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_ioctl.c,v 1.3 2008/02/06 08:59:28 dillon Exp $
39 static int hammer_ioc_prune(hammer_inode_t ip,
40 struct hammer_ioc_prune *prune);
41 static int hammer_ioc_gethistory(hammer_inode_t ip,
42 struct hammer_ioc_history *hist);
45 hammer_ioctl(hammer_inode_t ip, u_long com, caddr_t data, int fflag,
50 error = suser_cred(cred, PRISON_ROOT);
55 error = hammer_ioc_prune(ip,
56 (struct hammer_ioc_prune *)data);
59 case HAMMERIOC_GETHISTORY:
60 error = hammer_ioc_gethistory(ip,
61 (struct hammer_ioc_history *)data);
71 * Iterate through the specified range of object ids and remove any
72 * deleted records that fall entirely within a prune modulo.
74 * A reverse iteration is used to prevent overlapping records from being
75 * created during the iteration due to alignments. This also allows us
76 * to adjust alignments without blowing up the B-Tree.
78 static int check_prune(struct hammer_ioc_prune *prune, hammer_btree_elm_t elm,
79 int *realign_cre, int *realign_del);
80 static int realign_prune(struct hammer_ioc_prune *prune, hammer_cursor_t cursor,
81 int realign_cre, int realign_del);
84 hammer_ioc_prune(hammer_inode_t ip, struct hammer_ioc_prune *prune)
86 struct hammer_cursor cursor;
87 hammer_btree_elm_t elm;
93 if (prune->nelms < 0 || prune->nelms > HAMMER_MAX_PRUNE_ELMS)
95 if (prune->beg_obj_id >= prune->end_obj_id)
97 if ((prune->flags & HAMMER_IOC_PRUNE_ALL) && prune->nelms)
101 error = hammer_init_cursor_hmp(&cursor, NULL, ip->hmp);
103 hammer_done_cursor(&cursor);
106 cursor.key_beg.obj_id = prune->beg_obj_id;
107 cursor.key_beg.key = HAMMER_MIN_KEY;
108 cursor.key_beg.create_tid = 1;
109 cursor.key_beg.delete_tid = 0;
110 cursor.key_beg.rec_type = HAMMER_MIN_RECTYPE;
111 cursor.key_beg.obj_type = 0;
113 cursor.key_end.obj_id = prune->cur_obj_id;
114 cursor.key_end.key = prune->cur_key;
115 cursor.key_end.create_tid = HAMMER_MAX_TID - 1;
116 cursor.key_end.delete_tid = 0;
117 cursor.key_end.rec_type = HAMMER_MAX_RECTYPE;
118 cursor.key_end.obj_type = 0;
120 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
122 error = hammer_btree_last(&cursor);
124 elm = &cursor.node->ondisk->elms[cursor.index];
125 prune->cur_obj_id = elm->base.obj_id;
126 prune->cur_key = elm->base.key;
128 if (check_prune(prune, elm, &realign_cre, &realign_del) == 0) {
129 if (hammer_debug_general & 0x0200) {
130 kprintf("check %016llx %016llx: DELETE\n",
131 elm->base.obj_id, elm->base.key);
135 * NOTE: This can return EDEADLK
137 isdir = (elm->base.rec_type == HAMMER_RECTYPE_DIRENTRY);
139 error = hammer_delete_at_cursor(&cursor,
145 ++prune->stat_dirrecords;
147 ++prune->stat_rawrecords;
148 } else if (realign_cre >= 0 || realign_del >= 0) {
149 error = realign_prune(prune, &cursor,
150 realign_cre, realign_del);
152 cursor.flags |= HAMMER_CURSOR_ATEDISK;
153 if (hammer_debug_general & 0x0200) {
154 kprintf("check %016llx %016llx: "
161 cursor.flags |= HAMMER_CURSOR_ATEDISK;
162 if (hammer_debug_general & 0x0100) {
163 kprintf("check %016llx %016llx: SKIP\n",
164 elm->base.obj_id, elm->base.key);
168 error = hammer_btree_iterate_reverse(&cursor);
172 hammer_done_cursor(&cursor);
173 if (error == EDEADLK)
179 * Check pruning list. The list must be sorted in descending order.
182 check_prune(struct hammer_ioc_prune *prune, hammer_btree_elm_t elm,
183 int *realign_cre, int *realign_del)
185 struct hammer_ioc_prune_elm *scan;
192 * If pruning everything remove all records with a non-zero
195 if (prune->flags & HAMMER_IOC_PRUNE_ALL) {
196 if (elm->base.delete_tid != 0)
201 for (i = 0; i < prune->nelms; ++i) {
202 scan = &prune->elms[i];
205 * Locate the scan index covering the create and delete TIDs.
207 if (*realign_cre < 0 &&
208 elm->base.create_tid >= scan->beg_tid &&
209 elm->base.create_tid < scan->end_tid) {
212 if (*realign_del < 0 && elm->base.delete_tid &&
213 elm->base.delete_tid > scan->beg_tid &&
214 elm->base.delete_tid <= scan->end_tid) {
219 * Now check for loop termination.
221 if (elm->base.create_tid >= scan->end_tid ||
222 elm->base.delete_tid > scan->end_tid) {
227 * Now determine if we can delete the record.
229 if (elm->base.delete_tid &&
230 elm->base.create_tid >= scan->beg_tid &&
231 elm->base.delete_tid <= scan->end_tid &&
232 elm->base.create_tid / scan->mod_tid ==
233 elm->base.delete_tid / scan->mod_tid) {
241 * Align the record to cover any gaps created through the deletion of
242 * records within the pruning space. If we were to just delete the records
243 * there would be gaps which in turn would cause a snapshot that is NOT on
244 * a pruning boundary to appear corrupt to the user. Forcing alignment
245 * of the create_tid and delete_tid for retained records 'reconnects'
246 * the previously contiguous space, making it contiguous again after the
249 * The use of a reverse iteration allows us to safely align the records and
250 * related elements without creating temporary overlaps. XXX we should
251 * add ordering dependancies for record buffers to guarantee consistency
255 realign_prune(struct hammer_ioc_prune *prune,
256 hammer_cursor_t cursor, int realign_cre, int realign_del)
258 hammer_btree_elm_t elm;
264 hammer_cursor_downgrade(cursor);
266 elm = &cursor->node->ondisk->elms[cursor->index];
267 ++prune->stat_realignments;
270 * Align the create_tid. By doing a reverse iteration we guarantee
271 * that all records after our current record have already been
272 * aligned, allowing us to safely correct the right-hand-boundary
273 * (because no record to our right if otherwise exactly matching
274 * will have a create_tid to the left of our aligned create_tid).
276 * Ordering is important here XXX but disk write ordering for
277 * inter-cluster corrections is not currently guaranteed.
280 if (realign_cre >= 0) {
281 mod = prune->elms[realign_cre].mod_tid;
282 delta = elm->leaf.base.create_tid % mod;
284 tid = elm->leaf.base.create_tid - delta + mod;
287 error = hammer_btree_correct_rhb(cursor, tid + 1);
289 error = hammer_btree_extract(cursor,
290 HAMMER_CURSOR_GET_RECORD);
294 error = hammer_cursor_upgrade(cursor);
297 hammer_modify_buffer(cursor->record_buffer);
298 cursor->record->base.base.create_tid = tid;
299 hammer_modify_node(cursor->node);
300 elm->leaf.base.create_tid = tid;
306 * Align the delete_tid. This only occurs if the record is historical
307 * was deleted at some point. Realigning the delete_tid does not
308 * move the record within the B-Tree but may cause it to temporarily
309 * overlap a record that has not yet been pruned.
311 if (error == 0 && realign_del >= 0) {
312 mod = prune->elms[realign_del].mod_tid;
313 delta = elm->leaf.base.delete_tid % mod;
314 hammer_modify_node(cursor->node);
316 error = hammer_btree_extract(cursor,
317 HAMMER_CURSOR_GET_RECORD);
319 elm->leaf.base.delete_tid =
320 elm->leaf.base.delete_tid -
322 hammer_modify_buffer(cursor->record_buffer);
323 cursor->record->base.base.delete_tid =
324 elm->leaf.base.delete_tid;
332 * Iterate through an object's inode or an object's records and record
335 static void add_history(hammer_inode_t ip, struct hammer_ioc_history *hist,
336 hammer_btree_elm_t elm);
340 hammer_ioc_gethistory(hammer_inode_t ip, struct hammer_ioc_history *hist)
342 struct hammer_cursor cursor;
343 hammer_btree_elm_t elm;
347 * Validate the structure and initialize for return.
349 if (hist->beg_tid > hist->end_tid)
351 if (hist->flags & HAMMER_IOC_HISTORY_ATKEY) {
352 if (hist->key > hist->nxt_key)
356 hist->obj_id = ip->obj_id;
358 hist->nxt_tid = hist->end_tid;
359 hist->flags &= ~HAMMER_IOC_HISTORY_NEXT_TID;
360 hist->flags &= ~HAMMER_IOC_HISTORY_NEXT_KEY;
361 hist->flags &= ~HAMMER_IOC_HISTORY_EOF;
362 hist->flags &= ~HAMMER_IOC_HISTORY_UNSYNCED;
363 if ((ip->flags & HAMMER_INODE_MODMASK) & ~HAMMER_INODE_ITIMES)
364 hist->flags |= HAMMER_IOC_HISTORY_UNSYNCED;
367 * Setup the cursor. We can't handle undeletable records
368 * (create_tid of 0) at the moment. A create_tid of 0 has
369 * a special meaning and cannot be specified in the cursor.
371 error = hammer_init_cursor_hmp(&cursor, &ip->cache[0], ip->hmp);
373 hammer_done_cursor(&cursor);
377 cursor.key_beg.obj_id = hist->obj_id;
378 cursor.key_beg.create_tid = hist->beg_tid;
379 cursor.key_beg.delete_tid = 0;
380 cursor.key_beg.obj_type = 0;
381 if (cursor.key_beg.create_tid == HAMMER_MIN_TID)
382 cursor.key_beg.create_tid = 1;
384 cursor.key_end.obj_id = hist->obj_id;
385 cursor.key_end.create_tid = hist->end_tid;
386 cursor.key_end.delete_tid = 0;
387 cursor.key_end.obj_type = 0;
389 cursor.flags |= HAMMER_CURSOR_END_EXCLUSIVE;
391 if (hist->flags & HAMMER_IOC_HISTORY_ATKEY) {
393 * key-range within the file. For a regular file the
394 * on-disk key represents BASE+LEN, not BASE, so the
395 * first possible record containing the offset 'key'
396 * has an on-disk key of (key + 1).
398 cursor.key_beg.key = hist->key;
399 cursor.key_end.key = HAMMER_MAX_KEY;
401 switch(ip->ino_rec.base.base.obj_type) {
402 case HAMMER_OBJTYPE_REGFILE:
403 ++cursor.key_beg.key;
404 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
406 case HAMMER_OBJTYPE_DIRECTORY:
407 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
409 case HAMMER_OBJTYPE_DBFILE:
410 cursor.key_beg.rec_type = HAMMER_RECTYPE_DB;
416 cursor.key_end.rec_type = cursor.key_beg.rec_type;
421 cursor.key_beg.key = 0;
422 cursor.key_end.key = 0;
423 cursor.key_beg.rec_type = HAMMER_RECTYPE_INODE;
424 cursor.key_end.rec_type = HAMMER_RECTYPE_INODE;
427 error = hammer_btree_first(&cursor);
429 elm = &cursor.node->ondisk->elms[cursor.index];
431 add_history(ip, hist, elm);
432 if (hist->flags & (HAMMER_IOC_HISTORY_NEXT_TID |
433 HAMMER_IOC_HISTORY_NEXT_KEY |
434 HAMMER_IOC_HISTORY_EOF)) {
437 error = hammer_btree_iterate(&cursor);
439 if (error == ENOENT) {
440 hist->flags |= HAMMER_IOC_HISTORY_EOF;
443 hammer_done_cursor(&cursor);
448 * Add the scanned element to the ioctl return structure. Some special
449 * casing is required for regular files to accomodate how data ranges are
453 add_history(hammer_inode_t ip, struct hammer_ioc_history *hist,
454 hammer_btree_elm_t elm)
456 if (elm->base.btype != HAMMER_BTREE_TYPE_RECORD)
458 if ((hist->flags & HAMMER_IOC_HISTORY_ATKEY) &&
459 ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_REGFILE) {
463 if (hist->nxt_key > elm->leaf.base.key - elm->leaf.data_len &&
464 hist->key < elm->leaf.base.key - elm->leaf.data_len) {
465 hist->nxt_key = elm->leaf.base.key - elm->leaf.data_len;
467 if (hist->nxt_key > elm->leaf.base.key)
468 hist->nxt_key = elm->leaf.base.key;
471 * Record is beyond MAXPHYS, there won't be any more records
472 * in the iteration covering the requested offset (key).
474 if (elm->leaf.base.key >= MAXPHYS &&
475 elm->leaf.base.key - MAXPHYS > hist->key) {
476 hist->flags |= HAMMER_IOC_HISTORY_NEXT_KEY;
480 * Data-range of record does not cover the key.
482 if (elm->leaf.base.key - elm->leaf.data_len > hist->key)
485 } else if (hist->flags & HAMMER_IOC_HISTORY_ATKEY) {
489 if (hist->nxt_key > elm->leaf.base.key &&
490 hist->key < elm->leaf.base.key) {
491 hist->nxt_key = elm->leaf.base.key;
495 * Record is beyond the requested key.
497 if (elm->leaf.base.key > hist->key)
498 hist->flags |= HAMMER_IOC_HISTORY_NEXT_KEY;
502 * Add create_tid if it is in-bounds.
504 if ((hist->count == 0 ||
505 elm->leaf.base.create_tid != hist->tid_ary[hist->count - 1]) &&
506 elm->leaf.base.create_tid >= hist->beg_tid &&
507 elm->leaf.base.create_tid < hist->end_tid) {
508 if (hist->count == HAMMER_MAX_HISTORY_ELMS) {
509 hist->nxt_tid = elm->leaf.base.create_tid;
510 hist->flags |= HAMMER_IOC_HISTORY_NEXT_TID;
513 hist->tid_ary[hist->count++] = elm->leaf.base.create_tid;
517 * Add delete_tid if it is in-bounds. Note that different portions
518 * of the history may have overlapping data ranges with different
519 * delete_tid's. If this case occurs the delete_tid may match the
520 * create_tid of a following record. XXX
525 if (elm->leaf.base.delete_tid &&
526 elm->leaf.base.delete_tid >= hist->beg_tid &&
527 elm->leaf.base.delete_tid < hist->end_tid) {
528 if (hist->count == HAMMER_MAX_HISTORY_ELMS) {
529 hist->nxt_tid = elm->leaf.base.delete_tid;
530 hist->flags |= HAMMER_IOC_HISTORY_NEXT_TID;
533 hist->tid_ary[hist->count++] = elm->leaf.base.delete_tid;