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/obd_ost.h>
  52 #include <linux/lustre_dlm.h>
  53 #include <linux/lustre_fsfilt.h>
  54 #include <linux/lprocfs_status.h>
  55 #include <linux/lustre_log.h>
  56 #include <linux/lustre_commit_confd.h>
  57 #include <portals/list.h>
  58
  59 #include <linux/lustre_smfs.h>
  60 #include "filter_internal.h"
  61
  62 /* Group 0 is no longer a legal group, to catch uninitialized IDs */
  63 #define FILTER_MIN_GROUPS 3
  64
  65 static struct lvfs_callback_ops filter_lvfs_ops;
  66
  67 static int filter_destroy(struct obd_export *exp, struct obdo *oa,
  68                           struct lov_stripe_md *ea, struct obd_trans_info *);
  69 struct obd_llogs *filter_grab_llog_for_group(struct obd_device *,
  70                                              int, struct obd_export *);
  71
  72 static void filter_commit_cb(struct obd_device *obd, __u64 transno,
  73                              void *cb_data, int error)
  74 {
  75         obd_transno_commit_cb(obd, transno, error);
  76 }
  77
  78 /* Assumes caller has already pushed us into the kernel context. */
  79 int filter_finish_transno(struct obd_export *exp, struct obd_trans_info *oti,
  80                           int rc)
  81 {
  82         struct filter_obd *filter = &exp->exp_obd->u.filter;
  83         struct filter_export_data *fed = &exp->exp_filter_data;
  84         struct filter_client_data *fcd = fed->fed_fcd;
  85         __u64 last_rcvd;
  86         loff_t off;
  87         int err, log_pri = D_HA;
  88
  89         /* Propagate error code. */
  90         if (rc)
  91                 RETURN(rc);
  92
  93         if (!exp->exp_obd->obd_replayable || oti == NULL)
  94                 RETURN(rc);
  95
  96         /* we don't allocate new transnos for replayed requests */
  97         if (oti->oti_transno == 0) {
  98                 spin_lock(&filter->fo_translock);
  99                 last_rcvd = le64_to_cpu(filter->fo_fsd->fsd_last_transno) + 1;
 100                 filter->fo_fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
 101                 spin_unlock(&filter->fo_translock);
 102                 oti->oti_transno = last_rcvd;
 103         } else {
 104                 spin_lock(&filter->fo_translock);
 105                 last_rcvd = oti->oti_transno;
 106                 if (last_rcvd > le64_to_cpu(filter->fo_fsd->fsd_last_transno))
 107                         filter->fo_fsd->fsd_last_transno =
 108                                 cpu_to_le64(last_rcvd);
 109                 spin_unlock(&filter->fo_translock);
 110         }
 111         fcd->fcd_last_rcvd = cpu_to_le64(last_rcvd);
 112
 113         /* could get xid from oti, if it's ever needed */
 114         fcd->fcd_last_xid = 0;
 115
 116         off = fed->fed_lr_off;
 117
 118         fsfilt_add_journal_cb(exp->exp_obd, filter->fo_sb, last_rcvd,
 119                               oti->oti_handle, filter_commit_cb, NULL);
 120
 121         err = fsfilt_write_record(exp->exp_obd, filter->fo_rcvd_filp, fcd,
 122                                   sizeof(*fcd), &off, 0);
 123         if (err) {
 124                 log_pri = D_ERROR;
 125                 if (rc == 0)
 126                         rc = err;
 127         }
 128
 129         CDEBUG(log_pri, "wrote trans "LPU64" for client %s at #%d: err = %d\n",
 130                last_rcvd, fcd->fcd_uuid, fed->fed_lr_idx, err);
 131
 132         RETURN(rc);
 133 }
 134
 135 void f_dput(struct dentry *dentry)
 136 {
 137         /* Can't go inside filter_ddelete because it can block */
 138         CDEBUG(D_INODE, "putting %s: %p, count = %d\n",
 139                dentry->d_name.name, dentry, atomic_read(&dentry->d_count) - 1);
 140         LASSERT(atomic_read(&dentry->d_count) > 0);
 141
 142         dput(dentry);
 143 }
 144
 145 /* Add client data to the FILTER.  We use a bitmap to locate a free space
 146  * in the last_rcvd file if cl_idx is -1 (i.e. a new client).
 147  * Otherwise, we have just read the data from the last_rcvd file and
 148  * we know its offset. */
 149 static int filter_client_add(struct obd_device *obd, struct filter_obd *filter,
 150                              struct filter_export_data *fed, int cl_idx)
 151 {
 152         unsigned long *bitmap = filter->fo_last_rcvd_slots;
 153         int new_client = (cl_idx == -1);
 154         ENTRY;
 155
 156         LASSERT(bitmap != NULL);
 157
 158         /* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
 159         if (!strcmp(fed->fed_fcd->fcd_uuid, obd->obd_uuid.uuid))
 160                 RETURN(0);
 161
 162         /* the bitmap operations can handle cl_idx > sizeof(long) * 8, so
 163          * there's no need for extra complication here
 164          */
 165         if (new_client) {
 166                 cl_idx = find_first_zero_bit(bitmap, FILTER_LR_MAX_CLIENTS);
 167         repeat:
 168                 if (cl_idx >= FILTER_LR_MAX_CLIENTS) {
 169                         CERROR("no client slots - fix FILTER_LR_MAX_CLIENTS\n");
 170                         RETURN(-ENOMEM);
 171                 }
 172                 if (test_and_set_bit(cl_idx, bitmap)) {
 173                         CERROR("FILTER client %d: found bit is set in bitmap\n",
 174                                cl_idx);
 175                         cl_idx = find_next_zero_bit(bitmap,
 176                                                     FILTER_LR_MAX_CLIENTS,
 177                                                     cl_idx);
 178                         goto repeat;
 179                 }
 180         } else {
 181                 if (test_and_set_bit(cl_idx, bitmap)) {
 182                         CERROR("FILTER client %d: bit already set in bitmap!\n",
 183                                cl_idx);
 184                         LBUG();
 185                 }
 186         }
 187
 188         fed->fed_lr_idx = cl_idx;
 189         fed->fed_lr_off = le32_to_cpu(filter->fo_fsd->fsd_client_start) +
 190                 cl_idx * le16_to_cpu(filter->fo_fsd->fsd_client_size);
 191
 192         CDEBUG(D_INFO, "client at index %d (%llu) with UUID '%s' added\n",
 193                fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
 194
 195         if (new_client) {
 196                 struct lvfs_run_ctxt saved;
 197                 loff_t off = fed->fed_lr_off;
 198                 int err;
 199                 void *handle;
 200
 201                 CDEBUG(D_INFO, "writing client fcd at idx %u (%llu) (len %u)\n",
 202                        fed->fed_lr_idx,off,(unsigned int)sizeof(*fed->fed_fcd));
 203
 204                 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 205                 /* Transaction needed to fix bug 1403 */
 206                 handle = fsfilt_start(obd,
 207                                       filter->fo_rcvd_filp->f_dentry->d_inode,
 208                                       FSFILT_OP_SETATTR, NULL);
 209                 if (IS_ERR(handle)) {
 210                         err = PTR_ERR(handle);
 211                         CERROR("unable to start transaction: rc %d\n", err);
 212                 } else {
 213                         err = fsfilt_write_record(obd, filter->fo_rcvd_filp,
 214                                                   fed->fed_fcd,
 215                                                   sizeof(*fed->fed_fcd),
 216                                                   &off, 1);
 217                         fsfilt_commit(obd, filter->fo_sb,
 218                                       filter->fo_rcvd_filp->f_dentry->d_inode,
 219                                       handle, 1);
 220                 }
 221                 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 222
 223                 if (err) {
 224                         CERROR("error writing %s client idx %u: rc %d\n",
 225                                LAST_RCVD, fed->fed_lr_idx, err);
 226                         RETURN(err);
 227                 }
 228         }
 229         RETURN(0);
 230 }
 231
 232 static int filter_client_free(struct obd_export *exp, int flags)
 233 {
 234         struct filter_export_data *fed = &exp->exp_filter_data;
 235         struct filter_obd *filter = &exp->exp_obd->u.filter;
 236         struct obd_device *obd = exp->exp_obd;
 237         struct filter_client_data zero_fcd;
 238         struct lvfs_run_ctxt saved;
 239         int rc;
 240         loff_t off;
 241         ENTRY;
 242
 243         if (fed->fed_fcd == NULL)
 244                 RETURN(0);
 245
 246         if (flags & OBD_OPT_FAILOVER)
 247                 GOTO(free, 0);
 248
 249         /* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
 250         if (strcmp(fed->fed_fcd->fcd_uuid, obd->obd_uuid.uuid ) == 0)
 251                 GOTO(free, 0);
 252
 253         LASSERT(filter->fo_last_rcvd_slots != NULL);
 254
 255         off = fed->fed_lr_off;
 256
 257         CDEBUG(D_INFO, "freeing client at idx %u (%lld) with UUID '%s'\n",
 258                fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
 259
 260         /* Clear the bit _after_ zeroing out the client so we don't
 261            race with filter_client_add and zero out new clients.*/
 262         if (!test_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
 263                 CERROR("FILTER client %u: bit already clear in bitmap!!\n",
 264                        fed->fed_lr_idx);
 265                 LBUG();
 266         }
 267
 268         memset(&zero_fcd, 0, sizeof zero_fcd);
 269         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 270         rc = fsfilt_write_record(obd, filter->fo_rcvd_filp, &zero_fcd,
 271                                  sizeof(zero_fcd), &off, 1);
 272         if (rc == 0)
 273                 /* update server's transno */
 274                 filter_update_server_data(obd, filter->fo_rcvd_filp,
 275                                           filter->fo_fsd, 1);
 276         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 277
 278         CDEBUG(rc == 0 ? D_INFO : D_ERROR,
 279                "zeroing disconnecting client %s at idx %u (%llu) in %s rc %d\n",
 280                fed->fed_fcd->fcd_uuid, fed->fed_lr_idx, fed->fed_lr_off,
 281                LAST_RCVD, rc);
 282
 283         if (!test_and_clear_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
 284                 CERROR("FILTER client %u: bit already clear in bitmap!!\n",
 285                        fed->fed_lr_idx);
 286                 LBUG();
 287         }
 288
 289 free:
 290         OBD_FREE(fed->fed_fcd, sizeof(*fed->fed_fcd));
 291
 292         RETURN(0);
 293 }
 294
 295 static int filter_free_server_data(struct filter_obd *filter)
 296 {
 297         OBD_FREE(filter->fo_fsd, sizeof(*filter->fo_fsd));
 298         filter->fo_fsd = NULL;
 299         OBD_FREE(filter->fo_last_rcvd_slots,
 300                  FILTER_LR_MAX_CLIENT_WORDS * sizeof(unsigned long));
 301         filter->fo_last_rcvd_slots = NULL;
 302         return 0;
 303 }
 304
 305 /* assumes caller is already in kernel ctxt */
 306 int filter_update_server_data(struct obd_device *obd, struct file *filp,
 307                               struct filter_server_data *fsd, int force_sync)
 308 {
 309         loff_t off = 0;
 310         int rc;
 311         ENTRY;
 312
 313         CDEBUG(D_INODE, "server uuid      : %s\n", fsd->fsd_uuid);
 314         CDEBUG(D_INODE, "server last_rcvd : "LPU64"\n",
 315                le64_to_cpu(fsd->fsd_last_transno));
 316         CDEBUG(D_INODE, "server last_mount: "LPU64"\n",
 317                le64_to_cpu(fsd->fsd_mount_count));
 318
 319         rc = fsfilt_write_record(obd, filp, fsd, sizeof(*fsd), &off,force_sync);
 320         if (rc)
 321                 CERROR("error writing filter_server_data: rc = %d\n", rc);
 322
 323         RETURN(rc);
 324 }
 325
 326 int filter_update_last_objid(struct obd_device *obd, obd_gr group,
 327                              int force_sync)
 328 {
 329         struct filter_obd *filter = &obd->u.filter;
 330         __u64 tmp;
 331         loff_t off = 0;
 332         int rc;
 333         ENTRY;
 334
 335         if (filter->fo_last_objid_files[group] == NULL) {
 336                 CERROR("Object group "LPU64" not fully setup; not updating "
 337                        "last_objid\n", group);
 338                 RETURN(0);
 339         }
 340
 341         CDEBUG(D_INODE, "server last_objid for group "LPU64": "LPU64"\n",
 342                group, filter->fo_last_objids[group]);
 343
 344         tmp = cpu_to_le64(filter->fo_last_objids[group]);
 345         rc = fsfilt_write_record(obd, filter->fo_last_objid_files[group],
 346                                  &tmp, sizeof(tmp), &off, force_sync);
 347         if (rc)
 348                 CERROR("error writing group "LPU64" last objid: rc = %d\n",
 349                        group, rc);
 350         RETURN(rc);
 351 }
 352
 353 /* assumes caller has already in kernel ctxt */
 354 static int filter_init_server_data(struct obd_device *obd, struct file * filp)
 355 {
 356         struct filter_obd *filter = &obd->u.filter;
 357         struct filter_server_data *fsd;
 358         struct filter_client_data *fcd = NULL;
 359         struct inode *inode = filp->f_dentry->d_inode;
 360         unsigned long last_rcvd_size = inode->i_size;
 361         __u64 mount_count;
 362         int cl_idx;
 363         loff_t off = 0;
 364         int rc;
 365
 366         /* ensure padding in the struct is the correct size */
 367         LASSERT (offsetof(struct filter_server_data, fsd_padding) +
 368                  sizeof(fsd->fsd_padding) == FILTER_LR_SERVER_SIZE);
 369         LASSERT (offsetof(struct filter_client_data, fcd_padding) +
 370                  sizeof(fcd->fcd_padding) == FILTER_LR_CLIENT_SIZE);
 371
 372         OBD_ALLOC(fsd, sizeof(*fsd));
 373         if (!fsd)
 374                 RETURN(-ENOMEM);
 375         filter->fo_fsd = fsd;
 376
 377         OBD_ALLOC(filter->fo_last_rcvd_slots,
 378                   FILTER_LR_MAX_CLIENT_WORDS * sizeof(unsigned long));
 379         if (filter->fo_last_rcvd_slots == NULL) {
 380                 OBD_FREE(fsd, sizeof(*fsd));
 381                 RETURN(-ENOMEM);
 382         }
 383
 384         if (last_rcvd_size == 0) {
 385                 CWARN("%s: initializing new %s\n", obd->obd_name, LAST_RCVD);
 386
 387                 memcpy(fsd->fsd_uuid, obd->obd_uuid.uuid,sizeof(fsd->fsd_uuid));
 388                 fsd->fsd_last_transno = 0;
 389                 mount_count = fsd->fsd_mount_count = 0;
 390                 fsd->fsd_server_size = cpu_to_le32(FILTER_LR_SERVER_SIZE);
 391                 fsd->fsd_client_start = cpu_to_le32(FILTER_LR_CLIENT_START);
 392                 fsd->fsd_client_size = cpu_to_le16(FILTER_LR_CLIENT_SIZE);
 393                 fsd->fsd_subdir_count = cpu_to_le16(FILTER_SUBDIR_COUNT);
 394                 filter->fo_subdir_count = FILTER_SUBDIR_COUNT;
 395         } else {
 396                 rc = fsfilt_read_record(obd, filp, fsd, sizeof(*fsd), &off);
 397                 if (rc) {
 398                         CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
 399                                LAST_RCVD, rc);
 400                         GOTO(err_fsd, rc);
 401                 }
 402                 if (strcmp(fsd->fsd_uuid, obd->obd_uuid.uuid) != 0) {
 403                         CERROR("OBD UUID %s does not match last_rcvd UUID %s\n",
 404                                obd->obd_uuid.uuid, fsd->fsd_uuid);
 405                         GOTO(err_fsd, rc = -EINVAL);
 406                 }
 407                 mount_count = le64_to_cpu(fsd->fsd_mount_count);
 408                 filter->fo_subdir_count = le16_to_cpu(fsd->fsd_subdir_count);
 409         }
 410
 411         if (fsd->fsd_feature_incompat & ~cpu_to_le32(FILTER_INCOMPAT_SUPP)) {
 412                 CERROR("unsupported feature %x\n",
 413                        le32_to_cpu(fsd->fsd_feature_incompat) &
 414                        ~FILTER_INCOMPAT_SUPP);
 415                 GOTO(err_fsd, rc = -EINVAL);
 416         }
 417         if (fsd->fsd_feature_rocompat & ~cpu_to_le32(FILTER_ROCOMPAT_SUPP)) {
 418                 CERROR("read-only feature %x\n",
 419                        le32_to_cpu(fsd->fsd_feature_rocompat) &
 420                        ~FILTER_ROCOMPAT_SUPP);
 421                 /* Do something like remount filesystem read-only */
 422                 GOTO(err_fsd, rc = -EINVAL);
 423         }
 424
 425         CDEBUG(D_INODE, "%s: server last_rcvd : "LPU64"\n",
 426                obd->obd_name, le64_to_cpu(fsd->fsd_last_transno));
 427         CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
 428                obd->obd_name, mount_count + 1);
 429         CDEBUG(D_INODE, "%s: server data size: %u\n",
 430                obd->obd_name, le32_to_cpu(fsd->fsd_server_size));
 431         CDEBUG(D_INODE, "%s: per-client data start: %u\n",
 432                obd->obd_name, le32_to_cpu(fsd->fsd_client_start));
 433         CDEBUG(D_INODE, "%s: per-client data size: %u\n",
 434                obd->obd_name, le32_to_cpu(fsd->fsd_client_size));
 435         CDEBUG(D_INODE, "%s: server subdir_count: %u\n",
 436                obd->obd_name, le16_to_cpu(fsd->fsd_subdir_count));
 437         CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
 438                last_rcvd_size <= le32_to_cpu(fsd->fsd_client_start) ? 0 :
 439                (last_rcvd_size - le32_to_cpu(fsd->fsd_client_start)) /
 440                 le16_to_cpu(fsd->fsd_client_size));
 441
 442         if (!obd->obd_replayable) {
 443                 CWARN("%s: recovery support OFF\n", obd->obd_name);
 444                 GOTO(out, rc = 0);
 445         }
 446
 447         for (cl_idx = 0, off = le32_to_cpu(fsd->fsd_client_start);
 448              off < last_rcvd_size; cl_idx++) {
 449                 __u64 last_rcvd;
 450                 struct obd_export *exp;
 451                 struct filter_export_data *fed;
 452
 453                 if (!fcd) {
 454                         OBD_ALLOC(fcd, sizeof(*fcd));
 455                         if (!fcd)
 456                                 GOTO(err_client, rc = -ENOMEM);
 457                 }
 458
 459                 /* Don't assume off is incremented properly by
 460                  * fsfilt_read_record(), in case sizeof(*fcd)
 461                  * isn't the same as fsd->fsd_client_size.  */
 462                 off = le32_to_cpu(fsd->fsd_client_start) +
 463                         cl_idx * le16_to_cpu(fsd->fsd_client_size);
 464                 rc = fsfilt_read_record(obd, filp, fcd, sizeof(*fcd), &off);
 465                 if (rc) {
 466                         CERROR("error reading FILT %s idx %d off %llu: rc %d\n",
 467                                LAST_RCVD, cl_idx, off, rc);
 468                         break; /* read error shouldn't cause startup to fail */
 469                 }
 470
 471                 if (fcd->fcd_uuid[0] == '\0') {
 472                         CDEBUG(D_INFO, "skipping zeroed client at offset %d\n",
 473                                cl_idx);
 474                         continue;
 475                 }
 476
 477                 last_rcvd = le64_to_cpu(fcd->fcd_last_rcvd);
 478
 479                 /* These exports are cleaned up by filter_disconnect(), so they
 480                  * need to be set up like real exports as filter_connect() does.
 481                  */
 482                 exp = class_new_export(obd);
 483                 CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64
 484                        " srv lr: "LPU64"\n", fcd->fcd_uuid, cl_idx,
 485                        last_rcvd, le64_to_cpu(fsd->fsd_last_transno));
 486                 if (exp == NULL)
 487                         GOTO(err_client, rc = -ENOMEM);
 488
 489                 memcpy(&exp->exp_client_uuid.uuid, fcd->fcd_uuid,
 490                        sizeof exp->exp_client_uuid.uuid);
 491                 fed = &exp->exp_filter_data;
 492                 fed->fed_fcd = fcd;
 493                 filter_client_add(obd, filter, fed, cl_idx);
 494                 /* create helper if export init gets more complex */
 495                 spin_lock_init(&fed->fed_lock);
 496
 497                 fcd = NULL;
 498                 exp->exp_replay_needed = 1;
 499                 obd->obd_recoverable_clients++;
 500                 obd->obd_max_recoverable_clients++;
 501                 class_export_put(exp);
 502
 503                 CDEBUG(D_OTHER, "client at idx %d has last_rcvd = "LPU64"\n",
 504                        cl_idx, last_rcvd);
 505
 506                 if (last_rcvd > le64_to_cpu(fsd->fsd_last_transno))
 507                         fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
 508
 509         }
 510
 511         obd->obd_last_committed = le64_to_cpu(fsd->fsd_last_transno);
 512
 513         if (obd->obd_recoverable_clients) {
 514                 CWARN("RECOVERY: %d recoverable clients, last_rcvd "
 515                       LPU64"\n", obd->obd_recoverable_clients,
 516                       le64_to_cpu(fsd->fsd_last_transno));
 517                 obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
 518                 target_start_recovery_thread(obd, ost_handle);
 519         }
 520
 521         if (fcd)
 522                 OBD_FREE(fcd, sizeof(*fcd));
 523
 524 out:
 525         filter->fo_mount_count = mount_count + 1;
 526         fsd->fsd_mount_count = cpu_to_le64(filter->fo_mount_count);
 527
 528         /* save it, so mount count and last_transno is current */
 529         rc = filter_update_server_data(obd, filp, filter->fo_fsd, 1);
 530
 531         RETURN(rc);
 532
 533 err_client:
 534         class_disconnect_exports(obd, 0);
 535 err_fsd:
 536         filter_free_server_data(filter);
 537         RETURN(rc);
 538 }
 539
 540 static int filter_cleanup_groups(struct obd_device *obd)
 541 {
 542         struct filter_obd *filter = &obd->u.filter;
 543         struct dentry *dentry;
 544         int i, k;
 545         ENTRY;
 546
 547         for (i = 0; i < filter->fo_group_count; i++) {
 548                 if (filter->fo_subdirs != NULL) {
 549                         for (k = 0; k < filter->fo_subdir_count; k++) {
 550                                 dentry = filter->fo_subdirs[i].dentry[k];
 551                                 if (dentry == NULL)
 552                                         continue;
 553                                 f_dput(dentry);
 554                                 filter->fo_subdirs[i].dentry[k] = NULL;
 555                         }
 556                 }
 557                 if (filter->fo_last_objid_files[i] != NULL) {
 558                         filp_close(filter->fo_last_objid_files[i], 0);
 559                         filter->fo_last_objid_files[i] = NULL;
 560                 }
 561                 if (filter->fo_groups[i] != NULL) {
 562                         dput(filter->fo_groups[i]);
 563                         filter->fo_groups[i] = NULL;
 564                 }
 565         }
 566         if (filter->fo_subdirs != NULL)
 567                 OBD_FREE(filter->fo_subdirs,
 568                          filter->fo_group_count * sizeof(*filter->fo_subdirs));
 569         if (filter->fo_groups != NULL)
 570                 OBD_FREE(filter->fo_groups,
 571                          filter->fo_group_count * sizeof(*filter->fo_groups));
 572         if (filter->fo_last_objids != NULL)
 573                 OBD_FREE(filter->fo_last_objids,
 574                          filter->fo_group_count * sizeof(__u64));
 575         if (filter->fo_last_objid_files != NULL)
 576                 OBD_FREE(filter->fo_last_objid_files,
 577                          filter->fo_group_count * sizeof(struct file *));
 578         f_dput(filter->fo_dentry_O);
 579         RETURN(0);
 580 }
 581
 582 static int filter_update_last_group(struct obd_device *obd, int group)
 583 {
 584         struct filter_obd *filter = &obd->u.filter;
 585         struct file *filp = NULL;
 586         int last_group = 0, rc;
 587         loff_t off = 0;
 588         ENTRY;
 589
 590         if (group <= filter->fo_committed_group)
 591                 RETURN(0);
 592
 593         filp = filp_open("LAST_GROUP", O_RDWR, 0700);
 594         if (IS_ERR(filp)) {
 595                 rc = PTR_ERR(filp);
 596                 filp = NULL;
 597                 CERROR("cannot open LAST_GROUP: rc = %d\n", rc);
 598                 GOTO(cleanup, rc);
 599         }
 600
 601         rc = fsfilt_read_record(obd, filp, &last_group, sizeof(__u32), &off);
 602         if (rc) {
 603                 CDEBUG(D_INODE, "error reading LAST_GROUP: rc %d\n",rc);
 604                 GOTO(cleanup, rc);
 605         }
 606         LASSERT(off == 0 || last_group >= FILTER_MIN_GROUPS);
 607         CDEBUG(D_INODE, "%s: previous %d, new %d\n",
 608                obd->obd_name, last_group, group);
 609
 610         off = 0;
 611         last_group = group;
 612         /* must be sync: bXXXX */
 613         rc = fsfilt_write_record(obd, filp, &last_group, sizeof(__u32), &off, 1);
 614         if (rc) {
 615                 CDEBUG(D_INODE, "error updating LAST_GROUP: rc %d\n", rc);
 616                 GOTO(cleanup, rc);
 617         }
 618
 619         filter->fo_committed_group = group;
 620 cleanup:
 621         if (filp)
 622                 filp_close(filp, 0);
 623         RETURN(rc);
 624 }
 625
 626 static int filter_read_group_internal(struct obd_device *obd, int group,
 627                                       int create)
 628 {
 629         struct filter_obd *filter = &obd->u.filter;
 630         __u64 *new_objids = NULL;
 631         struct filter_subdirs *new_subdirs = NULL, *tmp_subdirs = NULL;
 632         struct dentry **new_groups = NULL;
 633         struct file **new_files = NULL;
 634         struct dentry *dentry;
 635         struct file *filp;
 636         int old_count = filter->fo_group_count, rc, stage = 0, i;
 637         char name[25];
 638         __u64 last_objid;
 639         loff_t off = 0;
 640
 641         snprintf(name, 24, "%d", group);
 642         name[24] = '\0';
 643
 644         if (!create) {
 645                 dentry = ll_lookup_one_len(name, filter->fo_dentry_O,
 646                                            strlen(name));
 647                 if (IS_ERR(dentry)) {
 648                         CERROR("Cannot lookup expected object group %d: %ld\n",
 649                                group, PTR_ERR(dentry));
 650                         RETURN(PTR_ERR(dentry));
 651                 }
 652         } else {
 653                 dentry = simple_mkdir(filter->fo_dentry_O, name, 0700, 1);
 654                 if (IS_ERR(dentry)) {
 655                         CERROR("cannot lookup/create O/%s: rc = %ld\n", name,
 656                                PTR_ERR(dentry));
 657                         RETURN(PTR_ERR(dentry));
 658                 }
 659         }
 660         stage = 1;
 661
 662         snprintf(name, 24, "O/%d/LAST_ID", group);
 663         name[24] = '\0';
 664         filp = filp_open(name, O_CREAT | O_RDWR, 0700);
 665         if (IS_ERR(filp)) {
 666                 CERROR("cannot create %s: rc = %ld\n", name, PTR_ERR(filp));
 667                 GOTO(cleanup, rc = PTR_ERR(filp));
 668         }
 669         stage = 2;
 670
 671         rc = fsfilt_read_record(obd, filp, &last_objid, sizeof(__u64), &off);
 672         if (rc) {
 673                 CDEBUG(D_INODE, "error reading %s: rc %d\n", name, rc);
 674                 GOTO(cleanup, rc);
 675         }
 676
 677         if (filter->fo_subdir_count) {
 678                 OBD_ALLOC(tmp_subdirs, sizeof(*tmp_subdirs));
 679                 if (tmp_subdirs == NULL)
 680                         GOTO(cleanup, rc = -ENOMEM);
 681                 stage = 3;
 682
 683                 for (i = 0; i < filter->fo_subdir_count; i++) {
 684                         char dir[20];
 685                         snprintf(dir, sizeof(dir), "d%u", i);
 686
 687                         tmp_subdirs->dentry[i] =
 688                                 simple_mkdir(dentry, dir, 0700, 1);
 689                         if (IS_ERR(tmp_subdirs->dentry[i])) {
 690                                 rc = PTR_ERR(tmp_subdirs->dentry[i]);
 691                                 CERROR("can't lookup/create O/%d/%s: rc = %d\n",
 692                                        group, dir, rc);
 693                                 GOTO(cleanup, rc);
 694                         }
 695                         CDEBUG(D_INODE, "got/created O/%d/%s: %p\n", group, dir,
 696                                tmp_subdirs->dentry[i]);
 697                 }
 698         }
 699
 700         /* 'group' is an index; we need an array of length 'group + 1' */
 701         if (group + 1 > old_count) {
 702                 int len = group + 1;
 703                 OBD_ALLOC(new_objids, len * sizeof(*new_objids));
 704                 OBD_ALLOC(new_subdirs, len * sizeof(*new_subdirs));
 705                 OBD_ALLOC(new_groups, len * sizeof(*new_groups));
 706                 OBD_ALLOC(new_files, len * sizeof(*new_files));
 707                 stage = 4;
 708                 if (new_objids == NULL || new_subdirs == NULL ||
 709                     new_groups == NULL || new_files == NULL)
 710                         GOTO(cleanup, rc = -ENOMEM);
 711
 712                 memcpy(new_objids, filter->fo_last_objids,
 713                        old_count * sizeof(*new_objids));
 714                 memcpy(new_subdirs, filter->fo_subdirs,
 715                        old_count * sizeof(*new_subdirs));
 716                 memcpy(new_groups, filter->fo_groups,
 717                        old_count * sizeof(*new_groups));
 718                 memcpy(new_files, filter->fo_last_objid_files,
 719                        old_count * sizeof(*new_files));
 720
 721                 if (old_count) {
 722                         OBD_FREE(filter->fo_last_objids,
 723                                  old_count * sizeof(*new_objids));
 724                         OBD_FREE(filter->fo_subdirs,
 725                                  old_count * sizeof(*new_subdirs));
 726                         OBD_FREE(filter->fo_groups,
 727                                  old_count * sizeof(*new_groups));
 728                         OBD_FREE(filter->fo_last_objid_files,
 729                                  old_count * sizeof(*new_files));
 730                 }
 731                 filter->fo_last_objids = new_objids;
 732                 filter->fo_subdirs = new_subdirs;
 733                 filter->fo_groups = new_groups;
 734                 filter->fo_last_objid_files = new_files;
 735                 filter->fo_group_count = len;
 736         }
 737
 738         filter->fo_groups[group] = dentry;
 739         filter->fo_last_objid_files[group] = filp;
 740         if (filter->fo_subdir_count) {
 741                 filter->fo_subdirs[group] = *tmp_subdirs;
 742                 OBD_FREE(tmp_subdirs, sizeof(*tmp_subdirs));
 743         }
 744
 745         filter_update_last_group(obd, group);
 746
 747         if (filp->f_dentry->d_inode->i_size == 0) {
 748                 filter->fo_last_objids[group] = FILTER_INIT_OBJID;
 749                 RETURN(0);
 750         }
 751
 752         filter->fo_last_objids[group] = le64_to_cpu(last_objid);
 753         CDEBUG(D_INODE, "%s: server last_objid group %d: "LPU64"\n",
 754                obd->obd_name, group, last_objid);
 755         RETURN(0);
 756  cleanup:
 757         switch (stage) {
 758         case 4:
 759                 if (new_objids != NULL)
 760                         OBD_FREE(new_objids, group * sizeof(*new_objids));
 761                 if (new_subdirs != NULL)
 762                         OBD_FREE(new_subdirs, group * sizeof(*new_subdirs));
 763                 if (new_groups != NULL)
 764                         OBD_FREE(new_groups, group * sizeof(*new_groups));
 765                 if (new_files != NULL)
 766                         OBD_FREE(new_files, group * sizeof(*new_files));
 767         case 3:
 768                 if (filter->fo_subdir_count) {
 769                         for (i = 0; i < filter->fo_subdir_count; i++) {
 770                                 if (tmp_subdirs->dentry[i] != NULL)
 771                                         dput(tmp_subdirs->dentry[i]);
 772                         }
 773                         OBD_FREE(tmp_subdirs, sizeof(*tmp_subdirs));
 774                 }
 775         case 2:
 776                 filp_close(filp, 0);
 777         case 1:
 778                 dput(dentry);
 779         }
 780         RETURN(rc);
 781 }
 782
 783 static int filter_read_groups(struct obd_device *obd, int last_group,
 784                               int create)
 785 {
 786         struct filter_obd *filter = &obd->u.filter;
 787         int old_count, group, rc = 0;
 788
 789         down(&filter->fo_init_lock);
 790         old_count = filter->fo_group_count;
 791         for (group = old_count; group <= last_group; group++) {
 792                 if (group == 0)
 793                         continue; /* no group zero */
 794
 795                 rc = filter_read_group_internal(obd, group, create);
 796                 if (rc != 0)
 797                         break;
 798         }
 799         up(&filter->fo_init_lock);
 800         return rc;
 801 }
 802
 803 static int filter_prep_groups(struct obd_device *obd)
 804 {
 805         struct filter_obd *filter = &obd->u.filter;
 806         struct dentry *dentry, *O_dentry;
 807         int rc = 0, cleanup_phase = 0;
 808         struct file *filp = NULL;
 809         int last_group;
 810         loff_t off = 0;
 811         ENTRY;
 812
 813         O_dentry = simple_mkdir(current->fs->pwd, "O", 0700, 1);
 814         CDEBUG(D_INODE, "got/created O: %p\n", O_dentry);
 815         if (IS_ERR(O_dentry)) {
 816                 rc = PTR_ERR(O_dentry);
 817                 CERROR("cannot open/create O: rc = %d\n", rc);
 818                 GOTO(cleanup, rc);
 819         }
 820         filter->fo_dentry_O = O_dentry;
 821         cleanup_phase = 1; /* O_dentry */
 822
 823         /* Lookup "R" to tell if we're on an old OST FS and need to convert
 824          * from O/R/<dir>/<objid> to O/0/<dir>/<objid>.  This can be removed
 825          * some time post 1.0 when all old-style OSTs have converted along
 826          * with the init_objid hack. */
 827         dentry = ll_lookup_one_len("R", O_dentry, 1);
 828         if (IS_ERR(dentry))
 829                 GOTO(cleanup, rc = PTR_ERR(dentry));
 830         if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
 831                 struct dentry *O0_dentry = lookup_one_len("0", O_dentry, 1);
 832                 ENTRY;
 833
 834                 CWARN("converting OST to new object layout\n");
 835                 if (IS_ERR(O0_dentry)) {
 836                         rc = PTR_ERR(O0_dentry);
 837                         CERROR("error looking up O/0: rc %d\n", rc);
 838                         GOTO(cleanup_R, rc);
 839                 }
 840
 841                 if (O0_dentry->d_inode) {
 842                         CERROR("Both O/R and O/0 exist. Fix manually.\n");
 843                         GOTO(cleanup_O0, rc = -EEXIST);
 844                 }
 845
 846                 down(&O_dentry->d_inode->i_sem);
 847                 rc = vfs_rename(O_dentry->d_inode, dentry,
 848                                 O_dentry->d_inode, O0_dentry);
 849                 up(&O_dentry->d_inode->i_sem);
 850
 851                 if (rc) {
 852                         CERROR("error renaming O/R to O/0: rc %d\n", rc);
 853                         GOTO(cleanup_O0, rc);
 854                 }
 855                 filter->fo_fsd->fsd_feature_incompat |=
 856                         cpu_to_le32(FILTER_INCOMPAT_GROUPS);
 857                 rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
 858                                                filter->fo_fsd, 1);
 859                 GOTO(cleanup_O0, rc);
 860
 861         cleanup_O0:
 862                 f_dput(O0_dentry);
 863         cleanup_R:
 864                 f_dput(dentry);
 865                 if (rc)
 866                         GOTO(cleanup, rc);
 867         } else {
 868                 f_dput(dentry);
 869         }
 870
 871         cleanup_phase = 2; /* groups */
 872
 873         /* we have to initialize all groups before first connections from
 874          * clients because they may send create/destroy for any group -bzzz */
 875         filp = filp_open("LAST_GROUP", O_CREAT | O_RDWR, 0700);
 876         if (IS_ERR(filp)) {
 877                 CERROR("cannot create LAST_GROUP: rc = %ld\n", PTR_ERR(filp));
 878                 GOTO(cleanup, rc = PTR_ERR(filp));
 879         }
 880         cleanup_phase = 3; /* filp */
 881
 882         rc = fsfilt_read_record(obd, filp, &last_group, sizeof(__u32), &off);
 883         if (rc) {
 884                 CDEBUG(D_INODE, "error reading LAST_GROUP: rc %d\n", rc);
 885                 GOTO(cleanup, rc);
 886         }
 887         if (off == 0) {
 888                 last_group = FILTER_MIN_GROUPS;
 889         } else {
 890                 LASSERT(last_group >= FILTER_MIN_GROUPS);
 891         }
 892
 893         CWARN("%s: initialize groups [%d,%d]\n", obd->obd_name,
 894               FILTER_MIN_GROUPS, last_group);
 895         filter->fo_committed_group = last_group;
 896         rc = filter_read_groups(obd, last_group, 1);
 897         if (rc)
 898                 GOTO(cleanup, rc);
 899
 900         filp_close(filp, 0);
 901         RETURN(0);
 902
 903  cleanup:
 904         switch (cleanup_phase) {
 905         case 3:
 906                 filp_close(filp, 0);
 907         case 2:
 908                 filter_cleanup_groups(obd);
 909         case 1:
 910                 f_dput(filter->fo_dentry_O);
 911                 filter->fo_dentry_O = NULL;
 912         default:
 913                 break;
 914         }
 915         return rc;
 916 }
 917
 918 /* setup the object store with correct subdirectories */
 919 static int filter_prep(struct obd_device *obd)
 920 {
 921         struct lvfs_run_ctxt saved;
 922         struct filter_obd *filter = &obd->u.filter;
 923         struct file *file;
 924         struct inode *inode;
 925         int rc = 0;
 926         ENTRY;
 927
 928         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 929         file = filp_open(LAST_RCVD, O_RDWR | O_CREAT | O_LARGEFILE, 0700);
 930         if (!file || IS_ERR(file)) {
 931                 rc = PTR_ERR(file);
 932                 CERROR("OBD filter: cannot open/create %s: rc = %d\n",
 933                        LAST_RCVD, rc);
 934                 GOTO(out, rc);
 935         }
 936
 937         if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
 938                 CERROR("%s is not a regular file!: mode = %o\n", LAST_RCVD,
 939                        file->f_dentry->d_inode->i_mode);
 940                 GOTO(err_filp, rc = -ENOENT);
 941         }
 942
 943         /* steal operations */
 944         inode = file->f_dentry->d_inode;
 945         filter->fo_fop = file->f_op;
 946         filter->fo_iop = inode->i_op;
 947         filter->fo_aops = inode->i_mapping->a_ops;
 948
 949         rc = filter_init_server_data(obd, file);
 950         if (rc) {
 951                 CERROR("cannot read %s: rc = %d\n", LAST_RCVD, rc);
 952                 GOTO(err_filp, rc);
 953         }
 954         filter->fo_rcvd_filp = file;
 955
 956         rc = filter_prep_groups(obd);
 957         if (rc)
 958                 GOTO(err_server_data, rc);
 959
 960  out:
 961         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 962
 963         return(rc);
 964
 965  err_server_data:
 966         //class_disconnect_exports(obd, 0);
 967         filter_free_server_data(filter);
 968  err_filp:
 969         if (filp_close(file, 0))
 970                 CERROR("can't close %s after error\n", LAST_RCVD);
 971         filter->fo_rcvd_filp = NULL;
 972         goto out;
 973 }
 974
 975 /* cleanup the filter: write last used object id to status file */
 976 static void filter_post(struct obd_device *obd)
 977 {
 978         struct lvfs_run_ctxt saved;
 979         struct filter_obd *filter = &obd->u.filter;
 980         int rc, i;
 981
 982         /* XXX: filter_update_lastobjid used to call fsync_dev.  It might be
 983          * best to start a transaction with h_sync, because we removed this
 984          * from lastobjid */
 985
 986         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 987         rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
 988                                        filter->fo_fsd, 0);
 989         if (rc)
 990                 CERROR("error writing server data: rc = %d\n", rc);
 991
 992         for (i = 1; i < filter->fo_group_count; i++) {
 993                 rc = filter_update_last_objid(obd, i,
 994                                              (i == filter->fo_group_count - 1));
 995                 if (rc)
 996                         CERROR("error writing group %d lastobjid: rc = %d\n",
 997                                i, rc);
 998         }
 999
1000         rc = filp_close(filter->fo_rcvd_filp, 0);
1001         filter->fo_rcvd_filp = NULL;
1002         if (rc)
1003                 CERROR("error closing %s: rc = %d\n", LAST_RCVD, rc);
1004
1005         filter_cleanup_groups(obd);
1006         filter_free_server_data(filter);
1007         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1008 }
1009
1010 static void filter_set_last_id(struct filter_obd *filter, int group, obd_id id)
1011 {
1012         LASSERT(filter->fo_fsd != NULL);
1013         LASSERT(group > 0);
1014         LASSERT(group < filter->fo_group_count);
1015
1016         spin_lock(&filter->fo_objidlock);
1017         filter->fo_last_objids[group] = id;
1018         spin_unlock(&filter->fo_objidlock);
1019 }
1020
1021 __u64 filter_last_id(struct filter_obd *filter, int group)
1022 {
1023         obd_id id;
1024         LASSERT(filter->fo_fsd != NULL);
1025         LASSERT(group > 0);
1026         LASSERT(group < filter->fo_group_count);
1027
1028         spin_lock(&filter->fo_objidlock);
1029         id = filter->fo_last_objids[group];
1030         spin_unlock(&filter->fo_objidlock);
1031
1032         return id;
1033 }
1034
1035 /* direct cut-n-paste of mds_blocking_ast() */
1036 static int filter_blocking_ast(struct ldlm_lock *lock,
1037                                struct ldlm_lock_desc *desc,
1038                                void *data, int flag)
1039 {
1040         int do_ast;
1041         ENTRY;
1042
1043         if (flag == LDLM_CB_CANCELING) {
1044                 /* Don't need to do anything here. */
1045                 RETURN(0);
1046         }
1047
1048         /* XXX layering violation!  -phil */
1049         l_lock(&lock->l_resource->lr_namespace->ns_lock);
1050         /* Get this: if filter_blocking_ast is racing with ldlm_intent_policy,
1051          * such that filter_blocking_ast is called just before l_i_p takes the
1052          * ns_lock, then by the time we get the lock, we might not be the
1053          * correct blocking function anymore.  So check, and return early, if
1054          * so. */
1055         if (lock->l_blocking_ast != filter_blocking_ast) {
1056                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
1057                 RETURN(0);
1058         }
1059
1060         lock->l_flags |= LDLM_FL_CBPENDING;
1061         do_ast = (!lock->l_readers && !lock->l_writers);
1062         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
1063
1064         if (do_ast) {
1065                 struct lustre_handle lockh;
1066                 int rc;
1067
1068                 LDLM_DEBUG(lock, "already unused, calling ldlm_cli_cancel");
1069                 ldlm_lock2handle(lock, &lockh);
1070                 rc = ldlm_cli_cancel(&lockh);
1071                 if (rc < 0)
1072                         CERROR("ldlm_cli_cancel: %d\n", rc);
1073         } else {
1074                 LDLM_DEBUG(lock, "Lock still has references, will be "
1075                            "cancelled later");
1076         }
1077         RETURN(0);
1078 }
1079
1080 extern void *lock_dir(struct inode *dir, struct qstr *name);
1081 extern void unlock_dir(struct inode *dir, void *lock);
1082
1083 static void * filter_lock_dentry(struct obd_device *obd,
1084                                  struct dentry *dparent, obd_id id)
1085 {
1086 #ifdef S_PDIROPS
1087         struct qstr qstr;
1088         char name[32];
1089         qstr.name = name;
1090         qstr.len = sprintf(name, LPU64, id);
1091         return lock_dir(dparent->d_inode, &qstr);
1092 #else
1093         down(&dparent->d_inode->i_sem);
1094 #endif
1095         return 0;
1096 }
1097
1098 /* We never dget the object parent, so DON'T dput it either */
1099 static void filter_parent_unlock(struct dentry *dparent, void *lock)
1100 {
1101 #ifdef S_PDIROPS
1102         LASSERT(lock != NULL);
1103         unlock_dir(dparent->d_inode, lock);
1104 #else
1105         up(&dparent->d_inode->i_sem);
1106 #endif
1107 }
1108
1109 /* We never dget the object parent, so DON'T dput it either */
1110 struct dentry *filter_parent(struct obd_device *obd, obd_gr group, obd_id objid)
1111 {
1112         struct filter_obd *filter = &obd->u.filter;
1113         LASSERT(group < filter->fo_group_count);
1114         LASSERT(group > 0);
1115
1116         if (filter->fo_subdir_count == 0)
1117                 return filter->fo_groups[group];
1118
1119         return filter->fo_subdirs[group].dentry[objid & (filter->fo_subdir_count - 1)];
1120 }
1121
1122 /* We never dget the object parent, so DON'T dput it either */
1123 struct dentry *filter_parent_lock(struct obd_device *obd, obd_gr group,
1124                                   obd_id objid, void **lock)
1125 {
1126         unsigned long now = jiffies;
1127         struct dentry *dparent = filter_parent(obd, group, objid);
1128
1129         if (IS_ERR(dparent))
1130                 return dparent;
1131
1132         LASSERT(dparent);
1133         LASSERT(dparent->d_inode);
1134
1135         *lock = filter_lock_dentry(obd, dparent, objid);
1136         if (time_after(jiffies, now + 15 * HZ))
1137                 CERROR("slow parent lock %lus\n", (jiffies - now) / HZ);
1138         return dparent;
1139 }
1140
1141 /* How to get files, dentries, inodes from object id's.
1142  *
1143  * If dir_dentry is passed, the caller has already locked the parent
1144  * appropriately for this operation (normally a write lock).  If
1145  * dir_dentry is NULL, we do a read lock while we do the lookup to
1146  * avoid races with create/destroy and such changing the directory
1147  * internal to the filesystem code. */
1148 struct dentry *filter_fid2dentry(struct obd_device *obd,
1149                                  struct dentry *dir_dentry,
1150                                  obd_gr group, obd_id id)
1151 {
1152         struct dentry *dparent = dir_dentry;
1153         struct dentry *dchild;
1154         void *lock = NULL;
1155         char name[32];
1156         int len;
1157         ENTRY;
1158
1159         if (id == 0) {
1160                 CERROR("fatal: invalid object id 0\n");
1161                 RETURN(ERR_PTR(-ESTALE));
1162         }
1163
1164         len = sprintf(name, LPU64, id);
1165         if (dir_dentry == NULL) {
1166                 dparent = filter_parent_lock(obd, group, id, &lock);
1167                 if (IS_ERR(dparent))
1168                         RETURN(dparent);
1169         }
1170         CDEBUG(D_INODE, "looking up object O/%*s/%s\n",
1171                dparent->d_name.len, dparent->d_name.name, name);
1172         dchild = /*ll_*/lookup_one_len(name, dparent, len);
1173         if (dir_dentry == NULL)
1174                 filter_parent_unlock(dparent, lock);
1175         if (IS_ERR(dchild)) {
1176                 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
1177                 RETURN(dchild);
1178         }
1179
1180         CDEBUG(D_INODE, "got child objid %s: %p, count = %d\n",
1181                name, dchild, atomic_read(&dchild->d_count));
1182
1183         LASSERT(atomic_read(&dchild->d_count) > 0);
1184
1185         RETURN(dchild);
1186 }
1187
1188 static int filter_prepare_destroy(struct obd_device *obd, obd_id objid,
1189                                   obd_id group)
1190 {
1191         struct lustre_handle lockh;
1192         int flags = LDLM_AST_DISCARD_DATA, rc;
1193         struct ldlm_res_id res_id = { .name = { objid, 0, group, 0 } };
1194         ldlm_policy_data_t policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
1195
1196         ENTRY;
1197         /* Tell the clients that the object is gone now and that they should
1198          * throw away any cached pages. */
1199         rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, res_id,
1200                               LDLM_EXTENT, &policy, LCK_PW,
1201                               &flags, filter_blocking_ast, ldlm_completion_ast,
1202                               NULL, NULL, NULL, 0, NULL, &lockh);
1203
1204         /* We only care about the side-effects, just drop the lock. */
1205         if (rc == ELDLM_OK)
1206                 ldlm_lock_decref(&lockh, LCK_PW);
1207
1208         RETURN(rc);
1209 }
1210
1211 /* Caller must hold LCK_PW on parent and push us into kernel context.
1212  * Caller is also required to ensure that dchild->d_inode exists. */
1213 static int filter_destroy_internal(struct obd_device *obd, obd_id objid,
1214                                    struct dentry *dparent,
1215                                    struct dentry *dchild)
1216 {
1217         struct inode *inode = dchild->d_inode;
1218         int rc;
1219         ENTRY;
1220
1221         if (inode->i_nlink != 1 || atomic_read(&inode->i_count) != 1) {
1222                 CERROR("destroying objid %*s nlink = %lu, count = %d\n",
1223                        dchild->d_name.len, dchild->d_name.name,
1224                        (unsigned long)inode->i_nlink,
1225                        atomic_read(&inode->i_count));
1226         }
1227
1228         rc = vfs_unlink(dparent->d_inode, dchild);
1229
1230         if (rc)
1231                 CERROR("error unlinking objid %*s: rc %d\n",
1232                        dchild->d_name.len, dchild->d_name.name, rc);
1233
1234         RETURN(rc);
1235 }
1236
1237 static int filter_intent_policy(struct ldlm_namespace *ns,
1238                                 struct ldlm_lock **lockp, void *req_cookie,
1239                                 ldlm_mode_t mode, int flags, void *data)
1240 {
1241         struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
1242         struct ptlrpc_request *req = req_cookie;
1243         struct ldlm_lock *lock = *lockp, *l = NULL;
1244         struct ldlm_resource *res = lock->l_resource;
1245         ldlm_processing_policy policy;
1246         struct ost_lvb *res_lvb, *reply_lvb;
1247         struct list_head *tmp;
1248         ldlm_error_t err;
1249         int tmpflags = 0, rc, repsize[2] = {sizeof(struct ldlm_reply),
1250                                             sizeof(struct ost_lvb) };
1251         ENTRY;
1252
1253         policy = ldlm_get_processing_policy(res);
1254         LASSERT(policy != NULL);
1255         LASSERT(req != NULL);
1256
1257         rc = lustre_pack_reply(req, 2, repsize, NULL);
1258         if (rc)
1259                 RETURN(req->rq_status = rc);
1260
1261         reply_lvb = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*reply_lvb));
1262         LASSERT(reply_lvb != NULL);
1263
1264         //fixup_handle_for_resent_req(req, lock, &lockh);
1265
1266         /* If we grant any lock at all, it will be a whole-file read lock.
1267          * Call the extent policy function to see if our request can be
1268          * granted, or is blocked. */
1269         lock->l_policy_data.l_extent.start = 0;
1270         lock->l_policy_data.l_extent.end = OBD_OBJECT_EOF;
1271         lock->l_req_mode = LCK_PR;
1272
1273         l_lock(&res->lr_namespace->ns_lock);
1274
1275         res->lr_tmp = &rpc_list;
1276         rc = policy(lock, &tmpflags, 0, &err);
1277         res->lr_tmp = NULL;
1278
1279         /* FIXME: we should change the policy function slightly, to not make
1280          * this list at all, since we just turn around and free it */
1281         while (!list_empty(&rpc_list)) {
1282                 struct ldlm_ast_work *w =
1283                         list_entry(rpc_list.next, struct ldlm_ast_work, w_list);
1284                 list_del(&w->w_list);
1285                 LDLM_LOCK_PUT(w->w_lock);
1286                 OBD_FREE(w, sizeof(*w));
1287         }
1288
1289         if (rc == LDLM_ITER_CONTINUE) {
1290                 /* The lock met with no resistance; we're finished. */
1291                 l_unlock(&res->lr_namespace->ns_lock);
1292                 RETURN(ELDLM_LOCK_REPLACED);
1293         }
1294
1295         /* Do not grant any lock, but instead send GL callbacks.  The extent
1296          * policy nicely created a list of all PW locks for us.  We will choose
1297          * the highest of those which are larger than the size in the LVB, if
1298          * any, and perform a glimpse callback. */
1299         down(&res->lr_lvb_sem);
1300         res_lvb = res->lr_lvb_data;
1301         LASSERT(res_lvb != NULL);
1302         reply_lvb->lvb_size = res_lvb->lvb_size;
1303         reply_lvb->lvb_blocks = res_lvb->lvb_blocks;
1304         up(&res->lr_lvb_sem);
1305
1306         list_for_each(tmp, &res->lr_granted) {
1307                 struct ldlm_lock *tmplock =
1308                         list_entry(tmp, struct ldlm_lock, l_res_link);
1309
1310                 if (tmplock->l_granted_mode == LCK_PR)
1311                         continue;
1312
1313                 if (tmplock->l_policy_data.l_extent.end <= reply_lvb->lvb_size)
1314                         continue;
1315
1316                 if (l == NULL) {
1317                         l = LDLM_LOCK_GET(tmplock);
1318                         continue;
1319                 }
1320
1321                 if (l->l_policy_data.l_extent.start >
1322                     tmplock->l_policy_data.l_extent.start)
1323                         continue;
1324
1325                 LDLM_LOCK_PUT(l);
1326                 l = LDLM_LOCK_GET(tmplock);
1327         }
1328         l_unlock(&res->lr_namespace->ns_lock);
1329
1330         /* There were no PW locks beyond the size in the LVB; finished. */
1331         if (l == NULL)
1332                 RETURN(ELDLM_LOCK_ABORTED);
1333
1334         LASSERT(l->l_glimpse_ast != NULL);
1335         rc = l->l_glimpse_ast(l, NULL); /* this will update the LVB */
1336         if (rc != 0 && res->lr_namespace->ns_lvbo &&
1337             res->lr_namespace->ns_lvbo->lvbo_update) {
1338                 res->lr_namespace->ns_lvbo->lvbo_update(res, NULL, 0, 1);
1339         }
1340
1341         down(&res->lr_lvb_sem);
1342         reply_lvb->lvb_size = res_lvb->lvb_size;
1343         reply_lvb->lvb_blocks = res_lvb->lvb_blocks;
1344         up(&res->lr_lvb_sem);
1345
1346         LDLM_LOCK_PUT(l);
1347
1348         RETURN(ELDLM_LOCK_ABORTED);
1349 }
1350
1351 static int filter_post_fs_cleanup(struct obd_device *obd)
1352 {
1353         int rc = 0;
1354
1355         rc = fsfilt_post_cleanup(obd);
1356
1357         RETURN(rc);
1358 }
1359
1360 static int filter_group_set_fs_flags(struct obd_device *obd, int group)
1361 {
1362         struct filter_obd *filter = &obd->u.filter;
1363         int rc = 0, i = 0;
1364         ENTRY;
1365
1366         /* zero group is not longer valid. */
1367         if (group== 0)
1368                 RETURN(rc);
1369         for (i = 0; i < filter->fo_subdir_count; i++) {
1370                 struct dentry *dentry;
1371                 dentry = (filter->fo_subdirs + group)->dentry[i];
1372                 rc = fsfilt_set_fs_flags(obd, dentry->d_inode,
1373                                          SM_DO_REC | SM_DO_COW);
1374                 if (rc)
1375                         RETURN(rc);
1376         }
1377         RETURN(rc);
1378 }
1379 static int filter_post_fs_setup(struct obd_device *obd)
1380 {
1381         struct filter_obd *filter = &obd->u.filter;
1382         int rc = 0, j = 0;
1383         struct llog_ctxt *ctxt = NULL;
1384
1385         rc = fsfilt_post_setup(obd, filter->fo_dentry_O);
1386         if (rc)
1387                 RETURN(rc);
1388
1389         for (j = 0; j < filter->fo_group_count; j++) {
1390                 rc = filter_group_set_fs_flags(obd, j);
1391                 if (rc)
1392                         return rc;
1393         }
1394
1395         fsfilt_get_reint_log_ctxt(obd, filter->fo_sb, &ctxt);
1396         if (ctxt) {
1397                 ctxt->loc_obd = obd;
1398                 ctxt->loc_idx = LLOG_REINT_ORIG_CTXT;
1399                 obd->obd_llog_ctxt[LLOG_REINT_ORIG_CTXT] = ctxt;
1400         }
1401         fsfilt_set_ost_flags(obd, filter->fo_sb);
1402         return rc;
1403 }
1404
1405 /* mount the file system (secretly) */
1406 int filter_common_setup(struct obd_device *obd, obd_count len,
1407                         void *buf, char *option)
1408 {
1409         struct lustre_cfg* lcfg = buf;
1410         struct filter_obd *filter = &obd->u.filter;
1411         struct vfsmount *mnt;
1412         char name[32] = "CATLIST";
1413         int rc = 0, i;
1414         ENTRY;
1415
1416         dev_clear_rdonly(2);
1417
1418         if (!lcfg->lcfg_inlbuf1 || !lcfg->lcfg_inlbuf2)
1419                 RETURN(-EINVAL);
1420
1421         obd->obd_fsops = fsfilt_get_ops(lcfg->lcfg_inlbuf2);
1422         if (IS_ERR(obd->obd_fsops))
1423                 RETURN(PTR_ERR(obd->obd_fsops));
1424
1425         mnt = do_kern_mount(lcfg->lcfg_inlbuf2, MS_NOATIME | MS_NODIRATIME,
1426                             lcfg->lcfg_inlbuf1, option);
1427         rc = PTR_ERR(mnt);
1428         if (IS_ERR(mnt))
1429                 GOTO(err_ops, rc);
1430
1431         if (lcfg->lcfg_inllen3 > 0 && lcfg->lcfg_inlbuf3) {
1432                 if (*lcfg->lcfg_inlbuf3 == 'f') {
1433                         obd->obd_replayable = 1;
1434                         obd_sync_filter = 1;
1435                         CWARN("%s: recovery enabled\n", obd->obd_name);
1436                 } else {
1437                         if (*lcfg->lcfg_inlbuf3 != 'n') {
1438                                 CERROR("unrecognised flag '%c'\n",
1439                                        *lcfg->lcfg_inlbuf3);
1440                         }
1441                         // XXX Robert? Why do we get errors here
1442                         // GOTO(err_mntput, rc = -EINVAL);
1443                 }
1444         }
1445
1446         filter->fo_vfsmnt = mnt;
1447         filter->fo_sb = mnt->mnt_sb;
1448         filter->fo_fstype = mnt->mnt_sb->s_type->name;
1449         CDEBUG(D_SUPER, "%s: mnt = %p\n", filter->fo_fstype, mnt);
1450
1451         OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1452         obd->obd_lvfs_ctxt.pwdmnt = mnt;
1453         obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
1454         obd->obd_lvfs_ctxt.fs = get_ds();
1455         obd->obd_lvfs_ctxt.cb_ops = filter_lvfs_ops;
1456
1457         rc = fsfilt_setup(obd, mnt->mnt_sb);
1458         if (rc)
1459                 GOTO(err_mntput, rc);
1460
1461         sema_init(&filter->fo_init_lock, 1);
1462         filter->fo_committed_group = 0;
1463         rc = filter_prep(obd);
1464         if (rc)
1465                 GOTO(err_mntput, rc);
1466
1467         filter->fo_destroys_in_progress = 0;
1468         for (i = 0; i < 32; i++)
1469                 sema_init(&filter->fo_create_locks[i], 1);
1470
1471         spin_lock_init(&filter->fo_translock);
1472         spin_lock_init(&filter->fo_objidlock);
1473         INIT_LIST_HEAD(&filter->fo_export_list);
1474         sema_init(&filter->fo_alloc_lock, 1);
1475         spin_lock_init(&filter->fo_r_pages.oh_lock);
1476         spin_lock_init(&filter->fo_w_pages.oh_lock);
1477         spin_lock_init(&filter->fo_r_discont_pages.oh_lock);
1478         spin_lock_init(&filter->fo_w_discont_pages.oh_lock);
1479         spin_lock_init(&filter->fo_r_discont_blocks.oh_lock);
1480         spin_lock_init(&filter->fo_w_discont_blocks.oh_lock);
1481         filter->fo_readcache_max_filesize = FILTER_MAX_CACHE_SIZE;
1482
1483         INIT_LIST_HEAD(&filter->fo_llog_list);
1484         spin_lock_init(&filter->fo_llog_list_lock);
1485
1486         obd->obd_namespace = ldlm_namespace_new("filter-tgt",
1487                                                 LDLM_NAMESPACE_SERVER);
1488         if (obd->obd_namespace == NULL)
1489                 GOTO(err_post, rc = -ENOMEM);
1490         obd->obd_namespace->ns_lvbp = obd;
1491         obd->obd_namespace->ns_lvbo = &filter_lvbo;
1492         ldlm_register_intent(obd->obd_namespace, filter_intent_policy);
1493
1494         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1495                            "filter_ldlm_cb_client", &obd->obd_ldlm_client);
1496
1497         rc = obd_llog_cat_initialize(obd, &obd->obd_llogs, 1, name);
1498         if (rc) {
1499                 CERROR("failed to setup llogging subsystems\n");
1500                 GOTO(err_post, rc);
1501         }
1502         RETURN(0);
1503
1504 err_post:
1505         filter_post(obd);
1506 err_mntput:
1507         unlock_kernel();
1508         mntput(mnt);
1509         filter->fo_sb = 0;
1510         lock_kernel();
1511 err_ops:
1512         fsfilt_put_ops(obd->obd_fsops);
1513         return rc;
1514 }
1515
1516 static int filter_attach(struct obd_device *obd, obd_count len, void *data)
1517 {
1518         struct lprocfs_static_vars lvars;
1519         int rc;
1520
1521         lprocfs_init_vars(filter, &lvars);
1522         rc = lprocfs_obd_attach(obd, lvars.obd_vars);
1523         if (rc != 0)
1524                 return rc;
1525
1526         rc = lprocfs_alloc_obd_stats(obd, LPROC_FILTER_LAST);
1527         if (rc != 0)
1528                 return rc;
1529
1530         /* Init obdfilter private stats here */
1531         lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_READ_BYTES,
1532                              LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes");
1533         lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_WRITE_BYTES,
1534                              LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes");
1535
1536         return lproc_filter_attach_seqstat(obd);
1537 }
1538
1539 static int filter_detach(struct obd_device *dev)
1540 {
1541         lprocfs_free_obd_stats(dev);
1542         return lprocfs_obd_detach(dev);
1543 }
1544
1545 static int filter_setup(struct obd_device *obd, obd_count len, void *buf)
1546 {
1547         struct lustre_cfg* lcfg = buf;
1548         int rc;
1549         ENTRY;
1550         /* all mount options including errors=remount-ro and asyncdel are passed
1551          * using 4th lcfg param. And it is good, finally we have got rid of
1552          * hardcoded fs types in the code. */
1553         rc = filter_common_setup(obd, len, buf, lcfg->lcfg_inlbuf4);
1554         if (rc)
1555                 RETURN(rc);
1556         rc = filter_post_fs_setup(obd);
1557         RETURN(rc);
1558 }
1559
1560 static int filter_cleanup(struct obd_device *obd, int flags)
1561 {
1562         struct filter_obd *filter = &obd->u.filter;
1563         ENTRY;
1564
1565         if (flags & OBD_OPT_FAILOVER)
1566                 CERROR("%s: shutting down for failover; client state will"
1567                        " be preserved.\n", obd->obd_name);
1568
1569         if (!list_empty(&obd->obd_exports)) {
1570                 CERROR("%s: still has clients!\n", obd->obd_name);
1571                 class_disconnect_exports(obd, flags);
1572                 if (!list_empty(&obd->obd_exports)) {
1573                         CERROR("still has exports after forced cleanup?\n");
1574                         RETURN(-EBUSY);
1575                 }
1576         }
1577
1578         ldlm_namespace_free(obd->obd_namespace, flags & OBD_OPT_FORCE);
1579
1580         if (filter->fo_sb == NULL)
1581                 RETURN(0);
1582
1583         filter_post_fs_cleanup(obd);
1584         filter_post(obd);
1585
1586         shrink_dcache_parent(filter->fo_sb->s_root);
1587         filter->fo_sb = 0;
1588
1589         if (atomic_read(&filter->fo_vfsmnt->mnt_count) > 1)
1590                 CERROR("%s: mount point %p busy, mnt_count: %d\n",
1591                        obd->obd_name, filter->fo_vfsmnt,
1592                        atomic_read(&filter->fo_vfsmnt->mnt_count));
1593
1594         unlock_kernel();
1595         mntput(filter->fo_vfsmnt);
1596         //destroy_buffers(filter->fo_sb->s_dev);
1597         filter->fo_sb = NULL;
1598         fsfilt_put_ops(obd->obd_fsops);
1599         lock_kernel();
1600
1601         dev_clear_rdonly(2);
1602
1603         RETURN(0);
1604 }
1605
1606 static int filter_connect_post(struct obd_export *exp)
1607 {
1608         struct obd_device *obd = exp->exp_obd;
1609         struct filter_export_data *fed;
1610         char str[PTL_NALFMT_SIZE];
1611         struct obd_llogs *llog;
1612         struct llog_ctxt *ctxt;
1613         int rc;
1614         ENTRY;
1615
1616         fed = &exp->exp_filter_data;
1617         if (fed->fed_group < FILTER_MIN_GROUPS)
1618                 RETURN(0);
1619
1620         /* initialize llogs for connections from MDS */
1621         llog = filter_grab_llog_for_group(obd, fed->fed_group, exp);
1622         LASSERT(llog != NULL);
1623
1624         ctxt = llog_get_context(llog, LLOG_UNLINK_REPL_CTXT);
1625         LASSERT(ctxt != NULL);
1626
1627         rc = llog_receptor_accept(ctxt, exp->exp_imp_reverse);
1628         portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
1629                         exp->exp_connection->c_peer.peer_nid, str);
1630         CDEBUG(D_OTHER, "%s: init llog ctxt for export "LPX64"/%s, group %d\n",
1631                obd->obd_name, exp->exp_connection->c_peer.peer_nid,
1632                str, fed->fed_group);
1633
1634         RETURN(rc);
1635 }
1636
1637 /* nearly identical to mds_connect */
1638 static int filter_connect(struct lustre_handle *conn, struct obd_device *obd,
1639                           struct obd_uuid *cluuid, unsigned long connect_flags)
1640 {
1641         struct obd_export *exp;
1642         struct filter_export_data *fed;
1643         struct filter_client_data *fcd = NULL;
1644         struct filter_obd *filter = &obd->u.filter;
1645         struct lvfs_run_ctxt saved;
1646         int rc, group;
1647         ENTRY;
1648
1649         if (conn == NULL || obd == NULL || cluuid == NULL)
1650                 RETURN(-EINVAL);
1651
1652         rc = class_connect(conn, obd, cluuid);
1653         if (rc)
1654                 RETURN(rc);
1655         exp = class_conn2export(conn);
1656         LASSERT(exp != NULL);
1657
1658         fed = &exp->exp_filter_data;
1659
1660         spin_lock_init(&fed->fed_lock);
1661
1662         if (obd->obd_replayable) {
1663                 OBD_ALLOC(fcd, sizeof(*fcd));
1664                 if (!fcd) {
1665                         CERROR("filter: out of memory for client data\n");
1666                         GOTO(cleanup, rc = -ENOMEM);
1667                 }
1668
1669                 memcpy(fcd->fcd_uuid, cluuid, sizeof(fcd->fcd_uuid));
1670                 fed->fed_fcd = fcd;
1671
1672                 rc = filter_client_add(obd, filter, fed, -1);
1673                 if (rc)
1674                         GOTO(cleanup, rc);
1675         }
1676
1677         if (connect_flags == 0)
1678                 GOTO(cleanup, rc);
1679
1680         /* connection from MDS */
1681         group = connect_flags;
1682         CWARN("%s: Received MDS connection ("LPX64"); group %d\n",
1683               obd->obd_name, exp->exp_handle.h_cookie, group);
1684
1685         if (fed->fed_group != 0 && fed->fed_group != group) {
1686                 char str[PTL_NALFMT_SIZE];
1687                 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
1688                                 exp->exp_connection->c_peer.peer_nid, str);
1689                 CERROR("!!! This export (nid "LPX64"/%s) used object group %d "
1690                        "earlier; now it's trying to use group %d!  This could "
1691                        "be a bug in the MDS.  Tell CFS.\n",
1692                        exp->exp_connection->c_peer.peer_nid, str,
1693                        fed->fed_group, group);
1694                 GOTO(cleanup, rc = -EPROTO);
1695         }
1696         fed->fed_group = group;
1697
1698         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1699         rc = filter_read_groups(obd, group, 1);
1700         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1701         if (rc != 0) {
1702                 CERROR("can't read group %u\n", group);
1703                 GOTO(cleanup, rc);
1704         }
1705         rc = filter_group_set_fs_flags(obd, group);
1706          if (rc != 0) {
1707                 CERROR("can't set kml flags %u\n", group);
1708                 GOTO(cleanup, rc);
1709         }
1710
1711 cleanup:
1712         if (rc) {
1713                 if (fcd)
1714                         OBD_FREE(fcd, sizeof(*fcd));
1715                 class_disconnect(exp, 0);
1716         } else {
1717                 class_export_put(exp);
1718         }
1719         return rc;
1720 }
1721
1722 static int filter_precleanup(struct obd_device *obd, int flags)
1723 {
1724         struct filter_group_llog *log;
1725         struct filter_obd *filter;
1726         int rc = 0;
1727         ENTRY;
1728
1729         filter = &obd->u.filter;
1730
1731         spin_lock(&filter->fo_llog_list_lock);
1732         while (!list_empty(&filter->fo_llog_list)) {
1733                 log = list_entry(filter->fo_llog_list.next,
1734                                  struct filter_group_llog, list);
1735                 list_del(&log->list);
1736                 spin_unlock(&filter->fo_llog_list_lock);
1737
1738                 rc = obd_llog_finish(obd, log->llogs, 0);
1739                 if (rc)
1740                         CERROR("failed to cleanup llogging subsystem for %u\n",
1741                                 log->group);
1742                 OBD_FREE(log->llogs, sizeof(*(log->llogs)));
1743                 OBD_FREE(log, sizeof(*log));
1744                 spin_lock(&filter->fo_llog_list_lock);
1745         }
1746         spin_unlock(&filter->fo_llog_list_lock);
1747
1748         rc = obd_llog_finish(obd, &obd->obd_llogs, 0);
1749         if (rc)
1750                 CERROR("failed to cleanup llogging subsystem\n");
1751
1752         RETURN(rc);
1753 }
1754
1755 /* Do extra sanity checks for grant accounting.  We do this at connect,
1756  * disconnect, and statfs RPC time, so it shouldn't be too bad.  We can
1757  * always get rid of it or turn it off when we know accounting is good. */
1758 static void filter_grant_sanity_check(struct obd_device *obd, const char *func)
1759 {
1760         struct filter_export_data *fed;
1761         struct obd_export *exp;
1762         obd_size maxsize = obd->obd_osfs.os_blocks * obd->obd_osfs.os_bsize;
1763         obd_size tot_dirty = 0, tot_pending = 0, tot_granted = 0;
1764         obd_size fo_tot_dirty, fo_tot_pending, fo_tot_granted;
1765
1766         if (list_empty(&obd->obd_exports))
1767                 return;
1768
1769         spin_lock(&obd->obd_osfs_lock);
1770         spin_lock(&obd->obd_dev_lock);
1771         list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
1772                 fed = &exp->exp_filter_data;
1773                 LASSERTF(fed->fed_grant + fed->fed_pending <= maxsize,
1774                          "cli %s/%p %lu+%lu > "LPU64"\n",
1775                          exp->exp_client_uuid.uuid, exp,
1776                          fed->fed_grant, fed->fed_pending, maxsize);
1777                 LASSERTF(fed->fed_dirty <= maxsize, "cli %s/%p %lu > "LPU64"\n",
1778                          exp->exp_client_uuid.uuid, exp,fed->fed_dirty,maxsize);
1779                 CDEBUG(D_CACHE,"%s: cli %s/%p dirty %lu pend %lu grant %lu\n",
1780                        obd->obd_name, exp->exp_client_uuid.uuid, exp,
1781                        fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1782                 tot_granted += fed->fed_grant + fed->fed_pending;
1783                 tot_pending += fed->fed_pending;
1784                 tot_dirty += fed->fed_dirty;
1785         }
1786         fo_tot_granted = obd->u.filter.fo_tot_granted;
1787         fo_tot_pending = obd->u.filter.fo_tot_pending;
1788         fo_tot_dirty = obd->u.filter.fo_tot_dirty;
1789         spin_unlock(&obd->obd_dev_lock);
1790         spin_unlock(&obd->obd_osfs_lock);
1791
1792         /* Do these assertions outside the spinlocks so we don't kill system */
1793         if (tot_granted != fo_tot_granted)
1794                 CERROR("%s: tot_granted "LPU64" != fo_tot_granted "LPU64"\n",
1795                        func, tot_granted, fo_tot_granted);
1796         if (tot_pending != fo_tot_pending)
1797                 CERROR("%s: tot_pending "LPU64" != fo_tot_pending "LPU64"\n",
1798                        func, tot_pending, fo_tot_pending);
1799         if (tot_dirty != fo_tot_dirty)
1800                 CERROR("%s: tot_dirty "LPU64" != fo_tot_dirty "LPU64"\n",
1801                        func, tot_dirty, fo_tot_dirty);
1802         if (tot_pending > tot_granted)
1803                 CERROR("%s: tot_pending "LPU64" > tot_granted "LPU64"\n",
1804                        func, tot_pending, tot_granted);
1805         if (tot_granted > maxsize)
1806                 CERROR("%s: tot_granted "LPU64" > maxsize "LPU64"\n",
1807                        func, tot_granted, maxsize);
1808         if (tot_dirty > maxsize)
1809                 CERROR("%s: tot_dirty "LPU64" > maxsize "LPU64"\n",
1810                        func, tot_dirty, maxsize);
1811 }
1812
1813 /* Remove this client from the grant accounting totals.  We also remove
1814  * the export from the obd device under the osfs and dev locks to ensure
1815  * that the filter_grant_sanity_check() calculations are always valid.
1816  * The client should do something similar when it invalidates its import. */
1817 static void filter_grant_discard(struct obd_export *exp)
1818 {
1819         struct obd_device *obd = exp->exp_obd;
1820         struct filter_obd *filter = &obd->u.filter;
1821         struct filter_export_data *fed = &exp->exp_filter_data;
1822
1823         spin_lock(&obd->obd_osfs_lock);
1824         spin_lock(&exp->exp_obd->obd_dev_lock);
1825         list_del_init(&exp->exp_obd_chain);
1826         spin_unlock(&exp->exp_obd->obd_dev_lock);
1827
1828         CDEBUG(D_CACHE, "%s: cli %s/%p dirty %lu pend %lu grant %lu\n",
1829                obd->obd_name, exp->exp_client_uuid.uuid, exp,
1830                fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1831
1832         LASSERTF(filter->fo_tot_granted >= fed->fed_grant,
1833                  "%s: tot_granted "LPU64" cli %s/%p fed_grant %lu\n",
1834                  obd->obd_name, filter->fo_tot_granted,
1835                  exp->exp_client_uuid.uuid, exp, fed->fed_grant);
1836         filter->fo_tot_granted -= fed->fed_grant;
1837         LASSERTF(exp->exp_obd->u.filter.fo_tot_pending >= fed->fed_pending,
1838                  "%s: tot_pending "LPU64" cli %s/%p fed_pending %lu\n",
1839                  obd->obd_name, filter->fo_tot_pending,
1840                  exp->exp_client_uuid.uuid, exp, fed->fed_pending);
1841         LASSERTF(filter->fo_tot_dirty >= fed->fed_dirty,
1842                  "%s: tot_dirty "LPU64" cli %s/%p fed_dirty %lu\n",
1843                  obd->obd_name, filter->fo_tot_dirty,
1844                  exp->exp_client_uuid.uuid, exp, fed->fed_dirty);
1845         filter->fo_tot_dirty -= fed->fed_dirty;
1846         fed->fed_dirty = 0;
1847         fed->fed_grant = 0;
1848
1849         spin_unlock(&obd->obd_osfs_lock);
1850 }
1851
1852 static int filter_destroy_export(struct obd_export *exp)
1853 {
1854         ENTRY;
1855
1856         if (exp->exp_filter_data.fed_pending)
1857                 CERROR("%s: cli %s/%p has %lu pending on destroyed export\n",
1858                        exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
1859                        exp, exp->exp_filter_data.fed_pending);
1860
1861         target_destroy_export(exp);
1862
1863         if (exp->exp_obd->obd_replayable)
1864                 filter_client_free(exp, exp->exp_flags);
1865
1866         filter_grant_discard(exp);
1867         if (!(exp->exp_flags & OBD_OPT_FORCE))
1868                 filter_grant_sanity_check(exp->exp_obd, __FUNCTION__);
1869
1870         RETURN(0);
1871 }
1872
1873 static void filter_sync_llogs(struct obd_device *obd, struct obd_export *dexp)
1874 {
1875         struct filter_group_llog *fglog, *nlog;
1876         struct filter_obd *filter;
1877         int worked = 0, group;
1878         struct llog_ctxt *ctxt;
1879         ENTRY;
1880
1881         filter = &obd->u.filter;
1882
1883         /* we can't sync log holding spinlock. also, we do not want to get
1884          * into livelock. so we do following: loop over MDS's exports in
1885          * group order and skip already synced llogs -bzzz */
1886         do {
1887                 /* look for group with min. number, but > worked */
1888                 fglog = NULL;
1889                 group = 1 << 30;
1890                 spin_lock(&filter->fo_llog_list_lock);
1891                 list_for_each_entry(nlog, &filter->fo_llog_list, list) {
1892
1893                         if (nlog->group <= worked) {
1894                                 /* this group is already synced */
1895                                 continue;
1896                         }
1897
1898                         if (group < nlog->group) {
1899                                 /* we have group with smaller number to sync */
1900                                 continue;
1901                         }
1902
1903                         /* store current minimal group */
1904                         fglog = nlog;
1905                         group = nlog->group;
1906                 }
1907                 spin_unlock(&filter->fo_llog_list_lock);
1908
1909                 if (fglog == NULL)
1910                         break;
1911
1912                 worked = fglog->group;
1913                 if (fglog->exp && (dexp == fglog->exp || dexp == NULL)) {
1914                         ctxt = llog_get_context(fglog->llogs,
1915                                         LLOG_UNLINK_REPL_CTXT);
1916                         LASSERT(ctxt != NULL);
1917                         llog_sync(ctxt, fglog->exp);
1918                 }
1919         } while (fglog != NULL);
1920 }
1921
1922 /* also incredibly similar to mds_disconnect */
1923 static int filter_disconnect(struct obd_export *exp, int flags)
1924 {
1925         struct obd_device *obd = exp->exp_obd;
1926         unsigned long irqflags;
1927         int rc;
1928         ENTRY;
1929
1930         LASSERT(exp);
1931         class_export_get(exp);
1932
1933         spin_lock_irqsave(&exp->exp_lock, irqflags);
1934         exp->exp_flags = flags;
1935         spin_unlock_irqrestore(&exp->exp_lock, irqflags);
1936
1937         if (!(flags & OBD_OPT_FORCE))
1938                 filter_grant_sanity_check(obd, __FUNCTION__);
1939         filter_grant_discard(exp);
1940
1941         /* Disconnect early so that clients can't keep using export */
1942         rc = class_disconnect(exp, flags);
1943
1944         ldlm_cancel_locks_for_export(exp);
1945
1946         fsfilt_sync(obd, obd->u.filter.fo_sb);
1947
1948         /* flush any remaining cancel messages out to the target */
1949         filter_sync_llogs(obd, exp);
1950
1951         class_export_put(exp);
1952         RETURN(rc);
1953 }
1954
1955 struct dentry *__filter_oa2dentry(struct obd_device *obd,
1956                                   struct obdo *oa, const char *what)
1957 {
1958         struct dentry *dchild = NULL;
1959         obd_gr group = 0;
1960
1961         if (oa->o_valid & OBD_MD_FLGROUP)
1962                 group = oa->o_gr;
1963
1964         dchild = filter_fid2dentry(obd, NULL, group, oa->o_id);
1965
1966         if (IS_ERR(dchild)) {
1967                 CERROR("%s error looking up object: "LPU64"\n", what, oa->o_id);
1968                 RETURN(dchild);
1969         }
1970
1971         if (dchild->d_inode == NULL) {
1972                 CERROR("%s: %s on non-existent object: "LPU64"\n",
1973                        obd->obd_name, what, oa->o_id);
1974                 f_dput(dchild);
1975                 RETURN(ERR_PTR(-ENOENT));
1976         }
1977
1978         return dchild;
1979 }
1980
1981 static int filter_getattr(struct obd_export *exp, struct obdo *oa,
1982                           struct lov_stripe_md *md)
1983 {
1984         struct dentry *dentry = NULL;
1985         struct obd_device *obd;
1986         int rc = 0;
1987         ENTRY;
1988
1989         obd = class_exp2obd(exp);
1990         if (obd == NULL) {
1991                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1992                        exp->exp_handle.h_cookie);
1993                 RETURN(-EINVAL);
1994         }
1995
1996         dentry = filter_oa2dentry(obd, oa);
1997         if (IS_ERR(dentry))
1998                 RETURN(PTR_ERR(dentry));
1999
2000         /* Limit the valid bits in the return data to what we actually use */
2001         oa->o_valid = OBD_MD_FLID;
2002         obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2003
2004         f_dput(dentry);
2005         RETURN(rc);
2006 }
2007
2008 /* this is called from filter_truncate() until we have filter_punch() */
2009 static int filter_setattr(struct obd_export *exp, struct obdo *oa,
2010                           struct lov_stripe_md *md, struct obd_trans_info *oti)
2011 {
2012         struct lvfs_run_ctxt saved;
2013         struct filter_obd *filter;
2014         struct dentry *dentry;
2015         struct iattr iattr;
2016         struct ldlm_res_id res_id = { .name = { oa->o_id, 0, oa->o_gr, 0 } };
2017         struct ldlm_resource *res;
2018         void *handle;
2019         int rc, rc2;
2020         ENTRY;
2021
2022         LASSERT(oti != NULL);
2023
2024         dentry = filter_oa2dentry(exp->exp_obd, oa);
2025         if (IS_ERR(dentry))
2026                 RETURN(PTR_ERR(dentry));
2027
2028         filter = &exp->exp_obd->u.filter;
2029
2030         iattr_from_obdo(&iattr, oa, oa->o_valid);
2031
2032         push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2033         lock_kernel();
2034
2035         if (iattr.ia_valid & ATTR_SIZE)
2036                 down(&dentry->d_inode->i_sem);
2037         handle = fsfilt_start(exp->exp_obd, dentry->d_inode, FSFILT_OP_SETATTR,
2038                               oti);
2039         if (IS_ERR(handle))
2040                 GOTO(out_unlock, rc = PTR_ERR(handle));
2041
2042         /* XXX this could be a rwsem instead, if filter_preprw played along */
2043         if (iattr.ia_valid & ATTR_ATTR_FLAG)
2044                 rc = fsfilt_iocontrol(exp->exp_obd, dentry->d_inode, NULL,
2045                                       EXT3_IOC_SETFLAGS,
2046                                       (long)&iattr.ia_attr_flags);
2047         else
2048                 rc = fsfilt_setattr(exp->exp_obd, dentry, handle, &iattr, 1);
2049         rc = filter_finish_transno(exp, oti, rc);
2050         rc2 = fsfilt_commit(exp->exp_obd, filter->fo_sb, dentry->d_inode, handle, 0);
2051         if (rc2) {
2052                 CERROR("error on commit, err = %d\n", rc2);
2053                 if (!rc)
2054                         rc = rc2;
2055         }
2056
2057         if (iattr.ia_valid & ATTR_SIZE) {
2058                 res = ldlm_resource_get(exp->exp_obd->obd_namespace, NULL,
2059                                         res_id, LDLM_EXTENT, 0);
2060                 if (res == NULL) {
2061                         CERROR("!!! resource_get failed for object "LPU64" -- "
2062                                "filter_setattr with no lock?\n", oa->o_id);
2063                 } else {
2064                         if (res->lr_namespace->ns_lvbo &&
2065                             res->lr_namespace->ns_lvbo->lvbo_update) {
2066                                 rc = res->lr_namespace->ns_lvbo->lvbo_update
2067                                         (res, NULL, 0, 0);
2068                         }
2069                         ldlm_resource_putref(res);
2070                 }
2071         }
2072
2073         oa->o_valid = OBD_MD_FLID;
2074         obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2075
2076 out_unlock:
2077         if (iattr.ia_valid & ATTR_SIZE)
2078                 up(&dentry->d_inode->i_sem);
2079         unlock_kernel();
2080         pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2081
2082         f_dput(dentry);
2083         RETURN(rc);
2084 }
2085
2086 /* XXX identical to osc_unpackmd */
2087 static int filter_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2088                            struct lov_mds_md *lmm, int lmm_bytes)
2089 {
2090         int lsm_size;
2091         ENTRY;
2092
2093         if (lmm != NULL) {
2094                 if (lmm_bytes < sizeof (*lmm)) {
2095                         CERROR("lov_mds_md too small: %d, need %d\n",
2096                                lmm_bytes, (int)sizeof(*lmm));
2097                         RETURN(-EINVAL);
2098                 }
2099                 /* XXX LOV_MAGIC etc check? */
2100
2101                 if (lmm->lmm_object_id == cpu_to_le64(0)) {
2102                         CERROR("lov_mds_md: zero lmm_object_id\n");
2103                         RETURN(-EINVAL);
2104                 }
2105         }
2106
2107         lsm_size = lov_stripe_md_size(1);
2108         if (lsmp == NULL)
2109                 RETURN(lsm_size);
2110
2111         if (*lsmp != NULL && lmm == NULL) {
2112                 OBD_FREE(*lsmp, lsm_size);
2113                 *lsmp = NULL;
2114                 RETURN(0);
2115         }
2116
2117         if (*lsmp == NULL) {
2118                 OBD_ALLOC(*lsmp, lsm_size);
2119                 if (*lsmp == NULL)
2120                         RETURN(-ENOMEM);
2121
2122                 loi_init((*lsmp)->lsm_oinfo);
2123         }
2124
2125         if (lmm != NULL) {
2126                 /* XXX zero *lsmp? */
2127                 (*lsmp)->lsm_object_id = le64_to_cpu (lmm->lmm_object_id);
2128                 LASSERT((*lsmp)->lsm_object_id);
2129         }
2130
2131         (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
2132
2133         RETURN(lsm_size);
2134 }
2135
2136 static void filter_destroy_precreated(struct obd_export *exp, struct obdo *oa,
2137                                       struct filter_obd *filter)
2138 {
2139         struct obdo doa = { 0, }; /* XXX obdo on stack */
2140         __u64 last, id;
2141         ENTRY;
2142         LASSERT(oa);
2143
2144         LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2145         LASSERT(oa->o_gr != 0);
2146         doa.o_mode = S_IFREG;
2147         doa.o_gr = oa->o_gr;
2148         doa.o_valid = oa->o_valid & (OBD_MD_FLGROUP | OBD_MD_FLID);
2149
2150         set_bit(doa.o_gr, &filter->fo_destroys_in_progress);
2151         down(&filter->fo_create_locks[doa.o_gr]);
2152         if (!test_bit(doa.o_gr, &filter->fo_destroys_in_progress)) {
2153                 CERROR("%s:["LPU64"] destroy_in_progress already cleared\n",
2154                        exp->exp_obd->obd_name, doa.o_gr);
2155                 up(&filter->fo_create_locks[doa.o_gr]);
2156                 EXIT;
2157                 return;
2158         }
2159
2160         last = filter_last_id(filter, doa.o_gr);
2161         CWARN("%s:["LPU64"] deleting orphan objects from "LPU64" to "LPU64"\n",
2162               exp->exp_obd->obd_name, doa.o_gr, oa->o_id + 1, last);
2163         for (id = oa->o_id + 1; id <= last; id++) {
2164                 doa.o_id = id;
2165                 filter_destroy(exp, &doa, NULL, NULL);
2166         }
2167
2168         CDEBUG(D_HA, "%s:["LPU64"] after destroy: set last_objids = "LPU64"\n",
2169                exp->exp_obd->obd_name, doa.o_gr, oa->o_id);
2170
2171         filter_set_last_id(filter, doa.o_gr, oa->o_id);
2172
2173         clear_bit(doa.o_gr, &filter->fo_destroys_in_progress);
2174         up(&filter->fo_create_locks[doa.o_gr]);
2175
2176         EXIT;
2177 }
2178
2179 /* returns a negative error or a nonnegative number of files to create */
2180 static int filter_should_precreate(struct obd_export *exp, struct obdo *oa,
2181                                    obd_gr group)
2182 {
2183         struct obd_device *obd = exp->exp_obd;
2184         struct filter_obd *filter = &obd->u.filter;
2185         int diff, rc;
2186         ENTRY;
2187
2188         diff = oa->o_id - filter_last_id(filter, oa->o_gr);
2189         CDEBUG(D_INFO, "filter_last_id() = "LPU64" -> diff = %d\n",
2190                filter_last_id(filter, oa->o_gr), diff);
2191
2192         /* delete orphans request */
2193         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
2194             (oa->o_flags & OBD_FL_DELORPHAN)) {
2195                 if (diff >= 0)
2196                         RETURN(diff);
2197                 if (-diff > 10000) { /* XXX make this smarter */
2198                         CERROR("ignoring bogus orphan destroy request: obdid "
2199                                LPU64" last_id "LPU64"\n",
2200                                oa->o_id, filter_last_id(filter, oa->o_gr));
2201                         RETURN(-EINVAL);
2202                 }
2203                 filter_destroy_precreated(exp, oa, filter);
2204                 rc = filter_update_last_objid(obd, group, 0);
2205                 if (rc)
2206                         CERROR("unable to write lastobjid, but orphans"
2207                                "were deleted\n");
2208                 RETURN(0);
2209         } else {
2210                 /* only precreate if group == 0 and o_id is specfied */
2211                 if (!(oa->o_valid & OBD_FL_DELORPHAN) &&
2212                     (/*group != 0 ||*/ oa->o_id == 0))
2213                         RETURN(1);
2214
2215                 LASSERT(diff >= 0);
2216                 RETURN(diff);
2217         }
2218 }
2219 static int filter_precreate_rec(struct obd_device *obd, struct dentry *dentry,
2220                                 int *number, struct obdo *oa)
2221 {
2222         int rc;
2223         ENTRY;
2224
2225         rc = fsfilt_precreate_rec(obd, dentry, number, oa);
2226
2227         RETURN(rc);
2228 }
2229
2230 /* We rely on the fact that only one thread will be creating files in a given
2231  * group at a time, which is why we don't need an atomic filter_get_new_id.
2232  * Even if we had that atomic function, the following race would exist:
2233  *
2234  * thread 1: gets id x from filter_next_id
2235  * thread 2: gets id (x + 1) from filter_next_id
2236  * thread 2: creates object (x + 1)
2237  * thread 1: tries to create object x, gets -ENOSPC
2238  */
2239 static int filter_precreate(struct obd_device *obd, struct obdo *oa,
2240                             obd_gr group, int *num)
2241 {
2242         struct dentry *dchild = NULL, *dparent = NULL;
2243         struct filter_obd *filter;
2244         int err = 0, rc = 0, recreate_obj = 0, i;
2245         __u64 next_id;
2246         void *handle = NULL;
2247         void *lock = NULL;
2248         ENTRY;
2249
2250         filter = &obd->u.filter;
2251
2252         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
2253             (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
2254                 recreate_obj = 1;
2255         }
2256
2257         CDEBUG(D_HA, "%s: precreating %d objects\n", obd->obd_name, *num);
2258
2259         down(&filter->fo_create_locks[group]);
2260
2261         for (i = 0; i < *num && err == 0; i++) {
2262                 int cleanup_phase = 0;
2263
2264                 if (test_bit(group, &filter->fo_destroys_in_progress)) {
2265                         CWARN("%s: precreate aborted by destroy\n",
2266                               obd->obd_name);
2267                         break;
2268                 }
2269
2270                 if (recreate_obj) {
2271                         __u64 last_id;
2272                         next_id = oa->o_id;
2273                         last_id = filter_last_id(filter, group);
2274                         if (next_id > last_id) {
2275                                 CERROR("Error: Trying to recreate obj greater"
2276                                        "than last id "LPD64" > "LPD64"\n",
2277                                        next_id, last_id);
2278                                 GOTO(cleanup, rc = -EINVAL);
2279                         }
2280                 } else {
2281                         next_id = filter_last_id(filter, group) + 1;
2282                 }
2283
2284                 CDEBUG(D_INFO, "precreate objid "LPU64"\n", next_id);
2285
2286                 dparent = filter_parent_lock(obd, group, next_id, &lock);
2287                 if (IS_ERR(dparent))
2288                         GOTO(cleanup, rc = PTR_ERR(dparent));
2289                 cleanup_phase = 1;
2290
2291                 /*only do precreate rec record. so clean kml flags here*/
2292                 fsfilt_clear_fs_flags(obd, dparent->d_inode, SM_DO_REC);
2293
2294                 dchild = filter_fid2dentry(obd, dparent, group, next_id);
2295                 if (IS_ERR(dchild))
2296                         GOTO(cleanup, rc = PTR_ERR(dchild));
2297                 cleanup_phase = 2;
2298
2299                 if (dchild->d_inode != NULL) {
2300                         /* This would only happen if lastobjid was bad on disk*/
2301                         /* Could also happen if recreating missing obj but
2302                          * already exists
2303                          */
2304                         if (recreate_obj) {
2305                                 CERROR("%s: Serious error: recreating obj %*s "
2306                                        "but obj already exists \n",
2307                                        obd->obd_name, dchild->d_name.len,
2308                                        dchild->d_name.name);
2309                                 LBUG();
2310                         } else {
2311                                 CERROR("%s: Serious error: objid %*s already "
2312                                        "exists; is this filesystem corrupt?\n",
2313                                        obd->obd_name, dchild->d_name.len,
2314                                        dchild->d_name.name);
2315                                 LBUG();
2316                         }
2317                         GOTO(cleanup, rc = -EEXIST);
2318                 }
2319
2320                 handle = fsfilt_start_log(obd, dparent->d_inode,
2321                                           FSFILT_OP_CREATE, NULL, 1);
2322                 if (IS_ERR(handle))
2323                         GOTO(cleanup, rc = PTR_ERR(handle));
2324                 cleanup_phase = 3;
2325
2326                 rc = ll_vfs_create(dparent->d_inode, dchild, S_IFREG, NULL);
2327                 if (rc) {
2328                         CERROR("create failed rc = %d\n", rc);
2329                         GOTO(cleanup, rc);
2330                 }
2331
2332                 if (!recreate_obj) {
2333                         filter_set_last_id(filter, group, next_id);
2334                         err = filter_update_last_objid(obd, group, 0);
2335                         if (err)
2336                                 CERROR("unable to write lastobjid "
2337                                        "but file created\n");
2338                 }
2339                 fsfilt_set_fs_flags(obd, dparent->d_inode, SM_DO_REC);
2340
2341         cleanup:
2342                 switch(cleanup_phase) {
2343                 case 3:
2344                         err = fsfilt_commit(obd, filter->fo_sb,
2345                                             dparent->d_inode, handle, 0);
2346                         if (err) {
2347                                 CERROR("error on commit, err = %d\n", err);
2348                                 if (!rc)
2349                                         rc = err;
2350                         }
2351                 case 2:
2352                         f_dput(dchild);
2353                 case 1:
2354                         filter_parent_unlock(dparent, lock);
2355                 case 0:
2356                         break;
2357                 }
2358
2359                 if (rc)
2360                         break;
2361         }
2362
2363         *num = i;
2364
2365         /* check if we have an error after ll_vfs_create(). It is possible that
2366          * there will be say -ENOSPC and we will leak it. */
2367         if (rc == 0)
2368                 rc = filter_precreate_rec(obd, dparent, num, oa);
2369
2370         up(&filter->fo_create_locks[group]);
2371
2372         CDEBUG(D_HA, "%s: server last_objid for group "LPU64": "LPU64"\n",
2373                obd->obd_name, group, filter->fo_last_objids[group]);
2374
2375         CDEBUG(D_HA, "%s: filter_precreate() created %d objects\n",
2376                obd->obd_name, i);
2377
2378         RETURN(rc);
2379 }
2380
2381 static int filter_create(struct obd_export *exp, struct obdo *oa,
2382                          struct lov_stripe_md **ea, struct obd_trans_info *oti)
2383 {
2384         struct obd_device *obd = NULL;
2385         struct filter_obd *filter;
2386         struct lvfs_run_ctxt saved;
2387         struct lov_stripe_md *lsm = NULL;
2388         struct filter_export_data *fed;
2389         char str[PTL_NALFMT_SIZE];
2390         int group = oa->o_gr, rc = 0, diff, recreate_objs = 0;
2391         ENTRY;
2392
2393         if (!(oa->o_valid & OBD_MD_FLGROUP) || group == 0) {
2394                 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
2395                                 exp->exp_connection->c_peer.peer_nid, str);
2396                 CERROR("!!! nid "LPX64"/%s sent invalid object group %d\n",
2397                        exp->exp_connection->c_peer.peer_nid, str, group);
2398                 RETURN(-EINVAL);
2399         }
2400
2401         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
2402             (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
2403                 recreate_objs = 1;
2404         }
2405
2406         obd = exp->exp_obd;
2407         fed = &exp->exp_filter_data;
2408         filter = &obd->u.filter;
2409
2410         if (fed->fed_group != group && !recreate_objs &&
2411             !(oa->o_valid & OBD_MD_REINT)) {
2412                 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
2413                                 exp->exp_connection->c_peer.peer_nid, str);
2414                 CERROR("!!! This export (nid "LPX64"/%s) used object group %d "
2415                        "earlier; now it's trying to use group %d!  This could "
2416                        "be a bug in the MDS.  Tell CFS.\n",
2417                        exp->exp_connection->c_peer.peer_nid, str,
2418                        fed->fed_group, group);
2419                 RETURN(-ENOTUNIQ);
2420         }
2421
2422         CDEBUG(D_INFO, "filter_create(od->o_gr=%d,od->o_id="LPU64")\n",
2423                group, oa->o_id);
2424         if (ea != NULL) {
2425                 lsm = *ea;
2426                 if (lsm == NULL) {
2427                         rc = obd_alloc_memmd(exp, &lsm);
2428                         if (rc < 0)
2429                                 RETURN(rc);
2430                 }
2431         }
2432
2433         obd = exp->exp_obd;
2434         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2435
2436         if (oa->o_valid & OBD_MD_REINT) {
2437                 int num = *((int*)oa->o_inline);
2438                 rc = filter_precreate(obd, oa, oa->o_gr, &num);
2439         } else if (recreate_objs) {
2440                 if (oa->o_id > filter_last_id(&obd->u.filter, group)) {
2441                         CERROR("recreate objid "LPU64" > last id "LPU64"\n",
2442                                oa->o_id, filter_last_id(&obd->u.filter, group));
2443                         rc = -EINVAL;
2444                 } else {
2445                         diff = 1;
2446                         rc = filter_precreate(obd, oa, group, &diff);
2447                 }
2448         } else {
2449                 diff = filter_should_precreate(exp, oa, group);
2450                 if (diff > 0) {
2451                         oa->o_id = filter_last_id(&obd->u.filter, group);
2452                         rc = filter_precreate(obd, oa, group, &diff);
2453                         oa->o_id += diff;
2454                         oa->o_valid = OBD_MD_FLID;
2455                 }
2456         }
2457
2458         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2459         if (rc && ea != NULL && *ea != lsm) {
2460                 obd_free_memmd(exp, &lsm);
2461         } else if (rc == 0 && ea != NULL) {
2462                 /* XXX LOV STACKING: the lsm that is passed to us from
2463                  * LOV does not have valid lsm_oinfo data structs, so
2464                  * don't go touching that.  This needs to be fixed in a
2465                  * big way. */
2466                 lsm->lsm_object_id = oa->o_id;
2467                 lsm->lsm_object_gr = oa->o_gr;
2468                 *ea = lsm;
2469         }
2470
2471         RETURN(rc);
2472 }
2473
2474 static int filter_destroy(struct obd_export *exp, struct obdo *oa,
2475                           struct lov_stripe_md *ea, struct obd_trans_info *oti)
2476 {
2477         struct obd_device *obd;
2478         struct filter_obd *filter;
2479         struct dentry *dchild = NULL, *dparent = NULL;
2480         struct lvfs_run_ctxt saved;
2481         void *handle = NULL;
2482         struct llog_cookie *fcc = NULL;
2483         int rc, rc2, cleanup_phase = 0, have_prepared = 0;
2484         void *lock = NULL;
2485         ENTRY;
2486
2487         LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2488
2489         obd = exp->exp_obd;
2490         filter = &obd->u.filter;
2491
2492         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2493
2494  acquire_locks:
2495         dparent = filter_parent_lock(obd, oa->o_gr, oa->o_id, &lock);
2496         if (IS_ERR(dparent))
2497                 GOTO(cleanup, rc = PTR_ERR(dparent));
2498         cleanup_phase = 1;
2499
2500         dchild = filter_fid2dentry(obd, dparent, oa->o_gr, oa->o_id);
2501         if (IS_ERR(dchild))
2502                 GOTO(cleanup, rc = -ENOENT);
2503         cleanup_phase = 2;
2504
2505         if (dchild->d_inode == NULL) {
2506                 CDEBUG(D_INODE, "destroying non-existent object "LPU64"\n",
2507                        oa->o_id);
2508                 GOTO(cleanup, rc = -ENOENT);
2509         }
2510
2511         if (!have_prepared) {
2512                 /* If we're really going to destroy the object, get ready
2513                  * by getting the clients to discard their cached data.
2514                  *
2515                  * We have to drop the parent lock, because
2516                  * filter_prepare_destroy will acquire a PW on the object, and
2517                  * we don't want to deadlock with an incoming write to the
2518                  * object, which has the extent PW and then wants to get the
2519                  * parent dentry to do the lookup.
2520                  *
2521                  * We dput the child because it's not worth the extra
2522                  * complication of condition the above code to skip it on the
2523                  * second time through. */
2524                 f_dput(dchild);
2525                 filter_parent_unlock(dparent, lock);
2526
2527                 filter_prepare_destroy(obd, oa->o_id, oa->o_gr);
2528                 have_prepared = 1;
2529                 goto acquire_locks;
2530         }
2531
2532         handle = fsfilt_start_log(obd, dparent->d_inode, FSFILT_OP_UNLINK, oti, 1);
2533         if (IS_ERR(handle))
2534                 GOTO(cleanup, rc = PTR_ERR(handle));
2535
2536         cleanup_phase = 3;
2537
2538         /* Our MDC connection is established by the MDS to us */
2539         if (oa->o_valid & OBD_MD_FLCOOKIE) {
2540                 OBD_ALLOC(fcc, sizeof(*fcc));
2541                 if (fcc != NULL)
2542                         memcpy(fcc, obdo_logcookie(oa), sizeof(*fcc));
2543         }
2544
2545         rc = filter_destroy_internal(obd, oa->o_id, dparent, dchild);
2546
2547 cleanup:
2548         switch(cleanup_phase) {
2549         case 3:
2550                 if (fcc != NULL) {
2551                         if (oti != NULL)
2552                                 fsfilt_add_journal_cb(obd, filter->fo_sb, 0,
2553                                                       oti->oti_handle,
2554                                                       filter_cancel_cookies_cb,
2555                                                       fcc);
2556                         else
2557                                 fsfilt_add_journal_cb(obd, filter->fo_sb, 0,
2558                                                       handle,
2559                                                       filter_cancel_cookies_cb,
2560                                                       fcc);
2561                 }
2562                 rc = filter_finish_transno(exp, oti, rc);
2563                 rc2 = fsfilt_commit(obd, filter->fo_sb, dparent->d_inode,
2564                                     handle, 0);
2565                 if (rc2) {
2566                         CERROR("error on commit, err = %d\n", rc2);
2567                         if (!rc)
2568                                 rc = rc2;
2569                 }
2570         case 2:
2571                 f_dput(dchild);
2572         case 1:
2573                 filter_parent_unlock(dparent, lock);
2574         case 0:
2575                 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2576                 break;
2577         default:
2578                 CERROR("invalid cleanup_phase %d\n", cleanup_phase);
2579                 LBUG();
2580         }
2581
2582         RETURN(rc);
2583 }
2584
2585 /* NB start and end are used for punch, but not truncate */
2586 static int filter_truncate(struct obd_export *exp, struct obdo *oa,
2587                            struct lov_stripe_md *lsm,
2588                            obd_off start, obd_off end,
2589                            struct obd_trans_info *oti)
2590 {
2591         int error;
2592         ENTRY;
2593
2594         if (end != OBD_OBJECT_EOF)
2595                 CERROR("PUNCH not supported, only truncate: end = "LPX64"\n",
2596                        end);
2597
2598         CDEBUG(D_INODE, "calling truncate for object "LPU64", valid = %x, "
2599                "o_size = "LPD64"\n", oa->o_id, oa->o_valid, start);
2600         oa->o_size = start;
2601         error = filter_setattr(exp, oa, NULL, oti);
2602         RETURN(error);
2603 }
2604
2605 static int filter_sync(struct obd_export *exp, struct obdo *oa,
2606                        struct lov_stripe_md *lsm, obd_off start, obd_off end)
2607 {
2608         struct obd_device *obd = exp->exp_obd;
2609         struct lvfs_run_ctxt saved;
2610         struct filter_obd *filter;
2611         struct dentry *dentry;
2612         int rc, rc2;
2613         ENTRY;
2614
2615         filter = &obd->u.filter;
2616
2617         /* an objid of zero is taken to mean "sync whole filesystem" */
2618         if (!oa || !(oa->o_valid & OBD_MD_FLID)) {
2619                 rc = fsfilt_sync(obd, filter->fo_sb);
2620                 /* flush any remaining cancel messages out to the target */
2621                 filter_sync_llogs(obd, NULL);
2622                 RETURN(rc);
2623         }
2624
2625         dentry = filter_oa2dentry(obd, oa);
2626         if (IS_ERR(dentry))
2627                 RETURN(PTR_ERR(dentry));
2628
2629         push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2630
2631         down(&dentry->d_inode->i_sem);
2632         rc = filemap_fdatasync(dentry->d_inode->i_mapping);
2633         if (rc == 0) {
2634                 /* just any file to grab fsync method - "file" arg unused */
2635                 struct file *file = filter->fo_rcvd_filp;
2636
2637                 if (file->f_op && file->f_op->fsync)
2638                         rc = file->f_op->fsync(NULL, dentry, 1);
2639
2640                 rc2 = filemap_fdatawait(dentry->d_inode->i_mapping);
2641                 if (!rc)
2642                         rc = rc2;
2643         }
2644         up(&dentry->d_inode->i_sem);
2645
2646         oa->o_valid = OBD_MD_FLID;
2647         obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2648
2649         pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2650
2651         f_dput(dentry);
2652         RETURN(rc);
2653 }
2654
2655 static int filter_statfs(struct obd_device *obd, struct obd_statfs *osfs,
2656                          unsigned long max_age)
2657 {
2658         struct filter_obd *filter = &obd->u.filter;
2659         int blockbits = filter->fo_sb->s_blocksize_bits;
2660         int rc;
2661         ENTRY;
2662
2663         /* at least try to account for cached pages.  its still racey and
2664          * might be under-reporting if clients haven't announced their
2665          * caches with brw recently */
2666         spin_lock(&obd->obd_osfs_lock);
2667         rc = fsfilt_statfs(obd, filter->fo_sb, max_age);
2668         memcpy(osfs, &obd->obd_osfs, sizeof(*osfs));
2669         spin_unlock(&obd->obd_osfs_lock);
2670
2671         CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
2672                " pending "LPU64" free "LPU64" avail "LPU64"\n",
2673                filter->fo_tot_dirty, filter->fo_tot_granted,
2674                filter->fo_tot_pending,
2675                osfs->os_bfree << blockbits, osfs->os_bavail << blockbits);
2676
2677         filter_grant_sanity_check(obd, __FUNCTION__);
2678
2679         osfs->os_bavail -= min(osfs->os_bavail,
2680                                (filter->fo_tot_dirty + filter->fo_tot_pending +
2681                                 osfs->os_bsize -1) >> blockbits);
2682
2683         RETURN(rc);
2684 }
2685
2686 static int filter_get_info(struct obd_export *exp, __u32 keylen,
2687                            void *key, __u32 *vallen, void *val)
2688 {
2689         struct filter_export_data *fed = &exp->exp_filter_data;
2690         struct obd_device *obd;
2691         ENTRY;
2692
2693         obd = class_exp2obd(exp);
2694         if (obd == NULL) {
2695                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2696                        exp->exp_handle.h_cookie);
2697                 RETURN(-EINVAL);
2698         }
2699
2700         if (keylen == strlen("blocksize") &&
2701             memcmp(key, "blocksize", keylen) == 0) {
2702                 __u32 *blocksize = val;
2703                 *vallen = sizeof(*blocksize);
2704                 *blocksize = obd->u.filter.fo_sb->s_blocksize;
2705                 RETURN(0);
2706         }
2707
2708         if (keylen == strlen("blocksize_bits") &&
2709             memcmp(key, "blocksize_bits", keylen) == 0) {
2710                 __u32 *blocksize_bits = val;
2711                 *vallen = sizeof(*blocksize_bits);
2712                 *blocksize_bits = obd->u.filter.fo_sb->s_blocksize_bits;
2713                 RETURN(0);
2714         }
2715
2716         if (keylen >= strlen("last_id") && memcmp(key, "last_id", 7) == 0) {
2717                 obd_id *last_id = val;
2718                 *last_id = filter_last_id(&obd->u.filter, fed->fed_group);
2719                 RETURN(0);
2720         }
2721         if (keylen >= strlen("reint_log") && memcmp(key, "reint_log", 9) == 0) {
2722                 /*Get log_context handle*/
2723                 unsigned long *llh_handle = val;
2724                 *vallen = sizeof(unsigned long);
2725                 *llh_handle = (unsigned long)obd->obd_llog_ctxt[LLOG_REINT_ORIG_CTXT];
2726                 RETURN(0);
2727         }
2728         if (keylen >= strlen("cache_sb") && memcmp(key, "cache_sb", 8) == 0) {
2729                 /*Get log_context handle*/
2730                 unsigned long *sb = val;
2731                 *vallen = sizeof(unsigned long);
2732                 *sb = (unsigned long)obd->u.filter.fo_sb;
2733                 RETURN(0);
2734         }
2735
2736         CDEBUG(D_IOCTL, "invalid key\n");
2737         RETURN(-EINVAL);
2738 }
2739
2740 struct obd_llogs *filter_grab_llog_for_group(struct obd_device *obd, int group,
2741                                              struct obd_export *export)
2742 {
2743         struct filter_group_llog *fglog, *nlog;
2744         char name[32] = "CATLIST";
2745         struct filter_obd *filter;
2746         struct llog_ctxt *ctxt;
2747         struct list_head *cur;
2748         int rc;
2749
2750         filter = &obd->u.filter;
2751
2752         spin_lock(&filter->fo_llog_list_lock);
2753         list_for_each(cur, &filter->fo_llog_list) {
2754                 fglog = list_entry(cur, struct filter_group_llog, list);
2755                 if (fglog->group == group) {
2756                         if (!(fglog->exp == NULL || fglog->exp == export || export == NULL))
2757                                 CWARN("%s: export for group %d changes: 0x%p -> 0x%p\n",
2758                                       obd->obd_name, group, fglog->exp, export);
2759                         spin_unlock(&filter->fo_llog_list_lock);
2760                         goto init;
2761                 }
2762         }
2763         spin_unlock(&filter->fo_llog_list_lock);
2764
2765         if (export == NULL)
2766                 RETURN(NULL);
2767
2768         OBD_ALLOC(fglog, sizeof(*fglog));
2769         if (fglog == NULL)
2770                 RETURN(NULL);
2771         fglog->group = group;
2772
2773         OBD_ALLOC(fglog->llogs, sizeof(struct obd_llogs));
2774         if (fglog->llogs == NULL) {
2775                 OBD_FREE(fglog, sizeof(*fglog));
2776                 RETURN(NULL);
2777         }
2778
2779         spin_lock(&filter->fo_llog_list_lock);
2780         list_for_each(cur, &filter->fo_llog_list) {
2781                 nlog = list_entry(cur, struct filter_group_llog, list);
2782                 LASSERT(nlog->group != group);
2783         }
2784         list_add(&fglog->list, &filter->fo_llog_list);
2785         spin_unlock(&filter->fo_llog_list_lock);
2786
2787         rc = obd_llog_cat_initialize(obd, fglog->llogs, 1, name);
2788         if (rc) {
2789                 OBD_FREE(fglog->llogs, sizeof(*(fglog->llogs)));
2790                 OBD_FREE(fglog, sizeof(*fglog));
2791                 RETURN(NULL);
2792         }
2793
2794 init:
2795         if (export) {
2796                 fglog->exp = export;
2797                 ctxt = llog_get_context(fglog->llogs, LLOG_UNLINK_REPL_CTXT);
2798                 LASSERT(ctxt != NULL);
2799
2800                 llog_receptor_accept(ctxt, export->exp_imp_reverse);
2801         }
2802         CDEBUG(D_OTHER, "%s: new llog 0x%p for group %u\n",
2803                obd->obd_name, fglog->llogs, group);
2804
2805         RETURN(fglog->llogs);
2806 }
2807
2808 int filter_iocontrol(unsigned int cmd, struct obd_export *exp,
2809                      int len, void *karg, void *uarg)
2810 {
2811         struct obd_device *obd = exp->exp_obd;
2812         struct obd_ioctl_data *data = karg;
2813         int rc = 0;
2814
2815         switch (cmd) {
2816         case OBD_IOC_ABORT_RECOVERY:
2817                 target_stop_recovery_thread(obd);
2818                 RETURN(0);
2819
2820         case OBD_IOC_SET_READONLY: {
2821                 void *handle;
2822                 struct super_block *sb = obd->u.filter.fo_sb;
2823                 struct inode *inode = sb->s_root->d_inode;
2824                 BDEVNAME_DECLARE_STORAGE(tmp);
2825                 CERROR("setting device %s read-only\n",
2826                        ll_bdevname(sb, tmp));
2827
2828                 handle = fsfilt_start(obd, inode, FSFILT_OP_MKNOD, NULL);
2829                 LASSERT(handle);
2830                 (void)fsfilt_commit(obd, sb, inode, handle, 1);
2831
2832                 dev_set_rdonly(ll_sbdev(obd->u.filter.fo_sb), 2);
2833                 RETURN(0);
2834         }
2835
2836         case OBD_IOC_CATLOGLIST: {
2837                 rc = llog_catalog_list(obd, 1, data);
2838                 RETURN(rc);
2839         }
2840         case OBD_IOC_LLOG_CANCEL:
2841         case OBD_IOC_LLOG_REMOVE:
2842         case OBD_IOC_LLOG_INFO:
2843         case OBD_IOC_LLOG_PRINT: {
2844                 /* FIXME to be finished */
2845                 RETURN(-EOPNOTSUPP);
2846 /*
2847                 struct llog_ctxt *ctxt = NULL;
2848
2849                 push_ctxt(&saved, &ctxt->loc_ctxt, NULL);
2850                 rc = llog_ioctl(ctxt, cmd, data);
2851                 pop_ctxt(&saved, &ctxt->loc_ctxt, NULL);
2852
2853                 RETURN(rc);
2854 */
2855         }
2856
2857
2858         default:
2859                 RETURN(-EINVAL);
2860         }
2861         RETURN(0);
2862 }
2863
2864 static struct llog_operations filter_unlink_repl_logops;
2865 static struct llog_operations filter_size_orig_logops = {
2866         lop_setup: llog_obd_origin_setup,
2867         lop_cleanup: llog_catalog_cleanup,
2868         lop_add: llog_catalog_add,
2869 };
2870
2871 static int filter_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
2872                             struct obd_device *tgt, int count,
2873                             struct llog_catid *catid)
2874 {
2875         struct llog_ctxt *ctxt;
2876         int rc;
2877         ENTRY;
2878
2879         filter_unlink_repl_logops = llog_client_ops;
2880         filter_unlink_repl_logops.lop_cancel = llog_obd_repl_cancel;
2881         filter_unlink_repl_logops.lop_connect = llog_repl_connect;
2882         filter_unlink_repl_logops.lop_sync = llog_obd_repl_sync;
2883
2884         rc = obd_llog_setup(obd, llogs, LLOG_UNLINK_REPL_CTXT, tgt, 0, NULL,
2885                         &filter_unlink_repl_logops);
2886         if (rc)
2887                 RETURN(rc);
2888         /* FIXME - assign unlink_cb for filter's recovery */
2889         ctxt = llog_get_context(llogs, LLOG_UNLINK_REPL_CTXT);
2890         ctxt->llog_proc_cb = filter_recov_log_unlink_cb;
2891
2892         /* FIXME - count should be 1 to setup size log */
2893         rc = obd_llog_setup(obd, llogs, LLOG_SIZE_ORIG_CTXT, tgt, 0,
2894                             &catid->lci_logid, &filter_size_orig_logops);
2895         RETURN(rc);
2896 }
2897
2898 static int filter_llog_finish(struct obd_device *obd,
2899                               struct obd_llogs *llogs, int count)
2900 {
2901         int rc;
2902         ENTRY;
2903
2904         rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_UNLINK_REPL_CTXT));
2905         if (rc)
2906                 RETURN(rc);
2907
2908         rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_SIZE_ORIG_CTXT));
2909         RETURN(rc);
2910 }
2911
2912 static int filter_llog_connect(struct obd_export *exp,
2913                                struct llogd_conn_body *body)
2914 {
2915         struct obd_device *obd = exp->exp_obd;
2916         struct llog_ctxt *ctxt;
2917         struct obd_llogs *llog;
2918         int rc;
2919         ENTRY;
2920
2921         CDEBUG(D_OTHER, "handle connect for %s: %u/%u/%u\n", obd->obd_name,
2922                (unsigned) body->lgdc_logid.lgl_ogr,
2923                (unsigned) body->lgdc_logid.lgl_oid,
2924                (unsigned) body->lgdc_logid.lgl_ogen);
2925         llog = filter_grab_llog_for_group(obd, body->lgdc_logid.lgl_ogr, exp);
2926         LASSERT(llog != NULL);
2927         ctxt = llog_get_context(llog, body->lgdc_ctxt_idx);
2928         rc = llog_connect(ctxt, 1, &body->lgdc_logid,
2929                           &body->lgdc_gen, NULL);
2930         if (rc != 0)
2931                 CERROR("failed to connect\n");
2932
2933         RETURN(rc);
2934 }
2935
2936 static struct dentry *filter_lvfs_fid2dentry(__u64 id, __u32 gen, __u64 gr,
2937                                              void *data)
2938 {
2939         return filter_fid2dentry(data, NULL, gr, id);
2940 }
2941
2942 static struct lvfs_callback_ops filter_lvfs_ops = {
2943         l_fid2dentry:     filter_lvfs_fid2dentry,
2944 };
2945
2946 static struct obd_ops filter_obd_ops = {
2947         .o_owner          = THIS_MODULE,
2948         .o_attach         = filter_attach,
2949         .o_detach         = filter_detach,
2950         .o_get_info       = filter_get_info,
2951         .o_setup          = filter_setup,
2952         .o_precleanup     = filter_precleanup,
2953         .o_cleanup        = filter_cleanup,
2954         .o_connect        = filter_connect,
2955         .o_connect_post   = filter_connect_post,
2956         .o_disconnect     = filter_disconnect,
2957         .o_statfs         = filter_statfs,
2958         .o_getattr        = filter_getattr,
2959         .o_unpackmd       = filter_unpackmd,
2960         .o_create         = filter_create,
2961         .o_setattr        = filter_setattr,
2962         .o_destroy        = filter_destroy,
2963         .o_brw            = filter_brw,
2964         .o_punch          = filter_truncate,
2965         .o_sync           = filter_sync,
2966         .o_preprw         = filter_preprw,
2967         .o_commitrw       = filter_commitrw,
2968         .o_do_cow         = filter_do_cow,
2969         .o_write_extents  = filter_write_extents,
2970         .o_destroy_export = filter_destroy_export,
2971         .o_llog_init      = filter_llog_init,
2972         .o_llog_finish    = filter_llog_finish,
2973         .o_llog_connect   = filter_llog_connect,
2974         .o_iocontrol      = filter_iocontrol,
2975 };
2976
2977 static struct obd_ops filter_sanobd_ops = {
2978         .o_owner          = THIS_MODULE,
2979         .o_attach         = filter_attach,
2980         .o_detach         = filter_detach,
2981         .o_get_info       = filter_get_info,
2982         .o_setup          = filter_san_setup,
2983         .o_precleanup     = filter_precleanup,
2984         .o_cleanup        = filter_cleanup,
2985         .o_connect        = filter_connect,
2986         .o_connect_post   = filter_connect_post,
2987         .o_disconnect     = filter_disconnect,
2988         .o_statfs         = filter_statfs,
2989         .o_getattr        = filter_getattr,
2990         .o_unpackmd       = filter_unpackmd,
2991         .o_create         = filter_create,
2992         .o_setattr        = filter_setattr,
2993         .o_destroy        = filter_destroy,
2994         .o_brw            = filter_brw,
2995         .o_punch          = filter_truncate,
2996         .o_sync           = filter_sync,
2997         .o_preprw         = filter_preprw,
2998         .o_commitrw       = filter_commitrw,
2999         .o_do_cow         = filter_do_cow,
3000         .o_write_extents  = filter_write_extents,
3001         .o_san_preprw     = filter_san_preprw,
3002         .o_destroy_export = filter_destroy_export,
3003         .o_llog_init      = filter_llog_init,
3004         .o_llog_finish    = filter_llog_finish,
3005         .o_llog_connect   = filter_llog_connect,
3006         .o_iocontrol      = filter_iocontrol,
3007 };
3008
3009 static int __init obdfilter_init(void)
3010 {
3011         struct lprocfs_static_vars lvars;
3012         int rc;
3013
3014         printk(KERN_INFO "Lustre: Filtering OBD driver; info@clusterfs.com\n");
3015
3016         lprocfs_init_vars(filter, &lvars);
3017
3018         rc = class_register_type(&filter_obd_ops, NULL, lvars.module_vars,
3019                                  OBD_FILTER_DEVICENAME);
3020         if (rc)
3021                 return rc;
3022
3023         rc = class_register_type(&filter_sanobd_ops, NULL, lvars.module_vars,
3024                                  OBD_FILTER_SAN_DEVICENAME);
3025         if (rc)
3026                 class_unregister_type(OBD_FILTER_DEVICENAME);
3027         return rc;
3028 }
3029
3030 static void __exit obdfilter_exit(void)
3031 {
3032         class_unregister_type(OBD_FILTER_SAN_DEVICENAME);
3033         class_unregister_type(OBD_FILTER_DEVICENAME);
3034 }
3035
3036 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
3037 MODULE_DESCRIPTION("Lustre Filtering OBD driver");
3038 MODULE_LICENSE("GPL");
3039
3040 module_init(obdfilter_init);
3041 module_exit(obdfilter_exit);