lustre/obdfilter/filter.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  linux/fs/obdfilter/filter.c
   5  *
   6  *  Copyright (c) 2001-2003 Cluster File Systems, Inc.
   7  *   Author: Peter Braam <braam@clusterfs.com>
   8  *   Author: Andreas Dilger <adilger@clusterfs.com>
   9  *
  10  *   This file is part of Lustre, http://www.lustre.org.
  11  *
  12  *   Lustre is free software; you can redistribute it and/or
  13  *   modify it under the terms of version 2 of the GNU General Public
  14  *   License as published by the Free Software Foundation.
  15  *
  16  *   Lustre is distributed in the hope that it will be useful,
  17  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  19  *   GNU General Public License for more details.
  20  *
  21  *   You should have received a copy of the GNU General Public License
  22  *   along with Lustre; if not, write to the Free Software
  23  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  24  */
  25
  26 /*
  27  * Invariant: Get O/R i_sem for lookup, if needed, before any journal ops
  28  *            (which need to get journal_lock, may block if journal full).
  29  *
  30  * Invariant: Call filter_start_transno() before any journal ops to avoid the
  31  *            same deadlock problem.  We can (and want) to get rid of the
  32  *            transno sem in favour of the dir/inode i_sem to avoid single
  33  *            threaded operation on the OST.
  34  */
  35
  36 #define DEBUG_SUBSYSTEM S_FILTER
  37
  38 #include <linux/config.h>
  39 #include <linux/module.h>
  40 #include <linux/fs.h>
  41 #include <linux/dcache.h>
  42 #include <linux/init.h>
  43 #include <linux/version.h>
  44 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
  45 # include <linux/mount.h>
  46 # include <linux/buffer_head.h>
  47 #endif
  48
  49 #include <linux/obd_class.h>
  50 #include <linux/obd_lov.h>
  51 #include <linux/lustre_dlm.h>
  52 #include <linux/lustre_fsfilt.h>
  53 #include <linux/lprocfs_status.h>
  54 #include <linux/lustre_log.h>
  55 #include <linux/lustre_commit_confd.h>
  56 #include <portals/list.h>
  57
  58 #include "filter_internal.h"
  59
  60 static struct lvfs_callback_ops filter_lvfs_ops;
  61
  62 static int filter_destroy(struct obd_export *exp, struct obdo *oa,
  63                           struct lov_stripe_md *ea, struct obd_trans_info *);
  64
  65 static void filter_commit_cb(struct obd_device *obd, __u64 transno,
  66                              void *cb_data, int error)
  67 {
  68         obd_transno_commit_cb(obd, transno, error);
  69 }
  70
  71
  72 /* Assumes caller has already pushed us into the kernel context. */
  73 int filter_finish_transno(struct obd_export *exp, struct obd_trans_info *oti,
  74                           int rc)
  75 {
  76         struct filter_obd *filter = &exp->exp_obd->u.filter;
  77         struct filter_export_data *fed = &exp->exp_filter_data;
  78         struct filter_client_data *fcd = fed->fed_fcd;
  79         __u64 last_rcvd;
  80         loff_t off;
  81         int err, log_pri = D_HA;
  82
  83         /* Propagate error code. */
  84         if (rc)
  85                 RETURN(rc);
  86
  87         if (!exp->exp_obd->obd_replayable || oti == NULL)
  88                 RETURN(rc);
  89
  90         /* we don't allocate new transnos for replayed requests */
  91         if (oti->oti_transno == 0) {
  92                 spin_lock(&filter->fo_translock);
  93                 last_rcvd = le64_to_cpu(filter->fo_fsd->fsd_last_transno) + 1;
  94                 filter->fo_fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
  95                 spin_unlock(&filter->fo_translock);
  96                 oti->oti_transno = last_rcvd;
  97         } else {
  98                 spin_lock(&filter->fo_translock);
  99                 last_rcvd = oti->oti_transno;
 100                 if (last_rcvd > le64_to_cpu(filter->fo_fsd->fsd_last_transno))
 101                         filter->fo_fsd->fsd_last_transno =
 102                                 cpu_to_le64(last_rcvd);
 103                 spin_unlock(&filter->fo_translock);
 104         }
 105         fcd->fcd_last_rcvd = cpu_to_le64(last_rcvd);
 106
 107         /* could get xid from oti, if it's ever needed */
 108         fcd->fcd_last_xid = 0;
 109
 110         off = fed->fed_lr_off;
 111         fsfilt_add_journal_cb(exp->exp_obd, last_rcvd, oti->oti_handle,
 112                               filter_commit_cb, NULL);
 113         err = fsfilt_write_record(exp->exp_obd, filter->fo_rcvd_filp, fcd,
 114                                   sizeof(*fcd), &off, 0);
 115         if (err) {
 116                 log_pri = D_ERROR;
 117                 if (rc == 0)
 118                         rc = err;
 119         }
 120
 121         CDEBUG(log_pri, "wrote trans "LPU64" for client %s at #%d: err = %d\n",
 122                last_rcvd, fcd->fcd_uuid, fed->fed_lr_idx, err);
 123
 124         RETURN(rc);
 125 }
 126
 127 void f_dput(struct dentry *dentry)
 128 {
 129         /* Can't go inside filter_ddelete because it can block */
 130         CDEBUG(D_INODE, "putting %s: %p, count = %d\n",
 131                dentry->d_name.name, dentry, atomic_read(&dentry->d_count) - 1);
 132         LASSERT(atomic_read(&dentry->d_count) > 0);
 133
 134         dput(dentry);
 135 }
 136
 137 /* Add client data to the FILTER.  We use a bitmap to locate a free space
 138  * in the last_rcvd file if cl_idx is -1 (i.e. a new client).
 139  * Otherwise, we have just read the data from the last_rcvd file and
 140  * we know its offset. */
 141 static int filter_client_add(struct obd_device *obd, struct filter_obd *filter,
 142                              struct filter_export_data *fed, int cl_idx)
 143 {
 144         unsigned long *bitmap = filter->fo_last_rcvd_slots;
 145         int new_client = (cl_idx == -1);
 146         ENTRY;
 147
 148         LASSERT(bitmap != NULL);
 149
 150         /* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
 151         if (!strcmp(fed->fed_fcd->fcd_uuid, obd->obd_uuid.uuid))
 152                 RETURN(0);
 153
 154         /* the bitmap operations can handle cl_idx > sizeof(long) * 8, so
 155          * there's no need for extra complication here
 156          */
 157         if (new_client) {
 158                 cl_idx = find_first_zero_bit(bitmap, FILTER_LR_MAX_CLIENTS);
 159         repeat:
 160                 if (cl_idx >= FILTER_LR_MAX_CLIENTS) {
 161                         CERROR("no client slots - fix FILTER_LR_MAX_CLIENTS\n");
 162                         RETURN(-ENOMEM);
 163                 }
 164                 if (test_and_set_bit(cl_idx, bitmap)) {
 165                         CERROR("FILTER client %d: found bit is set in bitmap\n",
 166                                cl_idx);
 167                         cl_idx = find_next_zero_bit(bitmap,
 168                                                     FILTER_LR_MAX_CLIENTS,
 169                                                     cl_idx);
 170                         goto repeat;
 171                 }
 172         } else {
 173                 if (test_and_set_bit(cl_idx, bitmap)) {
 174                         CERROR("FILTER client %d: bit already set in bitmap!\n",
 175                                cl_idx);
 176                         LBUG();
 177                 }
 178         }
 179
 180         fed->fed_lr_idx = cl_idx;
 181         fed->fed_lr_off = le32_to_cpu(filter->fo_fsd->fsd_client_start) +
 182                 cl_idx * le16_to_cpu(filter->fo_fsd->fsd_client_size);
 183
 184         CDEBUG(D_INFO, "client at index %d (%llu) with UUID '%s' added\n",
 185                fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
 186
 187         if (new_client) {
 188                 struct obd_run_ctxt saved;
 189                 loff_t off = fed->fed_lr_off;
 190                 int err;
 191                 void *handle;
 192
 193                 CDEBUG(D_INFO, "writing client fcd at idx %u (%llu) (len %u)\n",
 194                        fed->fed_lr_idx,off,(unsigned int)sizeof(*fed->fed_fcd));
 195
 196                 push_ctxt(&saved, &obd->obd_ctxt, NULL);
 197                 /* Transaction needed to fix bug 1403 */
 198                 handle = fsfilt_start(obd,
 199                                       filter->fo_rcvd_filp->f_dentry->d_inode,
 200                                       FSFILT_OP_SETATTR, NULL);
 201                 if (IS_ERR(handle)) {
 202                         err = PTR_ERR(handle);
 203                         CERROR("unable to start transaction: rc %d\n", err);
 204                 } else {
 205                         err = fsfilt_write_record(obd, filter->fo_rcvd_filp,
 206                                                   fed->fed_fcd,
 207                                                   sizeof(*fed->fed_fcd),
 208                                                   &off, 1);
 209                         fsfilt_commit(obd,
 210                                       filter->fo_rcvd_filp->f_dentry->d_inode,
 211                                       handle, 1);
 212                 }
 213                 pop_ctxt(&saved, &obd->obd_ctxt, NULL);
 214
 215                 if (err) {
 216                         CERROR("error writing %s client idx %u: rc %d\n",
 217                                LAST_RCVD, fed->fed_lr_idx, err);
 218                         RETURN(err);
 219                 }
 220         }
 221         RETURN(0);
 222 }
 223
 224 static int filter_client_free(struct obd_export *exp, int flags)
 225 {
 226         struct filter_export_data *fed = &exp->exp_filter_data;
 227         struct filter_obd *filter = &exp->exp_obd->u.filter;
 228         struct obd_device *obd = exp->exp_obd;
 229         struct filter_client_data zero_fcd;
 230         struct obd_run_ctxt saved;
 231         int rc;
 232         loff_t off;
 233         ENTRY;
 234
 235         if (fed->fed_fcd == NULL)
 236                 RETURN(0);
 237
 238         if (flags & OBD_OPT_FAILOVER)
 239                 GOTO(free, 0);
 240
 241         /* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
 242         if (strcmp(fed->fed_fcd->fcd_uuid, obd->obd_uuid.uuid ) == 0)
 243                 GOTO(free, 0);
 244
 245         LASSERT(filter->fo_last_rcvd_slots != NULL);
 246
 247         off = fed->fed_lr_off;
 248
 249         CDEBUG(D_INFO, "freeing client at idx %u (%lld) with UUID '%s'\n",
 250                fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
 251
 252         /* Clear the bit _after_ zeroing out the client so we don't
 253            race with filter_client_add and zero out new clients.*/
 254         if (!test_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
 255                 CERROR("FILTER client %u: bit already clear in bitmap!!\n",
 256                        fed->fed_lr_idx);
 257                 LBUG();
 258         }
 259
 260         memset(&zero_fcd, 0, sizeof zero_fcd);
 261         push_ctxt(&saved, &obd->obd_ctxt, NULL);
 262         rc = fsfilt_write_record(obd, filter->fo_rcvd_filp, &zero_fcd,
 263                                  sizeof(zero_fcd), &off, 1);
 264         pop_ctxt(&saved, &obd->obd_ctxt, NULL);
 265
 266         CDEBUG(rc == 0 ? D_INFO : D_ERROR,
 267                "zeroing disconnecting client %s at idx %u (%llu) in %s rc %d\n",
 268                fed->fed_fcd->fcd_uuid, fed->fed_lr_idx, fed->fed_lr_off,
 269                LAST_RCVD, rc);
 270
 271         if (!test_and_clear_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
 272                 CERROR("FILTER client %u: bit already clear in bitmap!!\n",
 273                        fed->fed_lr_idx);
 274                 LBUG();
 275         }
 276
 277 free:
 278         OBD_FREE(fed->fed_fcd, sizeof(*fed->fed_fcd));
 279
 280         RETURN(0);
 281 }
 282
 283 static int filter_free_server_data(struct filter_obd *filter)
 284 {
 285         OBD_FREE(filter->fo_fsd, sizeof(*filter->fo_fsd));
 286         filter->fo_fsd = NULL;
 287         OBD_FREE(filter->fo_last_rcvd_slots,
 288                  FILTER_LR_MAX_CLIENT_WORDS * sizeof(unsigned long));
 289         filter->fo_last_rcvd_slots = NULL;
 290         return 0;
 291 }
 292
 293 /* assumes caller is already in kernel ctxt */
 294 int filter_update_server_data(struct obd_device *obd, struct file *filp,
 295                               struct filter_server_data *fsd, int force_sync)
 296 {
 297         loff_t off = 0;
 298         int rc;
 299         ENTRY;
 300
 301         CDEBUG(D_INODE, "server uuid      : %s\n", fsd->fsd_uuid);
 302         CDEBUG(D_INODE, "server last_rcvd : "LPU64"\n",
 303                le64_to_cpu(fsd->fsd_last_transno));
 304         CDEBUG(D_INODE, "server last_mount: "LPU64"\n",
 305                le64_to_cpu(fsd->fsd_mount_count));
 306
 307         rc = fsfilt_write_record(obd, filp, fsd, sizeof(*fsd), &off,force_sync);
 308         if (rc)
 309                 CERROR("error writing filter_server_data: rc = %d\n", rc);
 310
 311         RETURN(rc);
 312 }
 313
 314 int filter_update_last_objid(struct obd_device *obd, obd_gr group,
 315                              int force_sync)
 316 {
 317         struct filter_obd *filter = &obd->u.filter;
 318         __u64 tmp;
 319         loff_t off = 0;
 320         int rc;
 321         ENTRY;
 322
 323         CDEBUG(D_INODE, "server last_objid for group "LPU64": "LPU64"\n",
 324                group, filter->fo_last_objids[group]);
 325
 326         tmp = cpu_to_le64(filter->fo_last_objids[group]);
 327         rc = fsfilt_write_record(obd, filter->fo_last_objid_files[group],
 328                                  &tmp, sizeof(tmp), &off, force_sync);
 329         if (rc)
 330                 CERROR("error writing group "LPU64" last objid: rc = %d\n",
 331                        group, rc);
 332         RETURN(rc);
 333 }
 334
 335 /* assumes caller has already in kernel ctxt */
 336 static int filter_init_server_data(struct obd_device *obd, struct file * filp)
 337 {
 338         struct filter_obd *filter = &obd->u.filter;
 339         struct filter_server_data *fsd;
 340         struct filter_client_data *fcd = NULL;
 341         struct inode *inode = filp->f_dentry->d_inode;
 342         unsigned long last_rcvd_size = inode->i_size;
 343         __u64 mount_count;
 344         int cl_idx;
 345         loff_t off = 0;
 346         int rc;
 347
 348         /* ensure padding in the struct is the correct size */
 349         LASSERT (offsetof(struct filter_server_data, fsd_padding) +
 350                  sizeof(fsd->fsd_padding) == FILTER_LR_SERVER_SIZE);
 351         LASSERT (offsetof(struct filter_client_data, fcd_padding) +
 352                  sizeof(fcd->fcd_padding) == FILTER_LR_CLIENT_SIZE);
 353
 354         OBD_ALLOC(fsd, sizeof(*fsd));
 355         if (!fsd)
 356                 RETURN(-ENOMEM);
 357         filter->fo_fsd = fsd;
 358
 359         OBD_ALLOC(filter->fo_last_rcvd_slots,
 360                   FILTER_LR_MAX_CLIENT_WORDS * sizeof(unsigned long));
 361         if (filter->fo_last_rcvd_slots == NULL) {
 362                 OBD_FREE(fsd, sizeof(*fsd));
 363                 RETURN(-ENOMEM);
 364         }
 365
 366         if (last_rcvd_size == 0) {
 367                 CWARN("%s: initializing new %s\n", obd->obd_name, LAST_RCVD);
 368
 369                 memcpy(fsd->fsd_uuid, obd->obd_uuid.uuid,sizeof(fsd->fsd_uuid));
 370                 fsd->fsd_last_transno = 0;
 371                 mount_count = fsd->fsd_mount_count = 0;
 372                 fsd->fsd_server_size = cpu_to_le32(FILTER_LR_SERVER_SIZE);
 373                 fsd->fsd_client_start = cpu_to_le32(FILTER_LR_CLIENT_START);
 374                 fsd->fsd_client_size = cpu_to_le16(FILTER_LR_CLIENT_SIZE);
 375                 fsd->fsd_subdir_count = cpu_to_le16(FILTER_SUBDIR_COUNT);
 376                 filter->fo_subdir_count = FILTER_SUBDIR_COUNT;
 377         } else {
 378                 rc = fsfilt_read_record(obd, filp, fsd, sizeof(*fsd), &off);
 379                 if (rc) {
 380                         CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
 381                                LAST_RCVD, rc);
 382                         GOTO(err_fsd, rc);
 383                 }
 384                 if (strcmp(fsd->fsd_uuid, obd->obd_uuid.uuid) != 0) {
 385                         CERROR("OBD UUID %s does not match last_rcvd UUID %s\n",
 386                                obd->obd_uuid.uuid, fsd->fsd_uuid);
 387                         GOTO(err_fsd, rc = -EINVAL);
 388                 }
 389                 mount_count = le64_to_cpu(fsd->fsd_mount_count);
 390                 filter->fo_subdir_count = le16_to_cpu(fsd->fsd_subdir_count);
 391         }
 392
 393         if (fsd->fsd_feature_incompat & ~cpu_to_le32(FILTER_INCOMPAT_SUPP)) {
 394                 CERROR("unsupported feature %x\n",
 395                        le32_to_cpu(fsd->fsd_feature_incompat) &
 396                        ~FILTER_INCOMPAT_SUPP);
 397                 GOTO(err_fsd, rc = -EINVAL);
 398         }
 399         if (fsd->fsd_feature_rocompat & ~cpu_to_le32(FILTER_ROCOMPAT_SUPP)) {
 400                 CERROR("read-only feature %x\n",
 401                        le32_to_cpu(fsd->fsd_feature_rocompat) &
 402                        ~FILTER_ROCOMPAT_SUPP);
 403                 /* Do something like remount filesystem read-only */
 404                 GOTO(err_fsd, rc = -EINVAL);
 405         }
 406
 407         CDEBUG(D_INODE, "%s: server last_rcvd : "LPU64"\n",
 408                obd->obd_name, le64_to_cpu(fsd->fsd_last_transno));
 409         CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
 410                obd->obd_name, mount_count + 1);
 411         CDEBUG(D_INODE, "%s: server data size: %u\n",
 412                obd->obd_name, le32_to_cpu(fsd->fsd_server_size));
 413         CDEBUG(D_INODE, "%s: per-client data start: %u\n",
 414                obd->obd_name, le32_to_cpu(fsd->fsd_client_start));
 415         CDEBUG(D_INODE, "%s: per-client data size: %u\n",
 416                obd->obd_name, le32_to_cpu(fsd->fsd_client_size));
 417         CDEBUG(D_INODE, "%s: server subdir_count: %u\n",
 418                obd->obd_name, le16_to_cpu(fsd->fsd_subdir_count));
 419         CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
 420                last_rcvd_size <= FILTER_LR_CLIENT_START ? 0 :
 421                (last_rcvd_size-FILTER_LR_CLIENT_START) /FILTER_LR_CLIENT_SIZE);
 422
 423         if (!obd->obd_replayable) {
 424                 CWARN("%s: recovery support OFF\n", obd->obd_name);
 425                 GOTO(out, rc = 0);
 426         }
 427
 428         for (cl_idx = 0, off = le32_to_cpu(fsd->fsd_client_start);
 429              off < last_rcvd_size; cl_idx++) {
 430                 __u64 last_rcvd;
 431                 struct obd_export *exp;
 432                 struct filter_export_data *fed;
 433
 434                 if (!fcd) {
 435                         OBD_ALLOC(fcd, sizeof(*fcd));
 436                         if (!fcd)
 437                                 GOTO(err_client, rc = -ENOMEM);
 438                 }
 439
 440                 /* Don't assume off is incremented properly by
 441                  * fsfilt_read_record(), in case sizeof(*fcd)
 442                  * isn't the same as fsd->fsd_client_size.  */
 443                 off = le32_to_cpu(fsd->fsd_client_start) +
 444                         cl_idx * le16_to_cpu(fsd->fsd_client_size);
 445                 rc = fsfilt_read_record(obd, filp, fcd, sizeof(*fcd), &off);
 446                 if (rc) {
 447                         CERROR("error reading FILT %s idx %d off %llu: rc %d\n",
 448                                LAST_RCVD, cl_idx, off, rc);
 449                         break; /* read error shouldn't cause startup to fail */
 450                 }
 451
 452                 if (fcd->fcd_uuid[0] == '\0') {
 453                         CDEBUG(D_INFO, "skipping zeroed client at offset %d\n",
 454                                cl_idx);
 455                         continue;
 456                 }
 457
 458                 last_rcvd = le64_to_cpu(fcd->fcd_last_rcvd);
 459
 460                 /* These exports are cleaned up by filter_disconnect(), so they
 461                  * need to be set up like real exports as filter_connect() does.
 462                  */
 463                 exp = class_new_export(obd);
 464                 CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64
 465                        " srv lr: "LPU64"\n", fcd->fcd_uuid, cl_idx,
 466                        last_rcvd, le64_to_cpu(fsd->fsd_last_transno));
 467                 if (exp == NULL)
 468                         GOTO(err_client, rc = -ENOMEM);
 469
 470                 memcpy(&exp->exp_client_uuid.uuid, fcd->fcd_uuid,
 471                        sizeof exp->exp_client_uuid.uuid);
 472                 fed = &exp->exp_filter_data;
 473                 fed->fed_fcd = fcd;
 474                 filter_client_add(obd, filter, fed, cl_idx);
 475                 /* create helper if export init gets more complex */
 476                 spin_lock_init(&fed->fed_lock);
 477
 478                 fcd = NULL;
 479                 obd->obd_recoverable_clients++;
 480                 class_export_put(exp);
 481
 482                 CDEBUG(D_OTHER, "client at idx %d has last_rcvd = "LPU64"\n",
 483                        cl_idx, last_rcvd);
 484
 485                 if (last_rcvd > le64_to_cpu(fsd->fsd_last_transno))
 486                         fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
 487
 488         }
 489
 490         obd->obd_last_committed = le64_to_cpu(fsd->fsd_last_transno);
 491
 492         if (obd->obd_recoverable_clients) {
 493                 CWARN("RECOVERY: %d recoverable clients, last_rcvd "
 494                        LPU64"\n", obd->obd_recoverable_clients,
 495                        le64_to_cpu(fsd->fsd_last_transno));
 496                 obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
 497                 obd->obd_recovering = 1;
 498         }
 499
 500         if (fcd)
 501                 OBD_FREE(fcd, sizeof(*fcd));
 502
 503 out:
 504         filter->fo_mount_count = mount_count + 1;
 505         fsd->fsd_mount_count = cpu_to_le64(filter->fo_mount_count);
 506
 507         /* save it, so mount count and last_transno is current */
 508         rc = filter_update_server_data(obd, filp, filter->fo_fsd, 1);
 509
 510         RETURN(rc);
 511
 512 err_client:
 513         class_disconnect_exports(obd, 0);
 514 err_fsd:
 515         filter_free_server_data(filter);
 516         RETURN(rc);
 517 }
 518
 519 static int filter_cleanup_groups(struct obd_device *obd)
 520 {
 521         struct filter_obd *filter = &obd->u.filter;
 522         int i;
 523         ENTRY;
 524
 525         if (filter->fo_dentry_O_groups != NULL &&
 526             filter->fo_last_objids != NULL &&
 527             filter->fo_last_objid_files != NULL) {
 528                 for (i = 0; i < FILTER_GROUPS; i++) {
 529                         struct dentry *dentry = filter->fo_dentry_O_groups[i];
 530                         struct file *filp = filter->fo_last_objid_files[i];
 531                         if (dentry != NULL) {
 532                                 f_dput(dentry);
 533                                 filter->fo_dentry_O_groups[i] = NULL;
 534                         }
 535                         if (filp != NULL) {
 536                                 filp_close(filp, 0);
 537                                 filter->fo_last_objid_files[i] = NULL;
 538                         }
 539                 }
 540         }
 541         if (filter->fo_dentry_O_sub != NULL && filter->fo_subdir_count) {
 542                 for (i = 0; i < filter->fo_subdir_count; i++) {
 543                         struct dentry *dentry = filter->fo_dentry_O_sub[i];
 544                         if (dentry != NULL) {
 545                                 f_dput(dentry);
 546                                 filter->fo_dentry_O_sub[i] = NULL;
 547                         }
 548                 }
 549                 OBD_FREE(filter->fo_dentry_O_sub,
 550                          filter->fo_subdir_count *
 551                          sizeof(*filter->fo_dentry_O_sub));
 552         }
 553         if (filter->fo_dentry_O_groups != NULL)
 554                 OBD_FREE(filter->fo_dentry_O_groups,
 555                          FILTER_GROUPS * sizeof(struct dentry *));
 556         if (filter->fo_last_objids != NULL)
 557                 OBD_FREE(filter->fo_last_objids,
 558                          FILTER_GROUPS * sizeof(__u64));
 559         if (filter->fo_last_objid_files != NULL)
 560                 OBD_FREE(filter->fo_last_objid_files,
 561                          FILTER_GROUPS * sizeof(struct file *));
 562         RETURN(0);
 563 }
 564
 565 /* FIXME: object groups */
 566 static int filter_prep_groups(struct obd_device *obd)
 567 {
 568         struct filter_obd *filter = &obd->u.filter;
 569         struct dentry *dentry, *O_dentry;
 570         struct file *filp;
 571         int i, rc = 0, cleanup_phase = 0;
 572         ENTRY;
 573
 574         O_dentry = simple_mkdir(current->fs->pwd, "O", 0700, 1);
 575         CDEBUG(D_INODE, "got/created O: %p\n", O_dentry);
 576         if (IS_ERR(O_dentry)) {
 577                 rc = PTR_ERR(O_dentry);
 578                 CERROR("cannot open/create O: rc = %d\n", rc);
 579                 GOTO(cleanup, rc);
 580         }
 581         filter->fo_dentry_O = O_dentry;
 582         cleanup_phase = 1; /* O_dentry */
 583
 584         /* Lookup "R" to tell if we're on an old OST FS and need to convert
 585          * from O/R/<dir>/<objid> to O/0/<dir>/<objid>.  This can be removed
 586          * some time post 1.0 when all old-style OSTs have converted along
 587          * with the init_objid hack. */
 588         dentry = ll_lookup_one_len("R", O_dentry, 1);
 589         if (IS_ERR(dentry))
 590                 GOTO(cleanup, rc = PTR_ERR(dentry));
 591         if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
 592                 struct dentry *O0_dentry = lookup_one_len("0", O_dentry, 1);
 593                 ENTRY;
 594
 595                 CWARN("converting OST to new object layout\n");
 596                 if (IS_ERR(O0_dentry)) {
 597                         rc = PTR_ERR(O0_dentry);
 598                         CERROR("error looking up O/0: rc %d\n", rc);
 599                         GOTO(cleanup_R, rc);
 600                 }
 601
 602                 if (O0_dentry->d_inode) {
 603                         CERROR("Both O/R and O/0 exist. Fix manually.\n");
 604                         GOTO(cleanup_O0, rc = -EEXIST);
 605                 }
 606
 607                 down(&O_dentry->d_inode->i_sem);
 608                 rc = vfs_rename(O_dentry->d_inode, dentry,
 609                                 O_dentry->d_inode, O0_dentry);
 610                 up(&O_dentry->d_inode->i_sem);
 611
 612                 if (rc) {
 613                         CERROR("error renaming O/R to O/0: rc %d\n", rc);
 614                         GOTO(cleanup_O0, rc);
 615                 }
 616                 filter->fo_fsd->fsd_feature_incompat |=
 617                         cpu_to_le32(FILTER_INCOMPAT_GROUPS);
 618                 rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
 619                                                filter->fo_fsd, 1);
 620                 GOTO(cleanup_O0, rc);
 621
 622         cleanup_O0:
 623                 dput(O0_dentry);
 624         cleanup_R:
 625                 dput(dentry);
 626                 if (rc)
 627                         GOTO(cleanup, rc);
 628         } else {
 629                 dput(dentry);
 630         }
 631
 632         OBD_ALLOC(filter->fo_last_objids, FILTER_GROUPS * sizeof(__u64));
 633         if (filter->fo_last_objids == NULL)
 634                 GOTO(cleanup, rc = -ENOMEM);
 635         cleanup_phase = 2; /* groups */
 636
 637         OBD_ALLOC(filter->fo_dentry_O_groups, FILTER_GROUPS * sizeof(dentry));
 638         if (filter->fo_dentry_O_groups == NULL)
 639                 GOTO(cleanup, rc = -ENOMEM);
 640         OBD_ALLOC(filter->fo_last_objid_files, FILTER_GROUPS * sizeof(filp));
 641         if (filter->fo_last_objid_files == NULL)
 642                 GOTO(cleanup, rc = -ENOMEM);
 643
 644         for (i = 0; i < FILTER_GROUPS; i++) {
 645                 char name[25];
 646                 loff_t off = 0;
 647
 648                 sprintf(name, "%d", i);
 649                 dentry = simple_mkdir(O_dentry, name, 0700, 1);
 650                 CDEBUG(D_INODE, "got/created O/%s: %p\n", name, dentry);
 651                 if (IS_ERR(dentry)) {
 652                         rc = PTR_ERR(dentry);
 653                         CERROR("cannot lookup/create O/%s: rc = %d\n",
 654                                name, rc);
 655                         GOTO(cleanup, rc);
 656                 }
 657                 filter->fo_dentry_O_groups[i] = dentry;
 658
 659                 sprintf(name, "O/%d/LAST_ID", i);
 660                 filp = filp_open(name, O_CREAT | O_RDWR, 0700);
 661                 if (IS_ERR(dentry)) {
 662                         rc = PTR_ERR(dentry);
 663                         CERROR("cannot create %s: rc = %d\n", name, rc);
 664                         GOTO(cleanup, rc);
 665                 }
 666                 filter->fo_last_objid_files[i] = filp;
 667
 668                 if (filp->f_dentry->d_inode->i_size == 0) {
 669                         if (i == 0 && filter->fo_fsd->fsd_unused != 0) {
 670                                 /* OST conversion, remove sometime post 1.0 */
 671                                 filter->fo_last_objids[i] =
 672                                         le64_to_cpu(filter->fo_fsd->fsd_unused);
 673                                 CWARN("saving old objid "LPU64" to LAST_ID\n",
 674                                       filter->fo_last_objids[i]);
 675                                 rc = filter_update_last_objid(obd, 0, 1);
 676                                 if (rc)
 677                                         GOTO(cleanup, rc);
 678                         } else {
 679                                 filter->fo_last_objids[i] = FILTER_INIT_OBJID;
 680                         }
 681                         continue;
 682                 }
 683
 684                 rc = fsfilt_read_record(obd, filp, &filter->fo_last_objids[i],
 685                                         sizeof(__u64), &off);
 686                 if (rc) {
 687                         CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
 688                                name, rc);
 689                         GOTO(cleanup, rc);
 690                 }
 691                 filter->fo_last_objids[i] =
 692                         le64_to_cpu(filter->fo_last_objids[i]);
 693                 CDEBUG(D_INODE, "%s: server last_objid group %d: "LPU64"\n",
 694                        obd->obd_name, i, filter->fo_last_objids[i]);
 695         }
 696
 697         if (filter->fo_subdir_count) {
 698                 O_dentry = filter->fo_dentry_O_groups[0];
 699                 OBD_ALLOC(filter->fo_dentry_O_sub,
 700                           filter->fo_subdir_count * sizeof(dentry));
 701                 if (filter->fo_dentry_O_sub == NULL)
 702                         GOTO(cleanup, rc = -ENOMEM);
 703
 704                 for (i = 0; i < filter->fo_subdir_count; i++) {
 705                         char dir[20];
 706                         snprintf(dir, sizeof(dir), "d%u", i);
 707
 708                         dentry = simple_mkdir(O_dentry, dir, 0700, 1);
 709                         CDEBUG(D_INODE, "got/created O/0/%s: %p\n", dir,dentry);
 710                         if (IS_ERR(dentry)) {
 711                                 rc = PTR_ERR(dentry);
 712                                 CERROR("can't lookup/create O/0/%s: rc = %d\n",
 713                                        dir, rc);
 714                                 GOTO(cleanup, rc);
 715                         }
 716                         filter->fo_dentry_O_sub[i] = dentry;
 717                 }
 718         }
 719         RETURN(0);
 720
 721  cleanup:
 722         switch (cleanup_phase) {
 723         case 2:
 724                 filter_cleanup_groups(obd);
 725         case 1:
 726                 f_dput(filter->fo_dentry_O);
 727                 filter->fo_dentry_O = NULL;
 728         default:
 729                 break;
 730         }
 731         return rc;
 732 }
 733
 734 /* setup the object store with correct subdirectories */
 735 static int filter_prep(struct obd_device *obd)
 736 {
 737         struct obd_run_ctxt saved;
 738         struct filter_obd *filter = &obd->u.filter;
 739         struct file *file;
 740         struct inode *inode;
 741         int rc = 0;
 742         ENTRY;
 743
 744         push_ctxt(&saved, &obd->obd_ctxt, NULL);
 745         file = filp_open(LAST_RCVD, O_RDWR | O_CREAT | O_LARGEFILE, 0700);
 746         if (!file || IS_ERR(file)) {
 747                 rc = PTR_ERR(file);
 748                 CERROR("OBD filter: cannot open/create %s: rc = %d\n",
 749                        LAST_RCVD, rc);
 750                 GOTO(out, rc);
 751         }
 752
 753         if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
 754                 CERROR("%s is not a regular file!: mode = %o\n", LAST_RCVD,
 755                        file->f_dentry->d_inode->i_mode);
 756                 GOTO(err_filp, rc = -ENOENT);
 757         }
 758
 759         /* steal operations */
 760         inode = file->f_dentry->d_inode;
 761         filter->fo_fop = file->f_op;
 762         filter->fo_iop = inode->i_op;
 763         filter->fo_aops = inode->i_mapping->a_ops;
 764
 765         rc = filter_init_server_data(obd, file);
 766         if (rc) {
 767                 CERROR("cannot read %s: rc = %d\n", LAST_RCVD, rc);
 768                 GOTO(err_filp, rc);
 769         }
 770         filter->fo_rcvd_filp = file;
 771
 772         rc = filter_prep_groups(obd);
 773         if (rc)
 774                 GOTO(err_server_data, rc);
 775
 776  out:
 777         pop_ctxt(&saved, &obd->obd_ctxt, NULL);
 778
 779         return(rc);
 780
 781  err_server_data:
 782         //class_disconnect_exports(obd, 0);
 783         filter_free_server_data(filter);
 784  err_filp:
 785         if (filp_close(file, 0))
 786                 CERROR("can't close %s after error\n", LAST_RCVD);
 787         filter->fo_rcvd_filp = NULL;
 788         goto out;
 789 }
 790
 791 /* cleanup the filter: write last used object id to status file */
 792 static void filter_post(struct obd_device *obd)
 793 {
 794         struct obd_run_ctxt saved;
 795         struct filter_obd *filter = &obd->u.filter;
 796         int rc, i;
 797
 798         /* XXX: filter_update_lastobjid used to call fsync_dev.  It might be
 799          * best to start a transaction with h_sync, because we removed this
 800          * from lastobjid */
 801
 802         push_ctxt(&saved, &obd->obd_ctxt, NULL);
 803         rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
 804                                        filter->fo_fsd, 0);
 805         if (rc)
 806                 CERROR("error writing server data: rc = %d\n", rc);
 807
 808         for (i = 0; i < FILTER_GROUPS; i++) {
 809                 rc = filter_update_last_objid(obd, i, (i == FILTER_GROUPS - 1));
 810                 if (rc)
 811                         CERROR("error writing group %d lastobjid: rc = %d\n",
 812                                i, rc);
 813         }
 814
 815         filp_close(filter->fo_rcvd_filp, 0);
 816         filter->fo_rcvd_filp = NULL;
 817         if (rc)
 818                 CERROR("error closing %s: rc = %d\n", LAST_RCVD, rc);
 819
 820         filter_cleanup_groups(obd);
 821         f_dput(filter->fo_dentry_O);
 822         filter_free_server_data(filter);
 823         pop_ctxt(&saved, &obd->obd_ctxt, NULL);
 824 }
 825
 826 static void filter_set_last_id(struct filter_obd *filter, struct obdo *oa,
 827                                obd_id id)
 828 {
 829         obd_gr group = 0;
 830         LASSERT(filter->fo_fsd != NULL);
 831
 832         if (oa != NULL) {
 833                 LASSERT(oa->o_gr <= FILTER_GROUPS);
 834                 group = oa->o_gr;
 835         }
 836
 837         spin_lock(&filter->fo_objidlock);
 838         filter->fo_last_objids[group] = id;
 839         spin_unlock(&filter->fo_objidlock);
 840 }
 841
 842 __u64 filter_last_id(struct filter_obd *filter, struct obdo *oa)
 843 {
 844         obd_id id;
 845         obd_gr group = 0;
 846         LASSERT(filter->fo_fsd != NULL);
 847
 848         if (oa != NULL) {
 849                 LASSERT(oa->o_gr <= FILTER_GROUPS);
 850                 group = oa->o_gr;
 851         }
 852
 853         /* FIXME: object groups */
 854         spin_lock(&filter->fo_objidlock);
 855         id = filter->fo_last_objids[group];
 856         spin_unlock(&filter->fo_objidlock);
 857
 858         return id;
 859 }
 860
 861 /* direct cut-n-paste of mds_blocking_ast() */
 862 static int filter_blocking_ast(struct ldlm_lock *lock,
 863                                struct ldlm_lock_desc *desc,
 864                                void *data, int flag)
 865 {
 866         int do_ast;
 867         ENTRY;
 868
 869         if (flag == LDLM_CB_CANCELING) {
 870                 /* Don't need to do anything here. */
 871                 RETURN(0);
 872         }
 873
 874         /* XXX layering violation!  -phil */
 875         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 876         /* Get this: if filter_blocking_ast is racing with ldlm_intent_policy,
 877          * such that filter_blocking_ast is called just before l_i_p takes the
 878          * ns_lock, then by the time we get the lock, we might not be the
 879          * correct blocking function anymore.  So check, and return early, if
 880          * so. */
 881         if (lock->l_blocking_ast != filter_blocking_ast) {
 882                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 883                 RETURN(0);
 884         }
 885
 886         lock->l_flags |= LDLM_FL_CBPENDING;
 887         do_ast = (!lock->l_readers && !lock->l_writers);
 888         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 889
 890         if (do_ast) {
 891                 struct lustre_handle lockh;
 892                 int rc;
 893
 894                 LDLM_DEBUG(lock, "already unused, calling ldlm_cli_cancel");
 895                 ldlm_lock2handle(lock, &lockh);
 896                 rc = ldlm_cli_cancel(&lockh);
 897                 if (rc < 0)
 898                         CERROR("ldlm_cli_cancel: %d\n", rc);
 899         } else {
 900                 LDLM_DEBUG(lock, "Lock still has references, will be "
 901                            "cancelled later");
 902         }
 903         RETURN(0);
 904 }
 905
 906 static int filter_lock_dentry(struct obd_device *obd, struct dentry *dparent)
 907 {
 908         down(&dparent->d_inode->i_sem);
 909         return 0;
 910 }
 911
 912 /* We never dget the object parent, so DON'T dput it either */
 913 static void filter_parent_unlock(struct dentry *dparent)
 914 {
 915         up(&dparent->d_inode->i_sem);
 916 }
 917
 918 /* We never dget the object parent, so DON'T dput it either */
 919 struct dentry *filter_parent(struct obd_device *obd, obd_gr group, obd_id objid)
 920 {
 921         struct filter_obd *filter = &obd->u.filter;
 922         LASSERT(group < FILTER_GROUPS); /* FIXME: object groups */
 923
 924         if (group > 0 || filter->fo_subdir_count == 0)
 925                 return filter->fo_dentry_O_groups[group];
 926
 927         return filter->fo_dentry_O_sub[objid & (filter->fo_subdir_count - 1)];
 928 }
 929
 930 /* We never dget the object parent, so DON'T dput it either */
 931 struct dentry *filter_parent_lock(struct obd_device *obd, obd_gr group,
 932                                   obd_id objid)
 933 {
 934         unsigned long now = jiffies;
 935         struct dentry *dparent = filter_parent(obd, group, objid);
 936         int rc;
 937
 938         if (IS_ERR(dparent))
 939                 return dparent;
 940
 941         rc = filter_lock_dentry(obd, dparent);
 942         if (time_after(jiffies, now + 15 * HZ))
 943                 CERROR("slow parent lock %lus\n", (jiffies - now) / HZ);
 944         return rc ? ERR_PTR(rc) : dparent;
 945 }
 946
 947 /* How to get files, dentries, inodes from object id's.
 948  *
 949  * If dir_dentry is passed, the caller has already locked the parent
 950  * appropriately for this operation (normally a write lock).  If
 951  * dir_dentry is NULL, we do a read lock while we do the lookup to
 952  * avoid races with create/destroy and such changing the directory
 953  * internal to the filesystem code. */
 954 struct dentry *filter_fid2dentry(struct obd_device *obd,
 955                                  struct dentry *dir_dentry,
 956                                  obd_gr group, obd_id id)
 957 {
 958         struct dentry *dparent = dir_dentry;
 959         struct dentry *dchild;
 960         char name[32];
 961         int len;
 962         ENTRY;
 963
 964         if (id == 0) {
 965                 CERROR("fatal: invalid object id 0\n");
 966                 RETURN(ERR_PTR(-ESTALE));
 967         }
 968
 969         len = sprintf(name, LPU64, id);
 970         if (dir_dentry == NULL) {
 971                 dparent = filter_parent_lock(obd, group, id);
 972                 if (IS_ERR(dparent))
 973                         RETURN(dparent);
 974         }
 975         CDEBUG(D_INODE, "looking up object O/%*s/%s\n",
 976                dparent->d_name.len, dparent->d_name.name, name);
 977         dchild = /*ll_*/lookup_one_len(name, dparent, len);
 978         if (dir_dentry == NULL)
 979                 filter_parent_unlock(dparent);
 980         if (IS_ERR(dchild)) {
 981                 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
 982                 RETURN(dchild);
 983         }
 984
 985         CDEBUG(D_INODE, "got child objid %s: %p, count = %d\n",
 986                name, dchild, atomic_read(&dchild->d_count));
 987
 988         LASSERT(atomic_read(&dchild->d_count) > 0);
 989
 990         RETURN(dchild);
 991 }
 992
 993 static int filter_prepare_destroy(struct obd_device *obd, obd_id objid)
 994 {
 995         struct lustre_handle lockh;
 996         int flags = LDLM_AST_DISCARD_DATA, rc;
 997         struct ldlm_res_id res_id = { .name = { objid } };
 998         ldlm_policy_data_t policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
 999
1000         ENTRY;
1001         /* Tell the clients that the object is gone now and that they should
1002          * throw away any cached pages. */
1003         rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, res_id,
1004                               LDLM_EXTENT, &policy, LCK_PW,
1005                               &flags, filter_blocking_ast, ldlm_completion_ast,
1006                               NULL, NULL, NULL, 0, NULL, &lockh);
1007
1008         /* We only care about the side-effects, just drop the lock. */
1009         if (rc == ELDLM_OK)
1010                 ldlm_lock_decref(&lockh, LCK_PW);
1011
1012         RETURN(rc);
1013 }
1014
1015 /* Caller must hold LCK_PW on parent and push us into kernel context.
1016  * Caller is also required to ensure that dchild->d_inode exists. */
1017 static int filter_destroy_internal(struct obd_device *obd, obd_id objid,
1018                                    struct dentry *dparent,
1019                                    struct dentry *dchild)
1020 {
1021         struct inode *inode = dchild->d_inode;
1022         int rc;
1023         ENTRY;
1024
1025         if (inode->i_nlink != 1 || atomic_read(&inode->i_count) != 1) {
1026                 CERROR("destroying objid %*s nlink = %lu, count = %d\n",
1027                        dchild->d_name.len, dchild->d_name.name,
1028                        (unsigned long)inode->i_nlink,
1029                        atomic_read(&inode->i_count));
1030         }
1031
1032         rc = vfs_unlink(dparent->d_inode, dchild);
1033
1034         if (rc)
1035                 CERROR("error unlinking objid %*s: rc %d\n",
1036                        dchild->d_name.len, dchild->d_name.name, rc);
1037
1038         RETURN(rc);
1039 }
1040
1041 static int filter_intent_policy(struct ldlm_namespace *ns,
1042                                 struct ldlm_lock **lockp, void *req_cookie,
1043                                 ldlm_mode_t mode, int flags, void *data)
1044 {
1045         struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
1046         struct ptlrpc_request *req = req_cookie;
1047         struct ldlm_lock *lock = *lockp, *l = NULL;
1048         struct ldlm_resource *res = lock->l_resource;
1049         ldlm_processing_policy policy;
1050         struct ost_lvb *res_lvb, *reply_lvb;
1051         struct list_head *tmp;
1052         ldlm_error_t err;
1053         int tmpflags = 0, rc, repsize[2] = {sizeof(struct ldlm_reply),
1054                                             sizeof(struct ost_lvb) };
1055         ENTRY;
1056
1057         policy = ldlm_get_processing_policy(res);
1058         LASSERT(policy != NULL);
1059         LASSERT(req != NULL);
1060
1061         rc = lustre_pack_reply(req, 2, repsize, NULL);
1062         if (rc)
1063                 RETURN(req->rq_status = rc);
1064
1065         reply_lvb = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*reply_lvb));
1066         LASSERT(reply_lvb != NULL);
1067
1068         //fixup_handle_for_resent_req(req, lock, &lockh);
1069
1070         /* If we grant any lock at all, it will be a whole-file read lock.
1071          * Call the extent policy function to see if our request can be
1072          * granted, or is blocked. */
1073         lock->l_policy_data.l_extent.start = 0;
1074         lock->l_policy_data.l_extent.end = OBD_OBJECT_EOF;
1075         lock->l_req_mode = LCK_PR;
1076
1077         l_lock(&res->lr_namespace->ns_lock);
1078
1079         res->lr_tmp = &rpc_list;
1080         rc = policy(lock, &tmpflags, 0, &err);
1081         res->lr_tmp = NULL;
1082
1083         /* FIXME: we should change the policy function slightly, to not make
1084          * this list at all, since we just turn around and free it */
1085         while (!list_empty(&rpc_list)) {
1086                 struct ldlm_ast_work *w =
1087                         list_entry(rpc_list.next, struct ldlm_ast_work, w_list);
1088                 list_del(&w->w_list);
1089                 LDLM_LOCK_PUT(w->w_lock);
1090                 OBD_FREE(w, sizeof(*w));
1091         }
1092
1093         if (rc == LDLM_ITER_CONTINUE) {
1094                 /* The lock met with no resistance; we're finished. */
1095                 l_unlock(&res->lr_namespace->ns_lock);
1096                 RETURN(ELDLM_LOCK_REPLACED);
1097         }
1098
1099         /* Do not grant any lock, but instead send GL callbacks.  The extent
1100          * policy nicely created a list of all PW locks for us.  We will choose
1101          * the highest of those which are larger than the size in the LVB, if
1102          * any, and perform a glimpse callback. */
1103         down(&res->lr_lvb_sem);
1104         res_lvb = res->lr_lvb_data;
1105         LASSERT(res_lvb != NULL);
1106         reply_lvb->lvb_size = res_lvb->lvb_size;
1107         up(&res->lr_lvb_sem);
1108
1109         list_for_each(tmp, &res->lr_granted) {
1110                 struct ldlm_lock *tmplock =
1111                         list_entry(tmp, struct ldlm_lock, l_res_link);
1112
1113                 if (tmplock->l_granted_mode == LCK_PR)
1114                         continue;
1115
1116                 if (tmplock->l_policy_data.l_extent.end <=
1117                     reply_lvb->lvb_size)
1118                         continue;
1119
1120                 if (l == NULL) {
1121                         l = LDLM_LOCK_GET(tmplock);
1122                         continue;
1123                 }
1124
1125                 if (l->l_policy_data.l_extent.start >
1126                     tmplock->l_policy_data.l_extent.start)
1127                         continue;
1128
1129                 LDLM_LOCK_PUT(l);
1130                 l = LDLM_LOCK_GET(tmplock);
1131         }
1132         l_unlock(&res->lr_namespace->ns_lock);
1133
1134         /* There were no PW locks beyond the size in the LVB; finished. */
1135         if (l == NULL)
1136                 RETURN(ELDLM_LOCK_ABORTED);
1137
1138         LASSERT(l->l_glimpse_ast != NULL);
1139         rc = l->l_glimpse_ast(l, NULL); /* this will update the LVB */
1140         if (rc != 0 && res->lr_namespace->ns_lvbo &&
1141             res->lr_namespace->ns_lvbo->lvbo_update) {
1142                 res->lr_namespace->ns_lvbo->lvbo_update(res, NULL, 0, 1);
1143         }
1144
1145         down(&res->lr_lvb_sem);
1146         reply_lvb->lvb_size = res_lvb->lvb_size;
1147         up(&res->lr_lvb_sem);
1148
1149         LDLM_LOCK_PUT(l);
1150
1151         RETURN(ELDLM_LOCK_ABORTED);
1152 }
1153
1154 /* mount the file system (secretly) */
1155 int filter_common_setup(struct obd_device *obd, obd_count len, void *buf,
1156                         char *option)
1157 {
1158         struct lustre_cfg* lcfg = buf;
1159         struct filter_obd *filter = &obd->u.filter;
1160         struct vfsmount *mnt;
1161         int rc = 0;
1162         ENTRY;
1163
1164         dev_clear_rdonly(2);
1165
1166         if (!lcfg->lcfg_inlbuf1 || !lcfg->lcfg_inlbuf2)
1167                 RETURN(-EINVAL);
1168
1169         obd->obd_fsops = fsfilt_get_ops(lcfg->lcfg_inlbuf2);
1170         if (IS_ERR(obd->obd_fsops))
1171                 RETURN(PTR_ERR(obd->obd_fsops));
1172
1173         mnt = do_kern_mount(lcfg->lcfg_inlbuf2, MS_NOATIME | MS_NODIRATIME,
1174                             lcfg->lcfg_inlbuf1, option);
1175         rc = PTR_ERR(mnt);
1176         if (IS_ERR(mnt))
1177                 GOTO(err_ops, rc);
1178
1179         if (lcfg->lcfg_inllen3 > 0 && lcfg->lcfg_inlbuf3) {
1180                 if (*lcfg->lcfg_inlbuf3 == 'f') {
1181                         obd->obd_replayable = 1;
1182                         obd_sync_filter = 1;
1183                         CWARN("%s: recovery enabled\n", obd->obd_name);
1184                 } else {
1185                         if (*lcfg->lcfg_inlbuf3 != 'n') {
1186                                 CERROR("unrecognised flag '%c'\n",
1187                                        *lcfg->lcfg_inlbuf3);
1188                         }
1189                         // XXX Robert? Why do we get errors here
1190                         // GOTO(err_mntput, rc = -EINVAL);
1191                 }
1192         }
1193
1194         filter->fo_vfsmnt = mnt;
1195         filter->fo_sb = mnt->mnt_sb;
1196         filter->fo_fstype = mnt->mnt_sb->s_type->name;
1197         CDEBUG(D_SUPER, "%s: mnt = %p\n", filter->fo_fstype, mnt);
1198
1199         OBD_SET_CTXT_MAGIC(&obd->obd_ctxt);
1200         obd->obd_ctxt.pwdmnt = mnt;
1201         obd->obd_ctxt.pwd = mnt->mnt_root;
1202         obd->obd_ctxt.fs = get_ds();
1203         obd->obd_ctxt.cb_ops = filter_lvfs_ops;
1204
1205         rc = filter_prep(obd);
1206         if (rc)
1207                 GOTO(err_mntput, rc);
1208
1209         spin_lock_init(&filter->fo_translock);
1210         spin_lock_init(&filter->fo_objidlock);
1211         INIT_LIST_HEAD(&filter->fo_export_list);
1212         sema_init(&filter->fo_alloc_lock, 1);
1213         spin_lock_init(&filter->fo_r_pages.oh_lock);
1214         spin_lock_init(&filter->fo_w_pages.oh_lock);
1215         spin_lock_init(&filter->fo_r_discont_pages.oh_lock);
1216         spin_lock_init(&filter->fo_w_discont_pages.oh_lock);
1217         spin_lock_init(&filter->fo_r_discont_blocks.oh_lock);
1218         spin_lock_init(&filter->fo_w_discont_blocks.oh_lock);
1219         filter->fo_readcache_max_filesize = FILTER_MAX_CACHE_SIZE;
1220
1221         obd->obd_namespace = ldlm_namespace_new("filter-tgt",
1222                                                 LDLM_NAMESPACE_SERVER);
1223         if (obd->obd_namespace == NULL)
1224                 GOTO(err_post, rc = -ENOMEM);
1225         obd->obd_namespace->ns_lvbp = obd;
1226         obd->obd_namespace->ns_lvbo = &filter_lvbo;
1227         ldlm_register_intent(obd->obd_namespace, filter_intent_policy);
1228
1229         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1230                            "filter_ldlm_cb_client", &obd->obd_ldlm_client);
1231
1232         rc = llog_cat_initialize(obd, 1);
1233         if (rc) {
1234                 CERROR("failed to setup llogging subsystems\n");
1235                 GOTO(err_post, rc);
1236         }
1237
1238         RETURN(0);
1239
1240 err_post:
1241         filter_post(obd);
1242 err_mntput:
1243         unlock_kernel();
1244         mntput(mnt);
1245         filter->fo_sb = 0;
1246         lock_kernel();
1247 err_ops:
1248         fsfilt_put_ops(obd->obd_fsops);
1249         return rc;
1250 }
1251
1252 static int filter_setup(struct obd_device *obd, obd_count len, void *buf)
1253 {
1254         struct lustre_cfg* lcfg = buf;
1255         const char *str = NULL;
1256         char *option = NULL;
1257         int n = 0;
1258         int rc;
1259
1260         if (!strcmp(lcfg->lcfg_inlbuf2, "ext3")) {
1261 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1262         /* bug 1577: implement async-delete for 2.5 */
1263                 str = "errors=remount-ro,asyncdel";
1264 #else
1265                 str = "errors=remount-ro";
1266 #endif
1267                 n = strlen(str) + 1;
1268                 OBD_ALLOC(option, n);
1269                 if (option == NULL)
1270                         RETURN(-ENOMEM);
1271                 strcpy(option, str);
1272         }
1273
1274         rc = filter_common_setup(obd, len, buf, option);
1275         if (option)
1276                 OBD_FREE(option, n);
1277         return rc;
1278 }
1279
1280 static int filter_cleanup(struct obd_device *obd, int flags)
1281 {
1282         struct filter_obd *filter = &obd->u.filter;
1283         ENTRY;
1284
1285         if (flags & OBD_OPT_FAILOVER)
1286                 CERROR("%s: shutting down for failover; client state will"
1287                        " be preserved.\n", obd->obd_name);
1288
1289         if (!list_empty(&obd->obd_exports)) {
1290                 CERROR("%s: still has clients!\n", obd->obd_name);
1291                 class_disconnect_exports(obd, flags);
1292                 if (!list_empty(&obd->obd_exports)) {
1293                         CERROR("still has exports after forced cleanup?\n");
1294                         RETURN(-EBUSY);
1295                 }
1296         }
1297
1298         ldlm_namespace_free(obd->obd_namespace, flags & OBD_OPT_FORCE);
1299
1300         if (filter->fo_sb == NULL)
1301                 RETURN(0);
1302
1303         filter_post(obd);
1304
1305         shrink_dcache_parent(filter->fo_sb->s_root);
1306         filter->fo_sb = 0;
1307
1308         if (atomic_read(&filter->fo_vfsmnt->mnt_count) > 1)
1309                 CERROR("%s: mount point %p busy, mnt_count: %d\n",
1310                        obd->obd_name, filter->fo_vfsmnt,
1311                        atomic_read(&filter->fo_vfsmnt->mnt_count));
1312
1313         unlock_kernel();
1314         mntput(filter->fo_vfsmnt);
1315         //destroy_buffers(filter->fo_sb->s_dev);
1316         filter->fo_sb = NULL;
1317         fsfilt_put_ops(obd->obd_fsops);
1318         lock_kernel();
1319
1320         dev_clear_rdonly(2);
1321
1322         RETURN(0);
1323 }
1324
1325 static int filter_attach(struct obd_device *obd, obd_count len, void *data)
1326 {
1327         struct lprocfs_static_vars lvars;
1328         int rc;
1329
1330         lprocfs_init_vars(filter, &lvars);
1331         rc = lprocfs_obd_attach(obd, lvars.obd_vars);
1332         if (rc != 0)
1333                 return rc;
1334
1335         rc = lprocfs_alloc_obd_stats(obd, LPROC_FILTER_LAST);
1336         if (rc != 0)
1337                 return rc;
1338
1339         /* Init obdfilter private stats here */
1340         lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_READ_BYTES,
1341                              LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes");
1342         lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_WRITE_BYTES,
1343                              LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes");
1344
1345         return lproc_filter_attach_seqstat(obd);
1346 }
1347
1348 static int filter_detach(struct obd_device *dev)
1349 {
1350         lprocfs_free_obd_stats(dev);
1351         return lprocfs_obd_detach(dev);
1352 }
1353
1354 /* nearly identical to mds_connect */
1355 static int filter_connect(struct lustre_handle *conn, struct obd_device *obd,
1356                           struct obd_uuid *cluuid)
1357 {
1358         struct obd_export *exp;
1359         struct filter_export_data *fed;
1360         struct filter_client_data *fcd = NULL;
1361         struct filter_obd *filter = &obd->u.filter;
1362         int rc;
1363         ENTRY;
1364
1365         if (conn == NULL || obd == NULL || cluuid == NULL)
1366                 RETURN(-EINVAL);
1367
1368         rc = class_connect(conn, obd, cluuid);
1369         if (rc)
1370                 RETURN(rc);
1371         exp = class_conn2export(conn);
1372         LASSERT(exp != NULL);
1373
1374         fed = &exp->exp_filter_data;
1375
1376         spin_lock_init(&fed->fed_lock);
1377
1378         if (!obd->obd_replayable)
1379                 GOTO(cleanup, rc = 0);
1380
1381         OBD_ALLOC(fcd, sizeof(*fcd));
1382         if (!fcd) {
1383                 CERROR("filter: out of memory for client data\n");
1384                 GOTO(cleanup, rc = -ENOMEM);
1385         }
1386
1387         memcpy(fcd->fcd_uuid, cluuid, sizeof(fcd->fcd_uuid));
1388         fed->fed_fcd = fcd;
1389
1390         rc = filter_client_add(obd, filter, fed, -1);
1391
1392 cleanup:
1393         if (rc) {
1394                 if (fcd)
1395                         OBD_FREE(fcd, sizeof(*fcd));
1396                 class_disconnect(exp, 0);
1397         } else {
1398                 class_export_put(exp);
1399         }
1400         return rc;
1401 }
1402
1403 static int filter_precleanup(struct obd_device *obd, int flags)
1404 {
1405         int rc = 0;
1406         ENTRY;
1407
1408         rc = obd_llog_finish(obd, 0);
1409         if (rc)
1410                 CERROR("failed to cleanup llogging subsystem\n");
1411
1412         RETURN(rc);
1413 }
1414
1415 /* Do extra sanity checks for grant accounting.  We do this at connect,
1416  * disconnect, and statfs RPC time, so it shouldn't be too bad.  We can
1417  * always get rid of it or turn it off when we know accounting is good. */
1418 static void filter_grant_sanity_check(struct obd_device *obd, char *func)
1419 {
1420         struct filter_export_data *fed;
1421         struct obd_export *exp;
1422         obd_size maxsize = obd->obd_osfs.os_blocks * obd->obd_osfs.os_bsize;
1423         obd_size tot_dirty = 0, tot_pending = 0, tot_granted = 0;
1424         obd_size fo_tot_dirty, fo_tot_pending, fo_tot_granted;
1425
1426         if (list_empty(&obd->obd_exports))
1427                 return;
1428
1429         spin_lock(&obd->obd_osfs_lock);
1430         spin_lock(&obd->obd_dev_lock);
1431         list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
1432                 fed = &exp->exp_filter_data;
1433                 LASSERTF(fed->fed_grant + fed->fed_pending <= maxsize,
1434                          "cli %s/%p %lu+%lu > "LPU64"\n",
1435                          exp->exp_client_uuid.uuid, exp,
1436                          fed->fed_grant, fed->fed_pending, maxsize);
1437                 LASSERTF(fed->fed_dirty <= maxsize, "cli %s/%p %lu > "LPU64"\n",
1438                          exp->exp_client_uuid.uuid, exp,fed->fed_dirty,maxsize);
1439                 CDEBUG(D_CACHE,"%s: cli %s/%p dirty %lu pend %lu grant %lu\n",
1440                        obd->obd_name, exp->exp_client_uuid.uuid, exp,
1441                        fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1442                 tot_granted += fed->fed_grant + fed->fed_pending;
1443                 tot_pending += fed->fed_pending;
1444                 tot_dirty += fed->fed_dirty;
1445         }
1446         fo_tot_granted = obd->u.filter.fo_tot_granted;
1447         fo_tot_pending = obd->u.filter.fo_tot_pending;
1448         fo_tot_dirty = obd->u.filter.fo_tot_dirty;
1449         spin_unlock(&obd->obd_dev_lock);
1450         spin_unlock(&obd->obd_osfs_lock);
1451
1452         /* Do these assertions outside the spinlocks so we don't kill system */
1453         CERROR("%s: tot_granted "LPU64" != fo_tot_granted "LPU64"\n",
1454                func, tot_granted, fo_tot_granted);
1455         CERROR("%s: tot_pending "LPU64" != fo_tot_pending "LPU64"\n",
1456                func, tot_pending, fo_tot_pending);
1457         CERROR("%s: tot_dirty "LPU64" != fo_tot_dirty "LPU64"\n",
1458                func, tot_dirty, fo_tot_dirty);
1459         CERROR("%s: tot_pending "LPU64" > tot_granted "LPU64"\n",
1460                func, tot_pending, tot_granted);
1461         CERROR("%s: tot_granted "LPU64" > maxsize "LPU64"\n",
1462                func, tot_granted, maxsize);
1463         CERROR("%s: tot_dirty "LPU64" > maxsize "LPU64"\n",
1464                func, tot_dirty, maxsize);
1465 }
1466
1467 /* Remove this client from the grant accounting totals.  We also remove
1468  * the export from the obd device under the osfs and dev locks to ensure
1469  * that the filter_grant_sanity_check() calculations are always valid.
1470  * The client should do something similar when it invalidates its import. */
1471 static void filter_grant_discard(struct obd_export *exp)
1472 {
1473         struct obd_device *obd = exp->exp_obd;
1474         struct filter_obd *filter = &obd->u.filter;
1475         struct filter_export_data *fed = &exp->exp_filter_data;
1476
1477         spin_lock(&obd->obd_osfs_lock);
1478         spin_lock(&exp->exp_obd->obd_dev_lock);
1479         list_del_init(&exp->exp_obd_chain);
1480         spin_unlock(&exp->exp_obd->obd_dev_lock);
1481
1482         CDEBUG(D_CACHE, "%s: cli %s/%p dirty %lu pend %lu grant %lu\n",
1483                obd->obd_name, exp->exp_client_uuid.uuid, exp,
1484                fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1485
1486         LASSERTF(filter->fo_tot_granted >= fed->fed_grant,
1487                  "%s: tot_granted "LPU64" cli %s/%p fed_grant %lu\n",
1488                  obd->obd_name, filter->fo_tot_granted,
1489                  exp->exp_client_uuid.uuid, exp, fed->fed_grant);
1490         filter->fo_tot_granted -= fed->fed_grant;
1491         LASSERTF(exp->exp_obd->u.filter.fo_tot_pending >= fed->fed_pending,
1492                  "%s: tot_pending "LPU64" cli %s/%p fed_pending %lu\n",
1493                  obd->obd_name, filter->fo_tot_pending,
1494                  exp->exp_client_uuid.uuid, exp, fed->fed_pending);
1495         LASSERTF(filter->fo_tot_dirty >= fed->fed_dirty,
1496                  "%s: tot_dirty "LPU64" cli %s/%p fed_dirty %lu\n",
1497                  obd->obd_name, filter->fo_tot_dirty,
1498                  exp->exp_client_uuid.uuid, exp, fed->fed_dirty);
1499         filter->fo_tot_dirty -= fed->fed_dirty;
1500         fed->fed_dirty = 0;
1501         fed->fed_grant = 0;
1502
1503         spin_unlock(&obd->obd_osfs_lock);
1504 }
1505
1506 static int filter_destroy_export(struct obd_export *exp)
1507 {
1508         ENTRY;
1509
1510         if (exp->exp_filter_data.fed_pending)
1511                 CERROR("%s: cli %s/%p has %lu pending on destroyed export\n",
1512                        exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
1513                        exp, exp->exp_filter_data.fed_pending);
1514
1515         target_destroy_export(exp);
1516
1517         if (exp->exp_obd->obd_replayable)
1518                 filter_client_free(exp, exp->exp_flags);
1519
1520         filter_grant_discard(exp);
1521         if (!(exp->exp_flags & OBD_OPT_FORCE))
1522                 filter_grant_sanity_check(exp->exp_obd, __FUNCTION__);
1523
1524         RETURN(0);
1525 }
1526
1527 /* also incredibly similar to mds_disconnect */
1528 static int filter_disconnect(struct obd_export *exp, int flags)
1529 {
1530         struct obd_device *obd = exp->exp_obd;
1531         unsigned long irqflags;
1532         struct llog_ctxt *ctxt;
1533         int rc;
1534         ENTRY;
1535
1536         LASSERT(exp);
1537         class_export_get(exp);
1538
1539         spin_lock_irqsave(&exp->exp_lock, irqflags);
1540         exp->exp_flags = flags;
1541         spin_unlock_irqrestore(&exp->exp_lock, irqflags);
1542
1543         if (!(flags & OBD_OPT_FORCE))
1544                 filter_grant_sanity_check(obd, __FUNCTION__);
1545         filter_grant_discard(exp);
1546
1547         /* Disconnect early so that clients can't keep using export */
1548         rc = class_disconnect(exp, flags);
1549
1550         ldlm_cancel_locks_for_export(exp);
1551
1552         fsfilt_sync(obd, obd->u.filter.fo_sb);
1553
1554         /* flush any remaining cancel messages out to the target */
1555         ctxt = llog_get_context(obd, LLOG_UNLINK_REPL_CTXT);
1556         llog_sync(ctxt, exp);
1557
1558         class_export_put(exp);
1559         RETURN(rc);
1560 }
1561
1562 struct dentry *__filter_oa2dentry(struct obd_device *obd,
1563                                   struct obdo *oa, const char *what)
1564 {
1565         struct dentry *dchild = NULL;
1566         obd_gr group = 0;
1567
1568         if (oa->o_valid & OBD_MD_FLGROUP)
1569                 group = oa->o_gr;
1570
1571         dchild = filter_fid2dentry(obd, NULL, group, oa->o_id);
1572
1573         if (IS_ERR(dchild)) {
1574                 CERROR("%s error looking up object: "LPU64"\n", what, oa->o_id);
1575                 RETURN(dchild);
1576         }
1577
1578         if (dchild->d_inode == NULL) {
1579                 CERROR("%s on non-existent object: "LPU64"\n", what, oa->o_id);
1580                 f_dput(dchild);
1581                 RETURN(ERR_PTR(-ENOENT));
1582         }
1583
1584         return dchild;
1585 }
1586
1587 static int filter_getattr(struct obd_export *exp, struct obdo *oa,
1588                           struct lov_stripe_md *md)
1589 {
1590         struct dentry *dentry = NULL;
1591         struct obd_device *obd;
1592         int rc = 0;
1593         ENTRY;
1594
1595         obd = class_exp2obd(exp);
1596         if (obd == NULL) {
1597                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1598                        exp->exp_handle.h_cookie);
1599                 RETURN(-EINVAL);
1600         }
1601
1602         dentry = filter_oa2dentry(obd, oa);
1603         if (IS_ERR(dentry))
1604                 RETURN(PTR_ERR(dentry));
1605
1606         /* Limit the valid bits in the return data to what we actually use */
1607         oa->o_valid = OBD_MD_FLID;
1608         obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
1609
1610         f_dput(dentry);
1611         RETURN(rc);
1612 }
1613
1614 /* this is called from filter_truncate() until we have filter_punch() */
1615 static int filter_setattr(struct obd_export *exp, struct obdo *oa,
1616                           struct lov_stripe_md *md, struct obd_trans_info *oti)
1617 {
1618         struct obd_run_ctxt saved;
1619         struct filter_obd *filter;
1620         struct dentry *dentry;
1621         struct iattr iattr;
1622         struct ldlm_res_id res_id = { .name = { oa->o_id } };
1623         struct ldlm_resource *res;
1624         void *handle;
1625         int rc, rc2;
1626         ENTRY;
1627
1628         LASSERT(oti != NULL);
1629
1630         dentry = filter_oa2dentry(exp->exp_obd, oa);
1631         if (IS_ERR(dentry))
1632                 RETURN(PTR_ERR(dentry));
1633
1634         filter = &exp->exp_obd->u.filter;
1635
1636         iattr_from_obdo(&iattr, oa, oa->o_valid);
1637
1638         push_ctxt(&saved, &exp->exp_obd->obd_ctxt, NULL);
1639         lock_kernel();
1640
1641         if (iattr.ia_valid & ATTR_SIZE)
1642                 down(&dentry->d_inode->i_sem);
1643         handle = fsfilt_start(exp->exp_obd, dentry->d_inode, FSFILT_OP_SETATTR,
1644                               oti);
1645         if (IS_ERR(handle))
1646                 GOTO(out_unlock, rc = PTR_ERR(handle));
1647
1648         /* XXX this could be a rwsem instead, if filter_preprw played along */
1649         if (iattr.ia_valid & ATTR_ATTR_FLAG)
1650                 rc = fsfilt_iocontrol(exp->exp_obd, dentry->d_inode, NULL,
1651                                       EXT3_IOC_SETFLAGS,
1652                                       (long)&iattr.ia_attr_flags);
1653         else
1654                 rc = fsfilt_setattr(exp->exp_obd, dentry, handle, &iattr, 1);
1655         rc = filter_finish_transno(exp, oti, rc);
1656         rc2 = fsfilt_commit(exp->exp_obd, dentry->d_inode, handle, 0);
1657         if (rc2) {
1658                 CERROR("error on commit, err = %d\n", rc2);
1659                 if (!rc)
1660                         rc = rc2;
1661         }
1662
1663         if (iattr.ia_valid & ATTR_SIZE) {
1664                 res = ldlm_resource_get(exp->exp_obd->obd_namespace, NULL,
1665                                         res_id, LDLM_EXTENT, 0);
1666                 if (res == NULL) {
1667                         CERROR("!!! resource_get failed for object "LPU64" -- "
1668                                "filter_setattr with no lock?\n", oa->o_id);
1669                 } else {
1670                         if (res->lr_namespace->ns_lvbo &&
1671                             res->lr_namespace->ns_lvbo->lvbo_update) {
1672                                 rc = res->lr_namespace->ns_lvbo->lvbo_update
1673                                         (res, NULL, 0, 0);
1674                         }
1675                         ldlm_resource_putref(res);
1676                 }
1677         }
1678
1679         oa->o_valid = OBD_MD_FLID;
1680         obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
1681
1682 out_unlock:
1683         if (iattr.ia_valid & ATTR_SIZE)
1684                 up(&dentry->d_inode->i_sem);
1685         unlock_kernel();
1686         pop_ctxt(&saved, &exp->exp_obd->obd_ctxt, NULL);
1687
1688         f_dput(dentry);
1689         RETURN(rc);
1690 }
1691
1692 /* XXX identical to osc_unpackmd */
1693 static int filter_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
1694                            struct lov_mds_md *lmm, int lmm_bytes)
1695 {
1696         int lsm_size;
1697         ENTRY;
1698
1699         if (lmm != NULL) {
1700                 if (lmm_bytes < sizeof (*lmm)) {
1701                         CERROR("lov_mds_md too small: %d, need %d\n",
1702                                lmm_bytes, (int)sizeof(*lmm));
1703                         RETURN(-EINVAL);
1704                 }
1705                 /* XXX LOV_MAGIC etc check? */
1706
1707                 if (lmm->lmm_object_id == cpu_to_le64(0)) {
1708                         CERROR("lov_mds_md: zero lmm_object_id\n");
1709                         RETURN(-EINVAL);
1710                 }
1711         }
1712
1713         lsm_size = lov_stripe_md_size(1);
1714         if (lsmp == NULL)
1715                 RETURN(lsm_size);
1716
1717         if (*lsmp != NULL && lmm == NULL) {
1718                 OBD_FREE(*lsmp, lsm_size);
1719                 *lsmp = NULL;
1720                 RETURN(0);
1721         }
1722
1723         if (*lsmp == NULL) {
1724                 OBD_ALLOC(*lsmp, lsm_size);
1725                 if (*lsmp == NULL)
1726                         RETURN(-ENOMEM);
1727
1728                 loi_init((*lsmp)->lsm_oinfo);
1729         }
1730
1731         if (lmm != NULL) {
1732                 /* XXX zero *lsmp? */
1733                 (*lsmp)->lsm_object_id = le64_to_cpu (lmm->lmm_object_id);
1734                 LASSERT((*lsmp)->lsm_object_id);
1735         }
1736
1737         (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
1738
1739         RETURN(lsm_size);
1740 }
1741
1742 static void filter_destroy_precreated(struct obd_export *exp, struct obdo *oa,
1743                                       struct filter_obd *filter)
1744 {
1745         struct obdo doa; /* XXX obdo on stack */
1746         __u64 last, id;
1747         ENTRY;
1748         LASSERT(oa);
1749
1750         memset(&doa, 0, sizeof(doa));
1751         if (oa->o_valid & OBD_MD_FLGROUP)
1752                 doa.o_gr = oa->o_gr;
1753         else
1754                 doa.o_gr = 0;
1755         doa.o_mode = S_IFREG;
1756         last = filter_last_id(filter, &doa); /* FIXME: object groups */
1757         CWARN("deleting orphan objects from "LPU64" to "LPU64"\n",
1758                oa->o_id + 1, last);
1759         for (id = oa->o_id + 1; id <= last; id++) {
1760                 doa.o_id = id;
1761                 filter_destroy(exp, &doa, NULL, NULL);
1762         }
1763         spin_lock(&filter->fo_objidlock);
1764         filter->fo_last_objids[0] = oa->o_id; /* FIXME: object groups */
1765         spin_unlock(&filter->fo_objidlock);
1766         EXIT;
1767 }
1768
1769 /* returns a negative error or a nonnegative number of files to create */
1770 static int filter_should_precreate(struct obd_export *exp, struct obdo *oa,
1771                                    int group)
1772 {
1773         struct obd_device *obd = exp->exp_obd;
1774         struct filter_obd *filter = &obd->u.filter;
1775         int diff, rc;
1776         ENTRY;
1777
1778         diff = oa->o_id - filter_last_id(filter, oa);
1779         CDEBUG(D_INFO, "filter_last_id() = "LPU64" -> diff = %d\n",
1780                filter_last_id(filter, oa), diff);
1781
1782         /* delete orphans request */
1783         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1784             (oa->o_flags & OBD_FL_DELORPHAN)) {
1785                 if (diff >= 0)
1786                         RETURN(diff);
1787                 if (-diff > 10000) { /* XXX make this smarter */
1788                         CERROR("ignoring bogus orphan destroy request: obdid "
1789                                LPU64" last_id "LPU64"\n",
1790                                oa->o_id, filter_last_id(filter, oa));
1791                         RETURN(-EINVAL);
1792                 }
1793                 filter_destroy_precreated(exp, oa, filter);
1794                 rc = filter_update_last_objid(obd, group, 0);
1795                 if (rc)
1796                         CERROR("unable to write lastobjid, but orphans"
1797                                "were deleted\n");
1798                 RETURN(0);
1799         } else {
1800                 /* only precreate if group == 0 and o_id is specfied */
1801                 if (!(oa->o_valid & OBD_FL_DELORPHAN) &&
1802                     (group != 0 || oa->o_id == 0))
1803                         RETURN(1);
1804
1805                 LASSERT(diff >= 0);
1806                 RETURN(diff);
1807         }
1808 }
1809
1810 /* We rely on the fact that only one thread will be creating files in a given
1811  * group at a time, which is why we don't need an atomic filter_get_new_id.
1812  * Even if we had that atomic function, the following race would exist:
1813  *
1814  * thread 1: gets id x from filter_next_id
1815  * thread 2: gets id (x + 1) from filter_next_id
1816  * thread 2: creates object (x + 1)
1817  * thread 1: tries to create object x, gets -ENOSPC
1818  */
1819 static int filter_precreate(struct obd_device *obd, struct obdo *oa,
1820                             obd_gr group, int *num)
1821 {
1822         struct dentry *dchild = NULL;
1823         struct filter_obd *filter;
1824         struct dentry *dparent;
1825         int err = 0, rc = 0, i;
1826         __u64 next_id;
1827         int recreate_obj = 0;
1828         void *handle = NULL;
1829         ENTRY;
1830
1831         filter = &obd->u.filter;
1832
1833         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1834             (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1835                 recreate_obj = 1;
1836         }
1837
1838         for (i = 0; i < *num && err == 0; i++) {
1839                 int cleanup_phase = 0;
1840
1841                 if (recreate_obj) {
1842                         __u64 last_id;
1843                         next_id = oa->o_id;
1844                         last_id = filter_last_id(filter, NULL);
1845                         if (next_id > last_id) {
1846                                 CERROR("Error: Trying to recreate obj greater"
1847                                        "than last id "LPD64" > "LPD64"\n",
1848                                        next_id, last_id);
1849                                 RETURN(-EINVAL);
1850                         }
1851                 } else
1852                         next_id = filter_last_id(filter, NULL) + 1;
1853
1854                 CDEBUG(D_INFO, "precreate objid "LPU64"\n", next_id);
1855
1856                 dparent = filter_parent_lock(obd, group, next_id);
1857                 if (IS_ERR(dparent))
1858                         GOTO(cleanup, rc = PTR_ERR(dparent));
1859                 cleanup_phase = 1;
1860
1861                 dchild = filter_fid2dentry(obd, dparent, group, next_id);
1862                 if (IS_ERR(dchild))
1863                         GOTO(cleanup, rc = PTR_ERR(dchild));
1864                 cleanup_phase = 2;
1865
1866                 if (dchild->d_inode != NULL) {
1867                         /* This would only happen if lastobjid was bad on disk*/
1868                         /* Could also happen if recreating missing obj but
1869                          * already exists
1870                          */
1871                         if (recreate_obj) {
1872                                 CERROR("Serious error: recreating obj %*s but "
1873                                        "obj already exists \n",
1874                                        dchild->d_name.len, dchild->d_name.name);
1875                         } else {
1876                                 CERROR("Serious error: objid %*s already "
1877                                        "exists; is this filesystem corrupt?\n",
1878                                         dchild->d_name.len, dchild->d_name.name);
1879                         }
1880                         GOTO(cleanup, rc = -EEXIST);
1881                 }
1882
1883                 handle = fsfilt_start(obd, dparent->d_inode,
1884                                       FSFILT_OP_CREATE_LOG, NULL);
1885                 if (IS_ERR(handle))
1886                         GOTO(cleanup, rc = PTR_ERR(handle));
1887                 cleanup_phase = 3;
1888
1889                 rc = ll_vfs_create(dparent->d_inode, dchild, S_IFREG, NULL);
1890                 if (rc) {
1891                         CERROR("create failed rc = %d\n", rc);
1892                         GOTO(cleanup, rc);
1893                 }
1894
1895                 if (!recreate_obj) {
1896                         filter_set_last_id(filter, NULL, next_id);
1897                         err = filter_update_last_objid(obd, group, 0);
1898                         if (err)
1899                                 CERROR("unable to write lastobjid "
1900                                        "but file created\n");
1901                 }
1902
1903         cleanup:
1904                 switch(cleanup_phase) {
1905                 case 3:
1906                         err = fsfilt_commit(obd, dparent->d_inode, handle, 0);
1907                         if (err) {
1908                                 CERROR("error on commit, err = %d\n", err);
1909                                 if (!rc)
1910                                         rc = err;
1911                         }
1912                 case 2:
1913                         f_dput(dchild);
1914                 case 1:
1915                         filter_parent_unlock(dparent);
1916                 case 0:
1917                         break;
1918                 }
1919
1920                 if (rc)
1921                         break;
1922         }
1923         *num = i;
1924
1925         CDEBUG(D_INFO, "filter_precreate() created %d objects\n", i);
1926         RETURN(rc);
1927 }
1928
1929 static int filter_create(struct obd_export *exp, struct obdo *oa,
1930                          struct lov_stripe_md **ea, struct obd_trans_info *oti)
1931 {
1932         struct obd_device *obd = NULL;
1933         struct obd_run_ctxt saved;
1934         struct lov_stripe_md *lsm = NULL;
1935         obd_gr group = 0;
1936         int rc = 0, diff;
1937         ENTRY;
1938
1939         if (oa->o_valid & OBD_MD_FLGROUP)
1940                 group = oa->o_gr;
1941
1942         CDEBUG(D_INFO, "filter_create(od->o_gr="LPU64",od->o_id="LPU64")\n",
1943                group, oa->o_id);
1944         if (ea != NULL) {
1945                 lsm = *ea;
1946                 if (lsm == NULL) {
1947                         rc = obd_alloc_memmd(exp, &lsm);
1948                         if (rc < 0)
1949                                 RETURN(rc);
1950                 }
1951         }
1952
1953         obd = exp->exp_obd;
1954         push_ctxt(&saved, &obd->obd_ctxt, NULL);
1955
1956         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1957             (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1958                 if (oa->o_id > filter_last_id(&obd->u.filter, oa)) {
1959                         CERROR("recreate objid "LPU64" > last id "LPU64"\n",
1960                                oa->o_id, filter_last_id(&obd->u.filter, oa));
1961                         rc = -EINVAL;
1962                 } else {
1963                         diff = 1;
1964                         rc = filter_precreate(obd, oa, group, &diff);
1965                 }
1966         } else {
1967                 diff = filter_should_precreate(exp, oa, group);
1968                 if (diff > 0) {
1969                         oa->o_id = filter_last_id(&obd->u.filter, oa);
1970                         rc = filter_precreate(obd, oa, group, &diff);
1971                         oa->o_id += diff;
1972                         oa->o_valid = OBD_MD_FLID;
1973                 }
1974         }
1975
1976         pop_ctxt(&saved, &obd->obd_ctxt, NULL);
1977         if (rc && ea != NULL && *ea != lsm) {
1978                 obd_free_memmd(exp, &lsm);
1979         } else if (rc == 0 && ea != NULL) {
1980                 /* XXX LOV STACKING: the lsm that is passed to us from
1981                  * LOV does not have valid lsm_oinfo data structs, so
1982                  * don't go touching that.  This needs to be fixed in a
1983                  * big way. */
1984                 lsm->lsm_object_id = oa->o_id;
1985                 *ea = lsm;
1986         }
1987
1988         RETURN(rc);
1989 }
1990
1991 static int filter_destroy(struct obd_export *exp, struct obdo *oa,
1992                           struct lov_stripe_md *ea, struct obd_trans_info *oti)
1993 {
1994         struct obd_device *obd;
1995         struct filter_obd *filter;
1996         struct dentry *dchild = NULL, *dparent = NULL;
1997         struct obd_run_ctxt saved;
1998         void *handle = NULL;
1999         struct llog_cookie *fcc = NULL;
2000         int rc, rc2, cleanup_phase = 0, have_prepared = 0;
2001         obd_gr group = 0;
2002         ENTRY;
2003
2004         if (oa->o_valid & OBD_MD_FLGROUP)
2005                 group = oa->o_gr;
2006
2007         obd = exp->exp_obd;
2008         filter = &obd->u.filter;
2009
2010         push_ctxt(&saved, &obd->obd_ctxt, NULL);
2011
2012  acquire_locks:
2013         dparent = filter_parent_lock(obd, group, oa->o_id);
2014         if (IS_ERR(dparent))
2015                 GOTO(cleanup, rc = PTR_ERR(dparent));
2016         cleanup_phase = 1;
2017
2018         dchild = filter_fid2dentry(obd, dparent, group, oa->o_id);
2019         if (IS_ERR(dchild))
2020                 GOTO(cleanup, rc = -ENOENT);
2021         cleanup_phase = 2;
2022
2023         if (dchild->d_inode == NULL) {
2024                 CDEBUG(D_INODE, "destroying non-existent object "LPU64"\n",
2025                        oa->o_id);
2026                 GOTO(cleanup, rc = -ENOENT);
2027         }
2028
2029         if (!have_prepared) {
2030                 /* If we're really going to destroy the object, get ready
2031                  * by getting the clients to discard their cached data.
2032                  *
2033                  * We have to drop the parent lock, because
2034                  * filter_prepare_destroy will acquire a PW on the object, and
2035                  * we don't want to deadlock with an incoming write to the
2036                  * object, which has the extent PW and then wants to get the
2037                  * parent dentry to do the lookup.
2038                  *
2039                  * We dput the child because it's not worth the extra
2040                  * complication of condition the above code to skip it on the
2041                  * second time through. */
2042                 f_dput(dchild);
2043                 filter_parent_unlock(dparent);
2044
2045                 filter_prepare_destroy(obd, oa->o_id);
2046                 have_prepared = 1;
2047                 goto acquire_locks;
2048         }
2049
2050         handle = fsfilt_start(obd, dparent->d_inode, FSFILT_OP_UNLINK_LOG, oti);
2051         if (IS_ERR(handle))
2052                 GOTO(cleanup, rc = PTR_ERR(handle));
2053         cleanup_phase = 3;
2054
2055         /* Our MDC connection is established by the MDS to us */
2056         if (oa->o_valid & OBD_MD_FLCOOKIE) {
2057                 OBD_ALLOC(fcc, sizeof(*fcc));
2058                 if (fcc != NULL)
2059                         memcpy(fcc, obdo_logcookie(oa), sizeof(*fcc));
2060         }
2061
2062         rc = filter_destroy_internal(obd, oa->o_id, dparent, dchild);
2063
2064 cleanup:
2065         switch(cleanup_phase) {
2066         case 3:
2067                 if (fcc != NULL) {
2068                         if (oti != NULL)
2069                                 fsfilt_add_journal_cb(obd, 0, oti->oti_handle,
2070                                                       filter_cancel_cookies_cb,
2071                                                       fcc);
2072                         else
2073                                 fsfilt_add_journal_cb(obd, 0, handle,
2074                                                       filter_cancel_cookies_cb,
2075                                                       fcc);
2076                 }
2077                 rc = filter_finish_transno(exp, oti, rc);
2078                 rc2 = fsfilt_commit(obd, dparent->d_inode, handle, 0);
2079                 if (rc2) {
2080                         CERROR("error on commit, err = %d\n", rc2);
2081                         if (!rc)
2082                                 rc = rc2;
2083                 }
2084         case 2:
2085                 f_dput(dchild);
2086         case 1:
2087                 filter_parent_unlock(dparent);
2088         case 0:
2089                 pop_ctxt(&saved, &obd->obd_ctxt, NULL);
2090                 break;
2091         default:
2092                 CERROR("invalid cleanup_phase %d\n", cleanup_phase);
2093                 LBUG();
2094         }
2095
2096         RETURN(rc);
2097 }
2098
2099 /* NB start and end are used for punch, but not truncate */
2100 static int filter_truncate(struct obd_export *exp, struct obdo *oa,
2101                            struct lov_stripe_md *lsm,
2102                            obd_off start, obd_off end,
2103                            struct obd_trans_info *oti)
2104 {
2105         int error;
2106         ENTRY;
2107
2108         if (end != OBD_OBJECT_EOF)
2109                 CERROR("PUNCH not supported, only truncate: end = "LPX64"\n",
2110                        end);
2111
2112         CDEBUG(D_INODE, "calling truncate for object "LPU64", valid = %x, "
2113                "o_size = "LPD64"\n", oa->o_id, oa->o_valid, start);
2114         oa->o_size = start;
2115         error = filter_setattr(exp, oa, NULL, oti);
2116         RETURN(error);
2117 }
2118
2119 static int filter_sync(struct obd_export *exp, struct obdo *oa,
2120                        struct lov_stripe_md *lsm, obd_off start, obd_off end)
2121 {
2122         struct obd_run_ctxt saved;
2123         struct filter_obd *filter;
2124         struct dentry *dentry;
2125         struct llog_ctxt *ctxt;
2126         int rc, rc2;
2127         ENTRY;
2128
2129         filter = &exp->exp_obd->u.filter;
2130
2131         /* an objid of zero is taken to mean "sync whole filesystem" */
2132         if (!oa || !(oa->o_valid & OBD_MD_FLID)) {
2133                 rc = fsfilt_sync(exp->exp_obd, filter->fo_sb);
2134                 /* flush any remaining cancel messages out to the target */
2135                 ctxt = llog_get_context(exp->exp_obd, LLOG_UNLINK_REPL_CTXT);
2136                 llog_sync(ctxt, exp);
2137                 RETURN(rc);
2138         }
2139
2140         dentry = filter_oa2dentry(exp->exp_obd, oa);
2141         if (IS_ERR(dentry))
2142                 RETURN(PTR_ERR(dentry));
2143
2144         push_ctxt(&saved, &exp->exp_obd->obd_ctxt, NULL);
2145
2146         down(&dentry->d_inode->i_sem);
2147         rc = filemap_fdatasync(dentry->d_inode->i_mapping);
2148         if (rc == 0) {
2149                 /* just any file to grab fsync method - "file" arg unused */
2150                 struct file *file = filter->fo_rcvd_filp;
2151
2152                 if (file->f_op && file->f_op->fsync)
2153                         rc = file->f_op->fsync(NULL, dentry, 1);
2154
2155                 rc2 = filemap_fdatawait(dentry->d_inode->i_mapping);
2156                 if (!rc)
2157                         rc = rc2;
2158         }
2159         up(&dentry->d_inode->i_sem);
2160
2161         oa->o_valid = OBD_MD_FLID;
2162         obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2163
2164         pop_ctxt(&saved, &exp->exp_obd->obd_ctxt, NULL);
2165
2166         f_dput(dentry);
2167         RETURN(rc);
2168 }
2169
2170 static int filter_statfs(struct obd_device *obd, struct obd_statfs *osfs,
2171                          unsigned long max_age)
2172 {
2173         struct filter_obd *filter = &obd->u.filter;
2174         int blockbits = filter->fo_sb->s_blocksize_bits;
2175         int rc;
2176         ENTRY;
2177
2178         /* at least try to account for cached pages.  its still racey and
2179          * might be under-reporting if clients haven't announced their
2180          * caches with brw recently */
2181         spin_lock(&obd->obd_osfs_lock);
2182         rc = fsfilt_statfs(obd, filter->fo_sb, max_age);
2183         memcpy(osfs, &obd->obd_osfs, sizeof(*osfs));
2184         spin_unlock(&obd->obd_osfs_lock);
2185
2186         CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
2187                " pending "LPU64" free "LPU64" avail "LPU64"\n",
2188                filter->fo_tot_dirty, filter->fo_tot_granted,
2189                filter->fo_tot_pending,
2190                osfs->os_bfree << blockbits, osfs->os_bavail << blockbits);
2191
2192         filter_grant_sanity_check(obd, __FUNCTION__);
2193
2194         osfs->os_bavail -= min(osfs->os_bavail,
2195                                (filter->fo_tot_dirty + filter->fo_tot_pending +
2196                                 osfs->os_bsize -1) >> blockbits);
2197
2198         RETURN(rc);
2199 }
2200
2201 static int filter_get_info(struct obd_export *exp, __u32 keylen,
2202                            void *key, __u32 *vallen, void *val)
2203 {
2204         struct obd_device *obd;
2205         ENTRY;
2206
2207         obd = class_exp2obd(exp);
2208         if (obd == NULL) {
2209                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2210                        exp->exp_handle.h_cookie);
2211                 RETURN(-EINVAL);
2212         }
2213
2214         if (keylen == strlen("blocksize") &&
2215             memcmp(key, "blocksize", keylen) == 0) {
2216                 __u32 *blocksize = val;
2217                 *vallen = sizeof(*blocksize);
2218                 *blocksize = obd->u.filter.fo_sb->s_blocksize;
2219                 RETURN(0);
2220         }
2221
2222         if (keylen == strlen("blocksize_bits") &&
2223             memcmp(key, "blocksize_bits", keylen) == 0) {
2224                 __u32 *blocksize_bits = val;
2225                 *vallen = sizeof(*blocksize_bits);
2226                 *blocksize_bits = obd->u.filter.fo_sb->s_blocksize_bits;
2227                 RETURN(0);
2228         }
2229
2230         if (keylen >= strlen("last_id") && memcmp(key, "last_id", 7) == 0) {
2231                 obd_id *last_id = val;
2232                 /* FIXME: object groups */
2233                 *last_id = filter_last_id(&obd->u.filter, 0);
2234                 RETURN(0);
2235         }
2236         CDEBUG(D_IOCTL, "invalid key\n");
2237         RETURN(-EINVAL);
2238 }
2239
2240 static int filter_set_info(struct obd_export *exp, __u32 keylen,
2241                            void *key, __u32 vallen, void *val)
2242 {
2243         struct obd_device *obd;
2244         struct lustre_handle conn;
2245         struct llog_ctxt *ctxt;
2246         int rc = 0;
2247         ENTRY;
2248
2249         conn.cookie = exp->exp_handle.h_cookie;
2250
2251         obd = exp->exp_obd;
2252         if (obd == NULL) {
2253                 CDEBUG(D_IOCTL, "invalid exp %p cookie "LPX64"\n",
2254                        exp, conn.cookie);
2255                 RETURN(-EINVAL);
2256         }
2257
2258         if (keylen < strlen("mds_conn") ||
2259             memcmp(key, "mds_conn", keylen) != 0)
2260                 RETURN(-EINVAL);
2261
2262         CWARN("Received MDS connection ("LPX64")\n", conn.cookie);
2263         memcpy(&obd->u.filter.fo_mdc_conn, &conn, sizeof(conn));
2264         ctxt = llog_get_context(obd, LLOG_UNLINK_REPL_CTXT);
2265         rc = llog_receptor_accept(ctxt, exp->exp_imp_reverse);
2266         RETURN(rc);
2267 }
2268
2269 int filter_iocontrol(unsigned int cmd, struct obd_export *exp,
2270                      int len, void *karg, void *uarg)
2271 {
2272         struct obd_device *obd = exp->exp_obd;
2273         struct obd_ioctl_data *data = karg;
2274         int rc = 0;
2275
2276         switch (cmd) {
2277         case OBD_IOC_ABORT_RECOVERY:
2278                 CERROR("aborting recovery for device %s\n", obd->obd_name);
2279                 target_abort_recovery(obd);
2280                 RETURN(0);
2281
2282         case OBD_IOC_SET_READONLY: {
2283                 void *handle;
2284                 struct super_block *sb = obd->u.filter.fo_sb;
2285                 struct inode *inode = sb->s_root->d_inode;
2286                 BDEVNAME_DECLARE_STORAGE(tmp);
2287                 CERROR("setting device %s read-only\n",
2288                        ll_bdevname(sb, tmp));
2289
2290                 handle = fsfilt_start(obd, inode, FSFILT_OP_MKNOD, NULL);
2291                 LASSERT(handle);
2292                 (void)fsfilt_commit(obd, inode, handle, 1);
2293
2294                 dev_set_rdonly(ll_sbdev(obd->u.filter.fo_sb), 2);
2295                 RETURN(0);
2296         }
2297
2298         case OBD_IOC_CATLOGLIST: {
2299                 rc = llog_catlog_list(obd, 1, data);
2300                 RETURN(rc);
2301         }
2302
2303         case OBD_IOC_LLOG_CANCEL:
2304         case OBD_IOC_LLOG_REMOVE:
2305         case OBD_IOC_LLOG_INFO:
2306         case OBD_IOC_LLOG_PRINT: {
2307                 /* FIXME to be finished */
2308                 RETURN(-EOPNOTSUPP);
2309 /*
2310                 struct llog_ctxt *ctxt = NULL;
2311
2312                 push_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_ctxt, NULL);
2313                 rc = llog_ioctl(ctxt, cmd, data);
2314                 pop_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_ctxt, NULL);
2315
2316                 RETURN(rc);
2317 */
2318         }
2319
2320
2321         default:
2322                 RETURN(-EINVAL);
2323         }
2324         RETURN(0);
2325 }
2326
2327 static struct llog_operations filter_unlink_repl_logops;
2328 static struct llog_operations filter_size_orig_logops = {
2329         lop_setup: llog_obd_origin_setup,
2330         lop_cleanup: llog_obd_origin_cleanup,
2331         lop_add: llog_obd_origin_add
2332 };
2333
2334 static int filter_llog_init(struct obd_device *obd, struct obd_device *tgt,
2335                             int count, struct llog_catid *logid)
2336 {
2337         struct llog_ctxt *ctxt;
2338         int rc;
2339         ENTRY;
2340
2341         filter_unlink_repl_logops = llog_client_ops;
2342         filter_unlink_repl_logops.lop_cancel = llog_obd_repl_cancel;
2343         filter_unlink_repl_logops.lop_connect = llog_repl_connect;
2344         filter_unlink_repl_logops.lop_sync = llog_obd_repl_sync;
2345
2346         rc = llog_setup(obd, LLOG_UNLINK_REPL_CTXT, tgt, 0, NULL,
2347                         &filter_unlink_repl_logops);
2348         if (rc)
2349                 RETURN(rc);
2350         /* FIXME - assign unlink_cb for filter's recovery */
2351         ctxt = llog_get_context(obd, LLOG_UNLINK_REPL_CTXT);
2352         ctxt->llog_proc_cb = filter_recov_log_unlink_cb;
2353
2354         rc = llog_setup(obd, LLOG_SIZE_ORIG_CTXT, tgt, 0, NULL,
2355                         &filter_size_orig_logops);
2356         RETURN(rc);
2357 }
2358
2359 static int filter_llog_finish(struct obd_device *obd, int count)
2360 {
2361         int rc;
2362         ENTRY;
2363
2364         rc = llog_cleanup(llog_get_context(obd, LLOG_UNLINK_REPL_CTXT));
2365         if (rc)
2366                 RETURN(rc);
2367
2368         rc = llog_cleanup(llog_get_context(obd, LLOG_SIZE_ORIG_CTXT));
2369         RETURN(rc);
2370 }
2371
2372 static struct dentry *filter_lvfs_fid2dentry(__u64 id, __u32 gen, __u64 gr,
2373                                              void *data)
2374 {
2375         return filter_fid2dentry(data, NULL, gr, id);
2376 }
2377
2378 static struct lvfs_callback_ops filter_lvfs_ops = {
2379         l_fid2dentry:     filter_lvfs_fid2dentry,
2380 };
2381
2382 static struct obd_ops filter_obd_ops = {
2383         o_owner:          THIS_MODULE,
2384         o_attach:         filter_attach,
2385         o_detach:         filter_detach,
2386         o_get_info:       filter_get_info,
2387         o_set_info:       filter_set_info,
2388         o_setup:          filter_setup,
2389         o_precleanup:     filter_precleanup,
2390         o_cleanup:        filter_cleanup,
2391         o_connect:        filter_connect,
2392         o_disconnect:     filter_disconnect,
2393         o_statfs:         filter_statfs,
2394         o_getattr:        filter_getattr,
2395         o_unpackmd:       filter_unpackmd,
2396         o_create:         filter_create,
2397         o_setattr:        filter_setattr,
2398         o_destroy:        filter_destroy,
2399         o_brw:            filter_brw,
2400         o_punch:          filter_truncate,
2401         o_sync:           filter_sync,
2402         o_preprw:         filter_preprw,
2403         o_commitrw:       filter_commitrw,
2404         o_destroy_export: filter_destroy_export,
2405         o_llog_init:      filter_llog_init,
2406         o_llog_finish:    filter_llog_finish,
2407         o_iocontrol:      filter_iocontrol,
2408 };
2409
2410 static struct obd_ops filter_sanobd_ops = {
2411         o_owner:          THIS_MODULE,
2412         o_attach:         filter_attach,
2413         o_detach:         filter_detach,
2414         o_get_info:       filter_get_info,
2415         o_set_info:       filter_set_info,
2416         o_setup:          filter_san_setup,
2417         o_precleanup:     filter_precleanup,
2418         o_cleanup:        filter_cleanup,
2419         o_connect:        filter_connect,
2420         o_disconnect:     filter_disconnect,
2421         o_statfs:         filter_statfs,
2422         o_getattr:        filter_getattr,
2423         o_unpackmd:       filter_unpackmd,
2424         o_create:         filter_create,
2425         o_setattr:        filter_setattr,
2426         o_destroy:        filter_destroy,
2427         o_brw:            filter_brw,
2428         o_punch:          filter_truncate,
2429         o_sync:           filter_sync,
2430         o_preprw:         filter_preprw,
2431         o_commitrw:       filter_commitrw,
2432         o_san_preprw:     filter_san_preprw,
2433         o_destroy_export: filter_destroy_export,
2434         o_llog_init:      filter_llog_init,
2435         o_llog_finish:    filter_llog_finish,
2436         o_iocontrol:      filter_iocontrol,
2437 };
2438
2439 static int __init obdfilter_init(void)
2440 {
2441         struct lprocfs_static_vars lvars;
2442         int rc;
2443
2444         printk(KERN_INFO "Lustre: Filtering OBD driver; info@clusterfs.com\n");
2445
2446         lprocfs_init_vars(filter, &lvars);
2447
2448         rc = class_register_type(&filter_obd_ops, lvars.module_vars,
2449                                  OBD_FILTER_DEVICENAME);
2450         if (rc)
2451                 return rc;
2452
2453         rc = class_register_type(&filter_sanobd_ops, lvars.module_vars,
2454                                  OBD_FILTER_SAN_DEVICENAME);
2455         if (rc)
2456                 class_unregister_type(OBD_FILTER_DEVICENAME);
2457         return rc;
2458 }
2459
2460 static void __exit obdfilter_exit(void)
2461 {
2462         class_unregister_type(OBD_FILTER_SAN_DEVICENAME);
2463         class_unregister_type(OBD_FILTER_DEVICENAME);
2464 }
2465
2466 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2467 MODULE_DESCRIPTION("Lustre Filtering OBD driver");
2468 MODULE_LICENSE("GPL");
2469
2470 module_init(obdfilter_init);
2471 module_exit(obdfilter_exit);