4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2015, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 * lustre/obdclass/llog_osd.c
35 * Low level llog routines on top of OSD API
37 * This file provides set of methods for llog operations on top of
38 * dt_device. It contains all supported llog_operations interfaces and
39 * supplimental functions.
41 * Author: Alexey Zhuravlev <alexey.zhuravlev@intel.com>
42 * Author: Mikhail Pershin <mike.pershin@intel.com>
45 #define DEBUG_SUBSYSTEM S_LOG
47 #include <dt_object.h>
48 #include <llog_swab.h>
49 #include <lustre_fid.h>
51 #include <obd_class.h>
53 #include "llog_internal.h"
54 #include "local_storage.h"
57 * Implementation of the llog_operations::lop_declare_create
59 * This function is a wrapper over local_storage API function
60 * local_object_declare_create().
62 * \param[in] env execution environment
63 * \param[in] los local_storage for bottom storage device
64 * \param[in] o dt_object to create
65 * \param[in] th current transaction handle
67 * \retval 0 on successful declaration of the new object
68 * \retval negative error if declaration was failed
70 static int llog_osd_declare_new_object(const struct lu_env *env,
71 struct local_oid_storage *los,
75 struct llog_thread_info *lgi = llog_info(env);
77 lgi->lgi_attr.la_valid = LA_MODE;
78 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
79 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
81 return local_object_declare_create(env, los, o, &lgi->lgi_attr,
86 * Implementation of the llog_operations::lop_create
88 * This function is a wrapper over local_storage API function
89 * local_object_create().
91 * \param[in] env execution environment
92 * \param[in] los local_storage for bottom storage device
93 * \param[in] o dt_object to create
94 * \param[in] th current transaction handle
96 * \retval 0 on successful creation of the new object
97 * \retval negative error if creation was failed
99 static int llog_osd_create_new_object(const struct lu_env *env,
100 struct local_oid_storage *los,
104 struct llog_thread_info *lgi = llog_info(env);
106 lgi->lgi_attr.la_valid = LA_MODE;
107 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
108 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
110 return local_object_create(env, los, o, &lgi->lgi_attr,
115 * Implementation of the llog_operations::lop_exist
117 * This function checks that llog exists on storage.
119 * \param[in] handle llog handle of the current llog
121 * \retval true if llog object exists and is not just destroyed
122 * \retval false if llog doesn't exist or just destroyed
124 static int llog_osd_exist(struct llog_handle *handle)
126 LASSERT(handle->lgh_obj);
127 return dt_object_exists(handle->lgh_obj) &&
128 !lu_object_is_dying(handle->lgh_obj->do_lu.lo_header);
132 * Write a padding record to the llog
134 * This function writes a padding record to the end of llog. That may
135 * be needed if llog contains records of variable size, e.g. config logs
137 * The padding record just aligns llog to the llog chunk_size boundary if
138 * the current record doesn't fit in the remaining space.
140 * It allocates full length to avoid two separate writes for header and tail.
141 * Such 2-steps scheme needs extra protection and complex error handling.
143 * \param[in] env execution environment
144 * \param[in] o dt_object to create
145 * \param[in,out] off pointer to the padding start offset
146 * \param[in] len padding length
147 * \param[in] index index of the padding record in a llog
148 * \param[in] th current transaction handle
150 * \retval 0 on successful padding write
151 * \retval negative error if write failed
153 static int llog_osd_pad(const struct lu_env *env, struct dt_object *o,
154 loff_t *off, int len, int index, struct thandle *th)
156 struct llog_thread_info *lgi = llog_info(env);
157 struct llog_rec_hdr *rec;
158 struct llog_rec_tail *tail;
165 LASSERT(len >= LLOG_MIN_REC_SIZE && (len & 0x7) == 0);
172 rec->lrh_index = index;
173 rec->lrh_type = LLOG_PAD_MAGIC;
175 tail = rec_tail(rec);
177 tail->lrt_index = index;
179 lgi->lgi_buf.lb_buf = rec;
180 lgi->lgi_buf.lb_len = len;
181 rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
183 CERROR("%s: error writing padding record: rc = %d\n",
184 o->do_lu.lo_dev->ld_obd->obd_name, rc);
191 * Implementation of the llog_operations::lop_read_header
193 * This function reads the current llog header from the bottom storage
196 * \param[in] env execution environment
197 * \param[in] handle llog handle of the current llog
199 * \retval 0 on successful header read
200 * \retval negative error if read failed
202 static int llog_osd_read_header(const struct lu_env *env,
203 struct llog_handle *handle)
205 struct llog_rec_hdr *llh_hdr;
207 struct llog_thread_info *lgi;
208 enum llog_flag flags;
216 lgi = llog_info(env);
218 rc = dt_attr_get(env, o, &lgi->lgi_attr);
222 LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
224 if (lgi->lgi_attr.la_size == 0) {
225 CDEBUG(D_HA, "not reading header from 0-byte log\n");
229 flags = handle->lgh_hdr->llh_flags;
232 lgi->lgi_buf.lb_buf = handle->lgh_hdr;
233 lgi->lgi_buf.lb_len = handle->lgh_hdr_size;
234 rc = dt_read(env, o, &lgi->lgi_buf, &lgi->lgi_off);
235 llh_hdr = &handle->lgh_hdr->llh_hdr;
236 if (rc < sizeof(*llh_hdr) || rc < llh_hdr->lrh_len) {
237 CERROR("%s: error reading "DFID" log header size %d: rc = %d\n",
238 o->do_lu.lo_dev->ld_obd->obd_name,
239 PFID(lu_object_fid(&o->do_lu)), rc < 0 ? 0 : rc,
248 if (LLOG_REC_HDR_NEEDS_SWABBING(llh_hdr))
249 lustre_swab_llog_hdr(handle->lgh_hdr);
251 if (llh_hdr->lrh_type != LLOG_HDR_MAGIC) {
252 CERROR("%s: bad log %s "DFID" header magic: %#x "
253 "(expected %#x)\n", o->do_lu.lo_dev->ld_obd->obd_name,
254 handle->lgh_name ? handle->lgh_name : "",
255 PFID(lu_object_fid(&o->do_lu)),
256 llh_hdr->lrh_type, LLOG_HDR_MAGIC);
258 } else if (llh_hdr->lrh_len < LLOG_MIN_CHUNK_SIZE ||
259 llh_hdr->lrh_len > handle->lgh_hdr_size) {
260 CERROR("%s: incorrectly sized log %s "DFID" header: "
261 "%#x (expected at least %#x)\n"
262 "you may need to re-run lconf --write_conf.\n",
263 o->do_lu.lo_dev->ld_obd->obd_name,
264 handle->lgh_name ? handle->lgh_name : "",
265 PFID(lu_object_fid(&o->do_lu)),
266 llh_hdr->lrh_len, LLOG_MIN_CHUNK_SIZE);
268 } else if (LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_index >
269 LLOG_HDR_BITMAP_SIZE(handle->lgh_hdr) ||
270 LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_len !=
272 CERROR("%s: incorrectly sized log %s "DFID" tailer: "
274 o->do_lu.lo_dev->ld_obd->obd_name,
275 handle->lgh_name ? handle->lgh_name : "",
276 PFID(lu_object_fid(&o->do_lu)),
277 LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_len, -EIO);
281 handle->lgh_hdr->llh_flags |= (flags & LLOG_F_EXT_MASK);
282 handle->lgh_last_idx = LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_index;
283 handle->lgh_write_offset = lgi->lgi_attr.la_size;
289 * Implementation of the llog_operations::lop_declare_write
291 * This function declares the new record write.
293 * \param[in] env execution environment
294 * \param[in] loghandle llog handle of the current llog
295 * \param[in] rec llog record header. This is a real header of the full
296 * llog record to write. This is the beginning of buffer
297 * to write, the length of buffer is stored in
299 * \param[in] idx index of the llog record. If \a idx == -1 then this is
300 * append case, otherwise \a idx is the index of record
302 * \param[in] th current transaction handle
304 * \retval 0 on successful declaration
305 * \retval negative error if declaration failed
307 static int llog_osd_declare_write_rec(const struct lu_env *env,
308 struct llog_handle *loghandle,
309 struct llog_rec_hdr *rec,
310 int idx, struct thandle *th)
312 struct llog_thread_info *lgi = llog_info(env);
323 LASSERT(rec->lrh_len <= loghandle->lgh_ctxt->loc_chunk_size);
325 o = loghandle->lgh_obj;
328 chunk_size = loghandle->lgh_ctxt->loc_chunk_size;
329 lgi->lgi_buf.lb_len = chunk_size;
330 lgi->lgi_buf.lb_buf = NULL;
331 /* each time we update header */
332 rc = dt_declare_record_write(env, o, &lgi->lgi_buf, 0,
334 if (rc || idx == 0) /* if error or just header */
338 * the pad record can be inserted so take into account double
341 lgi->lgi_buf.lb_len = chunk_size * 2;
342 lgi->lgi_buf.lb_buf = NULL;
343 /* XXX: implement declared window or multi-chunks approach */
344 rc = dt_declare_record_write(env, o, &lgi->lgi_buf, -1, th);
350 * Implementation of the llog_operations::lop_write
352 * This function writes the new record in the llog or modify the existed one.
354 * \param[in] env execution environment
355 * \param[in] loghandle llog handle of the current llog
356 * \param[in] rec llog record header. This is a real header of
357 * the full llog record to write. This is
358 * the beginning of buffer to write, the length
359 * of buffer is stored in \a rec::lrh_len
360 * \param[out] reccookie pointer to the cookie to return back if needed.
361 * It is used for further cancel of this llog
363 * \param[in] idx index of the llog record. If \a idx == -1 then
364 * this is append case, otherwise \a idx is
365 * the index of record to modify
366 * \param[in] th current transaction handle
368 * \retval 0 on successful write && \a reccookie == NULL
369 * 1 on successful write && \a reccookie != NULL
370 * \retval negative error if write failed
372 static int llog_osd_write_rec(const struct lu_env *env,
373 struct llog_handle *loghandle,
374 struct llog_rec_hdr *rec,
375 struct llog_cookie *reccookie,
376 int idx, struct thandle *th)
378 struct llog_thread_info *lgi = llog_info(env);
379 struct llog_log_hdr *llh;
380 int reclen = rec->lrh_len;
382 struct llog_rec_tail *lrt;
387 __u64 orig_write_offset;
391 llh = loghandle->lgh_hdr;
393 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 "DOSTID": idx:%d, "
507 "len:%u offset %llu\n",
508 POSTID(&loghandle->lgh_id.lgl_oi), 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 orig_write_offset = loghandle->lgh_write_offset;
558 lgi->lgi_off = lgi->lgi_attr.la_size;
559 left = chunk_size - (lgi->lgi_off & (chunk_size - 1));
560 /* NOTE: padding is a record, but no bit is set */
561 if (left != 0 && left != reclen &&
562 left < (reclen + LLOG_MIN_REC_SIZE)) {
563 index = loghandle->lgh_last_idx + 1;
564 rc = llog_osd_pad(env, o, &lgi->lgi_off, left, index, th);
568 if (dt_object_remote(o))
569 loghandle->lgh_write_offset = lgi->lgi_off;
571 loghandle->lgh_last_idx++; /* for pad rec */
573 /* if it's the last idx in log file, then return -ENOSPC
574 * or wrap around if a catalog */
575 if (llog_is_full(loghandle) ||
576 unlikely(llh->llh_flags & LLOG_F_IS_CAT &&
577 OBD_FAIL_PRECHECK(OBD_FAIL_CAT_RECORDS) &&
578 loghandle->lgh_last_idx >= cfs_fail_val)) {
579 if (llh->llh_flags & LLOG_F_IS_CAT)
580 loghandle->lgh_last_idx = 0;
585 /* increment the last_idx along with llh_tail index, they should
586 * be equal for a llog lifetime */
587 loghandle->lgh_last_idx++;
588 index = loghandle->lgh_last_idx;
589 LLOG_HDR_TAIL(llh)->lrt_index = index;
591 * NB: the caller should make sure only 1 process access
592 * the lgh_last_idx, e.g. append should be exclusive.
593 * Otherwise it might hit the assert.
595 LASSERT(index < LLOG_HDR_BITMAP_SIZE(llh));
596 rec->lrh_index = index;
598 lrt->lrt_len = rec->lrh_len;
599 lrt->lrt_index = rec->lrh_index;
601 /* the lgh_hdr_mutex protects llog header data from concurrent
602 * update/cancel, the llh_count and llh_bitmap are protected */
603 mutex_lock(&loghandle->lgh_hdr_mutex);
604 if (ext2_set_bit(index, LLOG_HDR_BITMAP(llh))) {
605 CERROR("%s: index %u already set in log bitmap\n",
606 o->do_lu.lo_dev->ld_obd->obd_name, index);
607 mutex_unlock(&loghandle->lgh_hdr_mutex);
608 LBUG(); /* should never happen */
612 if (!(llh->llh_flags & LLOG_F_IS_FIXSIZE)) {
613 /* Update the minimum size of the llog record */
614 if (llh->llh_size == 0)
615 llh->llh_size = reclen;
616 else if (reclen < llh->llh_size)
617 llh->llh_size = reclen;
620 if (lgi->lgi_attr.la_size == 0) {
622 lgi->lgi_buf.lb_len = llh->llh_hdr.lrh_len;
623 lgi->lgi_buf.lb_buf = &llh->llh_hdr;
624 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
626 GOTO(out_unlock, rc);
628 __u32 *bitmap = LLOG_HDR_BITMAP(llh);
630 /* Note: If this is not initialization (size == 0), then do not
631 * write the whole header (8k bytes), only update header/tail
632 * and bits needs to be updated. Because this update might be
633 * part of cross-MDT operation, which needs to write these
634 * updates into the update log(32KB limit) and also pack inside
635 * the RPC (1MB limit), if we write 8K for each operation, which
636 * will cost a lot space, and keep us adding more updates to one
639 lgi->lgi_buf.lb_len = llh->llh_bitmap_offset;
640 lgi->lgi_buf.lb_buf = &llh->llh_hdr;
641 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
643 GOTO(out_unlock, rc);
645 lgi->lgi_off = llh->llh_bitmap_offset +
646 (index / (sizeof(*bitmap) * 8)) * sizeof(*bitmap);
647 lgi->lgi_buf.lb_len = sizeof(*bitmap);
648 lgi->lgi_buf.lb_buf = &bitmap[index/(sizeof(*bitmap)*8)];
649 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
651 GOTO(out_unlock, rc);
653 lgi->lgi_off = (unsigned long)LLOG_HDR_TAIL(llh) -
655 lgi->lgi_buf.lb_len = sizeof(llh->llh_tail);
656 lgi->lgi_buf.lb_buf = LLOG_HDR_TAIL(llh);
657 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
659 GOTO(out_unlock, rc);
663 /* unlock here for remote object */
664 mutex_unlock(&loghandle->lgh_hdr_mutex);
668 /* computed index can be used to determine offset for fixed-size
669 * records. This also allows to handle Catalog wrap around case */
670 if (llh->llh_flags & LLOG_F_IS_FIXSIZE) {
671 lgi->lgi_off = llh->llh_hdr.lrh_len + (index - 1) * reclen;
672 } else if (dt_object_remote(o)) {
673 lgi->lgi_off = max_t(__u64, loghandle->lgh_write_offset,
676 rc = dt_attr_get(env, o, &lgi->lgi_attr);
680 LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
681 lgi->lgi_off = max_t(__u64, lgi->lgi_attr.la_size,
685 lgi->lgi_buf.lb_len = reclen;
686 lgi->lgi_buf.lb_buf = rec;
687 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
691 if (dt_object_remote(o))
692 loghandle->lgh_write_offset = lgi->lgi_off;
694 CDEBUG(D_HA, "added record "DFID": idx: %u, %u off"LPU64"\n",
695 PFID(lu_object_fid(&o->do_lu)), index, rec->lrh_len,
697 if (reccookie != NULL) {
698 reccookie->lgc_lgl = loghandle->lgh_id;
699 reccookie->lgc_index = index;
700 if ((rec->lrh_type == MDS_UNLINK_REC) ||
701 (rec->lrh_type == MDS_SETATTR64_REC))
702 reccookie->lgc_subsys = LLOG_MDS_OST_ORIG_CTXT;
703 else if (rec->lrh_type == OST_SZ_REC)
704 reccookie->lgc_subsys = LLOG_SIZE_ORIG_CTXT;
706 reccookie->lgc_subsys = -1;
711 /* cleanup llog for error case */
712 mutex_lock(&loghandle->lgh_hdr_mutex);
713 ext2_clear_bit(index, LLOG_HDR_BITMAP(llh));
715 mutex_unlock(&loghandle->lgh_hdr_mutex);
717 /* restore llog last_idx */
718 if (dt_object_remote(o)) {
719 loghandle->lgh_last_idx = orig_last_idx;
720 loghandle->lgh_write_offset = orig_write_offset;
721 } else if (--loghandle->lgh_last_idx == 0 &&
722 (llh->llh_flags & LLOG_F_IS_CAT) && llh->llh_cat_idx != 0) {
723 /* catalog had just wrap-around case */
724 loghandle->lgh_last_idx = LLOG_HDR_BITMAP_SIZE(llh) - 1;
727 LLOG_HDR_TAIL(llh)->lrt_index = loghandle->lgh_last_idx;
733 * We can skip reading at least as many log blocks as the number of
734 * minimum sized log records we are skipping. If it turns out
735 * that we are not far enough along the log (because the
736 * actual records are larger than minimum size) we just skip
739 * Note: in llog_process_thread, it will use bitmap offset as
740 * the index to locate the record, which also includs some pad
741 * records, whose record size is very small, and it also does not
742 * consider pad record when recording minimum record size (otherwise
743 * min_record size might be too small), so in some rare cases,
744 * it might skip too much record for @goal, see llog_osd_next_block().
746 * When force_mini_rec is true, it means we have to use LLOG_MIN_REC_SIZE
747 * as the min record size to skip over, usually because in the previous
748 * try, it skip too much record, see loog_osd_next(prev)_block().
750 static inline void llog_skip_over(struct llog_handle *lgh, __u64 *off,
751 int curr, int goal, __u32 chunk_size,
754 struct llog_log_hdr *llh = lgh->lgh_hdr;
756 /* Goal should not bigger than the record count */
757 if (goal > lgh->lgh_last_idx)
758 goal = lgh->lgh_last_idx;
761 if (llh->llh_flags & LLOG_F_IS_FIXSIZE) {
762 *off = chunk_size + (goal - 1) * llh->llh_size;
764 __u64 min_rec_size = LLOG_MIN_REC_SIZE;
766 if (llh->llh_size > 0 && !force_mini_rec)
767 min_rec_size = llh->llh_size;
769 *off = *off + (goal - curr - 1) * min_rec_size;
772 /* always align with lower chunk boundary*/
773 *off &= ~(chunk_size - 1);
777 * Remove optional fields that the client doesn't expect.
778 * This is typically in order to ensure compatibility with older clients.
779 * It is assumed that since we exclusively remove fields, the block will be
780 * big enough to handle the remapped records. It is also assumed that records
781 * of a block have the same format (i.e.: the same features enabled).
783 * \param[in,out] hdr Header of the block of records to remap.
784 * \param[in,out] last_hdr Last header, don't read past this point.
785 * \param[in] flags Flags describing the fields to keep.
787 static void changelog_block_trim_ext(struct llog_rec_hdr *hdr,
788 struct llog_rec_hdr *last_hdr,
789 enum changelog_rec_flags flags)
791 if (hdr->lrh_type != CHANGELOG_REC)
795 struct changelog_rec *rec = (struct changelog_rec *)(hdr + 1);
797 changelog_remap_rec(rec, rec->cr_flags & flags);
798 hdr = llog_rec_hdr_next(hdr);
799 } while ((char *)hdr <= (char *)last_hdr);
803 * Implementation of the llog_operations::lop_next_block
805 * This function finds the the next llog block to return which contains
806 * record with required index. It is main part of llog processing.
808 * \param[in] env execution environment
809 * \param[in] loghandle llog handle of the current llog
810 * \param[in,out] cur_idx index preceeding cur_offset
811 * \param[in] next_idx target index to find
812 * \param[in,out] cur_offset furtherst point read in the file
813 * \param[in] buf pointer to data buffer to fill
814 * \param[in] len required len to read, it is
815 * usually llog chunk_size.
817 * \retval 0 on successful buffer read
818 * \retval negative value on error
820 static int llog_osd_next_block(const struct lu_env *env,
821 struct llog_handle *loghandle, int *cur_idx,
822 int next_idx, __u64 *cur_offset, void *buf,
825 struct llog_thread_info *lgi = llog_info(env);
827 struct dt_device *dt;
830 int last_idx = *cur_idx;
831 __u64 last_offset = *cur_offset;
832 bool force_mini_rec = false;
839 chunk_size = loghandle->lgh_hdr->llh_hdr.lrh_len;
840 if (len == 0 || len & (chunk_size - 1))
844 LASSERT(loghandle->lgh_ctxt);
846 o = loghandle->lgh_obj;
848 LASSERT(dt_object_exists(o));
849 dt = lu2dt_dev(o->do_lu.lo_dev);
852 rc = dt_attr_get(env, o, &lgi->lgi_attr);
856 CDEBUG(D_OTHER, "looking for log index %u (cur idx %u off"
857 LPU64"), size %llu\n", next_idx, *cur_idx,
858 *cur_offset, lgi->lgi_attr.la_size);
860 while (*cur_offset < lgi->lgi_attr.la_size) {
861 struct llog_rec_hdr *rec, *last_rec;
862 struct llog_rec_tail *tail;
864 llog_skip_over(loghandle, cur_offset, *cur_idx,
865 next_idx, chunk_size, force_mini_rec);
867 /* read up to next llog chunk_size block */
868 lgi->lgi_buf.lb_len = chunk_size -
869 (*cur_offset & (chunk_size - 1));
870 lgi->lgi_buf.lb_buf = buf;
872 rc = dt_read(env, o, &lgi->lgi_buf, cur_offset);
874 if (rc == -EBADR && !force_mini_rec)
877 CERROR("%s: can't read llog block from log "DFID
878 " offset "LPU64": rc = %d\n",
879 o->do_lu.lo_dev->ld_obd->obd_name,
880 PFID(lu_object_fid(&o->do_lu)), *cur_offset,
886 /* signal the end of the valid buffer to
888 memset(buf + rc, 0, len - rc);
891 if (rc == 0) { /* end of file, nothing to do */
897 if (rc < sizeof(*tail)) {
901 CERROR("%s: invalid llog block at log id "DOSTID"/%u "
903 o->do_lu.lo_dev->ld_obd->obd_name,
904 POSTID(&loghandle->lgh_id.lgl_oi),
905 loghandle->lgh_id.lgl_ogen, *cur_offset);
906 GOTO(out, rc = -EINVAL);
910 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
911 lustre_swab_llog_rec(rec);
913 tail = (struct llog_rec_tail *)((char *)buf + rc -
914 sizeof(struct llog_rec_tail));
915 /* get the last record in block */
916 last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
919 if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
920 lustre_swab_llog_rec(last_rec);
922 if (last_rec->lrh_index != tail->lrt_index) {
923 CERROR("%s: invalid llog tail at log id "DOSTID"/%u "
924 "offset "LPU64" last_rec idx %u tail idx %u\n",
925 o->do_lu.lo_dev->ld_obd->obd_name,
926 POSTID(&loghandle->lgh_id.lgl_oi),
927 loghandle->lgh_id.lgl_ogen, *cur_offset,
928 last_rec->lrh_index, tail->lrt_index);
929 GOTO(out, rc = -EINVAL);
932 *cur_idx = tail->lrt_index;
934 /* this shouldn't happen */
935 if (tail->lrt_index == 0) {
936 CERROR("%s: invalid llog tail at log id "DOSTID"/%u "
937 "offset "LPU64" bytes %d\n",
938 o->do_lu.lo_dev->ld_obd->obd_name,
939 POSTID(&loghandle->lgh_id.lgl_oi),
940 loghandle->lgh_id.lgl_ogen, *cur_offset, rc);
941 GOTO(out, rc = -EINVAL);
943 if (tail->lrt_index < next_idx) {
945 last_offset = *cur_offset;
949 /* sanity check that the start of the new buffer is no farther
950 * than the record that we wanted. This shouldn't happen. */
951 if (rec->lrh_index > next_idx) {
952 if (!force_mini_rec && next_idx > last_idx)
955 CERROR("%s: missed desired record? %u > %u\n",
956 o->do_lu.lo_dev->ld_obd->obd_name,
957 rec->lrh_index, next_idx);
958 GOTO(out, rc = -ENOENT);
961 /* Trim unsupported extensions for compat w/ older clients */
962 if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_JOBID))
963 changelog_block_trim_ext(rec, last_rec,
964 CLF_VERSION | CLF_RENAME);
969 /* Note: because there are some pad records in the
970 * llog, so llog_skip_over() might skip too much
971 * records, let's try skip again with minimum record */
972 force_mini_rec = true;
973 *cur_offset = last_offset;
976 GOTO(out, rc = -EIO);
982 * Implementation of the llog_operations::lop_prev_block
984 * This function finds the llog block to return which contains
985 * record with required index but in reverse order - from end of llog
987 * It is main part of reverse llog processing.
989 * \param[in] env execution environment
990 * \param[in] loghandle llog handle of the current llog
991 * \param[in] prev_idx target index to find
992 * \param[in] buf pointer to data buffer to fill
993 * \param[in] len required len to read, it is llog_chunk_size usually.
995 * \retval 0 on successful buffer read
996 * \retval negative value on error
998 static int llog_osd_prev_block(const struct lu_env *env,
999 struct llog_handle *loghandle,
1000 int prev_idx, void *buf, int len)
1002 struct llog_thread_info *lgi = llog_info(env);
1003 struct dt_object *o;
1004 struct dt_device *dt;
1011 chunk_size = loghandle->lgh_hdr->llh_hdr.lrh_len;
1012 if (len == 0 || len & (chunk_size - 1))
1015 CDEBUG(D_OTHER, "looking for log index %u\n", prev_idx);
1018 LASSERT(loghandle->lgh_ctxt);
1020 o = loghandle->lgh_obj;
1022 LASSERT(dt_object_exists(o));
1023 dt = lu2dt_dev(o->do_lu.lo_dev);
1026 /* Let's only use mini record size for previous block read
1028 cur_offset = chunk_size;
1029 llog_skip_over(loghandle, &cur_offset, 0, prev_idx,
1032 rc = dt_attr_get(env, o, &lgi->lgi_attr);
1036 while (cur_offset < lgi->lgi_attr.la_size) {
1037 struct llog_rec_hdr *rec, *last_rec;
1038 struct llog_rec_tail *tail;
1040 lgi->lgi_buf.lb_len = len;
1041 lgi->lgi_buf.lb_buf = buf;
1042 rc = dt_read(env, o, &lgi->lgi_buf, &cur_offset);
1044 CERROR("%s: can't read llog block from log "DFID
1045 " offset "LPU64": rc = %d\n",
1046 o->do_lu.lo_dev->ld_obd->obd_name,
1047 PFID(lu_object_fid(&o->do_lu)), cur_offset, rc);
1051 if (rc == 0) /* end of file, nothing to do */
1054 if (rc < sizeof(*tail)) {
1055 CERROR("%s: invalid llog block at log id "DOSTID"/%u "
1057 o->do_lu.lo_dev->ld_obd->obd_name,
1058 POSTID(&loghandle->lgh_id.lgl_oi),
1059 loghandle->lgh_id.lgl_ogen, cur_offset);
1060 GOTO(out, rc = -EINVAL);
1064 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
1065 lustre_swab_llog_rec(rec);
1067 tail = (struct llog_rec_tail *)((char *)buf + rc -
1068 sizeof(struct llog_rec_tail));
1069 /* get the last record in block */
1070 last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
1071 le32_to_cpu(tail->lrt_len));
1073 if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
1074 lustre_swab_llog_rec(last_rec);
1075 LASSERT(last_rec->lrh_index == tail->lrt_index);
1077 /* this shouldn't happen */
1078 if (tail->lrt_index == 0) {
1079 CERROR("%s: invalid llog tail at log id "DOSTID"/%u "
1081 o->do_lu.lo_dev->ld_obd->obd_name,
1082 POSTID(&loghandle->lgh_id.lgl_oi),
1083 loghandle->lgh_id.lgl_ogen, cur_offset);
1084 GOTO(out, rc = -EINVAL);
1086 if (tail->lrt_index < prev_idx)
1089 /* sanity check that the start of the new buffer is no farther
1090 * than the record that we wanted. This shouldn't happen. */
1091 if (rec->lrh_index > prev_idx) {
1092 CERROR("%s: missed desired record? %u > %u\n",
1093 o->do_lu.lo_dev->ld_obd->obd_name,
1094 rec->lrh_index, prev_idx);
1095 GOTO(out, rc = -ENOENT);
1098 /* Trim unsupported extensions for compat w/ older clients */
1099 if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_JOBID))
1100 changelog_block_trim_ext(rec, last_rec,
1101 CLF_VERSION | CLF_RENAME);
1105 GOTO(out, rc = -EIO);
1111 * This is helper function to get llog directory object. It is used by named
1112 * llog operations to find/insert/delete llog entry from llog directory.
1114 * \param[in] env execution environment
1115 * \param[in] ctxt llog context
1117 * \retval dt_object of llog directory
1118 * \retval ERR_PTR of negative value on error
1120 static struct dt_object *llog_osd_dir_get(const struct lu_env *env,
1121 struct llog_ctxt *ctxt)
1123 struct dt_device *dt;
1124 struct dt_thread_info *dti = dt_info(env);
1125 struct dt_object *dir;
1128 dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
1129 if (ctxt->loc_dir == NULL) {
1130 rc = dt_root_get(env, dt, &dti->dti_fid);
1133 dir = dt_locate(env, dt, &dti->dti_fid);
1135 if (!IS_ERR(dir) && !dt_try_as_dir(env, dir)) {
1136 lu_object_put(env, &dir->do_lu);
1137 return ERR_PTR(-ENOTDIR);
1140 lu_object_get(&ctxt->loc_dir->do_lu);
1141 dir = ctxt->loc_dir;
1148 * Implementation of the llog_operations::lop_open
1150 * This function opens the llog by its logid or by name, it may open also
1151 * non existent llog and assing then new id to it.
1152 * The llog_open/llog_close pair works similar to lu_object_find/put,
1153 * the object may not exist prior open. The result of open is just dt_object
1154 * in the llog header.
1156 * \param[in] env execution environment
1157 * \param[in] handle llog handle of the current llog
1158 * \param[in] logid logid of llog to open (nameless llog)
1159 * \param[in] name name of llog to open (named llog)
1160 * \param[in] open_param
1161 * LLOG_OPEN_NEW - new llog, may not exist
1162 * LLOG_OPEN_EXIST - old llog, must exist
1164 * \retval 0 on successful open, llog_handle::lgh_obj
1165 * contains the dt_object of the llog.
1166 * \retval negative value on error
1168 static int llog_osd_open(const struct lu_env *env, struct llog_handle *handle,
1169 struct llog_logid *logid, char *name,
1170 enum llog_open_param open_param)
1172 struct llog_thread_info *lgi = llog_info(env);
1173 struct llog_ctxt *ctxt = handle->lgh_ctxt;
1174 struct dt_object *o;
1175 struct dt_device *dt;
1176 struct ls_device *ls;
1177 struct local_oid_storage *los = NULL;
1184 LASSERT(ctxt->loc_exp);
1185 LASSERT(ctxt->loc_exp->exp_obd);
1186 dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
1188 if (ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1189 struct lu_object_conf conf = { 0 };
1190 if (logid != NULL) {
1191 logid_to_fid(logid, &lgi->lgi_fid);
1193 /* If logid == NULL, then it means the caller needs
1194 * to allocate new FID (llog_cat_declare_add_rec()). */
1195 rc = obd_fid_alloc(env, ctxt->loc_exp,
1196 &lgi->lgi_fid, NULL);
1200 conf.loc_flags = LOC_F_NEW;
1203 o = dt_locate_at(env, dt, &lgi->lgi_fid,
1204 dt->dd_lu_dev.ld_site->ls_top_dev, &conf);
1211 ls = ls_device_get(dt);
1213 RETURN(PTR_ERR(ls));
1215 mutex_lock(&ls->ls_los_mutex);
1216 los = dt_los_find(ls, name != NULL ? FID_SEQ_LLOG_NAME : FID_SEQ_LLOG);
1217 mutex_unlock(&ls->ls_los_mutex);
1219 ls_device_put(env, ls);
1223 if (logid != NULL) {
1224 logid_to_fid(logid, &lgi->lgi_fid);
1226 struct dt_object *llog_dir;
1228 llog_dir = llog_osd_dir_get(env, ctxt);
1229 if (IS_ERR(llog_dir))
1230 GOTO(out, rc = PTR_ERR(llog_dir));
1231 dt_read_lock(env, llog_dir, 0);
1232 rc = dt_lookup_dir(env, llog_dir, name, &lgi->lgi_fid);
1233 dt_read_unlock(env, llog_dir);
1234 lu_object_put(env, &llog_dir->do_lu);
1235 if (rc == -ENOENT && open_param == LLOG_OPEN_NEW) {
1236 /* generate fid for new llog */
1237 rc = local_object_fid_generate(env, los,
1242 OBD_ALLOC(handle->lgh_name, strlen(name) + 1);
1243 if (handle->lgh_name)
1244 strcpy(handle->lgh_name, name);
1246 GOTO(out, rc = -ENOMEM);
1248 LASSERTF(open_param & LLOG_OPEN_NEW, "%#x\n", open_param);
1249 /* generate fid for new llog */
1250 rc = local_object_fid_generate(env, los, &lgi->lgi_fid);
1255 o = ls_locate(env, ls, &lgi->lgi_fid, NULL);
1257 GOTO(out_name, rc = PTR_ERR(o));
1260 /* No new llog is expected but doesn't exist */
1261 if (open_param != LLOG_OPEN_NEW && !dt_object_exists(o))
1262 GOTO(out_put, rc = -ENOENT);
1264 fid_to_logid(&lgi->lgi_fid, &handle->lgh_id);
1265 handle->lgh_obj = o;
1266 handle->private_data = los;
1267 LASSERT(handle->lgh_ctxt);
1272 lu_object_put(env, &o->do_lu);
1274 if (handle->lgh_name != NULL)
1275 OBD_FREE(handle->lgh_name, strlen(name) + 1);
1283 * Get dir for regular fid log object
1285 * Get directory for regular fid log object, and these regular fid log
1286 * object will be inserted under this directory, to satisfy the FS
1287 * consistency check, e2fsck etc.
1289 * \param [in] env execution environment
1290 * \param [in] dto llog object
1292 * \retval pointer to the directory if it is found.
1293 * \retval ERR_PTR(negative errno) if it fails.
1295 struct dt_object *llog_osd_get_regular_fid_dir(const struct lu_env *env,
1296 struct dt_object *dto)
1298 struct llog_thread_info *lgi = llog_info(env);
1299 struct seq_server_site *ss = dto->do_lu.lo_dev->ld_site->ld_seq_site;
1300 struct lu_seq_range *range = &lgi->lgi_range;
1301 struct lu_fid *dir_fid = &lgi->lgi_fid;
1302 struct dt_object *dir;
1306 fld_range_set_any(range);
1307 LASSERT(ss != NULL);
1308 rc = ss->ss_server_fld->lsf_seq_lookup(env, ss->ss_server_fld,
1309 fid_seq(lu_object_fid(&dto->do_lu)), range);
1311 RETURN(ERR_PTR(rc));
1313 lu_update_log_dir_fid(dir_fid, range->lsr_index);
1314 dir = dt_locate(env, lu2dt_dev(dto->do_lu.lo_dev), dir_fid);
1318 if (!dt_try_as_dir(env, dir)) {
1319 lu_object_put(env, &dir->do_lu);
1320 RETURN(ERR_PTR(-ENOTDIR));
1327 * Add llog object with regular FID to name entry
1329 * Add llog object with regular FID to name space, and each llog
1330 * object on each MDT will be /update_log_dir/[seq:oid:ver],
1331 * so to satisfy the namespace consistency check, e2fsck etc.
1333 * \param [in] env execution environment
1334 * \param [in] dto llog object
1335 * \param [in] th thandle
1336 * \param [in] declare if it is declare or execution
1338 * \retval 0 if insertion succeeds.
1339 * \retval negative errno if insertion fails.
1342 llog_osd_regular_fid_add_name_entry(const struct lu_env *env,
1343 struct dt_object *dto,
1344 struct thandle *th, bool declare)
1346 struct llog_thread_info *lgi = llog_info(env);
1347 const struct lu_fid *fid = lu_object_fid(&dto->do_lu);
1348 struct dt_insert_rec *rec = &lgi->lgi_dt_rec;
1349 struct dt_object *dir;
1350 char *name = lgi->lgi_name;
1354 if (!fid_is_norm(fid))
1357 dir = llog_osd_get_regular_fid_dir(env, dto);
1359 RETURN(PTR_ERR(dir));
1362 rec->rec_type = S_IFREG;
1363 snprintf(name, sizeof(lgi->lgi_name), DFID, PFID(fid));
1364 dt_write_lock(env, dir, 0);
1366 rc = dt_declare_insert(env, dir, (struct dt_rec *)rec,
1367 (struct dt_key *)name, th);
1369 rc = dt_insert(env, dir, (struct dt_rec *)rec,
1370 (struct dt_key *)name, th, 1);
1372 dt_write_unlock(env, dir);
1374 lu_object_put(env, &dir->do_lu);
1380 * Implementation of the llog_operations::lop_declare_create
1382 * This function declares the llog create. It declares also name insert
1383 * into llog directory in case of named llog.
1385 * \param[in] env execution environment
1386 * \param[in] res llog handle of the current llog
1387 * \param[in] th current transaction handle
1389 * \retval 0 on successful create declaration
1390 * \retval negative value on error
1392 static int llog_osd_declare_create(const struct lu_env *env,
1393 struct llog_handle *res, struct thandle *th)
1395 struct llog_thread_info *lgi = llog_info(env);
1396 struct dt_insert_rec *rec = &lgi->lgi_dt_rec;
1397 struct local_oid_storage *los;
1398 struct dt_object *o;
1403 LASSERT(res->lgh_obj);
1406 /* object can be created by another thread */
1408 if (dt_object_exists(o))
1411 if (res->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1412 struct llog_thread_info *lgi = llog_info(env);
1414 lgi->lgi_attr.la_valid = LA_MODE | LA_SIZE;
1415 lgi->lgi_attr.la_size = 0;
1416 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1417 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1419 rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL,
1425 rc = llog_osd_regular_fid_add_name_entry(env, o, th, true);
1429 los = res->private_data;
1432 rc = llog_osd_declare_new_object(env, los, o, th);
1436 /* do not declare header initialization here as it's declared
1437 * in llog_osd_declare_write_rec() which is always called */
1439 if (res->lgh_name) {
1440 struct dt_object *llog_dir;
1442 llog_dir = llog_osd_dir_get(env, res->lgh_ctxt);
1443 if (IS_ERR(llog_dir))
1444 RETURN(PTR_ERR(llog_dir));
1445 logid_to_fid(&res->lgh_id, &lgi->lgi_fid);
1446 rec->rec_fid = &lgi->lgi_fid;
1447 rec->rec_type = S_IFREG;
1448 rc = dt_declare_insert(env, llog_dir,
1449 (struct dt_rec *)rec,
1450 (struct dt_key *)res->lgh_name, th);
1451 lu_object_put(env, &llog_dir->do_lu);
1453 CERROR("%s: can't declare named llog %s: rc = %d\n",
1454 o->do_lu.lo_dev->ld_obd->obd_name,
1461 * Implementation of the llog_operations::lop_create
1463 * This function creates the llog according with llog_handle::lgh_obj
1464 * and llog_handle::lgh_name.
1466 * \param[in] env execution environment
1467 * \param[in] res llog handle of the current llog
1468 * \param[in] th current transaction handle
1470 * \retval 0 on successful create
1471 * \retval negative value on error
1473 static int llog_osd_create(const struct lu_env *env, struct llog_handle *res,
1476 struct llog_thread_info *lgi = llog_info(env);
1477 struct dt_insert_rec *rec = &lgi->lgi_dt_rec;
1478 struct local_oid_storage *los;
1479 struct dt_object *o;
1488 /* llog can be already created */
1489 if (dt_object_exists(o))
1492 if (res->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1493 struct llog_thread_info *lgi = llog_info(env);
1495 lgi->lgi_attr.la_valid = LA_MODE | LA_SIZE | LA_TYPE;
1496 lgi->lgi_attr.la_size = 0;
1497 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1498 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1500 dt_write_lock(env, o, 0);
1501 rc = dt_create(env, o, &lgi->lgi_attr, NULL,
1503 dt_write_unlock(env, o);
1507 rc = llog_osd_regular_fid_add_name_entry(env, o, th, false);
1512 los = res->private_data;
1515 dt_write_lock(env, o, 0);
1516 if (!dt_object_exists(o))
1517 rc = llog_osd_create_new_object(env, los, o, th);
1521 dt_write_unlock(env, o);
1525 if (res->lgh_name) {
1526 struct dt_object *llog_dir;
1528 llog_dir = llog_osd_dir_get(env, res->lgh_ctxt);
1529 if (IS_ERR(llog_dir))
1530 RETURN(PTR_ERR(llog_dir));
1532 logid_to_fid(&res->lgh_id, &lgi->lgi_fid);
1533 rec->rec_fid = &lgi->lgi_fid;
1534 rec->rec_type = S_IFREG;
1535 dt_read_lock(env, llog_dir, 0);
1536 rc = dt_insert(env, llog_dir, (struct dt_rec *)rec,
1537 (struct dt_key *)res->lgh_name,
1539 dt_read_unlock(env, llog_dir);
1540 lu_object_put(env, &llog_dir->do_lu);
1542 CERROR("%s: can't create named llog %s: rc = %d\n",
1543 o->do_lu.lo_dev->ld_obd->obd_name,
1550 * Implementation of the llog_operations::lop_close
1552 * This function closes the llog. It just put llog object and referenced
1555 * \param[in] env execution environment
1556 * \param[in] handle llog handle of the current llog
1558 * \retval 0 on successful llog close
1559 * \retval negative value on error
1561 static int llog_osd_close(const struct lu_env *env, struct llog_handle *handle)
1563 struct local_oid_storage *los;
1568 LASSERT(handle->lgh_obj);
1570 if (handle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1571 /* Remove the object from the cache, otherwise it may
1572 * hold LOD being released during cleanup process */
1573 lu_object_put_nocache(env, &handle->lgh_obj->do_lu);
1574 LASSERT(handle->private_data == NULL);
1577 lu_object_put(env, &handle->lgh_obj->do_lu);
1579 los = handle->private_data;
1583 if (handle->lgh_name)
1584 OBD_FREE(handle->lgh_name, strlen(handle->lgh_name) + 1);
1590 * delete llog object name entry
1592 * Delete llog object (with regular FID) from name space (under
1595 * \param [in] env execution environment
1596 * \param [in] dto llog object
1597 * \param [in] th thandle
1598 * \param [in] declare if it is declare or execution
1600 * \retval 0 if deletion succeeds.
1601 * \retval negative errno if deletion fails.
1604 llog_osd_regular_fid_del_name_entry(const struct lu_env *env,
1605 struct dt_object *dto,
1606 struct thandle *th, bool declare)
1608 struct llog_thread_info *lgi = llog_info(env);
1609 const struct lu_fid *fid = lu_object_fid(&dto->do_lu);
1610 struct dt_object *dir;
1611 char *name = lgi->lgi_name;
1615 if (!fid_is_norm(fid))
1618 dir = llog_osd_get_regular_fid_dir(env, dto);
1620 RETURN(PTR_ERR(dir));
1622 snprintf(name, sizeof(lgi->lgi_name), DFID, PFID(fid));
1623 dt_write_lock(env, dir, 0);
1625 rc = dt_declare_delete(env, dir, (struct dt_key *)name,
1628 rc = dt_delete(env, dir, (struct dt_key *)name, th);
1630 dt_write_unlock(env, dir);
1632 lu_object_put(env, &dir->do_lu);
1637 * Implementation of the llog_operations::lop_declare_destroy
1639 * This function declare destroys the llog and deletes also entry in the
1640 * llog directory in case of named llog. Llog should be opened prior that.
1642 * \param[in] env execution environment
1643 * \param[in] loghandle llog handle of the current llog
1645 * \retval 0 on successful destroy
1646 * \retval negative value on error
1648 static int llog_osd_declare_destroy(const struct lu_env *env,
1649 struct llog_handle *loghandle,
1652 struct llog_ctxt *ctxt;
1653 struct dt_object *o, *llog_dir = NULL;
1658 ctxt = loghandle->lgh_ctxt;
1661 o = loghandle->lgh_obj;
1664 if (loghandle->lgh_name) {
1665 llog_dir = llog_osd_dir_get(env, ctxt);
1666 if (IS_ERR(llog_dir))
1667 RETURN(PTR_ERR(llog_dir));
1669 rc = dt_declare_delete(env, llog_dir,
1670 (struct dt_key *)loghandle->lgh_name,
1676 rc = dt_declare_ref_del(env, o, th);
1680 rc = dt_declare_destroy(env, o, th);
1684 if (loghandle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1685 rc = llog_osd_regular_fid_del_name_entry(env, o, th, true);
1691 if (!(IS_ERR_OR_NULL(llog_dir)))
1692 lu_object_put(env, &llog_dir->do_lu);
1699 * Implementation of the llog_operations::lop_destroy
1701 * This function destroys the llog and deletes also entry in the
1702 * llog directory in case of named llog. Llog should be opened prior that.
1703 * Destroy method is not part of external transaction and does everything
1706 * \param[in] env execution environment
1707 * \param[in] loghandle llog handle of the current llog
1709 * \retval 0 on successful destroy
1710 * \retval negative value on error
1712 static int llog_osd_destroy(const struct lu_env *env,
1713 struct llog_handle *loghandle, struct thandle *th)
1715 struct llog_ctxt *ctxt;
1716 struct dt_object *o, *llog_dir = NULL;
1721 ctxt = loghandle->lgh_ctxt;
1722 LASSERT(ctxt != NULL);
1724 o = loghandle->lgh_obj;
1727 dt_write_lock(env, o, 0);
1728 if (!dt_object_exists(o))
1729 GOTO(out_unlock, rc = 0);
1731 if (loghandle->lgh_name) {
1732 llog_dir = llog_osd_dir_get(env, ctxt);
1733 if (IS_ERR(llog_dir))
1734 GOTO(out_unlock, rc = PTR_ERR(llog_dir));
1736 dt_read_lock(env, llog_dir, 0);
1737 rc = dt_delete(env, llog_dir,
1738 (struct dt_key *)loghandle->lgh_name,
1740 dt_read_unlock(env, llog_dir);
1742 CERROR("%s: can't remove llog %s: rc = %d\n",
1743 o->do_lu.lo_dev->ld_obd->obd_name,
1744 loghandle->lgh_name, rc);
1745 GOTO(out_unlock, rc);
1749 dt_ref_del(env, o, th);
1750 rc = dt_destroy(env, o, th);
1752 GOTO(out_unlock, rc);
1754 if (loghandle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1755 rc = llog_osd_regular_fid_del_name_entry(env, o, th, false);
1757 GOTO(out_unlock, rc);
1761 dt_write_unlock(env, o);
1762 if (!(IS_ERR_OR_NULL(llog_dir)))
1763 lu_object_put(env, &llog_dir->do_lu);
1768 * Implementation of the llog_operations::lop_setup
1770 * This function setup the llog on local storage.
1772 * \param[in] env execution environment
1773 * \param[in] obd obd device the llog belongs to
1774 * \param[in] olg the llog group, it is always zero group now.
1775 * \param[in] ctxt_idx the llog index, it defines the purpose of this llog.
1776 * Every new llog type have to use own index.
1777 * \param[in] disk_obd the storage obd, where llog is stored.
1779 * \retval 0 on successful llog setup
1780 * \retval negative value on error
1782 static int llog_osd_setup(const struct lu_env *env, struct obd_device *obd,
1783 struct obd_llog_group *olg, int ctxt_idx,
1784 struct obd_device *disk_obd)
1786 struct llog_thread_info *lgi = llog_info(env);
1787 struct llog_ctxt *ctxt;
1792 LASSERT(olg->olg_ctxts[ctxt_idx]);
1794 ctxt = llog_ctxt_get(olg->olg_ctxts[ctxt_idx]);
1797 if (disk_obd == NULL)
1800 /* initialize data allowing to generate new fids,
1801 * literally we need a sequece */
1802 lgi->lgi_fid.f_seq = FID_SEQ_LLOG;
1803 lgi->lgi_fid.f_oid = 1;
1804 lgi->lgi_fid.f_ver = 0;
1805 rc = local_oid_storage_init(env, disk_obd->obd_lvfs_ctxt.dt,
1807 &ctxt->loc_los_nameless);
1811 lgi->lgi_fid.f_seq = FID_SEQ_LLOG_NAME;
1812 lgi->lgi_fid.f_oid = 1;
1813 lgi->lgi_fid.f_ver = 0;
1814 rc = local_oid_storage_init(env, disk_obd->obd_lvfs_ctxt.dt,
1816 &ctxt->loc_los_named);
1818 local_oid_storage_fini(env, ctxt->loc_los_nameless);
1819 ctxt->loc_los_nameless = NULL;
1825 llog_ctxt_put(ctxt);
1830 * Implementation of the llog_operations::lop_cleanup
1832 * This function cleanups the llog on local storage.
1834 * \param[in] env execution environment
1835 * \param[in] ctxt the llog context
1839 static int llog_osd_cleanup(const struct lu_env *env, struct llog_ctxt *ctxt)
1841 if (ctxt->loc_los_nameless != NULL) {
1842 local_oid_storage_fini(env, ctxt->loc_los_nameless);
1843 ctxt->loc_los_nameless = NULL;
1846 if (ctxt->loc_los_named != NULL) {
1847 local_oid_storage_fini(env, ctxt->loc_los_named);
1848 ctxt->loc_los_named = NULL;
1854 struct llog_operations llog_osd_ops = {
1855 .lop_next_block = llog_osd_next_block,
1856 .lop_prev_block = llog_osd_prev_block,
1857 .lop_read_header = llog_osd_read_header,
1858 .lop_declare_destroy = llog_osd_declare_destroy,
1859 .lop_destroy = llog_osd_destroy,
1860 .lop_setup = llog_osd_setup,
1861 .lop_cleanup = llog_osd_cleanup,
1862 .lop_open = llog_osd_open,
1863 .lop_exist = llog_osd_exist,
1864 .lop_declare_create = llog_osd_declare_create,
1865 .lop_create = llog_osd_create,
1866 .lop_declare_write_rec = llog_osd_declare_write_rec,
1867 .lop_write_rec = llog_osd_write_rec,
1868 .lop_close = llog_osd_close,
1870 EXPORT_SYMBOL(llog_osd_ops);
1872 struct llog_operations llog_common_cat_ops = {
1873 .lop_next_block = llog_osd_next_block,
1874 .lop_prev_block = llog_osd_prev_block,
1875 .lop_read_header = llog_osd_read_header,
1876 .lop_declare_destroy = llog_osd_declare_destroy,
1877 .lop_destroy = llog_osd_destroy,
1878 .lop_setup = llog_osd_setup,
1879 .lop_cleanup = llog_osd_cleanup,
1880 .lop_open = llog_osd_open,
1881 .lop_exist = llog_osd_exist,
1882 .lop_declare_create = llog_osd_declare_create,
1883 .lop_create = llog_osd_create,
1884 .lop_declare_write_rec = llog_osd_declare_write_rec,
1885 .lop_write_rec = llog_osd_write_rec,
1886 .lop_close = llog_osd_close,
1887 .lop_add = llog_cat_add_rec,
1888 .lop_declare_add = llog_cat_declare_add_rec,
1890 EXPORT_SYMBOL(llog_common_cat_ops);
1893 * Read the special file which contains the list of llog catalogs IDs
1895 * This function reads the CATALOGS file which contains the array of llog
1896 * catalogs IDs. The main purpose of this file is to store OSP llogs indexed
1897 * by OST/MDT number.
1899 * \param[in] env execution environment
1900 * \param[in] d corresponding storage device
1901 * \param[in] idx position to start from, usually OST/MDT index
1902 * \param[in] count how many catalog IDs to read
1903 * \param[out] idarray the buffer for the data. If it is NULL then
1904 * function returns just number of catalog IDs
1906 * \param[in] fid LLOG_CATALOGS_OID for CATALOG object
1908 * \retval 0 on successful read of catalog IDs
1909 * \retval negative value on error
1910 * \retval positive value which is number of records in
1911 * the file if \a idarray is NULL
1913 int llog_osd_get_cat_list(const struct lu_env *env, struct dt_device *d,
1914 int idx, int count, struct llog_catid *idarray,
1915 const struct lu_fid *fid)
1917 struct llog_thread_info *lgi = llog_info(env);
1918 struct dt_object *o = NULL;
1926 size = sizeof(*idarray) * count;
1927 lgi->lgi_off = idx * sizeof(*idarray);
1929 lgi->lgi_fid = *fid;
1930 o = dt_locate(env, d, &lgi->lgi_fid);
1934 if (!dt_object_exists(o)) {
1935 th = dt_trans_create(env, d);
1937 GOTO(out, rc = PTR_ERR(th));
1939 lgi->lgi_attr.la_valid = LA_MODE;
1940 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1941 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1943 th->th_wait_submit = 1;
1944 /* Make the llog object creation synchronization, so
1945 * it will be reliable to the reference, especially
1946 * for remote reference */
1949 rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL,
1952 GOTO(out_trans, rc);
1954 rc = dt_trans_start_local(env, d, th);
1956 GOTO(out_trans, rc);
1958 dt_write_lock(env, o, 0);
1959 if (!dt_object_exists(o))
1960 rc = dt_create(env, o, &lgi->lgi_attr, NULL,
1962 dt_write_unlock(env, o);
1964 dt_trans_stop(env, d, th);
1969 rc = dt_attr_get(env, o, &lgi->lgi_attr);
1973 if (!S_ISREG(lgi->lgi_attr.la_mode)) {
1974 CERROR("%s: CATALOGS is not a regular file!: mode = %o\n",
1975 o->do_lu.lo_dev->ld_obd->obd_name,
1976 lgi->lgi_attr.la_mode);
1977 GOTO(out, rc = -ENOENT);
1980 CDEBUG(D_CONFIG, "cat list: disk size=%d, read=%d\n",
1981 (int)lgi->lgi_attr.la_size, size);
1983 /* return just number of llogs */
1984 if (idarray == NULL) {
1985 rc = lgi->lgi_attr.la_size / sizeof(*idarray);
1989 /* read for new ost index or for empty file */
1990 memset(idarray, 0, size);
1991 if (lgi->lgi_attr.la_size <= lgi->lgi_off)
1993 if (lgi->lgi_attr.la_size < lgi->lgi_off + size)
1994 size = lgi->lgi_attr.la_size - lgi->lgi_off;
1996 lgi->lgi_buf.lb_buf = idarray;
1997 lgi->lgi_buf.lb_len = size;
1998 rc = dt_record_read(env, o, &lgi->lgi_buf, &lgi->lgi_off);
1999 /* -EFAULT means the llog is a sparse file. This is not an error
2000 * after arbitrary OST index is supported. */
2001 if (rc < 0 && rc != -EFAULT) {
2002 CERROR("%s: error reading CATALOGS: rc = %d\n",
2003 o->do_lu.lo_dev->ld_obd->obd_name, rc);
2009 lu_object_put(env, &o->do_lu);
2012 EXPORT_SYMBOL(llog_osd_get_cat_list);
2015 * Write the special file which contains the list of llog catalogs IDs
2017 * This function writes the CATALOG file which contains the array of llog
2018 * catalogs IDs. It is used mostly to store OSP llogs indexed by OST/MDT
2021 * \param[in] env execution environment
2022 * \param[in] d corresponding storage device
2023 * \param[in] idx position to start from, usually OST/MDT index
2024 * \param[in] count how many catalog IDs to write
2025 * \param[out] idarray the buffer with the data to write.
2026 * \param[in] fid LLOG_CATALOGS_OID for CATALOG object
2028 * \retval 0 on successful write of catalog IDs
2029 * \retval negative value on error
2031 int llog_osd_put_cat_list(const struct lu_env *env, struct dt_device *d,
2032 int idx, int count, struct llog_catid *idarray,
2033 const struct lu_fid *fid)
2035 struct llog_thread_info *lgi = llog_info(env);
2036 struct dt_object *o = NULL;
2045 size = sizeof(*idarray) * count;
2046 lgi->lgi_off = idx * sizeof(*idarray);
2047 lgi->lgi_fid = *fid;
2049 o = dt_locate(env, d, &lgi->lgi_fid);
2053 if (!dt_object_exists(o))
2054 GOTO(out, rc = -ENOENT);
2056 rc = dt_attr_get(env, o, &lgi->lgi_attr);
2060 if (!S_ISREG(lgi->lgi_attr.la_mode)) {
2061 CERROR("%s: CATALOGS is not a regular file!: mode = %o\n",
2062 o->do_lu.lo_dev->ld_obd->obd_name,
2063 lgi->lgi_attr.la_mode);
2064 GOTO(out, rc = -ENOENT);
2067 th = dt_trans_create(env, d);
2069 GOTO(out, rc = PTR_ERR(th));
2071 lgi->lgi_buf.lb_len = size;
2072 lgi->lgi_buf.lb_buf = idarray;
2073 rc = dt_declare_record_write(env, o, &lgi->lgi_buf, lgi->lgi_off, th);
2075 GOTO(out_trans, rc);
2077 /* For update log, this happens during initialization,
2078 * see lod_sub_prep_llog(), and we need make sure catlog
2079 * file ID is written to catlist file(committed) before
2080 * cross-MDT operation write update records to catlog FILE,
2081 * otherwise, during failover these update records might
2083 if (fid_is_update_log(fid))
2086 rc = dt_trans_start_local(env, d, th);
2088 GOTO(out_trans, rc);
2090 th->th_wait_submit = 1;
2092 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
2094 CDEBUG(D_INODE, "can't write CATALOGS at index %d: rc = %d\n",
2097 dt_trans_stop(env, d, th);
2099 lu_object_put(env, &o->do_lu);
2102 EXPORT_SYMBOL(llog_osd_put_cat_list);