2 * Copyright (c) 2007-2008 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
37 static uint32_t ocp_allocbit(hammer_objid_cache_t ocp
, uint32_t n
);
41 * Start a standard transaction.
43 * May be called without fs_token
46 hammer_start_transaction(hammer_transaction_t trans
, hammer_mount_t hmp
)
51 trans
->type
= HAMMER_TRANS_STD
;
53 trans
->rootvol
= hammer_get_root_volume(hmp
, &error
);
56 trans
->sync_lock_refs
= 0;
60 trans
->time
= (unsigned long)ts
.tv_sec
* 1000000ULL +
62 trans
->time32
= (uint32_t)ts
.tv_sec
;
66 * Start a simple read-only transaction. This will not stall.
68 * May be called without fs_token
71 hammer_simple_transaction(hammer_transaction_t trans
, hammer_mount_t hmp
)
76 trans
->type
= HAMMER_TRANS_RO
;
78 trans
->rootvol
= hammer_get_root_volume(hmp
, &error
);
81 trans
->sync_lock_refs
= 0;
85 trans
->time
= (unsigned long)ts
.tv_sec
* 1000000ULL +
87 trans
->time32
= (uint32_t)ts
.tv_sec
;
91 * Start a transaction using a particular TID. Used by the sync code.
92 * This does not stall.
94 * This routine may only be called from the flusher thread. We predispose
95 * sync_lock_refs, implying serialization against the synchronization stage
96 * (which the flusher is responsible for).
99 hammer_start_transaction_fls(hammer_transaction_t trans
, hammer_mount_t hmp
)
104 bzero(trans
, sizeof(*trans
));
106 trans
->type
= HAMMER_TRANS_FLS
;
108 trans
->rootvol
= hammer_get_root_volume(hmp
, &error
);
109 KKASSERT(error
== 0);
110 trans
->tid
= hammer_alloc_tid(hmp
, 1);
111 trans
->sync_lock_refs
= 1;
115 trans
->time
= (unsigned long)ts
.tv_sec
* 1000000ULL +
117 trans
->time32
= (uint32_t)ts
.tv_sec
;
121 * May be called without fs_token
124 hammer_done_transaction(hammer_transaction_t trans
)
126 int expected_lock_refs __debugvar
;
128 hammer_rel_volume(trans
->rootvol
, 0);
129 trans
->rootvol
= NULL
;
130 expected_lock_refs
= (trans
->type
== HAMMER_TRANS_FLS
) ? 1 : 0;
131 KKASSERT(trans
->sync_lock_refs
== expected_lock_refs
);
132 trans
->sync_lock_refs
= 0;
133 if (trans
->type
!= HAMMER_TRANS_FLS
) {
134 if (trans
->flags
& HAMMER_TRANSF_NEWINODE
) {
135 lwkt_gettoken(&trans
->hmp
->fs_token
);
136 hammer_inode_waitreclaims(trans
);
137 lwkt_reltoken(&trans
->hmp
->fs_token
);
143 * Allocate (count) TIDs. If running in multi-master mode the returned
144 * base will be aligned to a 16-count plus the master id (0-15).
145 * Multi-master mode allows non-conflicting to run and new objects to be
146 * created on multiple masters in parallel. The transaction id identifies
147 * the original master. The object_id is also subject to this rule in
148 * order to allow objects to be created on multiple masters in parallel.
150 * Directories may pre-allocate a large number of object ids (100,000).
152 * NOTE: There is no longer a requirement that successive transaction
153 * ids be 2 apart for separator generation.
155 * NOTE: When called by pseudo-backends such as ioctls the allocated
156 * TID will be larger then the current flush TID, if a flush is running,
157 * so any mirroring will pick the records up on a later flush.
159 * NOTE: HAMMER1 does not support multi-master clustering as of 2015.
162 hammer_alloc_tid(hammer_mount_t hmp
, int count
)
166 if (hmp
->master_id
< 0) {
167 tid
= hmp
->next_tid
+ 1;
168 hmp
->next_tid
= tid
+ count
;
170 tid
= (hmp
->next_tid
+ HAMMER_MAX_MASTERS
) &
171 ~(hammer_tid_t
)(HAMMER_MAX_MASTERS
- 1);
172 hmp
->next_tid
= tid
+ count
* HAMMER_MAX_MASTERS
;
173 tid
|= hmp
->master_id
;
175 if (tid
>= 0xFFFFFFFFFF000000ULL
)
176 hpanic("Ran out of TIDs!");
177 if (hammer_debug_tid
)
178 hdkprintf("%016jx\n", (intmax_t)tid
);
183 * Allocate an object id.
185 * We use the upper OBJID_CACHE_BITS bits of the namekey to try to match
186 * the low bits of the objid we allocate.
189 hammer_alloc_objid(hammer_mount_t hmp
, hammer_inode_t dip
, int64_t namekey
)
191 hammer_objid_cache_t ocp
;
195 while ((ocp
= dip
->objid_cache
) == NULL
) {
196 if (hmp
->objid_cache_count
< OBJID_CACHE_SIZE
) {
197 ocp
= kmalloc(sizeof(*ocp
), hmp
->m_misc
,
199 ocp
->base_tid
= hammer_alloc_tid(hmp
,
200 OBJID_CACHE_BULK
* 2);
201 ocp
->base_tid
+= OBJID_CACHE_BULK_MASK64
;
202 ocp
->base_tid
&= ~OBJID_CACHE_BULK_MASK64
;
203 /* may have blocked, recheck */
204 if (dip
->objid_cache
== NULL
) {
205 TAILQ_INSERT_TAIL(&hmp
->objid_cache_list
,
207 ++hmp
->objid_cache_count
;
208 dip
->objid_cache
= ocp
;
211 kfree(ocp
, hmp
->m_misc
);
215 * Steal one from another directory?
217 * Throw away ocp's that are more then half full, they
218 * aren't worth stealing.
220 ocp
= TAILQ_FIRST(&hmp
->objid_cache_list
);
222 ocp
->dip
->objid_cache
= NULL
;
223 if (ocp
->count
>= OBJID_CACHE_BULK
/ 2) {
224 TAILQ_REMOVE(&hmp
->objid_cache_list
,
226 --hmp
->objid_cache_count
;
227 kfree(ocp
, hmp
->m_misc
);
229 dip
->objid_cache
= ocp
;
234 TAILQ_REMOVE(&hmp
->objid_cache_list
, ocp
, entry
);
237 * Allocate inode numbers uniformly.
240 n
= (namekey
>> (63 - OBJID_CACHE_BULK_BITS
)) & OBJID_CACHE_BULK_MASK
;
241 n
= ocp_allocbit(ocp
, n
);
242 tid
= ocp
->base_tid
+ n
;
246 * The TID is incremented by 1 or by 16 depending what mode the
247 * mount is operating in.
249 ocp
->next_tid
+= (hmp
->master_id
< 0) ? 1 : HAMMER_MAX_MASTERS
;
251 if (ocp
->count
>= OBJID_CACHE_BULK
* 3 / 4) {
252 dip
->objid_cache
= NULL
;
253 --hmp
->objid_cache_count
;
255 kfree(ocp
, hmp
->m_misc
);
257 TAILQ_INSERT_TAIL(&hmp
->objid_cache_list
, ocp
, entry
);
263 * Allocate a bit starting with bit n. Wrap if necessary.
265 * This routine is only ever called if a bit is available somewhere
269 ocp_allocbit(hammer_objid_cache_t ocp
, uint32_t n
)
276 while (ocp
->bm1
[n0
] & (1 << n
)) {
277 if (ocp
->bm0
& (1 << n0
)) {
280 } else if (++n
== 32) {
286 ocp
->bm1
[n0
] |= 1 << n
;
287 if (ocp
->bm1
[n0
] == 0xFFFFFFFFU
)
289 return((n0
<< 5) + n
);
293 hammer_clear_objid(hammer_inode_t dip
)
295 hammer_objid_cache_t ocp
;
297 if ((ocp
= dip
->objid_cache
) != NULL
) {
298 dip
->objid_cache
= NULL
;
300 TAILQ_REMOVE(&dip
->hmp
->objid_cache_list
, ocp
, entry
);
301 TAILQ_INSERT_HEAD(&dip
->hmp
->objid_cache_list
, ocp
, entry
);
306 hammer_destroy_objid_cache(hammer_mount_t hmp
)
308 hammer_objid_cache_t ocp
;
310 while ((ocp
= TAILQ_FIRST(&hmp
->objid_cache_list
)) != NULL
) {
311 TAILQ_REMOVE(&hmp
->objid_cache_list
, ocp
, entry
);
313 ocp
->dip
->objid_cache
= NULL
;
314 kfree(ocp
, hmp
->m_misc
);
315 --hmp
->objid_cache_count
;
317 KKASSERT(hmp
->objid_cache_count
== 0);