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;
198 lgi = llog_info(env);
200 rc = dt_attr_get(env, o, &lgi->lgi_attr);
204 LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
206 if (lgi->lgi_attr.la_size == 0) {
207 CDEBUG(D_HA, "not reading header from 0-byte log\n");
211 flags = handle->lgh_hdr->llh_flags;
214 lgi->lgi_buf.lb_buf = handle->lgh_hdr;
215 lgi->lgi_buf.lb_len = handle->lgh_hdr_size;
216 rc = dt_read(env, o, &lgi->lgi_buf, &lgi->lgi_off);
217 llh_hdr = &handle->lgh_hdr->llh_hdr;
218 if (rc < sizeof(*llh_hdr) || rc < llh_hdr->lrh_len) {
219 CERROR("%s: error reading "DFID" log header size %d: rc = %d\n",
220 o->do_lu.lo_dev->ld_obd->obd_name,
221 PFID(lu_object_fid(&o->do_lu)), rc < 0 ? 0 : rc,
230 if (LLOG_REC_HDR_NEEDS_SWABBING(llh_hdr))
231 lustre_swab_llog_hdr(handle->lgh_hdr);
233 if (llh_hdr->lrh_type != LLOG_HDR_MAGIC) {
234 CERROR("%s: bad log %s "DFID" header magic: %#x "
235 "(expected %#x)\n", o->do_lu.lo_dev->ld_obd->obd_name,
236 handle->lgh_name ? handle->lgh_name : "",
237 PFID(lu_object_fid(&o->do_lu)),
238 llh_hdr->lrh_type, LLOG_HDR_MAGIC);
240 } else if (llh_hdr->lrh_len < LLOG_MIN_CHUNK_SIZE ||
241 llh_hdr->lrh_len > handle->lgh_hdr_size) {
242 CERROR("%s: incorrectly sized log %s "DFID" header: "
243 "%#x (expected at least %#x)\n"
244 "you may need to re-run lconf --write_conf.\n",
245 o->do_lu.lo_dev->ld_obd->obd_name,
246 handle->lgh_name ? handle->lgh_name : "",
247 PFID(lu_object_fid(&o->do_lu)),
248 llh_hdr->lrh_len, LLOG_MIN_CHUNK_SIZE);
250 } else if (LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_index >
251 LLOG_HDR_BITMAP_SIZE(handle->lgh_hdr) ||
252 LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_len !=
254 CERROR("%s: incorrectly sized log %s "DFID" tailer: "
256 o->do_lu.lo_dev->ld_obd->obd_name,
257 handle->lgh_name ? handle->lgh_name : "",
258 PFID(lu_object_fid(&o->do_lu)),
259 LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_len, -EIO);
263 handle->lgh_hdr->llh_flags |= (flags & LLOG_F_EXT_MASK);
264 handle->lgh_last_idx = LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_index;
270 * Implementation of the llog_operations::lop_declare_write
272 * This function declares the new record write.
274 * \param[in] env execution environment
275 * \param[in] loghandle llog handle of the current llog
276 * \param[in] rec llog record header. This is a real header of the full
277 * llog record to write. This is the beginning of buffer
278 * to write, the length of buffer is stored in
280 * \param[in] idx index of the llog record. If \a idx == -1 then this is
281 * append case, otherwise \a idx is the index of record
283 * \param[in] th current transaction handle
285 * \retval 0 on successful declaration
286 * \retval negative error if declaration failed
288 static int llog_osd_declare_write_rec(const struct lu_env *env,
289 struct llog_handle *loghandle,
290 struct llog_rec_hdr *rec,
291 int idx, struct thandle *th)
293 struct llog_thread_info *lgi = llog_info(env);
304 LASSERT(rec->lrh_len <= loghandle->lgh_ctxt->loc_chunk_size);
306 o = loghandle->lgh_obj;
309 chunk_size = loghandle->lgh_ctxt->loc_chunk_size;
310 lgi->lgi_buf.lb_len = chunk_size;
311 lgi->lgi_buf.lb_buf = NULL;
312 /* each time we update header */
313 rc = dt_declare_record_write(env, o, &lgi->lgi_buf, 0,
315 if (rc || idx == 0) /* if error or just header */
319 * the pad record can be inserted so take into account double
322 lgi->lgi_buf.lb_len = chunk_size * 2;
323 lgi->lgi_buf.lb_buf = NULL;
324 /* XXX: implement declared window or multi-chunks approach */
325 rc = dt_declare_record_write(env, o, &lgi->lgi_buf, -1, th);
331 * Implementation of the llog_operations::lop_write
333 * This function writes the new record in the llog or modify the existed one.
335 * \param[in] env execution environment
336 * \param[in] loghandle llog handle of the current llog
337 * \param[in] rec llog record header. This is a real header of
338 * the full llog record to write. This is
339 * the beginning of buffer to write, the length
340 * of buffer is stored in \a rec::lrh_len
341 * \param[out] reccookie pointer to the cookie to return back if needed.
342 * It is used for further cancel of this llog
344 * \param[in] idx index of the llog record. If \a idx == -1 then
345 * this is append case, otherwise \a idx is
346 * the index of record to modify
347 * \param[in] th current transaction handle
349 * \retval 0 on successful write && \a reccookie == NULL
350 * 1 on successful write && \a reccookie != NULL
351 * \retval negative error if write failed
353 static int llog_osd_write_rec(const struct lu_env *env,
354 struct llog_handle *loghandle,
355 struct llog_rec_hdr *rec,
356 struct llog_cookie *reccookie,
357 int idx, struct thandle *th)
359 struct llog_thread_info *lgi = llog_info(env);
360 struct llog_log_hdr *llh;
361 int reclen = rec->lrh_len;
363 struct llog_rec_tail *lrt;
371 llh = loghandle->lgh_hdr;
373 o = loghandle->lgh_obj;
377 chunk_size = llh->llh_hdr.lrh_len;
378 CDEBUG(D_OTHER, "new record %x to "DFID"\n",
379 rec->lrh_type, PFID(lu_object_fid(&o->do_lu)));
381 /* record length should not bigger than */
382 if (reclen > loghandle->lgh_hdr->llh_hdr.lrh_len)
385 rc = dt_attr_get(env, o, &lgi->lgi_attr);
390 * The modification case.
391 * If idx set then the record with that index must be modified.
392 * There are three cases possible:
393 * 1) the common case is the llog header update (idx == 0)
394 * 2) the llog record modification during llog process.
395 * This is indicated by the \a loghandle::lgh_cur_idx > 0.
396 * In that case the \a loghandle::lgh_cur_offset
397 * 3) otherwise this is assumed that llog consist of records of
398 * fixed size, i.e. catalog. The llog header must has llh_size
399 * field equal to record size. The record offset is calculated
400 * just by /a idx value
402 * During modification we don't need extra header update because
403 * the bitmap and record count are not changed. The record header
404 * and tail remains the same too.
406 if (idx != LLOG_NEXT_IDX) {
407 /* llog can be empty only when first record is being written */
408 LASSERT(ergo(idx > 0, lgi->lgi_attr.la_size > 0));
410 if (!ext2_test_bit(idx, LLOG_HDR_BITMAP(llh))) {
411 CERROR("%s: modify unset record %u\n",
412 o->do_lu.lo_dev->ld_obd->obd_name, idx);
416 if (idx != rec->lrh_index) {
417 CERROR("%s: modify index mismatch %d %u\n",
418 o->do_lu.lo_dev->ld_obd->obd_name, idx,
423 if (idx == LLOG_HEADER_IDX) {
424 /* llog header update */
425 __u32 *bitmap = LLOG_HDR_BITMAP(llh);
429 /* If it does not indicate the bitmap index
430 * (reccookie == NULL), then it means update
431 * the whole update header. Otherwise only
432 * update header and bits needs to be updated,
433 * and in DNE cases, it will signaficantly
434 * shrink the RPC size.
435 * see distribute_txn_cancel_records()*/
436 if (reccookie == NULL) {
437 lgi->lgi_buf.lb_len = reclen;
438 lgi->lgi_buf.lb_buf = rec;
439 rc = dt_record_write(env, o, &lgi->lgi_buf,
444 /* update the header */
445 lgi->lgi_buf.lb_len = llh->llh_bitmap_offset;
446 lgi->lgi_buf.lb_buf = llh;
447 rc = dt_record_write(env, o, &lgi->lgi_buf,
452 /* update the bitmap */
453 index = reccookie->lgc_index;
454 lgi->lgi_off = llh->llh_bitmap_offset +
455 (index / (sizeof(*bitmap) * 8)) *
457 lgi->lgi_buf.lb_len = sizeof(*bitmap);
458 lgi->lgi_buf.lb_buf =
459 &bitmap[index/(sizeof(*bitmap)*8)];
460 rc = dt_record_write(env, o, &lgi->lgi_buf,
464 } else if (loghandle->lgh_cur_idx > 0) {
466 * The lgh_cur_offset can be used only if index is
469 if (idx != loghandle->lgh_cur_idx) {
470 CERROR("%s: modify index mismatch %d %d\n",
471 o->do_lu.lo_dev->ld_obd->obd_name, idx,
472 loghandle->lgh_cur_idx);
476 lgi->lgi_off = loghandle->lgh_cur_offset;
477 CDEBUG(D_OTHER, "modify record "DOSTID": idx:%d, "
478 "len:%u offset %llu\n",
479 POSTID(&loghandle->lgh_id.lgl_oi), idx,
480 rec->lrh_len, (long long)lgi->lgi_off);
481 } else if (llh->llh_flags & LLOG_F_IS_FIXSIZE) {
482 if (llh->llh_size == 0 ||
483 llh->llh_size != rec->lrh_len) {
484 CERROR("%s: wrong record size, llh_size is %u"
485 " but record size is %u\n",
486 o->do_lu.lo_dev->ld_obd->obd_name,
487 llh->llh_size, rec->lrh_len);
490 lgi->lgi_off = sizeof(*llh) + (idx - 1) * reclen;
492 /* This can be result of lgh_cur_idx is not set during
493 * llog processing or llh_size is not set to proper
494 * record size for fixed records llog. Therefore it is
495 * impossible to get record offset. */
496 CERROR("%s: can't get record offset, idx:%d, "
497 "len:%u.\n", o->do_lu.lo_dev->ld_obd->obd_name,
502 /* update only data, header and tail remain the same */
503 lgi->lgi_off += sizeof(struct llog_rec_hdr);
504 lgi->lgi_buf.lb_len = REC_DATA_LEN(rec);
505 lgi->lgi_buf.lb_buf = REC_DATA(rec);
506 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
507 if (rc == 0 && reccookie) {
508 reccookie->lgc_lgl = loghandle->lgh_id;
509 reccookie->lgc_index = idx;
517 * The most common case of using llog. The new index is assigned to
518 * the new record, new bit is set in llog bitmap and llog count is
521 * Make sure that records don't cross a chunk boundary, so we can
522 * process them page-at-a-time if needed. If it will cross a chunk
523 * boundary, write in a fake (but referenced) entry to pad the chunk.
525 LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
526 lgi->lgi_off = lgi->lgi_attr.la_size;
527 left = chunk_size - (lgi->lgi_off & (chunk_size - 1));
528 /* NOTE: padding is a record, but no bit is set */
529 if (left != 0 && left != reclen &&
530 left < (reclen + LLOG_MIN_REC_SIZE)) {
531 index = loghandle->lgh_last_idx + 1;
532 rc = llog_osd_pad(env, o, &lgi->lgi_off, left, index, th);
535 loghandle->lgh_last_idx++; /* for pad rec */
537 /* if it's the last idx in log file, then return -ENOSPC */
538 if (loghandle->lgh_last_idx >= LLOG_HDR_BITMAP_SIZE(llh) - 1)
541 /* increment the last_idx along with llh_tail index, they should
542 * be equal for a llog lifetime */
543 loghandle->lgh_last_idx++;
544 index = loghandle->lgh_last_idx;
545 LLOG_HDR_TAIL(llh)->lrt_index = index;
547 * NB: the caller should make sure only 1 process access
548 * the lgh_last_idx, e.g. append should be exclusive.
549 * Otherwise it might hit the assert.
551 LASSERT(index < LLOG_HDR_BITMAP_SIZE(llh));
552 rec->lrh_index = index;
554 lrt->lrt_len = rec->lrh_len;
555 lrt->lrt_index = rec->lrh_index;
557 /* the lgh_hdr_mutex protects llog header data from concurrent
558 * update/cancel, the llh_count and llh_bitmap are protected */
559 mutex_lock(&loghandle->lgh_hdr_mutex);
560 if (ext2_set_bit(index, LLOG_HDR_BITMAP(llh))) {
561 CERROR("%s: index %u already set in log bitmap\n",
562 o->do_lu.lo_dev->ld_obd->obd_name, index);
563 mutex_unlock(&loghandle->lgh_hdr_mutex);
564 LBUG(); /* should never happen */
568 if (llh->llh_flags & LLOG_F_IS_FIXSIZE) {
569 LASSERT(llh->llh_size == reclen);
571 /* Update the minimum size of the llog record */
572 if (llh->llh_size == 0)
573 llh->llh_size = reclen;
574 else if (reclen < llh->llh_size)
575 llh->llh_size = reclen;
578 if (lgi->lgi_attr.la_size == 0) {
580 lgi->lgi_buf.lb_len = llh->llh_hdr.lrh_len;
581 lgi->lgi_buf.lb_buf = &llh->llh_hdr;
582 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
584 GOTO(out_unlock, rc);
586 __u32 *bitmap = LLOG_HDR_BITMAP(llh);
588 /* Note: If this is not initialization (size == 0), then do not
589 * write the whole header (8k bytes), only update header/tail
590 * and bits needs to be updated. Because this update might be
591 * part of cross-MDT operation, which needs to write these
592 * updates into the update log(32KB limit) and also pack inside
593 * the RPC (1MB limit), if we write 8K for each operation, which
594 * will cost a lot space, and keep us adding more updates to one
597 lgi->lgi_buf.lb_len = llh->llh_bitmap_offset;
598 lgi->lgi_buf.lb_buf = &llh->llh_hdr;
599 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
601 GOTO(out_unlock, rc);
603 lgi->lgi_off = llh->llh_bitmap_offset +
604 (index / (sizeof(*bitmap) * 8)) * sizeof(*bitmap);
605 lgi->lgi_buf.lb_len = sizeof(*bitmap);
606 lgi->lgi_buf.lb_buf = &bitmap[index/(sizeof(*bitmap)*8)];
607 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
609 GOTO(out_unlock, rc);
611 lgi->lgi_off = (unsigned long)LLOG_HDR_TAIL(llh) -
613 lgi->lgi_buf.lb_len = sizeof(llh->llh_tail);
614 lgi->lgi_buf.lb_buf = LLOG_HDR_TAIL(llh);
615 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
617 GOTO(out_unlock, rc);
621 /* unlock here for remote object */
622 mutex_unlock(&loghandle->lgh_hdr_mutex);
626 rc = dt_attr_get(env, o, &lgi->lgi_attr);
630 LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
631 lgi->lgi_off = max_t(__u64, lgi->lgi_attr.la_size, lgi->lgi_off);
632 lgi->lgi_buf.lb_len = reclen;
633 lgi->lgi_buf.lb_buf = rec;
634 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
638 CDEBUG(D_OTHER, "added record "DOSTID": idx: %u, %u off"LPU64"\n",
639 POSTID(&loghandle->lgh_id.lgl_oi), index, rec->lrh_len,
641 if (reccookie != NULL) {
642 reccookie->lgc_lgl = loghandle->lgh_id;
643 reccookie->lgc_index = index;
644 if ((rec->lrh_type == MDS_UNLINK_REC) ||
645 (rec->lrh_type == MDS_SETATTR64_REC))
646 reccookie->lgc_subsys = LLOG_MDS_OST_ORIG_CTXT;
647 else if (rec->lrh_type == OST_SZ_REC)
648 reccookie->lgc_subsys = LLOG_SIZE_ORIG_CTXT;
650 reccookie->lgc_subsys = -1;
655 /* cleanup llog for error case */
656 mutex_lock(&loghandle->lgh_hdr_mutex);
657 ext2_clear_bit(index, LLOG_HDR_BITMAP(llh));
659 mutex_unlock(&loghandle->lgh_hdr_mutex);
661 /* restore llog last_idx */
662 loghandle->lgh_last_idx--;
663 LLOG_HDR_TAIL(llh)->lrt_index = loghandle->lgh_last_idx;
669 * We can skip reading at least as many log blocks as the number of
670 * minimum sized log records we are skipping. If it turns out
671 * that we are not far enough along the log (because the
672 * actual records are larger than minimum size) we just skip
675 * Note: in llog_process_thread, it will use bitmap offset as
676 * the index to locate the record, which also includs some pad
677 * records, whose record size is very small, and it also does not
678 * consider pad record when recording minimum record size (otherwise
679 * min_record size might be too small), so in some rare cases,
680 * it might skip too much record for @goal, see llog_osd_next_block().
682 * When force_mini_rec is true, it means we have to use LLOG_MIN_REC_SIZE
683 * as the min record size to skip over, usually because in the previous
684 * try, it skip too much record, see loog_osd_next(prev)_block().
686 static inline void llog_skip_over(struct llog_handle *lgh, __u64 *off,
687 int curr, int goal, __u32 chunk_size,
690 struct llog_log_hdr *llh = lgh->lgh_hdr;
692 /* Goal should not bigger than the record count */
693 if (goal > lgh->lgh_last_idx)
694 goal = lgh->lgh_last_idx;
697 if (llh->llh_flags & LLOG_F_IS_FIXSIZE) {
698 *off = chunk_size + (goal - 1) * llh->llh_size;
700 __u64 min_rec_size = LLOG_MIN_REC_SIZE;
702 if (llh->llh_size > 0 && !force_mini_rec)
703 min_rec_size = llh->llh_size;
705 *off = *off + (goal - curr - 1) * min_rec_size;
708 /* always align with lower chunk boundary*/
709 *off &= ~(chunk_size - 1);
713 * Remove optional fields that the client doesn't expect.
714 * This is typically in order to ensure compatibility with older clients.
715 * It is assumed that since we exclusively remove fields, the block will be
716 * big enough to handle the remapped records. It is also assumed that records
717 * of a block have the same format (i.e.: the same features enabled).
719 * \param[in,out] hdr Header of the block of records to remap.
720 * \param[in,out] last_hdr Last header, don't read past this point.
721 * \param[in] flags Flags describing the fields to keep.
723 static void changelog_block_trim_ext(struct llog_rec_hdr *hdr,
724 struct llog_rec_hdr *last_hdr,
725 enum changelog_rec_flags flags)
727 if (hdr->lrh_type != CHANGELOG_REC)
731 struct changelog_rec *rec = (struct changelog_rec *)(hdr + 1);
733 changelog_remap_rec(rec, rec->cr_flags & flags);
734 hdr = llog_rec_hdr_next(hdr);
735 } while ((char *)hdr <= (char *)last_hdr);
739 * Implementation of the llog_operations::lop_next_block
741 * This function finds the the next llog block to return which contains
742 * record with required index. It is main part of llog processing.
744 * \param[in] env execution environment
745 * \param[in] loghandle llog handle of the current llog
746 * \param[in,out] cur_idx index preceeding cur_offset
747 * \param[in] next_idx target index to find
748 * \param[in,out] cur_offset furtherst point read in the file
749 * \param[in] buf pointer to data buffer to fill
750 * \param[in] len required len to read, it is
751 * usually llog chunk_size.
753 * \retval 0 on successful buffer read
754 * \retval negative value on error
756 static int llog_osd_next_block(const struct lu_env *env,
757 struct llog_handle *loghandle, int *cur_idx,
758 int next_idx, __u64 *cur_offset, void *buf,
761 struct llog_thread_info *lgi = llog_info(env);
763 struct dt_device *dt;
766 int last_idx = *cur_idx;
767 __u64 last_offset = *cur_offset;
768 bool force_mini_rec = false;
775 chunk_size = loghandle->lgh_hdr->llh_hdr.lrh_len;
776 if (len == 0 || len & (chunk_size - 1))
779 CDEBUG(D_OTHER, "looking for log index %u (cur idx %u off "LPU64")\n",
780 next_idx, *cur_idx, *cur_offset);
783 LASSERT(loghandle->lgh_ctxt);
785 o = loghandle->lgh_obj;
787 LASSERT(dt_object_exists(o));
788 dt = lu2dt_dev(o->do_lu.lo_dev);
791 rc = dt_attr_get(env, o, &lgi->lgi_attr);
795 while (*cur_offset < lgi->lgi_attr.la_size) {
796 struct llog_rec_hdr *rec, *last_rec;
797 struct llog_rec_tail *tail;
799 llog_skip_over(loghandle, cur_offset, *cur_idx,
800 next_idx, chunk_size, force_mini_rec);
802 /* read up to next llog chunk_size block */
803 lgi->lgi_buf.lb_len = chunk_size -
804 (*cur_offset & (chunk_size - 1));
805 lgi->lgi_buf.lb_buf = buf;
807 rc = dt_read(env, o, &lgi->lgi_buf, cur_offset);
809 if (rc == -EBADR && !force_mini_rec)
812 CERROR("%s: can't read llog block from log "DFID
813 " offset "LPU64": rc = %d\n",
814 o->do_lu.lo_dev->ld_obd->obd_name,
815 PFID(lu_object_fid(&o->do_lu)), *cur_offset,
821 /* signal the end of the valid buffer to
823 memset(buf + rc, 0, len - rc);
826 if (rc == 0) { /* end of file, nothing to do */
832 if (rc < sizeof(*tail)) {
836 CERROR("%s: invalid llog block at log id "DOSTID"/%u "
838 o->do_lu.lo_dev->ld_obd->obd_name,
839 POSTID(&loghandle->lgh_id.lgl_oi),
840 loghandle->lgh_id.lgl_ogen, *cur_offset);
841 GOTO(out, rc = -EINVAL);
845 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
846 lustre_swab_llog_rec(rec);
848 tail = (struct llog_rec_tail *)((char *)buf + rc -
849 sizeof(struct llog_rec_tail));
850 /* get the last record in block */
851 last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
854 if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
855 lustre_swab_llog_rec(last_rec);
856 LASSERT(last_rec->lrh_index == tail->lrt_index);
858 *cur_idx = tail->lrt_index;
860 /* this shouldn't happen */
861 if (tail->lrt_index == 0) {
862 CERROR("%s: invalid llog tail at log id "DOSTID"/%u "
863 "offset "LPU64" bytes %d\n",
864 o->do_lu.lo_dev->ld_obd->obd_name,
865 POSTID(&loghandle->lgh_id.lgl_oi),
866 loghandle->lgh_id.lgl_ogen, *cur_offset, rc);
867 GOTO(out, rc = -EINVAL);
869 if (tail->lrt_index < next_idx) {
871 last_offset = *cur_offset;
875 /* sanity check that the start of the new buffer is no farther
876 * than the record that we wanted. This shouldn't happen. */
877 if (rec->lrh_index > next_idx) {
878 if (!force_mini_rec && next_idx > last_idx)
881 CERROR("%s: missed desired record? %u > %u\n",
882 o->do_lu.lo_dev->ld_obd->obd_name,
883 rec->lrh_index, next_idx);
884 GOTO(out, rc = -ENOENT);
887 /* Trim unsupported extensions for compat w/ older clients */
888 if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_JOBID))
889 changelog_block_trim_ext(rec, last_rec,
890 CLF_VERSION | CLF_RENAME);
895 /* Note: because there are some pad records in the
896 * llog, so llog_skip_over() might skip too much
897 * records, let's try skip again with minimum record */
898 force_mini_rec = true;
899 *cur_offset = last_offset;
902 GOTO(out, rc = -EIO);
908 * Implementation of the llog_operations::lop_prev_block
910 * This function finds the llog block to return which contains
911 * record with required index but in reverse order - from end of llog
913 * It is main part of reverse llog processing.
915 * \param[in] env execution environment
916 * \param[in] loghandle llog handle of the current llog
917 * \param[in] prev_idx target index to find
918 * \param[in] buf pointer to data buffer to fill
919 * \param[in] len required len to read, it is llog_chunk_size usually.
921 * \retval 0 on successful buffer read
922 * \retval negative value on error
924 static int llog_osd_prev_block(const struct lu_env *env,
925 struct llog_handle *loghandle,
926 int prev_idx, void *buf, int len)
928 struct llog_thread_info *lgi = llog_info(env);
930 struct dt_device *dt;
937 chunk_size = loghandle->lgh_hdr->llh_hdr.lrh_len;
938 if (len == 0 || len & (chunk_size - 1))
941 CDEBUG(D_OTHER, "looking for log index %u\n", prev_idx);
944 LASSERT(loghandle->lgh_ctxt);
946 o = loghandle->lgh_obj;
948 LASSERT(dt_object_exists(o));
949 dt = lu2dt_dev(o->do_lu.lo_dev);
952 /* Let's only use mini record size for previous block read
954 cur_offset = chunk_size;
955 llog_skip_over(loghandle, &cur_offset, 0, prev_idx,
958 rc = dt_attr_get(env, o, &lgi->lgi_attr);
962 while (cur_offset < lgi->lgi_attr.la_size) {
963 struct llog_rec_hdr *rec, *last_rec;
964 struct llog_rec_tail *tail;
966 lgi->lgi_buf.lb_len = len;
967 lgi->lgi_buf.lb_buf = buf;
968 rc = dt_read(env, o, &lgi->lgi_buf, &cur_offset);
970 CERROR("%s: can't read llog block from log "DFID
971 " offset "LPU64": rc = %d\n",
972 o->do_lu.lo_dev->ld_obd->obd_name,
973 PFID(lu_object_fid(&o->do_lu)), cur_offset, rc);
977 if (rc == 0) /* end of file, nothing to do */
980 if (rc < sizeof(*tail)) {
981 CERROR("%s: invalid llog block at log id "DOSTID"/%u "
983 o->do_lu.lo_dev->ld_obd->obd_name,
984 POSTID(&loghandle->lgh_id.lgl_oi),
985 loghandle->lgh_id.lgl_ogen, cur_offset);
986 GOTO(out, rc = -EINVAL);
990 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
991 lustre_swab_llog_rec(rec);
993 tail = (struct llog_rec_tail *)((char *)buf + rc -
994 sizeof(struct llog_rec_tail));
995 /* get the last record in block */
996 last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
997 le32_to_cpu(tail->lrt_len));
999 if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
1000 lustre_swab_llog_rec(last_rec);
1001 LASSERT(last_rec->lrh_index == tail->lrt_index);
1003 /* this shouldn't happen */
1004 if (tail->lrt_index == 0) {
1005 CERROR("%s: invalid llog tail at log id "DOSTID"/%u "
1007 o->do_lu.lo_dev->ld_obd->obd_name,
1008 POSTID(&loghandle->lgh_id.lgl_oi),
1009 loghandle->lgh_id.lgl_ogen, cur_offset);
1010 GOTO(out, rc = -EINVAL);
1012 if (tail->lrt_index < prev_idx)
1015 /* sanity check that the start of the new buffer is no farther
1016 * than the record that we wanted. This shouldn't happen. */
1017 if (rec->lrh_index > prev_idx) {
1018 CERROR("%s: missed desired record? %u > %u\n",
1019 o->do_lu.lo_dev->ld_obd->obd_name,
1020 rec->lrh_index, prev_idx);
1021 GOTO(out, rc = -ENOENT);
1024 /* Trim unsupported extensions for compat w/ older clients */
1025 if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_JOBID))
1026 changelog_block_trim_ext(rec, last_rec,
1027 CLF_VERSION | CLF_RENAME);
1031 GOTO(out, rc = -EIO);
1037 * This is helper function to get llog directory object. It is used by named
1038 * llog operations to find/insert/delete llog entry from llog directory.
1040 * \param[in] env execution environment
1041 * \param[in] ctxt llog context
1043 * \retval dt_object of llog directory
1044 * \retval ERR_PTR of negative value on error
1046 static struct dt_object *llog_osd_dir_get(const struct lu_env *env,
1047 struct llog_ctxt *ctxt)
1049 struct dt_device *dt;
1050 struct dt_thread_info *dti = dt_info(env);
1051 struct dt_object *dir;
1054 dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
1055 if (ctxt->loc_dir == NULL) {
1056 rc = dt_root_get(env, dt, &dti->dti_fid);
1059 dir = dt_locate(env, dt, &dti->dti_fid);
1061 if (!IS_ERR(dir) && !dt_try_as_dir(env, dir)) {
1062 lu_object_put(env, &dir->do_lu);
1063 return ERR_PTR(-ENOTDIR);
1066 lu_object_get(&ctxt->loc_dir->do_lu);
1067 dir = ctxt->loc_dir;
1074 * Implementation of the llog_operations::lop_open
1076 * This function opens the llog by its logid or by name, it may open also
1077 * non existent llog and assing then new id to it.
1078 * The llog_open/llog_close pair works similar to lu_object_find/put,
1079 * the object may not exist prior open. The result of open is just dt_object
1080 * in the llog header.
1082 * \param[in] env execution environment
1083 * \param[in] handle llog handle of the current llog
1084 * \param[in] logid logid of llog to open (nameless llog)
1085 * \param[in] name name of llog to open (named llog)
1086 * \param[in] open_param
1087 * LLOG_OPEN_NEW - new llog, may not exist
1088 * LLOG_OPEN_EXIST - old llog, must exist
1090 * \retval 0 on successful open, llog_handle::lgh_obj
1091 * contains the dt_object of the llog.
1092 * \retval negative value on error
1094 static int llog_osd_open(const struct lu_env *env, struct llog_handle *handle,
1095 struct llog_logid *logid, char *name,
1096 enum llog_open_param open_param)
1098 struct llog_thread_info *lgi = llog_info(env);
1099 struct llog_ctxt *ctxt = handle->lgh_ctxt;
1100 struct dt_object *o;
1101 struct dt_device *dt;
1102 struct ls_device *ls;
1103 struct local_oid_storage *los = NULL;
1110 LASSERT(ctxt->loc_exp);
1111 LASSERT(ctxt->loc_exp->exp_obd);
1112 dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
1114 if (ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1115 struct lu_object_conf conf = { 0 };
1116 if (logid != NULL) {
1117 logid_to_fid(logid, &lgi->lgi_fid);
1119 /* If logid == NULL, then it means the caller needs
1120 * to allocate new FID (llog_cat_declare_add_rec()). */
1121 rc = obd_fid_alloc(env, ctxt->loc_exp,
1122 &lgi->lgi_fid, NULL);
1126 conf.loc_flags = LOC_F_NEW;
1129 o = dt_locate_at(env, dt, &lgi->lgi_fid,
1130 dt->dd_lu_dev.ld_site->ls_top_dev, &conf);
1137 ls = ls_device_get(dt);
1139 RETURN(PTR_ERR(ls));
1141 mutex_lock(&ls->ls_los_mutex);
1142 los = dt_los_find(ls, name != NULL ? FID_SEQ_LLOG_NAME : FID_SEQ_LLOG);
1143 mutex_unlock(&ls->ls_los_mutex);
1145 ls_device_put(env, ls);
1149 if (logid != NULL) {
1150 logid_to_fid(logid, &lgi->lgi_fid);
1152 struct dt_object *llog_dir;
1154 llog_dir = llog_osd_dir_get(env, ctxt);
1155 if (IS_ERR(llog_dir))
1156 GOTO(out, rc = PTR_ERR(llog_dir));
1157 dt_read_lock(env, llog_dir, 0);
1158 rc = dt_lookup_dir(env, llog_dir, name, &lgi->lgi_fid);
1159 dt_read_unlock(env, llog_dir);
1160 lu_object_put(env, &llog_dir->do_lu);
1161 if (rc == -ENOENT && open_param == LLOG_OPEN_NEW) {
1162 /* generate fid for new llog */
1163 rc = local_object_fid_generate(env, los,
1168 OBD_ALLOC(handle->lgh_name, strlen(name) + 1);
1169 if (handle->lgh_name)
1170 strcpy(handle->lgh_name, name);
1172 GOTO(out, rc = -ENOMEM);
1174 LASSERTF(open_param & LLOG_OPEN_NEW, "%#x\n", open_param);
1175 /* generate fid for new llog */
1176 rc = local_object_fid_generate(env, los, &lgi->lgi_fid);
1181 o = ls_locate(env, ls, &lgi->lgi_fid, NULL);
1183 GOTO(out_name, rc = PTR_ERR(o));
1186 /* No new llog is expected but doesn't exist */
1187 if (open_param != LLOG_OPEN_NEW && !dt_object_exists(o))
1188 GOTO(out_put, rc = -ENOENT);
1190 fid_to_logid(&lgi->lgi_fid, &handle->lgh_id);
1191 handle->lgh_obj = o;
1192 handle->private_data = los;
1193 LASSERT(handle->lgh_ctxt);
1198 lu_object_put(env, &o->do_lu);
1200 if (handle->lgh_name != NULL)
1201 OBD_FREE(handle->lgh_name, strlen(name) + 1);
1209 * Implementation of the llog_operations::lop_exist
1211 * This function checks that llog exists on storage.
1213 * \param[in] handle llog handle of the current llog
1215 * \retval true if llog object exists and is not just destroyed
1216 * \retval false if llog doesn't exist or just destroyed
1218 static int llog_osd_exist(struct llog_handle *handle)
1220 LASSERT(handle->lgh_obj);
1221 return (dt_object_exists(handle->lgh_obj) &&
1222 !lu_object_is_dying(handle->lgh_obj->do_lu.lo_header));
1226 * Get dir for regular fid log object
1228 * Get directory for regular fid log object, and these regular fid log
1229 * object will be inserted under this directory, to satisfy the FS
1230 * consistency check, e2fsck etc.
1232 * \param [in] env execution environment
1233 * \param [in] dto llog object
1235 * \retval pointer to the directory if it is found.
1236 * \retval ERR_PTR(negative errno) if it fails.
1238 struct dt_object *llog_osd_get_regular_fid_dir(const struct lu_env *env,
1239 struct dt_object *dto)
1241 struct llog_thread_info *lgi = llog_info(env);
1242 struct seq_server_site *ss = dto->do_lu.lo_dev->ld_site->ld_seq_site;
1243 struct lu_seq_range *range = &lgi->lgi_range;
1244 struct lu_fid *dir_fid = &lgi->lgi_fid;
1245 struct dt_object *dir;
1249 fld_range_set_any(range);
1250 LASSERT(ss != NULL);
1251 rc = ss->ss_server_fld->lsf_seq_lookup(env, ss->ss_server_fld,
1252 fid_seq(lu_object_fid(&dto->do_lu)), range);
1254 RETURN(ERR_PTR(rc));
1256 lu_update_log_dir_fid(dir_fid, range->lsr_index);
1257 dir = dt_locate(env, lu2dt_dev(dto->do_lu.lo_dev), dir_fid);
1261 if (!dt_try_as_dir(env, dir)) {
1262 lu_object_put(env, &dir->do_lu);
1263 RETURN(ERR_PTR(-ENOTDIR));
1270 * Add llog object with regular FID to name entry
1272 * Add llog object with regular FID to name space, and each llog
1273 * object on each MDT will be /update_log_dir/[seq:oid:ver],
1274 * so to satisfy the namespace consistency check, e2fsck etc.
1276 * \param [in] env execution environment
1277 * \param [in] dto llog object
1278 * \param [in] th thandle
1279 * \param [in] declare if it is declare or execution
1281 * \retval 0 if insertion succeeds.
1282 * \retval negative errno if insertion fails.
1285 llog_osd_regular_fid_add_name_entry(const struct lu_env *env,
1286 struct dt_object *dto,
1287 struct thandle *th, bool declare)
1289 struct llog_thread_info *lgi = llog_info(env);
1290 const struct lu_fid *fid = lu_object_fid(&dto->do_lu);
1291 struct dt_insert_rec *rec = &lgi->lgi_dt_rec;
1292 struct dt_object *dir;
1293 char *name = lgi->lgi_name;
1297 if (!fid_is_norm(fid))
1300 dir = llog_osd_get_regular_fid_dir(env, dto);
1302 RETURN(PTR_ERR(dir));
1305 rec->rec_type = S_IFREG;
1306 snprintf(name, sizeof(lgi->lgi_name), DFID, PFID(fid));
1307 dt_write_lock(env, dir, 0);
1309 rc = dt_declare_insert(env, dir, (struct dt_rec *)rec,
1310 (struct dt_key *)name, th);
1312 rc = dt_insert(env, dir, (struct dt_rec *)rec,
1313 (struct dt_key *)name, th, 1);
1315 dt_write_unlock(env, dir);
1317 lu_object_put(env, &dir->do_lu);
1323 * Implementation of the llog_operations::lop_declare_create
1325 * This function declares the llog create. It declares also name insert
1326 * into llog directory in case of named llog.
1328 * \param[in] env execution environment
1329 * \param[in] res llog handle of the current llog
1330 * \param[in] th current transaction handle
1332 * \retval 0 on successful create declaration
1333 * \retval negative value on error
1335 static int llog_osd_declare_create(const struct lu_env *env,
1336 struct llog_handle *res, struct thandle *th)
1338 struct llog_thread_info *lgi = llog_info(env);
1339 struct dt_insert_rec *rec = &lgi->lgi_dt_rec;
1340 struct local_oid_storage *los;
1341 struct dt_object *o;
1346 LASSERT(res->lgh_obj);
1349 /* object can be created by another thread */
1351 if (dt_object_exists(o))
1354 if (res->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1355 struct llog_thread_info *lgi = llog_info(env);
1357 lgi->lgi_attr.la_valid = LA_MODE | LA_SIZE;
1358 lgi->lgi_attr.la_size = 0;
1359 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1360 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1362 rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL,
1368 rc = llog_osd_regular_fid_add_name_entry(env, o, th, true);
1372 los = res->private_data;
1375 rc = llog_osd_declare_new_object(env, los, o, th);
1379 /* do not declare header initialization here as it's declared
1380 * in llog_osd_declare_write_rec() which is always called */
1382 if (res->lgh_name) {
1383 struct dt_object *llog_dir;
1385 llog_dir = llog_osd_dir_get(env, res->lgh_ctxt);
1386 if (IS_ERR(llog_dir))
1387 RETURN(PTR_ERR(llog_dir));
1388 logid_to_fid(&res->lgh_id, &lgi->lgi_fid);
1389 rec->rec_fid = &lgi->lgi_fid;
1390 rec->rec_type = S_IFREG;
1391 rc = dt_declare_insert(env, llog_dir,
1392 (struct dt_rec *)rec,
1393 (struct dt_key *)res->lgh_name, th);
1394 lu_object_put(env, &llog_dir->do_lu);
1396 CERROR("%s: can't declare named llog %s: rc = %d\n",
1397 o->do_lu.lo_dev->ld_obd->obd_name,
1404 * Implementation of the llog_operations::lop_create
1406 * This function creates the llog according with llog_handle::lgh_obj
1407 * and llog_handle::lgh_name.
1409 * \param[in] env execution environment
1410 * \param[in] res llog handle of the current llog
1411 * \param[in] th current transaction handle
1413 * \retval 0 on successful create
1414 * \retval negative value on error
1416 static int llog_osd_create(const struct lu_env *env, struct llog_handle *res,
1419 struct llog_thread_info *lgi = llog_info(env);
1420 struct dt_insert_rec *rec = &lgi->lgi_dt_rec;
1421 struct local_oid_storage *los;
1422 struct dt_object *o;
1431 /* llog can be already created */
1432 if (dt_object_exists(o))
1435 if (res->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1436 struct llog_thread_info *lgi = llog_info(env);
1438 lgi->lgi_attr.la_valid = LA_MODE | LA_SIZE | LA_TYPE;
1439 lgi->lgi_attr.la_size = 0;
1440 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1441 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1443 dt_write_lock(env, o, 0);
1444 rc = dt_create(env, o, &lgi->lgi_attr, NULL,
1446 dt_write_unlock(env, o);
1450 rc = llog_osd_regular_fid_add_name_entry(env, o, th, false);
1455 los = res->private_data;
1458 dt_write_lock(env, o, 0);
1459 if (!dt_object_exists(o))
1460 rc = llog_osd_create_new_object(env, los, o, th);
1464 dt_write_unlock(env, o);
1468 if (res->lgh_name) {
1469 struct dt_object *llog_dir;
1471 llog_dir = llog_osd_dir_get(env, res->lgh_ctxt);
1472 if (IS_ERR(llog_dir))
1473 RETURN(PTR_ERR(llog_dir));
1475 logid_to_fid(&res->lgh_id, &lgi->lgi_fid);
1476 rec->rec_fid = &lgi->lgi_fid;
1477 rec->rec_type = S_IFREG;
1478 dt_read_lock(env, llog_dir, 0);
1479 rc = dt_insert(env, llog_dir, (struct dt_rec *)rec,
1480 (struct dt_key *)res->lgh_name,
1482 dt_read_unlock(env, llog_dir);
1483 lu_object_put(env, &llog_dir->do_lu);
1485 CERROR("%s: can't create named llog %s: rc = %d\n",
1486 o->do_lu.lo_dev->ld_obd->obd_name,
1493 * Implementation of the llog_operations::lop_close
1495 * This function closes the llog. It just put llog object and referenced
1498 * \param[in] env execution environment
1499 * \param[in] handle llog handle of the current llog
1501 * \retval 0 on successful llog close
1502 * \retval negative value on error
1504 static int llog_osd_close(const struct lu_env *env, struct llog_handle *handle)
1506 struct local_oid_storage *los;
1511 LASSERT(handle->lgh_obj);
1513 if (handle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1514 /* Remove the object from the cache, otherwise it may
1515 * hold LOD being released during cleanup process */
1516 lu_object_put_nocache(env, &handle->lgh_obj->do_lu);
1517 LASSERT(handle->private_data == NULL);
1520 lu_object_put(env, &handle->lgh_obj->do_lu);
1522 los = handle->private_data;
1526 if (handle->lgh_name)
1527 OBD_FREE(handle->lgh_name, strlen(handle->lgh_name) + 1);
1533 * delete llog object name entry
1535 * Delete llog object (with regular FID) from name space (under
1538 * \param [in] env execution environment
1539 * \param [in] dto llog object
1540 * \param [in] th thandle
1541 * \param [in] declare if it is declare or execution
1543 * \retval 0 if deletion succeeds.
1544 * \retval negative errno if deletion fails.
1547 llog_osd_regular_fid_del_name_entry(const struct lu_env *env,
1548 struct dt_object *dto,
1549 struct thandle *th, bool declare)
1551 struct llog_thread_info *lgi = llog_info(env);
1552 const struct lu_fid *fid = lu_object_fid(&dto->do_lu);
1553 struct dt_object *dir;
1554 char *name = lgi->lgi_name;
1558 if (!fid_is_norm(fid))
1561 dir = llog_osd_get_regular_fid_dir(env, dto);
1563 RETURN(PTR_ERR(dir));
1565 snprintf(name, sizeof(lgi->lgi_name), DFID, PFID(fid));
1566 dt_write_lock(env, dir, 0);
1568 rc = dt_declare_delete(env, dir, (struct dt_key *)name,
1571 rc = dt_delete(env, dir, (struct dt_key *)name, th);
1573 dt_write_unlock(env, dir);
1575 lu_object_put(env, &dir->do_lu);
1580 * Implementation of the llog_operations::lop_declare_destroy
1582 * This function declare destroys the llog and deletes also entry in the
1583 * llog directory in case of named llog. Llog should be opened prior that.
1585 * \param[in] env execution environment
1586 * \param[in] loghandle llog handle of the current llog
1588 * \retval 0 on successful destroy
1589 * \retval negative value on error
1591 static int llog_osd_declare_destroy(const struct lu_env *env,
1592 struct llog_handle *loghandle,
1595 struct llog_ctxt *ctxt;
1596 struct dt_object *o, *llog_dir = NULL;
1601 ctxt = loghandle->lgh_ctxt;
1604 o = loghandle->lgh_obj;
1607 if (loghandle->lgh_name) {
1608 llog_dir = llog_osd_dir_get(env, ctxt);
1609 if (IS_ERR(llog_dir))
1610 RETURN(PTR_ERR(llog_dir));
1612 rc = dt_declare_delete(env, llog_dir,
1613 (struct dt_key *)loghandle->lgh_name,
1619 rc = dt_declare_ref_del(env, o, th);
1623 rc = dt_declare_destroy(env, o, th);
1627 if (loghandle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1628 rc = llog_osd_regular_fid_del_name_entry(env, o, th, true);
1634 if (!(IS_ERR_OR_NULL(llog_dir)))
1635 lu_object_put(env, &llog_dir->do_lu);
1642 * Implementation of the llog_operations::lop_destroy
1644 * This function destroys the llog and deletes also entry in the
1645 * llog directory in case of named llog. Llog should be opened prior that.
1646 * Destroy method is not part of external transaction and does everything
1649 * \param[in] env execution environment
1650 * \param[in] loghandle llog handle of the current llog
1652 * \retval 0 on successful destroy
1653 * \retval negative value on error
1655 static int llog_osd_destroy(const struct lu_env *env,
1656 struct llog_handle *loghandle, struct thandle *th)
1658 struct llog_ctxt *ctxt;
1659 struct dt_object *o, *llog_dir = NULL;
1664 ctxt = loghandle->lgh_ctxt;
1665 LASSERT(ctxt != NULL);
1667 o = loghandle->lgh_obj;
1670 dt_write_lock(env, o, 0);
1671 if (!dt_object_exists(o))
1672 GOTO(out_unlock, rc = 0);
1674 if (loghandle->lgh_name) {
1675 llog_dir = llog_osd_dir_get(env, ctxt);
1676 if (IS_ERR(llog_dir))
1677 GOTO(out_unlock, rc = PTR_ERR(llog_dir));
1679 dt_read_lock(env, llog_dir, 0);
1680 rc = dt_delete(env, llog_dir,
1681 (struct dt_key *)loghandle->lgh_name,
1683 dt_read_unlock(env, llog_dir);
1685 CERROR("%s: can't remove llog %s: rc = %d\n",
1686 o->do_lu.lo_dev->ld_obd->obd_name,
1687 loghandle->lgh_name, rc);
1688 GOTO(out_unlock, rc);
1692 dt_ref_del(env, o, th);
1693 rc = dt_destroy(env, o, th);
1695 GOTO(out_unlock, rc);
1697 if (loghandle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1698 rc = llog_osd_regular_fid_del_name_entry(env, o, th, false);
1700 GOTO(out_unlock, rc);
1704 dt_write_unlock(env, o);
1705 if (!(IS_ERR_OR_NULL(llog_dir)))
1706 lu_object_put(env, &llog_dir->do_lu);
1711 * Implementation of the llog_operations::lop_setup
1713 * This function setup the llog on local storage.
1715 * \param[in] env execution environment
1716 * \param[in] obd obd device the llog belongs to
1717 * \param[in] olg the llog group, it is always zero group now.
1718 * \param[in] ctxt_idx the llog index, it defines the purpose of this llog.
1719 * Every new llog type have to use own index.
1720 * \param[in] disk_obd the storage obd, where llog is stored.
1722 * \retval 0 on successful llog setup
1723 * \retval negative value on error
1725 static int llog_osd_setup(const struct lu_env *env, struct obd_device *obd,
1726 struct obd_llog_group *olg, int ctxt_idx,
1727 struct obd_device *disk_obd)
1729 struct llog_thread_info *lgi = llog_info(env);
1730 struct llog_ctxt *ctxt;
1735 LASSERT(olg->olg_ctxts[ctxt_idx]);
1737 ctxt = llog_ctxt_get(olg->olg_ctxts[ctxt_idx]);
1740 if (disk_obd == NULL)
1743 /* initialize data allowing to generate new fids,
1744 * literally we need a sequece */
1745 lgi->lgi_fid.f_seq = FID_SEQ_LLOG;
1746 lgi->lgi_fid.f_oid = 1;
1747 lgi->lgi_fid.f_ver = 0;
1748 rc = local_oid_storage_init(env, disk_obd->obd_lvfs_ctxt.dt,
1750 &ctxt->loc_los_nameless);
1754 lgi->lgi_fid.f_seq = FID_SEQ_LLOG_NAME;
1755 lgi->lgi_fid.f_oid = 1;
1756 lgi->lgi_fid.f_ver = 0;
1757 rc = local_oid_storage_init(env, disk_obd->obd_lvfs_ctxt.dt,
1759 &ctxt->loc_los_named);
1761 local_oid_storage_fini(env, ctxt->loc_los_nameless);
1762 ctxt->loc_los_nameless = NULL;
1768 llog_ctxt_put(ctxt);
1773 * Implementation of the llog_operations::lop_cleanup
1775 * This function cleanups the llog on local storage.
1777 * \param[in] env execution environment
1778 * \param[in] ctxt the llog context
1782 static int llog_osd_cleanup(const struct lu_env *env, struct llog_ctxt *ctxt)
1784 if (ctxt->loc_los_nameless != NULL) {
1785 local_oid_storage_fini(env, ctxt->loc_los_nameless);
1786 ctxt->loc_los_nameless = NULL;
1789 if (ctxt->loc_los_named != NULL) {
1790 local_oid_storage_fini(env, ctxt->loc_los_named);
1791 ctxt->loc_los_named = NULL;
1797 struct llog_operations llog_osd_ops = {
1798 .lop_next_block = llog_osd_next_block,
1799 .lop_prev_block = llog_osd_prev_block,
1800 .lop_read_header = llog_osd_read_header,
1801 .lop_declare_destroy = llog_osd_declare_destroy,
1802 .lop_destroy = llog_osd_destroy,
1803 .lop_setup = llog_osd_setup,
1804 .lop_cleanup = llog_osd_cleanup,
1805 .lop_open = llog_osd_open,
1806 .lop_exist = llog_osd_exist,
1807 .lop_declare_create = llog_osd_declare_create,
1808 .lop_create = llog_osd_create,
1809 .lop_declare_write_rec = llog_osd_declare_write_rec,
1810 .lop_write_rec = llog_osd_write_rec,
1811 .lop_close = llog_osd_close,
1813 EXPORT_SYMBOL(llog_osd_ops);
1815 struct llog_operations llog_common_cat_ops = {
1816 .lop_next_block = llog_osd_next_block,
1817 .lop_prev_block = llog_osd_prev_block,
1818 .lop_read_header = llog_osd_read_header,
1819 .lop_declare_destroy = llog_osd_declare_destroy,
1820 .lop_destroy = llog_osd_destroy,
1821 .lop_setup = llog_osd_setup,
1822 .lop_cleanup = llog_osd_cleanup,
1823 .lop_open = llog_osd_open,
1824 .lop_exist = llog_osd_exist,
1825 .lop_declare_create = llog_osd_declare_create,
1826 .lop_create = llog_osd_create,
1827 .lop_declare_write_rec = llog_osd_declare_write_rec,
1828 .lop_write_rec = llog_osd_write_rec,
1829 .lop_close = llog_osd_close,
1830 .lop_add = llog_cat_add_rec,
1831 .lop_declare_add = llog_cat_declare_add_rec,
1833 EXPORT_SYMBOL(llog_common_cat_ops);
1836 * Read the special file which contains the list of llog catalogs IDs
1838 * This function reads the CATALOGS file which contains the array of llog
1839 * catalogs IDs. The main purpose of this file is to store OSP llogs indexed
1840 * by OST/MDT number.
1842 * \param[in] env execution environment
1843 * \param[in] d corresponding storage device
1844 * \param[in] idx position to start from, usually OST/MDT index
1845 * \param[in] count how many catalog IDs to read
1846 * \param[out] idarray the buffer for the data. If it is NULL then
1847 * function returns just number of catalog IDs
1849 * \param[in] fid LLOG_CATALOGS_OID for CATALOG object
1851 * \retval 0 on successful read of catalog IDs
1852 * \retval negative value on error
1853 * \retval positive value which is number of records in
1854 * the file if \a idarray is NULL
1856 int llog_osd_get_cat_list(const struct lu_env *env, struct dt_device *d,
1857 int idx, int count, struct llog_catid *idarray,
1858 const struct lu_fid *fid)
1860 struct llog_thread_info *lgi = llog_info(env);
1861 struct dt_object *o = NULL;
1869 size = sizeof(*idarray) * count;
1870 lgi->lgi_off = idx * sizeof(*idarray);
1872 lgi->lgi_fid = *fid;
1873 o = dt_locate(env, d, &lgi->lgi_fid);
1877 if (!dt_object_exists(o)) {
1878 th = dt_trans_create(env, d);
1880 GOTO(out, rc = PTR_ERR(th));
1882 lgi->lgi_attr.la_valid = LA_MODE;
1883 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1884 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1886 th->th_wait_submit = 1;
1887 /* Make the llog object creation synchronization, so
1888 * it will be reliable to the reference, especially
1889 * for remote reference */
1892 rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL,
1895 GOTO(out_trans, rc);
1897 rc = dt_trans_start_local(env, d, th);
1899 GOTO(out_trans, rc);
1901 dt_write_lock(env, o, 0);
1902 if (!dt_object_exists(o))
1903 rc = dt_create(env, o, &lgi->lgi_attr, NULL,
1905 dt_write_unlock(env, o);
1907 dt_trans_stop(env, d, th);
1912 rc = dt_attr_get(env, o, &lgi->lgi_attr);
1916 if (!S_ISREG(lgi->lgi_attr.la_mode)) {
1917 CERROR("%s: CATALOGS is not a regular file!: mode = %o\n",
1918 o->do_lu.lo_dev->ld_obd->obd_name,
1919 lgi->lgi_attr.la_mode);
1920 GOTO(out, rc = -ENOENT);
1923 CDEBUG(D_CONFIG, "cat list: disk size=%d, read=%d\n",
1924 (int)lgi->lgi_attr.la_size, size);
1926 /* return just number of llogs */
1927 if (idarray == NULL) {
1928 rc = lgi->lgi_attr.la_size / sizeof(*idarray);
1932 /* read for new ost index or for empty file */
1933 memset(idarray, 0, size);
1934 if (lgi->lgi_attr.la_size <= lgi->lgi_off)
1936 if (lgi->lgi_attr.la_size < lgi->lgi_off + size)
1937 size = lgi->lgi_attr.la_size - lgi->lgi_off;
1939 lgi->lgi_buf.lb_buf = idarray;
1940 lgi->lgi_buf.lb_len = size;
1941 rc = dt_record_read(env, o, &lgi->lgi_buf, &lgi->lgi_off);
1942 /* -EFAULT means the llog is a sparse file. This is not an error
1943 * after arbitrary OST index is supported. */
1944 if (rc < 0 && rc != -EFAULT) {
1945 CERROR("%s: error reading CATALOGS: rc = %d\n",
1946 o->do_lu.lo_dev->ld_obd->obd_name, rc);
1952 lu_object_put(env, &o->do_lu);
1955 EXPORT_SYMBOL(llog_osd_get_cat_list);
1958 * Write the special file which contains the list of llog catalogs IDs
1960 * This function writes the CATALOG file which contains the array of llog
1961 * catalogs IDs. It is used mostly to store OSP llogs indexed by OST/MDT
1964 * \param[in] env execution environment
1965 * \param[in] d corresponding storage device
1966 * \param[in] idx position to start from, usually OST/MDT index
1967 * \param[in] count how many catalog IDs to write
1968 * \param[out] idarray the buffer with the data to write.
1969 * \param[in] fid LLOG_CATALOGS_OID for CATALOG object
1971 * \retval 0 on successful write of catalog IDs
1972 * \retval negative value on error
1974 int llog_osd_put_cat_list(const struct lu_env *env, struct dt_device *d,
1975 int idx, int count, struct llog_catid *idarray,
1976 const struct lu_fid *fid)
1978 struct llog_thread_info *lgi = llog_info(env);
1979 struct dt_object *o = NULL;
1988 size = sizeof(*idarray) * count;
1989 lgi->lgi_off = idx * sizeof(*idarray);
1990 lgi->lgi_fid = *fid;
1992 o = dt_locate(env, d, &lgi->lgi_fid);
1996 if (!dt_object_exists(o))
1997 GOTO(out, rc = -ENOENT);
1999 rc = dt_attr_get(env, o, &lgi->lgi_attr);
2003 if (!S_ISREG(lgi->lgi_attr.la_mode)) {
2004 CERROR("%s: CATALOGS is not a regular file!: mode = %o\n",
2005 o->do_lu.lo_dev->ld_obd->obd_name,
2006 lgi->lgi_attr.la_mode);
2007 GOTO(out, rc = -ENOENT);
2010 th = dt_trans_create(env, d);
2012 GOTO(out, rc = PTR_ERR(th));
2014 lgi->lgi_buf.lb_len = size;
2015 lgi->lgi_buf.lb_buf = idarray;
2016 rc = dt_declare_record_write(env, o, &lgi->lgi_buf, lgi->lgi_off, th);
2018 GOTO(out_trans, rc);
2020 /* For update log, this happens during initialization,
2021 * see lod_sub_prep_llog(), and we need make sure catlog
2022 * file ID is written to catlist file(committed) before
2023 * cross-MDT operation write update records to catlog FILE,
2024 * otherwise, during failover these update records might
2026 if (fid_is_update_log(fid))
2029 rc = dt_trans_start_local(env, d, th);
2031 GOTO(out_trans, rc);
2033 th->th_wait_submit = 1;
2035 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
2037 CDEBUG(D_INODE, "can't write CATALOGS at index %d: rc = %d\n",
2040 dt_trans_stop(env, d, th);
2042 lu_object_put(env, &o->do_lu);
2045 EXPORT_SYMBOL(llog_osd_put_cat_list);