lustre/obdfilter/filter_io.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  linux/fs/obdfilter/filter_io.c
   5  *
   6  *  Copyright (c) 2001-2003 Cluster File Systems, Inc.
   7  *   Author: Peter Braam <braam@clusterfs.com>
   8  *   Author: Andreas Dilger <adilger@clusterfs.com>
   9  *   Author: Phil Schwan <phil@clusterfs.com>
  10  *
  11  *   This file is part of Lustre, http://www.lustre.org.
  12  *
  13  *   Lustre is free software; you can redistribute it and/or
  14  *   modify it under the terms of version 2 of the GNU General Public
  15  *   License as published by the Free Software Foundation.
  16  *
  17  *   Lustre is distributed in the hope that it will be useful,
  18  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  19  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  20  *   GNU General Public License for more details.
  21  *
  22  *   You should have received a copy of the GNU General Public License
  23  *   along with Lustre; if not, write to the Free Software
  24  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  25  */
  26
  27 #define DEBUG_SUBSYSTEM S_FILTER
  28
  29 #include <linux/config.h>
  30 #include <linux/module.h>
  31 #include <linux/pagemap.h> // XXX kill me soon
  32 #include <linux/version.h>
  33
  34 #include <linux/obd_class.h>
  35 #include <linux/lustre_fsfilt.h>
  36 #include "filter_internal.h"
  37
  38 int *obdfilter_created_scratchpad;
  39
  40 static int filter_alloc_dio_page(struct obd_device *obd, struct inode *inode,
  41                                  struct niobuf_local *lnb)
  42 {
  43         struct page *page;
  44
  45         page = alloc_pages(GFP_HIGHUSER, 0);
  46         if (page == NULL) {
  47                 CERROR("no memory for a temp page\n");
  48                 lnb->rc = -ENOMEM;
  49                 RETURN(-ENOMEM);
  50         }
  51 #if 0
  52         POISON_PAGE(page, 0xf1);
  53         if (lnb->len != PAGE_SIZE) {
  54                 memset(kmap(page) + lnb->len, 0, PAGE_SIZE - lnb->len);
  55                 kunmap(page);
  56         }
  57 #endif
  58         page->index = lnb->offset >> PAGE_SHIFT;
  59         lnb->page = page;
  60
  61         RETURN(0);
  62 }
  63
  64 void filter_free_dio_pages(int objcount, struct obd_ioobj *obj,
  65                            int niocount, struct niobuf_local *res)
  66 {
  67         int i, j;
  68
  69         for (i = 0; i < objcount; i++, obj++) {
  70                 for (j = 0 ; j < obj->ioo_bufcnt ; j++, res++) {
  71                         if (res->page != NULL) {
  72                                 __free_page(res->page);
  73                                 res->page = NULL;
  74                         }
  75                 }
  76         }
  77 }
  78
  79 /* Grab the dirty and seen grant announcements from the incoming obdo.
  80  * We will later calculate the clients new grant and return it.
  81  * Caller must hold osfs lock */
  82 static void filter_grant_incoming(struct obd_export *exp, struct obdo *oa)
  83 {
  84         struct filter_export_data *fed;
  85         struct obd_device *obd = exp->exp_obd;
  86         static unsigned long last_msg;
  87         static int last_count;
  88         int mask = D_CACHE;
  89         ENTRY;
  90
  91         LASSERT_SPIN_LOCKED(&obd->obd_osfs_lock);
  92
  93         if ((oa->o_valid & (OBD_MD_FLBLOCKS|OBD_MD_FLGRANT)) !=
  94                                         (OBD_MD_FLBLOCKS|OBD_MD_FLGRANT)) {
  95                 oa->o_valid &= ~OBD_MD_FLGRANT;
  96                 EXIT;
  97                 return;
  98         }
  99
 100         fed = &exp->exp_filter_data;
 101
 102         /* Don't print this to the console the first time it happens, since
 103          * it can happen legitimately on occasion, but only rarely. */
 104         if (time_after(jiffies, last_msg + 60 * HZ)) {
 105                 last_count = 0;
 106                 last_msg = jiffies;
 107         }
 108         if ((last_count & (-last_count)) == last_count)
 109                 mask = D_WARNING;
 110         last_count++;
 111
 112         /* Add some margin, since there is a small race if other RPCs arrive
 113          * out-or-order and have already consumed some grant.  We want to
 114          * leave this here in case there is a large error in accounting. */
 115         CDEBUG(oa->o_grant > fed->fed_grant + FILTER_GRANT_CHUNK ? mask:D_CACHE,
 116                "%s: cli %s/%p reports grant: "LPU64" dropped: %u, local: %lu\n",
 117                obd->obd_name, exp->exp_client_uuid.uuid, exp, oa->o_grant,
 118                oa->o_dropped, fed->fed_grant);
 119
 120         /* Update our accounting now so that statfs takes it into account.
 121          * Note that fed_dirty is only approximate and can become incorrect
 122          * if RPCs arrive out-of-order.  No important calculations depend
 123          * on fed_dirty however. */
 124         obd->u.filter.fo_tot_dirty += oa->o_dirty - fed->fed_dirty;
 125         if (fed->fed_grant < oa->o_dropped) {
 126                 CERROR("%s: cli %s/%p reports %u dropped > fed_grant %lu\n",
 127                        obd->obd_name, exp->exp_client_uuid.uuid, exp,
 128                        oa->o_dropped, fed->fed_grant);
 129                 oa->o_dropped = 0;
 130         }
 131         if (obd->u.filter.fo_tot_granted < oa->o_dropped) {
 132                 CERROR("%s: cli %s/%p reports %u dropped > tot_grant "LPU64"\n",
 133                        obd->obd_name, exp->exp_client_uuid.uuid, exp,
 134                        oa->o_dropped, obd->u.filter.fo_tot_granted);
 135                 oa->o_dropped = 0;
 136         }
 137         obd->u.filter.fo_tot_granted -= oa->o_dropped;
 138         fed->fed_grant -= oa->o_dropped;
 139         fed->fed_dirty = oa->o_dirty;
 140         if (fed->fed_dirty < 0 || fed->fed_grant < 0 || fed->fed_pending < 0) {
 141                 CERROR("%s: cli %s/%p dirty %ld pend %ld grant %ld\n",
 142                        obd->obd_name, exp->exp_client_uuid.uuid, exp,
 143                        fed->fed_dirty, fed->fed_pending, fed->fed_grant);
 144                 spin_unlock(&obd->obd_osfs_lock);
 145                 LBUG();
 146         }
 147         EXIT;
 148 }
 149
 150 #define GRANT_FOR_LLOG(obd) 16
 151
 152 /* Figure out how much space is available between what we've granted
 153  * and what remains in the filesystem.  Compensate for ext3 indirect
 154  * block overhead when computing how much free space is left ungranted.
 155  *
 156  * Caller must hold obd_osfs_lock. */
 157 obd_size filter_grant_space_left(struct obd_export *exp)
 158 {
 159         struct obd_device *obd = exp->exp_obd;
 160         int blockbits = obd->u.filter.fo_sb->s_blocksize_bits;
 161         obd_size tot_granted = obd->u.filter.fo_tot_granted, avail, left = 0;
 162         int rc, statfs_done = 0;
 163
 164         LASSERT_SPIN_LOCKED(&obd->obd_osfs_lock);
 165
 166         if (time_before(obd->obd_osfs_age, jiffies - HZ)) {
 167 restat:
 168                 rc = fsfilt_statfs(obd, obd->u.filter.fo_sb, jiffies + 1);
 169                 if (rc) /* N.B. statfs can't really fail */
 170                         RETURN(0);
 171                 statfs_done = 1;
 172         }
 173
 174         avail = obd->obd_osfs.os_bavail;
 175         left = avail - (avail >> (blockbits - 3)); /* (d)indirect */
 176         if (left > GRANT_FOR_LLOG(obd)) {
 177                 left = (left - GRANT_FOR_LLOG(obd)) << blockbits;
 178         } else {
 179                 left = 0 /* << blockbits */;
 180         }
 181
 182         if (!statfs_done && left < 32 * FILTER_GRANT_CHUNK + tot_granted) {
 183                 CDEBUG(D_CACHE, "fs has no space left and statfs too old\n");
 184                 goto restat;
 185         }
 186
 187         if (left >= tot_granted) {
 188                 left -= tot_granted;
 189         } else {
 190                 if (left < tot_granted - obd->u.filter.fo_tot_pending + 65536) {
 191                         CERROR("%s: cli %s/%p grant "LPU64" > available "
 192                                LPU64" and pending "LPU64"\n", obd->obd_name,
 193                                exp->exp_client_uuid.uuid, exp, tot_granted,
 194                                left, obd->u.filter.fo_tot_pending);
 195                 }
 196                 left = 0;
 197         }
 198
 199         CDEBUG(D_CACHE, "%s: cli %s/%p free: "LPU64" avail: "LPU64" grant "LPU64
 200                " left: "LPU64" pending: "LPU64"\n", obd->obd_name,
 201                exp->exp_client_uuid.uuid, exp,
 202                obd->obd_osfs.os_bfree << blockbits, avail << blockbits,
 203                tot_granted, left, obd->u.filter.fo_tot_pending);
 204
 205         return left;
 206 }
 207
 208 /* Calculate how much grant space to allocate to this client, based on how
 209  * much space is currently free and how much of that is already granted.
 210  *
 211  * Caller must hold obd_osfs_lock. */
 212 long filter_grant(struct obd_export *exp, obd_size current_grant,
 213                   obd_size want, obd_size fs_space_left)
 214 {
 215         struct obd_device *obd = exp->exp_obd;
 216         struct filter_export_data *fed = &exp->exp_filter_data;
 217         int blockbits = obd->u.filter.fo_sb->s_blocksize_bits;
 218         __u64 grant = 0;
 219
 220         LASSERT_SPIN_LOCKED(&obd->obd_osfs_lock);
 221
 222         /* Grant some fraction of the client's requested grant space so that
 223          * they are not always waiting for write credits (not all of it to
 224          * avoid overgranting in face of multiple RPCs in flight).  This
 225          * essentially will be able to control the OSC_MAX_RIF for a client.
 226          *
 227          * If we do have a large disparity between what the client thinks it
 228          * has and what we think it has, don't grant very much and let the
 229          * client consume its grant first.  Either it just has lots of RPCs
 230          * in flight, or it was evicted and its grants will soon be used up. */
 231         if (want > 0x7fffffff) {
 232                 CERROR("%s: client %s/%p requesting > 2GB grant "LPU64"\n",
 233                        obd->obd_name, exp->exp_client_uuid.uuid, exp, want);
 234         } else if (current_grant < want &&
 235                    current_grant < fed->fed_grant + FILTER_GRANT_CHUNK) {
 236                 grant = min((want >> blockbits) / 2,
 237                             (fs_space_left >> blockbits) / 8);
 238                 grant <<= blockbits;
 239
 240                 if (grant) {
 241                         if (grant > FILTER_GRANT_CHUNK)
 242                                 grant = FILTER_GRANT_CHUNK;
 243
 244                         obd->u.filter.fo_tot_granted += grant;
 245                         fed->fed_grant += grant;
 246                         if (fed->fed_grant < 0) {
 247                                 CERROR("%s: cli %s/%p grant %ld want "LPU64
 248                                        "current"LPU64"\n",
 249                                        obd->obd_name, exp->exp_client_uuid.uuid,
 250                                        exp, fed->fed_grant, want,current_grant);
 251                                 spin_unlock(&obd->obd_osfs_lock);
 252                                 LBUG();
 253                         }
 254                 }
 255         }
 256
 257         CDEBUG(D_CACHE,"%s: cli %s/%p wants: "LPU64" granting: "LPU64"\n",
 258                obd->obd_name, exp->exp_client_uuid.uuid, exp, want, grant);
 259         CDEBUG(D_CACHE,
 260                "%s: cli %s/%p tot cached:"LPU64" granted:"LPU64
 261                " num_exports: %d\n", obd->obd_name, exp->exp_client_uuid.uuid,
 262                exp, obd->u.filter.fo_tot_dirty,
 263                obd->u.filter.fo_tot_granted, obd->obd_num_exports);
 264
 265         return grant;
 266 }
 267
 268 static int filter_preprw_read(int cmd, struct obd_export *exp, struct obdo *oa,
 269                               int objcount, struct obd_ioobj *obj,
 270                               int niocount, struct niobuf_remote *nb,
 271                               struct niobuf_local *res,
 272                               struct obd_trans_info *oti)
 273 {
 274         struct obd_device *obd = exp->exp_obd;
 275         struct obd_run_ctxt saved;
 276         struct niobuf_remote *rnb;
 277         struct niobuf_local *lnb;
 278         struct dentry *dentry = NULL;
 279         struct inode *inode;
 280         void *iobuf = NULL;
 281         int rc = 0, i, tot_bytes = 0;
 282         unsigned long now = jiffies;
 283         ENTRY;
 284
 285         /* We are currently not supporting multi-obj BRW_READ RPCS at all.
 286          * When we do this function's dentry cleanup will need to be fixed */
 287         LASSERTF(objcount == 1, "%d\n", objcount);
 288         LASSERTF(obj->ioo_bufcnt > 0, "%d\n", obj->ioo_bufcnt);
 289
 290         if (oa && oa->o_valid & OBD_MD_FLGRANT) {
 291                 spin_lock(&obd->obd_osfs_lock);
 292                 filter_grant_incoming(exp, oa);
 293
 294                 oa->o_grant = 0;
 295                 spin_unlock(&obd->obd_osfs_lock);
 296         }
 297
 298         memset(res, 0, niocount * sizeof(*res));
 299
 300         push_ctxt(&saved, &exp->exp_obd->obd_ctxt, NULL);
 301
 302         rc = filter_alloc_iobuf(&obd->u.filter, OBD_BRW_READ, obj->ioo_bufcnt,
 303                                 &iobuf);
 304         if (rc)
 305                 GOTO(cleanup, rc);
 306
 307         dentry = filter_oa2dentry(obd, oa);
 308         if (IS_ERR(dentry)) {
 309                 rc = PTR_ERR(dentry);
 310                 dentry = NULL;
 311                 GOTO(cleanup, rc);
 312         }
 313
 314         inode = dentry->d_inode;
 315
 316         if (oa)
 317                 obdo_to_inode(inode, oa, OBD_MD_FLATIME);
 318
 319         fsfilt_check_slow(now, obd_timeout, "preprw_read setup");
 320
 321         for (i = 0, lnb = res, rnb = nb; i < obj->ioo_bufcnt;
 322              i++, rnb++, lnb++) {
 323                 lnb->dentry = dentry;
 324                 lnb->offset = rnb->offset;
 325                 lnb->len    = rnb->len;
 326                 lnb->flags  = rnb->flags;
 327
 328                 if (inode->i_size <= rnb->offset)
 329                       /* If there's no more data, abort early.
 330                       * lnb->page == NULL and lnb->rc == 0, so it's
 331                       * easy to detect later. */
 332                         break;
 333                 else
 334                         rc = filter_alloc_dio_page(obd, inode, lnb);
 335
 336                 if (rc) {
 337                         CDEBUG(rc == -ENOSPC ? D_INODE : D_ERROR,
 338                              "page err %u@"LPU64" %u/%u %p: rc %d\n",
 339                               lnb->len, lnb->offset, i, obj->ioo_bufcnt,
 340                               dentry, rc);
 341                         GOTO(cleanup, rc);
 342                 }
 343
 344                 if (inode->i_size < lnb->offset + lnb->len - 1)
 345                         lnb->rc = inode->i_size - lnb->offset;
 346                 else
 347                         lnb->rc = lnb->len;
 348
 349                 tot_bytes += lnb->rc;
 350
 351                 filter_iobuf_add_page(obd, iobuf, inode, lnb->page);
 352         }
 353
 354         fsfilt_check_slow(now, obd_timeout, "start_page_read");
 355
 356         rc = filter_direct_io(OBD_BRW_READ, dentry, iobuf, exp,
 357                               NULL, NULL, NULL);
 358         if (rc)
 359                 GOTO(cleanup, rc);
 360
 361         lprocfs_counter_add(obd->obd_stats, LPROC_FILTER_READ_BYTES, tot_bytes);
 362
 363         filter_tally_read(&exp->exp_obd->u.filter, res, niocount);
 364
 365         EXIT;
 366
 367  cleanup:
 368         if (rc != 0) {
 369                 filter_free_dio_pages(objcount, obj, niocount, res);
 370
 371                 if (dentry != NULL)
 372                         f_dput(dentry);
 373         }
 374
 375         if (iobuf != NULL)
 376                 filter_free_iobuf(iobuf);
 377
 378         pop_ctxt(&saved, &exp->exp_obd->obd_ctxt, NULL);
 379         if (rc)
 380                 CERROR("io error %d\n", rc);
 381
 382         return rc;
 383 }
 384
 385 /* When clients have dirtied as much space as they've been granted they
 386  * fall through to sync writes.  These sync writes haven't been expressed
 387  * in grants and need to error with ENOSPC when there isn't room in the
 388  * filesystem for them after grants are taken into account.  However,
 389  * writeback of the dirty data that was already granted space can write
 390  * right on through.
 391  *
 392  * Caller must hold obd_osfs_lock. */
 393 static int filter_grant_check(struct obd_export *exp, int objcount,
 394                               struct fsfilt_objinfo *fso, int niocount,
 395                               struct niobuf_remote *rnb,
 396                               struct niobuf_local *lnb, obd_size *left,
 397                               struct inode *inode)
 398 {
 399         struct filter_export_data *fed = &exp->exp_filter_data;
 400         int blocksize = exp->exp_obd->u.filter.fo_sb->s_blocksize;
 401         unsigned long used = 0, ungranted = 0, using;
 402         int i, rc = -ENOSPC, obj, n = 0, mask = D_CACHE;
 403
 404         LASSERT_SPIN_LOCKED(&exp->exp_obd->obd_osfs_lock);
 405
 406         for (obj = 0; obj < objcount; obj++) {
 407                 for (i = 0; i < fso[obj].fso_bufcnt; i++, n++) {
 408                         int tmp, bytes;
 409
 410                         /* should match the code in osc_exit_cache */
 411                         bytes = rnb[n].len;
 412                         bytes += rnb[n].offset & (blocksize - 1);
 413                         tmp = (rnb[n].offset + rnb[n].len) & (blocksize - 1);
 414                         if (tmp)
 415                                 bytes += blocksize - tmp;
 416
 417                         if (rnb[n].flags & OBD_BRW_FROM_GRANT) {
 418                                 if (fed->fed_grant < used + bytes) {
 419                                         CDEBUG(D_CACHE,
 420                                                "%s: cli %s/%p claims %ld+%d "
 421                                                "GRANT, real grant %lu idx %d\n",
 422                                                exp->exp_obd->obd_name,
 423                                                exp->exp_client_uuid.uuid, exp,
 424                                                used, bytes, fed->fed_grant, n);
 425                                         mask = D_ERROR;
 426                                 } else {
 427                                         used += bytes;
 428                                         rnb[n].flags |= OBD_BRW_GRANTED;
 429                                         lnb[n].lnb_grant_used = bytes;
 430                                         CDEBUG(0, "idx %d used=%lu\n", n, used);
 431                                         rc = 0;
 432                                         continue;
 433                                 }
 434                         }
 435                         if (*left > ungranted) {
 436                                 /* if enough space, pretend it was granted */
 437                                 ungranted += bytes;
 438                                 rnb[n].flags |= OBD_BRW_GRANTED;
 439                                 CDEBUG(0, "idx %d ungranted=%lu\n",n,ungranted);
 440                                 rc = 0;
 441                                 continue;
 442                         }
 443
 444                         /* We can't check for already-mapped blocks here, as
 445                          * it requires dropping the osfs lock to do the bmap.
 446                          * Instead, we return ENOSPC and in that case we need
 447                          * to go through and verify if all of the blocks not
 448                          * marked BRW_GRANTED are already mapped and we can
 449                          * ignore this error. */
 450                         lnb[n].rc = -ENOSPC;
 451                         rnb[n].flags &= OBD_BRW_GRANTED;
 452                         CDEBUG(D_CACHE,"%s: cli %s/%p idx %d no space for %d\n",
 453                                exp->exp_obd->obd_name,
 454                                exp->exp_client_uuid.uuid, exp, n, bytes);
 455                 }
 456         }
 457
 458         /* Now substract what client have used already.  We don't subtract
 459          * this from the tot_granted yet, so that other client's can't grab
 460          * that space before we have actually allocated our blocks.  That
 461          * happens in filter_grant_commit() after the writes are done. */
 462         *left -= ungranted;
 463         fed->fed_grant -= used;
 464         fed->fed_pending += used;
 465         exp->exp_obd->u.filter.fo_tot_pending += used;
 466
 467         CDEBUG(mask,
 468                "%s: cli %s/%p used: %lu ungranted: %lu grant: %lu dirty: %lu\n",
 469                exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp, used,
 470                ungranted, fed->fed_grant, fed->fed_dirty);
 471
 472         /* Rough calc in case we don't refresh cached statfs data */
 473         using = (used + ungranted + 1 ) >>
 474                 exp->exp_obd->u.filter.fo_sb->s_blocksize_bits;
 475         if (exp->exp_obd->obd_osfs.os_bavail > using)
 476                 exp->exp_obd->obd_osfs.os_bavail -= using;
 477         else
 478                 exp->exp_obd->obd_osfs.os_bavail = 0;
 479
 480         if (fed->fed_dirty < used) {
 481                 CERROR("%s: cli %s/%p claims used %lu > fed_dirty %lu\n",
 482                        exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp,
 483                        used, fed->fed_dirty);
 484                 used = fed->fed_dirty;
 485         }
 486         exp->exp_obd->u.filter.fo_tot_dirty -= used;
 487         fed->fed_dirty -= used;
 488
 489         if (fed->fed_dirty < 0 || fed->fed_grant < 0 || fed->fed_pending < 0) {
 490                 CERROR("%s: cli %s/%p dirty %ld pend %ld grant %ld\n",
 491                        exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp,
 492                        fed->fed_dirty, fed->fed_pending, fed->fed_grant);
 493                 spin_unlock(&exp->exp_obd->obd_osfs_lock);
 494                 LBUG();
 495         }
 496         return rc;
 497 }
 498
 499 /* If we ever start to support multi-object BRW RPCs, we will need to get locks
 500  * on mulitple inodes.  That isn't all, because there still exists the
 501  * possibility of a truncate starting a new transaction while holding the ext3
 502  * rwsem = write while some writes (which have started their transactions here)
 503  * blocking on the ext3 rwsem = read => lock inversion.
 504  *
 505  * The handling gets very ugly when dealing with locked pages.  It may be easier
 506  * to just get rid of the locked page code (which has problems of its own) and
 507  * either discover we do not need it anymore (i.e. it was a symptom of another
 508  * bug) or ensure we get the page locks in an appropriate order. */
 509 static int filter_preprw_write(int cmd, struct obd_export *exp, struct obdo *oa,
 510                                int objcount, struct obd_ioobj *obj,
 511                                int niocount, struct niobuf_remote *nb,
 512                                struct niobuf_local *res,
 513                                struct obd_trans_info *oti)
 514 {
 515         struct obd_run_ctxt saved;
 516         struct niobuf_remote *rnb;
 517         struct niobuf_local *lnb = res;
 518         struct fsfilt_objinfo fso;
 519         struct dentry *dentry = NULL;
 520         void *iobuf;
 521         obd_size left;
 522         unsigned long now = jiffies;
 523         int rc = 0, i, tot_bytes = 0, cleanup_phase = 0;
 524         ENTRY;
 525         LASSERT(objcount == 1);
 526         LASSERT(obj->ioo_bufcnt > 0);
 527
 528         memset(res, 0, niocount * sizeof(*res));
 529
 530         rc = filter_alloc_iobuf(&exp->exp_obd->u.filter, OBD_BRW_READ,
 531                                 obj->ioo_bufcnt, &iobuf);
 532         if (rc)
 533                 GOTO(cleanup, rc);
 534         cleanup_phase = 1;
 535
 536         push_ctxt(&saved, &exp->exp_obd->obd_ctxt, NULL);
 537         dentry = filter_fid2dentry(exp->exp_obd, NULL, obj->ioo_gr,
 538                                    obj->ioo_id);
 539         if (IS_ERR(dentry))
 540                 GOTO(cleanup, rc = PTR_ERR(dentry));
 541         cleanup_phase = 2;
 542
 543         if (dentry->d_inode == NULL) {
 544                 CERROR("trying to BRW to non-existent file "LPU64"\n",
 545                        obj->ioo_id);
 546                 GOTO(cleanup, rc = -ENOENT);
 547         }
 548
 549         fso.fso_dentry = dentry;
 550         fso.fso_bufcnt = obj->ioo_bufcnt;
 551
 552         fsfilt_check_slow(now, obd_timeout, "preprw_write setup");
 553
 554         spin_lock(&exp->exp_obd->obd_osfs_lock);
 555         if (oa) {
 556                 filter_grant_incoming(exp, oa);
 557                 obdo_to_inode(dentry->d_inode, oa,
 558                               OBD_MD_FLATIME | OBD_MD_FLMTIME | OBD_MD_FLCTIME);
 559         }
 560         cleanup_phase = 3;
 561
 562         left = filter_grant_space_left(exp);
 563
 564         rc = filter_grant_check(exp, objcount, &fso, niocount, nb, res,
 565                                 &left, dentry->d_inode);
 566
 567         /* We're finishing using body->oa as an input variable, so reset
 568          * o_valid here. */
 569         if (oa && oa->o_valid & OBD_MD_FLGRANT) {
 570                 oa->o_grant = filter_grant(exp,oa->o_grant,oa->o_undirty,left);
 571                 oa->o_valid = OBD_MD_FLGRANT;
 572         } else if (oa) {
 573                 oa->o_valid = 0;
 574         }
 575
 576         spin_unlock(&exp->exp_obd->obd_osfs_lock);
 577
 578         if (rc)
 579                 GOTO(cleanup, rc);
 580
 581         for (i = 0, rnb = nb, lnb = res; i < obj->ioo_bufcnt;
 582              i++, lnb++, rnb++) {
 583                 /* We still set up for ungranted pages so that granted pages
 584                  * can be written to disk as they were promised, and portals
 585                  * needs to keep the pages all aligned properly. */
 586                 lnb->dentry = dentry;
 587                 lnb->offset = rnb->offset;
 588                 lnb->len    = rnb->len;
 589                 lnb->flags  = rnb->flags;
 590
 591                 rc = filter_alloc_dio_page(exp->exp_obd, dentry->d_inode,lnb);
 592                 if (rc) {
 593                         CERROR("page err %u@"LPU64" %u/%u %p: rc %d\n",
 594                                lnb->len, lnb->offset,
 595                                i, obj->ioo_bufcnt, dentry, rc);
 596                         GOTO(cleanup, rc);
 597                 }
 598                 cleanup_phase = 4;
 599
 600                 /* If the filter writes a partial page, then has the file
 601                  * extended, the client will read in the whole page.  the
 602                  * filter has to be careful to zero the rest of the partial
 603                  * page on disk.  we do it by hand for partial extending
 604                  * writes, send_bio() is responsible for zeroing pages when
 605                  * asked to read unmapped blocks -- brw_kiovec() does this. */
 606                 if (lnb->len != PAGE_SIZE) {
 607                         __s64 maxidx;
 608
 609                         maxidx = ((dentry->d_inode->i_size + PAGE_SIZE - 1) >>
 610                                  PAGE_SHIFT) - 1;
 611                         if (maxidx >= lnb->page->index) {
 612                                 LL_CDEBUG_PAGE(D_PAGE, lnb->page, "write %u @ "
 613                                                LPU64" flg %x before EOF %llu\n",
 614                                                lnb->len, lnb->offset,lnb->flags,
 615                                                dentry->d_inode->i_size);
 616                                 filter_iobuf_add_page(exp->exp_obd, iobuf,
 617                                                       dentry->d_inode,
 618                                                       lnb->page);
 619                         } else {
 620                                 long off;
 621                                 char *p = kmap(lnb->page);
 622
 623                                 off = lnb->offset & ~PAGE_MASK;
 624                                 if (off)
 625                                         memset(p, 0, off);
 626                                 off = (lnb->offset + lnb->len) & ~PAGE_MASK;
 627                                 if (off)
 628                                         memset(p + off, 0, PAGE_SIZE - off);
 629                                 kunmap(lnb->page);
 630                         }
 631                 }
 632
 633                 if (lnb->rc == 0)
 634                         tot_bytes += lnb->len;
 635         }
 636
 637         rc = filter_direct_io(OBD_BRW_READ, dentry, iobuf, exp,
 638                               NULL, NULL, NULL);
 639
 640         fsfilt_check_slow(now, obd_timeout, "start_page_write");
 641
 642         lprocfs_counter_add(exp->exp_obd->obd_stats, LPROC_FILTER_WRITE_BYTES,
 643                             tot_bytes);
 644         EXIT;
 645 cleanup:
 646         switch(cleanup_phase) {
 647         case 4:
 648                 if (rc)
 649                         filter_free_dio_pages(objcount, obj, niocount, res);
 650         case 3:
 651                 pop_ctxt(&saved, &exp->exp_obd->obd_ctxt, NULL);
 652                 filter_free_iobuf(iobuf);
 653         case 2:
 654                 if (rc)
 655                         f_dput(dentry);
 656                 break;
 657         case 1:
 658                 spin_lock(&exp->exp_obd->obd_osfs_lock);
 659                 if (oa)
 660                         filter_grant_incoming(exp, oa);
 661                 spin_unlock(&exp->exp_obd->obd_osfs_lock);
 662                 pop_ctxt(&saved, &exp->exp_obd->obd_ctxt, NULL);
 663                 filter_free_iobuf(iobuf);
 664                 break;
 665         default:;
 666         }
 667         return rc;
 668 }
 669
 670 int filter_preprw(int cmd, struct obd_export *exp, struct obdo *oa,
 671                   int objcount, struct obd_ioobj *obj, int niocount,
 672                   struct niobuf_remote *nb, struct niobuf_local *res,
 673                   struct obd_trans_info *oti)
 674 {
 675         if (cmd == OBD_BRW_WRITE)
 676                 return filter_preprw_write(cmd, exp, oa, objcount, obj,
 677                                            niocount, nb, res, oti);
 678
 679         if (cmd == OBD_BRW_READ)
 680                 return filter_preprw_read(cmd, exp, oa, objcount, obj,
 681                                           niocount, nb, res, oti);
 682
 683         LBUG();
 684         return -EPROTO;
 685 }
 686
 687 void filter_release_read_page(struct filter_obd *filter, struct inode *inode,
 688                               struct page *page)
 689 {
 690         int drop = 0;
 691
 692         if (inode != NULL &&
 693             (inode->i_size > filter->fo_readcache_max_filesize))
 694                 drop = 1;
 695
 696         /* drop from cache like truncate_list_pages() */
 697         if (drop && !TryLockPage(page)) {
 698                 if (page->mapping)
 699                         ll_truncate_complete_page(page);
 700                 unlock_page(page);
 701         }
 702         page_cache_release(page);
 703 }
 704
 705 static int filter_commitrw_read(struct obd_export *exp, struct obdo *oa,
 706                                 int objcount, struct obd_ioobj *obj,
 707                                 int niocount, struct niobuf_local *res,
 708                                 struct obd_trans_info *oti, int rc)
 709 {
 710         struct inode *inode = NULL;
 711         ENTRY;
 712
 713         if (res->dentry != NULL)
 714                 inode = res->dentry->d_inode;
 715
 716         filter_free_dio_pages(objcount, obj, niocount, res);
 717
 718         if (res->dentry != NULL)
 719                 f_dput(res->dentry);
 720         RETURN(rc);
 721 }
 722
 723 void flip_into_page_cache(struct inode *inode, struct page *new_page)
 724 {
 725         struct page *old_page;
 726         int rc;
 727
 728         do {
 729                 /* the dlm is protecting us from read/write concurrency, so we
 730                  * expect this find_lock_page to return quickly.  even if we
 731                  * race with another writer it won't be doing much work with
 732                  * the page locked.  we do this 'cause t_c_p expects a
 733                  * locked page, and it wants to grab the pagecache lock
 734                  * as well. */
 735                 old_page = find_lock_page(inode->i_mapping, new_page->index);
 736                 if (old_page) {
 737                         ll_truncate_complete_page(old_page);
 738                         unlock_page(old_page);
 739                         page_cache_release(old_page);
 740                 }
 741
 742 #if 0 /* this should be a /proc tunable someday */
 743                 /* racing o_directs (no locking ioctl) could race adding
 744                  * their pages, so we repeat the page invalidation unless
 745                  * we successfully added our new page */
 746                 rc = add_to_page_cache_unique(new_page, inode->i_mapping,
 747                                               new_page->index,
 748                                               page_hash(inode->i_mapping,
 749                                                         new_page->index));
 750                 if (rc == 0) {
 751                         /* add_to_page_cache clears uptodate|dirty and locks
 752                          * the page */
 753                         SetPageUptodate(new_page);
 754                         unlock_page(new_page);
 755                 }
 756 #else
 757                 rc = 0;
 758 #endif
 759         } while (rc != 0);
 760 }
 761
 762 void filter_grant_commit(struct obd_export *exp, int niocount,
 763                          struct niobuf_local *res)
 764 {
 765         struct filter_obd *filter = &exp->exp_obd->u.filter;
 766         struct niobuf_local *lnb = res;
 767         unsigned long pending = 0;
 768         int i;
 769
 770         spin_lock(&exp->exp_obd->obd_osfs_lock);
 771         for (i = 0, lnb = res; i < niocount; i++, lnb++)
 772                 pending += lnb->lnb_grant_used;
 773
 774         LASSERTF(exp->exp_filter_data.fed_pending >= pending,
 775                  "%s: cli %s/%p fed_pending: %lu grant_used: %lu\n",
 776                  exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp,
 777                  exp->exp_filter_data.fed_pending, pending);
 778         exp->exp_filter_data.fed_pending -= pending;
 779         LASSERTF(filter->fo_tot_granted >= pending,
 780                  "%s: cli %s/%p tot_granted: "LPU64" grant_used: %lu\n",
 781                  exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp,
 782                  exp->exp_obd->u.filter.fo_tot_granted, pending);
 783         filter->fo_tot_granted -= pending;
 784         LASSERTF(filter->fo_tot_pending >= pending,
 785                  "%s: cli %s/%p tot_pending: "LPU64" grant_used: %lu\n",
 786                  exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp,
 787                  filter->fo_tot_pending, pending);
 788         filter->fo_tot_pending -= pending;
 789
 790         spin_unlock(&exp->exp_obd->obd_osfs_lock);
 791 }
 792
 793 int filter_commitrw(int cmd, struct obd_export *exp, struct obdo *oa,
 794                     int objcount, struct obd_ioobj *obj, int niocount,
 795                     struct niobuf_local *res, struct obd_trans_info *oti,int rc)
 796 {
 797         if (cmd == OBD_BRW_WRITE)
 798                 return filter_commitrw_write(exp, oa, objcount, obj, niocount,
 799                                              res, oti, rc);
 800         if (cmd == OBD_BRW_READ)
 801                 return filter_commitrw_read(exp, oa, objcount, obj, niocount,
 802                                             res, oti, rc);
 803         LBUG();
 804         return -EPROTO;
 805 }
 806
 807 int filter_brw(int cmd, struct obd_export *exp, struct obdo *oa,
 808                struct lov_stripe_md *lsm, obd_count oa_bufs,
 809                struct brw_page *pga, struct obd_trans_info *oti)
 810 {
 811         struct obd_ioobj ioo;
 812         struct niobuf_local *lnb;
 813         struct niobuf_remote *rnb;
 814         obd_count i;
 815         int ret = 0;
 816         ENTRY;
 817
 818         OBD_ALLOC(lnb, oa_bufs * sizeof(struct niobuf_local));
 819         OBD_ALLOC(rnb, oa_bufs * sizeof(struct niobuf_remote));
 820
 821         if (lnb == NULL || rnb == NULL)
 822                 GOTO(out, ret = -ENOMEM);
 823
 824         for (i = 0; i < oa_bufs; i++) {
 825                 rnb[i].offset = pga[i].off;
 826                 rnb[i].len = pga[i].count;
 827         }
 828
 829         obdo_to_ioobj(oa, &ioo);
 830         ioo.ioo_bufcnt = oa_bufs;
 831
 832         ret = filter_preprw(cmd, exp, oa, 1, &ioo, oa_bufs, rnb, lnb, oti);
 833         if (ret != 0)
 834                 GOTO(out, ret);
 835
 836         for (i = 0; i < oa_bufs; i++) {
 837                 void *virt;
 838                 obd_off off;
 839                 void *addr;
 840
 841                 if (lnb[i].page == NULL)
 842                         break;
 843
 844                 off = pga[i].off & ~PAGE_MASK;
 845                 virt = kmap(pga[i].pg);
 846                 addr = kmap(lnb[i].page);
 847
 848                 /* 2 kmaps == vanishingly small deadlock opportunity */
 849
 850                 if (cmd & OBD_BRW_WRITE)
 851                         memcpy(addr + off, virt + off, pga[i].count);
 852                 else
 853                         memcpy(virt + off, addr + off, pga[i].count);
 854
 855                 kunmap(lnb[i].page);
 856                 kunmap(pga[i].pg);
 857         }
 858
 859         ret = filter_commitrw(cmd, exp, oa, 1, &ioo, oa_bufs, lnb, oti, ret);
 860
 861 out:
 862         if (lnb)
 863                 OBD_FREE(lnb, oa_bufs * sizeof(struct niobuf_local));
 864         if (rnb)
 865                 OBD_FREE(rnb, oa_bufs * sizeof(struct niobuf_remote));
 866         RETURN(ret);
 867 }