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 #include <linux/jbd.h>
  45 #include <linux/ext3_fs.h>
  46 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
  47 # include <linux/mount.h>
  48 # include <linux/buffer_head.h>
  49 # include <linux/bio.h>
  50 #endif
  51
  52 #include <linux/obd_class.h>
  53 #include <linux/obd_lov.h>
  54 #include <linux/obd_ost.h>
  55 #include <linux/lustre_dlm.h>
  56 #include <linux/lustre_fsfilt.h>
  57 #include <linux/lprocfs_status.h>
  58 #include <linux/lustre_log.h>
  59 #include <linux/lustre_commit_confd.h>
  60 #include <portals/list.h>
  61
  62 #include <linux/lustre_smfs.h>
  63 #include "filter_internal.h"
  64
  65 /* Group 0 is no longer a legal group, to catch uninitialized IDs */
  66 #define FILTER_MIN_GROUPS 3
  67
  68 static struct lvfs_callback_ops filter_lvfs_ops;
  69
  70 static int filter_destroy(struct obd_export *exp, struct obdo *oa,
  71                           struct lov_stripe_md *ea, struct obd_trans_info *);
  72 struct obd_llogs *filter_grab_llog_for_group(struct obd_device *,
  73                                              int, struct obd_export *);
  74
  75 static void filter_commit_cb(struct obd_device *obd, __u64 transno,
  76                              void *cb_data, int error)
  77 {
  78         obd_transno_commit_cb(obd, transno, error);
  79 }
  80
  81 /* Assumes caller has already pushed us into the kernel context. */
  82 int filter_finish_transno(struct obd_export *exp, struct obd_trans_info *oti,
  83                           int rc)
  84 {
  85         struct filter_obd *filter = &exp->exp_obd->u.filter;
  86         struct filter_export_data *fed = &exp->exp_filter_data;
  87         struct filter_client_data *fcd = fed->fed_fcd;
  88         __u64 last_rcvd;
  89         loff_t off;
  90         int err, log_pri = D_HA;
  91
  92         /* Propagate error code. */
  93         if (rc)
  94                 RETURN(rc);
  95
  96         if (!exp->exp_obd->obd_replayable || oti == NULL)
  97                 RETURN(rc);
  98
  99         /* we don't allocate new transnos for replayed requests */
 100         if (oti->oti_transno == 0) {
 101                 spin_lock(&filter->fo_translock);
 102                 last_rcvd = le64_to_cpu(filter->fo_fsd->fsd_last_transno) + 1;
 103                 filter->fo_fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
 104                 spin_unlock(&filter->fo_translock);
 105                 oti->oti_transno = last_rcvd;
 106         } else {
 107                 spin_lock(&filter->fo_translock);
 108                 last_rcvd = oti->oti_transno;
 109                 if (last_rcvd > le64_to_cpu(filter->fo_fsd->fsd_last_transno))
 110                         filter->fo_fsd->fsd_last_transno =
 111                                 cpu_to_le64(last_rcvd);
 112                 spin_unlock(&filter->fo_translock);
 113         }
 114
 115         fcd->fcd_last_rcvd = cpu_to_le64(last_rcvd);
 116
 117         /* could get xid from oti, if it's ever needed */
 118         fcd->fcd_last_xid = 0;
 119
 120         off = fed->fed_lr_off;
 121
 122         fsfilt_add_journal_cb(exp->exp_obd, filter->fo_sb, last_rcvd,
 123                               oti->oti_handle, filter_commit_cb, NULL);
 124
 125         err = fsfilt_write_record(exp->exp_obd, filter->fo_rcvd_filp, fcd,
 126                                   sizeof(*fcd), &off, 0);
 127         if (err) {
 128                 log_pri = D_ERROR;
 129                 if (rc == 0)
 130                         rc = err;
 131         }
 132
 133         CDEBUG(log_pri, "wrote trans "LPU64" for client %s at #%d: err = %d\n",
 134                last_rcvd, fcd->fcd_uuid, fed->fed_lr_idx, err);
 135
 136         RETURN(rc);
 137 }
 138
 139 void f_dput(struct dentry *dentry)
 140 {
 141         /* Can't go inside filter_ddelete because it can block */
 142         CDEBUG(D_INODE, "putting %s: %p, count = %d\n",
 143                dentry->d_name.name, dentry, atomic_read(&dentry->d_count) - 1);
 144         LASSERT(atomic_read(&dentry->d_count) > 0);
 145
 146         dput(dentry);
 147 }
 148
 149 /* Add client data to the FILTER.  We use a bitmap to locate a free space
 150  * in the last_rcvd file if cl_idx is -1 (i.e. a new client).
 151  * Otherwise, we have just read the data from the last_rcvd file and
 152  * we know its offset. */
 153 static int filter_client_add(struct obd_device *obd, struct filter_obd *filter,
 154                              struct filter_export_data *fed, int cl_idx)
 155 {
 156         unsigned long *bitmap = filter->fo_last_rcvd_slots;
 157         int new_client = (cl_idx == -1);
 158         ENTRY;
 159
 160         LASSERT(bitmap != NULL);
 161
 162         /* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
 163         if (!strcmp(fed->fed_fcd->fcd_uuid, obd->obd_uuid.uuid))
 164                 RETURN(0);
 165
 166         /* the bitmap operations can handle cl_idx > sizeof(long) * 8, so
 167          * there's no need for extra complication here
 168          */
 169         if (new_client) {
 170                 cl_idx = find_first_zero_bit(bitmap, FILTER_LR_MAX_CLIENTS);
 171         repeat:
 172                 if (cl_idx >= FILTER_LR_MAX_CLIENTS) {
 173                         CERROR("no client slots - fix FILTER_LR_MAX_CLIENTS\n");
 174                         RETURN(-ENOMEM);
 175                 }
 176                 if (test_and_set_bit(cl_idx, bitmap)) {
 177                         CERROR("FILTER client %d: found bit is set in bitmap\n",
 178                                cl_idx);
 179                         cl_idx = find_next_zero_bit(bitmap,
 180                                                     FILTER_LR_MAX_CLIENTS,
 181                                                     cl_idx);
 182                         goto repeat;
 183                 }
 184         } else {
 185                 if (test_and_set_bit(cl_idx, bitmap)) {
 186                         CERROR("FILTER client %d: bit already set in bitmap!\n",
 187                                cl_idx);
 188                         LBUG();
 189                 }
 190         }
 191
 192         fed->fed_lr_idx = cl_idx;
 193         fed->fed_lr_off = le32_to_cpu(filter->fo_fsd->fsd_client_start) +
 194                 cl_idx * le16_to_cpu(filter->fo_fsd->fsd_client_size);
 195
 196         CDEBUG(D_INFO, "client at index %d (%llu) with UUID '%s' added\n",
 197                fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
 198
 199         if (new_client) {
 200                 struct lvfs_run_ctxt saved;
 201                 loff_t off = fed->fed_lr_off;
 202                 int err;
 203                 void *handle;
 204
 205                 CDEBUG(D_INFO, "writing client fcd at idx %u (%llu) (len %u)\n",
 206                        fed->fed_lr_idx,off,(unsigned int)sizeof(*fed->fed_fcd));
 207
 208                 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 209                 /* Transaction needed to fix bug 1403 */
 210                 handle = fsfilt_start(obd,
 211                                       filter->fo_rcvd_filp->f_dentry->d_inode,
 212                                       FSFILT_OP_SETATTR, NULL);
 213                 if (IS_ERR(handle)) {
 214                         err = PTR_ERR(handle);
 215                         CERROR("unable to start transaction: rc %d\n", err);
 216                 } else {
 217                         err = fsfilt_write_record(obd, filter->fo_rcvd_filp,
 218                                                   fed->fed_fcd,
 219                                                   sizeof(*fed->fed_fcd),
 220                                                   &off, 1);
 221                         fsfilt_commit(obd, filter->fo_sb,
 222                                       filter->fo_rcvd_filp->f_dentry->d_inode,
 223                                       handle, 1);
 224                 }
 225                 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 226
 227                 if (err) {
 228                         CERROR("error writing %s client idx %u: rc %d\n",
 229                                LAST_RCVD, fed->fed_lr_idx, err);
 230                         RETURN(err);
 231                 }
 232         }
 233         RETURN(0);
 234 }
 235
 236 static int filter_client_free(struct obd_export *exp, int flags)
 237 {
 238         struct filter_export_data *fed = &exp->exp_filter_data;
 239         struct filter_obd *filter = &exp->exp_obd->u.filter;
 240         struct obd_device *obd = exp->exp_obd;
 241         struct filter_client_data zero_fcd;
 242         struct lvfs_run_ctxt saved;
 243         int rc;
 244         loff_t off;
 245         ENTRY;
 246
 247         if (fed->fed_fcd == NULL)
 248                 RETURN(0);
 249
 250         if (flags & OBD_OPT_FAILOVER)
 251                 GOTO(free, 0);
 252
 253         /* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
 254         if (strcmp(fed->fed_fcd->fcd_uuid, obd->obd_uuid.uuid ) == 0)
 255                 GOTO(free, 0);
 256
 257         LASSERT(filter->fo_last_rcvd_slots != NULL);
 258
 259         off = fed->fed_lr_off;
 260
 261         CDEBUG(D_INFO, "freeing client at idx %u (%lld) with UUID '%s'\n",
 262                fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
 263
 264         /* Clear the bit _after_ zeroing out the client so we don't
 265            race with filter_client_add and zero out new clients.*/
 266         if (!test_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
 267                 CERROR("FILTER client %u: bit already clear in bitmap!!\n",
 268                        fed->fed_lr_idx);
 269                 LBUG();
 270         }
 271
 272         memset(&zero_fcd, 0, sizeof zero_fcd);
 273         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 274         rc = fsfilt_write_record(obd, filter->fo_rcvd_filp, &zero_fcd,
 275                                  sizeof(zero_fcd), &off, 1);
 276         if (rc == 0)
 277                 /* update server's transno */
 278                 filter_update_server_data(obd, filter->fo_rcvd_filp,
 279                                           filter->fo_fsd, 1);
 280         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 281
 282         CDEBUG(rc == 0 ? D_INFO : D_ERROR,
 283                "zeroing disconnecting client %s at idx %u (%llu) in %s rc %d\n",
 284                fed->fed_fcd->fcd_uuid, fed->fed_lr_idx, fed->fed_lr_off,
 285                LAST_RCVD, rc);
 286
 287         if (!test_and_clear_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
 288                 CERROR("FILTER client %u: bit already clear in bitmap!!\n",
 289                        fed->fed_lr_idx);
 290                 LBUG();
 291         }
 292
 293 free:
 294         OBD_FREE(fed->fed_fcd, sizeof(*fed->fed_fcd));
 295
 296         RETURN(0);
 297 }
 298
 299 static int filter_free_server_data(struct filter_obd *filter)
 300 {
 301         OBD_FREE(filter->fo_fsd, sizeof(*filter->fo_fsd));
 302         filter->fo_fsd = NULL;
 303         OBD_FREE(filter->fo_last_rcvd_slots,
 304                  FILTER_LR_MAX_CLIENT_WORDS * sizeof(unsigned long));
 305         filter->fo_last_rcvd_slots = NULL;
 306         return 0;
 307 }
 308
 309 /* assumes caller is already in kernel ctxt */
 310 int filter_update_server_data(struct obd_device *obd, struct file *filp,
 311                               struct filter_server_data *fsd, int force_sync)
 312 {
 313         loff_t off = 0;
 314         int rc;
 315         ENTRY;
 316
 317         CDEBUG(D_INODE, "server uuid      : %s\n", fsd->fsd_uuid);
 318         CDEBUG(D_INODE, "server last_rcvd : "LPU64"\n",
 319                le64_to_cpu(fsd->fsd_last_transno));
 320         CDEBUG(D_INODE, "server last_mount: "LPU64"\n",
 321                le64_to_cpu(fsd->fsd_mount_count));
 322
 323         rc = fsfilt_write_record(obd, filp, fsd, sizeof(*fsd), &off,force_sync);
 324         if (rc)
 325                 CERROR("error writing filter_server_data: rc = %d\n", rc);
 326
 327         RETURN(rc);
 328 }
 329
 330 int filter_update_last_objid(struct obd_device *obd, obd_gr group,
 331                              int force_sync)
 332 {
 333         struct filter_obd *filter = &obd->u.filter;
 334         __u64 tmp;
 335         loff_t off = 0;
 336         int rc;
 337         ENTRY;
 338
 339         if (filter->fo_last_objid_files[group] == NULL) {
 340                 CERROR("Object group "LPU64" not fully setup; not updating "
 341                        "last_objid\n", group);
 342                 RETURN(0);
 343         }
 344
 345         CDEBUG(D_INODE, "server last_objid for group "LPU64": "LPU64"\n",
 346                group, filter->fo_last_objids[group]);
 347
 348         tmp = cpu_to_le64(filter->fo_last_objids[group]);
 349         rc = fsfilt_write_record(obd, filter->fo_last_objid_files[group],
 350                                  &tmp, sizeof(tmp), &off, force_sync);
 351         if (rc)
 352                 CERROR("error writing group "LPU64" last objid: rc = %d\n",
 353                        group, rc);
 354         RETURN(rc);
 355 }
 356
 357 /* assumes caller has already in kernel ctxt */
 358 static int filter_init_server_data(struct obd_device *obd, struct file * filp)
 359 {
 360         struct filter_obd *filter = &obd->u.filter;
 361         struct filter_server_data *fsd;
 362         struct filter_client_data *fcd = NULL;
 363         struct inode *inode = filp->f_dentry->d_inode;
 364         unsigned long last_rcvd_size = inode->i_size;
 365         __u64 mount_count;
 366         int cl_idx;
 367         loff_t off = 0;
 368         int rc;
 369
 370         /* ensure padding in the struct is the correct size */
 371         LASSERT (offsetof(struct filter_server_data, fsd_padding) +
 372                  sizeof(fsd->fsd_padding) == FILTER_LR_SERVER_SIZE);
 373         LASSERT (offsetof(struct filter_client_data, fcd_padding) +
 374                  sizeof(fcd->fcd_padding) == FILTER_LR_CLIENT_SIZE);
 375
 376         OBD_ALLOC(fsd, sizeof(*fsd));
 377         if (!fsd)
 378                 RETURN(-ENOMEM);
 379         filter->fo_fsd = fsd;
 380
 381         OBD_ALLOC(filter->fo_last_rcvd_slots,
 382                   FILTER_LR_MAX_CLIENT_WORDS * sizeof(unsigned long));
 383         if (filter->fo_last_rcvd_slots == NULL) {
 384                 OBD_FREE(fsd, sizeof(*fsd));
 385                 RETURN(-ENOMEM);
 386         }
 387
 388         if (last_rcvd_size == 0) {
 389                 CWARN("%s: initializing new %s\n", obd->obd_name, LAST_RCVD);
 390
 391                 memcpy(fsd->fsd_uuid, obd->obd_uuid.uuid,sizeof(fsd->fsd_uuid));
 392                 fsd->fsd_last_transno = 0;
 393                 mount_count = fsd->fsd_mount_count = 0;
 394                 fsd->fsd_server_size = cpu_to_le32(FILTER_LR_SERVER_SIZE);
 395                 fsd->fsd_client_start = cpu_to_le32(FILTER_LR_CLIENT_START);
 396                 fsd->fsd_client_size = cpu_to_le16(FILTER_LR_CLIENT_SIZE);
 397                 fsd->fsd_subdir_count = cpu_to_le16(FILTER_SUBDIR_COUNT);
 398                 filter->fo_subdir_count = FILTER_SUBDIR_COUNT;
 399         } else {
 400                 rc = fsfilt_read_record(obd, filp, fsd, sizeof(*fsd), &off);
 401                 if (rc) {
 402                         CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
 403                                LAST_RCVD, rc);
 404                         GOTO(err_fsd, rc);
 405                 }
 406                 if (strcmp(fsd->fsd_uuid, obd->obd_uuid.uuid) != 0) {
 407                         CERROR("OBD UUID %s does not match last_rcvd UUID %s\n",
 408                                obd->obd_uuid.uuid, fsd->fsd_uuid);
 409                         GOTO(err_fsd, rc = -EINVAL);
 410                 }
 411                 mount_count = le64_to_cpu(fsd->fsd_mount_count);
 412                 filter->fo_subdir_count = le16_to_cpu(fsd->fsd_subdir_count);
 413         }
 414
 415         if (fsd->fsd_feature_incompat & ~cpu_to_le32(FILTER_INCOMPAT_SUPP)) {
 416                 CERROR("unsupported feature %x\n",
 417                        le32_to_cpu(fsd->fsd_feature_incompat) &
 418                        ~FILTER_INCOMPAT_SUPP);
 419                 GOTO(err_fsd, rc = -EINVAL);
 420         }
 421         if (fsd->fsd_feature_rocompat & ~cpu_to_le32(FILTER_ROCOMPAT_SUPP)) {
 422                 CERROR("read-only feature %x\n",
 423                        le32_to_cpu(fsd->fsd_feature_rocompat) &
 424                        ~FILTER_ROCOMPAT_SUPP);
 425                 /* Do something like remount filesystem read-only */
 426                 GOTO(err_fsd, rc = -EINVAL);
 427         }
 428
 429         CDEBUG(D_INODE, "%s: server last_rcvd : "LPU64"\n",
 430                obd->obd_name, le64_to_cpu(fsd->fsd_last_transno));
 431         CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
 432                obd->obd_name, mount_count + 1);
 433         CDEBUG(D_INODE, "%s: server data size: %u\n",
 434                obd->obd_name, le32_to_cpu(fsd->fsd_server_size));
 435         CDEBUG(D_INODE, "%s: per-client data start: %u\n",
 436                obd->obd_name, le32_to_cpu(fsd->fsd_client_start));
 437         CDEBUG(D_INODE, "%s: per-client data size: %u\n",
 438                obd->obd_name, le32_to_cpu(fsd->fsd_client_size));
 439         CDEBUG(D_INODE, "%s: server subdir_count: %u\n",
 440                obd->obd_name, le16_to_cpu(fsd->fsd_subdir_count));
 441         CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
 442                last_rcvd_size <= le32_to_cpu(fsd->fsd_client_start) ? 0 :
 443                (last_rcvd_size - le32_to_cpu(fsd->fsd_client_start)) /
 444                 le16_to_cpu(fsd->fsd_client_size));
 445
 446         if (!obd->obd_replayable) {
 447                 CWARN("%s: recovery support OFF\n", obd->obd_name);
 448                 GOTO(out, rc = 0);
 449         }
 450
 451         for (cl_idx = 0, off = le32_to_cpu(fsd->fsd_client_start);
 452              off < last_rcvd_size; cl_idx++) {
 453                 __u64 last_rcvd;
 454                 struct obd_export *exp;
 455                 struct filter_export_data *fed;
 456
 457                 if (!fcd) {
 458                         OBD_ALLOC(fcd, sizeof(*fcd));
 459                         if (!fcd)
 460                                 GOTO(err_client, rc = -ENOMEM);
 461                 }
 462
 463                 /* Don't assume off is incremented properly by
 464                  * fsfilt_read_record(), in case sizeof(*fcd)
 465                  * isn't the same as fsd->fsd_client_size.  */
 466                 off = le32_to_cpu(fsd->fsd_client_start) +
 467                         cl_idx * le16_to_cpu(fsd->fsd_client_size);
 468                 rc = fsfilt_read_record(obd, filp, fcd, sizeof(*fcd), &off);
 469                 if (rc) {
 470                         CERROR("error reading FILT %s idx %d off %llu: rc %d\n",
 471                                LAST_RCVD, cl_idx, off, rc);
 472                         break; /* read error shouldn't cause startup to fail */
 473                 }
 474
 475                 if (fcd->fcd_uuid[0] == '\0') {
 476                         CDEBUG(D_INFO, "skipping zeroed client at offset %d\n",
 477                                cl_idx);
 478                         continue;
 479                 }
 480
 481                 last_rcvd = le64_to_cpu(fcd->fcd_last_rcvd);
 482
 483                 /* These exports are cleaned up by filter_disconnect(), so they
 484                  * need to be set up like real exports as filter_connect() does.
 485                  */
 486                 exp = class_new_export(obd);
 487                 CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64" "
 488                        "srv lr: "LPU64" fcd_group %d \n", fcd->fcd_uuid, cl_idx,
 489                        last_rcvd, le64_to_cpu(fsd->fsd_last_transno),
 490                        le32_to_cpu(fcd->fcd_group));
 491                 if (exp == NULL)
 492                         GOTO(err_client, rc = -ENOMEM);
 493
 494                 memcpy(&exp->exp_client_uuid.uuid, fcd->fcd_uuid,
 495                        sizeof exp->exp_client_uuid.uuid);
 496                 fed = &exp->exp_filter_data;
 497                 fed->fed_fcd = fcd;
 498                 fed->fed_group = le32_to_cpu(fcd->fcd_group);
 499                 filter_client_add(obd, filter, fed, cl_idx);
 500                 /* create helper if export init gets more complex */
 501                 spin_lock_init(&fed->fed_lock);
 502
 503                 fcd = NULL;
 504                 exp->exp_replay_needed = 1;
 505                 obd->obd_recoverable_clients++;
 506                 obd->obd_max_recoverable_clients++;
 507                 class_export_put(exp);
 508
 509                 CDEBUG(D_OTHER, "client at idx %d has last_rcvd = "LPU64"\n",
 510                        cl_idx, last_rcvd);
 511
 512                 if (last_rcvd > le64_to_cpu(fsd->fsd_last_transno))
 513                         fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
 514
 515         }
 516
 517         obd->obd_last_committed = le64_to_cpu(fsd->fsd_last_transno);
 518
 519         if (obd->obd_recoverable_clients) {
 520                 CWARN("RECOVERY: service %s, %d recoverable clients, "
 521                       "last_transno "LPU64"\n", obd->obd_name,
 522                       obd->obd_recoverable_clients,
 523                       le64_to_cpu(fsd->fsd_last_transno));
 524                 obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
 525                 target_start_recovery_thread(obd, ost_handle);
 526         }
 527
 528         if (fcd)
 529                 OBD_FREE(fcd, sizeof(*fcd));
 530
 531 out:
 532         filter->fo_mount_count = mount_count + 1;
 533         fsd->fsd_mount_count = cpu_to_le64(filter->fo_mount_count);
 534
 535         /* save it, so mount count and last_transno is current */
 536         rc = filter_update_server_data(obd, filp, filter->fo_fsd, 1);
 537         if (rc)
 538                 GOTO(err_client, rc);
 539         RETURN(0);
 540
 541 err_client:
 542         class_disconnect_exports(obd, 0);
 543 err_fsd:
 544         filter_free_server_data(filter);
 545         RETURN(rc);
 546 }
 547
 548 static int filter_cleanup_groups(struct obd_device *obd)
 549 {
 550         struct filter_obd *filter = &obd->u.filter;
 551         struct dentry *dentry;
 552         int i, k;
 553         ENTRY;
 554
 555         for (i = 0; i < filter->fo_group_count; i++) {
 556                 if (filter->fo_subdirs != NULL) {
 557                         for (k = 0; k < filter->fo_subdir_count; k++) {
 558                                 dentry = filter->fo_subdirs[i].dentry[k];
 559                                 if (dentry == NULL)
 560                                         continue;
 561                                 f_dput(dentry);
 562                                 filter->fo_subdirs[i].dentry[k] = NULL;
 563                         }
 564                 }
 565                 if (filter->fo_last_objid_files[i] != NULL) {
 566                         filp_close(filter->fo_last_objid_files[i], 0);
 567                         filter->fo_last_objid_files[i] = NULL;
 568                 }
 569                 if (filter->fo_groups[i] != NULL) {
 570                         dput(filter->fo_groups[i]);
 571                         filter->fo_groups[i] = NULL;
 572                 }
 573         }
 574         if (filter->fo_subdirs != NULL)
 575                 OBD_FREE(filter->fo_subdirs,
 576                          filter->fo_group_count * sizeof(*filter->fo_subdirs));
 577         if (filter->fo_groups != NULL)
 578                 OBD_FREE(filter->fo_groups,
 579                          filter->fo_group_count * sizeof(*filter->fo_groups));
 580         if (filter->fo_last_objids != NULL)
 581                 OBD_FREE(filter->fo_last_objids,
 582                          filter->fo_group_count * sizeof(__u64));
 583         if (filter->fo_last_objid_files != NULL)
 584                 OBD_FREE(filter->fo_last_objid_files,
 585                          filter->fo_group_count * sizeof(struct file *));
 586         f_dput(filter->fo_dentry_O);
 587         RETURN(0);
 588 }
 589
 590 static int filter_update_last_group(struct obd_device *obd, int group)
 591 {
 592         struct filter_obd *filter = &obd->u.filter;
 593         struct file *filp = NULL;
 594         int last_group = 0, rc;
 595         loff_t off = 0;
 596         ENTRY;
 597
 598         if (group <= filter->fo_committed_group)
 599                 RETURN(0);
 600
 601         filp = filp_open("LAST_GROUP", O_RDWR, 0700);
 602         if (IS_ERR(filp)) {
 603                 rc = PTR_ERR(filp);
 604                 filp = NULL;
 605                 CERROR("cannot open LAST_GROUP: rc = %d\n", rc);
 606                 GOTO(cleanup, rc);
 607         }
 608
 609         rc = fsfilt_read_record(obd, filp, &last_group, sizeof(__u32), &off);
 610         if (rc) {
 611                 CDEBUG(D_INODE, "error reading LAST_GROUP: rc %d\n",rc);
 612                 GOTO(cleanup, rc);
 613         }
 614         LASSERT(off == 0 || last_group >= FILTER_MIN_GROUPS);
 615         CDEBUG(D_INODE, "%s: previous %d, new %d\n",
 616                obd->obd_name, last_group, group);
 617
 618         off = 0;
 619         last_group = group;
 620         /* must be sync: bXXXX */
 621         rc = fsfilt_write_record(obd, filp, &last_group, sizeof(__u32), &off, 1);
 622         if (rc) {
 623                 CDEBUG(D_INODE, "error updating LAST_GROUP: rc %d\n", rc);
 624                 GOTO(cleanup, rc);
 625         }
 626
 627         filter->fo_committed_group = group;
 628 cleanup:
 629         if (filp)
 630                 filp_close(filp, 0);
 631         RETURN(rc);
 632 }
 633
 634 static int filter_read_group_internal(struct obd_device *obd, int group,
 635                                       int create)
 636 {
 637         struct filter_obd *filter = &obd->u.filter;
 638         __u64 *new_objids = NULL;
 639         struct filter_subdirs *new_subdirs = NULL, *tmp_subdirs = NULL;
 640         struct dentry **new_groups = NULL;
 641         struct file **new_files = NULL;
 642         struct dentry *dentry;
 643         struct file *filp;
 644         int old_count = filter->fo_group_count, rc, stage = 0, i;
 645         char name[25];
 646         __u64 last_objid;
 647         loff_t off = 0;
 648
 649         snprintf(name, 24, "%d", group);
 650         name[24] = '\0';
 651
 652         if (!create) {
 653                 dentry = ll_lookup_one_len(name, filter->fo_dentry_O,
 654                                            strlen(name));
 655                 if (IS_ERR(dentry)) {
 656                         CERROR("Cannot lookup expected object group %d: %ld\n",
 657                                group, PTR_ERR(dentry));
 658                         RETURN(PTR_ERR(dentry));
 659                 }
 660         } else {
 661                 dentry = simple_mkdir(filter->fo_dentry_O, name, 0700, 1);
 662                 if (IS_ERR(dentry)) {
 663                         CERROR("cannot lookup/create O/%s: rc = %ld\n", name,
 664                                PTR_ERR(dentry));
 665                         RETURN(PTR_ERR(dentry));
 666                 }
 667         }
 668         stage = 1;
 669
 670         snprintf(name, 24, "O/%d/LAST_ID", group);
 671         name[24] = '\0';
 672         filp = filp_open(name, O_CREAT | O_RDWR, 0700);
 673         if (IS_ERR(filp)) {
 674                 CERROR("cannot create %s: rc = %ld\n", name, PTR_ERR(filp));
 675                 GOTO(cleanup, rc = PTR_ERR(filp));
 676         }
 677         stage = 2;
 678
 679         rc = fsfilt_read_record(obd, filp, &last_objid, sizeof(__u64), &off);
 680         if (rc) {
 681                 CDEBUG(D_INODE, "error reading %s: rc %d\n", name, rc);
 682                 GOTO(cleanup, rc);
 683         }
 684
 685         if (filter->fo_subdir_count) {
 686                 OBD_ALLOC(tmp_subdirs, sizeof(*tmp_subdirs));
 687                 if (tmp_subdirs == NULL)
 688                         GOTO(cleanup, rc = -ENOMEM);
 689                 stage = 3;
 690
 691                 for (i = 0; i < filter->fo_subdir_count; i++) {
 692                         char dir[20];
 693                         snprintf(dir, sizeof(dir), "d%u", i);
 694
 695                         tmp_subdirs->dentry[i] =
 696                                 simple_mkdir(dentry, dir, 0700, 1);
 697                         if (IS_ERR(tmp_subdirs->dentry[i])) {
 698                                 rc = PTR_ERR(tmp_subdirs->dentry[i]);
 699                                 CERROR("can't lookup/create O/%d/%s: rc = %d\n",
 700                                        group, dir, rc);
 701                                 GOTO(cleanup, rc);
 702                         }
 703                         CDEBUG(D_INODE, "got/created O/%d/%s: %p\n", group, dir,
 704                                tmp_subdirs->dentry[i]);
 705                 }
 706         }
 707
 708         /* 'group' is an index; we need an array of length 'group + 1' */
 709         if (group + 1 > old_count) {
 710                 int len = group + 1;
 711                 OBD_ALLOC(new_objids, len * sizeof(*new_objids));
 712                 OBD_ALLOC(new_subdirs, len * sizeof(*new_subdirs));
 713                 OBD_ALLOC(new_groups, len * sizeof(*new_groups));
 714                 OBD_ALLOC(new_files, len * sizeof(*new_files));
 715                 stage = 4;
 716                 if (new_objids == NULL || new_subdirs == NULL ||
 717                     new_groups == NULL || new_files == NULL)
 718                         GOTO(cleanup, rc = -ENOMEM);
 719
 720                 memcpy(new_objids, filter->fo_last_objids,
 721                        old_count * sizeof(*new_objids));
 722                 memcpy(new_subdirs, filter->fo_subdirs,
 723                        old_count * sizeof(*new_subdirs));
 724                 memcpy(new_groups, filter->fo_groups,
 725                        old_count * sizeof(*new_groups));
 726                 memcpy(new_files, filter->fo_last_objid_files,
 727                        old_count * sizeof(*new_files));
 728
 729                 if (old_count) {
 730                         OBD_FREE(filter->fo_last_objids,
 731                                  old_count * sizeof(*new_objids));
 732                         OBD_FREE(filter->fo_subdirs,
 733                                  old_count * sizeof(*new_subdirs));
 734                         OBD_FREE(filter->fo_groups,
 735                                  old_count * sizeof(*new_groups));
 736                         OBD_FREE(filter->fo_last_objid_files,
 737                                  old_count * sizeof(*new_files));
 738                 }
 739                 filter->fo_last_objids = new_objids;
 740                 filter->fo_subdirs = new_subdirs;
 741                 filter->fo_groups = new_groups;
 742                 filter->fo_last_objid_files = new_files;
 743                 filter->fo_group_count = len;
 744         }
 745
 746         filter->fo_groups[group] = dentry;
 747         filter->fo_last_objid_files[group] = filp;
 748         if (filter->fo_subdir_count) {
 749                 filter->fo_subdirs[group] = *tmp_subdirs;
 750                 OBD_FREE(tmp_subdirs, sizeof(*tmp_subdirs));
 751         }
 752
 753         filter_update_last_group(obd, group);
 754
 755         if (filp->f_dentry->d_inode->i_size == 0) {
 756                 filter->fo_last_objids[group] = FILTER_INIT_OBJID;
 757                 RETURN(0);
 758         }
 759
 760         filter->fo_last_objids[group] = le64_to_cpu(last_objid);
 761         CDEBUG(D_INODE, "%s: server last_objid group %d: "LPU64"\n",
 762                obd->obd_name, group, last_objid);
 763         RETURN(0);
 764  cleanup:
 765         switch (stage) {
 766         case 4:
 767                 if (new_objids != NULL)
 768                         OBD_FREE(new_objids, group * sizeof(*new_objids));
 769                 if (new_subdirs != NULL)
 770                         OBD_FREE(new_subdirs, group * sizeof(*new_subdirs));
 771                 if (new_groups != NULL)
 772                         OBD_FREE(new_groups, group * sizeof(*new_groups));
 773                 if (new_files != NULL)
 774                         OBD_FREE(new_files, group * sizeof(*new_files));
 775         case 3:
 776                 if (filter->fo_subdir_count) {
 777                         for (i = 0; i < filter->fo_subdir_count; i++) {
 778                                 if (tmp_subdirs->dentry[i] != NULL)
 779                                         dput(tmp_subdirs->dentry[i]);
 780                         }
 781                         OBD_FREE(tmp_subdirs, sizeof(*tmp_subdirs));
 782                 }
 783         case 2:
 784                 filp_close(filp, 0);
 785         case 1:
 786                 dput(dentry);
 787         }
 788         RETURN(rc);
 789 }
 790
 791 static int filter_read_groups(struct obd_device *obd, int last_group,
 792                               int create)
 793 {
 794         struct filter_obd *filter = &obd->u.filter;
 795         int old_count, group, rc = 0;
 796
 797         down(&filter->fo_init_lock);
 798         old_count = filter->fo_group_count;
 799         for (group = old_count; group <= last_group; group++) {
 800                 if (group == 0)
 801                         continue; /* no group zero */
 802
 803                 rc = filter_read_group_internal(obd, group, create);
 804                 if (rc != 0)
 805                         break;
 806         }
 807         up(&filter->fo_init_lock);
 808         return rc;
 809 }
 810
 811 static int filter_prep_groups(struct obd_device *obd)
 812 {
 813         struct filter_obd *filter = &obd->u.filter;
 814         struct dentry *dentry, *O_dentry;
 815         int rc = 0, cleanup_phase = 0;
 816         struct file *filp = NULL;
 817         int last_group;
 818         loff_t off = 0;
 819         ENTRY;
 820
 821         O_dentry = simple_mkdir(current->fs->pwd, "O", 0700, 1);
 822         CDEBUG(D_INODE, "got/created O: %p\n", O_dentry);
 823         if (IS_ERR(O_dentry)) {
 824                 rc = PTR_ERR(O_dentry);
 825                 CERROR("cannot open/create O: rc = %d\n", rc);
 826                 GOTO(cleanup, rc);
 827         }
 828         filter->fo_dentry_O = O_dentry;
 829         cleanup_phase = 1; /* O_dentry */
 830
 831         /* Lookup "R" to tell if we're on an old OST FS and need to convert
 832          * from O/R/<dir>/<objid> to O/0/<dir>/<objid>.  This can be removed
 833          * some time post 1.0 when all old-style OSTs have converted along
 834          * with the init_objid hack. */
 835         dentry = ll_lookup_one_len("R", O_dentry, 1);
 836         if (IS_ERR(dentry))
 837                 GOTO(cleanup, rc = PTR_ERR(dentry));
 838         if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
 839                 struct dentry *O0_dentry = lookup_one_len("0", O_dentry, 1);
 840                 ENTRY;
 841
 842                 CWARN("converting OST to new object layout\n");
 843                 if (IS_ERR(O0_dentry)) {
 844                         rc = PTR_ERR(O0_dentry);
 845                         CERROR("error looking up O/0: rc %d\n", rc);
 846                         GOTO(cleanup_R, rc);
 847                 }
 848
 849                 if (O0_dentry->d_inode) {
 850                         CERROR("Both O/R and O/0 exist. Fix manually.\n");
 851                         GOTO(cleanup_O0, rc = -EEXIST);
 852                 }
 853
 854                 down(&O_dentry->d_inode->i_sem);
 855                 rc = vfs_rename(O_dentry->d_inode, dentry,
 856                                 O_dentry->d_inode, O0_dentry);
 857                 up(&O_dentry->d_inode->i_sem);
 858
 859                 if (rc) {
 860                         CERROR("error renaming O/R to O/0: rc %d\n", rc);
 861                         GOTO(cleanup_O0, rc);
 862                 }
 863                 filter->fo_fsd->fsd_feature_incompat |=
 864                         cpu_to_le32(FILTER_INCOMPAT_GROUPS);
 865                 rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
 866                                                filter->fo_fsd, 1);
 867                 GOTO(cleanup_O0, rc);
 868
 869         cleanup_O0:
 870                 f_dput(O0_dentry);
 871         cleanup_R:
 872                 f_dput(dentry);
 873                 if (rc)
 874                         GOTO(cleanup, rc);
 875         } else {
 876                 f_dput(dentry);
 877         }
 878
 879         cleanup_phase = 2; /* groups */
 880
 881         /* we have to initialize all groups before first connections from
 882          * clients because they may send create/destroy for any group -bzzz */
 883         filp = filp_open("LAST_GROUP", O_CREAT | O_RDWR, 0700);
 884         if (IS_ERR(filp)) {
 885                 CERROR("cannot create LAST_GROUP: rc = %ld\n", PTR_ERR(filp));
 886                 GOTO(cleanup, rc = PTR_ERR(filp));
 887         }
 888         cleanup_phase = 3; /* filp */
 889
 890         rc = fsfilt_read_record(obd, filp, &last_group, sizeof(__u32), &off);
 891         if (rc) {
 892                 CDEBUG(D_INODE, "error reading LAST_GROUP: rc %d\n", rc);
 893                 GOTO(cleanup, rc);
 894         }
 895         if (off == 0) {
 896                 last_group = FILTER_MIN_GROUPS;
 897         } else {
 898                 LASSERT(last_group >= FILTER_MIN_GROUPS);
 899         }
 900
 901         CWARN("%s: initialize groups [%d,%d]\n", obd->obd_name,
 902               FILTER_MIN_GROUPS, last_group);
 903         filter->fo_committed_group = last_group;
 904         rc = filter_read_groups(obd, last_group, 1);
 905         if (rc)
 906                 GOTO(cleanup, rc);
 907
 908         filp_close(filp, 0);
 909         RETURN(0);
 910
 911  cleanup:
 912         switch (cleanup_phase) {
 913         case 3:
 914                 filp_close(filp, 0);
 915         case 2:
 916                 filter_cleanup_groups(obd);
 917         case 1:
 918                 f_dput(filter->fo_dentry_O);
 919                 filter->fo_dentry_O = NULL;
 920         default:
 921                 break;
 922         }
 923         return rc;
 924 }
 925
 926 /* setup the object store with correct subdirectories */
 927 static int filter_prep(struct obd_device *obd)
 928 {
 929         struct lvfs_run_ctxt saved;
 930         struct filter_obd *filter = &obd->u.filter;
 931         struct file *file;
 932         struct inode *inode;
 933         int rc = 0;
 934         ENTRY;
 935
 936         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 937         file = filp_open(LAST_RCVD, O_RDWR | O_CREAT | O_LARGEFILE, 0700);
 938         if (!file || IS_ERR(file)) {
 939                 rc = PTR_ERR(file);
 940                 CERROR("OBD filter: cannot open/create %s: rc = %d\n",
 941                        LAST_RCVD, rc);
 942                 GOTO(out, rc);
 943         }
 944
 945         if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
 946                 CERROR("%s is not a regular file!: mode = %o\n", LAST_RCVD,
 947                        file->f_dentry->d_inode->i_mode);
 948                 GOTO(err_filp, rc = -ENOENT);
 949         }
 950
 951         /* steal operations */
 952         inode = file->f_dentry->d_inode;
 953         filter->fo_fop = file->f_op;
 954         filter->fo_iop = inode->i_op;
 955         filter->fo_aops = inode->i_mapping->a_ops;
 956
 957         rc = filter_init_server_data(obd, file);
 958         if (rc) {
 959                 CERROR("cannot read %s: rc = %d\n", LAST_RCVD, rc);
 960                 GOTO(err_filp, rc);
 961         }
 962         filter->fo_rcvd_filp = file;
 963
 964         rc = filter_prep_groups(obd);
 965         if (rc)
 966                 GOTO(err_server_data, rc);
 967
 968  out:
 969         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 970
 971         return(rc);
 972
 973  err_server_data:
 974         //class_disconnect_exports(obd, 0);
 975         filter_free_server_data(filter);
 976  err_filp:
 977         if (filp_close(file, 0))
 978                 CERROR("can't close %s after error\n", LAST_RCVD);
 979         filter->fo_rcvd_filp = NULL;
 980         goto out;
 981 }
 982
 983 /* cleanup the filter: write last used object id to status file */
 984 static void filter_post(struct obd_device *obd)
 985 {
 986         struct lvfs_run_ctxt saved;
 987         struct filter_obd *filter = &obd->u.filter;
 988         int rc, i;
 989
 990         /* XXX: filter_update_lastobjid used to call fsync_dev.  It might be
 991          * best to start a transaction with h_sync, because we removed this
 992          * from lastobjid */
 993
 994         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 995         rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
 996                                        filter->fo_fsd, 0);
 997         if (rc)
 998                 CERROR("error writing server data: rc = %d\n", rc);
 999
1000         for (i = 1; i < filter->fo_group_count; i++) {
1001                 rc = filter_update_last_objid(obd, i,
1002                                              (i == filter->fo_group_count - 1));
1003                 if (rc)
1004                         CERROR("error writing group %d lastobjid: rc = %d\n",
1005                                i, rc);
1006         }
1007
1008         rc = filp_close(filter->fo_rcvd_filp, 0);
1009         filter->fo_rcvd_filp = NULL;
1010         if (rc)
1011                 CERROR("error closing %s: rc = %d\n", LAST_RCVD, rc);
1012
1013         filter_cleanup_groups(obd);
1014         filter_free_server_data(filter);
1015         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1016 }
1017
1018 static void filter_set_last_id(struct filter_obd *filter, int group, obd_id id)
1019 {
1020         LASSERT(filter->fo_fsd != NULL);
1021         LASSERT(group > 0);
1022         LASSERT(group < filter->fo_group_count);
1023
1024         spin_lock(&filter->fo_objidlock);
1025         filter->fo_last_objids[group] = id;
1026         spin_unlock(&filter->fo_objidlock);
1027 }
1028
1029 __u64 filter_last_id(struct filter_obd *filter, int group)
1030 {
1031         obd_id id;
1032         LASSERT(filter->fo_fsd != NULL);
1033         LASSERT(group > 0);
1034         LASSERT(group < filter->fo_group_count);
1035
1036         spin_lock(&filter->fo_objidlock);
1037         id = filter->fo_last_objids[group];
1038         spin_unlock(&filter->fo_objidlock);
1039
1040         return id;
1041 }
1042
1043 /* direct cut-n-paste of mds_blocking_ast() */
1044 static int filter_blocking_ast(struct ldlm_lock *lock,
1045                                struct ldlm_lock_desc *desc,
1046                                void *data, int flag)
1047 {
1048         int do_ast;
1049         ENTRY;
1050
1051         if (flag == LDLM_CB_CANCELING) {
1052                 /* Don't need to do anything here. */
1053                 RETURN(0);
1054         }
1055
1056         /* XXX layering violation!  -phil */
1057         l_lock(&lock->l_resource->lr_namespace->ns_lock);
1058         /* Get this: if filter_blocking_ast is racing with ldlm_intent_policy,
1059          * such that filter_blocking_ast is called just before l_i_p takes the
1060          * ns_lock, then by the time we get the lock, we might not be the
1061          * correct blocking function anymore.  So check, and return early, if
1062          * so. */
1063         if (lock->l_blocking_ast != filter_blocking_ast) {
1064                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
1065                 RETURN(0);
1066         }
1067
1068         lock->l_flags |= LDLM_FL_CBPENDING;
1069         do_ast = (!lock->l_readers && !lock->l_writers);
1070         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
1071
1072         if (do_ast) {
1073                 struct lustre_handle lockh;
1074                 int rc;
1075
1076                 LDLM_DEBUG(lock, "already unused, calling ldlm_cli_cancel");
1077                 ldlm_lock2handle(lock, &lockh);
1078                 rc = ldlm_cli_cancel(&lockh);
1079                 if (rc < 0)
1080                         CERROR("ldlm_cli_cancel: %d\n", rc);
1081         } else {
1082                 LDLM_DEBUG(lock, "Lock still has references, will be "
1083                            "cancelled later");
1084         }
1085         RETURN(0);
1086 }
1087
1088 extern void *lock_dir(struct inode *dir, struct qstr *name);
1089 extern void unlock_dir(struct inode *dir, void *lock);
1090
1091 static void * filter_lock_dentry(struct obd_device *obd,
1092                                  struct dentry *dparent, obd_id id)
1093 {
1094 #ifdef S_PDIROPS
1095         struct qstr qstr;
1096         char name[32];
1097         qstr.name = name;
1098         qstr.len = sprintf(name, LPU64, id);
1099         return lock_dir(dparent->d_inode, &qstr);
1100 #else
1101         down(&dparent->d_inode->i_sem);
1102 #endif
1103         return 0;
1104 }
1105
1106 /* We never dget the object parent, so DON'T dput it either */
1107 static void filter_parent_unlock(struct dentry *dparent, void *lock)
1108 {
1109 #ifdef S_PDIROPS
1110         LASSERT(lock != NULL);
1111         unlock_dir(dparent->d_inode, lock);
1112 #else
1113         up(&dparent->d_inode->i_sem);
1114 #endif
1115 }
1116
1117 /* We never dget the object parent, so DON'T dput it either */
1118 struct dentry *filter_parent(struct obd_device *obd, obd_gr group, obd_id objid)
1119 {
1120         struct filter_obd *filter = &obd->u.filter;
1121         LASSERT(group < filter->fo_group_count);
1122         LASSERT(group > 0);
1123
1124         if (filter->fo_subdir_count == 0)
1125                 return filter->fo_groups[group];
1126
1127         return filter->fo_subdirs[group].dentry[objid & (filter->fo_subdir_count - 1)];
1128 }
1129
1130 /* We never dget the object parent, so DON'T dput it either */
1131 struct dentry *filter_parent_lock(struct obd_device *obd, obd_gr group,
1132                                   obd_id objid, void **lock)
1133 {
1134         unsigned long now = jiffies;
1135         struct dentry *dparent = filter_parent(obd, group, objid);
1136
1137         if (IS_ERR(dparent))
1138                 return dparent;
1139
1140         LASSERT(dparent);
1141         LASSERT(dparent->d_inode);
1142
1143         *lock = filter_lock_dentry(obd, dparent, objid);
1144         if (time_after(jiffies, now + 15 * HZ))
1145                 CERROR("slow parent lock %lus\n", (jiffies - now) / HZ);
1146         return dparent;
1147 }
1148
1149 /* How to get files, dentries, inodes from object id's.
1150  *
1151  * If dir_dentry is passed, the caller has already locked the parent
1152  * appropriately for this operation (normally a write lock).  If
1153  * dir_dentry is NULL, we do a read lock while we do the lookup to
1154  * avoid races with create/destroy and such changing the directory
1155  * internal to the filesystem code. */
1156 struct dentry *filter_id2dentry(struct obd_device *obd,
1157                                 struct dentry *dir_dentry,
1158                                 obd_gr group, obd_id id)
1159 {
1160         struct dentry *dparent = dir_dentry;
1161         struct dentry *dchild;
1162         void *lock = NULL;
1163         char name[32];
1164         int len;
1165         ENTRY;
1166
1167         if (id == 0) {
1168                 CERROR("fatal: invalid object id 0\n");
1169                 RETURN(ERR_PTR(-ESTALE));
1170         }
1171
1172         len = sprintf(name, LPU64, id);
1173         if (dir_dentry == NULL) {
1174                 dparent = filter_parent_lock(obd, group, id, &lock);
1175                 if (IS_ERR(dparent))
1176                         RETURN(dparent);
1177         }
1178         CDEBUG(D_INODE, "looking up object O/%*s/%s\n",
1179                dparent->d_name.len, dparent->d_name.name, name);
1180         dchild = /*ll_*/lookup_one_len(name, dparent, len);
1181         if (dir_dentry == NULL)
1182                 filter_parent_unlock(dparent, lock);
1183         if (IS_ERR(dchild)) {
1184                 CERROR("%s: child lookup error %ld\n", obd->obd_name,
1185                        PTR_ERR(dchild));
1186                 RETURN(dchild);
1187         }
1188
1189         if (dchild->d_inode != NULL && is_bad_inode(dchild->d_inode)) {
1190                 CERROR("%s: got bad inode "LPU64"\n", obd->obd_name, id);
1191                 f_dput(dchild);
1192                 RETURN(ERR_PTR(-ENOENT));
1193         }
1194
1195         CDEBUG(D_INODE, "got child objid %s: %p, count = %d\n",
1196                name, dchild, atomic_read(&dchild->d_count));
1197
1198         LASSERT(atomic_read(&dchild->d_count) > 0);
1199
1200         RETURN(dchild);
1201 }
1202
1203 static int filter_prepare_destroy(struct obd_device *obd, obd_id objid,
1204                                   obd_id group)
1205 {
1206         struct lustre_handle lockh;
1207         int flags = LDLM_AST_DISCARD_DATA, rc;
1208         struct ldlm_res_id res_id = { .name = { objid, 0, group, 0 } };
1209         ldlm_policy_data_t policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
1210
1211         ENTRY;
1212         /* Tell the clients that the object is gone now and that they should
1213          * throw away any cached pages. */
1214         rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, res_id,
1215                               LDLM_EXTENT, &policy, LCK_PW,
1216                               &flags, filter_blocking_ast, ldlm_completion_ast,
1217                               NULL, NULL, NULL, 0, NULL, &lockh);
1218
1219         /* We only care about the side-effects, just drop the lock. */
1220         if (rc == ELDLM_OK)
1221                 ldlm_lock_decref(&lockh, LCK_PW);
1222
1223         RETURN(rc);
1224 }
1225
1226 /* Caller must hold LCK_PW on parent and push us into kernel context.
1227  * Caller is also required to ensure that dchild->d_inode exists. */
1228 static int filter_destroy_internal(struct obd_device *obd, obd_id objid,
1229                                    struct dentry *dparent,
1230                                    struct dentry *dchild)
1231 {
1232         struct inode *inode = dchild->d_inode;
1233         int rc;
1234         ENTRY;
1235
1236         if (inode->i_nlink != 1 || atomic_read(&inode->i_count) != 1) {
1237                 CERROR("destroying objid %*s nlink = %lu, count = %d\n",
1238                        dchild->d_name.len, dchild->d_name.name,
1239                        (unsigned long)inode->i_nlink,
1240                        atomic_read(&inode->i_count));
1241         }
1242
1243         rc = vfs_unlink(dparent->d_inode, dchild);
1244
1245         if (rc)
1246                 CERROR("error unlinking objid %*s: rc %d\n",
1247                        dchild->d_name.len, dchild->d_name.name, rc);
1248
1249         RETURN(rc);
1250 }
1251
1252 static int filter_intent_policy(struct ldlm_namespace *ns,
1253                                 struct ldlm_lock **lockp, void *req_cookie,
1254                                 ldlm_mode_t mode, int flags, void *data)
1255 {
1256         struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
1257         struct ptlrpc_request *req = req_cookie;
1258         struct ldlm_lock *lock = *lockp, *l = NULL;
1259         struct ldlm_resource *res = lock->l_resource;
1260         ldlm_processing_policy policy;
1261         struct ost_lvb *res_lvb, *reply_lvb;
1262         struct ldlm_reply *rep;
1263         struct list_head *tmp;
1264         ldlm_error_t err;
1265         int tmpflags = 0, rc, repsize[2] = {sizeof(*rep), sizeof(*reply_lvb)};
1266         ENTRY;
1267
1268         policy = ldlm_get_processing_policy(res);
1269         LASSERT(policy != NULL);
1270         LASSERT(req != NULL);
1271
1272         rc = lustre_pack_reply(req, 2, repsize, NULL);
1273         if (rc)
1274                 RETURN(req->rq_status = rc);
1275
1276         rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*rep));
1277         LASSERT(rep != NULL);
1278
1279         reply_lvb = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*reply_lvb));
1280         LASSERT(reply_lvb != NULL);
1281
1282         //fixup_handle_for_resent_req(req, lock, &lockh);
1283
1284         /* If we grant any lock at all, it will be a whole-file read lock.
1285          * Call the extent policy function to see if our request can be
1286          * granted, or is blocked. */
1287         lock->l_policy_data.l_extent.start = 0;
1288         lock->l_policy_data.l_extent.end = OBD_OBJECT_EOF;
1289         lock->l_req_mode = LCK_PR;
1290
1291         l_lock(&res->lr_namespace->ns_lock);
1292
1293         res->lr_tmp = &rpc_list;
1294         rc = policy(lock, &tmpflags, 0, &err);
1295         res->lr_tmp = NULL;
1296
1297         /* FIXME: we should change the policy function slightly, to not make
1298          * this list at all, since we just turn around and free it */
1299         while (!list_empty(&rpc_list)) {
1300                 struct ldlm_ast_work *w =
1301                         list_entry(rpc_list.next, struct ldlm_ast_work, w_list);
1302                 list_del(&w->w_list);
1303                 LDLM_LOCK_PUT(w->w_lock);
1304                 OBD_FREE(w, sizeof(*w));
1305         }
1306
1307         if (rc == LDLM_ITER_CONTINUE) {
1308                 /* The lock met with no resistance; we're finished. */
1309                 l_unlock(&res->lr_namespace->ns_lock);
1310                 RETURN(ELDLM_LOCK_REPLACED);
1311         }
1312
1313         /* Do not grant any lock, but instead send GL callbacks.  The extent
1314          * policy nicely created a list of all PW locks for us.  We will choose
1315          * the highest of those which are larger than the size in the LVB, if
1316          * any, and perform a glimpse callback. */
1317         down(&res->lr_lvb_sem);
1318         res_lvb = res->lr_lvb_data;
1319         LASSERT(res_lvb != NULL);
1320         *reply_lvb = *res_lvb;
1321         up(&res->lr_lvb_sem);
1322
1323         list_for_each(tmp, &res->lr_granted) {
1324                 struct ldlm_lock *tmplock =
1325                         list_entry(tmp, struct ldlm_lock, l_res_link);
1326
1327                 if (tmplock->l_granted_mode == LCK_PR)
1328                         continue;
1329
1330                 if (tmplock->l_policy_data.l_extent.end <= reply_lvb->lvb_size)
1331                         continue;
1332
1333                 if (l == NULL) {
1334                         l = LDLM_LOCK_GET(tmplock);
1335                         continue;
1336                 }
1337
1338                 if (l->l_policy_data.l_extent.start >
1339                     tmplock->l_policy_data.l_extent.start)
1340                         continue;
1341
1342                 LDLM_LOCK_PUT(l);
1343                 l = LDLM_LOCK_GET(tmplock);
1344         }
1345         l_unlock(&res->lr_namespace->ns_lock);
1346
1347         /* There were no PW locks beyond the size in the LVB; finished. */
1348         if (l == NULL)
1349                 RETURN(ELDLM_LOCK_ABORTED);
1350
1351         if (l->l_glimpse_ast == NULL) {
1352                 /* We are racing with unlink(); just return -ENOENT */
1353                 rep->lock_policy_res1 = -ENOENT;
1354                 goto out;
1355         }
1356
1357         LASSERTF(l->l_glimpse_ast != NULL, "l == %p", l);
1358
1359         rc = l->l_glimpse_ast(l, NULL); /* this will update the LVB */
1360         if (rc != 0 && res->lr_namespace->ns_lvbo &&
1361             res->lr_namespace->ns_lvbo->lvbo_update) {
1362                 res->lr_namespace->ns_lvbo->lvbo_update(res, NULL, 0, 1);
1363         }
1364
1365         down(&res->lr_lvb_sem);
1366         *reply_lvb = *res_lvb;
1367         up(&res->lr_lvb_sem);
1368 out:
1369         LDLM_LOCK_PUT(l);
1370
1371         RETURN(ELDLM_LOCK_ABORTED);
1372 }
1373
1374 static int filter_post_fs_cleanup(struct obd_device *obd)
1375 {
1376         int rc = 0;
1377
1378         rc = fsfilt_post_cleanup(obd);
1379
1380         RETURN(rc);
1381 }
1382
1383 static int filter_group_set_fs_flags(struct obd_device *obd, int group)
1384 {
1385         struct filter_obd *filter = &obd->u.filter;
1386         int rc = 0, i = 0;
1387         ENTRY;
1388
1389         /* zero group is not longer valid. */
1390         if (group== 0)
1391                 RETURN(rc);
1392         for (i = 0; i < filter->fo_subdir_count; i++) {
1393                 struct dentry *dentry;
1394                 dentry = (filter->fo_subdirs + group)->dentry[i];
1395                 rc = fsfilt_set_fs_flags(obd, dentry->d_inode,
1396                                          SM_DO_REC | SM_DO_COW);
1397                 if (rc)
1398                         RETURN(rc);
1399         }
1400         RETURN(rc);
1401 }
1402 static int filter_post_fs_setup(struct obd_device *obd)
1403 {
1404         struct filter_obd *filter = &obd->u.filter;
1405         int rc = 0, j = 0;
1406         struct llog_ctxt *ctxt = NULL;
1407
1408         rc = fsfilt_post_setup(obd, filter->fo_dentry_O);
1409         if (rc)
1410                 RETURN(rc);
1411
1412         for (j = 0; j < filter->fo_group_count; j++) {
1413                 rc = filter_group_set_fs_flags(obd, j);
1414                 if (rc)
1415                         return rc;
1416         }
1417
1418         fsfilt_get_reint_log_ctxt(obd, filter->fo_sb, &ctxt);
1419         if (ctxt) {
1420                 ctxt->loc_obd = obd;
1421                 ctxt->loc_idx = LLOG_REINT_ORIG_CTXT;
1422                 obd->obd_llog_ctxt[LLOG_REINT_ORIG_CTXT] = ctxt;
1423         }
1424         fsfilt_set_ost_flags(obd, filter->fo_sb);
1425         return rc;
1426 }
1427
1428 /* mount the file system (secretly) */
1429 int filter_common_setup(struct obd_device *obd, obd_count len, void *buf,
1430                         char *option)
1431 {
1432         struct lustre_cfg* lcfg = buf;
1433         struct filter_obd *filter = &obd->u.filter;
1434         struct vfsmount *mnt;
1435         char ns_name[48];
1436         int rc = 0, i;
1437         ENTRY;
1438
1439         dev_clear_rdonly(2);
1440
1441         if (!lcfg->lcfg_inlbuf1 || !lcfg->lcfg_inlbuf2)
1442                 RETURN(-EINVAL);
1443
1444         obd->obd_fsops = fsfilt_get_ops(lcfg->lcfg_inlbuf2);
1445         if (IS_ERR(obd->obd_fsops))
1446                 RETURN(PTR_ERR(obd->obd_fsops));
1447
1448         mnt = do_kern_mount(lcfg->lcfg_inlbuf2, MS_NOATIME | MS_NODIRATIME,
1449                             lcfg->lcfg_inlbuf1, option);
1450         rc = PTR_ERR(mnt);
1451         if (IS_ERR(mnt))
1452                 GOTO(err_ops, rc);
1453
1454         if (lcfg->lcfg_inllen3 > 0 && lcfg->lcfg_inlbuf3) {
1455                 if (*lcfg->lcfg_inlbuf3 == 'f') {
1456                         obd->obd_replayable = 1;
1457                         obd_sync_filter = 1;
1458                         CWARN("%s: recovery enabled\n", obd->obd_name);
1459                 } else {
1460                         if (*lcfg->lcfg_inlbuf3 != 'n') {
1461                                 CERROR("unrecognised flag '%c'\n",
1462                                        *lcfg->lcfg_inlbuf3);
1463                         }
1464                         // XXX Robert? Why do we get errors here
1465                         // GOTO(err_mntput, rc = -EINVAL);
1466                 }
1467         }
1468
1469         filter->fo_vfsmnt = mnt;
1470         filter->fo_sb = mnt->mnt_sb;
1471         filter->fo_fstype = mnt->mnt_sb->s_type->name;
1472         CDEBUG(D_SUPER, "%s: mnt = %p\n", filter->fo_fstype, mnt);
1473
1474         OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1475         obd->obd_lvfs_ctxt.pwdmnt = mnt;
1476         obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
1477         obd->obd_lvfs_ctxt.fs = get_ds();
1478         obd->obd_lvfs_ctxt.cb_ops = filter_lvfs_ops;
1479
1480         rc = fsfilt_setup(obd, mnt->mnt_sb);
1481         if (rc)
1482                 GOTO(err_mntput, rc);
1483
1484         sema_init(&filter->fo_init_lock, 1);
1485         filter->fo_committed_group = 0;
1486         rc = filter_prep(obd);
1487         if (rc)
1488                 GOTO(err_mntput, rc);
1489
1490         filter->fo_destroys_in_progress = 0;
1491         for (i = 0; i < 32; i++)
1492                 sema_init(&filter->fo_create_locks[i], 1);
1493
1494         spin_lock_init(&filter->fo_translock);
1495         spin_lock_init(&filter->fo_objidlock);
1496         INIT_LIST_HEAD(&filter->fo_export_list);
1497         sema_init(&filter->fo_alloc_lock, 1);
1498         spin_lock_init(&filter->fo_r_pages.oh_lock);
1499         spin_lock_init(&filter->fo_w_pages.oh_lock);
1500         spin_lock_init(&filter->fo_r_discont_pages.oh_lock);
1501         spin_lock_init(&filter->fo_w_discont_pages.oh_lock);
1502         spin_lock_init(&filter->fo_r_discont_blocks.oh_lock);
1503         spin_lock_init(&filter->fo_w_discont_blocks.oh_lock);
1504         filter->fo_readcache_max_filesize = FILTER_MAX_CACHE_SIZE;
1505
1506         INIT_LIST_HEAD(&filter->fo_llog_list);
1507         spin_lock_init(&filter->fo_llog_list_lock);
1508
1509         sprintf(ns_name, "filter-%s", obd->obd_uuid.uuid);
1510         obd->obd_namespace = ldlm_namespace_new(ns_name, LDLM_NAMESPACE_SERVER);
1511
1512         if (obd->obd_namespace == NULL)
1513                 GOTO(err_post, rc = -ENOMEM);
1514         obd->obd_namespace->ns_lvbp = obd;
1515         obd->obd_namespace->ns_lvbo = &filter_lvbo;
1516         ldlm_register_intent(obd->obd_namespace, filter_intent_policy);
1517
1518         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1519                            "filter_ldlm_cb_client", &obd->obd_ldlm_client);
1520
1521         rc = obd_llog_cat_initialize(obd, &obd->obd_llogs, 1, CATLIST);
1522         if (rc) {
1523                 CERROR("failed to setup llogging subsystems\n");
1524                 GOTO(err_post, rc);
1525         }
1526         RETURN(0);
1527
1528 err_post:
1529         filter_post(obd);
1530 err_mntput:
1531         unlock_kernel();
1532         mntput(mnt);
1533         filter->fo_sb = 0;
1534         lock_kernel();
1535 err_ops:
1536         fsfilt_put_ops(obd->obd_fsops);
1537         return rc;
1538 }
1539
1540 static int filter_attach(struct obd_device *obd, obd_count len, void *data)
1541 {
1542         struct lprocfs_static_vars lvars;
1543         int rc;
1544
1545         lprocfs_init_vars(filter, &lvars);
1546         rc = lprocfs_obd_attach(obd, lvars.obd_vars);
1547         if (rc != 0)
1548                 return rc;
1549
1550         rc = lprocfs_alloc_obd_stats(obd, LPROC_FILTER_LAST);
1551         if (rc != 0)
1552                 return rc;
1553
1554         /* Init obdfilter private stats here */
1555         lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_READ_BYTES,
1556                              LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes");
1557         lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_WRITE_BYTES,
1558                              LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes");
1559
1560         return lproc_filter_attach_seqstat(obd);
1561 }
1562
1563 static int filter_detach(struct obd_device *dev)
1564 {
1565         lprocfs_free_obd_stats(dev);
1566         return lprocfs_obd_detach(dev);
1567 }
1568
1569 static int filter_setup(struct obd_device *obd, obd_count len, void *buf)
1570 {
1571         struct lustre_cfg* lcfg = buf;
1572         int rc;
1573         ENTRY;
1574         /* all mount options including errors=remount-ro and asyncdel are passed
1575          * using 4th lcfg param. And it is good, finally we have got rid of
1576          * hardcoded fs types in the code. */
1577         rc = filter_common_setup(obd, len, buf, lcfg->lcfg_inlbuf4);
1578         if (rc)
1579                 RETURN(rc);
1580         rc = filter_post_fs_setup(obd);
1581         RETURN(rc);
1582 }
1583
1584 static int filter_cleanup(struct obd_device *obd, int flags)
1585 {
1586         struct filter_obd *filter = &obd->u.filter;
1587         ENTRY;
1588
1589         if (flags & OBD_OPT_FAILOVER)
1590                 CERROR("%s: shutting down for failover; client state will"
1591                        " be preserved.\n", obd->obd_name);
1592
1593         if (!list_empty(&obd->obd_exports)) {
1594                 CERROR("%s: still has clients!\n", obd->obd_name);
1595                 class_disconnect_exports(obd, flags);
1596                 if (!list_empty(&obd->obd_exports)) {
1597                         CERROR("still has exports after forced cleanup?\n");
1598                         RETURN(-EBUSY);
1599                 }
1600         }
1601
1602         target_cleanup_recovery(obd);
1603
1604         ldlm_namespace_free(obd->obd_namespace, flags & OBD_OPT_FORCE);
1605
1606         if (filter->fo_sb == NULL)
1607                 RETURN(0);
1608
1609         filter_post_fs_cleanup(obd);
1610         filter_post(obd);
1611
1612         shrink_dcache_parent(filter->fo_sb->s_root);
1613         filter->fo_sb = 0;
1614
1615         if (atomic_read(&filter->fo_vfsmnt->mnt_count) > 1)
1616                 CERROR("%s: mount point %p busy, mnt_count: %d\n",
1617                        obd->obd_name, filter->fo_vfsmnt,
1618                        atomic_read(&filter->fo_vfsmnt->mnt_count));
1619
1620         unlock_kernel();
1621         mntput(filter->fo_vfsmnt);
1622         //destroy_buffers(filter->fo_sb->s_dev);
1623         filter->fo_sb = NULL;
1624         fsfilt_put_ops(obd->obd_fsops);
1625         lock_kernel();
1626
1627         dev_clear_rdonly(2);
1628
1629         RETURN(0);
1630 }
1631
1632 static int filter_connect_post(struct obd_export *exp, unsigned long connect_flags)
1633 {
1634         struct obd_device *obd = exp->exp_obd;
1635         struct filter_export_data *fed;
1636         char str[PTL_NALFMT_SIZE];
1637         struct obd_llogs *llog;
1638         struct llog_ctxt *ctxt;
1639         int rc;
1640         ENTRY;
1641
1642         fed = &exp->exp_filter_data;
1643         if (fed->fed_group < FILTER_MIN_GROUPS)
1644                 RETURN(0);
1645
1646         /* initialize llogs for connections from MDS */
1647         llog = filter_grab_llog_for_group(obd, fed->fed_group, exp);
1648         LASSERT(llog != NULL);
1649
1650         ctxt = llog_get_context(llog, LLOG_UNLINK_REPL_CTXT);
1651         LASSERT(ctxt != NULL);
1652
1653         rc = llog_receptor_accept(ctxt, exp->exp_imp_reverse);
1654         portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
1655                         exp->exp_connection->c_peer.peer_id.nid, str);
1656         CDEBUG(D_OTHER, "%s: init llog ctxt for export "LPX64"/%s, group %d\n",
1657                obd->obd_name, exp->exp_connection->c_peer.peer_id.nid,
1658                str, fed->fed_group);
1659
1660         RETURN(rc);
1661 }
1662
1663 /* nearly identical to mds_connect */
1664 static int filter_connect(struct lustre_handle *conn, struct obd_device *obd,
1665                           struct obd_uuid *cluuid, unsigned long connect_flags)
1666 {
1667         struct obd_export *exp;
1668         struct filter_export_data *fed;
1669         struct filter_client_data *fcd = NULL;
1670         struct filter_obd *filter = &obd->u.filter;
1671         struct lvfs_run_ctxt saved;
1672         int rc, group;
1673         ENTRY;
1674
1675         if (conn == NULL || obd == NULL || cluuid == NULL)
1676                 RETURN(-EINVAL);
1677
1678         rc = class_connect(conn, obd, cluuid);
1679         if (rc)
1680                 RETURN(rc);
1681         exp = class_conn2export(conn);
1682         LASSERT(exp != NULL);
1683
1684         fed = &exp->exp_filter_data;
1685
1686         spin_lock_init(&fed->fed_lock);
1687
1688        /* connection from MDS */
1689         group = connect_flags;
1690         if (obd->obd_replayable) {
1691                 OBD_ALLOC(fcd, sizeof(*fcd));
1692                 if (!fcd) {
1693                         CERROR("filter: out of memory for client data\n");
1694                         GOTO(cleanup, rc = -ENOMEM);
1695                 }
1696
1697                 memcpy(fcd->fcd_uuid, cluuid, sizeof(fcd->fcd_uuid));
1698                 fed->fed_fcd = fcd;
1699                 fed->fed_fcd->fcd_group = group;
1700                 rc = filter_client_add(obd, filter, fed, -1);
1701                 if (rc)
1702                         GOTO(cleanup, rc);
1703         }
1704         CWARN("%s: Received MDS connection ("LPX64"); group %d\n",
1705               obd->obd_name, exp->exp_handle.h_cookie, group);
1706
1707         if (group == 0)
1708                 GOTO(cleanup, rc);
1709
1710         if (fed->fed_group != 0 && fed->fed_group != group) {
1711                 char str[PTL_NALFMT_SIZE];
1712                 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
1713                                 exp->exp_connection->c_peer.peer_id.nid, str);
1714                 CERROR("!!! This export (nid "LPX64"/%s) used object group %d "
1715                        "earlier; now it's trying to use group %d!  This could "
1716                        "be a bug in the MDS.  Tell CFS.\n",
1717                        exp->exp_connection->c_peer.peer_id.nid, str,
1718                        fed->fed_group, group);
1719                 GOTO(cleanup, rc = -EPROTO);
1720         }
1721         fed->fed_group = group;
1722
1723         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1724         rc = filter_read_groups(obd, group, 1);
1725         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1726         if (rc != 0) {
1727                 CERROR("can't read group %u\n", group);
1728                 GOTO(cleanup, rc);
1729         }
1730         rc = filter_group_set_fs_flags(obd, group);
1731         if (rc != 0) {
1732                 CERROR("can't set kml flags %u\n", group);
1733                 GOTO(cleanup, rc);
1734         }
1735
1736 cleanup:
1737         if (rc) {
1738                 if (fcd)
1739                         OBD_FREE(fcd, sizeof(*fcd));
1740                 class_disconnect(exp, 0);
1741         } else {
1742                 class_export_put(exp);
1743         }
1744         return rc;
1745 }
1746
1747 static int filter_precleanup(struct obd_device *obd, int flags)
1748 {
1749         struct filter_group_llog *log;
1750         struct filter_obd *filter;
1751         int rc = 0;
1752         ENTRY;
1753
1754         filter = &obd->u.filter;
1755
1756         spin_lock(&filter->fo_llog_list_lock);
1757         while (!list_empty(&filter->fo_llog_list)) {
1758                 log = list_entry(filter->fo_llog_list.next,
1759                                  struct filter_group_llog, list);
1760                 list_del(&log->list);
1761                 spin_unlock(&filter->fo_llog_list_lock);
1762
1763                 rc = obd_llog_finish(obd, log->llogs, 0);
1764                 if (rc)
1765                         CERROR("failed to cleanup llogging subsystem for %u\n",
1766                                 log->group);
1767                 OBD_FREE(log->llogs, sizeof(*(log->llogs)));
1768                 OBD_FREE(log, sizeof(*log));
1769                 spin_lock(&filter->fo_llog_list_lock);
1770         }
1771         spin_unlock(&filter->fo_llog_list_lock);
1772
1773         rc = obd_llog_finish(obd, &obd->obd_llogs, 0);
1774         if (rc)
1775                 CERROR("failed to cleanup llogging subsystem\n");
1776
1777         RETURN(rc);
1778 }
1779
1780 /* Do extra sanity checks for grant accounting.  We do this at connect,
1781  * disconnect, and statfs RPC time, so it shouldn't be too bad.  We can
1782  * always get rid of it or turn it off when we know accounting is good. */
1783 static void filter_grant_sanity_check(struct obd_device *obd, const char *func)
1784 {
1785         struct filter_export_data *fed;
1786         struct obd_export *exp;
1787         obd_size maxsize = obd->obd_osfs.os_blocks * obd->obd_osfs.os_bsize;
1788         obd_size tot_dirty = 0, tot_pending = 0, tot_granted = 0;
1789         obd_size fo_tot_dirty, fo_tot_pending, fo_tot_granted;
1790
1791         if (list_empty(&obd->obd_exports))
1792                 return;
1793
1794         spin_lock(&obd->obd_osfs_lock);
1795         spin_lock(&obd->obd_dev_lock);
1796         list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
1797                 fed = &exp->exp_filter_data;
1798                 LASSERTF(fed->fed_grant + fed->fed_pending <= maxsize,
1799                          "cli %s/%p %lu+%lu > "LPU64"\n",
1800                          exp->exp_client_uuid.uuid, exp,
1801                          fed->fed_grant, fed->fed_pending, maxsize);
1802                 LASSERTF(fed->fed_dirty <= maxsize, "cli %s/%p %lu > "LPU64"\n",
1803                          exp->exp_client_uuid.uuid, exp,fed->fed_dirty,maxsize);
1804                 CDEBUG(D_CACHE,"%s: cli %s/%p dirty %lu pend %lu grant %lu\n",
1805                        obd->obd_name, exp->exp_client_uuid.uuid, exp,
1806                        fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1807                 tot_granted += fed->fed_grant + fed->fed_pending;
1808                 tot_pending += fed->fed_pending;
1809                 tot_dirty += fed->fed_dirty;
1810         }
1811         fo_tot_granted = obd->u.filter.fo_tot_granted;
1812         fo_tot_pending = obd->u.filter.fo_tot_pending;
1813         fo_tot_dirty = obd->u.filter.fo_tot_dirty;
1814         spin_unlock(&obd->obd_dev_lock);
1815         spin_unlock(&obd->obd_osfs_lock);
1816
1817         /* Do these assertions outside the spinlocks so we don't kill system */
1818         if (tot_granted != fo_tot_granted)
1819                 CERROR("%s: tot_granted "LPU64" != fo_tot_granted "LPU64"\n",
1820                        func, tot_granted, fo_tot_granted);
1821         if (tot_pending != fo_tot_pending)
1822                 CERROR("%s: tot_pending "LPU64" != fo_tot_pending "LPU64"\n",
1823                        func, tot_pending, fo_tot_pending);
1824         if (tot_dirty != fo_tot_dirty)
1825                 CERROR("%s: tot_dirty "LPU64" != fo_tot_dirty "LPU64"\n",
1826                        func, tot_dirty, fo_tot_dirty);
1827         if (tot_pending > tot_granted)
1828                 CERROR("%s: tot_pending "LPU64" > tot_granted "LPU64"\n",
1829                        func, tot_pending, tot_granted);
1830         if (tot_granted > maxsize)
1831                 CERROR("%s: tot_granted "LPU64" > maxsize "LPU64"\n",
1832                        func, tot_granted, maxsize);
1833         if (tot_dirty > maxsize)
1834                 CERROR("%s: tot_dirty "LPU64" > maxsize "LPU64"\n",
1835                        func, tot_dirty, maxsize);
1836 }
1837
1838 /* Remove this client from the grant accounting totals.  We also remove
1839  * the export from the obd device under the osfs and dev locks to ensure
1840  * that the filter_grant_sanity_check() calculations are always valid.
1841  * The client should do something similar when it invalidates its import. */
1842 static void filter_grant_discard(struct obd_export *exp)
1843 {
1844         struct obd_device *obd = exp->exp_obd;
1845         struct filter_obd *filter = &obd->u.filter;
1846         struct filter_export_data *fed = &exp->exp_filter_data;
1847
1848         spin_lock(&obd->obd_osfs_lock);
1849         spin_lock(&exp->exp_obd->obd_dev_lock);
1850         list_del_init(&exp->exp_obd_chain);
1851         spin_unlock(&exp->exp_obd->obd_dev_lock);
1852
1853         CDEBUG(D_CACHE, "%s: cli %s/%p dirty %lu pend %lu grant %lu\n",
1854                obd->obd_name, exp->exp_client_uuid.uuid, exp,
1855                fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1856
1857         LASSERTF(filter->fo_tot_granted >= fed->fed_grant,
1858                  "%s: tot_granted "LPU64" cli %s/%p fed_grant %lu\n",
1859                  obd->obd_name, filter->fo_tot_granted,
1860                  exp->exp_client_uuid.uuid, exp, fed->fed_grant);
1861         filter->fo_tot_granted -= fed->fed_grant;
1862         LASSERTF(exp->exp_obd->u.filter.fo_tot_pending >= fed->fed_pending,
1863                  "%s: tot_pending "LPU64" cli %s/%p fed_pending %lu\n",
1864                  obd->obd_name, filter->fo_tot_pending,
1865                  exp->exp_client_uuid.uuid, exp, fed->fed_pending);
1866         LASSERTF(filter->fo_tot_dirty >= fed->fed_dirty,
1867                  "%s: tot_dirty "LPU64" cli %s/%p fed_dirty %lu\n",
1868                  obd->obd_name, filter->fo_tot_dirty,
1869                  exp->exp_client_uuid.uuid, exp, fed->fed_dirty);
1870         filter->fo_tot_dirty -= fed->fed_dirty;
1871         fed->fed_dirty = 0;
1872         fed->fed_grant = 0;
1873
1874         spin_unlock(&obd->obd_osfs_lock);
1875 }
1876
1877 static int filter_destroy_export(struct obd_export *exp)
1878 {
1879         ENTRY;
1880
1881         if (exp->exp_filter_data.fed_pending)
1882                 CERROR("%s: cli %s/%p has %lu pending on destroyed export\n",
1883                        exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
1884                        exp, exp->exp_filter_data.fed_pending);
1885
1886         target_destroy_export(exp);
1887
1888         if (exp->exp_obd->obd_replayable)
1889                 filter_client_free(exp, exp->exp_flags);
1890
1891         filter_grant_discard(exp);
1892         if (!(exp->exp_flags & OBD_OPT_FORCE))
1893                 filter_grant_sanity_check(exp->exp_obd, __FUNCTION__);
1894
1895         RETURN(0);
1896 }
1897
1898 static void filter_sync_llogs(struct obd_device *obd, struct obd_export *dexp)
1899 {
1900         struct filter_group_llog *fglog, *nlog;
1901         struct filter_obd *filter;
1902         int worked = 0, group;
1903         struct llog_ctxt *ctxt;
1904         ENTRY;
1905
1906         filter = &obd->u.filter;
1907
1908         /* we can't sync log holding spinlock. also, we do not want to get
1909          * into livelock. so we do following: loop over MDS's exports in
1910          * group order and skip already synced llogs -bzzz */
1911         do {
1912                 /* look for group with min. number, but > worked */
1913                 fglog = NULL;
1914                 group = 1 << 30;
1915                 spin_lock(&filter->fo_llog_list_lock);
1916                 list_for_each_entry(nlog, &filter->fo_llog_list, list) {
1917
1918                         if (nlog->group <= worked) {
1919                                 /* this group is already synced */
1920                                 continue;
1921                         }
1922
1923                         if (group < nlog->group) {
1924                                 /* we have group with smaller number to sync */
1925                                 continue;
1926                         }
1927
1928                         /* store current minimal group */
1929                         fglog = nlog;
1930                         group = nlog->group;
1931                 }
1932                 spin_unlock(&filter->fo_llog_list_lock);
1933
1934                 if (fglog == NULL)
1935                         break;
1936
1937                 worked = fglog->group;
1938                 if (fglog->exp && (dexp == fglog->exp || dexp == NULL)) {
1939                         ctxt = llog_get_context(fglog->llogs,
1940                                         LLOG_UNLINK_REPL_CTXT);
1941                         LASSERT(ctxt != NULL);
1942                         llog_sync(ctxt, fglog->exp);
1943                 }
1944         } while (fglog != NULL);
1945 }
1946
1947 /* also incredibly similar to mds_disconnect */
1948 static int filter_disconnect(struct obd_export *exp, unsigned long flags)
1949 {
1950         struct obd_device *obd = exp->exp_obd;
1951         unsigned long irqflags;
1952         int rc;
1953         ENTRY;
1954
1955         LASSERT(exp);
1956         class_export_get(exp);
1957
1958         spin_lock_irqsave(&exp->exp_lock, irqflags);
1959         exp->exp_flags = flags;
1960         spin_unlock_irqrestore(&exp->exp_lock, irqflags);
1961
1962         if (!(flags & OBD_OPT_FORCE))
1963                 filter_grant_sanity_check(obd, __FUNCTION__);
1964         filter_grant_discard(exp);
1965
1966         /* Disconnect early so that clients can't keep using export */
1967         rc = class_disconnect(exp, flags);
1968
1969         ldlm_cancel_locks_for_export(exp);
1970
1971         fsfilt_sync(obd, obd->u.filter.fo_sb);
1972
1973         /* flush any remaining cancel messages out to the target */
1974         filter_sync_llogs(obd, exp);
1975
1976         class_export_put(exp);
1977         RETURN(rc);
1978 }
1979
1980 struct dentry *__filter_oa2dentry(struct obd_device *obd,
1981                                   struct obdo *oa, const char *what)
1982 {
1983         struct dentry *dchild = NULL;
1984         obd_gr group = 0;
1985
1986         if (oa->o_valid & OBD_MD_FLGROUP)
1987                 group = oa->o_gr;
1988
1989         dchild = filter_id2dentry(obd, NULL, group, oa->o_id);
1990
1991         if (IS_ERR(dchild)) {
1992                 CERROR("%s error looking up object: "LPU64"\n", what, oa->o_id);
1993                 RETURN(dchild);
1994         }
1995
1996         if (dchild->d_inode == NULL) {
1997                 CERROR("%s: %s on non-existent object: "LPU64"\n",
1998                        obd->obd_name, what, oa->o_id);
1999                 f_dput(dchild);
2000                 RETURN(ERR_PTR(-ENOENT));
2001         }
2002
2003         return dchild;
2004 }
2005
2006 static int filter_getattr(struct obd_export *exp, struct obdo *oa,
2007                           struct lov_stripe_md *md)
2008 {
2009         struct dentry *dentry = NULL;
2010         struct obd_device *obd;
2011         int rc = 0;
2012         ENTRY;
2013
2014         obd = class_exp2obd(exp);
2015         if (obd == NULL) {
2016                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2017                        exp->exp_handle.h_cookie);
2018                 RETURN(-EINVAL);
2019         }
2020
2021         dentry = filter_oa2dentry(obd, oa);
2022         if (IS_ERR(dentry))
2023                 RETURN(PTR_ERR(dentry));
2024
2025         /* Limit the valid bits in the return data to what we actually use */
2026         oa->o_valid = OBD_MD_FLID;
2027         obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2028
2029         f_dput(dentry);
2030         RETURN(rc);
2031 }
2032
2033 /* this is called from filter_truncate() until we have filter_punch() */
2034 static int filter_setattr(struct obd_export *exp, struct obdo *oa,
2035                           struct lov_stripe_md *md, struct obd_trans_info *oti)
2036 {
2037         struct lvfs_run_ctxt saved;
2038         struct filter_obd *filter;
2039         struct dentry *dentry;
2040         struct iattr iattr;
2041         struct ldlm_res_id res_id = { .name = { oa->o_id, 0, oa->o_gr, 0 } };
2042         struct ldlm_resource *res;
2043         void *handle;
2044         int rc, rc2;
2045         ENTRY;
2046
2047         LASSERT(oti != NULL);
2048
2049         dentry = filter_oa2dentry(exp->exp_obd, oa);
2050         if (IS_ERR(dentry))
2051                 RETURN(PTR_ERR(dentry));
2052
2053         filter = &exp->exp_obd->u.filter;
2054
2055         iattr_from_obdo(&iattr, oa, oa->o_valid);
2056
2057         push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2058         lock_kernel();
2059
2060         if (iattr.ia_valid & ATTR_SIZE)
2061                 down(&dentry->d_inode->i_sem);
2062         handle = fsfilt_start(exp->exp_obd, dentry->d_inode, FSFILT_OP_SETATTR,
2063                               oti);
2064         if (IS_ERR(handle))
2065                 GOTO(out_unlock, rc = PTR_ERR(handle));
2066
2067         /* XXX this could be a rwsem instead, if filter_preprw played along */
2068         if (iattr.ia_valid & ATTR_ATTR_FLAG)
2069                 rc = fsfilt_iocontrol(exp->exp_obd, dentry->d_inode, NULL,
2070                                       EXT3_IOC_SETFLAGS,
2071                                       (long)&iattr.ia_attr_flags);
2072         else
2073                 rc = fsfilt_setattr(exp->exp_obd, dentry, handle, &iattr, 1);
2074         rc = filter_finish_transno(exp, oti, rc);
2075         rc2 = fsfilt_commit(exp->exp_obd, filter->fo_sb, dentry->d_inode, handle, 0);
2076         if (rc2) {
2077                 CERROR("error on commit, err = %d\n", rc2);
2078                 if (!rc)
2079                         rc = rc2;
2080         }
2081
2082         if (iattr.ia_valid & ATTR_SIZE) {
2083                 res = ldlm_resource_get(exp->exp_obd->obd_namespace, NULL,
2084                                         res_id, LDLM_EXTENT, 0);
2085                 if (res == NULL) {
2086                         CERROR("!!! resource_get failed for object "LPU64" -- "
2087                                "filter_setattr with no lock?\n", oa->o_id);
2088                 } else {
2089                         if (res->lr_namespace->ns_lvbo &&
2090                             res->lr_namespace->ns_lvbo->lvbo_update) {
2091                                 rc = res->lr_namespace->ns_lvbo->lvbo_update
2092                                         (res, NULL, 0, 0);
2093                         }
2094                         ldlm_resource_putref(res);
2095                 }
2096         }
2097
2098         oa->o_valid = OBD_MD_FLID;
2099         obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2100
2101 out_unlock:
2102         if (iattr.ia_valid & ATTR_SIZE)
2103                 up(&dentry->d_inode->i_sem);
2104         unlock_kernel();
2105         pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2106
2107         f_dput(dentry);
2108         RETURN(rc);
2109 }
2110
2111 /* XXX identical to osc_unpackmd */
2112 static int filter_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2113                            struct lov_mds_md *lmm, int lmm_bytes)
2114 {
2115         int lsm_size;
2116         ENTRY;
2117
2118         if (lmm != NULL) {
2119                 if (lmm_bytes < sizeof (*lmm)) {
2120                         CERROR("lov_mds_md too small: %d, need %d\n",
2121                                lmm_bytes, (int)sizeof(*lmm));
2122                         RETURN(-EINVAL);
2123                 }
2124                 /* XXX LOV_MAGIC etc check? */
2125
2126                 if (lmm->lmm_object_id == cpu_to_le64(0)) {
2127                         CERROR("lov_mds_md: zero lmm_object_id\n");
2128                         RETURN(-EINVAL);
2129                 }
2130         }
2131
2132         lsm_size = lov_stripe_md_size(1);
2133         if (lsmp == NULL)
2134                 RETURN(lsm_size);
2135
2136         if (*lsmp != NULL && lmm == NULL) {
2137                 OBD_FREE(*lsmp, lsm_size);
2138                 *lsmp = NULL;
2139                 RETURN(0);
2140         }
2141
2142         if (*lsmp == NULL) {
2143                 OBD_ALLOC(*lsmp, lsm_size);
2144                 if (*lsmp == NULL)
2145                         RETURN(-ENOMEM);
2146
2147                 loi_init((*lsmp)->lsm_oinfo);
2148         }
2149
2150         if (lmm != NULL) {
2151                 /* XXX zero *lsmp? */
2152                 (*lsmp)->lsm_object_id = le64_to_cpu (lmm->lmm_object_id);
2153                 LASSERT((*lsmp)->lsm_object_id);
2154         }
2155
2156         (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
2157
2158         RETURN(lsm_size);
2159 }
2160
2161 static void filter_destroy_precreated(struct obd_export *exp, struct obdo *oa,
2162                                       struct filter_obd *filter)
2163 {
2164         struct obdo *doa = NULL;
2165         __u64 last, id;
2166         ENTRY;
2167
2168         LASSERT(oa);
2169         LASSERT(oa->o_gr != 0);
2170         LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2171
2172         OBD_ALLOC(doa, sizeof(*doa));
2173         if (doa == NULL) {
2174                 CERROR("cannot allocate doa, error %d\n",
2175                        -ENOMEM);
2176                 EXIT;
2177                 return;
2178         }
2179
2180         memset(doa, 0, sizeof(*doa));
2181         doa->o_mode = S_IFREG;
2182         doa->o_gr = oa->o_gr;
2183         doa->o_valid = oa->o_valid & (OBD_MD_FLGROUP | OBD_MD_FLID);
2184
2185         set_bit(doa->o_gr, &filter->fo_destroys_in_progress);
2186         down(&filter->fo_create_locks[doa->o_gr]);
2187         if (!test_bit(doa->o_gr, &filter->fo_destroys_in_progress)) {
2188                 CERROR("%s:["LPU64"] destroy_in_progress already cleared\n",
2189                        exp->exp_obd->obd_name, doa->o_gr);
2190                 up(&filter->fo_create_locks[doa->o_gr]);
2191                 GOTO(out_free_doa, 0);
2192         }
2193
2194         last = filter_last_id(filter, doa->o_gr);
2195         CWARN("%s:["LPU64"] deleting orphan objects from "LPU64" to "LPU64"\n",
2196               exp->exp_obd->obd_name, doa->o_gr, oa->o_id + 1, last);
2197         for (id = oa->o_id + 1; id <= last; id++) {
2198                 doa->o_id = id;
2199                 filter_destroy(exp, doa, NULL, NULL);
2200         }
2201
2202         CDEBUG(D_HA, "%s:["LPU64"] after destroy: set last_objids = "LPU64"\n",
2203                exp->exp_obd->obd_name, doa->o_gr, oa->o_id);
2204
2205         filter_set_last_id(filter, doa->o_gr, oa->o_id);
2206
2207         clear_bit(doa->o_gr, &filter->fo_destroys_in_progress);
2208         up(&filter->fo_create_locks[doa->o_gr]);
2209
2210         EXIT;
2211 out_free_doa:
2212         OBD_FREE(doa, sizeof(*doa));
2213 }
2214
2215 /* returns a negative error or a nonnegative number of files to create */
2216 static int filter_should_precreate(struct obd_export *exp, struct obdo *oa,
2217                                    obd_gr group)
2218 {
2219         struct obd_device *obd = exp->exp_obd;
2220         struct filter_obd *filter = &obd->u.filter;
2221         int diff, rc;
2222         ENTRY;
2223
2224         diff = oa->o_id - filter_last_id(filter, oa->o_gr);
2225         CDEBUG(D_INFO, "filter_last_id() = "LPU64" -> diff = %d\n",
2226                filter_last_id(filter, oa->o_gr), diff);
2227
2228         /* delete orphans request */
2229         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
2230             (oa->o_flags & OBD_FL_DELORPHAN)) {
2231                 if (diff >= 0)
2232                         RETURN(diff);
2233                 if (-diff > 10000) { /* XXX make this smarter */
2234                         CERROR("ignoring bogus orphan destroy request: obdid "
2235                                LPU64" last_id "LPU64"\n",
2236                                oa->o_id, filter_last_id(filter, oa->o_gr));
2237                         RETURN(-EINVAL);
2238                 }
2239                 filter_destroy_precreated(exp, oa, filter);
2240                 rc = filter_update_last_objid(obd, group, 0);
2241                 if (rc)
2242                         CERROR("unable to write lastobjid, but orphans"
2243                                "were deleted\n");
2244                 RETURN(0);
2245         } else {
2246                 /* only precreate if group == 0 and o_id is specfied */
2247                 if (!(oa->o_valid & OBD_FL_DELORPHAN) &&
2248                     (/*group != 0 ||*/ oa->o_id == 0))
2249                         RETURN(1);
2250
2251                 LASSERT(diff >= 0);
2252                 RETURN(diff);
2253         }
2254 }
2255 static int filter_precreate_rec(struct obd_device *obd, struct dentry *dentry,
2256                                 int *number, struct obdo *oa)
2257 {
2258         int rc;
2259         ENTRY;
2260
2261         rc = fsfilt_precreate_rec(obd, dentry, number, oa);
2262
2263         RETURN(rc);
2264 }
2265
2266 /* We rely on the fact that only one thread will be creating files in a given
2267  * group at a time, which is why we don't need an atomic filter_get_new_id.
2268  * Even if we had that atomic function, the following race would exist:
2269  *
2270  * thread 1: gets id x from filter_next_id
2271  * thread 2: gets id (x + 1) from filter_next_id
2272  * thread 2: creates object (x + 1)
2273  * thread 1: tries to create object x, gets -ENOSPC
2274  */
2275 static int filter_precreate(struct obd_device *obd, struct obdo *oa,
2276                             obd_gr group, int *num)
2277 {
2278         struct dentry *dchild = NULL, *dparent = NULL;
2279         int err = 0, rc = 0, recreate_obj = 0, i;
2280         struct filter_obd *filter;
2281         void *handle = NULL;
2282         void *lock = NULL;
2283         __u64 next_id;
2284         ENTRY;
2285
2286         filter = &obd->u.filter;
2287
2288         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
2289             (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
2290                 recreate_obj = 1;
2291         }
2292
2293         CDEBUG(D_HA, "%s: precreating %d objects\n", obd->obd_name, *num);
2294
2295         down(&filter->fo_create_locks[group]);
2296
2297         for (i = 0; i < *num && err == 0; i++) {
2298                 int cleanup_phase = 0;
2299
2300                 if (test_bit(group, &filter->fo_destroys_in_progress)) {
2301                         CWARN("%s: precreate aborted by destroy\n",
2302                               obd->obd_name);
2303                         break;
2304                 }
2305
2306                 if (recreate_obj) {
2307                         __u64 last_id;
2308                         next_id = oa->o_id;
2309                         last_id = filter_last_id(filter, group);
2310                         if (next_id > last_id) {
2311                                 CERROR("Error: Trying to recreate obj greater"
2312                                        "than last id "LPD64" > "LPD64"\n",
2313                                        next_id, last_id);
2314                                 GOTO(cleanup, rc = -EINVAL);
2315                         }
2316                 } else {
2317                         next_id = filter_last_id(filter, group) + 1;
2318                 }
2319
2320                 CDEBUG(D_INFO, "precreate objid "LPU64"\n", next_id);
2321
2322                 dparent = filter_parent_lock(obd, group, next_id, &lock);
2323                 if (IS_ERR(dparent))
2324                         GOTO(cleanup, rc = PTR_ERR(dparent));
2325                 cleanup_phase = 1;
2326
2327                 /*only do precreate rec record. so clean kml flags here*/
2328                 fsfilt_clear_fs_flags(obd, dparent->d_inode, SM_DO_REC);
2329
2330                 dchild = filter_id2dentry(obd, dparent, group, next_id);
2331                 if (IS_ERR(dchild))
2332                         GOTO(cleanup, rc = PTR_ERR(dchild));
2333                 cleanup_phase = 2;
2334
2335                 if (dchild->d_inode != NULL) {
2336                         /* This would only happen if lastobjid was bad on disk*/
2337                         /* Could also happen if recreating missing obj but
2338                          * already exists
2339                          */
2340                         if (recreate_obj) {
2341                                 CERROR("%s: Serious error: recreating obj %*s "
2342                                        "but obj already exists \n",
2343                                        obd->obd_name, dchild->d_name.len,
2344                                        dchild->d_name.name);
2345                                 LBUG();
2346                         } else {
2347                                 CERROR("%s: Serious error: objid %*s already "
2348                                        "exists; is this filesystem corrupt?\n",
2349                                        obd->obd_name, dchild->d_name.len,
2350                                        dchild->d_name.name);
2351                                 LBUG();
2352                         }
2353                         GOTO(cleanup, rc = -EEXIST);
2354                 }
2355
2356                 handle = fsfilt_start_log(obd, dparent->d_inode,
2357                                           FSFILT_OP_CREATE, NULL, 1);
2358                 if (IS_ERR(handle))
2359                         GOTO(cleanup, rc = PTR_ERR(handle));
2360                 cleanup_phase = 3;
2361
2362                 rc = ll_vfs_create(dparent->d_inode, dchild, S_IFREG, NULL);
2363                 if (rc) {
2364                         CERROR("create failed rc = %d\n", rc);
2365                         GOTO(cleanup, rc);
2366                 }
2367
2368                 if (!recreate_obj) {
2369                         filter_set_last_id(filter, group, next_id);
2370                         err = filter_update_last_objid(obd, group, 0);
2371                         if (err)
2372                                 CERROR("unable to write lastobjid "
2373                                        "but file created\n");
2374                 }
2375                 fsfilt_set_fs_flags(obd, dparent->d_inode, SM_DO_REC);
2376
2377         cleanup:
2378                 switch(cleanup_phase) {
2379                 case 3:
2380                         err = fsfilt_commit(obd, filter->fo_sb,
2381                                             dparent->d_inode, handle, 0);
2382                         if (err) {
2383                                 CERROR("error on commit, err = %d\n", err);
2384                                 if (!rc)
2385                                         rc = err;
2386                         }
2387                 case 2:
2388                         f_dput(dchild);
2389                 case 1:
2390                         filter_parent_unlock(dparent, lock);
2391                 case 0:
2392                         break;
2393                 }
2394
2395                 if (rc)
2396                         break;
2397         }
2398
2399         *num = i;
2400
2401         /* check if we have an error after ll_vfs_create(). It is possible that
2402          * there will be say -ENOSPC and we will leak it. */
2403         if (rc == 0)
2404                 rc = filter_precreate_rec(obd, dparent, num, oa);
2405
2406         up(&filter->fo_create_locks[group]);
2407
2408         CDEBUG(D_HA, "%s: server last_objid for group "LPU64": "LPU64"\n",
2409                obd->obd_name, group, filter->fo_last_objids[group]);
2410
2411         CDEBUG(D_HA, "%s: filter_precreate() created %d objects\n",
2412                obd->obd_name, i);
2413
2414         RETURN(rc);
2415 }
2416
2417 static int filter_create(struct obd_export *exp, struct obdo *oa,
2418                          struct lov_stripe_md **ea, struct obd_trans_info *oti)
2419 {
2420         struct obd_device *obd = NULL;
2421         struct filter_obd *filter;
2422         struct lvfs_run_ctxt saved;
2423         struct lov_stripe_md *lsm = NULL;
2424         struct filter_export_data *fed;
2425         char str[PTL_NALFMT_SIZE];
2426         int group = oa->o_gr, rc = 0, diff, recreate_objs = 0;
2427         ENTRY;
2428
2429         if (!(oa->o_valid & OBD_MD_FLGROUP) || group == 0) {
2430                 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
2431                                 exp->exp_connection->c_peer.peer_id.nid, str);
2432                 CERROR("!!! nid "LPX64"/%s sent invalid object group %d\n",
2433                        exp->exp_connection->c_peer.peer_id.nid, str, group);
2434                 RETURN(-EINVAL);
2435         }
2436
2437         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
2438             (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
2439                 recreate_objs = 1;
2440         }
2441
2442         obd = exp->exp_obd;
2443         fed = &exp->exp_filter_data;
2444         filter = &obd->u.filter;
2445
2446         if (fed->fed_group != group && !recreate_objs &&
2447             !(oa->o_valid & OBD_MD_REINT)) {
2448                 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
2449                                 exp->exp_connection->c_peer.peer_id.nid, str);
2450                 CERROR("!!! This export (nid "LPX64"/%s) used object group %d "
2451                        "earlier; now it's trying to use group %d!  This could "
2452                        "be a bug in the MDS.  Tell CFS.\n",
2453                        exp->exp_connection->c_peer.peer_id.nid, str,
2454                        fed->fed_group, group);
2455                 RETURN(-ENOTUNIQ);
2456         }
2457
2458         CDEBUG(D_INFO, "filter_create(od->o_gr=%d,od->o_id="LPU64")\n",
2459                group, oa->o_id);
2460         if (ea != NULL) {
2461                 lsm = *ea;
2462                 if (lsm == NULL) {
2463                         rc = obd_alloc_memmd(exp, &lsm);
2464                         if (rc < 0)
2465                                 RETURN(rc);
2466                 }
2467         }
2468
2469         obd = exp->exp_obd;
2470         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2471
2472         if (oa->o_valid & OBD_MD_REINT) {
2473                 int num = *((int*)oa->o_inline);
2474                 rc = filter_precreate(obd, oa, oa->o_gr, &num);
2475         } else if (recreate_objs) {
2476                 if (oa->o_id > filter_last_id(&obd->u.filter, group)) {
2477                         CERROR("recreate objid "LPU64" > last id "LPU64"\n",
2478                                oa->o_id, filter_last_id(&obd->u.filter, group));
2479                         rc = -EINVAL;
2480                 } else {
2481                         diff = 1;
2482                         rc = filter_precreate(obd, oa, group, &diff);
2483                 }
2484         } else {
2485                 diff = filter_should_precreate(exp, oa, group);
2486                 if (diff > 0) {
2487                         oa->o_id = filter_last_id(&obd->u.filter, group);
2488                         rc = filter_precreate(obd, oa, group, &diff);
2489                         oa->o_id += diff;
2490                         oa->o_valid = OBD_MD_FLID;
2491                 }
2492         }
2493
2494         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2495         if (rc && ea != NULL && *ea != lsm) {
2496                 obd_free_memmd(exp, &lsm);
2497         } else if (rc == 0 && ea != NULL) {
2498                 /* XXX LOV STACKING: the lsm that is passed to us from
2499                  * LOV does not have valid lsm_oinfo data structs, so
2500                  * don't go touching that.  This needs to be fixed in a
2501                  * big way. */
2502                 lsm->lsm_object_id = oa->o_id;
2503                 lsm->lsm_object_gr = oa->o_gr;
2504                 *ea = lsm;
2505         }
2506
2507         RETURN(rc);
2508 }
2509
2510 static int filter_destroy(struct obd_export *exp, struct obdo *oa,
2511                           struct lov_stripe_md *ea, struct obd_trans_info *oti)
2512 {
2513         int rc, rc2, cleanup_phase = 0, have_prepared = 0;
2514         struct dentry *dchild = NULL, *dparent = NULL;
2515         struct llog_cookie *fcc = NULL;
2516         struct lvfs_run_ctxt saved;
2517         struct filter_obd *filter;
2518         struct obd_device *obd;
2519         void *handle = NULL;
2520         void *lock = NULL;
2521         struct iattr iattr;
2522         ENTRY;
2523
2524         LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2525
2526         obd = exp->exp_obd;
2527         filter = &obd->u.filter;
2528
2529         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2530
2531  acquire_locks:
2532         dparent = filter_parent_lock(obd, oa->o_gr, oa->o_id, &lock);
2533         if (IS_ERR(dparent))
2534                 GOTO(cleanup, rc = PTR_ERR(dparent));
2535         cleanup_phase = 1;
2536
2537         dchild = filter_id2dentry(obd, dparent, oa->o_gr, oa->o_id);
2538         if (IS_ERR(dchild))
2539                 GOTO(cleanup, rc = -ENOENT);
2540         cleanup_phase = 2;
2541
2542         if (dchild->d_inode == NULL) {
2543                 CDEBUG(D_INODE, "destroying non-existent object "LPU64"\n",
2544                        oa->o_id);
2545                 GOTO(cleanup, rc = -ENOENT);
2546         }
2547
2548         if (!have_prepared) {
2549                 /* If we're really going to destroy the object, get ready
2550                  * by getting the clients to discard their cached data.
2551                  *
2552                  * We have to drop the parent lock, because
2553                  * filter_prepare_destroy will acquire a PW on the object, and
2554                  * we don't want to deadlock with an incoming write to the
2555                  * object, which has the extent PW and then wants to get the
2556                  * parent dentry to do the lookup.
2557                  *
2558                  * We dput the child because it's not worth the extra
2559                  * complication of condition the above code to skip it on the
2560                  * second time through. */
2561                 f_dput(dchild);
2562                 filter_parent_unlock(dparent, lock);
2563
2564                 filter_prepare_destroy(obd, oa->o_id, oa->o_gr);
2565                 have_prepared = 1;
2566                 goto acquire_locks;
2567         }
2568
2569         /* Our MDC connection is established by the MDS to us */
2570         if (oa->o_valid & OBD_MD_FLCOOKIE) {
2571                 OBD_ALLOC(fcc, sizeof(*fcc));
2572                 if (fcc != NULL)
2573                         memcpy(fcc, obdo_logcookie(oa), sizeof(*fcc));
2574         }
2575
2576         /* we're gonna truncate it first in order to avoid possible
2577          * deadlock:
2578          *      P1                      P2
2579          * open trasaction      open transaction
2580          * down(i_zombie)       down(i_zombie)
2581          *                      restart transaction
2582          * (see BUG 4180) -bzzz
2583          */
2584         down(&dchild->d_inode->i_sem);
2585         handle = fsfilt_start_log(obd, dparent->d_inode,FSFILT_OP_SETATTR,NULL,1);
2586         if (IS_ERR(handle)) {
2587                 up(&dchild->d_inode->i_sem);
2588                 GOTO(cleanup, rc = PTR_ERR(handle));
2589         }
2590
2591         iattr.ia_valid = ATTR_SIZE;
2592         iattr.ia_size = 0;
2593         rc = fsfilt_setattr(obd, dchild, handle, &iattr, 1);
2594
2595         rc2 = fsfilt_commit(obd, filter->fo_sb, dparent->d_inode, handle, 0);
2596         up(&dchild->d_inode->i_sem);
2597         if (rc)
2598                 GOTO(cleanup, rc);
2599         if (rc2)
2600                 GOTO(cleanup, rc = rc2);
2601
2602         handle = fsfilt_start_log(obd, dparent->d_inode,FSFILT_OP_UNLINK,oti,1);
2603         if (IS_ERR(handle))
2604                 GOTO(cleanup, rc = PTR_ERR(handle));
2605
2606         cleanup_phase = 3;
2607
2608         rc = filter_destroy_internal(obd, oa->o_id, dparent, dchild);
2609
2610 cleanup:
2611         switch(cleanup_phase) {
2612         case 3:
2613                 if (fcc != NULL) {
2614                         if (oti != NULL)
2615                                 fsfilt_add_journal_cb(obd, filter->fo_sb, 0,
2616                                                       oti->oti_handle,
2617                                                       filter_cancel_cookies_cb,
2618                                                       fcc);
2619                         else
2620                                 fsfilt_add_journal_cb(obd, filter->fo_sb, 0,
2621                                                       handle,
2622                                                       filter_cancel_cookies_cb,
2623                                                       fcc);
2624                 }
2625                 rc = filter_finish_transno(exp, oti, rc);
2626                 rc2 = fsfilt_commit(obd, filter->fo_sb, dparent->d_inode,
2627                                     handle, 0);
2628                 if (rc2) {
2629                         CERROR("error on commit, err = %d\n", rc2);
2630                         if (!rc)
2631                                 rc = rc2;
2632                 }
2633         case 2:
2634                 f_dput(dchild);
2635         case 1:
2636                 filter_parent_unlock(dparent, lock);
2637         case 0:
2638                 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2639                 break;
2640         default:
2641                 CERROR("invalid cleanup_phase %d\n", cleanup_phase);
2642                 LBUG();
2643         }
2644
2645         RETURN(rc);
2646 }
2647
2648 /* NB start and end are used for punch, but not truncate */
2649 static int filter_truncate(struct obd_export *exp, struct obdo *oa,
2650                            struct lov_stripe_md *lsm,
2651                            obd_off start, obd_off end,
2652                            struct obd_trans_info *oti)
2653 {
2654         int error;
2655         ENTRY;
2656
2657         if (end != OBD_OBJECT_EOF)
2658                 CERROR("PUNCH not supported, only truncate: end = "LPX64"\n",
2659                        end);
2660
2661         CDEBUG(D_INODE, "calling truncate for object "LPU64", valid = "LPU64", "
2662                "o_size = "LPD64"\n", oa->o_id, oa->o_valid, start);
2663         oa->o_size = start;
2664         error = filter_setattr(exp, oa, NULL, oti);
2665         RETURN(error);
2666 }
2667
2668 static int filter_sync(struct obd_export *exp, struct obdo *oa,
2669                        struct lov_stripe_md *lsm, obd_off start, obd_off end)
2670 {
2671         struct obd_device *obd = exp->exp_obd;
2672         struct lvfs_run_ctxt saved;
2673         struct filter_obd *filter;
2674         struct dentry *dentry;
2675         int rc, rc2;
2676         ENTRY;
2677
2678         filter = &obd->u.filter;
2679
2680         /* an objid of zero is taken to mean "sync whole filesystem" */
2681         if (!oa || !(oa->o_valid & OBD_MD_FLID)) {
2682                 rc = fsfilt_sync(obd, filter->fo_sb);
2683                 /* flush any remaining cancel messages out to the target */
2684                 filter_sync_llogs(obd, NULL);
2685                 RETURN(rc);
2686         }
2687
2688         dentry = filter_oa2dentry(obd, oa);
2689         if (IS_ERR(dentry))
2690                 RETURN(PTR_ERR(dentry));
2691
2692         push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2693
2694         down(&dentry->d_inode->i_sem);
2695         rc = filemap_fdatasync(dentry->d_inode->i_mapping);
2696         if (rc == 0) {
2697                 /* just any file to grab fsync method - "file" arg unused */
2698                 struct file *file = filter->fo_rcvd_filp;
2699
2700                 if (file->f_op && file->f_op->fsync)
2701                         rc = file->f_op->fsync(NULL, dentry, 1);
2702
2703                 rc2 = filemap_fdatawait(dentry->d_inode->i_mapping);
2704                 if (!rc)
2705                         rc = rc2;
2706         }
2707         up(&dentry->d_inode->i_sem);
2708
2709         oa->o_valid = OBD_MD_FLID;
2710         obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2711
2712         pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2713
2714         f_dput(dentry);
2715         RETURN(rc);
2716 }
2717
2718 static int filter_statfs(struct obd_device *obd, struct obd_statfs *osfs,
2719                          unsigned long max_age)
2720 {
2721         struct filter_obd *filter = &obd->u.filter;
2722         int blockbits = filter->fo_sb->s_blocksize_bits;
2723         int rc;
2724         ENTRY;
2725
2726         /* at least try to account for cached pages.  its still racey and
2727          * might be under-reporting if clients haven't announced their
2728          * caches with brw recently */
2729         spin_lock(&obd->obd_osfs_lock);
2730         rc = fsfilt_statfs(obd, filter->fo_sb, max_age);
2731         memcpy(osfs, &obd->obd_osfs, sizeof(*osfs));
2732         spin_unlock(&obd->obd_osfs_lock);
2733
2734         CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
2735                " pending "LPU64" free "LPU64" avail "LPU64"\n",
2736                filter->fo_tot_dirty, filter->fo_tot_granted,
2737                filter->fo_tot_pending,
2738                osfs->os_bfree << blockbits, osfs->os_bavail << blockbits);
2739
2740         filter_grant_sanity_check(obd, __FUNCTION__);
2741
2742         osfs->os_bavail -= min(osfs->os_bavail,
2743                                (filter->fo_tot_dirty + filter->fo_tot_pending +
2744                                 osfs->os_bsize -1) >> blockbits);
2745
2746         RETURN(rc);
2747 }
2748
2749 static int filter_get_info(struct obd_export *exp, __u32 keylen,
2750                            void *key, __u32 *vallen, void *val)
2751 {
2752         struct filter_export_data *fed = &exp->exp_filter_data;
2753         struct obd_device *obd;
2754         ENTRY;
2755
2756         obd = class_exp2obd(exp);
2757         if (obd == NULL) {
2758                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2759                        exp->exp_handle.h_cookie);
2760                 RETURN(-EINVAL);
2761         }
2762
2763         if (keylen == strlen("blocksize") &&
2764             memcmp(key, "blocksize", keylen) == 0) {
2765                 __u32 *blocksize = val;
2766                 *vallen = sizeof(*blocksize);
2767                 *blocksize = obd->u.filter.fo_sb->s_blocksize;
2768                 RETURN(0);
2769         }
2770
2771         if (keylen == strlen("blocksize_bits") &&
2772             memcmp(key, "blocksize_bits", keylen) == 0) {
2773                 __u32 *blocksize_bits = val;
2774                 *vallen = sizeof(*blocksize_bits);
2775                 *blocksize_bits = obd->u.filter.fo_sb->s_blocksize_bits;
2776                 RETURN(0);
2777         }
2778
2779         if (keylen >= strlen("last_id") && memcmp(key, "last_id", 7) == 0) {
2780                 obd_id *last_id = val;
2781                 *last_id = filter_last_id(&obd->u.filter, fed->fed_group);
2782                 RETURN(0);
2783         }
2784         if (keylen >= strlen("reint_log") && memcmp(key, "reint_log", 9) == 0) {
2785                 /*Get log_context handle*/
2786                 unsigned long *llh_handle = val;
2787                 *vallen = sizeof(unsigned long);
2788                 *llh_handle = (unsigned long)obd->obd_llog_ctxt[LLOG_REINT_ORIG_CTXT];
2789                 RETURN(0);
2790         }
2791         if (keylen >= strlen("cache_sb") && memcmp(key, "cache_sb", 8) == 0) {
2792                 /*Get log_context handle*/
2793                 unsigned long *sb = val;
2794                 *vallen = sizeof(unsigned long);
2795                 *sb = (unsigned long)obd->u.filter.fo_sb;
2796                 RETURN(0);
2797         }
2798
2799         CDEBUG(D_IOCTL, "invalid key\n");
2800         RETURN(-EINVAL);
2801 }
2802
2803 struct obd_llogs *filter_grab_llog_for_group(struct obd_device *obd, int group,
2804                                              struct obd_export *export)
2805 {
2806         struct filter_group_llog *fglog, *nlog;
2807         char name[32] = "CATLIST";
2808         struct filter_obd *filter;
2809         struct llog_ctxt *ctxt;
2810         struct list_head *cur;
2811         int rc;
2812
2813         filter = &obd->u.filter;
2814
2815         spin_lock(&filter->fo_llog_list_lock);
2816         list_for_each(cur, &filter->fo_llog_list) {
2817                 fglog = list_entry(cur, struct filter_group_llog, list);
2818                 if (fglog->group == group) {
2819                         if (!(fglog->exp == NULL || fglog->exp == export || export == NULL))
2820                                 CWARN("%s: export for group %d changes: 0x%p -> 0x%p\n",
2821                                       obd->obd_name, group, fglog->exp, export);
2822                         spin_unlock(&filter->fo_llog_list_lock);
2823                         goto init;
2824                 }
2825         }
2826         spin_unlock(&filter->fo_llog_list_lock);
2827
2828         if (export == NULL)
2829                 RETURN(NULL);
2830
2831         OBD_ALLOC(fglog, sizeof(*fglog));
2832         if (fglog == NULL)
2833                 RETURN(NULL);
2834         fglog->group = group;
2835
2836         OBD_ALLOC(fglog->llogs, sizeof(struct obd_llogs));
2837         if (fglog->llogs == NULL) {
2838                 OBD_FREE(fglog, sizeof(*fglog));
2839                 RETURN(NULL);
2840         }
2841
2842         spin_lock(&filter->fo_llog_list_lock);
2843         list_for_each(cur, &filter->fo_llog_list) {
2844                 nlog = list_entry(cur, struct filter_group_llog, list);
2845                 LASSERT(nlog->group != group);
2846         }
2847         list_add(&fglog->list, &filter->fo_llog_list);
2848         spin_unlock(&filter->fo_llog_list_lock);
2849
2850         rc = obd_llog_cat_initialize(obd, fglog->llogs, 1, name);
2851         if (rc) {
2852                 OBD_FREE(fglog->llogs, sizeof(*(fglog->llogs)));
2853                 OBD_FREE(fglog, sizeof(*fglog));
2854                 RETURN(NULL);
2855         }
2856
2857 init:
2858         if (export) {
2859                 fglog->exp = export;
2860                 ctxt = llog_get_context(fglog->llogs, LLOG_UNLINK_REPL_CTXT);
2861                 LASSERT(ctxt != NULL);
2862
2863                 llog_receptor_accept(ctxt, export->exp_imp_reverse);
2864         }
2865         CDEBUG(D_OTHER, "%s: new llog 0x%p for group %u\n",
2866                obd->obd_name, fglog->llogs, group);
2867
2868         RETURN(fglog->llogs);
2869 }
2870
2871 int filter_iocontrol(unsigned int cmd, struct obd_export *exp,
2872                      int len, void *karg, void *uarg)
2873 {
2874         struct obd_device *obd = exp->exp_obd;
2875         struct obd_ioctl_data *data = karg;
2876         int rc = 0;
2877
2878         switch (cmd) {
2879         case OBD_IOC_ABORT_RECOVERY:
2880                 target_stop_recovery_thread(obd);
2881                 RETURN(0);
2882
2883         case OBD_IOC_SET_READONLY: {
2884                 void *handle;
2885                 struct super_block *sb = obd->u.filter.fo_sb;
2886                 struct inode *inode = sb->s_root->d_inode;
2887                 BDEVNAME_DECLARE_STORAGE(tmp);
2888                 CERROR("setting device %s read-only\n",
2889                        ll_bdevname(sb, tmp));
2890
2891                 handle = fsfilt_start(obd, inode, FSFILT_OP_MKNOD, NULL);
2892                 LASSERT(handle);
2893                 (void)fsfilt_commit(obd, sb, inode, handle, 1);
2894
2895                 dev_set_rdonly(ll_sbdev(obd->u.filter.fo_sb), 2);
2896                 RETURN(0);
2897         }
2898
2899         case OBD_IOC_CATLOGLIST: {
2900                 rc = llog_catalog_list(obd, 1, data);
2901                 RETURN(rc);
2902         }
2903         case OBD_IOC_LLOG_CANCEL:
2904         case OBD_IOC_LLOG_REMOVE:
2905         case OBD_IOC_LLOG_INFO:
2906         case OBD_IOC_LLOG_PRINT: {
2907                 /* FIXME to be finished */
2908                 RETURN(-EOPNOTSUPP);
2909 /*
2910                 struct llog_ctxt *ctxt = NULL;
2911
2912                 push_ctxt(&saved, &ctxt->loc_ctxt, NULL);
2913                 rc = llog_ioctl(ctxt, cmd, data);
2914                 pop_ctxt(&saved, &ctxt->loc_ctxt, NULL);
2915
2916                 RETURN(rc);
2917 */
2918         }
2919
2920
2921         default:
2922                 RETURN(-EINVAL);
2923         }
2924         RETURN(0);
2925 }
2926
2927 static struct llog_operations filter_unlink_repl_logops;
2928 static struct llog_operations filter_size_orig_logops = {
2929         lop_setup: llog_obd_origin_setup,
2930         lop_cleanup: llog_catalog_cleanup,
2931         lop_add: llog_catalog_add,
2932 };
2933
2934 static int filter_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
2935                             struct obd_device *tgt, int count,
2936                             struct llog_catid *catid)
2937 {
2938         struct llog_ctxt *ctxt;
2939         int rc;
2940         ENTRY;
2941
2942         filter_unlink_repl_logops = llog_client_ops;
2943         filter_unlink_repl_logops.lop_cancel = llog_obd_repl_cancel;
2944         filter_unlink_repl_logops.lop_connect = llog_repl_connect;
2945         filter_unlink_repl_logops.lop_sync = llog_obd_repl_sync;
2946
2947         rc = obd_llog_setup(obd, llogs, LLOG_UNLINK_REPL_CTXT, tgt, 0, NULL,
2948                         &filter_unlink_repl_logops);
2949         if (rc)
2950                 RETURN(rc);
2951         /* FIXME - assign unlink_cb for filter's recovery */
2952         ctxt = llog_get_context(llogs, LLOG_UNLINK_REPL_CTXT);
2953         ctxt->llog_proc_cb = filter_recov_log_unlink_cb;
2954
2955         /* FIXME - count should be 1 to setup size log */
2956         rc = obd_llog_setup(obd, llogs, LLOG_SIZE_ORIG_CTXT, tgt, 0,
2957                             &catid->lci_logid, &filter_size_orig_logops);
2958         RETURN(rc);
2959 }
2960
2961 static int filter_llog_finish(struct obd_device *obd,
2962                               struct obd_llogs *llogs, int count)
2963 {
2964         int rc;
2965         ENTRY;
2966
2967         rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_UNLINK_REPL_CTXT));
2968         if (rc)
2969                 RETURN(rc);
2970
2971         rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_SIZE_ORIG_CTXT));
2972         RETURN(rc);
2973 }
2974
2975 static int filter_llog_connect(struct obd_export *exp,
2976                                struct llogd_conn_body *body)
2977 {
2978         struct obd_device *obd = exp->exp_obd;
2979         struct llog_ctxt *ctxt;
2980         struct obd_llogs *llog;
2981         int rc;
2982         ENTRY;
2983
2984         CDEBUG(D_OTHER, "handle connect for %s: %u/%u/%u\n", obd->obd_name,
2985                (unsigned) body->lgdc_logid.lgl_ogr,
2986                (unsigned) body->lgdc_logid.lgl_oid,
2987                (unsigned) body->lgdc_logid.lgl_ogen);
2988         llog = filter_grab_llog_for_group(obd, body->lgdc_logid.lgl_ogr, exp);
2989         LASSERT(llog != NULL);
2990         ctxt = llog_get_context(llog, body->lgdc_ctxt_idx);
2991         rc = llog_connect(ctxt, 1, &body->lgdc_logid,
2992                           &body->lgdc_gen, NULL);
2993         if (rc != 0)
2994                 CERROR("failed to connect\n");
2995
2996         RETURN(rc);
2997 }
2998
2999 static struct dentry *filter_lvfs_id2dentry(__u64 id, __u32 gen,
3000                                             __u64 gr, void *data)
3001 {
3002         return filter_id2dentry(data, NULL, gr, id);
3003 }
3004
3005 static struct lvfs_callback_ops filter_lvfs_ops = {
3006         l_id2dentry:     filter_lvfs_id2dentry,
3007 };
3008
3009 static struct obd_ops filter_obd_ops = {
3010         .o_owner          = THIS_MODULE,
3011         .o_attach         = filter_attach,
3012         .o_detach         = filter_detach,
3013         .o_get_info       = filter_get_info,
3014         .o_setup          = filter_setup,
3015         .o_precleanup     = filter_precleanup,
3016         .o_cleanup        = filter_cleanup,
3017         .o_connect        = filter_connect,
3018         .o_connect_post   = filter_connect_post,
3019         .o_disconnect     = filter_disconnect,
3020         .o_statfs         = filter_statfs,
3021         .o_getattr        = filter_getattr,
3022         .o_unpackmd       = filter_unpackmd,
3023         .o_create         = filter_create,
3024         .o_setattr        = filter_setattr,
3025         .o_destroy        = filter_destroy,
3026         .o_brw            = filter_brw,
3027         .o_punch          = filter_truncate,
3028         .o_sync           = filter_sync,
3029         .o_preprw         = filter_preprw,
3030         .o_commitrw       = filter_commitrw,
3031         .o_do_cow         = filter_do_cow,
3032         .o_write_extents  = filter_write_extents,
3033         .o_destroy_export = filter_destroy_export,
3034         .o_llog_init      = filter_llog_init,
3035         .o_llog_finish    = filter_llog_finish,
3036         .o_llog_connect   = filter_llog_connect,
3037         .o_iocontrol      = filter_iocontrol,
3038 };
3039
3040 static struct obd_ops filter_sanobd_ops = {
3041         .o_owner          = THIS_MODULE,
3042         .o_attach         = filter_attach,
3043         .o_detach         = filter_detach,
3044         .o_get_info       = filter_get_info,
3045         .o_setup          = filter_san_setup,
3046         .o_precleanup     = filter_precleanup,
3047         .o_cleanup        = filter_cleanup,
3048         .o_connect        = filter_connect,
3049         .o_connect_post   = filter_connect_post,
3050         .o_disconnect     = filter_disconnect,
3051         .o_statfs         = filter_statfs,
3052         .o_getattr        = filter_getattr,
3053         .o_unpackmd       = filter_unpackmd,
3054         .o_create         = filter_create,
3055         .o_setattr        = filter_setattr,
3056         .o_destroy        = filter_destroy,
3057         .o_brw            = filter_brw,
3058         .o_punch          = filter_truncate,
3059         .o_sync           = filter_sync,
3060         .o_preprw         = filter_preprw,
3061         .o_commitrw       = filter_commitrw,
3062         .o_do_cow         = filter_do_cow,
3063         .o_write_extents  = filter_write_extents,
3064         .o_san_preprw     = filter_san_preprw,
3065         .o_destroy_export = filter_destroy_export,
3066         .o_llog_init      = filter_llog_init,
3067         .o_llog_finish    = filter_llog_finish,
3068         .o_llog_connect   = filter_llog_connect,
3069         .o_iocontrol      = filter_iocontrol,
3070 };
3071
3072 static int __init obdfilter_init(void)
3073 {
3074         struct lprocfs_static_vars lvars;
3075         int rc;
3076
3077         printk(KERN_INFO "Lustre: Filtering OBD driver; info@clusterfs.com\n");
3078
3079         lprocfs_init_vars(filter, &lvars);
3080
3081         rc = class_register_type(&filter_obd_ops, NULL, lvars.module_vars,
3082                                  OBD_FILTER_DEVICENAME);
3083         if (rc)
3084                 return rc;
3085
3086         rc = class_register_type(&filter_sanobd_ops, NULL, lvars.module_vars,
3087                                  OBD_FILTER_SAN_DEVICENAME);
3088         if (rc)
3089                 class_unregister_type(OBD_FILTER_DEVICENAME);
3090         return rc;
3091 }
3092
3093 static void __exit obdfilter_exit(void)
3094 {
3095         class_unregister_type(OBD_FILTER_SAN_DEVICENAME);
3096         class_unregister_type(OBD_FILTER_DEVICENAME);
3097 }
3098
3099 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
3100 MODULE_DESCRIPTION("Lustre Filtering OBD driver");
3101 MODULE_LICENSE("GPL");
3102
3103 module_init(obdfilter_init);
3104 module_exit(obdfilter_exit);