- Turn on IFM_HDX
[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 *
47197d71 34 * $DragonFly: src/sys/vfs/hammer/hammer_inode.c,v 1.29 2008/02/08 08:30:59 dillon Exp $
427e5fc6
MD
35 */
36
37#include "hammer.h"
38#include <sys/buf.h>
39#include <sys/buf2.h>
40
d113fda1
MD
41/*
42 * The kernel is not actively referencing this vnode but is still holding
43 * it cached.
44 */
427e5fc6
MD
45int
46hammer_vop_inactive(struct vop_inactive_args *ap)
47{
66325755 48 struct hammer_inode *ip = VTOI(ap->a_vp);
27ea2398 49
c0ade690
MD
50 /*
51 * Degenerate case
52 */
53 if (ip == NULL) {
66325755 54 vrecycle(ap->a_vp);
c0ade690
MD
55 return(0);
56 }
57
58 /*
59 * If the inode no longer has any references we recover its
60 * in-memory resources immediately.
61 */
d113fda1
MD
62 if (ip->ino_rec.ino_nlinks == 0)
63 vrecycle(ap->a_vp);
427e5fc6
MD
64 return(0);
65}
66
d113fda1
MD
67/*
68 * Release the vnode association. This is typically (but not always)
69 * the last reference on the inode and will flush the inode to the
70 * buffer cache.
71 *
72 * XXX Currently our sync code only runs through inodes with vnode
73 * associations, so we depend on hammer_rel_inode() to sync any inode
74 * record data to the block device prior to losing the association.
75 * Otherwise transactions that the user expected to be distinct by
76 * doing a manual sync may be merged.
77 */
427e5fc6
MD
78int
79hammer_vop_reclaim(struct vop_reclaim_args *ap)
80{
427e5fc6
MD
81 struct hammer_inode *ip;
82 struct vnode *vp;
83
84 vp = ap->a_vp;
c0ade690 85
a89aec1b
MD
86 if ((ip = vp->v_data) != NULL) {
87 vp->v_data = NULL;
88 ip->vp = NULL;
d113fda1 89 hammer_rel_inode(ip, 0);
a89aec1b 90 }
427e5fc6
MD
91 return(0);
92}
93
66325755
MD
94/*
95 * Return a locked vnode for the specified inode. The inode must be
96 * referenced but NOT LOCKED on entry and will remain referenced on
97 * return.
98 */
99int
100hammer_get_vnode(struct hammer_inode *ip, int lktype, struct vnode **vpp)
101{
102 struct vnode *vp;
103 int error = 0;
104
105 for (;;) {
106 if ((vp = ip->vp) == NULL) {
107 error = getnewvnode(VT_HAMMER, ip->hmp->mp, vpp, 0, 0);
108 if (error)
109 break;
8cd0a023
MD
110 hammer_lock_ex(&ip->lock);
111 if (ip->vp != NULL) {
112 hammer_unlock(&ip->lock);
113 vp->v_type = VBAD;
114 vx_put(vp);
115 continue;
66325755 116 }
8cd0a023
MD
117 hammer_ref(&ip->lock);
118 vp = *vpp;
119 ip->vp = vp;
120 vp->v_type = hammer_get_vnode_type(
121 ip->ino_rec.base.base.obj_type);
7a04d74f
MD
122
123 switch(ip->ino_rec.base.base.obj_type) {
124 case HAMMER_OBJTYPE_CDEV:
125 case HAMMER_OBJTYPE_BDEV:
126 vp->v_ops = &ip->hmp->mp->mnt_vn_spec_ops;
127 addaliasu(vp, ip->ino_data.rmajor,
128 ip->ino_data.rminor);
129 break;
130 case HAMMER_OBJTYPE_FIFO:
131 vp->v_ops = &ip->hmp->mp->mnt_vn_fifo_ops;
132 break;
133 default:
134 break;
135 }
42c7d26b
MD
136
137 /*
138 * Only mark as the root vnode if the ip is not
139 * historical, otherwise the VFS cache will get
140 * confused. The other half of the special handling
141 * is in hammer_vop_nlookupdotdot().
142 */
143 if (ip->obj_id == HAMMER_OBJID_ROOT &&
144 ip->obj_asof == ip->hmp->asof) {
7a04d74f 145 vp->v_flag |= VROOT;
42c7d26b 146 }
7a04d74f 147
8cd0a023
MD
148 vp->v_data = (void *)ip;
149 /* vnode locked by getnewvnode() */
150 /* make related vnode dirty if inode dirty? */
151 hammer_unlock(&ip->lock);
a89aec1b
MD
152 if (vp->v_type == VREG)
153 vinitvmio(vp, ip->ino_rec.ino_size);
8cd0a023
MD
154 break;
155 }
156
157 /*
158 * loop if the vget fails (aka races), or if the vp
159 * no longer matches ip->vp.
160 */
161 if (vget(vp, LK_EXCLUSIVE) == 0) {
162 if (vp == ip->vp)
163 break;
164 vput(vp);
66325755
MD
165 }
166 }
a89aec1b 167 *vpp = vp;
66325755
MD
168 return(error);
169}
170
171/*
8cd0a023
MD
172 * Acquire a HAMMER inode. The returned inode is not locked. These functions
173 * do not attach or detach the related vnode (use hammer_get_vnode() for
174 * that).
d113fda1
MD
175 *
176 * The flags argument is only applied for newly created inodes, and only
177 * certain flags are inherited.
66325755
MD
178 */
179struct hammer_inode *
61aeeb33
MD
180hammer_get_inode(struct hammer_mount *hmp, struct hammer_node **cache,
181 u_int64_t obj_id, hammer_tid_t asof, int flags, int *errorp)
66325755 182{
427e5fc6 183 struct hammer_inode_info iinfo;
8cd0a023 184 struct hammer_cursor cursor;
427e5fc6 185 struct hammer_inode *ip;
427e5fc6
MD
186
187 /*
188 * Determine if we already have an inode cached. If we do then
189 * we are golden.
190 */
66325755 191 iinfo.obj_id = obj_id;
7f7c1f84 192 iinfo.obj_asof = asof;
427e5fc6
MD
193loop:
194 ip = hammer_ino_rb_tree_RB_LOOKUP_INFO(&hmp->rb_inos_root, &iinfo);
195 if (ip) {
8cd0a023 196 hammer_ref(&ip->lock);
66325755
MD
197 *errorp = 0;
198 return(ip);
427e5fc6
MD
199 }
200
427e5fc6 201 ip = kmalloc(sizeof(*ip), M_HAMMER, M_WAITOK|M_ZERO);
b3deaf57 202 ++hammer_count_inodes;
66325755 203 ip->obj_id = obj_id;
27ea2398 204 ip->obj_asof = iinfo.obj_asof;
66325755 205 ip->hmp = hmp;
d113fda1
MD
206 ip->flags = flags & HAMMER_INODE_RO;
207 if (hmp->ronly)
208 ip->flags |= HAMMER_INODE_RO;
8cd0a023 209 RB_INIT(&ip->rec_tree);
427e5fc6
MD
210
211 /*
8cd0a023 212 * Locate the on-disk inode.
427e5fc6 213 */
6a37e7e4 214retry:
61aeeb33 215 hammer_init_cursor_hmp(&cursor, cache, hmp);
8cd0a023
MD
216 cursor.key_beg.obj_id = ip->obj_id;
217 cursor.key_beg.key = 0;
d5530d22 218 cursor.key_beg.create_tid = 0;
8cd0a023
MD
219 cursor.key_beg.delete_tid = 0;
220 cursor.key_beg.rec_type = HAMMER_RECTYPE_INODE;
221 cursor.key_beg.obj_type = 0;
d5530d22
MD
222 cursor.asof = iinfo.obj_asof;
223 cursor.flags = HAMMER_CURSOR_GET_RECORD | HAMMER_CURSOR_GET_DATA |
224 HAMMER_CURSOR_ASOF;
8cd0a023
MD
225
226 *errorp = hammer_btree_lookup(&cursor);
6a37e7e4
MD
227 if (*errorp == EDEADLK) {
228 hammer_done_cursor(&cursor);
229 goto retry;
230 }
427e5fc6
MD
231
232 /*
233 * On success the B-Tree lookup will hold the appropriate
234 * buffer cache buffers and provide a pointer to the requested
d113fda1
MD
235 * information. Copy the information to the in-memory inode
236 * and cache the B-Tree node to improve future operations.
427e5fc6 237 */
66325755 238 if (*errorp == 0) {
8cd0a023 239 ip->ino_rec = cursor.record->inode;
47197d71 240 ip->ino_data = cursor.data1->inode;
61aeeb33
MD
241 hammer_cache_node(cursor.node, &ip->cache[0]);
242 if (cache)
243 hammer_cache_node(cursor.node, cache);
427e5fc6 244 }
427e5fc6
MD
245
246 /*
247 * On success load the inode's record and data and insert the
248 * inode into the B-Tree. It is possible to race another lookup
249 * insertion of the same inode so deal with that condition too.
b3deaf57
MD
250 *
251 * The cursor's locked node interlocks against others creating and
252 * destroying ip while we were blocked.
427e5fc6 253 */
66325755 254 if (*errorp == 0) {
8cd0a023 255 hammer_ref(&ip->lock);
427e5fc6 256 if (RB_INSERT(hammer_ino_rb_tree, &hmp->rb_inos_root, ip)) {
61aeeb33
MD
257 hammer_uncache_node(&ip->cache[0]);
258 hammer_uncache_node(&ip->cache[1]);
8cd0a023 259 hammer_unref(&ip->lock);
b3deaf57 260 --hammer_count_inodes;
427e5fc6 261 kfree(ip, M_HAMMER);
b3deaf57 262 hammer_done_cursor(&cursor);
427e5fc6
MD
263 goto loop;
264 }
c0ade690 265 ip->flags |= HAMMER_INODE_ONDISK;
427e5fc6 266 } else {
b3deaf57 267 --hammer_count_inodes;
66325755
MD
268 kfree(ip, M_HAMMER);
269 ip = NULL;
427e5fc6 270 }
b3deaf57 271 hammer_done_cursor(&cursor);
66325755
MD
272 return (ip);
273}
274
8cd0a023
MD
275/*
276 * Create a new filesystem object, returning the inode in *ipp. The
277 * returned inode will be referenced but not locked.
278 *
279 * The inode is created in-memory and will be delay-synchronized to the
280 * disk.
281 */
282int
a89aec1b
MD
283hammer_create_inode(hammer_transaction_t trans, struct vattr *vap,
284 struct ucred *cred, hammer_inode_t dip,
8cd0a023 285 struct hammer_inode **ipp)
66325755 286{
a89aec1b
MD
287 hammer_mount_t hmp;
288 hammer_inode_t ip;
6b4f890b 289 uid_t xuid;
66325755 290
8cd0a023
MD
291 hmp = trans->hmp;
292 ip = kmalloc(sizeof(*ip), M_HAMMER, M_WAITOK|M_ZERO);
b3deaf57 293 ++hammer_count_inodes;
a89aec1b 294 ip->obj_id = hammer_alloc_tid(trans);
8cd0a023 295 KKASSERT(ip->obj_id != 0);
7f7c1f84 296 ip->obj_asof = hmp->asof;
8cd0a023
MD
297 ip->hmp = hmp;
298 ip->flags = HAMMER_INODE_DDIRTY | HAMMER_INODE_RDIRTY |
42c7d26b 299 HAMMER_INODE_ITIMES | HAMMER_INODE_TIDLOCKED;
8cd0a023
MD
300 ip->last_tid = trans->tid;
301
302 RB_INIT(&ip->rec_tree);
303
304 ip->ino_rec.ino_atime = trans->tid;
305 ip->ino_rec.ino_mtime = trans->tid;
306 ip->ino_rec.ino_size = 0;
307 ip->ino_rec.ino_nlinks = 0;
308 /* XXX */
fe7678ee 309 ip->ino_rec.base.base.btype = HAMMER_BTREE_TYPE_RECORD;
8cd0a023
MD
310 ip->ino_rec.base.base.obj_id = ip->obj_id;
311 ip->ino_rec.base.base.key = 0;
312 ip->ino_rec.base.base.create_tid = trans->tid;
313 ip->ino_rec.base.base.delete_tid = 0;
314 ip->ino_rec.base.base.rec_type = HAMMER_RECTYPE_INODE;
315 ip->ino_rec.base.base.obj_type = hammer_get_obj_type(vap->va_type);
316
317 ip->ino_data.version = HAMMER_INODE_DATA_VERSION;
318 ip->ino_data.mode = vap->va_mode;
319 ip->ino_data.ctime = trans->tid;
320 ip->ino_data.parent_obj_id = (dip) ? dip->ino_rec.base.base.obj_id : 0;
6b4f890b 321
7a04d74f
MD
322 switch(ip->ino_rec.base.base.obj_type) {
323 case HAMMER_OBJTYPE_CDEV:
324 case HAMMER_OBJTYPE_BDEV:
325 ip->ino_data.rmajor = vap->va_rmajor;
326 ip->ino_data.rminor = vap->va_rminor;
327 break;
328 default:
329 break;
330 }
331
6b4f890b
MD
332 /*
333 * Calculate default uid/gid and overwrite with information from
334 * the vap.
335 */
336 xuid = hammer_to_unix_xid(&dip->ino_data.uid);
337 ip->ino_data.gid = dip->ino_data.gid;
338 xuid = vop_helper_create_uid(hmp->mp, dip->ino_data.mode, xuid, cred,
339 &vap->va_mode);
340 ip->ino_data.mode = vap->va_mode;
341
8cd0a023
MD
342 if (vap->va_vaflags & VA_UID_UUID_VALID)
343 ip->ino_data.uid = vap->va_uid_uuid;
6b4f890b
MD
344 else if (vap->va_uid != (uid_t)VNOVAL)
345 hammer_guid_to_uuid(&ip->ino_data.uid, xuid);
8cd0a023
MD
346 if (vap->va_vaflags & VA_GID_UUID_VALID)
347 ip->ino_data.gid = vap->va_gid_uuid;
6b4f890b 348 else if (vap->va_gid != (gid_t)VNOVAL)
8cd0a023
MD
349 hammer_guid_to_uuid(&ip->ino_data.gid, vap->va_gid);
350
351 hammer_ref(&ip->lock);
352 if (RB_INSERT(hammer_ino_rb_tree, &hmp->rb_inos_root, ip)) {
353 hammer_unref(&ip->lock);
a89aec1b 354 panic("hammer_create_inode: duplicate obj_id %llx", ip->obj_id);
8cd0a023
MD
355 }
356 *ipp = ip;
357 return(0);
66325755
MD
358}
359
d113fda1
MD
360/*
361 * Called by hammer_sync_inode().
362 */
363static int
7f7c1f84 364hammer_update_inode(hammer_inode_t ip)
c0ade690
MD
365{
366 struct hammer_cursor cursor;
367 hammer_record_t record;
368 int error;
d113fda1 369 hammer_tid_t last_tid;
c0ade690
MD
370
371 /*
76376933 372 * Locate the record on-disk and mark it as deleted. Both the B-Tree
195c19a1
MD
373 * node and the record must be marked deleted. The record may or
374 * may not be physically deleted, depending on the retention policy.
76376933 375 *
195c19a1
MD
376 * If the inode has already been deleted on-disk we have nothing
377 * to do.
c0ade690
MD
378 *
379 * XXX Update the inode record and data in-place if the retention
380 * policy allows it.
381 */
d113fda1 382 last_tid = ip->last_tid;
d26d0ae9 383retry:
c0ade690
MD
384 error = 0;
385
76376933
MD
386 if ((ip->flags & (HAMMER_INODE_ONDISK|HAMMER_INODE_DELONDISK)) ==
387 HAMMER_INODE_ONDISK) {
61aeeb33 388 hammer_init_cursor_hmp(&cursor, &ip->cache[0], ip->hmp);
c0ade690
MD
389 cursor.key_beg.obj_id = ip->obj_id;
390 cursor.key_beg.key = 0;
d5530d22 391 cursor.key_beg.create_tid = 0;
c0ade690
MD
392 cursor.key_beg.delete_tid = 0;
393 cursor.key_beg.rec_type = HAMMER_RECTYPE_INODE;
394 cursor.key_beg.obj_type = 0;
d5530d22
MD
395 cursor.asof = ip->obj_asof;
396 cursor.flags |= HAMMER_CURSOR_GET_RECORD | HAMMER_CURSOR_ASOF;
c0ade690
MD
397
398 error = hammer_btree_lookup(&cursor);
399
400 if (error == 0) {
d113fda1 401 error = hammer_ip_delete_record(&cursor, last_tid);
195c19a1
MD
402 if (error == 0)
403 ip->flags |= HAMMER_INODE_DELONDISK;
6a37e7e4 404 hammer_cache_node(cursor.node, &ip->cache[0]);
c0ade690 405 }
c0ade690 406 hammer_done_cursor(&cursor);
6a37e7e4
MD
407 if (error == EDEADLK)
408 goto retry;
c0ade690
MD
409 }
410
411 /*
412 * Write out a new record if the in-memory inode is not marked
fbc6e32a
MD
413 * as having been deleted. Update our inode statistics if this
414 * is the first application of the inode on-disk.
76376933
MD
415 *
416 * If the inode has been deleted permanently, HAMMER_INODE_DELONDISK
417 * will remain set and prevent further updates.
c0ade690
MD
418 */
419 if (error == 0 && (ip->flags & HAMMER_INODE_DELETED) == 0) {
47197d71 420 record = hammer_alloc_mem_record(ip, sizeof(struct hammer_inode_record));
c0ade690 421 record->rec.inode = ip->ino_rec;
d113fda1 422 record->rec.inode.base.base.create_tid = last_tid;
c0ade690
MD
423 record->rec.inode.base.data_len = sizeof(ip->ino_data);
424 record->data = (void *)&ip->ino_data;
47197d71 425 error = hammer_ip_sync_record(record);
b3deaf57
MD
426 record->flags |= HAMMER_RECF_DELETED;
427 hammer_rel_mem_record(record);
d26d0ae9 428 if (error == 0) {
d113fda1
MD
429 ip->flags &= ~(HAMMER_INODE_RDIRTY |
430 HAMMER_INODE_DDIRTY |
431 HAMMER_INODE_DELONDISK |
432 HAMMER_INODE_ITIMES);
d26d0ae9 433 if ((ip->flags & HAMMER_INODE_ONDISK) == 0) {
47197d71 434 hammer_modify_volume(ip->hmp->rootvol, NULL, 0);
0b075555 435 ++ip->hmp->rootvol->ondisk->vol0_stat_inodes;
d26d0ae9
MD
436 ip->flags |= HAMMER_INODE_ONDISK;
437 }
42c7d26b
MD
438
439 /*
440 * Unlock the sync TID if it was locked, now that
441 * we have written it out to disk.
442 */
443 ip->flags &= ~HAMMER_INODE_TIDLOCKED;
fbc6e32a 444 }
c0ade690
MD
445 }
446 return(error);
447}
448
a89aec1b 449/*
d113fda1
MD
450 * Update only the itimes fields. This is done no-historically. The
451 * record is updated in-place on the disk.
452 */
453static int
454hammer_update_itimes(hammer_inode_t ip)
455{
456 struct hammer_cursor cursor;
457 struct hammer_inode_record *rec;
458 int error;
459
6a37e7e4 460retry:
d113fda1
MD
461 error = 0;
462 if ((ip->flags & (HAMMER_INODE_ONDISK|HAMMER_INODE_DELONDISK)) ==
463 HAMMER_INODE_ONDISK) {
61aeeb33 464 hammer_init_cursor_hmp(&cursor, &ip->cache[0], ip->hmp);
d113fda1
MD
465 cursor.key_beg.obj_id = ip->obj_id;
466 cursor.key_beg.key = 0;
d5530d22 467 cursor.key_beg.create_tid = 0;
d113fda1
MD
468 cursor.key_beg.delete_tid = 0;
469 cursor.key_beg.rec_type = HAMMER_RECTYPE_INODE;
470 cursor.key_beg.obj_type = 0;
d5530d22
MD
471 cursor.asof = ip->obj_asof;
472 cursor.flags |= HAMMER_CURSOR_GET_RECORD | HAMMER_CURSOR_ASOF;
d113fda1
MD
473
474 error = hammer_btree_lookup(&cursor);
d113fda1
MD
475 if (error == 0) {
476 rec = &cursor.record->inode;
47197d71 477 hammer_modify_buffer(cursor.record_buffer, NULL, 0);
d113fda1
MD
478 rec->ino_atime = ip->ino_rec.ino_atime;
479 rec->ino_mtime = ip->ino_rec.ino_mtime;
d113fda1
MD
480 ip->flags &= ~HAMMER_INODE_ITIMES;
481 /* XXX recalculate crc */
6a37e7e4 482 hammer_cache_node(cursor.node, &ip->cache[0]);
d113fda1 483 }
d113fda1 484 hammer_done_cursor(&cursor);
6a37e7e4
MD
485 if (error == EDEADLK)
486 goto retry;
d113fda1
MD
487 }
488 return(error);
489}
490
491/*
492 * Release a reference on an inode. If asked to flush the last release
493 * will flush the inode.
a89aec1b 494 */
66325755 495void
a89aec1b 496hammer_rel_inode(struct hammer_inode *ip, int flush)
66325755
MD
497{
498 hammer_unref(&ip->lock);
d113fda1 499 if (flush)
a89aec1b 500 ip->flags |= HAMMER_INODE_FLUSH;
d113fda1
MD
501 if (ip->lock.refs == 0) {
502 if (ip->flags & HAMMER_INODE_FLUSH)
503 hammer_unload_inode(ip, (void *)MNT_WAIT);
504 else
505 hammer_unload_inode(ip, (void *)MNT_NOWAIT);
506 }
427e5fc6
MD
507}
508
27ea2398 509/*
8cd0a023
MD
510 * Unload and destroy the specified inode.
511 *
b33e2cc0 512 * (typically called via RB_SCAN)
27ea2398
MD
513 */
514int
d113fda1 515hammer_unload_inode(struct hammer_inode *ip, void *data)
27ea2398 516{
c0ade690
MD
517 int error;
518
a89aec1b
MD
519 KASSERT(ip->lock.refs == 0,
520 ("hammer_unload_inode: %d refs\n", ip->lock.refs));
8cd0a023
MD
521 KKASSERT(ip->vp == NULL);
522 hammer_ref(&ip->lock);
6b4f890b 523
d113fda1 524 error = hammer_sync_inode(ip, (int)data, 1);
c0ade690
MD
525 if (error)
526 kprintf("hammer_sync_inode failed error %d\n", error);
d113fda1
MD
527 if (ip->lock.refs == 1) {
528 KKASSERT(RB_EMPTY(&ip->rec_tree));
529 RB_REMOVE(hammer_ino_rb_tree, &ip->hmp->rb_inos_root, ip);
8cd0a023 530
61aeeb33
MD
531 hammer_uncache_node(&ip->cache[0]);
532 hammer_uncache_node(&ip->cache[1]);
d113fda1
MD
533 --hammer_count_inodes;
534 kfree(ip, M_HAMMER);
535 } else {
536 hammer_unref(&ip->lock);
537 }
27ea2398
MD
538 return(0);
539}
540
427e5fc6 541/*
d113fda1
MD
542 * A transaction has modified an inode, requiring updates as specified by
543 * the passed flags.
7f7c1f84 544 *
d113fda1
MD
545 * HAMMER_INODE_RDIRTY: Inode record has been updated
546 * HAMMER_INODE_DDIRTY: Inode data has been updated
547 * HAMMER_INODE_DELETED: Inode record/data must be deleted
548 * HAMMER_INODE_ITIMES: mtime/atime has been updated
549 *
550 * last_tid is the TID to use to generate the correct TID when the inode
32c90105
MD
551 * is synced to disk. The first inode record laid out on disk must match
552 * the transaction id of the related directory entry so only update last_tid
553 * if that has already occured.
427e5fc6 554 */
66325755
MD
555void
556hammer_modify_inode(struct hammer_transaction *trans,
557 struct hammer_inode *ip, int flags)
427e5fc6 558{
d113fda1
MD
559 KKASSERT ((ip->flags & HAMMER_INODE_RO) == 0 ||
560 (HAMMER_INODE_RDIRTY|HAMMER_INODE_DDIRTY|
561 HAMMER_INODE_DELETED|HAMMER_INODE_ITIMES) == 0);
562
563 if (flags &
564 (HAMMER_INODE_RDIRTY|HAMMER_INODE_DDIRTY|HAMMER_INODE_DELETED)) {
565 if (hammer_debug_tid) {
566 kprintf("hammer_modify_inode: %016llx (%08x)\n",
567 trans->tid, (int)(trans->tid / 1000000000LL));
568 }
42c7d26b
MD
569
570 /*
571 * Update the inode sync transaction id unless it's locked
572 * due to some prior required synchroznization. Locking the
573 * tid in the new flags overrides this (used by rename).
574 */
575 if ((ip->flags & HAMMER_INODE_TIDLOCKED) == 0)
576 ip->last_tid = trans->tid;
577 else if (flags & HAMMER_INODE_TIDLOCKED)
32c90105 578 ip->last_tid = trans->tid;
7f7c1f84
MD
579 }
580 ip->flags |= flags;
c0ade690
MD
581}
582
583/*
584 * Sync any dirty buffers and records associated with an inode. The
585 * inode's last_tid field is used as the transaction id for the sync,
586 * overriding any intermediate TIDs that were used for records. Note
587 * that the dirty buffer cache buffers do not have any knowledge of
588 * the transaction id they were modified under.
589 */
590static int
47197d71 591hammer_sync_inode_callback(hammer_record_t rec, void *data __unused)
c0ade690
MD
592{
593 int error;
594
d26d0ae9 595 hammer_ref(&rec->lock);
47197d71 596 error = hammer_ip_sync_record(rec);
b3deaf57 597 hammer_rel_mem_record(rec);
c0ade690
MD
598
599 if (error) {
b3deaf57
MD
600 error = -error;
601 if (error != -ENOSPC) {
602 kprintf("hammer_sync_inode_callback: sync failed rec "
603 "%p, error %d\n", rec, error);
604 }
c0ade690 605 }
b3deaf57 606 return(error);
c0ade690
MD
607}
608
609/*
610 * XXX error handling
611 */
612int
613hammer_sync_inode(hammer_inode_t ip, int waitfor, int handle_delete)
614{
615 struct hammer_transaction trans;
616 int error;
c0ade690 617
d113fda1
MD
618 if ((ip->flags & HAMMER_INODE_MODMASK) == 0) {
619 return(0);
620 }
621
c0ade690 622 hammer_lock_ex(&ip->lock);
d113fda1
MD
623
624 /*
625 * Use the transaction id of the last operation to sync.
626 */
627 if (ip->last_tid)
628 hammer_start_transaction_tid(&trans, ip->hmp, ip->last_tid);
629 else
630 hammer_start_transaction(&trans, ip->hmp);
c0ade690
MD
631
632 /*
633 * If the inode has been deleted (nlinks == 0), and the OS no longer
634 * has any references to it (handle_delete != 0), clean up in-memory
635 * data.
636 *
637 * NOTE: We do not set the RDIRTY flag when updating the delete_tid,
638 * setting HAMMER_INODE_DELETED takes care of it.
7f7c1f84
MD
639 *
640 * NOTE: Because we may sync records within this new transaction,
641 * force the inode update later on to use our transaction id or
642 * the delete_tid of the inode may be less then the create_tid of
643 * the inode update. XXX shouldn't happen but don't take the chance.
c0ade690 644 */
d113fda1
MD
645 if (ip->ino_rec.ino_nlinks == 0 && handle_delete &&
646 (ip->flags & HAMMER_INODE_GONE) == 0) {
647 ip->flags |= HAMMER_INODE_GONE;
c0ade690
MD
648 if (ip->vp)
649 vtruncbuf(ip->vp, 0, HAMMER_BUFSIZE);
7a04d74f 650 error = hammer_ip_delete_range_all(&trans, ip);
c0ade690
MD
651 KKASSERT(RB_EMPTY(&ip->rec_tree));
652 ip->ino_rec.base.base.delete_tid = trans.tid;
d113fda1 653 hammer_modify_inode(&trans, ip, HAMMER_INODE_DELETED);
47197d71 654 hammer_modify_volume(ip->hmp->rootvol, NULL, 0);
0b075555 655 --ip->hmp->rootvol->ondisk->vol0_stat_inodes;
c0ade690
MD
656 }
657
658 /*
f3b0f382 659 * Sync the buffer cache.
c0ade690 660 */
f3b0f382 661 if (ip->vp != NULL) {
c0ade690 662 error = vfsync(ip->vp, waitfor, 1, NULL, NULL);
f3b0f382
MD
663 if (RB_ROOT(&ip->vp->v_rbdirty_tree) == NULL)
664 ip->flags &= ~HAMMER_INODE_BUFS;
665 } else {
c0ade690 666 error = 0;
f3b0f382
MD
667 }
668
c0ade690
MD
669
670 /*
671 * Now sync related records
672 */
d26d0ae9
MD
673 for (;;) {
674 error = RB_SCAN(hammer_rec_rb_tree, &ip->rec_tree, NULL,
47197d71 675 hammer_sync_inode_callback, NULL);
d26d0ae9
MD
676 KKASSERT(error <= 0);
677 if (error < 0)
678 error = -error;
d26d0ae9 679 break;
c0ade690 680 }
d113fda1
MD
681 if (RB_EMPTY(&ip->rec_tree))
682 ip->flags &= ~HAMMER_INODE_XDIRTY;
c0ade690
MD
683
684 /*
685 * Now update the inode's on-disk inode-data and/or on-disk record.
686 */
687 switch(ip->flags & (HAMMER_INODE_DELETED|HAMMER_INODE_ONDISK)) {
688 case HAMMER_INODE_DELETED|HAMMER_INODE_ONDISK:
689 /*
690 * If deleted and on-disk, don't set any additional flags.
691 * the delete flag takes care of things.
692 */
693 break;
694 case HAMMER_INODE_DELETED:
695 /*
696 * Take care of the case where a deleted inode was never
697 * flushed to the disk in the first place.
698 */
d113fda1
MD
699 ip->flags &= ~(HAMMER_INODE_RDIRTY|HAMMER_INODE_DDIRTY|
700 HAMMER_INODE_XDIRTY|HAMMER_INODE_ITIMES);
d26d0ae9
MD
701 while (RB_ROOT(&ip->rec_tree)) {
702 hammer_record_t rec = RB_ROOT(&ip->rec_tree);
703 hammer_ref(&rec->lock);
b3deaf57
MD
704 rec->flags |= HAMMER_RECF_DELETED;
705 hammer_rel_mem_record(rec);
d26d0ae9 706 }
c0ade690
MD
707 break;
708 case HAMMER_INODE_ONDISK:
709 /*
710 * If already on-disk, do not set any additional flags.
711 */
712 break;
713 default:
714 /*
715 * If not on-disk and not deleted, set both dirty flags
716 * to force an initial record to be written.
717 */
718 ip->flags |= HAMMER_INODE_RDIRTY | HAMMER_INODE_DDIRTY;
719 break;
720 }
721
722 /*
d113fda1
MD
723 * If RDIRTY or DDIRTY is set, write out a new record. If the inode
724 * is already on-disk the old record is marked as deleted.
725 *
726 * If DELETED is set hammer_update_inode() will delete the existing
727 * record without writing out a new one.
728 *
729 * If *ONLY* the ITIMES flag is set we can update the record in-place.
c0ade690 730 */
d113fda1
MD
731 if ((ip->flags & (HAMMER_INODE_RDIRTY | HAMMER_INODE_DDIRTY |
732 HAMMER_INODE_ITIMES | HAMMER_INODE_DELETED)) ==
733 HAMMER_INODE_ITIMES) {
734 error = hammer_update_itimes(ip);
735 } else
c0ade690 736 if (ip->flags & (HAMMER_INODE_RDIRTY | HAMMER_INODE_DDIRTY |
d113fda1 737 HAMMER_INODE_ITIMES | HAMMER_INODE_DELETED)) {
7f7c1f84 738 error = hammer_update_inode(ip);
c0ade690
MD
739 }
740 hammer_commit_transaction(&trans);
741 hammer_unlock(&ip->lock);
742 return(error);
8cd0a023
MD
743}
744