lustre/obdfilter/filter.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.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  *
  10  *   This file is part of Lustre, http://www.lustre.org.
  11  *
  12  *   Lustre is free software; you can redistribute it and/or
  13  *   modify it under the terms of version 2 of the GNU General Public
  14  *   License as published by the Free Software Foundation.
  15  *
  16  *   Lustre is distributed in the hope that it will be useful,
  17  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  19  *   GNU General Public License for more details.
  20  *
  21  *   You should have received a copy of the GNU General Public License
  22  *   along with Lustre; if not, write to the Free Software
  23  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  24  */
  25
  26 /*
  27  * Invariant: Get O/R i_sem for lookup, if needed, before any journal ops
  28  *            (which need to get journal_lock, may block if journal full).
  29  *
  30  * Invariant: Call filter_start_transno() before any journal ops to avoid the
  31  *            same deadlock problem.  We can (and want) to get rid of the
  32  *            transno sem in favour of the dir/inode i_sem to avoid single
  33  *            threaded operation on the OST.
  34  */
  35
  36 #define DEBUG_SUBSYSTEM S_FILTER
  37
  38 #include <linux/config.h>
  39 #include <linux/module.h>
  40 #include <linux/fs.h>
  41 #include <linux/dcache.h>
  42 #include <linux/init.h>
  43 #include <linux/version.h>
  44 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
  45 # include <linux/mount.h>
  46 # include <linux/buffer_head.h>
  47 #endif
  48
  49 #include <linux/obd_class.h>
  50 #include <linux/obd_lov.h>
  51 #include <linux/lustre_dlm.h>
  52 #include <linux/lustre_fsfilt.h>
  53 #include <linux/lprocfs_status.h>
  54 #include <linux/lustre_log.h>
  55 #include <linux/lustre_commit_confd.h>
  56 #include <portals/list.h>
  57
  58 #include "filter_internal.h"
  59
  60 static struct lvfs_callback_ops filter_lvfs_ops;
  61
  62 static int filter_destroy(struct obd_export *exp, struct obdo *oa,
  63                           struct lov_stripe_md *ea, struct obd_trans_info *);
  64
  65 static void filter_commit_cb(struct obd_device *obd, __u64 transno,
  66                              void *cb_data, int error)
  67 {
  68         obd_transno_commit_cb(obd, transno, error);
  69 }
  70
  71
  72 /* Assumes caller has already pushed us into the kernel context. */
  73 int filter_finish_transno(struct obd_export *exp, struct obd_trans_info *oti,
  74                           int rc)
  75 {
  76         struct filter_obd *filter = &exp->exp_obd->u.filter;
  77         struct filter_export_data *fed = &exp->exp_filter_data;
  78         struct filter_client_data *fcd = fed->fed_fcd;
  79         __u64 last_rcvd;
  80         loff_t off;
  81         int err, log_pri = D_HA;
  82
  83         /* Propagate error code. */
  84         if (rc)
  85                 RETURN(rc);
  86
  87         if (!exp->exp_obd->obd_replayable || oti == NULL)
  88                 RETURN(rc);
  89
  90         /* we don't allocate new transnos for replayed requests */
  91         if (oti->oti_transno == 0) {
  92                 spin_lock(&filter->fo_translock);
  93                 last_rcvd = le64_to_cpu(filter->fo_fsd->fsd_last_transno) + 1;
  94                 filter->fo_fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
  95                 spin_unlock(&filter->fo_translock);
  96                 oti->oti_transno = last_rcvd;
  97         } else {
  98                 spin_lock(&filter->fo_translock);
  99                 last_rcvd = oti->oti_transno;
 100                 if (last_rcvd > le64_to_cpu(filter->fo_fsd->fsd_last_transno))
 101                         filter->fo_fsd->fsd_last_transno =
 102                                 cpu_to_le64(last_rcvd);
 103                 spin_unlock(&filter->fo_translock);
 104         }
 105         fcd->fcd_last_rcvd = cpu_to_le64(last_rcvd);
 106
 107         /* could get xid from oti, if it's ever needed */
 108         fcd->fcd_last_xid = 0;
 109
 110         off = fed->fed_lr_off;
 111         fsfilt_add_journal_cb(exp->exp_obd, last_rcvd, oti->oti_handle,
 112                               filter_commit_cb, NULL);
 113         err = fsfilt_write_record(exp->exp_obd, filter->fo_rcvd_filp, fcd,
 114                                   sizeof(*fcd), &off, 0);
 115         if (err) {
 116                 log_pri = D_ERROR;
 117                 if (rc == 0)
 118                         rc = err;
 119         }
 120
 121         CDEBUG(log_pri, "wrote trans "LPU64" for client %s at #%d: err = %d\n",
 122                last_rcvd, fcd->fcd_uuid, fed->fed_lr_idx, err);
 123
 124         RETURN(rc);
 125 }
 126
 127 void f_dput(struct dentry *dentry)
 128 {
 129         /* Can't go inside filter_ddelete because it can block */
 130         CDEBUG(D_INODE, "putting %s: %p, count = %d\n",
 131                dentry->d_name.name, dentry, atomic_read(&dentry->d_count) - 1);
 132         LASSERT(atomic_read(&dentry->d_count) > 0);
 133
 134         dput(dentry);
 135 }
 136
 137 /* Add client data to the FILTER.  We use a bitmap to locate a free space
 138  * in the last_rcvd file if cl_idx is -1 (i.e. a new client).
 139  * Otherwise, we have just read the data from the last_rcvd file and
 140  * we know its offset. */
 141 static int filter_client_add(struct obd_device *obd, struct filter_obd *filter,
 142                              struct filter_export_data *fed, int cl_idx)
 143 {
 144         unsigned long *bitmap = filter->fo_last_rcvd_slots;
 145         int new_client = (cl_idx == -1);
 146         ENTRY;
 147
 148         LASSERT(bitmap != NULL);
 149
 150         /* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
 151         if (!strcmp(fed->fed_fcd->fcd_uuid, obd->obd_uuid.uuid))
 152                 RETURN(0);
 153
 154         /* the bitmap operations can handle cl_idx > sizeof(long) * 8, so
 155          * there's no need for extra complication here
 156          */
 157         if (new_client) {
 158                 cl_idx = find_first_zero_bit(bitmap, FILTER_LR_MAX_CLIENTS);
 159         repeat:
 160                 if (cl_idx >= FILTER_LR_MAX_CLIENTS) {
 161                         CERROR("no client slots - fix FILTER_LR_MAX_CLIENTS\n");
 162                         RETURN(-ENOMEM);
 163                 }
 164                 if (test_and_set_bit(cl_idx, bitmap)) {
 165                         CERROR("FILTER client %d: found bit is set in bitmap\n",
 166                                cl_idx);
 167                         cl_idx = find_next_zero_bit(bitmap,
 168                                                     FILTER_LR_MAX_CLIENTS,
 169                                                     cl_idx);
 170                         goto repeat;
 171                 }
 172         } else {
 173                 if (test_and_set_bit(cl_idx, bitmap)) {
 174                         CERROR("FILTER client %d: bit already set in bitmap!\n",
 175                                cl_idx);
 176                         LBUG();
 177                 }
 178         }
 179
 180         fed->fed_lr_idx = cl_idx;
 181         fed->fed_lr_off = le32_to_cpu(filter->fo_fsd->fsd_client_start) +
 182                 cl_idx * le16_to_cpu(filter->fo_fsd->fsd_client_size);
 183
 184         CDEBUG(D_INFO, "client at index %d (%llu) with UUID '%s' added\n",
 185                fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
 186
 187         if (new_client) {
 188                 struct obd_run_ctxt saved;
 189                 loff_t off = fed->fed_lr_off;
 190                 int err;
 191                 void *handle;
 192
 193                 CDEBUG(D_INFO, "writing client fcd at idx %u (%llu) (len %u)\n",
 194                        fed->fed_lr_idx,off,(unsigned int)sizeof(*fed->fed_fcd));
 195
 196                 push_ctxt(&saved, &obd->obd_ctxt, NULL);
 197                 /* Transaction needed to fix bug 1403 */
 198                 handle = fsfilt_start(obd,
 199                                       filter->fo_rcvd_filp->f_dentry->d_inode,
 200                                       FSFILT_OP_SETATTR, NULL);
 201                 if (IS_ERR(handle)) {
 202                         err = PTR_ERR(handle);
 203                         CERROR("unable to start transaction: rc %d\n", err);
 204                 } else {
 205                         err = fsfilt_write_record(obd, filter->fo_rcvd_filp,
 206                                                   fed->fed_fcd,
 207                                                   sizeof(*fed->fed_fcd),
 208                                                   &off, 1);
 209                         fsfilt_commit(obd,
 210                                       filter->fo_rcvd_filp->f_dentry->d_inode,
 211                                       handle, 1);
 212                 }
 213                 pop_ctxt(&saved, &obd->obd_ctxt, NULL);
 214
 215                 if (err) {
 216                         CERROR("error writing %s client idx %u: rc %d\n",
 217                                LAST_RCVD, fed->fed_lr_idx, err);
 218                         RETURN(err);
 219                 }
 220         }
 221         RETURN(0);
 222 }
 223
 224 static int filter_client_free(struct obd_export *exp, int flags)
 225 {
 226         struct filter_export_data *fed = &exp->exp_filter_data;
 227         struct filter_obd *filter = &exp->exp_obd->u.filter;
 228         struct obd_device *obd = exp->exp_obd;
 229         struct filter_client_data zero_fcd;
 230         struct obd_run_ctxt saved;
 231         int rc;
 232         loff_t off;
 233         ENTRY;
 234
 235         if (fed->fed_fcd == NULL)
 236                 RETURN(0);
 237
 238         if (flags & OBD_OPT_FAILOVER)
 239                 GOTO(free, 0);
 240
 241         /* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
 242         if (strcmp(fed->fed_fcd->fcd_uuid, obd->obd_uuid.uuid ) == 0)
 243                 GOTO(free, 0);
 244
 245         LASSERT(filter->fo_last_rcvd_slots != NULL);
 246
 247         off = fed->fed_lr_off;
 248
 249         CDEBUG(D_INFO, "freeing client at idx %u (%lld) with UUID '%s'\n",
 250                fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
 251
 252         /* Clear the bit _after_ zeroing out the client so we don't
 253            race with filter_client_add and zero out new clients.*/
 254         if (!test_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
 255                 CERROR("FILTER client %u: bit already clear in bitmap!!\n",
 256                        fed->fed_lr_idx);
 257                 LBUG();
 258         }
 259
 260         memset(&zero_fcd, 0, sizeof zero_fcd);
 261         push_ctxt(&saved, &obd->obd_ctxt, NULL);
 262         rc = fsfilt_write_record(obd, filter->fo_rcvd_filp, &zero_fcd,
 263                                  sizeof(zero_fcd), &off, 1);
 264         pop_ctxt(&saved, &obd->obd_ctxt, NULL);
 265
 266         CDEBUG(rc == 0 ? D_INFO : D_ERROR,
 267                "zeroing disconnecting client %s at idx %u (%llu) in %s rc %d\n",
 268                fed->fed_fcd->fcd_uuid, fed->fed_lr_idx, fed->fed_lr_off,
 269                LAST_RCVD, rc);
 270
 271         if (!test_and_clear_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
 272                 CERROR("FILTER client %u: bit already clear in bitmap!!\n",
 273                        fed->fed_lr_idx);
 274                 LBUG();
 275         }
 276
 277 free:
 278         OBD_FREE(fed->fed_fcd, sizeof(*fed->fed_fcd));
 279
 280         RETURN(0);
 281 }
 282
 283 static int filter_free_server_data(struct filter_obd *filter)
 284 {
 285         OBD_FREE(filter->fo_fsd, sizeof(*filter->fo_fsd));
 286         filter->fo_fsd = NULL;
 287         OBD_FREE(filter->fo_last_rcvd_slots,
 288                  FILTER_LR_MAX_CLIENT_WORDS * sizeof(unsigned long));
 289         filter->fo_last_rcvd_slots = NULL;
 290         return 0;
 291 }
 292
 293 /* assumes caller is already in kernel ctxt */
 294 int filter_update_server_data(struct obd_device *obd, struct file *filp,
 295                               struct filter_server_data *fsd, int force_sync)
 296 {
 297         loff_t off = 0;
 298         int rc;
 299         ENTRY;
 300
 301         CDEBUG(D_INODE, "server uuid      : %s\n", fsd->fsd_uuid);
 302         CDEBUG(D_INODE, "server last_rcvd : "LPU64"\n",
 303                le64_to_cpu(fsd->fsd_last_transno));
 304         CDEBUG(D_INODE, "server last_mount: "LPU64"\n",
 305                le64_to_cpu(fsd->fsd_mount_count));
 306
 307         rc = fsfilt_write_record(obd, filp, fsd, sizeof(*fsd), &off,force_sync);
 308         if (rc)
 309                 CERROR("error writing filter_server_data: rc = %d\n", rc);
 310
 311         RETURN(rc);
 312 }
 313
 314 int filter_update_last_objid(struct obd_device *obd, obd_gr group,
 315                              int force_sync)
 316 {
 317         struct filter_obd *filter = &obd->u.filter;
 318         __u64 tmp;
 319         loff_t off = 0;
 320         int rc;
 321         ENTRY;
 322
 323         CDEBUG(D_INODE, "server last_objid for group "LPU64": "LPU64"\n",
 324                group, filter->fo_last_objids[group]);
 325
 326         tmp = cpu_to_le64(filter->fo_last_objids[group]);
 327         rc = fsfilt_write_record(obd, filter->fo_last_objid_files[group],
 328                                  &tmp, sizeof(tmp), &off, force_sync);
 329         if (rc)
 330                 CERROR("error writing group "LPU64" last objid: rc = %d\n",
 331                        group, rc);
 332         RETURN(rc);
 333 }
 334
 335 /* assumes caller has already in kernel ctxt */
 336 static int filter_init_server_data(struct obd_device *obd, struct file * filp)
 337 {
 338         struct filter_obd *filter = &obd->u.filter;
 339         struct filter_server_data *fsd;
 340         struct filter_client_data *fcd = NULL;
 341         struct inode *inode = filp->f_dentry->d_inode;
 342         unsigned long last_rcvd_size = inode->i_size;
 343         __u64 mount_count;
 344         int cl_idx;
 345         loff_t off = 0;
 346         int rc;
 347
 348         /* ensure padding in the struct is the correct size */
 349         LASSERT (offsetof(struct filter_server_data, fsd_padding) +
 350                  sizeof(fsd->fsd_padding) == FILTER_LR_SERVER_SIZE);
 351         LASSERT (offsetof(struct filter_client_data, fcd_padding) +
 352                  sizeof(fcd->fcd_padding) == FILTER_LR_CLIENT_SIZE);
 353
 354         OBD_ALLOC(fsd, sizeof(*fsd));
 355         if (!fsd)
 356                 RETURN(-ENOMEM);
 357         filter->fo_fsd = fsd;
 358
 359         OBD_ALLOC(filter->fo_last_rcvd_slots,
 360                   FILTER_LR_MAX_CLIENT_WORDS * sizeof(unsigned long));
 361         if (filter->fo_last_rcvd_slots == NULL) {
 362                 OBD_FREE(fsd, sizeof(*fsd));
 363                 RETURN(-ENOMEM);
 364         }
 365
 366         if (last_rcvd_size == 0) {
 367                 CWARN("%s: initializing new %s\n", obd->obd_name, LAST_RCVD);
 368
 369                 memcpy(fsd->fsd_uuid, obd->obd_uuid.uuid,sizeof(fsd->fsd_uuid));
 370                 fsd->fsd_last_transno = 0;
 371                 mount_count = fsd->fsd_mount_count = 0;
 372                 fsd->fsd_server_size = cpu_to_le32(FILTER_LR_SERVER_SIZE);
 373                 fsd->fsd_client_start = cpu_to_le32(FILTER_LR_CLIENT_START);
 374                 fsd->fsd_client_size = cpu_to_le16(FILTER_LR_CLIENT_SIZE);
 375                 fsd->fsd_subdir_count = cpu_to_le16(FILTER_SUBDIR_COUNT);
 376                 filter->fo_subdir_count = FILTER_SUBDIR_COUNT;
 377         } else {
 378                 rc = fsfilt_read_record(obd, filp, fsd, sizeof(*fsd), &off);
 379                 if (rc) {
 380                         CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
 381                                LAST_RCVD, rc);
 382                         GOTO(err_fsd, rc);
 383                 }
 384                 if (strcmp(fsd->fsd_uuid, obd->obd_uuid.uuid) != 0) {
 385                         CERROR("OBD UUID %s does not match last_rcvd UUID %s\n",
 386                                obd->obd_uuid.uuid, fsd->fsd_uuid);
 387                         GOTO(err_fsd, rc = -EINVAL);
 388                 }
 389                 mount_count = le64_to_cpu(fsd->fsd_mount_count);
 390                 filter->fo_subdir_count = le16_to_cpu(fsd->fsd_subdir_count);
 391         }
 392
 393         if (fsd->fsd_feature_incompat & ~cpu_to_le32(FILTER_INCOMPAT_SUPP)) {
 394                 CERROR("unsupported feature %x\n",
 395                        le32_to_cpu(fsd->fsd_feature_incompat) &
 396                        ~FILTER_INCOMPAT_SUPP);
 397                 GOTO(err_fsd, rc = -EINVAL);
 398         }
 399         if (fsd->fsd_feature_rocompat & ~cpu_to_le32(FILTER_ROCOMPAT_SUPP)) {
 400                 CERROR("read-only feature %x\n",
 401                        le32_to_cpu(fsd->fsd_feature_rocompat) &
 402                        ~FILTER_ROCOMPAT_SUPP);
 403                 /* Do something like remount filesystem read-only */
 404                 GOTO(err_fsd, rc = -EINVAL);
 405         }
 406
 407         CDEBUG(D_INODE, "%s: server last_rcvd : "LPU64"\n",
 408                obd->obd_name, le64_to_cpu(fsd->fsd_last_transno));
 409         CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
 410                obd->obd_name, mount_count + 1);
 411         CDEBUG(D_INODE, "%s: server data size: %u\n",
 412                obd->obd_name, le32_to_cpu(fsd->fsd_server_size));
 413         CDEBUG(D_INODE, "%s: per-client data start: %u\n",
 414                obd->obd_name, le32_to_cpu(fsd->fsd_client_start));
 415         CDEBUG(D_INODE, "%s: per-client data size: %u\n",
 416                obd->obd_name, le32_to_cpu(fsd->fsd_client_size));
 417         CDEBUG(D_INODE, "%s: server subdir_count: %u\n",
 418                obd->obd_name, le16_to_cpu(fsd->fsd_subdir_count));
 419         CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
 420                last_rcvd_size <= le32_to_cpu(fsd->fsd_client_start) ? 0 :
 421                (last_rcvd_size - le32_to_cpu(fsd->fsd_client_start)) /
 422                 le16_to_cpu(fsd->fsd_client_size));
 423
 424         if (!obd->obd_replayable) {
 425                 CWARN("%s: recovery support OFF\n", obd->obd_name);
 426                 GOTO(out, rc = 0);
 427         }
 428
 429         for (cl_idx = 0, off = le32_to_cpu(fsd->fsd_client_start);
 430              off < last_rcvd_size; cl_idx++) {
 431                 __u64 last_rcvd;
 432                 struct obd_export *exp;
 433                 struct filter_export_data *fed;
 434
 435                 if (!fcd) {
 436                         OBD_ALLOC(fcd, sizeof(*fcd));
 437                         if (!fcd)
 438                                 GOTO(err_client, rc = -ENOMEM);
 439                 }
 440
 441                 /* Don't assume off is incremented properly by
 442                  * fsfilt_read_record(), in case sizeof(*fcd)
 443                  * isn't the same as fsd->fsd_client_size.  */
 444                 off = le32_to_cpu(fsd->fsd_client_start) +
 445                         cl_idx * le16_to_cpu(fsd->fsd_client_size);
 446                 rc = fsfilt_read_record(obd, filp, fcd, sizeof(*fcd), &off);
 447                 if (rc) {
 448                         CERROR("error reading FILT %s idx %d off %llu: rc %d\n",
 449                                LAST_RCVD, cl_idx, off, rc);
 450                         break; /* read error shouldn't cause startup to fail */
 451                 }
 452
 453                 if (fcd->fcd_uuid[0] == '\0') {
 454                         CDEBUG(D_INFO, "skipping zeroed client at offset %d\n",
 455                                cl_idx);
 456                         continue;
 457                 }
 458
 459                 last_rcvd = le64_to_cpu(fcd->fcd_last_rcvd);
 460
 461                 /* These exports are cleaned up by filter_disconnect(), so they
 462                  * need to be set up like real exports as filter_connect() does.
 463                  */
 464                 exp = class_new_export(obd);
 465                 CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64
 466                        " srv lr: "LPU64"\n", fcd->fcd_uuid, cl_idx,
 467                        last_rcvd, le64_to_cpu(fsd->fsd_last_transno));
 468                 if (exp == NULL)
 469                         GOTO(err_client, rc = -ENOMEM);
 470
 471                 memcpy(&exp->exp_client_uuid.uuid, fcd->fcd_uuid,
 472                        sizeof exp->exp_client_uuid.uuid);
 473                 fed = &exp->exp_filter_data;
 474                 fed->fed_fcd = fcd;
 475                 filter_client_add(obd, filter, fed, cl_idx);
 476                 /* create helper if export init gets more complex */
 477                 spin_lock_init(&fed->fed_lock);
 478
 479                 fcd = NULL;
 480                 obd->obd_recoverable_clients++;
 481                 class_export_put(exp);
 482
 483                 CDEBUG(D_OTHER, "client at idx %d has last_rcvd = "LPU64"\n",
 484                        cl_idx, last_rcvd);
 485
 486                 if (last_rcvd > le64_to_cpu(fsd->fsd_last_transno))
 487                         fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
 488
 489         }
 490
 491         obd->obd_last_committed = le64_to_cpu(fsd->fsd_last_transno);
 492
 493         if (obd->obd_recoverable_clients) {
 494                 CWARN("RECOVERY: %d recoverable clients, last_rcvd "
 495                       LPU64"\n", obd->obd_recoverable_clients,
 496                       le64_to_cpu(fsd->fsd_last_transno));
 497                 obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
 498                 obd->obd_recovering = 1;
 499         }
 500
 501         if (fcd)
 502                 OBD_FREE(fcd, sizeof(*fcd));
 503
 504 out:
 505         filter->fo_mount_count = mount_count + 1;
 506         fsd->fsd_mount_count = cpu_to_le64(filter->fo_mount_count);
 507
 508         /* save it, so mount count and last_transno is current */
 509         rc = filter_update_server_data(obd, filp, filter->fo_fsd, 1);
 510
 511         RETURN(rc);
 512
 513 err_client:
 514         class_disconnect_exports(obd, 0);
 515 err_fsd:
 516         filter_free_server_data(filter);
 517         RETURN(rc);
 518 }
 519
 520 static int filter_cleanup_groups(struct obd_device *obd)
 521 {
 522         struct filter_obd *filter = &obd->u.filter;
 523         struct file *filp;
 524         struct dentry *dentry;
 525         int i;
 526         ENTRY;
 527
 528         if (filter->fo_dentry_O_groups != NULL) {
 529                 for (i = 0; i < FILTER_GROUPS; i++) {
 530                         dentry = filter->fo_dentry_O_groups[i];
 531                         if (dentry != NULL)
 532                                 f_dput(dentry);
 533                 }
 534                 OBD_FREE(filter->fo_dentry_O_groups,
 535                          FILTER_GROUPS * sizeof(*filter->fo_dentry_O_groups));
 536                 filter->fo_dentry_O_groups = NULL;
 537         }
 538         if (filter->fo_last_objid_files != NULL) {
 539                 for (i = 0; i < FILTER_GROUPS; i++) {
 540                         filp = filter->fo_last_objid_files[i];
 541                         if (filp != NULL)
 542                                 filp_close(filp, 0);
 543                 }
 544                 OBD_FREE(filter->fo_last_objid_files,
 545                          FILTER_GROUPS * sizeof(*filter->fo_last_objid_files));
 546                 filter->fo_last_objid_files = NULL;
 547         }
 548         if (filter->fo_dentry_O_sub != NULL) {
 549                 for (i = 0; i < filter->fo_subdir_count; i++) {
 550                         dentry = filter->fo_dentry_O_sub[i];
 551                         if (dentry != NULL)
 552                                 f_dput(dentry);
 553                 }
 554                 OBD_FREE(filter->fo_dentry_O_sub,
 555                          filter->fo_subdir_count *
 556                          sizeof(*filter->fo_dentry_O_sub));
 557                 filter->fo_dentry_O_sub = NULL;
 558         }
 559         if (filter->fo_last_objids != NULL) {
 560                 OBD_FREE(filter->fo_last_objids,
 561                          FILTER_GROUPS * sizeof(*filter->fo_last_objids));
 562                 filter->fo_last_objids = NULL;
 563         }
 564         if (filter->fo_dentry_O != NULL) {
 565                 f_dput(filter->fo_dentry_O);
 566                 filter->fo_dentry_O = NULL;
 567         }
 568         RETURN(0);
 569 }
 570
 571 /* FIXME: object groups */
 572 static int filter_prep_groups(struct obd_device *obd)
 573 {
 574         struct filter_obd *filter = &obd->u.filter;
 575         struct dentry *dentry, *O_dentry;
 576         struct file *filp;
 577         int i, rc = 0, cleanup_phase = 0;
 578         ENTRY;
 579
 580         O_dentry = simple_mkdir(current->fs->pwd, "O", 0700, 1);
 581         CDEBUG(D_INODE, "got/created O: %p\n", O_dentry);
 582         if (IS_ERR(O_dentry)) {
 583                 rc = PTR_ERR(O_dentry);
 584                 CERROR("cannot open/create O: rc = %d\n", rc);
 585                 GOTO(cleanup, rc);
 586         }
 587         filter->fo_dentry_O = O_dentry;
 588         cleanup_phase = 1; /* O_dentry */
 589
 590         /* Lookup "R" to tell if we're on an old OST FS and need to convert
 591          * from O/R/<dir>/<objid> to O/0/<dir>/<objid>.  This can be removed
 592          * some time post 1.0 when all old-style OSTs have converted along
 593          * with the init_objid hack. */
 594         dentry = ll_lookup_one_len("R", O_dentry, 1);
 595         if (IS_ERR(dentry))
 596                 GOTO(cleanup, rc = PTR_ERR(dentry));
 597         if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
 598                 struct dentry *O0_dentry = lookup_one_len("0", O_dentry, 1);
 599                 ENTRY;
 600
 601                 CWARN("converting OST to new object layout\n");
 602                 if (IS_ERR(O0_dentry)) {
 603                         rc = PTR_ERR(O0_dentry);
 604                         CERROR("error looking up O/0: rc %d\n", rc);
 605                         GOTO(cleanup_R, rc);
 606                 }
 607
 608                 if (O0_dentry->d_inode) {
 609                         CERROR("Both O/R and O/0 exist. Fix manually.\n");
 610                         GOTO(cleanup_O0, rc = -EEXIST);
 611                 }
 612
 613                 down(&O_dentry->d_inode->i_sem);
 614                 rc = vfs_rename(O_dentry->d_inode, dentry,
 615                                 O_dentry->d_inode, O0_dentry);
 616                 up(&O_dentry->d_inode->i_sem);
 617
 618                 if (rc) {
 619                         CERROR("error renaming O/R to O/0: rc %d\n", rc);
 620                         GOTO(cleanup_O0, rc);
 621                 }
 622                 filter->fo_fsd->fsd_feature_incompat |=
 623                         cpu_to_le32(FILTER_INCOMPAT_GROUPS);
 624                 rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
 625                                                filter->fo_fsd, 1);
 626                 GOTO(cleanup_O0, rc);
 627
 628         cleanup_O0:
 629                 f_dput(O0_dentry);
 630         cleanup_R:
 631                 f_dput(dentry);
 632                 if (rc)
 633                         GOTO(cleanup, rc);
 634         } else {
 635                 f_dput(dentry);
 636         }
 637
 638         OBD_ALLOC(filter->fo_last_objids, FILTER_GROUPS * sizeof(__u64));
 639         if (filter->fo_last_objids == NULL)
 640                 GOTO(cleanup, rc = -ENOMEM);
 641         cleanup_phase = 2; /* groups */
 642
 643         OBD_ALLOC(filter->fo_dentry_O_groups, FILTER_GROUPS * sizeof(dentry));
 644         if (filter->fo_dentry_O_groups == NULL)
 645                 GOTO(cleanup, rc = -ENOMEM);
 646         OBD_ALLOC(filter->fo_last_objid_files, FILTER_GROUPS * sizeof(filp));
 647         if (filter->fo_last_objid_files == NULL)
 648                 GOTO(cleanup, rc = -ENOMEM);
 649
 650         for (i = 0; i < FILTER_GROUPS; i++) {
 651                 char name[25];
 652                 loff_t off = 0;
 653
 654                 sprintf(name, "%d", i);
 655                 dentry = simple_mkdir(O_dentry, name, 0700, 1);
 656                 CDEBUG(D_INODE, "got/created O/%s: %p\n", name, dentry);
 657                 if (IS_ERR(dentry)) {
 658                         rc = PTR_ERR(dentry);
 659                         CERROR("cannot lookup/create O/%s: rc = %d\n",
 660                                name, rc);
 661                         GOTO(cleanup, rc);
 662                 }
 663                 filter->fo_dentry_O_groups[i] = dentry;
 664
 665                 sprintf(name, "O/%d/LAST_ID", i);
 666                 filp = filp_open(name, O_CREAT | O_RDWR, 0700);
 667                 if (IS_ERR(filp)) {
 668                         rc = PTR_ERR(filp);
 669                         CERROR("cannot create %s: rc = %d\n", name, rc);
 670                         GOTO(cleanup, rc);
 671                 }
 672                 filter->fo_last_objid_files[i] = filp;
 673
 674                 if (filp->f_dentry->d_inode->i_size == 0) {
 675                         if (i == 0 && filter->fo_fsd->fsd_unused != 0) {
 676                                 /* OST conversion, remove sometime post 1.0 */
 677                                 filter->fo_last_objids[i] =
 678                                         le64_to_cpu(filter->fo_fsd->fsd_unused);
 679                                 CWARN("saving old objid "LPU64" to LAST_ID\n",
 680                                       filter->fo_last_objids[i]);
 681                                 rc = filter_update_last_objid(obd, 0, 1);
 682                                 if (rc)
 683                                         GOTO(cleanup, rc);
 684                         } else {
 685                                 filter->fo_last_objids[i] = FILTER_INIT_OBJID;
 686                         }
 687                         continue;
 688                 }
 689
 690                 rc = fsfilt_read_record(obd, filp, &filter->fo_last_objids[i],
 691                                         sizeof(__u64), &off);
 692                 if (rc) {
 693                         CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
 694                                name, rc);
 695                         GOTO(cleanup, rc);
 696                 }
 697                 filter->fo_last_objids[i] =
 698                         le64_to_cpu(filter->fo_last_objids[i]);
 699                 CDEBUG(D_INODE, "%s: server last_objid group %d: "LPU64"\n",
 700                        obd->obd_name, i, filter->fo_last_objids[i]);
 701         }
 702
 703         if (filter->fo_subdir_count) {
 704                 O_dentry = filter->fo_dentry_O_groups[0];
 705                 OBD_ALLOC(filter->fo_dentry_O_sub,
 706                           filter->fo_subdir_count * sizeof(dentry));
 707                 if (filter->fo_dentry_O_sub == NULL)
 708                         GOTO(cleanup, rc = -ENOMEM);
 709
 710                 for (i = 0; i < filter->fo_subdir_count; i++) {
 711                         char dir[20];
 712                         snprintf(dir, sizeof(dir), "d%u", i);
 713
 714                         dentry = simple_mkdir(O_dentry, dir, 0700, 1);
 715                         CDEBUG(D_INODE, "got/created O/0/%s: %p\n", dir,dentry);
 716                         if (IS_ERR(dentry)) {
 717                                 rc = PTR_ERR(dentry);
 718                                 CERROR("can't lookup/create O/0/%s: rc = %d\n",
 719                                        dir, rc);
 720                                 GOTO(cleanup, rc);
 721                         }
 722                         filter->fo_dentry_O_sub[i] = dentry;
 723                 }
 724         }
 725         RETURN(0);
 726
 727  cleanup:
 728         filter_cleanup_groups(obd);
 729         return rc;
 730 }
 731
 732 /* setup the object store with correct subdirectories */
 733 static int filter_prep(struct obd_device *obd)
 734 {
 735         struct obd_run_ctxt saved;
 736         struct filter_obd *filter = &obd->u.filter;
 737         struct file *file;
 738         struct inode *inode;
 739         int rc = 0;
 740         ENTRY;
 741
 742         push_ctxt(&saved, &obd->obd_ctxt, NULL);
 743         file = filp_open(LAST_RCVD, O_RDWR | O_CREAT | O_LARGEFILE, 0700);
 744         if (!file || IS_ERR(file)) {
 745                 rc = PTR_ERR(file);
 746                 CERROR("OBD filter: cannot open/create %s: rc = %d\n",
 747                        LAST_RCVD, rc);
 748                 GOTO(out, rc);
 749         }
 750
 751         if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
 752                 CERROR("%s is not a regular file!: mode = %o\n", LAST_RCVD,
 753                        file->f_dentry->d_inode->i_mode);
 754                 GOTO(err_filp, rc = -ENOENT);
 755         }
 756
 757         /* steal operations */
 758         inode = file->f_dentry->d_inode;
 759         filter->fo_fop = file->f_op;
 760         filter->fo_iop = inode->i_op;
 761         filter->fo_aops = inode->i_mapping->a_ops;
 762
 763         rc = filter_init_server_data(obd, file);
 764         if (rc) {
 765                 CERROR("cannot read %s: rc = %d\n", LAST_RCVD, rc);
 766                 GOTO(err_filp, rc);
 767         }
 768         filter->fo_rcvd_filp = file;
 769
 770         rc = filter_prep_groups(obd);
 771         if (rc)
 772                 GOTO(err_server_data, rc);
 773
 774  out:
 775         pop_ctxt(&saved, &obd->obd_ctxt, NULL);
 776
 777         return(rc);
 778
 779  err_server_data:
 780         //class_disconnect_exports(obd, 0);
 781         filter_free_server_data(filter);
 782  err_filp:
 783         if (filp_close(file, 0))
 784                 CERROR("can't close %s after error\n", LAST_RCVD);
 785         filter->fo_rcvd_filp = NULL;
 786         goto out;
 787 }
 788
 789 /* cleanup the filter: write last used object id to status file */
 790 static void filter_post(struct obd_device *obd)
 791 {
 792         struct obd_run_ctxt saved;
 793         struct filter_obd *filter = &obd->u.filter;
 794         int rc, i;
 795
 796         /* XXX: filter_update_lastobjid used to call fsync_dev.  It might be
 797          * best to start a transaction with h_sync, because we removed this
 798          * from lastobjid */
 799
 800         push_ctxt(&saved, &obd->obd_ctxt, NULL);
 801         rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
 802                                        filter->fo_fsd, 0);
 803         if (rc)
 804                 CERROR("error writing server data: rc = %d\n", rc);
 805
 806         for (i = 0; i < FILTER_GROUPS; i++) {
 807                 rc = filter_update_last_objid(obd, i, (i == FILTER_GROUPS - 1));
 808                 if (rc)
 809                         CERROR("error writing group %d lastobjid: rc = %d\n",
 810                                i, rc);
 811         }
 812
 813         rc = filp_close(filter->fo_rcvd_filp, 0);
 814         filter->fo_rcvd_filp = NULL;
 815         if (rc)
 816                 CERROR("error closing %s: rc = %d\n", LAST_RCVD, rc);
 817
 818         filter_cleanup_groups(obd);
 819         filter_free_server_data(filter);
 820         pop_ctxt(&saved, &obd->obd_ctxt, NULL);
 821 }
 822
 823 static void filter_set_last_id(struct filter_obd *filter, struct obdo *oa,
 824                                obd_id id)
 825 {
 826         obd_gr group = 0;
 827         LASSERT(filter->fo_fsd != NULL);
 828
 829         if (oa != NULL) {
 830                 LASSERT(oa->o_gr <= FILTER_GROUPS);
 831                 group = oa->o_gr;
 832         }
 833
 834         spin_lock(&filter->fo_objidlock);
 835         filter->fo_last_objids[group] = id;
 836         spin_unlock(&filter->fo_objidlock);
 837 }
 838
 839 __u64 filter_last_id(struct filter_obd *filter, struct obdo *oa)
 840 {
 841         obd_id id;
 842         obd_gr group = 0;
 843         LASSERT(filter->fo_fsd != NULL);
 844
 845         if (oa != NULL) {
 846                 LASSERT(oa->o_gr <= FILTER_GROUPS);
 847                 group = oa->o_gr;
 848         }
 849
 850         /* FIXME: object groups */
 851         spin_lock(&filter->fo_objidlock);
 852         id = filter->fo_last_objids[group];
 853         spin_unlock(&filter->fo_objidlock);
 854
 855         return id;
 856 }
 857
 858 /* direct cut-n-paste of mds_blocking_ast() */
 859 static int filter_blocking_ast(struct ldlm_lock *lock,
 860                                struct ldlm_lock_desc *desc,
 861                                void *data, int flag)
 862 {
 863         int do_ast;
 864         ENTRY;
 865
 866         if (flag == LDLM_CB_CANCELING) {
 867                 /* Don't need to do anything here. */
 868                 RETURN(0);
 869         }
 870
 871         /* XXX layering violation!  -phil */
 872         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 873         /* Get this: if filter_blocking_ast is racing with ldlm_intent_policy,
 874          * such that filter_blocking_ast is called just before l_i_p takes the
 875          * ns_lock, then by the time we get the lock, we might not be the
 876          * correct blocking function anymore.  So check, and return early, if
 877          * so. */
 878         if (lock->l_blocking_ast != filter_blocking_ast) {
 879                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 880                 RETURN(0);
 881         }
 882
 883         lock->l_flags |= LDLM_FL_CBPENDING;
 884         do_ast = (!lock->l_readers && !lock->l_writers);
 885         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 886
 887         if (do_ast) {
 888                 struct lustre_handle lockh;
 889                 int rc;
 890
 891                 LDLM_DEBUG(lock, "already unused, calling ldlm_cli_cancel");
 892                 ldlm_lock2handle(lock, &lockh);
 893                 rc = ldlm_cli_cancel(&lockh);
 894                 if (rc < 0)
 895                         CERROR("ldlm_cli_cancel: %d\n", rc);
 896         } else {
 897                 LDLM_DEBUG(lock, "Lock still has references, will be "
 898                            "cancelled later");
 899         }
 900         RETURN(0);
 901 }
 902
 903 static int filter_lock_dentry(struct obd_device *obd, struct dentry *dparent)
 904 {
 905         down(&dparent->d_inode->i_sem);
 906         return 0;
 907 }
 908
 909 /* We never dget the object parent, so DON'T dput it either */
 910 static void filter_parent_unlock(struct dentry *dparent)
 911 {
 912         up(&dparent->d_inode->i_sem);
 913 }
 914
 915 /* We never dget the object parent, so DON'T dput it either */
 916 struct dentry *filter_parent(struct obd_device *obd, obd_gr group, obd_id objid)
 917 {
 918         struct filter_obd *filter = &obd->u.filter;
 919         LASSERT(group < FILTER_GROUPS); /* FIXME: object groups */
 920
 921         if (group > 0 || filter->fo_subdir_count == 0)
 922                 return filter->fo_dentry_O_groups[group];
 923
 924         return filter->fo_dentry_O_sub[objid & (filter->fo_subdir_count - 1)];
 925 }
 926
 927 /* We never dget the object parent, so DON'T dput it either */
 928 struct dentry *filter_parent_lock(struct obd_device *obd, obd_gr group,
 929                                   obd_id objid)
 930 {
 931         unsigned long now = jiffies;
 932         struct dentry *dparent = filter_parent(obd, group, objid);
 933         int rc;
 934
 935         if (IS_ERR(dparent))
 936                 return dparent;
 937
 938         rc = filter_lock_dentry(obd, dparent);
 939         if (time_after(jiffies, now + 15 * HZ))
 940                 CERROR("slow parent lock %lus\n", (jiffies - now) / HZ);
 941         return rc ? ERR_PTR(rc) : dparent;
 942 }
 943
 944 /* How to get files, dentries, inodes from object id's.
 945  *
 946  * If dir_dentry is passed, the caller has already locked the parent
 947  * appropriately for this operation (normally a write lock).  If
 948  * dir_dentry is NULL, we do a read lock while we do the lookup to
 949  * avoid races with create/destroy and such changing the directory
 950  * internal to the filesystem code. */
 951 struct dentry *filter_fid2dentry(struct obd_device *obd,
 952                                  struct dentry *dir_dentry,
 953                                  obd_gr group, obd_id id)
 954 {
 955         struct dentry *dparent = dir_dentry;
 956         struct dentry *dchild;
 957         char name[32];
 958         int len;
 959         ENTRY;
 960
 961         if (id == 0) {
 962                 CERROR("fatal: invalid object id 0\n");
 963                 RETURN(ERR_PTR(-ESTALE));
 964         }
 965
 966         len = sprintf(name, LPU64, id);
 967         if (dir_dentry == NULL) {
 968                 dparent = filter_parent_lock(obd, group, id);
 969                 if (IS_ERR(dparent))
 970                         RETURN(dparent);
 971         }
 972         CDEBUG(D_INODE, "looking up object O/%*s/%s\n",
 973                dparent->d_name.len, dparent->d_name.name, name);
 974         dchild = /*ll_*/lookup_one_len(name, dparent, len);
 975         if (dir_dentry == NULL)
 976                 filter_parent_unlock(dparent);
 977         if (IS_ERR(dchild)) {
 978                 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
 979                 RETURN(dchild);
 980         }
 981
 982         CDEBUG(D_INODE, "got child objid %s: %p, count = %d\n",
 983                name, dchild, atomic_read(&dchild->d_count));
 984
 985         LASSERT(atomic_read(&dchild->d_count) > 0);
 986
 987         RETURN(dchild);
 988 }
 989
 990 static int filter_prepare_destroy(struct obd_device *obd, obd_id objid)
 991 {
 992         struct lustre_handle lockh;
 993         int flags = LDLM_AST_DISCARD_DATA, rc;
 994         struct ldlm_res_id res_id = { .name = { objid } };
 995         ldlm_policy_data_t policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
 996
 997         ENTRY;
 998         /* Tell the clients that the object is gone now and that they should
 999          * throw away any cached pages. */
1000         rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, res_id,
1001                               LDLM_EXTENT, &policy, LCK_PW,
1002                               &flags, filter_blocking_ast, ldlm_completion_ast,
1003                               NULL, NULL, NULL, 0, NULL, &lockh);
1004
1005         /* We only care about the side-effects, just drop the lock. */
1006         if (rc == ELDLM_OK)
1007                 ldlm_lock_decref(&lockh, LCK_PW);
1008
1009         RETURN(rc);
1010 }
1011
1012 /* Caller must hold LCK_PW on parent and push us into kernel context.
1013  * Caller is also required to ensure that dchild->d_inode exists. */
1014 static int filter_destroy_internal(struct obd_device *obd, obd_id objid,
1015                                    struct dentry *dparent,
1016                                    struct dentry *dchild)
1017 {
1018         struct inode *inode = dchild->d_inode;
1019         int rc;
1020         ENTRY;
1021
1022         if (inode->i_nlink != 1 || atomic_read(&inode->i_count) != 1) {
1023                 CERROR("destroying objid %*s nlink = %lu, count = %d\n",
1024                        dchild->d_name.len, dchild->d_name.name,
1025                        (unsigned long)inode->i_nlink,
1026                        atomic_read(&inode->i_count));
1027         }
1028
1029         rc = vfs_unlink(dparent->d_inode, dchild);
1030
1031         if (rc)
1032                 CERROR("error unlinking objid %*s: rc %d\n",
1033                        dchild->d_name.len, dchild->d_name.name, rc);
1034
1035         RETURN(rc);
1036 }
1037
1038 static int filter_intent_policy(struct ldlm_namespace *ns,
1039                                 struct ldlm_lock **lockp, void *req_cookie,
1040                                 ldlm_mode_t mode, int flags, void *data)
1041 {
1042         struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
1043         struct ptlrpc_request *req = req_cookie;
1044         struct ldlm_lock *lock = *lockp, *l = NULL;
1045         struct ldlm_resource *res = lock->l_resource;
1046         ldlm_processing_policy policy;
1047         struct ost_lvb *res_lvb, *reply_lvb;
1048         struct list_head *tmp;
1049         ldlm_error_t err;
1050         int tmpflags = 0, rc, repsize[2] = {sizeof(struct ldlm_reply),
1051                                             sizeof(struct ost_lvb) };
1052         ENTRY;
1053
1054         policy = ldlm_get_processing_policy(res);
1055         LASSERT(policy != NULL);
1056         LASSERT(req != NULL);
1057
1058         rc = lustre_pack_reply(req, 2, repsize, NULL);
1059         if (rc)
1060                 RETURN(req->rq_status = rc);
1061
1062         reply_lvb = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*reply_lvb));
1063         LASSERT(reply_lvb != NULL);
1064
1065         //fixup_handle_for_resent_req(req, lock, &lockh);
1066
1067         /* If we grant any lock at all, it will be a whole-file read lock.
1068          * Call the extent policy function to see if our request can be
1069          * granted, or is blocked. */
1070         lock->l_policy_data.l_extent.start = 0;
1071         lock->l_policy_data.l_extent.end = OBD_OBJECT_EOF;
1072         lock->l_req_mode = LCK_PR;
1073
1074         l_lock(&res->lr_namespace->ns_lock);
1075
1076         res->lr_tmp = &rpc_list;
1077         rc = policy(lock, &tmpflags, 0, &err);
1078         res->lr_tmp = NULL;
1079
1080         /* FIXME: we should change the policy function slightly, to not make
1081          * this list at all, since we just turn around and free it */
1082         while (!list_empty(&rpc_list)) {
1083                 struct ldlm_ast_work *w =
1084                         list_entry(rpc_list.next, struct ldlm_ast_work, w_list);
1085                 list_del(&w->w_list);
1086                 LDLM_LOCK_PUT(w->w_lock);
1087                 OBD_FREE(w, sizeof(*w));
1088         }
1089
1090         if (rc == LDLM_ITER_CONTINUE) {
1091                 /* The lock met with no resistance; we're finished. */
1092                 l_unlock(&res->lr_namespace->ns_lock);
1093                 RETURN(ELDLM_LOCK_REPLACED);
1094         }
1095
1096         /* Do not grant any lock, but instead send GL callbacks.  The extent
1097          * policy nicely created a list of all PW locks for us.  We will choose
1098          * the highest of those which are larger than the size in the LVB, if
1099          * any, and perform a glimpse callback. */
1100         down(&res->lr_lvb_sem);
1101         res_lvb = res->lr_lvb_data;
1102         LASSERT(res_lvb != NULL);
1103         reply_lvb->lvb_size = res_lvb->lvb_size;
1104         reply_lvb->lvb_blocks = res_lvb->lvb_blocks;
1105         up(&res->lr_lvb_sem);
1106
1107         list_for_each(tmp, &res->lr_granted) {
1108                 struct ldlm_lock *tmplock =
1109                         list_entry(tmp, struct ldlm_lock, l_res_link);
1110
1111                 if (tmplock->l_granted_mode == LCK_PR)
1112                         continue;
1113
1114                 if (tmplock->l_policy_data.l_extent.end <= reply_lvb->lvb_size)
1115                         continue;
1116
1117                 if (l == NULL) {
1118                         l = LDLM_LOCK_GET(tmplock);
1119                         continue;
1120                 }
1121
1122                 if (l->l_policy_data.l_extent.start >
1123                     tmplock->l_policy_data.l_extent.start)
1124                         continue;
1125
1126                 LDLM_LOCK_PUT(l);
1127                 l = LDLM_LOCK_GET(tmplock);
1128         }
1129         l_unlock(&res->lr_namespace->ns_lock);
1130
1131         /* There were no PW locks beyond the size in the LVB; finished. */
1132         if (l == NULL)
1133                 RETURN(ELDLM_LOCK_ABORTED);
1134
1135         LASSERT(l->l_glimpse_ast != NULL);
1136         rc = l->l_glimpse_ast(l, NULL); /* this will update the LVB */
1137         if (rc != 0 && res->lr_namespace->ns_lvbo &&
1138             res->lr_namespace->ns_lvbo->lvbo_update) {
1139                 res->lr_namespace->ns_lvbo->lvbo_update(res, NULL, 0, 1);
1140         }
1141
1142         down(&res->lr_lvb_sem);
1143         reply_lvb->lvb_size = res_lvb->lvb_size;
1144         reply_lvb->lvb_blocks = res_lvb->lvb_blocks;
1145         up(&res->lr_lvb_sem);
1146
1147         LDLM_LOCK_PUT(l);
1148
1149         RETURN(ELDLM_LOCK_ABORTED);
1150 }
1151
1152 /* mount the file system (secretly) */
1153 int filter_common_setup(struct obd_device *obd, obd_count len, void *buf,
1154                         char *option)
1155 {
1156         struct lustre_cfg* lcfg = buf;
1157         struct filter_obd *filter = &obd->u.filter;
1158         struct vfsmount *mnt;
1159         int rc = 0;
1160         ENTRY;
1161
1162         dev_clear_rdonly(2);
1163
1164         if (!lcfg->lcfg_inlbuf1 || !lcfg->lcfg_inlbuf2)
1165                 RETURN(-EINVAL);
1166
1167         obd->obd_fsops = fsfilt_get_ops(lcfg->lcfg_inlbuf2);
1168         if (IS_ERR(obd->obd_fsops))
1169                 RETURN(PTR_ERR(obd->obd_fsops));
1170
1171         mnt = do_kern_mount(lcfg->lcfg_inlbuf2, MS_NOATIME | MS_NODIRATIME,
1172                             lcfg->lcfg_inlbuf1, option);
1173         rc = PTR_ERR(mnt);
1174         if (IS_ERR(mnt))
1175                 GOTO(err_ops, rc);
1176
1177         if (lcfg->lcfg_inllen3 > 0 && lcfg->lcfg_inlbuf3) {
1178                 if (*lcfg->lcfg_inlbuf3 == 'f') {
1179                         obd->obd_replayable = 1;
1180                         obd_sync_filter = 1;
1181                         CWARN("%s: recovery enabled\n", obd->obd_name);
1182                 } else {
1183                         if (*lcfg->lcfg_inlbuf3 != 'n') {
1184                                 CERROR("unrecognised flag '%c'\n",
1185                                        *lcfg->lcfg_inlbuf3);
1186                         }
1187                         // XXX Robert? Why do we get errors here
1188                         // GOTO(err_mntput, rc = -EINVAL);
1189                 }
1190         }
1191
1192         filter->fo_vfsmnt = mnt;
1193         filter->fo_sb = mnt->mnt_sb;
1194         filter->fo_fstype = mnt->mnt_sb->s_type->name;
1195         CDEBUG(D_SUPER, "%s: mnt = %p\n", filter->fo_fstype, mnt);
1196
1197         OBD_SET_CTXT_MAGIC(&obd->obd_ctxt);
1198         obd->obd_ctxt.pwdmnt = mnt;
1199         obd->obd_ctxt.pwd = mnt->mnt_root;
1200         obd->obd_ctxt.fs = get_ds();
1201         obd->obd_ctxt.cb_ops = filter_lvfs_ops;
1202
1203         rc = filter_prep(obd);
1204         if (rc)
1205                 GOTO(err_mntput, rc);
1206
1207         spin_lock_init(&filter->fo_translock);
1208         spin_lock_init(&filter->fo_objidlock);
1209         INIT_LIST_HEAD(&filter->fo_export_list);
1210         sema_init(&filter->fo_alloc_lock, 1);
1211         spin_lock_init(&filter->fo_r_pages.oh_lock);
1212         spin_lock_init(&filter->fo_w_pages.oh_lock);
1213         spin_lock_init(&filter->fo_r_discont_pages.oh_lock);
1214         spin_lock_init(&filter->fo_w_discont_pages.oh_lock);
1215         spin_lock_init(&filter->fo_r_discont_blocks.oh_lock);
1216         spin_lock_init(&filter->fo_w_discont_blocks.oh_lock);
1217         filter->fo_readcache_max_filesize = FILTER_MAX_CACHE_SIZE;
1218
1219         obd->obd_namespace = ldlm_namespace_new("filter-tgt",
1220                                                 LDLM_NAMESPACE_SERVER);
1221         if (obd->obd_namespace == NULL)
1222                 GOTO(err_post, rc = -ENOMEM);
1223         obd->obd_namespace->ns_lvbp = obd;
1224         obd->obd_namespace->ns_lvbo = &filter_lvbo;
1225         ldlm_register_intent(obd->obd_namespace, filter_intent_policy);
1226
1227         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1228                            "filter_ldlm_cb_client", &obd->obd_ldlm_client);
1229
1230         rc = llog_cat_initialize(obd, 1);
1231         if (rc) {
1232                 CERROR("failed to setup llogging subsystems\n");
1233                 GOTO(err_post, rc);
1234         }
1235
1236         RETURN(0);
1237
1238 err_post:
1239         filter_post(obd);
1240 err_mntput:
1241         unlock_kernel();
1242         mntput(mnt);
1243         filter->fo_sb = 0;
1244         lock_kernel();
1245 err_ops:
1246         fsfilt_put_ops(obd->obd_fsops);
1247         return rc;
1248 }
1249
1250 static int filter_setup(struct obd_device *obd, obd_count len, void *buf)
1251 {
1252         struct lprocfs_static_vars lvars;
1253         struct lustre_cfg* lcfg = buf;
1254         const char *str = NULL;
1255         char *option = NULL;
1256         int n = 0;
1257         int rc;
1258
1259         if (!lcfg->lcfg_inlbuf1 || !lcfg->lcfg_inlbuf2)
1260                 RETURN(-EINVAL);
1261
1262         if (!strcmp(lcfg->lcfg_inlbuf2, "ext3")) {
1263 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1264         /* bug 1577: implement async-delete for 2.5 */
1265                 str = "errors=remount-ro,asyncdel";
1266 #else
1267                 str = "errors=remount-ro";
1268 #endif
1269                 n = strlen(str) + 1;
1270                 OBD_ALLOC(option, n);
1271                 if (option == NULL)
1272                         RETURN(-ENOMEM);
1273                 strcpy(option, str);
1274         }
1275
1276         rc = filter_common_setup(obd, len, buf, option);
1277         if (option)
1278                 OBD_FREE(option, n);
1279
1280         lprocfs_init_vars(filter, &lvars);
1281         if (rc == 0 && lprocfs_obd_setup(obd, lvars.obd_vars) == 0 &&
1282             lprocfs_alloc_obd_stats(obd, LPROC_FILTER_LAST) == 0) {
1283                 /* Init obdfilter private stats here */
1284                 lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_READ_BYTES,
1285                                      LPROCFS_CNTR_AVGMINMAX,
1286                                      "read_bytes", "bytes");
1287                 lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_WRITE_BYTES,
1288                                      LPROCFS_CNTR_AVGMINMAX,
1289                                      "write_bytes", "bytes");
1290
1291                 lproc_filter_attach_seqstat(obd);
1292         }
1293
1294         return rc;
1295 }
1296
1297 static int filter_cleanup(struct obd_device *obd, int flags)
1298 {
1299         struct filter_obd *filter = &obd->u.filter;
1300         ENTRY;
1301
1302         if (flags & OBD_OPT_FAILOVER)
1303                 CERROR("%s: shutting down for failover; client state will"
1304                        " be preserved.\n", obd->obd_name);
1305
1306         if (!list_empty(&obd->obd_exports)) {
1307                 CERROR("%s: still has clients!\n", obd->obd_name);
1308                 class_disconnect_exports(obd, flags);
1309                 if (!list_empty(&obd->obd_exports)) {
1310                         CERROR("still has exports after forced cleanup?\n");
1311                         RETURN(-EBUSY);
1312                 }
1313         }
1314
1315         ldlm_namespace_free(obd->obd_namespace, flags & OBD_OPT_FORCE);
1316
1317         if (filter->fo_sb == NULL)
1318                 RETURN(0);
1319
1320         lprocfs_free_obd_stats(obd);
1321         lprocfs_obd_cleanup(obd);
1322
1323         filter_post(obd);
1324
1325         shrink_dcache_parent(filter->fo_sb->s_root);
1326         filter->fo_sb = 0;
1327
1328         if (atomic_read(&filter->fo_vfsmnt->mnt_count) > 1)
1329                 CERROR("%s: mount point %p busy, mnt_count: %d\n",
1330                        obd->obd_name, filter->fo_vfsmnt,
1331                        atomic_read(&filter->fo_vfsmnt->mnt_count));
1332
1333         unlock_kernel();
1334         mntput(filter->fo_vfsmnt);
1335         //destroy_buffers(filter->fo_sb->s_dev);
1336         filter->fo_sb = NULL;
1337         fsfilt_put_ops(obd->obd_fsops);
1338         lock_kernel();
1339
1340         dev_clear_rdonly(2);
1341
1342         RETURN(0);
1343 }
1344
1345 /* nearly identical to mds_connect */
1346 static int filter_connect(struct lustre_handle *conn, struct obd_device *obd,
1347                           struct obd_uuid *cluuid)
1348 {
1349         struct obd_export *exp;
1350         struct filter_export_data *fed;
1351         struct filter_client_data *fcd = NULL;
1352         struct filter_obd *filter = &obd->u.filter;
1353         int rc;
1354         ENTRY;
1355
1356         if (conn == NULL || obd == NULL || cluuid == NULL)
1357                 RETURN(-EINVAL);
1358
1359         rc = class_connect(conn, obd, cluuid);
1360         if (rc)
1361                 RETURN(rc);
1362         exp = class_conn2export(conn);
1363         LASSERT(exp != NULL);
1364
1365         fed = &exp->exp_filter_data;
1366
1367         spin_lock_init(&fed->fed_lock);
1368
1369         if (!obd->obd_replayable)
1370                 GOTO(cleanup, rc = 0);
1371
1372         OBD_ALLOC(fcd, sizeof(*fcd));
1373         if (!fcd) {
1374                 CERROR("filter: out of memory for client data\n");
1375                 GOTO(cleanup, rc = -ENOMEM);
1376         }
1377
1378         memcpy(fcd->fcd_uuid, cluuid, sizeof(fcd->fcd_uuid));
1379         fed->fed_fcd = fcd;
1380
1381         rc = filter_client_add(obd, filter, fed, -1);
1382
1383 cleanup:
1384         if (rc) {
1385                 if (fcd)
1386                         OBD_FREE(fcd, sizeof(*fcd));
1387                 class_disconnect(exp, 0);
1388         } else {
1389                 class_export_put(exp);
1390         }
1391         return rc;
1392 }
1393
1394 static int filter_precleanup(struct obd_device *obd, int flags)
1395 {
1396         int rc = 0;
1397         ENTRY;
1398
1399         rc = obd_llog_finish(obd, 0);
1400         if (rc)
1401                 CERROR("failed to cleanup llogging subsystem\n");
1402
1403         RETURN(rc);
1404 }
1405
1406 /* Do extra sanity checks for grant accounting.  We do this at connect,
1407  * disconnect, and statfs RPC time, so it shouldn't be too bad.  We can
1408  * always get rid of it or turn it off when we know accounting is good. */
1409 static void filter_grant_sanity_check(struct obd_device *obd, const char *func)
1410 {
1411         struct filter_export_data *fed;
1412         struct obd_export *exp;
1413         obd_size maxsize = obd->obd_osfs.os_blocks * obd->obd_osfs.os_bsize;
1414         obd_size tot_dirty = 0, tot_pending = 0, tot_granted = 0;
1415         obd_size fo_tot_dirty, fo_tot_pending, fo_tot_granted;
1416
1417         if (list_empty(&obd->obd_exports))
1418                 return;
1419
1420         spin_lock(&obd->obd_osfs_lock);
1421         spin_lock(&obd->obd_dev_lock);
1422         list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
1423                 fed = &exp->exp_filter_data;
1424                 LASSERTF(fed->fed_grant + fed->fed_pending <= maxsize,
1425                          "cli %s/%p %lu+%lu > "LPU64"\n",
1426                          exp->exp_client_uuid.uuid, exp,
1427                          fed->fed_grant, fed->fed_pending, maxsize);
1428                 LASSERTF(fed->fed_dirty <= maxsize, "cli %s/%p %lu > "LPU64"\n",
1429                          exp->exp_client_uuid.uuid, exp,fed->fed_dirty,maxsize);
1430                 CDEBUG(D_CACHE,"%s: cli %s/%p dirty %lu pend %lu grant %lu\n",
1431                        obd->obd_name, exp->exp_client_uuid.uuid, exp,
1432                        fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1433                 tot_granted += fed->fed_grant + fed->fed_pending;
1434                 tot_pending += fed->fed_pending;
1435                 tot_dirty += fed->fed_dirty;
1436         }
1437         fo_tot_granted = obd->u.filter.fo_tot_granted;
1438         fo_tot_pending = obd->u.filter.fo_tot_pending;
1439         fo_tot_dirty = obd->u.filter.fo_tot_dirty;
1440         spin_unlock(&obd->obd_dev_lock);
1441         spin_unlock(&obd->obd_osfs_lock);
1442
1443         /* Do these assertions outside the spinlocks so we don't kill system */
1444         if (tot_granted != fo_tot_granted)
1445                 CERROR("%s: tot_granted "LPU64" != fo_tot_granted "LPU64"\n",
1446                        func, tot_granted, fo_tot_granted);
1447         if (tot_pending != fo_tot_pending)
1448                 CERROR("%s: tot_pending "LPU64" != fo_tot_pending "LPU64"\n",
1449                        func, tot_pending, fo_tot_pending);
1450         if (tot_dirty != fo_tot_dirty)
1451                 CERROR("%s: tot_dirty "LPU64" != fo_tot_dirty "LPU64"\n",
1452                        func, tot_dirty, fo_tot_dirty);
1453         if (tot_pending > tot_granted)
1454                 CERROR("%s: tot_pending "LPU64" > tot_granted "LPU64"\n",
1455                        func, tot_pending, tot_granted);
1456         if (tot_granted > maxsize)
1457                 CERROR("%s: tot_granted "LPU64" > maxsize "LPU64"\n",
1458                        func, tot_granted, maxsize);
1459         if (tot_dirty > maxsize)
1460                 CERROR("%s: tot_dirty "LPU64" > maxsize "LPU64"\n",
1461                        func, tot_dirty, maxsize);
1462 }
1463
1464 /* Remove this client from the grant accounting totals.  We also remove
1465  * the export from the obd device under the osfs and dev locks to ensure
1466  * that the filter_grant_sanity_check() calculations are always valid.
1467  * The client should do something similar when it invalidates its import. */
1468 static void filter_grant_discard(struct obd_export *exp)
1469 {
1470         struct obd_device *obd = exp->exp_obd;
1471         struct filter_obd *filter = &obd->u.filter;
1472         struct filter_export_data *fed = &exp->exp_filter_data;
1473
1474         spin_lock(&obd->obd_osfs_lock);
1475         spin_lock(&exp->exp_obd->obd_dev_lock);
1476         list_del_init(&exp->exp_obd_chain);
1477         spin_unlock(&exp->exp_obd->obd_dev_lock);
1478
1479         CDEBUG(D_CACHE, "%s: cli %s/%p dirty %lu pend %lu grant %lu\n",
1480                obd->obd_name, exp->exp_client_uuid.uuid, exp,
1481                fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1482
1483         LASSERTF(filter->fo_tot_granted >= fed->fed_grant,
1484                  "%s: tot_granted "LPU64" cli %s/%p fed_grant %lu\n",
1485                  obd->obd_name, filter->fo_tot_granted,
1486                  exp->exp_client_uuid.uuid, exp, fed->fed_grant);
1487         filter->fo_tot_granted -= fed->fed_grant;
1488         LASSERTF(exp->exp_obd->u.filter.fo_tot_pending >= fed->fed_pending,
1489                  "%s: tot_pending "LPU64" cli %s/%p fed_pending %lu\n",
1490                  obd->obd_name, filter->fo_tot_pending,
1491                  exp->exp_client_uuid.uuid, exp, fed->fed_pending);
1492         LASSERTF(filter->fo_tot_dirty >= fed->fed_dirty,
1493                  "%s: tot_dirty "LPU64" cli %s/%p fed_dirty %lu\n",
1494                  obd->obd_name, filter->fo_tot_dirty,
1495                  exp->exp_client_uuid.uuid, exp, fed->fed_dirty);
1496         filter->fo_tot_dirty -= fed->fed_dirty;
1497         fed->fed_dirty = 0;
1498         fed->fed_grant = 0;
1499
1500         spin_unlock(&obd->obd_osfs_lock);
1501 }
1502
1503 static int filter_destroy_export(struct obd_export *exp)
1504 {
1505         ENTRY;
1506
1507         if (exp->exp_filter_data.fed_pending)
1508                 CERROR("%s: cli %s/%p has %lu pending on destroyed export\n",
1509                        exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
1510                        exp, exp->exp_filter_data.fed_pending);
1511
1512         target_destroy_export(exp);
1513
1514         if (exp->exp_obd->obd_replayable)
1515                 filter_client_free(exp, exp->exp_flags);
1516
1517         filter_grant_discard(exp);
1518         if (!(exp->exp_flags & OBD_OPT_FORCE))
1519                 filter_grant_sanity_check(exp->exp_obd, __FUNCTION__);
1520
1521         RETURN(0);
1522 }
1523
1524 /* also incredibly similar to mds_disconnect */
1525 static int filter_disconnect(struct obd_export *exp, int flags)
1526 {
1527         struct obd_device *obd = exp->exp_obd;
1528         unsigned long irqflags;
1529         struct llog_ctxt *ctxt;
1530         int rc;
1531         ENTRY;
1532
1533         LASSERT(exp);
1534         class_export_get(exp);
1535
1536         spin_lock_irqsave(&exp->exp_lock, irqflags);
1537         exp->exp_flags = flags;
1538         spin_unlock_irqrestore(&exp->exp_lock, irqflags);
1539
1540         if (!(flags & OBD_OPT_FORCE))
1541                 filter_grant_sanity_check(obd, __FUNCTION__);
1542         filter_grant_discard(exp);
1543
1544         /* Disconnect early so that clients can't keep using export */
1545         rc = class_disconnect(exp, flags);
1546
1547         ldlm_cancel_locks_for_export(exp);
1548
1549         fsfilt_sync(obd, obd->u.filter.fo_sb);
1550
1551         /* flush any remaining cancel messages out to the target */
1552         ctxt = llog_get_context(obd, LLOG_UNLINK_REPL_CTXT);
1553         llog_sync(ctxt, exp);
1554
1555         class_export_put(exp);
1556         RETURN(rc);
1557 }
1558
1559 struct dentry *__filter_oa2dentry(struct obd_device *obd,
1560                                   struct obdo *oa, const char *what)
1561 {
1562         struct dentry *dchild = NULL;
1563         obd_gr group = 0;
1564
1565         if (oa->o_valid & OBD_MD_FLGROUP)
1566                 group = oa->o_gr;
1567
1568         dchild = filter_fid2dentry(obd, NULL, group, oa->o_id);
1569
1570         if (IS_ERR(dchild)) {
1571                 CERROR("%s error looking up object: "LPU64"\n", what, oa->o_id);
1572                 RETURN(dchild);
1573         }
1574
1575         if (dchild->d_inode == NULL) {
1576                 CERROR("%s on non-existent object: "LPU64"\n", what, oa->o_id);
1577                 f_dput(dchild);
1578                 RETURN(ERR_PTR(-ENOENT));
1579         }
1580
1581         return dchild;
1582 }
1583
1584 static int filter_getattr(struct obd_export *exp, struct obdo *oa,
1585                           struct lov_stripe_md *md)
1586 {
1587         struct dentry *dentry = NULL;
1588         struct obd_device *obd;
1589         int rc = 0;
1590         ENTRY;
1591
1592         obd = class_exp2obd(exp);
1593         if (obd == NULL) {
1594                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1595                        exp->exp_handle.h_cookie);
1596                 RETURN(-EINVAL);
1597         }
1598
1599         dentry = filter_oa2dentry(obd, oa);
1600         if (IS_ERR(dentry))
1601                 RETURN(PTR_ERR(dentry));
1602
1603         /* Limit the valid bits in the return data to what we actually use */
1604         oa->o_valid = OBD_MD_FLID;
1605         obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
1606
1607         f_dput(dentry);
1608         RETURN(rc);
1609 }
1610
1611 /* this is called from filter_truncate() until we have filter_punch() */
1612 static int filter_setattr(struct obd_export *exp, struct obdo *oa,
1613                           struct lov_stripe_md *md, struct obd_trans_info *oti)
1614 {
1615         struct obd_run_ctxt saved;
1616         struct filter_obd *filter;
1617         struct dentry *dentry;
1618         struct iattr iattr;
1619         struct ldlm_res_id res_id = { .name = { oa->o_id } };
1620         struct ldlm_resource *res;
1621         void *handle;
1622         int rc, rc2;
1623         ENTRY;
1624
1625         LASSERT(oti != NULL);
1626
1627         dentry = filter_oa2dentry(exp->exp_obd, oa);
1628         if (IS_ERR(dentry))
1629                 RETURN(PTR_ERR(dentry));
1630
1631         filter = &exp->exp_obd->u.filter;
1632
1633         iattr_from_obdo(&iattr, oa, oa->o_valid);
1634
1635         push_ctxt(&saved, &exp->exp_obd->obd_ctxt, NULL);
1636         lock_kernel();
1637
1638         if (iattr.ia_valid & ATTR_SIZE)
1639                 down(&dentry->d_inode->i_sem);
1640         handle = fsfilt_start(exp->exp_obd, dentry->d_inode, FSFILT_OP_SETATTR,
1641                               oti);
1642         if (IS_ERR(handle))
1643                 GOTO(out_unlock, rc = PTR_ERR(handle));
1644
1645         /* XXX this could be a rwsem instead, if filter_preprw played along */
1646         if (iattr.ia_valid & ATTR_ATTR_FLAG)
1647                 rc = fsfilt_iocontrol(exp->exp_obd, dentry->d_inode, NULL,
1648                                       EXT3_IOC_SETFLAGS,
1649                                       (long)&iattr.ia_attr_flags);
1650         else
1651                 rc = fsfilt_setattr(exp->exp_obd, dentry, handle, &iattr, 1);
1652         rc = filter_finish_transno(exp, oti, rc);
1653         rc2 = fsfilt_commit(exp->exp_obd, dentry->d_inode, handle, 0);
1654         if (rc2) {
1655                 CERROR("error on commit, err = %d\n", rc2);
1656                 if (!rc)
1657                         rc = rc2;
1658         }
1659
1660         if (iattr.ia_valid & ATTR_SIZE) {
1661                 res = ldlm_resource_get(exp->exp_obd->obd_namespace, NULL,
1662                                         res_id, LDLM_EXTENT, 0);
1663                 if (res == NULL) {
1664                         CERROR("!!! resource_get failed for object "LPU64" -- "
1665                                "filter_setattr with no lock?\n", oa->o_id);
1666                 } else {
1667                         if (res->lr_namespace->ns_lvbo &&
1668                             res->lr_namespace->ns_lvbo->lvbo_update) {
1669                                 rc = res->lr_namespace->ns_lvbo->lvbo_update
1670                                         (res, NULL, 0, 0);
1671                         }
1672                         ldlm_resource_putref(res);
1673                 }
1674         }
1675
1676         oa->o_valid = OBD_MD_FLID;
1677         obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
1678
1679 out_unlock:
1680         if (iattr.ia_valid & ATTR_SIZE)
1681                 up(&dentry->d_inode->i_sem);
1682         unlock_kernel();
1683         pop_ctxt(&saved, &exp->exp_obd->obd_ctxt, NULL);
1684
1685         f_dput(dentry);
1686         RETURN(rc);
1687 }
1688
1689 /* XXX identical to osc_unpackmd */
1690 static int filter_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
1691                            struct lov_mds_md *lmm, int lmm_bytes)
1692 {
1693         int lsm_size;
1694         ENTRY;
1695
1696         if (lmm != NULL) {
1697                 if (lmm_bytes < sizeof (*lmm)) {
1698                         CERROR("lov_mds_md too small: %d, need %d\n",
1699                                lmm_bytes, (int)sizeof(*lmm));
1700                         RETURN(-EINVAL);
1701                 }
1702                 /* XXX LOV_MAGIC etc check? */
1703
1704                 if (lmm->lmm_object_id == cpu_to_le64(0)) {
1705                         CERROR("lov_mds_md: zero lmm_object_id\n");
1706                         RETURN(-EINVAL);
1707                 }
1708         }
1709
1710         lsm_size = lov_stripe_md_size(1);
1711         if (lsmp == NULL)
1712                 RETURN(lsm_size);
1713
1714         if (*lsmp != NULL && lmm == NULL) {
1715                 OBD_FREE(*lsmp, lsm_size);
1716                 *lsmp = NULL;
1717                 RETURN(0);
1718         }
1719
1720         if (*lsmp == NULL) {
1721                 OBD_ALLOC(*lsmp, lsm_size);
1722                 if (*lsmp == NULL)
1723                         RETURN(-ENOMEM);
1724
1725                 loi_init((*lsmp)->lsm_oinfo);
1726         }
1727
1728         if (lmm != NULL) {
1729                 /* XXX zero *lsmp? */
1730                 (*lsmp)->lsm_object_id = le64_to_cpu (lmm->lmm_object_id);
1731                 LASSERT((*lsmp)->lsm_object_id);
1732         }
1733
1734         (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
1735
1736         RETURN(lsm_size);
1737 }
1738
1739 static void filter_destroy_precreated(struct obd_export *exp, struct obdo *oa,
1740                                       struct filter_obd *filter)
1741 {
1742         struct obdo doa; /* XXX obdo on stack */
1743         __u64 last, id;
1744         ENTRY;
1745         LASSERT(oa);
1746
1747         memset(&doa, 0, sizeof(doa));
1748         if (oa->o_valid & OBD_MD_FLGROUP)
1749                 doa.o_gr = oa->o_gr;
1750         else
1751                 doa.o_gr = 0;
1752         doa.o_mode = S_IFREG;
1753         last = filter_last_id(filter, &doa); /* FIXME: object groups */
1754         CWARN("deleting orphan objects from "LPU64" to "LPU64"\n",
1755                oa->o_id + 1, last);
1756         for (id = oa->o_id + 1; id <= last; id++) {
1757                 doa.o_id = id;
1758                 filter_destroy(exp, &doa, NULL, NULL);
1759         }
1760         spin_lock(&filter->fo_objidlock);
1761         filter->fo_last_objids[0] = oa->o_id; /* FIXME: object groups */
1762         spin_unlock(&filter->fo_objidlock);
1763         EXIT;
1764 }
1765
1766 /* returns a negative error or a nonnegative number of files to create */
1767 static int filter_should_precreate(struct obd_export *exp, struct obdo *oa,
1768                                    int group)
1769 {
1770         struct obd_device *obd = exp->exp_obd;
1771         struct filter_obd *filter = &obd->u.filter;
1772         int diff, rc;
1773         ENTRY;
1774
1775         diff = oa->o_id - filter_last_id(filter, oa);
1776         CDEBUG(D_INFO, "filter_last_id() = "LPU64" -> diff = %d\n",
1777                filter_last_id(filter, oa), diff);
1778
1779         /* delete orphans request */
1780         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1781             (oa->o_flags & OBD_FL_DELORPHAN)) {
1782                 if (diff >= 0)
1783                         RETURN(diff);
1784                 if (-diff > 10000) { /* XXX make this smarter */
1785                         CERROR("ignoring bogus orphan destroy request: obdid "
1786                                LPU64" last_id "LPU64"\n",
1787                                oa->o_id, filter_last_id(filter, oa));
1788                         RETURN(-EINVAL);
1789                 }
1790                 filter_destroy_precreated(exp, oa, filter);
1791                 rc = filter_update_last_objid(obd, group, 0);
1792                 if (rc)
1793                         CERROR("unable to write lastobjid, but orphans"
1794                                "were deleted\n");
1795                 RETURN(0);
1796         } else {
1797                 /* only precreate if group == 0 and o_id is specfied */
1798                 if (!(oa->o_valid & OBD_FL_DELORPHAN) &&
1799                     (group != 0 || oa->o_id == 0))
1800                         RETURN(1);
1801
1802                 LASSERT(diff >= 0);
1803                 RETURN(diff);
1804         }
1805 }
1806
1807 /* We rely on the fact that only one thread will be creating files in a given
1808  * group at a time, which is why we don't need an atomic filter_get_new_id.
1809  * Even if we had that atomic function, the following race would exist:
1810  *
1811  * thread 1: gets id x from filter_next_id
1812  * thread 2: gets id (x + 1) from filter_next_id
1813  * thread 2: creates object (x + 1)
1814  * thread 1: tries to create object x, gets -ENOSPC
1815  */
1816 static int filter_precreate(struct obd_device *obd, struct obdo *oa,
1817                             obd_gr group, int *num)
1818 {
1819         struct dentry *dchild = NULL;
1820         struct filter_obd *filter;
1821         struct dentry *dparent;
1822         int err = 0, rc = 0, i;
1823         __u64 next_id;
1824         int recreate_obj = 0;
1825         void *handle = NULL;
1826         ENTRY;
1827
1828         filter = &obd->u.filter;
1829
1830         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1831             (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1832                 recreate_obj = 1;
1833         }
1834
1835         for (i = 0; i < *num && err == 0; i++) {
1836                 int cleanup_phase = 0;
1837
1838                 if (recreate_obj) {
1839                         __u64 last_id;
1840                         next_id = oa->o_id;
1841                         last_id = filter_last_id(filter, NULL);
1842                         if (next_id > last_id) {
1843                                 CERROR("Error: Trying to recreate obj greater"
1844                                        "than last id "LPD64" > "LPD64"\n",
1845                                        next_id, last_id);
1846                                 RETURN(-EINVAL);
1847                         }
1848                 } else
1849                         next_id = filter_last_id(filter, NULL) + 1;
1850
1851                 CDEBUG(D_INFO, "precreate objid "LPU64"\n", next_id);
1852
1853                 dparent = filter_parent_lock(obd, group, next_id);
1854                 if (IS_ERR(dparent))
1855                         GOTO(cleanup, rc = PTR_ERR(dparent));
1856                 cleanup_phase = 1;
1857
1858                 dchild = filter_fid2dentry(obd, dparent, group, next_id);
1859                 if (IS_ERR(dchild))
1860                         GOTO(cleanup, rc = PTR_ERR(dchild));
1861                 cleanup_phase = 2;
1862
1863                 if (dchild->d_inode != NULL) {
1864                         /* This would only happen if lastobjid was bad on disk*/
1865                         /* Could also happen if recreating missing obj but
1866                          * already exists
1867                          */
1868                         if (recreate_obj) {
1869                                 CERROR("Serious error: recreating obj %*s but "
1870                                        "obj already exists \n",
1871                                        dchild->d_name.len, dchild->d_name.name);
1872                         } else {
1873                                 CERROR("Serious error: objid %*s already "
1874                                        "exists; is this filesystem corrupt?\n",
1875                                        dchild->d_name.len, dchild->d_name.name);
1876                         }
1877                         GOTO(cleanup, rc = -EEXIST);
1878                 }
1879
1880                 handle = fsfilt_start_log(obd, dparent->d_inode,
1881                                           FSFILT_OP_CREATE, NULL, 1);
1882                 if (IS_ERR(handle))
1883                         GOTO(cleanup, rc = PTR_ERR(handle));
1884                 cleanup_phase = 3;
1885
1886                 rc = ll_vfs_create(dparent->d_inode, dchild, S_IFREG, NULL);
1887                 if (rc) {
1888                         CERROR("create failed rc = %d\n", rc);
1889                         GOTO(cleanup, rc);
1890                 }
1891
1892                 if (!recreate_obj) {
1893                         filter_set_last_id(filter, NULL, next_id);
1894                         err = filter_update_last_objid(obd, group, 0);
1895                         if (err)
1896                                 CERROR("unable to write lastobjid "
1897                                        "but file created\n");
1898                 }
1899
1900         cleanup:
1901                 switch(cleanup_phase) {
1902                 case 3:
1903                         err = fsfilt_commit(obd, dparent->d_inode, handle, 0);
1904                         if (err) {
1905                                 CERROR("error on commit, err = %d\n", err);
1906                                 if (!rc)
1907                                         rc = err;
1908                         }
1909                 case 2:
1910                         f_dput(dchild);
1911                 case 1:
1912                         filter_parent_unlock(dparent);
1913                 case 0:
1914                         break;
1915                 }
1916
1917                 if (rc)
1918                         break;
1919         }
1920         *num = i;
1921
1922         CDEBUG(D_INFO, "filter_precreate() created %d objects\n", i);
1923         RETURN(rc);
1924 }
1925
1926 static int filter_create(struct obd_export *exp, struct obdo *oa,
1927                          struct lov_stripe_md **ea, struct obd_trans_info *oti)
1928 {
1929         struct obd_device *obd = NULL;
1930         struct obd_run_ctxt saved;
1931         struct lov_stripe_md *lsm = NULL;
1932         obd_gr group = 0;
1933         int rc = 0, diff;
1934         ENTRY;
1935
1936         if (oa->o_valid & OBD_MD_FLGROUP)
1937                 group = oa->o_gr;
1938
1939         CDEBUG(D_INFO, "filter_create(od->o_gr="LPU64",od->o_id="LPU64")\n",
1940                group, oa->o_id);
1941         if (ea != NULL) {
1942                 lsm = *ea;
1943                 if (lsm == NULL) {
1944                         rc = obd_alloc_memmd(exp, &lsm);
1945                         if (rc < 0)
1946                                 RETURN(rc);
1947                 }
1948         }
1949
1950         obd = exp->exp_obd;
1951         push_ctxt(&saved, &obd->obd_ctxt, NULL);
1952
1953         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1954             (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1955                 if (oa->o_id > filter_last_id(&obd->u.filter, oa)) {
1956                         CERROR("recreate objid "LPU64" > last id "LPU64"\n",
1957                                oa->o_id, filter_last_id(&obd->u.filter, oa));
1958                         rc = -EINVAL;
1959                 } else {
1960                         diff = 1;
1961                         rc = filter_precreate(obd, oa, group, &diff);
1962                 }
1963         } else {
1964                 diff = filter_should_precreate(exp, oa, group);
1965                 if (diff > 0) {
1966                         oa->o_id = filter_last_id(&obd->u.filter, oa);
1967                         rc = filter_precreate(obd, oa, group, &diff);
1968                         oa->o_id += diff;
1969                         oa->o_valid = OBD_MD_FLID;
1970                 }
1971         }
1972
1973         pop_ctxt(&saved, &obd->obd_ctxt, NULL);
1974         if (rc && ea != NULL && *ea != lsm) {
1975                 obd_free_memmd(exp, &lsm);
1976         } else if (rc == 0 && ea != NULL) {
1977                 /* XXX LOV STACKING: the lsm that is passed to us from
1978                  * LOV does not have valid lsm_oinfo data structs, so
1979                  * don't go touching that.  This needs to be fixed in a
1980                  * big way. */
1981                 lsm->lsm_object_id = oa->o_id;
1982                 *ea = lsm;
1983         }
1984
1985         RETURN(rc);
1986 }
1987
1988 static int filter_destroy(struct obd_export *exp, struct obdo *oa,
1989                           struct lov_stripe_md *ea, struct obd_trans_info *oti)
1990 {
1991         struct obd_device *obd;
1992         struct filter_obd *filter;
1993         struct dentry *dchild = NULL, *dparent = NULL;
1994         struct obd_run_ctxt saved;
1995         void *handle = NULL;
1996         struct llog_cookie *fcc = NULL;
1997         int rc, rc2, cleanup_phase = 0, have_prepared = 0;
1998         obd_gr group = 0;
1999         ENTRY;
2000
2001         if (oa->o_valid & OBD_MD_FLGROUP)
2002                 group = oa->o_gr;
2003
2004         obd = exp->exp_obd;
2005         filter = &obd->u.filter;
2006
2007         push_ctxt(&saved, &obd->obd_ctxt, NULL);
2008
2009  acquire_locks:
2010         dparent = filter_parent_lock(obd, group, oa->o_id);
2011         if (IS_ERR(dparent))
2012                 GOTO(cleanup, rc = PTR_ERR(dparent));
2013         cleanup_phase = 1;
2014
2015         dchild = filter_fid2dentry(obd, dparent, group, oa->o_id);
2016         if (IS_ERR(dchild))
2017                 GOTO(cleanup, rc = -ENOENT);
2018         cleanup_phase = 2;
2019
2020         if (dchild->d_inode == NULL) {
2021                 CDEBUG(D_INODE, "destroying non-existent object "LPU64"\n",
2022                        oa->o_id);
2023                 GOTO(cleanup, rc = -ENOENT);
2024         }
2025
2026         if (!have_prepared) {
2027                 /* If we're really going to destroy the object, get ready
2028                  * by getting the clients to discard their cached data.
2029                  *
2030                  * We have to drop the parent lock, because
2031                  * filter_prepare_destroy will acquire a PW on the object, and
2032                  * we don't want to deadlock with an incoming write to the
2033                  * object, which has the extent PW and then wants to get the
2034                  * parent dentry to do the lookup.
2035                  *
2036                  * We dput the child because it's not worth the extra
2037                  * complication of condition the above code to skip it on the
2038                  * second time through. */
2039                 f_dput(dchild);
2040                 filter_parent_unlock(dparent);
2041
2042                 filter_prepare_destroy(obd, oa->o_id);
2043                 have_prepared = 1;
2044                 goto acquire_locks;
2045         }
2046
2047         handle = fsfilt_start_log(obd, dparent->d_inode,FSFILT_OP_UNLINK,oti,1);
2048         if (IS_ERR(handle))
2049                 GOTO(cleanup, rc = PTR_ERR(handle));
2050         cleanup_phase = 3;
2051
2052         /* Our MDC connection is established by the MDS to us */
2053         if (oa->o_valid & OBD_MD_FLCOOKIE) {
2054                 OBD_ALLOC(fcc, sizeof(*fcc));
2055                 if (fcc != NULL)
2056                         memcpy(fcc, obdo_logcookie(oa), sizeof(*fcc));
2057         }
2058
2059         rc = filter_destroy_internal(obd, oa->o_id, dparent, dchild);
2060
2061 cleanup:
2062         switch(cleanup_phase) {
2063         case 3:
2064                 if (fcc != NULL) {
2065                         if (oti != NULL)
2066                                 fsfilt_add_journal_cb(obd, 0, oti->oti_handle,
2067                                                       filter_cancel_cookies_cb,
2068                                                       fcc);
2069                         else
2070                                 fsfilt_add_journal_cb(obd, 0, handle,
2071                                                       filter_cancel_cookies_cb,
2072                                                       fcc);
2073                 }
2074                 rc = filter_finish_transno(exp, oti, rc);
2075                 rc2 = fsfilt_commit(obd, dparent->d_inode, handle, 0);
2076                 if (rc2) {
2077                         CERROR("error on commit, err = %d\n", rc2);
2078                         if (!rc)
2079                                 rc = rc2;
2080                 }
2081         case 2:
2082                 f_dput(dchild);
2083         case 1:
2084                 filter_parent_unlock(dparent);
2085         case 0:
2086                 pop_ctxt(&saved, &obd->obd_ctxt, NULL);
2087                 break;
2088         default:
2089                 CERROR("invalid cleanup_phase %d\n", cleanup_phase);
2090                 LBUG();
2091         }
2092
2093         RETURN(rc);
2094 }
2095
2096 /* NB start and end are used for punch, but not truncate */
2097 static int filter_truncate(struct obd_export *exp, struct obdo *oa,
2098                            struct lov_stripe_md *lsm,
2099                            obd_off start, obd_off end,
2100                            struct obd_trans_info *oti)
2101 {
2102         int error;
2103         ENTRY;
2104
2105         if (end != OBD_OBJECT_EOF)
2106                 CERROR("PUNCH not supported, only truncate: end = "LPX64"\n",
2107                        end);
2108
2109         CDEBUG(D_INODE, "calling truncate for object "LPU64", valid = %x, "
2110                "o_size = "LPD64"\n", oa->o_id, oa->o_valid, start);
2111         oa->o_size = start;
2112         error = filter_setattr(exp, oa, NULL, oti);
2113         RETURN(error);
2114 }
2115
2116 static int filter_sync(struct obd_export *exp, struct obdo *oa,
2117                        struct lov_stripe_md *lsm, obd_off start, obd_off end)
2118 {
2119         struct obd_run_ctxt saved;
2120         struct filter_obd *filter;
2121         struct dentry *dentry;
2122         struct llog_ctxt *ctxt;
2123         int rc, rc2;
2124         ENTRY;
2125
2126         filter = &exp->exp_obd->u.filter;
2127
2128         /* an objid of zero is taken to mean "sync whole filesystem" */
2129         if (!oa || !(oa->o_valid & OBD_MD_FLID)) {
2130                 rc = fsfilt_sync(exp->exp_obd, filter->fo_sb);
2131                 /* flush any remaining cancel messages out to the target */
2132                 ctxt = llog_get_context(exp->exp_obd, LLOG_UNLINK_REPL_CTXT);
2133                 llog_sync(ctxt, exp);
2134                 RETURN(rc);
2135         }
2136
2137         dentry = filter_oa2dentry(exp->exp_obd, oa);
2138         if (IS_ERR(dentry))
2139                 RETURN(PTR_ERR(dentry));
2140
2141         push_ctxt(&saved, &exp->exp_obd->obd_ctxt, NULL);
2142
2143         down(&dentry->d_inode->i_sem);
2144         rc = filemap_fdatasync(dentry->d_inode->i_mapping);
2145         if (rc == 0) {
2146                 /* just any file to grab fsync method - "file" arg unused */
2147                 struct file *file = filter->fo_rcvd_filp;
2148
2149                 if (file->f_op && file->f_op->fsync)
2150                         rc = file->f_op->fsync(NULL, dentry, 1);
2151
2152                 rc2 = filemap_fdatawait(dentry->d_inode->i_mapping);
2153                 if (!rc)
2154                         rc = rc2;
2155         }
2156         up(&dentry->d_inode->i_sem);
2157
2158         oa->o_valid = OBD_MD_FLID;
2159         obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2160
2161         pop_ctxt(&saved, &exp->exp_obd->obd_ctxt, NULL);
2162
2163         f_dput(dentry);
2164         RETURN(rc);
2165 }
2166
2167 static int filter_statfs(struct obd_device *obd, struct obd_statfs *osfs,
2168                          unsigned long max_age)
2169 {
2170         struct filter_obd *filter = &obd->u.filter;
2171         int blockbits = filter->fo_sb->s_blocksize_bits;
2172         int rc;
2173         ENTRY;
2174
2175         /* at least try to account for cached pages.  its still racey and
2176          * might be under-reporting if clients haven't announced their
2177          * caches with brw recently */
2178         spin_lock(&obd->obd_osfs_lock);
2179         rc = fsfilt_statfs(obd, filter->fo_sb, max_age);
2180         memcpy(osfs, &obd->obd_osfs, sizeof(*osfs));
2181         spin_unlock(&obd->obd_osfs_lock);
2182
2183         CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
2184                " pending "LPU64" free "LPU64" avail "LPU64"\n",
2185                filter->fo_tot_dirty, filter->fo_tot_granted,
2186                filter->fo_tot_pending,
2187                osfs->os_bfree << blockbits, osfs->os_bavail << blockbits);
2188
2189         filter_grant_sanity_check(obd, __FUNCTION__);
2190
2191         osfs->os_bavail -= min(osfs->os_bavail,
2192                                (filter->fo_tot_dirty + filter->fo_tot_pending +
2193                                 osfs->os_bsize -1) >> blockbits);
2194
2195         RETURN(rc);
2196 }
2197
2198 static int filter_get_info(struct obd_export *exp, __u32 keylen,
2199                            void *key, __u32 *vallen, void *val)
2200 {
2201         struct obd_device *obd;
2202         ENTRY;
2203
2204         obd = class_exp2obd(exp);
2205         if (obd == NULL) {
2206                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2207                        exp->exp_handle.h_cookie);
2208                 RETURN(-EINVAL);
2209         }
2210
2211         if (keylen == strlen("blocksize") &&
2212             memcmp(key, "blocksize", keylen) == 0) {
2213                 __u32 *blocksize = val;
2214                 *vallen = sizeof(*blocksize);
2215                 *blocksize = obd->u.filter.fo_sb->s_blocksize;
2216                 RETURN(0);
2217         }
2218
2219         if (keylen == strlen("blocksize_bits") &&
2220             memcmp(key, "blocksize_bits", keylen) == 0) {
2221                 __u32 *blocksize_bits = val;
2222                 *vallen = sizeof(*blocksize_bits);
2223                 *blocksize_bits = obd->u.filter.fo_sb->s_blocksize_bits;
2224                 RETURN(0);
2225         }
2226
2227         if (keylen >= strlen("last_id") && memcmp(key, "last_id", 7) == 0) {
2228                 obd_id *last_id = val;
2229                 /* FIXME: object groups */
2230                 *last_id = filter_last_id(&obd->u.filter, 0);
2231                 RETURN(0);
2232         }
2233         CDEBUG(D_IOCTL, "invalid key\n");
2234         RETURN(-EINVAL);
2235 }
2236
2237 static int filter_set_info(struct obd_export *exp, __u32 keylen,
2238                            void *key, __u32 vallen, void *val)
2239 {
2240         struct obd_device *obd;
2241         struct lustre_handle conn;
2242         struct llog_ctxt *ctxt;
2243         int rc = 0;
2244         ENTRY;
2245
2246         conn.cookie = exp->exp_handle.h_cookie;
2247
2248         obd = exp->exp_obd;
2249         if (obd == NULL) {
2250                 CDEBUG(D_IOCTL, "invalid exp %p cookie "LPX64"\n",
2251                        exp, conn.cookie);
2252                 RETURN(-EINVAL);
2253         }
2254
2255         if (keylen < strlen("mds_conn") ||
2256             memcmp(key, "mds_conn", keylen) != 0)
2257                 RETURN(-EINVAL);
2258
2259         CWARN("Received MDS connection ("LPX64")\n", conn.cookie);
2260         memcpy(&obd->u.filter.fo_mdc_conn, &conn, sizeof(conn));
2261         ctxt = llog_get_context(obd, LLOG_UNLINK_REPL_CTXT);
2262         rc = llog_receptor_accept(ctxt, exp->exp_imp_reverse);
2263         RETURN(rc);
2264 }
2265
2266 int filter_iocontrol(unsigned int cmd, struct obd_export *exp,
2267                      int len, void *karg, void *uarg)
2268 {
2269         struct obd_device *obd = exp->exp_obd;
2270         struct obd_ioctl_data *data = karg;
2271         int rc = 0;
2272
2273         switch (cmd) {
2274         case OBD_IOC_ABORT_RECOVERY:
2275                 CERROR("aborting recovery for device %s\n", obd->obd_name);
2276                 target_abort_recovery(obd);
2277                 RETURN(0);
2278
2279         case OBD_IOC_SET_READONLY: {
2280                 void *handle;
2281                 struct super_block *sb = obd->u.filter.fo_sb;
2282                 struct inode *inode = sb->s_root->d_inode;
2283                 BDEVNAME_DECLARE_STORAGE(tmp);
2284                 CERROR("setting device %s read-only\n",
2285                        ll_bdevname(sb, tmp));
2286
2287                 handle = fsfilt_start(obd, inode, FSFILT_OP_MKNOD, NULL);
2288                 LASSERT(handle);
2289                 (void)fsfilt_commit(obd, inode, handle, 1);
2290
2291                 dev_set_rdonly(ll_sbdev(obd->u.filter.fo_sb), 2);
2292                 RETURN(0);
2293         }
2294
2295         case OBD_IOC_CATLOGLIST: {
2296                 rc = llog_catlog_list(obd, 1, data);
2297                 RETURN(rc);
2298         }
2299
2300         case OBD_IOC_LLOG_CANCEL:
2301         case OBD_IOC_LLOG_REMOVE:
2302         case OBD_IOC_LLOG_INFO:
2303         case OBD_IOC_LLOG_PRINT: {
2304                 /* FIXME to be finished */
2305                 RETURN(-EOPNOTSUPP);
2306 /*
2307                 struct llog_ctxt *ctxt = NULL;
2308
2309                 push_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_ctxt, NULL);
2310                 rc = llog_ioctl(ctxt, cmd, data);
2311                 pop_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_ctxt, NULL);
2312
2313                 RETURN(rc);
2314 */
2315         }
2316
2317
2318         default:
2319                 RETURN(-EINVAL);
2320         }
2321         RETURN(0);
2322 }
2323
2324 static struct llog_operations filter_unlink_repl_logops;
2325 static struct llog_operations filter_size_orig_logops = {
2326         lop_setup: llog_obd_origin_setup,
2327         lop_cleanup: llog_obd_origin_cleanup,
2328         lop_add: llog_obd_origin_add
2329 };
2330
2331 static int filter_llog_init(struct obd_device *obd, struct obd_device *tgt,
2332                             int count, struct llog_catid *logid)
2333 {
2334         struct llog_ctxt *ctxt;
2335         int rc;
2336         ENTRY;
2337
2338         filter_unlink_repl_logops = llog_client_ops;
2339         filter_unlink_repl_logops.lop_cancel = llog_obd_repl_cancel;
2340         filter_unlink_repl_logops.lop_connect = llog_repl_connect;
2341         filter_unlink_repl_logops.lop_sync = llog_obd_repl_sync;
2342
2343         rc = llog_setup(obd, LLOG_UNLINK_REPL_CTXT, tgt, 0, NULL,
2344                         &filter_unlink_repl_logops);
2345         if (rc)
2346                 RETURN(rc);
2347         /* FIXME - assign unlink_cb for filter's recovery */
2348         ctxt = llog_get_context(obd, LLOG_UNLINK_REPL_CTXT);
2349         ctxt->llog_proc_cb = filter_recov_log_unlink_cb;
2350
2351         rc = llog_setup(obd, LLOG_SIZE_ORIG_CTXT, tgt, 0, NULL,
2352                         &filter_size_orig_logops);
2353         RETURN(rc);
2354 }
2355
2356 static int filter_llog_finish(struct obd_device *obd, int count)
2357 {
2358         int rc;
2359         ENTRY;
2360
2361         rc = llog_cleanup(llog_get_context(obd, LLOG_UNLINK_REPL_CTXT));
2362         if (rc)
2363                 RETURN(rc);
2364
2365         rc = llog_cleanup(llog_get_context(obd, LLOG_SIZE_ORIG_CTXT));
2366         RETURN(rc);
2367 }
2368
2369 static struct dentry *filter_lvfs_fid2dentry(__u64 id, __u32 gen, __u64 gr,
2370                                              void *data)
2371 {
2372         return filter_fid2dentry(data, NULL, gr, id);
2373 }
2374
2375 static struct lvfs_callback_ops filter_lvfs_ops = {
2376         l_fid2dentry:     filter_lvfs_fid2dentry,
2377 };
2378
2379 static struct obd_ops filter_obd_ops = {
2380         o_owner:          THIS_MODULE,
2381         o_get_info:       filter_get_info,
2382         o_set_info:       filter_set_info,
2383         o_setup:          filter_setup,
2384         o_precleanup:     filter_precleanup,
2385         o_cleanup:        filter_cleanup,
2386         o_connect:        filter_connect,
2387         o_disconnect:     filter_disconnect,
2388         o_statfs:         filter_statfs,
2389         o_getattr:        filter_getattr,
2390         o_unpackmd:       filter_unpackmd,
2391         o_create:         filter_create,
2392         o_setattr:        filter_setattr,
2393         o_destroy:        filter_destroy,
2394         o_brw:            filter_brw,
2395         o_punch:          filter_truncate,
2396         o_sync:           filter_sync,
2397         o_preprw:         filter_preprw,
2398         o_commitrw:       filter_commitrw,
2399         o_destroy_export: filter_destroy_export,
2400         o_llog_init:      filter_llog_init,
2401         o_llog_finish:    filter_llog_finish,
2402         o_iocontrol:      filter_iocontrol,
2403 };
2404
2405 static struct obd_ops filter_sanobd_ops = {
2406         o_owner:          THIS_MODULE,
2407         o_get_info:       filter_get_info,
2408         o_set_info:       filter_set_info,
2409         o_setup:          filter_san_setup,
2410         o_precleanup:     filter_precleanup,
2411         o_cleanup:        filter_cleanup,
2412         o_connect:        filter_connect,
2413         o_disconnect:     filter_disconnect,
2414         o_statfs:         filter_statfs,
2415         o_getattr:        filter_getattr,
2416         o_unpackmd:       filter_unpackmd,
2417         o_create:         filter_create,
2418         o_setattr:        filter_setattr,
2419         o_destroy:        filter_destroy,
2420         o_brw:            filter_brw,
2421         o_punch:          filter_truncate,
2422         o_sync:           filter_sync,
2423         o_preprw:         filter_preprw,
2424         o_commitrw:       filter_commitrw,
2425         o_san_preprw:     filter_san_preprw,
2426         o_destroy_export: filter_destroy_export,
2427         o_llog_init:      filter_llog_init,
2428         o_llog_finish:    filter_llog_finish,
2429         o_iocontrol:      filter_iocontrol,
2430 };
2431
2432 static int __init obdfilter_init(void)
2433 {
2434         struct lprocfs_static_vars lvars;
2435         int rc;
2436
2437         printk(KERN_INFO "Lustre: Filtering OBD driver; info@clusterfs.com\n");
2438
2439         lprocfs_init_vars(filter, &lvars);
2440
2441         rc = class_register_type(&filter_obd_ops, lvars.module_vars,
2442                                  OBD_FILTER_DEVICENAME);
2443         if (rc)
2444                 return rc;
2445
2446         rc = class_register_type(&filter_sanobd_ops, lvars.module_vars,
2447                                  OBD_FILTER_SAN_DEVICENAME);
2448         if (rc)
2449                 class_unregister_type(OBD_FILTER_DEVICENAME);
2450         return rc;
2451 }
2452
2453 static void __exit obdfilter_exit(void)
2454 {
2455         class_unregister_type(OBD_FILTER_SAN_DEVICENAME);
2456         class_unregister_type(OBD_FILTER_DEVICENAME);
2457 }
2458
2459 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2460 MODULE_DESCRIPTION("Lustre Filtering OBD driver");
2461 MODULE_LICENSE("GPL");
2462
2463 module_init(obdfilter_init);
2464 module_exit(obdfilter_exit);