lustre/obdclass/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 the Lustre file system, http://www.lustre.org
   8  *   Lustre is a trademark of Cluster File Systems, Inc.
   9  *
  10  *   You may have signed or agreed to another license before downloading
  11  *   this software.  If so, you are bound by the terms and conditions
  12  *   of that agreement, and the following does not apply to you.  See the
  13  *   LICENSE file included with this distribution for more information.
  14  *
  15  *   If you did not agree to a different license, then this copy of Lustre
  16  *   is open source software; you can redistribute it and/or modify it
  17  *   under the terms of version 2 of the GNU General Public License as
  18  *   published by the Free Software Foundation.
  19  *
  20  *   In either case, Lustre is distributed in the hope that it will be
  21  *   useful, but WITHOUT ANY WARRANTY; without even the implied warranty
  22  *   of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  23  *   license text for more details.
  24  *
  25  * OST<->MDS recovery logging infrastructure.
  26  *
  27  * Invariants in implementation:
  28  * - we do not share logs among different OST<->MDS connections, so that
  29  *   if an OST or MDS fails it need only look at log(s) relevant to itself
  30  */
  31
  32 #define DEBUG_SUBSYSTEM S_LOG
  33
  34 #ifndef EXPORT_SYMTAB
  35 #define EXPORT_SYMTAB
  36 #endif
  37
  38 #ifndef __KERNEL__
  39 #include <liblustre.h>
  40 #endif
  41
  42 #include <obd.h>
  43 #include <obd_class.h>
  44 #include <lustre_log.h>
  45 #include <obd_ost.h>
  46 #include <lustre_mds.h>
  47 #include <libcfs/list.h>
  48 #include <lvfs.h>
  49 #include <lustre_fsfilt.h>
  50 #include <lustre_disk.h>
  51 #include "llog_internal.h"
  52
  53 #if defined(__KERNEL__) && defined(LLOG_LVFS)
  54
  55 static int llog_lvfs_pad(struct obd_device *obd, struct l_file *file,
  56                                 int len, int index)
  57 {
  58         struct llog_rec_hdr rec = { 0 };
  59         struct llog_rec_tail tail;
  60         int rc;
  61         ENTRY;
  62
  63         LASSERT(len >= LLOG_MIN_REC_SIZE && (len & 0x7) == 0);
  64
  65         tail.lrt_len = rec.lrh_len = len;
  66         tail.lrt_index = rec.lrh_index = index;
  67         rec.lrh_type = LLOG_PAD_MAGIC;
  68
  69         rc = fsfilt_write_record(obd, file, &rec, sizeof(rec), &file->f_pos, 0);
  70         if (rc) {
  71                 CERROR("error writing padding record: rc %d\n", rc);
  72                 goto out;
  73         }
  74
  75         file->f_pos += len - sizeof(rec) - sizeof(tail);
  76         rc = fsfilt_write_record(obd, file, &tail, sizeof(tail),&file->f_pos,0);
  77         if (rc) {
  78                 CERROR("error writing padding record: rc %d\n", rc);
  79                 goto out;
  80         }
  81
  82  out:
  83         RETURN(rc);
  84 }
  85
  86 static int llog_lvfs_write_blob(struct obd_device *obd, struct l_file *file,
  87                                 struct llog_rec_hdr *rec, void *buf, loff_t off)
  88 {
  89         int rc;
  90         struct llog_rec_tail end;
  91         loff_t saved_off = file->f_pos;
  92         int buflen = rec->lrh_len;
  93
  94         ENTRY;
  95         file->f_pos = off;
  96
  97         if (!buf) {
  98                 rc = fsfilt_write_record(obd, file, rec, buflen,&file->f_pos,0);
  99                 if (rc) {
 100                         CERROR("error writing log record: rc %d\n", rc);
 101                         goto out;
 102                 }
 103                 GOTO(out, rc = 0);
 104         }
 105
 106         /* the buf case */
 107         rec->lrh_len = sizeof(*rec) + buflen + sizeof(end);
 108         rc = fsfilt_write_record(obd, file, rec, sizeof(*rec), &file->f_pos, 0);
 109         if (rc) {
 110                 CERROR("error writing log hdr: rc %d\n", rc);
 111                 goto out;
 112         }
 113
 114         rc = fsfilt_write_record(obd, file, buf, buflen, &file->f_pos, 0);
 115         if (rc) {
 116                 CERROR("error writing log buffer: rc %d\n", rc);
 117                 goto out;
 118         }
 119
 120         end.lrt_len = rec->lrh_len;
 121         end.lrt_index = rec->lrh_index;
 122         rc = fsfilt_write_record(obd, file, &end, sizeof(end), &file->f_pos, 0);
 123         if (rc) {
 124                 CERROR("error writing log tail: rc %d\n", rc);
 125                 goto out;
 126         }
 127
 128         rc = 0;
 129  out:
 130         if (saved_off > file->f_pos)
 131                 file->f_pos = saved_off;
 132         LASSERT(rc <= 0);
 133         RETURN(rc);
 134 }
 135
 136 static int llog_lvfs_read_blob(struct obd_device *obd, struct l_file *file,
 137                                 void *buf, int size, loff_t off)
 138 {
 139         loff_t offset = off;
 140         int rc;
 141         ENTRY;
 142
 143         rc = fsfilt_read_record(obd, file, buf, size, &offset);
 144         if (rc) {
 145                 CERROR("error reading log record: rc %d\n", rc);
 146                 RETURN(rc);
 147         }
 148         RETURN(0);
 149 }
 150
 151 static int llog_lvfs_read_header(struct llog_handle *handle)
 152 {
 153         struct obd_device *obd;
 154         int rc;
 155         ENTRY;
 156
 157         LASSERT(sizeof(*handle->lgh_hdr) == LLOG_CHUNK_SIZE);
 158
 159         obd = handle->lgh_ctxt->loc_exp->exp_obd;
 160
 161         if (handle->lgh_file->f_dentry->d_inode->i_size == 0) {
 162                 CDEBUG(D_HA, "not reading header from 0-byte log\n");
 163                 RETURN(LLOG_EEMPTY);
 164         }
 165
 166         rc = llog_lvfs_read_blob(obd, handle->lgh_file, handle->lgh_hdr,
 167                                  LLOG_CHUNK_SIZE, 0);
 168         if (rc) {
 169                 CERROR("error reading log header from %.*s\n",
 170                        handle->lgh_file->f_dentry->d_name.len,
 171                        handle->lgh_file->f_dentry->d_name.name);
 172         } else {
 173                 struct llog_rec_hdr *llh_hdr = &handle->lgh_hdr->llh_hdr;
 174
 175                 if (LLOG_REC_HDR_NEEDS_SWABBING(llh_hdr))
 176                         lustre_swab_llog_hdr(handle->lgh_hdr);
 177
 178                 if (llh_hdr->lrh_type != LLOG_HDR_MAGIC) {
 179                         CERROR("bad log %.*s header magic: %#x (expected %#x)\n",
 180                                handle->lgh_file->f_dentry->d_name.len,
 181                                handle->lgh_file->f_dentry->d_name.name,
 182                                llh_hdr->lrh_type, LLOG_HDR_MAGIC);
 183                         rc = -EIO;
 184                 } else if (llh_hdr->lrh_len != LLOG_CHUNK_SIZE) {
 185                         CERROR("incorrectly sized log %.*s header: %#x "
 186                                "(expected %#x)\n",
 187                                handle->lgh_file->f_dentry->d_name.len,
 188                                handle->lgh_file->f_dentry->d_name.name,
 189                                llh_hdr->lrh_len, LLOG_CHUNK_SIZE);
 190                         CERROR("you may need to re-run lconf --write_conf.\n");
 191                         rc = -EIO;
 192                 }
 193         }
 194
 195         handle->lgh_last_idx = handle->lgh_hdr->llh_tail.lrt_index;
 196         handle->lgh_file->f_pos = handle->lgh_file->f_dentry->d_inode->i_size;
 197
 198         RETURN(rc);
 199 }
 200
 201 /* returns negative in on error; 0 if success && reccookie == 0; 1 otherwise */
 202 /* appends if idx == -1, otherwise overwrites record idx. */
 203 static int llog_lvfs_write_rec(struct llog_handle *loghandle,
 204                                struct llog_rec_hdr *rec,
 205                                struct llog_cookie *reccookie, int cookiecount,
 206                                void *buf, int idx)
 207 {
 208         struct llog_log_hdr *llh;
 209         int reclen = rec->lrh_len, index, rc;
 210         struct llog_rec_tail *lrt;
 211         struct obd_device *obd;
 212         struct file *file;
 213         size_t left;
 214         ENTRY;
 215
 216         llh = loghandle->lgh_hdr;
 217         file = loghandle->lgh_file;
 218         obd = loghandle->lgh_ctxt->loc_exp->exp_obd;
 219
 220         /* record length should not bigger than LLOG_CHUNK_SIZE */
 221         if (buf)
 222                 rc = (reclen > LLOG_CHUNK_SIZE - sizeof(struct llog_rec_hdr) -
 223                       sizeof(struct llog_rec_tail)) ? -E2BIG : 0;
 224         else
 225                 rc = (reclen > LLOG_CHUNK_SIZE) ? -E2BIG : 0;
 226         if (rc)
 227                 RETURN(rc);
 228
 229         if (idx != -1) {
 230                 loff_t saved_offset;
 231
 232                 /* no header: only allowed to insert record 1 */
 233                 if (idx != 1 && !file->f_dentry->d_inode->i_size) {
 234                         CERROR("idx != -1 in empty log\n");
 235                         LBUG();
 236                 }
 237
 238                 if (idx && llh->llh_size && llh->llh_size != reclen)
 239                         RETURN(-EINVAL);
 240
 241                 rc = llog_lvfs_write_blob(obd, file, &llh->llh_hdr, NULL, 0);
 242                 /* we are done if we only write the header or on error */
 243                 if (rc || idx == 0)
 244                         RETURN(rc);
 245
 246                 saved_offset = sizeof(*llh) + (idx-1)*rec->lrh_len;
 247                 rc = llog_lvfs_write_blob(obd, file, rec, buf, saved_offset);
 248                 if (rc == 0 && reccookie) {
 249                         reccookie->lgc_lgl = loghandle->lgh_id;
 250                         reccookie->lgc_index = idx;
 251                         rc = 1;
 252                 }
 253                 RETURN(rc);
 254         }
 255
 256         /* Make sure that records don't cross a chunk boundary, so we can
 257          * process them page-at-a-time if needed.  If it will cross a chunk
 258          * boundary, write in a fake (but referenced) entry to pad the chunk.
 259          *
 260          * We know that llog_current_log() will return a loghandle that is
 261          * big enough to hold reclen, so all we care about is padding here.
 262          */
 263         left = LLOG_CHUNK_SIZE - (file->f_pos & (LLOG_CHUNK_SIZE - 1));
 264         if (buf)
 265                 reclen = sizeof(*rec) + rec->lrh_len +
 266                         sizeof(struct llog_rec_tail);
 267
 268         /* NOTE: padding is a record, but no bit is set */
 269         if (left != 0 && left != reclen &&
 270             left < (reclen + LLOG_MIN_REC_SIZE)) {
 271                 loghandle->lgh_last_idx++;
 272                 rc = llog_lvfs_pad(obd, file, left, loghandle->lgh_last_idx);
 273                 if (rc)
 274                         RETURN(rc);
 275                 /* if it's the last idx in log file, then return -ENOSPC */
 276                 if (loghandle->lgh_last_idx == LLOG_BITMAP_SIZE(llh) - 1)
 277                         RETURN(-ENOSPC);
 278         }
 279
 280         loghandle->lgh_last_idx++;
 281         index = loghandle->lgh_last_idx;
 282         LASSERT(index < LLOG_BITMAP_SIZE(llh));
 283         rec->lrh_index = index;
 284         if (buf == NULL) {
 285                 lrt = (struct llog_rec_tail *)
 286                         ((char *)rec + rec->lrh_len - sizeof(*lrt));
 287                 lrt->lrt_len = rec->lrh_len;
 288                 lrt->lrt_index = rec->lrh_index;
 289         }
 290         if (ext2_set_bit(index, llh->llh_bitmap)) {
 291                 CERROR("argh, index %u already set in log bitmap?\n", index);
 292                 LBUG(); /* should never happen */
 293         }
 294         llh->llh_count++;
 295         llh->llh_tail.lrt_index = index;
 296
 297         rc = llog_lvfs_write_blob(obd, file, &llh->llh_hdr, NULL, 0);
 298         if (rc)
 299                 RETURN(rc);
 300
 301         rc = llog_lvfs_write_blob(obd, file, rec, buf, file->f_pos);
 302         if (rc)
 303                 RETURN(rc);
 304
 305         CDEBUG(D_HA, "added record "LPX64": idx: %u, %u bytes\n",
 306                loghandle->lgh_id.lgl_oid, index, rec->lrh_len);
 307         if (rc == 0 && reccookie) {
 308                 reccookie->lgc_lgl = loghandle->lgh_id;
 309                 reccookie->lgc_index = index;
 310                 if ((rec->lrh_type == MDS_UNLINK_REC) ||
 311                                 (rec->lrh_type == MDS_SETATTR_REC))
 312                         reccookie->lgc_subsys = LLOG_MDS_OST_ORIG_CTXT;
 313                 else if (rec->lrh_type == OST_SZ_REC)
 314                         reccookie->lgc_subsys = LLOG_SIZE_ORIG_CTXT;
 315                 else if (rec->lrh_type == OST_RAID1_REC)
 316                         reccookie->lgc_subsys = LLOG_RD1_ORIG_CTXT;
 317                 else
 318                         reccookie->lgc_subsys = -1;
 319                 rc = 1;
 320         }
 321         if (rc == 0 && rec->lrh_type == LLOG_GEN_REC)
 322                 rc = 1;
 323
 324         RETURN(rc);
 325 }
 326
 327 /* We can skip reading at least as many log blocks as the number of
 328 * minimum sized log records we are skipping.  If it turns out
 329 * that we are not far enough along the log (because the
 330 * actual records are larger than minimum size) we just skip
 331 * some more records. */
 332
 333 static void llog_skip_over(__u64 *off, int curr, int goal)
 334 {
 335         if (goal <= curr)
 336                 return;
 337         *off = (*off + (goal-curr-1) * LLOG_MIN_REC_SIZE) &
 338                 ~(LLOG_CHUNK_SIZE - 1);
 339 }
 340
 341
 342 /* sets:
 343  *  - cur_offset to the furthest point read in the log file
 344  *  - cur_idx to the log index preceeding cur_offset
 345  * returns -EIO/-EINVAL on error
 346  */
 347 static int llog_lvfs_next_block(struct llog_handle *loghandle, int *cur_idx,
 348                                 int next_idx, __u64 *cur_offset, void *buf,
 349                                 int len)
 350 {
 351         int rc;
 352         ENTRY;
 353
 354         if (len == 0 || len & (LLOG_CHUNK_SIZE - 1))
 355                 RETURN(-EINVAL);
 356
 357         CDEBUG(D_OTHER, "looking for log index %u (cur idx %u off "LPU64")\n",
 358                next_idx, *cur_idx, *cur_offset);
 359
 360         while (*cur_offset < loghandle->lgh_file->f_dentry->d_inode->i_size) {
 361                 struct llog_rec_hdr *rec;
 362                 struct llog_rec_tail *tail;
 363                 loff_t ppos;
 364
 365                 llog_skip_over(cur_offset, *cur_idx, next_idx);
 366
 367                 ppos = *cur_offset;
 368                 rc = fsfilt_read_record(loghandle->lgh_ctxt->loc_exp->exp_obd,
 369                                         loghandle->lgh_file, buf, len,
 370                                         &ppos);
 371                 if (rc) {
 372                         CERROR("Cant read llog block at log id "LPU64
 373                                "/%u offset "LPU64"\n",
 374                                loghandle->lgh_id.lgl_oid,
 375                                loghandle->lgh_id.lgl_ogen,
 376                                *cur_offset);
 377                         RETURN(rc);
 378                 }
 379
 380                 /* put number of bytes read into rc to make code simpler */
 381                 rc = ppos - *cur_offset;
 382                 *cur_offset = ppos;
 383
 384                 if (rc == 0) /* end of file, nothing to do */
 385                         RETURN(0);
 386
 387                 if (rc < sizeof(*tail)) {
 388                         CERROR("Invalid llog block at log id "LPU64"/%u offset "
 389                                LPU64"\n", loghandle->lgh_id.lgl_oid,
 390                                loghandle->lgh_id.lgl_ogen, *cur_offset);
 391                         RETURN(-EINVAL);
 392                 }
 393
 394                 rec = buf;
 395                 tail = (struct llog_rec_tail *)((char *)buf + rc -
 396                                                 sizeof(struct llog_rec_tail));
 397
 398                 if (LLOG_REC_HDR_NEEDS_SWABBING(rec)) {
 399                         lustre_swab_llog_rec(rec, tail);
 400                 }
 401
 402                 *cur_idx = tail->lrt_index;
 403
 404                 /* this shouldn't happen */
 405                 if (tail->lrt_index == 0) {
 406                         CERROR("Invalid llog tail at log id "LPU64"/%u offset "
 407                                LPU64"\n", loghandle->lgh_id.lgl_oid,
 408                                loghandle->lgh_id.lgl_ogen, *cur_offset);
 409                         RETURN(-EINVAL);
 410                 }
 411                 if (tail->lrt_index < next_idx)
 412                         continue;
 413
 414                 /* sanity check that the start of the new buffer is no farther
 415                  * than the record that we wanted.  This shouldn't happen. */
 416                 if (rec->lrh_index > next_idx) {
 417                         CERROR("missed desired record? %u > %u\n",
 418                                rec->lrh_index, next_idx);
 419                         RETURN(-ENOENT);
 420                 }
 421                 RETURN(0);
 422         }
 423         RETURN(-EIO);
 424 }
 425
 426 static int llog_lvfs_prev_block(struct llog_handle *loghandle,
 427                                 int prev_idx, void *buf, int len)
 428 {
 429         __u64 cur_offset;
 430         int rc;
 431         ENTRY;
 432
 433         if (len == 0 || len & (LLOG_CHUNK_SIZE - 1))
 434                 RETURN(-EINVAL);
 435
 436         CDEBUG(D_OTHER, "looking for log index %u n", prev_idx);
 437
 438         cur_offset = LLOG_CHUNK_SIZE;
 439         llog_skip_over(&cur_offset, 0, prev_idx);
 440
 441         while (cur_offset < loghandle->lgh_file->f_dentry->d_inode->i_size) {
 442                 struct llog_rec_hdr *rec;
 443                 struct llog_rec_tail *tail;
 444                 loff_t ppos;
 445
 446                 ppos = cur_offset;
 447
 448                 rc = fsfilt_read_record(loghandle->lgh_ctxt->loc_exp->exp_obd,
 449                                         loghandle->lgh_file, buf, len,
 450                                         &ppos);
 451                 if (rc) {
 452                         CERROR("Cant read llog block at log id "LPU64
 453                                "/%u offset "LPU64"\n",
 454                                loghandle->lgh_id.lgl_oid,
 455                                loghandle->lgh_id.lgl_ogen,
 456                                cur_offset);
 457                         RETURN(rc);
 458                 }
 459
 460                 /* put number of bytes read into rc to make code simpler */
 461                 rc = ppos - cur_offset;
 462                 cur_offset = ppos;
 463
 464                 if (rc == 0) /* end of file, nothing to do */
 465                         RETURN(0);
 466
 467                 if (rc < sizeof(*tail)) {
 468                         CERROR("Invalid llog block at log id "LPU64"/%u offset "
 469                                LPU64"\n", loghandle->lgh_id.lgl_oid,
 470                                loghandle->lgh_id.lgl_ogen, cur_offset);
 471                         RETURN(-EINVAL);
 472                 }
 473
 474                 tail = buf + rc - sizeof(struct llog_rec_tail);
 475
 476                 /* this shouldn't happen */
 477                 if (tail->lrt_index == 0) {
 478                         CERROR("Invalid llog tail at log id "LPU64"/%u offset "
 479                                LPU64"\n", loghandle->lgh_id.lgl_oid,
 480                                loghandle->lgh_id.lgl_ogen, cur_offset);
 481                         RETURN(-EINVAL);
 482                 }
 483                 if (le32_to_cpu(tail->lrt_index) < prev_idx)
 484                         continue;
 485
 486                 /* sanity check that the start of the new buffer is no farther
 487                  * than the record that we wanted.  This shouldn't happen. */
 488                 rec = buf;
 489                 if (le32_to_cpu(rec->lrh_index) > prev_idx) {
 490                         CERROR("missed desired record? %u > %u\n",
 491                                le32_to_cpu(rec->lrh_index), prev_idx);
 492                         RETURN(-ENOENT);
 493                 }
 494                 RETURN(0);
 495         }
 496         RETURN(-EIO);
 497 }
 498
 499 static struct file *llog_filp_open(char *dir, char *name, int flags, int mode)
 500 {
 501         char *logname;
 502         struct file *filp;
 503         int len;
 504
 505         OBD_ALLOC(logname, PATH_MAX);
 506         if (logname == NULL)
 507                 return ERR_PTR(-ENOMEM);
 508
 509         len = snprintf(logname, PATH_MAX, "%s/%s", dir, name);
 510         if (len >= PATH_MAX - 1) {
 511                 filp = ERR_PTR(-ENAMETOOLONG);
 512         } else {
 513                 filp = l_filp_open(logname, flags, mode);
 514                 if (IS_ERR(filp))
 515                         CERROR("logfile creation %s: %ld\n", logname,
 516                                PTR_ERR(filp));
 517         }
 518         OBD_FREE(logname, PATH_MAX);
 519         return filp;
 520 }
 521
 522 /* This is a callback from the llog_* functions.
 523  * Assumes caller has already pushed us into the kernel context. */
 524 static int llog_lvfs_create(struct llog_ctxt *ctxt, struct llog_handle **res,
 525                             struct llog_logid *logid, char *name)
 526 {
 527         struct llog_handle *handle;
 528         struct obd_device *obd;
 529         struct l_dentry *dchild = NULL;
 530         struct obdo *oa = NULL;
 531         int rc = 0, cleanup_phase = 1;
 532         int open_flags = O_RDWR | O_CREAT | O_LARGEFILE;
 533         ENTRY;
 534
 535         handle = llog_alloc_handle();
 536         if (handle == NULL)
 537                 RETURN(-ENOMEM);
 538         *res = handle;
 539
 540         LASSERT(ctxt);
 541         LASSERT(ctxt->loc_exp);
 542         obd = ctxt->loc_exp->exp_obd;
 543
 544         if (logid != NULL) {
 545                 dchild = obd_lvfs_fid2dentry(ctxt->loc_exp, logid->lgl_oid,
 546                                              logid->lgl_ogen, logid->lgl_ogr);
 547
 548                 if (IS_ERR(dchild)) {
 549                         rc = PTR_ERR(dchild);
 550                         CERROR("error looking up logfile "LPX64":0x%x: rc %d\n",
 551                                logid->lgl_oid, logid->lgl_ogen, rc);
 552                         GOTO(cleanup, rc);
 553                 }
 554
 555                 cleanup_phase = 2;
 556                 if (dchild->d_inode == NULL) {
 557                         rc = -ENOENT;
 558                         CERROR("nonexistent log file "LPX64":"LPX64": rc %d\n",
 559                                logid->lgl_oid, logid->lgl_ogr, rc);
 560                         GOTO(cleanup, rc);
 561                 }
 562
 563                 handle->lgh_file = l_dentry_open(&obd->obd_lvfs_ctxt, dchild,
 564                                                     O_RDWR | O_LARGEFILE);
 565                 if (IS_ERR(handle->lgh_file)) {
 566                         rc = PTR_ERR(handle->lgh_file);
 567                         CERROR("error opening logfile "LPX64"0x%x: rc %d\n",
 568                                logid->lgl_oid, logid->lgl_ogen, rc);
 569                         GOTO(cleanup, rc);
 570                 }
 571
 572                 /* assign the value of lgh_id for handle directly */
 573                 handle->lgh_id = *logid;
 574
 575         } else if (name) {
 576                 /* COMPAT_146 */
 577                 if (strcmp(obd->obd_type->typ_name, LUSTRE_MDS_NAME) == 0) {
 578                         handle->lgh_file = llog_filp_open(MDT_LOGS_DIR, name,
 579                                                           open_flags, 0644);
 580                 } else {
 581                         /* end COMPAT_146 */
 582                         handle->lgh_file = llog_filp_open(MOUNT_CONFIGS_DIR,
 583                                                           name, open_flags,
 584                                                           0644);
 585                 }
 586                 if (IS_ERR(handle->lgh_file))
 587                         GOTO(cleanup, rc = PTR_ERR(handle->lgh_file));
 588
 589                 handle->lgh_id.lgl_ogr = 1;
 590                 handle->lgh_id.lgl_oid =
 591                         handle->lgh_file->f_dentry->d_inode->i_ino;
 592                 handle->lgh_id.lgl_ogen =
 593                         handle->lgh_file->f_dentry->d_inode->i_generation;
 594         } else {
 595                 oa = obdo_alloc();
 596                 if (oa == NULL)
 597                         GOTO(cleanup, rc = -ENOMEM);
 598
 599                 oa->o_gr = FILTER_GROUP_LLOG;
 600                 oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLGROUP;
 601
 602                 rc = obd_create(ctxt->loc_exp, oa, NULL, NULL);
 603                 if (rc)
 604                         GOTO(cleanup, rc);
 605
 606                 dchild = obd_lvfs_fid2dentry(ctxt->loc_exp, oa->o_id,
 607                                              oa->o_generation, oa->o_gr);
 608
 609                 if (IS_ERR(dchild))
 610                         GOTO(cleanup, rc = PTR_ERR(dchild));
 611                 cleanup_phase = 2;
 612                 handle->lgh_file = l_dentry_open(&obd->obd_lvfs_ctxt, dchild,
 613                                                  open_flags);
 614                 if (IS_ERR(handle->lgh_file))
 615                         GOTO(cleanup, rc = PTR_ERR(handle->lgh_file));
 616
 617                 handle->lgh_id.lgl_ogr = oa->o_gr;
 618                 handle->lgh_id.lgl_oid = oa->o_id;
 619                 handle->lgh_id.lgl_ogen = oa->o_generation;
 620         }
 621
 622         handle->lgh_ctxt = ctxt;
 623  finish:
 624         if (oa)
 625                 obdo_free(oa);
 626         RETURN(rc);
 627 cleanup:
 628         switch (cleanup_phase) {
 629         case 2:
 630                 l_dput(dchild);
 631         case 1:
 632                 llog_free_handle(handle);
 633         }
 634         goto finish;
 635 }
 636
 637 static int llog_lvfs_close(struct llog_handle *handle)
 638 {
 639         int rc;
 640         ENTRY;
 641
 642         rc = filp_close(handle->lgh_file, 0);
 643         if (rc)
 644                 CERROR("error closing log: rc %d\n", rc);
 645         RETURN(rc);
 646 }
 647
 648 static int llog_lvfs_destroy(struct llog_handle *handle)
 649 {
 650         struct dentry *fdentry;
 651         struct obdo *oa;
 652         struct obd_device *obd = handle->lgh_ctxt->loc_exp->exp_obd;
 653         char *dir;
 654         int rc;
 655         ENTRY;
 656
 657         /* COMPAT_146 */
 658         if (strcmp(obd->obd_type->typ_name, LUSTRE_MDS_NAME) == 0)
 659                 dir = MDT_LOGS_DIR;
 660         else
 661                 /* end COMPAT_146 */
 662                 dir = MOUNT_CONFIGS_DIR;
 663
 664         fdentry = handle->lgh_file->f_dentry;
 665         if (strcmp(fdentry->d_parent->d_name.name, dir) == 0) {
 666                 struct inode *inode = fdentry->d_parent->d_inode;
 667                 struct lvfs_run_ctxt saved;
 668
 669                 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 670                 dget(fdentry);
 671                 rc = llog_lvfs_close(handle);
 672
 673                 if (rc == 0) {
 674                         LOCK_INODE_MUTEX(inode);
 675                         rc = vfs_unlink(inode, fdentry);
 676                         UNLOCK_INODE_MUTEX(inode);
 677                 }
 678
 679                 dput(fdentry);
 680                 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 681                 RETURN(rc);
 682         }
 683
 684         oa = obdo_alloc();
 685         if (oa == NULL)
 686                 RETURN(-ENOMEM);
 687
 688         oa->o_id = handle->lgh_id.lgl_oid;
 689         oa->o_gr = handle->lgh_id.lgl_ogr;
 690         oa->o_generation = handle->lgh_id.lgl_ogen;
 691         oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP | OBD_MD_FLGENER;
 692
 693         rc = llog_lvfs_close(handle);
 694         if (rc)
 695                 GOTO(out, rc);
 696
 697         rc = obd_destroy(handle->lgh_ctxt->loc_exp, oa, NULL, NULL, NULL);
 698  out:
 699         obdo_free(oa);
 700         RETURN(rc);
 701 }
 702
 703 /* reads the catalog list */
 704 int llog_get_cat_list(struct obd_device *obd, struct obd_device *disk_obd,
 705                       char *name, int count, struct llog_catid *idarray)
 706 {
 707         struct lvfs_run_ctxt saved;
 708         struct l_file *file;
 709         int rc;
 710         int size = sizeof(*idarray) * count;
 711         loff_t off = 0;
 712         ENTRY;
 713
 714         if (!count)
 715                 RETURN(0);
 716
 717         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 718         file = filp_open(name, O_RDWR | O_CREAT | O_LARGEFILE, 0700);
 719         if (!file || IS_ERR(file)) {
 720                 rc = PTR_ERR(file);
 721                 CERROR("OBD filter: cannot open/create %s: rc = %d\n",
 722                        name, rc);
 723                 GOTO(out, rc);
 724         }
 725
 726         if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
 727                 CERROR("%s is not a regular file!: mode = %o\n", name,
 728                        file->f_dentry->d_inode->i_mode);
 729                 GOTO(out, rc = -ENOENT);
 730         }
 731
 732         CDEBUG(D_CONFIG, "cat list: disk size=%d, read=%d\n",
 733                (int)file->f_dentry->d_inode->i_size, size);
 734
 735         rc = fsfilt_read_record(disk_obd, file, idarray, size, &off);
 736         if (rc) {
 737                 CERROR("OBD filter: error reading %s: rc %d\n", name, rc);
 738                 GOTO(out, rc);
 739         }
 740
 741         EXIT;
 742  out:
 743         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 744         if (file && !IS_ERR(file))
 745                 rc = filp_close(file, 0);
 746         return rc;
 747 }
 748 EXPORT_SYMBOL(llog_get_cat_list);
 749
 750 /* writes the cat list */
 751 int llog_put_cat_list(struct obd_device *obd, struct obd_device *disk_obd,
 752                       char *name, int count, struct llog_catid *idarray)
 753 {
 754         struct lvfs_run_ctxt saved;
 755         struct l_file *file;
 756         int rc;
 757         int size = sizeof(*idarray) * count;
 758         loff_t off = 0;
 759
 760         if (!count)
 761                 return (0);
 762
 763         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 764         file = filp_open(name, O_RDWR | O_CREAT | O_LARGEFILE, 0700);
 765         if (!file || IS_ERR(file)) {
 766                 rc = PTR_ERR(file);
 767                 CERROR("OBD filter: cannot open/create %s: rc = %d\n",
 768                        name, rc);
 769                 GOTO(out, rc);
 770         }
 771
 772         if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
 773                 CERROR("%s is not a regular file!: mode = %o\n", name,
 774                        file->f_dentry->d_inode->i_mode);
 775                 GOTO(out, rc = -ENOENT);
 776         }
 777
 778         rc = fsfilt_write_record(disk_obd, file, idarray, size, &off, 1);
 779         if (rc) {
 780                 CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
 781                        name, rc);
 782                 GOTO(out, rc);
 783         }
 784
 785  out:
 786         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 787         if (file && !IS_ERR(file))
 788                 rc = filp_close(file, 0);
 789         RETURN(rc);
 790 }
 791
 792 struct llog_operations llog_lvfs_ops = {
 793         lop_write_rec:   llog_lvfs_write_rec,
 794         lop_next_block:  llog_lvfs_next_block,
 795         lop_prev_block:  llog_lvfs_prev_block,
 796         lop_read_header: llog_lvfs_read_header,
 797         lop_create:      llog_lvfs_create,
 798         lop_destroy:     llog_lvfs_destroy,
 799         lop_close:       llog_lvfs_close,
 800         //        lop_cancel: llog_lvfs_cancel,
 801 };
 802
 803 EXPORT_SYMBOL(llog_lvfs_ops);
 804
 805 #else /* !__KERNEL__ */
 806
 807 static int llog_lvfs_read_header(struct llog_handle *handle)
 808 {
 809         LBUG();
 810         return 0;
 811 }
 812
 813 static int llog_lvfs_write_rec(struct llog_handle *loghandle,
 814                                struct llog_rec_hdr *rec,
 815                                struct llog_cookie *reccookie, int cookiecount,
 816                                void *buf, int idx)
 817 {
 818         LBUG();
 819         return 0;
 820 }
 821
 822 static int llog_lvfs_next_block(struct llog_handle *loghandle, int *cur_idx,
 823                                 int next_idx, __u64 *cur_offset, void *buf,
 824                                 int len)
 825 {
 826         LBUG();
 827         return 0;
 828 }
 829
 830 static int llog_lvfs_prev_block(struct llog_handle *loghandle,
 831                                 int prev_idx, void *buf, int len)
 832 {
 833         LBUG();
 834         return 0;
 835 }
 836
 837 static int llog_lvfs_create(struct llog_ctxt *ctxt, struct llog_handle **res,
 838                             struct llog_logid *logid, char *name)
 839 {
 840         LBUG();
 841         return 0;
 842 }
 843
 844 static int llog_lvfs_close(struct llog_handle *handle)
 845 {
 846         LBUG();
 847         return 0;
 848 }
 849
 850 static int llog_lvfs_destroy(struct llog_handle *handle)
 851 {
 852         LBUG();
 853         return 0;
 854 }
 855
 856 int llog_get_cat_list(struct obd_device *obd, struct obd_device *disk_obd,
 857                       char *name, int count, struct llog_catid *idarray)
 858 {
 859         LBUG();
 860         return 0;
 861 }
 862
 863 int llog_put_cat_list(struct obd_device *obd, struct obd_device *disk_obd,
 864                       char *name, int count, struct llog_catid *idarray)
 865 {
 866         LBUG();
 867         return 0;
 868 }
 869
 870 struct llog_operations llog_lvfs_ops = {
 871         lop_write_rec:   llog_lvfs_write_rec,
 872         lop_next_block:  llog_lvfs_next_block,
 873         lop_prev_block:  llog_lvfs_prev_block,
 874         lop_read_header: llog_lvfs_read_header,
 875         lop_create:      llog_lvfs_create,
 876         lop_destroy:     llog_lvfs_destroy,
 877         lop_close:       llog_lvfs_close,
 878 //        lop_cancel:      llog_lvfs_cancel,
 879 };
 880 #endif