lustre/lvfs/llog_lvfs.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  Copyright (C) 2001-2003 Cluster File Systems, Inc.
   5  *   Author: Andreas Dilger <adilger@clusterfs.com>
   6  *
   7  *   This file is part of Lustre, http://www.lustre.org.
   8  *
   9  *   Lustre is free software; you can redistribute it and/or
  10  *   modify it under the terms of version 2 of the GNU General Public
  11  *   License as published by the Free Software Foundation.
  12  *
  13  *   Lustre is distributed in the hope that it will be useful,
  14  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  16  *   GNU General Public License for more details.
  17  *
  18  *   You should have received a copy of the GNU General Public License
  19  *   along with Lustre; if not, write to the Free Software
  20  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  21  *
  22  * OST<->MDS recovery logging infrastructure.
  23  *
  24  * Invariants in implementation:
  25  * - we do not share logs among different OST<->MDS connections, so that
  26  *   if an OST or MDS fails it need only look at log(s) relevant to itself
  27  */
  28
  29 #define DEBUG_SUBSYSTEM S_LOG
  30
  31 #ifndef EXPORT_SYMTAB
  32 #define EXPORT_SYMTAB
  33 #endif
  34
  35 #ifdef __KERNEL__
  36 #include <linux/fs.h>
  37 #else
  38 #include <liblustre.h>
  39 #endif
  40
  41 #include <linux/lvfs.h>
  42 #include <linux/lustre_fsfilt.h>
  43 #include <linux/lustre_log.h>
  44
  45 #ifdef __KERNEL__
  46
  47 static int llog_lvfs_pad(struct llog_ctxt *ctxt, struct l_file *file,
  48                          int len, int index)
  49 {
  50         struct llog_rec_hdr rec;
  51         struct llog_rec_tail tail;
  52         int rc;
  53         ENTRY;
  54
  55         LASSERT(len >= LLOG_MIN_REC_SIZE && (len & 0x7) == 0);
  56
  57         tail.lrt_len = rec.lrh_len = cpu_to_le32(len);
  58         tail.lrt_index = rec.lrh_index = cpu_to_le32(index);
  59         rec.lrh_type = 0;
  60
  61         rc = llog_fsfilt_write_record(ctxt, file, &rec, sizeof(rec),
  62                                       &file->f_pos, 0);
  63         if (rc) {
  64                 CERROR("error writing padding record: rc %d\n", rc);
  65                 goto out;
  66         }
  67
  68         file->f_pos += len - sizeof(rec) - sizeof(tail);
  69         rc = llog_fsfilt_write_record(ctxt, file, &tail, sizeof(tail),
  70                                       &file->f_pos, 0);
  71         if (rc) {
  72                 CERROR("error writing padding record: rc %d\n", rc);
  73                 goto out;
  74         }
  75
  76  out:
  77         RETURN(rc);
  78 }
  79
  80 static int llog_lvfs_write_blob(struct llog_ctxt *ctxt, struct l_file *file,
  81                                 struct llog_rec_hdr *rec, void *buf, loff_t off)
  82 {
  83         int rc;
  84         struct llog_rec_tail end;
  85         loff_t saved_off = file->f_pos;
  86         int buflen = le32_to_cpu(rec->lrh_len);
  87
  88         ENTRY;
  89         file->f_pos = off;
  90
  91         if (!buf) {
  92                 rc = llog_fsfilt_write_record(ctxt, file, rec, buflen,
  93                                               &file->f_pos, 0);
  94                 if (rc) {
  95                         CERROR("error writing log record: rc %d\n", rc);
  96                         goto out;
  97                 }
  98                 GOTO(out, rc = 0);
  99         }
 100
 101         /* the buf case */
 102         rec->lrh_len = cpu_to_le32(sizeof(*rec) + buflen + sizeof(end));
 103         rc = llog_fsfilt_write_record(ctxt, file, rec, sizeof(*rec),
 104                                       &file->f_pos, 0);
 105         if (rc) {
 106                 CERROR("error writing log hdr: rc %d\n", rc);
 107                 goto out;
 108         }
 109
 110         rc = llog_fsfilt_write_record(ctxt, file, buf, buflen,
 111                                       &file->f_pos, 0);
 112         if (rc) {
 113                 CERROR("error writing log buffer: rc %d\n", rc);
 114                 goto out;
 115         }
 116
 117         end.lrt_len = rec->lrh_len;
 118         end.lrt_index = rec->lrh_index;
 119         rc = llog_fsfilt_write_record(ctxt, file, &end, sizeof(end),
 120                                       &file->f_pos, 0);
 121         if (rc) {
 122                 CERROR("error writing log tail: rc %d\n", rc);
 123                 goto out;
 124         }
 125
 126         rc = 0;
 127  out:
 128         if (saved_off > file->f_pos)
 129                 file->f_pos = saved_off;
 130         LASSERT(rc <= 0);
 131         RETURN(rc);
 132 }
 133
 134 static int llog_lvfs_read_blob(struct llog_ctxt *ctxt, struct l_file *file,
 135                                void *buf, int size, loff_t off)
 136 {
 137         loff_t offset = off;
 138         int rc;
 139         ENTRY;
 140
 141         rc = llog_fsfilt_read_record(ctxt, file, buf, size, &offset);
 142         if (rc) {
 143                 CERROR("error reading log record: rc %d\n", rc);
 144                 RETURN(rc);
 145         }
 146         RETURN(0);
 147 }
 148
 149 static int llog_lvfs_read_header(struct llog_handle *handle)
 150 {
 151         struct llog_ctxt *ctxt = handle->lgh_ctxt;
 152         int rc;
 153         ENTRY;
 154
 155         LASSERT(sizeof(*handle->lgh_hdr) == LLOG_CHUNK_SIZE);
 156         LASSERT(ctxt != NULL);
 157
 158         if (handle->lgh_file->f_dentry->d_inode->i_size == 0) {
 159                 CDEBUG(D_HA, "not reading header from 0-byte log\n");
 160                 RETURN(LLOG_EEMPTY);
 161         }
 162
 163         rc = llog_lvfs_read_blob(ctxt, handle->lgh_file, handle->lgh_hdr,
 164                                  LLOG_CHUNK_SIZE, 0);
 165         if (rc)
 166                 CERROR("error reading log header\n");
 167
 168         handle->lgh_last_idx = le32_to_cpu(handle->lgh_hdr->llh_tail.lrt_index);
 169         handle->lgh_file->f_pos = handle->lgh_file->f_dentry->d_inode->i_size;
 170
 171         RETURN(rc);
 172 }
 173
 174 /* returns negative in on error; 0 if success && reccookie == 0; 1 otherwise */
 175 /* appends if idx == -1, otherwise overwrites record idx. */
 176 static int llog_lvfs_write_rec(struct llog_handle *loghandle,
 177                                struct llog_rec_hdr *rec,
 178                                struct llog_cookie *reccookie,
 179                                int cookiecount,
 180                                void *buf, int idx)
 181 {
 182         struct llog_log_hdr *llh;
 183         int reclen = le32_to_cpu(rec->lrh_len), index, rc;
 184         struct llog_rec_tail *lrt;
 185         struct llog_ctxt *ctxt = loghandle->lgh_ctxt;
 186         struct file *file;
 187         loff_t offset;
 188         size_t left;
 189         ENTRY;
 190
 191         llh = loghandle->lgh_hdr;
 192         file = loghandle->lgh_file;
 193
 194         /* record length should not bigger than LLOG_CHUNK_SIZE */
 195         if (buf)
 196                 rc = (reclen > LLOG_CHUNK_SIZE - sizeof(struct llog_rec_hdr)
 197                       - sizeof(struct llog_rec_tail)) ? -E2BIG : 0;
 198         else
 199                 rc = (reclen > LLOG_CHUNK_SIZE) ? -E2BIG : 0;
 200         if (rc)
 201                 RETURN(rc);
 202
 203         if (idx != -1) {
 204                 loff_t saved_offset;
 205
 206                 /* no header: only allowed to insert record 1 */
 207                 if (idx > 1 && !file->f_dentry->d_inode->i_size) {
 208                         CERROR("idx != -1 in empty log\n");
 209                         LBUG();
 210                 }
 211
 212                 if (idx && llh->llh_size && llh->llh_size != reclen)
 213                         RETURN(-EINVAL);
 214
 215                 rc = llog_lvfs_write_blob(ctxt, file, &llh->llh_hdr, NULL, 0);
 216                 /* we are done if we only write the header or on error */
 217                 if (rc || idx == 0)
 218                         RETURN(rc);
 219
 220                 saved_offset = sizeof(*llh) + (idx-1)*le32_to_cpu(rec->lrh_len);
 221                 rc = llog_lvfs_write_blob(ctxt, file, rec, buf, saved_offset);
 222                 if (rc == 0 && reccookie) {
 223                         reccookie->lgc_lgl = loghandle->lgh_id;
 224                         reccookie->lgc_index = idx;
 225                         rc = 1;
 226                 }
 227                 RETURN(rc);
 228         }
 229
 230         /* Make sure that records don't cross a chunk boundary, so we can
 231          * process them page-at-a-time if needed.  If it will cross a chunk
 232          * boundary, write in a fake (but referenced) entry to pad the chunk.
 233          *
 234          * We know that llog_current_log() will return a loghandle that is
 235          * big enough to hold reclen, so all we care about is padding here.
 236          */
 237         left = LLOG_CHUNK_SIZE - (file->f_pos & (LLOG_CHUNK_SIZE - 1));
 238         if (buf)
 239                 reclen = sizeof(*rec) + le32_to_cpu(rec->lrh_len) +
 240                          sizeof(struct llog_rec_tail);
 241
 242         /* NOTE: padding is a record, but no bit is set */
 243         if (left != 0 && left != reclen &&
 244             left < (reclen + LLOG_MIN_REC_SIZE)) {
 245                 loghandle->lgh_last_idx++;
 246                 rc = llog_lvfs_pad(ctxt, file, left, loghandle->lgh_last_idx);
 247                 if (rc)
 248                         RETURN(rc);
 249                 /* if it's the last idx in log file, then return -ENOSPC */
 250                 if (loghandle->lgh_last_idx == LLOG_BITMAP_SIZE(llh) - 1)
 251                         RETURN(-ENOSPC);
 252         }
 253
 254         loghandle->lgh_last_idx++;
 255         index = loghandle->lgh_last_idx;
 256         rec->lrh_index = cpu_to_le32(index);
 257         if (buf == NULL) {
 258                 lrt = (void *)rec + le32_to_cpu(rec->lrh_len) - sizeof(*lrt);
 259                 lrt->lrt_len = rec->lrh_len;
 260                 lrt->lrt_index = rec->lrh_index;
 261         }
 262         if (ext2_set_bit(index, llh->llh_bitmap)) {
 263                 CERROR("argh, index %u already set in log bitmap?\n", index);
 264                 LBUG(); /* should never happen */
 265         }
 266         llh->llh_count = cpu_to_le32(le32_to_cpu(llh->llh_count) + 1);
 267         llh->llh_tail.lrt_index = cpu_to_le32(index);
 268
 269         offset = 0;
 270         rc = llog_lvfs_write_blob(ctxt, file, &llh->llh_hdr, NULL, 0);
 271         if (rc)
 272                 RETURN(rc);
 273
 274         CDEBUG(D_HA, "adding record "LPX64": idx: %u, %u bytes off: %lld\n",
 275                loghandle->lgh_id.lgl_oid, index, le32_to_cpu(rec->lrh_len),
 276                file->f_pos);
 277
 278         rc = llog_lvfs_write_blob(ctxt, file, rec, buf, file->f_pos);
 279         if (rc)
 280                 RETURN(rc);
 281
 282         if (rc == 0 && reccookie) {
 283                 if (llog_cookie_get_flags(reccookie) & LLOG_COOKIE_REPLAY) {
 284                         LASSERTF(EQ_LOGID(reccookie->lgc_lgl,loghandle->lgh_id),
 285                                  "lgc_lgl.oid/gr "LPU64"/"LPU64" lgh_id.oid/gr"
 286                                  LPU64"/"LPU64"\n",
 287                                  reccookie->lgc_lgl.lgl_oid,
 288                                  reccookie->lgc_lgl.lgl_ogr,
 289                                  loghandle->lgh_id.lgl_oid,
 290                                  loghandle->lgh_id.lgl_oid);
 291                         LASSERTF(reccookie->lgc_index == index,
 292                                  "lgc_index %u != index %u\n",
 293                                  reccookie->lgc_index, index);
 294                 } else {
 295                         reccookie->lgc_lgl = loghandle->lgh_id;
 296                         reccookie->lgc_index = index;
 297                         llog_cookie_add_flags(reccookie, LLOG_COOKIE_REPLAY);
 298                 }
 299
 300                 if (le32_to_cpu(rec->lrh_type) == MDS_UNLINK_REC)
 301                         reccookie->lgc_subsys = LLOG_UNLINK_ORIG_CTXT;
 302                 else if (le32_to_cpu(rec->lrh_type) == OST_SZ_REC)
 303                         reccookie->lgc_subsys = LLOG_SIZE_ORIG_CTXT;
 304                 else if (le32_to_cpu(rec->lrh_type) == OST_RAID1_REC)
 305                         reccookie->lgc_subsys = LLOG_RD1_ORIG_CTXT;
 306                 else
 307                         reccookie->lgc_subsys = -1;
 308                 rc = 1;
 309         }
 310         if (rc == 0 && (le32_to_cpu(rec->lrh_type) == LLOG_GEN_REC ||
 311             le32_to_cpu(rec->lrh_type) == SMFS_UPDATE_REC))
 312                 rc = 1;
 313
 314         RETURN(rc);
 315 }
 316
 317 /* We can skip reading at least as many log blocks as the number of
 318 * minimum sized log records we are skipping.  If it turns out
 319 * that we are not far enough along the log (because the
 320 * actual records are larger than minimum size) we just skip
 321 * some more records. */
 322
 323 static void llog_skip_over(__u64 *off, int curr, int goal)
 324 {
 325         if (goal <= curr)
 326                 return;
 327         *off = (*off + (goal-curr-1) * LLOG_MIN_REC_SIZE) &
 328                 ~(LLOG_CHUNK_SIZE - 1);
 329 }
 330
 331 /* sets:
 332  *  - curr_offset to the furthest point read in the log file
 333  *  - curr_idx to the log index preceeding curr_offset
 334  * returns -EIO/-EINVAL on error
 335  */
 336 static int llog_lvfs_next_block(struct llog_handle *loghandle, int *curr_idx,
 337                                 int next_idx, __u64 *curr_offset, void *buf,
 338                                 int len)
 339 {
 340         struct llog_ctxt *ctxt = loghandle->lgh_ctxt;
 341         ENTRY;
 342
 343         if (len == 0 || len & (LLOG_CHUNK_SIZE - 1))
 344                 RETURN(-EINVAL);
 345
 346         CDEBUG(D_OTHER, "looking for log index %u (cur idx %u off "LPU64")\n",
 347                next_idx, *curr_idx, *curr_offset);
 348
 349         while (*curr_offset < loghandle->lgh_file->f_dentry->d_inode->i_size) {
 350                 struct llog_rec_hdr *rec;
 351                 struct llog_rec_tail *tail;
 352                 loff_t ppos;
 353                 int nbytes, rc;
 354
 355                 llog_skip_over(curr_offset, *curr_idx, next_idx);
 356
 357                 ppos = *curr_offset;
 358                 rc = llog_fsfilt_read_record(ctxt, loghandle->lgh_file,
 359                                              buf, len, &ppos);
 360
 361                 if (rc) {
 362                         CERROR("Cant read llog block at log id "LPU64
 363                                "/%u offset "LPU64"\n",
 364                                loghandle->lgh_id.lgl_oid,
 365                                loghandle->lgh_id.lgl_ogen,
 366                                *curr_offset);
 367                         RETURN(rc);
 368                 }
 369
 370                 nbytes = ppos - *curr_offset;
 371                 *curr_offset = ppos;
 372
 373                 if (nbytes == 0) /* end of file, nothing to do */
 374                         RETURN(0);
 375
 376                 if (nbytes < sizeof(*tail)) {
 377                         CERROR("Invalid llog block at log id "LPU64"/%u offset "
 378                                LPU64"\n", loghandle->lgh_id.lgl_oid,
 379                                loghandle->lgh_id.lgl_ogen, *curr_offset);
 380                         RETURN(-EINVAL);
 381                 }
 382
 383                 tail = buf + nbytes - sizeof(struct llog_rec_tail);
 384                 *curr_idx = le32_to_cpu(tail->lrt_index);
 385
 386                 /* this shouldn't happen */
 387                 if (tail->lrt_index == 0) {
 388                         CERROR("Invalid llog tail at log id "LPU64"/%u offset "
 389                                LPU64"\n", loghandle->lgh_id.lgl_oid,
 390                                loghandle->lgh_id.lgl_ogen, *curr_offset);
 391                         RETURN(-EINVAL);
 392                 }
 393                 if (le32_to_cpu(tail->lrt_index) < next_idx)
 394                         continue;
 395
 396                 /* sanity check that the start of the new buffer is no farther
 397                  * than the record that we wanted.  This shouldn't happen. */
 398                 rec = buf;
 399                 if (le32_to_cpu(rec->lrh_index) > next_idx) {
 400                         CERROR("missed desired record? %u > %u\n",
 401                                le32_to_cpu(rec->lrh_index), next_idx);
 402                         RETURN(-ENOENT);
 403                 }
 404                 RETURN(0);
 405         }
 406         RETURN(-EIO);
 407 }
 408
 409 static int llog_lvfs_prev_block(struct llog_handle *loghandle,
 410                                 int prev_idx, void *buf, int len)
 411 {
 412         struct llog_ctxt *ctxt = loghandle->lgh_ctxt;
 413         __u64 curr_offset;
 414         int rc;
 415         ENTRY;
 416
 417         if (len == 0 || len & (LLOG_CHUNK_SIZE - 1))
 418                 RETURN(-EINVAL);
 419
 420         CDEBUG(D_OTHER, "looking for log index %u n", prev_idx);
 421
 422         curr_offset = LLOG_CHUNK_SIZE;
 423         llog_skip_over(&curr_offset, 0, prev_idx);
 424
 425         while (curr_offset < loghandle->lgh_file->f_dentry->d_inode->i_size) {
 426                 struct llog_rec_hdr *rec;
 427                 struct llog_rec_tail *tail;
 428                 loff_t ppos;
 429
 430                 ppos = curr_offset;
 431                 rc = llog_fsfilt_read_record(ctxt, loghandle->lgh_file,
 432                                              buf, len, &ppos);
 433
 434                 if (rc) {
 435                         CERROR("Cant read llog block at log id "LPU64
 436                                "/%u offset "LPU64"\n",
 437                                loghandle->lgh_id.lgl_oid,
 438                                loghandle->lgh_id.lgl_ogen,
 439                                curr_offset);
 440                         RETURN(rc);
 441                 }
 442
 443                 /* put number of bytes read into rc to make code simpler */
 444                 rc = ppos - curr_offset;
 445                 curr_offset = ppos;
 446
 447                 if (rc == 0) /* end of file, nothing to do */
 448                         RETURN(0);
 449
 450                 if (rc < sizeof(*tail)) {
 451                         CERROR("Invalid llog block at log id "LPU64"/%u offset "
 452                                LPU64"\n", loghandle->lgh_id.lgl_oid,
 453                                loghandle->lgh_id.lgl_ogen, curr_offset);
 454                         RETURN(-EINVAL);
 455                 }
 456
 457                 tail = buf + rc - sizeof(struct llog_rec_tail);
 458
 459                 /* this shouldn't happen */
 460                 if (tail->lrt_index == 0) {
 461                         CERROR("Invalid llog tail at log id "LPU64"/%u offset "
 462                                LPU64"\n", loghandle->lgh_id.lgl_oid,
 463                                loghandle->lgh_id.lgl_ogen, curr_offset);
 464                         RETURN(-EINVAL);
 465                 }
 466                 if (le32_to_cpu(tail->lrt_index) < prev_idx)
 467                         continue;
 468
 469                 /* sanity check that the start of the new buffer is no farther
 470                  * than the record that we wanted.  This shouldn't happen. */
 471                 rec = buf;
 472                 if (le32_to_cpu(rec->lrh_index) > prev_idx) {
 473                         CERROR("missed desired record? %u > %u\n",
 474                                le32_to_cpu(rec->lrh_index), prev_idx);
 475                         RETURN(-ENOENT);
 476                 }
 477                 RETURN(0);
 478         }
 479         RETURN(-EIO);
 480 }
 481
 482 static struct file *llog_filp_open(char *name, int flags, int mode)
 483 {
 484         char *logname;
 485         struct file *filp;
 486         int len;
 487
 488         OBD_ALLOC(logname, PATH_MAX);
 489         if (logname == NULL)
 490                 return ERR_PTR(-ENOMEM);
 491
 492         len = snprintf(logname, PATH_MAX, "LOGS/%s", name);
 493         if (len >= PATH_MAX - 1) {
 494                 filp = ERR_PTR(-ENAMETOOLONG);
 495         } else {
 496                 filp = l_filp_open(logname, flags, mode);
 497                 if (IS_ERR(filp)) {
 498                         CERROR("logfile %s(%s): %ld\n",
 499                                flags & O_CREAT ? "create" : "open", logname,
 500                                PTR_ERR(filp));
 501                 }
 502         }
 503
 504         OBD_FREE(logname, PATH_MAX);
 505         return filp;
 506 }
 507
 508 /* creates object for the case when we have no obd (smfs). */
 509 static struct file *
 510 llog_object_create_alone(struct llog_ctxt *ctxt, struct llog_logid *lgh_id)
 511 {
 512         struct file *filp;
 513         int rc = 0;
 514         ENTRY;
 515
 516         LASSERT(lgh_id != NULL);
 517         if (lgh_id->lgl_oid) {
 518                 struct dentry *dchild;
 519                 char fidname[LL_FID_NAMELEN];
 520                 int fidlen = 0;
 521
 522                 down(&ctxt->loc_objects_dir->d_inode->i_sem);
 523                 fidlen = ll_fid2str(fidname, lgh_id->lgl_oid, lgh_id->lgl_ogen);
 524                 dchild = lookup_one_len(fidname, ctxt->loc_objects_dir, fidlen);
 525                 if (IS_ERR(dchild)) {
 526                         up(&ctxt->loc_objects_dir->d_inode->i_sem);
 527                         RETURN((struct file *)dchild);
 528                 }
 529                 if (dchild->d_inode == NULL) {
 530                         struct dentry_params dp;
 531                         struct inode *inode;
 532
 533                         dchild->d_fsdata = (void *) &dp;
 534                         dp.p_ptr = NULL;
 535                         dp.p_inum = lgh_id->lgl_oid;
 536                         rc = ll_vfs_create(ctxt->loc_objects_dir->d_inode,
 537                                            dchild, S_IFREG, NULL);
 538                         if (dchild->d_fsdata == (void *)(unsigned long)lgh_id->lgl_oid)
 539                                 dchild->d_fsdata = NULL;
 540                         if (rc) {
 541                                 CDEBUG(D_INODE, "err during create: %d\n", rc);
 542                                 dput(dchild);
 543                                 up(&ctxt->loc_objects_dir->d_inode->i_sem);
 544                                 RETURN(ERR_PTR(rc));
 545                         }
 546                         inode = dchild->d_inode;
 547                         LASSERT(inode->i_ino == lgh_id->lgl_oid);
 548                         inode->i_generation = lgh_id->lgl_ogen;
 549                         CDEBUG(D_HA, "recreated ino %lu with gen %u\n",
 550                                inode->i_ino, inode->i_generation);
 551                         mark_inode_dirty(inode);
 552                 }
 553
 554                 mntget(ctxt->loc_lvfs_ctxt->pwdmnt);
 555                 filp = dentry_open(dchild, ctxt->loc_lvfs_ctxt->pwdmnt,
 556                                     O_RDWR | O_LARGEFILE);
 557                 if (IS_ERR(filp)) {
 558                         dput(dchild);
 559                         up(&ctxt->loc_objects_dir->d_inode->i_sem);
 560                         RETURN(filp);
 561                 }
 562                 if (!S_ISREG(filp->f_dentry->d_inode->i_mode)) {
 563                         CERROR("%s is not a regular file!: mode = %o\n", fidname,
 564                                filp->f_dentry->d_inode->i_mode);
 565                         filp_close(filp, 0);
 566                         up(&ctxt->loc_objects_dir->d_inode->i_sem);
 567                         RETURN(ERR_PTR(-ENOENT));
 568                 }
 569
 570                 up(&ctxt->loc_objects_dir->d_inode->i_sem);
 571                 RETURN(filp);
 572
 573         } else {
 574                 unsigned int tmpname = ll_insecure_random_int();
 575                 char fidname[LL_FID_NAMELEN];
 576                 struct dentry *new_child, *parent;
 577                 void *handle;
 578                 int err, namelen;
 579
 580                 sprintf(fidname, "OBJECTS/%u", tmpname);
 581                 filp = filp_open(fidname, O_CREAT | O_EXCL, 0644);
 582                 if (IS_ERR(filp)) {
 583                         rc = PTR_ERR(filp);
 584                         if (rc == -EEXIST) {
 585                                 CERROR("impossible object name collision %u\n",
 586                                         tmpname);
 587                                 LBUG();
 588                         }
 589                         CERROR("error creating tmp object %u: rc %d\n", tmpname, rc);
 590                         RETURN(filp);
 591                 }
 592
 593                 namelen = ll_fid2str(fidname, filp->f_dentry->d_inode->i_ino,
 594                                      filp->f_dentry->d_inode->i_generation);
 595                 parent = filp->f_dentry->d_parent;
 596                 down(&parent->d_inode->i_sem);
 597                 new_child = lookup_one_len(fidname, parent, namelen);
 598                 if (IS_ERR(new_child)) {
 599                         CERROR("getting neg dentry for obj rename: %d\n", rc);
 600                         GOTO(out_close, rc = PTR_ERR(new_child));
 601                 }
 602                 if (new_child->d_inode != NULL) {
 603                         CERROR("impossible non-negative obj dentry %lu:%u!\n",
 604                                 filp->f_dentry->d_inode->i_ino,
 605                                 filp->f_dentry->d_inode->i_generation);
 606                         LBUG();
 607                 }
 608
 609                 handle = llog_fsfilt_start(ctxt, parent->d_inode, FSFILT_OP_RENAME, NULL);
 610                 if (IS_ERR(handle))
 611                         GOTO(out_dput, rc = PTR_ERR(handle));
 612
 613                 lock_kernel();
 614                 rc = vfs_rename(parent->d_inode, filp->f_dentry,
 615                                 parent->d_inode, new_child);
 616                 unlock_kernel();
 617                 if (rc)
 618                         CERROR("error renaming new object %lu:%u: rc %d\n",
 619                                 filp->f_dentry->d_inode->i_ino,
 620                                 filp->f_dentry->d_inode->i_generation, rc);
 621
 622                 err = llog_fsfilt_commit(ctxt, parent->d_inode, handle, 0);
 623                 if (!rc)
 624                         rc = err;
 625
 626         out_dput:
 627                 dput(new_child);
 628         out_close:
 629                 up(&parent->d_inode->i_sem);
 630                 if (rc) {
 631                         filp_close(filp, 0);
 632                         filp = ERR_PTR(rc);
 633                 } else {
 634                         /* FIXME: is this group 1 is correct? */
 635                         lgh_id->lgl_ogr = 1;
 636                         lgh_id->lgl_oid = filp->f_dentry->d_inode->i_ino;
 637                         lgh_id->lgl_ogen = filp->f_dentry->d_inode->i_generation;
 638                 }
 639                 RETURN(filp);
 640         }
 641 }
 642
 643 /* creates object for generic case (obd exists) */
 644 static struct file *
 645 llog_object_create_generic(struct llog_ctxt *ctxt, struct llog_logid *lgh_id)
 646 {
 647         struct file *filp = NULL;
 648         struct dentry *dchild;
 649         struct obd_device *obd;
 650         struct obdo *oa = NULL;
 651         int open_flags = O_RDWR | O_LARGEFILE;
 652         int rc = 0;
 653         ENTRY;
 654
 655         obd = ctxt->loc_exp->exp_obd;
 656         LASSERT(obd != NULL);
 657
 658         if (lgh_id->lgl_oid) {
 659                 dchild = obd_lvfs_fid2dentry(ctxt->loc_exp, lgh_id->lgl_oid,
 660                                              lgh_id->lgl_ogen, lgh_id->lgl_ogr);
 661                 if (IS_ERR(dchild) == -ENOENT) {
 662                         OBD_ALLOC(oa, sizeof(*oa));
 663                         if (!oa)
 664                                 RETURN(ERR_PTR(-ENOMEM));
 665
 666                         oa->o_id = lgh_id->lgl_oid;
 667                         oa->o_generation = lgh_id->lgl_ogen;
 668                         oa->o_gr = lgh_id->lgl_ogr;
 669                         oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLGROUP;
 670                         rc = obd_create(ctxt->loc_exp, oa, NULL, NULL);
 671                         if (rc) {
 672                                 CDEBUG(D_INODE, "err during create: %d\n", rc);
 673                                 GOTO(out_free_oa, rc);
 674                         }
 675                         CDEBUG(D_HA, "re-create log object "LPX64":0x%x:"LPX64"\n",
 676                                lgh_id->lgl_oid, lgh_id->lgl_ogen, lgh_id->lgl_ogr);
 677
 678                         dchild = obd_lvfs_fid2dentry(ctxt->loc_exp, lgh_id->lgl_oid,
 679                                                      lgh_id->lgl_ogen, lgh_id->lgl_ogr);
 680                 } else if (IS_ERR(dchild)) {
 681                         CERROR("error looking up logfile "LPX64":0x%x: rc %d\n",
 682                                lgh_id->lgl_oid, lgh_id->lgl_ogen, rc);
 683                         RETURN((struct file *)dchild);
 684                 }
 685
 686                 filp = l_dentry_open(&obd->obd_lvfs_ctxt, dchild, open_flags);
 687                 if (IS_ERR(filp)) {
 688                         l_dput(dchild);
 689                         rc = PTR_ERR(filp);
 690                         CERROR("error opening logfile "LPX64"0x%x: rc %d\n",
 691                                lgh_id->lgl_oid, lgh_id->lgl_ogen, rc);
 692                 }
 693                 GOTO(out_free_oa, rc);
 694         } else {
 695                 /* this is important to work here over obd_create() as it manages
 696                   groups and we need it. Yet another reason is that mds_obd_create()
 697                  is fully the same as old version of this function and this helps
 698                  us to avoid code duplicating and layering violating. */
 699                 OBD_ALLOC(oa, sizeof(*oa));
 700                 if (!oa)
 701                         RETURN(ERR_PTR(-ENOMEM));
 702
 703                 oa->o_gr = FILTER_GROUP_LLOG;
 704                 oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLGROUP;
 705                 rc = obd_create(ctxt->loc_exp, oa, NULL, NULL);
 706                 if (rc)
 707                         GOTO(out_free_oa, rc);
 708
 709                 dchild = obd_lvfs_fid2dentry(ctxt->loc_exp, oa->o_id,
 710                                              oa->o_generation, oa->o_gr);
 711                 if (IS_ERR(dchild))
 712                         GOTO(out_free_oa, rc = PTR_ERR(dchild));
 713
 714                 filp = l_dentry_open(&obd->obd_lvfs_ctxt, dchild,
 715                                      open_flags);
 716                 if (IS_ERR(filp)) {
 717                         l_dput(dchild);
 718                         GOTO(out_free_oa, rc = PTR_ERR(filp));
 719                 }
 720
 721                 /* group 1 is not longer valid, we use the group which is set
 722                 by obd_create()->mds_obd_create(). */
 723                 lgh_id->lgl_ogr = oa->o_gr;
 724                 lgh_id->lgl_oid = oa->o_id;
 725                 lgh_id->lgl_ogen = oa->o_generation;
 726         }
 727
 728 out_free_oa:
 729         if (rc)
 730                 filp = ERR_PTR(rc);
 731         if (oa)
 732                 OBD_FREE(oa, sizeof(*oa));
 733         RETURN(filp);
 734 }
 735
 736 static struct file *
 737 llog_object_create(struct llog_ctxt *ctxt, struct llog_logid *lgh_id)
 738 {
 739         if (ctxt->loc_alone)
 740                 return llog_object_create_alone(ctxt, lgh_id);
 741         else
 742                 return llog_object_create_generic(ctxt, lgh_id);
 743 }
 744
 745 static int llog_add_link_object(struct llog_ctxt *ctxt, struct llog_logid logid,
 746                                 struct dentry *dentry)
 747 {
 748         struct dentry *new_child;
 749         char fidname[LL_FID_NAMELEN];
 750         void *handle;
 751         int namelen, rc = 0, err;
 752         ENTRY;
 753
 754         namelen = ll_fid2str(fidname, logid.lgl_oid, logid.lgl_ogen);
 755         down(&ctxt->loc_objects_dir->d_inode->i_sem);
 756         new_child = lookup_one_len(fidname, ctxt->loc_objects_dir, namelen);
 757         if (IS_ERR(new_child)) {
 758                 CERROR("getting neg dentry for obj rename: %d\n", rc);
 759                 GOTO(out, rc = PTR_ERR(new_child));
 760         }
 761         if (new_child->d_inode == dentry->d_inode)
 762                 GOTO(out_dput, rc);
 763         if (new_child->d_inode != NULL) {
 764                 CERROR("impossible non-negative obj dentry "LPX64":%u!\n",
 765                        logid.lgl_oid, logid.lgl_ogen);
 766                 LBUG();
 767         }
 768         handle = llog_fsfilt_start(ctxt, ctxt->loc_objects_dir->d_inode,
 769                                    FSFILT_OP_LINK, NULL);
 770         if (IS_ERR(handle))
 771                 GOTO(out_dput, rc = PTR_ERR(handle));
 772
 773         lock_kernel();
 774         rc = vfs_link(dentry, ctxt->loc_objects_dir->d_inode, new_child);
 775         unlock_kernel();
 776         if (rc)
 777                 CERROR("error link new object "LPX64":%u: rc %d\n",
 778                        logid.lgl_oid, logid.lgl_ogen, rc);
 779         err = llog_fsfilt_commit(ctxt, ctxt->loc_objects_dir->d_inode, handle, 0);
 780 out_dput:
 781         l_dput(new_child);
 782 out:
 783         up(&ctxt->loc_objects_dir->d_inode->i_sem);
 784         RETURN(rc);
 785 }
 786
 787 static int llog_lvfs_open(struct llog_ctxt *ctxt, struct llog_handle **res,
 788                           struct llog_logid *logid, char *name, int flags)
 789 {
 790         struct llog_handle *handle;
 791         struct lvfs_run_ctxt saved;
 792         int rc = 0;
 793         int open_flags = O_RDWR | O_LARGEFILE;
 794         ENTRY;
 795
 796         if (flags & OBD_LLOG_FL_CREATE)
 797                 open_flags |= O_CREAT;
 798
 799         handle = llog_alloc_handle();
 800         if (handle == NULL)
 801                 RETURN(-ENOMEM);
 802         *res = handle;
 803
 804         LASSERT(ctxt);
 805         if (ctxt->loc_lvfs_ctxt)
 806                 push_ctxt(&saved, ctxt->loc_lvfs_ctxt, NULL);
 807
 808         if (logid != NULL) {
 809                 handle->lgh_file = llog_object_create(ctxt, logid);
 810                 if (IS_ERR(handle->lgh_file)) {
 811                         CERROR("cannot create/open llog object "LPX64":%x "
 812                                "error = %ld", logid->lgl_oid, logid->lgl_ogen,
 813                                PTR_ERR(handle->lgh_file));
 814                         GOTO(cleanup, rc = PTR_ERR(handle->lgh_file));
 815                 }
 816                 handle->lgh_id = *logid;
 817
 818         } else if (name) {
 819                 handle->lgh_file = llog_filp_open(name, open_flags, 0644);
 820                 if (IS_ERR(handle->lgh_file)) {
 821                         CERROR("cannot open %s file, error = %ld\n",
 822                                name, PTR_ERR(handle->lgh_file));
 823                         GOTO(cleanup, rc = PTR_ERR(handle->lgh_file));
 824                 }
 825                 LASSERT(handle->lgh_file->f_dentry->d_parent == ctxt->loc_logs_dir);
 826
 827                 handle->lgh_id.lgl_ogr = 1;
 828                 handle->lgh_id.lgl_oid = handle->lgh_file->f_dentry->d_inode->i_ino;
 829                 handle->lgh_id.lgl_ogen = handle->lgh_file->f_dentry->d_inode->i_generation;
 830                 rc = llog_add_link_object(ctxt, handle->lgh_id, handle->lgh_file->f_dentry);
 831                 if (rc)
 832                         GOTO(cleanup, rc);
 833
 834         } else {
 835                 handle->lgh_file = llog_object_create(ctxt, &handle->lgh_id);
 836                 if (IS_ERR(handle->lgh_file)) {
 837                         CERROR("cannot create llog object, error = %ld\n",
 838                                PTR_ERR(handle->lgh_file));
 839                         GOTO(cleanup, rc = PTR_ERR(handle->lgh_file));
 840                 }
 841         }
 842
 843         handle->lgh_ctxt = ctxt;
 844 finish:
 845         if (ctxt->loc_lvfs_ctxt)
 846                 pop_ctxt(&saved, ctxt->loc_lvfs_ctxt, NULL);
 847         RETURN(rc);
 848 cleanup:
 849         llog_free_handle(handle);
 850         goto finish;
 851 }
 852
 853 static int llog_lvfs_close(struct llog_handle *handle)
 854 {
 855         int rc;
 856         ENTRY;
 857
 858         rc = filp_close(handle->lgh_file, 0);
 859         if (rc)
 860                 CERROR("error closing log: rc %d\n", rc);
 861         RETURN(rc);
 862 }
 863
 864 static int llog_lvfs_destroy(struct llog_handle *loghandle)
 865 {
 866         struct llog_ctxt *ctxt = loghandle->lgh_ctxt;
 867         struct lvfs_run_ctxt saved;
 868         struct dentry *fdentry;
 869         struct inode *parent_inode;
 870         char fidname[LL_FID_NAMELEN];
 871         void *handle;
 872         int rc = -EINVAL, err, namelen;
 873         ENTRY;
 874
 875         if (ctxt->loc_lvfs_ctxt)
 876                 push_ctxt(&saved, ctxt->loc_lvfs_ctxt, NULL);
 877
 878         fdentry = loghandle->lgh_file->f_dentry;
 879         parent_inode = fdentry->d_parent->d_inode;
 880
 881         if (!strcmp(fdentry->d_parent->d_name.name, "LOGS")) {
 882                 LASSERT(parent_inode == ctxt->loc_logs_dir->d_inode);
 883
 884                 namelen = ll_fid2str(fidname, fdentry->d_inode->i_ino,
 885                                      fdentry->d_inode->i_generation);
 886                 dget(fdentry);
 887                 rc = llog_lvfs_close(loghandle);
 888                 if (rc) {
 889                         dput(fdentry);
 890                         GOTO(out, rc);
 891                 }
 892
 893                 handle = llog_fsfilt_start(ctxt, parent_inode,
 894                                            FSFILT_OP_UNLINK, NULL);
 895                 if (IS_ERR(handle)) {
 896                         dput(fdentry);
 897                         GOTO(out, rc = PTR_ERR(handle));
 898                 }
 899
 900                 down(&parent_inode->i_sem);
 901                 rc = vfs_unlink(parent_inode, fdentry);
 902                 up(&parent_inode->i_sem);
 903                 dput(fdentry);
 904
 905                 if (!rc) {
 906                         down(&ctxt->loc_objects_dir->d_inode->i_sem);
 907                         fdentry = lookup_one_len(fidname, ctxt->loc_objects_dir,
 908                                                  namelen);
 909                         if (fdentry == NULL || fdentry->d_inode == NULL) {
 910                                 CERROR("destroy non_existent object %s\n", fidname);
 911                                 GOTO(out_err, rc = IS_ERR(fdentry) ?
 912                                      PTR_ERR(fdentry) : -ENOENT);
 913                         }
 914                         rc = vfs_unlink(ctxt->loc_objects_dir->d_inode, fdentry);
 915                         l_dput(fdentry);
 916 out_err:
 917                         up(&ctxt->loc_objects_dir->d_inode->i_sem);
 918                 }
 919                 err = llog_fsfilt_commit(ctxt, parent_inode, handle, 0);
 920                 if (err && !rc)
 921                         err = rc;
 922
 923                 GOTO(out, rc);
 924         }
 925         if (ctxt->loc_alone) {
 926                 if (!strcmp(fdentry->d_parent->d_name.name, "OBJECTS")) {
 927                         LASSERT(parent_inode == ctxt->loc_objects_dir->d_inode);
 928
 929                         dget(fdentry);
 930                         rc = llog_lvfs_close(loghandle);
 931                         if (rc == 0) {
 932                                 down(&parent_inode->i_sem);
 933                                 rc = vfs_unlink(parent_inode, fdentry);
 934                                 up(&parent_inode->i_sem);
 935                         }
 936                         dput(fdentry);
 937                 }
 938         } else {
 939                 struct obdo *oa = NULL;
 940
 941                 OBD_ALLOC(oa, sizeof(*oa));
 942                 if (!oa)
 943                         GOTO(out, rc = -ENOMEM);
 944
 945                 oa->o_id = loghandle->lgh_id.lgl_oid;
 946                 oa->o_gr = loghandle->lgh_id.lgl_ogr;
 947                 oa->o_generation = loghandle->lgh_id.lgl_ogen;
 948                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP | OBD_MD_FLGENER;
 949
 950                 rc = llog_lvfs_close(loghandle);
 951                 if (rc)
 952                         GOTO(out_free_oa, rc);
 953
 954                 rc = obd_destroy(loghandle->lgh_ctxt->loc_exp, oa, NULL, NULL);
 955 out_free_oa:
 956                 OBD_FREE(oa, sizeof(*oa));
 957         }
 958 out:
 959         if (ctxt->loc_lvfs_ctxt)
 960                 pop_ctxt(&saved, ctxt->loc_lvfs_ctxt, NULL);
 961         RETURN(rc);
 962 }
 963
 964 /* reads the catalog list */
 965 int llog_get_cat_list(struct lvfs_run_ctxt *ctxt,
 966                       struct fsfilt_operations *fsops, char *name,
 967                       int count, struct llog_catid *idarray)
 968 {
 969         struct lvfs_run_ctxt saved;
 970         struct l_file *file;
 971         int size = sizeof(*idarray) * count;
 972         loff_t off = 0;
 973         int rc;
 974
 975         LASSERT(count);
 976
 977         if (ctxt)
 978                 push_ctxt(&saved, ctxt, NULL);
 979         file = l_filp_open(name, O_RDWR | O_CREAT | O_LARGEFILE, 0700);
 980         if (!file || IS_ERR(file)) {
 981                 rc = PTR_ERR(file);
 982                 CERROR("OBD filter: cannot open/create %s: rc = %d\n",
 983                        name, rc);
 984                 GOTO(out, rc);
 985         }
 986
 987         if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
 988                 CERROR("%s is not a regular file!: mode = %o\n", name,
 989                        file->f_dentry->d_inode->i_mode);
 990                 GOTO(out, rc = -ENOENT);
 991         }
 992
 993         rc = fsops->fs_read_record(file, idarray, size, &off);
 994         if (rc) {
 995                 CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
 996                        name, rc);
 997                 GOTO(out, rc);
 998         }
 999
1000  out:
1001         if (file && !IS_ERR(file))
1002                 rc = filp_close(file, 0);
1003         if (ctxt)
1004                 pop_ctxt(&saved, ctxt, NULL);
1005         RETURN(rc);
1006 }
1007 EXPORT_SYMBOL(llog_get_cat_list);
1008
1009 /* writes the cat list */
1010 int llog_put_cat_list(struct lvfs_run_ctxt *ctxt,
1011                       struct fsfilt_operations *fsops, char *name,
1012                       int count, struct llog_catid *idarray)
1013 {
1014         struct lvfs_run_ctxt saved;
1015         struct l_file *file;
1016         int size = sizeof(*idarray) * count;
1017         loff_t off = 0;
1018         int rc;
1019
1020         LASSERT(count);
1021
1022         if (ctxt)
1023                 push_ctxt(&saved, ctxt, NULL);
1024         file = filp_open(name, O_RDWR | O_CREAT | O_LARGEFILE, 0700);
1025         if (!file || IS_ERR(file)) {
1026                 rc = PTR_ERR(file);
1027                 CERROR("OBD filter: cannot open/create %s: rc = %d\n",
1028                        name, rc);
1029                 GOTO(out, rc);
1030         }
1031
1032         if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1033                 CERROR("%s is not a regular file!: mode = %o\n", name,
1034                        file->f_dentry->d_inode->i_mode);
1035                 GOTO(out, rc = -ENOENT);
1036         }
1037
1038         rc = fsops->fs_write_record(file, idarray, size, &off, 1);
1039         if (rc) {
1040                 CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
1041                        name, rc);
1042                 GOTO(out, rc);
1043         }
1044
1045  out:
1046         if (file && !IS_ERR(file))
1047                 rc = filp_close(file, 0);
1048         if (ctxt)
1049                 pop_ctxt(&saved, ctxt, NULL);
1050         RETURN(rc);
1051 }
1052 EXPORT_SYMBOL(llog_put_cat_list);
1053
1054 struct llog_operations llog_lvfs_ops = {
1055         lop_open:        llog_lvfs_open,
1056         lop_destroy:     llog_lvfs_destroy,
1057         lop_close:       llog_lvfs_close,
1058         lop_read_header: llog_lvfs_read_header,
1059         lop_write_rec:   llog_lvfs_write_rec,
1060         lop_next_block:  llog_lvfs_next_block,
1061         lop_prev_block:  llog_lvfs_prev_block,
1062 };
1063 EXPORT_SYMBOL(llog_lvfs_ops);
1064
1065 #else /* !__KERNEL__ */
1066
1067 static int llog_lvfs_read_header(struct llog_handle *handle)
1068 {
1069         LBUG();
1070         return 0;
1071 }
1072
1073 static int llog_lvfs_write_rec(struct llog_handle *loghandle,
1074                                struct llog_rec_hdr *rec,
1075                                struct llog_cookie *reccookie, int cookiecount,
1076                                void *buf, int idx)
1077 {
1078         LBUG();
1079         return 0;
1080 }
1081
1082 static int llog_lvfs_open(struct llog_ctxt *ctxt, struct llog_handle **res,
1083                           struct llog_logid *logid, char *name, int flags)
1084 {
1085         LBUG();
1086         return 0;
1087 }
1088
1089 static int llog_lvfs_close(struct llog_handle *handle)
1090 {
1091         LBUG();
1092         return 0;
1093 }
1094
1095 static int llog_lvfs_destroy(struct llog_handle *handle)
1096 {
1097         LBUG();
1098         return 0;
1099 }
1100
1101 int llog_get_cat_list(struct lvfs_run_ctxt *ctxt,
1102                       struct fsfilt_operations *fsops, char *name,
1103                       int count, struct llog_catid *idarray)
1104 {
1105         LBUG();
1106         return 0;
1107 }
1108
1109 int llog_put_cat_list(struct lvfs_run_ctxt *ctxt,
1110                       struct fsfilt_operations *fsops, char *name,
1111                       int count, struct llog_catid *idarray)
1112 {
1113         LBUG();
1114         return 0;
1115 }
1116
1117 int llog_lvfs_prev_block(struct llog_handle *loghandle,
1118                          int prev_idx, void *buf, int len)
1119 {
1120         LBUG();
1121         return 0;
1122 }
1123
1124 int llog_lvfs_next_block(struct llog_handle *loghandle, int *curr_idx,
1125                          int next_idx, __u64 *offset, void *buf, int len)
1126 {
1127         LBUG();
1128         return 0;
1129 }
1130
1131 struct llog_operations llog_lvfs_ops = {
1132         lop_open:        llog_lvfs_open,
1133         lop_destroy:     llog_lvfs_destroy,
1134         lop_close:       llog_lvfs_close,
1135         lop_read_header: llog_lvfs_read_header,
1136         lop_write_rec:   llog_lvfs_write_rec,
1137         lop_next_block:  llog_lvfs_next_block,
1138         lop_prev_block:  llog_lvfs_prev_block,
1139 };
1140 #endif