xref: /dragonfly/sys/vfs/hammer2/hammer2_vnops.c (revision cae2835b)
1 /*
2  * Copyright (c) 2011-2015 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Matthew Dillon <dillon@dragonflybsd.org>
6  * by Venkatesh Srinivas <vsrinivas@dragonflybsd.org>
7  * by Daniel Flores (GSOC 2013 - mentored by Matthew Dillon, compression)
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  *
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in
17  *    the documentation and/or other materials provided with the
18  *    distribution.
19  * 3. Neither the name of The DragonFly Project nor the names of its
20  *    contributors may be used to endorse or promote products derived
21  *    from this software without specific, prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
26  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
27  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
28  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
29  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
30  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
31  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
32  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
33  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  */
36 /*
37  * Kernel Filesystem interface
38  *
39  * NOTE! local ipdata pointers must be reloaded on any modifying operation
40  *	 to the inode as its underlying chain may have changed.
41  */
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/kernel.h>
46 #include <sys/fcntl.h>
47 #include <sys/buf.h>
48 #include <sys/proc.h>
49 #include <sys/namei.h>
50 #include <sys/mount.h>
51 #include <sys/vnode.h>
52 #include <sys/mountctl.h>
53 #include <sys/dirent.h>
54 #include <sys/uio.h>
55 #include <sys/objcache.h>
56 #include <sys/event.h>
57 #include <sys/file.h>
58 #include <vfs/fifofs/fifo.h>
59 
60 #include "hammer2.h"
61 
62 static int hammer2_read_file(hammer2_inode_t *ip, struct uio *uio,
63 				int seqcount);
64 static int hammer2_write_file(hammer2_inode_t *ip, struct uio *uio,
65 				int ioflag, int seqcount);
66 static void hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize);
67 static void hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize);
68 
69 struct objcache *cache_xops;
70 
71 static __inline
72 void
73 hammer2_knote(struct vnode *vp, int flags)
74 {
75 	if (flags)
76 		KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
77 }
78 
79 /*
80  * Last reference to a vnode is going away but it is still cached.
81  */
82 static
83 int
84 hammer2_vop_inactive(struct vop_inactive_args *ap)
85 {
86 	hammer2_inode_t *ip;
87 	struct vnode *vp;
88 
89 	LOCKSTART;
90 	vp = ap->a_vp;
91 	ip = VTOI(vp);
92 
93 	/*
94 	 * Degenerate case
95 	 */
96 	if (ip == NULL) {
97 		vrecycle(vp);
98 		LOCKSTOP;
99 		return (0);
100 	}
101 
102 	/*
103 	 * Check for deleted inodes and recycle immediately on the last
104 	 * release.  Be sure to destroy any left-over buffer cache buffers
105 	 * so we do not waste time trying to flush them.
106 	 *
107 	 * Note that deleting the file block chains under the inode chain
108 	 * would just be a waste of energy, so don't do it.
109 	 *
110 	 * WARNING: nvtruncbuf() can only be safely called without the inode
111 	 *	    lock held due to the way our write thread works.
112 	 */
113 	if (ip->flags & HAMMER2_INODE_ISUNLINKED) {
114 		hammer2_key_t lbase;
115 		int nblksize;
116 
117 		/*
118 		 * Detect updates to the embedded data which may be
119 		 * synchronized by the strategy code.  Simply mark the
120 		 * inode modified so it gets picked up by our normal flush.
121 		 */
122 		nblksize = hammer2_calc_logical(ip, 0, &lbase, NULL);
123 		nvtruncbuf(vp, 0, nblksize, 0, 0);
124 		vrecycle(vp);
125 	}
126 	LOCKSTOP;
127 	return (0);
128 }
129 
130 /*
131  * Reclaim a vnode so that it can be reused; after the inode is
132  * disassociated, the filesystem must manage it alone.
133  */
134 static
135 int
136 hammer2_vop_reclaim(struct vop_reclaim_args *ap)
137 {
138 	hammer2_inode_t *ip;
139 	hammer2_pfs_t *pmp;
140 	struct vnode *vp;
141 
142 	LOCKSTART;
143 	vp = ap->a_vp;
144 	ip = VTOI(vp);
145 	if (ip == NULL) {
146 		LOCKSTOP;
147 		return(0);
148 	}
149 	pmp = ip->pmp;
150 
151 	/*
152 	 * The final close of a deleted file or directory marks it for
153 	 * destruction.  The DELETED flag allows the flusher to shortcut
154 	 * any modified blocks still unflushed (that is, just ignore them).
155 	 *
156 	 * HAMMER2 usually does not try to optimize the freemap by returning
157 	 * deleted blocks to it as it does not usually know how many snapshots
158 	 * might be referencing portions of the file/dir.
159 	 */
160 	vp->v_data = NULL;
161 	ip->vp = NULL;
162 
163 	/*
164 	 * NOTE! We do not attempt to flush chains here, flushing is
165 	 *	 really fragile and could also deadlock.
166 	 */
167 	vclrisdirty(vp);
168 
169 	/*
170 	 * This occurs if the inode was unlinked while open.  Reclamation of
171 	 * these inodes requires processing we cannot safely do here so add
172 	 * the inode to the sideq in that situation.
173 	 *
174 	 * A modified inode may require chain synchronization which will no
175 	 * longer be driven by a sync or fsync without the vnode, also use
176 	 * the sideq for that.
177 	 *
178 	 * A reclaim can occur at any time so we cannot safely start a
179 	 * transaction to handle reclamation of unlinked files.  Instead,
180 	 * the ip is left with a reference and placed on a linked list and
181 	 * handled later on.
182 	 */
183 
184 	if ((ip->flags & (HAMMER2_INODE_ISUNLINKED |
185 			  HAMMER2_INODE_MODIFIED |
186 			  HAMMER2_INODE_RESIZED)) &&
187 	    (ip->flags & HAMMER2_INODE_ISDELETED) == 0) {
188 		hammer2_inode_sideq_t *ipul;
189 
190 		ipul = kmalloc(sizeof(*ipul), pmp->minode, M_WAITOK | M_ZERO);
191 		ipul->ip = ip;
192 
193 		hammer2_spin_ex(&pmp->list_spin);
194 		if ((ip->flags & HAMMER2_INODE_ONSIDEQ) == 0) {
195 			/* ref -> sideq */
196 			atomic_set_int(&ip->flags, HAMMER2_INODE_ONSIDEQ);
197 			TAILQ_INSERT_TAIL(&pmp->sideq, ipul, entry);
198 			hammer2_spin_unex(&pmp->list_spin);
199 		} else {
200 			hammer2_spin_unex(&pmp->list_spin);
201 			kfree(ipul, pmp->minode);
202 			hammer2_inode_drop(ip);		/* vp ref */
203 		}
204 		/* retain ref from vp for ipul */
205 	} else {
206 		hammer2_inode_drop(ip);			/* vp ref */
207 	}
208 
209 	/*
210 	 * XXX handle background sync when ip dirty, kernel will no longer
211 	 * notify us regarding this inode because there is no longer a
212 	 * vnode attached to it.
213 	 */
214 
215 	LOCKSTOP;
216 	return (0);
217 }
218 
219 static
220 int
221 hammer2_vop_fsync(struct vop_fsync_args *ap)
222 {
223 	hammer2_inode_t *ip;
224 	struct vnode *vp;
225 
226 	LOCKSTART;
227 	vp = ap->a_vp;
228 	ip = VTOI(vp);
229 
230 #if 0
231 	/* XXX can't do this yet */
232 	hammer2_trans_init(ip->pmp, HAMMER2_TRANS_ISFLUSH);
233 	vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
234 #endif
235 	hammer2_trans_init(ip->pmp, 0);
236 	vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
237 
238 	/*
239 	 * Calling chain_flush here creates a lot of duplicative
240 	 * COW operations due to non-optimal vnode ordering.
241 	 *
242 	 * Only do it for an actual fsync() syscall.  The other forms
243 	 * which call this function will eventually call chain_flush
244 	 * on the volume root as a catch-all, which is far more optimal.
245 	 */
246 	hammer2_inode_lock(ip, 0);
247 	if (ip->flags & HAMMER2_INODE_MODIFIED)
248 		hammer2_inode_chain_sync(ip);
249 	hammer2_inode_unlock(ip);
250 	hammer2_trans_done(ip->pmp);
251 
252 	LOCKSTOP;
253 	return (0);
254 }
255 
256 static
257 int
258 hammer2_vop_access(struct vop_access_args *ap)
259 {
260 	hammer2_inode_t *ip = VTOI(ap->a_vp);
261 	uid_t uid;
262 	gid_t gid;
263 	int error;
264 
265 	LOCKSTART;
266 	hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
267 	uid = hammer2_to_unix_xid(&ip->meta.uid);
268 	gid = hammer2_to_unix_xid(&ip->meta.gid);
269 	error = vop_helper_access(ap, uid, gid, ip->meta.mode, ip->meta.uflags);
270 	hammer2_inode_unlock(ip);
271 
272 	LOCKSTOP;
273 	return (error);
274 }
275 
276 static
277 int
278 hammer2_vop_getattr(struct vop_getattr_args *ap)
279 {
280 	hammer2_pfs_t *pmp;
281 	hammer2_inode_t *ip;
282 	struct vnode *vp;
283 	struct vattr *vap;
284 	hammer2_chain_t *chain;
285 	int i;
286 
287 	LOCKSTART;
288 	vp = ap->a_vp;
289 	vap = ap->a_vap;
290 
291 	ip = VTOI(vp);
292 	pmp = ip->pmp;
293 
294 	hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
295 
296 	vap->va_fsid = pmp->mp->mnt_stat.f_fsid.val[0];
297 	vap->va_fileid = ip->meta.inum;
298 	vap->va_mode = ip->meta.mode;
299 	vap->va_nlink = ip->meta.nlinks;
300 	vap->va_uid = hammer2_to_unix_xid(&ip->meta.uid);
301 	vap->va_gid = hammer2_to_unix_xid(&ip->meta.gid);
302 	vap->va_rmajor = 0;
303 	vap->va_rminor = 0;
304 	vap->va_size = ip->meta.size;	/* protected by shared lock */
305 	vap->va_blocksize = HAMMER2_PBUFSIZE;
306 	vap->va_flags = ip->meta.uflags;
307 	hammer2_time_to_timespec(ip->meta.ctime, &vap->va_ctime);
308 	hammer2_time_to_timespec(ip->meta.mtime, &vap->va_mtime);
309 	hammer2_time_to_timespec(ip->meta.mtime, &vap->va_atime);
310 	vap->va_gen = 1;
311 	vap->va_bytes = 0;
312 	if (ip->meta.type == HAMMER2_OBJTYPE_DIRECTORY) {
313 		/*
314 		 * Can't really calculate directory use sans the files under
315 		 * it, just assume one block for now.
316 		 */
317 		vap->va_bytes += HAMMER2_INODE_BYTES;
318 	} else {
319 		for (i = 0; i < ip->cluster.nchains; ++i) {
320 			if ((chain = ip->cluster.array[i].chain) != NULL) {
321 				if (vap->va_bytes < chain->bref.data_count)
322 					vap->va_bytes = chain->bref.data_count;
323 			}
324 		}
325 	}
326 	vap->va_type = hammer2_get_vtype(ip->meta.type);
327 	vap->va_filerev = 0;
328 	vap->va_uid_uuid = ip->meta.uid;
329 	vap->va_gid_uuid = ip->meta.gid;
330 	vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
331 			  VA_FSID_UUID_VALID;
332 
333 	hammer2_inode_unlock(ip);
334 
335 	LOCKSTOP;
336 	return (0);
337 }
338 
339 static
340 int
341 hammer2_vop_setattr(struct vop_setattr_args *ap)
342 {
343 	hammer2_inode_t *ip;
344 	struct vnode *vp;
345 	struct vattr *vap;
346 	int error;
347 	int kflags = 0;
348 	uint64_t ctime;
349 
350 	LOCKSTART;
351 	vp = ap->a_vp;
352 	vap = ap->a_vap;
353 	hammer2_update_time(&ctime);
354 
355 	ip = VTOI(vp);
356 
357 	if (ip->pmp->ronly) {
358 		LOCKSTOP;
359 		return(EROFS);
360 	}
361 
362 	hammer2_pfs_memory_wait(ip->pmp);
363 	hammer2_trans_init(ip->pmp, 0);
364 	hammer2_inode_lock(ip, 0);
365 	error = 0;
366 
367 	if (vap->va_flags != VNOVAL) {
368 		uint32_t flags;
369 
370 		flags = ip->meta.uflags;
371 		error = vop_helper_setattr_flags(&flags, vap->va_flags,
372 				     hammer2_to_unix_xid(&ip->meta.uid),
373 				     ap->a_cred);
374 		if (error == 0) {
375 			if (ip->meta.uflags != flags) {
376 				hammer2_inode_modify(ip);
377 				ip->meta.uflags = flags;
378 				ip->meta.ctime = ctime;
379 				kflags |= NOTE_ATTRIB;
380 			}
381 			if (ip->meta.uflags & (IMMUTABLE | APPEND)) {
382 				error = 0;
383 				goto done;
384 			}
385 		}
386 		goto done;
387 	}
388 	if (ip->meta.uflags & (IMMUTABLE | APPEND)) {
389 		error = EPERM;
390 		goto done;
391 	}
392 	if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
393 		mode_t cur_mode = ip->meta.mode;
394 		uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid);
395 		gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid);
396 		uuid_t uuid_uid;
397 		uuid_t uuid_gid;
398 
399 		error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
400 					 ap->a_cred,
401 					 &cur_uid, &cur_gid, &cur_mode);
402 		if (error == 0) {
403 			hammer2_guid_to_uuid(&uuid_uid, cur_uid);
404 			hammer2_guid_to_uuid(&uuid_gid, cur_gid);
405 			if (bcmp(&uuid_uid, &ip->meta.uid, sizeof(uuid_uid)) ||
406 			    bcmp(&uuid_gid, &ip->meta.gid, sizeof(uuid_gid)) ||
407 			    ip->meta.mode != cur_mode
408 			) {
409 				hammer2_inode_modify(ip);
410 				ip->meta.uid = uuid_uid;
411 				ip->meta.gid = uuid_gid;
412 				ip->meta.mode = cur_mode;
413 				ip->meta.ctime = ctime;
414 			}
415 			kflags |= NOTE_ATTRIB;
416 		}
417 	}
418 
419 	/*
420 	 * Resize the file
421 	 */
422 	if (vap->va_size != VNOVAL && ip->meta.size != vap->va_size) {
423 		switch(vp->v_type) {
424 		case VREG:
425 			if (vap->va_size == ip->meta.size)
426 				break;
427 			if (vap->va_size < ip->meta.size) {
428 				hammer2_mtx_ex(&ip->truncate_lock);
429 				hammer2_truncate_file(ip, vap->va_size);
430 				hammer2_mtx_unlock(&ip->truncate_lock);
431 			} else {
432 				hammer2_extend_file(ip, vap->va_size);
433 			}
434 			hammer2_inode_modify(ip);
435 			ip->meta.mtime = ctime;
436 			break;
437 		default:
438 			error = EINVAL;
439 			goto done;
440 		}
441 	}
442 #if 0
443 	/* atime not supported */
444 	if (vap->va_atime.tv_sec != VNOVAL) {
445 		hammer2_inode_modify(ip);
446 		ip->meta.atime = hammer2_timespec_to_time(&vap->va_atime);
447 		kflags |= NOTE_ATTRIB;
448 	}
449 #endif
450 	if (vap->va_mode != (mode_t)VNOVAL) {
451 		mode_t cur_mode = ip->meta.mode;
452 		uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid);
453 		gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid);
454 
455 		error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
456 					 cur_uid, cur_gid, &cur_mode);
457 		if (error == 0 && ip->meta.mode != cur_mode) {
458 			hammer2_inode_modify(ip);
459 			ip->meta.mode = cur_mode;
460 			ip->meta.ctime = ctime;
461 			kflags |= NOTE_ATTRIB;
462 		}
463 	}
464 
465 	if (vap->va_mtime.tv_sec != VNOVAL) {
466 		hammer2_inode_modify(ip);
467 		ip->meta.mtime = hammer2_timespec_to_time(&vap->va_mtime);
468 		kflags |= NOTE_ATTRIB;
469 	}
470 
471 done:
472 	/*
473 	 * If a truncation occurred we must call inode_fsync() now in order
474 	 * to trim the related data chains, otherwise a later expansion can
475 	 * cause havoc.
476 	 *
477 	 * If an extend occured that changed the DIRECTDATA state, we must
478 	 * call inode_fsync now in order to prepare the inode's indirect
479 	 * block table.
480 	 */
481 	if (ip->flags & HAMMER2_INODE_RESIZED)
482 		hammer2_inode_chain_sync(ip);
483 
484 	/*
485 	 * Cleanup.
486 	 */
487 	hammer2_inode_unlock(ip);
488 	hammer2_trans_done(ip->pmp);
489 	hammer2_knote(ip->vp, kflags);
490 
491 	LOCKSTOP;
492 	return (error);
493 }
494 
495 static
496 int
497 hammer2_vop_readdir(struct vop_readdir_args *ap)
498 {
499 	hammer2_xop_readdir_t *xop;
500 	hammer2_blockref_t bref;
501 	hammer2_inode_t *ip;
502 	hammer2_tid_t inum;
503 	hammer2_key_t lkey;
504 	struct uio *uio;
505 	off_t *cookies;
506 	off_t saveoff;
507 	int cookie_index;
508 	int ncookies;
509 	int error;
510 	int eofflag;
511 	int dtype;
512 	int r;
513 
514 	LOCKSTART;
515 	ip = VTOI(ap->a_vp);
516 	uio = ap->a_uio;
517 	saveoff = uio->uio_offset;
518 	eofflag = 0;
519 	error = 0;
520 
521 	/*
522 	 * Setup cookies directory entry cookies if requested
523 	 */
524 	if (ap->a_ncookies) {
525 		ncookies = uio->uio_resid / 16 + 1;
526 		if (ncookies > 1024)
527 			ncookies = 1024;
528 		cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
529 	} else {
530 		ncookies = -1;
531 		cookies = NULL;
532 	}
533 	cookie_index = 0;
534 
535 	hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
536 
537 	/*
538 	 * Handle artificial entries.  To ensure that only positive 64 bit
539 	 * quantities are returned to userland we always strip off bit 63.
540 	 * The hash code is designed such that codes 0x0000-0x7FFF are not
541 	 * used, allowing us to use these codes for articial entries.
542 	 *
543 	 * Entry 0 is used for '.' and entry 1 is used for '..'.  Do not
544 	 * allow '..' to cross the mount point into (e.g.) the super-root.
545 	 */
546 	if (saveoff == 0) {
547 		inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK;
548 		r = vop_write_dirent(&error, uio, inum, DT_DIR, 1, ".");
549 		if (r)
550 			goto done;
551 		if (cookies)
552 			cookies[cookie_index] = saveoff;
553 		++saveoff;
554 		++cookie_index;
555 		if (cookie_index == ncookies)
556 			goto done;
557 	}
558 
559 	if (saveoff == 1) {
560 		/*
561 		 * Be careful with lockorder when accessing ".."
562 		 *
563 		 * (ip is the current dir. xip is the parent dir).
564 		 */
565 		inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK;
566 		if (ip != ip->pmp->iroot)
567 			inum = ip->meta.iparent & HAMMER2_DIRHASH_USERMSK;
568 		r = vop_write_dirent(&error, uio, inum, DT_DIR, 2, "..");
569 		if (r)
570 			goto done;
571 		if (cookies)
572 			cookies[cookie_index] = saveoff;
573 		++saveoff;
574 		++cookie_index;
575 		if (cookie_index == ncookies)
576 			goto done;
577 	}
578 
579 	lkey = saveoff | HAMMER2_DIRHASH_VISIBLE;
580 	if (hammer2_debug & 0x0020)
581 		kprintf("readdir: lkey %016jx\n", lkey);
582 	if (error)
583 		goto done;
584 
585 	/*
586 	 * Use XOP for cluster scan.
587 	 *
588 	 * parent is the inode cluster, already locked for us.  Don't
589 	 * double lock shared locks as this will screw up upgrades.
590 	 */
591 	xop = hammer2_xop_alloc(ip, 0);
592 	xop->lkey = lkey;
593 	hammer2_xop_start(&xop->head, hammer2_xop_readdir);
594 
595 	for (;;) {
596 		const hammer2_inode_data_t *ripdata;
597 
598 		error = hammer2_xop_collect(&xop->head, 0);
599 		if (error)
600 			break;
601 		if (cookie_index == ncookies)
602 			break;
603 		if (hammer2_debug & 0x0020)
604 		kprintf("cluster chain %p %p\n",
605 			xop->head.cluster.focus,
606 			(xop->head.cluster.focus ?
607 			 xop->head.cluster.focus->data : (void *)-1));
608 		ripdata = &hammer2_cluster_rdata(&xop->head.cluster)->ipdata;
609 		hammer2_cluster_bref(&xop->head.cluster, &bref);
610 		if (bref.type == HAMMER2_BREF_TYPE_INODE) {
611 			dtype = hammer2_get_dtype(ripdata);
612 			saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
613 			r = vop_write_dirent(&error, uio,
614 					     ripdata->meta.inum &
615 					      HAMMER2_DIRHASH_USERMSK,
616 					     dtype,
617 					     ripdata->meta.name_len,
618 					     ripdata->filename);
619 			if (r)
620 				break;
621 			if (cookies)
622 				cookies[cookie_index] = saveoff;
623 			++cookie_index;
624 		} else {
625 			/* XXX chain error */
626 			kprintf("bad chain type readdir %d\n", bref.type);
627 		}
628 	}
629 	hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
630 	if (error == ENOENT) {
631 		error = 0;
632 		eofflag = 1;
633 		saveoff = (hammer2_key_t)-1;
634 	} else {
635 		saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
636 	}
637 done:
638 	hammer2_inode_unlock(ip);
639 	if (ap->a_eofflag)
640 		*ap->a_eofflag = eofflag;
641 	if (hammer2_debug & 0x0020)
642 		kprintf("readdir: done at %016jx\n", saveoff);
643 	uio->uio_offset = saveoff & ~HAMMER2_DIRHASH_VISIBLE;
644 	if (error && cookie_index == 0) {
645 		if (cookies) {
646 			kfree(cookies, M_TEMP);
647 			*ap->a_ncookies = 0;
648 			*ap->a_cookies = NULL;
649 		}
650 	} else {
651 		if (cookies) {
652 			*ap->a_ncookies = cookie_index;
653 			*ap->a_cookies = cookies;
654 		}
655 	}
656 	LOCKSTOP;
657 	return (error);
658 }
659 
660 /*
661  * hammer2_vop_readlink { vp, uio, cred }
662  */
663 static
664 int
665 hammer2_vop_readlink(struct vop_readlink_args *ap)
666 {
667 	struct vnode *vp;
668 	hammer2_inode_t *ip;
669 	int error;
670 
671 	vp = ap->a_vp;
672 	if (vp->v_type != VLNK)
673 		return (EINVAL);
674 	ip = VTOI(vp);
675 
676 	error = hammer2_read_file(ip, ap->a_uio, 0);
677 	return (error);
678 }
679 
680 static
681 int
682 hammer2_vop_read(struct vop_read_args *ap)
683 {
684 	struct vnode *vp;
685 	hammer2_inode_t *ip;
686 	struct uio *uio;
687 	int error;
688 	int seqcount;
689 	int bigread;
690 
691 	/*
692 	 * Read operations supported on this vnode?
693 	 */
694 	vp = ap->a_vp;
695 	if (vp->v_type != VREG)
696 		return (EINVAL);
697 
698 	/*
699 	 * Misc
700 	 */
701 	ip = VTOI(vp);
702 	uio = ap->a_uio;
703 	error = 0;
704 
705 	seqcount = ap->a_ioflag >> 16;
706 	bigread = (uio->uio_resid > 100 * 1024 * 1024);
707 
708 	error = hammer2_read_file(ip, uio, seqcount);
709 	return (error);
710 }
711 
712 static
713 int
714 hammer2_vop_write(struct vop_write_args *ap)
715 {
716 	hammer2_inode_t *ip;
717 	thread_t td;
718 	struct vnode *vp;
719 	struct uio *uio;
720 	int error;
721 	int seqcount;
722 
723 	/*
724 	 * Read operations supported on this vnode?
725 	 */
726 	vp = ap->a_vp;
727 	if (vp->v_type != VREG)
728 		return (EINVAL);
729 
730 	/*
731 	 * Misc
732 	 */
733 	ip = VTOI(vp);
734 	uio = ap->a_uio;
735 	error = 0;
736 	if (ip->pmp->ronly) {
737 		return (EROFS);
738 	}
739 
740 	seqcount = ap->a_ioflag >> 16;
741 
742 	/*
743 	 * Check resource limit
744 	 */
745 	if (uio->uio_resid > 0 && (td = uio->uio_td) != NULL && td->td_proc &&
746 	    uio->uio_offset + uio->uio_resid >
747 	     td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
748 		lwpsignal(td->td_proc, td->td_lwp, SIGXFSZ);
749 		return (EFBIG);
750 	}
751 
752 	/*
753 	 * The transaction interlocks against flushes initiations
754 	 * (note: but will run concurrently with the actual flush).
755 	 */
756 	hammer2_trans_init(ip->pmp, 0);
757 	error = hammer2_write_file(ip, uio, ap->a_ioflag, seqcount);
758 	hammer2_trans_done(ip->pmp);
759 
760 	return (error);
761 }
762 
763 /*
764  * Perform read operations on a file or symlink given an UNLOCKED
765  * inode and uio.
766  *
767  * The passed ip is not locked.
768  */
769 static
770 int
771 hammer2_read_file(hammer2_inode_t *ip, struct uio *uio, int seqcount)
772 {
773 	hammer2_off_t size;
774 	struct buf *bp;
775 	int error;
776 
777 	error = 0;
778 
779 	/*
780 	 * UIO read loop.
781 	 *
782 	 * WARNING! Assumes that the kernel interlocks size changes at the
783 	 *	    vnode level.
784 	 */
785 	hammer2_mtx_sh(&ip->lock);
786 	hammer2_mtx_sh(&ip->truncate_lock);
787 	size = ip->meta.size;
788 	hammer2_mtx_unlock(&ip->lock);
789 
790 	while (uio->uio_resid > 0 && uio->uio_offset < size) {
791 		hammer2_key_t lbase;
792 		hammer2_key_t leof;
793 		int lblksize;
794 		int loff;
795 		int n;
796 
797 		lblksize = hammer2_calc_logical(ip, uio->uio_offset,
798 						&lbase, &leof);
799 
800 #if 1
801 		error = cluster_read(ip->vp, leof, lbase, lblksize,
802 				     uio->uio_resid, seqcount * MAXBSIZE,
803 				     &bp);
804 #else
805 		if (uio->uio_segflg == UIO_NOCOPY) {
806 			bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
807 			if (bp->b_flags & B_CACHE) {
808 				int i;
809 				int j = 0;
810 				if (bp->b_xio.xio_npages != 16)
811 					kprintf("NPAGES BAD\n");
812 				for (i = 0; i < bp->b_xio.xio_npages; ++i) {
813 					vm_page_t m;
814 					m = bp->b_xio.xio_pages[i];
815 					if (m == NULL || m->valid == 0) {
816 						kprintf("bp %016jx %016jx pg %d inv",
817 							lbase, leof, i);
818 						if (m)
819 							kprintf("m->object %p/%p", m->object, ip->vp->v_object);
820 						kprintf("\n");
821 						j = 1;
822 					}
823 				}
824 				if (j)
825 					kprintf("b_flags %08x, b_error %d\n", bp->b_flags, bp->b_error);
826 			}
827 			bqrelse(bp);
828 		}
829 		error = bread(ip->vp, lbase, lblksize, &bp);
830 #endif
831 		if (error) {
832 			brelse(bp);
833 			break;
834 		}
835 		loff = (int)(uio->uio_offset - lbase);
836 		n = lblksize - loff;
837 		if (n > uio->uio_resid)
838 			n = uio->uio_resid;
839 		if (n > size - uio->uio_offset)
840 			n = (int)(size - uio->uio_offset);
841 		bp->b_flags |= B_AGE;
842 		uiomovebp(bp, (char *)bp->b_data + loff, n, uio);
843 		bqrelse(bp);
844 	}
845 	hammer2_mtx_unlock(&ip->truncate_lock);
846 
847 	return (error);
848 }
849 
850 /*
851  * Write to the file represented by the inode via the logical buffer cache.
852  * The inode may represent a regular file or a symlink.
853  *
854  * The inode must not be locked.
855  */
856 static
857 int
858 hammer2_write_file(hammer2_inode_t *ip, struct uio *uio,
859 		   int ioflag, int seqcount)
860 {
861 	hammer2_key_t old_eof;
862 	hammer2_key_t new_eof;
863 	struct buf *bp;
864 	int kflags;
865 	int error;
866 	int modified;
867 
868 	/*
869 	 * Setup if append
870 	 *
871 	 * WARNING! Assumes that the kernel interlocks size changes at the
872 	 *	    vnode level.
873 	 */
874 	hammer2_mtx_ex(&ip->lock);
875 	hammer2_mtx_sh(&ip->truncate_lock);
876 	if (ioflag & IO_APPEND)
877 		uio->uio_offset = ip->meta.size;
878 	old_eof = ip->meta.size;
879 
880 	/*
881 	 * Extend the file if necessary.  If the write fails at some point
882 	 * we will truncate it back down to cover as much as we were able
883 	 * to write.
884 	 *
885 	 * Doing this now makes it easier to calculate buffer sizes in
886 	 * the loop.
887 	 */
888 	kflags = 0;
889 	error = 0;
890 	modified = 0;
891 
892 	if (uio->uio_offset + uio->uio_resid > old_eof) {
893 		new_eof = uio->uio_offset + uio->uio_resid;
894 		modified = 1;
895 		hammer2_extend_file(ip, new_eof);
896 		kflags |= NOTE_EXTEND;
897 	} else {
898 		new_eof = old_eof;
899 	}
900 	hammer2_mtx_unlock(&ip->lock);
901 
902 	/*
903 	 * UIO write loop
904 	 */
905 	while (uio->uio_resid > 0) {
906 		hammer2_key_t lbase;
907 		int trivial;
908 		int endofblk;
909 		int lblksize;
910 		int loff;
911 		int n;
912 
913 		/*
914 		 * Don't allow the buffer build to blow out the buffer
915 		 * cache.
916 		 */
917 		if ((ioflag & IO_RECURSE) == 0)
918 			bwillwrite(HAMMER2_PBUFSIZE);
919 
920 		/*
921 		 * This nominally tells us how much we can cluster and
922 		 * what the logical buffer size needs to be.  Currently
923 		 * we don't try to cluster the write and just handle one
924 		 * block at a time.
925 		 */
926 		lblksize = hammer2_calc_logical(ip, uio->uio_offset,
927 						&lbase, NULL);
928 		loff = (int)(uio->uio_offset - lbase);
929 
930 		KKASSERT(lblksize <= 65536);
931 
932 		/*
933 		 * Calculate bytes to copy this transfer and whether the
934 		 * copy completely covers the buffer or not.
935 		 */
936 		trivial = 0;
937 		n = lblksize - loff;
938 		if (n > uio->uio_resid) {
939 			n = uio->uio_resid;
940 			if (loff == lbase && uio->uio_offset + n == new_eof)
941 				trivial = 1;
942 			endofblk = 0;
943 		} else {
944 			if (loff == 0)
945 				trivial = 1;
946 			endofblk = 1;
947 		}
948 		if (lbase >= new_eof)
949 			trivial = 1;
950 
951 		/*
952 		 * Get the buffer
953 		 */
954 		if (uio->uio_segflg == UIO_NOCOPY) {
955 			/*
956 			 * Issuing a write with the same data backing the
957 			 * buffer.  Instantiate the buffer to collect the
958 			 * backing vm pages, then read-in any missing bits.
959 			 *
960 			 * This case is used by vop_stdputpages().
961 			 */
962 			bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
963 			if ((bp->b_flags & B_CACHE) == 0) {
964 				bqrelse(bp);
965 				error = bread(ip->vp, lbase, lblksize, &bp);
966 			}
967 		} else if (trivial) {
968 			/*
969 			 * Even though we are entirely overwriting the buffer
970 			 * we may still have to zero it out to avoid a
971 			 * mmap/write visibility issue.
972 			 */
973 			bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
974 			if ((bp->b_flags & B_CACHE) == 0)
975 				vfs_bio_clrbuf(bp);
976 		} else {
977 			/*
978 			 * Partial overwrite, read in any missing bits then
979 			 * replace the portion being written.
980 			 *
981 			 * (The strategy code will detect zero-fill physical
982 			 * blocks for this case).
983 			 */
984 			error = bread(ip->vp, lbase, lblksize, &bp);
985 			if (error == 0)
986 				bheavy(bp);
987 		}
988 
989 		if (error) {
990 			brelse(bp);
991 			break;
992 		}
993 
994 		/*
995 		 * Ok, copy the data in
996 		 */
997 		error = uiomovebp(bp, bp->b_data + loff, n, uio);
998 		kflags |= NOTE_WRITE;
999 		modified = 1;
1000 		if (error) {
1001 			brelse(bp);
1002 			break;
1003 		}
1004 
1005 		/*
1006 		 * WARNING: Pageout daemon will issue UIO_NOCOPY writes
1007 		 *	    with IO_SYNC or IO_ASYNC set.  These writes
1008 		 *	    must be handled as the pageout daemon expects.
1009 		 *
1010 		 * NOTE!    H2 relies on cluster_write() here because it
1011 		 *	    cannot preallocate disk blocks at the logical
1012 		 *	    level due to not knowing what the compression
1013 		 *	    size will be at this time.
1014 		 *
1015 		 *	    We must use cluster_write() here and we depend
1016 		 *	    on the write-behind feature to flush buffers
1017 		 *	    appropriately.  If we let the buffer daemons do
1018 		 *	    it the block allocations will be all over the
1019 		 *	    map.
1020 		 */
1021 		if (ioflag & IO_SYNC) {
1022 			bwrite(bp);
1023 		} else if ((ioflag & IO_DIRECT) && endofblk) {
1024 			bawrite(bp);
1025 		} else if (ioflag & IO_ASYNC) {
1026 			bawrite(bp);
1027 		} else if (ip->vp->v_mount->mnt_flag & MNT_NOCLUSTERW) {
1028 			bdwrite(bp);
1029 		} else {
1030 #if 1
1031 			bp->b_flags |= B_CLUSTEROK;
1032 			cluster_write(bp, new_eof, lblksize, seqcount);
1033 #else
1034 			bp->b_flags |= B_CLUSTEROK;
1035 			bdwrite(bp);
1036 #endif
1037 		}
1038 	}
1039 
1040 	/*
1041 	 * Cleanup.  If we extended the file EOF but failed to write through
1042 	 * the entire write is a failure and we have to back-up.
1043 	 */
1044 	if (error && new_eof != old_eof) {
1045 		hammer2_mtx_unlock(&ip->truncate_lock);
1046 		hammer2_mtx_ex(&ip->lock);
1047 		hammer2_mtx_ex(&ip->truncate_lock);
1048 		hammer2_truncate_file(ip, old_eof);
1049 		if (ip->flags & HAMMER2_INODE_MODIFIED)
1050 			hammer2_inode_chain_sync(ip);
1051 		hammer2_mtx_unlock(&ip->lock);
1052 	} else if (modified) {
1053 		hammer2_mtx_ex(&ip->lock);
1054 		hammer2_inode_modify(ip);
1055 		hammer2_update_time(&ip->meta.mtime);
1056 		if (ip->flags & HAMMER2_INODE_MODIFIED)
1057 			hammer2_inode_chain_sync(ip);
1058 		hammer2_mtx_unlock(&ip->lock);
1059 		hammer2_knote(ip->vp, kflags);
1060 	}
1061 	hammer2_trans_assert_strategy(ip->pmp);
1062 	hammer2_mtx_unlock(&ip->truncate_lock);
1063 
1064 	return error;
1065 }
1066 
1067 /*
1068  * Truncate the size of a file.  The inode must not be locked.
1069  *
1070  * We must unconditionally set HAMMER2_INODE_RESIZED to properly
1071  * ensure that any on-media data beyond the new file EOF has been destroyed.
1072  *
1073  * WARNING: nvtruncbuf() can only be safely called without the inode lock
1074  *	    held due to the way our write thread works.  If the truncation
1075  *	    occurs in the middle of a buffer, nvtruncbuf() is responsible
1076  *	    for dirtying that buffer and zeroing out trailing bytes.
1077  *
1078  * WARNING! Assumes that the kernel interlocks size changes at the
1079  *	    vnode level.
1080  *
1081  * WARNING! Caller assumes responsibility for removing dead blocks
1082  *	    if INODE_RESIZED is set.
1083  */
1084 static
1085 void
1086 hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1087 {
1088 	hammer2_key_t lbase;
1089 	int nblksize;
1090 
1091 	LOCKSTART;
1092 	hammer2_mtx_unlock(&ip->lock);
1093 	if (ip->vp) {
1094 		nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1095 		nvtruncbuf(ip->vp, nsize,
1096 			   nblksize, (int)nsize & (nblksize - 1),
1097 			   0);
1098 	}
1099 	hammer2_mtx_ex(&ip->lock);
1100 	KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0);
1101 	ip->osize = ip->meta.size;
1102 	ip->meta.size = nsize;
1103 	atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1104 	hammer2_inode_modify(ip);
1105 	LOCKSTOP;
1106 }
1107 
1108 /*
1109  * Extend the size of a file.  The inode must not be locked.
1110  *
1111  * Even though the file size is changing, we do not have to set the
1112  * INODE_RESIZED bit unless the file size crosses the EMBEDDED_BYTES
1113  * boundary.  When this occurs a hammer2_inode_chain_sync() is required
1114  * to prepare the inode cluster's indirect block table, otherwise
1115  * async execution of the strategy code will implode on us.
1116  *
1117  * WARNING! Assumes that the kernel interlocks size changes at the
1118  *	    vnode level.
1119  *
1120  * WARNING! Caller assumes responsibility for transitioning out
1121  *	    of the inode DIRECTDATA mode if INODE_RESIZED is set.
1122  */
1123 static
1124 void
1125 hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1126 {
1127 	hammer2_key_t lbase;
1128 	hammer2_key_t osize;
1129 	int oblksize;
1130 	int nblksize;
1131 
1132 	LOCKSTART;
1133 
1134 	KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0);
1135 	hammer2_inode_modify(ip);
1136 	osize = ip->meta.size;
1137 	ip->osize = osize;
1138 	ip->meta.size = nsize;
1139 
1140 	if (osize <= HAMMER2_EMBEDDED_BYTES && nsize > HAMMER2_EMBEDDED_BYTES) {
1141 		atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1142 		hammer2_inode_chain_sync(ip);
1143 	}
1144 
1145 	hammer2_mtx_unlock(&ip->lock);
1146 	if (ip->vp) {
1147 		oblksize = hammer2_calc_logical(ip, osize, &lbase, NULL);
1148 		nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1149 		nvextendbuf(ip->vp,
1150 			    osize, nsize,
1151 			    oblksize, nblksize,
1152 			    -1, -1, 0);
1153 	}
1154 	hammer2_mtx_ex(&ip->lock);
1155 
1156 	LOCKSTOP;
1157 }
1158 
1159 static
1160 int
1161 hammer2_vop_nresolve(struct vop_nresolve_args *ap)
1162 {
1163 	hammer2_xop_nresolve_t *xop;
1164 	hammer2_inode_t *ip;
1165 	hammer2_inode_t *dip;
1166 	struct namecache *ncp;
1167 	struct vnode *vp;
1168 	int error;
1169 
1170 	LOCKSTART;
1171 	dip = VTOI(ap->a_dvp);
1172 	xop = hammer2_xop_alloc(dip, 0);
1173 
1174 	ncp = ap->a_nch->ncp;
1175 	hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1176 
1177 	/*
1178 	 * Note: In DragonFly the kernel handles '.' and '..'.
1179 	 */
1180 	hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1181 	hammer2_xop_start(&xop->head, hammer2_xop_nresolve);
1182 
1183 	error = hammer2_xop_collect(&xop->head, 0);
1184 	if (error) {
1185 		ip = NULL;
1186 	} else {
1187 		ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1188 	}
1189 	hammer2_inode_unlock(dip);
1190 
1191 	/*
1192 	 * Acquire the related vnode
1193 	 *
1194 	 * NOTE: For error processing, only ENOENT resolves the namecache
1195 	 *	 entry to NULL, otherwise we just return the error and
1196 	 *	 leave the namecache unresolved.
1197 	 *
1198 	 * NOTE: multiple hammer2_inode structures can be aliased to the
1199 	 *	 same chain element, for example for hardlinks.  This
1200 	 *	 use case does not 'reattach' inode associations that
1201 	 *	 might already exist, but always allocates a new one.
1202 	 *
1203 	 * WARNING: inode structure is locked exclusively via inode_get
1204 	 *	    but chain was locked shared.  inode_unlock()
1205 	 *	    will handle it properly.
1206 	 */
1207 	if (ip) {
1208 		vp = hammer2_igetv(ip, &error);
1209 		if (error == 0) {
1210 			vn_unlock(vp);
1211 			cache_setvp(ap->a_nch, vp);
1212 		} else if (error == ENOENT) {
1213 			cache_setvp(ap->a_nch, NULL);
1214 		}
1215 		hammer2_inode_unlock(ip);
1216 
1217 		/*
1218 		 * The vp should not be released until after we've disposed
1219 		 * of our locks, because it might cause vop_inactive() to
1220 		 * be called.
1221 		 */
1222 		if (vp)
1223 			vrele(vp);
1224 	} else {
1225 		error = ENOENT;
1226 		cache_setvp(ap->a_nch, NULL);
1227 	}
1228 	hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1229 	KASSERT(error || ap->a_nch->ncp->nc_vp != NULL,
1230 		("resolve error %d/%p ap %p\n",
1231 		 error, ap->a_nch->ncp->nc_vp, ap));
1232 	LOCKSTOP;
1233 
1234 	return error;
1235 }
1236 
1237 static
1238 int
1239 hammer2_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
1240 {
1241 	hammer2_inode_t *dip;
1242 	hammer2_tid_t inum;
1243 	int error;
1244 
1245 	LOCKSTART;
1246 	dip = VTOI(ap->a_dvp);
1247 	inum = dip->meta.iparent;
1248 	*ap->a_vpp = NULL;
1249 
1250 	if (inum) {
1251 		error = hammer2_vfs_vget(ap->a_dvp->v_mount, NULL,
1252 					 inum, ap->a_vpp);
1253 	} else {
1254 		error = ENOENT;
1255 	}
1256 	LOCKSTOP;
1257 	return error;
1258 }
1259 
1260 static
1261 int
1262 hammer2_vop_nmkdir(struct vop_nmkdir_args *ap)
1263 {
1264 	hammer2_inode_t *dip;
1265 	hammer2_inode_t *nip;
1266 	struct namecache *ncp;
1267 	const uint8_t *name;
1268 	size_t name_len;
1269 	hammer2_tid_t inum;
1270 	int error;
1271 
1272 	LOCKSTART;
1273 	dip = VTOI(ap->a_dvp);
1274 	if (dip->pmp->ronly) {
1275 		LOCKSTOP;
1276 		return (EROFS);
1277 	}
1278 
1279 	ncp = ap->a_nch->ncp;
1280 	name = ncp->nc_name;
1281 	name_len = ncp->nc_nlen;
1282 
1283 	hammer2_pfs_memory_wait(dip->pmp);
1284 	hammer2_trans_init(dip->pmp, 0);
1285 
1286 	inum = hammer2_trans_newinum(dip->pmp);
1287 
1288 	/*
1289 	 * Create the actual inode as a hidden file in the iroot, then
1290 	 * create the directory entry as a hardlink to it.  The creation
1291 	 * of the actual inode sets its nlinks to 1 which is the value
1292 	 * we desire.
1293 	 */
1294 	nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
1295 				   NULL, 0, inum,
1296 				   inum, 0, 0,
1297 				   0, &error);
1298 	if (error == 0) {
1299 		hammer2_inode_create(dip, dip, NULL, NULL,
1300 				     name, name_len, 0,
1301 				     nip->meta.inum,
1302 				     HAMMER2_OBJTYPE_HARDLINK, nip->meta.type,
1303 				     0, &error);
1304 	}
1305 
1306 	if (error) {
1307 		KKASSERT(nip == NULL);
1308 		*ap->a_vpp = NULL;
1309 	} else {
1310 		*ap->a_vpp = hammer2_igetv(nip, &error);
1311 		hammer2_inode_unlock(nip);
1312 	}
1313 
1314 	/*
1315 	 * Update dip's mtime
1316 	 */
1317 	if (error == 0) {
1318 		uint64_t mtime;
1319 
1320 		hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1321 		hammer2_update_time(&mtime);
1322 		hammer2_inode_modify(dip);
1323 		dip->meta.mtime = mtime;
1324 		hammer2_inode_unlock(dip);
1325 	}
1326 
1327 	hammer2_trans_done(dip->pmp);
1328 
1329 	if (error == 0) {
1330 		cache_setunresolved(ap->a_nch);
1331 		cache_setvp(ap->a_nch, *ap->a_vpp);
1332 	}
1333 	LOCKSTOP;
1334 	return error;
1335 }
1336 
1337 static
1338 int
1339 hammer2_vop_open(struct vop_open_args *ap)
1340 {
1341 	return vop_stdopen(ap);
1342 }
1343 
1344 /*
1345  * hammer2_vop_advlock { vp, id, op, fl, flags }
1346  */
1347 static
1348 int
1349 hammer2_vop_advlock(struct vop_advlock_args *ap)
1350 {
1351 	hammer2_inode_t *ip = VTOI(ap->a_vp);
1352 	hammer2_off_t size;
1353 
1354 	size = ip->meta.size;
1355 	return (lf_advlock(ap, &ip->advlock, size));
1356 }
1357 
1358 static
1359 int
1360 hammer2_vop_close(struct vop_close_args *ap)
1361 {
1362 	return vop_stdclose(ap);
1363 }
1364 
1365 /*
1366  * hammer2_vop_nlink { nch, dvp, vp, cred }
1367  *
1368  * Create a hardlink from (vp) to {dvp, nch}.
1369  */
1370 static
1371 int
1372 hammer2_vop_nlink(struct vop_nlink_args *ap)
1373 {
1374 	hammer2_inode_t *tdip;	/* target directory to create link in */
1375 	hammer2_inode_t *ip;	/* inode we are hardlinking to */
1376 	struct namecache *ncp;
1377 	const uint8_t *name;
1378 	size_t name_len;
1379 	int error;
1380 
1381 	if (ap->a_dvp->v_mount != ap->a_vp->v_mount)
1382 		return(EXDEV);
1383 
1384 	LOCKSTART;
1385 	tdip = VTOI(ap->a_dvp);
1386 	if (tdip->pmp->ronly) {
1387 		LOCKSTOP;
1388 		return (EROFS);
1389 	}
1390 
1391 	ncp = ap->a_nch->ncp;
1392 	name = ncp->nc_name;
1393 	name_len = ncp->nc_nlen;
1394 
1395 	/*
1396 	 * ip represents the file being hardlinked.  The file could be a
1397 	 * normal file or a hardlink target if it has already been hardlinked.
1398 	 * (with the new semantics, it will almost always be a hardlink
1399 	 * target).
1400 	 *
1401 	 * Bump nlinks and potentially also create or move the hardlink
1402 	 * target in the parent directory common to (ip) and (tdip).  The
1403 	 * consolidation code can modify ip->cluster.  The returned cluster
1404 	 * is locked.
1405 	 */
1406 	ip = VTOI(ap->a_vp);
1407 	KASSERT(ip->pmp, ("ip->pmp is NULL %p %p", ip, ip->pmp));
1408 	hammer2_pfs_memory_wait(ip->pmp);
1409 	hammer2_trans_init(ip->pmp, 0);
1410 
1411 	/*
1412 	 * Target should be an indexed inode or there's no way we will ever
1413 	 * be able to find it!
1414 	 */
1415 	KKASSERT((ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE) == 0);
1416 
1417 	error = 0;
1418 
1419 	/*
1420 	 * Can return NULL and error == EXDEV if the common parent
1421 	 * crosses a directory with the xlink flag set.
1422 	 */
1423 	hammer2_inode_lock(tdip, 0);
1424 	hammer2_inode_lock(ip, 0);
1425 
1426 	/*
1427 	 * Create the hardlink target and bump nlinks.
1428 	 */
1429 	if (error == 0) {
1430 		hammer2_inode_create(tdip, tdip, NULL, NULL,
1431 				     name, name_len, 0,
1432 				     ip->meta.inum,
1433 				     HAMMER2_OBJTYPE_HARDLINK, ip->meta.type,
1434 				     0, &error);
1435 		hammer2_inode_modify(ip);
1436 		++ip->meta.nlinks;
1437 	}
1438 	if (error == 0) {
1439 		/*
1440 		 * Update dip's mtime
1441 		 */
1442 		uint64_t mtime;
1443 
1444 		hammer2_update_time(&mtime);
1445 		hammer2_inode_modify(tdip);
1446 		tdip->meta.mtime = mtime;
1447 
1448 		cache_setunresolved(ap->a_nch);
1449 		cache_setvp(ap->a_nch, ap->a_vp);
1450 	}
1451 	hammer2_inode_unlock(ip);
1452 	hammer2_inode_unlock(tdip);
1453 
1454 	hammer2_trans_done(ip->pmp);
1455 
1456 	LOCKSTOP;
1457 	return error;
1458 }
1459 
1460 /*
1461  * hammer2_vop_ncreate { nch, dvp, vpp, cred, vap }
1462  *
1463  * The operating system has already ensured that the directory entry
1464  * does not exist and done all appropriate namespace locking.
1465  */
1466 static
1467 int
1468 hammer2_vop_ncreate(struct vop_ncreate_args *ap)
1469 {
1470 	hammer2_inode_t *dip;
1471 	hammer2_inode_t *nip;
1472 	struct namecache *ncp;
1473 	const uint8_t *name;
1474 	size_t name_len;
1475 	hammer2_tid_t inum;
1476 	int error;
1477 
1478 	LOCKSTART;
1479 	dip = VTOI(ap->a_dvp);
1480 	if (dip->pmp->ronly) {
1481 		LOCKSTOP;
1482 		return (EROFS);
1483 	}
1484 
1485 	ncp = ap->a_nch->ncp;
1486 	name = ncp->nc_name;
1487 	name_len = ncp->nc_nlen;
1488 	hammer2_pfs_memory_wait(dip->pmp);
1489 	hammer2_trans_init(dip->pmp, 0);
1490 
1491 	inum = hammer2_trans_newinum(dip->pmp);
1492 
1493 	/*
1494 	 * Create the actual inode as a hidden file in the iroot, then
1495 	 * create the directory entry as a hardlink to it.  The creation
1496 	 * of the actual inode sets its nlinks to 1 which is the value
1497 	 * we desire.
1498 	 */
1499 	nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
1500 				   NULL, 0, inum,
1501 				   inum, 0, 0,
1502 				   0, &error);
1503 
1504 	if (error == 0) {
1505 		hammer2_inode_create(dip, dip, NULL, NULL,
1506 				     name, name_len, 0,
1507 				     nip->meta.inum,
1508 				     HAMMER2_OBJTYPE_HARDLINK, nip->meta.type,
1509 				     0, &error);
1510 	}
1511 	if (error) {
1512 		KKASSERT(nip == NULL);
1513 		*ap->a_vpp = NULL;
1514 	} else {
1515 		*ap->a_vpp = hammer2_igetv(nip, &error);
1516 		hammer2_inode_unlock(nip);
1517 	}
1518 
1519 	/*
1520 	 * Update dip's mtime
1521 	 */
1522 	if (error == 0) {
1523 		uint64_t mtime;
1524 
1525 		hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1526 		hammer2_update_time(&mtime);
1527 		hammer2_inode_modify(dip);
1528 		dip->meta.mtime = mtime;
1529 		hammer2_inode_unlock(dip);
1530 	}
1531 
1532 	hammer2_trans_done(dip->pmp);
1533 
1534 	if (error == 0) {
1535 		cache_setunresolved(ap->a_nch);
1536 		cache_setvp(ap->a_nch, *ap->a_vpp);
1537 	}
1538 	LOCKSTOP;
1539 	return error;
1540 }
1541 
1542 /*
1543  * Make a device node (typically a fifo)
1544  */
1545 static
1546 int
1547 hammer2_vop_nmknod(struct vop_nmknod_args *ap)
1548 {
1549 	hammer2_inode_t *dip;
1550 	hammer2_inode_t *nip;
1551 	struct namecache *ncp;
1552 	const uint8_t *name;
1553 	size_t name_len;
1554 	hammer2_tid_t inum;
1555 	int error;
1556 
1557 	LOCKSTART;
1558 	dip = VTOI(ap->a_dvp);
1559 	if (dip->pmp->ronly) {
1560 		LOCKSTOP;
1561 		return (EROFS);
1562 	}
1563 
1564 	ncp = ap->a_nch->ncp;
1565 	name = ncp->nc_name;
1566 	name_len = ncp->nc_nlen;
1567 	hammer2_pfs_memory_wait(dip->pmp);
1568 	hammer2_trans_init(dip->pmp, 0);
1569 
1570 	/*
1571 	 * The device node is entered as the directory entry itself and not
1572 	 * as a hardlink to an inode.  Since one cannot obtain a
1573 	 * file handle on the filesystem entry representing the device, we
1574 	 * do not have to worry about indexing its inode.
1575 	 */
1576 	inum = hammer2_trans_newinum(dip->pmp);
1577 	nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
1578 				   NULL, 0, inum,
1579 				   inum, 0, 0,
1580 				   0, &error);
1581 	if (error == 0) {
1582 		hammer2_inode_create(dip, dip, NULL, NULL,
1583 				     name, name_len, 0,
1584 				     nip->meta.inum,
1585 				     HAMMER2_OBJTYPE_HARDLINK, nip->meta.type,
1586 				     0, &error);
1587 	}
1588 
1589 
1590 	if (error) {
1591 		KKASSERT(nip == NULL);
1592 		*ap->a_vpp = NULL;
1593 	} else {
1594 		*ap->a_vpp = hammer2_igetv(nip, &error);
1595 		hammer2_inode_unlock(nip);
1596 	}
1597 
1598 	/*
1599 	 * Update dip's mtime
1600 	 */
1601 	if (error == 0) {
1602 		uint64_t mtime;
1603 
1604 		hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1605 		hammer2_update_time(&mtime);
1606 		hammer2_inode_modify(dip);
1607 		dip->meta.mtime = mtime;
1608 		hammer2_inode_unlock(dip);
1609 	}
1610 
1611 	hammer2_trans_done(dip->pmp);
1612 
1613 	if (error == 0) {
1614 		cache_setunresolved(ap->a_nch);
1615 		cache_setvp(ap->a_nch, *ap->a_vpp);
1616 	}
1617 	LOCKSTOP;
1618 	return error;
1619 }
1620 
1621 /*
1622  * hammer2_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1623  */
1624 static
1625 int
1626 hammer2_vop_nsymlink(struct vop_nsymlink_args *ap)
1627 {
1628 	hammer2_inode_t *dip;
1629 	hammer2_inode_t *nip;
1630 	struct namecache *ncp;
1631 	const uint8_t *name;
1632 	size_t name_len;
1633 	hammer2_tid_t inum;
1634 	int error;
1635 
1636 	dip = VTOI(ap->a_dvp);
1637 	if (dip->pmp->ronly)
1638 		return (EROFS);
1639 
1640 	ncp = ap->a_nch->ncp;
1641 	name = ncp->nc_name;
1642 	name_len = ncp->nc_nlen;
1643 	hammer2_pfs_memory_wait(dip->pmp);
1644 	hammer2_trans_init(dip->pmp, 0);
1645 
1646 	ap->a_vap->va_type = VLNK;	/* enforce type */
1647 
1648 	/*
1649 	 * The softlink is entered into the directory itself and not
1650 	 * as a hardlink to an inode.  Since one cannot obtain a
1651 	 * file handle on the softlink itself we do not have to worry
1652 	 * about indexing its inode.
1653 	 */
1654 	inum = hammer2_trans_newinum(dip->pmp);
1655 
1656 	nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
1657 				   NULL, 0, inum,
1658 				   inum, 0, 0,
1659 				   0, &error);
1660 	if (error == 0) {
1661 		hammer2_inode_create(dip, dip, NULL, NULL,
1662 				     name, name_len, 0,
1663 				     nip->meta.inum,
1664 				     HAMMER2_OBJTYPE_HARDLINK, nip->meta.type,
1665 				     0, &error);
1666 	}
1667 
1668 
1669 	if (error) {
1670 		KKASSERT(nip == NULL);
1671 		*ap->a_vpp = NULL;
1672 		hammer2_trans_done(dip->pmp);
1673 		return error;
1674 	}
1675 	*ap->a_vpp = hammer2_igetv(nip, &error);
1676 
1677 	/*
1678 	 * Build the softlink (~like file data) and finalize the namecache.
1679 	 */
1680 	if (error == 0) {
1681 		size_t bytes;
1682 		struct uio auio;
1683 		struct iovec aiov;
1684 
1685 		bytes = strlen(ap->a_target);
1686 
1687 		hammer2_inode_unlock(nip);
1688 		bzero(&auio, sizeof(auio));
1689 		bzero(&aiov, sizeof(aiov));
1690 		auio.uio_iov = &aiov;
1691 		auio.uio_segflg = UIO_SYSSPACE;
1692 		auio.uio_rw = UIO_WRITE;
1693 		auio.uio_resid = bytes;
1694 		auio.uio_iovcnt = 1;
1695 		auio.uio_td = curthread;
1696 		aiov.iov_base = ap->a_target;
1697 		aiov.iov_len = bytes;
1698 		error = hammer2_write_file(nip, &auio, IO_APPEND, 0);
1699 		/* XXX handle error */
1700 		error = 0;
1701 	} else {
1702 		hammer2_inode_unlock(nip);
1703 	}
1704 
1705 	/*
1706 	 * Update dip's mtime
1707 	 */
1708 	if (error == 0) {
1709 		uint64_t mtime;
1710 
1711 		hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1712 		hammer2_update_time(&mtime);
1713 		hammer2_inode_modify(dip);
1714 		dip->meta.mtime = mtime;
1715 		hammer2_inode_unlock(dip);
1716 	}
1717 
1718 	hammer2_trans_done(dip->pmp);
1719 
1720 	/*
1721 	 * Finalize namecache
1722 	 */
1723 	if (error == 0) {
1724 		cache_setunresolved(ap->a_nch);
1725 		cache_setvp(ap->a_nch, *ap->a_vpp);
1726 		/* hammer2_knote(ap->a_dvp, NOTE_WRITE); */
1727 	}
1728 	return error;
1729 }
1730 
1731 /*
1732  * hammer2_vop_nremove { nch, dvp, cred }
1733  */
1734 static
1735 int
1736 hammer2_vop_nremove(struct vop_nremove_args *ap)
1737 {
1738 	hammer2_xop_unlink_t *xop;
1739 	hammer2_inode_t *dip;
1740 	hammer2_inode_t *ip;
1741 	struct namecache *ncp;
1742 	int error;
1743 	int isopen;
1744 
1745 	LOCKSTART;
1746 	dip = VTOI(ap->a_dvp);
1747 	if (dip->pmp->ronly) {
1748 		LOCKSTOP;
1749 		return(EROFS);
1750 	}
1751 
1752 	ncp = ap->a_nch->ncp;
1753 
1754 	hammer2_pfs_memory_wait(dip->pmp);
1755 	hammer2_trans_init(dip->pmp, 0);
1756 	hammer2_inode_lock(dip, 0);
1757 
1758 	/*
1759 	 * The unlink XOP unlinks the path from the directory and
1760 	 * locates and returns the cluster associated with the real inode.
1761 	 * We have to handle nlinks here on the frontend.
1762 	 */
1763 	xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING);
1764 	hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1765 	isopen = cache_isopen(ap->a_nch);
1766 	xop->isdir = 0;
1767 	xop->dopermanent = 0;
1768 	hammer2_xop_start(&xop->head, hammer2_xop_unlink);
1769 
1770 	/*
1771 	 * Collect the real inode and adjust nlinks, destroy the real
1772 	 * inode if nlinks transitions to 0 and it was the real inode
1773 	 * (else it has already been removed).
1774 	 */
1775 	error = hammer2_xop_collect(&xop->head, 0);
1776 	hammer2_inode_unlock(dip);
1777 
1778 	if (error == 0) {
1779 		ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1780 		hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1781 		if (ip) {
1782 			hammer2_inode_unlink_finisher(ip, isopen);
1783 			hammer2_inode_unlock(ip);
1784 		}
1785 	} else {
1786 		hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1787 	}
1788 
1789 	/*
1790 	 * Update dip's mtime
1791 	 */
1792 	if (error == 0) {
1793 		uint64_t mtime;
1794 
1795 		hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1796 		hammer2_update_time(&mtime);
1797 		hammer2_inode_modify(dip);
1798 		dip->meta.mtime = mtime;
1799 		hammer2_inode_unlock(dip);
1800 	}
1801 
1802 	hammer2_inode_run_sideq(dip->pmp);
1803 	hammer2_trans_done(dip->pmp);
1804 	if (error == 0)
1805 		cache_unlink(ap->a_nch);
1806 	LOCKSTOP;
1807 	return (error);
1808 }
1809 
1810 /*
1811  * hammer2_vop_nrmdir { nch, dvp, cred }
1812  */
1813 static
1814 int
1815 hammer2_vop_nrmdir(struct vop_nrmdir_args *ap)
1816 {
1817 	hammer2_xop_unlink_t *xop;
1818 	hammer2_inode_t *dip;
1819 	hammer2_inode_t *ip;
1820 	struct namecache *ncp;
1821 	int isopen;
1822 	int error;
1823 
1824 	LOCKSTART;
1825 	dip = VTOI(ap->a_dvp);
1826 	if (dip->pmp->ronly) {
1827 		LOCKSTOP;
1828 		return(EROFS);
1829 	}
1830 
1831 	hammer2_pfs_memory_wait(dip->pmp);
1832 	hammer2_trans_init(dip->pmp, 0);
1833 	hammer2_inode_lock(dip, 0);
1834 
1835 	xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING);
1836 
1837 	ncp = ap->a_nch->ncp;
1838 	hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1839 	isopen = cache_isopen(ap->a_nch);
1840 	xop->isdir = 1;
1841 	xop->dopermanent = 0;
1842 	hammer2_xop_start(&xop->head, hammer2_xop_unlink);
1843 
1844 	/*
1845 	 * Collect the real inode and adjust nlinks, destroy the real
1846 	 * inode if nlinks transitions to 0 and it was the real inode
1847 	 * (else it has already been removed).
1848 	 */
1849 	error = hammer2_xop_collect(&xop->head, 0);
1850 	hammer2_inode_unlock(dip);
1851 
1852 	if (error == 0) {
1853 		ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1854 		hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1855 		if (ip) {
1856 			hammer2_inode_unlink_finisher(ip, isopen);
1857 			hammer2_inode_unlock(ip);
1858 		}
1859 	} else {
1860 		hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1861 	}
1862 
1863 	/*
1864 	 * Update dip's mtime
1865 	 */
1866 	if (error == 0) {
1867 		uint64_t mtime;
1868 
1869 		hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1870 		hammer2_update_time(&mtime);
1871 		hammer2_inode_modify(dip);
1872 		dip->meta.mtime = mtime;
1873 		hammer2_inode_unlock(dip);
1874 	}
1875 
1876 	hammer2_inode_run_sideq(dip->pmp);
1877 	hammer2_trans_done(dip->pmp);
1878 	if (error == 0)
1879 		cache_unlink(ap->a_nch);
1880 	LOCKSTOP;
1881 	return (error);
1882 }
1883 
1884 /*
1885  * hammer2_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1886  */
1887 static
1888 int
1889 hammer2_vop_nrename(struct vop_nrename_args *ap)
1890 {
1891 	struct namecache *fncp;
1892 	struct namecache *tncp;
1893 	hammer2_inode_t *fdip;
1894 	hammer2_inode_t *tdip;
1895 	hammer2_inode_t *ip;
1896 	const uint8_t *fname;
1897 	size_t fname_len;
1898 	const uint8_t *tname;
1899 	size_t tname_len;
1900 	int error;
1901 	int tnch_error;
1902 	int update_tdip;
1903 	int update_fdip;
1904 	hammer2_key_t tlhc;
1905 
1906 	if (ap->a_fdvp->v_mount != ap->a_tdvp->v_mount)
1907 		return(EXDEV);
1908 	if (ap->a_fdvp->v_mount != ap->a_fnch->ncp->nc_vp->v_mount)
1909 		return(EXDEV);
1910 
1911 	fdip = VTOI(ap->a_fdvp);	/* source directory */
1912 	tdip = VTOI(ap->a_tdvp);	/* target directory */
1913 
1914 	if (fdip->pmp->ronly)
1915 		return(EROFS);
1916 
1917 	LOCKSTART;
1918 	fncp = ap->a_fnch->ncp;		/* entry name in source */
1919 	fname = fncp->nc_name;
1920 	fname_len = fncp->nc_nlen;
1921 
1922 	tncp = ap->a_tnch->ncp;		/* entry name in target */
1923 	tname = tncp->nc_name;
1924 	tname_len = tncp->nc_nlen;
1925 
1926 	hammer2_pfs_memory_wait(tdip->pmp);
1927 	hammer2_trans_init(tdip->pmp, 0);
1928 
1929 	update_tdip = 0;
1930 	update_fdip = 0;
1931 
1932 	/*
1933 	 * ip is the inode being renamed.  If this is a hardlink then
1934 	 * ip represents the actual file and not the hardlink marker.
1935 	 */
1936 	ip = VTOI(fncp->nc_vp);
1937 
1938 	KKASSERT((ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE) == 0);
1939 
1940 	/*
1941 	 * Can return NULL and error == EXDEV if the common parent
1942 	 * crosses a directory with the xlink flag set.
1943 	 */
1944 	error = 0;
1945 	hammer2_inode_lock(fdip, 0);
1946 	hammer2_inode_lock(tdip, 0);
1947 	hammer2_inode_ref(ip);		/* extra ref */
1948 
1949 	hammer2_inode_lock(ip, 0);
1950 
1951 	/*
1952 	 * Delete the target namespace.
1953 	 */
1954 	{
1955 		hammer2_xop_unlink_t *xop2;
1956 		hammer2_inode_t *tip;
1957 		int isopen;
1958 
1959 		/*
1960 		 * The unlink XOP unlinks the path from the directory and
1961 		 * locates and returns the cluster associated with the real
1962 		 * inode.  We have to handle nlinks here on the frontend.
1963 		 */
1964 		xop2 = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING);
1965 		hammer2_xop_setname(&xop2->head, tname, tname_len);
1966 		isopen = cache_isopen(ap->a_tnch);
1967 		xop2->isdir = -1;
1968 		xop2->dopermanent = 0;
1969 		hammer2_xop_start(&xop2->head, hammer2_xop_unlink);
1970 
1971 		/*
1972 		 * Collect the real inode and adjust nlinks, destroy the real
1973 		 * inode if nlinks transitions to 0 and it was the real inode
1974 		 * (else it has already been removed).
1975 		 */
1976 		tnch_error = hammer2_xop_collect(&xop2->head, 0);
1977 		/* hammer2_inode_unlock(tdip); */
1978 
1979 		if (tnch_error == 0) {
1980 			tip = hammer2_inode_get(tdip->pmp, NULL,
1981 						&xop2->head.cluster, -1);
1982 			hammer2_xop_retire(&xop2->head, HAMMER2_XOPMASK_VOP);
1983 			if (tip) {
1984 				hammer2_inode_unlink_finisher(tip, isopen);
1985 				hammer2_inode_unlock(tip);
1986 			}
1987 		} else {
1988 			hammer2_xop_retire(&xop2->head, HAMMER2_XOPMASK_VOP);
1989 		}
1990 		/* hammer2_inode_lock(tdip, 0); */
1991 
1992 		if (tnch_error && tnch_error != ENOENT) {
1993 			error = tnch_error;
1994 			goto done2;
1995 		}
1996 		update_tdip = 1;
1997 	}
1998 
1999 	/*
2000 	 * Resolve the collision space for (tdip, tname, tname_len)
2001 	 *
2002 	 * tdip must be held exclusively locked to prevent races.
2003 	 */
2004 	{
2005 		hammer2_xop_scanlhc_t *sxop;
2006 		hammer2_tid_t lhcbase;
2007 
2008 		tlhc = hammer2_dirhash(tname, tname_len);
2009 		lhcbase = tlhc;
2010 		sxop = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING);
2011 		sxop->lhc = tlhc;
2012 		hammer2_xop_start(&sxop->head, hammer2_xop_scanlhc);
2013 		while ((error = hammer2_xop_collect(&sxop->head, 0)) == 0) {
2014 			if (tlhc != sxop->head.cluster.focus->bref.key)
2015 				break;
2016 			++tlhc;
2017 		}
2018 		hammer2_xop_retire(&sxop->head, HAMMER2_XOPMASK_VOP);
2019 
2020 		if (error) {
2021 			if (error != ENOENT)
2022 				goto done2;
2023 			++tlhc;
2024 			error = 0;
2025 		}
2026 		if ((lhcbase ^ tlhc) & ~HAMMER2_DIRHASH_LOMASK) {
2027 			error = ENOSPC;
2028 			goto done2;
2029 		}
2030 	}
2031 
2032 	/*
2033 	 * Everything is setup, do the rename.
2034 	 *
2035 	 * We have to synchronize ip->meta to the underlying operation.
2036 	 *
2037 	 * NOTE: To avoid deadlocks we cannot lock (ip) while we are
2038 	 *	 unlinking elements from their directories.  Locking
2039 	 *	 the nlinks field does not lock the whole inode.
2040 	 */
2041 	/* hammer2_inode_lock(ip, 0); */
2042 	if (error == 0) {
2043 		hammer2_xop_nrename_t *xop4;
2044 
2045 		xop4 = hammer2_xop_alloc(fdip, HAMMER2_XOP_MODIFYING);
2046 		xop4->lhc = tlhc;
2047 		xop4->ip_key = ip->meta.name_key;
2048 		hammer2_xop_setip2(&xop4->head, ip);
2049 		hammer2_xop_setip3(&xop4->head, tdip);
2050 		hammer2_xop_setname(&xop4->head, fname, fname_len);
2051 		hammer2_xop_setname2(&xop4->head, tname, tname_len);
2052 		hammer2_xop_start(&xop4->head, hammer2_xop_nrename);
2053 
2054 		error = hammer2_xop_collect(&xop4->head, 0);
2055 		hammer2_xop_retire(&xop4->head, HAMMER2_XOPMASK_VOP);
2056 
2057 		if (error == ENOENT)
2058 			error = 0;
2059 		if (error == 0 &&
2060 		    (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE)) {
2061 			hammer2_inode_modify(ip);
2062 			ip->meta.name_len = tname_len;
2063 			ip->meta.name_key = tlhc;
2064 
2065 		}
2066 		update_fdip = 1;
2067 		update_fdip = 1;
2068 	}
2069 
2070 done2:
2071 	/*
2072 	 * Update directory mtimes to represent the something changed.
2073 	 */
2074 	if (update_fdip || update_tdip) {
2075 		uint64_t mtime;
2076 
2077 		hammer2_update_time(&mtime);
2078 		if (update_fdip) {
2079 			hammer2_inode_modify(fdip);
2080 			fdip->meta.mtime = mtime;
2081 		}
2082 		if (update_tdip) {
2083 			hammer2_inode_modify(tdip);
2084 			tdip->meta.mtime = mtime;
2085 		}
2086 	}
2087 	hammer2_inode_unlock(ip);
2088 	hammer2_inode_unlock(tdip);
2089 	hammer2_inode_unlock(fdip);
2090 	hammer2_inode_drop(ip);
2091 	hammer2_inode_run_sideq(fdip->pmp);
2092 
2093 	hammer2_trans_done(tdip->pmp);
2094 
2095 	/*
2096 	 * Issue the namecache update after unlocking all the internal
2097 	 * hammer structures, otherwise we might deadlock.
2098 	 */
2099 	if (tnch_error == 0) {
2100 		cache_unlink(ap->a_tnch);
2101 		cache_setunresolved(ap->a_tnch);
2102 	}
2103 	if (error == 0)
2104 		cache_rename(ap->a_fnch, ap->a_tnch);
2105 
2106 	LOCKSTOP;
2107 	return (error);
2108 }
2109 
2110 /*
2111  * hammer2_vop_ioctl { vp, command, data, fflag, cred }
2112  */
2113 static
2114 int
2115 hammer2_vop_ioctl(struct vop_ioctl_args *ap)
2116 {
2117 	hammer2_inode_t *ip;
2118 	int error;
2119 
2120 	LOCKSTART;
2121 	ip = VTOI(ap->a_vp);
2122 
2123 	error = hammer2_ioctl(ip, ap->a_command, (void *)ap->a_data,
2124 			      ap->a_fflag, ap->a_cred);
2125 	LOCKSTOP;
2126 	return (error);
2127 }
2128 
2129 static
2130 int
2131 hammer2_vop_mountctl(struct vop_mountctl_args *ap)
2132 {
2133 	struct mount *mp;
2134 	hammer2_pfs_t *pmp;
2135 	int rc;
2136 
2137 	LOCKSTART;
2138 	switch (ap->a_op) {
2139 	case (MOUNTCTL_SET_EXPORT):
2140 		mp = ap->a_head.a_ops->head.vv_mount;
2141 		pmp = MPTOPMP(mp);
2142 
2143 		if (ap->a_ctllen != sizeof(struct export_args))
2144 			rc = (EINVAL);
2145 		else
2146 			rc = vfs_export(mp, &pmp->export,
2147 					(const struct export_args *)ap->a_ctl);
2148 		break;
2149 	default:
2150 		rc = vop_stdmountctl(ap);
2151 		break;
2152 	}
2153 	LOCKSTOP;
2154 	return (rc);
2155 }
2156 
2157 /*
2158  * KQFILTER
2159  */
2160 static void filt_hammer2detach(struct knote *kn);
2161 static int filt_hammer2read(struct knote *kn, long hint);
2162 static int filt_hammer2write(struct knote *kn, long hint);
2163 static int filt_hammer2vnode(struct knote *kn, long hint);
2164 
2165 static struct filterops hammer2read_filtops =
2166 	{ FILTEROP_ISFD | FILTEROP_MPSAFE,
2167 	  NULL, filt_hammer2detach, filt_hammer2read };
2168 static struct filterops hammer2write_filtops =
2169 	{ FILTEROP_ISFD | FILTEROP_MPSAFE,
2170 	  NULL, filt_hammer2detach, filt_hammer2write };
2171 static struct filterops hammer2vnode_filtops =
2172 	{ FILTEROP_ISFD | FILTEROP_MPSAFE,
2173 	  NULL, filt_hammer2detach, filt_hammer2vnode };
2174 
2175 static
2176 int
2177 hammer2_vop_kqfilter(struct vop_kqfilter_args *ap)
2178 {
2179 	struct vnode *vp = ap->a_vp;
2180 	struct knote *kn = ap->a_kn;
2181 
2182 	switch (kn->kn_filter) {
2183 	case EVFILT_READ:
2184 		kn->kn_fop = &hammer2read_filtops;
2185 		break;
2186 	case EVFILT_WRITE:
2187 		kn->kn_fop = &hammer2write_filtops;
2188 		break;
2189 	case EVFILT_VNODE:
2190 		kn->kn_fop = &hammer2vnode_filtops;
2191 		break;
2192 	default:
2193 		return (EOPNOTSUPP);
2194 	}
2195 
2196 	kn->kn_hook = (caddr_t)vp;
2197 
2198 	knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2199 
2200 	return(0);
2201 }
2202 
2203 static void
2204 filt_hammer2detach(struct knote *kn)
2205 {
2206 	struct vnode *vp = (void *)kn->kn_hook;
2207 
2208 	knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2209 }
2210 
2211 static int
2212 filt_hammer2read(struct knote *kn, long hint)
2213 {
2214 	struct vnode *vp = (void *)kn->kn_hook;
2215 	hammer2_inode_t *ip = VTOI(vp);
2216 	off_t off;
2217 
2218 	if (hint == NOTE_REVOKE) {
2219 		kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2220 		return(1);
2221 	}
2222 	off = ip->meta.size - kn->kn_fp->f_offset;
2223 	kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
2224 	if (kn->kn_sfflags & NOTE_OLDAPI)
2225 		return(1);
2226 	return (kn->kn_data != 0);
2227 }
2228 
2229 
2230 static int
2231 filt_hammer2write(struct knote *kn, long hint)
2232 {
2233 	if (hint == NOTE_REVOKE)
2234 		kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2235 	kn->kn_data = 0;
2236 	return (1);
2237 }
2238 
2239 static int
2240 filt_hammer2vnode(struct knote *kn, long hint)
2241 {
2242 	if (kn->kn_sfflags & hint)
2243 		kn->kn_fflags |= hint;
2244 	if (hint == NOTE_REVOKE) {
2245 		kn->kn_flags |= (EV_EOF | EV_NODATA);
2246 		return (1);
2247 	}
2248 	return (kn->kn_fflags != 0);
2249 }
2250 
2251 /*
2252  * FIFO VOPS
2253  */
2254 static
2255 int
2256 hammer2_vop_markatime(struct vop_markatime_args *ap)
2257 {
2258 	hammer2_inode_t *ip;
2259 	struct vnode *vp;
2260 
2261 	vp = ap->a_vp;
2262 	ip = VTOI(vp);
2263 
2264 	if (ip->pmp->ronly)
2265 		return(EROFS);
2266 	return(0);
2267 }
2268 
2269 static
2270 int
2271 hammer2_vop_fifokqfilter(struct vop_kqfilter_args *ap)
2272 {
2273 	int error;
2274 
2275 	error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2276 	if (error)
2277 		error = hammer2_vop_kqfilter(ap);
2278 	return(error);
2279 }
2280 
2281 /*
2282  * VOPS vector
2283  */
2284 struct vop_ops hammer2_vnode_vops = {
2285 	.vop_default	= vop_defaultop,
2286 	.vop_fsync	= hammer2_vop_fsync,
2287 	.vop_getpages	= vop_stdgetpages,
2288 	.vop_putpages	= vop_stdputpages,
2289 	.vop_access	= hammer2_vop_access,
2290 	.vop_advlock	= hammer2_vop_advlock,
2291 	.vop_close	= hammer2_vop_close,
2292 	.vop_nlink	= hammer2_vop_nlink,
2293 	.vop_ncreate	= hammer2_vop_ncreate,
2294 	.vop_nsymlink	= hammer2_vop_nsymlink,
2295 	.vop_nremove	= hammer2_vop_nremove,
2296 	.vop_nrmdir	= hammer2_vop_nrmdir,
2297 	.vop_nrename	= hammer2_vop_nrename,
2298 	.vop_getattr	= hammer2_vop_getattr,
2299 	.vop_setattr	= hammer2_vop_setattr,
2300 	.vop_readdir	= hammer2_vop_readdir,
2301 	.vop_readlink	= hammer2_vop_readlink,
2302 	.vop_getpages	= vop_stdgetpages,
2303 	.vop_putpages	= vop_stdputpages,
2304 	.vop_read	= hammer2_vop_read,
2305 	.vop_write	= hammer2_vop_write,
2306 	.vop_open	= hammer2_vop_open,
2307 	.vop_inactive	= hammer2_vop_inactive,
2308 	.vop_reclaim 	= hammer2_vop_reclaim,
2309 	.vop_nresolve	= hammer2_vop_nresolve,
2310 	.vop_nlookupdotdot = hammer2_vop_nlookupdotdot,
2311 	.vop_nmkdir 	= hammer2_vop_nmkdir,
2312 	.vop_nmknod 	= hammer2_vop_nmknod,
2313 	.vop_ioctl	= hammer2_vop_ioctl,
2314 	.vop_mountctl	= hammer2_vop_mountctl,
2315 	.vop_bmap	= hammer2_vop_bmap,
2316 	.vop_strategy	= hammer2_vop_strategy,
2317         .vop_kqfilter	= hammer2_vop_kqfilter
2318 };
2319 
2320 struct vop_ops hammer2_spec_vops = {
2321         .vop_default =          vop_defaultop,
2322         .vop_fsync =            hammer2_vop_fsync,
2323         .vop_read =             vop_stdnoread,
2324         .vop_write =            vop_stdnowrite,
2325         .vop_access =           hammer2_vop_access,
2326         .vop_close =            hammer2_vop_close,
2327         .vop_markatime =        hammer2_vop_markatime,
2328         .vop_getattr =          hammer2_vop_getattr,
2329         .vop_inactive =         hammer2_vop_inactive,
2330         .vop_reclaim =          hammer2_vop_reclaim,
2331         .vop_setattr =          hammer2_vop_setattr
2332 };
2333 
2334 struct vop_ops hammer2_fifo_vops = {
2335         .vop_default =          fifo_vnoperate,
2336         .vop_fsync =            hammer2_vop_fsync,
2337 #if 0
2338         .vop_read =             hammer2_vop_fiforead,
2339         .vop_write =            hammer2_vop_fifowrite,
2340 #endif
2341         .vop_access =           hammer2_vop_access,
2342 #if 0
2343         .vop_close =            hammer2_vop_fifoclose,
2344 #endif
2345         .vop_markatime =        hammer2_vop_markatime,
2346         .vop_getattr =          hammer2_vop_getattr,
2347         .vop_inactive =         hammer2_vop_inactive,
2348         .vop_reclaim =          hammer2_vop_reclaim,
2349         .vop_setattr =          hammer2_vop_setattr,
2350         .vop_kqfilter =         hammer2_vop_fifokqfilter
2351 };
2352 
2353