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, 2017, 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;
294 * Implementation of the llog_operations::lop_declare_write
296 * This function declares the new record write.
298 * \param[in] env execution environment
299 * \param[in] loghandle llog handle of the current llog
300 * \param[in] rec llog record header. This is a real header of the full
301 * llog record to write. This is the beginning of buffer
302 * to write, the length of buffer is stored in
304 * \param[in] idx index of the llog record. If \a idx == -1 then this is
305 * append case, otherwise \a idx is the index of record
307 * \param[in] th current transaction handle
309 * \retval 0 on successful declaration
310 * \retval negative error if declaration failed
312 static int llog_osd_declare_write_rec(const struct lu_env *env,
313 struct llog_handle *loghandle,
314 struct llog_rec_hdr *rec,
315 int idx, struct thandle *th)
317 struct llog_thread_info *lgi = llog_info(env);
328 LASSERT(rec->lrh_len <= loghandle->lgh_ctxt->loc_chunk_size);
330 o = loghandle->lgh_obj;
333 chunk_size = loghandle->lgh_ctxt->loc_chunk_size;
334 lgi->lgi_buf.lb_len = chunk_size;
335 lgi->lgi_buf.lb_buf = NULL;
336 /* each time we update header */
337 rc = dt_declare_record_write(env, o, &lgi->lgi_buf, 0,
339 if (rc || idx == 0) /* if error or just header */
343 * the pad record can be inserted so take into account double
346 lgi->lgi_buf.lb_len = chunk_size * 2;
347 lgi->lgi_buf.lb_buf = NULL;
348 /* XXX: implement declared window or multi-chunks approach */
349 rc = dt_declare_record_write(env, o, &lgi->lgi_buf, -1, th);
355 * Implementation of the llog_operations::lop_write
357 * This function writes the new record in the llog or modify the existed one.
359 * \param[in] env execution environment
360 * \param[in] loghandle llog handle of the current llog
361 * \param[in] rec llog record header. This is a real header of
362 * the full llog record to write. This is
363 * the beginning of buffer to write, the length
364 * of buffer is stored in \a rec::lrh_len
365 * \param[in,out] reccookie pointer to the cookie to return back if needed.
366 * It is used for further cancel of this llog
368 * \param[in] idx index of the llog record. If \a idx == -1 then
369 * this is append case, otherwise \a idx is
370 * the index of record to modify
371 * \param[in] th current transaction handle
373 * \retval 0 on successful write && \a reccookie == NULL
374 * 1 on successful write && \a reccookie != NULL
375 * \retval negative error if write failed
377 static int llog_osd_write_rec(const struct lu_env *env,
378 struct llog_handle *loghandle,
379 struct llog_rec_hdr *rec,
380 struct llog_cookie *reccookie,
381 int idx, struct thandle *th)
383 struct llog_thread_info *lgi = llog_info(env);
384 struct llog_log_hdr *llh;
385 int reclen = rec->lrh_len;
387 struct llog_rec_tail *lrt;
394 llh = loghandle->lgh_hdr;
395 o = loghandle->lgh_obj;
397 chunk_size = llh->llh_hdr.lrh_len;
398 CDEBUG(D_OTHER, "new record %x to "DFID"\n",
399 rec->lrh_type, PFID(lu_object_fid(&o->do_lu)));
401 if (!llog_osd_exist(loghandle))
404 /* record length should not bigger than */
405 if (reclen > loghandle->lgh_hdr->llh_hdr.lrh_len)
408 /* sanity check for fixed-records llog */
409 if (idx != LLOG_HEADER_IDX && (llh->llh_flags & LLOG_F_IS_FIXSIZE)) {
410 LASSERT(llh->llh_size != 0);
411 LASSERT(llh->llh_size == reclen);
414 rc = dt_attr_get(env, o, &lgi->lgi_attr);
419 * The modification case.
420 * If idx set then the record with that index must be modified.
421 * There are three cases possible:
422 * 1) the common case is the llog header update (idx == 0)
423 * 2) the llog record modification during llog process.
424 * This is indicated by the \a loghandle::lgh_cur_idx > 0.
425 * In that case the \a loghandle::lgh_cur_offset
426 * 3) otherwise this is assumed that llog consist of records of
427 * fixed size, i.e. catalog. The llog header must has llh_size
428 * field equal to record size. The record offset is calculated
429 * just by /a idx value
431 * During modification we don't need extra header update because
432 * the bitmap and record count are not changed. The record header
433 * and tail remains the same too.
435 if (idx != LLOG_NEXT_IDX) {
436 /* llog can be empty only when first record is being written */
437 LASSERT(ergo(idx > 0, lgi->lgi_attr.la_size > 0));
439 if (!ext2_test_bit(idx, LLOG_HDR_BITMAP(llh))) {
440 CERROR("%s: modify unset record %u\n",
441 o->do_lu.lo_dev->ld_obd->obd_name, idx);
445 if (idx != rec->lrh_index) {
446 CERROR("%s: modify index mismatch %d %u\n",
447 o->do_lu.lo_dev->ld_obd->obd_name, idx,
452 if (idx == LLOG_HEADER_IDX) {
453 /* llog header update */
454 __u32 *bitmap = LLOG_HDR_BITMAP(llh);
458 /* If it does not indicate the bitmap index
459 * (reccookie == NULL), then it means update
460 * the whole update header. Otherwise only
461 * update header and bits needs to be updated,
462 * and in DNE cases, it will signaficantly
463 * shrink the RPC size.
464 * see distribute_txn_cancel_records()*/
465 if (reccookie == NULL) {
466 lgi->lgi_buf.lb_len = reclen;
467 lgi->lgi_buf.lb_buf = rec;
468 rc = dt_record_write(env, o, &lgi->lgi_buf,
473 /* update the header */
474 lgi->lgi_buf.lb_len = llh->llh_bitmap_offset;
475 lgi->lgi_buf.lb_buf = llh;
476 rc = dt_record_write(env, o, &lgi->lgi_buf,
481 /* update the bitmap */
482 index = reccookie->lgc_index;
483 lgi->lgi_off = llh->llh_bitmap_offset +
484 (index / (sizeof(*bitmap) * 8)) *
486 lgi->lgi_buf.lb_len = sizeof(*bitmap);
487 lgi->lgi_buf.lb_buf =
488 &bitmap[index/(sizeof(*bitmap)*8)];
489 rc = dt_record_write(env, o, &lgi->lgi_buf,
493 } else if (llh->llh_flags & LLOG_F_IS_FIXSIZE) {
494 lgi->lgi_off = llh->llh_hdr.lrh_len +
496 } else if (reccookie != NULL && reccookie->lgc_index > 0) {
498 * The lgc_offset can be used only if index is
501 if (idx != reccookie->lgc_index) {
502 CERROR("%s: modify index mismatch %d %d\n",
503 o->do_lu.lo_dev->ld_obd->obd_name, idx,
504 reccookie->lgc_index);
508 lgi->lgi_off = reccookie->lgc_offset;
509 CDEBUG(D_OTHER, "modify record "DFID": idx:%u, "
510 "len:%u offset %llu\n",
511 PFID(&loghandle->lgh_id.lgl_oi.oi_fid), idx,
512 rec->lrh_len, (long long)lgi->lgi_off);
514 /* This can be result of lgh_cur_idx is not set during
515 * llog processing or llh_size is not set to proper
516 * record size for fixed records llog. Therefore it is
517 * impossible to get record offset. */
518 CERROR("%s: can't get record offset, idx:%d, "
519 "len:%u.\n", o->do_lu.lo_dev->ld_obd->obd_name,
524 /* update only data, header and tail remain the same */
525 lgi->lgi_off += sizeof(struct llog_rec_hdr);
526 lgi->lgi_buf.lb_len = REC_DATA_LEN(rec);
527 lgi->lgi_buf.lb_buf = REC_DATA(rec);
528 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
529 if (rc == 0 && reccookie) {
530 reccookie->lgc_lgl = loghandle->lgh_id;
531 reccookie->lgc_index = idx;
539 * The most common case of using llog. The new index is assigned to
540 * the new record, new bit is set in llog bitmap and llog count is
543 * Make sure that records don't cross a chunk boundary, so we can
544 * process them page-at-a-time if needed. If it will cross a chunk
545 * boundary, write in a fake (but referenced) entry to pad the chunk.
549 /* simulate ENOSPC when new plain llog is being added to the
551 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_LLOG_CREATE_FAILED2) &&
552 llh->llh_flags & LLOG_F_IS_CAT)
555 LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
556 orig_last_idx = loghandle->lgh_last_idx;
557 lgi->lgi_off = lgi->lgi_attr.la_size;
559 if (loghandle->lgh_max_size > 0 &&
560 lgi->lgi_off >= loghandle->lgh_max_size) {
561 CDEBUG(D_OTHER, "llog is getting too large (%u > %u) at %u "
562 DFID"\n", (unsigned)lgi->lgi_off,
563 loghandle->lgh_max_size, (int)loghandle->lgh_last_idx,
564 PFID(&loghandle->lgh_id.lgl_oi.oi_fid));
565 /* this is to signal that this llog is full */
566 loghandle->lgh_last_idx = LLOG_HDR_BITMAP_SIZE(llh) - 1;
570 left = chunk_size - (lgi->lgi_off & (chunk_size - 1));
571 /* NOTE: padding is a record, but no bit is set */
572 if (left != 0 && left != reclen &&
573 left < (reclen + LLOG_MIN_REC_SIZE)) {
574 index = loghandle->lgh_last_idx + 1;
575 rc = llog_osd_pad(env, o, &lgi->lgi_off, left, index, th);
579 loghandle->lgh_last_idx++; /* for pad rec */
581 /* if it's the last idx in log file, then return -ENOSPC
582 * or wrap around if a catalog */
583 if (llog_is_full(loghandle) ||
584 unlikely(llh->llh_flags & LLOG_F_IS_CAT &&
585 OBD_FAIL_PRECHECK(OBD_FAIL_CAT_RECORDS) &&
586 loghandle->lgh_last_idx >= cfs_fail_val)) {
587 if (llh->llh_flags & LLOG_F_IS_CAT)
588 loghandle->lgh_last_idx = 0;
593 /* increment the last_idx along with llh_tail index, they should
594 * be equal for a llog lifetime */
595 loghandle->lgh_last_idx++;
596 index = loghandle->lgh_last_idx;
597 LLOG_HDR_TAIL(llh)->lrt_index = index;
599 * NB: the caller should make sure only 1 process access
600 * the lgh_last_idx, e.g. append should be exclusive.
601 * Otherwise it might hit the assert.
603 LASSERT(index < LLOG_HDR_BITMAP_SIZE(llh));
604 rec->lrh_index = index;
606 lrt->lrt_len = rec->lrh_len;
607 lrt->lrt_index = rec->lrh_index;
609 /* the lgh_hdr_mutex protects llog header data from concurrent
610 * update/cancel, the llh_count and llh_bitmap are protected */
611 mutex_lock(&loghandle->lgh_hdr_mutex);
612 if (ext2_set_bit(index, LLOG_HDR_BITMAP(llh))) {
613 CERROR("%s: index %u already set in log bitmap\n",
614 o->do_lu.lo_dev->ld_obd->obd_name, index);
615 mutex_unlock(&loghandle->lgh_hdr_mutex);
616 LBUG(); /* should never happen */
620 if (!(llh->llh_flags & LLOG_F_IS_FIXSIZE)) {
621 /* Update the minimum size of the llog record */
622 if (llh->llh_size == 0)
623 llh->llh_size = reclen;
624 else if (reclen < llh->llh_size)
625 llh->llh_size = reclen;
628 if (lgi->lgi_attr.la_size == 0) {
630 lgi->lgi_buf.lb_len = llh->llh_hdr.lrh_len;
631 lgi->lgi_buf.lb_buf = &llh->llh_hdr;
632 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
634 GOTO(out_unlock, rc);
636 __u32 *bitmap = LLOG_HDR_BITMAP(llh);
638 /* Note: If this is not initialization (size == 0), then do not
639 * write the whole header (8k bytes), only update header/tail
640 * and bits needs to be updated. Because this update might be
641 * part of cross-MDT operation, which needs to write these
642 * updates into the update log(32KB limit) and also pack inside
643 * the RPC (1MB limit), if we write 8K for each operation, which
644 * will cost a lot space, and keep us adding more updates to one
647 lgi->lgi_buf.lb_len = llh->llh_bitmap_offset;
648 lgi->lgi_buf.lb_buf = &llh->llh_hdr;
649 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
651 GOTO(out_unlock, rc);
653 lgi->lgi_off = llh->llh_bitmap_offset +
654 (index / (sizeof(*bitmap) * 8)) * sizeof(*bitmap);
655 lgi->lgi_buf.lb_len = sizeof(*bitmap);
656 lgi->lgi_buf.lb_buf = &bitmap[index/(sizeof(*bitmap)*8)];
657 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
659 GOTO(out_unlock, rc);
661 lgi->lgi_off = (unsigned long)LLOG_HDR_TAIL(llh) -
663 lgi->lgi_buf.lb_len = sizeof(llh->llh_tail);
664 lgi->lgi_buf.lb_buf = LLOG_HDR_TAIL(llh);
665 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
667 GOTO(out_unlock, rc);
671 /* unlock here for remote object */
672 mutex_unlock(&loghandle->lgh_hdr_mutex);
676 /* computed index can be used to determine offset for fixed-size
677 * records. This also allows to handle Catalog wrap around case */
678 if (llh->llh_flags & LLOG_F_IS_FIXSIZE) {
679 lgi->lgi_off = llh->llh_hdr.lrh_len + (index - 1) * reclen;
681 rc = dt_attr_get(env, o, &lgi->lgi_attr);
685 LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
686 lgi->lgi_off = max_t(__u64, lgi->lgi_attr.la_size,
690 lgi->lgi_buf.lb_len = reclen;
691 lgi->lgi_buf.lb_buf = rec;
692 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
696 CDEBUG(D_HA, "added record "DFID".%u, %u off%llu\n",
697 PFID(lu_object_fid(&o->do_lu)), index, rec->lrh_len,
699 if (reccookie != NULL) {
700 reccookie->lgc_lgl = loghandle->lgh_id;
701 reccookie->lgc_index = index;
702 if ((rec->lrh_type == MDS_UNLINK_REC) ||
703 (rec->lrh_type == MDS_SETATTR64_REC))
704 reccookie->lgc_subsys = LLOG_MDS_OST_ORIG_CTXT;
705 else if (rec->lrh_type == OST_SZ_REC)
706 reccookie->lgc_subsys = LLOG_SIZE_ORIG_CTXT;
708 reccookie->lgc_subsys = -1;
713 /* cleanup llog for error case */
714 mutex_lock(&loghandle->lgh_hdr_mutex);
715 ext2_clear_bit(index, LLOG_HDR_BITMAP(llh));
717 mutex_unlock(&loghandle->lgh_hdr_mutex);
719 /* restore llog last_idx */
720 if (dt_object_remote(o)) {
721 loghandle->lgh_last_idx = orig_last_idx;
722 } else if (--loghandle->lgh_last_idx == 0 &&
723 (llh->llh_flags & LLOG_F_IS_CAT) && llh->llh_cat_idx != 0) {
724 /* catalog had just wrap-around case */
725 loghandle->lgh_last_idx = LLOG_HDR_BITMAP_SIZE(llh) - 1;
728 LLOG_HDR_TAIL(llh)->lrt_index = loghandle->lgh_last_idx;
734 * We can skip reading at least as many log blocks as the number of
735 * minimum sized log records we are skipping. If it turns out
736 * that we are not far enough along the log (because the
737 * actual records are larger than minimum size) we just skip
740 * Note: in llog_process_thread, it will use bitmap offset as
741 * the index to locate the record, which also includs some pad
742 * records, whose record size is very small, and it also does not
743 * consider pad record when recording minimum record size (otherwise
744 * min_record size might be too small), so in some rare cases,
745 * it might skip too much record for @goal, see llog_osd_next_block().
747 * When force_mini_rec is true, it means we have to use LLOG_MIN_REC_SIZE
748 * as the min record size to skip over, usually because in the previous
749 * try, it skip too much record, see loog_osd_next(prev)_block().
751 static inline void llog_skip_over(struct llog_handle *lgh, __u64 *off,
752 int curr, int goal, __u32 chunk_size,
755 struct llog_log_hdr *llh = lgh->lgh_hdr;
757 /* Goal should not bigger than the record count */
758 if (goal > lgh->lgh_last_idx)
759 goal = lgh->lgh_last_idx;
762 if (llh->llh_flags & LLOG_F_IS_FIXSIZE) {
763 *off = chunk_size + (goal - 1) * llh->llh_size;
765 __u64 min_rec_size = LLOG_MIN_REC_SIZE;
767 if (llh->llh_size > 0 && !force_mini_rec)
768 min_rec_size = llh->llh_size;
770 *off = *off + (goal - curr - 1) * min_rec_size;
773 /* always align with lower chunk boundary*/
774 *off &= ~(chunk_size - 1);
778 * Remove optional fields that the client doesn't expect.
779 * This is typically in order to ensure compatibility with older clients.
780 * It is assumed that since we exclusively remove fields, the block will be
781 * big enough to handle the remapped records. It is also assumed that records
782 * of a block have the same format (i.e.: the same features enabled).
784 * \param[in,out] hdr Header of the block of records to remap.
785 * \param[in,out] last_hdr Last header, don't read past this point.
786 * \param[in] flags Flags describing the fields to keep.
787 * \param[in] extra_flags Flags describing the extra fields to keep.
789 static void changelog_block_trim_ext(struct llog_rec_hdr *hdr,
790 struct llog_rec_hdr *last_hdr,
791 struct llog_handle *loghandle)
793 enum changelog_rec_flags flags = CLF_SUPPORTED;
794 enum changelog_rec_extra_flags extra_flags = CLFE_SUPPORTED;
796 if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_X_OMODE))
797 extra_flags &= ~CLFE_OPEN;
798 if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_X_NID))
799 extra_flags &= ~CLFE_NID;
800 if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_X_UIDGID))
801 extra_flags &= ~CLFE_UIDGID;
802 if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_EXTRA_FLAGS))
803 flags &= ~CLF_EXTRA_FLAGS;
804 if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_JOBID))
807 if (flags == CLF_SUPPORTED && extra_flags == CLFE_SUPPORTED)
810 if (hdr->lrh_type != CHANGELOG_REC)
814 struct changelog_rec *rec = (struct changelog_rec *)(hdr + 1);
815 enum changelog_rec_extra_flags xflag = CLFE_INVALID;
817 if (flags & CLF_EXTRA_FLAGS &&
818 rec->cr_flags & CLF_EXTRA_FLAGS) {
819 xflag = changelog_rec_extra_flags(rec)->cr_extra_flags &
823 changelog_remap_rec(rec, rec->cr_flags & flags, xflag);
824 hdr = llog_rec_hdr_next(hdr);
825 } while ((char *)hdr <= (char *)last_hdr);
829 * Implementation of the llog_operations::lop_next_block
831 * This function finds the the next llog block to return which contains
832 * record with required index. It is main part of llog processing.
834 * \param[in] env execution environment
835 * \param[in] loghandle llog handle of the current llog
836 * \param[in,out] cur_idx index preceeding cur_offset
837 * \param[in] next_idx target index to find
838 * \param[in,out] cur_offset furtherst point read in the file
839 * \param[in] buf pointer to data buffer to fill
840 * \param[in] len required len to read, it is
841 * usually llog chunk_size.
843 * \retval 0 on successful buffer read
844 * \retval negative value on error
846 static int llog_osd_next_block(const struct lu_env *env,
847 struct llog_handle *loghandle, int *cur_idx,
848 int next_idx, __u64 *cur_offset, void *buf,
851 struct llog_thread_info *lgi = llog_info(env);
853 struct dt_device *dt;
856 int last_idx = *cur_idx;
857 __u64 last_offset = *cur_offset;
858 bool force_mini_rec = false;
865 chunk_size = loghandle->lgh_hdr->llh_hdr.lrh_len;
866 if (len == 0 || len & (chunk_size - 1))
870 LASSERT(loghandle->lgh_ctxt);
872 o = loghandle->lgh_obj;
874 LASSERT(dt_object_exists(o));
875 dt = lu2dt_dev(o->do_lu.lo_dev);
878 rc = dt_attr_get(env, o, &lgi->lgi_attr);
882 CDEBUG(D_OTHER, "looking for log index %u (cur idx %u off"
883 "%llu), size %llu\n", next_idx, *cur_idx,
884 *cur_offset, lgi->lgi_attr.la_size);
886 while (*cur_offset < lgi->lgi_attr.la_size) {
887 struct llog_rec_hdr *rec, *last_rec;
888 struct llog_rec_tail *tail;
890 llog_skip_over(loghandle, cur_offset, *cur_idx,
891 next_idx, chunk_size, force_mini_rec);
893 /* read up to next llog chunk_size block */
894 lgi->lgi_buf.lb_len = chunk_size -
895 (*cur_offset & (chunk_size - 1));
896 lgi->lgi_buf.lb_buf = buf;
898 rc = dt_read(env, o, &lgi->lgi_buf, cur_offset);
900 if (rc == -EBADR && !force_mini_rec)
903 CERROR("%s: can't read llog block from log "DFID
904 " offset %llu: rc = %d\n",
905 o->do_lu.lo_dev->ld_obd->obd_name,
906 PFID(lu_object_fid(&o->do_lu)), *cur_offset,
912 /* signal the end of the valid buffer to
914 memset(buf + rc, 0, len - rc);
917 if (rc == 0) { /* end of file, nothing to do */
923 if (rc < sizeof(*tail)) {
927 CERROR("%s: invalid llog block at log id "DFID":%x "
929 o->do_lu.lo_dev->ld_obd->obd_name,
930 PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
931 loghandle->lgh_id.lgl_ogen, *cur_offset);
932 GOTO(out, rc = -EINVAL);
936 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
937 lustre_swab_llog_rec(rec);
939 tail = (struct llog_rec_tail *)((char *)buf + rc -
940 sizeof(struct llog_rec_tail));
941 /* get the last record in block */
942 last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
945 if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
946 lustre_swab_llog_rec(last_rec);
948 if (last_rec->lrh_index != tail->lrt_index) {
949 CERROR("%s: invalid llog tail at log id "DFID":%x "
950 "offset %llu last_rec idx %u tail idx %u"
951 "lrt len %u read_size %d\n",
952 o->do_lu.lo_dev->ld_obd->obd_name,
953 PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
954 loghandle->lgh_id.lgl_ogen, *cur_offset,
955 last_rec->lrh_index, tail->lrt_index,
957 GOTO(out, rc = -EINVAL);
960 *cur_idx = tail->lrt_index;
962 /* this shouldn't happen */
963 if (tail->lrt_index == 0) {
964 CERROR("%s: invalid llog tail at log id "DFID":%x "
965 "offset %llu bytes %d\n",
966 o->do_lu.lo_dev->ld_obd->obd_name,
967 PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
968 loghandle->lgh_id.lgl_ogen, *cur_offset, rc);
969 GOTO(out, rc = -EINVAL);
971 if (tail->lrt_index < next_idx) {
973 last_offset = *cur_offset;
977 /* sanity check that the start of the new buffer is no farther
978 * than the record that we wanted. This shouldn't happen. */
979 if (rec->lrh_index > next_idx) {
980 if (!force_mini_rec && next_idx > last_idx)
983 CERROR("%s: missed desired record? %u > %u\n",
984 o->do_lu.lo_dev->ld_obd->obd_name,
985 rec->lrh_index, next_idx);
986 GOTO(out, rc = -ENOENT);
989 /* Trim unsupported extensions for compat w/ older clients */
990 changelog_block_trim_ext(rec, last_rec, loghandle);
995 /* Note: because there are some pad records in the
996 * llog, so llog_skip_over() might skip too much
997 * records, let's try skip again with minimum record */
998 force_mini_rec = true;
999 *cur_offset = last_offset;
1000 *cur_idx = last_idx;
1002 GOTO(out, rc = -EIO);
1008 * Implementation of the llog_operations::lop_prev_block
1010 * This function finds the llog block to return which contains
1011 * record with required index but in reverse order - from end of llog
1013 * It is main part of reverse llog processing.
1015 * \param[in] env execution environment
1016 * \param[in] loghandle llog handle of the current llog
1017 * \param[in] prev_idx target index to find
1018 * \param[in] buf pointer to data buffer to fill
1019 * \param[in] len required len to read, it is llog_chunk_size usually.
1021 * \retval 0 on successful buffer read
1022 * \retval negative value on error
1024 static int llog_osd_prev_block(const struct lu_env *env,
1025 struct llog_handle *loghandle,
1026 int prev_idx, void *buf, int len)
1028 struct llog_thread_info *lgi = llog_info(env);
1029 struct dt_object *o;
1030 struct dt_device *dt;
1037 chunk_size = loghandle->lgh_hdr->llh_hdr.lrh_len;
1038 if (len == 0 || len & (chunk_size - 1))
1041 CDEBUG(D_OTHER, "looking for log index %u\n", prev_idx);
1044 LASSERT(loghandle->lgh_ctxt);
1046 o = loghandle->lgh_obj;
1048 LASSERT(dt_object_exists(o));
1049 dt = lu2dt_dev(o->do_lu.lo_dev);
1052 /* Let's only use mini record size for previous block read
1054 cur_offset = chunk_size;
1055 llog_skip_over(loghandle, &cur_offset, 0, prev_idx,
1058 rc = dt_attr_get(env, o, &lgi->lgi_attr);
1062 while (cur_offset < lgi->lgi_attr.la_size) {
1063 struct llog_rec_hdr *rec, *last_rec;
1064 struct llog_rec_tail *tail;
1066 lgi->lgi_buf.lb_len = len;
1067 lgi->lgi_buf.lb_buf = buf;
1068 rc = dt_read(env, o, &lgi->lgi_buf, &cur_offset);
1070 CERROR("%s: can't read llog block from log "DFID
1071 " offset %llu: rc = %d\n",
1072 o->do_lu.lo_dev->ld_obd->obd_name,
1073 PFID(lu_object_fid(&o->do_lu)), cur_offset, rc);
1077 if (rc == 0) /* end of file, nothing to do */
1080 if (rc < sizeof(*tail)) {
1081 CERROR("%s: invalid llog block at log id "DFID":%x "
1083 o->do_lu.lo_dev->ld_obd->obd_name,
1084 PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
1085 loghandle->lgh_id.lgl_ogen, cur_offset);
1086 GOTO(out, rc = -EINVAL);
1090 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
1091 lustre_swab_llog_rec(rec);
1093 tail = (struct llog_rec_tail *)((char *)buf + rc -
1094 sizeof(struct llog_rec_tail));
1095 /* get the last record in block */
1096 last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
1097 le32_to_cpu(tail->lrt_len));
1099 if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
1100 lustre_swab_llog_rec(last_rec);
1101 LASSERT(last_rec->lrh_index == tail->lrt_index);
1103 /* this shouldn't happen */
1104 if (tail->lrt_index == 0) {
1105 CERROR("%s: invalid llog tail at log id "DFID":%x "
1107 o->do_lu.lo_dev->ld_obd->obd_name,
1108 PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
1109 loghandle->lgh_id.lgl_ogen, cur_offset);
1110 GOTO(out, rc = -EINVAL);
1112 if (tail->lrt_index < prev_idx)
1115 /* sanity check that the start of the new buffer is no farther
1116 * than the record that we wanted. This shouldn't happen. */
1117 if (rec->lrh_index > prev_idx) {
1118 CERROR("%s: missed desired record? %u > %u\n",
1119 o->do_lu.lo_dev->ld_obd->obd_name,
1120 rec->lrh_index, prev_idx);
1121 GOTO(out, rc = -ENOENT);
1124 /* Trim unsupported extensions for compat w/ older clients */
1125 changelog_block_trim_ext(rec, last_rec, loghandle);
1129 GOTO(out, rc = -EIO);
1135 * This is helper function to get llog directory object. It is used by named
1136 * llog operations to find/insert/delete llog entry from llog directory.
1138 * \param[in] env execution environment
1139 * \param[in] ctxt llog context
1141 * \retval dt_object of llog directory
1142 * \retval ERR_PTR of negative value on error
1144 static struct dt_object *llog_osd_dir_get(const struct lu_env *env,
1145 struct llog_ctxt *ctxt)
1147 struct dt_device *dt;
1148 struct dt_thread_info *dti = dt_info(env);
1149 struct dt_object *dir;
1152 dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
1153 if (ctxt->loc_dir == NULL) {
1154 rc = dt_root_get(env, dt, &dti->dti_fid);
1157 dir = dt_locate(env, dt, &dti->dti_fid);
1159 if (!IS_ERR(dir) && !dt_try_as_dir(env, dir)) {
1160 dt_object_put(env, dir);
1161 return ERR_PTR(-ENOTDIR);
1164 lu_object_get(&ctxt->loc_dir->do_lu);
1165 dir = ctxt->loc_dir;
1172 * Implementation of the llog_operations::lop_open
1174 * This function opens the llog by its logid or by name, it may open also
1175 * non existent llog and assing then new id to it.
1176 * The llog_open/llog_close pair works similar to lu_object_find/put,
1177 * the object may not exist prior open. The result of open is just dt_object
1178 * in the llog header.
1180 * \param[in] env execution environment
1181 * \param[in] handle llog handle of the current llog
1182 * \param[in] logid logid of llog to open (nameless llog)
1183 * \param[in] name name of llog to open (named llog)
1184 * \param[in] open_param
1185 * LLOG_OPEN_NEW - new llog, may not exist
1186 * LLOG_OPEN_EXIST - old llog, must exist
1188 * \retval 0 on successful open, llog_handle::lgh_obj
1189 * contains the dt_object of the llog.
1190 * \retval negative value on error
1192 static int llog_osd_open(const struct lu_env *env, struct llog_handle *handle,
1193 struct llog_logid *logid, char *name,
1194 enum llog_open_param open_param)
1196 struct llog_thread_info *lgi = llog_info(env);
1197 struct llog_ctxt *ctxt = handle->lgh_ctxt;
1198 struct dt_object *o;
1199 struct dt_device *dt;
1200 struct ls_device *ls;
1201 struct local_oid_storage *los = NULL;
1203 bool new_id = false;
1209 LASSERT(ctxt->loc_exp);
1210 LASSERT(ctxt->loc_exp->exp_obd);
1211 dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
1213 if (ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1214 struct lu_object_conf conf = { 0 };
1215 if (logid != NULL) {
1216 logid_to_fid(logid, &lgi->lgi_fid);
1218 /* If logid == NULL, then it means the caller needs
1219 * to allocate new FID (llog_cat_declare_add_rec()). */
1220 rc = obd_fid_alloc(env, ctxt->loc_exp,
1221 &lgi->lgi_fid, NULL);
1225 conf.loc_flags = LOC_F_NEW;
1228 o = dt_locate_at(env, dt, &lgi->lgi_fid,
1229 dt->dd_lu_dev.ld_site->ls_top_dev, &conf);
1236 ls = ls_device_get(dt);
1238 RETURN(PTR_ERR(ls));
1240 mutex_lock(&ls->ls_los_mutex);
1241 los = dt_los_find(ls, name != NULL ? FID_SEQ_LLOG_NAME : FID_SEQ_LLOG);
1242 mutex_unlock(&ls->ls_los_mutex);
1244 ls_device_put(env, ls);
1248 if (logid != NULL) {
1249 logid_to_fid(logid, &lgi->lgi_fid);
1251 struct dt_object *llog_dir;
1253 llog_dir = llog_osd_dir_get(env, ctxt);
1254 if (IS_ERR(llog_dir))
1255 GOTO(out, rc = PTR_ERR(llog_dir));
1256 dt_read_lock(env, llog_dir, 0);
1257 rc = dt_lookup_dir(env, llog_dir, name, &lgi->lgi_fid);
1258 dt_read_unlock(env, llog_dir);
1259 dt_object_put(env, llog_dir);
1260 if (rc == -ENOENT && open_param == LLOG_OPEN_NEW) {
1261 /* generate fid for new llog */
1262 rc = local_object_fid_generate(env, los,
1268 OBD_ALLOC(handle->lgh_name, strlen(name) + 1);
1269 if (handle->lgh_name)
1270 strcpy(handle->lgh_name, name);
1272 GOTO(out, rc = -ENOMEM);
1274 LASSERTF(open_param & LLOG_OPEN_NEW, "%#x\n", open_param);
1275 /* generate fid for new llog */
1277 rc = local_object_fid_generate(env, los, &lgi->lgi_fid);
1283 o = ls_locate(env, ls, &lgi->lgi_fid, NULL);
1285 GOTO(out_name, rc = PTR_ERR(o));
1287 if (dt_object_exists(o) && new_id) {
1288 /* llog exists with just generated ID, e.g. some old llog file
1289 * still is in use or is orphan, drop a warn and skip it. */
1290 CDEBUG(D_INFO, "%s: llog exists with the same FID: "DFID
1292 o->do_lu.lo_dev->ld_obd->obd_name,
1293 PFID(lu_object_fid(&o->do_lu)));
1294 dt_object_put(env, o);
1295 /* just skip this llog ID, we shouldn't delete it because we
1296 * don't know exactly what is its purpose and state. */
1301 /* No new llog is expected but doesn't exist */
1302 if (open_param != LLOG_OPEN_NEW && !dt_object_exists(o)) {
1303 CDEBUG(D_INFO, "%s: llog FID: "DFID" obj %p doesn`t exist\n",
1304 o->do_lu.lo_dev->ld_obd->obd_name,
1305 PFID(lu_object_fid(&o->do_lu)), o);
1306 GOTO(out_put, rc = -ENOENT);
1308 fid_to_logid(&lgi->lgi_fid, &handle->lgh_id);
1309 handle->lgh_obj = o;
1310 handle->private_data = los;
1311 LASSERT(handle->lgh_ctxt);
1316 dt_object_put(env, o);
1318 if (handle->lgh_name != NULL)
1319 OBD_FREE(handle->lgh_name, strlen(name) + 1);
1327 * Get dir for regular fid log object
1329 * Get directory for regular fid log object, and these regular fid log
1330 * object will be inserted under this directory, to satisfy the FS
1331 * consistency check, e2fsck etc.
1333 * \param [in] env execution environment
1334 * \param [in] dto llog object
1336 * \retval pointer to the directory if it is found.
1337 * \retval ERR_PTR(negative errno) if it fails.
1339 struct dt_object *llog_osd_get_regular_fid_dir(const struct lu_env *env,
1340 struct dt_object *dto)
1342 struct llog_thread_info *lgi = llog_info(env);
1343 struct seq_server_site *ss = dto->do_lu.lo_dev->ld_site->ld_seq_site;
1344 struct lu_seq_range *range = &lgi->lgi_range;
1345 struct lu_fid *dir_fid = &lgi->lgi_fid;
1346 struct dt_object *dir;
1350 fld_range_set_any(range);
1351 LASSERT(ss != NULL);
1352 rc = ss->ss_server_fld->lsf_seq_lookup(env, ss->ss_server_fld,
1353 fid_seq(lu_object_fid(&dto->do_lu)), range);
1355 RETURN(ERR_PTR(rc));
1357 lu_update_log_dir_fid(dir_fid, range->lsr_index);
1358 dir = dt_locate(env, lu2dt_dev(dto->do_lu.lo_dev), dir_fid);
1362 if (!dt_try_as_dir(env, dir)) {
1363 dt_object_put(env, dir);
1364 RETURN(ERR_PTR(-ENOTDIR));
1371 * Add llog object with regular FID to name entry
1373 * Add llog object with regular FID to name space, and each llog
1374 * object on each MDT will be /update_log_dir/[seq:oid:ver],
1375 * so to satisfy the namespace consistency check, e2fsck etc.
1377 * \param [in] env execution environment
1378 * \param [in] dto llog object
1379 * \param [in] th thandle
1380 * \param [in] declare if it is declare or execution
1382 * \retval 0 if insertion succeeds.
1383 * \retval negative errno if insertion fails.
1386 llog_osd_regular_fid_add_name_entry(const struct lu_env *env,
1387 struct dt_object *dto,
1388 struct thandle *th, bool declare)
1390 struct llog_thread_info *lgi = llog_info(env);
1391 const struct lu_fid *fid = lu_object_fid(&dto->do_lu);
1392 struct dt_insert_rec *rec = &lgi->lgi_dt_rec;
1393 struct dt_object *dir;
1394 char *name = lgi->lgi_name;
1398 if (!fid_is_norm(fid))
1401 dir = llog_osd_get_regular_fid_dir(env, dto);
1403 RETURN(PTR_ERR(dir));
1406 rec->rec_type = S_IFREG;
1407 snprintf(name, sizeof(lgi->lgi_name), DFID, PFID(fid));
1408 dt_write_lock(env, dir, 0);
1410 rc = dt_declare_insert(env, dir, (struct dt_rec *)rec,
1411 (struct dt_key *)name, th);
1413 rc = dt_insert(env, dir, (struct dt_rec *)rec,
1414 (struct dt_key *)name, th, 1);
1416 dt_write_unlock(env, dir);
1418 dt_object_put(env, dir);
1424 * Implementation of the llog_operations::lop_declare_create
1426 * This function declares the llog create. It declares also name insert
1427 * into llog directory in case of named llog.
1429 * \param[in] env execution environment
1430 * \param[in] res llog handle of the current llog
1431 * \param[in] th current transaction handle
1433 * \retval 0 on successful create declaration
1434 * \retval negative value on error
1436 static int llog_osd_declare_create(const struct lu_env *env,
1437 struct llog_handle *res, struct thandle *th)
1439 struct llog_thread_info *lgi = llog_info(env);
1440 struct dt_insert_rec *rec = &lgi->lgi_dt_rec;
1441 struct local_oid_storage *los;
1442 struct dt_object *o;
1447 LASSERT(res->lgh_obj);
1450 /* object can be created by another thread */
1452 if (dt_object_exists(o))
1455 if (res->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1456 struct llog_thread_info *lgi = llog_info(env);
1458 lgi->lgi_attr.la_valid = LA_MODE | LA_SIZE;
1459 lgi->lgi_attr.la_size = 0;
1460 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1461 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1463 rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL,
1469 rc = llog_osd_regular_fid_add_name_entry(env, o, th, true);
1473 los = res->private_data;
1476 rc = llog_osd_declare_new_object(env, los, o, th);
1480 /* do not declare header initialization here as it's declared
1481 * in llog_osd_declare_write_rec() which is always called */
1483 if (res->lgh_name) {
1484 struct dt_object *llog_dir;
1486 llog_dir = llog_osd_dir_get(env, res->lgh_ctxt);
1487 if (IS_ERR(llog_dir))
1488 RETURN(PTR_ERR(llog_dir));
1489 logid_to_fid(&res->lgh_id, &lgi->lgi_fid);
1490 rec->rec_fid = &lgi->lgi_fid;
1491 rec->rec_type = S_IFREG;
1492 rc = dt_declare_insert(env, llog_dir,
1493 (struct dt_rec *)rec,
1494 (struct dt_key *)res->lgh_name, th);
1495 dt_object_put(env, llog_dir);
1497 CERROR("%s: can't declare named llog %s: rc = %d\n",
1498 o->do_lu.lo_dev->ld_obd->obd_name,
1505 * Implementation of the llog_operations::lop_create
1507 * This function creates the llog according with llog_handle::lgh_obj
1508 * and llog_handle::lgh_name.
1510 * \param[in] env execution environment
1511 * \param[in] res llog handle of the current llog
1512 * \param[in] th current transaction handle
1514 * \retval 0 on successful create
1515 * \retval negative value on error
1517 static int llog_osd_create(const struct lu_env *env, struct llog_handle *res,
1520 struct llog_thread_info *lgi = llog_info(env);
1521 struct dt_insert_rec *rec = &lgi->lgi_dt_rec;
1522 struct local_oid_storage *los;
1523 struct dt_object *o;
1532 /* llog can be already created */
1533 if (dt_object_exists(o))
1536 if (res->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1537 struct llog_thread_info *lgi = llog_info(env);
1539 lgi->lgi_attr.la_valid = LA_MODE | LA_SIZE | LA_TYPE;
1540 lgi->lgi_attr.la_size = 0;
1541 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1542 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1544 dt_write_lock(env, o, 0);
1545 rc = dt_create(env, o, &lgi->lgi_attr, NULL,
1547 dt_write_unlock(env, o);
1551 rc = llog_osd_regular_fid_add_name_entry(env, o, th, false);
1556 los = res->private_data;
1559 dt_write_lock(env, o, 0);
1560 if (!dt_object_exists(o))
1561 rc = llog_osd_create_new_object(env, los, o, th);
1565 dt_write_unlock(env, o);
1569 if (res->lgh_name) {
1570 struct dt_object *llog_dir;
1572 llog_dir = llog_osd_dir_get(env, res->lgh_ctxt);
1573 if (IS_ERR(llog_dir))
1574 RETURN(PTR_ERR(llog_dir));
1576 logid_to_fid(&res->lgh_id, &lgi->lgi_fid);
1577 rec->rec_fid = &lgi->lgi_fid;
1578 rec->rec_type = S_IFREG;
1579 dt_read_lock(env, llog_dir, 0);
1580 rc = dt_insert(env, llog_dir, (struct dt_rec *)rec,
1581 (struct dt_key *)res->lgh_name,
1583 dt_read_unlock(env, llog_dir);
1584 dt_object_put(env, llog_dir);
1586 CERROR("%s: can't create named llog %s: rc = %d\n",
1587 o->do_lu.lo_dev->ld_obd->obd_name,
1594 * Implementation of the llog_operations::lop_close
1596 * This function closes the llog. It just put llog object and referenced
1599 * \param[in] env execution environment
1600 * \param[in] handle llog handle of the current llog
1602 * \retval 0 on successful llog close
1603 * \retval negative value on error
1605 static int llog_osd_close(const struct lu_env *env, struct llog_handle *handle)
1607 struct local_oid_storage *los;
1612 LASSERT(handle->lgh_obj);
1614 if (handle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1615 /* Remove the object from the cache, otherwise it may
1616 * hold LOD being released during cleanup process */
1617 dt_object_put_nocache(env, handle->lgh_obj);
1618 LASSERT(handle->private_data == NULL);
1621 dt_object_put(env, handle->lgh_obj);
1623 los = handle->private_data;
1627 if (handle->lgh_name)
1628 OBD_FREE(handle->lgh_name, strlen(handle->lgh_name) + 1);
1634 * delete llog object name entry
1636 * Delete llog object (with regular FID) from name space (under
1639 * \param [in] env execution environment
1640 * \param [in] dto llog object
1641 * \param [in] th thandle
1642 * \param [in] declare if it is declare or execution
1644 * \retval 0 if deletion succeeds.
1645 * \retval negative errno if deletion fails.
1648 llog_osd_regular_fid_del_name_entry(const struct lu_env *env,
1649 struct dt_object *dto,
1650 struct thandle *th, bool declare)
1652 struct llog_thread_info *lgi = llog_info(env);
1653 const struct lu_fid *fid = lu_object_fid(&dto->do_lu);
1654 struct dt_object *dir;
1655 char *name = lgi->lgi_name;
1659 if (!fid_is_norm(fid))
1662 dir = llog_osd_get_regular_fid_dir(env, dto);
1664 RETURN(PTR_ERR(dir));
1666 snprintf(name, sizeof(lgi->lgi_name), DFID, PFID(fid));
1667 dt_write_lock(env, dir, 0);
1669 rc = dt_declare_delete(env, dir, (struct dt_key *)name,
1672 rc = dt_delete(env, dir, (struct dt_key *)name, th);
1674 dt_write_unlock(env, dir);
1676 dt_object_put(env, dir);
1681 * Implementation of the llog_operations::lop_declare_destroy
1683 * This function declare destroys the llog and deletes also entry in the
1684 * llog directory in case of named llog. Llog should be opened prior that.
1686 * \param[in] env execution environment
1687 * \param[in] loghandle llog handle of the current llog
1689 * \retval 0 on successful destroy
1690 * \retval negative value on error
1692 static int llog_osd_declare_destroy(const struct lu_env *env,
1693 struct llog_handle *loghandle,
1696 struct llog_ctxt *ctxt;
1697 struct dt_object *o, *llog_dir = NULL;
1702 ctxt = loghandle->lgh_ctxt;
1705 o = loghandle->lgh_obj;
1708 if (loghandle->lgh_name) {
1709 llog_dir = llog_osd_dir_get(env, ctxt);
1710 if (IS_ERR(llog_dir))
1711 RETURN(PTR_ERR(llog_dir));
1713 rc = dt_declare_delete(env, llog_dir,
1714 (struct dt_key *)loghandle->lgh_name,
1720 rc = dt_declare_ref_del(env, o, th);
1724 rc = dt_declare_destroy(env, o, th);
1728 if (loghandle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1729 rc = llog_osd_regular_fid_del_name_entry(env, o, th, true);
1735 if (!(IS_ERR_OR_NULL(llog_dir)))
1736 dt_object_put(env, llog_dir);
1743 * Implementation of the llog_operations::lop_destroy
1745 * This function destroys the llog and deletes also entry in the
1746 * llog directory in case of named llog. Llog should be opened prior that.
1747 * Destroy method is not part of external transaction and does everything
1750 * \param[in] env execution environment
1751 * \param[in] loghandle llog handle of the current llog
1753 * \retval 0 on successful destroy
1754 * \retval negative value on error
1756 static int llog_osd_destroy(const struct lu_env *env,
1757 struct llog_handle *loghandle, struct thandle *th)
1759 struct llog_ctxt *ctxt;
1760 struct dt_object *o, *llog_dir = NULL;
1765 ctxt = loghandle->lgh_ctxt;
1766 LASSERT(ctxt != NULL);
1768 o = loghandle->lgh_obj;
1771 dt_write_lock(env, o, 0);
1772 if (!dt_object_exists(o))
1773 GOTO(out_unlock, rc = 0);
1775 if (loghandle->lgh_name) {
1776 llog_dir = llog_osd_dir_get(env, ctxt);
1777 if (IS_ERR(llog_dir))
1778 GOTO(out_unlock, rc = PTR_ERR(llog_dir));
1780 dt_read_lock(env, llog_dir, 0);
1781 rc = dt_delete(env, llog_dir,
1782 (struct dt_key *)loghandle->lgh_name,
1784 dt_read_unlock(env, llog_dir);
1786 CERROR("%s: can't remove llog %s: rc = %d\n",
1787 o->do_lu.lo_dev->ld_obd->obd_name,
1788 loghandle->lgh_name, rc);
1789 GOTO(out_unlock, rc);
1793 dt_ref_del(env, o, th);
1794 rc = dt_destroy(env, o, th);
1796 GOTO(out_unlock, rc);
1798 if (loghandle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1799 rc = llog_osd_regular_fid_del_name_entry(env, o, th, false);
1801 GOTO(out_unlock, rc);
1805 dt_write_unlock(env, o);
1806 if (!(IS_ERR_OR_NULL(llog_dir)))
1807 dt_object_put(env, llog_dir);
1812 * Implementation of the llog_operations::lop_setup
1814 * This function setup the llog on local storage.
1816 * \param[in] env execution environment
1817 * \param[in] obd obd device the llog belongs to
1818 * \param[in] olg the llog group, it is always zero group now.
1819 * \param[in] ctxt_idx the llog index, it defines the purpose of this llog.
1820 * Every new llog type have to use own index.
1821 * \param[in] disk_obd the storage obd, where llog is stored.
1823 * \retval 0 on successful llog setup
1824 * \retval negative value on error
1826 static int llog_osd_setup(const struct lu_env *env, struct obd_device *obd,
1827 struct obd_llog_group *olg, int ctxt_idx,
1828 struct obd_device *disk_obd)
1830 struct llog_thread_info *lgi = llog_info(env);
1831 struct llog_ctxt *ctxt;
1836 LASSERT(olg->olg_ctxts[ctxt_idx]);
1838 ctxt = llog_ctxt_get(olg->olg_ctxts[ctxt_idx]);
1841 if (disk_obd == NULL)
1844 /* initialize data allowing to generate new fids,
1845 * literally we need a sequece */
1846 lgi->lgi_fid.f_seq = FID_SEQ_LLOG;
1847 lgi->lgi_fid.f_oid = 1;
1848 lgi->lgi_fid.f_ver = 0;
1849 rc = local_oid_storage_init(env, disk_obd->obd_lvfs_ctxt.dt,
1851 &ctxt->loc_los_nameless);
1855 lgi->lgi_fid.f_seq = FID_SEQ_LLOG_NAME;
1856 lgi->lgi_fid.f_oid = 1;
1857 lgi->lgi_fid.f_ver = 0;
1858 rc = local_oid_storage_init(env, disk_obd->obd_lvfs_ctxt.dt,
1860 &ctxt->loc_los_named);
1862 local_oid_storage_fini(env, ctxt->loc_los_nameless);
1863 ctxt->loc_los_nameless = NULL;
1869 llog_ctxt_put(ctxt);
1874 * Implementation of the llog_operations::lop_cleanup
1876 * This function cleanups the llog on local storage.
1878 * \param[in] env execution environment
1879 * \param[in] ctxt the llog context
1883 static int llog_osd_cleanup(const struct lu_env *env, struct llog_ctxt *ctxt)
1885 if (ctxt->loc_los_nameless != NULL) {
1886 local_oid_storage_fini(env, ctxt->loc_los_nameless);
1887 ctxt->loc_los_nameless = NULL;
1890 if (ctxt->loc_los_named != NULL) {
1891 local_oid_storage_fini(env, ctxt->loc_los_named);
1892 ctxt->loc_los_named = NULL;
1898 struct llog_operations llog_osd_ops = {
1899 .lop_next_block = llog_osd_next_block,
1900 .lop_prev_block = llog_osd_prev_block,
1901 .lop_read_header = llog_osd_read_header,
1902 .lop_declare_destroy = llog_osd_declare_destroy,
1903 .lop_destroy = llog_osd_destroy,
1904 .lop_setup = llog_osd_setup,
1905 .lop_cleanup = llog_osd_cleanup,
1906 .lop_open = llog_osd_open,
1907 .lop_exist = llog_osd_exist,
1908 .lop_declare_create = llog_osd_declare_create,
1909 .lop_create = llog_osd_create,
1910 .lop_declare_write_rec = llog_osd_declare_write_rec,
1911 .lop_write_rec = llog_osd_write_rec,
1912 .lop_close = llog_osd_close,
1914 EXPORT_SYMBOL(llog_osd_ops);
1916 struct llog_operations llog_common_cat_ops = {
1917 .lop_next_block = llog_osd_next_block,
1918 .lop_prev_block = llog_osd_prev_block,
1919 .lop_read_header = llog_osd_read_header,
1920 .lop_declare_destroy = llog_osd_declare_destroy,
1921 .lop_destroy = llog_osd_destroy,
1922 .lop_setup = llog_osd_setup,
1923 .lop_cleanup = llog_osd_cleanup,
1924 .lop_open = llog_osd_open,
1925 .lop_exist = llog_osd_exist,
1926 .lop_declare_create = llog_osd_declare_create,
1927 .lop_create = llog_osd_create,
1928 .lop_declare_write_rec = llog_osd_declare_write_rec,
1929 .lop_write_rec = llog_osd_write_rec,
1930 .lop_close = llog_osd_close,
1931 .lop_add = llog_cat_add_rec,
1932 .lop_declare_add = llog_cat_declare_add_rec,
1934 EXPORT_SYMBOL(llog_common_cat_ops);
1937 * Read the special file which contains the list of llog catalogs IDs
1939 * This function reads the CATALOGS file which contains the array of llog
1940 * catalogs IDs. The main purpose of this file is to store OSP llogs indexed
1941 * by OST/MDT number.
1943 * \param[in] env execution environment
1944 * \param[in] d corresponding storage device
1945 * \param[in] idx position to start from, usually OST/MDT index
1946 * \param[in] count how many catalog IDs to read
1947 * \param[out] idarray the buffer for the data. If it is NULL then
1948 * function returns just number of catalog IDs
1950 * \param[in] fid LLOG_CATALOGS_OID for CATALOG object
1952 * \retval 0 on successful read of catalog IDs
1953 * \retval negative value on error
1954 * \retval positive value which is number of records in
1955 * the file if \a idarray is NULL
1957 int llog_osd_get_cat_list(const struct lu_env *env, struct dt_device *d,
1958 int idx, int count, struct llog_catid *idarray,
1959 const struct lu_fid *fid)
1961 struct llog_thread_info *lgi = llog_info(env);
1962 struct dt_object *o = NULL;
1970 size = sizeof(*idarray) * count;
1971 lgi->lgi_off = idx * sizeof(*idarray);
1973 lgi->lgi_fid = *fid;
1974 o = dt_locate(env, d, &lgi->lgi_fid);
1978 if (!dt_object_exists(o)) {
1979 th = dt_trans_create(env, d);
1981 GOTO(out, rc = PTR_ERR(th));
1983 lgi->lgi_attr.la_valid = LA_MODE;
1984 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1985 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1987 th->th_wait_submit = 1;
1988 /* Make the llog object creation synchronization, so
1989 * it will be reliable to the reference, especially
1990 * for remote reference */
1993 rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL,
1996 GOTO(out_trans, rc);
1998 rc = dt_trans_start_local(env, d, th);
2000 GOTO(out_trans, rc);
2002 dt_write_lock(env, o, 0);
2003 if (!dt_object_exists(o))
2004 rc = dt_create(env, o, &lgi->lgi_attr, NULL,
2006 dt_write_unlock(env, o);
2008 dt_trans_stop(env, d, th);
2013 rc = dt_attr_get(env, o, &lgi->lgi_attr);
2017 if (!S_ISREG(lgi->lgi_attr.la_mode)) {
2018 CERROR("%s: CATALOGS is not a regular file!: mode = %o\n",
2019 o->do_lu.lo_dev->ld_obd->obd_name,
2020 lgi->lgi_attr.la_mode);
2021 GOTO(out, rc = -ENOENT);
2024 CDEBUG(D_CONFIG, "cat list: disk size=%d, read=%d\n",
2025 (int)lgi->lgi_attr.la_size, size);
2027 /* return just number of llogs */
2028 if (idarray == NULL) {
2029 rc = lgi->lgi_attr.la_size / sizeof(*idarray);
2033 /* read for new ost index or for empty file */
2034 memset(idarray, 0, size);
2035 if (lgi->lgi_attr.la_size <= lgi->lgi_off)
2037 if (lgi->lgi_attr.la_size < lgi->lgi_off + size)
2038 size = lgi->lgi_attr.la_size - lgi->lgi_off;
2040 lgi->lgi_buf.lb_buf = idarray;
2041 lgi->lgi_buf.lb_len = size;
2042 rc = dt_record_read(env, o, &lgi->lgi_buf, &lgi->lgi_off);
2043 /* -EFAULT means the llog is a sparse file. This is not an error
2044 * after arbitrary OST index is supported. */
2045 if (rc < 0 && rc != -EFAULT) {
2046 CERROR("%s: error reading CATALOGS: rc = %d\n",
2047 o->do_lu.lo_dev->ld_obd->obd_name, rc);
2053 dt_object_put(env, o);
2056 EXPORT_SYMBOL(llog_osd_get_cat_list);
2059 * Write the special file which contains the list of llog catalogs IDs
2061 * This function writes the CATALOG file which contains the array of llog
2062 * catalogs IDs. It is used mostly to store OSP llogs indexed by OST/MDT
2065 * \param[in] env execution environment
2066 * \param[in] d corresponding storage device
2067 * \param[in] idx position to start from, usually OST/MDT index
2068 * \param[in] count how many catalog IDs to write
2069 * \param[out] idarray the buffer with the data to write.
2070 * \param[in] fid LLOG_CATALOGS_OID for CATALOG object
2072 * \retval 0 on successful write of catalog IDs
2073 * \retval negative value on error
2075 int llog_osd_put_cat_list(const struct lu_env *env, struct dt_device *d,
2076 int idx, int count, struct llog_catid *idarray,
2077 const struct lu_fid *fid)
2079 struct llog_thread_info *lgi = llog_info(env);
2080 struct dt_object *o = NULL;
2089 size = sizeof(*idarray) * count;
2090 lgi->lgi_off = idx * sizeof(*idarray);
2091 lgi->lgi_fid = *fid;
2093 o = dt_locate(env, d, &lgi->lgi_fid);
2097 if (!dt_object_exists(o))
2098 GOTO(out, rc = -ENOENT);
2100 rc = dt_attr_get(env, o, &lgi->lgi_attr);
2104 if (!S_ISREG(lgi->lgi_attr.la_mode)) {
2105 CERROR("%s: CATALOGS is not a regular file!: mode = %o\n",
2106 o->do_lu.lo_dev->ld_obd->obd_name,
2107 lgi->lgi_attr.la_mode);
2108 GOTO(out, rc = -ENOENT);
2111 th = dt_trans_create(env, d);
2113 GOTO(out, rc = PTR_ERR(th));
2115 lgi->lgi_buf.lb_len = size;
2116 lgi->lgi_buf.lb_buf = idarray;
2117 rc = dt_declare_record_write(env, o, &lgi->lgi_buf, lgi->lgi_off, th);
2119 GOTO(out_trans, rc);
2121 /* For update log, this happens during initialization,
2122 * see lod_sub_prep_llog(), and we need make sure catlog
2123 * file ID is written to catlist file(committed) before
2124 * cross-MDT operation write update records to catlog FILE,
2125 * otherwise, during failover these update records might
2127 if (fid_is_update_log(fid))
2130 rc = dt_trans_start_local(env, d, th);
2132 GOTO(out_trans, rc);
2134 th->th_wait_submit = 1;
2136 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
2138 CDEBUG(D_INODE, "can't write CATALOGS at index %d: rc = %d\n",
2141 dt_trans_stop(env, d, th);
2143 dt_object_put(env, o);
2146 EXPORT_SYMBOL(llog_osd_put_cat_list);