static
int
hammer_ioc_get_data(hammer_transaction_t trans, hammer_inode_t ip,
struct hammer_ioc_data *data)
{
struct hammer_cursor cursor;
int bytes;
int error;
/* XXX cached inode ? */
error = hammer_init_cursor(trans, &cursor, NULL, NULL);
if (error)
goto failed;
cursor.key_beg = data->elm;
cursor.flags |= HAMMER_CURSOR_BACKEND;
error = hammer_btree_lookup(&cursor);
if (error == 0) {
error = hammer_btree_extract_data(&cursor);
if (error == 0) {
data->leaf = *cursor.leaf;
bytes = cursor.leaf->data_len;
if (bytes > data->size)
bytes = data->size;
error = copyout(cursor.data, data->ubuf, bytes);
}
}
failed:
hammer_done_cursor(&cursor);
return (error);
}
/*
* Reblock a record's data. Both the B-Tree element and record pointers
* to the data must be adjusted.
*/
static int
hammer_reblock_data(struct hammer_ioc_reblock *reblock,
hammer_cursor_t cursor, hammer_btree_elm_t elm)
{
hammer_buffer_t data_buffer = NULL;
hammer_off_t odata_offset;
hammer_off_t ndata_offset;
int error;
void *ndata;
error = hammer_btree_extract_data(cursor);
if (error)
return (error);
ndata = hammer_alloc_data(cursor->trans, elm->leaf.data_len,
elm->leaf.base.rec_type,
&ndata_offset, &data_buffer,
0, &error);
if (error)
goto done;
hammer_io_notmeta(data_buffer);
/*
* Move the data. Note that we must invalidate any cached
* data buffer in the cursor before calling blockmap_free.
* The blockmap_free may free up the entire big-block and
* will not be able to invalidate it if the cursor is holding
* a data buffer cached in that big-block.
*/
hammer_modify_buffer_noundo(cursor->trans, data_buffer);
bcopy(cursor->data, ndata, elm->leaf.data_len);
hammer_modify_buffer_done(data_buffer);
hammer_cursor_invalidate_cache(cursor);
hammer_blockmap_free(cursor->trans,
elm->leaf.data_offset, elm->leaf.data_len);
hammer_modify_node(cursor->trans, cursor->node,
&elm->leaf.data_offset, sizeof(hammer_off_t));
odata_offset = elm->leaf.data_offset;
elm->leaf.data_offset = ndata_offset;
hammer_modify_node_done(cursor->node);
if (hammer_debug_general & 0x4000) {
hdkprintf("%08x %016jx -> %016jx\n",
(elm ? elm->base.localization : -1),
(intmax_t)odata_offset,
(intmax_t)ndata_offset);
}
done:
if (data_buffer)
hammer_rel_buffer(data_buffer, 0);
return (error);
}
/*
* Retrieve the PFS hammer cleanup utility config record. This is
* different (newer than) the PFS config.
*/
static
int
hammer_ioc_get_config(hammer_transaction_t trans, hammer_inode_t ip,
struct hammer_ioc_config *config)
{
struct hammer_cursor cursor;
int error;
error = hammer_init_cursor(trans, &cursor, &ip->cache[0], NULL);
if (error) {
hammer_done_cursor(&cursor);
return(error);
}
cursor.key_beg.obj_id = HAMMER_OBJID_ROOT;
cursor.key_beg.create_tid = 0;
cursor.key_beg.delete_tid = 0;
cursor.key_beg.obj_type = 0;
cursor.key_beg.rec_type = HAMMER_RECTYPE_CONFIG;
cursor.key_beg.localization = ip->obj_localization | HAMMER_LOCALIZE_INODE;
cursor.key_beg.key = 0; /* config space page 0 */
cursor.asof = HAMMER_MAX_TID;
cursor.flags |= HAMMER_CURSOR_ASOF;
error = hammer_btree_lookup(&cursor);
if (error == 0) {
error = hammer_btree_extract_data(&cursor);
if (error == 0)
config->config = cursor.data->config;
}
/* error can be ENOENT */
config->head.error = error;
hammer_done_cursor(&cursor);
return(0);
}
/*
* All B-Tree records within the specified key range which also conform
* to the transaction id range are returned. Mirroring code keeps track
* of the last transaction id fully scanned and can efficiently pick up
* where it left off if interrupted.
*
* The PFS is identified in the mirror structure. The passed ip is just
* some directory in the overall HAMMER filesystem and has nothing to
* do with the PFS.
*/
int
hammer_ioc_mirror_read(hammer_transaction_t trans, hammer_inode_t ip,
struct hammer_ioc_mirror_rw *mirror)
{
struct hammer_cmirror cmirror;
struct hammer_cursor cursor;
union hammer_ioc_mrecord_any mrec;
hammer_btree_leaf_elm_t elm;
char *uptr;
int error;
int data_len;
int bytes;
int eatdisk;
int mrec_flags;
uint32_t localization;
hammer_crc_t rec_crc;
localization = pfs_to_lo(mirror->pfs_id);
if ((mirror->key_beg.localization | mirror->key_end.localization) &
HAMMER_LOCALIZE_PSEUDOFS_MASK) {
return(EINVAL);
}
if (hammer_btree_cmp(&mirror->key_beg, &mirror->key_end) > 0)
return(EINVAL);
mirror->key_cur = mirror->key_beg;
mirror->key_cur.localization &= HAMMER_LOCALIZE_MASK;
mirror->key_cur.localization |= localization;
bzero(&mrec, sizeof(mrec));
bzero(&cmirror, sizeof(cmirror));
/*
* Make CRC errors non-fatal (at least on data), causing an EDOM
* error instead of EIO.
*/
trans->flags |= HAMMER_TRANSF_CRCDOM;
retry:
error = hammer_init_cursor(trans, &cursor, NULL, NULL);
if (error) {
hammer_done_cursor(&cursor);
goto failed;
}
cursor.key_beg = mirror->key_cur;
cursor.key_end = mirror->key_end;
cursor.key_end.localization &= HAMMER_LOCALIZE_MASK;
cursor.key_end.localization |= localization;
cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
cursor.flags |= HAMMER_CURSOR_BACKEND;
/*
* This flag filters the search to only return elements whos create
* or delete TID is >= mirror_tid. The B-Tree uses the mirror_tid
* field stored with internal and leaf nodes to shortcut the scan.
*/
cursor.flags |= HAMMER_CURSOR_MIRROR_FILTERED;
cursor.cmirror = &cmirror;
cmirror.mirror_tid = mirror->tid_beg;
error = hammer_btree_first(&cursor);
while (error == 0) {
/*
* Yield to more important tasks
*/
if (error == 0) {
error = hammer_signal_check(trans->hmp);
if (error)
break;
}
/*
* An internal node can be returned in mirror-filtered
* mode and indicates that the scan is returning a skip
* range in the cursor->cmirror structure.
*/
uptr = (char *)mirror->ubuf + mirror->count;
if (cursor.node->ondisk->type == HAMMER_BTREE_TYPE_INTERNAL) {
/*
* Check space
*/
mirror->key_cur = cmirror.skip_beg;
bytes = sizeof(mrec.skip);
if (mirror->count + HAMMER_HEAD_DOALIGN(bytes) >
mirror->size) {
break;
}
/*
* Fill mrec
*/
mrec.head.signature = HAMMER_IOC_MIRROR_SIGNATURE;
mrec.head.type = HAMMER_MREC_TYPE_SKIP;
mrec.head.rec_size = bytes;
mrec.skip.skip_beg = cmirror.skip_beg;
mrec.skip.skip_end = cmirror.skip_end;
hammer_crc_set_mrec_head(&mrec.head, bytes);
error = copyout(&mrec, uptr, bytes);
eatdisk = 0;
goto didwrite;
}
/*
* Leaf node. In full-history mode we could filter out
* elements modified outside the user-requested TID range.
*
* However, such elements must be returned so the writer
* can compare them against the target to determine what
* needs to be deleted on the target, particular for
* no-history mirrors.
*/
KKASSERT(cursor.node->ondisk->type == HAMMER_BTREE_TYPE_LEAF);
elm = &cursor.node->ondisk->elms[cursor.index].leaf;
mirror->key_cur = elm->base;
/*
* If the record was created after our end point we just
* ignore it.
*/
if (elm->base.create_tid > mirror->tid_end) {
error = 0;
bytes = 0;
eatdisk = 1;
goto didwrite;
}
/*
* Determine if we should generate a PASS or a REC. PASS
* records are records without any data payload. Such
* records will be generated if the target is already expected
* to have the record, allowing it to delete the gaps.
*
* A PASS record is also used to perform deletions on the
* target.
*
* Such deletions are needed if the master or files on the
* master are no-history, or if the slave is so far behind
* the master has already been pruned.
*/
if (elm->base.create_tid < mirror->tid_beg) {
bytes = sizeof(mrec.rec);
if (mirror->count + HAMMER_HEAD_DOALIGN(bytes) >
mirror->size) {
break;
}
/*
* Fill mrec.
*/
mrec.head.signature = HAMMER_IOC_MIRROR_SIGNATURE;
mrec.head.type = HAMMER_MREC_TYPE_PASS;
mrec.head.rec_size = bytes;
mrec.rec.leaf = *elm;
hammer_crc_set_mrec_head(&mrec.head, bytes);
error = copyout(&mrec, uptr, bytes);
eatdisk = 1;
goto didwrite;
}
/*
* The core code exports the data to userland.
*
* CRC errors on data are reported but passed through,
* but the data must be washed by the user program.
*
* If userland just wants the btree records it can
* request that bulk data not be returned. This is
* use during mirror-stream histogram generation.
*/
mrec_flags = 0;
data_len = (elm->data_offset) ? elm->data_len : 0;
if (data_len &&
(mirror->head.flags & HAMMER_IOC_MIRROR_NODATA)) {
data_len = 0;
mrec_flags |= HAMMER_MRECF_NODATA;
}
if (data_len) {
error = hammer_btree_extract_data(&cursor);
if (error) {
if (error != EDOM)
break;
mrec_flags |= HAMMER_MRECF_CRC_ERROR |
HAMMER_MRECF_DATA_CRC_BAD;
}
}
bytes = sizeof(mrec.rec) + data_len;
if (mirror->count + HAMMER_HEAD_DOALIGN(bytes) > mirror->size)
break;
/*
* Construct the record for userland and copyout.
*
* The user is asking for a snapshot, if the record was
* deleted beyond the user-requested ending tid, the record
* is not considered deleted from the point of view of
* userland and delete_tid is cleared.
*/
mrec.head.signature = HAMMER_IOC_MIRROR_SIGNATURE;
mrec.head.type = HAMMER_MREC_TYPE_REC | mrec_flags;
mrec.head.rec_size = bytes;
mrec.rec.leaf = *elm;
if (elm->base.delete_tid > mirror->tid_end)
mrec.rec.leaf.base.delete_tid = 0;
rec_crc = hammer_crc_get_mrec_head(&mrec.head, sizeof(mrec.rec));
if (data_len)
rec_crc = crc32_ext(cursor.data, data_len, rec_crc);
mrec.head.rec_crc = rec_crc;
error = copyout(&mrec, uptr, sizeof(mrec.rec));
if (data_len && error == 0) {
error = copyout(cursor.data, uptr + sizeof(mrec.rec),
data_len);
}
eatdisk = 1;
/*
* eatdisk controls whether we skip the current cursor
* position on the next scan or not. If doing a SKIP
* the cursor is already positioned properly for the next
* scan and eatdisk will be 0.
*/
didwrite:
if (error == 0) {
mirror->count += HAMMER_HEAD_DOALIGN(bytes);
if (eatdisk)
cursor.flags |= HAMMER_CURSOR_ATEDISK;
else
cursor.flags &= ~HAMMER_CURSOR_ATEDISK;
error = hammer_btree_iterate(&cursor);
}
}
if (error == ENOENT) {
mirror->key_cur = mirror->key_end;
error = 0;
}
hammer_done_cursor(&cursor);
if (error == EDEADLK)
goto retry;
if (error == EINTR) {
mirror->head.flags |= HAMMER_IOC_HEAD_INTR;
error = 0;
}
failed:
mirror->key_cur.localization &= HAMMER_LOCALIZE_MASK;
return(error);
}
/*
* Retrieve the PFS hammer cleanup utility config record. This is
* different (newer than) the PFS config.
*
* This is kinda a hack.
*/
static
int
hammer_ioc_set_config(hammer_transaction_t trans, hammer_inode_t ip,
struct hammer_ioc_config *config)
{
struct hammer_btree_leaf_elm leaf;
struct hammer_cursor cursor;
hammer_mount_t hmp = ip->hmp;
int error;
again:
error = hammer_init_cursor(trans, &cursor, &ip->cache[0], NULL);
if (error) {
hammer_done_cursor(&cursor);
return(error);
}
bzero(&leaf, sizeof(leaf));
leaf.base.obj_id = HAMMER_OBJID_ROOT;
leaf.base.rec_type = HAMMER_RECTYPE_CONFIG;
leaf.base.create_tid = hammer_alloc_tid(hmp, 1);
leaf.base.btype = HAMMER_BTREE_TYPE_RECORD;
leaf.base.localization = ip->obj_localization | HAMMER_LOCALIZE_INODE;
leaf.base.key = 0; /* page 0 */
leaf.data_len = sizeof(struct hammer_config_data);
cursor.key_beg = leaf.base;
cursor.asof = HAMMER_MAX_TID;
cursor.flags |= HAMMER_CURSOR_BACKEND | HAMMER_CURSOR_ASOF;
error = hammer_btree_lookup(&cursor);
if (error == 0) {
error = hammer_btree_extract_data(&cursor);
error = hammer_delete_at_cursor(&cursor, HAMMER_DELETE_DESTROY,
0, 0, 0, NULL);
if (error == EDEADLK) {
hammer_done_cursor(&cursor);
goto again;
}
}
if (error == ENOENT)
error = 0;
if (error == 0) {
/*
* NOTE: Must reload key_beg after an ASOF search because
* the create_tid may have been modified during the
* search.
*/
cursor.flags &= ~HAMMER_CURSOR_ASOF;
cursor.key_beg = leaf.base;
error = hammer_create_at_cursor(&cursor, &leaf,
&config->config,
HAMMER_CREATE_MODE_SYS);
if (error == EDEADLK) {
hammer_done_cursor(&cursor);
goto again;
}
}
config->head.error = error;
hammer_done_cursor(&cursor);
return(0);
}
/*
* Retrieve as many snapshot ids as possible or until the array is
* full, starting after the last transaction id passed in. If count
* is 0 we retrieve starting at the beginning.
*
* NOTE: Because the b-tree key field is signed but transaction ids
* are unsigned the returned list will be signed-sorted instead
* of unsigned sorted. The Caller must still sort the aggregate
* results.
*/
static
int
hammer_ioc_get_snapshot(hammer_transaction_t trans, hammer_inode_t ip,
struct hammer_ioc_snapshot *snap)
{
struct hammer_cursor cursor;
int error;
/*
* Validate structure
*/
if (snap->index != 0)
return (EINVAL);
if (snap->count > HAMMER_SNAPS_PER_IOCTL)
return (EINVAL);
/*
* Look for keys starting after the previous iteration, or at
* the beginning if snap->count is 0.
*/
error = hammer_init_cursor(trans, &cursor, &ip->cache[0], NULL);
if (error) {
hammer_done_cursor(&cursor);
return(error);
}
cursor.key_beg.obj_id = HAMMER_OBJID_ROOT;
cursor.key_beg.create_tid = 0;
cursor.key_beg.delete_tid = 0;
cursor.key_beg.obj_type = 0;
cursor.key_beg.rec_type = HAMMER_RECTYPE_SNAPSHOT;
cursor.key_beg.localization = ip->obj_localization | HAMMER_LOCALIZE_INODE;
if (snap->count == 0)
cursor.key_beg.key = HAMMER_MIN_KEY;
else
cursor.key_beg.key = (int64_t)snap->snaps[snap->count - 1].tid + 1;
cursor.key_end = cursor.key_beg;
cursor.key_end.key = HAMMER_MAX_KEY;
cursor.asof = HAMMER_MAX_TID;
cursor.flags |= HAMMER_CURSOR_END_EXCLUSIVE | HAMMER_CURSOR_ASOF;
snap->count = 0;
error = hammer_btree_first(&cursor);
while (error == 0 && snap->count < HAMMER_SNAPS_PER_IOCTL) {
error = hammer_btree_extract_leaf(&cursor);
if (error)
break;
if (cursor.leaf->base.rec_type == HAMMER_RECTYPE_SNAPSHOT) {
error = hammer_btree_extract_data(&cursor);
snap->snaps[snap->count] = cursor.data->snap;
/*
* The snap data tid should match the key but might
* not due to a bug in the HAMMER v3 conversion code.
*
* This error will work itself out over time but we
* have to force a match or the snapshot will not
* be deletable.
*/
if (cursor.data->snap.tid !=
(hammer_tid_t)cursor.leaf->base.key) {
hkprintf("lo=%08x snapshot key "
"0x%016jx data mismatch 0x%016jx\n",
cursor.key_beg.localization,
(uintmax_t)cursor.data->snap.tid,
cursor.leaf->base.key);
hkprintf("Probably left over from the "
"original v3 conversion, hammer "
"cleanup should get it eventually\n");
snap->snaps[snap->count].tid =
cursor.leaf->base.key;
}
++snap->count;
}
error = hammer_btree_iterate(&cursor);
}
if (error == ENOENT) {
snap->head.flags |= HAMMER_IOC_SNAPSHOT_EOF;
error = 0;
}
snap->head.error = error;
hammer_done_cursor(&cursor);
return(0);
}
