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 <linux/lustre_sec.h>
  64 #include <linux/lustre_audit.h>
  65 #include "filter_internal.h"
  66
  67 /* Group 0 is no longer a legal group, to catch uninitialized IDs */
  68 #define FILTER_MIN_GROUPS 3
  69
  70 static struct lvfs_callback_ops filter_lvfs_ops;
  71
  72 static int filter_destroy(struct obd_export *exp, struct obdo *oa,
  73                           struct lov_stripe_md *ea, struct obd_trans_info *);
  74 struct obd_llogs *filter_grab_llog_for_group(struct obd_device *,
  75                                              int, struct obd_export *);
  76
  77 static void filter_commit_cb(struct obd_device *obd, __u64 transno,
  78                              void *cb_data, int error)
  79 {
  80         obd_transno_commit_cb(obd, transno, error);
  81 }
  82
  83 /* Assumes caller has already pushed us into the kernel context. */
  84 int filter_finish_transno(struct obd_export *exp, struct obd_trans_info *oti,
  85                           int rc)
  86 {
  87         struct filter_obd *filter = &exp->exp_obd->u.filter;
  88         struct filter_export_data *fed = &exp->exp_filter_data;
  89         struct filter_client_data *fcd = fed->fed_fcd;
  90         __u64 last_rcvd;
  91         loff_t off;
  92         int err, log_pri = D_HA;
  93
  94         /* Propagate error code. */
  95         if (rc)
  96                 RETURN(rc);
  97
  98         if (!exp->exp_obd->obd_replayable || oti == NULL)
  99                 RETURN(rc);
 100
 101         /* we don't allocate new transnos for replayed requests */
 102         if (oti->oti_transno == 0) {
 103                 spin_lock(&filter->fo_translock);
 104                 last_rcvd = le64_to_cpu(filter->fo_fsd->fsd_last_transno) + 1;
 105                 filter->fo_fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
 106                 spin_unlock(&filter->fo_translock);
 107                 oti->oti_transno = last_rcvd;
 108         } else {
 109                 spin_lock(&filter->fo_translock);
 110                 last_rcvd = oti->oti_transno;
 111                 if (last_rcvd > le64_to_cpu(filter->fo_fsd->fsd_last_transno))
 112                         filter->fo_fsd->fsd_last_transno =
 113                                 cpu_to_le64(last_rcvd);
 114                 spin_unlock(&filter->fo_translock);
 115         }
 116
 117         fcd->fcd_last_rcvd = cpu_to_le64(last_rcvd);
 118
 119         /* could get xid from oti, if it's ever needed */
 120         fcd->fcd_last_xid = 0;
 121
 122         off = fed->fed_lr_off;
 123
 124         fsfilt_add_journal_cb(exp->exp_obd, filter->fo_sb, last_rcvd,
 125                               oti->oti_handle, filter_commit_cb, NULL);
 126
 127         err = fsfilt_write_record(exp->exp_obd, filter->fo_rcvd_filp, fcd,
 128                                   sizeof(*fcd), &off, 0);
 129         if (err) {
 130                 log_pri = D_ERROR;
 131                 if (rc == 0)
 132                         rc = err;
 133         }
 134
 135         CDEBUG(log_pri, "wrote trans "LPU64" for client %s at #%d: err = %d\n",
 136                last_rcvd, fcd->fcd_uuid, fed->fed_lr_idx, err);
 137
 138         RETURN(rc);
 139 }
 140
 141 void f_dput(struct dentry *dentry)
 142 {
 143         /* Can't go inside filter_ddelete because it can block */
 144         CDEBUG(D_INODE, "putting %s: %p, count = %d\n",
 145                dentry->d_name.name, dentry, atomic_read(&dentry->d_count) - 1);
 146         LASSERT(atomic_read(&dentry->d_count) > 0);
 147
 148         dput(dentry);
 149 }
 150
 151 /* Add client data to the FILTER.  We use a bitmap to locate a free space
 152  * in the last_rcvd file if cl_idx is -1 (i.e. a new client).
 153  * Otherwise, we have just read the data from the last_rcvd file and
 154  * we know its offset. */
 155 static int filter_client_add(struct obd_device *obd, struct filter_obd *filter,
 156                              struct filter_export_data *fed, int cl_idx)
 157 {
 158         unsigned long *bitmap = filter->fo_last_rcvd_slots;
 159         int new_client = (cl_idx == -1);
 160         ENTRY;
 161
 162         LASSERT(bitmap != NULL);
 163
 164         /* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
 165         if (!strcmp((char *)fed->fed_fcd->fcd_uuid, (char *)obd->obd_uuid.uuid))
 166                 RETURN(0);
 167
 168         /* the bitmap operations can handle cl_idx > sizeof(long) * 8, so
 169          * there's no need for extra complication here
 170          */
 171         if (new_client) {
 172                 cl_idx = find_first_zero_bit(bitmap, FILTER_LR_MAX_CLIENTS);
 173         repeat:
 174                 if (cl_idx >= FILTER_LR_MAX_CLIENTS) {
 175                         CERROR("no client slots - fix FILTER_LR_MAX_CLIENTS\n");
 176                         RETURN(-ENOMEM);
 177                 }
 178                 if (test_and_set_bit(cl_idx, bitmap)) {
 179                         CERROR("FILTER client %d: found bit is set in bitmap\n",
 180                                cl_idx);
 181                         cl_idx = find_next_zero_bit(bitmap,
 182                                                     FILTER_LR_MAX_CLIENTS,
 183                                                     cl_idx);
 184                         goto repeat;
 185                 }
 186         } else {
 187                 if (test_and_set_bit(cl_idx, bitmap)) {
 188                         CERROR("FILTER client %d: bit already set in bitmap!\n",
 189                                cl_idx);
 190                         LBUG();
 191                 }
 192         }
 193
 194         fed->fed_lr_idx = cl_idx;
 195         fed->fed_lr_off = le32_to_cpu(filter->fo_fsd->fsd_client_start) +
 196                 cl_idx * le16_to_cpu(filter->fo_fsd->fsd_client_size);
 197
 198         CDEBUG(D_INFO, "client at index %d (%llu) with UUID '%s' added\n",
 199                fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
 200
 201         if (new_client) {
 202                 struct lvfs_run_ctxt saved;
 203                 loff_t off = fed->fed_lr_off;
 204                 int err;
 205                 void *handle;
 206
 207                 CDEBUG(D_INFO, "writing client fcd at idx %u (%llu) (len %u)\n",
 208                        fed->fed_lr_idx,off,(unsigned int)sizeof(*fed->fed_fcd));
 209
 210                 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 211                 /* Transaction needed to fix bug 1403 */
 212                 handle = fsfilt_start(obd,
 213                                       filter->fo_rcvd_filp->f_dentry->d_inode,
 214                                       FSFILT_OP_SETATTR, NULL);
 215                 if (IS_ERR(handle)) {
 216                         err = PTR_ERR(handle);
 217                         CERROR("unable to start transaction: rc %d\n", err);
 218                 } else {
 219                         err = fsfilt_write_record(obd, filter->fo_rcvd_filp,
 220                                                   fed->fed_fcd,
 221                                                   sizeof(*fed->fed_fcd),
 222                                                   &off, 1);
 223                         fsfilt_commit(obd, filter->fo_sb,
 224                                       filter->fo_rcvd_filp->f_dentry->d_inode,
 225                                       handle, 1);
 226                 }
 227                 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 228
 229                 if (err) {
 230                         CERROR("error writing %s client idx %u: rc %d\n",
 231                                LAST_RCVD, fed->fed_lr_idx, err);
 232                         RETURN(err);
 233                 }
 234         }
 235         RETURN(0);
 236 }
 237
 238 static int filter_client_free(struct obd_export *exp, int flags)
 239 {
 240         struct filter_export_data *fed = &exp->exp_filter_data;
 241         struct filter_obd *filter = &exp->exp_obd->u.filter;
 242         struct obd_device *obd = exp->exp_obd;
 243         struct filter_client_data zero_fcd;
 244         struct lvfs_run_ctxt saved;
 245         int rc;
 246         loff_t off;
 247         ENTRY;
 248
 249         if (fed->fed_fcd == NULL)
 250                 RETURN(0);
 251
 252         if (flags & OBD_OPT_FAILOVER)
 253                 GOTO(free, 0);
 254
 255         /* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
 256         if (!strcmp((char *)fed->fed_fcd->fcd_uuid, (char *)obd->obd_uuid.uuid))
 257                 GOTO(free, 0);
 258
 259         LASSERT(filter->fo_last_rcvd_slots != NULL);
 260
 261         off = fed->fed_lr_off;
 262
 263         CDEBUG(D_INFO, "freeing client at idx %u (%lld) with UUID '%s'\n",
 264                fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
 265
 266         /* Clear the bit _after_ zeroing out the client so we don't
 267            race with filter_client_add and zero out new clients.*/
 268         if (!test_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
 269                 CERROR("FILTER client %u: bit already clear in bitmap!!\n",
 270                        fed->fed_lr_idx);
 271                 LBUG();
 272         }
 273
 274         memset(&zero_fcd, 0, sizeof zero_fcd);
 275         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 276         rc = fsfilt_write_record(obd, filter->fo_rcvd_filp, &zero_fcd,
 277                                  sizeof(zero_fcd), &off, 1);
 278         if (rc == 0)
 279                 /* update server's transno */
 280                 filter_update_server_data(obd, filter->fo_rcvd_filp,
 281                                           filter->fo_fsd, 1);
 282         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 283
 284         CDEBUG(rc == 0 ? D_INFO : D_ERROR,
 285                "zeroing disconnecting client %s at idx %u (%llu) in %s rc %d\n",
 286                fed->fed_fcd->fcd_uuid, fed->fed_lr_idx, fed->fed_lr_off,
 287                LAST_RCVD, rc);
 288
 289         if (!test_and_clear_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
 290                 CERROR("FILTER client %u: bit already clear in bitmap!!\n",
 291                        fed->fed_lr_idx);
 292                 LBUG();
 293         }
 294
 295 free:
 296         OBD_FREE(fed->fed_fcd, sizeof(*fed->fed_fcd));
 297
 298         RETURN(0);
 299 }
 300
 301 static int filter_free_server_data(struct filter_obd *filter)
 302 {
 303         OBD_FREE(filter->fo_fsd, sizeof(*filter->fo_fsd));
 304         filter->fo_fsd = NULL;
 305         OBD_FREE(filter->fo_last_rcvd_slots, FILTER_LR_MAX_CLIENTS/8);
 306         filter->fo_last_rcvd_slots = NULL;
 307         return 0;
 308 }
 309
 310 /* assumes caller is already in kernel ctxt */
 311 int filter_update_server_data(struct obd_device *obd, struct file *filp,
 312                               struct filter_server_data *fsd, int force_sync)
 313 {
 314         loff_t off = 0;
 315         int rc;
 316         ENTRY;
 317
 318         CDEBUG(D_INODE, "server uuid      : %s\n", fsd->fsd_uuid);
 319         CDEBUG(D_INODE, "server last_rcvd : "LPU64"\n",
 320                le64_to_cpu(fsd->fsd_last_transno));
 321         CDEBUG(D_INODE, "server last_mount: "LPU64"\n",
 322                le64_to_cpu(fsd->fsd_mount_count));
 323
 324         rc = fsfilt_write_record(obd, filp, fsd, sizeof(*fsd), &off,force_sync);
 325         if (rc)
 326                 CERROR("error writing filter_server_data: rc = %d\n", rc);
 327
 328         RETURN(rc);
 329 }
 330
 331 int filter_update_last_objid(struct obd_device *obd, obd_gr group,
 332                              int force_sync)
 333 {
 334         struct filter_obd *filter = &obd->u.filter;
 335         __u64 tmp;
 336         loff_t off = 0;
 337         int rc;
 338         ENTRY;
 339
 340         if (filter->fo_last_objid_files[group] == NULL) {
 341                 CERROR("Object group "LPU64" not fully setup; not updating "
 342                        "last_objid\n", group);
 343                 RETURN(0);
 344         }
 345
 346         CDEBUG(D_INODE, "server last_objid for group "LPU64": "LPU64"\n",
 347                group, filter->fo_last_objids[group]);
 348
 349         tmp = cpu_to_le64(filter->fo_last_objids[group]);
 350         rc = fsfilt_write_record(obd, filter->fo_last_objid_files[group],
 351                                  &tmp, sizeof(tmp), &off, force_sync);
 352         if (rc)
 353                 CERROR("error writing group "LPU64" last objid: rc = %d\n",
 354                        group, rc);
 355         RETURN(rc);
 356 }
 357
 358 /* assumes caller has already in kernel ctxt */
 359 static int filter_init_server_data(struct obd_device *obd, struct file * filp)
 360 {
 361         struct filter_obd *filter = &obd->u.filter;
 362         struct filter_server_data *fsd;
 363         struct filter_client_data *fcd = NULL;
 364         struct inode *inode = filp->f_dentry->d_inode;
 365         unsigned long last_rcvd_size = inode->i_size;
 366         __u64 mount_count;
 367         int cl_idx;
 368         loff_t off = 0;
 369         int rc;
 370
 371         /* ensure padding in the struct is the correct size */
 372         LASSERT (offsetof(struct filter_server_data, fsd_padding) +
 373                  sizeof(fsd->fsd_padding) == FILTER_LR_SERVER_SIZE);
 374         LASSERT (offsetof(struct filter_client_data, fcd_padding) +
 375                  sizeof(fcd->fcd_padding) == FILTER_LR_CLIENT_SIZE);
 376
 377         OBD_ALLOC(fsd, sizeof(*fsd));
 378         if (!fsd)
 379                 RETURN(-ENOMEM);
 380         filter->fo_fsd = fsd;
 381
 382         OBD_ALLOC(filter->fo_last_rcvd_slots, FILTER_LR_MAX_CLIENTS/8);
 383         if (filter->fo_last_rcvd_slots == NULL) {
 384                 OBD_FREE(fsd, sizeof(*fsd));
 385                 RETURN(-ENOMEM);
 386         }
 387
 388         if (last_rcvd_size == 0) {
 389                 CWARN("%s: initializing new %s\n", obd->obd_name, LAST_RCVD);
 390
 391                 memcpy(fsd->fsd_uuid, obd->obd_uuid.uuid,sizeof(fsd->fsd_uuid));
 392                 fsd->fsd_last_transno = 0;
 393                 mount_count = fsd->fsd_mount_count = 0;
 394                 fsd->fsd_server_size = cpu_to_le32(FILTER_LR_SERVER_SIZE);
 395                 fsd->fsd_client_start = cpu_to_le32(FILTER_LR_CLIENT_START);
 396                 fsd->fsd_client_size = cpu_to_le16(FILTER_LR_CLIENT_SIZE);
 397                 fsd->fsd_subdir_count = cpu_to_le16(FILTER_SUBDIR_COUNT);
 398                 filter->fo_subdir_count = FILTER_SUBDIR_COUNT;
 399         } else {
 400                 rc = fsfilt_read_record(obd, filp, fsd, sizeof(*fsd), &off);
 401                 if (rc) {
 402                         CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
 403                                LAST_RCVD, rc);
 404                         GOTO(err_fsd, rc);
 405                 }
 406                 if (strcmp((char *)fsd->fsd_uuid, (char *)obd->obd_uuid.uuid)) {
 407                         CERROR("OBD UUID %s does not match last_rcvd UUID %s\n",
 408                                obd->obd_uuid.uuid, fsd->fsd_uuid);
 409                         GOTO(err_fsd, rc = -EINVAL);
 410                 }
 411                 mount_count = le64_to_cpu(fsd->fsd_mount_count);
 412                 filter->fo_subdir_count = le16_to_cpu(fsd->fsd_subdir_count);
 413         }
 414
 415         if (fsd->fsd_feature_incompat & ~cpu_to_le32(FILTER_INCOMPAT_SUPP)) {
 416                 CERROR("unsupported feature %x\n",
 417                        le32_to_cpu(fsd->fsd_feature_incompat) &
 418                        ~FILTER_INCOMPAT_SUPP);
 419                 GOTO(err_fsd, rc = -EINVAL);
 420         }
 421         if (fsd->fsd_feature_rocompat & ~cpu_to_le32(FILTER_ROCOMPAT_SUPP)) {
 422                 CERROR("read-only feature %x\n",
 423                        le32_to_cpu(fsd->fsd_feature_rocompat) &
 424                        ~FILTER_ROCOMPAT_SUPP);
 425                 /* Do something like remount filesystem read-only */
 426                 GOTO(err_fsd, rc = -EINVAL);
 427         }
 428
 429         CDEBUG(D_INODE, "%s: server last_rcvd : "LPU64"\n",
 430                obd->obd_name, le64_to_cpu(fsd->fsd_last_transno));
 431         CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
 432                obd->obd_name, mount_count + 1);
 433         CDEBUG(D_INODE, "%s: server data size: %u\n",
 434                obd->obd_name, le32_to_cpu(fsd->fsd_server_size));
 435         CDEBUG(D_INODE, "%s: per-client data start: %u\n",
 436                obd->obd_name, le32_to_cpu(fsd->fsd_client_start));
 437         CDEBUG(D_INODE, "%s: per-client data size: %u\n",
 438                obd->obd_name, le32_to_cpu(fsd->fsd_client_size));
 439         CDEBUG(D_INODE, "%s: server subdir_count: %u\n",
 440                obd->obd_name, le16_to_cpu(fsd->fsd_subdir_count));
 441         CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
 442                last_rcvd_size <= le32_to_cpu(fsd->fsd_client_start) ? 0 :
 443                (last_rcvd_size - le32_to_cpu(fsd->fsd_client_start)) /
 444                 le16_to_cpu(fsd->fsd_client_size));
 445
 446         if (!obd->obd_replayable) {
 447                 CWARN("%s: recovery support OFF\n", obd->obd_name);
 448                 GOTO(out, rc = 0);
 449         }
 450
 451         for (cl_idx = 0, off = le32_to_cpu(fsd->fsd_client_start);
 452              off < last_rcvd_size; cl_idx++) {
 453                 __u64 last_rcvd;
 454                 struct obd_export *exp;
 455                 struct filter_export_data *fed;
 456
 457                 if (!fcd) {
 458                         OBD_ALLOC(fcd, sizeof(*fcd));
 459                         if (!fcd)
 460                                 GOTO(err_client, rc = -ENOMEM);
 461                 }
 462
 463                 /* Don't assume off is incremented properly by
 464                  * fsfilt_read_record(), in case sizeof(*fcd)
 465                  * isn't the same as fsd->fsd_client_size.  */
 466                 off = le32_to_cpu(fsd->fsd_client_start) +
 467                         cl_idx * le16_to_cpu(fsd->fsd_client_size);
 468                 rc = fsfilt_read_record(obd, filp, fcd, sizeof(*fcd), &off);
 469                 if (rc) {
 470                         CERROR("error reading FILT %s idx %d off %llu: rc %d\n",
 471                                LAST_RCVD, cl_idx, off, rc);
 472                         break; /* read error shouldn't cause startup to fail */
 473                 }
 474
 475                 if (fcd->fcd_uuid[0] == '\0') {
 476                         CDEBUG(D_INFO, "skipping zeroed client at offset %d\n",
 477                                cl_idx);
 478                         continue;
 479                 }
 480
 481                 last_rcvd = le64_to_cpu(fcd->fcd_last_rcvd);
 482
 483                 /* These exports are cleaned up by filter_disconnect(), so they
 484                  * need to be set up like real exports as filter_connect() does.
 485                  */
 486                 exp = class_new_export(obd);
 487                 CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64" "
 488                        "srv lr: "LPU64" fcd_group %d \n", fcd->fcd_uuid, cl_idx,
 489                        last_rcvd, le64_to_cpu(fsd->fsd_last_transno),
 490                        le32_to_cpu(fcd->fcd_group));
 491                 if (exp == NULL)
 492                         GOTO(err_client, rc = -ENOMEM);
 493
 494                 memcpy(&exp->exp_client_uuid.uuid, fcd->fcd_uuid,
 495                        sizeof exp->exp_client_uuid.uuid);
 496                 fed = &exp->exp_filter_data;
 497                 fed->fed_fcd = fcd;
 498                 fed->fed_group = le32_to_cpu(fcd->fcd_group);
 499                 filter_client_add(obd, filter, fed, cl_idx);
 500                 /* create helper if export init gets more complex */
 501                 spin_lock_init(&fed->fed_lock);
 502
 503                 fcd = NULL;
 504                 exp->exp_connected = 0;
 505                 exp->exp_req_replay_needed = 1;
 506                 exp->exp_lock_replay_needed = 1;
 507                 atomic_inc(&obd->obd_req_replay_clients);
 508                 atomic_inc(&obd->obd_lock_replay_clients);
 509                 obd->obd_recoverable_clients++;
 510                 obd->obd_max_recoverable_clients++;
 511                 class_export_put(exp);
 512
 513                 CDEBUG(D_OTHER, "client at idx %d has last_rcvd = "LPU64"\n",
 514                        cl_idx, last_rcvd);
 515
 516                 if (last_rcvd > le64_to_cpu(fsd->fsd_last_transno))
 517                         fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
 518
 519         }
 520
 521         obd->obd_last_committed = le64_to_cpu(fsd->fsd_last_transno);
 522
 523         if (obd->obd_recoverable_clients) {
 524                 CWARN("RECOVERY: service %s, %d recoverable clients, "
 525                       "last_transno "LPU64"\n", obd->obd_name,
 526                       obd->obd_recoverable_clients,
 527                       le64_to_cpu(fsd->fsd_last_transno));
 528                 obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
 529                 target_start_recovery_thread(obd, ost_handle);
 530         }
 531
 532         if (fcd)
 533                 OBD_FREE(fcd, sizeof(*fcd));
 534
 535 out:
 536         filter->fo_mount_count = mount_count + 1;
 537         fsd->fsd_mount_count = cpu_to_le64(filter->fo_mount_count);
 538
 539         /* save it, so mount count and last_transno is current */
 540         rc = filter_update_server_data(obd, filp, filter->fo_fsd, 1);
 541         if (rc)
 542                 GOTO(err_client, rc);
 543         RETURN(0);
 544
 545 err_client:
 546         class_disconnect_exports(obd, 0);
 547 err_fsd:
 548         filter_free_server_data(filter);
 549         RETURN(rc);
 550 }
 551
 552 static int filter_cleanup_groups(struct obd_device *obd)
 553 {
 554         struct filter_obd *filter = &obd->u.filter;
 555         struct dentry *dentry;
 556         int i, k;
 557         ENTRY;
 558
 559         for (i = 0; i < filter->fo_group_count; i++) {
 560                 if (filter->fo_subdirs != NULL) {
 561                         for (k = 0; k < filter->fo_subdir_count; k++) {
 562                                 dentry = filter->fo_subdirs[i].dentry[k];
 563                                 if (dentry == NULL)
 564                                         continue;
 565                                 f_dput(dentry);
 566                                 filter->fo_subdirs[i].dentry[k] = NULL;
 567                         }
 568                 }
 569                 if (filter->fo_last_objid_files[i] != NULL) {
 570                         filp_close(filter->fo_last_objid_files[i], 0);
 571                         filter->fo_last_objid_files[i] = NULL;
 572                 }
 573                 if (filter->fo_groups[i] != NULL) {
 574                         dput(filter->fo_groups[i]);
 575                         filter->fo_groups[i] = NULL;
 576                 }
 577         }
 578         if (filter->fo_subdirs != NULL)
 579                 OBD_FREE(filter->fo_subdirs,
 580                          filter->fo_group_count * sizeof(*filter->fo_subdirs));
 581         if (filter->fo_groups != NULL)
 582                 OBD_FREE(filter->fo_groups,
 583                          filter->fo_group_count * sizeof(*filter->fo_groups));
 584         if (filter->fo_last_objids != NULL)
 585                 OBD_FREE(filter->fo_last_objids,
 586                          filter->fo_group_count * sizeof(__u64));
 587         if (filter->fo_last_objid_files != NULL)
 588                 OBD_FREE(filter->fo_last_objid_files,
 589                          filter->fo_group_count * sizeof(struct file *));
 590         f_dput(filter->fo_dentry_O);
 591         RETURN(0);
 592 }
 593
 594 static int filter_update_last_group(struct obd_device *obd, int group)
 595 {
 596         struct filter_obd *filter = &obd->u.filter;
 597         struct file *filp = NULL;
 598         int last_group = 0, rc;
 599         loff_t off = 0;
 600         ENTRY;
 601
 602         if (group <= filter->fo_committed_group)
 603                 RETURN(0);
 604
 605         filp = filp_open("LAST_GROUP", O_RDWR, 0700);
 606         if (IS_ERR(filp)) {
 607                 rc = PTR_ERR(filp);
 608                 filp = NULL;
 609                 CERROR("cannot open LAST_GROUP: rc = %d\n", rc);
 610                 GOTO(cleanup, rc);
 611         }
 612
 613         rc = fsfilt_read_record(obd, filp, &last_group, sizeof(__u32), &off);
 614         if (rc) {
 615                 CDEBUG(D_INODE, "error reading LAST_GROUP: rc %d\n",rc);
 616                 GOTO(cleanup, rc);
 617         }
 618         LASSERT(off == 0 || last_group >= FILTER_MIN_GROUPS);
 619         CDEBUG(D_INODE, "%s: previous %d, new %d\n",
 620                obd->obd_name, last_group, group);
 621
 622         off = 0;
 623         last_group = group;
 624         /* must be sync: bXXXX */
 625         rc = fsfilt_write_record(obd, filp, &last_group, sizeof(__u32), &off, 1);
 626         if (rc) {
 627                 CDEBUG(D_INODE, "error updating LAST_GROUP: rc %d\n", rc);
 628                 GOTO(cleanup, rc);
 629         }
 630
 631         filter->fo_committed_group = group;
 632 cleanup:
 633         if (filp)
 634                 filp_close(filp, 0);
 635         RETURN(rc);
 636 }
 637
 638 static int filter_read_group_internal(struct obd_device *obd, int group,
 639                                       int create)
 640 {
 641         struct filter_obd *filter = &obd->u.filter;
 642         __u64 *new_objids = NULL;
 643         struct filter_subdirs *new_subdirs = NULL, *tmp_subdirs = NULL;
 644         struct dentry **new_groups = NULL;
 645         struct file **new_files = NULL;
 646         struct dentry *dentry;
 647         struct file *filp;
 648         int old_count = filter->fo_group_count, rc, stage = 0, i;
 649         char name[25];
 650         __u64 last_objid;
 651         loff_t off = 0;
 652
 653         snprintf(name, 24, "%d", group);
 654         name[24] = '\0';
 655
 656         if (!create) {
 657                 dentry = ll_lookup_one_len(name, filter->fo_dentry_O,
 658                                            strlen(name));
 659                 if (IS_ERR(dentry)) {
 660                         CERROR("Cannot lookup expected object group %d: %ld\n",
 661                                group, PTR_ERR(dentry));
 662                         RETURN(PTR_ERR(dentry));
 663                 }
 664         } else {
 665                 dentry = simple_mkdir(filter->fo_dentry_O, name, 0700, 1);
 666                 if (IS_ERR(dentry)) {
 667                         CERROR("cannot lookup/create O/%s: rc = %ld\n", name,
 668                                PTR_ERR(dentry));
 669                         RETURN(PTR_ERR(dentry));
 670                 }
 671         }
 672         stage = 1;
 673
 674         snprintf(name, 24, "O/%d/LAST_ID", group);
 675         name[24] = '\0';
 676         filp = filp_open(name, O_CREAT | O_RDWR, 0700);
 677         if (IS_ERR(filp)) {
 678                 CERROR("cannot create %s: rc = %ld\n", name, PTR_ERR(filp));
 679                 GOTO(cleanup, rc = PTR_ERR(filp));
 680         }
 681         stage = 2;
 682
 683         rc = fsfilt_read_record(obd, filp, &last_objid, sizeof(__u64), &off);
 684         if (rc) {
 685                 CDEBUG(D_INODE, "error reading %s: rc %d\n", name, rc);
 686                 GOTO(cleanup, rc);
 687         }
 688
 689         if (filter->fo_subdir_count) {
 690                 OBD_ALLOC(tmp_subdirs, sizeof(*tmp_subdirs));
 691                 if (tmp_subdirs == NULL)
 692                         GOTO(cleanup, rc = -ENOMEM);
 693                 stage = 3;
 694
 695                 for (i = 0; i < filter->fo_subdir_count; i++) {
 696                         char dir[20];
 697                         snprintf(dir, sizeof(dir), "d%u", i);
 698
 699                         tmp_subdirs->dentry[i] =
 700                                 simple_mkdir(dentry, dir, 0700, 1);
 701                         if (IS_ERR(tmp_subdirs->dentry[i])) {
 702                                 rc = PTR_ERR(tmp_subdirs->dentry[i]);
 703                                 CERROR("can't lookup/create O/%d/%s: rc = %d\n",
 704                                        group, dir, rc);
 705                                 GOTO(cleanup, rc);
 706                         }
 707                         CDEBUG(D_INODE, "got/created O/%d/%s: %p\n", group, dir,
 708                                tmp_subdirs->dentry[i]);
 709                 }
 710         }
 711
 712         /* 'group' is an index; we need an array of length 'group + 1' */
 713         if (group + 1 > old_count) {
 714                 int len = group + 1;
 715                 OBD_ALLOC(new_objids, len * sizeof(*new_objids));
 716                 OBD_ALLOC(new_subdirs, len * sizeof(*new_subdirs));
 717                 OBD_ALLOC(new_groups, len * sizeof(*new_groups));
 718                 OBD_ALLOC(new_files, len * sizeof(*new_files));
 719                 stage = 4;
 720                 if (new_objids == NULL || new_subdirs == NULL ||
 721                     new_groups == NULL || new_files == NULL)
 722                         GOTO(cleanup, rc = -ENOMEM);
 723
 724                 memcpy(new_objids, filter->fo_last_objids,
 725                        old_count * sizeof(*new_objids));
 726                 memcpy(new_subdirs, filter->fo_subdirs,
 727                        old_count * sizeof(*new_subdirs));
 728                 memcpy(new_groups, filter->fo_groups,
 729                        old_count * sizeof(*new_groups));
 730                 memcpy(new_files, filter->fo_last_objid_files,
 731                        old_count * sizeof(*new_files));
 732
 733                 if (old_count) {
 734                         OBD_FREE(filter->fo_last_objids,
 735                                  old_count * sizeof(*new_objids));
 736                         OBD_FREE(filter->fo_subdirs,
 737                                  old_count * sizeof(*new_subdirs));
 738                         OBD_FREE(filter->fo_groups,
 739                                  old_count * sizeof(*new_groups));
 740                         OBD_FREE(filter->fo_last_objid_files,
 741                                  old_count * sizeof(*new_files));
 742                 }
 743                 filter->fo_last_objids = new_objids;
 744                 filter->fo_subdirs = new_subdirs;
 745                 filter->fo_groups = new_groups;
 746                 filter->fo_last_objid_files = new_files;
 747                 filter->fo_group_count = len;
 748         }
 749
 750         filter->fo_groups[group] = dentry;
 751         filter->fo_last_objid_files[group] = filp;
 752         if (filter->fo_subdir_count) {
 753                 filter->fo_subdirs[group] = *tmp_subdirs;
 754                 OBD_FREE(tmp_subdirs, sizeof(*tmp_subdirs));
 755         }
 756
 757         filter_update_last_group(obd, group);
 758
 759         if (filp->f_dentry->d_inode->i_size == 0) {
 760                 filter->fo_last_objids[group] = FILTER_INIT_OBJID;
 761                 RETURN(0);
 762         }
 763
 764         filter->fo_last_objids[group] = le64_to_cpu(last_objid);
 765         CDEBUG(D_INODE, "%s: server last_objid group %d: "LPU64"\n",
 766                obd->obd_name, group, last_objid);
 767         RETURN(0);
 768  cleanup:
 769         switch (stage) {
 770         case 4:
 771                 if (new_objids != NULL)
 772                         OBD_FREE(new_objids, group * sizeof(*new_objids));
 773                 if (new_subdirs != NULL)
 774                         OBD_FREE(new_subdirs, group * sizeof(*new_subdirs));
 775                 if (new_groups != NULL)
 776                         OBD_FREE(new_groups, group * sizeof(*new_groups));
 777                 if (new_files != NULL)
 778                         OBD_FREE(new_files, group * sizeof(*new_files));
 779         case 3:
 780                 if (filter->fo_subdir_count) {
 781                         for (i = 0; i < filter->fo_subdir_count; i++) {
 782                                 if (tmp_subdirs->dentry[i] != NULL)
 783                                         dput(tmp_subdirs->dentry[i]);
 784                         }
 785                         OBD_FREE(tmp_subdirs, sizeof(*tmp_subdirs));
 786                 }
 787         case 2:
 788                 filp_close(filp, 0);
 789         case 1:
 790                 dput(dentry);
 791         }
 792         RETURN(rc);
 793 }
 794
 795 static int filter_read_groups(struct obd_device *obd, int last_group,
 796                               int create)
 797 {
 798         struct filter_obd *filter = &obd->u.filter;
 799         int old_count, group, rc = 0;
 800
 801         down(&filter->fo_init_lock);
 802         old_count = filter->fo_group_count;
 803         for (group = old_count; group <= last_group; group++) {
 804                 if (group == 0)
 805                         continue; /* no group zero */
 806
 807                 rc = filter_read_group_internal(obd, group, create);
 808                 if (rc != 0)
 809                         break;
 810         }
 811         up(&filter->fo_init_lock);
 812         return rc;
 813 }
 814
 815 static int filter_prep_groups(struct obd_device *obd)
 816 {
 817         struct filter_obd *filter = &obd->u.filter;
 818         struct dentry *dentry, *O_dentry;
 819         int rc = 0, cleanup_phase = 0;
 820         struct file *filp = NULL;
 821         int last_group;
 822         loff_t off = 0;
 823         ENTRY;
 824
 825         O_dentry = simple_mkdir(current->fs->pwd, "O", 0700, 1);
 826         CDEBUG(D_INODE, "got/created O: %p\n", O_dentry);
 827         if (IS_ERR(O_dentry)) {
 828                 rc = PTR_ERR(O_dentry);
 829                 CERROR("cannot open/create O: rc = %d\n", rc);
 830                 GOTO(cleanup, rc);
 831         }
 832         filter->fo_dentry_O = O_dentry;
 833         cleanup_phase = 1; /* O_dentry */
 834
 835         /* Lookup "R" to tell if we're on an old OST FS and need to convert
 836          * from O/R/<dir>/<objid> to O/0/<dir>/<objid>.  This can be removed
 837          * some time post 1.0 when all old-style OSTs have converted along
 838          * with the init_objid hack. */
 839         dentry = ll_lookup_one_len("R", O_dentry, 1);
 840         if (IS_ERR(dentry))
 841                 GOTO(cleanup, rc = PTR_ERR(dentry));
 842         if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
 843                 struct dentry *O0_dentry = lookup_one_len("0", O_dentry, 1);
 844                 ENTRY;
 845
 846                 CWARN("converting OST to new object layout\n");
 847                 if (IS_ERR(O0_dentry)) {
 848                         rc = PTR_ERR(O0_dentry);
 849                         CERROR("error looking up O/0: rc %d\n", rc);
 850                         GOTO(cleanup_R, rc);
 851                 }
 852
 853                 if (O0_dentry->d_inode) {
 854                         CERROR("Both O/R and O/0 exist. Fix manually.\n");
 855                         GOTO(cleanup_O0, rc = -EEXIST);
 856                 }
 857
 858                 down(&O_dentry->d_inode->i_sem);
 859                 rc = vfs_rename(O_dentry->d_inode, dentry,
 860                                 O_dentry->d_inode, O0_dentry);
 861                 up(&O_dentry->d_inode->i_sem);
 862
 863                 if (rc) {
 864                         CERROR("error renaming O/R to O/0: rc %d\n", rc);
 865                         GOTO(cleanup_O0, rc);
 866                 }
 867                 filter->fo_fsd->fsd_feature_incompat |=
 868                         cpu_to_le32(FILTER_INCOMPAT_GROUPS);
 869                 rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
 870                                                filter->fo_fsd, 1);
 871                 GOTO(cleanup_O0, rc);
 872
 873         cleanup_O0:
 874                 f_dput(O0_dentry);
 875         cleanup_R:
 876                 f_dput(dentry);
 877                 if (rc)
 878                         GOTO(cleanup, rc);
 879         } else {
 880                 f_dput(dentry);
 881         }
 882
 883         cleanup_phase = 2; /* groups */
 884
 885         /* we have to initialize all groups before first connections from
 886          * clients because they may send create/destroy for any group -bzzz */
 887         filp = filp_open("LAST_GROUP", O_CREAT | O_RDWR, 0700);
 888         if (IS_ERR(filp)) {
 889                 CERROR("cannot create LAST_GROUP: rc = %ld\n", PTR_ERR(filp));
 890                 GOTO(cleanup, rc = PTR_ERR(filp));
 891         }
 892         cleanup_phase = 3; /* filp */
 893
 894         rc = fsfilt_read_record(obd, filp, &last_group, sizeof(__u32), &off);
 895         if (rc) {
 896                 CDEBUG(D_INODE, "error reading LAST_GROUP: rc %d\n", rc);
 897                 GOTO(cleanup, rc);
 898         }
 899         if (off == 0) {
 900                 last_group = FILTER_MIN_GROUPS;
 901         } else {
 902                 LASSERT(last_group >= FILTER_MIN_GROUPS);
 903         }
 904
 905         CWARN("%s: initialize groups [%d,%d]\n", obd->obd_name,
 906               FILTER_MIN_GROUPS, last_group);
 907         filter->fo_committed_group = last_group;
 908         rc = filter_read_groups(obd, last_group, 1);
 909         if (rc)
 910                 GOTO(cleanup, rc);
 911
 912         filp_close(filp, 0);
 913         RETURN(0);
 914
 915  cleanup:
 916         switch (cleanup_phase) {
 917         case 3:
 918                 filp_close(filp, 0);
 919         case 2:
 920                 filter_cleanup_groups(obd);
 921         case 1:
 922                 f_dput(filter->fo_dentry_O);
 923                 filter->fo_dentry_O = NULL;
 924         default:
 925                 break;
 926         }
 927         return rc;
 928 }
 929
 930 /* setup the object store with correct subdirectories */
 931 static int filter_prep(struct obd_device *obd)
 932 {
 933         struct lvfs_run_ctxt saved;
 934         struct filter_obd *filter = &obd->u.filter;
 935         struct file *file;
 936         struct inode *inode;
 937         int rc = 0;
 938         ENTRY;
 939
 940         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 941         file = filp_open(LAST_RCVD, O_RDWR | O_CREAT | O_LARGEFILE, 0700);
 942         if (!file || IS_ERR(file)) {
 943                 rc = PTR_ERR(file);
 944                 CERROR("OBD filter: cannot open/create %s: rc = %d\n",
 945                        LAST_RCVD, rc);
 946                 GOTO(out, rc);
 947         }
 948
 949         if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
 950                 CERROR("%s is not a regular file!: mode = %o\n", LAST_RCVD,
 951                        file->f_dentry->d_inode->i_mode);
 952                 GOTO(err_filp, rc = -ENOENT);
 953         }
 954
 955         /* steal operations */
 956         inode = file->f_dentry->d_inode;
 957         filter->fo_fop = file->f_op;
 958         filter->fo_iop = inode->i_op;
 959         filter->fo_aops = inode->i_mapping->a_ops;
 960
 961         rc = filter_init_server_data(obd, file);
 962         if (rc) {
 963                 CERROR("cannot read %s: rc = %d\n", LAST_RCVD, rc);
 964                 GOTO(err_filp, rc);
 965         }
 966         filter->fo_rcvd_filp = file;
 967
 968         rc = filter_prep_groups(obd);
 969         if (rc)
 970                 GOTO(err_server_data, rc);
 971
 972  out:
 973         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 974
 975         return(rc);
 976
 977  err_server_data:
 978         //class_disconnect_exports(obd, 0);
 979         filter_free_server_data(filter);
 980  err_filp:
 981         if (filp_close(file, 0))
 982                 CERROR("can't close %s after error\n", LAST_RCVD);
 983         filter->fo_rcvd_filp = NULL;
 984         goto out;
 985 }
 986
 987 /* cleanup the filter: write last used object id to status file */
 988 static void filter_post(struct obd_device *obd)
 989 {
 990         struct lvfs_run_ctxt saved;
 991         struct filter_obd *filter = &obd->u.filter;
 992         int rc, i;
 993
 994         /* XXX: filter_update_lastobjid used to call fsync_dev.  It might be
 995          * best to start a transaction with h_sync, because we removed this
 996          * from lastobjid */
 997
 998         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 999         rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
1000                                        filter->fo_fsd, 0);
1001         if (rc)
1002                 CERROR("error writing server data: rc = %d\n", rc);
1003
1004         for (i = 1; i < filter->fo_group_count; i++) {
1005                 rc = filter_update_last_objid(obd, i,
1006                                              (i == filter->fo_group_count - 1));
1007                 if (rc)
1008                         CERROR("error writing group %d lastobjid: rc = %d\n",
1009                                i, rc);
1010         }
1011
1012         rc = filp_close(filter->fo_rcvd_filp, 0);
1013         filter->fo_rcvd_filp = NULL;
1014         if (rc)
1015                 CERROR("error closing %s: rc = %d\n", LAST_RCVD, rc);
1016
1017         filter_cleanup_groups(obd);
1018         filter_free_server_data(filter);
1019         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1020 }
1021
1022 static void filter_set_last_id(struct filter_obd *filter, int group, obd_id id)
1023 {
1024         LASSERT(filter->fo_fsd != NULL);
1025         LASSERT(group > 0);
1026         LASSERT(group < filter->fo_group_count);
1027
1028         spin_lock(&filter->fo_objidlock);
1029         filter->fo_last_objids[group] = id;
1030         spin_unlock(&filter->fo_objidlock);
1031 }
1032
1033 __u64 filter_last_id(struct filter_obd *filter, int group)
1034 {
1035         obd_id id;
1036         LASSERT(filter->fo_fsd != NULL);
1037         LASSERT(group > 0);
1038         LASSERT(group < filter->fo_group_count);
1039
1040         spin_lock(&filter->fo_objidlock);
1041         id = filter->fo_last_objids[group];
1042         spin_unlock(&filter->fo_objidlock);
1043
1044         return id;
1045 }
1046
1047 static void filter_save_last_id(struct filter_obd *filter, int group, obd_id id)
1048 {
1049         LASSERT(group > 0);
1050         LASSERT(group < filter->fo_group_count);
1051
1052         spin_lock(&filter->fo_lastidlock);
1053         if (id > filter_last_id(filter, group))
1054                 filter_set_last_id(filter, group, id);
1055         spin_unlock(&filter->fo_lastidlock);
1056 }
1057
1058 /* direct cut-n-paste of mds_blocking_ast() */
1059 static int filter_blocking_ast(struct ldlm_lock *lock,
1060                                struct ldlm_lock_desc *desc,
1061                                void *data, int flag)
1062 {
1063         int do_ast;
1064         ENTRY;
1065
1066         if (flag == LDLM_CB_CANCELING) {
1067                 /* Don't need to do anything here. */
1068                 RETURN(0);
1069         }
1070
1071         /* XXX layering violation!  -phil */
1072         lock_res_and_lock(lock);
1073         /* Get this: if filter_blocking_ast is racing with ldlm_intent_policy,
1074          * such that filter_blocking_ast is called just before l_i_p takes the
1075          * ns_lock, then by the time we get the lock, we might not be the
1076          * correct blocking function anymore.  So check, and return early, if
1077          * so. */
1078         if (lock->l_blocking_ast != filter_blocking_ast) {
1079                 unlock_res_and_lock(lock);
1080                 RETURN(0);
1081         }
1082
1083         lock->l_flags |= LDLM_FL_CBPENDING;
1084         do_ast = (!lock->l_readers && !lock->l_writers);
1085         unlock_res_and_lock(lock);
1086
1087         if (do_ast) {
1088                 struct lustre_handle lockh;
1089                 int rc;
1090
1091                 LDLM_DEBUG(lock, "already unused, calling ldlm_cli_cancel");
1092                 ldlm_lock2handle(lock, &lockh);
1093                 rc = ldlm_cli_cancel(&lockh);
1094                 if (rc < 0)
1095                         CERROR("ldlm_cli_cancel: %d\n", rc);
1096         } else {
1097                 LDLM_DEBUG(lock, "Lock still has references, will be "
1098                            "cancelled later");
1099         }
1100         RETURN(0);
1101 }
1102
1103 extern void *lock_dir(struct inode *dir, struct qstr *name);
1104 extern void unlock_dir(struct inode *dir, void *lock);
1105
1106 static void *filter_lock_dentry(struct obd_device *obd,
1107                                 struct dentry *dparent,
1108                                 obd_id id)
1109 {
1110 #ifdef S_PDIROPS
1111         struct qstr qstr;
1112         char name[32];
1113         int len;
1114
1115         len = sprintf(name, LPU64, id);
1116         qstr_assign(&qstr, (char *)name, len);
1117         return lock_dir(dparent->d_inode, &qstr);
1118 #else
1119         down(&dparent->d_inode->i_sem);
1120 #endif
1121         return 0;
1122 }
1123
1124 /* We never dget the object parent, so DON'T dput it either */
1125 static void filter_parent_unlock(struct dentry *dparent, void *lock)
1126 {
1127 #ifdef S_PDIROPS
1128         LASSERT(lock != NULL);
1129         unlock_dir(dparent->d_inode, lock);
1130 #else
1131         up(&dparent->d_inode->i_sem);
1132 #endif
1133 }
1134
1135 /* We never dget the object parent, so DON'T dput it either */
1136 struct dentry *filter_parent(struct obd_device *obd, obd_gr group, obd_id objid)
1137 {
1138         struct filter_obd *filter = &obd->u.filter;
1139         LASSERT(group < filter->fo_group_count);
1140         LASSERT(group > 0);
1141
1142         if (filter->fo_subdir_count == 0)
1143                 return filter->fo_groups[group];
1144
1145         return filter->fo_subdirs[group].dentry[objid & (filter->fo_subdir_count - 1)];
1146 }
1147
1148 /* We never dget the object parent, so DON'T dput it either */
1149 struct dentry *filter_parent_lock(struct obd_device *obd, obd_gr group,
1150                                   obd_id objid, void **lock)
1151 {
1152         unsigned long now = jiffies;
1153         struct dentry *dparent = filter_parent(obd, group, objid);
1154
1155         if (IS_ERR(dparent))
1156                 return dparent;
1157
1158         LASSERT(dparent);
1159         LASSERT(dparent->d_inode);
1160
1161         *lock = filter_lock_dentry(obd, dparent, objid);
1162         fsfilt_check_slow(now, obd_timeout, "parent lock");
1163         return dparent;
1164 }
1165
1166 /* How to get files, dentries, inodes from object id's.
1167  *
1168  * If dir_dentry is passed, the caller has already locked the parent
1169  * appropriately for this operation (normally a write lock).  If
1170  * dir_dentry is NULL, we do a read lock while we do the lookup to
1171  * avoid races with create/destroy and such changing the directory
1172  * internal to the filesystem code. */
1173 struct dentry *filter_id2dentry(struct obd_device *obd,
1174                                 struct dentry *dir_dentry,
1175                                 obd_gr group, obd_id id)
1176 {
1177         struct dentry *dparent = dir_dentry;
1178         struct dentry *dchild;
1179         void *lock = NULL;
1180         char name[32];
1181         int len;
1182         ENTRY;
1183
1184         if (id == 0) {
1185                 CERROR("fatal: invalid object id 0\n");
1186                 RETURN(ERR_PTR(-ESTALE));
1187         }
1188
1189         len = sprintf(name, LPU64, id);
1190         if (dir_dentry == NULL) {
1191                 dparent = filter_parent_lock(obd, group, id, &lock);
1192                 if (IS_ERR(dparent)) {
1193                         CERROR("%s: error getting object "LPU64":"LPU64
1194                                " parent: rc %ld\n", obd->obd_name,
1195                                id, group, PTR_ERR(dparent));
1196                         RETURN(dparent);
1197                 }
1198         }
1199         CDEBUG(D_INODE, "looking up object O/%.*s/%s\n",
1200                dparent->d_name.len, dparent->d_name.name, name);
1201         dchild = /*ll_*/lookup_one_len(name, dparent, len);
1202         if (dir_dentry == NULL)
1203                 filter_parent_unlock(dparent, lock);
1204         if (IS_ERR(dchild)) {
1205                 CERROR("%s: child lookup error %ld\n", obd->obd_name,
1206                        PTR_ERR(dchild));
1207                 RETURN(dchild);
1208         }
1209
1210         if (dchild->d_inode != NULL && is_bad_inode(dchild->d_inode)) {
1211                 CERROR("%s: got bad inode "LPU64"\n", obd->obd_name, id);
1212                 f_dput(dchild);
1213                 RETURN(ERR_PTR(-ENOENT));
1214         }
1215
1216         CDEBUG(D_INODE, "got child objid %s: %p, count = %d\n",
1217                name, dchild, atomic_read(&dchild->d_count));
1218
1219         LASSERT(atomic_read(&dchild->d_count) > 0);
1220
1221         RETURN(dchild);
1222 }
1223
1224 static int filter_prepare_destroy(struct obd_device *obd, obd_id objid,
1225                                   obd_id group)
1226 {
1227         struct lustre_handle lockh;
1228         int flags = LDLM_AST_DISCARD_DATA, rc;
1229         struct ldlm_res_id res_id = { .name = { objid, 0, group, 0 } };
1230         ldlm_policy_data_t policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
1231
1232         ENTRY;
1233         /* Tell the clients that the object is gone now and that they should
1234          * throw away any cached pages. */
1235         rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, res_id,
1236                               LDLM_EXTENT, &policy, LCK_PW,
1237                               &flags, filter_blocking_ast, ldlm_completion_ast,
1238                               NULL, NULL, NULL, 0, NULL, &lockh);
1239
1240         /* We only care about the side-effects, just drop the lock. */
1241         if (rc == ELDLM_OK)
1242                 ldlm_lock_decref(&lockh, LCK_PW);
1243
1244         RETURN(rc);
1245 }
1246
1247 /* Caller must hold LCK_PW on parent and push us into kernel context.
1248  * Caller is also required to ensure that dchild->d_inode exists. */
1249 static int filter_destroy_internal(struct obd_device *obd, obd_id objid,
1250                                    struct dentry *dparent,
1251                                    struct dentry *dchild)
1252 {
1253         struct inode *inode = dchild->d_inode;
1254         int rc;
1255         ENTRY;
1256
1257         if (inode->i_nlink != 1 || atomic_read(&inode->i_count) != 1) {
1258                 CERROR("destroying objid %.*s nlink = %lu, count = %d\n",
1259                        dchild->d_name.len, dchild->d_name.name,
1260                        (unsigned long)inode->i_nlink,
1261                        atomic_read(&inode->i_count));
1262         }
1263
1264         rc = vfs_unlink(dparent->d_inode, dchild);
1265
1266         if (rc)
1267                 CERROR("error unlinking objid %.*s: rc %d\n",
1268                        dchild->d_name.len, dchild->d_name.name, rc);
1269
1270         RETURN(rc);
1271 }
1272
1273 static int filter_intent_policy(struct ldlm_namespace *ns,
1274                                 struct ldlm_lock **lockp, void *req_cookie,
1275                                 ldlm_mode_t mode, int flags, void *data)
1276 {
1277         struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
1278         struct ptlrpc_request *req = req_cookie;
1279         struct ldlm_lock *lock = *lockp, *l = NULL;
1280         struct ldlm_resource *res = lock->l_resource;
1281         ldlm_processing_policy policy;
1282         struct ost_lvb *res_lvb, *reply_lvb;
1283         struct ldlm_reply *rep;
1284         struct list_head *tmp;
1285         ldlm_error_t err;
1286         int tmpflags = 0, rc, repsize[2] = {sizeof(*rep), sizeof(*reply_lvb)};
1287         ENTRY;
1288
1289         policy = ldlm_get_processing_policy(res);
1290         LASSERT(policy != NULL);
1291         LASSERT(req != NULL);
1292
1293         rc = lustre_pack_reply(req, 2, repsize, NULL);
1294         if (rc)
1295                 RETURN(req->rq_status = rc);
1296
1297         rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*rep));
1298         LASSERT(rep != NULL);
1299
1300         reply_lvb = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*reply_lvb));
1301         LASSERT(reply_lvb != NULL);
1302
1303         //fixup_handle_for_resent_req(req, lock, &lockh);
1304
1305         /* If we grant any lock at all, it will be a whole-file read lock.
1306          * Call the extent policy function to see if our request can be
1307          * granted, or is blocked. */
1308         lock->l_policy_data.l_extent.start = 0;
1309         lock->l_policy_data.l_extent.end = OBD_OBJECT_EOF;
1310         lock->l_req_mode = LCK_PR;
1311
1312         lock_res_and_lock(lock);
1313         res = lock->l_resource;
1314         rc = policy(lock, &tmpflags, 0, &err, &rpc_list);
1315
1316         /* FIXME: we should change the policy function slightly, to not make
1317          * this list at all, since we just turn around and free it */
1318         while (!list_empty(&rpc_list)) {
1319                 struct ldlm_lock *wlock =
1320                         list_entry(rpc_list.next, struct ldlm_lock, l_cp_ast);
1321                 LASSERT((lock->l_flags & LDLM_FL_AST_SENT) == 0);
1322                 LASSERT(lock->l_flags & LDLM_FL_CP_REQD);
1323                 lock->l_flags &= ~LDLM_FL_CP_REQD;
1324                 list_del_init(&wlock->l_cp_ast);
1325                 LDLM_LOCK_PUT(wlock);
1326         }
1327
1328         if (rc == LDLM_ITER_CONTINUE) {
1329                 /* The lock met with no resistance; we're finished. */
1330                 unlock_res_and_lock(lock);
1331                 RETURN(ELDLM_LOCK_REPLACED);
1332         }
1333
1334         /* Do not grant any lock, but instead send GL callbacks.  The extent
1335          * policy nicely created a list of all PW locks for us.  We will choose
1336          * the highest of those which are larger than the size in the LVB, if
1337          * any, and perform a glimpse callback. */
1338         res_lvb = res->lr_lvb_data;
1339         LASSERT(res_lvb != NULL);
1340         *reply_lvb = *res_lvb;
1341
1342         list_for_each(tmp, &res->lr_granted) {
1343                 struct ldlm_lock *tmplock =
1344                         list_entry(tmp, struct ldlm_lock, l_res_link);
1345
1346                 if (tmplock->l_granted_mode == LCK_PR)
1347                         continue;
1348
1349                 if (tmplock->l_policy_data.l_extent.end <= reply_lvb->lvb_size)
1350                         continue;
1351
1352                 if (l == NULL) {
1353                         l = LDLM_LOCK_GET(tmplock);
1354                         continue;
1355                 }
1356
1357                 if (l->l_policy_data.l_extent.start >
1358                     tmplock->l_policy_data.l_extent.start)
1359                         continue;
1360
1361                 LDLM_LOCK_PUT(l);
1362                 l = LDLM_LOCK_GET(tmplock);
1363         }
1364         unlock_res(res);
1365
1366         /* There were no PW locks beyond the size in the LVB; finished. */
1367         if (l == NULL)
1368                 RETURN(ELDLM_LOCK_ABORTED);
1369
1370         if (l->l_glimpse_ast == NULL) {
1371                 /* We are racing with unlink(); just return -ENOENT */
1372                 CDEBUG(D_ERROR, "race with unlink: "LPX64"/"LPX64"\n",
1373                        res->lr_name.name[0], res->lr_name.name[1]);
1374                 rep->lock_policy_res1 = -ENOENT;
1375                 goto out;
1376         }
1377
1378         LASSERTF(l->l_glimpse_ast != NULL, "l == %p", l);
1379
1380         rc = l->l_glimpse_ast(l, NULL); /* this will update the LVB */
1381         if (rc != 0 && res->lr_namespace->ns_lvbo &&
1382             res->lr_namespace->ns_lvbo->lvbo_update) {
1383                 res->lr_namespace->ns_lvbo->lvbo_update(res, NULL, 0, 1);
1384         }
1385
1386         lock_res(res);
1387         *reply_lvb = *res_lvb;
1388         unlock_res(res);
1389 out:
1390         LDLM_LOCK_PUT(l);
1391
1392         RETURN(ELDLM_LOCK_ABORTED);
1393 }
1394
1395 static int filter_post_fs_cleanup(struct obd_device *obd)
1396 {
1397         int rc = 0;
1398
1399         rc = fsfilt_post_cleanup(obd);
1400
1401         RETURN(rc);
1402 }
1403
1404 #if 0
1405 static int filter_group_set_fs_flags(struct obd_device *obd, int group)
1406 {
1407         struct filter_obd *filter = &obd->u.filter;
1408         int rc = 0, i = 0;
1409         ENTRY;
1410
1411         /* zero group is not longer valid. */
1412         if (group== 0)
1413                 RETURN(rc);
1414         for (i = 0; i < filter->fo_subdir_count; i++) {
1415                 struct dentry *dentry;
1416                 dentry = (filter->fo_subdirs + group)->dentry[i];
1417                 rc = fsfilt_set_fs_flags(obd, dentry->d_inode,
1418                                          SM_DO_REC | SM_DO_COW);
1419                 if (rc)
1420                         RETURN(rc);
1421         }
1422         RETURN(rc);
1423 }
1424 #endif
1425
1426 static int filter_post_fs_setup(struct obd_device *obd)
1427 {
1428         struct filter_obd *filter = &obd->u.filter;
1429         int rc = 0;
1430
1431         rc = fsfilt_post_setup(obd, filter->fo_dentry_O);
1432
1433         return rc;
1434 }
1435
1436 /* mount the file system (secretly) */
1437 int filter_common_setup(struct obd_device *obd, obd_count len, void *buf,
1438                         char *option)
1439 {
1440         struct lustre_cfg *lcfg = buf;
1441         struct filter_obd *filter = &obd->u.filter;
1442         struct lvfs_obd_ctxt *lvfs_ctxt = NULL;
1443         struct vfsmount *mnt;
1444         char *str;
1445         char ns_name[48];
1446         int rc = 0, i;
1447         ENTRY;
1448
1449         if ((LUSTRE_CFG_BUFLEN(lcfg, 1)) < 1 ||
1450             (LUSTRE_CFG_BUFLEN(lcfg, 2) < 1))
1451                 RETURN(-EINVAL);
1452
1453         obd->obd_fsops = fsfilt_get_ops(lustre_cfg_string(lcfg, 2));
1454         if (IS_ERR(obd->obd_fsops))
1455                 RETURN(PTR_ERR(obd->obd_fsops));
1456
1457         rc = lvfs_mount_fs(lustre_cfg_string(lcfg, 1),
1458                            lustre_cfg_string(lcfg, 2),
1459                            option, MS_NOATIME | MS_NODIRATIME, &lvfs_ctxt);
1460         if (rc) {
1461                 CERROR("lvfs_mount_fs failed: rc = %d\n", rc);
1462                 GOTO(err_ops, rc);
1463         }
1464         LASSERT(lvfs_ctxt);
1465
1466         mnt = lvfs_ctxt->loc_mnt;
1467         filter->fo_lvfs_ctxt = lvfs_ctxt;
1468
1469         if (LUSTRE_CFG_BUFLEN(lcfg, 3) > 0 && lustre_cfg_buf(lcfg, 3)) {
1470                 str = lustre_cfg_string(lcfg, 3);
1471                 if (*str == 'f') {
1472                         obd->obd_replayable = 1;
1473                         obd_sync_filter = 1;
1474                         CWARN("%s: recovery enabled\n", obd->obd_name);
1475                 } else {
1476                         if (*str != 'n') {
1477                                 CERROR("unrecognised flag '%c'\n",
1478                                        *str);
1479                         }
1480                         // XXX Robert? Why do we get errors here
1481                         // GOTO(err_mntput, rc = -EINVAL);
1482                 }
1483         }
1484
1485         filter->fo_vfsmnt = mnt;
1486         filter->fo_sb = mnt->mnt_sb;
1487         filter->fo_fstype = mnt->mnt_sb->s_type->name;
1488         CDEBUG(D_SUPER, "%s: mnt = %p\n", filter->fo_fstype, mnt);
1489
1490         OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1491         obd->obd_lvfs_ctxt.pwdmnt = mnt;
1492         obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
1493         obd->obd_lvfs_ctxt.fs = get_ds();
1494         obd->obd_lvfs_ctxt.cb_ops = filter_lvfs_ops;
1495
1496         ll_clear_rdonly(ll_sbdev(filter->fo_sb));
1497
1498         rc = fsfilt_setup(obd, mnt->mnt_sb);
1499         if (rc)
1500                 GOTO(err_mntput, rc);
1501
1502         sema_init(&filter->fo_init_lock, 1);
1503         filter->fo_committed_group = 0;
1504         rc = filter_prep(obd);
1505         if (rc)
1506                 GOTO(err_mntput, rc);
1507
1508         filter->fo_destroys_in_progress = 0;
1509         for (i = 0; i < 32; i++)
1510                 sema_init(&filter->fo_create_locks[i], 1);
1511
1512         spin_lock_init(&filter->fo_translock);
1513         spin_lock_init(&filter->fo_objidlock);
1514         spin_lock_init(&filter->fo_lastidlock);
1515         INIT_LIST_HEAD(&filter->fo_export_list);
1516         sema_init(&filter->fo_alloc_lock, 1);
1517         spin_lock_init(&filter->fo_r_pages.oh_lock);
1518         spin_lock_init(&filter->fo_w_pages.oh_lock);
1519         spin_lock_init(&filter->fo_r_discont_pages.oh_lock);
1520         spin_lock_init(&filter->fo_w_discont_pages.oh_lock);
1521         spin_lock_init(&filter->fo_r_discont_blocks.oh_lock);
1522         spin_lock_init(&filter->fo_w_discont_blocks.oh_lock);
1523         filter->fo_readcache_max_filesize = FILTER_MAX_CACHE_SIZE;
1524
1525         INIT_LIST_HEAD(&filter->fo_llog_list);
1526         spin_lock_init(&filter->fo_llog_list_lock);
1527
1528         spin_lock_init(&filter->fo_denylist_lock);
1529         INIT_LIST_HEAD(&filter->fo_denylist);
1530
1531         sprintf(ns_name, "filter-%s", obd->obd_uuid.uuid);
1532         obd->obd_namespace = ldlm_namespace_new(ns_name, LDLM_NAMESPACE_SERVER);
1533
1534         if (obd->obd_namespace == NULL)
1535                 GOTO(err_post, rc = -ENOMEM);
1536         obd->obd_namespace->ns_lvbp = obd;
1537         obd->obd_namespace->ns_lvbo = &filter_lvbo;
1538         ldlm_register_intent(obd->obd_namespace, filter_intent_policy);
1539
1540         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1541                            "filter_ldlm_cb_client", &obd->obd_ldlm_client);
1542
1543         rc = obd_llog_cat_initialize(obd, &obd->obd_llogs, 1, CATLIST);
1544         if (rc) {
1545                 CERROR("failed to setup llogging subsystems\n");
1546                 GOTO(err_post, rc);
1547         }
1548         RETURN(0);
1549
1550 err_post:
1551         filter_post(obd);
1552 err_mntput:
1553         unlock_kernel();
1554         lvfs_umount_fs(filter->fo_lvfs_ctxt);
1555         filter->fo_sb = 0;
1556         lock_kernel();
1557 err_ops:
1558         fsfilt_put_ops(obd->obd_fsops);
1559         return rc;
1560 }
1561
1562 static int filter_attach(struct obd_device *obd, obd_count len, void *data)
1563 {
1564         struct lprocfs_static_vars lvars;
1565         int rc;
1566
1567         lprocfs_init_vars(filter, &lvars);
1568         rc = lprocfs_obd_attach(obd, lvars.obd_vars);
1569         if (rc != 0)
1570                 return rc;
1571
1572         rc = lprocfs_alloc_obd_stats(obd, LPROC_FILTER_LAST);
1573         if (rc != 0)
1574                 return rc;
1575
1576         /* Init obdfilter private stats here */
1577         lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_READ_BYTES,
1578                              LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes");
1579         lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_WRITE_BYTES,
1580                              LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes");
1581
1582         return lproc_filter_attach_seqstat(obd);
1583 }
1584
1585 static int filter_detach(struct obd_device *dev)
1586 {
1587         lprocfs_free_obd_stats(dev);
1588         return lprocfs_obd_detach(dev);
1589 }
1590
1591 static int filter_setup(struct obd_device *obd, obd_count len, void *buf)
1592 {
1593         struct lustre_cfg *lcfg = buf;
1594         unsigned long page;
1595         int rc;
1596         ENTRY;
1597
1598         /* 2.6.9 selinux wants a full option page for do_kern_mount (bug6471) */
1599         page = get_zeroed_page(GFP_KERNEL);
1600         if (!page)
1601                 RETURN(-ENOMEM);
1602
1603         memcpy((void *)page, lustre_cfg_buf(lcfg, 4),
1604                LUSTRE_CFG_BUFLEN(lcfg, 4));
1605
1606         /* all mount options including errors=remount-ro and asyncdel are passed
1607          * using 4th lcfg param. And it is good, finally we have got rid of
1608          * hardcoded fs types in the code. */
1609         rc = filter_common_setup(obd, len, buf, (void *)page);
1610         free_page(page);
1611
1612         if (rc)
1613                 RETURN(rc);
1614         rc = filter_post_fs_setup(obd);
1615         RETURN(rc);
1616 }
1617
1618 static int filter_cleanup(struct obd_device *obd, int flags)
1619 {
1620         struct filter_obd *filter = &obd->u.filter;
1621         ll_sbdev_type save_dev;
1622         ENTRY;
1623
1624         if (flags & OBD_OPT_FAILOVER)
1625                 CERROR("%s: shutting down for failover; client state will"
1626                        " be preserved.\n", obd->obd_name);
1627
1628         if (!list_empty(&obd->obd_exports)) {
1629                 CERROR("%s: still has clients!\n", obd->obd_name);
1630                 class_disconnect_exports(obd, flags);
1631                 if (!list_empty(&obd->obd_exports)) {
1632                         CERROR("still has exports after forced cleanup?\n");
1633                         RETURN(-EBUSY);
1634                 }
1635         }
1636
1637         target_cleanup_recovery(obd);
1638
1639         ldlm_namespace_free(obd->obd_namespace, flags & OBD_OPT_FORCE);
1640
1641         if (filter->fo_sb == NULL)
1642                 RETURN(0);
1643
1644         save_dev = ll_sbdev(filter->fo_sb);
1645         filter_post_fs_cleanup(obd);
1646         filter_post(obd);
1647
1648         shrink_dcache_parent(filter->fo_sb->s_root);
1649         filter->fo_sb = 0;
1650
1651         spin_lock(&filter->fo_denylist_lock);
1652         while (!list_empty(&filter->fo_denylist)) {
1653                 deny_sec_t *p_deny_sec = list_entry(filter->fo_denylist.next,
1654                                                     deny_sec_t, list);
1655                 list_del(&p_deny_sec->list);
1656                 OBD_FREE(p_deny_sec, sizeof(*p_deny_sec));
1657         }
1658         spin_unlock(&filter->fo_denylist_lock);
1659
1660         unlock_kernel();
1661         lvfs_umount_fs(filter->fo_lvfs_ctxt);
1662         //destroy_buffers(filter->fo_sb->s_dev);
1663         filter->fo_sb = NULL;
1664         fsfilt_put_ops(obd->obd_fsops);
1665         lock_kernel();
1666
1667         ll_clear_rdonly(save_dev);
1668
1669         RETURN(0);
1670 }
1671
1672 static int filter_process_config(struct obd_device *obd, obd_count len, void *buf)
1673 {
1674         struct lustre_cfg *lcfg = buf;
1675         struct filter_obd *filter = &obd->u.filter;
1676         int rc = 0;
1677         ENTRY;
1678
1679         switch(lcfg->lcfg_command) {
1680         case LCFG_SET_SECURITY: {
1681                 if ((LUSTRE_CFG_BUFLEN(lcfg, 1) == 0) ||
1682                     (LUSTRE_CFG_BUFLEN(lcfg, 2) == 0))
1683                         GOTO(out, rc = -EINVAL);
1684
1685                 if (!strcmp(lustre_cfg_string(lcfg, 1), "deny_sec")){
1686                         spin_lock(&filter->fo_denylist_lock);
1687                         rc = add_deny_security(lustre_cfg_string(lcfg, 2),
1688                                                &filter->fo_denylist);
1689                         spin_unlock(&filter->fo_denylist_lock);
1690                 }else {
1691                         CERROR("Unrecognized key\n");
1692                         rc = -EINVAL;
1693                 }
1694                 break;
1695         }
1696         default: {
1697                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1698                 GOTO(out, rc = -EINVAL);
1699
1700         }
1701         }
1702 out:
1703         RETURN(rc);
1704 }
1705
1706 static int filter_connect_post(struct obd_export *exp, unsigned initial,
1707                                unsigned long connect_flags)
1708 {
1709         struct obd_device *obd = exp->exp_obd;
1710         struct filter_export_data *fed;
1711         char str[PTL_NALFMT_SIZE];
1712         struct obd_llogs *llog;
1713         struct llog_ctxt *ctxt;
1714         int rc = 0;
1715         ENTRY;
1716
1717         fed = &exp->exp_filter_data;
1718         if (fed->fed_group < FILTER_MIN_GROUPS)
1719                 RETURN(0);
1720
1721         /* initialize llogs for connections from MDS */
1722         llog = filter_grab_llog_for_group(obd, fed->fed_group, exp);
1723         LASSERT(llog != NULL);
1724
1725         ctxt = llog_get_context(llog, LLOG_UNLINK_REPL_CTXT);
1726         LASSERT(ctxt != NULL);
1727
1728         rc = llog_receptor_accept(ctxt, exp->exp_imp_reverse);
1729
1730         portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
1731                         exp->exp_connection->c_peer.peer_id.nid, str);
1732
1733         CDEBUG(D_OTHER, "%s: init llog ctxt for export "LPX64"/%s, group %d\n",
1734                obd->obd_name, exp->exp_connection->c_peer.peer_id.nid,
1735                str, fed->fed_group);
1736
1737         RETURN(rc);
1738 }
1739
1740 /* nearly identical to mds_connect */
1741 static int filter_connect(struct lustre_handle *conn, struct obd_device *obd,
1742                           struct obd_uuid *cluuid,
1743                           struct obd_connect_data *data,
1744                           unsigned long connect_flags)
1745 {
1746         struct obd_export *exp;
1747         struct filter_export_data *fed;
1748         struct filter_client_data *fcd = NULL;
1749         struct filter_obd *filter = &obd->u.filter;
1750         struct lvfs_run_ctxt saved;
1751         int rc, group;
1752         ENTRY;
1753
1754         if (conn == NULL || obd == NULL || cluuid == NULL)
1755                 RETURN(-EINVAL);
1756
1757         rc = class_connect(conn, obd, cluuid);
1758         if (rc)
1759                 RETURN(rc);
1760         exp = class_conn2export(conn);
1761         LASSERT(exp != NULL);
1762
1763         fed = &exp->exp_filter_data;
1764
1765         spin_lock_init(&fed->fed_lock);
1766
1767        /* connection from MDS */
1768         group = connect_flags;
1769         if (obd->obd_replayable) {
1770                 OBD_ALLOC(fcd, sizeof(*fcd));
1771                 if (!fcd) {
1772                         CERROR("filter: out of memory for client data\n");
1773                         GOTO(cleanup, rc = -ENOMEM);
1774                 }
1775
1776                 memcpy(fcd->fcd_uuid, cluuid, sizeof(fcd->fcd_uuid));
1777                 fed->fed_fcd = fcd;
1778                 fed->fed_fcd->fcd_group = group;
1779                 rc = filter_client_add(obd, filter, fed, -1);
1780                 if (rc)
1781                         GOTO(cleanup, rc);
1782         }
1783         CWARN("%s: Received MDS connection ("LPX64"); group %d\n",
1784               obd->obd_name, exp->exp_handle.h_cookie, group);
1785
1786         if (group == 0)
1787                 GOTO(cleanup, rc);
1788
1789         if (fed->fed_group != 0 && fed->fed_group != group) {
1790                 char str[PTL_NALFMT_SIZE];
1791                 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
1792                                 exp->exp_connection->c_peer.peer_id.nid, str);
1793                 CERROR("!!! This export (nid "LPX64"/%s) used object group %d "
1794                        "earlier; now it's trying to use group %d!  This could "
1795                        "be a bug in the MDS.  Tell CFS.\n",
1796                        exp->exp_connection->c_peer.peer_id.nid, str,
1797                        fed->fed_group, group);
1798                 GOTO(cleanup, rc = -EPROTO);
1799         }
1800         fed->fed_group = group;
1801
1802         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1803         rc = filter_read_groups(obd, group, 1);
1804         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1805         if (rc != 0) {
1806                 CERROR("can't read group %u\n", group);
1807                 GOTO(cleanup, rc);
1808         }
1809 #if 0
1810         rc = filter_group_set_fs_flags(obd, group);
1811         if (rc != 0) {
1812                 CERROR("can't set kml flags %u\n", group);
1813                 GOTO(cleanup, rc);
1814         }
1815 #endif
1816 cleanup:
1817         if (rc) {
1818                 if (fcd)
1819                         OBD_FREE(fcd, sizeof(*fcd));
1820                 class_disconnect(exp, 0);
1821         } else {
1822                 class_export_put(exp);
1823         }
1824         return rc;
1825 }
1826
1827 static int filter_precleanup(struct obd_device *obd, int flags)
1828 {
1829         struct filter_group_llog *log;
1830         struct filter_obd *filter;
1831         int rc = 0;
1832         ENTRY;
1833
1834         filter = &obd->u.filter;
1835
1836         spin_lock(&filter->fo_llog_list_lock);
1837         while (!list_empty(&filter->fo_llog_list)) {
1838                 log = list_entry(filter->fo_llog_list.next,
1839                                  struct filter_group_llog, list);
1840                 list_del(&log->list);
1841                 spin_unlock(&filter->fo_llog_list_lock);
1842
1843                 rc = obd_llog_finish(obd, log->llogs, 0);
1844                 if (rc)
1845                         CERROR("failed to cleanup llogging subsystem for %u\n",
1846                                 log->group);
1847                 OBD_FREE(log->llogs, sizeof(*(log->llogs)));
1848                 OBD_FREE(log, sizeof(*log));
1849                 spin_lock(&filter->fo_llog_list_lock);
1850         }
1851         spin_unlock(&filter->fo_llog_list_lock);
1852
1853         rc = obd_llog_finish(obd, &obd->obd_llogs, 0);
1854         if (rc)
1855                 CERROR("failed to cleanup llogging subsystem\n");
1856
1857         RETURN(rc);
1858 }
1859
1860 /* Do extra sanity checks for grant accounting.  We do this at connect,
1861  * disconnect, and statfs RPC time, so it shouldn't be too bad.  We can
1862  * always get rid of it or turn it off when we know accounting is good. */
1863 static void filter_grant_sanity_check(struct obd_device *obd, const char *func)
1864 {
1865         struct filter_export_data *fed;
1866         struct obd_export *exp;
1867         obd_size maxsize = obd->obd_osfs.os_blocks * obd->obd_osfs.os_bsize;
1868         obd_size tot_dirty = 0, tot_pending = 0, tot_granted = 0;
1869         obd_size fo_tot_dirty, fo_tot_pending, fo_tot_granted;
1870         int level = D_CACHE;
1871
1872         if (list_empty(&obd->obd_exports))
1873                 return;
1874
1875         spin_lock(&obd->obd_osfs_lock);
1876         spin_lock(&obd->obd_dev_lock);
1877         list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
1878                 fed = &exp->exp_filter_data;
1879                 if (fed->fed_grant < 0 || fed->fed_pending < 0 ||
1880                     fed->fed_dirty < 0)
1881                         level = D_ERROR;
1882                 if (maxsize > 0) { /* we may not have done a statfs yet */
1883                         LASSERTF(fed->fed_grant + fed->fed_pending <= maxsize,
1884                                  "cli %s/%p %ld+%ld > "LPU64"\n",
1885                                  exp->exp_client_uuid.uuid, exp,
1886                                  fed->fed_grant, fed->fed_pending, maxsize);
1887                         LASSERTF(fed->fed_dirty <= maxsize,
1888                                  "cli %s/%p %ld > "LPU64"\n",
1889                                  exp->exp_client_uuid.uuid, exp,
1890                                  fed->fed_dirty, maxsize);
1891                 }
1892                 CDEBUG(level, "%s: cli %s/%p dirty %ld pend %ld grant %ld\n",
1893                        obd->obd_name, exp->exp_client_uuid.uuid, exp,
1894                        fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1895                 tot_granted += fed->fed_grant + fed->fed_pending;
1896                 tot_pending += fed->fed_pending;
1897                 tot_dirty += fed->fed_dirty;
1898         }
1899         fo_tot_granted = obd->u.filter.fo_tot_granted;
1900         fo_tot_pending = obd->u.filter.fo_tot_pending;
1901         fo_tot_dirty = obd->u.filter.fo_tot_dirty;
1902         spin_unlock(&obd->obd_dev_lock);
1903         spin_unlock(&obd->obd_osfs_lock);
1904
1905         /* Do these assertions outside the spinlocks so we don't kill system */
1906         if (tot_granted != fo_tot_granted)
1907                 CERROR("%s: tot_granted "LPU64" != fo_tot_granted "LPU64"\n",
1908                        func, tot_granted, fo_tot_granted);
1909         if (tot_pending != fo_tot_pending)
1910                 CERROR("%s: tot_pending "LPU64" != fo_tot_pending "LPU64"\n",
1911                        func, tot_pending, fo_tot_pending);
1912         if (tot_dirty != fo_tot_dirty)
1913                 CERROR("%s: tot_dirty "LPU64" != fo_tot_dirty "LPU64"\n",
1914                        func, tot_dirty, fo_tot_dirty);
1915         if (tot_pending > tot_granted)
1916                 CERROR("%s: tot_pending "LPU64" > tot_granted "LPU64"\n",
1917                        func, tot_pending, tot_granted);
1918         if (tot_granted > maxsize)
1919                 CERROR("%s: tot_granted "LPU64" > maxsize "LPU64"\n",
1920                        func, tot_granted, maxsize);
1921         if (tot_dirty > maxsize)
1922                 CERROR("%s: tot_dirty "LPU64" > maxsize "LPU64"\n",
1923                        func, tot_dirty, maxsize);
1924 }
1925
1926 /* Remove this client from the grant accounting totals.  We also remove
1927  * the export from the obd device under the osfs and dev locks to ensure
1928  * that the filter_grant_sanity_check() calculations are always valid.
1929  * The client should do something similar when it invalidates its import. */
1930 static void filter_grant_discard(struct obd_export *exp)
1931 {
1932         struct obd_device *obd = exp->exp_obd;
1933         struct filter_obd *filter = &obd->u.filter;
1934         struct filter_export_data *fed = &exp->exp_filter_data;
1935         int level = D_CACHE;
1936
1937         spin_lock(&obd->obd_osfs_lock);
1938         spin_lock(&exp->exp_obd->obd_dev_lock);
1939         list_del_init(&exp->exp_obd_chain);
1940         spin_unlock(&exp->exp_obd->obd_dev_lock);
1941
1942         if (fed->fed_dirty < 0 || fed->fed_grant < 0 || fed->fed_pending < 0)
1943                 level = D_ERROR;
1944         CDEBUG(level, "%s: cli %s/%p dirty %ld pend %ld grant %ld\n",
1945                obd->obd_name, exp->exp_client_uuid.uuid, exp,
1946                fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1947
1948         LASSERTF(filter->fo_tot_granted >= fed->fed_grant,
1949                  "%s: tot_granted "LPU64" cli %s/%p fed_grant %ld\n",
1950                  obd->obd_name, filter->fo_tot_granted,
1951                  exp->exp_client_uuid.uuid, exp, fed->fed_grant);
1952         filter->fo_tot_granted -= fed->fed_grant;
1953         LASSERTF(filter->fo_tot_pending >= fed->fed_pending,
1954                  "%s: tot_pending "LPU64" cli %s/%p fed_pending %ld\n",
1955                  obd->obd_name, filter->fo_tot_pending,
1956                  exp->exp_client_uuid.uuid, exp, fed->fed_pending);
1957         LASSERTF(filter->fo_tot_dirty >= fed->fed_dirty,
1958                  "%s: tot_dirty "LPU64" cli %s/%p fed_dirty %ld\n",
1959                  obd->obd_name, filter->fo_tot_dirty,
1960                  exp->exp_client_uuid.uuid, exp, fed->fed_dirty);
1961         filter->fo_tot_dirty -= fed->fed_dirty;
1962         fed->fed_dirty = 0;
1963         fed->fed_grant = 0;
1964
1965         spin_unlock(&obd->obd_osfs_lock);
1966 }
1967
1968 static int filter_destroy_export(struct obd_export *exp)
1969 {
1970         ENTRY;
1971
1972         if (exp->exp_filter_data.fed_pending)
1973                 CERROR("%s: cli %s/%p has %lu pending on destroyed export\n",
1974                        exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
1975                        exp, exp->exp_filter_data.fed_pending);
1976
1977         target_destroy_export(exp);
1978
1979         if (exp->exp_obd->obd_replayable)
1980                 filter_client_free(exp, exp->exp_flags);
1981
1982         filter_grant_discard(exp);
1983         if (!(exp->exp_flags & OBD_OPT_FORCE))
1984                 filter_grant_sanity_check(exp->exp_obd, __FUNCTION__);
1985
1986         RETURN(0);
1987 }
1988
1989 static void filter_sync_llogs(struct obd_device *obd, struct obd_export *dexp)
1990 {
1991         struct filter_group_llog *fglog, *nlog;
1992         struct filter_obd *filter;
1993         int worked = 0, group;
1994         struct llog_ctxt *ctxt;
1995         ENTRY;
1996
1997         filter = &obd->u.filter;
1998
1999         /* we can't sync log holding spinlock. also, we do not want to get
2000          * into livelock. so we do following: loop over MDS's exports in
2001          * group order and skip already synced llogs -bzzz */
2002         do {
2003                 /* look for group with min. number, but > worked */
2004                 fglog = NULL;
2005                 group = 1 << 30;
2006                 spin_lock(&filter->fo_llog_list_lock);
2007                 list_for_each_entry(nlog, &filter->fo_llog_list, list) {
2008
2009                         if (nlog->group <= worked) {
2010                                 /* this group is already synced */
2011                                 continue;
2012                         }
2013
2014                         if (group < nlog->group) {
2015                                 /* we have group with smaller number to sync */
2016                                 continue;
2017                         }
2018
2019                         /* store current minimal group */
2020                         fglog = nlog;
2021                         group = nlog->group;
2022                 }
2023                 spin_unlock(&filter->fo_llog_list_lock);
2024
2025                 if (fglog == NULL)
2026                         break;
2027
2028                 worked = fglog->group;
2029                 if (fglog->exp && (dexp == fglog->exp || dexp == NULL)) {
2030                         ctxt = llog_get_context(fglog->llogs,
2031                                         LLOG_UNLINK_REPL_CTXT);
2032                         LASSERT(ctxt != NULL);
2033                         llog_sync(ctxt, fglog->exp);
2034                 }
2035         } while (fglog != NULL);
2036 }
2037
2038 /* also incredibly similar to mds_disconnect */
2039 static int filter_disconnect(struct obd_export *exp, unsigned long flags)
2040 {
2041         struct obd_device *obd = exp->exp_obd;
2042         unsigned long irqflags;
2043         int rc;
2044         ENTRY;
2045
2046         LASSERT(exp);
2047         class_export_get(exp);
2048
2049         spin_lock_irqsave(&exp->exp_lock, irqflags);
2050         exp->exp_flags = flags;
2051         spin_unlock_irqrestore(&exp->exp_lock, irqflags);
2052
2053         if (!(flags & OBD_OPT_FORCE))
2054                 filter_grant_sanity_check(obd, __FUNCTION__);
2055         filter_grant_discard(exp);
2056
2057         /* Disconnect early so that clients can't keep using export */
2058         rc = class_disconnect(exp, flags);
2059
2060         ldlm_cancel_locks_for_export(exp);
2061
2062         fsfilt_sync(obd, obd->u.filter.fo_sb);
2063
2064         /* flush any remaining cancel messages out to the target */
2065         filter_sync_llogs(obd, exp);
2066         class_export_put(exp);
2067         RETURN(rc);
2068 }
2069
2070 struct dentry *__filter_oa2dentry(struct obd_device *obd,
2071                                   struct obdo *oa, const char *what)
2072 {
2073         struct dentry *dchild = NULL;
2074         obd_gr group = 0;
2075         ENTRY;
2076
2077         if (oa->o_valid & OBD_MD_FLGROUP)
2078                 group = oa->o_gr;
2079
2080         dchild = filter_id2dentry(obd, NULL, group, oa->o_id);
2081
2082         if (IS_ERR(dchild)) {
2083                 CERROR("%s error looking up object: "LPU64"\n",
2084                        what, oa->o_id);
2085                 RETURN(dchild);
2086         }
2087
2088         if (dchild->d_inode == NULL) {
2089                 CDEBUG(D_INFO, "%s: %s on non-existent object: "
2090                        LPU64"\n", obd->obd_name, what, oa->o_id);
2091                 f_dput(dchild);
2092                 RETURN(ERR_PTR(-ENOENT));
2093         }
2094
2095         RETURN(dchild);
2096 }
2097
2098 static int filter_getattr(struct obd_export *exp, struct obdo *oa,
2099                           struct lov_stripe_md *md)
2100 {
2101         struct dentry *dentry = NULL;
2102         struct obd_device *obd;
2103         int rc = 0;
2104         ENTRY;
2105
2106         obd = class_exp2obd(exp);
2107         if (obd == NULL) {
2108                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2109                        exp->exp_handle.h_cookie);
2110                 RETURN(-EINVAL);
2111         }
2112
2113         dentry = filter_oa2dentry(obd, oa);
2114         if (IS_ERR(dentry))
2115                 RETURN(PTR_ERR(dentry));
2116
2117         /* Limit the valid bits in the return data to what we actually use */
2118         oa->o_valid = OBD_MD_FLID;
2119         obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2120
2121         f_dput(dentry);
2122         RETURN(rc);
2123 }
2124
2125 int filter_setattr_internal(struct obd_export *exp, struct dentry *dentry,
2126                             struct obdo *oa, struct obd_trans_info *oti)
2127 {
2128         struct filter_obd *filter;
2129         struct iattr iattr;
2130         void *handle;
2131         int rc, err;
2132         ENTRY;
2133
2134         LASSERT(dentry != NULL);
2135         LASSERT(!IS_ERR(dentry));
2136         LASSERT(dentry->d_inode != NULL);
2137
2138         filter = &exp->exp_obd->u.filter;
2139         iattr_from_obdo(&iattr, oa, oa->o_valid);
2140
2141         if (iattr.ia_valid & ATTR_SIZE)
2142                 down(&dentry->d_inode->i_sem);
2143         handle = fsfilt_start(exp->exp_obd, dentry->d_inode,
2144                               FSFILT_OP_SETATTR, oti);
2145         if (IS_ERR(handle))
2146                 GOTO(out_unlock, rc = PTR_ERR(handle));
2147
2148         /* XXX this could be a rwsem instead, if filter_preprw played along */
2149         if (iattr.ia_valid & ATTR_ATTR_FLAG)
2150                 rc = fsfilt_iocontrol(exp->exp_obd, dentry->d_inode,
2151                                       NULL, EXT3_IOC_SETFLAGS,
2152                                       (long)&iattr.ia_attr_flags);
2153         else
2154                 rc = fsfilt_setattr(exp->exp_obd, dentry, handle,
2155                                     &iattr, 1);
2156
2157         rc = filter_finish_transno(exp, oti, rc);
2158
2159         err = fsfilt_commit(exp->exp_obd, filter->fo_sb,
2160                             dentry->d_inode, handle,
2161                             exp->exp_sync);
2162         if (err) {
2163                 CERROR("error on commit, err = %d\n", err);
2164                 if (!rc)
2165                         rc = err;
2166         }
2167         EXIT;
2168 out_unlock:
2169         if (iattr.ia_valid & ATTR_SIZE)
2170                 up(&dentry->d_inode->i_sem);
2171         return rc;
2172 }
2173
2174 /* this is called from filter_truncate() until we have filter_punch() */
2175 int filter_setattr(struct obd_export *exp, struct obdo *oa,
2176                    struct lov_stripe_md *md, struct obd_trans_info *oti)
2177 {
2178         struct ldlm_res_id res_id = { .name = { oa->o_id, 0, oa->o_gr, 0 } };
2179         struct ldlm_valblock_ops *ns_lvbo;
2180         struct lvfs_run_ctxt saved;
2181         struct filter_obd *filter;
2182         struct ldlm_resource *res;
2183         struct dentry *dentry;
2184         /*obd_uid uid;*/
2185         /*obd_gid gid;*/
2186         int rc;
2187         ENTRY;
2188
2189         LASSERT(oti != NULL);
2190
2191         filter = &exp->exp_obd->u.filter;
2192         push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2193
2194         /* make sure that object is allocated. */
2195         dentry = filter_crow_object(exp->exp_obd, oa);
2196         if (IS_ERR(dentry))
2197                 GOTO(out_pop, rc = PTR_ERR(dentry));
2198
2199         lock_kernel();
2200
2201         /* setting objects attributes (including owner/group) */
2202         rc = filter_setattr_internal(exp, dentry, oa, oti);
2203         if (rc)
2204                 GOTO(out_unlock, rc);
2205
2206         res = ldlm_resource_get(exp->exp_obd->obd_namespace, NULL,
2207                                 res_id, LDLM_EXTENT, 0);
2208
2209         if (res != NULL) {
2210                 ns_lvbo = res->lr_namespace->ns_lvbo;
2211                 if (ns_lvbo && ns_lvbo->lvbo_update)
2212                         rc = ns_lvbo->lvbo_update(res, NULL, 0, 0);
2213                 ldlm_resource_putref(res);
2214         }
2215
2216         oa->o_valid = OBD_MD_FLID;
2217         obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2218
2219         EXIT;
2220 out_unlock:
2221         unlock_kernel();
2222         f_dput(dentry);
2223 out_pop:
2224         pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2225         return rc;
2226 }
2227
2228 /* XXX identical to osc_unpackmd */
2229 static int filter_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2230                            struct lov_mds_md *lmm, int lmm_bytes)
2231 {
2232         int lsm_size;
2233         ENTRY;
2234
2235         if (lmm != NULL) {
2236                 if (lmm_bytes < sizeof (*lmm)) {
2237                         CERROR("lov_mds_md too small: %d, need %d\n",
2238                                lmm_bytes, (int)sizeof(*lmm));
2239                         RETURN(-EINVAL);
2240                 }
2241                 /* XXX LOV_MAGIC etc check? */
2242
2243                 if (lmm->lmm_object_id == cpu_to_le64(0)) {
2244                         CERROR("lov_mds_md: zero lmm_object_id\n");
2245                         RETURN(-EINVAL);
2246                 }
2247         }
2248
2249         lsm_size = lov_stripe_md_size(1);
2250         if (lsmp == NULL)
2251                 RETURN(lsm_size);
2252
2253         if (*lsmp != NULL && lmm == NULL) {
2254                 OBD_FREE(*lsmp, lsm_size);
2255                 *lsmp = NULL;
2256                 RETURN(0);
2257         }
2258
2259         if (*lsmp == NULL) {
2260                 OBD_ALLOC(*lsmp, lsm_size);
2261                 if (*lsmp == NULL)
2262                         RETURN(-ENOMEM);
2263
2264                 loi_init((*lsmp)->lsm_oinfo);
2265         }
2266
2267         if (lmm != NULL) {
2268                 /* XXX zero *lsmp? */
2269                 (*lsmp)->lsm_object_id = le64_to_cpu (lmm->lmm_object_id);
2270                 LASSERT((*lsmp)->lsm_object_id);
2271         }
2272
2273         (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
2274
2275         RETURN(lsm_size);
2276 }
2277
2278 static int filter_statfs(struct obd_device *obd, struct obd_statfs *osfs,
2279                          unsigned long max_age)
2280 {
2281         struct filter_obd *filter = &obd->u.filter;
2282         int blockbits = filter->fo_sb->s_blocksize_bits;
2283         int rc;
2284         ENTRY;
2285
2286         /* at least try to account for cached pages.  its still racey and
2287          * might be under-reporting if clients haven't announced their
2288          * caches with brw recently */
2289         spin_lock(&obd->obd_osfs_lock);
2290         rc = fsfilt_statfs(obd, filter->fo_sb, max_age);
2291         memcpy(osfs, &obd->obd_osfs, sizeof(*osfs));
2292         spin_unlock(&obd->obd_osfs_lock);
2293
2294         CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
2295                " pending "LPU64" free "LPU64" avail "LPU64"\n",
2296                filter->fo_tot_dirty, filter->fo_tot_granted,
2297                filter->fo_tot_pending,
2298                osfs->os_bfree << blockbits, osfs->os_bavail << blockbits);
2299
2300         filter_grant_sanity_check(obd, __FUNCTION__);
2301
2302         osfs->os_bavail -= min(osfs->os_bavail,
2303                                (filter->fo_tot_dirty + filter->fo_tot_pending +
2304                                 osfs->os_bsize -1) >> blockbits);
2305
2306         RETURN(rc);
2307 }
2308
2309 int filter_create_object(struct obd_device *obd, struct obdo *oa)
2310 {
2311         struct dentry *dparent = NULL;
2312         struct dentry *dchild = NULL;
2313         struct lvfs_ucred uc = {0,};
2314         struct lvfs_run_ctxt saved;
2315         struct filter_obd *filter;
2316         int cleanup_phase = 0;
2317         int err = 0, rc = 0;
2318         void *handle = NULL;
2319         void *lock = NULL;
2320         ENTRY;
2321
2322         filter = &obd->u.filter;
2323
2324         down(&filter->fo_create_locks[oa->o_gr]);
2325
2326         if (test_bit(oa->o_gr, &filter->fo_destroys_in_progress)) {
2327                 CWARN("%s: create aborted by destroy\n",
2328                       obd->obd_name);
2329                 GOTO(out, rc = -EALREADY);
2330         }
2331
2332         CDEBUG(D_INFO, "create objid "LPU64"\n", oa->o_id);
2333
2334         dparent = filter_parent_lock(obd, oa->o_gr, oa->o_id, &lock);
2335         if (IS_ERR(dparent))
2336                 GOTO(cleanup, rc = PTR_ERR(dparent));
2337         cleanup_phase = 1;
2338
2339         dchild = filter_id2dentry(obd, dparent, oa->o_gr, oa->o_id);
2340         if (IS_ERR(dchild))
2341                 GOTO(cleanup, rc = PTR_ERR(dchild));
2342         cleanup_phase = 2;
2343
2344         if (dchild->d_inode != NULL)
2345                 GOTO(cleanup, rc = 0);
2346
2347         handle = fsfilt_start_log(obd, dparent->d_inode,
2348                                   FSFILT_OP_CREATE, NULL, 1);
2349         if (IS_ERR(handle))
2350                 GOTO(cleanup, rc = PTR_ERR(handle));
2351         cleanup_phase = 3;
2352
2353         uc.luc_uid = oa->o_uid;
2354         uc.luc_gid = oa->o_gid;
2355         uc.luc_fsuid = oa->o_uid;
2356         uc.luc_fsgid = oa->o_gid;
2357         uc.luc_cap = current->cap_effective;
2358         cap_raise(uc.luc_cap, CAP_SYS_RESOURCE);
2359
2360         push_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
2361         rc = ll_vfs_create(dparent->d_inode, dchild, S_IFREG, NULL);
2362         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
2363
2364         if (rc) {
2365                 CERROR("create failed rc = %d\n", rc);
2366                 GOTO(cleanup, rc);
2367         }
2368
2369         /* nobody else is touching this newly created object */
2370         LASSERT(dchild->d_inode);
2371         if (oa->o_valid & OBD_MD_FLIFID) {
2372                 rc = fsfilt_set_md(obd, dchild->d_inode, handle, obdo_id(oa),
2373                                    sizeof(struct lustre_id), EA_SID);
2374                 if (rc) {
2375                         CERROR("store fid in object failed! rc:%d\n", rc);
2376                         GOTO(cleanup, rc);
2377                 }
2378         } else {
2379                 CWARN("create object without lustre id!\n");
2380         }
2381
2382
2383         fsfilt_set_fs_flags(obd, dparent->d_inode, SM_DO_REC);
2384
2385         /* save last created object id */
2386         filter_save_last_id(filter, oa->o_gr, oa->o_id);
2387
2388         rc = filter_update_last_objid(obd, oa->o_gr, 0);
2389         if (rc) {
2390                 CERROR("unable to write lastobjid, but "
2391                        "orphans were deleted, err = %d\n",
2392                        rc);
2393                 rc = 0;
2394         }
2395 cleanup:
2396         switch(cleanup_phase) {
2397         case 3:
2398                 err = fsfilt_commit(obd, filter->fo_sb,
2399                                     dparent->d_inode, handle, 0);
2400                 if (err) {
2401                         CERROR("error on commit, err = %d\n", err);
2402                         if (!rc)
2403                                 rc = err;
2404                 }
2405         case 2:
2406                 f_dput(dchild);
2407         case 1:
2408                 filter_parent_unlock(dparent, lock);
2409         case 0:
2410                 break;
2411         }
2412
2413         if (rc)
2414                 GOTO(out, rc);
2415
2416 out:
2417         up(&filter->fo_create_locks[oa->o_gr]);
2418         RETURN(rc);
2419 }
2420
2421 struct dentry *
2422 filter_crow_object(struct obd_device *obd, struct obdo *oa)
2423 {
2424         struct dentry *dentry;
2425         obd_uid uid;
2426         obd_gid gid;
2427         int rc = 0;
2428         ENTRY;
2429
2430         /* check if object is already allocated */
2431         dentry = filter_id2dentry(obd, NULL, oa->o_gr, oa->o_id);
2432         if (IS_ERR(dentry))
2433                 RETURN(dentry);
2434
2435         if (dentry->d_inode)
2436                 RETURN(dentry);
2437
2438         f_dput(dentry);
2439         CDEBUG(D_INODE, "OSS object "LPU64"/"LPU64
2440                " does not exists - allocate it now\n",
2441                oa->o_id, oa->o_gr);
2442
2443         uid = oa->o_valid & OBD_MD_FLUID ? oa->o_uid : 0;
2444         gid = oa->o_valid & OBD_MD_FLGID ? oa->o_gid : 0;
2445
2446         rc = filter_create_object(obd, oa);
2447         if (rc) {
2448                 CERROR("cannot create OSS object "LPU64"/"LPU64
2449                        ", err = %d\n", oa->o_id, oa->o_gr, rc);
2450                 RETURN(ERR_PTR(rc));
2451         }
2452
2453         /* lookup for just created object and return it to caller */
2454         dentry = filter_id2dentry(obd, NULL, oa->o_gr, oa->o_id);
2455         if (IS_ERR(dentry))
2456                 RETURN(dentry);
2457
2458         if (dentry->d_inode == NULL) {
2459                 f_dput(dentry);
2460                 dentry = ERR_PTR(-ENOENT);
2461                 CERROR("cannot find just created OSS object "
2462                        LPU64"/"LPU64" err = %d\n", oa->o_id,
2463                        oa->o_gr, (int)PTR_ERR(dentry));
2464                 RETURN(dentry);
2465         }
2466         RETURN(dentry);
2467 }
2468
2469 static int
2470 filter_clear_orphans(struct obd_export *exp, struct obdo *oa)
2471 {
2472         struct obd_device *obd = NULL;
2473         struct filter_obd *filter;
2474         struct obdo *doa = NULL;
2475         int rc = 0, orphans;
2476         __u64 last, id;
2477         ENTRY;
2478
2479         LASSERT(oa);
2480         LASSERT(oa->o_gr != 0);
2481         LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2482
2483         obd = exp->exp_obd;
2484         filter = &obd->u.filter;
2485
2486         last = filter_last_id(filter, oa->o_gr);
2487         orphans = last - oa->o_id;
2488
2489         if (orphans <= 0)
2490                 RETURN(0);
2491
2492         doa = obdo_alloc();
2493         if (doa == NULL)
2494                 RETURN(-ENOMEM);
2495
2496         doa->o_gr = oa->o_gr;
2497         doa->o_mode = S_IFREG;
2498         doa->o_valid = oa->o_valid & (OBD_MD_FLGROUP | OBD_MD_FLID);
2499
2500         set_bit(doa->o_gr, &filter->fo_destroys_in_progress);
2501         down(&filter->fo_create_locks[doa->o_gr]);
2502         if (!test_bit(doa->o_gr, &filter->fo_destroys_in_progress)) {
2503                 CERROR("%s:["LPU64"] destroy_in_progress already cleared\n",
2504                        exp->exp_obd->obd_name, doa->o_gr);
2505                 up(&filter->fo_create_locks[doa->o_gr]);
2506                 GOTO(out_free_doa, 0);
2507         }
2508
2509         CDEBUG(D_ERROR, "%s:["LPU64"] deleting orphan objects from "LPU64" to "
2510               LPU64"\n", exp->exp_obd->obd_name, doa->o_gr,
2511               oa->o_id + 1, last);
2512
2513         for (id = oa->o_id + 1; id <= last; id++) {
2514                 doa->o_id = id;
2515                 filter_destroy(exp, doa, NULL, NULL);
2516         }
2517
2518         CDEBUG(D_ERROR, "%s:["LPU64"] after destroy: set last_objids = "
2519                LPU64"\n", exp->exp_obd->obd_name, doa->o_gr, last + 1);
2520
2521         /* return next free id to be used as a new start of sequence -bzzz */
2522         oa->o_id = last + 1;
2523
2524         filter_set_last_id(filter, oa->o_gr, oa->o_id);
2525         clear_bit(doa->o_gr, &filter->fo_destroys_in_progress);
2526         up(&filter->fo_create_locks[oa->o_gr]);
2527
2528         EXIT;
2529 out_free_doa:
2530         obdo_free(doa);
2531         return rc;
2532 }
2533
2534 /*
2535  * by now this function is only needed as entry point for deleting orphanes on
2536  * OSS as objects are created on first write attempt. --umka
2537  */
2538 static int
2539 filter_create(struct obd_export *exp, struct obdo *oa, void *acl,
2540               int acl_size, struct lov_stripe_md **ea,
2541               struct obd_trans_info *oti)
2542 {
2543         struct filter_export_data *fed;
2544         struct obd_device *obd = NULL;
2545         int group = oa->o_gr, rc = 0;
2546         struct lvfs_run_ctxt saved;
2547         struct filter_obd *filter;
2548         char str[PTL_NALFMT_SIZE];
2549         ENTRY;
2550
2551         LASSERT(acl == NULL && acl_size == 0);
2552
2553         if (!(oa->o_valid & OBD_MD_FLGROUP) || group == 0) {
2554                 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
2555                                 exp->exp_connection->c_peer.peer_id.nid, str);
2556                 CERROR("!!! nid "LPX64"/%s sent invalid object group %d\n",
2557                        exp->exp_connection->c_peer.peer_id.nid, str, group);
2558                 RETURN(-EINVAL);
2559         }
2560
2561         obd = exp->exp_obd;
2562         fed = &exp->exp_filter_data;
2563         filter = &obd->u.filter;
2564
2565         if (fed->fed_group != group) {
2566                 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
2567                                 exp->exp_connection->c_peer.peer_id.nid, str);
2568                 CERROR("!!! this export (nid "LPX64"/%s) used object group %d "
2569                        "earlier; now it's trying to use group %d!  This could "
2570                        "be a bug in the MDS.  Tell CFS.\n",
2571                        exp->exp_connection->c_peer.peer_id.nid, str,
2572                        fed->fed_group, group);
2573                 RETURN(-ENOTUNIQ);
2574         }
2575
2576         CDEBUG(D_INFO, "filter_create(od->o_gr=%d,od->o_id="LPU64")\n",
2577                group, oa->o_id);
2578
2579         obd = exp->exp_obd;
2580         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2581
2582         LASSERT((oa->o_valid & OBD_MD_FLFLAGS) &&
2583                 (oa->o_flags == OBD_FL_DELORPHAN));
2584
2585         rc = filter_clear_orphans(exp, oa);
2586         if (rc) {
2587                 CERROR("cannot clear orphanes starting from "
2588                        LPU64", err = %d\n", oa->o_id, rc);
2589         } else {
2590                 rc = filter_update_last_objid(obd, group, 0);
2591                 if (rc) {
2592                         CERROR("unable to write lastobjid, but "
2593                                "orphans were deleted, err = %d\n",
2594                                rc);
2595                 }
2596         }
2597         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2598
2599         RETURN(0);
2600 }
2601
2602 static int filter_destroy(struct obd_export *exp, struct obdo *oa,
2603                           struct lov_stripe_md *ea, struct obd_trans_info *oti)
2604 {
2605         int rc, rc2, cleanup_phase = 0, have_prepared = 0;
2606         struct dentry *dchild = NULL, *dparent = NULL;
2607         struct llog_cookie *fcc = NULL;
2608         struct lvfs_run_ctxt saved;
2609         struct filter_obd *filter;
2610         struct obd_device *obd;
2611         void *handle = NULL;
2612         void *lock = NULL;
2613         struct iattr iattr;
2614         ENTRY;
2615
2616         LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2617
2618         obd = exp->exp_obd;
2619         filter = &obd->u.filter;
2620
2621         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2622
2623  acquire_locks:
2624         dparent = filter_parent_lock(obd, oa->o_gr, oa->o_id, &lock);
2625         if (IS_ERR(dparent))
2626                 GOTO(cleanup, rc = PTR_ERR(dparent));
2627         cleanup_phase = 1;
2628
2629         dchild = filter_id2dentry(obd, dparent, oa->o_gr, oa->o_id);
2630         if (IS_ERR(dchild))
2631                 GOTO(cleanup, rc = PTR_ERR(dchild));
2632         cleanup_phase = 2;
2633
2634         if (dchild->d_inode == NULL) {
2635                 CDEBUG(D_INODE, "destroying non-existent object "LPU64"\n",
2636                        oa->o_id);
2637                 if (have_prepared)
2638                         GOTO(cleanup, rc = -ENOENT);
2639         }
2640
2641         if (!have_prepared) {
2642                 /* If we're really going to destroy the object, get ready
2643                  * by getting the clients to discard their cached data.
2644                  *
2645                  * We have to drop the parent lock, because
2646                  * filter_prepare_destroy will acquire a PW on the object, and
2647                  * we don't want to deadlock with an incoming write to the
2648                  * object, which has the extent PW and then wants to get the
2649                  * parent dentry to do the lookup.
2650                  *
2651                  * We dput the child because it's not worth the extra
2652                  * complication of condition the above code to skip it on the
2653                  * second time through. */
2654                 f_dput(dchild);
2655                 filter_parent_unlock(dparent, lock);
2656
2657                 filter_prepare_destroy(obd, oa->o_id, oa->o_gr);
2658                 have_prepared = 1;
2659                 goto acquire_locks;
2660         }
2661
2662         /* Our MDC connection is established by the MDS to us */
2663         if (oa->o_valid & OBD_MD_FLCOOKIE) {
2664                 OBD_ALLOC(fcc, sizeof(*fcc));
2665                 if (fcc != NULL)
2666                         memcpy(fcc, obdo_logcookie(oa), sizeof(*fcc));
2667         }
2668
2669         /* we're gonna truncate it first in order to avoid possible
2670          * deadlock:
2671          *      P1                      P2
2672          * open trasaction      open transaction
2673          * down(i_zombie)       down(i_zombie)
2674          *                      restart transaction
2675          * (see BUG 4180) -bzzz
2676          */
2677         down(&dchild->d_inode->i_sem);
2678         handle = fsfilt_start_log(obd, dparent->d_inode,FSFILT_OP_SETATTR,NULL,1);
2679         if (IS_ERR(handle)) {
2680                 up(&dchild->d_inode->i_sem);
2681                 GOTO(cleanup, rc = PTR_ERR(handle));
2682         }
2683
2684         iattr.ia_valid = ATTR_SIZE;
2685         iattr.ia_size = 0;
2686         rc = fsfilt_setattr(obd, dchild, handle, &iattr, 1);
2687
2688         rc2 = fsfilt_commit(obd, filter->fo_sb, dparent->d_inode, handle, 0);
2689         up(&dchild->d_inode->i_sem);
2690         if (rc)
2691                 GOTO(cleanup, rc);
2692         if (rc2)
2693                 GOTO(cleanup, rc = rc2);
2694
2695         handle = fsfilt_start_log(obd, dparent->d_inode,FSFILT_OP_UNLINK,oti,1);
2696         if (IS_ERR(handle))
2697                 GOTO(cleanup, rc = PTR_ERR(handle));
2698
2699         cleanup_phase = 3;
2700
2701         rc = filter_destroy_internal(obd, oa->o_id, dparent, dchild);
2702
2703 cleanup:
2704         switch(cleanup_phase) {
2705         case 3:
2706                 if (fcc != NULL) {
2707                         fsfilt_add_journal_cb(obd, filter->fo_sb, 0,
2708                                               oti ? oti->oti_handle : handle,
2709                                               filter_cancel_cookies_cb, fcc);
2710                 }
2711                 rc = filter_finish_transno(exp, oti, rc);
2712                 rc2 = fsfilt_commit(obd, filter->fo_sb, dparent->d_inode,
2713                                     handle, exp->exp_sync);
2714                 if (rc2) {
2715                         CERROR("error on commit, err = %d\n", rc2);
2716                         if (!rc)
2717                                 rc = rc2;
2718                 }
2719         case 2:
2720                 f_dput(dchild);
2721         case 1:
2722                 filter_parent_unlock(dparent, lock);
2723         case 0:
2724                 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2725                 break;
2726         default:
2727                 CERROR("invalid cleanup_phase %d\n", cleanup_phase);
2728                 LBUG();
2729         }
2730
2731         RETURN(rc);
2732 }
2733
2734 /* NB start and end are used for punch, but not truncate */
2735 static int filter_truncate(struct obd_export *exp, struct obdo *oa,
2736                            struct lov_stripe_md *lsm,
2737                            obd_off start, obd_off end,
2738                            struct obd_trans_info *oti)
2739 {
2740         int error;
2741         ENTRY;
2742
2743         if (end != OBD_OBJECT_EOF)
2744                 CERROR("PUNCH not supported, only truncate: end = "LPX64"\n",
2745                        end);
2746
2747         CDEBUG(D_INODE, "calling truncate for object "LPU64", valid = "LPU64", "
2748                "o_size = "LPD64"\n", oa->o_id, oa->o_valid, start);
2749         oa->o_size = start;
2750         error = filter_setattr(exp, oa, NULL, oti);
2751         RETURN(error);
2752 }
2753
2754 static int filter_sync(struct obd_export *exp, struct obdo *oa,
2755                        struct lov_stripe_md *lsm, obd_off start, obd_off end)
2756 {
2757         struct obd_device *obd = exp->exp_obd;
2758         struct lvfs_run_ctxt saved;
2759         struct filter_obd *filter;
2760         struct dentry *dentry;
2761         int rc, rc2;
2762         ENTRY;
2763
2764         filter = &obd->u.filter;
2765
2766         /* an objid of zero is taken to mean "sync whole filesystem" */
2767         if (!oa || !(oa->o_valid & OBD_MD_FLID)) {
2768                 rc = fsfilt_sync(obd, filter->fo_sb);
2769                 /* flush any remaining cancel messages out to the target */
2770                 filter_sync_llogs(obd, NULL);
2771                 RETURN(rc);
2772         }
2773
2774         dentry = filter_oa2dentry(obd, oa);
2775         if (IS_ERR(dentry))
2776                 RETURN(PTR_ERR(dentry));
2777
2778         push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2779
2780         down(&dentry->d_inode->i_sem);
2781         rc = filemap_fdatawrite(dentry->d_inode->i_mapping);
2782         if (rc == 0) {
2783                 /* just any file to grab fsync method - "file" arg unused */
2784                 struct file *file = filter->fo_rcvd_filp;
2785
2786                 if (file->f_op && file->f_op->fsync)
2787                         rc = file->f_op->fsync(NULL, dentry, 1);
2788
2789                 rc2 = filemap_fdatawait(dentry->d_inode->i_mapping);
2790                 if (!rc)
2791                         rc = rc2;
2792         }
2793         up(&dentry->d_inode->i_sem);
2794
2795         oa->o_valid = OBD_MD_FLID;
2796         obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2797
2798         pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2799
2800         f_dput(dentry);
2801         RETURN(rc);
2802 }
2803
2804 static int filter_set_info(struct obd_export *exp, __u32 keylen,
2805                            void *key, __u32 vallen, void *val)
2806 {
2807         struct obd_device *obd;
2808         int rc = -EINVAL;
2809         ENTRY;
2810
2811         obd = class_exp2obd(exp);
2812         if (obd == NULL) {
2813                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2814                        exp->exp_handle.h_cookie);
2815                 RETURN(-EINVAL);
2816         }
2817
2818         if (keylen == 8 && memcmp(key, "auditlog", 8) == 0) {
2819
2820                 rc = fsfilt_set_info(obd, obd->u.filter.fo_sb, NULL,
2821                                      8, "auditlog", vallen, val);
2822                 RETURN(rc);
2823         } else if (keylen == 5 && strcmp(key, "audit") == 0) {
2824                 //set audit for whole FS on OSS
2825                 struct audit_attr_msg * msg = val;
2826
2827                 rc = fsfilt_set_info(obd, obd->u.filter.fo_sb, NULL,
2828                                      5, "audit", sizeof(msg->attr), &msg->attr);
2829                 RETURN(rc);
2830         } else if (keylen == 9 && strcmp(key, "audit_obj") == 0) {
2831                 struct obdo * oa = val;
2832                 struct dentry * dentry;
2833                 __u64 mask = oa->o_fid;
2834
2835                 dentry = filter_crow_object(obd, oa);
2836                 if (IS_ERR(dentry))
2837                         RETURN(PTR_ERR(dentry));
2838
2839                 rc = fsfilt_set_info(obd, obd->u.filter.fo_sb, dentry->d_inode,
2840                                      5, "audit", sizeof(mask), &mask);
2841
2842                 f_dput(dentry);
2843
2844                 RETURN(rc);
2845         }
2846
2847
2848         if (rc)
2849                 CDEBUG(D_IOCTL, "invalid key\n");
2850
2851         RETURN(rc);
2852 }
2853
2854 static int filter_get_info(struct obd_export *exp, __u32 keylen,
2855                            void *key, __u32 *vallen, void *val)
2856 {
2857         struct filter_export_data *fed = &exp->exp_filter_data;
2858         struct obd_device *obd;
2859         ENTRY;
2860
2861         obd = class_exp2obd(exp);
2862         if (obd == NULL) {
2863                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2864                        exp->exp_handle.h_cookie);
2865                 RETURN(-EINVAL);
2866         }
2867
2868         if (keylen == strlen("blocksize") &&
2869             memcmp(key, "blocksize", keylen) == 0) {
2870                 __u32 *blocksize = val;
2871                 *vallen = sizeof(*blocksize);
2872                 *blocksize = obd->u.filter.fo_sb->s_blocksize;
2873                 RETURN(0);
2874         }
2875
2876         if (keylen == strlen("blocksize_bits") &&
2877             memcmp(key, "blocksize_bits", keylen) == 0) {
2878                 __u32 *blocksize_bits = val;
2879                 *vallen = sizeof(*blocksize_bits);
2880                 *blocksize_bits = obd->u.filter.fo_sb->s_blocksize_bits;
2881                 RETURN(0);
2882         }
2883
2884         if (keylen >= strlen("last_id") && memcmp(key, "last_id", 7) == 0) {
2885                 obd_id *last_id = val;
2886                 *last_id = filter_last_id(&obd->u.filter, fed->fed_group);
2887                 RETURN(0);
2888         }
2889         if (keylen >= strlen("reint_log") && memcmp(key, "reint_log", 9) == 0) {
2890                 /*Get log_context handle*/
2891                 unsigned long *llh_handle = val;
2892                 *vallen = sizeof(unsigned long);
2893                 *llh_handle = (unsigned long)llog_get_context(&obd->obd_llogs,
2894                                                        LLOG_REINT_ORIG_CTXT);
2895                 RETURN(0);
2896         }
2897         if (keylen >= strlen("cache_sb") && memcmp(key, "cache_sb", 8) == 0) {
2898                 /*Get log_context handle*/
2899                 unsigned long *sb = val;
2900                 *vallen = sizeof(unsigned long);
2901                 *sb = (unsigned long)obd->u.filter.fo_sb;
2902                 RETURN(0);
2903         }
2904
2905         CDEBUG(D_IOCTL, "invalid key\n");
2906         RETURN(-EINVAL);
2907 }
2908
2909 struct obd_llogs *filter_grab_llog_for_group(struct obd_device *obd, int group,
2910                                              struct obd_export *export)
2911 {
2912         struct filter_group_llog *fglog, *nlog;
2913         char name[32] = "CATLIST";
2914         struct filter_obd *filter;
2915         struct llog_ctxt *ctxt;
2916         struct list_head *cur;
2917         int rc;
2918
2919         filter = &obd->u.filter;
2920
2921         spin_lock(&filter->fo_llog_list_lock);
2922         list_for_each(cur, &filter->fo_llog_list) {
2923                 fglog = list_entry(cur, struct filter_group_llog, list);
2924                 if (fglog->group == group) {
2925                         if (!(fglog->exp == NULL || fglog->exp == export || export == NULL))
2926                                 CWARN("%s: export for group %d changes: 0x%p -> 0x%p\n",
2927                                       obd->obd_name, group, fglog->exp, export);
2928                         spin_unlock(&filter->fo_llog_list_lock);
2929                         goto init;
2930                 }
2931         }
2932         spin_unlock(&filter->fo_llog_list_lock);
2933
2934         if (export == NULL)
2935                 RETURN(NULL);
2936
2937         OBD_ALLOC(fglog, sizeof(*fglog));
2938         if (fglog == NULL)
2939                 RETURN(NULL);
2940         fglog->group = group;
2941
2942         OBD_ALLOC(fglog->llogs, sizeof(struct obd_llogs));
2943         if (fglog->llogs == NULL) {
2944                 OBD_FREE(fglog, sizeof(*fglog));
2945                 RETURN(NULL);
2946         }
2947
2948         spin_lock(&filter->fo_llog_list_lock);
2949         list_for_each(cur, &filter->fo_llog_list) {
2950                 nlog = list_entry(cur, struct filter_group_llog, list);
2951                 LASSERT(nlog->group != group);
2952         }
2953         list_add(&fglog->list, &filter->fo_llog_list);
2954         spin_unlock(&filter->fo_llog_list_lock);
2955
2956         rc = obd_llog_cat_initialize(obd, fglog->llogs, 1, name);
2957         if (rc) {
2958                 OBD_FREE(fglog->llogs, sizeof(*(fglog->llogs)));
2959                 OBD_FREE(fglog, sizeof(*fglog));
2960                 RETURN(NULL);
2961         }
2962
2963 init:
2964         if (export) {
2965                 fglog->exp = export;
2966                 ctxt = llog_get_context(fglog->llogs, LLOG_UNLINK_REPL_CTXT);
2967                 LASSERT(ctxt != NULL);
2968
2969                 llog_receptor_accept(ctxt, export->exp_imp_reverse);
2970         }
2971         CDEBUG(D_OTHER, "%s: new llog 0x%p for group %u\n",
2972                obd->obd_name, fglog->llogs, group);
2973
2974         RETURN(fglog->llogs);
2975 }
2976
2977 int filter_iocontrol(unsigned int cmd, struct obd_export *exp,
2978                      int len, void *karg, void *uarg)
2979 {
2980         struct obd_device *obd = exp->exp_obd;
2981         struct obd_ioctl_data *data = karg;
2982         int rc = 0;
2983
2984         switch (cmd) {
2985         case OBD_IOC_ABORT_RECOVERY:
2986                 target_stop_recovery_thread(obd);
2987                 RETURN(0);
2988
2989         case OBD_IOC_SET_READONLY: {
2990                 void *handle;
2991                 struct super_block *sb = obd->u.filter.fo_sb;
2992                 struct inode *inode = sb->s_root->d_inode;
2993                 BDEVNAME_DECLARE_STORAGE(tmp);
2994                 CERROR("setting device %s read-only\n",
2995                        ll_bdevname(sb, tmp));
2996
2997                 handle = fsfilt_start(obd, inode, FSFILT_OP_MKNOD, NULL);
2998                 LASSERT(handle);
2999                 (void)fsfilt_commit(obd, sb, inode, handle, 1);
3000
3001                 ll_set_rdonly(ll_sbdev(obd->u.filter.fo_sb));
3002                 RETURN(0);
3003         }
3004
3005         case OBD_IOC_CATLOGLIST: {
3006                 rc = llog_catalog_list(obd, 1, data);
3007                 RETURN(rc);
3008         }
3009         case OBD_IOC_LLOG_CANCEL:
3010         case OBD_IOC_LLOG_REMOVE:
3011         case OBD_IOC_LLOG_INFO:
3012         case OBD_IOC_LLOG_PRINT: {
3013                 /* FIXME to be finished */
3014                 RETURN(-EOPNOTSUPP);
3015 /*
3016                 struct llog_ctxt *ctxt = NULL;
3017
3018                 push_ctxt(&saved, &ctxt->loc_ctxt, NULL);
3019                 rc = llog_ioctl(ctxt, cmd, data);
3020                 pop_ctxt(&saved, &ctxt->loc_ctxt, NULL);
3021
3022                 RETURN(rc);
3023 */
3024         }
3025
3026
3027         default:
3028                 RETURN(-EINVAL);
3029         }
3030         RETURN(0);
3031 }
3032
3033 static struct llog_operations filter_unlink_repl_logops;
3034 static struct llog_operations filter_size_orig_logops = {
3035         lop_setup: llog_obd_origin_setup,
3036         lop_cleanup: llog_catalog_cleanup,
3037         lop_add: llog_catalog_add,
3038 };
3039
3040 static int filter_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
3041                             struct obd_device *tgt, int count,
3042                             struct llog_catid *catid)
3043 {
3044         struct llog_ctxt *ctxt;
3045         int rc;
3046         ENTRY;
3047
3048         filter_unlink_repl_logops = llog_client_ops;
3049         filter_unlink_repl_logops.lop_cancel = llog_obd_repl_cancel;
3050         filter_unlink_repl_logops.lop_connect = llog_repl_connect;
3051         filter_unlink_repl_logops.lop_sync = llog_obd_repl_sync;
3052
3053         rc = obd_llog_setup(obd, llogs, LLOG_UNLINK_REPL_CTXT, tgt, 0, NULL,
3054                         &filter_unlink_repl_logops);
3055         if (rc)
3056                 RETURN(rc);
3057         /* FIXME - assign unlink_cb for filter's recovery */
3058         ctxt = llog_get_context(llogs, LLOG_UNLINK_REPL_CTXT);
3059         ctxt->llog_proc_cb = filter_recov_log_unlink_cb;
3060
3061         filter_size_orig_logops = llog_lvfs_ops;
3062 #if 0
3063         filter_size_orig_logops.lop_setup = llog_obd_origin_setup;
3064         filter_size_orig_logops.lop_cleanup = llog_catalog_cleanup;
3065         filter_size_orig_logops.lop_add = llog_catalog_add;
3066 #endif
3067         rc = obd_llog_setup(obd, llogs, LLOG_SIZE_ORIG_CTXT, tgt, 1,
3068                             &catid->lci_logid, &filter_size_orig_logops);
3069         RETURN(rc);
3070 }
3071
3072 static int filter_llog_finish(struct obd_device *obd,
3073                               struct obd_llogs *llogs, int count)
3074 {
3075         int rc;
3076         ENTRY;
3077
3078         rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_UNLINK_REPL_CTXT));
3079         if (rc)
3080                 RETURN(rc);
3081
3082         rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_SIZE_ORIG_CTXT));
3083         RETURN(rc);
3084 }
3085
3086 static int filter_llog_connect(struct obd_export *exp,
3087                                struct llogd_conn_body *body)
3088 {
3089         struct obd_device *obd = exp->exp_obd;
3090         struct llog_ctxt *ctxt;
3091         struct obd_llogs *llog;
3092         int rc;
3093         ENTRY;
3094
3095         CDEBUG(D_OTHER, "handle connect for %s: %u/%u/%u\n", obd->obd_name,
3096                (unsigned) body->lgdc_logid.lgl_ogr,
3097                (unsigned) body->lgdc_logid.lgl_oid,
3098                (unsigned) body->lgdc_logid.lgl_ogen);
3099         llog = filter_grab_llog_for_group(obd, body->lgdc_logid.lgl_ogr, exp);
3100         LASSERT(llog != NULL);
3101         ctxt = llog_get_context(llog, body->lgdc_ctxt_idx);
3102         rc = llog_connect(ctxt, 1, &body->lgdc_logid,
3103                           &body->lgdc_gen, NULL);
3104         if (rc != 0)
3105                 CERROR("failed to connect\n");
3106
3107         RETURN(rc);
3108 }
3109
3110 static struct dentry *filter_lvfs_id2dentry(__u64 id, __u32 gen,
3111                                             __u64 gr, void *data)
3112 {
3113         return filter_id2dentry(data, NULL, gr, id);
3114 }
3115
3116 static struct lvfs_callback_ops filter_lvfs_ops = {
3117         l_id2dentry:     filter_lvfs_id2dentry,
3118 };
3119
3120 static struct obd_ops filter_obd_ops = {
3121         .o_owner          = THIS_MODULE,
3122         .o_attach         = filter_attach,
3123         .o_detach         = filter_detach,
3124         .o_set_info       = filter_set_info,
3125         .o_get_info       = filter_get_info,
3126         .o_setup          = filter_setup,
3127         .o_precleanup     = filter_precleanup,
3128         .o_cleanup        = filter_cleanup,
3129         .o_process_config = filter_process_config,
3130         .o_connect        = filter_connect,
3131         .o_connect_post   = filter_connect_post,
3132         .o_disconnect     = filter_disconnect,
3133         .o_statfs         = filter_statfs,
3134         .o_getattr        = filter_getattr,
3135         .o_unpackmd       = filter_unpackmd,
3136         .o_create         = filter_create,
3137         .o_setattr        = filter_setattr,
3138         .o_destroy        = filter_destroy,
3139         .o_brw            = filter_brw,
3140         .o_punch          = filter_truncate,
3141         .o_sync           = filter_sync,
3142         .o_preprw         = filter_preprw,
3143         .o_commitrw       = filter_commitrw,
3144         .o_do_cow         = filter_do_cow,
3145         .o_write_extents  = filter_write_extents,
3146         .o_destroy_export = filter_destroy_export,
3147         .o_llog_init      = filter_llog_init,
3148         .o_llog_finish    = filter_llog_finish,
3149         .o_llog_connect   = filter_llog_connect,
3150         .o_iocontrol      = filter_iocontrol,
3151 };
3152
3153 static struct obd_ops filter_sanobd_ops = {
3154         .o_owner          = THIS_MODULE,
3155         .o_attach         = filter_attach,
3156         .o_detach         = filter_detach,
3157         .o_get_info       = filter_get_info,
3158         .o_setup          = filter_san_setup,
3159         .o_precleanup     = filter_precleanup,
3160         .o_cleanup        = filter_cleanup,
3161         .o_connect        = filter_connect,
3162         .o_connect_post   = filter_connect_post,
3163         .o_disconnect     = filter_disconnect,
3164         .o_statfs         = filter_statfs,
3165         .o_getattr        = filter_getattr,
3166         .o_unpackmd       = filter_unpackmd,
3167         .o_create         = filter_create,
3168         .o_setattr        = filter_setattr,
3169         .o_destroy        = filter_destroy,
3170         .o_brw            = filter_brw,
3171         .o_punch          = filter_truncate,
3172         .o_sync           = filter_sync,
3173         .o_preprw         = filter_preprw,
3174         .o_commitrw       = filter_commitrw,
3175         .o_do_cow         = filter_do_cow,
3176         .o_write_extents  = filter_write_extents,
3177         .o_san_preprw     = filter_san_preprw,
3178         .o_destroy_export = filter_destroy_export,
3179         .o_llog_init      = filter_llog_init,
3180         .o_llog_finish    = filter_llog_finish,
3181         .o_llog_connect   = filter_llog_connect,
3182         .o_iocontrol      = filter_iocontrol,
3183 };
3184
3185 static int __init obdfilter_init(void)
3186 {
3187         struct lprocfs_static_vars lvars;
3188         int size, rc;
3189
3190         printk(KERN_INFO "Lustre: Filtering OBD driver; info@clusterfs.com\n");
3191
3192         lprocfs_init_vars(filter, &lvars);
3193
3194         size = OBDFILTER_CREATED_SCRATCHPAD_ENTRIES *
3195                 sizeof(*obdfilter_created_scratchpad);
3196
3197         OBD_ALLOC(obdfilter_created_scratchpad, size);
3198         if (obdfilter_created_scratchpad == NULL) {
3199                 CERROR ("Can't allocate scratchpad\n");
3200                 return -ENOMEM;
3201         }
3202
3203         rc = class_register_type(&filter_obd_ops, NULL, lvars.module_vars,
3204                                  OBD_FILTER_DEVICENAME);
3205         if (rc) {
3206                 OBD_FREE(obdfilter_created_scratchpad, size);
3207                 return rc;
3208         }
3209
3210         rc = class_register_type(&filter_sanobd_ops, NULL, lvars.module_vars,
3211                                  OBD_FILTER_SAN_DEVICENAME);
3212         if (rc) {
3213                 class_unregister_type(OBD_FILTER_DEVICENAME);
3214                 OBD_FREE(obdfilter_created_scratchpad, size);
3215         }
3216         return rc;
3217 }
3218
3219 static void __exit obdfilter_exit(void)
3220 {
3221         class_unregister_type(OBD_FILTER_SAN_DEVICENAME);
3222         class_unregister_type(OBD_FILTER_DEVICENAME);
3223         OBD_FREE(obdfilter_created_scratchpad,
3224                  OBDFILTER_CREATED_SCRATCHPAD_ENTRIES *
3225                  sizeof(*obdfilter_created_scratchpad));
3226 }
3227
3228 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
3229 MODULE_DESCRIPTION("Lustre Filtering OBD driver");
3230 MODULE_LICENSE("GPL");
3231
3232 module_init(obdfilter_init);
3233 module_exit(obdfilter_exit);