2 * Copyright (c) 2008 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * $DragonFly: src/sys/vfs/hammer/hammer_mirror.c,v 1.17 2008/07/31 22:30:33 dillon Exp $
37 * HAMMER mirroring ioctls - serialize and deserialize modifications made
43 static int hammer_mirror_check(hammer_cursor_t cursor,
44 struct hammer_ioc_mrecord_rec *mrec);
45 static int hammer_mirror_update(hammer_cursor_t cursor,
46 struct hammer_ioc_mrecord_rec *mrec);
47 static int hammer_ioc_mirror_write_rec(hammer_cursor_t cursor,
48 struct hammer_ioc_mrecord_rec *mrec,
49 struct hammer_ioc_mirror_rw *mirror,
50 u_int32_t localization,
52 static int hammer_ioc_mirror_write_pass(hammer_cursor_t cursor,
53 struct hammer_ioc_mrecord_rec *mrec,
54 struct hammer_ioc_mirror_rw *mirror,
55 u_int32_t localization);
56 static int hammer_ioc_mirror_write_skip(hammer_cursor_t cursor,
57 struct hammer_ioc_mrecord_skip *mrec,
58 struct hammer_ioc_mirror_rw *mirror,
59 u_int32_t localization);
60 static int hammer_mirror_delete_to(hammer_cursor_t cursor,
61 struct hammer_ioc_mirror_rw *mirror);
62 static int hammer_mirror_nomirror(struct hammer_base_elm *base);
65 * All B-Tree records within the specified key range which also conform
66 * to the transaction id range are returned. Mirroring code keeps track
67 * of the last transaction id fully scanned and can efficiently pick up
68 * where it left off if interrupted.
70 * The PFS is identified in the mirror structure. The passed ip is just
71 * some directory in the overall HAMMER filesystem and has nothing to
75 hammer_ioc_mirror_read(hammer_transaction_t trans, hammer_inode_t ip,
76 struct hammer_ioc_mirror_rw *mirror)
78 struct hammer_cmirror cmirror;
79 struct hammer_cursor cursor;
80 union hammer_ioc_mrecord_any mrec;
81 hammer_btree_leaf_elm_t elm;
82 const int crc_start = HAMMER_MREC_CRCOFF;
89 u_int32_t localization;
92 localization = (u_int32_t)mirror->pfs_id << 16;
94 if ((mirror->key_beg.localization | mirror->key_end.localization) &
95 HAMMER_LOCALIZE_PSEUDOFS_MASK) {
98 if (hammer_btree_cmp(&mirror->key_beg, &mirror->key_end) > 0)
101 mirror->key_cur = mirror->key_beg;
102 mirror->key_cur.localization &= HAMMER_LOCALIZE_MASK;
103 mirror->key_cur.localization += localization;
104 bzero(&mrec, sizeof(mrec));
105 bzero(&cmirror, sizeof(cmirror));
108 * Make CRC errors non-fatal (at least on data), causing an EDOM
109 * error instead of EIO.
111 trans->flags |= HAMMER_TRANSF_CRCDOM;
114 error = hammer_init_cursor(trans, &cursor, NULL, NULL);
116 hammer_done_cursor(&cursor);
119 cursor.key_beg = mirror->key_cur;
120 cursor.key_end = mirror->key_end;
121 cursor.key_end.localization &= HAMMER_LOCALIZE_MASK;
122 cursor.key_end.localization += localization;
124 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
125 cursor.flags |= HAMMER_CURSOR_BACKEND;
128 * This flag filters the search to only return elements whos create
129 * or delete TID is >= mirror_tid. The B-Tree uses the mirror_tid
130 * field stored with internal and leaf nodes to shortcut the scan.
132 cursor.flags |= HAMMER_CURSOR_MIRROR_FILTERED;
133 cursor.cmirror = &cmirror;
134 cmirror.mirror_tid = mirror->tid_beg;
136 error = hammer_btree_first(&cursor);
139 * Yield to more important tasks
142 error = hammer_signal_check(trans->hmp);
148 * An internal node can be returned in mirror-filtered
149 * mode and indicates that the scan is returning a skip
150 * range in the cursor->cmirror structure.
152 uptr = (char *)mirror->ubuf + mirror->count;
153 if (cursor.node->ondisk->type == HAMMER_BTREE_TYPE_INTERNAL) {
157 mirror->key_cur = cmirror.skip_beg;
158 bytes = sizeof(mrec.skip);
159 if (mirror->count + HAMMER_HEAD_DOALIGN(bytes) >
167 mrec.head.signature = HAMMER_IOC_MIRROR_SIGNATURE;
168 mrec.head.type = HAMMER_MREC_TYPE_SKIP;
169 mrec.head.rec_size = bytes;
170 mrec.skip.skip_beg = cmirror.skip_beg;
171 mrec.skip.skip_end = cmirror.skip_end;
172 mrec.head.rec_crc = crc32(&mrec.head.rec_size,
174 error = copyout(&mrec, uptr, bytes);
180 * Leaf node. In full-history mode we could filter out
181 * elements modified outside the user-requested TID range.
183 * However, such elements must be returned so the writer
184 * can compare them against the target to determine what
185 * needs to be deleted on the target, particular for
186 * no-history mirrors.
188 KKASSERT(cursor.node->ondisk->type == HAMMER_BTREE_TYPE_LEAF);
189 elm = &cursor.node->ondisk->elms[cursor.index].leaf;
190 mirror->key_cur = elm->base;
193 * If the record was created after our end point we just
196 if (elm->base.create_tid > mirror->tid_end) {
204 * Determine if we should generate a PASS or a REC. PASS
205 * records are records without any data payload. Such
206 * records will be generated if the target is already expected
207 * to have the record, allowing it to delete the gaps.
209 * A PASS record is also used to perform deletions on the
212 * Such deletions are needed if the master or files on the
213 * master are no-history, or if the slave is so far behind
214 * the master has already been pruned.
216 if (elm->base.create_tid < mirror->tid_beg) {
217 bytes = sizeof(mrec.rec);
218 if (mirror->count + HAMMER_HEAD_DOALIGN(bytes) >
226 mrec.head.signature = HAMMER_IOC_MIRROR_SIGNATURE;
227 mrec.head.type = HAMMER_MREC_TYPE_PASS;
228 mrec.head.rec_size = bytes;
229 mrec.rec.leaf = *elm;
230 mrec.head.rec_crc = crc32(&mrec.head.rec_size,
232 error = copyout(&mrec, uptr, bytes);
239 * The core code exports the data to userland.
241 * CRC errors on data are reported but passed through,
242 * but the data must be washed by the user program.
245 data_len = (elm->data_offset) ? elm->data_len : 0;
247 error = hammer_btree_extract(&cursor,
248 HAMMER_CURSOR_GET_DATA);
252 mrec_flags |= HAMMER_MRECF_CRC_ERROR |
253 HAMMER_MRECF_DATA_CRC_BAD;
257 bytes = sizeof(mrec.rec) + data_len;
258 if (mirror->count + HAMMER_HEAD_DOALIGN(bytes) > mirror->size)
262 * Construct the record for userland and copyout.
264 * The user is asking for a snapshot, if the record was
265 * deleted beyond the user-requested ending tid, the record
266 * is not considered deleted from the point of view of
267 * userland and delete_tid is cleared.
269 mrec.head.signature = HAMMER_IOC_MIRROR_SIGNATURE;
270 mrec.head.type = HAMMER_MREC_TYPE_REC | mrec_flags;
271 mrec.head.rec_size = bytes;
272 mrec.rec.leaf = *elm;
274 if (elm->base.delete_tid > mirror->tid_end)
275 mrec.rec.leaf.base.delete_tid = 0;
276 rec_crc = crc32(&mrec.head.rec_size,
277 sizeof(mrec.rec) - crc_start);
279 rec_crc = crc32_ext(cursor.data, data_len, rec_crc);
280 mrec.head.rec_crc = rec_crc;
281 error = copyout(&mrec, uptr, sizeof(mrec.rec));
282 if (data_len && error == 0) {
283 error = copyout(cursor.data, uptr + sizeof(mrec.rec),
289 * eatdisk controls whether we skip the current cursor
290 * position on the next scan or not. If doing a SKIP
291 * the cursor is already positioned properly for the next
292 * scan and eatdisk will be 0.
296 mirror->count += HAMMER_HEAD_DOALIGN(bytes);
298 cursor.flags |= HAMMER_CURSOR_ATEDISK;
300 cursor.flags &= ~HAMMER_CURSOR_ATEDISK;
301 error = hammer_btree_iterate(&cursor);
304 if (error == ENOENT) {
305 mirror->key_cur = mirror->key_end;
308 hammer_done_cursor(&cursor);
309 if (error == EDEADLK)
311 if (error == EINTR) {
312 mirror->head.flags |= HAMMER_IOC_HEAD_INTR;
316 mirror->key_cur.localization &= HAMMER_LOCALIZE_MASK;
321 * Copy records from userland to the target mirror.
323 * The PFS is identified in the mirror structure. The passed ip is just
324 * some directory in the overall HAMMER filesystem and has nothing to
325 * do with the PFS. In fact, there might not even be a root directory for
329 hammer_ioc_mirror_write(hammer_transaction_t trans, hammer_inode_t ip,
330 struct hammer_ioc_mirror_rw *mirror)
332 union hammer_ioc_mrecord_any mrec;
333 struct hammer_cursor cursor;
334 u_int32_t localization;
335 int checkspace_count = 0;
341 localization = (u_int32_t)mirror->pfs_id << 16;
342 seq = trans->hmp->flusher.act;
345 * Validate the mirror structure and relocalize the tracking keys.
347 if (mirror->size < 0 || mirror->size > 0x70000000)
349 mirror->key_beg.localization &= HAMMER_LOCALIZE_MASK;
350 mirror->key_beg.localization += localization;
351 mirror->key_end.localization &= HAMMER_LOCALIZE_MASK;
352 mirror->key_end.localization += localization;
353 mirror->key_cur.localization &= HAMMER_LOCALIZE_MASK;
354 mirror->key_cur.localization += localization;
357 * Set up our tracking cursor for the loop. The tracking cursor
358 * is used to delete records that are no longer present on the
359 * master. The last handled record at key_cur must be skipped.
361 error = hammer_init_cursor(trans, &cursor, NULL, NULL);
363 cursor.key_beg = mirror->key_cur;
364 cursor.key_end = mirror->key_end;
365 cursor.flags |= HAMMER_CURSOR_BACKEND;
366 error = hammer_btree_first(&cursor);
368 cursor.flags |= HAMMER_CURSOR_ATEDISK;
373 * Loop until our input buffer has been exhausted.
376 mirror->count + sizeof(mrec.head) <= mirror->size) {
379 * Don't blow out the buffer cache. Leave room for frontend
382 * WARNING: See warnings in hammer_unlock_cursor() function.
384 while (hammer_flusher_meta_halflimit(trans->hmp) ||
385 hammer_flusher_undo_exhausted(trans, 2)) {
386 hammer_unlock_cursor(&cursor);
387 hammer_flusher_wait(trans->hmp, seq);
388 hammer_lock_cursor(&cursor);
389 seq = hammer_flusher_async_one(trans->hmp);
393 * If there is insufficient free space it may be due to
394 * reserved bigblocks, which flushing might fix.
396 if (hammer_checkspace(trans->hmp, HAMMER_CHKSPC_MIRROR)) {
397 if (++checkspace_count == 10) {
401 hammer_unlock_cursor(&cursor);
402 hammer_flusher_wait(trans->hmp, seq);
403 hammer_lock_cursor(&cursor);
404 seq = hammer_flusher_async(trans->hmp, NULL);
409 * Acquire and validate header
411 if ((bytes = mirror->size - mirror->count) > sizeof(mrec))
412 bytes = sizeof(mrec);
413 uptr = (char *)mirror->ubuf + mirror->count;
414 error = copyin(uptr, &mrec, bytes);
417 if (mrec.head.signature != HAMMER_IOC_MIRROR_SIGNATURE) {
421 if (mrec.head.rec_size < sizeof(mrec.head) ||
422 mrec.head.rec_size > sizeof(mrec) + HAMMER_XBUFSIZE ||
423 mirror->count + mrec.head.rec_size > mirror->size) {
428 switch(mrec.head.type & HAMMER_MRECF_TYPE_MASK) {
429 case HAMMER_MREC_TYPE_SKIP:
430 if (mrec.head.rec_size != sizeof(mrec.skip))
433 error = hammer_ioc_mirror_write_skip(&cursor, &mrec.skip, mirror, localization);
435 case HAMMER_MREC_TYPE_REC:
436 if (mrec.head.rec_size < sizeof(mrec.rec))
439 error = hammer_ioc_mirror_write_rec(&cursor, &mrec.rec, mirror, localization, uptr + sizeof(mrec.rec));
441 case HAMMER_MREC_TYPE_REC_BADCRC:
443 * Records with bad data payloads are ignored XXX.
445 if (mrec.head.rec_size < sizeof(mrec.rec))
448 case HAMMER_MREC_TYPE_PASS:
449 if (mrec.head.rec_size != sizeof(mrec.rec))
452 error = hammer_ioc_mirror_write_pass(&cursor, &mrec.rec, mirror, localization);
460 * Retry the current record on deadlock, otherwise setup
463 if (error == EDEADLK) {
464 while (error == EDEADLK) {
465 hammer_sync_lock_sh(trans);
466 hammer_recover_cursor(&cursor);
467 error = hammer_cursor_upgrade(&cursor);
468 hammer_sync_unlock(trans);
471 if (error == EALREADY)
475 HAMMER_HEAD_DOALIGN(mrec.head.rec_size);
479 hammer_done_cursor(&cursor);
485 mirror->head.flags |= HAMMER_IOC_HEAD_ERROR;
486 mirror->head.error = error;
490 * ioctls don't update the RW data structure if an error is returned,
497 * Handle skip records.
499 * We must iterate from the last resolved record position at mirror->key_cur
500 * to skip_beg non-inclusive and delete any records encountered.
502 * mirror->key_cur must be carefully set when we succeed in processing
506 hammer_ioc_mirror_write_skip(hammer_cursor_t cursor,
507 struct hammer_ioc_mrecord_skip *mrec,
508 struct hammer_ioc_mirror_rw *mirror,
509 u_int32_t localization)
514 * Relocalize the skip range
516 mrec->skip_beg.localization &= HAMMER_LOCALIZE_MASK;
517 mrec->skip_beg.localization += localization;
518 mrec->skip_end.localization &= HAMMER_LOCALIZE_MASK;
519 mrec->skip_end.localization += localization;
522 * Iterate from current position to skip_beg, deleting any records
523 * we encounter. The record at skip_beg is not included (it is
526 cursor->key_end = mrec->skip_beg;
527 cursor->flags &= ~HAMMER_CURSOR_END_INCLUSIVE;
528 cursor->flags |= HAMMER_CURSOR_BACKEND;
529 error = hammer_mirror_delete_to(cursor, mirror);
532 * Now skip past the skip (which is the whole point point of
533 * having a skip record). The sender has not sent us any records
534 * for the skip area so we wouldn't know what to keep and what
537 * Clear ATEDISK because skip_end is non-inclusive, so we can't
538 * count an exact match if we happened to get one.
541 mirror->key_cur = mrec->skip_end;
542 cursor->key_beg = mrec->skip_end;
543 error = hammer_btree_lookup(cursor);
544 cursor->flags &= ~HAMMER_CURSOR_ATEDISK;
552 * Handle B-Tree records.
554 * We must iterate to mrec->base.key (non-inclusively), and then process
555 * the record. We are allowed to write a new record or delete an existing
556 * record, but cannot replace an existing record.
558 * mirror->key_cur must be carefully set when we succeed in processing
562 hammer_ioc_mirror_write_rec(hammer_cursor_t cursor,
563 struct hammer_ioc_mrecord_rec *mrec,
564 struct hammer_ioc_mirror_rw *mirror,
565 u_int32_t localization,
568 hammer_transaction_t trans;
572 trans = cursor->trans;
573 rec_crc = crc32(mrec, sizeof(*mrec));
575 if (mrec->leaf.data_len < 0 ||
576 mrec->leaf.data_len > HAMMER_XBUFSIZE ||
577 mrec->leaf.data_len + sizeof(*mrec) > mrec->head.rec_size) {
582 * Re-localize for target. relocalization of data is handled
583 * by hammer_mirror_write().
585 mrec->leaf.base.localization &= HAMMER_LOCALIZE_MASK;
586 mrec->leaf.base.localization += localization;
589 * Delete records through until we reach (non-inclusively) the
592 cursor->key_end = mrec->leaf.base;
593 cursor->flags &= ~HAMMER_CURSOR_END_INCLUSIVE;
594 cursor->flags |= HAMMER_CURSOR_BACKEND;
595 error = hammer_mirror_delete_to(cursor, mirror);
598 * Certain records are not part of the mirroring operation
600 if (hammer_mirror_nomirror(&mrec->leaf.base))
606 * If the record exists only the delete_tid may be updated.
608 * If the record does not exist we can create it only if the
609 * create_tid is not too old. If the create_tid is too old
610 * it may have already been destroyed on the slave from pruning.
612 * Note that mirror operations are effectively as-of operations
613 * and delete_tid can be 0 for mirroring purposes even if it is
614 * not actually 0 at the originator.
616 * These functions can return EDEADLK
618 cursor->key_beg = mrec->leaf.base;
619 cursor->flags |= HAMMER_CURSOR_BACKEND;
620 cursor->flags &= ~HAMMER_CURSOR_INSERT;
621 error = hammer_btree_lookup(cursor);
623 if (error == 0 && hammer_mirror_check(cursor, mrec)) {
624 error = hammer_mirror_update(cursor, mrec);
625 } else if (error == ENOENT) {
626 if (mrec->leaf.base.create_tid >= mirror->tid_beg) {
627 error = hammer_create_at_cursor(
629 uptr, HAMMER_CREATE_MODE_UMIRROR);
634 if (error == 0 || error == EALREADY)
635 mirror->key_cur = mrec->leaf.base;
640 * This works like write_rec but no write or update is necessary,
641 * and no data payload is included so we couldn't do a write even
644 * We must still iterate for deletions, and we can validate the
645 * record header which is a good way to test for corrupted mirror
648 * mirror->key_cur must be carefully set when we succeed in processing
653 hammer_ioc_mirror_write_pass(hammer_cursor_t cursor,
654 struct hammer_ioc_mrecord_rec *mrec,
655 struct hammer_ioc_mirror_rw *mirror,
656 u_int32_t localization)
658 hammer_transaction_t trans;
662 trans = cursor->trans;
663 rec_crc = crc32(mrec, sizeof(*mrec));
666 * Re-localize for target. Relocalization of data is handled
667 * by hammer_mirror_write().
669 mrec->leaf.base.localization &= HAMMER_LOCALIZE_MASK;
670 mrec->leaf.base.localization += localization;
673 * Delete records through until we reach (non-inclusively) the
676 cursor->key_end = mrec->leaf.base;
677 cursor->flags &= ~HAMMER_CURSOR_END_INCLUSIVE;
678 cursor->flags |= HAMMER_CURSOR_BACKEND;
679 error = hammer_mirror_delete_to(cursor, mirror);
682 * Certain records are not part of the mirroring operation
684 if (hammer_mirror_nomirror(&mrec->leaf.base))
688 * Locate the record and get past it by setting ATEDISK. Perform
689 * any necessary deletions. We have no data payload and cannot
690 * create a new record.
693 mirror->key_cur = mrec->leaf.base;
694 cursor->key_beg = mrec->leaf.base;
695 cursor->flags |= HAMMER_CURSOR_BACKEND;
696 cursor->flags &= ~HAMMER_CURSOR_INSERT;
697 error = hammer_btree_lookup(cursor);
699 if (hammer_mirror_check(cursor, mrec))
700 error = hammer_mirror_update(cursor, mrec);
701 cursor->flags |= HAMMER_CURSOR_ATEDISK;
703 cursor->flags &= ~HAMMER_CURSOR_ATEDISK;
712 * As part of the mirror write we iterate across swaths of records
713 * on the target which no longer exist on the source, and mark them
716 * The caller has indexed the cursor and set up key_end. We iterate
717 * through to key_end.
719 * There is an edge case where the master has deleted a record whos
720 * create_tid exactly matches our end_tid. We cannot delete this
721 * record on the slave yet because we cannot assign delete_tid == create_tid.
722 * The deletion should be picked up on the next sequence since in order
723 * to have been deleted on the master a transaction must have occured with
724 * a TID greater then the create_tid of the record.
726 * To support incremental re-mirroring, just for robustness, we do not
727 * touch any records created beyond (or equal to) mirror->tid_end.
731 hammer_mirror_delete_to(hammer_cursor_t cursor,
732 struct hammer_ioc_mirror_rw *mirror)
734 hammer_btree_leaf_elm_t elm;
737 error = hammer_btree_iterate(cursor);
739 elm = &cursor->node->ondisk->elms[cursor->index].leaf;
740 KKASSERT(elm->base.btype == HAMMER_BTREE_TYPE_RECORD);
741 cursor->flags |= HAMMER_CURSOR_ATEDISK;
744 * Certain records are not part of the mirroring operation
746 if (hammer_mirror_nomirror(&elm->base)) {
747 error = hammer_btree_iterate(cursor);
752 * Note: Must still delete records with create_tid < tid_beg,
753 * as record may have been pruned-away on source.
755 if (elm->base.delete_tid == 0 &&
756 elm->base.create_tid < mirror->tid_end) {
757 error = hammer_delete_at_cursor(cursor,
758 HAMMER_DELETE_ADJUST,
764 error = hammer_btree_iterate(cursor);
772 * Check whether an update is needed in the case where a match already
773 * exists on the target. The only type of update allowed in this case
774 * is an update of the delete_tid.
776 * Return non-zero if the update should proceed.
780 hammer_mirror_check(hammer_cursor_t cursor, struct hammer_ioc_mrecord_rec *mrec)
782 hammer_btree_leaf_elm_t leaf = cursor->leaf;
784 if (leaf->base.delete_tid != mrec->leaf.base.delete_tid) {
785 if (mrec->leaf.base.delete_tid != 0)
792 * Filter out records which are never mirrored, such as configuration space
793 * records (for hammer cleanup).
795 * NOTE: We currently allow HAMMER_RECTYPE_SNAPSHOT records to be mirrored.
799 hammer_mirror_nomirror(struct hammer_base_elm *base)
802 * Certain types of records are never updated when mirroring.
803 * Slaves have their own configuration space.
805 if (base->rec_type == HAMMER_RECTYPE_CONFIG)
812 * Update a record in-place. Only the delete_tid can change, and
813 * only from zero to non-zero.
817 hammer_mirror_update(hammer_cursor_t cursor,
818 struct hammer_ioc_mrecord_rec *mrec)
823 * This case shouldn't occur.
825 if (mrec->leaf.base.delete_tid == 0)
829 * Mark the record deleted on the mirror target.
831 error = hammer_delete_at_cursor(cursor, HAMMER_DELETE_ADJUST,
832 mrec->leaf.base.delete_tid,
833 mrec->leaf.delete_ts,
835 cursor->flags |= HAMMER_CURSOR_ATEDISK;
841 * MOVED TO HAMMER_OBJECT.C: hammer_create_at_cursor()
844 static int hammer_mirror_localize_data(hammer_data_ondisk_t data,
845 hammer_btree_leaf_elm_t leaf);
848 * Write out a new record.
852 hammer_mirror_write(hammer_cursor_t cursor,
853 struct hammer_ioc_mrecord_rec *mrec,
856 hammer_transaction_t trans;
857 hammer_buffer_t data_buffer;
858 hammer_off_t ndata_offset;
859 hammer_tid_t high_tid;
864 trans = cursor->trans;
868 * Get the sync lock so the whole mess is atomic
870 hammer_sync_lock_sh(trans);
873 * Allocate and adjust data
875 if (mrec->leaf.data_len && mrec->leaf.data_offset) {
876 ndata = hammer_alloc_data(trans, mrec->leaf.data_len,
877 mrec->leaf.base.rec_type,
878 &ndata_offset, &data_buffer,
882 mrec->leaf.data_offset = ndata_offset;
883 hammer_modify_buffer(trans, data_buffer, NULL, 0);
884 error = copyin(udata, ndata, mrec->leaf.data_len);
886 if (hammer_crc_test_leaf(ndata, &mrec->leaf) == 0) {
887 kprintf("data crc mismatch on pipe\n");
890 error = hammer_mirror_localize_data(
894 hammer_modify_buffer_done(data_buffer);
896 mrec->leaf.data_offset = 0;
904 * Do the insertion. This can fail with a EDEADLK or EALREADY
906 cursor->flags |= HAMMER_CURSOR_INSERT;
907 error = hammer_btree_lookup(cursor);
908 if (error != ENOENT) {
914 error = hammer_btree_insert(cursor, &mrec->leaf, &doprop);
917 * Cursor is left on the current element, we want to skip it now.
919 cursor->flags |= HAMMER_CURSOR_ATEDISK;
920 cursor->flags &= ~HAMMER_CURSOR_INSERT;
923 * Track a count of active inodes.
926 mrec->leaf.base.rec_type == HAMMER_RECTYPE_INODE &&
927 mrec->leaf.base.delete_tid == 0) {
928 hammer_modify_volume_field(trans,
931 ++trans->hmp->rootvol->ondisk->vol0_stat_inodes;
932 hammer_modify_volume_done(trans->rootvol);
936 * vol0_next_tid must track the highest TID stored in the filesystem.
937 * We do not need to generate undo for this update.
939 high_tid = mrec->leaf.base.create_tid;
940 if (high_tid < mrec->leaf.base.delete_tid)
941 high_tid = mrec->leaf.base.delete_tid;
942 if (trans->rootvol->ondisk->vol0_next_tid < high_tid) {
943 hammer_modify_volume(trans, trans->rootvol, NULL, 0);
944 trans->rootvol->ondisk->vol0_next_tid = high_tid;
945 hammer_modify_volume_done(trans->rootvol);
949 * WARNING! cursor's leaf pointer may have changed after
950 * do_propagation returns.
952 if (error == 0 && doprop)
953 hammer_btree_do_propagation(cursor, NULL, &mrec->leaf);
959 if (error && mrec->leaf.data_offset) {
960 hammer_blockmap_free(cursor->trans,
961 mrec->leaf.data_offset,
962 mrec->leaf.data_len);
964 hammer_sync_unlock(trans);
966 hammer_rel_buffer(data_buffer, 0);
971 * Localize the data payload. Directory entries may need their
972 * localization adjusted.
976 hammer_mirror_localize_data(hammer_data_ondisk_t data,
977 hammer_btree_leaf_elm_t leaf)
979 u_int32_t localization;
981 if (leaf->base.rec_type == HAMMER_RECTYPE_DIRENTRY) {
982 localization = leaf->base.localization &
983 HAMMER_LOCALIZE_PSEUDOFS_MASK;
984 if (data->entry.localization != localization) {
985 data->entry.localization = localization;
986 hammer_crc_set_leaf(data, leaf);