
1396         fcb = &args->fcb;
1397
1398         fcb->i_mode = cpu_to_le16(inode->i_mode);
1399         fcb->i_uid = cpu_to_le32(i_uid_read(inode));
1400         fcb->i_gid = cpu_to_le32(i_gid_read(inode));
1401         fcb->i_links_count = cpu_to_le16(inode->i_nlink);
1402         fcb->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1403         fcb->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1404         fcb->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1405         oi->i_commit_size = i_size_read(inode);
1406         fcb->i_size = cpu_to_le64(oi->i_commit_size);
1407         fcb->i_generation = cpu_to_le32(inode->i_generation);
1408
1409         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1410                 if (old_valid_dev(inode->i_rdev)) {
1411                         fcb->i_data[0] =
1412                                 cpu_to_le32(old_encode_dev(inode->i_rdev));
1413                         fcb->i_data[1] = 0;
1414                 } else {
1415                         fcb->i_data[0] = 0;
1416                         fcb->i_data[1] =
1417                                 cpu_to_le32(new_encode_dev(inode->i_rdev));
1418                         fcb->i_data[2] = 0;
1419                 }
1420         } else
1421                 memcpy(fcb->i_data, oi->i_data, sizeof(fcb->i_data));
1422
1423         ret = ore_get_io_state(&sbi->layout, &oi->oc, &ios);
1424         if (unlikely(ret)) {
1425                 EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__);
1426                 goto free_args;
1427         }
1428
1429         attr = g_attr_inode_data;
1430         attr.val_ptr = fcb;
1431         ios->out_attr_len = 1;
1432         ios->out_attr = &attr;
1433
1434         wait_obj_created(oi);
1435
1436         if (!do_sync) {
1437                 args->sbi = sbi;
1438                 ios->done = updatei_done;
1439                 ios->private = args;
1440         }
1441
1442         ret = ore_write(ios);
1443         if (!do_sync && !ret) {
1444                 atomic_inc(&sbi->s_curr_pending);
1445                 goto out; /* deallocation in updatei_done */
1446         }
1447
1448         ore_put_io_state(ios);
1449 free_args:
1450         kfree(args);
1451 out:
1452         EXOFS_DBGMSG("(0x%lx) do_sync=%d ret=>%d\n",
1453                      inode->i_ino, do_sync, ret);
1454         return ret;
1455 }
1456
1457 int exofs_write_inode(struct inode *inode, struct writeback_control *wbc)
1458 {
1459         /* FIXME: fix fsync and use wbc->sync_mode == WB_SYNC_ALL */
1460         return exofs_update_inode(inode, 1);
1461 }
1462
1463 /*
1464  * Callback function from exofs_delete_inode() - don't have much cleaning up to
1465  * do.
1466  */
1467 static void delete_done(struct ore_io_state *ios, void *p)
1468 {
1469         struct exofs_sb_info *sbi = p;
1470
1471         ore_put_io_state(ios);
1472
1473         atomic_dec(&sbi->s_curr_pending);
1474 }
1475
1476 /*
1477  * Called when the refcount of an inode reaches zero.  We remove the object
1478  * from the OSD here.  We make sure the object was created before we try and
1479  * delete it.
1480  */
1481 void exofs_evict_inode(struct inode *inode)
1482 {
1483         struct exofs_i_info *oi = exofs_i(inode);
1484         struct super_block *sb = inode->i_sb;
1485         struct exofs_sb_info *sbi = sb->s_fs_info;
1486         struct ore_io_state *ios;
1487         int ret;
1488
1489         truncate_inode_pages(&inode->i_data, 0);
1490
1491         /* TODO: should do better here */
1492         if (inode->i_nlink || is_bad_inode(inode))
1493                 goto no_delete;
1494
1495         inode->i_size = 0;
1496         clear_inode(inode);
1497
1498         /* if we are deleting an obj that hasn't been created yet, wait.
1499          * This also makes sure that create_done cannot be called with an
1500          * already evicted inode.
1501          */
1502         wait_obj_created(oi);
1503         /* ignore the error, attempt a remove anyway */
1504
1505         /* Now Remove the OSD objects */
1506         ret = ore_get_io_state(&sbi->layout, &oi->oc, &ios);
1507         if (unlikely(ret)) {
1508                 EXOFS_ERR("%s: ore_get_io_state failed\n", __func__);
1509                 return;
1510         }
1511
1512         ios->done = delete_done;
1513         ios->private = sbi;
1514
1515         ret = ore_remove(ios);
1516         if (ret) {
1517                 EXOFS_ERR("%s: ore_remove failed\n", __func__);
1518                 ore_put_io_state(ios);
1519                 return;
1520         }
1521         atomic_inc(&sbi->s_curr_pending);
1522
1523         return;
1524
1525 no_delete:
1526         clear_inode(inode);
1527 }