HAMMER 25/many: Pruning code
[dragonfly/netmp.git] / sys / vfs / hammer / hammer_ioctl.c
blob347e6856eb5f9a952f3ce9e2012a06918d73264b
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:
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.
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.
34 * $DragonFly: src/sys/vfs/hammer/hammer_ioctl.c,v 1.2 2008/02/05 07:58:43 dillon Exp $
37 #include "hammer.h"
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);
44 int
45 hammer_ioctl(hammer_inode_t ip, u_long com, caddr_t data, int fflag,
46 struct ucred *cred)
48 int error;
50 error = suser_cred(cred, PRISON_ROOT);
52 switch(com) {
53 case HAMMERIOC_PRUNE:
54 if (error == 0) {
55 error = hammer_ioc_prune(ip,
56 (struct hammer_ioc_prune *)data);
58 break;
59 case HAMMERIOC_GETHISTORY:
60 error = hammer_ioc_gethistory(ip,
61 (struct hammer_ioc_history *)data);
62 break;
63 default:
64 error = EOPNOTSUPP;
65 break;
67 return (error);
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);
83 static int
84 hammer_ioc_prune(hammer_inode_t ip, struct hammer_ioc_prune *prune)
86 struct hammer_cursor cursor;
87 hammer_btree_elm_t elm;
88 int error;
89 int isdir;
90 int realign_cre;
91 int realign_del;
93 if (prune->nelms < 0 || prune->nelms > HAMMER_MAX_PRUNE_ELMS) {
94 return(EINVAL);
96 if (prune->beg_obj_id >= prune->end_obj_id) {
97 return(EINVAL);
100 retry:
101 error = hammer_init_cursor_hmp(&cursor, NULL, ip->hmp);
102 if (error) {
103 hammer_done_cursor(&cursor);
104 return(error);
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);
123 while (error == 0) {
124 elm = &cursor.node->ondisk->elms[cursor.index];
125 prune->cur_obj_id = elm->base.obj_id;
126 prune->cur_key = elm->base.key;
127 if (check_prune(prune, elm, &realign_cre, &realign_del) == 0) {
128 if (hammer_debug_general & 0x0200) {
129 kprintf("check %016llx %016llx: DELETE\n",
130 elm->base.obj_id, elm->base.key);
134 * NOTE: This can return EDEADLK
136 isdir = (elm->base.rec_type == HAMMER_RECTYPE_DIRENTRY);
138 error = hammer_delete_at_cursor(&cursor,
139 &prune->stat_bytes);
140 if (error)
141 break;
143 if (isdir)
144 ++prune->stat_dirrecords;
145 else
146 ++prune->stat_rawrecords;
147 } else if (realign_cre >= 0 || realign_del >= 0) {
148 error = realign_prune(prune, &cursor,
149 realign_cre, realign_del);
150 if (error == 0) {
151 cursor.flags |= HAMMER_CURSOR_ATEDISK;
152 if (hammer_debug_general & 0x0200) {
153 kprintf("check %016llx %016llx: "
154 "REALIGN\n",
155 elm->base.obj_id,
156 elm->base.key);
159 } else {
160 cursor.flags |= HAMMER_CURSOR_ATEDISK;
161 if (hammer_debug_general & 0x0100) {
162 kprintf("check %016llx %016llx: SKIP\n",
163 elm->base.obj_id, elm->base.key);
166 if (error == 0)
167 error = hammer_btree_iterate_reverse(&cursor);
169 if (error == ENOENT)
170 error = 0;
171 hammer_done_cursor(&cursor);
172 if (error == EDEADLK)
173 goto retry;
174 return(error);
178 * Check pruning list. The list must be sorted in descending order.
180 static int
181 check_prune(struct hammer_ioc_prune *prune, hammer_btree_elm_t elm,
182 int *realign_cre, int *realign_del)
184 struct hammer_ioc_prune_elm *scan;
185 int i;
187 *realign_cre = -1;
188 *realign_del = -1;
190 for (i = 0; i < prune->nelms; ++i) {
191 scan = &prune->elms[i];
194 * Locate the scan index covering the create and delete TIDs.
196 if (*realign_cre < 0 &&
197 elm->base.create_tid >= scan->beg_tid &&
198 elm->base.create_tid < scan->end_tid) {
199 *realign_cre = i;
201 if (*realign_del < 0 && elm->base.delete_tid &&
202 elm->base.delete_tid > scan->beg_tid &&
203 elm->base.delete_tid <= scan->end_tid) {
204 *realign_del = i;
208 * Now check for loop termination.
210 if (elm->base.create_tid >= scan->end_tid ||
211 elm->base.delete_tid > scan->end_tid) {
212 break;
216 * Now determine if we can delete the record.
218 if (elm->base.delete_tid &&
219 elm->base.create_tid >= scan->beg_tid &&
220 elm->base.delete_tid <= scan->end_tid &&
221 elm->base.create_tid / scan->mod_tid ==
222 elm->base.delete_tid / scan->mod_tid) {
223 return(0);
226 return(-1);
230 * Align the record to cover any gaps created through the deletion of
231 * records within the pruning space. If we were to just delete the records
232 * there would be gaps which in turn would cause a snapshot that is NOT on
233 * a pruning boundary to appear corrupt to the user. Forcing alignment
234 * of the create_tid and delete_tid for retained records 'reconnects'
235 * the previously contiguous space, making it contiguous again after the
236 * deletions.
238 * The use of a reverse iteration allows us to safely align the records and
239 * related elements without creating temporary overlaps. XXX we should
240 * add ordering dependancies for record buffers to guarantee consistency
241 * during recovery.
243 static int
244 realign_prune(struct hammer_ioc_prune *prune,
245 hammer_cursor_t cursor, int realign_cre, int realign_del)
247 hammer_btree_elm_t elm;
248 hammer_tid_t delta;
249 hammer_tid_t mod;
250 hammer_tid_t tid;
251 int error;
253 hammer_cursor_downgrade(cursor);
255 elm = &cursor->node->ondisk->elms[cursor->index];
256 ++prune->stat_realignments;
259 * Align the create_tid. By doing a reverse iteration we guarantee
260 * that all records after our current record have already been
261 * aligned, allowing us to safely correct the right-hand-boundary
262 * (because no record to our right if otherwise exactly matching
263 * will have a create_tid to the left of our aligned create_tid).
265 * Ordering is important here XXX but disk write ordering for
266 * inter-cluster corrections is not currently guaranteed.
268 error = 0;
269 if (realign_cre >= 0) {
270 mod = prune->elms[realign_cre].mod_tid;
271 delta = elm->leaf.base.create_tid % mod;
272 if (delta) {
273 tid = elm->leaf.base.create_tid - delta + mod;
275 /* can EDEADLK */
276 error = hammer_btree_correct_rhb(cursor, tid + 1);
277 if (error == 0) {
278 error = hammer_btree_extract(cursor,
279 HAMMER_CURSOR_GET_RECORD);
281 if (error == 0) {
282 /* can EDEADLK */
283 error = hammer_cursor_upgrade(cursor);
285 if (error == 0) {
286 hammer_modify_buffer(cursor->record_buffer);
287 cursor->record->base.base.create_tid = tid;
288 hammer_modify_node(cursor->node);
289 elm->leaf.base.create_tid = tid;
295 * Align the delete_tid. This only occurs if the record is historical
296 * was deleted at some point. Realigning the delete_tid does not
297 * move the record within the B-Tree but may cause it to temporarily
298 * overlap a record that has not yet been pruned.
300 if (error == 0 && realign_del >= 0) {
301 mod = prune->elms[realign_del].mod_tid;
302 delta = elm->leaf.base.delete_tid % mod;
303 hammer_modify_node(cursor->node);
304 if (delta) {
305 error = hammer_btree_extract(cursor,
306 HAMMER_CURSOR_GET_RECORD);
307 if (error == 0) {
308 elm->leaf.base.delete_tid =
309 elm->leaf.base.delete_tid -
310 delta + mod;
311 hammer_modify_buffer(cursor->record_buffer);
312 cursor->record->base.base.delete_tid =
313 elm->leaf.base.delete_tid;
317 return (error);
321 * Iterate through an object's inode or an object's records and record
322 * modification TIDs.
324 static void add_history(hammer_inode_t ip, struct hammer_ioc_history *hist,
325 hammer_btree_elm_t elm);
327 static
329 hammer_ioc_gethistory(hammer_inode_t ip, struct hammer_ioc_history *hist)
331 struct hammer_cursor cursor;
332 hammer_btree_elm_t elm;
333 int error;
336 * Validate the structure and initialize for return.
338 if (hist->beg_tid > hist->end_tid)
339 return(EINVAL);
340 if (hist->flags & HAMMER_IOC_HISTORY_ATKEY) {
341 if (hist->key > hist->nxt_key)
342 return(EINVAL);
345 hist->obj_id = ip->obj_id;
346 hist->count = 0;
347 hist->nxt_tid = hist->end_tid;
348 hist->flags &= ~HAMMER_IOC_HISTORY_NEXT_TID;
349 hist->flags &= ~HAMMER_IOC_HISTORY_NEXT_KEY;
350 hist->flags &= ~HAMMER_IOC_HISTORY_EOF;
351 hist->flags &= ~HAMMER_IOC_HISTORY_UNSYNCED;
352 if ((ip->flags & HAMMER_INODE_MODMASK) & ~HAMMER_INODE_ITIMES)
353 hist->flags |= HAMMER_IOC_HISTORY_UNSYNCED;
356 * Setup the cursor. We can't handle undeletable records
357 * (create_tid of 0) at the moment. A create_tid of 0 has
358 * a special meaning and cannot be specified in the cursor.
360 error = hammer_init_cursor_hmp(&cursor, &ip->cache[0], ip->hmp);
361 if (error) {
362 hammer_done_cursor(&cursor);
363 return(error);
366 cursor.key_beg.obj_id = hist->obj_id;
367 cursor.key_beg.create_tid = hist->beg_tid;
368 cursor.key_beg.delete_tid = 0;
369 cursor.key_beg.obj_type = 0;
370 if (cursor.key_beg.create_tid == HAMMER_MIN_TID)
371 cursor.key_beg.create_tid = 1;
373 cursor.key_end.obj_id = hist->obj_id;
374 cursor.key_end.create_tid = hist->end_tid;
375 cursor.key_end.delete_tid = 0;
376 cursor.key_end.obj_type = 0;
378 cursor.flags |= HAMMER_CURSOR_END_EXCLUSIVE;
380 if (hist->flags & HAMMER_IOC_HISTORY_ATKEY) {
382 * key-range within the file. For a regular file the
383 * on-disk key represents BASE+LEN, not BASE, so the
384 * first possible record containing the offset 'key'
385 * has an on-disk key of (key + 1).
387 cursor.key_beg.key = hist->key;
388 cursor.key_end.key = HAMMER_MAX_KEY;
390 switch(ip->ino_rec.base.base.obj_type) {
391 case HAMMER_OBJTYPE_REGFILE:
392 ++cursor.key_beg.key;
393 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
394 break;
395 case HAMMER_OBJTYPE_DIRECTORY:
396 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
397 break;
398 case HAMMER_OBJTYPE_DBFILE:
399 cursor.key_beg.rec_type = HAMMER_RECTYPE_DB;
400 break;
401 default:
402 error = EINVAL;
403 break;
405 cursor.key_end.rec_type = cursor.key_beg.rec_type;
406 } else {
408 * The inode itself.
410 cursor.key_beg.key = 0;
411 cursor.key_end.key = 0;
412 cursor.key_beg.rec_type = HAMMER_RECTYPE_INODE;
413 cursor.key_end.rec_type = HAMMER_RECTYPE_INODE;
416 error = hammer_btree_first(&cursor);
417 while (error == 0) {
418 elm = &cursor.node->ondisk->elms[cursor.index];
420 add_history(ip, hist, elm);
421 if (hist->flags & (HAMMER_IOC_HISTORY_NEXT_TID |
422 HAMMER_IOC_HISTORY_NEXT_KEY |
423 HAMMER_IOC_HISTORY_EOF)) {
424 break;
426 error = hammer_btree_iterate(&cursor);
428 if (error == ENOENT) {
429 hist->flags |= HAMMER_IOC_HISTORY_EOF;
430 error = 0;
432 hammer_done_cursor(&cursor);
433 return(error);
437 * Add the scanned element to the ioctl return structure. Some special
438 * casing is required for regular files to accomodate how data ranges are
439 * stored on-disk.
441 static void
442 add_history(hammer_inode_t ip, struct hammer_ioc_history *hist,
443 hammer_btree_elm_t elm)
445 if (elm->base.btype != HAMMER_BTREE_TYPE_RECORD)
446 return;
447 if ((hist->flags & HAMMER_IOC_HISTORY_ATKEY) &&
448 ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_REGFILE) {
450 * Adjust nxt_key
452 if (hist->nxt_key > elm->leaf.base.key - elm->leaf.data_len &&
453 hist->key < elm->leaf.base.key - elm->leaf.data_len) {
454 hist->nxt_key = elm->leaf.base.key - elm->leaf.data_len;
456 if (hist->nxt_key > elm->leaf.base.key)
457 hist->nxt_key = elm->leaf.base.key;
460 * Record is beyond MAXPHYS, there won't be any more records
461 * in the iteration covering the requested offset (key).
463 if (elm->leaf.base.key >= MAXPHYS &&
464 elm->leaf.base.key - MAXPHYS > hist->key) {
465 hist->flags |= HAMMER_IOC_HISTORY_NEXT_KEY;
469 * Data-range of record does not cover the key.
471 if (elm->leaf.base.key - elm->leaf.data_len > hist->key)
472 return;
474 } else if (hist->flags & HAMMER_IOC_HISTORY_ATKEY) {
476 * Adjust nxt_key
478 if (hist->nxt_key > elm->leaf.base.key &&
479 hist->key < elm->leaf.base.key) {
480 hist->nxt_key = elm->leaf.base.key;
484 * Record is beyond the requested key.
486 if (elm->leaf.base.key > hist->key)
487 hist->flags |= HAMMER_IOC_HISTORY_NEXT_KEY;
491 * Add create_tid if it is in-bounds.
493 if ((hist->count == 0 ||
494 elm->leaf.base.create_tid != hist->tid_ary[hist->count - 1]) &&
495 elm->leaf.base.create_tid >= hist->beg_tid &&
496 elm->leaf.base.create_tid < hist->end_tid) {
497 if (hist->count == HAMMER_MAX_HISTORY_ELMS) {
498 hist->nxt_tid = elm->leaf.base.create_tid;
499 hist->flags |= HAMMER_IOC_HISTORY_NEXT_TID;
500 return;
502 hist->tid_ary[hist->count++] = elm->leaf.base.create_tid;
506 * Add delete_tid if it is in-bounds. Note that different portions
507 * of the history may have overlapping data ranges with different
508 * delete_tid's. If this case occurs the delete_tid may match the
509 * create_tid of a following record. XXX
511 * [ ]
512 * [ ]
514 if (elm->leaf.base.delete_tid &&
515 elm->leaf.base.delete_tid >= hist->beg_tid &&
516 elm->leaf.base.delete_tid < hist->end_tid) {
517 if (hist->count == HAMMER_MAX_HISTORY_ELMS) {
518 hist->nxt_tid = elm->leaf.base.delete_tid;
519 hist->flags |= HAMMER_IOC_HISTORY_NEXT_TID;
520 return;
522 hist->tid_ary[hist->count++] = elm->leaf.base.delete_tid;