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->read_4_write)
 365                 _unlock_pcol_pages(pcol, ret, READ);
 366
 367         pcol_free(pcol);
 368
 369         kfree(pcol_copy);
 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_DBGMSG("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_DBGMSG("index=0x%lx uptodate=%d\n", index, *uptodate);
 600                 return page;
 601         } else {
 602                 EXOFS_DBGMSG("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_DBGMSG("index=0x%lx\n", page->index);
 615                 page_cache_release(page);
 616                 return;
 617         }
 618         EXOFS_DBGMSG("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         _unlock_pcol_pages(pcol, ret, WRITE);
 680         pcol_free(pcol);
 681         kfree(pcol_copy);
 682
 683         return ret;
 684 }
 685
 686 /* writepage_strip is called either directly from writepage() or by the VFS from
 687  * within write_cache_pages(), to add one more page to be written to storage.
 688  * It will try to collect as many contiguous pages as possible. If a
 689  * discontinuity is encountered or it runs out of resources it will submit the
 690  * previous segment and will start a new collection.
 691  * Eventually caller must submit the last segment if present.
 692  */
 693 static int writepage_strip(struct page *page,
 694                            struct writeback_control *wbc_unused, void *data)
 695 {
 696         struct page_collect *pcol = data;
 697         struct inode *inode = pcol->inode;
 698         struct exofs_i_info *oi = exofs_i(inode);
 699         loff_t i_size = i_size_read(inode);
 700         pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
 701         size_t len;
 702         int ret;
 703
 704         BUG_ON(!PageLocked(page));
 705
 706         ret = wait_obj_created(oi);
 707         if (unlikely(ret))
 708                 goto fail;
 709
 710         if (page->index < end_index)
 711                 /* in this case, the page is within the limits of the file */
 712                 len = PAGE_CACHE_SIZE;
 713         else {
 714                 len = i_size & ~PAGE_CACHE_MASK;
 715
 716                 if (page->index > end_index || !len) {
 717                         /* in this case, the page is outside the limits
 718                          * (truncate in progress)
 719                          */
 720                         ret = write_exec(pcol);
 721                         if (unlikely(ret))
 722                                 goto fail;
 723                         if (PageError(page))
 724                                 ClearPageError(page);
 725                         unlock_page(page);
 726                         EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) "
 727                                      "outside the limits\n",
 728                                      inode->i_ino, page->index);
 729                         return 0;
 730                 }
 731         }
 732
 733 try_again:
 734
 735         if (unlikely(pcol->pg_first == -1)) {
 736                 pcol->pg_first = page->index;
 737         } else if (unlikely((pcol->pg_first + pcol->nr_pages) !=
 738                    page->index)) {
 739                 /* Discontinuity detected, split the request */
 740                 ret = write_exec(pcol);
 741                 if (unlikely(ret))
 742                         goto fail;
 743
 744                 EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) Discontinuity\n",
 745                              inode->i_ino, page->index);
 746                 goto try_again;
 747         }
 748
 749         if (!pcol->pages) {
 750                 ret = pcol_try_alloc(pcol);
 751                 if (unlikely(ret))
 752                         goto fail;
 753         }
 754
 755         EXOFS_DBGMSG2("    writepage_strip(0x%lx, 0x%lx) len=0x%zx\n",
 756                      inode->i_ino, page->index, len);
 757
 758         ret = pcol_add_page(pcol, page, len);
 759         if (unlikely(ret)) {
 760                 EXOFS_DBGMSG2("Failed pcol_add_page "
 761                              "nr_pages=%u total_length=0x%lx\n",
 762                              pcol->nr_pages, pcol->length);
 763
 764                 /* split the request, next loop will start again */
 765                 ret = write_exec(pcol);
 766                 if (unlikely(ret)) {
 767                         EXOFS_DBGMSG("write_exec failed => %d", ret);
 768                         goto fail;
 769                 }
 770
 771                 goto try_again;
 772         }
 773
 774         BUG_ON(PageWriteback(page));
 775         set_page_writeback(page);
 776
 777         return 0;
 778
 779 fail:
 780         EXOFS_DBGMSG("Error: writepage_strip(0x%lx, 0x%lx)=>%d\n",
 781                      inode->i_ino, page->index, ret);
 782         set_bit(AS_EIO, &page->mapping->flags);
 783         unlock_page(page);
 784         return ret;
 785 }
 786
 787 static int exofs_writepages(struct address_space *mapping,
 788                        struct writeback_control *wbc)
 789 {
 790         struct page_collect pcol;
 791         long start, end, expected_pages;
 792         int ret;
 793
 794         start = wbc->range_start >> PAGE_CACHE_SHIFT;
 795         end = (wbc->range_end == LLONG_MAX) ?
 796                         start + mapping->nrpages :
 797                         wbc->range_end >> PAGE_CACHE_SHIFT;
 798
 799         if (start || end)
 800                 expected_pages = end - start + 1;
 801         else
 802                 expected_pages = mapping->nrpages;
 803
 804         if (expected_pages < 32L)
 805                 expected_pages = 32L;
 806
 807         EXOFS_DBGMSG2("inode(0x%lx) wbc->start=0x%llx wbc->end=0x%llx "
 808                      "nrpages=%lu start=0x%lx end=0x%lx expected_pages=%ld\n",
 809                      mapping->host->i_ino, wbc->range_start, wbc->range_end,
 810                      mapping->nrpages, start, end, expected_pages);
 811
 812         _pcol_init(&pcol, expected_pages, mapping->host);
 813
 814         ret = write_cache_pages(mapping, wbc, writepage_strip, &pcol);
 815         if (unlikely(ret)) {
 816                 EXOFS_ERR("write_cache_pages => %d\n", ret);
 817                 return ret;
 818         }
 819
 820         ret = write_exec(&pcol);
 821         if (unlikely(ret))
 822                 return ret;
 823
 824         if (wbc->sync_mode == WB_SYNC_ALL) {
 825                 return write_exec(&pcol); /* pump the last reminder */
 826         } else if (pcol.nr_pages) {
 827                 /* not SYNC let the reminder join the next writeout */
 828                 unsigned i;
 829
 830                 for (i = 0; i < pcol.nr_pages; i++) {
 831                         struct page *page = pcol.pages[i];
 832
 833                         end_page_writeback(page);
 834                         set_page_dirty(page);
 835                         unlock_page(page);
 836                 }
 837         }
 838         return 0;
 839 }
 840
 841 /*
 842 static int exofs_writepage(struct page *page, struct writeback_control *wbc)
 843 {
 844         struct page_collect pcol;
 845         int ret;
 846
 847         _pcol_init(&pcol, 1, page->mapping->host);
 848
 849         ret = writepage_strip(page, NULL, &pcol);
 850         if (ret) {
 851                 EXOFS_ERR("exofs_writepage => %d\n", ret);
 852                 return ret;
 853         }
 854
 855         return write_exec(&pcol);
 856 }
 857 */
 858 /* i_mutex held using inode->i_size directly */
 859 static void _write_failed(struct inode *inode, loff_t to)
 860 {
 861         if (to > inode->i_size)
 862                 truncate_pagecache(inode, to, inode->i_size);
 863 }
 864
 865 int exofs_write_begin(struct file *file, struct address_space *mapping,
 866                 loff_t pos, unsigned len, unsigned flags,
 867                 struct page **pagep, void **fsdata)
 868 {
 869         int ret = 0;
 870         struct page *page;
 871
 872         page = *pagep;
 873         if (page == NULL) {
 874                 ret = simple_write_begin(file, mapping, pos, len, flags, pagep,
 875                                          fsdata);
 876                 if (ret) {
 877                         EXOFS_DBGMSG("simple_write_begin failed\n");
 878                         goto out;
 879                 }
 880
 881                 page = *pagep;
 882         }
 883
 884          /* read modify write */
 885         if (!PageUptodate(page) && (len != PAGE_CACHE_SIZE)) {
 886                 loff_t i_size = i_size_read(mapping->host);
 887                 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
 888                 size_t rlen;
 889
 890                 if (page->index < end_index)
 891                         rlen = PAGE_CACHE_SIZE;
 892                 else if (page->index == end_index)
 893                         rlen = i_size & ~PAGE_CACHE_MASK;
 894                 else
 895                         rlen = 0;
 896
 897                 if (!rlen) {
 898                         clear_highpage(page);
 899                         SetPageUptodate(page);
 900                         goto out;
 901                 }
 902
 903                 ret = _readpage(page, true);
 904                 if (ret) {
 905                         /*SetPageError was done by _readpage. Is it ok?*/
 906                         unlock_page(page);
 907                         EXOFS_DBGMSG("__readpage failed\n");
 908                 }
 909         }
 910 out:
 911         if (unlikely(ret))
 912                 _write_failed(mapping->host, pos + len);
 913
 914         return ret;
 915 }
 916
 917 static int exofs_write_begin_export(struct file *file,
 918                 struct address_space *mapping,
 919                 loff_t pos, unsigned len, unsigned flags,
 920                 struct page **pagep, void **fsdata)
 921 {
 922         *pagep = NULL;
 923
 924         return exofs_write_begin(file, mapping, pos, len, flags, pagep,
 925                                         fsdata);
 926 }
 927
 928 static int exofs_write_end(struct file *file, struct address_space *mapping,
 929                         loff_t pos, unsigned len, unsigned copied,
 930                         struct page *page, void *fsdata)
 931 {
 932         struct inode *inode = mapping->host;
 933         /* According to comment in simple_write_end i_mutex is held */
 934         loff_t i_size = inode->i_size;
 935         int ret;
 936
 937         ret = simple_write_end(file, mapping,pos, len, copied, page, fsdata);
 938         if (unlikely(ret))
 939                 _write_failed(inode, pos + len);
 940
 941         /* TODO: once simple_write_end marks inode dirty remove */
 942         if (i_size != inode->i_size)
 943                 mark_inode_dirty(inode);
 944         return ret;
 945 }
 946
 947 static int exofs_releasepage(struct page *page, gfp_t gfp)
 948 {
 949         EXOFS_DBGMSG("page 0x%lx\n", page->index);
 950         WARN_ON(1);
 951         return 0;
 952 }
 953
 954 static void exofs_invalidatepage(struct page *page, unsigned long offset)
 955 {
 956         EXOFS_DBGMSG("page 0x%lx offset 0x%lx\n", page->index, offset);
 957         WARN_ON(1);
 958 }
 959
 960 const struct address_space_operations exofs_aops = {
 961         .readpage       = exofs_readpage,
 962         .readpages      = exofs_readpages,
 963         .writepage      = NULL,
 964         .writepages     = exofs_writepages,
 965         .write_begin    = exofs_write_begin_export,
 966         .write_end      = exofs_write_end,
 967         .releasepage    = exofs_releasepage,
 968         .set_page_dirty = __set_page_dirty_nobuffers,
 969         .invalidatepage = exofs_invalidatepage,
 970
 971         /* Not implemented Yet */
 972         .bmap           = NULL, /* TODO: use osd's OSD_ACT_READ_MAP */
 973         .direct_IO      = NULL, /* TODO: Should be trivial to do */
 974
 975         /* With these NULL has special meaning or default is not exported */
 976         .get_xip_mem    = NULL,
 977         .migratepage    = NULL,
 978         .launder_page   = NULL,
 979         .is_partially_uptodate = NULL,
 980         .error_remove_page = NULL,
 981 };
 982
 983 /******************************************************************************
 984  * INODE OPERATIONS
 985  *****************************************************************************/
 986
 987 /*
 988  * Test whether an inode is a fast symlink.
 989  */
 990 static inline int exofs_inode_is_fast_symlink(struct inode *inode)
 991 {
 992         struct exofs_i_info *oi = exofs_i(inode);
 993
 994         return S_ISLNK(inode->i_mode) && (oi->i_data[0] != 0);
 995 }
 996
 997 static int _do_truncate(struct inode *inode, loff_t newsize)
 998 {
 999         struct exofs_i_info *oi = exofs_i(inode);
1000         struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
1001         int ret;
1002
1003         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1004
1005         ret = ore_truncate(&sbi->layout, &oi->oc, (u64)newsize);
1006         if (likely(!ret))
1007                 truncate_setsize(inode, newsize);
1008
1009         EXOFS_DBGMSG("(0x%lx) size=0x%llx ret=>%d\n",
1010                      inode->i_ino, newsize, ret);
1011         return ret;
1012 }
1013
1014 /*
1015  * Set inode attributes - update size attribute on OSD if needed,
1016  *                        otherwise just call generic functions.
1017  */
1018 int exofs_setattr(struct dentry *dentry, struct iattr *iattr)
1019 {
1020         struct inode *inode = dentry->d_inode;
1021         int error;
1022
1023         /* if we are about to modify an object, and it hasn't been
1024          * created yet, wait
1025          */
1026         error = wait_obj_created(exofs_i(inode));
1027         if (unlikely(error))
1028                 return error;
1029
1030         error = inode_change_ok(inode, iattr);
1031         if (unlikely(error))
1032                 return error;
1033
1034         if ((iattr->ia_valid & ATTR_SIZE) &&
1035             iattr->ia_size != i_size_read(inode)) {
1036                 error = _do_truncate(inode, iattr->ia_size);
1037                 if (unlikely(error))
1038                         return error;
1039         }
1040
1041         setattr_copy(inode, iattr);
1042         mark_inode_dirty(inode);
1043         return 0;
1044 }
1045
1046 static const struct osd_attr g_attr_inode_file_layout = ATTR_DEF(
1047         EXOFS_APAGE_FS_DATA,
1048         EXOFS_ATTR_INODE_FILE_LAYOUT,
1049         0);
1050 static const struct osd_attr g_attr_inode_dir_layout = ATTR_DEF(
1051         EXOFS_APAGE_FS_DATA,
1052         EXOFS_ATTR_INODE_DIR_LAYOUT,
1053         0);
1054
1055 /*
1056  * Read the Linux inode info from the OSD, and return it as is. In exofs the
1057  * inode info is in an application specific page/attribute of the osd-object.
1058  */
1059 static int exofs_get_inode(struct super_block *sb, struct exofs_i_info *oi,
1060                     struct exofs_fcb *inode)
1061 {
1062         struct exofs_sb_info *sbi = sb->s_fs_info;
1063         struct osd_attr attrs[] = {
1064                 [0] = g_attr_inode_data,
1065                 [1] = g_attr_inode_file_layout,
1066                 [2] = g_attr_inode_dir_layout,
1067         };
1068         struct ore_io_state *ios;
1069         struct exofs_on_disk_inode_layout *layout;
1070         int ret;
1071
1072         ret = ore_get_io_state(&sbi->layout, &oi->oc, &ios);
1073         if (unlikely(ret)) {
1074                 EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__);
1075                 return ret;
1076         }
1077
1078         attrs[1].len = exofs_on_disk_inode_layout_size(sbi->oc.numdevs);
1079         attrs[2].len = exofs_on_disk_inode_layout_size(sbi->oc.numdevs);
1080
1081         ios->in_attr = attrs;
1082         ios->in_attr_len = ARRAY_SIZE(attrs);
1083
1084         ret = ore_read(ios);
1085         if (unlikely(ret)) {
1086                 EXOFS_ERR("object(0x%llx) corrupted, return empty file=>%d\n",
1087                           _LLU(oi->one_comp.obj.id), ret);
1088                 memset(inode, 0, sizeof(*inode));
1089                 inode->i_mode = 0040000 | (0777 & ~022);
1090                 /* If object is lost on target we might as well enable it's
1091                  * delete.
1092                  */
1093                 if ((ret == -ENOENT) || (ret == -EINVAL))
1094                         ret = 0;
1095                 goto out;
1096         }
1097
1098         ret = extract_attr_from_ios(ios, &attrs[0]);
1099         if (ret) {
1100                 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
1101                 goto out;
1102         }
1103         WARN_ON(attrs[0].len != EXOFS_INO_ATTR_SIZE);
1104         memcpy(inode, attrs[0].val_ptr, EXOFS_INO_ATTR_SIZE);
1105
1106         ret = extract_attr_from_ios(ios, &attrs[1]);
1107         if (ret) {
1108                 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
1109                 goto out;
1110         }
1111         if (attrs[1].len) {
1112                 layout = attrs[1].val_ptr;
1113                 if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) {
1114                         EXOFS_ERR("%s: unsupported files layout %d\n",
1115                                 __func__, layout->gen_func);
1116                         ret = -ENOTSUPP;
1117                         goto out;
1118                 }
1119         }
1120
1121         ret = extract_attr_from_ios(ios, &attrs[2]);
1122         if (ret) {
1123                 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
1124                 goto out;
1125         }
1126         if (attrs[2].len) {
1127                 layout = attrs[2].val_ptr;
1128                 if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) {
1129                         EXOFS_ERR("%s: unsupported meta-data layout %d\n",
1130                                 __func__, layout->gen_func);
1131                         ret = -ENOTSUPP;
1132                         goto out;
1133                 }
1134         }
1135
1136 out:
1137         ore_put_io_state(ios);
1138         return ret;
1139 }
1140
1141 static void __oi_init(struct exofs_i_info *oi)
1142 {
1143         init_waitqueue_head(&oi->i_wq);
1144         oi->i_flags = 0;
1145 }
1146 /*
1147  * Fill in an inode read from the OSD and set it up for use
1148  */
1149 struct inode *exofs_iget(struct super_block *sb, unsigned long ino)
1150 {
1151         struct exofs_i_info *oi;
1152         struct exofs_fcb fcb;
1153         struct inode *inode;
1154         int ret;
1155
1156         inode = iget_locked(sb, ino);
1157         if (!inode)
1158                 return ERR_PTR(-ENOMEM);
1159         if (!(inode->i_state & I_NEW))
1160                 return inode;
1161         oi = exofs_i(inode);
1162         __oi_init(oi);
1163         exofs_init_comps(&oi->oc, &oi->one_comp, sb->s_fs_info,
1164                          exofs_oi_objno(oi));
1165
1166         /* read the inode from the osd */
1167         ret = exofs_get_inode(sb, oi, &fcb);
1168         if (ret)
1169                 goto bad_inode;
1170
1171         set_obj_created(oi);
1172
1173         /* copy stuff from on-disk struct to in-memory struct */
1174         inode->i_mode = le16_to_cpu(fcb.i_mode);
1175         i_uid_write(inode, le32_to_cpu(fcb.i_uid));
1176         i_gid_write(inode, le32_to_cpu(fcb.i_gid));
1177         set_nlink(inode, le16_to_cpu(fcb.i_links_count));
1178         inode->i_ctime.tv_sec = (signed)le32_to_cpu(fcb.i_ctime);
1179         inode->i_atime.tv_sec = (signed)le32_to_cpu(fcb.i_atime);
1180         inode->i_mtime.tv_sec = (signed)le32_to_cpu(fcb.i_mtime);
1181         inode->i_ctime.tv_nsec =
1182                 inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = 0;
1183         oi->i_commit_size = le64_to_cpu(fcb.i_size);
1184         i_size_write(inode, oi->i_commit_size);
1185         inode->i_blkbits = EXOFS_BLKSHIFT;
1186         inode->i_generation = le32_to_cpu(fcb.i_generation);
1187
1188         oi->i_dir_start_lookup = 0;
1189
1190         if ((inode->i_nlink == 0) && (inode->i_mode == 0)) {
1191                 ret = -ESTALE;
1192                 goto bad_inode;
1193         }
1194
1195         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1196                 if (fcb.i_data[0])
1197                         inode->i_rdev =
1198                                 old_decode_dev(le32_to_cpu(fcb.i_data[0]));
1199                 else
1200                         inode->i_rdev =
1201                                 new_decode_dev(le32_to_cpu(fcb.i_data[1]));
1202         } else {
1203                 memcpy(oi->i_data, fcb.i_data, sizeof(fcb.i_data));
1204         }
1205
1206         inode->i_mapping->backing_dev_info = sb->s_bdi;
1207         if (S_ISREG(inode->i_mode)) {
1208                 inode->i_op = &exofs_file_inode_operations;
1209                 inode->i_fop = &exofs_file_operations;
1210                 inode->i_mapping->a_ops = &exofs_aops;
1211         } else if (S_ISDIR(inode->i_mode)) {
1212                 inode->i_op = &exofs_dir_inode_operations;
1213                 inode->i_fop = &exofs_dir_operations;
1214                 inode->i_mapping->a_ops = &exofs_aops;
1215         } else if (S_ISLNK(inode->i_mode)) {
1216                 if (exofs_inode_is_fast_symlink(inode))
1217                         inode->i_op = &exofs_fast_symlink_inode_operations;
1218                 else {
1219                         inode->i_op = &exofs_symlink_inode_operations;
1220                         inode->i_mapping->a_ops = &exofs_aops;
1221                 }
1222         } else {
1223                 inode->i_op = &exofs_special_inode_operations;
1224                 if (fcb.i_data[0])
1225                         init_special_inode(inode, inode->i_mode,
1226                            old_decode_dev(le32_to_cpu(fcb.i_data[0])));
1227                 else
1228                         init_special_inode(inode, inode->i_mode,
1229                            new_decode_dev(le32_to_cpu(fcb.i_data[1])));
1230         }
1231
1232         unlock_new_inode(inode);
1233         return inode;
1234
1235 bad_inode:
1236         iget_failed(inode);
1237         return ERR_PTR(ret);
1238 }
1239
1240 int __exofs_wait_obj_created(struct exofs_i_info *oi)
1241 {
1242         if (!obj_created(oi)) {
1243                 EXOFS_DBGMSG("!obj_created\n");
1244                 BUG_ON(!obj_2bcreated(oi));
1245                 wait_event(oi->i_wq, obj_created(oi));
1246                 EXOFS_DBGMSG("wait_event done\n");
1247         }
1248         return unlikely(is_bad_inode(&oi->vfs_inode)) ? -EIO : 0;
1249 }
1250
1251 /*
1252  * Callback function from exofs_new_inode().  The important thing is that we
1253  * set the obj_created flag so that other methods know that the object exists on
1254  * the OSD.
1255  */
1256 static void create_done(struct ore_io_state *ios, void *p)
1257 {
1258         struct inode *inode = p;
1259         struct exofs_i_info *oi = exofs_i(inode);
1260         struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
1261         int ret;
1262
1263         ret = ore_check_io(ios, NULL);
1264         ore_put_io_state(ios);
1265
1266         atomic_dec(&sbi->s_curr_pending);
1267
1268         if (unlikely(ret)) {
1269                 EXOFS_ERR("object=0x%llx creation failed in pid=0x%llx",
1270                           _LLU(exofs_oi_objno(oi)),
1271                           _LLU(oi->one_comp.obj.partition));
1272                 /*TODO: When FS is corrupted creation can fail, object already
1273                  * exist. Get rid of this asynchronous creation, if exist
1274                  * increment the obj counter and try the next object. Until we
1275                  * succeed. All these dangling objects will be made into lost
1276                  * files by chkfs.exofs
1277                  */
1278         }
1279
1280         set_obj_created(oi);
1281
1282         wake_up(&oi->i_wq);
1283 }
1284
1285 /*
1286  * Set up a new inode and create an object for it on the OSD
1287  */
1288 struct inode *exofs_new_inode(struct inode *dir, umode_t mode)
1289 {
1290         struct super_block *sb = dir->i_sb;
1291         struct exofs_sb_info *sbi = sb->s_fs_info;
1292         struct inode *inode;
1293         struct exofs_i_info *oi;
1294         struct ore_io_state *ios;
1295         int ret;
1296
1297         inode = new_inode(sb);
1298         if (!inode)
1299                 return ERR_PTR(-ENOMEM);
1300
1301         oi = exofs_i(inode);
1302         __oi_init(oi);
1303
1304         set_obj_2bcreated(oi);
1305
1306         inode->i_mapping->backing_dev_info = sb->s_bdi;
1307         inode_init_owner(inode, dir, mode);
1308         inode->i_ino = sbi->s_nextid++;
1309         inode->i_blkbits = EXOFS_BLKSHIFT;
1310         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1311         oi->i_commit_size = inode->i_size = 0;
1312         spin_lock(&sbi->s_next_gen_lock);
1313         inode->i_generation = sbi->s_next_generation++;
1314         spin_unlock(&sbi->s_next_gen_lock);
1315         insert_inode_hash(inode);
1316
1317         exofs_init_comps(&oi->oc, &oi->one_comp, sb->s_fs_info,
1318                          exofs_oi_objno(oi));
1319         exofs_sbi_write_stats(sbi); /* Make sure new sbi->s_nextid is on disk */
1320
1321         mark_inode_dirty(inode);
1322
1323         ret = ore_get_io_state(&sbi->layout, &oi->oc, &ios);
1324         if (unlikely(ret)) {
1325                 EXOFS_ERR("exofs_new_inode: ore_get_io_state failed\n");
1326                 return ERR_PTR(ret);
1327         }
1328
1329         ios->done = create_done;
1330         ios->private = inode;
1331
1332         ret = ore_create(ios);
1333         if (ret) {
1334                 ore_put_io_state(ios);
1335                 return ERR_PTR(ret);
1336         }
1337         atomic_inc(&sbi->s_curr_pending);
1338
1339         return inode;
1340 }
1341
1342 /*
1343  * struct to pass two arguments to update_inode's callback
1344  */
1345 struct updatei_args {
1346         struct exofs_sb_info    *sbi;
1347         struct exofs_fcb        fcb;
1348 };
1349
1350 /*
1351  * Callback function from exofs_update_inode().
1352  */
1353 static void updatei_done(struct ore_io_state *ios, void *p)
1354 {
1355         struct updatei_args *args = p;
1356
1357         ore_put_io_state(ios);
1358
1359         atomic_dec(&args->sbi->s_curr_pending);
1360
1361         kfree(args);
1362 }
1363
1364 /*
1365  * Write the inode to the OSD.  Just fill up the struct, and set the attribute
1366  * synchronously or asynchronously depending on the do_sync flag.
1367  */
1368 static int exofs_update_inode(struct inode *inode, int do_sync)
1369 {
1370         struct exofs_i_info *oi = exofs_i(inode);
1371         struct super_block *sb = inode->i_sb;
1372         struct exofs_sb_info *sbi = sb->s_fs_info;
1373         struct ore_io_state *ios;
1374         struct osd_attr attr;
1375         struct exofs_fcb *fcb;
1376         struct updatei_args *args;
1377         int ret;
1378
1379         args = kzalloc(sizeof(*args), GFP_KERNEL);
1380         if (!args) {
1381                 EXOFS_DBGMSG("Failed kzalloc of args\n");
1382                 return -ENOMEM;
1383         }
1384
1385         fcb = &args->fcb;
1386
1387         fcb->i_mode = cpu_to_le16(inode->i_mode);
1388         fcb->i_uid = cpu_to_le32(i_uid_read(inode));
1389         fcb->i_gid = cpu_to_le32(i_gid_read(inode));
1390         fcb->i_links_count = cpu_to_le16(inode->i_nlink);
1391         fcb->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1392         fcb->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1393         fcb->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1394         oi->i_commit_size = i_size_read(inode);
1395         fcb->i_size = cpu_to_le64(oi->i_commit_size);
1396         fcb->i_generation = cpu_to_le32(inode->i_generation);
1397
1398         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1399                 if (old_valid_dev(inode->i_rdev)) {
1400                         fcb->i_data[0] =
1401                                 cpu_to_le32(old_encode_dev(inode->i_rdev));
1402                         fcb->i_data[1] = 0;
1403                 } else {
1404                         fcb->i_data[0] = 0;
1405                         fcb->i_data[1] =
1406                                 cpu_to_le32(new_encode_dev(inode->i_rdev));
1407                         fcb->i_data[2] = 0;
1408                 }
1409         } else
1410                 memcpy(fcb->i_data, oi->i_data, sizeof(fcb->i_data));
1411
1412         ret = ore_get_io_state(&sbi->layout, &oi->oc, &ios);
1413         if (unlikely(ret)) {
1414                 EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__);
1415                 goto free_args;
1416         }
1417
1418         attr = g_attr_inode_data;
1419         attr.val_ptr = fcb;
1420         ios->out_attr_len = 1;
1421         ios->out_attr = &attr;
1422
1423         wait_obj_created(oi);
1424
1425         if (!do_sync) {
1426                 args->sbi = sbi;
1427                 ios->done = updatei_done;
1428                 ios->private = args;
1429         }
1430
1431         ret = ore_write(ios);
1432         if (!do_sync && !ret) {
1433                 atomic_inc(&sbi->s_curr_pending);
1434                 goto out; /* deallocation in updatei_done */
1435         }
1436
1437         ore_put_io_state(ios);
1438 free_args:
1439         kfree(args);
1440 out:
1441         EXOFS_DBGMSG("(0x%lx) do_sync=%d ret=>%d\n",
1442                      inode->i_ino, do_sync, ret);
1443         return ret;
1444 }
1445
1446 int exofs_write_inode(struct inode *inode, struct writeback_control *wbc)
1447 {
1448         /* FIXME: fix fsync and use wbc->sync_mode == WB_SYNC_ALL */
1449         return exofs_update_inode(inode, 1);
1450 }
1451
1452 /*
1453  * Callback function from exofs_delete_inode() - don't have much cleaning up to
1454  * do.
1455  */
1456 static void delete_done(struct ore_io_state *ios, void *p)
1457 {
1458         struct exofs_sb_info *sbi = p;
1459
1460         ore_put_io_state(ios);
1461
1462         atomic_dec(&sbi->s_curr_pending);
1463 }
1464
1465 /*
1466  * Called when the refcount of an inode reaches zero.  We remove the object
1467  * from the OSD here.  We make sure the object was created before we try and
1468  * delete it.
1469  */
1470 void exofs_evict_inode(struct inode *inode)
1471 {
1472         struct exofs_i_info *oi = exofs_i(inode);
1473         struct super_block *sb = inode->i_sb;
1474         struct exofs_sb_info *sbi = sb->s_fs_info;
1475         struct ore_io_state *ios;
1476         int ret;
1477
1478         truncate_inode_pages(&inode->i_data, 0);
1479
1480         /* TODO: should do better here */
1481         if (inode->i_nlink || is_bad_inode(inode))
1482                 goto no_delete;
1483
1484         inode->i_size = 0;
1485         clear_inode(inode);
1486
1487         /* if we are deleting an obj that hasn't been created yet, wait.
1488          * This also makes sure that create_done cannot be called with an
1489          * already evicted inode.
1490          */
1491         wait_obj_created(oi);
1492         /* ignore the error, attempt a remove anyway */
1493
1494         /* Now Remove the OSD objects */
1495         ret = ore_get_io_state(&sbi->layout, &oi->oc, &ios);
1496         if (unlikely(ret)) {
1497                 EXOFS_ERR("%s: ore_get_io_state failed\n", __func__);
1498                 return;
1499         }
1500
1501         ios->done = delete_done;
1502         ios->private = sbi;
1503
1504         ret = ore_remove(ios);
1505         if (ret) {
1506                 EXOFS_ERR("%s: ore_remove failed\n", __func__);
1507                 ore_put_io_state(ios);
1508                 return;
1509         }
1510         atomic_inc(&sbi->s_curr_pending);
1511
1512         return;
1513
1514 no_delete:
1515         clear_inode(inode);
1516 }