4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
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.
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).
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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2014 Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 * lustre/obdclass/llog_osd.c
35 * Low level llog routines on top of OSD API
37 * This file provides set of methods for llog operations on top of
38 * dt_device. It contains all supported llog_operations interfaces and
39 * supplimental functions.
41 * Author: Alexey Zhuravlev <alexey.zhuravlev@intel.com>
42 * Author: Mikhail Pershin <mike.pershin@intel.com>
45 #define DEBUG_SUBSYSTEM S_LOG
48 #include <obd_class.h>
49 #include <lustre_fid.h>
50 #include <dt_object.h>
52 #include "llog_internal.h"
53 #include "local_storage.h"
56 * Implementation of the llog_operations::lop_declare_create
58 * This function is a wrapper over local_storage API function
59 * local_object_declare_create().
61 * \param[in] env execution environment
62 * \param[in] los local_storage for bottom storage device
63 * \param[in] o dt_object to create
64 * \param[in] th current transaction handle
66 * \retval 0 on successful declaration of the new object
67 * \retval negative error if declaration was failed
69 static int llog_osd_declare_new_object(const struct lu_env *env,
70 struct local_oid_storage *los,
74 struct llog_thread_info *lgi = llog_info(env);
76 lgi->lgi_attr.la_valid = LA_MODE;
77 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
78 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
80 return local_object_declare_create(env, los, o, &lgi->lgi_attr,
85 * Implementation of the llog_operations::lop_create
87 * This function is a wrapper over local_storage API function
88 * local_object_create().
90 * \param[in] env execution environment
91 * \param[in] los local_storage for bottom storage device
92 * \param[in] o dt_object to create
93 * \param[in] th current transaction handle
95 * \retval 0 on successful creation of the new object
96 * \retval negative error if creation was failed
98 static int llog_osd_create_new_object(const struct lu_env *env,
99 struct local_oid_storage *los,
103 struct llog_thread_info *lgi = llog_info(env);
105 lgi->lgi_attr.la_valid = LA_MODE;
106 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
107 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
109 return local_object_create(env, los, o, &lgi->lgi_attr,
114 * Write a padding record to the llog
116 * This function writes a padding record to the end of llog. That may
117 * be needed if llog contains records of variable size, e.g. config logs
119 * The padding record just aligns llog to the LLOG_CHUNK_SIZE boundary if
120 * the current record doesn't fit in the remaining space.
122 * It allocates full length to avoid two separate writes for header and tail.
123 * Such 2-steps scheme needs extra protection and complex error handling.
125 * \param[in] env execution environment
126 * \param[in] o dt_object to create
127 * \param[in,out] off pointer to the padding start offset
128 * \param[in] len padding length
129 * \param[in] index index of the padding record in a llog
130 * \param[in] th current transaction handle
132 * \retval 0 on successful padding write
133 * \retval negative error if write failed
135 static int llog_osd_pad(const struct lu_env *env, struct dt_object *o,
136 loff_t *off, int len, int index, struct thandle *th)
138 struct llog_thread_info *lgi = llog_info(env);
139 struct llog_rec_hdr *rec;
140 struct llog_rec_tail *tail;
147 LASSERT(len >= LLOG_MIN_REC_SIZE && (len & 0x7) == 0);
154 rec->lrh_index = index;
155 rec->lrh_type = LLOG_PAD_MAGIC;
157 tail = rec_tail(rec);
159 tail->lrt_index = index;
161 lgi->lgi_buf.lb_buf = rec;
162 lgi->lgi_buf.lb_len = len;
163 rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
165 CERROR("%s: error writing padding record: rc = %d\n",
166 o->do_lu.lo_dev->ld_obd->obd_name, rc);
173 * Implementation of the llog_operations::lop_read_header
175 * This function reads the current llog header from the bottom storage
178 * \param[in] env execution environment
179 * \param[in] handle llog handle of the current llog
181 * \retval 0 on successful header read
182 * \retval negative error if read failed
184 static int llog_osd_read_header(const struct lu_env *env,
185 struct llog_handle *handle)
187 struct llog_rec_hdr *llh_hdr;
189 struct llog_thread_info *lgi;
190 enum llog_flag flags;
195 LASSERT(sizeof(*handle->lgh_hdr) == LLOG_CHUNK_SIZE);
200 lgi = llog_info(env);
202 rc = dt_attr_get(env, o, &lgi->lgi_attr);
206 LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
208 if (lgi->lgi_attr.la_size == 0) {
209 CDEBUG(D_HA, "not reading header from 0-byte log\n");
213 flags = handle->lgh_hdr->llh_flags;
216 lgi->lgi_buf.lb_buf = handle->lgh_hdr;
217 lgi->lgi_buf.lb_len = LLOG_CHUNK_SIZE;
219 rc = dt_record_read(env, o, &lgi->lgi_buf, &lgi->lgi_off);
221 CERROR("%s: error reading log header from "DFID": rc = %d\n",
222 o->do_lu.lo_dev->ld_obd->obd_name,
223 PFID(lu_object_fid(&o->do_lu)), rc);
227 llh_hdr = &handle->lgh_hdr->llh_hdr;
228 if (LLOG_REC_HDR_NEEDS_SWABBING(llh_hdr))
229 lustre_swab_llog_hdr(handle->lgh_hdr);
231 if (llh_hdr->lrh_type != LLOG_HDR_MAGIC) {
232 CERROR("%s: bad log %s "DFID" header magic: %#x "
233 "(expected %#x)\n", o->do_lu.lo_dev->ld_obd->obd_name,
234 handle->lgh_name ? handle->lgh_name : "",
235 PFID(lu_object_fid(&o->do_lu)),
236 llh_hdr->lrh_type, LLOG_HDR_MAGIC);
238 } else if (llh_hdr->lrh_len != LLOG_CHUNK_SIZE) {
239 CERROR("%s: incorrectly sized log %s "DFID" header: "
240 "%#x (expected %#x)\n"
241 "you may need to re-run lconf --write_conf.\n",
242 o->do_lu.lo_dev->ld_obd->obd_name,
243 handle->lgh_name ? handle->lgh_name : "",
244 PFID(lu_object_fid(&o->do_lu)),
245 llh_hdr->lrh_len, LLOG_CHUNK_SIZE);
249 handle->lgh_hdr->llh_flags |= (flags & LLOG_F_EXT_MASK);
250 handle->lgh_last_idx = handle->lgh_hdr->llh_tail.lrt_index;
256 * Implementation of the llog_operations::lop_declare_write
258 * This function declares the new record write.
260 * \param[in] env execution environment
261 * \param[in] loghandle llog handle of the current llog
262 * \param[in] rec llog record header. This is a real header of the full
263 * llog record to write. This is the beginning of buffer
264 * to write, the length of buffer is stored in
266 * \param[in] idx index of the llog record. If \a idx == -1 then this is
267 * append case, otherwise \a idx is the index of record
269 * \param[in] th current transaction handle
271 * \retval 0 on successful declaration
272 * \retval negative error if declaration failed
274 static int llog_osd_declare_write_rec(const struct lu_env *env,
275 struct llog_handle *loghandle,
276 struct llog_rec_hdr *rec,
277 int idx, struct thandle *th)
279 struct llog_thread_info *lgi = llog_info(env);
289 LASSERT(rec->lrh_len <= LLOG_CHUNK_SIZE);
291 o = loghandle->lgh_obj;
294 lgi->lgi_buf.lb_len = sizeof(struct llog_log_hdr);
295 lgi->lgi_buf.lb_buf = NULL;
296 /* each time we update header */
297 rc = dt_declare_record_write(env, o, &lgi->lgi_buf, 0,
299 if (rc || idx == 0) /* if error or just header */
303 * the pad record can be inserted so take into account double
306 lgi->lgi_buf.lb_len = rec->lrh_len * 2;
307 lgi->lgi_buf.lb_buf = NULL;
308 /* XXX: implement declared window or multi-chunks approach */
309 rc = dt_declare_record_write(env, o, &lgi->lgi_buf, -1, th);
315 * Implementation of the llog_operations::lop_write
317 * This function writes the new record in the llog or modify the existed one.
319 * \param[in] env execution environment
320 * \param[in] loghandle llog handle of the current llog
321 * \param[in] rec llog record header. This is a real header of
322 * the full llog record to write. This is
323 * the beginning of buffer to write, the length
324 * of buffer is stored in \a rec::lrh_len
325 * \param[out] reccookie pointer to the cookie to return back if needed.
326 * It is used for further cancel of this llog
328 * \param[in] idx index of the llog record. If \a idx == -1 then
329 * this is append case, otherwise \a idx is
330 * the index of record to modify
331 * \param[in] th current transaction handle
333 * \retval 0 on successful write && \a reccookie == NULL
334 * 1 on successful write && \a reccookie != NULL
335 * \retval negative error if write failed
337 static int llog_osd_write_rec(const struct lu_env *env,
338 struct llog_handle *loghandle,
339 struct llog_rec_hdr *rec,
340 struct llog_cookie *reccookie,
341 int idx, struct thandle *th)
343 struct llog_thread_info *lgi = llog_info(env);
344 struct llog_log_hdr *llh;
345 int reclen = rec->lrh_len;
347 struct llog_rec_tail *lrt;
354 llh = loghandle->lgh_hdr;
356 o = loghandle->lgh_obj;
360 CDEBUG(D_OTHER, "new record %x to "DFID"\n",
361 rec->lrh_type, PFID(lu_object_fid(&o->do_lu)));
363 /* record length should not bigger than LLOG_CHUNK_SIZE */
364 if (reclen > LLOG_CHUNK_SIZE)
367 rc = dt_attr_get(env, o, &lgi->lgi_attr);
372 * The modification case.
373 * If idx set then the record with that index must be modified.
374 * There are three cases possible:
375 * 1) the common case is the llog header update (idx == 0)
376 * 2) the llog record modification during llog process.
377 * This is indicated by the \a loghandle::lgh_cur_idx > 0.
378 * In that case the \a loghandle::lgh_cur_offset
379 * 3) otherwise this is assumed that llog consist of records of
380 * fixed size, i.e. catalog. The llog header must has llh_size
381 * field equal to record size. The record offset is calculated
382 * just by /a idx value
384 * During modification we don't need extra header update because
385 * the bitmap and record count are not changed. The record header
386 * and tail remains the same too.
388 if (idx != LLOG_NEXT_IDX) {
389 /* llog can be empty only when first record is being written */
390 LASSERT(ergo(idx > 0, lgi->lgi_attr.la_size > 0));
392 if (!ext2_test_bit(idx, llh->llh_bitmap)) {
393 CERROR("%s: modify unset record %u\n",
394 o->do_lu.lo_dev->ld_obd->obd_name, idx);
398 if (idx != rec->lrh_index) {
399 CERROR("%s: modify index mismatch %d %u\n",
400 o->do_lu.lo_dev->ld_obd->obd_name, idx,
405 if (idx == LLOG_HEADER_IDX) {
406 /* llog header update */
407 LASSERT(reclen == sizeof(struct llog_log_hdr));
408 LASSERT(rec == &llh->llh_hdr);
411 lgi->lgi_buf.lb_len = reclen;
412 lgi->lgi_buf.lb_buf = rec;
413 rc = dt_record_write(env, o, &lgi->lgi_buf,
416 } else if (loghandle->lgh_cur_idx > 0) {
418 * The lgh_cur_offset can be used only if index is
421 if (idx != loghandle->lgh_cur_idx) {
422 CERROR("%s: modify index mismatch %d %d\n",
423 o->do_lu.lo_dev->ld_obd->obd_name, idx,
424 loghandle->lgh_cur_idx);
428 lgi->lgi_off = loghandle->lgh_cur_offset;
429 CDEBUG(D_OTHER, "modify record "DOSTID": idx:%d, "
430 "len:%u offset %llu\n",
431 POSTID(&loghandle->lgh_id.lgl_oi), idx,
432 rec->lrh_len, (long long)lgi->lgi_off);
433 } else if (llh->llh_size > 0) {
434 if (llh->llh_size != rec->lrh_len) {
435 CERROR("%s: wrong record size, llh_size is %u"
436 " but record size is %u\n",
437 o->do_lu.lo_dev->ld_obd->obd_name,
438 llh->llh_size, rec->lrh_len);
441 lgi->lgi_off = sizeof(*llh) + (idx - 1) * reclen;
443 /* This can be result of lgh_cur_idx is not set during
444 * llog processing or llh_size is not set to proper
445 * record size for fixed records llog. Therefore it is
446 * impossible to get record offset. */
447 CERROR("%s: can't get record offset, idx:%d, "
448 "len:%u.\n", o->do_lu.lo_dev->ld_obd->obd_name,
453 /* update only data, header and tail remain the same */
454 lgi->lgi_off += sizeof(struct llog_rec_hdr);
455 lgi->lgi_buf.lb_len = REC_DATA_LEN(rec);
456 lgi->lgi_buf.lb_buf = REC_DATA(rec);
457 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
458 if (rc == 0 && reccookie) {
459 reccookie->lgc_lgl = loghandle->lgh_id;
460 reccookie->lgc_index = idx;
468 * The most common case of using llog. The new index is assigned to
469 * the new record, new bit is set in llog bitmap and llog count is
472 * Make sure that records don't cross a chunk boundary, so we can
473 * process them page-at-a-time if needed. If it will cross a chunk
474 * boundary, write in a fake (but referenced) entry to pad the chunk.
476 LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
477 lgi->lgi_off = lgi->lgi_attr.la_size;
478 left = LLOG_CHUNK_SIZE - (lgi->lgi_off & (LLOG_CHUNK_SIZE - 1));
479 /* NOTE: padding is a record, but no bit is set */
480 if (left != 0 && left != reclen &&
481 left < (reclen + LLOG_MIN_REC_SIZE)) {
482 index = loghandle->lgh_last_idx + 1;
483 rc = llog_osd_pad(env, o, &lgi->lgi_off, left, index, th);
486 loghandle->lgh_last_idx++; /* for pad rec */
488 /* if it's the last idx in log file, then return -ENOSPC */
489 if (loghandle->lgh_last_idx >= LLOG_BITMAP_SIZE(llh) - 1)
492 /* increment the last_idx along with llh_tail index, they should
493 * be equal for a llog lifetime */
494 loghandle->lgh_last_idx++;
495 index = loghandle->lgh_last_idx;
496 llh->llh_tail.lrt_index = index;
498 * NB: the caller should make sure only 1 process access
499 * the lgh_last_idx, e.g. append should be exclusive.
500 * Otherwise it might hit the assert.
502 LASSERT(index < LLOG_BITMAP_SIZE(llh));
503 rec->lrh_index = index;
505 lrt->lrt_len = rec->lrh_len;
506 lrt->lrt_index = rec->lrh_index;
508 /* the lgh_hdr_lock protects llog header data from concurrent
509 * update/cancel, the llh_count and llh_bitmap are protected */
510 down_write(&loghandle->lgh_hdr_lock);
511 if (ext2_set_bit(index, llh->llh_bitmap)) {
512 CERROR("%s: index %u already set in log bitmap\n",
513 o->do_lu.lo_dev->ld_obd->obd_name, index);
514 up_write(&loghandle->lgh_hdr_lock);
515 LBUG(); /* should never happen */
519 /* XXX It is a bit tricky here, if the log object is local,
520 * we do not need lock during write here, because if there is
521 * race, the transaction(jbd2, what about ZFS?) will make sure the
522 * conflicts will all committed in the same transaction group.
523 * But for remote object, we need lock the whole process, so to
524 * set the version of the remote transaction to make sure they
525 * are being sent in order. (see osp_md_write()) */
526 if (!dt_object_remote(o))
527 up_write(&loghandle->lgh_hdr_lock);
529 if (lgi->lgi_attr.la_size == 0) {
531 lgi->lgi_buf.lb_len = llh->llh_hdr.lrh_len;
532 lgi->lgi_buf.lb_buf = &llh->llh_hdr;
533 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
535 GOTO(out_remote_unlock, rc);
537 /* Note: If this is not initialization (size == 0), then do not
538 * write the whole header (8k bytes), only update header/tail
539 * and bits needs to be updated. Because this update might be
540 * part of cross-MDT operation, which needs to write these
541 * updates into the update log(32KB limit) and also pack inside
542 * the RPC (1MB limit), if we write 8K for each operation, which
543 * will cost a lot space, and keep us adding more updates to one
545 lgi->lgi_off = offsetof(typeof(*llh), llh_count);
546 lgi->lgi_buf.lb_len = sizeof(llh->llh_count);
547 lgi->lgi_buf.lb_buf = &llh->llh_count;
548 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
550 GOTO(out_remote_unlock, rc);
552 lgi->lgi_off = offsetof(typeof(*llh),
553 llh_bitmap[index / (sizeof(*llh->llh_bitmap) * 8)]);
554 lgi->lgi_buf.lb_len = sizeof(*llh->llh_bitmap);
555 lgi->lgi_buf.lb_buf =
556 &llh->llh_bitmap[index/(sizeof(*llh->llh_bitmap)*8)];
557 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
559 GOTO(out_remote_unlock, rc);
561 lgi->lgi_off = offsetof(typeof(*llh), llh_tail);
562 lgi->lgi_buf.lb_len = sizeof(llh->llh_tail);
563 lgi->lgi_buf.lb_buf = &llh->llh_tail;
564 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
566 GOTO(out_remote_unlock, rc);
570 /* unlock here for remote object */
571 if (dt_object_remote(o))
572 up_write(&loghandle->lgh_hdr_lock);
576 rc = dt_attr_get(env, o, &lgi->lgi_attr);
580 LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
581 lgi->lgi_off = max_t(__u64, lgi->lgi_attr.la_size, lgi->lgi_off);
582 lgi->lgi_buf.lb_len = reclen;
583 lgi->lgi_buf.lb_buf = rec;
584 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
588 CDEBUG(D_OTHER, "added record "DOSTID": idx: %u, %u\n",
589 POSTID(&loghandle->lgh_id.lgl_oi), index, rec->lrh_len);
590 if (reccookie != NULL) {
591 reccookie->lgc_lgl = loghandle->lgh_id;
592 reccookie->lgc_index = index;
593 if ((rec->lrh_type == MDS_UNLINK_REC) ||
594 (rec->lrh_type == MDS_SETATTR64_REC))
595 reccookie->lgc_subsys = LLOG_MDS_OST_ORIG_CTXT;
596 else if (rec->lrh_type == OST_SZ_REC)
597 reccookie->lgc_subsys = LLOG_SIZE_ORIG_CTXT;
599 reccookie->lgc_subsys = -1;
604 /* cleanup llog for error case */
605 down_write(&loghandle->lgh_hdr_lock);
606 ext2_clear_bit(index, llh->llh_bitmap);
608 up_write(&loghandle->lgh_hdr_lock);
610 /* restore llog last_idx */
611 loghandle->lgh_last_idx--;
612 llh->llh_tail.lrt_index = loghandle->lgh_last_idx;
618 * We can skip reading at least as many log blocks as the number of
619 * minimum sized log records we are skipping. If it turns out
620 * that we are not far enough along the log (because the
621 * actual records are larger than minimum size) we just skip
624 static inline void llog_skip_over(__u64 *off, int curr, int goal)
628 *off = (*off + (goal - curr - 1) * LLOG_MIN_REC_SIZE) &
629 ~(LLOG_CHUNK_SIZE - 1);
633 * Remove optional fields that the client doesn't expect.
634 * This is typically in order to ensure compatibility with older clients.
635 * It is assumed that since we exclusively remove fields, the block will be
636 * big enough to handle the remapped records. It is also assumed that records
637 * of a block have the same format (i.e.: the same features enabled).
639 * \param[in,out] hdr Header of the block of records to remap.
640 * \param[in,out] last_hdr Last header, don't read past this point.
641 * \param[in] flags Flags describing the fields to keep.
643 static void changelog_block_trim_ext(struct llog_rec_hdr *hdr,
644 struct llog_rec_hdr *last_hdr,
645 enum changelog_rec_flags flags)
647 if (hdr->lrh_type != CHANGELOG_REC)
651 struct changelog_rec *rec = (struct changelog_rec *)(hdr + 1);
653 changelog_remap_rec(rec, rec->cr_flags & flags);
654 hdr = llog_rec_hdr_next(hdr);
655 } while ((char *)hdr <= (char *)last_hdr);
659 * Implementation of the llog_operations::lop_next_block
661 * This function finds the the next llog block to return which contains
662 * record with required index. It is main part of llog processing.
664 * \param[in] env execution environment
665 * \param[in] loghandle llog handle of the current llog
666 * \param[in,out] cur_idx index preceeding cur_offset
667 * \param[in] next_idx target index to find
668 * \param[in,out] cur_offset furtherst point read in the file
669 * \param[in] buf pointer to data buffer to fill
670 * \param[in] len required len to read, it is
671 * LLOG_CHUNK_SIZE usually.
673 * \retval 0 on successful buffer read
674 * \retval negative value on error
676 static int llog_osd_next_block(const struct lu_env *env,
677 struct llog_handle *loghandle, int *cur_idx,
678 int next_idx, __u64 *cur_offset, void *buf,
681 struct llog_thread_info *lgi = llog_info(env);
683 struct dt_device *dt;
691 if (len == 0 || len & (LLOG_CHUNK_SIZE - 1))
694 CDEBUG(D_OTHER, "looking for log index %u (cur idx %u off "LPU64")\n",
695 next_idx, *cur_idx, *cur_offset);
698 LASSERT(loghandle->lgh_ctxt);
700 o = loghandle->lgh_obj;
702 LASSERT(dt_object_exists(o));
703 dt = lu2dt_dev(o->do_lu.lo_dev);
706 rc = dt_attr_get(env, o, &lgi->lgi_attr);
710 while (*cur_offset < lgi->lgi_attr.la_size) {
711 struct llog_rec_hdr *rec, *last_rec;
712 struct llog_rec_tail *tail;
714 llog_skip_over(cur_offset, *cur_idx, next_idx);
716 /* read up to next LLOG_CHUNK_SIZE block */
717 lgi->lgi_buf.lb_len = LLOG_CHUNK_SIZE -
718 (*cur_offset & (LLOG_CHUNK_SIZE - 1));
719 lgi->lgi_buf.lb_buf = buf;
721 rc = dt_read(env, o, &lgi->lgi_buf, cur_offset);
723 CERROR("%s: can't read llog block from log "DFID
724 " offset "LPU64": rc = %d\n",
725 o->do_lu.lo_dev->ld_obd->obd_name,
726 PFID(lu_object_fid(&o->do_lu)), *cur_offset,
732 /* signal the end of the valid buffer to
734 memset(buf + rc, 0, len - rc);
737 if (rc == 0) /* end of file, nothing to do */
740 if (rc < sizeof(*tail)) {
741 CERROR("%s: invalid llog block at log id "DOSTID"/%u "
743 o->do_lu.lo_dev->ld_obd->obd_name,
744 POSTID(&loghandle->lgh_id.lgl_oi),
745 loghandle->lgh_id.lgl_ogen, *cur_offset);
746 GOTO(out, rc = -EINVAL);
750 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
751 lustre_swab_llog_rec(rec);
753 tail = (struct llog_rec_tail *)((char *)buf + rc -
754 sizeof(struct llog_rec_tail));
755 /* get the last record in block */
756 last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
759 if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
760 lustre_swab_llog_rec(last_rec);
761 LASSERT(last_rec->lrh_index == tail->lrt_index);
763 *cur_idx = tail->lrt_index;
765 /* this shouldn't happen */
766 if (tail->lrt_index == 0) {
767 CERROR("%s: invalid llog tail at log id "DOSTID"/%u "
769 o->do_lu.lo_dev->ld_obd->obd_name,
770 POSTID(&loghandle->lgh_id.lgl_oi),
771 loghandle->lgh_id.lgl_ogen, *cur_offset);
772 GOTO(out, rc = -EINVAL);
774 if (tail->lrt_index < next_idx)
777 /* sanity check that the start of the new buffer is no farther
778 * than the record that we wanted. This shouldn't happen. */
779 if (rec->lrh_index > next_idx) {
780 CERROR("%s: missed desired record? %u > %u\n",
781 o->do_lu.lo_dev->ld_obd->obd_name,
782 rec->lrh_index, next_idx);
783 GOTO(out, rc = -ENOENT);
786 /* Trim unsupported extensions for compat w/ older clients */
787 if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_JOBID))
788 changelog_block_trim_ext(rec, last_rec,
789 CLF_VERSION | CLF_RENAME);
793 GOTO(out, rc = -EIO);
799 * Implementation of the llog_operations::lop_prev_block
801 * This function finds the llog block to return which contains
802 * record with required index but in reverse order - from end of llog
804 * It is main part of reverse llog processing.
806 * \param[in] env execution environment
807 * \param[in] loghandle llog handle of the current llog
808 * \param[in] prev_idx target index to find
809 * \param[in] buf pointer to data buffer to fill
810 * \param[in] len required len to read, it is LLOG_CHUNK_SIZE usually.
812 * \retval 0 on successful buffer read
813 * \retval negative value on error
815 static int llog_osd_prev_block(const struct lu_env *env,
816 struct llog_handle *loghandle,
817 int prev_idx, void *buf, int len)
819 struct llog_thread_info *lgi = llog_info(env);
821 struct dt_device *dt;
827 if (len == 0 || len & (LLOG_CHUNK_SIZE - 1))
830 CDEBUG(D_OTHER, "looking for log index %u\n", prev_idx);
833 LASSERT(loghandle->lgh_ctxt);
835 o = loghandle->lgh_obj;
837 LASSERT(dt_object_exists(o));
838 dt = lu2dt_dev(o->do_lu.lo_dev);
841 cur_offset = LLOG_CHUNK_SIZE;
842 llog_skip_over(&cur_offset, 0, prev_idx);
844 rc = dt_attr_get(env, o, &lgi->lgi_attr);
848 while (cur_offset < lgi->lgi_attr.la_size) {
849 struct llog_rec_hdr *rec, *last_rec;
850 struct llog_rec_tail *tail;
852 lgi->lgi_buf.lb_len = len;
853 lgi->lgi_buf.lb_buf = buf;
854 rc = dt_read(env, o, &lgi->lgi_buf, &cur_offset);
856 CERROR("%s: can't read llog block from log "DFID
857 " offset "LPU64": rc = %d\n",
858 o->do_lu.lo_dev->ld_obd->obd_name,
859 PFID(lu_object_fid(&o->do_lu)), cur_offset, rc);
863 if (rc == 0) /* end of file, nothing to do */
866 if (rc < sizeof(*tail)) {
867 CERROR("%s: invalid llog block at log id "DOSTID"/%u "
869 o->do_lu.lo_dev->ld_obd->obd_name,
870 POSTID(&loghandle->lgh_id.lgl_oi),
871 loghandle->lgh_id.lgl_ogen, cur_offset);
872 GOTO(out, rc = -EINVAL);
876 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
877 lustre_swab_llog_rec(rec);
879 tail = (struct llog_rec_tail *)((char *)buf + rc -
880 sizeof(struct llog_rec_tail));
881 /* get the last record in block */
882 last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
883 le32_to_cpu(tail->lrt_len));
885 if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
886 lustre_swab_llog_rec(last_rec);
887 LASSERT(last_rec->lrh_index == tail->lrt_index);
889 /* this shouldn't happen */
890 if (tail->lrt_index == 0) {
891 CERROR("%s: invalid llog tail at log id "DOSTID"/%u "
893 o->do_lu.lo_dev->ld_obd->obd_name,
894 POSTID(&loghandle->lgh_id.lgl_oi),
895 loghandle->lgh_id.lgl_ogen, cur_offset);
896 GOTO(out, rc = -EINVAL);
898 if (tail->lrt_index < prev_idx)
901 /* sanity check that the start of the new buffer is no farther
902 * than the record that we wanted. This shouldn't happen. */
903 if (rec->lrh_index > prev_idx) {
904 CERROR("%s: missed desired record? %u > %u\n",
905 o->do_lu.lo_dev->ld_obd->obd_name,
906 rec->lrh_index, prev_idx);
907 GOTO(out, rc = -ENOENT);
910 /* Trim unsupported extensions for compat w/ older clients */
911 if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_JOBID))
912 changelog_block_trim_ext(rec, last_rec,
913 CLF_VERSION | CLF_RENAME);
917 GOTO(out, rc = -EIO);
923 * This is helper function to get llog directory object. It is used by named
924 * llog operations to find/insert/delete llog entry from llog directory.
926 * \param[in] env execution environment
927 * \param[in] ctxt llog context
929 * \retval dt_object of llog directory
930 * \retval ERR_PTR of negative value on error
932 static struct dt_object *llog_osd_dir_get(const struct lu_env *env,
933 struct llog_ctxt *ctxt)
935 struct dt_device *dt;
936 struct dt_thread_info *dti = dt_info(env);
937 struct dt_object *dir;
940 dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
941 if (ctxt->loc_dir == NULL) {
942 rc = dt_root_get(env, dt, &dti->dti_fid);
945 dir = dt_locate(env, dt, &dti->dti_fid);
947 if (!IS_ERR(dir) && !dt_try_as_dir(env, dir)) {
948 lu_object_put(env, &dir->do_lu);
949 return ERR_PTR(-ENOTDIR);
952 lu_object_get(&ctxt->loc_dir->do_lu);
960 * Implementation of the llog_operations::lop_open
962 * This function opens the llog by its logid or by name, it may open also
963 * non existent llog and assing then new id to it.
964 * The llog_open/llog_close pair works similar to lu_object_find/put,
965 * the object may not exist prior open. The result of open is just dt_object
966 * in the llog header.
968 * \param[in] env execution environment
969 * \param[in] handle llog handle of the current llog
970 * \param[in] logid logid of llog to open (nameless llog)
971 * \param[in] name name of llog to open (named llog)
972 * \param[in] open_param
973 * LLOG_OPEN_NEW - new llog, may not exist
974 * LLOG_OPEN_EXIST - old llog, must exist
976 * \retval 0 on successful open, llog_handle::lgh_obj
977 * contains the dt_object of the llog.
978 * \retval negative value on error
980 static int llog_osd_open(const struct lu_env *env, struct llog_handle *handle,
981 struct llog_logid *logid, char *name,
982 enum llog_open_param open_param)
984 struct llog_thread_info *lgi = llog_info(env);
985 struct llog_ctxt *ctxt = handle->lgh_ctxt;
987 struct dt_device *dt;
988 struct ls_device *ls;
989 struct local_oid_storage *los = NULL;
996 LASSERT(ctxt->loc_exp);
997 LASSERT(ctxt->loc_exp->exp_obd);
998 dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
1000 if (ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1001 if (logid != NULL) {
1002 logid_to_fid(logid, &lgi->lgi_fid);
1004 /* If logid == NULL, then it means the caller needs
1005 * to allocate new FID (llog_cat_declare_add_rec()). */
1006 rc = obd_fid_alloc(env, ctxt->loc_exp,
1007 &lgi->lgi_fid, NULL);
1013 o = dt_locate(env, dt, &lgi->lgi_fid);
1020 ls = ls_device_get(dt);
1022 RETURN(PTR_ERR(ls));
1024 mutex_lock(&ls->ls_los_mutex);
1025 los = dt_los_find(ls, name != NULL ? FID_SEQ_LLOG_NAME : FID_SEQ_LLOG);
1026 mutex_unlock(&ls->ls_los_mutex);
1028 ls_device_put(env, ls);
1032 if (logid != NULL) {
1033 logid_to_fid(logid, &lgi->lgi_fid);
1035 struct dt_object *llog_dir;
1037 llog_dir = llog_osd_dir_get(env, ctxt);
1038 if (IS_ERR(llog_dir))
1039 GOTO(out, rc = PTR_ERR(llog_dir));
1040 dt_read_lock(env, llog_dir, 0);
1041 rc = dt_lookup_dir(env, llog_dir, name, &lgi->lgi_fid);
1042 dt_read_unlock(env, llog_dir);
1043 lu_object_put(env, &llog_dir->do_lu);
1044 if (rc == -ENOENT && open_param == LLOG_OPEN_NEW) {
1045 /* generate fid for new llog */
1046 rc = local_object_fid_generate(env, los,
1051 OBD_ALLOC(handle->lgh_name, strlen(name) + 1);
1052 if (handle->lgh_name)
1053 strcpy(handle->lgh_name, name);
1055 GOTO(out, rc = -ENOMEM);
1057 LASSERTF(open_param & LLOG_OPEN_NEW, "%#x\n", open_param);
1058 /* generate fid for new llog */
1059 rc = local_object_fid_generate(env, los, &lgi->lgi_fid);
1064 o = ls_locate(env, ls, &lgi->lgi_fid, NULL);
1066 GOTO(out_name, rc = PTR_ERR(o));
1069 /* No new llog is expected but doesn't exist */
1070 if (open_param != LLOG_OPEN_NEW && !dt_object_exists(o))
1071 GOTO(out_put, rc = -ENOENT);
1073 fid_to_logid(&lgi->lgi_fid, &handle->lgh_id);
1074 handle->lgh_obj = o;
1075 handle->private_data = los;
1076 LASSERT(handle->lgh_ctxt);
1081 lu_object_put(env, &o->do_lu);
1083 if (handle->lgh_name != NULL)
1084 OBD_FREE(handle->lgh_name, strlen(name) + 1);
1092 * Implementation of the llog_operations::lop_exist
1094 * This function checks that llog exists on storage.
1096 * \param[in] handle llog handle of the current llog
1098 * \retval true if llog object exists and is not just destroyed
1099 * \retval false if llog doesn't exist or just destroyed
1101 static int llog_osd_exist(struct llog_handle *handle)
1103 LASSERT(handle->lgh_obj);
1104 return (dt_object_exists(handle->lgh_obj) &&
1105 !lu_object_is_dying(handle->lgh_obj->do_lu.lo_header));
1109 * Get dir for regular fid log object
1111 * Get directory for regular fid log object, and these regular fid log
1112 * object will be inserted under this directory, to satisfy the FS
1113 * consistency check, e2fsck etc.
1115 * \param [in] env execution environment
1116 * \param [in] dto llog object
1118 * \retval pointer to the directory if it is found.
1119 * \retval ERR_PTR(negative errno) if it fails.
1121 struct dt_object *llog_osd_get_regular_fid_dir(const struct lu_env *env,
1122 struct dt_object *dto)
1124 struct llog_thread_info *lgi = llog_info(env);
1125 struct seq_server_site *ss = dto->do_lu.lo_dev->ld_site->ld_seq_site;
1126 struct lu_seq_range *range = &lgi->lgi_range;
1127 struct lu_fid *dir_fid = &lgi->lgi_fid;
1128 struct dt_object *dir;
1132 fld_range_set_any(range);
1133 LASSERT(ss != NULL);
1134 rc = ss->ss_server_fld->lsf_seq_lookup(env, ss->ss_server_fld,
1135 fid_seq(lu_object_fid(&dto->do_lu)), range);
1137 RETURN(ERR_PTR(rc));
1139 lu_update_log_dir_fid(dir_fid, range->lsr_index);
1140 dir = dt_locate(env, lu2dt_dev(dto->do_lu.lo_dev), dir_fid);
1144 if (!dt_try_as_dir(env, dir)) {
1145 lu_object_put(env, &dir->do_lu);
1146 RETURN(ERR_PTR(-ENOTDIR));
1153 * Add llog object with regular FID to name entry
1155 * Add llog object with regular FID to name space, and each llog
1156 * object on each MDT will be /update_log_dir/[seq:oid:ver],
1157 * so to satisfy the namespace consistency check, e2fsck etc.
1159 * \param [in] env execution environment
1160 * \param [in] dto llog object
1161 * \param [in] th thandle
1162 * \param [in] declare if it is declare or execution
1164 * \retval 0 if insertion succeeds.
1165 * \retval negative errno if insertion fails.
1168 llog_osd_regular_fid_add_name_entry(const struct lu_env *env,
1169 struct dt_object *dto,
1170 struct thandle *th, bool declare)
1172 struct llog_thread_info *lgi = llog_info(env);
1173 const struct lu_fid *fid = lu_object_fid(&dto->do_lu);
1174 struct dt_insert_rec *rec = &lgi->lgi_dt_rec;
1175 struct dt_object *dir;
1176 char *name = lgi->lgi_name;
1180 if (!fid_is_norm(fid))
1183 dir = llog_osd_get_regular_fid_dir(env, dto);
1185 RETURN(PTR_ERR(dir));
1188 rec->rec_type = S_IFREG;
1189 snprintf(name, sizeof(lgi->lgi_name), DFID, PFID(fid));
1190 dt_write_lock(env, dir, 0);
1192 rc = dt_declare_insert(env, dir, (struct dt_rec *)rec,
1193 (struct dt_key *)name, th);
1195 rc = dt_insert(env, dir, (struct dt_rec *)rec,
1196 (struct dt_key *)name, th, 1);
1198 dt_write_unlock(env, dir);
1200 lu_object_put(env, &dir->do_lu);
1206 * Implementation of the llog_operations::lop_declare_create
1208 * This function declares the llog create. It declares also name insert
1209 * into llog directory in case of named llog.
1211 * \param[in] env execution environment
1212 * \param[in] res llog handle of the current llog
1213 * \param[in] th current transaction handle
1215 * \retval 0 on successful create declaration
1216 * \retval negative value on error
1218 static int llog_osd_declare_create(const struct lu_env *env,
1219 struct llog_handle *res, struct thandle *th)
1221 struct llog_thread_info *lgi = llog_info(env);
1222 struct dt_insert_rec *rec = &lgi->lgi_dt_rec;
1223 struct local_oid_storage *los;
1224 struct dt_object *o;
1229 LASSERT(res->lgh_obj);
1232 /* object can be created by another thread */
1234 if (dt_object_exists(o))
1237 if (res->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1238 struct llog_thread_info *lgi = llog_info(env);
1240 lgi->lgi_attr.la_valid = LA_MODE | LA_SIZE;
1241 lgi->lgi_attr.la_size = 0;
1242 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1243 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1245 rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL,
1251 rc = llog_osd_regular_fid_add_name_entry(env, o, th, true);
1255 los = res->private_data;
1258 rc = llog_osd_declare_new_object(env, los, o, th);
1262 /* do not declare header initialization here as it's declared
1263 * in llog_osd_declare_write_rec() which is always called */
1265 if (res->lgh_name) {
1266 struct dt_object *llog_dir;
1268 llog_dir = llog_osd_dir_get(env, res->lgh_ctxt);
1269 if (IS_ERR(llog_dir))
1270 RETURN(PTR_ERR(llog_dir));
1271 logid_to_fid(&res->lgh_id, &lgi->lgi_fid);
1272 rec->rec_fid = &lgi->lgi_fid;
1273 rec->rec_type = S_IFREG;
1274 rc = dt_declare_insert(env, llog_dir,
1275 (struct dt_rec *)rec,
1276 (struct dt_key *)res->lgh_name, th);
1277 lu_object_put(env, &llog_dir->do_lu);
1279 CERROR("%s: can't declare named llog %s: rc = %d\n",
1280 o->do_lu.lo_dev->ld_obd->obd_name,
1287 * Implementation of the llog_operations::lop_create
1289 * This function creates the llog according with llog_handle::lgh_obj
1290 * and llog_handle::lgh_name.
1292 * \param[in] env execution environment
1293 * \param[in] res llog handle of the current llog
1294 * \param[in] th current transaction handle
1296 * \retval 0 on successful create
1297 * \retval negative value on error
1299 static int llog_osd_create(const struct lu_env *env, struct llog_handle *res,
1302 struct llog_thread_info *lgi = llog_info(env);
1303 struct dt_insert_rec *rec = &lgi->lgi_dt_rec;
1304 struct local_oid_storage *los;
1305 struct dt_object *o;
1314 /* llog can be already created */
1315 if (dt_object_exists(o))
1318 if (res->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1319 struct llog_thread_info *lgi = llog_info(env);
1321 lgi->lgi_attr.la_valid = LA_MODE | LA_SIZE | LA_TYPE;
1322 lgi->lgi_attr.la_size = 0;
1323 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1324 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1326 dt_write_lock(env, o, 0);
1327 rc = dt_create(env, o, &lgi->lgi_attr, NULL,
1329 dt_write_unlock(env, o);
1333 rc = llog_osd_regular_fid_add_name_entry(env, o, th, false);
1338 los = res->private_data;
1341 dt_write_lock(env, o, 0);
1342 if (!dt_object_exists(o))
1343 rc = llog_osd_create_new_object(env, los, o, th);
1347 dt_write_unlock(env, o);
1351 if (res->lgh_name) {
1352 struct dt_object *llog_dir;
1354 llog_dir = llog_osd_dir_get(env, res->lgh_ctxt);
1355 if (IS_ERR(llog_dir))
1356 RETURN(PTR_ERR(llog_dir));
1358 logid_to_fid(&res->lgh_id, &lgi->lgi_fid);
1359 rec->rec_fid = &lgi->lgi_fid;
1360 rec->rec_type = S_IFREG;
1361 dt_read_lock(env, llog_dir, 0);
1362 rc = dt_insert(env, llog_dir, (struct dt_rec *)rec,
1363 (struct dt_key *)res->lgh_name,
1365 dt_read_unlock(env, llog_dir);
1366 lu_object_put(env, &llog_dir->do_lu);
1368 CERROR("%s: can't create named llog %s: rc = %d\n",
1369 o->do_lu.lo_dev->ld_obd->obd_name,
1376 * Implementation of the llog_operations::lop_close
1378 * This function closes the llog. It just put llog object and referenced
1381 * \param[in] env execution environment
1382 * \param[in] handle llog handle of the current llog
1384 * \retval 0 on successful llog close
1385 * \retval negative value on error
1387 static int llog_osd_close(const struct lu_env *env, struct llog_handle *handle)
1389 struct local_oid_storage *los;
1394 LASSERT(handle->lgh_obj);
1396 if (handle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1397 /* Remove the object from the cache, otherwise it may
1398 * hold LOD being released during cleanup process */
1399 lu_object_put_nocache(env, &handle->lgh_obj->do_lu);
1400 LASSERT(handle->private_data == NULL);
1403 lu_object_put(env, &handle->lgh_obj->do_lu);
1405 los = handle->private_data;
1409 if (handle->lgh_name)
1410 OBD_FREE(handle->lgh_name, strlen(handle->lgh_name) + 1);
1416 * delete llog object name entry
1418 * Delete llog object (with regular FID) from name space (under
1421 * \param [in] env execution environment
1422 * \param [in] dto llog object
1423 * \param [in] th thandle
1424 * \param [in] declare if it is declare or execution
1426 * \retval 0 if deletion succeeds.
1427 * \retval negative errno if deletion fails.
1430 llog_osd_regular_fid_del_name_entry(const struct lu_env *env,
1431 struct dt_object *dto,
1432 struct thandle *th, bool declare)
1434 struct llog_thread_info *lgi = llog_info(env);
1435 const struct lu_fid *fid = lu_object_fid(&dto->do_lu);
1436 struct dt_object *dir;
1437 char *name = lgi->lgi_name;
1441 if (!fid_is_norm(fid))
1444 dir = llog_osd_get_regular_fid_dir(env, dto);
1446 RETURN(PTR_ERR(dir));
1448 snprintf(name, sizeof(lgi->lgi_name), DFID, PFID(fid));
1449 dt_write_lock(env, dir, 0);
1451 rc = dt_declare_delete(env, dir, (struct dt_key *)name,
1454 rc = dt_delete(env, dir, (struct dt_key *)name, th);
1456 dt_write_unlock(env, dir);
1458 lu_object_put(env, &dir->do_lu);
1464 * Implementation of the llog_operations::lop_destroy
1466 * This function destroys the llog and deletes also entry in the
1467 * llog directory in case of named llog. Llog should be opened prior that.
1468 * Destroy method is not part of external transaction and does everything
1471 * \param[in] env execution environment
1472 * \param[in] loghandle llog handle of the current llog
1474 * \retval 0 on successful destroy
1475 * \retval negative value on error
1477 static int llog_osd_destroy(const struct lu_env *env,
1478 struct llog_handle *loghandle)
1480 struct llog_ctxt *ctxt;
1481 struct dt_object *o, *llog_dir = NULL;
1482 struct dt_device *d;
1489 ctxt = loghandle->lgh_ctxt;
1492 o = loghandle->lgh_obj;
1495 d = lu2dt_dev(o->do_lu.lo_dev);
1497 LASSERT(d == ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt);
1499 th = dt_trans_create(env, d);
1501 RETURN(PTR_ERR(th));
1503 if (loghandle->lgh_name) {
1504 llog_dir = llog_osd_dir_get(env, ctxt);
1505 if (IS_ERR(llog_dir))
1506 GOTO(out_trans, rc = PTR_ERR(llog_dir));
1508 name = loghandle->lgh_name;
1509 rc = dt_declare_delete(env, llog_dir,
1510 (struct dt_key *)name, th);
1512 GOTO(out_trans, rc);
1515 rc = dt_declare_ref_del(env, o, th);
1517 GOTO(out_trans, rc);
1519 rc = dt_declare_destroy(env, o, th);
1521 GOTO(out_trans, rc);
1523 if (loghandle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1524 rc = llog_osd_regular_fid_del_name_entry(env, o, th, true);
1526 GOTO(out_trans, rc);
1529 rc = dt_trans_start_local(env, d, th);
1531 GOTO(out_trans, rc);
1533 th->th_wait_submit = 1;
1535 dt_write_lock(env, o, 0);
1536 if (dt_object_exists(o)) {
1538 dt_read_lock(env, llog_dir, 0);
1539 rc = dt_delete(env, llog_dir,
1540 (struct dt_key *) name,
1542 dt_read_unlock(env, llog_dir);
1544 CERROR("%s: can't remove llog %s: rc = %d\n",
1545 o->do_lu.lo_dev->ld_obd->obd_name,
1547 GOTO(out_unlock, rc);
1550 dt_ref_del(env, o, th);
1551 rc = dt_destroy(env, o, th);
1553 GOTO(out_unlock, rc);
1555 if (loghandle->lgh_ctxt->loc_flags &
1556 LLOG_CTXT_FLAG_NORMAL_FID) {
1557 rc = llog_osd_regular_fid_del_name_entry(env, o, th,
1560 GOTO(out_unlock, rc);
1564 dt_write_unlock(env, o);
1566 dt_trans_stop(env, d, th);
1567 if (!(IS_ERR_OR_NULL(llog_dir)))
1568 lu_object_put(env, &llog_dir->do_lu);
1573 * Implementation of the llog_operations::lop_setup
1575 * This function setup the llog on local storage.
1577 * \param[in] env execution environment
1578 * \param[in] obd obd device the llog belongs to
1579 * \param[in] olg the llog group, it is always zero group now.
1580 * \param[in] ctxt_idx the llog index, it defines the purpose of this llog.
1581 * Every new llog type have to use own index.
1582 * \param[in] disk_obd the storage obd, where llog is stored.
1584 * \retval 0 on successful llog setup
1585 * \retval negative value on error
1587 static int llog_osd_setup(const struct lu_env *env, struct obd_device *obd,
1588 struct obd_llog_group *olg, int ctxt_idx,
1589 struct obd_device *disk_obd)
1591 struct llog_thread_info *lgi = llog_info(env);
1592 struct llog_ctxt *ctxt;
1597 LASSERT(olg->olg_ctxts[ctxt_idx]);
1599 ctxt = llog_ctxt_get(olg->olg_ctxts[ctxt_idx]);
1602 if (disk_obd == NULL)
1605 /* initialize data allowing to generate new fids,
1606 * literally we need a sequece */
1607 lgi->lgi_fid.f_seq = FID_SEQ_LLOG;
1608 lgi->lgi_fid.f_oid = 1;
1609 lgi->lgi_fid.f_ver = 0;
1610 rc = local_oid_storage_init(env, disk_obd->obd_lvfs_ctxt.dt,
1612 &ctxt->loc_los_nameless);
1616 lgi->lgi_fid.f_seq = FID_SEQ_LLOG_NAME;
1617 lgi->lgi_fid.f_oid = 1;
1618 lgi->lgi_fid.f_ver = 0;
1619 rc = local_oid_storage_init(env, disk_obd->obd_lvfs_ctxt.dt,
1621 &ctxt->loc_los_named);
1623 local_oid_storage_fini(env, ctxt->loc_los_nameless);
1624 ctxt->loc_los_nameless = NULL;
1630 llog_ctxt_put(ctxt);
1635 * Implementation of the llog_operations::lop_cleanup
1637 * This function cleanups the llog on local storage.
1639 * \param[in] env execution environment
1640 * \param[in] ctxt the llog context
1644 static int llog_osd_cleanup(const struct lu_env *env, struct llog_ctxt *ctxt)
1646 if (ctxt->loc_los_nameless != NULL) {
1647 local_oid_storage_fini(env, ctxt->loc_los_nameless);
1648 ctxt->loc_los_nameless = NULL;
1651 if (ctxt->loc_los_named != NULL) {
1652 local_oid_storage_fini(env, ctxt->loc_los_named);
1653 ctxt->loc_los_named = NULL;
1659 struct llog_operations llog_osd_ops = {
1660 .lop_next_block = llog_osd_next_block,
1661 .lop_prev_block = llog_osd_prev_block,
1662 .lop_read_header = llog_osd_read_header,
1663 .lop_destroy = llog_osd_destroy,
1664 .lop_setup = llog_osd_setup,
1665 .lop_cleanup = llog_osd_cleanup,
1666 .lop_open = llog_osd_open,
1667 .lop_exist = llog_osd_exist,
1668 .lop_declare_create = llog_osd_declare_create,
1669 .lop_create = llog_osd_create,
1670 .lop_declare_write_rec = llog_osd_declare_write_rec,
1671 .lop_write_rec = llog_osd_write_rec,
1672 .lop_close = llog_osd_close,
1674 EXPORT_SYMBOL(llog_osd_ops);
1676 struct llog_operations llog_common_cat_ops = {
1677 .lop_next_block = llog_osd_next_block,
1678 .lop_prev_block = llog_osd_prev_block,
1679 .lop_read_header = llog_osd_read_header,
1680 .lop_destroy = llog_osd_destroy,
1681 .lop_setup = llog_osd_setup,
1682 .lop_cleanup = llog_osd_cleanup,
1683 .lop_open = llog_osd_open,
1684 .lop_exist = llog_osd_exist,
1685 .lop_declare_create = llog_osd_declare_create,
1686 .lop_create = llog_osd_create,
1687 .lop_declare_write_rec = llog_osd_declare_write_rec,
1688 .lop_write_rec = llog_osd_write_rec,
1689 .lop_close = llog_osd_close,
1690 .lop_add = llog_cat_add_rec,
1691 .lop_declare_add = llog_cat_declare_add_rec,
1693 EXPORT_SYMBOL(llog_common_cat_ops);
1696 * Read the special file which contains the list of llog catalogs IDs
1698 * This function reads the CATALOGS file which contains the array of llog
1699 * catalogs IDs. The main purpose of this file is to store OSP llogs indexed
1700 * by OST/MDT number.
1702 * \param[in] env execution environment
1703 * \param[in] d corresponding storage device
1704 * \param[in] idx position to start from, usually OST/MDT index
1705 * \param[in] count how many catalog IDs to read
1706 * \param[out] idarray the buffer for the data. If it is NULL then
1707 * function returns just number of catalog IDs
1709 * \param[in] fid LLOG_CATALOGS_OID for CATALOG object
1711 * \retval 0 on successful read of catalog IDs
1712 * \retval negative value on error
1713 * \retval positive value which is number of records in
1714 * the file if \a idarray is NULL
1716 int llog_osd_get_cat_list(const struct lu_env *env, struct dt_device *d,
1717 int idx, int count, struct llog_catid *idarray,
1718 const struct lu_fid *fid)
1720 struct llog_thread_info *lgi = llog_info(env);
1721 struct dt_object *o = NULL;
1729 size = sizeof(*idarray) * count;
1730 lgi->lgi_off = idx * sizeof(*idarray);
1732 lgi->lgi_fid = *fid;
1733 o = dt_locate(env, d, &lgi->lgi_fid);
1737 if (!dt_object_exists(o)) {
1738 th = dt_trans_create(env, d);
1740 GOTO(out, rc = PTR_ERR(th));
1742 lgi->lgi_attr.la_valid = LA_MODE;
1743 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1744 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1746 th->th_wait_submit = 1;
1747 /* Make the llog object creation synchronization, so
1748 * it will be reliable to the reference, especially
1749 * for remote reference */
1752 rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL,
1755 GOTO(out_trans, rc);
1757 rc = dt_trans_start_local(env, d, th);
1759 GOTO(out_trans, rc);
1761 dt_write_lock(env, o, 0);
1762 if (!dt_object_exists(o))
1763 rc = dt_create(env, o, &lgi->lgi_attr, NULL,
1765 dt_write_unlock(env, o);
1767 dt_trans_stop(env, d, th);
1772 rc = dt_attr_get(env, o, &lgi->lgi_attr);
1776 if (!S_ISREG(lgi->lgi_attr.la_mode)) {
1777 CERROR("%s: CATALOGS is not a regular file!: mode = %o\n",
1778 o->do_lu.lo_dev->ld_obd->obd_name,
1779 lgi->lgi_attr.la_mode);
1780 GOTO(out, rc = -ENOENT);
1783 CDEBUG(D_CONFIG, "cat list: disk size=%d, read=%d\n",
1784 (int)lgi->lgi_attr.la_size, size);
1786 /* return just number of llogs */
1787 if (idarray == NULL) {
1788 rc = lgi->lgi_attr.la_size / sizeof(*idarray);
1792 /* read for new ost index or for empty file */
1793 memset(idarray, 0, size);
1794 if (lgi->lgi_attr.la_size <= lgi->lgi_off)
1796 if (lgi->lgi_attr.la_size < lgi->lgi_off + size)
1797 size = lgi->lgi_attr.la_size - lgi->lgi_off;
1799 lgi->lgi_buf.lb_buf = idarray;
1800 lgi->lgi_buf.lb_len = size;
1801 rc = dt_record_read(env, o, &lgi->lgi_buf, &lgi->lgi_off);
1802 /* -EFAULT means the llog is a sparse file. This is not an error
1803 * after arbitrary OST index is supported. */
1804 if (rc < 0 && rc != -EFAULT) {
1805 CERROR("%s: error reading CATALOGS: rc = %d\n",
1806 o->do_lu.lo_dev->ld_obd->obd_name, rc);
1812 lu_object_put(env, &o->do_lu);
1815 EXPORT_SYMBOL(llog_osd_get_cat_list);
1818 * Write the special file which contains the list of llog catalogs IDs
1820 * This function writes the CATALOG file which contains the array of llog
1821 * catalogs IDs. It is used mostly to store OSP llogs indexed by OST/MDT
1824 * \param[in] env execution environment
1825 * \param[in] d corresponding storage device
1826 * \param[in] idx position to start from, usually OST/MDT index
1827 * \param[in] count how many catalog IDs to write
1828 * \param[out] idarray the buffer with the data to write.
1829 * \param[in] fid LLOG_CATALOGS_OID for CATALOG object
1831 * \retval 0 on successful write of catalog IDs
1832 * \retval negative value on error
1834 int llog_osd_put_cat_list(const struct lu_env *env, struct dt_device *d,
1835 int idx, int count, struct llog_catid *idarray,
1836 const struct lu_fid *fid)
1838 struct llog_thread_info *lgi = llog_info(env);
1839 struct dt_object *o = NULL;
1848 size = sizeof(*idarray) * count;
1849 lgi->lgi_off = idx * sizeof(*idarray);
1850 lgi->lgi_fid = *fid;
1852 o = dt_locate(env, d, &lgi->lgi_fid);
1856 if (!dt_object_exists(o))
1857 GOTO(out, rc = -ENOENT);
1859 rc = dt_attr_get(env, o, &lgi->lgi_attr);
1863 if (!S_ISREG(lgi->lgi_attr.la_mode)) {
1864 CERROR("%s: CATALOGS is not a regular file!: mode = %o\n",
1865 o->do_lu.lo_dev->ld_obd->obd_name,
1866 lgi->lgi_attr.la_mode);
1867 GOTO(out, rc = -ENOENT);
1870 th = dt_trans_create(env, d);
1872 GOTO(out, rc = PTR_ERR(th));
1874 lgi->lgi_buf.lb_len = size;
1875 lgi->lgi_buf.lb_buf = idarray;
1876 rc = dt_declare_record_write(env, o, &lgi->lgi_buf, lgi->lgi_off, th);
1880 rc = dt_trans_start_local(env, d, th);
1882 GOTO(out_trans, rc);
1884 th->th_wait_submit = 1;
1886 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
1888 CDEBUG(D_INODE, "can't write CATALOGS at index %d: rc = %d\n",
1891 dt_trans_stop(env, d, th);
1893 lu_object_put(env, &o->do_lu);
1896 EXPORT_SYMBOL(llog_osd_put_cat_list);