lustre/mds/mds_fs.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  mds/mds_fs.c
   5  *  Lustre Metadata Server (MDS) filesystem interface code
   6  *
   7  *  Copyright (C) 2002, 2003 Cluster File Systems, Inc.
   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 #ifndef EXPORT_SYMTAB
  27 # define EXPORT_SYMTAB
  28 #endif
  29 #define DEBUG_SUBSYSTEM S_MDS
  30
  31 #include <linux/module.h>
  32 #include <linux/kmod.h>
  33 #include <linux/version.h>
  34 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
  35 #include <linux/mount.h>
  36 #endif
  37 #include <linux/lustre_mds.h>
  38 #include <linux/obd_class.h>
  39 #include <linux/obd_support.h>
  40 #include <linux/lustre_lib.h>
  41 #include <linux/lustre_fsfilt.h>
  42 #include <libcfs/list.h>
  43
  44 #include <linux/lustre_smfs.h>
  45 #include "mds_internal.h"
  46
  47 /* This limit is arbitrary, but for now we fit it in 1 page (32k clients) */
  48 #define MDS_MAX_CLIENTS (PAGE_SIZE * 8)
  49
  50 #define LAST_RCVD "last_rcvd"
  51 #define LOV_OBJID "lov_objid"
  52 #define LAST_FID  "last_fid"
  53 #define VIRT_FID  "virt_fid"
  54 #define CAPA_KEYS  "capa_key"
  55
  56 struct fidmap_entry {
  57         struct hlist_node fm_hash;
  58         struct lustre_id  fm_id;
  59 };
  60
  61 int mds_fidmap_init(struct obd_device *obd, int size)
  62 {
  63         struct mds_obd *mds = &obd->u.mds;
  64         struct hlist_head *head;
  65         int i = 0;
  66         ENTRY;
  67
  68         LASSERT(size > sizeof(sizeof(struct hlist_head)));
  69         mds->mds_fidmap_size = size / sizeof(struct hlist_head);
  70
  71         CWARN("allocating %lu fid mapping entries\n",
  72               (unsigned long)mds->mds_fidmap_size);
  73
  74         OBD_ALLOC(mds->mds_fidmap_table, size);
  75         if (!mds->mds_fidmap_table)
  76                 RETURN(-ENOMEM);
  77
  78         i = mds->mds_fidmap_size;
  79         head = mds->mds_fidmap_table;
  80         do {
  81                 INIT_HLIST_HEAD(head);
  82                 head++;
  83                 i--;
  84         } while(i);
  85
  86         RETURN(0);
  87 }
  88
  89 int mds_fidmap_cleanup(struct obd_device *obd)
  90 {
  91         struct hlist_node *node = NULL, *tmp = NULL;
  92         struct mds_obd *mds = &obd->u.mds;
  93         struct fidmap_entry *entry;
  94         struct hlist_head *head;
  95         int i = 0;
  96         ENTRY;
  97
  98         spin_lock(&mds->mds_fidmap_lock);
  99         for (i = 0, head = mds->mds_fidmap_table;
 100              i < mds->mds_fidmap_size; i++, head++) {
 101                 hlist_for_each_safe(node, tmp, head) {
 102                         entry = hlist_entry(node, struct fidmap_entry, fm_hash);
 103                         hlist_del_init(&entry->fm_hash);
 104                         OBD_FREE(entry, sizeof(*entry));
 105                 }
 106         }
 107         spin_unlock(&mds->mds_fidmap_lock);
 108         OBD_FREE(mds->mds_fidmap_table, mds->mds_fidmap_size *
 109                  sizeof(struct hlist_head));
 110         RETURN(0);
 111 }
 112
 113 static inline unsigned long
 114 const hashfn(struct obd_device *obd, __u64 fid)
 115 {
 116         struct mds_obd *mds = &obd->u.mds;
 117         return (unsigned long)(fid & (mds->mds_fidmap_size - 1));
 118 }
 119
 120 static struct fidmap_entry *
 121 __mds_fidmap_find(struct obd_device *obd, __u64 fid)
 122 {
 123         struct fidmap_entry *entry = NULL;
 124         struct mds_obd *mds = &obd->u.mds;
 125         struct hlist_node *node = NULL;
 126         struct hlist_head *head;
 127         ENTRY;
 128
 129         head = mds->mds_fidmap_table + hashfn(obd, fid);
 130         hlist_for_each(node, head) {
 131                 entry = hlist_entry(node, struct fidmap_entry, fm_hash);
 132                 if (id_fid(&entry->fm_id) == fid)
 133                         RETURN(entry);
 134         }
 135         RETURN(NULL);
 136 }
 137
 138 struct fidmap_entry *
 139 mds_fidmap_find(struct obd_device *obd, __u64 fid)
 140 {
 141         struct mds_obd *mds = &obd->u.mds;
 142         struct fidmap_entry *entry;
 143         ENTRY;
 144
 145         spin_lock(&mds->mds_fidmap_lock);
 146         entry = __mds_fidmap_find(obd, fid);
 147         spin_unlock(&mds->mds_fidmap_lock);
 148
 149         RETURN(entry);
 150 }
 151
 152 static void __mds_fidmap_insert(struct obd_device *obd,
 153                                 struct fidmap_entry *entry)
 154 {
 155         struct mds_obd *mds = &obd->u.mds;
 156         struct hlist_head *head;
 157         unsigned long idx;
 158         ENTRY;
 159
 160         idx = hashfn(obd, id_fid(&entry->fm_id));
 161         head = mds->mds_fidmap_table + idx;
 162         hlist_add_head(&entry->fm_hash, head);
 163
 164         EXIT;
 165 }
 166
 167 void mds_fidmap_insert(struct obd_device *obd,
 168                        struct fidmap_entry *entry)
 169 {
 170         struct mds_obd *mds = &obd->u.mds;
 171         ENTRY;
 172
 173         spin_lock(&mds->mds_fidmap_lock);
 174         __mds_fidmap_insert(obd, entry);
 175         spin_unlock(&mds->mds_fidmap_lock);
 176
 177         EXIT;
 178 }
 179
 180 static void __mds_fidmap_remove(struct obd_device *obd,
 181                                 struct fidmap_entry *entry)
 182 {
 183         ENTRY;
 184         hlist_del_init(&entry->fm_hash);
 185         EXIT;
 186 }
 187
 188 void mds_fidmap_remove(struct obd_device *obd,
 189                        struct fidmap_entry *entry)
 190 {
 191         struct mds_obd *mds = &obd->u.mds;
 192         ENTRY;
 193
 194         spin_lock(&mds->mds_fidmap_lock);
 195         __mds_fidmap_remove(obd, entry);
 196         spin_unlock(&mds->mds_fidmap_lock);
 197
 198         EXIT;
 199 }
 200
 201 /* creates new mapping remote fid -> local inode store cookie. Both are saved in
 202  * lustre_id for better usability, as all mds function use lustre_id as input
 203  * params.*/
 204 int mds_fidmap_add(struct obd_device *obd,
 205                    struct lustre_id *id)
 206 {
 207         struct mds_obd *mds = &obd->u.mds;
 208         struct fidmap_entry *entry;
 209         ENTRY;
 210
 211         OBD_ALLOC(entry, sizeof(*entry));
 212         if (!entry)
 213                 RETURN(-ENOMEM);
 214
 215         entry->fm_id = *id;
 216
 217         spin_lock(&mds->mds_fidmap_lock);
 218         if (!__mds_fidmap_find(obd, id_fid(id))) {
 219                 __mds_fidmap_insert(obd, entry);
 220                 spin_unlock(&mds->mds_fidmap_lock);
 221                 CDEBUG(D_INODE, "added mapping to "DLID4"\n",
 222                        OLID4(id));
 223                 RETURN(1);
 224         }
 225         spin_unlock(&mds->mds_fidmap_lock);
 226         OBD_FREE(entry, sizeof(*entry));
 227
 228         RETURN(0);
 229 }
 230
 231 /* removes mapping using fid component from passed @id */
 232 void mds_fidmap_del(struct obd_device *obd,
 233                     struct lustre_id *id)
 234 {
 235         struct mds_obd *mds = &obd->u.mds;
 236         struct fidmap_entry *entry;
 237         ENTRY;
 238
 239         spin_lock(&mds->mds_fidmap_lock);
 240         entry = __mds_fidmap_find(obd, id_fid(id));
 241         if (entry) {
 242                 __mds_fidmap_remove(obd, entry);
 243                 spin_unlock(&mds->mds_fidmap_lock);
 244                 OBD_FREE(entry, sizeof(*entry));
 245                 CDEBUG(D_INODE, "removed mapping to "DLID4"\n",
 246                        OLID4(id));
 247                 goto out;
 248         }
 249         spin_unlock(&mds->mds_fidmap_lock);
 250 out:
 251         EXIT;
 252 }
 253
 254 struct lustre_id *mds_fidmap_lookup(struct obd_device *obd,
 255                                     struct lustre_id *id)
 256 {
 257         struct mds_obd *mds = &obd->u.mds;
 258         struct fidmap_entry *entry;
 259         ENTRY;
 260
 261         spin_lock(&mds->mds_fidmap_lock);
 262         entry = __mds_fidmap_find(obd, id_fid(id));
 263         spin_unlock(&mds->mds_fidmap_lock);
 264
 265         if (!entry)
 266                 RETURN(NULL);
 267
 268         RETURN(&entry->fm_id);
 269 }
 270
 271 /* Add client data to the MDS.  We use a bitmap to locate a free space
 272  * in the last_rcvd file if cl_off is -1 (i.e. a new client).
 273  * Otherwise, we have just read the data from the last_rcvd file and
 274  * we know its offset.
 275  */
 276 int mds_client_add(struct obd_device *obd, struct mds_obd *mds,
 277                    struct mds_export_data *med, int cl_idx)
 278 {
 279         unsigned long *bitmap = mds->mds_client_bitmap;
 280         int new_client = (cl_idx == -1);
 281         ENTRY;
 282
 283         LASSERT(bitmap != NULL);
 284
 285         /* XXX if mcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
 286         if (!strcmp((char *)med->med_mcd->mcd_uuid, (char *)obd->obd_uuid.uuid))
 287                 RETURN(0);
 288
 289         /* the bitmap operations can handle cl_idx > sizeof(long) * 8, so
 290          * there's no need for extra complication here
 291          */
 292         if (new_client) {
 293                 cl_idx = find_first_zero_bit(bitmap, MDS_MAX_CLIENTS);
 294         repeat:
 295                 if (cl_idx >= MDS_MAX_CLIENTS) {
 296                         CERROR("no room for clients - fix MDS_MAX_CLIENTS\n");
 297                         return -ENOMEM;
 298                 }
 299                 if (test_and_set_bit(cl_idx, bitmap)) {
 300                         cl_idx = find_next_zero_bit(bitmap, MDS_MAX_CLIENTS,
 301                                                     cl_idx);
 302                         goto repeat;
 303                 }
 304         } else {
 305                 if (test_and_set_bit(cl_idx, bitmap)) {
 306                         CERROR("MDS client %d: bit already set in bitmap!!\n",
 307                                cl_idx);
 308                         LBUG();
 309                 }
 310         }
 311
 312         CDEBUG(D_INFO, "client at idx %d with UUID '%s' added\n",
 313                cl_idx, med->med_mcd->mcd_uuid);
 314
 315         med->med_idx = cl_idx;
 316         med->med_off = le32_to_cpu(mds->mds_server_data->msd_client_start) +
 317                 (cl_idx * le16_to_cpu(mds->mds_server_data->msd_client_size));
 318
 319         if (new_client) {
 320                 struct file *file = mds->mds_rcvd_filp;
 321                 struct lvfs_run_ctxt saved;
 322                 loff_t off = med->med_off;
 323                 int rc;
 324
 325                 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 326                 rc = fsfilt_write_record(obd, file, med->med_mcd,
 327                                          sizeof(*med->med_mcd), &off, 1);
 328                 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 329
 330                 if (rc)
 331                         return rc;
 332                 CDEBUG(D_INFO, "wrote client mcd at idx %u off %llu (len %u)\n",
 333                        med->med_idx, med->med_off,
 334                        (unsigned int)sizeof(*med->med_mcd));
 335         }
 336         return 0;
 337 }
 338
 339 int mds_client_free(struct obd_export *exp, int clear_client)
 340 {
 341         struct mds_export_data *med = &exp->exp_mds_data;
 342         struct mds_obd *mds = &exp->exp_obd->u.mds;
 343         unsigned long *bitmap = mds->mds_client_bitmap;
 344         struct obd_device *obd = exp->exp_obd;
 345         struct mds_client_data zero_mcd;
 346         struct lvfs_run_ctxt saved;
 347         int rc;
 348
 349         if (!med->med_mcd)
 350                 RETURN(0);
 351
 352         /* XXX if mcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
 353         if (!strcmp((char *)med->med_mcd->mcd_uuid, (char *)obd->obd_uuid.uuid))
 354                 GOTO(free_and_out, 0);
 355
 356         CDEBUG(D_INFO, "freeing client at idx %u (%lld)with UUID '%s'\n",
 357                med->med_idx, med->med_off, med->med_mcd->mcd_uuid);
 358
 359         LASSERT(bitmap);
 360
 361         /* Clear the bit _after_ zeroing out the client so we don't
 362            race with mds_client_add and zero out new clients.*/
 363         if (!test_bit(med->med_idx, bitmap)) {
 364                 CERROR("MDS client %u: bit already clear in bitmap!!\n",
 365                        med->med_idx);
 366                 LBUG();
 367         }
 368
 369         if (clear_client) {
 370                 memset(&zero_mcd, 0, sizeof zero_mcd);
 371                 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 372                 rc = fsfilt_write_record(obd, mds->mds_rcvd_filp, &zero_mcd,
 373                                          sizeof(zero_mcd), &med->med_off, 1);
 374                 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 375
 376                 CDEBUG(rc == 0 ? D_INFO : D_ERROR,
 377                        "zeroing out client %s idx %u in %s rc %d\n",
 378                        med->med_mcd->mcd_uuid, med->med_idx, LAST_RCVD, rc);
 379         }
 380
 381         if (!test_and_clear_bit(med->med_idx, bitmap)) {
 382                 CERROR("MDS client %u: bit already clear in bitmap!!\n",
 383                        med->med_idx);
 384                 LBUG();
 385         }
 386
 387
 388         /* Make sure the server's last_transno is up to date. Do this
 389          * after the client is freed so we know all the client's
 390          * transactions have been committed. */
 391         mds_update_server_data(exp->exp_obd, 1);
 392
 393 free_and_out:
 394         OBD_FREE(med->med_mcd, sizeof(*med->med_mcd));
 395         med->med_mcd = NULL;
 396         return 0;
 397 }
 398
 399 static int mds_server_free_data(struct mds_obd *mds)
 400 {
 401         OBD_FREE(mds->mds_client_bitmap, MDS_MAX_CLIENTS / 8);
 402         OBD_FREE(mds->mds_server_data, sizeof(*mds->mds_server_data));
 403         mds->mds_server_data = NULL;
 404
 405         return 0;
 406 }
 407
 408 static int mds_read_last_fid(struct obd_device *obd, struct file *file)
 409 {
 410         int rc = 0;
 411         loff_t off = 0;
 412         struct mds_obd *mds = &obd->u.mds;
 413         unsigned long last_fid_size = file->f_dentry->d_inode->i_size;
 414         ENTRY;
 415
 416         if (last_fid_size == 0) {
 417                 CWARN("%s: initializing new %s\n", obd->obd_name,
 418                       file->f_dentry->d_name.name);
 419
 420                 /*
 421                  * as fid is used for forming res_id for locking, it should not
 422                  * be zero. This will keep us out of lots possible problems,
 423                  * asserts, etc.
 424                  */
 425                 mds_set_last_fid(obd, 0);
 426         } else {
 427                 __u64 lastfid;
 428
 429                 rc = fsfilt_read_record(obd, file, &lastfid,
 430                                         sizeof(lastfid), &off);
 431                 if (rc) {
 432                         CERROR("error reading MDS %s: rc = %d\n",
 433                                file->f_dentry->d_name.name, rc);
 434                         RETURN(rc);
 435                 }
 436
 437                 /*
 438                  * make sure, that fid is up-to-date.
 439                  */
 440                 mds_set_last_fid(obd, lastfid);
 441         }
 442
 443         CDEBUG(D_INODE, "%s: server last_fid: "LPU64"\n",
 444                obd->obd_name, mds->mds_last_fid);
 445
 446         rc = mds_update_last_fid(obd, NULL, 1);
 447         RETURN(rc);
 448 }
 449
 450 static int mds_read_last_rcvd(struct obd_device *obd, struct file *file)
 451 {
 452         unsigned long last_rcvd_size = file->f_dentry->d_inode->i_size;
 453         struct mds_obd *mds = &obd->u.mds;
 454         struct mds_server_data *msd = NULL;
 455         struct mds_client_data *mcd = NULL;
 456         loff_t off = 0;
 457         __u64 mount_count;
 458         int cl_idx, rc = 0;
 459         ENTRY;
 460
 461         /* ensure padding in the struct is the correct size */
 462         LASSERT(offsetof(struct mds_server_data, msd_padding) +
 463                 sizeof(msd->msd_padding) == MDS_LR_SERVER_SIZE);
 464         LASSERT(offsetof(struct mds_client_data, mcd_padding) +
 465                 sizeof(mcd->mcd_padding) == MDS_LR_CLIENT_SIZE);
 466
 467         OBD_ALLOC_WAIT(msd, sizeof(*msd));
 468         if (!msd)
 469                 RETURN(-ENOMEM);
 470
 471         OBD_ALLOC_WAIT(mds->mds_client_bitmap, MDS_MAX_CLIENTS / 8);
 472         if (!mds->mds_client_bitmap) {
 473                 OBD_FREE(msd, sizeof(*msd));
 474                 RETURN(-ENOMEM);
 475         }
 476
 477         mds->mds_server_data = msd;
 478
 479         if (last_rcvd_size == 0) {
 480                 CWARN("%s: initializing new %s\n", obd->obd_name,
 481                       file->f_dentry->d_name.name);
 482
 483                 memcpy(msd->msd_uuid, obd->obd_uuid.uuid,sizeof(msd->msd_uuid));
 484                 msd->msd_last_transno = 0;
 485                 mount_count = msd->msd_mount_count = 0;
 486                 msd->msd_server_size = cpu_to_le32(MDS_LR_SERVER_SIZE);
 487                 msd->msd_client_start = cpu_to_le32(MDS_LR_CLIENT_START);
 488                 msd->msd_client_size = cpu_to_le16(MDS_LR_CLIENT_SIZE);
 489                 msd->msd_feature_rocompat = cpu_to_le32(MDS_ROCOMPAT_LOVOBJID);
 490         } else {
 491                 rc = fsfilt_read_record(obd, file, msd, sizeof(*msd), &off);
 492                 if (rc) {
 493                         CERROR("error reading MDS %s: rc = %d\n",
 494                                file->f_dentry->d_name.name, rc);
 495                         GOTO(err_msd, rc);
 496                 }
 497                 if (strcmp((char *)msd->msd_uuid, (char *)obd->obd_uuid.uuid)) {
 498                         CERROR("OBD UUID %s does not match last_rcvd UUID %s\n",
 499                                obd->obd_uuid.uuid, msd->msd_uuid);
 500                         GOTO(err_msd, rc = -EINVAL);
 501                 }
 502                 mount_count = le64_to_cpu(msd->msd_mount_count);
 503         }
 504         if (msd->msd_feature_incompat & ~cpu_to_le32(MDS_INCOMPAT_SUPP)) {
 505                 CERROR("unsupported incompat feature %x\n",
 506                        le32_to_cpu(msd->msd_feature_incompat) &
 507                        ~MDS_INCOMPAT_SUPP);
 508                 GOTO(err_msd, rc = -EINVAL);
 509         }
 510         /* XXX updating existing b_devel fs only, can be removed in future */
 511         msd->msd_feature_rocompat = cpu_to_le32(MDS_ROCOMPAT_LOVOBJID);
 512         if (msd->msd_feature_rocompat & ~cpu_to_le32(MDS_ROCOMPAT_SUPP)) {
 513                 CERROR("unsupported read-only feature %x\n",
 514                        le32_to_cpu(msd->msd_feature_rocompat) &
 515                        ~MDS_ROCOMPAT_SUPP);
 516                 /* Do something like remount filesystem read-only */
 517                 GOTO(err_msd, rc = -EINVAL);
 518         }
 519
 520         mds->mds_last_transno = le64_to_cpu(msd->msd_last_transno);
 521
 522         CDEBUG(D_INODE, "%s: server last_transno: "LPU64"\n",
 523                obd->obd_name, mds->mds_last_transno);
 524         CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
 525                obd->obd_name, mount_count + 1);
 526         CDEBUG(D_INODE, "%s: server data size: %u\n",
 527                obd->obd_name, le32_to_cpu(msd->msd_server_size));
 528         CDEBUG(D_INODE, "%s: per-client data start: %u\n",
 529                obd->obd_name, le32_to_cpu(msd->msd_client_start));
 530         CDEBUG(D_INODE, "%s: per-client data size: %u\n",
 531                obd->obd_name, le32_to_cpu(msd->msd_client_size));
 532         CDEBUG(D_INODE, "%s: last_rcvd size: %lu\n",
 533                obd->obd_name, last_rcvd_size);
 534         CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
 535                last_rcvd_size <= le32_to_cpu(msd->msd_client_start) ? 0 :
 536                (last_rcvd_size - le32_to_cpu(msd->msd_client_start)) /
 537                 le16_to_cpu(msd->msd_client_size));
 538
 539         /* When we do a clean MDS shutdown, we save the last_transno into
 540          * the header.  If we find clients with higher last_transno values
 541          * then those clients may need recovery done. */
 542         for (cl_idx = 0, off = le32_to_cpu(msd->msd_client_start);
 543              off < last_rcvd_size; cl_idx++) {
 544                 __u64 last_transno;
 545                 struct obd_export *exp;
 546                 struct mds_export_data *med;
 547
 548                 if (!mcd) {
 549                         OBD_ALLOC_WAIT(mcd, sizeof(*mcd));
 550                         if (!mcd)
 551                                 GOTO(err_client, rc = -ENOMEM);
 552                 }
 553
 554                 /* Don't assume off is incremented properly by
 555                  * fsfilt_read_record(), in case sizeof(*mcd)
 556                  * isn't the same as msd->msd_client_size.  */
 557                 off = le32_to_cpu(msd->msd_client_start) +
 558                         cl_idx * le16_to_cpu(msd->msd_client_size);
 559                 rc = fsfilt_read_record(obd, file, mcd, sizeof(*mcd), &off);
 560                 if (rc) {
 561                         CERROR("error reading MDS %s idx %d, off %llu: rc %d\n",
 562                                file->f_dentry->d_name.name, cl_idx, off, rc);
 563                         break; /* read error shouldn't cause startup to fail */
 564                 }
 565
 566                 if (mcd->mcd_uuid[0] == '\0') {
 567                         CDEBUG(D_INFO, "skipping zeroed client at offset %d\n",
 568                                cl_idx);
 569                         continue;
 570                 }
 571
 572                 last_transno = le64_to_cpu(mcd->mcd_last_transno) >
 573                                le64_to_cpu(mcd->mcd_last_close_transno) ?
 574                                le64_to_cpu(mcd->mcd_last_transno) :
 575                                le64_to_cpu(mcd->mcd_last_close_transno);
 576
 577                 /* These exports are cleaned up by mds_disconnect(), so they
 578                  * need to be set up like real exports as mds_connect() does.
 579                  */
 580                 CDEBUG(D_HA|D_WARNING,"RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64
 581                        " srv lr: "LPU64" lx: "LPU64"\n", mcd->mcd_uuid, cl_idx,
 582                        last_transno, le64_to_cpu(msd->msd_last_transno),
 583                        mcd->mcd_last_xid);
 584
 585                 exp = class_new_export(obd);
 586                 if (exp == NULL)
 587                         GOTO(err_client, rc = -ENOMEM);
 588
 589                 memcpy(&exp->exp_client_uuid.uuid, mcd->mcd_uuid,
 590                        sizeof exp->exp_client_uuid.uuid);
 591                 med = &exp->exp_mds_data;
 592                 med->med_mcd = mcd;
 593                 mds_client_add(obd, mds, med, cl_idx);
 594                 /* create helper if export init gets more complex */
 595                 INIT_LIST_HEAD(&med->med_open_head);
 596                 spin_lock_init(&med->med_open_lock);
 597
 598                 mcd = NULL;
 599                 exp->exp_connected = 0;
 600                 exp->exp_req_replay_needed = 1;
 601                 obd->obd_recoverable_clients++;
 602                 obd->obd_max_recoverable_clients++;
 603
 604                 /* track clients to separate req replay
 605                  * from lock replay. bug 6063 */
 606                 atomic_inc(&obd->obd_req_replay_clients);
 607                 exp->exp_req_replay_needed = 1;
 608                 atomic_inc(&obd->obd_lock_replay_clients);
 609                 exp->exp_lock_replay_needed = 1;
 610
 611                 class_export_put(exp);
 612
 613                 CDEBUG(D_OTHER, "client at idx %d has last_transno = "LPU64"\n",
 614                        cl_idx, last_transno);
 615
 616                 if (last_transno > mds->mds_last_transno)
 617                        mds->mds_last_transno = last_transno;
 618         }
 619         if (mcd)
 620                 OBD_FREE(mcd, sizeof(*mcd));
 621         obd->obd_last_committed = mds->mds_last_transno;
 622         if (obd->obd_recoverable_clients) {
 623                 CWARN("RECOVERY: service %s, %d recoverable clients, "
 624                       "last_transno "LPU64"\n", obd->obd_name,
 625                       obd->obd_recoverable_clients, mds->mds_last_transno);
 626                 obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
 627                 target_start_recovery_thread(obd, mds_handle);
 628                 obd->obd_recovery_start = LTIME_S(CURRENT_TIME);
 629         }
 630
 631         mds->mds_mount_count = mount_count + 1;
 632         msd->msd_mount_count = cpu_to_le64(mds->mds_mount_count);
 633
 634         /* save it, so mount count and last_transno is current */
 635         rc = mds_update_server_data(obd, 1);
 636         if (rc)
 637                 GOTO(err_client, rc);
 638
 639         RETURN(0);
 640
 641 err_client:
 642         class_disconnect_exports(obd, 0);
 643 err_msd:
 644         mds_server_free_data(mds);
 645         RETURN(rc);
 646 }
 647
 648 /*
 649  * sets up root inode lustre_id. It tries to read it first from root inode and
 650  * if it is not there, new rootid is allocated and saved there.
 651  */
 652 int mds_fs_setup_rootid(struct obd_device *obd)
 653 {
 654         int rc = 0;
 655         void *handle;
 656         struct inode *inode;
 657         struct dentry *dentry;
 658         struct mds_obd *mds = &obd->u.mds;
 659         ENTRY;
 660
 661         /* getting root directory and setup its fid. */
 662         dentry = mds_id2dentry(obd, &mds->mds_rootid, NULL);
 663         if (IS_ERR(dentry)) {
 664                 CERROR("Can't find ROOT by "DLID4", err = %d\n",
 665                        OLID4(&mds->mds_rootid), (int)PTR_ERR(dentry));
 666                 RETURN(PTR_ERR(dentry));
 667         }
 668
 669         inode = dentry->d_inode;
 670         LASSERT(dentry->d_inode);
 671
 672         rc = mds_pack_inode2id(obd, &mds->mds_rootid, inode, 1);
 673         if (rc && rc != -ENODATA)
 674                 GOTO(out_dentry, rc);
 675
 676         if (rc) {
 677                 if (rc != -ENODATA)
 678                         GOTO(out_dentry, rc);
 679         } else {
 680                 /* rootid is filled by mds_read_inode_sid(), so we do not need
 681                  * to allocate it and update. */
 682                 LASSERT(id_group(&mds->mds_rootid) == mds->mds_num);
 683                 mds_set_last_fid(obd, id_fid(&mds->mds_rootid));
 684
 685                 rc = mds_fidmap_add(obd, &mds->mds_rootid);
 686                 if (rc > 0)
 687                         rc = 0;
 688
 689                 GOTO(out_dentry, rc);
 690         }
 691
 692         /* allocating new one, as it is not found in root inode. */
 693         handle = fsfilt_start(obd, inode,
 694                               FSFILT_OP_SETATTR, NULL);
 695
 696         if (IS_ERR(handle)) {
 697                 rc = PTR_ERR(handle);
 698                 CERROR("fsfilt_start() failed, rc = %d\n", rc);
 699                 GOTO(out_dentry, rc);
 700         }
 701
 702         mds_inode2id(obd, &mds->mds_rootid, inode, mds_alloc_fid(obd));
 703         rc = mds_update_inode_ids(obd, inode, handle, &mds->mds_rootid, NULL);
 704         if (rc) {
 705                 CERROR("mds_update_inode_ids() failed, rc = %d\n", rc);
 706                 GOTO(out_dentry, rc);
 707         }
 708
 709         rc = mds_fidmap_add(obd, &mds->mds_rootid);
 710         if (rc < 0)
 711                 GOTO(out_dentry, rc);
 712         else
 713                 rc = 0;
 714
 715         rc = fsfilt_commit(obd, mds->mds_sb, inode, handle, 0);
 716         if (rc)
 717                 CERROR("fsfilt_commit() failed, rc = %d\n", rc);
 718
 719         EXIT;
 720 out_dentry:
 721         l_dput(dentry);
 722         if (rc == 0)
 723                 CWARN("%s: rootid: "DLID4"\n", obd->obd_name,
 724                       OLID4(&mds->mds_rootid));
 725         return rc;
 726 }
 727
 728 static int mds_update_virtid_fid(struct obd_device *obd,
 729                                  void *handle, int force_sync)
 730 {
 731         struct mds_obd *mds = &obd->u.mds;
 732         struct file *filp = mds->mds_virtid_filp;
 733         struct lvfs_run_ctxt saved;
 734         loff_t off = 0;
 735         int rc = 0;
 736         ENTRY;
 737
 738         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 739         rc = fsfilt_write_record(obd, filp, &mds->mds_virtid_fid,
 740                                  sizeof(mds->mds_virtid_fid),
 741                                  &off, force_sync);
 742         if (rc) {
 743                 CERROR("error writing MDS virtid_fid #"LPU64
 744                        ", err = %d\n", mds->mds_virtid_fid, rc);
 745         }
 746
 747         CDEBUG(D_SUPER, "wrote virtid fid #"LPU64" at idx "
 748                "%llu: err = %d\n", mds->mds_virtid_fid,
 749                off, rc);
 750         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 751
 752         RETURN(rc);
 753 }
 754
 755 static int mds_read_virtid_fid(struct obd_device *obd, struct file *file)
 756 {
 757         int rc = 0;
 758         loff_t off = 0;
 759         struct mds_obd *mds = &obd->u.mds;
 760         unsigned long virtid_fid_size = file->f_dentry->d_inode->i_size;
 761         ENTRY;
 762
 763         if (virtid_fid_size == 0) {
 764                 mds->mds_virtid_fid = mds_alloc_fid(obd);
 765         } else {
 766                 rc = fsfilt_read_record(obd, file, &mds->mds_virtid_fid,
 767                                         sizeof(mds->mds_virtid_fid), &off);
 768                 if (rc) {
 769                         CERROR("error reading MDS %s: rc = %d\n",
 770                                file->f_dentry->d_name.name, rc);
 771                         RETURN(rc);
 772                 }
 773         }
 774         rc = mds_update_virtid_fid(obd, NULL, 1);
 775
 776         RETURN(rc);
 777 }
 778
 779 /*
 780  * initializes lustre_id for virtual id directory, it is needed sometimes, as it
 781  * is possible that it will be the parent for object an operations is going to
 782  * be performed on.
 783  */
 784 int mds_fs_setup_virtid(struct obd_device *obd)
 785 {
 786         int rc = 0;
 787         void *handle;
 788         struct lustre_id sid;
 789         struct mds_obd *mds = &obd->u.mds;
 790         struct inode *inode = mds->mds_id_dir->d_inode;
 791         ENTRY;
 792
 793         handle = fsfilt_start(obd, inode,
 794                               FSFILT_OP_SETATTR, NULL);
 795
 796         if (IS_ERR(handle)) {
 797                 rc = PTR_ERR(handle);
 798                 CERROR("fsfilt_start() failed, rc = %d\n", rc);
 799                 RETURN(rc);
 800         }
 801
 802         id_group(&sid) = mds->mds_num;
 803         id_fid(&sid) = mds->mds_virtid_fid;
 804
 805         id_ino(&sid) = inode->i_ino;
 806         id_gen(&sid) = inode->i_generation;
 807         id_type(&sid) = (S_IFMT & inode->i_mode);
 808
 809         rc = mds_update_inode_ids(obd, inode, handle, &sid, NULL);
 810
 811         if (rc) {
 812                 CERROR("mds_update_inode_ids() failed, rc = %d\n", rc);
 813                 RETURN(rc);
 814         }
 815
 816         rc = mds_fidmap_add(obd, &sid);
 817         if (rc < 0)
 818                 RETURN(rc);
 819         else
 820                 rc = 0;
 821
 822         rc = fsfilt_commit(obd, mds->mds_sb, inode, handle, 0);
 823         if (rc) {
 824                 CERROR("fsfilt_commit() failed, rc = %d\n", rc);
 825                 RETURN(rc);
 826         }
 827
 828         RETURN(rc);
 829 }
 830
 831 #define MDS_FIDMAP_SIZE (2*PAGE_SIZE)
 832
 833 int mds_fs_setup(struct obd_device *obd, struct vfsmount *mnt)
 834 {
 835         struct mds_obd *mds = &obd->u.mds;
 836         struct lvfs_run_ctxt saved;
 837         struct dentry *dentry;
 838         struct file *file;
 839         int rc;
 840         ENTRY;
 841
 842         rc = cleanup_group_info();
 843         if (rc)
 844                 RETURN(rc);
 845
 846         mds->mds_vfsmnt = mnt;
 847         mds->mds_sb = mnt->mnt_root->d_inode->i_sb;
 848
 849         fsfilt_setup(obd, mds->mds_sb);
 850
 851         OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
 852         obd->obd_lvfs_ctxt.pwdmnt = mnt;
 853         obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
 854         obd->obd_lvfs_ctxt.fs = get_ds();
 855         obd->obd_lvfs_ctxt.cb_ops = mds_lvfs_ops;
 856
 857         /* setup the directory tree */
 858         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 859         dentry = simple_mkdir(current->fs->pwd, "ROOT", 0755, 0);
 860         if (IS_ERR(dentry)) {
 861                 rc = PTR_ERR(dentry);
 862                 CERROR("cannot create ROOT directory: rc = %d\n", rc);
 863                 GOTO(err_pop, rc);
 864         }
 865
 866         mdc_pack_id(&mds->mds_rootid, dentry->d_inode->i_ino,
 867                     dentry->d_inode->i_generation, S_IFDIR, 0, 0);
 868
 869         dput(dentry);
 870
 871         dentry = lookup_one_len("__iopen__", current->fs->pwd,
 872                                 strlen("__iopen__"));
 873         if (IS_ERR(dentry)) {
 874                 rc = PTR_ERR(dentry);
 875                 CERROR("cannot lookup __iopen__ directory: rc = %d\n", rc);
 876                 GOTO(err_pop, rc);
 877         }
 878         mds->mds_id_de = dentry;
 879         if (!dentry->d_inode || is_bad_inode(dentry->d_inode)) {
 880                 rc = -ENOENT;
 881                 CERROR("__iopen__ directory has no inode? rc = %d\n", rc);
 882                 GOTO(err_id_de, rc);
 883         }
 884
 885         dentry = simple_mkdir(current->fs->pwd, "PENDING", 0777, 1);
 886         if (IS_ERR(dentry)) {
 887                 rc = PTR_ERR(dentry);
 888                 CERROR("cannot create PENDING directory: rc = %d\n", rc);
 889                 GOTO(err_id_de, rc);
 890         }
 891         mds->mds_pending_dir = dentry;
 892
 893         dentry = simple_mkdir(current->fs->pwd, "LOGS", 0777, 1);
 894         if (IS_ERR(dentry)) {
 895                 rc = PTR_ERR(dentry);
 896                 CERROR("cannot create LOGS directory: rc = %d\n", rc);
 897                 GOTO(err_pending, rc);
 898         }
 899         mds->mds_logs_dir = dentry;
 900
 901         dentry = simple_mkdir(current->fs->pwd, "OBJECTS", 0777, 1);
 902         if (IS_ERR(dentry)) {
 903                 rc = PTR_ERR(dentry);
 904                 CERROR("cannot create OBJECTS directory: rc = %d\n", rc);
 905                 GOTO(err_logs, rc);
 906         }
 907         mds->mds_objects_dir = dentry;
 908
 909         dentry = simple_mkdir(current->fs->pwd, "FIDS", 0777, 1);
 910         if (IS_ERR(dentry)) {
 911                 rc = PTR_ERR(dentry);
 912                 CERROR("cannot create FIDS directory: rc = %d\n", rc);
 913                 GOTO(err_objects, rc);
 914         }
 915         mds->mds_id_dir = dentry;
 916
 917         dentry = simple_mkdir(current->fs->pwd, "UNNAMED", 0777, 1);
 918         if (IS_ERR(dentry)) {
 919                 rc = PTR_ERR(dentry);
 920                 CERROR("cannot create UNNAMED directory: rc = %d\n", rc);
 921                 GOTO(err_id_dir, rc);
 922         }
 923         mds->mds_unnamed_dir = dentry;
 924
 925         /* open and test the last rcvd file */
 926         file = filp_open(LAST_RCVD, O_RDWR | O_CREAT, 0644);
 927         if (IS_ERR(file)) {
 928                 rc = PTR_ERR(file);
 929                 CERROR("cannot open/create %s file: rc = %d\n", LAST_RCVD, rc);
 930                 GOTO(err_unnamed, rc = PTR_ERR(file));
 931         }
 932         mds->mds_rcvd_filp = file;
 933
 934         if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
 935                 CERROR("%s is not a regular file!: mode = %o\n", LAST_RCVD,
 936                        file->f_dentry->d_inode->i_mode);
 937                 GOTO(err_last_rcvd, rc = -ENOENT);
 938         }
 939
 940         rc = mds_read_last_rcvd(obd, file);
 941         if (rc) {
 942                 CERROR("cannot read %s: rc = %d\n", LAST_RCVD, rc);
 943                 GOTO(err_last_rcvd, rc);
 944         }
 945
 946         /* open and test last fid file */
 947         file = filp_open(LAST_FID, O_RDWR | O_CREAT, 0644);
 948         if (IS_ERR(file)) {
 949                 rc = PTR_ERR(file);
 950                 CERROR("cannot open/create %s file: rc = %d\n",
 951                        LAST_FID, rc);
 952                 GOTO(err_client, rc = PTR_ERR(file));
 953         }
 954         mds->mds_fid_filp = file;
 955         if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
 956                 CERROR("%s is not a regular file!: mode = %o\n",
 957                        LAST_FID, file->f_dentry->d_inode->i_mode);
 958                 GOTO(err_last_fid, rc = -ENOENT);
 959         }
 960
 961         rc = mds_read_last_fid(obd, file);
 962         if (rc) {
 963                 CERROR("cannot read %s: rc = %d\n", LAST_FID, rc);
 964                 GOTO(err_last_fid, rc);
 965         }
 966
 967         /* open and test virtid fid file */
 968         file = filp_open(VIRT_FID, O_RDWR | O_CREAT, 0644);
 969         if (IS_ERR(file)) {
 970                 rc = PTR_ERR(file);
 971                 CERROR("cannot open/create %s file: rc = %d\n",
 972                        VIRT_FID, rc);
 973                 GOTO(err_last_fid, rc = PTR_ERR(file));
 974         }
 975         mds->mds_virtid_filp = file;
 976         if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
 977                 CERROR("%s is not a regular file!: mode = %o\n",
 978                        VIRT_FID, file->f_dentry->d_inode->i_mode);
 979                 GOTO(err_virtid_fid, rc = -ENOENT);
 980         }
 981
 982         rc = mds_read_virtid_fid(obd, file);
 983         if (rc) {
 984                 CERROR("cannot read %s: rc = %d\n", VIRT_FID, rc);
 985                 GOTO(err_virtid_fid, rc);
 986         }
 987
 988         /* open and test the lov objid file */
 989         file = filp_open(LOV_OBJID, O_RDWR | O_CREAT, 0644);
 990         if (IS_ERR(file)) {
 991                 rc = PTR_ERR(file);
 992                 CERROR("cannot open/create %s file: rc = %d\n", LOV_OBJID, rc);
 993                 GOTO(err_virtid_fid, rc = PTR_ERR(file));
 994         }
 995         mds->mds_dt_objid_filp = file;
 996         if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
 997                 CERROR("%s is not a regular file!: mode = %o\n", LOV_OBJID,
 998                        file->f_dentry->d_inode->i_mode);
 999                 GOTO(err_lov_objid, rc = -ENOENT);
1000         }
1001
1002         /* open and test capa keyid file */
1003         file = filp_open(CAPA_KEYS, O_RDWR | O_CREAT, 0644);
1004         if (IS_ERR(file)) {
1005                 rc = PTR_ERR(file);
1006                 CERROR("cannot open/create %s file: rc = %d\n",
1007                        CAPA_KEYS, rc);
1008                 GOTO(err_lov_objid, rc = PTR_ERR(file));
1009         }
1010         mds->mds_capa_keys_filp = file;
1011         if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1012                 CERROR("%s is not a regular file!: mode = %o\n",
1013                        CAPA_KEYS, file->f_dentry->d_inode->i_mode);
1014                 GOTO(err_capa_keyid, rc = -ENOENT);
1015         }
1016
1017         rc = mds_read_capa_key(obd, file);
1018         if (rc) {
1019                 CERROR("cannot read %s: rc = %d\n", CAPA_KEYS, rc);
1020                 GOTO(err_capa_keyid, rc);
1021         }
1022
1023         /* reint fidext thumb by last fid after root and virt are initialized */
1024         mds->mds_fidext_thumb = mds->mds_last_fid;
1025
1026         rc = mds_fidmap_init(obd, MDS_FIDMAP_SIZE);
1027         if (rc) {
1028                 CERROR("cannot init fid mapping tables, err %d\n", rc);
1029                 GOTO(err_capa_keyid, rc);
1030         }
1031
1032 err_pop:
1033         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1034         return rc;
1035
1036 err_capa_keyid:
1037         if (mds->mds_capa_keys_filp && filp_close(mds->mds_capa_keys_filp, 0))
1038                 CERROR("can't close %s after error\n", CAPA_KEYS);
1039 err_lov_objid:
1040         if (mds->mds_dt_objid_filp && filp_close(mds->mds_dt_objid_filp, 0))
1041                 CERROR("can't close %s after error\n", LOV_OBJID);
1042 err_virtid_fid:
1043         if (mds->mds_virtid_filp && filp_close(mds->mds_virtid_filp, 0))
1044                 CERROR("can't close %s after error\n", VIRT_FID);
1045 err_last_fid:
1046         if (mds->mds_fid_filp && filp_close(mds->mds_fid_filp, 0))
1047                 CERROR("can't close %s after error\n", LAST_FID);
1048 err_client:
1049         class_disconnect_exports(obd, 0);
1050 err_last_rcvd:
1051         if (mds->mds_rcvd_filp && filp_close(mds->mds_rcvd_filp, 0))
1052                 CERROR("can't close %s after error\n", LAST_RCVD);
1053 err_unnamed:
1054         dput(mds->mds_unnamed_dir);
1055 err_id_dir:
1056         dput(mds->mds_id_dir);
1057 err_objects:
1058         dput(mds->mds_objects_dir);
1059 err_logs:
1060         dput(mds->mds_logs_dir);
1061 err_pending:
1062         dput(mds->mds_pending_dir);
1063 err_id_de:
1064         dput(mds->mds_id_de);
1065         goto err_pop;
1066 }
1067
1068 static int  mds_fs_post_cleanup(struct obd_device *obd)
1069 {
1070         int    rc = 0;
1071         rc = fsfilt_post_cleanup(obd);
1072         return rc;
1073 }
1074
1075 int mds_fs_cleanup(struct obd_device *obd, int flags)
1076 {
1077         struct mds_obd *mds = &obd->u.mds;
1078         struct lvfs_run_ctxt saved;
1079         int rc = 0;
1080
1081         if (flags & OBD_OPT_FAILOVER)
1082                 CERROR("%s: shutting down for failover; client state will"
1083                        " be preserved.\n", obd->obd_name);
1084
1085         class_disconnect_exports(obd, flags); /* cleans up client info too */
1086         target_cleanup_recovery(obd);
1087         mds_server_free_data(mds);
1088         mds_fidmap_cleanup(obd);
1089
1090         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1091         if (mds->mds_virtid_filp) {
1092                 rc = filp_close(mds->mds_virtid_filp, 0);
1093                 mds->mds_virtid_filp = NULL;
1094                 if (rc)
1095                         CERROR("%s file won't close, rc = %d\n", VIRT_FID, rc);
1096         }
1097         if (mds->mds_fid_filp) {
1098                 rc = filp_close(mds->mds_fid_filp, 0);
1099                 mds->mds_fid_filp = NULL;
1100                 if (rc)
1101                         CERROR("%s file won't close, rc = %d\n", LAST_FID, rc);
1102         }
1103         if (mds->mds_rcvd_filp) {
1104                 rc = filp_close(mds->mds_rcvd_filp, 0);
1105                 mds->mds_rcvd_filp = NULL;
1106                 if (rc)
1107                         CERROR("%s file won't close, rc = %d\n", LAST_RCVD, rc);
1108         }
1109         if (mds->mds_dt_objid_filp) {
1110                 rc = filp_close(mds->mds_dt_objid_filp, 0);
1111                 mds->mds_dt_objid_filp = NULL;
1112                 if (rc)
1113                         CERROR("%s file won't close, rc=%d\n", LOV_OBJID, rc);
1114         }
1115         if (mds->mds_capa_keys_filp) {
1116                 rc = filp_close(mds->mds_capa_keys_filp, 0);
1117                 mds->mds_capa_keys_filp = NULL;
1118                 if (rc)
1119                         CERROR("%s file won't close, rc=%d\n", CAPA_KEYS, rc);
1120         }
1121         if (mds->mds_unnamed_dir != NULL) {
1122                 l_dput(mds->mds_unnamed_dir);
1123                 mds->mds_unnamed_dir = NULL;
1124         }
1125         if (mds->mds_id_dir != NULL) {
1126                 l_dput(mds->mds_id_dir);
1127                 mds->mds_id_dir = NULL;
1128         }
1129         if (mds->mds_objects_dir != NULL) {
1130                 l_dput(mds->mds_objects_dir);
1131                 mds->mds_objects_dir = NULL;
1132         }
1133         if (mds->mds_logs_dir) {
1134                 l_dput(mds->mds_logs_dir);
1135                 mds->mds_logs_dir = NULL;
1136         }
1137         if (mds->mds_pending_dir) {
1138                 l_dput(mds->mds_pending_dir);
1139                 mds->mds_pending_dir = NULL;
1140         }
1141         rc = mds_fs_post_cleanup(obd);
1142
1143         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1144         shrink_dcache_parent(mds->mds_id_de);
1145         dput(mds->mds_id_de);
1146
1147         return rc;
1148 }
1149
1150 /* Creates an object with the same name as its id.  Because this is not at all
1151  * performance sensitive, it is accomplished by creating a file, checking the
1152  * id, and renaming it. */
1153 int mds_obd_create(struct obd_export *exp, struct obdo *oa,
1154                    void *acl, int acl_size,
1155                    struct lov_stripe_md **ea, struct obd_trans_info *oti)
1156 {
1157         struct mds_obd *mds = &exp->exp_obd->u.mds;
1158         struct inode *parent_inode = mds->mds_objects_dir->d_inode;
1159         struct file *filp;
1160         struct dentry *dchild;
1161         struct lvfs_run_ctxt saved;
1162         char idname[LL_ID_NAMELEN];
1163         int rc = 0, err, idlen;
1164         void *handle;
1165         ENTRY;
1166
1167         push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
1168         down(&parent_inode->i_sem);
1169         if (oa->o_id) {
1170                 idlen = ll_id2str(idname, oa->o_id, oa->o_generation);
1171                 dchild = lookup_one_len(idname, mds->mds_objects_dir, idlen);
1172                 if (IS_ERR(dchild))
1173                         GOTO(out_pop, rc = PTR_ERR(dchild));
1174
1175                 if (dchild->d_inode == NULL) {
1176                         struct dentry_params dp;
1177                         struct inode *inode;
1178
1179                         CWARN("creating log with ID "LPU64"\n", oa->o_id);
1180
1181                         dchild->d_fsdata = (void *) &dp;
1182                         dp.p_ptr = NULL;
1183                         dp.p_inum = oa->o_id;
1184                         rc = ll_vfs_create(parent_inode, dchild, S_IFREG, NULL);
1185                         if (dchild->d_fsdata == (void *)(unsigned long)oa->o_id)
1186                                 dchild->d_fsdata = NULL;
1187                         if (rc) {
1188                                 CDEBUG(D_INODE, "err during create: %d\n", rc);
1189                                 dput(dchild);
1190                                 GOTO(out_pop, rc);
1191                         }
1192                         inode = dchild->d_inode;
1193                         LASSERT(inode->i_ino == oa->o_id);
1194                         inode->i_generation = oa->o_generation;
1195                         CDEBUG(D_HA, "recreated ino %lu with gen %u\n",
1196                                inode->i_ino, inode->i_generation);
1197                         mark_inode_dirty(inode);
1198                 } else {
1199                         CWARN("it should be here!\n");
1200                 }
1201                 GOTO(out_pop, rc);
1202         }
1203
1204         sprintf(idname, "OBJECTS/%u.%u", ll_insecure_random_int(), current->pid);
1205         filp = filp_open(idname, O_CREAT | O_EXCL, 0644);
1206         if (IS_ERR(filp)) {
1207                 rc = PTR_ERR(filp);
1208                 if (rc == -EEXIST) {
1209                         CERROR("impossible object name collision %s\n",
1210                                idname);
1211                         LBUG();
1212                 }
1213                 CERROR("error creating tmp object %s: rc %d\n",
1214                        idname, rc);
1215                 GOTO(out_pop, rc);
1216         }
1217
1218         LASSERT(mds->mds_objects_dir == filp->f_dentry->d_parent);
1219
1220         oa->o_id = filp->f_dentry->d_inode->i_ino;
1221         oa->o_generation = filp->f_dentry->d_inode->i_generation;
1222         idlen = ll_id2str(idname, oa->o_id, oa->o_generation);
1223
1224         CWARN("created log anonymous "LPU64"/%u\n",
1225               oa->o_id, oa->o_generation);
1226
1227         dchild = lookup_one_len(idname, mds->mds_objects_dir, idlen);
1228         if (IS_ERR(dchild)) {
1229                 CERROR("getting neg dentry for obj rename: %d\n", rc);
1230                 GOTO(out_close, rc = PTR_ERR(dchild));
1231         }
1232         if (dchild->d_inode != NULL) {
1233                 CERROR("impossible non-negative obj dentry " LPU64":%u!\n",
1234                        oa->o_id, oa->o_generation);
1235                 LBUG();
1236         }
1237
1238         handle = fsfilt_start(exp->exp_obd, mds->mds_objects_dir->d_inode,
1239                               FSFILT_OP_RENAME, NULL);
1240         if (IS_ERR(handle))
1241                 GOTO(out_dput, rc = PTR_ERR(handle));
1242
1243         lock_kernel();
1244         rc = vfs_rename(mds->mds_objects_dir->d_inode, filp->f_dentry,
1245                         mds->mds_objects_dir->d_inode, dchild);
1246         unlock_kernel();
1247         if (rc)
1248                 CERROR("error renaming new object "LPU64":%u: rc %d\n",
1249                        oa->o_id, oa->o_generation, rc);
1250
1251         err = fsfilt_commit(exp->exp_obd, mds->mds_sb,
1252                             mds->mds_objects_dir->d_inode, handle, 0);
1253         if (!err) {
1254                 oa->o_gr = FILTER_GROUP_FIRST_MDS + mds->mds_num;
1255                 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGENER | OBD_MD_FLGROUP;
1256         } else if (!rc)
1257                 rc = err;
1258 out_dput:
1259         dput(dchild);
1260 out_close:
1261         err = filp_close(filp, 0);
1262         if (err) {
1263                 CERROR("closing tmpfile %s: rc %d\n", idname, rc);
1264                 if (!rc)
1265                         rc = err;
1266         }
1267 out_pop:
1268         up(&parent_inode->i_sem);
1269         pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
1270         RETURN(rc);
1271 }
1272
1273 int mds_obd_destroy(struct obd_export *exp, struct obdo *oa,
1274                     struct lov_stripe_md *ea, struct obd_trans_info *oti)
1275 {
1276         struct mds_obd *mds = &exp->exp_obd->u.mds;
1277         struct inode *parent_inode = mds->mds_objects_dir->d_inode;
1278         struct obd_device *obd = exp->exp_obd;
1279         struct lvfs_run_ctxt saved;
1280         char idname[LL_ID_NAMELEN];
1281         struct dentry *de;
1282         void *handle;
1283         int err, idlen, rc = 0;
1284         ENTRY;
1285
1286         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1287
1288         idlen = ll_id2str(idname, oa->o_id, oa->o_generation);
1289
1290         down(&parent_inode->i_sem);
1291         de = lookup_one_len(idname, mds->mds_objects_dir, idlen);
1292         if (IS_ERR(de) || de->d_inode == NULL) {
1293                 rc = IS_ERR(de) ? PTR_ERR(de) : -ENOENT;
1294                 CERROR("destroying non-existent object "LPU64" %s: rc %d\n",
1295                        oa->o_id, idname, rc);
1296                 GOTO(out_dput, rc);
1297         }
1298         /* Stripe count is 1 here since this is some MDS specific stuff
1299            that is unlinked, not spanned across multiple OSTs */
1300         handle = fsfilt_start_log(obd, mds->mds_objects_dir->d_inode,
1301                                   FSFILT_OP_UNLINK, oti, 1);
1302
1303         if (IS_ERR(handle))
1304                 GOTO(out_dput, rc = PTR_ERR(handle));
1305
1306         rc = vfs_unlink(mds->mds_objects_dir->d_inode, de);
1307         if (rc)
1308                 CERROR("error destroying object "LPU64":%u: rc %d\n",
1309                        oa->o_id, oa->o_generation, rc);
1310
1311         err = fsfilt_commit(obd, mds->mds_sb, mds->mds_objects_dir->d_inode,
1312                             handle, exp->exp_sync);
1313         if (err && !rc)
1314                 rc = err;
1315 out_dput:
1316         if (de != NULL)
1317                 l_dput(de);
1318         up(&parent_inode->i_sem);
1319         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1320         RETURN(rc);
1321 }