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, 2016, 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[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 (loghandle->lgh_cur_idx > 0) {
495 * The lgh_cur_offset can be used only if index is
498 if (idx != loghandle->lgh_cur_idx) {
499 CERROR("%s: modify index mismatch %d %d\n",
500 o->do_lu.lo_dev->ld_obd->obd_name, idx,
501 loghandle->lgh_cur_idx);
505 lgi->lgi_off = loghandle->lgh_cur_offset;
506 CDEBUG(D_OTHER, "modify record "DFID": idx:%u, "
507 "len:%u offset %llu\n",
508 PFID(&loghandle->lgh_id.lgl_oi.oi_fid), idx,
509 rec->lrh_len, (long long)lgi->lgi_off);
510 } else if (llh->llh_flags & LLOG_F_IS_FIXSIZE) {
511 lgi->lgi_off = llh->llh_hdr.lrh_len +
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.
788 static void changelog_block_trim_ext(struct llog_rec_hdr *hdr,
789 struct llog_rec_hdr *last_hdr,
790 enum changelog_rec_flags flags)
792 if (hdr->lrh_type != CHANGELOG_REC)
796 struct changelog_rec *rec = (struct changelog_rec *)(hdr + 1);
798 changelog_remap_rec(rec, rec->cr_flags & flags);
799 hdr = llog_rec_hdr_next(hdr);
800 } while ((char *)hdr <= (char *)last_hdr);
804 * Implementation of the llog_operations::lop_next_block
806 * This function finds the the next llog block to return which contains
807 * record with required index. It is main part of llog processing.
809 * \param[in] env execution environment
810 * \param[in] loghandle llog handle of the current llog
811 * \param[in,out] cur_idx index preceeding cur_offset
812 * \param[in] next_idx target index to find
813 * \param[in,out] cur_offset furtherst point read in the file
814 * \param[in] buf pointer to data buffer to fill
815 * \param[in] len required len to read, it is
816 * usually llog chunk_size.
818 * \retval 0 on successful buffer read
819 * \retval negative value on error
821 static int llog_osd_next_block(const struct lu_env *env,
822 struct llog_handle *loghandle, int *cur_idx,
823 int next_idx, __u64 *cur_offset, void *buf,
826 struct llog_thread_info *lgi = llog_info(env);
828 struct dt_device *dt;
831 int last_idx = *cur_idx;
832 __u64 last_offset = *cur_offset;
833 bool force_mini_rec = false;
840 chunk_size = loghandle->lgh_hdr->llh_hdr.lrh_len;
841 if (len == 0 || len & (chunk_size - 1))
845 LASSERT(loghandle->lgh_ctxt);
847 o = loghandle->lgh_obj;
849 LASSERT(dt_object_exists(o));
850 dt = lu2dt_dev(o->do_lu.lo_dev);
853 rc = dt_attr_get(env, o, &lgi->lgi_attr);
857 CDEBUG(D_OTHER, "looking for log index %u (cur idx %u off"
858 "%llu), size %llu\n", next_idx, *cur_idx,
859 *cur_offset, lgi->lgi_attr.la_size);
861 while (*cur_offset < lgi->lgi_attr.la_size) {
862 struct llog_rec_hdr *rec, *last_rec;
863 struct llog_rec_tail *tail;
865 llog_skip_over(loghandle, cur_offset, *cur_idx,
866 next_idx, chunk_size, force_mini_rec);
868 /* read up to next llog chunk_size block */
869 lgi->lgi_buf.lb_len = chunk_size -
870 (*cur_offset & (chunk_size - 1));
871 lgi->lgi_buf.lb_buf = buf;
873 rc = dt_read(env, o, &lgi->lgi_buf, cur_offset);
875 if (rc == -EBADR && !force_mini_rec)
878 CERROR("%s: can't read llog block from log "DFID
879 " offset %llu: rc = %d\n",
880 o->do_lu.lo_dev->ld_obd->obd_name,
881 PFID(lu_object_fid(&o->do_lu)), *cur_offset,
887 /* signal the end of the valid buffer to
889 memset(buf + rc, 0, len - rc);
892 if (rc == 0) { /* end of file, nothing to do */
898 if (rc < sizeof(*tail)) {
902 CERROR("%s: invalid llog block at log id "DFID":%x "
904 o->do_lu.lo_dev->ld_obd->obd_name,
905 PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
906 loghandle->lgh_id.lgl_ogen, *cur_offset);
907 GOTO(out, rc = -EINVAL);
911 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
912 lustre_swab_llog_rec(rec);
914 tail = (struct llog_rec_tail *)((char *)buf + rc -
915 sizeof(struct llog_rec_tail));
916 /* get the last record in block */
917 last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
920 if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
921 lustre_swab_llog_rec(last_rec);
923 if (last_rec->lrh_index != tail->lrt_index) {
924 CERROR("%s: invalid llog tail at log id "DFID":%x "
925 "offset %llu last_rec idx %u tail idx %u"
926 "lrt len %u read_size %d\n",
927 o->do_lu.lo_dev->ld_obd->obd_name,
928 PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
929 loghandle->lgh_id.lgl_ogen, *cur_offset,
930 last_rec->lrh_index, tail->lrt_index,
932 GOTO(out, rc = -EINVAL);
935 *cur_idx = tail->lrt_index;
937 /* this shouldn't happen */
938 if (tail->lrt_index == 0) {
939 CERROR("%s: invalid llog tail at log id "DFID":%x "
940 "offset %llu bytes %d\n",
941 o->do_lu.lo_dev->ld_obd->obd_name,
942 PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
943 loghandle->lgh_id.lgl_ogen, *cur_offset, rc);
944 GOTO(out, rc = -EINVAL);
946 if (tail->lrt_index < next_idx) {
948 last_offset = *cur_offset;
952 /* sanity check that the start of the new buffer is no farther
953 * than the record that we wanted. This shouldn't happen. */
954 if (rec->lrh_index > next_idx) {
955 if (!force_mini_rec && next_idx > last_idx)
958 CERROR("%s: missed desired record? %u > %u\n",
959 o->do_lu.lo_dev->ld_obd->obd_name,
960 rec->lrh_index, next_idx);
961 GOTO(out, rc = -ENOENT);
964 /* Trim unsupported extensions for compat w/ older clients */
965 if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_JOBID))
966 changelog_block_trim_ext(rec, last_rec,
967 CLF_VERSION | CLF_RENAME);
972 /* Note: because there are some pad records in the
973 * llog, so llog_skip_over() might skip too much
974 * records, let's try skip again with minimum record */
975 force_mini_rec = true;
976 *cur_offset = last_offset;
979 GOTO(out, rc = -EIO);
985 * Implementation of the llog_operations::lop_prev_block
987 * This function finds the llog block to return which contains
988 * record with required index but in reverse order - from end of llog
990 * It is main part of reverse llog processing.
992 * \param[in] env execution environment
993 * \param[in] loghandle llog handle of the current llog
994 * \param[in] prev_idx target index to find
995 * \param[in] buf pointer to data buffer to fill
996 * \param[in] len required len to read, it is llog_chunk_size usually.
998 * \retval 0 on successful buffer read
999 * \retval negative value on error
1001 static int llog_osd_prev_block(const struct lu_env *env,
1002 struct llog_handle *loghandle,
1003 int prev_idx, void *buf, int len)
1005 struct llog_thread_info *lgi = llog_info(env);
1006 struct dt_object *o;
1007 struct dt_device *dt;
1014 chunk_size = loghandle->lgh_hdr->llh_hdr.lrh_len;
1015 if (len == 0 || len & (chunk_size - 1))
1018 CDEBUG(D_OTHER, "looking for log index %u\n", prev_idx);
1021 LASSERT(loghandle->lgh_ctxt);
1023 o = loghandle->lgh_obj;
1025 LASSERT(dt_object_exists(o));
1026 dt = lu2dt_dev(o->do_lu.lo_dev);
1029 /* Let's only use mini record size for previous block read
1031 cur_offset = chunk_size;
1032 llog_skip_over(loghandle, &cur_offset, 0, prev_idx,
1035 rc = dt_attr_get(env, o, &lgi->lgi_attr);
1039 while (cur_offset < lgi->lgi_attr.la_size) {
1040 struct llog_rec_hdr *rec, *last_rec;
1041 struct llog_rec_tail *tail;
1043 lgi->lgi_buf.lb_len = len;
1044 lgi->lgi_buf.lb_buf = buf;
1045 rc = dt_read(env, o, &lgi->lgi_buf, &cur_offset);
1047 CERROR("%s: can't read llog block from log "DFID
1048 " offset %llu: rc = %d\n",
1049 o->do_lu.lo_dev->ld_obd->obd_name,
1050 PFID(lu_object_fid(&o->do_lu)), cur_offset, rc);
1054 if (rc == 0) /* end of file, nothing to do */
1057 if (rc < sizeof(*tail)) {
1058 CERROR("%s: invalid llog block at log id "DFID":%x "
1060 o->do_lu.lo_dev->ld_obd->obd_name,
1061 PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
1062 loghandle->lgh_id.lgl_ogen, cur_offset);
1063 GOTO(out, rc = -EINVAL);
1067 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
1068 lustre_swab_llog_rec(rec);
1070 tail = (struct llog_rec_tail *)((char *)buf + rc -
1071 sizeof(struct llog_rec_tail));
1072 /* get the last record in block */
1073 last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
1074 le32_to_cpu(tail->lrt_len));
1076 if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
1077 lustre_swab_llog_rec(last_rec);
1078 LASSERT(last_rec->lrh_index == tail->lrt_index);
1080 /* this shouldn't happen */
1081 if (tail->lrt_index == 0) {
1082 CERROR("%s: invalid llog tail at log id "DFID":%x "
1084 o->do_lu.lo_dev->ld_obd->obd_name,
1085 PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
1086 loghandle->lgh_id.lgl_ogen, cur_offset);
1087 GOTO(out, rc = -EINVAL);
1089 if (tail->lrt_index < prev_idx)
1092 /* sanity check that the start of the new buffer is no farther
1093 * than the record that we wanted. This shouldn't happen. */
1094 if (rec->lrh_index > prev_idx) {
1095 CERROR("%s: missed desired record? %u > %u\n",
1096 o->do_lu.lo_dev->ld_obd->obd_name,
1097 rec->lrh_index, prev_idx);
1098 GOTO(out, rc = -ENOENT);
1101 /* Trim unsupported extensions for compat w/ older clients */
1102 if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_JOBID))
1103 changelog_block_trim_ext(rec, last_rec,
1104 CLF_VERSION | CLF_RENAME);
1108 GOTO(out, rc = -EIO);
1114 * This is helper function to get llog directory object. It is used by named
1115 * llog operations to find/insert/delete llog entry from llog directory.
1117 * \param[in] env execution environment
1118 * \param[in] ctxt llog context
1120 * \retval dt_object of llog directory
1121 * \retval ERR_PTR of negative value on error
1123 static struct dt_object *llog_osd_dir_get(const struct lu_env *env,
1124 struct llog_ctxt *ctxt)
1126 struct dt_device *dt;
1127 struct dt_thread_info *dti = dt_info(env);
1128 struct dt_object *dir;
1131 dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
1132 if (ctxt->loc_dir == NULL) {
1133 rc = dt_root_get(env, dt, &dti->dti_fid);
1136 dir = dt_locate(env, dt, &dti->dti_fid);
1138 if (!IS_ERR(dir) && !dt_try_as_dir(env, dir)) {
1139 dt_object_put(env, dir);
1140 return ERR_PTR(-ENOTDIR);
1143 lu_object_get(&ctxt->loc_dir->do_lu);
1144 dir = ctxt->loc_dir;
1151 * Implementation of the llog_operations::lop_open
1153 * This function opens the llog by its logid or by name, it may open also
1154 * non existent llog and assing then new id to it.
1155 * The llog_open/llog_close pair works similar to lu_object_find/put,
1156 * the object may not exist prior open. The result of open is just dt_object
1157 * in the llog header.
1159 * \param[in] env execution environment
1160 * \param[in] handle llog handle of the current llog
1161 * \param[in] logid logid of llog to open (nameless llog)
1162 * \param[in] name name of llog to open (named llog)
1163 * \param[in] open_param
1164 * LLOG_OPEN_NEW - new llog, may not exist
1165 * LLOG_OPEN_EXIST - old llog, must exist
1167 * \retval 0 on successful open, llog_handle::lgh_obj
1168 * contains the dt_object of the llog.
1169 * \retval negative value on error
1171 static int llog_osd_open(const struct lu_env *env, struct llog_handle *handle,
1172 struct llog_logid *logid, char *name,
1173 enum llog_open_param open_param)
1175 struct llog_thread_info *lgi = llog_info(env);
1176 struct llog_ctxt *ctxt = handle->lgh_ctxt;
1177 struct dt_object *o;
1178 struct dt_device *dt;
1179 struct ls_device *ls;
1180 struct local_oid_storage *los = NULL;
1182 bool new_id = false;
1188 LASSERT(ctxt->loc_exp);
1189 LASSERT(ctxt->loc_exp->exp_obd);
1190 dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
1192 if (ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1193 struct lu_object_conf conf = { 0 };
1194 if (logid != NULL) {
1195 logid_to_fid(logid, &lgi->lgi_fid);
1197 /* If logid == NULL, then it means the caller needs
1198 * to allocate new FID (llog_cat_declare_add_rec()). */
1199 rc = obd_fid_alloc(env, ctxt->loc_exp,
1200 &lgi->lgi_fid, NULL);
1204 conf.loc_flags = LOC_F_NEW;
1207 o = dt_locate_at(env, dt, &lgi->lgi_fid,
1208 dt->dd_lu_dev.ld_site->ls_top_dev, &conf);
1215 ls = ls_device_get(dt);
1217 RETURN(PTR_ERR(ls));
1219 mutex_lock(&ls->ls_los_mutex);
1220 los = dt_los_find(ls, name != NULL ? FID_SEQ_LLOG_NAME : FID_SEQ_LLOG);
1221 mutex_unlock(&ls->ls_los_mutex);
1223 ls_device_put(env, ls);
1227 if (logid != NULL) {
1228 logid_to_fid(logid, &lgi->lgi_fid);
1230 struct dt_object *llog_dir;
1232 llog_dir = llog_osd_dir_get(env, ctxt);
1233 if (IS_ERR(llog_dir))
1234 GOTO(out, rc = PTR_ERR(llog_dir));
1235 dt_read_lock(env, llog_dir, 0);
1236 rc = dt_lookup_dir(env, llog_dir, name, &lgi->lgi_fid);
1237 dt_read_unlock(env, llog_dir);
1238 dt_object_put(env, llog_dir);
1239 if (rc == -ENOENT && open_param == LLOG_OPEN_NEW) {
1240 /* generate fid for new llog */
1241 rc = local_object_fid_generate(env, los,
1247 OBD_ALLOC(handle->lgh_name, strlen(name) + 1);
1248 if (handle->lgh_name)
1249 strcpy(handle->lgh_name, name);
1251 GOTO(out, rc = -ENOMEM);
1253 LASSERTF(open_param & LLOG_OPEN_NEW, "%#x\n", open_param);
1254 /* generate fid for new llog */
1256 rc = local_object_fid_generate(env, los, &lgi->lgi_fid);
1262 o = ls_locate(env, ls, &lgi->lgi_fid, NULL);
1264 GOTO(out_name, rc = PTR_ERR(o));
1266 if (dt_object_exists(o) && new_id) {
1267 /* llog exists with just generated ID, e.g. some old llog file
1268 * still is in use or is orphan, drop a warn and skip it. */
1269 CDEBUG(D_INFO, "%s: llog exists with the same FID: "DFID
1271 o->do_lu.lo_dev->ld_obd->obd_name,
1272 PFID(lu_object_fid(&o->do_lu)));
1273 dt_object_put(env, o);
1274 /* just skip this llog ID, we shouldn't delete it because we
1275 * don't know exactly what is its purpose and state. */
1280 /* No new llog is expected but doesn't exist */
1281 if (open_param != LLOG_OPEN_NEW && !dt_object_exists(o)) {
1282 CDEBUG(D_INFO, "%s: llog FID: "DFID" obj %p doesn`t exist\n",
1283 o->do_lu.lo_dev->ld_obd->obd_name,
1284 PFID(lu_object_fid(&o->do_lu)), o);
1285 GOTO(out_put, rc = -ENOENT);
1287 fid_to_logid(&lgi->lgi_fid, &handle->lgh_id);
1288 handle->lgh_obj = o;
1289 handle->private_data = los;
1290 LASSERT(handle->lgh_ctxt);
1295 dt_object_put(env, o);
1297 if (handle->lgh_name != NULL)
1298 OBD_FREE(handle->lgh_name, strlen(name) + 1);
1306 * Get dir for regular fid log object
1308 * Get directory for regular fid log object, and these regular fid log
1309 * object will be inserted under this directory, to satisfy the FS
1310 * consistency check, e2fsck etc.
1312 * \param [in] env execution environment
1313 * \param [in] dto llog object
1315 * \retval pointer to the directory if it is found.
1316 * \retval ERR_PTR(negative errno) if it fails.
1318 struct dt_object *llog_osd_get_regular_fid_dir(const struct lu_env *env,
1319 struct dt_object *dto)
1321 struct llog_thread_info *lgi = llog_info(env);
1322 struct seq_server_site *ss = dto->do_lu.lo_dev->ld_site->ld_seq_site;
1323 struct lu_seq_range *range = &lgi->lgi_range;
1324 struct lu_fid *dir_fid = &lgi->lgi_fid;
1325 struct dt_object *dir;
1329 fld_range_set_any(range);
1330 LASSERT(ss != NULL);
1331 rc = ss->ss_server_fld->lsf_seq_lookup(env, ss->ss_server_fld,
1332 fid_seq(lu_object_fid(&dto->do_lu)), range);
1334 RETURN(ERR_PTR(rc));
1336 lu_update_log_dir_fid(dir_fid, range->lsr_index);
1337 dir = dt_locate(env, lu2dt_dev(dto->do_lu.lo_dev), dir_fid);
1341 if (!dt_try_as_dir(env, dir)) {
1342 dt_object_put(env, dir);
1343 RETURN(ERR_PTR(-ENOTDIR));
1350 * Add llog object with regular FID to name entry
1352 * Add llog object with regular FID to name space, and each llog
1353 * object on each MDT will be /update_log_dir/[seq:oid:ver],
1354 * so to satisfy the namespace consistency check, e2fsck etc.
1356 * \param [in] env execution environment
1357 * \param [in] dto llog object
1358 * \param [in] th thandle
1359 * \param [in] declare if it is declare or execution
1361 * \retval 0 if insertion succeeds.
1362 * \retval negative errno if insertion fails.
1365 llog_osd_regular_fid_add_name_entry(const struct lu_env *env,
1366 struct dt_object *dto,
1367 struct thandle *th, bool declare)
1369 struct llog_thread_info *lgi = llog_info(env);
1370 const struct lu_fid *fid = lu_object_fid(&dto->do_lu);
1371 struct dt_insert_rec *rec = &lgi->lgi_dt_rec;
1372 struct dt_object *dir;
1373 char *name = lgi->lgi_name;
1377 if (!fid_is_norm(fid))
1380 dir = llog_osd_get_regular_fid_dir(env, dto);
1382 RETURN(PTR_ERR(dir));
1385 rec->rec_type = S_IFREG;
1386 snprintf(name, sizeof(lgi->lgi_name), DFID, PFID(fid));
1387 dt_write_lock(env, dir, 0);
1389 rc = dt_declare_insert(env, dir, (struct dt_rec *)rec,
1390 (struct dt_key *)name, th);
1392 rc = dt_insert(env, dir, (struct dt_rec *)rec,
1393 (struct dt_key *)name, th, 1);
1395 dt_write_unlock(env, dir);
1397 dt_object_put(env, dir);
1403 * Implementation of the llog_operations::lop_declare_create
1405 * This function declares the llog create. It declares also name insert
1406 * into llog directory in case of named llog.
1408 * \param[in] env execution environment
1409 * \param[in] res llog handle of the current llog
1410 * \param[in] th current transaction handle
1412 * \retval 0 on successful create declaration
1413 * \retval negative value on error
1415 static int llog_osd_declare_create(const struct lu_env *env,
1416 struct llog_handle *res, struct thandle *th)
1418 struct llog_thread_info *lgi = llog_info(env);
1419 struct dt_insert_rec *rec = &lgi->lgi_dt_rec;
1420 struct local_oid_storage *los;
1421 struct dt_object *o;
1426 LASSERT(res->lgh_obj);
1429 /* object can be created by another thread */
1431 if (dt_object_exists(o))
1434 if (res->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1435 struct llog_thread_info *lgi = llog_info(env);
1437 lgi->lgi_attr.la_valid = LA_MODE | LA_SIZE;
1438 lgi->lgi_attr.la_size = 0;
1439 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1440 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1442 rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL,
1448 rc = llog_osd_regular_fid_add_name_entry(env, o, th, true);
1452 los = res->private_data;
1455 rc = llog_osd_declare_new_object(env, los, o, th);
1459 /* do not declare header initialization here as it's declared
1460 * in llog_osd_declare_write_rec() which is always called */
1462 if (res->lgh_name) {
1463 struct dt_object *llog_dir;
1465 llog_dir = llog_osd_dir_get(env, res->lgh_ctxt);
1466 if (IS_ERR(llog_dir))
1467 RETURN(PTR_ERR(llog_dir));
1468 logid_to_fid(&res->lgh_id, &lgi->lgi_fid);
1469 rec->rec_fid = &lgi->lgi_fid;
1470 rec->rec_type = S_IFREG;
1471 rc = dt_declare_insert(env, llog_dir,
1472 (struct dt_rec *)rec,
1473 (struct dt_key *)res->lgh_name, th);
1474 dt_object_put(env, llog_dir);
1476 CERROR("%s: can't declare named llog %s: rc = %d\n",
1477 o->do_lu.lo_dev->ld_obd->obd_name,
1484 * Implementation of the llog_operations::lop_create
1486 * This function creates the llog according with llog_handle::lgh_obj
1487 * and llog_handle::lgh_name.
1489 * \param[in] env execution environment
1490 * \param[in] res llog handle of the current llog
1491 * \param[in] th current transaction handle
1493 * \retval 0 on successful create
1494 * \retval negative value on error
1496 static int llog_osd_create(const struct lu_env *env, struct llog_handle *res,
1499 struct llog_thread_info *lgi = llog_info(env);
1500 struct dt_insert_rec *rec = &lgi->lgi_dt_rec;
1501 struct local_oid_storage *los;
1502 struct dt_object *o;
1511 /* llog can be already created */
1512 if (dt_object_exists(o))
1515 if (res->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1516 struct llog_thread_info *lgi = llog_info(env);
1518 lgi->lgi_attr.la_valid = LA_MODE | LA_SIZE | LA_TYPE;
1519 lgi->lgi_attr.la_size = 0;
1520 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1521 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1523 dt_write_lock(env, o, 0);
1524 rc = dt_create(env, o, &lgi->lgi_attr, NULL,
1526 dt_write_unlock(env, o);
1530 rc = llog_osd_regular_fid_add_name_entry(env, o, th, false);
1535 los = res->private_data;
1538 dt_write_lock(env, o, 0);
1539 if (!dt_object_exists(o))
1540 rc = llog_osd_create_new_object(env, los, o, th);
1544 dt_write_unlock(env, o);
1548 if (res->lgh_name) {
1549 struct dt_object *llog_dir;
1551 llog_dir = llog_osd_dir_get(env, res->lgh_ctxt);
1552 if (IS_ERR(llog_dir))
1553 RETURN(PTR_ERR(llog_dir));
1555 logid_to_fid(&res->lgh_id, &lgi->lgi_fid);
1556 rec->rec_fid = &lgi->lgi_fid;
1557 rec->rec_type = S_IFREG;
1558 dt_read_lock(env, llog_dir, 0);
1559 rc = dt_insert(env, llog_dir, (struct dt_rec *)rec,
1560 (struct dt_key *)res->lgh_name,
1562 dt_read_unlock(env, llog_dir);
1563 dt_object_put(env, llog_dir);
1565 CERROR("%s: can't create named llog %s: rc = %d\n",
1566 o->do_lu.lo_dev->ld_obd->obd_name,
1573 * Implementation of the llog_operations::lop_close
1575 * This function closes the llog. It just put llog object and referenced
1578 * \param[in] env execution environment
1579 * \param[in] handle llog handle of the current llog
1581 * \retval 0 on successful llog close
1582 * \retval negative value on error
1584 static int llog_osd_close(const struct lu_env *env, struct llog_handle *handle)
1586 struct local_oid_storage *los;
1591 LASSERT(handle->lgh_obj);
1593 if (handle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1594 /* Remove the object from the cache, otherwise it may
1595 * hold LOD being released during cleanup process */
1596 dt_object_put_nocache(env, handle->lgh_obj);
1597 LASSERT(handle->private_data == NULL);
1600 dt_object_put(env, handle->lgh_obj);
1602 los = handle->private_data;
1606 if (handle->lgh_name)
1607 OBD_FREE(handle->lgh_name, strlen(handle->lgh_name) + 1);
1613 * delete llog object name entry
1615 * Delete llog object (with regular FID) from name space (under
1618 * \param [in] env execution environment
1619 * \param [in] dto llog object
1620 * \param [in] th thandle
1621 * \param [in] declare if it is declare or execution
1623 * \retval 0 if deletion succeeds.
1624 * \retval negative errno if deletion fails.
1627 llog_osd_regular_fid_del_name_entry(const struct lu_env *env,
1628 struct dt_object *dto,
1629 struct thandle *th, bool declare)
1631 struct llog_thread_info *lgi = llog_info(env);
1632 const struct lu_fid *fid = lu_object_fid(&dto->do_lu);
1633 struct dt_object *dir;
1634 char *name = lgi->lgi_name;
1638 if (!fid_is_norm(fid))
1641 dir = llog_osd_get_regular_fid_dir(env, dto);
1643 RETURN(PTR_ERR(dir));
1645 snprintf(name, sizeof(lgi->lgi_name), DFID, PFID(fid));
1646 dt_write_lock(env, dir, 0);
1648 rc = dt_declare_delete(env, dir, (struct dt_key *)name,
1651 rc = dt_delete(env, dir, (struct dt_key *)name, th);
1653 dt_write_unlock(env, dir);
1655 dt_object_put(env, dir);
1660 * Implementation of the llog_operations::lop_declare_destroy
1662 * This function declare destroys the llog and deletes also entry in the
1663 * llog directory in case of named llog. Llog should be opened prior that.
1665 * \param[in] env execution environment
1666 * \param[in] loghandle llog handle of the current llog
1668 * \retval 0 on successful destroy
1669 * \retval negative value on error
1671 static int llog_osd_declare_destroy(const struct lu_env *env,
1672 struct llog_handle *loghandle,
1675 struct llog_ctxt *ctxt;
1676 struct dt_object *o, *llog_dir = NULL;
1681 ctxt = loghandle->lgh_ctxt;
1684 o = loghandle->lgh_obj;
1687 if (loghandle->lgh_name) {
1688 llog_dir = llog_osd_dir_get(env, ctxt);
1689 if (IS_ERR(llog_dir))
1690 RETURN(PTR_ERR(llog_dir));
1692 rc = dt_declare_delete(env, llog_dir,
1693 (struct dt_key *)loghandle->lgh_name,
1699 rc = dt_declare_ref_del(env, o, th);
1703 rc = dt_declare_destroy(env, o, th);
1707 if (loghandle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1708 rc = llog_osd_regular_fid_del_name_entry(env, o, th, true);
1714 if (!(IS_ERR_OR_NULL(llog_dir)))
1715 dt_object_put(env, llog_dir);
1722 * Implementation of the llog_operations::lop_destroy
1724 * This function destroys the llog and deletes also entry in the
1725 * llog directory in case of named llog. Llog should be opened prior that.
1726 * Destroy method is not part of external transaction and does everything
1729 * \param[in] env execution environment
1730 * \param[in] loghandle llog handle of the current llog
1732 * \retval 0 on successful destroy
1733 * \retval negative value on error
1735 static int llog_osd_destroy(const struct lu_env *env,
1736 struct llog_handle *loghandle, struct thandle *th)
1738 struct llog_ctxt *ctxt;
1739 struct dt_object *o, *llog_dir = NULL;
1744 ctxt = loghandle->lgh_ctxt;
1745 LASSERT(ctxt != NULL);
1747 o = loghandle->lgh_obj;
1750 dt_write_lock(env, o, 0);
1751 if (!dt_object_exists(o))
1752 GOTO(out_unlock, rc = 0);
1754 if (loghandle->lgh_name) {
1755 llog_dir = llog_osd_dir_get(env, ctxt);
1756 if (IS_ERR(llog_dir))
1757 GOTO(out_unlock, rc = PTR_ERR(llog_dir));
1759 dt_read_lock(env, llog_dir, 0);
1760 rc = dt_delete(env, llog_dir,
1761 (struct dt_key *)loghandle->lgh_name,
1763 dt_read_unlock(env, llog_dir);
1765 CERROR("%s: can't remove llog %s: rc = %d\n",
1766 o->do_lu.lo_dev->ld_obd->obd_name,
1767 loghandle->lgh_name, rc);
1768 GOTO(out_unlock, rc);
1772 dt_ref_del(env, o, th);
1773 rc = dt_destroy(env, o, th);
1775 GOTO(out_unlock, rc);
1777 if (loghandle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1778 rc = llog_osd_regular_fid_del_name_entry(env, o, th, false);
1780 GOTO(out_unlock, rc);
1784 dt_write_unlock(env, o);
1785 if (!(IS_ERR_OR_NULL(llog_dir)))
1786 dt_object_put(env, llog_dir);
1791 * Implementation of the llog_operations::lop_setup
1793 * This function setup the llog on local storage.
1795 * \param[in] env execution environment
1796 * \param[in] obd obd device the llog belongs to
1797 * \param[in] olg the llog group, it is always zero group now.
1798 * \param[in] ctxt_idx the llog index, it defines the purpose of this llog.
1799 * Every new llog type have to use own index.
1800 * \param[in] disk_obd the storage obd, where llog is stored.
1802 * \retval 0 on successful llog setup
1803 * \retval negative value on error
1805 static int llog_osd_setup(const struct lu_env *env, struct obd_device *obd,
1806 struct obd_llog_group *olg, int ctxt_idx,
1807 struct obd_device *disk_obd)
1809 struct llog_thread_info *lgi = llog_info(env);
1810 struct llog_ctxt *ctxt;
1815 LASSERT(olg->olg_ctxts[ctxt_idx]);
1817 ctxt = llog_ctxt_get(olg->olg_ctxts[ctxt_idx]);
1820 if (disk_obd == NULL)
1823 /* initialize data allowing to generate new fids,
1824 * literally we need a sequece */
1825 lgi->lgi_fid.f_seq = FID_SEQ_LLOG;
1826 lgi->lgi_fid.f_oid = 1;
1827 lgi->lgi_fid.f_ver = 0;
1828 rc = local_oid_storage_init(env, disk_obd->obd_lvfs_ctxt.dt,
1830 &ctxt->loc_los_nameless);
1834 lgi->lgi_fid.f_seq = FID_SEQ_LLOG_NAME;
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_named);
1841 local_oid_storage_fini(env, ctxt->loc_los_nameless);
1842 ctxt->loc_los_nameless = NULL;
1848 llog_ctxt_put(ctxt);
1853 * Implementation of the llog_operations::lop_cleanup
1855 * This function cleanups the llog on local storage.
1857 * \param[in] env execution environment
1858 * \param[in] ctxt the llog context
1862 static int llog_osd_cleanup(const struct lu_env *env, struct llog_ctxt *ctxt)
1864 if (ctxt->loc_los_nameless != NULL) {
1865 local_oid_storage_fini(env, ctxt->loc_los_nameless);
1866 ctxt->loc_los_nameless = NULL;
1869 if (ctxt->loc_los_named != NULL) {
1870 local_oid_storage_fini(env, ctxt->loc_los_named);
1871 ctxt->loc_los_named = NULL;
1877 struct llog_operations llog_osd_ops = {
1878 .lop_next_block = llog_osd_next_block,
1879 .lop_prev_block = llog_osd_prev_block,
1880 .lop_read_header = llog_osd_read_header,
1881 .lop_declare_destroy = llog_osd_declare_destroy,
1882 .lop_destroy = llog_osd_destroy,
1883 .lop_setup = llog_osd_setup,
1884 .lop_cleanup = llog_osd_cleanup,
1885 .lop_open = llog_osd_open,
1886 .lop_exist = llog_osd_exist,
1887 .lop_declare_create = llog_osd_declare_create,
1888 .lop_create = llog_osd_create,
1889 .lop_declare_write_rec = llog_osd_declare_write_rec,
1890 .lop_write_rec = llog_osd_write_rec,
1891 .lop_close = llog_osd_close,
1893 EXPORT_SYMBOL(llog_osd_ops);
1895 struct llog_operations llog_common_cat_ops = {
1896 .lop_next_block = llog_osd_next_block,
1897 .lop_prev_block = llog_osd_prev_block,
1898 .lop_read_header = llog_osd_read_header,
1899 .lop_declare_destroy = llog_osd_declare_destroy,
1900 .lop_destroy = llog_osd_destroy,
1901 .lop_setup = llog_osd_setup,
1902 .lop_cleanup = llog_osd_cleanup,
1903 .lop_open = llog_osd_open,
1904 .lop_exist = llog_osd_exist,
1905 .lop_declare_create = llog_osd_declare_create,
1906 .lop_create = llog_osd_create,
1907 .lop_declare_write_rec = llog_osd_declare_write_rec,
1908 .lop_write_rec = llog_osd_write_rec,
1909 .lop_close = llog_osd_close,
1910 .lop_add = llog_cat_add_rec,
1911 .lop_declare_add = llog_cat_declare_add_rec,
1913 EXPORT_SYMBOL(llog_common_cat_ops);
1916 * Read the special file which contains the list of llog catalogs IDs
1918 * This function reads the CATALOGS file which contains the array of llog
1919 * catalogs IDs. The main purpose of this file is to store OSP llogs indexed
1920 * by OST/MDT number.
1922 * \param[in] env execution environment
1923 * \param[in] d corresponding storage device
1924 * \param[in] idx position to start from, usually OST/MDT index
1925 * \param[in] count how many catalog IDs to read
1926 * \param[out] idarray the buffer for the data. If it is NULL then
1927 * function returns just number of catalog IDs
1929 * \param[in] fid LLOG_CATALOGS_OID for CATALOG object
1931 * \retval 0 on successful read of catalog IDs
1932 * \retval negative value on error
1933 * \retval positive value which is number of records in
1934 * the file if \a idarray is NULL
1936 int llog_osd_get_cat_list(const struct lu_env *env, struct dt_device *d,
1937 int idx, int count, struct llog_catid *idarray,
1938 const struct lu_fid *fid)
1940 struct llog_thread_info *lgi = llog_info(env);
1941 struct dt_object *o = NULL;
1949 size = sizeof(*idarray) * count;
1950 lgi->lgi_off = idx * sizeof(*idarray);
1952 lgi->lgi_fid = *fid;
1953 o = dt_locate(env, d, &lgi->lgi_fid);
1957 if (!dt_object_exists(o)) {
1958 th = dt_trans_create(env, d);
1960 GOTO(out, rc = PTR_ERR(th));
1962 lgi->lgi_attr.la_valid = LA_MODE;
1963 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1964 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1966 th->th_wait_submit = 1;
1967 /* Make the llog object creation synchronization, so
1968 * it will be reliable to the reference, especially
1969 * for remote reference */
1972 rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL,
1975 GOTO(out_trans, rc);
1977 rc = dt_trans_start_local(env, d, th);
1979 GOTO(out_trans, rc);
1981 dt_write_lock(env, o, 0);
1982 if (!dt_object_exists(o))
1983 rc = dt_create(env, o, &lgi->lgi_attr, NULL,
1985 dt_write_unlock(env, o);
1987 dt_trans_stop(env, d, th);
1992 rc = dt_attr_get(env, o, &lgi->lgi_attr);
1996 if (!S_ISREG(lgi->lgi_attr.la_mode)) {
1997 CERROR("%s: CATALOGS is not a regular file!: mode = %o\n",
1998 o->do_lu.lo_dev->ld_obd->obd_name,
1999 lgi->lgi_attr.la_mode);
2000 GOTO(out, rc = -ENOENT);
2003 CDEBUG(D_CONFIG, "cat list: disk size=%d, read=%d\n",
2004 (int)lgi->lgi_attr.la_size, size);
2006 /* return just number of llogs */
2007 if (idarray == NULL) {
2008 rc = lgi->lgi_attr.la_size / sizeof(*idarray);
2012 /* read for new ost index or for empty file */
2013 memset(idarray, 0, size);
2014 if (lgi->lgi_attr.la_size <= lgi->lgi_off)
2016 if (lgi->lgi_attr.la_size < lgi->lgi_off + size)
2017 size = lgi->lgi_attr.la_size - lgi->lgi_off;
2019 lgi->lgi_buf.lb_buf = idarray;
2020 lgi->lgi_buf.lb_len = size;
2021 rc = dt_record_read(env, o, &lgi->lgi_buf, &lgi->lgi_off);
2022 /* -EFAULT means the llog is a sparse file. This is not an error
2023 * after arbitrary OST index is supported. */
2024 if (rc < 0 && rc != -EFAULT) {
2025 CERROR("%s: error reading CATALOGS: rc = %d\n",
2026 o->do_lu.lo_dev->ld_obd->obd_name, rc);
2032 dt_object_put(env, o);
2035 EXPORT_SYMBOL(llog_osd_get_cat_list);
2038 * Write the special file which contains the list of llog catalogs IDs
2040 * This function writes the CATALOG file which contains the array of llog
2041 * catalogs IDs. It is used mostly to store OSP llogs indexed by OST/MDT
2044 * \param[in] env execution environment
2045 * \param[in] d corresponding storage device
2046 * \param[in] idx position to start from, usually OST/MDT index
2047 * \param[in] count how many catalog IDs to write
2048 * \param[out] idarray the buffer with the data to write.
2049 * \param[in] fid LLOG_CATALOGS_OID for CATALOG object
2051 * \retval 0 on successful write of catalog IDs
2052 * \retval negative value on error
2054 int llog_osd_put_cat_list(const struct lu_env *env, struct dt_device *d,
2055 int idx, int count, struct llog_catid *idarray,
2056 const struct lu_fid *fid)
2058 struct llog_thread_info *lgi = llog_info(env);
2059 struct dt_object *o = NULL;
2068 size = sizeof(*idarray) * count;
2069 lgi->lgi_off = idx * sizeof(*idarray);
2070 lgi->lgi_fid = *fid;
2072 o = dt_locate(env, d, &lgi->lgi_fid);
2076 if (!dt_object_exists(o))
2077 GOTO(out, rc = -ENOENT);
2079 rc = dt_attr_get(env, o, &lgi->lgi_attr);
2083 if (!S_ISREG(lgi->lgi_attr.la_mode)) {
2084 CERROR("%s: CATALOGS is not a regular file!: mode = %o\n",
2085 o->do_lu.lo_dev->ld_obd->obd_name,
2086 lgi->lgi_attr.la_mode);
2087 GOTO(out, rc = -ENOENT);
2090 th = dt_trans_create(env, d);
2092 GOTO(out, rc = PTR_ERR(th));
2094 lgi->lgi_buf.lb_len = size;
2095 lgi->lgi_buf.lb_buf = idarray;
2096 rc = dt_declare_record_write(env, o, &lgi->lgi_buf, lgi->lgi_off, th);
2098 GOTO(out_trans, rc);
2100 /* For update log, this happens during initialization,
2101 * see lod_sub_prep_llog(), and we need make sure catlog
2102 * file ID is written to catlist file(committed) before
2103 * cross-MDT operation write update records to catlog FILE,
2104 * otherwise, during failover these update records might
2106 if (fid_is_update_log(fid))
2109 rc = dt_trans_start_local(env, d, th);
2111 GOTO(out_trans, rc);
2113 th->th_wait_submit = 1;
2115 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
2117 CDEBUG(D_INODE, "can't write CATALOGS at index %d: rc = %d\n",
2120 dt_trans_stop(env, d, th);
2122 dt_object_put(env, o);
2125 EXPORT_SYMBOL(llog_osd_put_cat_list);