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
  57 #include "filter_internal.h"
  58
  59 static struct lvfs_callback_ops filter_lvfs_ops;
  60
  61 static int filter_destroy(struct obd_export *exp, struct obdo *oa,
  62                           struct lov_stripe_md *ea, struct obd_trans_info *);
  63
  64 static void filter_commit_cb(struct obd_device *obd, __u64 transno,
  65                              void *cb_data, int error)
  66 {
  67         obd_transno_commit_cb(obd, transno, error);
  68 }
  69
  70
  71 /* Assumes caller has already pushed us into the kernel context. */
  72 int filter_finish_transno(struct obd_export *exp, struct obd_trans_info *oti,
  73                           int rc)
  74 {
  75         struct filter_obd *filter = &exp->exp_obd->u.filter;
  76         struct filter_export_data *fed = &exp->exp_filter_data;
  77         struct filter_client_data *fcd = fed->fed_fcd;
  78         __u64 last_rcvd;
  79         loff_t off;
  80         int err, log_pri = D_HA;
  81
  82         /* Propagate error code. */
  83         if (rc)
  84                 RETURN(rc);
  85
  86         if (!exp->exp_obd->obd_replayable || oti == NULL)
  87                 RETURN(rc);
  88
  89         /* we don't allocate new transnos for replayed requests */
  90         if (oti->oti_transno == 0) {
  91                 spin_lock(&filter->fo_translock);
  92                 last_rcvd = le64_to_cpu(filter->fo_fsd->fsd_last_transno) + 1;
  93                 filter->fo_fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
  94                 spin_unlock(&filter->fo_translock);
  95                 oti->oti_transno = last_rcvd;
  96         } else {
  97                 spin_lock(&filter->fo_translock);
  98                 last_rcvd = oti->oti_transno;
  99                 if (last_rcvd > le64_to_cpu(filter->fo_fsd->fsd_last_transno))
 100                         filter->fo_fsd->fsd_last_transno =
 101                                 cpu_to_le64(last_rcvd);
 102                 spin_unlock(&filter->fo_translock);
 103         }
 104         fcd->fcd_last_rcvd = cpu_to_le64(last_rcvd);
 105
 106         /* could get xid from oti, if it's ever needed */
 107         fcd->fcd_last_xid = 0;
 108
 109         off = fed->fed_lr_off;
 110         fsfilt_add_journal_cb(exp->exp_obd, last_rcvd, oti->oti_handle,
 111                               filter_commit_cb, NULL);
 112         err = fsfilt_write_record(exp->exp_obd, filter->fo_rcvd_filp, fcd,
 113                                   sizeof(*fcd), &off, 0);
 114         if (err) {
 115                 log_pri = D_ERROR;
 116                 if (rc == 0)
 117                         rc = err;
 118         }
 119
 120         CDEBUG(log_pri, "wrote trans "LPU64" for client %s at #%d: err = %d\n",
 121                last_rcvd, fcd->fcd_uuid, fed->fed_lr_idx, err);
 122
 123         RETURN(rc);
 124 }
 125
 126 void f_dput(struct dentry *dentry)
 127 {
 128         /* Can't go inside filter_ddelete because it can block */
 129         CDEBUG(D_INODE, "putting %s: %p, count = %d\n",
 130                dentry->d_name.name, dentry, atomic_read(&dentry->d_count) - 1);
 131         LASSERT(atomic_read(&dentry->d_count) > 0);
 132
 133         dput(dentry);
 134 }
 135
 136 /* Add client data to the FILTER.  We use a bitmap to locate a free space
 137  * in the last_rcvd file if cl_idx is -1 (i.e. a new client).
 138  * Otherwise, we have just read the data from the last_rcvd file and
 139  * we know its offset. */
 140 static int filter_client_add(struct obd_device *obd, struct filter_obd *filter,
 141                              struct filter_export_data *fed, int cl_idx)
 142 {
 143         unsigned long *bitmap = filter->fo_last_rcvd_slots;
 144         int new_client = (cl_idx == -1);
 145         ENTRY;
 146
 147         LASSERT(bitmap != NULL);
 148
 149         /* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
 150         if (!strcmp(fed->fed_fcd->fcd_uuid, obd->obd_uuid.uuid))
 151                 RETURN(0);
 152
 153         /* the bitmap operations can handle cl_idx > sizeof(long) * 8, so
 154          * there's no need for extra complication here
 155          */
 156         if (new_client) {
 157                 cl_idx = find_first_zero_bit(bitmap, FILTER_LR_MAX_CLIENTS);
 158         repeat:
 159                 if (cl_idx >= FILTER_LR_MAX_CLIENTS) {
 160                         CERROR("no client slots - fix FILTER_LR_MAX_CLIENTS\n");
 161                         RETURN(-ENOMEM);
 162                 }
 163                 if (test_and_set_bit(cl_idx, bitmap)) {
 164                         CERROR("FILTER client %d: found bit is set in bitmap\n",
 165                                cl_idx);
 166                         cl_idx = find_next_zero_bit(bitmap,
 167                                                     FILTER_LR_MAX_CLIENTS,
 168                                                     cl_idx);
 169                         goto repeat;
 170                 }
 171         } else {
 172                 if (test_and_set_bit(cl_idx, bitmap)) {
 173                         CERROR("FILTER client %d: bit already set in bitmap!\n",
 174                                cl_idx);
 175                         LBUG();
 176                 }
 177         }
 178
 179         fed->fed_lr_idx = cl_idx;
 180         fed->fed_lr_off = le32_to_cpu(filter->fo_fsd->fsd_client_start) +
 181                 cl_idx * le16_to_cpu(filter->fo_fsd->fsd_client_size);
 182
 183         CDEBUG(D_INFO, "client at index %d (%llu) with UUID '%s' added\n",
 184                fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
 185
 186         if (new_client) {
 187                 struct obd_run_ctxt saved;
 188                 loff_t off = fed->fed_lr_off;
 189                 int err;
 190                 void *handle;
 191
 192                 CDEBUG(D_INFO, "writing client fcd at idx %u (%llu) (len %u)\n",
 193                        fed->fed_lr_idx,off,(unsigned int)sizeof(*fed->fed_fcd));
 194
 195                 push_ctxt(&saved, &obd->obd_ctxt, NULL);
 196                 /* Transaction needed to fix bug 1403 */
 197                 handle = fsfilt_start(obd,
 198                                       filter->fo_rcvd_filp->f_dentry->d_inode,
 199                                       FSFILT_OP_SETATTR, NULL);
 200                 if (IS_ERR(handle)) {
 201                         err = PTR_ERR(handle);
 202                         CERROR("unable to start transaction: rc %d\n", err);
 203                 } else {
 204                         err = fsfilt_write_record(obd, filter->fo_rcvd_filp,
 205                                                   fed->fed_fcd,
 206                                                   sizeof(*fed->fed_fcd),
 207                                                   &off, 1);
 208                         fsfilt_commit(obd,
 209                                       filter->fo_rcvd_filp->f_dentry->d_inode,
 210                                       handle, 1);
 211                 }
 212                 pop_ctxt(&saved, &obd->obd_ctxt, NULL);
 213
 214                 if (err) {
 215                         CERROR("error writing %s client idx %u: rc %d\n",
 216                                LAST_RCVD, fed->fed_lr_idx, err);
 217                         RETURN(err);
 218                 }
 219         }
 220         RETURN(0);
 221 }
 222
 223 static int filter_client_free(struct obd_export *exp, int flags)
 224 {
 225         struct filter_export_data *fed = &exp->exp_filter_data;
 226         struct filter_obd *filter = &exp->exp_obd->u.filter;
 227         struct obd_device *obd = exp->exp_obd;
 228         struct filter_client_data zero_fcd;
 229         struct obd_run_ctxt saved;
 230         int rc;
 231         loff_t off;
 232         ENTRY;
 233
 234         if (fed->fed_fcd == NULL)
 235                 RETURN(0);
 236
 237         if (flags & OBD_OPT_FAILOVER)
 238                 GOTO(free, 0);
 239
 240         /* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
 241         if (strcmp(fed->fed_fcd->fcd_uuid, obd->obd_uuid.uuid ) == 0)
 242                 GOTO(free, 0);
 243
 244         LASSERT(filter->fo_last_rcvd_slots != NULL);
 245
 246         off = fed->fed_lr_off;
 247
 248         CDEBUG(D_INFO, "freeing client at idx %u (%lld) with UUID '%s'\n",
 249                fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
 250
 251         /* Clear the bit _after_ zeroing out the client so we don't
 252            race with filter_client_add and zero out new clients.*/
 253         if (!test_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
 254                 CERROR("FILTER client %u: bit already clear in bitmap!!\n",
 255                        fed->fed_lr_idx);
 256                 LBUG();
 257         }
 258
 259         memset(&zero_fcd, 0, sizeof zero_fcd);
 260         push_ctxt(&saved, &obd->obd_ctxt, NULL);
 261         rc = fsfilt_write_record(obd, filter->fo_rcvd_filp, &zero_fcd,
 262                                  sizeof(zero_fcd), &off, 1);
 263         pop_ctxt(&saved, &obd->obd_ctxt, NULL);
 264
 265         CDEBUG(rc == 0 ? D_INFO : D_ERROR,
 266                "zeroing disconnecting client %s at idx %u (%llu) in %s rc %d\n",
 267                fed->fed_fcd->fcd_uuid, fed->fed_lr_idx, fed->fed_lr_off,
 268                LAST_RCVD, rc);
 269
 270         if (!test_and_clear_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
 271                 CERROR("FILTER client %u: bit already clear in bitmap!!\n",
 272                        fed->fed_lr_idx);
 273                 LBUG();
 274         }
 275
 276 free:
 277         OBD_FREE(fed->fed_fcd, sizeof(*fed->fed_fcd));
 278
 279         RETURN(0);
 280 }
 281
 282 static int filter_free_server_data(struct filter_obd *filter)
 283 {
 284         OBD_FREE(filter->fo_fsd, sizeof(*filter->fo_fsd));
 285         filter->fo_fsd = NULL;
 286         OBD_FREE(filter->fo_last_rcvd_slots,
 287                  FILTER_LR_MAX_CLIENT_WORDS * sizeof(unsigned long));
 288         filter->fo_last_rcvd_slots = NULL;
 289         return 0;
 290 }
 291
 292 /* assumes caller is already in kernel ctxt */
 293 int filter_update_server_data(struct obd_device *obd, struct file *filp,
 294                               struct filter_server_data *fsd, int force_sync)
 295 {
 296         loff_t off = 0;
 297         int rc;
 298         ENTRY;
 299
 300         CDEBUG(D_INODE, "server uuid      : %s\n", fsd->fsd_uuid);
 301         CDEBUG(D_INODE, "server last_rcvd : "LPU64"\n",
 302                le64_to_cpu(fsd->fsd_last_transno));
 303         CDEBUG(D_INODE, "server last_mount: "LPU64"\n",
 304                le64_to_cpu(fsd->fsd_mount_count));
 305
 306         rc = fsfilt_write_record(obd, filp, fsd, sizeof(*fsd), &off,force_sync);
 307         if (rc)
 308                 CERROR("error writing filter_server_data: rc = %d\n", rc);
 309
 310         RETURN(rc);
 311 }
 312
 313 int filter_update_last_objid(struct obd_device *obd, obd_gr group,
 314                              int force_sync)
 315 {
 316         struct filter_obd *filter = &obd->u.filter;
 317         __u64 tmp;
 318         loff_t off = 0;
 319         int rc;
 320         ENTRY;
 321
 322         CDEBUG(D_INODE, "server last_objid for group "LPU64": "LPU64"\n",
 323                group, filter->fo_last_objids[group]);
 324
 325         tmp = cpu_to_le64(filter->fo_last_objids[group]);
 326         rc = fsfilt_write_record(obd, filter->fo_last_objid_files[group],
 327                                  &tmp, sizeof(tmp), &off, force_sync);
 328         if (rc)
 329                 CERROR("error writing group "LPU64" last objid: rc = %d\n",
 330                        group, rc);
 331         RETURN(rc);
 332 }
 333
 334 /* assumes caller has already in kernel ctxt */
 335 static int filter_init_server_data(struct obd_device *obd, struct file * filp)
 336 {
 337         struct filter_obd *filter = &obd->u.filter;
 338         struct filter_server_data *fsd;
 339         struct filter_client_data *fcd = NULL;
 340         struct inode *inode = filp->f_dentry->d_inode;
 341         unsigned long last_rcvd_size = inode->i_size;
 342         __u64 mount_count;
 343         int cl_idx;
 344         loff_t off = 0;
 345         int rc;
 346
 347         /* ensure padding in the struct is the correct size */
 348         LASSERT (offsetof(struct filter_server_data, fsd_padding) +
 349                  sizeof(fsd->fsd_padding) == FILTER_LR_SERVER_SIZE);
 350         LASSERT (offsetof(struct filter_client_data, fcd_padding) +
 351                  sizeof(fcd->fcd_padding) == FILTER_LR_CLIENT_SIZE);
 352
 353         OBD_ALLOC(fsd, sizeof(*fsd));
 354         if (!fsd)
 355                 RETURN(-ENOMEM);
 356         filter->fo_fsd = fsd;
 357
 358         OBD_ALLOC(filter->fo_last_rcvd_slots,
 359                   FILTER_LR_MAX_CLIENT_WORDS * sizeof(unsigned long));
 360         if (filter->fo_last_rcvd_slots == NULL) {
 361                 OBD_FREE(fsd, sizeof(*fsd));
 362                 RETURN(-ENOMEM);
 363         }
 364
 365         if (last_rcvd_size == 0) {
 366                 CWARN("%s: initializing new %s\n", obd->obd_name, LAST_RCVD);
 367
 368                 memcpy(fsd->fsd_uuid, obd->obd_uuid.uuid,sizeof(fsd->fsd_uuid));
 369                 fsd->fsd_last_transno = 0;
 370                 mount_count = fsd->fsd_mount_count = 0;
 371                 fsd->fsd_server_size = cpu_to_le32(FILTER_LR_SERVER_SIZE);
 372                 fsd->fsd_client_start = cpu_to_le32(FILTER_LR_CLIENT_START);
 373                 fsd->fsd_client_size = cpu_to_le16(FILTER_LR_CLIENT_SIZE);
 374                 fsd->fsd_subdir_count = cpu_to_le16(FILTER_SUBDIR_COUNT);
 375                 filter->fo_subdir_count = FILTER_SUBDIR_COUNT;
 376         } else {
 377                 rc = fsfilt_read_record(obd, filp, fsd, sizeof(*fsd), &off);
 378                 if (rc) {
 379                         CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
 380                                LAST_RCVD, rc);
 381                         GOTO(err_fsd, rc);
 382                 }
 383                 if (strcmp(fsd->fsd_uuid, obd->obd_uuid.uuid) != 0) {
 384                         CERROR("OBD UUID %s does not match last_rcvd UUID %s\n",
 385                                obd->obd_uuid.uuid, fsd->fsd_uuid);
 386                         GOTO(err_fsd, rc = -EINVAL);
 387                 }
 388                 mount_count = le64_to_cpu(fsd->fsd_mount_count);
 389                 filter->fo_subdir_count = le16_to_cpu(fsd->fsd_subdir_count);
 390         }
 391
 392         if (fsd->fsd_feature_incompat & ~cpu_to_le32(FILTER_INCOMPAT_SUPP)) {
 393                 CERROR("unsupported feature %x\n",
 394                        le32_to_cpu(fsd->fsd_feature_incompat) &
 395                        ~FILTER_INCOMPAT_SUPP);
 396                 GOTO(err_fsd, rc = -EINVAL);
 397         }
 398         if (fsd->fsd_feature_rocompat & ~cpu_to_le32(FILTER_ROCOMPAT_SUPP)) {
 399                 CERROR("read-only feature %x\n",
 400                        le32_to_cpu(fsd->fsd_feature_rocompat) &
 401                        ~FILTER_ROCOMPAT_SUPP);
 402                 /* Do something like remount filesystem read-only */
 403                 GOTO(err_fsd, rc = -EINVAL);
 404         }
 405
 406         CDEBUG(D_INODE, "%s: server last_rcvd : "LPU64"\n",
 407                obd->obd_name, le64_to_cpu(fsd->fsd_last_transno));
 408         CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
 409                obd->obd_name, mount_count + 1);
 410         CDEBUG(D_INODE, "%s: server data size: %u\n",
 411                obd->obd_name, le32_to_cpu(fsd->fsd_server_size));
 412         CDEBUG(D_INODE, "%s: per-client data start: %u\n",
 413                obd->obd_name, le32_to_cpu(fsd->fsd_client_start));
 414         CDEBUG(D_INODE, "%s: per-client data size: %u\n",
 415                obd->obd_name, le32_to_cpu(fsd->fsd_client_size));
 416         CDEBUG(D_INODE, "%s: server subdir_count: %u\n",
 417                obd->obd_name, le16_to_cpu(fsd->fsd_subdir_count));
 418         CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
 419                last_rcvd_size <= FILTER_LR_CLIENT_START ? 0 :
 420                (last_rcvd_size-FILTER_LR_CLIENT_START) /FILTER_LR_CLIENT_SIZE);
 421
 422         if (!obd->obd_replayable) {
 423                 CWARN("%s: recovery support OFF\n", obd->obd_name);
 424                 GOTO(out, rc = 0);
 425         }
 426
 427         for (cl_idx = 0, off = le32_to_cpu(fsd->fsd_client_start);
 428              off < last_rcvd_size; cl_idx++) {
 429                 __u64 last_rcvd;
 430                 struct obd_export *exp;
 431                 struct filter_export_data *fed;
 432
 433                 if (!fcd) {
 434                         OBD_ALLOC(fcd, sizeof(*fcd));
 435                         if (!fcd)
 436                                 GOTO(err_client, rc = -ENOMEM);
 437                 }
 438
 439                 /* Don't assume off is incremented properly by
 440                  * fsfilt_read_record(), in case sizeof(*fcd)
 441                  * isn't the same as fsd->fsd_client_size.  */
 442                 off = le32_to_cpu(fsd->fsd_client_start) +
 443                         cl_idx * le16_to_cpu(fsd->fsd_client_size);
 444                 rc = fsfilt_read_record(obd, filp, fcd, sizeof(*fcd), &off);
 445                 if (rc) {
 446                         CERROR("error reading FILT %s idx %d off %llu: rc %d\n",
 447                                LAST_RCVD, cl_idx, off, rc);
 448                         break; /* read error shouldn't cause startup to fail */
 449                 }
 450
 451                 if (fcd->fcd_uuid[0] == '\0') {
 452                         CDEBUG(D_INFO, "skipping zeroed client at offset %d\n",
 453                                cl_idx);
 454                         continue;
 455                 }
 456
 457                 last_rcvd = le64_to_cpu(fcd->fcd_last_rcvd);
 458
 459                 /* These exports are cleaned up by filter_disconnect(), so they
 460                  * need to be set up like real exports as filter_connect() does.
 461                  */
 462                 exp = class_new_export(obd);
 463                 CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64
 464                        " srv lr: "LPU64"\n", fcd->fcd_uuid, cl_idx,
 465                        last_rcvd, le64_to_cpu(fsd->fsd_last_transno));
 466                 if (exp == NULL)
 467                         GOTO(err_client, rc = -ENOMEM);
 468
 469                 memcpy(&exp->exp_client_uuid.uuid, fcd->fcd_uuid,
 470                        sizeof exp->exp_client_uuid.uuid);
 471                 fed = &exp->exp_filter_data;
 472                 fed->fed_fcd = fcd;
 473                 filter_client_add(obd, filter, fed, cl_idx);
 474                 /* create helper if export init gets more complex */
 475                 spin_lock_init(&fed->fed_lock);
 476
 477                 fcd = NULL;
 478                 obd->obd_recoverable_clients++;
 479                 class_export_put(exp);
 480
 481                 CDEBUG(D_OTHER, "client at idx %d has last_rcvd = "LPU64"\n",
 482                        cl_idx, last_rcvd);
 483
 484                 if (last_rcvd > le64_to_cpu(fsd->fsd_last_transno))
 485                         fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
 486
 487         }
 488
 489         obd->obd_last_committed = le64_to_cpu(fsd->fsd_last_transno);
 490
 491         if (obd->obd_recoverable_clients) {
 492                 CWARN("RECOVERY: %d recoverable clients, last_rcvd "
 493                        LPU64"\n", obd->obd_recoverable_clients,
 494                        le64_to_cpu(fsd->fsd_last_transno));
 495                 obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
 496                 obd->obd_recovering = 1;
 497         }
 498
 499         if (fcd)
 500                 OBD_FREE(fcd, sizeof(*fcd));
 501
 502 out:
 503         filter->fo_mount_count = mount_count + 1;
 504         fsd->fsd_mount_count = cpu_to_le64(filter->fo_mount_count);
 505
 506         /* save it, so mount count and last_transno is current */
 507         rc = filter_update_server_data(obd, filp, filter->fo_fsd, 1);
 508
 509         RETURN(rc);
 510
 511 err_client:
 512         class_disconnect_exports(obd, 0);
 513 err_fsd:
 514         filter_free_server_data(filter);
 515         RETURN(rc);
 516 }
 517
 518 static int filter_cleanup_groups(struct obd_device *obd)
 519 {
 520         struct filter_obd *filter = &obd->u.filter;
 521         int i;
 522         ENTRY;
 523
 524         if (filter->fo_dentry_O_groups != NULL &&
 525             filter->fo_last_objids != NULL &&
 526             filter->fo_last_objid_files != NULL) {
 527                 for (i = 0; i < FILTER_GROUPS; i++) {
 528                         struct dentry *dentry = filter->fo_dentry_O_groups[i];
 529                         struct file *filp = filter->fo_last_objid_files[i];
 530                         if (dentry != NULL) {
 531                                 f_dput(dentry);
 532                                 filter->fo_dentry_O_groups[i] = NULL;
 533                         }
 534                         if (filp != NULL) {
 535                                 filp_close(filp, 0);
 536                                 filter->fo_last_objid_files[i] = NULL;
 537                         }
 538                 }
 539         }
 540         if (filter->fo_dentry_O_sub != NULL && filter->fo_subdir_count) {
 541                 for (i = 0; i < filter->fo_subdir_count; i++) {
 542                         struct dentry *dentry = filter->fo_dentry_O_sub[i];
 543                         if (dentry != NULL) {
 544                                 f_dput(dentry);
 545                                 filter->fo_dentry_O_sub[i] = NULL;
 546                         }
 547                 }
 548                 OBD_FREE(filter->fo_dentry_O_sub,
 549                          filter->fo_subdir_count *
 550                          sizeof(*filter->fo_dentry_O_sub));
 551         }
 552         if (filter->fo_dentry_O_groups != NULL)
 553                 OBD_FREE(filter->fo_dentry_O_groups,
 554                          FILTER_GROUPS * sizeof(struct dentry *));
 555         if (filter->fo_last_objids != NULL)
 556                 OBD_FREE(filter->fo_last_objids,
 557                          FILTER_GROUPS * sizeof(__u64));
 558         if (filter->fo_last_objid_files != NULL)
 559                 OBD_FREE(filter->fo_last_objid_files,
 560                          FILTER_GROUPS * sizeof(struct file *));
 561         RETURN(0);
 562 }
 563
 564 /* FIXME: object groups */
 565 static int filter_prep_groups(struct obd_device *obd)
 566 {
 567         struct filter_obd *filter = &obd->u.filter;
 568         struct dentry *dentry, *O_dentry;
 569         struct file *filp;
 570         int i, rc = 0, cleanup_phase = 0;
 571         ENTRY;
 572
 573         O_dentry = simple_mkdir(current->fs->pwd, "O", 0700, 1);
 574         CDEBUG(D_INODE, "got/created O: %p\n", O_dentry);
 575         if (IS_ERR(O_dentry)) {
 576                 rc = PTR_ERR(O_dentry);
 577                 CERROR("cannot open/create O: rc = %d\n", rc);
 578                 GOTO(cleanup, rc);
 579         }
 580         filter->fo_dentry_O = O_dentry;
 581         cleanup_phase = 1; /* O_dentry */
 582
 583         /* Lookup "R" to tell if we're on an old OST FS and need to convert
 584          * from O/R/<dir>/<objid> to O/0/<dir>/<objid>.  This can be removed
 585          * some time post 1.0 when all old-style OSTs have converted along
 586          * with the init_objid hack. */
 587         dentry = ll_lookup_one_len("R", O_dentry, 1);
 588         if (IS_ERR(dentry))
 589                 GOTO(cleanup, rc = PTR_ERR(dentry));
 590         if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
 591                 struct dentry *O0_dentry = lookup_one_len("0", O_dentry, 1);
 592                 ENTRY;
 593
 594                 CWARN("converting OST to new object layout\n");
 595                 if (IS_ERR(O0_dentry)) {
 596                         rc = PTR_ERR(O0_dentry);
 597                         CERROR("error looking up O/0: rc %d\n", rc);
 598                         GOTO(cleanup_R, rc);
 599                 }
 600
 601                 if (O0_dentry->d_inode) {
 602                         CERROR("Both O/R and O/0 exist. Fix manually.\n");
 603                         GOTO(cleanup_O0, rc = -EEXIST);
 604                 }
 605
 606                 down(&O_dentry->d_inode->i_sem);
 607                 rc = vfs_rename(O_dentry->d_inode, dentry,
 608                                 O_dentry->d_inode, O0_dentry);
 609                 up(&O_dentry->d_inode->i_sem);
 610
 611                 if (rc) {
 612                         CERROR("error renaming O/R to O/0: rc %d\n", rc);
 613                         GOTO(cleanup_O0, rc);
 614                 }
 615                 filter->fo_fsd->fsd_feature_incompat |=
 616                         cpu_to_le32(FILTER_INCOMPAT_GROUPS);
 617                 rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
 618                                                filter->fo_fsd, 1);
 619                 GOTO(cleanup_O0, rc);
 620
 621         cleanup_O0:
 622                 dput(O0_dentry);
 623         cleanup_R:
 624                 dput(dentry);
 625                 if (rc)
 626                         GOTO(cleanup, rc);
 627         } else {
 628                 dput(dentry);
 629         }
 630
 631         OBD_ALLOC(filter->fo_last_objids, FILTER_GROUPS * sizeof(__u64));
 632         if (filter->fo_last_objids == NULL)
 633                 GOTO(cleanup, rc = -ENOMEM);
 634         cleanup_phase = 2; /* groups */
 635
 636         OBD_ALLOC(filter->fo_dentry_O_groups, FILTER_GROUPS * sizeof(dentry));
 637         if (filter->fo_dentry_O_groups == NULL)
 638                 GOTO(cleanup, rc = -ENOMEM);
 639         OBD_ALLOC(filter->fo_last_objid_files, FILTER_GROUPS * sizeof(filp));
 640         if (filter->fo_last_objid_files == NULL)
 641                 GOTO(cleanup, rc = -ENOMEM);
 642
 643         for (i = 0; i < FILTER_GROUPS; i++) {
 644                 char name[25];
 645                 loff_t off = 0;
 646
 647                 sprintf(name, "%d", i);
 648                 dentry = simple_mkdir(O_dentry, name, 0700, 1);
 649                 CDEBUG(D_INODE, "got/created O/%s: %p\n", name, dentry);
 650                 if (IS_ERR(dentry)) {
 651                         rc = PTR_ERR(dentry);
 652                         CERROR("cannot lookup/create O/%s: rc = %d\n",
 653                                name, rc);
 654                         GOTO(cleanup, rc);
 655                 }
 656                 filter->fo_dentry_O_groups[i] = dentry;
 657
 658                 sprintf(name, "O/%d/LAST_ID", i);
 659                 filp = filp_open(name, O_CREAT | O_RDWR, 0700);
 660                 if (IS_ERR(dentry)) {
 661                         rc = PTR_ERR(dentry);
 662                         CERROR("cannot create %s: rc = %d\n", name, rc);
 663                         GOTO(cleanup, rc);
 664                 }
 665                 filter->fo_last_objid_files[i] = filp;
 666
 667                 if (filp->f_dentry->d_inode->i_size == 0) {
 668                         if (i == 0 && filter->fo_fsd->fsd_unused != 0) {
 669                                 /* OST conversion, remove sometime post 1.0 */
 670                                 filter->fo_last_objids[i] =
 671                                         le64_to_cpu(filter->fo_fsd->fsd_unused);
 672                                 CWARN("saving old objid "LPU64" to LAST_ID\n",
 673                                       filter->fo_last_objids[i]);
 674                                 rc = filter_update_last_objid(obd, 0, 1);
 675                                 if (rc)
 676                                         GOTO(cleanup, rc);
 677                         } else {
 678                                 filter->fo_last_objids[i] = FILTER_INIT_OBJID;
 679                         }
 680                         continue;
 681                 }
 682
 683                 rc = fsfilt_read_record(obd, filp, &filter->fo_last_objids[i],
 684                                         sizeof(__u64), &off);
 685                 if (rc) {
 686                         CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
 687                                name, rc);
 688                         GOTO(cleanup, rc);
 689                 }
 690                 filter->fo_last_objids[i] =
 691                         le64_to_cpu(filter->fo_last_objids[i]);
 692                 CDEBUG(D_INODE, "%s: server last_objid group %d: "LPU64"\n",
 693                        obd->obd_name, i, filter->fo_last_objids[i]);
 694         }
 695
 696         if (filter->fo_subdir_count) {
 697                 O_dentry = filter->fo_dentry_O_groups[0];
 698                 OBD_ALLOC(filter->fo_dentry_O_sub,
 699                           filter->fo_subdir_count * sizeof(dentry));
 700                 if (filter->fo_dentry_O_sub == NULL)
 701                         GOTO(cleanup, rc = -ENOMEM);
 702
 703                 for (i = 0; i < filter->fo_subdir_count; i++) {
 704                         char dir[20];
 705                         snprintf(dir, sizeof(dir), "d%u", i);
 706
 707                         dentry = simple_mkdir(O_dentry, dir, 0700, 1);
 708                         CDEBUG(D_INODE, "got/created O/0/%s: %p\n", dir,dentry);
 709                         if (IS_ERR(dentry)) {
 710                                 rc = PTR_ERR(dentry);
 711                                 CERROR("can't lookup/create O/0/%s: rc = %d\n",
 712                                        dir, rc);
 713                                 GOTO(cleanup, rc);
 714                         }
 715                         filter->fo_dentry_O_sub[i] = dentry;
 716                 }
 717         }
 718         RETURN(0);
 719
 720  cleanup:
 721         switch (cleanup_phase) {
 722         case 2:
 723                 filter_cleanup_groups(obd);
 724         case 1:
 725                 f_dput(filter->fo_dentry_O);
 726                 filter->fo_dentry_O = NULL;
 727         default:
 728                 break;
 729         }
 730         return rc;
 731 }
 732
 733 /* setup the object store with correct subdirectories */
 734 static int filter_prep(struct obd_device *obd)
 735 {
 736         struct obd_run_ctxt saved;
 737         struct filter_obd *filter = &obd->u.filter;
 738         struct file *file;
 739         struct inode *inode;
 740         int rc = 0;
 741         ENTRY;
 742
 743         push_ctxt(&saved, &obd->obd_ctxt, NULL);
 744         file = filp_open(LAST_RCVD, O_RDWR | O_CREAT | O_LARGEFILE, 0700);
 745         if (!file || IS_ERR(file)) {
 746                 rc = PTR_ERR(file);
 747                 CERROR("OBD filter: cannot open/create %s: rc = %d\n",
 748                        LAST_RCVD, rc);
 749                 GOTO(out, rc);
 750         }
 751
 752         if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
 753                 CERROR("%s is not a regular file!: mode = %o\n", LAST_RCVD,
 754                        file->f_dentry->d_inode->i_mode);
 755                 GOTO(err_filp, rc = -ENOENT);
 756         }
 757
 758         /* steal operations */
 759         inode = file->f_dentry->d_inode;
 760         filter->fo_fop = file->f_op;
 761         filter->fo_iop = inode->i_op;
 762         filter->fo_aops = inode->i_mapping->a_ops;
 763
 764         rc = filter_init_server_data(obd, file);
 765         if (rc) {
 766                 CERROR("cannot read %s: rc = %d\n", LAST_RCVD, rc);
 767                 GOTO(err_filp, rc);
 768         }
 769         filter->fo_rcvd_filp = file;
 770
 771         rc = filter_prep_groups(obd);
 772         if (rc)
 773                 GOTO(err_server_data, rc);
 774
 775  out:
 776         pop_ctxt(&saved, &obd->obd_ctxt, NULL);
 777
 778         return(rc);
 779
 780  err_server_data:
 781         //class_disconnect_exports(obd, 0);
 782         filter_free_server_data(filter);
 783  err_filp:
 784         if (filp_close(file, 0))
 785                 CERROR("can't close %s after error\n", LAST_RCVD);
 786         filter->fo_rcvd_filp = NULL;
 787         goto out;
 788 }
 789
 790 /* cleanup the filter: write last used object id to status file */
 791 static void filter_post(struct obd_device *obd)
 792 {
 793         struct obd_run_ctxt saved;
 794         struct filter_obd *filter = &obd->u.filter;
 795         int rc, i;
 796
 797         /* XXX: filter_update_lastobjid used to call fsync_dev.  It might be
 798          * best to start a transaction with h_sync, because we removed this
 799          * from lastobjid */
 800
 801         push_ctxt(&saved, &obd->obd_ctxt, NULL);
 802         rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
 803                                        filter->fo_fsd, 0);
 804         if (rc)
 805                 CERROR("error writing server data: rc = %d\n", rc);
 806
 807         for (i = 0; i < FILTER_GROUPS; i++) {
 808                 rc = filter_update_last_objid(obd, i, (i == FILTER_GROUPS - 1));
 809                 if (rc)
 810                         CERROR("error writing group %d lastobjid: rc = %d\n",
 811                                i, rc);
 812         }
 813
 814         filp_close(filter->fo_rcvd_filp, 0);
 815         filter->fo_rcvd_filp = NULL;
 816         if (rc)
 817                 CERROR("error closing %s: rc = %d\n", LAST_RCVD, rc);
 818
 819         filter_cleanup_groups(obd);
 820         f_dput(filter->fo_dentry_O);
 821         filter_free_server_data(filter);
 822         pop_ctxt(&saved, &obd->obd_ctxt, NULL);
 823 }
 824
 825 static void filter_set_last_id(struct filter_obd *filter, struct obdo *oa,
 826                                obd_id id)
 827 {
 828         obd_gr group = 0;
 829         LASSERT(filter->fo_fsd != NULL);
 830
 831         if (oa != NULL) {
 832                 LASSERT(oa->o_gr <= FILTER_GROUPS);
 833                 group = oa->o_gr;
 834         }
 835
 836         spin_lock(&filter->fo_objidlock);
 837         filter->fo_last_objids[group] = id;
 838         spin_unlock(&filter->fo_objidlock);
 839 }
 840
 841 __u64 filter_last_id(struct filter_obd *filter, struct obdo *oa)
 842 {
 843         obd_id id;
 844         obd_gr group = 0;
 845         LASSERT(filter->fo_fsd != NULL);
 846
 847         if (oa != NULL) {
 848                 LASSERT(oa->o_gr <= FILTER_GROUPS);
 849                 group = oa->o_gr;
 850         }
 851
 852         /* FIXME: object groups */
 853         spin_lock(&filter->fo_objidlock);
 854         id = filter->fo_last_objids[group];
 855         spin_unlock(&filter->fo_objidlock);
 856
 857         return id;
 858 }
 859
 860 /* direct cut-n-paste of mds_blocking_ast() */
 861 static int filter_blocking_ast(struct ldlm_lock *lock,
 862                                struct ldlm_lock_desc *desc,
 863                                void *data, int flag)
 864 {
 865         int do_ast;
 866         ENTRY;
 867
 868         if (flag == LDLM_CB_CANCELING) {
 869                 /* Don't need to do anything here. */
 870                 RETURN(0);
 871         }
 872
 873         /* XXX layering violation!  -phil */
 874         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 875         /* Get this: if filter_blocking_ast is racing with ldlm_intent_policy,
 876          * such that filter_blocking_ast is called just before l_i_p takes the
 877          * ns_lock, then by the time we get the lock, we might not be the
 878          * correct blocking function anymore.  So check, and return early, if
 879          * so. */
 880         if (lock->l_blocking_ast != filter_blocking_ast) {
 881                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 882                 RETURN(0);
 883         }
 884
 885         lock->l_flags |= LDLM_FL_CBPENDING;
 886         do_ast = (!lock->l_readers && !lock->l_writers);
 887         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 888
 889         if (do_ast) {
 890                 struct lustre_handle lockh;
 891                 int rc;
 892
 893                 LDLM_DEBUG(lock, "already unused, calling ldlm_cli_cancel");
 894                 ldlm_lock2handle(lock, &lockh);
 895                 rc = ldlm_cli_cancel(&lockh);
 896                 if (rc < 0)
 897                         CERROR("ldlm_cli_cancel: %d\n", rc);
 898         } else {
 899                 LDLM_DEBUG(lock, "Lock still has references, will be "
 900                            "cancelled later");
 901         }
 902         RETURN(0);
 903 }
 904
 905 static int filter_lock_dentry(struct obd_device *obd, struct dentry *dparent)
 906 {
 907         down(&dparent->d_inode->i_sem);
 908         return 0;
 909 }
 910
 911 /* We never dget the object parent, so DON'T dput it either */
 912 static void filter_parent_unlock(struct dentry *dparent)
 913 {
 914         up(&dparent->d_inode->i_sem);
 915 }
 916
 917 /* We never dget the object parent, so DON'T dput it either */
 918 struct dentry *filter_parent(struct obd_device *obd, obd_gr group, obd_id objid)
 919 {
 920         struct filter_obd *filter = &obd->u.filter;
 921         LASSERT(group < FILTER_GROUPS); /* FIXME: object groups */
 922
 923         if (group > 0 || filter->fo_subdir_count == 0)
 924                 return filter->fo_dentry_O_groups[group];
 925
 926         return filter->fo_dentry_O_sub[objid & (filter->fo_subdir_count - 1)];
 927 }
 928
 929 /* We never dget the object parent, so DON'T dput it either */
 930 struct dentry *filter_parent_lock(struct obd_device *obd, obd_gr group,
 931                                   obd_id objid)
 932 {
 933         unsigned long now = jiffies;
 934         struct dentry *dparent = filter_parent(obd, group, objid);
 935         int rc;
 936
 937         if (IS_ERR(dparent))
 938                 return dparent;
 939
 940         rc = filter_lock_dentry(obd, dparent);
 941         if (time_after(jiffies, now + 15 * HZ))
 942                 CERROR("slow parent lock %lus\n", (jiffies - now) / HZ);
 943         return rc ? ERR_PTR(rc) : dparent;
 944 }
 945
 946 /* How to get files, dentries, inodes from object id's.
 947  *
 948  * If dir_dentry is passed, the caller has already locked the parent
 949  * appropriately for this operation (normally a write lock).  If
 950  * dir_dentry is NULL, we do a read lock while we do the lookup to
 951  * avoid races with create/destroy and such changing the directory
 952  * internal to the filesystem code. */
 953 struct dentry *filter_fid2dentry(struct obd_device *obd,
 954                                  struct dentry *dir_dentry,
 955                                  obd_gr group, obd_id id)
 956 {
 957         struct dentry *dparent = dir_dentry;
 958         struct dentry *dchild;
 959         char name[32];
 960         int len;
 961         ENTRY;
 962
 963         if (id == 0) {
 964                 CERROR("fatal: invalid object id 0\n");
 965                 RETURN(ERR_PTR(-ESTALE));
 966         }
 967
 968         len = sprintf(name, LPU64, id);
 969         if (dir_dentry == NULL) {
 970                 dparent = filter_parent_lock(obd, group, id);
 971                 if (IS_ERR(dparent))
 972                         RETURN(dparent);
 973         }
 974         CDEBUG(D_INODE, "looking up object O/%*s/%s\n",
 975                dparent->d_name.len, dparent->d_name.name, name);
 976         dchild = /*ll_*/lookup_one_len(name, dparent, len);
 977         if (dir_dentry == NULL)
 978                 filter_parent_unlock(dparent);
 979         if (IS_ERR(dchild)) {
 980                 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
 981                 RETURN(dchild);
 982         }
 983
 984         CDEBUG(D_INODE, "got child objid %s: %p, count = %d\n",
 985                name, dchild, atomic_read(&dchild->d_count));
 986
 987         LASSERT(atomic_read(&dchild->d_count) > 0);
 988
 989         RETURN(dchild);
 990 }
 991
 992 static int filter_prepare_destroy(struct obd_device *obd, obd_id objid)
 993 {
 994         struct lustre_handle lockh;
 995         int flags = LDLM_AST_DISCARD_DATA, rc;
 996         struct ldlm_res_id res_id = { .name = { objid } };
 997         ldlm_policy_data_t policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
 998
 999         ENTRY;
1000         /* Tell the clients that the object is gone now and that they should
1001          * throw away any cached pages. */
1002         rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, res_id,
1003                               LDLM_EXTENT, &policy, LCK_PW,
1004                               &flags, filter_blocking_ast, ldlm_completion_ast,
1005                               NULL, NULL, NULL, 0, NULL, &lockh);
1006
1007         /* We only care about the side-effects, just drop the lock. */
1008         if (rc == ELDLM_OK)
1009                 ldlm_lock_decref(&lockh, LCK_PW);
1010
1011         RETURN(rc);
1012 }
1013
1014 /* Caller must hold LCK_PW on parent and push us into kernel context.
1015  * Caller is also required to ensure that dchild->d_inode exists. */
1016 static int filter_destroy_internal(struct obd_device *obd, obd_id objid,
1017                                    struct dentry *dparent,
1018                                    struct dentry *dchild)
1019 {
1020         struct inode *inode = dchild->d_inode;
1021         int rc;
1022         ENTRY;
1023
1024         if (inode->i_nlink != 1 || atomic_read(&inode->i_count) != 1) {
1025                 CERROR("destroying objid %*s nlink = %lu, count = %d\n",
1026                        dchild->d_name.len, dchild->d_name.name,
1027                        (unsigned long)inode->i_nlink,
1028                        atomic_read(&inode->i_count));
1029         }
1030
1031         rc = vfs_unlink(dparent->d_inode, dchild);
1032
1033         if (rc)
1034                 CERROR("error unlinking objid %*s: rc %d\n",
1035                        dchild->d_name.len, dchild->d_name.name, rc);
1036
1037         RETURN(rc);
1038 }
1039
1040 static int filter_intent_policy(struct ldlm_namespace *ns,
1041                                 struct ldlm_lock **lockp, void *req_cookie,
1042                                 ldlm_mode_t mode, int flags, void *data)
1043 {
1044         struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
1045         struct ptlrpc_request *req = req_cookie;
1046         struct ldlm_lock *lock = *lockp, *l = NULL;
1047         struct ldlm_resource *res = lock->l_resource;
1048         ldlm_processing_policy policy;
1049         struct ost_lvb *res_lvb, *reply_lvb;
1050         struct list_head *tmp;
1051         ldlm_error_t err;
1052         int tmpflags = 0, rc, repsize[2] = {sizeof(struct ldlm_reply),
1053                                             sizeof(struct ost_lvb) };
1054         ENTRY;
1055
1056         policy = ldlm_get_processing_policy(res);
1057         LASSERT(policy != NULL);
1058         LASSERT(req != NULL);
1059
1060         rc = lustre_pack_reply(req, 2, repsize, NULL);
1061         if (rc)
1062                 RETURN(req->rq_status = rc);
1063
1064         reply_lvb = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*reply_lvb));
1065         LASSERT(reply_lvb != NULL);
1066
1067         //fixup_handle_for_resent_req(req, lock, &lockh);
1068
1069         /* If we grant any lock at all, it will be a whole-file read lock.
1070          * Call the extent policy function to see if our request can be
1071          * granted, or is blocked. */
1072         lock->l_policy_data.l_extent.start = 0;
1073         lock->l_policy_data.l_extent.end = OBD_OBJECT_EOF;
1074         lock->l_req_mode = LCK_PR;
1075
1076         l_lock(&res->lr_namespace->ns_lock);
1077
1078         res->lr_tmp = &rpc_list;
1079         rc = policy(lock, &tmpflags, 0, &err);
1080         res->lr_tmp = NULL;
1081
1082         /* FIXME: we should change the policy function slightly, to not make
1083          * this list at all, since we just turn around and free it */
1084         while (!list_empty(&rpc_list)) {
1085                 struct ldlm_ast_work *w =
1086                         list_entry(rpc_list.next, struct ldlm_ast_work, w_list);
1087                 list_del(&w->w_list);
1088                 LDLM_LOCK_PUT(w->w_lock);
1089                 OBD_FREE(w, sizeof(*w));
1090         }
1091
1092         if (rc == LDLM_ITER_CONTINUE) {
1093                 /* The lock met with no resistance; we're finished. */
1094                 l_unlock(&res->lr_namespace->ns_lock);
1095                 RETURN(ELDLM_LOCK_REPLACED);
1096         }
1097
1098         /* Do not grant any lock, but instead send GL callbacks.  The extent
1099          * policy nicely created a list of all PW locks for us.  We will choose
1100          * the highest of those which are larger than the size in the LVB, if
1101          * any, and perform a glimpse callback. */
1102         down(&res->lr_lvb_sem);
1103         res_lvb = res->lr_lvb_data;
1104         LASSERT(res_lvb != NULL);
1105         reply_lvb->lvb_size = res_lvb->lvb_size;
1106         up(&res->lr_lvb_sem);
1107
1108         list_for_each(tmp, &res->lr_granted) {
1109                 struct ldlm_lock *tmplock =
1110                         list_entry(tmp, struct ldlm_lock, l_res_link);
1111
1112                 if (tmplock->l_granted_mode == LCK_PR)
1113                         continue;
1114
1115                 if (tmplock->l_policy_data.l_extent.end <=
1116                     reply_lvb->lvb_size)
1117                         continue;
1118
1119                 if (l == NULL) {
1120                         l = LDLM_LOCK_GET(tmplock);
1121                         continue;
1122                 }
1123
1124                 if (l->l_policy_data.l_extent.start >
1125                     tmplock->l_policy_data.l_extent.start)
1126                         continue;
1127
1128                 LDLM_LOCK_PUT(l);
1129                 l = LDLM_LOCK_GET(tmplock);
1130         }
1131         l_unlock(&res->lr_namespace->ns_lock);
1132
1133         /* There were no PW locks beyond the size in the LVB; finished. */
1134         if (l == NULL)
1135                 RETURN(ELDLM_LOCK_ABORTED);
1136
1137         LASSERT(l->l_glimpse_ast != NULL);
1138         rc = l->l_glimpse_ast(l, NULL); /* this will update the LVB */
1139
1140         down(&res->lr_lvb_sem);
1141 #if 0
1142         if (res_lvb->lvb_size == reply_lvb->lvb_size)
1143                 LDLM_ERROR(l, "we lost the glimpse race!");
1144 #endif
1145         reply_lvb->lvb_size = res_lvb->lvb_size;
1146         up(&res->lr_lvb_sem);
1147
1148         LDLM_LOCK_PUT(l);
1149
1150         RETURN(ELDLM_LOCK_ABORTED);
1151 }
1152
1153 /* mount the file system (secretly) */
1154 int filter_common_setup(struct obd_device *obd, obd_count len, void *buf,
1155                         char *option)
1156 {
1157         struct lustre_cfg* lcfg = buf;
1158         struct filter_obd *filter = &obd->u.filter;
1159         struct vfsmount *mnt;
1160         int rc = 0;
1161         ENTRY;
1162
1163         dev_clear_rdonly(2);
1164
1165         if (!lcfg->lcfg_inlbuf1 || !lcfg->lcfg_inlbuf2)
1166                 RETURN(-EINVAL);
1167
1168         obd->obd_fsops = fsfilt_get_ops(lcfg->lcfg_inlbuf2);
1169         if (IS_ERR(obd->obd_fsops))
1170                 RETURN(PTR_ERR(obd->obd_fsops));
1171
1172         mnt = do_kern_mount(lcfg->lcfg_inlbuf2, MS_NOATIME | MS_NODIRATIME,
1173                             lcfg->lcfg_inlbuf1, option);
1174         rc = PTR_ERR(mnt);
1175         if (IS_ERR(mnt))
1176                 GOTO(err_ops, rc);
1177
1178         if (lcfg->lcfg_inllen3 > 0 && lcfg->lcfg_inlbuf3) {
1179                 if (*lcfg->lcfg_inlbuf3 == 'f') {
1180                         obd->obd_replayable = 1;
1181                         obd_sync_filter = 1;
1182                         CWARN("%s: recovery enabled\n", obd->obd_name);
1183                 } else {
1184                         if (*lcfg->lcfg_inlbuf3 != 'n') {
1185                                 CERROR("unrecognised flag '%c'\n",
1186                                        *lcfg->lcfg_inlbuf3);
1187                         }
1188                         // XXX Robert? Why do we get errors here
1189                         // GOTO(err_mntput, rc = -EINVAL);
1190                 }
1191         }
1192
1193         filter->fo_vfsmnt = mnt;
1194         filter->fo_sb = mnt->mnt_sb;
1195         filter->fo_fstype = mnt->mnt_sb->s_type->name;
1196         CDEBUG(D_SUPER, "%s: mnt = %p\n", filter->fo_fstype, mnt);
1197
1198         OBD_SET_CTXT_MAGIC(&obd->obd_ctxt);
1199         obd->obd_ctxt.pwdmnt = mnt;
1200         obd->obd_ctxt.pwd = mnt->mnt_root;
1201         obd->obd_ctxt.fs = get_ds();
1202         obd->obd_ctxt.cb_ops = filter_lvfs_ops;
1203
1204         rc = filter_prep(obd);
1205         if (rc)
1206                 GOTO(err_mntput, rc);
1207
1208         spin_lock_init(&filter->fo_translock);
1209         spin_lock_init(&filter->fo_objidlock);
1210         INIT_LIST_HEAD(&filter->fo_export_list);
1211         sema_init(&filter->fo_alloc_lock, 1);
1212         spin_lock_init(&filter->fo_r_pages.oh_lock);
1213         spin_lock_init(&filter->fo_w_pages.oh_lock);
1214         spin_lock_init(&filter->fo_r_discont_pages.oh_lock);
1215         spin_lock_init(&filter->fo_w_discont_pages.oh_lock);
1216         spin_lock_init(&filter->fo_r_discont_blocks.oh_lock);
1217         spin_lock_init(&filter->fo_w_discont_blocks.oh_lock);
1218         filter->fo_readcache_max_filesize = FILTER_MAX_CACHE_SIZE;
1219
1220         obd->obd_namespace = ldlm_namespace_new("filter-tgt",
1221                                                 LDLM_NAMESPACE_SERVER);
1222         if (obd->obd_namespace == NULL)
1223                 GOTO(err_post, rc = -ENOMEM);
1224         obd->obd_namespace->ns_lvbp = obd;
1225         obd->obd_namespace->ns_lvbo = &filter_lvbo;
1226         ldlm_register_intent(obd->obd_namespace, filter_intent_policy);
1227
1228         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1229                            "filter_ldlm_cb_client", &obd->obd_ldlm_client);
1230
1231         rc = llog_cat_initialize(obd, 1);
1232         if (rc) {
1233                 CERROR("failed to setup llogging subsystems\n");
1234                 GOTO(err_post, rc);
1235         }
1236
1237         RETURN(0);
1238
1239 err_post:
1240         filter_post(obd);
1241 err_mntput:
1242         unlock_kernel();
1243         mntput(mnt);
1244         filter->fo_sb = 0;
1245         lock_kernel();
1246 err_ops:
1247         fsfilt_put_ops(obd->obd_fsops);
1248         return rc;
1249 }
1250
1251 static int filter_setup(struct obd_device *obd, obd_count len, void *buf)
1252 {
1253         struct lustre_cfg* lcfg = buf;
1254         const char *str = NULL;
1255         char *option = NULL;
1256         int n = 0;
1257         int rc;
1258
1259         if (!strcmp(lcfg->lcfg_inlbuf2, "ext3")) {
1260 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1261         /* bug 1577: implement async-delete for 2.5 */
1262                 str = "errors=remount-ro,asyncdel";
1263 #else
1264                 str = "errors=remount-ro";
1265 #endif
1266                 n = strlen(str) + 1;
1267                 OBD_ALLOC(option, n);
1268                 if (option == NULL)
1269                         RETURN(-ENOMEM);
1270                 strcpy(option, str);
1271         }
1272
1273         rc = filter_common_setup(obd, len, buf, option);
1274         if (option)
1275                 OBD_FREE(option, n);
1276         return rc;
1277 }
1278
1279 static int filter_cleanup(struct obd_device *obd, int flags)
1280 {
1281         struct filter_obd *filter = &obd->u.filter;
1282         ENTRY;
1283
1284         if (flags & OBD_OPT_FAILOVER)
1285                 CERROR("%s: shutting down for failover; client state will"
1286                        " be preserved.\n", obd->obd_name);
1287
1288         if (!list_empty(&obd->obd_exports)) {
1289                 CERROR("%s: still has clients!\n", obd->obd_name);
1290                 class_disconnect_exports(obd, flags);
1291                 if (!list_empty(&obd->obd_exports)) {
1292                         CERROR("still has exports after forced cleanup?\n");
1293                         RETURN(-EBUSY);
1294                 }
1295         }
1296
1297         ldlm_namespace_free(obd->obd_namespace, flags & OBD_OPT_FORCE);
1298
1299         if (filter->fo_sb == NULL)
1300                 RETURN(0);
1301
1302         filter_post(obd);
1303
1304         shrink_dcache_parent(filter->fo_sb->s_root);
1305         filter->fo_sb = 0;
1306
1307         if (atomic_read(&filter->fo_vfsmnt->mnt_count) > 1)
1308                 CERROR("%s: mount point %p busy, mnt_count: %d\n",
1309                        obd->obd_name, filter->fo_vfsmnt,
1310                        atomic_read(&filter->fo_vfsmnt->mnt_count));
1311
1312         unlock_kernel();
1313         mntput(filter->fo_vfsmnt);
1314         //destroy_buffers(filter->fo_sb->s_dev);
1315         filter->fo_sb = NULL;
1316         fsfilt_put_ops(obd->obd_fsops);
1317         lock_kernel();
1318
1319         dev_clear_rdonly(2);
1320
1321         RETURN(0);
1322 }
1323
1324 static int filter_attach(struct obd_device *obd, obd_count len, void *data)
1325 {
1326         struct lprocfs_static_vars lvars;
1327         int rc;
1328
1329         lprocfs_init_vars(filter, &lvars);
1330         rc = lprocfs_obd_attach(obd, lvars.obd_vars);
1331         if (rc != 0)
1332                 return rc;
1333
1334         rc = lprocfs_alloc_obd_stats(obd, LPROC_FILTER_LAST);
1335         if (rc != 0)
1336                 return rc;
1337
1338         /* Init obdfilter private stats here */
1339         lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_READ_BYTES,
1340                              LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes");
1341         lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_WRITE_BYTES,
1342                              LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes");
1343
1344         return lproc_filter_attach_seqstat(obd);
1345 }
1346
1347 static int filter_detach(struct obd_device *dev)
1348 {
1349         lprocfs_free_obd_stats(dev);
1350         return lprocfs_obd_detach(dev);
1351 }
1352
1353 /* nearly identical to mds_connect */
1354 static int filter_connect(struct lustre_handle *conn, struct obd_device *obd,
1355                           struct obd_uuid *cluuid)
1356 {
1357         struct obd_export *exp;
1358         struct filter_export_data *fed;
1359         struct filter_client_data *fcd = NULL;
1360         struct filter_obd *filter = &obd->u.filter;
1361         int rc;
1362         ENTRY;
1363
1364         if (conn == NULL || obd == NULL || cluuid == NULL)
1365                 RETURN(-EINVAL);
1366
1367         rc = class_connect(conn, obd, cluuid);
1368         if (rc)
1369                 RETURN(rc);
1370         exp = class_conn2export(conn);
1371         LASSERT(exp != NULL);
1372
1373         fed = &exp->exp_filter_data;
1374
1375         spin_lock_init(&fed->fed_lock);
1376
1377         if (!obd->obd_replayable)
1378                 GOTO(cleanup, rc = 0);
1379
1380         OBD_ALLOC(fcd, sizeof(*fcd));
1381         if (!fcd) {
1382                 CERROR("filter: out of memory for client data\n");
1383                 GOTO(cleanup, rc = -ENOMEM);
1384         }
1385
1386         memcpy(fcd->fcd_uuid, cluuid, sizeof(fcd->fcd_uuid));
1387         fed->fed_fcd = fcd;
1388
1389         rc = filter_client_add(obd, filter, fed, -1);
1390
1391 cleanup:
1392         if (rc) {
1393                 if (fcd)
1394                         OBD_FREE(fcd, sizeof(*fcd));
1395                 class_disconnect(exp, 0);
1396         } else {
1397                 class_export_put(exp);
1398         }
1399         return rc;
1400 }
1401
1402 static int filter_precleanup(struct obd_device *obd, int flags)
1403 {
1404         int rc = 0;
1405         ENTRY;
1406
1407         rc = obd_llog_finish(obd, 0);
1408         if (rc)
1409                 CERROR("failed to cleanup llogging subsystem\n");
1410
1411         RETURN(rc);
1412 }
1413
1414 /* Do extra sanity checks for grant accounting.  We do this at connect,
1415  * disconnect, and statfs RPC time, so it shouldn't be too bad.  We can
1416  * always get rid of it or turn it off when we know accounting is good. */
1417 static void filter_grant_sanity_check(struct obd_device *obd, char *func)
1418 {
1419         struct filter_export_data *fed;
1420         struct obd_export *exp;
1421         obd_size maxsize = obd->obd_osfs.os_blocks * obd->obd_osfs.os_bsize;
1422         obd_size tot_dirty = 0, tot_pending = 0, tot_granted = 0;
1423         obd_size fo_tot_dirty, fo_tot_pending, fo_tot_granted;
1424
1425         if (list_empty(&obd->obd_exports))
1426                 return;
1427
1428         spin_lock(&obd->obd_osfs_lock);
1429         spin_lock(&obd->obd_dev_lock);
1430         list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
1431                 fed = &exp->exp_filter_data;
1432                 LASSERTF(fed->fed_grant + fed->fed_pending <= maxsize,
1433                          "cli %s/%p %lu+%lu > "LPU64"\n",
1434                          exp->exp_client_uuid.uuid, exp,
1435                          fed->fed_grant, fed->fed_pending, maxsize);
1436                 LASSERTF(fed->fed_dirty <= maxsize, "cli %s/%p %lu > "LPU64"\n",
1437                          exp->exp_client_uuid.uuid, exp,fed->fed_dirty,maxsize);
1438                 CDEBUG(D_CACHE,"%s: cli %s/%p dirty %lu pend %lu grant %lu\n",
1439                        obd->obd_name, exp->exp_client_uuid.uuid, exp,
1440                        fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1441                 tot_granted += fed->fed_grant + fed->fed_pending;
1442                 tot_pending += fed->fed_pending;
1443                 tot_dirty += fed->fed_dirty;
1444         }
1445         fo_tot_granted = obd->u.filter.fo_tot_granted;
1446         fo_tot_pending = obd->u.filter.fo_tot_pending;
1447         fo_tot_dirty = obd->u.filter.fo_tot_dirty;
1448         spin_unlock(&obd->obd_dev_lock);
1449         spin_unlock(&obd->obd_osfs_lock);
1450
1451         /* Do these assertions outside the spinlocks so we don't kill system */
1452         LASSERTF(tot_granted == fo_tot_granted, "%s "LPU64" != "LPU64"\n",
1453                  func, tot_granted, fo_tot_granted);
1454         LASSERTF(tot_pending == fo_tot_pending, "%s "LPU64" != "LPU64"\n",
1455                  func, tot_pending, fo_tot_pending);
1456         LASSERTF(tot_dirty == fo_tot_dirty, "%s "LPU64" != "LPU64"\n",
1457                  func, tot_dirty, fo_tot_dirty);
1458         LASSERTF(tot_pending <= tot_granted, "%s "LPU64" > "LPU64"\n",
1459                  func, tot_pending, tot_granted);
1460         LASSERTF(tot_granted <= maxsize, "%s "LPU64" > "LPU64"\n",
1461                  func, tot_granted, maxsize);
1462         LASSERTF(tot_dirty <= maxsize, "%s "LPU64" > "LPU64"\n",
1463                  func, tot_dirty, maxsize);
1464 }
1465
1466 /* Remove this client from the grant accounting totals.  This is done at
1467  * disconnect time and also at export destroy time in case there was a race
1468  * between removing the export and an incoming BRW updating the client grant.
1469  * The client should do something similar when it invalidates its import. */
1470 static void filter_grant_discard(struct obd_export *exp)
1471 {
1472         struct obd_device *obd = exp->exp_obd;
1473         struct filter_obd *filter = &obd->u.filter;
1474         struct filter_export_data *fed = &exp->exp_filter_data;
1475
1476         spin_lock(&obd->obd_osfs_lock);
1477         CDEBUG(D_CACHE, "%s: cli %s/%p dirty %lu pend %lu grant %lu\n",
1478                obd->obd_name, exp->exp_client_uuid.uuid, exp,
1479                fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1480
1481         LASSERTF(filter->fo_tot_granted >= fed->fed_grant,
1482                  "%s: tot_granted "LPU64" cli %s/%p fed_grant %lu\n",
1483                  obd->obd_name, filter->fo_tot_granted,
1484                  exp->exp_client_uuid.uuid, exp, fed->fed_grant);
1485         filter->fo_tot_granted -= fed->fed_grant;
1486         LASSERTF(exp->exp_obd->u.filter.fo_tot_pending >= fed->fed_pending,
1487                  "%s: tot_pending "LPU64" cli %s/%p fed_pending %lu\n",
1488                  obd->obd_name, filter->fo_tot_pending,
1489                  exp->exp_client_uuid.uuid, exp, fed->fed_pending);
1490         LASSERTF(filter->fo_tot_dirty >= fed->fed_dirty,
1491                  "%s: tot_dirty "LPU64" cli %s/%p fed_dirty %lu\n",
1492                  obd->obd_name, filter->fo_tot_dirty,
1493                  exp->exp_client_uuid.uuid, exp, fed->fed_dirty);
1494         filter->fo_tot_dirty -= fed->fed_dirty;
1495         fed->fed_dirty = 0;
1496         fed->fed_grant = 0;
1497
1498         spin_unlock(&obd->obd_osfs_lock);
1499 }
1500
1501 static int filter_destroy_export(struct obd_export *exp)
1502 {
1503         ENTRY;
1504
1505         if (exp->exp_filter_data.fed_pending)
1506                 CERROR("%s: cli %s/%p has %lu pending on destroyed export\n",
1507                        exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
1508                        exp, exp->exp_filter_data.fed_pending);
1509
1510         target_destroy_export(exp);
1511
1512         if (exp->exp_obd->obd_replayable)
1513                 filter_client_free(exp, exp->exp_flags);
1514
1515         filter_grant_sanity_check(exp->exp_obd, __FUNCTION__);
1516
1517         RETURN(0);
1518 }
1519
1520 /* also incredibly similar to mds_disconnect */
1521 static int filter_disconnect(struct obd_export *exp, int flags)
1522 {
1523         struct obd_device *obd = exp->exp_obd;
1524         unsigned long irqflags;
1525         struct llog_ctxt *ctxt;
1526         int rc;
1527         ENTRY;
1528
1529         LASSERT(exp);
1530         class_export_get(exp);
1531
1532         spin_lock_irqsave(&exp->exp_lock, irqflags);
1533         exp->exp_flags = flags;
1534         spin_unlock_irqrestore(&exp->exp_lock, irqflags);
1535
1536         filter_grant_discard(exp);
1537
1538         /* Disconnect early so that clients can't keep using export */
1539         rc = class_disconnect(exp, flags);
1540
1541         /* Do this twice in case a BRW arrived between the first call and
1542          * the class_export_unlink() call (bug 2663) */
1543         filter_grant_discard(exp);
1544
1545         ldlm_cancel_locks_for_export(exp);
1546
1547         fsfilt_sync(obd, obd->u.filter.fo_sb);
1548
1549         /* flush any remaining cancel messages out to the target */
1550         ctxt = llog_get_context(obd, LLOG_UNLINK_REPL_CTXT);
1551         llog_sync(ctxt, exp);
1552
1553         class_export_put(exp);
1554         RETURN(rc);
1555 }
1556
1557 struct dentry *__filter_oa2dentry(struct obd_device *obd,
1558                                   struct obdo *oa, const char *what)
1559 {
1560         struct dentry *dchild = NULL;
1561         obd_gr group = 0;
1562
1563         if (oa->o_valid & OBD_MD_FLGROUP)
1564                 group = oa->o_gr;
1565
1566         dchild = filter_fid2dentry(obd, NULL, group, oa->o_id);
1567
1568         if (IS_ERR(dchild)) {
1569                 CERROR("%s error looking up object: "LPU64"\n", what, oa->o_id);
1570                 RETURN(dchild);
1571         }
1572
1573         if (dchild->d_inode == NULL) {
1574                 CERROR("%s on non-existent object: "LPU64"\n", what, oa->o_id);
1575                 f_dput(dchild);
1576                 RETURN(ERR_PTR(-ENOENT));
1577         }
1578
1579         return dchild;
1580 }
1581
1582 static int filter_getattr(struct obd_export *exp, struct obdo *oa,
1583                           struct lov_stripe_md *md)
1584 {
1585         struct dentry *dentry = NULL;
1586         struct obd_device *obd;
1587         int rc = 0;
1588         ENTRY;
1589
1590         obd = class_exp2obd(exp);
1591         if (obd == NULL) {
1592                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1593                        exp->exp_handle.h_cookie);
1594                 RETURN(-EINVAL);
1595         }
1596
1597         dentry = filter_oa2dentry(obd, oa);
1598         if (IS_ERR(dentry))
1599                 RETURN(PTR_ERR(dentry));
1600
1601         /* Limit the valid bits in the return data to what we actually use */
1602         oa->o_valid = OBD_MD_FLID;
1603         obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
1604
1605         f_dput(dentry);
1606         RETURN(rc);
1607 }
1608
1609 /* this is called from filter_truncate() until we have filter_punch() */
1610 static int filter_setattr(struct obd_export *exp, struct obdo *oa,
1611                           struct lov_stripe_md *md, struct obd_trans_info *oti)
1612 {
1613         struct obd_run_ctxt saved;
1614         struct filter_obd *filter;
1615         struct dentry *dentry;
1616         struct iattr iattr;
1617         struct ldlm_res_id res_id = { .name = { oa->o_id } };
1618         struct ldlm_resource *res;
1619         void *handle;
1620         int rc, rc2;
1621         ENTRY;
1622
1623         LASSERT(oti != NULL);
1624
1625         dentry = filter_oa2dentry(exp->exp_obd, oa);
1626         if (IS_ERR(dentry))
1627                 RETURN(PTR_ERR(dentry));
1628
1629         filter = &exp->exp_obd->u.filter;
1630
1631         iattr_from_obdo(&iattr, oa, oa->o_valid);
1632
1633         push_ctxt(&saved, &exp->exp_obd->obd_ctxt, NULL);
1634         lock_kernel();
1635
1636         if (iattr.ia_valid & ATTR_SIZE)
1637                 down(&dentry->d_inode->i_sem);
1638         handle = fsfilt_start(exp->exp_obd, dentry->d_inode, FSFILT_OP_SETATTR,
1639                               oti);
1640         if (IS_ERR(handle))
1641                 GOTO(out_unlock, rc = PTR_ERR(handle));
1642
1643         /* XXX this could be a rwsem instead, if filter_preprw played along */
1644         if (iattr.ia_valid & ATTR_ATTR_FLAG)
1645                 rc = fsfilt_iocontrol(exp->exp_obd, dentry->d_inode, NULL,
1646                                       EXT3_IOC_SETFLAGS,
1647                                       (long)&iattr.ia_attr_flags);
1648         else
1649                 rc = fsfilt_setattr(exp->exp_obd, dentry, handle, &iattr, 1);
1650         rc = filter_finish_transno(exp, oti, rc);
1651         rc2 = fsfilt_commit(exp->exp_obd, dentry->d_inode, handle, 0);
1652         if (rc2) {
1653                 CERROR("error on commit, err = %d\n", rc2);
1654                 if (!rc)
1655                         rc = rc2;
1656         }
1657
1658         if (iattr.ia_valid & ATTR_SIZE) {
1659                 res = ldlm_resource_get(exp->exp_obd->obd_namespace, NULL,
1660                                         res_id, LDLM_EXTENT, 0);
1661                 if (res == NULL) {
1662                         CERROR("!!! resource_get failed for object "LPU64" -- "
1663                                "filter_setattr with no lock?\n", oa->o_id);
1664                 } else {
1665                         if (res->lr_namespace->ns_lvbo &&
1666                             res->lr_namespace->ns_lvbo->lvbo_update) {
1667                                 rc = res->lr_namespace->ns_lvbo->lvbo_update
1668                                         (res, NULL, 0);
1669                         }
1670                         ldlm_resource_putref(res);
1671                 }
1672         }
1673
1674         oa->o_valid = OBD_MD_FLID;
1675         obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
1676
1677 out_unlock:
1678         if (iattr.ia_valid & ATTR_SIZE)
1679                 up(&dentry->d_inode->i_sem);
1680         unlock_kernel();
1681         pop_ctxt(&saved, &exp->exp_obd->obd_ctxt, NULL);
1682
1683         f_dput(dentry);
1684         RETURN(rc);
1685 }
1686
1687 /* XXX identical to osc_unpackmd */
1688 static int filter_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
1689                            struct lov_mds_md *lmm, int lmm_bytes)
1690 {
1691         int lsm_size;
1692         ENTRY;
1693
1694         if (lmm != NULL) {
1695                 if (lmm_bytes < sizeof (*lmm)) {
1696                         CERROR("lov_mds_md too small: %d, need %d\n",
1697                                lmm_bytes, (int)sizeof(*lmm));
1698                         RETURN(-EINVAL);
1699                 }
1700                 /* XXX LOV_MAGIC etc check? */
1701
1702                 if (lmm->lmm_object_id == cpu_to_le64(0)) {
1703                         CERROR("lov_mds_md: zero lmm_object_id\n");
1704                         RETURN(-EINVAL);
1705                 }
1706         }
1707
1708         lsm_size = lov_stripe_md_size(1);
1709         if (lsmp == NULL)
1710                 RETURN(lsm_size);
1711
1712         if (*lsmp != NULL && lmm == NULL) {
1713                 OBD_FREE(*lsmp, lsm_size);
1714                 *lsmp = NULL;
1715                 RETURN(0);
1716         }
1717
1718         if (*lsmp == NULL) {
1719                 OBD_ALLOC(*lsmp, lsm_size);
1720                 if (*lsmp == NULL)
1721                         RETURN(-ENOMEM);
1722
1723                 loi_init((*lsmp)->lsm_oinfo);
1724         }
1725
1726         if (lmm != NULL) {
1727                 /* XXX zero *lsmp? */
1728                 (*lsmp)->lsm_object_id = le64_to_cpu (lmm->lmm_object_id);
1729                 LASSERT((*lsmp)->lsm_object_id);
1730         }
1731
1732         (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
1733
1734         RETURN(lsm_size);
1735 }
1736
1737 static void filter_destroy_precreated(struct obd_export *exp, struct obdo *oa,
1738                                       struct filter_obd *filter)
1739 {
1740         struct obdo doa; /* XXX obdo on stack */
1741         __u64 last, id;
1742         ENTRY;
1743         LASSERT(oa);
1744
1745         memset(&doa, 0, sizeof(doa));
1746         if (oa->o_valid & OBD_MD_FLGROUP)
1747                 doa.o_gr = oa->o_gr;
1748         else
1749                 doa.o_gr = 0;
1750         doa.o_mode = S_IFREG;
1751         last = filter_last_id(filter, &doa); /* FIXME: object groups */
1752         CWARN("deleting orphan objects from "LPU64" to "LPU64"\n",
1753                oa->o_id + 1, last);
1754         for (id = oa->o_id + 1; id <= last; id++) {
1755                 doa.o_id = id;
1756                 filter_destroy(exp, &doa, NULL, NULL);
1757         }
1758         spin_lock(&filter->fo_objidlock);
1759         filter->fo_last_objids[0] = oa->o_id; /* FIXME: object groups */
1760         spin_unlock(&filter->fo_objidlock);
1761         EXIT;
1762 }
1763
1764 /* returns a negative error or a nonnegative number of files to create */
1765 static int filter_should_precreate(struct obd_export *exp, struct obdo *oa,
1766                                    int group)
1767 {
1768         struct obd_device *obd = exp->exp_obd;
1769         struct filter_obd *filter = &obd->u.filter;
1770         int diff, rc;
1771         ENTRY;
1772
1773         diff = oa->o_id - filter_last_id(filter, oa);
1774         CDEBUG(D_INFO, "filter_last_id() = "LPU64" -> diff = %d\n",
1775                filter_last_id(filter, oa), diff);
1776
1777         /* delete orphans request */
1778         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1779             (oa->o_flags & OBD_FL_DELORPHAN)) {
1780                 if (diff >= 0)
1781                         RETURN(diff);
1782                 if (-diff > 10000) { /* XXX make this smarter */
1783                         CERROR("ignoring bogus orphan destroy request: obdid "
1784                                LPU64" last_id "LPU64"\n",
1785                                oa->o_id, filter_last_id(filter, oa));
1786                         RETURN(-EINVAL);
1787                 }
1788                 filter_destroy_precreated(exp, oa, filter);
1789                 rc = filter_update_last_objid(obd, group, 0);
1790                 if (rc)
1791                         CERROR("unable to write lastobjid, but orphans"
1792                                "were deleted\n");
1793                 RETURN(0);
1794         } else {
1795                 /* only precreate if group == 0 and o_id is specfied */
1796                 if (!(oa->o_valid & OBD_FL_DELORPHAN) &&
1797                     (group != 0 || oa->o_id == 0))
1798                         RETURN(1);
1799
1800                 LASSERT(diff >= 0);
1801                 RETURN(diff);
1802         }
1803 }
1804
1805 /* We rely on the fact that only one thread will be creating files in a given
1806  * group at a time, which is why we don't need an atomic filter_get_new_id.
1807  * Even if we had that atomic function, the following race would exist:
1808  *
1809  * thread 1: gets id x from filter_next_id
1810  * thread 2: gets id (x + 1) from filter_next_id
1811  * thread 2: creates object (x + 1)
1812  * thread 1: tries to create object x, gets -ENOSPC
1813  */
1814 static int filter_precreate(struct obd_device *obd, struct obdo *oa,
1815                             obd_gr group, int *num)
1816 {
1817         struct dentry *dchild = NULL;
1818         struct filter_obd *filter;
1819         struct dentry *dparent;
1820         int err = 0, rc = 0, i;
1821         __u64 next_id;
1822         int recreate_obj = 0;
1823         void *handle = NULL;
1824         ENTRY;
1825
1826         filter = &obd->u.filter;
1827
1828         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1829             (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1830                 recreate_obj = 1;
1831         }
1832
1833         for (i = 0; i < *num && err == 0; i++) {
1834                 int cleanup_phase = 0;
1835
1836                 if (recreate_obj) {
1837                         __u64 last_id;
1838                         next_id = oa->o_id;
1839                         last_id = filter_last_id(filter, NULL);
1840                         if (next_id > last_id) {
1841                                 CERROR("Error: Trying to recreate obj greater"
1842                                        "than last id "LPD64" > "LPD64"\n",
1843                                        next_id, last_id);
1844                                 RETURN(-EINVAL);
1845                         }
1846                 } else
1847                         next_id = filter_last_id(filter, NULL) + 1;
1848
1849                 CDEBUG(D_INFO, "precreate objid "LPU64"\n", next_id);
1850
1851                 dparent = filter_parent_lock(obd, group, next_id);
1852                 if (IS_ERR(dparent))
1853                         GOTO(cleanup, rc = PTR_ERR(dparent));
1854                 cleanup_phase = 1;
1855
1856                 dchild = filter_fid2dentry(obd, dparent, group, next_id);
1857                 if (IS_ERR(dchild))
1858                         GOTO(cleanup, rc = PTR_ERR(dchild));
1859                 cleanup_phase = 2;
1860
1861                 if (dchild->d_inode != NULL) {
1862                         /* This would only happen if lastobjid was bad on disk*/
1863                         /* Could also happen if recreating missing obj but
1864                          * already exists
1865                          */
1866                         if (recreate_obj) {
1867                                 CERROR("Serious error: recreating obj %*s but "
1868                                        "obj already exists \n",
1869                                        dchild->d_name.len, dchild->d_name.name);
1870                         } else {
1871                                 CERROR("Serious error: objid %*s already "
1872                                        "exists; is this filesystem corrupt?\n",
1873                                         dchild->d_name.len, dchild->d_name.name);
1874                         }
1875                         GOTO(cleanup, rc = -EEXIST);
1876                 }
1877
1878                 handle = fsfilt_start(obd, dparent->d_inode,
1879                                       FSFILT_OP_CREATE_LOG, NULL);
1880                 if (IS_ERR(handle))
1881                         GOTO(cleanup, rc = PTR_ERR(handle));
1882                 cleanup_phase = 3;
1883
1884                 rc = ll_vfs_create(dparent->d_inode, dchild, S_IFREG, NULL);
1885                 if (rc) {
1886                         CERROR("create failed rc = %d\n", rc);
1887                         GOTO(cleanup, rc);
1888                 }
1889
1890                 if (!recreate_obj) {
1891                         filter_set_last_id(filter, NULL, next_id);
1892                         err = filter_update_last_objid(obd, group, 0);
1893                         if (err)
1894                                 CERROR("unable to write lastobjid "
1895                                        "but file created\n");
1896                 }
1897
1898         cleanup:
1899                 switch(cleanup_phase) {
1900                 case 3:
1901                         err = fsfilt_commit(obd, dparent->d_inode, handle, 0);
1902                         if (err) {
1903                                 CERROR("error on commit, err = %d\n", err);
1904                                 if (!rc)
1905                                         rc = err;
1906                         }
1907                 case 2:
1908                         f_dput(dchild);
1909                 case 1:
1910                         filter_parent_unlock(dparent);
1911                 case 0:
1912                         break;
1913                 }
1914
1915                 if (rc)
1916                         break;
1917         }
1918         *num = i;
1919
1920         CDEBUG(D_INFO, "filter_precreate() created %d objects\n", i);
1921         RETURN(rc);
1922 }
1923
1924 static int filter_create(struct obd_export *exp, struct obdo *oa,
1925                          struct lov_stripe_md **ea, struct obd_trans_info *oti)
1926 {
1927         struct obd_device *obd = NULL;
1928         struct obd_run_ctxt saved;
1929         struct lov_stripe_md *lsm = NULL;
1930         obd_gr group = 0;
1931         int rc = 0, diff;
1932         ENTRY;
1933
1934         if (oa->o_valid & OBD_MD_FLGROUP)
1935                 group = oa->o_gr;
1936
1937         CDEBUG(D_INFO, "filter_create(od->o_gr="LPU64",od->o_id="LPU64")\n",
1938                group, oa->o_id);
1939         if (ea != NULL) {
1940                 lsm = *ea;
1941                 if (lsm == NULL) {
1942                         rc = obd_alloc_memmd(exp, &lsm);
1943                         if (rc < 0)
1944                                 RETURN(rc);
1945                 }
1946         }
1947
1948         obd = exp->exp_obd;
1949         push_ctxt(&saved, &obd->obd_ctxt, NULL);
1950
1951         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1952             (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1953                 if (oa->o_id > filter_last_id(&obd->u.filter, oa)) {
1954                         CERROR("recreate objid "LPU64" > last id "LPU64"\n",
1955                                oa->o_id, filter_last_id(&obd->u.filter, oa));
1956                         rc = -EINVAL;
1957                 } else {
1958                         diff = 1;
1959                         rc = filter_precreate(obd, oa, group, &diff);
1960                 }
1961         } else {
1962                 diff = filter_should_precreate(exp, oa, group);
1963                 if (diff > 0) {
1964                         oa->o_id = filter_last_id(&obd->u.filter, oa);
1965                         rc = filter_precreate(obd, oa, group, &diff);
1966                         oa->o_id += diff;
1967                         oa->o_valid = OBD_MD_FLID;
1968                 }
1969         }
1970
1971         pop_ctxt(&saved, &obd->obd_ctxt, NULL);
1972         if (rc && ea != NULL && *ea != lsm) {
1973                 obd_free_memmd(exp, &lsm);
1974         } else if (rc == 0 && ea != NULL) {
1975                 /* XXX LOV STACKING: the lsm that is passed to us from
1976                  * LOV does not have valid lsm_oinfo data structs, so
1977                  * don't go touching that.  This needs to be fixed in a
1978                  * big way. */
1979                 lsm->lsm_object_id = oa->o_id;
1980                 *ea = lsm;
1981         }
1982
1983         RETURN(rc);
1984 }
1985
1986 static int filter_destroy(struct obd_export *exp, struct obdo *oa,
1987                           struct lov_stripe_md *ea, struct obd_trans_info *oti)
1988 {
1989         struct obd_device *obd;
1990         struct filter_obd *filter;
1991         struct dentry *dchild = NULL, *dparent = NULL;
1992         struct obd_run_ctxt saved;
1993         void *handle = NULL;
1994         struct llog_cookie *fcc = NULL;
1995         int rc, rc2, cleanup_phase = 0, have_prepared = 0;
1996         obd_gr group = 0;
1997         ENTRY;
1998
1999         if (oa->o_valid & OBD_MD_FLGROUP)
2000                 group = oa->o_gr;
2001
2002         obd = exp->exp_obd;
2003         filter = &obd->u.filter;
2004
2005         push_ctxt(&saved, &obd->obd_ctxt, NULL);
2006
2007  acquire_locks:
2008         dparent = filter_parent_lock(obd, group, oa->o_id);
2009         if (IS_ERR(dparent))
2010                 GOTO(cleanup, rc = PTR_ERR(dparent));
2011         cleanup_phase = 1;
2012
2013         dchild = filter_fid2dentry(obd, dparent, group, oa->o_id);
2014         if (IS_ERR(dchild))
2015                 GOTO(cleanup, rc = -ENOENT);
2016         cleanup_phase = 2;
2017
2018         if (dchild->d_inode == NULL) {
2019                 CDEBUG(D_INODE, "destroying non-existent object "LPU64"\n",
2020                        oa->o_id);
2021                 GOTO(cleanup, rc = -ENOENT);
2022         }
2023
2024         if (!have_prepared) {
2025                 /* If we're really going to destroy the object, get ready
2026                  * by getting the clients to discard their cached data.
2027                  *
2028                  * We have to drop the parent lock, because
2029                  * filter_prepare_destroy will acquire a PW on the object, and
2030                  * we don't want to deadlock with an incoming write to the
2031                  * object, which has the extent PW and then wants to get the
2032                  * parent dentry to do the lookup.
2033                  *
2034                  * We dput the child because it's not worth the extra
2035                  * complication of condition the above code to skip it on the
2036                  * second time through. */
2037                 f_dput(dchild);
2038                 filter_parent_unlock(dparent);
2039
2040                 filter_prepare_destroy(obd, oa->o_id);
2041                 have_prepared = 1;
2042                 goto acquire_locks;
2043         }
2044
2045         handle = fsfilt_start(obd, dparent->d_inode, FSFILT_OP_UNLINK_LOG, oti);
2046         if (IS_ERR(handle))
2047                 GOTO(cleanup, rc = PTR_ERR(handle));
2048         cleanup_phase = 3;
2049
2050         /* Our MDC connection is established by the MDS to us */
2051         if (oa->o_valid & OBD_MD_FLCOOKIE) {
2052                 OBD_ALLOC(fcc, sizeof(*fcc));
2053                 if (fcc != NULL)
2054                         memcpy(fcc, obdo_logcookie(oa), sizeof(*fcc));
2055         }
2056
2057         rc = filter_destroy_internal(obd, oa->o_id, dparent, dchild);
2058
2059 cleanup:
2060         switch(cleanup_phase) {
2061         case 3:
2062                 if (fcc != NULL) {
2063                         if (oti != NULL)
2064                                 fsfilt_add_journal_cb(obd, 0, oti->oti_handle,
2065                                                       filter_cancel_cookies_cb,
2066                                                       fcc);
2067                         else
2068                                 fsfilt_add_journal_cb(obd, 0, handle,
2069                                                       filter_cancel_cookies_cb,
2070                                                       fcc);
2071                 }
2072                 rc = filter_finish_transno(exp, oti, rc);
2073                 rc2 = fsfilt_commit(obd, dparent->d_inode, handle, 0);
2074                 if (rc2) {
2075                         CERROR("error on commit, err = %d\n", rc2);
2076                         if (!rc)
2077                                 rc = rc2;
2078                 }
2079         case 2:
2080                 f_dput(dchild);
2081         case 1:
2082                 filter_parent_unlock(dparent);
2083         case 0:
2084                 pop_ctxt(&saved, &obd->obd_ctxt, NULL);
2085                 break;
2086         default:
2087                 CERROR("invalid cleanup_phase %d\n", cleanup_phase);
2088                 LBUG();
2089         }
2090
2091         RETURN(rc);
2092 }
2093
2094 /* NB start and end are used for punch, but not truncate */
2095 static int filter_truncate(struct obd_export *exp, struct obdo *oa,
2096                            struct lov_stripe_md *lsm,
2097                            obd_off start, obd_off end,
2098                            struct obd_trans_info *oti)
2099 {
2100         int error;
2101         ENTRY;
2102
2103         if (end != OBD_OBJECT_EOF)
2104                 CERROR("PUNCH not supported, only truncate: end = "LPX64"\n",
2105                        end);
2106
2107         CDEBUG(D_INODE, "calling truncate for object "LPU64", valid = %x, "
2108                "o_size = "LPD64"\n", oa->o_id, oa->o_valid, start);
2109         oa->o_size = start;
2110         error = filter_setattr(exp, oa, NULL, oti);
2111         RETURN(error);
2112 }
2113
2114 static int filter_sync(struct obd_export *exp, struct obdo *oa,
2115                        struct lov_stripe_md *lsm, obd_off start, obd_off end)
2116 {
2117         struct obd_run_ctxt saved;
2118         struct filter_obd *filter;
2119         struct dentry *dentry;
2120         struct llog_ctxt *ctxt;
2121         int rc, rc2;
2122         ENTRY;
2123
2124         filter = &exp->exp_obd->u.filter;
2125
2126         /* an objid of zero is taken to mean "sync whole filesystem" */
2127         if (!oa || !(oa->o_valid & OBD_MD_FLID)) {
2128                 rc = fsfilt_sync(exp->exp_obd, filter->fo_sb);
2129                 /* flush any remaining cancel messages out to the target */
2130                 ctxt = llog_get_context(exp->exp_obd, LLOG_UNLINK_REPL_CTXT);
2131                 llog_sync(ctxt, exp);
2132                 RETURN(rc);
2133         }
2134
2135         dentry = filter_oa2dentry(exp->exp_obd, oa);
2136         if (IS_ERR(dentry))
2137                 RETURN(PTR_ERR(dentry));
2138
2139         push_ctxt(&saved, &exp->exp_obd->obd_ctxt, NULL);
2140
2141         down(&dentry->d_inode->i_sem);
2142         rc = filemap_fdatasync(dentry->d_inode->i_mapping);
2143         if (rc == 0) {
2144                 /* just any file to grab fsync method - "file" arg unused */
2145                 struct file *file = filter->fo_rcvd_filp;
2146
2147                 if (file->f_op && file->f_op->fsync)
2148                         rc = file->f_op->fsync(NULL, dentry, 1);
2149
2150                 rc2 = filemap_fdatawait(dentry->d_inode->i_mapping);
2151                 if (!rc)
2152                         rc = rc2;
2153         }
2154         up(&dentry->d_inode->i_sem);
2155
2156         oa->o_valid = OBD_MD_FLID;
2157         obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2158
2159         pop_ctxt(&saved, &exp->exp_obd->obd_ctxt, NULL);
2160
2161         f_dput(dentry);
2162         RETURN(rc);
2163 }
2164
2165 static int filter_statfs(struct obd_device *obd, struct obd_statfs *osfs,
2166                          unsigned long max_age)
2167 {
2168         struct filter_obd *filter = &obd->u.filter;
2169         int blockbits = filter->fo_sb->s_blocksize_bits;
2170         int rc;
2171         ENTRY;
2172
2173         /* at least try to account for cached pages.  its still racey and
2174          * might be under-reporting if clients haven't announced their
2175          * caches with brw recently */
2176         spin_lock(&obd->obd_osfs_lock);
2177         rc = fsfilt_statfs(obd, filter->fo_sb, max_age);
2178         memcpy(osfs, &obd->obd_osfs, sizeof(*osfs));
2179         spin_unlock(&obd->obd_osfs_lock);
2180
2181         CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
2182                " pending "LPU64" free "LPU64" avail "LPU64"\n",
2183                filter->fo_tot_dirty, filter->fo_tot_granted,
2184                filter->fo_tot_pending,
2185                osfs->os_bfree << blockbits, osfs->os_bavail << blockbits);
2186
2187         filter_grant_sanity_check(obd, __FUNCTION__);
2188
2189         osfs->os_bavail -= min(osfs->os_bavail,
2190                                (filter->fo_tot_dirty + filter->fo_tot_pending +
2191                                 osfs->os_bsize -1) >> blockbits);
2192
2193         RETURN(rc);
2194 }
2195
2196 static int filter_get_info(struct obd_export *exp, __u32 keylen,
2197                            void *key, __u32 *vallen, void *val)
2198 {
2199         struct obd_device *obd;
2200         ENTRY;
2201
2202         obd = class_exp2obd(exp);
2203         if (obd == NULL) {
2204                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2205                        exp->exp_handle.h_cookie);
2206                 RETURN(-EINVAL);
2207         }
2208
2209         if (keylen == strlen("blocksize") &&
2210             memcmp(key, "blocksize", keylen) == 0) {
2211                 __u32 *blocksize = val;
2212                 *vallen = sizeof(*blocksize);
2213                 *blocksize = obd->u.filter.fo_sb->s_blocksize;
2214                 RETURN(0);
2215         }
2216
2217         if (keylen == strlen("blocksize_bits") &&
2218             memcmp(key, "blocksize_bits", keylen) == 0) {
2219                 __u32 *blocksize_bits = val;
2220                 *vallen = sizeof(*blocksize_bits);
2221                 *blocksize_bits = obd->u.filter.fo_sb->s_blocksize_bits;
2222                 RETURN(0);
2223         }
2224
2225         if (keylen >= strlen("last_id") && memcmp(key, "last_id", 7) == 0) {
2226                 obd_id *last_id = val;
2227                 /* FIXME: object groups */
2228                 *last_id = filter_last_id(&obd->u.filter, 0);
2229                 RETURN(0);
2230         }
2231         CDEBUG(D_IOCTL, "invalid key\n");
2232         RETURN(-EINVAL);
2233 }
2234
2235 static int filter_set_info(struct obd_export *exp, __u32 keylen,
2236                            void *key, __u32 vallen, void *val)
2237 {
2238         struct obd_device *obd;
2239         struct lustre_handle conn;
2240         struct llog_ctxt *ctxt;
2241         int rc = 0;
2242         ENTRY;
2243
2244         conn.cookie = exp->exp_handle.h_cookie;
2245
2246         obd = exp->exp_obd;
2247         if (obd == NULL) {
2248                 CDEBUG(D_IOCTL, "invalid exp %p cookie "LPX64"\n",
2249                        exp, conn.cookie);
2250                 RETURN(-EINVAL);
2251         }
2252
2253         if (keylen < strlen("mds_conn") ||
2254             memcmp(key, "mds_conn", keylen) != 0)
2255                 RETURN(-EINVAL);
2256
2257         CWARN("Received MDS connection ("LPX64")\n", conn.cookie);
2258         memcpy(&obd->u.filter.fo_mdc_conn, &conn, sizeof(conn));
2259         ctxt = llog_get_context(obd, LLOG_UNLINK_REPL_CTXT);
2260         rc = llog_receptor_accept(ctxt, exp->exp_imp_reverse);
2261         RETURN(rc);
2262 }
2263
2264 int filter_iocontrol(unsigned int cmd, struct obd_export *exp,
2265                      int len, void *karg, void *uarg)
2266 {
2267         struct obd_device *obd = exp->exp_obd;
2268         struct obd_ioctl_data *data = karg;
2269         int rc = 0;
2270
2271         switch (cmd) {
2272         case OBD_IOC_ABORT_RECOVERY:
2273                 CERROR("aborting recovery for device %s\n", obd->obd_name);
2274                 target_abort_recovery(obd);
2275                 RETURN(0);
2276
2277         case OBD_IOC_SET_READONLY: {
2278                 void *handle;
2279                 struct super_block *sb = obd->u.filter.fo_sb;
2280                 struct inode *inode = sb->s_root->d_inode;
2281                 BDEVNAME_DECLARE_STORAGE(tmp);
2282                 CERROR("setting device %s read-only\n",
2283                        ll_bdevname(sb, tmp));
2284
2285                 handle = fsfilt_start(obd, inode, FSFILT_OP_MKNOD, NULL);
2286                 LASSERT(handle);
2287                 (void)fsfilt_commit(obd, inode, handle, 1);
2288
2289                 dev_set_rdonly(ll_sbdev(obd->u.filter.fo_sb), 2);
2290                 RETURN(0);
2291         }
2292
2293         case OBD_IOC_CATLOGLIST: {
2294                 rc = llog_catlog_list(obd, 1, data);
2295                 RETURN(rc);
2296         }
2297
2298         case OBD_IOC_LLOG_CANCEL:
2299         case OBD_IOC_LLOG_REMOVE:
2300         case OBD_IOC_LLOG_INFO:
2301         case OBD_IOC_LLOG_PRINT: {
2302                 /* FIXME to be finished */
2303                 RETURN(-EOPNOTSUPP);
2304 /*
2305                 struct llog_ctxt *ctxt = NULL;
2306
2307                 push_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_ctxt, NULL);
2308                 rc = llog_ioctl(ctxt, cmd, data);
2309                 pop_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_ctxt, NULL);
2310
2311                 RETURN(rc);
2312 */
2313         }
2314
2315
2316         default:
2317                 RETURN(-EINVAL);
2318         }
2319         RETURN(0);
2320 }
2321
2322 static struct llog_operations filter_unlink_repl_logops;
2323 static struct llog_operations filter_size_orig_logops = {
2324         lop_setup: llog_obd_origin_setup,
2325         lop_cleanup: llog_obd_origin_cleanup,
2326         lop_add: llog_obd_origin_add
2327 };
2328
2329 static int filter_llog_init(struct obd_device *obd, struct obd_device *tgt,
2330                             int count, struct llog_catid *logid)
2331 {
2332         struct llog_ctxt *ctxt;
2333         int rc;
2334         ENTRY;
2335
2336         filter_unlink_repl_logops = llog_client_ops;
2337         filter_unlink_repl_logops.lop_cancel = llog_obd_repl_cancel;
2338         filter_unlink_repl_logops.lop_connect = llog_repl_connect;
2339         filter_unlink_repl_logops.lop_sync = llog_obd_repl_sync;
2340
2341         rc = llog_setup(obd, LLOG_UNLINK_REPL_CTXT, tgt, 0, NULL,
2342                         &filter_unlink_repl_logops);
2343         if (rc)
2344                 RETURN(rc);
2345         /* FIXME - assign unlink_cb for filter's recovery */
2346         ctxt = llog_get_context(obd, LLOG_UNLINK_REPL_CTXT);
2347         ctxt->llog_proc_cb = filter_recov_log_unlink_cb;
2348
2349         rc = llog_setup(obd, LLOG_SIZE_ORIG_CTXT, tgt, 0, NULL,
2350                         &filter_size_orig_logops);
2351         RETURN(rc);
2352 }
2353
2354 static int filter_llog_finish(struct obd_device *obd, int count)
2355 {
2356         int rc;
2357         ENTRY;
2358
2359         rc = llog_cleanup(llog_get_context(obd, LLOG_UNLINK_REPL_CTXT));
2360         if (rc)
2361                 RETURN(rc);
2362
2363         rc = llog_cleanup(llog_get_context(obd, LLOG_SIZE_ORIG_CTXT));
2364         RETURN(rc);
2365 }
2366
2367 static struct dentry *filter_lvfs_fid2dentry(__u64 id, __u32 gen, __u64 gr,
2368                                              void *data)
2369 {
2370         return filter_fid2dentry(data, NULL, gr, id);
2371 }
2372
2373 static struct lvfs_callback_ops filter_lvfs_ops = {
2374         l_fid2dentry:     filter_lvfs_fid2dentry,
2375 };
2376
2377 static struct obd_ops filter_obd_ops = {
2378         o_owner:          THIS_MODULE,
2379         o_attach:         filter_attach,
2380         o_detach:         filter_detach,
2381         o_get_info:       filter_get_info,
2382         o_set_info:       filter_set_info,
2383         o_setup:          filter_setup,
2384         o_precleanup:     filter_precleanup,
2385         o_cleanup:        filter_cleanup,
2386         o_connect:        filter_connect,
2387         o_disconnect:     filter_disconnect,
2388         o_statfs:         filter_statfs,
2389         o_getattr:        filter_getattr,
2390         o_unpackmd:       filter_unpackmd,
2391         o_create:         filter_create,
2392         o_setattr:        filter_setattr,
2393         o_destroy:        filter_destroy,
2394         o_brw:            filter_brw,
2395         o_punch:          filter_truncate,
2396         o_sync:           filter_sync,
2397         o_preprw:         filter_preprw,
2398         o_commitrw:       filter_commitrw,
2399         o_destroy_export: filter_destroy_export,
2400         o_llog_init:      filter_llog_init,
2401         o_llog_finish:    filter_llog_finish,
2402         o_iocontrol:      filter_iocontrol,
2403 };
2404
2405 static struct obd_ops filter_sanobd_ops = {
2406         o_owner:          THIS_MODULE,
2407         o_attach:         filter_attach,
2408         o_detach:         filter_detach,
2409         o_get_info:       filter_get_info,
2410         o_set_info:       filter_set_info,
2411         o_setup:          filter_san_setup,
2412         o_precleanup:     filter_precleanup,
2413         o_cleanup:        filter_cleanup,
2414         o_connect:        filter_connect,
2415         o_disconnect:     filter_disconnect,
2416         o_statfs:         filter_statfs,
2417         o_getattr:        filter_getattr,
2418         o_unpackmd:       filter_unpackmd,
2419         o_create:         filter_create,
2420         o_setattr:        filter_setattr,
2421         o_destroy:        filter_destroy,
2422         o_brw:            filter_brw,
2423         o_punch:          filter_truncate,
2424         o_sync:           filter_sync,
2425         o_preprw:         filter_preprw,
2426         o_commitrw:       filter_commitrw,
2427         o_san_preprw:     filter_san_preprw,
2428         o_destroy_export: filter_destroy_export,
2429         o_llog_init:      filter_llog_init,
2430         o_llog_finish:    filter_llog_finish,
2431         o_iocontrol:      filter_iocontrol,
2432 };
2433
2434 static int __init obdfilter_init(void)
2435 {
2436         struct lprocfs_static_vars lvars;
2437         int rc;
2438
2439         printk(KERN_INFO "Lustre: Filtering OBD driver; info@clusterfs.com\n");
2440
2441         lprocfs_init_vars(filter, &lvars);
2442
2443         rc = class_register_type(&filter_obd_ops, lvars.module_vars,
2444                                  OBD_FILTER_DEVICENAME);
2445         if (rc)
2446                 return rc;
2447
2448         rc = class_register_type(&filter_sanobd_ops, lvars.module_vars,
2449                                  OBD_FILTER_SAN_DEVICENAME);
2450         if (rc)
2451                 class_unregister_type(OBD_FILTER_DEVICENAME);
2452         return rc;
2453 }
2454
2455 static void __exit obdfilter_exit(void)
2456 {
2457         class_unregister_type(OBD_FILTER_SAN_DEVICENAME);
2458         class_unregister_type(OBD_FILTER_DEVICENAME);
2459 }
2460
2461 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2462 MODULE_DESCRIPTION("Lustre Filtering OBD driver");
2463 MODULE_LICENSE("GPL");
2464
2465 module_init(obdfilter_init);
2466 module_exit(obdfilter_exit);