Catch vkernel divide-by-0 traps a bit earlier so they are reported properly.
[dragonfly.git] / sys / vfs / hammer / hammer_vnops.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 *
6b4f890b 34 * $DragonFly: src/sys/vfs/hammer/hammer_vnops.c,v 1.5 2007/11/20 22:55:40 dillon Exp $
427e5fc6
MD
35 */
36
37#include <sys/param.h>
38#include <sys/systm.h>
39#include <sys/kernel.h>
40#include <sys/fcntl.h>
41#include <sys/namecache.h>
42#include <sys/vnode.h>
43#include <sys/lockf.h>
44#include <sys/event.h>
45#include <sys/stat.h>
46#include "hammer.h"
47
48/*
49 * USERFS VNOPS
50 */
51/*static int hammer_vop_vnoperate(struct vop_generic_args *);*/
66325755
MD
52static int hammer_vop_fsync(struct vop_fsync_args *);
53static int hammer_vop_read(struct vop_read_args *);
54static int hammer_vop_write(struct vop_write_args *);
55static int hammer_vop_access(struct vop_access_args *);
56static int hammer_vop_advlock(struct vop_advlock_args *);
57static int hammer_vop_close(struct vop_close_args *);
58static int hammer_vop_ncreate(struct vop_ncreate_args *);
59static int hammer_vop_getattr(struct vop_getattr_args *);
60static int hammer_vop_nresolve(struct vop_nresolve_args *);
61static int hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *);
62static int hammer_vop_nlink(struct vop_nlink_args *);
63static int hammer_vop_nmkdir(struct vop_nmkdir_args *);
64static int hammer_vop_nmknod(struct vop_nmknod_args *);
65static int hammer_vop_open(struct vop_open_args *);
66static int hammer_vop_pathconf(struct vop_pathconf_args *);
67static int hammer_vop_print(struct vop_print_args *);
68static int hammer_vop_readdir(struct vop_readdir_args *);
69static int hammer_vop_readlink(struct vop_readlink_args *);
70static int hammer_vop_nremove(struct vop_nremove_args *);
71static int hammer_vop_nrename(struct vop_nrename_args *);
72static int hammer_vop_nrmdir(struct vop_nrmdir_args *);
73static int hammer_vop_setattr(struct vop_setattr_args *);
74static int hammer_vop_strategy(struct vop_strategy_args *);
75static int hammer_vop_nsymlink(struct vop_nsymlink_args *);
76static int hammer_vop_nwhiteout(struct vop_nwhiteout_args *);
427e5fc6
MD
77
78struct vop_ops hammer_vnode_vops = {
79 .vop_default = vop_defaultop,
80 .vop_fsync = hammer_vop_fsync,
81 .vop_read = hammer_vop_read,
82 .vop_write = hammer_vop_write,
83 .vop_access = hammer_vop_access,
84 .vop_advlock = hammer_vop_advlock,
85 .vop_close = hammer_vop_close,
86 .vop_ncreate = hammer_vop_ncreate,
87 .vop_getattr = hammer_vop_getattr,
88 .vop_inactive = hammer_vop_inactive,
89 .vop_reclaim = hammer_vop_reclaim,
90 .vop_nresolve = hammer_vop_nresolve,
91 .vop_nlookupdotdot = hammer_vop_nlookupdotdot,
92 .vop_nlink = hammer_vop_nlink,
93 .vop_nmkdir = hammer_vop_nmkdir,
94 .vop_nmknod = hammer_vop_nmknod,
95 .vop_open = hammer_vop_open,
96 .vop_pathconf = hammer_vop_pathconf,
97 .vop_print = hammer_vop_print,
98 .vop_readdir = hammer_vop_readdir,
99 .vop_readlink = hammer_vop_readlink,
100 .vop_nremove = hammer_vop_nremove,
101 .vop_nrename = hammer_vop_nrename,
102 .vop_nrmdir = hammer_vop_nrmdir,
103 .vop_setattr = hammer_vop_setattr,
104 .vop_strategy = hammer_vop_strategy,
105 .vop_nsymlink = hammer_vop_nsymlink,
106 .vop_nwhiteout = hammer_vop_nwhiteout
107};
108
8cd0a023
MD
109static int hammer_dounlink(struct nchandle *nch, struct vnode *dvp,
110 struct ucred *cred, int flags);
111static int hammer_vop_strategy_read(struct vop_strategy_args *ap);
112static int hammer_vop_strategy_write(struct vop_strategy_args *ap);
113
427e5fc6
MD
114#if 0
115static
116int
117hammer_vop_vnoperate(struct vop_generic_args *)
118{
119 return (VOCALL(&hammer_vnode_vops, ap));
120}
121#endif
122
66325755
MD
123/*
124 * hammer_vop_fsync { vp, waitfor }
125 */
427e5fc6
MD
126static
127int
66325755 128hammer_vop_fsync(struct vop_fsync_args *ap)
427e5fc6
MD
129{
130 return EOPNOTSUPP;
131}
132
66325755
MD
133/*
134 * hammer_vop_read { vp, uio, ioflag, cred }
135 */
427e5fc6
MD
136static
137int
66325755 138hammer_vop_read(struct vop_read_args *ap)
427e5fc6 139{
66325755
MD
140 struct hammer_transaction trans;
141 struct hammer_inode *ip;
142 off_t offset;
143 struct buf *bp;
144 struct uio *uio;
145 int error;
146 int n;
8cd0a023 147 int seqcount;
66325755
MD
148
149 if (ap->a_vp->v_type != VREG)
150 return (EINVAL);
151 ip = VTOI(ap->a_vp);
152 error = 0;
8cd0a023 153 seqcount = ap->a_ioflag >> 16;
66325755 154
8cd0a023 155 hammer_start_transaction(&trans, ip->hmp);
66325755
MD
156
157 /*
158 * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
159 */
160 uio = ap->a_uio;
161 while (uio->uio_resid > 0 && uio->uio_offset < ip->ino_rec.ino_size) {
162 offset = uio->uio_offset & HAMMER_BUFMASK;
8cd0a023
MD
163 error = cluster_read(ap->a_vp, ip->ino_rec.ino_size,
164 uio->uio_offset - offset, HAMMER_BUFSIZE,
165 MAXBSIZE, seqcount, &bp);
66325755
MD
166 if (error) {
167 brelse(bp);
168 break;
169 }
8cd0a023 170 bp->b_flags |= B_CLUSTEROK;
66325755
MD
171 n = HAMMER_BUFSIZE - offset;
172 if (n > uio->uio_resid)
173 n = uio->uio_resid;
174 if (n > ip->ino_rec.ino_size - uio->uio_offset)
175 n = (int)(ip->ino_rec.ino_size - uio->uio_offset);
176 error = uiomove((char *)bp->b_data + offset, n, uio);
177 if (error) {
8cd0a023 178 bqrelse(bp);
66325755
MD
179 break;
180 }
181 ip->ino_rec.ino_atime = trans.tid;
182 hammer_modify_inode(&trans, ip, HAMMER_INODE_ITIMES);
183 bqrelse(bp);
184 }
185 hammer_commit_transaction(&trans);
186 return (error);
427e5fc6
MD
187}
188
66325755
MD
189/*
190 * hammer_vop_write { vp, uio, ioflag, cred }
191 */
427e5fc6
MD
192static
193int
66325755 194hammer_vop_write(struct vop_write_args *ap)
427e5fc6 195{
66325755
MD
196 struct hammer_transaction trans;
197 struct hammer_inode *ip;
198 struct uio *uio;
199 off_t offset;
200 struct buf *bp;
201 int error;
202 int n;
203
204 if (ap->a_vp->v_type != VREG)
205 return (EINVAL);
206 ip = VTOI(ap->a_vp);
207 error = 0;
208
209 /*
210 * Create a transaction to cover the operations we perform.
211 */
8cd0a023 212 hammer_start_transaction(&trans, ip->hmp);
66325755
MD
213 uio = ap->a_uio;
214
215 /*
216 * Check append mode
217 */
218 if (ap->a_ioflag & IO_APPEND)
219 uio->uio_offset = ip->ino_rec.ino_size;
220
221 /*
222 * Check for illegal write offsets. Valid range is 0...2^63-1
223 */
224 if (uio->uio_offset < 0 || uio->uio_offset + uio->uio_resid <= 0)
225 return (EFBIG);
226
227 /*
228 * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
229 */
230 while (uio->uio_resid > 0) {
231 offset = uio->uio_offset & HAMMER_BUFMASK;
232 if (offset == 0 && uio->uio_resid >= HAMMER_BUFSIZE) {
233 bp = getblk(ap->a_vp, uio->uio_offset, HAMMER_BUFSIZE,
234 0, 0);
235 } else if (offset == 0 && uio->uio_offset >= ip->ino_rec.ino_size) {
236 bp = getblk(ap->a_vp, uio->uio_offset, HAMMER_BUFSIZE,
237 0, 0);
238 vfs_bio_clrbuf(bp);
239 } else {
240 error = bread(ap->a_vp, uio->uio_offset - offset,
241 HAMMER_BUFSIZE, &bp);
242 if (error) {
243 brelse(bp);
244 break;
245 }
246 }
247 n = HAMMER_BUFSIZE - offset;
248 if (n > uio->uio_resid)
249 n = uio->uio_resid;
250 error = uiomove((char *)bp->b_data + offset, n, uio);
251 if (error) {
252 brelse(bp);
253 break;
254 }
8cd0a023 255 bp->b_flags |= B_CLUSTEROK;
66325755
MD
256 if (ip->ino_rec.ino_size < uio->uio_offset) {
257 ip->ino_rec.ino_size = uio->uio_offset;
258 ip->ino_rec.ino_mtime = trans.tid;
259 hammer_modify_inode(&trans, ip,
260 HAMMER_INODE_RDIRTY | HAMMER_INODE_ITIMES);
261 }
262 if (ap->a_ioflag & IO_SYNC) {
263 bwrite(bp);
264 } else if (ap->a_ioflag & IO_DIRECT) {
66325755
MD
265 bawrite(bp);
266 } else {
66325755
MD
267 bdwrite(bp);
268 }
269 }
270 if (error)
271 hammer_abort_transaction(&trans);
272 else
273 hammer_commit_transaction(&trans);
274 return (error);
427e5fc6
MD
275}
276
66325755
MD
277/*
278 * hammer_vop_access { vp, mode, cred }
279 */
427e5fc6
MD
280static
281int
66325755 282hammer_vop_access(struct vop_access_args *ap)
427e5fc6 283{
66325755
MD
284 struct hammer_inode *ip = VTOI(ap->a_vp);
285 uid_t uid;
286 gid_t gid;
287 int error;
288
289 uid = hammer_to_unix_xid(&ip->ino_data.uid);
290 gid = hammer_to_unix_xid(&ip->ino_data.gid);
291
292 error = vop_helper_access(ap, uid, gid, ip->ino_data.mode,
293 ip->ino_data.uflags);
294 return (error);
427e5fc6
MD
295}
296
66325755
MD
297/*
298 * hammer_vop_advlock { vp, id, op, fl, flags }
299 */
427e5fc6
MD
300static
301int
66325755 302hammer_vop_advlock(struct vop_advlock_args *ap)
427e5fc6 303{
66325755
MD
304 struct hammer_inode *ip = VTOI(ap->a_vp);
305
306 return (lf_advlock(ap, &ip->advlock, ip->ino_rec.ino_size));
427e5fc6
MD
307}
308
66325755
MD
309/*
310 * hammer_vop_close { vp, fflag }
311 */
427e5fc6
MD
312static
313int
66325755 314hammer_vop_close(struct vop_close_args *ap)
427e5fc6 315{
a89aec1b 316 return (vop_stdclose(ap));
427e5fc6
MD
317}
318
66325755
MD
319/*
320 * hammer_vop_ncreate { nch, dvp, vpp, cred, vap }
321 *
322 * The operating system has already ensured that the directory entry
323 * does not exist and done all appropriate namespace locking.
324 */
427e5fc6
MD
325static
326int
66325755 327hammer_vop_ncreate(struct vop_ncreate_args *ap)
427e5fc6 328{
66325755
MD
329 struct hammer_transaction trans;
330 struct hammer_inode *dip;
331 struct hammer_inode *nip;
332 struct nchandle *nch;
333 int error;
334
335 nch = ap->a_nch;
336 dip = VTOI(ap->a_dvp);
337
338 /*
339 * Create a transaction to cover the operations we perform.
340 */
8cd0a023 341 hammer_start_transaction(&trans, dip->hmp);
66325755
MD
342
343 /*
344 * Create a new filesystem object of the requested type. The
8cd0a023 345 * returned inode will be referenceds but not locked.
66325755 346 */
8cd0a023
MD
347
348 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
66325755
MD
349 if (error) {
350 hammer_abort_transaction(&trans);
351 *ap->a_vpp = NULL;
352 return (error);
353 }
66325755
MD
354
355 /*
356 * Add the new filesystem object to the directory. This will also
357 * bump the inode's link count.
358 */
a89aec1b 359 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
66325755
MD
360
361 /*
362 * Finish up.
363 */
364 if (error) {
a89aec1b 365 hammer_rel_inode(nip, 0);
66325755
MD
366 hammer_abort_transaction(&trans);
367 *ap->a_vpp = NULL;
368 } else {
369 hammer_commit_transaction(&trans);
370 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
a89aec1b
MD
371 hammer_rel_inode(nip, 0);
372 if (error == 0) {
373 cache_setunresolved(ap->a_nch);
374 cache_setvp(ap->a_nch, *ap->a_vpp);
375 }
66325755
MD
376 }
377 return (error);
427e5fc6
MD
378}
379
66325755
MD
380/*
381 * hammer_vop_getattr { vp, vap }
382 */
427e5fc6
MD
383static
384int
66325755 385hammer_vop_getattr(struct vop_getattr_args *ap)
427e5fc6 386{
66325755
MD
387 struct hammer_inode *ip = VTOI(ap->a_vp);
388 struct vattr *vap = ap->a_vap;
389
390#if 0
391 if (cache_check_fsmid_vp(ap->a_vp, &ip->fsmid) &&
392 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0 &&
393 ip->obj_asof == 0
394 ) {
395 /* LAZYMOD XXX */
396 }
397 hammer_itimes(ap->a_vp);
398#endif
399
400 vap->va_fsid = ip->hmp->fsid_udev;
401 vap->va_fileid = ip->ino_rec.base.base.obj_id;
402 vap->va_mode = ip->ino_data.mode;
403 vap->va_nlink = ip->ino_rec.ino_nlinks;
404 vap->va_uid = hammer_to_unix_xid(&ip->ino_data.uid);
405 vap->va_gid = hammer_to_unix_xid(&ip->ino_data.gid);
406 vap->va_rmajor = 0;
407 vap->va_rminor = 0;
408 vap->va_size = ip->ino_rec.ino_size;
409 hammer_to_timespec(ip->ino_rec.ino_atime, &vap->va_atime);
410 hammer_to_timespec(ip->ino_rec.ino_mtime, &vap->va_mtime);
411 hammer_to_timespec(ip->ino_data.ctime, &vap->va_ctime);
412 vap->va_flags = ip->ino_data.uflags;
413 vap->va_gen = 1; /* hammer inums are unique for all time */
414 vap->va_blocksize = 32768; /* XXX - extract from root volume */
415 vap->va_bytes = ip->ino_rec.ino_size;
416 vap->va_type = hammer_get_vnode_type(ip->ino_rec.base.base.obj_type);
417 vap->va_filerev = 0; /* XXX */
418 /* mtime uniquely identifies any adjustments made to the file */
419 vap->va_fsmid = ip->ino_rec.ino_mtime;
420 vap->va_uid_uuid = ip->ino_data.uid;
421 vap->va_gid_uuid = ip->ino_data.gid;
422 vap->va_fsid_uuid = ip->hmp->fsid;
423 vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
424 VA_FSID_UUID_VALID;
425 return(0);
427e5fc6
MD
426}
427
66325755
MD
428/*
429 * hammer_vop_nresolve { nch, dvp, cred }
430 *
431 * Locate the requested directory entry.
432 */
427e5fc6
MD
433static
434int
66325755 435hammer_vop_nresolve(struct vop_nresolve_args *ap)
427e5fc6 436{
66325755
MD
437 struct namecache *ncp;
438 struct hammer_inode *dip;
8cd0a023
MD
439 struct hammer_cursor cursor;
440 union hammer_record_ondisk *rec;
66325755
MD
441 struct vnode *vp;
442 int64_t namekey;
443 int error;
66325755 444
8cd0a023
MD
445 /*
446 * Calculate the namekey and setup the key range for the scan. This
447 * works kinda like a chained hash table where the lower 32 bits
448 * of the namekey synthesize the chain.
449 *
450 * The key range is inclusive of both key_beg and key_end.
451 */
66325755
MD
452 dip = VTOI(ap->a_dvp);
453 ncp = ap->a_nch->ncp;
454 namekey = hammer_directory_namekey(ncp->nc_name, ncp->nc_nlen);
a89aec1b 455 kprintf("hammer_vop_nresolve %s dip %p\n", ncp->nc_name, dip);
66325755 456
8cd0a023
MD
457 hammer_init_cursor_ip(&cursor, dip);
458 cursor.key_beg.obj_id = dip->obj_id;
459 cursor.key_beg.key = namekey;
460 cursor.key_beg.create_tid = dip->obj_asof;
461 cursor.key_beg.delete_tid = 0;
462 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
463 cursor.key_beg.obj_type = 0;
66325755 464
8cd0a023
MD
465 cursor.key_end = cursor.key_beg;
466 cursor.key_end.key |= 0xFFFFFFFFULL;
66325755
MD
467
468 /*
8cd0a023 469 * Scan all matching records (the chain), locate the one matching
a89aec1b 470 * the requested path component.
8cd0a023
MD
471 *
472 * The hammer_ip_*() functions merge in-memory records with on-disk
473 * records for the purposes of the search.
66325755 474 */
a89aec1b
MD
475 error = hammer_ip_first(&cursor, dip);
476 while (error == 0) {
477 error = hammer_ip_resolve_data(&cursor);
478 if (error)
66325755 479 break;
a89aec1b 480 rec = cursor.record;
8cd0a023 481 if (ncp->nc_nlen == rec->entry.base.data_len &&
a89aec1b 482 bcmp(ncp->nc_name, cursor.data, ncp->nc_nlen) == 0) {
66325755
MD
483 break;
484 }
a89aec1b 485 error = hammer_ip_next(&cursor);
66325755
MD
486 }
487 if (error == 0) {
8cd0a023 488 error = hammer_vfs_vget(dip->hmp->mp, rec->entry.obj_id, &vp);
66325755
MD
489 if (error == 0) {
490 vn_unlock(vp);
491 cache_setvp(ap->a_nch, vp);
492 vrele(vp);
493 }
494 } else if (error == ENOENT) {
495 cache_setvp(ap->a_nch, NULL);
496 }
8cd0a023 497 hammer_done_cursor(&cursor);
a89aec1b 498 kprintf("hammer_vop_nresolve error %d\n", error);
66325755 499 return (error);
427e5fc6
MD
500}
501
66325755
MD
502/*
503 * hammer_vop_nlookupdotdot { dvp, vpp, cred }
504 *
505 * Locate the parent directory of a directory vnode.
506 *
507 * dvp is referenced but not locked. *vpp must be returned referenced and
508 * locked. A parent_obj_id of 0 does not necessarily indicate that we are
509 * at the root, instead it could indicate that the directory we were in was
510 * removed.
511 */
427e5fc6
MD
512static
513int
66325755 514hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
427e5fc6 515{
66325755
MD
516 struct hammer_inode *dip;
517 u_int64_t parent_obj_id;
518
519 dip = VTOI(ap->a_dvp);
520 if ((parent_obj_id = dip->ino_data.parent_obj_id) == 0) {
521 *ap->a_vpp = NULL;
522 return ENOENT;
523 }
524 return(hammer_vfs_vget(dip->hmp->mp, parent_obj_id, ap->a_vpp));
427e5fc6
MD
525}
526
66325755
MD
527/*
528 * hammer_vop_nlink { nch, dvp, vp, cred }
529 */
427e5fc6
MD
530static
531int
66325755 532hammer_vop_nlink(struct vop_nlink_args *ap)
427e5fc6 533{
66325755
MD
534 struct hammer_transaction trans;
535 struct hammer_inode *dip;
536 struct hammer_inode *ip;
537 struct nchandle *nch;
538 int error;
539
540 nch = ap->a_nch;
541 dip = VTOI(ap->a_dvp);
542 ip = VTOI(ap->a_vp);
543
544 /*
545 * Create a transaction to cover the operations we perform.
546 */
8cd0a023 547 hammer_start_transaction(&trans, dip->hmp);
66325755
MD
548
549 /*
550 * Add the filesystem object to the directory. Note that neither
551 * dip nor ip are referenced or locked, but their vnodes are
552 * referenced. This function will bump the inode's link count.
553 */
a89aec1b 554 error = hammer_ip_add_directory(&trans, dip, nch->ncp, ip);
66325755
MD
555
556 /*
557 * Finish up.
558 */
559 if (error) {
560 hammer_abort_transaction(&trans);
561 } else {
6b4f890b
MD
562 cache_setunresolved(nch);
563 cache_setvp(nch, ap->a_vp);
66325755
MD
564 hammer_commit_transaction(&trans);
565 }
566 return (error);
427e5fc6
MD
567}
568
66325755
MD
569/*
570 * hammer_vop_nmkdir { nch, dvp, vpp, cred, vap }
571 *
572 * The operating system has already ensured that the directory entry
573 * does not exist and done all appropriate namespace locking.
574 */
427e5fc6
MD
575static
576int
66325755 577hammer_vop_nmkdir(struct vop_nmkdir_args *ap)
427e5fc6 578{
66325755
MD
579 struct hammer_transaction trans;
580 struct hammer_inode *dip;
581 struct hammer_inode *nip;
582 struct nchandle *nch;
583 int error;
584
585 nch = ap->a_nch;
586 dip = VTOI(ap->a_dvp);
587
588 /*
589 * Create a transaction to cover the operations we perform.
590 */
8cd0a023 591 hammer_start_transaction(&trans, dip->hmp);
66325755
MD
592
593 /*
594 * Create a new filesystem object of the requested type. The
8cd0a023 595 * returned inode will be referenced but not locked.
66325755 596 */
8cd0a023 597 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
66325755
MD
598 if (error) {
599 hammer_abort_transaction(&trans);
600 *ap->a_vpp = NULL;
601 return (error);
602 }
66325755
MD
603
604 /*
605 * Add the new filesystem object to the directory. This will also
606 * bump the inode's link count.
607 */
a89aec1b 608 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
66325755
MD
609
610 /*
611 * Finish up.
612 */
613 if (error) {
a89aec1b 614 hammer_rel_inode(nip, 0);
66325755
MD
615 hammer_abort_transaction(&trans);
616 *ap->a_vpp = NULL;
617 } else {
618 hammer_commit_transaction(&trans);
619 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
a89aec1b
MD
620 hammer_rel_inode(nip, 0);
621 if (error == 0) {
622 cache_setunresolved(ap->a_nch);
623 cache_setvp(ap->a_nch, *ap->a_vpp);
624 }
66325755
MD
625 }
626 return (error);
427e5fc6
MD
627}
628
66325755
MD
629/*
630 * hammer_vop_nmknod { nch, dvp, vpp, cred, vap }
631 *
632 * The operating system has already ensured that the directory entry
633 * does not exist and done all appropriate namespace locking.
634 */
427e5fc6
MD
635static
636int
66325755 637hammer_vop_nmknod(struct vop_nmknod_args *ap)
427e5fc6 638{
66325755
MD
639 struct hammer_transaction trans;
640 struct hammer_inode *dip;
641 struct hammer_inode *nip;
642 struct nchandle *nch;
643 int error;
644
645 nch = ap->a_nch;
646 dip = VTOI(ap->a_dvp);
647
648 /*
649 * Create a transaction to cover the operations we perform.
650 */
8cd0a023 651 hammer_start_transaction(&trans, dip->hmp);
66325755
MD
652
653 /*
654 * Create a new filesystem object of the requested type. The
8cd0a023 655 * returned inode will be referenced but not locked.
66325755 656 */
8cd0a023 657 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
66325755
MD
658 if (error) {
659 hammer_abort_transaction(&trans);
660 *ap->a_vpp = NULL;
661 return (error);
662 }
66325755
MD
663
664 /*
665 * Add the new filesystem object to the directory. This will also
666 * bump the inode's link count.
667 */
a89aec1b 668 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
66325755
MD
669
670 /*
671 * Finish up.
672 */
673 if (error) {
a89aec1b 674 hammer_rel_inode(nip, 0);
66325755
MD
675 hammer_abort_transaction(&trans);
676 *ap->a_vpp = NULL;
677 } else {
678 hammer_commit_transaction(&trans);
679 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
a89aec1b
MD
680 hammer_rel_inode(nip, 0);
681 if (error == 0) {
682 cache_setunresolved(ap->a_nch);
683 cache_setvp(ap->a_nch, *ap->a_vpp);
684 }
66325755
MD
685 }
686 return (error);
427e5fc6
MD
687}
688
66325755
MD
689/*
690 * hammer_vop_open { vp, mode, cred, fp }
691 */
427e5fc6
MD
692static
693int
66325755 694hammer_vop_open(struct vop_open_args *ap)
427e5fc6 695{
a89aec1b
MD
696 kprintf("hammer_vop_open\n");
697 return(vop_stdopen(ap));
427e5fc6
MD
698}
699
66325755
MD
700/*
701 * hammer_vop_pathconf { vp, name, retval }
702 */
427e5fc6
MD
703static
704int
66325755 705hammer_vop_pathconf(struct vop_pathconf_args *ap)
427e5fc6
MD
706{
707 return EOPNOTSUPP;
708}
709
66325755
MD
710/*
711 * hammer_vop_print { vp }
712 */
427e5fc6
MD
713static
714int
66325755 715hammer_vop_print(struct vop_print_args *ap)
427e5fc6
MD
716{
717 return EOPNOTSUPP;
718}
719
66325755 720/*
6b4f890b 721 * hammer_vop_readdir { vp, uio, cred, *eofflag, *ncookies, off_t **cookies }
66325755 722 */
427e5fc6
MD
723static
724int
66325755 725hammer_vop_readdir(struct vop_readdir_args *ap)
427e5fc6 726{
6b4f890b
MD
727 struct hammer_cursor cursor;
728 struct hammer_inode *ip;
729 struct uio *uio;
730 hammer_record_ondisk_t rec;
731 hammer_base_elm_t base;
732 int error;
733 int cookie_index;
734 int ncookies;
735 off_t *cookies;
736 off_t saveoff;
737 int r;
738
739 ip = VTOI(ap->a_vp);
740 uio = ap->a_uio;
741 hammer_init_cursor_ip(&cursor, ip);
742
743 /*
744 * Key range (begin and end inclusive) to scan. Directory keys
745 * directly translate to a 64 bit 'seek' position.
746 */
747 cursor.key_beg.obj_id = ip->obj_id;
748 cursor.key_beg.create_tid = ip->obj_asof;
749 cursor.key_beg.delete_tid = 0;
750 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
751 cursor.key_beg.obj_type = 0;
752 cursor.key_beg.key = uio->uio_offset;
753
754 cursor.key_end = cursor.key_beg;
755 cursor.key_end.key = HAMMER_MAX_KEY;
756
757 if (ap->a_ncookies) {
758 ncookies = uio->uio_resid / 16 + 1;
759 if (ncookies > 1024)
760 ncookies = 1024;
761 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
762 cookie_index = 0;
763 } else {
764 ncookies = -1;
765 cookies = NULL;
766 cookie_index = 0;
767 }
768
769 saveoff = cursor.key_beg.key;
770 error = hammer_ip_first(&cursor, ip);
771
772 while (error == 0) {
773 error = hammer_ip_resolve_data(&cursor);
774 if (error)
775 break;
776 rec = cursor.record;
777 base = &rec->base.base;
778 saveoff = base->key;
779
780 r = vop_write_dirent(
781 &error, uio, rec->entry.obj_id,
782 rec->entry.base.data_len,
783 hammer_get_dtype(rec->entry.base.base.obj_type),
784 (void *)cursor.data);
785 if (r)
786 break;
787 ++saveoff;
788 if (cookies)
789 cookies[cookie_index] = base->key;
790 ++cookie_index;
791 if (cookie_index == ncookies)
792 break;
793 error = hammer_ip_next(&cursor);
794 }
795 hammer_done_cursor(&cursor);
796
797 if (ap->a_eofflag)
798 *ap->a_eofflag = (error == ENOENT);
799 if (error == ENOENT)
800 error = 0;
801 uio->uio_offset = saveoff;
802 if (error && cookie_index == 0) {
803 if (cookies) {
804 kfree(cookies, M_TEMP);
805 *ap->a_ncookies = 0;
806 *ap->a_cookies = NULL;
807 }
808 } else {
809 error = 0;
810 if (cookies) {
811 *ap->a_ncookies = cookie_index;
812 *ap->a_cookies = cookies;
813 }
814 }
815 return(error);
427e5fc6
MD
816}
817
66325755
MD
818/*
819 * hammer_vop_readlink { vp, uio, cred }
820 */
427e5fc6
MD
821static
822int
66325755 823hammer_vop_readlink(struct vop_readlink_args *ap)
427e5fc6
MD
824{
825 return EOPNOTSUPP;
826}
827
66325755
MD
828/*
829 * hammer_vop_nremove { nch, dvp, cred }
830 */
427e5fc6
MD
831static
832int
66325755 833hammer_vop_nremove(struct vop_nremove_args *ap)
427e5fc6 834{
8cd0a023 835 return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, 0));
427e5fc6
MD
836}
837
66325755
MD
838/*
839 * hammer_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
840 */
427e5fc6
MD
841static
842int
66325755 843hammer_vop_nrename(struct vop_nrename_args *ap)
427e5fc6 844{
8cd0a023
MD
845 struct hammer_transaction trans;
846 struct namecache *fncp;
847 struct namecache *tncp;
848 struct hammer_inode *fdip;
849 struct hammer_inode *tdip;
850 struct hammer_inode *ip;
851 struct hammer_cursor cursor;
852 union hammer_record_ondisk *rec;
853 int64_t namekey;
854 int error;
855
856 fdip = VTOI(ap->a_fdvp);
857 tdip = VTOI(ap->a_tdvp);
858 fncp = ap->a_fnch->ncp;
859 tncp = ap->a_tnch->ncp;
860 hammer_start_transaction(&trans, fdip->hmp);
861
862 /*
863 * Extract the hammer_inode from fncp and add link to the target
864 * directory.
865 */
866 ip = VTOI(fncp->nc_vp);
867 KKASSERT(ip != NULL);
868
a89aec1b 869 error = hammer_ip_add_directory(&trans, tdip, tncp, ip);
8cd0a023
MD
870
871 /*
872 * Locate the record in the originating directory and remove it.
873 *
874 * Calculate the namekey and setup the key range for the scan. This
875 * works kinda like a chained hash table where the lower 32 bits
876 * of the namekey synthesize the chain.
877 *
878 * The key range is inclusive of both key_beg and key_end.
879 */
880 namekey = hammer_directory_namekey(fncp->nc_name, fncp->nc_nlen);
881
882 hammer_init_cursor_ip(&cursor, fdip);
883 cursor.key_beg.obj_id = fdip->obj_id;
884 cursor.key_beg.key = namekey;
885 cursor.key_beg.create_tid = fdip->obj_asof;
886 cursor.key_beg.delete_tid = 0;
887 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
888 cursor.key_beg.obj_type = 0;
889
890 cursor.key_end = cursor.key_beg;
891 cursor.key_end.key |= 0xFFFFFFFFULL;
892
893 /*
894 * Scan all matching records (the chain), locate the one matching
a89aec1b 895 * the requested path component.
8cd0a023
MD
896 *
897 * The hammer_ip_*() functions merge in-memory records with on-disk
898 * records for the purposes of the search.
899 */
a89aec1b
MD
900 error = hammer_ip_first(&cursor, fdip);
901 while (error == 0) {
8cd0a023
MD
902 if (hammer_ip_resolve_data(&cursor) != 0)
903 break;
a89aec1b 904 rec = cursor.record;
8cd0a023
MD
905 if (fncp->nc_nlen == rec->entry.base.data_len &&
906 bcmp(fncp->nc_name, cursor.data, fncp->nc_nlen) == 0) {
907 break;
908 }
a89aec1b 909 error = hammer_ip_next(&cursor);
8cd0a023 910 }
8cd0a023
MD
911
912 /*
913 * If all is ok we have to get the inode so we can adjust nlinks.
914 */
915 if (error)
916 goto done;
a89aec1b 917 error = hammer_ip_del_directory(&trans, &cursor, fdip, ip);
8cd0a023
MD
918 if (error == 0) {
919 cache_rename(ap->a_fnch, ap->a_tnch);
920 cache_setvp(ap->a_tnch, ip->vp);
921 }
922done:
923 if (error == 0) {
924 hammer_commit_transaction(&trans);
925 } else {
926 hammer_abort_transaction(&trans);
927 }
928 hammer_done_cursor(&cursor);
929 return (error);
427e5fc6
MD
930}
931
66325755
MD
932/*
933 * hammer_vop_nrmdir { nch, dvp, cred }
934 */
427e5fc6
MD
935static
936int
66325755 937hammer_vop_nrmdir(struct vop_nrmdir_args *ap)
427e5fc6 938{
8cd0a023
MD
939 /* XXX check that directory is empty */
940
941 return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, 0));
427e5fc6
MD
942}
943
66325755
MD
944/*
945 * hammer_vop_setattr { vp, vap, cred }
946 */
427e5fc6
MD
947static
948int
66325755 949hammer_vop_setattr(struct vop_setattr_args *ap)
427e5fc6 950{
8cd0a023
MD
951 struct hammer_transaction trans;
952 struct vattr *vap;
953 struct hammer_inode *ip;
954 int modflags;
955 int error;
956 u_int32_t flags;
957 uuid_t uuid;
958
959 vap = ap->a_vap;
960 ip = ap->a_vp->v_data;
961 modflags = 0;
962
963 if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
964 return(EROFS);
965
966 hammer_start_transaction(&trans, ip->hmp);
967 error = 0;
968
969 if (vap->va_flags != VNOVAL) {
970 flags = ip->ino_data.uflags;
971 error = vop_helper_setattr_flags(&flags, vap->va_flags,
972 hammer_to_unix_xid(&ip->ino_data.uid),
973 ap->a_cred);
974 if (error == 0) {
975 if (ip->ino_data.uflags != flags) {
976 ip->ino_data.uflags = flags;
977 modflags |= HAMMER_INODE_DDIRTY;
978 }
979 if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
980 error = 0;
981 goto done;
982 }
983 }
984 goto done;
985 }
986 if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
987 error = EPERM;
988 goto done;
989 }
990 if (vap->va_uid != (uid_t)VNOVAL) {
991 hammer_guid_to_uuid(&uuid, vap->va_uid);
6b4f890b 992 if (bcmp(&uuid, &ip->ino_data.uid, sizeof(uuid)) != 0) {
8cd0a023
MD
993 ip->ino_data.uid = uuid;
994 modflags |= HAMMER_INODE_DDIRTY;
995 }
996 }
997 if (vap->va_gid != (uid_t)VNOVAL) {
6b4f890b
MD
998 hammer_guid_to_uuid(&uuid, vap->va_gid);
999 if (bcmp(&uuid, &ip->ino_data.gid, sizeof(uuid)) != 0) {
8cd0a023
MD
1000 ip->ino_data.gid = uuid;
1001 modflags |= HAMMER_INODE_DDIRTY;
1002 }
1003 }
1004 if (vap->va_size != VNOVAL) {
1005 switch(ap->a_vp->v_type) {
1006 case VREG:
1007 case VDATABASE:
a89aec1b 1008 error = hammer_ip_delete_range(&trans, ip,
8cd0a023
MD
1009 vap->va_size,
1010 0x7FFFFFFFFFFFFFFFLL);
1011 break;
1012 default:
1013 error = EINVAL;
1014 goto done;
1015 }
1016 }
1017 if (vap->va_atime.tv_sec != VNOVAL) {
1018 ip->ino_rec.ino_atime =
1019 hammer_timespec_to_transid(&vap->va_atime);
1020 modflags |= HAMMER_INODE_ITIMES;
1021 }
1022 if (vap->va_mtime.tv_sec != VNOVAL) {
1023 ip->ino_rec.ino_mtime =
1024 hammer_timespec_to_transid(&vap->va_mtime);
1025 modflags |= HAMMER_INODE_ITIMES;
1026 }
1027 if (vap->va_mode != (mode_t)VNOVAL) {
1028 if (ip->ino_data.mode != vap->va_mode) {
1029 ip->ino_data.mode = vap->va_mode;
1030 modflags |= HAMMER_INODE_DDIRTY;
1031 }
1032 }
1033done:
1034 if (error) {
1035 hammer_abort_transaction(&trans);
1036 } else {
1037 if (modflags)
1038 hammer_modify_inode(&trans, ip, modflags);
1039 hammer_commit_transaction(&trans);
1040 }
1041 return (error);
427e5fc6
MD
1042}
1043
66325755
MD
1044/*
1045 * hammer_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1046 */
427e5fc6
MD
1047static
1048int
66325755 1049hammer_vop_nsymlink(struct vop_nsymlink_args *ap)
427e5fc6
MD
1050{
1051 return EOPNOTSUPP;
1052}
1053
66325755
MD
1054/*
1055 * hammer_vop_nwhiteout { nch, dvp, cred, flags }
1056 */
427e5fc6
MD
1057static
1058int
66325755 1059hammer_vop_nwhiteout(struct vop_nwhiteout_args *ap)
427e5fc6 1060{
8cd0a023 1061 return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, ap->a_flags));
427e5fc6
MD
1062}
1063
66325755
MD
1064/*
1065 * hammer_vop_strategy { vp, bio }
8cd0a023
MD
1066 *
1067 * Strategy call, used for regular file read & write only. Note that the
1068 * bp may represent a cluster.
1069 *
1070 * To simplify operation and allow better optimizations in the future,
1071 * this code does not make any assumptions with regards to buffer alignment
1072 * or size.
66325755 1073 */
427e5fc6
MD
1074static
1075int
66325755 1076hammer_vop_strategy(struct vop_strategy_args *ap)
427e5fc6 1077{
8cd0a023
MD
1078 struct buf *bp;
1079 int error;
1080
1081 bp = ap->a_bio->bio_buf;
1082
1083 switch(bp->b_cmd) {
1084 case BUF_CMD_READ:
1085 error = hammer_vop_strategy_read(ap);
1086 break;
1087 case BUF_CMD_WRITE:
1088 error = hammer_vop_strategy_write(ap);
1089 break;
1090 default:
1091 error = EINVAL;
1092 break;
1093 }
1094 bp->b_error = error;
1095 if (error)
1096 bp->b_flags |= B_ERROR;
1097 biodone(ap->a_bio);
1098 return (error);
427e5fc6
MD
1099}
1100
8cd0a023
MD
1101/*
1102 * Read from a regular file. Iterate the related records and fill in the
1103 * BIO/BUF. Gaps are zero-filled.
1104 *
1105 * The support code in hammer_object.c should be used to deal with mixed
1106 * in-memory and on-disk records.
1107 *
1108 * XXX atime update
1109 */
1110static
1111int
1112hammer_vop_strategy_read(struct vop_strategy_args *ap)
1113{
1114 struct hammer_inode *ip = ap->a_vp->v_data;
1115 struct hammer_cursor cursor;
1116 hammer_record_ondisk_t rec;
1117 hammer_base_elm_t base;
1118 struct bio *bio;
1119 struct buf *bp;
1120 int64_t rec_offset;
a89aec1b 1121 int64_t ran_end;
8cd0a023
MD
1122 int error;
1123 int boff;
1124 int roff;
1125 int n;
1126
1127 bio = ap->a_bio;
1128 bp = bio->bio_buf;
1129
1130 hammer_init_cursor_ip(&cursor, ip);
1131
1132 /*
1133 * Key range (begin and end inclusive) to scan. Note that the key's
1134 * stored in the actual records represent the
1135 */
1136 cursor.key_beg.obj_id = ip->obj_id;
1137 cursor.key_beg.create_tid = ip->obj_asof;
1138 cursor.key_beg.delete_tid = 0;
8cd0a023
MD
1139 cursor.key_beg.obj_type = 0;
1140 cursor.key_beg.key = bio->bio_offset;
1141
1142 cursor.key_end = cursor.key_beg;
a89aec1b
MD
1143 if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1144 cursor.key_beg.rec_type = HAMMER_RECTYPE_DB;
1145 cursor.key_end.rec_type = HAMMER_RECTYPE_DB;
1146 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1147 } else {
1148 ran_end = bio->bio_offset + bp->b_bufsize - 1;
1149 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
1150 cursor.key_end.rec_type = HAMMER_RECTYPE_DATA;
1151 if (ran_end + MAXPHYS < ran_end)
1152 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1153 else
1154 cursor.key_end.key = ran_end + MAXPHYS;
1155 }
8cd0a023 1156
a89aec1b 1157 error = hammer_ip_first(&cursor, ip);
8cd0a023
MD
1158 boff = 0;
1159
a89aec1b
MD
1160 while (error == 0) {
1161 error = hammer_ip_resolve_data(&cursor);
1162 if (error)
66325755 1163 break;
a89aec1b 1164 rec = cursor.record;
8cd0a023
MD
1165 base = &rec->base.base;
1166
a89aec1b 1167 rec_offset = base->key - rec->data.base.data_len + 1;
8cd0a023 1168
66325755 1169 /*
a89aec1b 1170 * Calculate the gap, if any, and zero-fill it.
66325755 1171 */
8cd0a023
MD
1172 n = (int)(rec_offset - (bio->bio_offset + boff));
1173 if (n > 0) {
a89aec1b
MD
1174 if (n > bp->b_bufsize - boff)
1175 n = bp->b_bufsize - boff;
8cd0a023
MD
1176 kprintf("zfill %d bytes\n", n);
1177 bzero((char *)bp->b_data + boff, n);
1178 boff += n;
1179 n = 0;
66325755 1180 }
8cd0a023
MD
1181
1182 /*
1183 * Calculate the data offset in the record and the number
1184 * of bytes we can copy.
a89aec1b
MD
1185 *
1186 * Note there is a degenerate case here where boff may
1187 * already be at bp->b_bufsize.
8cd0a023
MD
1188 */
1189 roff = -n;
1190 n = rec->data.base.data_len - roff;
1191 KKASSERT(n > 0);
1192 if (n > bp->b_bufsize - boff)
1193 n = bp->b_bufsize - boff;
1194 bcopy((char *)cursor.data + roff, (char *)bp->b_data + boff, n);
1195 boff += n;
1196 if (boff == bp->b_bufsize)
66325755 1197 break;
a89aec1b 1198 error = hammer_ip_next(&cursor);
66325755 1199 }
8cd0a023 1200 hammer_done_cursor(&cursor);
66325755
MD
1201
1202 /*
8cd0a023 1203 * There may have been a gap after the last record
66325755 1204 */
8cd0a023
MD
1205 if (error == ENOENT)
1206 error = 0;
1207 if (error == 0 && boff != bp->b_bufsize) {
1208 bzero((char *)bp->b_data + boff, bp->b_bufsize - boff);
1209 /* boff = bp->b_bufsize; */
1210 }
1211 bp->b_resid = 0;
1212 return(error);
1213}
1214
1215/*
1216 * Write to a regular file. Iterate the related records and mark for
1217 * deletion. If existing edge records (left and right side) overlap our
1218 * write they have to be marked deleted and new records created, usually
1219 * referencing a portion of the original data. Then add a record to
1220 * represent the buffer.
1221 *
1222 * The support code in hammer_object.c should be used to deal with mixed
1223 * in-memory and on-disk records.
1224 */
1225static
1226int
1227hammer_vop_strategy_write(struct vop_strategy_args *ap)
1228{
1229 struct hammer_transaction trans;
1230 hammer_inode_t ip;
1231 struct bio *bio;
1232 struct buf *bp;
1233 int error;
1234
1235 bio = ap->a_bio;
1236 bp = bio->bio_buf;
1237 ip = ap->a_vp->v_data;
1238 hammer_start_transaction(&trans, ip->hmp);
1239
1240 /*
1241 * Delete any records overlapping our range. This function will
1242 * properly
1243 */
a89aec1b
MD
1244 if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1245 error = hammer_ip_delete_range(&trans, ip, bio->bio_offset,
1246 bio->bio_offset);
1247 } else {
1248 error = hammer_ip_delete_range(&trans, ip, bio->bio_offset,
1249 bio->bio_offset + bp->b_bufsize - 1);
1250 }
8cd0a023
MD
1251
1252 /*
1253 * Add a single record to cover the write
1254 */
1255 if (error == 0) {
a89aec1b
MD
1256 error = hammer_ip_add_data(&trans, ip, bio->bio_offset,
1257 bp->b_data, bp->b_bufsize);
66325755
MD
1258 }
1259
1260 /*
8cd0a023 1261 * If an error occured abort the transaction
66325755 1262 */
8cd0a023
MD
1263 if (error) {
1264 /* XXX undo deletion */
1265 hammer_abort_transaction(&trans);
1266 bp->b_resid = bp->b_bufsize;
1267 } else {
1268 hammer_commit_transaction(&trans);
1269 bp->b_resid = 0;
1270 }
1271 return(error);
1272}
1273
1274/*
1275 * dounlink - disconnect a directory entry
1276 *
1277 * XXX whiteout support not really in yet
1278 */
1279static int
1280hammer_dounlink(struct nchandle *nch, struct vnode *dvp, struct ucred *cred,
1281 int flags)
1282{
1283 struct hammer_transaction trans;
1284 struct namecache *ncp;
1285 hammer_inode_t dip;
1286 hammer_inode_t ip;
1287 hammer_record_ondisk_t rec;
1288 struct hammer_cursor cursor;
8cd0a023
MD
1289 int64_t namekey;
1290 int error;
1291
1292 /*
1293 * Calculate the namekey and setup the key range for the scan. This
1294 * works kinda like a chained hash table where the lower 32 bits
1295 * of the namekey synthesize the chain.
1296 *
1297 * The key range is inclusive of both key_beg and key_end.
1298 */
1299 dip = VTOI(dvp);
1300 ncp = nch->ncp;
1301 namekey = hammer_directory_namekey(ncp->nc_name, ncp->nc_nlen);
1302
1303 hammer_init_cursor_ip(&cursor, dip);
1304 cursor.key_beg.obj_id = dip->obj_id;
1305 cursor.key_beg.key = namekey;
1306 cursor.key_beg.create_tid = dip->obj_asof;
1307 cursor.key_beg.delete_tid = 0;
1308 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1309 cursor.key_beg.obj_type = 0;
1310
1311 cursor.key_end = cursor.key_beg;
1312 cursor.key_end.key |= 0xFFFFFFFFULL;
1313
1314 hammer_start_transaction(&trans, dip->hmp);
1315
1316 /*
1317 * Scan all matching records (the chain), locate the one matching
1318 * the requested path component. info->last_error contains the
1319 * error code on search termination and could be 0, ENOENT, or
1320 * something else.
1321 *
1322 * The hammer_ip_*() functions merge in-memory records with on-disk
1323 * records for the purposes of the search.
1324 */
a89aec1b
MD
1325 error = hammer_ip_first(&cursor, dip);
1326 while (error == 0) {
1327 error = hammer_ip_resolve_data(&cursor);
1328 if (error)
66325755 1329 break;
a89aec1b 1330 rec = cursor.record;
8cd0a023
MD
1331 if (ncp->nc_nlen == rec->entry.base.data_len &&
1332 bcmp(ncp->nc_name, cursor.data, ncp->nc_nlen) == 0) {
66325755
MD
1333 break;
1334 }
a89aec1b 1335 error = hammer_ip_next(&cursor);
66325755 1336 }
8cd0a023
MD
1337
1338 /*
1339 * If all is ok we have to get the inode so we can adjust nlinks.
1340 */
66325755 1341 if (error == 0) {
8cd0a023
MD
1342 ip = hammer_get_inode(dip->hmp, rec->entry.obj_id, &error);
1343 if (error == 0)
a89aec1b 1344 error = hammer_ip_del_directory(&trans, &cursor, dip, ip);
8cd0a023
MD
1345 if (error == 0) {
1346 cache_setunresolved(nch);
1347 cache_setvp(nch, NULL);
1348 /* XXX locking */
1349 if (ip->vp)
1350 cache_inval_vp(ip->vp, CINV_DESTROY);
1351 }
a89aec1b 1352 hammer_rel_inode(ip, 0);
8cd0a023 1353
66325755 1354 if (error == 0) {
8cd0a023
MD
1355 hammer_commit_transaction(&trans);
1356 } else {
1357 hammer_abort_transaction(&trans);
66325755 1358 }
66325755 1359 }
8cd0a023 1360 hammer_done_cursor(&cursor);
66325755 1361 return (error);
66325755
MD
1362}
1363