HAMMER 12/many - add VOPs for symlinks, device, and fifo support.
[dragonfly.git] / sys / vfs / hammer / hammer_inode.c
CommitLineData
427e5fc6
MD
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 *
7a04d74f 34 * $DragonFly: src/sys/vfs/hammer/hammer_inode.c,v 1.13 2007/12/30 00:47:22 dillon Exp $
427e5fc6
MD
35 */
36
37#include "hammer.h"
38#include <sys/buf.h>
39#include <sys/buf2.h>
40
427e5fc6
MD
41int
42hammer_vop_inactive(struct vop_inactive_args *ap)
43{
66325755 44 struct hammer_inode *ip = VTOI(ap->a_vp);
27ea2398 45
c0ade690
MD
46 /*
47 * Degenerate case
48 */
49 if (ip == NULL) {
66325755 50 vrecycle(ap->a_vp);
c0ade690
MD
51 return(0);
52 }
53
54 /*
55 * If the inode no longer has any references we recover its
56 * in-memory resources immediately.
57 */
58 if (ip->ino_rec.ino_nlinks == 0 &&
59 (ip->hmp->mp->mnt_flag & MNT_RDONLY) == 0) {
60 hammer_sync_inode(ip, MNT_NOWAIT, 1);
61 }
427e5fc6
MD
62 return(0);
63}
64
65int
66hammer_vop_reclaim(struct vop_reclaim_args *ap)
67{
427e5fc6
MD
68 struct hammer_inode *ip;
69 struct vnode *vp;
70
71 vp = ap->a_vp;
c0ade690
MD
72
73 /*
74 * Release the vnode association and ask that the inode be flushed.
75 */
a89aec1b
MD
76 if ((ip = vp->v_data) != NULL) {
77 vp->v_data = NULL;
78 ip->vp = NULL;
79 hammer_rel_inode(ip, 1);
80 }
427e5fc6
MD
81 return(0);
82}
83
84/*
66325755
MD
85 * Obtain a vnode for the specified inode number. An exclusively locked
86 * vnode is returned.
427e5fc6
MD
87 */
88int
89hammer_vfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
90{
91 struct hammer_mount *hmp = (void *)mp->mnt_data;
66325755
MD
92 struct hammer_inode *ip;
93 int error;
94
95 /*
96 * Get/allocate the hammer_inode structure. The structure must be
97 * unlocked while we manipulate the related vnode to avoid a
98 * deadlock.
99 */
7f7c1f84 100 ip = hammer_get_inode(hmp, ino, hmp->asof, &error);
66325755
MD
101 if (ip == NULL) {
102 *vpp = NULL;
103 return(error);
104 }
66325755 105 error = hammer_get_vnode(ip, LK_EXCLUSIVE, vpp);
a89aec1b 106 hammer_rel_inode(ip, 0);
66325755
MD
107 return (error);
108}
109
110/*
111 * Return a locked vnode for the specified inode. The inode must be
112 * referenced but NOT LOCKED on entry and will remain referenced on
113 * return.
114 */
115int
116hammer_get_vnode(struct hammer_inode *ip, int lktype, struct vnode **vpp)
117{
118 struct vnode *vp;
119 int error = 0;
120
121 for (;;) {
122 if ((vp = ip->vp) == NULL) {
123 error = getnewvnode(VT_HAMMER, ip->hmp->mp, vpp, 0, 0);
124 if (error)
125 break;
8cd0a023
MD
126 hammer_lock_ex(&ip->lock);
127 if (ip->vp != NULL) {
128 hammer_unlock(&ip->lock);
129 vp->v_type = VBAD;
130 vx_put(vp);
131 continue;
66325755 132 }
8cd0a023
MD
133 hammer_ref(&ip->lock);
134 vp = *vpp;
135 ip->vp = vp;
136 vp->v_type = hammer_get_vnode_type(
137 ip->ino_rec.base.base.obj_type);
7a04d74f
MD
138
139 switch(ip->ino_rec.base.base.obj_type) {
140 case HAMMER_OBJTYPE_CDEV:
141 case HAMMER_OBJTYPE_BDEV:
142 vp->v_ops = &ip->hmp->mp->mnt_vn_spec_ops;
143 addaliasu(vp, ip->ino_data.rmajor,
144 ip->ino_data.rminor);
145 break;
146 case HAMMER_OBJTYPE_FIFO:
147 vp->v_ops = &ip->hmp->mp->mnt_vn_fifo_ops;
148 break;
149 default:
150 break;
151 }
152 if (ip->obj_id == HAMMER_OBJID_ROOT)
153 vp->v_flag |= VROOT;
154
8cd0a023
MD
155 vp->v_data = (void *)ip;
156 /* vnode locked by getnewvnode() */
157 /* make related vnode dirty if inode dirty? */
158 hammer_unlock(&ip->lock);
a89aec1b
MD
159 if (vp->v_type == VREG)
160 vinitvmio(vp, ip->ino_rec.ino_size);
8cd0a023
MD
161 break;
162 }
163
164 /*
165 * loop if the vget fails (aka races), or if the vp
166 * no longer matches ip->vp.
167 */
168 if (vget(vp, LK_EXCLUSIVE) == 0) {
169 if (vp == ip->vp)
170 break;
171 vput(vp);
66325755
MD
172 }
173 }
a89aec1b 174 *vpp = vp;
66325755
MD
175 return(error);
176}
177
178/*
8cd0a023
MD
179 * Acquire a HAMMER inode. The returned inode is not locked. These functions
180 * do not attach or detach the related vnode (use hammer_get_vnode() for
181 * that).
66325755
MD
182 */
183struct hammer_inode *
7f7c1f84
MD
184hammer_get_inode(struct hammer_mount *hmp, u_int64_t obj_id, hammer_tid_t asof,
185 int *errorp)
66325755 186{
427e5fc6 187 struct hammer_inode_info iinfo;
8cd0a023 188 struct hammer_cursor cursor;
427e5fc6 189 struct hammer_inode *ip;
427e5fc6
MD
190
191 /*
192 * Determine if we already have an inode cached. If we do then
193 * we are golden.
194 */
66325755 195 iinfo.obj_id = obj_id;
7f7c1f84 196 iinfo.obj_asof = asof;
427e5fc6
MD
197loop:
198 ip = hammer_ino_rb_tree_RB_LOOKUP_INFO(&hmp->rb_inos_root, &iinfo);
199 if (ip) {
8cd0a023 200 hammer_ref(&ip->lock);
66325755
MD
201 *errorp = 0;
202 return(ip);
427e5fc6
MD
203 }
204
427e5fc6 205 ip = kmalloc(sizeof(*ip), M_HAMMER, M_WAITOK|M_ZERO);
66325755 206 ip->obj_id = obj_id;
27ea2398 207 ip->obj_asof = iinfo.obj_asof;
66325755 208 ip->hmp = hmp;
8cd0a023 209 RB_INIT(&ip->rec_tree);
427e5fc6
MD
210
211 /*
8cd0a023 212 * Locate the on-disk inode.
427e5fc6
MD
213 * If we do not have an inode cached search the HAMMER on-disk B-Tree
214 * for it.
215 */
427e5fc6 216
8cd0a023
MD
217 hammer_init_cursor_hmp(&cursor, hmp);
218 cursor.key_beg.obj_id = ip->obj_id;
219 cursor.key_beg.key = 0;
220 cursor.key_beg.create_tid = iinfo.obj_asof;
221 cursor.key_beg.delete_tid = 0;
222 cursor.key_beg.rec_type = HAMMER_RECTYPE_INODE;
223 cursor.key_beg.obj_type = 0;
a89aec1b 224 cursor.flags = HAMMER_CURSOR_GET_RECORD | HAMMER_CURSOR_GET_DATA;
8cd0a023
MD
225
226 *errorp = hammer_btree_lookup(&cursor);
427e5fc6
MD
227
228 /*
229 * On success the B-Tree lookup will hold the appropriate
230 * buffer cache buffers and provide a pointer to the requested
231 * information. Copy the information to the in-memory inode.
232 */
66325755 233 if (*errorp == 0) {
8cd0a023
MD
234 ip->ino_rec = cursor.record->inode;
235 ip->ino_data = cursor.data->inode;
427e5fc6 236 }
8cd0a023
MD
237 hammer_cache_node(cursor.node, &ip->cache);
238 hammer_done_cursor(&cursor);
427e5fc6
MD
239
240 /*
241 * On success load the inode's record and data and insert the
242 * inode into the B-Tree. It is possible to race another lookup
243 * insertion of the same inode so deal with that condition too.
244 */
66325755 245 if (*errorp == 0) {
8cd0a023 246 hammer_ref(&ip->lock);
427e5fc6 247 if (RB_INSERT(hammer_ino_rb_tree, &hmp->rb_inos_root, ip)) {
8cd0a023
MD
248 hammer_uncache_node(&ip->cache);
249 hammer_unref(&ip->lock);
427e5fc6
MD
250 kfree(ip, M_HAMMER);
251 goto loop;
252 }
c0ade690 253 ip->flags |= HAMMER_INODE_ONDISK;
427e5fc6 254 } else {
66325755
MD
255 kfree(ip, M_HAMMER);
256 ip = NULL;
427e5fc6 257 }
66325755
MD
258 return (ip);
259}
260
8cd0a023
MD
261/*
262 * Create a new filesystem object, returning the inode in *ipp. The
263 * returned inode will be referenced but not locked.
264 *
265 * The inode is created in-memory and will be delay-synchronized to the
266 * disk.
267 */
268int
a89aec1b
MD
269hammer_create_inode(hammer_transaction_t trans, struct vattr *vap,
270 struct ucred *cred, hammer_inode_t dip,
8cd0a023 271 struct hammer_inode **ipp)
66325755 272{
a89aec1b
MD
273 hammer_mount_t hmp;
274 hammer_inode_t ip;
6b4f890b 275 uid_t xuid;
66325755 276
8cd0a023
MD
277 hmp = trans->hmp;
278 ip = kmalloc(sizeof(*ip), M_HAMMER, M_WAITOK|M_ZERO);
a89aec1b 279 ip->obj_id = hammer_alloc_tid(trans);
8cd0a023 280 KKASSERT(ip->obj_id != 0);
7f7c1f84 281 ip->obj_asof = hmp->asof;
8cd0a023
MD
282 ip->hmp = hmp;
283 ip->flags = HAMMER_INODE_DDIRTY | HAMMER_INODE_RDIRTY |
284 HAMMER_INODE_ITIMES;
285 ip->last_tid = trans->tid;
286
287 RB_INIT(&ip->rec_tree);
288
289 ip->ino_rec.ino_atime = trans->tid;
290 ip->ino_rec.ino_mtime = trans->tid;
291 ip->ino_rec.ino_size = 0;
292 ip->ino_rec.ino_nlinks = 0;
293 /* XXX */
a89aec1b 294 ip->ino_rec.base.rec_id = hammer_alloc_recid(trans);
8cd0a023
MD
295 KKASSERT(ip->ino_rec.base.rec_id != 0);
296 ip->ino_rec.base.base.obj_id = ip->obj_id;
297 ip->ino_rec.base.base.key = 0;
298 ip->ino_rec.base.base.create_tid = trans->tid;
299 ip->ino_rec.base.base.delete_tid = 0;
300 ip->ino_rec.base.base.rec_type = HAMMER_RECTYPE_INODE;
301 ip->ino_rec.base.base.obj_type = hammer_get_obj_type(vap->va_type);
302
303 ip->ino_data.version = HAMMER_INODE_DATA_VERSION;
304 ip->ino_data.mode = vap->va_mode;
305 ip->ino_data.ctime = trans->tid;
306 ip->ino_data.parent_obj_id = (dip) ? dip->ino_rec.base.base.obj_id : 0;
6b4f890b 307
7a04d74f
MD
308 switch(ip->ino_rec.base.base.obj_type) {
309 case HAMMER_OBJTYPE_CDEV:
310 case HAMMER_OBJTYPE_BDEV:
311 ip->ino_data.rmajor = vap->va_rmajor;
312 ip->ino_data.rminor = vap->va_rminor;
313 break;
314 default:
315 break;
316 }
317
6b4f890b
MD
318 /*
319 * Calculate default uid/gid and overwrite with information from
320 * the vap.
321 */
322 xuid = hammer_to_unix_xid(&dip->ino_data.uid);
323 ip->ino_data.gid = dip->ino_data.gid;
324 xuid = vop_helper_create_uid(hmp->mp, dip->ino_data.mode, xuid, cred,
325 &vap->va_mode);
326 ip->ino_data.mode = vap->va_mode;
327
8cd0a023
MD
328 if (vap->va_vaflags & VA_UID_UUID_VALID)
329 ip->ino_data.uid = vap->va_uid_uuid;
6b4f890b
MD
330 else if (vap->va_uid != (uid_t)VNOVAL)
331 hammer_guid_to_uuid(&ip->ino_data.uid, xuid);
8cd0a023
MD
332 if (vap->va_vaflags & VA_GID_UUID_VALID)
333 ip->ino_data.gid = vap->va_gid_uuid;
6b4f890b 334 else if (vap->va_gid != (gid_t)VNOVAL)
8cd0a023
MD
335 hammer_guid_to_uuid(&ip->ino_data.gid, vap->va_gid);
336
337 hammer_ref(&ip->lock);
338 if (RB_INSERT(hammer_ino_rb_tree, &hmp->rb_inos_root, ip)) {
339 hammer_unref(&ip->lock);
a89aec1b 340 panic("hammer_create_inode: duplicate obj_id %llx", ip->obj_id);
8cd0a023
MD
341 }
342 *ipp = ip;
343 return(0);
66325755
MD
344}
345
c0ade690 346int
7f7c1f84 347hammer_update_inode(hammer_inode_t ip)
c0ade690
MD
348{
349 struct hammer_cursor cursor;
d26d0ae9 350 struct hammer_cursor *spike = NULL;
c0ade690
MD
351 hammer_record_t record;
352 int error;
353
354 /*
76376933 355 * Locate the record on-disk and mark it as deleted. Both the B-Tree
195c19a1
MD
356 * node and the record must be marked deleted. The record may or
357 * may not be physically deleted, depending on the retention policy.
76376933 358 *
195c19a1
MD
359 * If the inode has already been deleted on-disk we have nothing
360 * to do.
c0ade690
MD
361 *
362 * XXX Update the inode record and data in-place if the retention
363 * policy allows it.
364 */
d26d0ae9 365retry:
c0ade690
MD
366 error = 0;
367
76376933
MD
368 if ((ip->flags & (HAMMER_INODE_ONDISK|HAMMER_INODE_DELONDISK)) ==
369 HAMMER_INODE_ONDISK) {
c0ade690
MD
370 hammer_init_cursor_ip(&cursor, ip);
371 cursor.key_beg.obj_id = ip->obj_id;
372 cursor.key_beg.key = 0;
373 cursor.key_beg.create_tid = ip->obj_asof;
374 cursor.key_beg.delete_tid = 0;
375 cursor.key_beg.rec_type = HAMMER_RECTYPE_INODE;
376 cursor.key_beg.obj_type = 0;
377 cursor.flags = HAMMER_CURSOR_GET_RECORD;
378
379 error = hammer_btree_lookup(&cursor);
380
381 if (error == 0) {
195c19a1
MD
382 error = hammer_ip_delete_record(&cursor, ip->last_tid);
383 if (error == 0)
384 ip->flags |= HAMMER_INODE_DELONDISK;
c0ade690
MD
385 }
386 hammer_cache_node(cursor.node, &ip->cache);
387 hammer_done_cursor(&cursor);
388 }
389
390 /*
391 * Write out a new record if the in-memory inode is not marked
fbc6e32a
MD
392 * as having been deleted. Update our inode statistics if this
393 * is the first application of the inode on-disk.
76376933
MD
394 *
395 * If the inode has been deleted permanently, HAMMER_INODE_DELONDISK
396 * will remain set and prevent further updates.
c0ade690
MD
397 */
398 if (error == 0 && (ip->flags & HAMMER_INODE_DELETED) == 0) {
7f7c1f84 399 record = hammer_alloc_mem_record(ip);
c0ade690 400 record->rec.inode = ip->ino_rec;
7f7c1f84 401 record->rec.inode.base.base.create_tid = ip->last_tid;
c0ade690
MD
402 record->rec.inode.base.data_len = sizeof(ip->ino_data);
403 record->data = (void *)&ip->ino_data;
d26d0ae9
MD
404 error = hammer_ip_sync_record(record, &spike);
405 hammer_drop_mem_record(record, 1);
406 if (error == ENOSPC) {
407 error = hammer_spike(&spike);
408 if (error == 0)
409 goto retry;
410 }
411 KKASSERT(spike == NULL);
412 if (error == 0) {
413 ip->flags &= ~(HAMMER_INODE_RDIRTY|HAMMER_INODE_DDIRTY|
414 HAMMER_INODE_DELONDISK);
415 if ((ip->flags & HAMMER_INODE_ONDISK) == 0) {
416 ++ip->hmp->rootvol->ondisk->vol0_stat_inodes;
417 hammer_modify_volume(ip->hmp->rootvol);
418 ip->flags |= HAMMER_INODE_ONDISK;
419 }
fbc6e32a 420 }
c0ade690 421 }
d26d0ae9
MD
422 if (error == 0)
423 ip->flags &= ~HAMMER_INODE_TID;
c0ade690
MD
424 return(error);
425}
426
a89aec1b 427/*
6b4f890b 428 * Release a reference on an inode and unload it if told to flush.
a89aec1b 429 */
66325755 430void
a89aec1b 431hammer_rel_inode(struct hammer_inode *ip, int flush)
66325755
MD
432{
433 hammer_unref(&ip->lock);
6b4f890b 434 if (flush || ip->ino_rec.ino_nlinks == 0)
a89aec1b
MD
435 ip->flags |= HAMMER_INODE_FLUSH;
436 if (ip->lock.refs == 0 && (ip->flags & HAMMER_INODE_FLUSH))
437 hammer_unload_inode(ip, NULL);
427e5fc6
MD
438}
439
27ea2398 440/*
8cd0a023
MD
441 * Unload and destroy the specified inode.
442 *
27ea2398
MD
443 * (called via RB_SCAN)
444 */
445int
66325755 446hammer_unload_inode(struct hammer_inode *ip, void *data __unused)
27ea2398 447{
c0ade690
MD
448 int error;
449
a89aec1b
MD
450 KASSERT(ip->lock.refs == 0,
451 ("hammer_unload_inode: %d refs\n", ip->lock.refs));
8cd0a023
MD
452 KKASSERT(ip->vp == NULL);
453 hammer_ref(&ip->lock);
6b4f890b 454
c0ade690
MD
455 error = hammer_sync_inode(ip, MNT_WAIT, 1);
456 if (error)
457 kprintf("hammer_sync_inode failed error %d\n", error);
6b4f890b 458
66325755 459 RB_REMOVE(hammer_ino_rb_tree, &ip->hmp->rb_inos_root, ip);
8cd0a023
MD
460
461 hammer_uncache_node(&ip->cache);
27ea2398
MD
462 kfree(ip, M_HAMMER);
463 return(0);
464}
465
427e5fc6 466/*
66325755
MD
467 * A transaction has modified an inode, requiring a new record and possibly
468 * also data to be written out.
7f7c1f84
MD
469 *
470 * last_tid is the TID to use for the disk sync.
427e5fc6 471 */
66325755
MD
472void
473hammer_modify_inode(struct hammer_transaction *trans,
474 struct hammer_inode *ip, int flags)
427e5fc6 475{
7f7c1f84 476 if ((flags & HAMMER_INODE_TID) && (ip->flags & HAMMER_INODE_TID) == 0) {
c0ade690 477 ip->last_tid = trans->tid;
7f7c1f84
MD
478 }
479 ip->flags |= flags;
c0ade690
MD
480}
481
482/*
483 * Sync any dirty buffers and records associated with an inode. The
484 * inode's last_tid field is used as the transaction id for the sync,
485 * overriding any intermediate TIDs that were used for records. Note
486 * that the dirty buffer cache buffers do not have any knowledge of
487 * the transaction id they were modified under.
d26d0ae9
MD
488 *
489 * If we can't sync due to a cluster becoming full the spike structure
490 * will be filled in and ENOSPC returned. We must return -ENOSPC to
491 * terminate the RB_SCAN.
c0ade690
MD
492 */
493static int
d26d0ae9 494hammer_sync_inode_callback(hammer_record_t rec, void *data)
c0ade690 495{
d26d0ae9 496 struct hammer_cursor **spike = data;
c0ade690
MD
497 int error;
498
d26d0ae9
MD
499 hammer_ref(&rec->lock);
500 hammer_lock_ex(&rec->lock);
c0ade690 501 if ((rec->flags & HAMMER_RECF_DELETED) == 0)
d26d0ae9
MD
502 error = hammer_ip_sync_record(rec, spike);
503 else
504 error = 0;
505
506 if (error == ENOSPC) {
507 hammer_drop_mem_record(rec, 0);
508 return(-error);
509 }
c0ade690
MD
510
511 if (error) {
512 kprintf("hammer_sync_inode_callback: sync failed rec %p\n",
513 rec);
d26d0ae9
MD
514 hammer_drop_mem_record(rec, 0);
515 return(-error);
c0ade690 516 }
d26d0ae9 517 hammer_drop_mem_record(rec, 1); /* ref & lock eaten by call */
c0ade690
MD
518 return(0);
519}
520
521/*
522 * XXX error handling
523 */
524int
525hammer_sync_inode(hammer_inode_t ip, int waitfor, int handle_delete)
526{
527 struct hammer_transaction trans;
d26d0ae9 528 struct hammer_cursor *spike = NULL;
c0ade690 529 int error;
c0ade690
MD
530
531 hammer_lock_ex(&ip->lock);
532 hammer_start_transaction(&trans, ip->hmp);
533
534 /*
535 * If the inode has been deleted (nlinks == 0), and the OS no longer
536 * has any references to it (handle_delete != 0), clean up in-memory
537 * data.
538 *
539 * NOTE: We do not set the RDIRTY flag when updating the delete_tid,
540 * setting HAMMER_INODE_DELETED takes care of it.
7f7c1f84
MD
541 *
542 * NOTE: Because we may sync records within this new transaction,
543 * force the inode update later on to use our transaction id or
544 * the delete_tid of the inode may be less then the create_tid of
545 * the inode update. XXX shouldn't happen but don't take the chance.
d26d0ae9
MD
546 *
547 * NOTE: The call to hammer_ip_delete_range() cannot return ENOSPC
548 * so we can pass a NULL spike structure, because no partial data
549 * deletion can occur (yet).
c0ade690
MD
550 */
551 if (ip->ino_rec.ino_nlinks == 0 && handle_delete) {
552 if (ip->vp)
553 vtruncbuf(ip->vp, 0, HAMMER_BUFSIZE);
7a04d74f
MD
554 error = hammer_ip_delete_range_all(&trans, ip);
555 kprintf("delete_range_all error %d\n", error);
c0ade690 556 KKASSERT(RB_EMPTY(&ip->rec_tree));
7f7c1f84 557 ip->flags &= ~HAMMER_INODE_TID;
c0ade690
MD
558 ip->ino_rec.base.base.delete_tid = trans.tid;
559 hammer_modify_inode(&trans, ip,
560 HAMMER_INODE_DELETED | HAMMER_INODE_TID);
fbc6e32a
MD
561 --ip->hmp->rootvol->ondisk->vol0_stat_inodes;
562 hammer_modify_volume(ip->hmp->rootvol);
c0ade690
MD
563 }
564
565 /*
566 * Sync the buffer cache
567 */
568 if (ip->vp != NULL)
569 error = vfsync(ip->vp, waitfor, 1, NULL, NULL);
570 else
571 error = 0;
572
573 /*
574 * Now sync related records
575 */
d26d0ae9
MD
576 for (;;) {
577 error = RB_SCAN(hammer_rec_rb_tree, &ip->rec_tree, NULL,
578 hammer_sync_inode_callback, &spike);
579 KKASSERT(error <= 0);
580 if (error < 0)
581 error = -error;
582 if (error == ENOSPC) {
583 error = hammer_spike(&spike);
584 if (error == 0)
585 continue;
586 }
587 break;
c0ade690
MD
588 }
589
590 /*
591 * Now update the inode's on-disk inode-data and/or on-disk record.
592 */
593 switch(ip->flags & (HAMMER_INODE_DELETED|HAMMER_INODE_ONDISK)) {
594 case HAMMER_INODE_DELETED|HAMMER_INODE_ONDISK:
595 /*
596 * If deleted and on-disk, don't set any additional flags.
597 * the delete flag takes care of things.
598 */
599 break;
600 case HAMMER_INODE_DELETED:
601 /*
602 * Take care of the case where a deleted inode was never
603 * flushed to the disk in the first place.
604 */
605 ip->flags &= ~(HAMMER_INODE_RDIRTY|HAMMER_INODE_DDIRTY);
d26d0ae9
MD
606 while (RB_ROOT(&ip->rec_tree)) {
607 hammer_record_t rec = RB_ROOT(&ip->rec_tree);
608 hammer_ref(&rec->lock);
609 hammer_lock_ex(&rec->lock);
610 hammer_drop_mem_record(rec, 1);
611 }
c0ade690
MD
612 break;
613 case HAMMER_INODE_ONDISK:
614 /*
615 * If already on-disk, do not set any additional flags.
616 */
617 break;
618 default:
619 /*
620 * If not on-disk and not deleted, set both dirty flags
621 * to force an initial record to be written.
622 */
623 ip->flags |= HAMMER_INODE_RDIRTY | HAMMER_INODE_DDIRTY;
624 break;
625 }
626
627 /*
628 * If RDIRTY or DDIRTY is set, write out a new record. If the
629 * inode is already on-disk, the old record is marked as deleted.
630 */
631 if (ip->flags & (HAMMER_INODE_RDIRTY | HAMMER_INODE_DDIRTY |
632 HAMMER_INODE_DELETED)) {
7f7c1f84 633 error = hammer_update_inode(ip);
c0ade690
MD
634 }
635 hammer_commit_transaction(&trans);
636 hammer_unlock(&ip->lock);
637 return(error);
8cd0a023
MD
638}
639
427e5fc6
MD
640/*
641 * Access the filesystem buffer containing the cluster-relative byte
642 * offset, validate the buffer type, load *bufferp and return a
8cd0a023
MD
643 * pointer to the requested data. The buffer is reference and locked on
644 * return.
427e5fc6
MD
645 *
646 * If buf_type is 0 the buffer is assumed to be a pure-data buffer and
647 * no type or crc check is performed.
648 *
8cd0a023
MD
649 * If *bufferp is not NULL on entry it is assumed to contain a locked
650 * and referenced buffer which will then be replaced.
651 *
652 * If the caller is holding another unrelated buffer locked it must be
653 * passed in reorderbuf so we can properly order buffer locks.
654 *
427e5fc6
MD
655 * XXX add a flag for the buffer type and check the CRC here XXX
656 */
657void *
8cd0a023
MD
658hammer_bread(hammer_cluster_t cluster, int32_t cloff,
659 u_int64_t buf_type, int *errorp,
660 struct hammer_buffer **bufferp)
427e5fc6 661{
8cd0a023 662 hammer_buffer_t buffer;
427e5fc6
MD
663 int32_t buf_no;
664 int32_t buf_off;
665
666 /*
667 * Load the correct filesystem buffer, replacing *bufferp.
668 */
669 buf_no = cloff / HAMMER_BUFSIZE;
670 buffer = *bufferp;
671 if (buffer == NULL || buffer->cluster != cluster ||
672 buffer->buf_no != buf_no) {
8cd0a023 673 if (buffer) {
c0ade690 674 /*hammer_unlock(&buffer->io.lock);*/
8cd0a023
MD
675 hammer_rel_buffer(buffer, 0);
676 }
427e5fc6
MD
677 buffer = hammer_get_buffer(cluster, buf_no, 0, errorp);
678 *bufferp = buffer;
679 if (buffer == NULL)
680 return(NULL);
c0ade690 681 /*hammer_lock_ex(&buffer->io.lock);*/
427e5fc6
MD
682 }
683
684 /*
8cd0a023 685 * Validate the buffer type
427e5fc6
MD
686 */
687 buf_off = cloff & HAMMER_BUFMASK;
688 if (buf_type) {
689 if (buf_type != buffer->ondisk->head.buf_type) {
27ea2398
MD
690 kprintf("BUFFER HEAD TYPE MISMATCH %llx %llx\n",
691 buf_type, buffer->ondisk->head.buf_type);
427e5fc6
MD
692 *errorp = EIO;
693 return(NULL);
694 }
695 if (buf_off < sizeof(buffer->ondisk->head)) {
27ea2398 696 kprintf("BUFFER OFFSET TOO LOW %d\n", buf_off);
427e5fc6
MD
697 *errorp = EIO;
698 return(NULL);
699 }
427e5fc6
MD
700 }
701
702 /*
703 * Return a pointer to the buffer data.
704 */
705 *errorp = 0;
706 return((char *)buffer->ondisk + buf_off);
707}
708