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_pfs.c,v 1.5 2008/07/31 04:42:04 dillon Exp $
37 * HAMMER PFS ioctls - Manage pseudo-fs configurations
42 static int hammer_pfs_autodetect(struct hammer_ioc_pseudofs_rw *pfs,
44 static int hammer_pfs_rollback(hammer_transaction_t trans,
45 hammer_pseudofs_inmem_t pfsm,
46 hammer_tid_t trunc_tid);
47 static int hammer_pfs_delete_at_cursor(hammer_cursor_t cursor,
48 hammer_tid_t trunc_tid);
51 * Get mirroring/pseudo-fs information
53 * NOTE: The ip used for ioctl is not necessarily related to the PFS
56 hammer_ioc_get_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
57 struct hammer_ioc_pseudofs_rw *pfs)
59 hammer_pseudofs_inmem_t pfsm;
60 u_int32_t localization;
63 if ((error = hammer_pfs_autodetect(pfs, ip)) != 0)
65 localization = (u_int32_t)pfs->pfs_id << 16;
66 pfs->bytes = sizeof(struct hammer_pseudofs_data);
67 pfs->version = HAMMER_IOC_PSEUDOFS_VERSION;
69 pfsm = hammer_load_pseudofs(trans, localization, &error);
71 hammer_rel_pseudofs(trans->hmp, pfsm);
76 * If the PFS is a master the sync tid is set by normal operation
77 * rather then the mirroring code, and will always track the
78 * real HAMMER filesystem.
80 * We use flush_tid1, which is the highest fully committed TID.
81 * flush_tid2 is the TID most recently flushed, but the UNDO hasn't
82 * caught up to it yet so a crash will roll us back to flush_tid1.
84 if ((pfsm->pfsd.mirror_flags & HAMMER_PFSD_SLAVE) == 0)
85 pfsm->pfsd.sync_end_tid = trans->hmp->flush_tid1;
88 * Copy out to userland.
91 if (pfs->ondisk && error == 0)
92 error = copyout(&pfsm->pfsd, pfs->ondisk, sizeof(pfsm->pfsd));
93 hammer_rel_pseudofs(trans->hmp, pfsm);
98 * Set mirroring/pseudo-fs information
100 * NOTE: The ip used for ioctl is not necessarily related to the PFS
103 hammer_ioc_set_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
104 struct ucred *cred, struct hammer_ioc_pseudofs_rw *pfs)
106 hammer_pseudofs_inmem_t pfsm;
107 u_int32_t localization;
110 if ((error = hammer_pfs_autodetect(pfs, ip)) != 0)
112 localization = (u_int32_t)pfs->pfs_id << 16;
113 if (pfs->version != HAMMER_IOC_PSEUDOFS_VERSION)
115 localization = (u_int32_t)pfs->pfs_id << 16;
117 if (error == 0 && pfs->ondisk) {
119 * Load the PFS so we can modify our in-core copy. Ignore
122 pfsm = hammer_load_pseudofs(trans, localization, &error);
123 error = copyin(pfs->ondisk, &pfsm->pfsd, sizeof(pfsm->pfsd));
126 * Save it back, create a root inode if we are in master
127 * mode and no root exists.
129 * We do not create root inodes for slaves, the root inode
130 * must be mirrored from the master.
133 (pfsm->pfsd.mirror_flags & HAMMER_PFSD_SLAVE) == 0) {
134 error = hammer_mkroot_pseudofs(trans, cred, pfsm);
137 error = hammer_save_pseudofs(trans, pfsm);
140 * Wakeup anyone waiting for a TID update for this PFS
142 wakeup(&pfsm->pfsd.sync_end_tid);
143 hammer_rel_pseudofs(trans->hmp, pfsm);
149 * Upgrade a slave to a master
151 * This is fairly easy to do, but we must physically undo any partial syncs
152 * for transaction ids > sync_end_tid. Effective, we must do a partial
155 * NOTE: The ip used for ioctl is not necessarily related to the PFS
158 hammer_ioc_upgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
159 struct hammer_ioc_pseudofs_rw *pfs)
161 hammer_pseudofs_inmem_t pfsm;
162 u_int32_t localization;
165 if ((error = hammer_pfs_autodetect(pfs, ip)) != 0)
167 localization = (u_int32_t)pfs->pfs_id << 16;
168 if ((error = hammer_unload_pseudofs(trans, localization)) != 0)
172 * A master id must be set when upgrading
174 pfsm = hammer_load_pseudofs(trans, localization, &error);
176 if ((pfsm->pfsd.mirror_flags & HAMMER_PFSD_SLAVE) != 0) {
177 error = hammer_pfs_rollback(trans, pfsm,
178 pfsm->pfsd.sync_end_tid + 1);
180 pfsm->pfsd.mirror_flags &= ~HAMMER_PFSD_SLAVE;
181 error = hammer_save_pseudofs(trans, pfsm);
185 hammer_rel_pseudofs(trans->hmp, pfsm);
186 if (error == EINTR) {
187 pfs->head.flags |= HAMMER_IOC_HEAD_INTR;
194 * Downgrade a master to a slave
196 * This is really easy to do, just set the SLAVE flag.
198 * We also leave sync_end_tid intact... the field is not used in master
199 * mode (vol0_next_tid overrides it), but if someone switches to master
200 * mode accidently and then back to slave mode we don't want it to change.
201 * Eventually it will be used as the cross-synchronization TID in
202 * multi-master mode, and we don't want to mess with it for that feature
205 * NOTE: The ip used for ioctl is not necessarily related to the PFS
208 hammer_ioc_downgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
209 struct hammer_ioc_pseudofs_rw *pfs)
211 hammer_pseudofs_inmem_t pfsm;
212 u_int32_t localization;
215 if ((error = hammer_pfs_autodetect(pfs, ip)) != 0)
217 localization = (u_int32_t)pfs->pfs_id << 16;
218 if ((error = hammer_unload_pseudofs(trans, localization)) != 0)
221 pfsm = hammer_load_pseudofs(trans, localization, &error);
223 if ((pfsm->pfsd.mirror_flags & HAMMER_PFSD_SLAVE) == 0) {
224 pfsm->pfsd.mirror_flags |= HAMMER_PFSD_SLAVE;
225 error = hammer_save_pseudofs(trans, pfsm);
228 hammer_rel_pseudofs(trans->hmp, pfsm);
235 * We can destroy a PFS by scanning and deleting all of its records in the
236 * B-Tree. The hammer utility will delete the softlink in the primary
239 * NOTE: The ip used for ioctl is not necessarily related to the PFS
242 hammer_ioc_destroy_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
243 struct hammer_ioc_pseudofs_rw *pfs)
245 hammer_pseudofs_inmem_t pfsm;
246 u_int32_t localization;
249 if ((error = hammer_pfs_autodetect(pfs, ip)) != 0)
251 localization = (u_int32_t)pfs->pfs_id << 16;
253 if ((error = hammer_unload_pseudofs(trans, localization)) != 0)
256 pfsm = hammer_load_pseudofs(trans, localization, &error);
258 error = hammer_pfs_rollback(trans, pfsm, 0);
260 pfsm->pfsd.mirror_flags |= HAMMER_PFSD_DELETED;
261 error = hammer_save_pseudofs(trans, pfsm);
264 hammer_rel_pseudofs(trans->hmp, pfsm);
265 if (error == EINTR) {
266 pfs->head.flags |= HAMMER_IOC_HEAD_INTR;
273 * Wait for the PFS to sync past the specified TID
276 hammer_ioc_wait_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
277 struct hammer_ioc_pseudofs_rw *pfs)
279 hammer_pseudofs_inmem_t pfsm;
280 struct hammer_pseudofs_data pfsd;
281 u_int32_t localization;
286 if ((error = hammer_pfs_autodetect(pfs, ip)) != 0)
288 localization = (u_int32_t)pfs->pfs_id << 16;
290 if ((error = copyin(pfs->ondisk, &pfsd, sizeof(pfsd))) != 0)
293 pfsm = hammer_load_pseudofs(trans, localization, &error);
295 if (pfsm->pfsd.mirror_flags & HAMMER_PFSD_SLAVE) {
296 tid = pfsm->pfsd.sync_end_tid;
297 waitp = &pfsm->pfsd.sync_end_tid;
299 tid = trans->hmp->flush_tid1;
300 waitp = &trans->hmp->flush_tid1;
302 if (tid <= pfsd.sync_end_tid)
303 tsleep(waitp, PCATCH, "hmrmwt", 0);
305 hammer_rel_pseudofs(trans->hmp, pfsm);
306 if (error == EINTR) {
307 pfs->head.flags |= HAMMER_IOC_HEAD_INTR;
315 * Auto-detect the pseudofs and do basic bounds checking.
319 hammer_pfs_autodetect(struct hammer_ioc_pseudofs_rw *pfs, hammer_inode_t ip)
323 if (pfs->pfs_id == -1)
324 pfs->pfs_id = (int)(ip->obj_localization >> 16);
325 if (pfs->pfs_id < 0 || pfs->pfs_id >= HAMMER_MAX_PFS)
327 if (pfs->bytes < sizeof(struct hammer_pseudofs_data))
333 * Rollback the specified PFS to (trunc_tid - 1), removing everything
334 * greater or equal to trunc_tid. The PFS must not have been in no-mirror
335 * mode or the MIRROR_FILTERED scan will not work properly.
337 * This is typically used to remove any partial syncs when upgrading a
338 * slave to a master. It can theoretically also be used to rollback
339 * any PFS, including PFS#0, BUT ONLY TO POINTS THAT HAVE NOT YET BEEN
340 * PRUNED, and to points that are older only if they are on a retained
341 * (pruning softlink) boundary.
343 * Rollbacks destroy information. If you don't mind inode numbers changing
344 * a better way would be to cpdup a snapshot back onto the master.
348 hammer_pfs_rollback(hammer_transaction_t trans,
349 hammer_pseudofs_inmem_t pfsm,
350 hammer_tid_t trunc_tid)
352 struct hammer_cmirror cmirror;
353 struct hammer_cursor cursor;
354 struct hammer_base_elm key_cur;
358 bzero(&cmirror, sizeof(cmirror));
359 bzero(&key_cur, sizeof(key_cur));
360 key_cur.localization = HAMMER_MIN_LOCALIZATION + pfsm->localization;
361 key_cur.obj_id = HAMMER_MIN_OBJID;
362 key_cur.key = HAMMER_MIN_KEY;
363 key_cur.create_tid = 1;
364 key_cur.rec_type = HAMMER_MIN_RECTYPE;
366 seq = trans->hmp->flusher.act;
369 error = hammer_init_cursor(trans, &cursor, NULL, NULL);
371 hammer_done_cursor(&cursor);
374 cursor.key_beg = key_cur;
375 cursor.key_end.localization = HAMMER_MAX_LOCALIZATION +
377 cursor.key_end.obj_id = HAMMER_MAX_OBJID;
378 cursor.key_end.key = HAMMER_MAX_KEY;
379 cursor.key_end.create_tid = HAMMER_MAX_TID;
380 cursor.key_end.rec_type = HAMMER_MAX_RECTYPE;
382 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
383 cursor.flags |= HAMMER_CURSOR_BACKEND;
386 * Do an optimized scan of only records created or modified
387 * >= trunc_tid, so we can fix up those records. We must
388 * still check the TIDs but this greatly reduces the size of
391 cursor.flags |= HAMMER_CURSOR_MIRROR_FILTERED;
392 cursor.cmirror = &cmirror;
393 cmirror.mirror_tid = trunc_tid;
395 error = hammer_btree_first(&cursor);
398 * Abort the rollback.
401 error = hammer_signal_check(trans->hmp);
407 * We only care about leafs. Internal nodes can be returned
408 * in mirror-filtered mode (they are used to generate SKIP
409 * mrecords), but we don't need them for this code.
411 * WARNING: See warnings in hammer_unlock_cursor() function.
413 cursor.flags |= HAMMER_CURSOR_ATEDISK;
414 if (cursor.node->ondisk->type == HAMMER_BTREE_TYPE_LEAF) {
415 key_cur = cursor.node->ondisk->elms[cursor.index].base;
416 error = hammer_pfs_delete_at_cursor(&cursor, trunc_tid);
419 while (hammer_flusher_meta_halflimit(trans->hmp) ||
420 hammer_flusher_undo_exhausted(trans, 2)) {
421 hammer_unlock_cursor(&cursor);
422 hammer_flusher_wait(trans->hmp, seq);
423 hammer_lock_cursor(&cursor);
424 seq = hammer_flusher_async_one(trans->hmp);
428 error = hammer_btree_iterate(&cursor);
432 hammer_done_cursor(&cursor);
433 if (error == EDEADLK)
440 * Helper function - perform rollback on a B-Tree element given trunc_tid.
442 * If create_tid >= trunc_tid the record is physically destroyed.
443 * If delete_tid >= trunc_tid it will be set to 0, undeleting the record.
447 hammer_pfs_delete_at_cursor(hammer_cursor_t cursor, hammer_tid_t trunc_tid)
449 hammer_btree_leaf_elm_t elm;
450 hammer_transaction_t trans;
453 elm = &cursor->node->ondisk->elms[cursor->index].leaf;
454 if (elm->base.create_tid < trunc_tid &&
455 elm->base.delete_tid < trunc_tid) {
458 trans = cursor->trans;
460 if (elm->base.create_tid >= trunc_tid) {
461 error = hammer_delete_at_cursor(
462 cursor, HAMMER_DELETE_DESTROY,
463 cursor->trans->tid, cursor->trans->time32,
465 } else if (elm->base.delete_tid >= trunc_tid) {
466 error = hammer_delete_at_cursor(
467 cursor, HAMMER_DELETE_ADJUST,