lustre/obdclass/llog_lvfs.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  */
  30 /*
  31  * This file is part of Lustre, http://www.lustre.org/
  32  * Lustre is a trademark of Sun Microsystems, Inc.
  33  *
  34  * lustre/obdclass/llog_lvfs.c
  35  *
  36  * OST<->MDS recovery logging infrastructure.
  37  * Invariants in implementation:
  38  * - we do not share logs among different OST<->MDS connections, so that
  39  *   if an OST or MDS fails it need only look at log(s) relevant to itself
  40  *
  41  * Author: Andreas Dilger <adilger@clusterfs.com>
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_LOG
  45
  46 #ifndef __KERNEL__
  47 #include <liblustre.h>
  48 #endif
  49
  50 #include <obd.h>
  51 #include <obd_class.h>
  52 #include <lustre_log.h>
  53 #include <obd_ost.h>
  54 #include <libcfs/list.h>
  55 #include <lvfs.h>
  56 #include <lustre_fsfilt.h>
  57 #include <lustre_disk.h>
  58 #include "llog_internal.h"
  59
  60 #if defined(__KERNEL__) && defined(LLOG_LVFS)
  61
  62 static int llog_lvfs_pad(struct obd_device *obd, struct l_file *file,
  63                                 int len, int index)
  64 {
  65         struct llog_rec_hdr rec = { 0 };
  66         struct llog_rec_tail tail;
  67         int rc;
  68         ENTRY;
  69
  70         LASSERT(len >= LLOG_MIN_REC_SIZE && (len & 0x7) == 0);
  71
  72         tail.lrt_len = rec.lrh_len = len;
  73         tail.lrt_index = rec.lrh_index = index;
  74         rec.lrh_type = LLOG_PAD_MAGIC;
  75
  76         rc = fsfilt_write_record(obd, file, &rec, sizeof(rec), &file->f_pos, 0);
  77         if (rc) {
  78                 CERROR("error writing padding record: rc %d\n", rc);
  79                 goto out;
  80         }
  81
  82         file->f_pos += len - sizeof(rec) - sizeof(tail);
  83         rc = fsfilt_write_record(obd, file, &tail, sizeof(tail),&file->f_pos,0);
  84         if (rc) {
  85                 CERROR("error writing padding record: rc %d\n", rc);
  86                 goto out;
  87         }
  88
  89  out:
  90         RETURN(rc);
  91 }
  92
  93 static int llog_lvfs_write_blob(struct obd_device *obd, struct l_file *file,
  94                                 struct llog_rec_hdr *rec, void *buf, loff_t off)
  95 {
  96         int rc;
  97         struct llog_rec_tail end;
  98         loff_t saved_off = file->f_pos;
  99         int buflen = rec->lrh_len;
 100
 101         ENTRY;
 102
 103         file->f_pos = off;
 104
 105         if (buflen == 0)
 106                 CWARN("0-length record\n");
 107
 108         if (!buf) {
 109                 rc = fsfilt_write_record(obd, file, rec, buflen,&file->f_pos,0);
 110                 if (rc) {
 111                         CERROR("error writing log record: rc %d\n", rc);
 112                         goto out;
 113                 }
 114                 GOTO(out, rc = 0);
 115         }
 116
 117         /* the buf case */
 118         rec->lrh_len = sizeof(*rec) + buflen + sizeof(end);
 119         rc = fsfilt_write_record(obd, file, rec, sizeof(*rec), &file->f_pos, 0);
 120         if (rc) {
 121                 CERROR("error writing log hdr: rc %d\n", rc);
 122                 goto out;
 123         }
 124
 125         rc = fsfilt_write_record(obd, file, buf, buflen, &file->f_pos, 0);
 126         if (rc) {
 127                 CERROR("error writing log buffer: rc %d\n", rc);
 128                 goto out;
 129         }
 130
 131         end.lrt_len = rec->lrh_len;
 132         end.lrt_index = rec->lrh_index;
 133         rc = fsfilt_write_record(obd, file, &end, sizeof(end), &file->f_pos, 0);
 134         if (rc) {
 135                 CERROR("error writing log tail: rc %d\n", rc);
 136                 goto out;
 137         }
 138
 139         rc = 0;
 140  out:
 141         if (saved_off > file->f_pos)
 142                 file->f_pos = saved_off;
 143         LASSERT(rc <= 0);
 144         RETURN(rc);
 145 }
 146
 147 static int llog_lvfs_read_blob(struct obd_device *obd, struct l_file *file,
 148                                 void *buf, int size, loff_t off)
 149 {
 150         loff_t offset = off;
 151         int rc;
 152         ENTRY;
 153
 154         rc = fsfilt_read_record(obd, file, buf, size, &offset);
 155         if (rc) {
 156                 CERROR("error reading log record: rc %d\n", rc);
 157                 RETURN(rc);
 158         }
 159         RETURN(0);
 160 }
 161
 162 static int llog_lvfs_read_header(const struct lu_env *env,
 163                                  struct llog_handle *handle)
 164 {
 165         struct obd_device *obd;
 166         int rc;
 167         ENTRY;
 168
 169         LASSERT(sizeof(*handle->lgh_hdr) == LLOG_CHUNK_SIZE);
 170
 171         obd = handle->lgh_ctxt->loc_exp->exp_obd;
 172
 173         if (i_size_read(handle->lgh_file->f_dentry->d_inode) == 0) {
 174                 CDEBUG(D_HA, "not reading header from 0-byte log\n");
 175                 RETURN(LLOG_EEMPTY);
 176         }
 177
 178         rc = llog_lvfs_read_blob(obd, handle->lgh_file, handle->lgh_hdr,
 179                                  LLOG_CHUNK_SIZE, 0);
 180         if (rc) {
 181                 CERROR("error reading log header from %.*s\n",
 182                        handle->lgh_file->f_dentry->d_name.len,
 183                        handle->lgh_file->f_dentry->d_name.name);
 184         } else {
 185                 struct llog_rec_hdr *llh_hdr = &handle->lgh_hdr->llh_hdr;
 186
 187                 if (LLOG_REC_HDR_NEEDS_SWABBING(llh_hdr))
 188                         lustre_swab_llog_hdr(handle->lgh_hdr);
 189
 190                 if (llh_hdr->lrh_type != LLOG_HDR_MAGIC) {
 191                         CERROR("bad log %.*s header magic: %#x (expected %#x)\n",
 192                                handle->lgh_file->f_dentry->d_name.len,
 193                                handle->lgh_file->f_dentry->d_name.name,
 194                                llh_hdr->lrh_type, LLOG_HDR_MAGIC);
 195                         rc = -EIO;
 196                 } else if (llh_hdr->lrh_len != LLOG_CHUNK_SIZE) {
 197                         CERROR("incorrectly sized log %.*s header: %#x "
 198                                "(expected %#x)\n",
 199                                handle->lgh_file->f_dentry->d_name.len,
 200                                handle->lgh_file->f_dentry->d_name.name,
 201                                llh_hdr->lrh_len, LLOG_CHUNK_SIZE);
 202                         CERROR("you may need to re-run lconf --write_conf.\n");
 203                         rc = -EIO;
 204                 }
 205         }
 206
 207         handle->lgh_last_idx = handle->lgh_hdr->llh_tail.lrt_index;
 208         handle->lgh_file->f_pos = i_size_read(handle->lgh_file->f_dentry->d_inode);
 209
 210         RETURN(rc);
 211 }
 212
 213 /* returns negative in on error; 0 if success && reccookie == 0; 1 otherwise */
 214 /* appends if idx == -1, otherwise overwrites record idx. */
 215 static int llog_lvfs_write_rec(const struct lu_env *env,
 216                                struct llog_handle *loghandle,
 217                                struct llog_rec_hdr *rec,
 218                                struct llog_cookie *reccookie, int cookiecount,
 219                                void *buf, int idx)
 220 {
 221         struct llog_log_hdr *llh;
 222         int reclen = rec->lrh_len, index, rc;
 223         struct llog_rec_tail *lrt;
 224         struct obd_device *obd;
 225         struct file *file;
 226         size_t left;
 227         ENTRY;
 228
 229         llh = loghandle->lgh_hdr;
 230         file = loghandle->lgh_file;
 231         obd = loghandle->lgh_ctxt->loc_exp->exp_obd;
 232
 233         /* record length should not bigger than LLOG_CHUNK_SIZE */
 234         if (buf)
 235                 rc = (reclen > LLOG_CHUNK_SIZE - sizeof(struct llog_rec_hdr) -
 236                       sizeof(struct llog_rec_tail)) ? -E2BIG : 0;
 237         else
 238                 rc = (reclen > LLOG_CHUNK_SIZE) ? -E2BIG : 0;
 239         if (rc)
 240                 RETURN(rc);
 241
 242         if (buf)
 243                 /* write_blob adds header and tail to lrh_len. */
 244                 reclen = sizeof(*rec) + rec->lrh_len +
 245                          sizeof(struct llog_rec_tail);
 246
 247         if (idx != -1) {
 248                 loff_t saved_offset;
 249
 250                 /* no header: only allowed to insert record 1 */
 251                 if (idx != 1 && !i_size_read(file->f_dentry->d_inode)) {
 252                         CERROR("idx != -1 in empty log\n");
 253                         LBUG();
 254                 }
 255
 256                 if (idx && llh->llh_size && llh->llh_size != rec->lrh_len)
 257                         RETURN(-EINVAL);
 258
 259                 if (!ext2_test_bit(idx, llh->llh_bitmap))
 260                         CERROR("Modify unset record %u\n", idx);
 261                 if (idx != rec->lrh_index)
 262                         CERROR("Index mismatch %d %u\n", idx, rec->lrh_index);
 263
 264                 rc = llog_lvfs_write_blob(obd, file, &llh->llh_hdr, NULL, 0);
 265                 /* we are done if we only write the header or on error */
 266                 if (rc || idx == 0)
 267                         RETURN(rc);
 268
 269                 /* Assumes constant lrh_len */
 270                 saved_offset = sizeof(*llh) + (idx - 1) * reclen;
 271
 272                 if (buf) {
 273                         struct llog_rec_hdr check;
 274
 275                         /* We assume that caller has set lgh_cur_* */
 276                         saved_offset = loghandle->lgh_cur_offset;
 277                         CDEBUG(D_OTHER,
 278                                "modify record "LPX64": idx:%d/%u/%d, len:%u "
 279                                "offset %llu\n",
 280                                loghandle->lgh_id.lgl_oid, idx, rec->lrh_index,
 281                                loghandle->lgh_cur_idx, rec->lrh_len,
 282                                (long long)(saved_offset - sizeof(*llh)));
 283                         if (rec->lrh_index != loghandle->lgh_cur_idx) {
 284                                 CERROR("modify idx mismatch %u/%d\n",
 285                                        idx, loghandle->lgh_cur_idx);
 286                                 RETURN(-EFAULT);
 287                         }
 288 #if 1  /* FIXME remove this safety check at some point */
 289                         /* Verify that the record we're modifying is the
 290                            right one. */
 291                         rc = llog_lvfs_read_blob(obd, file, &check,
 292                                                  sizeof(check), saved_offset);
 293                         if (check.lrh_index != idx || check.lrh_len != reclen) {
 294                                 CERROR("Bad modify idx %u/%u size %u/%u (%d)\n",
 295                                        idx, check.lrh_index, reclen,
 296                                        check.lrh_len, rc);
 297                                 RETURN(-EFAULT);
 298                         }
 299 #endif
 300                 }
 301
 302                 rc = llog_lvfs_write_blob(obd, file, rec, buf, saved_offset);
 303                 if (rc == 0 && reccookie) {
 304                         reccookie->lgc_lgl = loghandle->lgh_id;
 305                         reccookie->lgc_index = idx;
 306                         rc = 1;
 307                 }
 308                 RETURN(rc);
 309         }
 310
 311         /* Make sure that records don't cross a chunk boundary, so we can
 312          * process them page-at-a-time if needed.  If it will cross a chunk
 313          * boundary, write in a fake (but referenced) entry to pad the chunk.
 314          *
 315          * We know that llog_current_log() will return a loghandle that is
 316          * big enough to hold reclen, so all we care about is padding here.
 317          */
 318         left = LLOG_CHUNK_SIZE - (file->f_pos & (LLOG_CHUNK_SIZE - 1));
 319
 320         /* NOTE: padding is a record, but no bit is set */
 321         if (left != 0 && left != reclen &&
 322             left < (reclen + LLOG_MIN_REC_SIZE)) {
 323                  index = loghandle->lgh_last_idx + 1;
 324                  rc = llog_lvfs_pad(obd, file, left, index);
 325                  if (rc)
 326                          RETURN(rc);
 327                  loghandle->lgh_last_idx++; /*for pad rec*/
 328          }
 329          /* if it's the last idx in log file, then return -ENOSPC */
 330          if (loghandle->lgh_last_idx >= LLOG_BITMAP_SIZE(llh) - 1)
 331                  RETURN(-ENOSPC);
 332         loghandle->lgh_last_idx++;
 333         index = loghandle->lgh_last_idx;
 334         LASSERT(index < LLOG_BITMAP_SIZE(llh));
 335         rec->lrh_index = index;
 336         if (buf == NULL) {
 337                 lrt = (struct llog_rec_tail *)
 338                         ((char *)rec + rec->lrh_len - sizeof(*lrt));
 339                 lrt->lrt_len = rec->lrh_len;
 340                 lrt->lrt_index = rec->lrh_index;
 341         }
 342         /*The caller should make sure only 1 process access the lgh_last_idx,
 343          *Otherwise it might hit the assert.*/
 344         LASSERT(index < LLOG_BITMAP_SIZE(llh));
 345         if (ext2_set_bit(index, llh->llh_bitmap)) {
 346                 CERROR("argh, index %u already set in log bitmap?\n", index);
 347                 LBUG(); /* should never happen */
 348         }
 349         llh->llh_count++;
 350         llh->llh_tail.lrt_index = index;
 351
 352         rc = llog_lvfs_write_blob(obd, file, &llh->llh_hdr, NULL, 0);
 353         if (rc)
 354                 RETURN(rc);
 355
 356         rc = llog_lvfs_write_blob(obd, file, rec, buf, file->f_pos);
 357         if (rc)
 358                 RETURN(rc);
 359
 360         CDEBUG(D_RPCTRACE, "added record "LPX64": idx: %u, %u \n",
 361                loghandle->lgh_id.lgl_oid, index, rec->lrh_len);
 362         if (rc == 0 && reccookie) {
 363                 reccookie->lgc_lgl = loghandle->lgh_id;
 364                 reccookie->lgc_index = index;
 365                 if ((rec->lrh_type == MDS_UNLINK_REC) ||
 366                     (rec->lrh_type == MDS_SETATTR64_REC))
 367                         reccookie->lgc_subsys = LLOG_MDS_OST_ORIG_CTXT;
 368                 else if (rec->lrh_type == OST_SZ_REC)
 369                         reccookie->lgc_subsys = LLOG_SIZE_ORIG_CTXT;
 370                 else
 371                         reccookie->lgc_subsys = -1;
 372                 rc = 1;
 373         }
 374         if (rc == 0 && rec->lrh_type == LLOG_GEN_REC)
 375                 rc = 1;
 376
 377         RETURN(rc);
 378 }
 379
 380 /* We can skip reading at least as many log blocks as the number of
 381 * minimum sized log records we are skipping.  If it turns out
 382 * that we are not far enough along the log (because the
 383 * actual records are larger than minimum size) we just skip
 384 * some more records. */
 385
 386 static void llog_skip_over(__u64 *off, int curr, int goal)
 387 {
 388         if (goal <= curr)
 389                 return;
 390         *off = (*off + (goal-curr-1) * LLOG_MIN_REC_SIZE) &
 391                 ~(LLOG_CHUNK_SIZE - 1);
 392 }
 393
 394
 395 /* sets:
 396  *  - cur_offset to the furthest point read in the log file
 397  *  - cur_idx to the log index preceeding cur_offset
 398  * returns -EIO/-EINVAL on error
 399  */
 400 static int llog_lvfs_next_block(const struct lu_env *env,
 401                                 struct llog_handle *loghandle, int *cur_idx,
 402                                 int next_idx, __u64 *cur_offset, void *buf,
 403                                 int len)
 404 {
 405         int rc;
 406         ENTRY;
 407
 408         if (len == 0 || len & (LLOG_CHUNK_SIZE - 1))
 409                 RETURN(-EINVAL);
 410
 411         CDEBUG(D_OTHER, "looking for log index %u (cur idx %u off "LPU64")\n",
 412                next_idx, *cur_idx, *cur_offset);
 413
 414         while (*cur_offset < i_size_read(loghandle->lgh_file->f_dentry->d_inode)) {
 415                 struct llog_rec_hdr *rec;
 416                 struct llog_rec_tail *tail;
 417                 loff_t ppos;
 418
 419                 llog_skip_over(cur_offset, *cur_idx, next_idx);
 420
 421                 ppos = *cur_offset;
 422                 rc = fsfilt_read_record(loghandle->lgh_ctxt->loc_exp->exp_obd,
 423                                         loghandle->lgh_file, buf, len,
 424                                         &ppos);
 425                 if (rc) {
 426                         CERROR("Cant read llog block at log id "LPU64
 427                                "/%u offset "LPU64"\n",
 428                                loghandle->lgh_id.lgl_oid,
 429                                loghandle->lgh_id.lgl_ogen,
 430                                *cur_offset);
 431                         RETURN(rc);
 432                 }
 433
 434                 /* put number of bytes read into rc to make code simpler */
 435                 rc = ppos - *cur_offset;
 436                 *cur_offset = ppos;
 437
 438                 if (rc < len) {
 439                         /* signal the end of the valid buffer to llog_process */
 440                         memset(buf + rc, 0, len - rc);
 441                 }
 442
 443                 if (rc == 0) /* end of file, nothing to do */
 444                         RETURN(0);
 445
 446                 if (rc < sizeof(*tail)) {
 447                         CERROR("Invalid llog block at log id "LPU64"/%u offset "
 448                                LPU64"\n", loghandle->lgh_id.lgl_oid,
 449                                loghandle->lgh_id.lgl_ogen, *cur_offset);
 450                         RETURN(-EINVAL);
 451                 }
 452
 453                 rec = buf;
 454                 tail = (struct llog_rec_tail *)((char *)buf + rc -
 455                                                 sizeof(struct llog_rec_tail));
 456
 457                 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
 458                         lustre_swab_llog_rec(rec);
 459
 460                 *cur_idx = tail->lrt_index;
 461
 462                 /* this shouldn't happen */
 463                 if (tail->lrt_index == 0) {
 464                         CERROR("Invalid llog tail at log id "LPU64"/%u offset "
 465                                LPU64"\n", loghandle->lgh_id.lgl_oid,
 466                                loghandle->lgh_id.lgl_ogen, *cur_offset);
 467                         RETURN(-EINVAL);
 468                 }
 469                 if (tail->lrt_index < next_idx)
 470                         continue;
 471
 472                 /* sanity check that the start of the new buffer is no farther
 473                  * than the record that we wanted.  This shouldn't happen. */
 474                 if (rec->lrh_index > next_idx) {
 475                         CERROR("missed desired record? %u > %u\n",
 476                                rec->lrh_index, next_idx);
 477                         RETURN(-ENOENT);
 478                 }
 479                 RETURN(0);
 480         }
 481         RETURN(-EIO);
 482 }
 483
 484 static int llog_lvfs_prev_block(const struct lu_env *env,
 485                                 struct llog_handle *loghandle,
 486                                 int prev_idx, void *buf, int len)
 487 {
 488         __u64 cur_offset;
 489         int rc;
 490         ENTRY;
 491
 492         if (len == 0 || len & (LLOG_CHUNK_SIZE - 1))
 493                 RETURN(-EINVAL);
 494
 495         CDEBUG(D_OTHER, "looking for log index %u\n", prev_idx);
 496
 497         cur_offset = LLOG_CHUNK_SIZE;
 498         llog_skip_over(&cur_offset, 0, prev_idx);
 499
 500         while (cur_offset < i_size_read(loghandle->lgh_file->f_dentry->d_inode)) {
 501                 struct llog_rec_hdr *rec;
 502                 struct llog_rec_tail *tail;
 503                 loff_t ppos;
 504
 505                 ppos = cur_offset;
 506
 507                 rc = fsfilt_read_record(loghandle->lgh_ctxt->loc_exp->exp_obd,
 508                                         loghandle->lgh_file, buf, len,
 509                                         &ppos);
 510                 if (rc) {
 511                         CERROR("Cant read llog block at log id "LPU64
 512                                "/%u offset "LPU64"\n",
 513                                loghandle->lgh_id.lgl_oid,
 514                                loghandle->lgh_id.lgl_ogen,
 515                                cur_offset);
 516                         RETURN(rc);
 517                 }
 518
 519                 /* put number of bytes read into rc to make code simpler */
 520                 rc = ppos - cur_offset;
 521                 cur_offset = ppos;
 522
 523                 if (rc == 0) /* end of file, nothing to do */
 524                         RETURN(0);
 525
 526                 if (rc < sizeof(*tail)) {
 527                         CERROR("Invalid llog block at log id "LPU64"/%u offset "
 528                                LPU64"\n", loghandle->lgh_id.lgl_oid,
 529                                loghandle->lgh_id.lgl_ogen, cur_offset);
 530                         RETURN(-EINVAL);
 531                 }
 532
 533                 tail = buf + rc - sizeof(struct llog_rec_tail);
 534
 535                 /* this shouldn't happen */
 536                 if (tail->lrt_index == 0) {
 537                         CERROR("Invalid llog tail at log id "LPU64"/%u offset "
 538                                LPU64"\n", loghandle->lgh_id.lgl_oid,
 539                                loghandle->lgh_id.lgl_ogen, cur_offset);
 540                         RETURN(-EINVAL);
 541                 }
 542                 if (le32_to_cpu(tail->lrt_index) < prev_idx)
 543                         continue;
 544
 545                 /* sanity check that the start of the new buffer is no farther
 546                  * than the record that we wanted.  This shouldn't happen. */
 547                 rec = buf;
 548                 if (le32_to_cpu(rec->lrh_index) > prev_idx) {
 549                         CERROR("missed desired record? %u > %u\n",
 550                                le32_to_cpu(rec->lrh_index), prev_idx);
 551                         RETURN(-ENOENT);
 552                 }
 553                 RETURN(0);
 554         }
 555         RETURN(-EIO);
 556 }
 557
 558 static struct file *llog_filp_open(char *dir, char *name, int flags, int mode)
 559 {
 560         char *logname;
 561         struct file *filp;
 562         int len;
 563
 564         OBD_ALLOC(logname, PATH_MAX);
 565         if (logname == NULL)
 566                 return ERR_PTR(-ENOMEM);
 567
 568         len = snprintf(logname, PATH_MAX, "%s/%s", dir, name);
 569         if (len >= PATH_MAX - 1) {
 570                 filp = ERR_PTR(-ENAMETOOLONG);
 571         } else {
 572                 filp = l_filp_open(logname, flags, mode);
 573                 if (IS_ERR(filp))
 574                         CERROR("logfile creation %s: %ld\n", logname,
 575                                PTR_ERR(filp));
 576         }
 577         OBD_FREE(logname, PATH_MAX);
 578         return filp;
 579 }
 580
 581 /* This is a callback from the llog_* functions.
 582  * Assumes caller has already pushed us into the kernel context. */
 583 static int llog_lvfs_create(const struct lu_env *env,
 584                             struct llog_ctxt *ctxt, struct llog_handle **res,
 585                             struct llog_logid *logid, char *name)
 586 {
 587         struct llog_handle *handle;
 588         struct obd_device *obd;
 589         struct l_dentry *dchild = NULL;
 590         struct obdo *oa = NULL;
 591         int rc = 0;
 592         int open_flags = O_RDWR | O_CREAT | O_LARGEFILE;
 593         ENTRY;
 594
 595         handle = llog_alloc_handle();
 596         if (handle == NULL)
 597                 RETURN(-ENOMEM);
 598         *res = handle;
 599
 600         LASSERT(ctxt);
 601         LASSERT(ctxt->loc_exp);
 602         obd = ctxt->loc_exp->exp_obd;
 603
 604         if (logid != NULL) {
 605                 dchild = obd_lvfs_fid2dentry(ctxt->loc_exp, logid->lgl_oid,
 606                                              logid->lgl_ogen, logid->lgl_oseq);
 607
 608                 if (IS_ERR(dchild)) {
 609                         rc = PTR_ERR(dchild);
 610                         CERROR("error looking up logfile "LPX64":0x%x: rc %d\n",
 611                                logid->lgl_oid, logid->lgl_ogen, rc);
 612                         GOTO(out, rc);
 613                 }
 614
 615                 if (dchild->d_inode == NULL) {
 616                         l_dput(dchild);
 617                         rc = -ENOENT;
 618                         CERROR("nonexistent log file "LPX64":"LPX64": rc %d\n",
 619                                logid->lgl_oid, logid->lgl_oseq, rc);
 620                         GOTO(out, rc);
 621                 }
 622
 623                 /* l_dentry_open will call dput(dchild) if there is an error */
 624                 handle->lgh_file = l_dentry_open(&obd->obd_lvfs_ctxt, dchild,
 625                                                     O_RDWR | O_LARGEFILE);
 626                 if (IS_ERR(handle->lgh_file)) {
 627                         rc = PTR_ERR(handle->lgh_file);
 628                         CERROR("error opening logfile "LPX64"0x%x: rc %d\n",
 629                                logid->lgl_oid, logid->lgl_ogen, rc);
 630                         GOTO(out, rc);
 631                 }
 632
 633                 /* assign the value of lgh_id for handle directly */
 634                 handle->lgh_id = *logid;
 635
 636         } else if (name) {
 637                 handle->lgh_file = llog_filp_open(MOUNT_CONFIGS_DIR,
 638                                                   name, open_flags, 0644);
 639                 if (IS_ERR(handle->lgh_file))
 640                         GOTO(out, rc = PTR_ERR(handle->lgh_file));
 641
 642                 handle->lgh_id.lgl_oseq = 1;
 643                 handle->lgh_id.lgl_oid =
 644                         handle->lgh_file->f_dentry->d_inode->i_ino;
 645                 handle->lgh_id.lgl_ogen =
 646                         handle->lgh_file->f_dentry->d_inode->i_generation;
 647         } else {
 648                 OBDO_ALLOC(oa);
 649                 if (oa == NULL)
 650                         GOTO(out, rc = -ENOMEM);
 651
 652                 oa->o_seq = FID_SEQ_LLOG;
 653                 oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLGROUP;
 654
 655                 rc = obd_create(NULL, ctxt->loc_exp, oa, NULL, NULL);
 656                 if (rc)
 657                         GOTO(out, rc);
 658
 659                 /* FIXME: rationalize the misuse of o_generation in
 660                  *        this API along with mds_obd_{create,destroy}.
 661                  *        Hopefully it is only an internal API issue. */
 662 #define o_generation o_parent_oid
 663                 dchild = obd_lvfs_fid2dentry(ctxt->loc_exp, oa->o_id,
 664                                              oa->o_generation, oa->o_seq);
 665
 666                 if (IS_ERR(dchild))
 667                         GOTO(out, rc = PTR_ERR(dchild));
 668
 669                 handle->lgh_file = l_dentry_open(&obd->obd_lvfs_ctxt, dchild,
 670                                                  open_flags);
 671                 if (IS_ERR(handle->lgh_file))
 672                         GOTO(out, rc = PTR_ERR(handle->lgh_file));
 673
 674                 handle->lgh_id.lgl_oseq = oa->o_seq;
 675                 handle->lgh_id.lgl_oid = oa->o_id;
 676                 handle->lgh_id.lgl_ogen = oa->o_generation;
 677         }
 678
 679         handle->lgh_ctxt = ctxt;
 680 out:
 681         if (rc)
 682                 llog_free_handle(handle);
 683
 684         if (oa)
 685                 OBDO_FREE(oa);
 686         RETURN(rc);
 687 }
 688
 689 static int llog_lvfs_close(const struct lu_env *env,
 690                            struct llog_handle *handle)
 691 {
 692         int rc;
 693         ENTRY;
 694
 695         rc = filp_close(handle->lgh_file, 0);
 696         if (rc)
 697                 CERROR("error closing log: rc %d\n", rc);
 698         RETURN(rc);
 699 }
 700
 701 static int llog_lvfs_destroy(const struct lu_env *env,
 702                              struct llog_handle *handle)
 703 {
 704         struct dentry *fdentry;
 705         struct obdo *oa;
 706         struct obd_device *obd = handle->lgh_ctxt->loc_exp->exp_obd;
 707         char *dir;
 708         void *th;
 709         struct inode *inode;
 710         int rc, rc1;
 711         ENTRY;
 712
 713         dir = MOUNT_CONFIGS_DIR;
 714
 715         fdentry = handle->lgh_file->f_dentry;
 716         inode = fdentry->d_parent->d_inode;
 717         if (strcmp(fdentry->d_parent->d_name.name, dir) == 0) {
 718                 struct lvfs_run_ctxt saved;
 719                 struct vfsmount *mnt = mntget(handle->lgh_file->f_vfsmnt);
 720
 721                 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 722                 dget(fdentry);
 723                 rc = llog_lvfs_close(env, handle);
 724                 if (rc == 0) {
 725                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
 726                         rc = ll_vfs_unlink(inode, fdentry, mnt);
 727                         mutex_unlock(&inode->i_mutex);
 728                 }
 729                 mntput(mnt);
 730
 731                 dput(fdentry);
 732                 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 733                 RETURN(rc);
 734         }
 735
 736         OBDO_ALLOC(oa);
 737         if (oa == NULL)
 738                 RETURN(-ENOMEM);
 739
 740         oa->o_id = handle->lgh_id.lgl_oid;
 741         oa->o_seq = handle->lgh_id.lgl_oseq;
 742         oa->o_generation = handle->lgh_id.lgl_ogen;
 743 #undef o_generation
 744         oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP | OBD_MD_FLGENER;
 745
 746         rc = llog_lvfs_close(env, handle);
 747         if (rc)
 748                 GOTO(out, rc);
 749
 750         th = fsfilt_start_log(obd, inode, FSFILT_OP_UNLINK, NULL, 1);
 751         if (IS_ERR(th)) {
 752                 CERROR("fsfilt_start failed: %ld\n", PTR_ERR(th));
 753                 GOTO(out, rc = PTR_ERR(th));
 754         }
 755
 756         rc = obd_destroy(NULL, handle->lgh_ctxt->loc_exp, oa,
 757                          NULL, NULL, NULL, NULL);
 758
 759         rc1 = fsfilt_commit(obd, inode, th, 0);
 760         if (rc == 0 && rc1 != 0)
 761                 rc = rc1;
 762  out:
 763         OBDO_FREE(oa);
 764         RETURN(rc);
 765 }
 766
 767 /* reads the catalog list */
 768 int llog_get_cat_list(struct obd_device *disk_obd,
 769                       char *name, int idx, int count, struct llog_catid *idarray)
 770 {
 771         struct lvfs_run_ctxt saved;
 772         struct l_file *file;
 773         int rc, rc1 = 0;
 774         int size = sizeof(*idarray) * count;
 775         loff_t off = idx *  sizeof(*idarray);
 776         ENTRY;
 777
 778         if (!count)
 779                 RETURN(0);
 780
 781         push_ctxt(&saved, &disk_obd->obd_lvfs_ctxt, NULL);
 782         file = filp_open(name, O_RDWR | O_CREAT | O_LARGEFILE, 0700);
 783         if (!file || IS_ERR(file)) {
 784                 rc = PTR_ERR(file);
 785                 CERROR("OBD filter: cannot open/create %s: rc = %d\n",
 786                        name, rc);
 787                 GOTO(out, rc);
 788         }
 789
 790         if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
 791                 CERROR("%s is not a regular file!: mode = %o\n", name,
 792                        file->f_dentry->d_inode->i_mode);
 793                 GOTO(out, rc = -ENOENT);
 794         }
 795
 796         CDEBUG(D_CONFIG, "cat list: disk size=%d, read=%d\n",
 797                (int)i_size_read(file->f_dentry->d_inode), size);
 798
 799         /* read for new ost index or for empty file */
 800         memset(idarray, 0, size);
 801         if (i_size_read(file->f_dentry->d_inode) < off)
 802                 GOTO(out, rc = 0);
 803
 804         rc = fsfilt_read_record(disk_obd, file, idarray, size, &off);
 805         if (rc) {
 806                 CERROR("OBD filter: error reading %s: rc %d\n", name, rc);
 807                 GOTO(out, rc);
 808         }
 809
 810         EXIT;
 811  out:
 812         pop_ctxt(&saved, &disk_obd->obd_lvfs_ctxt, NULL);
 813         if (file && !IS_ERR(file))
 814                 rc1 = filp_close(file, 0);
 815         if (rc == 0)
 816                 rc = rc1;
 817         return rc;
 818 }
 819 EXPORT_SYMBOL(llog_get_cat_list);
 820
 821 /* writes the cat list */
 822 int llog_put_cat_list(struct obd_device *disk_obd,
 823                       char *name, int idx, int count, struct llog_catid *idarray)
 824 {
 825         struct lvfs_run_ctxt saved;
 826         struct l_file *file;
 827         int rc, rc1 = 0;
 828         int size = sizeof(*idarray) * count;
 829         loff_t off = idx * sizeof(*idarray);
 830
 831         if (!count)
 832                 GOTO(out1, rc = 0);
 833
 834         push_ctxt(&saved, &disk_obd->obd_lvfs_ctxt, NULL);
 835         file = filp_open(name, O_RDWR | O_CREAT | O_LARGEFILE, 0700);
 836         if (!file || IS_ERR(file)) {
 837                 rc = PTR_ERR(file);
 838                 CERROR("OBD filter: cannot open/create %s: rc = %d\n",
 839                        name, rc);
 840                 GOTO(out, rc);
 841         }
 842
 843         if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
 844                 CERROR("%s is not a regular file!: mode = %o\n", name,
 845                        file->f_dentry->d_inode->i_mode);
 846                 GOTO(out, rc = -ENOENT);
 847         }
 848
 849         rc = fsfilt_write_record(disk_obd, file, idarray, size, &off, 1);
 850         if (rc) {
 851                 CDEBUG(D_INODE,"OBD filter: error writeing %s: rc %d\n",
 852                        name, rc);
 853                 GOTO(out, rc);
 854         }
 855
 856 out:
 857         pop_ctxt(&saved, &disk_obd->obd_lvfs_ctxt, NULL);
 858         if (file && !IS_ERR(file))
 859                 rc1 = filp_close(file, 0);
 860
 861         if (rc == 0)
 862                 rc = rc1;
 863 out1:
 864         RETURN(rc);
 865 }
 866 EXPORT_SYMBOL(llog_put_cat_list);
 867
 868 struct llog_operations llog_lvfs_ops = {
 869         lop_write_rec:   llog_lvfs_write_rec,
 870         lop_next_block:  llog_lvfs_next_block,
 871         lop_prev_block:  llog_lvfs_prev_block,
 872         lop_read_header: llog_lvfs_read_header,
 873         lop_create:      llog_lvfs_create,
 874         lop_destroy:     llog_lvfs_destroy,
 875         lop_close:       llog_lvfs_close,
 876         //        lop_cancel: llog_lvfs_cancel,
 877 };
 878
 879 EXPORT_SYMBOL(llog_lvfs_ops);
 880
 881 #else /* !__KERNEL__ */
 882
 883 static int llog_lvfs_read_header(const struct lu_env *env,
 884                                  struct llog_handle *handle)
 885 {
 886         LBUG();
 887         return 0;
 888 }
 889
 890 static int llog_lvfs_write_rec(const struct lu_env *env,
 891                                struct llog_handle *loghandle,
 892                                struct llog_rec_hdr *rec,
 893                                struct llog_cookie *reccookie, int cookiecount,
 894                                void *buf, int idx)
 895 {
 896         LBUG();
 897         return 0;
 898 }
 899
 900 static int llog_lvfs_next_block(const struct lu_env *env,
 901                                 struct llog_handle *loghandle, int *cur_idx,
 902                                 int next_idx, __u64 *cur_offset, void *buf,
 903                                 int len)
 904 {
 905         LBUG();
 906         return 0;
 907 }
 908
 909 static int llog_lvfs_prev_block(const struct lu_env *env,
 910                                 struct llog_handle *loghandle,
 911                                 int prev_idx, void *buf, int len)
 912 {
 913         LBUG();
 914         return 0;
 915 }
 916
 917 static int llog_lvfs_create(const struct lu_env *env,
 918                             struct llog_ctxt *ctxt, struct llog_handle **res,
 919                             struct llog_logid *logid, char *name)
 920 {
 921         LBUG();
 922         return 0;
 923 }
 924
 925 static int llog_lvfs_close(const struct lu_env *env,
 926                            struct llog_handle *handle)
 927 {
 928         LBUG();
 929         return 0;
 930 }
 931
 932 static int llog_lvfs_destroy(const struct lu_env *env,
 933                              struct llog_handle *handle)
 934 {
 935         LBUG();
 936         return 0;
 937 }
 938
 939 int llog_get_cat_list(struct obd_device *disk_obd,
 940                       char *name, int idx, int count, struct llog_catid *idarray)
 941 {
 942         LBUG();
 943         return 0;
 944 }
 945
 946 int llog_put_cat_list(struct obd_device *disk_obd,
 947                       char *name, int idx, int count, struct llog_catid *idarray)
 948 {
 949         LBUG();
 950         return 0;
 951 }
 952
 953 struct llog_operations llog_lvfs_ops = {
 954         lop_write_rec:   llog_lvfs_write_rec,
 955         lop_next_block:  llog_lvfs_next_block,
 956         lop_prev_block:  llog_lvfs_prev_block,
 957         lop_read_header: llog_lvfs_read_header,
 958         lop_create:      llog_lvfs_create,
 959         lop_destroy:     llog_lvfs_destroy,
 960         lop_close:       llog_lvfs_close,
 961 //        lop_cancel:      llog_lvfs_cancel,
 962 };
 963 #endif