xref: /dragonfly/sys/vfs/hammer/hammer_ioctl.c (revision dcd37f7d)
1 /*
2  * Copyright (c) 2008 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Matthew Dillon <dillon@backplane.com>
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  *
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
16  *    distribution.
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.
20  *
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
32  * SUCH DAMAGE.
33  *
34  * $DragonFly: src/sys/vfs/hammer/hammer_ioctl.c,v 1.32 2008/11/13 02:23:29 dillon Exp $
35  */
36 
37 #include "hammer.h"
38 
39 static int hammer_ioc_gethistory(hammer_transaction_t trans, hammer_inode_t ip,
40 				struct hammer_ioc_history *hist);
41 static int hammer_ioc_synctid(hammer_transaction_t trans, hammer_inode_t ip,
42 				struct hammer_ioc_synctid *std);
43 static int hammer_ioc_get_version(hammer_transaction_t trans,
44 				hammer_inode_t ip,
45 				struct hammer_ioc_version *ver);
46 static int hammer_ioc_set_version(hammer_transaction_t trans,
47 				hammer_inode_t ip,
48 				struct hammer_ioc_version *ver);
49 static int hammer_ioc_get_info(hammer_transaction_t trans,
50 				struct hammer_ioc_info *info);
51 static int hammer_ioc_add_snapshot(hammer_transaction_t trans, hammer_inode_t ip,
52 				struct hammer_ioc_snapshot *snap);
53 static int hammer_ioc_del_snapshot(hammer_transaction_t trans, hammer_inode_t ip,
54 				struct hammer_ioc_snapshot *snap);
55 static int hammer_ioc_get_snapshot(hammer_transaction_t trans, hammer_inode_t ip,
56 				struct hammer_ioc_snapshot *snap);
57 static int hammer_ioc_get_config(hammer_transaction_t trans, hammer_inode_t ip,
58 				struct hammer_ioc_config *snap);
59 static int hammer_ioc_set_config(hammer_transaction_t trans, hammer_inode_t ip,
60 				struct hammer_ioc_config *snap);
61 
62 int
63 hammer_ioctl(hammer_inode_t ip, u_long com, caddr_t data, int fflag,
64 	     struct ucred *cred)
65 {
66 	struct hammer_transaction trans;
67 	int error;
68 
69 	error = priv_check_cred(cred, PRIV_HAMMER_IOCTL, 0);
70 
71 	hammer_start_transaction(&trans, ip->hmp);
72 
73 	switch(com) {
74 	case HAMMERIOC_PRUNE:
75 		if (error == 0) {
76 			error = hammer_ioc_prune(&trans, ip,
77 					(struct hammer_ioc_prune *)data);
78 		}
79 		break;
80 	case HAMMERIOC_GETHISTORY:
81 		error = hammer_ioc_gethistory(&trans, ip,
82 					(struct hammer_ioc_history *)data);
83 		break;
84 	case HAMMERIOC_REBLOCK:
85 		if (error == 0) {
86 			error = hammer_ioc_reblock(&trans, ip,
87 					(struct hammer_ioc_reblock *)data);
88 		}
89 		break;
90 	case HAMMERIOC_REBALANCE:
91 		/*
92 		 * Rebalancing needs to lock a lot of B-Tree nodes.  The
93 		 * children and children's children.  Systems with very
94 		 * little memory will not be able to do it.
95 		 */
96 		if (error == 0 && nbuf < HAMMER_REBALANCE_MIN_BUFS) {
97 			kprintf("hammer: System has insufficient buffers "
98 				"to rebalance the tree.  nbuf < %d\n",
99 				HAMMER_REBALANCE_MIN_BUFS);
100 			error = ENOSPC;
101 		}
102 		if (error == 0) {
103 			error = hammer_ioc_rebalance(&trans, ip,
104 					(struct hammer_ioc_rebalance *)data);
105 		}
106 		break;
107 	case HAMMERIOC_SYNCTID:
108 		error = hammer_ioc_synctid(&trans, ip,
109 					(struct hammer_ioc_synctid *)data);
110 		break;
111 	case HAMMERIOC_GET_PSEUDOFS:
112 		error = hammer_ioc_get_pseudofs(&trans, ip,
113 				    (struct hammer_ioc_pseudofs_rw *)data);
114 		break;
115 	case HAMMERIOC_SET_PSEUDOFS:
116 		if (error == 0) {
117 			error = hammer_ioc_set_pseudofs(&trans, ip, cred,
118 				    (struct hammer_ioc_pseudofs_rw *)data);
119 		}
120 		break;
121 	case HAMMERIOC_UPG_PSEUDOFS:
122 		if (error == 0) {
123 			error = hammer_ioc_upgrade_pseudofs(&trans, ip,
124 				    (struct hammer_ioc_pseudofs_rw *)data);
125 		}
126 		break;
127 	case HAMMERIOC_DGD_PSEUDOFS:
128 		if (error == 0) {
129 			error = hammer_ioc_downgrade_pseudofs(&trans, ip,
130 				    (struct hammer_ioc_pseudofs_rw *)data);
131 		}
132 		break;
133 	case HAMMERIOC_RMR_PSEUDOFS:
134 		if (error == 0) {
135 			error = hammer_ioc_destroy_pseudofs(&trans, ip,
136 				    (struct hammer_ioc_pseudofs_rw *)data);
137 		}
138 		break;
139 	case HAMMERIOC_WAI_PSEUDOFS:
140 		if (error == 0) {
141 			error = hammer_ioc_wait_pseudofs(&trans, ip,
142 				    (struct hammer_ioc_pseudofs_rw *)data);
143 		}
144 		break;
145 	case HAMMERIOC_MIRROR_READ:
146 		if (error == 0) {
147 			error = hammer_ioc_mirror_read(&trans, ip,
148 				    (struct hammer_ioc_mirror_rw *)data);
149 		}
150 		break;
151 	case HAMMERIOC_MIRROR_WRITE:
152 		if (error == 0) {
153 			error = hammer_ioc_mirror_write(&trans, ip,
154 				    (struct hammer_ioc_mirror_rw *)data);
155 		}
156 		break;
157 	case HAMMERIOC_GET_VERSION:
158 		error = hammer_ioc_get_version(&trans, ip,
159 				    (struct hammer_ioc_version *)data);
160 		break;
161 	case HAMMERIOC_GET_INFO:
162 		error = hammer_ioc_get_info(&trans,
163 				    (struct hammer_ioc_info *)data);
164 		break;
165 	case HAMMERIOC_SET_VERSION:
166 		if (error == 0) {
167 			error = hammer_ioc_set_version(&trans, ip,
168 					    (struct hammer_ioc_version *)data);
169 		}
170 		break;
171 	case HAMMERIOC_ADD_VOLUME:
172 		if (error == 0) {
173 			error = priv_check_cred(cred, PRIV_HAMMER_VOLUME, 0);
174 			if (error == 0)
175 				error = hammer_ioc_volume_add(&trans, ip,
176 					    (struct hammer_ioc_volume *)data);
177 		}
178 		break;
179 	case HAMMERIOC_DEL_VOLUME:
180 		if (error == 0) {
181 			error = priv_check_cred(cred, PRIV_HAMMER_VOLUME, 0);
182 			if (error == 0)
183 				error = hammer_ioc_volume_del(&trans, ip,
184 					    (struct hammer_ioc_volume *)data);
185 		}
186 		break;
187 	case HAMMERIOC_ADD_SNAPSHOT:
188 		if (error == 0) {
189 			error = hammer_ioc_add_snapshot(
190 					&trans, ip, (struct hammer_ioc_snapshot *)data);
191 		}
192 		break;
193 	case HAMMERIOC_DEL_SNAPSHOT:
194 		if (error == 0) {
195 			error = hammer_ioc_del_snapshot(
196 					&trans, ip, (struct hammer_ioc_snapshot *)data);
197 		}
198 		break;
199 	case HAMMERIOC_GET_SNAPSHOT:
200 		error = hammer_ioc_get_snapshot(
201 					&trans, ip, (struct hammer_ioc_snapshot *)data);
202 		break;
203 	case HAMMERIOC_GET_CONFIG:
204 		error = hammer_ioc_get_config(
205 					&trans, ip, (struct hammer_ioc_config *)data);
206 		break;
207 	case HAMMERIOC_SET_CONFIG:
208 		if (error == 0) {
209 			error = hammer_ioc_set_config(
210 					&trans, ip, (struct hammer_ioc_config *)data);
211 		}
212 		break;
213 	default:
214 		error = EOPNOTSUPP;
215 		break;
216 	}
217 	hammer_done_transaction(&trans);
218 	return (error);
219 }
220 
221 /*
222  * Iterate through an object's inode or an object's records and record
223  * modification TIDs.
224  */
225 static void add_history(hammer_inode_t ip, struct hammer_ioc_history *hist,
226 			hammer_btree_elm_t elm);
227 
228 static
229 int
230 hammer_ioc_gethistory(hammer_transaction_t trans, hammer_inode_t ip,
231 		      struct hammer_ioc_history *hist)
232 {
233 	struct hammer_cursor cursor;
234 	hammer_btree_elm_t elm;
235 	int error;
236 
237 	/*
238 	 * Validate the structure and initialize for return.
239 	 */
240 	if (hist->beg_tid > hist->end_tid)
241 		return(EINVAL);
242 	if (hist->head.flags & HAMMER_IOC_HISTORY_ATKEY) {
243 		if (hist->key > hist->nxt_key)
244 			return(EINVAL);
245 	}
246 
247 	hist->obj_id = ip->obj_id;
248 	hist->count = 0;
249 	hist->nxt_tid = hist->end_tid;
250 	hist->head.flags &= ~HAMMER_IOC_HISTORY_NEXT_TID;
251 	hist->head.flags &= ~HAMMER_IOC_HISTORY_NEXT_KEY;
252 	hist->head.flags &= ~HAMMER_IOC_HISTORY_EOF;
253 	hist->head.flags &= ~HAMMER_IOC_HISTORY_UNSYNCED;
254 	if ((ip->flags & HAMMER_INODE_MODMASK) &
255 	    ~(HAMMER_INODE_ATIME | HAMMER_INODE_MTIME)) {
256 		hist->head.flags |= HAMMER_IOC_HISTORY_UNSYNCED;
257 	}
258 
259 	/*
260 	 * Setup the cursor.  We can't handle undeletable records
261 	 * (create_tid of 0) at the moment.  A create_tid of 0 has
262 	 * a special meaning and cannot be specified in the cursor.
263 	 */
264 	error = hammer_init_cursor(trans, &cursor, &ip->cache[0], NULL);
265 	if (error) {
266 		hammer_done_cursor(&cursor);
267 		return(error);
268 	}
269 
270 	cursor.key_beg.obj_id = hist->obj_id;
271 	cursor.key_beg.create_tid = hist->beg_tid;
272 	cursor.key_beg.delete_tid = 0;
273 	cursor.key_beg.obj_type = 0;
274 	if (cursor.key_beg.create_tid == HAMMER_MIN_TID)
275 		cursor.key_beg.create_tid = 1;
276 
277 	cursor.key_end.obj_id = hist->obj_id;
278 	cursor.key_end.create_tid = hist->end_tid;
279 	cursor.key_end.delete_tid = 0;
280 	cursor.key_end.obj_type = 0;
281 
282 	cursor.flags |= HAMMER_CURSOR_END_EXCLUSIVE;
283 
284 	if (hist->head.flags & HAMMER_IOC_HISTORY_ATKEY) {
285 		/*
286 		 * key-range within the file.  For a regular file the
287 		 * on-disk key represents BASE+LEN, not BASE, so the
288 		 * first possible record containing the offset 'key'
289 		 * has an on-disk key of (key + 1).
290 		 */
291 		cursor.key_beg.key = hist->key;
292 		cursor.key_end.key = HAMMER_MAX_KEY;
293 		cursor.key_beg.localization = ip->obj_localization +
294 					      HAMMER_LOCALIZE_MISC;
295 		cursor.key_end.localization = ip->obj_localization +
296 					      HAMMER_LOCALIZE_MISC;
297 
298 		switch(ip->ino_data.obj_type) {
299 		case HAMMER_OBJTYPE_REGFILE:
300 			++cursor.key_beg.key;
301 			cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
302 			break;
303 		case HAMMER_OBJTYPE_DIRECTORY:
304 			cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
305 			cursor.key_beg.localization = ip->obj_localization +
306 						hammer_dir_localization(ip);
307 			cursor.key_end.localization = ip->obj_localization +
308 						hammer_dir_localization(ip);
309 			break;
310 		case HAMMER_OBJTYPE_DBFILE:
311 			cursor.key_beg.rec_type = HAMMER_RECTYPE_DB;
312 			break;
313 		default:
314 			error = EINVAL;
315 			break;
316 		}
317 		cursor.key_end.rec_type = cursor.key_beg.rec_type;
318 	} else {
319 		/*
320 		 * The inode itself.
321 		 */
322 		cursor.key_beg.key = 0;
323 		cursor.key_end.key = 0;
324 		cursor.key_beg.rec_type = HAMMER_RECTYPE_INODE;
325 		cursor.key_end.rec_type = HAMMER_RECTYPE_INODE;
326 		cursor.key_beg.localization = ip->obj_localization +
327 					      HAMMER_LOCALIZE_INODE;
328 		cursor.key_end.localization = ip->obj_localization +
329 					      HAMMER_LOCALIZE_INODE;
330 	}
331 
332 	error = hammer_btree_first(&cursor);
333 	while (error == 0) {
334 		elm = &cursor.node->ondisk->elms[cursor.index];
335 
336 		add_history(ip, hist, elm);
337 		if (hist->head.flags & (HAMMER_IOC_HISTORY_NEXT_TID |
338 				        HAMMER_IOC_HISTORY_NEXT_KEY |
339 				        HAMMER_IOC_HISTORY_EOF)) {
340 			break;
341 		}
342 		error = hammer_btree_iterate(&cursor);
343 	}
344 	if (error == ENOENT) {
345 		hist->head.flags |= HAMMER_IOC_HISTORY_EOF;
346 		error = 0;
347 	}
348 	hammer_done_cursor(&cursor);
349 	return(error);
350 }
351 
352 /*
353  * Add the scanned element to the ioctl return structure.  Some special
354  * casing is required for regular files to accomodate how data ranges are
355  * stored on-disk.
356  */
357 static void
358 add_history(hammer_inode_t ip, struct hammer_ioc_history *hist,
359 	    hammer_btree_elm_t elm)
360 {
361 	int i;
362 
363 	if (elm->base.btype != HAMMER_BTREE_TYPE_RECORD)
364 		return;
365 	if ((hist->head.flags & HAMMER_IOC_HISTORY_ATKEY) &&
366 	    ip->ino_data.obj_type == HAMMER_OBJTYPE_REGFILE) {
367 		/*
368 		 * Adjust nxt_key
369 		 */
370 		if (hist->nxt_key > elm->leaf.base.key - elm->leaf.data_len &&
371 		    hist->key < elm->leaf.base.key - elm->leaf.data_len) {
372 			hist->nxt_key = elm->leaf.base.key - elm->leaf.data_len;
373 		}
374 		if (hist->nxt_key > elm->leaf.base.key)
375 			hist->nxt_key = elm->leaf.base.key;
376 
377 		/*
378 		 * Record is beyond MAXPHYS, there won't be any more records
379 		 * in the iteration covering the requested offset (key).
380 		 */
381 		if (elm->leaf.base.key >= MAXPHYS &&
382 		    elm->leaf.base.key - MAXPHYS > hist->key) {
383 			hist->head.flags |= HAMMER_IOC_HISTORY_NEXT_KEY;
384 		}
385 
386 		/*
387 		 * Data-range of record does not cover the key.
388 		 */
389 		if (elm->leaf.base.key - elm->leaf.data_len > hist->key)
390 			return;
391 
392 	} else if (hist->head.flags & HAMMER_IOC_HISTORY_ATKEY) {
393 		/*
394 		 * Adjust nxt_key
395 		 */
396 		if (hist->nxt_key > elm->leaf.base.key &&
397 		    hist->key < elm->leaf.base.key) {
398 			hist->nxt_key = elm->leaf.base.key;
399 		}
400 
401 		/*
402 		 * Record is beyond the requested key.
403 		 */
404 		if (elm->leaf.base.key > hist->key)
405 			hist->head.flags |= HAMMER_IOC_HISTORY_NEXT_KEY;
406 	}
407 
408 	/*
409 	 * Add create_tid if it is in-bounds.
410 	 */
411 	i = hist->count;
412 	if ((i == 0 ||
413 	     elm->leaf.base.create_tid != hist->hist_ary[i - 1].tid) &&
414 	    elm->leaf.base.create_tid >= hist->beg_tid &&
415 	    elm->leaf.base.create_tid < hist->end_tid) {
416 		if (hist->count == HAMMER_MAX_HISTORY_ELMS) {
417 			hist->nxt_tid = elm->leaf.base.create_tid;
418 			hist->head.flags |= HAMMER_IOC_HISTORY_NEXT_TID;
419 			return;
420 		}
421 		hist->hist_ary[i].tid = elm->leaf.base.create_tid;
422 		hist->hist_ary[i].time32 = elm->leaf.create_ts;
423 		++hist->count;
424 	}
425 
426 	/*
427 	 * Add delete_tid if it is in-bounds.  Note that different portions
428 	 * of the history may have overlapping data ranges with different
429 	 * delete_tid's.  If this case occurs the delete_tid may match the
430 	 * create_tid of a following record.  XXX
431 	 *
432 	 *	[        ]
433 	 *            [     ]
434 	 */
435 	i = hist->count;
436 	if (elm->leaf.base.delete_tid &&
437 	    elm->leaf.base.delete_tid >= hist->beg_tid &&
438 	    elm->leaf.base.delete_tid < hist->end_tid) {
439 		if (i == HAMMER_MAX_HISTORY_ELMS) {
440 			hist->nxt_tid = elm->leaf.base.delete_tid;
441 			hist->head.flags |= HAMMER_IOC_HISTORY_NEXT_TID;
442 			return;
443 		}
444 		hist->hist_ary[i].tid = elm->leaf.base.delete_tid;
445 		hist->hist_ary[i].time32 = elm->leaf.delete_ts;
446 		++hist->count;
447 	}
448 }
449 
450 /*
451  * Acquire synchronization TID
452  */
453 static
454 int
455 hammer_ioc_synctid(hammer_transaction_t trans, hammer_inode_t ip,
456 		   struct hammer_ioc_synctid *std)
457 {
458 	hammer_mount_t hmp = ip->hmp;
459 	int error = 0;
460 
461 	switch(std->op) {
462 	case HAMMER_SYNCTID_NONE:
463 		std->tid = hmp->flusher.tid;	/* inaccurate */
464 		break;
465 	case HAMMER_SYNCTID_ASYNC:
466 		hammer_queue_inodes_flusher(hmp, MNT_NOWAIT);
467 		hammer_flusher_async(hmp, NULL);
468 		std->tid = hmp->flusher.tid;	/* inaccurate */
469 		break;
470 	case HAMMER_SYNCTID_SYNC1:
471 		hammer_queue_inodes_flusher(hmp, MNT_WAIT);
472 		hammer_flusher_sync(hmp);
473 		std->tid = hmp->flusher.tid;
474 		break;
475 	case HAMMER_SYNCTID_SYNC2:
476 		hammer_queue_inodes_flusher(hmp, MNT_WAIT);
477 		hammer_flusher_sync(hmp);
478 		std->tid = hmp->flusher.tid;
479 		hammer_flusher_sync(hmp);
480 		break;
481 	default:
482 		error = EOPNOTSUPP;
483 		break;
484 	}
485 	return(error);
486 }
487 
488 /*
489  * Retrieve version info.
490  *
491  * Load min_version, wip_version, and max_versino.  If cur_version is passed
492  * as 0 then load the current version into cur_version.  Load the description
493  * for cur_version into the description array.
494  *
495  * Returns 0 on success, EINVAL if cur_version is non-zero and set to an
496  * unsupported value.
497  */
498 static
499 int
500 hammer_ioc_get_version(hammer_transaction_t trans, hammer_inode_t ip,
501 		   struct hammer_ioc_version *ver)
502 {
503 	int error = 0;
504 
505 	ver->min_version = HAMMER_VOL_VERSION_MIN;
506 	ver->wip_version = HAMMER_VOL_VERSION_WIP;
507 	ver->max_version = HAMMER_VOL_VERSION_MAX;
508 	if (ver->cur_version == 0)
509 		ver->cur_version = trans->hmp->version;
510 	switch(ver->cur_version) {
511 	case 1:
512 		ksnprintf(ver->description, sizeof(ver->description),
513 			 "First HAMMER release (DragonFly 2.0+)");
514 		break;
515 	case 2:
516 		ksnprintf(ver->description, sizeof(ver->description),
517 			 "New directory entry layout (DragonFly 2.3+)");
518 		break;
519 	case 3:
520 		ksnprintf(ver->description, sizeof(ver->description),
521 			 "New snapshot management (DragonFly 2.5+)");
522 		break;
523 	case 4:
524 		ksnprintf(ver->description, sizeof(ver->description),
525 			 "New undo/flush, faster flush/sync (DragonFly 2.5+)");
526 		break;
527 	default:
528 		ksnprintf(ver->description, sizeof(ver->description),
529 			 "Unknown");
530 		error = EINVAL;
531 		break;
532 	}
533 	return(error);
534 };
535 
536 /*
537  * Set version info
538  */
539 static
540 int
541 hammer_ioc_set_version(hammer_transaction_t trans, hammer_inode_t ip,
542 		   struct hammer_ioc_version *ver)
543 {
544 	hammer_mount_t hmp = trans->hmp;
545 	struct hammer_cursor cursor;
546 	hammer_volume_t volume;
547 	int error;
548 	int over = hmp->version;
549 
550 	/*
551 	 * Generally do not allow downgrades.  However, version 4 can
552 	 * be downgraded to version 3.
553 	 */
554 	if (ver->cur_version < hmp->version) {
555 		if (!(ver->cur_version == 3 && hmp->version == 4))
556 			return(EINVAL);
557 	}
558 	if (ver->cur_version == hmp->version)
559 		return(0);
560 	if (ver->cur_version > HAMMER_VOL_VERSION_MAX)
561 		return(EINVAL);
562 	if (hmp->ronly)
563 		return(EROFS);
564 
565 	/*
566 	 * Update the root volume header and the version cached in
567 	 * the hammer_mount structure.
568 	 */
569 	error = hammer_init_cursor(trans, &cursor, NULL, NULL);
570 	if (error)
571 		goto failed;
572 	hammer_lock_ex(&hmp->flusher.finalize_lock);
573 	hammer_sync_lock_ex(trans);
574 	hmp->version = ver->cur_version;
575 
576 	/*
577 	 * If upgrading from version < 4 to version >= 4 the UNDO FIFO
578 	 * must be reinitialized.
579 	 */
580 	if (over < HAMMER_VOL_VERSION_FOUR &&
581 	    ver->cur_version >= HAMMER_VOL_VERSION_FOUR) {
582 		kprintf("upgrade undo to version 4\n");
583 		error = hammer_upgrade_undo_4(trans);
584 		if (error)
585 			goto failed;
586 	}
587 
588 	/*
589 	 * Adjust the version in the volume header
590 	 */
591 	volume = hammer_get_root_volume(hmp, &error);
592 	KKASSERT(error == 0);
593 	hammer_modify_volume_field(cursor.trans, volume, vol_version);
594 	volume->ondisk->vol_version = ver->cur_version;
595 	hammer_modify_volume_done(volume);
596 	hammer_rel_volume(volume, 0);
597 
598 	hammer_sync_unlock(trans);
599 	hammer_unlock(&hmp->flusher.finalize_lock);
600 failed:
601 	ver->head.error = error;
602 	hammer_done_cursor(&cursor);
603 	return(0);
604 }
605 
606 /*
607  * Get information
608  */
609 static
610 int
611 hammer_ioc_get_info(hammer_transaction_t trans, struct hammer_ioc_info *info) {
612 
613 	struct hammer_volume_ondisk	*od = trans->hmp->rootvol->ondisk;
614 	struct hammer_mount 		*hm = trans->hmp;
615 
616 	/* Fill the structure with the necessary information */
617 	_hammer_checkspace(hm, HAMMER_CHKSPC_WRITE, &info->rsvbigblocks);
618 	info->rsvbigblocks = info->rsvbigblocks >> HAMMER_LARGEBLOCK_BITS;
619 	strlcpy(info->vol_name, od->vol_name, sizeof(od->vol_name));
620 
621 	info->vol_fsid = hm->fsid;
622 	info->vol_fstype = od->vol_fstype;
623 	info->version = hm->version;
624 
625 	info->inodes = od->vol0_stat_inodes;
626 	info->bigblocks = od->vol0_stat_bigblocks;
627 	info->freebigblocks = od->vol0_stat_freebigblocks;
628 	info->nvolumes = hm->nvolumes;
629 
630 	return 0;
631 }
632 
633 /*
634  * Add a snapshot transction id(s) to the list of snapshots.
635  *
636  * NOTE: Records are created with an allocated TID.  If a flush cycle
637  *	 is in progress the record may be synced in the current flush
638  *	 cycle and the volume header will reflect the allocation of the
639  *	 TID, but the synchronization point may not catch up to the
640  *	 TID until the next flush cycle.
641  */
642 static
643 int
644 hammer_ioc_add_snapshot(hammer_transaction_t trans, hammer_inode_t ip,
645 			struct hammer_ioc_snapshot *snap)
646 {
647 	hammer_mount_t hmp = ip->hmp;
648 	struct hammer_btree_leaf_elm leaf;
649 	struct hammer_cursor cursor;
650 	int error;
651 
652 	/*
653 	 * Validate structure
654 	 */
655 	if (snap->count > HAMMER_SNAPS_PER_IOCTL)
656 		return (EINVAL);
657 	if (snap->index > snap->count)
658 		return (EINVAL);
659 
660 	hammer_lock_ex(&hmp->snapshot_lock);
661 again:
662 	/*
663 	 * Look for keys starting after the previous iteration, or at
664 	 * the beginning if snap->count is 0.
665 	 */
666 	error = hammer_init_cursor(trans, &cursor, &ip->cache[0], NULL);
667 	if (error) {
668 		hammer_done_cursor(&cursor);
669 		return(error);
670 	}
671 
672 	cursor.asof = HAMMER_MAX_TID;
673 	cursor.flags |= HAMMER_CURSOR_BACKEND | HAMMER_CURSOR_ASOF;
674 
675 	bzero(&leaf, sizeof(leaf));
676 	leaf.base.obj_id = HAMMER_OBJID_ROOT;
677 	leaf.base.rec_type = HAMMER_RECTYPE_SNAPSHOT;
678 	leaf.base.create_tid = hammer_alloc_tid(hmp, 1);
679 	leaf.base.btype = HAMMER_BTREE_TYPE_RECORD;
680 	leaf.base.localization = ip->obj_localization + HAMMER_LOCALIZE_INODE;
681 	leaf.data_len = sizeof(struct hammer_snapshot_data);
682 
683 	while (snap->index < snap->count) {
684 		leaf.base.key = (int64_t)snap->snaps[snap->index].tid;
685 		cursor.key_beg = leaf.base;
686 		error = hammer_btree_lookup(&cursor);
687 		if (error == 0) {
688 			error = EEXIST;
689 			break;
690 		}
691 
692 		/*
693 		 * NOTE: Must reload key_beg after an ASOF search because
694 		 *	 the create_tid may have been modified during the
695 		 *	 search.
696 		 */
697 		cursor.flags &= ~HAMMER_CURSOR_ASOF;
698 		cursor.key_beg = leaf.base;
699 		error = hammer_create_at_cursor(&cursor, &leaf,
700 						&snap->snaps[snap->index],
701 						HAMMER_CREATE_MODE_SYS);
702 		if (error == EDEADLK) {
703 			hammer_done_cursor(&cursor);
704 			goto again;
705 		}
706 		cursor.flags |= HAMMER_CURSOR_ASOF;
707 		if (error)
708 			break;
709 		++snap->index;
710 	}
711 	snap->head.error = error;
712 	hammer_done_cursor(&cursor);
713 	hammer_unlock(&hmp->snapshot_lock);
714 	return(0);
715 }
716 
717 /*
718  * Delete snapshot transaction id(s) from the list of snapshots.
719  */
720 static
721 int
722 hammer_ioc_del_snapshot(hammer_transaction_t trans, hammer_inode_t ip,
723 			struct hammer_ioc_snapshot *snap)
724 {
725 	hammer_mount_t hmp = ip->hmp;
726 	struct hammer_cursor cursor;
727 	int error;
728 
729 	/*
730 	 * Validate structure
731 	 */
732 	if (snap->count > HAMMER_SNAPS_PER_IOCTL)
733 		return (EINVAL);
734 	if (snap->index > snap->count)
735 		return (EINVAL);
736 
737 	hammer_lock_ex(&hmp->snapshot_lock);
738 again:
739 	/*
740 	 * Look for keys starting after the previous iteration, or at
741 	 * the beginning if snap->count is 0.
742 	 */
743 	error = hammer_init_cursor(trans, &cursor, &ip->cache[0], NULL);
744 	if (error) {
745 		hammer_done_cursor(&cursor);
746 		return(error);
747 	}
748 
749 	cursor.key_beg.obj_id = HAMMER_OBJID_ROOT;
750 	cursor.key_beg.create_tid = 0;
751 	cursor.key_beg.delete_tid = 0;
752 	cursor.key_beg.obj_type = 0;
753 	cursor.key_beg.rec_type = HAMMER_RECTYPE_SNAPSHOT;
754 	cursor.key_beg.localization = ip->obj_localization + HAMMER_LOCALIZE_INODE;
755 	cursor.asof = HAMMER_MAX_TID;
756 	cursor.flags |= HAMMER_CURSOR_ASOF;
757 
758 	while (snap->index < snap->count) {
759 		cursor.key_beg.key = (int64_t)snap->snaps[snap->index].tid;
760 		error = hammer_btree_lookup(&cursor);
761 		if (error)
762 			break;
763 		error = hammer_btree_extract(&cursor, HAMMER_CURSOR_GET_LEAF);
764 		if (error)
765 			break;
766 		error = hammer_delete_at_cursor(&cursor, HAMMER_DELETE_DESTROY,
767 						0, 0, 0, NULL);
768 		if (error == EDEADLK) {
769 			hammer_done_cursor(&cursor);
770 			goto again;
771 		}
772 		if (error)
773 			break;
774 		++snap->index;
775 	}
776 	snap->head.error = error;
777 	hammer_done_cursor(&cursor);
778 	hammer_unlock(&hmp->snapshot_lock);
779 	return(0);
780 }
781 
782 /*
783  * Retrieve as many snapshot ids as possible or until the array is
784  * full, starting after the last transction id passed in.  If count
785  * is 0 we retrieve starting at the beginning.
786  *
787  * NOTE: Because the b-tree key field is signed but transaction ids
788  *       are unsigned the returned list will be signed-sorted instead
789  *	 of unsigned sorted.  The Caller must still sort the aggregate
790  *	 results.
791  */
792 static
793 int
794 hammer_ioc_get_snapshot(hammer_transaction_t trans, hammer_inode_t ip,
795 			struct hammer_ioc_snapshot *snap)
796 {
797 	struct hammer_cursor cursor;
798 	int error;
799 
800 	/*
801 	 * Validate structure
802 	 */
803 	if (snap->index != 0)
804 		return (EINVAL);
805 	if (snap->count > HAMMER_SNAPS_PER_IOCTL)
806 		return (EINVAL);
807 
808 	/*
809 	 * Look for keys starting after the previous iteration, or at
810 	 * the beginning if snap->count is 0.
811 	 */
812 	error = hammer_init_cursor(trans, &cursor, &ip->cache[0], NULL);
813 	if (error) {
814 		hammer_done_cursor(&cursor);
815 		return(error);
816 	}
817 
818 	cursor.key_beg.obj_id = HAMMER_OBJID_ROOT;
819 	cursor.key_beg.create_tid = 0;
820 	cursor.key_beg.delete_tid = 0;
821 	cursor.key_beg.obj_type = 0;
822 	cursor.key_beg.rec_type = HAMMER_RECTYPE_SNAPSHOT;
823 	cursor.key_beg.localization = ip->obj_localization + HAMMER_LOCALIZE_INODE;
824 	if (snap->count == 0)
825 		cursor.key_beg.key = HAMMER_MIN_KEY;
826 	else
827 		cursor.key_beg.key = (int64_t)snap->snaps[snap->count - 1].tid + 1;
828 
829 	cursor.key_end = cursor.key_beg;
830 	cursor.key_end.key = HAMMER_MAX_KEY;
831 	cursor.asof = HAMMER_MAX_TID;
832 	cursor.flags |= HAMMER_CURSOR_END_EXCLUSIVE | HAMMER_CURSOR_ASOF;
833 
834 	snap->count = 0;
835 
836 	error = hammer_btree_first(&cursor);
837 	while (error == 0 && snap->count < HAMMER_SNAPS_PER_IOCTL) {
838 		error = hammer_btree_extract(&cursor, HAMMER_CURSOR_GET_LEAF);
839 		if (error)
840 			break;
841 		if (cursor.leaf->base.rec_type == HAMMER_RECTYPE_SNAPSHOT) {
842 			error = hammer_btree_extract(
843 					     &cursor, HAMMER_CURSOR_GET_LEAF |
844 						      HAMMER_CURSOR_GET_DATA);
845 			snap->snaps[snap->count] = cursor.data->snap;
846 
847 			/*
848 			 * The snap data tid should match the key but might
849 			 * not due to a bug in the HAMMER v3 conversion code.
850 			 *
851 			 * This error will work itself out over time but we
852 			 * have to force a match or the snapshot will not
853 			 * be deletable.
854 			 */
855 			if (cursor.data->snap.tid !=
856 			    (hammer_tid_t)cursor.leaf->base.key) {
857 				kprintf("HAMMER: lo=%08x snapshot key "
858 					"0x%016jx data mismatch 0x%016jx\n",
859 					cursor.key_beg.localization,
860 					(uintmax_t)cursor.data->snap.tid,
861 					cursor.leaf->base.key);
862 				kprintf("HAMMER: Probably left over from the "
863 					"original v3 conversion, hammer "
864 					"cleanup should get it eventually\n");
865 				snap->snaps[snap->count].tid =
866 					cursor.leaf->base.key;
867 			}
868 			++snap->count;
869 		}
870 		error = hammer_btree_iterate(&cursor);
871 	}
872 
873 	if (error == ENOENT) {
874 		snap->head.flags |= HAMMER_IOC_SNAPSHOT_EOF;
875 		error = 0;
876 	}
877 	snap->head.error = error;
878 	hammer_done_cursor(&cursor);
879 	return(0);
880 }
881 
882 /*
883  * Retrieve the PFS hammer cleanup utility config record.  This is
884  * different (newer than) the PFS config.
885  */
886 static
887 int
888 hammer_ioc_get_config(hammer_transaction_t trans, hammer_inode_t ip,
889 			struct hammer_ioc_config *config)
890 {
891 	struct hammer_cursor cursor;
892 	int error;
893 
894 	error = hammer_init_cursor(trans, &cursor, &ip->cache[0], NULL);
895 	if (error) {
896 		hammer_done_cursor(&cursor);
897 		return(error);
898 	}
899 
900 	cursor.key_beg.obj_id = HAMMER_OBJID_ROOT;
901 	cursor.key_beg.create_tid = 0;
902 	cursor.key_beg.delete_tid = 0;
903 	cursor.key_beg.obj_type = 0;
904 	cursor.key_beg.rec_type = HAMMER_RECTYPE_CONFIG;
905 	cursor.key_beg.localization = ip->obj_localization + HAMMER_LOCALIZE_INODE;
906 	cursor.key_beg.key = 0;		/* config space page 0 */
907 
908 	cursor.asof = HAMMER_MAX_TID;
909 	cursor.flags |= HAMMER_CURSOR_ASOF;
910 
911 	error = hammer_btree_lookup(&cursor);
912 	if (error == 0) {
913 		error = hammer_btree_extract(&cursor, HAMMER_CURSOR_GET_LEAF |
914 						      HAMMER_CURSOR_GET_DATA);
915 		if (error == 0)
916 			config->config = cursor.data->config;
917 	}
918 	/* error can be ENOENT */
919 	config->head.error = error;
920 	hammer_done_cursor(&cursor);
921 	return(0);
922 }
923 
924 /*
925  * Retrieve the PFS hammer cleanup utility config record.  This is
926  * different (newer than) the PFS config.
927  *
928  * This is kinda a hack.
929  */
930 static
931 int
932 hammer_ioc_set_config(hammer_transaction_t trans, hammer_inode_t ip,
933 			struct hammer_ioc_config *config)
934 {
935 	struct hammer_btree_leaf_elm leaf;
936 	struct hammer_cursor cursor;
937 	hammer_mount_t hmp = ip->hmp;
938 	int error;
939 
940 again:
941 	error = hammer_init_cursor(trans, &cursor, &ip->cache[0], NULL);
942 	if (error) {
943 		hammer_done_cursor(&cursor);
944 		return(error);
945 	}
946 
947 	bzero(&leaf, sizeof(leaf));
948 	leaf.base.obj_id = HAMMER_OBJID_ROOT;
949 	leaf.base.rec_type = HAMMER_RECTYPE_CONFIG;
950 	leaf.base.create_tid = hammer_alloc_tid(hmp, 1);
951 	leaf.base.btype = HAMMER_BTREE_TYPE_RECORD;
952 	leaf.base.localization = ip->obj_localization + HAMMER_LOCALIZE_INODE;
953 	leaf.base.key = 0;	/* page 0 */
954 	leaf.data_len = sizeof(struct hammer_config_data);
955 
956 	cursor.key_beg = leaf.base;
957 
958 	cursor.asof = HAMMER_MAX_TID;
959 	cursor.flags |= HAMMER_CURSOR_BACKEND | HAMMER_CURSOR_ASOF;
960 
961 	error = hammer_btree_lookup(&cursor);
962 	if (error == 0) {
963 		error = hammer_btree_extract(&cursor, HAMMER_CURSOR_GET_LEAF |
964 						      HAMMER_CURSOR_GET_DATA);
965 		error = hammer_delete_at_cursor(&cursor, HAMMER_DELETE_DESTROY,
966 						0, 0, 0, NULL);
967 		if (error == EDEADLK) {
968 			hammer_done_cursor(&cursor);
969 			goto again;
970 		}
971 	}
972 	if (error == ENOENT)
973 		error = 0;
974 	if (error == 0) {
975 		/*
976 		 * NOTE: Must reload key_beg after an ASOF search because
977 		 *	 the create_tid may have been modified during the
978 		 *	 search.
979 		 */
980 		cursor.flags &= ~HAMMER_CURSOR_ASOF;
981 		cursor.key_beg = leaf.base;
982 		error = hammer_create_at_cursor(&cursor, &leaf,
983 						&config->config,
984 						HAMMER_CREATE_MODE_SYS);
985 		if (error == EDEADLK) {
986 			hammer_done_cursor(&cursor);
987 			goto again;
988 		}
989 	}
990 	config->head.error = error;
991 	hammer_done_cursor(&cursor);
992 	return(0);
993 }
994