4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2014 Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 * lustre/obdclass/llog_osd.c
35 * Low level llog routines on top of OSD API
37 * This file provides set of methods for llog operations on top of
38 * dt_device. It contains all supported llog_operations interfaces and
39 * supplimental functions.
41 * Author: Alexey Zhuravlev <alexey.zhuravlev@intel.com>
42 * Author: Mikhail Pershin <mike.pershin@intel.com>
45 #define DEBUG_SUBSYSTEM S_LOG
48 #include <obd_class.h>
49 #include <lustre_fid.h>
50 #include <dt_object.h>
52 #include "llog_internal.h"
53 #include "local_storage.h"
56 * Implementation of the llog_operations::lop_declare_create
58 * This function is a wrapper over local_storage API function
59 * local_object_declare_create().
61 * \param[in] env execution environment
62 * \param[in] los local_storage for bottom storage device
63 * \param[in] o dt_object to create
64 * \param[in] th current transaction handle
66 * \retval 0 on successful declaration of the new object
67 * \retval negative error if declaration was failed
69 static int llog_osd_declare_new_object(const struct lu_env *env,
70 struct local_oid_storage *los,
74 struct llog_thread_info *lgi = llog_info(env);
76 lgi->lgi_attr.la_valid = LA_MODE;
77 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
78 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
80 return local_object_declare_create(env, los, o, &lgi->lgi_attr,
85 * Implementation of the llog_operations::lop_create
87 * This function is a wrapper over local_storage API function
88 * local_object_create().
90 * \param[in] env execution environment
91 * \param[in] los local_storage for bottom storage device
92 * \param[in] o dt_object to create
93 * \param[in] th current transaction handle
95 * \retval 0 on successful creation of the new object
96 * \retval negative error if creation was failed
98 static int llog_osd_create_new_object(const struct lu_env *env,
99 struct local_oid_storage *los,
103 struct llog_thread_info *lgi = llog_info(env);
105 lgi->lgi_attr.la_valid = LA_MODE;
106 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
107 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
109 return local_object_create(env, los, o, &lgi->lgi_attr,
114 * Write a padding record to the llog
116 * This function writes a padding record to the end of llog. That may
117 * be needed if llog contains records of variable size, e.g. config logs
119 * The padding record just aligns llog to the llog chunk_size boundary if
120 * the current record doesn't fit in the remaining space.
122 * It allocates full length to avoid two separate writes for header and tail.
123 * Such 2-steps scheme needs extra protection and complex error handling.
125 * \param[in] env execution environment
126 * \param[in] o dt_object to create
127 * \param[in,out] off pointer to the padding start offset
128 * \param[in] len padding length
129 * \param[in] index index of the padding record in a llog
130 * \param[in] th current transaction handle
132 * \retval 0 on successful padding write
133 * \retval negative error if write failed
135 static int llog_osd_pad(const struct lu_env *env, struct dt_object *o,
136 loff_t *off, int len, int index, struct thandle *th)
138 struct llog_thread_info *lgi = llog_info(env);
139 struct llog_rec_hdr *rec;
140 struct llog_rec_tail *tail;
147 LASSERT(len >= LLOG_MIN_REC_SIZE && (len & 0x7) == 0);
154 rec->lrh_index = index;
155 rec->lrh_type = LLOG_PAD_MAGIC;
157 tail = rec_tail(rec);
159 tail->lrt_index = index;
161 lgi->lgi_buf.lb_buf = rec;
162 lgi->lgi_buf.lb_len = len;
163 rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
165 CERROR("%s: error writing padding record: rc = %d\n",
166 o->do_lu.lo_dev->ld_obd->obd_name, rc);
173 * Implementation of the llog_operations::lop_read_header
175 * This function reads the current llog header from the bottom storage
178 * \param[in] env execution environment
179 * \param[in] handle llog handle of the current llog
181 * \retval 0 on successful header read
182 * \retval negative error if read failed
184 static int llog_osd_read_header(const struct lu_env *env,
185 struct llog_handle *handle)
187 struct llog_rec_hdr *llh_hdr;
189 struct llog_thread_info *lgi;
190 enum llog_flag flags;
198 lgi = llog_info(env);
200 rc = dt_attr_get(env, o, &lgi->lgi_attr);
204 LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
206 if (lgi->lgi_attr.la_size == 0) {
207 CDEBUG(D_HA, "not reading header from 0-byte log\n");
211 flags = handle->lgh_hdr->llh_flags;
214 lgi->lgi_buf.lb_buf = handle->lgh_hdr;
215 lgi->lgi_buf.lb_len = handle->lgh_hdr_size;
216 rc = dt_read(env, o, &lgi->lgi_buf, &lgi->lgi_off);
217 llh_hdr = &handle->lgh_hdr->llh_hdr;
218 if (rc < sizeof(*llh_hdr) || rc < llh_hdr->lrh_len) {
219 CERROR("%s: error reading "DFID" log header size %d: rc = %d\n",
220 o->do_lu.lo_dev->ld_obd->obd_name,
221 PFID(lu_object_fid(&o->do_lu)), rc < 0 ? 0 : rc,
230 if (LLOG_REC_HDR_NEEDS_SWABBING(llh_hdr))
231 lustre_swab_llog_hdr(handle->lgh_hdr);
233 if (llh_hdr->lrh_type != LLOG_HDR_MAGIC) {
234 CERROR("%s: bad log %s "DFID" header magic: %#x "
235 "(expected %#x)\n", o->do_lu.lo_dev->ld_obd->obd_name,
236 handle->lgh_name ? handle->lgh_name : "",
237 PFID(lu_object_fid(&o->do_lu)),
238 llh_hdr->lrh_type, LLOG_HDR_MAGIC);
240 } else if (llh_hdr->lrh_len < LLOG_MIN_CHUNK_SIZE ||
241 llh_hdr->lrh_len > handle->lgh_hdr_size) {
242 CERROR("%s: incorrectly sized log %s "DFID" header: "
243 "%#x (expected at least %#x)\n"
244 "you may need to re-run lconf --write_conf.\n",
245 o->do_lu.lo_dev->ld_obd->obd_name,
246 handle->lgh_name ? handle->lgh_name : "",
247 PFID(lu_object_fid(&o->do_lu)),
248 llh_hdr->lrh_len, LLOG_MIN_CHUNK_SIZE);
250 } else if (LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_index >
251 LLOG_HDR_BITMAP_SIZE(handle->lgh_hdr) ||
252 LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_len !=
254 CERROR("%s: incorrectly sized log %s "DFID" tailer: "
256 o->do_lu.lo_dev->ld_obd->obd_name,
257 handle->lgh_name ? handle->lgh_name : "",
258 PFID(lu_object_fid(&o->do_lu)),
259 LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_len, -EIO);
263 handle->lgh_hdr->llh_flags |= (flags & LLOG_F_EXT_MASK);
264 handle->lgh_last_idx = LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_index;
270 * Implementation of the llog_operations::lop_declare_write
272 * This function declares the new record write.
274 * \param[in] env execution environment
275 * \param[in] loghandle llog handle of the current llog
276 * \param[in] rec llog record header. This is a real header of the full
277 * llog record to write. This is the beginning of buffer
278 * to write, the length of buffer is stored in
280 * \param[in] idx index of the llog record. If \a idx == -1 then this is
281 * append case, otherwise \a idx is the index of record
283 * \param[in] th current transaction handle
285 * \retval 0 on successful declaration
286 * \retval negative error if declaration failed
288 static int llog_osd_declare_write_rec(const struct lu_env *env,
289 struct llog_handle *loghandle,
290 struct llog_rec_hdr *rec,
291 int idx, struct thandle *th)
293 struct llog_thread_info *lgi = llog_info(env);
304 LASSERT(rec->lrh_len <= loghandle->lgh_ctxt->loc_chunk_size);
306 o = loghandle->lgh_obj;
309 chunk_size = loghandle->lgh_ctxt->loc_chunk_size;
310 lgi->lgi_buf.lb_len = chunk_size;
311 lgi->lgi_buf.lb_buf = NULL;
312 /* each time we update header */
313 rc = dt_declare_record_write(env, o, &lgi->lgi_buf, 0,
315 if (rc || idx == 0) /* if error or just header */
319 * the pad record can be inserted so take into account double
322 lgi->lgi_buf.lb_len = chunk_size * 2;
323 lgi->lgi_buf.lb_buf = NULL;
324 /* XXX: implement declared window or multi-chunks approach */
325 rc = dt_declare_record_write(env, o, &lgi->lgi_buf, -1, th);
331 * Implementation of the llog_operations::lop_write
333 * This function writes the new record in the llog or modify the existed one.
335 * \param[in] env execution environment
336 * \param[in] loghandle llog handle of the current llog
337 * \param[in] rec llog record header. This is a real header of
338 * the full llog record to write. This is
339 * the beginning of buffer to write, the length
340 * of buffer is stored in \a rec::lrh_len
341 * \param[out] reccookie pointer to the cookie to return back if needed.
342 * It is used for further cancel of this llog
344 * \param[in] idx index of the llog record. If \a idx == -1 then
345 * this is append case, otherwise \a idx is
346 * the index of record to modify
347 * \param[in] th current transaction handle
349 * \retval 0 on successful write && \a reccookie == NULL
350 * 1 on successful write && \a reccookie != NULL
351 * \retval negative error if write failed
353 static int llog_osd_write_rec(const struct lu_env *env,
354 struct llog_handle *loghandle,
355 struct llog_rec_hdr *rec,
356 struct llog_cookie *reccookie,
357 int idx, struct thandle *th)
359 struct llog_thread_info *lgi = llog_info(env);
360 struct llog_log_hdr *llh;
361 int reclen = rec->lrh_len;
363 struct llog_rec_tail *lrt;
371 llh = loghandle->lgh_hdr;
373 o = loghandle->lgh_obj;
377 chunk_size = llh->llh_hdr.lrh_len;
378 CDEBUG(D_OTHER, "new record %x to "DFID"\n",
379 rec->lrh_type, PFID(lu_object_fid(&o->do_lu)));
381 /* record length should not bigger than */
382 if (reclen > loghandle->lgh_hdr->llh_hdr.lrh_len)
385 /* sanity check for fixed-records llog */
386 if (idx != LLOG_HEADER_IDX && (llh->llh_flags & LLOG_F_IS_FIXSIZE)) {
387 LASSERT(llh->llh_size != 0);
388 LASSERT(llh->llh_size == reclen);
391 rc = dt_attr_get(env, o, &lgi->lgi_attr);
396 * The modification case.
397 * If idx set then the record with that index must be modified.
398 * There are three cases possible:
399 * 1) the common case is the llog header update (idx == 0)
400 * 2) the llog record modification during llog process.
401 * This is indicated by the \a loghandle::lgh_cur_idx > 0.
402 * In that case the \a loghandle::lgh_cur_offset
403 * 3) otherwise this is assumed that llog consist of records of
404 * fixed size, i.e. catalog. The llog header must has llh_size
405 * field equal to record size. The record offset is calculated
406 * just by /a idx value
408 * During modification we don't need extra header update because
409 * the bitmap and record count are not changed. The record header
410 * and tail remains the same too.
412 if (idx != LLOG_NEXT_IDX) {
413 /* llog can be empty only when first record is being written */
414 LASSERT(ergo(idx > 0, lgi->lgi_attr.la_size > 0));
416 if (!ext2_test_bit(idx, LLOG_HDR_BITMAP(llh))) {
417 CERROR("%s: modify unset record %u\n",
418 o->do_lu.lo_dev->ld_obd->obd_name, idx);
422 if (idx != rec->lrh_index) {
423 CERROR("%s: modify index mismatch %d %u\n",
424 o->do_lu.lo_dev->ld_obd->obd_name, idx,
429 if (idx == LLOG_HEADER_IDX) {
430 /* llog header update */
431 __u32 *bitmap = LLOG_HDR_BITMAP(llh);
435 /* If it does not indicate the bitmap index
436 * (reccookie == NULL), then it means update
437 * the whole update header. Otherwise only
438 * update header and bits needs to be updated,
439 * and in DNE cases, it will signaficantly
440 * shrink the RPC size.
441 * see distribute_txn_cancel_records()*/
442 if (reccookie == NULL) {
443 lgi->lgi_buf.lb_len = reclen;
444 lgi->lgi_buf.lb_buf = rec;
445 rc = dt_record_write(env, o, &lgi->lgi_buf,
450 /* update the header */
451 lgi->lgi_buf.lb_len = llh->llh_bitmap_offset;
452 lgi->lgi_buf.lb_buf = llh;
453 rc = dt_record_write(env, o, &lgi->lgi_buf,
458 /* update the bitmap */
459 index = reccookie->lgc_index;
460 lgi->lgi_off = llh->llh_bitmap_offset +
461 (index / (sizeof(*bitmap) * 8)) *
463 lgi->lgi_buf.lb_len = sizeof(*bitmap);
464 lgi->lgi_buf.lb_buf =
465 &bitmap[index/(sizeof(*bitmap)*8)];
466 rc = dt_record_write(env, o, &lgi->lgi_buf,
470 } else if (loghandle->lgh_cur_idx > 0) {
472 * The lgh_cur_offset can be used only if index is
475 if (idx != loghandle->lgh_cur_idx) {
476 CERROR("%s: modify index mismatch %d %d\n",
477 o->do_lu.lo_dev->ld_obd->obd_name, idx,
478 loghandle->lgh_cur_idx);
482 lgi->lgi_off = loghandle->lgh_cur_offset;
483 CDEBUG(D_OTHER, "modify record "DOSTID": idx:%d, "
484 "len:%u offset %llu\n",
485 POSTID(&loghandle->lgh_id.lgl_oi), idx,
486 rec->lrh_len, (long long)lgi->lgi_off);
487 } else if (llh->llh_flags & LLOG_F_IS_FIXSIZE) {
488 lgi->lgi_off = llh->llh_hdr.lrh_len +
491 /* This can be result of lgh_cur_idx is not set during
492 * llog processing or llh_size is not set to proper
493 * record size for fixed records llog. Therefore it is
494 * impossible to get record offset. */
495 CERROR("%s: can't get record offset, idx:%d, "
496 "len:%u.\n", o->do_lu.lo_dev->ld_obd->obd_name,
501 /* update only data, header and tail remain the same */
502 lgi->lgi_off += sizeof(struct llog_rec_hdr);
503 lgi->lgi_buf.lb_len = REC_DATA_LEN(rec);
504 lgi->lgi_buf.lb_buf = REC_DATA(rec);
505 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
506 if (rc == 0 && reccookie) {
507 reccookie->lgc_lgl = loghandle->lgh_id;
508 reccookie->lgc_index = idx;
516 * The most common case of using llog. The new index is assigned to
517 * the new record, new bit is set in llog bitmap and llog count is
520 * Make sure that records don't cross a chunk boundary, so we can
521 * process them page-at-a-time if needed. If it will cross a chunk
522 * boundary, write in a fake (but referenced) entry to pad the chunk.
524 LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
525 lgi->lgi_off = lgi->lgi_attr.la_size;
526 left = chunk_size - (lgi->lgi_off & (chunk_size - 1));
527 /* NOTE: padding is a record, but no bit is set */
528 if (left != 0 && left != reclen &&
529 left < (reclen + LLOG_MIN_REC_SIZE)) {
530 index = loghandle->lgh_last_idx + 1;
531 rc = llog_osd_pad(env, o, &lgi->lgi_off, left, index, th);
534 loghandle->lgh_last_idx++; /* for pad rec */
536 /* if it's the last idx in log file, then return -ENOSPC
537 * or wrap around if a catalog */
538 if ((loghandle->lgh_last_idx >= LLOG_HDR_BITMAP_SIZE(llh) - 1) ||
539 unlikely(llh->llh_flags & LLOG_F_IS_CAT &&
540 OBD_FAIL_PRECHECK(OBD_FAIL_CAT_RECORDS) &&
541 loghandle->lgh_last_idx >= cfs_fail_val)) {
542 if (llh->llh_flags & LLOG_F_IS_CAT)
543 loghandle->lgh_last_idx = 0;
548 /* increment the last_idx along with llh_tail index, they should
549 * be equal for a llog lifetime */
550 loghandle->lgh_last_idx++;
551 index = loghandle->lgh_last_idx;
552 LLOG_HDR_TAIL(llh)->lrt_index = index;
554 * NB: the caller should make sure only 1 process access
555 * the lgh_last_idx, e.g. append should be exclusive.
556 * Otherwise it might hit the assert.
558 LASSERT(index < LLOG_HDR_BITMAP_SIZE(llh));
559 rec->lrh_index = index;
561 lrt->lrt_len = rec->lrh_len;
562 lrt->lrt_index = rec->lrh_index;
564 /* the lgh_hdr_mutex protects llog header data from concurrent
565 * update/cancel, the llh_count and llh_bitmap are protected */
566 mutex_lock(&loghandle->lgh_hdr_mutex);
567 if (ext2_set_bit(index, LLOG_HDR_BITMAP(llh))) {
568 CERROR("%s: index %u already set in log bitmap\n",
569 o->do_lu.lo_dev->ld_obd->obd_name, index);
570 mutex_unlock(&loghandle->lgh_hdr_mutex);
571 LBUG(); /* should never happen */
575 if (!(llh->llh_flags & LLOG_F_IS_FIXSIZE)) {
576 /* Update the minimum size of the llog record */
577 if (llh->llh_size == 0)
578 llh->llh_size = reclen;
579 else if (reclen < llh->llh_size)
580 llh->llh_size = reclen;
583 if (lgi->lgi_attr.la_size == 0) {
585 lgi->lgi_buf.lb_len = llh->llh_hdr.lrh_len;
586 lgi->lgi_buf.lb_buf = &llh->llh_hdr;
587 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
589 GOTO(out_unlock, rc);
591 __u32 *bitmap = LLOG_HDR_BITMAP(llh);
593 /* Note: If this is not initialization (size == 0), then do not
594 * write the whole header (8k bytes), only update header/tail
595 * and bits needs to be updated. Because this update might be
596 * part of cross-MDT operation, which needs to write these
597 * updates into the update log(32KB limit) and also pack inside
598 * the RPC (1MB limit), if we write 8K for each operation, which
599 * will cost a lot space, and keep us adding more updates to one
602 lgi->lgi_buf.lb_len = llh->llh_bitmap_offset;
603 lgi->lgi_buf.lb_buf = &llh->llh_hdr;
604 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
606 GOTO(out_unlock, rc);
608 lgi->lgi_off = llh->llh_bitmap_offset +
609 (index / (sizeof(*bitmap) * 8)) * sizeof(*bitmap);
610 lgi->lgi_buf.lb_len = sizeof(*bitmap);
611 lgi->lgi_buf.lb_buf = &bitmap[index/(sizeof(*bitmap)*8)];
612 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
614 GOTO(out_unlock, rc);
616 lgi->lgi_off = (unsigned long)LLOG_HDR_TAIL(llh) -
618 lgi->lgi_buf.lb_len = sizeof(llh->llh_tail);
619 lgi->lgi_buf.lb_buf = LLOG_HDR_TAIL(llh);
620 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
622 GOTO(out_unlock, rc);
626 /* unlock here for remote object */
627 mutex_unlock(&loghandle->lgh_hdr_mutex);
631 /* computed index can be used to determine offset for fixed-size
632 * records. This also allows to handle Catalog wrap around case */
633 if (llh->llh_flags & LLOG_F_IS_FIXSIZE) {
634 lgi->lgi_off = llh->llh_hdr.lrh_len + (index - 1) * reclen;
636 rc = dt_attr_get(env, o, &lgi->lgi_attr);
640 LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
641 lgi->lgi_off = max_t(__u64, lgi->lgi_attr.la_size,
645 lgi->lgi_buf.lb_len = reclen;
646 lgi->lgi_buf.lb_buf = rec;
647 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
651 CDEBUG(D_OTHER, "added record "DOSTID": idx: %u, %u off"LPU64"\n",
652 POSTID(&loghandle->lgh_id.lgl_oi), index, rec->lrh_len,
654 if (reccookie != NULL) {
655 reccookie->lgc_lgl = loghandle->lgh_id;
656 reccookie->lgc_index = index;
657 if ((rec->lrh_type == MDS_UNLINK_REC) ||
658 (rec->lrh_type == MDS_SETATTR64_REC))
659 reccookie->lgc_subsys = LLOG_MDS_OST_ORIG_CTXT;
660 else if (rec->lrh_type == OST_SZ_REC)
661 reccookie->lgc_subsys = LLOG_SIZE_ORIG_CTXT;
663 reccookie->lgc_subsys = -1;
668 /* cleanup llog for error case */
669 mutex_lock(&loghandle->lgh_hdr_mutex);
670 ext2_clear_bit(index, LLOG_HDR_BITMAP(llh));
672 mutex_unlock(&loghandle->lgh_hdr_mutex);
674 /* restore llog last_idx */
675 if (--loghandle->lgh_last_idx == 0 &&
676 (llh->llh_flags & LLOG_F_IS_CAT) && llh->llh_cat_idx != 0) {
677 /* catalog had just wrap-around case */
678 loghandle->lgh_last_idx = LLOG_HDR_BITMAP_SIZE(llh) - 1;
680 LLOG_HDR_TAIL(llh)->lrt_index = loghandle->lgh_last_idx;
686 * We can skip reading at least as many log blocks as the number of
687 * minimum sized log records we are skipping. If it turns out
688 * that we are not far enough along the log (because the
689 * actual records are larger than minimum size) we just skip
692 * Note: in llog_process_thread, it will use bitmap offset as
693 * the index to locate the record, which also includs some pad
694 * records, whose record size is very small, and it also does not
695 * consider pad record when recording minimum record size (otherwise
696 * min_record size might be too small), so in some rare cases,
697 * it might skip too much record for @goal, see llog_osd_next_block().
699 * When force_mini_rec is true, it means we have to use LLOG_MIN_REC_SIZE
700 * as the min record size to skip over, usually because in the previous
701 * try, it skip too much record, see loog_osd_next(prev)_block().
703 static inline void llog_skip_over(struct llog_handle *lgh, __u64 *off,
704 int curr, int goal, __u32 chunk_size,
707 struct llog_log_hdr *llh = lgh->lgh_hdr;
709 /* Goal should not bigger than the record count */
710 if (goal > lgh->lgh_last_idx)
711 goal = lgh->lgh_last_idx;
714 if (llh->llh_flags & LLOG_F_IS_FIXSIZE) {
715 *off = chunk_size + (goal - 1) * llh->llh_size;
717 __u64 min_rec_size = LLOG_MIN_REC_SIZE;
719 if (llh->llh_size > 0 && !force_mini_rec)
720 min_rec_size = llh->llh_size;
722 *off = *off + (goal - curr - 1) * min_rec_size;
725 /* always align with lower chunk boundary*/
726 *off &= ~(chunk_size - 1);
730 * Remove optional fields that the client doesn't expect.
731 * This is typically in order to ensure compatibility with older clients.
732 * It is assumed that since we exclusively remove fields, the block will be
733 * big enough to handle the remapped records. It is also assumed that records
734 * of a block have the same format (i.e.: the same features enabled).
736 * \param[in,out] hdr Header of the block of records to remap.
737 * \param[in,out] last_hdr Last header, don't read past this point.
738 * \param[in] flags Flags describing the fields to keep.
740 static void changelog_block_trim_ext(struct llog_rec_hdr *hdr,
741 struct llog_rec_hdr *last_hdr,
742 enum changelog_rec_flags flags)
744 if (hdr->lrh_type != CHANGELOG_REC)
748 struct changelog_rec *rec = (struct changelog_rec *)(hdr + 1);
750 changelog_remap_rec(rec, rec->cr_flags & flags);
751 hdr = llog_rec_hdr_next(hdr);
752 } while ((char *)hdr <= (char *)last_hdr);
756 * Implementation of the llog_operations::lop_next_block
758 * This function finds the the next llog block to return which contains
759 * record with required index. It is main part of llog processing.
761 * \param[in] env execution environment
762 * \param[in] loghandle llog handle of the current llog
763 * \param[in,out] cur_idx index preceeding cur_offset
764 * \param[in] next_idx target index to find
765 * \param[in,out] cur_offset furtherst point read in the file
766 * \param[in] buf pointer to data buffer to fill
767 * \param[in] len required len to read, it is
768 * usually llog chunk_size.
770 * \retval 0 on successful buffer read
771 * \retval negative value on error
773 static int llog_osd_next_block(const struct lu_env *env,
774 struct llog_handle *loghandle, int *cur_idx,
775 int next_idx, __u64 *cur_offset, void *buf,
778 struct llog_thread_info *lgi = llog_info(env);
780 struct dt_device *dt;
783 int last_idx = *cur_idx;
784 __u64 last_offset = *cur_offset;
785 bool force_mini_rec = false;
792 chunk_size = loghandle->lgh_hdr->llh_hdr.lrh_len;
793 if (len == 0 || len & (chunk_size - 1))
796 CDEBUG(D_OTHER, "looking for log index %u (cur idx %u off "LPU64")\n",
797 next_idx, *cur_idx, *cur_offset);
800 LASSERT(loghandle->lgh_ctxt);
802 o = loghandle->lgh_obj;
804 LASSERT(dt_object_exists(o));
805 dt = lu2dt_dev(o->do_lu.lo_dev);
808 rc = dt_attr_get(env, o, &lgi->lgi_attr);
812 while (*cur_offset < lgi->lgi_attr.la_size) {
813 struct llog_rec_hdr *rec, *last_rec;
814 struct llog_rec_tail *tail;
816 llog_skip_over(loghandle, cur_offset, *cur_idx,
817 next_idx, chunk_size, force_mini_rec);
819 /* read up to next llog chunk_size block */
820 lgi->lgi_buf.lb_len = chunk_size -
821 (*cur_offset & (chunk_size - 1));
822 lgi->lgi_buf.lb_buf = buf;
824 rc = dt_read(env, o, &lgi->lgi_buf, cur_offset);
826 if (rc == -EBADR && !force_mini_rec)
829 CERROR("%s: can't read llog block from log "DFID
830 " offset "LPU64": rc = %d\n",
831 o->do_lu.lo_dev->ld_obd->obd_name,
832 PFID(lu_object_fid(&o->do_lu)), *cur_offset,
838 /* signal the end of the valid buffer to
840 memset(buf + rc, 0, len - rc);
843 if (rc == 0) { /* end of file, nothing to do */
849 if (rc < sizeof(*tail)) {
853 CERROR("%s: invalid llog block at log id "DOSTID"/%u "
855 o->do_lu.lo_dev->ld_obd->obd_name,
856 POSTID(&loghandle->lgh_id.lgl_oi),
857 loghandle->lgh_id.lgl_ogen, *cur_offset);
858 GOTO(out, rc = -EINVAL);
862 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
863 lustre_swab_llog_rec(rec);
865 tail = (struct llog_rec_tail *)((char *)buf + rc -
866 sizeof(struct llog_rec_tail));
867 /* get the last record in block */
868 last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
871 if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
872 lustre_swab_llog_rec(last_rec);
873 LASSERT(last_rec->lrh_index == tail->lrt_index);
875 *cur_idx = tail->lrt_index;
877 /* this shouldn't happen */
878 if (tail->lrt_index == 0) {
879 CERROR("%s: invalid llog tail at log id "DOSTID"/%u "
880 "offset "LPU64" bytes %d\n",
881 o->do_lu.lo_dev->ld_obd->obd_name,
882 POSTID(&loghandle->lgh_id.lgl_oi),
883 loghandle->lgh_id.lgl_ogen, *cur_offset, rc);
884 GOTO(out, rc = -EINVAL);
886 if (tail->lrt_index < next_idx) {
888 last_offset = *cur_offset;
892 /* sanity check that the start of the new buffer is no farther
893 * than the record that we wanted. This shouldn't happen. */
894 if (rec->lrh_index > next_idx) {
895 if (!force_mini_rec && next_idx > last_idx)
898 CERROR("%s: missed desired record? %u > %u\n",
899 o->do_lu.lo_dev->ld_obd->obd_name,
900 rec->lrh_index, next_idx);
901 GOTO(out, rc = -ENOENT);
904 /* Trim unsupported extensions for compat w/ older clients */
905 if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_JOBID))
906 changelog_block_trim_ext(rec, last_rec,
907 CLF_VERSION | CLF_RENAME);
912 /* Note: because there are some pad records in the
913 * llog, so llog_skip_over() might skip too much
914 * records, let's try skip again with minimum record */
915 force_mini_rec = true;
916 *cur_offset = last_offset;
919 GOTO(out, rc = -EIO);
925 * Implementation of the llog_operations::lop_prev_block
927 * This function finds the llog block to return which contains
928 * record with required index but in reverse order - from end of llog
930 * It is main part of reverse llog processing.
932 * \param[in] env execution environment
933 * \param[in] loghandle llog handle of the current llog
934 * \param[in] prev_idx target index to find
935 * \param[in] buf pointer to data buffer to fill
936 * \param[in] len required len to read, it is llog_chunk_size usually.
938 * \retval 0 on successful buffer read
939 * \retval negative value on error
941 static int llog_osd_prev_block(const struct lu_env *env,
942 struct llog_handle *loghandle,
943 int prev_idx, void *buf, int len)
945 struct llog_thread_info *lgi = llog_info(env);
947 struct dt_device *dt;
954 chunk_size = loghandle->lgh_hdr->llh_hdr.lrh_len;
955 if (len == 0 || len & (chunk_size - 1))
958 CDEBUG(D_OTHER, "looking for log index %u\n", prev_idx);
961 LASSERT(loghandle->lgh_ctxt);
963 o = loghandle->lgh_obj;
965 LASSERT(dt_object_exists(o));
966 dt = lu2dt_dev(o->do_lu.lo_dev);
969 /* Let's only use mini record size for previous block read
971 cur_offset = chunk_size;
972 llog_skip_over(loghandle, &cur_offset, 0, prev_idx,
975 rc = dt_attr_get(env, o, &lgi->lgi_attr);
979 while (cur_offset < lgi->lgi_attr.la_size) {
980 struct llog_rec_hdr *rec, *last_rec;
981 struct llog_rec_tail *tail;
983 lgi->lgi_buf.lb_len = len;
984 lgi->lgi_buf.lb_buf = buf;
985 rc = dt_read(env, o, &lgi->lgi_buf, &cur_offset);
987 CERROR("%s: can't read llog block from log "DFID
988 " offset "LPU64": rc = %d\n",
989 o->do_lu.lo_dev->ld_obd->obd_name,
990 PFID(lu_object_fid(&o->do_lu)), cur_offset, rc);
994 if (rc == 0) /* end of file, nothing to do */
997 if (rc < sizeof(*tail)) {
998 CERROR("%s: invalid llog block at log id "DOSTID"/%u "
1000 o->do_lu.lo_dev->ld_obd->obd_name,
1001 POSTID(&loghandle->lgh_id.lgl_oi),
1002 loghandle->lgh_id.lgl_ogen, cur_offset);
1003 GOTO(out, rc = -EINVAL);
1007 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
1008 lustre_swab_llog_rec(rec);
1010 tail = (struct llog_rec_tail *)((char *)buf + rc -
1011 sizeof(struct llog_rec_tail));
1012 /* get the last record in block */
1013 last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
1014 le32_to_cpu(tail->lrt_len));
1016 if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
1017 lustre_swab_llog_rec(last_rec);
1018 LASSERT(last_rec->lrh_index == tail->lrt_index);
1020 /* this shouldn't happen */
1021 if (tail->lrt_index == 0) {
1022 CERROR("%s: invalid llog tail at log id "DOSTID"/%u "
1024 o->do_lu.lo_dev->ld_obd->obd_name,
1025 POSTID(&loghandle->lgh_id.lgl_oi),
1026 loghandle->lgh_id.lgl_ogen, cur_offset);
1027 GOTO(out, rc = -EINVAL);
1029 if (tail->lrt_index < prev_idx)
1032 /* sanity check that the start of the new buffer is no farther
1033 * than the record that we wanted. This shouldn't happen. */
1034 if (rec->lrh_index > prev_idx) {
1035 CERROR("%s: missed desired record? %u > %u\n",
1036 o->do_lu.lo_dev->ld_obd->obd_name,
1037 rec->lrh_index, prev_idx);
1038 GOTO(out, rc = -ENOENT);
1041 /* Trim unsupported extensions for compat w/ older clients */
1042 if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_JOBID))
1043 changelog_block_trim_ext(rec, last_rec,
1044 CLF_VERSION | CLF_RENAME);
1048 GOTO(out, rc = -EIO);
1054 * This is helper function to get llog directory object. It is used by named
1055 * llog operations to find/insert/delete llog entry from llog directory.
1057 * \param[in] env execution environment
1058 * \param[in] ctxt llog context
1060 * \retval dt_object of llog directory
1061 * \retval ERR_PTR of negative value on error
1063 static struct dt_object *llog_osd_dir_get(const struct lu_env *env,
1064 struct llog_ctxt *ctxt)
1066 struct dt_device *dt;
1067 struct dt_thread_info *dti = dt_info(env);
1068 struct dt_object *dir;
1071 dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
1072 if (ctxt->loc_dir == NULL) {
1073 rc = dt_root_get(env, dt, &dti->dti_fid);
1076 dir = dt_locate(env, dt, &dti->dti_fid);
1078 if (!IS_ERR(dir) && !dt_try_as_dir(env, dir)) {
1079 lu_object_put(env, &dir->do_lu);
1080 return ERR_PTR(-ENOTDIR);
1083 lu_object_get(&ctxt->loc_dir->do_lu);
1084 dir = ctxt->loc_dir;
1091 * Implementation of the llog_operations::lop_open
1093 * This function opens the llog by its logid or by name, it may open also
1094 * non existent llog and assing then new id to it.
1095 * The llog_open/llog_close pair works similar to lu_object_find/put,
1096 * the object may not exist prior open. The result of open is just dt_object
1097 * in the llog header.
1099 * \param[in] env execution environment
1100 * \param[in] handle llog handle of the current llog
1101 * \param[in] logid logid of llog to open (nameless llog)
1102 * \param[in] name name of llog to open (named llog)
1103 * \param[in] open_param
1104 * LLOG_OPEN_NEW - new llog, may not exist
1105 * LLOG_OPEN_EXIST - old llog, must exist
1107 * \retval 0 on successful open, llog_handle::lgh_obj
1108 * contains the dt_object of the llog.
1109 * \retval negative value on error
1111 static int llog_osd_open(const struct lu_env *env, struct llog_handle *handle,
1112 struct llog_logid *logid, char *name,
1113 enum llog_open_param open_param)
1115 struct llog_thread_info *lgi = llog_info(env);
1116 struct llog_ctxt *ctxt = handle->lgh_ctxt;
1117 struct dt_object *o;
1118 struct dt_device *dt;
1119 struct ls_device *ls;
1120 struct local_oid_storage *los = NULL;
1127 LASSERT(ctxt->loc_exp);
1128 LASSERT(ctxt->loc_exp->exp_obd);
1129 dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
1131 if (ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1132 struct lu_object_conf conf = { 0 };
1133 if (logid != NULL) {
1134 logid_to_fid(logid, &lgi->lgi_fid);
1136 /* If logid == NULL, then it means the caller needs
1137 * to allocate new FID (llog_cat_declare_add_rec()). */
1138 rc = obd_fid_alloc(env, ctxt->loc_exp,
1139 &lgi->lgi_fid, NULL);
1143 conf.loc_flags = LOC_F_NEW;
1146 o = dt_locate_at(env, dt, &lgi->lgi_fid,
1147 dt->dd_lu_dev.ld_site->ls_top_dev, &conf);
1154 ls = ls_device_get(dt);
1156 RETURN(PTR_ERR(ls));
1158 mutex_lock(&ls->ls_los_mutex);
1159 los = dt_los_find(ls, name != NULL ? FID_SEQ_LLOG_NAME : FID_SEQ_LLOG);
1160 mutex_unlock(&ls->ls_los_mutex);
1162 ls_device_put(env, ls);
1166 if (logid != NULL) {
1167 logid_to_fid(logid, &lgi->lgi_fid);
1169 struct dt_object *llog_dir;
1171 llog_dir = llog_osd_dir_get(env, ctxt);
1172 if (IS_ERR(llog_dir))
1173 GOTO(out, rc = PTR_ERR(llog_dir));
1174 dt_read_lock(env, llog_dir, 0);
1175 rc = dt_lookup_dir(env, llog_dir, name, &lgi->lgi_fid);
1176 dt_read_unlock(env, llog_dir);
1177 lu_object_put(env, &llog_dir->do_lu);
1178 if (rc == -ENOENT && open_param == LLOG_OPEN_NEW) {
1179 /* generate fid for new llog */
1180 rc = local_object_fid_generate(env, los,
1185 OBD_ALLOC(handle->lgh_name, strlen(name) + 1);
1186 if (handle->lgh_name)
1187 strcpy(handle->lgh_name, name);
1189 GOTO(out, rc = -ENOMEM);
1191 LASSERTF(open_param & LLOG_OPEN_NEW, "%#x\n", open_param);
1192 /* generate fid for new llog */
1193 rc = local_object_fid_generate(env, los, &lgi->lgi_fid);
1198 o = ls_locate(env, ls, &lgi->lgi_fid, NULL);
1200 GOTO(out_name, rc = PTR_ERR(o));
1203 /* No new llog is expected but doesn't exist */
1204 if (open_param != LLOG_OPEN_NEW && !dt_object_exists(o))
1205 GOTO(out_put, rc = -ENOENT);
1207 fid_to_logid(&lgi->lgi_fid, &handle->lgh_id);
1208 handle->lgh_obj = o;
1209 handle->private_data = los;
1210 LASSERT(handle->lgh_ctxt);
1215 lu_object_put(env, &o->do_lu);
1217 if (handle->lgh_name != NULL)
1218 OBD_FREE(handle->lgh_name, strlen(name) + 1);
1226 * Implementation of the llog_operations::lop_exist
1228 * This function checks that llog exists on storage.
1230 * \param[in] handle llog handle of the current llog
1232 * \retval true if llog object exists and is not just destroyed
1233 * \retval false if llog doesn't exist or just destroyed
1235 static int llog_osd_exist(struct llog_handle *handle)
1237 LASSERT(handle->lgh_obj);
1238 return (dt_object_exists(handle->lgh_obj) &&
1239 !lu_object_is_dying(handle->lgh_obj->do_lu.lo_header));
1243 * Get dir for regular fid log object
1245 * Get directory for regular fid log object, and these regular fid log
1246 * object will be inserted under this directory, to satisfy the FS
1247 * consistency check, e2fsck etc.
1249 * \param [in] env execution environment
1250 * \param [in] dto llog object
1252 * \retval pointer to the directory if it is found.
1253 * \retval ERR_PTR(negative errno) if it fails.
1255 struct dt_object *llog_osd_get_regular_fid_dir(const struct lu_env *env,
1256 struct dt_object *dto)
1258 struct llog_thread_info *lgi = llog_info(env);
1259 struct seq_server_site *ss = dto->do_lu.lo_dev->ld_site->ld_seq_site;
1260 struct lu_seq_range *range = &lgi->lgi_range;
1261 struct lu_fid *dir_fid = &lgi->lgi_fid;
1262 struct dt_object *dir;
1266 fld_range_set_any(range);
1267 LASSERT(ss != NULL);
1268 rc = ss->ss_server_fld->lsf_seq_lookup(env, ss->ss_server_fld,
1269 fid_seq(lu_object_fid(&dto->do_lu)), range);
1271 RETURN(ERR_PTR(rc));
1273 lu_update_log_dir_fid(dir_fid, range->lsr_index);
1274 dir = dt_locate(env, lu2dt_dev(dto->do_lu.lo_dev), dir_fid);
1278 if (!dt_try_as_dir(env, dir)) {
1279 lu_object_put(env, &dir->do_lu);
1280 RETURN(ERR_PTR(-ENOTDIR));
1287 * Add llog object with regular FID to name entry
1289 * Add llog object with regular FID to name space, and each llog
1290 * object on each MDT will be /update_log_dir/[seq:oid:ver],
1291 * so to satisfy the namespace consistency check, e2fsck etc.
1293 * \param [in] env execution environment
1294 * \param [in] dto llog object
1295 * \param [in] th thandle
1296 * \param [in] declare if it is declare or execution
1298 * \retval 0 if insertion succeeds.
1299 * \retval negative errno if insertion fails.
1302 llog_osd_regular_fid_add_name_entry(const struct lu_env *env,
1303 struct dt_object *dto,
1304 struct thandle *th, bool declare)
1306 struct llog_thread_info *lgi = llog_info(env);
1307 const struct lu_fid *fid = lu_object_fid(&dto->do_lu);
1308 struct dt_insert_rec *rec = &lgi->lgi_dt_rec;
1309 struct dt_object *dir;
1310 char *name = lgi->lgi_name;
1314 if (!fid_is_norm(fid))
1317 dir = llog_osd_get_regular_fid_dir(env, dto);
1319 RETURN(PTR_ERR(dir));
1322 rec->rec_type = S_IFREG;
1323 snprintf(name, sizeof(lgi->lgi_name), DFID, PFID(fid));
1324 dt_write_lock(env, dir, 0);
1326 rc = dt_declare_insert(env, dir, (struct dt_rec *)rec,
1327 (struct dt_key *)name, th);
1329 rc = dt_insert(env, dir, (struct dt_rec *)rec,
1330 (struct dt_key *)name, th, 1);
1332 dt_write_unlock(env, dir);
1334 lu_object_put(env, &dir->do_lu);
1340 * Implementation of the llog_operations::lop_declare_create
1342 * This function declares the llog create. It declares also name insert
1343 * into llog directory in case of named llog.
1345 * \param[in] env execution environment
1346 * \param[in] res llog handle of the current llog
1347 * \param[in] th current transaction handle
1349 * \retval 0 on successful create declaration
1350 * \retval negative value on error
1352 static int llog_osd_declare_create(const struct lu_env *env,
1353 struct llog_handle *res, struct thandle *th)
1355 struct llog_thread_info *lgi = llog_info(env);
1356 struct dt_insert_rec *rec = &lgi->lgi_dt_rec;
1357 struct local_oid_storage *los;
1358 struct dt_object *o;
1363 LASSERT(res->lgh_obj);
1366 /* object can be created by another thread */
1368 if (dt_object_exists(o))
1371 if (res->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1372 struct llog_thread_info *lgi = llog_info(env);
1374 lgi->lgi_attr.la_valid = LA_MODE | LA_SIZE;
1375 lgi->lgi_attr.la_size = 0;
1376 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1377 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1379 rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL,
1385 rc = llog_osd_regular_fid_add_name_entry(env, o, th, true);
1389 los = res->private_data;
1392 rc = llog_osd_declare_new_object(env, los, o, th);
1396 /* do not declare header initialization here as it's declared
1397 * in llog_osd_declare_write_rec() which is always called */
1399 if (res->lgh_name) {
1400 struct dt_object *llog_dir;
1402 llog_dir = llog_osd_dir_get(env, res->lgh_ctxt);
1403 if (IS_ERR(llog_dir))
1404 RETURN(PTR_ERR(llog_dir));
1405 logid_to_fid(&res->lgh_id, &lgi->lgi_fid);
1406 rec->rec_fid = &lgi->lgi_fid;
1407 rec->rec_type = S_IFREG;
1408 rc = dt_declare_insert(env, llog_dir,
1409 (struct dt_rec *)rec,
1410 (struct dt_key *)res->lgh_name, th);
1411 lu_object_put(env, &llog_dir->do_lu);
1413 CERROR("%s: can't declare named llog %s: rc = %d\n",
1414 o->do_lu.lo_dev->ld_obd->obd_name,
1421 * Implementation of the llog_operations::lop_create
1423 * This function creates the llog according with llog_handle::lgh_obj
1424 * and llog_handle::lgh_name.
1426 * \param[in] env execution environment
1427 * \param[in] res llog handle of the current llog
1428 * \param[in] th current transaction handle
1430 * \retval 0 on successful create
1431 * \retval negative value on error
1433 static int llog_osd_create(const struct lu_env *env, struct llog_handle *res,
1436 struct llog_thread_info *lgi = llog_info(env);
1437 struct dt_insert_rec *rec = &lgi->lgi_dt_rec;
1438 struct local_oid_storage *los;
1439 struct dt_object *o;
1448 /* llog can be already created */
1449 if (dt_object_exists(o))
1452 if (res->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1453 struct llog_thread_info *lgi = llog_info(env);
1455 lgi->lgi_attr.la_valid = LA_MODE | LA_SIZE | LA_TYPE;
1456 lgi->lgi_attr.la_size = 0;
1457 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1458 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1460 dt_write_lock(env, o, 0);
1461 rc = dt_create(env, o, &lgi->lgi_attr, NULL,
1463 dt_write_unlock(env, o);
1467 rc = llog_osd_regular_fid_add_name_entry(env, o, th, false);
1472 los = res->private_data;
1475 dt_write_lock(env, o, 0);
1476 if (!dt_object_exists(o))
1477 rc = llog_osd_create_new_object(env, los, o, th);
1481 dt_write_unlock(env, o);
1485 if (res->lgh_name) {
1486 struct dt_object *llog_dir;
1488 llog_dir = llog_osd_dir_get(env, res->lgh_ctxt);
1489 if (IS_ERR(llog_dir))
1490 RETURN(PTR_ERR(llog_dir));
1492 logid_to_fid(&res->lgh_id, &lgi->lgi_fid);
1493 rec->rec_fid = &lgi->lgi_fid;
1494 rec->rec_type = S_IFREG;
1495 dt_read_lock(env, llog_dir, 0);
1496 rc = dt_insert(env, llog_dir, (struct dt_rec *)rec,
1497 (struct dt_key *)res->lgh_name,
1499 dt_read_unlock(env, llog_dir);
1500 lu_object_put(env, &llog_dir->do_lu);
1502 CERROR("%s: can't create named llog %s: rc = %d\n",
1503 o->do_lu.lo_dev->ld_obd->obd_name,
1510 * Implementation of the llog_operations::lop_close
1512 * This function closes the llog. It just put llog object and referenced
1515 * \param[in] env execution environment
1516 * \param[in] handle llog handle of the current llog
1518 * \retval 0 on successful llog close
1519 * \retval negative value on error
1521 static int llog_osd_close(const struct lu_env *env, struct llog_handle *handle)
1523 struct local_oid_storage *los;
1528 LASSERT(handle->lgh_obj);
1530 if (handle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1531 /* Remove the object from the cache, otherwise it may
1532 * hold LOD being released during cleanup process */
1533 lu_object_put_nocache(env, &handle->lgh_obj->do_lu);
1534 LASSERT(handle->private_data == NULL);
1537 lu_object_put(env, &handle->lgh_obj->do_lu);
1539 los = handle->private_data;
1543 if (handle->lgh_name)
1544 OBD_FREE(handle->lgh_name, strlen(handle->lgh_name) + 1);
1550 * delete llog object name entry
1552 * Delete llog object (with regular FID) from name space (under
1555 * \param [in] env execution environment
1556 * \param [in] dto llog object
1557 * \param [in] th thandle
1558 * \param [in] declare if it is declare or execution
1560 * \retval 0 if deletion succeeds.
1561 * \retval negative errno if deletion fails.
1564 llog_osd_regular_fid_del_name_entry(const struct lu_env *env,
1565 struct dt_object *dto,
1566 struct thandle *th, bool declare)
1568 struct llog_thread_info *lgi = llog_info(env);
1569 const struct lu_fid *fid = lu_object_fid(&dto->do_lu);
1570 struct dt_object *dir;
1571 char *name = lgi->lgi_name;
1575 if (!fid_is_norm(fid))
1578 dir = llog_osd_get_regular_fid_dir(env, dto);
1580 RETURN(PTR_ERR(dir));
1582 snprintf(name, sizeof(lgi->lgi_name), DFID, PFID(fid));
1583 dt_write_lock(env, dir, 0);
1585 rc = dt_declare_delete(env, dir, (struct dt_key *)name,
1588 rc = dt_delete(env, dir, (struct dt_key *)name, th);
1590 dt_write_unlock(env, dir);
1592 lu_object_put(env, &dir->do_lu);
1597 * Implementation of the llog_operations::lop_declare_destroy
1599 * This function declare destroys the llog and deletes also entry in the
1600 * llog directory in case of named llog. Llog should be opened prior that.
1602 * \param[in] env execution environment
1603 * \param[in] loghandle llog handle of the current llog
1605 * \retval 0 on successful destroy
1606 * \retval negative value on error
1608 static int llog_osd_declare_destroy(const struct lu_env *env,
1609 struct llog_handle *loghandle,
1612 struct llog_ctxt *ctxt;
1613 struct dt_object *o, *llog_dir = NULL;
1618 ctxt = loghandle->lgh_ctxt;
1621 o = loghandle->lgh_obj;
1624 if (loghandle->lgh_name) {
1625 llog_dir = llog_osd_dir_get(env, ctxt);
1626 if (IS_ERR(llog_dir))
1627 RETURN(PTR_ERR(llog_dir));
1629 rc = dt_declare_delete(env, llog_dir,
1630 (struct dt_key *)loghandle->lgh_name,
1636 rc = dt_declare_ref_del(env, o, th);
1640 rc = dt_declare_destroy(env, o, th);
1644 if (loghandle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1645 rc = llog_osd_regular_fid_del_name_entry(env, o, th, true);
1651 if (!(IS_ERR_OR_NULL(llog_dir)))
1652 lu_object_put(env, &llog_dir->do_lu);
1659 * Implementation of the llog_operations::lop_destroy
1661 * This function destroys the llog and deletes also entry in the
1662 * llog directory in case of named llog. Llog should be opened prior that.
1663 * Destroy method is not part of external transaction and does everything
1666 * \param[in] env execution environment
1667 * \param[in] loghandle llog handle of the current llog
1669 * \retval 0 on successful destroy
1670 * \retval negative value on error
1672 static int llog_osd_destroy(const struct lu_env *env,
1673 struct llog_handle *loghandle, struct thandle *th)
1675 struct llog_ctxt *ctxt;
1676 struct dt_object *o, *llog_dir = NULL;
1681 ctxt = loghandle->lgh_ctxt;
1682 LASSERT(ctxt != NULL);
1684 o = loghandle->lgh_obj;
1687 dt_write_lock(env, o, 0);
1688 if (!dt_object_exists(o))
1689 GOTO(out_unlock, rc = 0);
1691 if (loghandle->lgh_name) {
1692 llog_dir = llog_osd_dir_get(env, ctxt);
1693 if (IS_ERR(llog_dir))
1694 GOTO(out_unlock, rc = PTR_ERR(llog_dir));
1696 dt_read_lock(env, llog_dir, 0);
1697 rc = dt_delete(env, llog_dir,
1698 (struct dt_key *)loghandle->lgh_name,
1700 dt_read_unlock(env, llog_dir);
1702 CERROR("%s: can't remove llog %s: rc = %d\n",
1703 o->do_lu.lo_dev->ld_obd->obd_name,
1704 loghandle->lgh_name, rc);
1705 GOTO(out_unlock, rc);
1709 dt_ref_del(env, o, th);
1710 rc = dt_destroy(env, o, th);
1712 GOTO(out_unlock, rc);
1714 if (loghandle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1715 rc = llog_osd_regular_fid_del_name_entry(env, o, th, false);
1717 GOTO(out_unlock, rc);
1721 dt_write_unlock(env, o);
1722 if (!(IS_ERR_OR_NULL(llog_dir)))
1723 lu_object_put(env, &llog_dir->do_lu);
1728 * Implementation of the llog_operations::lop_setup
1730 * This function setup the llog on local storage.
1732 * \param[in] env execution environment
1733 * \param[in] obd obd device the llog belongs to
1734 * \param[in] olg the llog group, it is always zero group now.
1735 * \param[in] ctxt_idx the llog index, it defines the purpose of this llog.
1736 * Every new llog type have to use own index.
1737 * \param[in] disk_obd the storage obd, where llog is stored.
1739 * \retval 0 on successful llog setup
1740 * \retval negative value on error
1742 static int llog_osd_setup(const struct lu_env *env, struct obd_device *obd,
1743 struct obd_llog_group *olg, int ctxt_idx,
1744 struct obd_device *disk_obd)
1746 struct llog_thread_info *lgi = llog_info(env);
1747 struct llog_ctxt *ctxt;
1752 LASSERT(olg->olg_ctxts[ctxt_idx]);
1754 ctxt = llog_ctxt_get(olg->olg_ctxts[ctxt_idx]);
1757 if (disk_obd == NULL)
1760 /* initialize data allowing to generate new fids,
1761 * literally we need a sequece */
1762 lgi->lgi_fid.f_seq = FID_SEQ_LLOG;
1763 lgi->lgi_fid.f_oid = 1;
1764 lgi->lgi_fid.f_ver = 0;
1765 rc = local_oid_storage_init(env, disk_obd->obd_lvfs_ctxt.dt,
1767 &ctxt->loc_los_nameless);
1771 lgi->lgi_fid.f_seq = FID_SEQ_LLOG_NAME;
1772 lgi->lgi_fid.f_oid = 1;
1773 lgi->lgi_fid.f_ver = 0;
1774 rc = local_oid_storage_init(env, disk_obd->obd_lvfs_ctxt.dt,
1776 &ctxt->loc_los_named);
1778 local_oid_storage_fini(env, ctxt->loc_los_nameless);
1779 ctxt->loc_los_nameless = NULL;
1785 llog_ctxt_put(ctxt);
1790 * Implementation of the llog_operations::lop_cleanup
1792 * This function cleanups the llog on local storage.
1794 * \param[in] env execution environment
1795 * \param[in] ctxt the llog context
1799 static int llog_osd_cleanup(const struct lu_env *env, struct llog_ctxt *ctxt)
1801 if (ctxt->loc_los_nameless != NULL) {
1802 local_oid_storage_fini(env, ctxt->loc_los_nameless);
1803 ctxt->loc_los_nameless = NULL;
1806 if (ctxt->loc_los_named != NULL) {
1807 local_oid_storage_fini(env, ctxt->loc_los_named);
1808 ctxt->loc_los_named = NULL;
1814 struct llog_operations llog_osd_ops = {
1815 .lop_next_block = llog_osd_next_block,
1816 .lop_prev_block = llog_osd_prev_block,
1817 .lop_read_header = llog_osd_read_header,
1818 .lop_declare_destroy = llog_osd_declare_destroy,
1819 .lop_destroy = llog_osd_destroy,
1820 .lop_setup = llog_osd_setup,
1821 .lop_cleanup = llog_osd_cleanup,
1822 .lop_open = llog_osd_open,
1823 .lop_exist = llog_osd_exist,
1824 .lop_declare_create = llog_osd_declare_create,
1825 .lop_create = llog_osd_create,
1826 .lop_declare_write_rec = llog_osd_declare_write_rec,
1827 .lop_write_rec = llog_osd_write_rec,
1828 .lop_close = llog_osd_close,
1830 EXPORT_SYMBOL(llog_osd_ops);
1832 struct llog_operations llog_common_cat_ops = {
1833 .lop_next_block = llog_osd_next_block,
1834 .lop_prev_block = llog_osd_prev_block,
1835 .lop_read_header = llog_osd_read_header,
1836 .lop_declare_destroy = llog_osd_declare_destroy,
1837 .lop_destroy = llog_osd_destroy,
1838 .lop_setup = llog_osd_setup,
1839 .lop_cleanup = llog_osd_cleanup,
1840 .lop_open = llog_osd_open,
1841 .lop_exist = llog_osd_exist,
1842 .lop_declare_create = llog_osd_declare_create,
1843 .lop_create = llog_osd_create,
1844 .lop_declare_write_rec = llog_osd_declare_write_rec,
1845 .lop_write_rec = llog_osd_write_rec,
1846 .lop_close = llog_osd_close,
1847 .lop_add = llog_cat_add_rec,
1848 .lop_declare_add = llog_cat_declare_add_rec,
1850 EXPORT_SYMBOL(llog_common_cat_ops);
1853 * Read the special file which contains the list of llog catalogs IDs
1855 * This function reads the CATALOGS file which contains the array of llog
1856 * catalogs IDs. The main purpose of this file is to store OSP llogs indexed
1857 * by OST/MDT number.
1859 * \param[in] env execution environment
1860 * \param[in] d corresponding storage device
1861 * \param[in] idx position to start from, usually OST/MDT index
1862 * \param[in] count how many catalog IDs to read
1863 * \param[out] idarray the buffer for the data. If it is NULL then
1864 * function returns just number of catalog IDs
1866 * \param[in] fid LLOG_CATALOGS_OID for CATALOG object
1868 * \retval 0 on successful read of catalog IDs
1869 * \retval negative value on error
1870 * \retval positive value which is number of records in
1871 * the file if \a idarray is NULL
1873 int llog_osd_get_cat_list(const struct lu_env *env, struct dt_device *d,
1874 int idx, int count, struct llog_catid *idarray,
1875 const struct lu_fid *fid)
1877 struct llog_thread_info *lgi = llog_info(env);
1878 struct dt_object *o = NULL;
1886 size = sizeof(*idarray) * count;
1887 lgi->lgi_off = idx * sizeof(*idarray);
1889 lgi->lgi_fid = *fid;
1890 o = dt_locate(env, d, &lgi->lgi_fid);
1894 if (!dt_object_exists(o)) {
1895 th = dt_trans_create(env, d);
1897 GOTO(out, rc = PTR_ERR(th));
1899 lgi->lgi_attr.la_valid = LA_MODE;
1900 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1901 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1903 th->th_wait_submit = 1;
1904 /* Make the llog object creation synchronization, so
1905 * it will be reliable to the reference, especially
1906 * for remote reference */
1909 rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL,
1912 GOTO(out_trans, rc);
1914 rc = dt_trans_start_local(env, d, th);
1916 GOTO(out_trans, rc);
1918 dt_write_lock(env, o, 0);
1919 if (!dt_object_exists(o))
1920 rc = dt_create(env, o, &lgi->lgi_attr, NULL,
1922 dt_write_unlock(env, o);
1924 dt_trans_stop(env, d, th);
1929 rc = dt_attr_get(env, o, &lgi->lgi_attr);
1933 if (!S_ISREG(lgi->lgi_attr.la_mode)) {
1934 CERROR("%s: CATALOGS is not a regular file!: mode = %o\n",
1935 o->do_lu.lo_dev->ld_obd->obd_name,
1936 lgi->lgi_attr.la_mode);
1937 GOTO(out, rc = -ENOENT);
1940 CDEBUG(D_CONFIG, "cat list: disk size=%d, read=%d\n",
1941 (int)lgi->lgi_attr.la_size, size);
1943 /* return just number of llogs */
1944 if (idarray == NULL) {
1945 rc = lgi->lgi_attr.la_size / sizeof(*idarray);
1949 /* read for new ost index or for empty file */
1950 memset(idarray, 0, size);
1951 if (lgi->lgi_attr.la_size <= lgi->lgi_off)
1953 if (lgi->lgi_attr.la_size < lgi->lgi_off + size)
1954 size = lgi->lgi_attr.la_size - lgi->lgi_off;
1956 lgi->lgi_buf.lb_buf = idarray;
1957 lgi->lgi_buf.lb_len = size;
1958 rc = dt_record_read(env, o, &lgi->lgi_buf, &lgi->lgi_off);
1959 /* -EFAULT means the llog is a sparse file. This is not an error
1960 * after arbitrary OST index is supported. */
1961 if (rc < 0 && rc != -EFAULT) {
1962 CERROR("%s: error reading CATALOGS: rc = %d\n",
1963 o->do_lu.lo_dev->ld_obd->obd_name, rc);
1969 lu_object_put(env, &o->do_lu);
1972 EXPORT_SYMBOL(llog_osd_get_cat_list);
1975 * Write the special file which contains the list of llog catalogs IDs
1977 * This function writes the CATALOG file which contains the array of llog
1978 * catalogs IDs. It is used mostly to store OSP llogs indexed by OST/MDT
1981 * \param[in] env execution environment
1982 * \param[in] d corresponding storage device
1983 * \param[in] idx position to start from, usually OST/MDT index
1984 * \param[in] count how many catalog IDs to write
1985 * \param[out] idarray the buffer with the data to write.
1986 * \param[in] fid LLOG_CATALOGS_OID for CATALOG object
1988 * \retval 0 on successful write of catalog IDs
1989 * \retval negative value on error
1991 int llog_osd_put_cat_list(const struct lu_env *env, struct dt_device *d,
1992 int idx, int count, struct llog_catid *idarray,
1993 const struct lu_fid *fid)
1995 struct llog_thread_info *lgi = llog_info(env);
1996 struct dt_object *o = NULL;
2005 size = sizeof(*idarray) * count;
2006 lgi->lgi_off = idx * sizeof(*idarray);
2007 lgi->lgi_fid = *fid;
2009 o = dt_locate(env, d, &lgi->lgi_fid);
2013 if (!dt_object_exists(o))
2014 GOTO(out, rc = -ENOENT);
2016 rc = dt_attr_get(env, o, &lgi->lgi_attr);
2020 if (!S_ISREG(lgi->lgi_attr.la_mode)) {
2021 CERROR("%s: CATALOGS is not a regular file!: mode = %o\n",
2022 o->do_lu.lo_dev->ld_obd->obd_name,
2023 lgi->lgi_attr.la_mode);
2024 GOTO(out, rc = -ENOENT);
2027 th = dt_trans_create(env, d);
2029 GOTO(out, rc = PTR_ERR(th));
2031 lgi->lgi_buf.lb_len = size;
2032 lgi->lgi_buf.lb_buf = idarray;
2033 rc = dt_declare_record_write(env, o, &lgi->lgi_buf, lgi->lgi_off, th);
2035 GOTO(out_trans, rc);
2037 /* For update log, this happens during initialization,
2038 * see lod_sub_prep_llog(), and we need make sure catlog
2039 * file ID is written to catlist file(committed) before
2040 * cross-MDT operation write update records to catlog FILE,
2041 * otherwise, during failover these update records might
2043 if (fid_is_update_log(fid))
2046 rc = dt_trans_start_local(env, d, th);
2048 GOTO(out_trans, rc);
2050 th->th_wait_submit = 1;
2052 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
2054 CDEBUG(D_INODE, "can't write CATALOGS at index %d: rc = %d\n",
2057 dt_trans_stop(env, d, th);
2059 lu_object_put(env, &o->do_lu);
2062 EXPORT_SYMBOL(llog_osd_put_cat_list);