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 <libcfs/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, FILTER_LR_MAX_CLIENTS/8);
 304         filter->fo_last_rcvd_slots = NULL;
 305         return 0;
 306 }
 307
 308 /* assumes caller is already in kernel ctxt */
 309 int filter_update_server_data(struct obd_device *obd, struct file *filp,
 310                               struct filter_server_data *fsd, int force_sync)
 311 {
 312         loff_t off = 0;
 313         int rc;
 314         ENTRY;
 315
 316         CDEBUG(D_INODE, "server uuid      : %s\n", fsd->fsd_uuid);
 317         CDEBUG(D_INODE, "server last_rcvd : "LPU64"\n",
 318                le64_to_cpu(fsd->fsd_last_transno));
 319         CDEBUG(D_INODE, "server last_mount: "LPU64"\n",
 320                le64_to_cpu(fsd->fsd_mount_count));
 321
 322         rc = fsfilt_write_record(obd, filp, fsd, sizeof(*fsd), &off,force_sync);
 323         if (rc)
 324                 CERROR("error writing filter_server_data: rc = %d\n", rc);
 325
 326         RETURN(rc);
 327 }
 328
 329 int filter_update_last_objid(struct obd_device *obd, obd_gr group,
 330                              int force_sync)
 331 {
 332         struct filter_obd *filter = &obd->u.filter;
 333         __u64 tmp;
 334         loff_t off = 0;
 335         int rc;
 336         ENTRY;
 337
 338         if (filter->fo_last_objid_files[group] == NULL) {
 339                 CERROR("Object group "LPU64" not fully setup; not updating "
 340                        "last_objid\n", group);
 341                 RETURN(0);
 342         }
 343
 344         CDEBUG(D_INODE, "server last_objid for group "LPU64": "LPU64"\n",
 345                group, filter->fo_last_objids[group]);
 346
 347         tmp = cpu_to_le64(filter->fo_last_objids[group]);
 348         rc = fsfilt_write_record(obd, filter->fo_last_objid_files[group],
 349                                  &tmp, sizeof(tmp), &off, force_sync);
 350         if (rc)
 351                 CERROR("error writing group "LPU64" last objid: rc = %d\n",
 352                        group, rc);
 353         RETURN(rc);
 354 }
 355
 356 /* assumes caller has already in kernel ctxt */
 357 static int filter_init_server_data(struct obd_device *obd, struct file * filp)
 358 {
 359         struct filter_obd *filter = &obd->u.filter;
 360         struct filter_server_data *fsd;
 361         struct filter_client_data *fcd = NULL;
 362         struct inode *inode = filp->f_dentry->d_inode;
 363         unsigned long last_rcvd_size = inode->i_size;
 364         __u64 mount_count;
 365         int cl_idx;
 366         loff_t off = 0;
 367         int rc;
 368
 369         /* ensure padding in the struct is the correct size */
 370         LASSERT (offsetof(struct filter_server_data, fsd_padding) +
 371                  sizeof(fsd->fsd_padding) == FILTER_LR_SERVER_SIZE);
 372         LASSERT (offsetof(struct filter_client_data, fcd_padding) +
 373                  sizeof(fcd->fcd_padding) == FILTER_LR_CLIENT_SIZE);
 374
 375         OBD_ALLOC(fsd, sizeof(*fsd));
 376         if (!fsd)
 377                 RETURN(-ENOMEM);
 378         filter->fo_fsd = fsd;
 379
 380         OBD_ALLOC(filter->fo_last_rcvd_slots, FILTER_LR_MAX_CLIENTS/8);
 381         if (filter->fo_last_rcvd_slots == NULL) {
 382                 OBD_FREE(fsd, sizeof(*fsd));
 383                 RETURN(-ENOMEM);
 384         }
 385
 386         if (last_rcvd_size == 0) {
 387                 CWARN("%s: initializing new %s\n", obd->obd_name, LAST_RCVD);
 388
 389                 memcpy(fsd->fsd_uuid, obd->obd_uuid.uuid,sizeof(fsd->fsd_uuid));
 390                 fsd->fsd_last_transno = 0;
 391                 mount_count = fsd->fsd_mount_count = 0;
 392                 fsd->fsd_server_size = cpu_to_le32(FILTER_LR_SERVER_SIZE);
 393                 fsd->fsd_client_start = cpu_to_le32(FILTER_LR_CLIENT_START);
 394                 fsd->fsd_client_size = cpu_to_le16(FILTER_LR_CLIENT_SIZE);
 395                 fsd->fsd_subdir_count = cpu_to_le16(FILTER_SUBDIR_COUNT);
 396                 filter->fo_subdir_count = FILTER_SUBDIR_COUNT;
 397         } else {
 398                 rc = fsfilt_read_record(obd, filp, fsd, sizeof(*fsd), &off);
 399                 if (rc) {
 400                         CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
 401                                LAST_RCVD, rc);
 402                         GOTO(err_fsd, rc);
 403                 }
 404                 if (strcmp(fsd->fsd_uuid, obd->obd_uuid.uuid) != 0) {
 405                         CERROR("OBD UUID %s does not match last_rcvd UUID %s\n",
 406                                obd->obd_uuid.uuid, fsd->fsd_uuid);
 407                         GOTO(err_fsd, rc = -EINVAL);
 408                 }
 409                 mount_count = le64_to_cpu(fsd->fsd_mount_count);
 410                 filter->fo_subdir_count = le16_to_cpu(fsd->fsd_subdir_count);
 411         }
 412
 413         if (fsd->fsd_feature_incompat & ~cpu_to_le32(FILTER_INCOMPAT_SUPP)) {
 414                 CERROR("unsupported feature %x\n",
 415                        le32_to_cpu(fsd->fsd_feature_incompat) &
 416                        ~FILTER_INCOMPAT_SUPP);
 417                 GOTO(err_fsd, rc = -EINVAL);
 418         }
 419         if (fsd->fsd_feature_rocompat & ~cpu_to_le32(FILTER_ROCOMPAT_SUPP)) {
 420                 CERROR("read-only feature %x\n",
 421                        le32_to_cpu(fsd->fsd_feature_rocompat) &
 422                        ~FILTER_ROCOMPAT_SUPP);
 423                 /* Do something like remount filesystem read-only */
 424                 GOTO(err_fsd, rc = -EINVAL);
 425         }
 426
 427         CDEBUG(D_INODE, "%s: server last_rcvd : "LPU64"\n",
 428                obd->obd_name, le64_to_cpu(fsd->fsd_last_transno));
 429         CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
 430                obd->obd_name, mount_count + 1);
 431         CDEBUG(D_INODE, "%s: server data size: %u\n",
 432                obd->obd_name, le32_to_cpu(fsd->fsd_server_size));
 433         CDEBUG(D_INODE, "%s: per-client data start: %u\n",
 434                obd->obd_name, le32_to_cpu(fsd->fsd_client_start));
 435         CDEBUG(D_INODE, "%s: per-client data size: %u\n",
 436                obd->obd_name, le32_to_cpu(fsd->fsd_client_size));
 437         CDEBUG(D_INODE, "%s: server subdir_count: %u\n",
 438                obd->obd_name, le16_to_cpu(fsd->fsd_subdir_count));
 439         CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
 440                last_rcvd_size <= le32_to_cpu(fsd->fsd_client_start) ? 0 :
 441                (last_rcvd_size - le32_to_cpu(fsd->fsd_client_start)) /
 442                 le16_to_cpu(fsd->fsd_client_size));
 443
 444         if (!obd->obd_replayable) {
 445                 CWARN("%s: recovery support OFF\n", obd->obd_name);
 446                 GOTO(out, rc = 0);
 447         }
 448
 449         for (cl_idx = 0, off = le32_to_cpu(fsd->fsd_client_start);
 450              off < last_rcvd_size; cl_idx++) {
 451                 __u64 last_rcvd;
 452                 struct obd_export *exp;
 453                 struct filter_export_data *fed;
 454
 455                 if (!fcd) {
 456                         OBD_ALLOC(fcd, sizeof(*fcd));
 457                         if (!fcd)
 458                                 GOTO(err_client, rc = -ENOMEM);
 459                 }
 460
 461                 /* Don't assume off is incremented properly by
 462                  * fsfilt_read_record(), in case sizeof(*fcd)
 463                  * isn't the same as fsd->fsd_client_size.  */
 464                 off = le32_to_cpu(fsd->fsd_client_start) +
 465                         cl_idx * le16_to_cpu(fsd->fsd_client_size);
 466                 rc = fsfilt_read_record(obd, filp, fcd, sizeof(*fcd), &off);
 467                 if (rc) {
 468                         CERROR("error reading FILT %s idx %d off %llu: rc %d\n",
 469                                LAST_RCVD, cl_idx, off, rc);
 470                         break; /* read error shouldn't cause startup to fail */
 471                 }
 472
 473                 if (fcd->fcd_uuid[0] == '\0') {
 474                         CDEBUG(D_INFO, "skipping zeroed client at offset %d\n",
 475                                cl_idx);
 476                         continue;
 477                 }
 478
 479                 last_rcvd = le64_to_cpu(fcd->fcd_last_rcvd);
 480
 481                 /* These exports are cleaned up by filter_disconnect(), so they
 482                  * need to be set up like real exports as filter_connect() does.
 483                  */
 484                 exp = class_new_export(obd);
 485                 CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64" "
 486                        "srv lr: "LPU64" fcd_group %d \n", fcd->fcd_uuid, cl_idx,
 487                        last_rcvd, le64_to_cpu(fsd->fsd_last_transno),
 488                        le32_to_cpu(fcd->fcd_group));
 489                 if (exp == NULL)
 490                         GOTO(err_client, rc = -ENOMEM);
 491
 492                 memcpy(&exp->exp_client_uuid.uuid, fcd->fcd_uuid,
 493                        sizeof exp->exp_client_uuid.uuid);
 494                 fed = &exp->exp_filter_data;
 495                 fed->fed_fcd = fcd;
 496                 fed->fed_group = le32_to_cpu(fcd->fcd_group);
 497                 filter_client_add(obd, filter, fed, cl_idx);
 498                 /* create helper if export init gets more complex */
 499                 spin_lock_init(&fed->fed_lock);
 500
 501                 fcd = NULL;
 502                 exp->exp_req_replay_needed = 1;
 503                 exp->exp_lock_replay_needed = 1;
 504                 atomic_inc(&obd->obd_req_replay_clients);
 505                 atomic_inc(&obd->obd_lock_replay_clients);
 506                 obd->obd_recoverable_clients++;
 507                 obd->obd_max_recoverable_clients++;
 508                 class_export_put(exp);
 509
 510                 CDEBUG(D_OTHER, "client at idx %d has last_rcvd = "LPU64"\n",
 511                        cl_idx, last_rcvd);
 512
 513                 if (last_rcvd > le64_to_cpu(fsd->fsd_last_transno))
 514                         fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
 515
 516         }
 517
 518         obd->obd_last_committed = le64_to_cpu(fsd->fsd_last_transno);
 519
 520         if (obd->obd_recoverable_clients) {
 521                 CWARN("RECOVERY: service %s, %d recoverable clients, "
 522                       "last_transno "LPU64"\n", obd->obd_name,
 523                       obd->obd_recoverable_clients,
 524                       le64_to_cpu(fsd->fsd_last_transno));
 525                 obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
 526                 target_start_recovery_thread(obd, ost_handle);
 527         }
 528
 529         if (fcd)
 530                 OBD_FREE(fcd, sizeof(*fcd));
 531
 532 out:
 533         filter->fo_mount_count = mount_count + 1;
 534         fsd->fsd_mount_count = cpu_to_le64(filter->fo_mount_count);
 535
 536         /* save it, so mount count and last_transno is current */
 537         rc = filter_update_server_data(obd, filp, filter->fo_fsd, 1);
 538         if (rc)
 539                 GOTO(err_client, rc);
 540         RETURN(0);
 541
 542 err_client:
 543         class_disconnect_exports(obd, 0);
 544 err_fsd:
 545         filter_free_server_data(filter);
 546         RETURN(rc);
 547 }
 548
 549 static int filter_cleanup_groups(struct obd_device *obd)
 550 {
 551         struct filter_obd *filter = &obd->u.filter;
 552         struct dentry *dentry;
 553         int i, k;
 554         ENTRY;
 555
 556         for (i = 0; i < filter->fo_group_count; i++) {
 557                 if (filter->fo_subdirs != NULL) {
 558                         for (k = 0; k < filter->fo_subdir_count; k++) {
 559                                 dentry = filter->fo_subdirs[i].dentry[k];
 560                                 if (dentry == NULL)
 561                                         continue;
 562                                 f_dput(dentry);
 563                                 filter->fo_subdirs[i].dentry[k] = NULL;
 564                         }
 565                 }
 566                 if (filter->fo_last_objid_files[i] != NULL) {
 567                         filp_close(filter->fo_last_objid_files[i], 0);
 568                         filter->fo_last_objid_files[i] = NULL;
 569                 }
 570                 if (filter->fo_groups[i] != NULL) {
 571                         dput(filter->fo_groups[i]);
 572                         filter->fo_groups[i] = NULL;
 573                 }
 574         }
 575         if (filter->fo_subdirs != NULL)
 576                 OBD_FREE(filter->fo_subdirs,
 577                          filter->fo_group_count * sizeof(*filter->fo_subdirs));
 578         if (filter->fo_groups != NULL)
 579                 OBD_FREE(filter->fo_groups,
 580                          filter->fo_group_count * sizeof(*filter->fo_groups));
 581         if (filter->fo_last_objids != NULL)
 582                 OBD_FREE(filter->fo_last_objids,
 583                          filter->fo_group_count * sizeof(__u64));
 584         if (filter->fo_last_objid_files != NULL)
 585                 OBD_FREE(filter->fo_last_objid_files,
 586                          filter->fo_group_count * sizeof(struct file *));
 587         f_dput(filter->fo_dentry_O);
 588         RETURN(0);
 589 }
 590
 591 static int filter_update_last_group(struct obd_device *obd, int group)
 592 {
 593         struct filter_obd *filter = &obd->u.filter;
 594         struct file *filp = NULL;
 595         int last_group = 0, rc;
 596         loff_t off = 0;
 597         ENTRY;
 598
 599         if (group <= filter->fo_committed_group)
 600                 RETURN(0);
 601
 602         filp = filp_open("LAST_GROUP", O_RDWR, 0700);
 603         if (IS_ERR(filp)) {
 604                 rc = PTR_ERR(filp);
 605                 filp = NULL;
 606                 CERROR("cannot open LAST_GROUP: rc = %d\n", rc);
 607                 GOTO(cleanup, rc);
 608         }
 609
 610         rc = fsfilt_read_record(obd, filp, &last_group, sizeof(__u32), &off);
 611         if (rc) {
 612                 CDEBUG(D_INODE, "error reading LAST_GROUP: rc %d\n",rc);
 613                 GOTO(cleanup, rc);
 614         }
 615         LASSERT(off == 0 || last_group >= FILTER_MIN_GROUPS);
 616         CDEBUG(D_INODE, "%s: previous %d, new %d\n",
 617                obd->obd_name, last_group, group);
 618
 619         off = 0;
 620         last_group = group;
 621         /* must be sync: bXXXX */
 622         rc = fsfilt_write_record(obd, filp, &last_group, sizeof(__u32), &off, 1);
 623         if (rc) {
 624                 CDEBUG(D_INODE, "error updating LAST_GROUP: rc %d\n", rc);
 625                 GOTO(cleanup, rc);
 626         }
 627
 628         filter->fo_committed_group = group;
 629 cleanup:
 630         if (filp)
 631                 filp_close(filp, 0);
 632         RETURN(rc);
 633 }
 634
 635 static int filter_read_group_internal(struct obd_device *obd, int group,
 636                                       int create)
 637 {
 638         struct filter_obd *filter = &obd->u.filter;
 639         __u64 *new_objids = NULL;
 640         struct filter_subdirs *new_subdirs = NULL, *tmp_subdirs = NULL;
 641         struct dentry **new_groups = NULL;
 642         struct file **new_files = NULL;
 643         struct dentry *dentry;
 644         struct file *filp;
 645         int old_count = filter->fo_group_count, rc, stage = 0, i;
 646         char name[25];
 647         __u64 last_objid;
 648         loff_t off = 0;
 649
 650         snprintf(name, 24, "%d", group);
 651         name[24] = '\0';
 652
 653         if (!create) {
 654                 dentry = ll_lookup_one_len(name, filter->fo_dentry_O,
 655                                            strlen(name));
 656                 if (IS_ERR(dentry)) {
 657                         CERROR("Cannot lookup expected object group %d: %ld\n",
 658                                group, PTR_ERR(dentry));
 659                         RETURN(PTR_ERR(dentry));
 660                 }
 661         } else {
 662                 dentry = simple_mkdir(filter->fo_dentry_O, name, 0700, 1);
 663                 if (IS_ERR(dentry)) {
 664                         CERROR("cannot lookup/create O/%s: rc = %ld\n", name,
 665                                PTR_ERR(dentry));
 666                         RETURN(PTR_ERR(dentry));
 667                 }
 668         }
 669         stage = 1;
 670
 671         snprintf(name, 24, "O/%d/LAST_ID", group);
 672         name[24] = '\0';
 673         filp = filp_open(name, O_CREAT | O_RDWR, 0700);
 674         if (IS_ERR(filp)) {
 675                 CERROR("cannot create %s: rc = %ld\n", name, PTR_ERR(filp));
 676                 GOTO(cleanup, rc = PTR_ERR(filp));
 677         }
 678         stage = 2;
 679
 680         rc = fsfilt_read_record(obd, filp, &last_objid, sizeof(__u64), &off);
 681         if (rc) {
 682                 CDEBUG(D_INODE, "error reading %s: rc %d\n", name, rc);
 683                 GOTO(cleanup, rc);
 684         }
 685
 686         if (filter->fo_subdir_count) {
 687                 OBD_ALLOC(tmp_subdirs, sizeof(*tmp_subdirs));
 688                 if (tmp_subdirs == NULL)
 689                         GOTO(cleanup, rc = -ENOMEM);
 690                 stage = 3;
 691
 692                 for (i = 0; i < filter->fo_subdir_count; i++) {
 693                         char dir[20];
 694                         snprintf(dir, sizeof(dir), "d%u", i);
 695
 696                         tmp_subdirs->dentry[i] =
 697                                 simple_mkdir(dentry, dir, 0700, 1);
 698                         if (IS_ERR(tmp_subdirs->dentry[i])) {
 699                                 rc = PTR_ERR(tmp_subdirs->dentry[i]);
 700                                 CERROR("can't lookup/create O/%d/%s: rc = %d\n",
 701                                        group, dir, rc);
 702                                 GOTO(cleanup, rc);
 703                         }
 704                         CDEBUG(D_INODE, "got/created O/%d/%s: %p\n", group, dir,
 705                                tmp_subdirs->dentry[i]);
 706                 }
 707         }
 708
 709         /* 'group' is an index; we need an array of length 'group + 1' */
 710         if (group + 1 > old_count) {
 711                 int len = group + 1;
 712                 OBD_ALLOC(new_objids, len * sizeof(*new_objids));
 713                 OBD_ALLOC(new_subdirs, len * sizeof(*new_subdirs));
 714                 OBD_ALLOC(new_groups, len * sizeof(*new_groups));
 715                 OBD_ALLOC(new_files, len * sizeof(*new_files));
 716                 stage = 4;
 717                 if (new_objids == NULL || new_subdirs == NULL ||
 718                     new_groups == NULL || new_files == NULL)
 719                         GOTO(cleanup, rc = -ENOMEM);
 720
 721                 memcpy(new_objids, filter->fo_last_objids,
 722                        old_count * sizeof(*new_objids));
 723                 memcpy(new_subdirs, filter->fo_subdirs,
 724                        old_count * sizeof(*new_subdirs));
 725                 memcpy(new_groups, filter->fo_groups,
 726                        old_count * sizeof(*new_groups));
 727                 memcpy(new_files, filter->fo_last_objid_files,
 728                        old_count * sizeof(*new_files));
 729
 730                 if (old_count) {
 731                         OBD_FREE(filter->fo_last_objids,
 732                                  old_count * sizeof(*new_objids));
 733                         OBD_FREE(filter->fo_subdirs,
 734                                  old_count * sizeof(*new_subdirs));
 735                         OBD_FREE(filter->fo_groups,
 736                                  old_count * sizeof(*new_groups));
 737                         OBD_FREE(filter->fo_last_objid_files,
 738                                  old_count * sizeof(*new_files));
 739                 }
 740                 filter->fo_last_objids = new_objids;
 741                 filter->fo_subdirs = new_subdirs;
 742                 filter->fo_groups = new_groups;
 743                 filter->fo_last_objid_files = new_files;
 744                 filter->fo_group_count = len;
 745         }
 746
 747         filter->fo_groups[group] = dentry;
 748         filter->fo_last_objid_files[group] = filp;
 749         if (filter->fo_subdir_count) {
 750                 filter->fo_subdirs[group] = *tmp_subdirs;
 751                 OBD_FREE(tmp_subdirs, sizeof(*tmp_subdirs));
 752         }
 753
 754         filter_update_last_group(obd, group);
 755
 756         if (filp->f_dentry->d_inode->i_size == 0) {
 757                 filter->fo_last_objids[group] = FILTER_INIT_OBJID;
 758                 RETURN(0);
 759         }
 760
 761         filter->fo_last_objids[group] = le64_to_cpu(last_objid);
 762         CDEBUG(D_INODE, "%s: server last_objid group %d: "LPU64"\n",
 763                obd->obd_name, group, last_objid);
 764         RETURN(0);
 765  cleanup:
 766         switch (stage) {
 767         case 4:
 768                 if (new_objids != NULL)
 769                         OBD_FREE(new_objids, group * sizeof(*new_objids));
 770                 if (new_subdirs != NULL)
 771                         OBD_FREE(new_subdirs, group * sizeof(*new_subdirs));
 772                 if (new_groups != NULL)
 773                         OBD_FREE(new_groups, group * sizeof(*new_groups));
 774                 if (new_files != NULL)
 775                         OBD_FREE(new_files, group * sizeof(*new_files));
 776         case 3:
 777                 if (filter->fo_subdir_count) {
 778                         for (i = 0; i < filter->fo_subdir_count; i++) {
 779                                 if (tmp_subdirs->dentry[i] != NULL)
 780                                         dput(tmp_subdirs->dentry[i]);
 781                         }
 782                         OBD_FREE(tmp_subdirs, sizeof(*tmp_subdirs));
 783                 }
 784         case 2:
 785                 filp_close(filp, 0);
 786         case 1:
 787                 dput(dentry);
 788         }
 789         RETURN(rc);
 790 }
 791
 792 static int filter_read_groups(struct obd_device *obd, int last_group,
 793                               int create)
 794 {
 795         struct filter_obd *filter = &obd->u.filter;
 796         int old_count, group, rc = 0;
 797
 798         down(&filter->fo_init_lock);
 799         old_count = filter->fo_group_count;
 800         for (group = old_count; group <= last_group; group++) {
 801                 if (group == 0)
 802                         continue; /* no group zero */
 803
 804                 rc = filter_read_group_internal(obd, group, create);
 805                 if (rc != 0)
 806                         break;
 807         }
 808         up(&filter->fo_init_lock);
 809         return rc;
 810 }
 811
 812 static int filter_prep_groups(struct obd_device *obd)
 813 {
 814         struct filter_obd *filter = &obd->u.filter;
 815         struct dentry *dentry, *O_dentry;
 816         int rc = 0, cleanup_phase = 0;
 817         struct file *filp = NULL;
 818         int last_group;
 819         loff_t off = 0;
 820         ENTRY;
 821
 822         O_dentry = simple_mkdir(current->fs->pwd, "O", 0700, 1);
 823         CDEBUG(D_INODE, "got/created O: %p\n", O_dentry);
 824         if (IS_ERR(O_dentry)) {
 825                 rc = PTR_ERR(O_dentry);
 826                 CERROR("cannot open/create O: rc = %d\n", rc);
 827                 GOTO(cleanup, rc);
 828         }
 829         filter->fo_dentry_O = O_dentry;
 830         cleanup_phase = 1; /* O_dentry */
 831
 832         /* Lookup "R" to tell if we're on an old OST FS and need to convert
 833          * from O/R/<dir>/<objid> to O/0/<dir>/<objid>.  This can be removed
 834          * some time post 1.0 when all old-style OSTs have converted along
 835          * with the init_objid hack. */
 836         dentry = ll_lookup_one_len("R", O_dentry, 1);
 837         if (IS_ERR(dentry))
 838                 GOTO(cleanup, rc = PTR_ERR(dentry));
 839         if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
 840                 struct dentry *O0_dentry = lookup_one_len("0", O_dentry, 1);
 841                 ENTRY;
 842
 843                 CWARN("converting OST to new object layout\n");
 844                 if (IS_ERR(O0_dentry)) {
 845                         rc = PTR_ERR(O0_dentry);
 846                         CERROR("error looking up O/0: rc %d\n", rc);
 847                         GOTO(cleanup_R, rc);
 848                 }
 849
 850                 if (O0_dentry->d_inode) {
 851                         CERROR("Both O/R and O/0 exist. Fix manually.\n");
 852                         GOTO(cleanup_O0, rc = -EEXIST);
 853                 }
 854
 855                 down(&O_dentry->d_inode->i_sem);
 856                 rc = vfs_rename(O_dentry->d_inode, dentry,
 857                                 O_dentry->d_inode, O0_dentry);
 858                 up(&O_dentry->d_inode->i_sem);
 859
 860                 if (rc) {
 861                         CERROR("error renaming O/R to O/0: rc %d\n", rc);
 862                         GOTO(cleanup_O0, rc);
 863                 }
 864                 filter->fo_fsd->fsd_feature_incompat |=
 865                         cpu_to_le32(FILTER_INCOMPAT_GROUPS);
 866                 rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
 867                                                filter->fo_fsd, 1);
 868                 GOTO(cleanup_O0, rc);
 869
 870         cleanup_O0:
 871                 f_dput(O0_dentry);
 872         cleanup_R:
 873                 f_dput(dentry);
 874                 if (rc)
 875                         GOTO(cleanup, rc);
 876         } else {
 877                 f_dput(dentry);
 878         }
 879
 880         cleanup_phase = 2; /* groups */
 881
 882         /* we have to initialize all groups before first connections from
 883          * clients because they may send create/destroy for any group -bzzz */
 884         filp = filp_open("LAST_GROUP", O_CREAT | O_RDWR, 0700);
 885         if (IS_ERR(filp)) {
 886                 CERROR("cannot create LAST_GROUP: rc = %ld\n", PTR_ERR(filp));
 887                 GOTO(cleanup, rc = PTR_ERR(filp));
 888         }
 889         cleanup_phase = 3; /* filp */
 890
 891         rc = fsfilt_read_record(obd, filp, &last_group, sizeof(__u32), &off);
 892         if (rc) {
 893                 CDEBUG(D_INODE, "error reading LAST_GROUP: rc %d\n", rc);
 894                 GOTO(cleanup, rc);
 895         }
 896         if (off == 0) {
 897                 last_group = FILTER_MIN_GROUPS;
 898         } else {
 899                 LASSERT(last_group >= FILTER_MIN_GROUPS);
 900         }
 901
 902         CWARN("%s: initialize groups [%d,%d]\n", obd->obd_name,
 903               FILTER_MIN_GROUPS, last_group);
 904         filter->fo_committed_group = last_group;
 905         rc = filter_read_groups(obd, last_group, 1);
 906         if (rc)
 907                 GOTO(cleanup, rc);
 908
 909         filp_close(filp, 0);
 910         RETURN(0);
 911
 912  cleanup:
 913         switch (cleanup_phase) {
 914         case 3:
 915                 filp_close(filp, 0);
 916         case 2:
 917                 filter_cleanup_groups(obd);
 918         case 1:
 919                 f_dput(filter->fo_dentry_O);
 920                 filter->fo_dentry_O = NULL;
 921         default:
 922                 break;
 923         }
 924         return rc;
 925 }
 926
 927 /* setup the object store with correct subdirectories */
 928 static int filter_prep(struct obd_device *obd)
 929 {
 930         struct lvfs_run_ctxt saved;
 931         struct filter_obd *filter = &obd->u.filter;
 932         struct file *file;
 933         struct inode *inode;
 934         int rc = 0;
 935         ENTRY;
 936
 937         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 938         file = filp_open(LAST_RCVD, O_RDWR | O_CREAT | O_LARGEFILE, 0700);
 939         if (!file || IS_ERR(file)) {
 940                 rc = PTR_ERR(file);
 941                 CERROR("OBD filter: cannot open/create %s: rc = %d\n",
 942                        LAST_RCVD, rc);
 943                 GOTO(out, rc);
 944         }
 945
 946         if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
 947                 CERROR("%s is not a regular file!: mode = %o\n", LAST_RCVD,
 948                        file->f_dentry->d_inode->i_mode);
 949                 GOTO(err_filp, rc = -ENOENT);
 950         }
 951
 952         /* steal operations */
 953         inode = file->f_dentry->d_inode;
 954         filter->fo_fop = file->f_op;
 955         filter->fo_iop = inode->i_op;
 956         filter->fo_aops = inode->i_mapping->a_ops;
 957
 958         rc = filter_init_server_data(obd, file);
 959         if (rc) {
 960                 CERROR("cannot read %s: rc = %d\n", LAST_RCVD, rc);
 961                 GOTO(err_filp, rc);
 962         }
 963         filter->fo_rcvd_filp = file;
 964
 965         rc = filter_prep_groups(obd);
 966         if (rc)
 967                 GOTO(err_server_data, rc);
 968
 969  out:
 970         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 971
 972         return(rc);
 973
 974  err_server_data:
 975         //class_disconnect_exports(obd, 0);
 976         filter_free_server_data(filter);
 977  err_filp:
 978         if (filp_close(file, 0))
 979                 CERROR("can't close %s after error\n", LAST_RCVD);
 980         filter->fo_rcvd_filp = NULL;
 981         goto out;
 982 }
 983
 984 /* cleanup the filter: write last used object id to status file */
 985 static void filter_post(struct obd_device *obd)
 986 {
 987         struct lvfs_run_ctxt saved;
 988         struct filter_obd *filter = &obd->u.filter;
 989         int rc, i;
 990
 991         /* XXX: filter_update_lastobjid used to call fsync_dev.  It might be
 992          * best to start a transaction with h_sync, because we removed this
 993          * from lastobjid */
 994
 995         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 996         rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
 997                                        filter->fo_fsd, 0);
 998         if (rc)
 999                 CERROR("error writing server data: rc = %d\n", rc);
1000
1001         for (i = 1; i < filter->fo_group_count; i++) {
1002                 rc = filter_update_last_objid(obd, i,
1003                                              (i == filter->fo_group_count - 1));
1004                 if (rc)
1005                         CERROR("error writing group %d lastobjid: rc = %d\n",
1006                                i, rc);
1007         }
1008
1009         rc = filp_close(filter->fo_rcvd_filp, 0);
1010         filter->fo_rcvd_filp = NULL;
1011         if (rc)
1012                 CERROR("error closing %s: rc = %d\n", LAST_RCVD, rc);
1013
1014         filter_cleanup_groups(obd);
1015         filter_free_server_data(filter);
1016         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1017 }
1018
1019 static void filter_set_last_id(struct filter_obd *filter, int group, obd_id id)
1020 {
1021         LASSERT(filter->fo_fsd != NULL);
1022         LASSERT(group > 0);
1023         LASSERT(group < filter->fo_group_count);
1024
1025         spin_lock(&filter->fo_objidlock);
1026         filter->fo_last_objids[group] = id;
1027         spin_unlock(&filter->fo_objidlock);
1028 }
1029
1030 __u64 filter_last_id(struct filter_obd *filter, int group)
1031 {
1032         obd_id id;
1033         LASSERT(filter->fo_fsd != NULL);
1034         LASSERT(group > 0);
1035         LASSERT(group < filter->fo_group_count);
1036
1037         spin_lock(&filter->fo_objidlock);
1038         id = filter->fo_last_objids[group];
1039         spin_unlock(&filter->fo_objidlock);
1040
1041         return id;
1042 }
1043
1044 /* direct cut-n-paste of mds_blocking_ast() */
1045 static int filter_blocking_ast(struct ldlm_lock *lock,
1046                                struct ldlm_lock_desc *desc,
1047                                void *data, int flag)
1048 {
1049         int do_ast;
1050         ENTRY;
1051
1052         if (flag == LDLM_CB_CANCELING) {
1053                 /* Don't need to do anything here. */
1054                 RETURN(0);
1055         }
1056
1057         /* XXX layering violation!  -phil */
1058         l_lock(&lock->l_resource->lr_namespace->ns_lock);
1059         /* Get this: if filter_blocking_ast is racing with ldlm_intent_policy,
1060          * such that filter_blocking_ast is called just before l_i_p takes the
1061          * ns_lock, then by the time we get the lock, we might not be the
1062          * correct blocking function anymore.  So check, and return early, if
1063          * so. */
1064         if (lock->l_blocking_ast != filter_blocking_ast) {
1065                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
1066                 RETURN(0);
1067         }
1068
1069         lock->l_flags |= LDLM_FL_CBPENDING;
1070         do_ast = (!lock->l_readers && !lock->l_writers);
1071         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
1072
1073         if (do_ast) {
1074                 struct lustre_handle lockh;
1075                 int rc;
1076
1077                 LDLM_DEBUG(lock, "already unused, calling ldlm_cli_cancel");
1078                 ldlm_lock2handle(lock, &lockh);
1079                 rc = ldlm_cli_cancel(&lockh);
1080                 if (rc < 0)
1081                         CERROR("ldlm_cli_cancel: %d\n", rc);
1082         } else {
1083                 LDLM_DEBUG(lock, "Lock still has references, will be "
1084                            "cancelled later");
1085         }
1086         RETURN(0);
1087 }
1088
1089 extern void *lock_dir(struct inode *dir, struct qstr *name);
1090 extern void unlock_dir(struct inode *dir, void *lock);
1091
1092 static void * filter_lock_dentry(struct obd_device *obd,
1093                                  struct dentry *dparent, obd_id id)
1094 {
1095 #ifdef S_PDIROPS
1096         struct qstr qstr;
1097         char name[32];
1098         qstr.name = name;
1099         qstr.len = sprintf(name, LPU64, id);
1100         return lock_dir(dparent->d_inode, &qstr);
1101 #else
1102         down(&dparent->d_inode->i_sem);
1103 #endif
1104         return 0;
1105 }
1106
1107 /* We never dget the object parent, so DON'T dput it either */
1108 static void filter_parent_unlock(struct dentry *dparent, void *lock)
1109 {
1110 #ifdef S_PDIROPS
1111         LASSERT(lock != NULL);
1112         unlock_dir(dparent->d_inode, lock);
1113 #else
1114         up(&dparent->d_inode->i_sem);
1115 #endif
1116 }
1117
1118 /* We never dget the object parent, so DON'T dput it either */
1119 struct dentry *filter_parent(struct obd_device *obd, obd_gr group, obd_id objid)
1120 {
1121         struct filter_obd *filter = &obd->u.filter;
1122         LASSERT(group < filter->fo_group_count);
1123         LASSERT(group > 0);
1124
1125         if (filter->fo_subdir_count == 0)
1126                 return filter->fo_groups[group];
1127
1128         return filter->fo_subdirs[group].dentry[objid & (filter->fo_subdir_count - 1)];
1129 }
1130
1131 /* We never dget the object parent, so DON'T dput it either */
1132 struct dentry *filter_parent_lock(struct obd_device *obd, obd_gr group,
1133                                   obd_id objid, void **lock)
1134 {
1135         unsigned long now = jiffies;
1136         struct dentry *dparent = filter_parent(obd, group, objid);
1137
1138         if (IS_ERR(dparent))
1139                 return dparent;
1140
1141         LASSERT(dparent);
1142         LASSERT(dparent->d_inode);
1143
1144         *lock = filter_lock_dentry(obd, dparent, objid);
1145         fsfilt_check_slow(now, obd_timeout, "parent lock");
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                         CERROR("%s: error getting object "LPU64":"LPU64
1177                                " parent: rc %ld\n", obd->obd_name,
1178                                id, group, PTR_ERR(dparent));
1179                         RETURN(dparent);
1180                 }
1181         }
1182         CDEBUG(D_INODE, "looking up object O/%.*s/%s\n",
1183                dparent->d_name.len, dparent->d_name.name, name);
1184         dchild = /*ll_*/lookup_one_len(name, dparent, len);
1185         if (dir_dentry == NULL)
1186                 filter_parent_unlock(dparent, lock);
1187         if (IS_ERR(dchild)) {
1188                 CERROR("%s: child lookup error %ld\n", obd->obd_name,
1189                        PTR_ERR(dchild));
1190                 RETURN(dchild);
1191         }
1192
1193         if (dchild->d_inode != NULL && is_bad_inode(dchild->d_inode)) {
1194                 CERROR("%s: got bad inode "LPU64"\n", obd->obd_name, id);
1195                 f_dput(dchild);
1196                 RETURN(ERR_PTR(-ENOENT));
1197         }
1198
1199         CDEBUG(D_INODE, "got child objid %s: %p, count = %d\n",
1200                name, dchild, atomic_read(&dchild->d_count));
1201
1202         LASSERT(atomic_read(&dchild->d_count) > 0);
1203
1204         RETURN(dchild);
1205 }
1206
1207 static int filter_prepare_destroy(struct obd_device *obd, obd_id objid,
1208                                   obd_id group)
1209 {
1210         struct lustre_handle lockh;
1211         int flags = LDLM_AST_DISCARD_DATA, rc;
1212         struct ldlm_res_id res_id = { .name = { objid, 0, group, 0 } };
1213         ldlm_policy_data_t policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
1214
1215         ENTRY;
1216         /* Tell the clients that the object is gone now and that they should
1217          * throw away any cached pages. */
1218         rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, res_id,
1219                               LDLM_EXTENT, &policy, LCK_PW,
1220                               &flags, filter_blocking_ast, ldlm_completion_ast,
1221                               NULL, NULL, NULL, 0, NULL, &lockh);
1222
1223         /* We only care about the side-effects, just drop the lock. */
1224         if (rc == ELDLM_OK)
1225                 ldlm_lock_decref(&lockh, LCK_PW);
1226
1227         RETURN(rc);
1228 }
1229
1230 /* Caller must hold LCK_PW on parent and push us into kernel context.
1231  * Caller is also required to ensure that dchild->d_inode exists. */
1232 static int filter_destroy_internal(struct obd_device *obd, obd_id objid,
1233                                    struct dentry *dparent,
1234                                    struct dentry *dchild)
1235 {
1236         struct inode *inode = dchild->d_inode;
1237         int rc;
1238         ENTRY;
1239
1240         if (inode->i_nlink != 1 || atomic_read(&inode->i_count) != 1) {
1241                 CERROR("destroying objid %.*s nlink = %lu, count = %d\n",
1242                        dchild->d_name.len, dchild->d_name.name,
1243                        (unsigned long)inode->i_nlink,
1244                        atomic_read(&inode->i_count));
1245         }
1246
1247         rc = vfs_unlink(dparent->d_inode, dchild);
1248
1249         if (rc)
1250                 CERROR("error unlinking objid %.*s: rc %d\n",
1251                        dchild->d_name.len, dchild->d_name.name, rc);
1252
1253         RETURN(rc);
1254 }
1255
1256 static int filter_intent_policy(struct ldlm_namespace *ns,
1257                                 struct ldlm_lock **lockp, void *req_cookie,
1258                                 ldlm_mode_t mode, int flags, void *data)
1259 {
1260         struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
1261         struct ptlrpc_request *req = req_cookie;
1262         struct ldlm_lock *lock = *lockp, *l = NULL;
1263         struct ldlm_resource *res = lock->l_resource;
1264         ldlm_processing_policy policy;
1265         struct ost_lvb *res_lvb, *reply_lvb;
1266         struct ldlm_reply *rep;
1267         struct list_head *tmp;
1268         ldlm_error_t err;
1269         int tmpflags = 0, rc, repsize[2] = {sizeof(*rep), sizeof(*reply_lvb)};
1270         ENTRY;
1271
1272         policy = ldlm_get_processing_policy(res);
1273         LASSERT(policy != NULL);
1274         LASSERT(req != NULL);
1275
1276         rc = lustre_pack_reply(req, 2, repsize, NULL);
1277         if (rc)
1278                 RETURN(req->rq_status = rc);
1279
1280         rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*rep));
1281         LASSERT(rep != NULL);
1282
1283         reply_lvb = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*reply_lvb));
1284         LASSERT(reply_lvb != NULL);
1285
1286         //fixup_handle_for_resent_req(req, lock, &lockh);
1287
1288         /* If we grant any lock at all, it will be a whole-file read lock.
1289          * Call the extent policy function to see if our request can be
1290          * granted, or is blocked. */
1291         lock->l_policy_data.l_extent.start = 0;
1292         lock->l_policy_data.l_extent.end = OBD_OBJECT_EOF;
1293         lock->l_req_mode = LCK_PR;
1294
1295         l_lock(&res->lr_namespace->ns_lock);
1296
1297         res->lr_tmp = &rpc_list;
1298         rc = policy(lock, &tmpflags, 0, &err);
1299         res->lr_tmp = NULL;
1300
1301         /* FIXME: we should change the policy function slightly, to not make
1302          * this list at all, since we just turn around and free it */
1303         while (!list_empty(&rpc_list)) {
1304                 struct ldlm_ast_work *w =
1305                         list_entry(rpc_list.next, struct ldlm_ast_work, w_list);
1306                 list_del(&w->w_list);
1307                 LDLM_LOCK_PUT(w->w_lock);
1308                 OBD_FREE(w, sizeof(*w));
1309         }
1310
1311         if (rc == LDLM_ITER_CONTINUE) {
1312                 /* The lock met with no resistance; we're finished. */
1313                 l_unlock(&res->lr_namespace->ns_lock);
1314                 RETURN(ELDLM_LOCK_REPLACED);
1315         }
1316
1317         /* Do not grant any lock, but instead send GL callbacks.  The extent
1318          * policy nicely created a list of all PW locks for us.  We will choose
1319          * the highest of those which are larger than the size in the LVB, if
1320          * any, and perform a glimpse callback. */
1321         down(&res->lr_lvb_sem);
1322         res_lvb = res->lr_lvb_data;
1323         LASSERT(res_lvb != NULL);
1324         *reply_lvb = *res_lvb;
1325         up(&res->lr_lvb_sem);
1326
1327         list_for_each(tmp, &res->lr_granted) {
1328                 struct ldlm_lock *tmplock =
1329                         list_entry(tmp, struct ldlm_lock, l_res_link);
1330
1331                 if (tmplock->l_granted_mode == LCK_PR)
1332                         continue;
1333
1334                 if (tmplock->l_policy_data.l_extent.end <= reply_lvb->lvb_size)
1335                         continue;
1336
1337                 if (l == NULL) {
1338                         l = LDLM_LOCK_GET(tmplock);
1339                         continue;
1340                 }
1341
1342                 if (l->l_policy_data.l_extent.start >
1343                     tmplock->l_policy_data.l_extent.start)
1344                         continue;
1345
1346                 LDLM_LOCK_PUT(l);
1347                 l = LDLM_LOCK_GET(tmplock);
1348         }
1349         l_unlock(&res->lr_namespace->ns_lock);
1350
1351         /* There were no PW locks beyond the size in the LVB; finished. */
1352         if (l == NULL)
1353                 RETURN(ELDLM_LOCK_ABORTED);
1354
1355         if (l->l_glimpse_ast == NULL) {
1356                 /* We are racing with unlink(); just return -ENOENT */
1357                 rep->lock_policy_res1 = -ENOENT;
1358                 goto out;
1359         }
1360
1361         LASSERTF(l->l_glimpse_ast != NULL, "l == %p", l);
1362
1363         rc = l->l_glimpse_ast(l, NULL); /* this will update the LVB */
1364         if (rc != 0 && res->lr_namespace->ns_lvbo &&
1365             res->lr_namespace->ns_lvbo->lvbo_update) {
1366                 res->lr_namespace->ns_lvbo->lvbo_update(res, NULL, 0, 1);
1367         }
1368
1369         down(&res->lr_lvb_sem);
1370         *reply_lvb = *res_lvb;
1371         up(&res->lr_lvb_sem);
1372 out:
1373         LDLM_LOCK_PUT(l);
1374
1375         RETURN(ELDLM_LOCK_ABORTED);
1376 }
1377
1378 static int filter_post_fs_cleanup(struct obd_device *obd)
1379 {
1380         int rc = 0;
1381
1382         rc = fsfilt_post_cleanup(obd);
1383
1384         RETURN(rc);
1385 }
1386
1387 static int filter_group_set_fs_flags(struct obd_device *obd, int group)
1388 {
1389         struct filter_obd *filter = &obd->u.filter;
1390         int rc = 0, i = 0;
1391         ENTRY;
1392
1393         /* zero group is not longer valid. */
1394         if (group== 0)
1395                 RETURN(rc);
1396         for (i = 0; i < filter->fo_subdir_count; i++) {
1397                 struct dentry *dentry;
1398                 dentry = (filter->fo_subdirs + group)->dentry[i];
1399                 rc = fsfilt_set_fs_flags(obd, dentry->d_inode,
1400                                          SM_DO_REC | SM_DO_COW);
1401                 if (rc)
1402                         RETURN(rc);
1403         }
1404         RETURN(rc);
1405 }
1406 static int filter_post_fs_setup(struct obd_device *obd)
1407 {
1408         struct filter_obd *filter = &obd->u.filter;
1409         int rc = 0, j = 0;
1410         struct llog_ctxt *ctxt = NULL;
1411
1412         rc = fsfilt_post_setup(obd, filter->fo_dentry_O);
1413         if (rc)
1414                 RETURN(rc);
1415
1416         for (j = 0; j < filter->fo_group_count; j++) {
1417                 rc = filter_group_set_fs_flags(obd, j);
1418                 if (rc)
1419                         return rc;
1420         }
1421
1422         fsfilt_get_reint_log_ctxt(obd, filter->fo_sb, &ctxt);
1423         if (ctxt) {
1424                 ctxt->loc_obd = obd;
1425                 ctxt->loc_idx = LLOG_REINT_ORIG_CTXT;
1426                 obd->obd_llog_ctxt[LLOG_REINT_ORIG_CTXT] = ctxt;
1427         }
1428         fsfilt_set_ost_flags(obd, filter->fo_sb);
1429         return rc;
1430 }
1431
1432 /* mount the file system (secretly) */
1433 int filter_common_setup(struct obd_device *obd, obd_count len, void *buf,
1434                         char *option)
1435 {
1436         struct lustre_cfg* lcfg = buf;
1437         struct filter_obd *filter = &obd->u.filter;
1438         struct lvfs_obd_ctxt *lvfs_ctxt = NULL;
1439         struct vfsmount *mnt;
1440         char ns_name[48];
1441         int rc = 0, i;
1442         ENTRY;
1443
1444
1445         if (!lcfg->lcfg_inlbuf1 || !lcfg->lcfg_inlbuf2)
1446                 RETURN(-EINVAL);
1447
1448         obd->obd_fsops = fsfilt_get_ops(lcfg->lcfg_inlbuf2);
1449         if (IS_ERR(obd->obd_fsops))
1450                 RETURN(PTR_ERR(obd->obd_fsops));
1451
1452         rc = lvfs_mount_fs(lcfg->lcfg_inlbuf1, lcfg->lcfg_inlbuf2,
1453                            option, MS_NOATIME | MS_NODIRATIME, &lvfs_ctxt);
1454         if (rc) {
1455                 CERROR("lvfs_mount_fs failed: rc = %d\n", rc);
1456                 GOTO(err_ops, rc);
1457         }
1458         LASSERT(lvfs_ctxt);
1459
1460         mnt = lvfs_ctxt->loc_mnt;
1461         filter->fo_lvfs_ctxt = lvfs_ctxt;
1462
1463         if (lcfg->lcfg_inllen3 > 0 && lcfg->lcfg_inlbuf3) {
1464                 if (*lcfg->lcfg_inlbuf3 == 'f') {
1465                         obd->obd_replayable = 1;
1466                         obd_sync_filter = 1;
1467                         CWARN("%s: recovery enabled\n", obd->obd_name);
1468                 } else {
1469                         if (*lcfg->lcfg_inlbuf3 != 'n') {
1470                                 CERROR("unrecognised flag '%c'\n",
1471                                        *lcfg->lcfg_inlbuf3);
1472                         }
1473                         // XXX Robert? Why do we get errors here
1474                         // GOTO(err_mntput, rc = -EINVAL);
1475                 }
1476         }
1477
1478         filter->fo_vfsmnt = mnt;
1479         filter->fo_sb = mnt->mnt_sb;
1480         filter->fo_fstype = mnt->mnt_sb->s_type->name;
1481         CDEBUG(D_SUPER, "%s: mnt = %p\n", filter->fo_fstype, mnt);
1482
1483         OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1484         obd->obd_lvfs_ctxt.pwdmnt = mnt;
1485         obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
1486         obd->obd_lvfs_ctxt.fs = get_ds();
1487         obd->obd_lvfs_ctxt.cb_ops = filter_lvfs_ops;
1488
1489         ll_clear_rdonly(ll_sbdev(filter->fo_sb));
1490
1491         rc = fsfilt_setup(obd, mnt->mnt_sb);
1492         if (rc)
1493                 GOTO(err_mntput, rc);
1494
1495         sema_init(&filter->fo_init_lock, 1);
1496         filter->fo_committed_group = 0;
1497         rc = filter_prep(obd);
1498         if (rc)
1499                 GOTO(err_mntput, rc);
1500
1501         filter->fo_destroys_in_progress = 0;
1502         for (i = 0; i < 32; i++)
1503                 sema_init(&filter->fo_create_locks[i], 1);
1504
1505         spin_lock_init(&filter->fo_translock);
1506         spin_lock_init(&filter->fo_objidlock);
1507         INIT_LIST_HEAD(&filter->fo_export_list);
1508         sema_init(&filter->fo_alloc_lock, 1);
1509         spin_lock_init(&filter->fo_r_pages.oh_lock);
1510         spin_lock_init(&filter->fo_w_pages.oh_lock);
1511         spin_lock_init(&filter->fo_r_discont_pages.oh_lock);
1512         spin_lock_init(&filter->fo_w_discont_pages.oh_lock);
1513         spin_lock_init(&filter->fo_r_discont_blocks.oh_lock);
1514         spin_lock_init(&filter->fo_w_discont_blocks.oh_lock);
1515         filter->fo_readcache_max_filesize = FILTER_MAX_CACHE_SIZE;
1516
1517         INIT_LIST_HEAD(&filter->fo_llog_list);
1518         spin_lock_init(&filter->fo_llog_list_lock);
1519
1520         sprintf(ns_name, "filter-%s", obd->obd_uuid.uuid);
1521         obd->obd_namespace = ldlm_namespace_new(ns_name, LDLM_NAMESPACE_SERVER);
1522
1523         if (obd->obd_namespace == NULL)
1524                 GOTO(err_post, rc = -ENOMEM);
1525         obd->obd_namespace->ns_lvbp = obd;
1526         obd->obd_namespace->ns_lvbo = &filter_lvbo;
1527         ldlm_register_intent(obd->obd_namespace, filter_intent_policy);
1528
1529         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1530                            "filter_ldlm_cb_client", &obd->obd_ldlm_client);
1531
1532         rc = obd_llog_cat_initialize(obd, &obd->obd_llogs, 1, CATLIST);
1533         if (rc) {
1534                 CERROR("failed to setup llogging subsystems\n");
1535                 GOTO(err_post, rc);
1536         }
1537         RETURN(0);
1538
1539 err_post:
1540         filter_post(obd);
1541 err_mntput:
1542         unlock_kernel();
1543         lvfs_umount_fs(filter->fo_lvfs_ctxt);
1544         filter->fo_sb = 0;
1545         lock_kernel();
1546 err_ops:
1547         fsfilt_put_ops(obd->obd_fsops);
1548         return rc;
1549 }
1550
1551 static int filter_attach(struct obd_device *obd, obd_count len, void *data)
1552 {
1553         struct lprocfs_static_vars lvars;
1554         int rc;
1555
1556         lprocfs_init_vars(filter, &lvars);
1557         rc = lprocfs_obd_attach(obd, lvars.obd_vars);
1558         if (rc != 0)
1559                 return rc;
1560
1561         rc = lprocfs_alloc_obd_stats(obd, LPROC_FILTER_LAST);
1562         if (rc != 0)
1563                 return rc;
1564
1565         /* Init obdfilter private stats here */
1566         lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_READ_BYTES,
1567                              LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes");
1568         lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_WRITE_BYTES,
1569                              LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes");
1570
1571         return lproc_filter_attach_seqstat(obd);
1572 }
1573
1574 static int filter_detach(struct obd_device *dev)
1575 {
1576         lprocfs_free_obd_stats(dev);
1577         return lprocfs_obd_detach(dev);
1578 }
1579
1580 static int filter_setup(struct obd_device *obd, obd_count len, void *buf)
1581 {
1582         struct lustre_cfg* lcfg = buf;
1583         int rc;
1584         ENTRY;
1585         /* all mount options including errors=remount-ro and asyncdel are passed
1586          * using 4th lcfg param. And it is good, finally we have got rid of
1587          * hardcoded fs types in the code. */
1588         rc = filter_common_setup(obd, len, buf, lcfg->lcfg_inlbuf4);
1589         if (rc)
1590                 RETURN(rc);
1591         rc = filter_post_fs_setup(obd);
1592         RETURN(rc);
1593 }
1594
1595 static int filter_cleanup(struct obd_device *obd, int flags)
1596 {
1597         struct filter_obd *filter = &obd->u.filter;
1598         ll_sbdev_type save_dev;
1599         ENTRY;
1600
1601         if (flags & OBD_OPT_FAILOVER)
1602                 CERROR("%s: shutting down for failover; client state will"
1603                        " be preserved.\n", obd->obd_name);
1604
1605         if (!list_empty(&obd->obd_exports)) {
1606                 CERROR("%s: still has clients!\n", obd->obd_name);
1607                 class_disconnect_exports(obd, flags);
1608                 if (!list_empty(&obd->obd_exports)) {
1609                         CERROR("still has exports after forced cleanup?\n");
1610                         RETURN(-EBUSY);
1611                 }
1612         }
1613
1614         target_cleanup_recovery(obd);
1615
1616         ldlm_namespace_free(obd->obd_namespace, flags & OBD_OPT_FORCE);
1617
1618         if (filter->fo_sb == NULL)
1619                 RETURN(0);
1620
1621         save_dev = ll_sbdev(filter->fo_sb);
1622         filter_post_fs_cleanup(obd);
1623         filter_post(obd);
1624
1625         shrink_dcache_parent(filter->fo_sb->s_root);
1626         filter->fo_sb = 0;
1627
1628         unlock_kernel();
1629         lvfs_umount_fs(filter->fo_lvfs_ctxt);
1630         //destroy_buffers(filter->fo_sb->s_dev);
1631         filter->fo_sb = NULL;
1632         fsfilt_put_ops(obd->obd_fsops);
1633         lock_kernel();
1634
1635         ll_clear_rdonly(save_dev);
1636
1637         RETURN(0);
1638 }
1639
1640 static int filter_connect_post(struct obd_export *exp, unsigned initial,
1641                                unsigned long connect_flags)
1642 {
1643         struct obd_device *obd = exp->exp_obd;
1644         struct filter_export_data *fed;
1645         char str[PTL_NALFMT_SIZE];
1646         struct obd_llogs *llog;
1647         struct llog_ctxt *ctxt;
1648         int rc;
1649         ENTRY;
1650
1651         fed = &exp->exp_filter_data;
1652         if (fed->fed_group < FILTER_MIN_GROUPS)
1653                 RETURN(0);
1654
1655         /* initialize llogs for connections from MDS */
1656         llog = filter_grab_llog_for_group(obd, fed->fed_group, exp);
1657         LASSERT(llog != NULL);
1658
1659         ctxt = llog_get_context(llog, LLOG_UNLINK_REPL_CTXT);
1660         LASSERT(ctxt != NULL);
1661
1662         rc = llog_receptor_accept(ctxt, exp->exp_imp_reverse);
1663         portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
1664                         exp->exp_connection->c_peer.peer_id.nid, str);
1665         CDEBUG(D_OTHER, "%s: init llog ctxt for export "LPX64"/%s, group %d\n",
1666                obd->obd_name, exp->exp_connection->c_peer.peer_id.nid,
1667                str, fed->fed_group);
1668
1669         RETURN(rc);
1670 }
1671
1672 /* nearly identical to mds_connect */
1673 static int filter_connect(struct lustre_handle *conn, struct obd_device *obd,
1674                           struct obd_uuid *cluuid,
1675                           struct obd_connect_data *data,
1676                           unsigned long connect_flags)
1677 {
1678         struct obd_export *exp;
1679         struct filter_export_data *fed;
1680         struct filter_client_data *fcd = NULL;
1681         struct filter_obd *filter = &obd->u.filter;
1682         struct lvfs_run_ctxt saved;
1683         int rc, group;
1684         ENTRY;
1685
1686         if (conn == NULL || obd == NULL || cluuid == NULL)
1687                 RETURN(-EINVAL);
1688
1689         rc = class_connect(conn, obd, cluuid);
1690         if (rc)
1691                 RETURN(rc);
1692         exp = class_conn2export(conn);
1693         LASSERT(exp != NULL);
1694
1695         fed = &exp->exp_filter_data;
1696
1697         spin_lock_init(&fed->fed_lock);
1698
1699        /* connection from MDS */
1700         group = connect_flags;
1701         if (obd->obd_replayable) {
1702                 OBD_ALLOC(fcd, sizeof(*fcd));
1703                 if (!fcd) {
1704                         CERROR("filter: out of memory for client data\n");
1705                         GOTO(cleanup, rc = -ENOMEM);
1706                 }
1707
1708                 memcpy(fcd->fcd_uuid, cluuid, sizeof(fcd->fcd_uuid));
1709                 fed->fed_fcd = fcd;
1710                 fed->fed_fcd->fcd_group = group;
1711                 rc = filter_client_add(obd, filter, fed, -1);
1712                 if (rc)
1713                         GOTO(cleanup, rc);
1714         }
1715         CWARN("%s: Received MDS connection ("LPX64"); group %d\n",
1716               obd->obd_name, exp->exp_handle.h_cookie, group);
1717
1718         if (group == 0)
1719                 GOTO(cleanup, rc);
1720
1721         if (fed->fed_group != 0 && fed->fed_group != group) {
1722                 char str[PTL_NALFMT_SIZE];
1723                 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
1724                                 exp->exp_connection->c_peer.peer_id.nid, str);
1725                 CERROR("!!! This export (nid "LPX64"/%s) used object group %d "
1726                        "earlier; now it's trying to use group %d!  This could "
1727                        "be a bug in the MDS.  Tell CFS.\n",
1728                        exp->exp_connection->c_peer.peer_id.nid, str,
1729                        fed->fed_group, group);
1730                 GOTO(cleanup, rc = -EPROTO);
1731         }
1732         fed->fed_group = group;
1733
1734         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1735         rc = filter_read_groups(obd, group, 1);
1736         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1737         if (rc != 0) {
1738                 CERROR("can't read group %u\n", group);
1739                 GOTO(cleanup, rc);
1740         }
1741         rc = filter_group_set_fs_flags(obd, group);
1742         if (rc != 0) {
1743                 CERROR("can't set kml flags %u\n", group);
1744                 GOTO(cleanup, rc);
1745         }
1746
1747 cleanup:
1748         if (rc) {
1749                 if (fcd)
1750                         OBD_FREE(fcd, sizeof(*fcd));
1751                 class_disconnect(exp, 0);
1752         } else {
1753                 class_export_put(exp);
1754         }
1755         return rc;
1756 }
1757
1758 static int filter_precleanup(struct obd_device *obd, int flags)
1759 {
1760         struct filter_group_llog *log;
1761         struct filter_obd *filter;
1762         int rc = 0;
1763         ENTRY;
1764
1765         filter = &obd->u.filter;
1766
1767         spin_lock(&filter->fo_llog_list_lock);
1768         while (!list_empty(&filter->fo_llog_list)) {
1769                 log = list_entry(filter->fo_llog_list.next,
1770                                  struct filter_group_llog, list);
1771                 list_del(&log->list);
1772                 spin_unlock(&filter->fo_llog_list_lock);
1773
1774                 rc = obd_llog_finish(obd, log->llogs, 0);
1775                 if (rc)
1776                         CERROR("failed to cleanup llogging subsystem for %u\n",
1777                                 log->group);
1778                 OBD_FREE(log->llogs, sizeof(*(log->llogs)));
1779                 OBD_FREE(log, sizeof(*log));
1780                 spin_lock(&filter->fo_llog_list_lock);
1781         }
1782         spin_unlock(&filter->fo_llog_list_lock);
1783
1784         rc = obd_llog_finish(obd, &obd->obd_llogs, 0);
1785         if (rc)
1786                 CERROR("failed to cleanup llogging subsystem\n");
1787
1788         RETURN(rc);
1789 }
1790
1791 /* Do extra sanity checks for grant accounting.  We do this at connect,
1792  * disconnect, and statfs RPC time, so it shouldn't be too bad.  We can
1793  * always get rid of it or turn it off when we know accounting is good. */
1794 static void filter_grant_sanity_check(struct obd_device *obd, const char *func)
1795 {
1796         struct filter_export_data *fed;
1797         struct obd_export *exp;
1798         obd_size maxsize = obd->obd_osfs.os_blocks * obd->obd_osfs.os_bsize;
1799         obd_size tot_dirty = 0, tot_pending = 0, tot_granted = 0;
1800         obd_size fo_tot_dirty, fo_tot_pending, fo_tot_granted;
1801         int level = D_CACHE;
1802
1803         if (list_empty(&obd->obd_exports))
1804                 return;
1805
1806         spin_lock(&obd->obd_osfs_lock);
1807         spin_lock(&obd->obd_dev_lock);
1808         list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
1809                 fed = &exp->exp_filter_data;
1810                 if (fed->fed_grant < 0 || fed->fed_pending < 0 ||
1811                     fed->fed_dirty < 0)
1812                         level = D_ERROR;
1813                 if (maxsize > 0) { /* we may not have done a statfs yet */
1814                         LASSERTF(fed->fed_grant + fed->fed_pending <= maxsize,
1815                                  "cli %s/%p %ld+%ld > "LPU64"\n",
1816                                  exp->exp_client_uuid.uuid, exp,
1817                                  fed->fed_grant, fed->fed_pending, maxsize);
1818                         LASSERTF(fed->fed_dirty <= maxsize,
1819                                  "cli %s/%p %ld > "LPU64"\n",
1820                                  exp->exp_client_uuid.uuid, exp,
1821                                  fed->fed_dirty, maxsize);
1822                 }
1823                 CDEBUG(level, "%s: cli %s/%p dirty %ld pend %ld grant %ld\n",
1824                        obd->obd_name, exp->exp_client_uuid.uuid, exp,
1825                        fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1826                 tot_granted += fed->fed_grant + fed->fed_pending;
1827                 tot_pending += fed->fed_pending;
1828                 tot_dirty += fed->fed_dirty;
1829         }
1830         fo_tot_granted = obd->u.filter.fo_tot_granted;
1831         fo_tot_pending = obd->u.filter.fo_tot_pending;
1832         fo_tot_dirty = obd->u.filter.fo_tot_dirty;
1833         spin_unlock(&obd->obd_dev_lock);
1834         spin_unlock(&obd->obd_osfs_lock);
1835
1836         /* Do these assertions outside the spinlocks so we don't kill system */
1837         if (tot_granted != fo_tot_granted)
1838                 CERROR("%s: tot_granted "LPU64" != fo_tot_granted "LPU64"\n",
1839                        func, tot_granted, fo_tot_granted);
1840         if (tot_pending != fo_tot_pending)
1841                 CERROR("%s: tot_pending "LPU64" != fo_tot_pending "LPU64"\n",
1842                        func, tot_pending, fo_tot_pending);
1843         if (tot_dirty != fo_tot_dirty)
1844                 CERROR("%s: tot_dirty "LPU64" != fo_tot_dirty "LPU64"\n",
1845                        func, tot_dirty, fo_tot_dirty);
1846         if (tot_pending > tot_granted)
1847                 CERROR("%s: tot_pending "LPU64" > tot_granted "LPU64"\n",
1848                        func, tot_pending, tot_granted);
1849         if (tot_granted > maxsize)
1850                 CERROR("%s: tot_granted "LPU64" > maxsize "LPU64"\n",
1851                        func, tot_granted, maxsize);
1852         if (tot_dirty > maxsize)
1853                 CERROR("%s: tot_dirty "LPU64" > maxsize "LPU64"\n",
1854                        func, tot_dirty, maxsize);
1855 }
1856
1857 /* Remove this client from the grant accounting totals.  We also remove
1858  * the export from the obd device under the osfs and dev locks to ensure
1859  * that the filter_grant_sanity_check() calculations are always valid.
1860  * The client should do something similar when it invalidates its import. */
1861 static void filter_grant_discard(struct obd_export *exp)
1862 {
1863         struct obd_device *obd = exp->exp_obd;
1864         struct filter_obd *filter = &obd->u.filter;
1865         struct filter_export_data *fed = &exp->exp_filter_data;
1866         int level = D_CACHE;
1867
1868         spin_lock(&obd->obd_osfs_lock);
1869         spin_lock(&exp->exp_obd->obd_dev_lock);
1870         list_del_init(&exp->exp_obd_chain);
1871         spin_unlock(&exp->exp_obd->obd_dev_lock);
1872
1873         if (fed->fed_dirty < 0 || fed->fed_grant < 0 || fed->fed_pending < 0)
1874                 level = D_ERROR;
1875         CDEBUG(level, "%s: cli %s/%p dirty %ld pend %ld grant %ld\n",
1876                obd->obd_name, exp->exp_client_uuid.uuid, exp,
1877                fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1878
1879         LASSERTF(filter->fo_tot_granted >= fed->fed_grant,
1880                  "%s: tot_granted "LPU64" cli %s/%p fed_grant %ld\n",
1881                  obd->obd_name, filter->fo_tot_granted,
1882                  exp->exp_client_uuid.uuid, exp, fed->fed_grant);
1883         filter->fo_tot_granted -= fed->fed_grant;
1884         LASSERTF(filter->fo_tot_pending >= fed->fed_pending,
1885                  "%s: tot_pending "LPU64" cli %s/%p fed_pending %ld\n",
1886                  obd->obd_name, filter->fo_tot_pending,
1887                  exp->exp_client_uuid.uuid, exp, fed->fed_pending);
1888         LASSERTF(filter->fo_tot_dirty >= fed->fed_dirty,
1889                  "%s: tot_dirty "LPU64" cli %s/%p fed_dirty %ld\n",
1890                  obd->obd_name, filter->fo_tot_dirty,
1891                  exp->exp_client_uuid.uuid, exp, fed->fed_dirty);
1892         filter->fo_tot_dirty -= fed->fed_dirty;
1893         fed->fed_dirty = 0;
1894         fed->fed_grant = 0;
1895
1896         spin_unlock(&obd->obd_osfs_lock);
1897 }
1898
1899 static int filter_destroy_export(struct obd_export *exp)
1900 {
1901         ENTRY;
1902
1903         if (exp->exp_filter_data.fed_pending)
1904                 CERROR("%s: cli %s/%p has %lu pending on destroyed export\n",
1905                        exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
1906                        exp, exp->exp_filter_data.fed_pending);
1907
1908         target_destroy_export(exp);
1909
1910         if (exp->exp_obd->obd_replayable)
1911                 filter_client_free(exp, exp->exp_flags);
1912
1913         filter_grant_discard(exp);
1914         if (!(exp->exp_flags & OBD_OPT_FORCE))
1915                 filter_grant_sanity_check(exp->exp_obd, __FUNCTION__);
1916
1917         RETURN(0);
1918 }
1919
1920 static void filter_sync_llogs(struct obd_device *obd, struct obd_export *dexp)
1921 {
1922         struct filter_group_llog *fglog, *nlog;
1923         struct filter_obd *filter;
1924         int worked = 0, group;
1925         struct llog_ctxt *ctxt;
1926         ENTRY;
1927
1928         filter = &obd->u.filter;
1929
1930         /* we can't sync log holding spinlock. also, we do not want to get
1931          * into livelock. so we do following: loop over MDS's exports in
1932          * group order and skip already synced llogs -bzzz */
1933         do {
1934                 /* look for group with min. number, but > worked */
1935                 fglog = NULL;
1936                 group = 1 << 30;
1937                 spin_lock(&filter->fo_llog_list_lock);
1938                 list_for_each_entry(nlog, &filter->fo_llog_list, list) {
1939
1940                         if (nlog->group <= worked) {
1941                                 /* this group is already synced */
1942                                 continue;
1943                         }
1944
1945                         if (group < nlog->group) {
1946                                 /* we have group with smaller number to sync */
1947                                 continue;
1948                         }
1949
1950                         /* store current minimal group */
1951                         fglog = nlog;
1952                         group = nlog->group;
1953                 }
1954                 spin_unlock(&filter->fo_llog_list_lock);
1955
1956                 if (fglog == NULL)
1957                         break;
1958
1959                 worked = fglog->group;
1960                 if (fglog->exp && (dexp == fglog->exp || dexp == NULL)) {
1961                         ctxt = llog_get_context(fglog->llogs,
1962                                         LLOG_UNLINK_REPL_CTXT);
1963                         LASSERT(ctxt != NULL);
1964                         llog_sync(ctxt, fglog->exp);
1965                 }
1966         } while (fglog != NULL);
1967 }
1968
1969 /* also incredibly similar to mds_disconnect */
1970 static int filter_disconnect(struct obd_export *exp, unsigned long flags)
1971 {
1972         struct obd_device *obd = exp->exp_obd;
1973         unsigned long irqflags;
1974         int rc;
1975         ENTRY;
1976
1977         LASSERT(exp);
1978         class_export_get(exp);
1979
1980         spin_lock_irqsave(&exp->exp_lock, irqflags);
1981         exp->exp_flags = flags;
1982         spin_unlock_irqrestore(&exp->exp_lock, irqflags);
1983
1984         if (!(flags & OBD_OPT_FORCE))
1985                 filter_grant_sanity_check(obd, __FUNCTION__);
1986         filter_grant_discard(exp);
1987
1988         /* Disconnect early so that clients can't keep using export */
1989         rc = class_disconnect(exp, flags);
1990
1991         ldlm_cancel_locks_for_export(exp);
1992
1993         fsfilt_sync(obd, obd->u.filter.fo_sb);
1994
1995         /* flush any remaining cancel messages out to the target */
1996         filter_sync_llogs(obd, exp);
1997         class_export_put(exp);
1998         RETURN(rc);
1999 }
2000
2001 struct dentry *__filter_oa2dentry(struct obd_device *obd,
2002                                   struct obdo *oa, const char *what)
2003 {
2004         struct dentry *dchild = NULL;
2005         obd_gr group = 0;
2006
2007         if (oa->o_valid & OBD_MD_FLGROUP)
2008                 group = oa->o_gr;
2009
2010         dchild = filter_id2dentry(obd, NULL, group, oa->o_id);
2011
2012         if (IS_ERR(dchild)) {
2013                 CERROR("%s error looking up object: "LPU64"\n",
2014                        what, oa->o_id);
2015                 RETURN(dchild);
2016         }
2017
2018         if (dchild->d_inode == NULL) {
2019                 CERROR("%s: %s on non-existent object: "LPU64"\n",
2020                        obd->obd_name, what, oa->o_id);
2021                 f_dput(dchild);
2022                 RETURN(ERR_PTR(-ENOENT));
2023         }
2024
2025         return dchild;
2026 }
2027
2028 static int filter_getattr(struct obd_export *exp, struct obdo *oa,
2029                           struct lov_stripe_md *md)
2030 {
2031         struct dentry *dentry = NULL;
2032         struct obd_device *obd;
2033         int rc = 0;
2034         ENTRY;
2035
2036         obd = class_exp2obd(exp);
2037         if (obd == NULL) {
2038                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2039                        exp->exp_handle.h_cookie);
2040                 RETURN(-EINVAL);
2041         }
2042
2043         dentry = filter_oa2dentry(obd, oa);
2044         if (IS_ERR(dentry))
2045                 RETURN(PTR_ERR(dentry));
2046
2047         /* Limit the valid bits in the return data to what we actually use */
2048         oa->o_valid = OBD_MD_FLID;
2049         obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2050
2051         f_dput(dentry);
2052         RETURN(rc);
2053 }
2054
2055 /* this is called from filter_truncate() until we have filter_punch() */
2056 static int filter_setattr(struct obd_export *exp, struct obdo *oa,
2057                           struct lov_stripe_md *md, struct obd_trans_info *oti)
2058 {
2059         struct lvfs_run_ctxt saved;
2060         struct filter_obd *filter;
2061         struct dentry *dentry;
2062         struct iattr iattr;
2063         struct ldlm_res_id res_id = { .name = { oa->o_id, 0, oa->o_gr, 0 } };
2064         struct ldlm_resource *res;
2065         void *handle;
2066         int rc, rc2;
2067         ENTRY;
2068
2069         LASSERT(oti != NULL);
2070
2071         dentry = filter_oa2dentry(exp->exp_obd, oa);
2072         if (IS_ERR(dentry))
2073                 RETURN(PTR_ERR(dentry));
2074
2075         filter = &exp->exp_obd->u.filter;
2076
2077         iattr_from_obdo(&iattr, oa, oa->o_valid);
2078
2079         push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2080         lock_kernel();
2081
2082         if (iattr.ia_valid & ATTR_SIZE)
2083                 down(&dentry->d_inode->i_sem);
2084         handle = fsfilt_start(exp->exp_obd, dentry->d_inode, FSFILT_OP_SETATTR,
2085                               oti);
2086         if (IS_ERR(handle))
2087                 GOTO(out_unlock, rc = PTR_ERR(handle));
2088
2089         /* XXX this could be a rwsem instead, if filter_preprw played along */
2090         if (iattr.ia_valid & ATTR_ATTR_FLAG)
2091                 rc = fsfilt_iocontrol(exp->exp_obd, dentry->d_inode, NULL,
2092                                       EXT3_IOC_SETFLAGS,
2093                                       (long)&iattr.ia_attr_flags);
2094         else
2095                 rc = fsfilt_setattr(exp->exp_obd, dentry, handle, &iattr, 1);
2096         rc = filter_finish_transno(exp, oti, rc);
2097         rc2 = fsfilt_commit(exp->exp_obd, filter->fo_sb, dentry->d_inode,
2098                             handle, exp->exp_sync);
2099         if (rc2) {
2100                 CERROR("error on commit, err = %d\n", rc2);
2101                 if (!rc)
2102                         rc = rc2;
2103         }
2104
2105         res = ldlm_resource_get(exp->exp_obd->obd_namespace, NULL,
2106                                 res_id, LDLM_EXTENT, 0);
2107         if (res != NULL) {
2108                 if (res->lr_namespace->ns_lvbo &&
2109                     res->lr_namespace->ns_lvbo->lvbo_update)
2110                         rc = res->lr_namespace->ns_lvbo->lvbo_update(res, NULL,
2111                                                                      0, 0);
2112                 ldlm_resource_putref(res);
2113         } else if (iattr.ia_valid & ATTR_SIZE) {
2114                 /* called from MDS. */
2115         }
2116
2117         oa->o_valid = OBD_MD_FLID;
2118         obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2119
2120 out_unlock:
2121         if (iattr.ia_valid & ATTR_SIZE)
2122                 up(&dentry->d_inode->i_sem);
2123         unlock_kernel();
2124         pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2125
2126         f_dput(dentry);
2127         RETURN(rc);
2128 }
2129
2130 /* XXX identical to osc_unpackmd */
2131 static int filter_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2132                            struct lov_mds_md *lmm, int lmm_bytes)
2133 {
2134         int lsm_size;
2135         ENTRY;
2136
2137         if (lmm != NULL) {
2138                 if (lmm_bytes < sizeof (*lmm)) {
2139                         CERROR("lov_mds_md too small: %d, need %d\n",
2140                                lmm_bytes, (int)sizeof(*lmm));
2141                         RETURN(-EINVAL);
2142                 }
2143                 /* XXX LOV_MAGIC etc check? */
2144
2145                 if (lmm->lmm_object_id == cpu_to_le64(0)) {
2146                         CERROR("lov_mds_md: zero lmm_object_id\n");
2147                         RETURN(-EINVAL);
2148                 }
2149         }
2150
2151         lsm_size = lov_stripe_md_size(1);
2152         if (lsmp == NULL)
2153                 RETURN(lsm_size);
2154
2155         if (*lsmp != NULL && lmm == NULL) {
2156                 OBD_FREE(*lsmp, lsm_size);
2157                 *lsmp = NULL;
2158                 RETURN(0);
2159         }
2160
2161         if (*lsmp == NULL) {
2162                 OBD_ALLOC(*lsmp, lsm_size);
2163                 if (*lsmp == NULL)
2164                         RETURN(-ENOMEM);
2165
2166                 loi_init((*lsmp)->lsm_oinfo);
2167         }
2168
2169         if (lmm != NULL) {
2170                 /* XXX zero *lsmp? */
2171                 (*lsmp)->lsm_object_id = le64_to_cpu (lmm->lmm_object_id);
2172                 LASSERT((*lsmp)->lsm_object_id);
2173         }
2174
2175         (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
2176
2177         RETURN(lsm_size);
2178 }
2179
2180 static void filter_destroy_precreated(struct obd_export *exp, struct obdo *oa,
2181                                       struct filter_obd *filter)
2182 {
2183         struct obdo *doa = NULL;
2184         __u64 last, id;
2185         ENTRY;
2186
2187         LASSERT(oa);
2188         LASSERT(oa->o_gr != 0);
2189         LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2190
2191         doa = obdo_alloc();
2192         if (doa == NULL) {
2193                 CERROR("cannot allocate doa, error %d\n",
2194                        -ENOMEM);
2195                 EXIT;
2196                 return;
2197         }
2198
2199         doa->o_mode = S_IFREG;
2200         doa->o_gr = oa->o_gr;
2201         doa->o_valid = oa->o_valid & (OBD_MD_FLGROUP | OBD_MD_FLID);
2202
2203         set_bit(doa->o_gr, &filter->fo_destroys_in_progress);
2204         down(&filter->fo_create_locks[doa->o_gr]);
2205         if (!test_bit(doa->o_gr, &filter->fo_destroys_in_progress)) {
2206                 CERROR("%s:["LPU64"] destroy_in_progress already cleared\n",
2207                        exp->exp_obd->obd_name, doa->o_gr);
2208                 up(&filter->fo_create_locks[doa->o_gr]);
2209                 GOTO(out_free_doa, 0);
2210         }
2211
2212         last = filter_last_id(filter, doa->o_gr);
2213         CWARN("%s:["LPU64"] deleting orphan objects from "LPU64" to "LPU64"\n",
2214               exp->exp_obd->obd_name, doa->o_gr, oa->o_id + 1, last);
2215         for (id = oa->o_id + 1; id <= last; id++) {
2216                 doa->o_id = id;
2217                 filter_destroy(exp, doa, NULL, NULL);
2218         }
2219
2220         CDEBUG(D_HA, "%s:["LPU64"] after destroy: set last_objids = "LPU64"\n",
2221                exp->exp_obd->obd_name, doa->o_gr, oa->o_id);
2222
2223         filter_set_last_id(filter, doa->o_gr, oa->o_id);
2224
2225         clear_bit(doa->o_gr, &filter->fo_destroys_in_progress);
2226         up(&filter->fo_create_locks[doa->o_gr]);
2227
2228         EXIT;
2229 out_free_doa:
2230         obdo_free(doa);
2231 }
2232
2233 /* returns a negative error or a nonnegative number of files to create */
2234 static int filter_should_precreate(struct obd_export *exp, struct obdo *oa,
2235                                    obd_gr group)
2236 {
2237         struct obd_device *obd = exp->exp_obd;
2238         struct filter_obd *filter = &obd->u.filter;
2239         int diff, rc;
2240         ENTRY;
2241
2242         diff = oa->o_id - filter_last_id(filter, oa->o_gr);
2243         CDEBUG(D_INFO, "filter_last_id() = "LPU64" -> diff = %d\n",
2244                filter_last_id(filter, oa->o_gr), diff);
2245
2246         /* delete orphans request */
2247         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
2248             (oa->o_flags & OBD_FL_DELORPHAN)) {
2249                 if (diff >= 0)
2250                         RETURN(diff);
2251                 if (-diff > OST_MAX_PRECREATE) {
2252                         CERROR("ignoring bogus orphan destroy request: obdid "
2253                                LPU64" last_id "LPU64"\n",
2254                                oa->o_id, filter_last_id(filter, oa->o_gr));
2255                         RETURN(-EINVAL);
2256                 }
2257                 filter_destroy_precreated(exp, oa, filter);
2258                 rc = filter_update_last_objid(obd, group, 0);
2259                 if (rc)
2260                         CERROR("unable to write lastobjid, but orphans"
2261                                "were deleted\n");
2262                 RETURN(0);
2263         } else {
2264                 /* only precreate if group == 0 and o_id is specfied */
2265                 if (!(oa->o_valid & OBD_FL_DELORPHAN) &&
2266                     (/*group != 0 ||*/ oa->o_id == 0))
2267                         RETURN(1);
2268
2269                 LASSERTF(diff >= 0, LPU64" - "LPU64" = %d\n", oa->o_id,
2270                          filter_last_id(filter, oa->o_gr), diff);
2271                 RETURN(diff);
2272         }
2273 }
2274 static int filter_precreate_rec(struct obd_device *obd, struct dentry *dentry,
2275                                 int *number, struct obdo *oa)
2276 {
2277         int rc;
2278         ENTRY;
2279
2280         rc = fsfilt_precreate_rec(obd, dentry, number, oa);
2281
2282         RETURN(rc);
2283 }
2284
2285 static int filter_statfs(struct obd_device *obd, struct obd_statfs *osfs,
2286                          unsigned long max_age)
2287 {
2288         struct filter_obd *filter = &obd->u.filter;
2289         int blockbits = filter->fo_sb->s_blocksize_bits;
2290         int rc;
2291         ENTRY;
2292
2293         /* at least try to account for cached pages.  its still racey and
2294          * might be under-reporting if clients haven't announced their
2295          * caches with brw recently */
2296         spin_lock(&obd->obd_osfs_lock);
2297         rc = fsfilt_statfs(obd, filter->fo_sb, max_age);
2298         memcpy(osfs, &obd->obd_osfs, sizeof(*osfs));
2299         spin_unlock(&obd->obd_osfs_lock);
2300
2301         CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
2302                " pending "LPU64" free "LPU64" avail "LPU64"\n",
2303                filter->fo_tot_dirty, filter->fo_tot_granted,
2304                filter->fo_tot_pending,
2305                osfs->os_bfree << blockbits, osfs->os_bavail << blockbits);
2306
2307         filter_grant_sanity_check(obd, __FUNCTION__);
2308
2309         osfs->os_bavail -= min(osfs->os_bavail,
2310                                (filter->fo_tot_dirty + filter->fo_tot_pending +
2311                                 osfs->os_bsize -1) >> blockbits);
2312
2313         RETURN(rc);
2314 }
2315
2316 /* We rely on the fact that only one thread will be creating files in a given
2317  * group at a time, which is why we don't need an atomic filter_get_new_id.
2318  * Even if we had that atomic function, the following race would exist:
2319  *
2320  * thread 1: gets id x from filter_next_id
2321  * thread 2: gets id (x + 1) from filter_next_id
2322  * thread 2: creates object (x + 1)
2323  * thread 1: tries to create object x, gets -ENOSPC
2324  */
2325 static int filter_precreate(struct obd_device *obd, struct obdo *oa,
2326                             obd_gr group, int *num)
2327 {
2328         struct dentry *dchild = NULL, *dparent = NULL;
2329         int err = 0, rc = 0, recreate_obj = 0, i;
2330         struct filter_obd *filter;
2331         void *handle = NULL;
2332         void *lock = NULL;
2333         struct obd_statfs *osfs;
2334         unsigned long enough_time = jiffies + (obd_timeout * HZ) / 3;
2335         __u64 next_id;
2336         ENTRY;
2337
2338         filter = &obd->u.filter;
2339
2340         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
2341             (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
2342                 recreate_obj = 1;
2343         } else {
2344                 OBD_ALLOC(osfs, sizeof(*osfs));
2345                 if (osfs == NULL)
2346                         RETURN(-ENOMEM);
2347                 rc = filter_statfs(obd, osfs, jiffies-HZ);
2348                 if (rc == 0 && osfs->os_bavail < (osfs->os_blocks >> 10)) {
2349                         CDEBUG(D_HA, "OST out of space! avail "LPU64"\n",
2350                               osfs->os_bavail<<filter->fo_sb->s_blocksize_bits);
2351                         *num = 0;
2352                         rc = -ENOSPC;
2353                 }
2354                 OBD_FREE(osfs, sizeof(*osfs));
2355                 if (rc) {
2356                         RETURN(rc);
2357                 }
2358         }
2359
2360         CDEBUG(D_HA, "%s: precreating %d objects\n", obd->obd_name, *num);
2361
2362         down(&filter->fo_create_locks[group]);
2363
2364         for (i = 0; i < *num && err == 0; i++) {
2365                 int cleanup_phase = 0;
2366
2367                 if (test_bit(group, &filter->fo_destroys_in_progress)) {
2368                         CWARN("%s: precreate aborted by destroy\n",
2369                               obd->obd_name);
2370                         break;
2371                 }
2372
2373                 if (recreate_obj) {
2374                         __u64 last_id;
2375                         next_id = oa->o_id;
2376                         last_id = filter_last_id(filter, group);
2377                         if (next_id > last_id) {
2378                                 CERROR("Error: Trying to recreate obj greater"
2379                                        "than last id "LPD64" > "LPD64"\n",
2380                                        next_id, last_id);
2381                                 GOTO(cleanup, rc = -EINVAL);
2382                         }
2383                 } else {
2384                         next_id = filter_last_id(filter, group) + 1;
2385                 }
2386
2387                 CDEBUG(D_INFO, "precreate objid "LPU64"\n", next_id);
2388
2389                 dparent = filter_parent_lock(obd, group, next_id, &lock);
2390                 if (IS_ERR(dparent))
2391                         GOTO(cleanup, rc = PTR_ERR(dparent));
2392                 cleanup_phase = 1;
2393
2394                 /*only do precreate rec record. so clean kml flags here*/
2395                 fsfilt_clear_fs_flags(obd, dparent->d_inode, SM_DO_REC);
2396
2397                 dchild = filter_id2dentry(obd, dparent, group, next_id);
2398                 if (IS_ERR(dchild))
2399                         GOTO(cleanup, rc = PTR_ERR(dchild));
2400                 cleanup_phase = 2;
2401
2402                 if (dchild->d_inode != NULL) {
2403                         /* This would only happen if lastobjid was bad on disk*/
2404                         /* Could also happen if recreating missing obj but
2405                          * already exists
2406                          */
2407                         if (recreate_obj) {
2408                                 CERROR("%s: recreating existing object %.*s?\n",
2409                                        obd->obd_name, dchild->d_name.len,
2410                                        dchild->d_name.name);
2411                         } else {
2412                                 CERROR("%s: Serious error: objid %.*s already "
2413                                        "exists; is this filesystem corrupt?\n",
2414                                        obd->obd_name, dchild->d_name.len,
2415                                        dchild->d_name.name);
2416                                 LBUG();
2417                         }
2418                         GOTO(cleanup, rc = -EEXIST);
2419                 }
2420
2421                 handle = fsfilt_start_log(obd, dparent->d_inode,
2422                                           FSFILT_OP_CREATE, NULL, 1);
2423                 if (IS_ERR(handle))
2424                         GOTO(cleanup, rc = PTR_ERR(handle));
2425                 cleanup_phase = 3;
2426
2427                 rc = ll_vfs_create(dparent->d_inode, dchild, S_IFREG, NULL);
2428                 if (rc) {
2429                         CERROR("create failed rc = %d\n", rc);
2430                         GOTO(cleanup, rc);
2431                 }
2432
2433                 if (!recreate_obj) {
2434                         filter_set_last_id(filter, group, next_id);
2435                         err = filter_update_last_objid(obd, group, 0);
2436                         if (err)
2437                                 CERROR("unable to write lastobjid "
2438                                        "but file created\n");
2439                 }
2440                 fsfilt_set_fs_flags(obd, dparent->d_inode, SM_DO_REC);
2441
2442         cleanup:
2443                 switch(cleanup_phase) {
2444                 case 3:
2445                         err = fsfilt_commit(obd, filter->fo_sb,
2446                                             dparent->d_inode, handle, 0);
2447                         if (err) {
2448                                 CERROR("error on commit, err = %d\n", err);
2449                                 if (!rc)
2450                                         rc = err;
2451                         }
2452                 case 2:
2453                         f_dput(dchild);
2454                 case 1:
2455                         filter_parent_unlock(dparent, lock);
2456                 case 0:
2457                         break;
2458                 }
2459
2460                 if (rc)
2461                         break;
2462                 if (time_after(jiffies, enough_time)) {
2463                         CDEBUG(D_INODE,"%s: precreate slow - want %d got %d \n",
2464                                obd->obd_name, *num, i);
2465                         break;
2466                 }
2467         }
2468
2469         *num = i;
2470
2471         /* check if we have an error after ll_vfs_create(). It is possible that
2472          * there will be say -ENOSPC and we will leak it. */
2473         if (rc == 0)
2474                 rc = filter_precreate_rec(obd, dparent, num, oa);
2475
2476         up(&filter->fo_create_locks[group]);
2477
2478         CDEBUG(D_HA, "%s: server last_objid for group "LPU64": "LPU64"\n",
2479                obd->obd_name, group, filter->fo_last_objids[group]);
2480
2481         CDEBUG(D_HA, "%s: filter_precreate() created %d objects\n",
2482                obd->obd_name, i);
2483
2484         RETURN(rc);
2485 }
2486
2487 static int filter_create(struct obd_export *exp, struct obdo *oa,
2488                          void *acl, int acl_size,
2489                          struct lov_stripe_md **ea, struct obd_trans_info *oti)
2490 {
2491         struct obd_device *obd = NULL;
2492         struct filter_obd *filter;
2493         struct lvfs_run_ctxt saved;
2494         struct lov_stripe_md *lsm = NULL;
2495         struct filter_export_data *fed;
2496         char str[PTL_NALFMT_SIZE];
2497         int group = oa->o_gr, rc = 0, diff, recreate_objs = 0;
2498         ENTRY;
2499
2500         LASSERT(acl == NULL && acl_size == 0);
2501
2502         if (!(oa->o_valid & OBD_MD_FLGROUP) || group == 0) {
2503                 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
2504                                 exp->exp_connection->c_peer.peer_id.nid, str);
2505                 CERROR("!!! nid "LPX64"/%s sent invalid object group %d\n",
2506                        exp->exp_connection->c_peer.peer_id.nid, str, group);
2507                 RETURN(-EINVAL);
2508         }
2509
2510         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
2511             (oa->o_flags & OBD_FL_RECREATE_OBJS))
2512                 recreate_objs = 1;
2513
2514         obd = exp->exp_obd;
2515         fed = &exp->exp_filter_data;
2516         filter = &obd->u.filter;
2517
2518         if (fed->fed_group != group && !recreate_objs &&
2519             !(oa->o_valid & OBD_MD_REINT)) {
2520                 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
2521                                 exp->exp_connection->c_peer.peer_id.nid, str);
2522                 CERROR("!!! This export (nid "LPX64"/%s) used object group %d "
2523                        "earlier; now it's trying to use group %d!  This could "
2524                        "be a bug in the MDS.  Tell CFS.\n",
2525                        exp->exp_connection->c_peer.peer_id.nid, str,
2526                        fed->fed_group, group);
2527                 RETURN(-ENOTUNIQ);
2528         }
2529
2530         CDEBUG(D_INFO, "filter_create(od->o_gr=%d,od->o_id="LPU64")\n",
2531                group, oa->o_id);
2532         if (ea != NULL) {
2533                 lsm = *ea;
2534                 if (lsm == NULL) {
2535                         rc = obd_alloc_memmd(exp, &lsm);
2536                         if (rc < 0)
2537                                 RETURN(rc);
2538                 }
2539         }
2540
2541         obd = exp->exp_obd;
2542         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2543
2544         if (oa->o_valid & OBD_MD_REINT) {
2545                 int num = *((int*)oa->o_inline);
2546                 rc = filter_precreate(obd, oa, oa->o_gr, &num);
2547         } else if (recreate_objs) {
2548                 if (oa->o_id > filter_last_id(&obd->u.filter, group)) {
2549                         CERROR("recreate objid "LPU64" > last id "LPU64"\n",
2550                                oa->o_id, filter_last_id(&obd->u.filter, group));
2551                         rc = -EINVAL;
2552                 } else {
2553                         diff = 1;
2554                         rc = filter_precreate(obd, oa, group, &diff);
2555                 }
2556         } else {
2557                 diff = filter_should_precreate(exp, oa, group);
2558                 if (diff > 0) {
2559                         oa->o_id = filter_last_id(&obd->u.filter, group);
2560                         rc = filter_precreate(obd, oa, group, &diff);
2561                         oa->o_id = filter_last_id(&obd->u.filter, oa->o_gr);
2562                         oa->o_valid = OBD_MD_FLID;
2563                 }
2564         }
2565
2566         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2567         if (rc && ea != NULL && *ea != lsm) {
2568                 obd_free_memmd(exp, &lsm);
2569         } else if (rc == 0 && ea != NULL) {
2570                 /* XXX LOV STACKING: the lsm that is passed to us from
2571                  * LOV does not have valid lsm_oinfo data structs, so
2572                  * don't go touching that.  This needs to be fixed in a
2573                  * big way. */
2574                 lsm->lsm_object_id = oa->o_id;
2575                 lsm->lsm_object_gr = oa->o_gr;
2576                 *ea = lsm;
2577         }
2578
2579         RETURN(rc);
2580 }
2581
2582 static int filter_destroy(struct obd_export *exp, struct obdo *oa,
2583                           struct lov_stripe_md *ea, struct obd_trans_info *oti)
2584 {
2585         int rc, rc2, cleanup_phase = 0, have_prepared = 0;
2586         struct dentry *dchild = NULL, *dparent = NULL;
2587         struct llog_cookie *fcc = NULL;
2588         struct lvfs_run_ctxt saved;
2589         struct filter_obd *filter;
2590         struct obd_device *obd;
2591         void *handle = NULL;
2592         void *lock = NULL;
2593         struct iattr iattr;
2594         ENTRY;
2595
2596         LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2597
2598         obd = exp->exp_obd;
2599         filter = &obd->u.filter;
2600
2601         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2602
2603  acquire_locks:
2604         dparent = filter_parent_lock(obd, oa->o_gr, oa->o_id, &lock);
2605         if (IS_ERR(dparent))
2606                 GOTO(cleanup, rc = PTR_ERR(dparent));
2607         cleanup_phase = 1;
2608
2609         dchild = filter_id2dentry(obd, dparent, oa->o_gr, oa->o_id);
2610         if (IS_ERR(dchild))
2611                 GOTO(cleanup, rc = PTR_ERR(dchild));
2612         cleanup_phase = 2;
2613
2614         if (dchild->d_inode == NULL) {
2615                 CDEBUG(D_INODE, "destroying non-existent object "LPU64"\n",
2616                        oa->o_id);
2617                 GOTO(cleanup, rc = -ENOENT);
2618         }
2619
2620         if (!have_prepared) {
2621                 /* If we're really going to destroy the object, get ready
2622                  * by getting the clients to discard their cached data.
2623                  *
2624                  * We have to drop the parent lock, because
2625                  * filter_prepare_destroy will acquire a PW on the object, and
2626                  * we don't want to deadlock with an incoming write to the
2627                  * object, which has the extent PW and then wants to get the
2628                  * parent dentry to do the lookup.
2629                  *
2630                  * We dput the child because it's not worth the extra
2631                  * complication of condition the above code to skip it on the
2632                  * second time through. */
2633                 f_dput(dchild);
2634                 filter_parent_unlock(dparent, lock);
2635
2636                 filter_prepare_destroy(obd, oa->o_id, oa->o_gr);
2637                 have_prepared = 1;
2638                 goto acquire_locks;
2639         }
2640
2641         /* Our MDC connection is established by the MDS to us */
2642         if (oa->o_valid & OBD_MD_FLCOOKIE) {
2643                 OBD_ALLOC(fcc, sizeof(*fcc));
2644                 if (fcc != NULL)
2645                         memcpy(fcc, obdo_logcookie(oa), sizeof(*fcc));
2646         }
2647
2648         /* we're gonna truncate it first in order to avoid possible
2649          * deadlock:
2650          *      P1                      P2
2651          * open trasaction      open transaction
2652          * down(i_zombie)       down(i_zombie)
2653          *                      restart transaction
2654          * (see BUG 4180) -bzzz
2655          */
2656         down(&dchild->d_inode->i_sem);
2657         handle = fsfilt_start_log(obd, dparent->d_inode,FSFILT_OP_SETATTR,NULL,1);
2658         if (IS_ERR(handle)) {
2659                 up(&dchild->d_inode->i_sem);
2660                 GOTO(cleanup, rc = PTR_ERR(handle));
2661         }
2662
2663         iattr.ia_valid = ATTR_SIZE;
2664         iattr.ia_size = 0;
2665         rc = fsfilt_setattr(obd, dchild, handle, &iattr, 1);
2666
2667         rc2 = fsfilt_commit(obd, filter->fo_sb, dparent->d_inode, handle, 0);
2668         up(&dchild->d_inode->i_sem);
2669         if (rc)
2670                 GOTO(cleanup, rc);
2671         if (rc2)
2672                 GOTO(cleanup, rc = rc2);
2673
2674         handle = fsfilt_start_log(obd, dparent->d_inode,FSFILT_OP_UNLINK,oti,1);
2675         if (IS_ERR(handle))
2676                 GOTO(cleanup, rc = PTR_ERR(handle));
2677
2678         cleanup_phase = 3;
2679
2680         rc = filter_destroy_internal(obd, oa->o_id, dparent, dchild);
2681
2682 cleanup:
2683         switch(cleanup_phase) {
2684         case 3:
2685                 if (fcc != NULL) {
2686                         fsfilt_add_journal_cb(obd, filter->fo_sb, 0,
2687                                               oti ? oti->oti_handle : handle,
2688                                               filter_cancel_cookies_cb, fcc);
2689                 }
2690                 rc = filter_finish_transno(exp, oti, rc);
2691                 rc2 = fsfilt_commit(obd, filter->fo_sb, dparent->d_inode,
2692                                     handle, exp->exp_sync);
2693                 if (rc2) {
2694                         CERROR("error on commit, err = %d\n", rc2);
2695                         if (!rc)
2696                                 rc = rc2;
2697                 }
2698         case 2:
2699                 f_dput(dchild);
2700         case 1:
2701                 filter_parent_unlock(dparent, lock);
2702         case 0:
2703                 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2704                 break;
2705         default:
2706                 CERROR("invalid cleanup_phase %d\n", cleanup_phase);
2707                 LBUG();
2708         }
2709
2710         RETURN(rc);
2711 }
2712
2713 /* NB start and end are used for punch, but not truncate */
2714 static int filter_truncate(struct obd_export *exp, struct obdo *oa,
2715                            struct lov_stripe_md *lsm,
2716                            obd_off start, obd_off end,
2717                            struct obd_trans_info *oti)
2718 {
2719         int error;
2720         ENTRY;
2721
2722         if (end != OBD_OBJECT_EOF)
2723                 CERROR("PUNCH not supported, only truncate: end = "LPX64"\n",
2724                        end);
2725
2726         CDEBUG(D_INODE, "calling truncate for object "LPU64", valid = "LPU64", "
2727                "o_size = "LPD64"\n", oa->o_id, oa->o_valid, start);
2728         oa->o_size = start;
2729         error = filter_setattr(exp, oa, NULL, oti);
2730         RETURN(error);
2731 }
2732
2733 static int filter_sync(struct obd_export *exp, struct obdo *oa,
2734                        struct lov_stripe_md *lsm, obd_off start, obd_off end)
2735 {
2736         struct obd_device *obd = exp->exp_obd;
2737         struct lvfs_run_ctxt saved;
2738         struct filter_obd *filter;
2739         struct dentry *dentry;
2740         int rc, rc2;
2741         ENTRY;
2742
2743         filter = &obd->u.filter;
2744
2745         /* an objid of zero is taken to mean "sync whole filesystem" */
2746         if (!oa || !(oa->o_valid & OBD_MD_FLID)) {
2747                 rc = fsfilt_sync(obd, filter->fo_sb);
2748                 /* flush any remaining cancel messages out to the target */
2749                 filter_sync_llogs(obd, NULL);
2750                 RETURN(rc);
2751         }
2752
2753         dentry = filter_oa2dentry(obd, oa);
2754         if (IS_ERR(dentry))
2755                 RETURN(PTR_ERR(dentry));
2756
2757         push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2758
2759         down(&dentry->d_inode->i_sem);
2760         rc = filemap_fdatawrite(dentry->d_inode->i_mapping);
2761         if (rc == 0) {
2762                 /* just any file to grab fsync method - "file" arg unused */
2763                 struct file *file = filter->fo_rcvd_filp;
2764
2765                 if (file->f_op && file->f_op->fsync)
2766                         rc = file->f_op->fsync(NULL, dentry, 1);
2767
2768                 rc2 = filemap_fdatawait(dentry->d_inode->i_mapping);
2769                 if (!rc)
2770                         rc = rc2;
2771         }
2772         up(&dentry->d_inode->i_sem);
2773
2774         oa->o_valid = OBD_MD_FLID;
2775         obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2776
2777         pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2778
2779         f_dput(dentry);
2780         RETURN(rc);
2781 }
2782
2783 static int filter_get_info(struct obd_export *exp, __u32 keylen,
2784                            void *key, __u32 *vallen, void *val)
2785 {
2786         struct filter_export_data *fed = &exp->exp_filter_data;
2787         struct obd_device *obd;
2788         ENTRY;
2789
2790         obd = class_exp2obd(exp);
2791         if (obd == NULL) {
2792                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2793                        exp->exp_handle.h_cookie);
2794                 RETURN(-EINVAL);
2795         }
2796
2797         if (keylen == strlen("blocksize") &&
2798             memcmp(key, "blocksize", keylen) == 0) {
2799                 __u32 *blocksize = val;
2800                 *vallen = sizeof(*blocksize);
2801                 *blocksize = obd->u.filter.fo_sb->s_blocksize;
2802                 RETURN(0);
2803         }
2804
2805         if (keylen == strlen("blocksize_bits") &&
2806             memcmp(key, "blocksize_bits", keylen) == 0) {
2807                 __u32 *blocksize_bits = val;
2808                 *vallen = sizeof(*blocksize_bits);
2809                 *blocksize_bits = obd->u.filter.fo_sb->s_blocksize_bits;
2810                 RETURN(0);
2811         }
2812
2813         if (keylen >= strlen("last_id") && memcmp(key, "last_id", 7) == 0) {
2814                 obd_id *last_id = val;
2815                 *last_id = filter_last_id(&obd->u.filter, fed->fed_group);
2816                 RETURN(0);
2817         }
2818         if (keylen >= strlen("reint_log") && memcmp(key, "reint_log", 9) == 0) {
2819                 /*Get log_context handle*/
2820                 unsigned long *llh_handle = val;
2821                 *vallen = sizeof(unsigned long);
2822                 *llh_handle = (unsigned long)obd->obd_llog_ctxt[LLOG_REINT_ORIG_CTXT];
2823                 RETURN(0);
2824         }
2825         if (keylen >= strlen("cache_sb") && memcmp(key, "cache_sb", 8) == 0) {
2826                 /*Get log_context handle*/
2827                 unsigned long *sb = val;
2828                 *vallen = sizeof(unsigned long);
2829                 *sb = (unsigned long)obd->u.filter.fo_sb;
2830                 RETURN(0);
2831         }
2832
2833         CDEBUG(D_IOCTL, "invalid key\n");
2834         RETURN(-EINVAL);
2835 }
2836
2837 struct obd_llogs *filter_grab_llog_for_group(struct obd_device *obd, int group,
2838                                              struct obd_export *export)
2839 {
2840         struct filter_group_llog *fglog, *nlog;
2841         char name[32] = "CATLIST";
2842         struct filter_obd *filter;
2843         struct llog_ctxt *ctxt;
2844         struct list_head *cur;
2845         int rc;
2846
2847         filter = &obd->u.filter;
2848
2849         spin_lock(&filter->fo_llog_list_lock);
2850         list_for_each(cur, &filter->fo_llog_list) {
2851                 fglog = list_entry(cur, struct filter_group_llog, list);
2852                 if (fglog->group == group) {
2853                         if (!(fglog->exp == NULL || fglog->exp == export || export == NULL))
2854                                 CWARN("%s: export for group %d changes: 0x%p -> 0x%p\n",
2855                                       obd->obd_name, group, fglog->exp, export);
2856                         spin_unlock(&filter->fo_llog_list_lock);
2857                         goto init;
2858                 }
2859         }
2860         spin_unlock(&filter->fo_llog_list_lock);
2861
2862         if (export == NULL)
2863                 RETURN(NULL);
2864
2865         OBD_ALLOC(fglog, sizeof(*fglog));
2866         if (fglog == NULL)
2867                 RETURN(NULL);
2868         fglog->group = group;
2869
2870         OBD_ALLOC(fglog->llogs, sizeof(struct obd_llogs));
2871         if (fglog->llogs == NULL) {
2872                 OBD_FREE(fglog, sizeof(*fglog));
2873                 RETURN(NULL);
2874         }
2875
2876         spin_lock(&filter->fo_llog_list_lock);
2877         list_for_each(cur, &filter->fo_llog_list) {
2878                 nlog = list_entry(cur, struct filter_group_llog, list);
2879                 LASSERT(nlog->group != group);
2880         }
2881         list_add(&fglog->list, &filter->fo_llog_list);
2882         spin_unlock(&filter->fo_llog_list_lock);
2883
2884         rc = obd_llog_cat_initialize(obd, fglog->llogs, 1, name);
2885         if (rc) {
2886                 OBD_FREE(fglog->llogs, sizeof(*(fglog->llogs)));
2887                 OBD_FREE(fglog, sizeof(*fglog));
2888                 RETURN(NULL);
2889         }
2890
2891 init:
2892         if (export) {
2893                 fglog->exp = export;
2894                 ctxt = llog_get_context(fglog->llogs, LLOG_UNLINK_REPL_CTXT);
2895                 LASSERT(ctxt != NULL);
2896
2897                 llog_receptor_accept(ctxt, export->exp_imp_reverse);
2898         }
2899         CDEBUG(D_OTHER, "%s: new llog 0x%p for group %u\n",
2900                obd->obd_name, fglog->llogs, group);
2901
2902         RETURN(fglog->llogs);
2903 }
2904
2905 int filter_iocontrol(unsigned int cmd, struct obd_export *exp,
2906                      int len, void *karg, void *uarg)
2907 {
2908         struct obd_device *obd = exp->exp_obd;
2909         struct obd_ioctl_data *data = karg;
2910         int rc = 0;
2911
2912         switch (cmd) {
2913         case OBD_IOC_ABORT_RECOVERY:
2914                 target_stop_recovery_thread(obd);
2915                 RETURN(0);
2916
2917         case OBD_IOC_SET_READONLY: {
2918                 void *handle;
2919                 struct super_block *sb = obd->u.filter.fo_sb;
2920                 struct inode *inode = sb->s_root->d_inode;
2921                 BDEVNAME_DECLARE_STORAGE(tmp);
2922                 CERROR("setting device %s read-only\n",
2923                        ll_bdevname(sb, tmp));
2924
2925                 handle = fsfilt_start(obd, inode, FSFILT_OP_MKNOD, NULL);
2926                 LASSERT(handle);
2927                 (void)fsfilt_commit(obd, sb, inode, handle, 1);
2928
2929                 ll_set_rdonly(ll_sbdev(obd->u.filter.fo_sb));
2930                 RETURN(0);
2931         }
2932
2933         case OBD_IOC_CATLOGLIST: {
2934                 rc = llog_catalog_list(obd, 1, data);
2935                 RETURN(rc);
2936         }
2937         case OBD_IOC_LLOG_CANCEL:
2938         case OBD_IOC_LLOG_REMOVE:
2939         case OBD_IOC_LLOG_INFO:
2940         case OBD_IOC_LLOG_PRINT: {
2941                 /* FIXME to be finished */
2942                 RETURN(-EOPNOTSUPP);
2943 /*
2944                 struct llog_ctxt *ctxt = NULL;
2945
2946                 push_ctxt(&saved, &ctxt->loc_ctxt, NULL);
2947                 rc = llog_ioctl(ctxt, cmd, data);
2948                 pop_ctxt(&saved, &ctxt->loc_ctxt, NULL);
2949
2950                 RETURN(rc);
2951 */
2952         }
2953
2954
2955         default:
2956                 RETURN(-EINVAL);
2957         }
2958         RETURN(0);
2959 }
2960
2961 static struct llog_operations filter_unlink_repl_logops;
2962 static struct llog_operations filter_size_orig_logops = {
2963         lop_setup: llog_obd_origin_setup,
2964         lop_cleanup: llog_catalog_cleanup,
2965         lop_add: llog_catalog_add,
2966 };
2967
2968 static int filter_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
2969                             struct obd_device *tgt, int count,
2970                             struct llog_catid *catid)
2971 {
2972         struct llog_ctxt *ctxt;
2973         int rc;
2974         ENTRY;
2975
2976         filter_unlink_repl_logops = llog_client_ops;
2977         filter_unlink_repl_logops.lop_cancel = llog_obd_repl_cancel;
2978         filter_unlink_repl_logops.lop_connect = llog_repl_connect;
2979         filter_unlink_repl_logops.lop_sync = llog_obd_repl_sync;
2980
2981         rc = obd_llog_setup(obd, llogs, LLOG_UNLINK_REPL_CTXT, tgt, 0, NULL,
2982                         &filter_unlink_repl_logops);
2983         if (rc)
2984                 RETURN(rc);
2985         /* FIXME - assign unlink_cb for filter's recovery */
2986         ctxt = llog_get_context(llogs, LLOG_UNLINK_REPL_CTXT);
2987         ctxt->llog_proc_cb = filter_recov_log_unlink_cb;
2988
2989         /* FIXME - count should be 1 to setup size log */
2990         rc = obd_llog_setup(obd, llogs, LLOG_SIZE_ORIG_CTXT, tgt, 0,
2991                             &catid->lci_logid, &filter_size_orig_logops);
2992         RETURN(rc);
2993 }
2994
2995 static int filter_llog_finish(struct obd_device *obd,
2996                               struct obd_llogs *llogs, int count)
2997 {
2998         int rc;
2999         ENTRY;
3000
3001         rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_UNLINK_REPL_CTXT));
3002         if (rc)
3003                 RETURN(rc);
3004
3005         rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_SIZE_ORIG_CTXT));
3006         RETURN(rc);
3007 }
3008
3009 static int filter_llog_connect(struct obd_export *exp,
3010                                struct llogd_conn_body *body)
3011 {
3012         struct obd_device *obd = exp->exp_obd;
3013         struct llog_ctxt *ctxt;
3014         struct obd_llogs *llog;
3015         int rc;
3016         ENTRY;
3017
3018         CDEBUG(D_OTHER, "handle connect for %s: %u/%u/%u\n", obd->obd_name,
3019                (unsigned) body->lgdc_logid.lgl_ogr,
3020                (unsigned) body->lgdc_logid.lgl_oid,
3021                (unsigned) body->lgdc_logid.lgl_ogen);
3022         llog = filter_grab_llog_for_group(obd, body->lgdc_logid.lgl_ogr, exp);
3023         LASSERT(llog != NULL);
3024         ctxt = llog_get_context(llog, body->lgdc_ctxt_idx);
3025         rc = llog_connect(ctxt, 1, &body->lgdc_logid,
3026                           &body->lgdc_gen, NULL);
3027         if (rc != 0)
3028                 CERROR("failed to connect\n");
3029
3030         RETURN(rc);
3031 }
3032
3033 static struct dentry *filter_lvfs_id2dentry(__u64 id, __u32 gen,
3034                                             __u64 gr, void *data)
3035 {
3036         return filter_id2dentry(data, NULL, gr, id);
3037 }
3038
3039 static struct lvfs_callback_ops filter_lvfs_ops = {
3040         l_id2dentry:     filter_lvfs_id2dentry,
3041 };
3042
3043 static struct obd_ops filter_obd_ops = {
3044         .o_owner          = THIS_MODULE,
3045         .o_attach         = filter_attach,
3046         .o_detach         = filter_detach,
3047         .o_get_info       = filter_get_info,
3048         .o_setup          = filter_setup,
3049         .o_precleanup     = filter_precleanup,
3050         .o_cleanup        = filter_cleanup,
3051         .o_connect        = filter_connect,
3052         .o_connect_post   = filter_connect_post,
3053         .o_disconnect     = filter_disconnect,
3054         .o_statfs         = filter_statfs,
3055         .o_getattr        = filter_getattr,
3056         .o_unpackmd       = filter_unpackmd,
3057         .o_create         = filter_create,
3058         .o_setattr        = filter_setattr,
3059         .o_destroy        = filter_destroy,
3060         .o_brw            = filter_brw,
3061         .o_punch          = filter_truncate,
3062         .o_sync           = filter_sync,
3063         .o_preprw         = filter_preprw,
3064         .o_commitrw       = filter_commitrw,
3065         .o_do_cow         = filter_do_cow,
3066         .o_write_extents  = filter_write_extents,
3067         .o_destroy_export = filter_destroy_export,
3068         .o_llog_init      = filter_llog_init,
3069         .o_llog_finish    = filter_llog_finish,
3070         .o_llog_connect   = filter_llog_connect,
3071         .o_iocontrol      = filter_iocontrol,
3072 };
3073
3074 static struct obd_ops filter_sanobd_ops = {
3075         .o_owner          = THIS_MODULE,
3076         .o_attach         = filter_attach,
3077         .o_detach         = filter_detach,
3078         .o_get_info       = filter_get_info,
3079         .o_setup          = filter_san_setup,
3080         .o_precleanup     = filter_precleanup,
3081         .o_cleanup        = filter_cleanup,
3082         .o_connect        = filter_connect,
3083         .o_connect_post   = filter_connect_post,
3084         .o_disconnect     = filter_disconnect,
3085         .o_statfs         = filter_statfs,
3086         .o_getattr        = filter_getattr,
3087         .o_unpackmd       = filter_unpackmd,
3088         .o_create         = filter_create,
3089         .o_setattr        = filter_setattr,
3090         .o_destroy        = filter_destroy,
3091         .o_brw            = filter_brw,
3092         .o_punch          = filter_truncate,
3093         .o_sync           = filter_sync,
3094         .o_preprw         = filter_preprw,
3095         .o_commitrw       = filter_commitrw,
3096         .o_do_cow         = filter_do_cow,
3097         .o_write_extents  = filter_write_extents,
3098         .o_san_preprw     = filter_san_preprw,
3099         .o_destroy_export = filter_destroy_export,
3100         .o_llog_init      = filter_llog_init,
3101         .o_llog_finish    = filter_llog_finish,
3102         .o_llog_connect   = filter_llog_connect,
3103         .o_iocontrol      = filter_iocontrol,
3104 };
3105
3106 static int __init obdfilter_init(void)
3107 {
3108         struct lprocfs_static_vars lvars;
3109         int size, rc;
3110
3111         printk(KERN_INFO "Lustre: Filtering OBD driver; info@clusterfs.com\n");
3112
3113         lprocfs_init_vars(filter, &lvars);
3114
3115         size = OBDFILTER_CREATED_SCRATCHPAD_ENTRIES *
3116                 sizeof(*obdfilter_created_scratchpad);
3117
3118         OBD_ALLOC(obdfilter_created_scratchpad, size);
3119         if (obdfilter_created_scratchpad == NULL) {
3120                 CERROR ("Can't allocate scratchpad\n");
3121                 return -ENOMEM;
3122         }
3123
3124         rc = class_register_type(&filter_obd_ops, NULL, lvars.module_vars,
3125                                  OBD_FILTER_DEVICENAME);
3126         if (rc) {
3127                 OBD_FREE(obdfilter_created_scratchpad, size);
3128                 return rc;
3129         }
3130
3131         rc = class_register_type(&filter_sanobd_ops, NULL, lvars.module_vars,
3132                                  OBD_FILTER_SAN_DEVICENAME);
3133         if (rc) {
3134                 class_unregister_type(OBD_FILTER_DEVICENAME);
3135                 OBD_FREE(obdfilter_created_scratchpad, size);
3136         }
3137         return rc;
3138 }
3139
3140 static void __exit obdfilter_exit(void)
3141 {
3142         class_unregister_type(OBD_FILTER_SAN_DEVICENAME);
3143         class_unregister_type(OBD_FILTER_DEVICENAME);
3144         OBD_FREE(obdfilter_created_scratchpad,
3145                  OBDFILTER_CREATED_SCRATCHPAD_ENTRIES *
3146                  sizeof(*obdfilter_created_scratchpad));
3147 }
3148
3149 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
3150 MODULE_DESCRIPTION("Lustre Filtering OBD driver");
3151 MODULE_LICENSE("GPL");
3152
3153 module_init(obdfilter_init);
3154 module_exit(obdfilter_exit);