4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 * lustre/obdclass/llog_osd.c
35 * Low level llog routines on top of OSD API
37 * This file provides set of methods for llog operations on top of
38 * dt_device. It contains all supported llog_operations interfaces and
39 * supplimental functions.
41 * Author: Alexey Zhuravlev <alexey.zhuravlev@intel.com>
42 * Author: Mikhail Pershin <mike.pershin@intel.com>
45 #define DEBUG_SUBSYSTEM S_LOG
47 #include <linux/delay.h>
49 #include <dt_object.h>
50 #include <llog_swab.h>
51 #include <lustre_fid.h>
53 #include <obd_class.h>
55 #include "llog_internal.h"
56 #include "local_storage.h"
59 * Implementation of the llog_operations::lop_declare_create
61 * This function is a wrapper over local_storage API function
62 * local_object_declare_create().
64 * \param[in] env execution environment
65 * \param[in] los local_storage for bottom storage device
66 * \param[in] o dt_object to create
67 * \param[in] th current transaction handle
69 * \retval 0 on successful declaration of the new object
70 * \retval negative error if declaration was failed
72 static int llog_osd_declare_new_object(const struct lu_env *env,
73 struct local_oid_storage *los,
77 struct llog_thread_info *lgi = llog_info(env);
79 lgi->lgi_attr.la_valid = LA_MODE;
80 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
81 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
83 return local_object_declare_create(env, los, o, &lgi->lgi_attr,
88 * Implementation of the llog_operations::lop_create
90 * This function is a wrapper over local_storage API function
91 * local_object_create().
93 * \param[in] env execution environment
94 * \param[in] los local_storage for bottom storage device
95 * \param[in] o dt_object to create
96 * \param[in] th current transaction handle
98 * \retval 0 on successful creation of the new object
99 * \retval negative error if creation was failed
101 static int llog_osd_create_new_object(const struct lu_env *env,
102 struct local_oid_storage *los,
106 struct llog_thread_info *lgi = llog_info(env);
108 lgi->lgi_attr.la_valid = LA_MODE;
109 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
110 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
112 return local_object_create(env, los, o, &lgi->lgi_attr,
117 * Implementation of the llog_operations::lop_exist
119 * This function checks that llog exists on storage.
121 * \param[in] handle llog handle of the current llog
123 * \retval true if llog object exists and is not just destroyed
124 * \retval false if llog doesn't exist or just destroyed
126 static int llog_osd_exist(struct llog_handle *handle)
128 LASSERT(handle->lgh_obj);
129 return dt_object_exists(handle->lgh_obj) && !handle->lgh_destroyed;
132 static void *rec_tail(struct llog_rec_hdr *rec)
134 return (void *)((char *)rec + rec->lrh_len -
135 sizeof(struct llog_rec_tail));
139 * Write a padding record to the llog
141 * This function writes a padding record to the end of llog. That may
142 * be needed if llog contains records of variable size, e.g. config logs
144 * The padding record just aligns llog to the llog chunk_size boundary if
145 * the current record doesn't fit in the remaining space.
147 * It allocates full length to avoid two separate writes for header and tail.
148 * Such 2-steps scheme needs extra protection and complex error handling.
150 * \param[in] env execution environment
151 * \param[in] o dt_object to create
152 * \param[in,out] off pointer to the padding start offset
153 * \param[in] len padding length
154 * \param[in] index index of the padding record in a llog
155 * \param[in] th current transaction handle
157 * \retval 0 on successful padding write
158 * \retval negative error if write failed
160 static int llog_osd_pad(const struct lu_env *env, struct dt_object *o,
161 loff_t *off, int len, int index, struct thandle *th)
163 struct llog_thread_info *lgi = llog_info(env);
164 struct llog_rec_hdr *rec;
165 struct llog_rec_tail *tail;
172 LASSERT(len >= LLOG_MIN_REC_SIZE && (len & 0x7) == 0);
179 rec->lrh_index = index;
180 rec->lrh_type = LLOG_PAD_MAGIC;
182 tail = rec_tail(rec);
184 tail->lrt_index = index;
186 lgi->lgi_buf.lb_buf = rec;
187 lgi->lgi_buf.lb_len = len;
188 rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
190 CERROR("%s: error writing padding record: rc = %d\n",
191 o->do_lu.lo_dev->ld_obd->obd_name, rc);
198 * Implementation of the llog_operations::lop_read_header
200 * This function reads the current llog header from the bottom storage
203 * \param[in] env execution environment
204 * \param[in] handle llog handle of the current llog
206 * \retval 0 on successful header read
207 * \retval negative error if read failed
209 static int llog_osd_read_header(const struct lu_env *env,
210 struct llog_handle *handle)
212 struct llog_rec_hdr *llh_hdr;
214 struct llog_thread_info *lgi;
215 enum llog_flag flags;
223 lgi = llog_info(env);
225 rc = dt_attr_get(env, o, &lgi->lgi_attr);
229 LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
231 if (lgi->lgi_attr.la_size == 0) {
232 CDEBUG(D_HA, "not reading header from 0-byte log\n");
236 flags = handle->lgh_hdr->llh_flags;
239 lgi->lgi_buf.lb_buf = handle->lgh_hdr;
240 lgi->lgi_buf.lb_len = handle->lgh_hdr_size;
241 rc = dt_read(env, o, &lgi->lgi_buf, &lgi->lgi_off);
242 llh_hdr = &handle->lgh_hdr->llh_hdr;
243 if (rc < sizeof(*llh_hdr) || rc < llh_hdr->lrh_len) {
244 CERROR("%s: error reading "DFID" log header size %d: rc = %d\n",
245 o->do_lu.lo_dev->ld_obd->obd_name,
246 PFID(lu_object_fid(&o->do_lu)), rc < 0 ? 0 : rc,
255 if (LLOG_REC_HDR_NEEDS_SWABBING(llh_hdr))
256 lustre_swab_llog_hdr(handle->lgh_hdr);
258 if (llh_hdr->lrh_type != LLOG_HDR_MAGIC) {
259 CERROR("%s: bad log %s "DFID" header magic: %#x "
260 "(expected %#x)\n", o->do_lu.lo_dev->ld_obd->obd_name,
261 handle->lgh_name ? handle->lgh_name : "",
262 PFID(lu_object_fid(&o->do_lu)),
263 llh_hdr->lrh_type, LLOG_HDR_MAGIC);
265 } else if (llh_hdr->lrh_len < LLOG_MIN_CHUNK_SIZE ||
266 llh_hdr->lrh_len > handle->lgh_hdr_size) {
267 CERROR("%s: incorrectly sized log %s "DFID" header: "
268 "%#x (expected at least %#x)\n"
269 "you may need to re-run lconf --write_conf.\n",
270 o->do_lu.lo_dev->ld_obd->obd_name,
271 handle->lgh_name ? handle->lgh_name : "",
272 PFID(lu_object_fid(&o->do_lu)),
273 llh_hdr->lrh_len, LLOG_MIN_CHUNK_SIZE);
275 } else if (LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_index >
276 LLOG_HDR_BITMAP_SIZE(handle->lgh_hdr) ||
277 LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_len !=
279 CERROR("%s: incorrectly sized log %s "DFID" tailer: "
281 o->do_lu.lo_dev->ld_obd->obd_name,
282 handle->lgh_name ? handle->lgh_name : "",
283 PFID(lu_object_fid(&o->do_lu)),
284 LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_len, -EIO);
288 handle->lgh_hdr->llh_flags |= (flags & LLOG_F_EXT_MASK);
289 handle->lgh_last_idx = LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_index;
295 * Implementation of the llog_operations::lop_declare_write
297 * This function declares the new record write.
299 * \param[in] env execution environment
300 * \param[in] loghandle llog handle of the current llog
301 * \param[in] rec llog record header. This is a real header of the full
302 * llog record to write. This is the beginning of buffer
303 * to write, the length of buffer is stored in
305 * \param[in] idx index of the llog record. If \a idx == -1 then this is
306 * append case, otherwise \a idx is the index of record
308 * \param[in] th current transaction handle
310 * \retval 0 on successful declaration
311 * \retval negative error if declaration failed
313 static int llog_osd_declare_write_rec(const struct lu_env *env,
314 struct llog_handle *loghandle,
315 struct llog_rec_hdr *rec,
316 int idx, struct thandle *th)
318 struct llog_thread_info *lgi = llog_info(env);
329 LASSERT(rec->lrh_len <= loghandle->lgh_ctxt->loc_chunk_size);
331 o = loghandle->lgh_obj;
334 chunk_size = loghandle->lgh_ctxt->loc_chunk_size;
335 lgi->lgi_buf.lb_len = chunk_size;
336 lgi->lgi_buf.lb_buf = NULL;
337 /* each time we update header */
338 rc = dt_declare_record_write(env, o, &lgi->lgi_buf, 0,
340 if (rc || idx == 0) /* if error or just header */
344 * the pad record can be inserted so take into account double
347 lgi->lgi_buf.lb_len = chunk_size * 2;
348 lgi->lgi_buf.lb_buf = NULL;
349 /* XXX: implement declared window or multi-chunks approach */
350 rc = dt_declare_record_write(env, o, &lgi->lgi_buf, -1, th);
356 * Implementation of the llog_operations::lop_write
358 * This function writes the new record in the llog or modify the existed one.
360 * \param[in] env execution environment
361 * \param[in] loghandle llog handle of the current llog
362 * \param[in] rec llog record header. This is a real header of
363 * the full llog record to write. This is
364 * the beginning of buffer to write, the length
365 * of buffer is stored in \a rec::lrh_len
366 * \param[in,out] reccookie pointer to the cookie to return back if needed.
367 * It is used for further cancel of this llog
369 * \param[in] idx index of the llog record. If \a idx == -1 then
370 * this is append case, otherwise \a idx is
371 * the index of record to modify
372 * \param[in] th current transaction handle
374 * \retval 0 on successful write && \a reccookie == NULL
375 * 1 on successful write && \a reccookie != NULL
376 * \retval negative error if write failed
378 static int llog_osd_write_rec(const struct lu_env *env,
379 struct llog_handle *loghandle,
380 struct llog_rec_hdr *rec,
381 struct llog_cookie *reccookie,
382 int idx, struct thandle *th)
384 struct llog_thread_info *lgi = llog_info(env);
385 struct llog_log_hdr *llh;
386 int reclen = rec->lrh_len;
388 struct llog_rec_tail *lrt;
395 llh = loghandle->lgh_hdr;
396 o = loghandle->lgh_obj;
398 chunk_size = llh->llh_hdr.lrh_len;
399 CDEBUG(D_OTHER, "new record %x to "DFID"\n",
400 rec->lrh_type, PFID(lu_object_fid(&o->do_lu)));
402 if (!llog_osd_exist(loghandle))
405 /* record length should not bigger than */
406 if (reclen > loghandle->lgh_hdr->llh_hdr.lrh_len)
409 /* sanity check for fixed-records llog */
410 if (idx != LLOG_HEADER_IDX && (llh->llh_flags & LLOG_F_IS_FIXSIZE)) {
411 LASSERT(llh->llh_size != 0);
412 LASSERT(llh->llh_size == reclen);
415 rc = dt_attr_get(env, o, &lgi->lgi_attr);
420 * The modification case.
421 * If idx set then the record with that index must be modified.
422 * There are three cases possible:
423 * 1) the common case is the llog header update (idx == 0)
424 * 2) the llog record modification during llog process.
425 * This is indicated by the \a loghandle::lgh_cur_idx > 0.
426 * In that case the \a loghandle::lgh_cur_offset
427 * 3) otherwise this is assumed that llog consist of records of
428 * fixed size, i.e. catalog. The llog header must has llh_size
429 * field equal to record size. The record offset is calculated
430 * just by /a idx value
432 * During modification we don't need extra header update because
433 * the bitmap and record count are not changed. The record header
434 * and tail remains the same too.
436 if (idx != LLOG_NEXT_IDX) {
437 /* llog can be empty only when first record is being written */
438 LASSERT(ergo(idx > 0, lgi->lgi_attr.la_size > 0));
440 if (!test_bit_le(idx, LLOG_HDR_BITMAP(llh))) {
441 CERROR("%s: modify unset record %u\n",
442 o->do_lu.lo_dev->ld_obd->obd_name, idx);
446 if (idx != rec->lrh_index) {
447 CERROR("%s: modify index mismatch %d %u\n",
448 o->do_lu.lo_dev->ld_obd->obd_name, idx,
453 if (idx == LLOG_HEADER_IDX) {
454 /* llog header update */
455 __u32 *bitmap = LLOG_HDR_BITMAP(llh);
459 /* If it does not indicate the bitmap index
460 * (reccookie == NULL), then it means update
461 * the whole update header. Otherwise only
462 * update header and bits needs to be updated,
463 * and in DNE cases, it will signaficantly
464 * shrink the RPC size.
465 * see distribute_txn_cancel_records()*/
466 if (reccookie == NULL) {
467 lgi->lgi_buf.lb_len = reclen;
468 lgi->lgi_buf.lb_buf = rec;
469 rc = dt_record_write(env, o, &lgi->lgi_buf,
474 /* update the header */
475 lgi->lgi_buf.lb_len = llh->llh_bitmap_offset;
476 lgi->lgi_buf.lb_buf = llh;
477 rc = dt_record_write(env, o, &lgi->lgi_buf,
482 /* update the bitmap */
483 index = reccookie->lgc_index;
484 lgi->lgi_off = llh->llh_bitmap_offset +
485 (index / (sizeof(*bitmap) * 8)) *
487 lgi->lgi_buf.lb_len = sizeof(*bitmap);
488 lgi->lgi_buf.lb_buf =
489 &bitmap[index/(sizeof(*bitmap)*8)];
490 rc = dt_record_write(env, o, &lgi->lgi_buf,
494 } else if (llh->llh_flags & LLOG_F_IS_FIXSIZE) {
495 lgi->lgi_off = llh->llh_hdr.lrh_len +
497 } else if (reccookie != NULL && reccookie->lgc_index > 0) {
499 * The lgc_offset can be used only if index is
502 if (idx != reccookie->lgc_index) {
503 CERROR("%s: modify index mismatch %d %d\n",
504 o->do_lu.lo_dev->ld_obd->obd_name, idx,
505 reccookie->lgc_index);
509 lgi->lgi_off = reccookie->lgc_offset;
510 CDEBUG(D_OTHER, "modify record "DFID": idx:%u, "
511 "len:%u offset %llu\n",
512 PFID(&loghandle->lgh_id.lgl_oi.oi_fid), idx,
513 rec->lrh_len, (long long)lgi->lgi_off);
515 /* This can be result of lgh_cur_idx is not set during
516 * llog processing or llh_size is not set to proper
517 * record size for fixed records llog. Therefore it is
518 * impossible to get record offset. */
519 CERROR("%s: can't get record offset, idx:%d, "
520 "len:%u.\n", o->do_lu.lo_dev->ld_obd->obd_name,
525 /* update only data, header and tail remain the same */
526 lgi->lgi_off += sizeof(struct llog_rec_hdr);
527 lgi->lgi_buf.lb_len = REC_DATA_LEN(rec);
528 lgi->lgi_buf.lb_buf = REC_DATA(rec);
529 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
530 if (rc == 0 && reccookie) {
531 reccookie->lgc_lgl = loghandle->lgh_id;
532 reccookie->lgc_index = idx;
540 * The most common case of using llog. The new index is assigned to
541 * the new record, new bit is set in llog bitmap and llog count is
544 * Make sure that records don't cross a chunk boundary, so we can
545 * process them page-at-a-time if needed. If it will cross a chunk
546 * boundary, write in a fake (but referenced) entry to pad the chunk.
550 /* simulate ENOSPC when new plain llog is being added to the
552 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_LLOG_CREATE_FAILED2) &&
553 llh->llh_flags & LLOG_F_IS_CAT)
556 LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
557 orig_last_idx = loghandle->lgh_last_idx;
558 lgi->lgi_off = lgi->lgi_attr.la_size;
560 if (loghandle->lgh_max_size > 0 &&
561 lgi->lgi_off >= loghandle->lgh_max_size) {
562 CDEBUG(D_OTHER, "llog is getting too large (%u > %u) at %u "
563 DFID"\n", (unsigned)lgi->lgi_off,
564 loghandle->lgh_max_size, (int)loghandle->lgh_last_idx,
565 PFID(&loghandle->lgh_id.lgl_oi.oi_fid));
566 /* this is to signal that this llog is full */
567 loghandle->lgh_last_idx = LLOG_HDR_BITMAP_SIZE(llh) - 1;
571 left = chunk_size - (lgi->lgi_off & (chunk_size - 1));
572 /* NOTE: padding is a record, but no bit is set */
573 if (left != 0 && left != reclen &&
574 left < (reclen + LLOG_MIN_REC_SIZE)) {
575 index = loghandle->lgh_last_idx + 1;
576 rc = llog_osd_pad(env, o, &lgi->lgi_off, left, index, th);
580 loghandle->lgh_last_idx++; /* for pad rec */
582 /* if it's the last idx in log file, then return -ENOSPC
583 * or wrap around if a catalog */
584 if (llog_is_full(loghandle) ||
585 unlikely(llh->llh_flags & LLOG_F_IS_CAT &&
586 OBD_FAIL_PRECHECK(OBD_FAIL_CAT_RECORDS) &&
587 loghandle->lgh_last_idx >= cfs_fail_val)) {
588 if (llh->llh_flags & LLOG_F_IS_CAT)
589 loghandle->lgh_last_idx = 0;
594 down_write(&loghandle->lgh_last_sem);
595 /* increment the last_idx along with llh_tail index, they should
596 * be equal for a llog lifetime */
597 loghandle->lgh_last_idx++;
598 index = loghandle->lgh_last_idx;
599 LLOG_HDR_TAIL(llh)->lrt_index = index;
601 * NB: the caller should make sure only 1 process access
602 * the lgh_last_idx, e.g. append should be exclusive.
603 * Otherwise it might hit the assert.
605 LASSERT(index < LLOG_HDR_BITMAP_SIZE(llh));
606 rec->lrh_index = index;
608 lrt->lrt_len = rec->lrh_len;
609 lrt->lrt_index = rec->lrh_index;
611 /* the lgh_hdr_mutex protects llog header data from concurrent
612 * update/cancel, the llh_count and llh_bitmap are protected */
613 mutex_lock(&loghandle->lgh_hdr_mutex);
614 if (__test_and_set_bit_le(index, LLOG_HDR_BITMAP(llh))) {
615 CERROR("%s: index %u already set in log bitmap\n",
616 o->do_lu.lo_dev->ld_obd->obd_name, index);
617 mutex_unlock(&loghandle->lgh_hdr_mutex);
618 LBUG(); /* should never happen */
622 if (!(llh->llh_flags & LLOG_F_IS_FIXSIZE)) {
623 /* Update the minimum size of the llog record */
624 if (llh->llh_size == 0)
625 llh->llh_size = reclen;
626 else if (reclen < llh->llh_size)
627 llh->llh_size = reclen;
630 if (lgi->lgi_attr.la_size == 0) {
632 lgi->lgi_buf.lb_len = llh->llh_hdr.lrh_len;
633 lgi->lgi_buf.lb_buf = &llh->llh_hdr;
634 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
636 GOTO(out_unlock, rc);
638 __u32 *bitmap = LLOG_HDR_BITMAP(llh);
640 /* Note: If this is not initialization (size == 0), then do not
641 * write the whole header (8k bytes), only update header/tail
642 * and bits needs to be updated. Because this update might be
643 * part of cross-MDT operation, which needs to write these
644 * updates into the update log(32KB limit) and also pack inside
645 * the RPC (1MB limit), if we write 8K for each operation, which
646 * will cost a lot space, and keep us adding more updates to one
649 lgi->lgi_buf.lb_len = llh->llh_bitmap_offset;
650 lgi->lgi_buf.lb_buf = &llh->llh_hdr;
651 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
653 GOTO(out_unlock, rc);
655 lgi->lgi_off = llh->llh_bitmap_offset +
656 (index / (sizeof(*bitmap) * 8)) * sizeof(*bitmap);
657 lgi->lgi_buf.lb_len = sizeof(*bitmap);
658 lgi->lgi_buf.lb_buf = &bitmap[index/(sizeof(*bitmap)*8)];
659 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
661 GOTO(out_unlock, rc);
663 lgi->lgi_off = (unsigned long)LLOG_HDR_TAIL(llh) -
665 lgi->lgi_buf.lb_len = sizeof(llh->llh_tail);
666 lgi->lgi_buf.lb_buf = LLOG_HDR_TAIL(llh);
667 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
669 GOTO(out_unlock, rc);
673 /* unlock here for remote object */
674 mutex_unlock(&loghandle->lgh_hdr_mutex);
678 if (OBD_FAIL_PRECHECK(OBD_FAIL_LLOG_PROCESS_TIMEOUT) &&
679 cfs_fail_val == (unsigned int)(loghandle->lgh_id.lgl_oi.oi.oi_id &
681 OBD_RACE(OBD_FAIL_LLOG_PROCESS_TIMEOUT);
682 msleep(1 * MSEC_PER_SEC);
684 /* computed index can be used to determine offset for fixed-size
685 * records. This also allows to handle Catalog wrap around case */
686 if (llh->llh_flags & LLOG_F_IS_FIXSIZE) {
687 lgi->lgi_off = llh->llh_hdr.lrh_len + (index - 1) * reclen;
689 rc = dt_attr_get(env, o, &lgi->lgi_attr);
693 LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
694 lgi->lgi_off = max_t(__u64, lgi->lgi_attr.la_size,
698 lgi->lgi_buf.lb_len = reclen;
699 lgi->lgi_buf.lb_buf = rec;
700 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
704 up_write(&loghandle->lgh_last_sem);
706 CDEBUG(D_HA, "added record "DFID".%u, %u off%llu\n",
707 PFID(lu_object_fid(&o->do_lu)), index, rec->lrh_len,
709 if (reccookie != NULL) {
710 reccookie->lgc_lgl = loghandle->lgh_id;
711 reccookie->lgc_index = index;
712 if ((rec->lrh_type == MDS_UNLINK_REC) ||
713 (rec->lrh_type == MDS_SETATTR64_REC))
714 reccookie->lgc_subsys = LLOG_MDS_OST_ORIG_CTXT;
715 else if (rec->lrh_type == OST_SZ_REC)
716 reccookie->lgc_subsys = LLOG_SIZE_ORIG_CTXT;
718 reccookie->lgc_subsys = -1;
723 /* cleanup llog for error case */
724 mutex_lock(&loghandle->lgh_hdr_mutex);
725 clear_bit_le(index, LLOG_HDR_BITMAP(llh));
727 mutex_unlock(&loghandle->lgh_hdr_mutex);
729 /* restore llog last_idx */
730 if (dt_object_remote(o)) {
731 loghandle->lgh_last_idx = orig_last_idx;
732 } else if (--loghandle->lgh_last_idx == 0 &&
733 (llh->llh_flags & LLOG_F_IS_CAT) && llh->llh_cat_idx != 0) {
734 /* catalog had just wrap-around case */
735 loghandle->lgh_last_idx = LLOG_HDR_BITMAP_SIZE(llh) - 1;
738 LLOG_HDR_TAIL(llh)->lrt_index = loghandle->lgh_last_idx;
739 up_write(&loghandle->lgh_last_sem);
745 * We can skip reading at least as many log blocks as the number of
746 * minimum sized log records we are skipping. If it turns out
747 * that we are not far enough along the log (because the
748 * actual records are larger than minimum size) we just skip
751 * Note: in llog_process_thread, it will use bitmap offset as
752 * the index to locate the record, which also includs some pad
753 * records, whose record size is very small, and it also does not
754 * consider pad record when recording minimum record size (otherwise
755 * min_record size might be too small), so in some rare cases,
756 * it might skip too much record for @goal, see llog_osd_next_block().
758 * When force_mini_rec is true, it means we have to use LLOG_MIN_REC_SIZE
759 * as the min record size to skip over, usually because in the previous
760 * try, it skip too much record, see loog_osd_next(prev)_block().
762 static inline void llog_skip_over(struct llog_handle *lgh, __u64 *off,
763 int curr, int goal, __u32 chunk_size,
766 struct llog_log_hdr *llh = lgh->lgh_hdr;
768 /* Goal should not bigger than the record count */
769 if (goal > lgh->lgh_last_idx)
770 goal = lgh->lgh_last_idx;
773 if (llh->llh_flags & LLOG_F_IS_FIXSIZE) {
774 *off = chunk_size + (goal - 1) * llh->llh_size;
776 __u64 min_rec_size = LLOG_MIN_REC_SIZE;
778 if (llh->llh_size > 0 && !force_mini_rec)
779 min_rec_size = llh->llh_size;
781 *off = *off + (goal - curr - 1) * min_rec_size;
784 /* always align with lower chunk boundary*/
785 *off &= ~(chunk_size - 1);
789 * Remove optional fields that the client doesn't expect.
790 * This is typically in order to ensure compatibility with older clients.
791 * It is assumed that since we exclusively remove fields, the block will be
792 * big enough to handle the remapped records. It is also assumed that records
793 * of a block have the same format (i.e.: the same features enabled).
795 * \param[in,out] hdr Header of the block of records to remap.
796 * \param[in,out] last_hdr Last header, don't read past this point.
797 * \param[in] flags Flags describing the fields to keep.
798 * \param[in] extra_flags Flags describing the extra fields to keep.
800 static void changelog_block_trim_ext(struct llog_rec_hdr *hdr,
801 struct llog_rec_hdr *last_hdr,
802 struct llog_handle *loghandle)
804 enum changelog_rec_flags flags = CLF_SUPPORTED;
805 enum changelog_rec_extra_flags extra_flags = CLFE_SUPPORTED;
807 if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_X_XATTR))
808 extra_flags &= ~CLFE_XATTR;
809 if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_X_OMODE))
810 extra_flags &= ~CLFE_OPEN;
811 if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_X_NID))
812 extra_flags &= ~CLFE_NID;
813 if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_X_UIDGID))
814 extra_flags &= ~CLFE_UIDGID;
815 if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_EXTRA_FLAGS))
816 flags &= ~CLF_EXTRA_FLAGS;
817 if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_JOBID))
820 if (flags == CLF_SUPPORTED && extra_flags == CLFE_SUPPORTED)
823 if (hdr->lrh_type != CHANGELOG_REC)
827 struct changelog_rec *rec = (struct changelog_rec *)(hdr + 1);
828 enum changelog_rec_extra_flags xflag = CLFE_INVALID;
830 if (flags & CLF_EXTRA_FLAGS &&
831 rec->cr_flags & CLF_EXTRA_FLAGS) {
832 xflag = changelog_rec_extra_flags(rec)->cr_extra_flags &
836 if (unlikely(hdr->lrh_len == 0)) {
837 /* It is corruption case, we cannot know the next rec,
838 * jump to the last one directly to avoid dead loop. */
839 LCONSOLE(D_WARNING, "Hit invalid llog record: "
840 "idx %u, type %u, id %u\n",
841 hdr->lrh_index, hdr->lrh_type, hdr->lrh_id);
842 hdr = llog_rec_hdr_next(last_hdr);
843 if (unlikely(hdr == last_hdr))
844 LCONSOLE(D_WARNING, "The last record crashed: "
845 "idx %u, type %u, id %u\n",
846 hdr->lrh_index, hdr->lrh_type,
851 changelog_remap_rec(rec, rec->cr_flags & flags, xflag);
852 hdr = llog_rec_hdr_next(hdr);
853 /* Yield CPU to avoid soft-lockup if there are too many records
856 } while ((char *)hdr <= (char *)last_hdr);
860 * Implementation of the llog_operations::lop_next_block
862 * This function finds the the next llog block to return which contains
863 * record with required index. It is main part of llog processing.
865 * \param[in] env execution environment
866 * \param[in] loghandle llog handle of the current llog
867 * \param[in,out] cur_idx index preceeding cur_offset
868 * \param[in] next_idx target index to find
869 * \param[in,out] cur_offset furtherst point read in the file
870 * \param[in] buf pointer to data buffer to fill
871 * \param[in] len required len to read, it is
872 * usually llog chunk_size.
874 * \retval 0 on successful buffer read
875 * \retval negative value on error
877 static int llog_osd_next_block(const struct lu_env *env,
878 struct llog_handle *loghandle, int *cur_idx,
879 int next_idx, __u64 *cur_offset, void *buf,
882 struct llog_thread_info *lgi = llog_info(env);
884 struct dt_device *dt;
887 int last_idx = *cur_idx;
888 __u64 last_offset = *cur_offset;
889 bool force_mini_rec = false;
896 chunk_size = loghandle->lgh_hdr->llh_hdr.lrh_len;
897 if (len == 0 || len & (chunk_size - 1))
901 LASSERT(loghandle->lgh_ctxt);
903 o = loghandle->lgh_obj;
905 LASSERT(llog_osd_exist(loghandle));
906 dt = lu2dt_dev(o->do_lu.lo_dev);
909 rc = dt_attr_get(env, o, &lgi->lgi_attr);
914 "looking for log index %u (cur idx %u off %llu), size %llu\n",
916 *cur_offset, lgi->lgi_attr.la_size);
918 while (*cur_offset < lgi->lgi_attr.la_size) {
919 struct llog_rec_hdr *rec, *last_rec;
920 struct llog_rec_tail *tail;
922 llog_skip_over(loghandle, cur_offset, *cur_idx,
923 next_idx, chunk_size, force_mini_rec);
925 /* read up to next llog chunk_size block */
926 lgi->lgi_buf.lb_len = chunk_size -
927 (*cur_offset & (chunk_size - 1));
928 lgi->lgi_buf.lb_buf = buf;
930 rc = dt_read(env, o, &lgi->lgi_buf, cur_offset);
932 if (rc == -EBADR && !force_mini_rec)
935 CERROR("%s: can't read llog block from log "DFID
936 " offset %llu: rc = %d\n",
937 o->do_lu.lo_dev->ld_obd->obd_name,
938 PFID(lu_object_fid(&o->do_lu)), *cur_offset,
944 /* signal the end of the valid buffer to
946 memset(buf + rc, 0, len - rc);
949 if (rc == 0) { /* end of file, nothing to do */
955 if (rc < sizeof(*tail)) {
959 CERROR("%s: invalid llog block at log id "DFID":%x "
961 o->do_lu.lo_dev->ld_obd->obd_name,
962 PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
963 loghandle->lgh_id.lgl_ogen, *cur_offset);
964 GOTO(out, rc = -EINVAL);
968 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
969 lustre_swab_llog_rec(rec);
971 tail = (struct llog_rec_tail *)((char *)buf + rc -
972 sizeof(struct llog_rec_tail));
973 /* get the last record in block */
974 last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
977 if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
978 lustre_swab_llog_rec(last_rec);
980 if (last_rec->lrh_index != tail->lrt_index) {
981 CERROR("%s: invalid llog tail at log id "DFID":%x offset %llu last_rec idx %u tail idx %u lrt len %u read_size %d\n",
982 o->do_lu.lo_dev->ld_obd->obd_name,
983 PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
984 loghandle->lgh_id.lgl_ogen, *cur_offset,
985 last_rec->lrh_index, tail->lrt_index,
987 GOTO(out, rc = -EINVAL);
990 *cur_idx = tail->lrt_index;
992 /* this shouldn't happen */
993 if (tail->lrt_index == 0) {
994 CERROR("%s: invalid llog tail at log id "DFID":%x "
995 "offset %llu bytes %d\n",
996 o->do_lu.lo_dev->ld_obd->obd_name,
997 PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
998 loghandle->lgh_id.lgl_ogen, *cur_offset, rc);
999 GOTO(out, rc = -EINVAL);
1001 if (tail->lrt_index < next_idx) {
1002 last_idx = *cur_idx;
1003 last_offset = *cur_offset;
1007 /* sanity check that the start of the new buffer is no farther
1008 * than the record that we wanted. This shouldn't happen. */
1009 if (rec->lrh_index > next_idx) {
1010 if (!force_mini_rec && next_idx > last_idx)
1013 CERROR("%s: missed desired record? %u > %u\n",
1014 o->do_lu.lo_dev->ld_obd->obd_name,
1015 rec->lrh_index, next_idx);
1016 GOTO(out, rc = -ENOENT);
1019 /* Trim unsupported extensions for compat w/ older clients */
1020 changelog_block_trim_ext(rec, last_rec, loghandle);
1025 /* Note: because there are some pad records in the
1026 * llog, so llog_skip_over() might skip too much
1027 * records, let's try skip again with minimum record */
1028 force_mini_rec = true;
1029 *cur_offset = last_offset;
1030 *cur_idx = last_idx;
1032 GOTO(out, rc = -EIO);
1038 * Implementation of the llog_operations::lop_prev_block
1040 * This function finds the llog block to return which contains
1041 * record with required index but in reverse order - from end of llog
1043 * It is main part of reverse llog processing.
1045 * \param[in] env execution environment
1046 * \param[in] loghandle llog handle of the current llog
1047 * \param[in] prev_idx target index to find
1048 * \param[in] buf pointer to data buffer to fill
1049 * \param[in] len required len to read, it is llog_chunk_size usually.
1051 * \retval 0 on successful buffer read
1052 * \retval negative value on error
1054 static int llog_osd_prev_block(const struct lu_env *env,
1055 struct llog_handle *loghandle,
1056 int prev_idx, void *buf, int len)
1058 struct llog_thread_info *lgi = llog_info(env);
1059 struct dt_object *o;
1060 struct dt_device *dt;
1067 chunk_size = loghandle->lgh_hdr->llh_hdr.lrh_len;
1068 if (len == 0 || len & (chunk_size - 1))
1071 CDEBUG(D_OTHER, "looking for log index %u\n", prev_idx);
1074 LASSERT(loghandle->lgh_ctxt);
1076 o = loghandle->lgh_obj;
1078 LASSERT(llog_osd_exist(loghandle));
1079 dt = lu2dt_dev(o->do_lu.lo_dev);
1082 /* Let's only use mini record size for previous block read
1084 cur_offset = chunk_size;
1085 llog_skip_over(loghandle, &cur_offset, 0, prev_idx,
1088 rc = dt_attr_get(env, o, &lgi->lgi_attr);
1092 while (cur_offset < lgi->lgi_attr.la_size) {
1093 struct llog_rec_hdr *rec, *last_rec;
1094 struct llog_rec_tail *tail;
1096 lgi->lgi_buf.lb_len = len;
1097 lgi->lgi_buf.lb_buf = buf;
1098 rc = dt_read(env, o, &lgi->lgi_buf, &cur_offset);
1100 CERROR("%s: can't read llog block from log "DFID
1101 " offset %llu: rc = %d\n",
1102 o->do_lu.lo_dev->ld_obd->obd_name,
1103 PFID(lu_object_fid(&o->do_lu)), cur_offset, rc);
1107 if (rc == 0) /* end of file, nothing to do */
1110 if (rc < sizeof(*tail)) {
1111 CERROR("%s: invalid llog block at log id "DFID":%x "
1113 o->do_lu.lo_dev->ld_obd->obd_name,
1114 PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
1115 loghandle->lgh_id.lgl_ogen, cur_offset);
1116 GOTO(out, rc = -EINVAL);
1120 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
1121 lustre_swab_llog_rec(rec);
1123 tail = (struct llog_rec_tail *)((char *)buf + rc -
1124 sizeof(struct llog_rec_tail));
1125 /* get the last record in block */
1126 last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
1127 le32_to_cpu(tail->lrt_len));
1129 if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
1130 lustre_swab_llog_rec(last_rec);
1131 LASSERT(last_rec->lrh_index == tail->lrt_index);
1133 /* this shouldn't happen */
1134 if (tail->lrt_index == 0) {
1135 CERROR("%s: invalid llog tail at log id "DFID":%x "
1137 o->do_lu.lo_dev->ld_obd->obd_name,
1138 PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
1139 loghandle->lgh_id.lgl_ogen, cur_offset);
1140 GOTO(out, rc = -EINVAL);
1142 if (tail->lrt_index < prev_idx)
1145 /* sanity check that the start of the new buffer is no farther
1146 * than the record that we wanted. This shouldn't happen. */
1147 if (rec->lrh_index > prev_idx) {
1148 CERROR("%s: missed desired record? %u > %u\n",
1149 o->do_lu.lo_dev->ld_obd->obd_name,
1150 rec->lrh_index, prev_idx);
1151 GOTO(out, rc = -ENOENT);
1154 /* Trim unsupported extensions for compat w/ older clients */
1155 changelog_block_trim_ext(rec, last_rec, loghandle);
1159 GOTO(out, rc = -EIO);
1165 * This is helper function to get llog directory object. It is used by named
1166 * llog operations to find/insert/delete llog entry from llog directory.
1168 * \param[in] env execution environment
1169 * \param[in] ctxt llog context
1171 * \retval dt_object of llog directory
1172 * \retval ERR_PTR of negative value on error
1174 static struct dt_object *llog_osd_dir_get(const struct lu_env *env,
1175 struct llog_ctxt *ctxt)
1177 struct dt_device *dt;
1178 struct dt_thread_info *dti = dt_info(env);
1179 struct dt_object *dir;
1182 dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
1183 if (ctxt->loc_dir == NULL) {
1184 rc = dt_root_get(env, dt, &dti->dti_fid);
1187 dir = dt_locate(env, dt, &dti->dti_fid);
1189 if (!IS_ERR(dir) && !dt_try_as_dir(env, dir)) {
1190 dt_object_put(env, dir);
1191 return ERR_PTR(-ENOTDIR);
1194 lu_object_get(&ctxt->loc_dir->do_lu);
1195 dir = ctxt->loc_dir;
1202 * Implementation of the llog_operations::lop_open
1204 * This function opens the llog by its logid or by name, it may open also
1205 * non existent llog and assing then new id to it.
1206 * The llog_open/llog_close pair works similar to lu_object_find/put,
1207 * the object may not exist prior open. The result of open is just dt_object
1208 * in the llog header.
1210 * \param[in] env execution environment
1211 * \param[in] handle llog handle of the current llog
1212 * \param[in] logid logid of llog to open (nameless llog)
1213 * \param[in] name name of llog to open (named llog)
1214 * \param[in] open_param
1215 * LLOG_OPEN_NEW - new llog, may not exist
1216 * LLOG_OPEN_EXIST - old llog, must exist
1218 * \retval 0 on successful open, llog_handle::lgh_obj
1219 * contains the dt_object of the llog.
1220 * \retval negative value on error
1222 static int llog_osd_open(const struct lu_env *env, struct llog_handle *handle,
1223 struct llog_logid *logid, char *name,
1224 enum llog_open_param open_param)
1226 struct llog_thread_info *lgi = llog_info(env);
1227 struct llog_ctxt *ctxt = handle->lgh_ctxt;
1228 struct dt_object *o;
1229 struct dt_device *dt;
1230 struct ls_device *ls;
1231 struct local_oid_storage *los = NULL;
1233 bool new_id = false;
1239 LASSERT(ctxt->loc_exp);
1240 LASSERT(ctxt->loc_exp->exp_obd);
1241 dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
1243 if (ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1244 struct lu_object_conf conf = { 0 };
1245 if (logid != NULL) {
1246 logid_to_fid(logid, &lgi->lgi_fid);
1248 /* If logid == NULL, then it means the caller needs
1249 * to allocate new FID (llog_cat_declare_add_rec()). */
1250 rc = obd_fid_alloc(env, ctxt->loc_exp,
1251 &lgi->lgi_fid, NULL);
1255 conf.loc_flags = LOC_F_NEW;
1258 o = dt_locate_at(env, dt, &lgi->lgi_fid,
1259 dt->dd_lu_dev.ld_site->ls_top_dev, &conf);
1266 ls = ls_device_get(dt);
1268 RETURN(PTR_ERR(ls));
1270 mutex_lock(&ls->ls_los_mutex);
1271 los = dt_los_find(ls, name != NULL ? FID_SEQ_LLOG_NAME : FID_SEQ_LLOG);
1272 mutex_unlock(&ls->ls_los_mutex);
1274 ls_device_put(env, ls);
1278 if (logid != NULL) {
1279 logid_to_fid(logid, &lgi->lgi_fid);
1281 struct dt_object *llog_dir;
1283 llog_dir = llog_osd_dir_get(env, ctxt);
1284 if (IS_ERR(llog_dir))
1285 GOTO(out, rc = PTR_ERR(llog_dir));
1286 dt_read_lock(env, llog_dir, 0);
1287 rc = dt_lookup_dir(env, llog_dir, name, &lgi->lgi_fid);
1288 dt_read_unlock(env, llog_dir);
1289 dt_object_put(env, llog_dir);
1290 if (rc == -ENOENT && open_param == LLOG_OPEN_NEW) {
1291 /* generate fid for new llog */
1292 rc = local_object_fid_generate(env, los,
1298 OBD_ALLOC(handle->lgh_name, strlen(name) + 1);
1299 if (handle->lgh_name)
1300 strcpy(handle->lgh_name, name);
1302 GOTO(out, rc = -ENOMEM);
1304 LASSERTF(open_param & LLOG_OPEN_NEW, "%#x\n", open_param);
1305 /* generate fid for new llog */
1307 rc = local_object_fid_generate(env, los, &lgi->lgi_fid);
1312 if (OBD_FAIL_PRECHECK(OBD_FAIL_MDS_LLOG_UMOUNT_RACE) &&
1313 cfs_fail_val == 1) {
1315 OBD_RACE(OBD_FAIL_MDS_LLOG_UMOUNT_RACE);
1316 msleep(MSEC_PER_SEC);
1318 o = ls_locate(env, ls, &lgi->lgi_fid, NULL);
1320 GOTO(out_name, rc = PTR_ERR(o));
1322 if (dt_object_exists(o) && new_id) {
1323 /* llog exists with just generated ID, e.g. some old llog file
1324 * still is in use or is orphan, drop a warn and skip it. */
1325 CDEBUG(D_INFO, "%s: llog exists with the same FID: "DFID
1327 o->do_lu.lo_dev->ld_obd->obd_name,
1328 PFID(lu_object_fid(&o->do_lu)));
1329 dt_object_put(env, o);
1330 /* just skip this llog ID, we shouldn't delete it because we
1331 * don't know exactly what is its purpose and state. */
1336 /* No new llog is expected but doesn't exist */
1337 if (open_param != LLOG_OPEN_NEW && !dt_object_exists(o)) {
1338 CDEBUG(D_INFO, "%s: llog FID: "DFID" obj %p doesn`t exist\n",
1339 o->do_lu.lo_dev->ld_obd->obd_name,
1340 PFID(lu_object_fid(&o->do_lu)), o);
1341 GOTO(out_put, rc = -ENOENT);
1343 fid_to_logid(&lgi->lgi_fid, &handle->lgh_id);
1344 handle->lgh_obj = o;
1345 handle->private_data = los;
1346 LASSERT(handle->lgh_ctxt);
1351 dt_object_put(env, o);
1353 if (handle->lgh_name != NULL)
1354 OBD_FREE(handle->lgh_name, strlen(name) + 1);
1362 * Get dir for regular fid log object
1364 * Get directory for regular fid log object, and these regular fid log
1365 * object will be inserted under this directory, to satisfy the FS
1366 * consistency check, e2fsck etc.
1368 * \param [in] env execution environment
1369 * \param [in] dto llog object
1371 * \retval pointer to the directory if it is found.
1372 * \retval ERR_PTR(negative errno) if it fails.
1374 struct dt_object *llog_osd_get_regular_fid_dir(const struct lu_env *env,
1375 struct dt_object *dto)
1377 struct llog_thread_info *lgi = llog_info(env);
1378 struct seq_server_site *ss = dto->do_lu.lo_dev->ld_site->ld_seq_site;
1379 struct lu_seq_range *range = &lgi->lgi_range;
1380 struct lu_fid *dir_fid = &lgi->lgi_fid;
1381 struct dt_object *dir;
1385 fld_range_set_any(range);
1386 LASSERT(ss != NULL);
1387 rc = ss->ss_server_fld->lsf_seq_lookup(env, ss->ss_server_fld,
1388 fid_seq(lu_object_fid(&dto->do_lu)), range);
1390 RETURN(ERR_PTR(rc));
1392 lu_update_log_dir_fid(dir_fid, range->lsr_index);
1393 dir = dt_locate(env, lu2dt_dev(dto->do_lu.lo_dev), dir_fid);
1397 if (!dt_try_as_dir(env, dir)) {
1398 dt_object_put(env, dir);
1399 RETURN(ERR_PTR(-ENOTDIR));
1406 * Add llog object with regular FID to name entry
1408 * Add llog object with regular FID to name space, and each llog
1409 * object on each MDT will be /update_log_dir/[seq:oid:ver],
1410 * so to satisfy the namespace consistency check, e2fsck etc.
1412 * \param [in] env execution environment
1413 * \param [in] dto llog object
1414 * \param [in] th thandle
1415 * \param [in] declare if it is declare or execution
1417 * \retval 0 if insertion succeeds.
1418 * \retval negative errno if insertion fails.
1421 llog_osd_regular_fid_add_name_entry(const struct lu_env *env,
1422 struct dt_object *dto,
1423 struct thandle *th, bool declare)
1425 struct llog_thread_info *lgi = llog_info(env);
1426 const struct lu_fid *fid = lu_object_fid(&dto->do_lu);
1427 struct dt_insert_rec *rec = &lgi->lgi_dt_rec;
1428 struct dt_object *dir;
1429 char *name = lgi->lgi_name;
1433 if (!fid_is_norm(fid))
1436 dir = llog_osd_get_regular_fid_dir(env, dto);
1438 RETURN(PTR_ERR(dir));
1441 rec->rec_type = S_IFREG;
1442 snprintf(name, sizeof(lgi->lgi_name), DFID, PFID(fid));
1443 dt_write_lock(env, dir, 0);
1445 rc = dt_declare_insert(env, dir, (struct dt_rec *)rec,
1446 (struct dt_key *)name, th);
1448 rc = dt_insert(env, dir, (struct dt_rec *)rec,
1449 (struct dt_key *)name, th);
1451 dt_write_unlock(env, dir);
1453 dt_object_put(env, dir);
1459 * Implementation of the llog_operations::lop_declare_create
1461 * This function declares the llog create. It declares also name insert
1462 * into llog directory in case of named llog.
1464 * \param[in] env execution environment
1465 * \param[in] res llog handle of the current llog
1466 * \param[in] th current transaction handle
1468 * \retval 0 on successful create declaration
1469 * \retval negative value on error
1471 static int llog_osd_declare_create(const struct lu_env *env,
1472 struct llog_handle *res, struct thandle *th)
1474 struct llog_thread_info *lgi = llog_info(env);
1475 struct dt_insert_rec *rec = &lgi->lgi_dt_rec;
1476 struct local_oid_storage *los;
1477 struct dt_object *o;
1482 LASSERT(res->lgh_obj);
1485 /* object can be created by another thread */
1487 if (dt_object_exists(o))
1490 if (res->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1491 struct llog_thread_info *lgi = llog_info(env);
1493 lgi->lgi_attr.la_valid = LA_MODE | LA_SIZE;
1494 lgi->lgi_attr.la_size = 0;
1495 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1496 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1498 rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL,
1504 rc = llog_osd_regular_fid_add_name_entry(env, o, th, true);
1508 los = res->private_data;
1511 rc = llog_osd_declare_new_object(env, los, o, th);
1515 /* do not declare header initialization here as it's declared
1516 * in llog_osd_declare_write_rec() which is always called */
1518 if (res->lgh_name) {
1519 struct dt_object *llog_dir;
1521 llog_dir = llog_osd_dir_get(env, res->lgh_ctxt);
1522 if (IS_ERR(llog_dir))
1523 RETURN(PTR_ERR(llog_dir));
1524 logid_to_fid(&res->lgh_id, &lgi->lgi_fid);
1525 rec->rec_fid = &lgi->lgi_fid;
1526 rec->rec_type = S_IFREG;
1527 rc = dt_declare_insert(env, llog_dir,
1528 (struct dt_rec *)rec,
1529 (struct dt_key *)res->lgh_name, th);
1530 dt_object_put(env, llog_dir);
1532 CERROR("%s: can't declare named llog %s: rc = %d\n",
1533 o->do_lu.lo_dev->ld_obd->obd_name,
1540 * Implementation of the llog_operations::lop_create
1542 * This function creates the llog according with llog_handle::lgh_obj
1543 * and llog_handle::lgh_name.
1545 * \param[in] env execution environment
1546 * \param[in] res llog handle of the current llog
1547 * \param[in] th current transaction handle
1549 * \retval 0 on successful create
1550 * \retval negative value on error
1552 static int llog_osd_create(const struct lu_env *env, struct llog_handle *res,
1555 struct llog_thread_info *lgi = llog_info(env);
1556 struct dt_insert_rec *rec = &lgi->lgi_dt_rec;
1557 struct local_oid_storage *los;
1558 struct dt_object *o;
1567 /* llog can be already created */
1568 if (dt_object_exists(o))
1571 if (res->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1572 struct llog_thread_info *lgi = llog_info(env);
1574 lgi->lgi_attr.la_valid = LA_MODE | LA_SIZE | LA_TYPE;
1575 lgi->lgi_attr.la_size = 0;
1576 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1577 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1579 dt_write_lock(env, o, 0);
1580 rc = dt_create(env, o, &lgi->lgi_attr, NULL,
1582 dt_write_unlock(env, o);
1586 rc = llog_osd_regular_fid_add_name_entry(env, o, th, false);
1591 los = res->private_data;
1594 dt_write_lock(env, o, 0);
1595 if (!dt_object_exists(o))
1596 rc = llog_osd_create_new_object(env, los, o, th);
1600 dt_write_unlock(env, o);
1604 if (res->lgh_name) {
1605 struct dt_object *llog_dir;
1607 llog_dir = llog_osd_dir_get(env, res->lgh_ctxt);
1608 if (IS_ERR(llog_dir))
1609 RETURN(PTR_ERR(llog_dir));
1611 logid_to_fid(&res->lgh_id, &lgi->lgi_fid);
1612 rec->rec_fid = &lgi->lgi_fid;
1613 rec->rec_type = S_IFREG;
1614 dt_read_lock(env, llog_dir, 0);
1615 rc = dt_insert(env, llog_dir, (struct dt_rec *)rec,
1616 (struct dt_key *)res->lgh_name, th);
1617 dt_read_unlock(env, llog_dir);
1618 dt_object_put(env, llog_dir);
1620 CERROR("%s: can't create named llog %s: rc = %d\n",
1621 o->do_lu.lo_dev->ld_obd->obd_name,
1628 * Implementation of the llog_operations::lop_close
1630 * This function closes the llog. It just put llog object and referenced
1633 * \param[in] env execution environment
1634 * \param[in] handle llog handle of the current llog
1636 * \retval 0 on successful llog close
1637 * \retval negative value on error
1639 static int llog_osd_close(const struct lu_env *env, struct llog_handle *handle)
1641 struct local_oid_storage *los;
1646 LASSERT(handle->lgh_obj);
1648 if (handle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1649 /* Remove the object from the cache, otherwise it may
1650 * hold LOD being released during cleanup process */
1651 dt_object_put_nocache(env, handle->lgh_obj);
1652 LASSERT(handle->private_data == NULL);
1655 dt_object_put(env, handle->lgh_obj);
1657 los = handle->private_data;
1661 if (handle->lgh_name)
1662 OBD_FREE(handle->lgh_name, strlen(handle->lgh_name) + 1);
1668 * delete llog object name entry
1670 * Delete llog object (with regular FID) from name space (under
1673 * \param [in] env execution environment
1674 * \param [in] dto llog object
1675 * \param [in] th thandle
1676 * \param [in] declare if it is declare or execution
1678 * \retval 0 if deletion succeeds.
1679 * \retval negative errno if deletion fails.
1682 llog_osd_regular_fid_del_name_entry(const struct lu_env *env,
1683 struct dt_object *dto,
1684 struct thandle *th, bool declare)
1686 struct llog_thread_info *lgi = llog_info(env);
1687 const struct lu_fid *fid = lu_object_fid(&dto->do_lu);
1688 struct dt_object *dir;
1689 char *name = lgi->lgi_name;
1693 if (!fid_is_norm(fid))
1696 dir = llog_osd_get_regular_fid_dir(env, dto);
1698 RETURN(PTR_ERR(dir));
1700 snprintf(name, sizeof(lgi->lgi_name), DFID, PFID(fid));
1701 dt_write_lock(env, dir, 0);
1703 rc = dt_declare_delete(env, dir, (struct dt_key *)name,
1706 rc = dt_delete(env, dir, (struct dt_key *)name, th);
1708 dt_write_unlock(env, dir);
1710 dt_object_put(env, dir);
1715 * Implementation of the llog_operations::lop_declare_destroy
1717 * This function declare destroys the llog and deletes also entry in the
1718 * llog directory in case of named llog. Llog should be opened prior that.
1720 * \param[in] env execution environment
1721 * \param[in] loghandle llog handle of the current llog
1723 * \retval 0 on successful destroy
1724 * \retval negative value on error
1726 static int llog_osd_declare_destroy(const struct lu_env *env,
1727 struct llog_handle *loghandle,
1730 struct llog_ctxt *ctxt;
1731 struct dt_object *o, *llog_dir = NULL;
1736 ctxt = loghandle->lgh_ctxt;
1739 o = loghandle->lgh_obj;
1742 if (loghandle->lgh_name) {
1743 llog_dir = llog_osd_dir_get(env, ctxt);
1744 if (IS_ERR(llog_dir))
1745 RETURN(PTR_ERR(llog_dir));
1747 rc = dt_declare_delete(env, llog_dir,
1748 (struct dt_key *)loghandle->lgh_name,
1754 rc = dt_declare_ref_del(env, o, th);
1758 rc = dt_declare_destroy(env, o, th);
1762 if (loghandle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1763 rc = llog_osd_regular_fid_del_name_entry(env, o, th, true);
1769 if (!(IS_ERR_OR_NULL(llog_dir)))
1770 dt_object_put(env, llog_dir);
1777 * Implementation of the llog_operations::lop_destroy
1779 * This function destroys the llog and deletes also entry in the
1780 * llog directory in case of named llog. Llog should be opened prior that.
1781 * Destroy method is not part of external transaction and does everything
1784 * \param[in] env execution environment
1785 * \param[in] loghandle llog handle of the current llog
1787 * \retval 0 on successful destroy
1788 * \retval negative value on error
1790 static int llog_osd_destroy(const struct lu_env *env,
1791 struct llog_handle *loghandle, struct thandle *th)
1793 struct llog_ctxt *ctxt;
1794 struct dt_object *o, *llog_dir = NULL;
1799 ctxt = loghandle->lgh_ctxt;
1800 LASSERT(ctxt != NULL);
1802 o = loghandle->lgh_obj;
1805 dt_write_lock(env, o, 0);
1806 if (!llog_osd_exist(loghandle))
1807 GOTO(out_unlock, rc = 0);
1809 if (loghandle->lgh_name) {
1810 llog_dir = llog_osd_dir_get(env, ctxt);
1811 if (IS_ERR(llog_dir))
1812 GOTO(out_unlock, rc = PTR_ERR(llog_dir));
1814 dt_read_lock(env, llog_dir, 0);
1815 rc = dt_delete(env, llog_dir,
1816 (struct dt_key *)loghandle->lgh_name,
1818 dt_read_unlock(env, llog_dir);
1820 CERROR("%s: can't remove llog %s: rc = %d\n",
1821 o->do_lu.lo_dev->ld_obd->obd_name,
1822 loghandle->lgh_name, rc);
1823 GOTO(out_unlock, rc);
1827 dt_ref_del(env, o, th);
1828 rc = dt_destroy(env, o, th);
1830 GOTO(out_unlock, rc);
1832 loghandle->lgh_destroyed = true;
1833 if (loghandle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1834 rc = llog_osd_regular_fid_del_name_entry(env, o, th, false);
1836 GOTO(out_unlock, rc);
1840 dt_write_unlock(env, o);
1841 if (!(IS_ERR_OR_NULL(llog_dir)))
1842 dt_object_put(env, llog_dir);
1847 * Implementation of the llog_operations::lop_setup
1849 * This function setup the llog on local storage.
1851 * \param[in] env execution environment
1852 * \param[in] obd obd device the llog belongs to
1853 * \param[in] olg the llog group, it is always zero group now.
1854 * \param[in] ctxt_idx the llog index, it defines the purpose of this llog.
1855 * Every new llog type have to use own index.
1856 * \param[in] disk_obd the storage obd, where llog is stored.
1858 * \retval 0 on successful llog setup
1859 * \retval negative value on error
1861 static int llog_osd_setup(const struct lu_env *env, struct obd_device *obd,
1862 struct obd_llog_group *olg, int ctxt_idx,
1863 struct obd_device *disk_obd)
1865 struct llog_thread_info *lgi = llog_info(env);
1866 struct llog_ctxt *ctxt;
1871 LASSERT(olg->olg_ctxts[ctxt_idx]);
1873 ctxt = llog_ctxt_get(olg->olg_ctxts[ctxt_idx]);
1876 if (disk_obd == NULL)
1879 /* initialize data allowing to generate new fids,
1880 * literally we need a sequece */
1881 lgi->lgi_fid.f_seq = FID_SEQ_LLOG;
1882 lgi->lgi_fid.f_oid = 1;
1883 lgi->lgi_fid.f_ver = 0;
1884 rc = local_oid_storage_init(env, disk_obd->obd_lvfs_ctxt.dt,
1886 &ctxt->loc_los_nameless);
1890 lgi->lgi_fid.f_seq = FID_SEQ_LLOG_NAME;
1891 lgi->lgi_fid.f_oid = 1;
1892 lgi->lgi_fid.f_ver = 0;
1893 rc = local_oid_storage_init(env, disk_obd->obd_lvfs_ctxt.dt,
1895 &ctxt->loc_los_named);
1897 local_oid_storage_fini(env, ctxt->loc_los_nameless);
1898 ctxt->loc_los_nameless = NULL;
1904 llog_ctxt_put(ctxt);
1909 * Implementation of the llog_operations::lop_cleanup
1911 * This function cleanups the llog on local storage.
1913 * \param[in] env execution environment
1914 * \param[in] ctxt the llog context
1918 static int llog_osd_cleanup(const struct lu_env *env, struct llog_ctxt *ctxt)
1920 if (ctxt->loc_los_nameless != NULL) {
1921 local_oid_storage_fini(env, ctxt->loc_los_nameless);
1922 ctxt->loc_los_nameless = NULL;
1925 if (ctxt->loc_los_named != NULL) {
1926 local_oid_storage_fini(env, ctxt->loc_los_named);
1927 ctxt->loc_los_named = NULL;
1933 struct llog_operations llog_osd_ops = {
1934 .lop_next_block = llog_osd_next_block,
1935 .lop_prev_block = llog_osd_prev_block,
1936 .lop_read_header = llog_osd_read_header,
1937 .lop_declare_destroy = llog_osd_declare_destroy,
1938 .lop_destroy = llog_osd_destroy,
1939 .lop_setup = llog_osd_setup,
1940 .lop_cleanup = llog_osd_cleanup,
1941 .lop_open = llog_osd_open,
1942 .lop_exist = llog_osd_exist,
1943 .lop_declare_create = llog_osd_declare_create,
1944 .lop_create = llog_osd_create,
1945 .lop_declare_write_rec = llog_osd_declare_write_rec,
1946 .lop_write_rec = llog_osd_write_rec,
1947 .lop_close = llog_osd_close,
1949 EXPORT_SYMBOL(llog_osd_ops);
1951 struct llog_operations llog_common_cat_ops = {
1952 .lop_next_block = llog_osd_next_block,
1953 .lop_prev_block = llog_osd_prev_block,
1954 .lop_read_header = llog_osd_read_header,
1955 .lop_declare_destroy = llog_osd_declare_destroy,
1956 .lop_destroy = llog_osd_destroy,
1957 .lop_setup = llog_osd_setup,
1958 .lop_cleanup = llog_osd_cleanup,
1959 .lop_open = llog_osd_open,
1960 .lop_exist = llog_osd_exist,
1961 .lop_declare_create = llog_osd_declare_create,
1962 .lop_create = llog_osd_create,
1963 .lop_declare_write_rec = llog_osd_declare_write_rec,
1964 .lop_write_rec = llog_osd_write_rec,
1965 .lop_close = llog_osd_close,
1966 .lop_add = llog_cat_add_rec,
1967 .lop_declare_add = llog_cat_declare_add_rec,
1969 EXPORT_SYMBOL(llog_common_cat_ops);
1972 * Read the special file which contains the list of llog catalogs IDs
1974 * This function reads the CATALOGS file which contains the array of llog
1975 * catalogs IDs. The main purpose of this file is to store OSP llogs indexed
1976 * by OST/MDT number.
1978 * \param[in] env execution environment
1979 * \param[in] d corresponding storage device
1980 * \param[in] idx position to start from, usually OST/MDT index
1981 * \param[in] count how many catalog IDs to read
1982 * \param[out] idarray the buffer for the data. If it is NULL then
1983 * function returns just number of catalog IDs
1985 * \param[in] fid LLOG_CATALOGS_OID for CATALOG object
1987 * \retval 0 on successful read of catalog IDs
1988 * \retval negative value on error
1989 * \retval positive value which is number of records in
1990 * the file if \a idarray is NULL
1992 int llog_osd_get_cat_list(const struct lu_env *env, struct dt_device *d,
1993 int idx, int count, struct llog_catid *idarray,
1994 const struct lu_fid *fid)
1996 struct llog_thread_info *lgi = llog_info(env);
1997 struct dt_object *o = NULL;
2005 size = sizeof(*idarray) * count;
2006 lgi->lgi_off = idx * sizeof(*idarray);
2008 lgi->lgi_fid = *fid;
2009 o = dt_locate(env, d, &lgi->lgi_fid);
2013 if (!dt_object_exists(o)) {
2014 th = dt_trans_create(env, d);
2016 GOTO(out, rc = PTR_ERR(th));
2018 lgi->lgi_attr.la_valid = LA_MODE;
2019 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
2020 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
2022 th->th_wait_submit = 1;
2023 /* Make the llog object creation synchronization, so
2024 * it will be reliable to the reference, especially
2025 * for remote reference */
2028 rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL,
2031 GOTO(out_trans, rc);
2033 rc = dt_trans_start_local(env, d, th);
2035 GOTO(out_trans, rc);
2037 dt_write_lock(env, o, 0);
2038 if (!dt_object_exists(o))
2039 rc = dt_create(env, o, &lgi->lgi_attr, NULL,
2041 dt_write_unlock(env, o);
2043 dt_trans_stop(env, d, th);
2048 rc = dt_attr_get(env, o, &lgi->lgi_attr);
2052 if (!S_ISREG(lgi->lgi_attr.la_mode)) {
2053 CERROR("%s: CATALOGS is not a regular file!: mode = %o\n",
2054 o->do_lu.lo_dev->ld_obd->obd_name,
2055 lgi->lgi_attr.la_mode);
2056 GOTO(out, rc = -ENOENT);
2059 CDEBUG(D_CONFIG, "cat list: disk size=%d, read=%d\n",
2060 (int)lgi->lgi_attr.la_size, size);
2062 /* return just number of llogs */
2063 if (idarray == NULL) {
2064 rc = lgi->lgi_attr.la_size / sizeof(*idarray);
2068 /* read for new ost index or for empty file */
2069 memset(idarray, 0, size);
2070 if (lgi->lgi_attr.la_size <= lgi->lgi_off)
2072 if (lgi->lgi_attr.la_size < lgi->lgi_off + size)
2073 size = lgi->lgi_attr.la_size - lgi->lgi_off;
2075 lgi->lgi_buf.lb_buf = idarray;
2076 lgi->lgi_buf.lb_len = size;
2077 rc = dt_record_read(env, o, &lgi->lgi_buf, &lgi->lgi_off);
2078 /* -EFAULT means the llog is a sparse file. This is not an error
2079 * after arbitrary OST index is supported. */
2080 if (rc < 0 && rc != -EFAULT) {
2081 CERROR("%s: error reading CATALOGS: rc = %d\n",
2082 o->do_lu.lo_dev->ld_obd->obd_name, rc);
2088 dt_object_put(env, o);
2091 EXPORT_SYMBOL(llog_osd_get_cat_list);
2094 * Write the special file which contains the list of llog catalogs IDs
2096 * This function writes the CATALOG file which contains the array of llog
2097 * catalogs IDs. It is used mostly to store OSP llogs indexed by OST/MDT
2100 * \param[in] env execution environment
2101 * \param[in] d corresponding storage device
2102 * \param[in] idx position to start from, usually OST/MDT index
2103 * \param[in] count how many catalog IDs to write
2104 * \param[out] idarray the buffer with the data to write.
2105 * \param[in] fid LLOG_CATALOGS_OID for CATALOG object
2107 * \retval 0 on successful write of catalog IDs
2108 * \retval negative value on error
2110 int llog_osd_put_cat_list(const struct lu_env *env, struct dt_device *d,
2111 int idx, int count, struct llog_catid *idarray,
2112 const struct lu_fid *fid)
2114 struct llog_thread_info *lgi = llog_info(env);
2115 struct dt_object *o = NULL;
2124 size = sizeof(*idarray) * count;
2125 lgi->lgi_off = idx * sizeof(*idarray);
2126 lgi->lgi_fid = *fid;
2128 o = dt_locate(env, d, &lgi->lgi_fid);
2132 if (!dt_object_exists(o))
2133 GOTO(out, rc = -ENOENT);
2135 rc = dt_attr_get(env, o, &lgi->lgi_attr);
2139 if (!S_ISREG(lgi->lgi_attr.la_mode)) {
2140 CERROR("%s: CATALOGS is not a regular file!: mode = %o\n",
2141 o->do_lu.lo_dev->ld_obd->obd_name,
2142 lgi->lgi_attr.la_mode);
2143 GOTO(out, rc = -ENOENT);
2146 th = dt_trans_create(env, d);
2148 GOTO(out, rc = PTR_ERR(th));
2150 lgi->lgi_buf.lb_len = size;
2151 lgi->lgi_buf.lb_buf = idarray;
2152 rc = dt_declare_record_write(env, o, &lgi->lgi_buf, lgi->lgi_off, th);
2154 GOTO(out_trans, rc);
2156 /* For update log, this happens during initialization,
2157 * see lod_sub_prep_llog(), and we need make sure catlog
2158 * file ID is written to catlist file(committed) before
2159 * cross-MDT operation write update records to catlog FILE,
2160 * otherwise, during failover these update records might
2162 if (fid_is_update_log(fid))
2165 rc = dt_trans_start_local(env, d, th);
2167 GOTO(out_trans, rc);
2169 th->th_wait_submit = 1;
2171 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
2173 CDEBUG(D_INODE, "can't write CATALOGS at index %d: rc = %d\n",
2176 dt_trans_stop(env, d, th);
2178 dt_object_put(env, o);
2181 EXPORT_SYMBOL(llog_osd_put_cat_list);