HAMMER 2/many - core mount and unmount code now works, the B-Tree search
[dragonfly.git] / sys / vfs / hammer / hammer_ondisk.c
1 /*
2  * Copyright (c) 2007 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:
10  * 
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.
20  * 
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.
33  * 
34  * $DragonFly: src/sys/vfs/hammer/hammer_ondisk.c,v 1.2 2007/11/02 00:57:15 dillon Exp $
35  */
36 /*
37  * Manage HAMMER's on-disk structures.  These routines are primarily
38  * responsible for interfacing with the kernel's I/O subsystem and for
39  * managing in-memory structures.
40  */
41
42 #include "hammer.h"
43 #include <sys/fcntl.h>
44 #include <sys/nlookup.h>
45 #include <sys/buf.h>
46 #include <sys/buf2.h>
47
48 static void hammer_free_volume(struct hammer_volume *volume);
49 static void initbuffer(hammer_alist_t live, hammer_fsbuf_head_t head,
50                         u_int64_t type);
51 static void alloc_new_buffer(struct hammer_cluster *cluster,
52                         hammer_alist_t live, u_int64_t type, int32_t nelements,
53                         int32_t start,
54                         int *errorp, struct hammer_buffer **bufferp);
55 #if 0
56 static void readhammerbuf(struct hammer_volume *vol, void *data,
57                         int64_t offset);
58 static void writehammerbuf(struct hammer_volume *vol, const void *data,
59                         int64_t offset);
60 #endif
61 static int64_t calculate_cluster_offset(struct hammer_volume *vol,
62                         int32_t clu_no);
63 static int64_t calculate_supercl_offset(struct hammer_volume *vol,
64                         int32_t scl_no);
65
66
67 struct hammer_alist_config Buf_alist_config;
68 struct hammer_alist_config Vol_normal_alist_config;
69 struct hammer_alist_config Vol_super_alist_config;
70 struct hammer_alist_config Supercl_alist_config;
71 struct hammer_alist_config Clu_master_alist_config;
72 struct hammer_alist_config Clu_slave_alist_config;
73
74 /*
75  * Red-Black tree support for various structures
76  */
77 static int
78 hammer_ino_rb_compare(struct hammer_inode *ip1, struct hammer_inode *ip2)
79 {
80         if (ip1->obj_id < ip2->obj_id)
81                 return(-1);
82         if (ip1->obj_id > ip2->obj_id)
83                 return(1);
84         if (ip1->obj_asof < ip2->obj_asof)
85                 return(-1);
86         if (ip1->obj_asof > ip2->obj_asof)
87                 return(1);
88         return(0);
89 }
90
91 static int
92 hammer_inode_info_cmp(hammer_inode_info_t info, struct hammer_inode *ip)
93 {
94         if (info->obj_id < ip->obj_id)
95                 return(-1);
96         if (info->obj_id > ip->obj_id)
97                 return(1);
98         if (info->obj_asof < ip->obj_asof)
99                 return(-1);
100         if (info->obj_asof > ip->obj_asof)
101                 return(1);
102         return(0);
103 }
104
105 static int
106 hammer_vol_rb_compare(struct hammer_volume *vol1, struct hammer_volume *vol2)
107 {
108         if (vol1->vol_no < vol2->vol_no)
109                 return(-1);
110         if (vol1->vol_no > vol2->vol_no)
111                 return(1);
112         return(0);
113 }
114
115 static int
116 hammer_scl_rb_compare(struct hammer_supercl *cl1, struct hammer_supercl *cl2)
117 {
118         if (cl1->scl_no < cl2->scl_no)
119                 return(-1);
120         if (cl1->scl_no > cl2->scl_no)
121                 return(1);
122         return(0);
123 }
124
125 static int
126 hammer_clu_rb_compare(struct hammer_cluster *cl1, struct hammer_cluster *cl2)
127 {
128         if (cl1->clu_no < cl2->clu_no)
129                 return(-1);
130         if (cl1->clu_no > cl2->clu_no)
131                 return(1);
132         return(0);
133 }
134
135 static int
136 hammer_buf_rb_compare(struct hammer_buffer *buf1, struct hammer_buffer *buf2)
137 {
138         if (buf1->buf_no < buf2->buf_no)
139                 return(-1);
140         if (buf1->buf_no > buf2->buf_no)
141                 return(1);
142         return(0);
143 }
144
145 /*
146  * Note: The lookup function for hammer_ino_rb_tree winds up being named
147  * hammer_ino_rb_tree_RB_LOOKUP_INFO(root, info).  The other lookup
148  * functions are normal, e.g. hammer_clu_rb_tree_RB_LOOKUP(root, clu_no).
149  */
150 RB_GENERATE(hammer_ino_rb_tree, hammer_inode, rb_node, hammer_ino_rb_compare);
151 RB_GENERATE_XLOOKUP(hammer_ino_rb_tree, INFO, hammer_inode, rb_node,
152                 hammer_inode_info_cmp, hammer_inode_info_t);
153 RB_GENERATE2(hammer_vol_rb_tree, hammer_volume, rb_node,
154              hammer_vol_rb_compare, int32_t, vol_no);
155 RB_GENERATE2(hammer_scl_rb_tree, hammer_supercl, rb_node,
156              hammer_scl_rb_compare, int32_t, scl_no);
157 RB_GENERATE2(hammer_clu_rb_tree, hammer_cluster, rb_node,
158              hammer_clu_rb_compare, int32_t, clu_no);
159 RB_GENERATE2(hammer_buf_rb_tree, hammer_buffer, rb_node,
160              hammer_buf_rb_compare, int32_t, buf_no);
161
162 /*
163  * Load a HAMMER volume by name.  Returns 0 on success or a positive error
164  * code on failure.  Volumes must be loaded at mount time, get_volume() will
165  * not load a new volume.
166  *
167  * Calls made to hammer_load_volume() or single-threaded
168  */
169 int
170 hammer_load_volume(struct hammer_mount *hmp, const char *volname)
171 {
172         struct mount *mp;
173         struct hammer_volume *volume;
174         struct hammer_volume_ondisk *ondisk;
175         struct nlookupdata nd;
176         struct buf *bp = NULL;
177         int error;
178         int ronly;
179
180         mp = hmp->mp;
181         ronly = ((mp->mnt_flag & MNT_RDONLY) ? 1 : 0);
182
183         /*
184          * Allocate a volume structure
185          */
186         volume = kmalloc(sizeof(*volume), M_HAMMER, M_WAITOK|M_ZERO);
187         volume->vol_name = kstrdup(volname, M_HAMMER);
188         volume->hmp = hmp;
189
190         /*
191          * Get the device vnode
192          */
193         error = nlookup_init(&nd, volume->vol_name, UIO_SYSSPACE, NLC_FOLLOW);
194         if (error == 0)
195                 error = nlookup(&nd);
196         if (error == 0)
197                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &volume->devvp);
198         nlookup_done(&nd);
199         if (error == 0) {
200                 vn_isdisk(volume->devvp, &error);
201         }
202         if (error == 0) {
203                 vn_lock(volume->devvp, LK_EXCLUSIVE | LK_RETRY);
204                 error = VOP_OPEN(volume->devvp, (ronly ? FREAD : FREAD|FWRITE),
205                                  FSCRED, NULL);
206                 vn_unlock(volume->devvp);
207         }
208         if (error) {
209                 hammer_free_volume(volume);
210                 return(error);
211         }
212
213         /*
214          * Extract the volume number from the volume header and do various
215          * sanity checks.
216          */
217         error = bread(volume->devvp, 0LL, HAMMER_BUFSIZE, &bp);
218         if (error)
219                 goto late_failure;
220         ondisk = (void *)bp->b_data;
221         if (ondisk->head.buf_type != HAMMER_FSBUF_VOLUME) {
222                 kprintf("hammer_mount: volume %s has an invalid header\n",
223                         volume->vol_name);
224                 error = EFTYPE;
225                 goto late_failure;
226         }
227         volume->vol_no = ondisk->vol_no;
228         volume->cluster_base = ondisk->vol_beg;
229         volume->vol_clsize = ondisk->vol_clsize;
230         volume->vol_flags = ondisk->vol_flags;
231         RB_INIT(&volume->rb_clus_root);
232         RB_INIT(&volume->rb_scls_root);
233
234         if (RB_EMPTY(&hmp->rb_vols_root)) {
235                 hmp->fsid = ondisk->vol_fsid;
236         } else if (bcmp(&hmp->fsid, &ondisk->vol_fsid, sizeof(uuid_t))) {
237                 kprintf("hammer_mount: volume %s's fsid does not match "
238                         "other volumes\n", volume->vol_name);
239                 error = EFTYPE;
240                 goto late_failure;
241         }
242
243         /*
244          * Insert the volume structure into the red-black tree.
245          */
246         if (RB_INSERT(hammer_vol_rb_tree, &hmp->rb_vols_root, volume)) {
247                 kprintf("hammer_mount: volume %s has a duplicate vol_no %d\n",
248                         volume->vol_name, volume->vol_no);
249                 error = EEXIST;
250         }
251
252         /*
253          * Set the root volume and load the root cluster.
254          */
255         if (error == 0 && ondisk->vol_rootvol == ondisk->vol_no) {
256                 hmp->rootvol = volume;
257                 hmp->rootcl = hammer_get_cluster(volume,
258                                                  ondisk->vol0_root_clu_no,
259                                                  &error, 0);
260         }
261 late_failure:
262         if (bp)
263                 brelse(bp);
264         if (error) {
265                 /*vinvalbuf(volume->devvp, V_SAVE, 0, 0);*/
266                 VOP_CLOSE(volume->devvp, ronly ? FREAD : FREAD|FWRITE);
267                 hammer_free_volume(volume);
268         }
269         return (error);
270 }
271
272 /*
273  * Unload and free a HAMMER volume.  Must return >= 0 to continue scan
274  * so returns -1 on failure.
275  */
276 int
277 hammer_unload_volume(struct hammer_volume *volume, void *data __unused)
278 {
279         struct hammer_mount *hmp = volume->hmp;
280
281         /*
282          * Sync clusters, sync volume
283          */
284
285         /*
286          * Clean up
287          */
288         if (hmp->rootvol == volume) {
289                 if (hmp->rootcl) {
290                         hammer_put_cluster(hmp->rootcl);
291                         hmp->rootcl = NULL;
292                 }
293                 hmp->rootvol = NULL;
294         }
295
296         /* flush_volume */
297         KKASSERT(volume->bp == NULL);
298
299         /*
300          * Destroy the structure
301          */
302         RB_REMOVE(hammer_vol_rb_tree, &hmp->rb_vols_root, volume);
303         hammer_free_volume(volume);
304         return(0);
305 }
306
307 static
308 void
309 hammer_free_volume(struct hammer_volume *volume)
310 {
311         if (volume->vol_name) {
312                 kfree(volume->vol_name, M_HAMMER);
313                 volume->vol_name = NULL;
314         }
315         if (volume->devvp) {
316                 vrele(volume->devvp);
317                 volume->devvp = NULL;
318         }
319         kfree(volume, M_HAMMER);
320 }
321
322 /*
323  * Get a HAMMER volume.  The volume must already exist.
324  */
325 struct hammer_volume *
326 hammer_get_volume(struct hammer_mount *hmp, int32_t vol_no, int *errorp)
327 {
328         struct hammer_volume *volume;
329         struct hammer_volume_ondisk *ondisk;
330
331         /*
332          * Locate the volume structure
333          */
334         volume = RB_LOOKUP(hammer_vol_rb_tree, &hmp->rb_vols_root, vol_no);
335         if (volume == NULL) {
336                 *errorp = ENOENT;
337                 return(NULL);
338         }
339
340         /*
341          * Load the ondisk buffer if necessary
342          */
343         hammer_lock(&volume->lock);
344         if (volume->ondisk == NULL) {
345                 *errorp = bread(volume->devvp, 0LL, HAMMER_BUFSIZE,
346                                 &volume->bp);
347                 if (*errorp) {
348                         hammer_unlock(&volume->lock);
349                         return(NULL);
350                 }
351                 volume->ondisk = ondisk = (void *)volume->bp->b_data;
352
353                 /*
354                  * Configure the volume's A-lists.  These are used to
355                  * allocate clusters.
356                  */
357                 if (volume->vol_flags & HAMMER_VOLF_USINGSUPERCL) {
358                         volume->alist.config = &Vol_super_alist_config;
359                         volume->alist.meta = ondisk->vol_almeta.super;
360                         volume->alist.info = volume;
361                 } else {
362                         volume->alist.config = &Vol_normal_alist_config;
363                         volume->alist.meta = ondisk->vol_almeta.normal;
364                         volume->alist.info = NULL;
365                 }
366                 hammer_alist_init(&volume->alist);
367         }
368         *errorp = 0;
369         return(volume);
370 }
371
372 void
373 hammer_put_volume(struct hammer_volume *volume)
374 {
375         if (hammer_islastref(&volume->lock)) {
376                 if (volume->bp) {
377                         if (volume->modified) {
378                                 bdwrite(volume->bp);
379                         } else {
380                                 brelse(volume->bp);
381                         }
382                         volume->bp = NULL;
383                         volume->ondisk = NULL;
384                 }
385                 volume->modified = 0;
386         }
387         hammer_unlock(&volume->lock);
388 }
389
390 struct hammer_supercl *
391 hammer_get_supercl(struct hammer_volume *volume, int32_t scl_no,
392                    int *errorp, int isnew)
393 {
394         struct hammer_supercl_ondisk *ondisk;
395         struct hammer_supercl *supercl;
396
397         /*
398          * Locate and lock the super-cluster structure, creating one
399          * if necessary.
400          */
401 again:
402         supercl = RB_LOOKUP(hammer_scl_rb_tree, &volume->rb_scls_root, scl_no);
403         if (supercl == NULL) {
404                 supercl = kmalloc(sizeof(*supercl), M_HAMMER, M_WAITOK|M_ZERO);
405                 supercl->scl_no = scl_no;
406                 supercl->volume = volume;
407                 supercl->scl_offset = calculate_supercl_offset(volume, scl_no);
408                 hammer_lock(&supercl->lock);
409
410                 /*
411                  * Insert the cluster into the RB tree and handle late
412                  * collisions.
413                  */
414                 if (RB_INSERT(hammer_scl_rb_tree, &volume->rb_scls_root, supercl)) {
415                         hammer_unlock(&supercl->lock);
416                         kfree(supercl, M_HAMMER);
417                         goto again;
418                 }
419         } else {
420                 hammer_lock(&supercl->lock);
421         }
422
423         /*
424          * Load the cluster's on-disk info
425          */
426         *errorp = 0;
427         if (supercl->ondisk == NULL) {
428                 if (isnew) {
429                         supercl->bp = getblk(volume->devvp, supercl->scl_offset,
430                                              HAMMER_BUFSIZE, 0, 0);
431                         vfs_bio_clrbuf(supercl->bp);
432                         supercl->modified = 1;
433                 } else {
434                         *errorp = bread(volume->devvp, supercl->scl_offset,
435                                         HAMMER_BUFSIZE, &supercl->bp);
436                 }
437                 if (*errorp) {
438                         hammer_unlock(&supercl->lock);
439                         return(NULL);
440                 }
441                 BUF_KERNPROC(supercl->bp);
442                 supercl->ondisk = ondisk = (void *)supercl->bp->b_data;
443
444                 supercl->alist.config = &Supercl_alist_config;
445                 supercl->alist.meta = ondisk->scl_meta;
446                 supercl->alist.info = NULL;
447
448                 /*
449                  * If this is a new super-cluster we have to initialize
450                  * various ondisk structural elements.  The caller is
451                  * responsible for the remainder.
452                  */
453                 if (isnew) {
454                         struct hammer_alist_live dummy;
455
456                         dummy.config = &Buf_alist_config;
457                         dummy.meta = ondisk->head.buf_almeta;
458                         dummy.info = NULL;
459                         initbuffer(&dummy, &ondisk->head, HAMMER_FSBUF_SUPERCL);
460                         hammer_alist_init(&supercl->alist);
461                 }
462         } else if (isnew) {
463                 vfs_bio_clrbuf(supercl->bp);
464                 supercl->modified = 1;
465         }
466         return (supercl);
467 }
468
469 void
470 hammer_put_supercl(struct hammer_supercl *supercl)
471 {
472         if (hammer_islastref(&supercl->lock)) {
473                 if (supercl->bp) {
474                         if (supercl->modified) {
475                                 bdwrite(supercl->bp);
476                         } else {
477                                 brelse(supercl->bp);
478                         }
479                         supercl->bp = NULL;
480                         supercl->ondisk = NULL;
481                 }
482                 supercl->modified = 0;
483         }
484         hammer_unlock(&supercl->lock);
485 }
486
487 struct hammer_cluster *
488 hammer_get_cluster(struct hammer_volume *volume, int32_t clu_no,
489                    int *errorp, int isnew)
490 {
491         struct hammer_cluster_ondisk *ondisk;
492         struct hammer_cluster *cluster;
493
494         /*
495          * Locate and lock the cluster structure, creating one if necessary.
496          */
497 again:
498         cluster = RB_LOOKUP(hammer_clu_rb_tree, &volume->rb_clus_root, clu_no);
499         if (cluster == NULL) {
500                 cluster = kmalloc(sizeof(*cluster), M_HAMMER, M_WAITOK|M_ZERO);
501                 cluster->clu_no = clu_no;
502                 cluster->volume = volume;
503                 cluster->clu_offset = calculate_cluster_offset(volume, clu_no);
504                 RB_INIT(&cluster->rb_bufs_root);
505                 hammer_lock(&cluster->lock);
506
507                 /*
508                  * Insert the cluster into the RB tree and handle late
509                  * collisions.
510                  */
511                 if (RB_INSERT(hammer_clu_rb_tree, &volume->rb_clus_root, cluster)) {
512                         hammer_unlock(&cluster->lock);
513                         kfree(cluster, M_HAMMER);
514                         goto again;
515                 }
516         } else {
517                 hammer_lock(&cluster->lock);
518         }
519
520         /*
521          * Load the cluster's on-disk info
522          */
523         *errorp = 0;
524         if (cluster->ondisk == NULL) {
525                 if (isnew) {
526                         cluster->bp = getblk(volume->devvp, cluster->clu_offset,
527                                              HAMMER_BUFSIZE, 0, 0);
528                         vfs_bio_clrbuf(cluster->bp);
529                         cluster->modified = 1;
530                 } else {
531                         *errorp = bread(volume->devvp, cluster->clu_offset,
532                                         HAMMER_BUFSIZE, &cluster->bp);
533                 }
534                 if (*errorp) {
535                         hammer_unlock(&cluster->lock);
536                         return(NULL);
537                 }
538                 BUF_KERNPROC(cluster->bp);
539                 cluster->ondisk = ondisk = (void *)cluster->bp->b_data;
540
541                 cluster->alist_master.config = &Clu_master_alist_config;
542                 cluster->alist_master.meta = ondisk->clu_master_meta;
543                 cluster->alist_btree.config = &Clu_slave_alist_config;
544                 cluster->alist_btree.meta = ondisk->clu_btree_meta;
545                 cluster->alist_btree.info = cluster;
546                 cluster->alist_record.config = &Clu_slave_alist_config;
547                 cluster->alist_record.meta = ondisk->clu_record_meta;
548                 cluster->alist_record.info = cluster;
549                 cluster->alist_mdata.config = &Clu_slave_alist_config;
550                 cluster->alist_mdata.meta = ondisk->clu_mdata_meta;
551                 cluster->alist_mdata.info = cluster;
552
553                 /*
554                  * If this is a new cluster we have to initialize
555                  * various ondisk structural elements.  The caller is
556                  * responsible for the remainder.
557                  */
558                 if (isnew) {
559                         struct hammer_alist_live dummy;
560
561                         dummy.config = &Buf_alist_config;
562                         dummy.meta = ondisk->head.buf_almeta;
563                         dummy.info = NULL;
564                         initbuffer(&dummy, &ondisk->head, HAMMER_FSBUF_CLUSTER);
565
566                         hammer_alist_init(&cluster->alist_master);
567                         hammer_alist_init(&cluster->alist_btree);
568                         hammer_alist_init(&cluster->alist_record);
569                         hammer_alist_init(&cluster->alist_mdata);
570                 }
571         } else if (isnew) {
572                 vfs_bio_clrbuf(cluster->bp);
573                 cluster->modified = 1;
574         }
575         return(cluster);
576 }
577
578 void
579 hammer_put_cluster(struct hammer_cluster *cluster)
580 {
581         if (hammer_islastref(&cluster->lock)) {
582                 if (cluster->bp) {
583                         if (cluster->modified) {
584                                 bdwrite(cluster->bp);
585                         } else {
586                                 brelse(cluster->bp);
587                         }
588                         cluster->bp = NULL;
589                         cluster->ondisk = NULL;
590                 }
591                 cluster->modified = 0;
592         }
593         hammer_unlock(&cluster->lock);
594 }
595
596 /*
597  * Get a buffer from a cluster.  Note that buffer #0 is the cluster header
598  * itself and may not be retrieved with this function.
599  *
600  * If buf_type is 0 the buffer already exists in-memory or on-disk.
601  * Otherwise a new buffer is initialized with the specified buffer type.
602  */
603 struct hammer_buffer *
604 hammer_get_buffer(struct hammer_cluster *cluster, int32_t buf_no,
605                   int64_t buf_type, int *errorp)
606 {
607         hammer_fsbuf_ondisk_t ondisk;
608         struct hammer_buffer *buffer;
609
610         /*
611          * Find the buffer.  Note that buffer 0 corresponds to the cluster
612          * header and should never be requested.
613          */
614         KKASSERT(buf_no != 0);
615
616         /*
617          * Locate and lock the buffer structure, creating one if necessary.
618          */
619 again:
620         buffer = RB_LOOKUP(hammer_buf_rb_tree, &cluster->rb_bufs_root, buf_no);
621         if (buffer == NULL) {
622                 buffer = kmalloc(sizeof(*cluster), M_HAMMER, M_WAITOK|M_ZERO);
623                 buffer->buf_no = buf_no;
624                 buffer->cluster = cluster;
625                 buffer->volume = cluster->volume;
626                 buffer->buf_offset = cluster->clu_offset +
627                                      (buf_no * HAMMER_BUFSIZE);
628                 hammer_lock(&buffer->lock);
629
630                 /*
631                  * Insert the cluster into the RB tree and handle late
632                  * collisions.
633                  */
634                 if (RB_INSERT(hammer_buf_rb_tree, &cluster->rb_bufs_root, buffer)) {
635                         hammer_unlock(&buffer->lock);
636                         kfree(buffer, M_HAMMER);
637                         goto again;
638                 }
639         } else {
640                 hammer_lock(&buffer->lock);
641         }
642
643         *errorp = 0;
644         if (buffer->ondisk == NULL) {
645                 if (buf_type) {
646                         buffer->bp = getblk(buffer->volume->devvp,
647                                             buffer->buf_offset,
648                                             HAMMER_BUFSIZE, 0, 0);
649                         vfs_bio_clrbuf(buffer->bp);
650                         buffer->modified = 1;
651                 } else {
652                         *errorp = bread(buffer->volume->devvp,
653                                         buffer->buf_offset,
654                                         HAMMER_BUFSIZE, &buffer->bp);
655                 }
656                 if (*errorp) {
657                         hammer_unlock(&buffer->lock);
658                         return(NULL);
659                 }
660                 BUF_KERNPROC(buffer->bp);
661                 buffer->ondisk = ondisk = (void *)buffer->bp->b_data;
662                 buffer->alist.config = &Buf_alist_config;
663                 buffer->alist.meta = ondisk->head.buf_almeta;
664
665                 if (buf_type) {
666                         initbuffer(&buffer->alist, &ondisk->head, buf_type);
667                 }
668         } else if (buf_type) {
669                 vfs_bio_clrbuf(buffer->bp);
670                 buffer->modified = 1;
671         }
672         return (buffer);
673 }
674
675 void
676 hammer_put_buffer(struct hammer_buffer *buffer)
677 {
678         if (hammer_islastref(&buffer->lock)) {
679                 if (buffer->bp) {
680                         if (buffer->modified) {
681                                 bdwrite(buffer->bp);
682                         } else {
683                                 brelse(buffer->bp);
684                         }
685                         buffer->bp = NULL;
686                         buffer->ondisk = NULL;
687                 }
688                 buffer->modified = 0;
689         }
690         hammer_unlock(&buffer->lock);
691 }
692
693 void
694 hammer_dup_buffer(struct hammer_buffer **bufferp, struct hammer_buffer *buffer)
695 {
696         if (buffer != *bufferp) {
697                 if (buffer)
698                         hammer_lock(&buffer->lock);
699                 if (*bufferp)
700                         hammer_put_buffer(*bufferp);
701                 *bufferp = buffer;
702         }
703 }
704
705 void
706 hammer_dup_cluster(struct hammer_cluster **clusterp,
707                    struct hammer_cluster *cluster)
708 {
709         if (cluster != *clusterp) {
710                 if (cluster)
711                         hammer_lock(&cluster->lock);
712                 if (*clusterp)
713                         hammer_put_cluster(*clusterp);
714                 *clusterp = cluster;
715         }
716 }
717
718 /*
719  * Allocate HAMMER elements - btree nodes, data storage, and record elements
720  *
721  * The passed *bufferp should be initialized to NULL.  On successive calls
722  * *bufferp caches the most recent buffer used until put away by the caller.
723  * Note that previously returned pointers using the cached buffer become
724  * invalid on successive calls which reuse *bufferp.
725  *
726  * All allocations first attempt to use the block found at the specified
727  * iterator.  If that fails the first available block is used.  If that
728  * fails a new buffer is allocated and associated with the buffer type
729  * A-list and the element is allocated out of the new buffer.
730  */
731 void *
732 hammer_alloc_btree(struct hammer_cluster *cluster,
733                     int *errorp, struct hammer_buffer **bufferp)
734 {
735         struct hammer_buffer *buffer;
736         hammer_alist_t live;
737         int32_t elm_no;
738         int32_t buf_no;
739         void *item;
740
741         /*
742          * Allocate a B-Tree element
743          */
744         live = &cluster->alist_btree;
745         elm_no = hammer_alist_alloc_fwd(live, 1, cluster->ondisk->idx_index);
746         if (elm_no == HAMMER_ALIST_BLOCK_NONE)
747                 elm_no = hammer_alist_alloc_fwd(live, 1, 0);
748         if (elm_no == HAMMER_ALIST_BLOCK_NONE) {
749                 alloc_new_buffer(cluster, live,
750                                  HAMMER_FSBUF_BTREE, HAMMER_BTREE_NODES,
751                                  cluster->ondisk->idx_index, errorp, bufferp);
752                 elm_no = hammer_alist_alloc(live, 1);
753                 KKASSERT(elm_no != HAMMER_ALIST_BLOCK_NONE);
754         }
755         cluster->ondisk->idx_index = elm_no;
756
757         /*
758          * Load and return the B-Tree element
759          */
760         buf_no = elm_no / HAMMER_FSBUF_MAXBLKS;
761         buffer = *bufferp;
762         if (buffer == NULL || buffer->cluster != cluster ||
763             buffer->buf_no != buf_no) {
764                 if (buffer)
765                         hammer_put_buffer(buffer);
766                 buffer = hammer_get_buffer(cluster, buf_no, 0, errorp);
767                 *bufferp = buffer;
768         }
769         KKASSERT(buffer->ondisk->head.buf_type == HAMMER_FSBUF_BTREE);
770         item = &buffer->ondisk->btree.nodes[elm_no & HAMMER_FSBUF_BLKMASK];
771         bzero(item, sizeof(union hammer_btree_node));
772         return(item);
773 }
774
775 void *
776 hammer_alloc_data(struct hammer_cluster *cluster, int32_t bytes,
777                    int *errorp, struct hammer_buffer **bufferp)
778 {
779         struct hammer_buffer *buffer;
780         hammer_alist_t live;
781         int32_t elm_no;
782         int32_t buf_no;
783         int32_t nblks;
784         void *item;
785
786         /*
787          * Allocate a data element
788          */
789         nblks = (bytes + HAMMER_DATA_BLKMASK) & ~HAMMER_DATA_BLKMASK;
790         live = &cluster->alist_mdata;
791         elm_no = hammer_alist_alloc_fwd(live, nblks, cluster->ondisk->idx_data);
792         if (elm_no == HAMMER_ALIST_BLOCK_NONE)
793                 elm_no = hammer_alist_alloc_fwd(live, 1, 0);
794         if (elm_no == HAMMER_ALIST_BLOCK_NONE) {
795                 alloc_new_buffer(cluster, live,
796                                  HAMMER_FSBUF_DATA, HAMMER_DATA_NODES,
797                                  cluster->ondisk->idx_data, errorp, bufferp);
798                 elm_no = hammer_alist_alloc(live, nblks);
799                 KKASSERT(elm_no != HAMMER_ALIST_BLOCK_NONE);
800         }
801         cluster->ondisk->idx_index = elm_no;
802
803         /*
804          * Load and return the B-Tree element
805          */
806         buf_no = elm_no / HAMMER_FSBUF_MAXBLKS;
807         buffer = *bufferp;
808         if (buffer == NULL || buffer->cluster != cluster ||
809             buffer->buf_no != buf_no) {
810                 if (buffer)
811                         hammer_put_buffer(buffer);
812                 buffer = hammer_get_buffer(cluster, buf_no, 0, errorp);
813                 *bufferp = buffer;
814         }
815         KKASSERT(buffer->ondisk->head.buf_type == HAMMER_FSBUF_BTREE);
816         item = &buffer->ondisk->data.data[elm_no & HAMMER_FSBUF_BLKMASK];
817         bzero(item, nblks * HAMMER_DATA_BLKSIZE);
818         return(item);
819 }
820
821 void *
822 hammer_alloc_record(struct hammer_cluster *cluster,
823                      int *errorp, struct hammer_buffer **bufferp)
824 {
825         struct hammer_buffer *buffer;
826         hammer_alist_t live;
827         int32_t elm_no;
828         int32_t buf_no;
829         void *item;
830
831         /*
832          * Allocate a record element
833          */
834         live = &cluster->alist_record;
835         elm_no = hammer_alist_alloc_rev(live, 1, cluster->ondisk->idx_record);
836         if (elm_no == HAMMER_ALIST_BLOCK_NONE)
837                 elm_no = hammer_alist_alloc_rev(live, 1,HAMMER_ALIST_BLOCK_MAX);
838         if (elm_no == HAMMER_ALIST_BLOCK_NONE) {
839                 alloc_new_buffer(cluster, live,
840                                  HAMMER_FSBUF_RECORDS, HAMMER_RECORD_NODES,
841                                  cluster->ondisk->idx_record, errorp, bufferp);
842                 elm_no = hammer_alist_alloc(live, 1);
843                 KKASSERT(elm_no != HAMMER_ALIST_BLOCK_NONE);
844         }
845         cluster->ondisk->idx_record = elm_no;
846
847         /*
848          * Load and return the B-Tree element
849          */
850         buf_no = elm_no / HAMMER_FSBUF_MAXBLKS;
851         buffer = *bufferp;
852         if (buffer == NULL || buffer->cluster != cluster ||
853             buffer->buf_no != buf_no) {
854                 if (buffer)
855                         hammer_put_buffer(buffer);
856                 buffer = hammer_get_buffer(cluster, buf_no, 0, errorp);
857                 *bufferp = buffer;
858         }
859         KKASSERT(buffer->ondisk->head.buf_type != 0);
860         item = &buffer->ondisk->record.recs[elm_no & HAMMER_FSBUF_BLKMASK];
861         bzero(item, sizeof(union hammer_record_ondisk));
862         return(item);
863 }
864
865 /*
866  * Free HAMMER elements based on either a hammer_buffer and element pointer
867  * or a cluster-relative byte offset.
868  */
869 void
870 hammer_free_btree_ptr(struct hammer_buffer *buffer, hammer_btree_node_t node)
871 {
872         int32_t elm_no;
873         hammer_alist_t live;
874
875         elm_no = node - buffer->ondisk->btree.nodes;
876         KKASSERT(elm_no >= 0 && elm_no < HAMMER_BTREE_NODES);
877         elm_no += buffer->buf_no * HAMMER_FSBUF_MAXBLKS;
878         live = &buffer->cluster->alist_btree;
879         hammer_alist_free(live, elm_no, 1);
880 }
881
882 void
883 hammer_free_data_ptr(struct hammer_buffer *buffer, void *data, int bytes)
884 {
885         int32_t elm_no;
886         int32_t nblks;
887         hammer_alist_t live;
888
889         elm_no = ((char *)data - (char *)buffer->ondisk->data.data) /
890                  HAMMER_DATA_BLKSIZE;
891         KKASSERT(elm_no >= 0 && elm_no < HAMMER_DATA_NODES);
892         elm_no += buffer->buf_no * HAMMER_FSBUF_MAXBLKS;
893         nblks = (bytes + HAMMER_DATA_BLKMASK) & ~HAMMER_DATA_BLKMASK;
894         live = &buffer->cluster->alist_mdata;
895         hammer_alist_free(live, elm_no, nblks);
896 }
897
898 void
899 hammer_free_record_ptr(struct hammer_buffer *buffer,
900                        union hammer_record_ondisk *rec)
901 {
902         int32_t elm_no;
903         hammer_alist_t live;
904
905         elm_no = rec - &buffer->ondisk->record.recs[0];
906         KKASSERT(elm_no >= 0 && elm_no < HAMMER_BTREE_NODES);
907         elm_no += buffer->buf_no * HAMMER_FSBUF_MAXBLKS;
908         live = &buffer->cluster->alist_record;
909         hammer_alist_free(live, elm_no, 1);
910 }
911
912 void
913 hammer_free_btree(struct hammer_cluster *cluster, int32_t bclu_offset)
914 {
915         const int32_t blksize = sizeof(union hammer_btree_node);
916         int32_t fsbuf_offset = bclu_offset & HAMMER_BUFMASK;
917         hammer_alist_t live;
918         int32_t elm_no;
919
920         elm_no = bclu_offset / HAMMER_BUFSIZE * HAMMER_FSBUF_MAXBLKS;
921         fsbuf_offset -= offsetof(union hammer_fsbuf_ondisk, btree.nodes[0]);
922         live = &cluster->alist_btree;
923         KKASSERT(fsbuf_offset >= 0 && fsbuf_offset % blksize == 0);
924         elm_no += fsbuf_offset / blksize;
925         hammer_alist_free(live, elm_no, 1);
926 }
927
928 void
929 hammer_free_data(struct hammer_cluster *cluster, int32_t bclu_offset,
930                  int32_t bytes)
931 {
932         const int32_t blksize = HAMMER_DATA_BLKSIZE;
933         int32_t fsbuf_offset = bclu_offset & HAMMER_BUFMASK;
934         hammer_alist_t live;
935         int32_t elm_no;
936         int32_t nblks;
937
938         elm_no = bclu_offset / HAMMER_BUFSIZE * HAMMER_FSBUF_MAXBLKS;
939         fsbuf_offset -= offsetof(union hammer_fsbuf_ondisk, data.data[0][0]);
940         live = &cluster->alist_mdata;
941         nblks = (bytes + HAMMER_DATA_BLKMASK) & ~HAMMER_DATA_BLKMASK;
942         KKASSERT(fsbuf_offset >= 0 && fsbuf_offset % blksize == 0);
943         elm_no += fsbuf_offset / blksize;
944         hammer_alist_free(live, elm_no, nblks);
945 }
946
947 void
948 hammer_free_record(struct hammer_cluster *cluster, int32_t bclu_offset)
949 {
950         const int32_t blksize = sizeof(union hammer_record_ondisk);
951         int32_t fsbuf_offset = bclu_offset & HAMMER_BUFMASK;
952         hammer_alist_t live;
953         int32_t elm_no;
954
955         elm_no = bclu_offset / HAMMER_BUFSIZE * HAMMER_FSBUF_MAXBLKS;
956         fsbuf_offset -= offsetof(union hammer_fsbuf_ondisk, record.recs[0]);
957         live = &cluster->alist_record;
958         KKASSERT(fsbuf_offset >= 0 && fsbuf_offset % blksize == 0);
959         elm_no += fsbuf_offset / blksize;
960         hammer_alist_free(live, elm_no, 1);
961 }
962
963
964 /*
965  * Allocate a new filesystem buffer and assign it to the specified
966  * filesystem buffer type.  The new buffer will be added to the
967  * type-specific A-list and initialized.
968  */
969 static void
970 alloc_new_buffer(struct hammer_cluster *cluster, hammer_alist_t live,
971                  u_int64_t type, int32_t nelements,
972                  int start, int *errorp, struct hammer_buffer **bufferp)
973 {
974         struct hammer_buffer *buffer;
975         int32_t buf_no;
976
977         start = start / HAMMER_FSBUF_MAXBLKS;   /* convert to buf_no */
978
979         if (type == HAMMER_FSBUF_RECORDS) {
980                 buf_no = hammer_alist_alloc_rev(&cluster->alist_master,
981                                                 1, start);
982                 if (buf_no == HAMMER_ALIST_BLOCK_NONE) {
983                         buf_no = hammer_alist_alloc_rev(&cluster->alist_master,
984                                                 1, HAMMER_ALIST_BLOCK_MAX);
985                 }
986         } else {
987                 buf_no = hammer_alist_alloc_fwd(&cluster->alist_master,
988                                                 1, start);
989                 if (buf_no == HAMMER_ALIST_BLOCK_NONE) {
990                         buf_no = hammer_alist_alloc_fwd(&cluster->alist_master,
991                                                 1, 0);
992                 }
993         }
994         KKASSERT(buf_no != HAMMER_ALIST_BLOCK_NONE); /* XXX */
995
996         /*
997          * The new buffer must be initialized (type != 0) regardless of
998          * whether we already have it cached or not, so don't try to
999          * optimize the cached buffer check.  Just call hammer_get_buffer().
1000          */
1001         buffer = hammer_get_buffer(cluster, buf_no, type, errorp);
1002         if (*bufferp)
1003                 hammer_put_buffer(*bufferp);
1004         *bufferp = buffer;
1005 }
1006
1007 #if 0
1008
1009 /*
1010  * Flush various tracking structures to disk
1011  */
1012
1013 /*
1014  * Flush various tracking structures to disk
1015  */
1016 void
1017 flush_all_volumes(void)
1018 {
1019         struct hammer_volume *vol;
1020
1021         for (vol = VolBase; vol; vol = vol->next)
1022                 flush_volume(vol);
1023 }
1024
1025 void
1026 flush_volume(struct hammer_volume *vol)
1027 {
1028         struct hammer_supercl *supercl;
1029         struct hammer_cluster *cl;
1030
1031         for (supercl = vol->supercl_base; supercl; supercl = supercl->next)
1032                 flush_supercl(supercl);
1033         for (cl = vol->cluster_base; cl; cl = cl->next)
1034                 flush_cluster(cl);
1035         writehammerbuf(vol, vol->ondisk, 0);
1036 }
1037
1038 void
1039 flush_supercl(struct hammer_supercl *supercl)
1040 {
1041         int64_t supercl_offset;
1042
1043         supercl_offset = supercl->scl_offset;
1044         writehammerbuf(supercl->volume, supercl->ondisk, supercl_offset);
1045 }
1046
1047 void
1048 flush_cluster(struct hammer_cluster *cl)
1049 {
1050         struct hammer_buffer *buf;
1051         int64_t cluster_offset;
1052
1053         for (buf = cl->buffer_base; buf; buf = buf->next)
1054                 flush_buffer(buf);
1055         cluster_offset = cl->clu_offset;
1056         writehammerbuf(cl->volume, cl->ondisk, cluster_offset);
1057 }
1058
1059 void
1060 flush_buffer(struct hammer_buffer *buf)
1061 {
1062         int64_t buffer_offset;
1063
1064         buffer_offset = buf->buf_offset + buf->cluster->clu_offset;
1065         writehammerbuf(buf->volume, buf->ondisk, buffer_offset);
1066 }
1067
1068 #endif
1069
1070 /*
1071  * Generic buffer initialization
1072  */
1073 static void
1074 initbuffer(hammer_alist_t live, hammer_fsbuf_head_t head, u_int64_t type)
1075 {
1076         head->buf_type = type;
1077         hammer_alist_init(live);
1078 }
1079
1080 #if 0
1081 /*
1082  * Core I/O operations
1083  */
1084 static void
1085 readhammerbuf(struct hammer_volume *vol, void *data, int64_t offset)
1086 {
1087         ssize_t n;
1088
1089         n = pread(vol->fd, data, HAMMER_BUFSIZE, offset);
1090         if (n != HAMMER_BUFSIZE)
1091                 err(1, "Read volume %d (%s)", vol->vol_no, vol->name);
1092 }
1093
1094 static void
1095 writehammerbuf(struct hammer_volume *vol, const void *data, int64_t offset)
1096 {
1097         ssize_t n;
1098
1099         n = pwrite(vol->fd, data, HAMMER_BUFSIZE, offset);
1100         if (n != HAMMER_BUFSIZE)
1101                 err(1, "Write volume %d (%s)", vol->vol_no, vol->name);
1102 }
1103
1104 #endif
1105
1106 /*
1107  * Calculate the cluster's offset in the volume.  This calculation is
1108  * slightly more complex when using superclusters because superclusters
1109  * are grouped in blocks of 16, followed by 16 x N clusters where N
1110  * is the number of clusters a supercluster can manage.
1111  */
1112 static int64_t
1113 calculate_cluster_offset(struct hammer_volume *volume, int32_t clu_no)
1114 {
1115         int32_t scl_group;
1116         int64_t scl_group_size;
1117         int64_t off;
1118
1119         if (volume->vol_flags & HAMMER_VOLF_USINGSUPERCL) {
1120                 scl_group = clu_no / HAMMER_VOL_SUPERCLUSTER_GROUP /
1121                             HAMMER_SCL_MAXCLUSTERS;
1122                 scl_group_size = 
1123                             ((int64_t)HAMMER_BUFSIZE *
1124                              HAMMER_VOL_SUPERCLUSTER_GROUP) +
1125                             ((int64_t)HAMMER_VOL_SUPERCLUSTER_GROUP *
1126                              volume->vol_clsize * HAMMER_SCL_MAXCLUSTERS);
1127                 scl_group_size += 
1128                             HAMMER_VOL_SUPERCLUSTER_GROUP * HAMMER_BUFSIZE;
1129
1130                 off = volume->cluster_base +
1131                       scl_group * scl_group_size +
1132                       (HAMMER_BUFSIZE * HAMMER_VOL_SUPERCLUSTER_GROUP) +
1133                       ((int64_t)clu_no % ((int64_t)HAMMER_SCL_MAXCLUSTERS *
1134                        HAMMER_VOL_SUPERCLUSTER_GROUP))
1135                       * HAMMER_BUFSIZE;
1136         } else {
1137                 off = volume->cluster_base +
1138                       (int64_t)clu_no * volume->vol_clsize;
1139         }
1140         return(off);
1141 }
1142
1143 /*
1144  * Calculate a super-cluster's offset in the volume.
1145  */
1146 static int64_t
1147 calculate_supercl_offset(struct hammer_volume *volume, int32_t scl_no)
1148 {
1149         int64_t off;
1150         int32_t scl_group;
1151         int64_t scl_group_size;
1152
1153         KKASSERT (volume->vol_flags & HAMMER_VOLF_USINGSUPERCL);
1154         scl_group = scl_no / HAMMER_VOL_SUPERCLUSTER_GROUP;
1155         if (scl_group) {
1156                 scl_group_size = 
1157                             ((int64_t)HAMMER_BUFSIZE *
1158                              HAMMER_VOL_SUPERCLUSTER_GROUP) +
1159                             ((int64_t)HAMMER_VOL_SUPERCLUSTER_GROUP *
1160                              volume->vol_clsize * HAMMER_SCL_MAXCLUSTERS);
1161                 scl_group_size += 
1162                             HAMMER_VOL_SUPERCLUSTER_GROUP * HAMMER_BUFSIZE;
1163                 off = volume->cluster_base + (scl_group * scl_group_size) +
1164                       (scl_no % HAMMER_VOL_SUPERCLUSTER_GROUP) * HAMMER_BUFSIZE;
1165         } else {
1166                 off = volume->cluster_base + (scl_no * HAMMER_BUFSIZE);
1167         }
1168         return(off);
1169 }
1170
1171 /*
1172  * A-LIST SUPPORT
1173  *
1174  * Setup the parameters for the various A-lists we use in hammer.  The
1175  * supercluster A-list must be chained to the cluster A-list and cluster
1176  * slave A-lists are chained to buffer A-lists.
1177  *
1178  * See hammer_init_alist_config() below.
1179  */
1180
1181 /*
1182  * A-LIST - cluster recursion into a filesystem buffer
1183  */
1184 static int
1185 buffer_alist_init(void *info, int32_t blk, int32_t radix)
1186 {
1187         struct hammer_cluster *cluster = info;
1188         struct hammer_buffer *buffer;
1189         int32_t buf_no;
1190         int error = 0;
1191
1192         /*
1193          * Calculate the buffer number, initialize based on the buffer type.
1194          * The buffer has already been allocated so assert that it has been
1195          * initialized.
1196          */
1197         buf_no = blk / HAMMER_FSBUF_MAXBLKS;
1198         buffer = hammer_get_buffer(cluster, buf_no, 0, &error);
1199         if (buffer)
1200                 hammer_put_buffer(buffer);
1201         return (error);
1202 }
1203
1204 static int
1205 buffer_alist_destroy(void *info, int32_t blk, int32_t radix)
1206 {
1207         return (0);
1208 }
1209
1210 static int
1211 buffer_alist_alloc_fwd(void *info, int32_t blk, int32_t radix,
1212                       int32_t count, int32_t atblk, int32_t *fullp)
1213 {
1214         struct hammer_cluster *cluster = info;
1215         struct hammer_buffer *buffer;
1216         int32_t buf_no;
1217         int32_t r;
1218         int error = 0;
1219
1220         buf_no = blk / HAMMER_FSBUF_MAXBLKS;
1221         buffer = hammer_get_buffer(cluster, buf_no, 0, &error);
1222         if (buffer) {
1223                 KKASSERT(buffer->ondisk->head.buf_type != 0);
1224
1225                 r = hammer_alist_alloc_fwd(&buffer->alist, count, atblk - blk);
1226                 if (r != HAMMER_ALIST_BLOCK_NONE)
1227                         r += blk;
1228                 *fullp = hammer_alist_isfull(&buffer->alist);
1229                 hammer_put_buffer(buffer);
1230         } else {
1231                 r = HAMMER_ALIST_BLOCK_NONE;
1232         }
1233         return(r);
1234 }
1235
1236 static int
1237 buffer_alist_alloc_rev(void *info, int32_t blk, int32_t radix,
1238                       int32_t count, int32_t atblk, int32_t *fullp)
1239 {
1240         struct hammer_cluster *cluster = info;
1241         struct hammer_buffer *buffer;
1242         int32_t buf_no;
1243         int32_t r;
1244         int error = 0;
1245
1246         buf_no = blk / HAMMER_FSBUF_MAXBLKS;
1247         buffer = hammer_get_buffer(cluster, buf_no, 0, &error);
1248         if (buffer) {
1249                 KKASSERT(buffer->ondisk->head.buf_type != 0);
1250
1251                 r = hammer_alist_alloc_rev(&buffer->alist, count, atblk - blk);
1252                 if (r != HAMMER_ALIST_BLOCK_NONE)
1253                         r += blk;
1254                 *fullp = hammer_alist_isfull(&buffer->alist);
1255                 hammer_put_buffer(buffer);
1256         } else {
1257                 r = HAMMER_ALIST_BLOCK_NONE;
1258                 *fullp = 0;
1259         }
1260         return(r);
1261 }
1262
1263 static void
1264 buffer_alist_free(void *info, int32_t blk, int32_t radix,
1265                  int32_t base_blk, int32_t count, int32_t *emptyp)
1266 {
1267         struct hammer_cluster *cluster = info;
1268         struct hammer_buffer *buffer;
1269         int32_t buf_no;
1270         int error = 0;
1271
1272         buf_no = blk / HAMMER_FSBUF_MAXBLKS;
1273         buffer = hammer_get_buffer(cluster, buf_no, 0, &error);
1274         if (buffer) {
1275                 KKASSERT(buffer->ondisk->head.buf_type != 0);
1276                 hammer_alist_free(&buffer->alist, base_blk, count);
1277                 *emptyp = hammer_alist_isempty(&buffer->alist);
1278                 hammer_put_buffer(buffer);
1279         } else {
1280                 *emptyp = 0;
1281         }
1282 }
1283
1284 static void
1285 buffer_alist_print(void *info, int32_t blk, int32_t radix, int tab)
1286 {
1287 }
1288
1289 /*
1290  * A-LIST - super-cluster recursion into a cluster and cluster recursion
1291  * into a filesystem buffer.  A-List's are mostly self-contained entities,
1292  * but callbacks must be installed to recurse from one A-List to another.
1293  *
1294  * Implementing these callbacks allows us to operate a multi-layered A-List
1295  * as a single entity.
1296  */
1297 static int
1298 super_alist_init(void *info, int32_t blk, int32_t radix)
1299 {
1300         struct hammer_volume *volume = info;
1301         struct hammer_supercl *supercl;
1302         int32_t scl_no;
1303         int error = 0;
1304
1305         /*
1306          * Calculate the super-cluster number containing the cluster (blk)
1307          * and obtain the super-cluster buffer.
1308          */
1309         scl_no = blk / HAMMER_SCL_MAXCLUSTERS;
1310         supercl = hammer_get_supercl(volume, scl_no, &error, 1);
1311         if (supercl)
1312                 hammer_put_supercl(supercl);
1313         return (error);
1314 }
1315
1316 static int
1317 super_alist_destroy(void *info, int32_t blk, int32_t radix)
1318 {
1319         return(0);
1320 }
1321
1322 static int
1323 super_alist_alloc_fwd(void *info, int32_t blk, int32_t radix,
1324                       int32_t count, int32_t atblk, int32_t *fullp)
1325 {
1326         struct hammer_volume *volume = info;
1327         struct hammer_supercl *supercl;
1328         int32_t scl_no;
1329         int32_t r;
1330         int error = 0;
1331
1332         scl_no = blk / HAMMER_SCL_MAXCLUSTERS;
1333         supercl = hammer_get_supercl(volume, scl_no, &error, 1);
1334         if (supercl) {
1335                 r = hammer_alist_alloc_fwd(&supercl->alist, count, atblk - blk);
1336                 if (r != HAMMER_ALIST_BLOCK_NONE)
1337                         r += blk;
1338                 *fullp = hammer_alist_isfull(&supercl->alist);
1339                 hammer_put_supercl(supercl);
1340         } else {
1341                 r = HAMMER_ALIST_BLOCK_NONE;
1342                 *fullp = 0;
1343         }
1344         return(r);
1345 }
1346
1347 static int
1348 super_alist_alloc_rev(void *info, int32_t blk, int32_t radix,
1349                       int32_t count, int32_t atblk, int32_t *fullp)
1350 {
1351         struct hammer_volume *volume = info;
1352         struct hammer_supercl *supercl;
1353         int32_t scl_no;
1354         int32_t r;
1355         int error = 0;
1356
1357         scl_no = blk / HAMMER_SCL_MAXCLUSTERS;
1358         supercl = hammer_get_supercl(volume, scl_no, &error, 1);
1359         if (supercl) {
1360                 r = hammer_alist_alloc_rev(&supercl->alist, count, atblk - blk);
1361                 if (r != HAMMER_ALIST_BLOCK_NONE)
1362                         r += blk;
1363                 *fullp = hammer_alist_isfull(&supercl->alist);
1364                 hammer_put_supercl(supercl);
1365         } else { 
1366                 r = HAMMER_ALIST_BLOCK_NONE;
1367                 *fullp = 0;
1368         }
1369         return(r);
1370 }
1371
1372 static void
1373 super_alist_free(void *info, int32_t blk, int32_t radix,
1374                  int32_t base_blk, int32_t count, int32_t *emptyp)
1375 {
1376         struct hammer_volume *volume = info;
1377         struct hammer_supercl *supercl;
1378         int32_t scl_no;
1379         int error = 0;
1380
1381         scl_no = blk / HAMMER_SCL_MAXCLUSTERS;
1382         supercl = hammer_get_supercl(volume, scl_no, &error, 1);
1383         if (supercl) {
1384                 hammer_alist_free(&supercl->alist, base_blk, count);
1385                 *emptyp = hammer_alist_isempty(&supercl->alist);
1386                 hammer_put_supercl(supercl);
1387         } else {
1388                 *emptyp = 0;
1389         }
1390 }
1391
1392 static void
1393 super_alist_print(void *info, int32_t blk, int32_t radix, int tab)
1394 {
1395 }
1396
1397 void
1398 hammer_init_alist_config(void)
1399 {
1400         hammer_alist_config_t config;
1401
1402         hammer_alist_template(&Buf_alist_config, HAMMER_FSBUF_MAXBLKS,
1403                               1, HAMMER_FSBUF_METAELMS);
1404         hammer_alist_template(&Vol_normal_alist_config, HAMMER_VOL_MAXCLUSTERS,
1405                               1, HAMMER_VOL_METAELMS_1LYR);
1406         hammer_alist_template(&Vol_super_alist_config,
1407                               HAMMER_VOL_MAXSUPERCLUSTERS,
1408                               HAMMER_SCL_MAXCLUSTERS, HAMMER_VOL_METAELMS_2LYR);
1409         hammer_alist_template(&Supercl_alist_config, HAMMER_VOL_MAXCLUSTERS,
1410                               1, HAMMER_SUPERCL_METAELMS);
1411         hammer_alist_template(&Clu_master_alist_config, HAMMER_CLU_MAXBUFFERS,
1412                               1, HAMMER_CLU_MASTER_METAELMS);
1413         hammer_alist_template(&Clu_slave_alist_config, HAMMER_CLU_MAXBUFFERS,
1414                               HAMMER_FSBUF_MAXBLKS, HAMMER_CLU_SLAVE_METAELMS);
1415
1416         config = &Vol_super_alist_config;
1417         config->bl_radix_init = super_alist_init;
1418         config->bl_radix_destroy = super_alist_destroy;
1419         config->bl_radix_alloc_fwd = super_alist_alloc_fwd;
1420         config->bl_radix_alloc_rev = super_alist_alloc_rev;
1421         config->bl_radix_free = super_alist_free;
1422         config->bl_radix_print = super_alist_print;
1423
1424         config = &Clu_slave_alist_config;
1425         config->bl_radix_init = buffer_alist_init;
1426         config->bl_radix_destroy = buffer_alist_destroy;
1427         config->bl_radix_alloc_fwd = buffer_alist_alloc_fwd;
1428         config->bl_radix_alloc_rev = buffer_alist_alloc_rev;
1429         config->bl_radix_free = buffer_alist_free;
1430         config->bl_radix_print = buffer_alist_print;
1431 }
1432
1433