lustre/llite/rw.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * Lustre Lite I/O Page Cache
   5  *
   6  *  Copyright (c) 2001-2003 Cluster File Systems, Inc.
   7  *
   8  *   This file is part of Lustre, http://www.lustre.org.
   9  *
  10  *   Lustre is free software; you can redistribute it and/or
  11  *   modify it under the terms of version 2 of the GNU General Public
  12  *   License as published by the Free Software Foundation.
  13  *
  14  *   Lustre is distributed in the hope that it will be useful,
  15  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  *   GNU General Public License for more details.
  18  *
  19  *   You should have received a copy of the GNU General Public License
  20  *   along with Lustre; if not, write to the Free Software
  21  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  22  */
  23
  24 #include <linux/config.h>
  25 #include <linux/kernel.h>
  26 #include <linux/mm.h>
  27 #include <linux/string.h>
  28 #include <linux/stat.h>
  29 #include <linux/errno.h>
  30 #include <linux/smp_lock.h>
  31 #include <linux/unistd.h>
  32 #include <linux/version.h>
  33 #include <asm/system.h>
  34 #include <asm/uaccess.h>
  35 #include "llite_internal.h"
  36
  37 #include <linux/fs.h>
  38 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
  39 #include <linux/buffer_head.h>
  40 #else
  41 #include <linux/iobuf.h>
  42 #endif
  43 #include <linux/stat.h>
  44 #include <asm/uaccess.h>
  45 #include <asm/segment.h>
  46 #include <linux/mm.h>
  47 #include <linux/pagemap.h>
  48 #include <linux/smp_lock.h>
  49
  50 #define DEBUG_SUBSYSTEM S_LLITE
  51
  52 #include <linux/lustre_mds.h>
  53 #include <linux/lustre_lite.h>
  54 #include <linux/lustre_lib.h>
  55 #include <linux/lustre_compat25.h>
  56
  57 /*
  58  * Remove page from dirty list
  59  */
  60 static void __set_page_clean(struct page *page)
  61 {
  62         struct address_space *mapping = page->mapping;
  63         struct inode *inode;
  64
  65         if (!mapping)
  66                 return;
  67
  68         PGCACHE_WRLOCK(mapping);
  69
  70         list_del(&page->list);
  71         list_add(&page->list, &mapping->clean_pages);
  72
  73         /* XXX doesn't inode_lock protect i_state ? */
  74         inode = mapping->host;
  75         if (list_empty(&mapping->dirty_pages)) {
  76                 CDEBUG(D_INODE, "inode clean\n");
  77                 inode->i_state &= ~I_DIRTY_PAGES;
  78         }
  79
  80         PGCACHE_WRUNLOCK(mapping);
  81         EXIT;
  82 }
  83
  84 void set_page_clean(struct page *page)
  85 {
  86         if (PageDirty(page)) {
  87                 ClearPageDirty(page);
  88                 __set_page_clean(page);
  89         }
  90 }
  91
  92 /* SYNCHRONOUS I/O to object storage for an inode */
  93 static int ll_brw(int cmd, struct inode *inode, struct page *page, int flags)
  94 {
  95         struct ll_inode_info *lli = ll_i2info(inode);
  96         struct lov_stripe_md *lsm = lli->lli_smd;
  97         struct brw_page pg;
  98         int rc;
  99         ENTRY;
 100
 101         pg.pg = page;
 102         pg.off = ((obd_off)page->index) << PAGE_SHIFT;
 103
 104         if (cmd == OBD_BRW_WRITE && (pg.off + PAGE_SIZE > inode->i_size))
 105                 pg.count = inode->i_size % PAGE_SIZE;
 106         else
 107                 pg.count = PAGE_SIZE;
 108
 109         CDEBUG(D_PAGE, "%s %d bytes ino %lu at "LPU64"/"LPX64"\n",
 110                cmd & OBD_BRW_WRITE ? "write" : "read", pg.count, inode->i_ino,
 111                pg.off, pg.off);
 112         if (pg.count == 0) {
 113                 CERROR("ZERO COUNT: ino %lu: size %p:%Lu(%p:%Lu) idx %lu off "
 114                        LPU64"\n",
 115                        inode->i_ino, inode, inode->i_size, page->mapping->host,
 116                        page->mapping->host->i_size, page->index, pg.off);
 117         }
 118
 119         pg.flag = flags;
 120
 121         if (cmd == OBD_BRW_WRITE)
 122                 lprocfs_counter_add(ll_i2sbi(inode)->ll_stats,
 123                                     LPROC_LL_BRW_WRITE, pg.count);
 124         else
 125                 lprocfs_counter_add(ll_i2sbi(inode)->ll_stats,
 126                                     LPROC_LL_BRW_READ, pg.count);
 127         rc = obd_brw(cmd, ll_i2obdconn(inode), lsm, 1, &pg, NULL);
 128         if (rc)
 129                 CERROR("error from obd_brw: rc = %d\n", rc);
 130
 131         RETURN(rc);
 132 }
 133
 134 /*
 135  * we were asked to read a single page but we're going to try and read a batch
 136  * of pages all at once.  this vaguely simulates 2.5's readpages.
 137  */
 138 static int ll_readpage(struct file *file, struct page *first_page)
 139 {
 140         struct inode *inode = first_page->mapping->host;
 141         struct ll_inode_info *lli = ll_i2info(inode);
 142         struct page *page = first_page;
 143         struct list_head *pos;
 144         struct brw_page *pgs;
 145         unsigned long end_index, extent_end = 0;
 146         struct ptlrpc_request_set *set;
 147         int npgs = 0, rc = 0, max_pages;
 148         ENTRY;
 149
 150         LASSERT(PageLocked(page));
 151         LASSERT(!PageUptodate(page));
 152         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),offset="LPX64"\n",
 153                inode->i_ino, inode->i_generation, inode,
 154                (((obd_off)page->index) << PAGE_SHIFT));
 155         LASSERT(atomic_read(&file->f_dentry->d_inode->i_count) > 0);
 156
 157         if (inode->i_size <= ((obd_off)page->index) << PAGE_SHIFT) {
 158                 CERROR("reading beyond EOF\n");
 159                 memset(kmap(page), 0, PAGE_SIZE);
 160                 kunmap(page);
 161                 SetPageUptodate(page);
 162                 unlock_page(page);
 163                 RETURN(rc);
 164         }
 165
 166         /* try to read the file's preferred block size in a one-er */
 167         end_index = first_page->index +
 168                 (inode->i_blksize >> PAGE_CACHE_SHIFT);
 169         if (end_index > (inode->i_size >> PAGE_CACHE_SHIFT))
 170                 end_index = inode->i_size >> PAGE_CACHE_SHIFT;
 171
 172         max_pages = ((end_index - first_page->index) << PAGE_CACHE_SHIFT) >>
 173                 PAGE_SHIFT;
 174         pgs = kmalloc(max_pages * sizeof(*pgs), GFP_USER);
 175         if (pgs == NULL)
 176                 RETURN(-ENOMEM);
 177
 178         /*
 179          * find how far we're allowed to read under the extent ll_file_read
 180          * is passing us..
 181          */
 182         spin_lock(&lli->lli_read_extent_lock);
 183         list_for_each(pos, &lli->lli_read_extents) {
 184                 struct ll_read_extent *rextent;
 185                 rextent = list_entry(pos, struct ll_read_extent, re_lli_item);
 186                 if (rextent->re_task != current)
 187                         continue;
 188
 189                 if (rextent->re_extent.end + PAGE_SIZE < rextent->re_extent.end)
 190                         /* extent wrapping */
 191                         extent_end = ~0;
 192                 else {
 193                         extent_end = (rextent->re_extent.end + PAGE_SIZE)
 194                                                         << PAGE_CACHE_SHIFT;
 195                         /* 32bit indexes, 64bit extents.. */
 196                         if (((u64)extent_end >> PAGE_CACHE_SHIFT) <
 197                                         rextent->re_extent.end)
 198                                 extent_end = ~0;
 199                 }
 200                 break;
 201         }
 202         spin_unlock(&lli->lli_read_extent_lock);
 203
 204         if (extent_end == 0) {
 205                 static long next_print;
 206                 if (time_after(jiffies, next_print)) {
 207                         next_print = jiffies + 30 * HZ;
 208                         CDEBUG(D_INODE, "mmap readpage - check locks\n");
 209                 }
 210                 end_index = page->index + 1;
 211         } else if (extent_end < end_index)
 212                 end_index = extent_end;
 213
 214         /* to balance the find_get_page ref the other pages get that is
 215          * decrefed on teardown.. */
 216         page_cache_get(page);
 217         do {
 218                 unsigned long index ;
 219
 220                 pgs[npgs].pg = page;
 221                 pgs[npgs].off = ((obd_off)page->index) << PAGE_CACHE_SHIFT;
 222                 pgs[npgs].flag = 0;
 223                 pgs[npgs].count = PAGE_SIZE;
 224                 /* XXX Workaround for BA OSTs returning short reads at EOF.
 225                  * The linux OST will return the full page, zero-filled at the
 226                  * end, which will just overwrite the data we set here.  Bug
 227                  * 593 relates to fixing this properly.
 228                  */
 229                 if (inode->i_size < pgs[npgs].off + PAGE_SIZE) {
 230                         int count = inode->i_size - pgs[npgs].off;
 231                         void *addr = kmap(page);
 232                         pgs[npgs].count = count;
 233                         //POISON(addr, 0x7c, count);
 234                         memset(addr + count, 0, PAGE_SIZE - count);
 235                         kunmap(page);
 236                 }
 237
 238                 npgs++;
 239                 if (npgs == max_pages)
 240                         break;
 241
 242                 /*
 243                  * find pages ahead of us that we can read in.
 244                  * grab_cache_page waits on pages that are locked so
 245                  * we first try find_get_page, which doesn't.  this stops
 246                  * the worst case behaviour of racing threads waiting on
 247                  * each other, but doesn't remove it entirely.
 248                  */
 249                 for (index = page->index + 1, page = NULL;
 250                      page == NULL && index < end_index; index++) {
 251
 252                         /* see if the page already exists and needs updating */
 253                         page = find_get_page(inode->i_mapping, index);
 254                         if (page) {
 255                                 if (Page_Uptodate(page) || TryLockPage(page))
 256                                         goto out_release;
 257                                 if (!page->mapping || Page_Uptodate(page))
 258                                         goto out_unlock;
 259                         } else {
 260                                 /* ok, we have to create it.. */
 261                                 page = grab_cache_page(inode->i_mapping, index);
 262                                 if (page == NULL)
 263                                         continue;
 264                                 if (Page_Uptodate(page))
 265                                         goto out_unlock;
 266                         }
 267
 268                         break;
 269
 270                 out_unlock:
 271                         unlock_page(page);
 272                 out_release:
 273                         page_cache_release(page);
 274                         page = NULL;
 275                 }
 276
 277         } while (page);
 278
 279         set = ptlrpc_prep_set();
 280         if (set == NULL) {
 281                 CERROR("ENOMEM allocing request set\n");
 282                 rc = -ENOMEM;
 283         } else {
 284                 rc = obd_brw_async(OBD_BRW_READ, ll_i2obdconn(inode),
 285                                    ll_i2info(inode)->lli_smd, npgs, pgs,
 286                                    set, NULL);
 287                 if (rc == 0)
 288                         rc = ptlrpc_set_wait(set);
 289                 ptlrpc_set_destroy(set);
 290                 if (rc && rc != -EIO)
 291                         CERROR("error from obd_brw_async: rc = %d\n", rc);
 292         }
 293
 294         while (npgs-- > 0) {
 295                 page = pgs[npgs].pg;
 296
 297                 if (rc == 0)
 298                         SetPageUptodate(page);
 299                 unlock_page(page);
 300                 page_cache_release(page);
 301         }
 302
 303         kfree(pgs);
 304         RETURN(rc);
 305 } /* ll_readpage */
 306
 307 /* this isn't where truncate starts.   roughly:
 308  * sys_truncate->ll_setattr_raw->vmtruncate->ll_truncate
 309  * we grab the lock back in setattr_raw to avoid races. */
 310 void ll_truncate(struct inode *inode)
 311 {
 312         struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
 313         struct obdo oa = {0};
 314         int err;
 315         ENTRY;
 316         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
 317                inode->i_generation, inode);
 318
 319         if (!lsm) {
 320                 /* object not yet allocated */
 321                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 322                 EXIT;
 323                 return;
 324         }
 325
 326         /* vmtruncate will just throw away our dirty pages, make sure
 327          * we don't think they're still dirty, being careful to round
 328          * i_size to the first whole page that was tossed */
 329         err = ll_clear_dirty_pages(ll_i2obdconn(inode), lsm,
 330                         (inode->i_size + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT,
 331                         ~0);
 332
 333         oa.o_id = lsm->lsm_object_id;
 334         oa.o_mode = inode->i_mode;
 335         oa.o_valid = OBD_MD_FLID | OBD_MD_FLMODE | OBD_MD_FLTYPE;
 336
 337         CDEBUG(D_INFO, "calling punch for "LPX64" (all bytes after %Lu)\n",
 338                oa.o_id, inode->i_size);
 339
 340         /* truncate == punch from new size to absolute end of file */
 341         err = obd_punch(ll_i2obdconn(inode), &oa, lsm, inode->i_size,
 342                         OBD_OBJECT_EOF, NULL);
 343         if (err)
 344                 CERROR("obd_truncate fails (%d) ino %lu\n", err, inode->i_ino);
 345         else
 346                 obdo_to_inode(inode, &oa, oa.o_valid);
 347
 348         EXIT;
 349         return;
 350 } /* ll_truncate */
 351
 352 //#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 353
 354 static int ll_prepare_write(struct file *file, struct page *page, unsigned from,
 355                             unsigned to)
 356 {
 357         struct inode *inode = page->mapping->host;
 358         struct ll_inode_info *lli = ll_i2info(inode);
 359         struct lov_stripe_md *lsm = lli->lli_smd;
 360         obd_off offset = ((obd_off)page->index) << PAGE_SHIFT;
 361         struct brw_page pg;
 362         int rc = 0;
 363         ENTRY;
 364
 365         if (!PageLocked(page))
 366                 LBUG();
 367
 368         if (PageUptodate(page))
 369                 RETURN(0);
 370
 371         //POISON(addr + from, 0xca, to - from);
 372
 373         /* Check to see if we should return -EIO right away */
 374         pg.pg = page;
 375         pg.off = offset;
 376         pg.count = PAGE_SIZE;
 377         pg.flag = 0;
 378         rc = obd_brw(OBD_BRW_CHECK, ll_i2obdconn(inode), lsm, 1, &pg, NULL);
 379         if (rc)
 380                 RETURN(rc);
 381
 382         /* We're completely overwriting an existing page, so _don't_ set it up
 383          * to date until commit_write */
 384         if (from == 0 && to == PAGE_SIZE)
 385                 RETURN(0);
 386
 387         /* If are writing to a new page, no need to read old data.
 388          * the extent locking and getattr procedures in ll_file_write have
 389          * guaranteed that i_size is stable enough for our zeroing needs */
 390         if (inode->i_size <= offset) {
 391                 memset(kmap(page), 0, PAGE_SIZE);
 392                 kunmap(page);
 393                 GOTO(prepare_done, rc = 0);
 394         }
 395
 396         rc = ll_brw(OBD_BRW_READ, inode, page, 0);
 397
 398         EXIT;
 399  prepare_done:
 400         if (rc == 0)
 401                 SetPageUptodate(page);
 402
 403         return rc;
 404 }
 405
 406 /*
 407  * background file writeback.  This is called regularly from kupdated to write
 408  * dirty data, from kswapd when memory is low, and from filemap_fdatasync when
 409  * super blocks or inodes are synced..
 410  *
 411  * obd_brw errors down in _batch_writepage are ignored, so pages are always
 412  * unlocked.  Also, there is nobody to return an error code to from here - the
 413  * application may not even be running anymore.
 414  *
 415  * this should be async so that things like kswapd can have a chance to
 416  * free some more pages that our allocating writeback may need, but it isn't
 417  * yet.
 418  */
 419 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 420 static unsigned long ll_local_cache_dirty_pages;
 421 static unsigned long ll_max_dirty_pages = 20 * 1024 * 1024 / PAGE_SIZE;
 422
 423 static spinlock_t ll_local_cache_page_count_lock = SPIN_LOCK_UNLOCKED;
 424
 425 int ll_rd_dirty_pages(char *page, char **start, off_t off, int count, int *eof,
 426                       void *data)
 427 {
 428         unsigned long dirty_count;
 429         spin_lock(&ll_local_cache_page_count_lock);
 430         dirty_count = ll_local_cache_dirty_pages;
 431         spin_unlock(&ll_local_cache_page_count_lock);
 432         return snprintf(page, count, "%lu\n", dirty_count);
 433 }
 434
 435 int ll_rd_max_dirty_pages(char *page, char **start, off_t off, int count,
 436                           int *eof, void *data)
 437 {
 438         unsigned long max_dirty;
 439         spin_lock(&ll_local_cache_page_count_lock);
 440         max_dirty = ll_max_dirty_pages;
 441         spin_unlock(&ll_local_cache_page_count_lock);
 442         return snprintf(page, count, "%lu\n", max_dirty);
 443 }
 444
 445 int ll_wr_max_dirty_pages(struct file *file, const char *buffer,
 446                           unsigned long count, void *data)
 447 {
 448         unsigned long max_dirty;
 449         signed long max_dirty_signed;
 450         char kernbuf[20], *end;
 451
 452         if (count > (sizeof(kernbuf) - 1))
 453                 return -EINVAL;
 454
 455         if (copy_from_user(kernbuf, buffer, count))
 456                 return -EFAULT;
 457
 458         kernbuf[count] = '\0';
 459
 460         max_dirty_signed = simple_strtol(kernbuf, &end, 0);
 461         if (kernbuf == end)
 462                 return -EINVAL;
 463         max_dirty = (unsigned long)max_dirty_signed;
 464
 465 #if 0
 466         if (max_dirty < ll_local_cache_dirty_pages)
 467                 flush_to_new_max_dirty();
 468 #endif
 469
 470         spin_lock(&ll_local_cache_page_count_lock);
 471         CDEBUG(D_CACHE, "changing max_dirty from %lu to %lu\n",
 472                ll_max_dirty_pages, max_dirty);
 473         ll_max_dirty_pages = max_dirty;
 474         spin_unlock(&ll_local_cache_page_count_lock);
 475         return count;
 476 }
 477
 478 static int ll_local_cache_full(void)
 479 {
 480         int full = 0;
 481         spin_lock(&ll_local_cache_page_count_lock);
 482         if (ll_max_dirty_pages &&
 483             ll_local_cache_dirty_pages >= ll_max_dirty_pages) {
 484                 full = 1;
 485         }
 486         spin_unlock(&ll_local_cache_page_count_lock);
 487         /* XXX instrument? */
 488         /* XXX trigger async writeback when full, or 75% of full? */
 489         return full;
 490 }
 491
 492 static void ll_local_cache_flushed_pages(unsigned long pgcount)
 493 {
 494         unsigned long dirty_count;
 495         spin_lock(&ll_local_cache_page_count_lock);
 496         dirty_count = ll_local_cache_dirty_pages;
 497         ll_local_cache_dirty_pages -= pgcount;
 498         CDEBUG(D_CACHE, "dirty pages: %lu->%lu)\n",
 499                dirty_count, ll_local_cache_dirty_pages);
 500         spin_unlock(&ll_local_cache_page_count_lock);
 501         LASSERT(dirty_count >= pgcount);
 502 }
 503
 504 static void ll_local_cache_dirtied_pages(unsigned long pgcount)
 505 {
 506         unsigned long dirty_count;
 507         spin_lock(&ll_local_cache_page_count_lock);
 508         dirty_count = ll_local_cache_dirty_pages;
 509         ll_local_cache_dirty_pages += pgcount;
 510         CDEBUG(D_CACHE, "dirty pages: %lu->%lu\n",
 511                dirty_count, ll_local_cache_dirty_pages);
 512         spin_unlock(&ll_local_cache_page_count_lock);
 513         /* XXX track maximum cached, report to lprocfs */
 514 }
 515
 516 int ll_clear_dirty_pages(struct lustre_handle *conn, struct lov_stripe_md *lsm,
 517                          unsigned long start, unsigned long end)
 518 {
 519         unsigned long cleared;
 520         int rc;
 521
 522         ENTRY;
 523         rc = obd_clear_dirty_pages(conn, lsm, start, end, &cleared);
 524         if (!rc)
 525                 ll_local_cache_flushed_pages(cleared);
 526         RETURN(rc);
 527 }
 528
 529 int ll_mark_dirty_page(struct lustre_handle *conn, struct lov_stripe_md *lsm,
 530                        unsigned long index)
 531 {
 532         int rc;
 533
 534         ENTRY;
 535         if (ll_local_cache_full())
 536                 RETURN(-EDQUOT);
 537
 538         rc = obd_mark_page_dirty(conn, lsm, index);
 539         if (!rc)
 540                 ll_local_cache_dirtied_pages(1);
 541         RETURN(rc);
 542 }
 543
 544 static int ll_writepage(struct page *page)
 545 {
 546         struct inode *inode = page->mapping->host;
 547         ENTRY;
 548
 549         CDEBUG(D_CACHE, "page %p [lau %d] inode %p\n", page,
 550                         PageLaunder(page), inode);
 551         LASSERT(PageLocked(page));
 552
 553         /* XXX should obd_brw errors trickle up? */
 554         ll_batch_writepage(inode, page);
 555         RETURN(0);
 556 }
 557
 558 /*
 559  * we really don't want to start writeback here, we want to give callers some
 560  * time to further dirty the pages before we write them out.
 561  */
 562 static int ll_commit_write(struct file *file, struct page *page,
 563                            unsigned from, unsigned to)
 564 {
 565         struct inode *inode = page->mapping->host;
 566         loff_t size;
 567         int rc = 0;
 568         ENTRY;
 569
 570         LASSERT(inode == file->f_dentry->d_inode);
 571         LASSERT(PageLocked(page));
 572
 573         CDEBUG(D_INODE, "inode %p is writing page %p from %d to %d at %lu\n",
 574                inode, page, from, to, page->index);
 575         if (!PageDirty(page)) {
 576                 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats,
 577                                      LPROC_LL_DIRTY_MISSES);
 578                 rc = ll_mark_dirty_page(ll_i2obdconn(inode),
 579                                         ll_i2info(inode)->lli_smd,
 580                                         page->index);
 581                 if (rc < 0 && rc != -EDQUOT)
 582                         RETURN(rc); /* XXX lproc counter here? */
 583         } else {
 584                 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats,
 585                                      LPROC_LL_DIRTY_HITS);
 586         }
 587
 588         size = (((obd_off)page->index) << PAGE_SHIFT) + to;
 589         if (size > inode->i_size)
 590                 inode->i_size = size;
 591
 592         SetPageUptodate(page);
 593         set_page_dirty(page);
 594
 595         /* This means that we've hit either the local cache limit or the limit
 596          * of the OST's grant. */
 597         if (rc == -EDQUOT) {
 598                 int rc = ll_batch_writepage(inode, page);
 599                 lock_page(page); /* caller expects to unlock */
 600                 RETURN(rc);
 601         }
 602
 603         RETURN(0);
 604 } /* ll_commit_write */
 605 #else
 606 static int ll_writepage(struct page *page,
 607                         struct writeback_control *wbc)
 608 {
 609
 610         return 0;
 611 }
 612 static int ll_commit_write(struct file *file, struct page *page,
 613                            unsigned from, unsigned to)
 614 {
 615         return 0;
 616 }
 617 #endif
 618
 619 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 620 static int ll_direct_IO(int rw, struct inode *inode, struct kiobuf *iobuf,
 621                         unsigned long blocknr, int blocksize)
 622 {
 623         struct ll_inode_info *lli = ll_i2info(inode);
 624         struct lov_stripe_md *lsm = lli->lli_smd;
 625         struct brw_page *pga;
 626         struct ptlrpc_request_set *set;
 627         int length, i, flags, rc = 0;
 628         loff_t offset;
 629         ENTRY;
 630
 631         if (!lsm || !lsm->lsm_object_id)
 632                 RETURN(-ENOMEM);
 633
 634         if ((iobuf->offset & (blocksize - 1)) ||
 635             (iobuf->length & (blocksize - 1)))
 636                 RETURN(-EINVAL);
 637
 638         set = ptlrpc_prep_set();
 639         if (set == NULL)
 640                 RETURN(-ENOMEM);
 641
 642         OBD_ALLOC(pga, sizeof(*pga) * iobuf->nr_pages);
 643         if (!pga) {
 644                 ptlrpc_set_destroy(set);
 645                 RETURN(-ENOMEM);
 646         }
 647
 648         flags = (rw == WRITE ? OBD_BRW_CREATE : 0) /* | OBD_BRW_DIRECTIO */;
 649         offset = ((obd_off)blocknr << inode->i_blkbits);
 650         length = iobuf->length;
 651
 652         for (i = 0, length = iobuf->length; length > 0;
 653              length -= pga[i].count, offset += pga[i].count, i++) { /*i last!*/
 654                 pga[i].pg = iobuf->maplist[i];
 655                 pga[i].off = offset;
 656                 /* To the end of the page, or the length, whatever is less */
 657                 pga[i].count = min_t(int, PAGE_SIZE - (offset & ~PAGE_MASK),
 658                                      length);
 659                 pga[i].flag = flags;
 660                 if (rw == READ) {
 661                         //POISON(kmap(iobuf->maplist[i]), 0xc5, PAGE_SIZE);
 662                         //kunmap(iobuf->maplist[i]);
 663                 }
 664         }
 665
 666         if (rw == WRITE)
 667                 lprocfs_counter_add(ll_i2sbi(inode)->ll_stats,
 668                                     LPROC_LL_DIRECT_WRITE, iobuf->length);
 669         else
 670                 lprocfs_counter_add(ll_i2sbi(inode)->ll_stats,
 671                                     LPROC_LL_DIRECT_READ, iobuf->length);
 672         rc = obd_brw_async(rw == WRITE ? OBD_BRW_WRITE : OBD_BRW_READ,
 673                            ll_i2obdconn(inode), lsm, iobuf->nr_pages, pga, set,
 674                            NULL);
 675         if (rc) {
 676                 CDEBUG(rc == -ENOSPC ? D_INODE : D_ERROR,
 677                        "error from obd_brw_async: rc = %d\n", rc);
 678         } else {
 679                 rc = ptlrpc_set_wait(set);
 680                 if (rc)
 681                         CERROR("error from callback: rc = %d\n", rc);
 682         }
 683         ptlrpc_set_destroy(set);
 684         if (rc == 0)
 685                 rc = iobuf->length;
 686
 687         OBD_FREE(pga, sizeof(*pga) * iobuf->nr_pages);
 688         RETURN(rc);
 689 }
 690 #endif
 691
 692 //#endif
 693
 694 struct address_space_operations ll_aops = {
 695         readpage: ll_readpage,
 696 #if (LINUX_VERSION_CODE <= KERNEL_VERSION(2,5,0))
 697         direct_IO: ll_direct_IO,
 698 #endif
 699         writepage: ll_writepage,
 700         sync_page: block_sync_page,
 701         prepare_write: ll_prepare_write,
 702         commit_write: ll_commit_write,
 703         bmap: NULL
 704 //#endif
 705 };