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_mirror_write(hammer_cursor_t cursor,
48 struct hammer_ioc_mrecord_rec *mrec,
50 static int hammer_ioc_mirror_write_rec(hammer_cursor_t cursor,
51 struct hammer_ioc_mrecord_rec *mrec,
52 struct hammer_ioc_mirror_rw *mirror,
53 u_int32_t localization,
55 static int hammer_ioc_mirror_write_pass(hammer_cursor_t cursor,
56 struct hammer_ioc_mrecord_rec *mrec,
57 struct hammer_ioc_mirror_rw *mirror,
58 u_int32_t localization);
59 static int hammer_ioc_mirror_write_skip(hammer_cursor_t cursor,
60 struct hammer_ioc_mrecord_skip *mrec,
61 struct hammer_ioc_mirror_rw *mirror,
62 u_int32_t localization);
63 static int hammer_mirror_delete_to(hammer_cursor_t cursor,
64 struct hammer_ioc_mirror_rw *mirror);
65 static int hammer_mirror_localize_data(hammer_data_ondisk_t data,
66 hammer_btree_leaf_elm_t leaf);
69 * All B-Tree records within the specified key range which also conform
70 * to the transaction id range are returned. Mirroring code keeps track
71 * of the last transaction id fully scanned and can efficiently pick up
72 * where it left off if interrupted.
74 * The PFS is identified in the mirror structure. The passed ip is just
75 * some directory in the overall HAMMER filesystem and has nothing to
79 hammer_ioc_mirror_read(hammer_transaction_t trans, hammer_inode_t ip,
80 struct hammer_ioc_mirror_rw *mirror)
82 struct hammer_cmirror cmirror;
83 struct hammer_cursor cursor;
84 union hammer_ioc_mrecord_any mrec;
85 hammer_btree_leaf_elm_t elm;
86 const int crc_start = HAMMER_MREC_CRCOFF;
93 u_int32_t localization;
96 localization = (u_int32_t)mirror->pfs_id << 16;
98 if ((mirror->key_beg.localization | mirror->key_end.localization) &
99 HAMMER_LOCALIZE_PSEUDOFS_MASK) {
102 if (hammer_btree_cmp(&mirror->key_beg, &mirror->key_end) > 0)
105 mirror->key_cur = mirror->key_beg;
106 mirror->key_cur.localization &= HAMMER_LOCALIZE_MASK;
107 mirror->key_cur.localization += localization;
108 bzero(&mrec, sizeof(mrec));
109 bzero(&cmirror, sizeof(cmirror));
112 * Make CRC errors non-fatal (at least on data), causing an EDOM
113 * error instead of EIO.
115 trans->flags |= HAMMER_TRANSF_CRCDOM;
118 error = hammer_init_cursor(trans, &cursor, NULL, NULL);
120 hammer_done_cursor(&cursor);
123 cursor.key_beg = mirror->key_cur;
124 cursor.key_end = mirror->key_end;
125 cursor.key_end.localization &= HAMMER_LOCALIZE_MASK;
126 cursor.key_end.localization += localization;
128 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
129 cursor.flags |= HAMMER_CURSOR_BACKEND;
132 * This flag filters the search to only return elements whos create
133 * or delete TID is >= mirror_tid. The B-Tree uses the mirror_tid
134 * field stored with internal and leaf nodes to shortcut the scan.
136 cursor.flags |= HAMMER_CURSOR_MIRROR_FILTERED;
137 cursor.cmirror = &cmirror;
138 cmirror.mirror_tid = mirror->tid_beg;
140 error = hammer_btree_first(&cursor);
143 * Yield to more important tasks
146 error = hammer_signal_check(trans->hmp);
152 * An internal node can be returned in mirror-filtered
153 * mode and indicates that the scan is returning a skip
154 * range in the cursor->cmirror structure.
156 uptr = (char *)mirror->ubuf + mirror->count;
157 if (cursor.node->ondisk->type == HAMMER_BTREE_TYPE_INTERNAL) {
161 mirror->key_cur = cmirror.skip_beg;
162 bytes = sizeof(mrec.skip);
163 if (mirror->count + HAMMER_HEAD_DOALIGN(bytes) >
171 mrec.head.signature = HAMMER_IOC_MIRROR_SIGNATURE;
172 mrec.head.type = HAMMER_MREC_TYPE_SKIP;
173 mrec.head.rec_size = bytes;
174 mrec.skip.skip_beg = cmirror.skip_beg;
175 mrec.skip.skip_end = cmirror.skip_end;
176 mrec.head.rec_crc = crc32(&mrec.head.rec_size,
178 error = copyout(&mrec, uptr, bytes);
184 * Leaf node. In full-history mode we could filter out
185 * elements modified outside the user-requested TID range.
187 * However, such elements must be returned so the writer
188 * can compare them against the target to determine what
189 * needs to be deleted on the target, particular for
190 * no-history mirrors.
192 KKASSERT(cursor.node->ondisk->type == HAMMER_BTREE_TYPE_LEAF);
193 elm = &cursor.node->ondisk->elms[cursor.index].leaf;
194 mirror->key_cur = elm->base;
197 * If the record was created after our end point we just
200 if (elm->base.create_tid > mirror->tid_end) {
208 * Determine if we should generate a PASS or a REC. PASS
209 * records are records without any data payload. Such
210 * records will be generated if the target is already expected
211 * to have the record, allowing it to delete the gaps.
213 * A PASS record is also used to perform deletions on the
216 * Such deletions are needed if the master or files on the
217 * master are no-history, or if the slave is so far behind
218 * the master has already been pruned.
220 if (elm->base.create_tid < mirror->tid_beg) {
221 bytes = sizeof(mrec.rec);
222 if (mirror->count + HAMMER_HEAD_DOALIGN(bytes) >
230 mrec.head.signature = HAMMER_IOC_MIRROR_SIGNATURE;
231 mrec.head.type = HAMMER_MREC_TYPE_PASS;
232 mrec.head.rec_size = bytes;
233 mrec.rec.leaf = *elm;
234 mrec.head.rec_crc = crc32(&mrec.head.rec_size,
236 error = copyout(&mrec, uptr, bytes);
243 * The core code exports the data to userland.
245 * CRC errors on data are reported but passed through,
246 * but the data must be washed by the user program.
249 data_len = (elm->data_offset) ? elm->data_len : 0;
251 error = hammer_btree_extract(&cursor,
252 HAMMER_CURSOR_GET_DATA);
256 mrec_flags |= HAMMER_MRECF_CRC_ERROR |
257 HAMMER_MRECF_DATA_CRC_BAD;
261 bytes = sizeof(mrec.rec) + data_len;
262 if (mirror->count + HAMMER_HEAD_DOALIGN(bytes) > mirror->size)
266 * Construct the record for userland and copyout.
268 * The user is asking for a snapshot, if the record was
269 * deleted beyond the user-requested ending tid, the record
270 * is not considered deleted from the point of view of
271 * userland and delete_tid is cleared.
273 mrec.head.signature = HAMMER_IOC_MIRROR_SIGNATURE;
274 mrec.head.type = HAMMER_MREC_TYPE_REC | mrec_flags;
275 mrec.head.rec_size = bytes;
276 mrec.rec.leaf = *elm;
278 if (elm->base.delete_tid > mirror->tid_end)
279 mrec.rec.leaf.base.delete_tid = 0;
280 rec_crc = crc32(&mrec.head.rec_size,
281 sizeof(mrec.rec) - crc_start);
283 rec_crc = crc32_ext(cursor.data, data_len, rec_crc);
284 mrec.head.rec_crc = rec_crc;
285 error = copyout(&mrec, uptr, sizeof(mrec.rec));
286 if (data_len && error == 0) {
287 error = copyout(cursor.data, uptr + sizeof(mrec.rec),
293 * eatdisk controls whether we skip the current cursor
294 * position on the next scan or not. If doing a SKIP
295 * the cursor is already positioned properly for the next
296 * scan and eatdisk will be 0.
300 mirror->count += HAMMER_HEAD_DOALIGN(bytes);
302 cursor.flags |= HAMMER_CURSOR_ATEDISK;
304 cursor.flags &= ~HAMMER_CURSOR_ATEDISK;
305 error = hammer_btree_iterate(&cursor);
308 if (error == ENOENT) {
309 mirror->key_cur = mirror->key_end;
312 hammer_done_cursor(&cursor);
313 if (error == EDEADLK)
315 if (error == EINTR) {
316 mirror->head.flags |= HAMMER_IOC_HEAD_INTR;
320 mirror->key_cur.localization &= HAMMER_LOCALIZE_MASK;
325 * Copy records from userland to the target mirror.
327 * The PFS is identified in the mirror structure. The passed ip is just
328 * some directory in the overall HAMMER filesystem and has nothing to
329 * do with the PFS. In fact, there might not even be a root directory for
333 hammer_ioc_mirror_write(hammer_transaction_t trans, hammer_inode_t ip,
334 struct hammer_ioc_mirror_rw *mirror)
336 union hammer_ioc_mrecord_any mrec;
337 struct hammer_cursor cursor;
338 u_int32_t localization;
339 int checkspace_count = 0;
345 localization = (u_int32_t)mirror->pfs_id << 16;
346 seq = trans->hmp->flusher.act;
349 * Validate the mirror structure and relocalize the tracking keys.
351 if (mirror->size < 0 || mirror->size > 0x70000000)
353 mirror->key_beg.localization &= HAMMER_LOCALIZE_MASK;
354 mirror->key_beg.localization += localization;
355 mirror->key_end.localization &= HAMMER_LOCALIZE_MASK;
356 mirror->key_end.localization += localization;
357 mirror->key_cur.localization &= HAMMER_LOCALIZE_MASK;
358 mirror->key_cur.localization += localization;
361 * Set up our tracking cursor for the loop. The tracking cursor
362 * is used to delete records that are no longer present on the
363 * master. The last handled record at key_cur must be skipped.
365 error = hammer_init_cursor(trans, &cursor, NULL, NULL);
367 cursor.key_beg = mirror->key_cur;
368 cursor.key_end = mirror->key_end;
369 cursor.flags |= HAMMER_CURSOR_BACKEND;
370 error = hammer_btree_first(&cursor);
372 cursor.flags |= HAMMER_CURSOR_ATEDISK;
377 * Loop until our input buffer has been exhausted.
380 mirror->count + sizeof(mrec.head) <= mirror->size) {
383 * Don't blow out the buffer cache. Leave room for frontend
386 while (hammer_flusher_meta_halflimit(trans->hmp) ||
387 hammer_flusher_undo_exhausted(trans, 2)) {
388 hammer_unlock_cursor(&cursor);
389 hammer_flusher_wait(trans->hmp, seq);
390 hammer_lock_cursor(&cursor);
391 seq = hammer_flusher_async_one(trans->hmp);
395 * If there is insufficient free space it may be due to
396 * reserved bigblocks, which flushing might fix.
398 if (hammer_checkspace(trans->hmp, HAMMER_CHKSPC_MIRROR)) {
399 if (++checkspace_count == 10) {
403 hammer_unlock_cursor(&cursor);
404 hammer_flusher_wait(trans->hmp, seq);
405 hammer_lock_cursor(&cursor);
406 seq = hammer_flusher_async(trans->hmp, NULL);
411 * Acquire and validate header
413 if ((bytes = mirror->size - mirror->count) > sizeof(mrec))
414 bytes = sizeof(mrec);
415 uptr = (char *)mirror->ubuf + mirror->count;
416 error = copyin(uptr, &mrec, bytes);
419 if (mrec.head.signature != HAMMER_IOC_MIRROR_SIGNATURE) {
423 if (mrec.head.rec_size < sizeof(mrec.head) ||
424 mrec.head.rec_size > sizeof(mrec) + HAMMER_XBUFSIZE ||
425 mirror->count + mrec.head.rec_size > mirror->size) {
430 switch(mrec.head.type & HAMMER_MRECF_TYPE_MASK) {
431 case HAMMER_MREC_TYPE_SKIP:
432 if (mrec.head.rec_size != sizeof(mrec.skip))
435 error = hammer_ioc_mirror_write_skip(&cursor, &mrec.skip, mirror, localization);
437 case HAMMER_MREC_TYPE_REC:
438 if (mrec.head.rec_size < sizeof(mrec.rec))
441 error = hammer_ioc_mirror_write_rec(&cursor, &mrec.rec, mirror, localization, uptr + sizeof(mrec.rec));
443 case HAMMER_MREC_TYPE_REC_BADCRC:
445 * Records with bad data payloads are ignored XXX.
447 if (mrec.head.rec_size < sizeof(mrec.rec))
450 case HAMMER_MREC_TYPE_PASS:
451 if (mrec.head.rec_size != sizeof(mrec.rec))
454 error = hammer_ioc_mirror_write_pass(&cursor, &mrec.rec, mirror, localization);
462 * Retry the current record on deadlock, otherwise setup
465 if (error == EDEADLK) {
466 while (error == EDEADLK) {
467 hammer_recover_cursor(&cursor);
468 error = hammer_cursor_upgrade(&cursor);
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);
600 * If the record exists only the delete_tid may be updated.
602 * If the record does not exist we can create it only if the
603 * create_tid is not too old. If the create_tid is too old
604 * it may have already been destroyed on the slave from pruning.
606 * Note that mirror operations are effectively as-of operations
607 * and delete_tid can be 0 for mirroring purposes even if it is
608 * not actually 0 at the originator.
610 * These functions can return EDEADLK
612 cursor->key_beg = mrec->leaf.base;
613 cursor->flags |= HAMMER_CURSOR_BACKEND;
614 cursor->flags &= ~HAMMER_CURSOR_INSERT;
615 error = hammer_btree_lookup(cursor);
617 if (error == 0 && hammer_mirror_check(cursor, mrec)) {
618 error = hammer_mirror_update(cursor, mrec);
619 } else if (error == ENOENT) {
620 if (mrec->leaf.base.create_tid >= mirror->tid_beg)
621 error = hammer_mirror_write(cursor, mrec, uptr);
625 if (error == 0 || error == EALREADY)
626 mirror->key_cur = mrec->leaf.base;
631 * This works like write_rec but no write or update is necessary,
632 * and no data payload is included so we couldn't do a write even
635 * We must still iterate for deletions, and we can validate the
636 * record header which is a good way to test for corrupted mirror
639 * mirror->key_cur must be carefully set when we succeed in processing
644 hammer_ioc_mirror_write_pass(hammer_cursor_t cursor,
645 struct hammer_ioc_mrecord_rec *mrec,
646 struct hammer_ioc_mirror_rw *mirror,
647 u_int32_t localization)
649 hammer_transaction_t trans;
653 trans = cursor->trans;
654 rec_crc = crc32(mrec, sizeof(*mrec));
657 * Re-localize for target. Relocalization of data is handled
658 * by hammer_mirror_write().
660 mrec->leaf.base.localization &= HAMMER_LOCALIZE_MASK;
661 mrec->leaf.base.localization += localization;
664 * Delete records through until we reach (non-inclusively) the
667 cursor->key_end = mrec->leaf.base;
668 cursor->flags &= ~HAMMER_CURSOR_END_INCLUSIVE;
669 cursor->flags |= HAMMER_CURSOR_BACKEND;
670 error = hammer_mirror_delete_to(cursor, mirror);
673 * Locate the record and get past it by setting ATEDISK. Perform
674 * any necessary deletions. We have no data payload and cannot
675 * create a new record.
678 mirror->key_cur = mrec->leaf.base;
679 cursor->key_beg = mrec->leaf.base;
680 cursor->flags |= HAMMER_CURSOR_BACKEND;
681 cursor->flags &= ~HAMMER_CURSOR_INSERT;
682 error = hammer_btree_lookup(cursor);
684 if (hammer_mirror_check(cursor, mrec))
685 error = hammer_mirror_update(cursor, mrec);
686 cursor->flags |= HAMMER_CURSOR_ATEDISK;
688 cursor->flags &= ~HAMMER_CURSOR_ATEDISK;
697 * As part of the mirror write we iterate across swaths of records
698 * on the target which no longer exist on the source, and mark them
701 * The caller has indexed the cursor and set up key_end. We iterate
702 * through to key_end.
704 * There is an edge case where the master has deleted a record whos
705 * create_tid exactly matches our end_tid. We cannot delete this
706 * record on the slave yet because we cannot assign delete_tid == create_tid.
707 * The deletion should be picked up on the next sequence since in order
708 * to have been deleted on the master a transaction must have occured with
709 * a TID greater then the create_tid of the record.
711 * To support incremental re-mirroring, just for robustness, we do not
712 * touch any records created beyond (or equal to) mirror->tid_end.
716 hammer_mirror_delete_to(hammer_cursor_t cursor,
717 struct hammer_ioc_mirror_rw *mirror)
719 hammer_btree_leaf_elm_t elm;
722 error = hammer_btree_iterate(cursor);
724 elm = &cursor->node->ondisk->elms[cursor->index].leaf;
725 KKASSERT(elm->base.btype == HAMMER_BTREE_TYPE_RECORD);
726 cursor->flags |= HAMMER_CURSOR_ATEDISK;
729 * Note: Must still delete records with create_tid < tid_beg,
730 * as record may have been pruned-away on source.
732 if (elm->base.delete_tid == 0 &&
733 elm->base.create_tid < mirror->tid_end) {
734 error = hammer_delete_at_cursor(cursor,
735 HAMMER_DELETE_ADJUST,
741 error = hammer_btree_iterate(cursor);
749 * Check whether an update is needed in the case where a match already
750 * exists on the target. The only type of update allowed in this case
751 * is an update of the delete_tid.
753 * Return non-zero if the update should proceed.
757 hammer_mirror_check(hammer_cursor_t cursor, struct hammer_ioc_mrecord_rec *mrec)
759 hammer_btree_leaf_elm_t leaf = cursor->leaf;
761 if (leaf->base.delete_tid != mrec->leaf.base.delete_tid) {
762 if (mrec->leaf.base.delete_tid != 0)
769 * Update a record in-place. Only the delete_tid can change, and
770 * only from zero to non-zero.
774 hammer_mirror_update(hammer_cursor_t cursor,
775 struct hammer_ioc_mrecord_rec *mrec)
780 * This case shouldn't occur.
782 if (mrec->leaf.base.delete_tid == 0)
786 * Mark the record deleted on the mirror target.
788 error = hammer_delete_at_cursor(cursor, HAMMER_DELETE_ADJUST,
789 mrec->leaf.base.delete_tid,
790 mrec->leaf.delete_ts,
792 cursor->flags |= HAMMER_CURSOR_ATEDISK;
797 * Write out a new record.
801 hammer_mirror_write(hammer_cursor_t cursor,
802 struct hammer_ioc_mrecord_rec *mrec,
805 hammer_transaction_t trans;
806 hammer_buffer_t data_buffer;
807 hammer_off_t ndata_offset;
808 hammer_tid_t high_tid;
813 trans = cursor->trans;
817 * Get the sync lock so the whole mess is atomic
819 hammer_sync_lock_sh(trans);
822 * Allocate and adjust data
824 if (mrec->leaf.data_len && mrec->leaf.data_offset) {
825 ndata = hammer_alloc_data(trans, mrec->leaf.data_len,
826 mrec->leaf.base.rec_type,
827 &ndata_offset, &data_buffer,
831 mrec->leaf.data_offset = ndata_offset;
832 hammer_modify_buffer(trans, data_buffer, NULL, 0);
833 error = copyin(udata, ndata, mrec->leaf.data_len);
835 if (hammer_crc_test_leaf(ndata, &mrec->leaf) == 0) {
836 kprintf("data crc mismatch on pipe\n");
839 error = hammer_mirror_localize_data(
843 hammer_modify_buffer_done(data_buffer);
845 mrec->leaf.data_offset = 0;
853 * Do the insertion. This can fail with a EDEADLK or EALREADY
855 cursor->flags |= HAMMER_CURSOR_INSERT;
856 error = hammer_btree_lookup(cursor);
857 if (error != ENOENT) {
863 error = hammer_btree_insert(cursor, &mrec->leaf, &doprop);
866 * Cursor is left on the current element, we want to skip it now.
868 cursor->flags |= HAMMER_CURSOR_ATEDISK;
869 cursor->flags &= ~HAMMER_CURSOR_INSERT;
872 * Track a count of active inodes.
875 mrec->leaf.base.rec_type == HAMMER_RECTYPE_INODE &&
876 mrec->leaf.base.delete_tid == 0) {
877 hammer_modify_volume_field(trans,
880 ++trans->hmp->rootvol->ondisk->vol0_stat_inodes;
881 hammer_modify_volume_done(trans->rootvol);
885 * vol0_next_tid must track the highest TID stored in the filesystem.
886 * We do not need to generate undo for this update.
888 high_tid = mrec->leaf.base.create_tid;
889 if (high_tid < mrec->leaf.base.delete_tid)
890 high_tid = mrec->leaf.base.delete_tid;
891 if (trans->rootvol->ondisk->vol0_next_tid < high_tid) {
892 hammer_modify_volume(trans, trans->rootvol, NULL, 0);
893 trans->rootvol->ondisk->vol0_next_tid = high_tid;
894 hammer_modify_volume_done(trans->rootvol);
897 if (error == 0 && doprop)
898 hammer_btree_do_propagation(cursor, NULL, &mrec->leaf);
904 if (error && mrec->leaf.data_offset) {
905 hammer_blockmap_free(cursor->trans,
906 mrec->leaf.data_offset,
907 mrec->leaf.data_len);
909 hammer_sync_unlock(trans);
911 hammer_rel_buffer(data_buffer, 0);
916 * Localize the data payload. Directory entries may need their
917 * localization adjusted.
919 * PFS directory entries must be skipped entirely (return EALREADY).
923 hammer_mirror_localize_data(hammer_data_ondisk_t data,
924 hammer_btree_leaf_elm_t leaf)
926 u_int32_t localization;
928 if (leaf->base.rec_type == HAMMER_RECTYPE_DIRENTRY) {
929 if (data->entry.obj_id == HAMMER_OBJID_ROOT)
931 localization = leaf->base.localization &
932 HAMMER_LOCALIZE_PSEUDOFS_MASK;
933 if (data->entry.localization != localization) {
934 data->entry.localization = localization;
935 hammer_crc_set_leaf(data, leaf);