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/writeback.h>
  35 #include <linux/buffer_head.h>
  36 #include <scsi/scsi_device.h>
  37
  38 #include "exofs.h"
  39
  40 #ifdef CONFIG_EXOFS_DEBUG
  41 #  define EXOFS_DEBUG_OBJ_ISIZE 1
  42 #endif
  43
  44 #define EXOFS_DBGMSG2(M...) do {} while (0)
  45
  46 struct page_collect {
  47         struct exofs_sb_info *sbi;
  48         struct request_queue *req_q;
  49         struct inode *inode;
  50         unsigned expected_pages;
  51
  52         struct bio *bio;
  53         unsigned nr_pages;
  54         unsigned long length;
  55         loff_t pg_first; /* keep 64bit also in 32-arches */
  56 };
  57
  58 static void _pcol_init(struct page_collect *pcol, unsigned expected_pages,
  59                 struct inode *inode)
  60 {
  61         struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
  62
  63         pcol->sbi = sbi;
  64         pcol->req_q = osd_request_queue(sbi->s_dev);
  65         pcol->inode = inode;
  66         pcol->expected_pages = expected_pages;
  67
  68         pcol->bio = NULL;
  69         pcol->nr_pages = 0;
  70         pcol->length = 0;
  71         pcol->pg_first = -1;
  72
  73         EXOFS_DBGMSG("_pcol_init ino=0x%lx expected_pages=%u\n", inode->i_ino,
  74                      expected_pages);
  75 }
  76
  77 static void _pcol_reset(struct page_collect *pcol)
  78 {
  79         pcol->expected_pages -= min(pcol->nr_pages, pcol->expected_pages);
  80
  81         pcol->bio = NULL;
  82         pcol->nr_pages = 0;
  83         pcol->length = 0;
  84         pcol->pg_first = -1;
  85         EXOFS_DBGMSG("_pcol_reset ino=0x%lx expected_pages=%u\n",
  86                      pcol->inode->i_ino, pcol->expected_pages);
  87
  88         /* this is probably the end of the loop but in writes
  89          * it might not end here. don't be left with nothing
  90          */
  91         if (!pcol->expected_pages)
  92                 pcol->expected_pages = 128;
  93 }
  94
  95 static int pcol_try_alloc(struct page_collect *pcol)
  96 {
  97         int pages = min_t(unsigned, pcol->expected_pages, BIO_MAX_PAGES);
  98
  99         for (; pages; pages >>= 1) {
 100                 pcol->bio = bio_alloc(GFP_KERNEL, pages);
 101                 if (likely(pcol->bio))
 102                         return 0;
 103         }
 104
 105         EXOFS_ERR("Failed to kcalloc expected_pages=%u\n",
 106                   pcol->expected_pages);
 107         return -ENOMEM;
 108 }
 109
 110 static void pcol_free(struct page_collect *pcol)
 111 {
 112         bio_put(pcol->bio);
 113         pcol->bio = NULL;
 114 }
 115
 116 static int pcol_add_page(struct page_collect *pcol, struct page *page,
 117                          unsigned len)
 118 {
 119         int added_len = bio_add_pc_page(pcol->req_q, pcol->bio, page, len, 0);
 120         if (unlikely(len != added_len))
 121                 return -ENOMEM;
 122
 123         ++pcol->nr_pages;
 124         pcol->length += len;
 125         return 0;
 126 }
 127
 128 static int update_read_page(struct page *page, int ret)
 129 {
 130         if (ret == 0) {
 131                 /* Everything is OK */
 132                 SetPageUptodate(page);
 133                 if (PageError(page))
 134                         ClearPageError(page);
 135         } else if (ret == -EFAULT) {
 136                 /* In this case we were trying to read something that wasn't on
 137                  * disk yet - return a page full of zeroes.  This should be OK,
 138                  * because the object should be empty (if there was a write
 139                  * before this read, the read would be waiting with the page
 140                  * locked */
 141                 clear_highpage(page);
 142
 143                 SetPageUptodate(page);
 144                 if (PageError(page))
 145                         ClearPageError(page);
 146                 ret = 0; /* recovered error */
 147                 EXOFS_DBGMSG("recovered read error\n");
 148         } else /* Error */
 149                 SetPageError(page);
 150
 151         return ret;
 152 }
 153
 154 static void update_write_page(struct page *page, int ret)
 155 {
 156         if (ret) {
 157                 mapping_set_error(page->mapping, ret);
 158                 SetPageError(page);
 159         }
 160         end_page_writeback(page);
 161 }
 162
 163 /* Called at the end of reads, to optionally unlock pages and update their
 164  * status.
 165  */
 166 static int __readpages_done(struct osd_request *or, struct page_collect *pcol,
 167                             bool do_unlock)
 168 {
 169         struct bio_vec *bvec;
 170         int i;
 171         u64 resid;
 172         u64 good_bytes;
 173         u64 length = 0;
 174         int ret = exofs_check_ok_resid(or, &resid, NULL);
 175
 176         osd_end_request(or);
 177
 178         if (likely(!ret))
 179                 good_bytes = pcol->length;
 180         else if (!resid)
 181                 good_bytes = 0;
 182         else
 183                 good_bytes = pcol->length - resid;
 184
 185         EXOFS_DBGMSG("readpages_done(0x%lx) good_bytes=0x%llx"
 186                      " length=0x%lx nr_pages=%u\n",
 187                      pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
 188                      pcol->nr_pages);
 189
 190         __bio_for_each_segment(bvec, pcol->bio, i, 0) {
 191                 struct page *page = bvec->bv_page;
 192                 struct inode *inode = page->mapping->host;
 193                 int page_stat;
 194
 195                 if (inode != pcol->inode)
 196                         continue; /* osd might add more pages at end */
 197
 198                 if (likely(length < good_bytes))
 199                         page_stat = 0;
 200                 else
 201                         page_stat = ret;
 202
 203                 EXOFS_DBGMSG2("    readpages_done(0x%lx, 0x%lx) %s\n",
 204                           inode->i_ino, page->index,
 205                           page_stat ? "bad_bytes" : "good_bytes");
 206
 207                 ret = update_read_page(page, page_stat);
 208                 if (do_unlock)
 209                         unlock_page(page);
 210                 length += bvec->bv_len;
 211         }
 212
 213         pcol_free(pcol);
 214         EXOFS_DBGMSG("readpages_done END\n");
 215         return ret;
 216 }
 217
 218 /* callback of async reads */
 219 static void readpages_done(struct osd_request *or, void *p)
 220 {
 221         struct page_collect *pcol = p;
 222
 223         __readpages_done(or, pcol, true);
 224         atomic_dec(&pcol->sbi->s_curr_pending);
 225         kfree(p);
 226 }
 227
 228 static void _unlock_pcol_pages(struct page_collect *pcol, int ret, int rw)
 229 {
 230         struct bio_vec *bvec;
 231         int i;
 232
 233         __bio_for_each_segment(bvec, pcol->bio, i, 0) {
 234                 struct page *page = bvec->bv_page;
 235
 236                 if (rw == READ)
 237                         update_read_page(page, ret);
 238                 else
 239                         update_write_page(page, ret);
 240
 241                 unlock_page(page);
 242         }
 243         pcol_free(pcol);
 244 }
 245
 246 static int read_exec(struct page_collect *pcol, bool is_sync)
 247 {
 248         struct exofs_i_info *oi = exofs_i(pcol->inode);
 249         struct osd_obj_id obj = {pcol->sbi->s_pid,
 250                                         pcol->inode->i_ino + EXOFS_OBJ_OFF};
 251         struct osd_request *or = NULL;
 252         struct page_collect *pcol_copy = NULL;
 253         loff_t i_start = pcol->pg_first << PAGE_CACHE_SHIFT;
 254         int ret;
 255
 256         if (!pcol->bio)
 257                 return 0;
 258
 259         /* see comment in _readpage() about sync reads */
 260         WARN_ON(is_sync && (pcol->nr_pages != 1));
 261
 262         or = osd_start_request(pcol->sbi->s_dev, GFP_KERNEL);
 263         if (unlikely(!or)) {
 264                 ret = -ENOMEM;
 265                 goto err;
 266         }
 267
 268         osd_req_read(or, &obj, i_start, pcol->bio, pcol->length);
 269
 270         if (is_sync) {
 271                 exofs_sync_op(or, pcol->sbi->s_timeout, oi->i_cred);
 272                 return __readpages_done(or, pcol, false);
 273         }
 274
 275         pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
 276         if (!pcol_copy) {
 277                 ret = -ENOMEM;
 278                 goto err;
 279         }
 280
 281         *pcol_copy = *pcol;
 282         ret = exofs_async_op(or, readpages_done, pcol_copy, oi->i_cred);
 283         if (unlikely(ret))
 284                 goto err;
 285
 286         atomic_inc(&pcol->sbi->s_curr_pending);
 287
 288         EXOFS_DBGMSG("read_exec obj=0x%llx start=0x%llx length=0x%lx\n",
 289                   obj.id, _LLU(i_start), pcol->length);
 290
 291         /* pages ownership was passed to pcol_copy */
 292         _pcol_reset(pcol);
 293         return 0;
 294
 295 err:
 296         if (!is_sync)
 297                 _unlock_pcol_pages(pcol, ret, READ);
 298         else /* Pages unlocked by caller in sync mode only free bio */
 299                 pcol_free(pcol);
 300
 301         kfree(pcol_copy);
 302         if (or)
 303                 osd_end_request(or);
 304         return ret;
 305 }
 306
 307 /* readpage_strip is called either directly from readpage() or by the VFS from
 308  * within read_cache_pages(), to add one more page to be read. It will try to
 309  * collect as many contiguous pages as posible. If a discontinuity is
 310  * encountered, or it runs out of resources, it will submit the previous segment
 311  * and will start a new collection. Eventually caller must submit the last
 312  * segment if present.
 313  */
 314 static int readpage_strip(void *data, struct page *page)
 315 {
 316         struct page_collect *pcol = data;
 317         struct inode *inode = pcol->inode;
 318         struct exofs_i_info *oi = exofs_i(inode);
 319         loff_t i_size = i_size_read(inode);
 320         pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
 321         size_t len;
 322         int ret;
 323
 324         /* FIXME: Just for debugging, will be removed */
 325         if (PageUptodate(page))
 326                 EXOFS_ERR("PageUptodate(0x%lx, 0x%lx)\n", pcol->inode->i_ino,
 327                           page->index);
 328
 329         if (page->index < end_index)
 330                 len = PAGE_CACHE_SIZE;
 331         else if (page->index == end_index)
 332                 len = i_size & ~PAGE_CACHE_MASK;
 333         else
 334                 len = 0;
 335
 336         if (!len || !obj_created(oi)) {
 337                 /* this will be out of bounds, or doesn't exist yet.
 338                  * Current page is cleared and the request is split
 339                  */
 340                 clear_highpage(page);
 341
 342                 SetPageUptodate(page);
 343                 if (PageError(page))
 344                         ClearPageError(page);
 345
 346                 unlock_page(page);
 347                 EXOFS_DBGMSG("readpage_strip(0x%lx, 0x%lx) empty page,"
 348                              " splitting\n", inode->i_ino, page->index);
 349
 350                 return read_exec(pcol, false);
 351         }
 352
 353 try_again:
 354
 355         if (unlikely(pcol->pg_first == -1)) {
 356                 pcol->pg_first = page->index;
 357         } else if (unlikely((pcol->pg_first + pcol->nr_pages) !=
 358                    page->index)) {
 359                 /* Discontinuity detected, split the request */
 360                 ret = read_exec(pcol, false);
 361                 if (unlikely(ret))
 362                         goto fail;
 363                 goto try_again;
 364         }
 365
 366         if (!pcol->bio) {
 367                 ret = pcol_try_alloc(pcol);
 368                 if (unlikely(ret))
 369                         goto fail;
 370         }
 371
 372         if (len != PAGE_CACHE_SIZE)
 373                 zero_user(page, len, PAGE_CACHE_SIZE - len);
 374
 375         EXOFS_DBGMSG2("    readpage_strip(0x%lx, 0x%lx) len=0x%zx\n",
 376                      inode->i_ino, page->index, len);
 377
 378         ret = pcol_add_page(pcol, page, len);
 379         if (ret) {
 380                 EXOFS_DBGMSG2("Failed pcol_add_page pages[i]=%p "
 381                           "this_len=0x%zx nr_pages=%u length=0x%lx\n",
 382                           page, len, pcol->nr_pages, pcol->length);
 383
 384                 /* split the request, and start again with current page */
 385                 ret = read_exec(pcol, false);
 386                 if (unlikely(ret))
 387                         goto fail;
 388
 389                 goto try_again;
 390         }
 391
 392         return 0;
 393
 394 fail:
 395         /* SetPageError(page); ??? */
 396         unlock_page(page);
 397         return ret;
 398 }
 399
 400 static int exofs_readpages(struct file *file, struct address_space *mapping,
 401                            struct list_head *pages, unsigned nr_pages)
 402 {
 403         struct page_collect pcol;
 404         int ret;
 405
 406         _pcol_init(&pcol, nr_pages, mapping->host);
 407
 408         ret = read_cache_pages(mapping, pages, readpage_strip, &pcol);
 409         if (ret) {
 410                 EXOFS_ERR("read_cache_pages => %d\n", ret);
 411                 return ret;
 412         }
 413
 414         return read_exec(&pcol, false);
 415 }
 416
 417 static int _readpage(struct page *page, bool is_sync)
 418 {
 419         struct page_collect pcol;
 420         int ret;
 421
 422         _pcol_init(&pcol, 1, page->mapping->host);
 423
 424         /* readpage_strip might call read_exec(,async) inside at several places
 425          * but this is safe for is_async=0 since read_exec will not do anything
 426          * when we have a single page.
 427          */
 428         ret = readpage_strip(&pcol, page);
 429         if (ret) {
 430                 EXOFS_ERR("_readpage => %d\n", ret);
 431                 return ret;
 432         }
 433
 434         return read_exec(&pcol, is_sync);
 435 }
 436
 437 /*
 438  * We don't need the file
 439  */
 440 static int exofs_readpage(struct file *file, struct page *page)
 441 {
 442         return _readpage(page, false);
 443 }
 444
 445 /* Callback for osd_write. All writes are asynchronouse */
 446 static void writepages_done(struct osd_request *or, void *p)
 447 {
 448         struct page_collect *pcol = p;
 449         struct bio_vec *bvec;
 450         int i;
 451         u64 resid;
 452         u64  good_bytes;
 453         u64  length = 0;
 454
 455         int ret = exofs_check_ok_resid(or, NULL, &resid);
 456
 457         osd_end_request(or);
 458         atomic_dec(&pcol->sbi->s_curr_pending);
 459
 460         if (likely(!ret))
 461                 good_bytes = pcol->length;
 462         else if (!resid)
 463                 good_bytes = 0;
 464         else
 465                 good_bytes = pcol->length - resid;
 466
 467         EXOFS_DBGMSG("writepages_done(0x%lx) good_bytes=0x%llx"
 468                      " length=0x%lx nr_pages=%u\n",
 469                      pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
 470                      pcol->nr_pages);
 471
 472         __bio_for_each_segment(bvec, pcol->bio, i, 0) {
 473                 struct page *page = bvec->bv_page;
 474                 struct inode *inode = page->mapping->host;
 475                 int page_stat;
 476
 477                 if (inode != pcol->inode)
 478                         continue; /* osd might add more pages to a bio */
 479
 480                 if (likely(length < good_bytes))
 481                         page_stat = 0;
 482                 else
 483                         page_stat = ret;
 484
 485                 update_write_page(page, page_stat);
 486                 unlock_page(page);
 487                 EXOFS_DBGMSG2("    writepages_done(0x%lx, 0x%lx) status=%d\n",
 488                              inode->i_ino, page->index, page_stat);
 489
 490                 length += bvec->bv_len;
 491         }
 492
 493         pcol_free(pcol);
 494         kfree(pcol);
 495         EXOFS_DBGMSG("writepages_done END\n");
 496 }
 497
 498 static int write_exec(struct page_collect *pcol)
 499 {
 500         struct exofs_i_info *oi = exofs_i(pcol->inode);
 501         struct osd_obj_id obj = {pcol->sbi->s_pid,
 502                                         pcol->inode->i_ino + EXOFS_OBJ_OFF};
 503         struct osd_request *or = NULL;
 504         struct page_collect *pcol_copy = NULL;
 505         loff_t i_start = pcol->pg_first << PAGE_CACHE_SHIFT;
 506         int ret;
 507
 508         if (!pcol->bio)
 509                 return 0;
 510
 511         or = osd_start_request(pcol->sbi->s_dev, GFP_KERNEL);
 512         if (unlikely(!or)) {
 513                 EXOFS_ERR("write_exec: Faild to osd_start_request()\n");
 514                 ret = -ENOMEM;
 515                 goto err;
 516         }
 517
 518         pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
 519         if (!pcol_copy) {
 520                 EXOFS_ERR("write_exec: Faild to kmalloc(pcol)\n");
 521                 ret = -ENOMEM;
 522                 goto err;
 523         }
 524
 525         *pcol_copy = *pcol;
 526
 527         pcol_copy->bio->bi_rw |= (1 << BIO_RW); /* FIXME: bio_set_dir() */
 528         osd_req_write(or, &obj, i_start, pcol_copy->bio, pcol_copy->length);
 529         ret = exofs_async_op(or, writepages_done, pcol_copy, oi->i_cred);
 530         if (unlikely(ret)) {
 531                 EXOFS_ERR("write_exec: exofs_async_op() Faild\n");
 532                 goto err;
 533         }
 534
 535         atomic_inc(&pcol->sbi->s_curr_pending);
 536         EXOFS_DBGMSG("write_exec(0x%lx, 0x%llx) start=0x%llx length=0x%lx\n",
 537                   pcol->inode->i_ino, pcol->pg_first, _LLU(i_start),
 538                   pcol->length);
 539         /* pages ownership was passed to pcol_copy */
 540         _pcol_reset(pcol);
 541         return 0;
 542
 543 err:
 544         _unlock_pcol_pages(pcol, ret, WRITE);
 545         kfree(pcol_copy);
 546         if (or)
 547                 osd_end_request(or);
 548         return ret;
 549 }
 550
 551 /* writepage_strip is called either directly from writepage() or by the VFS from
 552  * within write_cache_pages(), to add one more page to be written to storage.
 553  * It will try to collect as many contiguous pages as possible. If a
 554  * discontinuity is encountered or it runs out of resources it will submit the
 555  * previous segment and will start a new collection.
 556  * Eventually caller must submit the last segment if present.
 557  */
 558 static int writepage_strip(struct page *page,
 559                            struct writeback_control *wbc_unused, void *data)
 560 {
 561         struct page_collect *pcol = data;
 562         struct inode *inode = pcol->inode;
 563         struct exofs_i_info *oi = exofs_i(inode);
 564         loff_t i_size = i_size_read(inode);
 565         pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
 566         size_t len;
 567         int ret;
 568
 569         BUG_ON(!PageLocked(page));
 570
 571         ret = wait_obj_created(oi);
 572         if (unlikely(ret))
 573                 goto fail;
 574
 575         if (page->index < end_index)
 576                 /* in this case, the page is within the limits of the file */
 577                 len = PAGE_CACHE_SIZE;
 578         else {
 579                 len = i_size & ~PAGE_CACHE_MASK;
 580
 581                 if (page->index > end_index || !len) {
 582                         /* in this case, the page is outside the limits
 583                          * (truncate in progress)
 584                          */
 585                         ret = write_exec(pcol);
 586                         if (unlikely(ret))
 587                                 goto fail;
 588                         if (PageError(page))
 589                                 ClearPageError(page);
 590                         unlock_page(page);
 591                         return 0;
 592                 }
 593         }
 594
 595 try_again:
 596
 597         if (unlikely(pcol->pg_first == -1)) {
 598                 pcol->pg_first = page->index;
 599         } else if (unlikely((pcol->pg_first + pcol->nr_pages) !=
 600                    page->index)) {
 601                 /* Discontinuity detected, split the request */
 602                 ret = write_exec(pcol);
 603                 if (unlikely(ret))
 604                         goto fail;
 605                 goto try_again;
 606         }
 607
 608         if (!pcol->bio) {
 609                 ret = pcol_try_alloc(pcol);
 610                 if (unlikely(ret))
 611                         goto fail;
 612         }
 613
 614         EXOFS_DBGMSG2("    writepage_strip(0x%lx, 0x%lx) len=0x%zx\n",
 615                      inode->i_ino, page->index, len);
 616
 617         ret = pcol_add_page(pcol, page, len);
 618         if (unlikely(ret)) {
 619                 EXOFS_DBGMSG("Failed pcol_add_page "
 620                              "nr_pages=%u total_length=0x%lx\n",
 621                              pcol->nr_pages, pcol->length);
 622
 623                 /* split the request, next loop will start again */
 624                 ret = write_exec(pcol);
 625                 if (unlikely(ret)) {
 626                         EXOFS_DBGMSG("write_exec faild => %d", ret);
 627                         goto fail;
 628                 }
 629
 630                 goto try_again;
 631         }
 632
 633         BUG_ON(PageWriteback(page));
 634         set_page_writeback(page);
 635
 636         return 0;
 637
 638 fail:
 639         set_bit(AS_EIO, &page->mapping->flags);
 640         unlock_page(page);
 641         return ret;
 642 }
 643
 644 static int exofs_writepages(struct address_space *mapping,
 645                        struct writeback_control *wbc)
 646 {
 647         struct page_collect pcol;
 648         long start, end, expected_pages;
 649         int ret;
 650
 651         start = wbc->range_start >> PAGE_CACHE_SHIFT;
 652         end = (wbc->range_end == LLONG_MAX) ?
 653                         start + mapping->nrpages :
 654                         wbc->range_end >> PAGE_CACHE_SHIFT;
 655
 656         if (start || end)
 657                 expected_pages = min(end - start + 1, 32L);
 658         else
 659                 expected_pages = mapping->nrpages;
 660
 661         EXOFS_DBGMSG("inode(0x%lx) wbc->start=0x%llx wbc->end=0x%llx"
 662                      " m->nrpages=%lu start=0x%lx end=0x%lx\n",
 663                      mapping->host->i_ino, wbc->range_start, wbc->range_end,
 664                      mapping->nrpages, start, end);
 665
 666         _pcol_init(&pcol, expected_pages, mapping->host);
 667
 668         ret = write_cache_pages(mapping, wbc, writepage_strip, &pcol);
 669         if (ret) {
 670                 EXOFS_ERR("write_cache_pages => %d\n", ret);
 671                 return ret;
 672         }
 673
 674         return write_exec(&pcol);
 675 }
 676
 677 static int exofs_writepage(struct page *page, struct writeback_control *wbc)
 678 {
 679         struct page_collect pcol;
 680         int ret;
 681
 682         _pcol_init(&pcol, 1, page->mapping->host);
 683
 684         ret = writepage_strip(page, NULL, &pcol);
 685         if (ret) {
 686                 EXOFS_ERR("exofs_writepage => %d\n", ret);
 687                 return ret;
 688         }
 689
 690         return write_exec(&pcol);
 691 }
 692
 693 int exofs_write_begin(struct file *file, struct address_space *mapping,
 694                 loff_t pos, unsigned len, unsigned flags,
 695                 struct page **pagep, void **fsdata)
 696 {
 697         int ret = 0;
 698         struct page *page;
 699
 700         page = *pagep;
 701         if (page == NULL) {
 702                 ret = simple_write_begin(file, mapping, pos, len, flags, pagep,
 703                                          fsdata);
 704                 if (ret) {
 705                         EXOFS_DBGMSG("simple_write_begin faild\n");
 706                         return ret;
 707                 }
 708
 709                 page = *pagep;
 710         }
 711
 712          /* read modify write */
 713         if (!PageUptodate(page) && (len != PAGE_CACHE_SIZE)) {
 714                 ret = _readpage(page, true);
 715                 if (ret) {
 716                         /*SetPageError was done by _readpage. Is it ok?*/
 717                         unlock_page(page);
 718                         EXOFS_DBGMSG("__readpage_filler faild\n");
 719                 }
 720         }
 721
 722         return ret;
 723 }
 724
 725 static int exofs_write_begin_export(struct file *file,
 726                 struct address_space *mapping,
 727                 loff_t pos, unsigned len, unsigned flags,
 728                 struct page **pagep, void **fsdata)
 729 {
 730         *pagep = NULL;
 731
 732         return exofs_write_begin(file, mapping, pos, len, flags, pagep,
 733                                         fsdata);
 734 }
 735
 736 const struct address_space_operations exofs_aops = {
 737         .readpage       = exofs_readpage,
 738         .readpages      = exofs_readpages,
 739         .writepage      = exofs_writepage,
 740         .writepages     = exofs_writepages,
 741         .write_begin    = exofs_write_begin_export,
 742         .write_end      = simple_write_end,
 743 };
 744
 745 /******************************************************************************
 746  * INODE OPERATIONS
 747  *****************************************************************************/
 748
 749 /*
 750  * Test whether an inode is a fast symlink.
 751  */
 752 static inline int exofs_inode_is_fast_symlink(struct inode *inode)
 753 {
 754         struct exofs_i_info *oi = exofs_i(inode);
 755
 756         return S_ISLNK(inode->i_mode) && (oi->i_data[0] != 0);
 757 }
 758
 759 /*
 760  * get_block_t - Fill in a buffer_head
 761  * An OSD takes care of block allocation so we just fake an allocation by
 762  * putting in the inode's sector_t in the buffer_head.
 763  * TODO: What about the case of create==0 and @iblock does not exist in the
 764  * object?
 765  */
 766 static int exofs_get_block(struct inode *inode, sector_t iblock,
 767                     struct buffer_head *bh_result, int create)
 768 {
 769         map_bh(bh_result, inode->i_sb, iblock);
 770         return 0;
 771 }
 772
 773 const struct osd_attr g_attr_logical_length = ATTR_DEF(
 774         OSD_APAGE_OBJECT_INFORMATION, OSD_ATTR_OI_LOGICAL_LENGTH, 8);
 775
 776 /*
 777  * Truncate a file to the specified size - all we have to do is set the size
 778  * attribute.  We make sure the object exists first.
 779  */
 780 void exofs_truncate(struct inode *inode)
 781 {
 782         struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
 783         struct exofs_i_info *oi = exofs_i(inode);
 784         struct osd_obj_id obj = {sbi->s_pid, inode->i_ino + EXOFS_OBJ_OFF};
 785         struct osd_request *or;
 786         struct osd_attr attr;
 787         loff_t isize = i_size_read(inode);
 788         __be64 newsize;
 789         int ret;
 790
 791         if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
 792              || S_ISLNK(inode->i_mode)))
 793                 return;
 794         if (exofs_inode_is_fast_symlink(inode))
 795                 return;
 796         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 797                 return;
 798         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 799
 800         nobh_truncate_page(inode->i_mapping, isize, exofs_get_block);
 801
 802         or = osd_start_request(sbi->s_dev, GFP_KERNEL);
 803         if (unlikely(!or)) {
 804                 EXOFS_ERR("ERROR: exofs_truncate: osd_start_request failed\n");
 805                 goto fail;
 806         }
 807
 808         osd_req_set_attributes(or, &obj);
 809
 810         newsize = cpu_to_be64((u64)isize);
 811         attr = g_attr_logical_length;
 812         attr.val_ptr = &newsize;
 813         osd_req_add_set_attr_list(or, &attr, 1);
 814
 815         /* if we are about to truncate an object, and it hasn't been
 816          * created yet, wait
 817          */
 818         if (unlikely(wait_obj_created(oi)))
 819                 goto fail;
 820
 821         ret = exofs_sync_op(or, sbi->s_timeout, oi->i_cred);
 822         osd_end_request(or);
 823         if (ret)
 824                 goto fail;
 825
 826 out:
 827         mark_inode_dirty(inode);
 828         return;
 829 fail:
 830         make_bad_inode(inode);
 831         goto out;
 832 }
 833
 834 /*
 835  * Set inode attributes - just call generic functions.
 836  */
 837 int exofs_setattr(struct dentry *dentry, struct iattr *iattr)
 838 {
 839         struct inode *inode = dentry->d_inode;
 840         int error;
 841
 842         error = inode_change_ok(inode, iattr);
 843         if (error)
 844                 return error;
 845
 846         error = inode_setattr(inode, iattr);
 847         return error;
 848 }
 849
 850 /*
 851  * Read an inode from the OSD, and return it as is.  We also return the size
 852  * attribute in the 'sanity' argument if we got compiled with debugging turned
 853  * on.
 854  */
 855 static int exofs_get_inode(struct super_block *sb, struct exofs_i_info *oi,
 856                     struct exofs_fcb *inode, uint64_t *sanity)
 857 {
 858         struct exofs_sb_info *sbi = sb->s_fs_info;
 859         struct osd_request *or;
 860         struct osd_attr attr;
 861         struct osd_obj_id obj = {sbi->s_pid,
 862                                  oi->vfs_inode.i_ino + EXOFS_OBJ_OFF};
 863         int ret;
 864
 865         exofs_make_credential(oi->i_cred, &obj);
 866
 867         or = osd_start_request(sbi->s_dev, GFP_KERNEL);
 868         if (unlikely(!or)) {
 869                 EXOFS_ERR("exofs_get_inode: osd_start_request failed.\n");
 870                 return -ENOMEM;
 871         }
 872         osd_req_get_attributes(or, &obj);
 873
 874         /* we need the inode attribute */
 875         osd_req_add_get_attr_list(or, &g_attr_inode_data, 1);
 876
 877 #ifdef EXOFS_DEBUG_OBJ_ISIZE
 878         /* we get the size attributes to do a sanity check */
 879         osd_req_add_get_attr_list(or, &g_attr_logical_length, 1);
 880 #endif
 881
 882         ret = exofs_sync_op(or, sbi->s_timeout, oi->i_cred);
 883         if (ret)
 884                 goto out;
 885
 886         attr = g_attr_inode_data;
 887         ret = extract_attr_from_req(or, &attr);
 888         if (ret) {
 889                 EXOFS_ERR("exofs_get_inode: extract_attr_from_req failed\n");
 890                 goto out;
 891         }
 892
 893         WARN_ON(attr.len != EXOFS_INO_ATTR_SIZE);
 894         memcpy(inode, attr.val_ptr, EXOFS_INO_ATTR_SIZE);
 895
 896 #ifdef EXOFS_DEBUG_OBJ_ISIZE
 897         attr = g_attr_logical_length;
 898         ret = extract_attr_from_req(or, &attr);
 899         if (ret) {
 900                 EXOFS_ERR("ERROR: extract attr from or failed\n");
 901                 goto out;
 902         }
 903         *sanity = get_unaligned_be64(attr.val_ptr);
 904 #endif
 905
 906 out:
 907         osd_end_request(or);
 908         return ret;
 909 }
 910
 911 /*
 912  * Fill in an inode read from the OSD and set it up for use
 913  */
 914 struct inode *exofs_iget(struct super_block *sb, unsigned long ino)
 915 {
 916         struct exofs_i_info *oi;
 917         struct exofs_fcb fcb;
 918         struct inode *inode;
 919         uint64_t uninitialized_var(sanity);
 920         int ret;
 921
 922         inode = iget_locked(sb, ino);
 923         if (!inode)
 924                 return ERR_PTR(-ENOMEM);
 925         if (!(inode->i_state & I_NEW))
 926                 return inode;
 927         oi = exofs_i(inode);
 928
 929         /* read the inode from the osd */
 930         ret = exofs_get_inode(sb, oi, &fcb, &sanity);
 931         if (ret)
 932                 goto bad_inode;
 933
 934         init_waitqueue_head(&oi->i_wq);
 935         set_obj_created(oi);
 936
 937         /* copy stuff from on-disk struct to in-memory struct */
 938         inode->i_mode = le16_to_cpu(fcb.i_mode);
 939         inode->i_uid = le32_to_cpu(fcb.i_uid);
 940         inode->i_gid = le32_to_cpu(fcb.i_gid);
 941         inode->i_nlink = le16_to_cpu(fcb.i_links_count);
 942         inode->i_ctime.tv_sec = (signed)le32_to_cpu(fcb.i_ctime);
 943         inode->i_atime.tv_sec = (signed)le32_to_cpu(fcb.i_atime);
 944         inode->i_mtime.tv_sec = (signed)le32_to_cpu(fcb.i_mtime);
 945         inode->i_ctime.tv_nsec =
 946                 inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = 0;
 947         oi->i_commit_size = le64_to_cpu(fcb.i_size);
 948         i_size_write(inode, oi->i_commit_size);
 949         inode->i_blkbits = EXOFS_BLKSHIFT;
 950         inode->i_generation = le32_to_cpu(fcb.i_generation);
 951
 952 #ifdef EXOFS_DEBUG_OBJ_ISIZE
 953         if ((inode->i_size != sanity) &&
 954                 (!exofs_inode_is_fast_symlink(inode))) {
 955                 EXOFS_ERR("WARNING: Size of inode=%llu != object=%llu\n",
 956                           inode->i_size, _LLU(sanity));
 957         }
 958 #endif
 959
 960         oi->i_dir_start_lookup = 0;
 961
 962         if ((inode->i_nlink == 0) && (inode->i_mode == 0)) {
 963                 ret = -ESTALE;
 964                 goto bad_inode;
 965         }
 966
 967         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
 968                 if (fcb.i_data[0])
 969                         inode->i_rdev =
 970                                 old_decode_dev(le32_to_cpu(fcb.i_data[0]));
 971                 else
 972                         inode->i_rdev =
 973                                 new_decode_dev(le32_to_cpu(fcb.i_data[1]));
 974         } else {
 975                 memcpy(oi->i_data, fcb.i_data, sizeof(fcb.i_data));
 976         }
 977
 978         if (S_ISREG(inode->i_mode)) {
 979                 inode->i_op = &exofs_file_inode_operations;
 980                 inode->i_fop = &exofs_file_operations;
 981                 inode->i_mapping->a_ops = &exofs_aops;
 982         } else if (S_ISDIR(inode->i_mode)) {
 983                 inode->i_op = &exofs_dir_inode_operations;
 984                 inode->i_fop = &exofs_dir_operations;
 985                 inode->i_mapping->a_ops = &exofs_aops;
 986         } else if (S_ISLNK(inode->i_mode)) {
 987                 if (exofs_inode_is_fast_symlink(inode))
 988                         inode->i_op = &exofs_fast_symlink_inode_operations;
 989                 else {
 990                         inode->i_op = &exofs_symlink_inode_operations;
 991                         inode->i_mapping->a_ops = &exofs_aops;
 992                 }
 993         } else {
 994                 inode->i_op = &exofs_special_inode_operations;
 995                 if (fcb.i_data[0])
 996                         init_special_inode(inode, inode->i_mode,
 997                            old_decode_dev(le32_to_cpu(fcb.i_data[0])));
 998                 else
 999                         init_special_inode(inode, inode->i_mode,
1000                            new_decode_dev(le32_to_cpu(fcb.i_data[1])));
1001         }
1002
1003         unlock_new_inode(inode);
1004         return inode;
1005
1006 bad_inode:
1007         iget_failed(inode);
1008         return ERR_PTR(ret);
1009 }
1010
1011 int __exofs_wait_obj_created(struct exofs_i_info *oi)
1012 {
1013         if (!obj_created(oi)) {
1014                 BUG_ON(!obj_2bcreated(oi));
1015                 wait_event(oi->i_wq, obj_created(oi));
1016         }
1017         return unlikely(is_bad_inode(&oi->vfs_inode)) ? -EIO : 0;
1018 }
1019 /*
1020  * Callback function from exofs_new_inode().  The important thing is that we
1021  * set the obj_created flag so that other methods know that the object exists on
1022  * the OSD.
1023  */
1024 static void create_done(struct osd_request *or, void *p)
1025 {
1026         struct inode *inode = p;
1027         struct exofs_i_info *oi = exofs_i(inode);
1028         struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
1029         int ret;
1030
1031         ret = exofs_check_ok(or);
1032         osd_end_request(or);
1033         atomic_dec(&sbi->s_curr_pending);
1034
1035         if (unlikely(ret)) {
1036                 EXOFS_ERR("object=0x%llx creation faild in pid=0x%llx",
1037                           _LLU(sbi->s_pid), _LLU(inode->i_ino + EXOFS_OBJ_OFF));
1038                 make_bad_inode(inode);
1039         } else
1040                 set_obj_created(oi);
1041
1042         atomic_dec(&inode->i_count);
1043         wake_up(&oi->i_wq);
1044 }
1045
1046 /*
1047  * Set up a new inode and create an object for it on the OSD
1048  */
1049 struct inode *exofs_new_inode(struct inode *dir, int mode)
1050 {
1051         struct super_block *sb;
1052         struct inode *inode;
1053         struct exofs_i_info *oi;
1054         struct exofs_sb_info *sbi;
1055         struct osd_request *or;
1056         struct osd_obj_id obj;
1057         int ret;
1058
1059         sb = dir->i_sb;
1060         inode = new_inode(sb);
1061         if (!inode)
1062                 return ERR_PTR(-ENOMEM);
1063
1064         oi = exofs_i(inode);
1065
1066         init_waitqueue_head(&oi->i_wq);
1067         set_obj_2bcreated(oi);
1068
1069         sbi = sb->s_fs_info;
1070
1071         sb->s_dirt = 1;
1072         inode->i_uid = current->cred->fsuid;
1073         if (dir->i_mode & S_ISGID) {
1074                 inode->i_gid = dir->i_gid;
1075                 if (S_ISDIR(mode))
1076                         mode |= S_ISGID;
1077         } else {
1078                 inode->i_gid = current->cred->fsgid;
1079         }
1080         inode->i_mode = mode;
1081
1082         inode->i_ino = sbi->s_nextid++;
1083         inode->i_blkbits = EXOFS_BLKSHIFT;
1084         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1085         oi->i_commit_size = inode->i_size = 0;
1086         spin_lock(&sbi->s_next_gen_lock);
1087         inode->i_generation = sbi->s_next_generation++;
1088         spin_unlock(&sbi->s_next_gen_lock);
1089         insert_inode_hash(inode);
1090
1091         mark_inode_dirty(inode);
1092
1093         obj.partition = sbi->s_pid;
1094         obj.id = inode->i_ino + EXOFS_OBJ_OFF;
1095         exofs_make_credential(oi->i_cred, &obj);
1096
1097         or = osd_start_request(sbi->s_dev, GFP_KERNEL);
1098         if (unlikely(!or)) {
1099                 EXOFS_ERR("exofs_new_inode: osd_start_request failed\n");
1100                 return ERR_PTR(-ENOMEM);
1101         }
1102
1103         osd_req_create_object(or, &obj);
1104
1105         /* increment the refcount so that the inode will still be around when we
1106          * reach the callback
1107          */
1108         atomic_inc(&inode->i_count);
1109
1110         ret = exofs_async_op(or, create_done, inode, oi->i_cred);
1111         if (ret) {
1112                 atomic_dec(&inode->i_count);
1113                 osd_end_request(or);
1114                 return ERR_PTR(-EIO);
1115         }
1116         atomic_inc(&sbi->s_curr_pending);
1117
1118         return inode;
1119 }
1120
1121 /*
1122  * struct to pass two arguments to update_inode's callback
1123  */
1124 struct updatei_args {
1125         struct exofs_sb_info    *sbi;
1126         struct exofs_fcb        fcb;
1127 };
1128
1129 /*
1130  * Callback function from exofs_update_inode().
1131  */
1132 static void updatei_done(struct osd_request *or, void *p)
1133 {
1134         struct updatei_args *args = p;
1135
1136         osd_end_request(or);
1137
1138         atomic_dec(&args->sbi->s_curr_pending);
1139
1140         kfree(args);
1141 }
1142
1143 /*
1144  * Write the inode to the OSD.  Just fill up the struct, and set the attribute
1145  * synchronously or asynchronously depending on the do_sync flag.
1146  */
1147 static int exofs_update_inode(struct inode *inode, int do_sync)
1148 {
1149         struct exofs_i_info *oi = exofs_i(inode);
1150         struct super_block *sb = inode->i_sb;
1151         struct exofs_sb_info *sbi = sb->s_fs_info;
1152         struct osd_obj_id obj = {sbi->s_pid, inode->i_ino + EXOFS_OBJ_OFF};
1153         struct osd_request *or;
1154         struct osd_attr attr;
1155         struct exofs_fcb *fcb;
1156         struct updatei_args *args;
1157         int ret;
1158
1159         args = kzalloc(sizeof(*args), GFP_KERNEL);
1160         if (!args)
1161                 return -ENOMEM;
1162
1163         fcb = &args->fcb;
1164
1165         fcb->i_mode = cpu_to_le16(inode->i_mode);
1166         fcb->i_uid = cpu_to_le32(inode->i_uid);
1167         fcb->i_gid = cpu_to_le32(inode->i_gid);
1168         fcb->i_links_count = cpu_to_le16(inode->i_nlink);
1169         fcb->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1170         fcb->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1171         fcb->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1172         oi->i_commit_size = i_size_read(inode);
1173         fcb->i_size = cpu_to_le64(oi->i_commit_size);
1174         fcb->i_generation = cpu_to_le32(inode->i_generation);
1175
1176         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1177                 if (old_valid_dev(inode->i_rdev)) {
1178                         fcb->i_data[0] =
1179                                 cpu_to_le32(old_encode_dev(inode->i_rdev));
1180                         fcb->i_data[1] = 0;
1181                 } else {
1182                         fcb->i_data[0] = 0;
1183                         fcb->i_data[1] =
1184                                 cpu_to_le32(new_encode_dev(inode->i_rdev));
1185                         fcb->i_data[2] = 0;
1186                 }
1187         } else
1188                 memcpy(fcb->i_data, oi->i_data, sizeof(fcb->i_data));
1189
1190         or = osd_start_request(sbi->s_dev, GFP_KERNEL);
1191         if (unlikely(!or)) {
1192                 EXOFS_ERR("exofs_update_inode: osd_start_request failed.\n");
1193                 ret = -ENOMEM;
1194                 goto free_args;
1195         }
1196
1197         osd_req_set_attributes(or, &obj);
1198
1199         attr = g_attr_inode_data;
1200         attr.val_ptr = fcb;
1201         osd_req_add_set_attr_list(or, &attr, 1);
1202
1203         if (!obj_created(oi)) {
1204                 EXOFS_DBGMSG("!obj_created\n");
1205                 BUG_ON(!obj_2bcreated(oi));
1206                 wait_event(oi->i_wq, obj_created(oi));
1207                 EXOFS_DBGMSG("wait_event done\n");
1208         }
1209
1210         if (do_sync) {
1211                 ret = exofs_sync_op(or, sbi->s_timeout, oi->i_cred);
1212                 osd_end_request(or);
1213                 goto free_args;
1214         } else {
1215                 args->sbi = sbi;
1216
1217                 ret = exofs_async_op(or, updatei_done, args, oi->i_cred);
1218                 if (ret) {
1219                         osd_end_request(or);
1220                         goto free_args;
1221                 }
1222                 atomic_inc(&sbi->s_curr_pending);
1223                 goto out; /* deallocation in updatei_done */
1224         }
1225
1226 free_args:
1227         kfree(args);
1228 out:
1229         EXOFS_DBGMSG("ret=>%d\n", ret);
1230         return ret;
1231 }
1232
1233 int exofs_write_inode(struct inode *inode, int wait)
1234 {
1235         return exofs_update_inode(inode, wait);
1236 }
1237
1238 /*
1239  * Callback function from exofs_delete_inode() - don't have much cleaning up to
1240  * do.
1241  */
1242 static void delete_done(struct osd_request *or, void *p)
1243 {
1244         struct exofs_sb_info *sbi;
1245         osd_end_request(or);
1246         sbi = p;
1247         atomic_dec(&sbi->s_curr_pending);
1248 }
1249
1250 /*
1251  * Called when the refcount of an inode reaches zero.  We remove the object
1252  * from the OSD here.  We make sure the object was created before we try and
1253  * delete it.
1254  */
1255 void exofs_delete_inode(struct inode *inode)
1256 {
1257         struct exofs_i_info *oi = exofs_i(inode);
1258         struct super_block *sb = inode->i_sb;
1259         struct exofs_sb_info *sbi = sb->s_fs_info;
1260         struct osd_obj_id obj = {sbi->s_pid, inode->i_ino + EXOFS_OBJ_OFF};
1261         struct osd_request *or;
1262         int ret;
1263
1264         truncate_inode_pages(&inode->i_data, 0);
1265
1266         if (is_bad_inode(inode))
1267                 goto no_delete;
1268
1269         mark_inode_dirty(inode);
1270         exofs_update_inode(inode, inode_needs_sync(inode));
1271
1272         inode->i_size = 0;
1273         if (inode->i_blocks)
1274                 exofs_truncate(inode);
1275
1276         clear_inode(inode);
1277
1278         or = osd_start_request(sbi->s_dev, GFP_KERNEL);
1279         if (unlikely(!or)) {
1280                 EXOFS_ERR("exofs_delete_inode: osd_start_request failed\n");
1281                 return;
1282         }
1283
1284         osd_req_remove_object(or, &obj);
1285
1286         /* if we are deleting an obj that hasn't been created yet, wait */
1287         if (!obj_created(oi)) {
1288                 BUG_ON(!obj_2bcreated(oi));
1289                 wait_event(oi->i_wq, obj_created(oi));
1290         }
1291
1292         ret = exofs_async_op(or, delete_done, sbi, oi->i_cred);
1293         if (ret) {
1294                 EXOFS_ERR(
1295                        "ERROR: @exofs_delete_inode exofs_async_op failed\n");
1296                 osd_end_request(or);
1297                 return;
1298         }
1299         atomic_inc(&sbi->s_curr_pending);
1300
1301         return;
1302
1303 no_delete:
1304         clear_inode(inode);
1305 }