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
35 * HAMMER PFS ioctls - Manage pseudo-fs configurations
40 static int hammer_pfs_autodetect(struct hammer_ioc_pseudofs_rw *pfs,
42 static int hammer_pfs_rollback(hammer_transaction_t trans,
43 hammer_pseudofs_inmem_t pfsm,
44 hammer_tid_t trunc_tid);
45 static int hammer_pfs_delete_at_cursor(hammer_cursor_t cursor,
46 hammer_tid_t trunc_tid);
49 * Get mirroring/pseudo-fs information
51 * NOTE: The ip used for ioctl is not necessarily related to the PFS
52 * since this ioctl only requires PFS id (or upper 16 bits of ip localization).
55 hammer_ioc_get_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
56 struct hammer_ioc_pseudofs_rw *pfs)
58 hammer_pseudofs_inmem_t pfsm;
59 u_int32_t localization;
62 if ((error = hammer_pfs_autodetect(pfs, ip)) != 0)
64 localization = (u_int32_t)pfs->pfs_id << 16;
65 pfs->bytes = sizeof(struct hammer_pseudofs_data);
66 pfs->version = HAMMER_IOC_PSEUDOFS_VERSION;
68 pfsm = hammer_load_pseudofs(trans, localization, &error);
70 hammer_rel_pseudofs(trans->hmp, pfsm);
75 * If the PFS is a master the sync tid is set by normal operation
76 * rather than the mirroring code, and will always track the
77 * real HAMMER filesystem.
79 * We use flush_tid1, which is the highest fully committed TID.
80 * flush_tid2 is the TID most recently flushed, but the UNDO hasn't
81 * caught up to it yet so a crash will roll us back to flush_tid1.
83 if ((pfsm->pfsd.mirror_flags & HAMMER_PFSD_SLAVE) == 0)
84 pfsm->pfsd.sync_end_tid = trans->hmp->flush_tid1;
87 * Copy out to userland.
90 if (pfs->ondisk && error == 0)
91 error = copyout(&pfsm->pfsd, pfs->ondisk, sizeof(pfsm->pfsd));
92 hammer_rel_pseudofs(trans->hmp, pfsm);
97 * Set mirroring/pseudo-fs information
99 * NOTE: The ip used for ioctl is not necessarily related to the PFS
100 * since this ioctl only requires PFS id (or upper 16 bits of ip localization).
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)
116 if (error == 0 && pfs->ondisk) {
118 * Load the PFS so we can modify our in-core copy. Ignore
121 pfsm = hammer_load_pseudofs(trans, localization, &error);
122 error = copyin(pfs->ondisk, &pfsm->pfsd, sizeof(pfsm->pfsd));
125 * Save it back, create a root inode if we are in master
126 * mode and no root exists.
128 * We do not create root inodes for slaves, the root inode
129 * must be mirrored from the master.
132 (pfsm->pfsd.mirror_flags & HAMMER_PFSD_SLAVE) == 0) {
133 error = hammer_mkroot_pseudofs(trans, cred, pfsm);
136 error = hammer_save_pseudofs(trans, pfsm);
139 * Wakeup anyone waiting for a TID update for this PFS
141 wakeup(&pfsm->pfsd.sync_end_tid);
142 hammer_rel_pseudofs(trans->hmp, pfsm);
148 * Upgrade a slave to a master
150 * This is fairly easy to do, but we must physically undo any partial syncs
151 * for transaction ids > sync_end_tid. Effective, we must do a partial
154 * NOTE: The ip used for ioctl is not necessarily related to the PFS
155 * since this ioctl only requires PFS id (or upper 16 bits of ip localization).
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 and update sync_end_tid.
198 * We previously did not update sync_end_tid in consideration for a slave
199 * upgraded to a master and then downgraded again, but this completely breaks
200 * the case where one starts with a master and then downgrades to a slave,
201 * then upgrades again.
203 * NOTE: The ip used for ioctl is not necessarily related to the PFS
204 * since this ioctl only requires PFS id (or upper 16 bits of ip localization).
207 hammer_ioc_downgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
208 struct hammer_ioc_pseudofs_rw *pfs)
210 hammer_mount_t hmp = trans->hmp;
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 if (pfsm->pfsd.sync_end_tid < hmp->flush_tid1)
226 pfsm->pfsd.sync_end_tid = hmp->flush_tid1;
227 error = hammer_save_pseudofs(trans, pfsm);
230 hammer_rel_pseudofs(trans->hmp, pfsm);
237 * We can destroy a PFS by scanning and deleting all of its records in the
238 * B-Tree. The hammer utility will delete the softlink in the primary
241 * NOTE: The ip used for ioctl is not necessarily related to the PFS
242 * since this ioctl only requires PFS id (or upper 16 bits of ip localization).
245 hammer_ioc_destroy_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
246 struct hammer_ioc_pseudofs_rw *pfs)
248 hammer_pseudofs_inmem_t pfsm;
249 u_int32_t localization;
252 if ((error = hammer_pfs_autodetect(pfs, ip)) != 0)
254 localization = (u_int32_t)pfs->pfs_id << 16;
256 if ((error = hammer_unload_pseudofs(trans, localization)) != 0)
259 pfsm = hammer_load_pseudofs(trans, localization, &error);
261 error = hammer_pfs_rollback(trans, pfsm, 0);
263 pfsm->pfsd.mirror_flags |= HAMMER_PFSD_DELETED;
264 error = hammer_save_pseudofs(trans, pfsm);
267 hammer_rel_pseudofs(trans->hmp, pfsm);
268 if (error == EINTR) {
269 pfs->head.flags |= HAMMER_IOC_HEAD_INTR;
276 * Wait for the PFS to sync past the specified TID
279 hammer_ioc_wait_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
280 struct hammer_ioc_pseudofs_rw *pfs)
282 hammer_pseudofs_inmem_t pfsm;
283 struct hammer_pseudofs_data pfsd;
284 u_int32_t localization;
289 if ((error = hammer_pfs_autodetect(pfs, ip)) != 0)
291 localization = (u_int32_t)pfs->pfs_id << 16;
293 if ((error = copyin(pfs->ondisk, &pfsd, sizeof(pfsd))) != 0)
296 pfsm = hammer_load_pseudofs(trans, localization, &error);
298 if (pfsm->pfsd.mirror_flags & HAMMER_PFSD_SLAVE) {
299 tid = pfsm->pfsd.sync_end_tid;
300 waitp = &pfsm->pfsd.sync_end_tid;
302 tid = trans->hmp->flush_tid1;
303 waitp = &trans->hmp->flush_tid1;
305 if (tid <= pfsd.sync_end_tid)
306 tsleep(waitp, PCATCH, "hmrmwt", 0);
308 hammer_rel_pseudofs(trans->hmp, pfsm);
309 if (error == EINTR) {
310 pfs->head.flags |= HAMMER_IOC_HEAD_INTR;
318 * Auto-detect the pseudofs and do basic bounds checking.
322 hammer_pfs_autodetect(struct hammer_ioc_pseudofs_rw *pfs, hammer_inode_t ip)
326 if (pfs->pfs_id == -1)
327 pfs->pfs_id = (int)(ip->obj_localization >> 16);
328 if (pfs->pfs_id < 0 || pfs->pfs_id >= HAMMER_MAX_PFS)
330 if (pfs->bytes < sizeof(struct hammer_pseudofs_data))
336 * Rollback the specified PFS to (trunc_tid - 1), removing everything
337 * greater or equal to trunc_tid. The PFS must not have been in no-mirror
338 * mode or the MIRROR_FILTERED scan will not work properly.
340 * This is typically used to remove any partial syncs when upgrading a
341 * slave to a master. It can theoretically also be used to rollback
342 * any PFS, including PFS#0, BUT ONLY TO POINTS THAT HAVE NOT YET BEEN
343 * PRUNED, and to points that are older only if they are on a retained
344 * (pruning softlink) boundary.
346 * Rollbacks destroy information. If you don't mind inode numbers changing
347 * a better way would be to cpdup a snapshot back onto the master.
351 hammer_pfs_rollback(hammer_transaction_t trans,
352 hammer_pseudofs_inmem_t pfsm,
353 hammer_tid_t trunc_tid)
355 struct hammer_cmirror cmirror;
356 struct hammer_cursor cursor;
357 struct hammer_base_elm key_cur;
361 bzero(&cmirror, sizeof(cmirror));
362 bzero(&key_cur, sizeof(key_cur));
363 key_cur.localization = HAMMER_MIN_LOCALIZATION + pfsm->localization;
364 key_cur.obj_id = HAMMER_MIN_OBJID;
365 key_cur.key = HAMMER_MIN_KEY;
366 key_cur.create_tid = 1;
367 key_cur.rec_type = HAMMER_MIN_RECTYPE;
369 seq = trans->hmp->flusher.done;
372 error = hammer_init_cursor(trans, &cursor, NULL, NULL);
374 hammer_done_cursor(&cursor);
377 cursor.key_beg = key_cur;
378 cursor.key_end.localization = HAMMER_MAX_LOCALIZATION +
380 cursor.key_end.obj_id = HAMMER_MAX_OBJID;
381 cursor.key_end.key = HAMMER_MAX_KEY;
382 cursor.key_end.create_tid = HAMMER_MAX_TID;
383 cursor.key_end.rec_type = HAMMER_MAX_RECTYPE;
385 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
386 cursor.flags |= HAMMER_CURSOR_BACKEND;
389 * Do an optimized scan of only records created or modified
390 * >= trunc_tid, so we can fix up those records. We must
391 * still check the TIDs but this greatly reduces the size of
394 cursor.flags |= HAMMER_CURSOR_MIRROR_FILTERED;
395 cursor.cmirror = &cmirror;
396 cmirror.mirror_tid = trunc_tid;
398 error = hammer_btree_first(&cursor);
401 * Abort the rollback.
404 error = hammer_signal_check(trans->hmp);
410 * We only care about leafs. Internal nodes can be returned
411 * in mirror-filtered mode (they are used to generate SKIP
412 * mrecords), but we don't need them for this code.
414 * WARNING: See warnings in hammer_unlock_cursor() function.
416 cursor.flags |= HAMMER_CURSOR_ATEDISK;
417 if (cursor.node->ondisk->type == HAMMER_BTREE_TYPE_LEAF) {
418 key_cur = cursor.node->ondisk->elms[cursor.index].base;
419 error = hammer_pfs_delete_at_cursor(&cursor, trunc_tid);
422 while (hammer_flusher_meta_halflimit(trans->hmp) ||
423 hammer_flusher_undo_exhausted(trans, 2)) {
424 hammer_unlock_cursor(&cursor);
425 hammer_flusher_wait(trans->hmp, seq);
426 hammer_lock_cursor(&cursor);
427 seq = hammer_flusher_async_one(trans->hmp);
431 error = hammer_btree_iterate(&cursor);
435 hammer_done_cursor(&cursor);
436 if (error == EDEADLK)
443 * Helper function - perform rollback on a B-Tree element given trunc_tid.
445 * If create_tid >= trunc_tid the record is physically destroyed.
446 * If delete_tid >= trunc_tid it will be set to 0, undeleting the record.
450 hammer_pfs_delete_at_cursor(hammer_cursor_t cursor, hammer_tid_t trunc_tid)
452 hammer_btree_leaf_elm_t elm;
455 elm = &cursor->node->ondisk->elms[cursor->index].leaf;
456 if (elm->base.create_tid < trunc_tid &&
457 elm->base.delete_tid < trunc_tid) {
461 if (elm->base.create_tid >= trunc_tid) {
462 error = hammer_delete_at_cursor(
463 cursor, HAMMER_DELETE_DESTROY,
464 cursor->trans->tid, cursor->trans->time32,
466 } else if (elm->base.delete_tid >= trunc_tid) {
467 error = hammer_delete_at_cursor(
468 cursor, HAMMER_DELETE_ADJUST,