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, 2015, 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
47 #include <dt_object.h>
48 #include <llog_swab.h>
49 #include <lustre_fid.h>
51 #include <obd_class.h>
53 #include "llog_internal.h"
54 #include "local_storage.h"
57 * Implementation of the llog_operations::lop_declare_create
59 * This function is a wrapper over local_storage API function
60 * local_object_declare_create().
62 * \param[in] env execution environment
63 * \param[in] los local_storage for bottom storage device
64 * \param[in] o dt_object to create
65 * \param[in] th current transaction handle
67 * \retval 0 on successful declaration of the new object
68 * \retval negative error if declaration was failed
70 static int llog_osd_declare_new_object(const struct lu_env *env,
71 struct local_oid_storage *los,
75 struct llog_thread_info *lgi = llog_info(env);
77 lgi->lgi_attr.la_valid = LA_MODE;
78 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
79 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
81 return local_object_declare_create(env, los, o, &lgi->lgi_attr,
86 * Implementation of the llog_operations::lop_create
88 * This function is a wrapper over local_storage API function
89 * local_object_create().
91 * \param[in] env execution environment
92 * \param[in] los local_storage for bottom storage device
93 * \param[in] o dt_object to create
94 * \param[in] th current transaction handle
96 * \retval 0 on successful creation of the new object
97 * \retval negative error if creation was failed
99 static int llog_osd_create_new_object(const struct lu_env *env,
100 struct local_oid_storage *los,
104 struct llog_thread_info *lgi = llog_info(env);
106 lgi->lgi_attr.la_valid = LA_MODE;
107 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
108 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
110 return local_object_create(env, los, o, &lgi->lgi_attr,
115 * Implementation of the llog_operations::lop_exist
117 * This function checks that llog exists on storage.
119 * \param[in] handle llog handle of the current llog
121 * \retval true if llog object exists and is not just destroyed
122 * \retval false if llog doesn't exist or just destroyed
124 static int llog_osd_exist(struct llog_handle *handle)
126 LASSERT(handle->lgh_obj);
127 return dt_object_exists(handle->lgh_obj) &&
128 !lu_object_is_dying(handle->lgh_obj->do_lu.lo_header);
131 static void *rec_tail(struct llog_rec_hdr *rec)
133 return (void *)((char *)rec + rec->lrh_len -
134 sizeof(struct llog_rec_tail));
138 * Write a padding record to the llog
140 * This function writes a padding record to the end of llog. That may
141 * be needed if llog contains records of variable size, e.g. config logs
143 * The padding record just aligns llog to the llog chunk_size boundary if
144 * the current record doesn't fit in the remaining space.
146 * It allocates full length to avoid two separate writes for header and tail.
147 * Such 2-steps scheme needs extra protection and complex error handling.
149 * \param[in] env execution environment
150 * \param[in] o dt_object to create
151 * \param[in,out] off pointer to the padding start offset
152 * \param[in] len padding length
153 * \param[in] index index of the padding record in a llog
154 * \param[in] th current transaction handle
156 * \retval 0 on successful padding write
157 * \retval negative error if write failed
159 static int llog_osd_pad(const struct lu_env *env, struct dt_object *o,
160 loff_t *off, int len, int index, struct thandle *th)
162 struct llog_thread_info *lgi = llog_info(env);
163 struct llog_rec_hdr *rec;
164 struct llog_rec_tail *tail;
171 LASSERT(len >= LLOG_MIN_REC_SIZE && (len & 0x7) == 0);
178 rec->lrh_index = index;
179 rec->lrh_type = LLOG_PAD_MAGIC;
181 tail = rec_tail(rec);
183 tail->lrt_index = index;
185 lgi->lgi_buf.lb_buf = rec;
186 lgi->lgi_buf.lb_len = len;
187 rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
189 CERROR("%s: error writing padding record: rc = %d\n",
190 o->do_lu.lo_dev->ld_obd->obd_name, rc);
197 * Implementation of the llog_operations::lop_read_header
199 * This function reads the current llog header from the bottom storage
202 * \param[in] env execution environment
203 * \param[in] handle llog handle of the current llog
205 * \retval 0 on successful header read
206 * \retval negative error if read failed
208 static int llog_osd_read_header(const struct lu_env *env,
209 struct llog_handle *handle)
211 struct llog_rec_hdr *llh_hdr;
213 struct llog_thread_info *lgi;
214 enum llog_flag flags;
222 lgi = llog_info(env);
224 rc = dt_attr_get(env, o, &lgi->lgi_attr);
228 LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
230 if (lgi->lgi_attr.la_size == 0) {
231 CDEBUG(D_HA, "not reading header from 0-byte log\n");
235 flags = handle->lgh_hdr->llh_flags;
238 lgi->lgi_buf.lb_buf = handle->lgh_hdr;
239 lgi->lgi_buf.lb_len = handle->lgh_hdr_size;
240 rc = dt_read(env, o, &lgi->lgi_buf, &lgi->lgi_off);
241 llh_hdr = &handle->lgh_hdr->llh_hdr;
242 if (rc < sizeof(*llh_hdr) || rc < llh_hdr->lrh_len) {
243 CERROR("%s: error reading "DFID" log header size %d: rc = %d\n",
244 o->do_lu.lo_dev->ld_obd->obd_name,
245 PFID(lu_object_fid(&o->do_lu)), rc < 0 ? 0 : rc,
254 if (LLOG_REC_HDR_NEEDS_SWABBING(llh_hdr))
255 lustre_swab_llog_hdr(handle->lgh_hdr);
257 if (llh_hdr->lrh_type != LLOG_HDR_MAGIC) {
258 CERROR("%s: bad log %s "DFID" header magic: %#x "
259 "(expected %#x)\n", o->do_lu.lo_dev->ld_obd->obd_name,
260 handle->lgh_name ? handle->lgh_name : "",
261 PFID(lu_object_fid(&o->do_lu)),
262 llh_hdr->lrh_type, LLOG_HDR_MAGIC);
264 } else if (llh_hdr->lrh_len < LLOG_MIN_CHUNK_SIZE ||
265 llh_hdr->lrh_len > handle->lgh_hdr_size) {
266 CERROR("%s: incorrectly sized log %s "DFID" header: "
267 "%#x (expected at least %#x)\n"
268 "you may need to re-run lconf --write_conf.\n",
269 o->do_lu.lo_dev->ld_obd->obd_name,
270 handle->lgh_name ? handle->lgh_name : "",
271 PFID(lu_object_fid(&o->do_lu)),
272 llh_hdr->lrh_len, LLOG_MIN_CHUNK_SIZE);
274 } else if (LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_index >
275 LLOG_HDR_BITMAP_SIZE(handle->lgh_hdr) ||
276 LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_len !=
278 CERROR("%s: incorrectly sized log %s "DFID" tailer: "
280 o->do_lu.lo_dev->ld_obd->obd_name,
281 handle->lgh_name ? handle->lgh_name : "",
282 PFID(lu_object_fid(&o->do_lu)),
283 LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_len, -EIO);
287 handle->lgh_hdr->llh_flags |= (flags & LLOG_F_EXT_MASK);
288 handle->lgh_last_idx = LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_index;
289 handle->lgh_write_offset = lgi->lgi_attr.la_size;
295 * Implementation of the llog_operations::lop_declare_write
297 * This function declares the new record write.
299 * \param[in] env execution environment
300 * \param[in] loghandle llog handle of the current llog
301 * \param[in] rec llog record header. This is a real header of the full
302 * llog record to write. This is the beginning of buffer
303 * to write, the length of buffer is stored in
305 * \param[in] idx index of the llog record. If \a idx == -1 then this is
306 * append case, otherwise \a idx is the index of record
308 * \param[in] th current transaction handle
310 * \retval 0 on successful declaration
311 * \retval negative error if declaration failed
313 static int llog_osd_declare_write_rec(const struct lu_env *env,
314 struct llog_handle *loghandle,
315 struct llog_rec_hdr *rec,
316 int idx, struct thandle *th)
318 struct llog_thread_info *lgi = llog_info(env);
329 LASSERT(rec->lrh_len <= loghandle->lgh_ctxt->loc_chunk_size);
331 o = loghandle->lgh_obj;
334 chunk_size = loghandle->lgh_ctxt->loc_chunk_size;
335 lgi->lgi_buf.lb_len = chunk_size;
336 lgi->lgi_buf.lb_buf = NULL;
337 /* each time we update header */
338 rc = dt_declare_record_write(env, o, &lgi->lgi_buf, 0,
340 if (rc || idx == 0) /* if error or just header */
344 * the pad record can be inserted so take into account double
347 lgi->lgi_buf.lb_len = chunk_size * 2;
348 lgi->lgi_buf.lb_buf = NULL;
349 /* XXX: implement declared window or multi-chunks approach */
350 rc = dt_declare_record_write(env, o, &lgi->lgi_buf, -1, th);
356 * Implementation of the llog_operations::lop_write
358 * This function writes the new record in the llog or modify the existed one.
360 * \param[in] env execution environment
361 * \param[in] loghandle llog handle of the current llog
362 * \param[in] rec llog record header. This is a real header of
363 * the full llog record to write. This is
364 * the beginning of buffer to write, the length
365 * of buffer is stored in \a rec::lrh_len
366 * \param[out] reccookie pointer to the cookie to return back if needed.
367 * It is used for further cancel of this llog
369 * \param[in] idx index of the llog record. If \a idx == -1 then
370 * this is append case, otherwise \a idx is
371 * the index of record to modify
372 * \param[in] th current transaction handle
374 * \retval 0 on successful write && \a reccookie == NULL
375 * 1 on successful write && \a reccookie != NULL
376 * \retval negative error if write failed
378 static int llog_osd_write_rec(const struct lu_env *env,
379 struct llog_handle *loghandle,
380 struct llog_rec_hdr *rec,
381 struct llog_cookie *reccookie,
382 int idx, struct thandle *th)
384 struct llog_thread_info *lgi = llog_info(env);
385 struct llog_log_hdr *llh;
386 int reclen = rec->lrh_len;
388 struct llog_rec_tail *lrt;
393 __u64 orig_write_offset;
396 llh = loghandle->lgh_hdr;
397 o = loghandle->lgh_obj;
399 chunk_size = llh->llh_hdr.lrh_len;
400 CDEBUG(D_OTHER, "new record %x to "DFID"\n",
401 rec->lrh_type, PFID(lu_object_fid(&o->do_lu)));
403 if (!llog_osd_exist(loghandle))
406 /* record length should not bigger than */
407 if (reclen > loghandle->lgh_hdr->llh_hdr.lrh_len)
410 /* sanity check for fixed-records llog */
411 if (idx != LLOG_HEADER_IDX && (llh->llh_flags & LLOG_F_IS_FIXSIZE)) {
412 LASSERT(llh->llh_size != 0);
413 LASSERT(llh->llh_size == reclen);
416 rc = dt_attr_get(env, o, &lgi->lgi_attr);
421 * The modification case.
422 * If idx set then the record with that index must be modified.
423 * There are three cases possible:
424 * 1) the common case is the llog header update (idx == 0)
425 * 2) the llog record modification during llog process.
426 * This is indicated by the \a loghandle::lgh_cur_idx > 0.
427 * In that case the \a loghandle::lgh_cur_offset
428 * 3) otherwise this is assumed that llog consist of records of
429 * fixed size, i.e. catalog. The llog header must has llh_size
430 * field equal to record size. The record offset is calculated
431 * just by /a idx value
433 * During modification we don't need extra header update because
434 * the bitmap and record count are not changed. The record header
435 * and tail remains the same too.
437 if (idx != LLOG_NEXT_IDX) {
438 /* llog can be empty only when first record is being written */
439 LASSERT(ergo(idx > 0, lgi->lgi_attr.la_size > 0));
441 if (!ext2_test_bit(idx, LLOG_HDR_BITMAP(llh))) {
442 CERROR("%s: modify unset record %u\n",
443 o->do_lu.lo_dev->ld_obd->obd_name, idx);
447 if (idx != rec->lrh_index) {
448 CERROR("%s: modify index mismatch %d %u\n",
449 o->do_lu.lo_dev->ld_obd->obd_name, idx,
454 if (idx == LLOG_HEADER_IDX) {
455 /* llog header update */
456 __u32 *bitmap = LLOG_HDR_BITMAP(llh);
460 /* If it does not indicate the bitmap index
461 * (reccookie == NULL), then it means update
462 * the whole update header. Otherwise only
463 * update header and bits needs to be updated,
464 * and in DNE cases, it will signaficantly
465 * shrink the RPC size.
466 * see distribute_txn_cancel_records()*/
467 if (reccookie == NULL) {
468 lgi->lgi_buf.lb_len = reclen;
469 lgi->lgi_buf.lb_buf = rec;
470 rc = dt_record_write(env, o, &lgi->lgi_buf,
475 /* update the header */
476 lgi->lgi_buf.lb_len = llh->llh_bitmap_offset;
477 lgi->lgi_buf.lb_buf = llh;
478 rc = dt_record_write(env, o, &lgi->lgi_buf,
483 /* update the bitmap */
484 index = reccookie->lgc_index;
485 lgi->lgi_off = llh->llh_bitmap_offset +
486 (index / (sizeof(*bitmap) * 8)) *
488 lgi->lgi_buf.lb_len = sizeof(*bitmap);
489 lgi->lgi_buf.lb_buf =
490 &bitmap[index/(sizeof(*bitmap)*8)];
491 rc = dt_record_write(env, o, &lgi->lgi_buf,
495 } else if (loghandle->lgh_cur_idx > 0) {
497 * The lgh_cur_offset can be used only if index is
500 if (idx != loghandle->lgh_cur_idx) {
501 CERROR("%s: modify index mismatch %d %d\n",
502 o->do_lu.lo_dev->ld_obd->obd_name, idx,
503 loghandle->lgh_cur_idx);
507 lgi->lgi_off = loghandle->lgh_cur_offset;
508 CDEBUG(D_OTHER, "modify record "DOSTID": idx:%d, "
509 "len:%u offset %llu\n",
510 POSTID(&loghandle->lgh_id.lgl_oi), idx,
511 rec->lrh_len, (long long)lgi->lgi_off);
512 } else if (llh->llh_flags & LLOG_F_IS_FIXSIZE) {
513 lgi->lgi_off = llh->llh_hdr.lrh_len +
516 /* This can be result of lgh_cur_idx is not set during
517 * llog processing or llh_size is not set to proper
518 * record size for fixed records llog. Therefore it is
519 * impossible to get record offset. */
520 CERROR("%s: can't get record offset, idx:%d, "
521 "len:%u.\n", o->do_lu.lo_dev->ld_obd->obd_name,
526 /* update only data, header and tail remain the same */
527 lgi->lgi_off += sizeof(struct llog_rec_hdr);
528 lgi->lgi_buf.lb_len = REC_DATA_LEN(rec);
529 lgi->lgi_buf.lb_buf = REC_DATA(rec);
530 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
531 if (rc == 0 && reccookie) {
532 reccookie->lgc_lgl = loghandle->lgh_id;
533 reccookie->lgc_index = idx;
541 * The most common case of using llog. The new index is assigned to
542 * the new record, new bit is set in llog bitmap and llog count is
545 * Make sure that records don't cross a chunk boundary, so we can
546 * process them page-at-a-time if needed. If it will cross a chunk
547 * boundary, write in a fake (but referenced) entry to pad the chunk.
551 /* simulate ENOSPC when new plain llog is being added to the
553 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_LLOG_CREATE_FAILED2) &&
554 llh->llh_flags & LLOG_F_IS_CAT)
557 LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
558 orig_last_idx = loghandle->lgh_last_idx;
559 orig_write_offset = loghandle->lgh_write_offset;
560 lgi->lgi_off = lgi->lgi_attr.la_size;
562 if (loghandle->lgh_max_size > 0 &&
563 lgi->lgi_off >= loghandle->lgh_max_size) {
564 CDEBUG(D_OTHER, "llog is getting too large (%u > %u) at %u "
565 DOSTID"\n", (unsigned)lgi->lgi_off,
566 loghandle->lgh_max_size,
567 (int)loghandle->lgh_last_idx,
568 POSTID(&loghandle->lgh_id.lgl_oi));
569 /* this is to signal that this llog is full */
570 loghandle->lgh_last_idx = LLOG_HDR_BITMAP_SIZE(llh) - 1;
574 left = chunk_size - (lgi->lgi_off & (chunk_size - 1));
575 /* NOTE: padding is a record, but no bit is set */
576 if (left != 0 && left != reclen &&
577 left < (reclen + LLOG_MIN_REC_SIZE)) {
578 index = loghandle->lgh_last_idx + 1;
579 rc = llog_osd_pad(env, o, &lgi->lgi_off, left, index, th);
583 if (dt_object_remote(o))
584 loghandle->lgh_write_offset = lgi->lgi_off;
586 loghandle->lgh_last_idx++; /* for pad rec */
588 /* if it's the last idx in log file, then return -ENOSPC
589 * or wrap around if a catalog */
590 if (llog_is_full(loghandle) ||
591 unlikely(llh->llh_flags & LLOG_F_IS_CAT &&
592 OBD_FAIL_PRECHECK(OBD_FAIL_CAT_RECORDS) &&
593 loghandle->lgh_last_idx >= cfs_fail_val)) {
594 if (llh->llh_flags & LLOG_F_IS_CAT)
595 loghandle->lgh_last_idx = 0;
600 /* increment the last_idx along with llh_tail index, they should
601 * be equal for a llog lifetime */
602 loghandle->lgh_last_idx++;
603 index = loghandle->lgh_last_idx;
604 LLOG_HDR_TAIL(llh)->lrt_index = index;
606 * NB: the caller should make sure only 1 process access
607 * the lgh_last_idx, e.g. append should be exclusive.
608 * Otherwise it might hit the assert.
610 LASSERT(index < LLOG_HDR_BITMAP_SIZE(llh));
611 rec->lrh_index = index;
613 lrt->lrt_len = rec->lrh_len;
614 lrt->lrt_index = rec->lrh_index;
616 /* the lgh_hdr_mutex protects llog header data from concurrent
617 * update/cancel, the llh_count and llh_bitmap are protected */
618 mutex_lock(&loghandle->lgh_hdr_mutex);
619 if (ext2_set_bit(index, LLOG_HDR_BITMAP(llh))) {
620 CERROR("%s: index %u already set in log bitmap\n",
621 o->do_lu.lo_dev->ld_obd->obd_name, index);
622 mutex_unlock(&loghandle->lgh_hdr_mutex);
623 LBUG(); /* should never happen */
627 if (!(llh->llh_flags & LLOG_F_IS_FIXSIZE)) {
628 /* Update the minimum size of the llog record */
629 if (llh->llh_size == 0)
630 llh->llh_size = reclen;
631 else if (reclen < llh->llh_size)
632 llh->llh_size = reclen;
635 if (lgi->lgi_attr.la_size == 0) {
637 lgi->lgi_buf.lb_len = llh->llh_hdr.lrh_len;
638 lgi->lgi_buf.lb_buf = &llh->llh_hdr;
639 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
641 GOTO(out_unlock, rc);
643 __u32 *bitmap = LLOG_HDR_BITMAP(llh);
645 /* Note: If this is not initialization (size == 0), then do not
646 * write the whole header (8k bytes), only update header/tail
647 * and bits needs to be updated. Because this update might be
648 * part of cross-MDT operation, which needs to write these
649 * updates into the update log(32KB limit) and also pack inside
650 * the RPC (1MB limit), if we write 8K for each operation, which
651 * will cost a lot space, and keep us adding more updates to one
654 lgi->lgi_buf.lb_len = llh->llh_bitmap_offset;
655 lgi->lgi_buf.lb_buf = &llh->llh_hdr;
656 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
658 GOTO(out_unlock, rc);
660 lgi->lgi_off = llh->llh_bitmap_offset +
661 (index / (sizeof(*bitmap) * 8)) * sizeof(*bitmap);
662 lgi->lgi_buf.lb_len = sizeof(*bitmap);
663 lgi->lgi_buf.lb_buf = &bitmap[index/(sizeof(*bitmap)*8)];
664 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
666 GOTO(out_unlock, rc);
668 lgi->lgi_off = (unsigned long)LLOG_HDR_TAIL(llh) -
670 lgi->lgi_buf.lb_len = sizeof(llh->llh_tail);
671 lgi->lgi_buf.lb_buf = LLOG_HDR_TAIL(llh);
672 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
674 GOTO(out_unlock, rc);
678 /* unlock here for remote object */
679 mutex_unlock(&loghandle->lgh_hdr_mutex);
683 /* computed index can be used to determine offset for fixed-size
684 * records. This also allows to handle Catalog wrap around case */
685 if (llh->llh_flags & LLOG_F_IS_FIXSIZE) {
686 lgi->lgi_off = llh->llh_hdr.lrh_len + (index - 1) * reclen;
687 } else if (dt_object_remote(o)) {
688 lgi->lgi_off = max_t(__u64, loghandle->lgh_write_offset,
691 rc = dt_attr_get(env, o, &lgi->lgi_attr);
695 LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
696 lgi->lgi_off = max_t(__u64, lgi->lgi_attr.la_size,
700 lgi->lgi_buf.lb_len = reclen;
701 lgi->lgi_buf.lb_buf = rec;
702 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
706 if (dt_object_remote(o))
707 loghandle->lgh_write_offset = lgi->lgi_off;
709 CDEBUG(D_HA, "added record "DFID": idx: %u, %u off%llu\n",
710 PFID(lu_object_fid(&o->do_lu)), index, rec->lrh_len,
712 if (reccookie != NULL) {
713 reccookie->lgc_lgl = loghandle->lgh_id;
714 reccookie->lgc_index = index;
715 if ((rec->lrh_type == MDS_UNLINK_REC) ||
716 (rec->lrh_type == MDS_SETATTR64_REC))
717 reccookie->lgc_subsys = LLOG_MDS_OST_ORIG_CTXT;
718 else if (rec->lrh_type == OST_SZ_REC)
719 reccookie->lgc_subsys = LLOG_SIZE_ORIG_CTXT;
721 reccookie->lgc_subsys = -1;
726 /* cleanup llog for error case */
727 mutex_lock(&loghandle->lgh_hdr_mutex);
728 ext2_clear_bit(index, LLOG_HDR_BITMAP(llh));
730 mutex_unlock(&loghandle->lgh_hdr_mutex);
732 /* restore llog last_idx */
733 if (dt_object_remote(o)) {
734 loghandle->lgh_last_idx = orig_last_idx;
735 loghandle->lgh_write_offset = orig_write_offset;
736 } else if (--loghandle->lgh_last_idx == 0 &&
737 (llh->llh_flags & LLOG_F_IS_CAT) && llh->llh_cat_idx != 0) {
738 /* catalog had just wrap-around case */
739 loghandle->lgh_last_idx = LLOG_HDR_BITMAP_SIZE(llh) - 1;
742 LLOG_HDR_TAIL(llh)->lrt_index = loghandle->lgh_last_idx;
748 * We can skip reading at least as many log blocks as the number of
749 * minimum sized log records we are skipping. If it turns out
750 * that we are not far enough along the log (because the
751 * actual records are larger than minimum size) we just skip
754 * Note: in llog_process_thread, it will use bitmap offset as
755 * the index to locate the record, which also includs some pad
756 * records, whose record size is very small, and it also does not
757 * consider pad record when recording minimum record size (otherwise
758 * min_record size might be too small), so in some rare cases,
759 * it might skip too much record for @goal, see llog_osd_next_block().
761 * When force_mini_rec is true, it means we have to use LLOG_MIN_REC_SIZE
762 * as the min record size to skip over, usually because in the previous
763 * try, it skip too much record, see loog_osd_next(prev)_block().
765 static inline void llog_skip_over(struct llog_handle *lgh, __u64 *off,
766 int curr, int goal, __u32 chunk_size,
769 struct llog_log_hdr *llh = lgh->lgh_hdr;
771 /* Goal should not bigger than the record count */
772 if (goal > lgh->lgh_last_idx)
773 goal = lgh->lgh_last_idx;
776 if (llh->llh_flags & LLOG_F_IS_FIXSIZE) {
777 *off = chunk_size + (goal - 1) * llh->llh_size;
779 __u64 min_rec_size = LLOG_MIN_REC_SIZE;
781 if (llh->llh_size > 0 && !force_mini_rec)
782 min_rec_size = llh->llh_size;
784 *off = *off + (goal - curr - 1) * min_rec_size;
787 /* always align with lower chunk boundary*/
788 *off &= ~(chunk_size - 1);
792 * Remove optional fields that the client doesn't expect.
793 * This is typically in order to ensure compatibility with older clients.
794 * It is assumed that since we exclusively remove fields, the block will be
795 * big enough to handle the remapped records. It is also assumed that records
796 * of a block have the same format (i.e.: the same features enabled).
798 * \param[in,out] hdr Header of the block of records to remap.
799 * \param[in,out] last_hdr Last header, don't read past this point.
800 * \param[in] flags Flags describing the fields to keep.
802 static void changelog_block_trim_ext(struct llog_rec_hdr *hdr,
803 struct llog_rec_hdr *last_hdr,
804 enum changelog_rec_flags flags)
806 if (hdr->lrh_type != CHANGELOG_REC)
810 struct changelog_rec *rec = (struct changelog_rec *)(hdr + 1);
812 changelog_remap_rec(rec, rec->cr_flags & flags);
813 hdr = llog_rec_hdr_next(hdr);
814 } while ((char *)hdr <= (char *)last_hdr);
818 * Implementation of the llog_operations::lop_next_block
820 * This function finds the the next llog block to return which contains
821 * record with required index. It is main part of llog processing.
823 * \param[in] env execution environment
824 * \param[in] loghandle llog handle of the current llog
825 * \param[in,out] cur_idx index preceeding cur_offset
826 * \param[in] next_idx target index to find
827 * \param[in,out] cur_offset furtherst point read in the file
828 * \param[in] buf pointer to data buffer to fill
829 * \param[in] len required len to read, it is
830 * usually llog chunk_size.
832 * \retval 0 on successful buffer read
833 * \retval negative value on error
835 static int llog_osd_next_block(const struct lu_env *env,
836 struct llog_handle *loghandle, int *cur_idx,
837 int next_idx, __u64 *cur_offset, void *buf,
840 struct llog_thread_info *lgi = llog_info(env);
842 struct dt_device *dt;
845 int last_idx = *cur_idx;
846 __u64 last_offset = *cur_offset;
847 bool force_mini_rec = false;
854 chunk_size = loghandle->lgh_hdr->llh_hdr.lrh_len;
855 if (len == 0 || len & (chunk_size - 1))
859 LASSERT(loghandle->lgh_ctxt);
861 o = loghandle->lgh_obj;
863 LASSERT(dt_object_exists(o));
864 dt = lu2dt_dev(o->do_lu.lo_dev);
867 rc = dt_attr_get(env, o, &lgi->lgi_attr);
871 CDEBUG(D_OTHER, "looking for log index %u (cur idx %u off"
872 "%llu), size %llu\n", next_idx, *cur_idx,
873 *cur_offset, lgi->lgi_attr.la_size);
875 while (*cur_offset < lgi->lgi_attr.la_size) {
876 struct llog_rec_hdr *rec, *last_rec;
877 struct llog_rec_tail *tail;
879 llog_skip_over(loghandle, cur_offset, *cur_idx,
880 next_idx, chunk_size, force_mini_rec);
882 /* read up to next llog chunk_size block */
883 lgi->lgi_buf.lb_len = chunk_size -
884 (*cur_offset & (chunk_size - 1));
885 lgi->lgi_buf.lb_buf = buf;
887 rc = dt_read(env, o, &lgi->lgi_buf, cur_offset);
889 if (rc == -EBADR && !force_mini_rec)
892 CERROR("%s: can't read llog block from log "DFID
893 " offset %llu: rc = %d\n",
894 o->do_lu.lo_dev->ld_obd->obd_name,
895 PFID(lu_object_fid(&o->do_lu)), *cur_offset,
901 /* signal the end of the valid buffer to
903 memset(buf + rc, 0, len - rc);
906 if (rc == 0) { /* end of file, nothing to do */
912 if (rc < sizeof(*tail)) {
916 CERROR("%s: invalid llog block at log id "DOSTID"/%u "
918 o->do_lu.lo_dev->ld_obd->obd_name,
919 POSTID(&loghandle->lgh_id.lgl_oi),
920 loghandle->lgh_id.lgl_ogen, *cur_offset);
921 GOTO(out, rc = -EINVAL);
925 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
926 lustre_swab_llog_rec(rec);
928 tail = (struct llog_rec_tail *)((char *)buf + rc -
929 sizeof(struct llog_rec_tail));
930 /* get the last record in block */
931 last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
934 if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
935 lustre_swab_llog_rec(last_rec);
937 if (last_rec->lrh_index != tail->lrt_index) {
938 CERROR("%s: invalid llog tail at log id "DOSTID"/%u "
939 "offset %llu last_rec idx %u tail idx %u\n",
940 o->do_lu.lo_dev->ld_obd->obd_name,
941 POSTID(&loghandle->lgh_id.lgl_oi),
942 loghandle->lgh_id.lgl_ogen, *cur_offset,
943 last_rec->lrh_index, tail->lrt_index);
944 GOTO(out, rc = -EINVAL);
947 *cur_idx = tail->lrt_index;
949 /* this shouldn't happen */
950 if (tail->lrt_index == 0) {
951 CERROR("%s: invalid llog tail at log id "DOSTID"/%u "
952 "offset %llu bytes %d\n",
953 o->do_lu.lo_dev->ld_obd->obd_name,
954 POSTID(&loghandle->lgh_id.lgl_oi),
955 loghandle->lgh_id.lgl_ogen, *cur_offset, rc);
956 GOTO(out, rc = -EINVAL);
958 if (tail->lrt_index < next_idx) {
960 last_offset = *cur_offset;
964 /* sanity check that the start of the new buffer is no farther
965 * than the record that we wanted. This shouldn't happen. */
966 if (rec->lrh_index > next_idx) {
967 if (!force_mini_rec && next_idx > last_idx)
970 CERROR("%s: missed desired record? %u > %u\n",
971 o->do_lu.lo_dev->ld_obd->obd_name,
972 rec->lrh_index, next_idx);
973 GOTO(out, rc = -ENOENT);
976 /* Trim unsupported extensions for compat w/ older clients */
977 if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_JOBID))
978 changelog_block_trim_ext(rec, last_rec,
979 CLF_VERSION | CLF_RENAME);
984 /* Note: because there are some pad records in the
985 * llog, so llog_skip_over() might skip too much
986 * records, let's try skip again with minimum record */
987 force_mini_rec = true;
988 *cur_offset = last_offset;
991 GOTO(out, rc = -EIO);
997 * Implementation of the llog_operations::lop_prev_block
999 * This function finds the llog block to return which contains
1000 * record with required index but in reverse order - from end of llog
1002 * It is main part of reverse llog processing.
1004 * \param[in] env execution environment
1005 * \param[in] loghandle llog handle of the current llog
1006 * \param[in] prev_idx target index to find
1007 * \param[in] buf pointer to data buffer to fill
1008 * \param[in] len required len to read, it is llog_chunk_size usually.
1010 * \retval 0 on successful buffer read
1011 * \retval negative value on error
1013 static int llog_osd_prev_block(const struct lu_env *env,
1014 struct llog_handle *loghandle,
1015 int prev_idx, void *buf, int len)
1017 struct llog_thread_info *lgi = llog_info(env);
1018 struct dt_object *o;
1019 struct dt_device *dt;
1026 chunk_size = loghandle->lgh_hdr->llh_hdr.lrh_len;
1027 if (len == 0 || len & (chunk_size - 1))
1030 CDEBUG(D_OTHER, "looking for log index %u\n", prev_idx);
1033 LASSERT(loghandle->lgh_ctxt);
1035 o = loghandle->lgh_obj;
1037 LASSERT(dt_object_exists(o));
1038 dt = lu2dt_dev(o->do_lu.lo_dev);
1041 /* Let's only use mini record size for previous block read
1043 cur_offset = chunk_size;
1044 llog_skip_over(loghandle, &cur_offset, 0, prev_idx,
1047 rc = dt_attr_get(env, o, &lgi->lgi_attr);
1051 while (cur_offset < lgi->lgi_attr.la_size) {
1052 struct llog_rec_hdr *rec, *last_rec;
1053 struct llog_rec_tail *tail;
1055 lgi->lgi_buf.lb_len = len;
1056 lgi->lgi_buf.lb_buf = buf;
1057 rc = dt_read(env, o, &lgi->lgi_buf, &cur_offset);
1059 CERROR("%s: can't read llog block from log "DFID
1060 " offset %llu: rc = %d\n",
1061 o->do_lu.lo_dev->ld_obd->obd_name,
1062 PFID(lu_object_fid(&o->do_lu)), cur_offset, rc);
1066 if (rc == 0) /* end of file, nothing to do */
1069 if (rc < sizeof(*tail)) {
1070 CERROR("%s: invalid llog block at log id "DOSTID"/%u "
1072 o->do_lu.lo_dev->ld_obd->obd_name,
1073 POSTID(&loghandle->lgh_id.lgl_oi),
1074 loghandle->lgh_id.lgl_ogen, cur_offset);
1075 GOTO(out, rc = -EINVAL);
1079 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
1080 lustre_swab_llog_rec(rec);
1082 tail = (struct llog_rec_tail *)((char *)buf + rc -
1083 sizeof(struct llog_rec_tail));
1084 /* get the last record in block */
1085 last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
1086 le32_to_cpu(tail->lrt_len));
1088 if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
1089 lustre_swab_llog_rec(last_rec);
1090 LASSERT(last_rec->lrh_index == tail->lrt_index);
1092 /* this shouldn't happen */
1093 if (tail->lrt_index == 0) {
1094 CERROR("%s: invalid llog tail at log id "DOSTID"/%u "
1096 o->do_lu.lo_dev->ld_obd->obd_name,
1097 POSTID(&loghandle->lgh_id.lgl_oi),
1098 loghandle->lgh_id.lgl_ogen, cur_offset);
1099 GOTO(out, rc = -EINVAL);
1101 if (tail->lrt_index < prev_idx)
1104 /* sanity check that the start of the new buffer is no farther
1105 * than the record that we wanted. This shouldn't happen. */
1106 if (rec->lrh_index > prev_idx) {
1107 CERROR("%s: missed desired record? %u > %u\n",
1108 o->do_lu.lo_dev->ld_obd->obd_name,
1109 rec->lrh_index, prev_idx);
1110 GOTO(out, rc = -ENOENT);
1113 /* Trim unsupported extensions for compat w/ older clients */
1114 if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_JOBID))
1115 changelog_block_trim_ext(rec, last_rec,
1116 CLF_VERSION | CLF_RENAME);
1120 GOTO(out, rc = -EIO);
1126 * This is helper function to get llog directory object. It is used by named
1127 * llog operations to find/insert/delete llog entry from llog directory.
1129 * \param[in] env execution environment
1130 * \param[in] ctxt llog context
1132 * \retval dt_object of llog directory
1133 * \retval ERR_PTR of negative value on error
1135 static struct dt_object *llog_osd_dir_get(const struct lu_env *env,
1136 struct llog_ctxt *ctxt)
1138 struct dt_device *dt;
1139 struct dt_thread_info *dti = dt_info(env);
1140 struct dt_object *dir;
1143 dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
1144 if (ctxt->loc_dir == NULL) {
1145 rc = dt_root_get(env, dt, &dti->dti_fid);
1148 dir = dt_locate(env, dt, &dti->dti_fid);
1150 if (!IS_ERR(dir) && !dt_try_as_dir(env, dir)) {
1151 lu_object_put(env, &dir->do_lu);
1152 return ERR_PTR(-ENOTDIR);
1155 lu_object_get(&ctxt->loc_dir->do_lu);
1156 dir = ctxt->loc_dir;
1163 * Implementation of the llog_operations::lop_open
1165 * This function opens the llog by its logid or by name, it may open also
1166 * non existent llog and assing then new id to it.
1167 * The llog_open/llog_close pair works similar to lu_object_find/put,
1168 * the object may not exist prior open. The result of open is just dt_object
1169 * in the llog header.
1171 * \param[in] env execution environment
1172 * \param[in] handle llog handle of the current llog
1173 * \param[in] logid logid of llog to open (nameless llog)
1174 * \param[in] name name of llog to open (named llog)
1175 * \param[in] open_param
1176 * LLOG_OPEN_NEW - new llog, may not exist
1177 * LLOG_OPEN_EXIST - old llog, must exist
1179 * \retval 0 on successful open, llog_handle::lgh_obj
1180 * contains the dt_object of the llog.
1181 * \retval negative value on error
1183 static int llog_osd_open(const struct lu_env *env, struct llog_handle *handle,
1184 struct llog_logid *logid, char *name,
1185 enum llog_open_param open_param)
1187 struct llog_thread_info *lgi = llog_info(env);
1188 struct llog_ctxt *ctxt = handle->lgh_ctxt;
1189 struct dt_object *o;
1190 struct dt_device *dt;
1191 struct ls_device *ls;
1192 struct local_oid_storage *los = NULL;
1194 bool new_id = false;
1200 LASSERT(ctxt->loc_exp);
1201 LASSERT(ctxt->loc_exp->exp_obd);
1202 dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
1204 if (ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1205 struct lu_object_conf conf = { 0 };
1206 if (logid != NULL) {
1207 logid_to_fid(logid, &lgi->lgi_fid);
1209 /* If logid == NULL, then it means the caller needs
1210 * to allocate new FID (llog_cat_declare_add_rec()). */
1211 rc = obd_fid_alloc(env, ctxt->loc_exp,
1212 &lgi->lgi_fid, NULL);
1216 conf.loc_flags = LOC_F_NEW;
1219 o = dt_locate_at(env, dt, &lgi->lgi_fid,
1220 dt->dd_lu_dev.ld_site->ls_top_dev, &conf);
1227 ls = ls_device_get(dt);
1229 RETURN(PTR_ERR(ls));
1231 mutex_lock(&ls->ls_los_mutex);
1232 los = dt_los_find(ls, name != NULL ? FID_SEQ_LLOG_NAME : FID_SEQ_LLOG);
1233 mutex_unlock(&ls->ls_los_mutex);
1235 ls_device_put(env, ls);
1239 if (logid != NULL) {
1240 logid_to_fid(logid, &lgi->lgi_fid);
1242 struct dt_object *llog_dir;
1244 llog_dir = llog_osd_dir_get(env, ctxt);
1245 if (IS_ERR(llog_dir))
1246 GOTO(out, rc = PTR_ERR(llog_dir));
1247 dt_read_lock(env, llog_dir, 0);
1248 rc = dt_lookup_dir(env, llog_dir, name, &lgi->lgi_fid);
1249 dt_read_unlock(env, llog_dir);
1250 lu_object_put(env, &llog_dir->do_lu);
1251 if (rc == -ENOENT && open_param == LLOG_OPEN_NEW) {
1252 /* generate fid for new llog */
1253 rc = local_object_fid_generate(env, los,
1259 OBD_ALLOC(handle->lgh_name, strlen(name) + 1);
1260 if (handle->lgh_name)
1261 strcpy(handle->lgh_name, name);
1263 GOTO(out, rc = -ENOMEM);
1265 LASSERTF(open_param & LLOG_OPEN_NEW, "%#x\n", open_param);
1266 /* generate fid for new llog */
1268 rc = local_object_fid_generate(env, los, &lgi->lgi_fid);
1274 o = ls_locate(env, ls, &lgi->lgi_fid, NULL);
1276 GOTO(out_name, rc = PTR_ERR(o));
1278 if (dt_object_exists(o) && new_id) {
1279 /* llog exists with just generated ID, e.g. some old llog file
1280 * still is in use or is orphan, drop a warn and skip it. */
1281 CDEBUG(D_INFO, "%s: llog exists with the same FID: "DFID
1283 o->do_lu.lo_dev->ld_obd->obd_name,
1284 PFID(lu_object_fid(&o->do_lu)));
1285 lu_object_put(env, &o->do_lu);
1286 /* just skip this llog ID, we shouldn't delete it because we
1287 * don't know exactly what is its purpose and state. */
1292 /* No new llog is expected but doesn't exist */
1293 if (open_param != LLOG_OPEN_NEW && !dt_object_exists(o))
1294 GOTO(out_put, rc = -ENOENT);
1296 fid_to_logid(&lgi->lgi_fid, &handle->lgh_id);
1297 handle->lgh_obj = o;
1298 handle->private_data = los;
1299 LASSERT(handle->lgh_ctxt);
1304 lu_object_put(env, &o->do_lu);
1306 if (handle->lgh_name != NULL)
1307 OBD_FREE(handle->lgh_name, strlen(name) + 1);
1315 * Get dir for regular fid log object
1317 * Get directory for regular fid log object, and these regular fid log
1318 * object will be inserted under this directory, to satisfy the FS
1319 * consistency check, e2fsck etc.
1321 * \param [in] env execution environment
1322 * \param [in] dto llog object
1324 * \retval pointer to the directory if it is found.
1325 * \retval ERR_PTR(negative errno) if it fails.
1327 struct dt_object *llog_osd_get_regular_fid_dir(const struct lu_env *env,
1328 struct dt_object *dto)
1330 struct llog_thread_info *lgi = llog_info(env);
1331 struct seq_server_site *ss = dto->do_lu.lo_dev->ld_site->ld_seq_site;
1332 struct lu_seq_range *range = &lgi->lgi_range;
1333 struct lu_fid *dir_fid = &lgi->lgi_fid;
1334 struct dt_object *dir;
1338 fld_range_set_any(range);
1339 LASSERT(ss != NULL);
1340 rc = ss->ss_server_fld->lsf_seq_lookup(env, ss->ss_server_fld,
1341 fid_seq(lu_object_fid(&dto->do_lu)), range);
1343 RETURN(ERR_PTR(rc));
1345 lu_update_log_dir_fid(dir_fid, range->lsr_index);
1346 dir = dt_locate(env, lu2dt_dev(dto->do_lu.lo_dev), dir_fid);
1350 if (!dt_try_as_dir(env, dir)) {
1351 lu_object_put(env, &dir->do_lu);
1352 RETURN(ERR_PTR(-ENOTDIR));
1359 * Add llog object with regular FID to name entry
1361 * Add llog object with regular FID to name space, and each llog
1362 * object on each MDT will be /update_log_dir/[seq:oid:ver],
1363 * so to satisfy the namespace consistency check, e2fsck etc.
1365 * \param [in] env execution environment
1366 * \param [in] dto llog object
1367 * \param [in] th thandle
1368 * \param [in] declare if it is declare or execution
1370 * \retval 0 if insertion succeeds.
1371 * \retval negative errno if insertion fails.
1374 llog_osd_regular_fid_add_name_entry(const struct lu_env *env,
1375 struct dt_object *dto,
1376 struct thandle *th, bool declare)
1378 struct llog_thread_info *lgi = llog_info(env);
1379 const struct lu_fid *fid = lu_object_fid(&dto->do_lu);
1380 struct dt_insert_rec *rec = &lgi->lgi_dt_rec;
1381 struct dt_object *dir;
1382 char *name = lgi->lgi_name;
1386 if (!fid_is_norm(fid))
1389 dir = llog_osd_get_regular_fid_dir(env, dto);
1391 RETURN(PTR_ERR(dir));
1394 rec->rec_type = S_IFREG;
1395 snprintf(name, sizeof(lgi->lgi_name), DFID, PFID(fid));
1396 dt_write_lock(env, dir, 0);
1398 rc = dt_declare_insert(env, dir, (struct dt_rec *)rec,
1399 (struct dt_key *)name, th);
1401 rc = dt_insert(env, dir, (struct dt_rec *)rec,
1402 (struct dt_key *)name, th, 1);
1404 dt_write_unlock(env, dir);
1406 lu_object_put(env, &dir->do_lu);
1412 * Implementation of the llog_operations::lop_declare_create
1414 * This function declares the llog create. It declares also name insert
1415 * into llog directory in case of named llog.
1417 * \param[in] env execution environment
1418 * \param[in] res llog handle of the current llog
1419 * \param[in] th current transaction handle
1421 * \retval 0 on successful create declaration
1422 * \retval negative value on error
1424 static int llog_osd_declare_create(const struct lu_env *env,
1425 struct llog_handle *res, struct thandle *th)
1427 struct llog_thread_info *lgi = llog_info(env);
1428 struct dt_insert_rec *rec = &lgi->lgi_dt_rec;
1429 struct local_oid_storage *los;
1430 struct dt_object *o;
1435 LASSERT(res->lgh_obj);
1438 /* object can be created by another thread */
1440 if (dt_object_exists(o))
1443 if (res->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1444 struct llog_thread_info *lgi = llog_info(env);
1446 lgi->lgi_attr.la_valid = LA_MODE | LA_SIZE;
1447 lgi->lgi_attr.la_size = 0;
1448 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1449 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1451 rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL,
1457 rc = llog_osd_regular_fid_add_name_entry(env, o, th, true);
1461 los = res->private_data;
1464 rc = llog_osd_declare_new_object(env, los, o, th);
1468 /* do not declare header initialization here as it's declared
1469 * in llog_osd_declare_write_rec() which is always called */
1471 if (res->lgh_name) {
1472 struct dt_object *llog_dir;
1474 llog_dir = llog_osd_dir_get(env, res->lgh_ctxt);
1475 if (IS_ERR(llog_dir))
1476 RETURN(PTR_ERR(llog_dir));
1477 logid_to_fid(&res->lgh_id, &lgi->lgi_fid);
1478 rec->rec_fid = &lgi->lgi_fid;
1479 rec->rec_type = S_IFREG;
1480 rc = dt_declare_insert(env, llog_dir,
1481 (struct dt_rec *)rec,
1482 (struct dt_key *)res->lgh_name, th);
1483 lu_object_put(env, &llog_dir->do_lu);
1485 CERROR("%s: can't declare named llog %s: rc = %d\n",
1486 o->do_lu.lo_dev->ld_obd->obd_name,
1493 * Implementation of the llog_operations::lop_create
1495 * This function creates the llog according with llog_handle::lgh_obj
1496 * and llog_handle::lgh_name.
1498 * \param[in] env execution environment
1499 * \param[in] res llog handle of the current llog
1500 * \param[in] th current transaction handle
1502 * \retval 0 on successful create
1503 * \retval negative value on error
1505 static int llog_osd_create(const struct lu_env *env, struct llog_handle *res,
1508 struct llog_thread_info *lgi = llog_info(env);
1509 struct dt_insert_rec *rec = &lgi->lgi_dt_rec;
1510 struct local_oid_storage *los;
1511 struct dt_object *o;
1520 /* llog can be already created */
1521 if (dt_object_exists(o))
1524 if (res->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1525 struct llog_thread_info *lgi = llog_info(env);
1527 lgi->lgi_attr.la_valid = LA_MODE | LA_SIZE | LA_TYPE;
1528 lgi->lgi_attr.la_size = 0;
1529 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1530 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1532 dt_write_lock(env, o, 0);
1533 rc = dt_create(env, o, &lgi->lgi_attr, NULL,
1535 dt_write_unlock(env, o);
1539 rc = llog_osd_regular_fid_add_name_entry(env, o, th, false);
1544 los = res->private_data;
1547 dt_write_lock(env, o, 0);
1548 if (!dt_object_exists(o))
1549 rc = llog_osd_create_new_object(env, los, o, th);
1553 dt_write_unlock(env, o);
1557 if (res->lgh_name) {
1558 struct dt_object *llog_dir;
1560 llog_dir = llog_osd_dir_get(env, res->lgh_ctxt);
1561 if (IS_ERR(llog_dir))
1562 RETURN(PTR_ERR(llog_dir));
1564 logid_to_fid(&res->lgh_id, &lgi->lgi_fid);
1565 rec->rec_fid = &lgi->lgi_fid;
1566 rec->rec_type = S_IFREG;
1567 dt_read_lock(env, llog_dir, 0);
1568 rc = dt_insert(env, llog_dir, (struct dt_rec *)rec,
1569 (struct dt_key *)res->lgh_name,
1571 dt_read_unlock(env, llog_dir);
1572 lu_object_put(env, &llog_dir->do_lu);
1574 CERROR("%s: can't create named llog %s: rc = %d\n",
1575 o->do_lu.lo_dev->ld_obd->obd_name,
1582 * Implementation of the llog_operations::lop_close
1584 * This function closes the llog. It just put llog object and referenced
1587 * \param[in] env execution environment
1588 * \param[in] handle llog handle of the current llog
1590 * \retval 0 on successful llog close
1591 * \retval negative value on error
1593 static int llog_osd_close(const struct lu_env *env, struct llog_handle *handle)
1595 struct local_oid_storage *los;
1600 LASSERT(handle->lgh_obj);
1602 if (handle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1603 /* Remove the object from the cache, otherwise it may
1604 * hold LOD being released during cleanup process */
1605 lu_object_put_nocache(env, &handle->lgh_obj->do_lu);
1606 LASSERT(handle->private_data == NULL);
1609 lu_object_put(env, &handle->lgh_obj->do_lu);
1611 los = handle->private_data;
1615 if (handle->lgh_name)
1616 OBD_FREE(handle->lgh_name, strlen(handle->lgh_name) + 1);
1622 * delete llog object name entry
1624 * Delete llog object (with regular FID) from name space (under
1627 * \param [in] env execution environment
1628 * \param [in] dto llog object
1629 * \param [in] th thandle
1630 * \param [in] declare if it is declare or execution
1632 * \retval 0 if deletion succeeds.
1633 * \retval negative errno if deletion fails.
1636 llog_osd_regular_fid_del_name_entry(const struct lu_env *env,
1637 struct dt_object *dto,
1638 struct thandle *th, bool declare)
1640 struct llog_thread_info *lgi = llog_info(env);
1641 const struct lu_fid *fid = lu_object_fid(&dto->do_lu);
1642 struct dt_object *dir;
1643 char *name = lgi->lgi_name;
1647 if (!fid_is_norm(fid))
1650 dir = llog_osd_get_regular_fid_dir(env, dto);
1652 RETURN(PTR_ERR(dir));
1654 snprintf(name, sizeof(lgi->lgi_name), DFID, PFID(fid));
1655 dt_write_lock(env, dir, 0);
1657 rc = dt_declare_delete(env, dir, (struct dt_key *)name,
1660 rc = dt_delete(env, dir, (struct dt_key *)name, th);
1662 dt_write_unlock(env, dir);
1664 lu_object_put(env, &dir->do_lu);
1669 * Implementation of the llog_operations::lop_declare_destroy
1671 * This function declare destroys the llog and deletes also entry in the
1672 * llog directory in case of named llog. Llog should be opened prior that.
1674 * \param[in] env execution environment
1675 * \param[in] loghandle llog handle of the current llog
1677 * \retval 0 on successful destroy
1678 * \retval negative value on error
1680 static int llog_osd_declare_destroy(const struct lu_env *env,
1681 struct llog_handle *loghandle,
1684 struct llog_ctxt *ctxt;
1685 struct dt_object *o, *llog_dir = NULL;
1690 ctxt = loghandle->lgh_ctxt;
1693 o = loghandle->lgh_obj;
1696 if (loghandle->lgh_name) {
1697 llog_dir = llog_osd_dir_get(env, ctxt);
1698 if (IS_ERR(llog_dir))
1699 RETURN(PTR_ERR(llog_dir));
1701 rc = dt_declare_delete(env, llog_dir,
1702 (struct dt_key *)loghandle->lgh_name,
1708 rc = dt_declare_ref_del(env, o, th);
1712 rc = dt_declare_destroy(env, o, th);
1716 if (loghandle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1717 rc = llog_osd_regular_fid_del_name_entry(env, o, th, true);
1723 if (!(IS_ERR_OR_NULL(llog_dir)))
1724 lu_object_put(env, &llog_dir->do_lu);
1731 * Implementation of the llog_operations::lop_destroy
1733 * This function destroys the llog and deletes also entry in the
1734 * llog directory in case of named llog. Llog should be opened prior that.
1735 * Destroy method is not part of external transaction and does everything
1738 * \param[in] env execution environment
1739 * \param[in] loghandle llog handle of the current llog
1741 * \retval 0 on successful destroy
1742 * \retval negative value on error
1744 static int llog_osd_destroy(const struct lu_env *env,
1745 struct llog_handle *loghandle, struct thandle *th)
1747 struct llog_ctxt *ctxt;
1748 struct dt_object *o, *llog_dir = NULL;
1753 ctxt = loghandle->lgh_ctxt;
1754 LASSERT(ctxt != NULL);
1756 o = loghandle->lgh_obj;
1759 dt_write_lock(env, o, 0);
1760 if (!dt_object_exists(o))
1761 GOTO(out_unlock, rc = 0);
1763 if (loghandle->lgh_name) {
1764 llog_dir = llog_osd_dir_get(env, ctxt);
1765 if (IS_ERR(llog_dir))
1766 GOTO(out_unlock, rc = PTR_ERR(llog_dir));
1768 dt_read_lock(env, llog_dir, 0);
1769 rc = dt_delete(env, llog_dir,
1770 (struct dt_key *)loghandle->lgh_name,
1772 dt_read_unlock(env, llog_dir);
1774 CERROR("%s: can't remove llog %s: rc = %d\n",
1775 o->do_lu.lo_dev->ld_obd->obd_name,
1776 loghandle->lgh_name, rc);
1777 GOTO(out_unlock, rc);
1781 dt_ref_del(env, o, th);
1782 rc = dt_destroy(env, o, th);
1784 GOTO(out_unlock, rc);
1786 if (loghandle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1787 rc = llog_osd_regular_fid_del_name_entry(env, o, th, false);
1789 GOTO(out_unlock, rc);
1793 dt_write_unlock(env, o);
1794 if (!(IS_ERR_OR_NULL(llog_dir)))
1795 lu_object_put(env, &llog_dir->do_lu);
1800 * Implementation of the llog_operations::lop_setup
1802 * This function setup the llog on local storage.
1804 * \param[in] env execution environment
1805 * \param[in] obd obd device the llog belongs to
1806 * \param[in] olg the llog group, it is always zero group now.
1807 * \param[in] ctxt_idx the llog index, it defines the purpose of this llog.
1808 * Every new llog type have to use own index.
1809 * \param[in] disk_obd the storage obd, where llog is stored.
1811 * \retval 0 on successful llog setup
1812 * \retval negative value on error
1814 static int llog_osd_setup(const struct lu_env *env, struct obd_device *obd,
1815 struct obd_llog_group *olg, int ctxt_idx,
1816 struct obd_device *disk_obd)
1818 struct llog_thread_info *lgi = llog_info(env);
1819 struct llog_ctxt *ctxt;
1824 LASSERT(olg->olg_ctxts[ctxt_idx]);
1826 ctxt = llog_ctxt_get(olg->olg_ctxts[ctxt_idx]);
1829 if (disk_obd == NULL)
1832 /* initialize data allowing to generate new fids,
1833 * literally we need a sequece */
1834 lgi->lgi_fid.f_seq = FID_SEQ_LLOG;
1835 lgi->lgi_fid.f_oid = 1;
1836 lgi->lgi_fid.f_ver = 0;
1837 rc = local_oid_storage_init(env, disk_obd->obd_lvfs_ctxt.dt,
1839 &ctxt->loc_los_nameless);
1843 lgi->lgi_fid.f_seq = FID_SEQ_LLOG_NAME;
1844 lgi->lgi_fid.f_oid = 1;
1845 lgi->lgi_fid.f_ver = 0;
1846 rc = local_oid_storage_init(env, disk_obd->obd_lvfs_ctxt.dt,
1848 &ctxt->loc_los_named);
1850 local_oid_storage_fini(env, ctxt->loc_los_nameless);
1851 ctxt->loc_los_nameless = NULL;
1857 llog_ctxt_put(ctxt);
1862 * Implementation of the llog_operations::lop_cleanup
1864 * This function cleanups the llog on local storage.
1866 * \param[in] env execution environment
1867 * \param[in] ctxt the llog context
1871 static int llog_osd_cleanup(const struct lu_env *env, struct llog_ctxt *ctxt)
1873 if (ctxt->loc_los_nameless != NULL) {
1874 local_oid_storage_fini(env, ctxt->loc_los_nameless);
1875 ctxt->loc_los_nameless = NULL;
1878 if (ctxt->loc_los_named != NULL) {
1879 local_oid_storage_fini(env, ctxt->loc_los_named);
1880 ctxt->loc_los_named = NULL;
1886 struct llog_operations llog_osd_ops = {
1887 .lop_next_block = llog_osd_next_block,
1888 .lop_prev_block = llog_osd_prev_block,
1889 .lop_read_header = llog_osd_read_header,
1890 .lop_declare_destroy = llog_osd_declare_destroy,
1891 .lop_destroy = llog_osd_destroy,
1892 .lop_setup = llog_osd_setup,
1893 .lop_cleanup = llog_osd_cleanup,
1894 .lop_open = llog_osd_open,
1895 .lop_exist = llog_osd_exist,
1896 .lop_declare_create = llog_osd_declare_create,
1897 .lop_create = llog_osd_create,
1898 .lop_declare_write_rec = llog_osd_declare_write_rec,
1899 .lop_write_rec = llog_osd_write_rec,
1900 .lop_close = llog_osd_close,
1902 EXPORT_SYMBOL(llog_osd_ops);
1904 struct llog_operations llog_common_cat_ops = {
1905 .lop_next_block = llog_osd_next_block,
1906 .lop_prev_block = llog_osd_prev_block,
1907 .lop_read_header = llog_osd_read_header,
1908 .lop_declare_destroy = llog_osd_declare_destroy,
1909 .lop_destroy = llog_osd_destroy,
1910 .lop_setup = llog_osd_setup,
1911 .lop_cleanup = llog_osd_cleanup,
1912 .lop_open = llog_osd_open,
1913 .lop_exist = llog_osd_exist,
1914 .lop_declare_create = llog_osd_declare_create,
1915 .lop_create = llog_osd_create,
1916 .lop_declare_write_rec = llog_osd_declare_write_rec,
1917 .lop_write_rec = llog_osd_write_rec,
1918 .lop_close = llog_osd_close,
1919 .lop_add = llog_cat_add_rec,
1920 .lop_declare_add = llog_cat_declare_add_rec,
1922 EXPORT_SYMBOL(llog_common_cat_ops);
1925 * Read the special file which contains the list of llog catalogs IDs
1927 * This function reads the CATALOGS file which contains the array of llog
1928 * catalogs IDs. The main purpose of this file is to store OSP llogs indexed
1929 * by OST/MDT number.
1931 * \param[in] env execution environment
1932 * \param[in] d corresponding storage device
1933 * \param[in] idx position to start from, usually OST/MDT index
1934 * \param[in] count how many catalog IDs to read
1935 * \param[out] idarray the buffer for the data. If it is NULL then
1936 * function returns just number of catalog IDs
1938 * \param[in] fid LLOG_CATALOGS_OID for CATALOG object
1940 * \retval 0 on successful read of catalog IDs
1941 * \retval negative value on error
1942 * \retval positive value which is number of records in
1943 * the file if \a idarray is NULL
1945 int llog_osd_get_cat_list(const struct lu_env *env, struct dt_device *d,
1946 int idx, int count, struct llog_catid *idarray,
1947 const struct lu_fid *fid)
1949 struct llog_thread_info *lgi = llog_info(env);
1950 struct dt_object *o = NULL;
1958 size = sizeof(*idarray) * count;
1959 lgi->lgi_off = idx * sizeof(*idarray);
1961 lgi->lgi_fid = *fid;
1962 o = dt_locate(env, d, &lgi->lgi_fid);
1966 if (!dt_object_exists(o)) {
1967 th = dt_trans_create(env, d);
1969 GOTO(out, rc = PTR_ERR(th));
1971 lgi->lgi_attr.la_valid = LA_MODE;
1972 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1973 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1975 th->th_wait_submit = 1;
1976 /* Make the llog object creation synchronization, so
1977 * it will be reliable to the reference, especially
1978 * for remote reference */
1981 rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL,
1984 GOTO(out_trans, rc);
1986 rc = dt_trans_start_local(env, d, th);
1988 GOTO(out_trans, rc);
1990 dt_write_lock(env, o, 0);
1991 if (!dt_object_exists(o))
1992 rc = dt_create(env, o, &lgi->lgi_attr, NULL,
1994 dt_write_unlock(env, o);
1996 dt_trans_stop(env, d, th);
2001 rc = dt_attr_get(env, o, &lgi->lgi_attr);
2005 if (!S_ISREG(lgi->lgi_attr.la_mode)) {
2006 CERROR("%s: CATALOGS is not a regular file!: mode = %o\n",
2007 o->do_lu.lo_dev->ld_obd->obd_name,
2008 lgi->lgi_attr.la_mode);
2009 GOTO(out, rc = -ENOENT);
2012 CDEBUG(D_CONFIG, "cat list: disk size=%d, read=%d\n",
2013 (int)lgi->lgi_attr.la_size, size);
2015 /* return just number of llogs */
2016 if (idarray == NULL) {
2017 rc = lgi->lgi_attr.la_size / sizeof(*idarray);
2021 /* read for new ost index or for empty file */
2022 memset(idarray, 0, size);
2023 if (lgi->lgi_attr.la_size <= lgi->lgi_off)
2025 if (lgi->lgi_attr.la_size < lgi->lgi_off + size)
2026 size = lgi->lgi_attr.la_size - lgi->lgi_off;
2028 lgi->lgi_buf.lb_buf = idarray;
2029 lgi->lgi_buf.lb_len = size;
2030 rc = dt_record_read(env, o, &lgi->lgi_buf, &lgi->lgi_off);
2031 /* -EFAULT means the llog is a sparse file. This is not an error
2032 * after arbitrary OST index is supported. */
2033 if (rc < 0 && rc != -EFAULT) {
2034 CERROR("%s: error reading CATALOGS: rc = %d\n",
2035 o->do_lu.lo_dev->ld_obd->obd_name, rc);
2041 lu_object_put(env, &o->do_lu);
2044 EXPORT_SYMBOL(llog_osd_get_cat_list);
2047 * Write the special file which contains the list of llog catalogs IDs
2049 * This function writes the CATALOG file which contains the array of llog
2050 * catalogs IDs. It is used mostly to store OSP llogs indexed by OST/MDT
2053 * \param[in] env execution environment
2054 * \param[in] d corresponding storage device
2055 * \param[in] idx position to start from, usually OST/MDT index
2056 * \param[in] count how many catalog IDs to write
2057 * \param[out] idarray the buffer with the data to write.
2058 * \param[in] fid LLOG_CATALOGS_OID for CATALOG object
2060 * \retval 0 on successful write of catalog IDs
2061 * \retval negative value on error
2063 int llog_osd_put_cat_list(const struct lu_env *env, struct dt_device *d,
2064 int idx, int count, struct llog_catid *idarray,
2065 const struct lu_fid *fid)
2067 struct llog_thread_info *lgi = llog_info(env);
2068 struct dt_object *o = NULL;
2077 size = sizeof(*idarray) * count;
2078 lgi->lgi_off = idx * sizeof(*idarray);
2079 lgi->lgi_fid = *fid;
2081 o = dt_locate(env, d, &lgi->lgi_fid);
2085 if (!dt_object_exists(o))
2086 GOTO(out, rc = -ENOENT);
2088 rc = dt_attr_get(env, o, &lgi->lgi_attr);
2092 if (!S_ISREG(lgi->lgi_attr.la_mode)) {
2093 CERROR("%s: CATALOGS is not a regular file!: mode = %o\n",
2094 o->do_lu.lo_dev->ld_obd->obd_name,
2095 lgi->lgi_attr.la_mode);
2096 GOTO(out, rc = -ENOENT);
2099 th = dt_trans_create(env, d);
2101 GOTO(out, rc = PTR_ERR(th));
2103 lgi->lgi_buf.lb_len = size;
2104 lgi->lgi_buf.lb_buf = idarray;
2105 rc = dt_declare_record_write(env, o, &lgi->lgi_buf, lgi->lgi_off, th);
2107 GOTO(out_trans, rc);
2109 /* For update log, this happens during initialization,
2110 * see lod_sub_prep_llog(), and we need make sure catlog
2111 * file ID is written to catlist file(committed) before
2112 * cross-MDT operation write update records to catlog FILE,
2113 * otherwise, during failover these update records might
2115 if (fid_is_update_log(fid))
2118 rc = dt_trans_start_local(env, d, th);
2120 GOTO(out_trans, rc);
2122 th->th_wait_submit = 1;
2124 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
2126 CDEBUG(D_INODE, "can't write CATALOGS at index %d: rc = %d\n",
2129 dt_trans_stop(env, d, th);
2131 lu_object_put(env, &o->do_lu);
2134 EXPORT_SYMBOL(llog_osd_put_cat_list);