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_flusher.c,v 1.8 2008/04/29 04:43:08 dillon Exp $
37 * HAMMER dependancy flusher thread
39 * Meta data updates create buffer dependancies which are arranged as a
45 static void hammer_flusher_thread(void *arg);
46 static void hammer_flusher_clean_loose_ios(hammer_mount_t hmp);
47 static void hammer_flusher_flush(hammer_mount_t hmp);
48 static int hammer_must_finalize_undo(hammer_mount_t hmp);
49 static void hammer_flusher_finalize(hammer_mount_t hmp,
50 hammer_volume_t root_volume, hammer_off_t start_offset);
53 hammer_flusher_sync(hammer_mount_t hmp)
57 if (hmp->flusher_td) {
58 seq = ++hmp->flusher_seq;
59 wakeup(&hmp->flusher_seq);
60 while ((int)(seq - hmp->flusher_act) > 0)
61 tsleep(&hmp->flusher_act, 0, "hmrfls", 0);
66 hammer_flusher_async(hammer_mount_t hmp)
68 if (hmp->flusher_td) {
70 wakeup(&hmp->flusher_seq);
75 hammer_flusher_create(hammer_mount_t hmp)
77 lwkt_create(hammer_flusher_thread, hmp, &hmp->flusher_td, NULL,
82 hammer_flusher_destroy(hammer_mount_t hmp)
84 if (hmp->flusher_td) {
85 hmp->flusher_exiting = 1;
87 wakeup(&hmp->flusher_seq);
88 while (hmp->flusher_td)
89 tsleep(&hmp->flusher_exiting, 0, "hmrwex", 0);
94 hammer_flusher_thread(void *arg)
96 hammer_mount_t hmp = arg;
99 hmp->flusher_demark = kmalloc(sizeof(struct hammer_inode),
100 M_HAMMER, M_WAITOK | M_ZERO);
101 TAILQ_INSERT_TAIL(&hmp->flush_list, hmp->flusher_demark, flush_entry);
104 seq = hmp->flusher_seq;
105 hammer_flusher_clean_loose_ios(hmp);
106 hammer_flusher_flush(hmp);
107 hammer_flusher_clean_loose_ios(hmp);
108 hmp->flusher_act = seq;
109 wakeup(&hmp->flusher_act);
112 * Loop if more got queued after our demark.
114 if (TAILQ_NEXT(hmp->flusher_demark, flush_entry))
117 if (hmp->flusher_exiting)
119 while (hmp->flusher_seq == hmp->flusher_act)
120 tsleep(&hmp->flusher_seq, 0, "hmrwwa", 0);
122 TAILQ_REMOVE(&hmp->flush_list, hmp->flusher_demark, flush_entry);
123 kfree(hmp->flusher_demark, M_HAMMER);
124 hmp->flusher_demark = NULL;
125 hmp->flusher_td = NULL;
126 wakeup(&hmp->flusher_exiting);
131 hammer_flusher_clean_loose_ios(hammer_mount_t hmp)
133 hammer_buffer_t buffer;
137 * loose ends - buffers without bp's aren't tracked by the kernel
138 * and can build up, so clean them out. This can occur when an
139 * IO completes on a buffer with no references left.
141 while ((io = TAILQ_FIRST(&hmp->lose_list)) != NULL) {
142 KKASSERT(io->mod_list == &hmp->lose_list);
143 TAILQ_REMOVE(io->mod_list, io, mod_entry);
145 hammer_ref(&io->lock);
147 hammer_rel_buffer(buffer, 0);
155 hammer_flusher_flush(hammer_mount_t hmp)
157 hammer_volume_t root_volume;
158 hammer_blockmap_t rootmap;
160 hammer_off_t start_offset;
163 root_volume = hammer_get_root_volume(hmp, &error);
164 rootmap = &hmp->blockmap[HAMMER_ZONE_UNDO_INDEX];
165 start_offset = rootmap->next_offset;
167 if (hammer_debug_general & 0x00010000)
170 TAILQ_REMOVE(&hmp->flush_list, hmp->flusher_demark, flush_entry);
171 TAILQ_INSERT_TAIL(&hmp->flush_list, hmp->flusher_demark, flush_entry);
173 while ((ip = TAILQ_FIRST(&hmp->flush_list)) != hmp->flusher_demark) {
174 TAILQ_REMOVE(&hmp->flush_list, ip, flush_entry);
177 * We inherit the inode ref from the flush list
179 ip->error = hammer_sync_inode(ip, (ip->vp ? 0 : 1));
180 hammer_flush_inode_done(ip);
181 if (hmp->locked_dirty_count > 64 ||
182 hammer_must_finalize_undo(hmp)) {
183 hammer_flusher_finalize(hmp, root_volume, start_offset);
184 start_offset = rootmap->next_offset;
187 hammer_flusher_finalize(hmp, root_volume, start_offset);
188 hammer_rel_volume(root_volume, 0);
192 * If the UNDO area gets over half full we have to flush it. We can't
193 * afford the UNDO area becoming completely full as that would break
194 * the crash recovery atomicy.
198 hammer_must_finalize_undo(hammer_mount_t hmp)
200 hammer_blockmap_t rootmap;
204 rootmap = &hmp->blockmap[HAMMER_ZONE_UNDO_INDEX];
206 if (rootmap->first_offset <= rootmap->next_offset) {
207 bytes = (int)(rootmap->next_offset - rootmap->first_offset);
209 bytes = (int)(rootmap->alloc_offset - rootmap->first_offset +
210 rootmap->next_offset);
212 max_bytes = (int)(rootmap->alloc_offset & HAMMER_OFF_SHORT_MASK);
213 if (bytes > max_bytes / 2)
215 return (bytes > max_bytes / 2);
219 * To finalize the flush we finish flushing all undo and data buffers
220 * still present, then we update the volume header and flush it,
221 * then we flush out the mata-data (that can now be undone).
223 * Note that as long as the undo fifo's start and end points do not
224 * match, we always must at least update the volume header.
226 * The sync_lock is used by other threads to issue modifying operations
227 * to HAMMER media without crossing a synchronization boundary or messing
228 * up the media synchronization operation. Specifically, the pruning
229 * the reblocking ioctls, and allowing the frontend strategy code to
230 * allocate media data space.
234 hammer_flusher_finalize(hammer_mount_t hmp, hammer_volume_t root_volume,
235 hammer_off_t start_offset)
237 hammer_blockmap_t rootmap;
240 hammer_lock_ex(&hmp->sync_lock);
245 while ((io = TAILQ_FIRST(&hmp->undo_list)) != NULL) {
246 KKASSERT(io->modify_refs == 0);
247 hammer_ref(&io->lock);
248 KKASSERT(io->type != HAMMER_STRUCTURE_VOLUME);
250 hammer_rel_buffer((hammer_buffer_t)io, 1);
256 while ((io = TAILQ_FIRST(&hmp->data_list)) != NULL) {
257 KKASSERT(io->modify_refs == 0);
258 hammer_ref(&io->lock);
259 KKASSERT(io->type != HAMMER_STRUCTURE_VOLUME);
261 hammer_rel_buffer((hammer_buffer_t)io, 1);
265 * Wait for I/O to complete
268 while (hmp->io_running_count) {
269 kprintf("W[%d]", hmp->io_running_count);
270 tsleep(&hmp->io_running_count, 0, "hmrfl1", 0);
275 * Update the volume header
277 rootmap = &hmp->blockmap[HAMMER_ZONE_UNDO_INDEX];
278 if (rootmap->first_offset != start_offset) {
279 hammer_modify_volume(NULL, root_volume, NULL, 0);
280 rootmap->first_offset = start_offset;
281 hammer_modify_volume_done(root_volume);
283 if (root_volume->ondisk->vol0_next_tid != hmp->next_tid) {
284 hammer_modify_volume(NULL, root_volume, NULL, 0);
285 root_volume->ondisk->vol0_next_tid = hmp->next_tid;
286 hammer_modify_volume_done(root_volume);
290 * Sync our cached blockmap array with the one in the root
293 if (root_volume->io.modified) {
294 bcopy(hmp->blockmap, root_volume->ondisk->vol0_blockmap,
295 sizeof(hmp->blockmap));
296 hammer_io_flush(&root_volume->io);
300 * Wait for I/O to complete
303 while (hmp->io_running_count) {
304 tsleep(&hmp->io_running_count, 0, "hmrfl2", 0);
311 while ((io = TAILQ_FIRST(&hmp->meta_list)) != NULL) {
312 KKASSERT(io->modify_refs == 0);
313 hammer_ref(&io->lock);
314 KKASSERT(io->type != HAMMER_STRUCTURE_VOLUME);
316 hammer_rel_buffer((hammer_buffer_t)io, 1);
318 hammer_unlock(&hmp->sync_lock);